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/slab.h>
30 #include <linux/migrate.h>
31 #include <linux/ratelimit.h>
32 #include <linux/uuid.h>
33 #include <linux/semaphore.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "print-tree.h"
42 #include "async-thread.h"
45 #include "free-space-cache.h"
46 #include "inode-map.h"
47 #include "check-integrity.h"
48 #include "rcu-string.h"
49 #include "dev-replace.h"
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_destroy_delalloc_inodes(struct btrfs_root *root);
68 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
69 struct extent_io_tree *dirty_pages,
71 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
72 struct extent_io_tree *pinned_extents);
73 static int btrfs_cleanup_transaction(struct btrfs_root *root);
74 static void btrfs_error_commit_super(struct btrfs_root *root);
77 * end_io_wq structs are used to do processing in task context when an IO is
78 * complete. This is used during reads to verify checksums, and it is used
79 * by writes to insert metadata for new file extents after IO is complete.
85 struct btrfs_fs_info *info;
88 struct list_head list;
89 struct btrfs_work work;
93 * async submit bios are used to offload expensive checksumming
94 * onto the worker threads. They checksum file and metadata bios
95 * just before they are sent down the IO stack.
97 struct async_submit_bio {
100 struct list_head list;
101 extent_submit_bio_hook_t *submit_bio_start;
102 extent_submit_bio_hook_t *submit_bio_done;
105 unsigned long bio_flags;
107 * bio_offset is optional, can be used if the pages in the bio
108 * can't tell us where in the file the bio should go
111 struct btrfs_work work;
116 * Lockdep class keys for extent_buffer->lock's in this root. For a given
117 * eb, the lockdep key is determined by the btrfs_root it belongs to and
118 * the level the eb occupies in the tree.
120 * Different roots are used for different purposes and may nest inside each
121 * other and they require separate keysets. As lockdep keys should be
122 * static, assign keysets according to the purpose of the root as indicated
123 * by btrfs_root->objectid. This ensures that all special purpose roots
124 * have separate keysets.
126 * Lock-nesting across peer nodes is always done with the immediate parent
127 * node locked thus preventing deadlock. As lockdep doesn't know this, use
128 * subclass to avoid triggering lockdep warning in such cases.
130 * The key is set by the readpage_end_io_hook after the buffer has passed
131 * csum validation but before the pages are unlocked. It is also set by
132 * btrfs_init_new_buffer on freshly allocated blocks.
134 * We also add a check to make sure the highest level of the tree is the
135 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
136 * needs update as well.
138 #ifdef CONFIG_DEBUG_LOCK_ALLOC
139 # if BTRFS_MAX_LEVEL != 8
143 static struct btrfs_lockdep_keyset {
144 u64 id; /* root objectid */
145 const char *name_stem; /* lock name stem */
146 char names[BTRFS_MAX_LEVEL + 1][20];
147 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
148 } btrfs_lockdep_keysets[] = {
149 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
150 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
151 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
152 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
153 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
154 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
155 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
156 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
157 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
158 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
159 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
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 btrfs_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
304 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
306 root->fs_info->sb->s_id, buf->start,
307 val, found, btrfs_header_level(buf));
308 if (result != (char *)&inline_result)
313 write_extent_buffer(buf, result, 0, csum_size);
315 if (result != (char *)&inline_result)
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
326 static int verify_parent_transid(struct extent_io_tree *io_tree,
327 struct extent_buffer *eb, u64 parent_transid,
330 struct extent_state *cached_state = NULL;
333 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
339 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
341 if (extent_buffer_uptodate(eb) &&
342 btrfs_header_generation(eb) == parent_transid) {
346 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
348 eb->start, parent_transid, btrfs_header_generation(eb));
350 clear_extent_buffer_uptodate(eb);
352 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
353 &cached_state, GFP_NOFS);
358 * Return 0 if the superblock checksum type matches the checksum value of that
359 * algorithm. Pass the raw disk superblock data.
361 static int btrfs_check_super_csum(char *raw_disk_sb)
363 struct btrfs_super_block *disk_sb =
364 (struct btrfs_super_block *)raw_disk_sb;
365 u16 csum_type = btrfs_super_csum_type(disk_sb);
368 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
370 const int csum_size = sizeof(crc);
371 char result[csum_size];
374 * The super_block structure does not span the whole
375 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
376 * is filled with zeros and is included in the checkum.
378 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
379 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
380 btrfs_csum_final(crc, result);
382 if (memcmp(raw_disk_sb, result, csum_size))
385 if (ret && btrfs_super_generation(disk_sb) < 10) {
387 "BTRFS: super block crcs don't match, older mkfs detected\n");
392 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
393 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
402 * helper to read a given tree block, doing retries as required when
403 * the checksums don't match and we have alternate mirrors to try.
405 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
406 struct extent_buffer *eb,
407 u64 start, u64 parent_transid)
409 struct extent_io_tree *io_tree;
414 int failed_mirror = 0;
416 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
417 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
419 ret = read_extent_buffer_pages(io_tree, eb, start,
421 btree_get_extent, mirror_num);
423 if (!verify_parent_transid(io_tree, eb,
431 * This buffer's crc is fine, but its contents are corrupted, so
432 * there is no reason to read the other copies, they won't be
435 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
438 num_copies = btrfs_num_copies(root->fs_info,
443 if (!failed_mirror) {
445 failed_mirror = eb->read_mirror;
449 if (mirror_num == failed_mirror)
452 if (mirror_num > num_copies)
456 if (failed && !ret && failed_mirror)
457 repair_eb_io_failure(root, eb, failed_mirror);
463 * checksum a dirty tree block before IO. This has extra checks to make sure
464 * we only fill in the checksum field in the first page of a multi-page block
467 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
469 u64 start = page_offset(page);
471 struct extent_buffer *eb;
473 eb = (struct extent_buffer *)page->private;
474 if (page != eb->pages[0])
476 found_start = btrfs_header_bytenr(eb);
477 if (WARN_ON(found_start != start || !PageUptodate(page)))
479 csum_tree_block(root, eb, 0);
483 static int check_tree_block_fsid(struct btrfs_root *root,
484 struct extent_buffer *eb)
486 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
487 u8 fsid[BTRFS_UUID_SIZE];
490 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
492 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
496 fs_devices = fs_devices->seed;
501 #define CORRUPT(reason, eb, root, slot) \
502 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
503 "root=%llu, slot=%d", reason, \
504 btrfs_header_bytenr(eb), root->objectid, slot)
506 static noinline int check_leaf(struct btrfs_root *root,
507 struct extent_buffer *leaf)
509 struct btrfs_key key;
510 struct btrfs_key leaf_key;
511 u32 nritems = btrfs_header_nritems(leaf);
517 /* Check the 0 item */
518 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
519 BTRFS_LEAF_DATA_SIZE(root)) {
520 CORRUPT("invalid item offset size pair", leaf, root, 0);
525 * Check to make sure each items keys are in the correct order and their
526 * offsets make sense. We only have to loop through nritems-1 because
527 * we check the current slot against the next slot, which verifies the
528 * next slot's offset+size makes sense and that the current's slot
531 for (slot = 0; slot < nritems - 1; slot++) {
532 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
533 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
535 /* Make sure the keys are in the right order */
536 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
537 CORRUPT("bad key order", leaf, root, slot);
542 * Make sure the offset and ends are right, remember that the
543 * item data starts at the end of the leaf and grows towards the
546 if (btrfs_item_offset_nr(leaf, slot) !=
547 btrfs_item_end_nr(leaf, slot + 1)) {
548 CORRUPT("slot offset bad", leaf, root, slot);
553 * Check to make sure that we don't point outside of the leaf,
554 * just incase all the items are consistent to eachother, but
555 * all point outside of the leaf.
557 if (btrfs_item_end_nr(leaf, slot) >
558 BTRFS_LEAF_DATA_SIZE(root)) {
559 CORRUPT("slot end outside of leaf", leaf, root, slot);
567 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
568 u64 phy_offset, struct page *page,
569 u64 start, u64 end, int mirror)
573 struct extent_buffer *eb;
574 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
581 eb = (struct extent_buffer *)page->private;
583 /* the pending IO might have been the only thing that kept this buffer
584 * in memory. Make sure we have a ref for all this other checks
586 extent_buffer_get(eb);
588 reads_done = atomic_dec_and_test(&eb->io_pages);
592 eb->read_mirror = mirror;
593 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
598 found_start = btrfs_header_bytenr(eb);
599 if (found_start != eb->start) {
600 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start "
602 found_start, eb->start);
606 if (check_tree_block_fsid(root, eb)) {
607 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n",
612 found_level = btrfs_header_level(eb);
613 if (found_level >= BTRFS_MAX_LEVEL) {
614 btrfs_info(root->fs_info, "bad tree block level %d",
615 (int)btrfs_header_level(eb));
620 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
623 ret = csum_tree_block(root, eb, 1);
630 * If this is a leaf block and it is corrupt, set the corrupt bit so
631 * that we don't try and read the other copies of this block, just
634 if (found_level == 0 && check_leaf(root, eb)) {
635 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
640 set_extent_buffer_uptodate(eb);
643 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
644 btree_readahead_hook(root, eb, eb->start, ret);
648 * our io error hook is going to dec the io pages
649 * again, we have to make sure it has something
652 atomic_inc(&eb->io_pages);
653 clear_extent_buffer_uptodate(eb);
655 free_extent_buffer(eb);
660 static int btree_io_failed_hook(struct page *page, int failed_mirror)
662 struct extent_buffer *eb;
663 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
665 eb = (struct extent_buffer *)page->private;
666 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
667 eb->read_mirror = failed_mirror;
668 atomic_dec(&eb->io_pages);
669 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
670 btree_readahead_hook(root, eb, eb->start, -EIO);
671 return -EIO; /* we fixed nothing */
674 static void end_workqueue_bio(struct bio *bio, int err)
676 struct end_io_wq *end_io_wq = bio->bi_private;
677 struct btrfs_fs_info *fs_info;
679 fs_info = end_io_wq->info;
680 end_io_wq->error = err;
681 end_io_wq->work.func = end_workqueue_fn;
682 end_io_wq->work.flags = 0;
684 if (bio->bi_rw & REQ_WRITE) {
685 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
686 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
688 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
689 btrfs_queue_worker(&fs_info->endio_freespace_worker,
691 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
692 btrfs_queue_worker(&fs_info->endio_raid56_workers,
695 btrfs_queue_worker(&fs_info->endio_write_workers,
698 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
699 btrfs_queue_worker(&fs_info->endio_raid56_workers,
701 else if (end_io_wq->metadata)
702 btrfs_queue_worker(&fs_info->endio_meta_workers,
705 btrfs_queue_worker(&fs_info->endio_workers,
711 * For the metadata arg you want
714 * 1 - if normal metadta
715 * 2 - if writing to the free space cache area
716 * 3 - raid parity work
718 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
721 struct end_io_wq *end_io_wq;
722 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
726 end_io_wq->private = bio->bi_private;
727 end_io_wq->end_io = bio->bi_end_io;
728 end_io_wq->info = info;
729 end_io_wq->error = 0;
730 end_io_wq->bio = bio;
731 end_io_wq->metadata = metadata;
733 bio->bi_private = end_io_wq;
734 bio->bi_end_io = end_workqueue_bio;
738 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
740 unsigned long limit = min_t(unsigned long,
741 info->workers.max_workers,
742 info->fs_devices->open_devices);
746 static void run_one_async_start(struct btrfs_work *work)
748 struct async_submit_bio *async;
751 async = container_of(work, struct async_submit_bio, work);
752 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
753 async->mirror_num, async->bio_flags,
759 static void run_one_async_done(struct btrfs_work *work)
761 struct btrfs_fs_info *fs_info;
762 struct async_submit_bio *async;
765 async = container_of(work, struct async_submit_bio, work);
766 fs_info = BTRFS_I(async->inode)->root->fs_info;
768 limit = btrfs_async_submit_limit(fs_info);
769 limit = limit * 2 / 3;
771 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
772 waitqueue_active(&fs_info->async_submit_wait))
773 wake_up(&fs_info->async_submit_wait);
775 /* If an error occured we just want to clean up the bio and move on */
777 bio_endio(async->bio, async->error);
781 async->submit_bio_done(async->inode, async->rw, async->bio,
782 async->mirror_num, async->bio_flags,
786 static void run_one_async_free(struct btrfs_work *work)
788 struct async_submit_bio *async;
790 async = container_of(work, struct async_submit_bio, work);
794 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
795 int rw, struct bio *bio, int mirror_num,
796 unsigned long bio_flags,
798 extent_submit_bio_hook_t *submit_bio_start,
799 extent_submit_bio_hook_t *submit_bio_done)
801 struct async_submit_bio *async;
803 async = kmalloc(sizeof(*async), GFP_NOFS);
807 async->inode = inode;
810 async->mirror_num = mirror_num;
811 async->submit_bio_start = submit_bio_start;
812 async->submit_bio_done = submit_bio_done;
814 async->work.func = run_one_async_start;
815 async->work.ordered_func = run_one_async_done;
816 async->work.ordered_free = run_one_async_free;
818 async->work.flags = 0;
819 async->bio_flags = bio_flags;
820 async->bio_offset = bio_offset;
824 atomic_inc(&fs_info->nr_async_submits);
827 btrfs_set_work_high_prio(&async->work);
829 btrfs_queue_worker(&fs_info->workers, &async->work);
831 while (atomic_read(&fs_info->async_submit_draining) &&
832 atomic_read(&fs_info->nr_async_submits)) {
833 wait_event(fs_info->async_submit_wait,
834 (atomic_read(&fs_info->nr_async_submits) == 0));
840 static int btree_csum_one_bio(struct bio *bio)
842 struct bio_vec *bvec = bio->bi_io_vec;
844 struct btrfs_root *root;
847 WARN_ON(bio->bi_vcnt <= 0);
848 while (bio_index < bio->bi_vcnt) {
849 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
850 ret = csum_dirty_buffer(root, bvec->bv_page);
859 static int __btree_submit_bio_start(struct inode *inode, int rw,
860 struct bio *bio, int mirror_num,
861 unsigned long bio_flags,
865 * when we're called for a write, we're already in the async
866 * submission context. Just jump into btrfs_map_bio
868 return btree_csum_one_bio(bio);
871 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
872 int mirror_num, unsigned long bio_flags,
878 * when we're called for a write, we're already in the async
879 * submission context. Just jump into btrfs_map_bio
881 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
887 static int check_async_write(struct inode *inode, unsigned long bio_flags)
889 if (bio_flags & EXTENT_BIO_TREE_LOG)
898 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
899 int mirror_num, unsigned long bio_flags,
902 int async = check_async_write(inode, bio_flags);
905 if (!(rw & REQ_WRITE)) {
907 * called for a read, do the setup so that checksum validation
908 * can happen in the async kernel threads
910 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
914 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
917 ret = btree_csum_one_bio(bio);
920 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
924 * kthread helpers are used to submit writes so that
925 * checksumming can happen in parallel across all CPUs
927 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
928 inode, rw, bio, mirror_num, 0,
930 __btree_submit_bio_start,
931 __btree_submit_bio_done);
941 #ifdef CONFIG_MIGRATION
942 static int btree_migratepage(struct address_space *mapping,
943 struct page *newpage, struct page *page,
944 enum migrate_mode mode)
947 * we can't safely write a btree page from here,
948 * we haven't done the locking hook
953 * Buffers may be managed in a filesystem specific way.
954 * We must have no buffers or drop them.
956 if (page_has_private(page) &&
957 !try_to_release_page(page, GFP_KERNEL))
959 return migrate_page(mapping, newpage, page, mode);
964 static int btree_writepages(struct address_space *mapping,
965 struct writeback_control *wbc)
967 struct btrfs_fs_info *fs_info;
970 if (wbc->sync_mode == WB_SYNC_NONE) {
972 if (wbc->for_kupdate)
975 fs_info = BTRFS_I(mapping->host)->root->fs_info;
976 /* this is a bit racy, but that's ok */
977 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
978 BTRFS_DIRTY_METADATA_THRESH);
982 return btree_write_cache_pages(mapping, wbc);
985 static int btree_readpage(struct file *file, struct page *page)
987 struct extent_io_tree *tree;
988 tree = &BTRFS_I(page->mapping->host)->io_tree;
989 return extent_read_full_page(tree, page, btree_get_extent, 0);
992 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
994 if (PageWriteback(page) || PageDirty(page))
997 return try_release_extent_buffer(page);
1000 static void btree_invalidatepage(struct page *page, unsigned int offset,
1001 unsigned int length)
1003 struct extent_io_tree *tree;
1004 tree = &BTRFS_I(page->mapping->host)->io_tree;
1005 extent_invalidatepage(tree, page, offset);
1006 btree_releasepage(page, GFP_NOFS);
1007 if (PagePrivate(page)) {
1008 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1009 "page private not zero on page %llu",
1010 (unsigned long long)page_offset(page));
1011 ClearPagePrivate(page);
1012 set_page_private(page, 0);
1013 page_cache_release(page);
1017 static int btree_set_page_dirty(struct page *page)
1020 struct extent_buffer *eb;
1022 BUG_ON(!PagePrivate(page));
1023 eb = (struct extent_buffer *)page->private;
1025 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1026 BUG_ON(!atomic_read(&eb->refs));
1027 btrfs_assert_tree_locked(eb);
1029 return __set_page_dirty_nobuffers(page);
1032 static const struct address_space_operations btree_aops = {
1033 .readpage = btree_readpage,
1034 .writepages = btree_writepages,
1035 .releasepage = btree_releasepage,
1036 .invalidatepage = btree_invalidatepage,
1037 #ifdef CONFIG_MIGRATION
1038 .migratepage = btree_migratepage,
1040 .set_page_dirty = btree_set_page_dirty,
1043 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1046 struct extent_buffer *buf = NULL;
1047 struct inode *btree_inode = root->fs_info->btree_inode;
1050 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1053 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1054 buf, 0, WAIT_NONE, btree_get_extent, 0);
1055 free_extent_buffer(buf);
1059 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1060 int mirror_num, struct extent_buffer **eb)
1062 struct extent_buffer *buf = NULL;
1063 struct inode *btree_inode = root->fs_info->btree_inode;
1064 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1067 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1071 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1073 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1074 btree_get_extent, mirror_num);
1076 free_extent_buffer(buf);
1080 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1081 free_extent_buffer(buf);
1083 } else if (extent_buffer_uptodate(buf)) {
1086 free_extent_buffer(buf);
1091 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1092 u64 bytenr, u32 blocksize)
1094 return find_extent_buffer(root->fs_info, bytenr);
1097 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1098 u64 bytenr, u32 blocksize)
1100 return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1104 int btrfs_write_tree_block(struct extent_buffer *buf)
1106 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1107 buf->start + buf->len - 1);
1110 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1112 return filemap_fdatawait_range(buf->pages[0]->mapping,
1113 buf->start, buf->start + buf->len - 1);
1116 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1117 u32 blocksize, u64 parent_transid)
1119 struct extent_buffer *buf = NULL;
1122 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1126 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1128 free_extent_buffer(buf);
1135 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1136 struct extent_buffer *buf)
1138 struct btrfs_fs_info *fs_info = root->fs_info;
1140 if (btrfs_header_generation(buf) ==
1141 fs_info->running_transaction->transid) {
1142 btrfs_assert_tree_locked(buf);
1144 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1145 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1147 fs_info->dirty_metadata_batch);
1148 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1149 btrfs_set_lock_blocking(buf);
1150 clear_extent_buffer_dirty(buf);
1155 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1156 u32 stripesize, struct btrfs_root *root,
1157 struct btrfs_fs_info *fs_info,
1161 root->commit_root = NULL;
1162 root->sectorsize = sectorsize;
1163 root->nodesize = nodesize;
1164 root->leafsize = leafsize;
1165 root->stripesize = stripesize;
1167 root->track_dirty = 0;
1169 root->orphan_item_inserted = 0;
1170 root->orphan_cleanup_state = 0;
1172 root->objectid = objectid;
1173 root->last_trans = 0;
1174 root->highest_objectid = 0;
1175 root->nr_delalloc_inodes = 0;
1176 root->nr_ordered_extents = 0;
1178 root->inode_tree = RB_ROOT;
1179 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1180 root->block_rsv = NULL;
1181 root->orphan_block_rsv = NULL;
1183 INIT_LIST_HEAD(&root->dirty_list);
1184 INIT_LIST_HEAD(&root->root_list);
1185 INIT_LIST_HEAD(&root->delalloc_inodes);
1186 INIT_LIST_HEAD(&root->delalloc_root);
1187 INIT_LIST_HEAD(&root->ordered_extents);
1188 INIT_LIST_HEAD(&root->ordered_root);
1189 INIT_LIST_HEAD(&root->logged_list[0]);
1190 INIT_LIST_HEAD(&root->logged_list[1]);
1191 spin_lock_init(&root->orphan_lock);
1192 spin_lock_init(&root->inode_lock);
1193 spin_lock_init(&root->delalloc_lock);
1194 spin_lock_init(&root->ordered_extent_lock);
1195 spin_lock_init(&root->accounting_lock);
1196 spin_lock_init(&root->log_extents_lock[0]);
1197 spin_lock_init(&root->log_extents_lock[1]);
1198 mutex_init(&root->objectid_mutex);
1199 mutex_init(&root->log_mutex);
1200 init_waitqueue_head(&root->log_writer_wait);
1201 init_waitqueue_head(&root->log_commit_wait[0]);
1202 init_waitqueue_head(&root->log_commit_wait[1]);
1203 atomic_set(&root->log_commit[0], 0);
1204 atomic_set(&root->log_commit[1], 0);
1205 atomic_set(&root->log_writers, 0);
1206 atomic_set(&root->log_batch, 0);
1207 atomic_set(&root->orphan_inodes, 0);
1208 atomic_set(&root->refs, 1);
1209 root->log_transid = 0;
1210 root->last_log_commit = 0;
1212 extent_io_tree_init(&root->dirty_log_pages,
1213 fs_info->btree_inode->i_mapping);
1215 memset(&root->root_key, 0, sizeof(root->root_key));
1216 memset(&root->root_item, 0, sizeof(root->root_item));
1217 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1218 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1220 root->defrag_trans_start = fs_info->generation;
1222 root->defrag_trans_start = 0;
1223 init_completion(&root->kobj_unregister);
1224 root->defrag_running = 0;
1225 root->root_key.objectid = objectid;
1228 spin_lock_init(&root->root_item_lock);
1231 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1233 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1235 root->fs_info = fs_info;
1239 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1240 /* Should only be used by the testing infrastructure */
1241 struct btrfs_root *btrfs_alloc_dummy_root(void)
1243 struct btrfs_root *root;
1245 root = btrfs_alloc_root(NULL);
1247 return ERR_PTR(-ENOMEM);
1248 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1249 root->dummy_root = 1;
1255 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1256 struct btrfs_fs_info *fs_info,
1259 struct extent_buffer *leaf;
1260 struct btrfs_root *tree_root = fs_info->tree_root;
1261 struct btrfs_root *root;
1262 struct btrfs_key key;
1266 root = btrfs_alloc_root(fs_info);
1268 return ERR_PTR(-ENOMEM);
1270 __setup_root(tree_root->nodesize, tree_root->leafsize,
1271 tree_root->sectorsize, tree_root->stripesize,
1272 root, fs_info, objectid);
1273 root->root_key.objectid = objectid;
1274 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1275 root->root_key.offset = 0;
1277 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1278 0, objectid, NULL, 0, 0, 0);
1280 ret = PTR_ERR(leaf);
1285 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1286 btrfs_set_header_bytenr(leaf, leaf->start);
1287 btrfs_set_header_generation(leaf, trans->transid);
1288 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1289 btrfs_set_header_owner(leaf, objectid);
1292 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1294 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1295 btrfs_header_chunk_tree_uuid(leaf),
1297 btrfs_mark_buffer_dirty(leaf);
1299 root->commit_root = btrfs_root_node(root);
1300 root->track_dirty = 1;
1303 root->root_item.flags = 0;
1304 root->root_item.byte_limit = 0;
1305 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1306 btrfs_set_root_generation(&root->root_item, trans->transid);
1307 btrfs_set_root_level(&root->root_item, 0);
1308 btrfs_set_root_refs(&root->root_item, 1);
1309 btrfs_set_root_used(&root->root_item, leaf->len);
1310 btrfs_set_root_last_snapshot(&root->root_item, 0);
1311 btrfs_set_root_dirid(&root->root_item, 0);
1313 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1314 root->root_item.drop_level = 0;
1316 key.objectid = objectid;
1317 key.type = BTRFS_ROOT_ITEM_KEY;
1319 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1323 btrfs_tree_unlock(leaf);
1329 btrfs_tree_unlock(leaf);
1330 free_extent_buffer(leaf);
1334 return ERR_PTR(ret);
1337 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1338 struct btrfs_fs_info *fs_info)
1340 struct btrfs_root *root;
1341 struct btrfs_root *tree_root = fs_info->tree_root;
1342 struct extent_buffer *leaf;
1344 root = btrfs_alloc_root(fs_info);
1346 return ERR_PTR(-ENOMEM);
1348 __setup_root(tree_root->nodesize, tree_root->leafsize,
1349 tree_root->sectorsize, tree_root->stripesize,
1350 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1352 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1353 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1354 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1356 * log trees do not get reference counted because they go away
1357 * before a real commit is actually done. They do store pointers
1358 * to file data extents, and those reference counts still get
1359 * updated (along with back refs to the log tree).
1363 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1364 BTRFS_TREE_LOG_OBJECTID, NULL,
1368 return ERR_CAST(leaf);
1371 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1372 btrfs_set_header_bytenr(leaf, leaf->start);
1373 btrfs_set_header_generation(leaf, trans->transid);
1374 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1375 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1378 write_extent_buffer(root->node, root->fs_info->fsid,
1379 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1380 btrfs_mark_buffer_dirty(root->node);
1381 btrfs_tree_unlock(root->node);
1385 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1386 struct btrfs_fs_info *fs_info)
1388 struct btrfs_root *log_root;
1390 log_root = alloc_log_tree(trans, fs_info);
1391 if (IS_ERR(log_root))
1392 return PTR_ERR(log_root);
1393 WARN_ON(fs_info->log_root_tree);
1394 fs_info->log_root_tree = log_root;
1398 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1399 struct btrfs_root *root)
1401 struct btrfs_root *log_root;
1402 struct btrfs_inode_item *inode_item;
1404 log_root = alloc_log_tree(trans, root->fs_info);
1405 if (IS_ERR(log_root))
1406 return PTR_ERR(log_root);
1408 log_root->last_trans = trans->transid;
1409 log_root->root_key.offset = root->root_key.objectid;
1411 inode_item = &log_root->root_item.inode;
1412 btrfs_set_stack_inode_generation(inode_item, 1);
1413 btrfs_set_stack_inode_size(inode_item, 3);
1414 btrfs_set_stack_inode_nlink(inode_item, 1);
1415 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1416 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1418 btrfs_set_root_node(&log_root->root_item, log_root->node);
1420 WARN_ON(root->log_root);
1421 root->log_root = log_root;
1422 root->log_transid = 0;
1423 root->last_log_commit = 0;
1427 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1428 struct btrfs_key *key)
1430 struct btrfs_root *root;
1431 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1432 struct btrfs_path *path;
1437 path = btrfs_alloc_path();
1439 return ERR_PTR(-ENOMEM);
1441 root = btrfs_alloc_root(fs_info);
1447 __setup_root(tree_root->nodesize, tree_root->leafsize,
1448 tree_root->sectorsize, tree_root->stripesize,
1449 root, fs_info, key->objectid);
1451 ret = btrfs_find_root(tree_root, key, path,
1452 &root->root_item, &root->root_key);
1459 generation = btrfs_root_generation(&root->root_item);
1460 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1461 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1462 blocksize, generation);
1466 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1470 root->commit_root = btrfs_root_node(root);
1472 btrfs_free_path(path);
1476 free_extent_buffer(root->node);
1480 root = ERR_PTR(ret);
1484 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1485 struct btrfs_key *location)
1487 struct btrfs_root *root;
1489 root = btrfs_read_tree_root(tree_root, location);
1493 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1495 btrfs_check_and_init_root_item(&root->root_item);
1501 int btrfs_init_fs_root(struct btrfs_root *root)
1505 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1506 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1508 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1513 btrfs_init_free_ino_ctl(root);
1514 mutex_init(&root->fs_commit_mutex);
1515 spin_lock_init(&root->cache_lock);
1516 init_waitqueue_head(&root->cache_wait);
1518 ret = get_anon_bdev(&root->anon_dev);
1523 kfree(root->free_ino_ctl);
1524 kfree(root->free_ino_pinned);
1528 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1531 struct btrfs_root *root;
1533 spin_lock(&fs_info->fs_roots_radix_lock);
1534 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1535 (unsigned long)root_id);
1536 spin_unlock(&fs_info->fs_roots_radix_lock);
1540 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1541 struct btrfs_root *root)
1545 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1549 spin_lock(&fs_info->fs_roots_radix_lock);
1550 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1551 (unsigned long)root->root_key.objectid,
1555 spin_unlock(&fs_info->fs_roots_radix_lock);
1556 radix_tree_preload_end();
1561 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1562 struct btrfs_key *location,
1565 struct btrfs_root *root;
1568 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1569 return fs_info->tree_root;
1570 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1571 return fs_info->extent_root;
1572 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1573 return fs_info->chunk_root;
1574 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1575 return fs_info->dev_root;
1576 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1577 return fs_info->csum_root;
1578 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1579 return fs_info->quota_root ? fs_info->quota_root :
1581 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1582 return fs_info->uuid_root ? fs_info->uuid_root :
1585 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1587 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1588 return ERR_PTR(-ENOENT);
1592 root = btrfs_read_fs_root(fs_info->tree_root, location);
1596 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1601 ret = btrfs_init_fs_root(root);
1605 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1606 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1610 root->orphan_item_inserted = 1;
1612 ret = btrfs_insert_fs_root(fs_info, root);
1614 if (ret == -EEXIST) {
1623 return ERR_PTR(ret);
1626 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1628 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1630 struct btrfs_device *device;
1631 struct backing_dev_info *bdi;
1634 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1637 bdi = blk_get_backing_dev_info(device->bdev);
1638 if (bdi && bdi_congested(bdi, bdi_bits)) {
1648 * If this fails, caller must call bdi_destroy() to get rid of the
1651 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1655 bdi->capabilities = BDI_CAP_MAP_COPY;
1656 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1660 bdi->ra_pages = default_backing_dev_info.ra_pages;
1661 bdi->congested_fn = btrfs_congested_fn;
1662 bdi->congested_data = info;
1667 * called by the kthread helper functions to finally call the bio end_io
1668 * functions. This is where read checksum verification actually happens
1670 static void end_workqueue_fn(struct btrfs_work *work)
1673 struct end_io_wq *end_io_wq;
1676 end_io_wq = container_of(work, struct end_io_wq, work);
1677 bio = end_io_wq->bio;
1679 error = end_io_wq->error;
1680 bio->bi_private = end_io_wq->private;
1681 bio->bi_end_io = end_io_wq->end_io;
1683 bio_endio(bio, error);
1686 static int cleaner_kthread(void *arg)
1688 struct btrfs_root *root = arg;
1694 /* Make the cleaner go to sleep early. */
1695 if (btrfs_need_cleaner_sleep(root))
1698 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1702 * Avoid the problem that we change the status of the fs
1703 * during the above check and trylock.
1705 if (btrfs_need_cleaner_sleep(root)) {
1706 mutex_unlock(&root->fs_info->cleaner_mutex);
1710 btrfs_run_delayed_iputs(root);
1711 again = btrfs_clean_one_deleted_snapshot(root);
1712 mutex_unlock(&root->fs_info->cleaner_mutex);
1715 * The defragger has dealt with the R/O remount and umount,
1716 * needn't do anything special here.
1718 btrfs_run_defrag_inodes(root->fs_info);
1720 if (!try_to_freeze() && !again) {
1721 set_current_state(TASK_INTERRUPTIBLE);
1722 if (!kthread_should_stop())
1724 __set_current_state(TASK_RUNNING);
1726 } while (!kthread_should_stop());
1730 static int transaction_kthread(void *arg)
1732 struct btrfs_root *root = arg;
1733 struct btrfs_trans_handle *trans;
1734 struct btrfs_transaction *cur;
1737 unsigned long delay;
1741 cannot_commit = false;
1742 delay = HZ * root->fs_info->commit_interval;
1743 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1745 spin_lock(&root->fs_info->trans_lock);
1746 cur = root->fs_info->running_transaction;
1748 spin_unlock(&root->fs_info->trans_lock);
1752 now = get_seconds();
1753 if (cur->state < TRANS_STATE_BLOCKED &&
1754 (now < cur->start_time ||
1755 now - cur->start_time < root->fs_info->commit_interval)) {
1756 spin_unlock(&root->fs_info->trans_lock);
1760 transid = cur->transid;
1761 spin_unlock(&root->fs_info->trans_lock);
1763 /* If the file system is aborted, this will always fail. */
1764 trans = btrfs_attach_transaction(root);
1765 if (IS_ERR(trans)) {
1766 if (PTR_ERR(trans) != -ENOENT)
1767 cannot_commit = true;
1770 if (transid == trans->transid) {
1771 btrfs_commit_transaction(trans, root);
1773 btrfs_end_transaction(trans, root);
1776 wake_up_process(root->fs_info->cleaner_kthread);
1777 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1779 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1780 &root->fs_info->fs_state)))
1781 btrfs_cleanup_transaction(root);
1782 if (!try_to_freeze()) {
1783 set_current_state(TASK_INTERRUPTIBLE);
1784 if (!kthread_should_stop() &&
1785 (!btrfs_transaction_blocked(root->fs_info) ||
1787 schedule_timeout(delay);
1788 __set_current_state(TASK_RUNNING);
1790 } while (!kthread_should_stop());
1795 * this will find the highest generation in the array of
1796 * root backups. The index of the highest array is returned,
1797 * or -1 if we can't find anything.
1799 * We check to make sure the array is valid by comparing the
1800 * generation of the latest root in the array with the generation
1801 * in the super block. If they don't match we pitch it.
1803 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1806 int newest_index = -1;
1807 struct btrfs_root_backup *root_backup;
1810 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1811 root_backup = info->super_copy->super_roots + i;
1812 cur = btrfs_backup_tree_root_gen(root_backup);
1813 if (cur == newest_gen)
1817 /* check to see if we actually wrapped around */
1818 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1819 root_backup = info->super_copy->super_roots;
1820 cur = btrfs_backup_tree_root_gen(root_backup);
1821 if (cur == newest_gen)
1824 return newest_index;
1829 * find the oldest backup so we know where to store new entries
1830 * in the backup array. This will set the backup_root_index
1831 * field in the fs_info struct
1833 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1836 int newest_index = -1;
1838 newest_index = find_newest_super_backup(info, newest_gen);
1839 /* if there was garbage in there, just move along */
1840 if (newest_index == -1) {
1841 info->backup_root_index = 0;
1843 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1848 * copy all the root pointers into the super backup array.
1849 * this will bump the backup pointer by one when it is
1852 static void backup_super_roots(struct btrfs_fs_info *info)
1855 struct btrfs_root_backup *root_backup;
1858 next_backup = info->backup_root_index;
1859 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1860 BTRFS_NUM_BACKUP_ROOTS;
1863 * just overwrite the last backup if we're at the same generation
1864 * this happens only at umount
1866 root_backup = info->super_for_commit->super_roots + last_backup;
1867 if (btrfs_backup_tree_root_gen(root_backup) ==
1868 btrfs_header_generation(info->tree_root->node))
1869 next_backup = last_backup;
1871 root_backup = info->super_for_commit->super_roots + next_backup;
1874 * make sure all of our padding and empty slots get zero filled
1875 * regardless of which ones we use today
1877 memset(root_backup, 0, sizeof(*root_backup));
1879 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1881 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1882 btrfs_set_backup_tree_root_gen(root_backup,
1883 btrfs_header_generation(info->tree_root->node));
1885 btrfs_set_backup_tree_root_level(root_backup,
1886 btrfs_header_level(info->tree_root->node));
1888 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1889 btrfs_set_backup_chunk_root_gen(root_backup,
1890 btrfs_header_generation(info->chunk_root->node));
1891 btrfs_set_backup_chunk_root_level(root_backup,
1892 btrfs_header_level(info->chunk_root->node));
1894 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1895 btrfs_set_backup_extent_root_gen(root_backup,
1896 btrfs_header_generation(info->extent_root->node));
1897 btrfs_set_backup_extent_root_level(root_backup,
1898 btrfs_header_level(info->extent_root->node));
1901 * we might commit during log recovery, which happens before we set
1902 * the fs_root. Make sure it is valid before we fill it in.
1904 if (info->fs_root && info->fs_root->node) {
1905 btrfs_set_backup_fs_root(root_backup,
1906 info->fs_root->node->start);
1907 btrfs_set_backup_fs_root_gen(root_backup,
1908 btrfs_header_generation(info->fs_root->node));
1909 btrfs_set_backup_fs_root_level(root_backup,
1910 btrfs_header_level(info->fs_root->node));
1913 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1914 btrfs_set_backup_dev_root_gen(root_backup,
1915 btrfs_header_generation(info->dev_root->node));
1916 btrfs_set_backup_dev_root_level(root_backup,
1917 btrfs_header_level(info->dev_root->node));
1919 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1920 btrfs_set_backup_csum_root_gen(root_backup,
1921 btrfs_header_generation(info->csum_root->node));
1922 btrfs_set_backup_csum_root_level(root_backup,
1923 btrfs_header_level(info->csum_root->node));
1925 btrfs_set_backup_total_bytes(root_backup,
1926 btrfs_super_total_bytes(info->super_copy));
1927 btrfs_set_backup_bytes_used(root_backup,
1928 btrfs_super_bytes_used(info->super_copy));
1929 btrfs_set_backup_num_devices(root_backup,
1930 btrfs_super_num_devices(info->super_copy));
1933 * if we don't copy this out to the super_copy, it won't get remembered
1934 * for the next commit
1936 memcpy(&info->super_copy->super_roots,
1937 &info->super_for_commit->super_roots,
1938 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1942 * this copies info out of the root backup array and back into
1943 * the in-memory super block. It is meant to help iterate through
1944 * the array, so you send it the number of backups you've already
1945 * tried and the last backup index you used.
1947 * this returns -1 when it has tried all the backups
1949 static noinline int next_root_backup(struct btrfs_fs_info *info,
1950 struct btrfs_super_block *super,
1951 int *num_backups_tried, int *backup_index)
1953 struct btrfs_root_backup *root_backup;
1954 int newest = *backup_index;
1956 if (*num_backups_tried == 0) {
1957 u64 gen = btrfs_super_generation(super);
1959 newest = find_newest_super_backup(info, gen);
1963 *backup_index = newest;
1964 *num_backups_tried = 1;
1965 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1966 /* we've tried all the backups, all done */
1969 /* jump to the next oldest backup */
1970 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1971 BTRFS_NUM_BACKUP_ROOTS;
1972 *backup_index = newest;
1973 *num_backups_tried += 1;
1975 root_backup = super->super_roots + newest;
1977 btrfs_set_super_generation(super,
1978 btrfs_backup_tree_root_gen(root_backup));
1979 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1980 btrfs_set_super_root_level(super,
1981 btrfs_backup_tree_root_level(root_backup));
1982 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1985 * fixme: the total bytes and num_devices need to match or we should
1988 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1989 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1993 /* helper to cleanup workers */
1994 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1996 btrfs_stop_workers(&fs_info->generic_worker);
1997 btrfs_stop_workers(&fs_info->fixup_workers);
1998 btrfs_stop_workers(&fs_info->delalloc_workers);
1999 btrfs_stop_workers(&fs_info->workers);
2000 btrfs_stop_workers(&fs_info->endio_workers);
2001 btrfs_stop_workers(&fs_info->endio_meta_workers);
2002 btrfs_stop_workers(&fs_info->endio_raid56_workers);
2003 btrfs_stop_workers(&fs_info->rmw_workers);
2004 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2005 btrfs_stop_workers(&fs_info->endio_write_workers);
2006 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2007 btrfs_stop_workers(&fs_info->submit_workers);
2008 btrfs_stop_workers(&fs_info->delayed_workers);
2009 btrfs_stop_workers(&fs_info->caching_workers);
2010 btrfs_stop_workers(&fs_info->readahead_workers);
2011 btrfs_stop_workers(&fs_info->flush_workers);
2012 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
2015 static void free_root_extent_buffers(struct btrfs_root *root)
2018 free_extent_buffer(root->node);
2019 free_extent_buffer(root->commit_root);
2021 root->commit_root = NULL;
2025 /* helper to cleanup tree roots */
2026 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2028 free_root_extent_buffers(info->tree_root);
2030 free_root_extent_buffers(info->dev_root);
2031 free_root_extent_buffers(info->extent_root);
2032 free_root_extent_buffers(info->csum_root);
2033 free_root_extent_buffers(info->quota_root);
2034 free_root_extent_buffers(info->uuid_root);
2036 free_root_extent_buffers(info->chunk_root);
2039 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2042 struct btrfs_root *gang[8];
2045 while (!list_empty(&fs_info->dead_roots)) {
2046 gang[0] = list_entry(fs_info->dead_roots.next,
2047 struct btrfs_root, root_list);
2048 list_del(&gang[0]->root_list);
2050 if (gang[0]->in_radix) {
2051 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2053 free_extent_buffer(gang[0]->node);
2054 free_extent_buffer(gang[0]->commit_root);
2055 btrfs_put_fs_root(gang[0]);
2060 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2065 for (i = 0; i < ret; i++)
2066 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2069 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2070 btrfs_free_log_root_tree(NULL, fs_info);
2071 btrfs_destroy_pinned_extent(fs_info->tree_root,
2072 fs_info->pinned_extents);
2076 int open_ctree(struct super_block *sb,
2077 struct btrfs_fs_devices *fs_devices,
2087 struct btrfs_key location;
2088 struct buffer_head *bh;
2089 struct btrfs_super_block *disk_super;
2090 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2091 struct btrfs_root *tree_root;
2092 struct btrfs_root *extent_root;
2093 struct btrfs_root *csum_root;
2094 struct btrfs_root *chunk_root;
2095 struct btrfs_root *dev_root;
2096 struct btrfs_root *quota_root;
2097 struct btrfs_root *uuid_root;
2098 struct btrfs_root *log_tree_root;
2101 int num_backups_tried = 0;
2102 int backup_index = 0;
2103 bool create_uuid_tree;
2104 bool check_uuid_tree;
2106 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2107 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2108 if (!tree_root || !chunk_root) {
2113 ret = init_srcu_struct(&fs_info->subvol_srcu);
2119 ret = setup_bdi(fs_info, &fs_info->bdi);
2125 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2130 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2131 (1 + ilog2(nr_cpu_ids));
2133 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2136 goto fail_dirty_metadata_bytes;
2139 ret = percpu_counter_init(&fs_info->bio_counter, 0);
2142 goto fail_delalloc_bytes;
2145 fs_info->btree_inode = new_inode(sb);
2146 if (!fs_info->btree_inode) {
2148 goto fail_bio_counter;
2151 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2153 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2154 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2155 INIT_LIST_HEAD(&fs_info->trans_list);
2156 INIT_LIST_HEAD(&fs_info->dead_roots);
2157 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2158 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2159 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2160 spin_lock_init(&fs_info->delalloc_root_lock);
2161 spin_lock_init(&fs_info->trans_lock);
2162 spin_lock_init(&fs_info->fs_roots_radix_lock);
2163 spin_lock_init(&fs_info->delayed_iput_lock);
2164 spin_lock_init(&fs_info->defrag_inodes_lock);
2165 spin_lock_init(&fs_info->free_chunk_lock);
2166 spin_lock_init(&fs_info->tree_mod_seq_lock);
2167 spin_lock_init(&fs_info->super_lock);
2168 spin_lock_init(&fs_info->buffer_lock);
2169 rwlock_init(&fs_info->tree_mod_log_lock);
2170 mutex_init(&fs_info->reloc_mutex);
2171 seqlock_init(&fs_info->profiles_lock);
2173 init_completion(&fs_info->kobj_unregister);
2174 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2175 INIT_LIST_HEAD(&fs_info->space_info);
2176 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2177 btrfs_mapping_init(&fs_info->mapping_tree);
2178 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2179 BTRFS_BLOCK_RSV_GLOBAL);
2180 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2181 BTRFS_BLOCK_RSV_DELALLOC);
2182 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2183 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2184 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2185 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2186 BTRFS_BLOCK_RSV_DELOPS);
2187 atomic_set(&fs_info->nr_async_submits, 0);
2188 atomic_set(&fs_info->async_delalloc_pages, 0);
2189 atomic_set(&fs_info->async_submit_draining, 0);
2190 atomic_set(&fs_info->nr_async_bios, 0);
2191 atomic_set(&fs_info->defrag_running, 0);
2192 atomic64_set(&fs_info->tree_mod_seq, 0);
2194 fs_info->max_inline = 8192 * 1024;
2195 fs_info->metadata_ratio = 0;
2196 fs_info->defrag_inodes = RB_ROOT;
2197 fs_info->free_chunk_space = 0;
2198 fs_info->tree_mod_log = RB_ROOT;
2199 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2200 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2201 /* readahead state */
2202 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2203 spin_lock_init(&fs_info->reada_lock);
2205 fs_info->thread_pool_size = min_t(unsigned long,
2206 num_online_cpus() + 2, 8);
2208 INIT_LIST_HEAD(&fs_info->ordered_roots);
2209 spin_lock_init(&fs_info->ordered_root_lock);
2210 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2212 if (!fs_info->delayed_root) {
2216 btrfs_init_delayed_root(fs_info->delayed_root);
2218 mutex_init(&fs_info->scrub_lock);
2219 atomic_set(&fs_info->scrubs_running, 0);
2220 atomic_set(&fs_info->scrub_pause_req, 0);
2221 atomic_set(&fs_info->scrubs_paused, 0);
2222 atomic_set(&fs_info->scrub_cancel_req, 0);
2223 init_waitqueue_head(&fs_info->replace_wait);
2224 init_waitqueue_head(&fs_info->scrub_pause_wait);
2225 fs_info->scrub_workers_refcnt = 0;
2226 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2227 fs_info->check_integrity_print_mask = 0;
2230 spin_lock_init(&fs_info->balance_lock);
2231 mutex_init(&fs_info->balance_mutex);
2232 atomic_set(&fs_info->balance_running, 0);
2233 atomic_set(&fs_info->balance_pause_req, 0);
2234 atomic_set(&fs_info->balance_cancel_req, 0);
2235 fs_info->balance_ctl = NULL;
2236 init_waitqueue_head(&fs_info->balance_wait_q);
2238 sb->s_blocksize = 4096;
2239 sb->s_blocksize_bits = blksize_bits(4096);
2240 sb->s_bdi = &fs_info->bdi;
2242 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2243 set_nlink(fs_info->btree_inode, 1);
2245 * we set the i_size on the btree inode to the max possible int.
2246 * the real end of the address space is determined by all of
2247 * the devices in the system
2249 fs_info->btree_inode->i_size = OFFSET_MAX;
2250 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2251 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2253 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2254 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2255 fs_info->btree_inode->i_mapping);
2256 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2257 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2259 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2261 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2262 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2263 sizeof(struct btrfs_key));
2264 set_bit(BTRFS_INODE_DUMMY,
2265 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2266 btrfs_insert_inode_hash(fs_info->btree_inode);
2268 spin_lock_init(&fs_info->block_group_cache_lock);
2269 fs_info->block_group_cache_tree = RB_ROOT;
2270 fs_info->first_logical_byte = (u64)-1;
2272 extent_io_tree_init(&fs_info->freed_extents[0],
2273 fs_info->btree_inode->i_mapping);
2274 extent_io_tree_init(&fs_info->freed_extents[1],
2275 fs_info->btree_inode->i_mapping);
2276 fs_info->pinned_extents = &fs_info->freed_extents[0];
2277 fs_info->do_barriers = 1;
2280 mutex_init(&fs_info->ordered_operations_mutex);
2281 mutex_init(&fs_info->ordered_extent_flush_mutex);
2282 mutex_init(&fs_info->tree_log_mutex);
2283 mutex_init(&fs_info->chunk_mutex);
2284 mutex_init(&fs_info->transaction_kthread_mutex);
2285 mutex_init(&fs_info->cleaner_mutex);
2286 mutex_init(&fs_info->volume_mutex);
2287 init_rwsem(&fs_info->extent_commit_sem);
2288 init_rwsem(&fs_info->cleanup_work_sem);
2289 init_rwsem(&fs_info->subvol_sem);
2290 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2291 fs_info->dev_replace.lock_owner = 0;
2292 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2293 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2294 mutex_init(&fs_info->dev_replace.lock_management_lock);
2295 mutex_init(&fs_info->dev_replace.lock);
2297 spin_lock_init(&fs_info->qgroup_lock);
2298 mutex_init(&fs_info->qgroup_ioctl_lock);
2299 fs_info->qgroup_tree = RB_ROOT;
2300 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2301 fs_info->qgroup_seq = 1;
2302 fs_info->quota_enabled = 0;
2303 fs_info->pending_quota_state = 0;
2304 fs_info->qgroup_ulist = NULL;
2305 mutex_init(&fs_info->qgroup_rescan_lock);
2307 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2308 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2310 init_waitqueue_head(&fs_info->transaction_throttle);
2311 init_waitqueue_head(&fs_info->transaction_wait);
2312 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2313 init_waitqueue_head(&fs_info->async_submit_wait);
2315 ret = btrfs_alloc_stripe_hash_table(fs_info);
2321 __setup_root(4096, 4096, 4096, 4096, tree_root,
2322 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2324 invalidate_bdev(fs_devices->latest_bdev);
2327 * Read super block and check the signature bytes only
2329 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2336 * We want to check superblock checksum, the type is stored inside.
2337 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2339 if (btrfs_check_super_csum(bh->b_data)) {
2340 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2346 * super_copy is zeroed at allocation time and we never touch the
2347 * following bytes up to INFO_SIZE, the checksum is calculated from
2348 * the whole block of INFO_SIZE
2350 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2351 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2352 sizeof(*fs_info->super_for_commit));
2355 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2357 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2359 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2364 disk_super = fs_info->super_copy;
2365 if (!btrfs_super_root(disk_super))
2368 /* check FS state, whether FS is broken. */
2369 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2370 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2373 * run through our array of backup supers and setup
2374 * our ring pointer to the oldest one
2376 generation = btrfs_super_generation(disk_super);
2377 find_oldest_super_backup(fs_info, generation);
2380 * In the long term, we'll store the compression type in the super
2381 * block, and it'll be used for per file compression control.
2383 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2385 ret = btrfs_parse_options(tree_root, options);
2391 features = btrfs_super_incompat_flags(disk_super) &
2392 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2394 printk(KERN_ERR "BTRFS: couldn't mount because of "
2395 "unsupported optional features (%Lx).\n",
2401 if (btrfs_super_leafsize(disk_super) !=
2402 btrfs_super_nodesize(disk_super)) {
2403 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2404 "blocksizes don't match. node %d leaf %d\n",
2405 btrfs_super_nodesize(disk_super),
2406 btrfs_super_leafsize(disk_super));
2410 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2411 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2412 "blocksize (%d) was too large\n",
2413 btrfs_super_leafsize(disk_super));
2418 features = btrfs_super_incompat_flags(disk_super);
2419 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2420 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2421 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2423 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2424 printk(KERN_ERR "BTRFS: has skinny extents\n");
2427 * flag our filesystem as having big metadata blocks if
2428 * they are bigger than the page size
2430 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2431 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2432 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2433 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2436 nodesize = btrfs_super_nodesize(disk_super);
2437 leafsize = btrfs_super_leafsize(disk_super);
2438 sectorsize = btrfs_super_sectorsize(disk_super);
2439 stripesize = btrfs_super_stripesize(disk_super);
2440 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2441 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2444 * mixed block groups end up with duplicate but slightly offset
2445 * extent buffers for the same range. It leads to corruptions
2447 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2448 (sectorsize != leafsize)) {
2449 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2450 "are not allowed for mixed block groups on %s\n",
2456 * Needn't use the lock because there is no other task which will
2459 btrfs_set_super_incompat_flags(disk_super, features);
2461 features = btrfs_super_compat_ro_flags(disk_super) &
2462 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2463 if (!(sb->s_flags & MS_RDONLY) && features) {
2464 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2465 "unsupported option features (%Lx).\n",
2471 btrfs_init_workers(&fs_info->generic_worker,
2472 "genwork", 1, NULL);
2474 btrfs_init_workers(&fs_info->workers, "worker",
2475 fs_info->thread_pool_size,
2476 &fs_info->generic_worker);
2478 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2479 fs_info->thread_pool_size, NULL);
2481 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2482 fs_info->thread_pool_size, NULL);
2484 btrfs_init_workers(&fs_info->submit_workers, "submit",
2485 min_t(u64, fs_devices->num_devices,
2486 fs_info->thread_pool_size), NULL);
2488 btrfs_init_workers(&fs_info->caching_workers, "cache",
2489 fs_info->thread_pool_size, NULL);
2491 /* a higher idle thresh on the submit workers makes it much more
2492 * likely that bios will be send down in a sane order to the
2495 fs_info->submit_workers.idle_thresh = 64;
2497 fs_info->workers.idle_thresh = 16;
2498 fs_info->workers.ordered = 1;
2500 fs_info->delalloc_workers.idle_thresh = 2;
2501 fs_info->delalloc_workers.ordered = 1;
2503 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2504 &fs_info->generic_worker);
2505 btrfs_init_workers(&fs_info->endio_workers, "endio",
2506 fs_info->thread_pool_size,
2507 &fs_info->generic_worker);
2508 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2509 fs_info->thread_pool_size,
2510 &fs_info->generic_worker);
2511 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2512 "endio-meta-write", fs_info->thread_pool_size,
2513 &fs_info->generic_worker);
2514 btrfs_init_workers(&fs_info->endio_raid56_workers,
2515 "endio-raid56", fs_info->thread_pool_size,
2516 &fs_info->generic_worker);
2517 btrfs_init_workers(&fs_info->rmw_workers,
2518 "rmw", fs_info->thread_pool_size,
2519 &fs_info->generic_worker);
2520 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2521 fs_info->thread_pool_size,
2522 &fs_info->generic_worker);
2523 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2524 1, &fs_info->generic_worker);
2525 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2526 fs_info->thread_pool_size,
2527 &fs_info->generic_worker);
2528 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2529 fs_info->thread_pool_size,
2530 &fs_info->generic_worker);
2531 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2532 &fs_info->generic_worker);
2535 * endios are largely parallel and should have a very
2538 fs_info->endio_workers.idle_thresh = 4;
2539 fs_info->endio_meta_workers.idle_thresh = 4;
2540 fs_info->endio_raid56_workers.idle_thresh = 4;
2541 fs_info->rmw_workers.idle_thresh = 2;
2543 fs_info->endio_write_workers.idle_thresh = 2;
2544 fs_info->endio_meta_write_workers.idle_thresh = 2;
2545 fs_info->readahead_workers.idle_thresh = 2;
2548 * btrfs_start_workers can really only fail because of ENOMEM so just
2549 * return -ENOMEM if any of these fail.
2551 ret = btrfs_start_workers(&fs_info->workers);
2552 ret |= btrfs_start_workers(&fs_info->generic_worker);
2553 ret |= btrfs_start_workers(&fs_info->submit_workers);
2554 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2555 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2556 ret |= btrfs_start_workers(&fs_info->endio_workers);
2557 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2558 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2559 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2560 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2561 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2562 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2563 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2564 ret |= btrfs_start_workers(&fs_info->caching_workers);
2565 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2566 ret |= btrfs_start_workers(&fs_info->flush_workers);
2567 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2570 goto fail_sb_buffer;
2573 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2574 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2575 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2577 tree_root->nodesize = nodesize;
2578 tree_root->leafsize = leafsize;
2579 tree_root->sectorsize = sectorsize;
2580 tree_root->stripesize = stripesize;
2582 sb->s_blocksize = sectorsize;
2583 sb->s_blocksize_bits = blksize_bits(sectorsize);
2585 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2586 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2587 goto fail_sb_buffer;
2590 if (sectorsize != PAGE_SIZE) {
2591 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2592 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2593 goto fail_sb_buffer;
2596 mutex_lock(&fs_info->chunk_mutex);
2597 ret = btrfs_read_sys_array(tree_root);
2598 mutex_unlock(&fs_info->chunk_mutex);
2600 printk(KERN_WARNING "BTRFS: failed to read the system "
2601 "array on %s\n", sb->s_id);
2602 goto fail_sb_buffer;
2605 blocksize = btrfs_level_size(tree_root,
2606 btrfs_super_chunk_root_level(disk_super));
2607 generation = btrfs_super_chunk_root_generation(disk_super);
2609 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2610 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2612 chunk_root->node = read_tree_block(chunk_root,
2613 btrfs_super_chunk_root(disk_super),
2614 blocksize, generation);
2615 if (!chunk_root->node ||
2616 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2617 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2619 goto fail_tree_roots;
2621 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2622 chunk_root->commit_root = btrfs_root_node(chunk_root);
2624 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2625 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2627 ret = btrfs_read_chunk_tree(chunk_root);
2629 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2631 goto fail_tree_roots;
2635 * keep the device that is marked to be the target device for the
2636 * dev_replace procedure
2638 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2640 if (!fs_devices->latest_bdev) {
2641 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2643 goto fail_tree_roots;
2647 blocksize = btrfs_level_size(tree_root,
2648 btrfs_super_root_level(disk_super));
2649 generation = btrfs_super_generation(disk_super);
2651 tree_root->node = read_tree_block(tree_root,
2652 btrfs_super_root(disk_super),
2653 blocksize, generation);
2654 if (!tree_root->node ||
2655 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2656 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2659 goto recovery_tree_root;
2662 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2663 tree_root->commit_root = btrfs_root_node(tree_root);
2664 btrfs_set_root_refs(&tree_root->root_item, 1);
2666 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2667 location.type = BTRFS_ROOT_ITEM_KEY;
2668 location.offset = 0;
2670 extent_root = btrfs_read_tree_root(tree_root, &location);
2671 if (IS_ERR(extent_root)) {
2672 ret = PTR_ERR(extent_root);
2673 goto recovery_tree_root;
2675 extent_root->track_dirty = 1;
2676 fs_info->extent_root = extent_root;
2678 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2679 dev_root = btrfs_read_tree_root(tree_root, &location);
2680 if (IS_ERR(dev_root)) {
2681 ret = PTR_ERR(dev_root);
2682 goto recovery_tree_root;
2684 dev_root->track_dirty = 1;
2685 fs_info->dev_root = dev_root;
2686 btrfs_init_devices_late(fs_info);
2688 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2689 csum_root = btrfs_read_tree_root(tree_root, &location);
2690 if (IS_ERR(csum_root)) {
2691 ret = PTR_ERR(csum_root);
2692 goto recovery_tree_root;
2694 csum_root->track_dirty = 1;
2695 fs_info->csum_root = csum_root;
2697 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2698 quota_root = btrfs_read_tree_root(tree_root, &location);
2699 if (!IS_ERR(quota_root)) {
2700 quota_root->track_dirty = 1;
2701 fs_info->quota_enabled = 1;
2702 fs_info->pending_quota_state = 1;
2703 fs_info->quota_root = quota_root;
2706 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2707 uuid_root = btrfs_read_tree_root(tree_root, &location);
2708 if (IS_ERR(uuid_root)) {
2709 ret = PTR_ERR(uuid_root);
2711 goto recovery_tree_root;
2712 create_uuid_tree = true;
2713 check_uuid_tree = false;
2715 uuid_root->track_dirty = 1;
2716 fs_info->uuid_root = uuid_root;
2717 create_uuid_tree = false;
2719 generation != btrfs_super_uuid_tree_generation(disk_super);
2722 fs_info->generation = generation;
2723 fs_info->last_trans_committed = generation;
2725 ret = btrfs_recover_balance(fs_info);
2727 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2728 goto fail_block_groups;
2731 ret = btrfs_init_dev_stats(fs_info);
2733 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2735 goto fail_block_groups;
2738 ret = btrfs_init_dev_replace(fs_info);
2740 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2741 goto fail_block_groups;
2744 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2746 ret = btrfs_sysfs_add_one(fs_info);
2748 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2749 goto fail_block_groups;
2752 ret = btrfs_init_space_info(fs_info);
2754 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2758 ret = btrfs_read_block_groups(extent_root);
2760 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2763 fs_info->num_tolerated_disk_barrier_failures =
2764 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2765 if (fs_info->fs_devices->missing_devices >
2766 fs_info->num_tolerated_disk_barrier_failures &&
2767 !(sb->s_flags & MS_RDONLY)) {
2768 printk(KERN_WARNING "BTRFS: "
2769 "too many missing devices, writeable mount is not allowed\n");
2773 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2775 if (IS_ERR(fs_info->cleaner_kthread))
2778 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2780 "btrfs-transaction");
2781 if (IS_ERR(fs_info->transaction_kthread))
2784 if (!btrfs_test_opt(tree_root, SSD) &&
2785 !btrfs_test_opt(tree_root, NOSSD) &&
2786 !fs_info->fs_devices->rotating) {
2787 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2789 btrfs_set_opt(fs_info->mount_opt, SSD);
2792 /* Set the real inode map cache flag */
2793 if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2794 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2796 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2797 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2798 ret = btrfsic_mount(tree_root, fs_devices,
2799 btrfs_test_opt(tree_root,
2800 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2802 fs_info->check_integrity_print_mask);
2804 printk(KERN_WARNING "BTRFS: failed to initialize"
2805 " integrity check module %s\n", sb->s_id);
2808 ret = btrfs_read_qgroup_config(fs_info);
2810 goto fail_trans_kthread;
2812 /* do not make disk changes in broken FS */
2813 if (btrfs_super_log_root(disk_super) != 0) {
2814 u64 bytenr = btrfs_super_log_root(disk_super);
2816 if (fs_devices->rw_devices == 0) {
2817 printk(KERN_WARNING "BTRFS: log replay required "
2823 btrfs_level_size(tree_root,
2824 btrfs_super_log_root_level(disk_super));
2826 log_tree_root = btrfs_alloc_root(fs_info);
2827 if (!log_tree_root) {
2832 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2833 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2835 log_tree_root->node = read_tree_block(tree_root, bytenr,
2838 if (!log_tree_root->node ||
2839 !extent_buffer_uptodate(log_tree_root->node)) {
2840 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2841 free_extent_buffer(log_tree_root->node);
2842 kfree(log_tree_root);
2843 goto fail_trans_kthread;
2845 /* returns with log_tree_root freed on success */
2846 ret = btrfs_recover_log_trees(log_tree_root);
2848 btrfs_error(tree_root->fs_info, ret,
2849 "Failed to recover log tree");
2850 free_extent_buffer(log_tree_root->node);
2851 kfree(log_tree_root);
2852 goto fail_trans_kthread;
2855 if (sb->s_flags & MS_RDONLY) {
2856 ret = btrfs_commit_super(tree_root);
2858 goto fail_trans_kthread;
2862 ret = btrfs_find_orphan_roots(tree_root);
2864 goto fail_trans_kthread;
2866 if (!(sb->s_flags & MS_RDONLY)) {
2867 ret = btrfs_cleanup_fs_roots(fs_info);
2869 goto fail_trans_kthread;
2871 ret = btrfs_recover_relocation(tree_root);
2874 "BTRFS: failed to recover relocation\n");
2880 location.objectid = BTRFS_FS_TREE_OBJECTID;
2881 location.type = BTRFS_ROOT_ITEM_KEY;
2882 location.offset = 0;
2884 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2885 if (IS_ERR(fs_info->fs_root)) {
2886 err = PTR_ERR(fs_info->fs_root);
2890 if (sb->s_flags & MS_RDONLY)
2893 down_read(&fs_info->cleanup_work_sem);
2894 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2895 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2896 up_read(&fs_info->cleanup_work_sem);
2897 close_ctree(tree_root);
2900 up_read(&fs_info->cleanup_work_sem);
2902 ret = btrfs_resume_balance_async(fs_info);
2904 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2905 close_ctree(tree_root);
2909 ret = btrfs_resume_dev_replace_async(fs_info);
2911 pr_warn("BTRFS: failed to resume dev_replace\n");
2912 close_ctree(tree_root);
2916 btrfs_qgroup_rescan_resume(fs_info);
2918 if (create_uuid_tree) {
2919 pr_info("BTRFS: creating UUID tree\n");
2920 ret = btrfs_create_uuid_tree(fs_info);
2922 pr_warn("BTRFS: failed to create the UUID tree %d\n",
2924 close_ctree(tree_root);
2927 } else if (check_uuid_tree ||
2928 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2929 pr_info("BTRFS: checking UUID tree\n");
2930 ret = btrfs_check_uuid_tree(fs_info);
2932 pr_warn("BTRFS: failed to check the UUID tree %d\n",
2934 close_ctree(tree_root);
2938 fs_info->update_uuid_tree_gen = 1;
2944 btrfs_free_qgroup_config(fs_info);
2946 kthread_stop(fs_info->transaction_kthread);
2947 btrfs_cleanup_transaction(fs_info->tree_root);
2948 del_fs_roots(fs_info);
2950 kthread_stop(fs_info->cleaner_kthread);
2953 * make sure we're done with the btree inode before we stop our
2956 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2959 btrfs_sysfs_remove_one(fs_info);
2962 btrfs_put_block_group_cache(fs_info);
2963 btrfs_free_block_groups(fs_info);
2966 free_root_pointers(fs_info, 1);
2967 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2970 btrfs_stop_all_workers(fs_info);
2973 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2975 iput(fs_info->btree_inode);
2977 percpu_counter_destroy(&fs_info->bio_counter);
2978 fail_delalloc_bytes:
2979 percpu_counter_destroy(&fs_info->delalloc_bytes);
2980 fail_dirty_metadata_bytes:
2981 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2983 bdi_destroy(&fs_info->bdi);
2985 cleanup_srcu_struct(&fs_info->subvol_srcu);
2987 btrfs_free_stripe_hash_table(fs_info);
2988 btrfs_close_devices(fs_info->fs_devices);
2992 if (!btrfs_test_opt(tree_root, RECOVERY))
2993 goto fail_tree_roots;
2995 free_root_pointers(fs_info, 0);
2997 /* don't use the log in recovery mode, it won't be valid */
2998 btrfs_set_super_log_root(disk_super, 0);
3000 /* we can't trust the free space cache either */
3001 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3003 ret = next_root_backup(fs_info, fs_info->super_copy,
3004 &num_backups_tried, &backup_index);
3006 goto fail_block_groups;
3007 goto retry_root_backup;
3010 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3013 set_buffer_uptodate(bh);
3015 struct btrfs_device *device = (struct btrfs_device *)
3018 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3019 "I/O error on %s\n",
3020 rcu_str_deref(device->name));
3021 /* note, we dont' set_buffer_write_io_error because we have
3022 * our own ways of dealing with the IO errors
3024 clear_buffer_uptodate(bh);
3025 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3031 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3033 struct buffer_head *bh;
3034 struct buffer_head *latest = NULL;
3035 struct btrfs_super_block *super;
3040 /* we would like to check all the supers, but that would make
3041 * a btrfs mount succeed after a mkfs from a different FS.
3042 * So, we need to add a special mount option to scan for
3043 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3045 for (i = 0; i < 1; i++) {
3046 bytenr = btrfs_sb_offset(i);
3047 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3048 i_size_read(bdev->bd_inode))
3050 bh = __bread(bdev, bytenr / 4096,
3051 BTRFS_SUPER_INFO_SIZE);
3055 super = (struct btrfs_super_block *)bh->b_data;
3056 if (btrfs_super_bytenr(super) != bytenr ||
3057 btrfs_super_magic(super) != BTRFS_MAGIC) {
3062 if (!latest || btrfs_super_generation(super) > transid) {
3065 transid = btrfs_super_generation(super);
3074 * this should be called twice, once with wait == 0 and
3075 * once with wait == 1. When wait == 0 is done, all the buffer heads
3076 * we write are pinned.
3078 * They are released when wait == 1 is done.
3079 * max_mirrors must be the same for both runs, and it indicates how
3080 * many supers on this one device should be written.
3082 * max_mirrors == 0 means to write them all.
3084 static int write_dev_supers(struct btrfs_device *device,
3085 struct btrfs_super_block *sb,
3086 int do_barriers, int wait, int max_mirrors)
3088 struct buffer_head *bh;
3095 if (max_mirrors == 0)
3096 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3098 for (i = 0; i < max_mirrors; i++) {
3099 bytenr = btrfs_sb_offset(i);
3100 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3104 bh = __find_get_block(device->bdev, bytenr / 4096,
3105 BTRFS_SUPER_INFO_SIZE);
3111 if (!buffer_uptodate(bh))
3114 /* drop our reference */
3117 /* drop the reference from the wait == 0 run */
3121 btrfs_set_super_bytenr(sb, bytenr);
3124 crc = btrfs_csum_data((char *)sb +
3125 BTRFS_CSUM_SIZE, crc,
3126 BTRFS_SUPER_INFO_SIZE -
3128 btrfs_csum_final(crc, sb->csum);
3131 * one reference for us, and we leave it for the
3134 bh = __getblk(device->bdev, bytenr / 4096,
3135 BTRFS_SUPER_INFO_SIZE);
3137 printk(KERN_ERR "BTRFS: couldn't get super "
3138 "buffer head for bytenr %Lu\n", bytenr);
3143 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3145 /* one reference for submit_bh */
3148 set_buffer_uptodate(bh);
3150 bh->b_end_io = btrfs_end_buffer_write_sync;
3151 bh->b_private = device;
3155 * we fua the first super. The others we allow
3159 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3161 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3165 return errors < i ? 0 : -1;
3169 * endio for the write_dev_flush, this will wake anyone waiting
3170 * for the barrier when it is done
3172 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3175 if (err == -EOPNOTSUPP)
3176 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3177 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3179 if (bio->bi_private)
3180 complete(bio->bi_private);
3185 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3186 * sent down. With wait == 1, it waits for the previous flush.
3188 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3191 static int write_dev_flush(struct btrfs_device *device, int wait)
3196 if (device->nobarriers)
3200 bio = device->flush_bio;
3204 wait_for_completion(&device->flush_wait);
3206 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3207 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3208 rcu_str_deref(device->name));
3209 device->nobarriers = 1;
3210 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3212 btrfs_dev_stat_inc_and_print(device,
3213 BTRFS_DEV_STAT_FLUSH_ERRS);
3216 /* drop the reference from the wait == 0 run */
3218 device->flush_bio = NULL;
3224 * one reference for us, and we leave it for the
3227 device->flush_bio = NULL;
3228 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3232 bio->bi_end_io = btrfs_end_empty_barrier;
3233 bio->bi_bdev = device->bdev;
3234 init_completion(&device->flush_wait);
3235 bio->bi_private = &device->flush_wait;
3236 device->flush_bio = bio;
3239 btrfsic_submit_bio(WRITE_FLUSH, bio);
3245 * send an empty flush down to each device in parallel,
3246 * then wait for them
3248 static int barrier_all_devices(struct btrfs_fs_info *info)
3250 struct list_head *head;
3251 struct btrfs_device *dev;
3252 int errors_send = 0;
3253 int errors_wait = 0;
3256 /* send down all the barriers */
3257 head = &info->fs_devices->devices;
3258 list_for_each_entry_rcu(dev, head, dev_list) {
3265 if (!dev->in_fs_metadata || !dev->writeable)
3268 ret = write_dev_flush(dev, 0);
3273 /* wait for all the barriers */
3274 list_for_each_entry_rcu(dev, head, dev_list) {
3281 if (!dev->in_fs_metadata || !dev->writeable)
3284 ret = write_dev_flush(dev, 1);
3288 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3289 errors_wait > info->num_tolerated_disk_barrier_failures)
3294 int btrfs_calc_num_tolerated_disk_barrier_failures(
3295 struct btrfs_fs_info *fs_info)
3297 struct btrfs_ioctl_space_info space;
3298 struct btrfs_space_info *sinfo;
3299 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3300 BTRFS_BLOCK_GROUP_SYSTEM,
3301 BTRFS_BLOCK_GROUP_METADATA,
3302 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3306 int num_tolerated_disk_barrier_failures =
3307 (int)fs_info->fs_devices->num_devices;
3309 for (i = 0; i < num_types; i++) {
3310 struct btrfs_space_info *tmp;
3314 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3315 if (tmp->flags == types[i]) {
3325 down_read(&sinfo->groups_sem);
3326 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3327 if (!list_empty(&sinfo->block_groups[c])) {
3330 btrfs_get_block_group_info(
3331 &sinfo->block_groups[c], &space);
3332 if (space.total_bytes == 0 ||
3333 space.used_bytes == 0)
3335 flags = space.flags;
3338 * 0: if dup, single or RAID0 is configured for
3339 * any of metadata, system or data, else
3340 * 1: if RAID5 is configured, or if RAID1 or
3341 * RAID10 is configured and only two mirrors
3343 * 2: if RAID6 is configured, else
3344 * num_mirrors - 1: if RAID1 or RAID10 is
3345 * configured and more than
3346 * 2 mirrors are used.
3348 if (num_tolerated_disk_barrier_failures > 0 &&
3349 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3350 BTRFS_BLOCK_GROUP_RAID0)) ||
3351 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3353 num_tolerated_disk_barrier_failures = 0;
3354 else if (num_tolerated_disk_barrier_failures > 1) {
3355 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3356 BTRFS_BLOCK_GROUP_RAID5 |
3357 BTRFS_BLOCK_GROUP_RAID10)) {
3358 num_tolerated_disk_barrier_failures = 1;
3360 BTRFS_BLOCK_GROUP_RAID6) {
3361 num_tolerated_disk_barrier_failures = 2;
3366 up_read(&sinfo->groups_sem);
3369 return num_tolerated_disk_barrier_failures;
3372 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3374 struct list_head *head;
3375 struct btrfs_device *dev;
3376 struct btrfs_super_block *sb;
3377 struct btrfs_dev_item *dev_item;
3381 int total_errors = 0;
3384 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3385 backup_super_roots(root->fs_info);
3387 sb = root->fs_info->super_for_commit;
3388 dev_item = &sb->dev_item;
3390 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3391 head = &root->fs_info->fs_devices->devices;
3392 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3395 ret = barrier_all_devices(root->fs_info);
3398 &root->fs_info->fs_devices->device_list_mutex);
3399 btrfs_error(root->fs_info, ret,
3400 "errors while submitting device barriers.");
3405 list_for_each_entry_rcu(dev, head, dev_list) {
3410 if (!dev->in_fs_metadata || !dev->writeable)
3413 btrfs_set_stack_device_generation(dev_item, 0);
3414 btrfs_set_stack_device_type(dev_item, dev->type);
3415 btrfs_set_stack_device_id(dev_item, dev->devid);
3416 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3417 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3418 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3419 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3420 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3421 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3422 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3424 flags = btrfs_super_flags(sb);
3425 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3427 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3431 if (total_errors > max_errors) {
3432 btrfs_err(root->fs_info, "%d errors while writing supers",
3434 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3436 /* FUA is masked off if unsupported and can't be the reason */
3437 btrfs_error(root->fs_info, -EIO,
3438 "%d errors while writing supers", total_errors);
3443 list_for_each_entry_rcu(dev, head, dev_list) {
3446 if (!dev->in_fs_metadata || !dev->writeable)
3449 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3453 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3454 if (total_errors > max_errors) {
3455 btrfs_error(root->fs_info, -EIO,
3456 "%d errors while writing supers", total_errors);
3462 int write_ctree_super(struct btrfs_trans_handle *trans,
3463 struct btrfs_root *root, int max_mirrors)
3465 return write_all_supers(root, max_mirrors);
3468 /* Drop a fs root from the radix tree and free it. */
3469 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3470 struct btrfs_root *root)
3472 spin_lock(&fs_info->fs_roots_radix_lock);
3473 radix_tree_delete(&fs_info->fs_roots_radix,
3474 (unsigned long)root->root_key.objectid);
3475 spin_unlock(&fs_info->fs_roots_radix_lock);
3477 if (btrfs_root_refs(&root->root_item) == 0)
3478 synchronize_srcu(&fs_info->subvol_srcu);
3480 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3481 btrfs_free_log(NULL, root);
3483 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3484 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3488 static void free_fs_root(struct btrfs_root *root)
3490 iput(root->cache_inode);
3491 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3492 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3493 root->orphan_block_rsv = NULL;
3495 free_anon_bdev(root->anon_dev);
3496 free_extent_buffer(root->node);
3497 free_extent_buffer(root->commit_root);
3498 kfree(root->free_ino_ctl);
3499 kfree(root->free_ino_pinned);
3501 btrfs_put_fs_root(root);
3504 void btrfs_free_fs_root(struct btrfs_root *root)
3509 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3511 u64 root_objectid = 0;
3512 struct btrfs_root *gang[8];
3517 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3518 (void **)gang, root_objectid,
3523 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3524 for (i = 0; i < ret; i++) {
3527 root_objectid = gang[i]->root_key.objectid;
3528 err = btrfs_orphan_cleanup(gang[i]);
3537 int btrfs_commit_super(struct btrfs_root *root)
3539 struct btrfs_trans_handle *trans;
3541 mutex_lock(&root->fs_info->cleaner_mutex);
3542 btrfs_run_delayed_iputs(root);
3543 mutex_unlock(&root->fs_info->cleaner_mutex);
3544 wake_up_process(root->fs_info->cleaner_kthread);
3546 /* wait until ongoing cleanup work done */
3547 down_write(&root->fs_info->cleanup_work_sem);
3548 up_write(&root->fs_info->cleanup_work_sem);
3550 trans = btrfs_join_transaction(root);
3552 return PTR_ERR(trans);
3553 return btrfs_commit_transaction(trans, root);
3556 int close_ctree(struct btrfs_root *root)
3558 struct btrfs_fs_info *fs_info = root->fs_info;
3561 fs_info->closing = 1;
3564 /* wait for the uuid_scan task to finish */
3565 down(&fs_info->uuid_tree_rescan_sem);
3566 /* avoid complains from lockdep et al., set sem back to initial state */
3567 up(&fs_info->uuid_tree_rescan_sem);
3569 /* pause restriper - we want to resume on mount */
3570 btrfs_pause_balance(fs_info);
3572 btrfs_dev_replace_suspend_for_unmount(fs_info);
3574 btrfs_scrub_cancel(fs_info);
3576 /* wait for any defraggers to finish */
3577 wait_event(fs_info->transaction_wait,
3578 (atomic_read(&fs_info->defrag_running) == 0));
3580 /* clear out the rbtree of defraggable inodes */
3581 btrfs_cleanup_defrag_inodes(fs_info);
3583 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3584 ret = btrfs_commit_super(root);
3586 btrfs_err(root->fs_info, "commit super ret %d", ret);
3589 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3590 btrfs_error_commit_super(root);
3592 kthread_stop(fs_info->transaction_kthread);
3593 kthread_stop(fs_info->cleaner_kthread);
3595 fs_info->closing = 2;
3598 btrfs_free_qgroup_config(root->fs_info);
3600 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3601 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3602 percpu_counter_sum(&fs_info->delalloc_bytes));
3605 btrfs_sysfs_remove_one(fs_info);
3607 del_fs_roots(fs_info);
3609 btrfs_put_block_group_cache(fs_info);
3611 btrfs_free_block_groups(fs_info);
3613 btrfs_stop_all_workers(fs_info);
3615 free_root_pointers(fs_info, 1);
3617 iput(fs_info->btree_inode);
3619 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3620 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3621 btrfsic_unmount(root, fs_info->fs_devices);
3624 btrfs_close_devices(fs_info->fs_devices);
3625 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3627 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3628 percpu_counter_destroy(&fs_info->delalloc_bytes);
3629 percpu_counter_destroy(&fs_info->bio_counter);
3630 bdi_destroy(&fs_info->bdi);
3631 cleanup_srcu_struct(&fs_info->subvol_srcu);
3633 btrfs_free_stripe_hash_table(fs_info);
3635 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3636 root->orphan_block_rsv = NULL;
3641 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3645 struct inode *btree_inode = buf->pages[0]->mapping->host;
3647 ret = extent_buffer_uptodate(buf);
3651 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3652 parent_transid, atomic);
3658 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3660 return set_extent_buffer_uptodate(buf);
3663 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3665 struct btrfs_root *root;
3666 u64 transid = btrfs_header_generation(buf);
3669 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3671 * This is a fast path so only do this check if we have sanity tests
3672 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3673 * outside of the sanity tests.
3675 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3678 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3679 btrfs_assert_tree_locked(buf);
3680 if (transid != root->fs_info->generation)
3681 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3682 "found %llu running %llu\n",
3683 buf->start, transid, root->fs_info->generation);
3684 was_dirty = set_extent_buffer_dirty(buf);
3686 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3688 root->fs_info->dirty_metadata_batch);
3691 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3695 * looks as though older kernels can get into trouble with
3696 * this code, they end up stuck in balance_dirty_pages forever
3700 if (current->flags & PF_MEMALLOC)
3704 btrfs_balance_delayed_items(root);
3706 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3707 BTRFS_DIRTY_METADATA_THRESH);
3709 balance_dirty_pages_ratelimited(
3710 root->fs_info->btree_inode->i_mapping);
3715 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3717 __btrfs_btree_balance_dirty(root, 1);
3720 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3722 __btrfs_btree_balance_dirty(root, 0);
3725 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3727 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3728 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3731 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3735 * Placeholder for checks
3740 static void btrfs_error_commit_super(struct btrfs_root *root)
3742 mutex_lock(&root->fs_info->cleaner_mutex);
3743 btrfs_run_delayed_iputs(root);
3744 mutex_unlock(&root->fs_info->cleaner_mutex);
3746 down_write(&root->fs_info->cleanup_work_sem);
3747 up_write(&root->fs_info->cleanup_work_sem);
3749 /* cleanup FS via transaction */
3750 btrfs_cleanup_transaction(root);
3753 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3754 struct btrfs_root *root)
3756 struct btrfs_inode *btrfs_inode;
3757 struct list_head splice;
3759 INIT_LIST_HEAD(&splice);
3761 mutex_lock(&root->fs_info->ordered_operations_mutex);
3762 spin_lock(&root->fs_info->ordered_root_lock);
3764 list_splice_init(&t->ordered_operations, &splice);
3765 while (!list_empty(&splice)) {
3766 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3767 ordered_operations);
3769 list_del_init(&btrfs_inode->ordered_operations);
3770 spin_unlock(&root->fs_info->ordered_root_lock);
3772 btrfs_invalidate_inodes(btrfs_inode->root);
3774 spin_lock(&root->fs_info->ordered_root_lock);
3777 spin_unlock(&root->fs_info->ordered_root_lock);
3778 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3781 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3783 struct btrfs_ordered_extent *ordered;
3785 spin_lock(&root->ordered_extent_lock);
3787 * This will just short circuit the ordered completion stuff which will
3788 * make sure the ordered extent gets properly cleaned up.
3790 list_for_each_entry(ordered, &root->ordered_extents,
3792 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3793 spin_unlock(&root->ordered_extent_lock);
3796 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3798 struct btrfs_root *root;
3799 struct list_head splice;
3801 INIT_LIST_HEAD(&splice);
3803 spin_lock(&fs_info->ordered_root_lock);
3804 list_splice_init(&fs_info->ordered_roots, &splice);
3805 while (!list_empty(&splice)) {
3806 root = list_first_entry(&splice, struct btrfs_root,
3808 list_move_tail(&root->ordered_root,
3809 &fs_info->ordered_roots);
3811 spin_unlock(&fs_info->ordered_root_lock);
3812 btrfs_destroy_ordered_extents(root);
3815 spin_lock(&fs_info->ordered_root_lock);
3817 spin_unlock(&fs_info->ordered_root_lock);
3820 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3821 struct btrfs_root *root)
3823 struct rb_node *node;
3824 struct btrfs_delayed_ref_root *delayed_refs;
3825 struct btrfs_delayed_ref_node *ref;
3828 delayed_refs = &trans->delayed_refs;
3830 spin_lock(&delayed_refs->lock);
3831 if (atomic_read(&delayed_refs->num_entries) == 0) {
3832 spin_unlock(&delayed_refs->lock);
3833 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3837 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3838 struct btrfs_delayed_ref_head *head;
3839 bool pin_bytes = false;
3841 head = rb_entry(node, struct btrfs_delayed_ref_head,
3843 if (!mutex_trylock(&head->mutex)) {
3844 atomic_inc(&head->node.refs);
3845 spin_unlock(&delayed_refs->lock);
3847 mutex_lock(&head->mutex);
3848 mutex_unlock(&head->mutex);
3849 btrfs_put_delayed_ref(&head->node);
3850 spin_lock(&delayed_refs->lock);
3853 spin_lock(&head->lock);
3854 while ((node = rb_first(&head->ref_root)) != NULL) {
3855 ref = rb_entry(node, struct btrfs_delayed_ref_node,
3858 rb_erase(&ref->rb_node, &head->ref_root);
3859 atomic_dec(&delayed_refs->num_entries);
3860 btrfs_put_delayed_ref(ref);
3862 if (head->must_insert_reserved)
3864 btrfs_free_delayed_extent_op(head->extent_op);
3865 delayed_refs->num_heads--;
3866 if (head->processing == 0)
3867 delayed_refs->num_heads_ready--;
3868 atomic_dec(&delayed_refs->num_entries);
3869 head->node.in_tree = 0;
3870 rb_erase(&head->href_node, &delayed_refs->href_root);
3871 spin_unlock(&head->lock);
3872 spin_unlock(&delayed_refs->lock);
3873 mutex_unlock(&head->mutex);
3876 btrfs_pin_extent(root, head->node.bytenr,
3877 head->node.num_bytes, 1);
3878 btrfs_put_delayed_ref(&head->node);
3880 spin_lock(&delayed_refs->lock);
3883 spin_unlock(&delayed_refs->lock);
3888 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3890 struct btrfs_inode *btrfs_inode;
3891 struct list_head splice;
3893 INIT_LIST_HEAD(&splice);
3895 spin_lock(&root->delalloc_lock);
3896 list_splice_init(&root->delalloc_inodes, &splice);
3898 while (!list_empty(&splice)) {
3899 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3902 list_del_init(&btrfs_inode->delalloc_inodes);
3903 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3904 &btrfs_inode->runtime_flags);
3905 spin_unlock(&root->delalloc_lock);
3907 btrfs_invalidate_inodes(btrfs_inode->root);
3909 spin_lock(&root->delalloc_lock);
3912 spin_unlock(&root->delalloc_lock);
3915 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3917 struct btrfs_root *root;
3918 struct list_head splice;
3920 INIT_LIST_HEAD(&splice);
3922 spin_lock(&fs_info->delalloc_root_lock);
3923 list_splice_init(&fs_info->delalloc_roots, &splice);
3924 while (!list_empty(&splice)) {
3925 root = list_first_entry(&splice, struct btrfs_root,
3927 list_del_init(&root->delalloc_root);
3928 root = btrfs_grab_fs_root(root);
3930 spin_unlock(&fs_info->delalloc_root_lock);
3932 btrfs_destroy_delalloc_inodes(root);
3933 btrfs_put_fs_root(root);
3935 spin_lock(&fs_info->delalloc_root_lock);
3937 spin_unlock(&fs_info->delalloc_root_lock);
3940 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3941 struct extent_io_tree *dirty_pages,
3945 struct extent_buffer *eb;
3950 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3955 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3956 while (start <= end) {
3957 eb = btrfs_find_tree_block(root, start,
3959 start += root->leafsize;
3962 wait_on_extent_buffer_writeback(eb);
3964 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3966 clear_extent_buffer_dirty(eb);
3967 free_extent_buffer_stale(eb);
3974 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3975 struct extent_io_tree *pinned_extents)
3977 struct extent_io_tree *unpin;
3983 unpin = pinned_extents;
3986 ret = find_first_extent_bit(unpin, 0, &start, &end,
3987 EXTENT_DIRTY, NULL);
3992 if (btrfs_test_opt(root, DISCARD))
3993 ret = btrfs_error_discard_extent(root, start,
3997 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3998 btrfs_error_unpin_extent_range(root, start, end);
4003 if (unpin == &root->fs_info->freed_extents[0])
4004 unpin = &root->fs_info->freed_extents[1];
4006 unpin = &root->fs_info->freed_extents[0];
4014 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4015 struct btrfs_root *root)
4017 btrfs_destroy_ordered_operations(cur_trans, root);
4019 btrfs_destroy_delayed_refs(cur_trans, root);
4021 cur_trans->state = TRANS_STATE_COMMIT_START;
4022 wake_up(&root->fs_info->transaction_blocked_wait);
4024 cur_trans->state = TRANS_STATE_UNBLOCKED;
4025 wake_up(&root->fs_info->transaction_wait);
4027 btrfs_destroy_delayed_inodes(root);
4028 btrfs_assert_delayed_root_empty(root);
4030 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4032 btrfs_destroy_pinned_extent(root,
4033 root->fs_info->pinned_extents);
4035 cur_trans->state =TRANS_STATE_COMPLETED;
4036 wake_up(&cur_trans->commit_wait);
4039 memset(cur_trans, 0, sizeof(*cur_trans));
4040 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4044 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4046 struct btrfs_transaction *t;
4048 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4050 spin_lock(&root->fs_info->trans_lock);
4051 while (!list_empty(&root->fs_info->trans_list)) {
4052 t = list_first_entry(&root->fs_info->trans_list,
4053 struct btrfs_transaction, list);
4054 if (t->state >= TRANS_STATE_COMMIT_START) {
4055 atomic_inc(&t->use_count);
4056 spin_unlock(&root->fs_info->trans_lock);
4057 btrfs_wait_for_commit(root, t->transid);
4058 btrfs_put_transaction(t);
4059 spin_lock(&root->fs_info->trans_lock);
4062 if (t == root->fs_info->running_transaction) {
4063 t->state = TRANS_STATE_COMMIT_DOING;
4064 spin_unlock(&root->fs_info->trans_lock);
4066 * We wait for 0 num_writers since we don't hold a trans
4067 * handle open currently for this transaction.
4069 wait_event(t->writer_wait,
4070 atomic_read(&t->num_writers) == 0);
4072 spin_unlock(&root->fs_info->trans_lock);
4074 btrfs_cleanup_one_transaction(t, root);
4076 spin_lock(&root->fs_info->trans_lock);
4077 if (t == root->fs_info->running_transaction)
4078 root->fs_info->running_transaction = NULL;
4079 list_del_init(&t->list);
4080 spin_unlock(&root->fs_info->trans_lock);
4082 btrfs_put_transaction(t);
4083 trace_btrfs_transaction_commit(root);
4084 spin_lock(&root->fs_info->trans_lock);
4086 spin_unlock(&root->fs_info->trans_lock);
4087 btrfs_destroy_all_ordered_extents(root->fs_info);
4088 btrfs_destroy_delayed_inodes(root);
4089 btrfs_assert_delayed_root_empty(root);
4090 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4091 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4092 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4097 static struct extent_io_ops btree_extent_io_ops = {
4098 .readpage_end_io_hook = btree_readpage_end_io_hook,
4099 .readpage_io_failed_hook = btree_io_failed_hook,
4100 .submit_bio_hook = btree_submit_bio_hook,
4101 /* note we're sharing with inode.c for the merge bio hook */
4102 .merge_bio_hook = btrfs_merge_bio_hook,