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_struct 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->thread_pool_size,
742 info->fs_devices->open_devices);
746 static void run_one_async_start(struct btrfs_work_struct *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_struct *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_struct *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 btrfs_init_work(&async->work, run_one_async_start,
815 run_one_async_done, run_one_async_free);
817 async->bio_flags = bio_flags;
818 async->bio_offset = bio_offset;
822 atomic_inc(&fs_info->nr_async_submits);
825 btrfs_set_work_high_priority(&async->work);
827 btrfs_queue_work(fs_info->workers, &async->work);
829 while (atomic_read(&fs_info->async_submit_draining) &&
830 atomic_read(&fs_info->nr_async_submits)) {
831 wait_event(fs_info->async_submit_wait,
832 (atomic_read(&fs_info->nr_async_submits) == 0));
838 static int btree_csum_one_bio(struct bio *bio)
840 struct bio_vec *bvec = bio->bi_io_vec;
842 struct btrfs_root *root;
845 WARN_ON(bio->bi_vcnt <= 0);
846 while (bio_index < bio->bi_vcnt) {
847 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
848 ret = csum_dirty_buffer(root, bvec->bv_page);
857 static int __btree_submit_bio_start(struct inode *inode, int rw,
858 struct bio *bio, int mirror_num,
859 unsigned long bio_flags,
863 * when we're called for a write, we're already in the async
864 * submission context. Just jump into btrfs_map_bio
866 return btree_csum_one_bio(bio);
869 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
870 int mirror_num, unsigned long bio_flags,
876 * when we're called for a write, we're already in the async
877 * submission context. Just jump into btrfs_map_bio
879 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
885 static int check_async_write(struct inode *inode, unsigned long bio_flags)
887 if (bio_flags & EXTENT_BIO_TREE_LOG)
896 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
897 int mirror_num, unsigned long bio_flags,
900 int async = check_async_write(inode, bio_flags);
903 if (!(rw & REQ_WRITE)) {
905 * called for a read, do the setup so that checksum validation
906 * can happen in the async kernel threads
908 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
912 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
915 ret = btree_csum_one_bio(bio);
918 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
922 * kthread helpers are used to submit writes so that
923 * checksumming can happen in parallel across all CPUs
925 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
926 inode, rw, bio, mirror_num, 0,
928 __btree_submit_bio_start,
929 __btree_submit_bio_done);
939 #ifdef CONFIG_MIGRATION
940 static int btree_migratepage(struct address_space *mapping,
941 struct page *newpage, struct page *page,
942 enum migrate_mode mode)
945 * we can't safely write a btree page from here,
946 * we haven't done the locking hook
951 * Buffers may be managed in a filesystem specific way.
952 * We must have no buffers or drop them.
954 if (page_has_private(page) &&
955 !try_to_release_page(page, GFP_KERNEL))
957 return migrate_page(mapping, newpage, page, mode);
962 static int btree_writepages(struct address_space *mapping,
963 struct writeback_control *wbc)
965 struct btrfs_fs_info *fs_info;
968 if (wbc->sync_mode == WB_SYNC_NONE) {
970 if (wbc->for_kupdate)
973 fs_info = BTRFS_I(mapping->host)->root->fs_info;
974 /* this is a bit racy, but that's ok */
975 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
976 BTRFS_DIRTY_METADATA_THRESH);
980 return btree_write_cache_pages(mapping, wbc);
983 static int btree_readpage(struct file *file, struct page *page)
985 struct extent_io_tree *tree;
986 tree = &BTRFS_I(page->mapping->host)->io_tree;
987 return extent_read_full_page(tree, page, btree_get_extent, 0);
990 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
992 if (PageWriteback(page) || PageDirty(page))
995 return try_release_extent_buffer(page);
998 static void btree_invalidatepage(struct page *page, unsigned int offset,
1001 struct extent_io_tree *tree;
1002 tree = &BTRFS_I(page->mapping->host)->io_tree;
1003 extent_invalidatepage(tree, page, offset);
1004 btree_releasepage(page, GFP_NOFS);
1005 if (PagePrivate(page)) {
1006 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1007 "page private not zero on page %llu",
1008 (unsigned long long)page_offset(page));
1009 ClearPagePrivate(page);
1010 set_page_private(page, 0);
1011 page_cache_release(page);
1015 static int btree_set_page_dirty(struct page *page)
1018 struct extent_buffer *eb;
1020 BUG_ON(!PagePrivate(page));
1021 eb = (struct extent_buffer *)page->private;
1023 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1024 BUG_ON(!atomic_read(&eb->refs));
1025 btrfs_assert_tree_locked(eb);
1027 return __set_page_dirty_nobuffers(page);
1030 static const struct address_space_operations btree_aops = {
1031 .readpage = btree_readpage,
1032 .writepages = btree_writepages,
1033 .releasepage = btree_releasepage,
1034 .invalidatepage = btree_invalidatepage,
1035 #ifdef CONFIG_MIGRATION
1036 .migratepage = btree_migratepage,
1038 .set_page_dirty = btree_set_page_dirty,
1041 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1044 struct extent_buffer *buf = NULL;
1045 struct inode *btree_inode = root->fs_info->btree_inode;
1048 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1051 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1052 buf, 0, WAIT_NONE, btree_get_extent, 0);
1053 free_extent_buffer(buf);
1057 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1058 int mirror_num, struct extent_buffer **eb)
1060 struct extent_buffer *buf = NULL;
1061 struct inode *btree_inode = root->fs_info->btree_inode;
1062 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1065 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1069 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1071 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1072 btree_get_extent, mirror_num);
1074 free_extent_buffer(buf);
1078 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1079 free_extent_buffer(buf);
1081 } else if (extent_buffer_uptodate(buf)) {
1084 free_extent_buffer(buf);
1089 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1090 u64 bytenr, u32 blocksize)
1092 return find_extent_buffer(root->fs_info, bytenr);
1095 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1096 u64 bytenr, u32 blocksize)
1098 return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1102 int btrfs_write_tree_block(struct extent_buffer *buf)
1104 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1105 buf->start + buf->len - 1);
1108 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1110 return filemap_fdatawait_range(buf->pages[0]->mapping,
1111 buf->start, buf->start + buf->len - 1);
1114 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1115 u32 blocksize, u64 parent_transid)
1117 struct extent_buffer *buf = NULL;
1120 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1124 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1126 free_extent_buffer(buf);
1133 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1134 struct extent_buffer *buf)
1136 struct btrfs_fs_info *fs_info = root->fs_info;
1138 if (btrfs_header_generation(buf) ==
1139 fs_info->running_transaction->transid) {
1140 btrfs_assert_tree_locked(buf);
1142 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1143 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1145 fs_info->dirty_metadata_batch);
1146 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1147 btrfs_set_lock_blocking(buf);
1148 clear_extent_buffer_dirty(buf);
1153 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1154 u32 stripesize, struct btrfs_root *root,
1155 struct btrfs_fs_info *fs_info,
1159 root->commit_root = NULL;
1160 root->sectorsize = sectorsize;
1161 root->nodesize = nodesize;
1162 root->leafsize = leafsize;
1163 root->stripesize = stripesize;
1165 root->track_dirty = 0;
1167 root->orphan_item_inserted = 0;
1168 root->orphan_cleanup_state = 0;
1170 root->objectid = objectid;
1171 root->last_trans = 0;
1172 root->highest_objectid = 0;
1173 root->nr_delalloc_inodes = 0;
1174 root->nr_ordered_extents = 0;
1176 root->inode_tree = RB_ROOT;
1177 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1178 root->block_rsv = NULL;
1179 root->orphan_block_rsv = NULL;
1181 INIT_LIST_HEAD(&root->dirty_list);
1182 INIT_LIST_HEAD(&root->root_list);
1183 INIT_LIST_HEAD(&root->delalloc_inodes);
1184 INIT_LIST_HEAD(&root->delalloc_root);
1185 INIT_LIST_HEAD(&root->ordered_extents);
1186 INIT_LIST_HEAD(&root->ordered_root);
1187 INIT_LIST_HEAD(&root->logged_list[0]);
1188 INIT_LIST_HEAD(&root->logged_list[1]);
1189 spin_lock_init(&root->orphan_lock);
1190 spin_lock_init(&root->inode_lock);
1191 spin_lock_init(&root->delalloc_lock);
1192 spin_lock_init(&root->ordered_extent_lock);
1193 spin_lock_init(&root->accounting_lock);
1194 spin_lock_init(&root->log_extents_lock[0]);
1195 spin_lock_init(&root->log_extents_lock[1]);
1196 mutex_init(&root->objectid_mutex);
1197 mutex_init(&root->log_mutex);
1198 init_waitqueue_head(&root->log_writer_wait);
1199 init_waitqueue_head(&root->log_commit_wait[0]);
1200 init_waitqueue_head(&root->log_commit_wait[1]);
1201 INIT_LIST_HEAD(&root->log_ctxs[0]);
1202 INIT_LIST_HEAD(&root->log_ctxs[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->log_transid_committed = -1;
1211 root->last_log_commit = 0;
1213 extent_io_tree_init(&root->dirty_log_pages,
1214 fs_info->btree_inode->i_mapping);
1216 memset(&root->root_key, 0, sizeof(root->root_key));
1217 memset(&root->root_item, 0, sizeof(root->root_item));
1218 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1219 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1221 root->defrag_trans_start = fs_info->generation;
1223 root->defrag_trans_start = 0;
1224 init_completion(&root->kobj_unregister);
1225 root->defrag_running = 0;
1226 root->root_key.objectid = objectid;
1229 spin_lock_init(&root->root_item_lock);
1232 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1234 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1236 root->fs_info = fs_info;
1240 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1241 /* Should only be used by the testing infrastructure */
1242 struct btrfs_root *btrfs_alloc_dummy_root(void)
1244 struct btrfs_root *root;
1246 root = btrfs_alloc_root(NULL);
1248 return ERR_PTR(-ENOMEM);
1249 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1250 root->dummy_root = 1;
1256 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1257 struct btrfs_fs_info *fs_info,
1260 struct extent_buffer *leaf;
1261 struct btrfs_root *tree_root = fs_info->tree_root;
1262 struct btrfs_root *root;
1263 struct btrfs_key key;
1267 root = btrfs_alloc_root(fs_info);
1269 return ERR_PTR(-ENOMEM);
1271 __setup_root(tree_root->nodesize, tree_root->leafsize,
1272 tree_root->sectorsize, tree_root->stripesize,
1273 root, fs_info, objectid);
1274 root->root_key.objectid = objectid;
1275 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1276 root->root_key.offset = 0;
1278 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1279 0, objectid, NULL, 0, 0, 0);
1281 ret = PTR_ERR(leaf);
1286 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1287 btrfs_set_header_bytenr(leaf, leaf->start);
1288 btrfs_set_header_generation(leaf, trans->transid);
1289 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1290 btrfs_set_header_owner(leaf, objectid);
1293 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1295 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1296 btrfs_header_chunk_tree_uuid(leaf),
1298 btrfs_mark_buffer_dirty(leaf);
1300 root->commit_root = btrfs_root_node(root);
1301 root->track_dirty = 1;
1304 root->root_item.flags = 0;
1305 root->root_item.byte_limit = 0;
1306 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1307 btrfs_set_root_generation(&root->root_item, trans->transid);
1308 btrfs_set_root_level(&root->root_item, 0);
1309 btrfs_set_root_refs(&root->root_item, 1);
1310 btrfs_set_root_used(&root->root_item, leaf->len);
1311 btrfs_set_root_last_snapshot(&root->root_item, 0);
1312 btrfs_set_root_dirid(&root->root_item, 0);
1314 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1315 root->root_item.drop_level = 0;
1317 key.objectid = objectid;
1318 key.type = BTRFS_ROOT_ITEM_KEY;
1320 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1324 btrfs_tree_unlock(leaf);
1330 btrfs_tree_unlock(leaf);
1331 free_extent_buffer(leaf);
1335 return ERR_PTR(ret);
1338 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1339 struct btrfs_fs_info *fs_info)
1341 struct btrfs_root *root;
1342 struct btrfs_root *tree_root = fs_info->tree_root;
1343 struct extent_buffer *leaf;
1345 root = btrfs_alloc_root(fs_info);
1347 return ERR_PTR(-ENOMEM);
1349 __setup_root(tree_root->nodesize, tree_root->leafsize,
1350 tree_root->sectorsize, tree_root->stripesize,
1351 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1353 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1354 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1355 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1357 * log trees do not get reference counted because they go away
1358 * before a real commit is actually done. They do store pointers
1359 * to file data extents, and those reference counts still get
1360 * updated (along with back refs to the log tree).
1364 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1365 BTRFS_TREE_LOG_OBJECTID, NULL,
1369 return ERR_CAST(leaf);
1372 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1373 btrfs_set_header_bytenr(leaf, leaf->start);
1374 btrfs_set_header_generation(leaf, trans->transid);
1375 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1376 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1379 write_extent_buffer(root->node, root->fs_info->fsid,
1380 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1381 btrfs_mark_buffer_dirty(root->node);
1382 btrfs_tree_unlock(root->node);
1386 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1387 struct btrfs_fs_info *fs_info)
1389 struct btrfs_root *log_root;
1391 log_root = alloc_log_tree(trans, fs_info);
1392 if (IS_ERR(log_root))
1393 return PTR_ERR(log_root);
1394 WARN_ON(fs_info->log_root_tree);
1395 fs_info->log_root_tree = log_root;
1399 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1400 struct btrfs_root *root)
1402 struct btrfs_root *log_root;
1403 struct btrfs_inode_item *inode_item;
1405 log_root = alloc_log_tree(trans, root->fs_info);
1406 if (IS_ERR(log_root))
1407 return PTR_ERR(log_root);
1409 log_root->last_trans = trans->transid;
1410 log_root->root_key.offset = root->root_key.objectid;
1412 inode_item = &log_root->root_item.inode;
1413 btrfs_set_stack_inode_generation(inode_item, 1);
1414 btrfs_set_stack_inode_size(inode_item, 3);
1415 btrfs_set_stack_inode_nlink(inode_item, 1);
1416 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1417 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1419 btrfs_set_root_node(&log_root->root_item, log_root->node);
1421 WARN_ON(root->log_root);
1422 root->log_root = log_root;
1423 root->log_transid = 0;
1424 root->log_transid_committed = -1;
1425 root->last_log_commit = 0;
1429 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1430 struct btrfs_key *key)
1432 struct btrfs_root *root;
1433 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1434 struct btrfs_path *path;
1439 path = btrfs_alloc_path();
1441 return ERR_PTR(-ENOMEM);
1443 root = btrfs_alloc_root(fs_info);
1449 __setup_root(tree_root->nodesize, tree_root->leafsize,
1450 tree_root->sectorsize, tree_root->stripesize,
1451 root, fs_info, key->objectid);
1453 ret = btrfs_find_root(tree_root, key, path,
1454 &root->root_item, &root->root_key);
1461 generation = btrfs_root_generation(&root->root_item);
1462 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1463 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1464 blocksize, generation);
1468 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1472 root->commit_root = btrfs_root_node(root);
1474 btrfs_free_path(path);
1478 free_extent_buffer(root->node);
1482 root = ERR_PTR(ret);
1486 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1487 struct btrfs_key *location)
1489 struct btrfs_root *root;
1491 root = btrfs_read_tree_root(tree_root, location);
1495 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1497 btrfs_check_and_init_root_item(&root->root_item);
1503 int btrfs_init_fs_root(struct btrfs_root *root)
1507 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1508 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1510 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1515 btrfs_init_free_ino_ctl(root);
1516 mutex_init(&root->fs_commit_mutex);
1517 spin_lock_init(&root->cache_lock);
1518 init_waitqueue_head(&root->cache_wait);
1520 ret = get_anon_bdev(&root->anon_dev);
1525 kfree(root->free_ino_ctl);
1526 kfree(root->free_ino_pinned);
1530 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1533 struct btrfs_root *root;
1535 spin_lock(&fs_info->fs_roots_radix_lock);
1536 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1537 (unsigned long)root_id);
1538 spin_unlock(&fs_info->fs_roots_radix_lock);
1542 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1543 struct btrfs_root *root)
1547 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1551 spin_lock(&fs_info->fs_roots_radix_lock);
1552 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1553 (unsigned long)root->root_key.objectid,
1557 spin_unlock(&fs_info->fs_roots_radix_lock);
1558 radix_tree_preload_end();
1563 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1564 struct btrfs_key *location,
1567 struct btrfs_root *root;
1570 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1571 return fs_info->tree_root;
1572 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1573 return fs_info->extent_root;
1574 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1575 return fs_info->chunk_root;
1576 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1577 return fs_info->dev_root;
1578 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1579 return fs_info->csum_root;
1580 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1581 return fs_info->quota_root ? fs_info->quota_root :
1583 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1584 return fs_info->uuid_root ? fs_info->uuid_root :
1587 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1589 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1590 return ERR_PTR(-ENOENT);
1594 root = btrfs_read_fs_root(fs_info->tree_root, location);
1598 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1603 ret = btrfs_init_fs_root(root);
1607 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1608 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1612 root->orphan_item_inserted = 1;
1614 ret = btrfs_insert_fs_root(fs_info, root);
1616 if (ret == -EEXIST) {
1625 return ERR_PTR(ret);
1628 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1630 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1632 struct btrfs_device *device;
1633 struct backing_dev_info *bdi;
1636 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1639 bdi = blk_get_backing_dev_info(device->bdev);
1640 if (bdi && bdi_congested(bdi, bdi_bits)) {
1650 * If this fails, caller must call bdi_destroy() to get rid of the
1653 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1657 bdi->capabilities = BDI_CAP_MAP_COPY;
1658 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1662 bdi->ra_pages = default_backing_dev_info.ra_pages;
1663 bdi->congested_fn = btrfs_congested_fn;
1664 bdi->congested_data = info;
1669 * called by the kthread helper functions to finally call the bio end_io
1670 * functions. This is where read checksum verification actually happens
1672 static void end_workqueue_fn(struct btrfs_work *work)
1675 struct end_io_wq *end_io_wq;
1678 end_io_wq = container_of(work, struct end_io_wq, work);
1679 bio = end_io_wq->bio;
1681 error = end_io_wq->error;
1682 bio->bi_private = end_io_wq->private;
1683 bio->bi_end_io = end_io_wq->end_io;
1685 bio_endio(bio, error);
1688 static int cleaner_kthread(void *arg)
1690 struct btrfs_root *root = arg;
1696 /* Make the cleaner go to sleep early. */
1697 if (btrfs_need_cleaner_sleep(root))
1700 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1704 * Avoid the problem that we change the status of the fs
1705 * during the above check and trylock.
1707 if (btrfs_need_cleaner_sleep(root)) {
1708 mutex_unlock(&root->fs_info->cleaner_mutex);
1712 btrfs_run_delayed_iputs(root);
1713 again = btrfs_clean_one_deleted_snapshot(root);
1714 mutex_unlock(&root->fs_info->cleaner_mutex);
1717 * The defragger has dealt with the R/O remount and umount,
1718 * needn't do anything special here.
1720 btrfs_run_defrag_inodes(root->fs_info);
1722 if (!try_to_freeze() && !again) {
1723 set_current_state(TASK_INTERRUPTIBLE);
1724 if (!kthread_should_stop())
1726 __set_current_state(TASK_RUNNING);
1728 } while (!kthread_should_stop());
1732 static int transaction_kthread(void *arg)
1734 struct btrfs_root *root = arg;
1735 struct btrfs_trans_handle *trans;
1736 struct btrfs_transaction *cur;
1739 unsigned long delay;
1743 cannot_commit = false;
1744 delay = HZ * root->fs_info->commit_interval;
1745 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1747 spin_lock(&root->fs_info->trans_lock);
1748 cur = root->fs_info->running_transaction;
1750 spin_unlock(&root->fs_info->trans_lock);
1754 now = get_seconds();
1755 if (cur->state < TRANS_STATE_BLOCKED &&
1756 (now < cur->start_time ||
1757 now - cur->start_time < root->fs_info->commit_interval)) {
1758 spin_unlock(&root->fs_info->trans_lock);
1762 transid = cur->transid;
1763 spin_unlock(&root->fs_info->trans_lock);
1765 /* If the file system is aborted, this will always fail. */
1766 trans = btrfs_attach_transaction(root);
1767 if (IS_ERR(trans)) {
1768 if (PTR_ERR(trans) != -ENOENT)
1769 cannot_commit = true;
1772 if (transid == trans->transid) {
1773 btrfs_commit_transaction(trans, root);
1775 btrfs_end_transaction(trans, root);
1778 wake_up_process(root->fs_info->cleaner_kthread);
1779 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1781 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1782 &root->fs_info->fs_state)))
1783 btrfs_cleanup_transaction(root);
1784 if (!try_to_freeze()) {
1785 set_current_state(TASK_INTERRUPTIBLE);
1786 if (!kthread_should_stop() &&
1787 (!btrfs_transaction_blocked(root->fs_info) ||
1789 schedule_timeout(delay);
1790 __set_current_state(TASK_RUNNING);
1792 } while (!kthread_should_stop());
1797 * this will find the highest generation in the array of
1798 * root backups. The index of the highest array is returned,
1799 * or -1 if we can't find anything.
1801 * We check to make sure the array is valid by comparing the
1802 * generation of the latest root in the array with the generation
1803 * in the super block. If they don't match we pitch it.
1805 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1808 int newest_index = -1;
1809 struct btrfs_root_backup *root_backup;
1812 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1813 root_backup = info->super_copy->super_roots + i;
1814 cur = btrfs_backup_tree_root_gen(root_backup);
1815 if (cur == newest_gen)
1819 /* check to see if we actually wrapped around */
1820 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1821 root_backup = info->super_copy->super_roots;
1822 cur = btrfs_backup_tree_root_gen(root_backup);
1823 if (cur == newest_gen)
1826 return newest_index;
1831 * find the oldest backup so we know where to store new entries
1832 * in the backup array. This will set the backup_root_index
1833 * field in the fs_info struct
1835 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1838 int newest_index = -1;
1840 newest_index = find_newest_super_backup(info, newest_gen);
1841 /* if there was garbage in there, just move along */
1842 if (newest_index == -1) {
1843 info->backup_root_index = 0;
1845 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1850 * copy all the root pointers into the super backup array.
1851 * this will bump the backup pointer by one when it is
1854 static void backup_super_roots(struct btrfs_fs_info *info)
1857 struct btrfs_root_backup *root_backup;
1860 next_backup = info->backup_root_index;
1861 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1862 BTRFS_NUM_BACKUP_ROOTS;
1865 * just overwrite the last backup if we're at the same generation
1866 * this happens only at umount
1868 root_backup = info->super_for_commit->super_roots + last_backup;
1869 if (btrfs_backup_tree_root_gen(root_backup) ==
1870 btrfs_header_generation(info->tree_root->node))
1871 next_backup = last_backup;
1873 root_backup = info->super_for_commit->super_roots + next_backup;
1876 * make sure all of our padding and empty slots get zero filled
1877 * regardless of which ones we use today
1879 memset(root_backup, 0, sizeof(*root_backup));
1881 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1883 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1884 btrfs_set_backup_tree_root_gen(root_backup,
1885 btrfs_header_generation(info->tree_root->node));
1887 btrfs_set_backup_tree_root_level(root_backup,
1888 btrfs_header_level(info->tree_root->node));
1890 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1891 btrfs_set_backup_chunk_root_gen(root_backup,
1892 btrfs_header_generation(info->chunk_root->node));
1893 btrfs_set_backup_chunk_root_level(root_backup,
1894 btrfs_header_level(info->chunk_root->node));
1896 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1897 btrfs_set_backup_extent_root_gen(root_backup,
1898 btrfs_header_generation(info->extent_root->node));
1899 btrfs_set_backup_extent_root_level(root_backup,
1900 btrfs_header_level(info->extent_root->node));
1903 * we might commit during log recovery, which happens before we set
1904 * the fs_root. Make sure it is valid before we fill it in.
1906 if (info->fs_root && info->fs_root->node) {
1907 btrfs_set_backup_fs_root(root_backup,
1908 info->fs_root->node->start);
1909 btrfs_set_backup_fs_root_gen(root_backup,
1910 btrfs_header_generation(info->fs_root->node));
1911 btrfs_set_backup_fs_root_level(root_backup,
1912 btrfs_header_level(info->fs_root->node));
1915 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1916 btrfs_set_backup_dev_root_gen(root_backup,
1917 btrfs_header_generation(info->dev_root->node));
1918 btrfs_set_backup_dev_root_level(root_backup,
1919 btrfs_header_level(info->dev_root->node));
1921 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1922 btrfs_set_backup_csum_root_gen(root_backup,
1923 btrfs_header_generation(info->csum_root->node));
1924 btrfs_set_backup_csum_root_level(root_backup,
1925 btrfs_header_level(info->csum_root->node));
1927 btrfs_set_backup_total_bytes(root_backup,
1928 btrfs_super_total_bytes(info->super_copy));
1929 btrfs_set_backup_bytes_used(root_backup,
1930 btrfs_super_bytes_used(info->super_copy));
1931 btrfs_set_backup_num_devices(root_backup,
1932 btrfs_super_num_devices(info->super_copy));
1935 * if we don't copy this out to the super_copy, it won't get remembered
1936 * for the next commit
1938 memcpy(&info->super_copy->super_roots,
1939 &info->super_for_commit->super_roots,
1940 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1944 * this copies info out of the root backup array and back into
1945 * the in-memory super block. It is meant to help iterate through
1946 * the array, so you send it the number of backups you've already
1947 * tried and the last backup index you used.
1949 * this returns -1 when it has tried all the backups
1951 static noinline int next_root_backup(struct btrfs_fs_info *info,
1952 struct btrfs_super_block *super,
1953 int *num_backups_tried, int *backup_index)
1955 struct btrfs_root_backup *root_backup;
1956 int newest = *backup_index;
1958 if (*num_backups_tried == 0) {
1959 u64 gen = btrfs_super_generation(super);
1961 newest = find_newest_super_backup(info, gen);
1965 *backup_index = newest;
1966 *num_backups_tried = 1;
1967 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1968 /* we've tried all the backups, all done */
1971 /* jump to the next oldest backup */
1972 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1973 BTRFS_NUM_BACKUP_ROOTS;
1974 *backup_index = newest;
1975 *num_backups_tried += 1;
1977 root_backup = super->super_roots + newest;
1979 btrfs_set_super_generation(super,
1980 btrfs_backup_tree_root_gen(root_backup));
1981 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1982 btrfs_set_super_root_level(super,
1983 btrfs_backup_tree_root_level(root_backup));
1984 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1987 * fixme: the total bytes and num_devices need to match or we should
1990 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1991 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1995 /* helper to cleanup workers */
1996 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1998 btrfs_stop_workers(&fs_info->generic_worker);
1999 btrfs_stop_workers(&fs_info->fixup_workers);
2000 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2001 btrfs_destroy_workqueue(fs_info->workers);
2002 btrfs_stop_workers(&fs_info->endio_workers);
2003 btrfs_stop_workers(&fs_info->endio_meta_workers);
2004 btrfs_stop_workers(&fs_info->endio_raid56_workers);
2005 btrfs_stop_workers(&fs_info->rmw_workers);
2006 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2007 btrfs_stop_workers(&fs_info->endio_write_workers);
2008 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2009 btrfs_stop_workers(&fs_info->submit_workers);
2010 btrfs_stop_workers(&fs_info->delayed_workers);
2011 btrfs_stop_workers(&fs_info->caching_workers);
2012 btrfs_stop_workers(&fs_info->readahead_workers);
2013 btrfs_stop_workers(&fs_info->flush_workers);
2014 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
2017 static void free_root_extent_buffers(struct btrfs_root *root)
2020 free_extent_buffer(root->node);
2021 free_extent_buffer(root->commit_root);
2023 root->commit_root = NULL;
2027 /* helper to cleanup tree roots */
2028 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2030 free_root_extent_buffers(info->tree_root);
2032 free_root_extent_buffers(info->dev_root);
2033 free_root_extent_buffers(info->extent_root);
2034 free_root_extent_buffers(info->csum_root);
2035 free_root_extent_buffers(info->quota_root);
2036 free_root_extent_buffers(info->uuid_root);
2038 free_root_extent_buffers(info->chunk_root);
2041 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2044 struct btrfs_root *gang[8];
2047 while (!list_empty(&fs_info->dead_roots)) {
2048 gang[0] = list_entry(fs_info->dead_roots.next,
2049 struct btrfs_root, root_list);
2050 list_del(&gang[0]->root_list);
2052 if (gang[0]->in_radix) {
2053 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2055 free_extent_buffer(gang[0]->node);
2056 free_extent_buffer(gang[0]->commit_root);
2057 btrfs_put_fs_root(gang[0]);
2062 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2067 for (i = 0; i < ret; i++)
2068 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2071 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2072 btrfs_free_log_root_tree(NULL, fs_info);
2073 btrfs_destroy_pinned_extent(fs_info->tree_root,
2074 fs_info->pinned_extents);
2078 int open_ctree(struct super_block *sb,
2079 struct btrfs_fs_devices *fs_devices,
2089 struct btrfs_key location;
2090 struct buffer_head *bh;
2091 struct btrfs_super_block *disk_super;
2092 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2093 struct btrfs_root *tree_root;
2094 struct btrfs_root *extent_root;
2095 struct btrfs_root *csum_root;
2096 struct btrfs_root *chunk_root;
2097 struct btrfs_root *dev_root;
2098 struct btrfs_root *quota_root;
2099 struct btrfs_root *uuid_root;
2100 struct btrfs_root *log_tree_root;
2103 int num_backups_tried = 0;
2104 int backup_index = 0;
2106 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2107 bool create_uuid_tree;
2108 bool check_uuid_tree;
2110 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2111 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2112 if (!tree_root || !chunk_root) {
2117 ret = init_srcu_struct(&fs_info->subvol_srcu);
2123 ret = setup_bdi(fs_info, &fs_info->bdi);
2129 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2134 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2135 (1 + ilog2(nr_cpu_ids));
2137 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2140 goto fail_dirty_metadata_bytes;
2143 ret = percpu_counter_init(&fs_info->bio_counter, 0);
2146 goto fail_delalloc_bytes;
2149 fs_info->btree_inode = new_inode(sb);
2150 if (!fs_info->btree_inode) {
2152 goto fail_bio_counter;
2155 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2157 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2158 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2159 INIT_LIST_HEAD(&fs_info->trans_list);
2160 INIT_LIST_HEAD(&fs_info->dead_roots);
2161 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2162 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2163 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2164 spin_lock_init(&fs_info->delalloc_root_lock);
2165 spin_lock_init(&fs_info->trans_lock);
2166 spin_lock_init(&fs_info->fs_roots_radix_lock);
2167 spin_lock_init(&fs_info->delayed_iput_lock);
2168 spin_lock_init(&fs_info->defrag_inodes_lock);
2169 spin_lock_init(&fs_info->free_chunk_lock);
2170 spin_lock_init(&fs_info->tree_mod_seq_lock);
2171 spin_lock_init(&fs_info->super_lock);
2172 spin_lock_init(&fs_info->buffer_lock);
2173 rwlock_init(&fs_info->tree_mod_log_lock);
2174 mutex_init(&fs_info->reloc_mutex);
2175 seqlock_init(&fs_info->profiles_lock);
2177 init_completion(&fs_info->kobj_unregister);
2178 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2179 INIT_LIST_HEAD(&fs_info->space_info);
2180 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2181 btrfs_mapping_init(&fs_info->mapping_tree);
2182 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2183 BTRFS_BLOCK_RSV_GLOBAL);
2184 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2185 BTRFS_BLOCK_RSV_DELALLOC);
2186 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2187 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2188 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2189 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2190 BTRFS_BLOCK_RSV_DELOPS);
2191 atomic_set(&fs_info->nr_async_submits, 0);
2192 atomic_set(&fs_info->async_delalloc_pages, 0);
2193 atomic_set(&fs_info->async_submit_draining, 0);
2194 atomic_set(&fs_info->nr_async_bios, 0);
2195 atomic_set(&fs_info->defrag_running, 0);
2196 atomic64_set(&fs_info->tree_mod_seq, 0);
2198 fs_info->max_inline = 8192 * 1024;
2199 fs_info->metadata_ratio = 0;
2200 fs_info->defrag_inodes = RB_ROOT;
2201 fs_info->free_chunk_space = 0;
2202 fs_info->tree_mod_log = RB_ROOT;
2203 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2204 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2205 /* readahead state */
2206 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2207 spin_lock_init(&fs_info->reada_lock);
2209 fs_info->thread_pool_size = min_t(unsigned long,
2210 num_online_cpus() + 2, 8);
2212 INIT_LIST_HEAD(&fs_info->ordered_roots);
2213 spin_lock_init(&fs_info->ordered_root_lock);
2214 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2216 if (!fs_info->delayed_root) {
2220 btrfs_init_delayed_root(fs_info->delayed_root);
2222 mutex_init(&fs_info->scrub_lock);
2223 atomic_set(&fs_info->scrubs_running, 0);
2224 atomic_set(&fs_info->scrub_pause_req, 0);
2225 atomic_set(&fs_info->scrubs_paused, 0);
2226 atomic_set(&fs_info->scrub_cancel_req, 0);
2227 init_waitqueue_head(&fs_info->replace_wait);
2228 init_waitqueue_head(&fs_info->scrub_pause_wait);
2229 fs_info->scrub_workers_refcnt = 0;
2230 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2231 fs_info->check_integrity_print_mask = 0;
2234 spin_lock_init(&fs_info->balance_lock);
2235 mutex_init(&fs_info->balance_mutex);
2236 atomic_set(&fs_info->balance_running, 0);
2237 atomic_set(&fs_info->balance_pause_req, 0);
2238 atomic_set(&fs_info->balance_cancel_req, 0);
2239 fs_info->balance_ctl = NULL;
2240 init_waitqueue_head(&fs_info->balance_wait_q);
2242 sb->s_blocksize = 4096;
2243 sb->s_blocksize_bits = blksize_bits(4096);
2244 sb->s_bdi = &fs_info->bdi;
2246 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2247 set_nlink(fs_info->btree_inode, 1);
2249 * we set the i_size on the btree inode to the max possible int.
2250 * the real end of the address space is determined by all of
2251 * the devices in the system
2253 fs_info->btree_inode->i_size = OFFSET_MAX;
2254 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2255 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2257 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2258 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2259 fs_info->btree_inode->i_mapping);
2260 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2261 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2263 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2265 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2266 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2267 sizeof(struct btrfs_key));
2268 set_bit(BTRFS_INODE_DUMMY,
2269 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2270 btrfs_insert_inode_hash(fs_info->btree_inode);
2272 spin_lock_init(&fs_info->block_group_cache_lock);
2273 fs_info->block_group_cache_tree = RB_ROOT;
2274 fs_info->first_logical_byte = (u64)-1;
2276 extent_io_tree_init(&fs_info->freed_extents[0],
2277 fs_info->btree_inode->i_mapping);
2278 extent_io_tree_init(&fs_info->freed_extents[1],
2279 fs_info->btree_inode->i_mapping);
2280 fs_info->pinned_extents = &fs_info->freed_extents[0];
2281 fs_info->do_barriers = 1;
2284 mutex_init(&fs_info->ordered_operations_mutex);
2285 mutex_init(&fs_info->ordered_extent_flush_mutex);
2286 mutex_init(&fs_info->tree_log_mutex);
2287 mutex_init(&fs_info->chunk_mutex);
2288 mutex_init(&fs_info->transaction_kthread_mutex);
2289 mutex_init(&fs_info->cleaner_mutex);
2290 mutex_init(&fs_info->volume_mutex);
2291 init_rwsem(&fs_info->extent_commit_sem);
2292 init_rwsem(&fs_info->cleanup_work_sem);
2293 init_rwsem(&fs_info->subvol_sem);
2294 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2295 fs_info->dev_replace.lock_owner = 0;
2296 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2297 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2298 mutex_init(&fs_info->dev_replace.lock_management_lock);
2299 mutex_init(&fs_info->dev_replace.lock);
2301 spin_lock_init(&fs_info->qgroup_lock);
2302 mutex_init(&fs_info->qgroup_ioctl_lock);
2303 fs_info->qgroup_tree = RB_ROOT;
2304 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2305 fs_info->qgroup_seq = 1;
2306 fs_info->quota_enabled = 0;
2307 fs_info->pending_quota_state = 0;
2308 fs_info->qgroup_ulist = NULL;
2309 mutex_init(&fs_info->qgroup_rescan_lock);
2311 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2312 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2314 init_waitqueue_head(&fs_info->transaction_throttle);
2315 init_waitqueue_head(&fs_info->transaction_wait);
2316 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2317 init_waitqueue_head(&fs_info->async_submit_wait);
2319 ret = btrfs_alloc_stripe_hash_table(fs_info);
2325 __setup_root(4096, 4096, 4096, 4096, tree_root,
2326 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2328 invalidate_bdev(fs_devices->latest_bdev);
2331 * Read super block and check the signature bytes only
2333 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2340 * We want to check superblock checksum, the type is stored inside.
2341 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2343 if (btrfs_check_super_csum(bh->b_data)) {
2344 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2350 * super_copy is zeroed at allocation time and we never touch the
2351 * following bytes up to INFO_SIZE, the checksum is calculated from
2352 * the whole block of INFO_SIZE
2354 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2355 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2356 sizeof(*fs_info->super_for_commit));
2359 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2361 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2363 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2368 disk_super = fs_info->super_copy;
2369 if (!btrfs_super_root(disk_super))
2372 /* check FS state, whether FS is broken. */
2373 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2374 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2377 * run through our array of backup supers and setup
2378 * our ring pointer to the oldest one
2380 generation = btrfs_super_generation(disk_super);
2381 find_oldest_super_backup(fs_info, generation);
2384 * In the long term, we'll store the compression type in the super
2385 * block, and it'll be used for per file compression control.
2387 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2389 ret = btrfs_parse_options(tree_root, options);
2395 features = btrfs_super_incompat_flags(disk_super) &
2396 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2398 printk(KERN_ERR "BTRFS: couldn't mount because of "
2399 "unsupported optional features (%Lx).\n",
2405 if (btrfs_super_leafsize(disk_super) !=
2406 btrfs_super_nodesize(disk_super)) {
2407 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2408 "blocksizes don't match. node %d leaf %d\n",
2409 btrfs_super_nodesize(disk_super),
2410 btrfs_super_leafsize(disk_super));
2414 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2415 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2416 "blocksize (%d) was too large\n",
2417 btrfs_super_leafsize(disk_super));
2422 features = btrfs_super_incompat_flags(disk_super);
2423 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2424 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2425 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2427 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2428 printk(KERN_ERR "BTRFS: has skinny extents\n");
2431 * flag our filesystem as having big metadata blocks if
2432 * they are bigger than the page size
2434 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2435 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2436 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2437 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2440 nodesize = btrfs_super_nodesize(disk_super);
2441 leafsize = btrfs_super_leafsize(disk_super);
2442 sectorsize = btrfs_super_sectorsize(disk_super);
2443 stripesize = btrfs_super_stripesize(disk_super);
2444 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2445 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2448 * mixed block groups end up with duplicate but slightly offset
2449 * extent buffers for the same range. It leads to corruptions
2451 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2452 (sectorsize != leafsize)) {
2453 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2454 "are not allowed for mixed block groups on %s\n",
2460 * Needn't use the lock because there is no other task which will
2463 btrfs_set_super_incompat_flags(disk_super, features);
2465 features = btrfs_super_compat_ro_flags(disk_super) &
2466 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2467 if (!(sb->s_flags & MS_RDONLY) && features) {
2468 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2469 "unsupported option features (%Lx).\n",
2475 max_active = fs_info->thread_pool_size;
2476 btrfs_init_workers(&fs_info->generic_worker,
2477 "genwork", 1, NULL);
2480 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2483 fs_info->delalloc_workers =
2484 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2486 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2487 fs_info->thread_pool_size, NULL);
2489 btrfs_init_workers(&fs_info->submit_workers, "submit",
2490 min_t(u64, fs_devices->num_devices,
2491 fs_info->thread_pool_size), NULL);
2493 btrfs_init_workers(&fs_info->caching_workers, "cache",
2494 fs_info->thread_pool_size, NULL);
2496 /* a higher idle thresh on the submit workers makes it much more
2497 * likely that bios will be send down in a sane order to the
2500 fs_info->submit_workers.idle_thresh = 64;
2502 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2503 &fs_info->generic_worker);
2504 btrfs_init_workers(&fs_info->endio_workers, "endio",
2505 fs_info->thread_pool_size,
2506 &fs_info->generic_worker);
2507 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2508 fs_info->thread_pool_size,
2509 &fs_info->generic_worker);
2510 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2511 "endio-meta-write", fs_info->thread_pool_size,
2512 &fs_info->generic_worker);
2513 btrfs_init_workers(&fs_info->endio_raid56_workers,
2514 "endio-raid56", fs_info->thread_pool_size,
2515 &fs_info->generic_worker);
2516 btrfs_init_workers(&fs_info->rmw_workers,
2517 "rmw", fs_info->thread_pool_size,
2518 &fs_info->generic_worker);
2519 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2520 fs_info->thread_pool_size,
2521 &fs_info->generic_worker);
2522 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2523 1, &fs_info->generic_worker);
2524 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2525 fs_info->thread_pool_size,
2526 &fs_info->generic_worker);
2527 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2528 fs_info->thread_pool_size,
2529 &fs_info->generic_worker);
2530 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2531 &fs_info->generic_worker);
2534 * endios are largely parallel and should have a very
2537 fs_info->endio_workers.idle_thresh = 4;
2538 fs_info->endio_meta_workers.idle_thresh = 4;
2539 fs_info->endio_raid56_workers.idle_thresh = 4;
2540 fs_info->rmw_workers.idle_thresh = 2;
2542 fs_info->endio_write_workers.idle_thresh = 2;
2543 fs_info->endio_meta_write_workers.idle_thresh = 2;
2544 fs_info->readahead_workers.idle_thresh = 2;
2547 * btrfs_start_workers can really only fail because of ENOMEM so just
2548 * return -ENOMEM if any of these fail.
2550 ret = btrfs_start_workers(&fs_info->generic_worker);
2551 ret |= btrfs_start_workers(&fs_info->submit_workers);
2552 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2553 ret |= btrfs_start_workers(&fs_info->endio_workers);
2554 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2555 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2556 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2557 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2558 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2559 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2560 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2561 ret |= btrfs_start_workers(&fs_info->caching_workers);
2562 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2563 ret |= btrfs_start_workers(&fs_info->flush_workers);
2564 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2567 goto fail_sb_buffer;
2569 if (!(fs_info->workers && fs_info->delalloc_workers)) {
2571 goto fail_sb_buffer;
2574 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2575 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2576 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2578 tree_root->nodesize = nodesize;
2579 tree_root->leafsize = leafsize;
2580 tree_root->sectorsize = sectorsize;
2581 tree_root->stripesize = stripesize;
2583 sb->s_blocksize = sectorsize;
2584 sb->s_blocksize_bits = blksize_bits(sectorsize);
2586 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2587 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2588 goto fail_sb_buffer;
2591 if (sectorsize != PAGE_SIZE) {
2592 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2593 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2594 goto fail_sb_buffer;
2597 mutex_lock(&fs_info->chunk_mutex);
2598 ret = btrfs_read_sys_array(tree_root);
2599 mutex_unlock(&fs_info->chunk_mutex);
2601 printk(KERN_WARNING "BTRFS: failed to read the system "
2602 "array on %s\n", sb->s_id);
2603 goto fail_sb_buffer;
2606 blocksize = btrfs_level_size(tree_root,
2607 btrfs_super_chunk_root_level(disk_super));
2608 generation = btrfs_super_chunk_root_generation(disk_super);
2610 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2611 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2613 chunk_root->node = read_tree_block(chunk_root,
2614 btrfs_super_chunk_root(disk_super),
2615 blocksize, generation);
2616 if (!chunk_root->node ||
2617 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2618 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2620 goto fail_tree_roots;
2622 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2623 chunk_root->commit_root = btrfs_root_node(chunk_root);
2625 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2626 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2628 ret = btrfs_read_chunk_tree(chunk_root);
2630 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2632 goto fail_tree_roots;
2636 * keep the device that is marked to be the target device for the
2637 * dev_replace procedure
2639 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2641 if (!fs_devices->latest_bdev) {
2642 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2644 goto fail_tree_roots;
2648 blocksize = btrfs_level_size(tree_root,
2649 btrfs_super_root_level(disk_super));
2650 generation = btrfs_super_generation(disk_super);
2652 tree_root->node = read_tree_block(tree_root,
2653 btrfs_super_root(disk_super),
2654 blocksize, generation);
2655 if (!tree_root->node ||
2656 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2657 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2660 goto recovery_tree_root;
2663 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2664 tree_root->commit_root = btrfs_root_node(tree_root);
2665 btrfs_set_root_refs(&tree_root->root_item, 1);
2667 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2668 location.type = BTRFS_ROOT_ITEM_KEY;
2669 location.offset = 0;
2671 extent_root = btrfs_read_tree_root(tree_root, &location);
2672 if (IS_ERR(extent_root)) {
2673 ret = PTR_ERR(extent_root);
2674 goto recovery_tree_root;
2676 extent_root->track_dirty = 1;
2677 fs_info->extent_root = extent_root;
2679 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2680 dev_root = btrfs_read_tree_root(tree_root, &location);
2681 if (IS_ERR(dev_root)) {
2682 ret = PTR_ERR(dev_root);
2683 goto recovery_tree_root;
2685 dev_root->track_dirty = 1;
2686 fs_info->dev_root = dev_root;
2687 btrfs_init_devices_late(fs_info);
2689 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2690 csum_root = btrfs_read_tree_root(tree_root, &location);
2691 if (IS_ERR(csum_root)) {
2692 ret = PTR_ERR(csum_root);
2693 goto recovery_tree_root;
2695 csum_root->track_dirty = 1;
2696 fs_info->csum_root = csum_root;
2698 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2699 quota_root = btrfs_read_tree_root(tree_root, &location);
2700 if (!IS_ERR(quota_root)) {
2701 quota_root->track_dirty = 1;
2702 fs_info->quota_enabled = 1;
2703 fs_info->pending_quota_state = 1;
2704 fs_info->quota_root = quota_root;
2707 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2708 uuid_root = btrfs_read_tree_root(tree_root, &location);
2709 if (IS_ERR(uuid_root)) {
2710 ret = PTR_ERR(uuid_root);
2712 goto recovery_tree_root;
2713 create_uuid_tree = true;
2714 check_uuid_tree = false;
2716 uuid_root->track_dirty = 1;
2717 fs_info->uuid_root = uuid_root;
2718 create_uuid_tree = false;
2720 generation != btrfs_super_uuid_tree_generation(disk_super);
2723 fs_info->generation = generation;
2724 fs_info->last_trans_committed = generation;
2726 ret = btrfs_recover_balance(fs_info);
2728 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2729 goto fail_block_groups;
2732 ret = btrfs_init_dev_stats(fs_info);
2734 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2736 goto fail_block_groups;
2739 ret = btrfs_init_dev_replace(fs_info);
2741 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2742 goto fail_block_groups;
2745 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2747 ret = btrfs_sysfs_add_one(fs_info);
2749 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2750 goto fail_block_groups;
2753 ret = btrfs_init_space_info(fs_info);
2755 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2759 ret = btrfs_read_block_groups(extent_root);
2761 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2764 fs_info->num_tolerated_disk_barrier_failures =
2765 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2766 if (fs_info->fs_devices->missing_devices >
2767 fs_info->num_tolerated_disk_barrier_failures &&
2768 !(sb->s_flags & MS_RDONLY)) {
2769 printk(KERN_WARNING "BTRFS: "
2770 "too many missing devices, writeable mount is not allowed\n");
2774 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2776 if (IS_ERR(fs_info->cleaner_kthread))
2779 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2781 "btrfs-transaction");
2782 if (IS_ERR(fs_info->transaction_kthread))
2785 if (!btrfs_test_opt(tree_root, SSD) &&
2786 !btrfs_test_opt(tree_root, NOSSD) &&
2787 !fs_info->fs_devices->rotating) {
2788 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2790 btrfs_set_opt(fs_info->mount_opt, SSD);
2793 /* Set the real inode map cache flag */
2794 if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2795 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2797 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2798 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2799 ret = btrfsic_mount(tree_root, fs_devices,
2800 btrfs_test_opt(tree_root,
2801 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2803 fs_info->check_integrity_print_mask);
2805 printk(KERN_WARNING "BTRFS: failed to initialize"
2806 " integrity check module %s\n", sb->s_id);
2809 ret = btrfs_read_qgroup_config(fs_info);
2811 goto fail_trans_kthread;
2813 /* do not make disk changes in broken FS */
2814 if (btrfs_super_log_root(disk_super) != 0) {
2815 u64 bytenr = btrfs_super_log_root(disk_super);
2817 if (fs_devices->rw_devices == 0) {
2818 printk(KERN_WARNING "BTRFS: log replay required "
2824 btrfs_level_size(tree_root,
2825 btrfs_super_log_root_level(disk_super));
2827 log_tree_root = btrfs_alloc_root(fs_info);
2828 if (!log_tree_root) {
2833 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2834 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2836 log_tree_root->node = read_tree_block(tree_root, bytenr,
2839 if (!log_tree_root->node ||
2840 !extent_buffer_uptodate(log_tree_root->node)) {
2841 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2842 free_extent_buffer(log_tree_root->node);
2843 kfree(log_tree_root);
2844 goto fail_trans_kthread;
2846 /* returns with log_tree_root freed on success */
2847 ret = btrfs_recover_log_trees(log_tree_root);
2849 btrfs_error(tree_root->fs_info, ret,
2850 "Failed to recover log tree");
2851 free_extent_buffer(log_tree_root->node);
2852 kfree(log_tree_root);
2853 goto fail_trans_kthread;
2856 if (sb->s_flags & MS_RDONLY) {
2857 ret = btrfs_commit_super(tree_root);
2859 goto fail_trans_kthread;
2863 ret = btrfs_find_orphan_roots(tree_root);
2865 goto fail_trans_kthread;
2867 if (!(sb->s_flags & MS_RDONLY)) {
2868 ret = btrfs_cleanup_fs_roots(fs_info);
2870 goto fail_trans_kthread;
2872 ret = btrfs_recover_relocation(tree_root);
2875 "BTRFS: failed to recover relocation\n");
2881 location.objectid = BTRFS_FS_TREE_OBJECTID;
2882 location.type = BTRFS_ROOT_ITEM_KEY;
2883 location.offset = 0;
2885 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2886 if (IS_ERR(fs_info->fs_root)) {
2887 err = PTR_ERR(fs_info->fs_root);
2891 if (sb->s_flags & MS_RDONLY)
2894 down_read(&fs_info->cleanup_work_sem);
2895 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2896 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2897 up_read(&fs_info->cleanup_work_sem);
2898 close_ctree(tree_root);
2901 up_read(&fs_info->cleanup_work_sem);
2903 ret = btrfs_resume_balance_async(fs_info);
2905 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2906 close_ctree(tree_root);
2910 ret = btrfs_resume_dev_replace_async(fs_info);
2912 pr_warn("BTRFS: failed to resume dev_replace\n");
2913 close_ctree(tree_root);
2917 btrfs_qgroup_rescan_resume(fs_info);
2919 if (create_uuid_tree) {
2920 pr_info("BTRFS: creating UUID tree\n");
2921 ret = btrfs_create_uuid_tree(fs_info);
2923 pr_warn("BTRFS: failed to create the UUID tree %d\n",
2925 close_ctree(tree_root);
2928 } else if (check_uuid_tree ||
2929 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2930 pr_info("BTRFS: checking UUID tree\n");
2931 ret = btrfs_check_uuid_tree(fs_info);
2933 pr_warn("BTRFS: failed to check the UUID tree %d\n",
2935 close_ctree(tree_root);
2939 fs_info->update_uuid_tree_gen = 1;
2945 btrfs_free_qgroup_config(fs_info);
2947 kthread_stop(fs_info->transaction_kthread);
2948 btrfs_cleanup_transaction(fs_info->tree_root);
2949 del_fs_roots(fs_info);
2951 kthread_stop(fs_info->cleaner_kthread);
2954 * make sure we're done with the btree inode before we stop our
2957 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2960 btrfs_sysfs_remove_one(fs_info);
2963 btrfs_put_block_group_cache(fs_info);
2964 btrfs_free_block_groups(fs_info);
2967 free_root_pointers(fs_info, 1);
2968 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2971 btrfs_stop_all_workers(fs_info);
2974 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2976 iput(fs_info->btree_inode);
2978 percpu_counter_destroy(&fs_info->bio_counter);
2979 fail_delalloc_bytes:
2980 percpu_counter_destroy(&fs_info->delalloc_bytes);
2981 fail_dirty_metadata_bytes:
2982 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2984 bdi_destroy(&fs_info->bdi);
2986 cleanup_srcu_struct(&fs_info->subvol_srcu);
2988 btrfs_free_stripe_hash_table(fs_info);
2989 btrfs_close_devices(fs_info->fs_devices);
2993 if (!btrfs_test_opt(tree_root, RECOVERY))
2994 goto fail_tree_roots;
2996 free_root_pointers(fs_info, 0);
2998 /* don't use the log in recovery mode, it won't be valid */
2999 btrfs_set_super_log_root(disk_super, 0);
3001 /* we can't trust the free space cache either */
3002 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3004 ret = next_root_backup(fs_info, fs_info->super_copy,
3005 &num_backups_tried, &backup_index);
3007 goto fail_block_groups;
3008 goto retry_root_backup;
3011 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3014 set_buffer_uptodate(bh);
3016 struct btrfs_device *device = (struct btrfs_device *)
3019 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3020 "I/O error on %s\n",
3021 rcu_str_deref(device->name));
3022 /* note, we dont' set_buffer_write_io_error because we have
3023 * our own ways of dealing with the IO errors
3025 clear_buffer_uptodate(bh);
3026 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3032 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3034 struct buffer_head *bh;
3035 struct buffer_head *latest = NULL;
3036 struct btrfs_super_block *super;
3041 /* we would like to check all the supers, but that would make
3042 * a btrfs mount succeed after a mkfs from a different FS.
3043 * So, we need to add a special mount option to scan for
3044 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3046 for (i = 0; i < 1; i++) {
3047 bytenr = btrfs_sb_offset(i);
3048 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3049 i_size_read(bdev->bd_inode))
3051 bh = __bread(bdev, bytenr / 4096,
3052 BTRFS_SUPER_INFO_SIZE);
3056 super = (struct btrfs_super_block *)bh->b_data;
3057 if (btrfs_super_bytenr(super) != bytenr ||
3058 btrfs_super_magic(super) != BTRFS_MAGIC) {
3063 if (!latest || btrfs_super_generation(super) > transid) {
3066 transid = btrfs_super_generation(super);
3075 * this should be called twice, once with wait == 0 and
3076 * once with wait == 1. When wait == 0 is done, all the buffer heads
3077 * we write are pinned.
3079 * They are released when wait == 1 is done.
3080 * max_mirrors must be the same for both runs, and it indicates how
3081 * many supers on this one device should be written.
3083 * max_mirrors == 0 means to write them all.
3085 static int write_dev_supers(struct btrfs_device *device,
3086 struct btrfs_super_block *sb,
3087 int do_barriers, int wait, int max_mirrors)
3089 struct buffer_head *bh;
3096 if (max_mirrors == 0)
3097 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3099 for (i = 0; i < max_mirrors; i++) {
3100 bytenr = btrfs_sb_offset(i);
3101 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3105 bh = __find_get_block(device->bdev, bytenr / 4096,
3106 BTRFS_SUPER_INFO_SIZE);
3112 if (!buffer_uptodate(bh))
3115 /* drop our reference */
3118 /* drop the reference from the wait == 0 run */
3122 btrfs_set_super_bytenr(sb, bytenr);
3125 crc = btrfs_csum_data((char *)sb +
3126 BTRFS_CSUM_SIZE, crc,
3127 BTRFS_SUPER_INFO_SIZE -
3129 btrfs_csum_final(crc, sb->csum);
3132 * one reference for us, and we leave it for the
3135 bh = __getblk(device->bdev, bytenr / 4096,
3136 BTRFS_SUPER_INFO_SIZE);
3138 printk(KERN_ERR "BTRFS: couldn't get super "
3139 "buffer head for bytenr %Lu\n", bytenr);
3144 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3146 /* one reference for submit_bh */
3149 set_buffer_uptodate(bh);
3151 bh->b_end_io = btrfs_end_buffer_write_sync;
3152 bh->b_private = device;
3156 * we fua the first super. The others we allow
3160 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3162 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3166 return errors < i ? 0 : -1;
3170 * endio for the write_dev_flush, this will wake anyone waiting
3171 * for the barrier when it is done
3173 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3176 if (err == -EOPNOTSUPP)
3177 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3178 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3180 if (bio->bi_private)
3181 complete(bio->bi_private);
3186 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3187 * sent down. With wait == 1, it waits for the previous flush.
3189 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3192 static int write_dev_flush(struct btrfs_device *device, int wait)
3197 if (device->nobarriers)
3201 bio = device->flush_bio;
3205 wait_for_completion(&device->flush_wait);
3207 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3208 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3209 rcu_str_deref(device->name));
3210 device->nobarriers = 1;
3211 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3213 btrfs_dev_stat_inc_and_print(device,
3214 BTRFS_DEV_STAT_FLUSH_ERRS);
3217 /* drop the reference from the wait == 0 run */
3219 device->flush_bio = NULL;
3225 * one reference for us, and we leave it for the
3228 device->flush_bio = NULL;
3229 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3233 bio->bi_end_io = btrfs_end_empty_barrier;
3234 bio->bi_bdev = device->bdev;
3235 init_completion(&device->flush_wait);
3236 bio->bi_private = &device->flush_wait;
3237 device->flush_bio = bio;
3240 btrfsic_submit_bio(WRITE_FLUSH, bio);
3246 * send an empty flush down to each device in parallel,
3247 * then wait for them
3249 static int barrier_all_devices(struct btrfs_fs_info *info)
3251 struct list_head *head;
3252 struct btrfs_device *dev;
3253 int errors_send = 0;
3254 int errors_wait = 0;
3257 /* send down all the barriers */
3258 head = &info->fs_devices->devices;
3259 list_for_each_entry_rcu(dev, head, dev_list) {
3266 if (!dev->in_fs_metadata || !dev->writeable)
3269 ret = write_dev_flush(dev, 0);
3274 /* wait for all the barriers */
3275 list_for_each_entry_rcu(dev, head, dev_list) {
3282 if (!dev->in_fs_metadata || !dev->writeable)
3285 ret = write_dev_flush(dev, 1);
3289 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3290 errors_wait > info->num_tolerated_disk_barrier_failures)
3295 int btrfs_calc_num_tolerated_disk_barrier_failures(
3296 struct btrfs_fs_info *fs_info)
3298 struct btrfs_ioctl_space_info space;
3299 struct btrfs_space_info *sinfo;
3300 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3301 BTRFS_BLOCK_GROUP_SYSTEM,
3302 BTRFS_BLOCK_GROUP_METADATA,
3303 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3307 int num_tolerated_disk_barrier_failures =
3308 (int)fs_info->fs_devices->num_devices;
3310 for (i = 0; i < num_types; i++) {
3311 struct btrfs_space_info *tmp;
3315 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3316 if (tmp->flags == types[i]) {
3326 down_read(&sinfo->groups_sem);
3327 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3328 if (!list_empty(&sinfo->block_groups[c])) {
3331 btrfs_get_block_group_info(
3332 &sinfo->block_groups[c], &space);
3333 if (space.total_bytes == 0 ||
3334 space.used_bytes == 0)
3336 flags = space.flags;
3339 * 0: if dup, single or RAID0 is configured for
3340 * any of metadata, system or data, else
3341 * 1: if RAID5 is configured, or if RAID1 or
3342 * RAID10 is configured and only two mirrors
3344 * 2: if RAID6 is configured, else
3345 * num_mirrors - 1: if RAID1 or RAID10 is
3346 * configured and more than
3347 * 2 mirrors are used.
3349 if (num_tolerated_disk_barrier_failures > 0 &&
3350 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3351 BTRFS_BLOCK_GROUP_RAID0)) ||
3352 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3354 num_tolerated_disk_barrier_failures = 0;
3355 else if (num_tolerated_disk_barrier_failures > 1) {
3356 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3357 BTRFS_BLOCK_GROUP_RAID5 |
3358 BTRFS_BLOCK_GROUP_RAID10)) {
3359 num_tolerated_disk_barrier_failures = 1;
3361 BTRFS_BLOCK_GROUP_RAID6) {
3362 num_tolerated_disk_barrier_failures = 2;
3367 up_read(&sinfo->groups_sem);
3370 return num_tolerated_disk_barrier_failures;
3373 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3375 struct list_head *head;
3376 struct btrfs_device *dev;
3377 struct btrfs_super_block *sb;
3378 struct btrfs_dev_item *dev_item;
3382 int total_errors = 0;
3385 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3386 backup_super_roots(root->fs_info);
3388 sb = root->fs_info->super_for_commit;
3389 dev_item = &sb->dev_item;
3391 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3392 head = &root->fs_info->fs_devices->devices;
3393 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3396 ret = barrier_all_devices(root->fs_info);
3399 &root->fs_info->fs_devices->device_list_mutex);
3400 btrfs_error(root->fs_info, ret,
3401 "errors while submitting device barriers.");
3406 list_for_each_entry_rcu(dev, head, dev_list) {
3411 if (!dev->in_fs_metadata || !dev->writeable)
3414 btrfs_set_stack_device_generation(dev_item, 0);
3415 btrfs_set_stack_device_type(dev_item, dev->type);
3416 btrfs_set_stack_device_id(dev_item, dev->devid);
3417 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3418 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3419 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3420 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3421 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3422 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3423 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3425 flags = btrfs_super_flags(sb);
3426 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3428 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3432 if (total_errors > max_errors) {
3433 btrfs_err(root->fs_info, "%d errors while writing supers",
3435 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3437 /* FUA is masked off if unsupported and can't be the reason */
3438 btrfs_error(root->fs_info, -EIO,
3439 "%d errors while writing supers", total_errors);
3444 list_for_each_entry_rcu(dev, head, dev_list) {
3447 if (!dev->in_fs_metadata || !dev->writeable)
3450 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3454 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3455 if (total_errors > max_errors) {
3456 btrfs_error(root->fs_info, -EIO,
3457 "%d errors while writing supers", total_errors);
3463 int write_ctree_super(struct btrfs_trans_handle *trans,
3464 struct btrfs_root *root, int max_mirrors)
3466 return write_all_supers(root, max_mirrors);
3469 /* Drop a fs root from the radix tree and free it. */
3470 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3471 struct btrfs_root *root)
3473 spin_lock(&fs_info->fs_roots_radix_lock);
3474 radix_tree_delete(&fs_info->fs_roots_radix,
3475 (unsigned long)root->root_key.objectid);
3476 spin_unlock(&fs_info->fs_roots_radix_lock);
3478 if (btrfs_root_refs(&root->root_item) == 0)
3479 synchronize_srcu(&fs_info->subvol_srcu);
3481 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3482 btrfs_free_log(NULL, root);
3484 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3485 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3489 static void free_fs_root(struct btrfs_root *root)
3491 iput(root->cache_inode);
3492 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3493 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3494 root->orphan_block_rsv = NULL;
3496 free_anon_bdev(root->anon_dev);
3497 free_extent_buffer(root->node);
3498 free_extent_buffer(root->commit_root);
3499 kfree(root->free_ino_ctl);
3500 kfree(root->free_ino_pinned);
3502 btrfs_put_fs_root(root);
3505 void btrfs_free_fs_root(struct btrfs_root *root)
3510 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3512 u64 root_objectid = 0;
3513 struct btrfs_root *gang[8];
3518 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3519 (void **)gang, root_objectid,
3524 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3525 for (i = 0; i < ret; i++) {
3528 root_objectid = gang[i]->root_key.objectid;
3529 err = btrfs_orphan_cleanup(gang[i]);
3538 int btrfs_commit_super(struct btrfs_root *root)
3540 struct btrfs_trans_handle *trans;
3542 mutex_lock(&root->fs_info->cleaner_mutex);
3543 btrfs_run_delayed_iputs(root);
3544 mutex_unlock(&root->fs_info->cleaner_mutex);
3545 wake_up_process(root->fs_info->cleaner_kthread);
3547 /* wait until ongoing cleanup work done */
3548 down_write(&root->fs_info->cleanup_work_sem);
3549 up_write(&root->fs_info->cleanup_work_sem);
3551 trans = btrfs_join_transaction(root);
3553 return PTR_ERR(trans);
3554 return btrfs_commit_transaction(trans, root);
3557 int close_ctree(struct btrfs_root *root)
3559 struct btrfs_fs_info *fs_info = root->fs_info;
3562 fs_info->closing = 1;
3565 /* wait for the uuid_scan task to finish */
3566 down(&fs_info->uuid_tree_rescan_sem);
3567 /* avoid complains from lockdep et al., set sem back to initial state */
3568 up(&fs_info->uuid_tree_rescan_sem);
3570 /* pause restriper - we want to resume on mount */
3571 btrfs_pause_balance(fs_info);
3573 btrfs_dev_replace_suspend_for_unmount(fs_info);
3575 btrfs_scrub_cancel(fs_info);
3577 /* wait for any defraggers to finish */
3578 wait_event(fs_info->transaction_wait,
3579 (atomic_read(&fs_info->defrag_running) == 0));
3581 /* clear out the rbtree of defraggable inodes */
3582 btrfs_cleanup_defrag_inodes(fs_info);
3584 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3585 ret = btrfs_commit_super(root);
3587 btrfs_err(root->fs_info, "commit super ret %d", ret);
3590 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3591 btrfs_error_commit_super(root);
3593 kthread_stop(fs_info->transaction_kthread);
3594 kthread_stop(fs_info->cleaner_kthread);
3596 fs_info->closing = 2;
3599 btrfs_free_qgroup_config(root->fs_info);
3601 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3602 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3603 percpu_counter_sum(&fs_info->delalloc_bytes));
3606 btrfs_sysfs_remove_one(fs_info);
3608 del_fs_roots(fs_info);
3610 btrfs_put_block_group_cache(fs_info);
3612 btrfs_free_block_groups(fs_info);
3614 btrfs_stop_all_workers(fs_info);
3616 free_root_pointers(fs_info, 1);
3618 iput(fs_info->btree_inode);
3620 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3621 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3622 btrfsic_unmount(root, fs_info->fs_devices);
3625 btrfs_close_devices(fs_info->fs_devices);
3626 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3628 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3629 percpu_counter_destroy(&fs_info->delalloc_bytes);
3630 percpu_counter_destroy(&fs_info->bio_counter);
3631 bdi_destroy(&fs_info->bdi);
3632 cleanup_srcu_struct(&fs_info->subvol_srcu);
3634 btrfs_free_stripe_hash_table(fs_info);
3636 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3637 root->orphan_block_rsv = NULL;
3642 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3646 struct inode *btree_inode = buf->pages[0]->mapping->host;
3648 ret = extent_buffer_uptodate(buf);
3652 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3653 parent_transid, atomic);
3659 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3661 return set_extent_buffer_uptodate(buf);
3664 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3666 struct btrfs_root *root;
3667 u64 transid = btrfs_header_generation(buf);
3670 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3672 * This is a fast path so only do this check if we have sanity tests
3673 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3674 * outside of the sanity tests.
3676 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3679 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3680 btrfs_assert_tree_locked(buf);
3681 if (transid != root->fs_info->generation)
3682 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3683 "found %llu running %llu\n",
3684 buf->start, transid, root->fs_info->generation);
3685 was_dirty = set_extent_buffer_dirty(buf);
3687 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3689 root->fs_info->dirty_metadata_batch);
3692 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3696 * looks as though older kernels can get into trouble with
3697 * this code, they end up stuck in balance_dirty_pages forever
3701 if (current->flags & PF_MEMALLOC)
3705 btrfs_balance_delayed_items(root);
3707 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3708 BTRFS_DIRTY_METADATA_THRESH);
3710 balance_dirty_pages_ratelimited(
3711 root->fs_info->btree_inode->i_mapping);
3716 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3718 __btrfs_btree_balance_dirty(root, 1);
3721 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3723 __btrfs_btree_balance_dirty(root, 0);
3726 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3728 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3729 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3732 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3736 * Placeholder for checks
3741 static void btrfs_error_commit_super(struct btrfs_root *root)
3743 mutex_lock(&root->fs_info->cleaner_mutex);
3744 btrfs_run_delayed_iputs(root);
3745 mutex_unlock(&root->fs_info->cleaner_mutex);
3747 down_write(&root->fs_info->cleanup_work_sem);
3748 up_write(&root->fs_info->cleanup_work_sem);
3750 /* cleanup FS via transaction */
3751 btrfs_cleanup_transaction(root);
3754 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3755 struct btrfs_root *root)
3757 struct btrfs_inode *btrfs_inode;
3758 struct list_head splice;
3760 INIT_LIST_HEAD(&splice);
3762 mutex_lock(&root->fs_info->ordered_operations_mutex);
3763 spin_lock(&root->fs_info->ordered_root_lock);
3765 list_splice_init(&t->ordered_operations, &splice);
3766 while (!list_empty(&splice)) {
3767 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3768 ordered_operations);
3770 list_del_init(&btrfs_inode->ordered_operations);
3771 spin_unlock(&root->fs_info->ordered_root_lock);
3773 btrfs_invalidate_inodes(btrfs_inode->root);
3775 spin_lock(&root->fs_info->ordered_root_lock);
3778 spin_unlock(&root->fs_info->ordered_root_lock);
3779 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3782 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3784 struct btrfs_ordered_extent *ordered;
3786 spin_lock(&root->ordered_extent_lock);
3788 * This will just short circuit the ordered completion stuff which will
3789 * make sure the ordered extent gets properly cleaned up.
3791 list_for_each_entry(ordered, &root->ordered_extents,
3793 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3794 spin_unlock(&root->ordered_extent_lock);
3797 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3799 struct btrfs_root *root;
3800 struct list_head splice;
3802 INIT_LIST_HEAD(&splice);
3804 spin_lock(&fs_info->ordered_root_lock);
3805 list_splice_init(&fs_info->ordered_roots, &splice);
3806 while (!list_empty(&splice)) {
3807 root = list_first_entry(&splice, struct btrfs_root,
3809 list_move_tail(&root->ordered_root,
3810 &fs_info->ordered_roots);
3812 spin_unlock(&fs_info->ordered_root_lock);
3813 btrfs_destroy_ordered_extents(root);
3816 spin_lock(&fs_info->ordered_root_lock);
3818 spin_unlock(&fs_info->ordered_root_lock);
3821 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3822 struct btrfs_root *root)
3824 struct rb_node *node;
3825 struct btrfs_delayed_ref_root *delayed_refs;
3826 struct btrfs_delayed_ref_node *ref;
3829 delayed_refs = &trans->delayed_refs;
3831 spin_lock(&delayed_refs->lock);
3832 if (atomic_read(&delayed_refs->num_entries) == 0) {
3833 spin_unlock(&delayed_refs->lock);
3834 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3838 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3839 struct btrfs_delayed_ref_head *head;
3840 bool pin_bytes = false;
3842 head = rb_entry(node, struct btrfs_delayed_ref_head,
3844 if (!mutex_trylock(&head->mutex)) {
3845 atomic_inc(&head->node.refs);
3846 spin_unlock(&delayed_refs->lock);
3848 mutex_lock(&head->mutex);
3849 mutex_unlock(&head->mutex);
3850 btrfs_put_delayed_ref(&head->node);
3851 spin_lock(&delayed_refs->lock);
3854 spin_lock(&head->lock);
3855 while ((node = rb_first(&head->ref_root)) != NULL) {
3856 ref = rb_entry(node, struct btrfs_delayed_ref_node,
3859 rb_erase(&ref->rb_node, &head->ref_root);
3860 atomic_dec(&delayed_refs->num_entries);
3861 btrfs_put_delayed_ref(ref);
3863 if (head->must_insert_reserved)
3865 btrfs_free_delayed_extent_op(head->extent_op);
3866 delayed_refs->num_heads--;
3867 if (head->processing == 0)
3868 delayed_refs->num_heads_ready--;
3869 atomic_dec(&delayed_refs->num_entries);
3870 head->node.in_tree = 0;
3871 rb_erase(&head->href_node, &delayed_refs->href_root);
3872 spin_unlock(&head->lock);
3873 spin_unlock(&delayed_refs->lock);
3874 mutex_unlock(&head->mutex);
3877 btrfs_pin_extent(root, head->node.bytenr,
3878 head->node.num_bytes, 1);
3879 btrfs_put_delayed_ref(&head->node);
3881 spin_lock(&delayed_refs->lock);
3884 spin_unlock(&delayed_refs->lock);
3889 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3891 struct btrfs_inode *btrfs_inode;
3892 struct list_head splice;
3894 INIT_LIST_HEAD(&splice);
3896 spin_lock(&root->delalloc_lock);
3897 list_splice_init(&root->delalloc_inodes, &splice);
3899 while (!list_empty(&splice)) {
3900 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3903 list_del_init(&btrfs_inode->delalloc_inodes);
3904 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3905 &btrfs_inode->runtime_flags);
3906 spin_unlock(&root->delalloc_lock);
3908 btrfs_invalidate_inodes(btrfs_inode->root);
3910 spin_lock(&root->delalloc_lock);
3913 spin_unlock(&root->delalloc_lock);
3916 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3918 struct btrfs_root *root;
3919 struct list_head splice;
3921 INIT_LIST_HEAD(&splice);
3923 spin_lock(&fs_info->delalloc_root_lock);
3924 list_splice_init(&fs_info->delalloc_roots, &splice);
3925 while (!list_empty(&splice)) {
3926 root = list_first_entry(&splice, struct btrfs_root,
3928 list_del_init(&root->delalloc_root);
3929 root = btrfs_grab_fs_root(root);
3931 spin_unlock(&fs_info->delalloc_root_lock);
3933 btrfs_destroy_delalloc_inodes(root);
3934 btrfs_put_fs_root(root);
3936 spin_lock(&fs_info->delalloc_root_lock);
3938 spin_unlock(&fs_info->delalloc_root_lock);
3941 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3942 struct extent_io_tree *dirty_pages,
3946 struct extent_buffer *eb;
3951 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3956 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3957 while (start <= end) {
3958 eb = btrfs_find_tree_block(root, start,
3960 start += root->leafsize;
3963 wait_on_extent_buffer_writeback(eb);
3965 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3967 clear_extent_buffer_dirty(eb);
3968 free_extent_buffer_stale(eb);
3975 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3976 struct extent_io_tree *pinned_extents)
3978 struct extent_io_tree *unpin;
3984 unpin = pinned_extents;
3987 ret = find_first_extent_bit(unpin, 0, &start, &end,
3988 EXTENT_DIRTY, NULL);
3993 if (btrfs_test_opt(root, DISCARD))
3994 ret = btrfs_error_discard_extent(root, start,
3998 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3999 btrfs_error_unpin_extent_range(root, start, end);
4004 if (unpin == &root->fs_info->freed_extents[0])
4005 unpin = &root->fs_info->freed_extents[1];
4007 unpin = &root->fs_info->freed_extents[0];
4015 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4016 struct btrfs_root *root)
4018 btrfs_destroy_ordered_operations(cur_trans, root);
4020 btrfs_destroy_delayed_refs(cur_trans, root);
4022 cur_trans->state = TRANS_STATE_COMMIT_START;
4023 wake_up(&root->fs_info->transaction_blocked_wait);
4025 cur_trans->state = TRANS_STATE_UNBLOCKED;
4026 wake_up(&root->fs_info->transaction_wait);
4028 btrfs_destroy_delayed_inodes(root);
4029 btrfs_assert_delayed_root_empty(root);
4031 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4033 btrfs_destroy_pinned_extent(root,
4034 root->fs_info->pinned_extents);
4036 cur_trans->state =TRANS_STATE_COMPLETED;
4037 wake_up(&cur_trans->commit_wait);
4040 memset(cur_trans, 0, sizeof(*cur_trans));
4041 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4045 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4047 struct btrfs_transaction *t;
4049 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4051 spin_lock(&root->fs_info->trans_lock);
4052 while (!list_empty(&root->fs_info->trans_list)) {
4053 t = list_first_entry(&root->fs_info->trans_list,
4054 struct btrfs_transaction, list);
4055 if (t->state >= TRANS_STATE_COMMIT_START) {
4056 atomic_inc(&t->use_count);
4057 spin_unlock(&root->fs_info->trans_lock);
4058 btrfs_wait_for_commit(root, t->transid);
4059 btrfs_put_transaction(t);
4060 spin_lock(&root->fs_info->trans_lock);
4063 if (t == root->fs_info->running_transaction) {
4064 t->state = TRANS_STATE_COMMIT_DOING;
4065 spin_unlock(&root->fs_info->trans_lock);
4067 * We wait for 0 num_writers since we don't hold a trans
4068 * handle open currently for this transaction.
4070 wait_event(t->writer_wait,
4071 atomic_read(&t->num_writers) == 0);
4073 spin_unlock(&root->fs_info->trans_lock);
4075 btrfs_cleanup_one_transaction(t, root);
4077 spin_lock(&root->fs_info->trans_lock);
4078 if (t == root->fs_info->running_transaction)
4079 root->fs_info->running_transaction = NULL;
4080 list_del_init(&t->list);
4081 spin_unlock(&root->fs_info->trans_lock);
4083 btrfs_put_transaction(t);
4084 trace_btrfs_transaction_commit(root);
4085 spin_lock(&root->fs_info->trans_lock);
4087 spin_unlock(&root->fs_info->trans_lock);
4088 btrfs_destroy_all_ordered_extents(root->fs_info);
4089 btrfs_destroy_delayed_inodes(root);
4090 btrfs_assert_delayed_root_empty(root);
4091 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4092 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4093 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4098 static struct extent_io_ops btree_extent_io_ops = {
4099 .readpage_end_io_hook = btree_readpage_end_io_hook,
4100 .readpage_io_failed_hook = btree_io_failed_hook,
4101 .submit_bio_hook = btree_submit_bio_hook,
4102 /* note we're sharing with inode.c for the merge bio hook */
4103 .merge_bio_hook = btrfs_merge_bio_hook,