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"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
54 #include <asm/cpufeature.h>
57 static const 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_extents(struct btrfs_root *root);
63 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
64 struct btrfs_root *root);
65 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
66 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
67 struct extent_io_tree *dirty_pages,
69 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
70 struct extent_io_tree *pinned_extents);
71 static int btrfs_cleanup_transaction(struct btrfs_root *root);
72 static void btrfs_error_commit_super(struct btrfs_root *root);
75 * btrfs_end_io_wq structs are used to do processing in task context when an IO
76 * is complete. This is used during reads to verify checksums, and it is used
77 * by writes to insert metadata for new file extents after IO is complete.
79 struct btrfs_end_io_wq {
83 struct btrfs_fs_info *info;
85 enum btrfs_wq_endio_type metadata;
86 struct list_head list;
87 struct btrfs_work work;
90 static struct kmem_cache *btrfs_end_io_wq_cache;
92 int __init btrfs_end_io_wq_init(void)
94 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
95 sizeof(struct btrfs_end_io_wq),
97 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
99 if (!btrfs_end_io_wq_cache)
104 void btrfs_end_io_wq_exit(void)
106 if (btrfs_end_io_wq_cache)
107 kmem_cache_destroy(btrfs_end_io_wq_cache);
111 * async submit bios are used to offload expensive checksumming
112 * onto the worker threads. They checksum file and metadata bios
113 * just before they are sent down the IO stack.
115 struct async_submit_bio {
118 struct list_head list;
119 extent_submit_bio_hook_t *submit_bio_start;
120 extent_submit_bio_hook_t *submit_bio_done;
123 unsigned long bio_flags;
125 * bio_offset is optional, can be used if the pages in the bio
126 * can't tell us where in the file the bio should go
129 struct btrfs_work work;
134 * Lockdep class keys for extent_buffer->lock's in this root. For a given
135 * eb, the lockdep key is determined by the btrfs_root it belongs to and
136 * the level the eb occupies in the tree.
138 * Different roots are used for different purposes and may nest inside each
139 * other and they require separate keysets. As lockdep keys should be
140 * static, assign keysets according to the purpose of the root as indicated
141 * by btrfs_root->objectid. This ensures that all special purpose roots
142 * have separate keysets.
144 * Lock-nesting across peer nodes is always done with the immediate parent
145 * node locked thus preventing deadlock. As lockdep doesn't know this, use
146 * subclass to avoid triggering lockdep warning in such cases.
148 * The key is set by the readpage_end_io_hook after the buffer has passed
149 * csum validation but before the pages are unlocked. It is also set by
150 * btrfs_init_new_buffer on freshly allocated blocks.
152 * We also add a check to make sure the highest level of the tree is the
153 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
154 * needs update as well.
156 #ifdef CONFIG_DEBUG_LOCK_ALLOC
157 # if BTRFS_MAX_LEVEL != 8
161 static struct btrfs_lockdep_keyset {
162 u64 id; /* root objectid */
163 const char *name_stem; /* lock name stem */
164 char names[BTRFS_MAX_LEVEL + 1][20];
165 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
166 } btrfs_lockdep_keysets[] = {
167 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
168 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
169 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
170 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
171 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
172 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
173 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
174 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
175 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
176 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
177 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
178 { .id = 0, .name_stem = "tree" },
181 void __init btrfs_init_lockdep(void)
185 /* initialize lockdep class names */
186 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
187 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
189 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
190 snprintf(ks->names[j], sizeof(ks->names[j]),
191 "btrfs-%s-%02d", ks->name_stem, j);
195 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
198 struct btrfs_lockdep_keyset *ks;
200 BUG_ON(level >= ARRAY_SIZE(ks->keys));
202 /* find the matching keyset, id 0 is the default entry */
203 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
204 if (ks->id == objectid)
207 lockdep_set_class_and_name(&eb->lock,
208 &ks->keys[level], ks->names[level]);
214 * extents on the btree inode are pretty simple, there's one extent
215 * that covers the entire device
217 static struct extent_map *btree_get_extent(struct inode *inode,
218 struct page *page, size_t pg_offset, u64 start, u64 len,
221 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
222 struct extent_map *em;
225 read_lock(&em_tree->lock);
226 em = lookup_extent_mapping(em_tree, start, len);
229 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
230 read_unlock(&em_tree->lock);
233 read_unlock(&em_tree->lock);
235 em = alloc_extent_map();
237 em = ERR_PTR(-ENOMEM);
242 em->block_len = (u64)-1;
244 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
246 write_lock(&em_tree->lock);
247 ret = add_extent_mapping(em_tree, em, 0);
248 if (ret == -EEXIST) {
250 em = lookup_extent_mapping(em_tree, start, len);
257 write_unlock(&em_tree->lock);
263 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
265 return btrfs_crc32c(seed, data, len);
268 void btrfs_csum_final(u32 crc, char *result)
270 put_unaligned_le32(~crc, result);
274 * compute the csum for a btree block, and either verify it or write it
275 * into the csum field of the block.
277 static int csum_tree_block(struct btrfs_fs_info *fs_info,
278 struct extent_buffer *buf,
281 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
284 unsigned long cur_len;
285 unsigned long offset = BTRFS_CSUM_SIZE;
287 unsigned long map_start;
288 unsigned long map_len;
291 unsigned long inline_result;
293 len = buf->len - offset;
295 err = map_private_extent_buffer(buf, offset, 32,
296 &kaddr, &map_start, &map_len);
299 cur_len = min(len, map_len - (offset - map_start));
300 crc = btrfs_csum_data(kaddr + offset - map_start,
305 if (csum_size > sizeof(inline_result)) {
306 result = kzalloc(csum_size, GFP_NOFS);
310 result = (char *)&inline_result;
313 btrfs_csum_final(crc, result);
316 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
319 memcpy(&found, result, csum_size);
321 read_extent_buffer(buf, &val, 0, csum_size);
322 btrfs_warn_rl(fs_info,
323 "%s checksum verify failed on %llu wanted %X found %X "
325 fs_info->sb->s_id, buf->start,
326 val, found, btrfs_header_level(buf));
327 if (result != (char *)&inline_result)
332 write_extent_buffer(buf, result, 0, csum_size);
334 if (result != (char *)&inline_result)
340 * we can't consider a given block up to date unless the transid of the
341 * block matches the transid in the parent node's pointer. This is how we
342 * detect blocks that either didn't get written at all or got written
343 * in the wrong place.
345 static int verify_parent_transid(struct extent_io_tree *io_tree,
346 struct extent_buffer *eb, u64 parent_transid,
349 struct extent_state *cached_state = NULL;
351 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
353 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
360 btrfs_tree_read_lock(eb);
361 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
364 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
366 if (extent_buffer_uptodate(eb) &&
367 btrfs_header_generation(eb) == parent_transid) {
371 btrfs_err_rl(eb->fs_info,
372 "parent transid verify failed on %llu wanted %llu found %llu",
374 parent_transid, btrfs_header_generation(eb));
378 * Things reading via commit roots that don't have normal protection,
379 * like send, can have a really old block in cache that may point at a
380 * block that has been free'd and re-allocated. So don't clear uptodate
381 * if we find an eb that is under IO (dirty/writeback) because we could
382 * end up reading in the stale data and then writing it back out and
383 * making everybody very sad.
385 if (!extent_buffer_under_io(eb))
386 clear_extent_buffer_uptodate(eb);
388 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
389 &cached_state, GFP_NOFS);
391 btrfs_tree_read_unlock_blocking(eb);
396 * Return 0 if the superblock checksum type matches the checksum value of that
397 * algorithm. Pass the raw disk superblock data.
399 static int btrfs_check_super_csum(char *raw_disk_sb)
401 struct btrfs_super_block *disk_sb =
402 (struct btrfs_super_block *)raw_disk_sb;
403 u16 csum_type = btrfs_super_csum_type(disk_sb);
406 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
408 const int csum_size = sizeof(crc);
409 char result[csum_size];
412 * The super_block structure does not span the whole
413 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
414 * is filled with zeros and is included in the checkum.
416 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
417 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
418 btrfs_csum_final(crc, result);
420 if (memcmp(raw_disk_sb, result, csum_size))
424 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
425 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
434 * helper to read a given tree block, doing retries as required when
435 * the checksums don't match and we have alternate mirrors to try.
437 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
438 struct extent_buffer *eb,
439 u64 start, u64 parent_transid)
441 struct extent_io_tree *io_tree;
446 int failed_mirror = 0;
448 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
449 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
451 ret = read_extent_buffer_pages(io_tree, eb, start,
453 btree_get_extent, mirror_num);
455 if (!verify_parent_transid(io_tree, eb,
463 * This buffer's crc is fine, but its contents are corrupted, so
464 * there is no reason to read the other copies, they won't be
467 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
470 num_copies = btrfs_num_copies(root->fs_info,
475 if (!failed_mirror) {
477 failed_mirror = eb->read_mirror;
481 if (mirror_num == failed_mirror)
484 if (mirror_num > num_copies)
488 if (failed && !ret && failed_mirror)
489 repair_eb_io_failure(root, eb, failed_mirror);
495 * checksum a dirty tree block before IO. This has extra checks to make sure
496 * we only fill in the checksum field in the first page of a multi-page block
499 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
501 u64 start = page_offset(page);
503 struct extent_buffer *eb;
505 eb = (struct extent_buffer *)page->private;
506 if (page != eb->pages[0])
508 found_start = btrfs_header_bytenr(eb);
509 if (WARN_ON(found_start != start || !PageUptodate(page)))
511 csum_tree_block(fs_info, eb, 0);
515 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
516 struct extent_buffer *eb)
518 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
519 u8 fsid[BTRFS_UUID_SIZE];
522 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
524 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
528 fs_devices = fs_devices->seed;
533 #define CORRUPT(reason, eb, root, slot) \
534 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
535 "root=%llu, slot=%d", reason, \
536 btrfs_header_bytenr(eb), root->objectid, slot)
538 static noinline int check_leaf(struct btrfs_root *root,
539 struct extent_buffer *leaf)
541 struct btrfs_key key;
542 struct btrfs_key leaf_key;
543 u32 nritems = btrfs_header_nritems(leaf);
549 /* Check the 0 item */
550 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
551 BTRFS_LEAF_DATA_SIZE(root)) {
552 CORRUPT("invalid item offset size pair", leaf, root, 0);
557 * Check to make sure each items keys are in the correct order and their
558 * offsets make sense. We only have to loop through nritems-1 because
559 * we check the current slot against the next slot, which verifies the
560 * next slot's offset+size makes sense and that the current's slot
563 for (slot = 0; slot < nritems - 1; slot++) {
564 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
565 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
567 /* Make sure the keys are in the right order */
568 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
569 CORRUPT("bad key order", leaf, root, slot);
574 * Make sure the offset and ends are right, remember that the
575 * item data starts at the end of the leaf and grows towards the
578 if (btrfs_item_offset_nr(leaf, slot) !=
579 btrfs_item_end_nr(leaf, slot + 1)) {
580 CORRUPT("slot offset bad", leaf, root, slot);
585 * Check to make sure that we don't point outside of the leaf,
586 * just incase all the items are consistent to eachother, but
587 * all point outside of the leaf.
589 if (btrfs_item_end_nr(leaf, slot) >
590 BTRFS_LEAF_DATA_SIZE(root)) {
591 CORRUPT("slot end outside of leaf", leaf, root, slot);
599 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
600 u64 phy_offset, struct page *page,
601 u64 start, u64 end, int mirror)
605 struct extent_buffer *eb;
606 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
613 eb = (struct extent_buffer *)page->private;
615 /* the pending IO might have been the only thing that kept this buffer
616 * in memory. Make sure we have a ref for all this other checks
618 extent_buffer_get(eb);
620 reads_done = atomic_dec_and_test(&eb->io_pages);
624 eb->read_mirror = mirror;
625 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
630 found_start = btrfs_header_bytenr(eb);
631 if (found_start != eb->start) {
632 btrfs_err_rl(eb->fs_info, "bad tree block start %llu %llu",
633 found_start, eb->start);
637 if (check_tree_block_fsid(root->fs_info, eb)) {
638 btrfs_err_rl(eb->fs_info, "bad fsid on block %llu",
643 found_level = btrfs_header_level(eb);
644 if (found_level >= BTRFS_MAX_LEVEL) {
645 btrfs_err(root->fs_info, "bad tree block level %d",
646 (int)btrfs_header_level(eb));
651 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
654 ret = csum_tree_block(root->fs_info, eb, 1);
661 * If this is a leaf block and it is corrupt, set the corrupt bit so
662 * that we don't try and read the other copies of this block, just
665 if (found_level == 0 && check_leaf(root, eb)) {
666 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
671 set_extent_buffer_uptodate(eb);
674 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
675 btree_readahead_hook(root, eb, eb->start, ret);
679 * our io error hook is going to dec the io pages
680 * again, we have to make sure it has something
683 atomic_inc(&eb->io_pages);
684 clear_extent_buffer_uptodate(eb);
686 free_extent_buffer(eb);
691 static int btree_io_failed_hook(struct page *page, int failed_mirror)
693 struct extent_buffer *eb;
694 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
696 eb = (struct extent_buffer *)page->private;
697 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
698 eb->read_mirror = failed_mirror;
699 atomic_dec(&eb->io_pages);
700 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
701 btree_readahead_hook(root, eb, eb->start, -EIO);
702 return -EIO; /* we fixed nothing */
705 static void end_workqueue_bio(struct bio *bio)
707 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
708 struct btrfs_fs_info *fs_info;
709 struct btrfs_workqueue *wq;
710 btrfs_work_func_t func;
712 fs_info = end_io_wq->info;
713 end_io_wq->error = bio->bi_error;
715 if (bio->bi_rw & REQ_WRITE) {
716 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
717 wq = fs_info->endio_meta_write_workers;
718 func = btrfs_endio_meta_write_helper;
719 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
720 wq = fs_info->endio_freespace_worker;
721 func = btrfs_freespace_write_helper;
722 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
723 wq = fs_info->endio_raid56_workers;
724 func = btrfs_endio_raid56_helper;
726 wq = fs_info->endio_write_workers;
727 func = btrfs_endio_write_helper;
730 if (unlikely(end_io_wq->metadata ==
731 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
732 wq = fs_info->endio_repair_workers;
733 func = btrfs_endio_repair_helper;
734 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
735 wq = fs_info->endio_raid56_workers;
736 func = btrfs_endio_raid56_helper;
737 } else if (end_io_wq->metadata) {
738 wq = fs_info->endio_meta_workers;
739 func = btrfs_endio_meta_helper;
741 wq = fs_info->endio_workers;
742 func = btrfs_endio_helper;
746 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
747 btrfs_queue_work(wq, &end_io_wq->work);
750 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
751 enum btrfs_wq_endio_type metadata)
753 struct btrfs_end_io_wq *end_io_wq;
755 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
759 end_io_wq->private = bio->bi_private;
760 end_io_wq->end_io = bio->bi_end_io;
761 end_io_wq->info = info;
762 end_io_wq->error = 0;
763 end_io_wq->bio = bio;
764 end_io_wq->metadata = metadata;
766 bio->bi_private = end_io_wq;
767 bio->bi_end_io = end_workqueue_bio;
771 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
773 unsigned long limit = min_t(unsigned long,
774 info->thread_pool_size,
775 info->fs_devices->open_devices);
779 static void run_one_async_start(struct btrfs_work *work)
781 struct async_submit_bio *async;
784 async = container_of(work, struct async_submit_bio, work);
785 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
786 async->mirror_num, async->bio_flags,
792 static void run_one_async_done(struct btrfs_work *work)
794 struct btrfs_fs_info *fs_info;
795 struct async_submit_bio *async;
798 async = container_of(work, struct async_submit_bio, work);
799 fs_info = BTRFS_I(async->inode)->root->fs_info;
801 limit = btrfs_async_submit_limit(fs_info);
802 limit = limit * 2 / 3;
805 * atomic_dec_return implies a barrier for waitqueue_active
807 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
808 waitqueue_active(&fs_info->async_submit_wait))
809 wake_up(&fs_info->async_submit_wait);
811 /* If an error occured we just want to clean up the bio and move on */
813 async->bio->bi_error = async->error;
814 bio_endio(async->bio);
818 async->submit_bio_done(async->inode, async->rw, async->bio,
819 async->mirror_num, async->bio_flags,
823 static void run_one_async_free(struct btrfs_work *work)
825 struct async_submit_bio *async;
827 async = container_of(work, struct async_submit_bio, work);
831 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
832 int rw, struct bio *bio, int mirror_num,
833 unsigned long bio_flags,
835 extent_submit_bio_hook_t *submit_bio_start,
836 extent_submit_bio_hook_t *submit_bio_done)
838 struct async_submit_bio *async;
840 async = kmalloc(sizeof(*async), GFP_NOFS);
844 async->inode = inode;
847 async->mirror_num = mirror_num;
848 async->submit_bio_start = submit_bio_start;
849 async->submit_bio_done = submit_bio_done;
851 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
852 run_one_async_done, run_one_async_free);
854 async->bio_flags = bio_flags;
855 async->bio_offset = bio_offset;
859 atomic_inc(&fs_info->nr_async_submits);
862 btrfs_set_work_high_priority(&async->work);
864 btrfs_queue_work(fs_info->workers, &async->work);
866 while (atomic_read(&fs_info->async_submit_draining) &&
867 atomic_read(&fs_info->nr_async_submits)) {
868 wait_event(fs_info->async_submit_wait,
869 (atomic_read(&fs_info->nr_async_submits) == 0));
875 static int btree_csum_one_bio(struct bio *bio)
877 struct bio_vec *bvec;
878 struct btrfs_root *root;
881 bio_for_each_segment_all(bvec, bio, i) {
882 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
883 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
891 static int __btree_submit_bio_start(struct inode *inode, int rw,
892 struct bio *bio, int mirror_num,
893 unsigned long bio_flags,
897 * when we're called for a write, we're already in the async
898 * submission context. Just jump into btrfs_map_bio
900 return btree_csum_one_bio(bio);
903 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
904 int mirror_num, unsigned long bio_flags,
910 * when we're called for a write, we're already in the async
911 * submission context. Just jump into btrfs_map_bio
913 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
921 static int check_async_write(struct inode *inode, unsigned long bio_flags)
923 if (bio_flags & EXTENT_BIO_TREE_LOG)
932 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
933 int mirror_num, unsigned long bio_flags,
936 int async = check_async_write(inode, bio_flags);
939 if (!(rw & REQ_WRITE)) {
941 * called for a read, do the setup so that checksum validation
942 * can happen in the async kernel threads
944 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
945 bio, BTRFS_WQ_ENDIO_METADATA);
948 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
951 ret = btree_csum_one_bio(bio);
954 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
958 * kthread helpers are used to submit writes so that
959 * checksumming can happen in parallel across all CPUs
961 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
962 inode, rw, bio, mirror_num, 0,
964 __btree_submit_bio_start,
965 __btree_submit_bio_done);
978 #ifdef CONFIG_MIGRATION
979 static int btree_migratepage(struct address_space *mapping,
980 struct page *newpage, struct page *page,
981 enum migrate_mode mode)
984 * we can't safely write a btree page from here,
985 * we haven't done the locking hook
990 * Buffers may be managed in a filesystem specific way.
991 * We must have no buffers or drop them.
993 if (page_has_private(page) &&
994 !try_to_release_page(page, GFP_KERNEL))
996 return migrate_page(mapping, newpage, page, mode);
1001 static int btree_writepages(struct address_space *mapping,
1002 struct writeback_control *wbc)
1004 struct btrfs_fs_info *fs_info;
1007 if (wbc->sync_mode == WB_SYNC_NONE) {
1009 if (wbc->for_kupdate)
1012 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1013 /* this is a bit racy, but that's ok */
1014 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1015 BTRFS_DIRTY_METADATA_THRESH);
1019 return btree_write_cache_pages(mapping, wbc);
1022 static int btree_readpage(struct file *file, struct page *page)
1024 struct extent_io_tree *tree;
1025 tree = &BTRFS_I(page->mapping->host)->io_tree;
1026 return extent_read_full_page(tree, page, btree_get_extent, 0);
1029 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1031 if (PageWriteback(page) || PageDirty(page))
1034 return try_release_extent_buffer(page);
1037 static void btree_invalidatepage(struct page *page, unsigned int offset,
1038 unsigned int length)
1040 struct extent_io_tree *tree;
1041 tree = &BTRFS_I(page->mapping->host)->io_tree;
1042 extent_invalidatepage(tree, page, offset);
1043 btree_releasepage(page, GFP_NOFS);
1044 if (PagePrivate(page)) {
1045 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1046 "page private not zero on page %llu",
1047 (unsigned long long)page_offset(page));
1048 ClearPagePrivate(page);
1049 set_page_private(page, 0);
1050 page_cache_release(page);
1054 static int btree_set_page_dirty(struct page *page)
1057 struct extent_buffer *eb;
1059 BUG_ON(!PagePrivate(page));
1060 eb = (struct extent_buffer *)page->private;
1062 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1063 BUG_ON(!atomic_read(&eb->refs));
1064 btrfs_assert_tree_locked(eb);
1066 return __set_page_dirty_nobuffers(page);
1069 static const struct address_space_operations btree_aops = {
1070 .readpage = btree_readpage,
1071 .writepages = btree_writepages,
1072 .releasepage = btree_releasepage,
1073 .invalidatepage = btree_invalidatepage,
1074 #ifdef CONFIG_MIGRATION
1075 .migratepage = btree_migratepage,
1077 .set_page_dirty = btree_set_page_dirty,
1080 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1082 struct extent_buffer *buf = NULL;
1083 struct inode *btree_inode = root->fs_info->btree_inode;
1085 buf = btrfs_find_create_tree_block(root, bytenr);
1088 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1089 buf, 0, WAIT_NONE, btree_get_extent, 0);
1090 free_extent_buffer(buf);
1093 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1094 int mirror_num, struct extent_buffer **eb)
1096 struct extent_buffer *buf = NULL;
1097 struct inode *btree_inode = root->fs_info->btree_inode;
1098 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1101 buf = btrfs_find_create_tree_block(root, bytenr);
1105 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1107 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1108 btree_get_extent, mirror_num);
1110 free_extent_buffer(buf);
1114 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1115 free_extent_buffer(buf);
1117 } else if (extent_buffer_uptodate(buf)) {
1120 free_extent_buffer(buf);
1125 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1128 return find_extent_buffer(fs_info, bytenr);
1131 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1134 if (btrfs_test_is_dummy_root(root))
1135 return alloc_test_extent_buffer(root->fs_info, bytenr);
1136 return alloc_extent_buffer(root->fs_info, bytenr);
1140 int btrfs_write_tree_block(struct extent_buffer *buf)
1142 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1143 buf->start + buf->len - 1);
1146 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1148 return filemap_fdatawait_range(buf->pages[0]->mapping,
1149 buf->start, buf->start + buf->len - 1);
1152 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1155 struct extent_buffer *buf = NULL;
1158 buf = btrfs_find_create_tree_block(root, bytenr);
1160 return ERR_PTR(-ENOMEM);
1162 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1164 free_extent_buffer(buf);
1165 return ERR_PTR(ret);
1171 void clean_tree_block(struct btrfs_trans_handle *trans,
1172 struct btrfs_fs_info *fs_info,
1173 struct extent_buffer *buf)
1175 if (btrfs_header_generation(buf) ==
1176 fs_info->running_transaction->transid) {
1177 btrfs_assert_tree_locked(buf);
1179 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1180 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1182 fs_info->dirty_metadata_batch);
1183 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1184 btrfs_set_lock_blocking(buf);
1185 clear_extent_buffer_dirty(buf);
1190 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1192 struct btrfs_subvolume_writers *writers;
1195 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1197 return ERR_PTR(-ENOMEM);
1199 ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1202 return ERR_PTR(ret);
1205 init_waitqueue_head(&writers->wait);
1210 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1212 percpu_counter_destroy(&writers->counter);
1216 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1217 struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1221 root->commit_root = NULL;
1222 root->sectorsize = sectorsize;
1223 root->nodesize = nodesize;
1224 root->stripesize = stripesize;
1226 root->orphan_cleanup_state = 0;
1228 root->objectid = objectid;
1229 root->last_trans = 0;
1230 root->highest_objectid = 0;
1231 root->nr_delalloc_inodes = 0;
1232 root->nr_ordered_extents = 0;
1234 root->inode_tree = RB_ROOT;
1235 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1236 root->block_rsv = NULL;
1237 root->orphan_block_rsv = NULL;
1239 INIT_LIST_HEAD(&root->dirty_list);
1240 INIT_LIST_HEAD(&root->root_list);
1241 INIT_LIST_HEAD(&root->delalloc_inodes);
1242 INIT_LIST_HEAD(&root->delalloc_root);
1243 INIT_LIST_HEAD(&root->ordered_extents);
1244 INIT_LIST_HEAD(&root->ordered_root);
1245 INIT_LIST_HEAD(&root->logged_list[0]);
1246 INIT_LIST_HEAD(&root->logged_list[1]);
1247 spin_lock_init(&root->orphan_lock);
1248 spin_lock_init(&root->inode_lock);
1249 spin_lock_init(&root->delalloc_lock);
1250 spin_lock_init(&root->ordered_extent_lock);
1251 spin_lock_init(&root->accounting_lock);
1252 spin_lock_init(&root->log_extents_lock[0]);
1253 spin_lock_init(&root->log_extents_lock[1]);
1254 mutex_init(&root->objectid_mutex);
1255 mutex_init(&root->log_mutex);
1256 mutex_init(&root->ordered_extent_mutex);
1257 mutex_init(&root->delalloc_mutex);
1258 init_waitqueue_head(&root->log_writer_wait);
1259 init_waitqueue_head(&root->log_commit_wait[0]);
1260 init_waitqueue_head(&root->log_commit_wait[1]);
1261 INIT_LIST_HEAD(&root->log_ctxs[0]);
1262 INIT_LIST_HEAD(&root->log_ctxs[1]);
1263 atomic_set(&root->log_commit[0], 0);
1264 atomic_set(&root->log_commit[1], 0);
1265 atomic_set(&root->log_writers, 0);
1266 atomic_set(&root->log_batch, 0);
1267 atomic_set(&root->orphan_inodes, 0);
1268 atomic_set(&root->refs, 1);
1269 atomic_set(&root->will_be_snapshoted, 0);
1270 atomic_set(&root->qgroup_meta_rsv, 0);
1271 root->log_transid = 0;
1272 root->log_transid_committed = -1;
1273 root->last_log_commit = 0;
1275 extent_io_tree_init(&root->dirty_log_pages,
1276 fs_info->btree_inode->i_mapping);
1278 memset(&root->root_key, 0, sizeof(root->root_key));
1279 memset(&root->root_item, 0, sizeof(root->root_item));
1280 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1282 root->defrag_trans_start = fs_info->generation;
1284 root->defrag_trans_start = 0;
1285 root->root_key.objectid = objectid;
1288 spin_lock_init(&root->root_item_lock);
1291 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1293 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1295 root->fs_info = fs_info;
1299 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1300 /* Should only be used by the testing infrastructure */
1301 struct btrfs_root *btrfs_alloc_dummy_root(void)
1303 struct btrfs_root *root;
1305 root = btrfs_alloc_root(NULL);
1307 return ERR_PTR(-ENOMEM);
1308 __setup_root(4096, 4096, 4096, root, NULL, 1);
1309 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1310 root->alloc_bytenr = 0;
1316 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1317 struct btrfs_fs_info *fs_info,
1320 struct extent_buffer *leaf;
1321 struct btrfs_root *tree_root = fs_info->tree_root;
1322 struct btrfs_root *root;
1323 struct btrfs_key key;
1327 root = btrfs_alloc_root(fs_info);
1329 return ERR_PTR(-ENOMEM);
1331 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1332 tree_root->stripesize, root, fs_info, objectid);
1333 root->root_key.objectid = objectid;
1334 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1335 root->root_key.offset = 0;
1337 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1339 ret = PTR_ERR(leaf);
1344 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1345 btrfs_set_header_bytenr(leaf, leaf->start);
1346 btrfs_set_header_generation(leaf, trans->transid);
1347 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1348 btrfs_set_header_owner(leaf, objectid);
1351 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1353 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1354 btrfs_header_chunk_tree_uuid(leaf),
1356 btrfs_mark_buffer_dirty(leaf);
1358 root->commit_root = btrfs_root_node(root);
1359 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1361 root->root_item.flags = 0;
1362 root->root_item.byte_limit = 0;
1363 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1364 btrfs_set_root_generation(&root->root_item, trans->transid);
1365 btrfs_set_root_level(&root->root_item, 0);
1366 btrfs_set_root_refs(&root->root_item, 1);
1367 btrfs_set_root_used(&root->root_item, leaf->len);
1368 btrfs_set_root_last_snapshot(&root->root_item, 0);
1369 btrfs_set_root_dirid(&root->root_item, 0);
1371 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1372 root->root_item.drop_level = 0;
1374 key.objectid = objectid;
1375 key.type = BTRFS_ROOT_ITEM_KEY;
1377 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1381 btrfs_tree_unlock(leaf);
1387 btrfs_tree_unlock(leaf);
1388 free_extent_buffer(root->commit_root);
1389 free_extent_buffer(leaf);
1393 return ERR_PTR(ret);
1396 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1397 struct btrfs_fs_info *fs_info)
1399 struct btrfs_root *root;
1400 struct btrfs_root *tree_root = fs_info->tree_root;
1401 struct extent_buffer *leaf;
1403 root = btrfs_alloc_root(fs_info);
1405 return ERR_PTR(-ENOMEM);
1407 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1408 tree_root->stripesize, root, fs_info,
1409 BTRFS_TREE_LOG_OBJECTID);
1411 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1412 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1413 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1416 * DON'T set REF_COWS for log trees
1418 * log trees do not get reference counted because they go away
1419 * before a real commit is actually done. They do store pointers
1420 * to file data extents, and those reference counts still get
1421 * updated (along with back refs to the log tree).
1424 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1428 return ERR_CAST(leaf);
1431 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1432 btrfs_set_header_bytenr(leaf, leaf->start);
1433 btrfs_set_header_generation(leaf, trans->transid);
1434 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1435 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1438 write_extent_buffer(root->node, root->fs_info->fsid,
1439 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1440 btrfs_mark_buffer_dirty(root->node);
1441 btrfs_tree_unlock(root->node);
1445 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1446 struct btrfs_fs_info *fs_info)
1448 struct btrfs_root *log_root;
1450 log_root = alloc_log_tree(trans, fs_info);
1451 if (IS_ERR(log_root))
1452 return PTR_ERR(log_root);
1453 WARN_ON(fs_info->log_root_tree);
1454 fs_info->log_root_tree = log_root;
1458 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1459 struct btrfs_root *root)
1461 struct btrfs_root *log_root;
1462 struct btrfs_inode_item *inode_item;
1464 log_root = alloc_log_tree(trans, root->fs_info);
1465 if (IS_ERR(log_root))
1466 return PTR_ERR(log_root);
1468 log_root->last_trans = trans->transid;
1469 log_root->root_key.offset = root->root_key.objectid;
1471 inode_item = &log_root->root_item.inode;
1472 btrfs_set_stack_inode_generation(inode_item, 1);
1473 btrfs_set_stack_inode_size(inode_item, 3);
1474 btrfs_set_stack_inode_nlink(inode_item, 1);
1475 btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1476 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1478 btrfs_set_root_node(&log_root->root_item, log_root->node);
1480 WARN_ON(root->log_root);
1481 root->log_root = log_root;
1482 root->log_transid = 0;
1483 root->log_transid_committed = -1;
1484 root->last_log_commit = 0;
1488 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1489 struct btrfs_key *key)
1491 struct btrfs_root *root;
1492 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1493 struct btrfs_path *path;
1497 path = btrfs_alloc_path();
1499 return ERR_PTR(-ENOMEM);
1501 root = btrfs_alloc_root(fs_info);
1507 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1508 tree_root->stripesize, root, fs_info, key->objectid);
1510 ret = btrfs_find_root(tree_root, key, path,
1511 &root->root_item, &root->root_key);
1518 generation = btrfs_root_generation(&root->root_item);
1519 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1521 if (IS_ERR(root->node)) {
1522 ret = PTR_ERR(root->node);
1524 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1526 free_extent_buffer(root->node);
1529 root->commit_root = btrfs_root_node(root);
1531 btrfs_free_path(path);
1537 root = ERR_PTR(ret);
1541 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1542 struct btrfs_key *location)
1544 struct btrfs_root *root;
1546 root = btrfs_read_tree_root(tree_root, location);
1550 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1551 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1552 btrfs_check_and_init_root_item(&root->root_item);
1558 int btrfs_init_fs_root(struct btrfs_root *root)
1561 struct btrfs_subvolume_writers *writers;
1563 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1564 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1566 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1571 writers = btrfs_alloc_subvolume_writers();
1572 if (IS_ERR(writers)) {
1573 ret = PTR_ERR(writers);
1576 root->subv_writers = writers;
1578 btrfs_init_free_ino_ctl(root);
1579 spin_lock_init(&root->ino_cache_lock);
1580 init_waitqueue_head(&root->ino_cache_wait);
1582 ret = get_anon_bdev(&root->anon_dev);
1586 mutex_lock(&root->objectid_mutex);
1587 ret = btrfs_find_highest_objectid(root,
1588 &root->highest_objectid);
1590 mutex_unlock(&root->objectid_mutex);
1594 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1596 mutex_unlock(&root->objectid_mutex);
1601 free_anon_bdev(root->anon_dev);
1603 btrfs_free_subvolume_writers(root->subv_writers);
1605 kfree(root->free_ino_ctl);
1606 kfree(root->free_ino_pinned);
1610 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1613 struct btrfs_root *root;
1615 spin_lock(&fs_info->fs_roots_radix_lock);
1616 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1617 (unsigned long)root_id);
1618 spin_unlock(&fs_info->fs_roots_radix_lock);
1622 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1623 struct btrfs_root *root)
1627 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1631 spin_lock(&fs_info->fs_roots_radix_lock);
1632 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1633 (unsigned long)root->root_key.objectid,
1636 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1637 spin_unlock(&fs_info->fs_roots_radix_lock);
1638 radix_tree_preload_end();
1643 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1644 struct btrfs_key *location,
1647 struct btrfs_root *root;
1648 struct btrfs_path *path;
1649 struct btrfs_key key;
1652 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1653 return fs_info->tree_root;
1654 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1655 return fs_info->extent_root;
1656 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1657 return fs_info->chunk_root;
1658 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1659 return fs_info->dev_root;
1660 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1661 return fs_info->csum_root;
1662 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1663 return fs_info->quota_root ? fs_info->quota_root :
1665 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1666 return fs_info->uuid_root ? fs_info->uuid_root :
1669 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1671 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1672 return ERR_PTR(-ENOENT);
1676 root = btrfs_read_fs_root(fs_info->tree_root, location);
1680 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1685 ret = btrfs_init_fs_root(root);
1689 path = btrfs_alloc_path();
1694 key.objectid = BTRFS_ORPHAN_OBJECTID;
1695 key.type = BTRFS_ORPHAN_ITEM_KEY;
1696 key.offset = location->objectid;
1698 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1699 btrfs_free_path(path);
1703 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1705 ret = btrfs_insert_fs_root(fs_info, root);
1707 if (ret == -EEXIST) {
1716 return ERR_PTR(ret);
1719 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1721 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1723 struct btrfs_device *device;
1724 struct backing_dev_info *bdi;
1727 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1730 bdi = blk_get_backing_dev_info(device->bdev);
1731 if (bdi_congested(bdi, bdi_bits)) {
1740 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1744 err = bdi_setup_and_register(bdi, "btrfs");
1748 bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE;
1749 bdi->congested_fn = btrfs_congested_fn;
1750 bdi->congested_data = info;
1751 bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1756 * called by the kthread helper functions to finally call the bio end_io
1757 * functions. This is where read checksum verification actually happens
1759 static void end_workqueue_fn(struct btrfs_work *work)
1762 struct btrfs_end_io_wq *end_io_wq;
1764 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1765 bio = end_io_wq->bio;
1767 bio->bi_error = end_io_wq->error;
1768 bio->bi_private = end_io_wq->private;
1769 bio->bi_end_io = end_io_wq->end_io;
1770 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1774 static int cleaner_kthread(void *arg)
1776 struct btrfs_root *root = arg;
1778 struct btrfs_trans_handle *trans;
1783 /* Make the cleaner go to sleep early. */
1784 if (btrfs_need_cleaner_sleep(root))
1787 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1791 * Avoid the problem that we change the status of the fs
1792 * during the above check and trylock.
1794 if (btrfs_need_cleaner_sleep(root)) {
1795 mutex_unlock(&root->fs_info->cleaner_mutex);
1799 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1800 btrfs_run_delayed_iputs(root);
1801 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1803 again = btrfs_clean_one_deleted_snapshot(root);
1804 mutex_unlock(&root->fs_info->cleaner_mutex);
1807 * The defragger has dealt with the R/O remount and umount,
1808 * needn't do anything special here.
1810 btrfs_run_defrag_inodes(root->fs_info);
1813 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1814 * with relocation (btrfs_relocate_chunk) and relocation
1815 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1816 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1817 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1818 * unused block groups.
1820 btrfs_delete_unused_bgs(root->fs_info);
1822 if (!try_to_freeze() && !again) {
1823 set_current_state(TASK_INTERRUPTIBLE);
1824 if (!kthread_should_stop())
1826 __set_current_state(TASK_RUNNING);
1828 } while (!kthread_should_stop());
1831 * Transaction kthread is stopped before us and wakes us up.
1832 * However we might have started a new transaction and COWed some
1833 * tree blocks when deleting unused block groups for example. So
1834 * make sure we commit the transaction we started to have a clean
1835 * shutdown when evicting the btree inode - if it has dirty pages
1836 * when we do the final iput() on it, eviction will trigger a
1837 * writeback for it which will fail with null pointer dereferences
1838 * since work queues and other resources were already released and
1839 * destroyed by the time the iput/eviction/writeback is made.
1841 trans = btrfs_attach_transaction(root);
1842 if (IS_ERR(trans)) {
1843 if (PTR_ERR(trans) != -ENOENT)
1844 btrfs_err(root->fs_info,
1845 "cleaner transaction attach returned %ld",
1850 ret = btrfs_commit_transaction(trans, root);
1852 btrfs_err(root->fs_info,
1853 "cleaner open transaction commit returned %d",
1860 static int transaction_kthread(void *arg)
1862 struct btrfs_root *root = arg;
1863 struct btrfs_trans_handle *trans;
1864 struct btrfs_transaction *cur;
1867 unsigned long delay;
1871 cannot_commit = false;
1872 delay = HZ * root->fs_info->commit_interval;
1873 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1875 spin_lock(&root->fs_info->trans_lock);
1876 cur = root->fs_info->running_transaction;
1878 spin_unlock(&root->fs_info->trans_lock);
1882 now = get_seconds();
1883 if (cur->state < TRANS_STATE_BLOCKED &&
1884 (now < cur->start_time ||
1885 now - cur->start_time < root->fs_info->commit_interval)) {
1886 spin_unlock(&root->fs_info->trans_lock);
1890 transid = cur->transid;
1891 spin_unlock(&root->fs_info->trans_lock);
1893 /* If the file system is aborted, this will always fail. */
1894 trans = btrfs_attach_transaction(root);
1895 if (IS_ERR(trans)) {
1896 if (PTR_ERR(trans) != -ENOENT)
1897 cannot_commit = true;
1900 if (transid == trans->transid) {
1901 btrfs_commit_transaction(trans, root);
1903 btrfs_end_transaction(trans, root);
1906 wake_up_process(root->fs_info->cleaner_kthread);
1907 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1909 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1910 &root->fs_info->fs_state)))
1911 btrfs_cleanup_transaction(root);
1912 if (!try_to_freeze()) {
1913 set_current_state(TASK_INTERRUPTIBLE);
1914 if (!kthread_should_stop() &&
1915 (!btrfs_transaction_blocked(root->fs_info) ||
1917 schedule_timeout(delay);
1918 __set_current_state(TASK_RUNNING);
1920 } while (!kthread_should_stop());
1925 * this will find the highest generation in the array of
1926 * root backups. The index of the highest array is returned,
1927 * or -1 if we can't find anything.
1929 * We check to make sure the array is valid by comparing the
1930 * generation of the latest root in the array with the generation
1931 * in the super block. If they don't match we pitch it.
1933 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1936 int newest_index = -1;
1937 struct btrfs_root_backup *root_backup;
1940 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1941 root_backup = info->super_copy->super_roots + i;
1942 cur = btrfs_backup_tree_root_gen(root_backup);
1943 if (cur == newest_gen)
1947 /* check to see if we actually wrapped around */
1948 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1949 root_backup = info->super_copy->super_roots;
1950 cur = btrfs_backup_tree_root_gen(root_backup);
1951 if (cur == newest_gen)
1954 return newest_index;
1959 * find the oldest backup so we know where to store new entries
1960 * in the backup array. This will set the backup_root_index
1961 * field in the fs_info struct
1963 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1966 int newest_index = -1;
1968 newest_index = find_newest_super_backup(info, newest_gen);
1969 /* if there was garbage in there, just move along */
1970 if (newest_index == -1) {
1971 info->backup_root_index = 0;
1973 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1978 * copy all the root pointers into the super backup array.
1979 * this will bump the backup pointer by one when it is
1982 static void backup_super_roots(struct btrfs_fs_info *info)
1985 struct btrfs_root_backup *root_backup;
1988 next_backup = info->backup_root_index;
1989 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1990 BTRFS_NUM_BACKUP_ROOTS;
1993 * just overwrite the last backup if we're at the same generation
1994 * this happens only at umount
1996 root_backup = info->super_for_commit->super_roots + last_backup;
1997 if (btrfs_backup_tree_root_gen(root_backup) ==
1998 btrfs_header_generation(info->tree_root->node))
1999 next_backup = last_backup;
2001 root_backup = info->super_for_commit->super_roots + next_backup;
2004 * make sure all of our padding and empty slots get zero filled
2005 * regardless of which ones we use today
2007 memset(root_backup, 0, sizeof(*root_backup));
2009 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2011 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2012 btrfs_set_backup_tree_root_gen(root_backup,
2013 btrfs_header_generation(info->tree_root->node));
2015 btrfs_set_backup_tree_root_level(root_backup,
2016 btrfs_header_level(info->tree_root->node));
2018 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2019 btrfs_set_backup_chunk_root_gen(root_backup,
2020 btrfs_header_generation(info->chunk_root->node));
2021 btrfs_set_backup_chunk_root_level(root_backup,
2022 btrfs_header_level(info->chunk_root->node));
2024 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2025 btrfs_set_backup_extent_root_gen(root_backup,
2026 btrfs_header_generation(info->extent_root->node));
2027 btrfs_set_backup_extent_root_level(root_backup,
2028 btrfs_header_level(info->extent_root->node));
2031 * we might commit during log recovery, which happens before we set
2032 * the fs_root. Make sure it is valid before we fill it in.
2034 if (info->fs_root && info->fs_root->node) {
2035 btrfs_set_backup_fs_root(root_backup,
2036 info->fs_root->node->start);
2037 btrfs_set_backup_fs_root_gen(root_backup,
2038 btrfs_header_generation(info->fs_root->node));
2039 btrfs_set_backup_fs_root_level(root_backup,
2040 btrfs_header_level(info->fs_root->node));
2043 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2044 btrfs_set_backup_dev_root_gen(root_backup,
2045 btrfs_header_generation(info->dev_root->node));
2046 btrfs_set_backup_dev_root_level(root_backup,
2047 btrfs_header_level(info->dev_root->node));
2049 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2050 btrfs_set_backup_csum_root_gen(root_backup,
2051 btrfs_header_generation(info->csum_root->node));
2052 btrfs_set_backup_csum_root_level(root_backup,
2053 btrfs_header_level(info->csum_root->node));
2055 btrfs_set_backup_total_bytes(root_backup,
2056 btrfs_super_total_bytes(info->super_copy));
2057 btrfs_set_backup_bytes_used(root_backup,
2058 btrfs_super_bytes_used(info->super_copy));
2059 btrfs_set_backup_num_devices(root_backup,
2060 btrfs_super_num_devices(info->super_copy));
2063 * if we don't copy this out to the super_copy, it won't get remembered
2064 * for the next commit
2066 memcpy(&info->super_copy->super_roots,
2067 &info->super_for_commit->super_roots,
2068 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2072 * this copies info out of the root backup array and back into
2073 * the in-memory super block. It is meant to help iterate through
2074 * the array, so you send it the number of backups you've already
2075 * tried and the last backup index you used.
2077 * this returns -1 when it has tried all the backups
2079 static noinline int next_root_backup(struct btrfs_fs_info *info,
2080 struct btrfs_super_block *super,
2081 int *num_backups_tried, int *backup_index)
2083 struct btrfs_root_backup *root_backup;
2084 int newest = *backup_index;
2086 if (*num_backups_tried == 0) {
2087 u64 gen = btrfs_super_generation(super);
2089 newest = find_newest_super_backup(info, gen);
2093 *backup_index = newest;
2094 *num_backups_tried = 1;
2095 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2096 /* we've tried all the backups, all done */
2099 /* jump to the next oldest backup */
2100 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2101 BTRFS_NUM_BACKUP_ROOTS;
2102 *backup_index = newest;
2103 *num_backups_tried += 1;
2105 root_backup = super->super_roots + newest;
2107 btrfs_set_super_generation(super,
2108 btrfs_backup_tree_root_gen(root_backup));
2109 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2110 btrfs_set_super_root_level(super,
2111 btrfs_backup_tree_root_level(root_backup));
2112 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2115 * fixme: the total bytes and num_devices need to match or we should
2118 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2119 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2123 /* helper to cleanup workers */
2124 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2126 btrfs_destroy_workqueue(fs_info->fixup_workers);
2127 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2128 btrfs_destroy_workqueue(fs_info->workers);
2129 btrfs_destroy_workqueue(fs_info->endio_workers);
2130 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2131 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2132 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2133 btrfs_destroy_workqueue(fs_info->rmw_workers);
2134 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2135 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2136 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2137 btrfs_destroy_workqueue(fs_info->submit_workers);
2138 btrfs_destroy_workqueue(fs_info->delayed_workers);
2139 btrfs_destroy_workqueue(fs_info->caching_workers);
2140 btrfs_destroy_workqueue(fs_info->readahead_workers);
2141 btrfs_destroy_workqueue(fs_info->flush_workers);
2142 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2143 btrfs_destroy_workqueue(fs_info->extent_workers);
2146 static void free_root_extent_buffers(struct btrfs_root *root)
2149 free_extent_buffer(root->node);
2150 free_extent_buffer(root->commit_root);
2152 root->commit_root = NULL;
2156 /* helper to cleanup tree roots */
2157 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2159 free_root_extent_buffers(info->tree_root);
2161 free_root_extent_buffers(info->dev_root);
2162 free_root_extent_buffers(info->extent_root);
2163 free_root_extent_buffers(info->csum_root);
2164 free_root_extent_buffers(info->quota_root);
2165 free_root_extent_buffers(info->uuid_root);
2167 free_root_extent_buffers(info->chunk_root);
2170 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2173 struct btrfs_root *gang[8];
2176 while (!list_empty(&fs_info->dead_roots)) {
2177 gang[0] = list_entry(fs_info->dead_roots.next,
2178 struct btrfs_root, root_list);
2179 list_del(&gang[0]->root_list);
2181 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2182 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2184 free_extent_buffer(gang[0]->node);
2185 free_extent_buffer(gang[0]->commit_root);
2186 btrfs_put_fs_root(gang[0]);
2191 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2196 for (i = 0; i < ret; i++)
2197 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2200 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2201 btrfs_free_log_root_tree(NULL, fs_info);
2202 btrfs_destroy_pinned_extent(fs_info->tree_root,
2203 fs_info->pinned_extents);
2207 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2209 mutex_init(&fs_info->scrub_lock);
2210 atomic_set(&fs_info->scrubs_running, 0);
2211 atomic_set(&fs_info->scrub_pause_req, 0);
2212 atomic_set(&fs_info->scrubs_paused, 0);
2213 atomic_set(&fs_info->scrub_cancel_req, 0);
2214 init_waitqueue_head(&fs_info->scrub_pause_wait);
2215 fs_info->scrub_workers_refcnt = 0;
2218 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2220 spin_lock_init(&fs_info->balance_lock);
2221 mutex_init(&fs_info->balance_mutex);
2222 atomic_set(&fs_info->balance_running, 0);
2223 atomic_set(&fs_info->balance_pause_req, 0);
2224 atomic_set(&fs_info->balance_cancel_req, 0);
2225 fs_info->balance_ctl = NULL;
2226 init_waitqueue_head(&fs_info->balance_wait_q);
2229 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2230 struct btrfs_root *tree_root)
2232 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2233 set_nlink(fs_info->btree_inode, 1);
2235 * we set the i_size on the btree inode to the max possible int.
2236 * the real end of the address space is determined by all of
2237 * the devices in the system
2239 fs_info->btree_inode->i_size = OFFSET_MAX;
2240 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2242 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2243 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2244 fs_info->btree_inode->i_mapping);
2245 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2246 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2248 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2250 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2251 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2252 sizeof(struct btrfs_key));
2253 set_bit(BTRFS_INODE_DUMMY,
2254 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2255 btrfs_insert_inode_hash(fs_info->btree_inode);
2258 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2260 fs_info->dev_replace.lock_owner = 0;
2261 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2262 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2263 mutex_init(&fs_info->dev_replace.lock_management_lock);
2264 mutex_init(&fs_info->dev_replace.lock);
2265 init_waitqueue_head(&fs_info->replace_wait);
2268 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2270 spin_lock_init(&fs_info->qgroup_lock);
2271 mutex_init(&fs_info->qgroup_ioctl_lock);
2272 fs_info->qgroup_tree = RB_ROOT;
2273 fs_info->qgroup_op_tree = RB_ROOT;
2274 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2275 fs_info->qgroup_seq = 1;
2276 fs_info->quota_enabled = 0;
2277 fs_info->pending_quota_state = 0;
2278 fs_info->qgroup_ulist = NULL;
2279 fs_info->qgroup_rescan_running = false;
2280 mutex_init(&fs_info->qgroup_rescan_lock);
2283 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2284 struct btrfs_fs_devices *fs_devices)
2286 int max_active = fs_info->thread_pool_size;
2287 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2290 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2293 fs_info->delalloc_workers =
2294 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2296 fs_info->flush_workers =
2297 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2299 fs_info->caching_workers =
2300 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2303 * a higher idle thresh on the submit workers makes it much more
2304 * likely that bios will be send down in a sane order to the
2307 fs_info->submit_workers =
2308 btrfs_alloc_workqueue("submit", flags,
2309 min_t(u64, fs_devices->num_devices,
2312 fs_info->fixup_workers =
2313 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2316 * endios are largely parallel and should have a very
2319 fs_info->endio_workers =
2320 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2321 fs_info->endio_meta_workers =
2322 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2323 fs_info->endio_meta_write_workers =
2324 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2325 fs_info->endio_raid56_workers =
2326 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2327 fs_info->endio_repair_workers =
2328 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2329 fs_info->rmw_workers =
2330 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2331 fs_info->endio_write_workers =
2332 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2333 fs_info->endio_freespace_worker =
2334 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2335 fs_info->delayed_workers =
2336 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2337 fs_info->readahead_workers =
2338 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2339 fs_info->qgroup_rescan_workers =
2340 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2341 fs_info->extent_workers =
2342 btrfs_alloc_workqueue("extent-refs", flags,
2343 min_t(u64, fs_devices->num_devices,
2346 if (!(fs_info->workers && fs_info->delalloc_workers &&
2347 fs_info->submit_workers && fs_info->flush_workers &&
2348 fs_info->endio_workers && fs_info->endio_meta_workers &&
2349 fs_info->endio_meta_write_workers &&
2350 fs_info->endio_repair_workers &&
2351 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2352 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2353 fs_info->caching_workers && fs_info->readahead_workers &&
2354 fs_info->fixup_workers && fs_info->delayed_workers &&
2355 fs_info->extent_workers &&
2356 fs_info->qgroup_rescan_workers)) {
2363 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2364 struct btrfs_fs_devices *fs_devices)
2367 struct btrfs_root *tree_root = fs_info->tree_root;
2368 struct btrfs_root *log_tree_root;
2369 struct btrfs_super_block *disk_super = fs_info->super_copy;
2370 u64 bytenr = btrfs_super_log_root(disk_super);
2372 if (fs_devices->rw_devices == 0) {
2373 btrfs_warn(fs_info, "log replay required on RO media");
2377 log_tree_root = btrfs_alloc_root(fs_info);
2381 __setup_root(tree_root->nodesize, tree_root->sectorsize,
2382 tree_root->stripesize, log_tree_root, fs_info,
2383 BTRFS_TREE_LOG_OBJECTID);
2385 log_tree_root->node = read_tree_block(tree_root, bytenr,
2386 fs_info->generation + 1);
2387 if (IS_ERR(log_tree_root->node)) {
2388 btrfs_warn(fs_info, "failed to read log tree");
2389 ret = PTR_ERR(log_tree_root->node);
2390 kfree(log_tree_root);
2392 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2393 btrfs_err(fs_info, "failed to read log tree");
2394 free_extent_buffer(log_tree_root->node);
2395 kfree(log_tree_root);
2398 /* returns with log_tree_root freed on success */
2399 ret = btrfs_recover_log_trees(log_tree_root);
2401 btrfs_std_error(tree_root->fs_info, ret,
2402 "Failed to recover log tree");
2403 free_extent_buffer(log_tree_root->node);
2404 kfree(log_tree_root);
2408 if (fs_info->sb->s_flags & MS_RDONLY) {
2409 ret = btrfs_commit_super(tree_root);
2417 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2418 struct btrfs_root *tree_root)
2420 struct btrfs_root *root;
2421 struct btrfs_key location;
2424 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2425 location.type = BTRFS_ROOT_ITEM_KEY;
2426 location.offset = 0;
2428 root = btrfs_read_tree_root(tree_root, &location);
2430 return PTR_ERR(root);
2431 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2432 fs_info->extent_root = root;
2434 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2435 root = btrfs_read_tree_root(tree_root, &location);
2437 return PTR_ERR(root);
2438 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2439 fs_info->dev_root = root;
2440 btrfs_init_devices_late(fs_info);
2442 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2443 root = btrfs_read_tree_root(tree_root, &location);
2445 return PTR_ERR(root);
2446 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2447 fs_info->csum_root = root;
2449 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2450 root = btrfs_read_tree_root(tree_root, &location);
2451 if (!IS_ERR(root)) {
2452 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2453 fs_info->quota_enabled = 1;
2454 fs_info->pending_quota_state = 1;
2455 fs_info->quota_root = root;
2458 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2459 root = btrfs_read_tree_root(tree_root, &location);
2461 ret = PTR_ERR(root);
2465 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2466 fs_info->uuid_root = root;
2472 int open_ctree(struct super_block *sb,
2473 struct btrfs_fs_devices *fs_devices,
2481 struct btrfs_key location;
2482 struct buffer_head *bh;
2483 struct btrfs_super_block *disk_super;
2484 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2485 struct btrfs_root *tree_root;
2486 struct btrfs_root *chunk_root;
2489 int num_backups_tried = 0;
2490 int backup_index = 0;
2493 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2494 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2495 if (!tree_root || !chunk_root) {
2500 ret = init_srcu_struct(&fs_info->subvol_srcu);
2506 ret = setup_bdi(fs_info, &fs_info->bdi);
2512 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2517 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2518 (1 + ilog2(nr_cpu_ids));
2520 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2523 goto fail_dirty_metadata_bytes;
2526 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2529 goto fail_delalloc_bytes;
2532 fs_info->btree_inode = new_inode(sb);
2533 if (!fs_info->btree_inode) {
2535 goto fail_bio_counter;
2538 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2540 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2541 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2542 INIT_LIST_HEAD(&fs_info->trans_list);
2543 INIT_LIST_HEAD(&fs_info->dead_roots);
2544 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2545 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2546 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2547 spin_lock_init(&fs_info->delalloc_root_lock);
2548 spin_lock_init(&fs_info->trans_lock);
2549 spin_lock_init(&fs_info->fs_roots_radix_lock);
2550 spin_lock_init(&fs_info->delayed_iput_lock);
2551 spin_lock_init(&fs_info->defrag_inodes_lock);
2552 spin_lock_init(&fs_info->free_chunk_lock);
2553 spin_lock_init(&fs_info->tree_mod_seq_lock);
2554 spin_lock_init(&fs_info->super_lock);
2555 spin_lock_init(&fs_info->qgroup_op_lock);
2556 spin_lock_init(&fs_info->buffer_lock);
2557 spin_lock_init(&fs_info->unused_bgs_lock);
2558 rwlock_init(&fs_info->tree_mod_log_lock);
2559 mutex_init(&fs_info->unused_bg_unpin_mutex);
2560 mutex_init(&fs_info->delete_unused_bgs_mutex);
2561 mutex_init(&fs_info->reloc_mutex);
2562 mutex_init(&fs_info->delalloc_root_mutex);
2563 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2564 seqlock_init(&fs_info->profiles_lock);
2566 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2567 INIT_LIST_HEAD(&fs_info->space_info);
2568 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2569 INIT_LIST_HEAD(&fs_info->unused_bgs);
2570 btrfs_mapping_init(&fs_info->mapping_tree);
2571 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2572 BTRFS_BLOCK_RSV_GLOBAL);
2573 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2574 BTRFS_BLOCK_RSV_DELALLOC);
2575 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2576 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2577 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2578 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2579 BTRFS_BLOCK_RSV_DELOPS);
2580 atomic_set(&fs_info->nr_async_submits, 0);
2581 atomic_set(&fs_info->async_delalloc_pages, 0);
2582 atomic_set(&fs_info->async_submit_draining, 0);
2583 atomic_set(&fs_info->nr_async_bios, 0);
2584 atomic_set(&fs_info->defrag_running, 0);
2585 atomic_set(&fs_info->qgroup_op_seq, 0);
2586 atomic64_set(&fs_info->tree_mod_seq, 0);
2588 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2589 fs_info->metadata_ratio = 0;
2590 fs_info->defrag_inodes = RB_ROOT;
2591 fs_info->free_chunk_space = 0;
2592 fs_info->tree_mod_log = RB_ROOT;
2593 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2594 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2595 /* readahead state */
2596 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2597 spin_lock_init(&fs_info->reada_lock);
2599 fs_info->thread_pool_size = min_t(unsigned long,
2600 num_online_cpus() + 2, 8);
2602 INIT_LIST_HEAD(&fs_info->ordered_roots);
2603 spin_lock_init(&fs_info->ordered_root_lock);
2604 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2606 if (!fs_info->delayed_root) {
2610 btrfs_init_delayed_root(fs_info->delayed_root);
2612 btrfs_init_scrub(fs_info);
2613 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2614 fs_info->check_integrity_print_mask = 0;
2616 btrfs_init_balance(fs_info);
2617 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2619 sb->s_blocksize = 4096;
2620 sb->s_blocksize_bits = blksize_bits(4096);
2621 sb->s_bdi = &fs_info->bdi;
2623 btrfs_init_btree_inode(fs_info, tree_root);
2625 spin_lock_init(&fs_info->block_group_cache_lock);
2626 fs_info->block_group_cache_tree = RB_ROOT;
2627 fs_info->first_logical_byte = (u64)-1;
2629 extent_io_tree_init(&fs_info->freed_extents[0],
2630 fs_info->btree_inode->i_mapping);
2631 extent_io_tree_init(&fs_info->freed_extents[1],
2632 fs_info->btree_inode->i_mapping);
2633 fs_info->pinned_extents = &fs_info->freed_extents[0];
2634 fs_info->do_barriers = 1;
2637 mutex_init(&fs_info->ordered_operations_mutex);
2638 mutex_init(&fs_info->tree_log_mutex);
2639 mutex_init(&fs_info->chunk_mutex);
2640 mutex_init(&fs_info->transaction_kthread_mutex);
2641 mutex_init(&fs_info->cleaner_mutex);
2642 mutex_init(&fs_info->volume_mutex);
2643 mutex_init(&fs_info->ro_block_group_mutex);
2644 init_rwsem(&fs_info->commit_root_sem);
2645 init_rwsem(&fs_info->cleanup_work_sem);
2646 init_rwsem(&fs_info->subvol_sem);
2647 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2649 btrfs_init_dev_replace_locks(fs_info);
2650 btrfs_init_qgroup(fs_info);
2652 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2653 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2655 init_waitqueue_head(&fs_info->transaction_throttle);
2656 init_waitqueue_head(&fs_info->transaction_wait);
2657 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2658 init_waitqueue_head(&fs_info->async_submit_wait);
2660 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2662 ret = btrfs_alloc_stripe_hash_table(fs_info);
2668 __setup_root(4096, 4096, 4096, tree_root,
2669 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2671 invalidate_bdev(fs_devices->latest_bdev);
2674 * Read super block and check the signature bytes only
2676 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2683 * We want to check superblock checksum, the type is stored inside.
2684 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2686 if (btrfs_check_super_csum(bh->b_data)) {
2687 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2694 * super_copy is zeroed at allocation time and we never touch the
2695 * following bytes up to INFO_SIZE, the checksum is calculated from
2696 * the whole block of INFO_SIZE
2698 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2699 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2700 sizeof(*fs_info->super_for_commit));
2703 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2705 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2707 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2712 disk_super = fs_info->super_copy;
2713 if (!btrfs_super_root(disk_super))
2716 /* check FS state, whether FS is broken. */
2717 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2718 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2721 * run through our array of backup supers and setup
2722 * our ring pointer to the oldest one
2724 generation = btrfs_super_generation(disk_super);
2725 find_oldest_super_backup(fs_info, generation);
2728 * In the long term, we'll store the compression type in the super
2729 * block, and it'll be used for per file compression control.
2731 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2733 ret = btrfs_parse_options(tree_root, options);
2739 features = btrfs_super_incompat_flags(disk_super) &
2740 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2742 printk(KERN_ERR "BTRFS: couldn't mount because of "
2743 "unsupported optional features (%Lx).\n",
2750 * Leafsize and nodesize were always equal, this is only a sanity check.
2752 if (le32_to_cpu(disk_super->__unused_leafsize) !=
2753 btrfs_super_nodesize(disk_super)) {
2754 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2755 "blocksizes don't match. node %d leaf %d\n",
2756 btrfs_super_nodesize(disk_super),
2757 le32_to_cpu(disk_super->__unused_leafsize));
2761 if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2762 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2763 "blocksize (%d) was too large\n",
2764 btrfs_super_nodesize(disk_super));
2769 features = btrfs_super_incompat_flags(disk_super);
2770 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2771 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2772 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2774 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2775 printk(KERN_INFO "BTRFS: has skinny extents\n");
2778 * flag our filesystem as having big metadata blocks if
2779 * they are bigger than the page size
2781 if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2782 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2783 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2784 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2787 nodesize = btrfs_super_nodesize(disk_super);
2788 sectorsize = btrfs_super_sectorsize(disk_super);
2789 stripesize = btrfs_super_stripesize(disk_super);
2790 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2791 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2794 * mixed block groups end up with duplicate but slightly offset
2795 * extent buffers for the same range. It leads to corruptions
2797 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2798 (sectorsize != nodesize)) {
2799 printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2800 "are not allowed for mixed block groups on %s\n",
2806 * Needn't use the lock because there is no other task which will
2809 btrfs_set_super_incompat_flags(disk_super, features);
2811 features = btrfs_super_compat_ro_flags(disk_super) &
2812 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2813 if (!(sb->s_flags & MS_RDONLY) && features) {
2814 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2815 "unsupported option features (%Lx).\n",
2821 max_active = fs_info->thread_pool_size;
2823 ret = btrfs_init_workqueues(fs_info, fs_devices);
2826 goto fail_sb_buffer;
2829 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2830 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2831 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2833 tree_root->nodesize = nodesize;
2834 tree_root->sectorsize = sectorsize;
2835 tree_root->stripesize = stripesize;
2837 sb->s_blocksize = sectorsize;
2838 sb->s_blocksize_bits = blksize_bits(sectorsize);
2840 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2841 printk(KERN_ERR "BTRFS: valid FS not found on %s\n", sb->s_id);
2842 goto fail_sb_buffer;
2845 if (sectorsize != PAGE_SIZE) {
2846 printk(KERN_ERR "BTRFS: incompatible sector size (%lu) "
2847 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2848 goto fail_sb_buffer;
2851 mutex_lock(&fs_info->chunk_mutex);
2852 ret = btrfs_read_sys_array(tree_root);
2853 mutex_unlock(&fs_info->chunk_mutex);
2855 printk(KERN_ERR "BTRFS: failed to read the system "
2856 "array on %s\n", sb->s_id);
2857 goto fail_sb_buffer;
2860 generation = btrfs_super_chunk_root_generation(disk_super);
2862 __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2863 fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2865 chunk_root->node = read_tree_block(chunk_root,
2866 btrfs_super_chunk_root(disk_super),
2868 if (IS_ERR(chunk_root->node) ||
2869 !extent_buffer_uptodate(chunk_root->node)) {
2870 printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2872 if (!IS_ERR(chunk_root->node))
2873 free_extent_buffer(chunk_root->node);
2874 chunk_root->node = NULL;
2875 goto fail_tree_roots;
2877 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2878 chunk_root->commit_root = btrfs_root_node(chunk_root);
2880 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2881 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2883 ret = btrfs_read_chunk_tree(chunk_root);
2885 printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2887 goto fail_tree_roots;
2891 * keep the device that is marked to be the target device for the
2892 * dev_replace procedure
2894 btrfs_close_extra_devices(fs_devices, 0);
2896 if (!fs_devices->latest_bdev) {
2897 printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2899 goto fail_tree_roots;
2903 generation = btrfs_super_generation(disk_super);
2905 tree_root->node = read_tree_block(tree_root,
2906 btrfs_super_root(disk_super),
2908 if (IS_ERR(tree_root->node) ||
2909 !extent_buffer_uptodate(tree_root->node)) {
2910 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2912 if (!IS_ERR(tree_root->node))
2913 free_extent_buffer(tree_root->node);
2914 tree_root->node = NULL;
2915 goto recovery_tree_root;
2918 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2919 tree_root->commit_root = btrfs_root_node(tree_root);
2920 btrfs_set_root_refs(&tree_root->root_item, 1);
2922 mutex_lock(&tree_root->objectid_mutex);
2923 ret = btrfs_find_highest_objectid(tree_root,
2924 &tree_root->highest_objectid);
2926 mutex_unlock(&tree_root->objectid_mutex);
2927 goto recovery_tree_root;
2930 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2932 mutex_unlock(&tree_root->objectid_mutex);
2934 ret = btrfs_read_roots(fs_info, tree_root);
2936 goto recovery_tree_root;
2938 fs_info->generation = generation;
2939 fs_info->last_trans_committed = generation;
2941 ret = btrfs_recover_balance(fs_info);
2943 printk(KERN_ERR "BTRFS: failed to recover balance\n");
2944 goto fail_block_groups;
2947 ret = btrfs_init_dev_stats(fs_info);
2949 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2951 goto fail_block_groups;
2954 ret = btrfs_init_dev_replace(fs_info);
2956 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2957 goto fail_block_groups;
2960 btrfs_close_extra_devices(fs_devices, 1);
2962 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2964 pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret);
2965 goto fail_block_groups;
2968 ret = btrfs_sysfs_add_device(fs_devices);
2970 pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret);
2971 goto fail_fsdev_sysfs;
2974 ret = btrfs_sysfs_add_mounted(fs_info);
2976 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2977 goto fail_fsdev_sysfs;
2980 ret = btrfs_init_space_info(fs_info);
2982 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2986 ret = btrfs_read_block_groups(fs_info->extent_root);
2988 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2991 fs_info->num_tolerated_disk_barrier_failures =
2992 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2993 if (fs_info->fs_devices->missing_devices >
2994 fs_info->num_tolerated_disk_barrier_failures &&
2995 !(sb->s_flags & MS_RDONLY)) {
2996 pr_warn("BTRFS: missing devices(%llu) exceeds the limit(%d), writeable mount is not allowed\n",
2997 fs_info->fs_devices->missing_devices,
2998 fs_info->num_tolerated_disk_barrier_failures);
3002 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3004 if (IS_ERR(fs_info->cleaner_kthread))
3007 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3009 "btrfs-transaction");
3010 if (IS_ERR(fs_info->transaction_kthread))
3013 if (!btrfs_test_opt(tree_root, SSD) &&
3014 !btrfs_test_opt(tree_root, NOSSD) &&
3015 !fs_info->fs_devices->rotating) {
3016 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
3018 btrfs_set_opt(fs_info->mount_opt, SSD);
3022 * Mount does not set all options immediatelly, we can do it now and do
3023 * not have to wait for transaction commit
3025 btrfs_apply_pending_changes(fs_info);
3027 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3028 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
3029 ret = btrfsic_mount(tree_root, fs_devices,
3030 btrfs_test_opt(tree_root,
3031 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3033 fs_info->check_integrity_print_mask);
3035 printk(KERN_WARNING "BTRFS: failed to initialize"
3036 " integrity check module %s\n", sb->s_id);
3039 ret = btrfs_read_qgroup_config(fs_info);
3041 goto fail_trans_kthread;
3043 /* do not make disk changes in broken FS */
3044 if (btrfs_super_log_root(disk_super) != 0) {
3045 ret = btrfs_replay_log(fs_info, fs_devices);
3052 ret = btrfs_find_orphan_roots(tree_root);
3056 if (!(sb->s_flags & MS_RDONLY)) {
3057 ret = btrfs_cleanup_fs_roots(fs_info);
3061 mutex_lock(&fs_info->cleaner_mutex);
3062 ret = btrfs_recover_relocation(tree_root);
3063 mutex_unlock(&fs_info->cleaner_mutex);
3066 "BTRFS: failed to recover relocation\n");
3072 location.objectid = BTRFS_FS_TREE_OBJECTID;
3073 location.type = BTRFS_ROOT_ITEM_KEY;
3074 location.offset = 0;
3076 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3077 if (IS_ERR(fs_info->fs_root)) {
3078 err = PTR_ERR(fs_info->fs_root);
3082 if (sb->s_flags & MS_RDONLY)
3085 down_read(&fs_info->cleanup_work_sem);
3086 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3087 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3088 up_read(&fs_info->cleanup_work_sem);
3089 close_ctree(tree_root);
3092 up_read(&fs_info->cleanup_work_sem);
3094 ret = btrfs_resume_balance_async(fs_info);
3096 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
3097 close_ctree(tree_root);
3101 ret = btrfs_resume_dev_replace_async(fs_info);
3103 pr_warn("BTRFS: failed to resume dev_replace\n");
3104 close_ctree(tree_root);
3108 btrfs_qgroup_rescan_resume(fs_info);
3110 if (!fs_info->uuid_root) {
3111 pr_info("BTRFS: creating UUID tree\n");
3112 ret = btrfs_create_uuid_tree(fs_info);
3114 pr_warn("BTRFS: failed to create the UUID tree %d\n",
3116 close_ctree(tree_root);
3119 } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3120 fs_info->generation !=
3121 btrfs_super_uuid_tree_generation(disk_super)) {
3122 pr_info("BTRFS: checking UUID tree\n");
3123 ret = btrfs_check_uuid_tree(fs_info);
3125 pr_warn("BTRFS: failed to check the UUID tree %d\n",
3127 close_ctree(tree_root);
3131 fs_info->update_uuid_tree_gen = 1;
3139 btrfs_free_qgroup_config(fs_info);
3141 kthread_stop(fs_info->transaction_kthread);
3142 btrfs_cleanup_transaction(fs_info->tree_root);
3143 btrfs_free_fs_roots(fs_info);
3145 kthread_stop(fs_info->cleaner_kthread);
3148 * make sure we're done with the btree inode before we stop our
3151 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3154 btrfs_sysfs_remove_mounted(fs_info);
3157 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3160 btrfs_put_block_group_cache(fs_info);
3161 btrfs_free_block_groups(fs_info);
3164 free_root_pointers(fs_info, 1);
3165 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3168 btrfs_stop_all_workers(fs_info);
3171 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3173 iput(fs_info->btree_inode);
3175 percpu_counter_destroy(&fs_info->bio_counter);
3176 fail_delalloc_bytes:
3177 percpu_counter_destroy(&fs_info->delalloc_bytes);
3178 fail_dirty_metadata_bytes:
3179 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3181 bdi_destroy(&fs_info->bdi);
3183 cleanup_srcu_struct(&fs_info->subvol_srcu);
3185 btrfs_free_stripe_hash_table(fs_info);
3186 btrfs_close_devices(fs_info->fs_devices);
3190 if (!btrfs_test_opt(tree_root, RECOVERY))
3191 goto fail_tree_roots;
3193 free_root_pointers(fs_info, 0);
3195 /* don't use the log in recovery mode, it won't be valid */
3196 btrfs_set_super_log_root(disk_super, 0);
3198 /* we can't trust the free space cache either */
3199 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3201 ret = next_root_backup(fs_info, fs_info->super_copy,
3202 &num_backups_tried, &backup_index);
3204 goto fail_block_groups;
3205 goto retry_root_backup;
3208 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3211 set_buffer_uptodate(bh);
3213 struct btrfs_device *device = (struct btrfs_device *)
3216 btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3217 "lost page write due to IO error on %s",
3218 rcu_str_deref(device->name));
3219 /* note, we dont' set_buffer_write_io_error because we have
3220 * our own ways of dealing with the IO errors
3222 clear_buffer_uptodate(bh);
3223 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3229 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3230 struct buffer_head **bh_ret)
3232 struct buffer_head *bh;
3233 struct btrfs_super_block *super;
3236 bytenr = btrfs_sb_offset(copy_num);
3237 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3240 bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3242 * If we fail to read from the underlying devices, as of now
3243 * the best option we have is to mark it EIO.
3248 super = (struct btrfs_super_block *)bh->b_data;
3249 if (btrfs_super_bytenr(super) != bytenr ||
3250 btrfs_super_magic(super) != BTRFS_MAGIC) {
3260 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3262 struct buffer_head *bh;
3263 struct buffer_head *latest = NULL;
3264 struct btrfs_super_block *super;
3269 /* we would like to check all the supers, but that would make
3270 * a btrfs mount succeed after a mkfs from a different FS.
3271 * So, we need to add a special mount option to scan for
3272 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3274 for (i = 0; i < 1; i++) {
3275 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3279 super = (struct btrfs_super_block *)bh->b_data;
3281 if (!latest || btrfs_super_generation(super) > transid) {
3284 transid = btrfs_super_generation(super);
3291 return ERR_PTR(ret);
3297 * this should be called twice, once with wait == 0 and
3298 * once with wait == 1. When wait == 0 is done, all the buffer heads
3299 * we write are pinned.
3301 * They are released when wait == 1 is done.
3302 * max_mirrors must be the same for both runs, and it indicates how
3303 * many supers on this one device should be written.
3305 * max_mirrors == 0 means to write them all.
3307 static int write_dev_supers(struct btrfs_device *device,
3308 struct btrfs_super_block *sb,
3309 int do_barriers, int wait, int max_mirrors)
3311 struct buffer_head *bh;
3318 if (max_mirrors == 0)
3319 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3321 for (i = 0; i < max_mirrors; i++) {
3322 bytenr = btrfs_sb_offset(i);
3323 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3324 device->commit_total_bytes)
3328 bh = __find_get_block(device->bdev, bytenr / 4096,
3329 BTRFS_SUPER_INFO_SIZE);
3335 if (!buffer_uptodate(bh))
3338 /* drop our reference */
3341 /* drop the reference from the wait == 0 run */
3345 btrfs_set_super_bytenr(sb, bytenr);
3348 crc = btrfs_csum_data((char *)sb +
3349 BTRFS_CSUM_SIZE, crc,
3350 BTRFS_SUPER_INFO_SIZE -
3352 btrfs_csum_final(crc, sb->csum);
3355 * one reference for us, and we leave it for the
3358 bh = __getblk(device->bdev, bytenr / 4096,
3359 BTRFS_SUPER_INFO_SIZE);
3361 btrfs_err(device->dev_root->fs_info,
3362 "couldn't get super buffer head for bytenr %llu",
3368 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3370 /* one reference for submit_bh */
3373 set_buffer_uptodate(bh);
3375 bh->b_end_io = btrfs_end_buffer_write_sync;
3376 bh->b_private = device;
3380 * we fua the first super. The others we allow
3384 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3386 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3390 return errors < i ? 0 : -1;
3394 * endio for the write_dev_flush, this will wake anyone waiting
3395 * for the barrier when it is done
3397 static void btrfs_end_empty_barrier(struct bio *bio)
3399 if (bio->bi_private)
3400 complete(bio->bi_private);
3405 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3406 * sent down. With wait == 1, it waits for the previous flush.
3408 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3411 static int write_dev_flush(struct btrfs_device *device, int wait)
3416 if (device->nobarriers)
3420 bio = device->flush_bio;
3424 wait_for_completion(&device->flush_wait);
3426 if (bio->bi_error) {
3427 ret = bio->bi_error;
3428 btrfs_dev_stat_inc_and_print(device,
3429 BTRFS_DEV_STAT_FLUSH_ERRS);
3432 /* drop the reference from the wait == 0 run */
3434 device->flush_bio = NULL;
3440 * one reference for us, and we leave it for the
3443 device->flush_bio = NULL;
3444 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3448 bio->bi_end_io = btrfs_end_empty_barrier;
3449 bio->bi_bdev = device->bdev;
3450 init_completion(&device->flush_wait);
3451 bio->bi_private = &device->flush_wait;
3452 device->flush_bio = bio;
3455 btrfsic_submit_bio(WRITE_FLUSH, bio);
3461 * send an empty flush down to each device in parallel,
3462 * then wait for them
3464 static int barrier_all_devices(struct btrfs_fs_info *info)
3466 struct list_head *head;
3467 struct btrfs_device *dev;
3468 int errors_send = 0;
3469 int errors_wait = 0;
3472 /* send down all the barriers */
3473 head = &info->fs_devices->devices;
3474 list_for_each_entry_rcu(dev, head, dev_list) {
3481 if (!dev->in_fs_metadata || !dev->writeable)
3484 ret = write_dev_flush(dev, 0);
3489 /* wait for all the barriers */
3490 list_for_each_entry_rcu(dev, head, dev_list) {
3497 if (!dev->in_fs_metadata || !dev->writeable)
3500 ret = write_dev_flush(dev, 1);
3504 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3505 errors_wait > info->num_tolerated_disk_barrier_failures)
3510 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3513 int min_tolerated = INT_MAX;
3515 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3516 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3517 min_tolerated = min(min_tolerated,
3518 btrfs_raid_array[BTRFS_RAID_SINGLE].
3519 tolerated_failures);
3521 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3522 if (raid_type == BTRFS_RAID_SINGLE)
3524 if (!(flags & btrfs_raid_group[raid_type]))
3526 min_tolerated = min(min_tolerated,
3527 btrfs_raid_array[raid_type].
3528 tolerated_failures);
3531 if (min_tolerated == INT_MAX) {
3532 pr_warn("BTRFS: unknown raid flag: %llu\n", flags);
3536 return min_tolerated;
3539 int btrfs_calc_num_tolerated_disk_barrier_failures(
3540 struct btrfs_fs_info *fs_info)
3542 struct btrfs_ioctl_space_info space;
3543 struct btrfs_space_info *sinfo;
3544 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3545 BTRFS_BLOCK_GROUP_SYSTEM,
3546 BTRFS_BLOCK_GROUP_METADATA,
3547 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3550 int num_tolerated_disk_barrier_failures =
3551 (int)fs_info->fs_devices->num_devices;
3553 for (i = 0; i < ARRAY_SIZE(types); i++) {
3554 struct btrfs_space_info *tmp;
3558 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3559 if (tmp->flags == types[i]) {
3569 down_read(&sinfo->groups_sem);
3570 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3573 if (list_empty(&sinfo->block_groups[c]))
3576 btrfs_get_block_group_info(&sinfo->block_groups[c],
3578 if (space.total_bytes == 0 || space.used_bytes == 0)
3580 flags = space.flags;
3582 num_tolerated_disk_barrier_failures = min(
3583 num_tolerated_disk_barrier_failures,
3584 btrfs_get_num_tolerated_disk_barrier_failures(
3587 up_read(&sinfo->groups_sem);
3590 return num_tolerated_disk_barrier_failures;
3593 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3595 struct list_head *head;
3596 struct btrfs_device *dev;
3597 struct btrfs_super_block *sb;
3598 struct btrfs_dev_item *dev_item;
3602 int total_errors = 0;
3605 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3606 backup_super_roots(root->fs_info);
3608 sb = root->fs_info->super_for_commit;
3609 dev_item = &sb->dev_item;
3611 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3612 head = &root->fs_info->fs_devices->devices;
3613 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3616 ret = barrier_all_devices(root->fs_info);
3619 &root->fs_info->fs_devices->device_list_mutex);
3620 btrfs_std_error(root->fs_info, ret,
3621 "errors while submitting device barriers.");
3626 list_for_each_entry_rcu(dev, head, dev_list) {
3631 if (!dev->in_fs_metadata || !dev->writeable)
3634 btrfs_set_stack_device_generation(dev_item, 0);
3635 btrfs_set_stack_device_type(dev_item, dev->type);
3636 btrfs_set_stack_device_id(dev_item, dev->devid);
3637 btrfs_set_stack_device_total_bytes(dev_item,
3638 dev->commit_total_bytes);
3639 btrfs_set_stack_device_bytes_used(dev_item,
3640 dev->commit_bytes_used);
3641 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3642 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3643 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3644 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3645 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3647 flags = btrfs_super_flags(sb);
3648 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3650 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3654 if (total_errors > max_errors) {
3655 btrfs_err(root->fs_info, "%d errors while writing supers",
3657 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3659 /* FUA is masked off if unsupported and can't be the reason */
3660 btrfs_std_error(root->fs_info, -EIO,
3661 "%d errors while writing supers", total_errors);
3666 list_for_each_entry_rcu(dev, head, dev_list) {
3669 if (!dev->in_fs_metadata || !dev->writeable)
3672 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3676 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3677 if (total_errors > max_errors) {
3678 btrfs_std_error(root->fs_info, -EIO,
3679 "%d errors while writing supers", total_errors);
3685 int write_ctree_super(struct btrfs_trans_handle *trans,
3686 struct btrfs_root *root, int max_mirrors)
3688 return write_all_supers(root, max_mirrors);
3691 /* Drop a fs root from the radix tree and free it. */
3692 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3693 struct btrfs_root *root)
3695 spin_lock(&fs_info->fs_roots_radix_lock);
3696 radix_tree_delete(&fs_info->fs_roots_radix,
3697 (unsigned long)root->root_key.objectid);
3698 spin_unlock(&fs_info->fs_roots_radix_lock);
3700 if (btrfs_root_refs(&root->root_item) == 0)
3701 synchronize_srcu(&fs_info->subvol_srcu);
3703 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3704 btrfs_free_log(NULL, root);
3706 if (root->free_ino_pinned)
3707 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3708 if (root->free_ino_ctl)
3709 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3713 static void free_fs_root(struct btrfs_root *root)
3715 iput(root->ino_cache_inode);
3716 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3717 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3718 root->orphan_block_rsv = NULL;
3720 free_anon_bdev(root->anon_dev);
3721 if (root->subv_writers)
3722 btrfs_free_subvolume_writers(root->subv_writers);
3723 free_extent_buffer(root->node);
3724 free_extent_buffer(root->commit_root);
3725 kfree(root->free_ino_ctl);
3726 kfree(root->free_ino_pinned);
3728 btrfs_put_fs_root(root);
3731 void btrfs_free_fs_root(struct btrfs_root *root)
3736 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3738 u64 root_objectid = 0;
3739 struct btrfs_root *gang[8];
3742 unsigned int ret = 0;
3746 index = srcu_read_lock(&fs_info->subvol_srcu);
3747 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3748 (void **)gang, root_objectid,
3751 srcu_read_unlock(&fs_info->subvol_srcu, index);
3754 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3756 for (i = 0; i < ret; i++) {
3757 /* Avoid to grab roots in dead_roots */
3758 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3762 /* grab all the search result for later use */
3763 gang[i] = btrfs_grab_fs_root(gang[i]);
3765 srcu_read_unlock(&fs_info->subvol_srcu, index);
3767 for (i = 0; i < ret; i++) {
3770 root_objectid = gang[i]->root_key.objectid;
3771 err = btrfs_orphan_cleanup(gang[i]);
3774 btrfs_put_fs_root(gang[i]);
3779 /* release the uncleaned roots due to error */
3780 for (; i < ret; i++) {
3782 btrfs_put_fs_root(gang[i]);
3787 int btrfs_commit_super(struct btrfs_root *root)
3789 struct btrfs_trans_handle *trans;
3791 mutex_lock(&root->fs_info->cleaner_mutex);
3792 btrfs_run_delayed_iputs(root);
3793 mutex_unlock(&root->fs_info->cleaner_mutex);
3794 wake_up_process(root->fs_info->cleaner_kthread);
3796 /* wait until ongoing cleanup work done */
3797 down_write(&root->fs_info->cleanup_work_sem);
3798 up_write(&root->fs_info->cleanup_work_sem);
3800 trans = btrfs_join_transaction(root);
3802 return PTR_ERR(trans);
3803 return btrfs_commit_transaction(trans, root);
3806 void close_ctree(struct btrfs_root *root)
3808 struct btrfs_fs_info *fs_info = root->fs_info;
3811 fs_info->closing = 1;
3814 /* wait for the qgroup rescan worker to stop */
3815 btrfs_qgroup_wait_for_completion(fs_info, false);
3817 /* wait for the uuid_scan task to finish */
3818 down(&fs_info->uuid_tree_rescan_sem);
3819 /* avoid complains from lockdep et al., set sem back to initial state */
3820 up(&fs_info->uuid_tree_rescan_sem);
3822 /* pause restriper - we want to resume on mount */
3823 btrfs_pause_balance(fs_info);
3825 btrfs_dev_replace_suspend_for_unmount(fs_info);
3827 btrfs_scrub_cancel(fs_info);
3829 /* wait for any defraggers to finish */
3830 wait_event(fs_info->transaction_wait,
3831 (atomic_read(&fs_info->defrag_running) == 0));
3833 /* clear out the rbtree of defraggable inodes */
3834 btrfs_cleanup_defrag_inodes(fs_info);
3836 cancel_work_sync(&fs_info->async_reclaim_work);
3838 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3840 * If the cleaner thread is stopped and there are
3841 * block groups queued for removal, the deletion will be
3842 * skipped when we quit the cleaner thread.
3844 btrfs_delete_unused_bgs(root->fs_info);
3846 ret = btrfs_commit_super(root);
3848 btrfs_err(fs_info, "commit super ret %d", ret);
3851 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3852 btrfs_error_commit_super(root);
3854 kthread_stop(fs_info->transaction_kthread);
3855 kthread_stop(fs_info->cleaner_kthread);
3857 fs_info->closing = 2;
3860 btrfs_free_qgroup_config(fs_info);
3862 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3863 btrfs_info(fs_info, "at unmount delalloc count %lld",
3864 percpu_counter_sum(&fs_info->delalloc_bytes));
3867 btrfs_sysfs_remove_mounted(fs_info);
3868 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3870 btrfs_free_fs_roots(fs_info);
3872 btrfs_put_block_group_cache(fs_info);
3874 btrfs_free_block_groups(fs_info);
3877 * we must make sure there is not any read request to
3878 * submit after we stopping all workers.
3880 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3881 btrfs_stop_all_workers(fs_info);
3884 free_root_pointers(fs_info, 1);
3886 iput(fs_info->btree_inode);
3888 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3889 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3890 btrfsic_unmount(root, fs_info->fs_devices);
3893 btrfs_close_devices(fs_info->fs_devices);
3894 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3896 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3897 percpu_counter_destroy(&fs_info->delalloc_bytes);
3898 percpu_counter_destroy(&fs_info->bio_counter);
3899 bdi_destroy(&fs_info->bdi);
3900 cleanup_srcu_struct(&fs_info->subvol_srcu);
3902 btrfs_free_stripe_hash_table(fs_info);
3904 __btrfs_free_block_rsv(root->orphan_block_rsv);
3905 root->orphan_block_rsv = NULL;
3908 while (!list_empty(&fs_info->pinned_chunks)) {
3909 struct extent_map *em;
3911 em = list_first_entry(&fs_info->pinned_chunks,
3912 struct extent_map, list);
3913 list_del_init(&em->list);
3914 free_extent_map(em);
3916 unlock_chunks(root);
3919 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3923 struct inode *btree_inode = buf->pages[0]->mapping->host;
3925 ret = extent_buffer_uptodate(buf);
3929 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3930 parent_transid, atomic);
3936 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3938 return set_extent_buffer_uptodate(buf);
3941 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3943 struct btrfs_root *root;
3944 u64 transid = btrfs_header_generation(buf);
3947 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3949 * This is a fast path so only do this check if we have sanity tests
3950 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3951 * outside of the sanity tests.
3953 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3956 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3957 btrfs_assert_tree_locked(buf);
3958 if (transid != root->fs_info->generation)
3959 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3960 "found %llu running %llu\n",
3961 buf->start, transid, root->fs_info->generation);
3962 was_dirty = set_extent_buffer_dirty(buf);
3964 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3966 root->fs_info->dirty_metadata_batch);
3967 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3968 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3969 btrfs_print_leaf(root, buf);
3975 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3979 * looks as though older kernels can get into trouble with
3980 * this code, they end up stuck in balance_dirty_pages forever
3984 if (current->flags & PF_MEMALLOC)
3988 btrfs_balance_delayed_items(root);
3990 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3991 BTRFS_DIRTY_METADATA_THRESH);
3993 balance_dirty_pages_ratelimited(
3994 root->fs_info->btree_inode->i_mapping);
3999 void btrfs_btree_balance_dirty(struct btrfs_root *root)
4001 __btrfs_btree_balance_dirty(root, 1);
4004 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
4006 __btrfs_btree_balance_dirty(root, 0);
4009 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4011 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4012 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
4015 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4018 struct btrfs_super_block *sb = fs_info->super_copy;
4021 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4022 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
4023 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4026 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4027 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
4028 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4031 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4032 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
4033 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4038 * The common minimum, we don't know if we can trust the nodesize/sectorsize
4039 * items yet, they'll be verified later. Issue just a warning.
4041 if (!IS_ALIGNED(btrfs_super_root(sb), 4096))
4042 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
4043 btrfs_super_root(sb));
4044 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096))
4045 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
4046 btrfs_super_chunk_root(sb));
4047 if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096))
4048 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
4049 btrfs_super_log_root(sb));
4052 * Check the lower bound, the alignment and other constraints are
4055 if (btrfs_super_nodesize(sb) < 4096) {
4056 printk(KERN_ERR "BTRFS: nodesize too small: %u < 4096\n",
4057 btrfs_super_nodesize(sb));
4060 if (btrfs_super_sectorsize(sb) < 4096) {
4061 printk(KERN_ERR "BTRFS: sectorsize too small: %u < 4096\n",
4062 btrfs_super_sectorsize(sb));
4066 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4067 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4068 fs_info->fsid, sb->dev_item.fsid);
4073 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4076 if (btrfs_super_num_devices(sb) > (1UL << 31))
4077 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4078 btrfs_super_num_devices(sb));
4079 if (btrfs_super_num_devices(sb) == 0) {
4080 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4084 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4085 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4086 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4091 * Obvious sys_chunk_array corruptions, it must hold at least one key
4094 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4095 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4096 btrfs_super_sys_array_size(sb),
4097 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4100 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4101 + sizeof(struct btrfs_chunk)) {
4102 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4103 btrfs_super_sys_array_size(sb),
4104 sizeof(struct btrfs_disk_key)
4105 + sizeof(struct btrfs_chunk));
4110 * The generation is a global counter, we'll trust it more than the others
4111 * but it's still possible that it's the one that's wrong.
4113 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4115 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4116 btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4117 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4118 && btrfs_super_cache_generation(sb) != (u64)-1)
4120 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4121 btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4126 static void btrfs_error_commit_super(struct btrfs_root *root)
4128 mutex_lock(&root->fs_info->cleaner_mutex);
4129 btrfs_run_delayed_iputs(root);
4130 mutex_unlock(&root->fs_info->cleaner_mutex);
4132 down_write(&root->fs_info->cleanup_work_sem);
4133 up_write(&root->fs_info->cleanup_work_sem);
4135 /* cleanup FS via transaction */
4136 btrfs_cleanup_transaction(root);
4139 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4141 struct btrfs_ordered_extent *ordered;
4143 spin_lock(&root->ordered_extent_lock);
4145 * This will just short circuit the ordered completion stuff which will
4146 * make sure the ordered extent gets properly cleaned up.
4148 list_for_each_entry(ordered, &root->ordered_extents,
4150 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4151 spin_unlock(&root->ordered_extent_lock);
4154 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4156 struct btrfs_root *root;
4157 struct list_head splice;
4159 INIT_LIST_HEAD(&splice);
4161 spin_lock(&fs_info->ordered_root_lock);
4162 list_splice_init(&fs_info->ordered_roots, &splice);
4163 while (!list_empty(&splice)) {
4164 root = list_first_entry(&splice, struct btrfs_root,
4166 list_move_tail(&root->ordered_root,
4167 &fs_info->ordered_roots);
4169 spin_unlock(&fs_info->ordered_root_lock);
4170 btrfs_destroy_ordered_extents(root);
4173 spin_lock(&fs_info->ordered_root_lock);
4175 spin_unlock(&fs_info->ordered_root_lock);
4178 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4179 struct btrfs_root *root)
4181 struct rb_node *node;
4182 struct btrfs_delayed_ref_root *delayed_refs;
4183 struct btrfs_delayed_ref_node *ref;
4186 delayed_refs = &trans->delayed_refs;
4188 spin_lock(&delayed_refs->lock);
4189 if (atomic_read(&delayed_refs->num_entries) == 0) {
4190 spin_unlock(&delayed_refs->lock);
4191 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4195 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4196 struct btrfs_delayed_ref_head *head;
4197 struct btrfs_delayed_ref_node *tmp;
4198 bool pin_bytes = false;
4200 head = rb_entry(node, struct btrfs_delayed_ref_head,
4202 if (!mutex_trylock(&head->mutex)) {
4203 atomic_inc(&head->node.refs);
4204 spin_unlock(&delayed_refs->lock);
4206 mutex_lock(&head->mutex);
4207 mutex_unlock(&head->mutex);
4208 btrfs_put_delayed_ref(&head->node);
4209 spin_lock(&delayed_refs->lock);
4212 spin_lock(&head->lock);
4213 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4216 list_del(&ref->list);
4217 atomic_dec(&delayed_refs->num_entries);
4218 btrfs_put_delayed_ref(ref);
4220 if (head->must_insert_reserved)
4222 btrfs_free_delayed_extent_op(head->extent_op);
4223 delayed_refs->num_heads--;
4224 if (head->processing == 0)
4225 delayed_refs->num_heads_ready--;
4226 atomic_dec(&delayed_refs->num_entries);
4227 head->node.in_tree = 0;
4228 rb_erase(&head->href_node, &delayed_refs->href_root);
4229 spin_unlock(&head->lock);
4230 spin_unlock(&delayed_refs->lock);
4231 mutex_unlock(&head->mutex);
4234 btrfs_pin_extent(root, head->node.bytenr,
4235 head->node.num_bytes, 1);
4236 btrfs_put_delayed_ref(&head->node);
4238 spin_lock(&delayed_refs->lock);
4241 spin_unlock(&delayed_refs->lock);
4246 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4248 struct btrfs_inode *btrfs_inode;
4249 struct list_head splice;
4251 INIT_LIST_HEAD(&splice);
4253 spin_lock(&root->delalloc_lock);
4254 list_splice_init(&root->delalloc_inodes, &splice);
4256 while (!list_empty(&splice)) {
4257 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4260 list_del_init(&btrfs_inode->delalloc_inodes);
4261 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4262 &btrfs_inode->runtime_flags);
4263 spin_unlock(&root->delalloc_lock);
4265 btrfs_invalidate_inodes(btrfs_inode->root);
4267 spin_lock(&root->delalloc_lock);
4270 spin_unlock(&root->delalloc_lock);
4273 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4275 struct btrfs_root *root;
4276 struct list_head splice;
4278 INIT_LIST_HEAD(&splice);
4280 spin_lock(&fs_info->delalloc_root_lock);
4281 list_splice_init(&fs_info->delalloc_roots, &splice);
4282 while (!list_empty(&splice)) {
4283 root = list_first_entry(&splice, struct btrfs_root,
4285 list_del_init(&root->delalloc_root);
4286 root = btrfs_grab_fs_root(root);
4288 spin_unlock(&fs_info->delalloc_root_lock);
4290 btrfs_destroy_delalloc_inodes(root);
4291 btrfs_put_fs_root(root);
4293 spin_lock(&fs_info->delalloc_root_lock);
4295 spin_unlock(&fs_info->delalloc_root_lock);
4298 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4299 struct extent_io_tree *dirty_pages,
4303 struct extent_buffer *eb;
4308 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4313 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4314 while (start <= end) {
4315 eb = btrfs_find_tree_block(root->fs_info, start);
4316 start += root->nodesize;
4319 wait_on_extent_buffer_writeback(eb);
4321 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4323 clear_extent_buffer_dirty(eb);
4324 free_extent_buffer_stale(eb);
4331 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4332 struct extent_io_tree *pinned_extents)
4334 struct extent_io_tree *unpin;
4340 unpin = pinned_extents;
4343 ret = find_first_extent_bit(unpin, 0, &start, &end,
4344 EXTENT_DIRTY, NULL);
4348 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4349 btrfs_error_unpin_extent_range(root, start, end);
4354 if (unpin == &root->fs_info->freed_extents[0])
4355 unpin = &root->fs_info->freed_extents[1];
4357 unpin = &root->fs_info->freed_extents[0];
4365 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4366 struct btrfs_root *root)
4368 btrfs_destroy_delayed_refs(cur_trans, root);
4370 cur_trans->state = TRANS_STATE_COMMIT_START;
4371 wake_up(&root->fs_info->transaction_blocked_wait);
4373 cur_trans->state = TRANS_STATE_UNBLOCKED;
4374 wake_up(&root->fs_info->transaction_wait);
4376 btrfs_destroy_delayed_inodes(root);
4377 btrfs_assert_delayed_root_empty(root);
4379 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4381 btrfs_destroy_pinned_extent(root,
4382 root->fs_info->pinned_extents);
4384 cur_trans->state =TRANS_STATE_COMPLETED;
4385 wake_up(&cur_trans->commit_wait);
4388 memset(cur_trans, 0, sizeof(*cur_trans));
4389 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4393 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4395 struct btrfs_transaction *t;
4397 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4399 spin_lock(&root->fs_info->trans_lock);
4400 while (!list_empty(&root->fs_info->trans_list)) {
4401 t = list_first_entry(&root->fs_info->trans_list,
4402 struct btrfs_transaction, list);
4403 if (t->state >= TRANS_STATE_COMMIT_START) {
4404 atomic_inc(&t->use_count);
4405 spin_unlock(&root->fs_info->trans_lock);
4406 btrfs_wait_for_commit(root, t->transid);
4407 btrfs_put_transaction(t);
4408 spin_lock(&root->fs_info->trans_lock);
4411 if (t == root->fs_info->running_transaction) {
4412 t->state = TRANS_STATE_COMMIT_DOING;
4413 spin_unlock(&root->fs_info->trans_lock);
4415 * We wait for 0 num_writers since we don't hold a trans
4416 * handle open currently for this transaction.
4418 wait_event(t->writer_wait,
4419 atomic_read(&t->num_writers) == 0);
4421 spin_unlock(&root->fs_info->trans_lock);
4423 btrfs_cleanup_one_transaction(t, root);
4425 spin_lock(&root->fs_info->trans_lock);
4426 if (t == root->fs_info->running_transaction)
4427 root->fs_info->running_transaction = NULL;
4428 list_del_init(&t->list);
4429 spin_unlock(&root->fs_info->trans_lock);
4431 btrfs_put_transaction(t);
4432 trace_btrfs_transaction_commit(root);
4433 spin_lock(&root->fs_info->trans_lock);
4435 spin_unlock(&root->fs_info->trans_lock);
4436 btrfs_destroy_all_ordered_extents(root->fs_info);
4437 btrfs_destroy_delayed_inodes(root);
4438 btrfs_assert_delayed_root_empty(root);
4439 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4440 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4441 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4446 static const struct extent_io_ops btree_extent_io_ops = {
4447 .readpage_end_io_hook = btree_readpage_end_io_hook,
4448 .readpage_io_failed_hook = btree_io_failed_hook,
4449 .submit_bio_hook = btree_submit_bio_hook,
4450 /* note we're sharing with inode.c for the merge bio hook */
4451 .merge_bio_hook = btrfs_merge_bio_hook,