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/slab.h>
29 #include <linux/migrate.h>
30 #include <linux/ratelimit.h>
31 #include <linux/uuid.h>
32 #include <linux/semaphore.h>
33 #include <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
43 #include "free-space-cache.h"
44 #include "free-space-tree.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
52 #include "compression.h"
55 #include <asm/cpufeature.h>
58 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
59 BTRFS_HEADER_FLAG_RELOC |\
60 BTRFS_SUPER_FLAG_ERROR |\
61 BTRFS_SUPER_FLAG_SEEDING |\
62 BTRFS_SUPER_FLAG_METADUMP)
64 static const struct extent_io_ops btree_extent_io_ops;
65 static void end_workqueue_fn(struct btrfs_work *work);
66 static void free_fs_root(struct btrfs_root *root);
67 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
69 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
70 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
71 struct btrfs_root *root);
72 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
73 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
74 struct extent_io_tree *dirty_pages,
76 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
77 struct extent_io_tree *pinned_extents);
78 static int btrfs_cleanup_transaction(struct btrfs_root *root);
79 static void btrfs_error_commit_super(struct btrfs_root *root);
82 * btrfs_end_io_wq structs are used to do processing in task context when an IO
83 * is complete. This is used during reads to verify checksums, and it is used
84 * by writes to insert metadata for new file extents after IO is complete.
86 struct btrfs_end_io_wq {
90 struct btrfs_fs_info *info;
92 enum btrfs_wq_endio_type metadata;
93 struct list_head list;
94 struct btrfs_work work;
97 static struct kmem_cache *btrfs_end_io_wq_cache;
99 int __init btrfs_end_io_wq_init(void)
101 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
102 sizeof(struct btrfs_end_io_wq),
104 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
106 if (!btrfs_end_io_wq_cache)
111 void btrfs_end_io_wq_exit(void)
113 kmem_cache_destroy(btrfs_end_io_wq_cache);
117 * async submit bios are used to offload expensive checksumming
118 * onto the worker threads. They checksum file and metadata bios
119 * just before they are sent down the IO stack.
121 struct async_submit_bio {
124 struct list_head list;
125 extent_submit_bio_hook_t *submit_bio_start;
126 extent_submit_bio_hook_t *submit_bio_done;
128 unsigned long bio_flags;
130 * bio_offset is optional, can be used if the pages in the bio
131 * can't tell us where in the file the bio should go
134 struct btrfs_work work;
139 * Lockdep class keys for extent_buffer->lock's in this root. For a given
140 * eb, the lockdep key is determined by the btrfs_root it belongs to and
141 * the level the eb occupies in the tree.
143 * Different roots are used for different purposes and may nest inside each
144 * other and they require separate keysets. As lockdep keys should be
145 * static, assign keysets according to the purpose of the root as indicated
146 * by btrfs_root->objectid. This ensures that all special purpose roots
147 * have separate keysets.
149 * Lock-nesting across peer nodes is always done with the immediate parent
150 * node locked thus preventing deadlock. As lockdep doesn't know this, use
151 * subclass to avoid triggering lockdep warning in such cases.
153 * The key is set by the readpage_end_io_hook after the buffer has passed
154 * csum validation but before the pages are unlocked. It is also set by
155 * btrfs_init_new_buffer on freshly allocated blocks.
157 * We also add a check to make sure the highest level of the tree is the
158 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
159 * needs update as well.
161 #ifdef CONFIG_DEBUG_LOCK_ALLOC
162 # if BTRFS_MAX_LEVEL != 8
166 static struct btrfs_lockdep_keyset {
167 u64 id; /* root objectid */
168 const char *name_stem; /* lock name stem */
169 char names[BTRFS_MAX_LEVEL + 1][20];
170 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
171 } btrfs_lockdep_keysets[] = {
172 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
173 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
174 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
175 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
176 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
177 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
178 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
179 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
180 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
181 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
182 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
183 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
184 { .id = 0, .name_stem = "tree" },
187 void __init btrfs_init_lockdep(void)
191 /* initialize lockdep class names */
192 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
193 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
195 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
196 snprintf(ks->names[j], sizeof(ks->names[j]),
197 "btrfs-%s-%02d", ks->name_stem, j);
201 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
204 struct btrfs_lockdep_keyset *ks;
206 BUG_ON(level >= ARRAY_SIZE(ks->keys));
208 /* find the matching keyset, id 0 is the default entry */
209 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
210 if (ks->id == objectid)
213 lockdep_set_class_and_name(&eb->lock,
214 &ks->keys[level], ks->names[level]);
220 * extents on the btree inode are pretty simple, there's one extent
221 * that covers the entire device
223 static struct extent_map *btree_get_extent(struct inode *inode,
224 struct page *page, size_t pg_offset, u64 start, u64 len,
227 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
228 struct extent_map *em;
231 read_lock(&em_tree->lock);
232 em = lookup_extent_mapping(em_tree, start, len);
235 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
236 read_unlock(&em_tree->lock);
239 read_unlock(&em_tree->lock);
241 em = alloc_extent_map();
243 em = ERR_PTR(-ENOMEM);
248 em->block_len = (u64)-1;
250 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
252 write_lock(&em_tree->lock);
253 ret = add_extent_mapping(em_tree, em, 0);
254 if (ret == -EEXIST) {
256 em = lookup_extent_mapping(em_tree, start, len);
263 write_unlock(&em_tree->lock);
269 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
271 return btrfs_crc32c(seed, data, len);
274 void btrfs_csum_final(u32 crc, char *result)
276 put_unaligned_le32(~crc, result);
280 * compute the csum for a btree block, and either verify it or write it
281 * into the csum field of the block.
283 static int csum_tree_block(struct btrfs_fs_info *fs_info,
284 struct extent_buffer *buf,
287 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
290 unsigned long cur_len;
291 unsigned long offset = BTRFS_CSUM_SIZE;
293 unsigned long map_start;
294 unsigned long map_len;
297 unsigned long inline_result;
299 len = buf->len - offset;
301 err = map_private_extent_buffer(buf, offset, 32,
302 &kaddr, &map_start, &map_len);
305 cur_len = min(len, map_len - (offset - map_start));
306 crc = btrfs_csum_data(kaddr + offset - map_start,
311 if (csum_size > sizeof(inline_result)) {
312 result = kzalloc(csum_size, GFP_NOFS);
316 result = (char *)&inline_result;
319 btrfs_csum_final(crc, result);
322 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
325 memcpy(&found, result, csum_size);
327 read_extent_buffer(buf, &val, 0, csum_size);
328 btrfs_warn_rl(fs_info,
329 "%s checksum verify failed on %llu wanted %X found %X "
331 fs_info->sb->s_id, buf->start,
332 val, found, btrfs_header_level(buf));
333 if (result != (char *)&inline_result)
338 write_extent_buffer(buf, result, 0, csum_size);
340 if (result != (char *)&inline_result)
346 * we can't consider a given block up to date unless the transid of the
347 * block matches the transid in the parent node's pointer. This is how we
348 * detect blocks that either didn't get written at all or got written
349 * in the wrong place.
351 static int verify_parent_transid(struct extent_io_tree *io_tree,
352 struct extent_buffer *eb, u64 parent_transid,
355 struct extent_state *cached_state = NULL;
357 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
359 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
366 btrfs_tree_read_lock(eb);
367 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
370 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
372 if (extent_buffer_uptodate(eb) &&
373 btrfs_header_generation(eb) == parent_transid) {
377 btrfs_err_rl(eb->fs_info,
378 "parent transid verify failed on %llu wanted %llu found %llu",
380 parent_transid, btrfs_header_generation(eb));
384 * Things reading via commit roots that don't have normal protection,
385 * like send, can have a really old block in cache that may point at a
386 * block that has been freed and re-allocated. So don't clear uptodate
387 * if we find an eb that is under IO (dirty/writeback) because we could
388 * end up reading in the stale data and then writing it back out and
389 * making everybody very sad.
391 if (!extent_buffer_under_io(eb))
392 clear_extent_buffer_uptodate(eb);
394 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
395 &cached_state, GFP_NOFS);
397 btrfs_tree_read_unlock_blocking(eb);
402 * Return 0 if the superblock checksum type matches the checksum value of that
403 * algorithm. Pass the raw disk superblock data.
405 static int btrfs_check_super_csum(char *raw_disk_sb)
407 struct btrfs_super_block *disk_sb =
408 (struct btrfs_super_block *)raw_disk_sb;
409 u16 csum_type = btrfs_super_csum_type(disk_sb);
412 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
414 const int csum_size = sizeof(crc);
415 char result[csum_size];
418 * The super_block structure does not span the whole
419 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
420 * is filled with zeros and is included in the checksum.
422 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
423 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
424 btrfs_csum_final(crc, result);
426 if (memcmp(raw_disk_sb, result, csum_size))
430 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
431 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
440 * helper to read a given tree block, doing retries as required when
441 * the checksums don't match and we have alternate mirrors to try.
443 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
444 struct extent_buffer *eb,
445 u64 start, u64 parent_transid)
447 struct extent_io_tree *io_tree;
452 int failed_mirror = 0;
454 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
455 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
457 ret = read_extent_buffer_pages(io_tree, eb, start,
459 btree_get_extent, mirror_num);
461 if (!verify_parent_transid(io_tree, eb,
469 * This buffer's crc is fine, but its contents are corrupted, so
470 * there is no reason to read the other copies, they won't be
473 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
476 num_copies = btrfs_num_copies(root->fs_info,
481 if (!failed_mirror) {
483 failed_mirror = eb->read_mirror;
487 if (mirror_num == failed_mirror)
490 if (mirror_num > num_copies)
494 if (failed && !ret && failed_mirror)
495 repair_eb_io_failure(root, eb, failed_mirror);
501 * checksum a dirty tree block before IO. This has extra checks to make sure
502 * we only fill in the checksum field in the first page of a multi-page block
505 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
507 u64 start = page_offset(page);
509 struct extent_buffer *eb;
511 eb = (struct extent_buffer *)page->private;
512 if (page != eb->pages[0])
515 found_start = btrfs_header_bytenr(eb);
517 * Please do not consolidate these warnings into a single if.
518 * It is useful to know what went wrong.
520 if (WARN_ON(found_start != start))
522 if (WARN_ON(!PageUptodate(page)))
525 ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
526 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
528 return csum_tree_block(fs_info, eb, 0);
531 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
532 struct extent_buffer *eb)
534 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
535 u8 fsid[BTRFS_UUID_SIZE];
538 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
540 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
544 fs_devices = fs_devices->seed;
549 #define CORRUPT(reason, eb, root, slot) \
550 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
551 "root=%llu, slot=%d", reason, \
552 btrfs_header_bytenr(eb), root->objectid, slot)
554 static noinline int check_leaf(struct btrfs_root *root,
555 struct extent_buffer *leaf)
557 struct btrfs_key key;
558 struct btrfs_key leaf_key;
559 u32 nritems = btrfs_header_nritems(leaf);
565 /* Check the 0 item */
566 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
567 BTRFS_LEAF_DATA_SIZE(root)) {
568 CORRUPT("invalid item offset size pair", leaf, root, 0);
573 * Check to make sure each items keys are in the correct order and their
574 * offsets make sense. We only have to loop through nritems-1 because
575 * we check the current slot against the next slot, which verifies the
576 * next slot's offset+size makes sense and that the current's slot
579 for (slot = 0; slot < nritems - 1; slot++) {
580 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
581 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
583 /* Make sure the keys are in the right order */
584 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
585 CORRUPT("bad key order", leaf, root, slot);
590 * Make sure the offset and ends are right, remember that the
591 * item data starts at the end of the leaf and grows towards the
594 if (btrfs_item_offset_nr(leaf, slot) !=
595 btrfs_item_end_nr(leaf, slot + 1)) {
596 CORRUPT("slot offset bad", leaf, root, slot);
601 * Check to make sure that we don't point outside of the leaf,
602 * just in case all the items are consistent to each other, but
603 * all point outside of the leaf.
605 if (btrfs_item_end_nr(leaf, slot) >
606 BTRFS_LEAF_DATA_SIZE(root)) {
607 CORRUPT("slot end outside of leaf", leaf, root, slot);
615 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
616 u64 phy_offset, struct page *page,
617 u64 start, u64 end, int mirror)
621 struct extent_buffer *eb;
622 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
623 struct btrfs_fs_info *fs_info = root->fs_info;
630 eb = (struct extent_buffer *)page->private;
632 /* the pending IO might have been the only thing that kept this buffer
633 * in memory. Make sure we have a ref for all this other checks
635 extent_buffer_get(eb);
637 reads_done = atomic_dec_and_test(&eb->io_pages);
641 eb->read_mirror = mirror;
642 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
647 found_start = btrfs_header_bytenr(eb);
648 if (found_start != eb->start) {
649 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
650 found_start, eb->start);
654 if (check_tree_block_fsid(fs_info, eb)) {
655 btrfs_err_rl(fs_info, "bad fsid on block %llu",
660 found_level = btrfs_header_level(eb);
661 if (found_level >= BTRFS_MAX_LEVEL) {
662 btrfs_err(fs_info, "bad tree block level %d",
663 (int)btrfs_header_level(eb));
668 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
671 ret = csum_tree_block(fs_info, eb, 1);
676 * If this is a leaf block and it is corrupt, set the corrupt bit so
677 * that we don't try and read the other copies of this block, just
680 if (found_level == 0 && check_leaf(root, eb)) {
681 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
686 set_extent_buffer_uptodate(eb);
689 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
690 btree_readahead_hook(fs_info, eb, eb->start, ret);
694 * our io error hook is going to dec the io pages
695 * again, we have to make sure it has something
698 atomic_inc(&eb->io_pages);
699 clear_extent_buffer_uptodate(eb);
701 free_extent_buffer(eb);
706 static int btree_io_failed_hook(struct page *page, int failed_mirror)
708 struct extent_buffer *eb;
710 eb = (struct extent_buffer *)page->private;
711 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
712 eb->read_mirror = failed_mirror;
713 atomic_dec(&eb->io_pages);
714 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
715 btree_readahead_hook(eb->fs_info, eb, eb->start, -EIO);
716 return -EIO; /* we fixed nothing */
719 static void end_workqueue_bio(struct bio *bio)
721 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
722 struct btrfs_fs_info *fs_info;
723 struct btrfs_workqueue *wq;
724 btrfs_work_func_t func;
726 fs_info = end_io_wq->info;
727 end_io_wq->error = bio->bi_error;
729 if (bio_op(bio) == REQ_OP_WRITE) {
730 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
731 wq = fs_info->endio_meta_write_workers;
732 func = btrfs_endio_meta_write_helper;
733 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
734 wq = fs_info->endio_freespace_worker;
735 func = btrfs_freespace_write_helper;
736 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
737 wq = fs_info->endio_raid56_workers;
738 func = btrfs_endio_raid56_helper;
740 wq = fs_info->endio_write_workers;
741 func = btrfs_endio_write_helper;
744 if (unlikely(end_io_wq->metadata ==
745 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
746 wq = fs_info->endio_repair_workers;
747 func = btrfs_endio_repair_helper;
748 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
749 wq = fs_info->endio_raid56_workers;
750 func = btrfs_endio_raid56_helper;
751 } else if (end_io_wq->metadata) {
752 wq = fs_info->endio_meta_workers;
753 func = btrfs_endio_meta_helper;
755 wq = fs_info->endio_workers;
756 func = btrfs_endio_helper;
760 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
761 btrfs_queue_work(wq, &end_io_wq->work);
764 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
765 enum btrfs_wq_endio_type metadata)
767 struct btrfs_end_io_wq *end_io_wq;
769 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
773 end_io_wq->private = bio->bi_private;
774 end_io_wq->end_io = bio->bi_end_io;
775 end_io_wq->info = info;
776 end_io_wq->error = 0;
777 end_io_wq->bio = bio;
778 end_io_wq->metadata = metadata;
780 bio->bi_private = end_io_wq;
781 bio->bi_end_io = end_workqueue_bio;
785 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
787 unsigned long limit = min_t(unsigned long,
788 info->thread_pool_size,
789 info->fs_devices->open_devices);
793 static void run_one_async_start(struct btrfs_work *work)
795 struct async_submit_bio *async;
798 async = container_of(work, struct async_submit_bio, work);
799 ret = async->submit_bio_start(async->inode, async->bio,
800 async->mirror_num, async->bio_flags,
806 static void run_one_async_done(struct btrfs_work *work)
808 struct btrfs_fs_info *fs_info;
809 struct async_submit_bio *async;
812 async = container_of(work, struct async_submit_bio, work);
813 fs_info = BTRFS_I(async->inode)->root->fs_info;
815 limit = btrfs_async_submit_limit(fs_info);
816 limit = limit * 2 / 3;
819 * atomic_dec_return implies a barrier for waitqueue_active
821 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
822 waitqueue_active(&fs_info->async_submit_wait))
823 wake_up(&fs_info->async_submit_wait);
825 /* If an error occurred we just want to clean up the bio and move on */
827 async->bio->bi_error = async->error;
828 bio_endio(async->bio);
832 async->submit_bio_done(async->inode, async->bio, async->mirror_num,
833 async->bio_flags, async->bio_offset);
836 static void run_one_async_free(struct btrfs_work *work)
838 struct async_submit_bio *async;
840 async = container_of(work, struct async_submit_bio, work);
844 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
845 struct bio *bio, int mirror_num,
846 unsigned long bio_flags,
848 extent_submit_bio_hook_t *submit_bio_start,
849 extent_submit_bio_hook_t *submit_bio_done)
851 struct async_submit_bio *async;
853 async = kmalloc(sizeof(*async), GFP_NOFS);
857 async->inode = inode;
859 async->mirror_num = mirror_num;
860 async->submit_bio_start = submit_bio_start;
861 async->submit_bio_done = submit_bio_done;
863 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
864 run_one_async_done, run_one_async_free);
866 async->bio_flags = bio_flags;
867 async->bio_offset = bio_offset;
871 atomic_inc(&fs_info->nr_async_submits);
873 if (bio->bi_rw & REQ_SYNC)
874 btrfs_set_work_high_priority(&async->work);
876 btrfs_queue_work(fs_info->workers, &async->work);
878 while (atomic_read(&fs_info->async_submit_draining) &&
879 atomic_read(&fs_info->nr_async_submits)) {
880 wait_event(fs_info->async_submit_wait,
881 (atomic_read(&fs_info->nr_async_submits) == 0));
887 static int btree_csum_one_bio(struct bio *bio)
889 struct bio_vec *bvec;
890 struct btrfs_root *root;
893 bio_for_each_segment_all(bvec, bio, i) {
894 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
895 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
903 static int __btree_submit_bio_start(struct inode *inode, struct bio *bio,
904 int mirror_num, unsigned long bio_flags,
908 * when we're called for a write, we're already in the async
909 * submission context. Just jump into btrfs_map_bio
911 return btree_csum_one_bio(bio);
914 static int __btree_submit_bio_done(struct inode *inode, struct bio *bio,
915 int mirror_num, unsigned long bio_flags,
921 * when we're called for a write, we're already in the async
922 * submission context. Just jump into btrfs_map_bio
924 ret = btrfs_map_bio(BTRFS_I(inode)->root, bio, mirror_num, 1);
932 static int check_async_write(struct inode *inode, unsigned long bio_flags)
934 if (bio_flags & EXTENT_BIO_TREE_LOG)
937 if (static_cpu_has(X86_FEATURE_XMM4_2))
943 static int btree_submit_bio_hook(struct inode *inode, struct bio *bio,
944 int mirror_num, unsigned long bio_flags,
947 int async = check_async_write(inode, bio_flags);
950 if (bio_op(bio) != REQ_OP_WRITE) {
952 * called for a read, do the setup so that checksum validation
953 * can happen in the async kernel threads
955 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
956 bio, BTRFS_WQ_ENDIO_METADATA);
959 ret = btrfs_map_bio(BTRFS_I(inode)->root, bio, mirror_num, 0);
961 ret = btree_csum_one_bio(bio);
964 ret = btrfs_map_bio(BTRFS_I(inode)->root, bio, mirror_num, 0);
967 * kthread helpers are used to submit writes so that
968 * checksumming can happen in parallel across all CPUs
970 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
971 inode, bio, mirror_num, 0,
973 __btree_submit_bio_start,
974 __btree_submit_bio_done);
987 #ifdef CONFIG_MIGRATION
988 static int btree_migratepage(struct address_space *mapping,
989 struct page *newpage, struct page *page,
990 enum migrate_mode mode)
993 * we can't safely write a btree page from here,
994 * we haven't done the locking hook
999 * Buffers may be managed in a filesystem specific way.
1000 * We must have no buffers or drop them.
1002 if (page_has_private(page) &&
1003 !try_to_release_page(page, GFP_KERNEL))
1005 return migrate_page(mapping, newpage, page, mode);
1010 static int btree_writepages(struct address_space *mapping,
1011 struct writeback_control *wbc)
1013 struct btrfs_fs_info *fs_info;
1016 if (wbc->sync_mode == WB_SYNC_NONE) {
1018 if (wbc->for_kupdate)
1021 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1022 /* this is a bit racy, but that's ok */
1023 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1024 BTRFS_DIRTY_METADATA_THRESH);
1028 return btree_write_cache_pages(mapping, wbc);
1031 static int btree_readpage(struct file *file, struct page *page)
1033 struct extent_io_tree *tree;
1034 tree = &BTRFS_I(page->mapping->host)->io_tree;
1035 return extent_read_full_page(tree, page, btree_get_extent, 0);
1038 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1040 if (PageWriteback(page) || PageDirty(page))
1043 return try_release_extent_buffer(page);
1046 static void btree_invalidatepage(struct page *page, unsigned int offset,
1047 unsigned int length)
1049 struct extent_io_tree *tree;
1050 tree = &BTRFS_I(page->mapping->host)->io_tree;
1051 extent_invalidatepage(tree, page, offset);
1052 btree_releasepage(page, GFP_NOFS);
1053 if (PagePrivate(page)) {
1054 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1055 "page private not zero on page %llu",
1056 (unsigned long long)page_offset(page));
1057 ClearPagePrivate(page);
1058 set_page_private(page, 0);
1063 static int btree_set_page_dirty(struct page *page)
1066 struct extent_buffer *eb;
1068 BUG_ON(!PagePrivate(page));
1069 eb = (struct extent_buffer *)page->private;
1071 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1072 BUG_ON(!atomic_read(&eb->refs));
1073 btrfs_assert_tree_locked(eb);
1075 return __set_page_dirty_nobuffers(page);
1078 static const struct address_space_operations btree_aops = {
1079 .readpage = btree_readpage,
1080 .writepages = btree_writepages,
1081 .releasepage = btree_releasepage,
1082 .invalidatepage = btree_invalidatepage,
1083 #ifdef CONFIG_MIGRATION
1084 .migratepage = btree_migratepage,
1086 .set_page_dirty = btree_set_page_dirty,
1089 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1091 struct extent_buffer *buf = NULL;
1092 struct inode *btree_inode = root->fs_info->btree_inode;
1094 buf = btrfs_find_create_tree_block(root, bytenr);
1097 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1098 buf, 0, WAIT_NONE, btree_get_extent, 0);
1099 free_extent_buffer(buf);
1102 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1103 int mirror_num, struct extent_buffer **eb)
1105 struct extent_buffer *buf = NULL;
1106 struct inode *btree_inode = root->fs_info->btree_inode;
1107 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1110 buf = btrfs_find_create_tree_block(root, bytenr);
1114 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1116 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1117 btree_get_extent, mirror_num);
1119 free_extent_buffer(buf);
1123 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1124 free_extent_buffer(buf);
1126 } else if (extent_buffer_uptodate(buf)) {
1129 free_extent_buffer(buf);
1134 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1137 return find_extent_buffer(fs_info, bytenr);
1140 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1143 if (btrfs_test_is_dummy_root(root))
1144 return alloc_test_extent_buffer(root->fs_info, bytenr);
1145 return alloc_extent_buffer(root->fs_info, bytenr);
1149 int btrfs_write_tree_block(struct extent_buffer *buf)
1151 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1152 buf->start + buf->len - 1);
1155 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1157 return filemap_fdatawait_range(buf->pages[0]->mapping,
1158 buf->start, buf->start + buf->len - 1);
1161 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1164 struct extent_buffer *buf = NULL;
1167 buf = btrfs_find_create_tree_block(root, bytenr);
1169 return ERR_PTR(-ENOMEM);
1171 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1173 free_extent_buffer(buf);
1174 return ERR_PTR(ret);
1180 void clean_tree_block(struct btrfs_trans_handle *trans,
1181 struct btrfs_fs_info *fs_info,
1182 struct extent_buffer *buf)
1184 if (btrfs_header_generation(buf) ==
1185 fs_info->running_transaction->transid) {
1186 btrfs_assert_tree_locked(buf);
1188 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1189 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1191 fs_info->dirty_metadata_batch);
1192 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1193 btrfs_set_lock_blocking(buf);
1194 clear_extent_buffer_dirty(buf);
1199 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1201 struct btrfs_subvolume_writers *writers;
1204 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1206 return ERR_PTR(-ENOMEM);
1208 ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1211 return ERR_PTR(ret);
1214 init_waitqueue_head(&writers->wait);
1219 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1221 percpu_counter_destroy(&writers->counter);
1225 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1226 struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1230 root->commit_root = NULL;
1231 root->sectorsize = sectorsize;
1232 root->nodesize = nodesize;
1233 root->stripesize = stripesize;
1235 root->orphan_cleanup_state = 0;
1237 root->objectid = objectid;
1238 root->last_trans = 0;
1239 root->highest_objectid = 0;
1240 root->nr_delalloc_inodes = 0;
1241 root->nr_ordered_extents = 0;
1243 root->inode_tree = RB_ROOT;
1244 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1245 root->block_rsv = NULL;
1246 root->orphan_block_rsv = NULL;
1248 INIT_LIST_HEAD(&root->dirty_list);
1249 INIT_LIST_HEAD(&root->root_list);
1250 INIT_LIST_HEAD(&root->delalloc_inodes);
1251 INIT_LIST_HEAD(&root->delalloc_root);
1252 INIT_LIST_HEAD(&root->ordered_extents);
1253 INIT_LIST_HEAD(&root->ordered_root);
1254 INIT_LIST_HEAD(&root->logged_list[0]);
1255 INIT_LIST_HEAD(&root->logged_list[1]);
1256 spin_lock_init(&root->orphan_lock);
1257 spin_lock_init(&root->inode_lock);
1258 spin_lock_init(&root->delalloc_lock);
1259 spin_lock_init(&root->ordered_extent_lock);
1260 spin_lock_init(&root->accounting_lock);
1261 spin_lock_init(&root->log_extents_lock[0]);
1262 spin_lock_init(&root->log_extents_lock[1]);
1263 mutex_init(&root->objectid_mutex);
1264 mutex_init(&root->log_mutex);
1265 mutex_init(&root->ordered_extent_mutex);
1266 mutex_init(&root->delalloc_mutex);
1267 init_waitqueue_head(&root->log_writer_wait);
1268 init_waitqueue_head(&root->log_commit_wait[0]);
1269 init_waitqueue_head(&root->log_commit_wait[1]);
1270 INIT_LIST_HEAD(&root->log_ctxs[0]);
1271 INIT_LIST_HEAD(&root->log_ctxs[1]);
1272 atomic_set(&root->log_commit[0], 0);
1273 atomic_set(&root->log_commit[1], 0);
1274 atomic_set(&root->log_writers, 0);
1275 atomic_set(&root->log_batch, 0);
1276 atomic_set(&root->orphan_inodes, 0);
1277 atomic_set(&root->refs, 1);
1278 atomic_set(&root->will_be_snapshoted, 0);
1279 atomic_set(&root->qgroup_meta_rsv, 0);
1280 root->log_transid = 0;
1281 root->log_transid_committed = -1;
1282 root->last_log_commit = 0;
1284 extent_io_tree_init(&root->dirty_log_pages,
1285 fs_info->btree_inode->i_mapping);
1287 memset(&root->root_key, 0, sizeof(root->root_key));
1288 memset(&root->root_item, 0, sizeof(root->root_item));
1289 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1291 root->defrag_trans_start = fs_info->generation;
1293 root->defrag_trans_start = 0;
1294 root->root_key.objectid = objectid;
1297 spin_lock_init(&root->root_item_lock);
1300 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1303 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1305 root->fs_info = fs_info;
1309 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1310 /* Should only be used by the testing infrastructure */
1311 struct btrfs_root *btrfs_alloc_dummy_root(void)
1313 struct btrfs_root *root;
1315 root = btrfs_alloc_root(NULL, GFP_KERNEL);
1317 return ERR_PTR(-ENOMEM);
1318 __setup_root(4096, 4096, 4096, root, NULL, 1);
1319 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1320 root->alloc_bytenr = 0;
1326 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1327 struct btrfs_fs_info *fs_info,
1330 struct extent_buffer *leaf;
1331 struct btrfs_root *tree_root = fs_info->tree_root;
1332 struct btrfs_root *root;
1333 struct btrfs_key key;
1337 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1339 return ERR_PTR(-ENOMEM);
1341 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1342 tree_root->stripesize, root, fs_info, objectid);
1343 root->root_key.objectid = objectid;
1344 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1345 root->root_key.offset = 0;
1347 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1349 ret = PTR_ERR(leaf);
1354 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1355 btrfs_set_header_bytenr(leaf, leaf->start);
1356 btrfs_set_header_generation(leaf, trans->transid);
1357 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1358 btrfs_set_header_owner(leaf, objectid);
1361 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1363 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1364 btrfs_header_chunk_tree_uuid(leaf),
1366 btrfs_mark_buffer_dirty(leaf);
1368 root->commit_root = btrfs_root_node(root);
1369 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1371 root->root_item.flags = 0;
1372 root->root_item.byte_limit = 0;
1373 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1374 btrfs_set_root_generation(&root->root_item, trans->transid);
1375 btrfs_set_root_level(&root->root_item, 0);
1376 btrfs_set_root_refs(&root->root_item, 1);
1377 btrfs_set_root_used(&root->root_item, leaf->len);
1378 btrfs_set_root_last_snapshot(&root->root_item, 0);
1379 btrfs_set_root_dirid(&root->root_item, 0);
1381 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1382 root->root_item.drop_level = 0;
1384 key.objectid = objectid;
1385 key.type = BTRFS_ROOT_ITEM_KEY;
1387 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1391 btrfs_tree_unlock(leaf);
1397 btrfs_tree_unlock(leaf);
1398 free_extent_buffer(root->commit_root);
1399 free_extent_buffer(leaf);
1403 return ERR_PTR(ret);
1406 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1407 struct btrfs_fs_info *fs_info)
1409 struct btrfs_root *root;
1410 struct btrfs_root *tree_root = fs_info->tree_root;
1411 struct extent_buffer *leaf;
1413 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1415 return ERR_PTR(-ENOMEM);
1417 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1418 tree_root->stripesize, root, fs_info,
1419 BTRFS_TREE_LOG_OBJECTID);
1421 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1422 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1423 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1426 * DON'T set REF_COWS for log trees
1428 * log trees do not get reference counted because they go away
1429 * before a real commit is actually done. They do store pointers
1430 * to file data extents, and those reference counts still get
1431 * updated (along with back refs to the log tree).
1434 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1438 return ERR_CAST(leaf);
1441 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1442 btrfs_set_header_bytenr(leaf, leaf->start);
1443 btrfs_set_header_generation(leaf, trans->transid);
1444 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1445 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1448 write_extent_buffer(root->node, root->fs_info->fsid,
1449 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1450 btrfs_mark_buffer_dirty(root->node);
1451 btrfs_tree_unlock(root->node);
1455 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1456 struct btrfs_fs_info *fs_info)
1458 struct btrfs_root *log_root;
1460 log_root = alloc_log_tree(trans, fs_info);
1461 if (IS_ERR(log_root))
1462 return PTR_ERR(log_root);
1463 WARN_ON(fs_info->log_root_tree);
1464 fs_info->log_root_tree = log_root;
1468 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1469 struct btrfs_root *root)
1471 struct btrfs_root *log_root;
1472 struct btrfs_inode_item *inode_item;
1474 log_root = alloc_log_tree(trans, root->fs_info);
1475 if (IS_ERR(log_root))
1476 return PTR_ERR(log_root);
1478 log_root->last_trans = trans->transid;
1479 log_root->root_key.offset = root->root_key.objectid;
1481 inode_item = &log_root->root_item.inode;
1482 btrfs_set_stack_inode_generation(inode_item, 1);
1483 btrfs_set_stack_inode_size(inode_item, 3);
1484 btrfs_set_stack_inode_nlink(inode_item, 1);
1485 btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1486 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1488 btrfs_set_root_node(&log_root->root_item, log_root->node);
1490 WARN_ON(root->log_root);
1491 root->log_root = log_root;
1492 root->log_transid = 0;
1493 root->log_transid_committed = -1;
1494 root->last_log_commit = 0;
1498 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1499 struct btrfs_key *key)
1501 struct btrfs_root *root;
1502 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1503 struct btrfs_path *path;
1507 path = btrfs_alloc_path();
1509 return ERR_PTR(-ENOMEM);
1511 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1517 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1518 tree_root->stripesize, root, fs_info, key->objectid);
1520 ret = btrfs_find_root(tree_root, key, path,
1521 &root->root_item, &root->root_key);
1528 generation = btrfs_root_generation(&root->root_item);
1529 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1531 if (IS_ERR(root->node)) {
1532 ret = PTR_ERR(root->node);
1534 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1536 free_extent_buffer(root->node);
1539 root->commit_root = btrfs_root_node(root);
1541 btrfs_free_path(path);
1547 root = ERR_PTR(ret);
1551 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1552 struct btrfs_key *location)
1554 struct btrfs_root *root;
1556 root = btrfs_read_tree_root(tree_root, location);
1560 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1561 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1562 btrfs_check_and_init_root_item(&root->root_item);
1568 int btrfs_init_fs_root(struct btrfs_root *root)
1571 struct btrfs_subvolume_writers *writers;
1573 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1574 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1576 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1581 writers = btrfs_alloc_subvolume_writers();
1582 if (IS_ERR(writers)) {
1583 ret = PTR_ERR(writers);
1586 root->subv_writers = writers;
1588 btrfs_init_free_ino_ctl(root);
1589 spin_lock_init(&root->ino_cache_lock);
1590 init_waitqueue_head(&root->ino_cache_wait);
1592 ret = get_anon_bdev(&root->anon_dev);
1596 mutex_lock(&root->objectid_mutex);
1597 ret = btrfs_find_highest_objectid(root,
1598 &root->highest_objectid);
1600 mutex_unlock(&root->objectid_mutex);
1604 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1606 mutex_unlock(&root->objectid_mutex);
1611 free_anon_bdev(root->anon_dev);
1613 btrfs_free_subvolume_writers(root->subv_writers);
1615 kfree(root->free_ino_ctl);
1616 kfree(root->free_ino_pinned);
1620 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1623 struct btrfs_root *root;
1625 spin_lock(&fs_info->fs_roots_radix_lock);
1626 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1627 (unsigned long)root_id);
1628 spin_unlock(&fs_info->fs_roots_radix_lock);
1632 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1633 struct btrfs_root *root)
1637 ret = radix_tree_preload(GFP_NOFS);
1641 spin_lock(&fs_info->fs_roots_radix_lock);
1642 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1643 (unsigned long)root->root_key.objectid,
1646 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1647 spin_unlock(&fs_info->fs_roots_radix_lock);
1648 radix_tree_preload_end();
1653 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1654 struct btrfs_key *location,
1657 struct btrfs_root *root;
1658 struct btrfs_path *path;
1659 struct btrfs_key key;
1662 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1663 return fs_info->tree_root;
1664 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1665 return fs_info->extent_root;
1666 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1667 return fs_info->chunk_root;
1668 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1669 return fs_info->dev_root;
1670 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1671 return fs_info->csum_root;
1672 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1673 return fs_info->quota_root ? fs_info->quota_root :
1675 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1676 return fs_info->uuid_root ? fs_info->uuid_root :
1678 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1679 return fs_info->free_space_root ? fs_info->free_space_root :
1682 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1684 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1685 return ERR_PTR(-ENOENT);
1689 root = btrfs_read_fs_root(fs_info->tree_root, location);
1693 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1698 ret = btrfs_init_fs_root(root);
1702 path = btrfs_alloc_path();
1707 key.objectid = BTRFS_ORPHAN_OBJECTID;
1708 key.type = BTRFS_ORPHAN_ITEM_KEY;
1709 key.offset = location->objectid;
1711 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1712 btrfs_free_path(path);
1716 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1718 ret = btrfs_insert_fs_root(fs_info, root);
1720 if (ret == -EEXIST) {
1729 return ERR_PTR(ret);
1732 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1734 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1736 struct btrfs_device *device;
1737 struct backing_dev_info *bdi;
1740 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1743 bdi = blk_get_backing_dev_info(device->bdev);
1744 if (bdi_congested(bdi, bdi_bits)) {
1753 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1757 err = bdi_setup_and_register(bdi, "btrfs");
1761 bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1762 bdi->congested_fn = btrfs_congested_fn;
1763 bdi->congested_data = info;
1764 bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1769 * called by the kthread helper functions to finally call the bio end_io
1770 * functions. This is where read checksum verification actually happens
1772 static void end_workqueue_fn(struct btrfs_work *work)
1775 struct btrfs_end_io_wq *end_io_wq;
1777 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1778 bio = end_io_wq->bio;
1780 bio->bi_error = end_io_wq->error;
1781 bio->bi_private = end_io_wq->private;
1782 bio->bi_end_io = end_io_wq->end_io;
1783 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1787 static int cleaner_kthread(void *arg)
1789 struct btrfs_root *root = arg;
1791 struct btrfs_trans_handle *trans;
1796 /* Make the cleaner go to sleep early. */
1797 if (btrfs_need_cleaner_sleep(root))
1800 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1804 * Avoid the problem that we change the status of the fs
1805 * during the above check and trylock.
1807 if (btrfs_need_cleaner_sleep(root)) {
1808 mutex_unlock(&root->fs_info->cleaner_mutex);
1812 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1813 btrfs_run_delayed_iputs(root);
1814 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1816 again = btrfs_clean_one_deleted_snapshot(root);
1817 mutex_unlock(&root->fs_info->cleaner_mutex);
1820 * The defragger has dealt with the R/O remount and umount,
1821 * needn't do anything special here.
1823 btrfs_run_defrag_inodes(root->fs_info);
1826 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1827 * with relocation (btrfs_relocate_chunk) and relocation
1828 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1829 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1830 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1831 * unused block groups.
1833 btrfs_delete_unused_bgs(root->fs_info);
1836 set_current_state(TASK_INTERRUPTIBLE);
1837 if (!kthread_should_stop())
1839 __set_current_state(TASK_RUNNING);
1841 } while (!kthread_should_stop());
1844 * Transaction kthread is stopped before us and wakes us up.
1845 * However we might have started a new transaction and COWed some
1846 * tree blocks when deleting unused block groups for example. So
1847 * make sure we commit the transaction we started to have a clean
1848 * shutdown when evicting the btree inode - if it has dirty pages
1849 * when we do the final iput() on it, eviction will trigger a
1850 * writeback for it which will fail with null pointer dereferences
1851 * since work queues and other resources were already released and
1852 * destroyed by the time the iput/eviction/writeback is made.
1854 trans = btrfs_attach_transaction(root);
1855 if (IS_ERR(trans)) {
1856 if (PTR_ERR(trans) != -ENOENT)
1857 btrfs_err(root->fs_info,
1858 "cleaner transaction attach returned %ld",
1863 ret = btrfs_commit_transaction(trans, root);
1865 btrfs_err(root->fs_info,
1866 "cleaner open transaction commit returned %d",
1873 static int transaction_kthread(void *arg)
1875 struct btrfs_root *root = arg;
1876 struct btrfs_trans_handle *trans;
1877 struct btrfs_transaction *cur;
1880 unsigned long delay;
1884 cannot_commit = false;
1885 delay = HZ * root->fs_info->commit_interval;
1886 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1888 spin_lock(&root->fs_info->trans_lock);
1889 cur = root->fs_info->running_transaction;
1891 spin_unlock(&root->fs_info->trans_lock);
1895 now = get_seconds();
1896 if (cur->state < TRANS_STATE_BLOCKED &&
1897 (now < cur->start_time ||
1898 now - cur->start_time < root->fs_info->commit_interval)) {
1899 spin_unlock(&root->fs_info->trans_lock);
1903 transid = cur->transid;
1904 spin_unlock(&root->fs_info->trans_lock);
1906 /* If the file system is aborted, this will always fail. */
1907 trans = btrfs_attach_transaction(root);
1908 if (IS_ERR(trans)) {
1909 if (PTR_ERR(trans) != -ENOENT)
1910 cannot_commit = true;
1913 if (transid == trans->transid) {
1914 btrfs_commit_transaction(trans, root);
1916 btrfs_end_transaction(trans, root);
1919 wake_up_process(root->fs_info->cleaner_kthread);
1920 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1922 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1923 &root->fs_info->fs_state)))
1924 btrfs_cleanup_transaction(root);
1925 set_current_state(TASK_INTERRUPTIBLE);
1926 if (!kthread_should_stop() &&
1927 (!btrfs_transaction_blocked(root->fs_info) ||
1929 schedule_timeout(delay);
1930 __set_current_state(TASK_RUNNING);
1931 } while (!kthread_should_stop());
1936 * this will find the highest generation in the array of
1937 * root backups. The index of the highest array is returned,
1938 * or -1 if we can't find anything.
1940 * We check to make sure the array is valid by comparing the
1941 * generation of the latest root in the array with the generation
1942 * in the super block. If they don't match we pitch it.
1944 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1947 int newest_index = -1;
1948 struct btrfs_root_backup *root_backup;
1951 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1952 root_backup = info->super_copy->super_roots + i;
1953 cur = btrfs_backup_tree_root_gen(root_backup);
1954 if (cur == newest_gen)
1958 /* check to see if we actually wrapped around */
1959 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1960 root_backup = info->super_copy->super_roots;
1961 cur = btrfs_backup_tree_root_gen(root_backup);
1962 if (cur == newest_gen)
1965 return newest_index;
1970 * find the oldest backup so we know where to store new entries
1971 * in the backup array. This will set the backup_root_index
1972 * field in the fs_info struct
1974 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1977 int newest_index = -1;
1979 newest_index = find_newest_super_backup(info, newest_gen);
1980 /* if there was garbage in there, just move along */
1981 if (newest_index == -1) {
1982 info->backup_root_index = 0;
1984 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1989 * copy all the root pointers into the super backup array.
1990 * this will bump the backup pointer by one when it is
1993 static void backup_super_roots(struct btrfs_fs_info *info)
1996 struct btrfs_root_backup *root_backup;
1999 next_backup = info->backup_root_index;
2000 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2001 BTRFS_NUM_BACKUP_ROOTS;
2004 * just overwrite the last backup if we're at the same generation
2005 * this happens only at umount
2007 root_backup = info->super_for_commit->super_roots + last_backup;
2008 if (btrfs_backup_tree_root_gen(root_backup) ==
2009 btrfs_header_generation(info->tree_root->node))
2010 next_backup = last_backup;
2012 root_backup = info->super_for_commit->super_roots + next_backup;
2015 * make sure all of our padding and empty slots get zero filled
2016 * regardless of which ones we use today
2018 memset(root_backup, 0, sizeof(*root_backup));
2020 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2022 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2023 btrfs_set_backup_tree_root_gen(root_backup,
2024 btrfs_header_generation(info->tree_root->node));
2026 btrfs_set_backup_tree_root_level(root_backup,
2027 btrfs_header_level(info->tree_root->node));
2029 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2030 btrfs_set_backup_chunk_root_gen(root_backup,
2031 btrfs_header_generation(info->chunk_root->node));
2032 btrfs_set_backup_chunk_root_level(root_backup,
2033 btrfs_header_level(info->chunk_root->node));
2035 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2036 btrfs_set_backup_extent_root_gen(root_backup,
2037 btrfs_header_generation(info->extent_root->node));
2038 btrfs_set_backup_extent_root_level(root_backup,
2039 btrfs_header_level(info->extent_root->node));
2042 * we might commit during log recovery, which happens before we set
2043 * the fs_root. Make sure it is valid before we fill it in.
2045 if (info->fs_root && info->fs_root->node) {
2046 btrfs_set_backup_fs_root(root_backup,
2047 info->fs_root->node->start);
2048 btrfs_set_backup_fs_root_gen(root_backup,
2049 btrfs_header_generation(info->fs_root->node));
2050 btrfs_set_backup_fs_root_level(root_backup,
2051 btrfs_header_level(info->fs_root->node));
2054 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2055 btrfs_set_backup_dev_root_gen(root_backup,
2056 btrfs_header_generation(info->dev_root->node));
2057 btrfs_set_backup_dev_root_level(root_backup,
2058 btrfs_header_level(info->dev_root->node));
2060 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2061 btrfs_set_backup_csum_root_gen(root_backup,
2062 btrfs_header_generation(info->csum_root->node));
2063 btrfs_set_backup_csum_root_level(root_backup,
2064 btrfs_header_level(info->csum_root->node));
2066 btrfs_set_backup_total_bytes(root_backup,
2067 btrfs_super_total_bytes(info->super_copy));
2068 btrfs_set_backup_bytes_used(root_backup,
2069 btrfs_super_bytes_used(info->super_copy));
2070 btrfs_set_backup_num_devices(root_backup,
2071 btrfs_super_num_devices(info->super_copy));
2074 * if we don't copy this out to the super_copy, it won't get remembered
2075 * for the next commit
2077 memcpy(&info->super_copy->super_roots,
2078 &info->super_for_commit->super_roots,
2079 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2083 * this copies info out of the root backup array and back into
2084 * the in-memory super block. It is meant to help iterate through
2085 * the array, so you send it the number of backups you've already
2086 * tried and the last backup index you used.
2088 * this returns -1 when it has tried all the backups
2090 static noinline int next_root_backup(struct btrfs_fs_info *info,
2091 struct btrfs_super_block *super,
2092 int *num_backups_tried, int *backup_index)
2094 struct btrfs_root_backup *root_backup;
2095 int newest = *backup_index;
2097 if (*num_backups_tried == 0) {
2098 u64 gen = btrfs_super_generation(super);
2100 newest = find_newest_super_backup(info, gen);
2104 *backup_index = newest;
2105 *num_backups_tried = 1;
2106 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2107 /* we've tried all the backups, all done */
2110 /* jump to the next oldest backup */
2111 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2112 BTRFS_NUM_BACKUP_ROOTS;
2113 *backup_index = newest;
2114 *num_backups_tried += 1;
2116 root_backup = super->super_roots + newest;
2118 btrfs_set_super_generation(super,
2119 btrfs_backup_tree_root_gen(root_backup));
2120 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2121 btrfs_set_super_root_level(super,
2122 btrfs_backup_tree_root_level(root_backup));
2123 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2126 * fixme: the total bytes and num_devices need to match or we should
2129 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2130 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2134 /* helper to cleanup workers */
2135 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2137 btrfs_destroy_workqueue(fs_info->fixup_workers);
2138 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2139 btrfs_destroy_workqueue(fs_info->workers);
2140 btrfs_destroy_workqueue(fs_info->endio_workers);
2141 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2142 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2143 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2144 btrfs_destroy_workqueue(fs_info->rmw_workers);
2145 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2146 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2147 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2148 btrfs_destroy_workqueue(fs_info->submit_workers);
2149 btrfs_destroy_workqueue(fs_info->delayed_workers);
2150 btrfs_destroy_workqueue(fs_info->caching_workers);
2151 btrfs_destroy_workqueue(fs_info->readahead_workers);
2152 btrfs_destroy_workqueue(fs_info->flush_workers);
2153 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2154 btrfs_destroy_workqueue(fs_info->extent_workers);
2157 static void free_root_extent_buffers(struct btrfs_root *root)
2160 free_extent_buffer(root->node);
2161 free_extent_buffer(root->commit_root);
2163 root->commit_root = NULL;
2167 /* helper to cleanup tree roots */
2168 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2170 free_root_extent_buffers(info->tree_root);
2172 free_root_extent_buffers(info->dev_root);
2173 free_root_extent_buffers(info->extent_root);
2174 free_root_extent_buffers(info->csum_root);
2175 free_root_extent_buffers(info->quota_root);
2176 free_root_extent_buffers(info->uuid_root);
2178 free_root_extent_buffers(info->chunk_root);
2179 free_root_extent_buffers(info->free_space_root);
2182 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2185 struct btrfs_root *gang[8];
2188 while (!list_empty(&fs_info->dead_roots)) {
2189 gang[0] = list_entry(fs_info->dead_roots.next,
2190 struct btrfs_root, root_list);
2191 list_del(&gang[0]->root_list);
2193 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2194 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2196 free_extent_buffer(gang[0]->node);
2197 free_extent_buffer(gang[0]->commit_root);
2198 btrfs_put_fs_root(gang[0]);
2203 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2208 for (i = 0; i < ret; i++)
2209 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2212 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2213 btrfs_free_log_root_tree(NULL, fs_info);
2214 btrfs_destroy_pinned_extent(fs_info->tree_root,
2215 fs_info->pinned_extents);
2219 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2221 mutex_init(&fs_info->scrub_lock);
2222 atomic_set(&fs_info->scrubs_running, 0);
2223 atomic_set(&fs_info->scrub_pause_req, 0);
2224 atomic_set(&fs_info->scrubs_paused, 0);
2225 atomic_set(&fs_info->scrub_cancel_req, 0);
2226 init_waitqueue_head(&fs_info->scrub_pause_wait);
2227 fs_info->scrub_workers_refcnt = 0;
2230 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2232 spin_lock_init(&fs_info->balance_lock);
2233 mutex_init(&fs_info->balance_mutex);
2234 atomic_set(&fs_info->balance_running, 0);
2235 atomic_set(&fs_info->balance_pause_req, 0);
2236 atomic_set(&fs_info->balance_cancel_req, 0);
2237 fs_info->balance_ctl = NULL;
2238 init_waitqueue_head(&fs_info->balance_wait_q);
2241 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2242 struct btrfs_root *tree_root)
2244 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2245 set_nlink(fs_info->btree_inode, 1);
2247 * we set the i_size on the btree inode to the max possible int.
2248 * the real end of the address space is determined by all of
2249 * the devices in the system
2251 fs_info->btree_inode->i_size = OFFSET_MAX;
2252 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2254 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2255 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2256 fs_info->btree_inode->i_mapping);
2257 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2258 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2260 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2262 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2263 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2264 sizeof(struct btrfs_key));
2265 set_bit(BTRFS_INODE_DUMMY,
2266 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2267 btrfs_insert_inode_hash(fs_info->btree_inode);
2270 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2272 fs_info->dev_replace.lock_owner = 0;
2273 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2274 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2275 rwlock_init(&fs_info->dev_replace.lock);
2276 atomic_set(&fs_info->dev_replace.read_locks, 0);
2277 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2278 init_waitqueue_head(&fs_info->replace_wait);
2279 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2282 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2284 spin_lock_init(&fs_info->qgroup_lock);
2285 mutex_init(&fs_info->qgroup_ioctl_lock);
2286 fs_info->qgroup_tree = RB_ROOT;
2287 fs_info->qgroup_op_tree = RB_ROOT;
2288 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2289 fs_info->qgroup_seq = 1;
2290 fs_info->quota_enabled = 0;
2291 fs_info->pending_quota_state = 0;
2292 fs_info->qgroup_ulist = NULL;
2293 mutex_init(&fs_info->qgroup_rescan_lock);
2296 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2297 struct btrfs_fs_devices *fs_devices)
2299 int max_active = fs_info->thread_pool_size;
2300 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2303 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2306 fs_info->delalloc_workers =
2307 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2309 fs_info->flush_workers =
2310 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2312 fs_info->caching_workers =
2313 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2316 * a higher idle thresh on the submit workers makes it much more
2317 * likely that bios will be send down in a sane order to the
2320 fs_info->submit_workers =
2321 btrfs_alloc_workqueue("submit", flags,
2322 min_t(u64, fs_devices->num_devices,
2325 fs_info->fixup_workers =
2326 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2329 * endios are largely parallel and should have a very
2332 fs_info->endio_workers =
2333 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2334 fs_info->endio_meta_workers =
2335 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2336 fs_info->endio_meta_write_workers =
2337 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2338 fs_info->endio_raid56_workers =
2339 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2340 fs_info->endio_repair_workers =
2341 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2342 fs_info->rmw_workers =
2343 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2344 fs_info->endio_write_workers =
2345 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2346 fs_info->endio_freespace_worker =
2347 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2348 fs_info->delayed_workers =
2349 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2350 fs_info->readahead_workers =
2351 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2352 fs_info->qgroup_rescan_workers =
2353 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2354 fs_info->extent_workers =
2355 btrfs_alloc_workqueue("extent-refs", flags,
2356 min_t(u64, fs_devices->num_devices,
2359 if (!(fs_info->workers && fs_info->delalloc_workers &&
2360 fs_info->submit_workers && fs_info->flush_workers &&
2361 fs_info->endio_workers && fs_info->endio_meta_workers &&
2362 fs_info->endio_meta_write_workers &&
2363 fs_info->endio_repair_workers &&
2364 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2365 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2366 fs_info->caching_workers && fs_info->readahead_workers &&
2367 fs_info->fixup_workers && fs_info->delayed_workers &&
2368 fs_info->extent_workers &&
2369 fs_info->qgroup_rescan_workers)) {
2376 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2377 struct btrfs_fs_devices *fs_devices)
2380 struct btrfs_root *tree_root = fs_info->tree_root;
2381 struct btrfs_root *log_tree_root;
2382 struct btrfs_super_block *disk_super = fs_info->super_copy;
2383 u64 bytenr = btrfs_super_log_root(disk_super);
2385 if (fs_devices->rw_devices == 0) {
2386 btrfs_warn(fs_info, "log replay required on RO media");
2390 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2394 __setup_root(tree_root->nodesize, tree_root->sectorsize,
2395 tree_root->stripesize, log_tree_root, fs_info,
2396 BTRFS_TREE_LOG_OBJECTID);
2398 log_tree_root->node = read_tree_block(tree_root, bytenr,
2399 fs_info->generation + 1);
2400 if (IS_ERR(log_tree_root->node)) {
2401 btrfs_warn(fs_info, "failed to read log tree");
2402 ret = PTR_ERR(log_tree_root->node);
2403 kfree(log_tree_root);
2405 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2406 btrfs_err(fs_info, "failed to read log tree");
2407 free_extent_buffer(log_tree_root->node);
2408 kfree(log_tree_root);
2411 /* returns with log_tree_root freed on success */
2412 ret = btrfs_recover_log_trees(log_tree_root);
2414 btrfs_handle_fs_error(tree_root->fs_info, ret,
2415 "Failed to recover log tree");
2416 free_extent_buffer(log_tree_root->node);
2417 kfree(log_tree_root);
2421 if (fs_info->sb->s_flags & MS_RDONLY) {
2422 ret = btrfs_commit_super(tree_root);
2430 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2431 struct btrfs_root *tree_root)
2433 struct btrfs_root *root;
2434 struct btrfs_key location;
2437 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2438 location.type = BTRFS_ROOT_ITEM_KEY;
2439 location.offset = 0;
2441 root = btrfs_read_tree_root(tree_root, &location);
2443 return PTR_ERR(root);
2444 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2445 fs_info->extent_root = root;
2447 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2448 root = btrfs_read_tree_root(tree_root, &location);
2450 return PTR_ERR(root);
2451 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2452 fs_info->dev_root = root;
2453 btrfs_init_devices_late(fs_info);
2455 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2456 root = btrfs_read_tree_root(tree_root, &location);
2458 return PTR_ERR(root);
2459 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2460 fs_info->csum_root = root;
2462 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2463 root = btrfs_read_tree_root(tree_root, &location);
2464 if (!IS_ERR(root)) {
2465 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2466 fs_info->quota_enabled = 1;
2467 fs_info->pending_quota_state = 1;
2468 fs_info->quota_root = root;
2471 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2472 root = btrfs_read_tree_root(tree_root, &location);
2474 ret = PTR_ERR(root);
2478 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2479 fs_info->uuid_root = root;
2482 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2483 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2484 root = btrfs_read_tree_root(tree_root, &location);
2486 return PTR_ERR(root);
2487 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2488 fs_info->free_space_root = root;
2494 int open_ctree(struct super_block *sb,
2495 struct btrfs_fs_devices *fs_devices,
2503 struct btrfs_key location;
2504 struct buffer_head *bh;
2505 struct btrfs_super_block *disk_super;
2506 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2507 struct btrfs_root *tree_root;
2508 struct btrfs_root *chunk_root;
2511 int num_backups_tried = 0;
2512 int backup_index = 0;
2514 bool cleaner_mutex_locked = false;
2516 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2517 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2518 if (!tree_root || !chunk_root) {
2523 ret = init_srcu_struct(&fs_info->subvol_srcu);
2529 ret = setup_bdi(fs_info, &fs_info->bdi);
2535 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2540 fs_info->dirty_metadata_batch = PAGE_SIZE *
2541 (1 + ilog2(nr_cpu_ids));
2543 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2546 goto fail_dirty_metadata_bytes;
2549 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2552 goto fail_delalloc_bytes;
2555 fs_info->btree_inode = new_inode(sb);
2556 if (!fs_info->btree_inode) {
2558 goto fail_bio_counter;
2561 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2563 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2564 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2565 INIT_LIST_HEAD(&fs_info->trans_list);
2566 INIT_LIST_HEAD(&fs_info->dead_roots);
2567 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2568 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2569 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2570 spin_lock_init(&fs_info->delalloc_root_lock);
2571 spin_lock_init(&fs_info->trans_lock);
2572 spin_lock_init(&fs_info->fs_roots_radix_lock);
2573 spin_lock_init(&fs_info->delayed_iput_lock);
2574 spin_lock_init(&fs_info->defrag_inodes_lock);
2575 spin_lock_init(&fs_info->free_chunk_lock);
2576 spin_lock_init(&fs_info->tree_mod_seq_lock);
2577 spin_lock_init(&fs_info->super_lock);
2578 spin_lock_init(&fs_info->qgroup_op_lock);
2579 spin_lock_init(&fs_info->buffer_lock);
2580 spin_lock_init(&fs_info->unused_bgs_lock);
2581 rwlock_init(&fs_info->tree_mod_log_lock);
2582 mutex_init(&fs_info->unused_bg_unpin_mutex);
2583 mutex_init(&fs_info->delete_unused_bgs_mutex);
2584 mutex_init(&fs_info->reloc_mutex);
2585 mutex_init(&fs_info->delalloc_root_mutex);
2586 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2587 seqlock_init(&fs_info->profiles_lock);
2589 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2590 INIT_LIST_HEAD(&fs_info->space_info);
2591 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2592 INIT_LIST_HEAD(&fs_info->unused_bgs);
2593 btrfs_mapping_init(&fs_info->mapping_tree);
2594 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2595 BTRFS_BLOCK_RSV_GLOBAL);
2596 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2597 BTRFS_BLOCK_RSV_DELALLOC);
2598 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2599 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2600 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2601 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2602 BTRFS_BLOCK_RSV_DELOPS);
2603 atomic_set(&fs_info->nr_async_submits, 0);
2604 atomic_set(&fs_info->async_delalloc_pages, 0);
2605 atomic_set(&fs_info->async_submit_draining, 0);
2606 atomic_set(&fs_info->nr_async_bios, 0);
2607 atomic_set(&fs_info->defrag_running, 0);
2608 atomic_set(&fs_info->qgroup_op_seq, 0);
2609 atomic_set(&fs_info->reada_works_cnt, 0);
2610 atomic64_set(&fs_info->tree_mod_seq, 0);
2612 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2613 fs_info->metadata_ratio = 0;
2614 fs_info->defrag_inodes = RB_ROOT;
2615 fs_info->free_chunk_space = 0;
2616 fs_info->tree_mod_log = RB_ROOT;
2617 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2618 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2619 /* readahead state */
2620 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2621 spin_lock_init(&fs_info->reada_lock);
2623 fs_info->thread_pool_size = min_t(unsigned long,
2624 num_online_cpus() + 2, 8);
2626 INIT_LIST_HEAD(&fs_info->ordered_roots);
2627 spin_lock_init(&fs_info->ordered_root_lock);
2628 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2630 if (!fs_info->delayed_root) {
2634 btrfs_init_delayed_root(fs_info->delayed_root);
2636 btrfs_init_scrub(fs_info);
2637 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2638 fs_info->check_integrity_print_mask = 0;
2640 btrfs_init_balance(fs_info);
2641 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2643 sb->s_blocksize = 4096;
2644 sb->s_blocksize_bits = blksize_bits(4096);
2645 sb->s_bdi = &fs_info->bdi;
2647 btrfs_init_btree_inode(fs_info, tree_root);
2649 spin_lock_init(&fs_info->block_group_cache_lock);
2650 fs_info->block_group_cache_tree = RB_ROOT;
2651 fs_info->first_logical_byte = (u64)-1;
2653 extent_io_tree_init(&fs_info->freed_extents[0],
2654 fs_info->btree_inode->i_mapping);
2655 extent_io_tree_init(&fs_info->freed_extents[1],
2656 fs_info->btree_inode->i_mapping);
2657 fs_info->pinned_extents = &fs_info->freed_extents[0];
2658 fs_info->do_barriers = 1;
2661 mutex_init(&fs_info->ordered_operations_mutex);
2662 mutex_init(&fs_info->tree_log_mutex);
2663 mutex_init(&fs_info->chunk_mutex);
2664 mutex_init(&fs_info->transaction_kthread_mutex);
2665 mutex_init(&fs_info->cleaner_mutex);
2666 mutex_init(&fs_info->volume_mutex);
2667 mutex_init(&fs_info->ro_block_group_mutex);
2668 init_rwsem(&fs_info->commit_root_sem);
2669 init_rwsem(&fs_info->cleanup_work_sem);
2670 init_rwsem(&fs_info->subvol_sem);
2671 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2673 btrfs_init_dev_replace_locks(fs_info);
2674 btrfs_init_qgroup(fs_info);
2676 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2677 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2679 init_waitqueue_head(&fs_info->transaction_throttle);
2680 init_waitqueue_head(&fs_info->transaction_wait);
2681 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2682 init_waitqueue_head(&fs_info->async_submit_wait);
2684 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2686 ret = btrfs_alloc_stripe_hash_table(fs_info);
2692 __setup_root(4096, 4096, 4096, tree_root,
2693 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2695 invalidate_bdev(fs_devices->latest_bdev);
2698 * Read super block and check the signature bytes only
2700 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2707 * We want to check superblock checksum, the type is stored inside.
2708 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2710 if (btrfs_check_super_csum(bh->b_data)) {
2711 btrfs_err(fs_info, "superblock checksum mismatch");
2718 * super_copy is zeroed at allocation time and we never touch the
2719 * following bytes up to INFO_SIZE, the checksum is calculated from
2720 * the whole block of INFO_SIZE
2722 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2723 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2724 sizeof(*fs_info->super_for_commit));
2727 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2729 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2731 btrfs_err(fs_info, "superblock contains fatal errors");
2736 disk_super = fs_info->super_copy;
2737 if (!btrfs_super_root(disk_super))
2740 /* check FS state, whether FS is broken. */
2741 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2742 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2745 * run through our array of backup supers and setup
2746 * our ring pointer to the oldest one
2748 generation = btrfs_super_generation(disk_super);
2749 find_oldest_super_backup(fs_info, generation);
2752 * In the long term, we'll store the compression type in the super
2753 * block, and it'll be used for per file compression control.
2755 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2757 ret = btrfs_parse_options(tree_root, options, sb->s_flags);
2763 features = btrfs_super_incompat_flags(disk_super) &
2764 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2767 "cannot mount because of unsupported optional features (%llx)",
2773 features = btrfs_super_incompat_flags(disk_super);
2774 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2775 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2776 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2778 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2779 btrfs_info(fs_info, "has skinny extents");
2782 * flag our filesystem as having big metadata blocks if
2783 * they are bigger than the page size
2785 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2786 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2788 "flagging fs with big metadata feature");
2789 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2792 nodesize = btrfs_super_nodesize(disk_super);
2793 sectorsize = btrfs_super_sectorsize(disk_super);
2794 stripesize = btrfs_super_stripesize(disk_super);
2795 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2796 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2799 * mixed block groups end up with duplicate but slightly offset
2800 * extent buffers for the same range. It leads to corruptions
2802 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2803 (sectorsize != nodesize)) {
2805 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2806 nodesize, sectorsize);
2811 * Needn't use the lock because there is no other task which will
2814 btrfs_set_super_incompat_flags(disk_super, features);
2816 features = btrfs_super_compat_ro_flags(disk_super) &
2817 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2818 if (!(sb->s_flags & MS_RDONLY) && features) {
2820 "cannot mount read-write because of unsupported optional features (%llx)",
2826 max_active = fs_info->thread_pool_size;
2828 ret = btrfs_init_workqueues(fs_info, fs_devices);
2831 goto fail_sb_buffer;
2834 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2835 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2838 tree_root->nodesize = nodesize;
2839 tree_root->sectorsize = sectorsize;
2840 tree_root->stripesize = stripesize;
2842 sb->s_blocksize = sectorsize;
2843 sb->s_blocksize_bits = blksize_bits(sectorsize);
2845 mutex_lock(&fs_info->chunk_mutex);
2846 ret = btrfs_read_sys_array(tree_root);
2847 mutex_unlock(&fs_info->chunk_mutex);
2849 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2850 goto fail_sb_buffer;
2853 generation = btrfs_super_chunk_root_generation(disk_super);
2855 __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2856 fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2858 chunk_root->node = read_tree_block(chunk_root,
2859 btrfs_super_chunk_root(disk_super),
2861 if (IS_ERR(chunk_root->node) ||
2862 !extent_buffer_uptodate(chunk_root->node)) {
2863 btrfs_err(fs_info, "failed to read chunk root");
2864 if (!IS_ERR(chunk_root->node))
2865 free_extent_buffer(chunk_root->node);
2866 chunk_root->node = NULL;
2867 goto fail_tree_roots;
2869 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2870 chunk_root->commit_root = btrfs_root_node(chunk_root);
2872 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2873 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2875 ret = btrfs_read_chunk_tree(chunk_root);
2877 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2878 goto fail_tree_roots;
2882 * keep the device that is marked to be the target device for the
2883 * dev_replace procedure
2885 btrfs_close_extra_devices(fs_devices, 0);
2887 if (!fs_devices->latest_bdev) {
2888 btrfs_err(fs_info, "failed to read devices");
2889 goto fail_tree_roots;
2893 generation = btrfs_super_generation(disk_super);
2895 tree_root->node = read_tree_block(tree_root,
2896 btrfs_super_root(disk_super),
2898 if (IS_ERR(tree_root->node) ||
2899 !extent_buffer_uptodate(tree_root->node)) {
2900 btrfs_warn(fs_info, "failed to read tree root");
2901 if (!IS_ERR(tree_root->node))
2902 free_extent_buffer(tree_root->node);
2903 tree_root->node = NULL;
2904 goto recovery_tree_root;
2907 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2908 tree_root->commit_root = btrfs_root_node(tree_root);
2909 btrfs_set_root_refs(&tree_root->root_item, 1);
2911 mutex_lock(&tree_root->objectid_mutex);
2912 ret = btrfs_find_highest_objectid(tree_root,
2913 &tree_root->highest_objectid);
2915 mutex_unlock(&tree_root->objectid_mutex);
2916 goto recovery_tree_root;
2919 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2921 mutex_unlock(&tree_root->objectid_mutex);
2923 ret = btrfs_read_roots(fs_info, tree_root);
2925 goto recovery_tree_root;
2927 fs_info->generation = generation;
2928 fs_info->last_trans_committed = generation;
2930 ret = btrfs_recover_balance(fs_info);
2932 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2933 goto fail_block_groups;
2936 ret = btrfs_init_dev_stats(fs_info);
2938 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2939 goto fail_block_groups;
2942 ret = btrfs_init_dev_replace(fs_info);
2944 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2945 goto fail_block_groups;
2948 btrfs_close_extra_devices(fs_devices, 1);
2950 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2952 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2954 goto fail_block_groups;
2957 ret = btrfs_sysfs_add_device(fs_devices);
2959 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2961 goto fail_fsdev_sysfs;
2964 ret = btrfs_sysfs_add_mounted(fs_info);
2966 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2967 goto fail_fsdev_sysfs;
2970 ret = btrfs_init_space_info(fs_info);
2972 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2976 ret = btrfs_read_block_groups(fs_info->extent_root);
2978 btrfs_err(fs_info, "failed to read block groups: %d", ret);
2981 fs_info->num_tolerated_disk_barrier_failures =
2982 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2983 if (fs_info->fs_devices->missing_devices >
2984 fs_info->num_tolerated_disk_barrier_failures &&
2985 !(sb->s_flags & MS_RDONLY)) {
2987 "missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
2988 fs_info->fs_devices->missing_devices,
2989 fs_info->num_tolerated_disk_barrier_failures);
2994 * Hold the cleaner_mutex thread here so that we don't block
2995 * for a long time on btrfs_recover_relocation. cleaner_kthread
2996 * will wait for us to finish mounting the filesystem.
2998 mutex_lock(&fs_info->cleaner_mutex);
2999 cleaner_mutex_locked = true;
3000 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3002 if (IS_ERR(fs_info->cleaner_kthread))
3005 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3007 "btrfs-transaction");
3008 if (IS_ERR(fs_info->transaction_kthread))
3011 if (!btrfs_test_opt(tree_root, SSD) &&
3012 !btrfs_test_opt(tree_root, NOSSD) &&
3013 !fs_info->fs_devices->rotating) {
3014 btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3015 btrfs_set_opt(fs_info->mount_opt, SSD);
3019 * Mount does not set all options immediately, we can do it now and do
3020 * not have to wait for transaction commit
3022 btrfs_apply_pending_changes(fs_info);
3024 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3025 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
3026 ret = btrfsic_mount(tree_root, fs_devices,
3027 btrfs_test_opt(tree_root,
3028 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3030 fs_info->check_integrity_print_mask);
3033 "failed to initialize integrity check module: %d",
3037 ret = btrfs_read_qgroup_config(fs_info);
3039 goto fail_trans_kthread;
3041 /* do not make disk changes in broken FS or nologreplay is given */
3042 if (btrfs_super_log_root(disk_super) != 0 &&
3043 !btrfs_test_opt(tree_root, NOLOGREPLAY)) {
3044 ret = btrfs_replay_log(fs_info, fs_devices);
3051 ret = btrfs_find_orphan_roots(tree_root);
3055 if (!(sb->s_flags & MS_RDONLY)) {
3056 ret = btrfs_cleanup_fs_roots(fs_info);
3059 /* We locked cleaner_mutex before creating cleaner_kthread. */
3060 ret = btrfs_recover_relocation(tree_root);
3062 btrfs_warn(fs_info, "failed to recover relocation: %d",
3068 mutex_unlock(&fs_info->cleaner_mutex);
3069 cleaner_mutex_locked = false;
3071 location.objectid = BTRFS_FS_TREE_OBJECTID;
3072 location.type = BTRFS_ROOT_ITEM_KEY;
3073 location.offset = 0;
3075 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3076 if (IS_ERR(fs_info->fs_root)) {
3077 err = PTR_ERR(fs_info->fs_root);
3081 if (sb->s_flags & MS_RDONLY)
3084 if (btrfs_test_opt(tree_root, FREE_SPACE_TREE) &&
3085 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3086 btrfs_info(fs_info, "creating free space tree");
3087 ret = btrfs_create_free_space_tree(fs_info);
3090 "failed to create free space tree: %d", ret);
3091 close_ctree(tree_root);
3096 down_read(&fs_info->cleanup_work_sem);
3097 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3098 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3099 up_read(&fs_info->cleanup_work_sem);
3100 close_ctree(tree_root);
3103 up_read(&fs_info->cleanup_work_sem);
3105 ret = btrfs_resume_balance_async(fs_info);
3107 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3108 close_ctree(tree_root);
3112 ret = btrfs_resume_dev_replace_async(fs_info);
3114 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3115 close_ctree(tree_root);
3119 btrfs_qgroup_rescan_resume(fs_info);
3121 if (btrfs_test_opt(tree_root, CLEAR_CACHE) &&
3122 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3123 btrfs_info(fs_info, "clearing free space tree");
3124 ret = btrfs_clear_free_space_tree(fs_info);
3127 "failed to clear free space tree: %d", ret);
3128 close_ctree(tree_root);
3133 if (!fs_info->uuid_root) {
3134 btrfs_info(fs_info, "creating UUID tree");
3135 ret = btrfs_create_uuid_tree(fs_info);
3138 "failed to create the UUID tree: %d", ret);
3139 close_ctree(tree_root);
3142 } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3143 fs_info->generation !=
3144 btrfs_super_uuid_tree_generation(disk_super)) {
3145 btrfs_info(fs_info, "checking UUID tree");
3146 ret = btrfs_check_uuid_tree(fs_info);
3149 "failed to check the UUID tree: %d", ret);
3150 close_ctree(tree_root);
3154 fs_info->update_uuid_tree_gen = 1;
3160 * backuproot only affect mount behavior, and if open_ctree succeeded,
3161 * no need to keep the flag
3163 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3168 btrfs_free_qgroup_config(fs_info);
3170 kthread_stop(fs_info->transaction_kthread);
3171 btrfs_cleanup_transaction(fs_info->tree_root);
3172 btrfs_free_fs_roots(fs_info);
3174 kthread_stop(fs_info->cleaner_kthread);
3177 * make sure we're done with the btree inode before we stop our
3180 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3183 if (cleaner_mutex_locked) {
3184 mutex_unlock(&fs_info->cleaner_mutex);
3185 cleaner_mutex_locked = false;
3187 btrfs_sysfs_remove_mounted(fs_info);
3190 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3193 btrfs_put_block_group_cache(fs_info);
3194 btrfs_free_block_groups(fs_info);
3197 free_root_pointers(fs_info, 1);
3198 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3201 btrfs_stop_all_workers(fs_info);
3204 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3206 iput(fs_info->btree_inode);
3208 percpu_counter_destroy(&fs_info->bio_counter);
3209 fail_delalloc_bytes:
3210 percpu_counter_destroy(&fs_info->delalloc_bytes);
3211 fail_dirty_metadata_bytes:
3212 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3214 bdi_destroy(&fs_info->bdi);
3216 cleanup_srcu_struct(&fs_info->subvol_srcu);
3218 btrfs_free_stripe_hash_table(fs_info);
3219 btrfs_close_devices(fs_info->fs_devices);
3223 if (!btrfs_test_opt(tree_root, USEBACKUPROOT))
3224 goto fail_tree_roots;
3226 free_root_pointers(fs_info, 0);
3228 /* don't use the log in recovery mode, it won't be valid */
3229 btrfs_set_super_log_root(disk_super, 0);
3231 /* we can't trust the free space cache either */
3232 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3234 ret = next_root_backup(fs_info, fs_info->super_copy,
3235 &num_backups_tried, &backup_index);
3237 goto fail_block_groups;
3238 goto retry_root_backup;
3241 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3244 set_buffer_uptodate(bh);
3246 struct btrfs_device *device = (struct btrfs_device *)
3249 btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3250 "lost page write due to IO error on %s",
3251 rcu_str_deref(device->name));
3252 /* note, we don't set_buffer_write_io_error because we have
3253 * our own ways of dealing with the IO errors
3255 clear_buffer_uptodate(bh);
3256 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3262 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3263 struct buffer_head **bh_ret)
3265 struct buffer_head *bh;
3266 struct btrfs_super_block *super;
3269 bytenr = btrfs_sb_offset(copy_num);
3270 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3273 bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3275 * If we fail to read from the underlying devices, as of now
3276 * the best option we have is to mark it EIO.
3281 super = (struct btrfs_super_block *)bh->b_data;
3282 if (btrfs_super_bytenr(super) != bytenr ||
3283 btrfs_super_magic(super) != BTRFS_MAGIC) {
3293 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3295 struct buffer_head *bh;
3296 struct buffer_head *latest = NULL;
3297 struct btrfs_super_block *super;
3302 /* we would like to check all the supers, but that would make
3303 * a btrfs mount succeed after a mkfs from a different FS.
3304 * So, we need to add a special mount option to scan for
3305 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3307 for (i = 0; i < 1; i++) {
3308 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3312 super = (struct btrfs_super_block *)bh->b_data;
3314 if (!latest || btrfs_super_generation(super) > transid) {
3317 transid = btrfs_super_generation(super);
3324 return ERR_PTR(ret);
3330 * this should be called twice, once with wait == 0 and
3331 * once with wait == 1. When wait == 0 is done, all the buffer heads
3332 * we write are pinned.
3334 * They are released when wait == 1 is done.
3335 * max_mirrors must be the same for both runs, and it indicates how
3336 * many supers on this one device should be written.
3338 * max_mirrors == 0 means to write them all.
3340 static int write_dev_supers(struct btrfs_device *device,
3341 struct btrfs_super_block *sb,
3342 int do_barriers, int wait, int max_mirrors)
3344 struct buffer_head *bh;
3351 if (max_mirrors == 0)
3352 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3354 for (i = 0; i < max_mirrors; i++) {
3355 bytenr = btrfs_sb_offset(i);
3356 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3357 device->commit_total_bytes)
3361 bh = __find_get_block(device->bdev, bytenr / 4096,
3362 BTRFS_SUPER_INFO_SIZE);
3368 if (!buffer_uptodate(bh))
3371 /* drop our reference */
3374 /* drop the reference from the wait == 0 run */
3378 btrfs_set_super_bytenr(sb, bytenr);
3381 crc = btrfs_csum_data((char *)sb +
3382 BTRFS_CSUM_SIZE, crc,
3383 BTRFS_SUPER_INFO_SIZE -
3385 btrfs_csum_final(crc, sb->csum);
3388 * one reference for us, and we leave it for the
3391 bh = __getblk(device->bdev, bytenr / 4096,
3392 BTRFS_SUPER_INFO_SIZE);
3394 btrfs_err(device->dev_root->fs_info,
3395 "couldn't get super buffer head for bytenr %llu",
3401 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3403 /* one reference for submit_bh */
3406 set_buffer_uptodate(bh);
3408 bh->b_end_io = btrfs_end_buffer_write_sync;
3409 bh->b_private = device;
3413 * we fua the first super. The others we allow
3417 ret = btrfsic_submit_bh(REQ_OP_WRITE, WRITE_FUA, bh);
3419 ret = btrfsic_submit_bh(REQ_OP_WRITE, WRITE_SYNC, bh);
3423 return errors < i ? 0 : -1;
3427 * endio for the write_dev_flush, this will wake anyone waiting
3428 * for the barrier when it is done
3430 static void btrfs_end_empty_barrier(struct bio *bio)
3432 if (bio->bi_private)
3433 complete(bio->bi_private);
3438 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3439 * sent down. With wait == 1, it waits for the previous flush.
3441 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3444 static int write_dev_flush(struct btrfs_device *device, int wait)
3449 if (device->nobarriers)
3453 bio = device->flush_bio;
3457 wait_for_completion(&device->flush_wait);
3459 if (bio->bi_error) {
3460 ret = bio->bi_error;
3461 btrfs_dev_stat_inc_and_print(device,
3462 BTRFS_DEV_STAT_FLUSH_ERRS);
3465 /* drop the reference from the wait == 0 run */
3467 device->flush_bio = NULL;
3473 * one reference for us, and we leave it for the
3476 device->flush_bio = NULL;
3477 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3481 bio->bi_end_io = btrfs_end_empty_barrier;
3482 bio->bi_bdev = device->bdev;
3483 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
3484 init_completion(&device->flush_wait);
3485 bio->bi_private = &device->flush_wait;
3486 device->flush_bio = bio;
3489 btrfsic_submit_bio(bio);
3495 * send an empty flush down to each device in parallel,
3496 * then wait for them
3498 static int barrier_all_devices(struct btrfs_fs_info *info)
3500 struct list_head *head;
3501 struct btrfs_device *dev;
3502 int errors_send = 0;
3503 int errors_wait = 0;
3506 /* send down all the barriers */
3507 head = &info->fs_devices->devices;
3508 list_for_each_entry_rcu(dev, head, dev_list) {
3515 if (!dev->in_fs_metadata || !dev->writeable)
3518 ret = write_dev_flush(dev, 0);
3523 /* wait for all the barriers */
3524 list_for_each_entry_rcu(dev, head, dev_list) {
3531 if (!dev->in_fs_metadata || !dev->writeable)
3534 ret = write_dev_flush(dev, 1);
3538 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3539 errors_wait > info->num_tolerated_disk_barrier_failures)
3544 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3547 int min_tolerated = INT_MAX;
3549 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3550 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3551 min_tolerated = min(min_tolerated,
3552 btrfs_raid_array[BTRFS_RAID_SINGLE].
3553 tolerated_failures);
3555 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3556 if (raid_type == BTRFS_RAID_SINGLE)
3558 if (!(flags & btrfs_raid_group[raid_type]))
3560 min_tolerated = min(min_tolerated,
3561 btrfs_raid_array[raid_type].
3562 tolerated_failures);
3565 if (min_tolerated == INT_MAX) {
3566 pr_warn("BTRFS: unknown raid flag: %llu\n", flags);
3570 return min_tolerated;
3573 int btrfs_calc_num_tolerated_disk_barrier_failures(
3574 struct btrfs_fs_info *fs_info)
3576 struct btrfs_ioctl_space_info space;
3577 struct btrfs_space_info *sinfo;
3578 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3579 BTRFS_BLOCK_GROUP_SYSTEM,
3580 BTRFS_BLOCK_GROUP_METADATA,
3581 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3584 int num_tolerated_disk_barrier_failures =
3585 (int)fs_info->fs_devices->num_devices;
3587 for (i = 0; i < ARRAY_SIZE(types); i++) {
3588 struct btrfs_space_info *tmp;
3592 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3593 if (tmp->flags == types[i]) {
3603 down_read(&sinfo->groups_sem);
3604 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3607 if (list_empty(&sinfo->block_groups[c]))
3610 btrfs_get_block_group_info(&sinfo->block_groups[c],
3612 if (space.total_bytes == 0 || space.used_bytes == 0)
3614 flags = space.flags;
3616 num_tolerated_disk_barrier_failures = min(
3617 num_tolerated_disk_barrier_failures,
3618 btrfs_get_num_tolerated_disk_barrier_failures(
3621 up_read(&sinfo->groups_sem);
3624 return num_tolerated_disk_barrier_failures;
3627 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3629 struct list_head *head;
3630 struct btrfs_device *dev;
3631 struct btrfs_super_block *sb;
3632 struct btrfs_dev_item *dev_item;
3636 int total_errors = 0;
3639 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3640 backup_super_roots(root->fs_info);
3642 sb = root->fs_info->super_for_commit;
3643 dev_item = &sb->dev_item;
3645 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3646 head = &root->fs_info->fs_devices->devices;
3647 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3650 ret = barrier_all_devices(root->fs_info);
3653 &root->fs_info->fs_devices->device_list_mutex);
3654 btrfs_handle_fs_error(root->fs_info, ret,
3655 "errors while submitting device barriers.");
3660 list_for_each_entry_rcu(dev, head, dev_list) {
3665 if (!dev->in_fs_metadata || !dev->writeable)
3668 btrfs_set_stack_device_generation(dev_item, 0);
3669 btrfs_set_stack_device_type(dev_item, dev->type);
3670 btrfs_set_stack_device_id(dev_item, dev->devid);
3671 btrfs_set_stack_device_total_bytes(dev_item,
3672 dev->commit_total_bytes);
3673 btrfs_set_stack_device_bytes_used(dev_item,
3674 dev->commit_bytes_used);
3675 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3676 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3677 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3678 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3679 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3681 flags = btrfs_super_flags(sb);
3682 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3684 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3688 if (total_errors > max_errors) {
3689 btrfs_err(root->fs_info, "%d errors while writing supers",
3691 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3693 /* FUA is masked off if unsupported and can't be the reason */
3694 btrfs_handle_fs_error(root->fs_info, -EIO,
3695 "%d errors while writing supers", total_errors);
3700 list_for_each_entry_rcu(dev, head, dev_list) {
3703 if (!dev->in_fs_metadata || !dev->writeable)
3706 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3710 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3711 if (total_errors > max_errors) {
3712 btrfs_handle_fs_error(root->fs_info, -EIO,
3713 "%d errors while writing supers", total_errors);
3719 int write_ctree_super(struct btrfs_trans_handle *trans,
3720 struct btrfs_root *root, int max_mirrors)
3722 return write_all_supers(root, max_mirrors);
3725 /* Drop a fs root from the radix tree and free it. */
3726 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3727 struct btrfs_root *root)
3729 spin_lock(&fs_info->fs_roots_radix_lock);
3730 radix_tree_delete(&fs_info->fs_roots_radix,
3731 (unsigned long)root->root_key.objectid);
3732 spin_unlock(&fs_info->fs_roots_radix_lock);
3734 if (btrfs_root_refs(&root->root_item) == 0)
3735 synchronize_srcu(&fs_info->subvol_srcu);
3737 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3738 btrfs_free_log(NULL, root);
3740 if (root->free_ino_pinned)
3741 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3742 if (root->free_ino_ctl)
3743 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3747 static void free_fs_root(struct btrfs_root *root)
3749 iput(root->ino_cache_inode);
3750 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3751 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3752 root->orphan_block_rsv = NULL;
3754 free_anon_bdev(root->anon_dev);
3755 if (root->subv_writers)
3756 btrfs_free_subvolume_writers(root->subv_writers);
3757 free_extent_buffer(root->node);
3758 free_extent_buffer(root->commit_root);
3759 kfree(root->free_ino_ctl);
3760 kfree(root->free_ino_pinned);
3762 btrfs_put_fs_root(root);
3765 void btrfs_free_fs_root(struct btrfs_root *root)
3770 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3772 u64 root_objectid = 0;
3773 struct btrfs_root *gang[8];
3776 unsigned int ret = 0;
3780 index = srcu_read_lock(&fs_info->subvol_srcu);
3781 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3782 (void **)gang, root_objectid,
3785 srcu_read_unlock(&fs_info->subvol_srcu, index);
3788 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3790 for (i = 0; i < ret; i++) {
3791 /* Avoid to grab roots in dead_roots */
3792 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3796 /* grab all the search result for later use */
3797 gang[i] = btrfs_grab_fs_root(gang[i]);
3799 srcu_read_unlock(&fs_info->subvol_srcu, index);
3801 for (i = 0; i < ret; i++) {
3804 root_objectid = gang[i]->root_key.objectid;
3805 err = btrfs_orphan_cleanup(gang[i]);
3808 btrfs_put_fs_root(gang[i]);
3813 /* release the uncleaned roots due to error */
3814 for (; i < ret; i++) {
3816 btrfs_put_fs_root(gang[i]);
3821 int btrfs_commit_super(struct btrfs_root *root)
3823 struct btrfs_trans_handle *trans;
3825 mutex_lock(&root->fs_info->cleaner_mutex);
3826 btrfs_run_delayed_iputs(root);
3827 mutex_unlock(&root->fs_info->cleaner_mutex);
3828 wake_up_process(root->fs_info->cleaner_kthread);
3830 /* wait until ongoing cleanup work done */
3831 down_write(&root->fs_info->cleanup_work_sem);
3832 up_write(&root->fs_info->cleanup_work_sem);
3834 trans = btrfs_join_transaction(root);
3836 return PTR_ERR(trans);
3837 return btrfs_commit_transaction(trans, root);
3840 void close_ctree(struct btrfs_root *root)
3842 struct btrfs_fs_info *fs_info = root->fs_info;
3845 fs_info->closing = 1;
3848 /* wait for the qgroup rescan worker to stop */
3849 btrfs_qgroup_wait_for_completion(fs_info);
3851 /* wait for the uuid_scan task to finish */
3852 down(&fs_info->uuid_tree_rescan_sem);
3853 /* avoid complains from lockdep et al., set sem back to initial state */
3854 up(&fs_info->uuid_tree_rescan_sem);
3856 /* pause restriper - we want to resume on mount */
3857 btrfs_pause_balance(fs_info);
3859 btrfs_dev_replace_suspend_for_unmount(fs_info);
3861 btrfs_scrub_cancel(fs_info);
3863 /* wait for any defraggers to finish */
3864 wait_event(fs_info->transaction_wait,
3865 (atomic_read(&fs_info->defrag_running) == 0));
3867 /* clear out the rbtree of defraggable inodes */
3868 btrfs_cleanup_defrag_inodes(fs_info);
3870 cancel_work_sync(&fs_info->async_reclaim_work);
3872 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3874 * If the cleaner thread is stopped and there are
3875 * block groups queued for removal, the deletion will be
3876 * skipped when we quit the cleaner thread.
3878 btrfs_delete_unused_bgs(root->fs_info);
3880 ret = btrfs_commit_super(root);
3882 btrfs_err(fs_info, "commit super ret %d", ret);
3885 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3886 btrfs_error_commit_super(root);
3888 kthread_stop(fs_info->transaction_kthread);
3889 kthread_stop(fs_info->cleaner_kthread);
3891 fs_info->closing = 2;
3894 btrfs_free_qgroup_config(fs_info);
3896 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3897 btrfs_info(fs_info, "at unmount delalloc count %lld",
3898 percpu_counter_sum(&fs_info->delalloc_bytes));
3901 btrfs_sysfs_remove_mounted(fs_info);
3902 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3904 btrfs_free_fs_roots(fs_info);
3906 btrfs_put_block_group_cache(fs_info);
3908 btrfs_free_block_groups(fs_info);
3911 * we must make sure there is not any read request to
3912 * submit after we stopping all workers.
3914 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3915 btrfs_stop_all_workers(fs_info);
3918 free_root_pointers(fs_info, 1);
3920 iput(fs_info->btree_inode);
3922 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3923 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3924 btrfsic_unmount(root, fs_info->fs_devices);
3927 btrfs_close_devices(fs_info->fs_devices);
3928 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3930 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3931 percpu_counter_destroy(&fs_info->delalloc_bytes);
3932 percpu_counter_destroy(&fs_info->bio_counter);
3933 bdi_destroy(&fs_info->bdi);
3934 cleanup_srcu_struct(&fs_info->subvol_srcu);
3936 btrfs_free_stripe_hash_table(fs_info);
3938 __btrfs_free_block_rsv(root->orphan_block_rsv);
3939 root->orphan_block_rsv = NULL;
3942 while (!list_empty(&fs_info->pinned_chunks)) {
3943 struct extent_map *em;
3945 em = list_first_entry(&fs_info->pinned_chunks,
3946 struct extent_map, list);
3947 list_del_init(&em->list);
3948 free_extent_map(em);
3950 unlock_chunks(root);
3953 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3957 struct inode *btree_inode = buf->pages[0]->mapping->host;
3959 ret = extent_buffer_uptodate(buf);
3963 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3964 parent_transid, atomic);
3970 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3972 struct btrfs_root *root;
3973 u64 transid = btrfs_header_generation(buf);
3976 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3978 * This is a fast path so only do this check if we have sanity tests
3979 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3980 * outside of the sanity tests.
3982 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3985 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3986 btrfs_assert_tree_locked(buf);
3987 if (transid != root->fs_info->generation)
3988 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3989 "found %llu running %llu\n",
3990 buf->start, transid, root->fs_info->generation);
3991 was_dirty = set_extent_buffer_dirty(buf);
3993 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3995 root->fs_info->dirty_metadata_batch);
3996 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3997 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3998 btrfs_print_leaf(root, buf);
4004 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
4008 * looks as though older kernels can get into trouble with
4009 * this code, they end up stuck in balance_dirty_pages forever
4013 if (current->flags & PF_MEMALLOC)
4017 btrfs_balance_delayed_items(root);
4019 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
4020 BTRFS_DIRTY_METADATA_THRESH);
4022 balance_dirty_pages_ratelimited(
4023 root->fs_info->btree_inode->i_mapping);
4027 void btrfs_btree_balance_dirty(struct btrfs_root *root)
4029 __btrfs_btree_balance_dirty(root, 1);
4032 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
4034 __btrfs_btree_balance_dirty(root, 0);
4037 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4039 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4040 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
4043 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4046 struct btrfs_super_block *sb = fs_info->super_copy;
4047 u64 nodesize = btrfs_super_nodesize(sb);
4048 u64 sectorsize = btrfs_super_sectorsize(sb);
4051 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4052 printk(KERN_ERR "BTRFS: no valid FS found\n");
4055 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4056 printk(KERN_WARNING "BTRFS: unrecognized super flag: %llu\n",
4057 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4058 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4059 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
4060 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4063 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4064 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
4065 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4068 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4069 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
4070 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4075 * Check sectorsize and nodesize first, other check will need it.
4076 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4078 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4079 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4080 printk(KERN_ERR "BTRFS: invalid sectorsize %llu\n", sectorsize);
4083 /* Only PAGE SIZE is supported yet */
4084 if (sectorsize != PAGE_SIZE) {
4085 printk(KERN_ERR "BTRFS: sectorsize %llu not supported yet, only support %lu\n",
4086 sectorsize, PAGE_SIZE);
4089 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4090 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4091 printk(KERN_ERR "BTRFS: invalid nodesize %llu\n", nodesize);
4094 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4095 printk(KERN_ERR "BTRFS: invalid leafsize %u, should be %llu\n",
4096 le32_to_cpu(sb->__unused_leafsize),
4101 /* Root alignment check */
4102 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4103 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
4104 btrfs_super_root(sb));
4107 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4108 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
4109 btrfs_super_chunk_root(sb));
4112 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4113 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
4114 btrfs_super_log_root(sb));
4118 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4119 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4120 fs_info->fsid, sb->dev_item.fsid);
4125 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4128 if (btrfs_super_num_devices(sb) > (1UL << 31))
4129 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4130 btrfs_super_num_devices(sb));
4131 if (btrfs_super_num_devices(sb) == 0) {
4132 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4136 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4137 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4138 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4143 * Obvious sys_chunk_array corruptions, it must hold at least one key
4146 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4147 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4148 btrfs_super_sys_array_size(sb),
4149 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4152 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4153 + sizeof(struct btrfs_chunk)) {
4154 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4155 btrfs_super_sys_array_size(sb),
4156 sizeof(struct btrfs_disk_key)
4157 + sizeof(struct btrfs_chunk));
4162 * The generation is a global counter, we'll trust it more than the others
4163 * but it's still possible that it's the one that's wrong.
4165 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4167 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4168 btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4169 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4170 && btrfs_super_cache_generation(sb) != (u64)-1)
4172 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4173 btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4178 static void btrfs_error_commit_super(struct btrfs_root *root)
4180 mutex_lock(&root->fs_info->cleaner_mutex);
4181 btrfs_run_delayed_iputs(root);
4182 mutex_unlock(&root->fs_info->cleaner_mutex);
4184 down_write(&root->fs_info->cleanup_work_sem);
4185 up_write(&root->fs_info->cleanup_work_sem);
4187 /* cleanup FS via transaction */
4188 btrfs_cleanup_transaction(root);
4191 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4193 struct btrfs_ordered_extent *ordered;
4195 spin_lock(&root->ordered_extent_lock);
4197 * This will just short circuit the ordered completion stuff which will
4198 * make sure the ordered extent gets properly cleaned up.
4200 list_for_each_entry(ordered, &root->ordered_extents,
4202 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4203 spin_unlock(&root->ordered_extent_lock);
4206 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4208 struct btrfs_root *root;
4209 struct list_head splice;
4211 INIT_LIST_HEAD(&splice);
4213 spin_lock(&fs_info->ordered_root_lock);
4214 list_splice_init(&fs_info->ordered_roots, &splice);
4215 while (!list_empty(&splice)) {
4216 root = list_first_entry(&splice, struct btrfs_root,
4218 list_move_tail(&root->ordered_root,
4219 &fs_info->ordered_roots);
4221 spin_unlock(&fs_info->ordered_root_lock);
4222 btrfs_destroy_ordered_extents(root);
4225 spin_lock(&fs_info->ordered_root_lock);
4227 spin_unlock(&fs_info->ordered_root_lock);
4230 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4231 struct btrfs_root *root)
4233 struct rb_node *node;
4234 struct btrfs_delayed_ref_root *delayed_refs;
4235 struct btrfs_delayed_ref_node *ref;
4238 delayed_refs = &trans->delayed_refs;
4240 spin_lock(&delayed_refs->lock);
4241 if (atomic_read(&delayed_refs->num_entries) == 0) {
4242 spin_unlock(&delayed_refs->lock);
4243 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4247 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4248 struct btrfs_delayed_ref_head *head;
4249 struct btrfs_delayed_ref_node *tmp;
4250 bool pin_bytes = false;
4252 head = rb_entry(node, struct btrfs_delayed_ref_head,
4254 if (!mutex_trylock(&head->mutex)) {
4255 atomic_inc(&head->node.refs);
4256 spin_unlock(&delayed_refs->lock);
4258 mutex_lock(&head->mutex);
4259 mutex_unlock(&head->mutex);
4260 btrfs_put_delayed_ref(&head->node);
4261 spin_lock(&delayed_refs->lock);
4264 spin_lock(&head->lock);
4265 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4268 list_del(&ref->list);
4269 atomic_dec(&delayed_refs->num_entries);
4270 btrfs_put_delayed_ref(ref);
4272 if (head->must_insert_reserved)
4274 btrfs_free_delayed_extent_op(head->extent_op);
4275 delayed_refs->num_heads--;
4276 if (head->processing == 0)
4277 delayed_refs->num_heads_ready--;
4278 atomic_dec(&delayed_refs->num_entries);
4279 head->node.in_tree = 0;
4280 rb_erase(&head->href_node, &delayed_refs->href_root);
4281 spin_unlock(&head->lock);
4282 spin_unlock(&delayed_refs->lock);
4283 mutex_unlock(&head->mutex);
4286 btrfs_pin_extent(root, head->node.bytenr,
4287 head->node.num_bytes, 1);
4288 btrfs_put_delayed_ref(&head->node);
4290 spin_lock(&delayed_refs->lock);
4293 spin_unlock(&delayed_refs->lock);
4298 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4300 struct btrfs_inode *btrfs_inode;
4301 struct list_head splice;
4303 INIT_LIST_HEAD(&splice);
4305 spin_lock(&root->delalloc_lock);
4306 list_splice_init(&root->delalloc_inodes, &splice);
4308 while (!list_empty(&splice)) {
4309 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4312 list_del_init(&btrfs_inode->delalloc_inodes);
4313 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4314 &btrfs_inode->runtime_flags);
4315 spin_unlock(&root->delalloc_lock);
4317 btrfs_invalidate_inodes(btrfs_inode->root);
4319 spin_lock(&root->delalloc_lock);
4322 spin_unlock(&root->delalloc_lock);
4325 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4327 struct btrfs_root *root;
4328 struct list_head splice;
4330 INIT_LIST_HEAD(&splice);
4332 spin_lock(&fs_info->delalloc_root_lock);
4333 list_splice_init(&fs_info->delalloc_roots, &splice);
4334 while (!list_empty(&splice)) {
4335 root = list_first_entry(&splice, struct btrfs_root,
4337 list_del_init(&root->delalloc_root);
4338 root = btrfs_grab_fs_root(root);
4340 spin_unlock(&fs_info->delalloc_root_lock);
4342 btrfs_destroy_delalloc_inodes(root);
4343 btrfs_put_fs_root(root);
4345 spin_lock(&fs_info->delalloc_root_lock);
4347 spin_unlock(&fs_info->delalloc_root_lock);
4350 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4351 struct extent_io_tree *dirty_pages,
4355 struct extent_buffer *eb;
4360 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4365 clear_extent_bits(dirty_pages, start, end, mark);
4366 while (start <= end) {
4367 eb = btrfs_find_tree_block(root->fs_info, start);
4368 start += root->nodesize;
4371 wait_on_extent_buffer_writeback(eb);
4373 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4375 clear_extent_buffer_dirty(eb);
4376 free_extent_buffer_stale(eb);
4383 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4384 struct extent_io_tree *pinned_extents)
4386 struct extent_io_tree *unpin;
4392 unpin = pinned_extents;
4395 ret = find_first_extent_bit(unpin, 0, &start, &end,
4396 EXTENT_DIRTY, NULL);
4400 clear_extent_dirty(unpin, start, end);
4401 btrfs_error_unpin_extent_range(root, start, end);
4406 if (unpin == &root->fs_info->freed_extents[0])
4407 unpin = &root->fs_info->freed_extents[1];
4409 unpin = &root->fs_info->freed_extents[0];
4417 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4418 struct btrfs_root *root)
4420 btrfs_destroy_delayed_refs(cur_trans, root);
4422 cur_trans->state = TRANS_STATE_COMMIT_START;
4423 wake_up(&root->fs_info->transaction_blocked_wait);
4425 cur_trans->state = TRANS_STATE_UNBLOCKED;
4426 wake_up(&root->fs_info->transaction_wait);
4428 btrfs_destroy_delayed_inodes(root);
4429 btrfs_assert_delayed_root_empty(root);
4431 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4433 btrfs_destroy_pinned_extent(root,
4434 root->fs_info->pinned_extents);
4436 cur_trans->state =TRANS_STATE_COMPLETED;
4437 wake_up(&cur_trans->commit_wait);
4440 memset(cur_trans, 0, sizeof(*cur_trans));
4441 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4445 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4447 struct btrfs_transaction *t;
4449 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4451 spin_lock(&root->fs_info->trans_lock);
4452 while (!list_empty(&root->fs_info->trans_list)) {
4453 t = list_first_entry(&root->fs_info->trans_list,
4454 struct btrfs_transaction, list);
4455 if (t->state >= TRANS_STATE_COMMIT_START) {
4456 atomic_inc(&t->use_count);
4457 spin_unlock(&root->fs_info->trans_lock);
4458 btrfs_wait_for_commit(root, t->transid);
4459 btrfs_put_transaction(t);
4460 spin_lock(&root->fs_info->trans_lock);
4463 if (t == root->fs_info->running_transaction) {
4464 t->state = TRANS_STATE_COMMIT_DOING;
4465 spin_unlock(&root->fs_info->trans_lock);
4467 * We wait for 0 num_writers since we don't hold a trans
4468 * handle open currently for this transaction.
4470 wait_event(t->writer_wait,
4471 atomic_read(&t->num_writers) == 0);
4473 spin_unlock(&root->fs_info->trans_lock);
4475 btrfs_cleanup_one_transaction(t, root);
4477 spin_lock(&root->fs_info->trans_lock);
4478 if (t == root->fs_info->running_transaction)
4479 root->fs_info->running_transaction = NULL;
4480 list_del_init(&t->list);
4481 spin_unlock(&root->fs_info->trans_lock);
4483 btrfs_put_transaction(t);
4484 trace_btrfs_transaction_commit(root);
4485 spin_lock(&root->fs_info->trans_lock);
4487 spin_unlock(&root->fs_info->trans_lock);
4488 btrfs_destroy_all_ordered_extents(root->fs_info);
4489 btrfs_destroy_delayed_inodes(root);
4490 btrfs_assert_delayed_root_empty(root);
4491 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4492 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4493 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4498 static const struct extent_io_ops btree_extent_io_ops = {
4499 .readpage_end_io_hook = btree_readpage_end_io_hook,
4500 .readpage_io_failed_hook = btree_io_failed_hook,
4501 .submit_bio_hook = btree_submit_bio_hook,
4502 /* note we're sharing with inode.c for the merge bio hook */
4503 .merge_bio_hook = btrfs_merge_bio_hook,