2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/slab.h>
30 #include <linux/migrate.h>
31 #include <linux/ratelimit.h>
32 #include <linux/uuid.h>
33 #include <linux/semaphore.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "print-tree.h"
44 #include "free-space-cache.h"
45 #include "free-space-tree.h"
46 #include "inode-map.h"
47 #include "check-integrity.h"
48 #include "rcu-string.h"
49 #include "dev-replace.h"
53 #include "compression.h"
56 #include <asm/cpufeature.h>
59 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
60 BTRFS_HEADER_FLAG_RELOC |\
61 BTRFS_SUPER_FLAG_ERROR |\
62 BTRFS_SUPER_FLAG_SEEDING |\
63 BTRFS_SUPER_FLAG_METADUMP)
65 static const struct extent_io_ops btree_extent_io_ops;
66 static void end_workqueue_fn(struct btrfs_work *work);
67 static void free_fs_root(struct btrfs_root *root);
68 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
70 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
71 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
72 struct btrfs_root *root);
73 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
74 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
75 struct extent_io_tree *dirty_pages,
77 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
78 struct extent_io_tree *pinned_extents);
79 static int btrfs_cleanup_transaction(struct btrfs_root *root);
80 static void btrfs_error_commit_super(struct btrfs_root *root);
83 * btrfs_end_io_wq structs are used to do processing in task context when an IO
84 * is complete. This is used during reads to verify checksums, and it is used
85 * by writes to insert metadata for new file extents after IO is complete.
87 struct btrfs_end_io_wq {
91 struct btrfs_fs_info *info;
93 enum btrfs_wq_endio_type metadata;
94 struct list_head list;
95 struct btrfs_work work;
98 static struct kmem_cache *btrfs_end_io_wq_cache;
100 int __init btrfs_end_io_wq_init(void)
102 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
103 sizeof(struct btrfs_end_io_wq),
105 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
107 if (!btrfs_end_io_wq_cache)
112 void btrfs_end_io_wq_exit(void)
114 kmem_cache_destroy(btrfs_end_io_wq_cache);
118 * async submit bios are used to offload expensive checksumming
119 * onto the worker threads. They checksum file and metadata bios
120 * just before they are sent down the IO stack.
122 struct async_submit_bio {
125 struct list_head list;
126 extent_submit_bio_hook_t *submit_bio_start;
127 extent_submit_bio_hook_t *submit_bio_done;
130 unsigned long bio_flags;
132 * bio_offset is optional, can be used if the pages in the bio
133 * can't tell us where in the file the bio should go
136 struct btrfs_work work;
141 * Lockdep class keys for extent_buffer->lock's in this root. For a given
142 * eb, the lockdep key is determined by the btrfs_root it belongs to and
143 * the level the eb occupies in the tree.
145 * Different roots are used for different purposes and may nest inside each
146 * other and they require separate keysets. As lockdep keys should be
147 * static, assign keysets according to the purpose of the root as indicated
148 * by btrfs_root->objectid. This ensures that all special purpose roots
149 * have separate keysets.
151 * Lock-nesting across peer nodes is always done with the immediate parent
152 * node locked thus preventing deadlock. As lockdep doesn't know this, use
153 * subclass to avoid triggering lockdep warning in such cases.
155 * The key is set by the readpage_end_io_hook after the buffer has passed
156 * csum validation but before the pages are unlocked. It is also set by
157 * btrfs_init_new_buffer on freshly allocated blocks.
159 * We also add a check to make sure the highest level of the tree is the
160 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
161 * needs update as well.
163 #ifdef CONFIG_DEBUG_LOCK_ALLOC
164 # if BTRFS_MAX_LEVEL != 8
168 static struct btrfs_lockdep_keyset {
169 u64 id; /* root objectid */
170 const char *name_stem; /* lock name stem */
171 char names[BTRFS_MAX_LEVEL + 1][20];
172 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
173 } btrfs_lockdep_keysets[] = {
174 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
175 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
176 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
177 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
178 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
179 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
180 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
181 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
182 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
183 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
184 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
185 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
186 { .id = 0, .name_stem = "tree" },
189 void __init btrfs_init_lockdep(void)
193 /* initialize lockdep class names */
194 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
195 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
197 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
198 snprintf(ks->names[j], sizeof(ks->names[j]),
199 "btrfs-%s-%02d", ks->name_stem, j);
203 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
206 struct btrfs_lockdep_keyset *ks;
208 BUG_ON(level >= ARRAY_SIZE(ks->keys));
210 /* find the matching keyset, id 0 is the default entry */
211 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
212 if (ks->id == objectid)
215 lockdep_set_class_and_name(&eb->lock,
216 &ks->keys[level], ks->names[level]);
222 * extents on the btree inode are pretty simple, there's one extent
223 * that covers the entire device
225 static struct extent_map *btree_get_extent(struct inode *inode,
226 struct page *page, size_t pg_offset, u64 start, u64 len,
229 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
230 struct extent_map *em;
233 read_lock(&em_tree->lock);
234 em = lookup_extent_mapping(em_tree, start, len);
237 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
238 read_unlock(&em_tree->lock);
241 read_unlock(&em_tree->lock);
243 em = alloc_extent_map();
245 em = ERR_PTR(-ENOMEM);
250 em->block_len = (u64)-1;
252 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
254 write_lock(&em_tree->lock);
255 ret = add_extent_mapping(em_tree, em, 0);
256 if (ret == -EEXIST) {
258 em = lookup_extent_mapping(em_tree, start, len);
265 write_unlock(&em_tree->lock);
271 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
273 return btrfs_crc32c(seed, data, len);
276 void btrfs_csum_final(u32 crc, char *result)
278 put_unaligned_le32(~crc, result);
282 * compute the csum for a btree block, and either verify it or write it
283 * into the csum field of the block.
285 static int csum_tree_block(struct btrfs_fs_info *fs_info,
286 struct extent_buffer *buf,
289 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
292 unsigned long cur_len;
293 unsigned long offset = BTRFS_CSUM_SIZE;
295 unsigned long map_start;
296 unsigned long map_len;
299 unsigned long inline_result;
301 len = buf->len - offset;
303 err = map_private_extent_buffer(buf, offset, 32,
304 &kaddr, &map_start, &map_len);
307 cur_len = min(len, map_len - (offset - map_start));
308 crc = btrfs_csum_data(kaddr + offset - map_start,
313 if (csum_size > sizeof(inline_result)) {
314 result = kzalloc(csum_size, GFP_NOFS);
318 result = (char *)&inline_result;
321 btrfs_csum_final(crc, result);
324 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
327 memcpy(&found, result, csum_size);
329 read_extent_buffer(buf, &val, 0, csum_size);
330 btrfs_warn_rl(fs_info,
331 "%s checksum verify failed on %llu wanted %X found %X "
333 fs_info->sb->s_id, buf->start,
334 val, found, btrfs_header_level(buf));
335 if (result != (char *)&inline_result)
340 write_extent_buffer(buf, result, 0, csum_size);
342 if (result != (char *)&inline_result)
348 * we can't consider a given block up to date unless the transid of the
349 * block matches the transid in the parent node's pointer. This is how we
350 * detect blocks that either didn't get written at all or got written
351 * in the wrong place.
353 static int verify_parent_transid(struct extent_io_tree *io_tree,
354 struct extent_buffer *eb, u64 parent_transid,
357 struct extent_state *cached_state = NULL;
359 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
361 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
368 btrfs_tree_read_lock(eb);
369 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
372 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
374 if (extent_buffer_uptodate(eb) &&
375 btrfs_header_generation(eb) == parent_transid) {
379 btrfs_err_rl(eb->fs_info,
380 "parent transid verify failed on %llu wanted %llu found %llu",
382 parent_transid, btrfs_header_generation(eb));
386 * Things reading via commit roots that don't have normal protection,
387 * like send, can have a really old block in cache that may point at a
388 * block that has been free'd and re-allocated. So don't clear uptodate
389 * if we find an eb that is under IO (dirty/writeback) because we could
390 * end up reading in the stale data and then writing it back out and
391 * making everybody very sad.
393 if (!extent_buffer_under_io(eb))
394 clear_extent_buffer_uptodate(eb);
396 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
397 &cached_state, GFP_NOFS);
399 btrfs_tree_read_unlock_blocking(eb);
404 * Return 0 if the superblock checksum type matches the checksum value of that
405 * algorithm. Pass the raw disk superblock data.
407 static int btrfs_check_super_csum(char *raw_disk_sb)
409 struct btrfs_super_block *disk_sb =
410 (struct btrfs_super_block *)raw_disk_sb;
411 u16 csum_type = btrfs_super_csum_type(disk_sb);
414 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
416 const int csum_size = sizeof(crc);
417 char result[csum_size];
420 * The super_block structure does not span the whole
421 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
422 * is filled with zeros and is included in the checkum.
424 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
425 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
426 btrfs_csum_final(crc, result);
428 if (memcmp(raw_disk_sb, result, csum_size))
432 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
433 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
442 * helper to read a given tree block, doing retries as required when
443 * the checksums don't match and we have alternate mirrors to try.
445 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
446 struct extent_buffer *eb,
447 u64 start, u64 parent_transid)
449 struct extent_io_tree *io_tree;
454 int failed_mirror = 0;
456 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
457 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
459 ret = read_extent_buffer_pages(io_tree, eb, start,
461 btree_get_extent, mirror_num);
463 if (!verify_parent_transid(io_tree, eb,
471 * This buffer's crc is fine, but its contents are corrupted, so
472 * there is no reason to read the other copies, they won't be
475 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
478 num_copies = btrfs_num_copies(root->fs_info,
483 if (!failed_mirror) {
485 failed_mirror = eb->read_mirror;
489 if (mirror_num == failed_mirror)
492 if (mirror_num > num_copies)
496 if (failed && !ret && failed_mirror)
497 repair_eb_io_failure(root, eb, failed_mirror);
503 * checksum a dirty tree block before IO. This has extra checks to make sure
504 * we only fill in the checksum field in the first page of a multi-page block
507 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
509 u64 start = page_offset(page);
511 struct extent_buffer *eb;
513 eb = (struct extent_buffer *)page->private;
514 if (page != eb->pages[0])
516 found_start = btrfs_header_bytenr(eb);
517 if (WARN_ON(found_start != start || !PageUptodate(page)))
519 csum_tree_block(fs_info, eb, 0);
523 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
524 struct extent_buffer *eb)
526 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
527 u8 fsid[BTRFS_UUID_SIZE];
530 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
532 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
536 fs_devices = fs_devices->seed;
541 #define CORRUPT(reason, eb, root, slot) \
542 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
543 "root=%llu, slot=%d", reason, \
544 btrfs_header_bytenr(eb), root->objectid, slot)
546 static noinline int check_leaf(struct btrfs_root *root,
547 struct extent_buffer *leaf)
549 struct btrfs_key key;
550 struct btrfs_key leaf_key;
551 u32 nritems = btrfs_header_nritems(leaf);
557 /* Check the 0 item */
558 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
559 BTRFS_LEAF_DATA_SIZE(root)) {
560 CORRUPT("invalid item offset size pair", leaf, root, 0);
565 * Check to make sure each items keys are in the correct order and their
566 * offsets make sense. We only have to loop through nritems-1 because
567 * we check the current slot against the next slot, which verifies the
568 * next slot's offset+size makes sense and that the current's slot
571 for (slot = 0; slot < nritems - 1; slot++) {
572 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
573 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
575 /* Make sure the keys are in the right order */
576 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
577 CORRUPT("bad key order", leaf, root, slot);
582 * Make sure the offset and ends are right, remember that the
583 * item data starts at the end of the leaf and grows towards the
586 if (btrfs_item_offset_nr(leaf, slot) !=
587 btrfs_item_end_nr(leaf, slot + 1)) {
588 CORRUPT("slot offset bad", leaf, root, slot);
593 * Check to make sure that we don't point outside of the leaf,
594 * just incase all the items are consistent to eachother, but
595 * all point outside of the leaf.
597 if (btrfs_item_end_nr(leaf, slot) >
598 BTRFS_LEAF_DATA_SIZE(root)) {
599 CORRUPT("slot end outside of leaf", leaf, root, slot);
607 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
608 u64 phy_offset, struct page *page,
609 u64 start, u64 end, int mirror)
613 struct extent_buffer *eb;
614 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
615 struct btrfs_fs_info *fs_info = root->fs_info;
622 eb = (struct extent_buffer *)page->private;
624 /* the pending IO might have been the only thing that kept this buffer
625 * in memory. Make sure we have a ref for all this other checks
627 extent_buffer_get(eb);
629 reads_done = atomic_dec_and_test(&eb->io_pages);
633 eb->read_mirror = mirror;
634 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
639 found_start = btrfs_header_bytenr(eb);
640 if (found_start != eb->start) {
641 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
642 found_start, eb->start);
646 if (check_tree_block_fsid(fs_info, eb)) {
647 btrfs_err_rl(fs_info, "bad fsid on block %llu",
652 found_level = btrfs_header_level(eb);
653 if (found_level >= BTRFS_MAX_LEVEL) {
654 btrfs_err(fs_info, "bad tree block level %d",
655 (int)btrfs_header_level(eb));
660 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
663 ret = csum_tree_block(fs_info, eb, 1);
670 * If this is a leaf block and it is corrupt, set the corrupt bit so
671 * that we don't try and read the other copies of this block, just
674 if (found_level == 0 && check_leaf(root, eb)) {
675 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
680 set_extent_buffer_uptodate(eb);
683 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
684 btree_readahead_hook(fs_info, eb, eb->start, ret);
688 * our io error hook is going to dec the io pages
689 * again, we have to make sure it has something
692 atomic_inc(&eb->io_pages);
693 clear_extent_buffer_uptodate(eb);
695 free_extent_buffer(eb);
700 static int btree_io_failed_hook(struct page *page, int failed_mirror)
702 struct extent_buffer *eb;
704 eb = (struct extent_buffer *)page->private;
705 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
706 eb->read_mirror = failed_mirror;
707 atomic_dec(&eb->io_pages);
708 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
709 btree_readahead_hook(eb->fs_info, eb, eb->start, -EIO);
710 return -EIO; /* we fixed nothing */
713 static void end_workqueue_bio(struct bio *bio)
715 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
716 struct btrfs_fs_info *fs_info;
717 struct btrfs_workqueue *wq;
718 btrfs_work_func_t func;
720 fs_info = end_io_wq->info;
721 end_io_wq->error = bio->bi_error;
723 if (bio->bi_rw & REQ_WRITE) {
724 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
725 wq = fs_info->endio_meta_write_workers;
726 func = btrfs_endio_meta_write_helper;
727 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
728 wq = fs_info->endio_freespace_worker;
729 func = btrfs_freespace_write_helper;
730 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
731 wq = fs_info->endio_raid56_workers;
732 func = btrfs_endio_raid56_helper;
734 wq = fs_info->endio_write_workers;
735 func = btrfs_endio_write_helper;
738 if (unlikely(end_io_wq->metadata ==
739 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
740 wq = fs_info->endio_repair_workers;
741 func = btrfs_endio_repair_helper;
742 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
743 wq = fs_info->endio_raid56_workers;
744 func = btrfs_endio_raid56_helper;
745 } else if (end_io_wq->metadata) {
746 wq = fs_info->endio_meta_workers;
747 func = btrfs_endio_meta_helper;
749 wq = fs_info->endio_workers;
750 func = btrfs_endio_helper;
754 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
755 btrfs_queue_work(wq, &end_io_wq->work);
758 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
759 enum btrfs_wq_endio_type metadata)
761 struct btrfs_end_io_wq *end_io_wq;
763 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
767 end_io_wq->private = bio->bi_private;
768 end_io_wq->end_io = bio->bi_end_io;
769 end_io_wq->info = info;
770 end_io_wq->error = 0;
771 end_io_wq->bio = bio;
772 end_io_wq->metadata = metadata;
774 bio->bi_private = end_io_wq;
775 bio->bi_end_io = end_workqueue_bio;
779 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
781 unsigned long limit = min_t(unsigned long,
782 info->thread_pool_size,
783 info->fs_devices->open_devices);
787 static void run_one_async_start(struct btrfs_work *work)
789 struct async_submit_bio *async;
792 async = container_of(work, struct async_submit_bio, work);
793 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
794 async->mirror_num, async->bio_flags,
800 static void run_one_async_done(struct btrfs_work *work)
802 struct btrfs_fs_info *fs_info;
803 struct async_submit_bio *async;
806 async = container_of(work, struct async_submit_bio, work);
807 fs_info = BTRFS_I(async->inode)->root->fs_info;
809 limit = btrfs_async_submit_limit(fs_info);
810 limit = limit * 2 / 3;
813 * atomic_dec_return implies a barrier for waitqueue_active
815 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
816 waitqueue_active(&fs_info->async_submit_wait))
817 wake_up(&fs_info->async_submit_wait);
819 /* If an error occured we just want to clean up the bio and move on */
821 async->bio->bi_error = async->error;
822 bio_endio(async->bio);
826 async->submit_bio_done(async->inode, async->rw, async->bio,
827 async->mirror_num, async->bio_flags,
831 static void run_one_async_free(struct btrfs_work *work)
833 struct async_submit_bio *async;
835 async = container_of(work, struct async_submit_bio, work);
839 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
840 int rw, struct bio *bio, int mirror_num,
841 unsigned long bio_flags,
843 extent_submit_bio_hook_t *submit_bio_start,
844 extent_submit_bio_hook_t *submit_bio_done)
846 struct async_submit_bio *async;
848 async = kmalloc(sizeof(*async), GFP_NOFS);
852 async->inode = inode;
855 async->mirror_num = mirror_num;
856 async->submit_bio_start = submit_bio_start;
857 async->submit_bio_done = submit_bio_done;
859 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
860 run_one_async_done, run_one_async_free);
862 async->bio_flags = bio_flags;
863 async->bio_offset = bio_offset;
867 atomic_inc(&fs_info->nr_async_submits);
870 btrfs_set_work_high_priority(&async->work);
872 btrfs_queue_work(fs_info->workers, &async->work);
874 while (atomic_read(&fs_info->async_submit_draining) &&
875 atomic_read(&fs_info->nr_async_submits)) {
876 wait_event(fs_info->async_submit_wait,
877 (atomic_read(&fs_info->nr_async_submits) == 0));
883 static int btree_csum_one_bio(struct bio *bio)
885 struct bio_vec *bvec;
886 struct btrfs_root *root;
889 bio_for_each_segment_all(bvec, bio, i) {
890 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
891 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
899 static int __btree_submit_bio_start(struct inode *inode, int rw,
900 struct bio *bio, int mirror_num,
901 unsigned long bio_flags,
905 * when we're called for a write, we're already in the async
906 * submission context. Just jump into btrfs_map_bio
908 return btree_csum_one_bio(bio);
911 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
912 int mirror_num, unsigned long bio_flags,
918 * when we're called for a write, we're already in the async
919 * submission context. Just jump into btrfs_map_bio
921 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
929 static int check_async_write(struct inode *inode, unsigned long bio_flags)
931 if (bio_flags & EXTENT_BIO_TREE_LOG)
934 if (static_cpu_has_safe(X86_FEATURE_XMM4_2))
940 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
941 int mirror_num, unsigned long bio_flags,
944 int async = check_async_write(inode, bio_flags);
947 if (!(rw & REQ_WRITE)) {
949 * called for a read, do the setup so that checksum validation
950 * can happen in the async kernel threads
952 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
953 bio, BTRFS_WQ_ENDIO_METADATA);
956 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
959 ret = btree_csum_one_bio(bio);
962 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
966 * kthread helpers are used to submit writes so that
967 * checksumming can happen in parallel across all CPUs
969 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
970 inode, rw, bio, mirror_num, 0,
972 __btree_submit_bio_start,
973 __btree_submit_bio_done);
986 #ifdef CONFIG_MIGRATION
987 static int btree_migratepage(struct address_space *mapping,
988 struct page *newpage, struct page *page,
989 enum migrate_mode mode)
992 * we can't safely write a btree page from here,
993 * we haven't done the locking hook
998 * Buffers may be managed in a filesystem specific way.
999 * We must have no buffers or drop them.
1001 if (page_has_private(page) &&
1002 !try_to_release_page(page, GFP_KERNEL))
1004 return migrate_page(mapping, newpage, page, mode);
1009 static int btree_writepages(struct address_space *mapping,
1010 struct writeback_control *wbc)
1012 struct btrfs_fs_info *fs_info;
1015 if (wbc->sync_mode == WB_SYNC_NONE) {
1017 if (wbc->for_kupdate)
1020 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1021 /* this is a bit racy, but that's ok */
1022 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1023 BTRFS_DIRTY_METADATA_THRESH);
1027 return btree_write_cache_pages(mapping, wbc);
1030 static int btree_readpage(struct file *file, struct page *page)
1032 struct extent_io_tree *tree;
1033 tree = &BTRFS_I(page->mapping->host)->io_tree;
1034 return extent_read_full_page(tree, page, btree_get_extent, 0);
1037 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1039 if (PageWriteback(page) || PageDirty(page))
1042 return try_release_extent_buffer(page);
1045 static void btree_invalidatepage(struct page *page, unsigned int offset,
1046 unsigned int length)
1048 struct extent_io_tree *tree;
1049 tree = &BTRFS_I(page->mapping->host)->io_tree;
1050 extent_invalidatepage(tree, page, offset);
1051 btree_releasepage(page, GFP_NOFS);
1052 if (PagePrivate(page)) {
1053 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1054 "page private not zero on page %llu",
1055 (unsigned long long)page_offset(page));
1056 ClearPagePrivate(page);
1057 set_page_private(page, 0);
1058 page_cache_release(page);
1062 static int btree_set_page_dirty(struct page *page)
1065 struct extent_buffer *eb;
1067 BUG_ON(!PagePrivate(page));
1068 eb = (struct extent_buffer *)page->private;
1070 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1071 BUG_ON(!atomic_read(&eb->refs));
1072 btrfs_assert_tree_locked(eb);
1074 return __set_page_dirty_nobuffers(page);
1077 static const struct address_space_operations btree_aops = {
1078 .readpage = btree_readpage,
1079 .writepages = btree_writepages,
1080 .releasepage = btree_releasepage,
1081 .invalidatepage = btree_invalidatepage,
1082 #ifdef CONFIG_MIGRATION
1083 .migratepage = btree_migratepage,
1085 .set_page_dirty = btree_set_page_dirty,
1088 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1090 struct extent_buffer *buf = NULL;
1091 struct inode *btree_inode = root->fs_info->btree_inode;
1093 buf = btrfs_find_create_tree_block(root, bytenr);
1096 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1097 buf, 0, WAIT_NONE, btree_get_extent, 0);
1098 free_extent_buffer(buf);
1101 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1102 int mirror_num, struct extent_buffer **eb)
1104 struct extent_buffer *buf = NULL;
1105 struct inode *btree_inode = root->fs_info->btree_inode;
1106 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1109 buf = btrfs_find_create_tree_block(root, bytenr);
1113 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1115 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1116 btree_get_extent, mirror_num);
1118 free_extent_buffer(buf);
1122 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1123 free_extent_buffer(buf);
1125 } else if (extent_buffer_uptodate(buf)) {
1128 free_extent_buffer(buf);
1133 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1136 return find_extent_buffer(fs_info, bytenr);
1139 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1142 if (btrfs_test_is_dummy_root(root))
1143 return alloc_test_extent_buffer(root->fs_info, bytenr);
1144 return alloc_extent_buffer(root->fs_info, bytenr);
1148 int btrfs_write_tree_block(struct extent_buffer *buf)
1150 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1151 buf->start + buf->len - 1);
1154 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1156 return filemap_fdatawait_range(buf->pages[0]->mapping,
1157 buf->start, buf->start + buf->len - 1);
1160 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1163 struct extent_buffer *buf = NULL;
1166 buf = btrfs_find_create_tree_block(root, bytenr);
1168 return ERR_PTR(-ENOMEM);
1170 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1172 free_extent_buffer(buf);
1173 return ERR_PTR(ret);
1179 void clean_tree_block(struct btrfs_trans_handle *trans,
1180 struct btrfs_fs_info *fs_info,
1181 struct extent_buffer *buf)
1183 if (btrfs_header_generation(buf) ==
1184 fs_info->running_transaction->transid) {
1185 btrfs_assert_tree_locked(buf);
1187 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1188 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1190 fs_info->dirty_metadata_batch);
1191 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1192 btrfs_set_lock_blocking(buf);
1193 clear_extent_buffer_dirty(buf);
1198 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1200 struct btrfs_subvolume_writers *writers;
1203 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1205 return ERR_PTR(-ENOMEM);
1207 ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1210 return ERR_PTR(ret);
1213 init_waitqueue_head(&writers->wait);
1218 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1220 percpu_counter_destroy(&writers->counter);
1224 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1225 struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1229 root->commit_root = NULL;
1230 root->sectorsize = sectorsize;
1231 root->nodesize = nodesize;
1232 root->stripesize = stripesize;
1234 root->orphan_cleanup_state = 0;
1236 root->objectid = objectid;
1237 root->last_trans = 0;
1238 root->highest_objectid = 0;
1239 root->nr_delalloc_inodes = 0;
1240 root->nr_ordered_extents = 0;
1242 root->inode_tree = RB_ROOT;
1243 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1244 root->block_rsv = NULL;
1245 root->orphan_block_rsv = NULL;
1247 INIT_LIST_HEAD(&root->dirty_list);
1248 INIT_LIST_HEAD(&root->root_list);
1249 INIT_LIST_HEAD(&root->delalloc_inodes);
1250 INIT_LIST_HEAD(&root->delalloc_root);
1251 INIT_LIST_HEAD(&root->ordered_extents);
1252 INIT_LIST_HEAD(&root->ordered_root);
1253 INIT_LIST_HEAD(&root->logged_list[0]);
1254 INIT_LIST_HEAD(&root->logged_list[1]);
1255 spin_lock_init(&root->orphan_lock);
1256 spin_lock_init(&root->inode_lock);
1257 spin_lock_init(&root->delalloc_lock);
1258 spin_lock_init(&root->ordered_extent_lock);
1259 spin_lock_init(&root->accounting_lock);
1260 spin_lock_init(&root->log_extents_lock[0]);
1261 spin_lock_init(&root->log_extents_lock[1]);
1262 mutex_init(&root->objectid_mutex);
1263 mutex_init(&root->log_mutex);
1264 mutex_init(&root->ordered_extent_mutex);
1265 mutex_init(&root->delalloc_mutex);
1266 init_waitqueue_head(&root->log_writer_wait);
1267 init_waitqueue_head(&root->log_commit_wait[0]);
1268 init_waitqueue_head(&root->log_commit_wait[1]);
1269 INIT_LIST_HEAD(&root->log_ctxs[0]);
1270 INIT_LIST_HEAD(&root->log_ctxs[1]);
1271 atomic_set(&root->log_commit[0], 0);
1272 atomic_set(&root->log_commit[1], 0);
1273 atomic_set(&root->log_writers, 0);
1274 atomic_set(&root->log_batch, 0);
1275 atomic_set(&root->orphan_inodes, 0);
1276 atomic_set(&root->refs, 1);
1277 atomic_set(&root->will_be_snapshoted, 0);
1278 atomic_set(&root->qgroup_meta_rsv, 0);
1279 root->log_transid = 0;
1280 root->log_transid_committed = -1;
1281 root->last_log_commit = 0;
1283 extent_io_tree_init(&root->dirty_log_pages,
1284 fs_info->btree_inode->i_mapping);
1286 memset(&root->root_key, 0, sizeof(root->root_key));
1287 memset(&root->root_item, 0, sizeof(root->root_item));
1288 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1290 root->defrag_trans_start = fs_info->generation;
1292 root->defrag_trans_start = 0;
1293 root->root_key.objectid = objectid;
1296 spin_lock_init(&root->root_item_lock);
1299 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1302 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1304 root->fs_info = fs_info;
1308 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1309 /* Should only be used by the testing infrastructure */
1310 struct btrfs_root *btrfs_alloc_dummy_root(void)
1312 struct btrfs_root *root;
1314 root = btrfs_alloc_root(NULL, GFP_KERNEL);
1316 return ERR_PTR(-ENOMEM);
1317 __setup_root(4096, 4096, 4096, root, NULL, 1);
1318 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1319 root->alloc_bytenr = 0;
1325 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1326 struct btrfs_fs_info *fs_info,
1329 struct extent_buffer *leaf;
1330 struct btrfs_root *tree_root = fs_info->tree_root;
1331 struct btrfs_root *root;
1332 struct btrfs_key key;
1336 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1338 return ERR_PTR(-ENOMEM);
1340 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1341 tree_root->stripesize, root, fs_info, objectid);
1342 root->root_key.objectid = objectid;
1343 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1344 root->root_key.offset = 0;
1346 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1348 ret = PTR_ERR(leaf);
1353 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1354 btrfs_set_header_bytenr(leaf, leaf->start);
1355 btrfs_set_header_generation(leaf, trans->transid);
1356 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1357 btrfs_set_header_owner(leaf, objectid);
1360 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1362 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1363 btrfs_header_chunk_tree_uuid(leaf),
1365 btrfs_mark_buffer_dirty(leaf);
1367 root->commit_root = btrfs_root_node(root);
1368 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1370 root->root_item.flags = 0;
1371 root->root_item.byte_limit = 0;
1372 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1373 btrfs_set_root_generation(&root->root_item, trans->transid);
1374 btrfs_set_root_level(&root->root_item, 0);
1375 btrfs_set_root_refs(&root->root_item, 1);
1376 btrfs_set_root_used(&root->root_item, leaf->len);
1377 btrfs_set_root_last_snapshot(&root->root_item, 0);
1378 btrfs_set_root_dirid(&root->root_item, 0);
1380 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1381 root->root_item.drop_level = 0;
1383 key.objectid = objectid;
1384 key.type = BTRFS_ROOT_ITEM_KEY;
1386 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1390 btrfs_tree_unlock(leaf);
1396 btrfs_tree_unlock(leaf);
1397 free_extent_buffer(root->commit_root);
1398 free_extent_buffer(leaf);
1402 return ERR_PTR(ret);
1405 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1406 struct btrfs_fs_info *fs_info)
1408 struct btrfs_root *root;
1409 struct btrfs_root *tree_root = fs_info->tree_root;
1410 struct extent_buffer *leaf;
1412 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1414 return ERR_PTR(-ENOMEM);
1416 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1417 tree_root->stripesize, root, fs_info,
1418 BTRFS_TREE_LOG_OBJECTID);
1420 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1421 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1422 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1425 * DON'T set REF_COWS for log trees
1427 * log trees do not get reference counted because they go away
1428 * before a real commit is actually done. They do store pointers
1429 * to file data extents, and those reference counts still get
1430 * updated (along with back refs to the log tree).
1433 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1437 return ERR_CAST(leaf);
1440 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1441 btrfs_set_header_bytenr(leaf, leaf->start);
1442 btrfs_set_header_generation(leaf, trans->transid);
1443 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1444 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1447 write_extent_buffer(root->node, root->fs_info->fsid,
1448 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1449 btrfs_mark_buffer_dirty(root->node);
1450 btrfs_tree_unlock(root->node);
1454 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1455 struct btrfs_fs_info *fs_info)
1457 struct btrfs_root *log_root;
1459 log_root = alloc_log_tree(trans, fs_info);
1460 if (IS_ERR(log_root))
1461 return PTR_ERR(log_root);
1462 WARN_ON(fs_info->log_root_tree);
1463 fs_info->log_root_tree = log_root;
1467 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1468 struct btrfs_root *root)
1470 struct btrfs_root *log_root;
1471 struct btrfs_inode_item *inode_item;
1473 log_root = alloc_log_tree(trans, root->fs_info);
1474 if (IS_ERR(log_root))
1475 return PTR_ERR(log_root);
1477 log_root->last_trans = trans->transid;
1478 log_root->root_key.offset = root->root_key.objectid;
1480 inode_item = &log_root->root_item.inode;
1481 btrfs_set_stack_inode_generation(inode_item, 1);
1482 btrfs_set_stack_inode_size(inode_item, 3);
1483 btrfs_set_stack_inode_nlink(inode_item, 1);
1484 btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1485 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1487 btrfs_set_root_node(&log_root->root_item, log_root->node);
1489 WARN_ON(root->log_root);
1490 root->log_root = log_root;
1491 root->log_transid = 0;
1492 root->log_transid_committed = -1;
1493 root->last_log_commit = 0;
1497 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1498 struct btrfs_key *key)
1500 struct btrfs_root *root;
1501 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1502 struct btrfs_path *path;
1506 path = btrfs_alloc_path();
1508 return ERR_PTR(-ENOMEM);
1510 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1516 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1517 tree_root->stripesize, root, fs_info, key->objectid);
1519 ret = btrfs_find_root(tree_root, key, path,
1520 &root->root_item, &root->root_key);
1527 generation = btrfs_root_generation(&root->root_item);
1528 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1530 if (IS_ERR(root->node)) {
1531 ret = PTR_ERR(root->node);
1533 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1535 free_extent_buffer(root->node);
1538 root->commit_root = btrfs_root_node(root);
1540 btrfs_free_path(path);
1546 root = ERR_PTR(ret);
1550 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1551 struct btrfs_key *location)
1553 struct btrfs_root *root;
1555 root = btrfs_read_tree_root(tree_root, location);
1559 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1560 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1561 btrfs_check_and_init_root_item(&root->root_item);
1567 int btrfs_init_fs_root(struct btrfs_root *root)
1570 struct btrfs_subvolume_writers *writers;
1572 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1573 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1575 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1580 writers = btrfs_alloc_subvolume_writers();
1581 if (IS_ERR(writers)) {
1582 ret = PTR_ERR(writers);
1585 root->subv_writers = writers;
1587 btrfs_init_free_ino_ctl(root);
1588 spin_lock_init(&root->ino_cache_lock);
1589 init_waitqueue_head(&root->ino_cache_wait);
1591 ret = get_anon_bdev(&root->anon_dev);
1595 mutex_lock(&root->objectid_mutex);
1596 ret = btrfs_find_highest_objectid(root,
1597 &root->highest_objectid);
1599 mutex_unlock(&root->objectid_mutex);
1603 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1605 mutex_unlock(&root->objectid_mutex);
1610 free_anon_bdev(root->anon_dev);
1612 btrfs_free_subvolume_writers(root->subv_writers);
1614 kfree(root->free_ino_ctl);
1615 kfree(root->free_ino_pinned);
1619 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1622 struct btrfs_root *root;
1624 spin_lock(&fs_info->fs_roots_radix_lock);
1625 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1626 (unsigned long)root_id);
1627 spin_unlock(&fs_info->fs_roots_radix_lock);
1631 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1632 struct btrfs_root *root)
1636 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1640 spin_lock(&fs_info->fs_roots_radix_lock);
1641 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1642 (unsigned long)root->root_key.objectid,
1645 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1646 spin_unlock(&fs_info->fs_roots_radix_lock);
1647 radix_tree_preload_end();
1652 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1653 struct btrfs_key *location,
1656 struct btrfs_root *root;
1657 struct btrfs_path *path;
1658 struct btrfs_key key;
1661 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1662 return fs_info->tree_root;
1663 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1664 return fs_info->extent_root;
1665 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1666 return fs_info->chunk_root;
1667 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1668 return fs_info->dev_root;
1669 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1670 return fs_info->csum_root;
1671 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1672 return fs_info->quota_root ? fs_info->quota_root :
1674 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1675 return fs_info->uuid_root ? fs_info->uuid_root :
1677 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1678 return fs_info->free_space_root ? fs_info->free_space_root :
1681 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1683 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1684 return ERR_PTR(-ENOENT);
1688 root = btrfs_read_fs_root(fs_info->tree_root, location);
1692 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1697 ret = btrfs_init_fs_root(root);
1701 path = btrfs_alloc_path();
1706 key.objectid = BTRFS_ORPHAN_OBJECTID;
1707 key.type = BTRFS_ORPHAN_ITEM_KEY;
1708 key.offset = location->objectid;
1710 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1711 btrfs_free_path(path);
1715 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1717 ret = btrfs_insert_fs_root(fs_info, root);
1719 if (ret == -EEXIST) {
1728 return ERR_PTR(ret);
1731 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1733 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1735 struct btrfs_device *device;
1736 struct backing_dev_info *bdi;
1739 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1742 bdi = blk_get_backing_dev_info(device->bdev);
1743 if (bdi_congested(bdi, bdi_bits)) {
1752 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1756 err = bdi_setup_and_register(bdi, "btrfs");
1760 bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE;
1761 bdi->congested_fn = btrfs_congested_fn;
1762 bdi->congested_data = info;
1763 bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1768 * called by the kthread helper functions to finally call the bio end_io
1769 * functions. This is where read checksum verification actually happens
1771 static void end_workqueue_fn(struct btrfs_work *work)
1774 struct btrfs_end_io_wq *end_io_wq;
1776 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1777 bio = end_io_wq->bio;
1779 bio->bi_error = end_io_wq->error;
1780 bio->bi_private = end_io_wq->private;
1781 bio->bi_end_io = end_io_wq->end_io;
1782 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1786 static int cleaner_kthread(void *arg)
1788 struct btrfs_root *root = arg;
1790 struct btrfs_trans_handle *trans;
1795 /* Make the cleaner go to sleep early. */
1796 if (btrfs_need_cleaner_sleep(root))
1799 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1803 * Avoid the problem that we change the status of the fs
1804 * during the above check and trylock.
1806 if (btrfs_need_cleaner_sleep(root)) {
1807 mutex_unlock(&root->fs_info->cleaner_mutex);
1811 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1812 btrfs_run_delayed_iputs(root);
1813 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1815 again = btrfs_clean_one_deleted_snapshot(root);
1816 mutex_unlock(&root->fs_info->cleaner_mutex);
1819 * The defragger has dealt with the R/O remount and umount,
1820 * needn't do anything special here.
1822 btrfs_run_defrag_inodes(root->fs_info);
1825 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1826 * with relocation (btrfs_relocate_chunk) and relocation
1827 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1828 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1829 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1830 * unused block groups.
1832 btrfs_delete_unused_bgs(root->fs_info);
1834 if (!try_to_freeze() && !again) {
1835 set_current_state(TASK_INTERRUPTIBLE);
1836 if (!kthread_should_stop())
1838 __set_current_state(TASK_RUNNING);
1840 } while (!kthread_should_stop());
1843 * Transaction kthread is stopped before us and wakes us up.
1844 * However we might have started a new transaction and COWed some
1845 * tree blocks when deleting unused block groups for example. So
1846 * make sure we commit the transaction we started to have a clean
1847 * shutdown when evicting the btree inode - if it has dirty pages
1848 * when we do the final iput() on it, eviction will trigger a
1849 * writeback for it which will fail with null pointer dereferences
1850 * since work queues and other resources were already released and
1851 * destroyed by the time the iput/eviction/writeback is made.
1853 trans = btrfs_attach_transaction(root);
1854 if (IS_ERR(trans)) {
1855 if (PTR_ERR(trans) != -ENOENT)
1856 btrfs_err(root->fs_info,
1857 "cleaner transaction attach returned %ld",
1862 ret = btrfs_commit_transaction(trans, root);
1864 btrfs_err(root->fs_info,
1865 "cleaner open transaction commit returned %d",
1872 static int transaction_kthread(void *arg)
1874 struct btrfs_root *root = arg;
1875 struct btrfs_trans_handle *trans;
1876 struct btrfs_transaction *cur;
1879 unsigned long delay;
1883 cannot_commit = false;
1884 delay = HZ * root->fs_info->commit_interval;
1885 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1887 spin_lock(&root->fs_info->trans_lock);
1888 cur = root->fs_info->running_transaction;
1890 spin_unlock(&root->fs_info->trans_lock);
1894 now = get_seconds();
1895 if (cur->state < TRANS_STATE_BLOCKED &&
1896 (now < cur->start_time ||
1897 now - cur->start_time < root->fs_info->commit_interval)) {
1898 spin_unlock(&root->fs_info->trans_lock);
1902 transid = cur->transid;
1903 spin_unlock(&root->fs_info->trans_lock);
1905 /* If the file system is aborted, this will always fail. */
1906 trans = btrfs_attach_transaction(root);
1907 if (IS_ERR(trans)) {
1908 if (PTR_ERR(trans) != -ENOENT)
1909 cannot_commit = true;
1912 if (transid == trans->transid) {
1913 btrfs_commit_transaction(trans, root);
1915 btrfs_end_transaction(trans, root);
1918 wake_up_process(root->fs_info->cleaner_kthread);
1919 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1921 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1922 &root->fs_info->fs_state)))
1923 btrfs_cleanup_transaction(root);
1924 if (!try_to_freeze()) {
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);
1932 } while (!kthread_should_stop());
1937 * this will find the highest generation in the array of
1938 * root backups. The index of the highest array is returned,
1939 * or -1 if we can't find anything.
1941 * We check to make sure the array is valid by comparing the
1942 * generation of the latest root in the array with the generation
1943 * in the super block. If they don't match we pitch it.
1945 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1948 int newest_index = -1;
1949 struct btrfs_root_backup *root_backup;
1952 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1953 root_backup = info->super_copy->super_roots + i;
1954 cur = btrfs_backup_tree_root_gen(root_backup);
1955 if (cur == newest_gen)
1959 /* check to see if we actually wrapped around */
1960 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1961 root_backup = info->super_copy->super_roots;
1962 cur = btrfs_backup_tree_root_gen(root_backup);
1963 if (cur == newest_gen)
1966 return newest_index;
1971 * find the oldest backup so we know where to store new entries
1972 * in the backup array. This will set the backup_root_index
1973 * field in the fs_info struct
1975 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1978 int newest_index = -1;
1980 newest_index = find_newest_super_backup(info, newest_gen);
1981 /* if there was garbage in there, just move along */
1982 if (newest_index == -1) {
1983 info->backup_root_index = 0;
1985 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1990 * copy all the root pointers into the super backup array.
1991 * this will bump the backup pointer by one when it is
1994 static void backup_super_roots(struct btrfs_fs_info *info)
1997 struct btrfs_root_backup *root_backup;
2000 next_backup = info->backup_root_index;
2001 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2002 BTRFS_NUM_BACKUP_ROOTS;
2005 * just overwrite the last backup if we're at the same generation
2006 * this happens only at umount
2008 root_backup = info->super_for_commit->super_roots + last_backup;
2009 if (btrfs_backup_tree_root_gen(root_backup) ==
2010 btrfs_header_generation(info->tree_root->node))
2011 next_backup = last_backup;
2013 root_backup = info->super_for_commit->super_roots + next_backup;
2016 * make sure all of our padding and empty slots get zero filled
2017 * regardless of which ones we use today
2019 memset(root_backup, 0, sizeof(*root_backup));
2021 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2023 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2024 btrfs_set_backup_tree_root_gen(root_backup,
2025 btrfs_header_generation(info->tree_root->node));
2027 btrfs_set_backup_tree_root_level(root_backup,
2028 btrfs_header_level(info->tree_root->node));
2030 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2031 btrfs_set_backup_chunk_root_gen(root_backup,
2032 btrfs_header_generation(info->chunk_root->node));
2033 btrfs_set_backup_chunk_root_level(root_backup,
2034 btrfs_header_level(info->chunk_root->node));
2036 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2037 btrfs_set_backup_extent_root_gen(root_backup,
2038 btrfs_header_generation(info->extent_root->node));
2039 btrfs_set_backup_extent_root_level(root_backup,
2040 btrfs_header_level(info->extent_root->node));
2043 * we might commit during log recovery, which happens before we set
2044 * the fs_root. Make sure it is valid before we fill it in.
2046 if (info->fs_root && info->fs_root->node) {
2047 btrfs_set_backup_fs_root(root_backup,
2048 info->fs_root->node->start);
2049 btrfs_set_backup_fs_root_gen(root_backup,
2050 btrfs_header_generation(info->fs_root->node));
2051 btrfs_set_backup_fs_root_level(root_backup,
2052 btrfs_header_level(info->fs_root->node));
2055 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2056 btrfs_set_backup_dev_root_gen(root_backup,
2057 btrfs_header_generation(info->dev_root->node));
2058 btrfs_set_backup_dev_root_level(root_backup,
2059 btrfs_header_level(info->dev_root->node));
2061 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2062 btrfs_set_backup_csum_root_gen(root_backup,
2063 btrfs_header_generation(info->csum_root->node));
2064 btrfs_set_backup_csum_root_level(root_backup,
2065 btrfs_header_level(info->csum_root->node));
2067 btrfs_set_backup_total_bytes(root_backup,
2068 btrfs_super_total_bytes(info->super_copy));
2069 btrfs_set_backup_bytes_used(root_backup,
2070 btrfs_super_bytes_used(info->super_copy));
2071 btrfs_set_backup_num_devices(root_backup,
2072 btrfs_super_num_devices(info->super_copy));
2075 * if we don't copy this out to the super_copy, it won't get remembered
2076 * for the next commit
2078 memcpy(&info->super_copy->super_roots,
2079 &info->super_for_commit->super_roots,
2080 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2084 * this copies info out of the root backup array and back into
2085 * the in-memory super block. It is meant to help iterate through
2086 * the array, so you send it the number of backups you've already
2087 * tried and the last backup index you used.
2089 * this returns -1 when it has tried all the backups
2091 static noinline int next_root_backup(struct btrfs_fs_info *info,
2092 struct btrfs_super_block *super,
2093 int *num_backups_tried, int *backup_index)
2095 struct btrfs_root_backup *root_backup;
2096 int newest = *backup_index;
2098 if (*num_backups_tried == 0) {
2099 u64 gen = btrfs_super_generation(super);
2101 newest = find_newest_super_backup(info, gen);
2105 *backup_index = newest;
2106 *num_backups_tried = 1;
2107 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2108 /* we've tried all the backups, all done */
2111 /* jump to the next oldest backup */
2112 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2113 BTRFS_NUM_BACKUP_ROOTS;
2114 *backup_index = newest;
2115 *num_backups_tried += 1;
2117 root_backup = super->super_roots + newest;
2119 btrfs_set_super_generation(super,
2120 btrfs_backup_tree_root_gen(root_backup));
2121 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2122 btrfs_set_super_root_level(super,
2123 btrfs_backup_tree_root_level(root_backup));
2124 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2127 * fixme: the total bytes and num_devices need to match or we should
2130 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2131 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2135 /* helper to cleanup workers */
2136 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2138 btrfs_destroy_workqueue(fs_info->fixup_workers);
2139 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2140 btrfs_destroy_workqueue(fs_info->workers);
2141 btrfs_destroy_workqueue(fs_info->endio_workers);
2142 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2143 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2144 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2145 btrfs_destroy_workqueue(fs_info->rmw_workers);
2146 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2147 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2148 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2149 btrfs_destroy_workqueue(fs_info->submit_workers);
2150 btrfs_destroy_workqueue(fs_info->delayed_workers);
2151 btrfs_destroy_workqueue(fs_info->caching_workers);
2152 btrfs_destroy_workqueue(fs_info->readahead_workers);
2153 btrfs_destroy_workqueue(fs_info->flush_workers);
2154 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2155 btrfs_destroy_workqueue(fs_info->extent_workers);
2158 static void free_root_extent_buffers(struct btrfs_root *root)
2161 free_extent_buffer(root->node);
2162 free_extent_buffer(root->commit_root);
2164 root->commit_root = NULL;
2168 /* helper to cleanup tree roots */
2169 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2171 free_root_extent_buffers(info->tree_root);
2173 free_root_extent_buffers(info->dev_root);
2174 free_root_extent_buffers(info->extent_root);
2175 free_root_extent_buffers(info->csum_root);
2176 free_root_extent_buffers(info->quota_root);
2177 free_root_extent_buffers(info->uuid_root);
2179 free_root_extent_buffers(info->chunk_root);
2180 free_root_extent_buffers(info->free_space_root);
2183 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2186 struct btrfs_root *gang[8];
2189 while (!list_empty(&fs_info->dead_roots)) {
2190 gang[0] = list_entry(fs_info->dead_roots.next,
2191 struct btrfs_root, root_list);
2192 list_del(&gang[0]->root_list);
2194 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2195 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2197 free_extent_buffer(gang[0]->node);
2198 free_extent_buffer(gang[0]->commit_root);
2199 btrfs_put_fs_root(gang[0]);
2204 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2209 for (i = 0; i < ret; i++)
2210 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2213 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2214 btrfs_free_log_root_tree(NULL, fs_info);
2215 btrfs_destroy_pinned_extent(fs_info->tree_root,
2216 fs_info->pinned_extents);
2220 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2222 mutex_init(&fs_info->scrub_lock);
2223 atomic_set(&fs_info->scrubs_running, 0);
2224 atomic_set(&fs_info->scrub_pause_req, 0);
2225 atomic_set(&fs_info->scrubs_paused, 0);
2226 atomic_set(&fs_info->scrub_cancel_req, 0);
2227 init_waitqueue_head(&fs_info->scrub_pause_wait);
2228 fs_info->scrub_workers_refcnt = 0;
2231 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2233 spin_lock_init(&fs_info->balance_lock);
2234 mutex_init(&fs_info->balance_mutex);
2235 atomic_set(&fs_info->balance_running, 0);
2236 atomic_set(&fs_info->balance_pause_req, 0);
2237 atomic_set(&fs_info->balance_cancel_req, 0);
2238 fs_info->balance_ctl = NULL;
2239 init_waitqueue_head(&fs_info->balance_wait_q);
2242 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2243 struct btrfs_root *tree_root)
2245 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2246 set_nlink(fs_info->btree_inode, 1);
2248 * we set the i_size on the btree inode to the max possible int.
2249 * the real end of the address space is determined by all of
2250 * the devices in the system
2252 fs_info->btree_inode->i_size = OFFSET_MAX;
2253 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2255 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2256 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2257 fs_info->btree_inode->i_mapping);
2258 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2259 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2261 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2263 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2264 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2265 sizeof(struct btrfs_key));
2266 set_bit(BTRFS_INODE_DUMMY,
2267 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2268 btrfs_insert_inode_hash(fs_info->btree_inode);
2271 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2273 fs_info->dev_replace.lock_owner = 0;
2274 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2275 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2276 rwlock_init(&fs_info->dev_replace.lock);
2277 atomic_set(&fs_info->dev_replace.read_locks, 0);
2278 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2279 init_waitqueue_head(&fs_info->replace_wait);
2280 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2283 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2285 spin_lock_init(&fs_info->qgroup_lock);
2286 mutex_init(&fs_info->qgroup_ioctl_lock);
2287 fs_info->qgroup_tree = RB_ROOT;
2288 fs_info->qgroup_op_tree = RB_ROOT;
2289 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2290 fs_info->qgroup_seq = 1;
2291 fs_info->quota_enabled = 0;
2292 fs_info->pending_quota_state = 0;
2293 fs_info->qgroup_ulist = NULL;
2294 mutex_init(&fs_info->qgroup_rescan_lock);
2297 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2298 struct btrfs_fs_devices *fs_devices)
2300 int max_active = fs_info->thread_pool_size;
2301 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2304 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2307 fs_info->delalloc_workers =
2308 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2310 fs_info->flush_workers =
2311 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2313 fs_info->caching_workers =
2314 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2317 * a higher idle thresh on the submit workers makes it much more
2318 * likely that bios will be send down in a sane order to the
2321 fs_info->submit_workers =
2322 btrfs_alloc_workqueue("submit", flags,
2323 min_t(u64, fs_devices->num_devices,
2326 fs_info->fixup_workers =
2327 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2330 * endios are largely parallel and should have a very
2333 fs_info->endio_workers =
2334 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2335 fs_info->endio_meta_workers =
2336 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2337 fs_info->endio_meta_write_workers =
2338 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2339 fs_info->endio_raid56_workers =
2340 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2341 fs_info->endio_repair_workers =
2342 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2343 fs_info->rmw_workers =
2344 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2345 fs_info->endio_write_workers =
2346 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2347 fs_info->endio_freespace_worker =
2348 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2349 fs_info->delayed_workers =
2350 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2351 fs_info->readahead_workers =
2352 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2353 fs_info->qgroup_rescan_workers =
2354 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2355 fs_info->extent_workers =
2356 btrfs_alloc_workqueue("extent-refs", flags,
2357 min_t(u64, fs_devices->num_devices,
2360 if (!(fs_info->workers && fs_info->delalloc_workers &&
2361 fs_info->submit_workers && fs_info->flush_workers &&
2362 fs_info->endio_workers && fs_info->endio_meta_workers &&
2363 fs_info->endio_meta_write_workers &&
2364 fs_info->endio_repair_workers &&
2365 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2366 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2367 fs_info->caching_workers && fs_info->readahead_workers &&
2368 fs_info->fixup_workers && fs_info->delayed_workers &&
2369 fs_info->extent_workers &&
2370 fs_info->qgroup_rescan_workers)) {
2377 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2378 struct btrfs_fs_devices *fs_devices)
2381 struct btrfs_root *tree_root = fs_info->tree_root;
2382 struct btrfs_root *log_tree_root;
2383 struct btrfs_super_block *disk_super = fs_info->super_copy;
2384 u64 bytenr = btrfs_super_log_root(disk_super);
2386 if (fs_devices->rw_devices == 0) {
2387 btrfs_warn(fs_info, "log replay required on RO media");
2391 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2395 __setup_root(tree_root->nodesize, tree_root->sectorsize,
2396 tree_root->stripesize, log_tree_root, fs_info,
2397 BTRFS_TREE_LOG_OBJECTID);
2399 log_tree_root->node = read_tree_block(tree_root, bytenr,
2400 fs_info->generation + 1);
2401 if (IS_ERR(log_tree_root->node)) {
2402 btrfs_warn(fs_info, "failed to read log tree");
2403 ret = PTR_ERR(log_tree_root->node);
2404 kfree(log_tree_root);
2406 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2407 btrfs_err(fs_info, "failed to read log tree");
2408 free_extent_buffer(log_tree_root->node);
2409 kfree(log_tree_root);
2412 /* returns with log_tree_root freed on success */
2413 ret = btrfs_recover_log_trees(log_tree_root);
2415 btrfs_std_error(tree_root->fs_info, ret,
2416 "Failed to recover log tree");
2417 free_extent_buffer(log_tree_root->node);
2418 kfree(log_tree_root);
2422 if (fs_info->sb->s_flags & MS_RDONLY) {
2423 ret = btrfs_commit_super(tree_root);
2431 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2432 struct btrfs_root *tree_root)
2434 struct btrfs_root *root;
2435 struct btrfs_key location;
2438 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2439 location.type = BTRFS_ROOT_ITEM_KEY;
2440 location.offset = 0;
2442 root = btrfs_read_tree_root(tree_root, &location);
2444 return PTR_ERR(root);
2445 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2446 fs_info->extent_root = root;
2448 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2449 root = btrfs_read_tree_root(tree_root, &location);
2451 return PTR_ERR(root);
2452 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2453 fs_info->dev_root = root;
2454 btrfs_init_devices_late(fs_info);
2456 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2457 root = btrfs_read_tree_root(tree_root, &location);
2459 return PTR_ERR(root);
2460 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2461 fs_info->csum_root = root;
2463 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2464 root = btrfs_read_tree_root(tree_root, &location);
2465 if (!IS_ERR(root)) {
2466 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2467 fs_info->quota_enabled = 1;
2468 fs_info->pending_quota_state = 1;
2469 fs_info->quota_root = root;
2472 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2473 root = btrfs_read_tree_root(tree_root, &location);
2475 ret = PTR_ERR(root);
2479 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2480 fs_info->uuid_root = root;
2483 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2484 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2485 root = btrfs_read_tree_root(tree_root, &location);
2487 return PTR_ERR(root);
2488 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2489 fs_info->free_space_root = root;
2495 int open_ctree(struct super_block *sb,
2496 struct btrfs_fs_devices *fs_devices,
2504 struct btrfs_key location;
2505 struct buffer_head *bh;
2506 struct btrfs_super_block *disk_super;
2507 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2508 struct btrfs_root *tree_root;
2509 struct btrfs_root *chunk_root;
2512 int num_backups_tried = 0;
2513 int backup_index = 0;
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_CACHE_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 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
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 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
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;
2766 printk(KERN_ERR "BTRFS: couldn't mount because of "
2767 "unsupported optional features (%Lx).\n",
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 printk(KERN_INFO "BTRFS: has skinny extents\n");
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_CACHE_SIZE) {
2786 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2787 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2788 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2791 nodesize = btrfs_super_nodesize(disk_super);
2792 sectorsize = btrfs_super_sectorsize(disk_super);
2793 stripesize = btrfs_super_stripesize(disk_super);
2794 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2795 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2798 * mixed block groups end up with duplicate but slightly offset
2799 * extent buffers for the same range. It leads to corruptions
2801 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2802 (sectorsize != nodesize)) {
2803 printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2804 "are not allowed for mixed block groups on %s\n",
2810 * Needn't use the lock because there is no other task which will
2813 btrfs_set_super_incompat_flags(disk_super, features);
2815 features = btrfs_super_compat_ro_flags(disk_super) &
2816 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2817 if (!(sb->s_flags & MS_RDONLY) && features) {
2818 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2819 "unsupported option features (%Lx).\n",
2825 max_active = fs_info->thread_pool_size;
2827 ret = btrfs_init_workqueues(fs_info, fs_devices);
2830 goto fail_sb_buffer;
2833 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2834 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2835 SZ_4M / PAGE_CACHE_SIZE);
2837 tree_root->nodesize = nodesize;
2838 tree_root->sectorsize = sectorsize;
2839 tree_root->stripesize = stripesize;
2841 sb->s_blocksize = sectorsize;
2842 sb->s_blocksize_bits = blksize_bits(sectorsize);
2844 mutex_lock(&fs_info->chunk_mutex);
2845 ret = btrfs_read_sys_array(tree_root);
2846 mutex_unlock(&fs_info->chunk_mutex);
2848 printk(KERN_ERR "BTRFS: failed to read the system "
2849 "array on %s\n", sb->s_id);
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 printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2865 if (!IS_ERR(chunk_root->node))
2866 free_extent_buffer(chunk_root->node);
2867 chunk_root->node = NULL;
2868 goto fail_tree_roots;
2870 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2871 chunk_root->commit_root = btrfs_root_node(chunk_root);
2873 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2874 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2876 ret = btrfs_read_chunk_tree(chunk_root);
2878 printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2880 goto fail_tree_roots;
2884 * keep the device that is marked to be the target device for the
2885 * dev_replace procedure
2887 btrfs_close_extra_devices(fs_devices, 0);
2889 if (!fs_devices->latest_bdev) {
2890 printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2892 goto fail_tree_roots;
2896 generation = btrfs_super_generation(disk_super);
2898 tree_root->node = read_tree_block(tree_root,
2899 btrfs_super_root(disk_super),
2901 if (IS_ERR(tree_root->node) ||
2902 !extent_buffer_uptodate(tree_root->node)) {
2903 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2905 if (!IS_ERR(tree_root->node))
2906 free_extent_buffer(tree_root->node);
2907 tree_root->node = NULL;
2908 goto recovery_tree_root;
2911 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2912 tree_root->commit_root = btrfs_root_node(tree_root);
2913 btrfs_set_root_refs(&tree_root->root_item, 1);
2915 mutex_lock(&tree_root->objectid_mutex);
2916 ret = btrfs_find_highest_objectid(tree_root,
2917 &tree_root->highest_objectid);
2919 mutex_unlock(&tree_root->objectid_mutex);
2920 goto recovery_tree_root;
2923 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2925 mutex_unlock(&tree_root->objectid_mutex);
2927 ret = btrfs_read_roots(fs_info, tree_root);
2929 goto recovery_tree_root;
2931 fs_info->generation = generation;
2932 fs_info->last_trans_committed = generation;
2934 ret = btrfs_recover_balance(fs_info);
2936 printk(KERN_ERR "BTRFS: failed to recover balance\n");
2937 goto fail_block_groups;
2940 ret = btrfs_init_dev_stats(fs_info);
2942 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2944 goto fail_block_groups;
2947 ret = btrfs_init_dev_replace(fs_info);
2949 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2950 goto fail_block_groups;
2953 btrfs_close_extra_devices(fs_devices, 1);
2955 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2957 pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret);
2958 goto fail_block_groups;
2961 ret = btrfs_sysfs_add_device(fs_devices);
2963 pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret);
2964 goto fail_fsdev_sysfs;
2967 ret = btrfs_sysfs_add_mounted(fs_info);
2969 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2970 goto fail_fsdev_sysfs;
2973 ret = btrfs_init_space_info(fs_info);
2975 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2979 ret = btrfs_read_block_groups(fs_info->extent_root);
2981 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2984 fs_info->num_tolerated_disk_barrier_failures =
2985 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2986 if (fs_info->fs_devices->missing_devices >
2987 fs_info->num_tolerated_disk_barrier_failures &&
2988 !(sb->s_flags & MS_RDONLY)) {
2989 pr_warn("BTRFS: missing devices(%llu) exceeds the limit(%d), writeable mount is not allowed\n",
2990 fs_info->fs_devices->missing_devices,
2991 fs_info->num_tolerated_disk_barrier_failures);
2995 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2997 if (IS_ERR(fs_info->cleaner_kthread))
3000 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3002 "btrfs-transaction");
3003 if (IS_ERR(fs_info->transaction_kthread))
3006 if (!btrfs_test_opt(tree_root, SSD) &&
3007 !btrfs_test_opt(tree_root, NOSSD) &&
3008 !fs_info->fs_devices->rotating) {
3009 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
3011 btrfs_set_opt(fs_info->mount_opt, SSD);
3015 * Mount does not set all options immediatelly, we can do it now and do
3016 * not have to wait for transaction commit
3018 btrfs_apply_pending_changes(fs_info);
3020 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3021 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
3022 ret = btrfsic_mount(tree_root, fs_devices,
3023 btrfs_test_opt(tree_root,
3024 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3026 fs_info->check_integrity_print_mask);
3028 printk(KERN_WARNING "BTRFS: failed to initialize"
3029 " integrity check module %s\n", sb->s_id);
3032 ret = btrfs_read_qgroup_config(fs_info);
3034 goto fail_trans_kthread;
3036 /* do not make disk changes in broken FS or nologreplay is given */
3037 if (btrfs_super_log_root(disk_super) != 0 &&
3038 !btrfs_test_opt(tree_root, NOLOGREPLAY)) {
3039 ret = btrfs_replay_log(fs_info, fs_devices);
3046 ret = btrfs_find_orphan_roots(tree_root);
3050 if (!(sb->s_flags & MS_RDONLY)) {
3051 ret = btrfs_cleanup_fs_roots(fs_info);
3055 mutex_lock(&fs_info->cleaner_mutex);
3056 ret = btrfs_recover_relocation(tree_root);
3057 mutex_unlock(&fs_info->cleaner_mutex);
3060 "BTRFS: failed to recover relocation\n");
3066 location.objectid = BTRFS_FS_TREE_OBJECTID;
3067 location.type = BTRFS_ROOT_ITEM_KEY;
3068 location.offset = 0;
3070 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3071 if (IS_ERR(fs_info->fs_root)) {
3072 err = PTR_ERR(fs_info->fs_root);
3076 if (sb->s_flags & MS_RDONLY)
3079 if (btrfs_test_opt(tree_root, FREE_SPACE_TREE) &&
3080 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3081 pr_info("BTRFS: creating free space tree\n");
3082 ret = btrfs_create_free_space_tree(fs_info);
3084 pr_warn("BTRFS: failed to create free space tree %d\n",
3086 close_ctree(tree_root);
3091 down_read(&fs_info->cleanup_work_sem);
3092 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3093 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3094 up_read(&fs_info->cleanup_work_sem);
3095 close_ctree(tree_root);
3098 up_read(&fs_info->cleanup_work_sem);
3100 ret = btrfs_resume_balance_async(fs_info);
3102 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
3103 close_ctree(tree_root);
3107 ret = btrfs_resume_dev_replace_async(fs_info);
3109 pr_warn("BTRFS: failed to resume dev_replace\n");
3110 close_ctree(tree_root);
3114 btrfs_qgroup_rescan_resume(fs_info);
3116 if (btrfs_test_opt(tree_root, CLEAR_CACHE) &&
3117 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3118 pr_info("BTRFS: clearing free space tree\n");
3119 ret = btrfs_clear_free_space_tree(fs_info);
3121 pr_warn("BTRFS: failed to clear free space tree %d\n",
3123 close_ctree(tree_root);
3128 if (!fs_info->uuid_root) {
3129 pr_info("BTRFS: creating UUID tree\n");
3130 ret = btrfs_create_uuid_tree(fs_info);
3132 pr_warn("BTRFS: failed to create the UUID tree %d\n",
3134 close_ctree(tree_root);
3137 } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3138 fs_info->generation !=
3139 btrfs_super_uuid_tree_generation(disk_super)) {
3140 pr_info("BTRFS: checking UUID tree\n");
3141 ret = btrfs_check_uuid_tree(fs_info);
3143 pr_warn("BTRFS: failed to check the UUID tree %d\n",
3145 close_ctree(tree_root);
3149 fs_info->update_uuid_tree_gen = 1;
3155 * backuproot only affect mount behavior, and if open_ctree succeeded,
3156 * no need to keep the flag
3158 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3163 btrfs_free_qgroup_config(fs_info);
3165 kthread_stop(fs_info->transaction_kthread);
3166 btrfs_cleanup_transaction(fs_info->tree_root);
3167 btrfs_free_fs_roots(fs_info);
3169 kthread_stop(fs_info->cleaner_kthread);
3172 * make sure we're done with the btree inode before we stop our
3175 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3178 btrfs_sysfs_remove_mounted(fs_info);
3181 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3184 btrfs_put_block_group_cache(fs_info);
3185 btrfs_free_block_groups(fs_info);
3188 free_root_pointers(fs_info, 1);
3189 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3192 btrfs_stop_all_workers(fs_info);
3195 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3197 iput(fs_info->btree_inode);
3199 percpu_counter_destroy(&fs_info->bio_counter);
3200 fail_delalloc_bytes:
3201 percpu_counter_destroy(&fs_info->delalloc_bytes);
3202 fail_dirty_metadata_bytes:
3203 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3205 bdi_destroy(&fs_info->bdi);
3207 cleanup_srcu_struct(&fs_info->subvol_srcu);
3209 btrfs_free_stripe_hash_table(fs_info);
3210 btrfs_close_devices(fs_info->fs_devices);
3214 if (!btrfs_test_opt(tree_root, USEBACKUPROOT))
3215 goto fail_tree_roots;
3217 free_root_pointers(fs_info, 0);
3219 /* don't use the log in recovery mode, it won't be valid */
3220 btrfs_set_super_log_root(disk_super, 0);
3222 /* we can't trust the free space cache either */
3223 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3225 ret = next_root_backup(fs_info, fs_info->super_copy,
3226 &num_backups_tried, &backup_index);
3228 goto fail_block_groups;
3229 goto retry_root_backup;
3232 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3235 set_buffer_uptodate(bh);
3237 struct btrfs_device *device = (struct btrfs_device *)
3240 btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3241 "lost page write due to IO error on %s",
3242 rcu_str_deref(device->name));
3243 /* note, we dont' set_buffer_write_io_error because we have
3244 * our own ways of dealing with the IO errors
3246 clear_buffer_uptodate(bh);
3247 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3253 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3254 struct buffer_head **bh_ret)
3256 struct buffer_head *bh;
3257 struct btrfs_super_block *super;
3260 bytenr = btrfs_sb_offset(copy_num);
3261 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3264 bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3266 * If we fail to read from the underlying devices, as of now
3267 * the best option we have is to mark it EIO.
3272 super = (struct btrfs_super_block *)bh->b_data;
3273 if (btrfs_super_bytenr(super) != bytenr ||
3274 btrfs_super_magic(super) != BTRFS_MAGIC) {
3284 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3286 struct buffer_head *bh;
3287 struct buffer_head *latest = NULL;
3288 struct btrfs_super_block *super;
3293 /* we would like to check all the supers, but that would make
3294 * a btrfs mount succeed after a mkfs from a different FS.
3295 * So, we need to add a special mount option to scan for
3296 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3298 for (i = 0; i < 1; i++) {
3299 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3303 super = (struct btrfs_super_block *)bh->b_data;
3305 if (!latest || btrfs_super_generation(super) > transid) {
3308 transid = btrfs_super_generation(super);
3315 return ERR_PTR(ret);
3321 * this should be called twice, once with wait == 0 and
3322 * once with wait == 1. When wait == 0 is done, all the buffer heads
3323 * we write are pinned.
3325 * They are released when wait == 1 is done.
3326 * max_mirrors must be the same for both runs, and it indicates how
3327 * many supers on this one device should be written.
3329 * max_mirrors == 0 means to write them all.
3331 static int write_dev_supers(struct btrfs_device *device,
3332 struct btrfs_super_block *sb,
3333 int do_barriers, int wait, int max_mirrors)
3335 struct buffer_head *bh;
3342 if (max_mirrors == 0)
3343 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3345 for (i = 0; i < max_mirrors; i++) {
3346 bytenr = btrfs_sb_offset(i);
3347 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3348 device->commit_total_bytes)
3352 bh = __find_get_block(device->bdev, bytenr / 4096,
3353 BTRFS_SUPER_INFO_SIZE);
3359 if (!buffer_uptodate(bh))
3362 /* drop our reference */
3365 /* drop the reference from the wait == 0 run */
3369 btrfs_set_super_bytenr(sb, bytenr);
3372 crc = btrfs_csum_data((char *)sb +
3373 BTRFS_CSUM_SIZE, crc,
3374 BTRFS_SUPER_INFO_SIZE -
3376 btrfs_csum_final(crc, sb->csum);
3379 * one reference for us, and we leave it for the
3382 bh = __getblk(device->bdev, bytenr / 4096,
3383 BTRFS_SUPER_INFO_SIZE);
3385 btrfs_err(device->dev_root->fs_info,
3386 "couldn't get super buffer head for bytenr %llu",
3392 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3394 /* one reference for submit_bh */
3397 set_buffer_uptodate(bh);
3399 bh->b_end_io = btrfs_end_buffer_write_sync;
3400 bh->b_private = device;
3404 * we fua the first super. The others we allow
3408 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3410 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3414 return errors < i ? 0 : -1;
3418 * endio for the write_dev_flush, this will wake anyone waiting
3419 * for the barrier when it is done
3421 static void btrfs_end_empty_barrier(struct bio *bio)
3423 if (bio->bi_private)
3424 complete(bio->bi_private);
3429 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3430 * sent down. With wait == 1, it waits for the previous flush.
3432 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3435 static int write_dev_flush(struct btrfs_device *device, int wait)
3440 if (device->nobarriers)
3444 bio = device->flush_bio;
3448 wait_for_completion(&device->flush_wait);
3450 if (bio->bi_error) {
3451 ret = bio->bi_error;
3452 btrfs_dev_stat_inc_and_print(device,
3453 BTRFS_DEV_STAT_FLUSH_ERRS);
3456 /* drop the reference from the wait == 0 run */
3458 device->flush_bio = NULL;
3464 * one reference for us, and we leave it for the
3467 device->flush_bio = NULL;
3468 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3472 bio->bi_end_io = btrfs_end_empty_barrier;
3473 bio->bi_bdev = device->bdev;
3474 init_completion(&device->flush_wait);
3475 bio->bi_private = &device->flush_wait;
3476 device->flush_bio = bio;
3479 btrfsic_submit_bio(WRITE_FLUSH, bio);
3485 * send an empty flush down to each device in parallel,
3486 * then wait for them
3488 static int barrier_all_devices(struct btrfs_fs_info *info)
3490 struct list_head *head;
3491 struct btrfs_device *dev;
3492 int errors_send = 0;
3493 int errors_wait = 0;
3496 /* send down all the barriers */
3497 head = &info->fs_devices->devices;
3498 list_for_each_entry_rcu(dev, head, dev_list) {
3505 if (!dev->in_fs_metadata || !dev->writeable)
3508 ret = write_dev_flush(dev, 0);
3513 /* wait for all the barriers */
3514 list_for_each_entry_rcu(dev, head, dev_list) {
3521 if (!dev->in_fs_metadata || !dev->writeable)
3524 ret = write_dev_flush(dev, 1);
3528 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3529 errors_wait > info->num_tolerated_disk_barrier_failures)
3534 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3537 int min_tolerated = INT_MAX;
3539 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3540 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3541 min_tolerated = min(min_tolerated,
3542 btrfs_raid_array[BTRFS_RAID_SINGLE].
3543 tolerated_failures);
3545 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3546 if (raid_type == BTRFS_RAID_SINGLE)
3548 if (!(flags & btrfs_raid_group[raid_type]))
3550 min_tolerated = min(min_tolerated,
3551 btrfs_raid_array[raid_type].
3552 tolerated_failures);
3555 if (min_tolerated == INT_MAX) {
3556 pr_warn("BTRFS: unknown raid flag: %llu\n", flags);
3560 return min_tolerated;
3563 int btrfs_calc_num_tolerated_disk_barrier_failures(
3564 struct btrfs_fs_info *fs_info)
3566 struct btrfs_ioctl_space_info space;
3567 struct btrfs_space_info *sinfo;
3568 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3569 BTRFS_BLOCK_GROUP_SYSTEM,
3570 BTRFS_BLOCK_GROUP_METADATA,
3571 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3574 int num_tolerated_disk_barrier_failures =
3575 (int)fs_info->fs_devices->num_devices;
3577 for (i = 0; i < ARRAY_SIZE(types); i++) {
3578 struct btrfs_space_info *tmp;
3582 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3583 if (tmp->flags == types[i]) {
3593 down_read(&sinfo->groups_sem);
3594 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3597 if (list_empty(&sinfo->block_groups[c]))
3600 btrfs_get_block_group_info(&sinfo->block_groups[c],
3602 if (space.total_bytes == 0 || space.used_bytes == 0)
3604 flags = space.flags;
3606 num_tolerated_disk_barrier_failures = min(
3607 num_tolerated_disk_barrier_failures,
3608 btrfs_get_num_tolerated_disk_barrier_failures(
3611 up_read(&sinfo->groups_sem);
3614 return num_tolerated_disk_barrier_failures;
3617 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3619 struct list_head *head;
3620 struct btrfs_device *dev;
3621 struct btrfs_super_block *sb;
3622 struct btrfs_dev_item *dev_item;
3626 int total_errors = 0;
3629 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3630 backup_super_roots(root->fs_info);
3632 sb = root->fs_info->super_for_commit;
3633 dev_item = &sb->dev_item;
3635 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3636 head = &root->fs_info->fs_devices->devices;
3637 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3640 ret = barrier_all_devices(root->fs_info);
3643 &root->fs_info->fs_devices->device_list_mutex);
3644 btrfs_std_error(root->fs_info, ret,
3645 "errors while submitting device barriers.");
3650 list_for_each_entry_rcu(dev, head, dev_list) {
3655 if (!dev->in_fs_metadata || !dev->writeable)
3658 btrfs_set_stack_device_generation(dev_item, 0);
3659 btrfs_set_stack_device_type(dev_item, dev->type);
3660 btrfs_set_stack_device_id(dev_item, dev->devid);
3661 btrfs_set_stack_device_total_bytes(dev_item,
3662 dev->commit_total_bytes);
3663 btrfs_set_stack_device_bytes_used(dev_item,
3664 dev->commit_bytes_used);
3665 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3666 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3667 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3668 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3669 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3671 flags = btrfs_super_flags(sb);
3672 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3674 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3678 if (total_errors > max_errors) {
3679 btrfs_err(root->fs_info, "%d errors while writing supers",
3681 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3683 /* FUA is masked off if unsupported and can't be the reason */
3684 btrfs_std_error(root->fs_info, -EIO,
3685 "%d errors while writing supers", total_errors);
3690 list_for_each_entry_rcu(dev, head, dev_list) {
3693 if (!dev->in_fs_metadata || !dev->writeable)
3696 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3700 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3701 if (total_errors > max_errors) {
3702 btrfs_std_error(root->fs_info, -EIO,
3703 "%d errors while writing supers", total_errors);
3709 int write_ctree_super(struct btrfs_trans_handle *trans,
3710 struct btrfs_root *root, int max_mirrors)
3712 return write_all_supers(root, max_mirrors);
3715 /* Drop a fs root from the radix tree and free it. */
3716 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3717 struct btrfs_root *root)
3719 spin_lock(&fs_info->fs_roots_radix_lock);
3720 radix_tree_delete(&fs_info->fs_roots_radix,
3721 (unsigned long)root->root_key.objectid);
3722 spin_unlock(&fs_info->fs_roots_radix_lock);
3724 if (btrfs_root_refs(&root->root_item) == 0)
3725 synchronize_srcu(&fs_info->subvol_srcu);
3727 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3728 btrfs_free_log(NULL, root);
3730 if (root->free_ino_pinned)
3731 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3732 if (root->free_ino_ctl)
3733 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3737 static void free_fs_root(struct btrfs_root *root)
3739 iput(root->ino_cache_inode);
3740 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3741 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3742 root->orphan_block_rsv = NULL;
3744 free_anon_bdev(root->anon_dev);
3745 if (root->subv_writers)
3746 btrfs_free_subvolume_writers(root->subv_writers);
3747 free_extent_buffer(root->node);
3748 free_extent_buffer(root->commit_root);
3749 kfree(root->free_ino_ctl);
3750 kfree(root->free_ino_pinned);
3752 btrfs_put_fs_root(root);
3755 void btrfs_free_fs_root(struct btrfs_root *root)
3760 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3762 u64 root_objectid = 0;
3763 struct btrfs_root *gang[8];
3766 unsigned int ret = 0;
3770 index = srcu_read_lock(&fs_info->subvol_srcu);
3771 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3772 (void **)gang, root_objectid,
3775 srcu_read_unlock(&fs_info->subvol_srcu, index);
3778 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3780 for (i = 0; i < ret; i++) {
3781 /* Avoid to grab roots in dead_roots */
3782 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3786 /* grab all the search result for later use */
3787 gang[i] = btrfs_grab_fs_root(gang[i]);
3789 srcu_read_unlock(&fs_info->subvol_srcu, index);
3791 for (i = 0; i < ret; i++) {
3794 root_objectid = gang[i]->root_key.objectid;
3795 err = btrfs_orphan_cleanup(gang[i]);
3798 btrfs_put_fs_root(gang[i]);
3803 /* release the uncleaned roots due to error */
3804 for (; i < ret; i++) {
3806 btrfs_put_fs_root(gang[i]);
3811 int btrfs_commit_super(struct btrfs_root *root)
3813 struct btrfs_trans_handle *trans;
3815 mutex_lock(&root->fs_info->cleaner_mutex);
3816 btrfs_run_delayed_iputs(root);
3817 mutex_unlock(&root->fs_info->cleaner_mutex);
3818 wake_up_process(root->fs_info->cleaner_kthread);
3820 /* wait until ongoing cleanup work done */
3821 down_write(&root->fs_info->cleanup_work_sem);
3822 up_write(&root->fs_info->cleanup_work_sem);
3824 trans = btrfs_join_transaction(root);
3826 return PTR_ERR(trans);
3827 return btrfs_commit_transaction(trans, root);
3830 void close_ctree(struct btrfs_root *root)
3832 struct btrfs_fs_info *fs_info = root->fs_info;
3835 fs_info->closing = 1;
3838 /* wait for the qgroup rescan worker to stop */
3839 btrfs_qgroup_wait_for_completion(fs_info);
3841 /* wait for the uuid_scan task to finish */
3842 down(&fs_info->uuid_tree_rescan_sem);
3843 /* avoid complains from lockdep et al., set sem back to initial state */
3844 up(&fs_info->uuid_tree_rescan_sem);
3846 /* pause restriper - we want to resume on mount */
3847 btrfs_pause_balance(fs_info);
3849 btrfs_dev_replace_suspend_for_unmount(fs_info);
3851 btrfs_scrub_cancel(fs_info);
3853 /* wait for any defraggers to finish */
3854 wait_event(fs_info->transaction_wait,
3855 (atomic_read(&fs_info->defrag_running) == 0));
3857 /* clear out the rbtree of defraggable inodes */
3858 btrfs_cleanup_defrag_inodes(fs_info);
3860 cancel_work_sync(&fs_info->async_reclaim_work);
3862 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3864 * If the cleaner thread is stopped and there are
3865 * block groups queued for removal, the deletion will be
3866 * skipped when we quit the cleaner thread.
3868 btrfs_delete_unused_bgs(root->fs_info);
3870 ret = btrfs_commit_super(root);
3872 btrfs_err(fs_info, "commit super ret %d", ret);
3875 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3876 btrfs_error_commit_super(root);
3878 kthread_stop(fs_info->transaction_kthread);
3879 kthread_stop(fs_info->cleaner_kthread);
3881 fs_info->closing = 2;
3884 btrfs_free_qgroup_config(fs_info);
3886 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3887 btrfs_info(fs_info, "at unmount delalloc count %lld",
3888 percpu_counter_sum(&fs_info->delalloc_bytes));
3891 btrfs_sysfs_remove_mounted(fs_info);
3892 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3894 btrfs_free_fs_roots(fs_info);
3896 btrfs_put_block_group_cache(fs_info);
3898 btrfs_free_block_groups(fs_info);
3901 * we must make sure there is not any read request to
3902 * submit after we stopping all workers.
3904 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3905 btrfs_stop_all_workers(fs_info);
3908 free_root_pointers(fs_info, 1);
3910 iput(fs_info->btree_inode);
3912 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3913 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3914 btrfsic_unmount(root, fs_info->fs_devices);
3917 btrfs_close_devices(fs_info->fs_devices);
3918 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3920 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3921 percpu_counter_destroy(&fs_info->delalloc_bytes);
3922 percpu_counter_destroy(&fs_info->bio_counter);
3923 bdi_destroy(&fs_info->bdi);
3924 cleanup_srcu_struct(&fs_info->subvol_srcu);
3926 btrfs_free_stripe_hash_table(fs_info);
3928 __btrfs_free_block_rsv(root->orphan_block_rsv);
3929 root->orphan_block_rsv = NULL;
3932 while (!list_empty(&fs_info->pinned_chunks)) {
3933 struct extent_map *em;
3935 em = list_first_entry(&fs_info->pinned_chunks,
3936 struct extent_map, list);
3937 list_del_init(&em->list);
3938 free_extent_map(em);
3940 unlock_chunks(root);
3943 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3947 struct inode *btree_inode = buf->pages[0]->mapping->host;
3949 ret = extent_buffer_uptodate(buf);
3953 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3954 parent_transid, atomic);
3960 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3962 struct btrfs_root *root;
3963 u64 transid = btrfs_header_generation(buf);
3966 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3968 * This is a fast path so only do this check if we have sanity tests
3969 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3970 * outside of the sanity tests.
3972 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3975 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3976 btrfs_assert_tree_locked(buf);
3977 if (transid != root->fs_info->generation)
3978 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3979 "found %llu running %llu\n",
3980 buf->start, transid, root->fs_info->generation);
3981 was_dirty = set_extent_buffer_dirty(buf);
3983 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3985 root->fs_info->dirty_metadata_batch);
3986 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3987 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3988 btrfs_print_leaf(root, buf);
3994 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3998 * looks as though older kernels can get into trouble with
3999 * this code, they end up stuck in balance_dirty_pages forever
4003 if (current->flags & PF_MEMALLOC)
4007 btrfs_balance_delayed_items(root);
4009 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
4010 BTRFS_DIRTY_METADATA_THRESH);
4012 balance_dirty_pages_ratelimited(
4013 root->fs_info->btree_inode->i_mapping);
4017 void btrfs_btree_balance_dirty(struct btrfs_root *root)
4019 __btrfs_btree_balance_dirty(root, 1);
4022 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
4024 __btrfs_btree_balance_dirty(root, 0);
4027 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4029 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4030 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
4033 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4036 struct btrfs_super_block *sb = fs_info->super_copy;
4037 u64 nodesize = btrfs_super_nodesize(sb);
4038 u64 sectorsize = btrfs_super_sectorsize(sb);
4041 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4042 printk(KERN_ERR "BTRFS: no valid FS found\n");
4045 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4046 printk(KERN_WARNING "BTRFS: unrecognized super flag: %llu\n",
4047 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4048 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4049 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
4050 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4053 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4054 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
4055 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4058 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4059 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
4060 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4065 * Check sectorsize and nodesize first, other check will need it.
4066 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4068 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4069 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4070 printk(KERN_ERR "BTRFS: invalid sectorsize %llu\n", sectorsize);
4073 /* Only PAGE SIZE is supported yet */
4074 if (sectorsize != PAGE_CACHE_SIZE) {
4075 printk(KERN_ERR "BTRFS: sectorsize %llu not supported yet, only support %lu\n",
4076 sectorsize, PAGE_CACHE_SIZE);
4079 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4080 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4081 printk(KERN_ERR "BTRFS: invalid nodesize %llu\n", nodesize);
4084 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4085 printk(KERN_ERR "BTRFS: invalid leafsize %u, should be %llu\n",
4086 le32_to_cpu(sb->__unused_leafsize),
4091 /* Root alignment check */
4092 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4093 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
4094 btrfs_super_root(sb));
4097 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4098 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
4099 btrfs_super_chunk_root(sb));
4102 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4103 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
4104 btrfs_super_log_root(sb));
4108 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4109 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4110 fs_info->fsid, sb->dev_item.fsid);
4115 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4118 if (btrfs_super_num_devices(sb) > (1UL << 31))
4119 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4120 btrfs_super_num_devices(sb));
4121 if (btrfs_super_num_devices(sb) == 0) {
4122 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4126 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4127 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4128 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4133 * Obvious sys_chunk_array corruptions, it must hold at least one key
4136 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4137 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4138 btrfs_super_sys_array_size(sb),
4139 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4142 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4143 + sizeof(struct btrfs_chunk)) {
4144 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4145 btrfs_super_sys_array_size(sb),
4146 sizeof(struct btrfs_disk_key)
4147 + sizeof(struct btrfs_chunk));
4152 * The generation is a global counter, we'll trust it more than the others
4153 * but it's still possible that it's the one that's wrong.
4155 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4157 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4158 btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4159 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4160 && btrfs_super_cache_generation(sb) != (u64)-1)
4162 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4163 btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4168 static void btrfs_error_commit_super(struct btrfs_root *root)
4170 mutex_lock(&root->fs_info->cleaner_mutex);
4171 btrfs_run_delayed_iputs(root);
4172 mutex_unlock(&root->fs_info->cleaner_mutex);
4174 down_write(&root->fs_info->cleanup_work_sem);
4175 up_write(&root->fs_info->cleanup_work_sem);
4177 /* cleanup FS via transaction */
4178 btrfs_cleanup_transaction(root);
4181 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4183 struct btrfs_ordered_extent *ordered;
4185 spin_lock(&root->ordered_extent_lock);
4187 * This will just short circuit the ordered completion stuff which will
4188 * make sure the ordered extent gets properly cleaned up.
4190 list_for_each_entry(ordered, &root->ordered_extents,
4192 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4193 spin_unlock(&root->ordered_extent_lock);
4196 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4198 struct btrfs_root *root;
4199 struct list_head splice;
4201 INIT_LIST_HEAD(&splice);
4203 spin_lock(&fs_info->ordered_root_lock);
4204 list_splice_init(&fs_info->ordered_roots, &splice);
4205 while (!list_empty(&splice)) {
4206 root = list_first_entry(&splice, struct btrfs_root,
4208 list_move_tail(&root->ordered_root,
4209 &fs_info->ordered_roots);
4211 spin_unlock(&fs_info->ordered_root_lock);
4212 btrfs_destroy_ordered_extents(root);
4215 spin_lock(&fs_info->ordered_root_lock);
4217 spin_unlock(&fs_info->ordered_root_lock);
4220 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4221 struct btrfs_root *root)
4223 struct rb_node *node;
4224 struct btrfs_delayed_ref_root *delayed_refs;
4225 struct btrfs_delayed_ref_node *ref;
4228 delayed_refs = &trans->delayed_refs;
4230 spin_lock(&delayed_refs->lock);
4231 if (atomic_read(&delayed_refs->num_entries) == 0) {
4232 spin_unlock(&delayed_refs->lock);
4233 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4237 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4238 struct btrfs_delayed_ref_head *head;
4239 struct btrfs_delayed_ref_node *tmp;
4240 bool pin_bytes = false;
4242 head = rb_entry(node, struct btrfs_delayed_ref_head,
4244 if (!mutex_trylock(&head->mutex)) {
4245 atomic_inc(&head->node.refs);
4246 spin_unlock(&delayed_refs->lock);
4248 mutex_lock(&head->mutex);
4249 mutex_unlock(&head->mutex);
4250 btrfs_put_delayed_ref(&head->node);
4251 spin_lock(&delayed_refs->lock);
4254 spin_lock(&head->lock);
4255 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4258 list_del(&ref->list);
4259 atomic_dec(&delayed_refs->num_entries);
4260 btrfs_put_delayed_ref(ref);
4262 if (head->must_insert_reserved)
4264 btrfs_free_delayed_extent_op(head->extent_op);
4265 delayed_refs->num_heads--;
4266 if (head->processing == 0)
4267 delayed_refs->num_heads_ready--;
4268 atomic_dec(&delayed_refs->num_entries);
4269 head->node.in_tree = 0;
4270 rb_erase(&head->href_node, &delayed_refs->href_root);
4271 spin_unlock(&head->lock);
4272 spin_unlock(&delayed_refs->lock);
4273 mutex_unlock(&head->mutex);
4276 btrfs_pin_extent(root, head->node.bytenr,
4277 head->node.num_bytes, 1);
4278 btrfs_put_delayed_ref(&head->node);
4280 spin_lock(&delayed_refs->lock);
4283 spin_unlock(&delayed_refs->lock);
4288 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4290 struct btrfs_inode *btrfs_inode;
4291 struct list_head splice;
4293 INIT_LIST_HEAD(&splice);
4295 spin_lock(&root->delalloc_lock);
4296 list_splice_init(&root->delalloc_inodes, &splice);
4298 while (!list_empty(&splice)) {
4299 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4302 list_del_init(&btrfs_inode->delalloc_inodes);
4303 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4304 &btrfs_inode->runtime_flags);
4305 spin_unlock(&root->delalloc_lock);
4307 btrfs_invalidate_inodes(btrfs_inode->root);
4309 spin_lock(&root->delalloc_lock);
4312 spin_unlock(&root->delalloc_lock);
4315 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4317 struct btrfs_root *root;
4318 struct list_head splice;
4320 INIT_LIST_HEAD(&splice);
4322 spin_lock(&fs_info->delalloc_root_lock);
4323 list_splice_init(&fs_info->delalloc_roots, &splice);
4324 while (!list_empty(&splice)) {
4325 root = list_first_entry(&splice, struct btrfs_root,
4327 list_del_init(&root->delalloc_root);
4328 root = btrfs_grab_fs_root(root);
4330 spin_unlock(&fs_info->delalloc_root_lock);
4332 btrfs_destroy_delalloc_inodes(root);
4333 btrfs_put_fs_root(root);
4335 spin_lock(&fs_info->delalloc_root_lock);
4337 spin_unlock(&fs_info->delalloc_root_lock);
4340 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4341 struct extent_io_tree *dirty_pages,
4345 struct extent_buffer *eb;
4350 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4355 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4356 while (start <= end) {
4357 eb = btrfs_find_tree_block(root->fs_info, start);
4358 start += root->nodesize;
4361 wait_on_extent_buffer_writeback(eb);
4363 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4365 clear_extent_buffer_dirty(eb);
4366 free_extent_buffer_stale(eb);
4373 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4374 struct extent_io_tree *pinned_extents)
4376 struct extent_io_tree *unpin;
4382 unpin = pinned_extents;
4385 ret = find_first_extent_bit(unpin, 0, &start, &end,
4386 EXTENT_DIRTY, NULL);
4390 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4391 btrfs_error_unpin_extent_range(root, start, end);
4396 if (unpin == &root->fs_info->freed_extents[0])
4397 unpin = &root->fs_info->freed_extents[1];
4399 unpin = &root->fs_info->freed_extents[0];
4407 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4408 struct btrfs_root *root)
4410 btrfs_destroy_delayed_refs(cur_trans, root);
4412 cur_trans->state = TRANS_STATE_COMMIT_START;
4413 wake_up(&root->fs_info->transaction_blocked_wait);
4415 cur_trans->state = TRANS_STATE_UNBLOCKED;
4416 wake_up(&root->fs_info->transaction_wait);
4418 btrfs_destroy_delayed_inodes(root);
4419 btrfs_assert_delayed_root_empty(root);
4421 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4423 btrfs_destroy_pinned_extent(root,
4424 root->fs_info->pinned_extents);
4426 cur_trans->state =TRANS_STATE_COMPLETED;
4427 wake_up(&cur_trans->commit_wait);
4430 memset(cur_trans, 0, sizeof(*cur_trans));
4431 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4435 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4437 struct btrfs_transaction *t;
4439 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4441 spin_lock(&root->fs_info->trans_lock);
4442 while (!list_empty(&root->fs_info->trans_list)) {
4443 t = list_first_entry(&root->fs_info->trans_list,
4444 struct btrfs_transaction, list);
4445 if (t->state >= TRANS_STATE_COMMIT_START) {
4446 atomic_inc(&t->use_count);
4447 spin_unlock(&root->fs_info->trans_lock);
4448 btrfs_wait_for_commit(root, t->transid);
4449 btrfs_put_transaction(t);
4450 spin_lock(&root->fs_info->trans_lock);
4453 if (t == root->fs_info->running_transaction) {
4454 t->state = TRANS_STATE_COMMIT_DOING;
4455 spin_unlock(&root->fs_info->trans_lock);
4457 * We wait for 0 num_writers since we don't hold a trans
4458 * handle open currently for this transaction.
4460 wait_event(t->writer_wait,
4461 atomic_read(&t->num_writers) == 0);
4463 spin_unlock(&root->fs_info->trans_lock);
4465 btrfs_cleanup_one_transaction(t, root);
4467 spin_lock(&root->fs_info->trans_lock);
4468 if (t == root->fs_info->running_transaction)
4469 root->fs_info->running_transaction = NULL;
4470 list_del_init(&t->list);
4471 spin_unlock(&root->fs_info->trans_lock);
4473 btrfs_put_transaction(t);
4474 trace_btrfs_transaction_commit(root);
4475 spin_lock(&root->fs_info->trans_lock);
4477 spin_unlock(&root->fs_info->trans_lock);
4478 btrfs_destroy_all_ordered_extents(root->fs_info);
4479 btrfs_destroy_delayed_inodes(root);
4480 btrfs_assert_delayed_root_empty(root);
4481 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4482 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4483 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4488 static const struct extent_io_ops btree_extent_io_ops = {
4489 .readpage_end_io_hook = btree_readpage_end_io_hook,
4490 .readpage_io_failed_hook = btree_io_failed_hook,
4491 .submit_bio_hook = btree_submit_bio_hook,
4492 /* note we're sharing with inode.c for the merge bio hook */
4493 .merge_bio_hook = btrfs_merge_bio_hook,