2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/slab.h>
30 #include <linux/migrate.h>
31 #include <linux/ratelimit.h>
32 #include <linux/uuid.h>
33 #include <linux/semaphore.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "print-tree.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
54 #include <asm/cpufeature.h>
57 static const struct extent_io_ops btree_extent_io_ops;
58 static void end_workqueue_fn(struct btrfs_work *work);
59 static void free_fs_root(struct btrfs_root *root);
60 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
62 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
63 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
64 struct btrfs_root *root);
65 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
66 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
67 struct extent_io_tree *dirty_pages,
69 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
70 struct extent_io_tree *pinned_extents);
71 static int btrfs_cleanup_transaction(struct btrfs_root *root);
72 static void btrfs_error_commit_super(struct btrfs_root *root);
75 * btrfs_end_io_wq structs are used to do processing in task context when an IO
76 * is complete. This is used during reads to verify checksums, and it is used
77 * by writes to insert metadata for new file extents after IO is complete.
79 struct btrfs_end_io_wq {
83 struct btrfs_fs_info *info;
85 enum btrfs_wq_endio_type metadata;
86 struct list_head list;
87 struct btrfs_work work;
90 static struct kmem_cache *btrfs_end_io_wq_cache;
92 int __init btrfs_end_io_wq_init(void)
94 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
95 sizeof(struct btrfs_end_io_wq),
97 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
99 if (!btrfs_end_io_wq_cache)
104 void btrfs_end_io_wq_exit(void)
106 if (btrfs_end_io_wq_cache)
107 kmem_cache_destroy(btrfs_end_io_wq_cache);
111 * async submit bios are used to offload expensive checksumming
112 * onto the worker threads. They checksum file and metadata bios
113 * just before they are sent down the IO stack.
115 struct async_submit_bio {
118 struct list_head list;
119 extent_submit_bio_hook_t *submit_bio_start;
120 extent_submit_bio_hook_t *submit_bio_done;
123 unsigned long bio_flags;
125 * bio_offset is optional, can be used if the pages in the bio
126 * can't tell us where in the file the bio should go
129 struct btrfs_work work;
134 * Lockdep class keys for extent_buffer->lock's in this root. For a given
135 * eb, the lockdep key is determined by the btrfs_root it belongs to and
136 * the level the eb occupies in the tree.
138 * Different roots are used for different purposes and may nest inside each
139 * other and they require separate keysets. As lockdep keys should be
140 * static, assign keysets according to the purpose of the root as indicated
141 * by btrfs_root->objectid. This ensures that all special purpose roots
142 * have separate keysets.
144 * Lock-nesting across peer nodes is always done with the immediate parent
145 * node locked thus preventing deadlock. As lockdep doesn't know this, use
146 * subclass to avoid triggering lockdep warning in such cases.
148 * The key is set by the readpage_end_io_hook after the buffer has passed
149 * csum validation but before the pages are unlocked. It is also set by
150 * btrfs_init_new_buffer on freshly allocated blocks.
152 * We also add a check to make sure the highest level of the tree is the
153 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
154 * needs update as well.
156 #ifdef CONFIG_DEBUG_LOCK_ALLOC
157 # if BTRFS_MAX_LEVEL != 8
161 static struct btrfs_lockdep_keyset {
162 u64 id; /* root objectid */
163 const char *name_stem; /* lock name stem */
164 char names[BTRFS_MAX_LEVEL + 1][20];
165 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
166 } btrfs_lockdep_keysets[] = {
167 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
168 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
169 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
170 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
171 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
172 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
173 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
174 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
175 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
176 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
177 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
178 { .id = 0, .name_stem = "tree" },
181 void __init btrfs_init_lockdep(void)
185 /* initialize lockdep class names */
186 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
187 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
189 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
190 snprintf(ks->names[j], sizeof(ks->names[j]),
191 "btrfs-%s-%02d", ks->name_stem, j);
195 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
198 struct btrfs_lockdep_keyset *ks;
200 BUG_ON(level >= ARRAY_SIZE(ks->keys));
202 /* find the matching keyset, id 0 is the default entry */
203 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
204 if (ks->id == objectid)
207 lockdep_set_class_and_name(&eb->lock,
208 &ks->keys[level], ks->names[level]);
214 * extents on the btree inode are pretty simple, there's one extent
215 * that covers the entire device
217 static struct extent_map *btree_get_extent(struct inode *inode,
218 struct page *page, size_t pg_offset, u64 start, u64 len,
221 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
222 struct extent_map *em;
225 read_lock(&em_tree->lock);
226 em = lookup_extent_mapping(em_tree, start, len);
229 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
230 read_unlock(&em_tree->lock);
233 read_unlock(&em_tree->lock);
235 em = alloc_extent_map();
237 em = ERR_PTR(-ENOMEM);
242 em->block_len = (u64)-1;
244 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
246 write_lock(&em_tree->lock);
247 ret = add_extent_mapping(em_tree, em, 0);
248 if (ret == -EEXIST) {
250 em = lookup_extent_mapping(em_tree, start, len);
257 write_unlock(&em_tree->lock);
263 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
265 return btrfs_crc32c(seed, data, len);
268 void btrfs_csum_final(u32 crc, char *result)
270 put_unaligned_le32(~crc, result);
274 * compute the csum for a btree block, and either verify it or write it
275 * into the csum field of the block.
277 static int csum_tree_block(struct btrfs_fs_info *fs_info,
278 struct extent_buffer *buf,
281 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
284 unsigned long cur_len;
285 unsigned long offset = BTRFS_CSUM_SIZE;
287 unsigned long map_start;
288 unsigned long map_len;
291 unsigned long inline_result;
293 len = buf->len - offset;
295 err = map_private_extent_buffer(buf, offset, 32,
296 &kaddr, &map_start, &map_len);
299 cur_len = min(len, map_len - (offset - map_start));
300 crc = btrfs_csum_data(kaddr + offset - map_start,
305 if (csum_size > sizeof(inline_result)) {
306 result = kzalloc(csum_size, GFP_NOFS);
310 result = (char *)&inline_result;
313 btrfs_csum_final(crc, result);
316 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
319 memcpy(&found, result, csum_size);
321 read_extent_buffer(buf, &val, 0, csum_size);
322 printk_ratelimited(KERN_WARNING
323 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
325 fs_info->sb->s_id, buf->start,
326 val, found, btrfs_header_level(buf));
327 if (result != (char *)&inline_result)
332 write_extent_buffer(buf, result, 0, csum_size);
334 if (result != (char *)&inline_result)
340 * we can't consider a given block up to date unless the transid of the
341 * block matches the transid in the parent node's pointer. This is how we
342 * detect blocks that either didn't get written at all or got written
343 * in the wrong place.
345 static int verify_parent_transid(struct extent_io_tree *io_tree,
346 struct extent_buffer *eb, u64 parent_transid,
349 struct extent_state *cached_state = NULL;
351 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
353 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
360 btrfs_tree_read_lock(eb);
361 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
364 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
366 if (extent_buffer_uptodate(eb) &&
367 btrfs_header_generation(eb) == parent_transid) {
371 printk_ratelimited(KERN_ERR
372 "BTRFS (device %s): parent transid verify failed on %llu wanted %llu found %llu\n",
373 eb->fs_info->sb->s_id, eb->start,
374 parent_transid, btrfs_header_generation(eb));
378 * Things reading via commit roots that don't have normal protection,
379 * like send, can have a really old block in cache that may point at a
380 * block that has been free'd and re-allocated. So don't clear uptodate
381 * if we find an eb that is under IO (dirty/writeback) because we could
382 * end up reading in the stale data and then writing it back out and
383 * making everybody very sad.
385 if (!extent_buffer_under_io(eb))
386 clear_extent_buffer_uptodate(eb);
388 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
389 &cached_state, GFP_NOFS);
391 btrfs_tree_read_unlock_blocking(eb);
396 * Return 0 if the superblock checksum type matches the checksum value of that
397 * algorithm. Pass the raw disk superblock data.
399 static int btrfs_check_super_csum(char *raw_disk_sb)
401 struct btrfs_super_block *disk_sb =
402 (struct btrfs_super_block *)raw_disk_sb;
403 u16 csum_type = btrfs_super_csum_type(disk_sb);
406 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
408 const int csum_size = sizeof(crc);
409 char result[csum_size];
412 * The super_block structure does not span the whole
413 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
414 * is filled with zeros and is included in the checkum.
416 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
417 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
418 btrfs_csum_final(crc, result);
420 if (memcmp(raw_disk_sb, result, csum_size))
424 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
425 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
434 * helper to read a given tree block, doing retries as required when
435 * the checksums don't match and we have alternate mirrors to try.
437 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
438 struct extent_buffer *eb,
439 u64 start, u64 parent_transid)
441 struct extent_io_tree *io_tree;
446 int failed_mirror = 0;
448 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
449 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
451 ret = read_extent_buffer_pages(io_tree, eb, start,
453 btree_get_extent, mirror_num);
455 if (!verify_parent_transid(io_tree, eb,
463 * This buffer's crc is fine, but its contents are corrupted, so
464 * there is no reason to read the other copies, they won't be
467 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
470 num_copies = btrfs_num_copies(root->fs_info,
475 if (!failed_mirror) {
477 failed_mirror = eb->read_mirror;
481 if (mirror_num == failed_mirror)
484 if (mirror_num > num_copies)
488 if (failed && !ret && failed_mirror)
489 repair_eb_io_failure(root, eb, failed_mirror);
495 * checksum a dirty tree block before IO. This has extra checks to make sure
496 * we only fill in the checksum field in the first page of a multi-page block
499 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
501 u64 start = page_offset(page);
503 struct extent_buffer *eb;
505 eb = (struct extent_buffer *)page->private;
506 if (page != eb->pages[0])
508 found_start = btrfs_header_bytenr(eb);
509 if (WARN_ON(found_start != start || !PageUptodate(page)))
511 csum_tree_block(fs_info, eb, 0);
515 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
516 struct extent_buffer *eb)
518 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
519 u8 fsid[BTRFS_UUID_SIZE];
522 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
524 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
528 fs_devices = fs_devices->seed;
533 #define CORRUPT(reason, eb, root, slot) \
534 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
535 "root=%llu, slot=%d", reason, \
536 btrfs_header_bytenr(eb), root->objectid, slot)
538 static noinline int check_leaf(struct btrfs_root *root,
539 struct extent_buffer *leaf)
541 struct btrfs_key key;
542 struct btrfs_key leaf_key;
543 u32 nritems = btrfs_header_nritems(leaf);
549 /* Check the 0 item */
550 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
551 BTRFS_LEAF_DATA_SIZE(root)) {
552 CORRUPT("invalid item offset size pair", leaf, root, 0);
557 * Check to make sure each items keys are in the correct order and their
558 * offsets make sense. We only have to loop through nritems-1 because
559 * we check the current slot against the next slot, which verifies the
560 * next slot's offset+size makes sense and that the current's slot
563 for (slot = 0; slot < nritems - 1; slot++) {
564 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
565 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
567 /* Make sure the keys are in the right order */
568 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
569 CORRUPT("bad key order", leaf, root, slot);
574 * Make sure the offset and ends are right, remember that the
575 * item data starts at the end of the leaf and grows towards the
578 if (btrfs_item_offset_nr(leaf, slot) !=
579 btrfs_item_end_nr(leaf, slot + 1)) {
580 CORRUPT("slot offset bad", leaf, root, slot);
585 * Check to make sure that we don't point outside of the leaf,
586 * just incase all the items are consistent to eachother, but
587 * all point outside of the leaf.
589 if (btrfs_item_end_nr(leaf, slot) >
590 BTRFS_LEAF_DATA_SIZE(root)) {
591 CORRUPT("slot end outside of leaf", leaf, root, slot);
599 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
600 u64 phy_offset, struct page *page,
601 u64 start, u64 end, int mirror)
605 struct extent_buffer *eb;
606 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
613 eb = (struct extent_buffer *)page->private;
615 /* the pending IO might have been the only thing that kept this buffer
616 * in memory. Make sure we have a ref for all this other checks
618 extent_buffer_get(eb);
620 reads_done = atomic_dec_and_test(&eb->io_pages);
624 eb->read_mirror = mirror;
625 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
630 found_start = btrfs_header_bytenr(eb);
631 if (found_start != eb->start) {
632 printk_ratelimited(KERN_ERR "BTRFS (device %s): bad tree block start "
634 eb->fs_info->sb->s_id, found_start, eb->start);
638 if (check_tree_block_fsid(root->fs_info, eb)) {
639 printk_ratelimited(KERN_ERR "BTRFS (device %s): bad fsid on block %llu\n",
640 eb->fs_info->sb->s_id, eb->start);
644 found_level = btrfs_header_level(eb);
645 if (found_level >= BTRFS_MAX_LEVEL) {
646 btrfs_err(root->fs_info, "bad tree block level %d",
647 (int)btrfs_header_level(eb));
652 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
655 ret = csum_tree_block(root->fs_info, eb, 1);
662 * If this is a leaf block and it is corrupt, set the corrupt bit so
663 * that we don't try and read the other copies of this block, just
666 if (found_level == 0 && check_leaf(root, eb)) {
667 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
672 set_extent_buffer_uptodate(eb);
675 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
676 btree_readahead_hook(root, eb, eb->start, ret);
680 * our io error hook is going to dec the io pages
681 * again, we have to make sure it has something
684 atomic_inc(&eb->io_pages);
685 clear_extent_buffer_uptodate(eb);
687 free_extent_buffer(eb);
692 static int btree_io_failed_hook(struct page *page, int failed_mirror)
694 struct extent_buffer *eb;
695 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
697 eb = (struct extent_buffer *)page->private;
698 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
699 eb->read_mirror = failed_mirror;
700 atomic_dec(&eb->io_pages);
701 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
702 btree_readahead_hook(root, eb, eb->start, -EIO);
703 return -EIO; /* we fixed nothing */
706 static void end_workqueue_bio(struct bio *bio)
708 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
709 struct btrfs_fs_info *fs_info;
710 struct btrfs_workqueue *wq;
711 btrfs_work_func_t func;
713 fs_info = end_io_wq->info;
714 end_io_wq->error = bio->bi_error;
716 if (bio->bi_rw & REQ_WRITE) {
717 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
718 wq = fs_info->endio_meta_write_workers;
719 func = btrfs_endio_meta_write_helper;
720 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
721 wq = fs_info->endio_freespace_worker;
722 func = btrfs_freespace_write_helper;
723 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
724 wq = fs_info->endio_raid56_workers;
725 func = btrfs_endio_raid56_helper;
727 wq = fs_info->endio_write_workers;
728 func = btrfs_endio_write_helper;
731 if (unlikely(end_io_wq->metadata ==
732 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
733 wq = fs_info->endio_repair_workers;
734 func = btrfs_endio_repair_helper;
735 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
736 wq = fs_info->endio_raid56_workers;
737 func = btrfs_endio_raid56_helper;
738 } else if (end_io_wq->metadata) {
739 wq = fs_info->endio_meta_workers;
740 func = btrfs_endio_meta_helper;
742 wq = fs_info->endio_workers;
743 func = btrfs_endio_helper;
747 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
748 btrfs_queue_work(wq, &end_io_wq->work);
751 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
752 enum btrfs_wq_endio_type metadata)
754 struct btrfs_end_io_wq *end_io_wq;
756 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
760 end_io_wq->private = bio->bi_private;
761 end_io_wq->end_io = bio->bi_end_io;
762 end_io_wq->info = info;
763 end_io_wq->error = 0;
764 end_io_wq->bio = bio;
765 end_io_wq->metadata = metadata;
767 bio->bi_private = end_io_wq;
768 bio->bi_end_io = end_workqueue_bio;
772 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
774 unsigned long limit = min_t(unsigned long,
775 info->thread_pool_size,
776 info->fs_devices->open_devices);
780 static void run_one_async_start(struct btrfs_work *work)
782 struct async_submit_bio *async;
785 async = container_of(work, struct async_submit_bio, work);
786 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
787 async->mirror_num, async->bio_flags,
793 static void run_one_async_done(struct btrfs_work *work)
795 struct btrfs_fs_info *fs_info;
796 struct async_submit_bio *async;
799 async = container_of(work, struct async_submit_bio, work);
800 fs_info = BTRFS_I(async->inode)->root->fs_info;
802 limit = btrfs_async_submit_limit(fs_info);
803 limit = limit * 2 / 3;
805 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
806 waitqueue_active(&fs_info->async_submit_wait))
807 wake_up(&fs_info->async_submit_wait);
809 /* If an error occured we just want to clean up the bio and move on */
811 async->bio->bi_error = async->error;
812 bio_endio(async->bio);
816 async->submit_bio_done(async->inode, async->rw, async->bio,
817 async->mirror_num, async->bio_flags,
821 static void run_one_async_free(struct btrfs_work *work)
823 struct async_submit_bio *async;
825 async = container_of(work, struct async_submit_bio, work);
829 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
830 int rw, struct bio *bio, int mirror_num,
831 unsigned long bio_flags,
833 extent_submit_bio_hook_t *submit_bio_start,
834 extent_submit_bio_hook_t *submit_bio_done)
836 struct async_submit_bio *async;
838 async = kmalloc(sizeof(*async), GFP_NOFS);
842 async->inode = inode;
845 async->mirror_num = mirror_num;
846 async->submit_bio_start = submit_bio_start;
847 async->submit_bio_done = submit_bio_done;
849 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
850 run_one_async_done, run_one_async_free);
852 async->bio_flags = bio_flags;
853 async->bio_offset = bio_offset;
857 atomic_inc(&fs_info->nr_async_submits);
860 btrfs_set_work_high_priority(&async->work);
862 btrfs_queue_work(fs_info->workers, &async->work);
864 while (atomic_read(&fs_info->async_submit_draining) &&
865 atomic_read(&fs_info->nr_async_submits)) {
866 wait_event(fs_info->async_submit_wait,
867 (atomic_read(&fs_info->nr_async_submits) == 0));
873 static int btree_csum_one_bio(struct bio *bio)
875 struct bio_vec *bvec;
876 struct btrfs_root *root;
879 bio_for_each_segment_all(bvec, bio, i) {
880 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
881 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
889 static int __btree_submit_bio_start(struct inode *inode, int rw,
890 struct bio *bio, int mirror_num,
891 unsigned long bio_flags,
895 * when we're called for a write, we're already in the async
896 * submission context. Just jump into btrfs_map_bio
898 return btree_csum_one_bio(bio);
901 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
902 int mirror_num, unsigned long bio_flags,
908 * when we're called for a write, we're already in the async
909 * submission context. Just jump into btrfs_map_bio
911 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
919 static int check_async_write(struct inode *inode, unsigned long bio_flags)
921 if (bio_flags & EXTENT_BIO_TREE_LOG)
930 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
931 int mirror_num, unsigned long bio_flags,
934 int async = check_async_write(inode, bio_flags);
937 if (!(rw & REQ_WRITE)) {
939 * called for a read, do the setup so that checksum validation
940 * can happen in the async kernel threads
942 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
943 bio, BTRFS_WQ_ENDIO_METADATA);
946 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
949 ret = btree_csum_one_bio(bio);
952 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
956 * kthread helpers are used to submit writes so that
957 * checksumming can happen in parallel across all CPUs
959 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
960 inode, rw, bio, mirror_num, 0,
962 __btree_submit_bio_start,
963 __btree_submit_bio_done);
976 #ifdef CONFIG_MIGRATION
977 static int btree_migratepage(struct address_space *mapping,
978 struct page *newpage, struct page *page,
979 enum migrate_mode mode)
982 * we can't safely write a btree page from here,
983 * we haven't done the locking hook
988 * Buffers may be managed in a filesystem specific way.
989 * We must have no buffers or drop them.
991 if (page_has_private(page) &&
992 !try_to_release_page(page, GFP_KERNEL))
994 return migrate_page(mapping, newpage, page, mode);
999 static int btree_writepages(struct address_space *mapping,
1000 struct writeback_control *wbc)
1002 struct btrfs_fs_info *fs_info;
1005 if (wbc->sync_mode == WB_SYNC_NONE) {
1007 if (wbc->for_kupdate)
1010 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1011 /* this is a bit racy, but that's ok */
1012 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1013 BTRFS_DIRTY_METADATA_THRESH);
1017 return btree_write_cache_pages(mapping, wbc);
1020 static int btree_readpage(struct file *file, struct page *page)
1022 struct extent_io_tree *tree;
1023 tree = &BTRFS_I(page->mapping->host)->io_tree;
1024 return extent_read_full_page(tree, page, btree_get_extent, 0);
1027 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1029 if (PageWriteback(page) || PageDirty(page))
1032 return try_release_extent_buffer(page);
1035 static void btree_invalidatepage(struct page *page, unsigned int offset,
1036 unsigned int length)
1038 struct extent_io_tree *tree;
1039 tree = &BTRFS_I(page->mapping->host)->io_tree;
1040 extent_invalidatepage(tree, page, offset);
1041 btree_releasepage(page, GFP_NOFS);
1042 if (PagePrivate(page)) {
1043 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1044 "page private not zero on page %llu",
1045 (unsigned long long)page_offset(page));
1046 ClearPagePrivate(page);
1047 set_page_private(page, 0);
1048 page_cache_release(page);
1052 static int btree_set_page_dirty(struct page *page)
1055 struct extent_buffer *eb;
1057 BUG_ON(!PagePrivate(page));
1058 eb = (struct extent_buffer *)page->private;
1060 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1061 BUG_ON(!atomic_read(&eb->refs));
1062 btrfs_assert_tree_locked(eb);
1064 return __set_page_dirty_nobuffers(page);
1067 static const struct address_space_operations btree_aops = {
1068 .readpage = btree_readpage,
1069 .writepages = btree_writepages,
1070 .releasepage = btree_releasepage,
1071 .invalidatepage = btree_invalidatepage,
1072 #ifdef CONFIG_MIGRATION
1073 .migratepage = btree_migratepage,
1075 .set_page_dirty = btree_set_page_dirty,
1078 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1080 struct extent_buffer *buf = NULL;
1081 struct inode *btree_inode = root->fs_info->btree_inode;
1083 buf = btrfs_find_create_tree_block(root, bytenr);
1086 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1087 buf, 0, WAIT_NONE, btree_get_extent, 0);
1088 free_extent_buffer(buf);
1091 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1092 int mirror_num, struct extent_buffer **eb)
1094 struct extent_buffer *buf = NULL;
1095 struct inode *btree_inode = root->fs_info->btree_inode;
1096 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1099 buf = btrfs_find_create_tree_block(root, bytenr);
1103 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1105 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1106 btree_get_extent, mirror_num);
1108 free_extent_buffer(buf);
1112 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1113 free_extent_buffer(buf);
1115 } else if (extent_buffer_uptodate(buf)) {
1118 free_extent_buffer(buf);
1123 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1126 return find_extent_buffer(fs_info, bytenr);
1129 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1132 if (btrfs_test_is_dummy_root(root))
1133 return alloc_test_extent_buffer(root->fs_info, bytenr);
1134 return alloc_extent_buffer(root->fs_info, bytenr);
1138 int btrfs_write_tree_block(struct extent_buffer *buf)
1140 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1141 buf->start + buf->len - 1);
1144 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1146 return filemap_fdatawait_range(buf->pages[0]->mapping,
1147 buf->start, buf->start + buf->len - 1);
1150 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1153 struct extent_buffer *buf = NULL;
1156 buf = btrfs_find_create_tree_block(root, bytenr);
1158 return ERR_PTR(-ENOMEM);
1160 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1162 free_extent_buffer(buf);
1163 return ERR_PTR(ret);
1169 void clean_tree_block(struct btrfs_trans_handle *trans,
1170 struct btrfs_fs_info *fs_info,
1171 struct extent_buffer *buf)
1173 if (btrfs_header_generation(buf) ==
1174 fs_info->running_transaction->transid) {
1175 btrfs_assert_tree_locked(buf);
1177 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1178 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1180 fs_info->dirty_metadata_batch);
1181 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1182 btrfs_set_lock_blocking(buf);
1183 clear_extent_buffer_dirty(buf);
1188 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1190 struct btrfs_subvolume_writers *writers;
1193 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1195 return ERR_PTR(-ENOMEM);
1197 ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1200 return ERR_PTR(ret);
1203 init_waitqueue_head(&writers->wait);
1208 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1210 percpu_counter_destroy(&writers->counter);
1214 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1215 struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1219 root->commit_root = NULL;
1220 root->sectorsize = sectorsize;
1221 root->nodesize = nodesize;
1222 root->stripesize = stripesize;
1224 root->orphan_cleanup_state = 0;
1226 root->objectid = objectid;
1227 root->last_trans = 0;
1228 root->highest_objectid = 0;
1229 root->nr_delalloc_inodes = 0;
1230 root->nr_ordered_extents = 0;
1232 root->inode_tree = RB_ROOT;
1233 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1234 root->block_rsv = NULL;
1235 root->orphan_block_rsv = NULL;
1237 INIT_LIST_HEAD(&root->dirty_list);
1238 INIT_LIST_HEAD(&root->root_list);
1239 INIT_LIST_HEAD(&root->delalloc_inodes);
1240 INIT_LIST_HEAD(&root->delalloc_root);
1241 INIT_LIST_HEAD(&root->ordered_extents);
1242 INIT_LIST_HEAD(&root->ordered_root);
1243 INIT_LIST_HEAD(&root->logged_list[0]);
1244 INIT_LIST_HEAD(&root->logged_list[1]);
1245 spin_lock_init(&root->orphan_lock);
1246 spin_lock_init(&root->inode_lock);
1247 spin_lock_init(&root->delalloc_lock);
1248 spin_lock_init(&root->ordered_extent_lock);
1249 spin_lock_init(&root->accounting_lock);
1250 spin_lock_init(&root->log_extents_lock[0]);
1251 spin_lock_init(&root->log_extents_lock[1]);
1252 mutex_init(&root->objectid_mutex);
1253 mutex_init(&root->log_mutex);
1254 mutex_init(&root->ordered_extent_mutex);
1255 mutex_init(&root->delalloc_mutex);
1256 init_waitqueue_head(&root->log_writer_wait);
1257 init_waitqueue_head(&root->log_commit_wait[0]);
1258 init_waitqueue_head(&root->log_commit_wait[1]);
1259 INIT_LIST_HEAD(&root->log_ctxs[0]);
1260 INIT_LIST_HEAD(&root->log_ctxs[1]);
1261 atomic_set(&root->log_commit[0], 0);
1262 atomic_set(&root->log_commit[1], 0);
1263 atomic_set(&root->log_writers, 0);
1264 atomic_set(&root->log_batch, 0);
1265 atomic_set(&root->orphan_inodes, 0);
1266 atomic_set(&root->refs, 1);
1267 atomic_set(&root->will_be_snapshoted, 0);
1268 root->log_transid = 0;
1269 root->log_transid_committed = -1;
1270 root->last_log_commit = 0;
1272 extent_io_tree_init(&root->dirty_log_pages,
1273 fs_info->btree_inode->i_mapping);
1275 memset(&root->root_key, 0, sizeof(root->root_key));
1276 memset(&root->root_item, 0, sizeof(root->root_item));
1277 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1279 root->defrag_trans_start = fs_info->generation;
1281 root->defrag_trans_start = 0;
1282 root->root_key.objectid = objectid;
1285 spin_lock_init(&root->root_item_lock);
1288 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1290 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1292 root->fs_info = fs_info;
1296 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1297 /* Should only be used by the testing infrastructure */
1298 struct btrfs_root *btrfs_alloc_dummy_root(void)
1300 struct btrfs_root *root;
1302 root = btrfs_alloc_root(NULL);
1304 return ERR_PTR(-ENOMEM);
1305 __setup_root(4096, 4096, 4096, root, NULL, 1);
1306 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1307 root->alloc_bytenr = 0;
1313 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1314 struct btrfs_fs_info *fs_info,
1317 struct extent_buffer *leaf;
1318 struct btrfs_root *tree_root = fs_info->tree_root;
1319 struct btrfs_root *root;
1320 struct btrfs_key key;
1324 root = btrfs_alloc_root(fs_info);
1326 return ERR_PTR(-ENOMEM);
1328 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1329 tree_root->stripesize, root, fs_info, objectid);
1330 root->root_key.objectid = objectid;
1331 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1332 root->root_key.offset = 0;
1334 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1336 ret = PTR_ERR(leaf);
1341 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1342 btrfs_set_header_bytenr(leaf, leaf->start);
1343 btrfs_set_header_generation(leaf, trans->transid);
1344 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1345 btrfs_set_header_owner(leaf, objectid);
1348 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1350 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1351 btrfs_header_chunk_tree_uuid(leaf),
1353 btrfs_mark_buffer_dirty(leaf);
1355 root->commit_root = btrfs_root_node(root);
1356 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1358 root->root_item.flags = 0;
1359 root->root_item.byte_limit = 0;
1360 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1361 btrfs_set_root_generation(&root->root_item, trans->transid);
1362 btrfs_set_root_level(&root->root_item, 0);
1363 btrfs_set_root_refs(&root->root_item, 1);
1364 btrfs_set_root_used(&root->root_item, leaf->len);
1365 btrfs_set_root_last_snapshot(&root->root_item, 0);
1366 btrfs_set_root_dirid(&root->root_item, 0);
1368 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1369 root->root_item.drop_level = 0;
1371 key.objectid = objectid;
1372 key.type = BTRFS_ROOT_ITEM_KEY;
1374 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1378 btrfs_tree_unlock(leaf);
1384 btrfs_tree_unlock(leaf);
1385 free_extent_buffer(root->commit_root);
1386 free_extent_buffer(leaf);
1390 return ERR_PTR(ret);
1393 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1394 struct btrfs_fs_info *fs_info)
1396 struct btrfs_root *root;
1397 struct btrfs_root *tree_root = fs_info->tree_root;
1398 struct extent_buffer *leaf;
1400 root = btrfs_alloc_root(fs_info);
1402 return ERR_PTR(-ENOMEM);
1404 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1405 tree_root->stripesize, root, fs_info,
1406 BTRFS_TREE_LOG_OBJECTID);
1408 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1409 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1410 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1413 * DON'T set REF_COWS for log trees
1415 * log trees do not get reference counted because they go away
1416 * before a real commit is actually done. They do store pointers
1417 * to file data extents, and those reference counts still get
1418 * updated (along with back refs to the log tree).
1421 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1425 return ERR_CAST(leaf);
1428 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1429 btrfs_set_header_bytenr(leaf, leaf->start);
1430 btrfs_set_header_generation(leaf, trans->transid);
1431 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1432 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1435 write_extent_buffer(root->node, root->fs_info->fsid,
1436 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1437 btrfs_mark_buffer_dirty(root->node);
1438 btrfs_tree_unlock(root->node);
1442 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1443 struct btrfs_fs_info *fs_info)
1445 struct btrfs_root *log_root;
1447 log_root = alloc_log_tree(trans, fs_info);
1448 if (IS_ERR(log_root))
1449 return PTR_ERR(log_root);
1450 WARN_ON(fs_info->log_root_tree);
1451 fs_info->log_root_tree = log_root;
1455 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1456 struct btrfs_root *root)
1458 struct btrfs_root *log_root;
1459 struct btrfs_inode_item *inode_item;
1461 log_root = alloc_log_tree(trans, root->fs_info);
1462 if (IS_ERR(log_root))
1463 return PTR_ERR(log_root);
1465 log_root->last_trans = trans->transid;
1466 log_root->root_key.offset = root->root_key.objectid;
1468 inode_item = &log_root->root_item.inode;
1469 btrfs_set_stack_inode_generation(inode_item, 1);
1470 btrfs_set_stack_inode_size(inode_item, 3);
1471 btrfs_set_stack_inode_nlink(inode_item, 1);
1472 btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1473 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1475 btrfs_set_root_node(&log_root->root_item, log_root->node);
1477 WARN_ON(root->log_root);
1478 root->log_root = log_root;
1479 root->log_transid = 0;
1480 root->log_transid_committed = -1;
1481 root->last_log_commit = 0;
1485 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1486 struct btrfs_key *key)
1488 struct btrfs_root *root;
1489 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1490 struct btrfs_path *path;
1494 path = btrfs_alloc_path();
1496 return ERR_PTR(-ENOMEM);
1498 root = btrfs_alloc_root(fs_info);
1504 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1505 tree_root->stripesize, root, fs_info, key->objectid);
1507 ret = btrfs_find_root(tree_root, key, path,
1508 &root->root_item, &root->root_key);
1515 generation = btrfs_root_generation(&root->root_item);
1516 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1518 if (IS_ERR(root->node)) {
1519 ret = PTR_ERR(root->node);
1521 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1523 free_extent_buffer(root->node);
1526 root->commit_root = btrfs_root_node(root);
1528 btrfs_free_path(path);
1534 root = ERR_PTR(ret);
1538 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1539 struct btrfs_key *location)
1541 struct btrfs_root *root;
1543 root = btrfs_read_tree_root(tree_root, location);
1547 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1548 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1549 btrfs_check_and_init_root_item(&root->root_item);
1555 int btrfs_init_fs_root(struct btrfs_root *root)
1558 struct btrfs_subvolume_writers *writers;
1560 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1561 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1563 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1568 writers = btrfs_alloc_subvolume_writers();
1569 if (IS_ERR(writers)) {
1570 ret = PTR_ERR(writers);
1573 root->subv_writers = writers;
1575 btrfs_init_free_ino_ctl(root);
1576 spin_lock_init(&root->ino_cache_lock);
1577 init_waitqueue_head(&root->ino_cache_wait);
1579 ret = get_anon_bdev(&root->anon_dev);
1585 btrfs_free_subvolume_writers(root->subv_writers);
1587 kfree(root->free_ino_ctl);
1588 kfree(root->free_ino_pinned);
1592 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1595 struct btrfs_root *root;
1597 spin_lock(&fs_info->fs_roots_radix_lock);
1598 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1599 (unsigned long)root_id);
1600 spin_unlock(&fs_info->fs_roots_radix_lock);
1604 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1605 struct btrfs_root *root)
1609 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1613 spin_lock(&fs_info->fs_roots_radix_lock);
1614 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1615 (unsigned long)root->root_key.objectid,
1618 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1619 spin_unlock(&fs_info->fs_roots_radix_lock);
1620 radix_tree_preload_end();
1625 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1626 struct btrfs_key *location,
1629 struct btrfs_root *root;
1630 struct btrfs_path *path;
1631 struct btrfs_key key;
1634 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1635 return fs_info->tree_root;
1636 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1637 return fs_info->extent_root;
1638 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1639 return fs_info->chunk_root;
1640 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1641 return fs_info->dev_root;
1642 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1643 return fs_info->csum_root;
1644 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1645 return fs_info->quota_root ? fs_info->quota_root :
1647 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1648 return fs_info->uuid_root ? fs_info->uuid_root :
1651 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1653 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1654 return ERR_PTR(-ENOENT);
1658 root = btrfs_read_fs_root(fs_info->tree_root, location);
1662 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1667 ret = btrfs_init_fs_root(root);
1671 path = btrfs_alloc_path();
1676 key.objectid = BTRFS_ORPHAN_OBJECTID;
1677 key.type = BTRFS_ORPHAN_ITEM_KEY;
1678 key.offset = location->objectid;
1680 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1681 btrfs_free_path(path);
1685 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1687 ret = btrfs_insert_fs_root(fs_info, root);
1689 if (ret == -EEXIST) {
1698 return ERR_PTR(ret);
1701 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1703 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1705 struct btrfs_device *device;
1706 struct backing_dev_info *bdi;
1709 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1712 bdi = blk_get_backing_dev_info(device->bdev);
1713 if (bdi_congested(bdi, bdi_bits)) {
1722 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1726 err = bdi_setup_and_register(bdi, "btrfs");
1730 bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE;
1731 bdi->congested_fn = btrfs_congested_fn;
1732 bdi->congested_data = info;
1737 * called by the kthread helper functions to finally call the bio end_io
1738 * functions. This is where read checksum verification actually happens
1740 static void end_workqueue_fn(struct btrfs_work *work)
1743 struct btrfs_end_io_wq *end_io_wq;
1745 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1746 bio = end_io_wq->bio;
1748 bio->bi_error = end_io_wq->error;
1749 bio->bi_private = end_io_wq->private;
1750 bio->bi_end_io = end_io_wq->end_io;
1751 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1755 static int cleaner_kthread(void *arg)
1757 struct btrfs_root *root = arg;
1759 struct btrfs_trans_handle *trans;
1764 /* Make the cleaner go to sleep early. */
1765 if (btrfs_need_cleaner_sleep(root))
1768 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1772 * Avoid the problem that we change the status of the fs
1773 * during the above check and trylock.
1775 if (btrfs_need_cleaner_sleep(root)) {
1776 mutex_unlock(&root->fs_info->cleaner_mutex);
1780 btrfs_run_delayed_iputs(root);
1781 again = btrfs_clean_one_deleted_snapshot(root);
1782 mutex_unlock(&root->fs_info->cleaner_mutex);
1785 * The defragger has dealt with the R/O remount and umount,
1786 * needn't do anything special here.
1788 btrfs_run_defrag_inodes(root->fs_info);
1791 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1792 * with relocation (btrfs_relocate_chunk) and relocation
1793 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1794 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1795 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1796 * unused block groups.
1798 btrfs_delete_unused_bgs(root->fs_info);
1800 if (!try_to_freeze() && !again) {
1801 set_current_state(TASK_INTERRUPTIBLE);
1802 if (!kthread_should_stop())
1804 __set_current_state(TASK_RUNNING);
1806 } while (!kthread_should_stop());
1809 * Transaction kthread is stopped before us and wakes us up.
1810 * However we might have started a new transaction and COWed some
1811 * tree blocks when deleting unused block groups for example. So
1812 * make sure we commit the transaction we started to have a clean
1813 * shutdown when evicting the btree inode - if it has dirty pages
1814 * when we do the final iput() on it, eviction will trigger a
1815 * writeback for it which will fail with null pointer dereferences
1816 * since work queues and other resources were already released and
1817 * destroyed by the time the iput/eviction/writeback is made.
1819 trans = btrfs_attach_transaction(root);
1820 if (IS_ERR(trans)) {
1821 if (PTR_ERR(trans) != -ENOENT)
1822 btrfs_err(root->fs_info,
1823 "cleaner transaction attach returned %ld",
1828 ret = btrfs_commit_transaction(trans, root);
1830 btrfs_err(root->fs_info,
1831 "cleaner open transaction commit returned %d",
1838 static int transaction_kthread(void *arg)
1840 struct btrfs_root *root = arg;
1841 struct btrfs_trans_handle *trans;
1842 struct btrfs_transaction *cur;
1845 unsigned long delay;
1849 cannot_commit = false;
1850 delay = HZ * root->fs_info->commit_interval;
1851 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1853 spin_lock(&root->fs_info->trans_lock);
1854 cur = root->fs_info->running_transaction;
1856 spin_unlock(&root->fs_info->trans_lock);
1860 now = get_seconds();
1861 if (cur->state < TRANS_STATE_BLOCKED &&
1862 (now < cur->start_time ||
1863 now - cur->start_time < root->fs_info->commit_interval)) {
1864 spin_unlock(&root->fs_info->trans_lock);
1868 transid = cur->transid;
1869 spin_unlock(&root->fs_info->trans_lock);
1871 /* If the file system is aborted, this will always fail. */
1872 trans = btrfs_attach_transaction(root);
1873 if (IS_ERR(trans)) {
1874 if (PTR_ERR(trans) != -ENOENT)
1875 cannot_commit = true;
1878 if (transid == trans->transid) {
1879 btrfs_commit_transaction(trans, root);
1881 btrfs_end_transaction(trans, root);
1884 wake_up_process(root->fs_info->cleaner_kthread);
1885 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1887 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1888 &root->fs_info->fs_state)))
1889 btrfs_cleanup_transaction(root);
1890 if (!try_to_freeze()) {
1891 set_current_state(TASK_INTERRUPTIBLE);
1892 if (!kthread_should_stop() &&
1893 (!btrfs_transaction_blocked(root->fs_info) ||
1895 schedule_timeout(delay);
1896 __set_current_state(TASK_RUNNING);
1898 } while (!kthread_should_stop());
1903 * this will find the highest generation in the array of
1904 * root backups. The index of the highest array is returned,
1905 * or -1 if we can't find anything.
1907 * We check to make sure the array is valid by comparing the
1908 * generation of the latest root in the array with the generation
1909 * in the super block. If they don't match we pitch it.
1911 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1914 int newest_index = -1;
1915 struct btrfs_root_backup *root_backup;
1918 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1919 root_backup = info->super_copy->super_roots + i;
1920 cur = btrfs_backup_tree_root_gen(root_backup);
1921 if (cur == newest_gen)
1925 /* check to see if we actually wrapped around */
1926 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1927 root_backup = info->super_copy->super_roots;
1928 cur = btrfs_backup_tree_root_gen(root_backup);
1929 if (cur == newest_gen)
1932 return newest_index;
1937 * find the oldest backup so we know where to store new entries
1938 * in the backup array. This will set the backup_root_index
1939 * field in the fs_info struct
1941 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1944 int newest_index = -1;
1946 newest_index = find_newest_super_backup(info, newest_gen);
1947 /* if there was garbage in there, just move along */
1948 if (newest_index == -1) {
1949 info->backup_root_index = 0;
1951 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1956 * copy all the root pointers into the super backup array.
1957 * this will bump the backup pointer by one when it is
1960 static void backup_super_roots(struct btrfs_fs_info *info)
1963 struct btrfs_root_backup *root_backup;
1966 next_backup = info->backup_root_index;
1967 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1968 BTRFS_NUM_BACKUP_ROOTS;
1971 * just overwrite the last backup if we're at the same generation
1972 * this happens only at umount
1974 root_backup = info->super_for_commit->super_roots + last_backup;
1975 if (btrfs_backup_tree_root_gen(root_backup) ==
1976 btrfs_header_generation(info->tree_root->node))
1977 next_backup = last_backup;
1979 root_backup = info->super_for_commit->super_roots + next_backup;
1982 * make sure all of our padding and empty slots get zero filled
1983 * regardless of which ones we use today
1985 memset(root_backup, 0, sizeof(*root_backup));
1987 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1989 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1990 btrfs_set_backup_tree_root_gen(root_backup,
1991 btrfs_header_generation(info->tree_root->node));
1993 btrfs_set_backup_tree_root_level(root_backup,
1994 btrfs_header_level(info->tree_root->node));
1996 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1997 btrfs_set_backup_chunk_root_gen(root_backup,
1998 btrfs_header_generation(info->chunk_root->node));
1999 btrfs_set_backup_chunk_root_level(root_backup,
2000 btrfs_header_level(info->chunk_root->node));
2002 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2003 btrfs_set_backup_extent_root_gen(root_backup,
2004 btrfs_header_generation(info->extent_root->node));
2005 btrfs_set_backup_extent_root_level(root_backup,
2006 btrfs_header_level(info->extent_root->node));
2009 * we might commit during log recovery, which happens before we set
2010 * the fs_root. Make sure it is valid before we fill it in.
2012 if (info->fs_root && info->fs_root->node) {
2013 btrfs_set_backup_fs_root(root_backup,
2014 info->fs_root->node->start);
2015 btrfs_set_backup_fs_root_gen(root_backup,
2016 btrfs_header_generation(info->fs_root->node));
2017 btrfs_set_backup_fs_root_level(root_backup,
2018 btrfs_header_level(info->fs_root->node));
2021 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2022 btrfs_set_backup_dev_root_gen(root_backup,
2023 btrfs_header_generation(info->dev_root->node));
2024 btrfs_set_backup_dev_root_level(root_backup,
2025 btrfs_header_level(info->dev_root->node));
2027 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2028 btrfs_set_backup_csum_root_gen(root_backup,
2029 btrfs_header_generation(info->csum_root->node));
2030 btrfs_set_backup_csum_root_level(root_backup,
2031 btrfs_header_level(info->csum_root->node));
2033 btrfs_set_backup_total_bytes(root_backup,
2034 btrfs_super_total_bytes(info->super_copy));
2035 btrfs_set_backup_bytes_used(root_backup,
2036 btrfs_super_bytes_used(info->super_copy));
2037 btrfs_set_backup_num_devices(root_backup,
2038 btrfs_super_num_devices(info->super_copy));
2041 * if we don't copy this out to the super_copy, it won't get remembered
2042 * for the next commit
2044 memcpy(&info->super_copy->super_roots,
2045 &info->super_for_commit->super_roots,
2046 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2050 * this copies info out of the root backup array and back into
2051 * the in-memory super block. It is meant to help iterate through
2052 * the array, so you send it the number of backups you've already
2053 * tried and the last backup index you used.
2055 * this returns -1 when it has tried all the backups
2057 static noinline int next_root_backup(struct btrfs_fs_info *info,
2058 struct btrfs_super_block *super,
2059 int *num_backups_tried, int *backup_index)
2061 struct btrfs_root_backup *root_backup;
2062 int newest = *backup_index;
2064 if (*num_backups_tried == 0) {
2065 u64 gen = btrfs_super_generation(super);
2067 newest = find_newest_super_backup(info, gen);
2071 *backup_index = newest;
2072 *num_backups_tried = 1;
2073 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2074 /* we've tried all the backups, all done */
2077 /* jump to the next oldest backup */
2078 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2079 BTRFS_NUM_BACKUP_ROOTS;
2080 *backup_index = newest;
2081 *num_backups_tried += 1;
2083 root_backup = super->super_roots + newest;
2085 btrfs_set_super_generation(super,
2086 btrfs_backup_tree_root_gen(root_backup));
2087 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2088 btrfs_set_super_root_level(super,
2089 btrfs_backup_tree_root_level(root_backup));
2090 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2093 * fixme: the total bytes and num_devices need to match or we should
2096 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2097 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2101 /* helper to cleanup workers */
2102 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2104 btrfs_destroy_workqueue(fs_info->fixup_workers);
2105 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2106 btrfs_destroy_workqueue(fs_info->workers);
2107 btrfs_destroy_workqueue(fs_info->endio_workers);
2108 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2109 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2110 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2111 btrfs_destroy_workqueue(fs_info->rmw_workers);
2112 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2113 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2114 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2115 btrfs_destroy_workqueue(fs_info->submit_workers);
2116 btrfs_destroy_workqueue(fs_info->delayed_workers);
2117 btrfs_destroy_workqueue(fs_info->caching_workers);
2118 btrfs_destroy_workqueue(fs_info->readahead_workers);
2119 btrfs_destroy_workqueue(fs_info->flush_workers);
2120 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2121 btrfs_destroy_workqueue(fs_info->extent_workers);
2124 static void free_root_extent_buffers(struct btrfs_root *root)
2127 free_extent_buffer(root->node);
2128 free_extent_buffer(root->commit_root);
2130 root->commit_root = NULL;
2134 /* helper to cleanup tree roots */
2135 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2137 free_root_extent_buffers(info->tree_root);
2139 free_root_extent_buffers(info->dev_root);
2140 free_root_extent_buffers(info->extent_root);
2141 free_root_extent_buffers(info->csum_root);
2142 free_root_extent_buffers(info->quota_root);
2143 free_root_extent_buffers(info->uuid_root);
2145 free_root_extent_buffers(info->chunk_root);
2148 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2151 struct btrfs_root *gang[8];
2154 while (!list_empty(&fs_info->dead_roots)) {
2155 gang[0] = list_entry(fs_info->dead_roots.next,
2156 struct btrfs_root, root_list);
2157 list_del(&gang[0]->root_list);
2159 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2160 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2162 free_extent_buffer(gang[0]->node);
2163 free_extent_buffer(gang[0]->commit_root);
2164 btrfs_put_fs_root(gang[0]);
2169 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2174 for (i = 0; i < ret; i++)
2175 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2178 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2179 btrfs_free_log_root_tree(NULL, fs_info);
2180 btrfs_destroy_pinned_extent(fs_info->tree_root,
2181 fs_info->pinned_extents);
2185 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2187 mutex_init(&fs_info->scrub_lock);
2188 atomic_set(&fs_info->scrubs_running, 0);
2189 atomic_set(&fs_info->scrub_pause_req, 0);
2190 atomic_set(&fs_info->scrubs_paused, 0);
2191 atomic_set(&fs_info->scrub_cancel_req, 0);
2192 init_waitqueue_head(&fs_info->scrub_pause_wait);
2193 fs_info->scrub_workers_refcnt = 0;
2196 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2198 spin_lock_init(&fs_info->balance_lock);
2199 mutex_init(&fs_info->balance_mutex);
2200 atomic_set(&fs_info->balance_running, 0);
2201 atomic_set(&fs_info->balance_pause_req, 0);
2202 atomic_set(&fs_info->balance_cancel_req, 0);
2203 fs_info->balance_ctl = NULL;
2204 init_waitqueue_head(&fs_info->balance_wait_q);
2207 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2208 struct btrfs_root *tree_root)
2210 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2211 set_nlink(fs_info->btree_inode, 1);
2213 * we set the i_size on the btree inode to the max possible int.
2214 * the real end of the address space is determined by all of
2215 * the devices in the system
2217 fs_info->btree_inode->i_size = OFFSET_MAX;
2218 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2220 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2221 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2222 fs_info->btree_inode->i_mapping);
2223 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2224 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2226 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2228 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2229 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2230 sizeof(struct btrfs_key));
2231 set_bit(BTRFS_INODE_DUMMY,
2232 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2233 btrfs_insert_inode_hash(fs_info->btree_inode);
2236 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2238 fs_info->dev_replace.lock_owner = 0;
2239 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2240 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2241 mutex_init(&fs_info->dev_replace.lock_management_lock);
2242 mutex_init(&fs_info->dev_replace.lock);
2243 init_waitqueue_head(&fs_info->replace_wait);
2246 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2248 spin_lock_init(&fs_info->qgroup_lock);
2249 mutex_init(&fs_info->qgroup_ioctl_lock);
2250 fs_info->qgroup_tree = RB_ROOT;
2251 fs_info->qgroup_op_tree = RB_ROOT;
2252 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2253 fs_info->qgroup_seq = 1;
2254 fs_info->quota_enabled = 0;
2255 fs_info->pending_quota_state = 0;
2256 fs_info->qgroup_ulist = NULL;
2257 mutex_init(&fs_info->qgroup_rescan_lock);
2260 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2261 struct btrfs_fs_devices *fs_devices)
2263 int max_active = fs_info->thread_pool_size;
2264 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2267 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2270 fs_info->delalloc_workers =
2271 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2273 fs_info->flush_workers =
2274 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2276 fs_info->caching_workers =
2277 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2280 * a higher idle thresh on the submit workers makes it much more
2281 * likely that bios will be send down in a sane order to the
2284 fs_info->submit_workers =
2285 btrfs_alloc_workqueue("submit", flags,
2286 min_t(u64, fs_devices->num_devices,
2289 fs_info->fixup_workers =
2290 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2293 * endios are largely parallel and should have a very
2296 fs_info->endio_workers =
2297 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2298 fs_info->endio_meta_workers =
2299 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2300 fs_info->endio_meta_write_workers =
2301 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2302 fs_info->endio_raid56_workers =
2303 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2304 fs_info->endio_repair_workers =
2305 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2306 fs_info->rmw_workers =
2307 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2308 fs_info->endio_write_workers =
2309 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2310 fs_info->endio_freespace_worker =
2311 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2312 fs_info->delayed_workers =
2313 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2314 fs_info->readahead_workers =
2315 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2316 fs_info->qgroup_rescan_workers =
2317 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2318 fs_info->extent_workers =
2319 btrfs_alloc_workqueue("extent-refs", flags,
2320 min_t(u64, fs_devices->num_devices,
2323 if (!(fs_info->workers && fs_info->delalloc_workers &&
2324 fs_info->submit_workers && fs_info->flush_workers &&
2325 fs_info->endio_workers && fs_info->endio_meta_workers &&
2326 fs_info->endio_meta_write_workers &&
2327 fs_info->endio_repair_workers &&
2328 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2329 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2330 fs_info->caching_workers && fs_info->readahead_workers &&
2331 fs_info->fixup_workers && fs_info->delayed_workers &&
2332 fs_info->extent_workers &&
2333 fs_info->qgroup_rescan_workers)) {
2340 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2341 struct btrfs_fs_devices *fs_devices)
2344 struct btrfs_root *tree_root = fs_info->tree_root;
2345 struct btrfs_root *log_tree_root;
2346 struct btrfs_super_block *disk_super = fs_info->super_copy;
2347 u64 bytenr = btrfs_super_log_root(disk_super);
2349 if (fs_devices->rw_devices == 0) {
2350 printk(KERN_WARNING "BTRFS: log replay required "
2355 log_tree_root = btrfs_alloc_root(fs_info);
2359 __setup_root(tree_root->nodesize, tree_root->sectorsize,
2360 tree_root->stripesize, log_tree_root, fs_info,
2361 BTRFS_TREE_LOG_OBJECTID);
2363 log_tree_root->node = read_tree_block(tree_root, bytenr,
2364 fs_info->generation + 1);
2365 if (IS_ERR(log_tree_root->node)) {
2366 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2367 ret = PTR_ERR(log_tree_root->node);
2368 kfree(log_tree_root);
2370 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2371 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2372 free_extent_buffer(log_tree_root->node);
2373 kfree(log_tree_root);
2376 /* returns with log_tree_root freed on success */
2377 ret = btrfs_recover_log_trees(log_tree_root);
2379 btrfs_error(tree_root->fs_info, ret,
2380 "Failed to recover log tree");
2381 free_extent_buffer(log_tree_root->node);
2382 kfree(log_tree_root);
2386 if (fs_info->sb->s_flags & MS_RDONLY) {
2387 ret = btrfs_commit_super(tree_root);
2395 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2396 struct btrfs_root *tree_root)
2398 struct btrfs_root *root;
2399 struct btrfs_key location;
2402 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2403 location.type = BTRFS_ROOT_ITEM_KEY;
2404 location.offset = 0;
2406 root = btrfs_read_tree_root(tree_root, &location);
2408 return PTR_ERR(root);
2409 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2410 fs_info->extent_root = root;
2412 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2413 root = btrfs_read_tree_root(tree_root, &location);
2415 return PTR_ERR(root);
2416 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2417 fs_info->dev_root = root;
2418 btrfs_init_devices_late(fs_info);
2420 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2421 root = btrfs_read_tree_root(tree_root, &location);
2423 return PTR_ERR(root);
2424 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2425 fs_info->csum_root = root;
2427 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2428 root = btrfs_read_tree_root(tree_root, &location);
2429 if (!IS_ERR(root)) {
2430 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2431 fs_info->quota_enabled = 1;
2432 fs_info->pending_quota_state = 1;
2433 fs_info->quota_root = root;
2436 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2437 root = btrfs_read_tree_root(tree_root, &location);
2439 ret = PTR_ERR(root);
2443 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2444 fs_info->uuid_root = root;
2450 int open_ctree(struct super_block *sb,
2451 struct btrfs_fs_devices *fs_devices,
2459 struct btrfs_key location;
2460 struct buffer_head *bh;
2461 struct btrfs_super_block *disk_super;
2462 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2463 struct btrfs_root *tree_root;
2464 struct btrfs_root *chunk_root;
2467 int num_backups_tried = 0;
2468 int backup_index = 0;
2471 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2472 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2473 if (!tree_root || !chunk_root) {
2478 ret = init_srcu_struct(&fs_info->subvol_srcu);
2484 ret = setup_bdi(fs_info, &fs_info->bdi);
2490 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2495 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2496 (1 + ilog2(nr_cpu_ids));
2498 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2501 goto fail_dirty_metadata_bytes;
2504 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2507 goto fail_delalloc_bytes;
2510 fs_info->btree_inode = new_inode(sb);
2511 if (!fs_info->btree_inode) {
2513 goto fail_bio_counter;
2516 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2518 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2519 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2520 INIT_LIST_HEAD(&fs_info->trans_list);
2521 INIT_LIST_HEAD(&fs_info->dead_roots);
2522 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2523 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2524 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2525 spin_lock_init(&fs_info->delalloc_root_lock);
2526 spin_lock_init(&fs_info->trans_lock);
2527 spin_lock_init(&fs_info->fs_roots_radix_lock);
2528 spin_lock_init(&fs_info->delayed_iput_lock);
2529 spin_lock_init(&fs_info->defrag_inodes_lock);
2530 spin_lock_init(&fs_info->free_chunk_lock);
2531 spin_lock_init(&fs_info->tree_mod_seq_lock);
2532 spin_lock_init(&fs_info->super_lock);
2533 spin_lock_init(&fs_info->qgroup_op_lock);
2534 spin_lock_init(&fs_info->buffer_lock);
2535 spin_lock_init(&fs_info->unused_bgs_lock);
2536 rwlock_init(&fs_info->tree_mod_log_lock);
2537 mutex_init(&fs_info->unused_bg_unpin_mutex);
2538 mutex_init(&fs_info->delete_unused_bgs_mutex);
2539 mutex_init(&fs_info->reloc_mutex);
2540 mutex_init(&fs_info->delalloc_root_mutex);
2541 seqlock_init(&fs_info->profiles_lock);
2542 init_rwsem(&fs_info->delayed_iput_sem);
2544 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2545 INIT_LIST_HEAD(&fs_info->space_info);
2546 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2547 INIT_LIST_HEAD(&fs_info->unused_bgs);
2548 btrfs_mapping_init(&fs_info->mapping_tree);
2549 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2550 BTRFS_BLOCK_RSV_GLOBAL);
2551 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2552 BTRFS_BLOCK_RSV_DELALLOC);
2553 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2554 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2555 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2556 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2557 BTRFS_BLOCK_RSV_DELOPS);
2558 atomic_set(&fs_info->nr_async_submits, 0);
2559 atomic_set(&fs_info->async_delalloc_pages, 0);
2560 atomic_set(&fs_info->async_submit_draining, 0);
2561 atomic_set(&fs_info->nr_async_bios, 0);
2562 atomic_set(&fs_info->defrag_running, 0);
2563 atomic_set(&fs_info->qgroup_op_seq, 0);
2564 atomic64_set(&fs_info->tree_mod_seq, 0);
2566 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2567 fs_info->metadata_ratio = 0;
2568 fs_info->defrag_inodes = RB_ROOT;
2569 fs_info->free_chunk_space = 0;
2570 fs_info->tree_mod_log = RB_ROOT;
2571 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2572 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2573 /* readahead state */
2574 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2575 spin_lock_init(&fs_info->reada_lock);
2577 fs_info->thread_pool_size = min_t(unsigned long,
2578 num_online_cpus() + 2, 8);
2580 INIT_LIST_HEAD(&fs_info->ordered_roots);
2581 spin_lock_init(&fs_info->ordered_root_lock);
2582 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2584 if (!fs_info->delayed_root) {
2588 btrfs_init_delayed_root(fs_info->delayed_root);
2590 btrfs_init_scrub(fs_info);
2591 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2592 fs_info->check_integrity_print_mask = 0;
2594 btrfs_init_balance(fs_info);
2595 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2597 sb->s_blocksize = 4096;
2598 sb->s_blocksize_bits = blksize_bits(4096);
2599 sb->s_bdi = &fs_info->bdi;
2601 btrfs_init_btree_inode(fs_info, tree_root);
2603 spin_lock_init(&fs_info->block_group_cache_lock);
2604 fs_info->block_group_cache_tree = RB_ROOT;
2605 fs_info->first_logical_byte = (u64)-1;
2607 extent_io_tree_init(&fs_info->freed_extents[0],
2608 fs_info->btree_inode->i_mapping);
2609 extent_io_tree_init(&fs_info->freed_extents[1],
2610 fs_info->btree_inode->i_mapping);
2611 fs_info->pinned_extents = &fs_info->freed_extents[0];
2612 fs_info->do_barriers = 1;
2615 mutex_init(&fs_info->ordered_operations_mutex);
2616 mutex_init(&fs_info->ordered_extent_flush_mutex);
2617 mutex_init(&fs_info->tree_log_mutex);
2618 mutex_init(&fs_info->chunk_mutex);
2619 mutex_init(&fs_info->transaction_kthread_mutex);
2620 mutex_init(&fs_info->cleaner_mutex);
2621 mutex_init(&fs_info->volume_mutex);
2622 mutex_init(&fs_info->ro_block_group_mutex);
2623 init_rwsem(&fs_info->commit_root_sem);
2624 init_rwsem(&fs_info->cleanup_work_sem);
2625 init_rwsem(&fs_info->subvol_sem);
2626 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2628 btrfs_init_dev_replace_locks(fs_info);
2629 btrfs_init_qgroup(fs_info);
2631 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2632 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2634 init_waitqueue_head(&fs_info->transaction_throttle);
2635 init_waitqueue_head(&fs_info->transaction_wait);
2636 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2637 init_waitqueue_head(&fs_info->async_submit_wait);
2639 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2641 ret = btrfs_alloc_stripe_hash_table(fs_info);
2647 __setup_root(4096, 4096, 4096, tree_root,
2648 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2650 invalidate_bdev(fs_devices->latest_bdev);
2653 * Read super block and check the signature bytes only
2655 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2662 * We want to check superblock checksum, the type is stored inside.
2663 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2665 if (btrfs_check_super_csum(bh->b_data)) {
2666 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2672 * super_copy is zeroed at allocation time and we never touch the
2673 * following bytes up to INFO_SIZE, the checksum is calculated from
2674 * the whole block of INFO_SIZE
2676 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2677 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2678 sizeof(*fs_info->super_for_commit));
2681 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2683 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2685 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2690 disk_super = fs_info->super_copy;
2691 if (!btrfs_super_root(disk_super))
2694 /* check FS state, whether FS is broken. */
2695 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2696 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2699 * run through our array of backup supers and setup
2700 * our ring pointer to the oldest one
2702 generation = btrfs_super_generation(disk_super);
2703 find_oldest_super_backup(fs_info, generation);
2706 * In the long term, we'll store the compression type in the super
2707 * block, and it'll be used for per file compression control.
2709 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2711 ret = btrfs_parse_options(tree_root, options);
2717 features = btrfs_super_incompat_flags(disk_super) &
2718 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2720 printk(KERN_ERR "BTRFS: couldn't mount because of "
2721 "unsupported optional features (%Lx).\n",
2728 * Leafsize and nodesize were always equal, this is only a sanity check.
2730 if (le32_to_cpu(disk_super->__unused_leafsize) !=
2731 btrfs_super_nodesize(disk_super)) {
2732 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2733 "blocksizes don't match. node %d leaf %d\n",
2734 btrfs_super_nodesize(disk_super),
2735 le32_to_cpu(disk_super->__unused_leafsize));
2739 if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2740 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2741 "blocksize (%d) was too large\n",
2742 btrfs_super_nodesize(disk_super));
2747 features = btrfs_super_incompat_flags(disk_super);
2748 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2749 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2750 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2752 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2753 printk(KERN_INFO "BTRFS: has skinny extents\n");
2756 * flag our filesystem as having big metadata blocks if
2757 * they are bigger than the page size
2759 if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2760 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2761 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2762 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2765 nodesize = btrfs_super_nodesize(disk_super);
2766 sectorsize = btrfs_super_sectorsize(disk_super);
2767 stripesize = btrfs_super_stripesize(disk_super);
2768 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2769 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2772 * mixed block groups end up with duplicate but slightly offset
2773 * extent buffers for the same range. It leads to corruptions
2775 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2776 (sectorsize != nodesize)) {
2777 printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2778 "are not allowed for mixed block groups on %s\n",
2784 * Needn't use the lock because there is no other task which will
2787 btrfs_set_super_incompat_flags(disk_super, features);
2789 features = btrfs_super_compat_ro_flags(disk_super) &
2790 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2791 if (!(sb->s_flags & MS_RDONLY) && features) {
2792 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2793 "unsupported option features (%Lx).\n",
2799 max_active = fs_info->thread_pool_size;
2801 ret = btrfs_init_workqueues(fs_info, fs_devices);
2804 goto fail_sb_buffer;
2807 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2808 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2809 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2811 tree_root->nodesize = nodesize;
2812 tree_root->sectorsize = sectorsize;
2813 tree_root->stripesize = stripesize;
2815 sb->s_blocksize = sectorsize;
2816 sb->s_blocksize_bits = blksize_bits(sectorsize);
2818 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2819 printk(KERN_ERR "BTRFS: valid FS not found on %s\n", sb->s_id);
2820 goto fail_sb_buffer;
2823 if (sectorsize != PAGE_SIZE) {
2824 printk(KERN_ERR "BTRFS: incompatible sector size (%lu) "
2825 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2826 goto fail_sb_buffer;
2829 mutex_lock(&fs_info->chunk_mutex);
2830 ret = btrfs_read_sys_array(tree_root);
2831 mutex_unlock(&fs_info->chunk_mutex);
2833 printk(KERN_ERR "BTRFS: failed to read the system "
2834 "array on %s\n", sb->s_id);
2835 goto fail_sb_buffer;
2838 generation = btrfs_super_chunk_root_generation(disk_super);
2840 __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2841 fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2843 chunk_root->node = read_tree_block(chunk_root,
2844 btrfs_super_chunk_root(disk_super),
2846 if (IS_ERR(chunk_root->node) ||
2847 !extent_buffer_uptodate(chunk_root->node)) {
2848 printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2850 chunk_root->node = NULL;
2851 goto fail_tree_roots;
2853 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2854 chunk_root->commit_root = btrfs_root_node(chunk_root);
2856 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2857 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2859 ret = btrfs_read_chunk_tree(chunk_root);
2861 printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2863 goto fail_tree_roots;
2867 * keep the device that is marked to be the target device for the
2868 * dev_replace procedure
2870 btrfs_close_extra_devices(fs_devices, 0);
2872 if (!fs_devices->latest_bdev) {
2873 printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2875 goto fail_tree_roots;
2879 generation = btrfs_super_generation(disk_super);
2881 tree_root->node = read_tree_block(tree_root,
2882 btrfs_super_root(disk_super),
2884 if (IS_ERR(tree_root->node) ||
2885 !extent_buffer_uptodate(tree_root->node)) {
2886 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2888 tree_root->node = NULL;
2889 goto recovery_tree_root;
2892 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2893 tree_root->commit_root = btrfs_root_node(tree_root);
2894 btrfs_set_root_refs(&tree_root->root_item, 1);
2896 ret = btrfs_read_roots(fs_info, tree_root);
2898 goto recovery_tree_root;
2900 fs_info->generation = generation;
2901 fs_info->last_trans_committed = generation;
2903 ret = btrfs_recover_balance(fs_info);
2905 printk(KERN_ERR "BTRFS: failed to recover balance\n");
2906 goto fail_block_groups;
2909 ret = btrfs_init_dev_stats(fs_info);
2911 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2913 goto fail_block_groups;
2916 ret = btrfs_init_dev_replace(fs_info);
2918 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2919 goto fail_block_groups;
2922 btrfs_close_extra_devices(fs_devices, 1);
2924 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2926 pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret);
2927 goto fail_block_groups;
2930 ret = btrfs_sysfs_add_device(fs_devices);
2932 pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret);
2933 goto fail_fsdev_sysfs;
2936 ret = btrfs_sysfs_add_one(fs_info);
2938 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2939 goto fail_fsdev_sysfs;
2942 ret = btrfs_init_space_info(fs_info);
2944 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2948 ret = btrfs_read_block_groups(fs_info->extent_root);
2950 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2953 fs_info->num_tolerated_disk_barrier_failures =
2954 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2955 if (fs_info->fs_devices->missing_devices >
2956 fs_info->num_tolerated_disk_barrier_failures &&
2957 !(sb->s_flags & MS_RDONLY)) {
2958 printk(KERN_WARNING "BTRFS: "
2959 "too many missing devices, writeable mount is not allowed\n");
2963 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2965 if (IS_ERR(fs_info->cleaner_kthread))
2968 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2970 "btrfs-transaction");
2971 if (IS_ERR(fs_info->transaction_kthread))
2974 if (!btrfs_test_opt(tree_root, SSD) &&
2975 !btrfs_test_opt(tree_root, NOSSD) &&
2976 !fs_info->fs_devices->rotating) {
2977 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2979 btrfs_set_opt(fs_info->mount_opt, SSD);
2983 * Mount does not set all options immediatelly, we can do it now and do
2984 * not have to wait for transaction commit
2986 btrfs_apply_pending_changes(fs_info);
2988 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2989 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2990 ret = btrfsic_mount(tree_root, fs_devices,
2991 btrfs_test_opt(tree_root,
2992 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2994 fs_info->check_integrity_print_mask);
2996 printk(KERN_WARNING "BTRFS: failed to initialize"
2997 " integrity check module %s\n", sb->s_id);
3000 ret = btrfs_read_qgroup_config(fs_info);
3002 goto fail_trans_kthread;
3004 /* do not make disk changes in broken FS */
3005 if (btrfs_super_log_root(disk_super) != 0) {
3006 ret = btrfs_replay_log(fs_info, fs_devices);
3013 ret = btrfs_find_orphan_roots(tree_root);
3017 if (!(sb->s_flags & MS_RDONLY)) {
3018 ret = btrfs_cleanup_fs_roots(fs_info);
3022 mutex_lock(&fs_info->cleaner_mutex);
3023 ret = btrfs_recover_relocation(tree_root);
3024 mutex_unlock(&fs_info->cleaner_mutex);
3027 "BTRFS: failed to recover relocation\n");
3033 location.objectid = BTRFS_FS_TREE_OBJECTID;
3034 location.type = BTRFS_ROOT_ITEM_KEY;
3035 location.offset = 0;
3037 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3038 if (IS_ERR(fs_info->fs_root)) {
3039 err = PTR_ERR(fs_info->fs_root);
3043 if (sb->s_flags & MS_RDONLY)
3046 down_read(&fs_info->cleanup_work_sem);
3047 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3048 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3049 up_read(&fs_info->cleanup_work_sem);
3050 close_ctree(tree_root);
3053 up_read(&fs_info->cleanup_work_sem);
3055 ret = btrfs_resume_balance_async(fs_info);
3057 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
3058 close_ctree(tree_root);
3062 ret = btrfs_resume_dev_replace_async(fs_info);
3064 pr_warn("BTRFS: failed to resume dev_replace\n");
3065 close_ctree(tree_root);
3069 btrfs_qgroup_rescan_resume(fs_info);
3071 if (!fs_info->uuid_root) {
3072 pr_info("BTRFS: creating UUID tree\n");
3073 ret = btrfs_create_uuid_tree(fs_info);
3075 pr_warn("BTRFS: failed to create the UUID tree %d\n",
3077 close_ctree(tree_root);
3080 } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3081 fs_info->generation !=
3082 btrfs_super_uuid_tree_generation(disk_super)) {
3083 pr_info("BTRFS: checking UUID tree\n");
3084 ret = btrfs_check_uuid_tree(fs_info);
3086 pr_warn("BTRFS: failed to check the UUID tree %d\n",
3088 close_ctree(tree_root);
3092 fs_info->update_uuid_tree_gen = 1;
3100 btrfs_free_qgroup_config(fs_info);
3102 kthread_stop(fs_info->transaction_kthread);
3103 btrfs_cleanup_transaction(fs_info->tree_root);
3104 btrfs_free_fs_roots(fs_info);
3106 kthread_stop(fs_info->cleaner_kthread);
3109 * make sure we're done with the btree inode before we stop our
3112 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3115 btrfs_sysfs_remove_one(fs_info);
3118 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3121 btrfs_put_block_group_cache(fs_info);
3122 btrfs_free_block_groups(fs_info);
3125 free_root_pointers(fs_info, 1);
3126 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3129 btrfs_stop_all_workers(fs_info);
3132 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3134 iput(fs_info->btree_inode);
3136 percpu_counter_destroy(&fs_info->bio_counter);
3137 fail_delalloc_bytes:
3138 percpu_counter_destroy(&fs_info->delalloc_bytes);
3139 fail_dirty_metadata_bytes:
3140 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3142 bdi_destroy(&fs_info->bdi);
3144 cleanup_srcu_struct(&fs_info->subvol_srcu);
3146 btrfs_free_stripe_hash_table(fs_info);
3147 btrfs_close_devices(fs_info->fs_devices);
3151 if (!btrfs_test_opt(tree_root, RECOVERY))
3152 goto fail_tree_roots;
3154 free_root_pointers(fs_info, 0);
3156 /* don't use the log in recovery mode, it won't be valid */
3157 btrfs_set_super_log_root(disk_super, 0);
3159 /* we can't trust the free space cache either */
3160 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3162 ret = next_root_backup(fs_info, fs_info->super_copy,
3163 &num_backups_tried, &backup_index);
3165 goto fail_block_groups;
3166 goto retry_root_backup;
3169 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3172 set_buffer_uptodate(bh);
3174 struct btrfs_device *device = (struct btrfs_device *)
3177 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3178 "I/O error on %s\n",
3179 rcu_str_deref(device->name));
3180 /* note, we dont' set_buffer_write_io_error because we have
3181 * our own ways of dealing with the IO errors
3183 clear_buffer_uptodate(bh);
3184 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3190 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3192 struct buffer_head *bh;
3193 struct buffer_head *latest = NULL;
3194 struct btrfs_super_block *super;
3199 /* we would like to check all the supers, but that would make
3200 * a btrfs mount succeed after a mkfs from a different FS.
3201 * So, we need to add a special mount option to scan for
3202 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3204 for (i = 0; i < 1; i++) {
3205 bytenr = btrfs_sb_offset(i);
3206 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3207 i_size_read(bdev->bd_inode))
3209 bh = __bread(bdev, bytenr / 4096,
3210 BTRFS_SUPER_INFO_SIZE);
3214 super = (struct btrfs_super_block *)bh->b_data;
3215 if (btrfs_super_bytenr(super) != bytenr ||
3216 btrfs_super_magic(super) != BTRFS_MAGIC) {
3221 if (!latest || btrfs_super_generation(super) > transid) {
3224 transid = btrfs_super_generation(super);
3233 * this should be called twice, once with wait == 0 and
3234 * once with wait == 1. When wait == 0 is done, all the buffer heads
3235 * we write are pinned.
3237 * They are released when wait == 1 is done.
3238 * max_mirrors must be the same for both runs, and it indicates how
3239 * many supers on this one device should be written.
3241 * max_mirrors == 0 means to write them all.
3243 static int write_dev_supers(struct btrfs_device *device,
3244 struct btrfs_super_block *sb,
3245 int do_barriers, int wait, int max_mirrors)
3247 struct buffer_head *bh;
3254 if (max_mirrors == 0)
3255 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3257 for (i = 0; i < max_mirrors; i++) {
3258 bytenr = btrfs_sb_offset(i);
3259 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3260 device->commit_total_bytes)
3264 bh = __find_get_block(device->bdev, bytenr / 4096,
3265 BTRFS_SUPER_INFO_SIZE);
3271 if (!buffer_uptodate(bh))
3274 /* drop our reference */
3277 /* drop the reference from the wait == 0 run */
3281 btrfs_set_super_bytenr(sb, bytenr);
3284 crc = btrfs_csum_data((char *)sb +
3285 BTRFS_CSUM_SIZE, crc,
3286 BTRFS_SUPER_INFO_SIZE -
3288 btrfs_csum_final(crc, sb->csum);
3291 * one reference for us, and we leave it for the
3294 bh = __getblk(device->bdev, bytenr / 4096,
3295 BTRFS_SUPER_INFO_SIZE);
3297 printk(KERN_ERR "BTRFS: couldn't get super "
3298 "buffer head for bytenr %Lu\n", bytenr);
3303 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3305 /* one reference for submit_bh */
3308 set_buffer_uptodate(bh);
3310 bh->b_end_io = btrfs_end_buffer_write_sync;
3311 bh->b_private = device;
3315 * we fua the first super. The others we allow
3319 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3321 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3325 return errors < i ? 0 : -1;
3329 * endio for the write_dev_flush, this will wake anyone waiting
3330 * for the barrier when it is done
3332 static void btrfs_end_empty_barrier(struct bio *bio)
3334 if (bio->bi_private)
3335 complete(bio->bi_private);
3340 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3341 * sent down. With wait == 1, it waits for the previous flush.
3343 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3346 static int write_dev_flush(struct btrfs_device *device, int wait)
3351 if (device->nobarriers)
3355 bio = device->flush_bio;
3359 wait_for_completion(&device->flush_wait);
3361 if (bio->bi_error) {
3362 ret = bio->bi_error;
3363 btrfs_dev_stat_inc_and_print(device,
3364 BTRFS_DEV_STAT_FLUSH_ERRS);
3367 /* drop the reference from the wait == 0 run */
3369 device->flush_bio = NULL;
3375 * one reference for us, and we leave it for the
3378 device->flush_bio = NULL;
3379 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3383 bio->bi_end_io = btrfs_end_empty_barrier;
3384 bio->bi_bdev = device->bdev;
3385 init_completion(&device->flush_wait);
3386 bio->bi_private = &device->flush_wait;
3387 device->flush_bio = bio;
3390 btrfsic_submit_bio(WRITE_FLUSH, bio);
3396 * send an empty flush down to each device in parallel,
3397 * then wait for them
3399 static int barrier_all_devices(struct btrfs_fs_info *info)
3401 struct list_head *head;
3402 struct btrfs_device *dev;
3403 int errors_send = 0;
3404 int errors_wait = 0;
3407 /* send down all the barriers */
3408 head = &info->fs_devices->devices;
3409 list_for_each_entry_rcu(dev, head, dev_list) {
3416 if (!dev->in_fs_metadata || !dev->writeable)
3419 ret = write_dev_flush(dev, 0);
3424 /* wait for all the barriers */
3425 list_for_each_entry_rcu(dev, head, dev_list) {
3432 if (!dev->in_fs_metadata || !dev->writeable)
3435 ret = write_dev_flush(dev, 1);
3439 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3440 errors_wait > info->num_tolerated_disk_barrier_failures)
3445 int btrfs_calc_num_tolerated_disk_barrier_failures(
3446 struct btrfs_fs_info *fs_info)
3448 struct btrfs_ioctl_space_info space;
3449 struct btrfs_space_info *sinfo;
3450 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3451 BTRFS_BLOCK_GROUP_SYSTEM,
3452 BTRFS_BLOCK_GROUP_METADATA,
3453 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3457 int num_tolerated_disk_barrier_failures =
3458 (int)fs_info->fs_devices->num_devices;
3460 for (i = 0; i < num_types; i++) {
3461 struct btrfs_space_info *tmp;
3465 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3466 if (tmp->flags == types[i]) {
3476 down_read(&sinfo->groups_sem);
3477 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3478 if (!list_empty(&sinfo->block_groups[c])) {
3481 btrfs_get_block_group_info(
3482 &sinfo->block_groups[c], &space);
3483 if (space.total_bytes == 0 ||
3484 space.used_bytes == 0)
3486 flags = space.flags;
3489 * 0: if dup, single or RAID0 is configured for
3490 * any of metadata, system or data, else
3491 * 1: if RAID5 is configured, or if RAID1 or
3492 * RAID10 is configured and only two mirrors
3494 * 2: if RAID6 is configured, else
3495 * num_mirrors - 1: if RAID1 or RAID10 is
3496 * configured and more than
3497 * 2 mirrors are used.
3499 if (num_tolerated_disk_barrier_failures > 0 &&
3500 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3501 BTRFS_BLOCK_GROUP_RAID0)) ||
3502 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3504 num_tolerated_disk_barrier_failures = 0;
3505 else if (num_tolerated_disk_barrier_failures > 1) {
3506 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3507 BTRFS_BLOCK_GROUP_RAID5 |
3508 BTRFS_BLOCK_GROUP_RAID10)) {
3509 num_tolerated_disk_barrier_failures = 1;
3511 BTRFS_BLOCK_GROUP_RAID6) {
3512 num_tolerated_disk_barrier_failures = 2;
3517 up_read(&sinfo->groups_sem);
3520 return num_tolerated_disk_barrier_failures;
3523 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3525 struct list_head *head;
3526 struct btrfs_device *dev;
3527 struct btrfs_super_block *sb;
3528 struct btrfs_dev_item *dev_item;
3532 int total_errors = 0;
3535 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3536 backup_super_roots(root->fs_info);
3538 sb = root->fs_info->super_for_commit;
3539 dev_item = &sb->dev_item;
3541 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3542 head = &root->fs_info->fs_devices->devices;
3543 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3546 ret = barrier_all_devices(root->fs_info);
3549 &root->fs_info->fs_devices->device_list_mutex);
3550 btrfs_error(root->fs_info, ret,
3551 "errors while submitting device barriers.");
3556 list_for_each_entry_rcu(dev, head, dev_list) {
3561 if (!dev->in_fs_metadata || !dev->writeable)
3564 btrfs_set_stack_device_generation(dev_item, 0);
3565 btrfs_set_stack_device_type(dev_item, dev->type);
3566 btrfs_set_stack_device_id(dev_item, dev->devid);
3567 btrfs_set_stack_device_total_bytes(dev_item,
3568 dev->commit_total_bytes);
3569 btrfs_set_stack_device_bytes_used(dev_item,
3570 dev->commit_bytes_used);
3571 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3572 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3573 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3574 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3575 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3577 flags = btrfs_super_flags(sb);
3578 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3580 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3584 if (total_errors > max_errors) {
3585 btrfs_err(root->fs_info, "%d errors while writing supers",
3587 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3589 /* FUA is masked off if unsupported and can't be the reason */
3590 btrfs_error(root->fs_info, -EIO,
3591 "%d errors while writing supers", total_errors);
3596 list_for_each_entry_rcu(dev, head, dev_list) {
3599 if (!dev->in_fs_metadata || !dev->writeable)
3602 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3606 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3607 if (total_errors > max_errors) {
3608 btrfs_error(root->fs_info, -EIO,
3609 "%d errors while writing supers", total_errors);
3615 int write_ctree_super(struct btrfs_trans_handle *trans,
3616 struct btrfs_root *root, int max_mirrors)
3618 return write_all_supers(root, max_mirrors);
3621 /* Drop a fs root from the radix tree and free it. */
3622 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3623 struct btrfs_root *root)
3625 spin_lock(&fs_info->fs_roots_radix_lock);
3626 radix_tree_delete(&fs_info->fs_roots_radix,
3627 (unsigned long)root->root_key.objectid);
3628 spin_unlock(&fs_info->fs_roots_radix_lock);
3630 if (btrfs_root_refs(&root->root_item) == 0)
3631 synchronize_srcu(&fs_info->subvol_srcu);
3633 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3634 btrfs_free_log(NULL, root);
3636 if (root->free_ino_pinned)
3637 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3638 if (root->free_ino_ctl)
3639 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3643 static void free_fs_root(struct btrfs_root *root)
3645 iput(root->ino_cache_inode);
3646 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3647 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3648 root->orphan_block_rsv = NULL;
3650 free_anon_bdev(root->anon_dev);
3651 if (root->subv_writers)
3652 btrfs_free_subvolume_writers(root->subv_writers);
3653 free_extent_buffer(root->node);
3654 free_extent_buffer(root->commit_root);
3655 kfree(root->free_ino_ctl);
3656 kfree(root->free_ino_pinned);
3658 btrfs_put_fs_root(root);
3661 void btrfs_free_fs_root(struct btrfs_root *root)
3666 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3668 u64 root_objectid = 0;
3669 struct btrfs_root *gang[8];
3672 unsigned int ret = 0;
3676 index = srcu_read_lock(&fs_info->subvol_srcu);
3677 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3678 (void **)gang, root_objectid,
3681 srcu_read_unlock(&fs_info->subvol_srcu, index);
3684 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3686 for (i = 0; i < ret; i++) {
3687 /* Avoid to grab roots in dead_roots */
3688 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3692 /* grab all the search result for later use */
3693 gang[i] = btrfs_grab_fs_root(gang[i]);
3695 srcu_read_unlock(&fs_info->subvol_srcu, index);
3697 for (i = 0; i < ret; i++) {
3700 root_objectid = gang[i]->root_key.objectid;
3701 err = btrfs_orphan_cleanup(gang[i]);
3704 btrfs_put_fs_root(gang[i]);
3709 /* release the uncleaned roots due to error */
3710 for (; i < ret; i++) {
3712 btrfs_put_fs_root(gang[i]);
3717 int btrfs_commit_super(struct btrfs_root *root)
3719 struct btrfs_trans_handle *trans;
3721 mutex_lock(&root->fs_info->cleaner_mutex);
3722 btrfs_run_delayed_iputs(root);
3723 mutex_unlock(&root->fs_info->cleaner_mutex);
3724 wake_up_process(root->fs_info->cleaner_kthread);
3726 /* wait until ongoing cleanup work done */
3727 down_write(&root->fs_info->cleanup_work_sem);
3728 up_write(&root->fs_info->cleanup_work_sem);
3730 trans = btrfs_join_transaction(root);
3732 return PTR_ERR(trans);
3733 return btrfs_commit_transaction(trans, root);
3736 void close_ctree(struct btrfs_root *root)
3738 struct btrfs_fs_info *fs_info = root->fs_info;
3741 fs_info->closing = 1;
3744 /* wait for the uuid_scan task to finish */
3745 down(&fs_info->uuid_tree_rescan_sem);
3746 /* avoid complains from lockdep et al., set sem back to initial state */
3747 up(&fs_info->uuid_tree_rescan_sem);
3749 /* pause restriper - we want to resume on mount */
3750 btrfs_pause_balance(fs_info);
3752 btrfs_dev_replace_suspend_for_unmount(fs_info);
3754 btrfs_scrub_cancel(fs_info);
3756 /* wait for any defraggers to finish */
3757 wait_event(fs_info->transaction_wait,
3758 (atomic_read(&fs_info->defrag_running) == 0));
3760 /* clear out the rbtree of defraggable inodes */
3761 btrfs_cleanup_defrag_inodes(fs_info);
3763 cancel_work_sync(&fs_info->async_reclaim_work);
3765 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3766 ret = btrfs_commit_super(root);
3768 btrfs_err(fs_info, "commit super ret %d", ret);
3771 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3772 btrfs_error_commit_super(root);
3774 kthread_stop(fs_info->transaction_kthread);
3775 kthread_stop(fs_info->cleaner_kthread);
3777 fs_info->closing = 2;
3780 btrfs_free_qgroup_config(fs_info);
3782 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3783 btrfs_info(fs_info, "at unmount delalloc count %lld",
3784 percpu_counter_sum(&fs_info->delalloc_bytes));
3787 btrfs_sysfs_remove_one(fs_info);
3788 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3790 btrfs_free_fs_roots(fs_info);
3792 btrfs_put_block_group_cache(fs_info);
3794 btrfs_free_block_groups(fs_info);
3797 * we must make sure there is not any read request to
3798 * submit after we stopping all workers.
3800 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3801 btrfs_stop_all_workers(fs_info);
3804 free_root_pointers(fs_info, 1);
3806 iput(fs_info->btree_inode);
3808 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3809 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3810 btrfsic_unmount(root, fs_info->fs_devices);
3813 btrfs_close_devices(fs_info->fs_devices);
3814 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3816 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3817 percpu_counter_destroy(&fs_info->delalloc_bytes);
3818 percpu_counter_destroy(&fs_info->bio_counter);
3819 bdi_destroy(&fs_info->bdi);
3820 cleanup_srcu_struct(&fs_info->subvol_srcu);
3822 btrfs_free_stripe_hash_table(fs_info);
3824 __btrfs_free_block_rsv(root->orphan_block_rsv);
3825 root->orphan_block_rsv = NULL;
3828 while (!list_empty(&fs_info->pinned_chunks)) {
3829 struct extent_map *em;
3831 em = list_first_entry(&fs_info->pinned_chunks,
3832 struct extent_map, list);
3833 list_del_init(&em->list);
3834 free_extent_map(em);
3836 unlock_chunks(root);
3839 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3843 struct inode *btree_inode = buf->pages[0]->mapping->host;
3845 ret = extent_buffer_uptodate(buf);
3849 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3850 parent_transid, atomic);
3856 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3858 return set_extent_buffer_uptodate(buf);
3861 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3863 struct btrfs_root *root;
3864 u64 transid = btrfs_header_generation(buf);
3867 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3869 * This is a fast path so only do this check if we have sanity tests
3870 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3871 * outside of the sanity tests.
3873 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3876 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3877 btrfs_assert_tree_locked(buf);
3878 if (transid != root->fs_info->generation)
3879 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3880 "found %llu running %llu\n",
3881 buf->start, transid, root->fs_info->generation);
3882 was_dirty = set_extent_buffer_dirty(buf);
3884 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3886 root->fs_info->dirty_metadata_batch);
3887 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3888 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3889 btrfs_print_leaf(root, buf);
3895 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3899 * looks as though older kernels can get into trouble with
3900 * this code, they end up stuck in balance_dirty_pages forever
3904 if (current->flags & PF_MEMALLOC)
3908 btrfs_balance_delayed_items(root);
3910 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3911 BTRFS_DIRTY_METADATA_THRESH);
3913 balance_dirty_pages_ratelimited(
3914 root->fs_info->btree_inode->i_mapping);
3919 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3921 __btrfs_btree_balance_dirty(root, 1);
3924 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3926 __btrfs_btree_balance_dirty(root, 0);
3929 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3931 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3932 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3935 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3938 struct btrfs_super_block *sb = fs_info->super_copy;
3941 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3942 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
3943 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3946 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3947 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
3948 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3951 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3952 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
3953 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3958 * The common minimum, we don't know if we can trust the nodesize/sectorsize
3959 * items yet, they'll be verified later. Issue just a warning.
3961 if (!IS_ALIGNED(btrfs_super_root(sb), 4096))
3962 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
3963 btrfs_super_root(sb));
3964 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096))
3965 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
3966 btrfs_super_chunk_root(sb));
3967 if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096))
3968 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
3969 btrfs_super_log_root(sb));
3972 * Check the lower bound, the alignment and other constraints are
3975 if (btrfs_super_nodesize(sb) < 4096) {
3976 printk(KERN_ERR "BTRFS: nodesize too small: %u < 4096\n",
3977 btrfs_super_nodesize(sb));
3980 if (btrfs_super_sectorsize(sb) < 4096) {
3981 printk(KERN_ERR "BTRFS: sectorsize too small: %u < 4096\n",
3982 btrfs_super_sectorsize(sb));
3986 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
3987 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
3988 fs_info->fsid, sb->dev_item.fsid);
3993 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3996 if (btrfs_super_num_devices(sb) > (1UL << 31))
3997 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
3998 btrfs_super_num_devices(sb));
3999 if (btrfs_super_num_devices(sb) == 0) {
4000 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4004 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4005 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4006 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4011 * Obvious sys_chunk_array corruptions, it must hold at least one key
4014 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4015 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4016 btrfs_super_sys_array_size(sb),
4017 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4020 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4021 + sizeof(struct btrfs_chunk)) {
4022 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4023 btrfs_super_sys_array_size(sb),
4024 sizeof(struct btrfs_disk_key)
4025 + sizeof(struct btrfs_chunk));
4030 * The generation is a global counter, we'll trust it more than the others
4031 * but it's still possible that it's the one that's wrong.
4033 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4035 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4036 btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4037 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4038 && btrfs_super_cache_generation(sb) != (u64)-1)
4040 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4041 btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4046 static void btrfs_error_commit_super(struct btrfs_root *root)
4048 mutex_lock(&root->fs_info->cleaner_mutex);
4049 btrfs_run_delayed_iputs(root);
4050 mutex_unlock(&root->fs_info->cleaner_mutex);
4052 down_write(&root->fs_info->cleanup_work_sem);
4053 up_write(&root->fs_info->cleanup_work_sem);
4055 /* cleanup FS via transaction */
4056 btrfs_cleanup_transaction(root);
4059 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4061 struct btrfs_ordered_extent *ordered;
4063 spin_lock(&root->ordered_extent_lock);
4065 * This will just short circuit the ordered completion stuff which will
4066 * make sure the ordered extent gets properly cleaned up.
4068 list_for_each_entry(ordered, &root->ordered_extents,
4070 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4071 spin_unlock(&root->ordered_extent_lock);
4074 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4076 struct btrfs_root *root;
4077 struct list_head splice;
4079 INIT_LIST_HEAD(&splice);
4081 spin_lock(&fs_info->ordered_root_lock);
4082 list_splice_init(&fs_info->ordered_roots, &splice);
4083 while (!list_empty(&splice)) {
4084 root = list_first_entry(&splice, struct btrfs_root,
4086 list_move_tail(&root->ordered_root,
4087 &fs_info->ordered_roots);
4089 spin_unlock(&fs_info->ordered_root_lock);
4090 btrfs_destroy_ordered_extents(root);
4093 spin_lock(&fs_info->ordered_root_lock);
4095 spin_unlock(&fs_info->ordered_root_lock);
4098 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4099 struct btrfs_root *root)
4101 struct rb_node *node;
4102 struct btrfs_delayed_ref_root *delayed_refs;
4103 struct btrfs_delayed_ref_node *ref;
4106 delayed_refs = &trans->delayed_refs;
4108 spin_lock(&delayed_refs->lock);
4109 if (atomic_read(&delayed_refs->num_entries) == 0) {
4110 spin_unlock(&delayed_refs->lock);
4111 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4115 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4116 struct btrfs_delayed_ref_head *head;
4117 struct btrfs_delayed_ref_node *tmp;
4118 bool pin_bytes = false;
4120 head = rb_entry(node, struct btrfs_delayed_ref_head,
4122 if (!mutex_trylock(&head->mutex)) {
4123 atomic_inc(&head->node.refs);
4124 spin_unlock(&delayed_refs->lock);
4126 mutex_lock(&head->mutex);
4127 mutex_unlock(&head->mutex);
4128 btrfs_put_delayed_ref(&head->node);
4129 spin_lock(&delayed_refs->lock);
4132 spin_lock(&head->lock);
4133 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4136 list_del(&ref->list);
4137 atomic_dec(&delayed_refs->num_entries);
4138 btrfs_put_delayed_ref(ref);
4140 if (head->must_insert_reserved)
4142 btrfs_free_delayed_extent_op(head->extent_op);
4143 delayed_refs->num_heads--;
4144 if (head->processing == 0)
4145 delayed_refs->num_heads_ready--;
4146 atomic_dec(&delayed_refs->num_entries);
4147 head->node.in_tree = 0;
4148 rb_erase(&head->href_node, &delayed_refs->href_root);
4149 spin_unlock(&head->lock);
4150 spin_unlock(&delayed_refs->lock);
4151 mutex_unlock(&head->mutex);
4154 btrfs_pin_extent(root, head->node.bytenr,
4155 head->node.num_bytes, 1);
4156 btrfs_put_delayed_ref(&head->node);
4158 spin_lock(&delayed_refs->lock);
4161 spin_unlock(&delayed_refs->lock);
4166 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4168 struct btrfs_inode *btrfs_inode;
4169 struct list_head splice;
4171 INIT_LIST_HEAD(&splice);
4173 spin_lock(&root->delalloc_lock);
4174 list_splice_init(&root->delalloc_inodes, &splice);
4176 while (!list_empty(&splice)) {
4177 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4180 list_del_init(&btrfs_inode->delalloc_inodes);
4181 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4182 &btrfs_inode->runtime_flags);
4183 spin_unlock(&root->delalloc_lock);
4185 btrfs_invalidate_inodes(btrfs_inode->root);
4187 spin_lock(&root->delalloc_lock);
4190 spin_unlock(&root->delalloc_lock);
4193 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4195 struct btrfs_root *root;
4196 struct list_head splice;
4198 INIT_LIST_HEAD(&splice);
4200 spin_lock(&fs_info->delalloc_root_lock);
4201 list_splice_init(&fs_info->delalloc_roots, &splice);
4202 while (!list_empty(&splice)) {
4203 root = list_first_entry(&splice, struct btrfs_root,
4205 list_del_init(&root->delalloc_root);
4206 root = btrfs_grab_fs_root(root);
4208 spin_unlock(&fs_info->delalloc_root_lock);
4210 btrfs_destroy_delalloc_inodes(root);
4211 btrfs_put_fs_root(root);
4213 spin_lock(&fs_info->delalloc_root_lock);
4215 spin_unlock(&fs_info->delalloc_root_lock);
4218 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4219 struct extent_io_tree *dirty_pages,
4223 struct extent_buffer *eb;
4228 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4233 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4234 while (start <= end) {
4235 eb = btrfs_find_tree_block(root->fs_info, start);
4236 start += root->nodesize;
4239 wait_on_extent_buffer_writeback(eb);
4241 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4243 clear_extent_buffer_dirty(eb);
4244 free_extent_buffer_stale(eb);
4251 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4252 struct extent_io_tree *pinned_extents)
4254 struct extent_io_tree *unpin;
4260 unpin = pinned_extents;
4263 ret = find_first_extent_bit(unpin, 0, &start, &end,
4264 EXTENT_DIRTY, NULL);
4268 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4269 btrfs_error_unpin_extent_range(root, start, end);
4274 if (unpin == &root->fs_info->freed_extents[0])
4275 unpin = &root->fs_info->freed_extents[1];
4277 unpin = &root->fs_info->freed_extents[0];
4285 static void btrfs_free_pending_ordered(struct btrfs_transaction *cur_trans,
4286 struct btrfs_fs_info *fs_info)
4288 struct btrfs_ordered_extent *ordered;
4290 spin_lock(&fs_info->trans_lock);
4291 while (!list_empty(&cur_trans->pending_ordered)) {
4292 ordered = list_first_entry(&cur_trans->pending_ordered,
4293 struct btrfs_ordered_extent,
4295 list_del_init(&ordered->trans_list);
4296 spin_unlock(&fs_info->trans_lock);
4298 btrfs_put_ordered_extent(ordered);
4299 spin_lock(&fs_info->trans_lock);
4301 spin_unlock(&fs_info->trans_lock);
4304 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4305 struct btrfs_root *root)
4307 btrfs_destroy_delayed_refs(cur_trans, root);
4309 cur_trans->state = TRANS_STATE_COMMIT_START;
4310 wake_up(&root->fs_info->transaction_blocked_wait);
4312 cur_trans->state = TRANS_STATE_UNBLOCKED;
4313 wake_up(&root->fs_info->transaction_wait);
4315 btrfs_free_pending_ordered(cur_trans, root->fs_info);
4316 btrfs_destroy_delayed_inodes(root);
4317 btrfs_assert_delayed_root_empty(root);
4319 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4321 btrfs_destroy_pinned_extent(root,
4322 root->fs_info->pinned_extents);
4324 cur_trans->state =TRANS_STATE_COMPLETED;
4325 wake_up(&cur_trans->commit_wait);
4328 memset(cur_trans, 0, sizeof(*cur_trans));
4329 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4333 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4335 struct btrfs_transaction *t;
4337 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4339 spin_lock(&root->fs_info->trans_lock);
4340 while (!list_empty(&root->fs_info->trans_list)) {
4341 t = list_first_entry(&root->fs_info->trans_list,
4342 struct btrfs_transaction, list);
4343 if (t->state >= TRANS_STATE_COMMIT_START) {
4344 atomic_inc(&t->use_count);
4345 spin_unlock(&root->fs_info->trans_lock);
4346 btrfs_wait_for_commit(root, t->transid);
4347 btrfs_put_transaction(t);
4348 spin_lock(&root->fs_info->trans_lock);
4351 if (t == root->fs_info->running_transaction) {
4352 t->state = TRANS_STATE_COMMIT_DOING;
4353 spin_unlock(&root->fs_info->trans_lock);
4355 * We wait for 0 num_writers since we don't hold a trans
4356 * handle open currently for this transaction.
4358 wait_event(t->writer_wait,
4359 atomic_read(&t->num_writers) == 0);
4361 spin_unlock(&root->fs_info->trans_lock);
4363 btrfs_cleanup_one_transaction(t, root);
4365 spin_lock(&root->fs_info->trans_lock);
4366 if (t == root->fs_info->running_transaction)
4367 root->fs_info->running_transaction = NULL;
4368 list_del_init(&t->list);
4369 spin_unlock(&root->fs_info->trans_lock);
4371 btrfs_put_transaction(t);
4372 trace_btrfs_transaction_commit(root);
4373 spin_lock(&root->fs_info->trans_lock);
4375 spin_unlock(&root->fs_info->trans_lock);
4376 btrfs_destroy_all_ordered_extents(root->fs_info);
4377 btrfs_destroy_delayed_inodes(root);
4378 btrfs_assert_delayed_root_empty(root);
4379 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4380 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4381 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4386 static const struct extent_io_ops btree_extent_io_ops = {
4387 .readpage_end_io_hook = btree_readpage_end_io_hook,
4388 .readpage_io_failed_hook = btree_io_failed_hook,
4389 .submit_bio_hook = btree_submit_bio_hook,
4390 /* note we're sharing with inode.c for the merge bio hook */
4391 .merge_bio_hook = btrfs_merge_bio_hook,