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, int err)
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 = err;
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 bio_endio(async->bio, async->error);
815 async->submit_bio_done(async->inode, async->rw, async->bio,
816 async->mirror_num, async->bio_flags,
820 static void run_one_async_free(struct btrfs_work *work)
822 struct async_submit_bio *async;
824 async = container_of(work, struct async_submit_bio, work);
828 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
829 int rw, struct bio *bio, int mirror_num,
830 unsigned long bio_flags,
832 extent_submit_bio_hook_t *submit_bio_start,
833 extent_submit_bio_hook_t *submit_bio_done)
835 struct async_submit_bio *async;
837 async = kmalloc(sizeof(*async), GFP_NOFS);
841 async->inode = inode;
844 async->mirror_num = mirror_num;
845 async->submit_bio_start = submit_bio_start;
846 async->submit_bio_done = submit_bio_done;
848 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
849 run_one_async_done, run_one_async_free);
851 async->bio_flags = bio_flags;
852 async->bio_offset = bio_offset;
856 atomic_inc(&fs_info->nr_async_submits);
859 btrfs_set_work_high_priority(&async->work);
861 btrfs_queue_work(fs_info->workers, &async->work);
863 while (atomic_read(&fs_info->async_submit_draining) &&
864 atomic_read(&fs_info->nr_async_submits)) {
865 wait_event(fs_info->async_submit_wait,
866 (atomic_read(&fs_info->nr_async_submits) == 0));
872 static int btree_csum_one_bio(struct bio *bio)
874 struct bio_vec *bvec;
875 struct btrfs_root *root;
878 bio_for_each_segment_all(bvec, bio, i) {
879 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
880 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
888 static int __btree_submit_bio_start(struct inode *inode, int rw,
889 struct bio *bio, int mirror_num,
890 unsigned long bio_flags,
894 * when we're called for a write, we're already in the async
895 * submission context. Just jump into btrfs_map_bio
897 return btree_csum_one_bio(bio);
900 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
901 int mirror_num, unsigned long bio_flags,
907 * when we're called for a write, we're already in the async
908 * submission context. Just jump into btrfs_map_bio
910 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
916 static int check_async_write(struct inode *inode, unsigned long bio_flags)
918 if (bio_flags & EXTENT_BIO_TREE_LOG)
927 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
928 int mirror_num, unsigned long bio_flags,
931 int async = check_async_write(inode, bio_flags);
934 if (!(rw & REQ_WRITE)) {
936 * called for a read, do the setup so that checksum validation
937 * can happen in the async kernel threads
939 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
940 bio, BTRFS_WQ_ENDIO_METADATA);
943 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
946 ret = btree_csum_one_bio(bio);
949 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
953 * kthread helpers are used to submit writes so that
954 * checksumming can happen in parallel across all CPUs
956 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
957 inode, rw, bio, mirror_num, 0,
959 __btree_submit_bio_start,
960 __btree_submit_bio_done);
970 #ifdef CONFIG_MIGRATION
971 static int btree_migratepage(struct address_space *mapping,
972 struct page *newpage, struct page *page,
973 enum migrate_mode mode)
976 * we can't safely write a btree page from here,
977 * we haven't done the locking hook
982 * Buffers may be managed in a filesystem specific way.
983 * We must have no buffers or drop them.
985 if (page_has_private(page) &&
986 !try_to_release_page(page, GFP_KERNEL))
988 return migrate_page(mapping, newpage, page, mode);
993 static int btree_writepages(struct address_space *mapping,
994 struct writeback_control *wbc)
996 struct btrfs_fs_info *fs_info;
999 if (wbc->sync_mode == WB_SYNC_NONE) {
1001 if (wbc->for_kupdate)
1004 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1005 /* this is a bit racy, but that's ok */
1006 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1007 BTRFS_DIRTY_METADATA_THRESH);
1011 return btree_write_cache_pages(mapping, wbc);
1014 static int btree_readpage(struct file *file, struct page *page)
1016 struct extent_io_tree *tree;
1017 tree = &BTRFS_I(page->mapping->host)->io_tree;
1018 return extent_read_full_page(tree, page, btree_get_extent, 0);
1021 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1023 if (PageWriteback(page) || PageDirty(page))
1026 return try_release_extent_buffer(page);
1029 static void btree_invalidatepage(struct page *page, unsigned int offset,
1030 unsigned int length)
1032 struct extent_io_tree *tree;
1033 tree = &BTRFS_I(page->mapping->host)->io_tree;
1034 extent_invalidatepage(tree, page, offset);
1035 btree_releasepage(page, GFP_NOFS);
1036 if (PagePrivate(page)) {
1037 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1038 "page private not zero on page %llu",
1039 (unsigned long long)page_offset(page));
1040 ClearPagePrivate(page);
1041 set_page_private(page, 0);
1042 page_cache_release(page);
1046 static int btree_set_page_dirty(struct page *page)
1049 struct extent_buffer *eb;
1051 BUG_ON(!PagePrivate(page));
1052 eb = (struct extent_buffer *)page->private;
1054 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1055 BUG_ON(!atomic_read(&eb->refs));
1056 btrfs_assert_tree_locked(eb);
1058 return __set_page_dirty_nobuffers(page);
1061 static const struct address_space_operations btree_aops = {
1062 .readpage = btree_readpage,
1063 .writepages = btree_writepages,
1064 .releasepage = btree_releasepage,
1065 .invalidatepage = btree_invalidatepage,
1066 #ifdef CONFIG_MIGRATION
1067 .migratepage = btree_migratepage,
1069 .set_page_dirty = btree_set_page_dirty,
1072 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1074 struct extent_buffer *buf = NULL;
1075 struct inode *btree_inode = root->fs_info->btree_inode;
1077 buf = btrfs_find_create_tree_block(root, bytenr);
1080 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1081 buf, 0, WAIT_NONE, btree_get_extent, 0);
1082 free_extent_buffer(buf);
1085 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1086 int mirror_num, struct extent_buffer **eb)
1088 struct extent_buffer *buf = NULL;
1089 struct inode *btree_inode = root->fs_info->btree_inode;
1090 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1093 buf = btrfs_find_create_tree_block(root, bytenr);
1097 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1099 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1100 btree_get_extent, mirror_num);
1102 free_extent_buffer(buf);
1106 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1107 free_extent_buffer(buf);
1109 } else if (extent_buffer_uptodate(buf)) {
1112 free_extent_buffer(buf);
1117 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1120 return find_extent_buffer(fs_info, bytenr);
1123 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1126 if (btrfs_test_is_dummy_root(root))
1127 return alloc_test_extent_buffer(root->fs_info, bytenr);
1128 return alloc_extent_buffer(root->fs_info, bytenr);
1132 int btrfs_write_tree_block(struct extent_buffer *buf)
1134 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1135 buf->start + buf->len - 1);
1138 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1140 return filemap_fdatawait_range(buf->pages[0]->mapping,
1141 buf->start, buf->start + buf->len - 1);
1144 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1147 struct extent_buffer *buf = NULL;
1150 buf = btrfs_find_create_tree_block(root, bytenr);
1154 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1156 free_extent_buffer(buf);
1163 void clean_tree_block(struct btrfs_trans_handle *trans,
1164 struct btrfs_fs_info *fs_info,
1165 struct extent_buffer *buf)
1167 if (btrfs_header_generation(buf) ==
1168 fs_info->running_transaction->transid) {
1169 btrfs_assert_tree_locked(buf);
1171 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1172 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1174 fs_info->dirty_metadata_batch);
1175 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1176 btrfs_set_lock_blocking(buf);
1177 clear_extent_buffer_dirty(buf);
1182 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1184 struct btrfs_subvolume_writers *writers;
1187 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1189 return ERR_PTR(-ENOMEM);
1191 ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1194 return ERR_PTR(ret);
1197 init_waitqueue_head(&writers->wait);
1202 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1204 percpu_counter_destroy(&writers->counter);
1208 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1209 struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1213 root->commit_root = NULL;
1214 root->sectorsize = sectorsize;
1215 root->nodesize = nodesize;
1216 root->stripesize = stripesize;
1218 root->orphan_cleanup_state = 0;
1220 root->objectid = objectid;
1221 root->last_trans = 0;
1222 root->highest_objectid = 0;
1223 root->nr_delalloc_inodes = 0;
1224 root->nr_ordered_extents = 0;
1226 root->inode_tree = RB_ROOT;
1227 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1228 root->block_rsv = NULL;
1229 root->orphan_block_rsv = NULL;
1231 INIT_LIST_HEAD(&root->dirty_list);
1232 INIT_LIST_HEAD(&root->root_list);
1233 INIT_LIST_HEAD(&root->delalloc_inodes);
1234 INIT_LIST_HEAD(&root->delalloc_root);
1235 INIT_LIST_HEAD(&root->ordered_extents);
1236 INIT_LIST_HEAD(&root->ordered_root);
1237 INIT_LIST_HEAD(&root->logged_list[0]);
1238 INIT_LIST_HEAD(&root->logged_list[1]);
1239 spin_lock_init(&root->orphan_lock);
1240 spin_lock_init(&root->inode_lock);
1241 spin_lock_init(&root->delalloc_lock);
1242 spin_lock_init(&root->ordered_extent_lock);
1243 spin_lock_init(&root->accounting_lock);
1244 spin_lock_init(&root->log_extents_lock[0]);
1245 spin_lock_init(&root->log_extents_lock[1]);
1246 mutex_init(&root->objectid_mutex);
1247 mutex_init(&root->log_mutex);
1248 mutex_init(&root->ordered_extent_mutex);
1249 mutex_init(&root->delalloc_mutex);
1250 init_waitqueue_head(&root->log_writer_wait);
1251 init_waitqueue_head(&root->log_commit_wait[0]);
1252 init_waitqueue_head(&root->log_commit_wait[1]);
1253 INIT_LIST_HEAD(&root->log_ctxs[0]);
1254 INIT_LIST_HEAD(&root->log_ctxs[1]);
1255 atomic_set(&root->log_commit[0], 0);
1256 atomic_set(&root->log_commit[1], 0);
1257 atomic_set(&root->log_writers, 0);
1258 atomic_set(&root->log_batch, 0);
1259 atomic_set(&root->orphan_inodes, 0);
1260 atomic_set(&root->refs, 1);
1261 atomic_set(&root->will_be_snapshoted, 0);
1262 root->log_transid = 0;
1263 root->log_transid_committed = -1;
1264 root->last_log_commit = 0;
1266 extent_io_tree_init(&root->dirty_log_pages,
1267 fs_info->btree_inode->i_mapping);
1269 memset(&root->root_key, 0, sizeof(root->root_key));
1270 memset(&root->root_item, 0, sizeof(root->root_item));
1271 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1273 root->defrag_trans_start = fs_info->generation;
1275 root->defrag_trans_start = 0;
1276 root->root_key.objectid = objectid;
1279 spin_lock_init(&root->root_item_lock);
1282 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1284 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1286 root->fs_info = fs_info;
1290 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1291 /* Should only be used by the testing infrastructure */
1292 struct btrfs_root *btrfs_alloc_dummy_root(void)
1294 struct btrfs_root *root;
1296 root = btrfs_alloc_root(NULL);
1298 return ERR_PTR(-ENOMEM);
1299 __setup_root(4096, 4096, 4096, root, NULL, 1);
1300 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1301 root->alloc_bytenr = 0;
1307 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1308 struct btrfs_fs_info *fs_info,
1311 struct extent_buffer *leaf;
1312 struct btrfs_root *tree_root = fs_info->tree_root;
1313 struct btrfs_root *root;
1314 struct btrfs_key key;
1318 root = btrfs_alloc_root(fs_info);
1320 return ERR_PTR(-ENOMEM);
1322 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1323 tree_root->stripesize, root, fs_info, objectid);
1324 root->root_key.objectid = objectid;
1325 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1326 root->root_key.offset = 0;
1328 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1330 ret = PTR_ERR(leaf);
1335 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1336 btrfs_set_header_bytenr(leaf, leaf->start);
1337 btrfs_set_header_generation(leaf, trans->transid);
1338 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1339 btrfs_set_header_owner(leaf, objectid);
1342 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1344 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1345 btrfs_header_chunk_tree_uuid(leaf),
1347 btrfs_mark_buffer_dirty(leaf);
1349 root->commit_root = btrfs_root_node(root);
1350 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1352 root->root_item.flags = 0;
1353 root->root_item.byte_limit = 0;
1354 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1355 btrfs_set_root_generation(&root->root_item, trans->transid);
1356 btrfs_set_root_level(&root->root_item, 0);
1357 btrfs_set_root_refs(&root->root_item, 1);
1358 btrfs_set_root_used(&root->root_item, leaf->len);
1359 btrfs_set_root_last_snapshot(&root->root_item, 0);
1360 btrfs_set_root_dirid(&root->root_item, 0);
1362 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1363 root->root_item.drop_level = 0;
1365 key.objectid = objectid;
1366 key.type = BTRFS_ROOT_ITEM_KEY;
1368 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1372 btrfs_tree_unlock(leaf);
1378 btrfs_tree_unlock(leaf);
1379 free_extent_buffer(root->commit_root);
1380 free_extent_buffer(leaf);
1384 return ERR_PTR(ret);
1387 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1388 struct btrfs_fs_info *fs_info)
1390 struct btrfs_root *root;
1391 struct btrfs_root *tree_root = fs_info->tree_root;
1392 struct extent_buffer *leaf;
1394 root = btrfs_alloc_root(fs_info);
1396 return ERR_PTR(-ENOMEM);
1398 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1399 tree_root->stripesize, root, fs_info,
1400 BTRFS_TREE_LOG_OBJECTID);
1402 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1403 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1404 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1407 * DON'T set REF_COWS for log trees
1409 * log trees do not get reference counted because they go away
1410 * before a real commit is actually done. They do store pointers
1411 * to file data extents, and those reference counts still get
1412 * updated (along with back refs to the log tree).
1415 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1419 return ERR_CAST(leaf);
1422 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1423 btrfs_set_header_bytenr(leaf, leaf->start);
1424 btrfs_set_header_generation(leaf, trans->transid);
1425 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1426 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1429 write_extent_buffer(root->node, root->fs_info->fsid,
1430 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1431 btrfs_mark_buffer_dirty(root->node);
1432 btrfs_tree_unlock(root->node);
1436 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1437 struct btrfs_fs_info *fs_info)
1439 struct btrfs_root *log_root;
1441 log_root = alloc_log_tree(trans, fs_info);
1442 if (IS_ERR(log_root))
1443 return PTR_ERR(log_root);
1444 WARN_ON(fs_info->log_root_tree);
1445 fs_info->log_root_tree = log_root;
1449 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1450 struct btrfs_root *root)
1452 struct btrfs_root *log_root;
1453 struct btrfs_inode_item *inode_item;
1455 log_root = alloc_log_tree(trans, root->fs_info);
1456 if (IS_ERR(log_root))
1457 return PTR_ERR(log_root);
1459 log_root->last_trans = trans->transid;
1460 log_root->root_key.offset = root->root_key.objectid;
1462 inode_item = &log_root->root_item.inode;
1463 btrfs_set_stack_inode_generation(inode_item, 1);
1464 btrfs_set_stack_inode_size(inode_item, 3);
1465 btrfs_set_stack_inode_nlink(inode_item, 1);
1466 btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1467 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1469 btrfs_set_root_node(&log_root->root_item, log_root->node);
1471 WARN_ON(root->log_root);
1472 root->log_root = log_root;
1473 root->log_transid = 0;
1474 root->log_transid_committed = -1;
1475 root->last_log_commit = 0;
1479 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1480 struct btrfs_key *key)
1482 struct btrfs_root *root;
1483 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1484 struct btrfs_path *path;
1488 path = btrfs_alloc_path();
1490 return ERR_PTR(-ENOMEM);
1492 root = btrfs_alloc_root(fs_info);
1498 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1499 tree_root->stripesize, root, fs_info, key->objectid);
1501 ret = btrfs_find_root(tree_root, key, path,
1502 &root->root_item, &root->root_key);
1509 generation = btrfs_root_generation(&root->root_item);
1510 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1515 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1519 root->commit_root = btrfs_root_node(root);
1521 btrfs_free_path(path);
1525 free_extent_buffer(root->node);
1529 root = ERR_PTR(ret);
1533 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1534 struct btrfs_key *location)
1536 struct btrfs_root *root;
1538 root = btrfs_read_tree_root(tree_root, location);
1542 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1543 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1544 btrfs_check_and_init_root_item(&root->root_item);
1550 int btrfs_init_fs_root(struct btrfs_root *root)
1553 struct btrfs_subvolume_writers *writers;
1555 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1556 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1558 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1563 writers = btrfs_alloc_subvolume_writers();
1564 if (IS_ERR(writers)) {
1565 ret = PTR_ERR(writers);
1568 root->subv_writers = writers;
1570 btrfs_init_free_ino_ctl(root);
1571 spin_lock_init(&root->ino_cache_lock);
1572 init_waitqueue_head(&root->ino_cache_wait);
1574 ret = get_anon_bdev(&root->anon_dev);
1580 btrfs_free_subvolume_writers(root->subv_writers);
1582 kfree(root->free_ino_ctl);
1583 kfree(root->free_ino_pinned);
1587 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1590 struct btrfs_root *root;
1592 spin_lock(&fs_info->fs_roots_radix_lock);
1593 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1594 (unsigned long)root_id);
1595 spin_unlock(&fs_info->fs_roots_radix_lock);
1599 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1600 struct btrfs_root *root)
1604 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1608 spin_lock(&fs_info->fs_roots_radix_lock);
1609 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1610 (unsigned long)root->root_key.objectid,
1613 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1614 spin_unlock(&fs_info->fs_roots_radix_lock);
1615 radix_tree_preload_end();
1620 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1621 struct btrfs_key *location,
1624 struct btrfs_root *root;
1625 struct btrfs_path *path;
1626 struct btrfs_key key;
1629 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1630 return fs_info->tree_root;
1631 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1632 return fs_info->extent_root;
1633 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1634 return fs_info->chunk_root;
1635 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1636 return fs_info->dev_root;
1637 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1638 return fs_info->csum_root;
1639 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1640 return fs_info->quota_root ? fs_info->quota_root :
1642 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1643 return fs_info->uuid_root ? fs_info->uuid_root :
1646 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1648 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1649 return ERR_PTR(-ENOENT);
1653 root = btrfs_read_fs_root(fs_info->tree_root, location);
1657 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1662 ret = btrfs_init_fs_root(root);
1666 path = btrfs_alloc_path();
1671 key.objectid = BTRFS_ORPHAN_OBJECTID;
1672 key.type = BTRFS_ORPHAN_ITEM_KEY;
1673 key.offset = location->objectid;
1675 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1676 btrfs_free_path(path);
1680 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1682 ret = btrfs_insert_fs_root(fs_info, root);
1684 if (ret == -EEXIST) {
1693 return ERR_PTR(ret);
1696 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1698 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1700 struct btrfs_device *device;
1701 struct backing_dev_info *bdi;
1704 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1707 bdi = blk_get_backing_dev_info(device->bdev);
1708 if (bdi_congested(bdi, bdi_bits)) {
1717 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1721 err = bdi_setup_and_register(bdi, "btrfs");
1725 bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE;
1726 bdi->congested_fn = btrfs_congested_fn;
1727 bdi->congested_data = info;
1732 * called by the kthread helper functions to finally call the bio end_io
1733 * functions. This is where read checksum verification actually happens
1735 static void end_workqueue_fn(struct btrfs_work *work)
1738 struct btrfs_end_io_wq *end_io_wq;
1741 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1742 bio = end_io_wq->bio;
1744 error = end_io_wq->error;
1745 bio->bi_private = end_io_wq->private;
1746 bio->bi_end_io = end_io_wq->end_io;
1747 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1748 bio_endio_nodec(bio, error);
1751 static int cleaner_kthread(void *arg)
1753 struct btrfs_root *root = arg;
1759 /* Make the cleaner go to sleep early. */
1760 if (btrfs_need_cleaner_sleep(root))
1763 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1767 * Avoid the problem that we change the status of the fs
1768 * during the above check and trylock.
1770 if (btrfs_need_cleaner_sleep(root)) {
1771 mutex_unlock(&root->fs_info->cleaner_mutex);
1775 btrfs_run_delayed_iputs(root);
1776 btrfs_delete_unused_bgs(root->fs_info);
1777 again = btrfs_clean_one_deleted_snapshot(root);
1778 mutex_unlock(&root->fs_info->cleaner_mutex);
1781 * The defragger has dealt with the R/O remount and umount,
1782 * needn't do anything special here.
1784 btrfs_run_defrag_inodes(root->fs_info);
1786 if (!try_to_freeze() && !again) {
1787 set_current_state(TASK_INTERRUPTIBLE);
1788 if (!kthread_should_stop())
1790 __set_current_state(TASK_RUNNING);
1792 } while (!kthread_should_stop());
1796 static int transaction_kthread(void *arg)
1798 struct btrfs_root *root = arg;
1799 struct btrfs_trans_handle *trans;
1800 struct btrfs_transaction *cur;
1803 unsigned long delay;
1807 cannot_commit = false;
1808 delay = HZ * root->fs_info->commit_interval;
1809 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1811 spin_lock(&root->fs_info->trans_lock);
1812 cur = root->fs_info->running_transaction;
1814 spin_unlock(&root->fs_info->trans_lock);
1818 now = get_seconds();
1819 if (cur->state < TRANS_STATE_BLOCKED &&
1820 (now < cur->start_time ||
1821 now - cur->start_time < root->fs_info->commit_interval)) {
1822 spin_unlock(&root->fs_info->trans_lock);
1826 transid = cur->transid;
1827 spin_unlock(&root->fs_info->trans_lock);
1829 /* If the file system is aborted, this will always fail. */
1830 trans = btrfs_attach_transaction(root);
1831 if (IS_ERR(trans)) {
1832 if (PTR_ERR(trans) != -ENOENT)
1833 cannot_commit = true;
1836 if (transid == trans->transid) {
1837 btrfs_commit_transaction(trans, root);
1839 btrfs_end_transaction(trans, root);
1842 wake_up_process(root->fs_info->cleaner_kthread);
1843 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1845 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1846 &root->fs_info->fs_state)))
1847 btrfs_cleanup_transaction(root);
1848 if (!try_to_freeze()) {
1849 set_current_state(TASK_INTERRUPTIBLE);
1850 if (!kthread_should_stop() &&
1851 (!btrfs_transaction_blocked(root->fs_info) ||
1853 schedule_timeout(delay);
1854 __set_current_state(TASK_RUNNING);
1856 } while (!kthread_should_stop());
1861 * this will find the highest generation in the array of
1862 * root backups. The index of the highest array is returned,
1863 * or -1 if we can't find anything.
1865 * We check to make sure the array is valid by comparing the
1866 * generation of the latest root in the array with the generation
1867 * in the super block. If they don't match we pitch it.
1869 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1872 int newest_index = -1;
1873 struct btrfs_root_backup *root_backup;
1876 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1877 root_backup = info->super_copy->super_roots + i;
1878 cur = btrfs_backup_tree_root_gen(root_backup);
1879 if (cur == newest_gen)
1883 /* check to see if we actually wrapped around */
1884 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1885 root_backup = info->super_copy->super_roots;
1886 cur = btrfs_backup_tree_root_gen(root_backup);
1887 if (cur == newest_gen)
1890 return newest_index;
1895 * find the oldest backup so we know where to store new entries
1896 * in the backup array. This will set the backup_root_index
1897 * field in the fs_info struct
1899 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1902 int newest_index = -1;
1904 newest_index = find_newest_super_backup(info, newest_gen);
1905 /* if there was garbage in there, just move along */
1906 if (newest_index == -1) {
1907 info->backup_root_index = 0;
1909 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1914 * copy all the root pointers into the super backup array.
1915 * this will bump the backup pointer by one when it is
1918 static void backup_super_roots(struct btrfs_fs_info *info)
1921 struct btrfs_root_backup *root_backup;
1924 next_backup = info->backup_root_index;
1925 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1926 BTRFS_NUM_BACKUP_ROOTS;
1929 * just overwrite the last backup if we're at the same generation
1930 * this happens only at umount
1932 root_backup = info->super_for_commit->super_roots + last_backup;
1933 if (btrfs_backup_tree_root_gen(root_backup) ==
1934 btrfs_header_generation(info->tree_root->node))
1935 next_backup = last_backup;
1937 root_backup = info->super_for_commit->super_roots + next_backup;
1940 * make sure all of our padding and empty slots get zero filled
1941 * regardless of which ones we use today
1943 memset(root_backup, 0, sizeof(*root_backup));
1945 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1947 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1948 btrfs_set_backup_tree_root_gen(root_backup,
1949 btrfs_header_generation(info->tree_root->node));
1951 btrfs_set_backup_tree_root_level(root_backup,
1952 btrfs_header_level(info->tree_root->node));
1954 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1955 btrfs_set_backup_chunk_root_gen(root_backup,
1956 btrfs_header_generation(info->chunk_root->node));
1957 btrfs_set_backup_chunk_root_level(root_backup,
1958 btrfs_header_level(info->chunk_root->node));
1960 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1961 btrfs_set_backup_extent_root_gen(root_backup,
1962 btrfs_header_generation(info->extent_root->node));
1963 btrfs_set_backup_extent_root_level(root_backup,
1964 btrfs_header_level(info->extent_root->node));
1967 * we might commit during log recovery, which happens before we set
1968 * the fs_root. Make sure it is valid before we fill it in.
1970 if (info->fs_root && info->fs_root->node) {
1971 btrfs_set_backup_fs_root(root_backup,
1972 info->fs_root->node->start);
1973 btrfs_set_backup_fs_root_gen(root_backup,
1974 btrfs_header_generation(info->fs_root->node));
1975 btrfs_set_backup_fs_root_level(root_backup,
1976 btrfs_header_level(info->fs_root->node));
1979 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1980 btrfs_set_backup_dev_root_gen(root_backup,
1981 btrfs_header_generation(info->dev_root->node));
1982 btrfs_set_backup_dev_root_level(root_backup,
1983 btrfs_header_level(info->dev_root->node));
1985 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1986 btrfs_set_backup_csum_root_gen(root_backup,
1987 btrfs_header_generation(info->csum_root->node));
1988 btrfs_set_backup_csum_root_level(root_backup,
1989 btrfs_header_level(info->csum_root->node));
1991 btrfs_set_backup_total_bytes(root_backup,
1992 btrfs_super_total_bytes(info->super_copy));
1993 btrfs_set_backup_bytes_used(root_backup,
1994 btrfs_super_bytes_used(info->super_copy));
1995 btrfs_set_backup_num_devices(root_backup,
1996 btrfs_super_num_devices(info->super_copy));
1999 * if we don't copy this out to the super_copy, it won't get remembered
2000 * for the next commit
2002 memcpy(&info->super_copy->super_roots,
2003 &info->super_for_commit->super_roots,
2004 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2008 * this copies info out of the root backup array and back into
2009 * the in-memory super block. It is meant to help iterate through
2010 * the array, so you send it the number of backups you've already
2011 * tried and the last backup index you used.
2013 * this returns -1 when it has tried all the backups
2015 static noinline int next_root_backup(struct btrfs_fs_info *info,
2016 struct btrfs_super_block *super,
2017 int *num_backups_tried, int *backup_index)
2019 struct btrfs_root_backup *root_backup;
2020 int newest = *backup_index;
2022 if (*num_backups_tried == 0) {
2023 u64 gen = btrfs_super_generation(super);
2025 newest = find_newest_super_backup(info, gen);
2029 *backup_index = newest;
2030 *num_backups_tried = 1;
2031 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2032 /* we've tried all the backups, all done */
2035 /* jump to the next oldest backup */
2036 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2037 BTRFS_NUM_BACKUP_ROOTS;
2038 *backup_index = newest;
2039 *num_backups_tried += 1;
2041 root_backup = super->super_roots + newest;
2043 btrfs_set_super_generation(super,
2044 btrfs_backup_tree_root_gen(root_backup));
2045 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2046 btrfs_set_super_root_level(super,
2047 btrfs_backup_tree_root_level(root_backup));
2048 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2051 * fixme: the total bytes and num_devices need to match or we should
2054 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2055 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2059 /* helper to cleanup workers */
2060 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2062 btrfs_destroy_workqueue(fs_info->fixup_workers);
2063 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2064 btrfs_destroy_workqueue(fs_info->workers);
2065 btrfs_destroy_workqueue(fs_info->endio_workers);
2066 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2067 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2068 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2069 btrfs_destroy_workqueue(fs_info->rmw_workers);
2070 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2071 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2072 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2073 btrfs_destroy_workqueue(fs_info->submit_workers);
2074 btrfs_destroy_workqueue(fs_info->delayed_workers);
2075 btrfs_destroy_workqueue(fs_info->caching_workers);
2076 btrfs_destroy_workqueue(fs_info->readahead_workers);
2077 btrfs_destroy_workqueue(fs_info->flush_workers);
2078 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2079 btrfs_destroy_workqueue(fs_info->extent_workers);
2082 static void free_root_extent_buffers(struct btrfs_root *root)
2085 free_extent_buffer(root->node);
2086 free_extent_buffer(root->commit_root);
2088 root->commit_root = NULL;
2092 /* helper to cleanup tree roots */
2093 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2095 free_root_extent_buffers(info->tree_root);
2097 free_root_extent_buffers(info->dev_root);
2098 free_root_extent_buffers(info->extent_root);
2099 free_root_extent_buffers(info->csum_root);
2100 free_root_extent_buffers(info->quota_root);
2101 free_root_extent_buffers(info->uuid_root);
2103 free_root_extent_buffers(info->chunk_root);
2106 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2109 struct btrfs_root *gang[8];
2112 while (!list_empty(&fs_info->dead_roots)) {
2113 gang[0] = list_entry(fs_info->dead_roots.next,
2114 struct btrfs_root, root_list);
2115 list_del(&gang[0]->root_list);
2117 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2118 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2120 free_extent_buffer(gang[0]->node);
2121 free_extent_buffer(gang[0]->commit_root);
2122 btrfs_put_fs_root(gang[0]);
2127 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2132 for (i = 0; i < ret; i++)
2133 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2136 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2137 btrfs_free_log_root_tree(NULL, fs_info);
2138 btrfs_destroy_pinned_extent(fs_info->tree_root,
2139 fs_info->pinned_extents);
2143 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2145 mutex_init(&fs_info->scrub_lock);
2146 atomic_set(&fs_info->scrubs_running, 0);
2147 atomic_set(&fs_info->scrub_pause_req, 0);
2148 atomic_set(&fs_info->scrubs_paused, 0);
2149 atomic_set(&fs_info->scrub_cancel_req, 0);
2150 init_waitqueue_head(&fs_info->scrub_pause_wait);
2151 fs_info->scrub_workers_refcnt = 0;
2154 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2156 spin_lock_init(&fs_info->balance_lock);
2157 mutex_init(&fs_info->balance_mutex);
2158 atomic_set(&fs_info->balance_running, 0);
2159 atomic_set(&fs_info->balance_pause_req, 0);
2160 atomic_set(&fs_info->balance_cancel_req, 0);
2161 fs_info->balance_ctl = NULL;
2162 init_waitqueue_head(&fs_info->balance_wait_q);
2165 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2166 struct btrfs_root *tree_root)
2168 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2169 set_nlink(fs_info->btree_inode, 1);
2171 * we set the i_size on the btree inode to the max possible int.
2172 * the real end of the address space is determined by all of
2173 * the devices in the system
2175 fs_info->btree_inode->i_size = OFFSET_MAX;
2176 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2178 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2179 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2180 fs_info->btree_inode->i_mapping);
2181 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2182 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2184 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2186 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2187 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2188 sizeof(struct btrfs_key));
2189 set_bit(BTRFS_INODE_DUMMY,
2190 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2191 btrfs_insert_inode_hash(fs_info->btree_inode);
2194 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2196 fs_info->dev_replace.lock_owner = 0;
2197 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2198 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2199 mutex_init(&fs_info->dev_replace.lock_management_lock);
2200 mutex_init(&fs_info->dev_replace.lock);
2201 init_waitqueue_head(&fs_info->replace_wait);
2204 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2206 spin_lock_init(&fs_info->qgroup_lock);
2207 mutex_init(&fs_info->qgroup_ioctl_lock);
2208 fs_info->qgroup_tree = RB_ROOT;
2209 fs_info->qgroup_op_tree = RB_ROOT;
2210 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2211 fs_info->qgroup_seq = 1;
2212 fs_info->quota_enabled = 0;
2213 fs_info->pending_quota_state = 0;
2214 fs_info->qgroup_ulist = NULL;
2215 mutex_init(&fs_info->qgroup_rescan_lock);
2218 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2219 struct btrfs_fs_devices *fs_devices)
2221 int max_active = fs_info->thread_pool_size;
2222 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2225 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2228 fs_info->delalloc_workers =
2229 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2231 fs_info->flush_workers =
2232 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2234 fs_info->caching_workers =
2235 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2238 * a higher idle thresh on the submit workers makes it much more
2239 * likely that bios will be send down in a sane order to the
2242 fs_info->submit_workers =
2243 btrfs_alloc_workqueue("submit", flags,
2244 min_t(u64, fs_devices->num_devices,
2247 fs_info->fixup_workers =
2248 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2251 * endios are largely parallel and should have a very
2254 fs_info->endio_workers =
2255 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2256 fs_info->endio_meta_workers =
2257 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2258 fs_info->endio_meta_write_workers =
2259 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2260 fs_info->endio_raid56_workers =
2261 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2262 fs_info->endio_repair_workers =
2263 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2264 fs_info->rmw_workers =
2265 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2266 fs_info->endio_write_workers =
2267 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2268 fs_info->endio_freespace_worker =
2269 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2270 fs_info->delayed_workers =
2271 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2272 fs_info->readahead_workers =
2273 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2274 fs_info->qgroup_rescan_workers =
2275 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2276 fs_info->extent_workers =
2277 btrfs_alloc_workqueue("extent-refs", flags,
2278 min_t(u64, fs_devices->num_devices,
2281 if (!(fs_info->workers && fs_info->delalloc_workers &&
2282 fs_info->submit_workers && fs_info->flush_workers &&
2283 fs_info->endio_workers && fs_info->endio_meta_workers &&
2284 fs_info->endio_meta_write_workers &&
2285 fs_info->endio_repair_workers &&
2286 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2287 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2288 fs_info->caching_workers && fs_info->readahead_workers &&
2289 fs_info->fixup_workers && fs_info->delayed_workers &&
2290 fs_info->extent_workers &&
2291 fs_info->qgroup_rescan_workers)) {
2298 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2299 struct btrfs_fs_devices *fs_devices)
2302 struct btrfs_root *tree_root = fs_info->tree_root;
2303 struct btrfs_root *log_tree_root;
2304 struct btrfs_super_block *disk_super = fs_info->super_copy;
2305 u64 bytenr = btrfs_super_log_root(disk_super);
2307 if (fs_devices->rw_devices == 0) {
2308 printk(KERN_WARNING "BTRFS: log replay required "
2313 log_tree_root = btrfs_alloc_root(fs_info);
2317 __setup_root(tree_root->nodesize, tree_root->sectorsize,
2318 tree_root->stripesize, log_tree_root, fs_info,
2319 BTRFS_TREE_LOG_OBJECTID);
2321 log_tree_root->node = read_tree_block(tree_root, bytenr,
2322 fs_info->generation + 1);
2323 if (!log_tree_root->node ||
2324 !extent_buffer_uptodate(log_tree_root->node)) {
2325 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2326 free_extent_buffer(log_tree_root->node);
2327 kfree(log_tree_root);
2330 /* returns with log_tree_root freed on success */
2331 ret = btrfs_recover_log_trees(log_tree_root);
2333 btrfs_error(tree_root->fs_info, ret,
2334 "Failed to recover log tree");
2335 free_extent_buffer(log_tree_root->node);
2336 kfree(log_tree_root);
2340 if (fs_info->sb->s_flags & MS_RDONLY) {
2341 ret = btrfs_commit_super(tree_root);
2349 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2350 struct btrfs_root *tree_root)
2352 struct btrfs_root *root;
2353 struct btrfs_key location;
2356 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2357 location.type = BTRFS_ROOT_ITEM_KEY;
2358 location.offset = 0;
2360 root = btrfs_read_tree_root(tree_root, &location);
2362 return PTR_ERR(root);
2363 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2364 fs_info->extent_root = root;
2366 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2367 root = btrfs_read_tree_root(tree_root, &location);
2369 return PTR_ERR(root);
2370 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2371 fs_info->dev_root = root;
2372 btrfs_init_devices_late(fs_info);
2374 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2375 root = btrfs_read_tree_root(tree_root, &location);
2377 return PTR_ERR(root);
2378 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2379 fs_info->csum_root = root;
2381 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2382 root = btrfs_read_tree_root(tree_root, &location);
2383 if (!IS_ERR(root)) {
2384 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2385 fs_info->quota_enabled = 1;
2386 fs_info->pending_quota_state = 1;
2387 fs_info->quota_root = root;
2390 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2391 root = btrfs_read_tree_root(tree_root, &location);
2393 ret = PTR_ERR(root);
2397 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2398 fs_info->uuid_root = root;
2404 int open_ctree(struct super_block *sb,
2405 struct btrfs_fs_devices *fs_devices,
2413 struct btrfs_key location;
2414 struct buffer_head *bh;
2415 struct btrfs_super_block *disk_super;
2416 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2417 struct btrfs_root *tree_root;
2418 struct btrfs_root *chunk_root;
2421 int num_backups_tried = 0;
2422 int backup_index = 0;
2425 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2426 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2427 if (!tree_root || !chunk_root) {
2432 ret = init_srcu_struct(&fs_info->subvol_srcu);
2438 ret = setup_bdi(fs_info, &fs_info->bdi);
2444 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2449 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2450 (1 + ilog2(nr_cpu_ids));
2452 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2455 goto fail_dirty_metadata_bytes;
2458 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2461 goto fail_delalloc_bytes;
2464 fs_info->btree_inode = new_inode(sb);
2465 if (!fs_info->btree_inode) {
2467 goto fail_bio_counter;
2470 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2472 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2473 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2474 INIT_LIST_HEAD(&fs_info->trans_list);
2475 INIT_LIST_HEAD(&fs_info->dead_roots);
2476 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2477 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2478 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2479 spin_lock_init(&fs_info->delalloc_root_lock);
2480 spin_lock_init(&fs_info->trans_lock);
2481 spin_lock_init(&fs_info->fs_roots_radix_lock);
2482 spin_lock_init(&fs_info->delayed_iput_lock);
2483 spin_lock_init(&fs_info->defrag_inodes_lock);
2484 spin_lock_init(&fs_info->free_chunk_lock);
2485 spin_lock_init(&fs_info->tree_mod_seq_lock);
2486 spin_lock_init(&fs_info->super_lock);
2487 spin_lock_init(&fs_info->qgroup_op_lock);
2488 spin_lock_init(&fs_info->buffer_lock);
2489 spin_lock_init(&fs_info->unused_bgs_lock);
2490 mutex_init(&fs_info->unused_bg_unpin_mutex);
2491 rwlock_init(&fs_info->tree_mod_log_lock);
2492 mutex_init(&fs_info->reloc_mutex);
2493 mutex_init(&fs_info->delalloc_root_mutex);
2494 seqlock_init(&fs_info->profiles_lock);
2496 init_completion(&fs_info->kobj_unregister);
2497 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2498 INIT_LIST_HEAD(&fs_info->space_info);
2499 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2500 INIT_LIST_HEAD(&fs_info->unused_bgs);
2501 btrfs_mapping_init(&fs_info->mapping_tree);
2502 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2503 BTRFS_BLOCK_RSV_GLOBAL);
2504 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2505 BTRFS_BLOCK_RSV_DELALLOC);
2506 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2507 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2508 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2509 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2510 BTRFS_BLOCK_RSV_DELOPS);
2511 atomic_set(&fs_info->nr_async_submits, 0);
2512 atomic_set(&fs_info->async_delalloc_pages, 0);
2513 atomic_set(&fs_info->async_submit_draining, 0);
2514 atomic_set(&fs_info->nr_async_bios, 0);
2515 atomic_set(&fs_info->defrag_running, 0);
2516 atomic_set(&fs_info->qgroup_op_seq, 0);
2517 atomic64_set(&fs_info->tree_mod_seq, 0);
2519 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2520 fs_info->metadata_ratio = 0;
2521 fs_info->defrag_inodes = RB_ROOT;
2522 fs_info->free_chunk_space = 0;
2523 fs_info->tree_mod_log = RB_ROOT;
2524 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2525 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2526 /* readahead state */
2527 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2528 spin_lock_init(&fs_info->reada_lock);
2530 fs_info->thread_pool_size = min_t(unsigned long,
2531 num_online_cpus() + 2, 8);
2533 INIT_LIST_HEAD(&fs_info->ordered_roots);
2534 spin_lock_init(&fs_info->ordered_root_lock);
2535 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2537 if (!fs_info->delayed_root) {
2541 btrfs_init_delayed_root(fs_info->delayed_root);
2543 btrfs_init_scrub(fs_info);
2544 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2545 fs_info->check_integrity_print_mask = 0;
2547 btrfs_init_balance(fs_info);
2548 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2550 sb->s_blocksize = 4096;
2551 sb->s_blocksize_bits = blksize_bits(4096);
2552 sb->s_bdi = &fs_info->bdi;
2554 btrfs_init_btree_inode(fs_info, tree_root);
2556 spin_lock_init(&fs_info->block_group_cache_lock);
2557 fs_info->block_group_cache_tree = RB_ROOT;
2558 fs_info->first_logical_byte = (u64)-1;
2560 extent_io_tree_init(&fs_info->freed_extents[0],
2561 fs_info->btree_inode->i_mapping);
2562 extent_io_tree_init(&fs_info->freed_extents[1],
2563 fs_info->btree_inode->i_mapping);
2564 fs_info->pinned_extents = &fs_info->freed_extents[0];
2565 fs_info->do_barriers = 1;
2568 mutex_init(&fs_info->ordered_operations_mutex);
2569 mutex_init(&fs_info->ordered_extent_flush_mutex);
2570 mutex_init(&fs_info->tree_log_mutex);
2571 mutex_init(&fs_info->chunk_mutex);
2572 mutex_init(&fs_info->transaction_kthread_mutex);
2573 mutex_init(&fs_info->cleaner_mutex);
2574 mutex_init(&fs_info->volume_mutex);
2575 init_rwsem(&fs_info->commit_root_sem);
2576 init_rwsem(&fs_info->cleanup_work_sem);
2577 init_rwsem(&fs_info->subvol_sem);
2578 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2580 btrfs_init_dev_replace_locks(fs_info);
2581 btrfs_init_qgroup(fs_info);
2583 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2584 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2586 init_waitqueue_head(&fs_info->transaction_throttle);
2587 init_waitqueue_head(&fs_info->transaction_wait);
2588 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2589 init_waitqueue_head(&fs_info->async_submit_wait);
2591 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2593 ret = btrfs_alloc_stripe_hash_table(fs_info);
2599 __setup_root(4096, 4096, 4096, tree_root,
2600 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2602 invalidate_bdev(fs_devices->latest_bdev);
2605 * Read super block and check the signature bytes only
2607 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2614 * We want to check superblock checksum, the type is stored inside.
2615 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2617 if (btrfs_check_super_csum(bh->b_data)) {
2618 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2624 * super_copy is zeroed at allocation time and we never touch the
2625 * following bytes up to INFO_SIZE, the checksum is calculated from
2626 * the whole block of INFO_SIZE
2628 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2629 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2630 sizeof(*fs_info->super_for_commit));
2633 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2635 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2637 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2642 disk_super = fs_info->super_copy;
2643 if (!btrfs_super_root(disk_super))
2646 /* check FS state, whether FS is broken. */
2647 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2648 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2651 * run through our array of backup supers and setup
2652 * our ring pointer to the oldest one
2654 generation = btrfs_super_generation(disk_super);
2655 find_oldest_super_backup(fs_info, generation);
2658 * In the long term, we'll store the compression type in the super
2659 * block, and it'll be used for per file compression control.
2661 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2663 ret = btrfs_parse_options(tree_root, options);
2669 features = btrfs_super_incompat_flags(disk_super) &
2670 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2672 printk(KERN_ERR "BTRFS: couldn't mount because of "
2673 "unsupported optional features (%Lx).\n",
2680 * Leafsize and nodesize were always equal, this is only a sanity check.
2682 if (le32_to_cpu(disk_super->__unused_leafsize) !=
2683 btrfs_super_nodesize(disk_super)) {
2684 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2685 "blocksizes don't match. node %d leaf %d\n",
2686 btrfs_super_nodesize(disk_super),
2687 le32_to_cpu(disk_super->__unused_leafsize));
2691 if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2692 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2693 "blocksize (%d) was too large\n",
2694 btrfs_super_nodesize(disk_super));
2699 features = btrfs_super_incompat_flags(disk_super);
2700 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2701 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2702 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2704 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2705 printk(KERN_INFO "BTRFS: has skinny extents\n");
2708 * flag our filesystem as having big metadata blocks if
2709 * they are bigger than the page size
2711 if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2712 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2713 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2714 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2717 nodesize = btrfs_super_nodesize(disk_super);
2718 sectorsize = btrfs_super_sectorsize(disk_super);
2719 stripesize = btrfs_super_stripesize(disk_super);
2720 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2721 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2724 * mixed block groups end up with duplicate but slightly offset
2725 * extent buffers for the same range. It leads to corruptions
2727 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2728 (sectorsize != nodesize)) {
2729 printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2730 "are not allowed for mixed block groups on %s\n",
2736 * Needn't use the lock because there is no other task which will
2739 btrfs_set_super_incompat_flags(disk_super, features);
2741 features = btrfs_super_compat_ro_flags(disk_super) &
2742 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2743 if (!(sb->s_flags & MS_RDONLY) && features) {
2744 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2745 "unsupported option features (%Lx).\n",
2751 max_active = fs_info->thread_pool_size;
2753 ret = btrfs_init_workqueues(fs_info, fs_devices);
2756 goto fail_sb_buffer;
2759 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2760 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2761 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2763 tree_root->nodesize = nodesize;
2764 tree_root->sectorsize = sectorsize;
2765 tree_root->stripesize = stripesize;
2767 sb->s_blocksize = sectorsize;
2768 sb->s_blocksize_bits = blksize_bits(sectorsize);
2770 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2771 printk(KERN_ERR "BTRFS: valid FS not found on %s\n", sb->s_id);
2772 goto fail_sb_buffer;
2775 if (sectorsize != PAGE_SIZE) {
2776 printk(KERN_ERR "BTRFS: incompatible sector size (%lu) "
2777 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2778 goto fail_sb_buffer;
2781 mutex_lock(&fs_info->chunk_mutex);
2782 ret = btrfs_read_sys_array(tree_root);
2783 mutex_unlock(&fs_info->chunk_mutex);
2785 printk(KERN_ERR "BTRFS: failed to read the system "
2786 "array on %s\n", sb->s_id);
2787 goto fail_sb_buffer;
2790 generation = btrfs_super_chunk_root_generation(disk_super);
2792 __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2793 fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2795 chunk_root->node = read_tree_block(chunk_root,
2796 btrfs_super_chunk_root(disk_super),
2798 if (!chunk_root->node ||
2799 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2800 printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2802 goto fail_tree_roots;
2804 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2805 chunk_root->commit_root = btrfs_root_node(chunk_root);
2807 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2808 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2810 ret = btrfs_read_chunk_tree(chunk_root);
2812 printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2814 goto fail_tree_roots;
2818 * keep the device that is marked to be the target device for the
2819 * dev_replace procedure
2821 btrfs_close_extra_devices(fs_devices, 0);
2823 if (!fs_devices->latest_bdev) {
2824 printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2826 goto fail_tree_roots;
2830 generation = btrfs_super_generation(disk_super);
2832 tree_root->node = read_tree_block(tree_root,
2833 btrfs_super_root(disk_super),
2835 if (!tree_root->node ||
2836 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2837 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2840 goto recovery_tree_root;
2843 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2844 tree_root->commit_root = btrfs_root_node(tree_root);
2845 btrfs_set_root_refs(&tree_root->root_item, 1);
2847 ret = btrfs_read_roots(fs_info, tree_root);
2849 goto recovery_tree_root;
2851 fs_info->generation = generation;
2852 fs_info->last_trans_committed = generation;
2854 ret = btrfs_recover_balance(fs_info);
2856 printk(KERN_ERR "BTRFS: failed to recover balance\n");
2857 goto fail_block_groups;
2860 ret = btrfs_init_dev_stats(fs_info);
2862 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2864 goto fail_block_groups;
2867 ret = btrfs_init_dev_replace(fs_info);
2869 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2870 goto fail_block_groups;
2873 btrfs_close_extra_devices(fs_devices, 1);
2875 ret = btrfs_sysfs_add_one(fs_info);
2877 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2878 goto fail_block_groups;
2881 ret = btrfs_init_space_info(fs_info);
2883 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2887 ret = btrfs_read_block_groups(fs_info->extent_root);
2889 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2892 fs_info->num_tolerated_disk_barrier_failures =
2893 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2894 if (fs_info->fs_devices->missing_devices >
2895 fs_info->num_tolerated_disk_barrier_failures &&
2896 !(sb->s_flags & MS_RDONLY)) {
2897 printk(KERN_WARNING "BTRFS: "
2898 "too many missing devices, writeable mount is not allowed\n");
2902 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2904 if (IS_ERR(fs_info->cleaner_kthread))
2907 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2909 "btrfs-transaction");
2910 if (IS_ERR(fs_info->transaction_kthread))
2913 if (!btrfs_test_opt(tree_root, SSD) &&
2914 !btrfs_test_opt(tree_root, NOSSD) &&
2915 !fs_info->fs_devices->rotating) {
2916 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2918 btrfs_set_opt(fs_info->mount_opt, SSD);
2922 * Mount does not set all options immediatelly, we can do it now and do
2923 * not have to wait for transaction commit
2925 btrfs_apply_pending_changes(fs_info);
2927 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2928 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2929 ret = btrfsic_mount(tree_root, fs_devices,
2930 btrfs_test_opt(tree_root,
2931 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2933 fs_info->check_integrity_print_mask);
2935 printk(KERN_WARNING "BTRFS: failed to initialize"
2936 " integrity check module %s\n", sb->s_id);
2939 ret = btrfs_read_qgroup_config(fs_info);
2941 goto fail_trans_kthread;
2943 /* do not make disk changes in broken FS */
2944 if (btrfs_super_log_root(disk_super) != 0) {
2945 ret = btrfs_replay_log(fs_info, fs_devices);
2952 ret = btrfs_find_orphan_roots(tree_root);
2956 if (!(sb->s_flags & MS_RDONLY)) {
2957 ret = btrfs_cleanup_fs_roots(fs_info);
2961 mutex_lock(&fs_info->cleaner_mutex);
2962 ret = btrfs_recover_relocation(tree_root);
2963 mutex_unlock(&fs_info->cleaner_mutex);
2966 "BTRFS: failed to recover relocation\n");
2972 location.objectid = BTRFS_FS_TREE_OBJECTID;
2973 location.type = BTRFS_ROOT_ITEM_KEY;
2974 location.offset = 0;
2976 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2977 if (IS_ERR(fs_info->fs_root)) {
2978 err = PTR_ERR(fs_info->fs_root);
2982 if (sb->s_flags & MS_RDONLY)
2985 down_read(&fs_info->cleanup_work_sem);
2986 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2987 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2988 up_read(&fs_info->cleanup_work_sem);
2989 close_ctree(tree_root);
2992 up_read(&fs_info->cleanup_work_sem);
2994 ret = btrfs_resume_balance_async(fs_info);
2996 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2997 close_ctree(tree_root);
3001 ret = btrfs_resume_dev_replace_async(fs_info);
3003 pr_warn("BTRFS: failed to resume dev_replace\n");
3004 close_ctree(tree_root);
3008 btrfs_qgroup_rescan_resume(fs_info);
3010 if (!fs_info->uuid_root) {
3011 pr_info("BTRFS: creating UUID tree\n");
3012 ret = btrfs_create_uuid_tree(fs_info);
3014 pr_warn("BTRFS: failed to create the UUID tree %d\n",
3016 close_ctree(tree_root);
3019 } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3020 fs_info->generation !=
3021 btrfs_super_uuid_tree_generation(disk_super)) {
3022 pr_info("BTRFS: checking UUID tree\n");
3023 ret = btrfs_check_uuid_tree(fs_info);
3025 pr_warn("BTRFS: failed to check the UUID tree %d\n",
3027 close_ctree(tree_root);
3031 fs_info->update_uuid_tree_gen = 1;
3039 btrfs_free_qgroup_config(fs_info);
3041 kthread_stop(fs_info->transaction_kthread);
3042 btrfs_cleanup_transaction(fs_info->tree_root);
3043 btrfs_free_fs_roots(fs_info);
3045 kthread_stop(fs_info->cleaner_kthread);
3048 * make sure we're done with the btree inode before we stop our
3051 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3054 btrfs_sysfs_remove_one(fs_info);
3057 btrfs_put_block_group_cache(fs_info);
3058 btrfs_free_block_groups(fs_info);
3061 free_root_pointers(fs_info, 1);
3062 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3065 btrfs_stop_all_workers(fs_info);
3068 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3070 iput(fs_info->btree_inode);
3072 percpu_counter_destroy(&fs_info->bio_counter);
3073 fail_delalloc_bytes:
3074 percpu_counter_destroy(&fs_info->delalloc_bytes);
3075 fail_dirty_metadata_bytes:
3076 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3078 bdi_destroy(&fs_info->bdi);
3080 cleanup_srcu_struct(&fs_info->subvol_srcu);
3082 btrfs_free_stripe_hash_table(fs_info);
3083 btrfs_close_devices(fs_info->fs_devices);
3087 if (!btrfs_test_opt(tree_root, RECOVERY))
3088 goto fail_tree_roots;
3090 free_root_pointers(fs_info, 0);
3092 /* don't use the log in recovery mode, it won't be valid */
3093 btrfs_set_super_log_root(disk_super, 0);
3095 /* we can't trust the free space cache either */
3096 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3098 ret = next_root_backup(fs_info, fs_info->super_copy,
3099 &num_backups_tried, &backup_index);
3101 goto fail_block_groups;
3102 goto retry_root_backup;
3105 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3108 set_buffer_uptodate(bh);
3110 struct btrfs_device *device = (struct btrfs_device *)
3113 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3114 "I/O error on %s\n",
3115 rcu_str_deref(device->name));
3116 /* note, we dont' set_buffer_write_io_error because we have
3117 * our own ways of dealing with the IO errors
3119 clear_buffer_uptodate(bh);
3120 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3126 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3128 struct buffer_head *bh;
3129 struct buffer_head *latest = NULL;
3130 struct btrfs_super_block *super;
3135 /* we would like to check all the supers, but that would make
3136 * a btrfs mount succeed after a mkfs from a different FS.
3137 * So, we need to add a special mount option to scan for
3138 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3140 for (i = 0; i < 1; i++) {
3141 bytenr = btrfs_sb_offset(i);
3142 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3143 i_size_read(bdev->bd_inode))
3145 bh = __bread(bdev, bytenr / 4096,
3146 BTRFS_SUPER_INFO_SIZE);
3150 super = (struct btrfs_super_block *)bh->b_data;
3151 if (btrfs_super_bytenr(super) != bytenr ||
3152 btrfs_super_magic(super) != BTRFS_MAGIC) {
3157 if (!latest || btrfs_super_generation(super) > transid) {
3160 transid = btrfs_super_generation(super);
3169 * this should be called twice, once with wait == 0 and
3170 * once with wait == 1. When wait == 0 is done, all the buffer heads
3171 * we write are pinned.
3173 * They are released when wait == 1 is done.
3174 * max_mirrors must be the same for both runs, and it indicates how
3175 * many supers on this one device should be written.
3177 * max_mirrors == 0 means to write them all.
3179 static int write_dev_supers(struct btrfs_device *device,
3180 struct btrfs_super_block *sb,
3181 int do_barriers, int wait, int max_mirrors)
3183 struct buffer_head *bh;
3190 if (max_mirrors == 0)
3191 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3193 for (i = 0; i < max_mirrors; i++) {
3194 bytenr = btrfs_sb_offset(i);
3195 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3196 device->commit_total_bytes)
3200 bh = __find_get_block(device->bdev, bytenr / 4096,
3201 BTRFS_SUPER_INFO_SIZE);
3207 if (!buffer_uptodate(bh))
3210 /* drop our reference */
3213 /* drop the reference from the wait == 0 run */
3217 btrfs_set_super_bytenr(sb, bytenr);
3220 crc = btrfs_csum_data((char *)sb +
3221 BTRFS_CSUM_SIZE, crc,
3222 BTRFS_SUPER_INFO_SIZE -
3224 btrfs_csum_final(crc, sb->csum);
3227 * one reference for us, and we leave it for the
3230 bh = __getblk(device->bdev, bytenr / 4096,
3231 BTRFS_SUPER_INFO_SIZE);
3233 printk(KERN_ERR "BTRFS: couldn't get super "
3234 "buffer head for bytenr %Lu\n", bytenr);
3239 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3241 /* one reference for submit_bh */
3244 set_buffer_uptodate(bh);
3246 bh->b_end_io = btrfs_end_buffer_write_sync;
3247 bh->b_private = device;
3251 * we fua the first super. The others we allow
3255 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3257 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3261 return errors < i ? 0 : -1;
3265 * endio for the write_dev_flush, this will wake anyone waiting
3266 * for the barrier when it is done
3268 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3271 if (err == -EOPNOTSUPP)
3272 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3273 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3275 if (bio->bi_private)
3276 complete(bio->bi_private);
3281 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3282 * sent down. With wait == 1, it waits for the previous flush.
3284 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3287 static int write_dev_flush(struct btrfs_device *device, int wait)
3292 if (device->nobarriers)
3296 bio = device->flush_bio;
3300 wait_for_completion(&device->flush_wait);
3302 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3303 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3304 rcu_str_deref(device->name));
3305 device->nobarriers = 1;
3306 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3308 btrfs_dev_stat_inc_and_print(device,
3309 BTRFS_DEV_STAT_FLUSH_ERRS);
3312 /* drop the reference from the wait == 0 run */
3314 device->flush_bio = NULL;
3320 * one reference for us, and we leave it for the
3323 device->flush_bio = NULL;
3324 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3328 bio->bi_end_io = btrfs_end_empty_barrier;
3329 bio->bi_bdev = device->bdev;
3330 init_completion(&device->flush_wait);
3331 bio->bi_private = &device->flush_wait;
3332 device->flush_bio = bio;
3335 btrfsic_submit_bio(WRITE_FLUSH, bio);
3341 * send an empty flush down to each device in parallel,
3342 * then wait for them
3344 static int barrier_all_devices(struct btrfs_fs_info *info)
3346 struct list_head *head;
3347 struct btrfs_device *dev;
3348 int errors_send = 0;
3349 int errors_wait = 0;
3352 /* send down all the barriers */
3353 head = &info->fs_devices->devices;
3354 list_for_each_entry_rcu(dev, head, dev_list) {
3361 if (!dev->in_fs_metadata || !dev->writeable)
3364 ret = write_dev_flush(dev, 0);
3369 /* wait for all the barriers */
3370 list_for_each_entry_rcu(dev, head, dev_list) {
3377 if (!dev->in_fs_metadata || !dev->writeable)
3380 ret = write_dev_flush(dev, 1);
3384 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3385 errors_wait > info->num_tolerated_disk_barrier_failures)
3390 int btrfs_calc_num_tolerated_disk_barrier_failures(
3391 struct btrfs_fs_info *fs_info)
3393 struct btrfs_ioctl_space_info space;
3394 struct btrfs_space_info *sinfo;
3395 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3396 BTRFS_BLOCK_GROUP_SYSTEM,
3397 BTRFS_BLOCK_GROUP_METADATA,
3398 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3402 int num_tolerated_disk_barrier_failures =
3403 (int)fs_info->fs_devices->num_devices;
3405 for (i = 0; i < num_types; i++) {
3406 struct btrfs_space_info *tmp;
3410 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3411 if (tmp->flags == types[i]) {
3421 down_read(&sinfo->groups_sem);
3422 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3423 if (!list_empty(&sinfo->block_groups[c])) {
3426 btrfs_get_block_group_info(
3427 &sinfo->block_groups[c], &space);
3428 if (space.total_bytes == 0 ||
3429 space.used_bytes == 0)
3431 flags = space.flags;
3434 * 0: if dup, single or RAID0 is configured for
3435 * any of metadata, system or data, else
3436 * 1: if RAID5 is configured, or if RAID1 or
3437 * RAID10 is configured and only two mirrors
3439 * 2: if RAID6 is configured, else
3440 * num_mirrors - 1: if RAID1 or RAID10 is
3441 * configured and more than
3442 * 2 mirrors are used.
3444 if (num_tolerated_disk_barrier_failures > 0 &&
3445 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3446 BTRFS_BLOCK_GROUP_RAID0)) ||
3447 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3449 num_tolerated_disk_barrier_failures = 0;
3450 else if (num_tolerated_disk_barrier_failures > 1) {
3451 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3452 BTRFS_BLOCK_GROUP_RAID5 |
3453 BTRFS_BLOCK_GROUP_RAID10)) {
3454 num_tolerated_disk_barrier_failures = 1;
3456 BTRFS_BLOCK_GROUP_RAID6) {
3457 num_tolerated_disk_barrier_failures = 2;
3462 up_read(&sinfo->groups_sem);
3465 return num_tolerated_disk_barrier_failures;
3468 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3470 struct list_head *head;
3471 struct btrfs_device *dev;
3472 struct btrfs_super_block *sb;
3473 struct btrfs_dev_item *dev_item;
3477 int total_errors = 0;
3480 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3481 backup_super_roots(root->fs_info);
3483 sb = root->fs_info->super_for_commit;
3484 dev_item = &sb->dev_item;
3486 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3487 head = &root->fs_info->fs_devices->devices;
3488 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3491 ret = barrier_all_devices(root->fs_info);
3494 &root->fs_info->fs_devices->device_list_mutex);
3495 btrfs_error(root->fs_info, ret,
3496 "errors while submitting device barriers.");
3501 list_for_each_entry_rcu(dev, head, dev_list) {
3506 if (!dev->in_fs_metadata || !dev->writeable)
3509 btrfs_set_stack_device_generation(dev_item, 0);
3510 btrfs_set_stack_device_type(dev_item, dev->type);
3511 btrfs_set_stack_device_id(dev_item, dev->devid);
3512 btrfs_set_stack_device_total_bytes(dev_item,
3513 dev->commit_total_bytes);
3514 btrfs_set_stack_device_bytes_used(dev_item,
3515 dev->commit_bytes_used);
3516 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3517 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3518 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3519 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3520 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3522 flags = btrfs_super_flags(sb);
3523 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3525 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3529 if (total_errors > max_errors) {
3530 btrfs_err(root->fs_info, "%d errors while writing supers",
3532 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3534 /* FUA is masked off if unsupported and can't be the reason */
3535 btrfs_error(root->fs_info, -EIO,
3536 "%d errors while writing supers", total_errors);
3541 list_for_each_entry_rcu(dev, head, dev_list) {
3544 if (!dev->in_fs_metadata || !dev->writeable)
3547 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3551 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3552 if (total_errors > max_errors) {
3553 btrfs_error(root->fs_info, -EIO,
3554 "%d errors while writing supers", total_errors);
3560 int write_ctree_super(struct btrfs_trans_handle *trans,
3561 struct btrfs_root *root, int max_mirrors)
3563 return write_all_supers(root, max_mirrors);
3566 /* Drop a fs root from the radix tree and free it. */
3567 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3568 struct btrfs_root *root)
3570 spin_lock(&fs_info->fs_roots_radix_lock);
3571 radix_tree_delete(&fs_info->fs_roots_radix,
3572 (unsigned long)root->root_key.objectid);
3573 spin_unlock(&fs_info->fs_roots_radix_lock);
3575 if (btrfs_root_refs(&root->root_item) == 0)
3576 synchronize_srcu(&fs_info->subvol_srcu);
3578 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3579 btrfs_free_log(NULL, root);
3581 if (root->free_ino_pinned)
3582 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3583 if (root->free_ino_ctl)
3584 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3588 static void free_fs_root(struct btrfs_root *root)
3590 iput(root->ino_cache_inode);
3591 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3592 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3593 root->orphan_block_rsv = NULL;
3595 free_anon_bdev(root->anon_dev);
3596 if (root->subv_writers)
3597 btrfs_free_subvolume_writers(root->subv_writers);
3598 free_extent_buffer(root->node);
3599 free_extent_buffer(root->commit_root);
3600 kfree(root->free_ino_ctl);
3601 kfree(root->free_ino_pinned);
3603 btrfs_put_fs_root(root);
3606 void btrfs_free_fs_root(struct btrfs_root *root)
3611 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3613 u64 root_objectid = 0;
3614 struct btrfs_root *gang[8];
3617 unsigned int ret = 0;
3621 index = srcu_read_lock(&fs_info->subvol_srcu);
3622 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3623 (void **)gang, root_objectid,
3626 srcu_read_unlock(&fs_info->subvol_srcu, index);
3629 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3631 for (i = 0; i < ret; i++) {
3632 /* Avoid to grab roots in dead_roots */
3633 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3637 /* grab all the search result for later use */
3638 gang[i] = btrfs_grab_fs_root(gang[i]);
3640 srcu_read_unlock(&fs_info->subvol_srcu, index);
3642 for (i = 0; i < ret; i++) {
3645 root_objectid = gang[i]->root_key.objectid;
3646 err = btrfs_orphan_cleanup(gang[i]);
3649 btrfs_put_fs_root(gang[i]);
3654 /* release the uncleaned roots due to error */
3655 for (; i < ret; i++) {
3657 btrfs_put_fs_root(gang[i]);
3662 int btrfs_commit_super(struct btrfs_root *root)
3664 struct btrfs_trans_handle *trans;
3666 mutex_lock(&root->fs_info->cleaner_mutex);
3667 btrfs_run_delayed_iputs(root);
3668 mutex_unlock(&root->fs_info->cleaner_mutex);
3669 wake_up_process(root->fs_info->cleaner_kthread);
3671 /* wait until ongoing cleanup work done */
3672 down_write(&root->fs_info->cleanup_work_sem);
3673 up_write(&root->fs_info->cleanup_work_sem);
3675 trans = btrfs_join_transaction(root);
3677 return PTR_ERR(trans);
3678 return btrfs_commit_transaction(trans, root);
3681 void close_ctree(struct btrfs_root *root)
3683 struct btrfs_fs_info *fs_info = root->fs_info;
3686 fs_info->closing = 1;
3689 /* wait for the uuid_scan task to finish */
3690 down(&fs_info->uuid_tree_rescan_sem);
3691 /* avoid complains from lockdep et al., set sem back to initial state */
3692 up(&fs_info->uuid_tree_rescan_sem);
3694 /* pause restriper - we want to resume on mount */
3695 btrfs_pause_balance(fs_info);
3697 btrfs_dev_replace_suspend_for_unmount(fs_info);
3699 btrfs_scrub_cancel(fs_info);
3701 /* wait for any defraggers to finish */
3702 wait_event(fs_info->transaction_wait,
3703 (atomic_read(&fs_info->defrag_running) == 0));
3705 /* clear out the rbtree of defraggable inodes */
3706 btrfs_cleanup_defrag_inodes(fs_info);
3708 cancel_work_sync(&fs_info->async_reclaim_work);
3710 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3711 ret = btrfs_commit_super(root);
3713 btrfs_err(fs_info, "commit super ret %d", ret);
3716 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3717 btrfs_error_commit_super(root);
3719 kthread_stop(fs_info->transaction_kthread);
3720 kthread_stop(fs_info->cleaner_kthread);
3722 fs_info->closing = 2;
3725 btrfs_free_qgroup_config(fs_info);
3727 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3728 btrfs_info(fs_info, "at unmount delalloc count %lld",
3729 percpu_counter_sum(&fs_info->delalloc_bytes));
3732 btrfs_sysfs_remove_one(fs_info);
3734 btrfs_free_fs_roots(fs_info);
3736 btrfs_put_block_group_cache(fs_info);
3738 btrfs_free_block_groups(fs_info);
3741 * we must make sure there is not any read request to
3742 * submit after we stopping all workers.
3744 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3745 btrfs_stop_all_workers(fs_info);
3748 free_root_pointers(fs_info, 1);
3750 iput(fs_info->btree_inode);
3752 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3753 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3754 btrfsic_unmount(root, fs_info->fs_devices);
3757 btrfs_close_devices(fs_info->fs_devices);
3758 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3760 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3761 percpu_counter_destroy(&fs_info->delalloc_bytes);
3762 percpu_counter_destroy(&fs_info->bio_counter);
3763 bdi_destroy(&fs_info->bdi);
3764 cleanup_srcu_struct(&fs_info->subvol_srcu);
3766 btrfs_free_stripe_hash_table(fs_info);
3768 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3769 root->orphan_block_rsv = NULL;
3772 while (!list_empty(&fs_info->pinned_chunks)) {
3773 struct extent_map *em;
3775 em = list_first_entry(&fs_info->pinned_chunks,
3776 struct extent_map, list);
3777 list_del_init(&em->list);
3778 free_extent_map(em);
3780 unlock_chunks(root);
3783 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3787 struct inode *btree_inode = buf->pages[0]->mapping->host;
3789 ret = extent_buffer_uptodate(buf);
3793 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3794 parent_transid, atomic);
3800 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3802 return set_extent_buffer_uptodate(buf);
3805 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3807 struct btrfs_root *root;
3808 u64 transid = btrfs_header_generation(buf);
3811 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3813 * This is a fast path so only do this check if we have sanity tests
3814 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3815 * outside of the sanity tests.
3817 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3820 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3821 btrfs_assert_tree_locked(buf);
3822 if (transid != root->fs_info->generation)
3823 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3824 "found %llu running %llu\n",
3825 buf->start, transid, root->fs_info->generation);
3826 was_dirty = set_extent_buffer_dirty(buf);
3828 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3830 root->fs_info->dirty_metadata_batch);
3831 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3832 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3833 btrfs_print_leaf(root, buf);
3839 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3843 * looks as though older kernels can get into trouble with
3844 * this code, they end up stuck in balance_dirty_pages forever
3848 if (current->flags & PF_MEMALLOC)
3852 btrfs_balance_delayed_items(root);
3854 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3855 BTRFS_DIRTY_METADATA_THRESH);
3857 balance_dirty_pages_ratelimited(
3858 root->fs_info->btree_inode->i_mapping);
3863 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3865 __btrfs_btree_balance_dirty(root, 1);
3868 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3870 __btrfs_btree_balance_dirty(root, 0);
3873 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3875 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3876 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3879 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3882 struct btrfs_super_block *sb = fs_info->super_copy;
3885 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3886 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
3887 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3890 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3891 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
3892 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3895 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3896 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
3897 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3902 * The common minimum, we don't know if we can trust the nodesize/sectorsize
3903 * items yet, they'll be verified later. Issue just a warning.
3905 if (!IS_ALIGNED(btrfs_super_root(sb), 4096))
3906 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
3907 btrfs_super_root(sb));
3908 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096))
3909 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
3910 btrfs_super_chunk_root(sb));
3911 if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096))
3912 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
3913 btrfs_super_log_root(sb));
3916 * Check the lower bound, the alignment and other constraints are
3919 if (btrfs_super_nodesize(sb) < 4096) {
3920 printk(KERN_ERR "BTRFS: nodesize too small: %u < 4096\n",
3921 btrfs_super_nodesize(sb));
3924 if (btrfs_super_sectorsize(sb) < 4096) {
3925 printk(KERN_ERR "BTRFS: sectorsize too small: %u < 4096\n",
3926 btrfs_super_sectorsize(sb));
3930 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
3931 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
3932 fs_info->fsid, sb->dev_item.fsid);
3937 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3940 if (btrfs_super_num_devices(sb) > (1UL << 31))
3941 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
3942 btrfs_super_num_devices(sb));
3943 if (btrfs_super_num_devices(sb) == 0) {
3944 printk(KERN_ERR "BTRFS: number of devices is 0\n");
3948 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
3949 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
3950 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
3955 * Obvious sys_chunk_array corruptions, it must hold at least one key
3958 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
3959 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
3960 btrfs_super_sys_array_size(sb),
3961 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
3964 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
3965 + sizeof(struct btrfs_chunk)) {
3966 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
3967 btrfs_super_sys_array_size(sb),
3968 sizeof(struct btrfs_disk_key)
3969 + sizeof(struct btrfs_chunk));
3974 * The generation is a global counter, we'll trust it more than the others
3975 * but it's still possible that it's the one that's wrong.
3977 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
3979 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
3980 btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
3981 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
3982 && btrfs_super_cache_generation(sb) != (u64)-1)
3984 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
3985 btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
3990 static void btrfs_error_commit_super(struct btrfs_root *root)
3992 mutex_lock(&root->fs_info->cleaner_mutex);
3993 btrfs_run_delayed_iputs(root);
3994 mutex_unlock(&root->fs_info->cleaner_mutex);
3996 down_write(&root->fs_info->cleanup_work_sem);
3997 up_write(&root->fs_info->cleanup_work_sem);
3999 /* cleanup FS via transaction */
4000 btrfs_cleanup_transaction(root);
4003 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4005 struct btrfs_ordered_extent *ordered;
4007 spin_lock(&root->ordered_extent_lock);
4009 * This will just short circuit the ordered completion stuff which will
4010 * make sure the ordered extent gets properly cleaned up.
4012 list_for_each_entry(ordered, &root->ordered_extents,
4014 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4015 spin_unlock(&root->ordered_extent_lock);
4018 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4020 struct btrfs_root *root;
4021 struct list_head splice;
4023 INIT_LIST_HEAD(&splice);
4025 spin_lock(&fs_info->ordered_root_lock);
4026 list_splice_init(&fs_info->ordered_roots, &splice);
4027 while (!list_empty(&splice)) {
4028 root = list_first_entry(&splice, struct btrfs_root,
4030 list_move_tail(&root->ordered_root,
4031 &fs_info->ordered_roots);
4033 spin_unlock(&fs_info->ordered_root_lock);
4034 btrfs_destroy_ordered_extents(root);
4037 spin_lock(&fs_info->ordered_root_lock);
4039 spin_unlock(&fs_info->ordered_root_lock);
4042 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4043 struct btrfs_root *root)
4045 struct rb_node *node;
4046 struct btrfs_delayed_ref_root *delayed_refs;
4047 struct btrfs_delayed_ref_node *ref;
4050 delayed_refs = &trans->delayed_refs;
4052 spin_lock(&delayed_refs->lock);
4053 if (atomic_read(&delayed_refs->num_entries) == 0) {
4054 spin_unlock(&delayed_refs->lock);
4055 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4059 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4060 struct btrfs_delayed_ref_head *head;
4061 bool pin_bytes = false;
4063 head = rb_entry(node, struct btrfs_delayed_ref_head,
4065 if (!mutex_trylock(&head->mutex)) {
4066 atomic_inc(&head->node.refs);
4067 spin_unlock(&delayed_refs->lock);
4069 mutex_lock(&head->mutex);
4070 mutex_unlock(&head->mutex);
4071 btrfs_put_delayed_ref(&head->node);
4072 spin_lock(&delayed_refs->lock);
4075 spin_lock(&head->lock);
4076 while ((node = rb_first(&head->ref_root)) != NULL) {
4077 ref = rb_entry(node, struct btrfs_delayed_ref_node,
4080 rb_erase(&ref->rb_node, &head->ref_root);
4081 atomic_dec(&delayed_refs->num_entries);
4082 btrfs_put_delayed_ref(ref);
4084 if (head->must_insert_reserved)
4086 btrfs_free_delayed_extent_op(head->extent_op);
4087 delayed_refs->num_heads--;
4088 if (head->processing == 0)
4089 delayed_refs->num_heads_ready--;
4090 atomic_dec(&delayed_refs->num_entries);
4091 head->node.in_tree = 0;
4092 rb_erase(&head->href_node, &delayed_refs->href_root);
4093 spin_unlock(&head->lock);
4094 spin_unlock(&delayed_refs->lock);
4095 mutex_unlock(&head->mutex);
4098 btrfs_pin_extent(root, head->node.bytenr,
4099 head->node.num_bytes, 1);
4100 btrfs_put_delayed_ref(&head->node);
4102 spin_lock(&delayed_refs->lock);
4105 spin_unlock(&delayed_refs->lock);
4110 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4112 struct btrfs_inode *btrfs_inode;
4113 struct list_head splice;
4115 INIT_LIST_HEAD(&splice);
4117 spin_lock(&root->delalloc_lock);
4118 list_splice_init(&root->delalloc_inodes, &splice);
4120 while (!list_empty(&splice)) {
4121 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4124 list_del_init(&btrfs_inode->delalloc_inodes);
4125 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4126 &btrfs_inode->runtime_flags);
4127 spin_unlock(&root->delalloc_lock);
4129 btrfs_invalidate_inodes(btrfs_inode->root);
4131 spin_lock(&root->delalloc_lock);
4134 spin_unlock(&root->delalloc_lock);
4137 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4139 struct btrfs_root *root;
4140 struct list_head splice;
4142 INIT_LIST_HEAD(&splice);
4144 spin_lock(&fs_info->delalloc_root_lock);
4145 list_splice_init(&fs_info->delalloc_roots, &splice);
4146 while (!list_empty(&splice)) {
4147 root = list_first_entry(&splice, struct btrfs_root,
4149 list_del_init(&root->delalloc_root);
4150 root = btrfs_grab_fs_root(root);
4152 spin_unlock(&fs_info->delalloc_root_lock);
4154 btrfs_destroy_delalloc_inodes(root);
4155 btrfs_put_fs_root(root);
4157 spin_lock(&fs_info->delalloc_root_lock);
4159 spin_unlock(&fs_info->delalloc_root_lock);
4162 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4163 struct extent_io_tree *dirty_pages,
4167 struct extent_buffer *eb;
4172 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4177 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4178 while (start <= end) {
4179 eb = btrfs_find_tree_block(root->fs_info, start);
4180 start += root->nodesize;
4183 wait_on_extent_buffer_writeback(eb);
4185 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4187 clear_extent_buffer_dirty(eb);
4188 free_extent_buffer_stale(eb);
4195 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4196 struct extent_io_tree *pinned_extents)
4198 struct extent_io_tree *unpin;
4204 unpin = pinned_extents;
4207 ret = find_first_extent_bit(unpin, 0, &start, &end,
4208 EXTENT_DIRTY, NULL);
4212 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4213 btrfs_error_unpin_extent_range(root, start, end);
4218 if (unpin == &root->fs_info->freed_extents[0])
4219 unpin = &root->fs_info->freed_extents[1];
4221 unpin = &root->fs_info->freed_extents[0];
4229 static void btrfs_free_pending_ordered(struct btrfs_transaction *cur_trans,
4230 struct btrfs_fs_info *fs_info)
4232 struct btrfs_ordered_extent *ordered;
4234 spin_lock(&fs_info->trans_lock);
4235 while (!list_empty(&cur_trans->pending_ordered)) {
4236 ordered = list_first_entry(&cur_trans->pending_ordered,
4237 struct btrfs_ordered_extent,
4239 list_del_init(&ordered->trans_list);
4240 spin_unlock(&fs_info->trans_lock);
4242 btrfs_put_ordered_extent(ordered);
4243 spin_lock(&fs_info->trans_lock);
4245 spin_unlock(&fs_info->trans_lock);
4248 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4249 struct btrfs_root *root)
4251 btrfs_destroy_delayed_refs(cur_trans, root);
4253 cur_trans->state = TRANS_STATE_COMMIT_START;
4254 wake_up(&root->fs_info->transaction_blocked_wait);
4256 cur_trans->state = TRANS_STATE_UNBLOCKED;
4257 wake_up(&root->fs_info->transaction_wait);
4259 btrfs_free_pending_ordered(cur_trans, root->fs_info);
4260 btrfs_destroy_delayed_inodes(root);
4261 btrfs_assert_delayed_root_empty(root);
4263 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4265 btrfs_destroy_pinned_extent(root,
4266 root->fs_info->pinned_extents);
4268 cur_trans->state =TRANS_STATE_COMPLETED;
4269 wake_up(&cur_trans->commit_wait);
4272 memset(cur_trans, 0, sizeof(*cur_trans));
4273 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4277 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4279 struct btrfs_transaction *t;
4281 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4283 spin_lock(&root->fs_info->trans_lock);
4284 while (!list_empty(&root->fs_info->trans_list)) {
4285 t = list_first_entry(&root->fs_info->trans_list,
4286 struct btrfs_transaction, list);
4287 if (t->state >= TRANS_STATE_COMMIT_START) {
4288 atomic_inc(&t->use_count);
4289 spin_unlock(&root->fs_info->trans_lock);
4290 btrfs_wait_for_commit(root, t->transid);
4291 btrfs_put_transaction(t);
4292 spin_lock(&root->fs_info->trans_lock);
4295 if (t == root->fs_info->running_transaction) {
4296 t->state = TRANS_STATE_COMMIT_DOING;
4297 spin_unlock(&root->fs_info->trans_lock);
4299 * We wait for 0 num_writers since we don't hold a trans
4300 * handle open currently for this transaction.
4302 wait_event(t->writer_wait,
4303 atomic_read(&t->num_writers) == 0);
4305 spin_unlock(&root->fs_info->trans_lock);
4307 btrfs_cleanup_one_transaction(t, root);
4309 spin_lock(&root->fs_info->trans_lock);
4310 if (t == root->fs_info->running_transaction)
4311 root->fs_info->running_transaction = NULL;
4312 list_del_init(&t->list);
4313 spin_unlock(&root->fs_info->trans_lock);
4315 btrfs_put_transaction(t);
4316 trace_btrfs_transaction_commit(root);
4317 spin_lock(&root->fs_info->trans_lock);
4319 spin_unlock(&root->fs_info->trans_lock);
4320 btrfs_destroy_all_ordered_extents(root->fs_info);
4321 btrfs_destroy_delayed_inodes(root);
4322 btrfs_assert_delayed_root_empty(root);
4323 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4324 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4325 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4330 static const struct extent_io_ops btree_extent_io_ops = {
4331 .readpage_end_io_hook = btree_readpage_end_io_hook,
4332 .readpage_io_failed_hook = btree_io_failed_hook,
4333 .submit_bio_hook = btree_submit_bio_hook,
4334 /* note we're sharing with inode.c for the merge bio hook */
4335 .merge_bio_hook = btrfs_merge_bio_hook,