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 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_root *root, struct extent_buffer *buf,
280 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
283 unsigned long cur_len;
284 unsigned long offset = BTRFS_CSUM_SIZE;
286 unsigned long map_start;
287 unsigned long map_len;
290 unsigned long inline_result;
292 len = buf->len - offset;
294 err = map_private_extent_buffer(buf, offset, 32,
295 &kaddr, &map_start, &map_len);
298 cur_len = min(len, map_len - (offset - map_start));
299 crc = btrfs_csum_data(kaddr + offset - map_start,
304 if (csum_size > sizeof(inline_result)) {
305 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
309 result = (char *)&inline_result;
312 btrfs_csum_final(crc, result);
315 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
318 memcpy(&found, result, csum_size);
320 read_extent_buffer(buf, &val, 0, csum_size);
321 printk_ratelimited(KERN_INFO
322 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
324 root->fs_info->sb->s_id, buf->start,
325 val, found, btrfs_header_level(buf));
326 if (result != (char *)&inline_result)
331 write_extent_buffer(buf, result, 0, csum_size);
333 if (result != (char *)&inline_result)
339 * we can't consider a given block up to date unless the transid of the
340 * block matches the transid in the parent node's pointer. This is how we
341 * detect blocks that either didn't get written at all or got written
342 * in the wrong place.
344 static int verify_parent_transid(struct extent_io_tree *io_tree,
345 struct extent_buffer *eb, u64 parent_transid,
348 struct extent_state *cached_state = NULL;
350 bool need_lock = (current->journal_info ==
351 (void *)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_INFO "BTRFS (device %s): parent transid verify failed on %llu wanted %llu found %llu\n",
372 eb->fs_info->sb->s_id, eb->start,
373 parent_transid, btrfs_header_generation(eb));
377 * Things reading via commit roots that don't have normal protection,
378 * like send, can have a really old block in cache that may point at a
379 * block that has been free'd and re-allocated. So don't clear uptodate
380 * if we find an eb that is under IO (dirty/writeback) because we could
381 * end up reading in the stale data and then writing it back out and
382 * making everybody very sad.
384 if (!extent_buffer_under_io(eb))
385 clear_extent_buffer_uptodate(eb);
387 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
388 &cached_state, GFP_NOFS);
390 btrfs_tree_read_unlock_blocking(eb);
395 * Return 0 if the superblock checksum type matches the checksum value of that
396 * algorithm. Pass the raw disk superblock data.
398 static int btrfs_check_super_csum(char *raw_disk_sb)
400 struct btrfs_super_block *disk_sb =
401 (struct btrfs_super_block *)raw_disk_sb;
402 u16 csum_type = btrfs_super_csum_type(disk_sb);
405 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
407 const int csum_size = sizeof(crc);
408 char result[csum_size];
411 * The super_block structure does not span the whole
412 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
413 * is filled with zeros and is included in the checkum.
415 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
416 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
417 btrfs_csum_final(crc, result);
419 if (memcmp(raw_disk_sb, result, csum_size))
422 if (ret && btrfs_super_generation(disk_sb) < 10) {
424 "BTRFS: super block crcs don't match, older mkfs detected\n");
429 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
430 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
439 * helper to read a given tree block, doing retries as required when
440 * the checksums don't match and we have alternate mirrors to try.
442 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
443 struct extent_buffer *eb,
444 u64 start, u64 parent_transid)
446 struct extent_io_tree *io_tree;
451 int failed_mirror = 0;
453 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
454 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
456 ret = read_extent_buffer_pages(io_tree, eb, start,
458 btree_get_extent, mirror_num);
460 if (!verify_parent_transid(io_tree, eb,
468 * This buffer's crc is fine, but its contents are corrupted, so
469 * there is no reason to read the other copies, they won't be
472 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
475 num_copies = btrfs_num_copies(root->fs_info,
480 if (!failed_mirror) {
482 failed_mirror = eb->read_mirror;
486 if (mirror_num == failed_mirror)
489 if (mirror_num > num_copies)
493 if (failed && !ret && failed_mirror)
494 repair_eb_io_failure(root, eb, failed_mirror);
500 * checksum a dirty tree block before IO. This has extra checks to make sure
501 * we only fill in the checksum field in the first page of a multi-page block
504 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
506 u64 start = page_offset(page);
508 struct extent_buffer *eb;
510 eb = (struct extent_buffer *)page->private;
511 if (page != eb->pages[0])
513 found_start = btrfs_header_bytenr(eb);
514 if (WARN_ON(found_start != start || !PageUptodate(page)))
516 csum_tree_block(root, eb, 0);
520 static int check_tree_block_fsid(struct btrfs_root *root,
521 struct extent_buffer *eb)
523 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
524 u8 fsid[BTRFS_UUID_SIZE];
527 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
529 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
533 fs_devices = fs_devices->seed;
538 #define CORRUPT(reason, eb, root, slot) \
539 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
540 "root=%llu, slot=%d", reason, \
541 btrfs_header_bytenr(eb), root->objectid, slot)
543 static noinline int check_leaf(struct btrfs_root *root,
544 struct extent_buffer *leaf)
546 struct btrfs_key key;
547 struct btrfs_key leaf_key;
548 u32 nritems = btrfs_header_nritems(leaf);
554 /* Check the 0 item */
555 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
556 BTRFS_LEAF_DATA_SIZE(root)) {
557 CORRUPT("invalid item offset size pair", leaf, root, 0);
562 * Check to make sure each items keys are in the correct order and their
563 * offsets make sense. We only have to loop through nritems-1 because
564 * we check the current slot against the next slot, which verifies the
565 * next slot's offset+size makes sense and that the current's slot
568 for (slot = 0; slot < nritems - 1; slot++) {
569 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
570 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
572 /* Make sure the keys are in the right order */
573 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
574 CORRUPT("bad key order", leaf, root, slot);
579 * Make sure the offset and ends are right, remember that the
580 * item data starts at the end of the leaf and grows towards the
583 if (btrfs_item_offset_nr(leaf, slot) !=
584 btrfs_item_end_nr(leaf, slot + 1)) {
585 CORRUPT("slot offset bad", leaf, root, slot);
590 * Check to make sure that we don't point outside of the leaf,
591 * just incase all the items are consistent to eachother, but
592 * all point outside of the leaf.
594 if (btrfs_item_end_nr(leaf, slot) >
595 BTRFS_LEAF_DATA_SIZE(root)) {
596 CORRUPT("slot end outside of leaf", leaf, root, slot);
604 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
605 u64 phy_offset, struct page *page,
606 u64 start, u64 end, int mirror)
610 struct extent_buffer *eb;
611 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
618 eb = (struct extent_buffer *)page->private;
620 /* the pending IO might have been the only thing that kept this buffer
621 * in memory. Make sure we have a ref for all this other checks
623 extent_buffer_get(eb);
625 reads_done = atomic_dec_and_test(&eb->io_pages);
629 eb->read_mirror = mirror;
630 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
635 found_start = btrfs_header_bytenr(eb);
636 if (found_start != eb->start) {
637 printk_ratelimited(KERN_INFO "BTRFS (device %s): bad tree block start "
639 eb->fs_info->sb->s_id, found_start, eb->start);
643 if (check_tree_block_fsid(root, eb)) {
644 printk_ratelimited(KERN_INFO "BTRFS (device %s): bad fsid on block %llu\n",
645 eb->fs_info->sb->s_id, eb->start);
649 found_level = btrfs_header_level(eb);
650 if (found_level >= BTRFS_MAX_LEVEL) {
651 btrfs_info(root->fs_info, "bad tree block level %d",
652 (int)btrfs_header_level(eb));
657 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
660 ret = csum_tree_block(root, eb, 1);
667 * If this is a leaf block and it is corrupt, set the corrupt bit so
668 * that we don't try and read the other copies of this block, just
671 if (found_level == 0 && check_leaf(root, eb)) {
672 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
677 set_extent_buffer_uptodate(eb);
680 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
681 btree_readahead_hook(root, eb, eb->start, ret);
685 * our io error hook is going to dec the io pages
686 * again, we have to make sure it has something
689 atomic_inc(&eb->io_pages);
690 clear_extent_buffer_uptodate(eb);
692 free_extent_buffer(eb);
697 static int btree_io_failed_hook(struct page *page, int failed_mirror)
699 struct extent_buffer *eb;
700 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
702 eb = (struct extent_buffer *)page->private;
703 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
704 eb->read_mirror = failed_mirror;
705 atomic_dec(&eb->io_pages);
706 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
707 btree_readahead_hook(root, eb, eb->start, -EIO);
708 return -EIO; /* we fixed nothing */
711 static void end_workqueue_bio(struct bio *bio, int err)
713 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
714 struct btrfs_fs_info *fs_info;
715 struct btrfs_workqueue *wq;
716 btrfs_work_func_t func;
718 fs_info = end_io_wq->info;
719 end_io_wq->error = err;
721 if (bio->bi_rw & REQ_WRITE) {
722 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
723 wq = fs_info->endio_meta_write_workers;
724 func = btrfs_endio_meta_write_helper;
725 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
726 wq = fs_info->endio_freespace_worker;
727 func = btrfs_freespace_write_helper;
728 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
729 wq = fs_info->endio_raid56_workers;
730 func = btrfs_endio_raid56_helper;
732 wq = fs_info->endio_write_workers;
733 func = btrfs_endio_write_helper;
736 if (unlikely(end_io_wq->metadata ==
737 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
738 wq = fs_info->endio_repair_workers;
739 func = btrfs_endio_repair_helper;
740 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
741 wq = fs_info->endio_raid56_workers;
742 func = btrfs_endio_raid56_helper;
743 } else if (end_io_wq->metadata) {
744 wq = fs_info->endio_meta_workers;
745 func = btrfs_endio_meta_helper;
747 wq = fs_info->endio_workers;
748 func = btrfs_endio_helper;
752 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
753 btrfs_queue_work(wq, &end_io_wq->work);
756 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
757 enum btrfs_wq_endio_type metadata)
759 struct btrfs_end_io_wq *end_io_wq;
761 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
765 end_io_wq->private = bio->bi_private;
766 end_io_wq->end_io = bio->bi_end_io;
767 end_io_wq->info = info;
768 end_io_wq->error = 0;
769 end_io_wq->bio = bio;
770 end_io_wq->metadata = metadata;
772 bio->bi_private = end_io_wq;
773 bio->bi_end_io = end_workqueue_bio;
777 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
779 unsigned long limit = min_t(unsigned long,
780 info->thread_pool_size,
781 info->fs_devices->open_devices);
785 static void run_one_async_start(struct btrfs_work *work)
787 struct async_submit_bio *async;
790 async = container_of(work, struct async_submit_bio, work);
791 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
792 async->mirror_num, async->bio_flags,
798 static void run_one_async_done(struct btrfs_work *work)
800 struct btrfs_fs_info *fs_info;
801 struct async_submit_bio *async;
804 async = container_of(work, struct async_submit_bio, work);
805 fs_info = BTRFS_I(async->inode)->root->fs_info;
807 limit = btrfs_async_submit_limit(fs_info);
808 limit = limit * 2 / 3;
810 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
811 waitqueue_active(&fs_info->async_submit_wait))
812 wake_up(&fs_info->async_submit_wait);
814 /* If an error occured we just want to clean up the bio and move on */
816 bio_endio(async->bio, async->error);
820 async->submit_bio_done(async->inode, async->rw, async->bio,
821 async->mirror_num, async->bio_flags,
825 static void run_one_async_free(struct btrfs_work *work)
827 struct async_submit_bio *async;
829 async = container_of(work, struct async_submit_bio, work);
833 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
834 int rw, struct bio *bio, int mirror_num,
835 unsigned long bio_flags,
837 extent_submit_bio_hook_t *submit_bio_start,
838 extent_submit_bio_hook_t *submit_bio_done)
840 struct async_submit_bio *async;
842 async = kmalloc(sizeof(*async), GFP_NOFS);
846 async->inode = inode;
849 async->mirror_num = mirror_num;
850 async->submit_bio_start = submit_bio_start;
851 async->submit_bio_done = submit_bio_done;
853 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
854 run_one_async_done, run_one_async_free);
856 async->bio_flags = bio_flags;
857 async->bio_offset = bio_offset;
861 atomic_inc(&fs_info->nr_async_submits);
864 btrfs_set_work_high_priority(&async->work);
866 btrfs_queue_work(fs_info->workers, &async->work);
868 while (atomic_read(&fs_info->async_submit_draining) &&
869 atomic_read(&fs_info->nr_async_submits)) {
870 wait_event(fs_info->async_submit_wait,
871 (atomic_read(&fs_info->nr_async_submits) == 0));
877 static int btree_csum_one_bio(struct bio *bio)
879 struct bio_vec *bvec;
880 struct btrfs_root *root;
883 bio_for_each_segment_all(bvec, bio, i) {
884 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
885 ret = csum_dirty_buffer(root, bvec->bv_page);
893 static int __btree_submit_bio_start(struct inode *inode, int rw,
894 struct bio *bio, int mirror_num,
895 unsigned long bio_flags,
899 * when we're called for a write, we're already in the async
900 * submission context. Just jump into btrfs_map_bio
902 return btree_csum_one_bio(bio);
905 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
906 int mirror_num, unsigned long bio_flags,
912 * when we're called for a write, we're already in the async
913 * submission context. Just jump into btrfs_map_bio
915 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
921 static int check_async_write(struct inode *inode, unsigned long bio_flags)
923 if (bio_flags & EXTENT_BIO_TREE_LOG)
932 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
933 int mirror_num, unsigned long bio_flags,
936 int async = check_async_write(inode, bio_flags);
939 if (!(rw & REQ_WRITE)) {
941 * called for a read, do the setup so that checksum validation
942 * can happen in the async kernel threads
944 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
945 bio, BTRFS_WQ_ENDIO_METADATA);
948 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
951 ret = btree_csum_one_bio(bio);
954 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
958 * kthread helpers are used to submit writes so that
959 * checksumming can happen in parallel across all CPUs
961 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
962 inode, rw, bio, mirror_num, 0,
964 __btree_submit_bio_start,
965 __btree_submit_bio_done);
975 #ifdef CONFIG_MIGRATION
976 static int btree_migratepage(struct address_space *mapping,
977 struct page *newpage, struct page *page,
978 enum migrate_mode mode)
981 * we can't safely write a btree page from here,
982 * we haven't done the locking hook
987 * Buffers may be managed in a filesystem specific way.
988 * We must have no buffers or drop them.
990 if (page_has_private(page) &&
991 !try_to_release_page(page, GFP_KERNEL))
993 return migrate_page(mapping, newpage, page, mode);
998 static int btree_writepages(struct address_space *mapping,
999 struct writeback_control *wbc)
1001 struct btrfs_fs_info *fs_info;
1004 if (wbc->sync_mode == WB_SYNC_NONE) {
1006 if (wbc->for_kupdate)
1009 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1010 /* this is a bit racy, but that's ok */
1011 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1012 BTRFS_DIRTY_METADATA_THRESH);
1016 return btree_write_cache_pages(mapping, wbc);
1019 static int btree_readpage(struct file *file, struct page *page)
1021 struct extent_io_tree *tree;
1022 tree = &BTRFS_I(page->mapping->host)->io_tree;
1023 return extent_read_full_page(tree, page, btree_get_extent, 0);
1026 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1028 if (PageWriteback(page) || PageDirty(page))
1031 return try_release_extent_buffer(page);
1034 static void btree_invalidatepage(struct page *page, unsigned int offset,
1035 unsigned int length)
1037 struct extent_io_tree *tree;
1038 tree = &BTRFS_I(page->mapping->host)->io_tree;
1039 extent_invalidatepage(tree, page, offset);
1040 btree_releasepage(page, GFP_NOFS);
1041 if (PagePrivate(page)) {
1042 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1043 "page private not zero on page %llu",
1044 (unsigned long long)page_offset(page));
1045 ClearPagePrivate(page);
1046 set_page_private(page, 0);
1047 page_cache_release(page);
1051 static int btree_set_page_dirty(struct page *page)
1054 struct extent_buffer *eb;
1056 BUG_ON(!PagePrivate(page));
1057 eb = (struct extent_buffer *)page->private;
1059 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1060 BUG_ON(!atomic_read(&eb->refs));
1061 btrfs_assert_tree_locked(eb);
1063 return __set_page_dirty_nobuffers(page);
1066 static const struct address_space_operations btree_aops = {
1067 .readpage = btree_readpage,
1068 .writepages = btree_writepages,
1069 .releasepage = btree_releasepage,
1070 .invalidatepage = btree_invalidatepage,
1071 #ifdef CONFIG_MIGRATION
1072 .migratepage = btree_migratepage,
1074 .set_page_dirty = btree_set_page_dirty,
1077 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1080 struct extent_buffer *buf = NULL;
1081 struct inode *btree_inode = root->fs_info->btree_inode;
1084 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1087 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1088 buf, 0, WAIT_NONE, btree_get_extent, 0);
1089 free_extent_buffer(buf);
1093 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1094 int mirror_num, struct extent_buffer **eb)
1096 struct extent_buffer *buf = NULL;
1097 struct inode *btree_inode = root->fs_info->btree_inode;
1098 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1101 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1105 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1107 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1108 btree_get_extent, mirror_num);
1110 free_extent_buffer(buf);
1114 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1115 free_extent_buffer(buf);
1117 } else if (extent_buffer_uptodate(buf)) {
1120 free_extent_buffer(buf);
1125 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1126 u64 bytenr, u32 blocksize)
1128 return find_extent_buffer(root->fs_info, bytenr);
1131 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1132 u64 bytenr, u32 blocksize)
1134 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1135 if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state)))
1136 return alloc_test_extent_buffer(root->fs_info, bytenr,
1139 return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1143 int btrfs_write_tree_block(struct extent_buffer *buf)
1145 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1146 buf->start + buf->len - 1);
1149 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1151 return filemap_fdatawait_range(buf->pages[0]->mapping,
1152 buf->start, buf->start + buf->len - 1);
1155 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1156 u32 blocksize, u64 parent_transid)
1158 struct extent_buffer *buf = NULL;
1161 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1165 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1167 free_extent_buffer(buf);
1174 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1175 struct extent_buffer *buf)
1177 struct btrfs_fs_info *fs_info = root->fs_info;
1179 if (btrfs_header_generation(buf) ==
1180 fs_info->running_transaction->transid) {
1181 btrfs_assert_tree_locked(buf);
1183 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1184 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1186 fs_info->dirty_metadata_batch);
1187 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1188 btrfs_set_lock_blocking(buf);
1189 clear_extent_buffer_dirty(buf);
1194 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1196 struct btrfs_subvolume_writers *writers;
1199 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1201 return ERR_PTR(-ENOMEM);
1203 ret = percpu_counter_init(&writers->counter, 0);
1206 return ERR_PTR(ret);
1209 init_waitqueue_head(&writers->wait);
1214 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1216 percpu_counter_destroy(&writers->counter);
1220 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1221 struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1225 root->commit_root = NULL;
1226 root->sectorsize = sectorsize;
1227 root->nodesize = nodesize;
1228 root->stripesize = stripesize;
1230 root->orphan_cleanup_state = 0;
1232 root->objectid = objectid;
1233 root->last_trans = 0;
1234 root->highest_objectid = 0;
1235 root->nr_delalloc_inodes = 0;
1236 root->nr_ordered_extents = 0;
1238 root->inode_tree = RB_ROOT;
1239 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1240 root->block_rsv = NULL;
1241 root->orphan_block_rsv = NULL;
1243 INIT_LIST_HEAD(&root->dirty_list);
1244 INIT_LIST_HEAD(&root->root_list);
1245 INIT_LIST_HEAD(&root->delalloc_inodes);
1246 INIT_LIST_HEAD(&root->delalloc_root);
1247 INIT_LIST_HEAD(&root->ordered_extents);
1248 INIT_LIST_HEAD(&root->ordered_root);
1249 INIT_LIST_HEAD(&root->logged_list[0]);
1250 INIT_LIST_HEAD(&root->logged_list[1]);
1251 spin_lock_init(&root->orphan_lock);
1252 spin_lock_init(&root->inode_lock);
1253 spin_lock_init(&root->delalloc_lock);
1254 spin_lock_init(&root->ordered_extent_lock);
1255 spin_lock_init(&root->accounting_lock);
1256 spin_lock_init(&root->log_extents_lock[0]);
1257 spin_lock_init(&root->log_extents_lock[1]);
1258 mutex_init(&root->objectid_mutex);
1259 mutex_init(&root->log_mutex);
1260 mutex_init(&root->ordered_extent_mutex);
1261 mutex_init(&root->delalloc_mutex);
1262 init_waitqueue_head(&root->log_writer_wait);
1263 init_waitqueue_head(&root->log_commit_wait[0]);
1264 init_waitqueue_head(&root->log_commit_wait[1]);
1265 INIT_LIST_HEAD(&root->log_ctxs[0]);
1266 INIT_LIST_HEAD(&root->log_ctxs[1]);
1267 atomic_set(&root->log_commit[0], 0);
1268 atomic_set(&root->log_commit[1], 0);
1269 atomic_set(&root->log_writers, 0);
1270 atomic_set(&root->log_batch, 0);
1271 atomic_set(&root->orphan_inodes, 0);
1272 atomic_set(&root->refs, 1);
1273 atomic_set(&root->will_be_snapshoted, 0);
1274 root->log_transid = 0;
1275 root->log_transid_committed = -1;
1276 root->last_log_commit = 0;
1278 extent_io_tree_init(&root->dirty_log_pages,
1279 fs_info->btree_inode->i_mapping);
1281 memset(&root->root_key, 0, sizeof(root->root_key));
1282 memset(&root->root_item, 0, sizeof(root->root_item));
1283 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1284 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1286 root->defrag_trans_start = fs_info->generation;
1288 root->defrag_trans_start = 0;
1289 init_completion(&root->kobj_unregister);
1290 root->root_key.objectid = objectid;
1293 spin_lock_init(&root->root_item_lock);
1296 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1298 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1300 root->fs_info = fs_info;
1304 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1305 /* Should only be used by the testing infrastructure */
1306 struct btrfs_root *btrfs_alloc_dummy_root(void)
1308 struct btrfs_root *root;
1310 root = btrfs_alloc_root(NULL);
1312 return ERR_PTR(-ENOMEM);
1313 __setup_root(4096, 4096, 4096, root, NULL, 1);
1314 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1315 root->alloc_bytenr = 0;
1321 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1322 struct btrfs_fs_info *fs_info,
1325 struct extent_buffer *leaf;
1326 struct btrfs_root *tree_root = fs_info->tree_root;
1327 struct btrfs_root *root;
1328 struct btrfs_key key;
1332 root = btrfs_alloc_root(fs_info);
1334 return ERR_PTR(-ENOMEM);
1336 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1337 tree_root->stripesize, root, fs_info, objectid);
1338 root->root_key.objectid = objectid;
1339 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1340 root->root_key.offset = 0;
1342 leaf = btrfs_alloc_free_block(trans, root, root->nodesize,
1343 0, objectid, NULL, 0, 0, 0);
1345 ret = PTR_ERR(leaf);
1350 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1351 btrfs_set_header_bytenr(leaf, leaf->start);
1352 btrfs_set_header_generation(leaf, trans->transid);
1353 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1354 btrfs_set_header_owner(leaf, objectid);
1357 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1359 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1360 btrfs_header_chunk_tree_uuid(leaf),
1362 btrfs_mark_buffer_dirty(leaf);
1364 root->commit_root = btrfs_root_node(root);
1365 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1367 root->root_item.flags = 0;
1368 root->root_item.byte_limit = 0;
1369 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1370 btrfs_set_root_generation(&root->root_item, trans->transid);
1371 btrfs_set_root_level(&root->root_item, 0);
1372 btrfs_set_root_refs(&root->root_item, 1);
1373 btrfs_set_root_used(&root->root_item, leaf->len);
1374 btrfs_set_root_last_snapshot(&root->root_item, 0);
1375 btrfs_set_root_dirid(&root->root_item, 0);
1377 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1378 root->root_item.drop_level = 0;
1380 key.objectid = objectid;
1381 key.type = BTRFS_ROOT_ITEM_KEY;
1383 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1387 btrfs_tree_unlock(leaf);
1393 btrfs_tree_unlock(leaf);
1394 free_extent_buffer(root->commit_root);
1395 free_extent_buffer(leaf);
1399 return ERR_PTR(ret);
1402 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1403 struct btrfs_fs_info *fs_info)
1405 struct btrfs_root *root;
1406 struct btrfs_root *tree_root = fs_info->tree_root;
1407 struct extent_buffer *leaf;
1409 root = btrfs_alloc_root(fs_info);
1411 return ERR_PTR(-ENOMEM);
1413 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1414 tree_root->stripesize, root, fs_info,
1415 BTRFS_TREE_LOG_OBJECTID);
1417 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1418 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1419 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1422 * DON'T set REF_COWS for log trees
1424 * log trees do not get reference counted because they go away
1425 * before a real commit is actually done. They do store pointers
1426 * to file data extents, and those reference counts still get
1427 * updated (along with back refs to the log tree).
1430 leaf = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
1431 BTRFS_TREE_LOG_OBJECTID, NULL,
1435 return ERR_CAST(leaf);
1438 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1439 btrfs_set_header_bytenr(leaf, leaf->start);
1440 btrfs_set_header_generation(leaf, trans->transid);
1441 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1442 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1445 write_extent_buffer(root->node, root->fs_info->fsid,
1446 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1447 btrfs_mark_buffer_dirty(root->node);
1448 btrfs_tree_unlock(root->node);
1452 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1453 struct btrfs_fs_info *fs_info)
1455 struct btrfs_root *log_root;
1457 log_root = alloc_log_tree(trans, fs_info);
1458 if (IS_ERR(log_root))
1459 return PTR_ERR(log_root);
1460 WARN_ON(fs_info->log_root_tree);
1461 fs_info->log_root_tree = log_root;
1465 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1466 struct btrfs_root *root)
1468 struct btrfs_root *log_root;
1469 struct btrfs_inode_item *inode_item;
1471 log_root = alloc_log_tree(trans, root->fs_info);
1472 if (IS_ERR(log_root))
1473 return PTR_ERR(log_root);
1475 log_root->last_trans = trans->transid;
1476 log_root->root_key.offset = root->root_key.objectid;
1478 inode_item = &log_root->root_item.inode;
1479 btrfs_set_stack_inode_generation(inode_item, 1);
1480 btrfs_set_stack_inode_size(inode_item, 3);
1481 btrfs_set_stack_inode_nlink(inode_item, 1);
1482 btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1483 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1485 btrfs_set_root_node(&log_root->root_item, log_root->node);
1487 WARN_ON(root->log_root);
1488 root->log_root = log_root;
1489 root->log_transid = 0;
1490 root->log_transid_committed = -1;
1491 root->last_log_commit = 0;
1495 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1496 struct btrfs_key *key)
1498 struct btrfs_root *root;
1499 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1500 struct btrfs_path *path;
1505 path = btrfs_alloc_path();
1507 return ERR_PTR(-ENOMEM);
1509 root = btrfs_alloc_root(fs_info);
1515 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1516 tree_root->stripesize, root, fs_info, key->objectid);
1518 ret = btrfs_find_root(tree_root, key, path,
1519 &root->root_item, &root->root_key);
1526 generation = btrfs_root_generation(&root->root_item);
1527 blocksize = root->nodesize;
1528 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1529 blocksize, generation);
1533 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1537 root->commit_root = btrfs_root_node(root);
1539 btrfs_free_path(path);
1543 free_extent_buffer(root->node);
1547 root = ERR_PTR(ret);
1551 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1552 struct btrfs_key *location)
1554 struct btrfs_root *root;
1556 root = btrfs_read_tree_root(tree_root, location);
1560 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1561 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1562 btrfs_check_and_init_root_item(&root->root_item);
1568 int btrfs_init_fs_root(struct btrfs_root *root)
1571 struct btrfs_subvolume_writers *writers;
1573 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1574 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1576 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1581 writers = btrfs_alloc_subvolume_writers();
1582 if (IS_ERR(writers)) {
1583 ret = PTR_ERR(writers);
1586 root->subv_writers = writers;
1588 btrfs_init_free_ino_ctl(root);
1589 spin_lock_init(&root->ino_cache_lock);
1590 init_waitqueue_head(&root->ino_cache_wait);
1592 ret = get_anon_bdev(&root->anon_dev);
1598 btrfs_free_subvolume_writers(root->subv_writers);
1600 kfree(root->free_ino_ctl);
1601 kfree(root->free_ino_pinned);
1605 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1608 struct btrfs_root *root;
1610 spin_lock(&fs_info->fs_roots_radix_lock);
1611 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1612 (unsigned long)root_id);
1613 spin_unlock(&fs_info->fs_roots_radix_lock);
1617 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1618 struct btrfs_root *root)
1622 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1626 spin_lock(&fs_info->fs_roots_radix_lock);
1627 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1628 (unsigned long)root->root_key.objectid,
1631 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1632 spin_unlock(&fs_info->fs_roots_radix_lock);
1633 radix_tree_preload_end();
1638 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1639 struct btrfs_key *location,
1642 struct btrfs_root *root;
1645 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1646 return fs_info->tree_root;
1647 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1648 return fs_info->extent_root;
1649 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1650 return fs_info->chunk_root;
1651 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1652 return fs_info->dev_root;
1653 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1654 return fs_info->csum_root;
1655 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1656 return fs_info->quota_root ? fs_info->quota_root :
1658 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1659 return fs_info->uuid_root ? fs_info->uuid_root :
1662 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1664 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1665 return ERR_PTR(-ENOENT);
1669 root = btrfs_read_fs_root(fs_info->tree_root, location);
1673 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1678 ret = btrfs_init_fs_root(root);
1682 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1683 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1687 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1689 ret = btrfs_insert_fs_root(fs_info, root);
1691 if (ret == -EEXIST) {
1700 return ERR_PTR(ret);
1703 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1705 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1707 struct btrfs_device *device;
1708 struct backing_dev_info *bdi;
1711 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1714 bdi = blk_get_backing_dev_info(device->bdev);
1715 if (bdi && bdi_congested(bdi, bdi_bits)) {
1724 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1728 bdi->capabilities = BDI_CAP_MAP_COPY;
1729 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1733 bdi->ra_pages = default_backing_dev_info.ra_pages;
1734 bdi->congested_fn = btrfs_congested_fn;
1735 bdi->congested_data = info;
1740 * called by the kthread helper functions to finally call the bio end_io
1741 * functions. This is where read checksum verification actually happens
1743 static void end_workqueue_fn(struct btrfs_work *work)
1746 struct btrfs_end_io_wq *end_io_wq;
1749 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1750 bio = end_io_wq->bio;
1752 error = end_io_wq->error;
1753 bio->bi_private = end_io_wq->private;
1754 bio->bi_end_io = end_io_wq->end_io;
1755 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1756 bio_endio_nodec(bio, error);
1759 static int cleaner_kthread(void *arg)
1761 struct btrfs_root *root = arg;
1767 /* Make the cleaner go to sleep early. */
1768 if (btrfs_need_cleaner_sleep(root))
1771 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1775 * Avoid the problem that we change the status of the fs
1776 * during the above check and trylock.
1778 if (btrfs_need_cleaner_sleep(root)) {
1779 mutex_unlock(&root->fs_info->cleaner_mutex);
1783 btrfs_run_delayed_iputs(root);
1784 btrfs_delete_unused_bgs(root->fs_info);
1785 again = btrfs_clean_one_deleted_snapshot(root);
1786 mutex_unlock(&root->fs_info->cleaner_mutex);
1789 * The defragger has dealt with the R/O remount and umount,
1790 * needn't do anything special here.
1792 btrfs_run_defrag_inodes(root->fs_info);
1794 if (!try_to_freeze() && !again) {
1795 set_current_state(TASK_INTERRUPTIBLE);
1796 if (!kthread_should_stop())
1798 __set_current_state(TASK_RUNNING);
1800 } while (!kthread_should_stop());
1804 static int transaction_kthread(void *arg)
1806 struct btrfs_root *root = arg;
1807 struct btrfs_trans_handle *trans;
1808 struct btrfs_transaction *cur;
1811 unsigned long delay;
1815 cannot_commit = false;
1816 delay = HZ * root->fs_info->commit_interval;
1817 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1819 spin_lock(&root->fs_info->trans_lock);
1820 cur = root->fs_info->running_transaction;
1822 spin_unlock(&root->fs_info->trans_lock);
1826 now = get_seconds();
1827 if (cur->state < TRANS_STATE_BLOCKED &&
1828 (now < cur->start_time ||
1829 now - cur->start_time < root->fs_info->commit_interval)) {
1830 spin_unlock(&root->fs_info->trans_lock);
1834 transid = cur->transid;
1835 spin_unlock(&root->fs_info->trans_lock);
1837 /* If the file system is aborted, this will always fail. */
1838 trans = btrfs_attach_transaction(root);
1839 if (IS_ERR(trans)) {
1840 if (PTR_ERR(trans) != -ENOENT)
1841 cannot_commit = true;
1844 if (transid == trans->transid) {
1845 btrfs_commit_transaction(trans, root);
1847 btrfs_end_transaction(trans, root);
1850 wake_up_process(root->fs_info->cleaner_kthread);
1851 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1853 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1854 &root->fs_info->fs_state)))
1855 btrfs_cleanup_transaction(root);
1856 if (!try_to_freeze()) {
1857 set_current_state(TASK_INTERRUPTIBLE);
1858 if (!kthread_should_stop() &&
1859 (!btrfs_transaction_blocked(root->fs_info) ||
1861 schedule_timeout(delay);
1862 __set_current_state(TASK_RUNNING);
1864 } while (!kthread_should_stop());
1869 * this will find the highest generation in the array of
1870 * root backups. The index of the highest array is returned,
1871 * or -1 if we can't find anything.
1873 * We check to make sure the array is valid by comparing the
1874 * generation of the latest root in the array with the generation
1875 * in the super block. If they don't match we pitch it.
1877 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1880 int newest_index = -1;
1881 struct btrfs_root_backup *root_backup;
1884 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1885 root_backup = info->super_copy->super_roots + i;
1886 cur = btrfs_backup_tree_root_gen(root_backup);
1887 if (cur == newest_gen)
1891 /* check to see if we actually wrapped around */
1892 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1893 root_backup = info->super_copy->super_roots;
1894 cur = btrfs_backup_tree_root_gen(root_backup);
1895 if (cur == newest_gen)
1898 return newest_index;
1903 * find the oldest backup so we know where to store new entries
1904 * in the backup array. This will set the backup_root_index
1905 * field in the fs_info struct
1907 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1910 int newest_index = -1;
1912 newest_index = find_newest_super_backup(info, newest_gen);
1913 /* if there was garbage in there, just move along */
1914 if (newest_index == -1) {
1915 info->backup_root_index = 0;
1917 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1922 * copy all the root pointers into the super backup array.
1923 * this will bump the backup pointer by one when it is
1926 static void backup_super_roots(struct btrfs_fs_info *info)
1929 struct btrfs_root_backup *root_backup;
1932 next_backup = info->backup_root_index;
1933 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1934 BTRFS_NUM_BACKUP_ROOTS;
1937 * just overwrite the last backup if we're at the same generation
1938 * this happens only at umount
1940 root_backup = info->super_for_commit->super_roots + last_backup;
1941 if (btrfs_backup_tree_root_gen(root_backup) ==
1942 btrfs_header_generation(info->tree_root->node))
1943 next_backup = last_backup;
1945 root_backup = info->super_for_commit->super_roots + next_backup;
1948 * make sure all of our padding and empty slots get zero filled
1949 * regardless of which ones we use today
1951 memset(root_backup, 0, sizeof(*root_backup));
1953 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1955 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1956 btrfs_set_backup_tree_root_gen(root_backup,
1957 btrfs_header_generation(info->tree_root->node));
1959 btrfs_set_backup_tree_root_level(root_backup,
1960 btrfs_header_level(info->tree_root->node));
1962 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1963 btrfs_set_backup_chunk_root_gen(root_backup,
1964 btrfs_header_generation(info->chunk_root->node));
1965 btrfs_set_backup_chunk_root_level(root_backup,
1966 btrfs_header_level(info->chunk_root->node));
1968 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1969 btrfs_set_backup_extent_root_gen(root_backup,
1970 btrfs_header_generation(info->extent_root->node));
1971 btrfs_set_backup_extent_root_level(root_backup,
1972 btrfs_header_level(info->extent_root->node));
1975 * we might commit during log recovery, which happens before we set
1976 * the fs_root. Make sure it is valid before we fill it in.
1978 if (info->fs_root && info->fs_root->node) {
1979 btrfs_set_backup_fs_root(root_backup,
1980 info->fs_root->node->start);
1981 btrfs_set_backup_fs_root_gen(root_backup,
1982 btrfs_header_generation(info->fs_root->node));
1983 btrfs_set_backup_fs_root_level(root_backup,
1984 btrfs_header_level(info->fs_root->node));
1987 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1988 btrfs_set_backup_dev_root_gen(root_backup,
1989 btrfs_header_generation(info->dev_root->node));
1990 btrfs_set_backup_dev_root_level(root_backup,
1991 btrfs_header_level(info->dev_root->node));
1993 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1994 btrfs_set_backup_csum_root_gen(root_backup,
1995 btrfs_header_generation(info->csum_root->node));
1996 btrfs_set_backup_csum_root_level(root_backup,
1997 btrfs_header_level(info->csum_root->node));
1999 btrfs_set_backup_total_bytes(root_backup,
2000 btrfs_super_total_bytes(info->super_copy));
2001 btrfs_set_backup_bytes_used(root_backup,
2002 btrfs_super_bytes_used(info->super_copy));
2003 btrfs_set_backup_num_devices(root_backup,
2004 btrfs_super_num_devices(info->super_copy));
2007 * if we don't copy this out to the super_copy, it won't get remembered
2008 * for the next commit
2010 memcpy(&info->super_copy->super_roots,
2011 &info->super_for_commit->super_roots,
2012 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2016 * this copies info out of the root backup array and back into
2017 * the in-memory super block. It is meant to help iterate through
2018 * the array, so you send it the number of backups you've already
2019 * tried and the last backup index you used.
2021 * this returns -1 when it has tried all the backups
2023 static noinline int next_root_backup(struct btrfs_fs_info *info,
2024 struct btrfs_super_block *super,
2025 int *num_backups_tried, int *backup_index)
2027 struct btrfs_root_backup *root_backup;
2028 int newest = *backup_index;
2030 if (*num_backups_tried == 0) {
2031 u64 gen = btrfs_super_generation(super);
2033 newest = find_newest_super_backup(info, gen);
2037 *backup_index = newest;
2038 *num_backups_tried = 1;
2039 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2040 /* we've tried all the backups, all done */
2043 /* jump to the next oldest backup */
2044 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2045 BTRFS_NUM_BACKUP_ROOTS;
2046 *backup_index = newest;
2047 *num_backups_tried += 1;
2049 root_backup = super->super_roots + newest;
2051 btrfs_set_super_generation(super,
2052 btrfs_backup_tree_root_gen(root_backup));
2053 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2054 btrfs_set_super_root_level(super,
2055 btrfs_backup_tree_root_level(root_backup));
2056 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2059 * fixme: the total bytes and num_devices need to match or we should
2062 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2063 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2067 /* helper to cleanup workers */
2068 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2070 btrfs_destroy_workqueue(fs_info->fixup_workers);
2071 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2072 btrfs_destroy_workqueue(fs_info->workers);
2073 btrfs_destroy_workqueue(fs_info->endio_workers);
2074 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2075 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2076 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2077 btrfs_destroy_workqueue(fs_info->rmw_workers);
2078 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2079 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2080 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2081 btrfs_destroy_workqueue(fs_info->submit_workers);
2082 btrfs_destroy_workqueue(fs_info->delayed_workers);
2083 btrfs_destroy_workqueue(fs_info->caching_workers);
2084 btrfs_destroy_workqueue(fs_info->readahead_workers);
2085 btrfs_destroy_workqueue(fs_info->flush_workers);
2086 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2087 btrfs_destroy_workqueue(fs_info->extent_workers);
2090 static void free_root_extent_buffers(struct btrfs_root *root)
2093 free_extent_buffer(root->node);
2094 free_extent_buffer(root->commit_root);
2096 root->commit_root = NULL;
2100 /* helper to cleanup tree roots */
2101 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2103 free_root_extent_buffers(info->tree_root);
2105 free_root_extent_buffers(info->dev_root);
2106 free_root_extent_buffers(info->extent_root);
2107 free_root_extent_buffers(info->csum_root);
2108 free_root_extent_buffers(info->quota_root);
2109 free_root_extent_buffers(info->uuid_root);
2111 free_root_extent_buffers(info->chunk_root);
2114 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2117 struct btrfs_root *gang[8];
2120 while (!list_empty(&fs_info->dead_roots)) {
2121 gang[0] = list_entry(fs_info->dead_roots.next,
2122 struct btrfs_root, root_list);
2123 list_del(&gang[0]->root_list);
2125 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2126 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2128 free_extent_buffer(gang[0]->node);
2129 free_extent_buffer(gang[0]->commit_root);
2130 btrfs_put_fs_root(gang[0]);
2135 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2140 for (i = 0; i < ret; i++)
2141 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2144 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2145 btrfs_free_log_root_tree(NULL, fs_info);
2146 btrfs_destroy_pinned_extent(fs_info->tree_root,
2147 fs_info->pinned_extents);
2151 int open_ctree(struct super_block *sb,
2152 struct btrfs_fs_devices *fs_devices,
2161 struct btrfs_key location;
2162 struct buffer_head *bh;
2163 struct btrfs_super_block *disk_super;
2164 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2165 struct btrfs_root *tree_root;
2166 struct btrfs_root *extent_root;
2167 struct btrfs_root *csum_root;
2168 struct btrfs_root *chunk_root;
2169 struct btrfs_root *dev_root;
2170 struct btrfs_root *quota_root;
2171 struct btrfs_root *uuid_root;
2172 struct btrfs_root *log_tree_root;
2175 int num_backups_tried = 0;
2176 int backup_index = 0;
2178 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2179 bool create_uuid_tree;
2180 bool check_uuid_tree;
2182 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2183 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2184 if (!tree_root || !chunk_root) {
2189 ret = init_srcu_struct(&fs_info->subvol_srcu);
2195 ret = setup_bdi(fs_info, &fs_info->bdi);
2201 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2206 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2207 (1 + ilog2(nr_cpu_ids));
2209 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2212 goto fail_dirty_metadata_bytes;
2215 ret = percpu_counter_init(&fs_info->bio_counter, 0);
2218 goto fail_delalloc_bytes;
2221 fs_info->btree_inode = new_inode(sb);
2222 if (!fs_info->btree_inode) {
2224 goto fail_bio_counter;
2227 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2229 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2230 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2231 INIT_LIST_HEAD(&fs_info->trans_list);
2232 INIT_LIST_HEAD(&fs_info->dead_roots);
2233 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2234 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2235 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2236 spin_lock_init(&fs_info->delalloc_root_lock);
2237 spin_lock_init(&fs_info->trans_lock);
2238 spin_lock_init(&fs_info->fs_roots_radix_lock);
2239 spin_lock_init(&fs_info->delayed_iput_lock);
2240 spin_lock_init(&fs_info->defrag_inodes_lock);
2241 spin_lock_init(&fs_info->free_chunk_lock);
2242 spin_lock_init(&fs_info->tree_mod_seq_lock);
2243 spin_lock_init(&fs_info->super_lock);
2244 spin_lock_init(&fs_info->qgroup_op_lock);
2245 spin_lock_init(&fs_info->buffer_lock);
2246 spin_lock_init(&fs_info->unused_bgs_lock);
2247 rwlock_init(&fs_info->tree_mod_log_lock);
2248 mutex_init(&fs_info->reloc_mutex);
2249 mutex_init(&fs_info->delalloc_root_mutex);
2250 seqlock_init(&fs_info->profiles_lock);
2252 init_completion(&fs_info->kobj_unregister);
2253 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2254 INIT_LIST_HEAD(&fs_info->space_info);
2255 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2256 INIT_LIST_HEAD(&fs_info->unused_bgs);
2257 btrfs_mapping_init(&fs_info->mapping_tree);
2258 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2259 BTRFS_BLOCK_RSV_GLOBAL);
2260 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2261 BTRFS_BLOCK_RSV_DELALLOC);
2262 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2263 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2264 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2265 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2266 BTRFS_BLOCK_RSV_DELOPS);
2267 atomic_set(&fs_info->nr_async_submits, 0);
2268 atomic_set(&fs_info->async_delalloc_pages, 0);
2269 atomic_set(&fs_info->async_submit_draining, 0);
2270 atomic_set(&fs_info->nr_async_bios, 0);
2271 atomic_set(&fs_info->defrag_running, 0);
2272 atomic_set(&fs_info->qgroup_op_seq, 0);
2273 atomic64_set(&fs_info->tree_mod_seq, 0);
2275 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2276 fs_info->metadata_ratio = 0;
2277 fs_info->defrag_inodes = RB_ROOT;
2278 fs_info->free_chunk_space = 0;
2279 fs_info->tree_mod_log = RB_ROOT;
2280 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2281 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2282 /* readahead state */
2283 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2284 spin_lock_init(&fs_info->reada_lock);
2286 fs_info->thread_pool_size = min_t(unsigned long,
2287 num_online_cpus() + 2, 8);
2289 INIT_LIST_HEAD(&fs_info->ordered_roots);
2290 spin_lock_init(&fs_info->ordered_root_lock);
2291 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2293 if (!fs_info->delayed_root) {
2297 btrfs_init_delayed_root(fs_info->delayed_root);
2299 mutex_init(&fs_info->scrub_lock);
2300 atomic_set(&fs_info->scrubs_running, 0);
2301 atomic_set(&fs_info->scrub_pause_req, 0);
2302 atomic_set(&fs_info->scrubs_paused, 0);
2303 atomic_set(&fs_info->scrub_cancel_req, 0);
2304 init_waitqueue_head(&fs_info->replace_wait);
2305 init_waitqueue_head(&fs_info->scrub_pause_wait);
2306 fs_info->scrub_workers_refcnt = 0;
2307 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2308 fs_info->check_integrity_print_mask = 0;
2311 spin_lock_init(&fs_info->balance_lock);
2312 mutex_init(&fs_info->balance_mutex);
2313 atomic_set(&fs_info->balance_running, 0);
2314 atomic_set(&fs_info->balance_pause_req, 0);
2315 atomic_set(&fs_info->balance_cancel_req, 0);
2316 fs_info->balance_ctl = NULL;
2317 init_waitqueue_head(&fs_info->balance_wait_q);
2318 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2320 sb->s_blocksize = 4096;
2321 sb->s_blocksize_bits = blksize_bits(4096);
2322 sb->s_bdi = &fs_info->bdi;
2324 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2325 set_nlink(fs_info->btree_inode, 1);
2327 * we set the i_size on the btree inode to the max possible int.
2328 * the real end of the address space is determined by all of
2329 * the devices in the system
2331 fs_info->btree_inode->i_size = OFFSET_MAX;
2332 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2333 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2335 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2336 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2337 fs_info->btree_inode->i_mapping);
2338 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2339 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2341 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2343 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2344 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2345 sizeof(struct btrfs_key));
2346 set_bit(BTRFS_INODE_DUMMY,
2347 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2348 btrfs_insert_inode_hash(fs_info->btree_inode);
2350 spin_lock_init(&fs_info->block_group_cache_lock);
2351 fs_info->block_group_cache_tree = RB_ROOT;
2352 fs_info->first_logical_byte = (u64)-1;
2354 extent_io_tree_init(&fs_info->freed_extents[0],
2355 fs_info->btree_inode->i_mapping);
2356 extent_io_tree_init(&fs_info->freed_extents[1],
2357 fs_info->btree_inode->i_mapping);
2358 fs_info->pinned_extents = &fs_info->freed_extents[0];
2359 fs_info->do_barriers = 1;
2362 mutex_init(&fs_info->ordered_operations_mutex);
2363 mutex_init(&fs_info->ordered_extent_flush_mutex);
2364 mutex_init(&fs_info->tree_log_mutex);
2365 mutex_init(&fs_info->chunk_mutex);
2366 mutex_init(&fs_info->transaction_kthread_mutex);
2367 mutex_init(&fs_info->cleaner_mutex);
2368 mutex_init(&fs_info->volume_mutex);
2369 init_rwsem(&fs_info->commit_root_sem);
2370 init_rwsem(&fs_info->cleanup_work_sem);
2371 init_rwsem(&fs_info->subvol_sem);
2372 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2373 fs_info->dev_replace.lock_owner = 0;
2374 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2375 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2376 mutex_init(&fs_info->dev_replace.lock_management_lock);
2377 mutex_init(&fs_info->dev_replace.lock);
2379 spin_lock_init(&fs_info->qgroup_lock);
2380 mutex_init(&fs_info->qgroup_ioctl_lock);
2381 fs_info->qgroup_tree = RB_ROOT;
2382 fs_info->qgroup_op_tree = RB_ROOT;
2383 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2384 fs_info->qgroup_seq = 1;
2385 fs_info->quota_enabled = 0;
2386 fs_info->pending_quota_state = 0;
2387 fs_info->qgroup_ulist = NULL;
2388 mutex_init(&fs_info->qgroup_rescan_lock);
2390 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2391 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2393 init_waitqueue_head(&fs_info->transaction_throttle);
2394 init_waitqueue_head(&fs_info->transaction_wait);
2395 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2396 init_waitqueue_head(&fs_info->async_submit_wait);
2398 ret = btrfs_alloc_stripe_hash_table(fs_info);
2404 __setup_root(4096, 4096, 4096, tree_root,
2405 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2407 invalidate_bdev(fs_devices->latest_bdev);
2410 * Read super block and check the signature bytes only
2412 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2419 * We want to check superblock checksum, the type is stored inside.
2420 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2422 if (btrfs_check_super_csum(bh->b_data)) {
2423 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2429 * super_copy is zeroed at allocation time and we never touch the
2430 * following bytes up to INFO_SIZE, the checksum is calculated from
2431 * the whole block of INFO_SIZE
2433 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2434 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2435 sizeof(*fs_info->super_for_commit));
2438 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2440 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2442 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2447 disk_super = fs_info->super_copy;
2448 if (!btrfs_super_root(disk_super))
2451 /* check FS state, whether FS is broken. */
2452 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2453 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2456 * run through our array of backup supers and setup
2457 * our ring pointer to the oldest one
2459 generation = btrfs_super_generation(disk_super);
2460 find_oldest_super_backup(fs_info, generation);
2463 * In the long term, we'll store the compression type in the super
2464 * block, and it'll be used for per file compression control.
2466 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2468 ret = btrfs_parse_options(tree_root, options);
2474 features = btrfs_super_incompat_flags(disk_super) &
2475 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2477 printk(KERN_ERR "BTRFS: couldn't mount because of "
2478 "unsupported optional features (%Lx).\n",
2485 * Leafsize and nodesize were always equal, this is only a sanity check.
2487 if (le32_to_cpu(disk_super->__unused_leafsize) !=
2488 btrfs_super_nodesize(disk_super)) {
2489 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2490 "blocksizes don't match. node %d leaf %d\n",
2491 btrfs_super_nodesize(disk_super),
2492 le32_to_cpu(disk_super->__unused_leafsize));
2496 if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2497 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2498 "blocksize (%d) was too large\n",
2499 btrfs_super_nodesize(disk_super));
2504 features = btrfs_super_incompat_flags(disk_super);
2505 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2506 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2507 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2509 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2510 printk(KERN_ERR "BTRFS: has skinny extents\n");
2513 * flag our filesystem as having big metadata blocks if
2514 * they are bigger than the page size
2516 if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2517 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2518 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2519 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2522 nodesize = btrfs_super_nodesize(disk_super);
2523 sectorsize = btrfs_super_sectorsize(disk_super);
2524 stripesize = btrfs_super_stripesize(disk_super);
2525 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2526 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2529 * mixed block groups end up with duplicate but slightly offset
2530 * extent buffers for the same range. It leads to corruptions
2532 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2533 (sectorsize != nodesize)) {
2534 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2535 "are not allowed for mixed block groups on %s\n",
2541 * Needn't use the lock because there is no other task which will
2544 btrfs_set_super_incompat_flags(disk_super, features);
2546 features = btrfs_super_compat_ro_flags(disk_super) &
2547 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2548 if (!(sb->s_flags & MS_RDONLY) && features) {
2549 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2550 "unsupported option features (%Lx).\n",
2556 max_active = fs_info->thread_pool_size;
2559 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2562 fs_info->delalloc_workers =
2563 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2565 fs_info->flush_workers =
2566 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2568 fs_info->caching_workers =
2569 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2572 * a higher idle thresh on the submit workers makes it much more
2573 * likely that bios will be send down in a sane order to the
2576 fs_info->submit_workers =
2577 btrfs_alloc_workqueue("submit", flags,
2578 min_t(u64, fs_devices->num_devices,
2581 fs_info->fixup_workers =
2582 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2585 * endios are largely parallel and should have a very
2588 fs_info->endio_workers =
2589 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2590 fs_info->endio_meta_workers =
2591 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2592 fs_info->endio_meta_write_workers =
2593 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2594 fs_info->endio_raid56_workers =
2595 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2596 fs_info->endio_repair_workers =
2597 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2598 fs_info->rmw_workers =
2599 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2600 fs_info->endio_write_workers =
2601 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2602 fs_info->endio_freespace_worker =
2603 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2604 fs_info->delayed_workers =
2605 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2606 fs_info->readahead_workers =
2607 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2608 fs_info->qgroup_rescan_workers =
2609 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2610 fs_info->extent_workers =
2611 btrfs_alloc_workqueue("extent-refs", flags,
2612 min_t(u64, fs_devices->num_devices,
2615 if (!(fs_info->workers && fs_info->delalloc_workers &&
2616 fs_info->submit_workers && fs_info->flush_workers &&
2617 fs_info->endio_workers && fs_info->endio_meta_workers &&
2618 fs_info->endio_meta_write_workers &&
2619 fs_info->endio_repair_workers &&
2620 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2621 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2622 fs_info->caching_workers && fs_info->readahead_workers &&
2623 fs_info->fixup_workers && fs_info->delayed_workers &&
2624 fs_info->extent_workers &&
2625 fs_info->qgroup_rescan_workers)) {
2627 goto fail_sb_buffer;
2630 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2631 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2632 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2634 tree_root->nodesize = nodesize;
2635 tree_root->sectorsize = sectorsize;
2636 tree_root->stripesize = stripesize;
2638 sb->s_blocksize = sectorsize;
2639 sb->s_blocksize_bits = blksize_bits(sectorsize);
2641 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2642 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2643 goto fail_sb_buffer;
2646 if (sectorsize != PAGE_SIZE) {
2647 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2648 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2649 goto fail_sb_buffer;
2652 mutex_lock(&fs_info->chunk_mutex);
2653 ret = btrfs_read_sys_array(tree_root);
2654 mutex_unlock(&fs_info->chunk_mutex);
2656 printk(KERN_WARNING "BTRFS: failed to read the system "
2657 "array on %s\n", sb->s_id);
2658 goto fail_sb_buffer;
2661 blocksize = tree_root->nodesize;
2662 generation = btrfs_super_chunk_root_generation(disk_super);
2664 __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2665 fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2667 chunk_root->node = read_tree_block(chunk_root,
2668 btrfs_super_chunk_root(disk_super),
2669 blocksize, generation);
2670 if (!chunk_root->node ||
2671 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2672 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2674 goto fail_tree_roots;
2676 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2677 chunk_root->commit_root = btrfs_root_node(chunk_root);
2679 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2680 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2682 ret = btrfs_read_chunk_tree(chunk_root);
2684 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2686 goto fail_tree_roots;
2690 * keep the device that is marked to be the target device for the
2691 * dev_replace procedure
2693 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2695 if (!fs_devices->latest_bdev) {
2696 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2698 goto fail_tree_roots;
2702 blocksize = tree_root->nodesize;
2703 generation = btrfs_super_generation(disk_super);
2705 tree_root->node = read_tree_block(tree_root,
2706 btrfs_super_root(disk_super),
2707 blocksize, generation);
2708 if (!tree_root->node ||
2709 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2710 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2713 goto recovery_tree_root;
2716 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2717 tree_root->commit_root = btrfs_root_node(tree_root);
2718 btrfs_set_root_refs(&tree_root->root_item, 1);
2720 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2721 location.type = BTRFS_ROOT_ITEM_KEY;
2722 location.offset = 0;
2724 extent_root = btrfs_read_tree_root(tree_root, &location);
2725 if (IS_ERR(extent_root)) {
2726 ret = PTR_ERR(extent_root);
2727 goto recovery_tree_root;
2729 set_bit(BTRFS_ROOT_TRACK_DIRTY, &extent_root->state);
2730 fs_info->extent_root = extent_root;
2732 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2733 dev_root = btrfs_read_tree_root(tree_root, &location);
2734 if (IS_ERR(dev_root)) {
2735 ret = PTR_ERR(dev_root);
2736 goto recovery_tree_root;
2738 set_bit(BTRFS_ROOT_TRACK_DIRTY, &dev_root->state);
2739 fs_info->dev_root = dev_root;
2740 btrfs_init_devices_late(fs_info);
2742 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2743 csum_root = btrfs_read_tree_root(tree_root, &location);
2744 if (IS_ERR(csum_root)) {
2745 ret = PTR_ERR(csum_root);
2746 goto recovery_tree_root;
2748 set_bit(BTRFS_ROOT_TRACK_DIRTY, &csum_root->state);
2749 fs_info->csum_root = csum_root;
2751 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2752 quota_root = btrfs_read_tree_root(tree_root, &location);
2753 if (!IS_ERR(quota_root)) {
2754 set_bit(BTRFS_ROOT_TRACK_DIRTY, "a_root->state);
2755 fs_info->quota_enabled = 1;
2756 fs_info->pending_quota_state = 1;
2757 fs_info->quota_root = quota_root;
2760 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2761 uuid_root = btrfs_read_tree_root(tree_root, &location);
2762 if (IS_ERR(uuid_root)) {
2763 ret = PTR_ERR(uuid_root);
2765 goto recovery_tree_root;
2766 create_uuid_tree = true;
2767 check_uuid_tree = false;
2769 set_bit(BTRFS_ROOT_TRACK_DIRTY, &uuid_root->state);
2770 fs_info->uuid_root = uuid_root;
2771 create_uuid_tree = false;
2773 generation != btrfs_super_uuid_tree_generation(disk_super);
2776 fs_info->generation = generation;
2777 fs_info->last_trans_committed = generation;
2779 ret = btrfs_recover_balance(fs_info);
2781 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2782 goto fail_block_groups;
2785 ret = btrfs_init_dev_stats(fs_info);
2787 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2789 goto fail_block_groups;
2792 ret = btrfs_init_dev_replace(fs_info);
2794 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2795 goto fail_block_groups;
2798 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2800 ret = btrfs_sysfs_add_one(fs_info);
2802 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2803 goto fail_block_groups;
2806 ret = btrfs_init_space_info(fs_info);
2808 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2812 ret = btrfs_read_block_groups(extent_root);
2814 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2817 fs_info->num_tolerated_disk_barrier_failures =
2818 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2819 if (fs_info->fs_devices->missing_devices >
2820 fs_info->num_tolerated_disk_barrier_failures &&
2821 !(sb->s_flags & MS_RDONLY)) {
2822 printk(KERN_WARNING "BTRFS: "
2823 "too many missing devices, writeable mount is not allowed\n");
2827 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2829 if (IS_ERR(fs_info->cleaner_kthread))
2832 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2834 "btrfs-transaction");
2835 if (IS_ERR(fs_info->transaction_kthread))
2838 if (!btrfs_test_opt(tree_root, SSD) &&
2839 !btrfs_test_opt(tree_root, NOSSD) &&
2840 !fs_info->fs_devices->rotating) {
2841 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2843 btrfs_set_opt(fs_info->mount_opt, SSD);
2846 /* Set the real inode map cache flag */
2847 if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2848 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2850 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2851 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2852 ret = btrfsic_mount(tree_root, fs_devices,
2853 btrfs_test_opt(tree_root,
2854 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2856 fs_info->check_integrity_print_mask);
2858 printk(KERN_WARNING "BTRFS: failed to initialize"
2859 " integrity check module %s\n", sb->s_id);
2862 ret = btrfs_read_qgroup_config(fs_info);
2864 goto fail_trans_kthread;
2866 /* do not make disk changes in broken FS */
2867 if (btrfs_super_log_root(disk_super) != 0) {
2868 u64 bytenr = btrfs_super_log_root(disk_super);
2870 if (fs_devices->rw_devices == 0) {
2871 printk(KERN_WARNING "BTRFS: log replay required "
2876 blocksize = tree_root->nodesize;
2878 log_tree_root = btrfs_alloc_root(fs_info);
2879 if (!log_tree_root) {
2884 __setup_root(nodesize, sectorsize, stripesize,
2885 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2887 log_tree_root->node = read_tree_block(tree_root, bytenr,
2890 if (!log_tree_root->node ||
2891 !extent_buffer_uptodate(log_tree_root->node)) {
2892 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2893 free_extent_buffer(log_tree_root->node);
2894 kfree(log_tree_root);
2897 /* returns with log_tree_root freed on success */
2898 ret = btrfs_recover_log_trees(log_tree_root);
2900 btrfs_error(tree_root->fs_info, ret,
2901 "Failed to recover log tree");
2902 free_extent_buffer(log_tree_root->node);
2903 kfree(log_tree_root);
2907 if (sb->s_flags & MS_RDONLY) {
2908 ret = btrfs_commit_super(tree_root);
2914 ret = btrfs_find_orphan_roots(tree_root);
2918 if (!(sb->s_flags & MS_RDONLY)) {
2919 ret = btrfs_cleanup_fs_roots(fs_info);
2923 mutex_lock(&fs_info->cleaner_mutex);
2924 ret = btrfs_recover_relocation(tree_root);
2925 mutex_unlock(&fs_info->cleaner_mutex);
2928 "BTRFS: failed to recover relocation\n");
2934 location.objectid = BTRFS_FS_TREE_OBJECTID;
2935 location.type = BTRFS_ROOT_ITEM_KEY;
2936 location.offset = 0;
2938 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2939 if (IS_ERR(fs_info->fs_root)) {
2940 err = PTR_ERR(fs_info->fs_root);
2944 if (sb->s_flags & MS_RDONLY)
2947 down_read(&fs_info->cleanup_work_sem);
2948 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2949 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2950 up_read(&fs_info->cleanup_work_sem);
2951 close_ctree(tree_root);
2954 up_read(&fs_info->cleanup_work_sem);
2956 ret = btrfs_resume_balance_async(fs_info);
2958 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2959 close_ctree(tree_root);
2963 ret = btrfs_resume_dev_replace_async(fs_info);
2965 pr_warn("BTRFS: failed to resume dev_replace\n");
2966 close_ctree(tree_root);
2970 btrfs_qgroup_rescan_resume(fs_info);
2972 if (create_uuid_tree) {
2973 pr_info("BTRFS: creating UUID tree\n");
2974 ret = btrfs_create_uuid_tree(fs_info);
2976 pr_warn("BTRFS: failed to create the UUID tree %d\n",
2978 close_ctree(tree_root);
2981 } else if (check_uuid_tree ||
2982 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2983 pr_info("BTRFS: checking UUID tree\n");
2984 ret = btrfs_check_uuid_tree(fs_info);
2986 pr_warn("BTRFS: failed to check the UUID tree %d\n",
2988 close_ctree(tree_root);
2992 fs_info->update_uuid_tree_gen = 1;
3000 btrfs_free_qgroup_config(fs_info);
3002 kthread_stop(fs_info->transaction_kthread);
3003 btrfs_cleanup_transaction(fs_info->tree_root);
3004 btrfs_free_fs_roots(fs_info);
3006 kthread_stop(fs_info->cleaner_kthread);
3009 * make sure we're done with the btree inode before we stop our
3012 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3015 btrfs_sysfs_remove_one(fs_info);
3018 btrfs_put_block_group_cache(fs_info);
3019 btrfs_free_block_groups(fs_info);
3022 free_root_pointers(fs_info, 1);
3023 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3026 btrfs_stop_all_workers(fs_info);
3029 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3031 iput(fs_info->btree_inode);
3033 percpu_counter_destroy(&fs_info->bio_counter);
3034 fail_delalloc_bytes:
3035 percpu_counter_destroy(&fs_info->delalloc_bytes);
3036 fail_dirty_metadata_bytes:
3037 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3039 bdi_destroy(&fs_info->bdi);
3041 cleanup_srcu_struct(&fs_info->subvol_srcu);
3043 btrfs_free_stripe_hash_table(fs_info);
3044 btrfs_close_devices(fs_info->fs_devices);
3048 if (!btrfs_test_opt(tree_root, RECOVERY))
3049 goto fail_tree_roots;
3051 free_root_pointers(fs_info, 0);
3053 /* don't use the log in recovery mode, it won't be valid */
3054 btrfs_set_super_log_root(disk_super, 0);
3056 /* we can't trust the free space cache either */
3057 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3059 ret = next_root_backup(fs_info, fs_info->super_copy,
3060 &num_backups_tried, &backup_index);
3062 goto fail_block_groups;
3063 goto retry_root_backup;
3066 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3069 set_buffer_uptodate(bh);
3071 struct btrfs_device *device = (struct btrfs_device *)
3074 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3075 "I/O error on %s\n",
3076 rcu_str_deref(device->name));
3077 /* note, we dont' set_buffer_write_io_error because we have
3078 * our own ways of dealing with the IO errors
3080 clear_buffer_uptodate(bh);
3081 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3087 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3089 struct buffer_head *bh;
3090 struct buffer_head *latest = NULL;
3091 struct btrfs_super_block *super;
3096 /* we would like to check all the supers, but that would make
3097 * a btrfs mount succeed after a mkfs from a different FS.
3098 * So, we need to add a special mount option to scan for
3099 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3101 for (i = 0; i < 1; i++) {
3102 bytenr = btrfs_sb_offset(i);
3103 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3104 i_size_read(bdev->bd_inode))
3106 bh = __bread(bdev, bytenr / 4096,
3107 BTRFS_SUPER_INFO_SIZE);
3111 super = (struct btrfs_super_block *)bh->b_data;
3112 if (btrfs_super_bytenr(super) != bytenr ||
3113 btrfs_super_magic(super) != BTRFS_MAGIC) {
3118 if (!latest || btrfs_super_generation(super) > transid) {
3121 transid = btrfs_super_generation(super);
3130 * this should be called twice, once with wait == 0 and
3131 * once with wait == 1. When wait == 0 is done, all the buffer heads
3132 * we write are pinned.
3134 * They are released when wait == 1 is done.
3135 * max_mirrors must be the same for both runs, and it indicates how
3136 * many supers on this one device should be written.
3138 * max_mirrors == 0 means to write them all.
3140 static int write_dev_supers(struct btrfs_device *device,
3141 struct btrfs_super_block *sb,
3142 int do_barriers, int wait, int max_mirrors)
3144 struct buffer_head *bh;
3151 if (max_mirrors == 0)
3152 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3154 for (i = 0; i < max_mirrors; i++) {
3155 bytenr = btrfs_sb_offset(i);
3156 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3157 device->commit_total_bytes)
3161 bh = __find_get_block(device->bdev, bytenr / 4096,
3162 BTRFS_SUPER_INFO_SIZE);
3168 if (!buffer_uptodate(bh))
3171 /* drop our reference */
3174 /* drop the reference from the wait == 0 run */
3178 btrfs_set_super_bytenr(sb, bytenr);
3181 crc = btrfs_csum_data((char *)sb +
3182 BTRFS_CSUM_SIZE, crc,
3183 BTRFS_SUPER_INFO_SIZE -
3185 btrfs_csum_final(crc, sb->csum);
3188 * one reference for us, and we leave it for the
3191 bh = __getblk(device->bdev, bytenr / 4096,
3192 BTRFS_SUPER_INFO_SIZE);
3194 printk(KERN_ERR "BTRFS: couldn't get super "
3195 "buffer head for bytenr %Lu\n", bytenr);
3200 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3202 /* one reference for submit_bh */
3205 set_buffer_uptodate(bh);
3207 bh->b_end_io = btrfs_end_buffer_write_sync;
3208 bh->b_private = device;
3212 * we fua the first super. The others we allow
3216 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3218 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3222 return errors < i ? 0 : -1;
3226 * endio for the write_dev_flush, this will wake anyone waiting
3227 * for the barrier when it is done
3229 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3232 if (err == -EOPNOTSUPP)
3233 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3234 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3236 if (bio->bi_private)
3237 complete(bio->bi_private);
3242 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3243 * sent down. With wait == 1, it waits for the previous flush.
3245 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3248 static int write_dev_flush(struct btrfs_device *device, int wait)
3253 if (device->nobarriers)
3257 bio = device->flush_bio;
3261 wait_for_completion(&device->flush_wait);
3263 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3264 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3265 rcu_str_deref(device->name));
3266 device->nobarriers = 1;
3267 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3269 btrfs_dev_stat_inc_and_print(device,
3270 BTRFS_DEV_STAT_FLUSH_ERRS);
3273 /* drop the reference from the wait == 0 run */
3275 device->flush_bio = NULL;
3281 * one reference for us, and we leave it for the
3284 device->flush_bio = NULL;
3285 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3289 bio->bi_end_io = btrfs_end_empty_barrier;
3290 bio->bi_bdev = device->bdev;
3291 init_completion(&device->flush_wait);
3292 bio->bi_private = &device->flush_wait;
3293 device->flush_bio = bio;
3296 btrfsic_submit_bio(WRITE_FLUSH, bio);
3302 * send an empty flush down to each device in parallel,
3303 * then wait for them
3305 static int barrier_all_devices(struct btrfs_fs_info *info)
3307 struct list_head *head;
3308 struct btrfs_device *dev;
3309 int errors_send = 0;
3310 int errors_wait = 0;
3313 /* send down all the barriers */
3314 head = &info->fs_devices->devices;
3315 list_for_each_entry_rcu(dev, head, dev_list) {
3322 if (!dev->in_fs_metadata || !dev->writeable)
3325 ret = write_dev_flush(dev, 0);
3330 /* wait for all the barriers */
3331 list_for_each_entry_rcu(dev, head, dev_list) {
3338 if (!dev->in_fs_metadata || !dev->writeable)
3341 ret = write_dev_flush(dev, 1);
3345 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3346 errors_wait > info->num_tolerated_disk_barrier_failures)
3351 int btrfs_calc_num_tolerated_disk_barrier_failures(
3352 struct btrfs_fs_info *fs_info)
3354 struct btrfs_ioctl_space_info space;
3355 struct btrfs_space_info *sinfo;
3356 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3357 BTRFS_BLOCK_GROUP_SYSTEM,
3358 BTRFS_BLOCK_GROUP_METADATA,
3359 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3363 int num_tolerated_disk_barrier_failures =
3364 (int)fs_info->fs_devices->num_devices;
3366 for (i = 0; i < num_types; i++) {
3367 struct btrfs_space_info *tmp;
3371 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3372 if (tmp->flags == types[i]) {
3382 down_read(&sinfo->groups_sem);
3383 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3384 if (!list_empty(&sinfo->block_groups[c])) {
3387 btrfs_get_block_group_info(
3388 &sinfo->block_groups[c], &space);
3389 if (space.total_bytes == 0 ||
3390 space.used_bytes == 0)
3392 flags = space.flags;
3395 * 0: if dup, single or RAID0 is configured for
3396 * any of metadata, system or data, else
3397 * 1: if RAID5 is configured, or if RAID1 or
3398 * RAID10 is configured and only two mirrors
3400 * 2: if RAID6 is configured, else
3401 * num_mirrors - 1: if RAID1 or RAID10 is
3402 * configured and more than
3403 * 2 mirrors are used.
3405 if (num_tolerated_disk_barrier_failures > 0 &&
3406 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3407 BTRFS_BLOCK_GROUP_RAID0)) ||
3408 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3410 num_tolerated_disk_barrier_failures = 0;
3411 else if (num_tolerated_disk_barrier_failures > 1) {
3412 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3413 BTRFS_BLOCK_GROUP_RAID5 |
3414 BTRFS_BLOCK_GROUP_RAID10)) {
3415 num_tolerated_disk_barrier_failures = 1;
3417 BTRFS_BLOCK_GROUP_RAID6) {
3418 num_tolerated_disk_barrier_failures = 2;
3423 up_read(&sinfo->groups_sem);
3426 return num_tolerated_disk_barrier_failures;
3429 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3431 struct list_head *head;
3432 struct btrfs_device *dev;
3433 struct btrfs_super_block *sb;
3434 struct btrfs_dev_item *dev_item;
3438 int total_errors = 0;
3441 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3442 backup_super_roots(root->fs_info);
3444 sb = root->fs_info->super_for_commit;
3445 dev_item = &sb->dev_item;
3447 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3448 head = &root->fs_info->fs_devices->devices;
3449 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3452 ret = barrier_all_devices(root->fs_info);
3455 &root->fs_info->fs_devices->device_list_mutex);
3456 btrfs_error(root->fs_info, ret,
3457 "errors while submitting device barriers.");
3462 list_for_each_entry_rcu(dev, head, dev_list) {
3467 if (!dev->in_fs_metadata || !dev->writeable)
3470 btrfs_set_stack_device_generation(dev_item, 0);
3471 btrfs_set_stack_device_type(dev_item, dev->type);
3472 btrfs_set_stack_device_id(dev_item, dev->devid);
3473 btrfs_set_stack_device_total_bytes(dev_item,
3474 dev->commit_total_bytes);
3475 btrfs_set_stack_device_bytes_used(dev_item,
3476 dev->commit_bytes_used);
3477 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3478 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3479 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3480 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3481 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3483 flags = btrfs_super_flags(sb);
3484 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3486 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3490 if (total_errors > max_errors) {
3491 btrfs_err(root->fs_info, "%d errors while writing supers",
3493 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3495 /* FUA is masked off if unsupported and can't be the reason */
3496 btrfs_error(root->fs_info, -EIO,
3497 "%d errors while writing supers", total_errors);
3502 list_for_each_entry_rcu(dev, head, dev_list) {
3505 if (!dev->in_fs_metadata || !dev->writeable)
3508 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3512 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3513 if (total_errors > max_errors) {
3514 btrfs_error(root->fs_info, -EIO,
3515 "%d errors while writing supers", total_errors);
3521 int write_ctree_super(struct btrfs_trans_handle *trans,
3522 struct btrfs_root *root, int max_mirrors)
3524 return write_all_supers(root, max_mirrors);
3527 /* Drop a fs root from the radix tree and free it. */
3528 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3529 struct btrfs_root *root)
3531 spin_lock(&fs_info->fs_roots_radix_lock);
3532 radix_tree_delete(&fs_info->fs_roots_radix,
3533 (unsigned long)root->root_key.objectid);
3534 spin_unlock(&fs_info->fs_roots_radix_lock);
3536 if (btrfs_root_refs(&root->root_item) == 0)
3537 synchronize_srcu(&fs_info->subvol_srcu);
3539 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3540 btrfs_free_log(NULL, root);
3542 if (root->free_ino_pinned)
3543 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3544 if (root->free_ino_ctl)
3545 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3549 static void free_fs_root(struct btrfs_root *root)
3551 iput(root->ino_cache_inode);
3552 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3553 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3554 root->orphan_block_rsv = NULL;
3556 free_anon_bdev(root->anon_dev);
3557 if (root->subv_writers)
3558 btrfs_free_subvolume_writers(root->subv_writers);
3559 free_extent_buffer(root->node);
3560 free_extent_buffer(root->commit_root);
3561 kfree(root->free_ino_ctl);
3562 kfree(root->free_ino_pinned);
3564 btrfs_put_fs_root(root);
3567 void btrfs_free_fs_root(struct btrfs_root *root)
3572 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3574 u64 root_objectid = 0;
3575 struct btrfs_root *gang[8];
3578 unsigned int ret = 0;
3582 index = srcu_read_lock(&fs_info->subvol_srcu);
3583 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3584 (void **)gang, root_objectid,
3587 srcu_read_unlock(&fs_info->subvol_srcu, index);
3590 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3592 for (i = 0; i < ret; i++) {
3593 /* Avoid to grab roots in dead_roots */
3594 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3598 /* grab all the search result for later use */
3599 gang[i] = btrfs_grab_fs_root(gang[i]);
3601 srcu_read_unlock(&fs_info->subvol_srcu, index);
3603 for (i = 0; i < ret; i++) {
3606 root_objectid = gang[i]->root_key.objectid;
3607 err = btrfs_orphan_cleanup(gang[i]);
3610 btrfs_put_fs_root(gang[i]);
3615 /* release the uncleaned roots due to error */
3616 for (; i < ret; i++) {
3618 btrfs_put_fs_root(gang[i]);
3623 int btrfs_commit_super(struct btrfs_root *root)
3625 struct btrfs_trans_handle *trans;
3627 mutex_lock(&root->fs_info->cleaner_mutex);
3628 btrfs_run_delayed_iputs(root);
3629 mutex_unlock(&root->fs_info->cleaner_mutex);
3630 wake_up_process(root->fs_info->cleaner_kthread);
3632 /* wait until ongoing cleanup work done */
3633 down_write(&root->fs_info->cleanup_work_sem);
3634 up_write(&root->fs_info->cleanup_work_sem);
3636 trans = btrfs_join_transaction(root);
3638 return PTR_ERR(trans);
3639 return btrfs_commit_transaction(trans, root);
3642 void close_ctree(struct btrfs_root *root)
3644 struct btrfs_fs_info *fs_info = root->fs_info;
3647 fs_info->closing = 1;
3650 /* wait for the uuid_scan task to finish */
3651 down(&fs_info->uuid_tree_rescan_sem);
3652 /* avoid complains from lockdep et al., set sem back to initial state */
3653 up(&fs_info->uuid_tree_rescan_sem);
3655 /* pause restriper - we want to resume on mount */
3656 btrfs_pause_balance(fs_info);
3658 btrfs_dev_replace_suspend_for_unmount(fs_info);
3660 btrfs_scrub_cancel(fs_info);
3662 /* wait for any defraggers to finish */
3663 wait_event(fs_info->transaction_wait,
3664 (atomic_read(&fs_info->defrag_running) == 0));
3666 /* clear out the rbtree of defraggable inodes */
3667 btrfs_cleanup_defrag_inodes(fs_info);
3669 cancel_work_sync(&fs_info->async_reclaim_work);
3671 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3672 ret = btrfs_commit_super(root);
3674 btrfs_err(root->fs_info, "commit super ret %d", ret);
3677 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3678 btrfs_error_commit_super(root);
3680 kthread_stop(fs_info->transaction_kthread);
3681 kthread_stop(fs_info->cleaner_kthread);
3683 fs_info->closing = 2;
3686 btrfs_free_qgroup_config(root->fs_info);
3688 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3689 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3690 percpu_counter_sum(&fs_info->delalloc_bytes));
3693 btrfs_sysfs_remove_one(fs_info);
3695 btrfs_free_fs_roots(fs_info);
3697 btrfs_put_block_group_cache(fs_info);
3699 btrfs_free_block_groups(fs_info);
3702 * we must make sure there is not any read request to
3703 * submit after we stopping all workers.
3705 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3706 btrfs_stop_all_workers(fs_info);
3709 free_root_pointers(fs_info, 1);
3711 iput(fs_info->btree_inode);
3713 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3714 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3715 btrfsic_unmount(root, fs_info->fs_devices);
3718 btrfs_close_devices(fs_info->fs_devices);
3719 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3721 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3722 percpu_counter_destroy(&fs_info->delalloc_bytes);
3723 percpu_counter_destroy(&fs_info->bio_counter);
3724 bdi_destroy(&fs_info->bdi);
3725 cleanup_srcu_struct(&fs_info->subvol_srcu);
3727 btrfs_free_stripe_hash_table(fs_info);
3729 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3730 root->orphan_block_rsv = NULL;
3733 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3737 struct inode *btree_inode = buf->pages[0]->mapping->host;
3739 ret = extent_buffer_uptodate(buf);
3743 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3744 parent_transid, atomic);
3750 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3752 return set_extent_buffer_uptodate(buf);
3755 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3757 struct btrfs_root *root;
3758 u64 transid = btrfs_header_generation(buf);
3761 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3763 * This is a fast path so only do this check if we have sanity tests
3764 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3765 * outside of the sanity tests.
3767 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3770 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3771 btrfs_assert_tree_locked(buf);
3772 if (transid != root->fs_info->generation)
3773 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3774 "found %llu running %llu\n",
3775 buf->start, transid, root->fs_info->generation);
3776 was_dirty = set_extent_buffer_dirty(buf);
3778 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3780 root->fs_info->dirty_metadata_batch);
3781 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3782 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3783 btrfs_print_leaf(root, buf);
3789 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3793 * looks as though older kernels can get into trouble with
3794 * this code, they end up stuck in balance_dirty_pages forever
3798 if (current->flags & PF_MEMALLOC)
3802 btrfs_balance_delayed_items(root);
3804 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3805 BTRFS_DIRTY_METADATA_THRESH);
3807 balance_dirty_pages_ratelimited(
3808 root->fs_info->btree_inode->i_mapping);
3813 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3815 __btrfs_btree_balance_dirty(root, 1);
3818 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3820 __btrfs_btree_balance_dirty(root, 0);
3823 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3825 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3826 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3829 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3833 * Placeholder for checks
3838 static void btrfs_error_commit_super(struct btrfs_root *root)
3840 mutex_lock(&root->fs_info->cleaner_mutex);
3841 btrfs_run_delayed_iputs(root);
3842 mutex_unlock(&root->fs_info->cleaner_mutex);
3844 down_write(&root->fs_info->cleanup_work_sem);
3845 up_write(&root->fs_info->cleanup_work_sem);
3847 /* cleanup FS via transaction */
3848 btrfs_cleanup_transaction(root);
3851 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3853 struct btrfs_ordered_extent *ordered;
3855 spin_lock(&root->ordered_extent_lock);
3857 * This will just short circuit the ordered completion stuff which will
3858 * make sure the ordered extent gets properly cleaned up.
3860 list_for_each_entry(ordered, &root->ordered_extents,
3862 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3863 spin_unlock(&root->ordered_extent_lock);
3866 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3868 struct btrfs_root *root;
3869 struct list_head splice;
3871 INIT_LIST_HEAD(&splice);
3873 spin_lock(&fs_info->ordered_root_lock);
3874 list_splice_init(&fs_info->ordered_roots, &splice);
3875 while (!list_empty(&splice)) {
3876 root = list_first_entry(&splice, struct btrfs_root,
3878 list_move_tail(&root->ordered_root,
3879 &fs_info->ordered_roots);
3881 spin_unlock(&fs_info->ordered_root_lock);
3882 btrfs_destroy_ordered_extents(root);
3885 spin_lock(&fs_info->ordered_root_lock);
3887 spin_unlock(&fs_info->ordered_root_lock);
3890 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3891 struct btrfs_root *root)
3893 struct rb_node *node;
3894 struct btrfs_delayed_ref_root *delayed_refs;
3895 struct btrfs_delayed_ref_node *ref;
3898 delayed_refs = &trans->delayed_refs;
3900 spin_lock(&delayed_refs->lock);
3901 if (atomic_read(&delayed_refs->num_entries) == 0) {
3902 spin_unlock(&delayed_refs->lock);
3903 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3907 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3908 struct btrfs_delayed_ref_head *head;
3909 bool pin_bytes = false;
3911 head = rb_entry(node, struct btrfs_delayed_ref_head,
3913 if (!mutex_trylock(&head->mutex)) {
3914 atomic_inc(&head->node.refs);
3915 spin_unlock(&delayed_refs->lock);
3917 mutex_lock(&head->mutex);
3918 mutex_unlock(&head->mutex);
3919 btrfs_put_delayed_ref(&head->node);
3920 spin_lock(&delayed_refs->lock);
3923 spin_lock(&head->lock);
3924 while ((node = rb_first(&head->ref_root)) != NULL) {
3925 ref = rb_entry(node, struct btrfs_delayed_ref_node,
3928 rb_erase(&ref->rb_node, &head->ref_root);
3929 atomic_dec(&delayed_refs->num_entries);
3930 btrfs_put_delayed_ref(ref);
3932 if (head->must_insert_reserved)
3934 btrfs_free_delayed_extent_op(head->extent_op);
3935 delayed_refs->num_heads--;
3936 if (head->processing == 0)
3937 delayed_refs->num_heads_ready--;
3938 atomic_dec(&delayed_refs->num_entries);
3939 head->node.in_tree = 0;
3940 rb_erase(&head->href_node, &delayed_refs->href_root);
3941 spin_unlock(&head->lock);
3942 spin_unlock(&delayed_refs->lock);
3943 mutex_unlock(&head->mutex);
3946 btrfs_pin_extent(root, head->node.bytenr,
3947 head->node.num_bytes, 1);
3948 btrfs_put_delayed_ref(&head->node);
3950 spin_lock(&delayed_refs->lock);
3953 spin_unlock(&delayed_refs->lock);
3958 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3960 struct btrfs_inode *btrfs_inode;
3961 struct list_head splice;
3963 INIT_LIST_HEAD(&splice);
3965 spin_lock(&root->delalloc_lock);
3966 list_splice_init(&root->delalloc_inodes, &splice);
3968 while (!list_empty(&splice)) {
3969 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3972 list_del_init(&btrfs_inode->delalloc_inodes);
3973 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3974 &btrfs_inode->runtime_flags);
3975 spin_unlock(&root->delalloc_lock);
3977 btrfs_invalidate_inodes(btrfs_inode->root);
3979 spin_lock(&root->delalloc_lock);
3982 spin_unlock(&root->delalloc_lock);
3985 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3987 struct btrfs_root *root;
3988 struct list_head splice;
3990 INIT_LIST_HEAD(&splice);
3992 spin_lock(&fs_info->delalloc_root_lock);
3993 list_splice_init(&fs_info->delalloc_roots, &splice);
3994 while (!list_empty(&splice)) {
3995 root = list_first_entry(&splice, struct btrfs_root,
3997 list_del_init(&root->delalloc_root);
3998 root = btrfs_grab_fs_root(root);
4000 spin_unlock(&fs_info->delalloc_root_lock);
4002 btrfs_destroy_delalloc_inodes(root);
4003 btrfs_put_fs_root(root);
4005 spin_lock(&fs_info->delalloc_root_lock);
4007 spin_unlock(&fs_info->delalloc_root_lock);
4010 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4011 struct extent_io_tree *dirty_pages,
4015 struct extent_buffer *eb;
4020 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4025 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4026 while (start <= end) {
4027 eb = btrfs_find_tree_block(root, start,
4029 start += root->nodesize;
4032 wait_on_extent_buffer_writeback(eb);
4034 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4036 clear_extent_buffer_dirty(eb);
4037 free_extent_buffer_stale(eb);
4044 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4045 struct extent_io_tree *pinned_extents)
4047 struct extent_io_tree *unpin;
4053 unpin = pinned_extents;
4056 ret = find_first_extent_bit(unpin, 0, &start, &end,
4057 EXTENT_DIRTY, NULL);
4062 if (btrfs_test_opt(root, DISCARD))
4063 ret = btrfs_error_discard_extent(root, start,
4067 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4068 btrfs_error_unpin_extent_range(root, start, end);
4073 if (unpin == &root->fs_info->freed_extents[0])
4074 unpin = &root->fs_info->freed_extents[1];
4076 unpin = &root->fs_info->freed_extents[0];
4084 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4085 struct btrfs_root *root)
4087 btrfs_destroy_delayed_refs(cur_trans, root);
4089 cur_trans->state = TRANS_STATE_COMMIT_START;
4090 wake_up(&root->fs_info->transaction_blocked_wait);
4092 cur_trans->state = TRANS_STATE_UNBLOCKED;
4093 wake_up(&root->fs_info->transaction_wait);
4095 btrfs_destroy_delayed_inodes(root);
4096 btrfs_assert_delayed_root_empty(root);
4098 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4100 btrfs_destroy_pinned_extent(root,
4101 root->fs_info->pinned_extents);
4103 cur_trans->state =TRANS_STATE_COMPLETED;
4104 wake_up(&cur_trans->commit_wait);
4107 memset(cur_trans, 0, sizeof(*cur_trans));
4108 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4112 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4114 struct btrfs_transaction *t;
4116 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4118 spin_lock(&root->fs_info->trans_lock);
4119 while (!list_empty(&root->fs_info->trans_list)) {
4120 t = list_first_entry(&root->fs_info->trans_list,
4121 struct btrfs_transaction, list);
4122 if (t->state >= TRANS_STATE_COMMIT_START) {
4123 atomic_inc(&t->use_count);
4124 spin_unlock(&root->fs_info->trans_lock);
4125 btrfs_wait_for_commit(root, t->transid);
4126 btrfs_put_transaction(t);
4127 spin_lock(&root->fs_info->trans_lock);
4130 if (t == root->fs_info->running_transaction) {
4131 t->state = TRANS_STATE_COMMIT_DOING;
4132 spin_unlock(&root->fs_info->trans_lock);
4134 * We wait for 0 num_writers since we don't hold a trans
4135 * handle open currently for this transaction.
4137 wait_event(t->writer_wait,
4138 atomic_read(&t->num_writers) == 0);
4140 spin_unlock(&root->fs_info->trans_lock);
4142 btrfs_cleanup_one_transaction(t, root);
4144 spin_lock(&root->fs_info->trans_lock);
4145 if (t == root->fs_info->running_transaction)
4146 root->fs_info->running_transaction = NULL;
4147 list_del_init(&t->list);
4148 spin_unlock(&root->fs_info->trans_lock);
4150 btrfs_put_transaction(t);
4151 trace_btrfs_transaction_commit(root);
4152 spin_lock(&root->fs_info->trans_lock);
4154 spin_unlock(&root->fs_info->trans_lock);
4155 btrfs_destroy_all_ordered_extents(root->fs_info);
4156 btrfs_destroy_delayed_inodes(root);
4157 btrfs_assert_delayed_root_empty(root);
4158 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4159 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4160 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4165 static struct extent_io_ops btree_extent_io_ops = {
4166 .readpage_end_io_hook = btree_readpage_end_io_hook,
4167 .readpage_io_failed_hook = btree_io_failed_hook,
4168 .submit_bio_hook = btree_submit_bio_hook,
4169 /* note we're sharing with inode.c for the merge bio hook */
4170 .merge_bio_hook = btrfs_merge_bio_hook,