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/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <linux/uuid.h>
34 #include <linux/semaphore.h>
35 #include <asm/unaligned.h>
39 #include "transaction.h"
40 #include "btrfs_inode.h"
42 #include "print-tree.h"
43 #include "async-thread.h"
46 #include "free-space-cache.h"
47 #include "inode-map.h"
48 #include "check-integrity.h"
49 #include "rcu-string.h"
50 #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_operations(struct btrfs_transaction *t,
63 struct btrfs_root *root);
64 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
65 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
66 struct btrfs_root *root);
67 static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
68 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
69 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
70 struct extent_io_tree *dirty_pages,
72 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
73 struct extent_io_tree *pinned_extents);
74 static int btrfs_cleanup_transaction(struct btrfs_root *root);
75 static void btrfs_error_commit_super(struct btrfs_root *root);
78 * end_io_wq structs are used to do processing in task context when an IO is
79 * complete. This is used during reads to verify checksums, and it is used
80 * by writes to insert metadata for new file extents after IO is complete.
86 struct btrfs_fs_info *info;
89 struct list_head list;
90 struct btrfs_work work;
94 * async submit bios are used to offload expensive checksumming
95 * onto the worker threads. They checksum file and metadata bios
96 * just before they are sent down the IO stack.
98 struct async_submit_bio {
101 struct list_head list;
102 extent_submit_bio_hook_t *submit_bio_start;
103 extent_submit_bio_hook_t *submit_bio_done;
106 unsigned long bio_flags;
108 * bio_offset is optional, can be used if the pages in the bio
109 * can't tell us where in the file the bio should go
112 struct btrfs_work work;
117 * Lockdep class keys for extent_buffer->lock's in this root. For a given
118 * eb, the lockdep key is determined by the btrfs_root it belongs to and
119 * the level the eb occupies in the tree.
121 * Different roots are used for different purposes and may nest inside each
122 * other and they require separate keysets. As lockdep keys should be
123 * static, assign keysets according to the purpose of the root as indicated
124 * by btrfs_root->objectid. This ensures that all special purpose roots
125 * have separate keysets.
127 * Lock-nesting across peer nodes is always done with the immediate parent
128 * node locked thus preventing deadlock. As lockdep doesn't know this, use
129 * subclass to avoid triggering lockdep warning in such cases.
131 * The key is set by the readpage_end_io_hook after the buffer has passed
132 * csum validation but before the pages are unlocked. It is also set by
133 * btrfs_init_new_buffer on freshly allocated blocks.
135 * We also add a check to make sure the highest level of the tree is the
136 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
137 * needs update as well.
139 #ifdef CONFIG_DEBUG_LOCK_ALLOC
140 # if BTRFS_MAX_LEVEL != 8
144 static struct btrfs_lockdep_keyset {
145 u64 id; /* root objectid */
146 const char *name_stem; /* lock name stem */
147 char names[BTRFS_MAX_LEVEL + 1][20];
148 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
149 } btrfs_lockdep_keysets[] = {
150 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
151 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
152 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
153 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
154 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
155 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
156 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
157 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
158 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
159 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
160 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
161 { .id = 0, .name_stem = "tree" },
164 void __init btrfs_init_lockdep(void)
168 /* initialize lockdep class names */
169 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
170 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
172 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
173 snprintf(ks->names[j], sizeof(ks->names[j]),
174 "btrfs-%s-%02d", ks->name_stem, j);
178 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
181 struct btrfs_lockdep_keyset *ks;
183 BUG_ON(level >= ARRAY_SIZE(ks->keys));
185 /* find the matching keyset, id 0 is the default entry */
186 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
187 if (ks->id == objectid)
190 lockdep_set_class_and_name(&eb->lock,
191 &ks->keys[level], ks->names[level]);
197 * extents on the btree inode are pretty simple, there's one extent
198 * that covers the entire device
200 static struct extent_map *btree_get_extent(struct inode *inode,
201 struct page *page, size_t pg_offset, u64 start, u64 len,
204 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
205 struct extent_map *em;
208 read_lock(&em_tree->lock);
209 em = lookup_extent_mapping(em_tree, start, len);
212 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
213 read_unlock(&em_tree->lock);
216 read_unlock(&em_tree->lock);
218 em = alloc_extent_map();
220 em = ERR_PTR(-ENOMEM);
225 em->block_len = (u64)-1;
227 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
229 write_lock(&em_tree->lock);
230 ret = add_extent_mapping(em_tree, em, 0);
231 if (ret == -EEXIST) {
233 em = lookup_extent_mapping(em_tree, start, len);
240 write_unlock(&em_tree->lock);
246 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
248 return crc32c(seed, data, len);
251 void btrfs_csum_final(u32 crc, char *result)
253 put_unaligned_le32(~crc, result);
257 * compute the csum for a btree block, and either verify it or write it
258 * into the csum field of the block.
260 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
263 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
266 unsigned long cur_len;
267 unsigned long offset = BTRFS_CSUM_SIZE;
269 unsigned long map_start;
270 unsigned long map_len;
273 unsigned long inline_result;
275 len = buf->len - offset;
277 err = map_private_extent_buffer(buf, offset, 32,
278 &kaddr, &map_start, &map_len);
281 cur_len = min(len, map_len - (offset - map_start));
282 crc = btrfs_csum_data(kaddr + offset - map_start,
287 if (csum_size > sizeof(inline_result)) {
288 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
292 result = (char *)&inline_result;
295 btrfs_csum_final(crc, result);
298 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
301 memcpy(&found, result, csum_size);
303 read_extent_buffer(buf, &val, 0, csum_size);
304 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
305 "failed on %llu wanted %X found %X "
307 root->fs_info->sb->s_id, buf->start,
308 val, found, btrfs_header_level(buf));
309 if (result != (char *)&inline_result)
314 write_extent_buffer(buf, result, 0, csum_size);
316 if (result != (char *)&inline_result)
322 * we can't consider a given block up to date unless the transid of the
323 * block matches the transid in the parent node's pointer. This is how we
324 * detect blocks that either didn't get written at all or got written
325 * in the wrong place.
327 static int verify_parent_transid(struct extent_io_tree *io_tree,
328 struct extent_buffer *eb, u64 parent_transid,
331 struct extent_state *cached_state = NULL;
334 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
340 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
342 if (extent_buffer_uptodate(eb) &&
343 btrfs_header_generation(eb) == parent_transid) {
347 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
349 eb->start, parent_transid, btrfs_header_generation(eb));
351 clear_extent_buffer_uptodate(eb);
353 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
354 &cached_state, GFP_NOFS);
359 * Return 0 if the superblock checksum type matches the checksum value of that
360 * algorithm. Pass the raw disk superblock data.
362 static int btrfs_check_super_csum(char *raw_disk_sb)
364 struct btrfs_super_block *disk_sb =
365 (struct btrfs_super_block *)raw_disk_sb;
366 u16 csum_type = btrfs_super_csum_type(disk_sb);
369 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
371 const int csum_size = sizeof(crc);
372 char result[csum_size];
375 * The super_block structure does not span the whole
376 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
377 * is filled with zeros and is included in the checkum.
379 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
380 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
381 btrfs_csum_final(crc, result);
383 if (memcmp(raw_disk_sb, result, csum_size))
386 if (ret && btrfs_super_generation(disk_sb) < 10) {
387 printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
392 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
393 printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
402 * helper to read a given tree block, doing retries as required when
403 * the checksums don't match and we have alternate mirrors to try.
405 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
406 struct extent_buffer *eb,
407 u64 start, u64 parent_transid)
409 struct extent_io_tree *io_tree;
414 int failed_mirror = 0;
416 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
417 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
419 ret = read_extent_buffer_pages(io_tree, eb, start,
421 btree_get_extent, mirror_num);
423 if (!verify_parent_transid(io_tree, eb,
431 * This buffer's crc is fine, but its contents are corrupted, so
432 * there is no reason to read the other copies, they won't be
435 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
438 num_copies = btrfs_num_copies(root->fs_info,
443 if (!failed_mirror) {
445 failed_mirror = eb->read_mirror;
449 if (mirror_num == failed_mirror)
452 if (mirror_num > num_copies)
456 if (failed && !ret && failed_mirror)
457 repair_eb_io_failure(root, eb, failed_mirror);
463 * checksum a dirty tree block before IO. This has extra checks to make sure
464 * we only fill in the checksum field in the first page of a multi-page block
467 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
469 struct extent_io_tree *tree;
470 u64 start = page_offset(page);
472 struct extent_buffer *eb;
474 tree = &BTRFS_I(page->mapping->host)->io_tree;
476 eb = (struct extent_buffer *)page->private;
477 if (page != eb->pages[0])
479 found_start = btrfs_header_bytenr(eb);
480 if (found_start != start) {
484 if (!PageUptodate(page)) {
488 csum_tree_block(root, eb, 0);
492 static int check_tree_block_fsid(struct btrfs_root *root,
493 struct extent_buffer *eb)
495 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
496 u8 fsid[BTRFS_UUID_SIZE];
499 read_extent_buffer(eb, fsid, btrfs_header_fsid(eb), BTRFS_FSID_SIZE);
501 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
505 fs_devices = fs_devices->seed;
510 #define CORRUPT(reason, eb, root, slot) \
511 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
512 "root=%llu, slot=%d\n", reason, \
513 btrfs_header_bytenr(eb), root->objectid, slot)
515 static noinline int check_leaf(struct btrfs_root *root,
516 struct extent_buffer *leaf)
518 struct btrfs_key key;
519 struct btrfs_key leaf_key;
520 u32 nritems = btrfs_header_nritems(leaf);
526 /* Check the 0 item */
527 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
528 BTRFS_LEAF_DATA_SIZE(root)) {
529 CORRUPT("invalid item offset size pair", leaf, root, 0);
534 * Check to make sure each items keys are in the correct order and their
535 * offsets make sense. We only have to loop through nritems-1 because
536 * we check the current slot against the next slot, which verifies the
537 * next slot's offset+size makes sense and that the current's slot
540 for (slot = 0; slot < nritems - 1; slot++) {
541 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
542 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
544 /* Make sure the keys are in the right order */
545 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
546 CORRUPT("bad key order", leaf, root, slot);
551 * Make sure the offset and ends are right, remember that the
552 * item data starts at the end of the leaf and grows towards the
555 if (btrfs_item_offset_nr(leaf, slot) !=
556 btrfs_item_end_nr(leaf, slot + 1)) {
557 CORRUPT("slot offset bad", leaf, root, slot);
562 * Check to make sure that we don't point outside of the leaf,
563 * just incase all the items are consistent to eachother, but
564 * all point outside of the leaf.
566 if (btrfs_item_end_nr(leaf, slot) >
567 BTRFS_LEAF_DATA_SIZE(root)) {
568 CORRUPT("slot end outside of leaf", leaf, root, slot);
576 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
577 u64 phy_offset, struct page *page,
578 u64 start, u64 end, int mirror)
580 struct extent_io_tree *tree;
583 struct extent_buffer *eb;
584 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
591 tree = &BTRFS_I(page->mapping->host)->io_tree;
592 eb = (struct extent_buffer *)page->private;
594 /* the pending IO might have been the only thing that kept this buffer
595 * in memory. Make sure we have a ref for all this other checks
597 extent_buffer_get(eb);
599 reads_done = atomic_dec_and_test(&eb->io_pages);
603 eb->read_mirror = mirror;
604 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
609 found_start = btrfs_header_bytenr(eb);
610 if (found_start != eb->start) {
611 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
613 found_start, eb->start);
617 if (check_tree_block_fsid(root, eb)) {
618 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
623 found_level = btrfs_header_level(eb);
624 if (found_level >= BTRFS_MAX_LEVEL) {
625 btrfs_info(root->fs_info, "bad tree block level %d\n",
626 (int)btrfs_header_level(eb));
631 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
634 ret = csum_tree_block(root, eb, 1);
641 * If this is a leaf block and it is corrupt, set the corrupt bit so
642 * that we don't try and read the other copies of this block, just
645 if (found_level == 0 && check_leaf(root, eb)) {
646 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
651 set_extent_buffer_uptodate(eb);
654 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
655 btree_readahead_hook(root, eb, eb->start, ret);
659 * our io error hook is going to dec the io pages
660 * again, we have to make sure it has something
663 atomic_inc(&eb->io_pages);
664 clear_extent_buffer_uptodate(eb);
666 free_extent_buffer(eb);
671 static int btree_io_failed_hook(struct page *page, int failed_mirror)
673 struct extent_buffer *eb;
674 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
676 eb = (struct extent_buffer *)page->private;
677 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
678 eb->read_mirror = failed_mirror;
679 atomic_dec(&eb->io_pages);
680 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
681 btree_readahead_hook(root, eb, eb->start, -EIO);
682 return -EIO; /* we fixed nothing */
685 static void end_workqueue_bio(struct bio *bio, int err)
687 struct end_io_wq *end_io_wq = bio->bi_private;
688 struct btrfs_fs_info *fs_info;
690 fs_info = end_io_wq->info;
691 end_io_wq->error = err;
692 end_io_wq->work.func = end_workqueue_fn;
693 end_io_wq->work.flags = 0;
695 if (bio->bi_rw & REQ_WRITE) {
696 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
697 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
699 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
700 btrfs_queue_worker(&fs_info->endio_freespace_worker,
702 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
703 btrfs_queue_worker(&fs_info->endio_raid56_workers,
706 btrfs_queue_worker(&fs_info->endio_write_workers,
709 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
710 btrfs_queue_worker(&fs_info->endio_raid56_workers,
712 else if (end_io_wq->metadata)
713 btrfs_queue_worker(&fs_info->endio_meta_workers,
716 btrfs_queue_worker(&fs_info->endio_workers,
722 * For the metadata arg you want
725 * 1 - if normal metadta
726 * 2 - if writing to the free space cache area
727 * 3 - raid parity work
729 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
732 struct end_io_wq *end_io_wq;
733 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
737 end_io_wq->private = bio->bi_private;
738 end_io_wq->end_io = bio->bi_end_io;
739 end_io_wq->info = info;
740 end_io_wq->error = 0;
741 end_io_wq->bio = bio;
742 end_io_wq->metadata = metadata;
744 bio->bi_private = end_io_wq;
745 bio->bi_end_io = end_workqueue_bio;
749 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
751 unsigned long limit = min_t(unsigned long,
752 info->workers.max_workers,
753 info->fs_devices->open_devices);
757 static void run_one_async_start(struct btrfs_work *work)
759 struct async_submit_bio *async;
762 async = container_of(work, struct async_submit_bio, work);
763 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
764 async->mirror_num, async->bio_flags,
770 static void run_one_async_done(struct btrfs_work *work)
772 struct btrfs_fs_info *fs_info;
773 struct async_submit_bio *async;
776 async = container_of(work, struct async_submit_bio, work);
777 fs_info = BTRFS_I(async->inode)->root->fs_info;
779 limit = btrfs_async_submit_limit(fs_info);
780 limit = limit * 2 / 3;
782 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
783 waitqueue_active(&fs_info->async_submit_wait))
784 wake_up(&fs_info->async_submit_wait);
786 /* If an error occured we just want to clean up the bio and move on */
788 bio_endio(async->bio, async->error);
792 async->submit_bio_done(async->inode, async->rw, async->bio,
793 async->mirror_num, async->bio_flags,
797 static void run_one_async_free(struct btrfs_work *work)
799 struct async_submit_bio *async;
801 async = container_of(work, struct async_submit_bio, work);
805 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
806 int rw, struct bio *bio, int mirror_num,
807 unsigned long bio_flags,
809 extent_submit_bio_hook_t *submit_bio_start,
810 extent_submit_bio_hook_t *submit_bio_done)
812 struct async_submit_bio *async;
814 async = kmalloc(sizeof(*async), GFP_NOFS);
818 async->inode = inode;
821 async->mirror_num = mirror_num;
822 async->submit_bio_start = submit_bio_start;
823 async->submit_bio_done = submit_bio_done;
825 async->work.func = run_one_async_start;
826 async->work.ordered_func = run_one_async_done;
827 async->work.ordered_free = run_one_async_free;
829 async->work.flags = 0;
830 async->bio_flags = bio_flags;
831 async->bio_offset = bio_offset;
835 atomic_inc(&fs_info->nr_async_submits);
838 btrfs_set_work_high_prio(&async->work);
840 btrfs_queue_worker(&fs_info->workers, &async->work);
842 while (atomic_read(&fs_info->async_submit_draining) &&
843 atomic_read(&fs_info->nr_async_submits)) {
844 wait_event(fs_info->async_submit_wait,
845 (atomic_read(&fs_info->nr_async_submits) == 0));
851 static int btree_csum_one_bio(struct bio *bio)
853 struct bio_vec *bvec = bio->bi_io_vec;
855 struct btrfs_root *root;
858 WARN_ON(bio->bi_vcnt <= 0);
859 while (bio_index < bio->bi_vcnt) {
860 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
861 ret = csum_dirty_buffer(root, bvec->bv_page);
870 static int __btree_submit_bio_start(struct inode *inode, int rw,
871 struct bio *bio, int mirror_num,
872 unsigned long bio_flags,
876 * when we're called for a write, we're already in the async
877 * submission context. Just jump into btrfs_map_bio
879 return btree_csum_one_bio(bio);
882 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
883 int mirror_num, unsigned long bio_flags,
889 * when we're called for a write, we're already in the async
890 * submission context. Just jump into btrfs_map_bio
892 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
898 static int check_async_write(struct inode *inode, unsigned long bio_flags)
900 if (bio_flags & EXTENT_BIO_TREE_LOG)
909 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
910 int mirror_num, unsigned long bio_flags,
913 int async = check_async_write(inode, bio_flags);
916 if (!(rw & REQ_WRITE)) {
918 * called for a read, do the setup so that checksum validation
919 * can happen in the async kernel threads
921 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
925 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
928 ret = btree_csum_one_bio(bio);
931 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
935 * kthread helpers are used to submit writes so that
936 * checksumming can happen in parallel across all CPUs
938 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
939 inode, rw, bio, mirror_num, 0,
941 __btree_submit_bio_start,
942 __btree_submit_bio_done);
952 #ifdef CONFIG_MIGRATION
953 static int btree_migratepage(struct address_space *mapping,
954 struct page *newpage, struct page *page,
955 enum migrate_mode mode)
958 * we can't safely write a btree page from here,
959 * we haven't done the locking hook
964 * Buffers may be managed in a filesystem specific way.
965 * We must have no buffers or drop them.
967 if (page_has_private(page) &&
968 !try_to_release_page(page, GFP_KERNEL))
970 return migrate_page(mapping, newpage, page, mode);
975 static int btree_writepages(struct address_space *mapping,
976 struct writeback_control *wbc)
978 struct extent_io_tree *tree;
979 struct btrfs_fs_info *fs_info;
982 tree = &BTRFS_I(mapping->host)->io_tree;
983 if (wbc->sync_mode == WB_SYNC_NONE) {
985 if (wbc->for_kupdate)
988 fs_info = BTRFS_I(mapping->host)->root->fs_info;
989 /* this is a bit racy, but that's ok */
990 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
991 BTRFS_DIRTY_METADATA_THRESH);
995 return btree_write_cache_pages(mapping, wbc);
998 static int btree_readpage(struct file *file, struct page *page)
1000 struct extent_io_tree *tree;
1001 tree = &BTRFS_I(page->mapping->host)->io_tree;
1002 return extent_read_full_page(tree, page, btree_get_extent, 0);
1005 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1007 if (PageWriteback(page) || PageDirty(page))
1010 return try_release_extent_buffer(page);
1013 static void btree_invalidatepage(struct page *page, unsigned int offset,
1014 unsigned int length)
1016 struct extent_io_tree *tree;
1017 tree = &BTRFS_I(page->mapping->host)->io_tree;
1018 extent_invalidatepage(tree, page, offset);
1019 btree_releasepage(page, GFP_NOFS);
1020 if (PagePrivate(page)) {
1021 printk(KERN_WARNING "btrfs warning page private not zero "
1022 "on page %llu\n", (unsigned long long)page_offset(page));
1023 ClearPagePrivate(page);
1024 set_page_private(page, 0);
1025 page_cache_release(page);
1029 static int btree_set_page_dirty(struct page *page)
1032 struct extent_buffer *eb;
1034 BUG_ON(!PagePrivate(page));
1035 eb = (struct extent_buffer *)page->private;
1037 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1038 BUG_ON(!atomic_read(&eb->refs));
1039 btrfs_assert_tree_locked(eb);
1041 return __set_page_dirty_nobuffers(page);
1044 static const struct address_space_operations btree_aops = {
1045 .readpage = btree_readpage,
1046 .writepages = btree_writepages,
1047 .releasepage = btree_releasepage,
1048 .invalidatepage = btree_invalidatepage,
1049 #ifdef CONFIG_MIGRATION
1050 .migratepage = btree_migratepage,
1052 .set_page_dirty = btree_set_page_dirty,
1055 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1058 struct extent_buffer *buf = NULL;
1059 struct inode *btree_inode = root->fs_info->btree_inode;
1062 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1065 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1066 buf, 0, WAIT_NONE, btree_get_extent, 0);
1067 free_extent_buffer(buf);
1071 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1072 int mirror_num, struct extent_buffer **eb)
1074 struct extent_buffer *buf = NULL;
1075 struct inode *btree_inode = root->fs_info->btree_inode;
1076 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1079 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1083 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1085 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1086 btree_get_extent, mirror_num);
1088 free_extent_buffer(buf);
1092 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1093 free_extent_buffer(buf);
1095 } else if (extent_buffer_uptodate(buf)) {
1098 free_extent_buffer(buf);
1103 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1104 u64 bytenr, u32 blocksize)
1106 struct inode *btree_inode = root->fs_info->btree_inode;
1107 struct extent_buffer *eb;
1108 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1113 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1114 u64 bytenr, u32 blocksize)
1116 struct inode *btree_inode = root->fs_info->btree_inode;
1117 struct extent_buffer *eb;
1119 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1125 int btrfs_write_tree_block(struct extent_buffer *buf)
1127 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1128 buf->start + buf->len - 1);
1131 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1133 return filemap_fdatawait_range(buf->pages[0]->mapping,
1134 buf->start, buf->start + buf->len - 1);
1137 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1138 u32 blocksize, u64 parent_transid)
1140 struct extent_buffer *buf = NULL;
1143 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1147 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1149 free_extent_buffer(buf);
1156 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1157 struct extent_buffer *buf)
1159 struct btrfs_fs_info *fs_info = root->fs_info;
1161 if (btrfs_header_generation(buf) ==
1162 fs_info->running_transaction->transid) {
1163 btrfs_assert_tree_locked(buf);
1165 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1166 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1168 fs_info->dirty_metadata_batch);
1169 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1170 btrfs_set_lock_blocking(buf);
1171 clear_extent_buffer_dirty(buf);
1176 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1177 u32 stripesize, struct btrfs_root *root,
1178 struct btrfs_fs_info *fs_info,
1182 root->commit_root = NULL;
1183 root->sectorsize = sectorsize;
1184 root->nodesize = nodesize;
1185 root->leafsize = leafsize;
1186 root->stripesize = stripesize;
1188 root->track_dirty = 0;
1190 root->orphan_item_inserted = 0;
1191 root->orphan_cleanup_state = 0;
1193 root->objectid = objectid;
1194 root->last_trans = 0;
1195 root->highest_objectid = 0;
1196 root->nr_delalloc_inodes = 0;
1197 root->nr_ordered_extents = 0;
1199 root->inode_tree = RB_ROOT;
1200 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1201 root->block_rsv = NULL;
1202 root->orphan_block_rsv = NULL;
1204 INIT_LIST_HEAD(&root->dirty_list);
1205 INIT_LIST_HEAD(&root->root_list);
1206 INIT_LIST_HEAD(&root->delalloc_inodes);
1207 INIT_LIST_HEAD(&root->delalloc_root);
1208 INIT_LIST_HEAD(&root->ordered_extents);
1209 INIT_LIST_HEAD(&root->ordered_root);
1210 INIT_LIST_HEAD(&root->logged_list[0]);
1211 INIT_LIST_HEAD(&root->logged_list[1]);
1212 spin_lock_init(&root->orphan_lock);
1213 spin_lock_init(&root->inode_lock);
1214 spin_lock_init(&root->delalloc_lock);
1215 spin_lock_init(&root->ordered_extent_lock);
1216 spin_lock_init(&root->accounting_lock);
1217 spin_lock_init(&root->log_extents_lock[0]);
1218 spin_lock_init(&root->log_extents_lock[1]);
1219 mutex_init(&root->objectid_mutex);
1220 mutex_init(&root->log_mutex);
1221 init_waitqueue_head(&root->log_writer_wait);
1222 init_waitqueue_head(&root->log_commit_wait[0]);
1223 init_waitqueue_head(&root->log_commit_wait[1]);
1224 atomic_set(&root->log_commit[0], 0);
1225 atomic_set(&root->log_commit[1], 0);
1226 atomic_set(&root->log_writers, 0);
1227 atomic_set(&root->log_batch, 0);
1228 atomic_set(&root->orphan_inodes, 0);
1229 atomic_set(&root->refs, 1);
1230 root->log_transid = 0;
1231 root->last_log_commit = 0;
1233 extent_io_tree_init(&root->dirty_log_pages,
1234 fs_info->btree_inode->i_mapping);
1236 memset(&root->root_key, 0, sizeof(root->root_key));
1237 memset(&root->root_item, 0, sizeof(root->root_item));
1238 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1239 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1241 root->defrag_trans_start = fs_info->generation;
1243 root->defrag_trans_start = 0;
1244 init_completion(&root->kobj_unregister);
1245 root->defrag_running = 0;
1246 root->root_key.objectid = objectid;
1249 spin_lock_init(&root->root_item_lock);
1252 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1254 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1256 root->fs_info = fs_info;
1260 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1261 /* Should only be used by the testing infrastructure */
1262 struct btrfs_root *btrfs_alloc_dummy_root(void)
1264 struct btrfs_root *root;
1266 root = btrfs_alloc_root(NULL);
1268 return ERR_PTR(-ENOMEM);
1269 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1270 root->dummy_root = 1;
1276 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1277 struct btrfs_fs_info *fs_info,
1280 struct extent_buffer *leaf;
1281 struct btrfs_root *tree_root = fs_info->tree_root;
1282 struct btrfs_root *root;
1283 struct btrfs_key key;
1288 root = btrfs_alloc_root(fs_info);
1290 return ERR_PTR(-ENOMEM);
1292 __setup_root(tree_root->nodesize, tree_root->leafsize,
1293 tree_root->sectorsize, tree_root->stripesize,
1294 root, fs_info, objectid);
1295 root->root_key.objectid = objectid;
1296 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1297 root->root_key.offset = 0;
1299 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1300 0, objectid, NULL, 0, 0, 0);
1302 ret = PTR_ERR(leaf);
1307 bytenr = leaf->start;
1308 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1309 btrfs_set_header_bytenr(leaf, leaf->start);
1310 btrfs_set_header_generation(leaf, trans->transid);
1311 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1312 btrfs_set_header_owner(leaf, objectid);
1315 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(leaf),
1317 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1318 btrfs_header_chunk_tree_uuid(leaf),
1320 btrfs_mark_buffer_dirty(leaf);
1322 root->commit_root = btrfs_root_node(root);
1323 root->track_dirty = 1;
1326 root->root_item.flags = 0;
1327 root->root_item.byte_limit = 0;
1328 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1329 btrfs_set_root_generation(&root->root_item, trans->transid);
1330 btrfs_set_root_level(&root->root_item, 0);
1331 btrfs_set_root_refs(&root->root_item, 1);
1332 btrfs_set_root_used(&root->root_item, leaf->len);
1333 btrfs_set_root_last_snapshot(&root->root_item, 0);
1334 btrfs_set_root_dirid(&root->root_item, 0);
1336 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1337 root->root_item.drop_level = 0;
1339 key.objectid = objectid;
1340 key.type = BTRFS_ROOT_ITEM_KEY;
1342 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1346 btrfs_tree_unlock(leaf);
1352 btrfs_tree_unlock(leaf);
1353 free_extent_buffer(leaf);
1357 return ERR_PTR(ret);
1360 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1361 struct btrfs_fs_info *fs_info)
1363 struct btrfs_root *root;
1364 struct btrfs_root *tree_root = fs_info->tree_root;
1365 struct extent_buffer *leaf;
1367 root = btrfs_alloc_root(fs_info);
1369 return ERR_PTR(-ENOMEM);
1371 __setup_root(tree_root->nodesize, tree_root->leafsize,
1372 tree_root->sectorsize, tree_root->stripesize,
1373 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1375 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1376 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1377 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1379 * log trees do not get reference counted because they go away
1380 * before a real commit is actually done. They do store pointers
1381 * to file data extents, and those reference counts still get
1382 * updated (along with back refs to the log tree).
1386 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1387 BTRFS_TREE_LOG_OBJECTID, NULL,
1391 return ERR_CAST(leaf);
1394 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1395 btrfs_set_header_bytenr(leaf, leaf->start);
1396 btrfs_set_header_generation(leaf, trans->transid);
1397 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1398 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1401 write_extent_buffer(root->node, root->fs_info->fsid,
1402 btrfs_header_fsid(root->node), BTRFS_FSID_SIZE);
1403 btrfs_mark_buffer_dirty(root->node);
1404 btrfs_tree_unlock(root->node);
1408 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1409 struct btrfs_fs_info *fs_info)
1411 struct btrfs_root *log_root;
1413 log_root = alloc_log_tree(trans, fs_info);
1414 if (IS_ERR(log_root))
1415 return PTR_ERR(log_root);
1416 WARN_ON(fs_info->log_root_tree);
1417 fs_info->log_root_tree = log_root;
1421 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1422 struct btrfs_root *root)
1424 struct btrfs_root *log_root;
1425 struct btrfs_inode_item *inode_item;
1427 log_root = alloc_log_tree(trans, root->fs_info);
1428 if (IS_ERR(log_root))
1429 return PTR_ERR(log_root);
1431 log_root->last_trans = trans->transid;
1432 log_root->root_key.offset = root->root_key.objectid;
1434 inode_item = &log_root->root_item.inode;
1435 btrfs_set_stack_inode_generation(inode_item, 1);
1436 btrfs_set_stack_inode_size(inode_item, 3);
1437 btrfs_set_stack_inode_nlink(inode_item, 1);
1438 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1439 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1441 btrfs_set_root_node(&log_root->root_item, log_root->node);
1443 WARN_ON(root->log_root);
1444 root->log_root = log_root;
1445 root->log_transid = 0;
1446 root->last_log_commit = 0;
1450 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1451 struct btrfs_key *key)
1453 struct btrfs_root *root;
1454 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1455 struct btrfs_path *path;
1460 path = btrfs_alloc_path();
1462 return ERR_PTR(-ENOMEM);
1464 root = btrfs_alloc_root(fs_info);
1470 __setup_root(tree_root->nodesize, tree_root->leafsize,
1471 tree_root->sectorsize, tree_root->stripesize,
1472 root, fs_info, key->objectid);
1474 ret = btrfs_find_root(tree_root, key, path,
1475 &root->root_item, &root->root_key);
1482 generation = btrfs_root_generation(&root->root_item);
1483 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1484 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1485 blocksize, generation);
1489 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1493 root->commit_root = btrfs_root_node(root);
1495 btrfs_free_path(path);
1499 free_extent_buffer(root->node);
1503 root = ERR_PTR(ret);
1507 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1508 struct btrfs_key *location)
1510 struct btrfs_root *root;
1512 root = btrfs_read_tree_root(tree_root, location);
1516 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1518 btrfs_check_and_init_root_item(&root->root_item);
1524 int btrfs_init_fs_root(struct btrfs_root *root)
1528 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1529 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1531 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1536 btrfs_init_free_ino_ctl(root);
1537 mutex_init(&root->fs_commit_mutex);
1538 spin_lock_init(&root->cache_lock);
1539 init_waitqueue_head(&root->cache_wait);
1541 ret = get_anon_bdev(&root->anon_dev);
1546 kfree(root->free_ino_ctl);
1547 kfree(root->free_ino_pinned);
1551 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1554 struct btrfs_root *root;
1556 spin_lock(&fs_info->fs_roots_radix_lock);
1557 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1558 (unsigned long)root_id);
1559 spin_unlock(&fs_info->fs_roots_radix_lock);
1563 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1564 struct btrfs_root *root)
1568 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1572 spin_lock(&fs_info->fs_roots_radix_lock);
1573 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1574 (unsigned long)root->root_key.objectid,
1578 spin_unlock(&fs_info->fs_roots_radix_lock);
1579 radix_tree_preload_end();
1584 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1585 struct btrfs_key *location,
1588 struct btrfs_root *root;
1591 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1592 return fs_info->tree_root;
1593 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1594 return fs_info->extent_root;
1595 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1596 return fs_info->chunk_root;
1597 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1598 return fs_info->dev_root;
1599 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1600 return fs_info->csum_root;
1601 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1602 return fs_info->quota_root ? fs_info->quota_root :
1604 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1605 return fs_info->uuid_root ? fs_info->uuid_root :
1608 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1610 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1611 return ERR_PTR(-ENOENT);
1615 root = btrfs_read_fs_root(fs_info->tree_root, location);
1619 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1624 ret = btrfs_init_fs_root(root);
1628 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1632 root->orphan_item_inserted = 1;
1634 ret = btrfs_insert_fs_root(fs_info, root);
1636 if (ret == -EEXIST) {
1645 return ERR_PTR(ret);
1648 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1650 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1652 struct btrfs_device *device;
1653 struct backing_dev_info *bdi;
1656 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1659 bdi = blk_get_backing_dev_info(device->bdev);
1660 if (bdi && bdi_congested(bdi, bdi_bits)) {
1670 * If this fails, caller must call bdi_destroy() to get rid of the
1673 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1677 bdi->capabilities = BDI_CAP_MAP_COPY;
1678 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1682 bdi->ra_pages = default_backing_dev_info.ra_pages;
1683 bdi->congested_fn = btrfs_congested_fn;
1684 bdi->congested_data = info;
1689 * called by the kthread helper functions to finally call the bio end_io
1690 * functions. This is where read checksum verification actually happens
1692 static void end_workqueue_fn(struct btrfs_work *work)
1695 struct end_io_wq *end_io_wq;
1696 struct btrfs_fs_info *fs_info;
1699 end_io_wq = container_of(work, struct end_io_wq, work);
1700 bio = end_io_wq->bio;
1701 fs_info = end_io_wq->info;
1703 error = end_io_wq->error;
1704 bio->bi_private = end_io_wq->private;
1705 bio->bi_end_io = end_io_wq->end_io;
1707 bio_endio(bio, error);
1710 static int cleaner_kthread(void *arg)
1712 struct btrfs_root *root = arg;
1718 /* Make the cleaner go to sleep early. */
1719 if (btrfs_need_cleaner_sleep(root))
1722 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1726 * Avoid the problem that we change the status of the fs
1727 * during the above check and trylock.
1729 if (btrfs_need_cleaner_sleep(root)) {
1730 mutex_unlock(&root->fs_info->cleaner_mutex);
1734 btrfs_run_delayed_iputs(root);
1735 again = btrfs_clean_one_deleted_snapshot(root);
1736 mutex_unlock(&root->fs_info->cleaner_mutex);
1739 * The defragger has dealt with the R/O remount and umount,
1740 * needn't do anything special here.
1742 btrfs_run_defrag_inodes(root->fs_info);
1744 if (!try_to_freeze() && !again) {
1745 set_current_state(TASK_INTERRUPTIBLE);
1746 if (!kthread_should_stop())
1748 __set_current_state(TASK_RUNNING);
1750 } while (!kthread_should_stop());
1754 static int transaction_kthread(void *arg)
1756 struct btrfs_root *root = arg;
1757 struct btrfs_trans_handle *trans;
1758 struct btrfs_transaction *cur;
1761 unsigned long delay;
1765 cannot_commit = false;
1766 delay = HZ * root->fs_info->commit_interval;
1767 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1769 spin_lock(&root->fs_info->trans_lock);
1770 cur = root->fs_info->running_transaction;
1772 spin_unlock(&root->fs_info->trans_lock);
1776 now = get_seconds();
1777 if (cur->state < TRANS_STATE_BLOCKED &&
1778 (now < cur->start_time ||
1779 now - cur->start_time < root->fs_info->commit_interval)) {
1780 spin_unlock(&root->fs_info->trans_lock);
1784 transid = cur->transid;
1785 spin_unlock(&root->fs_info->trans_lock);
1787 /* If the file system is aborted, this will always fail. */
1788 trans = btrfs_attach_transaction(root);
1789 if (IS_ERR(trans)) {
1790 if (PTR_ERR(trans) != -ENOENT)
1791 cannot_commit = true;
1794 if (transid == trans->transid) {
1795 btrfs_commit_transaction(trans, root);
1797 btrfs_end_transaction(trans, root);
1800 wake_up_process(root->fs_info->cleaner_kthread);
1801 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1803 if (!try_to_freeze()) {
1804 set_current_state(TASK_INTERRUPTIBLE);
1805 if (!kthread_should_stop() &&
1806 (!btrfs_transaction_blocked(root->fs_info) ||
1808 schedule_timeout(delay);
1809 __set_current_state(TASK_RUNNING);
1811 } while (!kthread_should_stop());
1816 * this will find the highest generation in the array of
1817 * root backups. The index of the highest array is returned,
1818 * or -1 if we can't find anything.
1820 * We check to make sure the array is valid by comparing the
1821 * generation of the latest root in the array with the generation
1822 * in the super block. If they don't match we pitch it.
1824 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1827 int newest_index = -1;
1828 struct btrfs_root_backup *root_backup;
1831 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1832 root_backup = info->super_copy->super_roots + i;
1833 cur = btrfs_backup_tree_root_gen(root_backup);
1834 if (cur == newest_gen)
1838 /* check to see if we actually wrapped around */
1839 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1840 root_backup = info->super_copy->super_roots;
1841 cur = btrfs_backup_tree_root_gen(root_backup);
1842 if (cur == newest_gen)
1845 return newest_index;
1850 * find the oldest backup so we know where to store new entries
1851 * in the backup array. This will set the backup_root_index
1852 * field in the fs_info struct
1854 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1857 int newest_index = -1;
1859 newest_index = find_newest_super_backup(info, newest_gen);
1860 /* if there was garbage in there, just move along */
1861 if (newest_index == -1) {
1862 info->backup_root_index = 0;
1864 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1869 * copy all the root pointers into the super backup array.
1870 * this will bump the backup pointer by one when it is
1873 static void backup_super_roots(struct btrfs_fs_info *info)
1876 struct btrfs_root_backup *root_backup;
1879 next_backup = info->backup_root_index;
1880 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1881 BTRFS_NUM_BACKUP_ROOTS;
1884 * just overwrite the last backup if we're at the same generation
1885 * this happens only at umount
1887 root_backup = info->super_for_commit->super_roots + last_backup;
1888 if (btrfs_backup_tree_root_gen(root_backup) ==
1889 btrfs_header_generation(info->tree_root->node))
1890 next_backup = last_backup;
1892 root_backup = info->super_for_commit->super_roots + next_backup;
1895 * make sure all of our padding and empty slots get zero filled
1896 * regardless of which ones we use today
1898 memset(root_backup, 0, sizeof(*root_backup));
1900 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1902 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1903 btrfs_set_backup_tree_root_gen(root_backup,
1904 btrfs_header_generation(info->tree_root->node));
1906 btrfs_set_backup_tree_root_level(root_backup,
1907 btrfs_header_level(info->tree_root->node));
1909 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1910 btrfs_set_backup_chunk_root_gen(root_backup,
1911 btrfs_header_generation(info->chunk_root->node));
1912 btrfs_set_backup_chunk_root_level(root_backup,
1913 btrfs_header_level(info->chunk_root->node));
1915 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1916 btrfs_set_backup_extent_root_gen(root_backup,
1917 btrfs_header_generation(info->extent_root->node));
1918 btrfs_set_backup_extent_root_level(root_backup,
1919 btrfs_header_level(info->extent_root->node));
1922 * we might commit during log recovery, which happens before we set
1923 * the fs_root. Make sure it is valid before we fill it in.
1925 if (info->fs_root && info->fs_root->node) {
1926 btrfs_set_backup_fs_root(root_backup,
1927 info->fs_root->node->start);
1928 btrfs_set_backup_fs_root_gen(root_backup,
1929 btrfs_header_generation(info->fs_root->node));
1930 btrfs_set_backup_fs_root_level(root_backup,
1931 btrfs_header_level(info->fs_root->node));
1934 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1935 btrfs_set_backup_dev_root_gen(root_backup,
1936 btrfs_header_generation(info->dev_root->node));
1937 btrfs_set_backup_dev_root_level(root_backup,
1938 btrfs_header_level(info->dev_root->node));
1940 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1941 btrfs_set_backup_csum_root_gen(root_backup,
1942 btrfs_header_generation(info->csum_root->node));
1943 btrfs_set_backup_csum_root_level(root_backup,
1944 btrfs_header_level(info->csum_root->node));
1946 btrfs_set_backup_total_bytes(root_backup,
1947 btrfs_super_total_bytes(info->super_copy));
1948 btrfs_set_backup_bytes_used(root_backup,
1949 btrfs_super_bytes_used(info->super_copy));
1950 btrfs_set_backup_num_devices(root_backup,
1951 btrfs_super_num_devices(info->super_copy));
1954 * if we don't copy this out to the super_copy, it won't get remembered
1955 * for the next commit
1957 memcpy(&info->super_copy->super_roots,
1958 &info->super_for_commit->super_roots,
1959 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1963 * this copies info out of the root backup array and back into
1964 * the in-memory super block. It is meant to help iterate through
1965 * the array, so you send it the number of backups you've already
1966 * tried and the last backup index you used.
1968 * this returns -1 when it has tried all the backups
1970 static noinline int next_root_backup(struct btrfs_fs_info *info,
1971 struct btrfs_super_block *super,
1972 int *num_backups_tried, int *backup_index)
1974 struct btrfs_root_backup *root_backup;
1975 int newest = *backup_index;
1977 if (*num_backups_tried == 0) {
1978 u64 gen = btrfs_super_generation(super);
1980 newest = find_newest_super_backup(info, gen);
1984 *backup_index = newest;
1985 *num_backups_tried = 1;
1986 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1987 /* we've tried all the backups, all done */
1990 /* jump to the next oldest backup */
1991 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1992 BTRFS_NUM_BACKUP_ROOTS;
1993 *backup_index = newest;
1994 *num_backups_tried += 1;
1996 root_backup = super->super_roots + newest;
1998 btrfs_set_super_generation(super,
1999 btrfs_backup_tree_root_gen(root_backup));
2000 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2001 btrfs_set_super_root_level(super,
2002 btrfs_backup_tree_root_level(root_backup));
2003 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2006 * fixme: the total bytes and num_devices need to match or we should
2009 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2010 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2014 /* helper to cleanup workers */
2015 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2017 btrfs_stop_workers(&fs_info->generic_worker);
2018 btrfs_stop_workers(&fs_info->fixup_workers);
2019 btrfs_stop_workers(&fs_info->delalloc_workers);
2020 btrfs_stop_workers(&fs_info->workers);
2021 btrfs_stop_workers(&fs_info->endio_workers);
2022 btrfs_stop_workers(&fs_info->endio_meta_workers);
2023 btrfs_stop_workers(&fs_info->endio_raid56_workers);
2024 btrfs_stop_workers(&fs_info->rmw_workers);
2025 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2026 btrfs_stop_workers(&fs_info->endio_write_workers);
2027 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2028 btrfs_stop_workers(&fs_info->submit_workers);
2029 btrfs_stop_workers(&fs_info->delayed_workers);
2030 btrfs_stop_workers(&fs_info->caching_workers);
2031 btrfs_stop_workers(&fs_info->readahead_workers);
2032 btrfs_stop_workers(&fs_info->flush_workers);
2033 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
2036 /* helper to cleanup tree roots */
2037 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2039 free_extent_buffer(info->tree_root->node);
2040 free_extent_buffer(info->tree_root->commit_root);
2041 info->tree_root->node = NULL;
2042 info->tree_root->commit_root = NULL;
2044 if (info->dev_root) {
2045 free_extent_buffer(info->dev_root->node);
2046 free_extent_buffer(info->dev_root->commit_root);
2047 info->dev_root->node = NULL;
2048 info->dev_root->commit_root = NULL;
2050 if (info->extent_root) {
2051 free_extent_buffer(info->extent_root->node);
2052 free_extent_buffer(info->extent_root->commit_root);
2053 info->extent_root->node = NULL;
2054 info->extent_root->commit_root = NULL;
2056 if (info->csum_root) {
2057 free_extent_buffer(info->csum_root->node);
2058 free_extent_buffer(info->csum_root->commit_root);
2059 info->csum_root->node = NULL;
2060 info->csum_root->commit_root = NULL;
2062 if (info->quota_root) {
2063 free_extent_buffer(info->quota_root->node);
2064 free_extent_buffer(info->quota_root->commit_root);
2065 info->quota_root->node = NULL;
2066 info->quota_root->commit_root = NULL;
2068 if (info->uuid_root) {
2069 free_extent_buffer(info->uuid_root->node);
2070 free_extent_buffer(info->uuid_root->commit_root);
2071 info->uuid_root->node = NULL;
2072 info->uuid_root->commit_root = NULL;
2075 free_extent_buffer(info->chunk_root->node);
2076 free_extent_buffer(info->chunk_root->commit_root);
2077 info->chunk_root->node = NULL;
2078 info->chunk_root->commit_root = NULL;
2082 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2085 struct btrfs_root *gang[8];
2088 while (!list_empty(&fs_info->dead_roots)) {
2089 gang[0] = list_entry(fs_info->dead_roots.next,
2090 struct btrfs_root, root_list);
2091 list_del(&gang[0]->root_list);
2093 if (gang[0]->in_radix) {
2094 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2096 free_extent_buffer(gang[0]->node);
2097 free_extent_buffer(gang[0]->commit_root);
2098 btrfs_put_fs_root(gang[0]);
2103 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2108 for (i = 0; i < ret; i++)
2109 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2113 int open_ctree(struct super_block *sb,
2114 struct btrfs_fs_devices *fs_devices,
2124 struct btrfs_key location;
2125 struct buffer_head *bh;
2126 struct btrfs_super_block *disk_super;
2127 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2128 struct btrfs_root *tree_root;
2129 struct btrfs_root *extent_root;
2130 struct btrfs_root *csum_root;
2131 struct btrfs_root *chunk_root;
2132 struct btrfs_root *dev_root;
2133 struct btrfs_root *quota_root;
2134 struct btrfs_root *uuid_root;
2135 struct btrfs_root *log_tree_root;
2138 int num_backups_tried = 0;
2139 int backup_index = 0;
2140 bool create_uuid_tree;
2141 bool check_uuid_tree;
2143 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2144 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2145 if (!tree_root || !chunk_root) {
2150 ret = init_srcu_struct(&fs_info->subvol_srcu);
2156 ret = setup_bdi(fs_info, &fs_info->bdi);
2162 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2167 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2168 (1 + ilog2(nr_cpu_ids));
2170 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2173 goto fail_dirty_metadata_bytes;
2176 fs_info->btree_inode = new_inode(sb);
2177 if (!fs_info->btree_inode) {
2179 goto fail_delalloc_bytes;
2182 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2184 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2185 INIT_LIST_HEAD(&fs_info->trans_list);
2186 INIT_LIST_HEAD(&fs_info->dead_roots);
2187 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2188 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2189 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2190 spin_lock_init(&fs_info->delalloc_root_lock);
2191 spin_lock_init(&fs_info->trans_lock);
2192 spin_lock_init(&fs_info->fs_roots_radix_lock);
2193 spin_lock_init(&fs_info->delayed_iput_lock);
2194 spin_lock_init(&fs_info->defrag_inodes_lock);
2195 spin_lock_init(&fs_info->free_chunk_lock);
2196 spin_lock_init(&fs_info->tree_mod_seq_lock);
2197 spin_lock_init(&fs_info->super_lock);
2198 rwlock_init(&fs_info->tree_mod_log_lock);
2199 mutex_init(&fs_info->reloc_mutex);
2200 seqlock_init(&fs_info->profiles_lock);
2202 init_completion(&fs_info->kobj_unregister);
2203 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2204 INIT_LIST_HEAD(&fs_info->space_info);
2205 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2206 btrfs_mapping_init(&fs_info->mapping_tree);
2207 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2208 BTRFS_BLOCK_RSV_GLOBAL);
2209 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2210 BTRFS_BLOCK_RSV_DELALLOC);
2211 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2212 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2213 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2214 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2215 BTRFS_BLOCK_RSV_DELOPS);
2216 atomic_set(&fs_info->nr_async_submits, 0);
2217 atomic_set(&fs_info->async_delalloc_pages, 0);
2218 atomic_set(&fs_info->async_submit_draining, 0);
2219 atomic_set(&fs_info->nr_async_bios, 0);
2220 atomic_set(&fs_info->defrag_running, 0);
2221 atomic64_set(&fs_info->tree_mod_seq, 0);
2223 fs_info->max_inline = 8192 * 1024;
2224 fs_info->metadata_ratio = 0;
2225 fs_info->defrag_inodes = RB_ROOT;
2226 fs_info->free_chunk_space = 0;
2227 fs_info->tree_mod_log = RB_ROOT;
2228 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2230 /* readahead state */
2231 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2232 spin_lock_init(&fs_info->reada_lock);
2234 fs_info->thread_pool_size = min_t(unsigned long,
2235 num_online_cpus() + 2, 8);
2237 INIT_LIST_HEAD(&fs_info->ordered_roots);
2238 spin_lock_init(&fs_info->ordered_root_lock);
2239 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2241 if (!fs_info->delayed_root) {
2245 btrfs_init_delayed_root(fs_info->delayed_root);
2247 mutex_init(&fs_info->scrub_lock);
2248 atomic_set(&fs_info->scrubs_running, 0);
2249 atomic_set(&fs_info->scrub_pause_req, 0);
2250 atomic_set(&fs_info->scrubs_paused, 0);
2251 atomic_set(&fs_info->scrub_cancel_req, 0);
2252 init_waitqueue_head(&fs_info->scrub_pause_wait);
2253 init_rwsem(&fs_info->scrub_super_lock);
2254 fs_info->scrub_workers_refcnt = 0;
2255 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2256 fs_info->check_integrity_print_mask = 0;
2259 spin_lock_init(&fs_info->balance_lock);
2260 mutex_init(&fs_info->balance_mutex);
2261 atomic_set(&fs_info->balance_running, 0);
2262 atomic_set(&fs_info->balance_pause_req, 0);
2263 atomic_set(&fs_info->balance_cancel_req, 0);
2264 fs_info->balance_ctl = NULL;
2265 init_waitqueue_head(&fs_info->balance_wait_q);
2267 sb->s_blocksize = 4096;
2268 sb->s_blocksize_bits = blksize_bits(4096);
2269 sb->s_bdi = &fs_info->bdi;
2271 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2272 set_nlink(fs_info->btree_inode, 1);
2274 * we set the i_size on the btree inode to the max possible int.
2275 * the real end of the address space is determined by all of
2276 * the devices in the system
2278 fs_info->btree_inode->i_size = OFFSET_MAX;
2279 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2280 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2282 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2283 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2284 fs_info->btree_inode->i_mapping);
2285 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2286 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2288 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2290 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2291 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2292 sizeof(struct btrfs_key));
2293 set_bit(BTRFS_INODE_DUMMY,
2294 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2295 insert_inode_hash(fs_info->btree_inode);
2297 spin_lock_init(&fs_info->block_group_cache_lock);
2298 fs_info->block_group_cache_tree = RB_ROOT;
2299 fs_info->first_logical_byte = (u64)-1;
2301 extent_io_tree_init(&fs_info->freed_extents[0],
2302 fs_info->btree_inode->i_mapping);
2303 extent_io_tree_init(&fs_info->freed_extents[1],
2304 fs_info->btree_inode->i_mapping);
2305 fs_info->pinned_extents = &fs_info->freed_extents[0];
2306 fs_info->do_barriers = 1;
2309 mutex_init(&fs_info->ordered_operations_mutex);
2310 mutex_init(&fs_info->ordered_extent_flush_mutex);
2311 mutex_init(&fs_info->tree_log_mutex);
2312 mutex_init(&fs_info->chunk_mutex);
2313 mutex_init(&fs_info->transaction_kthread_mutex);
2314 mutex_init(&fs_info->cleaner_mutex);
2315 mutex_init(&fs_info->volume_mutex);
2316 init_rwsem(&fs_info->extent_commit_sem);
2317 init_rwsem(&fs_info->cleanup_work_sem);
2318 init_rwsem(&fs_info->subvol_sem);
2319 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2320 fs_info->dev_replace.lock_owner = 0;
2321 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2322 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2323 mutex_init(&fs_info->dev_replace.lock_management_lock);
2324 mutex_init(&fs_info->dev_replace.lock);
2326 spin_lock_init(&fs_info->qgroup_lock);
2327 mutex_init(&fs_info->qgroup_ioctl_lock);
2328 fs_info->qgroup_tree = RB_ROOT;
2329 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2330 fs_info->qgroup_seq = 1;
2331 fs_info->quota_enabled = 0;
2332 fs_info->pending_quota_state = 0;
2333 fs_info->qgroup_ulist = NULL;
2334 mutex_init(&fs_info->qgroup_rescan_lock);
2336 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2337 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2339 init_waitqueue_head(&fs_info->transaction_throttle);
2340 init_waitqueue_head(&fs_info->transaction_wait);
2341 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2342 init_waitqueue_head(&fs_info->async_submit_wait);
2344 ret = btrfs_alloc_stripe_hash_table(fs_info);
2350 __setup_root(4096, 4096, 4096, 4096, tree_root,
2351 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2353 invalidate_bdev(fs_devices->latest_bdev);
2356 * Read super block and check the signature bytes only
2358 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2365 * We want to check superblock checksum, the type is stored inside.
2366 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2368 if (btrfs_check_super_csum(bh->b_data)) {
2369 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2375 * super_copy is zeroed at allocation time and we never touch the
2376 * following bytes up to INFO_SIZE, the checksum is calculated from
2377 * the whole block of INFO_SIZE
2379 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2380 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2381 sizeof(*fs_info->super_for_commit));
2384 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2386 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2388 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2393 disk_super = fs_info->super_copy;
2394 if (!btrfs_super_root(disk_super))
2397 /* check FS state, whether FS is broken. */
2398 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2399 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2402 * run through our array of backup supers and setup
2403 * our ring pointer to the oldest one
2405 generation = btrfs_super_generation(disk_super);
2406 find_oldest_super_backup(fs_info, generation);
2409 * In the long term, we'll store the compression type in the super
2410 * block, and it'll be used for per file compression control.
2412 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2414 ret = btrfs_parse_options(tree_root, options);
2420 features = btrfs_super_incompat_flags(disk_super) &
2421 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2423 printk(KERN_ERR "BTRFS: couldn't mount because of "
2424 "unsupported optional features (%Lx).\n",
2430 if (btrfs_super_leafsize(disk_super) !=
2431 btrfs_super_nodesize(disk_super)) {
2432 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2433 "blocksizes don't match. node %d leaf %d\n",
2434 btrfs_super_nodesize(disk_super),
2435 btrfs_super_leafsize(disk_super));
2439 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2440 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2441 "blocksize (%d) was too large\n",
2442 btrfs_super_leafsize(disk_super));
2447 features = btrfs_super_incompat_flags(disk_super);
2448 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2449 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2450 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2452 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2453 printk(KERN_ERR "btrfs: has skinny extents\n");
2456 * flag our filesystem as having big metadata blocks if
2457 * they are bigger than the page size
2459 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2460 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2461 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2462 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2465 nodesize = btrfs_super_nodesize(disk_super);
2466 leafsize = btrfs_super_leafsize(disk_super);
2467 sectorsize = btrfs_super_sectorsize(disk_super);
2468 stripesize = btrfs_super_stripesize(disk_super);
2469 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2470 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2473 * mixed block groups end up with duplicate but slightly offset
2474 * extent buffers for the same range. It leads to corruptions
2476 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2477 (sectorsize != leafsize)) {
2478 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2479 "are not allowed for mixed block groups on %s\n",
2485 * Needn't use the lock because there is no other task which will
2488 btrfs_set_super_incompat_flags(disk_super, features);
2490 features = btrfs_super_compat_ro_flags(disk_super) &
2491 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2492 if (!(sb->s_flags & MS_RDONLY) && features) {
2493 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2494 "unsupported option features (%Lx).\n",
2500 btrfs_init_workers(&fs_info->generic_worker,
2501 "genwork", 1, NULL);
2503 btrfs_init_workers(&fs_info->workers, "worker",
2504 fs_info->thread_pool_size,
2505 &fs_info->generic_worker);
2507 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2508 fs_info->thread_pool_size, NULL);
2510 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2511 fs_info->thread_pool_size, NULL);
2513 btrfs_init_workers(&fs_info->submit_workers, "submit",
2514 min_t(u64, fs_devices->num_devices,
2515 fs_info->thread_pool_size), NULL);
2517 btrfs_init_workers(&fs_info->caching_workers, "cache",
2518 fs_info->thread_pool_size, NULL);
2520 /* a higher idle thresh on the submit workers makes it much more
2521 * likely that bios will be send down in a sane order to the
2524 fs_info->submit_workers.idle_thresh = 64;
2526 fs_info->workers.idle_thresh = 16;
2527 fs_info->workers.ordered = 1;
2529 fs_info->delalloc_workers.idle_thresh = 2;
2530 fs_info->delalloc_workers.ordered = 1;
2532 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2533 &fs_info->generic_worker);
2534 btrfs_init_workers(&fs_info->endio_workers, "endio",
2535 fs_info->thread_pool_size,
2536 &fs_info->generic_worker);
2537 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2538 fs_info->thread_pool_size,
2539 &fs_info->generic_worker);
2540 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2541 "endio-meta-write", fs_info->thread_pool_size,
2542 &fs_info->generic_worker);
2543 btrfs_init_workers(&fs_info->endio_raid56_workers,
2544 "endio-raid56", fs_info->thread_pool_size,
2545 &fs_info->generic_worker);
2546 btrfs_init_workers(&fs_info->rmw_workers,
2547 "rmw", fs_info->thread_pool_size,
2548 &fs_info->generic_worker);
2549 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2550 fs_info->thread_pool_size,
2551 &fs_info->generic_worker);
2552 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2553 1, &fs_info->generic_worker);
2554 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2555 fs_info->thread_pool_size,
2556 &fs_info->generic_worker);
2557 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2558 fs_info->thread_pool_size,
2559 &fs_info->generic_worker);
2560 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2561 &fs_info->generic_worker);
2564 * endios are largely parallel and should have a very
2567 fs_info->endio_workers.idle_thresh = 4;
2568 fs_info->endio_meta_workers.idle_thresh = 4;
2569 fs_info->endio_raid56_workers.idle_thresh = 4;
2570 fs_info->rmw_workers.idle_thresh = 2;
2572 fs_info->endio_write_workers.idle_thresh = 2;
2573 fs_info->endio_meta_write_workers.idle_thresh = 2;
2574 fs_info->readahead_workers.idle_thresh = 2;
2577 * btrfs_start_workers can really only fail because of ENOMEM so just
2578 * return -ENOMEM if any of these fail.
2580 ret = btrfs_start_workers(&fs_info->workers);
2581 ret |= btrfs_start_workers(&fs_info->generic_worker);
2582 ret |= btrfs_start_workers(&fs_info->submit_workers);
2583 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2584 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2585 ret |= btrfs_start_workers(&fs_info->endio_workers);
2586 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2587 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2588 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2589 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2590 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2591 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2592 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2593 ret |= btrfs_start_workers(&fs_info->caching_workers);
2594 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2595 ret |= btrfs_start_workers(&fs_info->flush_workers);
2596 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2599 goto fail_sb_buffer;
2602 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2603 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2604 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2606 tree_root->nodesize = nodesize;
2607 tree_root->leafsize = leafsize;
2608 tree_root->sectorsize = sectorsize;
2609 tree_root->stripesize = stripesize;
2611 sb->s_blocksize = sectorsize;
2612 sb->s_blocksize_bits = blksize_bits(sectorsize);
2614 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2615 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2616 goto fail_sb_buffer;
2619 if (sectorsize != PAGE_SIZE) {
2620 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2621 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2622 goto fail_sb_buffer;
2625 mutex_lock(&fs_info->chunk_mutex);
2626 ret = btrfs_read_sys_array(tree_root);
2627 mutex_unlock(&fs_info->chunk_mutex);
2629 printk(KERN_WARNING "btrfs: failed to read the system "
2630 "array on %s\n", sb->s_id);
2631 goto fail_sb_buffer;
2634 blocksize = btrfs_level_size(tree_root,
2635 btrfs_super_chunk_root_level(disk_super));
2636 generation = btrfs_super_chunk_root_generation(disk_super);
2638 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2639 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2641 chunk_root->node = read_tree_block(chunk_root,
2642 btrfs_super_chunk_root(disk_super),
2643 blocksize, generation);
2644 if (!chunk_root->node ||
2645 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2646 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2648 goto fail_tree_roots;
2650 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2651 chunk_root->commit_root = btrfs_root_node(chunk_root);
2653 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2654 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2656 ret = btrfs_read_chunk_tree(chunk_root);
2658 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2660 goto fail_tree_roots;
2664 * keep the device that is marked to be the target device for the
2665 * dev_replace procedure
2667 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2669 if (!fs_devices->latest_bdev) {
2670 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2672 goto fail_tree_roots;
2676 blocksize = btrfs_level_size(tree_root,
2677 btrfs_super_root_level(disk_super));
2678 generation = btrfs_super_generation(disk_super);
2680 tree_root->node = read_tree_block(tree_root,
2681 btrfs_super_root(disk_super),
2682 blocksize, generation);
2683 if (!tree_root->node ||
2684 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2685 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2688 goto recovery_tree_root;
2691 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2692 tree_root->commit_root = btrfs_root_node(tree_root);
2693 btrfs_set_root_refs(&tree_root->root_item, 1);
2695 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2696 location.type = BTRFS_ROOT_ITEM_KEY;
2697 location.offset = 0;
2699 extent_root = btrfs_read_tree_root(tree_root, &location);
2700 if (IS_ERR(extent_root)) {
2701 ret = PTR_ERR(extent_root);
2702 goto recovery_tree_root;
2704 extent_root->track_dirty = 1;
2705 fs_info->extent_root = extent_root;
2707 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2708 dev_root = btrfs_read_tree_root(tree_root, &location);
2709 if (IS_ERR(dev_root)) {
2710 ret = PTR_ERR(dev_root);
2711 goto recovery_tree_root;
2713 dev_root->track_dirty = 1;
2714 fs_info->dev_root = dev_root;
2715 btrfs_init_devices_late(fs_info);
2717 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2718 csum_root = btrfs_read_tree_root(tree_root, &location);
2719 if (IS_ERR(csum_root)) {
2720 ret = PTR_ERR(csum_root);
2721 goto recovery_tree_root;
2723 csum_root->track_dirty = 1;
2724 fs_info->csum_root = csum_root;
2726 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2727 quota_root = btrfs_read_tree_root(tree_root, &location);
2728 if (!IS_ERR(quota_root)) {
2729 quota_root->track_dirty = 1;
2730 fs_info->quota_enabled = 1;
2731 fs_info->pending_quota_state = 1;
2732 fs_info->quota_root = quota_root;
2735 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2736 uuid_root = btrfs_read_tree_root(tree_root, &location);
2737 if (IS_ERR(uuid_root)) {
2738 ret = PTR_ERR(uuid_root);
2740 goto recovery_tree_root;
2741 create_uuid_tree = true;
2742 check_uuid_tree = false;
2744 uuid_root->track_dirty = 1;
2745 fs_info->uuid_root = uuid_root;
2746 create_uuid_tree = false;
2748 generation != btrfs_super_uuid_tree_generation(disk_super);
2751 fs_info->generation = generation;
2752 fs_info->last_trans_committed = generation;
2754 ret = btrfs_recover_balance(fs_info);
2756 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2757 goto fail_block_groups;
2760 ret = btrfs_init_dev_stats(fs_info);
2762 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2764 goto fail_block_groups;
2767 ret = btrfs_init_dev_replace(fs_info);
2769 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2770 goto fail_block_groups;
2773 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2775 ret = btrfs_init_space_info(fs_info);
2777 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2778 goto fail_block_groups;
2781 ret = btrfs_read_block_groups(extent_root);
2783 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2784 goto fail_block_groups;
2786 fs_info->num_tolerated_disk_barrier_failures =
2787 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2788 if (fs_info->fs_devices->missing_devices >
2789 fs_info->num_tolerated_disk_barrier_failures &&
2790 !(sb->s_flags & MS_RDONLY)) {
2792 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2793 goto fail_block_groups;
2796 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2798 if (IS_ERR(fs_info->cleaner_kthread))
2799 goto fail_block_groups;
2801 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2803 "btrfs-transaction");
2804 if (IS_ERR(fs_info->transaction_kthread))
2807 if (!btrfs_test_opt(tree_root, SSD) &&
2808 !btrfs_test_opt(tree_root, NOSSD) &&
2809 !fs_info->fs_devices->rotating) {
2810 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2812 btrfs_set_opt(fs_info->mount_opt, SSD);
2815 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2816 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2817 ret = btrfsic_mount(tree_root, fs_devices,
2818 btrfs_test_opt(tree_root,
2819 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2821 fs_info->check_integrity_print_mask);
2823 printk(KERN_WARNING "btrfs: failed to initialize"
2824 " integrity check module %s\n", sb->s_id);
2827 ret = btrfs_read_qgroup_config(fs_info);
2829 goto fail_trans_kthread;
2831 /* do not make disk changes in broken FS */
2832 if (btrfs_super_log_root(disk_super) != 0) {
2833 u64 bytenr = btrfs_super_log_root(disk_super);
2835 if (fs_devices->rw_devices == 0) {
2836 printk(KERN_WARNING "Btrfs log replay required "
2842 btrfs_level_size(tree_root,
2843 btrfs_super_log_root_level(disk_super));
2845 log_tree_root = btrfs_alloc_root(fs_info);
2846 if (!log_tree_root) {
2851 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2852 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2854 log_tree_root->node = read_tree_block(tree_root, bytenr,
2857 if (!log_tree_root->node ||
2858 !extent_buffer_uptodate(log_tree_root->node)) {
2859 printk(KERN_ERR "btrfs: failed to read log tree\n");
2860 free_extent_buffer(log_tree_root->node);
2861 kfree(log_tree_root);
2862 goto fail_trans_kthread;
2864 /* returns with log_tree_root freed on success */
2865 ret = btrfs_recover_log_trees(log_tree_root);
2867 btrfs_error(tree_root->fs_info, ret,
2868 "Failed to recover log tree");
2869 free_extent_buffer(log_tree_root->node);
2870 kfree(log_tree_root);
2871 goto fail_trans_kthread;
2874 if (sb->s_flags & MS_RDONLY) {
2875 ret = btrfs_commit_super(tree_root);
2877 goto fail_trans_kthread;
2881 ret = btrfs_find_orphan_roots(tree_root);
2883 goto fail_trans_kthread;
2885 if (!(sb->s_flags & MS_RDONLY)) {
2886 ret = btrfs_cleanup_fs_roots(fs_info);
2888 goto fail_trans_kthread;
2890 ret = btrfs_recover_relocation(tree_root);
2893 "btrfs: failed to recover relocation\n");
2899 location.objectid = BTRFS_FS_TREE_OBJECTID;
2900 location.type = BTRFS_ROOT_ITEM_KEY;
2901 location.offset = 0;
2903 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2904 if (IS_ERR(fs_info->fs_root)) {
2905 err = PTR_ERR(fs_info->fs_root);
2909 if (sb->s_flags & MS_RDONLY)
2912 down_read(&fs_info->cleanup_work_sem);
2913 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2914 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2915 up_read(&fs_info->cleanup_work_sem);
2916 close_ctree(tree_root);
2919 up_read(&fs_info->cleanup_work_sem);
2921 ret = btrfs_resume_balance_async(fs_info);
2923 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2924 close_ctree(tree_root);
2928 ret = btrfs_resume_dev_replace_async(fs_info);
2930 pr_warn("btrfs: failed to resume dev_replace\n");
2931 close_ctree(tree_root);
2935 btrfs_qgroup_rescan_resume(fs_info);
2937 if (create_uuid_tree) {
2938 pr_info("btrfs: creating UUID tree\n");
2939 ret = btrfs_create_uuid_tree(fs_info);
2941 pr_warn("btrfs: failed to create the UUID tree %d\n",
2943 close_ctree(tree_root);
2946 } else if (check_uuid_tree ||
2947 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2948 pr_info("btrfs: checking UUID tree\n");
2949 ret = btrfs_check_uuid_tree(fs_info);
2951 pr_warn("btrfs: failed to check the UUID tree %d\n",
2953 close_ctree(tree_root);
2957 fs_info->update_uuid_tree_gen = 1;
2963 btrfs_free_qgroup_config(fs_info);
2965 kthread_stop(fs_info->transaction_kthread);
2966 btrfs_cleanup_transaction(fs_info->tree_root);
2967 del_fs_roots(fs_info);
2969 kthread_stop(fs_info->cleaner_kthread);
2972 * make sure we're done with the btree inode before we stop our
2975 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2978 btrfs_put_block_group_cache(fs_info);
2979 btrfs_free_block_groups(fs_info);
2982 free_root_pointers(fs_info, 1);
2983 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2986 btrfs_stop_all_workers(fs_info);
2989 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2991 iput(fs_info->btree_inode);
2992 fail_delalloc_bytes:
2993 percpu_counter_destroy(&fs_info->delalloc_bytes);
2994 fail_dirty_metadata_bytes:
2995 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2997 bdi_destroy(&fs_info->bdi);
2999 cleanup_srcu_struct(&fs_info->subvol_srcu);
3001 btrfs_free_stripe_hash_table(fs_info);
3002 btrfs_close_devices(fs_info->fs_devices);
3006 if (!btrfs_test_opt(tree_root, RECOVERY))
3007 goto fail_tree_roots;
3009 free_root_pointers(fs_info, 0);
3011 /* don't use the log in recovery mode, it won't be valid */
3012 btrfs_set_super_log_root(disk_super, 0);
3014 /* we can't trust the free space cache either */
3015 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3017 ret = next_root_backup(fs_info, fs_info->super_copy,
3018 &num_backups_tried, &backup_index);
3020 goto fail_block_groups;
3021 goto retry_root_backup;
3024 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3027 set_buffer_uptodate(bh);
3029 struct btrfs_device *device = (struct btrfs_device *)
3032 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
3033 "I/O error on %s\n",
3034 rcu_str_deref(device->name));
3035 /* note, we dont' set_buffer_write_io_error because we have
3036 * our own ways of dealing with the IO errors
3038 clear_buffer_uptodate(bh);
3039 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3045 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3047 struct buffer_head *bh;
3048 struct buffer_head *latest = NULL;
3049 struct btrfs_super_block *super;
3054 /* we would like to check all the supers, but that would make
3055 * a btrfs mount succeed after a mkfs from a different FS.
3056 * So, we need to add a special mount option to scan for
3057 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3059 for (i = 0; i < 1; i++) {
3060 bytenr = btrfs_sb_offset(i);
3061 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3062 i_size_read(bdev->bd_inode))
3064 bh = __bread(bdev, bytenr / 4096,
3065 BTRFS_SUPER_INFO_SIZE);
3069 super = (struct btrfs_super_block *)bh->b_data;
3070 if (btrfs_super_bytenr(super) != bytenr ||
3071 btrfs_super_magic(super) != BTRFS_MAGIC) {
3076 if (!latest || btrfs_super_generation(super) > transid) {
3079 transid = btrfs_super_generation(super);
3088 * this should be called twice, once with wait == 0 and
3089 * once with wait == 1. When wait == 0 is done, all the buffer heads
3090 * we write are pinned.
3092 * They are released when wait == 1 is done.
3093 * max_mirrors must be the same for both runs, and it indicates how
3094 * many supers on this one device should be written.
3096 * max_mirrors == 0 means to write them all.
3098 static int write_dev_supers(struct btrfs_device *device,
3099 struct btrfs_super_block *sb,
3100 int do_barriers, int wait, int max_mirrors)
3102 struct buffer_head *bh;
3109 if (max_mirrors == 0)
3110 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3112 for (i = 0; i < max_mirrors; i++) {
3113 bytenr = btrfs_sb_offset(i);
3114 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3118 bh = __find_get_block(device->bdev, bytenr / 4096,
3119 BTRFS_SUPER_INFO_SIZE);
3125 if (!buffer_uptodate(bh))
3128 /* drop our reference */
3131 /* drop the reference from the wait == 0 run */
3135 btrfs_set_super_bytenr(sb, bytenr);
3138 crc = btrfs_csum_data((char *)sb +
3139 BTRFS_CSUM_SIZE, crc,
3140 BTRFS_SUPER_INFO_SIZE -
3142 btrfs_csum_final(crc, sb->csum);
3145 * one reference for us, and we leave it for the
3148 bh = __getblk(device->bdev, bytenr / 4096,
3149 BTRFS_SUPER_INFO_SIZE);
3151 printk(KERN_ERR "btrfs: couldn't get super "
3152 "buffer head for bytenr %Lu\n", bytenr);
3157 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3159 /* one reference for submit_bh */
3162 set_buffer_uptodate(bh);
3164 bh->b_end_io = btrfs_end_buffer_write_sync;
3165 bh->b_private = device;
3169 * we fua the first super. The others we allow
3172 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3176 return errors < i ? 0 : -1;
3180 * endio for the write_dev_flush, this will wake anyone waiting
3181 * for the barrier when it is done
3183 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3186 if (err == -EOPNOTSUPP)
3187 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3188 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3190 if (bio->bi_private)
3191 complete(bio->bi_private);
3196 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3197 * sent down. With wait == 1, it waits for the previous flush.
3199 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3202 static int write_dev_flush(struct btrfs_device *device, int wait)
3207 if (device->nobarriers)
3211 bio = device->flush_bio;
3215 wait_for_completion(&device->flush_wait);
3217 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3218 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3219 rcu_str_deref(device->name));
3220 device->nobarriers = 1;
3221 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3223 btrfs_dev_stat_inc_and_print(device,
3224 BTRFS_DEV_STAT_FLUSH_ERRS);
3227 /* drop the reference from the wait == 0 run */
3229 device->flush_bio = NULL;
3235 * one reference for us, and we leave it for the
3238 device->flush_bio = NULL;
3239 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3243 bio->bi_end_io = btrfs_end_empty_barrier;
3244 bio->bi_bdev = device->bdev;
3245 init_completion(&device->flush_wait);
3246 bio->bi_private = &device->flush_wait;
3247 device->flush_bio = bio;
3250 btrfsic_submit_bio(WRITE_FLUSH, bio);
3256 * send an empty flush down to each device in parallel,
3257 * then wait for them
3259 static int barrier_all_devices(struct btrfs_fs_info *info)
3261 struct list_head *head;
3262 struct btrfs_device *dev;
3263 int errors_send = 0;
3264 int errors_wait = 0;
3267 /* send down all the barriers */
3268 head = &info->fs_devices->devices;
3269 list_for_each_entry_rcu(dev, head, dev_list) {
3274 if (!dev->in_fs_metadata || !dev->writeable)
3277 ret = write_dev_flush(dev, 0);
3282 /* wait for all the barriers */
3283 list_for_each_entry_rcu(dev, head, dev_list) {
3288 if (!dev->in_fs_metadata || !dev->writeable)
3291 ret = write_dev_flush(dev, 1);
3295 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3296 errors_wait > info->num_tolerated_disk_barrier_failures)
3301 int btrfs_calc_num_tolerated_disk_barrier_failures(
3302 struct btrfs_fs_info *fs_info)
3304 struct btrfs_ioctl_space_info space;
3305 struct btrfs_space_info *sinfo;
3306 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3307 BTRFS_BLOCK_GROUP_SYSTEM,
3308 BTRFS_BLOCK_GROUP_METADATA,
3309 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3313 int num_tolerated_disk_barrier_failures =
3314 (int)fs_info->fs_devices->num_devices;
3316 for (i = 0; i < num_types; i++) {
3317 struct btrfs_space_info *tmp;
3321 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3322 if (tmp->flags == types[i]) {
3332 down_read(&sinfo->groups_sem);
3333 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3334 if (!list_empty(&sinfo->block_groups[c])) {
3337 btrfs_get_block_group_info(
3338 &sinfo->block_groups[c], &space);
3339 if (space.total_bytes == 0 ||
3340 space.used_bytes == 0)
3342 flags = space.flags;
3345 * 0: if dup, single or RAID0 is configured for
3346 * any of metadata, system or data, else
3347 * 1: if RAID5 is configured, or if RAID1 or
3348 * RAID10 is configured and only two mirrors
3350 * 2: if RAID6 is configured, else
3351 * num_mirrors - 1: if RAID1 or RAID10 is
3352 * configured and more than
3353 * 2 mirrors are used.
3355 if (num_tolerated_disk_barrier_failures > 0 &&
3356 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3357 BTRFS_BLOCK_GROUP_RAID0)) ||
3358 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3360 num_tolerated_disk_barrier_failures = 0;
3361 else if (num_tolerated_disk_barrier_failures > 1) {
3362 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3363 BTRFS_BLOCK_GROUP_RAID5 |
3364 BTRFS_BLOCK_GROUP_RAID10)) {
3365 num_tolerated_disk_barrier_failures = 1;
3367 BTRFS_BLOCK_GROUP_RAID6) {
3368 num_tolerated_disk_barrier_failures = 2;
3373 up_read(&sinfo->groups_sem);
3376 return num_tolerated_disk_barrier_failures;
3379 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3381 struct list_head *head;
3382 struct btrfs_device *dev;
3383 struct btrfs_super_block *sb;
3384 struct btrfs_dev_item *dev_item;
3388 int total_errors = 0;
3391 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3392 backup_super_roots(root->fs_info);
3394 sb = root->fs_info->super_for_commit;
3395 dev_item = &sb->dev_item;
3397 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3398 head = &root->fs_info->fs_devices->devices;
3399 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3402 ret = barrier_all_devices(root->fs_info);
3405 &root->fs_info->fs_devices->device_list_mutex);
3406 btrfs_error(root->fs_info, ret,
3407 "errors while submitting device barriers.");
3412 list_for_each_entry_rcu(dev, head, dev_list) {
3417 if (!dev->in_fs_metadata || !dev->writeable)
3420 btrfs_set_stack_device_generation(dev_item, 0);
3421 btrfs_set_stack_device_type(dev_item, dev->type);
3422 btrfs_set_stack_device_id(dev_item, dev->devid);
3423 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3424 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3425 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3426 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3427 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3428 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3429 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3431 flags = btrfs_super_flags(sb);
3432 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3434 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3438 if (total_errors > max_errors) {
3439 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3441 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3443 /* FUA is masked off if unsupported and can't be the reason */
3444 btrfs_error(root->fs_info, -EIO,
3445 "%d errors while writing supers", total_errors);
3450 list_for_each_entry_rcu(dev, head, dev_list) {
3453 if (!dev->in_fs_metadata || !dev->writeable)
3456 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3460 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3461 if (total_errors > max_errors) {
3462 btrfs_error(root->fs_info, -EIO,
3463 "%d errors while writing supers", total_errors);
3469 int write_ctree_super(struct btrfs_trans_handle *trans,
3470 struct btrfs_root *root, int max_mirrors)
3474 ret = write_all_supers(root, max_mirrors);
3478 /* Drop a fs root from the radix tree and free it. */
3479 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3480 struct btrfs_root *root)
3482 spin_lock(&fs_info->fs_roots_radix_lock);
3483 radix_tree_delete(&fs_info->fs_roots_radix,
3484 (unsigned long)root->root_key.objectid);
3485 spin_unlock(&fs_info->fs_roots_radix_lock);
3487 if (btrfs_root_refs(&root->root_item) == 0)
3488 synchronize_srcu(&fs_info->subvol_srcu);
3490 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3491 btrfs_free_log(NULL, root);
3492 btrfs_free_log_root_tree(NULL, fs_info);
3495 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3496 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3500 static void free_fs_root(struct btrfs_root *root)
3502 iput(root->cache_inode);
3503 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3504 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3505 root->orphan_block_rsv = NULL;
3507 free_anon_bdev(root->anon_dev);
3508 free_extent_buffer(root->node);
3509 free_extent_buffer(root->commit_root);
3510 kfree(root->free_ino_ctl);
3511 kfree(root->free_ino_pinned);
3513 btrfs_put_fs_root(root);
3516 void btrfs_free_fs_root(struct btrfs_root *root)
3521 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3523 u64 root_objectid = 0;
3524 struct btrfs_root *gang[8];
3529 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3530 (void **)gang, root_objectid,
3535 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3536 for (i = 0; i < ret; i++) {
3539 root_objectid = gang[i]->root_key.objectid;
3540 err = btrfs_orphan_cleanup(gang[i]);
3549 int btrfs_commit_super(struct btrfs_root *root)
3551 struct btrfs_trans_handle *trans;
3554 mutex_lock(&root->fs_info->cleaner_mutex);
3555 btrfs_run_delayed_iputs(root);
3556 mutex_unlock(&root->fs_info->cleaner_mutex);
3557 wake_up_process(root->fs_info->cleaner_kthread);
3559 /* wait until ongoing cleanup work done */
3560 down_write(&root->fs_info->cleanup_work_sem);
3561 up_write(&root->fs_info->cleanup_work_sem);
3563 trans = btrfs_join_transaction(root);
3565 return PTR_ERR(trans);
3566 ret = btrfs_commit_transaction(trans, root);
3569 /* run commit again to drop the original snapshot */
3570 trans = btrfs_join_transaction(root);
3572 return PTR_ERR(trans);
3573 ret = btrfs_commit_transaction(trans, root);
3576 ret = btrfs_write_and_wait_transaction(NULL, root);
3578 btrfs_error(root->fs_info, ret,
3579 "Failed to sync btree inode to disk.");
3583 ret = write_ctree_super(NULL, root, 0);
3587 int close_ctree(struct btrfs_root *root)
3589 struct btrfs_fs_info *fs_info = root->fs_info;
3592 fs_info->closing = 1;
3595 /* wait for the uuid_scan task to finish */
3596 down(&fs_info->uuid_tree_rescan_sem);
3597 /* avoid complains from lockdep et al., set sem back to initial state */
3598 up(&fs_info->uuid_tree_rescan_sem);
3600 /* pause restriper - we want to resume on mount */
3601 btrfs_pause_balance(fs_info);
3603 btrfs_dev_replace_suspend_for_unmount(fs_info);
3605 btrfs_scrub_cancel(fs_info);
3607 /* wait for any defraggers to finish */
3608 wait_event(fs_info->transaction_wait,
3609 (atomic_read(&fs_info->defrag_running) == 0));
3611 /* clear out the rbtree of defraggable inodes */
3612 btrfs_cleanup_defrag_inodes(fs_info);
3614 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3615 ret = btrfs_commit_super(root);
3617 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3620 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3621 btrfs_error_commit_super(root);
3623 btrfs_put_block_group_cache(fs_info);
3625 kthread_stop(fs_info->transaction_kthread);
3626 kthread_stop(fs_info->cleaner_kthread);
3628 fs_info->closing = 2;
3631 btrfs_free_qgroup_config(root->fs_info);
3633 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3634 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3635 percpu_counter_sum(&fs_info->delalloc_bytes));
3638 btrfs_free_block_groups(fs_info);
3640 btrfs_stop_all_workers(fs_info);
3642 del_fs_roots(fs_info);
3644 free_root_pointers(fs_info, 1);
3646 iput(fs_info->btree_inode);
3648 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3649 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3650 btrfsic_unmount(root, fs_info->fs_devices);
3653 btrfs_close_devices(fs_info->fs_devices);
3654 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3656 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3657 percpu_counter_destroy(&fs_info->delalloc_bytes);
3658 bdi_destroy(&fs_info->bdi);
3659 cleanup_srcu_struct(&fs_info->subvol_srcu);
3661 btrfs_free_stripe_hash_table(fs_info);
3663 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3664 root->orphan_block_rsv = NULL;
3669 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3673 struct inode *btree_inode = buf->pages[0]->mapping->host;
3675 ret = extent_buffer_uptodate(buf);
3679 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3680 parent_transid, atomic);
3686 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3688 return set_extent_buffer_uptodate(buf);
3691 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3693 struct btrfs_root *root;
3694 u64 transid = btrfs_header_generation(buf);
3697 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3699 * This is a fast path so only do this check if we have sanity tests
3700 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3701 * outside of the sanity tests.
3703 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3706 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3707 btrfs_assert_tree_locked(buf);
3708 if (transid != root->fs_info->generation)
3709 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3710 "found %llu running %llu\n",
3711 buf->start, transid, root->fs_info->generation);
3712 was_dirty = set_extent_buffer_dirty(buf);
3714 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3716 root->fs_info->dirty_metadata_batch);
3719 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3723 * looks as though older kernels can get into trouble with
3724 * this code, they end up stuck in balance_dirty_pages forever
3728 if (current->flags & PF_MEMALLOC)
3732 btrfs_balance_delayed_items(root);
3734 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3735 BTRFS_DIRTY_METADATA_THRESH);
3737 balance_dirty_pages_ratelimited(
3738 root->fs_info->btree_inode->i_mapping);
3743 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3745 __btrfs_btree_balance_dirty(root, 1);
3748 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3750 __btrfs_btree_balance_dirty(root, 0);
3753 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3755 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3756 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3759 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3763 * Placeholder for checks
3768 static void btrfs_error_commit_super(struct btrfs_root *root)
3770 mutex_lock(&root->fs_info->cleaner_mutex);
3771 btrfs_run_delayed_iputs(root);
3772 mutex_unlock(&root->fs_info->cleaner_mutex);
3774 down_write(&root->fs_info->cleanup_work_sem);
3775 up_write(&root->fs_info->cleanup_work_sem);
3777 /* cleanup FS via transaction */
3778 btrfs_cleanup_transaction(root);
3781 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3782 struct btrfs_root *root)
3784 struct btrfs_inode *btrfs_inode;
3785 struct list_head splice;
3787 INIT_LIST_HEAD(&splice);
3789 mutex_lock(&root->fs_info->ordered_operations_mutex);
3790 spin_lock(&root->fs_info->ordered_root_lock);
3792 list_splice_init(&t->ordered_operations, &splice);
3793 while (!list_empty(&splice)) {
3794 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3795 ordered_operations);
3797 list_del_init(&btrfs_inode->ordered_operations);
3798 spin_unlock(&root->fs_info->ordered_root_lock);
3800 btrfs_invalidate_inodes(btrfs_inode->root);
3802 spin_lock(&root->fs_info->ordered_root_lock);
3805 spin_unlock(&root->fs_info->ordered_root_lock);
3806 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3809 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3811 struct btrfs_ordered_extent *ordered;
3813 spin_lock(&root->ordered_extent_lock);
3815 * This will just short circuit the ordered completion stuff which will
3816 * make sure the ordered extent gets properly cleaned up.
3818 list_for_each_entry(ordered, &root->ordered_extents,
3820 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3821 spin_unlock(&root->ordered_extent_lock);
3824 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3826 struct btrfs_root *root;
3827 struct list_head splice;
3829 INIT_LIST_HEAD(&splice);
3831 spin_lock(&fs_info->ordered_root_lock);
3832 list_splice_init(&fs_info->ordered_roots, &splice);
3833 while (!list_empty(&splice)) {
3834 root = list_first_entry(&splice, struct btrfs_root,
3836 list_del_init(&root->ordered_root);
3838 btrfs_destroy_ordered_extents(root);
3840 cond_resched_lock(&fs_info->ordered_root_lock);
3842 spin_unlock(&fs_info->ordered_root_lock);
3845 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3846 struct btrfs_root *root)
3848 struct rb_node *node;
3849 struct btrfs_delayed_ref_root *delayed_refs;
3850 struct btrfs_delayed_ref_node *ref;
3853 delayed_refs = &trans->delayed_refs;
3855 spin_lock(&delayed_refs->lock);
3856 if (delayed_refs->num_entries == 0) {
3857 spin_unlock(&delayed_refs->lock);
3858 printk(KERN_INFO "delayed_refs has NO entry\n");
3862 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3863 struct btrfs_delayed_ref_head *head = NULL;
3864 bool pin_bytes = false;
3866 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3867 atomic_set(&ref->refs, 1);
3868 if (btrfs_delayed_ref_is_head(ref)) {
3870 head = btrfs_delayed_node_to_head(ref);
3871 if (!mutex_trylock(&head->mutex)) {
3872 atomic_inc(&ref->refs);
3873 spin_unlock(&delayed_refs->lock);
3875 /* Need to wait for the delayed ref to run */
3876 mutex_lock(&head->mutex);
3877 mutex_unlock(&head->mutex);
3878 btrfs_put_delayed_ref(ref);
3880 spin_lock(&delayed_refs->lock);
3884 if (head->must_insert_reserved)
3886 btrfs_free_delayed_extent_op(head->extent_op);
3887 delayed_refs->num_heads--;
3888 if (list_empty(&head->cluster))
3889 delayed_refs->num_heads_ready--;
3890 list_del_init(&head->cluster);
3894 rb_erase(&ref->rb_node, &delayed_refs->root);
3895 delayed_refs->num_entries--;
3896 spin_unlock(&delayed_refs->lock);
3899 btrfs_pin_extent(root, ref->bytenr,
3901 mutex_unlock(&head->mutex);
3903 btrfs_put_delayed_ref(ref);
3906 spin_lock(&delayed_refs->lock);
3909 spin_unlock(&delayed_refs->lock);
3914 static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
3916 struct btrfs_pending_snapshot *snapshot;
3917 struct list_head splice;
3919 INIT_LIST_HEAD(&splice);
3921 list_splice_init(&t->pending_snapshots, &splice);
3923 while (!list_empty(&splice)) {
3924 snapshot = list_entry(splice.next,
3925 struct btrfs_pending_snapshot,
3927 snapshot->error = -ECANCELED;
3928 list_del_init(&snapshot->list);
3932 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3934 struct btrfs_inode *btrfs_inode;
3935 struct list_head splice;
3937 INIT_LIST_HEAD(&splice);
3939 spin_lock(&root->delalloc_lock);
3940 list_splice_init(&root->delalloc_inodes, &splice);
3942 while (!list_empty(&splice)) {
3943 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3946 list_del_init(&btrfs_inode->delalloc_inodes);
3947 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3948 &btrfs_inode->runtime_flags);
3949 spin_unlock(&root->delalloc_lock);
3951 btrfs_invalidate_inodes(btrfs_inode->root);
3953 spin_lock(&root->delalloc_lock);
3956 spin_unlock(&root->delalloc_lock);
3959 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3961 struct btrfs_root *root;
3962 struct list_head splice;
3964 INIT_LIST_HEAD(&splice);
3966 spin_lock(&fs_info->delalloc_root_lock);
3967 list_splice_init(&fs_info->delalloc_roots, &splice);
3968 while (!list_empty(&splice)) {
3969 root = list_first_entry(&splice, struct btrfs_root,
3971 list_del_init(&root->delalloc_root);
3972 root = btrfs_grab_fs_root(root);
3974 spin_unlock(&fs_info->delalloc_root_lock);
3976 btrfs_destroy_delalloc_inodes(root);
3977 btrfs_put_fs_root(root);
3979 spin_lock(&fs_info->delalloc_root_lock);
3981 spin_unlock(&fs_info->delalloc_root_lock);
3984 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3985 struct extent_io_tree *dirty_pages,
3989 struct extent_buffer *eb;
3994 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3999 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4000 while (start <= end) {
4001 eb = btrfs_find_tree_block(root, start,
4003 start += root->leafsize;
4006 wait_on_extent_buffer_writeback(eb);
4008 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4010 clear_extent_buffer_dirty(eb);
4011 free_extent_buffer_stale(eb);
4018 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4019 struct extent_io_tree *pinned_extents)
4021 struct extent_io_tree *unpin;
4027 unpin = pinned_extents;
4030 ret = find_first_extent_bit(unpin, 0, &start, &end,
4031 EXTENT_DIRTY, NULL);
4036 if (btrfs_test_opt(root, DISCARD))
4037 ret = btrfs_error_discard_extent(root, start,
4041 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4042 btrfs_error_unpin_extent_range(root, start, end);
4047 if (unpin == &root->fs_info->freed_extents[0])
4048 unpin = &root->fs_info->freed_extents[1];
4050 unpin = &root->fs_info->freed_extents[0];
4058 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4059 struct btrfs_root *root)
4061 btrfs_destroy_delayed_refs(cur_trans, root);
4062 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
4063 cur_trans->dirty_pages.dirty_bytes);
4065 cur_trans->state = TRANS_STATE_COMMIT_START;
4066 wake_up(&root->fs_info->transaction_blocked_wait);
4068 btrfs_evict_pending_snapshots(cur_trans);
4070 cur_trans->state = TRANS_STATE_UNBLOCKED;
4071 wake_up(&root->fs_info->transaction_wait);
4073 btrfs_destroy_delayed_inodes(root);
4074 btrfs_assert_delayed_root_empty(root);
4076 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4078 btrfs_destroy_pinned_extent(root,
4079 root->fs_info->pinned_extents);
4081 cur_trans->state =TRANS_STATE_COMPLETED;
4082 wake_up(&cur_trans->commit_wait);
4085 memset(cur_trans, 0, sizeof(*cur_trans));
4086 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4090 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4092 struct btrfs_transaction *t;
4095 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4097 spin_lock(&root->fs_info->trans_lock);
4098 list_splice_init(&root->fs_info->trans_list, &list);
4099 root->fs_info->running_transaction = NULL;
4100 spin_unlock(&root->fs_info->trans_lock);
4102 while (!list_empty(&list)) {
4103 t = list_entry(list.next, struct btrfs_transaction, list);
4105 btrfs_destroy_ordered_operations(t, root);
4107 btrfs_destroy_all_ordered_extents(root->fs_info);
4109 btrfs_destroy_delayed_refs(t, root);
4112 * FIXME: cleanup wait for commit
4113 * We needn't acquire the lock here, because we are during
4114 * the umount, there is no other task which will change it.
4116 t->state = TRANS_STATE_COMMIT_START;
4118 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
4119 wake_up(&root->fs_info->transaction_blocked_wait);
4121 btrfs_evict_pending_snapshots(t);
4123 t->state = TRANS_STATE_UNBLOCKED;
4125 if (waitqueue_active(&root->fs_info->transaction_wait))
4126 wake_up(&root->fs_info->transaction_wait);
4128 btrfs_destroy_delayed_inodes(root);
4129 btrfs_assert_delayed_root_empty(root);
4131 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4133 btrfs_destroy_marked_extents(root, &t->dirty_pages,
4136 btrfs_destroy_pinned_extent(root,
4137 root->fs_info->pinned_extents);
4139 t->state = TRANS_STATE_COMPLETED;
4141 if (waitqueue_active(&t->commit_wait))
4142 wake_up(&t->commit_wait);
4144 atomic_set(&t->use_count, 0);
4145 list_del_init(&t->list);
4146 memset(t, 0, sizeof(*t));
4147 kmem_cache_free(btrfs_transaction_cachep, t);
4150 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4155 static struct extent_io_ops btree_extent_io_ops = {
4156 .readpage_end_io_hook = btree_readpage_end_io_hook,
4157 .readpage_io_failed_hook = btree_io_failed_hook,
4158 .submit_bio_hook = btree_submit_bio_hook,
4159 /* note we're sharing with inode.c for the merge bio hook */
4160 .merge_bio_hook = btrfs_merge_bio_hook,