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>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "print-tree.h"
42 #include "async-thread.h"
45 #include "free-space-cache.h"
46 #include "inode-map.h"
47 #include "check-integrity.h"
48 #include "rcu-string.h"
49 #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_destroy_delalloc_inodes(struct btrfs_root *root);
68 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
69 struct extent_io_tree *dirty_pages,
71 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
72 struct extent_io_tree *pinned_extents);
73 static int btrfs_cleanup_transaction(struct btrfs_root *root);
74 static void btrfs_error_commit_super(struct btrfs_root *root);
77 * end_io_wq structs are used to do processing in task context when an IO is
78 * complete. This is used during reads to verify checksums, and it is used
79 * by writes to insert metadata for new file extents after IO is complete.
85 struct btrfs_fs_info *info;
88 struct list_head list;
89 struct btrfs_work work;
93 * async submit bios are used to offload expensive checksumming
94 * onto the worker threads. They checksum file and metadata bios
95 * just before they are sent down the IO stack.
97 struct async_submit_bio {
100 struct list_head list;
101 extent_submit_bio_hook_t *submit_bio_start;
102 extent_submit_bio_hook_t *submit_bio_done;
105 unsigned long bio_flags;
107 * bio_offset is optional, can be used if the pages in the bio
108 * can't tell us where in the file the bio should go
111 struct btrfs_work work;
116 * Lockdep class keys for extent_buffer->lock's in this root. For a given
117 * eb, the lockdep key is determined by the btrfs_root it belongs to and
118 * the level the eb occupies in the tree.
120 * Different roots are used for different purposes and may nest inside each
121 * other and they require separate keysets. As lockdep keys should be
122 * static, assign keysets according to the purpose of the root as indicated
123 * by btrfs_root->objectid. This ensures that all special purpose roots
124 * have separate keysets.
126 * Lock-nesting across peer nodes is always done with the immediate parent
127 * node locked thus preventing deadlock. As lockdep doesn't know this, use
128 * subclass to avoid triggering lockdep warning in such cases.
130 * The key is set by the readpage_end_io_hook after the buffer has passed
131 * csum validation but before the pages are unlocked. It is also set by
132 * btrfs_init_new_buffer on freshly allocated blocks.
134 * We also add a check to make sure the highest level of the tree is the
135 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
136 * needs update as well.
138 #ifdef CONFIG_DEBUG_LOCK_ALLOC
139 # if BTRFS_MAX_LEVEL != 8
143 static struct btrfs_lockdep_keyset {
144 u64 id; /* root objectid */
145 const char *name_stem; /* lock name stem */
146 char names[BTRFS_MAX_LEVEL + 1][20];
147 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
148 } btrfs_lockdep_keysets[] = {
149 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
150 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
151 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
152 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
153 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
154 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
155 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
156 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
157 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
158 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
159 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
160 { .id = 0, .name_stem = "tree" },
163 void __init btrfs_init_lockdep(void)
167 /* initialize lockdep class names */
168 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
169 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
171 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
172 snprintf(ks->names[j], sizeof(ks->names[j]),
173 "btrfs-%s-%02d", ks->name_stem, j);
177 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
180 struct btrfs_lockdep_keyset *ks;
182 BUG_ON(level >= ARRAY_SIZE(ks->keys));
184 /* find the matching keyset, id 0 is the default entry */
185 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
186 if (ks->id == objectid)
189 lockdep_set_class_and_name(&eb->lock,
190 &ks->keys[level], ks->names[level]);
196 * extents on the btree inode are pretty simple, there's one extent
197 * that covers the entire device
199 static struct extent_map *btree_get_extent(struct inode *inode,
200 struct page *page, size_t pg_offset, u64 start, u64 len,
203 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
204 struct extent_map *em;
207 read_lock(&em_tree->lock);
208 em = lookup_extent_mapping(em_tree, start, len);
211 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
212 read_unlock(&em_tree->lock);
215 read_unlock(&em_tree->lock);
217 em = alloc_extent_map();
219 em = ERR_PTR(-ENOMEM);
224 em->block_len = (u64)-1;
226 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
228 write_lock(&em_tree->lock);
229 ret = add_extent_mapping(em_tree, em, 0);
230 if (ret == -EEXIST) {
232 em = lookup_extent_mapping(em_tree, start, len);
239 write_unlock(&em_tree->lock);
245 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
247 return crc32c(seed, data, len);
250 void btrfs_csum_final(u32 crc, char *result)
252 put_unaligned_le32(~crc, result);
256 * compute the csum for a btree block, and either verify it or write it
257 * into the csum field of the block.
259 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
262 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
265 unsigned long cur_len;
266 unsigned long offset = BTRFS_CSUM_SIZE;
268 unsigned long map_start;
269 unsigned long map_len;
272 unsigned long inline_result;
274 len = buf->len - offset;
276 err = map_private_extent_buffer(buf, offset, 32,
277 &kaddr, &map_start, &map_len);
280 cur_len = min(len, map_len - (offset - map_start));
281 crc = btrfs_csum_data(kaddr + offset - map_start,
286 if (csum_size > sizeof(inline_result)) {
287 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
291 result = (char *)&inline_result;
294 btrfs_csum_final(crc, result);
297 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
300 memcpy(&found, result, csum_size);
302 read_extent_buffer(buf, &val, 0, csum_size);
303 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
304 "failed on %llu wanted %X found %X "
306 root->fs_info->sb->s_id, buf->start,
307 val, found, btrfs_header_level(buf));
308 if (result != (char *)&inline_result)
313 write_extent_buffer(buf, result, 0, csum_size);
315 if (result != (char *)&inline_result)
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
326 static int verify_parent_transid(struct extent_io_tree *io_tree,
327 struct extent_buffer *eb, u64 parent_transid,
330 struct extent_state *cached_state = NULL;
333 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
339 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
341 if (extent_buffer_uptodate(eb) &&
342 btrfs_header_generation(eb) == parent_transid) {
346 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
348 eb->start, parent_transid, btrfs_header_generation(eb));
350 clear_extent_buffer_uptodate(eb);
352 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
353 &cached_state, GFP_NOFS);
358 * Return 0 if the superblock checksum type matches the checksum value of that
359 * algorithm. Pass the raw disk superblock data.
361 static int btrfs_check_super_csum(char *raw_disk_sb)
363 struct btrfs_super_block *disk_sb =
364 (struct btrfs_super_block *)raw_disk_sb;
365 u16 csum_type = btrfs_super_csum_type(disk_sb);
368 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
370 const int csum_size = sizeof(crc);
371 char result[csum_size];
374 * The super_block structure does not span the whole
375 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
376 * is filled with zeros and is included in the checkum.
378 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
379 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
380 btrfs_csum_final(crc, result);
382 if (memcmp(raw_disk_sb, result, csum_size))
385 if (ret && btrfs_super_generation(disk_sb) < 10) {
386 printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
391 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
392 printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
401 * helper to read a given tree block, doing retries as required when
402 * the checksums don't match and we have alternate mirrors to try.
404 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
405 struct extent_buffer *eb,
406 u64 start, u64 parent_transid)
408 struct extent_io_tree *io_tree;
413 int failed_mirror = 0;
415 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
416 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
418 ret = read_extent_buffer_pages(io_tree, eb, start,
420 btree_get_extent, mirror_num);
422 if (!verify_parent_transid(io_tree, eb,
430 * This buffer's crc is fine, but its contents are corrupted, so
431 * there is no reason to read the other copies, they won't be
434 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
437 num_copies = btrfs_num_copies(root->fs_info,
442 if (!failed_mirror) {
444 failed_mirror = eb->read_mirror;
448 if (mirror_num == failed_mirror)
451 if (mirror_num > num_copies)
455 if (failed && !ret && failed_mirror)
456 repair_eb_io_failure(root, eb, failed_mirror);
462 * checksum a dirty tree block before IO. This has extra checks to make sure
463 * we only fill in the checksum field in the first page of a multi-page block
466 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
468 struct extent_io_tree *tree;
469 u64 start = page_offset(page);
471 struct extent_buffer *eb;
473 tree = &BTRFS_I(page->mapping->host)->io_tree;
475 eb = (struct extent_buffer *)page->private;
476 if (page != eb->pages[0])
478 found_start = btrfs_header_bytenr(eb);
479 if (WARN_ON(found_start != start || !PageUptodate(page)))
481 csum_tree_block(root, eb, 0);
485 static int check_tree_block_fsid(struct btrfs_root *root,
486 struct extent_buffer *eb)
488 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
489 u8 fsid[BTRFS_UUID_SIZE];
492 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
494 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
498 fs_devices = fs_devices->seed;
503 #define CORRUPT(reason, eb, root, slot) \
504 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
505 "root=%llu, slot=%d\n", reason, \
506 btrfs_header_bytenr(eb), root->objectid, slot)
508 static noinline int check_leaf(struct btrfs_root *root,
509 struct extent_buffer *leaf)
511 struct btrfs_key key;
512 struct btrfs_key leaf_key;
513 u32 nritems = btrfs_header_nritems(leaf);
519 /* Check the 0 item */
520 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
521 BTRFS_LEAF_DATA_SIZE(root)) {
522 CORRUPT("invalid item offset size pair", leaf, root, 0);
527 * Check to make sure each items keys are in the correct order and their
528 * offsets make sense. We only have to loop through nritems-1 because
529 * we check the current slot against the next slot, which verifies the
530 * next slot's offset+size makes sense and that the current's slot
533 for (slot = 0; slot < nritems - 1; slot++) {
534 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
535 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
537 /* Make sure the keys are in the right order */
538 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
539 CORRUPT("bad key order", leaf, root, slot);
544 * Make sure the offset and ends are right, remember that the
545 * item data starts at the end of the leaf and grows towards the
548 if (btrfs_item_offset_nr(leaf, slot) !=
549 btrfs_item_end_nr(leaf, slot + 1)) {
550 CORRUPT("slot offset bad", leaf, root, slot);
555 * Check to make sure that we don't point outside of the leaf,
556 * just incase all the items are consistent to eachother, but
557 * all point outside of the leaf.
559 if (btrfs_item_end_nr(leaf, slot) >
560 BTRFS_LEAF_DATA_SIZE(root)) {
561 CORRUPT("slot end outside of leaf", leaf, root, slot);
569 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
570 u64 phy_offset, struct page *page,
571 u64 start, u64 end, int mirror)
573 struct extent_io_tree *tree;
576 struct extent_buffer *eb;
577 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
584 tree = &BTRFS_I(page->mapping->host)->io_tree;
585 eb = (struct extent_buffer *)page->private;
587 /* the pending IO might have been the only thing that kept this buffer
588 * in memory. Make sure we have a ref for all this other checks
590 extent_buffer_get(eb);
592 reads_done = atomic_dec_and_test(&eb->io_pages);
596 eb->read_mirror = mirror;
597 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
602 found_start = btrfs_header_bytenr(eb);
603 if (found_start != eb->start) {
604 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
606 found_start, eb->start);
610 if (check_tree_block_fsid(root, eb)) {
611 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
616 found_level = btrfs_header_level(eb);
617 if (found_level >= BTRFS_MAX_LEVEL) {
618 btrfs_info(root->fs_info, "bad tree block level %d\n",
619 (int)btrfs_header_level(eb));
624 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
627 ret = csum_tree_block(root, eb, 1);
634 * If this is a leaf block and it is corrupt, set the corrupt bit so
635 * that we don't try and read the other copies of this block, just
638 if (found_level == 0 && check_leaf(root, eb)) {
639 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
644 set_extent_buffer_uptodate(eb);
647 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
648 btree_readahead_hook(root, eb, eb->start, ret);
652 * our io error hook is going to dec the io pages
653 * again, we have to make sure it has something
656 atomic_inc(&eb->io_pages);
657 clear_extent_buffer_uptodate(eb);
659 free_extent_buffer(eb);
664 static int btree_io_failed_hook(struct page *page, int failed_mirror)
666 struct extent_buffer *eb;
667 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
669 eb = (struct extent_buffer *)page->private;
670 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
671 eb->read_mirror = failed_mirror;
672 atomic_dec(&eb->io_pages);
673 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
674 btree_readahead_hook(root, eb, eb->start, -EIO);
675 return -EIO; /* we fixed nothing */
678 static void end_workqueue_bio(struct bio *bio, int err)
680 struct end_io_wq *end_io_wq = bio->bi_private;
681 struct btrfs_fs_info *fs_info;
683 fs_info = end_io_wq->info;
684 end_io_wq->error = err;
685 end_io_wq->work.func = end_workqueue_fn;
686 end_io_wq->work.flags = 0;
688 if (bio->bi_rw & REQ_WRITE) {
689 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
690 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
692 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
693 btrfs_queue_worker(&fs_info->endio_freespace_worker,
695 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
696 btrfs_queue_worker(&fs_info->endio_raid56_workers,
699 btrfs_queue_worker(&fs_info->endio_write_workers,
702 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
703 btrfs_queue_worker(&fs_info->endio_raid56_workers,
705 else if (end_io_wq->metadata)
706 btrfs_queue_worker(&fs_info->endio_meta_workers,
709 btrfs_queue_worker(&fs_info->endio_workers,
715 * For the metadata arg you want
718 * 1 - if normal metadta
719 * 2 - if writing to the free space cache area
720 * 3 - raid parity work
722 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
725 struct end_io_wq *end_io_wq;
726 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
730 end_io_wq->private = bio->bi_private;
731 end_io_wq->end_io = bio->bi_end_io;
732 end_io_wq->info = info;
733 end_io_wq->error = 0;
734 end_io_wq->bio = bio;
735 end_io_wq->metadata = metadata;
737 bio->bi_private = end_io_wq;
738 bio->bi_end_io = end_workqueue_bio;
742 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
744 unsigned long limit = min_t(unsigned long,
745 info->workers.max_workers,
746 info->fs_devices->open_devices);
750 static void run_one_async_start(struct btrfs_work *work)
752 struct async_submit_bio *async;
755 async = container_of(work, struct async_submit_bio, work);
756 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
757 async->mirror_num, async->bio_flags,
763 static void run_one_async_done(struct btrfs_work *work)
765 struct btrfs_fs_info *fs_info;
766 struct async_submit_bio *async;
769 async = container_of(work, struct async_submit_bio, work);
770 fs_info = BTRFS_I(async->inode)->root->fs_info;
772 limit = btrfs_async_submit_limit(fs_info);
773 limit = limit * 2 / 3;
775 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
776 waitqueue_active(&fs_info->async_submit_wait))
777 wake_up(&fs_info->async_submit_wait);
779 /* If an error occured we just want to clean up the bio and move on */
781 bio_endio(async->bio, async->error);
785 async->submit_bio_done(async->inode, async->rw, async->bio,
786 async->mirror_num, async->bio_flags,
790 static void run_one_async_free(struct btrfs_work *work)
792 struct async_submit_bio *async;
794 async = container_of(work, struct async_submit_bio, work);
798 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
799 int rw, struct bio *bio, int mirror_num,
800 unsigned long bio_flags,
802 extent_submit_bio_hook_t *submit_bio_start,
803 extent_submit_bio_hook_t *submit_bio_done)
805 struct async_submit_bio *async;
807 async = kmalloc(sizeof(*async), GFP_NOFS);
811 async->inode = inode;
814 async->mirror_num = mirror_num;
815 async->submit_bio_start = submit_bio_start;
816 async->submit_bio_done = submit_bio_done;
818 async->work.func = run_one_async_start;
819 async->work.ordered_func = run_one_async_done;
820 async->work.ordered_free = run_one_async_free;
822 async->work.flags = 0;
823 async->bio_flags = bio_flags;
824 async->bio_offset = bio_offset;
828 atomic_inc(&fs_info->nr_async_submits);
831 btrfs_set_work_high_prio(&async->work);
833 btrfs_queue_worker(&fs_info->workers, &async->work);
835 while (atomic_read(&fs_info->async_submit_draining) &&
836 atomic_read(&fs_info->nr_async_submits)) {
837 wait_event(fs_info->async_submit_wait,
838 (atomic_read(&fs_info->nr_async_submits) == 0));
844 static int btree_csum_one_bio(struct bio *bio)
846 struct bio_vec *bvec = bio->bi_io_vec;
848 struct btrfs_root *root;
851 WARN_ON(bio->bi_vcnt <= 0);
852 while (bio_index < bio->bi_vcnt) {
853 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
854 ret = csum_dirty_buffer(root, bvec->bv_page);
863 static int __btree_submit_bio_start(struct inode *inode, int rw,
864 struct bio *bio, int mirror_num,
865 unsigned long bio_flags,
869 * when we're called for a write, we're already in the async
870 * submission context. Just jump into btrfs_map_bio
872 return btree_csum_one_bio(bio);
875 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
876 int mirror_num, unsigned long bio_flags,
882 * when we're called for a write, we're already in the async
883 * submission context. Just jump into btrfs_map_bio
885 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
891 static int check_async_write(struct inode *inode, unsigned long bio_flags)
893 if (bio_flags & EXTENT_BIO_TREE_LOG)
902 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
903 int mirror_num, unsigned long bio_flags,
906 int async = check_async_write(inode, bio_flags);
909 if (!(rw & REQ_WRITE)) {
911 * called for a read, do the setup so that checksum validation
912 * can happen in the async kernel threads
914 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
918 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
921 ret = btree_csum_one_bio(bio);
924 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
928 * kthread helpers are used to submit writes so that
929 * checksumming can happen in parallel across all CPUs
931 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
932 inode, rw, bio, mirror_num, 0,
934 __btree_submit_bio_start,
935 __btree_submit_bio_done);
945 #ifdef CONFIG_MIGRATION
946 static int btree_migratepage(struct address_space *mapping,
947 struct page *newpage, struct page *page,
948 enum migrate_mode mode)
951 * we can't safely write a btree page from here,
952 * we haven't done the locking hook
957 * Buffers may be managed in a filesystem specific way.
958 * We must have no buffers or drop them.
960 if (page_has_private(page) &&
961 !try_to_release_page(page, GFP_KERNEL))
963 return migrate_page(mapping, newpage, page, mode);
968 static int btree_writepages(struct address_space *mapping,
969 struct writeback_control *wbc)
971 struct extent_io_tree *tree;
972 struct btrfs_fs_info *fs_info;
975 tree = &BTRFS_I(mapping->host)->io_tree;
976 if (wbc->sync_mode == WB_SYNC_NONE) {
978 if (wbc->for_kupdate)
981 fs_info = BTRFS_I(mapping->host)->root->fs_info;
982 /* this is a bit racy, but that's ok */
983 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
984 BTRFS_DIRTY_METADATA_THRESH);
988 return btree_write_cache_pages(mapping, wbc);
991 static int btree_readpage(struct file *file, struct page *page)
993 struct extent_io_tree *tree;
994 tree = &BTRFS_I(page->mapping->host)->io_tree;
995 return extent_read_full_page(tree, page, btree_get_extent, 0);
998 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1000 if (PageWriteback(page) || PageDirty(page))
1003 return try_release_extent_buffer(page);
1006 static void btree_invalidatepage(struct page *page, unsigned int offset,
1007 unsigned int length)
1009 struct extent_io_tree *tree;
1010 tree = &BTRFS_I(page->mapping->host)->io_tree;
1011 extent_invalidatepage(tree, page, offset);
1012 btree_releasepage(page, GFP_NOFS);
1013 if (PagePrivate(page)) {
1014 printk(KERN_WARNING "btrfs warning page private not zero "
1015 "on page %llu\n", (unsigned long long)page_offset(page));
1016 ClearPagePrivate(page);
1017 set_page_private(page, 0);
1018 page_cache_release(page);
1022 static int btree_set_page_dirty(struct page *page)
1025 struct extent_buffer *eb;
1027 BUG_ON(!PagePrivate(page));
1028 eb = (struct extent_buffer *)page->private;
1030 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1031 BUG_ON(!atomic_read(&eb->refs));
1032 btrfs_assert_tree_locked(eb);
1034 return __set_page_dirty_nobuffers(page);
1037 static const struct address_space_operations btree_aops = {
1038 .readpage = btree_readpage,
1039 .writepages = btree_writepages,
1040 .releasepage = btree_releasepage,
1041 .invalidatepage = btree_invalidatepage,
1042 #ifdef CONFIG_MIGRATION
1043 .migratepage = btree_migratepage,
1045 .set_page_dirty = btree_set_page_dirty,
1048 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1051 struct extent_buffer *buf = NULL;
1052 struct inode *btree_inode = root->fs_info->btree_inode;
1055 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1058 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1059 buf, 0, WAIT_NONE, btree_get_extent, 0);
1060 free_extent_buffer(buf);
1064 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1065 int mirror_num, struct extent_buffer **eb)
1067 struct extent_buffer *buf = NULL;
1068 struct inode *btree_inode = root->fs_info->btree_inode;
1069 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1072 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1076 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1078 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1079 btree_get_extent, mirror_num);
1081 free_extent_buffer(buf);
1085 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1086 free_extent_buffer(buf);
1088 } else if (extent_buffer_uptodate(buf)) {
1091 free_extent_buffer(buf);
1096 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1097 u64 bytenr, u32 blocksize)
1099 struct inode *btree_inode = root->fs_info->btree_inode;
1100 struct extent_buffer *eb;
1101 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree, bytenr);
1105 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1106 u64 bytenr, u32 blocksize)
1108 struct inode *btree_inode = root->fs_info->btree_inode;
1109 struct extent_buffer *eb;
1111 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1117 int btrfs_write_tree_block(struct extent_buffer *buf)
1119 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1120 buf->start + buf->len - 1);
1123 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1125 return filemap_fdatawait_range(buf->pages[0]->mapping,
1126 buf->start, buf->start + buf->len - 1);
1129 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1130 u32 blocksize, u64 parent_transid)
1132 struct extent_buffer *buf = NULL;
1135 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1139 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1141 free_extent_buffer(buf);
1148 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1149 struct extent_buffer *buf)
1151 struct btrfs_fs_info *fs_info = root->fs_info;
1153 if (btrfs_header_generation(buf) ==
1154 fs_info->running_transaction->transid) {
1155 btrfs_assert_tree_locked(buf);
1157 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1158 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1160 fs_info->dirty_metadata_batch);
1161 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1162 btrfs_set_lock_blocking(buf);
1163 clear_extent_buffer_dirty(buf);
1168 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1169 u32 stripesize, struct btrfs_root *root,
1170 struct btrfs_fs_info *fs_info,
1174 root->commit_root = NULL;
1175 root->sectorsize = sectorsize;
1176 root->nodesize = nodesize;
1177 root->leafsize = leafsize;
1178 root->stripesize = stripesize;
1180 root->track_dirty = 0;
1182 root->orphan_item_inserted = 0;
1183 root->orphan_cleanup_state = 0;
1185 root->objectid = objectid;
1186 root->last_trans = 0;
1187 root->highest_objectid = 0;
1188 root->nr_delalloc_inodes = 0;
1189 root->nr_ordered_extents = 0;
1191 root->inode_tree = RB_ROOT;
1192 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1193 root->block_rsv = NULL;
1194 root->orphan_block_rsv = NULL;
1196 INIT_LIST_HEAD(&root->dirty_list);
1197 INIT_LIST_HEAD(&root->root_list);
1198 INIT_LIST_HEAD(&root->delalloc_inodes);
1199 INIT_LIST_HEAD(&root->delalloc_root);
1200 INIT_LIST_HEAD(&root->ordered_extents);
1201 INIT_LIST_HEAD(&root->ordered_root);
1202 INIT_LIST_HEAD(&root->logged_list[0]);
1203 INIT_LIST_HEAD(&root->logged_list[1]);
1204 spin_lock_init(&root->orphan_lock);
1205 spin_lock_init(&root->inode_lock);
1206 spin_lock_init(&root->delalloc_lock);
1207 spin_lock_init(&root->ordered_extent_lock);
1208 spin_lock_init(&root->accounting_lock);
1209 spin_lock_init(&root->log_extents_lock[0]);
1210 spin_lock_init(&root->log_extents_lock[1]);
1211 mutex_init(&root->objectid_mutex);
1212 mutex_init(&root->log_mutex);
1213 init_waitqueue_head(&root->log_writer_wait);
1214 init_waitqueue_head(&root->log_commit_wait[0]);
1215 init_waitqueue_head(&root->log_commit_wait[1]);
1216 atomic_set(&root->log_commit[0], 0);
1217 atomic_set(&root->log_commit[1], 0);
1218 atomic_set(&root->log_writers, 0);
1219 atomic_set(&root->log_batch, 0);
1220 atomic_set(&root->orphan_inodes, 0);
1221 atomic_set(&root->refs, 1);
1222 root->log_transid = 0;
1223 root->last_log_commit = 0;
1225 extent_io_tree_init(&root->dirty_log_pages,
1226 fs_info->btree_inode->i_mapping);
1228 memset(&root->root_key, 0, sizeof(root->root_key));
1229 memset(&root->root_item, 0, sizeof(root->root_item));
1230 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1231 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1233 root->defrag_trans_start = fs_info->generation;
1235 root->defrag_trans_start = 0;
1236 init_completion(&root->kobj_unregister);
1237 root->defrag_running = 0;
1238 root->root_key.objectid = objectid;
1241 spin_lock_init(&root->root_item_lock);
1244 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1246 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1248 root->fs_info = fs_info;
1252 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1253 /* Should only be used by the testing infrastructure */
1254 struct btrfs_root *btrfs_alloc_dummy_root(void)
1256 struct btrfs_root *root;
1258 root = btrfs_alloc_root(NULL);
1260 return ERR_PTR(-ENOMEM);
1261 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1262 root->dummy_root = 1;
1268 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1269 struct btrfs_fs_info *fs_info,
1272 struct extent_buffer *leaf;
1273 struct btrfs_root *tree_root = fs_info->tree_root;
1274 struct btrfs_root *root;
1275 struct btrfs_key key;
1280 root = btrfs_alloc_root(fs_info);
1282 return ERR_PTR(-ENOMEM);
1284 __setup_root(tree_root->nodesize, tree_root->leafsize,
1285 tree_root->sectorsize, tree_root->stripesize,
1286 root, fs_info, objectid);
1287 root->root_key.objectid = objectid;
1288 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1289 root->root_key.offset = 0;
1291 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1292 0, objectid, NULL, 0, 0, 0);
1294 ret = PTR_ERR(leaf);
1299 bytenr = leaf->start;
1300 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1301 btrfs_set_header_bytenr(leaf, leaf->start);
1302 btrfs_set_header_generation(leaf, trans->transid);
1303 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1304 btrfs_set_header_owner(leaf, objectid);
1307 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1309 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1310 btrfs_header_chunk_tree_uuid(leaf),
1312 btrfs_mark_buffer_dirty(leaf);
1314 root->commit_root = btrfs_root_node(root);
1315 root->track_dirty = 1;
1318 root->root_item.flags = 0;
1319 root->root_item.byte_limit = 0;
1320 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1321 btrfs_set_root_generation(&root->root_item, trans->transid);
1322 btrfs_set_root_level(&root->root_item, 0);
1323 btrfs_set_root_refs(&root->root_item, 1);
1324 btrfs_set_root_used(&root->root_item, leaf->len);
1325 btrfs_set_root_last_snapshot(&root->root_item, 0);
1326 btrfs_set_root_dirid(&root->root_item, 0);
1328 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1329 root->root_item.drop_level = 0;
1331 key.objectid = objectid;
1332 key.type = BTRFS_ROOT_ITEM_KEY;
1334 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1338 btrfs_tree_unlock(leaf);
1344 btrfs_tree_unlock(leaf);
1345 free_extent_buffer(leaf);
1349 return ERR_PTR(ret);
1352 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1353 struct btrfs_fs_info *fs_info)
1355 struct btrfs_root *root;
1356 struct btrfs_root *tree_root = fs_info->tree_root;
1357 struct extent_buffer *leaf;
1359 root = btrfs_alloc_root(fs_info);
1361 return ERR_PTR(-ENOMEM);
1363 __setup_root(tree_root->nodesize, tree_root->leafsize,
1364 tree_root->sectorsize, tree_root->stripesize,
1365 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1367 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1368 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1369 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1371 * log trees do not get reference counted because they go away
1372 * before a real commit is actually done. They do store pointers
1373 * to file data extents, and those reference counts still get
1374 * updated (along with back refs to the log tree).
1378 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1379 BTRFS_TREE_LOG_OBJECTID, NULL,
1383 return ERR_CAST(leaf);
1386 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1387 btrfs_set_header_bytenr(leaf, leaf->start);
1388 btrfs_set_header_generation(leaf, trans->transid);
1389 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1390 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1393 write_extent_buffer(root->node, root->fs_info->fsid,
1394 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1395 btrfs_mark_buffer_dirty(root->node);
1396 btrfs_tree_unlock(root->node);
1400 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1401 struct btrfs_fs_info *fs_info)
1403 struct btrfs_root *log_root;
1405 log_root = alloc_log_tree(trans, fs_info);
1406 if (IS_ERR(log_root))
1407 return PTR_ERR(log_root);
1408 WARN_ON(fs_info->log_root_tree);
1409 fs_info->log_root_tree = log_root;
1413 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1414 struct btrfs_root *root)
1416 struct btrfs_root *log_root;
1417 struct btrfs_inode_item *inode_item;
1419 log_root = alloc_log_tree(trans, root->fs_info);
1420 if (IS_ERR(log_root))
1421 return PTR_ERR(log_root);
1423 log_root->last_trans = trans->transid;
1424 log_root->root_key.offset = root->root_key.objectid;
1426 inode_item = &log_root->root_item.inode;
1427 btrfs_set_stack_inode_generation(inode_item, 1);
1428 btrfs_set_stack_inode_size(inode_item, 3);
1429 btrfs_set_stack_inode_nlink(inode_item, 1);
1430 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1431 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1433 btrfs_set_root_node(&log_root->root_item, log_root->node);
1435 WARN_ON(root->log_root);
1436 root->log_root = log_root;
1437 root->log_transid = 0;
1438 root->last_log_commit = 0;
1442 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1443 struct btrfs_key *key)
1445 struct btrfs_root *root;
1446 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1447 struct btrfs_path *path;
1452 path = btrfs_alloc_path();
1454 return ERR_PTR(-ENOMEM);
1456 root = btrfs_alloc_root(fs_info);
1462 __setup_root(tree_root->nodesize, tree_root->leafsize,
1463 tree_root->sectorsize, tree_root->stripesize,
1464 root, fs_info, key->objectid);
1466 ret = btrfs_find_root(tree_root, key, path,
1467 &root->root_item, &root->root_key);
1474 generation = btrfs_root_generation(&root->root_item);
1475 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1476 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1477 blocksize, generation);
1481 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1485 root->commit_root = btrfs_root_node(root);
1487 btrfs_free_path(path);
1491 free_extent_buffer(root->node);
1495 root = ERR_PTR(ret);
1499 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1500 struct btrfs_key *location)
1502 struct btrfs_root *root;
1504 root = btrfs_read_tree_root(tree_root, location);
1508 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1510 btrfs_check_and_init_root_item(&root->root_item);
1516 int btrfs_init_fs_root(struct btrfs_root *root)
1520 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1521 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1523 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1528 btrfs_init_free_ino_ctl(root);
1529 mutex_init(&root->fs_commit_mutex);
1530 spin_lock_init(&root->cache_lock);
1531 init_waitqueue_head(&root->cache_wait);
1533 ret = get_anon_bdev(&root->anon_dev);
1538 kfree(root->free_ino_ctl);
1539 kfree(root->free_ino_pinned);
1543 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1546 struct btrfs_root *root;
1548 spin_lock(&fs_info->fs_roots_radix_lock);
1549 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1550 (unsigned long)root_id);
1551 spin_unlock(&fs_info->fs_roots_radix_lock);
1555 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1556 struct btrfs_root *root)
1560 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1564 spin_lock(&fs_info->fs_roots_radix_lock);
1565 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1566 (unsigned long)root->root_key.objectid,
1570 spin_unlock(&fs_info->fs_roots_radix_lock);
1571 radix_tree_preload_end();
1576 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1577 struct btrfs_key *location,
1580 struct btrfs_root *root;
1583 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1584 return fs_info->tree_root;
1585 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1586 return fs_info->extent_root;
1587 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1588 return fs_info->chunk_root;
1589 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1590 return fs_info->dev_root;
1591 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1592 return fs_info->csum_root;
1593 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1594 return fs_info->quota_root ? fs_info->quota_root :
1596 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1597 return fs_info->uuid_root ? fs_info->uuid_root :
1600 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1602 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1603 return ERR_PTR(-ENOENT);
1607 root = btrfs_read_fs_root(fs_info->tree_root, location);
1611 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1616 ret = btrfs_init_fs_root(root);
1620 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1624 root->orphan_item_inserted = 1;
1626 ret = btrfs_insert_fs_root(fs_info, root);
1628 if (ret == -EEXIST) {
1637 return ERR_PTR(ret);
1640 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1642 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1644 struct btrfs_device *device;
1645 struct backing_dev_info *bdi;
1648 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1651 bdi = blk_get_backing_dev_info(device->bdev);
1652 if (bdi && bdi_congested(bdi, bdi_bits)) {
1662 * If this fails, caller must call bdi_destroy() to get rid of the
1665 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1669 bdi->capabilities = BDI_CAP_MAP_COPY;
1670 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1674 bdi->ra_pages = default_backing_dev_info.ra_pages;
1675 bdi->congested_fn = btrfs_congested_fn;
1676 bdi->congested_data = info;
1681 * called by the kthread helper functions to finally call the bio end_io
1682 * functions. This is where read checksum verification actually happens
1684 static void end_workqueue_fn(struct btrfs_work *work)
1687 struct end_io_wq *end_io_wq;
1688 struct btrfs_fs_info *fs_info;
1691 end_io_wq = container_of(work, struct end_io_wq, work);
1692 bio = end_io_wq->bio;
1693 fs_info = end_io_wq->info;
1695 error = end_io_wq->error;
1696 bio->bi_private = end_io_wq->private;
1697 bio->bi_end_io = end_io_wq->end_io;
1699 bio_endio(bio, error);
1702 static int cleaner_kthread(void *arg)
1704 struct btrfs_root *root = arg;
1710 /* Make the cleaner go to sleep early. */
1711 if (btrfs_need_cleaner_sleep(root))
1714 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1718 * Avoid the problem that we change the status of the fs
1719 * during the above check and trylock.
1721 if (btrfs_need_cleaner_sleep(root)) {
1722 mutex_unlock(&root->fs_info->cleaner_mutex);
1726 btrfs_run_delayed_iputs(root);
1727 again = btrfs_clean_one_deleted_snapshot(root);
1728 mutex_unlock(&root->fs_info->cleaner_mutex);
1731 * The defragger has dealt with the R/O remount and umount,
1732 * needn't do anything special here.
1734 btrfs_run_defrag_inodes(root->fs_info);
1736 if (!try_to_freeze() && !again) {
1737 set_current_state(TASK_INTERRUPTIBLE);
1738 if (!kthread_should_stop())
1740 __set_current_state(TASK_RUNNING);
1742 } while (!kthread_should_stop());
1746 static int transaction_kthread(void *arg)
1748 struct btrfs_root *root = arg;
1749 struct btrfs_trans_handle *trans;
1750 struct btrfs_transaction *cur;
1753 unsigned long delay;
1757 cannot_commit = false;
1758 delay = HZ * root->fs_info->commit_interval;
1759 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1761 spin_lock(&root->fs_info->trans_lock);
1762 cur = root->fs_info->running_transaction;
1764 spin_unlock(&root->fs_info->trans_lock);
1768 now = get_seconds();
1769 if (cur->state < TRANS_STATE_BLOCKED &&
1770 (now < cur->start_time ||
1771 now - cur->start_time < root->fs_info->commit_interval)) {
1772 spin_unlock(&root->fs_info->trans_lock);
1776 transid = cur->transid;
1777 spin_unlock(&root->fs_info->trans_lock);
1779 /* If the file system is aborted, this will always fail. */
1780 trans = btrfs_attach_transaction(root);
1781 if (IS_ERR(trans)) {
1782 if (PTR_ERR(trans) != -ENOENT)
1783 cannot_commit = true;
1786 if (transid == trans->transid) {
1787 btrfs_commit_transaction(trans, root);
1789 btrfs_end_transaction(trans, root);
1792 wake_up_process(root->fs_info->cleaner_kthread);
1793 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1795 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1796 &root->fs_info->fs_state)))
1797 btrfs_cleanup_transaction(root);
1798 if (!try_to_freeze()) {
1799 set_current_state(TASK_INTERRUPTIBLE);
1800 if (!kthread_should_stop() &&
1801 (!btrfs_transaction_blocked(root->fs_info) ||
1803 schedule_timeout(delay);
1804 __set_current_state(TASK_RUNNING);
1806 } while (!kthread_should_stop());
1811 * this will find the highest generation in the array of
1812 * root backups. The index of the highest array is returned,
1813 * or -1 if we can't find anything.
1815 * We check to make sure the array is valid by comparing the
1816 * generation of the latest root in the array with the generation
1817 * in the super block. If they don't match we pitch it.
1819 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1822 int newest_index = -1;
1823 struct btrfs_root_backup *root_backup;
1826 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1827 root_backup = info->super_copy->super_roots + i;
1828 cur = btrfs_backup_tree_root_gen(root_backup);
1829 if (cur == newest_gen)
1833 /* check to see if we actually wrapped around */
1834 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1835 root_backup = info->super_copy->super_roots;
1836 cur = btrfs_backup_tree_root_gen(root_backup);
1837 if (cur == newest_gen)
1840 return newest_index;
1845 * find the oldest backup so we know where to store new entries
1846 * in the backup array. This will set the backup_root_index
1847 * field in the fs_info struct
1849 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1852 int newest_index = -1;
1854 newest_index = find_newest_super_backup(info, newest_gen);
1855 /* if there was garbage in there, just move along */
1856 if (newest_index == -1) {
1857 info->backup_root_index = 0;
1859 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1864 * copy all the root pointers into the super backup array.
1865 * this will bump the backup pointer by one when it is
1868 static void backup_super_roots(struct btrfs_fs_info *info)
1871 struct btrfs_root_backup *root_backup;
1874 next_backup = info->backup_root_index;
1875 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1876 BTRFS_NUM_BACKUP_ROOTS;
1879 * just overwrite the last backup if we're at the same generation
1880 * this happens only at umount
1882 root_backup = info->super_for_commit->super_roots + last_backup;
1883 if (btrfs_backup_tree_root_gen(root_backup) ==
1884 btrfs_header_generation(info->tree_root->node))
1885 next_backup = last_backup;
1887 root_backup = info->super_for_commit->super_roots + next_backup;
1890 * make sure all of our padding and empty slots get zero filled
1891 * regardless of which ones we use today
1893 memset(root_backup, 0, sizeof(*root_backup));
1895 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1897 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1898 btrfs_set_backup_tree_root_gen(root_backup,
1899 btrfs_header_generation(info->tree_root->node));
1901 btrfs_set_backup_tree_root_level(root_backup,
1902 btrfs_header_level(info->tree_root->node));
1904 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1905 btrfs_set_backup_chunk_root_gen(root_backup,
1906 btrfs_header_generation(info->chunk_root->node));
1907 btrfs_set_backup_chunk_root_level(root_backup,
1908 btrfs_header_level(info->chunk_root->node));
1910 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1911 btrfs_set_backup_extent_root_gen(root_backup,
1912 btrfs_header_generation(info->extent_root->node));
1913 btrfs_set_backup_extent_root_level(root_backup,
1914 btrfs_header_level(info->extent_root->node));
1917 * we might commit during log recovery, which happens before we set
1918 * the fs_root. Make sure it is valid before we fill it in.
1920 if (info->fs_root && info->fs_root->node) {
1921 btrfs_set_backup_fs_root(root_backup,
1922 info->fs_root->node->start);
1923 btrfs_set_backup_fs_root_gen(root_backup,
1924 btrfs_header_generation(info->fs_root->node));
1925 btrfs_set_backup_fs_root_level(root_backup,
1926 btrfs_header_level(info->fs_root->node));
1929 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1930 btrfs_set_backup_dev_root_gen(root_backup,
1931 btrfs_header_generation(info->dev_root->node));
1932 btrfs_set_backup_dev_root_level(root_backup,
1933 btrfs_header_level(info->dev_root->node));
1935 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1936 btrfs_set_backup_csum_root_gen(root_backup,
1937 btrfs_header_generation(info->csum_root->node));
1938 btrfs_set_backup_csum_root_level(root_backup,
1939 btrfs_header_level(info->csum_root->node));
1941 btrfs_set_backup_total_bytes(root_backup,
1942 btrfs_super_total_bytes(info->super_copy));
1943 btrfs_set_backup_bytes_used(root_backup,
1944 btrfs_super_bytes_used(info->super_copy));
1945 btrfs_set_backup_num_devices(root_backup,
1946 btrfs_super_num_devices(info->super_copy));
1949 * if we don't copy this out to the super_copy, it won't get remembered
1950 * for the next commit
1952 memcpy(&info->super_copy->super_roots,
1953 &info->super_for_commit->super_roots,
1954 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1958 * this copies info out of the root backup array and back into
1959 * the in-memory super block. It is meant to help iterate through
1960 * the array, so you send it the number of backups you've already
1961 * tried and the last backup index you used.
1963 * this returns -1 when it has tried all the backups
1965 static noinline int next_root_backup(struct btrfs_fs_info *info,
1966 struct btrfs_super_block *super,
1967 int *num_backups_tried, int *backup_index)
1969 struct btrfs_root_backup *root_backup;
1970 int newest = *backup_index;
1972 if (*num_backups_tried == 0) {
1973 u64 gen = btrfs_super_generation(super);
1975 newest = find_newest_super_backup(info, gen);
1979 *backup_index = newest;
1980 *num_backups_tried = 1;
1981 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1982 /* we've tried all the backups, all done */
1985 /* jump to the next oldest backup */
1986 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1987 BTRFS_NUM_BACKUP_ROOTS;
1988 *backup_index = newest;
1989 *num_backups_tried += 1;
1991 root_backup = super->super_roots + newest;
1993 btrfs_set_super_generation(super,
1994 btrfs_backup_tree_root_gen(root_backup));
1995 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1996 btrfs_set_super_root_level(super,
1997 btrfs_backup_tree_root_level(root_backup));
1998 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2001 * fixme: the total bytes and num_devices need to match or we should
2004 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2005 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2009 /* helper to cleanup workers */
2010 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2012 btrfs_stop_workers(&fs_info->generic_worker);
2013 btrfs_stop_workers(&fs_info->fixup_workers);
2014 btrfs_stop_workers(&fs_info->delalloc_workers);
2015 btrfs_stop_workers(&fs_info->workers);
2016 btrfs_stop_workers(&fs_info->endio_workers);
2017 btrfs_stop_workers(&fs_info->endio_meta_workers);
2018 btrfs_stop_workers(&fs_info->endio_raid56_workers);
2019 btrfs_stop_workers(&fs_info->rmw_workers);
2020 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2021 btrfs_stop_workers(&fs_info->endio_write_workers);
2022 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2023 btrfs_stop_workers(&fs_info->submit_workers);
2024 btrfs_stop_workers(&fs_info->delayed_workers);
2025 btrfs_stop_workers(&fs_info->caching_workers);
2026 btrfs_stop_workers(&fs_info->readahead_workers);
2027 btrfs_stop_workers(&fs_info->flush_workers);
2028 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
2031 static void free_root_extent_buffers(struct btrfs_root *root)
2034 free_extent_buffer(root->node);
2035 free_extent_buffer(root->commit_root);
2037 root->commit_root = NULL;
2041 /* helper to cleanup tree roots */
2042 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2044 free_root_extent_buffers(info->tree_root);
2046 free_root_extent_buffers(info->dev_root);
2047 free_root_extent_buffers(info->extent_root);
2048 free_root_extent_buffers(info->csum_root);
2049 free_root_extent_buffers(info->quota_root);
2050 free_root_extent_buffers(info->uuid_root);
2052 free_root_extent_buffers(info->chunk_root);
2055 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2058 struct btrfs_root *gang[8];
2061 while (!list_empty(&fs_info->dead_roots)) {
2062 gang[0] = list_entry(fs_info->dead_roots.next,
2063 struct btrfs_root, root_list);
2064 list_del(&gang[0]->root_list);
2066 if (gang[0]->in_radix) {
2067 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2069 free_extent_buffer(gang[0]->node);
2070 free_extent_buffer(gang[0]->commit_root);
2071 btrfs_put_fs_root(gang[0]);
2076 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2081 for (i = 0; i < ret; i++)
2082 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2086 int open_ctree(struct super_block *sb,
2087 struct btrfs_fs_devices *fs_devices,
2097 struct btrfs_key location;
2098 struct buffer_head *bh;
2099 struct btrfs_super_block *disk_super;
2100 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2101 struct btrfs_root *tree_root;
2102 struct btrfs_root *extent_root;
2103 struct btrfs_root *csum_root;
2104 struct btrfs_root *chunk_root;
2105 struct btrfs_root *dev_root;
2106 struct btrfs_root *quota_root;
2107 struct btrfs_root *uuid_root;
2108 struct btrfs_root *log_tree_root;
2111 int num_backups_tried = 0;
2112 int backup_index = 0;
2113 bool create_uuid_tree;
2114 bool check_uuid_tree;
2116 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2117 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2118 if (!tree_root || !chunk_root) {
2123 ret = init_srcu_struct(&fs_info->subvol_srcu);
2129 ret = setup_bdi(fs_info, &fs_info->bdi);
2135 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2140 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2141 (1 + ilog2(nr_cpu_ids));
2143 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2146 goto fail_dirty_metadata_bytes;
2149 fs_info->btree_inode = new_inode(sb);
2150 if (!fs_info->btree_inode) {
2152 goto fail_delalloc_bytes;
2155 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2157 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2158 INIT_LIST_HEAD(&fs_info->trans_list);
2159 INIT_LIST_HEAD(&fs_info->dead_roots);
2160 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2161 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2162 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2163 spin_lock_init(&fs_info->delalloc_root_lock);
2164 spin_lock_init(&fs_info->trans_lock);
2165 spin_lock_init(&fs_info->fs_roots_radix_lock);
2166 spin_lock_init(&fs_info->delayed_iput_lock);
2167 spin_lock_init(&fs_info->defrag_inodes_lock);
2168 spin_lock_init(&fs_info->free_chunk_lock);
2169 spin_lock_init(&fs_info->tree_mod_seq_lock);
2170 spin_lock_init(&fs_info->super_lock);
2171 rwlock_init(&fs_info->tree_mod_log_lock);
2172 mutex_init(&fs_info->reloc_mutex);
2173 seqlock_init(&fs_info->profiles_lock);
2175 init_completion(&fs_info->kobj_unregister);
2176 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2177 INIT_LIST_HEAD(&fs_info->space_info);
2178 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2179 btrfs_mapping_init(&fs_info->mapping_tree);
2180 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2181 BTRFS_BLOCK_RSV_GLOBAL);
2182 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2183 BTRFS_BLOCK_RSV_DELALLOC);
2184 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2185 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2186 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2187 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2188 BTRFS_BLOCK_RSV_DELOPS);
2189 atomic_set(&fs_info->nr_async_submits, 0);
2190 atomic_set(&fs_info->async_delalloc_pages, 0);
2191 atomic_set(&fs_info->async_submit_draining, 0);
2192 atomic_set(&fs_info->nr_async_bios, 0);
2193 atomic_set(&fs_info->defrag_running, 0);
2194 atomic64_set(&fs_info->tree_mod_seq, 0);
2196 fs_info->max_inline = 8192 * 1024;
2197 fs_info->metadata_ratio = 0;
2198 fs_info->defrag_inodes = RB_ROOT;
2199 fs_info->free_chunk_space = 0;
2200 fs_info->tree_mod_log = RB_ROOT;
2201 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2203 /* readahead state */
2204 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2205 spin_lock_init(&fs_info->reada_lock);
2207 fs_info->thread_pool_size = min_t(unsigned long,
2208 num_online_cpus() + 2, 8);
2210 INIT_LIST_HEAD(&fs_info->ordered_roots);
2211 spin_lock_init(&fs_info->ordered_root_lock);
2212 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2214 if (!fs_info->delayed_root) {
2218 btrfs_init_delayed_root(fs_info->delayed_root);
2220 mutex_init(&fs_info->scrub_lock);
2221 atomic_set(&fs_info->scrubs_running, 0);
2222 atomic_set(&fs_info->scrub_pause_req, 0);
2223 atomic_set(&fs_info->scrubs_paused, 0);
2224 atomic_set(&fs_info->scrub_cancel_req, 0);
2225 init_waitqueue_head(&fs_info->scrub_pause_wait);
2226 fs_info->scrub_workers_refcnt = 0;
2227 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2228 fs_info->check_integrity_print_mask = 0;
2231 spin_lock_init(&fs_info->balance_lock);
2232 mutex_init(&fs_info->balance_mutex);
2233 atomic_set(&fs_info->balance_running, 0);
2234 atomic_set(&fs_info->balance_pause_req, 0);
2235 atomic_set(&fs_info->balance_cancel_req, 0);
2236 fs_info->balance_ctl = NULL;
2237 init_waitqueue_head(&fs_info->balance_wait_q);
2239 sb->s_blocksize = 4096;
2240 sb->s_blocksize_bits = blksize_bits(4096);
2241 sb->s_bdi = &fs_info->bdi;
2243 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2244 set_nlink(fs_info->btree_inode, 1);
2246 * we set the i_size on the btree inode to the max possible int.
2247 * the real end of the address space is determined by all of
2248 * the devices in the system
2250 fs_info->btree_inode->i_size = OFFSET_MAX;
2251 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2252 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2254 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2255 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2256 fs_info->btree_inode->i_mapping);
2257 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2258 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2260 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2262 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2263 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2264 sizeof(struct btrfs_key));
2265 set_bit(BTRFS_INODE_DUMMY,
2266 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2267 btrfs_insert_inode_hash(fs_info->btree_inode);
2269 spin_lock_init(&fs_info->block_group_cache_lock);
2270 fs_info->block_group_cache_tree = RB_ROOT;
2271 fs_info->first_logical_byte = (u64)-1;
2273 extent_io_tree_init(&fs_info->freed_extents[0],
2274 fs_info->btree_inode->i_mapping);
2275 extent_io_tree_init(&fs_info->freed_extents[1],
2276 fs_info->btree_inode->i_mapping);
2277 fs_info->pinned_extents = &fs_info->freed_extents[0];
2278 fs_info->do_barriers = 1;
2281 mutex_init(&fs_info->ordered_operations_mutex);
2282 mutex_init(&fs_info->ordered_extent_flush_mutex);
2283 mutex_init(&fs_info->tree_log_mutex);
2284 mutex_init(&fs_info->chunk_mutex);
2285 mutex_init(&fs_info->transaction_kthread_mutex);
2286 mutex_init(&fs_info->cleaner_mutex);
2287 mutex_init(&fs_info->volume_mutex);
2288 init_rwsem(&fs_info->extent_commit_sem);
2289 init_rwsem(&fs_info->cleanup_work_sem);
2290 init_rwsem(&fs_info->subvol_sem);
2291 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2292 fs_info->dev_replace.lock_owner = 0;
2293 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2294 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2295 mutex_init(&fs_info->dev_replace.lock_management_lock);
2296 mutex_init(&fs_info->dev_replace.lock);
2298 spin_lock_init(&fs_info->qgroup_lock);
2299 mutex_init(&fs_info->qgroup_ioctl_lock);
2300 fs_info->qgroup_tree = RB_ROOT;
2301 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2302 fs_info->qgroup_seq = 1;
2303 fs_info->quota_enabled = 0;
2304 fs_info->pending_quota_state = 0;
2305 fs_info->qgroup_ulist = NULL;
2306 mutex_init(&fs_info->qgroup_rescan_lock);
2308 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2309 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2311 init_waitqueue_head(&fs_info->transaction_throttle);
2312 init_waitqueue_head(&fs_info->transaction_wait);
2313 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2314 init_waitqueue_head(&fs_info->async_submit_wait);
2316 ret = btrfs_alloc_stripe_hash_table(fs_info);
2322 __setup_root(4096, 4096, 4096, 4096, tree_root,
2323 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2325 invalidate_bdev(fs_devices->latest_bdev);
2328 * Read super block and check the signature bytes only
2330 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2337 * We want to check superblock checksum, the type is stored inside.
2338 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2340 if (btrfs_check_super_csum(bh->b_data)) {
2341 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2347 * super_copy is zeroed at allocation time and we never touch the
2348 * following bytes up to INFO_SIZE, the checksum is calculated from
2349 * the whole block of INFO_SIZE
2351 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2352 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2353 sizeof(*fs_info->super_for_commit));
2356 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2358 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2360 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2365 disk_super = fs_info->super_copy;
2366 if (!btrfs_super_root(disk_super))
2369 /* check FS state, whether FS is broken. */
2370 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2371 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2374 * run through our array of backup supers and setup
2375 * our ring pointer to the oldest one
2377 generation = btrfs_super_generation(disk_super);
2378 find_oldest_super_backup(fs_info, generation);
2381 * In the long term, we'll store the compression type in the super
2382 * block, and it'll be used for per file compression control.
2384 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2386 ret = btrfs_parse_options(tree_root, options);
2392 features = btrfs_super_incompat_flags(disk_super) &
2393 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2395 printk(KERN_ERR "BTRFS: couldn't mount because of "
2396 "unsupported optional features (%Lx).\n",
2402 if (btrfs_super_leafsize(disk_super) !=
2403 btrfs_super_nodesize(disk_super)) {
2404 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2405 "blocksizes don't match. node %d leaf %d\n",
2406 btrfs_super_nodesize(disk_super),
2407 btrfs_super_leafsize(disk_super));
2411 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2412 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2413 "blocksize (%d) was too large\n",
2414 btrfs_super_leafsize(disk_super));
2419 features = btrfs_super_incompat_flags(disk_super);
2420 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2421 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2422 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2424 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2425 printk(KERN_ERR "btrfs: has skinny extents\n");
2428 * flag our filesystem as having big metadata blocks if
2429 * they are bigger than the page size
2431 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2432 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2433 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2434 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2437 nodesize = btrfs_super_nodesize(disk_super);
2438 leafsize = btrfs_super_leafsize(disk_super);
2439 sectorsize = btrfs_super_sectorsize(disk_super);
2440 stripesize = btrfs_super_stripesize(disk_super);
2441 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2442 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2445 * mixed block groups end up with duplicate but slightly offset
2446 * extent buffers for the same range. It leads to corruptions
2448 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2449 (sectorsize != leafsize)) {
2450 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2451 "are not allowed for mixed block groups on %s\n",
2457 * Needn't use the lock because there is no other task which will
2460 btrfs_set_super_incompat_flags(disk_super, features);
2462 features = btrfs_super_compat_ro_flags(disk_super) &
2463 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2464 if (!(sb->s_flags & MS_RDONLY) && features) {
2465 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2466 "unsupported option features (%Lx).\n",
2472 btrfs_init_workers(&fs_info->generic_worker,
2473 "genwork", 1, NULL);
2475 btrfs_init_workers(&fs_info->workers, "worker",
2476 fs_info->thread_pool_size,
2477 &fs_info->generic_worker);
2479 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2480 fs_info->thread_pool_size, NULL);
2482 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2483 fs_info->thread_pool_size, NULL);
2485 btrfs_init_workers(&fs_info->submit_workers, "submit",
2486 min_t(u64, fs_devices->num_devices,
2487 fs_info->thread_pool_size), NULL);
2489 btrfs_init_workers(&fs_info->caching_workers, "cache",
2490 fs_info->thread_pool_size, NULL);
2492 /* a higher idle thresh on the submit workers makes it much more
2493 * likely that bios will be send down in a sane order to the
2496 fs_info->submit_workers.idle_thresh = 64;
2498 fs_info->workers.idle_thresh = 16;
2499 fs_info->workers.ordered = 1;
2501 fs_info->delalloc_workers.idle_thresh = 2;
2502 fs_info->delalloc_workers.ordered = 1;
2504 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2505 &fs_info->generic_worker);
2506 btrfs_init_workers(&fs_info->endio_workers, "endio",
2507 fs_info->thread_pool_size,
2508 &fs_info->generic_worker);
2509 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2510 fs_info->thread_pool_size,
2511 &fs_info->generic_worker);
2512 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2513 "endio-meta-write", fs_info->thread_pool_size,
2514 &fs_info->generic_worker);
2515 btrfs_init_workers(&fs_info->endio_raid56_workers,
2516 "endio-raid56", fs_info->thread_pool_size,
2517 &fs_info->generic_worker);
2518 btrfs_init_workers(&fs_info->rmw_workers,
2519 "rmw", fs_info->thread_pool_size,
2520 &fs_info->generic_worker);
2521 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2522 fs_info->thread_pool_size,
2523 &fs_info->generic_worker);
2524 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2525 1, &fs_info->generic_worker);
2526 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2527 fs_info->thread_pool_size,
2528 &fs_info->generic_worker);
2529 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2530 fs_info->thread_pool_size,
2531 &fs_info->generic_worker);
2532 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2533 &fs_info->generic_worker);
2536 * endios are largely parallel and should have a very
2539 fs_info->endio_workers.idle_thresh = 4;
2540 fs_info->endio_meta_workers.idle_thresh = 4;
2541 fs_info->endio_raid56_workers.idle_thresh = 4;
2542 fs_info->rmw_workers.idle_thresh = 2;
2544 fs_info->endio_write_workers.idle_thresh = 2;
2545 fs_info->endio_meta_write_workers.idle_thresh = 2;
2546 fs_info->readahead_workers.idle_thresh = 2;
2549 * btrfs_start_workers can really only fail because of ENOMEM so just
2550 * return -ENOMEM if any of these fail.
2552 ret = btrfs_start_workers(&fs_info->workers);
2553 ret |= btrfs_start_workers(&fs_info->generic_worker);
2554 ret |= btrfs_start_workers(&fs_info->submit_workers);
2555 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2556 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2557 ret |= btrfs_start_workers(&fs_info->endio_workers);
2558 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2559 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2560 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2561 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2562 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2563 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2564 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2565 ret |= btrfs_start_workers(&fs_info->caching_workers);
2566 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2567 ret |= btrfs_start_workers(&fs_info->flush_workers);
2568 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2571 goto fail_sb_buffer;
2574 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2575 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2576 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2578 tree_root->nodesize = nodesize;
2579 tree_root->leafsize = leafsize;
2580 tree_root->sectorsize = sectorsize;
2581 tree_root->stripesize = stripesize;
2583 sb->s_blocksize = sectorsize;
2584 sb->s_blocksize_bits = blksize_bits(sectorsize);
2586 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2587 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2588 goto fail_sb_buffer;
2591 if (sectorsize != PAGE_SIZE) {
2592 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2593 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2594 goto fail_sb_buffer;
2597 mutex_lock(&fs_info->chunk_mutex);
2598 ret = btrfs_read_sys_array(tree_root);
2599 mutex_unlock(&fs_info->chunk_mutex);
2601 printk(KERN_WARNING "btrfs: failed to read the system "
2602 "array on %s\n", sb->s_id);
2603 goto fail_sb_buffer;
2606 blocksize = btrfs_level_size(tree_root,
2607 btrfs_super_chunk_root_level(disk_super));
2608 generation = btrfs_super_chunk_root_generation(disk_super);
2610 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2611 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2613 chunk_root->node = read_tree_block(chunk_root,
2614 btrfs_super_chunk_root(disk_super),
2615 blocksize, generation);
2616 if (!chunk_root->node ||
2617 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2618 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2620 goto fail_tree_roots;
2622 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2623 chunk_root->commit_root = btrfs_root_node(chunk_root);
2625 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2626 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2628 ret = btrfs_read_chunk_tree(chunk_root);
2630 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2632 goto fail_tree_roots;
2636 * keep the device that is marked to be the target device for the
2637 * dev_replace procedure
2639 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2641 if (!fs_devices->latest_bdev) {
2642 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2644 goto fail_tree_roots;
2648 blocksize = btrfs_level_size(tree_root,
2649 btrfs_super_root_level(disk_super));
2650 generation = btrfs_super_generation(disk_super);
2652 tree_root->node = read_tree_block(tree_root,
2653 btrfs_super_root(disk_super),
2654 blocksize, generation);
2655 if (!tree_root->node ||
2656 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2657 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2660 goto recovery_tree_root;
2663 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2664 tree_root->commit_root = btrfs_root_node(tree_root);
2665 btrfs_set_root_refs(&tree_root->root_item, 1);
2667 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2668 location.type = BTRFS_ROOT_ITEM_KEY;
2669 location.offset = 0;
2671 extent_root = btrfs_read_tree_root(tree_root, &location);
2672 if (IS_ERR(extent_root)) {
2673 ret = PTR_ERR(extent_root);
2674 goto recovery_tree_root;
2676 extent_root->track_dirty = 1;
2677 fs_info->extent_root = extent_root;
2679 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2680 dev_root = btrfs_read_tree_root(tree_root, &location);
2681 if (IS_ERR(dev_root)) {
2682 ret = PTR_ERR(dev_root);
2683 goto recovery_tree_root;
2685 dev_root->track_dirty = 1;
2686 fs_info->dev_root = dev_root;
2687 btrfs_init_devices_late(fs_info);
2689 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2690 csum_root = btrfs_read_tree_root(tree_root, &location);
2691 if (IS_ERR(csum_root)) {
2692 ret = PTR_ERR(csum_root);
2693 goto recovery_tree_root;
2695 csum_root->track_dirty = 1;
2696 fs_info->csum_root = csum_root;
2698 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2699 quota_root = btrfs_read_tree_root(tree_root, &location);
2700 if (!IS_ERR(quota_root)) {
2701 quota_root->track_dirty = 1;
2702 fs_info->quota_enabled = 1;
2703 fs_info->pending_quota_state = 1;
2704 fs_info->quota_root = quota_root;
2707 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2708 uuid_root = btrfs_read_tree_root(tree_root, &location);
2709 if (IS_ERR(uuid_root)) {
2710 ret = PTR_ERR(uuid_root);
2712 goto recovery_tree_root;
2713 create_uuid_tree = true;
2714 check_uuid_tree = false;
2716 uuid_root->track_dirty = 1;
2717 fs_info->uuid_root = uuid_root;
2718 create_uuid_tree = false;
2720 generation != btrfs_super_uuid_tree_generation(disk_super);
2723 fs_info->generation = generation;
2724 fs_info->last_trans_committed = generation;
2726 ret = btrfs_recover_balance(fs_info);
2728 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2729 goto fail_block_groups;
2732 ret = btrfs_init_dev_stats(fs_info);
2734 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2736 goto fail_block_groups;
2739 ret = btrfs_init_dev_replace(fs_info);
2741 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2742 goto fail_block_groups;
2745 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2747 ret = btrfs_sysfs_add_one(fs_info);
2749 pr_err("btrfs: failed to init sysfs interface: %d\n", ret);
2750 goto fail_block_groups;
2753 ret = btrfs_init_space_info(fs_info);
2755 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2756 goto fail_block_groups;
2759 ret = btrfs_read_block_groups(extent_root);
2761 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2762 goto fail_block_groups;
2764 fs_info->num_tolerated_disk_barrier_failures =
2765 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2766 if (fs_info->fs_devices->missing_devices >
2767 fs_info->num_tolerated_disk_barrier_failures &&
2768 !(sb->s_flags & MS_RDONLY)) {
2770 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2771 goto fail_block_groups;
2774 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2776 if (IS_ERR(fs_info->cleaner_kthread))
2777 goto fail_block_groups;
2779 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2781 "btrfs-transaction");
2782 if (IS_ERR(fs_info->transaction_kthread))
2785 if (!btrfs_test_opt(tree_root, SSD) &&
2786 !btrfs_test_opt(tree_root, NOSSD) &&
2787 !fs_info->fs_devices->rotating) {
2788 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2790 btrfs_set_opt(fs_info->mount_opt, SSD);
2793 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2794 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2795 ret = btrfsic_mount(tree_root, fs_devices,
2796 btrfs_test_opt(tree_root,
2797 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2799 fs_info->check_integrity_print_mask);
2801 printk(KERN_WARNING "btrfs: failed to initialize"
2802 " integrity check module %s\n", sb->s_id);
2805 ret = btrfs_read_qgroup_config(fs_info);
2807 goto fail_trans_kthread;
2809 /* do not make disk changes in broken FS */
2810 if (btrfs_super_log_root(disk_super) != 0) {
2811 u64 bytenr = btrfs_super_log_root(disk_super);
2813 if (fs_devices->rw_devices == 0) {
2814 printk(KERN_WARNING "Btrfs log replay required "
2820 btrfs_level_size(tree_root,
2821 btrfs_super_log_root_level(disk_super));
2823 log_tree_root = btrfs_alloc_root(fs_info);
2824 if (!log_tree_root) {
2829 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2830 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2832 log_tree_root->node = read_tree_block(tree_root, bytenr,
2835 if (!log_tree_root->node ||
2836 !extent_buffer_uptodate(log_tree_root->node)) {
2837 printk(KERN_ERR "btrfs: failed to read log tree\n");
2838 free_extent_buffer(log_tree_root->node);
2839 kfree(log_tree_root);
2840 goto fail_trans_kthread;
2842 /* returns with log_tree_root freed on success */
2843 ret = btrfs_recover_log_trees(log_tree_root);
2845 btrfs_error(tree_root->fs_info, ret,
2846 "Failed to recover log tree");
2847 free_extent_buffer(log_tree_root->node);
2848 kfree(log_tree_root);
2849 goto fail_trans_kthread;
2852 if (sb->s_flags & MS_RDONLY) {
2853 ret = btrfs_commit_super(tree_root);
2855 goto fail_trans_kthread;
2859 ret = btrfs_find_orphan_roots(tree_root);
2861 goto fail_trans_kthread;
2863 if (!(sb->s_flags & MS_RDONLY)) {
2864 ret = btrfs_cleanup_fs_roots(fs_info);
2866 goto fail_trans_kthread;
2868 ret = btrfs_recover_relocation(tree_root);
2871 "btrfs: failed to recover relocation\n");
2877 location.objectid = BTRFS_FS_TREE_OBJECTID;
2878 location.type = BTRFS_ROOT_ITEM_KEY;
2879 location.offset = 0;
2881 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2882 if (IS_ERR(fs_info->fs_root)) {
2883 err = PTR_ERR(fs_info->fs_root);
2887 if (sb->s_flags & MS_RDONLY)
2890 down_read(&fs_info->cleanup_work_sem);
2891 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2892 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2893 up_read(&fs_info->cleanup_work_sem);
2894 close_ctree(tree_root);
2897 up_read(&fs_info->cleanup_work_sem);
2899 ret = btrfs_resume_balance_async(fs_info);
2901 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2902 close_ctree(tree_root);
2906 ret = btrfs_resume_dev_replace_async(fs_info);
2908 pr_warn("btrfs: failed to resume dev_replace\n");
2909 close_ctree(tree_root);
2913 btrfs_qgroup_rescan_resume(fs_info);
2915 if (create_uuid_tree) {
2916 pr_info("btrfs: creating UUID tree\n");
2917 ret = btrfs_create_uuid_tree(fs_info);
2919 pr_warn("btrfs: failed to create the UUID tree %d\n",
2921 close_ctree(tree_root);
2924 } else if (check_uuid_tree ||
2925 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2926 pr_info("btrfs: checking UUID tree\n");
2927 ret = btrfs_check_uuid_tree(fs_info);
2929 pr_warn("btrfs: failed to check the UUID tree %d\n",
2931 close_ctree(tree_root);
2935 fs_info->update_uuid_tree_gen = 1;
2941 btrfs_free_qgroup_config(fs_info);
2943 kthread_stop(fs_info->transaction_kthread);
2944 btrfs_cleanup_transaction(fs_info->tree_root);
2945 del_fs_roots(fs_info);
2947 kthread_stop(fs_info->cleaner_kthread);
2950 * make sure we're done with the btree inode before we stop our
2953 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2956 btrfs_put_block_group_cache(fs_info);
2957 btrfs_free_block_groups(fs_info);
2960 free_root_pointers(fs_info, 1);
2961 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2964 btrfs_stop_all_workers(fs_info);
2967 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2969 iput(fs_info->btree_inode);
2970 fail_delalloc_bytes:
2971 percpu_counter_destroy(&fs_info->delalloc_bytes);
2972 fail_dirty_metadata_bytes:
2973 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2975 bdi_destroy(&fs_info->bdi);
2977 cleanup_srcu_struct(&fs_info->subvol_srcu);
2979 btrfs_free_stripe_hash_table(fs_info);
2980 btrfs_close_devices(fs_info->fs_devices);
2984 if (!btrfs_test_opt(tree_root, RECOVERY))
2985 goto fail_tree_roots;
2987 free_root_pointers(fs_info, 0);
2989 /* don't use the log in recovery mode, it won't be valid */
2990 btrfs_set_super_log_root(disk_super, 0);
2992 /* we can't trust the free space cache either */
2993 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2995 ret = next_root_backup(fs_info, fs_info->super_copy,
2996 &num_backups_tried, &backup_index);
2998 goto fail_block_groups;
2999 goto retry_root_backup;
3002 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3005 set_buffer_uptodate(bh);
3007 struct btrfs_device *device = (struct btrfs_device *)
3010 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
3011 "I/O error on %s\n",
3012 rcu_str_deref(device->name));
3013 /* note, we dont' set_buffer_write_io_error because we have
3014 * our own ways of dealing with the IO errors
3016 clear_buffer_uptodate(bh);
3017 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3023 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3025 struct buffer_head *bh;
3026 struct buffer_head *latest = NULL;
3027 struct btrfs_super_block *super;
3032 /* we would like to check all the supers, but that would make
3033 * a btrfs mount succeed after a mkfs from a different FS.
3034 * So, we need to add a special mount option to scan for
3035 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3037 for (i = 0; i < 1; i++) {
3038 bytenr = btrfs_sb_offset(i);
3039 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3040 i_size_read(bdev->bd_inode))
3042 bh = __bread(bdev, bytenr / 4096,
3043 BTRFS_SUPER_INFO_SIZE);
3047 super = (struct btrfs_super_block *)bh->b_data;
3048 if (btrfs_super_bytenr(super) != bytenr ||
3049 btrfs_super_magic(super) != BTRFS_MAGIC) {
3054 if (!latest || btrfs_super_generation(super) > transid) {
3057 transid = btrfs_super_generation(super);
3066 * this should be called twice, once with wait == 0 and
3067 * once with wait == 1. When wait == 0 is done, all the buffer heads
3068 * we write are pinned.
3070 * They are released when wait == 1 is done.
3071 * max_mirrors must be the same for both runs, and it indicates how
3072 * many supers on this one device should be written.
3074 * max_mirrors == 0 means to write them all.
3076 static int write_dev_supers(struct btrfs_device *device,
3077 struct btrfs_super_block *sb,
3078 int do_barriers, int wait, int max_mirrors)
3080 struct buffer_head *bh;
3087 if (max_mirrors == 0)
3088 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3090 for (i = 0; i < max_mirrors; i++) {
3091 bytenr = btrfs_sb_offset(i);
3092 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3096 bh = __find_get_block(device->bdev, bytenr / 4096,
3097 BTRFS_SUPER_INFO_SIZE);
3103 if (!buffer_uptodate(bh))
3106 /* drop our reference */
3109 /* drop the reference from the wait == 0 run */
3113 btrfs_set_super_bytenr(sb, bytenr);
3116 crc = btrfs_csum_data((char *)sb +
3117 BTRFS_CSUM_SIZE, crc,
3118 BTRFS_SUPER_INFO_SIZE -
3120 btrfs_csum_final(crc, sb->csum);
3123 * one reference for us, and we leave it for the
3126 bh = __getblk(device->bdev, bytenr / 4096,
3127 BTRFS_SUPER_INFO_SIZE);
3129 printk(KERN_ERR "btrfs: couldn't get super "
3130 "buffer head for bytenr %Lu\n", bytenr);
3135 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3137 /* one reference for submit_bh */
3140 set_buffer_uptodate(bh);
3142 bh->b_end_io = btrfs_end_buffer_write_sync;
3143 bh->b_private = device;
3147 * we fua the first super. The others we allow
3150 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3154 return errors < i ? 0 : -1;
3158 * endio for the write_dev_flush, this will wake anyone waiting
3159 * for the barrier when it is done
3161 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3164 if (err == -EOPNOTSUPP)
3165 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3166 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3168 if (bio->bi_private)
3169 complete(bio->bi_private);
3174 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3175 * sent down. With wait == 1, it waits for the previous flush.
3177 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3180 static int write_dev_flush(struct btrfs_device *device, int wait)
3185 if (device->nobarriers)
3189 bio = device->flush_bio;
3193 wait_for_completion(&device->flush_wait);
3195 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3196 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3197 rcu_str_deref(device->name));
3198 device->nobarriers = 1;
3199 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3201 btrfs_dev_stat_inc_and_print(device,
3202 BTRFS_DEV_STAT_FLUSH_ERRS);
3205 /* drop the reference from the wait == 0 run */
3207 device->flush_bio = NULL;
3213 * one reference for us, and we leave it for the
3216 device->flush_bio = NULL;
3217 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3221 bio->bi_end_io = btrfs_end_empty_barrier;
3222 bio->bi_bdev = device->bdev;
3223 init_completion(&device->flush_wait);
3224 bio->bi_private = &device->flush_wait;
3225 device->flush_bio = bio;
3228 btrfsic_submit_bio(WRITE_FLUSH, bio);
3234 * send an empty flush down to each device in parallel,
3235 * then wait for them
3237 static int barrier_all_devices(struct btrfs_fs_info *info)
3239 struct list_head *head;
3240 struct btrfs_device *dev;
3241 int errors_send = 0;
3242 int errors_wait = 0;
3245 /* send down all the barriers */
3246 head = &info->fs_devices->devices;
3247 list_for_each_entry_rcu(dev, head, dev_list) {
3252 if (!dev->in_fs_metadata || !dev->writeable)
3255 ret = write_dev_flush(dev, 0);
3260 /* wait for all the barriers */
3261 list_for_each_entry_rcu(dev, head, dev_list) {
3266 if (!dev->in_fs_metadata || !dev->writeable)
3269 ret = write_dev_flush(dev, 1);
3273 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3274 errors_wait > info->num_tolerated_disk_barrier_failures)
3279 int btrfs_calc_num_tolerated_disk_barrier_failures(
3280 struct btrfs_fs_info *fs_info)
3282 struct btrfs_ioctl_space_info space;
3283 struct btrfs_space_info *sinfo;
3284 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3285 BTRFS_BLOCK_GROUP_SYSTEM,
3286 BTRFS_BLOCK_GROUP_METADATA,
3287 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3291 int num_tolerated_disk_barrier_failures =
3292 (int)fs_info->fs_devices->num_devices;
3294 for (i = 0; i < num_types; i++) {
3295 struct btrfs_space_info *tmp;
3299 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3300 if (tmp->flags == types[i]) {
3310 down_read(&sinfo->groups_sem);
3311 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3312 if (!list_empty(&sinfo->block_groups[c])) {
3315 btrfs_get_block_group_info(
3316 &sinfo->block_groups[c], &space);
3317 if (space.total_bytes == 0 ||
3318 space.used_bytes == 0)
3320 flags = space.flags;
3323 * 0: if dup, single or RAID0 is configured for
3324 * any of metadata, system or data, else
3325 * 1: if RAID5 is configured, or if RAID1 or
3326 * RAID10 is configured and only two mirrors
3328 * 2: if RAID6 is configured, else
3329 * num_mirrors - 1: if RAID1 or RAID10 is
3330 * configured and more than
3331 * 2 mirrors are used.
3333 if (num_tolerated_disk_barrier_failures > 0 &&
3334 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3335 BTRFS_BLOCK_GROUP_RAID0)) ||
3336 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3338 num_tolerated_disk_barrier_failures = 0;
3339 else if (num_tolerated_disk_barrier_failures > 1) {
3340 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3341 BTRFS_BLOCK_GROUP_RAID5 |
3342 BTRFS_BLOCK_GROUP_RAID10)) {
3343 num_tolerated_disk_barrier_failures = 1;
3345 BTRFS_BLOCK_GROUP_RAID6) {
3346 num_tolerated_disk_barrier_failures = 2;
3351 up_read(&sinfo->groups_sem);
3354 return num_tolerated_disk_barrier_failures;
3357 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3359 struct list_head *head;
3360 struct btrfs_device *dev;
3361 struct btrfs_super_block *sb;
3362 struct btrfs_dev_item *dev_item;
3366 int total_errors = 0;
3369 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3370 backup_super_roots(root->fs_info);
3372 sb = root->fs_info->super_for_commit;
3373 dev_item = &sb->dev_item;
3375 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3376 head = &root->fs_info->fs_devices->devices;
3377 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3380 ret = barrier_all_devices(root->fs_info);
3383 &root->fs_info->fs_devices->device_list_mutex);
3384 btrfs_error(root->fs_info, ret,
3385 "errors while submitting device barriers.");
3390 list_for_each_entry_rcu(dev, head, dev_list) {
3395 if (!dev->in_fs_metadata || !dev->writeable)
3398 btrfs_set_stack_device_generation(dev_item, 0);
3399 btrfs_set_stack_device_type(dev_item, dev->type);
3400 btrfs_set_stack_device_id(dev_item, dev->devid);
3401 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3402 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3403 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3404 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3405 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3406 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3407 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3409 flags = btrfs_super_flags(sb);
3410 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3412 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3416 if (total_errors > max_errors) {
3417 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3419 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3421 /* FUA is masked off if unsupported and can't be the reason */
3422 btrfs_error(root->fs_info, -EIO,
3423 "%d errors while writing supers", total_errors);
3428 list_for_each_entry_rcu(dev, head, dev_list) {
3431 if (!dev->in_fs_metadata || !dev->writeable)
3434 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3438 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3439 if (total_errors > max_errors) {
3440 btrfs_error(root->fs_info, -EIO,
3441 "%d errors while writing supers", total_errors);
3447 int write_ctree_super(struct btrfs_trans_handle *trans,
3448 struct btrfs_root *root, int max_mirrors)
3450 return write_all_supers(root, max_mirrors);
3453 /* Drop a fs root from the radix tree and free it. */
3454 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3455 struct btrfs_root *root)
3457 spin_lock(&fs_info->fs_roots_radix_lock);
3458 radix_tree_delete(&fs_info->fs_roots_radix,
3459 (unsigned long)root->root_key.objectid);
3460 spin_unlock(&fs_info->fs_roots_radix_lock);
3462 if (btrfs_root_refs(&root->root_item) == 0)
3463 synchronize_srcu(&fs_info->subvol_srcu);
3465 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3466 btrfs_free_log(NULL, root);
3467 btrfs_free_log_root_tree(NULL, fs_info);
3470 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3471 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3475 static void free_fs_root(struct btrfs_root *root)
3477 iput(root->cache_inode);
3478 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3479 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3480 root->orphan_block_rsv = NULL;
3482 free_anon_bdev(root->anon_dev);
3483 free_extent_buffer(root->node);
3484 free_extent_buffer(root->commit_root);
3485 kfree(root->free_ino_ctl);
3486 kfree(root->free_ino_pinned);
3488 btrfs_put_fs_root(root);
3491 void btrfs_free_fs_root(struct btrfs_root *root)
3496 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3498 u64 root_objectid = 0;
3499 struct btrfs_root *gang[8];
3504 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3505 (void **)gang, root_objectid,
3510 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3511 for (i = 0; i < ret; i++) {
3514 root_objectid = gang[i]->root_key.objectid;
3515 err = btrfs_orphan_cleanup(gang[i]);
3524 int btrfs_commit_super(struct btrfs_root *root)
3526 struct btrfs_trans_handle *trans;
3528 mutex_lock(&root->fs_info->cleaner_mutex);
3529 btrfs_run_delayed_iputs(root);
3530 mutex_unlock(&root->fs_info->cleaner_mutex);
3531 wake_up_process(root->fs_info->cleaner_kthread);
3533 /* wait until ongoing cleanup work done */
3534 down_write(&root->fs_info->cleanup_work_sem);
3535 up_write(&root->fs_info->cleanup_work_sem);
3537 trans = btrfs_join_transaction(root);
3539 return PTR_ERR(trans);
3540 return btrfs_commit_transaction(trans, root);
3543 int close_ctree(struct btrfs_root *root)
3545 struct btrfs_fs_info *fs_info = root->fs_info;
3548 fs_info->closing = 1;
3551 /* wait for the uuid_scan task to finish */
3552 down(&fs_info->uuid_tree_rescan_sem);
3553 /* avoid complains from lockdep et al., set sem back to initial state */
3554 up(&fs_info->uuid_tree_rescan_sem);
3556 /* pause restriper - we want to resume on mount */
3557 btrfs_pause_balance(fs_info);
3559 btrfs_dev_replace_suspend_for_unmount(fs_info);
3561 btrfs_scrub_cancel(fs_info);
3563 /* wait for any defraggers to finish */
3564 wait_event(fs_info->transaction_wait,
3565 (atomic_read(&fs_info->defrag_running) == 0));
3567 /* clear out the rbtree of defraggable inodes */
3568 btrfs_cleanup_defrag_inodes(fs_info);
3570 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3571 ret = btrfs_commit_super(root);
3573 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3576 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3577 btrfs_error_commit_super(root);
3579 btrfs_put_block_group_cache(fs_info);
3581 kthread_stop(fs_info->transaction_kthread);
3582 kthread_stop(fs_info->cleaner_kthread);
3584 fs_info->closing = 2;
3587 btrfs_free_qgroup_config(root->fs_info);
3589 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3590 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3591 percpu_counter_sum(&fs_info->delalloc_bytes));
3594 btrfs_sysfs_remove_one(fs_info);
3596 del_fs_roots(fs_info);
3598 btrfs_free_block_groups(fs_info);
3600 btrfs_stop_all_workers(fs_info);
3602 free_root_pointers(fs_info, 1);
3604 iput(fs_info->btree_inode);
3606 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3607 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3608 btrfsic_unmount(root, fs_info->fs_devices);
3611 btrfs_close_devices(fs_info->fs_devices);
3612 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3614 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3615 percpu_counter_destroy(&fs_info->delalloc_bytes);
3616 bdi_destroy(&fs_info->bdi);
3617 cleanup_srcu_struct(&fs_info->subvol_srcu);
3619 btrfs_free_stripe_hash_table(fs_info);
3621 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3622 root->orphan_block_rsv = NULL;
3627 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3631 struct inode *btree_inode = buf->pages[0]->mapping->host;
3633 ret = extent_buffer_uptodate(buf);
3637 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3638 parent_transid, atomic);
3644 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3646 return set_extent_buffer_uptodate(buf);
3649 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3651 struct btrfs_root *root;
3652 u64 transid = btrfs_header_generation(buf);
3655 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3657 * This is a fast path so only do this check if we have sanity tests
3658 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3659 * outside of the sanity tests.
3661 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3664 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3665 btrfs_assert_tree_locked(buf);
3666 if (transid != root->fs_info->generation)
3667 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3668 "found %llu running %llu\n",
3669 buf->start, transid, root->fs_info->generation);
3670 was_dirty = set_extent_buffer_dirty(buf);
3672 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3674 root->fs_info->dirty_metadata_batch);
3677 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3681 * looks as though older kernels can get into trouble with
3682 * this code, they end up stuck in balance_dirty_pages forever
3686 if (current->flags & PF_MEMALLOC)
3690 btrfs_balance_delayed_items(root);
3692 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3693 BTRFS_DIRTY_METADATA_THRESH);
3695 balance_dirty_pages_ratelimited(
3696 root->fs_info->btree_inode->i_mapping);
3701 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3703 __btrfs_btree_balance_dirty(root, 1);
3706 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3708 __btrfs_btree_balance_dirty(root, 0);
3711 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3713 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3714 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3717 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3721 * Placeholder for checks
3726 static void btrfs_error_commit_super(struct btrfs_root *root)
3728 mutex_lock(&root->fs_info->cleaner_mutex);
3729 btrfs_run_delayed_iputs(root);
3730 mutex_unlock(&root->fs_info->cleaner_mutex);
3732 down_write(&root->fs_info->cleanup_work_sem);
3733 up_write(&root->fs_info->cleanup_work_sem);
3735 /* cleanup FS via transaction */
3736 btrfs_cleanup_transaction(root);
3739 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3740 struct btrfs_root *root)
3742 struct btrfs_inode *btrfs_inode;
3743 struct list_head splice;
3745 INIT_LIST_HEAD(&splice);
3747 mutex_lock(&root->fs_info->ordered_operations_mutex);
3748 spin_lock(&root->fs_info->ordered_root_lock);
3750 list_splice_init(&t->ordered_operations, &splice);
3751 while (!list_empty(&splice)) {
3752 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3753 ordered_operations);
3755 list_del_init(&btrfs_inode->ordered_operations);
3756 spin_unlock(&root->fs_info->ordered_root_lock);
3758 btrfs_invalidate_inodes(btrfs_inode->root);
3760 spin_lock(&root->fs_info->ordered_root_lock);
3763 spin_unlock(&root->fs_info->ordered_root_lock);
3764 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3767 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3769 struct btrfs_ordered_extent *ordered;
3771 spin_lock(&root->ordered_extent_lock);
3773 * This will just short circuit the ordered completion stuff which will
3774 * make sure the ordered extent gets properly cleaned up.
3776 list_for_each_entry(ordered, &root->ordered_extents,
3778 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3779 spin_unlock(&root->ordered_extent_lock);
3782 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3784 struct btrfs_root *root;
3785 struct list_head splice;
3787 INIT_LIST_HEAD(&splice);
3789 spin_lock(&fs_info->ordered_root_lock);
3790 list_splice_init(&fs_info->ordered_roots, &splice);
3791 while (!list_empty(&splice)) {
3792 root = list_first_entry(&splice, struct btrfs_root,
3794 list_move_tail(&root->ordered_root,
3795 &fs_info->ordered_roots);
3797 btrfs_destroy_ordered_extents(root);
3799 cond_resched_lock(&fs_info->ordered_root_lock);
3801 spin_unlock(&fs_info->ordered_root_lock);
3804 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3805 struct btrfs_root *root)
3807 struct rb_node *node;
3808 struct btrfs_delayed_ref_root *delayed_refs;
3809 struct btrfs_delayed_ref_node *ref;
3812 delayed_refs = &trans->delayed_refs;
3814 spin_lock(&delayed_refs->lock);
3815 if (delayed_refs->num_entries == 0) {
3816 spin_unlock(&delayed_refs->lock);
3817 printk(KERN_INFO "delayed_refs has NO entry\n");
3821 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3822 struct btrfs_delayed_ref_head *head = NULL;
3823 bool pin_bytes = false;
3825 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3826 atomic_set(&ref->refs, 1);
3827 if (btrfs_delayed_ref_is_head(ref)) {
3829 head = btrfs_delayed_node_to_head(ref);
3830 if (!mutex_trylock(&head->mutex)) {
3831 atomic_inc(&ref->refs);
3832 spin_unlock(&delayed_refs->lock);
3834 /* Need to wait for the delayed ref to run */
3835 mutex_lock(&head->mutex);
3836 mutex_unlock(&head->mutex);
3837 btrfs_put_delayed_ref(ref);
3839 spin_lock(&delayed_refs->lock);
3843 if (head->must_insert_reserved)
3845 btrfs_free_delayed_extent_op(head->extent_op);
3846 delayed_refs->num_heads--;
3847 if (list_empty(&head->cluster))
3848 delayed_refs->num_heads_ready--;
3849 list_del_init(&head->cluster);
3853 rb_erase(&ref->rb_node, &delayed_refs->root);
3855 rb_erase(&head->href_node, &delayed_refs->href_root);
3857 delayed_refs->num_entries--;
3858 spin_unlock(&delayed_refs->lock);
3861 btrfs_pin_extent(root, ref->bytenr,
3863 mutex_unlock(&head->mutex);
3865 btrfs_put_delayed_ref(ref);
3868 spin_lock(&delayed_refs->lock);
3871 spin_unlock(&delayed_refs->lock);
3876 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3878 struct btrfs_inode *btrfs_inode;
3879 struct list_head splice;
3881 INIT_LIST_HEAD(&splice);
3883 spin_lock(&root->delalloc_lock);
3884 list_splice_init(&root->delalloc_inodes, &splice);
3886 while (!list_empty(&splice)) {
3887 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3890 list_del_init(&btrfs_inode->delalloc_inodes);
3891 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3892 &btrfs_inode->runtime_flags);
3893 spin_unlock(&root->delalloc_lock);
3895 btrfs_invalidate_inodes(btrfs_inode->root);
3897 spin_lock(&root->delalloc_lock);
3900 spin_unlock(&root->delalloc_lock);
3903 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3905 struct btrfs_root *root;
3906 struct list_head splice;
3908 INIT_LIST_HEAD(&splice);
3910 spin_lock(&fs_info->delalloc_root_lock);
3911 list_splice_init(&fs_info->delalloc_roots, &splice);
3912 while (!list_empty(&splice)) {
3913 root = list_first_entry(&splice, struct btrfs_root,
3915 list_del_init(&root->delalloc_root);
3916 root = btrfs_grab_fs_root(root);
3918 spin_unlock(&fs_info->delalloc_root_lock);
3920 btrfs_destroy_delalloc_inodes(root);
3921 btrfs_put_fs_root(root);
3923 spin_lock(&fs_info->delalloc_root_lock);
3925 spin_unlock(&fs_info->delalloc_root_lock);
3928 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3929 struct extent_io_tree *dirty_pages,
3933 struct extent_buffer *eb;
3938 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3943 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3944 while (start <= end) {
3945 eb = btrfs_find_tree_block(root, start,
3947 start += root->leafsize;
3950 wait_on_extent_buffer_writeback(eb);
3952 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3954 clear_extent_buffer_dirty(eb);
3955 free_extent_buffer_stale(eb);
3962 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3963 struct extent_io_tree *pinned_extents)
3965 struct extent_io_tree *unpin;
3971 unpin = pinned_extents;
3974 ret = find_first_extent_bit(unpin, 0, &start, &end,
3975 EXTENT_DIRTY, NULL);
3980 if (btrfs_test_opt(root, DISCARD))
3981 ret = btrfs_error_discard_extent(root, start,
3985 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3986 btrfs_error_unpin_extent_range(root, start, end);
3991 if (unpin == &root->fs_info->freed_extents[0])
3992 unpin = &root->fs_info->freed_extents[1];
3994 unpin = &root->fs_info->freed_extents[0];
4002 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4003 struct btrfs_root *root)
4005 btrfs_destroy_ordered_operations(cur_trans, root);
4007 btrfs_destroy_delayed_refs(cur_trans, root);
4009 cur_trans->state = TRANS_STATE_COMMIT_START;
4010 wake_up(&root->fs_info->transaction_blocked_wait);
4012 cur_trans->state = TRANS_STATE_UNBLOCKED;
4013 wake_up(&root->fs_info->transaction_wait);
4015 btrfs_destroy_delayed_inodes(root);
4016 btrfs_assert_delayed_root_empty(root);
4018 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4020 btrfs_destroy_pinned_extent(root,
4021 root->fs_info->pinned_extents);
4023 cur_trans->state =TRANS_STATE_COMPLETED;
4024 wake_up(&cur_trans->commit_wait);
4027 memset(cur_trans, 0, sizeof(*cur_trans));
4028 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4032 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4034 struct btrfs_transaction *t;
4036 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4038 spin_lock(&root->fs_info->trans_lock);
4039 while (!list_empty(&root->fs_info->trans_list)) {
4040 t = list_first_entry(&root->fs_info->trans_list,
4041 struct btrfs_transaction, list);
4042 if (t->state >= TRANS_STATE_COMMIT_START) {
4043 atomic_inc(&t->use_count);
4044 spin_unlock(&root->fs_info->trans_lock);
4045 btrfs_wait_for_commit(root, t->transid);
4046 btrfs_put_transaction(t);
4047 spin_lock(&root->fs_info->trans_lock);
4050 if (t == root->fs_info->running_transaction) {
4051 t->state = TRANS_STATE_COMMIT_DOING;
4052 spin_unlock(&root->fs_info->trans_lock);
4054 * We wait for 0 num_writers since we don't hold a trans
4055 * handle open currently for this transaction.
4057 wait_event(t->writer_wait,
4058 atomic_read(&t->num_writers) == 0);
4060 spin_unlock(&root->fs_info->trans_lock);
4062 btrfs_cleanup_one_transaction(t, root);
4064 spin_lock(&root->fs_info->trans_lock);
4065 if (t == root->fs_info->running_transaction)
4066 root->fs_info->running_transaction = NULL;
4067 list_del_init(&t->list);
4068 spin_unlock(&root->fs_info->trans_lock);
4070 btrfs_put_transaction(t);
4071 trace_btrfs_transaction_commit(root);
4072 spin_lock(&root->fs_info->trans_lock);
4074 spin_unlock(&root->fs_info->trans_lock);
4075 btrfs_destroy_all_ordered_extents(root->fs_info);
4076 btrfs_destroy_delayed_inodes(root);
4077 btrfs_assert_delayed_root_empty(root);
4078 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4079 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4080 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4085 static struct extent_io_ops btree_extent_io_ops = {
4086 .readpage_end_io_hook = btree_readpage_end_io_hook,
4087 .readpage_io_failed_hook = btree_io_failed_hook,
4088 .submit_bio_hook = btree_submit_bio_hook,
4089 /* note we're sharing with inode.c for the merge bio hook */
4090 .merge_bio_hook = btrfs_merge_bio_hook,