2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/slab.h>
30 #include <linux/migrate.h>
31 #include <linux/ratelimit.h>
32 #include <linux/uuid.h>
33 #include <linux/semaphore.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "print-tree.h"
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 btrfs_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
304 "BTRFS: %s checksum verify 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) {
387 "BTRFS: super block crcs don't match, older mkfs detected\n");
392 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
393 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
402 * helper to read a given tree block, doing retries as required when
403 * the checksums don't match and we have alternate mirrors to try.
405 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
406 struct extent_buffer *eb,
407 u64 start, u64 parent_transid)
409 struct extent_io_tree *io_tree;
414 int failed_mirror = 0;
416 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
417 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
419 ret = read_extent_buffer_pages(io_tree, eb, start,
421 btree_get_extent, mirror_num);
423 if (!verify_parent_transid(io_tree, eb,
431 * This buffer's crc is fine, but its contents are corrupted, so
432 * there is no reason to read the other copies, they won't be
435 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
438 num_copies = btrfs_num_copies(root->fs_info,
443 if (!failed_mirror) {
445 failed_mirror = eb->read_mirror;
449 if (mirror_num == failed_mirror)
452 if (mirror_num > num_copies)
456 if (failed && !ret && failed_mirror)
457 repair_eb_io_failure(root, eb, failed_mirror);
463 * checksum a dirty tree block before IO. This has extra checks to make sure
464 * we only fill in the checksum field in the first page of a multi-page block
467 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
469 u64 start = page_offset(page);
471 struct extent_buffer *eb;
473 eb = (struct extent_buffer *)page->private;
474 if (page != eb->pages[0])
476 found_start = btrfs_header_bytenr(eb);
477 if (WARN_ON(found_start != start || !PageUptodate(page)))
479 csum_tree_block(root, eb, 0);
483 static int check_tree_block_fsid(struct btrfs_root *root,
484 struct extent_buffer *eb)
486 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
487 u8 fsid[BTRFS_UUID_SIZE];
490 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
492 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
496 fs_devices = fs_devices->seed;
501 #define CORRUPT(reason, eb, root, slot) \
502 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
503 "root=%llu, slot=%d", reason, \
504 btrfs_header_bytenr(eb), root->objectid, slot)
506 static noinline int check_leaf(struct btrfs_root *root,
507 struct extent_buffer *leaf)
509 struct btrfs_key key;
510 struct btrfs_key leaf_key;
511 u32 nritems = btrfs_header_nritems(leaf);
517 /* Check the 0 item */
518 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
519 BTRFS_LEAF_DATA_SIZE(root)) {
520 CORRUPT("invalid item offset size pair", leaf, root, 0);
525 * Check to make sure each items keys are in the correct order and their
526 * offsets make sense. We only have to loop through nritems-1 because
527 * we check the current slot against the next slot, which verifies the
528 * next slot's offset+size makes sense and that the current's slot
531 for (slot = 0; slot < nritems - 1; slot++) {
532 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
533 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
535 /* Make sure the keys are in the right order */
536 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
537 CORRUPT("bad key order", leaf, root, slot);
542 * Make sure the offset and ends are right, remember that the
543 * item data starts at the end of the leaf and grows towards the
546 if (btrfs_item_offset_nr(leaf, slot) !=
547 btrfs_item_end_nr(leaf, slot + 1)) {
548 CORRUPT("slot offset bad", leaf, root, slot);
553 * Check to make sure that we don't point outside of the leaf,
554 * just incase all the items are consistent to eachother, but
555 * all point outside of the leaf.
557 if (btrfs_item_end_nr(leaf, slot) >
558 BTRFS_LEAF_DATA_SIZE(root)) {
559 CORRUPT("slot end outside of leaf", leaf, root, slot);
567 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
568 u64 phy_offset, struct page *page,
569 u64 start, u64 end, int mirror)
573 struct extent_buffer *eb;
574 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
581 eb = (struct extent_buffer *)page->private;
583 /* the pending IO might have been the only thing that kept this buffer
584 * in memory. Make sure we have a ref for all this other checks
586 extent_buffer_get(eb);
588 reads_done = atomic_dec_and_test(&eb->io_pages);
592 eb->read_mirror = mirror;
593 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
598 found_start = btrfs_header_bytenr(eb);
599 if (found_start != eb->start) {
600 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start "
602 found_start, eb->start);
606 if (check_tree_block_fsid(root, eb)) {
607 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n",
612 found_level = btrfs_header_level(eb);
613 if (found_level >= BTRFS_MAX_LEVEL) {
614 btrfs_info(root->fs_info, "bad tree block level %d",
615 (int)btrfs_header_level(eb));
620 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
623 ret = csum_tree_block(root, eb, 1);
630 * If this is a leaf block and it is corrupt, set the corrupt bit so
631 * that we don't try and read the other copies of this block, just
634 if (found_level == 0 && check_leaf(root, eb)) {
635 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
640 set_extent_buffer_uptodate(eb);
643 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
644 btree_readahead_hook(root, eb, eb->start, ret);
648 * our io error hook is going to dec the io pages
649 * again, we have to make sure it has something
652 atomic_inc(&eb->io_pages);
653 clear_extent_buffer_uptodate(eb);
655 free_extent_buffer(eb);
660 static int btree_io_failed_hook(struct page *page, int failed_mirror)
662 struct extent_buffer *eb;
663 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
665 eb = (struct extent_buffer *)page->private;
666 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
667 eb->read_mirror = failed_mirror;
668 atomic_dec(&eb->io_pages);
669 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
670 btree_readahead_hook(root, eb, eb->start, -EIO);
671 return -EIO; /* we fixed nothing */
674 static void end_workqueue_bio(struct bio *bio, int err)
676 struct end_io_wq *end_io_wq = bio->bi_private;
677 struct btrfs_fs_info *fs_info;
679 fs_info = end_io_wq->info;
680 end_io_wq->error = err;
681 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
683 if (bio->bi_rw & REQ_WRITE) {
684 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
685 btrfs_queue_work(fs_info->endio_meta_write_workers,
687 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
688 btrfs_queue_work(fs_info->endio_freespace_worker,
690 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
691 btrfs_queue_work(fs_info->endio_raid56_workers,
694 btrfs_queue_work(fs_info->endio_write_workers,
697 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
698 btrfs_queue_work(fs_info->endio_raid56_workers,
700 else if (end_io_wq->metadata)
701 btrfs_queue_work(fs_info->endio_meta_workers,
704 btrfs_queue_work(fs_info->endio_workers,
710 * For the metadata arg you want
713 * 1 - if normal metadta
714 * 2 - if writing to the free space cache area
715 * 3 - raid parity work
717 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
720 struct end_io_wq *end_io_wq;
721 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
725 end_io_wq->private = bio->bi_private;
726 end_io_wq->end_io = bio->bi_end_io;
727 end_io_wq->info = info;
728 end_io_wq->error = 0;
729 end_io_wq->bio = bio;
730 end_io_wq->metadata = metadata;
732 bio->bi_private = end_io_wq;
733 bio->bi_end_io = end_workqueue_bio;
737 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
739 unsigned long limit = min_t(unsigned long,
740 info->thread_pool_size,
741 info->fs_devices->open_devices);
745 static void run_one_async_start(struct btrfs_work *work)
747 struct async_submit_bio *async;
750 async = container_of(work, struct async_submit_bio, work);
751 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
752 async->mirror_num, async->bio_flags,
758 static void run_one_async_done(struct btrfs_work *work)
760 struct btrfs_fs_info *fs_info;
761 struct async_submit_bio *async;
764 async = container_of(work, struct async_submit_bio, work);
765 fs_info = BTRFS_I(async->inode)->root->fs_info;
767 limit = btrfs_async_submit_limit(fs_info);
768 limit = limit * 2 / 3;
770 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
771 waitqueue_active(&fs_info->async_submit_wait))
772 wake_up(&fs_info->async_submit_wait);
774 /* If an error occured we just want to clean up the bio and move on */
776 bio_endio(async->bio, async->error);
780 async->submit_bio_done(async->inode, async->rw, async->bio,
781 async->mirror_num, async->bio_flags,
785 static void run_one_async_free(struct btrfs_work *work)
787 struct async_submit_bio *async;
789 async = container_of(work, struct async_submit_bio, work);
793 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
794 int rw, struct bio *bio, int mirror_num,
795 unsigned long bio_flags,
797 extent_submit_bio_hook_t *submit_bio_start,
798 extent_submit_bio_hook_t *submit_bio_done)
800 struct async_submit_bio *async;
802 async = kmalloc(sizeof(*async), GFP_NOFS);
806 async->inode = inode;
809 async->mirror_num = mirror_num;
810 async->submit_bio_start = submit_bio_start;
811 async->submit_bio_done = submit_bio_done;
813 btrfs_init_work(&async->work, run_one_async_start,
814 run_one_async_done, run_one_async_free);
816 async->bio_flags = bio_flags;
817 async->bio_offset = bio_offset;
821 atomic_inc(&fs_info->nr_async_submits);
824 btrfs_set_work_high_priority(&async->work);
826 btrfs_queue_work(fs_info->workers, &async->work);
828 while (atomic_read(&fs_info->async_submit_draining) &&
829 atomic_read(&fs_info->nr_async_submits)) {
830 wait_event(fs_info->async_submit_wait,
831 (atomic_read(&fs_info->nr_async_submits) == 0));
837 static int btree_csum_one_bio(struct bio *bio)
839 struct bio_vec *bvec = bio->bi_io_vec;
841 struct btrfs_root *root;
844 WARN_ON(bio->bi_vcnt <= 0);
845 while (bio_index < bio->bi_vcnt) {
846 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
847 ret = csum_dirty_buffer(root, bvec->bv_page);
856 static int __btree_submit_bio_start(struct inode *inode, int rw,
857 struct bio *bio, int mirror_num,
858 unsigned long bio_flags,
862 * when we're called for a write, we're already in the async
863 * submission context. Just jump into btrfs_map_bio
865 return btree_csum_one_bio(bio);
868 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
869 int mirror_num, unsigned long bio_flags,
875 * when we're called for a write, we're already in the async
876 * submission context. Just jump into btrfs_map_bio
878 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
884 static int check_async_write(struct inode *inode, unsigned long bio_flags)
886 if (bio_flags & EXTENT_BIO_TREE_LOG)
895 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
896 int mirror_num, unsigned long bio_flags,
899 int async = check_async_write(inode, bio_flags);
902 if (!(rw & REQ_WRITE)) {
904 * called for a read, do the setup so that checksum validation
905 * can happen in the async kernel threads
907 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
911 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
914 ret = btree_csum_one_bio(bio);
917 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
921 * kthread helpers are used to submit writes so that
922 * checksumming can happen in parallel across all CPUs
924 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
925 inode, rw, bio, mirror_num, 0,
927 __btree_submit_bio_start,
928 __btree_submit_bio_done);
938 #ifdef CONFIG_MIGRATION
939 static int btree_migratepage(struct address_space *mapping,
940 struct page *newpage, struct page *page,
941 enum migrate_mode mode)
944 * we can't safely write a btree page from here,
945 * we haven't done the locking hook
950 * Buffers may be managed in a filesystem specific way.
951 * We must have no buffers or drop them.
953 if (page_has_private(page) &&
954 !try_to_release_page(page, GFP_KERNEL))
956 return migrate_page(mapping, newpage, page, mode);
961 static int btree_writepages(struct address_space *mapping,
962 struct writeback_control *wbc)
964 struct btrfs_fs_info *fs_info;
967 if (wbc->sync_mode == WB_SYNC_NONE) {
969 if (wbc->for_kupdate)
972 fs_info = BTRFS_I(mapping->host)->root->fs_info;
973 /* this is a bit racy, but that's ok */
974 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
975 BTRFS_DIRTY_METADATA_THRESH);
979 return btree_write_cache_pages(mapping, wbc);
982 static int btree_readpage(struct file *file, struct page *page)
984 struct extent_io_tree *tree;
985 tree = &BTRFS_I(page->mapping->host)->io_tree;
986 return extent_read_full_page(tree, page, btree_get_extent, 0);
989 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
991 if (PageWriteback(page) || PageDirty(page))
994 return try_release_extent_buffer(page);
997 static void btree_invalidatepage(struct page *page, unsigned int offset,
1000 struct extent_io_tree *tree;
1001 tree = &BTRFS_I(page->mapping->host)->io_tree;
1002 extent_invalidatepage(tree, page, offset);
1003 btree_releasepage(page, GFP_NOFS);
1004 if (PagePrivate(page)) {
1005 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1006 "page private not zero on page %llu",
1007 (unsigned long long)page_offset(page));
1008 ClearPagePrivate(page);
1009 set_page_private(page, 0);
1010 page_cache_release(page);
1014 static int btree_set_page_dirty(struct page *page)
1017 struct extent_buffer *eb;
1019 BUG_ON(!PagePrivate(page));
1020 eb = (struct extent_buffer *)page->private;
1022 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1023 BUG_ON(!atomic_read(&eb->refs));
1024 btrfs_assert_tree_locked(eb);
1026 return __set_page_dirty_nobuffers(page);
1029 static const struct address_space_operations btree_aops = {
1030 .readpage = btree_readpage,
1031 .writepages = btree_writepages,
1032 .releasepage = btree_releasepage,
1033 .invalidatepage = btree_invalidatepage,
1034 #ifdef CONFIG_MIGRATION
1035 .migratepage = btree_migratepage,
1037 .set_page_dirty = btree_set_page_dirty,
1040 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1043 struct extent_buffer *buf = NULL;
1044 struct inode *btree_inode = root->fs_info->btree_inode;
1047 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1050 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1051 buf, 0, WAIT_NONE, btree_get_extent, 0);
1052 free_extent_buffer(buf);
1056 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1057 int mirror_num, struct extent_buffer **eb)
1059 struct extent_buffer *buf = NULL;
1060 struct inode *btree_inode = root->fs_info->btree_inode;
1061 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1064 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1068 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1070 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1071 btree_get_extent, mirror_num);
1073 free_extent_buffer(buf);
1077 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1078 free_extent_buffer(buf);
1080 } else if (extent_buffer_uptodate(buf)) {
1083 free_extent_buffer(buf);
1088 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1089 u64 bytenr, u32 blocksize)
1091 return find_extent_buffer(root->fs_info, bytenr);
1094 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1095 u64 bytenr, u32 blocksize)
1097 return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1101 int btrfs_write_tree_block(struct extent_buffer *buf)
1103 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1104 buf->start + buf->len - 1);
1107 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1109 return filemap_fdatawait_range(buf->pages[0]->mapping,
1110 buf->start, buf->start + buf->len - 1);
1113 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1114 u32 blocksize, u64 parent_transid)
1116 struct extent_buffer *buf = NULL;
1119 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1123 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1125 free_extent_buffer(buf);
1132 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1133 struct extent_buffer *buf)
1135 struct btrfs_fs_info *fs_info = root->fs_info;
1137 if (btrfs_header_generation(buf) ==
1138 fs_info->running_transaction->transid) {
1139 btrfs_assert_tree_locked(buf);
1141 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1142 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1144 fs_info->dirty_metadata_batch);
1145 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1146 btrfs_set_lock_blocking(buf);
1147 clear_extent_buffer_dirty(buf);
1152 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1154 struct btrfs_subvolume_writers *writers;
1157 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1159 return ERR_PTR(-ENOMEM);
1161 ret = percpu_counter_init(&writers->counter, 0);
1164 return ERR_PTR(ret);
1167 init_waitqueue_head(&writers->wait);
1172 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1174 percpu_counter_destroy(&writers->counter);
1178 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1179 u32 stripesize, struct btrfs_root *root,
1180 struct btrfs_fs_info *fs_info,
1184 root->commit_root = NULL;
1185 root->sectorsize = sectorsize;
1186 root->nodesize = nodesize;
1187 root->leafsize = leafsize;
1188 root->stripesize = stripesize;
1190 root->track_dirty = 0;
1192 root->orphan_item_inserted = 0;
1193 root->orphan_cleanup_state = 0;
1195 root->objectid = objectid;
1196 root->last_trans = 0;
1197 root->highest_objectid = 0;
1198 root->nr_delalloc_inodes = 0;
1199 root->nr_ordered_extents = 0;
1201 root->inode_tree = RB_ROOT;
1202 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1203 root->block_rsv = NULL;
1204 root->orphan_block_rsv = NULL;
1206 INIT_LIST_HEAD(&root->dirty_list);
1207 INIT_LIST_HEAD(&root->root_list);
1208 INIT_LIST_HEAD(&root->delalloc_inodes);
1209 INIT_LIST_HEAD(&root->delalloc_root);
1210 INIT_LIST_HEAD(&root->ordered_extents);
1211 INIT_LIST_HEAD(&root->ordered_root);
1212 INIT_LIST_HEAD(&root->logged_list[0]);
1213 INIT_LIST_HEAD(&root->logged_list[1]);
1214 spin_lock_init(&root->orphan_lock);
1215 spin_lock_init(&root->inode_lock);
1216 spin_lock_init(&root->delalloc_lock);
1217 spin_lock_init(&root->ordered_extent_lock);
1218 spin_lock_init(&root->accounting_lock);
1219 spin_lock_init(&root->log_extents_lock[0]);
1220 spin_lock_init(&root->log_extents_lock[1]);
1221 mutex_init(&root->objectid_mutex);
1222 mutex_init(&root->log_mutex);
1223 mutex_init(&root->ordered_extent_mutex);
1224 mutex_init(&root->delalloc_mutex);
1225 init_waitqueue_head(&root->log_writer_wait);
1226 init_waitqueue_head(&root->log_commit_wait[0]);
1227 init_waitqueue_head(&root->log_commit_wait[1]);
1228 INIT_LIST_HEAD(&root->log_ctxs[0]);
1229 INIT_LIST_HEAD(&root->log_ctxs[1]);
1230 atomic_set(&root->log_commit[0], 0);
1231 atomic_set(&root->log_commit[1], 0);
1232 atomic_set(&root->log_writers, 0);
1233 atomic_set(&root->log_batch, 0);
1234 atomic_set(&root->orphan_inodes, 0);
1235 atomic_set(&root->refs, 1);
1236 atomic_set(&root->will_be_snapshoted, 0);
1237 root->log_transid = 0;
1238 root->log_transid_committed = -1;
1239 root->last_log_commit = 0;
1241 extent_io_tree_init(&root->dirty_log_pages,
1242 fs_info->btree_inode->i_mapping);
1244 memset(&root->root_key, 0, sizeof(root->root_key));
1245 memset(&root->root_item, 0, sizeof(root->root_item));
1246 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1247 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1249 root->defrag_trans_start = fs_info->generation;
1251 root->defrag_trans_start = 0;
1252 init_completion(&root->kobj_unregister);
1253 root->defrag_running = 0;
1254 root->root_key.objectid = objectid;
1257 spin_lock_init(&root->root_item_lock);
1260 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1262 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1264 root->fs_info = fs_info;
1268 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1269 /* Should only be used by the testing infrastructure */
1270 struct btrfs_root *btrfs_alloc_dummy_root(void)
1272 struct btrfs_root *root;
1274 root = btrfs_alloc_root(NULL);
1276 return ERR_PTR(-ENOMEM);
1277 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1278 root->dummy_root = 1;
1284 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1285 struct btrfs_fs_info *fs_info,
1288 struct extent_buffer *leaf;
1289 struct btrfs_root *tree_root = fs_info->tree_root;
1290 struct btrfs_root *root;
1291 struct btrfs_key key;
1295 root = btrfs_alloc_root(fs_info);
1297 return ERR_PTR(-ENOMEM);
1299 __setup_root(tree_root->nodesize, tree_root->leafsize,
1300 tree_root->sectorsize, tree_root->stripesize,
1301 root, fs_info, objectid);
1302 root->root_key.objectid = objectid;
1303 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1304 root->root_key.offset = 0;
1306 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1307 0, objectid, NULL, 0, 0, 0);
1309 ret = PTR_ERR(leaf);
1314 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1315 btrfs_set_header_bytenr(leaf, leaf->start);
1316 btrfs_set_header_generation(leaf, trans->transid);
1317 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1318 btrfs_set_header_owner(leaf, objectid);
1321 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1323 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1324 btrfs_header_chunk_tree_uuid(leaf),
1326 btrfs_mark_buffer_dirty(leaf);
1328 root->commit_root = btrfs_root_node(root);
1329 root->track_dirty = 1;
1332 root->root_item.flags = 0;
1333 root->root_item.byte_limit = 0;
1334 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1335 btrfs_set_root_generation(&root->root_item, trans->transid);
1336 btrfs_set_root_level(&root->root_item, 0);
1337 btrfs_set_root_refs(&root->root_item, 1);
1338 btrfs_set_root_used(&root->root_item, leaf->len);
1339 btrfs_set_root_last_snapshot(&root->root_item, 0);
1340 btrfs_set_root_dirid(&root->root_item, 0);
1342 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1343 root->root_item.drop_level = 0;
1345 key.objectid = objectid;
1346 key.type = BTRFS_ROOT_ITEM_KEY;
1348 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1352 btrfs_tree_unlock(leaf);
1358 btrfs_tree_unlock(leaf);
1359 free_extent_buffer(leaf);
1363 return ERR_PTR(ret);
1366 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1367 struct btrfs_fs_info *fs_info)
1369 struct btrfs_root *root;
1370 struct btrfs_root *tree_root = fs_info->tree_root;
1371 struct extent_buffer *leaf;
1373 root = btrfs_alloc_root(fs_info);
1375 return ERR_PTR(-ENOMEM);
1377 __setup_root(tree_root->nodesize, tree_root->leafsize,
1378 tree_root->sectorsize, tree_root->stripesize,
1379 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1381 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1382 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1383 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1385 * log trees do not get reference counted because they go away
1386 * before a real commit is actually done. They do store pointers
1387 * to file data extents, and those reference counts still get
1388 * updated (along with back refs to the log tree).
1392 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1393 BTRFS_TREE_LOG_OBJECTID, NULL,
1397 return ERR_CAST(leaf);
1400 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1401 btrfs_set_header_bytenr(leaf, leaf->start);
1402 btrfs_set_header_generation(leaf, trans->transid);
1403 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1404 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1407 write_extent_buffer(root->node, root->fs_info->fsid,
1408 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1409 btrfs_mark_buffer_dirty(root->node);
1410 btrfs_tree_unlock(root->node);
1414 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1415 struct btrfs_fs_info *fs_info)
1417 struct btrfs_root *log_root;
1419 log_root = alloc_log_tree(trans, fs_info);
1420 if (IS_ERR(log_root))
1421 return PTR_ERR(log_root);
1422 WARN_ON(fs_info->log_root_tree);
1423 fs_info->log_root_tree = log_root;
1427 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1428 struct btrfs_root *root)
1430 struct btrfs_root *log_root;
1431 struct btrfs_inode_item *inode_item;
1433 log_root = alloc_log_tree(trans, root->fs_info);
1434 if (IS_ERR(log_root))
1435 return PTR_ERR(log_root);
1437 log_root->last_trans = trans->transid;
1438 log_root->root_key.offset = root->root_key.objectid;
1440 inode_item = &log_root->root_item.inode;
1441 btrfs_set_stack_inode_generation(inode_item, 1);
1442 btrfs_set_stack_inode_size(inode_item, 3);
1443 btrfs_set_stack_inode_nlink(inode_item, 1);
1444 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1445 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1447 btrfs_set_root_node(&log_root->root_item, log_root->node);
1449 WARN_ON(root->log_root);
1450 root->log_root = log_root;
1451 root->log_transid = 0;
1452 root->log_transid_committed = -1;
1453 root->last_log_commit = 0;
1457 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1458 struct btrfs_key *key)
1460 struct btrfs_root *root;
1461 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1462 struct btrfs_path *path;
1467 path = btrfs_alloc_path();
1469 return ERR_PTR(-ENOMEM);
1471 root = btrfs_alloc_root(fs_info);
1477 __setup_root(tree_root->nodesize, tree_root->leafsize,
1478 tree_root->sectorsize, tree_root->stripesize,
1479 root, fs_info, key->objectid);
1481 ret = btrfs_find_root(tree_root, key, path,
1482 &root->root_item, &root->root_key);
1489 generation = btrfs_root_generation(&root->root_item);
1490 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1491 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1492 blocksize, generation);
1496 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1500 root->commit_root = btrfs_root_node(root);
1502 btrfs_free_path(path);
1506 free_extent_buffer(root->node);
1510 root = ERR_PTR(ret);
1514 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1515 struct btrfs_key *location)
1517 struct btrfs_root *root;
1519 root = btrfs_read_tree_root(tree_root, location);
1523 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1525 btrfs_check_and_init_root_item(&root->root_item);
1531 int btrfs_init_fs_root(struct btrfs_root *root)
1534 struct btrfs_subvolume_writers *writers;
1536 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1537 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1539 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1544 writers = btrfs_alloc_subvolume_writers();
1545 if (IS_ERR(writers)) {
1546 ret = PTR_ERR(writers);
1549 root->subv_writers = writers;
1551 btrfs_init_free_ino_ctl(root);
1552 mutex_init(&root->fs_commit_mutex);
1553 spin_lock_init(&root->cache_lock);
1554 init_waitqueue_head(&root->cache_wait);
1556 ret = get_anon_bdev(&root->anon_dev);
1562 btrfs_free_subvolume_writers(root->subv_writers);
1564 kfree(root->free_ino_ctl);
1565 kfree(root->free_ino_pinned);
1569 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1572 struct btrfs_root *root;
1574 spin_lock(&fs_info->fs_roots_radix_lock);
1575 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1576 (unsigned long)root_id);
1577 spin_unlock(&fs_info->fs_roots_radix_lock);
1581 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1582 struct btrfs_root *root)
1586 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1590 spin_lock(&fs_info->fs_roots_radix_lock);
1591 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1592 (unsigned long)root->root_key.objectid,
1596 spin_unlock(&fs_info->fs_roots_radix_lock);
1597 radix_tree_preload_end();
1602 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1603 struct btrfs_key *location,
1606 struct btrfs_root *root;
1609 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1610 return fs_info->tree_root;
1611 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1612 return fs_info->extent_root;
1613 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1614 return fs_info->chunk_root;
1615 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1616 return fs_info->dev_root;
1617 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1618 return fs_info->csum_root;
1619 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1620 return fs_info->quota_root ? fs_info->quota_root :
1622 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1623 return fs_info->uuid_root ? fs_info->uuid_root :
1626 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1628 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1629 return ERR_PTR(-ENOENT);
1633 root = btrfs_read_fs_root(fs_info->tree_root, location);
1637 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1642 ret = btrfs_init_fs_root(root);
1646 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1647 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1651 root->orphan_item_inserted = 1;
1653 ret = btrfs_insert_fs_root(fs_info, root);
1655 if (ret == -EEXIST) {
1664 return ERR_PTR(ret);
1667 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1669 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1671 struct btrfs_device *device;
1672 struct backing_dev_info *bdi;
1675 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1678 bdi = blk_get_backing_dev_info(device->bdev);
1679 if (bdi && bdi_congested(bdi, bdi_bits)) {
1689 * If this fails, caller must call bdi_destroy() to get rid of the
1692 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1696 bdi->capabilities = BDI_CAP_MAP_COPY;
1697 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1701 bdi->ra_pages = default_backing_dev_info.ra_pages;
1702 bdi->congested_fn = btrfs_congested_fn;
1703 bdi->congested_data = info;
1708 * called by the kthread helper functions to finally call the bio end_io
1709 * functions. This is where read checksum verification actually happens
1711 static void end_workqueue_fn(struct btrfs_work *work)
1714 struct end_io_wq *end_io_wq;
1717 end_io_wq = container_of(work, struct end_io_wq, work);
1718 bio = end_io_wq->bio;
1720 error = end_io_wq->error;
1721 bio->bi_private = end_io_wq->private;
1722 bio->bi_end_io = end_io_wq->end_io;
1724 bio_endio(bio, error);
1727 static int cleaner_kthread(void *arg)
1729 struct btrfs_root *root = arg;
1735 /* Make the cleaner go to sleep early. */
1736 if (btrfs_need_cleaner_sleep(root))
1739 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1743 * Avoid the problem that we change the status of the fs
1744 * during the above check and trylock.
1746 if (btrfs_need_cleaner_sleep(root)) {
1747 mutex_unlock(&root->fs_info->cleaner_mutex);
1751 btrfs_run_delayed_iputs(root);
1752 again = btrfs_clean_one_deleted_snapshot(root);
1753 mutex_unlock(&root->fs_info->cleaner_mutex);
1756 * The defragger has dealt with the R/O remount and umount,
1757 * needn't do anything special here.
1759 btrfs_run_defrag_inodes(root->fs_info);
1761 if (!try_to_freeze() && !again) {
1762 set_current_state(TASK_INTERRUPTIBLE);
1763 if (!kthread_should_stop())
1765 __set_current_state(TASK_RUNNING);
1767 } while (!kthread_should_stop());
1771 static int transaction_kthread(void *arg)
1773 struct btrfs_root *root = arg;
1774 struct btrfs_trans_handle *trans;
1775 struct btrfs_transaction *cur;
1778 unsigned long delay;
1782 cannot_commit = false;
1783 delay = HZ * root->fs_info->commit_interval;
1784 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1786 spin_lock(&root->fs_info->trans_lock);
1787 cur = root->fs_info->running_transaction;
1789 spin_unlock(&root->fs_info->trans_lock);
1793 now = get_seconds();
1794 if (cur->state < TRANS_STATE_BLOCKED &&
1795 (now < cur->start_time ||
1796 now - cur->start_time < root->fs_info->commit_interval)) {
1797 spin_unlock(&root->fs_info->trans_lock);
1801 transid = cur->transid;
1802 spin_unlock(&root->fs_info->trans_lock);
1804 /* If the file system is aborted, this will always fail. */
1805 trans = btrfs_attach_transaction(root);
1806 if (IS_ERR(trans)) {
1807 if (PTR_ERR(trans) != -ENOENT)
1808 cannot_commit = true;
1811 if (transid == trans->transid) {
1812 btrfs_commit_transaction(trans, root);
1814 btrfs_end_transaction(trans, root);
1817 wake_up_process(root->fs_info->cleaner_kthread);
1818 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1820 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1821 &root->fs_info->fs_state)))
1822 btrfs_cleanup_transaction(root);
1823 if (!try_to_freeze()) {
1824 set_current_state(TASK_INTERRUPTIBLE);
1825 if (!kthread_should_stop() &&
1826 (!btrfs_transaction_blocked(root->fs_info) ||
1828 schedule_timeout(delay);
1829 __set_current_state(TASK_RUNNING);
1831 } while (!kthread_should_stop());
1836 * this will find the highest generation in the array of
1837 * root backups. The index of the highest array is returned,
1838 * or -1 if we can't find anything.
1840 * We check to make sure the array is valid by comparing the
1841 * generation of the latest root in the array with the generation
1842 * in the super block. If they don't match we pitch it.
1844 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1847 int newest_index = -1;
1848 struct btrfs_root_backup *root_backup;
1851 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1852 root_backup = info->super_copy->super_roots + i;
1853 cur = btrfs_backup_tree_root_gen(root_backup);
1854 if (cur == newest_gen)
1858 /* check to see if we actually wrapped around */
1859 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1860 root_backup = info->super_copy->super_roots;
1861 cur = btrfs_backup_tree_root_gen(root_backup);
1862 if (cur == newest_gen)
1865 return newest_index;
1870 * find the oldest backup so we know where to store new entries
1871 * in the backup array. This will set the backup_root_index
1872 * field in the fs_info struct
1874 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1877 int newest_index = -1;
1879 newest_index = find_newest_super_backup(info, newest_gen);
1880 /* if there was garbage in there, just move along */
1881 if (newest_index == -1) {
1882 info->backup_root_index = 0;
1884 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1889 * copy all the root pointers into the super backup array.
1890 * this will bump the backup pointer by one when it is
1893 static void backup_super_roots(struct btrfs_fs_info *info)
1896 struct btrfs_root_backup *root_backup;
1899 next_backup = info->backup_root_index;
1900 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1901 BTRFS_NUM_BACKUP_ROOTS;
1904 * just overwrite the last backup if we're at the same generation
1905 * this happens only at umount
1907 root_backup = info->super_for_commit->super_roots + last_backup;
1908 if (btrfs_backup_tree_root_gen(root_backup) ==
1909 btrfs_header_generation(info->tree_root->node))
1910 next_backup = last_backup;
1912 root_backup = info->super_for_commit->super_roots + next_backup;
1915 * make sure all of our padding and empty slots get zero filled
1916 * regardless of which ones we use today
1918 memset(root_backup, 0, sizeof(*root_backup));
1920 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1922 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1923 btrfs_set_backup_tree_root_gen(root_backup,
1924 btrfs_header_generation(info->tree_root->node));
1926 btrfs_set_backup_tree_root_level(root_backup,
1927 btrfs_header_level(info->tree_root->node));
1929 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1930 btrfs_set_backup_chunk_root_gen(root_backup,
1931 btrfs_header_generation(info->chunk_root->node));
1932 btrfs_set_backup_chunk_root_level(root_backup,
1933 btrfs_header_level(info->chunk_root->node));
1935 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1936 btrfs_set_backup_extent_root_gen(root_backup,
1937 btrfs_header_generation(info->extent_root->node));
1938 btrfs_set_backup_extent_root_level(root_backup,
1939 btrfs_header_level(info->extent_root->node));
1942 * we might commit during log recovery, which happens before we set
1943 * the fs_root. Make sure it is valid before we fill it in.
1945 if (info->fs_root && info->fs_root->node) {
1946 btrfs_set_backup_fs_root(root_backup,
1947 info->fs_root->node->start);
1948 btrfs_set_backup_fs_root_gen(root_backup,
1949 btrfs_header_generation(info->fs_root->node));
1950 btrfs_set_backup_fs_root_level(root_backup,
1951 btrfs_header_level(info->fs_root->node));
1954 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1955 btrfs_set_backup_dev_root_gen(root_backup,
1956 btrfs_header_generation(info->dev_root->node));
1957 btrfs_set_backup_dev_root_level(root_backup,
1958 btrfs_header_level(info->dev_root->node));
1960 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1961 btrfs_set_backup_csum_root_gen(root_backup,
1962 btrfs_header_generation(info->csum_root->node));
1963 btrfs_set_backup_csum_root_level(root_backup,
1964 btrfs_header_level(info->csum_root->node));
1966 btrfs_set_backup_total_bytes(root_backup,
1967 btrfs_super_total_bytes(info->super_copy));
1968 btrfs_set_backup_bytes_used(root_backup,
1969 btrfs_super_bytes_used(info->super_copy));
1970 btrfs_set_backup_num_devices(root_backup,
1971 btrfs_super_num_devices(info->super_copy));
1974 * if we don't copy this out to the super_copy, it won't get remembered
1975 * for the next commit
1977 memcpy(&info->super_copy->super_roots,
1978 &info->super_for_commit->super_roots,
1979 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1983 * this copies info out of the root backup array and back into
1984 * the in-memory super block. It is meant to help iterate through
1985 * the array, so you send it the number of backups you've already
1986 * tried and the last backup index you used.
1988 * this returns -1 when it has tried all the backups
1990 static noinline int next_root_backup(struct btrfs_fs_info *info,
1991 struct btrfs_super_block *super,
1992 int *num_backups_tried, int *backup_index)
1994 struct btrfs_root_backup *root_backup;
1995 int newest = *backup_index;
1997 if (*num_backups_tried == 0) {
1998 u64 gen = btrfs_super_generation(super);
2000 newest = find_newest_super_backup(info, gen);
2004 *backup_index = newest;
2005 *num_backups_tried = 1;
2006 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2007 /* we've tried all the backups, all done */
2010 /* jump to the next oldest backup */
2011 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2012 BTRFS_NUM_BACKUP_ROOTS;
2013 *backup_index = newest;
2014 *num_backups_tried += 1;
2016 root_backup = super->super_roots + newest;
2018 btrfs_set_super_generation(super,
2019 btrfs_backup_tree_root_gen(root_backup));
2020 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2021 btrfs_set_super_root_level(super,
2022 btrfs_backup_tree_root_level(root_backup));
2023 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2026 * fixme: the total bytes and num_devices need to match or we should
2029 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2030 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2034 /* helper to cleanup workers */
2035 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2037 btrfs_destroy_workqueue(fs_info->fixup_workers);
2038 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2039 btrfs_destroy_workqueue(fs_info->workers);
2040 btrfs_destroy_workqueue(fs_info->endio_workers);
2041 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2042 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2043 btrfs_destroy_workqueue(fs_info->rmw_workers);
2044 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2045 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2046 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2047 btrfs_destroy_workqueue(fs_info->submit_workers);
2048 btrfs_destroy_workqueue(fs_info->delayed_workers);
2049 btrfs_destroy_workqueue(fs_info->caching_workers);
2050 btrfs_destroy_workqueue(fs_info->readahead_workers);
2051 btrfs_destroy_workqueue(fs_info->flush_workers);
2052 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2055 static void free_root_extent_buffers(struct btrfs_root *root)
2058 free_extent_buffer(root->node);
2059 free_extent_buffer(root->commit_root);
2061 root->commit_root = NULL;
2065 /* helper to cleanup tree roots */
2066 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2068 free_root_extent_buffers(info->tree_root);
2070 free_root_extent_buffers(info->dev_root);
2071 free_root_extent_buffers(info->extent_root);
2072 free_root_extent_buffers(info->csum_root);
2073 free_root_extent_buffers(info->quota_root);
2074 free_root_extent_buffers(info->uuid_root);
2076 free_root_extent_buffers(info->chunk_root);
2079 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2082 struct btrfs_root *gang[8];
2085 while (!list_empty(&fs_info->dead_roots)) {
2086 gang[0] = list_entry(fs_info->dead_roots.next,
2087 struct btrfs_root, root_list);
2088 list_del(&gang[0]->root_list);
2090 if (gang[0]->in_radix) {
2091 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2093 free_extent_buffer(gang[0]->node);
2094 free_extent_buffer(gang[0]->commit_root);
2095 btrfs_put_fs_root(gang[0]);
2100 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2105 for (i = 0; i < ret; i++)
2106 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2109 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2110 btrfs_free_log_root_tree(NULL, fs_info);
2111 btrfs_destroy_pinned_extent(fs_info->tree_root,
2112 fs_info->pinned_extents);
2116 int open_ctree(struct super_block *sb,
2117 struct btrfs_fs_devices *fs_devices,
2127 struct btrfs_key location;
2128 struct buffer_head *bh;
2129 struct btrfs_super_block *disk_super;
2130 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2131 struct btrfs_root *tree_root;
2132 struct btrfs_root *extent_root;
2133 struct btrfs_root *csum_root;
2134 struct btrfs_root *chunk_root;
2135 struct btrfs_root *dev_root;
2136 struct btrfs_root *quota_root;
2137 struct btrfs_root *uuid_root;
2138 struct btrfs_root *log_tree_root;
2141 int num_backups_tried = 0;
2142 int backup_index = 0;
2144 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2145 bool create_uuid_tree;
2146 bool check_uuid_tree;
2148 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2149 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2150 if (!tree_root || !chunk_root) {
2155 ret = init_srcu_struct(&fs_info->subvol_srcu);
2161 ret = setup_bdi(fs_info, &fs_info->bdi);
2167 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2172 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2173 (1 + ilog2(nr_cpu_ids));
2175 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2178 goto fail_dirty_metadata_bytes;
2181 ret = percpu_counter_init(&fs_info->bio_counter, 0);
2184 goto fail_delalloc_bytes;
2187 fs_info->btree_inode = new_inode(sb);
2188 if (!fs_info->btree_inode) {
2190 goto fail_bio_counter;
2193 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2195 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2196 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2197 INIT_LIST_HEAD(&fs_info->trans_list);
2198 INIT_LIST_HEAD(&fs_info->dead_roots);
2199 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2200 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2201 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2202 spin_lock_init(&fs_info->delalloc_root_lock);
2203 spin_lock_init(&fs_info->trans_lock);
2204 spin_lock_init(&fs_info->fs_roots_radix_lock);
2205 spin_lock_init(&fs_info->delayed_iput_lock);
2206 spin_lock_init(&fs_info->defrag_inodes_lock);
2207 spin_lock_init(&fs_info->free_chunk_lock);
2208 spin_lock_init(&fs_info->tree_mod_seq_lock);
2209 spin_lock_init(&fs_info->super_lock);
2210 spin_lock_init(&fs_info->buffer_lock);
2211 rwlock_init(&fs_info->tree_mod_log_lock);
2212 mutex_init(&fs_info->reloc_mutex);
2213 mutex_init(&fs_info->delalloc_root_mutex);
2214 seqlock_init(&fs_info->profiles_lock);
2216 init_completion(&fs_info->kobj_unregister);
2217 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2218 INIT_LIST_HEAD(&fs_info->space_info);
2219 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2220 btrfs_mapping_init(&fs_info->mapping_tree);
2221 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2222 BTRFS_BLOCK_RSV_GLOBAL);
2223 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2224 BTRFS_BLOCK_RSV_DELALLOC);
2225 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2226 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2227 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2228 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2229 BTRFS_BLOCK_RSV_DELOPS);
2230 atomic_set(&fs_info->nr_async_submits, 0);
2231 atomic_set(&fs_info->async_delalloc_pages, 0);
2232 atomic_set(&fs_info->async_submit_draining, 0);
2233 atomic_set(&fs_info->nr_async_bios, 0);
2234 atomic_set(&fs_info->defrag_running, 0);
2235 atomic64_set(&fs_info->tree_mod_seq, 0);
2237 fs_info->max_inline = 8192 * 1024;
2238 fs_info->metadata_ratio = 0;
2239 fs_info->defrag_inodes = RB_ROOT;
2240 fs_info->free_chunk_space = 0;
2241 fs_info->tree_mod_log = RB_ROOT;
2242 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2243 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2244 /* readahead state */
2245 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2246 spin_lock_init(&fs_info->reada_lock);
2248 fs_info->thread_pool_size = min_t(unsigned long,
2249 num_online_cpus() + 2, 8);
2251 INIT_LIST_HEAD(&fs_info->ordered_roots);
2252 spin_lock_init(&fs_info->ordered_root_lock);
2253 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2255 if (!fs_info->delayed_root) {
2259 btrfs_init_delayed_root(fs_info->delayed_root);
2261 mutex_init(&fs_info->scrub_lock);
2262 atomic_set(&fs_info->scrubs_running, 0);
2263 atomic_set(&fs_info->scrub_pause_req, 0);
2264 atomic_set(&fs_info->scrubs_paused, 0);
2265 atomic_set(&fs_info->scrub_cancel_req, 0);
2266 init_waitqueue_head(&fs_info->replace_wait);
2267 init_waitqueue_head(&fs_info->scrub_pause_wait);
2268 fs_info->scrub_workers_refcnt = 0;
2269 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2270 fs_info->check_integrity_print_mask = 0;
2273 spin_lock_init(&fs_info->balance_lock);
2274 mutex_init(&fs_info->balance_mutex);
2275 atomic_set(&fs_info->balance_running, 0);
2276 atomic_set(&fs_info->balance_pause_req, 0);
2277 atomic_set(&fs_info->balance_cancel_req, 0);
2278 fs_info->balance_ctl = NULL;
2279 init_waitqueue_head(&fs_info->balance_wait_q);
2281 sb->s_blocksize = 4096;
2282 sb->s_blocksize_bits = blksize_bits(4096);
2283 sb->s_bdi = &fs_info->bdi;
2285 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2286 set_nlink(fs_info->btree_inode, 1);
2288 * we set the i_size on the btree inode to the max possible int.
2289 * the real end of the address space is determined by all of
2290 * the devices in the system
2292 fs_info->btree_inode->i_size = OFFSET_MAX;
2293 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2294 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2296 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2297 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2298 fs_info->btree_inode->i_mapping);
2299 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2300 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2302 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2304 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2305 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2306 sizeof(struct btrfs_key));
2307 set_bit(BTRFS_INODE_DUMMY,
2308 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2309 btrfs_insert_inode_hash(fs_info->btree_inode);
2311 spin_lock_init(&fs_info->block_group_cache_lock);
2312 fs_info->block_group_cache_tree = RB_ROOT;
2313 fs_info->first_logical_byte = (u64)-1;
2315 extent_io_tree_init(&fs_info->freed_extents[0],
2316 fs_info->btree_inode->i_mapping);
2317 extent_io_tree_init(&fs_info->freed_extents[1],
2318 fs_info->btree_inode->i_mapping);
2319 fs_info->pinned_extents = &fs_info->freed_extents[0];
2320 fs_info->do_barriers = 1;
2323 mutex_init(&fs_info->ordered_operations_mutex);
2324 mutex_init(&fs_info->ordered_extent_flush_mutex);
2325 mutex_init(&fs_info->tree_log_mutex);
2326 mutex_init(&fs_info->chunk_mutex);
2327 mutex_init(&fs_info->transaction_kthread_mutex);
2328 mutex_init(&fs_info->cleaner_mutex);
2329 mutex_init(&fs_info->volume_mutex);
2330 init_rwsem(&fs_info->extent_commit_sem);
2331 init_rwsem(&fs_info->cleanup_work_sem);
2332 init_rwsem(&fs_info->subvol_sem);
2333 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2334 fs_info->dev_replace.lock_owner = 0;
2335 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2336 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2337 mutex_init(&fs_info->dev_replace.lock_management_lock);
2338 mutex_init(&fs_info->dev_replace.lock);
2340 spin_lock_init(&fs_info->qgroup_lock);
2341 mutex_init(&fs_info->qgroup_ioctl_lock);
2342 fs_info->qgroup_tree = RB_ROOT;
2343 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2344 fs_info->qgroup_seq = 1;
2345 fs_info->quota_enabled = 0;
2346 fs_info->pending_quota_state = 0;
2347 fs_info->qgroup_ulist = NULL;
2348 mutex_init(&fs_info->qgroup_rescan_lock);
2350 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2351 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2353 init_waitqueue_head(&fs_info->transaction_throttle);
2354 init_waitqueue_head(&fs_info->transaction_wait);
2355 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2356 init_waitqueue_head(&fs_info->async_submit_wait);
2358 ret = btrfs_alloc_stripe_hash_table(fs_info);
2364 __setup_root(4096, 4096, 4096, 4096, tree_root,
2365 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2367 invalidate_bdev(fs_devices->latest_bdev);
2370 * Read super block and check the signature bytes only
2372 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2379 * We want to check superblock checksum, the type is stored inside.
2380 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2382 if (btrfs_check_super_csum(bh->b_data)) {
2383 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2389 * super_copy is zeroed at allocation time and we never touch the
2390 * following bytes up to INFO_SIZE, the checksum is calculated from
2391 * the whole block of INFO_SIZE
2393 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2394 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2395 sizeof(*fs_info->super_for_commit));
2398 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2400 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2402 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2407 disk_super = fs_info->super_copy;
2408 if (!btrfs_super_root(disk_super))
2411 /* check FS state, whether FS is broken. */
2412 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2413 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2416 * run through our array of backup supers and setup
2417 * our ring pointer to the oldest one
2419 generation = btrfs_super_generation(disk_super);
2420 find_oldest_super_backup(fs_info, generation);
2423 * In the long term, we'll store the compression type in the super
2424 * block, and it'll be used for per file compression control.
2426 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2428 ret = btrfs_parse_options(tree_root, options);
2434 features = btrfs_super_incompat_flags(disk_super) &
2435 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2437 printk(KERN_ERR "BTRFS: couldn't mount because of "
2438 "unsupported optional features (%Lx).\n",
2444 if (btrfs_super_leafsize(disk_super) !=
2445 btrfs_super_nodesize(disk_super)) {
2446 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2447 "blocksizes don't match. node %d leaf %d\n",
2448 btrfs_super_nodesize(disk_super),
2449 btrfs_super_leafsize(disk_super));
2453 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2454 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2455 "blocksize (%d) was too large\n",
2456 btrfs_super_leafsize(disk_super));
2461 features = btrfs_super_incompat_flags(disk_super);
2462 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2463 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2464 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2466 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2467 printk(KERN_ERR "BTRFS: has skinny extents\n");
2470 * flag our filesystem as having big metadata blocks if
2471 * they are bigger than the page size
2473 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2474 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2475 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2476 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2479 nodesize = btrfs_super_nodesize(disk_super);
2480 leafsize = btrfs_super_leafsize(disk_super);
2481 sectorsize = btrfs_super_sectorsize(disk_super);
2482 stripesize = btrfs_super_stripesize(disk_super);
2483 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2484 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2487 * mixed block groups end up with duplicate but slightly offset
2488 * extent buffers for the same range. It leads to corruptions
2490 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2491 (sectorsize != leafsize)) {
2492 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2493 "are not allowed for mixed block groups on %s\n",
2499 * Needn't use the lock because there is no other task which will
2502 btrfs_set_super_incompat_flags(disk_super, features);
2504 features = btrfs_super_compat_ro_flags(disk_super) &
2505 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2506 if (!(sb->s_flags & MS_RDONLY) && features) {
2507 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2508 "unsupported option features (%Lx).\n",
2514 max_active = fs_info->thread_pool_size;
2517 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2520 fs_info->delalloc_workers =
2521 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2523 fs_info->flush_workers =
2524 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2526 fs_info->caching_workers =
2527 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2530 * a higher idle thresh on the submit workers makes it much more
2531 * likely that bios will be send down in a sane order to the
2534 fs_info->submit_workers =
2535 btrfs_alloc_workqueue("submit", flags,
2536 min_t(u64, fs_devices->num_devices,
2539 fs_info->fixup_workers =
2540 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2543 * endios are largely parallel and should have a very
2546 fs_info->endio_workers =
2547 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2548 fs_info->endio_meta_workers =
2549 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2550 fs_info->endio_meta_write_workers =
2551 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2552 fs_info->endio_raid56_workers =
2553 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2554 fs_info->rmw_workers =
2555 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2556 fs_info->endio_write_workers =
2557 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2558 fs_info->endio_freespace_worker =
2559 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2560 fs_info->delayed_workers =
2561 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2562 fs_info->readahead_workers =
2563 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2564 fs_info->qgroup_rescan_workers =
2565 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2567 if (!(fs_info->workers && fs_info->delalloc_workers &&
2568 fs_info->submit_workers && fs_info->flush_workers &&
2569 fs_info->endio_workers && fs_info->endio_meta_workers &&
2570 fs_info->endio_meta_write_workers &&
2571 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2572 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2573 fs_info->caching_workers && fs_info->readahead_workers &&
2574 fs_info->fixup_workers && fs_info->delayed_workers &&
2575 fs_info->qgroup_rescan_workers)) {
2577 goto fail_sb_buffer;
2580 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2581 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2582 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2584 tree_root->nodesize = nodesize;
2585 tree_root->leafsize = leafsize;
2586 tree_root->sectorsize = sectorsize;
2587 tree_root->stripesize = stripesize;
2589 sb->s_blocksize = sectorsize;
2590 sb->s_blocksize_bits = blksize_bits(sectorsize);
2592 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2593 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2594 goto fail_sb_buffer;
2597 if (sectorsize != PAGE_SIZE) {
2598 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2599 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2600 goto fail_sb_buffer;
2603 mutex_lock(&fs_info->chunk_mutex);
2604 ret = btrfs_read_sys_array(tree_root);
2605 mutex_unlock(&fs_info->chunk_mutex);
2607 printk(KERN_WARNING "BTRFS: failed to read the system "
2608 "array on %s\n", sb->s_id);
2609 goto fail_sb_buffer;
2612 blocksize = btrfs_level_size(tree_root,
2613 btrfs_super_chunk_root_level(disk_super));
2614 generation = btrfs_super_chunk_root_generation(disk_super);
2616 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2617 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2619 chunk_root->node = read_tree_block(chunk_root,
2620 btrfs_super_chunk_root(disk_super),
2621 blocksize, generation);
2622 if (!chunk_root->node ||
2623 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2624 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2626 goto fail_tree_roots;
2628 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2629 chunk_root->commit_root = btrfs_root_node(chunk_root);
2631 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2632 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2634 ret = btrfs_read_chunk_tree(chunk_root);
2636 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2638 goto fail_tree_roots;
2642 * keep the device that is marked to be the target device for the
2643 * dev_replace procedure
2645 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2647 if (!fs_devices->latest_bdev) {
2648 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2650 goto fail_tree_roots;
2654 blocksize = btrfs_level_size(tree_root,
2655 btrfs_super_root_level(disk_super));
2656 generation = btrfs_super_generation(disk_super);
2658 tree_root->node = read_tree_block(tree_root,
2659 btrfs_super_root(disk_super),
2660 blocksize, generation);
2661 if (!tree_root->node ||
2662 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2663 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2666 goto recovery_tree_root;
2669 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2670 tree_root->commit_root = btrfs_root_node(tree_root);
2671 btrfs_set_root_refs(&tree_root->root_item, 1);
2673 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2674 location.type = BTRFS_ROOT_ITEM_KEY;
2675 location.offset = 0;
2677 extent_root = btrfs_read_tree_root(tree_root, &location);
2678 if (IS_ERR(extent_root)) {
2679 ret = PTR_ERR(extent_root);
2680 goto recovery_tree_root;
2682 extent_root->track_dirty = 1;
2683 fs_info->extent_root = extent_root;
2685 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2686 dev_root = btrfs_read_tree_root(tree_root, &location);
2687 if (IS_ERR(dev_root)) {
2688 ret = PTR_ERR(dev_root);
2689 goto recovery_tree_root;
2691 dev_root->track_dirty = 1;
2692 fs_info->dev_root = dev_root;
2693 btrfs_init_devices_late(fs_info);
2695 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2696 csum_root = btrfs_read_tree_root(tree_root, &location);
2697 if (IS_ERR(csum_root)) {
2698 ret = PTR_ERR(csum_root);
2699 goto recovery_tree_root;
2701 csum_root->track_dirty = 1;
2702 fs_info->csum_root = csum_root;
2704 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2705 quota_root = btrfs_read_tree_root(tree_root, &location);
2706 if (!IS_ERR(quota_root)) {
2707 quota_root->track_dirty = 1;
2708 fs_info->quota_enabled = 1;
2709 fs_info->pending_quota_state = 1;
2710 fs_info->quota_root = quota_root;
2713 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2714 uuid_root = btrfs_read_tree_root(tree_root, &location);
2715 if (IS_ERR(uuid_root)) {
2716 ret = PTR_ERR(uuid_root);
2718 goto recovery_tree_root;
2719 create_uuid_tree = true;
2720 check_uuid_tree = false;
2722 uuid_root->track_dirty = 1;
2723 fs_info->uuid_root = uuid_root;
2724 create_uuid_tree = false;
2726 generation != btrfs_super_uuid_tree_generation(disk_super);
2729 fs_info->generation = generation;
2730 fs_info->last_trans_committed = generation;
2732 ret = btrfs_recover_balance(fs_info);
2734 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2735 goto fail_block_groups;
2738 ret = btrfs_init_dev_stats(fs_info);
2740 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2742 goto fail_block_groups;
2745 ret = btrfs_init_dev_replace(fs_info);
2747 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2748 goto fail_block_groups;
2751 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2753 ret = btrfs_sysfs_add_one(fs_info);
2755 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2756 goto fail_block_groups;
2759 ret = btrfs_init_space_info(fs_info);
2761 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2765 ret = btrfs_read_block_groups(extent_root);
2767 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2770 fs_info->num_tolerated_disk_barrier_failures =
2771 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2772 if (fs_info->fs_devices->missing_devices >
2773 fs_info->num_tolerated_disk_barrier_failures &&
2774 !(sb->s_flags & MS_RDONLY)) {
2775 printk(KERN_WARNING "BTRFS: "
2776 "too many missing devices, writeable mount is not allowed\n");
2780 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2782 if (IS_ERR(fs_info->cleaner_kthread))
2785 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2787 "btrfs-transaction");
2788 if (IS_ERR(fs_info->transaction_kthread))
2791 if (!btrfs_test_opt(tree_root, SSD) &&
2792 !btrfs_test_opt(tree_root, NOSSD) &&
2793 !fs_info->fs_devices->rotating) {
2794 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2796 btrfs_set_opt(fs_info->mount_opt, SSD);
2799 /* Set the real inode map cache flag */
2800 if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2801 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2803 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2804 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2805 ret = btrfsic_mount(tree_root, fs_devices,
2806 btrfs_test_opt(tree_root,
2807 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2809 fs_info->check_integrity_print_mask);
2811 printk(KERN_WARNING "BTRFS: failed to initialize"
2812 " integrity check module %s\n", sb->s_id);
2815 ret = btrfs_read_qgroup_config(fs_info);
2817 goto fail_trans_kthread;
2819 /* do not make disk changes in broken FS */
2820 if (btrfs_super_log_root(disk_super) != 0) {
2821 u64 bytenr = btrfs_super_log_root(disk_super);
2823 if (fs_devices->rw_devices == 0) {
2824 printk(KERN_WARNING "BTRFS: log replay required "
2830 btrfs_level_size(tree_root,
2831 btrfs_super_log_root_level(disk_super));
2833 log_tree_root = btrfs_alloc_root(fs_info);
2834 if (!log_tree_root) {
2839 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2840 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2842 log_tree_root->node = read_tree_block(tree_root, bytenr,
2845 if (!log_tree_root->node ||
2846 !extent_buffer_uptodate(log_tree_root->node)) {
2847 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2848 free_extent_buffer(log_tree_root->node);
2849 kfree(log_tree_root);
2850 goto fail_trans_kthread;
2852 /* returns with log_tree_root freed on success */
2853 ret = btrfs_recover_log_trees(log_tree_root);
2855 btrfs_error(tree_root->fs_info, ret,
2856 "Failed to recover log tree");
2857 free_extent_buffer(log_tree_root->node);
2858 kfree(log_tree_root);
2859 goto fail_trans_kthread;
2862 if (sb->s_flags & MS_RDONLY) {
2863 ret = btrfs_commit_super(tree_root);
2865 goto fail_trans_kthread;
2869 ret = btrfs_find_orphan_roots(tree_root);
2871 goto fail_trans_kthread;
2873 if (!(sb->s_flags & MS_RDONLY)) {
2874 ret = btrfs_cleanup_fs_roots(fs_info);
2876 goto fail_trans_kthread;
2878 ret = btrfs_recover_relocation(tree_root);
2881 "BTRFS: failed to recover relocation\n");
2887 location.objectid = BTRFS_FS_TREE_OBJECTID;
2888 location.type = BTRFS_ROOT_ITEM_KEY;
2889 location.offset = 0;
2891 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2892 if (IS_ERR(fs_info->fs_root)) {
2893 err = PTR_ERR(fs_info->fs_root);
2897 if (sb->s_flags & MS_RDONLY)
2900 down_read(&fs_info->cleanup_work_sem);
2901 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2902 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2903 up_read(&fs_info->cleanup_work_sem);
2904 close_ctree(tree_root);
2907 up_read(&fs_info->cleanup_work_sem);
2909 ret = btrfs_resume_balance_async(fs_info);
2911 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2912 close_ctree(tree_root);
2916 ret = btrfs_resume_dev_replace_async(fs_info);
2918 pr_warn("BTRFS: failed to resume dev_replace\n");
2919 close_ctree(tree_root);
2923 btrfs_qgroup_rescan_resume(fs_info);
2925 if (create_uuid_tree) {
2926 pr_info("BTRFS: creating UUID tree\n");
2927 ret = btrfs_create_uuid_tree(fs_info);
2929 pr_warn("BTRFS: failed to create the UUID tree %d\n",
2931 close_ctree(tree_root);
2934 } else if (check_uuid_tree ||
2935 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2936 pr_info("BTRFS: checking UUID tree\n");
2937 ret = btrfs_check_uuid_tree(fs_info);
2939 pr_warn("BTRFS: failed to check the UUID tree %d\n",
2941 close_ctree(tree_root);
2945 fs_info->update_uuid_tree_gen = 1;
2951 btrfs_free_qgroup_config(fs_info);
2953 kthread_stop(fs_info->transaction_kthread);
2954 btrfs_cleanup_transaction(fs_info->tree_root);
2955 del_fs_roots(fs_info);
2957 kthread_stop(fs_info->cleaner_kthread);
2960 * make sure we're done with the btree inode before we stop our
2963 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2966 btrfs_sysfs_remove_one(fs_info);
2969 btrfs_put_block_group_cache(fs_info);
2970 btrfs_free_block_groups(fs_info);
2973 free_root_pointers(fs_info, 1);
2974 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2977 btrfs_stop_all_workers(fs_info);
2980 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2982 iput(fs_info->btree_inode);
2984 percpu_counter_destroy(&fs_info->bio_counter);
2985 fail_delalloc_bytes:
2986 percpu_counter_destroy(&fs_info->delalloc_bytes);
2987 fail_dirty_metadata_bytes:
2988 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2990 bdi_destroy(&fs_info->bdi);
2992 cleanup_srcu_struct(&fs_info->subvol_srcu);
2994 btrfs_free_stripe_hash_table(fs_info);
2995 btrfs_close_devices(fs_info->fs_devices);
2999 if (!btrfs_test_opt(tree_root, RECOVERY))
3000 goto fail_tree_roots;
3002 free_root_pointers(fs_info, 0);
3004 /* don't use the log in recovery mode, it won't be valid */
3005 btrfs_set_super_log_root(disk_super, 0);
3007 /* we can't trust the free space cache either */
3008 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3010 ret = next_root_backup(fs_info, fs_info->super_copy,
3011 &num_backups_tried, &backup_index);
3013 goto fail_block_groups;
3014 goto retry_root_backup;
3017 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3020 set_buffer_uptodate(bh);
3022 struct btrfs_device *device = (struct btrfs_device *)
3025 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3026 "I/O error on %s\n",
3027 rcu_str_deref(device->name));
3028 /* note, we dont' set_buffer_write_io_error because we have
3029 * our own ways of dealing with the IO errors
3031 clear_buffer_uptodate(bh);
3032 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3038 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3040 struct buffer_head *bh;
3041 struct buffer_head *latest = NULL;
3042 struct btrfs_super_block *super;
3047 /* we would like to check all the supers, but that would make
3048 * a btrfs mount succeed after a mkfs from a different FS.
3049 * So, we need to add a special mount option to scan for
3050 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3052 for (i = 0; i < 1; i++) {
3053 bytenr = btrfs_sb_offset(i);
3054 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3055 i_size_read(bdev->bd_inode))
3057 bh = __bread(bdev, bytenr / 4096,
3058 BTRFS_SUPER_INFO_SIZE);
3062 super = (struct btrfs_super_block *)bh->b_data;
3063 if (btrfs_super_bytenr(super) != bytenr ||
3064 btrfs_super_magic(super) != BTRFS_MAGIC) {
3069 if (!latest || btrfs_super_generation(super) > transid) {
3072 transid = btrfs_super_generation(super);
3081 * this should be called twice, once with wait == 0 and
3082 * once with wait == 1. When wait == 0 is done, all the buffer heads
3083 * we write are pinned.
3085 * They are released when wait == 1 is done.
3086 * max_mirrors must be the same for both runs, and it indicates how
3087 * many supers on this one device should be written.
3089 * max_mirrors == 0 means to write them all.
3091 static int write_dev_supers(struct btrfs_device *device,
3092 struct btrfs_super_block *sb,
3093 int do_barriers, int wait, int max_mirrors)
3095 struct buffer_head *bh;
3102 if (max_mirrors == 0)
3103 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3105 for (i = 0; i < max_mirrors; i++) {
3106 bytenr = btrfs_sb_offset(i);
3107 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3111 bh = __find_get_block(device->bdev, bytenr / 4096,
3112 BTRFS_SUPER_INFO_SIZE);
3118 if (!buffer_uptodate(bh))
3121 /* drop our reference */
3124 /* drop the reference from the wait == 0 run */
3128 btrfs_set_super_bytenr(sb, bytenr);
3131 crc = btrfs_csum_data((char *)sb +
3132 BTRFS_CSUM_SIZE, crc,
3133 BTRFS_SUPER_INFO_SIZE -
3135 btrfs_csum_final(crc, sb->csum);
3138 * one reference for us, and we leave it for the
3141 bh = __getblk(device->bdev, bytenr / 4096,
3142 BTRFS_SUPER_INFO_SIZE);
3144 printk(KERN_ERR "BTRFS: couldn't get super "
3145 "buffer head for bytenr %Lu\n", bytenr);
3150 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3152 /* one reference for submit_bh */
3155 set_buffer_uptodate(bh);
3157 bh->b_end_io = btrfs_end_buffer_write_sync;
3158 bh->b_private = device;
3162 * we fua the first super. The others we allow
3166 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3168 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3172 return errors < i ? 0 : -1;
3176 * endio for the write_dev_flush, this will wake anyone waiting
3177 * for the barrier when it is done
3179 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3182 if (err == -EOPNOTSUPP)
3183 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3184 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3186 if (bio->bi_private)
3187 complete(bio->bi_private);
3192 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3193 * sent down. With wait == 1, it waits for the previous flush.
3195 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3198 static int write_dev_flush(struct btrfs_device *device, int wait)
3203 if (device->nobarriers)
3207 bio = device->flush_bio;
3211 wait_for_completion(&device->flush_wait);
3213 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3214 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3215 rcu_str_deref(device->name));
3216 device->nobarriers = 1;
3217 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3219 btrfs_dev_stat_inc_and_print(device,
3220 BTRFS_DEV_STAT_FLUSH_ERRS);
3223 /* drop the reference from the wait == 0 run */
3225 device->flush_bio = NULL;
3231 * one reference for us, and we leave it for the
3234 device->flush_bio = NULL;
3235 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3239 bio->bi_end_io = btrfs_end_empty_barrier;
3240 bio->bi_bdev = device->bdev;
3241 init_completion(&device->flush_wait);
3242 bio->bi_private = &device->flush_wait;
3243 device->flush_bio = bio;
3246 btrfsic_submit_bio(WRITE_FLUSH, bio);
3252 * send an empty flush down to each device in parallel,
3253 * then wait for them
3255 static int barrier_all_devices(struct btrfs_fs_info *info)
3257 struct list_head *head;
3258 struct btrfs_device *dev;
3259 int errors_send = 0;
3260 int errors_wait = 0;
3263 /* send down all the barriers */
3264 head = &info->fs_devices->devices;
3265 list_for_each_entry_rcu(dev, head, dev_list) {
3272 if (!dev->in_fs_metadata || !dev->writeable)
3275 ret = write_dev_flush(dev, 0);
3280 /* wait for all the barriers */
3281 list_for_each_entry_rcu(dev, head, dev_list) {
3288 if (!dev->in_fs_metadata || !dev->writeable)
3291 ret = write_dev_flush(dev, 1);
3295 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3296 errors_wait > info->num_tolerated_disk_barrier_failures)
3301 int btrfs_calc_num_tolerated_disk_barrier_failures(
3302 struct btrfs_fs_info *fs_info)
3304 struct btrfs_ioctl_space_info space;
3305 struct btrfs_space_info *sinfo;
3306 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3307 BTRFS_BLOCK_GROUP_SYSTEM,
3308 BTRFS_BLOCK_GROUP_METADATA,
3309 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3313 int num_tolerated_disk_barrier_failures =
3314 (int)fs_info->fs_devices->num_devices;
3316 for (i = 0; i < num_types; i++) {
3317 struct btrfs_space_info *tmp;
3321 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3322 if (tmp->flags == types[i]) {
3332 down_read(&sinfo->groups_sem);
3333 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3334 if (!list_empty(&sinfo->block_groups[c])) {
3337 btrfs_get_block_group_info(
3338 &sinfo->block_groups[c], &space);
3339 if (space.total_bytes == 0 ||
3340 space.used_bytes == 0)
3342 flags = space.flags;
3345 * 0: if dup, single or RAID0 is configured for
3346 * any of metadata, system or data, else
3347 * 1: if RAID5 is configured, or if RAID1 or
3348 * RAID10 is configured and only two mirrors
3350 * 2: if RAID6 is configured, else
3351 * num_mirrors - 1: if RAID1 or RAID10 is
3352 * configured and more than
3353 * 2 mirrors are used.
3355 if (num_tolerated_disk_barrier_failures > 0 &&
3356 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3357 BTRFS_BLOCK_GROUP_RAID0)) ||
3358 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3360 num_tolerated_disk_barrier_failures = 0;
3361 else if (num_tolerated_disk_barrier_failures > 1) {
3362 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3363 BTRFS_BLOCK_GROUP_RAID5 |
3364 BTRFS_BLOCK_GROUP_RAID10)) {
3365 num_tolerated_disk_barrier_failures = 1;
3367 BTRFS_BLOCK_GROUP_RAID6) {
3368 num_tolerated_disk_barrier_failures = 2;
3373 up_read(&sinfo->groups_sem);
3376 return num_tolerated_disk_barrier_failures;
3379 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3381 struct list_head *head;
3382 struct btrfs_device *dev;
3383 struct btrfs_super_block *sb;
3384 struct btrfs_dev_item *dev_item;
3388 int total_errors = 0;
3391 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3392 backup_super_roots(root->fs_info);
3394 sb = root->fs_info->super_for_commit;
3395 dev_item = &sb->dev_item;
3397 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3398 head = &root->fs_info->fs_devices->devices;
3399 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3402 ret = barrier_all_devices(root->fs_info);
3405 &root->fs_info->fs_devices->device_list_mutex);
3406 btrfs_error(root->fs_info, ret,
3407 "errors while submitting device barriers.");
3412 list_for_each_entry_rcu(dev, head, dev_list) {
3417 if (!dev->in_fs_metadata || !dev->writeable)
3420 btrfs_set_stack_device_generation(dev_item, 0);
3421 btrfs_set_stack_device_type(dev_item, dev->type);
3422 btrfs_set_stack_device_id(dev_item, dev->devid);
3423 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3424 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3425 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3426 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3427 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3428 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3429 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3431 flags = btrfs_super_flags(sb);
3432 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3434 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3438 if (total_errors > max_errors) {
3439 btrfs_err(root->fs_info, "%d errors while writing supers",
3441 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3443 /* FUA is masked off if unsupported and can't be the reason */
3444 btrfs_error(root->fs_info, -EIO,
3445 "%d errors while writing supers", total_errors);
3450 list_for_each_entry_rcu(dev, head, dev_list) {
3453 if (!dev->in_fs_metadata || !dev->writeable)
3456 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3460 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3461 if (total_errors > max_errors) {
3462 btrfs_error(root->fs_info, -EIO,
3463 "%d errors while writing supers", total_errors);
3469 int write_ctree_super(struct btrfs_trans_handle *trans,
3470 struct btrfs_root *root, int max_mirrors)
3472 return write_all_supers(root, max_mirrors);
3475 /* Drop a fs root from the radix tree and free it. */
3476 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3477 struct btrfs_root *root)
3479 spin_lock(&fs_info->fs_roots_radix_lock);
3480 radix_tree_delete(&fs_info->fs_roots_radix,
3481 (unsigned long)root->root_key.objectid);
3482 spin_unlock(&fs_info->fs_roots_radix_lock);
3484 if (btrfs_root_refs(&root->root_item) == 0)
3485 synchronize_srcu(&fs_info->subvol_srcu);
3487 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3488 btrfs_free_log(NULL, root);
3490 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3491 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3495 static void free_fs_root(struct btrfs_root *root)
3497 iput(root->cache_inode);
3498 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3499 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3500 root->orphan_block_rsv = NULL;
3502 free_anon_bdev(root->anon_dev);
3503 if (root->subv_writers)
3504 btrfs_free_subvolume_writers(root->subv_writers);
3505 free_extent_buffer(root->node);
3506 free_extent_buffer(root->commit_root);
3507 kfree(root->free_ino_ctl);
3508 kfree(root->free_ino_pinned);
3510 btrfs_put_fs_root(root);
3513 void btrfs_free_fs_root(struct btrfs_root *root)
3518 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3520 u64 root_objectid = 0;
3521 struct btrfs_root *gang[8];
3526 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3527 (void **)gang, root_objectid,
3532 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3533 for (i = 0; i < ret; i++) {
3536 root_objectid = gang[i]->root_key.objectid;
3537 err = btrfs_orphan_cleanup(gang[i]);
3546 int btrfs_commit_super(struct btrfs_root *root)
3548 struct btrfs_trans_handle *trans;
3550 mutex_lock(&root->fs_info->cleaner_mutex);
3551 btrfs_run_delayed_iputs(root);
3552 mutex_unlock(&root->fs_info->cleaner_mutex);
3553 wake_up_process(root->fs_info->cleaner_kthread);
3555 /* wait until ongoing cleanup work done */
3556 down_write(&root->fs_info->cleanup_work_sem);
3557 up_write(&root->fs_info->cleanup_work_sem);
3559 trans = btrfs_join_transaction(root);
3561 return PTR_ERR(trans);
3562 return btrfs_commit_transaction(trans, root);
3565 int close_ctree(struct btrfs_root *root)
3567 struct btrfs_fs_info *fs_info = root->fs_info;
3570 fs_info->closing = 1;
3573 /* wait for the uuid_scan task to finish */
3574 down(&fs_info->uuid_tree_rescan_sem);
3575 /* avoid complains from lockdep et al., set sem back to initial state */
3576 up(&fs_info->uuid_tree_rescan_sem);
3578 /* pause restriper - we want to resume on mount */
3579 btrfs_pause_balance(fs_info);
3581 btrfs_dev_replace_suspend_for_unmount(fs_info);
3583 btrfs_scrub_cancel(fs_info);
3585 /* wait for any defraggers to finish */
3586 wait_event(fs_info->transaction_wait,
3587 (atomic_read(&fs_info->defrag_running) == 0));
3589 /* clear out the rbtree of defraggable inodes */
3590 btrfs_cleanup_defrag_inodes(fs_info);
3592 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3593 ret = btrfs_commit_super(root);
3595 btrfs_err(root->fs_info, "commit super ret %d", ret);
3598 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3599 btrfs_error_commit_super(root);
3601 kthread_stop(fs_info->transaction_kthread);
3602 kthread_stop(fs_info->cleaner_kthread);
3604 fs_info->closing = 2;
3607 btrfs_free_qgroup_config(root->fs_info);
3609 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3610 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3611 percpu_counter_sum(&fs_info->delalloc_bytes));
3614 btrfs_sysfs_remove_one(fs_info);
3616 del_fs_roots(fs_info);
3618 btrfs_put_block_group_cache(fs_info);
3620 btrfs_free_block_groups(fs_info);
3622 btrfs_stop_all_workers(fs_info);
3624 free_root_pointers(fs_info, 1);
3626 iput(fs_info->btree_inode);
3628 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3629 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3630 btrfsic_unmount(root, fs_info->fs_devices);
3633 btrfs_close_devices(fs_info->fs_devices);
3634 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3636 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3637 percpu_counter_destroy(&fs_info->delalloc_bytes);
3638 percpu_counter_destroy(&fs_info->bio_counter);
3639 bdi_destroy(&fs_info->bdi);
3640 cleanup_srcu_struct(&fs_info->subvol_srcu);
3642 btrfs_free_stripe_hash_table(fs_info);
3644 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3645 root->orphan_block_rsv = NULL;
3650 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3654 struct inode *btree_inode = buf->pages[0]->mapping->host;
3656 ret = extent_buffer_uptodate(buf);
3660 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3661 parent_transid, atomic);
3667 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3669 return set_extent_buffer_uptodate(buf);
3672 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3674 struct btrfs_root *root;
3675 u64 transid = btrfs_header_generation(buf);
3678 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3680 * This is a fast path so only do this check if we have sanity tests
3681 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3682 * outside of the sanity tests.
3684 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3687 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3688 btrfs_assert_tree_locked(buf);
3689 if (transid != root->fs_info->generation)
3690 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3691 "found %llu running %llu\n",
3692 buf->start, transid, root->fs_info->generation);
3693 was_dirty = set_extent_buffer_dirty(buf);
3695 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3697 root->fs_info->dirty_metadata_batch);
3700 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3704 * looks as though older kernels can get into trouble with
3705 * this code, they end up stuck in balance_dirty_pages forever
3709 if (current->flags & PF_MEMALLOC)
3713 btrfs_balance_delayed_items(root);
3715 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3716 BTRFS_DIRTY_METADATA_THRESH);
3718 balance_dirty_pages_ratelimited(
3719 root->fs_info->btree_inode->i_mapping);
3724 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3726 __btrfs_btree_balance_dirty(root, 1);
3729 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3731 __btrfs_btree_balance_dirty(root, 0);
3734 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3736 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3737 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3740 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3744 * Placeholder for checks
3749 static void btrfs_error_commit_super(struct btrfs_root *root)
3751 mutex_lock(&root->fs_info->cleaner_mutex);
3752 btrfs_run_delayed_iputs(root);
3753 mutex_unlock(&root->fs_info->cleaner_mutex);
3755 down_write(&root->fs_info->cleanup_work_sem);
3756 up_write(&root->fs_info->cleanup_work_sem);
3758 /* cleanup FS via transaction */
3759 btrfs_cleanup_transaction(root);
3762 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3763 struct btrfs_root *root)
3765 struct btrfs_inode *btrfs_inode;
3766 struct list_head splice;
3768 INIT_LIST_HEAD(&splice);
3770 mutex_lock(&root->fs_info->ordered_operations_mutex);
3771 spin_lock(&root->fs_info->ordered_root_lock);
3773 list_splice_init(&t->ordered_operations, &splice);
3774 while (!list_empty(&splice)) {
3775 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3776 ordered_operations);
3778 list_del_init(&btrfs_inode->ordered_operations);
3779 spin_unlock(&root->fs_info->ordered_root_lock);
3781 btrfs_invalidate_inodes(btrfs_inode->root);
3783 spin_lock(&root->fs_info->ordered_root_lock);
3786 spin_unlock(&root->fs_info->ordered_root_lock);
3787 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3790 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3792 struct btrfs_ordered_extent *ordered;
3794 spin_lock(&root->ordered_extent_lock);
3796 * This will just short circuit the ordered completion stuff which will
3797 * make sure the ordered extent gets properly cleaned up.
3799 list_for_each_entry(ordered, &root->ordered_extents,
3801 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3802 spin_unlock(&root->ordered_extent_lock);
3805 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3807 struct btrfs_root *root;
3808 struct list_head splice;
3810 INIT_LIST_HEAD(&splice);
3812 spin_lock(&fs_info->ordered_root_lock);
3813 list_splice_init(&fs_info->ordered_roots, &splice);
3814 while (!list_empty(&splice)) {
3815 root = list_first_entry(&splice, struct btrfs_root,
3817 list_move_tail(&root->ordered_root,
3818 &fs_info->ordered_roots);
3820 spin_unlock(&fs_info->ordered_root_lock);
3821 btrfs_destroy_ordered_extents(root);
3824 spin_lock(&fs_info->ordered_root_lock);
3826 spin_unlock(&fs_info->ordered_root_lock);
3829 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3830 struct btrfs_root *root)
3832 struct rb_node *node;
3833 struct btrfs_delayed_ref_root *delayed_refs;
3834 struct btrfs_delayed_ref_node *ref;
3837 delayed_refs = &trans->delayed_refs;
3839 spin_lock(&delayed_refs->lock);
3840 if (atomic_read(&delayed_refs->num_entries) == 0) {
3841 spin_unlock(&delayed_refs->lock);
3842 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3846 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3847 struct btrfs_delayed_ref_head *head;
3848 bool pin_bytes = false;
3850 head = rb_entry(node, struct btrfs_delayed_ref_head,
3852 if (!mutex_trylock(&head->mutex)) {
3853 atomic_inc(&head->node.refs);
3854 spin_unlock(&delayed_refs->lock);
3856 mutex_lock(&head->mutex);
3857 mutex_unlock(&head->mutex);
3858 btrfs_put_delayed_ref(&head->node);
3859 spin_lock(&delayed_refs->lock);
3862 spin_lock(&head->lock);
3863 while ((node = rb_first(&head->ref_root)) != NULL) {
3864 ref = rb_entry(node, struct btrfs_delayed_ref_node,
3867 rb_erase(&ref->rb_node, &head->ref_root);
3868 atomic_dec(&delayed_refs->num_entries);
3869 btrfs_put_delayed_ref(ref);
3871 if (head->must_insert_reserved)
3873 btrfs_free_delayed_extent_op(head->extent_op);
3874 delayed_refs->num_heads--;
3875 if (head->processing == 0)
3876 delayed_refs->num_heads_ready--;
3877 atomic_dec(&delayed_refs->num_entries);
3878 head->node.in_tree = 0;
3879 rb_erase(&head->href_node, &delayed_refs->href_root);
3880 spin_unlock(&head->lock);
3881 spin_unlock(&delayed_refs->lock);
3882 mutex_unlock(&head->mutex);
3885 btrfs_pin_extent(root, head->node.bytenr,
3886 head->node.num_bytes, 1);
3887 btrfs_put_delayed_ref(&head->node);
3889 spin_lock(&delayed_refs->lock);
3892 spin_unlock(&delayed_refs->lock);
3897 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3899 struct btrfs_inode *btrfs_inode;
3900 struct list_head splice;
3902 INIT_LIST_HEAD(&splice);
3904 spin_lock(&root->delalloc_lock);
3905 list_splice_init(&root->delalloc_inodes, &splice);
3907 while (!list_empty(&splice)) {
3908 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3911 list_del_init(&btrfs_inode->delalloc_inodes);
3912 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3913 &btrfs_inode->runtime_flags);
3914 spin_unlock(&root->delalloc_lock);
3916 btrfs_invalidate_inodes(btrfs_inode->root);
3918 spin_lock(&root->delalloc_lock);
3921 spin_unlock(&root->delalloc_lock);
3924 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3926 struct btrfs_root *root;
3927 struct list_head splice;
3929 INIT_LIST_HEAD(&splice);
3931 spin_lock(&fs_info->delalloc_root_lock);
3932 list_splice_init(&fs_info->delalloc_roots, &splice);
3933 while (!list_empty(&splice)) {
3934 root = list_first_entry(&splice, struct btrfs_root,
3936 list_del_init(&root->delalloc_root);
3937 root = btrfs_grab_fs_root(root);
3939 spin_unlock(&fs_info->delalloc_root_lock);
3941 btrfs_destroy_delalloc_inodes(root);
3942 btrfs_put_fs_root(root);
3944 spin_lock(&fs_info->delalloc_root_lock);
3946 spin_unlock(&fs_info->delalloc_root_lock);
3949 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3950 struct extent_io_tree *dirty_pages,
3954 struct extent_buffer *eb;
3959 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3964 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3965 while (start <= end) {
3966 eb = btrfs_find_tree_block(root, start,
3968 start += root->leafsize;
3971 wait_on_extent_buffer_writeback(eb);
3973 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3975 clear_extent_buffer_dirty(eb);
3976 free_extent_buffer_stale(eb);
3983 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3984 struct extent_io_tree *pinned_extents)
3986 struct extent_io_tree *unpin;
3992 unpin = pinned_extents;
3995 ret = find_first_extent_bit(unpin, 0, &start, &end,
3996 EXTENT_DIRTY, NULL);
4001 if (btrfs_test_opt(root, DISCARD))
4002 ret = btrfs_error_discard_extent(root, start,
4006 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4007 btrfs_error_unpin_extent_range(root, start, end);
4012 if (unpin == &root->fs_info->freed_extents[0])
4013 unpin = &root->fs_info->freed_extents[1];
4015 unpin = &root->fs_info->freed_extents[0];
4023 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4024 struct btrfs_root *root)
4026 btrfs_destroy_ordered_operations(cur_trans, root);
4028 btrfs_destroy_delayed_refs(cur_trans, root);
4030 cur_trans->state = TRANS_STATE_COMMIT_START;
4031 wake_up(&root->fs_info->transaction_blocked_wait);
4033 cur_trans->state = TRANS_STATE_UNBLOCKED;
4034 wake_up(&root->fs_info->transaction_wait);
4036 btrfs_destroy_delayed_inodes(root);
4037 btrfs_assert_delayed_root_empty(root);
4039 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4041 btrfs_destroy_pinned_extent(root,
4042 root->fs_info->pinned_extents);
4044 cur_trans->state =TRANS_STATE_COMPLETED;
4045 wake_up(&cur_trans->commit_wait);
4048 memset(cur_trans, 0, sizeof(*cur_trans));
4049 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4053 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4055 struct btrfs_transaction *t;
4057 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4059 spin_lock(&root->fs_info->trans_lock);
4060 while (!list_empty(&root->fs_info->trans_list)) {
4061 t = list_first_entry(&root->fs_info->trans_list,
4062 struct btrfs_transaction, list);
4063 if (t->state >= TRANS_STATE_COMMIT_START) {
4064 atomic_inc(&t->use_count);
4065 spin_unlock(&root->fs_info->trans_lock);
4066 btrfs_wait_for_commit(root, t->transid);
4067 btrfs_put_transaction(t);
4068 spin_lock(&root->fs_info->trans_lock);
4071 if (t == root->fs_info->running_transaction) {
4072 t->state = TRANS_STATE_COMMIT_DOING;
4073 spin_unlock(&root->fs_info->trans_lock);
4075 * We wait for 0 num_writers since we don't hold a trans
4076 * handle open currently for this transaction.
4078 wait_event(t->writer_wait,
4079 atomic_read(&t->num_writers) == 0);
4081 spin_unlock(&root->fs_info->trans_lock);
4083 btrfs_cleanup_one_transaction(t, root);
4085 spin_lock(&root->fs_info->trans_lock);
4086 if (t == root->fs_info->running_transaction)
4087 root->fs_info->running_transaction = NULL;
4088 list_del_init(&t->list);
4089 spin_unlock(&root->fs_info->trans_lock);
4091 btrfs_put_transaction(t);
4092 trace_btrfs_transaction_commit(root);
4093 spin_lock(&root->fs_info->trans_lock);
4095 spin_unlock(&root->fs_info->trans_lock);
4096 btrfs_destroy_all_ordered_extents(root->fs_info);
4097 btrfs_destroy_delayed_inodes(root);
4098 btrfs_assert_delayed_root_empty(root);
4099 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4100 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4101 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4106 static struct extent_io_ops btree_extent_io_ops = {
4107 .readpage_end_io_hook = btree_readpage_end_io_hook,
4108 .readpage_io_failed_hook = btree_io_failed_hook,
4109 .submit_bio_hook = btree_submit_bio_hook,
4110 /* note we're sharing with inode.c for the merge bio hook */
4111 .merge_bio_hook = btrfs_merge_bio_hook,