2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <linux/uuid.h>
34 #include <linux/semaphore.h>
35 #include <asm/unaligned.h>
39 #include "transaction.h"
40 #include "btrfs_inode.h"
42 #include "print-tree.h"
43 #include "async-thread.h"
46 #include "free-space-cache.h"
47 #include "inode-map.h"
48 #include "check-integrity.h"
49 #include "rcu-string.h"
50 #include "dev-replace.h"
54 #include <asm/cpufeature.h>
57 static struct extent_io_ops btree_extent_io_ops;
58 static void end_workqueue_fn(struct btrfs_work *work);
59 static void free_fs_root(struct btrfs_root *root);
60 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
62 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
63 struct btrfs_root *root);
64 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
65 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
66 struct btrfs_root *root);
67 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
68 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
69 struct extent_io_tree *dirty_pages,
71 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
72 struct extent_io_tree *pinned_extents);
73 static int btrfs_cleanup_transaction(struct btrfs_root *root);
74 static void btrfs_error_commit_super(struct btrfs_root *root);
77 * end_io_wq structs are used to do processing in task context when an IO is
78 * complete. This is used during reads to verify checksums, and it is used
79 * by writes to insert metadata for new file extents after IO is complete.
85 struct btrfs_fs_info *info;
88 struct list_head list;
89 struct btrfs_work work;
93 * async submit bios are used to offload expensive checksumming
94 * onto the worker threads. They checksum file and metadata bios
95 * just before they are sent down the IO stack.
97 struct async_submit_bio {
100 struct list_head list;
101 extent_submit_bio_hook_t *submit_bio_start;
102 extent_submit_bio_hook_t *submit_bio_done;
105 unsigned long bio_flags;
107 * bio_offset is optional, can be used if the pages in the bio
108 * can't tell us where in the file the bio should go
111 struct btrfs_work work;
116 * Lockdep class keys for extent_buffer->lock's in this root. For a given
117 * eb, the lockdep key is determined by the btrfs_root it belongs to and
118 * the level the eb occupies in the tree.
120 * Different roots are used for different purposes and may nest inside each
121 * other and they require separate keysets. As lockdep keys should be
122 * static, assign keysets according to the purpose of the root as indicated
123 * by btrfs_root->objectid. This ensures that all special purpose roots
124 * have separate keysets.
126 * Lock-nesting across peer nodes is always done with the immediate parent
127 * node locked thus preventing deadlock. As lockdep doesn't know this, use
128 * subclass to avoid triggering lockdep warning in such cases.
130 * The key is set by the readpage_end_io_hook after the buffer has passed
131 * csum validation but before the pages are unlocked. It is also set by
132 * btrfs_init_new_buffer on freshly allocated blocks.
134 * We also add a check to make sure the highest level of the tree is the
135 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
136 * needs update as well.
138 #ifdef CONFIG_DEBUG_LOCK_ALLOC
139 # if BTRFS_MAX_LEVEL != 8
143 static struct btrfs_lockdep_keyset {
144 u64 id; /* root objectid */
145 const char *name_stem; /* lock name stem */
146 char names[BTRFS_MAX_LEVEL + 1][20];
147 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
148 } btrfs_lockdep_keysets[] = {
149 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
150 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
151 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
152 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
153 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
154 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
155 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
156 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
157 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
158 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
159 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
160 { .id = 0, .name_stem = "tree" },
163 void __init btrfs_init_lockdep(void)
167 /* initialize lockdep class names */
168 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
169 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
171 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
172 snprintf(ks->names[j], sizeof(ks->names[j]),
173 "btrfs-%s-%02d", ks->name_stem, j);
177 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
180 struct btrfs_lockdep_keyset *ks;
182 BUG_ON(level >= ARRAY_SIZE(ks->keys));
184 /* find the matching keyset, id 0 is the default entry */
185 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
186 if (ks->id == objectid)
189 lockdep_set_class_and_name(&eb->lock,
190 &ks->keys[level], ks->names[level]);
196 * extents on the btree inode are pretty simple, there's one extent
197 * that covers the entire device
199 static struct extent_map *btree_get_extent(struct inode *inode,
200 struct page *page, size_t pg_offset, u64 start, u64 len,
203 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
204 struct extent_map *em;
207 read_lock(&em_tree->lock);
208 em = lookup_extent_mapping(em_tree, start, len);
211 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
212 read_unlock(&em_tree->lock);
215 read_unlock(&em_tree->lock);
217 em = alloc_extent_map();
219 em = ERR_PTR(-ENOMEM);
224 em->block_len = (u64)-1;
226 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
228 write_lock(&em_tree->lock);
229 ret = add_extent_mapping(em_tree, em, 0);
230 if (ret == -EEXIST) {
232 em = lookup_extent_mapping(em_tree, start, len);
239 write_unlock(&em_tree->lock);
245 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
247 return crc32c(seed, data, len);
250 void btrfs_csum_final(u32 crc, char *result)
252 put_unaligned_le32(~crc, result);
256 * compute the csum for a btree block, and either verify it or write it
257 * into the csum field of the block.
259 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
262 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
265 unsigned long cur_len;
266 unsigned long offset = BTRFS_CSUM_SIZE;
268 unsigned long map_start;
269 unsigned long map_len;
272 unsigned long inline_result;
274 len = buf->len - offset;
276 err = map_private_extent_buffer(buf, offset, 32,
277 &kaddr, &map_start, &map_len);
280 cur_len = min(len, map_len - (offset - map_start));
281 crc = btrfs_csum_data(kaddr + offset - map_start,
286 if (csum_size > sizeof(inline_result)) {
287 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
291 result = (char *)&inline_result;
294 btrfs_csum_final(crc, result);
297 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
300 memcpy(&found, result, csum_size);
302 read_extent_buffer(buf, &val, 0, csum_size);
303 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
304 "failed on %llu wanted %X found %X "
306 root->fs_info->sb->s_id, buf->start,
307 val, found, btrfs_header_level(buf));
308 if (result != (char *)&inline_result)
313 write_extent_buffer(buf, result, 0, csum_size);
315 if (result != (char *)&inline_result)
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
326 static int verify_parent_transid(struct extent_io_tree *io_tree,
327 struct extent_buffer *eb, u64 parent_transid,
330 struct extent_state *cached_state = NULL;
333 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
339 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
341 if (extent_buffer_uptodate(eb) &&
342 btrfs_header_generation(eb) == parent_transid) {
346 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
348 eb->start, parent_transid, btrfs_header_generation(eb));
350 clear_extent_buffer_uptodate(eb);
352 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
353 &cached_state, GFP_NOFS);
358 * Return 0 if the superblock checksum type matches the checksum value of that
359 * algorithm. Pass the raw disk superblock data.
361 static int btrfs_check_super_csum(char *raw_disk_sb)
363 struct btrfs_super_block *disk_sb =
364 (struct btrfs_super_block *)raw_disk_sb;
365 u16 csum_type = btrfs_super_csum_type(disk_sb);
368 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
370 const int csum_size = sizeof(crc);
371 char result[csum_size];
374 * The super_block structure does not span the whole
375 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
376 * is filled with zeros and is included in the checkum.
378 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
379 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
380 btrfs_csum_final(crc, result);
382 if (memcmp(raw_disk_sb, result, csum_size))
385 if (ret && btrfs_super_generation(disk_sb) < 10) {
386 printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
391 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
392 printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
401 * helper to read a given tree block, doing retries as required when
402 * the checksums don't match and we have alternate mirrors to try.
404 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
405 struct extent_buffer *eb,
406 u64 start, u64 parent_transid)
408 struct extent_io_tree *io_tree;
413 int failed_mirror = 0;
415 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
416 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
418 ret = read_extent_buffer_pages(io_tree, eb, start,
420 btree_get_extent, mirror_num);
422 if (!verify_parent_transid(io_tree, eb,
430 * This buffer's crc is fine, but its contents are corrupted, so
431 * there is no reason to read the other copies, they won't be
434 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
437 num_copies = btrfs_num_copies(root->fs_info,
442 if (!failed_mirror) {
444 failed_mirror = eb->read_mirror;
448 if (mirror_num == failed_mirror)
451 if (mirror_num > num_copies)
455 if (failed && !ret && failed_mirror)
456 repair_eb_io_failure(root, eb, failed_mirror);
462 * checksum a dirty tree block before IO. This has extra checks to make sure
463 * we only fill in the checksum field in the first page of a multi-page block
466 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
468 struct extent_io_tree *tree;
469 u64 start = page_offset(page);
471 struct extent_buffer *eb;
473 tree = &BTRFS_I(page->mapping->host)->io_tree;
475 eb = (struct extent_buffer *)page->private;
476 if (page != eb->pages[0])
478 found_start = btrfs_header_bytenr(eb);
479 if (found_start != start) {
483 if (!PageUptodate(page)) {
487 csum_tree_block(root, eb, 0);
491 static int check_tree_block_fsid(struct btrfs_root *root,
492 struct extent_buffer *eb)
494 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
495 u8 fsid[BTRFS_UUID_SIZE];
498 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
500 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
504 fs_devices = fs_devices->seed;
509 #define CORRUPT(reason, eb, root, slot) \
510 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
511 "root=%llu, slot=%d\n", reason, \
512 btrfs_header_bytenr(eb), root->objectid, slot)
514 static noinline int check_leaf(struct btrfs_root *root,
515 struct extent_buffer *leaf)
517 struct btrfs_key key;
518 struct btrfs_key leaf_key;
519 u32 nritems = btrfs_header_nritems(leaf);
525 /* Check the 0 item */
526 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
527 BTRFS_LEAF_DATA_SIZE(root)) {
528 CORRUPT("invalid item offset size pair", leaf, root, 0);
533 * Check to make sure each items keys are in the correct order and their
534 * offsets make sense. We only have to loop through nritems-1 because
535 * we check the current slot against the next slot, which verifies the
536 * next slot's offset+size makes sense and that the current's slot
539 for (slot = 0; slot < nritems - 1; slot++) {
540 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
541 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
543 /* Make sure the keys are in the right order */
544 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
545 CORRUPT("bad key order", leaf, root, slot);
550 * Make sure the offset and ends are right, remember that the
551 * item data starts at the end of the leaf and grows towards the
554 if (btrfs_item_offset_nr(leaf, slot) !=
555 btrfs_item_end_nr(leaf, slot + 1)) {
556 CORRUPT("slot offset bad", leaf, root, slot);
561 * Check to make sure that we don't point outside of the leaf,
562 * just incase all the items are consistent to eachother, but
563 * all point outside of the leaf.
565 if (btrfs_item_end_nr(leaf, slot) >
566 BTRFS_LEAF_DATA_SIZE(root)) {
567 CORRUPT("slot end outside of leaf", leaf, root, slot);
575 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
576 u64 phy_offset, struct page *page,
577 u64 start, u64 end, int mirror)
579 struct extent_io_tree *tree;
582 struct extent_buffer *eb;
583 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
590 tree = &BTRFS_I(page->mapping->host)->io_tree;
591 eb = (struct extent_buffer *)page->private;
593 /* the pending IO might have been the only thing that kept this buffer
594 * in memory. Make sure we have a ref for all this other checks
596 extent_buffer_get(eb);
598 reads_done = atomic_dec_and_test(&eb->io_pages);
602 eb->read_mirror = mirror;
603 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
608 found_start = btrfs_header_bytenr(eb);
609 if (found_start != eb->start) {
610 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
612 found_start, eb->start);
616 if (check_tree_block_fsid(root, eb)) {
617 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
622 found_level = btrfs_header_level(eb);
623 if (found_level >= BTRFS_MAX_LEVEL) {
624 btrfs_info(root->fs_info, "bad tree block level %d\n",
625 (int)btrfs_header_level(eb));
630 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
633 ret = csum_tree_block(root, eb, 1);
640 * If this is a leaf block and it is corrupt, set the corrupt bit so
641 * that we don't try and read the other copies of this block, just
644 if (found_level == 0 && check_leaf(root, eb)) {
645 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
650 set_extent_buffer_uptodate(eb);
653 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
654 btree_readahead_hook(root, eb, eb->start, ret);
658 * our io error hook is going to dec the io pages
659 * again, we have to make sure it has something
662 atomic_inc(&eb->io_pages);
663 clear_extent_buffer_uptodate(eb);
665 free_extent_buffer(eb);
670 static int btree_io_failed_hook(struct page *page, int failed_mirror)
672 struct extent_buffer *eb;
673 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
675 eb = (struct extent_buffer *)page->private;
676 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
677 eb->read_mirror = failed_mirror;
678 atomic_dec(&eb->io_pages);
679 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
680 btree_readahead_hook(root, eb, eb->start, -EIO);
681 return -EIO; /* we fixed nothing */
684 static void end_workqueue_bio(struct bio *bio, int err)
686 struct end_io_wq *end_io_wq = bio->bi_private;
687 struct btrfs_fs_info *fs_info;
689 fs_info = end_io_wq->info;
690 end_io_wq->error = err;
691 end_io_wq->work.func = end_workqueue_fn;
692 end_io_wq->work.flags = 0;
694 if (bio->bi_rw & REQ_WRITE) {
695 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
696 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
698 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
699 btrfs_queue_worker(&fs_info->endio_freespace_worker,
701 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
702 btrfs_queue_worker(&fs_info->endio_raid56_workers,
705 btrfs_queue_worker(&fs_info->endio_write_workers,
708 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
709 btrfs_queue_worker(&fs_info->endio_raid56_workers,
711 else if (end_io_wq->metadata)
712 btrfs_queue_worker(&fs_info->endio_meta_workers,
715 btrfs_queue_worker(&fs_info->endio_workers,
721 * For the metadata arg you want
724 * 1 - if normal metadta
725 * 2 - if writing to the free space cache area
726 * 3 - raid parity work
728 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
731 struct end_io_wq *end_io_wq;
732 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
736 end_io_wq->private = bio->bi_private;
737 end_io_wq->end_io = bio->bi_end_io;
738 end_io_wq->info = info;
739 end_io_wq->error = 0;
740 end_io_wq->bio = bio;
741 end_io_wq->metadata = metadata;
743 bio->bi_private = end_io_wq;
744 bio->bi_end_io = end_workqueue_bio;
748 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
750 unsigned long limit = min_t(unsigned long,
751 info->workers.max_workers,
752 info->fs_devices->open_devices);
756 static void run_one_async_start(struct btrfs_work *work)
758 struct async_submit_bio *async;
761 async = container_of(work, struct async_submit_bio, work);
762 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
763 async->mirror_num, async->bio_flags,
769 static void run_one_async_done(struct btrfs_work *work)
771 struct btrfs_fs_info *fs_info;
772 struct async_submit_bio *async;
775 async = container_of(work, struct async_submit_bio, work);
776 fs_info = BTRFS_I(async->inode)->root->fs_info;
778 limit = btrfs_async_submit_limit(fs_info);
779 limit = limit * 2 / 3;
781 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
782 waitqueue_active(&fs_info->async_submit_wait))
783 wake_up(&fs_info->async_submit_wait);
785 /* If an error occured we just want to clean up the bio and move on */
787 bio_endio(async->bio, async->error);
791 async->submit_bio_done(async->inode, async->rw, async->bio,
792 async->mirror_num, async->bio_flags,
796 static void run_one_async_free(struct btrfs_work *work)
798 struct async_submit_bio *async;
800 async = container_of(work, struct async_submit_bio, work);
804 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
805 int rw, struct bio *bio, int mirror_num,
806 unsigned long bio_flags,
808 extent_submit_bio_hook_t *submit_bio_start,
809 extent_submit_bio_hook_t *submit_bio_done)
811 struct async_submit_bio *async;
813 async = kmalloc(sizeof(*async), GFP_NOFS);
817 async->inode = inode;
820 async->mirror_num = mirror_num;
821 async->submit_bio_start = submit_bio_start;
822 async->submit_bio_done = submit_bio_done;
824 async->work.func = run_one_async_start;
825 async->work.ordered_func = run_one_async_done;
826 async->work.ordered_free = run_one_async_free;
828 async->work.flags = 0;
829 async->bio_flags = bio_flags;
830 async->bio_offset = bio_offset;
834 atomic_inc(&fs_info->nr_async_submits);
837 btrfs_set_work_high_prio(&async->work);
839 btrfs_queue_worker(&fs_info->workers, &async->work);
841 while (atomic_read(&fs_info->async_submit_draining) &&
842 atomic_read(&fs_info->nr_async_submits)) {
843 wait_event(fs_info->async_submit_wait,
844 (atomic_read(&fs_info->nr_async_submits) == 0));
850 static int btree_csum_one_bio(struct bio *bio)
852 struct bio_vec *bvec = bio->bi_io_vec;
854 struct btrfs_root *root;
857 WARN_ON(bio->bi_vcnt <= 0);
858 while (bio_index < bio->bi_vcnt) {
859 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
860 ret = csum_dirty_buffer(root, bvec->bv_page);
869 static int __btree_submit_bio_start(struct inode *inode, int rw,
870 struct bio *bio, int mirror_num,
871 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 return btree_csum_one_bio(bio);
881 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
882 int mirror_num, unsigned long bio_flags,
888 * when we're called for a write, we're already in the async
889 * submission context. Just jump into btrfs_map_bio
891 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
897 static int check_async_write(struct inode *inode, unsigned long bio_flags)
899 if (bio_flags & EXTENT_BIO_TREE_LOG)
908 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
909 int mirror_num, unsigned long bio_flags,
912 int async = check_async_write(inode, bio_flags);
915 if (!(rw & REQ_WRITE)) {
917 * called for a read, do the setup so that checksum validation
918 * can happen in the async kernel threads
920 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
924 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
927 ret = btree_csum_one_bio(bio);
930 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
934 * kthread helpers are used to submit writes so that
935 * checksumming can happen in parallel across all CPUs
937 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
938 inode, rw, bio, mirror_num, 0,
940 __btree_submit_bio_start,
941 __btree_submit_bio_done);
951 #ifdef CONFIG_MIGRATION
952 static int btree_migratepage(struct address_space *mapping,
953 struct page *newpage, struct page *page,
954 enum migrate_mode mode)
957 * we can't safely write a btree page from here,
958 * we haven't done the locking hook
963 * Buffers may be managed in a filesystem specific way.
964 * We must have no buffers or drop them.
966 if (page_has_private(page) &&
967 !try_to_release_page(page, GFP_KERNEL))
969 return migrate_page(mapping, newpage, page, mode);
974 static int btree_writepages(struct address_space *mapping,
975 struct writeback_control *wbc)
977 struct extent_io_tree *tree;
978 struct btrfs_fs_info *fs_info;
981 tree = &BTRFS_I(mapping->host)->io_tree;
982 if (wbc->sync_mode == WB_SYNC_NONE) {
984 if (wbc->for_kupdate)
987 fs_info = BTRFS_I(mapping->host)->root->fs_info;
988 /* this is a bit racy, but that's ok */
989 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
990 BTRFS_DIRTY_METADATA_THRESH);
994 return btree_write_cache_pages(mapping, wbc);
997 static int btree_readpage(struct file *file, struct page *page)
999 struct extent_io_tree *tree;
1000 tree = &BTRFS_I(page->mapping->host)->io_tree;
1001 return extent_read_full_page(tree, page, btree_get_extent, 0);
1004 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1006 if (PageWriteback(page) || PageDirty(page))
1009 return try_release_extent_buffer(page);
1012 static void btree_invalidatepage(struct page *page, unsigned int offset,
1013 unsigned int length)
1015 struct extent_io_tree *tree;
1016 tree = &BTRFS_I(page->mapping->host)->io_tree;
1017 extent_invalidatepage(tree, page, offset);
1018 btree_releasepage(page, GFP_NOFS);
1019 if (PagePrivate(page)) {
1020 printk(KERN_WARNING "btrfs warning page private not zero "
1021 "on page %llu\n", (unsigned long long)page_offset(page));
1022 ClearPagePrivate(page);
1023 set_page_private(page, 0);
1024 page_cache_release(page);
1028 static int btree_set_page_dirty(struct page *page)
1031 struct extent_buffer *eb;
1033 BUG_ON(!PagePrivate(page));
1034 eb = (struct extent_buffer *)page->private;
1036 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1037 BUG_ON(!atomic_read(&eb->refs));
1038 btrfs_assert_tree_locked(eb);
1040 return __set_page_dirty_nobuffers(page);
1043 static const struct address_space_operations btree_aops = {
1044 .readpage = btree_readpage,
1045 .writepages = btree_writepages,
1046 .releasepage = btree_releasepage,
1047 .invalidatepage = btree_invalidatepage,
1048 #ifdef CONFIG_MIGRATION
1049 .migratepage = btree_migratepage,
1051 .set_page_dirty = btree_set_page_dirty,
1054 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1057 struct extent_buffer *buf = NULL;
1058 struct inode *btree_inode = root->fs_info->btree_inode;
1061 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1064 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1065 buf, 0, WAIT_NONE, btree_get_extent, 0);
1066 free_extent_buffer(buf);
1070 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1071 int mirror_num, struct extent_buffer **eb)
1073 struct extent_buffer *buf = NULL;
1074 struct inode *btree_inode = root->fs_info->btree_inode;
1075 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1078 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1082 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1084 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1085 btree_get_extent, mirror_num);
1087 free_extent_buffer(buf);
1091 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1092 free_extent_buffer(buf);
1094 } else if (extent_buffer_uptodate(buf)) {
1097 free_extent_buffer(buf);
1102 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1103 u64 bytenr, u32 blocksize)
1105 struct inode *btree_inode = root->fs_info->btree_inode;
1106 struct extent_buffer *eb;
1107 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1112 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1113 u64 bytenr, u32 blocksize)
1115 struct inode *btree_inode = root->fs_info->btree_inode;
1116 struct extent_buffer *eb;
1118 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1124 int btrfs_write_tree_block(struct extent_buffer *buf)
1126 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1127 buf->start + buf->len - 1);
1130 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1132 return filemap_fdatawait_range(buf->pages[0]->mapping,
1133 buf->start, buf->start + buf->len - 1);
1136 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1137 u32 blocksize, u64 parent_transid)
1139 struct extent_buffer *buf = NULL;
1142 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1146 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1148 free_extent_buffer(buf);
1155 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1156 struct extent_buffer *buf)
1158 struct btrfs_fs_info *fs_info = root->fs_info;
1160 if (btrfs_header_generation(buf) ==
1161 fs_info->running_transaction->transid) {
1162 btrfs_assert_tree_locked(buf);
1164 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1165 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1167 fs_info->dirty_metadata_batch);
1168 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1169 btrfs_set_lock_blocking(buf);
1170 clear_extent_buffer_dirty(buf);
1175 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1176 u32 stripesize, struct btrfs_root *root,
1177 struct btrfs_fs_info *fs_info,
1181 root->commit_root = NULL;
1182 root->sectorsize = sectorsize;
1183 root->nodesize = nodesize;
1184 root->leafsize = leafsize;
1185 root->stripesize = stripesize;
1187 root->track_dirty = 0;
1189 root->orphan_item_inserted = 0;
1190 root->orphan_cleanup_state = 0;
1192 root->objectid = objectid;
1193 root->last_trans = 0;
1194 root->highest_objectid = 0;
1195 root->nr_delalloc_inodes = 0;
1196 root->nr_ordered_extents = 0;
1198 root->inode_tree = RB_ROOT;
1199 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1200 root->block_rsv = NULL;
1201 root->orphan_block_rsv = NULL;
1203 INIT_LIST_HEAD(&root->dirty_list);
1204 INIT_LIST_HEAD(&root->root_list);
1205 INIT_LIST_HEAD(&root->delalloc_inodes);
1206 INIT_LIST_HEAD(&root->delalloc_root);
1207 INIT_LIST_HEAD(&root->ordered_extents);
1208 INIT_LIST_HEAD(&root->ordered_root);
1209 INIT_LIST_HEAD(&root->logged_list[0]);
1210 INIT_LIST_HEAD(&root->logged_list[1]);
1211 spin_lock_init(&root->orphan_lock);
1212 spin_lock_init(&root->inode_lock);
1213 spin_lock_init(&root->delalloc_lock);
1214 spin_lock_init(&root->ordered_extent_lock);
1215 spin_lock_init(&root->accounting_lock);
1216 spin_lock_init(&root->log_extents_lock[0]);
1217 spin_lock_init(&root->log_extents_lock[1]);
1218 mutex_init(&root->objectid_mutex);
1219 mutex_init(&root->log_mutex);
1220 init_waitqueue_head(&root->log_writer_wait);
1221 init_waitqueue_head(&root->log_commit_wait[0]);
1222 init_waitqueue_head(&root->log_commit_wait[1]);
1223 atomic_set(&root->log_commit[0], 0);
1224 atomic_set(&root->log_commit[1], 0);
1225 atomic_set(&root->log_writers, 0);
1226 atomic_set(&root->log_batch, 0);
1227 atomic_set(&root->orphan_inodes, 0);
1228 atomic_set(&root->refs, 1);
1229 root->log_transid = 0;
1230 root->last_log_commit = 0;
1232 extent_io_tree_init(&root->dirty_log_pages,
1233 fs_info->btree_inode->i_mapping);
1235 memset(&root->root_key, 0, sizeof(root->root_key));
1236 memset(&root->root_item, 0, sizeof(root->root_item));
1237 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1238 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1240 root->defrag_trans_start = fs_info->generation;
1242 root->defrag_trans_start = 0;
1243 init_completion(&root->kobj_unregister);
1244 root->defrag_running = 0;
1245 root->root_key.objectid = objectid;
1248 spin_lock_init(&root->root_item_lock);
1251 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1253 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1255 root->fs_info = fs_info;
1259 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1260 /* Should only be used by the testing infrastructure */
1261 struct btrfs_root *btrfs_alloc_dummy_root(void)
1263 struct btrfs_root *root;
1265 root = btrfs_alloc_root(NULL);
1267 return ERR_PTR(-ENOMEM);
1268 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1269 root->dummy_root = 1;
1275 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1276 struct btrfs_fs_info *fs_info,
1279 struct extent_buffer *leaf;
1280 struct btrfs_root *tree_root = fs_info->tree_root;
1281 struct btrfs_root *root;
1282 struct btrfs_key key;
1287 root = btrfs_alloc_root(fs_info);
1289 return ERR_PTR(-ENOMEM);
1291 __setup_root(tree_root->nodesize, tree_root->leafsize,
1292 tree_root->sectorsize, tree_root->stripesize,
1293 root, fs_info, objectid);
1294 root->root_key.objectid = objectid;
1295 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1296 root->root_key.offset = 0;
1298 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1299 0, objectid, NULL, 0, 0, 0);
1301 ret = PTR_ERR(leaf);
1306 bytenr = leaf->start;
1307 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1308 btrfs_set_header_bytenr(leaf, leaf->start);
1309 btrfs_set_header_generation(leaf, trans->transid);
1310 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1311 btrfs_set_header_owner(leaf, objectid);
1314 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1316 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1317 btrfs_header_chunk_tree_uuid(leaf),
1319 btrfs_mark_buffer_dirty(leaf);
1321 root->commit_root = btrfs_root_node(root);
1322 root->track_dirty = 1;
1325 root->root_item.flags = 0;
1326 root->root_item.byte_limit = 0;
1327 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1328 btrfs_set_root_generation(&root->root_item, trans->transid);
1329 btrfs_set_root_level(&root->root_item, 0);
1330 btrfs_set_root_refs(&root->root_item, 1);
1331 btrfs_set_root_used(&root->root_item, leaf->len);
1332 btrfs_set_root_last_snapshot(&root->root_item, 0);
1333 btrfs_set_root_dirid(&root->root_item, 0);
1335 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1336 root->root_item.drop_level = 0;
1338 key.objectid = objectid;
1339 key.type = BTRFS_ROOT_ITEM_KEY;
1341 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1345 btrfs_tree_unlock(leaf);
1351 btrfs_tree_unlock(leaf);
1352 free_extent_buffer(leaf);
1356 return ERR_PTR(ret);
1359 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1360 struct btrfs_fs_info *fs_info)
1362 struct btrfs_root *root;
1363 struct btrfs_root *tree_root = fs_info->tree_root;
1364 struct extent_buffer *leaf;
1366 root = btrfs_alloc_root(fs_info);
1368 return ERR_PTR(-ENOMEM);
1370 __setup_root(tree_root->nodesize, tree_root->leafsize,
1371 tree_root->sectorsize, tree_root->stripesize,
1372 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1374 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1375 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1376 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1378 * log trees do not get reference counted because they go away
1379 * before a real commit is actually done. They do store pointers
1380 * to file data extents, and those reference counts still get
1381 * updated (along with back refs to the log tree).
1385 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1386 BTRFS_TREE_LOG_OBJECTID, NULL,
1390 return ERR_CAST(leaf);
1393 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1394 btrfs_set_header_bytenr(leaf, leaf->start);
1395 btrfs_set_header_generation(leaf, trans->transid);
1396 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1397 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1400 write_extent_buffer(root->node, root->fs_info->fsid,
1401 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1402 btrfs_mark_buffer_dirty(root->node);
1403 btrfs_tree_unlock(root->node);
1407 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1408 struct btrfs_fs_info *fs_info)
1410 struct btrfs_root *log_root;
1412 log_root = alloc_log_tree(trans, fs_info);
1413 if (IS_ERR(log_root))
1414 return PTR_ERR(log_root);
1415 WARN_ON(fs_info->log_root_tree);
1416 fs_info->log_root_tree = log_root;
1420 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1421 struct btrfs_root *root)
1423 struct btrfs_root *log_root;
1424 struct btrfs_inode_item *inode_item;
1426 log_root = alloc_log_tree(trans, root->fs_info);
1427 if (IS_ERR(log_root))
1428 return PTR_ERR(log_root);
1430 log_root->last_trans = trans->transid;
1431 log_root->root_key.offset = root->root_key.objectid;
1433 inode_item = &log_root->root_item.inode;
1434 btrfs_set_stack_inode_generation(inode_item, 1);
1435 btrfs_set_stack_inode_size(inode_item, 3);
1436 btrfs_set_stack_inode_nlink(inode_item, 1);
1437 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1438 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1440 btrfs_set_root_node(&log_root->root_item, log_root->node);
1442 WARN_ON(root->log_root);
1443 root->log_root = log_root;
1444 root->log_transid = 0;
1445 root->last_log_commit = 0;
1449 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1450 struct btrfs_key *key)
1452 struct btrfs_root *root;
1453 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1454 struct btrfs_path *path;
1459 path = btrfs_alloc_path();
1461 return ERR_PTR(-ENOMEM);
1463 root = btrfs_alloc_root(fs_info);
1469 __setup_root(tree_root->nodesize, tree_root->leafsize,
1470 tree_root->sectorsize, tree_root->stripesize,
1471 root, fs_info, key->objectid);
1473 ret = btrfs_find_root(tree_root, key, path,
1474 &root->root_item, &root->root_key);
1481 generation = btrfs_root_generation(&root->root_item);
1482 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1483 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1484 blocksize, generation);
1488 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1492 root->commit_root = btrfs_root_node(root);
1494 btrfs_free_path(path);
1498 free_extent_buffer(root->node);
1502 root = ERR_PTR(ret);
1506 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1507 struct btrfs_key *location)
1509 struct btrfs_root *root;
1511 root = btrfs_read_tree_root(tree_root, location);
1515 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1517 btrfs_check_and_init_root_item(&root->root_item);
1523 int btrfs_init_fs_root(struct btrfs_root *root)
1527 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1528 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1530 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1535 btrfs_init_free_ino_ctl(root);
1536 mutex_init(&root->fs_commit_mutex);
1537 spin_lock_init(&root->cache_lock);
1538 init_waitqueue_head(&root->cache_wait);
1540 ret = get_anon_bdev(&root->anon_dev);
1545 kfree(root->free_ino_ctl);
1546 kfree(root->free_ino_pinned);
1550 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1553 struct btrfs_root *root;
1555 spin_lock(&fs_info->fs_roots_radix_lock);
1556 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1557 (unsigned long)root_id);
1558 spin_unlock(&fs_info->fs_roots_radix_lock);
1562 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1563 struct btrfs_root *root)
1567 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1571 spin_lock(&fs_info->fs_roots_radix_lock);
1572 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1573 (unsigned long)root->root_key.objectid,
1577 spin_unlock(&fs_info->fs_roots_radix_lock);
1578 radix_tree_preload_end();
1583 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1584 struct btrfs_key *location,
1587 struct btrfs_root *root;
1590 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1591 return fs_info->tree_root;
1592 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1593 return fs_info->extent_root;
1594 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1595 return fs_info->chunk_root;
1596 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1597 return fs_info->dev_root;
1598 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1599 return fs_info->csum_root;
1600 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1601 return fs_info->quota_root ? fs_info->quota_root :
1603 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1604 return fs_info->uuid_root ? fs_info->uuid_root :
1607 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1609 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1610 return ERR_PTR(-ENOENT);
1614 root = btrfs_read_fs_root(fs_info->tree_root, location);
1618 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1623 ret = btrfs_init_fs_root(root);
1627 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1631 root->orphan_item_inserted = 1;
1633 ret = btrfs_insert_fs_root(fs_info, root);
1635 if (ret == -EEXIST) {
1644 return ERR_PTR(ret);
1647 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1649 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1651 struct btrfs_device *device;
1652 struct backing_dev_info *bdi;
1655 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1658 bdi = blk_get_backing_dev_info(device->bdev);
1659 if (bdi && bdi_congested(bdi, bdi_bits)) {
1669 * If this fails, caller must call bdi_destroy() to get rid of the
1672 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1676 bdi->capabilities = BDI_CAP_MAP_COPY;
1677 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1681 bdi->ra_pages = default_backing_dev_info.ra_pages;
1682 bdi->congested_fn = btrfs_congested_fn;
1683 bdi->congested_data = info;
1688 * called by the kthread helper functions to finally call the bio end_io
1689 * functions. This is where read checksum verification actually happens
1691 static void end_workqueue_fn(struct btrfs_work *work)
1694 struct end_io_wq *end_io_wq;
1695 struct btrfs_fs_info *fs_info;
1698 end_io_wq = container_of(work, struct end_io_wq, work);
1699 bio = end_io_wq->bio;
1700 fs_info = end_io_wq->info;
1702 error = end_io_wq->error;
1703 bio->bi_private = end_io_wq->private;
1704 bio->bi_end_io = end_io_wq->end_io;
1706 bio_endio(bio, error);
1709 static int cleaner_kthread(void *arg)
1711 struct btrfs_root *root = arg;
1717 /* Make the cleaner go to sleep early. */
1718 if (btrfs_need_cleaner_sleep(root))
1721 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1725 * Avoid the problem that we change the status of the fs
1726 * during the above check and trylock.
1728 if (btrfs_need_cleaner_sleep(root)) {
1729 mutex_unlock(&root->fs_info->cleaner_mutex);
1733 btrfs_run_delayed_iputs(root);
1734 again = btrfs_clean_one_deleted_snapshot(root);
1735 mutex_unlock(&root->fs_info->cleaner_mutex);
1738 * The defragger has dealt with the R/O remount and umount,
1739 * needn't do anything special here.
1741 btrfs_run_defrag_inodes(root->fs_info);
1743 if (!try_to_freeze() && !again) {
1744 set_current_state(TASK_INTERRUPTIBLE);
1745 if (!kthread_should_stop())
1747 __set_current_state(TASK_RUNNING);
1749 } while (!kthread_should_stop());
1753 static int transaction_kthread(void *arg)
1755 struct btrfs_root *root = arg;
1756 struct btrfs_trans_handle *trans;
1757 struct btrfs_transaction *cur;
1760 unsigned long delay;
1764 cannot_commit = false;
1765 delay = HZ * root->fs_info->commit_interval;
1766 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1768 spin_lock(&root->fs_info->trans_lock);
1769 cur = root->fs_info->running_transaction;
1771 spin_unlock(&root->fs_info->trans_lock);
1775 now = get_seconds();
1776 if (cur->state < TRANS_STATE_BLOCKED &&
1777 (now < cur->start_time ||
1778 now - cur->start_time < root->fs_info->commit_interval)) {
1779 spin_unlock(&root->fs_info->trans_lock);
1783 transid = cur->transid;
1784 spin_unlock(&root->fs_info->trans_lock);
1786 /* If the file system is aborted, this will always fail. */
1787 trans = btrfs_attach_transaction(root);
1788 if (IS_ERR(trans)) {
1789 if (PTR_ERR(trans) != -ENOENT)
1790 cannot_commit = true;
1793 if (transid == trans->transid) {
1794 btrfs_commit_transaction(trans, root);
1796 btrfs_end_transaction(trans, root);
1799 wake_up_process(root->fs_info->cleaner_kthread);
1800 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1802 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1803 &root->fs_info->fs_state)))
1804 btrfs_cleanup_transaction(root);
1805 if (!try_to_freeze()) {
1806 set_current_state(TASK_INTERRUPTIBLE);
1807 if (!kthread_should_stop() &&
1808 (!btrfs_transaction_blocked(root->fs_info) ||
1810 schedule_timeout(delay);
1811 __set_current_state(TASK_RUNNING);
1813 } while (!kthread_should_stop());
1818 * this will find the highest generation in the array of
1819 * root backups. The index of the highest array is returned,
1820 * or -1 if we can't find anything.
1822 * We check to make sure the array is valid by comparing the
1823 * generation of the latest root in the array with the generation
1824 * in the super block. If they don't match we pitch it.
1826 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1829 int newest_index = -1;
1830 struct btrfs_root_backup *root_backup;
1833 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1834 root_backup = info->super_copy->super_roots + i;
1835 cur = btrfs_backup_tree_root_gen(root_backup);
1836 if (cur == newest_gen)
1840 /* check to see if we actually wrapped around */
1841 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1842 root_backup = info->super_copy->super_roots;
1843 cur = btrfs_backup_tree_root_gen(root_backup);
1844 if (cur == newest_gen)
1847 return newest_index;
1852 * find the oldest backup so we know where to store new entries
1853 * in the backup array. This will set the backup_root_index
1854 * field in the fs_info struct
1856 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1859 int newest_index = -1;
1861 newest_index = find_newest_super_backup(info, newest_gen);
1862 /* if there was garbage in there, just move along */
1863 if (newest_index == -1) {
1864 info->backup_root_index = 0;
1866 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1871 * copy all the root pointers into the super backup array.
1872 * this will bump the backup pointer by one when it is
1875 static void backup_super_roots(struct btrfs_fs_info *info)
1878 struct btrfs_root_backup *root_backup;
1881 next_backup = info->backup_root_index;
1882 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1883 BTRFS_NUM_BACKUP_ROOTS;
1886 * just overwrite the last backup if we're at the same generation
1887 * this happens only at umount
1889 root_backup = info->super_for_commit->super_roots + last_backup;
1890 if (btrfs_backup_tree_root_gen(root_backup) ==
1891 btrfs_header_generation(info->tree_root->node))
1892 next_backup = last_backup;
1894 root_backup = info->super_for_commit->super_roots + next_backup;
1897 * make sure all of our padding and empty slots get zero filled
1898 * regardless of which ones we use today
1900 memset(root_backup, 0, sizeof(*root_backup));
1902 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1904 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1905 btrfs_set_backup_tree_root_gen(root_backup,
1906 btrfs_header_generation(info->tree_root->node));
1908 btrfs_set_backup_tree_root_level(root_backup,
1909 btrfs_header_level(info->tree_root->node));
1911 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1912 btrfs_set_backup_chunk_root_gen(root_backup,
1913 btrfs_header_generation(info->chunk_root->node));
1914 btrfs_set_backup_chunk_root_level(root_backup,
1915 btrfs_header_level(info->chunk_root->node));
1917 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1918 btrfs_set_backup_extent_root_gen(root_backup,
1919 btrfs_header_generation(info->extent_root->node));
1920 btrfs_set_backup_extent_root_level(root_backup,
1921 btrfs_header_level(info->extent_root->node));
1924 * we might commit during log recovery, which happens before we set
1925 * the fs_root. Make sure it is valid before we fill it in.
1927 if (info->fs_root && info->fs_root->node) {
1928 btrfs_set_backup_fs_root(root_backup,
1929 info->fs_root->node->start);
1930 btrfs_set_backup_fs_root_gen(root_backup,
1931 btrfs_header_generation(info->fs_root->node));
1932 btrfs_set_backup_fs_root_level(root_backup,
1933 btrfs_header_level(info->fs_root->node));
1936 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1937 btrfs_set_backup_dev_root_gen(root_backup,
1938 btrfs_header_generation(info->dev_root->node));
1939 btrfs_set_backup_dev_root_level(root_backup,
1940 btrfs_header_level(info->dev_root->node));
1942 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1943 btrfs_set_backup_csum_root_gen(root_backup,
1944 btrfs_header_generation(info->csum_root->node));
1945 btrfs_set_backup_csum_root_level(root_backup,
1946 btrfs_header_level(info->csum_root->node));
1948 btrfs_set_backup_total_bytes(root_backup,
1949 btrfs_super_total_bytes(info->super_copy));
1950 btrfs_set_backup_bytes_used(root_backup,
1951 btrfs_super_bytes_used(info->super_copy));
1952 btrfs_set_backup_num_devices(root_backup,
1953 btrfs_super_num_devices(info->super_copy));
1956 * if we don't copy this out to the super_copy, it won't get remembered
1957 * for the next commit
1959 memcpy(&info->super_copy->super_roots,
1960 &info->super_for_commit->super_roots,
1961 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1965 * this copies info out of the root backup array and back into
1966 * the in-memory super block. It is meant to help iterate through
1967 * the array, so you send it the number of backups you've already
1968 * tried and the last backup index you used.
1970 * this returns -1 when it has tried all the backups
1972 static noinline int next_root_backup(struct btrfs_fs_info *info,
1973 struct btrfs_super_block *super,
1974 int *num_backups_tried, int *backup_index)
1976 struct btrfs_root_backup *root_backup;
1977 int newest = *backup_index;
1979 if (*num_backups_tried == 0) {
1980 u64 gen = btrfs_super_generation(super);
1982 newest = find_newest_super_backup(info, gen);
1986 *backup_index = newest;
1987 *num_backups_tried = 1;
1988 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1989 /* we've tried all the backups, all done */
1992 /* jump to the next oldest backup */
1993 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1994 BTRFS_NUM_BACKUP_ROOTS;
1995 *backup_index = newest;
1996 *num_backups_tried += 1;
1998 root_backup = super->super_roots + newest;
2000 btrfs_set_super_generation(super,
2001 btrfs_backup_tree_root_gen(root_backup));
2002 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2003 btrfs_set_super_root_level(super,
2004 btrfs_backup_tree_root_level(root_backup));
2005 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2008 * fixme: the total bytes and num_devices need to match or we should
2011 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2012 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2016 /* helper to cleanup workers */
2017 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2019 btrfs_stop_workers(&fs_info->generic_worker);
2020 btrfs_stop_workers(&fs_info->fixup_workers);
2021 btrfs_stop_workers(&fs_info->delalloc_workers);
2022 btrfs_stop_workers(&fs_info->workers);
2023 btrfs_stop_workers(&fs_info->endio_workers);
2024 btrfs_stop_workers(&fs_info->endio_meta_workers);
2025 btrfs_stop_workers(&fs_info->endio_raid56_workers);
2026 btrfs_stop_workers(&fs_info->rmw_workers);
2027 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2028 btrfs_stop_workers(&fs_info->endio_write_workers);
2029 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2030 btrfs_stop_workers(&fs_info->submit_workers);
2031 btrfs_stop_workers(&fs_info->delayed_workers);
2032 btrfs_stop_workers(&fs_info->caching_workers);
2033 btrfs_stop_workers(&fs_info->readahead_workers);
2034 btrfs_stop_workers(&fs_info->flush_workers);
2035 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
2038 /* helper to cleanup tree roots */
2039 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2041 free_extent_buffer(info->tree_root->node);
2042 free_extent_buffer(info->tree_root->commit_root);
2043 info->tree_root->node = NULL;
2044 info->tree_root->commit_root = NULL;
2046 if (info->dev_root) {
2047 free_extent_buffer(info->dev_root->node);
2048 free_extent_buffer(info->dev_root->commit_root);
2049 info->dev_root->node = NULL;
2050 info->dev_root->commit_root = NULL;
2052 if (info->extent_root) {
2053 free_extent_buffer(info->extent_root->node);
2054 free_extent_buffer(info->extent_root->commit_root);
2055 info->extent_root->node = NULL;
2056 info->extent_root->commit_root = NULL;
2058 if (info->csum_root) {
2059 free_extent_buffer(info->csum_root->node);
2060 free_extent_buffer(info->csum_root->commit_root);
2061 info->csum_root->node = NULL;
2062 info->csum_root->commit_root = NULL;
2064 if (info->quota_root) {
2065 free_extent_buffer(info->quota_root->node);
2066 free_extent_buffer(info->quota_root->commit_root);
2067 info->quota_root->node = NULL;
2068 info->quota_root->commit_root = NULL;
2070 if (info->uuid_root) {
2071 free_extent_buffer(info->uuid_root->node);
2072 free_extent_buffer(info->uuid_root->commit_root);
2073 info->uuid_root->node = NULL;
2074 info->uuid_root->commit_root = NULL;
2077 free_extent_buffer(info->chunk_root->node);
2078 free_extent_buffer(info->chunk_root->commit_root);
2079 info->chunk_root->node = NULL;
2080 info->chunk_root->commit_root = NULL;
2084 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2087 struct btrfs_root *gang[8];
2090 while (!list_empty(&fs_info->dead_roots)) {
2091 gang[0] = list_entry(fs_info->dead_roots.next,
2092 struct btrfs_root, root_list);
2093 list_del(&gang[0]->root_list);
2095 if (gang[0]->in_radix) {
2096 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2098 free_extent_buffer(gang[0]->node);
2099 free_extent_buffer(gang[0]->commit_root);
2100 btrfs_put_fs_root(gang[0]);
2105 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2110 for (i = 0; i < ret; i++)
2111 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2115 int open_ctree(struct super_block *sb,
2116 struct btrfs_fs_devices *fs_devices,
2126 struct btrfs_key location;
2127 struct buffer_head *bh;
2128 struct btrfs_super_block *disk_super;
2129 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2130 struct btrfs_root *tree_root;
2131 struct btrfs_root *extent_root;
2132 struct btrfs_root *csum_root;
2133 struct btrfs_root *chunk_root;
2134 struct btrfs_root *dev_root;
2135 struct btrfs_root *quota_root;
2136 struct btrfs_root *uuid_root;
2137 struct btrfs_root *log_tree_root;
2140 int num_backups_tried = 0;
2141 int backup_index = 0;
2142 bool create_uuid_tree;
2143 bool check_uuid_tree;
2145 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2146 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2147 if (!tree_root || !chunk_root) {
2152 ret = init_srcu_struct(&fs_info->subvol_srcu);
2158 ret = setup_bdi(fs_info, &fs_info->bdi);
2164 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2169 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2170 (1 + ilog2(nr_cpu_ids));
2172 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2175 goto fail_dirty_metadata_bytes;
2178 fs_info->btree_inode = new_inode(sb);
2179 if (!fs_info->btree_inode) {
2181 goto fail_delalloc_bytes;
2184 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2186 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2187 INIT_LIST_HEAD(&fs_info->trans_list);
2188 INIT_LIST_HEAD(&fs_info->dead_roots);
2189 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2190 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2191 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2192 spin_lock_init(&fs_info->delalloc_root_lock);
2193 spin_lock_init(&fs_info->trans_lock);
2194 spin_lock_init(&fs_info->fs_roots_radix_lock);
2195 spin_lock_init(&fs_info->delayed_iput_lock);
2196 spin_lock_init(&fs_info->defrag_inodes_lock);
2197 spin_lock_init(&fs_info->free_chunk_lock);
2198 spin_lock_init(&fs_info->tree_mod_seq_lock);
2199 spin_lock_init(&fs_info->super_lock);
2200 rwlock_init(&fs_info->tree_mod_log_lock);
2201 mutex_init(&fs_info->reloc_mutex);
2202 seqlock_init(&fs_info->profiles_lock);
2204 init_completion(&fs_info->kobj_unregister);
2205 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2206 INIT_LIST_HEAD(&fs_info->space_info);
2207 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2208 btrfs_mapping_init(&fs_info->mapping_tree);
2209 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2210 BTRFS_BLOCK_RSV_GLOBAL);
2211 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2212 BTRFS_BLOCK_RSV_DELALLOC);
2213 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2214 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2215 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2216 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2217 BTRFS_BLOCK_RSV_DELOPS);
2218 atomic_set(&fs_info->nr_async_submits, 0);
2219 atomic_set(&fs_info->async_delalloc_pages, 0);
2220 atomic_set(&fs_info->async_submit_draining, 0);
2221 atomic_set(&fs_info->nr_async_bios, 0);
2222 atomic_set(&fs_info->defrag_running, 0);
2223 atomic64_set(&fs_info->tree_mod_seq, 0);
2225 fs_info->max_inline = 8192 * 1024;
2226 fs_info->metadata_ratio = 0;
2227 fs_info->defrag_inodes = RB_ROOT;
2228 fs_info->free_chunk_space = 0;
2229 fs_info->tree_mod_log = RB_ROOT;
2230 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2232 /* readahead state */
2233 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2234 spin_lock_init(&fs_info->reada_lock);
2236 fs_info->thread_pool_size = min_t(unsigned long,
2237 num_online_cpus() + 2, 8);
2239 INIT_LIST_HEAD(&fs_info->ordered_roots);
2240 spin_lock_init(&fs_info->ordered_root_lock);
2241 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2243 if (!fs_info->delayed_root) {
2247 btrfs_init_delayed_root(fs_info->delayed_root);
2249 mutex_init(&fs_info->scrub_lock);
2250 atomic_set(&fs_info->scrubs_running, 0);
2251 atomic_set(&fs_info->scrub_pause_req, 0);
2252 atomic_set(&fs_info->scrubs_paused, 0);
2253 atomic_set(&fs_info->scrub_cancel_req, 0);
2254 init_waitqueue_head(&fs_info->scrub_pause_wait);
2255 init_rwsem(&fs_info->scrub_super_lock);
2256 fs_info->scrub_workers_refcnt = 0;
2257 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2258 fs_info->check_integrity_print_mask = 0;
2261 spin_lock_init(&fs_info->balance_lock);
2262 mutex_init(&fs_info->balance_mutex);
2263 atomic_set(&fs_info->balance_running, 0);
2264 atomic_set(&fs_info->balance_pause_req, 0);
2265 atomic_set(&fs_info->balance_cancel_req, 0);
2266 fs_info->balance_ctl = NULL;
2267 init_waitqueue_head(&fs_info->balance_wait_q);
2269 sb->s_blocksize = 4096;
2270 sb->s_blocksize_bits = blksize_bits(4096);
2271 sb->s_bdi = &fs_info->bdi;
2273 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2274 set_nlink(fs_info->btree_inode, 1);
2276 * we set the i_size on the btree inode to the max possible int.
2277 * the real end of the address space is determined by all of
2278 * the devices in the system
2280 fs_info->btree_inode->i_size = OFFSET_MAX;
2281 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2282 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2284 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2285 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2286 fs_info->btree_inode->i_mapping);
2287 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2288 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2290 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2292 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2293 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2294 sizeof(struct btrfs_key));
2295 set_bit(BTRFS_INODE_DUMMY,
2296 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2297 insert_inode_hash(fs_info->btree_inode);
2299 spin_lock_init(&fs_info->block_group_cache_lock);
2300 fs_info->block_group_cache_tree = RB_ROOT;
2301 fs_info->first_logical_byte = (u64)-1;
2303 extent_io_tree_init(&fs_info->freed_extents[0],
2304 fs_info->btree_inode->i_mapping);
2305 extent_io_tree_init(&fs_info->freed_extents[1],
2306 fs_info->btree_inode->i_mapping);
2307 fs_info->pinned_extents = &fs_info->freed_extents[0];
2308 fs_info->do_barriers = 1;
2311 mutex_init(&fs_info->ordered_operations_mutex);
2312 mutex_init(&fs_info->ordered_extent_flush_mutex);
2313 mutex_init(&fs_info->tree_log_mutex);
2314 mutex_init(&fs_info->chunk_mutex);
2315 mutex_init(&fs_info->transaction_kthread_mutex);
2316 mutex_init(&fs_info->cleaner_mutex);
2317 mutex_init(&fs_info->volume_mutex);
2318 init_rwsem(&fs_info->extent_commit_sem);
2319 init_rwsem(&fs_info->cleanup_work_sem);
2320 init_rwsem(&fs_info->subvol_sem);
2321 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2322 fs_info->dev_replace.lock_owner = 0;
2323 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2324 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2325 mutex_init(&fs_info->dev_replace.lock_management_lock);
2326 mutex_init(&fs_info->dev_replace.lock);
2328 spin_lock_init(&fs_info->qgroup_lock);
2329 mutex_init(&fs_info->qgroup_ioctl_lock);
2330 fs_info->qgroup_tree = RB_ROOT;
2331 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2332 fs_info->qgroup_seq = 1;
2333 fs_info->quota_enabled = 0;
2334 fs_info->pending_quota_state = 0;
2335 fs_info->qgroup_ulist = NULL;
2336 mutex_init(&fs_info->qgroup_rescan_lock);
2338 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2339 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2341 init_waitqueue_head(&fs_info->transaction_throttle);
2342 init_waitqueue_head(&fs_info->transaction_wait);
2343 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2344 init_waitqueue_head(&fs_info->async_submit_wait);
2346 ret = btrfs_alloc_stripe_hash_table(fs_info);
2352 __setup_root(4096, 4096, 4096, 4096, tree_root,
2353 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2355 invalidate_bdev(fs_devices->latest_bdev);
2358 * Read super block and check the signature bytes only
2360 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2367 * We want to check superblock checksum, the type is stored inside.
2368 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2370 if (btrfs_check_super_csum(bh->b_data)) {
2371 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2377 * super_copy is zeroed at allocation time and we never touch the
2378 * following bytes up to INFO_SIZE, the checksum is calculated from
2379 * the whole block of INFO_SIZE
2381 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2382 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2383 sizeof(*fs_info->super_for_commit));
2386 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2388 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2390 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2395 disk_super = fs_info->super_copy;
2396 if (!btrfs_super_root(disk_super))
2399 /* check FS state, whether FS is broken. */
2400 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2401 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2404 * run through our array of backup supers and setup
2405 * our ring pointer to the oldest one
2407 generation = btrfs_super_generation(disk_super);
2408 find_oldest_super_backup(fs_info, generation);
2411 * In the long term, we'll store the compression type in the super
2412 * block, and it'll be used for per file compression control.
2414 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2416 ret = btrfs_parse_options(tree_root, options);
2422 features = btrfs_super_incompat_flags(disk_super) &
2423 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2425 printk(KERN_ERR "BTRFS: couldn't mount because of "
2426 "unsupported optional features (%Lx).\n",
2432 if (btrfs_super_leafsize(disk_super) !=
2433 btrfs_super_nodesize(disk_super)) {
2434 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2435 "blocksizes don't match. node %d leaf %d\n",
2436 btrfs_super_nodesize(disk_super),
2437 btrfs_super_leafsize(disk_super));
2441 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2442 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2443 "blocksize (%d) was too large\n",
2444 btrfs_super_leafsize(disk_super));
2449 features = btrfs_super_incompat_flags(disk_super);
2450 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2451 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2452 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2454 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2455 printk(KERN_ERR "btrfs: has skinny extents\n");
2458 * flag our filesystem as having big metadata blocks if
2459 * they are bigger than the page size
2461 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2462 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2463 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2464 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2467 nodesize = btrfs_super_nodesize(disk_super);
2468 leafsize = btrfs_super_leafsize(disk_super);
2469 sectorsize = btrfs_super_sectorsize(disk_super);
2470 stripesize = btrfs_super_stripesize(disk_super);
2471 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2472 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2475 * mixed block groups end up with duplicate but slightly offset
2476 * extent buffers for the same range. It leads to corruptions
2478 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2479 (sectorsize != leafsize)) {
2480 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2481 "are not allowed for mixed block groups on %s\n",
2487 * Needn't use the lock because there is no other task which will
2490 btrfs_set_super_incompat_flags(disk_super, features);
2492 features = btrfs_super_compat_ro_flags(disk_super) &
2493 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2494 if (!(sb->s_flags & MS_RDONLY) && features) {
2495 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2496 "unsupported option features (%Lx).\n",
2502 btrfs_init_workers(&fs_info->generic_worker,
2503 "genwork", 1, NULL);
2505 btrfs_init_workers(&fs_info->workers, "worker",
2506 fs_info->thread_pool_size,
2507 &fs_info->generic_worker);
2509 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2510 fs_info->thread_pool_size, NULL);
2512 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2513 fs_info->thread_pool_size, NULL);
2515 btrfs_init_workers(&fs_info->submit_workers, "submit",
2516 min_t(u64, fs_devices->num_devices,
2517 fs_info->thread_pool_size), NULL);
2519 btrfs_init_workers(&fs_info->caching_workers, "cache",
2520 fs_info->thread_pool_size, NULL);
2522 /* a higher idle thresh on the submit workers makes it much more
2523 * likely that bios will be send down in a sane order to the
2526 fs_info->submit_workers.idle_thresh = 64;
2528 fs_info->workers.idle_thresh = 16;
2529 fs_info->workers.ordered = 1;
2531 fs_info->delalloc_workers.idle_thresh = 2;
2532 fs_info->delalloc_workers.ordered = 1;
2534 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2535 &fs_info->generic_worker);
2536 btrfs_init_workers(&fs_info->endio_workers, "endio",
2537 fs_info->thread_pool_size,
2538 &fs_info->generic_worker);
2539 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2540 fs_info->thread_pool_size,
2541 &fs_info->generic_worker);
2542 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2543 "endio-meta-write", fs_info->thread_pool_size,
2544 &fs_info->generic_worker);
2545 btrfs_init_workers(&fs_info->endio_raid56_workers,
2546 "endio-raid56", fs_info->thread_pool_size,
2547 &fs_info->generic_worker);
2548 btrfs_init_workers(&fs_info->rmw_workers,
2549 "rmw", fs_info->thread_pool_size,
2550 &fs_info->generic_worker);
2551 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2552 fs_info->thread_pool_size,
2553 &fs_info->generic_worker);
2554 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2555 1, &fs_info->generic_worker);
2556 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2557 fs_info->thread_pool_size,
2558 &fs_info->generic_worker);
2559 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2560 fs_info->thread_pool_size,
2561 &fs_info->generic_worker);
2562 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2563 &fs_info->generic_worker);
2566 * endios are largely parallel and should have a very
2569 fs_info->endio_workers.idle_thresh = 4;
2570 fs_info->endio_meta_workers.idle_thresh = 4;
2571 fs_info->endio_raid56_workers.idle_thresh = 4;
2572 fs_info->rmw_workers.idle_thresh = 2;
2574 fs_info->endio_write_workers.idle_thresh = 2;
2575 fs_info->endio_meta_write_workers.idle_thresh = 2;
2576 fs_info->readahead_workers.idle_thresh = 2;
2579 * btrfs_start_workers can really only fail because of ENOMEM so just
2580 * return -ENOMEM if any of these fail.
2582 ret = btrfs_start_workers(&fs_info->workers);
2583 ret |= btrfs_start_workers(&fs_info->generic_worker);
2584 ret |= btrfs_start_workers(&fs_info->submit_workers);
2585 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2586 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2587 ret |= btrfs_start_workers(&fs_info->endio_workers);
2588 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2589 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2590 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2591 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2592 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2593 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2594 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2595 ret |= btrfs_start_workers(&fs_info->caching_workers);
2596 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2597 ret |= btrfs_start_workers(&fs_info->flush_workers);
2598 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2601 goto fail_sb_buffer;
2604 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2605 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2606 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2608 tree_root->nodesize = nodesize;
2609 tree_root->leafsize = leafsize;
2610 tree_root->sectorsize = sectorsize;
2611 tree_root->stripesize = stripesize;
2613 sb->s_blocksize = sectorsize;
2614 sb->s_blocksize_bits = blksize_bits(sectorsize);
2616 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2617 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2618 goto fail_sb_buffer;
2621 if (sectorsize != PAGE_SIZE) {
2622 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2623 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2624 goto fail_sb_buffer;
2627 mutex_lock(&fs_info->chunk_mutex);
2628 ret = btrfs_read_sys_array(tree_root);
2629 mutex_unlock(&fs_info->chunk_mutex);
2631 printk(KERN_WARNING "btrfs: failed to read the system "
2632 "array on %s\n", sb->s_id);
2633 goto fail_sb_buffer;
2636 blocksize = btrfs_level_size(tree_root,
2637 btrfs_super_chunk_root_level(disk_super));
2638 generation = btrfs_super_chunk_root_generation(disk_super);
2640 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2641 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2643 chunk_root->node = read_tree_block(chunk_root,
2644 btrfs_super_chunk_root(disk_super),
2645 blocksize, generation);
2646 if (!chunk_root->node ||
2647 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2648 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2650 goto fail_tree_roots;
2652 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2653 chunk_root->commit_root = btrfs_root_node(chunk_root);
2655 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2656 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2658 ret = btrfs_read_chunk_tree(chunk_root);
2660 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2662 goto fail_tree_roots;
2666 * keep the device that is marked to be the target device for the
2667 * dev_replace procedure
2669 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2671 if (!fs_devices->latest_bdev) {
2672 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2674 goto fail_tree_roots;
2678 blocksize = btrfs_level_size(tree_root,
2679 btrfs_super_root_level(disk_super));
2680 generation = btrfs_super_generation(disk_super);
2682 tree_root->node = read_tree_block(tree_root,
2683 btrfs_super_root(disk_super),
2684 blocksize, generation);
2685 if (!tree_root->node ||
2686 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2687 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2690 goto recovery_tree_root;
2693 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2694 tree_root->commit_root = btrfs_root_node(tree_root);
2695 btrfs_set_root_refs(&tree_root->root_item, 1);
2697 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2698 location.type = BTRFS_ROOT_ITEM_KEY;
2699 location.offset = 0;
2701 extent_root = btrfs_read_tree_root(tree_root, &location);
2702 if (IS_ERR(extent_root)) {
2703 ret = PTR_ERR(extent_root);
2704 goto recovery_tree_root;
2706 extent_root->track_dirty = 1;
2707 fs_info->extent_root = extent_root;
2709 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2710 dev_root = btrfs_read_tree_root(tree_root, &location);
2711 if (IS_ERR(dev_root)) {
2712 ret = PTR_ERR(dev_root);
2713 goto recovery_tree_root;
2715 dev_root->track_dirty = 1;
2716 fs_info->dev_root = dev_root;
2717 btrfs_init_devices_late(fs_info);
2719 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2720 csum_root = btrfs_read_tree_root(tree_root, &location);
2721 if (IS_ERR(csum_root)) {
2722 ret = PTR_ERR(csum_root);
2723 goto recovery_tree_root;
2725 csum_root->track_dirty = 1;
2726 fs_info->csum_root = csum_root;
2728 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2729 quota_root = btrfs_read_tree_root(tree_root, &location);
2730 if (!IS_ERR(quota_root)) {
2731 quota_root->track_dirty = 1;
2732 fs_info->quota_enabled = 1;
2733 fs_info->pending_quota_state = 1;
2734 fs_info->quota_root = quota_root;
2737 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2738 uuid_root = btrfs_read_tree_root(tree_root, &location);
2739 if (IS_ERR(uuid_root)) {
2740 ret = PTR_ERR(uuid_root);
2742 goto recovery_tree_root;
2743 create_uuid_tree = true;
2744 check_uuid_tree = false;
2746 uuid_root->track_dirty = 1;
2747 fs_info->uuid_root = uuid_root;
2748 create_uuid_tree = false;
2750 generation != btrfs_super_uuid_tree_generation(disk_super);
2753 fs_info->generation = generation;
2754 fs_info->last_trans_committed = generation;
2756 ret = btrfs_recover_balance(fs_info);
2758 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2759 goto fail_block_groups;
2762 ret = btrfs_init_dev_stats(fs_info);
2764 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2766 goto fail_block_groups;
2769 ret = btrfs_init_dev_replace(fs_info);
2771 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2772 goto fail_block_groups;
2775 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2777 ret = btrfs_init_space_info(fs_info);
2779 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2780 goto fail_block_groups;
2783 ret = btrfs_read_block_groups(extent_root);
2785 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2786 goto fail_block_groups;
2788 fs_info->num_tolerated_disk_barrier_failures =
2789 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2790 if (fs_info->fs_devices->missing_devices >
2791 fs_info->num_tolerated_disk_barrier_failures &&
2792 !(sb->s_flags & MS_RDONLY)) {
2794 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2795 goto fail_block_groups;
2798 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2800 if (IS_ERR(fs_info->cleaner_kthread))
2801 goto fail_block_groups;
2803 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2805 "btrfs-transaction");
2806 if (IS_ERR(fs_info->transaction_kthread))
2809 if (!btrfs_test_opt(tree_root, SSD) &&
2810 !btrfs_test_opt(tree_root, NOSSD) &&
2811 !fs_info->fs_devices->rotating) {
2812 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2814 btrfs_set_opt(fs_info->mount_opt, SSD);
2817 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2818 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2819 ret = btrfsic_mount(tree_root, fs_devices,
2820 btrfs_test_opt(tree_root,
2821 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2823 fs_info->check_integrity_print_mask);
2825 printk(KERN_WARNING "btrfs: failed to initialize"
2826 " integrity check module %s\n", sb->s_id);
2829 ret = btrfs_read_qgroup_config(fs_info);
2831 goto fail_trans_kthread;
2833 /* do not make disk changes in broken FS */
2834 if (btrfs_super_log_root(disk_super) != 0) {
2835 u64 bytenr = btrfs_super_log_root(disk_super);
2837 if (fs_devices->rw_devices == 0) {
2838 printk(KERN_WARNING "Btrfs log replay required "
2844 btrfs_level_size(tree_root,
2845 btrfs_super_log_root_level(disk_super));
2847 log_tree_root = btrfs_alloc_root(fs_info);
2848 if (!log_tree_root) {
2853 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2854 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2856 log_tree_root->node = read_tree_block(tree_root, bytenr,
2859 if (!log_tree_root->node ||
2860 !extent_buffer_uptodate(log_tree_root->node)) {
2861 printk(KERN_ERR "btrfs: failed to read log tree\n");
2862 free_extent_buffer(log_tree_root->node);
2863 kfree(log_tree_root);
2864 goto fail_trans_kthread;
2866 /* returns with log_tree_root freed on success */
2867 ret = btrfs_recover_log_trees(log_tree_root);
2869 btrfs_error(tree_root->fs_info, ret,
2870 "Failed to recover log tree");
2871 free_extent_buffer(log_tree_root->node);
2872 kfree(log_tree_root);
2873 goto fail_trans_kthread;
2876 if (sb->s_flags & MS_RDONLY) {
2877 ret = btrfs_commit_super(tree_root);
2879 goto fail_trans_kthread;
2883 ret = btrfs_find_orphan_roots(tree_root);
2885 goto fail_trans_kthread;
2887 if (!(sb->s_flags & MS_RDONLY)) {
2888 ret = btrfs_cleanup_fs_roots(fs_info);
2890 goto fail_trans_kthread;
2892 ret = btrfs_recover_relocation(tree_root);
2895 "btrfs: failed to recover relocation\n");
2901 location.objectid = BTRFS_FS_TREE_OBJECTID;
2902 location.type = BTRFS_ROOT_ITEM_KEY;
2903 location.offset = 0;
2905 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2906 if (IS_ERR(fs_info->fs_root)) {
2907 err = PTR_ERR(fs_info->fs_root);
2911 if (sb->s_flags & MS_RDONLY)
2914 down_read(&fs_info->cleanup_work_sem);
2915 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2916 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2917 up_read(&fs_info->cleanup_work_sem);
2918 close_ctree(tree_root);
2921 up_read(&fs_info->cleanup_work_sem);
2923 ret = btrfs_resume_balance_async(fs_info);
2925 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2926 close_ctree(tree_root);
2930 ret = btrfs_resume_dev_replace_async(fs_info);
2932 pr_warn("btrfs: failed to resume dev_replace\n");
2933 close_ctree(tree_root);
2937 btrfs_qgroup_rescan_resume(fs_info);
2939 if (create_uuid_tree) {
2940 pr_info("btrfs: creating UUID tree\n");
2941 ret = btrfs_create_uuid_tree(fs_info);
2943 pr_warn("btrfs: failed to create the UUID tree %d\n",
2945 close_ctree(tree_root);
2948 } else if (check_uuid_tree ||
2949 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2950 pr_info("btrfs: checking UUID tree\n");
2951 ret = btrfs_check_uuid_tree(fs_info);
2953 pr_warn("btrfs: failed to check the UUID tree %d\n",
2955 close_ctree(tree_root);
2959 fs_info->update_uuid_tree_gen = 1;
2965 btrfs_free_qgroup_config(fs_info);
2967 kthread_stop(fs_info->transaction_kthread);
2968 btrfs_cleanup_transaction(fs_info->tree_root);
2969 del_fs_roots(fs_info);
2971 kthread_stop(fs_info->cleaner_kthread);
2974 * make sure we're done with the btree inode before we stop our
2977 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2980 btrfs_put_block_group_cache(fs_info);
2981 btrfs_free_block_groups(fs_info);
2984 free_root_pointers(fs_info, 1);
2985 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2988 btrfs_stop_all_workers(fs_info);
2991 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2993 iput(fs_info->btree_inode);
2994 fail_delalloc_bytes:
2995 percpu_counter_destroy(&fs_info->delalloc_bytes);
2996 fail_dirty_metadata_bytes:
2997 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2999 bdi_destroy(&fs_info->bdi);
3001 cleanup_srcu_struct(&fs_info->subvol_srcu);
3003 btrfs_free_stripe_hash_table(fs_info);
3004 btrfs_close_devices(fs_info->fs_devices);
3008 if (!btrfs_test_opt(tree_root, RECOVERY))
3009 goto fail_tree_roots;
3011 free_root_pointers(fs_info, 0);
3013 /* don't use the log in recovery mode, it won't be valid */
3014 btrfs_set_super_log_root(disk_super, 0);
3016 /* we can't trust the free space cache either */
3017 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3019 ret = next_root_backup(fs_info, fs_info->super_copy,
3020 &num_backups_tried, &backup_index);
3022 goto fail_block_groups;
3023 goto retry_root_backup;
3026 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3029 set_buffer_uptodate(bh);
3031 struct btrfs_device *device = (struct btrfs_device *)
3034 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
3035 "I/O error on %s\n",
3036 rcu_str_deref(device->name));
3037 /* note, we dont' set_buffer_write_io_error because we have
3038 * our own ways of dealing with the IO errors
3040 clear_buffer_uptodate(bh);
3041 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3047 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3049 struct buffer_head *bh;
3050 struct buffer_head *latest = NULL;
3051 struct btrfs_super_block *super;
3056 /* we would like to check all the supers, but that would make
3057 * a btrfs mount succeed after a mkfs from a different FS.
3058 * So, we need to add a special mount option to scan for
3059 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3061 for (i = 0; i < 1; i++) {
3062 bytenr = btrfs_sb_offset(i);
3063 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3064 i_size_read(bdev->bd_inode))
3066 bh = __bread(bdev, bytenr / 4096,
3067 BTRFS_SUPER_INFO_SIZE);
3071 super = (struct btrfs_super_block *)bh->b_data;
3072 if (btrfs_super_bytenr(super) != bytenr ||
3073 btrfs_super_magic(super) != BTRFS_MAGIC) {
3078 if (!latest || btrfs_super_generation(super) > transid) {
3081 transid = btrfs_super_generation(super);
3090 * this should be called twice, once with wait == 0 and
3091 * once with wait == 1. When wait == 0 is done, all the buffer heads
3092 * we write are pinned.
3094 * They are released when wait == 1 is done.
3095 * max_mirrors must be the same for both runs, and it indicates how
3096 * many supers on this one device should be written.
3098 * max_mirrors == 0 means to write them all.
3100 static int write_dev_supers(struct btrfs_device *device,
3101 struct btrfs_super_block *sb,
3102 int do_barriers, int wait, int max_mirrors)
3104 struct buffer_head *bh;
3111 if (max_mirrors == 0)
3112 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3114 for (i = 0; i < max_mirrors; i++) {
3115 bytenr = btrfs_sb_offset(i);
3116 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3120 bh = __find_get_block(device->bdev, bytenr / 4096,
3121 BTRFS_SUPER_INFO_SIZE);
3127 if (!buffer_uptodate(bh))
3130 /* drop our reference */
3133 /* drop the reference from the wait == 0 run */
3137 btrfs_set_super_bytenr(sb, bytenr);
3140 crc = btrfs_csum_data((char *)sb +
3141 BTRFS_CSUM_SIZE, crc,
3142 BTRFS_SUPER_INFO_SIZE -
3144 btrfs_csum_final(crc, sb->csum);
3147 * one reference for us, and we leave it for the
3150 bh = __getblk(device->bdev, bytenr / 4096,
3151 BTRFS_SUPER_INFO_SIZE);
3153 printk(KERN_ERR "btrfs: couldn't get super "
3154 "buffer head for bytenr %Lu\n", bytenr);
3159 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3161 /* one reference for submit_bh */
3164 set_buffer_uptodate(bh);
3166 bh->b_end_io = btrfs_end_buffer_write_sync;
3167 bh->b_private = device;
3171 * we fua the first super. The others we allow
3174 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3178 return errors < i ? 0 : -1;
3182 * endio for the write_dev_flush, this will wake anyone waiting
3183 * for the barrier when it is done
3185 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3188 if (err == -EOPNOTSUPP)
3189 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3190 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3192 if (bio->bi_private)
3193 complete(bio->bi_private);
3198 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3199 * sent down. With wait == 1, it waits for the previous flush.
3201 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3204 static int write_dev_flush(struct btrfs_device *device, int wait)
3209 if (device->nobarriers)
3213 bio = device->flush_bio;
3217 wait_for_completion(&device->flush_wait);
3219 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3220 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3221 rcu_str_deref(device->name));
3222 device->nobarriers = 1;
3223 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3225 btrfs_dev_stat_inc_and_print(device,
3226 BTRFS_DEV_STAT_FLUSH_ERRS);
3229 /* drop the reference from the wait == 0 run */
3231 device->flush_bio = NULL;
3237 * one reference for us, and we leave it for the
3240 device->flush_bio = NULL;
3241 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3245 bio->bi_end_io = btrfs_end_empty_barrier;
3246 bio->bi_bdev = device->bdev;
3247 init_completion(&device->flush_wait);
3248 bio->bi_private = &device->flush_wait;
3249 device->flush_bio = bio;
3252 btrfsic_submit_bio(WRITE_FLUSH, bio);
3258 * send an empty flush down to each device in parallel,
3259 * then wait for them
3261 static int barrier_all_devices(struct btrfs_fs_info *info)
3263 struct list_head *head;
3264 struct btrfs_device *dev;
3265 int errors_send = 0;
3266 int errors_wait = 0;
3269 /* send down all the barriers */
3270 head = &info->fs_devices->devices;
3271 list_for_each_entry_rcu(dev, head, dev_list) {
3276 if (!dev->in_fs_metadata || !dev->writeable)
3279 ret = write_dev_flush(dev, 0);
3284 /* wait for all the barriers */
3285 list_for_each_entry_rcu(dev, head, dev_list) {
3290 if (!dev->in_fs_metadata || !dev->writeable)
3293 ret = write_dev_flush(dev, 1);
3297 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3298 errors_wait > info->num_tolerated_disk_barrier_failures)
3303 int btrfs_calc_num_tolerated_disk_barrier_failures(
3304 struct btrfs_fs_info *fs_info)
3306 struct btrfs_ioctl_space_info space;
3307 struct btrfs_space_info *sinfo;
3308 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3309 BTRFS_BLOCK_GROUP_SYSTEM,
3310 BTRFS_BLOCK_GROUP_METADATA,
3311 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3315 int num_tolerated_disk_barrier_failures =
3316 (int)fs_info->fs_devices->num_devices;
3318 for (i = 0; i < num_types; i++) {
3319 struct btrfs_space_info *tmp;
3323 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3324 if (tmp->flags == types[i]) {
3334 down_read(&sinfo->groups_sem);
3335 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3336 if (!list_empty(&sinfo->block_groups[c])) {
3339 btrfs_get_block_group_info(
3340 &sinfo->block_groups[c], &space);
3341 if (space.total_bytes == 0 ||
3342 space.used_bytes == 0)
3344 flags = space.flags;
3347 * 0: if dup, single or RAID0 is configured for
3348 * any of metadata, system or data, else
3349 * 1: if RAID5 is configured, or if RAID1 or
3350 * RAID10 is configured and only two mirrors
3352 * 2: if RAID6 is configured, else
3353 * num_mirrors - 1: if RAID1 or RAID10 is
3354 * configured and more than
3355 * 2 mirrors are used.
3357 if (num_tolerated_disk_barrier_failures > 0 &&
3358 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3359 BTRFS_BLOCK_GROUP_RAID0)) ||
3360 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3362 num_tolerated_disk_barrier_failures = 0;
3363 else if (num_tolerated_disk_barrier_failures > 1) {
3364 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3365 BTRFS_BLOCK_GROUP_RAID5 |
3366 BTRFS_BLOCK_GROUP_RAID10)) {
3367 num_tolerated_disk_barrier_failures = 1;
3369 BTRFS_BLOCK_GROUP_RAID6) {
3370 num_tolerated_disk_barrier_failures = 2;
3375 up_read(&sinfo->groups_sem);
3378 return num_tolerated_disk_barrier_failures;
3381 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3383 struct list_head *head;
3384 struct btrfs_device *dev;
3385 struct btrfs_super_block *sb;
3386 struct btrfs_dev_item *dev_item;
3390 int total_errors = 0;
3393 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3394 backup_super_roots(root->fs_info);
3396 sb = root->fs_info->super_for_commit;
3397 dev_item = &sb->dev_item;
3399 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3400 head = &root->fs_info->fs_devices->devices;
3401 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3404 ret = barrier_all_devices(root->fs_info);
3407 &root->fs_info->fs_devices->device_list_mutex);
3408 btrfs_error(root->fs_info, ret,
3409 "errors while submitting device barriers.");
3414 list_for_each_entry_rcu(dev, head, dev_list) {
3419 if (!dev->in_fs_metadata || !dev->writeable)
3422 btrfs_set_stack_device_generation(dev_item, 0);
3423 btrfs_set_stack_device_type(dev_item, dev->type);
3424 btrfs_set_stack_device_id(dev_item, dev->devid);
3425 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3426 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3427 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3428 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3429 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3430 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3431 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3433 flags = btrfs_super_flags(sb);
3434 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3436 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3440 if (total_errors > max_errors) {
3441 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3443 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3445 /* FUA is masked off if unsupported and can't be the reason */
3446 btrfs_error(root->fs_info, -EIO,
3447 "%d errors while writing supers", total_errors);
3452 list_for_each_entry_rcu(dev, head, dev_list) {
3455 if (!dev->in_fs_metadata || !dev->writeable)
3458 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3462 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3463 if (total_errors > max_errors) {
3464 btrfs_error(root->fs_info, -EIO,
3465 "%d errors while writing supers", total_errors);
3471 int write_ctree_super(struct btrfs_trans_handle *trans,
3472 struct btrfs_root *root, int max_mirrors)
3476 ret = write_all_supers(root, max_mirrors);
3480 /* Drop a fs root from the radix tree and free it. */
3481 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3482 struct btrfs_root *root)
3484 spin_lock(&fs_info->fs_roots_radix_lock);
3485 radix_tree_delete(&fs_info->fs_roots_radix,
3486 (unsigned long)root->root_key.objectid);
3487 spin_unlock(&fs_info->fs_roots_radix_lock);
3489 if (btrfs_root_refs(&root->root_item) == 0)
3490 synchronize_srcu(&fs_info->subvol_srcu);
3492 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3493 btrfs_free_log(NULL, root);
3494 btrfs_free_log_root_tree(NULL, fs_info);
3497 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3498 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3502 static void free_fs_root(struct btrfs_root *root)
3504 iput(root->cache_inode);
3505 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3506 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3507 root->orphan_block_rsv = NULL;
3509 free_anon_bdev(root->anon_dev);
3510 free_extent_buffer(root->node);
3511 free_extent_buffer(root->commit_root);
3512 kfree(root->free_ino_ctl);
3513 kfree(root->free_ino_pinned);
3515 btrfs_put_fs_root(root);
3518 void btrfs_free_fs_root(struct btrfs_root *root)
3523 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3525 u64 root_objectid = 0;
3526 struct btrfs_root *gang[8];
3531 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3532 (void **)gang, root_objectid,
3537 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3538 for (i = 0; i < ret; i++) {
3541 root_objectid = gang[i]->root_key.objectid;
3542 err = btrfs_orphan_cleanup(gang[i]);
3551 int btrfs_commit_super(struct btrfs_root *root)
3553 struct btrfs_trans_handle *trans;
3556 mutex_lock(&root->fs_info->cleaner_mutex);
3557 btrfs_run_delayed_iputs(root);
3558 mutex_unlock(&root->fs_info->cleaner_mutex);
3559 wake_up_process(root->fs_info->cleaner_kthread);
3561 /* wait until ongoing cleanup work done */
3562 down_write(&root->fs_info->cleanup_work_sem);
3563 up_write(&root->fs_info->cleanup_work_sem);
3565 trans = btrfs_join_transaction(root);
3567 return PTR_ERR(trans);
3568 ret = btrfs_commit_transaction(trans, root);
3571 /* run commit again to drop the original snapshot */
3572 trans = btrfs_join_transaction(root);
3574 return PTR_ERR(trans);
3575 ret = btrfs_commit_transaction(trans, root);
3578 ret = btrfs_write_and_wait_transaction(NULL, root);
3580 btrfs_error(root->fs_info, ret,
3581 "Failed to sync btree inode to disk.");
3585 ret = write_ctree_super(NULL, root, 0);
3589 int close_ctree(struct btrfs_root *root)
3591 struct btrfs_fs_info *fs_info = root->fs_info;
3594 fs_info->closing = 1;
3597 /* wait for the uuid_scan task to finish */
3598 down(&fs_info->uuid_tree_rescan_sem);
3599 /* avoid complains from lockdep et al., set sem back to initial state */
3600 up(&fs_info->uuid_tree_rescan_sem);
3602 /* pause restriper - we want to resume on mount */
3603 btrfs_pause_balance(fs_info);
3605 btrfs_dev_replace_suspend_for_unmount(fs_info);
3607 btrfs_scrub_cancel(fs_info);
3609 /* wait for any defraggers to finish */
3610 wait_event(fs_info->transaction_wait,
3611 (atomic_read(&fs_info->defrag_running) == 0));
3613 /* clear out the rbtree of defraggable inodes */
3614 btrfs_cleanup_defrag_inodes(fs_info);
3616 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3617 ret = btrfs_commit_super(root);
3619 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3622 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3623 btrfs_error_commit_super(root);
3625 btrfs_put_block_group_cache(fs_info);
3627 kthread_stop(fs_info->transaction_kthread);
3628 kthread_stop(fs_info->cleaner_kthread);
3630 fs_info->closing = 2;
3633 btrfs_free_qgroup_config(root->fs_info);
3635 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3636 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3637 percpu_counter_sum(&fs_info->delalloc_bytes));
3640 btrfs_free_block_groups(fs_info);
3642 btrfs_stop_all_workers(fs_info);
3644 del_fs_roots(fs_info);
3646 free_root_pointers(fs_info, 1);
3648 iput(fs_info->btree_inode);
3650 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3651 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3652 btrfsic_unmount(root, fs_info->fs_devices);
3655 btrfs_close_devices(fs_info->fs_devices);
3656 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3658 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3659 percpu_counter_destroy(&fs_info->delalloc_bytes);
3660 bdi_destroy(&fs_info->bdi);
3661 cleanup_srcu_struct(&fs_info->subvol_srcu);
3663 btrfs_free_stripe_hash_table(fs_info);
3665 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3666 root->orphan_block_rsv = NULL;
3671 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3675 struct inode *btree_inode = buf->pages[0]->mapping->host;
3677 ret = extent_buffer_uptodate(buf);
3681 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3682 parent_transid, atomic);
3688 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3690 return set_extent_buffer_uptodate(buf);
3693 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3695 struct btrfs_root *root;
3696 u64 transid = btrfs_header_generation(buf);
3699 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3701 * This is a fast path so only do this check if we have sanity tests
3702 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3703 * outside of the sanity tests.
3705 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3708 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3709 btrfs_assert_tree_locked(buf);
3710 if (transid != root->fs_info->generation)
3711 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3712 "found %llu running %llu\n",
3713 buf->start, transid, root->fs_info->generation);
3714 was_dirty = set_extent_buffer_dirty(buf);
3716 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3718 root->fs_info->dirty_metadata_batch);
3721 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3725 * looks as though older kernels can get into trouble with
3726 * this code, they end up stuck in balance_dirty_pages forever
3730 if (current->flags & PF_MEMALLOC)
3734 btrfs_balance_delayed_items(root);
3736 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3737 BTRFS_DIRTY_METADATA_THRESH);
3739 balance_dirty_pages_ratelimited(
3740 root->fs_info->btree_inode->i_mapping);
3745 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3747 __btrfs_btree_balance_dirty(root, 1);
3750 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3752 __btrfs_btree_balance_dirty(root, 0);
3755 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3757 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3758 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3761 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3765 * Placeholder for checks
3770 static void btrfs_error_commit_super(struct btrfs_root *root)
3772 mutex_lock(&root->fs_info->cleaner_mutex);
3773 btrfs_run_delayed_iputs(root);
3774 mutex_unlock(&root->fs_info->cleaner_mutex);
3776 down_write(&root->fs_info->cleanup_work_sem);
3777 up_write(&root->fs_info->cleanup_work_sem);
3779 /* cleanup FS via transaction */
3780 btrfs_cleanup_transaction(root);
3783 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3784 struct btrfs_root *root)
3786 struct btrfs_inode *btrfs_inode;
3787 struct list_head splice;
3789 INIT_LIST_HEAD(&splice);
3791 mutex_lock(&root->fs_info->ordered_operations_mutex);
3792 spin_lock(&root->fs_info->ordered_root_lock);
3794 list_splice_init(&t->ordered_operations, &splice);
3795 while (!list_empty(&splice)) {
3796 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3797 ordered_operations);
3799 list_del_init(&btrfs_inode->ordered_operations);
3800 spin_unlock(&root->fs_info->ordered_root_lock);
3802 btrfs_invalidate_inodes(btrfs_inode->root);
3804 spin_lock(&root->fs_info->ordered_root_lock);
3807 spin_unlock(&root->fs_info->ordered_root_lock);
3808 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3811 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3813 struct btrfs_ordered_extent *ordered;
3815 spin_lock(&root->ordered_extent_lock);
3817 * This will just short circuit the ordered completion stuff which will
3818 * make sure the ordered extent gets properly cleaned up.
3820 list_for_each_entry(ordered, &root->ordered_extents,
3822 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3823 spin_unlock(&root->ordered_extent_lock);
3826 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3828 struct btrfs_root *root;
3829 struct list_head splice;
3831 INIT_LIST_HEAD(&splice);
3833 spin_lock(&fs_info->ordered_root_lock);
3834 list_splice_init(&fs_info->ordered_roots, &splice);
3835 while (!list_empty(&splice)) {
3836 root = list_first_entry(&splice, struct btrfs_root,
3838 list_move_tail(&root->ordered_root,
3839 &fs_info->ordered_roots);
3841 btrfs_destroy_ordered_extents(root);
3843 cond_resched_lock(&fs_info->ordered_root_lock);
3845 spin_unlock(&fs_info->ordered_root_lock);
3848 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3849 struct btrfs_root *root)
3851 struct rb_node *node;
3852 struct btrfs_delayed_ref_root *delayed_refs;
3853 struct btrfs_delayed_ref_node *ref;
3856 delayed_refs = &trans->delayed_refs;
3858 spin_lock(&delayed_refs->lock);
3859 if (delayed_refs->num_entries == 0) {
3860 spin_unlock(&delayed_refs->lock);
3861 printk(KERN_INFO "delayed_refs has NO entry\n");
3865 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3866 struct btrfs_delayed_ref_head *head = NULL;
3867 bool pin_bytes = false;
3869 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3870 atomic_set(&ref->refs, 1);
3871 if (btrfs_delayed_ref_is_head(ref)) {
3873 head = btrfs_delayed_node_to_head(ref);
3874 if (!mutex_trylock(&head->mutex)) {
3875 atomic_inc(&ref->refs);
3876 spin_unlock(&delayed_refs->lock);
3878 /* Need to wait for the delayed ref to run */
3879 mutex_lock(&head->mutex);
3880 mutex_unlock(&head->mutex);
3881 btrfs_put_delayed_ref(ref);
3883 spin_lock(&delayed_refs->lock);
3887 if (head->must_insert_reserved)
3889 btrfs_free_delayed_extent_op(head->extent_op);
3890 delayed_refs->num_heads--;
3891 if (list_empty(&head->cluster))
3892 delayed_refs->num_heads_ready--;
3893 list_del_init(&head->cluster);
3897 rb_erase(&ref->rb_node, &delayed_refs->root);
3898 delayed_refs->num_entries--;
3899 spin_unlock(&delayed_refs->lock);
3902 btrfs_pin_extent(root, ref->bytenr,
3904 mutex_unlock(&head->mutex);
3906 btrfs_put_delayed_ref(ref);
3909 spin_lock(&delayed_refs->lock);
3912 spin_unlock(&delayed_refs->lock);
3917 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3919 struct btrfs_inode *btrfs_inode;
3920 struct list_head splice;
3922 INIT_LIST_HEAD(&splice);
3924 spin_lock(&root->delalloc_lock);
3925 list_splice_init(&root->delalloc_inodes, &splice);
3927 while (!list_empty(&splice)) {
3928 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3931 list_del_init(&btrfs_inode->delalloc_inodes);
3932 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3933 &btrfs_inode->runtime_flags);
3934 spin_unlock(&root->delalloc_lock);
3936 btrfs_invalidate_inodes(btrfs_inode->root);
3938 spin_lock(&root->delalloc_lock);
3941 spin_unlock(&root->delalloc_lock);
3944 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3946 struct btrfs_root *root;
3947 struct list_head splice;
3949 INIT_LIST_HEAD(&splice);
3951 spin_lock(&fs_info->delalloc_root_lock);
3952 list_splice_init(&fs_info->delalloc_roots, &splice);
3953 while (!list_empty(&splice)) {
3954 root = list_first_entry(&splice, struct btrfs_root,
3956 list_del_init(&root->delalloc_root);
3957 root = btrfs_grab_fs_root(root);
3959 spin_unlock(&fs_info->delalloc_root_lock);
3961 btrfs_destroy_delalloc_inodes(root);
3962 btrfs_put_fs_root(root);
3964 spin_lock(&fs_info->delalloc_root_lock);
3966 spin_unlock(&fs_info->delalloc_root_lock);
3969 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3970 struct extent_io_tree *dirty_pages,
3974 struct extent_buffer *eb;
3979 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3984 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3985 while (start <= end) {
3986 eb = btrfs_find_tree_block(root, start,
3988 start += root->leafsize;
3991 wait_on_extent_buffer_writeback(eb);
3993 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3995 clear_extent_buffer_dirty(eb);
3996 free_extent_buffer_stale(eb);
4003 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4004 struct extent_io_tree *pinned_extents)
4006 struct extent_io_tree *unpin;
4012 unpin = pinned_extents;
4015 ret = find_first_extent_bit(unpin, 0, &start, &end,
4016 EXTENT_DIRTY, NULL);
4021 if (btrfs_test_opt(root, DISCARD))
4022 ret = btrfs_error_discard_extent(root, start,
4026 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4027 btrfs_error_unpin_extent_range(root, start, end);
4032 if (unpin == &root->fs_info->freed_extents[0])
4033 unpin = &root->fs_info->freed_extents[1];
4035 unpin = &root->fs_info->freed_extents[0];
4043 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4044 struct btrfs_root *root)
4046 btrfs_destroy_ordered_operations(cur_trans, root);
4048 btrfs_destroy_delayed_refs(cur_trans, root);
4049 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
4050 cur_trans->dirty_pages.dirty_bytes);
4052 cur_trans->state = TRANS_STATE_COMMIT_START;
4053 wake_up(&root->fs_info->transaction_blocked_wait);
4055 cur_trans->state = TRANS_STATE_UNBLOCKED;
4056 wake_up(&root->fs_info->transaction_wait);
4058 btrfs_destroy_delayed_inodes(root);
4059 btrfs_assert_delayed_root_empty(root);
4061 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4063 btrfs_destroy_pinned_extent(root,
4064 root->fs_info->pinned_extents);
4066 cur_trans->state =TRANS_STATE_COMPLETED;
4067 wake_up(&cur_trans->commit_wait);
4070 memset(cur_trans, 0, sizeof(*cur_trans));
4071 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4075 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4077 struct btrfs_transaction *t;
4079 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4081 spin_lock(&root->fs_info->trans_lock);
4082 while (!list_empty(&root->fs_info->trans_list)) {
4083 t = list_first_entry(&root->fs_info->trans_list,
4084 struct btrfs_transaction, list);
4085 if (t->state >= TRANS_STATE_COMMIT_START) {
4086 atomic_inc(&t->use_count);
4087 spin_unlock(&root->fs_info->trans_lock);
4088 btrfs_wait_for_commit(root, t->transid);
4089 btrfs_put_transaction(t);
4090 spin_lock(&root->fs_info->trans_lock);
4093 if (t == root->fs_info->running_transaction) {
4094 t->state = TRANS_STATE_COMMIT_DOING;
4095 spin_unlock(&root->fs_info->trans_lock);
4097 * We wait for 0 num_writers since we don't hold a trans
4098 * handle open currently for this transaction.
4100 wait_event(t->writer_wait,
4101 atomic_read(&t->num_writers) == 0);
4103 spin_unlock(&root->fs_info->trans_lock);
4105 btrfs_cleanup_one_transaction(t, root);
4107 spin_lock(&root->fs_info->trans_lock);
4108 if (t == root->fs_info->running_transaction)
4109 root->fs_info->running_transaction = NULL;
4110 list_del_init(&t->list);
4111 spin_unlock(&root->fs_info->trans_lock);
4113 btrfs_put_transaction(t);
4114 trace_btrfs_transaction_commit(root);
4115 spin_lock(&root->fs_info->trans_lock);
4117 spin_unlock(&root->fs_info->trans_lock);
4118 btrfs_destroy_all_ordered_extents(root->fs_info);
4119 btrfs_destroy_delayed_inodes(root);
4120 btrfs_assert_delayed_root_empty(root);
4121 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4122 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4123 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4128 static struct extent_io_ops btree_extent_io_ops = {
4129 .readpage_end_io_hook = btree_readpage_end_io_hook,
4130 .readpage_io_failed_hook = btree_io_failed_hook,
4131 .submit_bio_hook = btree_submit_bio_hook,
4132 /* note we're sharing with inode.c for the merge bio hook */
4133 .merge_bio_hook = btrfs_merge_bio_hook,