2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/slab.h>
30 #include <linux/migrate.h>
31 #include <linux/ratelimit.h>
32 #include <linux/uuid.h>
33 #include <linux/semaphore.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "print-tree.h"
42 #include "async-thread.h"
45 #include "free-space-cache.h"
46 #include "inode-map.h"
47 #include "check-integrity.h"
48 #include "rcu-string.h"
49 #include "dev-replace.h"
54 #include <asm/cpufeature.h>
57 static struct extent_io_ops btree_extent_io_ops;
58 static void end_workqueue_fn(struct btrfs_work *work);
59 static void free_fs_root(struct btrfs_root *root);
60 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
62 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
63 struct btrfs_root *root);
64 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
65 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
66 struct btrfs_root *root);
67 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
68 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
69 struct extent_io_tree *dirty_pages,
71 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
72 struct extent_io_tree *pinned_extents);
73 static int btrfs_cleanup_transaction(struct btrfs_root *root);
74 static void btrfs_error_commit_super(struct btrfs_root *root);
77 * end_io_wq structs are used to do processing in task context when an IO is
78 * complete. This is used during reads to verify checksums, and it is used
79 * by writes to insert metadata for new file extents after IO is complete.
85 struct btrfs_fs_info *info;
88 struct list_head list;
89 struct btrfs_work work;
93 * async submit bios are used to offload expensive checksumming
94 * onto the worker threads. They checksum file and metadata bios
95 * just before they are sent down the IO stack.
97 struct async_submit_bio {
100 struct list_head list;
101 extent_submit_bio_hook_t *submit_bio_start;
102 extent_submit_bio_hook_t *submit_bio_done;
105 unsigned long bio_flags;
107 * bio_offset is optional, can be used if the pages in the bio
108 * can't tell us where in the file the bio should go
111 struct btrfs_work work;
116 * Lockdep class keys for extent_buffer->lock's in this root. For a given
117 * eb, the lockdep key is determined by the btrfs_root it belongs to and
118 * the level the eb occupies in the tree.
120 * Different roots are used for different purposes and may nest inside each
121 * other and they require separate keysets. As lockdep keys should be
122 * static, assign keysets according to the purpose of the root as indicated
123 * by btrfs_root->objectid. This ensures that all special purpose roots
124 * have separate keysets.
126 * Lock-nesting across peer nodes is always done with the immediate parent
127 * node locked thus preventing deadlock. As lockdep doesn't know this, use
128 * subclass to avoid triggering lockdep warning in such cases.
130 * The key is set by the readpage_end_io_hook after the buffer has passed
131 * csum validation but before the pages are unlocked. It is also set by
132 * btrfs_init_new_buffer on freshly allocated blocks.
134 * We also add a check to make sure the highest level of the tree is the
135 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
136 * needs update as well.
138 #ifdef CONFIG_DEBUG_LOCK_ALLOC
139 # if BTRFS_MAX_LEVEL != 8
143 static struct btrfs_lockdep_keyset {
144 u64 id; /* root objectid */
145 const char *name_stem; /* lock name stem */
146 char names[BTRFS_MAX_LEVEL + 1][20];
147 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
148 } btrfs_lockdep_keysets[] = {
149 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
150 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
151 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
152 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
153 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
154 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
155 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
156 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
157 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
158 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
159 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
160 { .id = 0, .name_stem = "tree" },
163 void __init btrfs_init_lockdep(void)
167 /* initialize lockdep class names */
168 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
169 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
171 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
172 snprintf(ks->names[j], sizeof(ks->names[j]),
173 "btrfs-%s-%02d", ks->name_stem, j);
177 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
180 struct btrfs_lockdep_keyset *ks;
182 BUG_ON(level >= ARRAY_SIZE(ks->keys));
184 /* find the matching keyset, id 0 is the default entry */
185 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
186 if (ks->id == objectid)
189 lockdep_set_class_and_name(&eb->lock,
190 &ks->keys[level], ks->names[level]);
196 * extents on the btree inode are pretty simple, there's one extent
197 * that covers the entire device
199 static struct extent_map *btree_get_extent(struct inode *inode,
200 struct page *page, size_t pg_offset, u64 start, u64 len,
203 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
204 struct extent_map *em;
207 read_lock(&em_tree->lock);
208 em = lookup_extent_mapping(em_tree, start, len);
211 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
212 read_unlock(&em_tree->lock);
215 read_unlock(&em_tree->lock);
217 em = alloc_extent_map();
219 em = ERR_PTR(-ENOMEM);
224 em->block_len = (u64)-1;
226 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
228 write_lock(&em_tree->lock);
229 ret = add_extent_mapping(em_tree, em, 0);
230 if (ret == -EEXIST) {
232 em = lookup_extent_mapping(em_tree, start, len);
239 write_unlock(&em_tree->lock);
245 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
247 return btrfs_crc32c(seed, data, len);
250 void btrfs_csum_final(u32 crc, char *result)
252 put_unaligned_le32(~crc, result);
256 * compute the csum for a btree block, and either verify it or write it
257 * into the csum field of the block.
259 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
262 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
265 unsigned long cur_len;
266 unsigned long offset = BTRFS_CSUM_SIZE;
268 unsigned long map_start;
269 unsigned long map_len;
272 unsigned long inline_result;
274 len = buf->len - offset;
276 err = map_private_extent_buffer(buf, offset, 32,
277 &kaddr, &map_start, &map_len);
280 cur_len = min(len, map_len - (offset - map_start));
281 crc = btrfs_csum_data(kaddr + offset - map_start,
286 if (csum_size > sizeof(inline_result)) {
287 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
291 result = (char *)&inline_result;
294 btrfs_csum_final(crc, result);
297 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
300 memcpy(&found, result, csum_size);
302 read_extent_buffer(buf, &val, 0, csum_size);
303 printk_ratelimited(KERN_INFO
304 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
306 root->fs_info->sb->s_id, buf->start,
307 val, found, btrfs_header_level(buf));
308 if (result != (char *)&inline_result)
313 write_extent_buffer(buf, result, 0, csum_size);
315 if (result != (char *)&inline_result)
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
326 static int verify_parent_transid(struct extent_io_tree *io_tree,
327 struct extent_buffer *eb, u64 parent_transid,
330 struct extent_state *cached_state = NULL;
332 bool need_lock = (current->journal_info ==
333 (void *)BTRFS_SEND_TRANS_STUB);
335 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
342 btrfs_tree_read_lock(eb);
343 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
346 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
348 if (extent_buffer_uptodate(eb) &&
349 btrfs_header_generation(eb) == parent_transid) {
353 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
355 eb->start, parent_transid, btrfs_header_generation(eb));
359 * Things reading via commit roots that don't have normal protection,
360 * like send, can have a really old block in cache that may point at a
361 * block that has been free'd and re-allocated. So don't clear uptodate
362 * if we find an eb that is under IO (dirty/writeback) because we could
363 * end up reading in the stale data and then writing it back out and
364 * making everybody very sad.
366 if (!extent_buffer_under_io(eb))
367 clear_extent_buffer_uptodate(eb);
369 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
370 &cached_state, GFP_NOFS);
371 btrfs_tree_read_unlock_blocking(eb);
376 * Return 0 if the superblock checksum type matches the checksum value of that
377 * algorithm. Pass the raw disk superblock data.
379 static int btrfs_check_super_csum(char *raw_disk_sb)
381 struct btrfs_super_block *disk_sb =
382 (struct btrfs_super_block *)raw_disk_sb;
383 u16 csum_type = btrfs_super_csum_type(disk_sb);
386 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
388 const int csum_size = sizeof(crc);
389 char result[csum_size];
392 * The super_block structure does not span the whole
393 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
394 * is filled with zeros and is included in the checkum.
396 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
397 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
398 btrfs_csum_final(crc, result);
400 if (memcmp(raw_disk_sb, result, csum_size))
403 if (ret && btrfs_super_generation(disk_sb) < 10) {
405 "BTRFS: super block crcs don't match, older mkfs detected\n");
410 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
411 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
420 * helper to read a given tree block, doing retries as required when
421 * the checksums don't match and we have alternate mirrors to try.
423 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
424 struct extent_buffer *eb,
425 u64 start, u64 parent_transid)
427 struct extent_io_tree *io_tree;
432 int failed_mirror = 0;
434 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
435 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
437 ret = read_extent_buffer_pages(io_tree, eb, start,
439 btree_get_extent, mirror_num);
441 if (!verify_parent_transid(io_tree, eb,
449 * This buffer's crc is fine, but its contents are corrupted, so
450 * there is no reason to read the other copies, they won't be
453 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
456 num_copies = btrfs_num_copies(root->fs_info,
461 if (!failed_mirror) {
463 failed_mirror = eb->read_mirror;
467 if (mirror_num == failed_mirror)
470 if (mirror_num > num_copies)
474 if (failed && !ret && failed_mirror)
475 repair_eb_io_failure(root, eb, failed_mirror);
481 * checksum a dirty tree block before IO. This has extra checks to make sure
482 * we only fill in the checksum field in the first page of a multi-page block
485 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
487 u64 start = page_offset(page);
489 struct extent_buffer *eb;
491 eb = (struct extent_buffer *)page->private;
492 if (page != eb->pages[0])
494 found_start = btrfs_header_bytenr(eb);
495 if (WARN_ON(found_start != start || !PageUptodate(page)))
497 csum_tree_block(root, eb, 0);
501 static int check_tree_block_fsid(struct btrfs_root *root,
502 struct extent_buffer *eb)
504 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
505 u8 fsid[BTRFS_UUID_SIZE];
508 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
510 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
514 fs_devices = fs_devices->seed;
519 #define CORRUPT(reason, eb, root, slot) \
520 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
521 "root=%llu, slot=%d", reason, \
522 btrfs_header_bytenr(eb), root->objectid, slot)
524 static noinline int check_leaf(struct btrfs_root *root,
525 struct extent_buffer *leaf)
527 struct btrfs_key key;
528 struct btrfs_key leaf_key;
529 u32 nritems = btrfs_header_nritems(leaf);
535 /* Check the 0 item */
536 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
537 BTRFS_LEAF_DATA_SIZE(root)) {
538 CORRUPT("invalid item offset size pair", leaf, root, 0);
543 * Check to make sure each items keys are in the correct order and their
544 * offsets make sense. We only have to loop through nritems-1 because
545 * we check the current slot against the next slot, which verifies the
546 * next slot's offset+size makes sense and that the current's slot
549 for (slot = 0; slot < nritems - 1; slot++) {
550 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
551 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
553 /* Make sure the keys are in the right order */
554 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
555 CORRUPT("bad key order", leaf, root, slot);
560 * Make sure the offset and ends are right, remember that the
561 * item data starts at the end of the leaf and grows towards the
564 if (btrfs_item_offset_nr(leaf, slot) !=
565 btrfs_item_end_nr(leaf, slot + 1)) {
566 CORRUPT("slot offset bad", leaf, root, slot);
571 * Check to make sure that we don't point outside of the leaf,
572 * just incase all the items are consistent to eachother, but
573 * all point outside of the leaf.
575 if (btrfs_item_end_nr(leaf, slot) >
576 BTRFS_LEAF_DATA_SIZE(root)) {
577 CORRUPT("slot end outside of leaf", leaf, root, slot);
585 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
586 u64 phy_offset, struct page *page,
587 u64 start, u64 end, int mirror)
591 struct extent_buffer *eb;
592 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
599 eb = (struct extent_buffer *)page->private;
601 /* the pending IO might have been the only thing that kept this buffer
602 * in memory. Make sure we have a ref for all this other checks
604 extent_buffer_get(eb);
606 reads_done = atomic_dec_and_test(&eb->io_pages);
610 eb->read_mirror = mirror;
611 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
616 found_start = btrfs_header_bytenr(eb);
617 if (found_start != eb->start) {
618 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start "
620 found_start, eb->start);
624 if (check_tree_block_fsid(root, eb)) {
625 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n",
630 found_level = btrfs_header_level(eb);
631 if (found_level >= BTRFS_MAX_LEVEL) {
632 btrfs_info(root->fs_info, "bad tree block level %d",
633 (int)btrfs_header_level(eb));
638 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
641 ret = csum_tree_block(root, eb, 1);
648 * If this is a leaf block and it is corrupt, set the corrupt bit so
649 * that we don't try and read the other copies of this block, just
652 if (found_level == 0 && check_leaf(root, eb)) {
653 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
658 set_extent_buffer_uptodate(eb);
661 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
662 btree_readahead_hook(root, eb, eb->start, ret);
666 * our io error hook is going to dec the io pages
667 * again, we have to make sure it has something
670 atomic_inc(&eb->io_pages);
671 clear_extent_buffer_uptodate(eb);
673 free_extent_buffer(eb);
678 static int btree_io_failed_hook(struct page *page, int failed_mirror)
680 struct extent_buffer *eb;
681 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
683 eb = (struct extent_buffer *)page->private;
684 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
685 eb->read_mirror = failed_mirror;
686 atomic_dec(&eb->io_pages);
687 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
688 btree_readahead_hook(root, eb, eb->start, -EIO);
689 return -EIO; /* we fixed nothing */
692 static void end_workqueue_bio(struct bio *bio, int err)
694 struct end_io_wq *end_io_wq = bio->bi_private;
695 struct btrfs_fs_info *fs_info;
697 fs_info = end_io_wq->info;
698 end_io_wq->error = err;
699 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
701 if (bio->bi_rw & REQ_WRITE) {
702 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
703 btrfs_queue_work(fs_info->endio_meta_write_workers,
705 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
706 btrfs_queue_work(fs_info->endio_freespace_worker,
708 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
709 btrfs_queue_work(fs_info->endio_raid56_workers,
712 btrfs_queue_work(fs_info->endio_write_workers,
715 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
716 btrfs_queue_work(fs_info->endio_raid56_workers,
718 else if (end_io_wq->metadata)
719 btrfs_queue_work(fs_info->endio_meta_workers,
722 btrfs_queue_work(fs_info->endio_workers,
728 * For the metadata arg you want
731 * 1 - if normal metadta
732 * 2 - if writing to the free space cache area
733 * 3 - raid parity work
735 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
738 struct end_io_wq *end_io_wq;
739 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
743 end_io_wq->private = bio->bi_private;
744 end_io_wq->end_io = bio->bi_end_io;
745 end_io_wq->info = info;
746 end_io_wq->error = 0;
747 end_io_wq->bio = bio;
748 end_io_wq->metadata = metadata;
750 bio->bi_private = end_io_wq;
751 bio->bi_end_io = end_workqueue_bio;
755 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
757 unsigned long limit = min_t(unsigned long,
758 info->thread_pool_size,
759 info->fs_devices->open_devices);
763 static void run_one_async_start(struct btrfs_work *work)
765 struct async_submit_bio *async;
768 async = container_of(work, struct async_submit_bio, work);
769 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
770 async->mirror_num, async->bio_flags,
776 static void run_one_async_done(struct btrfs_work *work)
778 struct btrfs_fs_info *fs_info;
779 struct async_submit_bio *async;
782 async = container_of(work, struct async_submit_bio, work);
783 fs_info = BTRFS_I(async->inode)->root->fs_info;
785 limit = btrfs_async_submit_limit(fs_info);
786 limit = limit * 2 / 3;
788 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
789 waitqueue_active(&fs_info->async_submit_wait))
790 wake_up(&fs_info->async_submit_wait);
792 /* If an error occured we just want to clean up the bio and move on */
794 bio_endio(async->bio, async->error);
798 async->submit_bio_done(async->inode, async->rw, async->bio,
799 async->mirror_num, async->bio_flags,
803 static void run_one_async_free(struct btrfs_work *work)
805 struct async_submit_bio *async;
807 async = container_of(work, struct async_submit_bio, work);
811 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
812 int rw, struct bio *bio, int mirror_num,
813 unsigned long bio_flags,
815 extent_submit_bio_hook_t *submit_bio_start,
816 extent_submit_bio_hook_t *submit_bio_done)
818 struct async_submit_bio *async;
820 async = kmalloc(sizeof(*async), GFP_NOFS);
824 async->inode = inode;
827 async->mirror_num = mirror_num;
828 async->submit_bio_start = submit_bio_start;
829 async->submit_bio_done = submit_bio_done;
831 btrfs_init_work(&async->work, run_one_async_start,
832 run_one_async_done, run_one_async_free);
834 async->bio_flags = bio_flags;
835 async->bio_offset = bio_offset;
839 atomic_inc(&fs_info->nr_async_submits);
842 btrfs_set_work_high_priority(&async->work);
844 btrfs_queue_work(fs_info->workers, &async->work);
846 while (atomic_read(&fs_info->async_submit_draining) &&
847 atomic_read(&fs_info->nr_async_submits)) {
848 wait_event(fs_info->async_submit_wait,
849 (atomic_read(&fs_info->nr_async_submits) == 0));
855 static int btree_csum_one_bio(struct bio *bio)
857 struct bio_vec *bvec;
858 struct btrfs_root *root;
861 bio_for_each_segment_all(bvec, bio, i) {
862 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
863 ret = csum_dirty_buffer(root, bvec->bv_page);
871 static int __btree_submit_bio_start(struct inode *inode, int rw,
872 struct bio *bio, int mirror_num,
873 unsigned long bio_flags,
877 * when we're called for a write, we're already in the async
878 * submission context. Just jump into btrfs_map_bio
880 return btree_csum_one_bio(bio);
883 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
884 int mirror_num, unsigned long bio_flags,
890 * when we're called for a write, we're already in the async
891 * submission context. Just jump into btrfs_map_bio
893 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
899 static int check_async_write(struct inode *inode, unsigned long bio_flags)
901 if (bio_flags & EXTENT_BIO_TREE_LOG)
910 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
911 int mirror_num, unsigned long bio_flags,
914 int async = check_async_write(inode, bio_flags);
917 if (!(rw & REQ_WRITE)) {
919 * called for a read, do the setup so that checksum validation
920 * can happen in the async kernel threads
922 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
926 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
929 ret = btree_csum_one_bio(bio);
932 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
936 * kthread helpers are used to submit writes so that
937 * checksumming can happen in parallel across all CPUs
939 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
940 inode, rw, bio, mirror_num, 0,
942 __btree_submit_bio_start,
943 __btree_submit_bio_done);
953 #ifdef CONFIG_MIGRATION
954 static int btree_migratepage(struct address_space *mapping,
955 struct page *newpage, struct page *page,
956 enum migrate_mode mode)
959 * we can't safely write a btree page from here,
960 * we haven't done the locking hook
965 * Buffers may be managed in a filesystem specific way.
966 * We must have no buffers or drop them.
968 if (page_has_private(page) &&
969 !try_to_release_page(page, GFP_KERNEL))
971 return migrate_page(mapping, newpage, page, mode);
976 static int btree_writepages(struct address_space *mapping,
977 struct writeback_control *wbc)
979 struct btrfs_fs_info *fs_info;
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 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1021 "page private not zero on page %llu",
1022 (unsigned long long)page_offset(page));
1023 ClearPagePrivate(page);
1024 set_page_private(page, 0);
1025 page_cache_release(page);
1029 static int btree_set_page_dirty(struct page *page)
1032 struct extent_buffer *eb;
1034 BUG_ON(!PagePrivate(page));
1035 eb = (struct extent_buffer *)page->private;
1037 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1038 BUG_ON(!atomic_read(&eb->refs));
1039 btrfs_assert_tree_locked(eb);
1041 return __set_page_dirty_nobuffers(page);
1044 static const struct address_space_operations btree_aops = {
1045 .readpage = btree_readpage,
1046 .writepages = btree_writepages,
1047 .releasepage = btree_releasepage,
1048 .invalidatepage = btree_invalidatepage,
1049 #ifdef CONFIG_MIGRATION
1050 .migratepage = btree_migratepage,
1052 .set_page_dirty = btree_set_page_dirty,
1055 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1058 struct extent_buffer *buf = NULL;
1059 struct inode *btree_inode = root->fs_info->btree_inode;
1062 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1065 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1066 buf, 0, WAIT_NONE, btree_get_extent, 0);
1067 free_extent_buffer(buf);
1071 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1072 int mirror_num, struct extent_buffer **eb)
1074 struct extent_buffer *buf = NULL;
1075 struct inode *btree_inode = root->fs_info->btree_inode;
1076 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1079 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1083 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1085 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1086 btree_get_extent, mirror_num);
1088 free_extent_buffer(buf);
1092 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1093 free_extent_buffer(buf);
1095 } else if (extent_buffer_uptodate(buf)) {
1098 free_extent_buffer(buf);
1103 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1104 u64 bytenr, u32 blocksize)
1106 return find_extent_buffer(root->fs_info, bytenr);
1109 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1110 u64 bytenr, u32 blocksize)
1112 return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1116 int btrfs_write_tree_block(struct extent_buffer *buf)
1118 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1119 buf->start + buf->len - 1);
1122 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1124 return filemap_fdatawait_range(buf->pages[0]->mapping,
1125 buf->start, buf->start + buf->len - 1);
1128 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1129 u32 blocksize, u64 parent_transid)
1131 struct extent_buffer *buf = NULL;
1134 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1138 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1140 free_extent_buffer(buf);
1147 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1148 struct extent_buffer *buf)
1150 struct btrfs_fs_info *fs_info = root->fs_info;
1152 if (btrfs_header_generation(buf) ==
1153 fs_info->running_transaction->transid) {
1154 btrfs_assert_tree_locked(buf);
1156 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1157 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1159 fs_info->dirty_metadata_batch);
1160 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1161 btrfs_set_lock_blocking(buf);
1162 clear_extent_buffer_dirty(buf);
1167 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1169 struct btrfs_subvolume_writers *writers;
1172 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1174 return ERR_PTR(-ENOMEM);
1176 ret = percpu_counter_init(&writers->counter, 0);
1179 return ERR_PTR(ret);
1182 init_waitqueue_head(&writers->wait);
1187 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1189 percpu_counter_destroy(&writers->counter);
1193 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1194 u32 stripesize, struct btrfs_root *root,
1195 struct btrfs_fs_info *fs_info,
1199 root->commit_root = NULL;
1200 root->sectorsize = sectorsize;
1201 root->nodesize = nodesize;
1202 root->leafsize = leafsize;
1203 root->stripesize = stripesize;
1205 root->orphan_cleanup_state = 0;
1207 root->objectid = objectid;
1208 root->last_trans = 0;
1209 root->highest_objectid = 0;
1210 root->nr_delalloc_inodes = 0;
1211 root->nr_ordered_extents = 0;
1213 root->inode_tree = RB_ROOT;
1214 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1215 root->block_rsv = NULL;
1216 root->orphan_block_rsv = NULL;
1218 INIT_LIST_HEAD(&root->dirty_list);
1219 INIT_LIST_HEAD(&root->root_list);
1220 INIT_LIST_HEAD(&root->delalloc_inodes);
1221 INIT_LIST_HEAD(&root->delalloc_root);
1222 INIT_LIST_HEAD(&root->ordered_extents);
1223 INIT_LIST_HEAD(&root->ordered_root);
1224 INIT_LIST_HEAD(&root->logged_list[0]);
1225 INIT_LIST_HEAD(&root->logged_list[1]);
1226 spin_lock_init(&root->orphan_lock);
1227 spin_lock_init(&root->inode_lock);
1228 spin_lock_init(&root->delalloc_lock);
1229 spin_lock_init(&root->ordered_extent_lock);
1230 spin_lock_init(&root->accounting_lock);
1231 spin_lock_init(&root->log_extents_lock[0]);
1232 spin_lock_init(&root->log_extents_lock[1]);
1233 mutex_init(&root->objectid_mutex);
1234 mutex_init(&root->log_mutex);
1235 mutex_init(&root->ordered_extent_mutex);
1236 mutex_init(&root->delalloc_mutex);
1237 init_waitqueue_head(&root->log_writer_wait);
1238 init_waitqueue_head(&root->log_commit_wait[0]);
1239 init_waitqueue_head(&root->log_commit_wait[1]);
1240 INIT_LIST_HEAD(&root->log_ctxs[0]);
1241 INIT_LIST_HEAD(&root->log_ctxs[1]);
1242 atomic_set(&root->log_commit[0], 0);
1243 atomic_set(&root->log_commit[1], 0);
1244 atomic_set(&root->log_writers, 0);
1245 atomic_set(&root->log_batch, 0);
1246 atomic_set(&root->orphan_inodes, 0);
1247 atomic_set(&root->refs, 1);
1248 atomic_set(&root->will_be_snapshoted, 0);
1249 root->log_transid = 0;
1250 root->log_transid_committed = -1;
1251 root->last_log_commit = 0;
1253 extent_io_tree_init(&root->dirty_log_pages,
1254 fs_info->btree_inode->i_mapping);
1256 memset(&root->root_key, 0, sizeof(root->root_key));
1257 memset(&root->root_item, 0, sizeof(root->root_item));
1258 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1259 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1261 root->defrag_trans_start = fs_info->generation;
1263 root->defrag_trans_start = 0;
1264 init_completion(&root->kobj_unregister);
1265 root->root_key.objectid = objectid;
1268 spin_lock_init(&root->root_item_lock);
1271 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1273 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1275 root->fs_info = fs_info;
1279 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1280 /* Should only be used by the testing infrastructure */
1281 struct btrfs_root *btrfs_alloc_dummy_root(void)
1283 struct btrfs_root *root;
1285 root = btrfs_alloc_root(NULL);
1287 return ERR_PTR(-ENOMEM);
1288 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1289 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1295 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1296 struct btrfs_fs_info *fs_info,
1299 struct extent_buffer *leaf;
1300 struct btrfs_root *tree_root = fs_info->tree_root;
1301 struct btrfs_root *root;
1302 struct btrfs_key key;
1306 root = btrfs_alloc_root(fs_info);
1308 return ERR_PTR(-ENOMEM);
1310 __setup_root(tree_root->nodesize, tree_root->leafsize,
1311 tree_root->sectorsize, tree_root->stripesize,
1312 root, fs_info, objectid);
1313 root->root_key.objectid = objectid;
1314 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1315 root->root_key.offset = 0;
1317 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1318 0, objectid, NULL, 0, 0, 0);
1320 ret = PTR_ERR(leaf);
1325 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1326 btrfs_set_header_bytenr(leaf, leaf->start);
1327 btrfs_set_header_generation(leaf, trans->transid);
1328 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1329 btrfs_set_header_owner(leaf, objectid);
1332 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1334 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1335 btrfs_header_chunk_tree_uuid(leaf),
1337 btrfs_mark_buffer_dirty(leaf);
1339 root->commit_root = btrfs_root_node(root);
1340 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1342 root->root_item.flags = 0;
1343 root->root_item.byte_limit = 0;
1344 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1345 btrfs_set_root_generation(&root->root_item, trans->transid);
1346 btrfs_set_root_level(&root->root_item, 0);
1347 btrfs_set_root_refs(&root->root_item, 1);
1348 btrfs_set_root_used(&root->root_item, leaf->len);
1349 btrfs_set_root_last_snapshot(&root->root_item, 0);
1350 btrfs_set_root_dirid(&root->root_item, 0);
1352 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1353 root->root_item.drop_level = 0;
1355 key.objectid = objectid;
1356 key.type = BTRFS_ROOT_ITEM_KEY;
1358 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1362 btrfs_tree_unlock(leaf);
1368 btrfs_tree_unlock(leaf);
1369 free_extent_buffer(root->commit_root);
1370 free_extent_buffer(leaf);
1374 return ERR_PTR(ret);
1377 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1378 struct btrfs_fs_info *fs_info)
1380 struct btrfs_root *root;
1381 struct btrfs_root *tree_root = fs_info->tree_root;
1382 struct extent_buffer *leaf;
1384 root = btrfs_alloc_root(fs_info);
1386 return ERR_PTR(-ENOMEM);
1388 __setup_root(tree_root->nodesize, tree_root->leafsize,
1389 tree_root->sectorsize, tree_root->stripesize,
1390 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1392 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1393 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1394 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1397 * DON'T set REF_COWS for log trees
1399 * log trees do not get reference counted because they go away
1400 * before a real commit is actually done. They do store pointers
1401 * to file data extents, and those reference counts still get
1402 * updated (along with back refs to the log tree).
1405 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1406 BTRFS_TREE_LOG_OBJECTID, NULL,
1410 return ERR_CAST(leaf);
1413 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1414 btrfs_set_header_bytenr(leaf, leaf->start);
1415 btrfs_set_header_generation(leaf, trans->transid);
1416 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1417 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1420 write_extent_buffer(root->node, root->fs_info->fsid,
1421 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1422 btrfs_mark_buffer_dirty(root->node);
1423 btrfs_tree_unlock(root->node);
1427 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1428 struct btrfs_fs_info *fs_info)
1430 struct btrfs_root *log_root;
1432 log_root = alloc_log_tree(trans, fs_info);
1433 if (IS_ERR(log_root))
1434 return PTR_ERR(log_root);
1435 WARN_ON(fs_info->log_root_tree);
1436 fs_info->log_root_tree = log_root;
1440 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1441 struct btrfs_root *root)
1443 struct btrfs_root *log_root;
1444 struct btrfs_inode_item *inode_item;
1446 log_root = alloc_log_tree(trans, root->fs_info);
1447 if (IS_ERR(log_root))
1448 return PTR_ERR(log_root);
1450 log_root->last_trans = trans->transid;
1451 log_root->root_key.offset = root->root_key.objectid;
1453 inode_item = &log_root->root_item.inode;
1454 btrfs_set_stack_inode_generation(inode_item, 1);
1455 btrfs_set_stack_inode_size(inode_item, 3);
1456 btrfs_set_stack_inode_nlink(inode_item, 1);
1457 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1458 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1460 btrfs_set_root_node(&log_root->root_item, log_root->node);
1462 WARN_ON(root->log_root);
1463 root->log_root = log_root;
1464 root->log_transid = 0;
1465 root->log_transid_committed = -1;
1466 root->last_log_commit = 0;
1470 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1471 struct btrfs_key *key)
1473 struct btrfs_root *root;
1474 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1475 struct btrfs_path *path;
1480 path = btrfs_alloc_path();
1482 return ERR_PTR(-ENOMEM);
1484 root = btrfs_alloc_root(fs_info);
1490 __setup_root(tree_root->nodesize, tree_root->leafsize,
1491 tree_root->sectorsize, tree_root->stripesize,
1492 root, fs_info, key->objectid);
1494 ret = btrfs_find_root(tree_root, key, path,
1495 &root->root_item, &root->root_key);
1502 generation = btrfs_root_generation(&root->root_item);
1503 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1504 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1505 blocksize, generation);
1509 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1513 root->commit_root = btrfs_root_node(root);
1515 btrfs_free_path(path);
1519 free_extent_buffer(root->node);
1523 root = ERR_PTR(ret);
1527 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1528 struct btrfs_key *location)
1530 struct btrfs_root *root;
1532 root = btrfs_read_tree_root(tree_root, location);
1536 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1537 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1538 btrfs_check_and_init_root_item(&root->root_item);
1544 int btrfs_init_fs_root(struct btrfs_root *root)
1547 struct btrfs_subvolume_writers *writers;
1549 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1550 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1552 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1557 writers = btrfs_alloc_subvolume_writers();
1558 if (IS_ERR(writers)) {
1559 ret = PTR_ERR(writers);
1562 root->subv_writers = writers;
1564 btrfs_init_free_ino_ctl(root);
1565 spin_lock_init(&root->cache_lock);
1566 init_waitqueue_head(&root->cache_wait);
1568 ret = get_anon_bdev(&root->anon_dev);
1574 btrfs_free_subvolume_writers(root->subv_writers);
1576 kfree(root->free_ino_ctl);
1577 kfree(root->free_ino_pinned);
1581 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1584 struct btrfs_root *root;
1586 spin_lock(&fs_info->fs_roots_radix_lock);
1587 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1588 (unsigned long)root_id);
1589 spin_unlock(&fs_info->fs_roots_radix_lock);
1593 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1594 struct btrfs_root *root)
1598 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1602 spin_lock(&fs_info->fs_roots_radix_lock);
1603 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1604 (unsigned long)root->root_key.objectid,
1607 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1608 spin_unlock(&fs_info->fs_roots_radix_lock);
1609 radix_tree_preload_end();
1614 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1615 struct btrfs_key *location,
1618 struct btrfs_root *root;
1621 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1622 return fs_info->tree_root;
1623 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1624 return fs_info->extent_root;
1625 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1626 return fs_info->chunk_root;
1627 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1628 return fs_info->dev_root;
1629 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1630 return fs_info->csum_root;
1631 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1632 return fs_info->quota_root ? fs_info->quota_root :
1634 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1635 return fs_info->uuid_root ? fs_info->uuid_root :
1638 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1640 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1641 return ERR_PTR(-ENOENT);
1645 root = btrfs_read_fs_root(fs_info->tree_root, location);
1649 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1654 ret = btrfs_init_fs_root(root);
1658 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1659 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1663 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1665 ret = btrfs_insert_fs_root(fs_info, root);
1667 if (ret == -EEXIST) {
1676 return ERR_PTR(ret);
1679 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1681 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1683 struct btrfs_device *device;
1684 struct backing_dev_info *bdi;
1687 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1690 bdi = blk_get_backing_dev_info(device->bdev);
1691 if (bdi && bdi_congested(bdi, bdi_bits)) {
1701 * If this fails, caller must call bdi_destroy() to get rid of the
1704 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1708 bdi->capabilities = BDI_CAP_MAP_COPY;
1709 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1713 bdi->ra_pages = default_backing_dev_info.ra_pages;
1714 bdi->congested_fn = btrfs_congested_fn;
1715 bdi->congested_data = info;
1720 * called by the kthread helper functions to finally call the bio end_io
1721 * functions. This is where read checksum verification actually happens
1723 static void end_workqueue_fn(struct btrfs_work *work)
1726 struct end_io_wq *end_io_wq;
1729 end_io_wq = container_of(work, struct end_io_wq, work);
1730 bio = end_io_wq->bio;
1732 error = end_io_wq->error;
1733 bio->bi_private = end_io_wq->private;
1734 bio->bi_end_io = end_io_wq->end_io;
1736 bio_endio_nodec(bio, error);
1739 static int cleaner_kthread(void *arg)
1741 struct btrfs_root *root = arg;
1747 /* Make the cleaner go to sleep early. */
1748 if (btrfs_need_cleaner_sleep(root))
1751 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1755 * Avoid the problem that we change the status of the fs
1756 * during the above check and trylock.
1758 if (btrfs_need_cleaner_sleep(root)) {
1759 mutex_unlock(&root->fs_info->cleaner_mutex);
1763 btrfs_run_delayed_iputs(root);
1764 again = btrfs_clean_one_deleted_snapshot(root);
1765 mutex_unlock(&root->fs_info->cleaner_mutex);
1768 * The defragger has dealt with the R/O remount and umount,
1769 * needn't do anything special here.
1771 btrfs_run_defrag_inodes(root->fs_info);
1773 if (!try_to_freeze() && !again) {
1774 set_current_state(TASK_INTERRUPTIBLE);
1775 if (!kthread_should_stop())
1777 __set_current_state(TASK_RUNNING);
1779 } while (!kthread_should_stop());
1783 static int transaction_kthread(void *arg)
1785 struct btrfs_root *root = arg;
1786 struct btrfs_trans_handle *trans;
1787 struct btrfs_transaction *cur;
1790 unsigned long delay;
1794 cannot_commit = false;
1795 delay = HZ * root->fs_info->commit_interval;
1796 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1798 spin_lock(&root->fs_info->trans_lock);
1799 cur = root->fs_info->running_transaction;
1801 spin_unlock(&root->fs_info->trans_lock);
1805 now = get_seconds();
1806 if (cur->state < TRANS_STATE_BLOCKED &&
1807 (now < cur->start_time ||
1808 now - cur->start_time < root->fs_info->commit_interval)) {
1809 spin_unlock(&root->fs_info->trans_lock);
1813 transid = cur->transid;
1814 spin_unlock(&root->fs_info->trans_lock);
1816 /* If the file system is aborted, this will always fail. */
1817 trans = btrfs_attach_transaction(root);
1818 if (IS_ERR(trans)) {
1819 if (PTR_ERR(trans) != -ENOENT)
1820 cannot_commit = true;
1823 if (transid == trans->transid) {
1824 btrfs_commit_transaction(trans, root);
1826 btrfs_end_transaction(trans, root);
1829 wake_up_process(root->fs_info->cleaner_kthread);
1830 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1832 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1833 &root->fs_info->fs_state)))
1834 btrfs_cleanup_transaction(root);
1835 if (!try_to_freeze()) {
1836 set_current_state(TASK_INTERRUPTIBLE);
1837 if (!kthread_should_stop() &&
1838 (!btrfs_transaction_blocked(root->fs_info) ||
1840 schedule_timeout(delay);
1841 __set_current_state(TASK_RUNNING);
1843 } while (!kthread_should_stop());
1848 * this will find the highest generation in the array of
1849 * root backups. The index of the highest array is returned,
1850 * or -1 if we can't find anything.
1852 * We check to make sure the array is valid by comparing the
1853 * generation of the latest root in the array with the generation
1854 * in the super block. If they don't match we pitch it.
1856 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1859 int newest_index = -1;
1860 struct btrfs_root_backup *root_backup;
1863 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1864 root_backup = info->super_copy->super_roots + i;
1865 cur = btrfs_backup_tree_root_gen(root_backup);
1866 if (cur == newest_gen)
1870 /* check to see if we actually wrapped around */
1871 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1872 root_backup = info->super_copy->super_roots;
1873 cur = btrfs_backup_tree_root_gen(root_backup);
1874 if (cur == newest_gen)
1877 return newest_index;
1882 * find the oldest backup so we know where to store new entries
1883 * in the backup array. This will set the backup_root_index
1884 * field in the fs_info struct
1886 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1889 int newest_index = -1;
1891 newest_index = find_newest_super_backup(info, newest_gen);
1892 /* if there was garbage in there, just move along */
1893 if (newest_index == -1) {
1894 info->backup_root_index = 0;
1896 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1901 * copy all the root pointers into the super backup array.
1902 * this will bump the backup pointer by one when it is
1905 static void backup_super_roots(struct btrfs_fs_info *info)
1908 struct btrfs_root_backup *root_backup;
1911 next_backup = info->backup_root_index;
1912 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1913 BTRFS_NUM_BACKUP_ROOTS;
1916 * just overwrite the last backup if we're at the same generation
1917 * this happens only at umount
1919 root_backup = info->super_for_commit->super_roots + last_backup;
1920 if (btrfs_backup_tree_root_gen(root_backup) ==
1921 btrfs_header_generation(info->tree_root->node))
1922 next_backup = last_backup;
1924 root_backup = info->super_for_commit->super_roots + next_backup;
1927 * make sure all of our padding and empty slots get zero filled
1928 * regardless of which ones we use today
1930 memset(root_backup, 0, sizeof(*root_backup));
1932 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1934 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1935 btrfs_set_backup_tree_root_gen(root_backup,
1936 btrfs_header_generation(info->tree_root->node));
1938 btrfs_set_backup_tree_root_level(root_backup,
1939 btrfs_header_level(info->tree_root->node));
1941 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1942 btrfs_set_backup_chunk_root_gen(root_backup,
1943 btrfs_header_generation(info->chunk_root->node));
1944 btrfs_set_backup_chunk_root_level(root_backup,
1945 btrfs_header_level(info->chunk_root->node));
1947 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1948 btrfs_set_backup_extent_root_gen(root_backup,
1949 btrfs_header_generation(info->extent_root->node));
1950 btrfs_set_backup_extent_root_level(root_backup,
1951 btrfs_header_level(info->extent_root->node));
1954 * we might commit during log recovery, which happens before we set
1955 * the fs_root. Make sure it is valid before we fill it in.
1957 if (info->fs_root && info->fs_root->node) {
1958 btrfs_set_backup_fs_root(root_backup,
1959 info->fs_root->node->start);
1960 btrfs_set_backup_fs_root_gen(root_backup,
1961 btrfs_header_generation(info->fs_root->node));
1962 btrfs_set_backup_fs_root_level(root_backup,
1963 btrfs_header_level(info->fs_root->node));
1966 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1967 btrfs_set_backup_dev_root_gen(root_backup,
1968 btrfs_header_generation(info->dev_root->node));
1969 btrfs_set_backup_dev_root_level(root_backup,
1970 btrfs_header_level(info->dev_root->node));
1972 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1973 btrfs_set_backup_csum_root_gen(root_backup,
1974 btrfs_header_generation(info->csum_root->node));
1975 btrfs_set_backup_csum_root_level(root_backup,
1976 btrfs_header_level(info->csum_root->node));
1978 btrfs_set_backup_total_bytes(root_backup,
1979 btrfs_super_total_bytes(info->super_copy));
1980 btrfs_set_backup_bytes_used(root_backup,
1981 btrfs_super_bytes_used(info->super_copy));
1982 btrfs_set_backup_num_devices(root_backup,
1983 btrfs_super_num_devices(info->super_copy));
1986 * if we don't copy this out to the super_copy, it won't get remembered
1987 * for the next commit
1989 memcpy(&info->super_copy->super_roots,
1990 &info->super_for_commit->super_roots,
1991 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1995 * this copies info out of the root backup array and back into
1996 * the in-memory super block. It is meant to help iterate through
1997 * the array, so you send it the number of backups you've already
1998 * tried and the last backup index you used.
2000 * this returns -1 when it has tried all the backups
2002 static noinline int next_root_backup(struct btrfs_fs_info *info,
2003 struct btrfs_super_block *super,
2004 int *num_backups_tried, int *backup_index)
2006 struct btrfs_root_backup *root_backup;
2007 int newest = *backup_index;
2009 if (*num_backups_tried == 0) {
2010 u64 gen = btrfs_super_generation(super);
2012 newest = find_newest_super_backup(info, gen);
2016 *backup_index = newest;
2017 *num_backups_tried = 1;
2018 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2019 /* we've tried all the backups, all done */
2022 /* jump to the next oldest backup */
2023 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2024 BTRFS_NUM_BACKUP_ROOTS;
2025 *backup_index = newest;
2026 *num_backups_tried += 1;
2028 root_backup = super->super_roots + newest;
2030 btrfs_set_super_generation(super,
2031 btrfs_backup_tree_root_gen(root_backup));
2032 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2033 btrfs_set_super_root_level(super,
2034 btrfs_backup_tree_root_level(root_backup));
2035 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2038 * fixme: the total bytes and num_devices need to match or we should
2041 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2042 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2046 /* helper to cleanup workers */
2047 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2049 btrfs_destroy_workqueue(fs_info->fixup_workers);
2050 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2051 btrfs_destroy_workqueue(fs_info->workers);
2052 btrfs_destroy_workqueue(fs_info->endio_workers);
2053 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2054 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2055 btrfs_destroy_workqueue(fs_info->rmw_workers);
2056 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2057 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2058 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2059 btrfs_destroy_workqueue(fs_info->submit_workers);
2060 btrfs_destroy_workqueue(fs_info->delayed_workers);
2061 btrfs_destroy_workqueue(fs_info->caching_workers);
2062 btrfs_destroy_workqueue(fs_info->readahead_workers);
2063 btrfs_destroy_workqueue(fs_info->flush_workers);
2064 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2067 static void free_root_extent_buffers(struct btrfs_root *root)
2070 free_extent_buffer(root->node);
2071 free_extent_buffer(root->commit_root);
2073 root->commit_root = NULL;
2077 /* helper to cleanup tree roots */
2078 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2080 free_root_extent_buffers(info->tree_root);
2082 free_root_extent_buffers(info->dev_root);
2083 free_root_extent_buffers(info->extent_root);
2084 free_root_extent_buffers(info->csum_root);
2085 free_root_extent_buffers(info->quota_root);
2086 free_root_extent_buffers(info->uuid_root);
2088 free_root_extent_buffers(info->chunk_root);
2091 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2094 struct btrfs_root *gang[8];
2097 while (!list_empty(&fs_info->dead_roots)) {
2098 gang[0] = list_entry(fs_info->dead_roots.next,
2099 struct btrfs_root, root_list);
2100 list_del(&gang[0]->root_list);
2102 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2103 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2105 free_extent_buffer(gang[0]->node);
2106 free_extent_buffer(gang[0]->commit_root);
2107 btrfs_put_fs_root(gang[0]);
2112 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2117 for (i = 0; i < ret; i++)
2118 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2121 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2122 btrfs_free_log_root_tree(NULL, fs_info);
2123 btrfs_destroy_pinned_extent(fs_info->tree_root,
2124 fs_info->pinned_extents);
2128 int open_ctree(struct super_block *sb,
2129 struct btrfs_fs_devices *fs_devices,
2139 struct btrfs_key location;
2140 struct buffer_head *bh;
2141 struct btrfs_super_block *disk_super;
2142 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2143 struct btrfs_root *tree_root;
2144 struct btrfs_root *extent_root;
2145 struct btrfs_root *csum_root;
2146 struct btrfs_root *chunk_root;
2147 struct btrfs_root *dev_root;
2148 struct btrfs_root *quota_root;
2149 struct btrfs_root *uuid_root;
2150 struct btrfs_root *log_tree_root;
2153 int num_backups_tried = 0;
2154 int backup_index = 0;
2156 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2157 bool create_uuid_tree;
2158 bool check_uuid_tree;
2160 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2161 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2162 if (!tree_root || !chunk_root) {
2167 ret = init_srcu_struct(&fs_info->subvol_srcu);
2173 ret = setup_bdi(fs_info, &fs_info->bdi);
2179 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2184 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2185 (1 + ilog2(nr_cpu_ids));
2187 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2190 goto fail_dirty_metadata_bytes;
2193 ret = percpu_counter_init(&fs_info->bio_counter, 0);
2196 goto fail_delalloc_bytes;
2199 fs_info->btree_inode = new_inode(sb);
2200 if (!fs_info->btree_inode) {
2202 goto fail_bio_counter;
2205 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2207 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2208 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2209 INIT_LIST_HEAD(&fs_info->trans_list);
2210 INIT_LIST_HEAD(&fs_info->dead_roots);
2211 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2212 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2213 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2214 spin_lock_init(&fs_info->delalloc_root_lock);
2215 spin_lock_init(&fs_info->trans_lock);
2216 spin_lock_init(&fs_info->fs_roots_radix_lock);
2217 spin_lock_init(&fs_info->delayed_iput_lock);
2218 spin_lock_init(&fs_info->defrag_inodes_lock);
2219 spin_lock_init(&fs_info->free_chunk_lock);
2220 spin_lock_init(&fs_info->tree_mod_seq_lock);
2221 spin_lock_init(&fs_info->super_lock);
2222 spin_lock_init(&fs_info->buffer_lock);
2223 rwlock_init(&fs_info->tree_mod_log_lock);
2224 mutex_init(&fs_info->reloc_mutex);
2225 mutex_init(&fs_info->delalloc_root_mutex);
2226 seqlock_init(&fs_info->profiles_lock);
2228 init_completion(&fs_info->kobj_unregister);
2229 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2230 INIT_LIST_HEAD(&fs_info->space_info);
2231 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2232 btrfs_mapping_init(&fs_info->mapping_tree);
2233 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2234 BTRFS_BLOCK_RSV_GLOBAL);
2235 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2236 BTRFS_BLOCK_RSV_DELALLOC);
2237 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2238 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2239 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2240 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2241 BTRFS_BLOCK_RSV_DELOPS);
2242 atomic_set(&fs_info->nr_async_submits, 0);
2243 atomic_set(&fs_info->async_delalloc_pages, 0);
2244 atomic_set(&fs_info->async_submit_draining, 0);
2245 atomic_set(&fs_info->nr_async_bios, 0);
2246 atomic_set(&fs_info->defrag_running, 0);
2247 atomic64_set(&fs_info->tree_mod_seq, 0);
2249 fs_info->max_inline = 8192 * 1024;
2250 fs_info->metadata_ratio = 0;
2251 fs_info->defrag_inodes = RB_ROOT;
2252 fs_info->free_chunk_space = 0;
2253 fs_info->tree_mod_log = RB_ROOT;
2254 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2255 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2256 /* readahead state */
2257 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2258 spin_lock_init(&fs_info->reada_lock);
2260 fs_info->thread_pool_size = min_t(unsigned long,
2261 num_online_cpus() + 2, 8);
2263 INIT_LIST_HEAD(&fs_info->ordered_roots);
2264 spin_lock_init(&fs_info->ordered_root_lock);
2265 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2267 if (!fs_info->delayed_root) {
2271 btrfs_init_delayed_root(fs_info->delayed_root);
2273 mutex_init(&fs_info->scrub_lock);
2274 atomic_set(&fs_info->scrubs_running, 0);
2275 atomic_set(&fs_info->scrub_pause_req, 0);
2276 atomic_set(&fs_info->scrubs_paused, 0);
2277 atomic_set(&fs_info->scrub_cancel_req, 0);
2278 init_waitqueue_head(&fs_info->replace_wait);
2279 init_waitqueue_head(&fs_info->scrub_pause_wait);
2280 fs_info->scrub_workers_refcnt = 0;
2281 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2282 fs_info->check_integrity_print_mask = 0;
2285 spin_lock_init(&fs_info->balance_lock);
2286 mutex_init(&fs_info->balance_mutex);
2287 atomic_set(&fs_info->balance_running, 0);
2288 atomic_set(&fs_info->balance_pause_req, 0);
2289 atomic_set(&fs_info->balance_cancel_req, 0);
2290 fs_info->balance_ctl = NULL;
2291 init_waitqueue_head(&fs_info->balance_wait_q);
2292 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2294 sb->s_blocksize = 4096;
2295 sb->s_blocksize_bits = blksize_bits(4096);
2296 sb->s_bdi = &fs_info->bdi;
2298 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2299 set_nlink(fs_info->btree_inode, 1);
2301 * we set the i_size on the btree inode to the max possible int.
2302 * the real end of the address space is determined by all of
2303 * the devices in the system
2305 fs_info->btree_inode->i_size = OFFSET_MAX;
2306 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2307 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2309 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2310 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2311 fs_info->btree_inode->i_mapping);
2312 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2313 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2315 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2317 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2318 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2319 sizeof(struct btrfs_key));
2320 set_bit(BTRFS_INODE_DUMMY,
2321 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2322 btrfs_insert_inode_hash(fs_info->btree_inode);
2324 spin_lock_init(&fs_info->block_group_cache_lock);
2325 fs_info->block_group_cache_tree = RB_ROOT;
2326 fs_info->first_logical_byte = (u64)-1;
2328 extent_io_tree_init(&fs_info->freed_extents[0],
2329 fs_info->btree_inode->i_mapping);
2330 extent_io_tree_init(&fs_info->freed_extents[1],
2331 fs_info->btree_inode->i_mapping);
2332 fs_info->pinned_extents = &fs_info->freed_extents[0];
2333 fs_info->do_barriers = 1;
2336 mutex_init(&fs_info->ordered_operations_mutex);
2337 mutex_init(&fs_info->ordered_extent_flush_mutex);
2338 mutex_init(&fs_info->tree_log_mutex);
2339 mutex_init(&fs_info->chunk_mutex);
2340 mutex_init(&fs_info->transaction_kthread_mutex);
2341 mutex_init(&fs_info->cleaner_mutex);
2342 mutex_init(&fs_info->volume_mutex);
2343 init_rwsem(&fs_info->commit_root_sem);
2344 init_rwsem(&fs_info->cleanup_work_sem);
2345 init_rwsem(&fs_info->subvol_sem);
2346 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2347 fs_info->dev_replace.lock_owner = 0;
2348 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2349 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2350 mutex_init(&fs_info->dev_replace.lock_management_lock);
2351 mutex_init(&fs_info->dev_replace.lock);
2353 spin_lock_init(&fs_info->qgroup_lock);
2354 mutex_init(&fs_info->qgroup_ioctl_lock);
2355 fs_info->qgroup_tree = RB_ROOT;
2356 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2357 fs_info->qgroup_seq = 1;
2358 fs_info->quota_enabled = 0;
2359 fs_info->pending_quota_state = 0;
2360 fs_info->qgroup_ulist = NULL;
2361 mutex_init(&fs_info->qgroup_rescan_lock);
2363 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2364 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2366 init_waitqueue_head(&fs_info->transaction_throttle);
2367 init_waitqueue_head(&fs_info->transaction_wait);
2368 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2369 init_waitqueue_head(&fs_info->async_submit_wait);
2371 ret = btrfs_alloc_stripe_hash_table(fs_info);
2377 __setup_root(4096, 4096, 4096, 4096, tree_root,
2378 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2380 invalidate_bdev(fs_devices->latest_bdev);
2383 * Read super block and check the signature bytes only
2385 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2392 * We want to check superblock checksum, the type is stored inside.
2393 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2395 if (btrfs_check_super_csum(bh->b_data)) {
2396 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2402 * super_copy is zeroed at allocation time and we never touch the
2403 * following bytes up to INFO_SIZE, the checksum is calculated from
2404 * the whole block of INFO_SIZE
2406 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2407 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2408 sizeof(*fs_info->super_for_commit));
2411 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2413 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2415 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2420 disk_super = fs_info->super_copy;
2421 if (!btrfs_super_root(disk_super))
2424 /* check FS state, whether FS is broken. */
2425 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2426 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2429 * run through our array of backup supers and setup
2430 * our ring pointer to the oldest one
2432 generation = btrfs_super_generation(disk_super);
2433 find_oldest_super_backup(fs_info, generation);
2436 * In the long term, we'll store the compression type in the super
2437 * block, and it'll be used for per file compression control.
2439 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2441 ret = btrfs_parse_options(tree_root, options);
2447 features = btrfs_super_incompat_flags(disk_super) &
2448 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2450 printk(KERN_ERR "BTRFS: couldn't mount because of "
2451 "unsupported optional features (%Lx).\n",
2457 if (btrfs_super_leafsize(disk_super) !=
2458 btrfs_super_nodesize(disk_super)) {
2459 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2460 "blocksizes don't match. node %d leaf %d\n",
2461 btrfs_super_nodesize(disk_super),
2462 btrfs_super_leafsize(disk_super));
2466 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2467 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2468 "blocksize (%d) was too large\n",
2469 btrfs_super_leafsize(disk_super));
2474 features = btrfs_super_incompat_flags(disk_super);
2475 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2476 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2477 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2479 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2480 printk(KERN_ERR "BTRFS: has skinny extents\n");
2483 * flag our filesystem as having big metadata blocks if
2484 * they are bigger than the page size
2486 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2487 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2488 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2489 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2492 nodesize = btrfs_super_nodesize(disk_super);
2493 leafsize = btrfs_super_leafsize(disk_super);
2494 sectorsize = btrfs_super_sectorsize(disk_super);
2495 stripesize = btrfs_super_stripesize(disk_super);
2496 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2497 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2500 * mixed block groups end up with duplicate but slightly offset
2501 * extent buffers for the same range. It leads to corruptions
2503 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2504 (sectorsize != leafsize)) {
2505 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2506 "are not allowed for mixed block groups on %s\n",
2512 * Needn't use the lock because there is no other task which will
2515 btrfs_set_super_incompat_flags(disk_super, features);
2517 features = btrfs_super_compat_ro_flags(disk_super) &
2518 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2519 if (!(sb->s_flags & MS_RDONLY) && features) {
2520 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2521 "unsupported option features (%Lx).\n",
2527 max_active = fs_info->thread_pool_size;
2530 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2533 fs_info->delalloc_workers =
2534 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2536 fs_info->flush_workers =
2537 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2539 fs_info->caching_workers =
2540 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2543 * a higher idle thresh on the submit workers makes it much more
2544 * likely that bios will be send down in a sane order to the
2547 fs_info->submit_workers =
2548 btrfs_alloc_workqueue("submit", flags,
2549 min_t(u64, fs_devices->num_devices,
2552 fs_info->fixup_workers =
2553 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2556 * endios are largely parallel and should have a very
2559 fs_info->endio_workers =
2560 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2561 fs_info->endio_meta_workers =
2562 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2563 fs_info->endio_meta_write_workers =
2564 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2565 fs_info->endio_raid56_workers =
2566 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2567 fs_info->rmw_workers =
2568 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2569 fs_info->endio_write_workers =
2570 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2571 fs_info->endio_freespace_worker =
2572 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2573 fs_info->delayed_workers =
2574 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2575 fs_info->readahead_workers =
2576 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2577 fs_info->qgroup_rescan_workers =
2578 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2580 if (!(fs_info->workers && fs_info->delalloc_workers &&
2581 fs_info->submit_workers && fs_info->flush_workers &&
2582 fs_info->endio_workers && fs_info->endio_meta_workers &&
2583 fs_info->endio_meta_write_workers &&
2584 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2585 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2586 fs_info->caching_workers && fs_info->readahead_workers &&
2587 fs_info->fixup_workers && fs_info->delayed_workers &&
2588 fs_info->qgroup_rescan_workers)) {
2590 goto fail_sb_buffer;
2593 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2594 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2595 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2597 tree_root->nodesize = nodesize;
2598 tree_root->leafsize = leafsize;
2599 tree_root->sectorsize = sectorsize;
2600 tree_root->stripesize = stripesize;
2602 sb->s_blocksize = sectorsize;
2603 sb->s_blocksize_bits = blksize_bits(sectorsize);
2605 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2606 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2607 goto fail_sb_buffer;
2610 if (sectorsize != PAGE_SIZE) {
2611 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2612 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2613 goto fail_sb_buffer;
2616 mutex_lock(&fs_info->chunk_mutex);
2617 ret = btrfs_read_sys_array(tree_root);
2618 mutex_unlock(&fs_info->chunk_mutex);
2620 printk(KERN_WARNING "BTRFS: failed to read the system "
2621 "array on %s\n", sb->s_id);
2622 goto fail_sb_buffer;
2625 blocksize = btrfs_level_size(tree_root,
2626 btrfs_super_chunk_root_level(disk_super));
2627 generation = btrfs_super_chunk_root_generation(disk_super);
2629 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2630 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2632 chunk_root->node = read_tree_block(chunk_root,
2633 btrfs_super_chunk_root(disk_super),
2634 blocksize, generation);
2635 if (!chunk_root->node ||
2636 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2637 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2639 goto fail_tree_roots;
2641 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2642 chunk_root->commit_root = btrfs_root_node(chunk_root);
2644 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2645 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2647 ret = btrfs_read_chunk_tree(chunk_root);
2649 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2651 goto fail_tree_roots;
2655 * keep the device that is marked to be the target device for the
2656 * dev_replace procedure
2658 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2660 if (!fs_devices->latest_bdev) {
2661 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2663 goto fail_tree_roots;
2667 blocksize = btrfs_level_size(tree_root,
2668 btrfs_super_root_level(disk_super));
2669 generation = btrfs_super_generation(disk_super);
2671 tree_root->node = read_tree_block(tree_root,
2672 btrfs_super_root(disk_super),
2673 blocksize, generation);
2674 if (!tree_root->node ||
2675 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2676 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2679 goto recovery_tree_root;
2682 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2683 tree_root->commit_root = btrfs_root_node(tree_root);
2684 btrfs_set_root_refs(&tree_root->root_item, 1);
2686 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2687 location.type = BTRFS_ROOT_ITEM_KEY;
2688 location.offset = 0;
2690 extent_root = btrfs_read_tree_root(tree_root, &location);
2691 if (IS_ERR(extent_root)) {
2692 ret = PTR_ERR(extent_root);
2693 goto recovery_tree_root;
2695 set_bit(BTRFS_ROOT_TRACK_DIRTY, &extent_root->state);
2696 fs_info->extent_root = extent_root;
2698 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2699 dev_root = btrfs_read_tree_root(tree_root, &location);
2700 if (IS_ERR(dev_root)) {
2701 ret = PTR_ERR(dev_root);
2702 goto recovery_tree_root;
2704 set_bit(BTRFS_ROOT_TRACK_DIRTY, &dev_root->state);
2705 fs_info->dev_root = dev_root;
2706 btrfs_init_devices_late(fs_info);
2708 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2709 csum_root = btrfs_read_tree_root(tree_root, &location);
2710 if (IS_ERR(csum_root)) {
2711 ret = PTR_ERR(csum_root);
2712 goto recovery_tree_root;
2714 set_bit(BTRFS_ROOT_TRACK_DIRTY, &csum_root->state);
2715 fs_info->csum_root = csum_root;
2717 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2718 quota_root = btrfs_read_tree_root(tree_root, &location);
2719 if (!IS_ERR(quota_root)) {
2720 set_bit(BTRFS_ROOT_TRACK_DIRTY, "a_root->state);
2721 fs_info->quota_enabled = 1;
2722 fs_info->pending_quota_state = 1;
2723 fs_info->quota_root = quota_root;
2726 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2727 uuid_root = btrfs_read_tree_root(tree_root, &location);
2728 if (IS_ERR(uuid_root)) {
2729 ret = PTR_ERR(uuid_root);
2731 goto recovery_tree_root;
2732 create_uuid_tree = true;
2733 check_uuid_tree = false;
2735 set_bit(BTRFS_ROOT_TRACK_DIRTY, &uuid_root->state);
2736 fs_info->uuid_root = uuid_root;
2737 create_uuid_tree = false;
2739 generation != btrfs_super_uuid_tree_generation(disk_super);
2742 fs_info->generation = generation;
2743 fs_info->last_trans_committed = generation;
2745 ret = btrfs_recover_balance(fs_info);
2747 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2748 goto fail_block_groups;
2751 ret = btrfs_init_dev_stats(fs_info);
2753 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2755 goto fail_block_groups;
2758 ret = btrfs_init_dev_replace(fs_info);
2760 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2761 goto fail_block_groups;
2764 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2766 ret = btrfs_sysfs_add_one(fs_info);
2768 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2769 goto fail_block_groups;
2772 ret = btrfs_init_space_info(fs_info);
2774 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2778 ret = btrfs_read_block_groups(extent_root);
2780 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2783 fs_info->num_tolerated_disk_barrier_failures =
2784 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2785 if (fs_info->fs_devices->missing_devices >
2786 fs_info->num_tolerated_disk_barrier_failures &&
2787 !(sb->s_flags & MS_RDONLY)) {
2788 printk(KERN_WARNING "BTRFS: "
2789 "too many missing devices, writeable mount is not allowed\n");
2793 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2795 if (IS_ERR(fs_info->cleaner_kthread))
2798 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2800 "btrfs-transaction");
2801 if (IS_ERR(fs_info->transaction_kthread))
2804 if (!btrfs_test_opt(tree_root, SSD) &&
2805 !btrfs_test_opt(tree_root, NOSSD) &&
2806 !fs_info->fs_devices->rotating) {
2807 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2809 btrfs_set_opt(fs_info->mount_opt, SSD);
2812 /* Set the real inode map cache flag */
2813 if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2814 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2816 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2817 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2818 ret = btrfsic_mount(tree_root, fs_devices,
2819 btrfs_test_opt(tree_root,
2820 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2822 fs_info->check_integrity_print_mask);
2824 printk(KERN_WARNING "BTRFS: failed to initialize"
2825 " integrity check module %s\n", sb->s_id);
2828 ret = btrfs_read_qgroup_config(fs_info);
2830 goto fail_trans_kthread;
2832 /* do not make disk changes in broken FS */
2833 if (btrfs_super_log_root(disk_super) != 0) {
2834 u64 bytenr = btrfs_super_log_root(disk_super);
2836 if (fs_devices->rw_devices == 0) {
2837 printk(KERN_WARNING "BTRFS: log replay required "
2843 btrfs_level_size(tree_root,
2844 btrfs_super_log_root_level(disk_super));
2846 log_tree_root = btrfs_alloc_root(fs_info);
2847 if (!log_tree_root) {
2852 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2853 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2855 log_tree_root->node = read_tree_block(tree_root, bytenr,
2858 if (!log_tree_root->node ||
2859 !extent_buffer_uptodate(log_tree_root->node)) {
2860 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2861 free_extent_buffer(log_tree_root->node);
2862 kfree(log_tree_root);
2865 /* returns with log_tree_root freed on success */
2866 ret = btrfs_recover_log_trees(log_tree_root);
2868 btrfs_error(tree_root->fs_info, ret,
2869 "Failed to recover log tree");
2870 free_extent_buffer(log_tree_root->node);
2871 kfree(log_tree_root);
2875 if (sb->s_flags & MS_RDONLY) {
2876 ret = btrfs_commit_super(tree_root);
2882 ret = btrfs_find_orphan_roots(tree_root);
2886 if (!(sb->s_flags & MS_RDONLY)) {
2887 ret = btrfs_cleanup_fs_roots(fs_info);
2891 ret = btrfs_recover_relocation(tree_root);
2894 "BTRFS: failed to recover relocation\n");
2900 location.objectid = BTRFS_FS_TREE_OBJECTID;
2901 location.type = BTRFS_ROOT_ITEM_KEY;
2902 location.offset = 0;
2904 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2905 if (IS_ERR(fs_info->fs_root)) {
2906 err = PTR_ERR(fs_info->fs_root);
2910 if (sb->s_flags & MS_RDONLY)
2913 down_read(&fs_info->cleanup_work_sem);
2914 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2915 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2916 up_read(&fs_info->cleanup_work_sem);
2917 close_ctree(tree_root);
2920 up_read(&fs_info->cleanup_work_sem);
2922 ret = btrfs_resume_balance_async(fs_info);
2924 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2925 close_ctree(tree_root);
2929 ret = btrfs_resume_dev_replace_async(fs_info);
2931 pr_warn("BTRFS: failed to resume dev_replace\n");
2932 close_ctree(tree_root);
2936 btrfs_qgroup_rescan_resume(fs_info);
2938 if (create_uuid_tree) {
2939 pr_info("BTRFS: creating UUID tree\n");
2940 ret = btrfs_create_uuid_tree(fs_info);
2942 pr_warn("BTRFS: failed to create the UUID tree %d\n",
2944 close_ctree(tree_root);
2947 } else if (check_uuid_tree ||
2948 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2949 pr_info("BTRFS: checking UUID tree\n");
2950 ret = btrfs_check_uuid_tree(fs_info);
2952 pr_warn("BTRFS: failed to check the UUID tree %d\n",
2954 close_ctree(tree_root);
2958 fs_info->update_uuid_tree_gen = 1;
2964 btrfs_free_qgroup_config(fs_info);
2966 kthread_stop(fs_info->transaction_kthread);
2967 btrfs_cleanup_transaction(fs_info->tree_root);
2968 del_fs_roots(fs_info);
2970 kthread_stop(fs_info->cleaner_kthread);
2973 * make sure we're done with the btree inode before we stop our
2976 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2979 btrfs_sysfs_remove_one(fs_info);
2982 btrfs_put_block_group_cache(fs_info);
2983 btrfs_free_block_groups(fs_info);
2986 free_root_pointers(fs_info, 1);
2987 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2990 btrfs_stop_all_workers(fs_info);
2993 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2995 iput(fs_info->btree_inode);
2997 percpu_counter_destroy(&fs_info->bio_counter);
2998 fail_delalloc_bytes:
2999 percpu_counter_destroy(&fs_info->delalloc_bytes);
3000 fail_dirty_metadata_bytes:
3001 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3003 bdi_destroy(&fs_info->bdi);
3005 cleanup_srcu_struct(&fs_info->subvol_srcu);
3007 btrfs_free_stripe_hash_table(fs_info);
3008 btrfs_close_devices(fs_info->fs_devices);
3012 if (!btrfs_test_opt(tree_root, RECOVERY))
3013 goto fail_tree_roots;
3015 free_root_pointers(fs_info, 0);
3017 /* don't use the log in recovery mode, it won't be valid */
3018 btrfs_set_super_log_root(disk_super, 0);
3020 /* we can't trust the free space cache either */
3021 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3023 ret = next_root_backup(fs_info, fs_info->super_copy,
3024 &num_backups_tried, &backup_index);
3026 goto fail_block_groups;
3027 goto retry_root_backup;
3030 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3033 set_buffer_uptodate(bh);
3035 struct btrfs_device *device = (struct btrfs_device *)
3038 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3039 "I/O error on %s\n",
3040 rcu_str_deref(device->name));
3041 /* note, we dont' set_buffer_write_io_error because we have
3042 * our own ways of dealing with the IO errors
3044 clear_buffer_uptodate(bh);
3045 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3051 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3053 struct buffer_head *bh;
3054 struct buffer_head *latest = NULL;
3055 struct btrfs_super_block *super;
3060 /* we would like to check all the supers, but that would make
3061 * a btrfs mount succeed after a mkfs from a different FS.
3062 * So, we need to add a special mount option to scan for
3063 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3065 for (i = 0; i < 1; i++) {
3066 bytenr = btrfs_sb_offset(i);
3067 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3068 i_size_read(bdev->bd_inode))
3070 bh = __bread(bdev, bytenr / 4096,
3071 BTRFS_SUPER_INFO_SIZE);
3075 super = (struct btrfs_super_block *)bh->b_data;
3076 if (btrfs_super_bytenr(super) != bytenr ||
3077 btrfs_super_magic(super) != BTRFS_MAGIC) {
3082 if (!latest || btrfs_super_generation(super) > transid) {
3085 transid = btrfs_super_generation(super);
3094 * this should be called twice, once with wait == 0 and
3095 * once with wait == 1. When wait == 0 is done, all the buffer heads
3096 * we write are pinned.
3098 * They are released when wait == 1 is done.
3099 * max_mirrors must be the same for both runs, and it indicates how
3100 * many supers on this one device should be written.
3102 * max_mirrors == 0 means to write them all.
3104 static int write_dev_supers(struct btrfs_device *device,
3105 struct btrfs_super_block *sb,
3106 int do_barriers, int wait, int max_mirrors)
3108 struct buffer_head *bh;
3115 if (max_mirrors == 0)
3116 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3118 for (i = 0; i < max_mirrors; i++) {
3119 bytenr = btrfs_sb_offset(i);
3120 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3124 bh = __find_get_block(device->bdev, bytenr / 4096,
3125 BTRFS_SUPER_INFO_SIZE);
3131 if (!buffer_uptodate(bh))
3134 /* drop our reference */
3137 /* drop the reference from the wait == 0 run */
3141 btrfs_set_super_bytenr(sb, bytenr);
3144 crc = btrfs_csum_data((char *)sb +
3145 BTRFS_CSUM_SIZE, crc,
3146 BTRFS_SUPER_INFO_SIZE -
3148 btrfs_csum_final(crc, sb->csum);
3151 * one reference for us, and we leave it for the
3154 bh = __getblk(device->bdev, bytenr / 4096,
3155 BTRFS_SUPER_INFO_SIZE);
3157 printk(KERN_ERR "BTRFS: couldn't get super "
3158 "buffer head for bytenr %Lu\n", bytenr);
3163 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3165 /* one reference for submit_bh */
3168 set_buffer_uptodate(bh);
3170 bh->b_end_io = btrfs_end_buffer_write_sync;
3171 bh->b_private = device;
3175 * we fua the first super. The others we allow
3179 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3181 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3185 return errors < i ? 0 : -1;
3189 * endio for the write_dev_flush, this will wake anyone waiting
3190 * for the barrier when it is done
3192 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3195 if (err == -EOPNOTSUPP)
3196 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3197 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3199 if (bio->bi_private)
3200 complete(bio->bi_private);
3205 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3206 * sent down. With wait == 1, it waits for the previous flush.
3208 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3211 static int write_dev_flush(struct btrfs_device *device, int wait)
3216 if (device->nobarriers)
3220 bio = device->flush_bio;
3224 wait_for_completion(&device->flush_wait);
3226 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3227 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3228 rcu_str_deref(device->name));
3229 device->nobarriers = 1;
3230 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3232 btrfs_dev_stat_inc_and_print(device,
3233 BTRFS_DEV_STAT_FLUSH_ERRS);
3236 /* drop the reference from the wait == 0 run */
3238 device->flush_bio = NULL;
3244 * one reference for us, and we leave it for the
3247 device->flush_bio = NULL;
3248 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3252 bio->bi_end_io = btrfs_end_empty_barrier;
3253 bio->bi_bdev = device->bdev;
3254 init_completion(&device->flush_wait);
3255 bio->bi_private = &device->flush_wait;
3256 device->flush_bio = bio;
3259 btrfsic_submit_bio(WRITE_FLUSH, bio);
3265 * send an empty flush down to each device in parallel,
3266 * then wait for them
3268 static int barrier_all_devices(struct btrfs_fs_info *info)
3270 struct list_head *head;
3271 struct btrfs_device *dev;
3272 int errors_send = 0;
3273 int errors_wait = 0;
3276 /* send down all the barriers */
3277 head = &info->fs_devices->devices;
3278 list_for_each_entry_rcu(dev, head, dev_list) {
3285 if (!dev->in_fs_metadata || !dev->writeable)
3288 ret = write_dev_flush(dev, 0);
3293 /* wait for all the barriers */
3294 list_for_each_entry_rcu(dev, head, dev_list) {
3301 if (!dev->in_fs_metadata || !dev->writeable)
3304 ret = write_dev_flush(dev, 1);
3308 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3309 errors_wait > info->num_tolerated_disk_barrier_failures)
3314 int btrfs_calc_num_tolerated_disk_barrier_failures(
3315 struct btrfs_fs_info *fs_info)
3317 struct btrfs_ioctl_space_info space;
3318 struct btrfs_space_info *sinfo;
3319 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3320 BTRFS_BLOCK_GROUP_SYSTEM,
3321 BTRFS_BLOCK_GROUP_METADATA,
3322 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3326 int num_tolerated_disk_barrier_failures =
3327 (int)fs_info->fs_devices->num_devices;
3329 for (i = 0; i < num_types; i++) {
3330 struct btrfs_space_info *tmp;
3334 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3335 if (tmp->flags == types[i]) {
3345 down_read(&sinfo->groups_sem);
3346 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3347 if (!list_empty(&sinfo->block_groups[c])) {
3350 btrfs_get_block_group_info(
3351 &sinfo->block_groups[c], &space);
3352 if (space.total_bytes == 0 ||
3353 space.used_bytes == 0)
3355 flags = space.flags;
3358 * 0: if dup, single or RAID0 is configured for
3359 * any of metadata, system or data, else
3360 * 1: if RAID5 is configured, or if RAID1 or
3361 * RAID10 is configured and only two mirrors
3363 * 2: if RAID6 is configured, else
3364 * num_mirrors - 1: if RAID1 or RAID10 is
3365 * configured and more than
3366 * 2 mirrors are used.
3368 if (num_tolerated_disk_barrier_failures > 0 &&
3369 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3370 BTRFS_BLOCK_GROUP_RAID0)) ||
3371 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3373 num_tolerated_disk_barrier_failures = 0;
3374 else if (num_tolerated_disk_barrier_failures > 1) {
3375 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3376 BTRFS_BLOCK_GROUP_RAID5 |
3377 BTRFS_BLOCK_GROUP_RAID10)) {
3378 num_tolerated_disk_barrier_failures = 1;
3380 BTRFS_BLOCK_GROUP_RAID6) {
3381 num_tolerated_disk_barrier_failures = 2;
3386 up_read(&sinfo->groups_sem);
3389 return num_tolerated_disk_barrier_failures;
3392 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3394 struct list_head *head;
3395 struct btrfs_device *dev;
3396 struct btrfs_super_block *sb;
3397 struct btrfs_dev_item *dev_item;
3401 int total_errors = 0;
3404 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3405 backup_super_roots(root->fs_info);
3407 sb = root->fs_info->super_for_commit;
3408 dev_item = &sb->dev_item;
3410 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3411 head = &root->fs_info->fs_devices->devices;
3412 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3415 ret = barrier_all_devices(root->fs_info);
3418 &root->fs_info->fs_devices->device_list_mutex);
3419 btrfs_error(root->fs_info, ret,
3420 "errors while submitting device barriers.");
3425 list_for_each_entry_rcu(dev, head, dev_list) {
3430 if (!dev->in_fs_metadata || !dev->writeable)
3433 btrfs_set_stack_device_generation(dev_item, 0);
3434 btrfs_set_stack_device_type(dev_item, dev->type);
3435 btrfs_set_stack_device_id(dev_item, dev->devid);
3436 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3437 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3438 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3439 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3440 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3441 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3442 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3444 flags = btrfs_super_flags(sb);
3445 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3447 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3451 if (total_errors > max_errors) {
3452 btrfs_err(root->fs_info, "%d errors while writing supers",
3454 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3456 /* FUA is masked off if unsupported and can't be the reason */
3457 btrfs_error(root->fs_info, -EIO,
3458 "%d errors while writing supers", total_errors);
3463 list_for_each_entry_rcu(dev, head, dev_list) {
3466 if (!dev->in_fs_metadata || !dev->writeable)
3469 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3473 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3474 if (total_errors > max_errors) {
3475 btrfs_error(root->fs_info, -EIO,
3476 "%d errors while writing supers", total_errors);
3482 int write_ctree_super(struct btrfs_trans_handle *trans,
3483 struct btrfs_root *root, int max_mirrors)
3485 return write_all_supers(root, max_mirrors);
3488 /* Drop a fs root from the radix tree and free it. */
3489 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3490 struct btrfs_root *root)
3492 spin_lock(&fs_info->fs_roots_radix_lock);
3493 radix_tree_delete(&fs_info->fs_roots_radix,
3494 (unsigned long)root->root_key.objectid);
3495 spin_unlock(&fs_info->fs_roots_radix_lock);
3497 if (btrfs_root_refs(&root->root_item) == 0)
3498 synchronize_srcu(&fs_info->subvol_srcu);
3500 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3501 btrfs_free_log(NULL, root);
3503 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3504 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3508 static void free_fs_root(struct btrfs_root *root)
3510 iput(root->cache_inode);
3511 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3512 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3513 root->orphan_block_rsv = NULL;
3515 free_anon_bdev(root->anon_dev);
3516 if (root->subv_writers)
3517 btrfs_free_subvolume_writers(root->subv_writers);
3518 free_extent_buffer(root->node);
3519 free_extent_buffer(root->commit_root);
3520 kfree(root->free_ino_ctl);
3521 kfree(root->free_ino_pinned);
3523 btrfs_put_fs_root(root);
3526 void btrfs_free_fs_root(struct btrfs_root *root)
3531 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3533 u64 root_objectid = 0;
3534 struct btrfs_root *gang[8];
3537 unsigned int ret = 0;
3541 index = srcu_read_lock(&fs_info->subvol_srcu);
3542 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3543 (void **)gang, root_objectid,
3546 srcu_read_unlock(&fs_info->subvol_srcu, index);
3549 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3551 for (i = 0; i < ret; i++) {
3552 /* Avoid to grab roots in dead_roots */
3553 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3557 /* grab all the search result for later use */
3558 gang[i] = btrfs_grab_fs_root(gang[i]);
3560 srcu_read_unlock(&fs_info->subvol_srcu, index);
3562 for (i = 0; i < ret; i++) {
3565 root_objectid = gang[i]->root_key.objectid;
3566 err = btrfs_orphan_cleanup(gang[i]);
3569 btrfs_put_fs_root(gang[i]);
3574 /* release the uncleaned roots due to error */
3575 for (; i < ret; i++) {
3577 btrfs_put_fs_root(gang[i]);
3582 int btrfs_commit_super(struct btrfs_root *root)
3584 struct btrfs_trans_handle *trans;
3586 mutex_lock(&root->fs_info->cleaner_mutex);
3587 btrfs_run_delayed_iputs(root);
3588 mutex_unlock(&root->fs_info->cleaner_mutex);
3589 wake_up_process(root->fs_info->cleaner_kthread);
3591 /* wait until ongoing cleanup work done */
3592 down_write(&root->fs_info->cleanup_work_sem);
3593 up_write(&root->fs_info->cleanup_work_sem);
3595 trans = btrfs_join_transaction(root);
3597 return PTR_ERR(trans);
3598 return btrfs_commit_transaction(trans, root);
3601 int close_ctree(struct btrfs_root *root)
3603 struct btrfs_fs_info *fs_info = root->fs_info;
3606 fs_info->closing = 1;
3609 /* wait for the uuid_scan task to finish */
3610 down(&fs_info->uuid_tree_rescan_sem);
3611 /* avoid complains from lockdep et al., set sem back to initial state */
3612 up(&fs_info->uuid_tree_rescan_sem);
3614 /* pause restriper - we want to resume on mount */
3615 btrfs_pause_balance(fs_info);
3617 btrfs_dev_replace_suspend_for_unmount(fs_info);
3619 btrfs_scrub_cancel(fs_info);
3621 /* wait for any defraggers to finish */
3622 wait_event(fs_info->transaction_wait,
3623 (atomic_read(&fs_info->defrag_running) == 0));
3625 /* clear out the rbtree of defraggable inodes */
3626 btrfs_cleanup_defrag_inodes(fs_info);
3628 cancel_work_sync(&fs_info->async_reclaim_work);
3630 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3631 ret = btrfs_commit_super(root);
3633 btrfs_err(root->fs_info, "commit super ret %d", ret);
3636 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3637 btrfs_error_commit_super(root);
3639 kthread_stop(fs_info->transaction_kthread);
3640 kthread_stop(fs_info->cleaner_kthread);
3642 fs_info->closing = 2;
3645 btrfs_free_qgroup_config(root->fs_info);
3647 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3648 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3649 percpu_counter_sum(&fs_info->delalloc_bytes));
3652 btrfs_sysfs_remove_one(fs_info);
3654 del_fs_roots(fs_info);
3656 btrfs_put_block_group_cache(fs_info);
3658 btrfs_free_block_groups(fs_info);
3661 * we must make sure there is not any read request to
3662 * submit after we stopping all workers.
3664 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3665 btrfs_stop_all_workers(fs_info);
3667 free_root_pointers(fs_info, 1);
3669 iput(fs_info->btree_inode);
3671 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3672 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3673 btrfsic_unmount(root, fs_info->fs_devices);
3676 btrfs_close_devices(fs_info->fs_devices);
3677 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3679 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3680 percpu_counter_destroy(&fs_info->delalloc_bytes);
3681 percpu_counter_destroy(&fs_info->bio_counter);
3682 bdi_destroy(&fs_info->bdi);
3683 cleanup_srcu_struct(&fs_info->subvol_srcu);
3685 btrfs_free_stripe_hash_table(fs_info);
3687 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3688 root->orphan_block_rsv = NULL;
3693 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3697 struct inode *btree_inode = buf->pages[0]->mapping->host;
3699 ret = extent_buffer_uptodate(buf);
3703 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3704 parent_transid, atomic);
3710 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3712 return set_extent_buffer_uptodate(buf);
3715 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3717 struct btrfs_root *root;
3718 u64 transid = btrfs_header_generation(buf);
3721 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3723 * This is a fast path so only do this check if we have sanity tests
3724 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3725 * outside of the sanity tests.
3727 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3730 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3731 btrfs_assert_tree_locked(buf);
3732 if (transid != root->fs_info->generation)
3733 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3734 "found %llu running %llu\n",
3735 buf->start, transid, root->fs_info->generation);
3736 was_dirty = set_extent_buffer_dirty(buf);
3738 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3740 root->fs_info->dirty_metadata_batch);
3741 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3742 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3743 btrfs_print_leaf(root, buf);
3749 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3753 * looks as though older kernels can get into trouble with
3754 * this code, they end up stuck in balance_dirty_pages forever
3758 if (current->flags & PF_MEMALLOC)
3762 btrfs_balance_delayed_items(root);
3764 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3765 BTRFS_DIRTY_METADATA_THRESH);
3767 balance_dirty_pages_ratelimited(
3768 root->fs_info->btree_inode->i_mapping);
3773 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3775 __btrfs_btree_balance_dirty(root, 1);
3778 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3780 __btrfs_btree_balance_dirty(root, 0);
3783 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3785 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3786 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3789 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3793 * Placeholder for checks
3798 static void btrfs_error_commit_super(struct btrfs_root *root)
3800 mutex_lock(&root->fs_info->cleaner_mutex);
3801 btrfs_run_delayed_iputs(root);
3802 mutex_unlock(&root->fs_info->cleaner_mutex);
3804 down_write(&root->fs_info->cleanup_work_sem);
3805 up_write(&root->fs_info->cleanup_work_sem);
3807 /* cleanup FS via transaction */
3808 btrfs_cleanup_transaction(root);
3811 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3812 struct btrfs_root *root)
3814 struct btrfs_inode *btrfs_inode;
3815 struct list_head splice;
3817 INIT_LIST_HEAD(&splice);
3819 mutex_lock(&root->fs_info->ordered_operations_mutex);
3820 spin_lock(&root->fs_info->ordered_root_lock);
3822 list_splice_init(&t->ordered_operations, &splice);
3823 while (!list_empty(&splice)) {
3824 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3825 ordered_operations);
3827 list_del_init(&btrfs_inode->ordered_operations);
3828 spin_unlock(&root->fs_info->ordered_root_lock);
3830 btrfs_invalidate_inodes(btrfs_inode->root);
3832 spin_lock(&root->fs_info->ordered_root_lock);
3835 spin_unlock(&root->fs_info->ordered_root_lock);
3836 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3839 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3841 struct btrfs_ordered_extent *ordered;
3843 spin_lock(&root->ordered_extent_lock);
3845 * This will just short circuit the ordered completion stuff which will
3846 * make sure the ordered extent gets properly cleaned up.
3848 list_for_each_entry(ordered, &root->ordered_extents,
3850 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3851 spin_unlock(&root->ordered_extent_lock);
3854 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3856 struct btrfs_root *root;
3857 struct list_head splice;
3859 INIT_LIST_HEAD(&splice);
3861 spin_lock(&fs_info->ordered_root_lock);
3862 list_splice_init(&fs_info->ordered_roots, &splice);
3863 while (!list_empty(&splice)) {
3864 root = list_first_entry(&splice, struct btrfs_root,
3866 list_move_tail(&root->ordered_root,
3867 &fs_info->ordered_roots);
3869 spin_unlock(&fs_info->ordered_root_lock);
3870 btrfs_destroy_ordered_extents(root);
3873 spin_lock(&fs_info->ordered_root_lock);
3875 spin_unlock(&fs_info->ordered_root_lock);
3878 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3879 struct btrfs_root *root)
3881 struct rb_node *node;
3882 struct btrfs_delayed_ref_root *delayed_refs;
3883 struct btrfs_delayed_ref_node *ref;
3886 delayed_refs = &trans->delayed_refs;
3888 spin_lock(&delayed_refs->lock);
3889 if (atomic_read(&delayed_refs->num_entries) == 0) {
3890 spin_unlock(&delayed_refs->lock);
3891 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3895 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3896 struct btrfs_delayed_ref_head *head;
3897 bool pin_bytes = false;
3899 head = rb_entry(node, struct btrfs_delayed_ref_head,
3901 if (!mutex_trylock(&head->mutex)) {
3902 atomic_inc(&head->node.refs);
3903 spin_unlock(&delayed_refs->lock);
3905 mutex_lock(&head->mutex);
3906 mutex_unlock(&head->mutex);
3907 btrfs_put_delayed_ref(&head->node);
3908 spin_lock(&delayed_refs->lock);
3911 spin_lock(&head->lock);
3912 while ((node = rb_first(&head->ref_root)) != NULL) {
3913 ref = rb_entry(node, struct btrfs_delayed_ref_node,
3916 rb_erase(&ref->rb_node, &head->ref_root);
3917 atomic_dec(&delayed_refs->num_entries);
3918 btrfs_put_delayed_ref(ref);
3920 if (head->must_insert_reserved)
3922 btrfs_free_delayed_extent_op(head->extent_op);
3923 delayed_refs->num_heads--;
3924 if (head->processing == 0)
3925 delayed_refs->num_heads_ready--;
3926 atomic_dec(&delayed_refs->num_entries);
3927 head->node.in_tree = 0;
3928 rb_erase(&head->href_node, &delayed_refs->href_root);
3929 spin_unlock(&head->lock);
3930 spin_unlock(&delayed_refs->lock);
3931 mutex_unlock(&head->mutex);
3934 btrfs_pin_extent(root, head->node.bytenr,
3935 head->node.num_bytes, 1);
3936 btrfs_put_delayed_ref(&head->node);
3938 spin_lock(&delayed_refs->lock);
3941 spin_unlock(&delayed_refs->lock);
3946 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3948 struct btrfs_inode *btrfs_inode;
3949 struct list_head splice;
3951 INIT_LIST_HEAD(&splice);
3953 spin_lock(&root->delalloc_lock);
3954 list_splice_init(&root->delalloc_inodes, &splice);
3956 while (!list_empty(&splice)) {
3957 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3960 list_del_init(&btrfs_inode->delalloc_inodes);
3961 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3962 &btrfs_inode->runtime_flags);
3963 spin_unlock(&root->delalloc_lock);
3965 btrfs_invalidate_inodes(btrfs_inode->root);
3967 spin_lock(&root->delalloc_lock);
3970 spin_unlock(&root->delalloc_lock);
3973 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3975 struct btrfs_root *root;
3976 struct list_head splice;
3978 INIT_LIST_HEAD(&splice);
3980 spin_lock(&fs_info->delalloc_root_lock);
3981 list_splice_init(&fs_info->delalloc_roots, &splice);
3982 while (!list_empty(&splice)) {
3983 root = list_first_entry(&splice, struct btrfs_root,
3985 list_del_init(&root->delalloc_root);
3986 root = btrfs_grab_fs_root(root);
3988 spin_unlock(&fs_info->delalloc_root_lock);
3990 btrfs_destroy_delalloc_inodes(root);
3991 btrfs_put_fs_root(root);
3993 spin_lock(&fs_info->delalloc_root_lock);
3995 spin_unlock(&fs_info->delalloc_root_lock);
3998 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3999 struct extent_io_tree *dirty_pages,
4003 struct extent_buffer *eb;
4008 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4013 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4014 while (start <= end) {
4015 eb = btrfs_find_tree_block(root, start,
4017 start += root->leafsize;
4020 wait_on_extent_buffer_writeback(eb);
4022 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4024 clear_extent_buffer_dirty(eb);
4025 free_extent_buffer_stale(eb);
4032 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4033 struct extent_io_tree *pinned_extents)
4035 struct extent_io_tree *unpin;
4041 unpin = pinned_extents;
4044 ret = find_first_extent_bit(unpin, 0, &start, &end,
4045 EXTENT_DIRTY, NULL);
4050 if (btrfs_test_opt(root, DISCARD))
4051 ret = btrfs_error_discard_extent(root, start,
4055 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4056 btrfs_error_unpin_extent_range(root, start, end);
4061 if (unpin == &root->fs_info->freed_extents[0])
4062 unpin = &root->fs_info->freed_extents[1];
4064 unpin = &root->fs_info->freed_extents[0];
4072 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4073 struct btrfs_root *root)
4075 btrfs_destroy_ordered_operations(cur_trans, root);
4077 btrfs_destroy_delayed_refs(cur_trans, root);
4079 cur_trans->state = TRANS_STATE_COMMIT_START;
4080 wake_up(&root->fs_info->transaction_blocked_wait);
4082 cur_trans->state = TRANS_STATE_UNBLOCKED;
4083 wake_up(&root->fs_info->transaction_wait);
4085 btrfs_destroy_delayed_inodes(root);
4086 btrfs_assert_delayed_root_empty(root);
4088 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4090 btrfs_destroy_pinned_extent(root,
4091 root->fs_info->pinned_extents);
4093 cur_trans->state =TRANS_STATE_COMPLETED;
4094 wake_up(&cur_trans->commit_wait);
4097 memset(cur_trans, 0, sizeof(*cur_trans));
4098 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4102 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4104 struct btrfs_transaction *t;
4106 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4108 spin_lock(&root->fs_info->trans_lock);
4109 while (!list_empty(&root->fs_info->trans_list)) {
4110 t = list_first_entry(&root->fs_info->trans_list,
4111 struct btrfs_transaction, list);
4112 if (t->state >= TRANS_STATE_COMMIT_START) {
4113 atomic_inc(&t->use_count);
4114 spin_unlock(&root->fs_info->trans_lock);
4115 btrfs_wait_for_commit(root, t->transid);
4116 btrfs_put_transaction(t);
4117 spin_lock(&root->fs_info->trans_lock);
4120 if (t == root->fs_info->running_transaction) {
4121 t->state = TRANS_STATE_COMMIT_DOING;
4122 spin_unlock(&root->fs_info->trans_lock);
4124 * We wait for 0 num_writers since we don't hold a trans
4125 * handle open currently for this transaction.
4127 wait_event(t->writer_wait,
4128 atomic_read(&t->num_writers) == 0);
4130 spin_unlock(&root->fs_info->trans_lock);
4132 btrfs_cleanup_one_transaction(t, root);
4134 spin_lock(&root->fs_info->trans_lock);
4135 if (t == root->fs_info->running_transaction)
4136 root->fs_info->running_transaction = NULL;
4137 list_del_init(&t->list);
4138 spin_unlock(&root->fs_info->trans_lock);
4140 btrfs_put_transaction(t);
4141 trace_btrfs_transaction_commit(root);
4142 spin_lock(&root->fs_info->trans_lock);
4144 spin_unlock(&root->fs_info->trans_lock);
4145 btrfs_destroy_all_ordered_extents(root->fs_info);
4146 btrfs_destroy_delayed_inodes(root);
4147 btrfs_assert_delayed_root_empty(root);
4148 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4149 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4150 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4155 static struct extent_io_ops btree_extent_io_ops = {
4156 .readpage_end_io_hook = btree_readpage_end_io_hook,
4157 .readpage_io_failed_hook = btree_io_failed_hook,
4158 .submit_bio_hook = btree_submit_bio_hook,
4159 /* note we're sharing with inode.c for the merge bio hook */
4160 .merge_bio_hook = btrfs_merge_bio_hook,