1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19#include <linux/fs.h>
20#include <linux/slab.h>
21#include <linux/sched.h>
22#include <linux/writeback.h>
23#include <linux/pagemap.h>
24#include <linux/blkdev.h>
25#include <linux/uuid.h>
26#include "ctree.h"
27#include "disk-io.h"
28#include "transaction.h"
29#include "locking.h"
30#include "tree-log.h"
31#include "inode-map.h"
32#include "volumes.h"
33#include "dev-replace.h"
34
35#define BTRFS_ROOT_TRANS_TAG 0
36
37static void put_transaction(struct btrfs_transaction *transaction)
38{
39 WARN_ON(atomic_read(&transaction->use_count) == 0);
40 if (atomic_dec_and_test(&transaction->use_count)) {
41 BUG_ON(!list_empty(&transaction->list));
42 WARN_ON(transaction->delayed_refs.root.rb_node);
43 kmem_cache_free(btrfs_transaction_cachep, transaction);
44 }
45}
46
47static noinline void switch_commit_root(struct btrfs_root *root)
48{
49 free_extent_buffer(root->commit_root);
50 root->commit_root = btrfs_root_node(root);
51}
52
53static inline int can_join_transaction(struct btrfs_transaction *trans,
54 int type)
55{
56 return !(trans->in_commit &&
57 type != TRANS_JOIN &&
58 type != TRANS_JOIN_NOLOCK);
59}
60
61
62
63
64static noinline int join_transaction(struct btrfs_root *root, int type)
65{
66 struct btrfs_transaction *cur_trans;
67 struct btrfs_fs_info *fs_info = root->fs_info;
68
69 spin_lock(&fs_info->trans_lock);
70loop:
71
72 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
73 spin_unlock(&fs_info->trans_lock);
74 return -EROFS;
75 }
76
77 if (fs_info->trans_no_join) {
78
79
80
81
82
83
84 if (type != TRANS_JOIN_NOLOCK) {
85 spin_unlock(&fs_info->trans_lock);
86 return -EBUSY;
87 }
88 }
89
90 cur_trans = fs_info->running_transaction;
91 if (cur_trans) {
92 if (cur_trans->aborted) {
93 spin_unlock(&fs_info->trans_lock);
94 return cur_trans->aborted;
95 }
96 if (!can_join_transaction(cur_trans, type)) {
97 spin_unlock(&fs_info->trans_lock);
98 return -EBUSY;
99 }
100 atomic_inc(&cur_trans->use_count);
101 atomic_inc(&cur_trans->num_writers);
102 cur_trans->num_joined++;
103 spin_unlock(&fs_info->trans_lock);
104 return 0;
105 }
106 spin_unlock(&fs_info->trans_lock);
107
108
109
110
111
112 if (type == TRANS_ATTACH)
113 return -ENOENT;
114
115 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS);
116 if (!cur_trans)
117 return -ENOMEM;
118
119 spin_lock(&fs_info->trans_lock);
120 if (fs_info->running_transaction) {
121
122
123
124
125 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
126 goto loop;
127 } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
128 spin_unlock(&fs_info->trans_lock);
129 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
130 return -EROFS;
131 }
132
133 atomic_set(&cur_trans->num_writers, 1);
134 cur_trans->num_joined = 0;
135 init_waitqueue_head(&cur_trans->writer_wait);
136 init_waitqueue_head(&cur_trans->commit_wait);
137 cur_trans->in_commit = 0;
138 cur_trans->blocked = 0;
139
140
141
142
143 atomic_set(&cur_trans->use_count, 2);
144 cur_trans->commit_done = 0;
145 cur_trans->start_time = get_seconds();
146
147 cur_trans->delayed_refs.root = RB_ROOT;
148 cur_trans->delayed_refs.num_entries = 0;
149 cur_trans->delayed_refs.num_heads_ready = 0;
150 cur_trans->delayed_refs.num_heads = 0;
151 cur_trans->delayed_refs.flushing = 0;
152 cur_trans->delayed_refs.run_delayed_start = 0;
153
154
155
156
157
158 smp_mb();
159 if (!list_empty(&fs_info->tree_mod_seq_list))
160 WARN(1, KERN_ERR "btrfs: tree_mod_seq_list not empty when "
161 "creating a fresh transaction\n");
162 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log))
163 WARN(1, KERN_ERR "btrfs: tree_mod_log rb tree not empty when "
164 "creating a fresh transaction\n");
165 atomic64_set(&fs_info->tree_mod_seq, 0);
166
167 spin_lock_init(&cur_trans->commit_lock);
168 spin_lock_init(&cur_trans->delayed_refs.lock);
169 atomic_set(&cur_trans->delayed_refs.procs_running_refs, 0);
170 atomic_set(&cur_trans->delayed_refs.ref_seq, 0);
171 init_waitqueue_head(&cur_trans->delayed_refs.wait);
172
173 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
174 INIT_LIST_HEAD(&cur_trans->ordered_operations);
175 list_add_tail(&cur_trans->list, &fs_info->trans_list);
176 extent_io_tree_init(&cur_trans->dirty_pages,
177 fs_info->btree_inode->i_mapping);
178 fs_info->generation++;
179 cur_trans->transid = fs_info->generation;
180 fs_info->running_transaction = cur_trans;
181 cur_trans->aborted = 0;
182 spin_unlock(&fs_info->trans_lock);
183
184 return 0;
185}
186
187
188
189
190
191
192
193static int record_root_in_trans(struct btrfs_trans_handle *trans,
194 struct btrfs_root *root)
195{
196 if (root->ref_cows && root->last_trans < trans->transid) {
197 WARN_ON(root == root->fs_info->extent_root);
198 WARN_ON(root->commit_root != root->node);
199
200
201
202
203
204
205 root->in_trans_setup = 1;
206
207
208
209
210 smp_wmb();
211
212 spin_lock(&root->fs_info->fs_roots_radix_lock);
213 if (root->last_trans == trans->transid) {
214 spin_unlock(&root->fs_info->fs_roots_radix_lock);
215 return 0;
216 }
217 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
218 (unsigned long)root->root_key.objectid,
219 BTRFS_ROOT_TRANS_TAG);
220 spin_unlock(&root->fs_info->fs_roots_radix_lock);
221 root->last_trans = trans->transid;
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242 btrfs_init_reloc_root(trans, root);
243 smp_wmb();
244 root->in_trans_setup = 0;
245 }
246 return 0;
247}
248
249
250int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
251 struct btrfs_root *root)
252{
253 if (!root->ref_cows)
254 return 0;
255
256
257
258
259
260 smp_rmb();
261 if (root->last_trans == trans->transid &&
262 !root->in_trans_setup)
263 return 0;
264
265 mutex_lock(&root->fs_info->reloc_mutex);
266 record_root_in_trans(trans, root);
267 mutex_unlock(&root->fs_info->reloc_mutex);
268
269 return 0;
270}
271
272
273
274
275
276static void wait_current_trans(struct btrfs_root *root)
277{
278 struct btrfs_transaction *cur_trans;
279
280 spin_lock(&root->fs_info->trans_lock);
281 cur_trans = root->fs_info->running_transaction;
282 if (cur_trans && cur_trans->blocked) {
283 atomic_inc(&cur_trans->use_count);
284 spin_unlock(&root->fs_info->trans_lock);
285
286 wait_event(root->fs_info->transaction_wait,
287 !cur_trans->blocked);
288 put_transaction(cur_trans);
289 } else {
290 spin_unlock(&root->fs_info->trans_lock);
291 }
292}
293
294static int may_wait_transaction(struct btrfs_root *root, int type)
295{
296 if (root->fs_info->log_root_recovering)
297 return 0;
298
299 if (type == TRANS_USERSPACE)
300 return 1;
301
302 if (type == TRANS_START &&
303 !atomic_read(&root->fs_info->open_ioctl_trans))
304 return 1;
305
306 return 0;
307}
308
309static struct btrfs_trans_handle *
310start_transaction(struct btrfs_root *root, u64 num_items, int type,
311 enum btrfs_reserve_flush_enum flush)
312{
313 struct btrfs_trans_handle *h;
314 struct btrfs_transaction *cur_trans;
315 u64 num_bytes = 0;
316 int ret;
317 u64 qgroup_reserved = 0;
318
319 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
320 return ERR_PTR(-EROFS);
321
322 if (current->journal_info) {
323 WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK);
324 h = current->journal_info;
325 h->use_count++;
326 WARN_ON(h->use_count > 2);
327 h->orig_rsv = h->block_rsv;
328 h->block_rsv = NULL;
329 goto got_it;
330 }
331
332
333
334
335
336 if (num_items > 0 && root != root->fs_info->chunk_root) {
337 if (root->fs_info->quota_enabled &&
338 is_fstree(root->root_key.objectid)) {
339 qgroup_reserved = num_items * root->leafsize;
340 ret = btrfs_qgroup_reserve(root, qgroup_reserved);
341 if (ret)
342 return ERR_PTR(ret);
343 }
344
345 num_bytes = btrfs_calc_trans_metadata_size(root, num_items);
346 ret = btrfs_block_rsv_add(root,
347 &root->fs_info->trans_block_rsv,
348 num_bytes, flush);
349 if (ret)
350 goto reserve_fail;
351 }
352again:
353 h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
354 if (!h) {
355 ret = -ENOMEM;
356 goto alloc_fail;
357 }
358
359
360
361
362
363
364
365
366
367
368
369 if (type < TRANS_JOIN_NOLOCK)
370 sb_start_intwrite(root->fs_info->sb);
371
372 if (may_wait_transaction(root, type))
373 wait_current_trans(root);
374
375 do {
376 ret = join_transaction(root, type);
377 if (ret == -EBUSY) {
378 wait_current_trans(root);
379 if (unlikely(type == TRANS_ATTACH))
380 ret = -ENOENT;
381 }
382 } while (ret == -EBUSY);
383
384 if (ret < 0) {
385
386 BUG_ON(type == TRANS_JOIN_NOLOCK);
387 goto join_fail;
388 }
389
390 cur_trans = root->fs_info->running_transaction;
391
392 h->transid = cur_trans->transid;
393 h->transaction = cur_trans;
394 h->blocks_used = 0;
395 h->bytes_reserved = 0;
396 h->root = root;
397 h->delayed_ref_updates = 0;
398 h->use_count = 1;
399 h->adding_csums = 0;
400 h->block_rsv = NULL;
401 h->orig_rsv = NULL;
402 h->aborted = 0;
403 h->qgroup_reserved = 0;
404 h->delayed_ref_elem.seq = 0;
405 h->type = type;
406 h->allocating_chunk = false;
407 INIT_LIST_HEAD(&h->qgroup_ref_list);
408 INIT_LIST_HEAD(&h->new_bgs);
409
410 smp_mb();
411 if (cur_trans->blocked && may_wait_transaction(root, type)) {
412 btrfs_commit_transaction(h, root);
413 goto again;
414 }
415
416 if (num_bytes) {
417 trace_btrfs_space_reservation(root->fs_info, "transaction",
418 h->transid, num_bytes, 1);
419 h->block_rsv = &root->fs_info->trans_block_rsv;
420 h->bytes_reserved = num_bytes;
421 }
422 h->qgroup_reserved = qgroup_reserved;
423
424got_it:
425 btrfs_record_root_in_trans(h, root);
426
427 if (!current->journal_info && type != TRANS_USERSPACE)
428 current->journal_info = h;
429 return h;
430
431join_fail:
432 if (type < TRANS_JOIN_NOLOCK)
433 sb_end_intwrite(root->fs_info->sb);
434 kmem_cache_free(btrfs_trans_handle_cachep, h);
435alloc_fail:
436 if (num_bytes)
437 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
438 num_bytes);
439reserve_fail:
440 if (qgroup_reserved)
441 btrfs_qgroup_free(root, qgroup_reserved);
442 return ERR_PTR(ret);
443}
444
445struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
446 int num_items)
447{
448 return start_transaction(root, num_items, TRANS_START,
449 BTRFS_RESERVE_FLUSH_ALL);
450}
451
452struct btrfs_trans_handle *btrfs_start_transaction_lflush(
453 struct btrfs_root *root, int num_items)
454{
455 return start_transaction(root, num_items, TRANS_START,
456 BTRFS_RESERVE_FLUSH_LIMIT);
457}
458
459struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root)
460{
461 return start_transaction(root, 0, TRANS_JOIN, 0);
462}
463
464struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root)
465{
466 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0);
467}
468
469struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root)
470{
471 return start_transaction(root, 0, TRANS_USERSPACE, 0);
472}
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root)
488{
489 return start_transaction(root, 0, TRANS_ATTACH, 0);
490}
491
492
493
494
495
496
497
498
499struct btrfs_trans_handle *
500btrfs_attach_transaction_barrier(struct btrfs_root *root)
501{
502 struct btrfs_trans_handle *trans;
503
504 trans = start_transaction(root, 0, TRANS_ATTACH, 0);
505 if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT)
506 btrfs_wait_for_commit(root, 0);
507
508 return trans;
509}
510
511
512static noinline void wait_for_commit(struct btrfs_root *root,
513 struct btrfs_transaction *commit)
514{
515 wait_event(commit->commit_wait, commit->commit_done);
516}
517
518int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid)
519{
520 struct btrfs_transaction *cur_trans = NULL, *t;
521 int ret = 0;
522
523 if (transid) {
524 if (transid <= root->fs_info->last_trans_committed)
525 goto out;
526
527 ret = -EINVAL;
528
529 spin_lock(&root->fs_info->trans_lock);
530 list_for_each_entry(t, &root->fs_info->trans_list, list) {
531 if (t->transid == transid) {
532 cur_trans = t;
533 atomic_inc(&cur_trans->use_count);
534 ret = 0;
535 break;
536 }
537 if (t->transid > transid) {
538 ret = 0;
539 break;
540 }
541 }
542 spin_unlock(&root->fs_info->trans_lock);
543
544 if (!cur_trans)
545 goto out;
546 } else {
547
548 spin_lock(&root->fs_info->trans_lock);
549 list_for_each_entry_reverse(t, &root->fs_info->trans_list,
550 list) {
551 if (t->in_commit) {
552 if (t->commit_done)
553 break;
554 cur_trans = t;
555 atomic_inc(&cur_trans->use_count);
556 break;
557 }
558 }
559 spin_unlock(&root->fs_info->trans_lock);
560 if (!cur_trans)
561 goto out;
562 }
563
564 wait_for_commit(root, cur_trans);
565 put_transaction(cur_trans);
566out:
567 return ret;
568}
569
570void btrfs_throttle(struct btrfs_root *root)
571{
572 if (!atomic_read(&root->fs_info->open_ioctl_trans))
573 wait_current_trans(root);
574}
575
576static int should_end_transaction(struct btrfs_trans_handle *trans,
577 struct btrfs_root *root)
578{
579 int ret;
580
581 ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5);
582 return ret ? 1 : 0;
583}
584
585int btrfs_should_end_transaction(struct btrfs_trans_handle *trans,
586 struct btrfs_root *root)
587{
588 struct btrfs_transaction *cur_trans = trans->transaction;
589 int updates;
590 int err;
591
592 smp_mb();
593 if (cur_trans->blocked || cur_trans->delayed_refs.flushing)
594 return 1;
595
596 updates = trans->delayed_ref_updates;
597 trans->delayed_ref_updates = 0;
598 if (updates) {
599 err = btrfs_run_delayed_refs(trans, root, updates);
600 if (err)
601 return err;
602 }
603
604 return should_end_transaction(trans, root);
605}
606
607static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
608 struct btrfs_root *root, int throttle)
609{
610 struct btrfs_transaction *cur_trans = trans->transaction;
611 struct btrfs_fs_info *info = root->fs_info;
612 int count = 0;
613 int lock = (trans->type != TRANS_JOIN_NOLOCK);
614 int err = 0;
615
616 if (--trans->use_count) {
617 trans->block_rsv = trans->orig_rsv;
618 return 0;
619 }
620
621
622
623
624 err = btrfs_delayed_refs_qgroup_accounting(trans, info);
625
626 btrfs_trans_release_metadata(trans, root);
627 trans->block_rsv = NULL;
628
629 if (trans->qgroup_reserved) {
630
631
632
633
634 btrfs_qgroup_free(trans->root, trans->qgroup_reserved);
635 trans->qgroup_reserved = 0;
636 }
637
638 if (!list_empty(&trans->new_bgs))
639 btrfs_create_pending_block_groups(trans, root);
640
641 while (count < 1) {
642 unsigned long cur = trans->delayed_ref_updates;
643 trans->delayed_ref_updates = 0;
644 if (cur &&
645 trans->transaction->delayed_refs.num_heads_ready > 64) {
646 trans->delayed_ref_updates = 0;
647 btrfs_run_delayed_refs(trans, root, cur);
648 } else {
649 break;
650 }
651 count++;
652 }
653
654 btrfs_trans_release_metadata(trans, root);
655 trans->block_rsv = NULL;
656
657 if (!list_empty(&trans->new_bgs))
658 btrfs_create_pending_block_groups(trans, root);
659
660 if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) &&
661 should_end_transaction(trans, root)) {
662 trans->transaction->blocked = 1;
663 smp_wmb();
664 }
665
666 if (lock && cur_trans->blocked && !cur_trans->in_commit) {
667 if (throttle) {
668
669
670
671
672
673 trans->use_count++;
674 return btrfs_commit_transaction(trans, root);
675 } else {
676 wake_up_process(info->transaction_kthread);
677 }
678 }
679
680 if (trans->type < TRANS_JOIN_NOLOCK)
681 sb_end_intwrite(root->fs_info->sb);
682
683 WARN_ON(cur_trans != info->running_transaction);
684 WARN_ON(atomic_read(&cur_trans->num_writers) < 1);
685 atomic_dec(&cur_trans->num_writers);
686
687 smp_mb();
688 if (waitqueue_active(&cur_trans->writer_wait))
689 wake_up(&cur_trans->writer_wait);
690 put_transaction(cur_trans);
691
692 if (current->journal_info == trans)
693 current->journal_info = NULL;
694
695 if (throttle)
696 btrfs_run_delayed_iputs(root);
697
698 if (trans->aborted ||
699 test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
700 err = -EIO;
701 assert_qgroups_uptodate(trans);
702
703 kmem_cache_free(btrfs_trans_handle_cachep, trans);
704 return err;
705}
706
707int btrfs_end_transaction(struct btrfs_trans_handle *trans,
708 struct btrfs_root *root)
709{
710 return __btrfs_end_transaction(trans, root, 0);
711}
712
713int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
714 struct btrfs_root *root)
715{
716 return __btrfs_end_transaction(trans, root, 1);
717}
718
719int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans,
720 struct btrfs_root *root)
721{
722 return __btrfs_end_transaction(trans, root, 1);
723}
724
725
726
727
728
729
730int btrfs_write_marked_extents(struct btrfs_root *root,
731 struct extent_io_tree *dirty_pages, int mark)
732{
733 int err = 0;
734 int werr = 0;
735 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
736 struct extent_state *cached_state = NULL;
737 u64 start = 0;
738 u64 end;
739 struct blk_plug plug;
740
741 blk_start_plug(&plug);
742 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
743 mark, &cached_state)) {
744 convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
745 mark, &cached_state, GFP_NOFS);
746 cached_state = NULL;
747 err = filemap_fdatawrite_range(mapping, start, end);
748 if (err)
749 werr = err;
750 cond_resched();
751 start = end + 1;
752 }
753 if (err)
754 werr = err;
755 blk_finish_plug(&plug);
756 return werr;
757}
758
759
760
761
762
763
764
765int btrfs_wait_marked_extents(struct btrfs_root *root,
766 struct extent_io_tree *dirty_pages, int mark)
767{
768 int err = 0;
769 int werr = 0;
770 struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
771 struct extent_state *cached_state = NULL;
772 u64 start = 0;
773 u64 end;
774
775 while (!find_first_extent_bit(dirty_pages, start, &start, &end,
776 EXTENT_NEED_WAIT, &cached_state)) {
777 clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT,
778 0, 0, &cached_state, GFP_NOFS);
779 err = filemap_fdatawait_range(mapping, start, end);
780 if (err)
781 werr = err;
782 cond_resched();
783 start = end + 1;
784 }
785 if (err)
786 werr = err;
787 return werr;
788}
789
790
791
792
793
794
795int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
796 struct extent_io_tree *dirty_pages, int mark)
797{
798 int ret;
799 int ret2;
800
801 ret = btrfs_write_marked_extents(root, dirty_pages, mark);
802 ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
803
804 if (ret)
805 return ret;
806 if (ret2)
807 return ret2;
808 return 0;
809}
810
811int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
812 struct btrfs_root *root)
813{
814 if (!trans || !trans->transaction) {
815 struct inode *btree_inode;
816 btree_inode = root->fs_info->btree_inode;
817 return filemap_write_and_wait(btree_inode->i_mapping);
818 }
819 return btrfs_write_and_wait_marked_extents(root,
820 &trans->transaction->dirty_pages,
821 EXTENT_DIRTY);
822}
823
824
825
826
827
828
829
830
831
832
833
834static int update_cowonly_root(struct btrfs_trans_handle *trans,
835 struct btrfs_root *root)
836{
837 int ret;
838 u64 old_root_bytenr;
839 u64 old_root_used;
840 struct btrfs_root *tree_root = root->fs_info->tree_root;
841
842 old_root_used = btrfs_root_used(&root->root_item);
843 btrfs_write_dirty_block_groups(trans, root);
844
845 while (1) {
846 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
847 if (old_root_bytenr == root->node->start &&
848 old_root_used == btrfs_root_used(&root->root_item))
849 break;
850
851 btrfs_set_root_node(&root->root_item, root->node);
852 ret = btrfs_update_root(trans, tree_root,
853 &root->root_key,
854 &root->root_item);
855 if (ret)
856 return ret;
857
858 old_root_used = btrfs_root_used(&root->root_item);
859 ret = btrfs_write_dirty_block_groups(trans, root);
860 if (ret)
861 return ret;
862 }
863
864 if (root != root->fs_info->extent_root)
865 switch_commit_root(root);
866
867 return 0;
868}
869
870
871
872
873
874
875
876
877static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
878 struct btrfs_root *root)
879{
880 struct btrfs_fs_info *fs_info = root->fs_info;
881 struct list_head *next;
882 struct extent_buffer *eb;
883 int ret;
884
885 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
886 if (ret)
887 return ret;
888
889 eb = btrfs_lock_root_node(fs_info->tree_root);
890 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL,
891 0, &eb);
892 btrfs_tree_unlock(eb);
893 free_extent_buffer(eb);
894
895 if (ret)
896 return ret;
897
898 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
899 if (ret)
900 return ret;
901
902 ret = btrfs_run_dev_stats(trans, root->fs_info);
903 WARN_ON(ret);
904 ret = btrfs_run_dev_replace(trans, root->fs_info);
905 WARN_ON(ret);
906
907 ret = btrfs_run_qgroups(trans, root->fs_info);
908 BUG_ON(ret);
909
910
911 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
912 BUG_ON(ret);
913
914 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
915 next = fs_info->dirty_cowonly_roots.next;
916 list_del_init(next);
917 root = list_entry(next, struct btrfs_root, dirty_list);
918
919 ret = update_cowonly_root(trans, root);
920 if (ret)
921 return ret;
922 }
923
924 down_write(&fs_info->extent_commit_sem);
925 switch_commit_root(fs_info->extent_root);
926 up_write(&fs_info->extent_commit_sem);
927
928 btrfs_after_dev_replace_commit(fs_info);
929
930 return 0;
931}
932
933
934
935
936
937
938int btrfs_add_dead_root(struct btrfs_root *root)
939{
940 spin_lock(&root->fs_info->trans_lock);
941 list_add_tail(&root->root_list, &root->fs_info->dead_roots);
942 spin_unlock(&root->fs_info->trans_lock);
943 return 0;
944}
945
946
947
948
949static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
950 struct btrfs_root *root)
951{
952 struct btrfs_root *gang[8];
953 struct btrfs_fs_info *fs_info = root->fs_info;
954 int i;
955 int ret;
956 int err = 0;
957
958 spin_lock(&fs_info->fs_roots_radix_lock);
959 while (1) {
960 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix,
961 (void **)gang, 0,
962 ARRAY_SIZE(gang),
963 BTRFS_ROOT_TRANS_TAG);
964 if (ret == 0)
965 break;
966 for (i = 0; i < ret; i++) {
967 root = gang[i];
968 radix_tree_tag_clear(&fs_info->fs_roots_radix,
969 (unsigned long)root->root_key.objectid,
970 BTRFS_ROOT_TRANS_TAG);
971 spin_unlock(&fs_info->fs_roots_radix_lock);
972
973 btrfs_free_log(trans, root);
974 btrfs_update_reloc_root(trans, root);
975 btrfs_orphan_commit_root(trans, root);
976
977 btrfs_save_ino_cache(root, trans);
978
979
980 root->force_cow = 0;
981 smp_wmb();
982
983 if (root->commit_root != root->node) {
984 mutex_lock(&root->fs_commit_mutex);
985 switch_commit_root(root);
986 btrfs_unpin_free_ino(root);
987 mutex_unlock(&root->fs_commit_mutex);
988
989 btrfs_set_root_node(&root->root_item,
990 root->node);
991 }
992
993 err = btrfs_update_root(trans, fs_info->tree_root,
994 &root->root_key,
995 &root->root_item);
996 spin_lock(&fs_info->fs_roots_radix_lock);
997 if (err)
998 break;
999 }
1000 }
1001 spin_unlock(&fs_info->fs_roots_radix_lock);
1002 return err;
1003}
1004
1005
1006
1007
1008
1009int btrfs_defrag_root(struct btrfs_root *root)
1010{
1011 struct btrfs_fs_info *info = root->fs_info;
1012 struct btrfs_trans_handle *trans;
1013 int ret;
1014
1015 if (xchg(&root->defrag_running, 1))
1016 return 0;
1017
1018 while (1) {
1019 trans = btrfs_start_transaction(root, 0);
1020 if (IS_ERR(trans))
1021 return PTR_ERR(trans);
1022
1023 ret = btrfs_defrag_leaves(trans, root);
1024
1025 btrfs_end_transaction(trans, root);
1026 btrfs_btree_balance_dirty(info->tree_root);
1027 cond_resched();
1028
1029 if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN)
1030 break;
1031
1032 if (btrfs_defrag_cancelled(root->fs_info)) {
1033 printk(KERN_DEBUG "btrfs: defrag_root cancelled\n");
1034 ret = -EAGAIN;
1035 break;
1036 }
1037 }
1038 root->defrag_running = 0;
1039 return ret;
1040}
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
1052 struct btrfs_fs_info *fs_info,
1053 struct btrfs_pending_snapshot *pending)
1054{
1055 struct btrfs_key key;
1056 struct btrfs_root_item *new_root_item;
1057 struct btrfs_root *tree_root = fs_info->tree_root;
1058 struct btrfs_root *root = pending->root;
1059 struct btrfs_root *parent_root;
1060 struct btrfs_block_rsv *rsv;
1061 struct inode *parent_inode;
1062 struct btrfs_path *path;
1063 struct btrfs_dir_item *dir_item;
1064 struct dentry *dentry;
1065 struct extent_buffer *tmp;
1066 struct extent_buffer *old;
1067 struct timespec cur_time = CURRENT_TIME;
1068 int ret = 0;
1069 u64 to_reserve = 0;
1070 u64 index = 0;
1071 u64 objectid;
1072 u64 root_flags;
1073 uuid_le new_uuid;
1074
1075 path = btrfs_alloc_path();
1076 if (!path) {
1077 pending->error = -ENOMEM;
1078 return 0;
1079 }
1080
1081 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
1082 if (!new_root_item) {
1083 pending->error = -ENOMEM;
1084 goto root_item_alloc_fail;
1085 }
1086
1087 pending->error = btrfs_find_free_objectid(tree_root, &objectid);
1088 if (pending->error)
1089 goto no_free_objectid;
1090
1091 btrfs_reloc_pre_snapshot(trans, pending, &to_reserve);
1092
1093 if (to_reserve > 0) {
1094 pending->error = btrfs_block_rsv_add(root,
1095 &pending->block_rsv,
1096 to_reserve,
1097 BTRFS_RESERVE_NO_FLUSH);
1098 if (pending->error)
1099 goto no_free_objectid;
1100 }
1101
1102 pending->error = btrfs_qgroup_inherit(trans, fs_info,
1103 root->root_key.objectid,
1104 objectid, pending->inherit);
1105 if (pending->error)
1106 goto no_free_objectid;
1107
1108 key.objectid = objectid;
1109 key.offset = (u64)-1;
1110 key.type = BTRFS_ROOT_ITEM_KEY;
1111
1112 rsv = trans->block_rsv;
1113 trans->block_rsv = &pending->block_rsv;
1114 trans->bytes_reserved = trans->block_rsv->reserved;
1115
1116 dentry = pending->dentry;
1117 parent_inode = pending->dir;
1118 parent_root = BTRFS_I(parent_inode)->root;
1119 record_root_in_trans(trans, parent_root);
1120
1121
1122
1123
1124 ret = btrfs_set_inode_index(parent_inode, &index);
1125 BUG_ON(ret);
1126
1127
1128 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path,
1129 btrfs_ino(parent_inode),
1130 dentry->d_name.name,
1131 dentry->d_name.len, 0);
1132 if (dir_item != NULL && !IS_ERR(dir_item)) {
1133 pending->error = -EEXIST;
1134 goto dir_item_existed;
1135 } else if (IS_ERR(dir_item)) {
1136 ret = PTR_ERR(dir_item);
1137 btrfs_abort_transaction(trans, root, ret);
1138 goto fail;
1139 }
1140 btrfs_release_path(path);
1141
1142
1143
1144
1145
1146
1147
1148 ret = btrfs_run_delayed_items(trans, root);
1149 if (ret) {
1150 btrfs_abort_transaction(trans, root, ret);
1151 goto fail;
1152 }
1153
1154 record_root_in_trans(trans, root);
1155 btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
1156 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
1157 btrfs_check_and_init_root_item(new_root_item);
1158
1159 root_flags = btrfs_root_flags(new_root_item);
1160 if (pending->readonly)
1161 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY;
1162 else
1163 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY;
1164 btrfs_set_root_flags(new_root_item, root_flags);
1165
1166 btrfs_set_root_generation_v2(new_root_item,
1167 trans->transid);
1168 uuid_le_gen(&new_uuid);
1169 memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE);
1170 memcpy(new_root_item->parent_uuid, root->root_item.uuid,
1171 BTRFS_UUID_SIZE);
1172 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) {
1173 memset(new_root_item->received_uuid, 0,
1174 sizeof(new_root_item->received_uuid));
1175 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime));
1176 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime));
1177 btrfs_set_root_stransid(new_root_item, 0);
1178 btrfs_set_root_rtransid(new_root_item, 0);
1179 }
1180 new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec);
1181 new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec);
1182 btrfs_set_root_otransid(new_root_item, trans->transid);
1183
1184 old = btrfs_lock_root_node(root);
1185 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old);
1186 if (ret) {
1187 btrfs_tree_unlock(old);
1188 free_extent_buffer(old);
1189 btrfs_abort_transaction(trans, root, ret);
1190 goto fail;
1191 }
1192
1193 btrfs_set_lock_blocking(old);
1194
1195 ret = btrfs_copy_root(trans, root, old, &tmp, objectid);
1196
1197 btrfs_tree_unlock(old);
1198 free_extent_buffer(old);
1199 if (ret) {
1200 btrfs_abort_transaction(trans, root, ret);
1201 goto fail;
1202 }
1203
1204
1205 root->force_cow = 1;
1206 smp_wmb();
1207
1208 btrfs_set_root_node(new_root_item, tmp);
1209
1210 key.offset = trans->transid;
1211 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item);
1212 btrfs_tree_unlock(tmp);
1213 free_extent_buffer(tmp);
1214 if (ret) {
1215 btrfs_abort_transaction(trans, root, ret);
1216 goto fail;
1217 }
1218
1219
1220
1221
1222 ret = btrfs_add_root_ref(trans, tree_root, objectid,
1223 parent_root->root_key.objectid,
1224 btrfs_ino(parent_inode), index,
1225 dentry->d_name.name, dentry->d_name.len);
1226 if (ret) {
1227 btrfs_abort_transaction(trans, root, ret);
1228 goto fail;
1229 }
1230
1231 key.offset = (u64)-1;
1232 pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key);
1233 if (IS_ERR(pending->snap)) {
1234 ret = PTR_ERR(pending->snap);
1235 btrfs_abort_transaction(trans, root, ret);
1236 goto fail;
1237 }
1238
1239 ret = btrfs_reloc_post_snapshot(trans, pending);
1240 if (ret) {
1241 btrfs_abort_transaction(trans, root, ret);
1242 goto fail;
1243 }
1244
1245 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1246 if (ret) {
1247 btrfs_abort_transaction(trans, root, ret);
1248 goto fail;
1249 }
1250
1251 ret = btrfs_insert_dir_item(trans, parent_root,
1252 dentry->d_name.name, dentry->d_name.len,
1253 parent_inode, &key,
1254 BTRFS_FT_DIR, index);
1255
1256 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW);
1257 if (ret) {
1258 btrfs_abort_transaction(trans, root, ret);
1259 goto fail;
1260 }
1261
1262 btrfs_i_size_write(parent_inode, parent_inode->i_size +
1263 dentry->d_name.len * 2);
1264 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1265 ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode);
1266 if (ret)
1267 btrfs_abort_transaction(trans, root, ret);
1268fail:
1269 pending->error = ret;
1270dir_item_existed:
1271 trans->block_rsv = rsv;
1272 trans->bytes_reserved = 0;
1273no_free_objectid:
1274 kfree(new_root_item);
1275root_item_alloc_fail:
1276 btrfs_free_path(path);
1277 return ret;
1278}
1279
1280
1281
1282
1283static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
1284 struct btrfs_fs_info *fs_info)
1285{
1286 struct btrfs_pending_snapshot *pending, *next;
1287 struct list_head *head = &trans->transaction->pending_snapshots;
1288 int ret = 0;
1289
1290 list_for_each_entry_safe(pending, next, head, list) {
1291 list_del(&pending->list);
1292 ret = create_pending_snapshot(trans, fs_info, pending);
1293 if (ret)
1294 break;
1295 }
1296 return ret;
1297}
1298
1299static void update_super_roots(struct btrfs_root *root)
1300{
1301 struct btrfs_root_item *root_item;
1302 struct btrfs_super_block *super;
1303
1304 super = root->fs_info->super_copy;
1305
1306 root_item = &root->fs_info->chunk_root->root_item;
1307 super->chunk_root = root_item->bytenr;
1308 super->chunk_root_generation = root_item->generation;
1309 super->chunk_root_level = root_item->level;
1310
1311 root_item = &root->fs_info->tree_root->root_item;
1312 super->root = root_item->bytenr;
1313 super->generation = root_item->generation;
1314 super->root_level = root_item->level;
1315 if (btrfs_test_opt(root, SPACE_CACHE))
1316 super->cache_generation = root_item->generation;
1317}
1318
1319int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
1320{
1321 int ret = 0;
1322 spin_lock(&info->trans_lock);
1323 if (info->running_transaction)
1324 ret = info->running_transaction->in_commit;
1325 spin_unlock(&info->trans_lock);
1326 return ret;
1327}
1328
1329int btrfs_transaction_blocked(struct btrfs_fs_info *info)
1330{
1331 int ret = 0;
1332 spin_lock(&info->trans_lock);
1333 if (info->running_transaction)
1334 ret = info->running_transaction->blocked;
1335 spin_unlock(&info->trans_lock);
1336 return ret;
1337}
1338
1339
1340
1341
1342
1343static void wait_current_trans_commit_start(struct btrfs_root *root,
1344 struct btrfs_transaction *trans)
1345{
1346 wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit);
1347}
1348
1349
1350
1351
1352
1353static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root,
1354 struct btrfs_transaction *trans)
1355{
1356 wait_event(root->fs_info->transaction_wait,
1357 trans->commit_done || (trans->in_commit && !trans->blocked));
1358}
1359
1360
1361
1362
1363
1364struct btrfs_async_commit {
1365 struct btrfs_trans_handle *newtrans;
1366 struct btrfs_root *root;
1367 struct work_struct work;
1368};
1369
1370static void do_async_commit(struct work_struct *work)
1371{
1372 struct btrfs_async_commit *ac =
1373 container_of(work, struct btrfs_async_commit, work);
1374
1375
1376
1377
1378
1379 if (ac->newtrans->type < TRANS_JOIN_NOLOCK)
1380 rwsem_acquire_read(
1381 &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1382 0, 1, _THIS_IP_);
1383
1384 current->journal_info = ac->newtrans;
1385
1386 btrfs_commit_transaction(ac->newtrans, ac->root);
1387 kfree(ac);
1388}
1389
1390int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
1391 struct btrfs_root *root,
1392 int wait_for_unblock)
1393{
1394 struct btrfs_async_commit *ac;
1395 struct btrfs_transaction *cur_trans;
1396
1397 ac = kmalloc(sizeof(*ac), GFP_NOFS);
1398 if (!ac)
1399 return -ENOMEM;
1400
1401 INIT_WORK(&ac->work, do_async_commit);
1402 ac->root = root;
1403 ac->newtrans = btrfs_join_transaction(root);
1404 if (IS_ERR(ac->newtrans)) {
1405 int err = PTR_ERR(ac->newtrans);
1406 kfree(ac);
1407 return err;
1408 }
1409
1410
1411 cur_trans = trans->transaction;
1412 atomic_inc(&cur_trans->use_count);
1413
1414 btrfs_end_transaction(trans, root);
1415
1416
1417
1418
1419
1420 if (trans->type < TRANS_JOIN_NOLOCK)
1421 rwsem_release(
1422 &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1],
1423 1, _THIS_IP_);
1424
1425 schedule_work(&ac->work);
1426
1427
1428 if (wait_for_unblock)
1429 wait_current_trans_commit_start_and_unblock(root, cur_trans);
1430 else
1431 wait_current_trans_commit_start(root, cur_trans);
1432
1433 if (current->journal_info == trans)
1434 current->journal_info = NULL;
1435
1436 put_transaction(cur_trans);
1437 return 0;
1438}
1439
1440
1441static void cleanup_transaction(struct btrfs_trans_handle *trans,
1442 struct btrfs_root *root, int err)
1443{
1444 struct btrfs_transaction *cur_trans = trans->transaction;
1445 DEFINE_WAIT(wait);
1446
1447 WARN_ON(trans->use_count > 1);
1448
1449 btrfs_abort_transaction(trans, root, err);
1450
1451 spin_lock(&root->fs_info->trans_lock);
1452
1453 if (list_empty(&cur_trans->list)) {
1454 spin_unlock(&root->fs_info->trans_lock);
1455 btrfs_end_transaction(trans, root);
1456 return;
1457 }
1458
1459 list_del_init(&cur_trans->list);
1460 if (cur_trans == root->fs_info->running_transaction) {
1461 root->fs_info->trans_no_join = 1;
1462 spin_unlock(&root->fs_info->trans_lock);
1463 wait_event(cur_trans->writer_wait,
1464 atomic_read(&cur_trans->num_writers) == 1);
1465
1466 spin_lock(&root->fs_info->trans_lock);
1467 root->fs_info->running_transaction = NULL;
1468 }
1469 spin_unlock(&root->fs_info->trans_lock);
1470
1471 btrfs_cleanup_one_transaction(trans->transaction, root);
1472
1473 put_transaction(cur_trans);
1474 put_transaction(cur_trans);
1475
1476 trace_btrfs_transaction_commit(root);
1477
1478 btrfs_scrub_continue(root);
1479
1480 if (current->journal_info == trans)
1481 current->journal_info = NULL;
1482
1483 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1484
1485 spin_lock(&root->fs_info->trans_lock);
1486 root->fs_info->trans_no_join = 0;
1487 spin_unlock(&root->fs_info->trans_lock);
1488}
1489
1490static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans,
1491 struct btrfs_root *root)
1492{
1493 int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT);
1494 int snap_pending = 0;
1495 int ret;
1496
1497 if (!flush_on_commit) {
1498 spin_lock(&root->fs_info->trans_lock);
1499 if (!list_empty(&trans->transaction->pending_snapshots))
1500 snap_pending = 1;
1501 spin_unlock(&root->fs_info->trans_lock);
1502 }
1503
1504 if (flush_on_commit || snap_pending) {
1505 ret = btrfs_start_delalloc_inodes(root, 1);
1506 if (ret)
1507 return ret;
1508 btrfs_wait_ordered_extents(root, 1);
1509 }
1510
1511 ret = btrfs_run_delayed_items(trans, root);
1512 if (ret)
1513 return ret;
1514
1515
1516
1517
1518
1519
1520 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
1521
1522
1523
1524
1525
1526
1527
1528
1529 ret = btrfs_run_ordered_operations(trans, root, 1);
1530
1531 return ret;
1532}
1533
1534
1535
1536
1537
1538
1539
1540
1541int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1542 struct btrfs_root *root)
1543{
1544 unsigned long joined = 0;
1545 struct btrfs_transaction *cur_trans = trans->transaction;
1546 struct btrfs_transaction *prev_trans = NULL;
1547 DEFINE_WAIT(wait);
1548 int ret;
1549 int should_grow = 0;
1550 unsigned long now = get_seconds();
1551
1552 ret = btrfs_run_ordered_operations(trans, root, 0);
1553 if (ret) {
1554 btrfs_abort_transaction(trans, root, ret);
1555 btrfs_end_transaction(trans, root);
1556 return ret;
1557 }
1558
1559
1560 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1561 ret = cur_trans->aborted;
1562 btrfs_end_transaction(trans, root);
1563 return ret;
1564 }
1565
1566
1567
1568
1569 ret = btrfs_run_delayed_refs(trans, root, 0);
1570 if (ret) {
1571 btrfs_end_transaction(trans, root);
1572 return ret;
1573 }
1574
1575 btrfs_trans_release_metadata(trans, root);
1576 trans->block_rsv = NULL;
1577 if (trans->qgroup_reserved) {
1578 btrfs_qgroup_free(root, trans->qgroup_reserved);
1579 trans->qgroup_reserved = 0;
1580 }
1581
1582 cur_trans = trans->transaction;
1583
1584
1585
1586
1587
1588 cur_trans->delayed_refs.flushing = 1;
1589
1590 if (!list_empty(&trans->new_bgs))
1591 btrfs_create_pending_block_groups(trans, root);
1592
1593 ret = btrfs_run_delayed_refs(trans, root, 0);
1594 if (ret) {
1595 btrfs_end_transaction(trans, root);
1596 return ret;
1597 }
1598
1599 spin_lock(&cur_trans->commit_lock);
1600 if (cur_trans->in_commit) {
1601 spin_unlock(&cur_trans->commit_lock);
1602 atomic_inc(&cur_trans->use_count);
1603 ret = btrfs_end_transaction(trans, root);
1604
1605 wait_for_commit(root, cur_trans);
1606
1607 put_transaction(cur_trans);
1608
1609 return ret;
1610 }
1611
1612 trans->transaction->in_commit = 1;
1613 trans->transaction->blocked = 1;
1614 spin_unlock(&cur_trans->commit_lock);
1615 wake_up(&root->fs_info->transaction_blocked_wait);
1616
1617 spin_lock(&root->fs_info->trans_lock);
1618 if (cur_trans->list.prev != &root->fs_info->trans_list) {
1619 prev_trans = list_entry(cur_trans->list.prev,
1620 struct btrfs_transaction, list);
1621 if (!prev_trans->commit_done) {
1622 atomic_inc(&prev_trans->use_count);
1623 spin_unlock(&root->fs_info->trans_lock);
1624
1625 wait_for_commit(root, prev_trans);
1626
1627 put_transaction(prev_trans);
1628 } else {
1629 spin_unlock(&root->fs_info->trans_lock);
1630 }
1631 } else {
1632 spin_unlock(&root->fs_info->trans_lock);
1633 }
1634
1635 if (!btrfs_test_opt(root, SSD) &&
1636 (now < cur_trans->start_time || now - cur_trans->start_time < 1))
1637 should_grow = 1;
1638
1639 do {
1640 joined = cur_trans->num_joined;
1641
1642 WARN_ON(cur_trans != trans->transaction);
1643
1644 ret = btrfs_flush_all_pending_stuffs(trans, root);
1645 if (ret)
1646 goto cleanup_transaction;
1647
1648 prepare_to_wait(&cur_trans->writer_wait, &wait,
1649 TASK_UNINTERRUPTIBLE);
1650
1651 if (atomic_read(&cur_trans->num_writers) > 1)
1652 schedule_timeout(MAX_SCHEDULE_TIMEOUT);
1653 else if (should_grow)
1654 schedule_timeout(1);
1655
1656 finish_wait(&cur_trans->writer_wait, &wait);
1657 } while (atomic_read(&cur_trans->num_writers) > 1 ||
1658 (should_grow && cur_trans->num_joined != joined));
1659
1660 ret = btrfs_flush_all_pending_stuffs(trans, root);
1661 if (ret)
1662 goto cleanup_transaction;
1663
1664
1665
1666
1667
1668
1669 spin_lock(&root->fs_info->trans_lock);
1670 root->fs_info->trans_no_join = 1;
1671 spin_unlock(&root->fs_info->trans_lock);
1672 wait_event(cur_trans->writer_wait,
1673 atomic_read(&cur_trans->num_writers) == 1);
1674
1675
1676 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1677 ret = cur_trans->aborted;
1678 goto cleanup_transaction;
1679 }
1680
1681
1682
1683
1684
1685 mutex_lock(&root->fs_info->reloc_mutex);
1686
1687
1688
1689
1690
1691
1692 ret = create_pending_snapshots(trans, root->fs_info);
1693 if (ret) {
1694 mutex_unlock(&root->fs_info->reloc_mutex);
1695 goto cleanup_transaction;
1696 }
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708 ret = btrfs_run_delayed_items(trans, root);
1709 if (ret) {
1710 mutex_unlock(&root->fs_info->reloc_mutex);
1711 goto cleanup_transaction;
1712 }
1713
1714 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1715 if (ret) {
1716 mutex_unlock(&root->fs_info->reloc_mutex);
1717 goto cleanup_transaction;
1718 }
1719
1720
1721
1722
1723
1724 btrfs_assert_delayed_root_empty(root);
1725
1726 WARN_ON(cur_trans != trans->transaction);
1727
1728 btrfs_scrub_pause(root);
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742 mutex_lock(&root->fs_info->tree_log_mutex);
1743
1744 ret = commit_fs_roots(trans, root);
1745 if (ret) {
1746 mutex_unlock(&root->fs_info->tree_log_mutex);
1747 mutex_unlock(&root->fs_info->reloc_mutex);
1748 goto cleanup_transaction;
1749 }
1750
1751
1752
1753
1754 btrfs_free_log_root_tree(trans, root->fs_info);
1755
1756 ret = commit_cowonly_roots(trans, root);
1757 if (ret) {
1758 mutex_unlock(&root->fs_info->tree_log_mutex);
1759 mutex_unlock(&root->fs_info->reloc_mutex);
1760 goto cleanup_transaction;
1761 }
1762
1763
1764
1765
1766
1767 if (unlikely(ACCESS_ONCE(cur_trans->aborted))) {
1768 ret = cur_trans->aborted;
1769 mutex_unlock(&root->fs_info->tree_log_mutex);
1770 mutex_unlock(&root->fs_info->reloc_mutex);
1771 goto cleanup_transaction;
1772 }
1773
1774 btrfs_prepare_extent_commit(trans, root);
1775
1776 cur_trans = root->fs_info->running_transaction;
1777
1778 btrfs_set_root_node(&root->fs_info->tree_root->root_item,
1779 root->fs_info->tree_root->node);
1780 switch_commit_root(root->fs_info->tree_root);
1781
1782 btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
1783 root->fs_info->chunk_root->node);
1784 switch_commit_root(root->fs_info->chunk_root);
1785
1786 assert_qgroups_uptodate(trans);
1787 update_super_roots(root);
1788
1789 if (!root->fs_info->log_root_recovering) {
1790 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
1791 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
1792 }
1793
1794 memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy,
1795 sizeof(*root->fs_info->super_copy));
1796
1797 trans->transaction->blocked = 0;
1798 spin_lock(&root->fs_info->trans_lock);
1799 root->fs_info->running_transaction = NULL;
1800 root->fs_info->trans_no_join = 0;
1801 spin_unlock(&root->fs_info->trans_lock);
1802 mutex_unlock(&root->fs_info->reloc_mutex);
1803
1804 wake_up(&root->fs_info->transaction_wait);
1805
1806 ret = btrfs_write_and_wait_transaction(trans, root);
1807 if (ret) {
1808 btrfs_error(root->fs_info, ret,
1809 "Error while writing out transaction");
1810 mutex_unlock(&root->fs_info->tree_log_mutex);
1811 goto cleanup_transaction;
1812 }
1813
1814 ret = write_ctree_super(trans, root, 0);
1815 if (ret) {
1816 mutex_unlock(&root->fs_info->tree_log_mutex);
1817 goto cleanup_transaction;
1818 }
1819
1820
1821
1822
1823
1824 mutex_unlock(&root->fs_info->tree_log_mutex);
1825
1826 btrfs_finish_extent_commit(trans, root);
1827
1828 cur_trans->commit_done = 1;
1829
1830 root->fs_info->last_trans_committed = cur_trans->transid;
1831
1832 wake_up(&cur_trans->commit_wait);
1833
1834 spin_lock(&root->fs_info->trans_lock);
1835 list_del_init(&cur_trans->list);
1836 spin_unlock(&root->fs_info->trans_lock);
1837
1838 put_transaction(cur_trans);
1839 put_transaction(cur_trans);
1840
1841 if (trans->type < TRANS_JOIN_NOLOCK)
1842 sb_end_intwrite(root->fs_info->sb);
1843
1844 trace_btrfs_transaction_commit(root);
1845
1846 btrfs_scrub_continue(root);
1847
1848 if (current->journal_info == trans)
1849 current->journal_info = NULL;
1850
1851 kmem_cache_free(btrfs_trans_handle_cachep, trans);
1852
1853 if (current != root->fs_info->transaction_kthread)
1854 btrfs_run_delayed_iputs(root);
1855
1856 return ret;
1857
1858cleanup_transaction:
1859 btrfs_trans_release_metadata(trans, root);
1860 trans->block_rsv = NULL;
1861 if (trans->qgroup_reserved) {
1862 btrfs_qgroup_free(root, trans->qgroup_reserved);
1863 trans->qgroup_reserved = 0;
1864 }
1865 btrfs_warn(root->fs_info, "Skipping commit of aborted transaction.");
1866 if (current->journal_info == trans)
1867 current->journal_info = NULL;
1868 cleanup_transaction(trans, root, ret);
1869
1870 return ret;
1871}
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root)
1884{
1885 int ret;
1886 struct btrfs_fs_info *fs_info = root->fs_info;
1887
1888 if (fs_info->sb->s_flags & MS_RDONLY) {
1889 pr_debug("btrfs: cleaner called for RO fs!\n");
1890 return 0;
1891 }
1892
1893 spin_lock(&fs_info->trans_lock);
1894 if (list_empty(&fs_info->dead_roots)) {
1895 spin_unlock(&fs_info->trans_lock);
1896 return 0;
1897 }
1898 root = list_first_entry(&fs_info->dead_roots,
1899 struct btrfs_root, root_list);
1900 list_del(&root->root_list);
1901 spin_unlock(&fs_info->trans_lock);
1902
1903 pr_debug("btrfs: cleaner removing %llu\n",
1904 (unsigned long long)root->objectid);
1905
1906 btrfs_kill_all_delayed_nodes(root);
1907
1908 if (btrfs_header_backref_rev(root->node) <
1909 BTRFS_MIXED_BACKREF_REV)
1910 ret = btrfs_drop_snapshot(root, NULL, 0, 0);
1911 else
1912 ret = btrfs_drop_snapshot(root, NULL, 1, 0);
1913
1914
1915
1916
1917 BUG_ON(ret < 0 && ret != -EAGAIN && ret != -EROFS);
1918 return 1;
1919}
1920