aboutsummaryrefslogtreecommitdiffstats
path: root/fs/btrfs/transaction.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/btrfs/transaction.c')
-rw-r--r--fs/btrfs/transaction.c950
1 files changed, 950 insertions, 0 deletions
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
new file mode 100644
index 000000000000..151b00d52593
--- /dev/null
+++ b/fs/btrfs/transaction.c
@@ -0,0 +1,950 @@
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
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.
7 *
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.
12 *
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.
17 */
18
19#include <linux/fs.h>
20#include <linux/sched.h>
21#include <linux/writeback.h>
22#include <linux/pagemap.h>
23#include "ctree.h"
24#include "disk-io.h"
25#include "transaction.h"
26#include "locking.h"
27#include "ref-cache.h"
28#include "tree-log.h"
29
30static int total_trans = 0;
31extern struct kmem_cache *btrfs_trans_handle_cachep;
32extern struct kmem_cache *btrfs_transaction_cachep;
33
34#define BTRFS_ROOT_TRANS_TAG 0
35
36static noinline void put_transaction(struct btrfs_transaction *transaction)
37{
38 WARN_ON(transaction->use_count == 0);
39 transaction->use_count--;
40 if (transaction->use_count == 0) {
41 WARN_ON(total_trans == 0);
42 total_trans--;
43 list_del_init(&transaction->list);
44 memset(transaction, 0, sizeof(*transaction));
45 kmem_cache_free(btrfs_transaction_cachep, transaction);
46 }
47}
48
49static noinline int join_transaction(struct btrfs_root *root)
50{
51 struct btrfs_transaction *cur_trans;
52 cur_trans = root->fs_info->running_transaction;
53 if (!cur_trans) {
54 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
55 GFP_NOFS);
56 total_trans++;
57 BUG_ON(!cur_trans);
58 root->fs_info->generation++;
59 root->fs_info->last_alloc = 0;
60 root->fs_info->last_data_alloc = 0;
61 root->fs_info->last_log_alloc = 0;
62 cur_trans->num_writers = 1;
63 cur_trans->num_joined = 0;
64 cur_trans->transid = root->fs_info->generation;
65 init_waitqueue_head(&cur_trans->writer_wait);
66 init_waitqueue_head(&cur_trans->commit_wait);
67 cur_trans->in_commit = 0;
68 cur_trans->blocked = 0;
69 cur_trans->use_count = 1;
70 cur_trans->commit_done = 0;
71 cur_trans->start_time = get_seconds();
72 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
73 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
74 extent_io_tree_init(&cur_trans->dirty_pages,
75 root->fs_info->btree_inode->i_mapping,
76 GFP_NOFS);
77 spin_lock(&root->fs_info->new_trans_lock);
78 root->fs_info->running_transaction = cur_trans;
79 spin_unlock(&root->fs_info->new_trans_lock);
80 } else {
81 cur_trans->num_writers++;
82 cur_trans->num_joined++;
83 }
84
85 return 0;
86}
87
88noinline int btrfs_record_root_in_trans(struct btrfs_root *root)
89{
90 struct btrfs_dirty_root *dirty;
91 u64 running_trans_id = root->fs_info->running_transaction->transid;
92 if (root->ref_cows && root->last_trans < running_trans_id) {
93 WARN_ON(root == root->fs_info->extent_root);
94 if (root->root_item.refs != 0) {
95 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
96 (unsigned long)root->root_key.objectid,
97 BTRFS_ROOT_TRANS_TAG);
98
99 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
100 BUG_ON(!dirty);
101 dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
102 BUG_ON(!dirty->root);
103 dirty->latest_root = root;
104 INIT_LIST_HEAD(&dirty->list);
105
106 root->commit_root = btrfs_root_node(root);
107
108 memcpy(dirty->root, root, sizeof(*root));
109 spin_lock_init(&dirty->root->node_lock);
110 spin_lock_init(&dirty->root->list_lock);
111 mutex_init(&dirty->root->objectid_mutex);
112 INIT_LIST_HEAD(&dirty->root->dead_list);
113 dirty->root->node = root->commit_root;
114 dirty->root->commit_root = NULL;
115
116 spin_lock(&root->list_lock);
117 list_add(&dirty->root->dead_list, &root->dead_list);
118 spin_unlock(&root->list_lock);
119
120 root->dirty_root = dirty;
121 } else {
122 WARN_ON(1);
123 }
124 root->last_trans = running_trans_id;
125 }
126 return 0;
127}
128
129static void wait_current_trans(struct btrfs_root *root)
130{
131 struct btrfs_transaction *cur_trans;
132
133 cur_trans = root->fs_info->running_transaction;
134 if (cur_trans && cur_trans->blocked) {
135 DEFINE_WAIT(wait);
136 cur_trans->use_count++;
137 while(1) {
138 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
139 TASK_UNINTERRUPTIBLE);
140 if (cur_trans->blocked) {
141 mutex_unlock(&root->fs_info->trans_mutex);
142 schedule();
143 mutex_lock(&root->fs_info->trans_mutex);
144 finish_wait(&root->fs_info->transaction_wait,
145 &wait);
146 } else {
147 finish_wait(&root->fs_info->transaction_wait,
148 &wait);
149 break;
150 }
151 }
152 put_transaction(cur_trans);
153 }
154}
155
156static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
157 int num_blocks, int wait)
158{
159 struct btrfs_trans_handle *h =
160 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
161 int ret;
162
163 mutex_lock(&root->fs_info->trans_mutex);
164 if (!root->fs_info->log_root_recovering &&
165 ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2))
166 wait_current_trans(root);
167 ret = join_transaction(root);
168 BUG_ON(ret);
169
170 btrfs_record_root_in_trans(root);
171 h->transid = root->fs_info->running_transaction->transid;
172 h->transaction = root->fs_info->running_transaction;
173 h->blocks_reserved = num_blocks;
174 h->blocks_used = 0;
175 h->block_group = NULL;
176 h->alloc_exclude_nr = 0;
177 h->alloc_exclude_start = 0;
178 root->fs_info->running_transaction->use_count++;
179 mutex_unlock(&root->fs_info->trans_mutex);
180 return h;
181}
182
183struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
184 int num_blocks)
185{
186 return start_transaction(root, num_blocks, 1);
187}
188struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
189 int num_blocks)
190{
191 return start_transaction(root, num_blocks, 0);
192}
193
194struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
195 int num_blocks)
196{
197 return start_transaction(r, num_blocks, 2);
198}
199
200
201static noinline int wait_for_commit(struct btrfs_root *root,
202 struct btrfs_transaction *commit)
203{
204 DEFINE_WAIT(wait);
205 mutex_lock(&root->fs_info->trans_mutex);
206 while(!commit->commit_done) {
207 prepare_to_wait(&commit->commit_wait, &wait,
208 TASK_UNINTERRUPTIBLE);
209 if (commit->commit_done)
210 break;
211 mutex_unlock(&root->fs_info->trans_mutex);
212 schedule();
213 mutex_lock(&root->fs_info->trans_mutex);
214 }
215 mutex_unlock(&root->fs_info->trans_mutex);
216 finish_wait(&commit->commit_wait, &wait);
217 return 0;
218}
219
220static void throttle_on_drops(struct btrfs_root *root)
221{
222 struct btrfs_fs_info *info = root->fs_info;
223 int harder_count = 0;
224
225harder:
226 if (atomic_read(&info->throttles)) {
227 DEFINE_WAIT(wait);
228 int thr;
229 thr = atomic_read(&info->throttle_gen);
230
231 do {
232 prepare_to_wait(&info->transaction_throttle,
233 &wait, TASK_UNINTERRUPTIBLE);
234 if (!atomic_read(&info->throttles)) {
235 finish_wait(&info->transaction_throttle, &wait);
236 break;
237 }
238 schedule();
239 finish_wait(&info->transaction_throttle, &wait);
240 } while (thr == atomic_read(&info->throttle_gen));
241 harder_count++;
242
243 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
244 harder_count < 2)
245 goto harder;
246
247 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
248 harder_count < 10)
249 goto harder;
250
251 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
252 harder_count < 20)
253 goto harder;
254 }
255}
256
257void btrfs_throttle(struct btrfs_root *root)
258{
259 mutex_lock(&root->fs_info->trans_mutex);
260 if (!root->fs_info->open_ioctl_trans)
261 wait_current_trans(root);
262 mutex_unlock(&root->fs_info->trans_mutex);
263
264 throttle_on_drops(root);
265}
266
267static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
268 struct btrfs_root *root, int throttle)
269{
270 struct btrfs_transaction *cur_trans;
271 struct btrfs_fs_info *info = root->fs_info;
272
273 mutex_lock(&info->trans_mutex);
274 cur_trans = info->running_transaction;
275 WARN_ON(cur_trans != trans->transaction);
276 WARN_ON(cur_trans->num_writers < 1);
277 cur_trans->num_writers--;
278
279 if (waitqueue_active(&cur_trans->writer_wait))
280 wake_up(&cur_trans->writer_wait);
281 put_transaction(cur_trans);
282 mutex_unlock(&info->trans_mutex);
283 memset(trans, 0, sizeof(*trans));
284 kmem_cache_free(btrfs_trans_handle_cachep, trans);
285
286 if (throttle)
287 throttle_on_drops(root);
288
289 return 0;
290}
291
292int btrfs_end_transaction(struct btrfs_trans_handle *trans,
293 struct btrfs_root *root)
294{
295 return __btrfs_end_transaction(trans, root, 0);
296}
297
298int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
299 struct btrfs_root *root)
300{
301 return __btrfs_end_transaction(trans, root, 1);
302}
303
304
305int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
306 struct extent_io_tree *dirty_pages)
307{
308 int ret;
309 int err = 0;
310 int werr = 0;
311 struct page *page;
312 struct inode *btree_inode = root->fs_info->btree_inode;
313 u64 start = 0;
314 u64 end;
315 unsigned long index;
316
317 while(1) {
318 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
319 EXTENT_DIRTY);
320 if (ret)
321 break;
322 while(start <= end) {
323 cond_resched();
324
325 index = start >> PAGE_CACHE_SHIFT;
326 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
327 page = find_get_page(btree_inode->i_mapping, index);
328 if (!page)
329 continue;
330
331 btree_lock_page_hook(page);
332 if (!page->mapping) {
333 unlock_page(page);
334 page_cache_release(page);
335 continue;
336 }
337
338 if (PageWriteback(page)) {
339 if (PageDirty(page))
340 wait_on_page_writeback(page);
341 else {
342 unlock_page(page);
343 page_cache_release(page);
344 continue;
345 }
346 }
347 err = write_one_page(page, 0);
348 if (err)
349 werr = err;
350 page_cache_release(page);
351 }
352 }
353 while(1) {
354 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
355 EXTENT_DIRTY);
356 if (ret)
357 break;
358
359 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
360 while(start <= end) {
361 index = start >> PAGE_CACHE_SHIFT;
362 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
363 page = find_get_page(btree_inode->i_mapping, index);
364 if (!page)
365 continue;
366 if (PageDirty(page)) {
367 btree_lock_page_hook(page);
368 wait_on_page_writeback(page);
369 err = write_one_page(page, 0);
370 if (err)
371 werr = err;
372 }
373 wait_on_page_writeback(page);
374 page_cache_release(page);
375 cond_resched();
376 }
377 }
378 if (err)
379 werr = err;
380 return werr;
381}
382
383int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
384 struct btrfs_root *root)
385{
386 if (!trans || !trans->transaction) {
387 struct inode *btree_inode;
388 btree_inode = root->fs_info->btree_inode;
389 return filemap_write_and_wait(btree_inode->i_mapping);
390 }
391 return btrfs_write_and_wait_marked_extents(root,
392 &trans->transaction->dirty_pages);
393}
394
395static int update_cowonly_root(struct btrfs_trans_handle *trans,
396 struct btrfs_root *root)
397{
398 int ret;
399 u64 old_root_bytenr;
400 struct btrfs_root *tree_root = root->fs_info->tree_root;
401
402 btrfs_write_dirty_block_groups(trans, root);
403 while(1) {
404 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
405 if (old_root_bytenr == root->node->start)
406 break;
407 btrfs_set_root_bytenr(&root->root_item,
408 root->node->start);
409 btrfs_set_root_level(&root->root_item,
410 btrfs_header_level(root->node));
411 ret = btrfs_update_root(trans, tree_root,
412 &root->root_key,
413 &root->root_item);
414 BUG_ON(ret);
415 btrfs_write_dirty_block_groups(trans, root);
416 }
417 return 0;
418}
419
420int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
421 struct btrfs_root *root)
422{
423 struct btrfs_fs_info *fs_info = root->fs_info;
424 struct list_head *next;
425
426 while(!list_empty(&fs_info->dirty_cowonly_roots)) {
427 next = fs_info->dirty_cowonly_roots.next;
428 list_del_init(next);
429 root = list_entry(next, struct btrfs_root, dirty_list);
430 update_cowonly_root(trans, root);
431 }
432 return 0;
433}
434
435int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest)
436{
437 struct btrfs_dirty_root *dirty;
438
439 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
440 if (!dirty)
441 return -ENOMEM;
442 dirty->root = root;
443 dirty->latest_root = latest;
444
445 mutex_lock(&root->fs_info->trans_mutex);
446 list_add(&dirty->list, &latest->fs_info->dead_roots);
447 mutex_unlock(&root->fs_info->trans_mutex);
448 return 0;
449}
450
451static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
452 struct radix_tree_root *radix,
453 struct list_head *list)
454{
455 struct btrfs_dirty_root *dirty;
456 struct btrfs_root *gang[8];
457 struct btrfs_root *root;
458 int i;
459 int ret;
460 int err = 0;
461 u32 refs;
462
463 while(1) {
464 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
465 ARRAY_SIZE(gang),
466 BTRFS_ROOT_TRANS_TAG);
467 if (ret == 0)
468 break;
469 for (i = 0; i < ret; i++) {
470 root = gang[i];
471 radix_tree_tag_clear(radix,
472 (unsigned long)root->root_key.objectid,
473 BTRFS_ROOT_TRANS_TAG);
474
475 BUG_ON(!root->ref_tree);
476 dirty = root->dirty_root;
477
478 btrfs_free_log(trans, root);
479
480 if (root->commit_root == root->node) {
481 WARN_ON(root->node->start !=
482 btrfs_root_bytenr(&root->root_item));
483
484 free_extent_buffer(root->commit_root);
485 root->commit_root = NULL;
486 root->dirty_root = NULL;
487
488 spin_lock(&root->list_lock);
489 list_del_init(&dirty->root->dead_list);
490 spin_unlock(&root->list_lock);
491
492 kfree(dirty->root);
493 kfree(dirty);
494
495 /* make sure to update the root on disk
496 * so we get any updates to the block used
497 * counts
498 */
499 err = btrfs_update_root(trans,
500 root->fs_info->tree_root,
501 &root->root_key,
502 &root->root_item);
503 continue;
504 }
505
506 memset(&root->root_item.drop_progress, 0,
507 sizeof(struct btrfs_disk_key));
508 root->root_item.drop_level = 0;
509 root->commit_root = NULL;
510 root->dirty_root = NULL;
511 root->root_key.offset = root->fs_info->generation;
512 btrfs_set_root_bytenr(&root->root_item,
513 root->node->start);
514 btrfs_set_root_level(&root->root_item,
515 btrfs_header_level(root->node));
516 err = btrfs_insert_root(trans, root->fs_info->tree_root,
517 &root->root_key,
518 &root->root_item);
519 if (err)
520 break;
521
522 refs = btrfs_root_refs(&dirty->root->root_item);
523 btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
524 err = btrfs_update_root(trans, root->fs_info->tree_root,
525 &dirty->root->root_key,
526 &dirty->root->root_item);
527
528 BUG_ON(err);
529 if (refs == 1) {
530 list_add(&dirty->list, list);
531 } else {
532 WARN_ON(1);
533 free_extent_buffer(dirty->root->node);
534 kfree(dirty->root);
535 kfree(dirty);
536 }
537 }
538 }
539 return err;
540}
541
542int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
543{
544 struct btrfs_fs_info *info = root->fs_info;
545 int ret;
546 struct btrfs_trans_handle *trans;
547 unsigned long nr;
548
549 smp_mb();
550 if (root->defrag_running)
551 return 0;
552 trans = btrfs_start_transaction(root, 1);
553 while (1) {
554 root->defrag_running = 1;
555 ret = btrfs_defrag_leaves(trans, root, cacheonly);
556 nr = trans->blocks_used;
557 btrfs_end_transaction(trans, root);
558 btrfs_btree_balance_dirty(info->tree_root, nr);
559 cond_resched();
560
561 trans = btrfs_start_transaction(root, 1);
562 if (root->fs_info->closing || ret != -EAGAIN)
563 break;
564 }
565 root->defrag_running = 0;
566 smp_mb();
567 btrfs_end_transaction(trans, root);
568 return 0;
569}
570
571static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
572 struct list_head *list)
573{
574 struct btrfs_dirty_root *dirty;
575 struct btrfs_trans_handle *trans;
576 unsigned long nr;
577 u64 num_bytes;
578 u64 bytes_used;
579 u64 max_useless;
580 int ret = 0;
581 int err;
582
583 while(!list_empty(list)) {
584 struct btrfs_root *root;
585
586 dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
587 list_del_init(&dirty->list);
588
589 num_bytes = btrfs_root_used(&dirty->root->root_item);
590 root = dirty->latest_root;
591 atomic_inc(&root->fs_info->throttles);
592
593 mutex_lock(&root->fs_info->drop_mutex);
594 while(1) {
595 trans = btrfs_start_transaction(tree_root, 1);
596 ret = btrfs_drop_snapshot(trans, dirty->root);
597 if (ret != -EAGAIN) {
598 break;
599 }
600
601 err = btrfs_update_root(trans,
602 tree_root,
603 &dirty->root->root_key,
604 &dirty->root->root_item);
605 if (err)
606 ret = err;
607 nr = trans->blocks_used;
608 ret = btrfs_end_transaction(trans, tree_root);
609 BUG_ON(ret);
610
611 mutex_unlock(&root->fs_info->drop_mutex);
612 btrfs_btree_balance_dirty(tree_root, nr);
613 cond_resched();
614 mutex_lock(&root->fs_info->drop_mutex);
615 }
616 BUG_ON(ret);
617 atomic_dec(&root->fs_info->throttles);
618 wake_up(&root->fs_info->transaction_throttle);
619
620 mutex_lock(&root->fs_info->alloc_mutex);
621 num_bytes -= btrfs_root_used(&dirty->root->root_item);
622 bytes_used = btrfs_root_used(&root->root_item);
623 if (num_bytes) {
624 btrfs_record_root_in_trans(root);
625 btrfs_set_root_used(&root->root_item,
626 bytes_used - num_bytes);
627 }
628 mutex_unlock(&root->fs_info->alloc_mutex);
629
630 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
631 if (ret) {
632 BUG();
633 break;
634 }
635 mutex_unlock(&root->fs_info->drop_mutex);
636
637 spin_lock(&root->list_lock);
638 list_del_init(&dirty->root->dead_list);
639 if (!list_empty(&root->dead_list)) {
640 struct btrfs_root *oldest;
641 oldest = list_entry(root->dead_list.prev,
642 struct btrfs_root, dead_list);
643 max_useless = oldest->root_key.offset - 1;
644 } else {
645 max_useless = root->root_key.offset - 1;
646 }
647 spin_unlock(&root->list_lock);
648
649 nr = trans->blocks_used;
650 ret = btrfs_end_transaction(trans, tree_root);
651 BUG_ON(ret);
652
653 ret = btrfs_remove_leaf_refs(root, max_useless);
654 BUG_ON(ret);
655
656 free_extent_buffer(dirty->root->node);
657 kfree(dirty->root);
658 kfree(dirty);
659
660 btrfs_btree_balance_dirty(tree_root, nr);
661 cond_resched();
662 }
663 return ret;
664}
665
666static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
667 struct btrfs_fs_info *fs_info,
668 struct btrfs_pending_snapshot *pending)
669{
670 struct btrfs_key key;
671 struct btrfs_root_item *new_root_item;
672 struct btrfs_root *tree_root = fs_info->tree_root;
673 struct btrfs_root *root = pending->root;
674 struct extent_buffer *tmp;
675 struct extent_buffer *old;
676 int ret;
677 int namelen;
678 u64 objectid;
679
680 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
681 if (!new_root_item) {
682 ret = -ENOMEM;
683 goto fail;
684 }
685 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
686 if (ret)
687 goto fail;
688
689 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
690
691 key.objectid = objectid;
692 key.offset = 1;
693 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
694
695 old = btrfs_lock_root_node(root);
696 btrfs_cow_block(trans, root, old, NULL, 0, &old, 0);
697
698 btrfs_copy_root(trans, root, old, &tmp, objectid);
699 btrfs_tree_unlock(old);
700 free_extent_buffer(old);
701
702 btrfs_set_root_bytenr(new_root_item, tmp->start);
703 btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
704 ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
705 new_root_item);
706 btrfs_tree_unlock(tmp);
707 free_extent_buffer(tmp);
708 if (ret)
709 goto fail;
710
711 /*
712 * insert the directory item
713 */
714 key.offset = (u64)-1;
715 namelen = strlen(pending->name);
716 ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
717 pending->name, namelen,
718 root->fs_info->sb->s_root->d_inode->i_ino,
719 &key, BTRFS_FT_DIR, 0);
720
721 if (ret)
722 goto fail;
723
724 ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
725 pending->name, strlen(pending->name), objectid,
726 root->fs_info->sb->s_root->d_inode->i_ino, 0);
727
728 /* Invalidate existing dcache entry for new snapshot. */
729 btrfs_invalidate_dcache_root(root, pending->name, namelen);
730
731fail:
732 kfree(new_root_item);
733 return ret;
734}
735
736static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
737 struct btrfs_fs_info *fs_info)
738{
739 struct btrfs_pending_snapshot *pending;
740 struct list_head *head = &trans->transaction->pending_snapshots;
741 int ret;
742
743 while(!list_empty(head)) {
744 pending = list_entry(head->next,
745 struct btrfs_pending_snapshot, list);
746 ret = create_pending_snapshot(trans, fs_info, pending);
747 BUG_ON(ret);
748 list_del(&pending->list);
749 kfree(pending->name);
750 kfree(pending);
751 }
752 return 0;
753}
754
755int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
756 struct btrfs_root *root)
757{
758 unsigned long joined = 0;
759 unsigned long timeout = 1;
760 struct btrfs_transaction *cur_trans;
761 struct btrfs_transaction *prev_trans = NULL;
762 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
763 struct list_head dirty_fs_roots;
764 struct extent_io_tree *pinned_copy;
765 DEFINE_WAIT(wait);
766 int ret;
767
768 INIT_LIST_HEAD(&dirty_fs_roots);
769 mutex_lock(&root->fs_info->trans_mutex);
770 if (trans->transaction->in_commit) {
771 cur_trans = trans->transaction;
772 trans->transaction->use_count++;
773 mutex_unlock(&root->fs_info->trans_mutex);
774 btrfs_end_transaction(trans, root);
775
776 ret = wait_for_commit(root, cur_trans);
777 BUG_ON(ret);
778
779 mutex_lock(&root->fs_info->trans_mutex);
780 put_transaction(cur_trans);
781 mutex_unlock(&root->fs_info->trans_mutex);
782
783 return 0;
784 }
785
786 pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
787 if (!pinned_copy)
788 return -ENOMEM;
789
790 extent_io_tree_init(pinned_copy,
791 root->fs_info->btree_inode->i_mapping, GFP_NOFS);
792
793 trans->transaction->in_commit = 1;
794 trans->transaction->blocked = 1;
795 cur_trans = trans->transaction;
796 if (cur_trans->list.prev != &root->fs_info->trans_list) {
797 prev_trans = list_entry(cur_trans->list.prev,
798 struct btrfs_transaction, list);
799 if (!prev_trans->commit_done) {
800 prev_trans->use_count++;
801 mutex_unlock(&root->fs_info->trans_mutex);
802
803 wait_for_commit(root, prev_trans);
804
805 mutex_lock(&root->fs_info->trans_mutex);
806 put_transaction(prev_trans);
807 }
808 }
809
810 do {
811 int snap_pending = 0;
812 joined = cur_trans->num_joined;
813 if (!list_empty(&trans->transaction->pending_snapshots))
814 snap_pending = 1;
815
816 WARN_ON(cur_trans != trans->transaction);
817 prepare_to_wait(&cur_trans->writer_wait, &wait,
818 TASK_UNINTERRUPTIBLE);
819
820 if (cur_trans->num_writers > 1)
821 timeout = MAX_SCHEDULE_TIMEOUT;
822 else
823 timeout = 1;
824
825 mutex_unlock(&root->fs_info->trans_mutex);
826
827 if (snap_pending) {
828 ret = btrfs_wait_ordered_extents(root, 1);
829 BUG_ON(ret);
830 }
831
832 schedule_timeout(timeout);
833
834 mutex_lock(&root->fs_info->trans_mutex);
835 finish_wait(&cur_trans->writer_wait, &wait);
836 } while (cur_trans->num_writers > 1 ||
837 (cur_trans->num_joined != joined));
838
839 ret = create_pending_snapshots(trans, root->fs_info);
840 BUG_ON(ret);
841
842 WARN_ON(cur_trans != trans->transaction);
843
844 /* btrfs_commit_tree_roots is responsible for getting the
845 * various roots consistent with each other. Every pointer
846 * in the tree of tree roots has to point to the most up to date
847 * root for every subvolume and other tree. So, we have to keep
848 * the tree logging code from jumping in and changing any
849 * of the trees.
850 *
851 * At this point in the commit, there can't be any tree-log
852 * writers, but a little lower down we drop the trans mutex
853 * and let new people in. By holding the tree_log_mutex
854 * from now until after the super is written, we avoid races
855 * with the tree-log code.
856 */
857 mutex_lock(&root->fs_info->tree_log_mutex);
858
859 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
860 &dirty_fs_roots);
861 BUG_ON(ret);
862
863 /* add_dirty_roots gets rid of all the tree log roots, it is now
864 * safe to free the root of tree log roots
865 */
866 btrfs_free_log_root_tree(trans, root->fs_info);
867
868 ret = btrfs_commit_tree_roots(trans, root);
869 BUG_ON(ret);
870
871 cur_trans = root->fs_info->running_transaction;
872 spin_lock(&root->fs_info->new_trans_lock);
873 root->fs_info->running_transaction = NULL;
874 spin_unlock(&root->fs_info->new_trans_lock);
875 btrfs_set_super_generation(&root->fs_info->super_copy,
876 cur_trans->transid);
877 btrfs_set_super_root(&root->fs_info->super_copy,
878 root->fs_info->tree_root->node->start);
879 btrfs_set_super_root_level(&root->fs_info->super_copy,
880 btrfs_header_level(root->fs_info->tree_root->node));
881
882 btrfs_set_super_chunk_root(&root->fs_info->super_copy,
883 chunk_root->node->start);
884 btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
885 btrfs_header_level(chunk_root->node));
886
887 if (!root->fs_info->log_root_recovering) {
888 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
889 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
890 }
891
892 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
893 sizeof(root->fs_info->super_copy));
894
895 btrfs_copy_pinned(root, pinned_copy);
896
897 trans->transaction->blocked = 0;
898 wake_up(&root->fs_info->transaction_throttle);
899 wake_up(&root->fs_info->transaction_wait);
900
901 mutex_unlock(&root->fs_info->trans_mutex);
902 ret = btrfs_write_and_wait_transaction(trans, root);
903 BUG_ON(ret);
904 write_ctree_super(trans, root);
905
906 /*
907 * the super is written, we can safely allow the tree-loggers
908 * to go about their business
909 */
910 mutex_unlock(&root->fs_info->tree_log_mutex);
911
912 btrfs_finish_extent_commit(trans, root, pinned_copy);
913 mutex_lock(&root->fs_info->trans_mutex);
914
915 kfree(pinned_copy);
916
917 cur_trans->commit_done = 1;
918 root->fs_info->last_trans_committed = cur_trans->transid;
919 wake_up(&cur_trans->commit_wait);
920 put_transaction(cur_trans);
921 put_transaction(cur_trans);
922
923 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
924 if (root->fs_info->closing)
925 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
926
927 mutex_unlock(&root->fs_info->trans_mutex);
928 kmem_cache_free(btrfs_trans_handle_cachep, trans);
929
930 if (root->fs_info->closing) {
931 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
932 }
933 return ret;
934}
935
936int btrfs_clean_old_snapshots(struct btrfs_root *root)
937{
938 struct list_head dirty_roots;
939 INIT_LIST_HEAD(&dirty_roots);
940again:
941 mutex_lock(&root->fs_info->trans_mutex);
942 list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
943 mutex_unlock(&root->fs_info->trans_mutex);
944
945 if (!list_empty(&dirty_roots)) {
946 drop_dirty_roots(root, &dirty_roots);
947 goto again;
948 }
949 return 0;
950}
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-25 20:15:52 -0500