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.c1023
1 files changed, 1023 insertions, 0 deletions
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
new file mode 100644
index 000000000000..5ecc24d634a2
--- /dev/null
+++ b/fs/btrfs/transaction.c
@@ -0,0 +1,1023 @@
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
49/*
50 * either allocate a new transaction or hop into the existing one
51 */
52static noinline int join_transaction(struct btrfs_root *root)
53{
54 struct btrfs_transaction *cur_trans;
55 cur_trans = root->fs_info->running_transaction;
56 if (!cur_trans) {
57 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
58 GFP_NOFS);
59 total_trans++;
60 BUG_ON(!cur_trans);
61 root->fs_info->generation++;
62 root->fs_info->last_alloc = 0;
63 root->fs_info->last_data_alloc = 0;
64 cur_trans->num_writers = 1;
65 cur_trans->num_joined = 0;
66 cur_trans->transid = root->fs_info->generation;
67 init_waitqueue_head(&cur_trans->writer_wait);
68 init_waitqueue_head(&cur_trans->commit_wait);
69 cur_trans->in_commit = 0;
70 cur_trans->blocked = 0;
71 cur_trans->use_count = 1;
72 cur_trans->commit_done = 0;
73 cur_trans->start_time = get_seconds();
74 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
75 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
76 extent_io_tree_init(&cur_trans->dirty_pages,
77 root->fs_info->btree_inode->i_mapping,
78 GFP_NOFS);
79 spin_lock(&root->fs_info->new_trans_lock);
80 root->fs_info->running_transaction = cur_trans;
81 spin_unlock(&root->fs_info->new_trans_lock);
82 } else {
83 cur_trans->num_writers++;
84 cur_trans->num_joined++;
85 }
86
87 return 0;
88}
89
90/*
91 * this does all the record keeping required to make sure that a
92 * reference counted root is properly recorded in a given transaction.
93 * This is required to make sure the old root from before we joined the transaction
94 * is deleted when the transaction commits
95 */
96noinline int btrfs_record_root_in_trans(struct btrfs_root *root)
97{
98 struct btrfs_dirty_root *dirty;
99 u64 running_trans_id = root->fs_info->running_transaction->transid;
100 if (root->ref_cows && root->last_trans < running_trans_id) {
101 WARN_ON(root == root->fs_info->extent_root);
102 if (root->root_item.refs != 0) {
103 radix_tree_tag_set(&root->fs_info->fs_roots_radix,
104 (unsigned long)root->root_key.objectid,
105 BTRFS_ROOT_TRANS_TAG);
106
107 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
108 BUG_ON(!dirty);
109 dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
110 BUG_ON(!dirty->root);
111 dirty->latest_root = root;
112 INIT_LIST_HEAD(&dirty->list);
113
114 root->commit_root = btrfs_root_node(root);
115
116 memcpy(dirty->root, root, sizeof(*root));
117 spin_lock_init(&dirty->root->node_lock);
118 spin_lock_init(&dirty->root->list_lock);
119 mutex_init(&dirty->root->objectid_mutex);
120 mutex_init(&dirty->root->log_mutex);
121 INIT_LIST_HEAD(&dirty->root->dead_list);
122 dirty->root->node = root->commit_root;
123 dirty->root->commit_root = NULL;
124
125 spin_lock(&root->list_lock);
126 list_add(&dirty->root->dead_list, &root->dead_list);
127 spin_unlock(&root->list_lock);
128
129 root->dirty_root = dirty;
130 } else {
131 WARN_ON(1);
132 }
133 root->last_trans = running_trans_id;
134 }
135 return 0;
136}
137
138/* wait for commit against the current transaction to become unblocked
139 * when this is done, it is safe to start a new transaction, but the current
140 * transaction might not be fully on disk.
141 */
142static void wait_current_trans(struct btrfs_root *root)
143{
144 struct btrfs_transaction *cur_trans;
145
146 cur_trans = root->fs_info->running_transaction;
147 if (cur_trans && cur_trans->blocked) {
148 DEFINE_WAIT(wait);
149 cur_trans->use_count++;
150 while(1) {
151 prepare_to_wait(&root->fs_info->transaction_wait, &wait,
152 TASK_UNINTERRUPTIBLE);
153 if (cur_trans->blocked) {
154 mutex_unlock(&root->fs_info->trans_mutex);
155 schedule();
156 mutex_lock(&root->fs_info->trans_mutex);
157 finish_wait(&root->fs_info->transaction_wait,
158 &wait);
159 } else {
160 finish_wait(&root->fs_info->transaction_wait,
161 &wait);
162 break;
163 }
164 }
165 put_transaction(cur_trans);
166 }
167}
168
169static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
170 int num_blocks, int wait)
171{
172 struct btrfs_trans_handle *h =
173 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
174 int ret;
175
176 mutex_lock(&root->fs_info->trans_mutex);
177 if (!root->fs_info->log_root_recovering &&
178 ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2))
179 wait_current_trans(root);
180 ret = join_transaction(root);
181 BUG_ON(ret);
182
183 btrfs_record_root_in_trans(root);
184 h->transid = root->fs_info->running_transaction->transid;
185 h->transaction = root->fs_info->running_transaction;
186 h->blocks_reserved = num_blocks;
187 h->blocks_used = 0;
188 h->block_group = NULL;
189 h->alloc_exclude_nr = 0;
190 h->alloc_exclude_start = 0;
191 root->fs_info->running_transaction->use_count++;
192 mutex_unlock(&root->fs_info->trans_mutex);
193 return h;
194}
195
196struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
197 int num_blocks)
198{
199 return start_transaction(root, num_blocks, 1);
200}
201struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
202 int num_blocks)
203{
204 return start_transaction(root, num_blocks, 0);
205}
206
207struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
208 int num_blocks)
209{
210 return start_transaction(r, num_blocks, 2);
211}
212
213/* wait for a transaction commit to be fully complete */
214static noinline int wait_for_commit(struct btrfs_root *root,
215 struct btrfs_transaction *commit)
216{
217 DEFINE_WAIT(wait);
218 mutex_lock(&root->fs_info->trans_mutex);
219 while(!commit->commit_done) {
220 prepare_to_wait(&commit->commit_wait, &wait,
221 TASK_UNINTERRUPTIBLE);
222 if (commit->commit_done)
223 break;
224 mutex_unlock(&root->fs_info->trans_mutex);
225 schedule();
226 mutex_lock(&root->fs_info->trans_mutex);
227 }
228 mutex_unlock(&root->fs_info->trans_mutex);
229 finish_wait(&commit->commit_wait, &wait);
230 return 0;
231}
232
233/*
234 * rate limit against the drop_snapshot code. This helps to slow down new operations
235 * if the drop_snapshot code isn't able to keep up.
236 */
237static void throttle_on_drops(struct btrfs_root *root)
238{
239 struct btrfs_fs_info *info = root->fs_info;
240 int harder_count = 0;
241
242harder:
243 if (atomic_read(&info->throttles)) {
244 DEFINE_WAIT(wait);
245 int thr;
246 thr = atomic_read(&info->throttle_gen);
247
248 do {
249 prepare_to_wait(&info->transaction_throttle,
250 &wait, TASK_UNINTERRUPTIBLE);
251 if (!atomic_read(&info->throttles)) {
252 finish_wait(&info->transaction_throttle, &wait);
253 break;
254 }
255 schedule();
256 finish_wait(&info->transaction_throttle, &wait);
257 } while (thr == atomic_read(&info->throttle_gen));
258 harder_count++;
259
260 if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
261 harder_count < 2)
262 goto harder;
263
264 if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
265 harder_count < 10)
266 goto harder;
267
268 if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
269 harder_count < 20)
270 goto harder;
271 }
272}
273
274void btrfs_throttle(struct btrfs_root *root)
275{
276 mutex_lock(&root->fs_info->trans_mutex);
277 if (!root->fs_info->open_ioctl_trans)
278 wait_current_trans(root);
279 mutex_unlock(&root->fs_info->trans_mutex);
280
281 throttle_on_drops(root);
282}
283
284static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
285 struct btrfs_root *root, int throttle)
286{
287 struct btrfs_transaction *cur_trans;
288 struct btrfs_fs_info *info = root->fs_info;
289
290 mutex_lock(&info->trans_mutex);
291 cur_trans = info->running_transaction;
292 WARN_ON(cur_trans != trans->transaction);
293 WARN_ON(cur_trans->num_writers < 1);
294 cur_trans->num_writers--;
295
296 if (waitqueue_active(&cur_trans->writer_wait))
297 wake_up(&cur_trans->writer_wait);
298 put_transaction(cur_trans);
299 mutex_unlock(&info->trans_mutex);
300 memset(trans, 0, sizeof(*trans));
301 kmem_cache_free(btrfs_trans_handle_cachep, trans);
302
303 if (throttle)
304 throttle_on_drops(root);
305
306 return 0;
307}
308
309int btrfs_end_transaction(struct btrfs_trans_handle *trans,
310 struct btrfs_root *root)
311{
312 return __btrfs_end_transaction(trans, root, 0);
313}
314
315int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
316 struct btrfs_root *root)
317{
318 return __btrfs_end_transaction(trans, root, 1);
319}
320
321/*
322 * when btree blocks are allocated, they have some corresponding bits set for
323 * them in one of two extent_io trees. This is used to make sure all of
324 * those extents are on disk for transaction or log commit
325 */
326int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
327 struct extent_io_tree *dirty_pages)
328{
329 int ret;
330 int err = 0;
331 int werr = 0;
332 struct page *page;
333 struct inode *btree_inode = root->fs_info->btree_inode;
334 u64 start = 0;
335 u64 end;
336 unsigned long index;
337
338 while(1) {
339 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
340 EXTENT_DIRTY);
341 if (ret)
342 break;
343 while(start <= end) {
344 cond_resched();
345
346 index = start >> PAGE_CACHE_SHIFT;
347 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
348 page = find_get_page(btree_inode->i_mapping, index);
349 if (!page)
350 continue;
351
352 btree_lock_page_hook(page);
353 if (!page->mapping) {
354 unlock_page(page);
355 page_cache_release(page);
356 continue;
357 }
358
359 if (PageWriteback(page)) {
360 if (PageDirty(page))
361 wait_on_page_writeback(page);
362 else {
363 unlock_page(page);
364 page_cache_release(page);
365 continue;
366 }
367 }
368 err = write_one_page(page, 0);
369 if (err)
370 werr = err;
371 page_cache_release(page);
372 }
373 }
374 while(1) {
375 ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
376 EXTENT_DIRTY);
377 if (ret)
378 break;
379
380 clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
381 while(start <= end) {
382 index = start >> PAGE_CACHE_SHIFT;
383 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
384 page = find_get_page(btree_inode->i_mapping, index);
385 if (!page)
386 continue;
387 if (PageDirty(page)) {
388 btree_lock_page_hook(page);
389 wait_on_page_writeback(page);
390 err = write_one_page(page, 0);
391 if (err)
392 werr = err;
393 }
394 wait_on_page_writeback(page);
395 page_cache_release(page);
396 cond_resched();
397 }
398 }
399 if (err)
400 werr = err;
401 return werr;
402}
403
404int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
405 struct btrfs_root *root)
406{
407 if (!trans || !trans->transaction) {
408 struct inode *btree_inode;
409 btree_inode = root->fs_info->btree_inode;
410 return filemap_write_and_wait(btree_inode->i_mapping);
411 }
412 return btrfs_write_and_wait_marked_extents(root,
413 &trans->transaction->dirty_pages);
414}
415
416/*
417 * this is used to update the root pointer in the tree of tree roots.
418 *
419 * But, in the case of the extent allocation tree, updating the root
420 * pointer may allocate blocks which may change the root of the extent
421 * allocation tree.
422 *
423 * So, this loops and repeats and makes sure the cowonly root didn't
424 * change while the root pointer was being updated in the metadata.
425 */
426static int update_cowonly_root(struct btrfs_trans_handle *trans,
427 struct btrfs_root *root)
428{
429 int ret;
430 u64 old_root_bytenr;
431 struct btrfs_root *tree_root = root->fs_info->tree_root;
432
433 btrfs_write_dirty_block_groups(trans, root);
434 while(1) {
435 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
436 if (old_root_bytenr == root->node->start)
437 break;
438 btrfs_set_root_bytenr(&root->root_item,
439 root->node->start);
440 btrfs_set_root_level(&root->root_item,
441 btrfs_header_level(root->node));
442 ret = btrfs_update_root(trans, tree_root,
443 &root->root_key,
444 &root->root_item);
445 BUG_ON(ret);
446 btrfs_write_dirty_block_groups(trans, root);
447 }
448 return 0;
449}
450
451/*
452 * update all the cowonly tree roots on disk
453 */
454int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
455 struct btrfs_root *root)
456{
457 struct btrfs_fs_info *fs_info = root->fs_info;
458 struct list_head *next;
459
460 while(!list_empty(&fs_info->dirty_cowonly_roots)) {
461 next = fs_info->dirty_cowonly_roots.next;
462 list_del_init(next);
463 root = list_entry(next, struct btrfs_root, dirty_list);
464 update_cowonly_root(trans, root);
465 }
466 return 0;
467}
468
469/*
470 * dead roots are old snapshots that need to be deleted. This allocates
471 * a dirty root struct and adds it into the list of dead roots that need to
472 * be deleted
473 */
474int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest)
475{
476 struct btrfs_dirty_root *dirty;
477
478 dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
479 if (!dirty)
480 return -ENOMEM;
481 dirty->root = root;
482 dirty->latest_root = latest;
483
484 mutex_lock(&root->fs_info->trans_mutex);
485 list_add(&dirty->list, &latest->fs_info->dead_roots);
486 mutex_unlock(&root->fs_info->trans_mutex);
487 return 0;
488}
489
490/*
491 * at transaction commit time we need to schedule the old roots for
492 * deletion via btrfs_drop_snapshot. This runs through all the
493 * reference counted roots that were modified in the current
494 * transaction and puts them into the drop list
495 */
496static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
497 struct radix_tree_root *radix,
498 struct list_head *list)
499{
500 struct btrfs_dirty_root *dirty;
501 struct btrfs_root *gang[8];
502 struct btrfs_root *root;
503 int i;
504 int ret;
505 int err = 0;
506 u32 refs;
507
508 while(1) {
509 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
510 ARRAY_SIZE(gang),
511 BTRFS_ROOT_TRANS_TAG);
512 if (ret == 0)
513 break;
514 for (i = 0; i < ret; i++) {
515 root = gang[i];
516 radix_tree_tag_clear(radix,
517 (unsigned long)root->root_key.objectid,
518 BTRFS_ROOT_TRANS_TAG);
519
520 BUG_ON(!root->ref_tree);
521 dirty = root->dirty_root;
522
523 btrfs_free_log(trans, root);
524 btrfs_free_reloc_root(root);
525
526 if (root->commit_root == root->node) {
527 WARN_ON(root->node->start !=
528 btrfs_root_bytenr(&root->root_item));
529
530 free_extent_buffer(root->commit_root);
531 root->commit_root = NULL;
532 root->dirty_root = NULL;
533
534 spin_lock(&root->list_lock);
535 list_del_init(&dirty->root->dead_list);
536 spin_unlock(&root->list_lock);
537
538 kfree(dirty->root);
539 kfree(dirty);
540
541 /* make sure to update the root on disk
542 * so we get any updates to the block used
543 * counts
544 */
545 err = btrfs_update_root(trans,
546 root->fs_info->tree_root,
547 &root->root_key,
548 &root->root_item);
549 continue;
550 }
551
552 memset(&root->root_item.drop_progress, 0,
553 sizeof(struct btrfs_disk_key));
554 root->root_item.drop_level = 0;
555 root->commit_root = NULL;
556 root->dirty_root = NULL;
557 root->root_key.offset = root->fs_info->generation;
558 btrfs_set_root_bytenr(&root->root_item,
559 root->node->start);
560 btrfs_set_root_level(&root->root_item,
561 btrfs_header_level(root->node));
562 err = btrfs_insert_root(trans, root->fs_info->tree_root,
563 &root->root_key,
564 &root->root_item);
565 if (err)
566 break;
567
568 refs = btrfs_root_refs(&dirty->root->root_item);
569 btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
570 err = btrfs_update_root(trans, root->fs_info->tree_root,
571 &dirty->root->root_key,
572 &dirty->root->root_item);
573
574 BUG_ON(err);
575 if (refs == 1) {
576 list_add(&dirty->list, list);
577 } else {
578 WARN_ON(1);
579 free_extent_buffer(dirty->root->node);
580 kfree(dirty->root);
581 kfree(dirty);
582 }
583 }
584 }
585 return err;
586}
587
588/*
589 * defrag a given btree. If cacheonly == 1, this won't read from the disk,
590 * otherwise every leaf in the btree is read and defragged.
591 */
592int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
593{
594 struct btrfs_fs_info *info = root->fs_info;
595 int ret;
596 struct btrfs_trans_handle *trans;
597 unsigned long nr;
598
599 smp_mb();
600 if (root->defrag_running)
601 return 0;
602 trans = btrfs_start_transaction(root, 1);
603 while (1) {
604 root->defrag_running = 1;
605 ret = btrfs_defrag_leaves(trans, root, cacheonly);
606 nr = trans->blocks_used;
607 btrfs_end_transaction(trans, root);
608 btrfs_btree_balance_dirty(info->tree_root, nr);
609 cond_resched();
610
611 trans = btrfs_start_transaction(root, 1);
612 if (root->fs_info->closing || ret != -EAGAIN)
613 break;
614 }
615 root->defrag_running = 0;
616 smp_mb();
617 btrfs_end_transaction(trans, root);
618 return 0;
619}
620
621/*
622 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
623 * all of them
624 */
625static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
626 struct list_head *list)
627{
628 struct btrfs_dirty_root *dirty;
629 struct btrfs_trans_handle *trans;
630 unsigned long nr;
631 u64 num_bytes;
632 u64 bytes_used;
633 u64 max_useless;
634 int ret = 0;
635 int err;
636
637 while(!list_empty(list)) {
638 struct btrfs_root *root;
639
640 dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
641 list_del_init(&dirty->list);
642
643 num_bytes = btrfs_root_used(&dirty->root->root_item);
644 root = dirty->latest_root;
645 atomic_inc(&root->fs_info->throttles);
646
647 while(1) {
648 trans = btrfs_start_transaction(tree_root, 1);
649 mutex_lock(&root->fs_info->drop_mutex);
650 ret = btrfs_drop_snapshot(trans, dirty->root);
651 if (ret != -EAGAIN) {
652 break;
653 }
654 mutex_unlock(&root->fs_info->drop_mutex);
655
656 err = btrfs_update_root(trans,
657 tree_root,
658 &dirty->root->root_key,
659 &dirty->root->root_item);
660 if (err)
661 ret = err;
662 nr = trans->blocks_used;
663 ret = btrfs_end_transaction(trans, tree_root);
664 BUG_ON(ret);
665
666 btrfs_btree_balance_dirty(tree_root, nr);
667 cond_resched();
668 }
669 BUG_ON(ret);
670 atomic_dec(&root->fs_info->throttles);
671 wake_up(&root->fs_info->transaction_throttle);
672
673 mutex_lock(&root->fs_info->alloc_mutex);
674 num_bytes -= btrfs_root_used(&dirty->root->root_item);
675 bytes_used = btrfs_root_used(&root->root_item);
676 if (num_bytes) {
677 btrfs_record_root_in_trans(root);
678 btrfs_set_root_used(&root->root_item,
679 bytes_used - num_bytes);
680 }
681 mutex_unlock(&root->fs_info->alloc_mutex);
682
683 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
684 if (ret) {
685 BUG();
686 break;
687 }
688 mutex_unlock(&root->fs_info->drop_mutex);
689
690 spin_lock(&root->list_lock);
691 list_del_init(&dirty->root->dead_list);
692 if (!list_empty(&root->dead_list)) {
693 struct btrfs_root *oldest;
694 oldest = list_entry(root->dead_list.prev,
695 struct btrfs_root, dead_list);
696 max_useless = oldest->root_key.offset - 1;
697 } else {
698 max_useless = root->root_key.offset - 1;
699 }
700 spin_unlock(&root->list_lock);
701
702 nr = trans->blocks_used;
703 ret = btrfs_end_transaction(trans, tree_root);
704 BUG_ON(ret);
705
706 ret = btrfs_remove_leaf_refs(root, max_useless, 0);
707 BUG_ON(ret);
708
709 free_extent_buffer(dirty->root->node);
710 kfree(dirty->root);
711 kfree(dirty);
712
713 btrfs_btree_balance_dirty(tree_root, nr);
714 cond_resched();
715 }
716 return ret;
717}
718
719/*
720 * new snapshots need to be created at a very specific time in the
721 * transaction commit. This does the actual creation
722 */
723static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
724 struct btrfs_fs_info *fs_info,
725 struct btrfs_pending_snapshot *pending)
726{
727 struct btrfs_key key;
728 struct btrfs_root_item *new_root_item;
729 struct btrfs_root *tree_root = fs_info->tree_root;
730 struct btrfs_root *root = pending->root;
731 struct extent_buffer *tmp;
732 struct extent_buffer *old;
733 int ret;
734 int namelen;
735 u64 objectid;
736
737 new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
738 if (!new_root_item) {
739 ret = -ENOMEM;
740 goto fail;
741 }
742 ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
743 if (ret)
744 goto fail;
745
746 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
747
748 key.objectid = objectid;
749 key.offset = trans->transid;
750 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
751
752 old = btrfs_lock_root_node(root);
753 btrfs_cow_block(trans, root, old, NULL, 0, &old, 0);
754
755 btrfs_copy_root(trans, root, old, &tmp, objectid);
756 btrfs_tree_unlock(old);
757 free_extent_buffer(old);
758
759 btrfs_set_root_bytenr(new_root_item, tmp->start);
760 btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
761 ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
762 new_root_item);
763 btrfs_tree_unlock(tmp);
764 free_extent_buffer(tmp);
765 if (ret)
766 goto fail;
767
768 /*
769 * insert the directory item
770 */
771 key.offset = (u64)-1;
772 namelen = strlen(pending->name);
773 ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
774 pending->name, namelen,
775 root->fs_info->sb->s_root->d_inode->i_ino,
776 &key, BTRFS_FT_DIR, 0);
777
778 if (ret)
779 goto fail;
780
781 ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
782 pending->name, strlen(pending->name), objectid,
783 root->fs_info->sb->s_root->d_inode->i_ino, 0);
784
785 /* Invalidate existing dcache entry for new snapshot. */
786 btrfs_invalidate_dcache_root(root, pending->name, namelen);
787
788fail:
789 kfree(new_root_item);
790 return ret;
791}
792
793/*
794 * create all the snapshots we've scheduled for creation
795 */
796static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
797 struct btrfs_fs_info *fs_info)
798{
799 struct btrfs_pending_snapshot *pending;
800 struct list_head *head = &trans->transaction->pending_snapshots;
801 int ret;
802
803 while(!list_empty(head)) {
804 pending = list_entry(head->next,
805 struct btrfs_pending_snapshot, list);
806 ret = create_pending_snapshot(trans, fs_info, pending);
807 BUG_ON(ret);
808 list_del(&pending->list);
809 kfree(pending->name);
810 kfree(pending);
811 }
812 return 0;
813}
814
815int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
816 struct btrfs_root *root)
817{
818 unsigned long joined = 0;
819 unsigned long timeout = 1;
820 struct btrfs_transaction *cur_trans;
821 struct btrfs_transaction *prev_trans = NULL;
822 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
823 struct list_head dirty_fs_roots;
824 struct extent_io_tree *pinned_copy;
825 DEFINE_WAIT(wait);
826 int ret;
827
828 INIT_LIST_HEAD(&dirty_fs_roots);
829 mutex_lock(&root->fs_info->trans_mutex);
830 if (trans->transaction->in_commit) {
831 cur_trans = trans->transaction;
832 trans->transaction->use_count++;
833 mutex_unlock(&root->fs_info->trans_mutex);
834 btrfs_end_transaction(trans, root);
835
836 ret = wait_for_commit(root, cur_trans);
837 BUG_ON(ret);
838
839 mutex_lock(&root->fs_info->trans_mutex);
840 put_transaction(cur_trans);
841 mutex_unlock(&root->fs_info->trans_mutex);
842
843 return 0;
844 }
845
846 pinned_copy = kmalloc(sizeof(*pinned_copy), GFP_NOFS);
847 if (!pinned_copy)
848 return -ENOMEM;
849
850 extent_io_tree_init(pinned_copy,
851 root->fs_info->btree_inode->i_mapping, GFP_NOFS);
852
853 trans->transaction->in_commit = 1;
854 trans->transaction->blocked = 1;
855 cur_trans = trans->transaction;
856 if (cur_trans->list.prev != &root->fs_info->trans_list) {
857 prev_trans = list_entry(cur_trans->list.prev,
858 struct btrfs_transaction, list);
859 if (!prev_trans->commit_done) {
860 prev_trans->use_count++;
861 mutex_unlock(&root->fs_info->trans_mutex);
862
863 wait_for_commit(root, prev_trans);
864
865 mutex_lock(&root->fs_info->trans_mutex);
866 put_transaction(prev_trans);
867 }
868 }
869
870 do {
871 int snap_pending = 0;
872 joined = cur_trans->num_joined;
873 if (!list_empty(&trans->transaction->pending_snapshots))
874 snap_pending = 1;
875
876 WARN_ON(cur_trans != trans->transaction);
877 prepare_to_wait(&cur_trans->writer_wait, &wait,
878 TASK_UNINTERRUPTIBLE);
879
880 if (cur_trans->num_writers > 1)
881 timeout = MAX_SCHEDULE_TIMEOUT;
882 else
883 timeout = 1;
884
885 mutex_unlock(&root->fs_info->trans_mutex);
886
887 if (snap_pending) {
888 ret = btrfs_wait_ordered_extents(root, 1);
889 BUG_ON(ret);
890 }
891
892 schedule_timeout(timeout);
893
894 mutex_lock(&root->fs_info->trans_mutex);
895 finish_wait(&cur_trans->writer_wait, &wait);
896 } while (cur_trans->num_writers > 1 ||
897 (cur_trans->num_joined != joined));
898
899 ret = create_pending_snapshots(trans, root->fs_info);
900 BUG_ON(ret);
901
902 WARN_ON(cur_trans != trans->transaction);
903
904 /* btrfs_commit_tree_roots is responsible for getting the
905 * various roots consistent with each other. Every pointer
906 * in the tree of tree roots has to point to the most up to date
907 * root for every subvolume and other tree. So, we have to keep
908 * the tree logging code from jumping in and changing any
909 * of the trees.
910 *
911 * At this point in the commit, there can't be any tree-log
912 * writers, but a little lower down we drop the trans mutex
913 * and let new people in. By holding the tree_log_mutex
914 * from now until after the super is written, we avoid races
915 * with the tree-log code.
916 */
917 mutex_lock(&root->fs_info->tree_log_mutex);
918 /*
919 * keep tree reloc code from adding new reloc trees
920 */
921 mutex_lock(&root->fs_info->tree_reloc_mutex);
922
923
924 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
925 &dirty_fs_roots);
926 BUG_ON(ret);
927
928 /* add_dirty_roots gets rid of all the tree log roots, it is now
929 * safe to free the root of tree log roots
930 */
931 btrfs_free_log_root_tree(trans, root->fs_info);
932
933 btrfs_free_reloc_mappings(root);
934
935 ret = btrfs_commit_tree_roots(trans, root);
936 BUG_ON(ret);
937
938 cur_trans = root->fs_info->running_transaction;
939 spin_lock(&root->fs_info->new_trans_lock);
940 root->fs_info->running_transaction = NULL;
941 spin_unlock(&root->fs_info->new_trans_lock);
942 btrfs_set_super_generation(&root->fs_info->super_copy,
943 cur_trans->transid);
944 btrfs_set_super_root(&root->fs_info->super_copy,
945 root->fs_info->tree_root->node->start);
946 btrfs_set_super_root_level(&root->fs_info->super_copy,
947 btrfs_header_level(root->fs_info->tree_root->node));
948
949 btrfs_set_super_chunk_root(&root->fs_info->super_copy,
950 chunk_root->node->start);
951 btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
952 btrfs_header_level(chunk_root->node));
953
954 if (!root->fs_info->log_root_recovering) {
955 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
956 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
957 }
958
959 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
960 sizeof(root->fs_info->super_copy));
961
962 btrfs_copy_pinned(root, pinned_copy);
963
964 trans->transaction->blocked = 0;
965 wake_up(&root->fs_info->transaction_throttle);
966 wake_up(&root->fs_info->transaction_wait);
967
968 mutex_unlock(&root->fs_info->trans_mutex);
969 ret = btrfs_write_and_wait_transaction(trans, root);
970 BUG_ON(ret);
971 write_ctree_super(trans, root);
972
973 /*
974 * the super is written, we can safely allow the tree-loggers
975 * to go about their business
976 */
977 mutex_unlock(&root->fs_info->tree_log_mutex);
978
979 btrfs_finish_extent_commit(trans, root, pinned_copy);
980 kfree(pinned_copy);
981
982 btrfs_drop_dead_reloc_roots(root);
983 mutex_unlock(&root->fs_info->tree_reloc_mutex);
984
985 mutex_lock(&root->fs_info->trans_mutex);
986
987 cur_trans->commit_done = 1;
988 root->fs_info->last_trans_committed = cur_trans->transid;
989 wake_up(&cur_trans->commit_wait);
990 put_transaction(cur_trans);
991 put_transaction(cur_trans);
992
993 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
994 if (root->fs_info->closing)
995 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
996
997 mutex_unlock(&root->fs_info->trans_mutex);
998 kmem_cache_free(btrfs_trans_handle_cachep, trans);
999
1000 if (root->fs_info->closing) {
1001 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
1002 }
1003 return ret;
1004}
1005
1006/*
1007 * interface function to delete all the snapshots we have scheduled for deletion
1008 */
1009int btrfs_clean_old_snapshots(struct btrfs_root *root)
1010{
1011 struct list_head dirty_roots;
1012 INIT_LIST_HEAD(&dirty_roots);
1013again:
1014 mutex_lock(&root->fs_info->trans_mutex);
1015 list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
1016 mutex_unlock(&root->fs_info->trans_mutex);
1017
1018 if (!list_empty(&dirty_roots)) {
1019 drop_dirty_roots(root, &dirty_roots);
1020 goto again;
1021 }
1022 return 0;
1023}
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-01 08:24:29 -0500