]> git.karo-electronics.de Git - karo-tx-linux.git/blob - fs/btrfs/transaction.c
Btrfs: Do snapshot deletion in smaller chunks.
[karo-tx-linux.git] / fs / btrfs / transaction.c
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 "ctree.h"
22 #include "disk-io.h"
23 #include "transaction.h"
24
25 static int total_trans = 0;
26 extern struct kmem_cache *btrfs_trans_handle_cachep;
27 extern struct kmem_cache *btrfs_transaction_cachep;
28
29 static struct workqueue_struct *trans_wq;
30
31 #define BTRFS_ROOT_TRANS_TAG 0
32
33 static void put_transaction(struct btrfs_transaction *transaction)
34 {
35         WARN_ON(transaction->use_count == 0);
36         transaction->use_count--;
37         if (transaction->use_count == 0) {
38                 WARN_ON(total_trans == 0);
39                 total_trans--;
40                 list_del_init(&transaction->list);
41                 memset(transaction, 0, sizeof(*transaction));
42                 kmem_cache_free(btrfs_transaction_cachep, transaction);
43         }
44 }
45
46 static int join_transaction(struct btrfs_root *root)
47 {
48         struct btrfs_transaction *cur_trans;
49         cur_trans = root->fs_info->running_transaction;
50         if (!cur_trans) {
51                 cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
52                                              GFP_NOFS);
53                 total_trans++;
54                 BUG_ON(!cur_trans);
55                 root->fs_info->generation++;
56                 root->fs_info->running_transaction = cur_trans;
57                 cur_trans->num_writers = 0;
58                 cur_trans->transid = root->fs_info->generation;
59                 init_waitqueue_head(&cur_trans->writer_wait);
60                 init_waitqueue_head(&cur_trans->commit_wait);
61                 cur_trans->in_commit = 0;
62                 cur_trans->use_count = 1;
63                 cur_trans->commit_done = 0;
64                 cur_trans->start_time = get_seconds();
65                 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
66                 init_bit_radix(&cur_trans->dirty_pages);
67         }
68         cur_trans->num_writers++;
69         return 0;
70 }
71
72 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
73                                                    int num_blocks)
74 {
75         struct btrfs_trans_handle *h =
76                 kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
77         int ret;
78         u64 running_trans_id;
79
80         mutex_lock(&root->fs_info->trans_mutex);
81         ret = join_transaction(root);
82         BUG_ON(ret);
83         running_trans_id = root->fs_info->running_transaction->transid;
84
85         if (root != root->fs_info->tree_root && root->last_trans <
86             running_trans_id) {
87                 WARN_ON(root == root->fs_info->extent_root);
88                 WARN_ON(root->ref_cows != 1);
89                 if (root->root_item.refs != 0) {
90                         radix_tree_tag_set(&root->fs_info->fs_roots_radix,
91                                            (unsigned long)root->root_key.objectid,
92                                            BTRFS_ROOT_TRANS_TAG);
93                         root->commit_root = root->node;
94                         get_bh(root->node);
95                 } else {
96                         WARN_ON(1);
97                 }
98         }
99         root->last_trans = running_trans_id;
100         h->transid = running_trans_id;
101         h->transaction = root->fs_info->running_transaction;
102         h->blocks_reserved = num_blocks;
103         h->blocks_used = 0;
104         h->block_group = NULL;
105         root->fs_info->running_transaction->use_count++;
106         mutex_unlock(&root->fs_info->trans_mutex);
107         return h;
108 }
109
110 int btrfs_end_transaction(struct btrfs_trans_handle *trans,
111                           struct btrfs_root *root)
112 {
113         struct btrfs_transaction *cur_trans;
114
115         mutex_lock(&root->fs_info->trans_mutex);
116         cur_trans = root->fs_info->running_transaction;
117         WARN_ON(cur_trans != trans->transaction);
118         WARN_ON(cur_trans->num_writers < 1);
119         cur_trans->num_writers--;
120         if (waitqueue_active(&cur_trans->writer_wait))
121                 wake_up(&cur_trans->writer_wait);
122         put_transaction(cur_trans);
123         mutex_unlock(&root->fs_info->trans_mutex);
124         memset(trans, 0, sizeof(*trans));
125         kmem_cache_free(btrfs_trans_handle_cachep, trans);
126         return 0;
127 }
128
129
130 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
131                                      struct btrfs_root *root)
132 {
133         unsigned long gang[16];
134         int ret;
135         int i;
136         int err;
137         int werr = 0;
138         struct page *page;
139         struct radix_tree_root *dirty_pages;
140         struct inode *btree_inode = root->fs_info->btree_inode;
141
142         if (!trans || !trans->transaction) {
143                 return filemap_write_and_wait(btree_inode->i_mapping);
144         }
145         dirty_pages = &trans->transaction->dirty_pages;
146         while(1) {
147                 ret = find_first_radix_bit(dirty_pages, gang,
148                                            0, ARRAY_SIZE(gang));
149                 if (!ret)
150                         break;
151                 for (i = 0; i < ret; i++) {
152                         /* FIXME EIO */
153                         clear_radix_bit(dirty_pages, gang[i]);
154                         page = find_lock_page(btree_inode->i_mapping,
155                                               gang[i]);
156                         if (!page)
157                                 continue;
158                         err = write_one_page(page, 0);
159                         if (err)
160                                 werr = err;
161                         page_cache_release(page);
162                 }
163         }
164         err = filemap_fdatawait(btree_inode->i_mapping);
165         if (err)
166                 werr = err;
167         return werr;
168 }
169
170 int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
171                             struct btrfs_root *root)
172 {
173         int ret;
174         u64 old_extent_block;
175         struct btrfs_fs_info *fs_info = root->fs_info;
176         struct btrfs_root *tree_root = fs_info->tree_root;
177         struct btrfs_root *extent_root = fs_info->extent_root;
178
179         btrfs_write_dirty_block_groups(trans, extent_root);
180         while(1) {
181                 old_extent_block = btrfs_root_blocknr(&extent_root->root_item);
182                 if (old_extent_block == bh_blocknr(extent_root->node))
183                         break;
184                 btrfs_set_root_blocknr(&extent_root->root_item,
185                                        bh_blocknr(extent_root->node));
186                 ret = btrfs_update_root(trans, tree_root,
187                                         &extent_root->root_key,
188                                         &extent_root->root_item);
189                 BUG_ON(ret);
190                 btrfs_write_dirty_block_groups(trans, extent_root);
191         }
192         return 0;
193 }
194
195 static int wait_for_commit(struct btrfs_root *root,
196                            struct btrfs_transaction *commit)
197 {
198         DEFINE_WAIT(wait);
199         mutex_lock(&root->fs_info->trans_mutex);
200         while(!commit->commit_done) {
201                 prepare_to_wait(&commit->commit_wait, &wait,
202                                 TASK_UNINTERRUPTIBLE);
203                 if (commit->commit_done)
204                         break;
205                 mutex_unlock(&root->fs_info->trans_mutex);
206                 schedule();
207                 mutex_lock(&root->fs_info->trans_mutex);
208         }
209         mutex_unlock(&root->fs_info->trans_mutex);
210         finish_wait(&commit->commit_wait, &wait);
211         return 0;
212 }
213
214 struct dirty_root {
215         struct list_head list;
216         struct btrfs_root *root;
217 };
218
219 int btrfs_add_dead_root(struct btrfs_root *root, struct list_head *dead_list)
220 {
221         struct dirty_root *dirty;
222
223         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
224         if (!dirty)
225                 return -ENOMEM;
226         dirty->root = root;
227         list_add(&dirty->list, dead_list);
228         return 0;
229 }
230
231 static int add_dirty_roots(struct btrfs_trans_handle *trans,
232                            struct radix_tree_root *radix,
233                            struct list_head *list)
234 {
235         struct dirty_root *dirty;
236         struct btrfs_root *gang[8];
237         struct btrfs_root *root;
238         int i;
239         int ret;
240         int err = 0;
241         u32 refs;
242
243         while(1) {
244                 ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
245                                                  ARRAY_SIZE(gang),
246                                                  BTRFS_ROOT_TRANS_TAG);
247                 if (ret == 0)
248                         break;
249                 for (i = 0; i < ret; i++) {
250                         root = gang[i];
251                         radix_tree_tag_clear(radix,
252                                      (unsigned long)root->root_key.objectid,
253                                      BTRFS_ROOT_TRANS_TAG);
254                         if (root->commit_root == root->node) {
255                                 WARN_ON(bh_blocknr(root->node) !=
256                                         btrfs_root_blocknr(&root->root_item));
257                                 brelse(root->commit_root);
258                                 root->commit_root = NULL;
259                                 continue;
260                         }
261                         dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
262                         BUG_ON(!dirty);
263                         dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
264                         BUG_ON(!dirty->root);
265
266                         memset(&root->root_item.drop_progress, 0,
267                                sizeof(struct btrfs_disk_key));
268                         root->root_item.drop_level = 0;
269
270                         memcpy(dirty->root, root, sizeof(*root));
271                         dirty->root->node = root->commit_root;
272                         root->commit_root = NULL;
273
274                         root->root_key.offset = root->fs_info->generation;
275                         btrfs_set_root_blocknr(&root->root_item,
276                                                bh_blocknr(root->node));
277                         err = btrfs_insert_root(trans, root->fs_info->tree_root,
278                                                 &root->root_key,
279                                                 &root->root_item);
280                         if (err)
281                                 break;
282
283                         refs = btrfs_root_refs(&dirty->root->root_item);
284                         btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
285                         err = btrfs_update_root(trans, root->fs_info->tree_root,
286                                                 &dirty->root->root_key,
287                                                 &dirty->root->root_item);
288
289                         BUG_ON(err);
290                         if (refs == 1) {
291                                 list_add(&dirty->list, list);
292                         } else {
293                                 WARN_ON(1);
294                                 kfree(dirty->root);
295                                 kfree(dirty);
296                         }
297                 }
298         }
299         return err;
300 }
301
302 static int drop_dirty_roots(struct btrfs_root *tree_root,
303                             struct list_head *list)
304 {
305         struct dirty_root *dirty;
306         struct btrfs_trans_handle *trans;
307         int ret = 0;
308         int err;
309
310         while(!list_empty(list)) {
311                 mutex_lock(&tree_root->fs_info->fs_mutex);
312                 dirty = list_entry(list->next, struct dirty_root, list);
313                 list_del_init(&dirty->list);
314
315                 while(1) {
316                         trans = btrfs_start_transaction(tree_root, 1);
317                         ret = btrfs_drop_snapshot(trans, dirty->root);
318                         if (ret != -EAGAIN) {
319                                 break;
320                         }
321                         err = btrfs_update_root(trans,
322                                         tree_root,
323                                         &dirty->root->root_key,
324                                         &dirty->root->root_item);
325                         if (err)
326                                 ret = err;
327                         ret = btrfs_end_transaction(trans, tree_root);
328                         BUG_ON(ret);
329                 }
330                 BUG_ON(ret);
331                 ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
332                 if (ret)
333                         break;
334                 ret = btrfs_end_transaction(trans, tree_root);
335                 BUG_ON(ret);
336
337                 kfree(dirty->root);
338                 kfree(dirty);
339                 mutex_unlock(&tree_root->fs_info->fs_mutex);
340                 btrfs_btree_balance_dirty(tree_root);
341         }
342         return ret;
343 }
344
345 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
346                              struct btrfs_root *root)
347 {
348         int ret = 0;
349         struct btrfs_transaction *cur_trans;
350         struct btrfs_transaction *prev_trans = NULL;
351         struct list_head dirty_fs_roots;
352         struct radix_tree_root pinned_copy;
353         DEFINE_WAIT(wait);
354
355         init_bit_radix(&pinned_copy);
356         INIT_LIST_HEAD(&dirty_fs_roots);
357
358         mutex_lock(&root->fs_info->trans_mutex);
359         if (trans->transaction->in_commit) {
360                 cur_trans = trans->transaction;
361                 trans->transaction->use_count++;
362                 mutex_unlock(&root->fs_info->trans_mutex);
363                 btrfs_end_transaction(trans, root);
364
365                 mutex_unlock(&root->fs_info->fs_mutex);
366                 ret = wait_for_commit(root, cur_trans);
367                 BUG_ON(ret);
368                 put_transaction(cur_trans);
369                 mutex_lock(&root->fs_info->fs_mutex);
370                 return 0;
371         }
372         trans->transaction->in_commit = 1;
373         cur_trans = trans->transaction;
374         if (cur_trans->list.prev != &root->fs_info->trans_list) {
375                 prev_trans = list_entry(cur_trans->list.prev,
376                                         struct btrfs_transaction, list);
377                 if (!prev_trans->commit_done) {
378                         prev_trans->use_count++;
379                         mutex_unlock(&root->fs_info->fs_mutex);
380                         mutex_unlock(&root->fs_info->trans_mutex);
381
382                         wait_for_commit(root, prev_trans);
383                         put_transaction(prev_trans);
384
385                         mutex_lock(&root->fs_info->fs_mutex);
386                         mutex_lock(&root->fs_info->trans_mutex);
387                 }
388         }
389         while (trans->transaction->num_writers > 1) {
390                 WARN_ON(cur_trans != trans->transaction);
391                 prepare_to_wait(&trans->transaction->writer_wait, &wait,
392                                 TASK_UNINTERRUPTIBLE);
393                 if (trans->transaction->num_writers <= 1)
394                         break;
395                 mutex_unlock(&root->fs_info->fs_mutex);
396                 mutex_unlock(&root->fs_info->trans_mutex);
397                 schedule();
398                 mutex_lock(&root->fs_info->fs_mutex);
399                 mutex_lock(&root->fs_info->trans_mutex);
400                 finish_wait(&trans->transaction->writer_wait, &wait);
401         }
402         finish_wait(&trans->transaction->writer_wait, &wait);
403         WARN_ON(cur_trans != trans->transaction);
404         ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
405                               &dirty_fs_roots);
406         BUG_ON(ret);
407
408         ret = btrfs_commit_tree_roots(trans, root);
409         BUG_ON(ret);
410
411         cur_trans = root->fs_info->running_transaction;
412         root->fs_info->running_transaction = NULL;
413         btrfs_set_super_generation(&root->fs_info->super_copy,
414                                    cur_trans->transid);
415         btrfs_set_super_root(&root->fs_info->super_copy,
416                              bh_blocknr(root->fs_info->tree_root->node));
417         memcpy(root->fs_info->disk_super, &root->fs_info->super_copy,
418                sizeof(root->fs_info->super_copy));
419
420         btrfs_copy_pinned(root, &pinned_copy);
421
422         mutex_unlock(&root->fs_info->trans_mutex);
423         mutex_unlock(&root->fs_info->fs_mutex);
424         ret = btrfs_write_and_wait_transaction(trans, root);
425         BUG_ON(ret);
426         write_ctree_super(trans, root);
427         mutex_lock(&root->fs_info->fs_mutex);
428         btrfs_finish_extent_commit(trans, root, &pinned_copy);
429         mutex_lock(&root->fs_info->trans_mutex);
430         cur_trans->commit_done = 1;
431         wake_up(&cur_trans->commit_wait);
432         put_transaction(cur_trans);
433         put_transaction(cur_trans);
434         if (root->fs_info->closing)
435                 list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
436         else
437                 list_splice_init(&dirty_fs_roots, &root->fs_info->dead_roots);
438         mutex_unlock(&root->fs_info->trans_mutex);
439         kmem_cache_free(btrfs_trans_handle_cachep, trans);
440
441         if (root->fs_info->closing) {
442                 mutex_unlock(&root->fs_info->fs_mutex);
443                 drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
444                 mutex_lock(&root->fs_info->fs_mutex);
445         }
446         return ret;
447 }
448
449 void btrfs_transaction_cleaner(struct work_struct *work)
450 {
451         struct btrfs_fs_info *fs_info = container_of(work,
452                                                      struct btrfs_fs_info,
453                                                      trans_work.work);
454
455         struct btrfs_root *root = fs_info->tree_root;
456         struct btrfs_transaction *cur;
457         struct btrfs_trans_handle *trans;
458         struct list_head dirty_roots;
459         unsigned long now;
460         unsigned long delay = HZ * 30;
461         int ret;
462
463         INIT_LIST_HEAD(&dirty_roots);
464         mutex_lock(&root->fs_info->fs_mutex);
465         mutex_lock(&root->fs_info->trans_mutex);
466         cur = root->fs_info->running_transaction;
467         if (!cur) {
468                 mutex_unlock(&root->fs_info->trans_mutex);
469                 goto out;
470         }
471         now = get_seconds();
472         if (now < cur->start_time || now - cur->start_time < 30) {
473                 mutex_unlock(&root->fs_info->trans_mutex);
474                 delay = HZ * 5;
475                 goto out;
476         }
477         mutex_unlock(&root->fs_info->trans_mutex);
478         trans = btrfs_start_transaction(root, 1);
479         ret = btrfs_commit_transaction(trans, root);
480 out:
481         mutex_unlock(&root->fs_info->fs_mutex);
482
483         mutex_lock(&root->fs_info->trans_mutex);
484         list_splice_init(&root->fs_info->dead_roots, &dirty_roots);
485         mutex_unlock(&root->fs_info->trans_mutex);
486
487         if (!list_empty(&dirty_roots)) {
488                 drop_dirty_roots(root, &dirty_roots);
489         }
490         btrfs_transaction_queue_work(root, delay);
491 }
492
493 void btrfs_transaction_queue_work(struct btrfs_root *root, int delay)
494 {
495         queue_delayed_work(trans_wq, &root->fs_info->trans_work, delay);
496 }
497
498 void btrfs_transaction_flush_work(struct btrfs_root *root)
499 {
500         cancel_rearming_delayed_workqueue(trans_wq, &root->fs_info->trans_work);
501         flush_workqueue(trans_wq);
502 }
503
504 void __init btrfs_init_transaction_sys(void)
505 {
506         trans_wq = create_workqueue("btrfs");
507 }
508
509 void __exit btrfs_exit_transaction_sys(void)
510 {
511         destroy_workqueue(trans_wq);
512 }
513