1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83#ifndef __KERNEL__
84#include "jfs_user.h"
85#else
86#include <linux/time.h>
87#include <linux/fs.h>
88#include <linux/jbd2.h>
89#include <linux/errno.h>
90#include <linux/slab.h>
91#include <linux/list.h>
92#include <linux/init.h>
93#include <linux/bio.h>
94#include <linux/log2.h>
95#include <linux/hash.h>
96#endif
97
98static struct kmem_cache *jbd2_revoke_record_cache;
99static struct kmem_cache *jbd2_revoke_table_cache;
100
101
102
103
104
105struct jbd2_revoke_record_s
106{
107 struct list_head hash;
108 tid_t sequence;
109 unsigned long long blocknr;
110};
111
112
113
114struct jbd2_revoke_table_s
115{
116
117
118 int hash_size;
119 int hash_shift;
120 struct list_head *hash_table;
121};
122
123
124#ifdef __KERNEL__
125static void write_one_revoke_record(journal_t *, transaction_t *,
126 struct list_head *,
127 struct buffer_head **, int *,
128 struct jbd2_revoke_record_s *, int);
129static void flush_descriptor(journal_t *, struct buffer_head *, int, int);
130#endif
131
132
133
134static inline int hash(journal_t *journal, unsigned long long block)
135{
136 return hash_64(block, journal->j_revoke->hash_shift);
137}
138
139static int insert_revoke_hash(journal_t *journal, unsigned long long blocknr,
140 tid_t seq)
141{
142 struct list_head *hash_list;
143 struct jbd2_revoke_record_s *record;
144 gfp_t gfp_mask = GFP_NOFS;
145
146 if (journal_oom_retry)
147 gfp_mask |= __GFP_NOFAIL;
148 record = kmem_cache_alloc(jbd2_revoke_record_cache, gfp_mask);
149 if (!record)
150 return -ENOMEM;
151
152 record->sequence = seq;
153 record->blocknr = blocknr;
154 hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)];
155 spin_lock(&journal->j_revoke_lock);
156 list_add(&record->hash, hash_list);
157 spin_unlock(&journal->j_revoke_lock);
158 return 0;
159}
160
161
162
163static struct jbd2_revoke_record_s *find_revoke_record(journal_t *journal,
164 unsigned long long blocknr)
165{
166 struct list_head *hash_list;
167 struct jbd2_revoke_record_s *record;
168
169 hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)];
170
171 spin_lock(&journal->j_revoke_lock);
172 record = (struct jbd2_revoke_record_s *) hash_list->next;
173 while (&(record->hash) != hash_list) {
174 if (record->blocknr == blocknr) {
175 spin_unlock(&journal->j_revoke_lock);
176 return record;
177 }
178 record = (struct jbd2_revoke_record_s *) record->hash.next;
179 }
180 spin_unlock(&journal->j_revoke_lock);
181 return NULL;
182}
183
184void jbd2_journal_destroy_revoke_caches(void)
185{
186 if (jbd2_revoke_record_cache) {
187 kmem_cache_destroy(jbd2_revoke_record_cache);
188 jbd2_revoke_record_cache = NULL;
189 }
190 if (jbd2_revoke_table_cache) {
191 kmem_cache_destroy(jbd2_revoke_table_cache);
192 jbd2_revoke_table_cache = NULL;
193 }
194}
195
196int __init jbd2_journal_init_revoke_caches(void)
197{
198 J_ASSERT(!jbd2_revoke_record_cache);
199 J_ASSERT(!jbd2_revoke_table_cache);
200
201 jbd2_revoke_record_cache = KMEM_CACHE(jbd2_revoke_record_s,
202 SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY);
203 if (!jbd2_revoke_record_cache)
204 goto record_cache_failure;
205
206 jbd2_revoke_table_cache = KMEM_CACHE(jbd2_revoke_table_s,
207 SLAB_TEMPORARY);
208 if (!jbd2_revoke_table_cache)
209 goto table_cache_failure;
210 return 0;
211table_cache_failure:
212 jbd2_journal_destroy_revoke_caches();
213record_cache_failure:
214 return -ENOMEM;
215}
216
217static struct jbd2_revoke_table_s *jbd2_journal_init_revoke_table(int hash_size)
218{
219 int shift = 0;
220 int tmp = hash_size;
221 struct jbd2_revoke_table_s *table;
222
223 table = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL);
224 if (!table)
225 goto out;
226
227 while((tmp >>= 1UL) != 0UL)
228 shift++;
229
230 table->hash_size = hash_size;
231 table->hash_shift = shift;
232 table->hash_table =
233 kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL);
234 if (!table->hash_table) {
235 kmem_cache_free(jbd2_revoke_table_cache, table);
236 table = NULL;
237 goto out;
238 }
239
240 for (tmp = 0; tmp < hash_size; tmp++)
241 INIT_LIST_HEAD(&table->hash_table[tmp]);
242
243out:
244 return table;
245}
246
247static void jbd2_journal_destroy_revoke_table(struct jbd2_revoke_table_s *table)
248{
249 int i;
250 struct list_head *hash_list;
251
252 for (i = 0; i < table->hash_size; i++) {
253 hash_list = &table->hash_table[i];
254 J_ASSERT(list_empty(hash_list));
255 }
256
257 kfree(table->hash_table);
258 kmem_cache_free(jbd2_revoke_table_cache, table);
259}
260
261
262int jbd2_journal_init_revoke(journal_t *journal, int hash_size)
263{
264 J_ASSERT(journal->j_revoke_table[0] == NULL);
265 J_ASSERT(is_power_of_2(hash_size));
266
267 journal->j_revoke_table[0] = jbd2_journal_init_revoke_table(hash_size);
268 if (!journal->j_revoke_table[0])
269 goto fail0;
270
271 journal->j_revoke_table[1] = jbd2_journal_init_revoke_table(hash_size);
272 if (!journal->j_revoke_table[1])
273 goto fail1;
274
275 journal->j_revoke = journal->j_revoke_table[1];
276
277 spin_lock_init(&journal->j_revoke_lock);
278
279 return 0;
280
281fail1:
282 jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]);
283fail0:
284 return -ENOMEM;
285}
286
287
288void jbd2_journal_destroy_revoke(journal_t *journal)
289{
290 journal->j_revoke = NULL;
291 if (journal->j_revoke_table[0])
292 jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]);
293 if (journal->j_revoke_table[1])
294 jbd2_journal_destroy_revoke_table(journal->j_revoke_table[1]);
295}
296
297
298#ifdef __KERNEL__
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr,
325 struct buffer_head *bh_in)
326{
327 struct buffer_head *bh = NULL;
328 journal_t *journal;
329 struct block_device *bdev;
330 int err;
331
332 might_sleep();
333 if (bh_in)
334 BUFFER_TRACE(bh_in, "enter");
335
336 journal = handle->h_transaction->t_journal;
337 if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){
338 J_ASSERT (!"Cannot set revoke feature!");
339 return -EINVAL;
340 }
341
342 bdev = journal->j_fs_dev;
343 bh = bh_in;
344
345 if (!bh) {
346 bh = __find_get_block(bdev, blocknr, journal->j_blocksize);
347 if (bh)
348 BUFFER_TRACE(bh, "found on hash");
349 }
350#ifdef JBD2_EXPENSIVE_CHECKING
351 else {
352 struct buffer_head *bh2;
353
354
355
356 bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize);
357 if (bh2) {
358
359 if (bh2 != bh && buffer_revokevalid(bh2))
360
361
362
363
364
365
366 J_ASSERT_BH(bh2, buffer_revoked(bh2));
367 put_bh(bh2);
368 }
369 }
370#endif
371
372
373
374
375 if (bh) {
376 if (!J_EXPECT_BH(bh, !buffer_revoked(bh),
377 "inconsistent data on disk")) {
378 if (!bh_in)
379 brelse(bh);
380 return -EIO;
381 }
382 set_buffer_revoked(bh);
383 set_buffer_revokevalid(bh);
384 if (bh_in) {
385 BUFFER_TRACE(bh_in, "call jbd2_journal_forget");
386 jbd2_journal_forget(handle, bh_in);
387 } else {
388 BUFFER_TRACE(bh, "call brelse");
389 __brelse(bh);
390 }
391 }
392
393 jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in);
394 err = insert_revoke_hash(journal, blocknr,
395 handle->h_transaction->t_tid);
396 BUFFER_TRACE(bh_in, "exit");
397 return err;
398}
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh)
416{
417 struct jbd2_revoke_record_s *record;
418 journal_t *journal = handle->h_transaction->t_journal;
419 int need_cancel;
420 int did_revoke = 0;
421 struct buffer_head *bh = jh2bh(jh);
422
423 jbd_debug(4, "journal_head %p, cancelling revoke\n", jh);
424
425
426
427
428
429 if (test_set_buffer_revokevalid(bh)) {
430 need_cancel = test_clear_buffer_revoked(bh);
431 } else {
432 need_cancel = 1;
433 clear_buffer_revoked(bh);
434 }
435
436 if (need_cancel) {
437 record = find_revoke_record(journal, bh->b_blocknr);
438 if (record) {
439 jbd_debug(4, "cancelled existing revoke on "
440 "blocknr %llu\n", (unsigned long long)bh->b_blocknr);
441 spin_lock(&journal->j_revoke_lock);
442 list_del(&record->hash);
443 spin_unlock(&journal->j_revoke_lock);
444 kmem_cache_free(jbd2_revoke_record_cache, record);
445 did_revoke = 1;
446 }
447 }
448
449#ifdef JBD2_EXPENSIVE_CHECKING
450
451 record = find_revoke_record(journal, bh->b_blocknr);
452 J_ASSERT_JH(jh, record == NULL);
453#endif
454
455
456
457
458
459 if (need_cancel) {
460 struct buffer_head *bh2;
461 bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size);
462 if (bh2) {
463 if (bh2 != bh)
464 clear_buffer_revoked(bh2);
465 __brelse(bh2);
466 }
467 }
468 return did_revoke;
469}
470
471
472
473
474
475
476void jbd2_clear_buffer_revoked_flags(journal_t *journal)
477{
478 struct jbd2_revoke_table_s *revoke = journal->j_revoke;
479 int i = 0;
480
481 for (i = 0; i < revoke->hash_size; i++) {
482 struct list_head *hash_list;
483 struct list_head *list_entry;
484 hash_list = &revoke->hash_table[i];
485
486 list_for_each(list_entry, hash_list) {
487 struct jbd2_revoke_record_s *record;
488 struct buffer_head *bh;
489 record = (struct jbd2_revoke_record_s *)list_entry;
490 bh = __find_get_block(journal->j_fs_dev,
491 record->blocknr,
492 journal->j_blocksize);
493 if (bh) {
494 clear_buffer_revoked(bh);
495 __brelse(bh);
496 }
497 }
498 }
499}
500
501
502
503
504
505void jbd2_journal_switch_revoke_table(journal_t *journal)
506{
507 int i;
508
509 if (journal->j_revoke == journal->j_revoke_table[0])
510 journal->j_revoke = journal->j_revoke_table[1];
511 else
512 journal->j_revoke = journal->j_revoke_table[0];
513
514 for (i = 0; i < journal->j_revoke->hash_size; i++)
515 INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]);
516}
517
518
519
520
521
522void jbd2_journal_write_revoke_records(journal_t *journal,
523 transaction_t *transaction,
524 struct list_head *log_bufs,
525 int write_op)
526{
527 struct buffer_head *descriptor;
528 struct jbd2_revoke_record_s *record;
529 struct jbd2_revoke_table_s *revoke;
530 struct list_head *hash_list;
531 int i, offset, count;
532
533 descriptor = NULL;
534 offset = 0;
535 count = 0;
536
537
538 revoke = journal->j_revoke == journal->j_revoke_table[0] ?
539 journal->j_revoke_table[1] : journal->j_revoke_table[0];
540
541 for (i = 0; i < revoke->hash_size; i++) {
542 hash_list = &revoke->hash_table[i];
543
544 while (!list_empty(hash_list)) {
545 record = (struct jbd2_revoke_record_s *)
546 hash_list->next;
547 write_one_revoke_record(journal, transaction, log_bufs,
548 &descriptor, &offset,
549 record, write_op);
550 count++;
551 list_del(&record->hash);
552 kmem_cache_free(jbd2_revoke_record_cache, record);
553 }
554 }
555 if (descriptor)
556 flush_descriptor(journal, descriptor, offset, write_op);
557 jbd_debug(1, "Wrote %d revoke records\n", count);
558}
559
560
561
562
563
564
565static void write_one_revoke_record(journal_t *journal,
566 transaction_t *transaction,
567 struct list_head *log_bufs,
568 struct buffer_head **descriptorp,
569 int *offsetp,
570 struct jbd2_revoke_record_s *record,
571 int write_op)
572{
573 int csum_size = 0;
574 struct buffer_head *descriptor;
575 int sz, offset;
576 journal_header_t *header;
577
578
579
580
581
582 if (is_journal_aborted(journal))
583 return;
584
585 descriptor = *descriptorp;
586 offset = *offsetp;
587
588
589 if (jbd2_journal_has_csum_v2or3(journal))
590 csum_size = sizeof(struct jbd2_journal_revoke_tail);
591
592 if (jbd2_has_feature_64bit(journal))
593 sz = 8;
594 else
595 sz = 4;
596
597
598 if (descriptor) {
599 if (offset + sz > journal->j_blocksize - csum_size) {
600 flush_descriptor(journal, descriptor, offset, write_op);
601 descriptor = NULL;
602 }
603 }
604
605 if (!descriptor) {
606 descriptor = jbd2_journal_get_descriptor_buffer(journal);
607 if (!descriptor)
608 return;
609 header = (journal_header_t *)descriptor->b_data;
610 header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
611 header->h_blocktype = cpu_to_be32(JBD2_REVOKE_BLOCK);
612 header->h_sequence = cpu_to_be32(transaction->t_tid);
613
614
615 BUFFER_TRACE(descriptor, "file in log_bufs");
616 jbd2_file_log_bh(log_bufs, descriptor);
617
618 offset = sizeof(jbd2_journal_revoke_header_t);
619 *descriptorp = descriptor;
620 }
621
622 if (jbd2_has_feature_64bit(journal))
623 * ((__be64 *)(&descriptor->b_data[offset])) =
624 cpu_to_be64(record->blocknr);
625 else
626 * ((__be32 *)(&descriptor->b_data[offset])) =
627 cpu_to_be32(record->blocknr);
628 offset += sz;
629
630 *offsetp = offset;
631}
632
633static void jbd2_revoke_csum_set(journal_t *j, struct buffer_head *bh)
634{
635 struct jbd2_journal_revoke_tail *tail;
636 __u32 csum;
637
638 if (!jbd2_journal_has_csum_v2or3(j))
639 return;
640
641 tail = (struct jbd2_journal_revoke_tail *)(bh->b_data + j->j_blocksize -
642 sizeof(struct jbd2_journal_revoke_tail));
643 tail->r_checksum = 0;
644 csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
645 tail->r_checksum = cpu_to_be32(csum);
646}
647
648
649
650
651
652
653
654
655static void flush_descriptor(journal_t *journal,
656 struct buffer_head *descriptor,
657 int offset, int write_op)
658{
659 jbd2_journal_revoke_header_t *header;
660
661 if (is_journal_aborted(journal)) {
662 put_bh(descriptor);
663 return;
664 }
665
666 header = (jbd2_journal_revoke_header_t *)descriptor->b_data;
667 header->r_count = cpu_to_be32(offset);
668 jbd2_revoke_csum_set(journal, descriptor);
669
670 set_buffer_jwrite(descriptor);
671 BUFFER_TRACE(descriptor, "write");
672 set_buffer_dirty(descriptor);
673 write_dirty_buffer(descriptor, write_op);
674}
675#endif
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699int jbd2_journal_set_revoke(journal_t *journal,
700 unsigned long long blocknr,
701 tid_t sequence)
702{
703 struct jbd2_revoke_record_s *record;
704
705 record = find_revoke_record(journal, blocknr);
706 if (record) {
707
708
709 if (tid_gt(sequence, record->sequence))
710 record->sequence = sequence;
711 return 0;
712 }
713 return insert_revoke_hash(journal, blocknr, sequence);
714}
715
716
717
718
719
720
721
722
723int jbd2_journal_test_revoke(journal_t *journal,
724 unsigned long long blocknr,
725 tid_t sequence)
726{
727 struct jbd2_revoke_record_s *record;
728
729 record = find_revoke_record(journal, blocknr);
730 if (!record)
731 return 0;
732 if (tid_gt(sequence, record->sequence))
733 return 0;
734 return 1;
735}
736
737
738
739
740
741
742void jbd2_journal_clear_revoke(journal_t *journal)
743{
744 int i;
745 struct list_head *hash_list;
746 struct jbd2_revoke_record_s *record;
747 struct jbd2_revoke_table_s *revoke;
748
749 revoke = journal->j_revoke;
750
751 for (i = 0; i < revoke->hash_size; i++) {
752 hash_list = &revoke->hash_table[i];
753 while (!list_empty(hash_list)) {
754 record = (struct jbd2_revoke_record_s*) hash_list->next;
755 list_del(&record->hash);
756 kmem_cache_free(jbd2_revoke_record_cache, record);
757 }
758 }
759}
760