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