[PATCH] dm mirror log: sector size fix
[pandora-kernel.git] / drivers / md / dm-log.c
1 /*
2  * Copyright (C) 2003 Sistina Software
3  *
4  * This file is released under the LGPL.
5  */
6
7 #include <linux/init.h>
8 #include <linux/slab.h>
9 #include <linux/module.h>
10 #include <linux/vmalloc.h>
11
12 #include "dm-log.h"
13 #include "dm-io.h"
14
15 static LIST_HEAD(_log_types);
16 static DEFINE_SPINLOCK(_lock);
17
18 int dm_register_dirty_log_type(struct dirty_log_type *type)
19 {
20         spin_lock(&_lock);
21         type->use_count = 0;
22         list_add(&type->list, &_log_types);
23         spin_unlock(&_lock);
24
25         return 0;
26 }
27
28 int dm_unregister_dirty_log_type(struct dirty_log_type *type)
29 {
30         spin_lock(&_lock);
31
32         if (type->use_count)
33                 DMWARN("Attempt to unregister a log type that is still in use");
34         else
35                 list_del(&type->list);
36
37         spin_unlock(&_lock);
38
39         return 0;
40 }
41
42 static struct dirty_log_type *get_type(const char *type_name)
43 {
44         struct dirty_log_type *type;
45
46         spin_lock(&_lock);
47         list_for_each_entry (type, &_log_types, list)
48                 if (!strcmp(type_name, type->name)) {
49                         if (!type->use_count && !try_module_get(type->module)){
50                                 spin_unlock(&_lock);
51                                 return NULL;
52                         }
53                         type->use_count++;
54                         spin_unlock(&_lock);
55                         return type;
56                 }
57
58         spin_unlock(&_lock);
59         return NULL;
60 }
61
62 static void put_type(struct dirty_log_type *type)
63 {
64         spin_lock(&_lock);
65         if (!--type->use_count)
66                 module_put(type->module);
67         spin_unlock(&_lock);
68 }
69
70 struct dirty_log *dm_create_dirty_log(const char *type_name, struct dm_target *ti,
71                                       unsigned int argc, char **argv)
72 {
73         struct dirty_log_type *type;
74         struct dirty_log *log;
75
76         log = kmalloc(sizeof(*log), GFP_KERNEL);
77         if (!log)
78                 return NULL;
79
80         type = get_type(type_name);
81         if (!type) {
82                 kfree(log);
83                 return NULL;
84         }
85
86         log->type = type;
87         if (type->ctr(log, ti, argc, argv)) {
88                 kfree(log);
89                 put_type(type);
90                 return NULL;
91         }
92
93         return log;
94 }
95
96 void dm_destroy_dirty_log(struct dirty_log *log)
97 {
98         log->type->dtr(log);
99         put_type(log->type);
100         kfree(log);
101 }
102
103 /*-----------------------------------------------------------------
104  * Persistent and core logs share a lot of their implementation.
105  * FIXME: need a reload method to be called from a resume
106  *---------------------------------------------------------------*/
107 /*
108  * Magic for persistent mirrors: "MiRr"
109  */
110 #define MIRROR_MAGIC 0x4D695272
111
112 /*
113  * The on-disk version of the metadata.
114  */
115 #define MIRROR_DISK_VERSION 2
116 #define LOG_OFFSET 2
117
118 struct log_header {
119         uint32_t magic;
120
121         /*
122          * Simple, incrementing version. no backward
123          * compatibility.
124          */
125         uint32_t version;
126         sector_t nr_regions;
127 };
128
129 struct log_c {
130         struct dm_target *ti;
131         int touched;
132         uint32_t region_size;
133         unsigned int region_count;
134         region_t sync_count;
135
136         unsigned bitset_uint32_count;
137         uint32_t *clean_bits;
138         uint32_t *sync_bits;
139         uint32_t *recovering_bits;      /* FIXME: this seems excessive */
140
141         int sync_search;
142
143         /* Resync flag */
144         enum sync {
145                 DEFAULTSYNC,    /* Synchronize if necessary */
146                 NOSYNC,         /* Devices known to be already in sync */
147                 FORCESYNC,      /* Force a sync to happen */
148         } sync;
149
150         /*
151          * Disk log fields
152          */
153         struct dm_dev *log_dev;
154         struct log_header header;
155
156         struct io_region header_location;
157         struct log_header *disk_header;
158 };
159
160 /*
161  * The touched member needs to be updated every time we access
162  * one of the bitsets.
163  */
164 static  inline int log_test_bit(uint32_t *bs, unsigned bit)
165 {
166         return ext2_test_bit(bit, (unsigned long *) bs) ? 1 : 0;
167 }
168
169 static inline void log_set_bit(struct log_c *l,
170                                uint32_t *bs, unsigned bit)
171 {
172         ext2_set_bit(bit, (unsigned long *) bs);
173         l->touched = 1;
174 }
175
176 static inline void log_clear_bit(struct log_c *l,
177                                  uint32_t *bs, unsigned bit)
178 {
179         ext2_clear_bit(bit, (unsigned long *) bs);
180         l->touched = 1;
181 }
182
183 /*----------------------------------------------------------------
184  * Header IO
185  *--------------------------------------------------------------*/
186 static void header_to_disk(struct log_header *core, struct log_header *disk)
187 {
188         disk->magic = cpu_to_le32(core->magic);
189         disk->version = cpu_to_le32(core->version);
190         disk->nr_regions = cpu_to_le64(core->nr_regions);
191 }
192
193 static void header_from_disk(struct log_header *core, struct log_header *disk)
194 {
195         core->magic = le32_to_cpu(disk->magic);
196         core->version = le32_to_cpu(disk->version);
197         core->nr_regions = le64_to_cpu(disk->nr_regions);
198 }
199
200 static int read_header(struct log_c *log)
201 {
202         int r;
203         unsigned long ebits;
204
205         r = dm_io_sync_vm(1, &log->header_location, READ,
206                           log->disk_header, &ebits);
207         if (r)
208                 return r;
209
210         header_from_disk(&log->header, log->disk_header);
211
212         /* New log required? */
213         if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
214                 log->header.magic = MIRROR_MAGIC;
215                 log->header.version = MIRROR_DISK_VERSION;
216                 log->header.nr_regions = 0;
217         }
218
219 #ifdef __LITTLE_ENDIAN
220         if (log->header.version == 1)
221                 log->header.version = 2;
222 #endif
223
224         if (log->header.version != MIRROR_DISK_VERSION) {
225                 DMWARN("incompatible disk log version");
226                 return -EINVAL;
227         }
228
229         return 0;
230 }
231
232 static inline int write_header(struct log_c *log)
233 {
234         unsigned long ebits;
235
236         header_to_disk(&log->header, log->disk_header);
237         return dm_io_sync_vm(1, &log->header_location, WRITE,
238                              log->disk_header, &ebits);
239 }
240
241 /*----------------------------------------------------------------
242  * core log constructor/destructor
243  *
244  * argv contains region_size followed optionally by [no]sync
245  *--------------------------------------------------------------*/
246 #define BYTE_SHIFT 3
247 static int core_ctr(struct dirty_log *log, struct dm_target *ti,
248                     unsigned int argc, char **argv)
249 {
250         enum sync sync = DEFAULTSYNC;
251
252         struct log_c *lc;
253         uint32_t region_size;
254         unsigned int region_count;
255         size_t bitset_size;
256
257         if (argc < 1 || argc > 2) {
258                 DMWARN("wrong number of arguments to mirror log");
259                 return -EINVAL;
260         }
261
262         if (argc > 1) {
263                 if (!strcmp(argv[1], "sync"))
264                         sync = FORCESYNC;
265                 else if (!strcmp(argv[1], "nosync"))
266                         sync = NOSYNC;
267                 else {
268                         DMWARN("unrecognised sync argument to mirror log: %s",
269                                argv[1]);
270                         return -EINVAL;
271                 }
272         }
273
274         if (sscanf(argv[0], "%u", &region_size) != 1) {
275                 DMWARN("invalid region size string");
276                 return -EINVAL;
277         }
278
279         region_count = dm_sector_div_up(ti->len, region_size);
280
281         lc = kmalloc(sizeof(*lc), GFP_KERNEL);
282         if (!lc) {
283                 DMWARN("couldn't allocate core log");
284                 return -ENOMEM;
285         }
286
287         lc->ti = ti;
288         lc->touched = 0;
289         lc->region_size = region_size;
290         lc->region_count = region_count;
291         lc->sync = sync;
292
293         /*
294          * Work out how many "unsigned long"s we need to hold the bitset.
295          */
296         bitset_size = dm_round_up(region_count,
297                                   sizeof(unsigned long) << BYTE_SHIFT);
298         bitset_size >>= BYTE_SHIFT;
299
300         lc->bitset_uint32_count = bitset_size / 4;
301         lc->clean_bits = vmalloc(bitset_size);
302         if (!lc->clean_bits) {
303                 DMWARN("couldn't allocate clean bitset");
304                 kfree(lc);
305                 return -ENOMEM;
306         }
307         memset(lc->clean_bits, -1, bitset_size);
308
309         lc->sync_bits = vmalloc(bitset_size);
310         if (!lc->sync_bits) {
311                 DMWARN("couldn't allocate sync bitset");
312                 vfree(lc->clean_bits);
313                 kfree(lc);
314                 return -ENOMEM;
315         }
316         memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
317         lc->sync_count = (sync == NOSYNC) ? region_count : 0;
318
319         lc->recovering_bits = vmalloc(bitset_size);
320         if (!lc->recovering_bits) {
321                 DMWARN("couldn't allocate sync bitset");
322                 vfree(lc->sync_bits);
323                 vfree(lc->clean_bits);
324                 kfree(lc);
325                 return -ENOMEM;
326         }
327         memset(lc->recovering_bits, 0, bitset_size);
328         lc->sync_search = 0;
329         log->context = lc;
330         return 0;
331 }
332
333 static void core_dtr(struct dirty_log *log)
334 {
335         struct log_c *lc = (struct log_c *) log->context;
336         vfree(lc->clean_bits);
337         vfree(lc->sync_bits);
338         vfree(lc->recovering_bits);
339         kfree(lc);
340 }
341
342 /*----------------------------------------------------------------
343  * disk log constructor/destructor
344  *
345  * argv contains log_device region_size followed optionally by [no]sync
346  *--------------------------------------------------------------*/
347 static int disk_ctr(struct dirty_log *log, struct dm_target *ti,
348                     unsigned int argc, char **argv)
349 {
350         int r;
351         size_t size, bitset_size;
352         struct log_c *lc;
353         struct dm_dev *dev;
354         uint32_t *clean_bits;
355
356         if (argc < 2 || argc > 3) {
357                 DMWARN("wrong number of arguments to disk mirror log");
358                 return -EINVAL;
359         }
360
361         r = dm_get_device(ti, argv[0], 0, 0 /* FIXME */,
362                           FMODE_READ | FMODE_WRITE, &dev);
363         if (r)
364                 return r;
365
366         r = core_ctr(log, ti, argc - 1, argv + 1);
367         if (r) {
368                 dm_put_device(ti, dev);
369                 return r;
370         }
371
372         lc = (struct log_c *) log->context;
373         lc->log_dev = dev;
374
375         /* setup the disk header fields */
376         lc->header_location.bdev = lc->log_dev->bdev;
377         lc->header_location.sector = 0;
378
379         /* Include both the header and the bitset in one buffer. */
380         bitset_size = lc->bitset_uint32_count * sizeof(uint32_t);
381         size = dm_round_up((LOG_OFFSET << SECTOR_SHIFT) + bitset_size,
382                            ti->limits.hardsect_size);
383         lc->header_location.count = size >> SECTOR_SHIFT;
384
385         lc->disk_header = vmalloc(size);
386         if (!lc->disk_header)
387                 goto bad;
388
389         /*
390          * Deallocate the clean_bits buffer that was allocated in core_ctr()
391          * and point it at the appropriate place in the disk_header buffer.
392          */
393         clean_bits = lc->clean_bits;
394         lc->clean_bits = (void *)lc->disk_header + (LOG_OFFSET << SECTOR_SHIFT);
395         memcpy(lc->clean_bits, clean_bits, bitset_size);
396         vfree(clean_bits);
397
398         return 0;
399
400  bad:
401         dm_put_device(ti, lc->log_dev);
402         core_dtr(log);
403         return -ENOMEM;
404 }
405
406 static void disk_dtr(struct dirty_log *log)
407 {
408         struct log_c *lc = (struct log_c *) log->context;
409         dm_put_device(lc->ti, lc->log_dev);
410         vfree(lc->disk_header);
411         lc->clean_bits = NULL;
412         core_dtr(log);
413 }
414
415 static int count_bits32(uint32_t *addr, unsigned size)
416 {
417         int count = 0, i;
418
419         for (i = 0; i < size; i++) {
420                 count += hweight32(*(addr+i));
421         }
422         return count;
423 }
424
425 static int disk_resume(struct dirty_log *log)
426 {
427         int r;
428         unsigned i;
429         struct log_c *lc = (struct log_c *) log->context;
430         size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
431
432         /* read the disk header */
433         r = read_header(lc);
434         if (r)
435                 return r;
436
437         /* set or clear any new bits */
438         if (lc->sync == NOSYNC)
439                 for (i = lc->header.nr_regions; i < lc->region_count; i++)
440                         /* FIXME: amazingly inefficient */
441                         log_set_bit(lc, lc->clean_bits, i);
442         else
443                 for (i = lc->header.nr_regions; i < lc->region_count; i++)
444                         /* FIXME: amazingly inefficient */
445                         log_clear_bit(lc, lc->clean_bits, i);
446
447         /* copy clean across to sync */
448         memcpy(lc->sync_bits, lc->clean_bits, size);
449         lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count);
450
451         /* set the correct number of regions in the header */
452         lc->header.nr_regions = lc->region_count;
453
454         /* write the new header */
455         return write_header(lc);
456 }
457
458 static uint32_t core_get_region_size(struct dirty_log *log)
459 {
460         struct log_c *lc = (struct log_c *) log->context;
461         return lc->region_size;
462 }
463
464 static int core_is_clean(struct dirty_log *log, region_t region)
465 {
466         struct log_c *lc = (struct log_c *) log->context;
467         return log_test_bit(lc->clean_bits, region);
468 }
469
470 static int core_in_sync(struct dirty_log *log, region_t region, int block)
471 {
472         struct log_c *lc = (struct log_c *) log->context;
473         return log_test_bit(lc->sync_bits, region);
474 }
475
476 static int core_flush(struct dirty_log *log)
477 {
478         /* no op */
479         return 0;
480 }
481
482 static int disk_flush(struct dirty_log *log)
483 {
484         int r;
485         struct log_c *lc = (struct log_c *) log->context;
486
487         /* only write if the log has changed */
488         if (!lc->touched)
489                 return 0;
490
491         r = write_header(lc);
492         if (!r)
493                 lc->touched = 0;
494
495         return r;
496 }
497
498 static void core_mark_region(struct dirty_log *log, region_t region)
499 {
500         struct log_c *lc = (struct log_c *) log->context;
501         log_clear_bit(lc, lc->clean_bits, region);
502 }
503
504 static void core_clear_region(struct dirty_log *log, region_t region)
505 {
506         struct log_c *lc = (struct log_c *) log->context;
507         log_set_bit(lc, lc->clean_bits, region);
508 }
509
510 static int core_get_resync_work(struct dirty_log *log, region_t *region)
511 {
512         struct log_c *lc = (struct log_c *) log->context;
513
514         if (lc->sync_search >= lc->region_count)
515                 return 0;
516
517         do {
518                 *region = ext2_find_next_zero_bit(
519                                              (unsigned long *) lc->sync_bits,
520                                              lc->region_count,
521                                              lc->sync_search);
522                 lc->sync_search = *region + 1;
523
524                 if (*region >= lc->region_count)
525                         return 0;
526
527         } while (log_test_bit(lc->recovering_bits, *region));
528
529         log_set_bit(lc, lc->recovering_bits, *region);
530         return 1;
531 }
532
533 static void core_complete_resync_work(struct dirty_log *log, region_t region,
534                                       int success)
535 {
536         struct log_c *lc = (struct log_c *) log->context;
537
538         log_clear_bit(lc, lc->recovering_bits, region);
539         if (success) {
540                 log_set_bit(lc, lc->sync_bits, region);
541                 lc->sync_count++;
542         }
543 }
544
545 static region_t core_get_sync_count(struct dirty_log *log)
546 {
547         struct log_c *lc = (struct log_c *) log->context;
548
549         return lc->sync_count;
550 }
551
552 #define DMEMIT_SYNC \
553         if (lc->sync != DEFAULTSYNC) \
554                 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
555
556 static int core_status(struct dirty_log *log, status_type_t status,
557                        char *result, unsigned int maxlen)
558 {
559         int sz = 0;
560         struct log_c *lc = log->context;
561
562         switch(status) {
563         case STATUSTYPE_INFO:
564                 break;
565
566         case STATUSTYPE_TABLE:
567                 DMEMIT("%s %u %u ", log->type->name,
568                        lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
569                 DMEMIT_SYNC;
570         }
571
572         return sz;
573 }
574
575 static int disk_status(struct dirty_log *log, status_type_t status,
576                        char *result, unsigned int maxlen)
577 {
578         int sz = 0;
579         char buffer[16];
580         struct log_c *lc = log->context;
581
582         switch(status) {
583         case STATUSTYPE_INFO:
584                 break;
585
586         case STATUSTYPE_TABLE:
587                 format_dev_t(buffer, lc->log_dev->bdev->bd_dev);
588                 DMEMIT("%s %u %s %u ", log->type->name,
589                        lc->sync == DEFAULTSYNC ? 2 : 3, buffer,
590                        lc->region_size);
591                 DMEMIT_SYNC;
592         }
593
594         return sz;
595 }
596
597 static struct dirty_log_type _core_type = {
598         .name = "core",
599         .module = THIS_MODULE,
600         .ctr = core_ctr,
601         .dtr = core_dtr,
602         .get_region_size = core_get_region_size,
603         .is_clean = core_is_clean,
604         .in_sync = core_in_sync,
605         .flush = core_flush,
606         .mark_region = core_mark_region,
607         .clear_region = core_clear_region,
608         .get_resync_work = core_get_resync_work,
609         .complete_resync_work = core_complete_resync_work,
610         .get_sync_count = core_get_sync_count,
611         .status = core_status,
612 };
613
614 static struct dirty_log_type _disk_type = {
615         .name = "disk",
616         .module = THIS_MODULE,
617         .ctr = disk_ctr,
618         .dtr = disk_dtr,
619         .suspend = disk_flush,
620         .resume = disk_resume,
621         .get_region_size = core_get_region_size,
622         .is_clean = core_is_clean,
623         .in_sync = core_in_sync,
624         .flush = disk_flush,
625         .mark_region = core_mark_region,
626         .clear_region = core_clear_region,
627         .get_resync_work = core_get_resync_work,
628         .complete_resync_work = core_complete_resync_work,
629         .get_sync_count = core_get_sync_count,
630         .status = disk_status,
631 };
632
633 int __init dm_dirty_log_init(void)
634 {
635         int r;
636
637         r = dm_register_dirty_log_type(&_core_type);
638         if (r)
639                 DMWARN("couldn't register core log");
640
641         r = dm_register_dirty_log_type(&_disk_type);
642         if (r) {
643                 DMWARN("couldn't register disk type");
644                 dm_unregister_dirty_log_type(&_core_type);
645         }
646
647         return r;
648 }
649
650 void dm_dirty_log_exit(void)
651 {
652         dm_unregister_dirty_log_type(&_disk_type);
653         dm_unregister_dirty_log_type(&_core_type);
654 }
655
656 EXPORT_SYMBOL(dm_register_dirty_log_type);
657 EXPORT_SYMBOL(dm_unregister_dirty_log_type);
658 EXPORT_SYMBOL(dm_create_dirty_log);
659 EXPORT_SYMBOL(dm_destroy_dirty_log);