]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/blob - drivers/md/dm-log.c
Merge branch 'smsc47b397-new-id' into release
[linux-2.6-omap-h63xx.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 #define DM_MSG_PREFIX "mirror log"
16
17 static LIST_HEAD(_log_types);
18 static DEFINE_SPINLOCK(_lock);
19
20 int dm_register_dirty_log_type(struct dirty_log_type *type)
21 {
22         spin_lock(&_lock);
23         type->use_count = 0;
24         list_add(&type->list, &_log_types);
25         spin_unlock(&_lock);
26
27         return 0;
28 }
29
30 int dm_unregister_dirty_log_type(struct dirty_log_type *type)
31 {
32         spin_lock(&_lock);
33
34         if (type->use_count)
35                 DMWARN("Attempt to unregister a log type that is still in use");
36         else
37                 list_del(&type->list);
38
39         spin_unlock(&_lock);
40
41         return 0;
42 }
43
44 static struct dirty_log_type *_get_type(const char *type_name)
45 {
46         struct dirty_log_type *type;
47
48         spin_lock(&_lock);
49         list_for_each_entry (type, &_log_types, list)
50                 if (!strcmp(type_name, type->name)) {
51                         if (!type->use_count && !try_module_get(type->module)){
52                                 spin_unlock(&_lock);
53                                 return NULL;
54                         }
55                         type->use_count++;
56                         spin_unlock(&_lock);
57                         return type;
58                 }
59
60         spin_unlock(&_lock);
61         return NULL;
62 }
63
64 /*
65  * get_type
66  * @type_name
67  *
68  * Attempt to retrieve the dirty_log_type by name.  If not already
69  * available, attempt to load the appropriate module.
70  *
71  * Log modules are named "dm-log-" followed by the 'type_name'.
72  * Modules may contain multiple types.
73  * This function will first try the module "dm-log-<type_name>",
74  * then truncate 'type_name' on the last '-' and try again.
75  *
76  * For example, if type_name was "clustered-disk", it would search
77  * 'dm-log-clustered-disk' then 'dm-log-clustered'.
78  *
79  * Returns: dirty_log_type* on success, NULL on failure
80  */
81 static struct dirty_log_type *get_type(const char *type_name)
82 {
83         char *p, *type_name_dup;
84         struct dirty_log_type *type;
85
86         type = _get_type(type_name);
87         if (type)
88                 return type;
89
90         type_name_dup = kstrdup(type_name, GFP_KERNEL);
91         if (!type_name_dup) {
92                 DMWARN("No memory left to attempt log module load for \"%s\"",
93                        type_name);
94                 return NULL;
95         }
96
97         while (request_module("dm-log-%s", type_name_dup) ||
98                !(type = _get_type(type_name))) {
99                 p = strrchr(type_name_dup, '-');
100                 if (!p)
101                         break;
102                 p[0] = '\0';
103         }
104
105         if (!type)
106                 DMWARN("Module for logging type \"%s\" not found.", type_name);
107
108         kfree(type_name_dup);
109
110         return type;
111 }
112
113 static void put_type(struct dirty_log_type *type)
114 {
115         spin_lock(&_lock);
116         if (!--type->use_count)
117                 module_put(type->module);
118         spin_unlock(&_lock);
119 }
120
121 struct dirty_log *dm_create_dirty_log(const char *type_name, struct dm_target *ti,
122                                       unsigned int argc, char **argv)
123 {
124         struct dirty_log_type *type;
125         struct dirty_log *log;
126
127         log = kmalloc(sizeof(*log), GFP_KERNEL);
128         if (!log)
129                 return NULL;
130
131         type = get_type(type_name);
132         if (!type) {
133                 kfree(log);
134                 return NULL;
135         }
136
137         log->type = type;
138         if (type->ctr(log, ti, argc, argv)) {
139                 kfree(log);
140                 put_type(type);
141                 return NULL;
142         }
143
144         return log;
145 }
146
147 void dm_destroy_dirty_log(struct dirty_log *log)
148 {
149         log->type->dtr(log);
150         put_type(log->type);
151         kfree(log);
152 }
153
154 /*-----------------------------------------------------------------
155  * Persistent and core logs share a lot of their implementation.
156  * FIXME: need a reload method to be called from a resume
157  *---------------------------------------------------------------*/
158 /*
159  * Magic for persistent mirrors: "MiRr"
160  */
161 #define MIRROR_MAGIC 0x4D695272
162
163 /*
164  * The on-disk version of the metadata.
165  */
166 #define MIRROR_DISK_VERSION 2
167 #define LOG_OFFSET 2
168
169 struct log_header {
170         uint32_t magic;
171
172         /*
173          * Simple, incrementing version. no backward
174          * compatibility.
175          */
176         uint32_t version;
177         sector_t nr_regions;
178 };
179
180 struct log_c {
181         struct dm_target *ti;
182         int touched;
183         uint32_t region_size;
184         unsigned int region_count;
185         region_t sync_count;
186
187         unsigned bitset_uint32_count;
188         uint32_t *clean_bits;
189         uint32_t *sync_bits;
190         uint32_t *recovering_bits;      /* FIXME: this seems excessive */
191
192         int sync_search;
193
194         /* Resync flag */
195         enum sync {
196                 DEFAULTSYNC,    /* Synchronize if necessary */
197                 NOSYNC,         /* Devices known to be already in sync */
198                 FORCESYNC,      /* Force a sync to happen */
199         } sync;
200
201         struct dm_io_request io_req;
202
203         /*
204          * Disk log fields
205          */
206         int log_dev_failed;
207         struct dm_dev *log_dev;
208         struct log_header header;
209
210         struct io_region header_location;
211         struct log_header *disk_header;
212 };
213
214 /*
215  * The touched member needs to be updated every time we access
216  * one of the bitsets.
217  */
218 static  inline int log_test_bit(uint32_t *bs, unsigned bit)
219 {
220         return ext2_test_bit(bit, (unsigned long *) bs) ? 1 : 0;
221 }
222
223 static inline void log_set_bit(struct log_c *l,
224                                uint32_t *bs, unsigned bit)
225 {
226         ext2_set_bit(bit, (unsigned long *) bs);
227         l->touched = 1;
228 }
229
230 static inline void log_clear_bit(struct log_c *l,
231                                  uint32_t *bs, unsigned bit)
232 {
233         ext2_clear_bit(bit, (unsigned long *) bs);
234         l->touched = 1;
235 }
236
237 /*----------------------------------------------------------------
238  * Header IO
239  *--------------------------------------------------------------*/
240 static void header_to_disk(struct log_header *core, struct log_header *disk)
241 {
242         disk->magic = cpu_to_le32(core->magic);
243         disk->version = cpu_to_le32(core->version);
244         disk->nr_regions = cpu_to_le64(core->nr_regions);
245 }
246
247 static void header_from_disk(struct log_header *core, struct log_header *disk)
248 {
249         core->magic = le32_to_cpu(disk->magic);
250         core->version = le32_to_cpu(disk->version);
251         core->nr_regions = le64_to_cpu(disk->nr_regions);
252 }
253
254 static int rw_header(struct log_c *lc, int rw)
255 {
256         lc->io_req.bi_rw = rw;
257         lc->io_req.mem.ptr.vma = lc->disk_header;
258         lc->io_req.notify.fn = NULL;
259
260         return dm_io(&lc->io_req, 1, &lc->header_location, NULL);
261 }
262
263 static int read_header(struct log_c *log)
264 {
265         int r;
266
267         r = rw_header(log, READ);
268         if (r)
269                 return r;
270
271         header_from_disk(&log->header, log->disk_header);
272
273         /* New log required? */
274         if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
275                 log->header.magic = MIRROR_MAGIC;
276                 log->header.version = MIRROR_DISK_VERSION;
277                 log->header.nr_regions = 0;
278         }
279
280 #ifdef __LITTLE_ENDIAN
281         if (log->header.version == 1)
282                 log->header.version = 2;
283 #endif
284
285         if (log->header.version != MIRROR_DISK_VERSION) {
286                 DMWARN("incompatible disk log version");
287                 return -EINVAL;
288         }
289
290         return 0;
291 }
292
293 static inline int write_header(struct log_c *log)
294 {
295         header_to_disk(&log->header, log->disk_header);
296         return rw_header(log, WRITE);
297 }
298
299 /*----------------------------------------------------------------
300  * core log constructor/destructor
301  *
302  * argv contains region_size followed optionally by [no]sync
303  *--------------------------------------------------------------*/
304 #define BYTE_SHIFT 3
305 static int create_log_context(struct dirty_log *log, struct dm_target *ti,
306                               unsigned int argc, char **argv,
307                               struct dm_dev *dev)
308 {
309         enum sync sync = DEFAULTSYNC;
310
311         struct log_c *lc;
312         uint32_t region_size;
313         unsigned int region_count;
314         size_t bitset_size, buf_size;
315         int r;
316
317         if (argc < 1 || argc > 2) {
318                 DMWARN("wrong number of arguments to mirror log");
319                 return -EINVAL;
320         }
321
322         if (argc > 1) {
323                 if (!strcmp(argv[1], "sync"))
324                         sync = FORCESYNC;
325                 else if (!strcmp(argv[1], "nosync"))
326                         sync = NOSYNC;
327                 else {
328                         DMWARN("unrecognised sync argument to mirror log: %s",
329                                argv[1]);
330                         return -EINVAL;
331                 }
332         }
333
334         if (sscanf(argv[0], "%u", &region_size) != 1) {
335                 DMWARN("invalid region size string");
336                 return -EINVAL;
337         }
338
339         region_count = dm_sector_div_up(ti->len, region_size);
340
341         lc = kmalloc(sizeof(*lc), GFP_KERNEL);
342         if (!lc) {
343                 DMWARN("couldn't allocate core log");
344                 return -ENOMEM;
345         }
346
347         lc->ti = ti;
348         lc->touched = 0;
349         lc->region_size = region_size;
350         lc->region_count = region_count;
351         lc->sync = sync;
352
353         /*
354          * Work out how many "unsigned long"s we need to hold the bitset.
355          */
356         bitset_size = dm_round_up(region_count,
357                                   sizeof(*lc->clean_bits) << BYTE_SHIFT);
358         bitset_size >>= BYTE_SHIFT;
359
360         lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
361
362         /*
363          * Disk log?
364          */
365         if (!dev) {
366                 lc->clean_bits = vmalloc(bitset_size);
367                 if (!lc->clean_bits) {
368                         DMWARN("couldn't allocate clean bitset");
369                         kfree(lc);
370                         return -ENOMEM;
371                 }
372                 lc->disk_header = NULL;
373         } else {
374                 lc->log_dev = dev;
375                 lc->log_dev_failed = 0;
376                 lc->header_location.bdev = lc->log_dev->bdev;
377                 lc->header_location.sector = 0;
378
379                 /*
380                  * Buffer holds both header and bitset.
381                  */
382                 buf_size = dm_round_up((LOG_OFFSET << SECTOR_SHIFT) +
383                                        bitset_size, ti->limits.hardsect_size);
384                 lc->header_location.count = buf_size >> SECTOR_SHIFT;
385                 lc->io_req.mem.type = DM_IO_VMA;
386                 lc->io_req.client = dm_io_client_create(dm_div_up(buf_size,
387                                                                    PAGE_SIZE));
388                 if (IS_ERR(lc->io_req.client)) {
389                         r = PTR_ERR(lc->io_req.client);
390                         DMWARN("couldn't allocate disk io client");
391                         kfree(lc);
392                         return -ENOMEM;
393                 }
394
395                 lc->disk_header = vmalloc(buf_size);
396                 if (!lc->disk_header) {
397                         DMWARN("couldn't allocate disk log buffer");
398                         kfree(lc);
399                         return -ENOMEM;
400                 }
401
402                 lc->clean_bits = (void *)lc->disk_header +
403                                  (LOG_OFFSET << SECTOR_SHIFT);
404         }
405
406         memset(lc->clean_bits, -1, bitset_size);
407
408         lc->sync_bits = vmalloc(bitset_size);
409         if (!lc->sync_bits) {
410                 DMWARN("couldn't allocate sync bitset");
411                 if (!dev)
412                         vfree(lc->clean_bits);
413                 vfree(lc->disk_header);
414                 kfree(lc);
415                 return -ENOMEM;
416         }
417         memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
418         lc->sync_count = (sync == NOSYNC) ? region_count : 0;
419
420         lc->recovering_bits = vmalloc(bitset_size);
421         if (!lc->recovering_bits) {
422                 DMWARN("couldn't allocate sync bitset");
423                 vfree(lc->sync_bits);
424                 if (!dev)
425                         vfree(lc->clean_bits);
426                 vfree(lc->disk_header);
427                 kfree(lc);
428                 return -ENOMEM;
429         }
430         memset(lc->recovering_bits, 0, bitset_size);
431         lc->sync_search = 0;
432         log->context = lc;
433
434         return 0;
435 }
436
437 static int core_ctr(struct dirty_log *log, struct dm_target *ti,
438                     unsigned int argc, char **argv)
439 {
440         return create_log_context(log, ti, argc, argv, NULL);
441 }
442
443 static void destroy_log_context(struct log_c *lc)
444 {
445         vfree(lc->sync_bits);
446         vfree(lc->recovering_bits);
447         kfree(lc);
448 }
449
450 static void core_dtr(struct dirty_log *log)
451 {
452         struct log_c *lc = (struct log_c *) log->context;
453
454         vfree(lc->clean_bits);
455         destroy_log_context(lc);
456 }
457
458 /*----------------------------------------------------------------
459  * disk log constructor/destructor
460  *
461  * argv contains log_device region_size followed optionally by [no]sync
462  *--------------------------------------------------------------*/
463 static int disk_ctr(struct dirty_log *log, struct dm_target *ti,
464                     unsigned int argc, char **argv)
465 {
466         int r;
467         struct dm_dev *dev;
468
469         if (argc < 2 || argc > 3) {
470                 DMWARN("wrong number of arguments to disk mirror log");
471                 return -EINVAL;
472         }
473
474         r = dm_get_device(ti, argv[0], 0, 0 /* FIXME */,
475                           FMODE_READ | FMODE_WRITE, &dev);
476         if (r)
477                 return r;
478
479         r = create_log_context(log, ti, argc - 1, argv + 1, dev);
480         if (r) {
481                 dm_put_device(ti, dev);
482                 return r;
483         }
484
485         return 0;
486 }
487
488 static void disk_dtr(struct dirty_log *log)
489 {
490         struct log_c *lc = (struct log_c *) log->context;
491
492         dm_put_device(lc->ti, lc->log_dev);
493         vfree(lc->disk_header);
494         dm_io_client_destroy(lc->io_req.client);
495         destroy_log_context(lc);
496 }
497
498 static int count_bits32(uint32_t *addr, unsigned size)
499 {
500         int count = 0, i;
501
502         for (i = 0; i < size; i++) {
503                 count += hweight32(*(addr+i));
504         }
505         return count;
506 }
507
508 static void fail_log_device(struct log_c *lc)
509 {
510         if (lc->log_dev_failed)
511                 return;
512
513         lc->log_dev_failed = 1;
514         dm_table_event(lc->ti->table);
515 }
516
517 static int disk_resume(struct dirty_log *log)
518 {
519         int r;
520         unsigned i;
521         struct log_c *lc = (struct log_c *) log->context;
522         size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
523
524         /* read the disk header */
525         r = read_header(lc);
526         if (r) {
527                 DMWARN("%s: Failed to read header on mirror log device",
528                        lc->log_dev->name);
529                 fail_log_device(lc);
530                 /*
531                  * If the log device cannot be read, we must assume
532                  * all regions are out-of-sync.  If we simply return
533                  * here, the state will be uninitialized and could
534                  * lead us to return 'in-sync' status for regions
535                  * that are actually 'out-of-sync'.
536                  */
537                 lc->header.nr_regions = 0;
538         }
539
540         /* set or clear any new bits -- device has grown */
541         if (lc->sync == NOSYNC)
542                 for (i = lc->header.nr_regions; i < lc->region_count; i++)
543                         /* FIXME: amazingly inefficient */
544                         log_set_bit(lc, lc->clean_bits, i);
545         else
546                 for (i = lc->header.nr_regions; i < lc->region_count; i++)
547                         /* FIXME: amazingly inefficient */
548                         log_clear_bit(lc, lc->clean_bits, i);
549
550         /* clear any old bits -- device has shrunk */
551         for (i = lc->region_count; i % (sizeof(*lc->clean_bits) << BYTE_SHIFT); i++)
552                 log_clear_bit(lc, lc->clean_bits, i);
553
554         /* copy clean across to sync */
555         memcpy(lc->sync_bits, lc->clean_bits, size);
556         lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count);
557         lc->sync_search = 0;
558
559         /* set the correct number of regions in the header */
560         lc->header.nr_regions = lc->region_count;
561
562         /* write the new header */
563         r = write_header(lc);
564         if (r) {
565                 DMWARN("%s: Failed to write header on mirror log device",
566                        lc->log_dev->name);
567                 fail_log_device(lc);
568         }
569
570         return r;
571 }
572
573 static uint32_t core_get_region_size(struct dirty_log *log)
574 {
575         struct log_c *lc = (struct log_c *) log->context;
576         return lc->region_size;
577 }
578
579 static int core_resume(struct dirty_log *log)
580 {
581         struct log_c *lc = (struct log_c *) log->context;
582         lc->sync_search = 0;
583         return 0;
584 }
585
586 static int core_is_clean(struct dirty_log *log, region_t region)
587 {
588         struct log_c *lc = (struct log_c *) log->context;
589         return log_test_bit(lc->clean_bits, region);
590 }
591
592 static int core_in_sync(struct dirty_log *log, region_t region, int block)
593 {
594         struct log_c *lc = (struct log_c *) log->context;
595         return log_test_bit(lc->sync_bits, region);
596 }
597
598 static int core_flush(struct dirty_log *log)
599 {
600         /* no op */
601         return 0;
602 }
603
604 static int disk_flush(struct dirty_log *log)
605 {
606         int r;
607         struct log_c *lc = (struct log_c *) log->context;
608
609         /* only write if the log has changed */
610         if (!lc->touched)
611                 return 0;
612
613         r = write_header(lc);
614         if (r)
615                 fail_log_device(lc);
616         else
617                 lc->touched = 0;
618
619         return r;
620 }
621
622 static void core_mark_region(struct dirty_log *log, region_t region)
623 {
624         struct log_c *lc = (struct log_c *) log->context;
625         log_clear_bit(lc, lc->clean_bits, region);
626 }
627
628 static void core_clear_region(struct dirty_log *log, region_t region)
629 {
630         struct log_c *lc = (struct log_c *) log->context;
631         log_set_bit(lc, lc->clean_bits, region);
632 }
633
634 static int core_get_resync_work(struct dirty_log *log, region_t *region)
635 {
636         struct log_c *lc = (struct log_c *) log->context;
637
638         if (lc->sync_search >= lc->region_count)
639                 return 0;
640
641         do {
642                 *region = ext2_find_next_zero_bit(
643                                              (unsigned long *) lc->sync_bits,
644                                              lc->region_count,
645                                              lc->sync_search);
646                 lc->sync_search = *region + 1;
647
648                 if (*region >= lc->region_count)
649                         return 0;
650
651         } while (log_test_bit(lc->recovering_bits, *region));
652
653         log_set_bit(lc, lc->recovering_bits, *region);
654         return 1;
655 }
656
657 static void core_set_region_sync(struct dirty_log *log, region_t region,
658                                  int in_sync)
659 {
660         struct log_c *lc = (struct log_c *) log->context;
661
662         log_clear_bit(lc, lc->recovering_bits, region);
663         if (in_sync) {
664                 log_set_bit(lc, lc->sync_bits, region);
665                 lc->sync_count++;
666         } else if (log_test_bit(lc->sync_bits, region)) {
667                 lc->sync_count--;
668                 log_clear_bit(lc, lc->sync_bits, region);
669         }
670 }
671
672 static region_t core_get_sync_count(struct dirty_log *log)
673 {
674         struct log_c *lc = (struct log_c *) log->context;
675
676         return lc->sync_count;
677 }
678
679 #define DMEMIT_SYNC \
680         if (lc->sync != DEFAULTSYNC) \
681                 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
682
683 static int core_status(struct dirty_log *log, status_type_t status,
684                        char *result, unsigned int maxlen)
685 {
686         int sz = 0;
687         struct log_c *lc = log->context;
688
689         switch(status) {
690         case STATUSTYPE_INFO:
691                 DMEMIT("1 %s", log->type->name);
692                 break;
693
694         case STATUSTYPE_TABLE:
695                 DMEMIT("%s %u %u ", log->type->name,
696                        lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
697                 DMEMIT_SYNC;
698         }
699
700         return sz;
701 }
702
703 static int disk_status(struct dirty_log *log, status_type_t status,
704                        char *result, unsigned int maxlen)
705 {
706         int sz = 0;
707         struct log_c *lc = log->context;
708
709         switch(status) {
710         case STATUSTYPE_INFO:
711                 DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name,
712                        lc->log_dev_failed ? 'D' : 'A');
713                 break;
714
715         case STATUSTYPE_TABLE:
716                 DMEMIT("%s %u %s %u ", log->type->name,
717                        lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name,
718                        lc->region_size);
719                 DMEMIT_SYNC;
720         }
721
722         return sz;
723 }
724
725 static struct dirty_log_type _core_type = {
726         .name = "core",
727         .module = THIS_MODULE,
728         .ctr = core_ctr,
729         .dtr = core_dtr,
730         .resume = core_resume,
731         .get_region_size = core_get_region_size,
732         .is_clean = core_is_clean,
733         .in_sync = core_in_sync,
734         .flush = core_flush,
735         .mark_region = core_mark_region,
736         .clear_region = core_clear_region,
737         .get_resync_work = core_get_resync_work,
738         .set_region_sync = core_set_region_sync,
739         .get_sync_count = core_get_sync_count,
740         .status = core_status,
741 };
742
743 static struct dirty_log_type _disk_type = {
744         .name = "disk",
745         .module = THIS_MODULE,
746         .ctr = disk_ctr,
747         .dtr = disk_dtr,
748         .postsuspend = disk_flush,
749         .resume = disk_resume,
750         .get_region_size = core_get_region_size,
751         .is_clean = core_is_clean,
752         .in_sync = core_in_sync,
753         .flush = disk_flush,
754         .mark_region = core_mark_region,
755         .clear_region = core_clear_region,
756         .get_resync_work = core_get_resync_work,
757         .set_region_sync = core_set_region_sync,
758         .get_sync_count = core_get_sync_count,
759         .status = disk_status,
760 };
761
762 int __init dm_dirty_log_init(void)
763 {
764         int r;
765
766         r = dm_register_dirty_log_type(&_core_type);
767         if (r)
768                 DMWARN("couldn't register core log");
769
770         r = dm_register_dirty_log_type(&_disk_type);
771         if (r) {
772                 DMWARN("couldn't register disk type");
773                 dm_unregister_dirty_log_type(&_core_type);
774         }
775
776         return r;
777 }
778
779 void dm_dirty_log_exit(void)
780 {
781         dm_unregister_dirty_log_type(&_disk_type);
782         dm_unregister_dirty_log_type(&_core_type);
783 }
784
785 EXPORT_SYMBOL(dm_register_dirty_log_type);
786 EXPORT_SYMBOL(dm_unregister_dirty_log_type);
787 EXPORT_SYMBOL(dm_create_dirty_log);
788 EXPORT_SYMBOL(dm_destroy_dirty_log);