[PATCH] md: Update status_resync to handle LARGE devices
[linux-2.6] / 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         struct io_region bits_location;
160 };
161
162 /*
163  * The touched member needs to be updated every time we access
164  * one of the bitsets.
165  */
166 static  inline int log_test_bit(uint32_t *bs, unsigned bit)
167 {
168         return ext2_test_bit(bit, (unsigned long *) bs) ? 1 : 0;
169 }
170
171 static inline void log_set_bit(struct log_c *l,
172                                uint32_t *bs, unsigned bit)
173 {
174         ext2_set_bit(bit, (unsigned long *) bs);
175         l->touched = 1;
176 }
177
178 static inline void log_clear_bit(struct log_c *l,
179                                  uint32_t *bs, unsigned bit)
180 {
181         ext2_clear_bit(bit, (unsigned long *) bs);
182         l->touched = 1;
183 }
184
185 /*----------------------------------------------------------------
186  * Header IO
187  *--------------------------------------------------------------*/
188 static void header_to_disk(struct log_header *core, struct log_header *disk)
189 {
190         disk->magic = cpu_to_le32(core->magic);
191         disk->version = cpu_to_le32(core->version);
192         disk->nr_regions = cpu_to_le64(core->nr_regions);
193 }
194
195 static void header_from_disk(struct log_header *core, struct log_header *disk)
196 {
197         core->magic = le32_to_cpu(disk->magic);
198         core->version = le32_to_cpu(disk->version);
199         core->nr_regions = le64_to_cpu(disk->nr_regions);
200 }
201
202 static int read_header(struct log_c *log)
203 {
204         int r;
205         unsigned long ebits;
206
207         r = dm_io_sync_vm(1, &log->header_location, READ,
208                           log->disk_header, &ebits);
209         if (r)
210                 return r;
211
212         header_from_disk(&log->header, log->disk_header);
213
214         /* New log required? */
215         if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
216                 log->header.magic = MIRROR_MAGIC;
217                 log->header.version = MIRROR_DISK_VERSION;
218                 log->header.nr_regions = 0;
219         }
220
221 #ifdef __LITTLE_ENDIAN
222         if (log->header.version == 1)
223                 log->header.version = 2;
224 #endif
225
226         if (log->header.version != MIRROR_DISK_VERSION) {
227                 DMWARN("incompatible disk log version");
228                 return -EINVAL;
229         }
230
231         return 0;
232 }
233
234 static inline int write_header(struct log_c *log)
235 {
236         unsigned long ebits;
237
238         header_to_disk(&log->header, log->disk_header);
239         return dm_io_sync_vm(1, &log->header_location, WRITE,
240                              log->disk_header, &ebits);
241 }
242
243 /*----------------------------------------------------------------
244  * Bits IO
245  *--------------------------------------------------------------*/
246 static int read_bits(struct log_c *log)
247 {
248         int r;
249         unsigned long ebits;
250
251         r = dm_io_sync_vm(1, &log->bits_location, READ,
252                           log->clean_bits, &ebits);
253         if (r)
254                 return r;
255
256         return 0;
257 }
258
259 static int write_bits(struct log_c *log)
260 {
261         unsigned long ebits;
262         return dm_io_sync_vm(1, &log->bits_location, WRITE,
263                              log->clean_bits, &ebits);
264 }
265
266 /*----------------------------------------------------------------
267  * core log constructor/destructor
268  *
269  * argv contains region_size followed optionally by [no]sync
270  *--------------------------------------------------------------*/
271 #define BYTE_SHIFT 3
272 static int core_ctr(struct dirty_log *log, struct dm_target *ti,
273                     unsigned int argc, char **argv)
274 {
275         enum sync sync = DEFAULTSYNC;
276
277         struct log_c *lc;
278         uint32_t region_size;
279         unsigned int region_count;
280         size_t bitset_size;
281
282         if (argc < 1 || argc > 2) {
283                 DMWARN("wrong number of arguments to mirror log");
284                 return -EINVAL;
285         }
286
287         if (argc > 1) {
288                 if (!strcmp(argv[1], "sync"))
289                         sync = FORCESYNC;
290                 else if (!strcmp(argv[1], "nosync"))
291                         sync = NOSYNC;
292                 else {
293                         DMWARN("unrecognised sync argument to mirror log: %s",
294                                argv[1]);
295                         return -EINVAL;
296                 }
297         }
298
299         if (sscanf(argv[0], "%u", &region_size) != 1) {
300                 DMWARN("invalid region size string");
301                 return -EINVAL;
302         }
303
304         region_count = dm_sector_div_up(ti->len, region_size);
305
306         lc = kmalloc(sizeof(*lc), GFP_KERNEL);
307         if (!lc) {
308                 DMWARN("couldn't allocate core log");
309                 return -ENOMEM;
310         }
311
312         lc->ti = ti;
313         lc->touched = 0;
314         lc->region_size = region_size;
315         lc->region_count = region_count;
316         lc->sync = sync;
317
318         /*
319          * Work out how many "unsigned long"s we need to hold the bitset.
320          */
321         bitset_size = dm_round_up(region_count,
322                                   sizeof(unsigned long) << BYTE_SHIFT);
323         bitset_size >>= BYTE_SHIFT;
324
325         lc->bitset_uint32_count = bitset_size / 4;
326         lc->clean_bits = vmalloc(bitset_size);
327         if (!lc->clean_bits) {
328                 DMWARN("couldn't allocate clean bitset");
329                 kfree(lc);
330                 return -ENOMEM;
331         }
332         memset(lc->clean_bits, -1, bitset_size);
333
334         lc->sync_bits = vmalloc(bitset_size);
335         if (!lc->sync_bits) {
336                 DMWARN("couldn't allocate sync bitset");
337                 vfree(lc->clean_bits);
338                 kfree(lc);
339                 return -ENOMEM;
340         }
341         memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
342         lc->sync_count = (sync == NOSYNC) ? region_count : 0;
343
344         lc->recovering_bits = vmalloc(bitset_size);
345         if (!lc->recovering_bits) {
346                 DMWARN("couldn't allocate sync bitset");
347                 vfree(lc->sync_bits);
348                 vfree(lc->clean_bits);
349                 kfree(lc);
350                 return -ENOMEM;
351         }
352         memset(lc->recovering_bits, 0, bitset_size);
353         lc->sync_search = 0;
354         log->context = lc;
355         return 0;
356 }
357
358 static void core_dtr(struct dirty_log *log)
359 {
360         struct log_c *lc = (struct log_c *) log->context;
361         vfree(lc->clean_bits);
362         vfree(lc->sync_bits);
363         vfree(lc->recovering_bits);
364         kfree(lc);
365 }
366
367 /*----------------------------------------------------------------
368  * disk log constructor/destructor
369  *
370  * argv contains log_device region_size followed optionally by [no]sync
371  *--------------------------------------------------------------*/
372 static int disk_ctr(struct dirty_log *log, struct dm_target *ti,
373                     unsigned int argc, char **argv)
374 {
375         int r;
376         size_t size;
377         struct log_c *lc;
378         struct dm_dev *dev;
379
380         if (argc < 2 || argc > 3) {
381                 DMWARN("wrong number of arguments to disk mirror log");
382                 return -EINVAL;
383         }
384
385         r = dm_get_device(ti, argv[0], 0, 0 /* FIXME */,
386                           FMODE_READ | FMODE_WRITE, &dev);
387         if (r)
388                 return r;
389
390         r = core_ctr(log, ti, argc - 1, argv + 1);
391         if (r) {
392                 dm_put_device(ti, dev);
393                 return r;
394         }
395
396         lc = (struct log_c *) log->context;
397         lc->log_dev = dev;
398
399         /* setup the disk header fields */
400         lc->header_location.bdev = lc->log_dev->bdev;
401         lc->header_location.sector = 0;
402         lc->header_location.count = 1;
403
404         /*
405          * We can't read less than this amount, even though we'll
406          * not be using most of this space.
407          */
408         lc->disk_header = vmalloc(1 << SECTOR_SHIFT);
409         if (!lc->disk_header)
410                 goto bad;
411
412         /* setup the disk bitset fields */
413         lc->bits_location.bdev = lc->log_dev->bdev;
414         lc->bits_location.sector = LOG_OFFSET;
415
416         size = dm_round_up(lc->bitset_uint32_count * sizeof(uint32_t),
417                            1 << SECTOR_SHIFT);
418         lc->bits_location.count = size >> SECTOR_SHIFT;
419         return 0;
420
421  bad:
422         dm_put_device(ti, lc->log_dev);
423         core_dtr(log);
424         return -ENOMEM;
425 }
426
427 static void disk_dtr(struct dirty_log *log)
428 {
429         struct log_c *lc = (struct log_c *) log->context;
430         dm_put_device(lc->ti, lc->log_dev);
431         vfree(lc->disk_header);
432         core_dtr(log);
433 }
434
435 static int count_bits32(uint32_t *addr, unsigned size)
436 {
437         int count = 0, i;
438
439         for (i = 0; i < size; i++) {
440                 count += hweight32(*(addr+i));
441         }
442         return count;
443 }
444
445 static int disk_resume(struct dirty_log *log)
446 {
447         int r;
448         unsigned i;
449         struct log_c *lc = (struct log_c *) log->context;
450         size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
451
452         /* read the disk header */
453         r = read_header(lc);
454         if (r)
455                 return r;
456
457         /* read the bits */
458         r = read_bits(lc);
459         if (r)
460                 return r;
461
462         /* set or clear any new bits */
463         if (lc->sync == NOSYNC)
464                 for (i = lc->header.nr_regions; i < lc->region_count; i++)
465                         /* FIXME: amazingly inefficient */
466                         log_set_bit(lc, lc->clean_bits, i);
467         else
468                 for (i = lc->header.nr_regions; i < lc->region_count; i++)
469                         /* FIXME: amazingly inefficient */
470                         log_clear_bit(lc, lc->clean_bits, i);
471
472         /* copy clean across to sync */
473         memcpy(lc->sync_bits, lc->clean_bits, size);
474         lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count);
475
476         /* write the bits */
477         r = write_bits(lc);
478         if (r)
479                 return r;
480
481         /* set the correct number of regions in the header */
482         lc->header.nr_regions = lc->region_count;
483
484         /* write the new header */
485         return write_header(lc);
486 }
487
488 static uint32_t core_get_region_size(struct dirty_log *log)
489 {
490         struct log_c *lc = (struct log_c *) log->context;
491         return lc->region_size;
492 }
493
494 static int core_is_clean(struct dirty_log *log, region_t region)
495 {
496         struct log_c *lc = (struct log_c *) log->context;
497         return log_test_bit(lc->clean_bits, region);
498 }
499
500 static int core_in_sync(struct dirty_log *log, region_t region, int block)
501 {
502         struct log_c *lc = (struct log_c *) log->context;
503         return log_test_bit(lc->sync_bits, region);
504 }
505
506 static int core_flush(struct dirty_log *log)
507 {
508         /* no op */
509         return 0;
510 }
511
512 static int disk_flush(struct dirty_log *log)
513 {
514         int r;
515         struct log_c *lc = (struct log_c *) log->context;
516
517         /* only write if the log has changed */
518         if (!lc->touched)
519                 return 0;
520
521         r = write_bits(lc);
522         if (!r)
523                 lc->touched = 0;
524
525         return r;
526 }
527
528 static void core_mark_region(struct dirty_log *log, region_t region)
529 {
530         struct log_c *lc = (struct log_c *) log->context;
531         log_clear_bit(lc, lc->clean_bits, region);
532 }
533
534 static void core_clear_region(struct dirty_log *log, region_t region)
535 {
536         struct log_c *lc = (struct log_c *) log->context;
537         log_set_bit(lc, lc->clean_bits, region);
538 }
539
540 static int core_get_resync_work(struct dirty_log *log, region_t *region)
541 {
542         struct log_c *lc = (struct log_c *) log->context;
543
544         if (lc->sync_search >= lc->region_count)
545                 return 0;
546
547         do {
548                 *region = ext2_find_next_zero_bit(
549                                              (unsigned long *) lc->sync_bits,
550                                              lc->region_count,
551                                              lc->sync_search);
552                 lc->sync_search = *region + 1;
553
554                 if (*region >= lc->region_count)
555                         return 0;
556
557         } while (log_test_bit(lc->recovering_bits, *region));
558
559         log_set_bit(lc, lc->recovering_bits, *region);
560         return 1;
561 }
562
563 static void core_complete_resync_work(struct dirty_log *log, region_t region,
564                                       int success)
565 {
566         struct log_c *lc = (struct log_c *) log->context;
567
568         log_clear_bit(lc, lc->recovering_bits, region);
569         if (success) {
570                 log_set_bit(lc, lc->sync_bits, region);
571                 lc->sync_count++;
572         }
573 }
574
575 static region_t core_get_sync_count(struct dirty_log *log)
576 {
577         struct log_c *lc = (struct log_c *) log->context;
578
579         return lc->sync_count;
580 }
581
582 #define DMEMIT_SYNC \
583         if (lc->sync != DEFAULTSYNC) \
584                 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
585
586 static int core_status(struct dirty_log *log, status_type_t status,
587                        char *result, unsigned int maxlen)
588 {
589         int sz = 0;
590         struct log_c *lc = log->context;
591
592         switch(status) {
593         case STATUSTYPE_INFO:
594                 break;
595
596         case STATUSTYPE_TABLE:
597                 DMEMIT("%s %u %u ", log->type->name,
598                        lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
599                 DMEMIT_SYNC;
600         }
601
602         return sz;
603 }
604
605 static int disk_status(struct dirty_log *log, status_type_t status,
606                        char *result, unsigned int maxlen)
607 {
608         int sz = 0;
609         char buffer[16];
610         struct log_c *lc = log->context;
611
612         switch(status) {
613         case STATUSTYPE_INFO:
614                 break;
615
616         case STATUSTYPE_TABLE:
617                 format_dev_t(buffer, lc->log_dev->bdev->bd_dev);
618                 DMEMIT("%s %u %s %u ", log->type->name,
619                        lc->sync == DEFAULTSYNC ? 2 : 3, buffer,
620                        lc->region_size);
621                 DMEMIT_SYNC;
622         }
623
624         return sz;
625 }
626
627 static struct dirty_log_type _core_type = {
628         .name = "core",
629         .module = THIS_MODULE,
630         .ctr = core_ctr,
631         .dtr = core_dtr,
632         .get_region_size = core_get_region_size,
633         .is_clean = core_is_clean,
634         .in_sync = core_in_sync,
635         .flush = core_flush,
636         .mark_region = core_mark_region,
637         .clear_region = core_clear_region,
638         .get_resync_work = core_get_resync_work,
639         .complete_resync_work = core_complete_resync_work,
640         .get_sync_count = core_get_sync_count,
641         .status = core_status,
642 };
643
644 static struct dirty_log_type _disk_type = {
645         .name = "disk",
646         .module = THIS_MODULE,
647         .ctr = disk_ctr,
648         .dtr = disk_dtr,
649         .suspend = disk_flush,
650         .resume = disk_resume,
651         .get_region_size = core_get_region_size,
652         .is_clean = core_is_clean,
653         .in_sync = core_in_sync,
654         .flush = disk_flush,
655         .mark_region = core_mark_region,
656         .clear_region = core_clear_region,
657         .get_resync_work = core_get_resync_work,
658         .complete_resync_work = core_complete_resync_work,
659         .get_sync_count = core_get_sync_count,
660         .status = disk_status,
661 };
662
663 int __init dm_dirty_log_init(void)
664 {
665         int r;
666
667         r = dm_register_dirty_log_type(&_core_type);
668         if (r)
669                 DMWARN("couldn't register core log");
670
671         r = dm_register_dirty_log_type(&_disk_type);
672         if (r) {
673                 DMWARN("couldn't register disk type");
674                 dm_unregister_dirty_log_type(&_core_type);
675         }
676
677         return r;
678 }
679
680 void dm_dirty_log_exit(void)
681 {
682         dm_unregister_dirty_log_type(&_disk_type);
683         dm_unregister_dirty_log_type(&_core_type);
684 }
685
686 EXPORT_SYMBOL(dm_register_dirty_log_type);
687 EXPORT_SYMBOL(dm_unregister_dirty_log_type);
688 EXPORT_SYMBOL(dm_create_dirty_log);
689 EXPORT_SYMBOL(dm_destroy_dirty_log);