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