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         struct dm_io_request io_req;
 153
 154         /*
 155          * Disk log fields
 156          */
 157         int log_dev_failed;
 158         struct dm_dev *log_dev;
 159         struct log_header header;
 160
 161         struct io_region header_location;
 162         struct log_header *disk_header;
 163 };
 164
 165 /*
 166  * The touched member needs to be updated every time we access
 167  * one of the bitsets.
 168  */
 169 static  inline int log_test_bit(uint32_t *bs, unsigned bit)
 170 {
 171         return ext2_test_bit(bit, (unsigned long *) bs) ? 1 : 0;
 172 }
 173
 174 static inline void log_set_bit(struct log_c *l,
 175                                uint32_t *bs, unsigned bit)
 176 {
 177         ext2_set_bit(bit, (unsigned long *) bs);
 178         l->touched = 1;
 179 }
 180
 181 static inline void log_clear_bit(struct log_c *l,
 182                                  uint32_t *bs, unsigned bit)
 183 {
 184         ext2_clear_bit(bit, (unsigned long *) bs);
 185         l->touched = 1;
 186 }
 187
 188 /*----------------------------------------------------------------
 189  * Header IO
 190  *--------------------------------------------------------------*/
 191 static void header_to_disk(struct log_header *core, struct log_header *disk)
 192 {
 193         disk->magic = cpu_to_le32(core->magic);
 194         disk->version = cpu_to_le32(core->version);
 195         disk->nr_regions = cpu_to_le64(core->nr_regions);
 196 }
 197
 198 static void header_from_disk(struct log_header *core, struct log_header *disk)
 199 {
 200         core->magic = le32_to_cpu(disk->magic);
 201         core->version = le32_to_cpu(disk->version);
 202         core->nr_regions = le64_to_cpu(disk->nr_regions);
 203 }
 204
 205 static int rw_header(struct log_c *lc, int rw)
 206 {
 207         lc->io_req.bi_rw = rw;
 208         lc->io_req.mem.ptr.vma = lc->disk_header;
 209         lc->io_req.notify.fn = NULL;
 210
 211         return dm_io(&lc->io_req, 1, &lc->header_location, NULL);
 212 }
 213
 214 static int read_header(struct log_c *log)
 215 {
 216         int r;
 217
 218         r = rw_header(log, READ);
 219         if (r)
 220                 return r;
 221
 222         header_from_disk(&log->header, log->disk_header);
 223
 224         /* New log required? */
 225         if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
 226                 log->header.magic = MIRROR_MAGIC;
 227                 log->header.version = MIRROR_DISK_VERSION;
 228                 log->header.nr_regions = 0;
 229         }
 230
 231 #ifdef __LITTLE_ENDIAN
 232         if (log->header.version == 1)
 233                 log->header.version = 2;
 234 #endif
 235
 236         if (log->header.version != MIRROR_DISK_VERSION) {
 237                 DMWARN("incompatible disk log version");
 238                 return -EINVAL;
 239         }
 240
 241         return 0;
 242 }
 243
 244 static inline int write_header(struct log_c *log)
 245 {
 246         header_to_disk(&log->header, log->disk_header);
 247         return rw_header(log, WRITE);
 248 }
 249
 250 /*----------------------------------------------------------------
 251  * core log constructor/destructor
 252  *
 253  * argv contains region_size followed optionally by [no]sync
 254  *--------------------------------------------------------------*/
 255 #define BYTE_SHIFT 3
 256 static int create_log_context(struct dirty_log *log, struct dm_target *ti,
 257                               unsigned int argc, char **argv,
 258                               struct dm_dev *dev)
 259 {
 260         enum sync sync = DEFAULTSYNC;
 261
 262         struct log_c *lc;
 263         uint32_t region_size;
 264         unsigned int region_count;
 265         size_t bitset_size, buf_size;
 266         int r;
 267
 268         if (argc < 1 || argc > 2) {
 269                 DMWARN("wrong number of arguments to mirror log");
 270                 return -EINVAL;
 271         }
 272
 273         if (argc > 1) {
 274                 if (!strcmp(argv[1], "sync"))
 275                         sync = FORCESYNC;
 276                 else if (!strcmp(argv[1], "nosync"))
 277                         sync = NOSYNC;
 278                 else {
 279                         DMWARN("unrecognised sync argument to mirror log: %s",
 280                                argv[1]);
 281                         return -EINVAL;
 282                 }
 283         }
 284
 285         if (sscanf(argv[0], "%u", &region_size) != 1) {
 286                 DMWARN("invalid region size string");
 287                 return -EINVAL;
 288         }
 289
 290         region_count = dm_sector_div_up(ti->len, region_size);
 291
 292         lc = kmalloc(sizeof(*lc), GFP_KERNEL);
 293         if (!lc) {
 294                 DMWARN("couldn't allocate core log");
 295                 return -ENOMEM;
 296         }
 297
 298         lc->ti = ti;
 299         lc->touched = 0;
 300         lc->region_size = region_size;
 301         lc->region_count = region_count;
 302         lc->sync = sync;
 303
 304         /*
 305          * Work out how many "unsigned long"s we need to hold the bitset.
 306          */
 307         bitset_size = dm_round_up(region_count,
 308                                   sizeof(*lc->clean_bits) << BYTE_SHIFT);
 309         bitset_size >>= BYTE_SHIFT;
 310
 311         lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
 312
 313         /*
 314          * Disk log?
 315          */
 316         if (!dev) {
 317                 lc->clean_bits = vmalloc(bitset_size);
 318                 if (!lc->clean_bits) {
 319                         DMWARN("couldn't allocate clean bitset");
 320                         kfree(lc);
 321                         return -ENOMEM;
 322                 }
 323                 lc->disk_header = NULL;
 324         } else {
 325                 lc->log_dev = dev;
 326                 lc->log_dev_failed = 0;
 327                 lc->header_location.bdev = lc->log_dev->bdev;
 328                 lc->header_location.sector = 0;
 329
 330                 /*
 331                  * Buffer holds both header and bitset.
 332                  */
 333                 buf_size = dm_round_up((LOG_OFFSET << SECTOR_SHIFT) +
 334                                        bitset_size, ti->limits.hardsect_size);
 335                 lc->header_location.count = buf_size >> SECTOR_SHIFT;
 336                 lc->io_req.mem.type = DM_IO_VMA;
 337                 lc->io_req.client = dm_io_client_create(dm_div_up(buf_size,
 338                                                                    PAGE_SIZE));
 339                 if (IS_ERR(lc->io_req.client)) {
 340                         r = PTR_ERR(lc->io_req.client);
 341                         DMWARN("couldn't allocate disk io client");
 342                         kfree(lc);
 343                         return -ENOMEM;
 344                 }
 345
 346                 lc->disk_header = vmalloc(buf_size);
 347                 if (!lc->disk_header) {
 348                         DMWARN("couldn't allocate disk log buffer");
 349                         kfree(lc);
 350                         return -ENOMEM;
 351                 }
 352
 353                 lc->clean_bits = (void *)lc->disk_header +
 354                                  (LOG_OFFSET << SECTOR_SHIFT);
 355         }
 356
 357         memset(lc->clean_bits, -1, bitset_size);
 358
 359         lc->sync_bits = vmalloc(bitset_size);
 360         if (!lc->sync_bits) {
 361                 DMWARN("couldn't allocate sync bitset");
 362                 if (!dev)
 363                         vfree(lc->clean_bits);
 364                 vfree(lc->disk_header);
 365                 kfree(lc);
 366                 return -ENOMEM;
 367         }
 368         memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
 369         lc->sync_count = (sync == NOSYNC) ? region_count : 0;
 370
 371         lc->recovering_bits = vmalloc(bitset_size);
 372         if (!lc->recovering_bits) {
 373                 DMWARN("couldn't allocate sync bitset");
 374                 vfree(lc->sync_bits);
 375                 if (!dev)
 376                         vfree(lc->clean_bits);
 377                 vfree(lc->disk_header);
 378                 kfree(lc);
 379                 return -ENOMEM;
 380         }
 381         memset(lc->recovering_bits, 0, bitset_size);
 382         lc->sync_search = 0;
 383         log->context = lc;
 384
 385         return 0;
 386 }
 387
 388 static int core_ctr(struct dirty_log *log, struct dm_target *ti,
 389                     unsigned int argc, char **argv)
 390 {
 391         return create_log_context(log, ti, argc, argv, NULL);
 392 }
 393
 394 static void destroy_log_context(struct log_c *lc)
 395 {
 396         vfree(lc->sync_bits);
 397         vfree(lc->recovering_bits);
 398         kfree(lc);
 399 }
 400
 401 static void core_dtr(struct dirty_log *log)
 402 {
 403         struct log_c *lc = (struct log_c *) log->context;
 404
 405         vfree(lc->clean_bits);
 406         destroy_log_context(lc);
 407 }
 408
 409 /*----------------------------------------------------------------
 410  * disk log constructor/destructor
 411  *
 412  * argv contains log_device region_size followed optionally by [no]sync
 413  *--------------------------------------------------------------*/
 414 static int disk_ctr(struct dirty_log *log, struct dm_target *ti,
 415                     unsigned int argc, char **argv)
 416 {
 417         int r;
 418         struct dm_dev *dev;
 419
 420         if (argc < 2 || argc > 3) {
 421                 DMWARN("wrong number of arguments to disk mirror log");
 422                 return -EINVAL;
 423         }
 424
 425         r = dm_get_device(ti, argv[0], 0, 0 /* FIXME */,
 426                           FMODE_READ | FMODE_WRITE, &dev);
 427         if (r)
 428                 return r;
 429
 430         r = create_log_context(log, ti, argc - 1, argv + 1, dev);
 431         if (r) {
 432                 dm_put_device(ti, dev);
 433                 return r;
 434         }
 435
 436         return 0;
 437 }
 438
 439 static void disk_dtr(struct dirty_log *log)
 440 {
 441         struct log_c *lc = (struct log_c *) log->context;
 442
 443         dm_put_device(lc->ti, lc->log_dev);
 444         vfree(lc->disk_header);
 445         dm_io_client_destroy(lc->io_req.client);
 446         destroy_log_context(lc);
 447 }
 448
 449 static int count_bits32(uint32_t *addr, unsigned size)
 450 {
 451         int count = 0, i;
 452
 453         for (i = 0; i < size; i++) {
 454                 count += hweight32(*(addr+i));
 455         }
 456         return count;
 457 }
 458
 459 static void fail_log_device(struct log_c *lc)
 460 {
 461         if (lc->log_dev_failed)
 462                 return;
 463
 464         lc->log_dev_failed = 1;
 465         dm_table_event(lc->ti->table);
 466 }
 467
 468 static int disk_resume(struct dirty_log *log)
 469 {
 470         int r;
 471         unsigned i;
 472         struct log_c *lc = (struct log_c *) log->context;
 473         size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
 474
 475         /* read the disk header */
 476         r = read_header(lc);
 477         if (r) {
 478                 DMWARN("%s: Failed to read header on mirror log device",
 479                        lc->log_dev->name);
 480                 fail_log_device(lc);
 481                 /*
 482                  * If the log device cannot be read, we must assume
 483                  * all regions are out-of-sync.  If we simply return
 484                  * here, the state will be uninitialized and could
 485                  * lead us to return 'in-sync' status for regions
 486                  * that are actually 'out-of-sync'.
 487                  */
 488                 lc->header.nr_regions = 0;
 489         }
 490
 491         /* set or clear any new bits -- device has grown */
 492         if (lc->sync == NOSYNC)
 493                 for (i = lc->header.nr_regions; i < lc->region_count; i++)
 494                         /* FIXME: amazingly inefficient */
 495                         log_set_bit(lc, lc->clean_bits, i);
 496         else
 497                 for (i = lc->header.nr_regions; i < lc->region_count; i++)
 498                         /* FIXME: amazingly inefficient */
 499                         log_clear_bit(lc, lc->clean_bits, i);
 500
 501         /* clear any old bits -- device has shrunk */
 502         for (i = lc->region_count; i % (sizeof(*lc->clean_bits) << BYTE_SHIFT); i++)
 503                 log_clear_bit(lc, lc->clean_bits, i);
 504
 505         /* copy clean across to sync */
 506         memcpy(lc->sync_bits, lc->clean_bits, size);
 507         lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count);
 508         lc->sync_search = 0;
 509
 510         /* set the correct number of regions in the header */
 511         lc->header.nr_regions = lc->region_count;
 512
 513         /* write the new header */
 514         r = write_header(lc);
 515         if (r) {
 516                 DMWARN("%s: Failed to write header on mirror log device",
 517                        lc->log_dev->name);
 518                 fail_log_device(lc);
 519         }
 520
 521         return r;
 522 }
 523
 524 static uint32_t core_get_region_size(struct dirty_log *log)
 525 {
 526         struct log_c *lc = (struct log_c *) log->context;
 527         return lc->region_size;
 528 }
 529
 530 static int core_resume(struct dirty_log *log)
 531 {
 532         struct log_c *lc = (struct log_c *) log->context;
 533         lc->sync_search = 0;
 534         return 0;
 535 }
 536
 537 static int core_is_clean(struct dirty_log *log, region_t region)
 538 {
 539         struct log_c *lc = (struct log_c *) log->context;
 540         return log_test_bit(lc->clean_bits, region);
 541 }
 542
 543 static int core_in_sync(struct dirty_log *log, region_t region, int block)
 544 {
 545         struct log_c *lc = (struct log_c *) log->context;
 546         return log_test_bit(lc->sync_bits, region);
 547 }
 548
 549 static int core_flush(struct dirty_log *log)
 550 {
 551         /* no op */
 552         return 0;
 553 }
 554
 555 static int disk_flush(struct dirty_log *log)
 556 {
 557         int r;
 558         struct log_c *lc = (struct log_c *) log->context;
 559
 560         /* only write if the log has changed */
 561         if (!lc->touched)
 562                 return 0;
 563
 564         r = write_header(lc);
 565         if (r)
 566                 fail_log_device(lc);
 567         else
 568                 lc->touched = 0;
 569
 570         return r;
 571 }
 572
 573 static void core_mark_region(struct dirty_log *log, region_t region)
 574 {
 575         struct log_c *lc = (struct log_c *) log->context;
 576         log_clear_bit(lc, lc->clean_bits, region);
 577 }
 578
 579 static void core_clear_region(struct dirty_log *log, region_t region)
 580 {
 581         struct log_c *lc = (struct log_c *) log->context;
 582         log_set_bit(lc, lc->clean_bits, region);
 583 }
 584
 585 static int core_get_resync_work(struct dirty_log *log, region_t *region)
 586 {
 587         struct log_c *lc = (struct log_c *) log->context;
 588
 589         if (lc->sync_search >= lc->region_count)
 590                 return 0;
 591
 592         do {
 593                 *region = ext2_find_next_zero_bit(
 594                                              (unsigned long *) lc->sync_bits,
 595                                              lc->region_count,
 596                                              lc->sync_search);
 597                 lc->sync_search = *region + 1;
 598
 599                 if (*region >= lc->region_count)
 600                         return 0;
 601
 602         } while (log_test_bit(lc->recovering_bits, *region));
 603
 604         log_set_bit(lc, lc->recovering_bits, *region);
 605         return 1;
 606 }
 607
 608 static void core_set_region_sync(struct dirty_log *log, region_t region,
 609                                  int in_sync)
 610 {
 611         struct log_c *lc = (struct log_c *) log->context;
 612
 613         log_clear_bit(lc, lc->recovering_bits, region);
 614         if (in_sync) {
 615                 log_set_bit(lc, lc->sync_bits, region);
 616                 lc->sync_count++;
 617         } else if (log_test_bit(lc->sync_bits, region)) {
 618                 lc->sync_count--;
 619                 log_clear_bit(lc, lc->sync_bits, region);
 620         }
 621 }
 622
 623 static region_t core_get_sync_count(struct dirty_log *log)
 624 {
 625         struct log_c *lc = (struct log_c *) log->context;
 626
 627         return lc->sync_count;
 628 }
 629
 630 #define DMEMIT_SYNC \
 631         if (lc->sync != DEFAULTSYNC) \
 632                 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
 633
 634 static int core_status(struct dirty_log *log, status_type_t status,
 635                        char *result, unsigned int maxlen)
 636 {
 637         int sz = 0;
 638         struct log_c *lc = log->context;
 639
 640         switch(status) {
 641         case STATUSTYPE_INFO:
 642                 DMEMIT("1 %s", log->type->name);
 643                 break;
 644
 645         case STATUSTYPE_TABLE:
 646                 DMEMIT("%s %u %u ", log->type->name,
 647                        lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
 648                 DMEMIT_SYNC;
 649         }
 650
 651         return sz;
 652 }
 653
 654 static int disk_status(struct dirty_log *log, status_type_t status,
 655                        char *result, unsigned int maxlen)
 656 {
 657         int sz = 0;
 658         struct log_c *lc = log->context;
 659
 660         switch(status) {
 661         case STATUSTYPE_INFO:
 662                 DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name,
 663                        lc->log_dev_failed ? 'D' : 'A');
 664                 break;
 665
 666         case STATUSTYPE_TABLE:
 667                 DMEMIT("%s %u %s %u ", log->type->name,
 668                        lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name,
 669                        lc->region_size);
 670                 DMEMIT_SYNC;
 671         }
 672
 673         return sz;
 674 }
 675
 676 static struct dirty_log_type _core_type = {
 677         .name = "core",
 678         .module = THIS_MODULE,
 679         .ctr = core_ctr,
 680         .dtr = core_dtr,
 681         .resume = core_resume,
 682         .get_region_size = core_get_region_size,
 683         .is_clean = core_is_clean,
 684         .in_sync = core_in_sync,
 685         .flush = core_flush,
 686         .mark_region = core_mark_region,
 687         .clear_region = core_clear_region,
 688         .get_resync_work = core_get_resync_work,
 689         .set_region_sync = core_set_region_sync,
 690         .get_sync_count = core_get_sync_count,
 691         .status = core_status,
 692 };
 693
 694 static struct dirty_log_type _disk_type = {
 695         .name = "disk",
 696         .module = THIS_MODULE,
 697         .ctr = disk_ctr,
 698         .dtr = disk_dtr,
 699         .postsuspend = disk_flush,
 700         .resume = disk_resume,
 701         .get_region_size = core_get_region_size,
 702         .is_clean = core_is_clean,
 703         .in_sync = core_in_sync,
 704         .flush = disk_flush,
 705         .mark_region = core_mark_region,
 706         .clear_region = core_clear_region,
 707         .get_resync_work = core_get_resync_work,
 708         .set_region_sync = core_set_region_sync,
 709         .get_sync_count = core_get_sync_count,
 710         .status = disk_status,
 711 };
 712
 713 int __init dm_dirty_log_init(void)
 714 {
 715         int r;
 716
 717         r = dm_register_dirty_log_type(&_core_type);
 718         if (r)
 719                 DMWARN("couldn't register core log");
 720
 721         r = dm_register_dirty_log_type(&_disk_type);
 722         if (r) {
 723                 DMWARN("couldn't register disk type");
 724                 dm_unregister_dirty_log_type(&_core_type);
 725         }
 726
 727         return r;
 728 }
 729
 730 void dm_dirty_log_exit(void)
 731 {
 732         dm_unregister_dirty_log_type(&_disk_type);
 733         dm_unregister_dirty_log_type(&_core_type);
 734 }
 735
 736 EXPORT_SYMBOL(dm_register_dirty_log_type);
 737 EXPORT_SYMBOL(dm_unregister_dirty_log_type);
 738 EXPORT_SYMBOL(dm_create_dirty_log);
 739 EXPORT_SYMBOL(dm_destroy_dirty_log);