drivers/md/dm-log.c

   1 /*
   2  * Copyright (C) 2003 Sistina Software
   3  * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
   4  *
   5  * This file is released under the LGPL.
   6  */
   7
   8 #include <linux/init.h>
   9 #include <linux/slab.h>
  10 #include <linux/module.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/dm-io.h>
  13 #include <linux/dm-dirty-log.h>
  14
  15 #include <linux/device-mapper.h>
  16
  17 #define DM_MSG_PREFIX "dirty region log"
  18
  19 static LIST_HEAD(_log_types);
  20 static DEFINE_SPINLOCK(_lock);
  21
  22 static struct dm_dirty_log_type *__find_dirty_log_type(const char *name)
  23 {
  24         struct dm_dirty_log_type *log_type;
  25
  26         list_for_each_entry(log_type, &_log_types, list)
  27                 if (!strcmp(name, log_type->name))
  28                         return log_type;
  29
  30         return NULL;
  31 }
  32
  33 static struct dm_dirty_log_type *_get_dirty_log_type(const char *name)
  34 {
  35         struct dm_dirty_log_type *log_type;
  36
  37         spin_lock(&_lock);
  38
  39         log_type = __find_dirty_log_type(name);
  40         if (log_type && !try_module_get(log_type->module))
  41                 log_type = NULL;
  42
  43         spin_unlock(&_lock);
  44
  45         return log_type;
  46 }
  47
  48 /*
  49  * get_type
  50  * @type_name
  51  *
  52  * Attempt to retrieve the dm_dirty_log_type by name.  If not already
  53  * available, attempt to load the appropriate module.
  54  *
  55  * Log modules are named "dm-log-" followed by the 'type_name'.
  56  * Modules may contain multiple types.
  57  * This function will first try the module "dm-log-<type_name>",
  58  * then truncate 'type_name' on the last '-' and try again.
  59  *
  60  * For example, if type_name was "clustered-disk", it would search
  61  * 'dm-log-clustered-disk' then 'dm-log-clustered'.
  62  *
  63  * Returns: dirty_log_type* on success, NULL on failure
  64  */
  65 static struct dm_dirty_log_type *get_type(const char *type_name)
  66 {
  67         char *p, *type_name_dup;
  68         struct dm_dirty_log_type *log_type;
  69
  70         if (!type_name)
  71                 return NULL;
  72
  73         log_type = _get_dirty_log_type(type_name);
  74         if (log_type)
  75                 return log_type;
  76
  77         type_name_dup = kstrdup(type_name, GFP_KERNEL);
  78         if (!type_name_dup) {
  79                 DMWARN("No memory left to attempt log module load for \"%s\"",
  80                        type_name);
  81                 return NULL;
  82         }
  83
  84         while (request_module("dm-log-%s", type_name_dup) ||
  85                !(log_type = _get_dirty_log_type(type_name))) {
  86                 p = strrchr(type_name_dup, '-');
  87                 if (!p)
  88                         break;
  89                 p[0] = '\0';
  90         }
  91
  92         if (!log_type)
  93                 DMWARN("Module for logging type \"%s\" not found.", type_name);
  94
  95         kfree(type_name_dup);
  96
  97         return log_type;
  98 }
  99
 100 static void put_type(struct dm_dirty_log_type *type)
 101 {
 102         if (!type)
 103                 return;
 104
 105         spin_lock(&_lock);
 106         if (!__find_dirty_log_type(type->name))
 107                 goto out;
 108
 109         module_put(type->module);
 110
 111 out:
 112         spin_unlock(&_lock);
 113 }
 114
 115 int dm_dirty_log_type_register(struct dm_dirty_log_type *type)
 116 {
 117         int r = 0;
 118
 119         spin_lock(&_lock);
 120         if (!__find_dirty_log_type(type->name))
 121                 list_add(&type->list, &_log_types);
 122         else
 123                 r = -EEXIST;
 124         spin_unlock(&_lock);
 125
 126         return r;
 127 }
 128 EXPORT_SYMBOL(dm_dirty_log_type_register);
 129
 130 int dm_dirty_log_type_unregister(struct dm_dirty_log_type *type)
 131 {
 132         spin_lock(&_lock);
 133
 134         if (!__find_dirty_log_type(type->name)) {
 135                 spin_unlock(&_lock);
 136                 return -EINVAL;
 137         }
 138
 139         list_del(&type->list);
 140
 141         spin_unlock(&_lock);
 142
 143         return 0;
 144 }
 145 EXPORT_SYMBOL(dm_dirty_log_type_unregister);
 146
 147 struct dm_dirty_log *dm_dirty_log_create(const char *type_name,
 148                         struct dm_target *ti,
 149                         int (*flush_callback_fn)(struct dm_target *ti),
 150                         unsigned int argc, char **argv)
 151 {
 152         struct dm_dirty_log_type *type;
 153         struct dm_dirty_log *log;
 154
 155         log = kmalloc(sizeof(*log), GFP_KERNEL);
 156         if (!log)
 157                 return NULL;
 158
 159         type = get_type(type_name);
 160         if (!type) {
 161                 kfree(log);
 162                 return NULL;
 163         }
 164
 165         log->flush_callback_fn = flush_callback_fn;
 166         log->type = type;
 167         if (type->ctr(log, ti, argc, argv)) {
 168                 kfree(log);
 169                 put_type(type);
 170                 return NULL;
 171         }
 172
 173         return log;
 174 }
 175 EXPORT_SYMBOL(dm_dirty_log_create);
 176
 177 void dm_dirty_log_destroy(struct dm_dirty_log *log)
 178 {
 179         log->type->dtr(log);
 180         put_type(log->type);
 181         kfree(log);
 182 }
 183 EXPORT_SYMBOL(dm_dirty_log_destroy);
 184
 185 /*-----------------------------------------------------------------
 186  * Persistent and core logs share a lot of their implementation.
 187  * FIXME: need a reload method to be called from a resume
 188  *---------------------------------------------------------------*/
 189 /*
 190  * Magic for persistent mirrors: "MiRr"
 191  */
 192 #define MIRROR_MAGIC 0x4D695272
 193
 194 /*
 195  * The on-disk version of the metadata.
 196  */
 197 #define MIRROR_DISK_VERSION 2
 198 #define LOG_OFFSET 2
 199
 200 struct log_header_disk {
 201         __le32 magic;
 202
 203         /*
 204          * Simple, incrementing version. no backward
 205          * compatibility.
 206          */
 207         __le32 version;
 208         __le64 nr_regions;
 209 } __packed;
 210
 211 struct log_header_core {
 212         uint32_t magic;
 213         uint32_t version;
 214         uint64_t nr_regions;
 215 };
 216
 217 struct log_c {
 218         struct dm_target *ti;
 219         int touched_dirtied;
 220         int touched_cleaned;
 221         int flush_failed;
 222         uint32_t region_size;
 223         unsigned int region_count;
 224         region_t sync_count;
 225
 226         unsigned bitset_uint32_count;
 227         uint32_t *clean_bits;
 228         uint32_t *sync_bits;
 229         uint32_t *recovering_bits;      /* FIXME: this seems excessive */
 230
 231         int sync_search;
 232
 233         /* Resync flag */
 234         enum sync {
 235                 DEFAULTSYNC,    /* Synchronize if necessary */
 236                 NOSYNC,         /* Devices known to be already in sync */
 237                 FORCESYNC,      /* Force a sync to happen */
 238         } sync;
 239
 240         struct dm_io_request io_req;
 241
 242         /*
 243          * Disk log fields
 244          */
 245         int log_dev_failed;
 246         int log_dev_flush_failed;
 247         struct dm_dev *log_dev;
 248         struct log_header_core header;
 249
 250         struct dm_io_region header_location;
 251         struct log_header_disk *disk_header;
 252 };
 253
 254 /*
 255  * The touched member needs to be updated every time we access
 256  * one of the bitsets.
 257  */
 258 static inline int log_test_bit(uint32_t *bs, unsigned bit)
 259 {
 260         return test_bit_le(bit, (unsigned long *) bs) ? 1 : 0;
 261 }
 262
 263 static inline void log_set_bit(struct log_c *l,
 264                                uint32_t *bs, unsigned bit)
 265 {
 266         __test_and_set_bit_le(bit, (unsigned long *) bs);
 267         l->touched_cleaned = 1;
 268 }
 269
 270 static inline void log_clear_bit(struct log_c *l,
 271                                  uint32_t *bs, unsigned bit)
 272 {
 273         __test_and_clear_bit_le(bit, (unsigned long *) bs);
 274         l->touched_dirtied = 1;
 275 }
 276
 277 /*----------------------------------------------------------------
 278  * Header IO
 279  *--------------------------------------------------------------*/
 280 static void header_to_disk(struct log_header_core *core, struct log_header_disk *disk)
 281 {
 282         disk->magic = cpu_to_le32(core->magic);
 283         disk->version = cpu_to_le32(core->version);
 284         disk->nr_regions = cpu_to_le64(core->nr_regions);
 285 }
 286
 287 static void header_from_disk(struct log_header_core *core, struct log_header_disk *disk)
 288 {
 289         core->magic = le32_to_cpu(disk->magic);
 290         core->version = le32_to_cpu(disk->version);
 291         core->nr_regions = le64_to_cpu(disk->nr_regions);
 292 }
 293
 294 static int rw_header(struct log_c *lc, int rw)
 295 {
 296         lc->io_req.bi_rw = rw;
 297
 298         return dm_io(&lc->io_req, 1, &lc->header_location, NULL);
 299 }
 300
 301 static int flush_header(struct log_c *lc)
 302 {
 303         struct dm_io_region null_location = {
 304                 .bdev = lc->header_location.bdev,
 305                 .sector = 0,
 306                 .count = 0,
 307         };
 308
 309         lc->io_req.bi_rw = WRITE_FLUSH;
 310
 311         return dm_io(&lc->io_req, 1, &null_location, NULL);
 312 }
 313
 314 static int read_header(struct log_c *log)
 315 {
 316         int r;
 317
 318         r = rw_header(log, READ);
 319         if (r)
 320                 return r;
 321
 322         header_from_disk(&log->header, log->disk_header);
 323
 324         /* New log required? */
 325         if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) {
 326                 log->header.magic = MIRROR_MAGIC;
 327                 log->header.version = MIRROR_DISK_VERSION;
 328                 log->header.nr_regions = 0;
 329         }
 330
 331 #ifdef __LITTLE_ENDIAN
 332         if (log->header.version == 1)
 333                 log->header.version = 2;
 334 #endif
 335
 336         if (log->header.version != MIRROR_DISK_VERSION) {
 337                 DMWARN("incompatible disk log version");
 338                 return -EINVAL;
 339         }
 340
 341         return 0;
 342 }
 343
 344 static int _check_region_size(struct dm_target *ti, uint32_t region_size)
 345 {
 346         if (region_size < 2 || region_size > ti->len)
 347                 return 0;
 348
 349         if (!is_power_of_2(region_size))
 350                 return 0;
 351
 352         return 1;
 353 }
 354
 355 /*----------------------------------------------------------------
 356  * core log constructor/destructor
 357  *
 358  * argv contains region_size followed optionally by [no]sync
 359  *--------------------------------------------------------------*/
 360 #define BYTE_SHIFT 3
 361 static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti,
 362                               unsigned int argc, char **argv,
 363                               struct dm_dev *dev)
 364 {
 365         enum sync sync = DEFAULTSYNC;
 366
 367         struct log_c *lc;
 368         uint32_t region_size;
 369         unsigned int region_count;
 370         size_t bitset_size, buf_size;
 371         int r;
 372
 373         if (argc < 1 || argc > 2) {
 374                 DMWARN("wrong number of arguments to dirty region log");
 375                 return -EINVAL;
 376         }
 377
 378         if (argc > 1) {
 379                 if (!strcmp(argv[1], "sync"))
 380                         sync = FORCESYNC;
 381                 else if (!strcmp(argv[1], "nosync"))
 382                         sync = NOSYNC;
 383                 else {
 384                         DMWARN("unrecognised sync argument to "
 385                                "dirty region log: %s", argv[1]);
 386                         return -EINVAL;
 387                 }
 388         }
 389
 390         if (sscanf(argv[0], "%u", &region_size) != 1 ||
 391             !_check_region_size(ti, region_size)) {
 392                 DMWARN("invalid region size %s", argv[0]);
 393                 return -EINVAL;
 394         }
 395
 396         region_count = dm_sector_div_up(ti->len, region_size);
 397
 398         lc = kmalloc(sizeof(*lc), GFP_KERNEL);
 399         if (!lc) {
 400                 DMWARN("couldn't allocate core log");
 401                 return -ENOMEM;
 402         }
 403
 404         lc->ti = ti;
 405         lc->touched_dirtied = 0;
 406         lc->touched_cleaned = 0;
 407         lc->flush_failed = 0;
 408         lc->region_size = region_size;
 409         lc->region_count = region_count;
 410         lc->sync = sync;
 411
 412         /*
 413          * Work out how many "unsigned long"s we need to hold the bitset.
 414          */
 415         bitset_size = dm_round_up(region_count,
 416                                   sizeof(*lc->clean_bits) << BYTE_SHIFT);
 417         bitset_size >>= BYTE_SHIFT;
 418
 419         lc->bitset_uint32_count = bitset_size / sizeof(*lc->clean_bits);
 420
 421         /*
 422          * Disk log?
 423          */
 424         if (!dev) {
 425                 lc->clean_bits = vmalloc(bitset_size);
 426                 if (!lc->clean_bits) {
 427                         DMWARN("couldn't allocate clean bitset");
 428                         kfree(lc);
 429                         return -ENOMEM;
 430                 }
 431                 lc->disk_header = NULL;
 432         } else {
 433                 lc->log_dev = dev;
 434                 lc->log_dev_failed = 0;
 435                 lc->log_dev_flush_failed = 0;
 436                 lc->header_location.bdev = lc->log_dev->bdev;
 437                 lc->header_location.sector = 0;
 438
 439                 /*
 440                  * Buffer holds both header and bitset.
 441                  */
 442                 buf_size =
 443                     dm_round_up((LOG_OFFSET << SECTOR_SHIFT) + bitset_size,
 444                                 bdev_logical_block_size(lc->header_location.
 445                                                             bdev));
 446
 447                 if (buf_size > i_size_read(dev->bdev->bd_inode)) {
 448                         DMWARN("log device %s too small: need %llu bytes",
 449                                 dev->name, (unsigned long long)buf_size);
 450                         kfree(lc);
 451                         return -EINVAL;
 452                 }
 453
 454                 lc->header_location.count = buf_size >> SECTOR_SHIFT;
 455
 456                 lc->io_req.mem.type = DM_IO_VMA;
 457                 lc->io_req.notify.fn = NULL;
 458                 lc->io_req.client = dm_io_client_create();
 459                 if (IS_ERR(lc->io_req.client)) {
 460                         r = PTR_ERR(lc->io_req.client);
 461                         DMWARN("couldn't allocate disk io client");
 462                         kfree(lc);
 463                         return r;
 464                 }
 465
 466                 lc->disk_header = vmalloc(buf_size);
 467                 if (!lc->disk_header) {
 468                         DMWARN("couldn't allocate disk log buffer");
 469                         dm_io_client_destroy(lc->io_req.client);
 470                         kfree(lc);
 471                         return -ENOMEM;
 472                 }
 473
 474                 lc->io_req.mem.ptr.vma = lc->disk_header;
 475                 lc->clean_bits = (void *)lc->disk_header +
 476                                  (LOG_OFFSET << SECTOR_SHIFT);
 477         }
 478
 479         memset(lc->clean_bits, -1, bitset_size);
 480
 481         lc->sync_bits = vmalloc(bitset_size);
 482         if (!lc->sync_bits) {
 483                 DMWARN("couldn't allocate sync bitset");
 484                 if (!dev)
 485                         vfree(lc->clean_bits);
 486                 else
 487                         dm_io_client_destroy(lc->io_req.client);
 488                 vfree(lc->disk_header);
 489                 kfree(lc);
 490                 return -ENOMEM;
 491         }
 492         memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size);
 493         lc->sync_count = (sync == NOSYNC) ? region_count : 0;
 494
 495         lc->recovering_bits = vmalloc(bitset_size);
 496         if (!lc->recovering_bits) {
 497                 DMWARN("couldn't allocate sync bitset");
 498                 vfree(lc->sync_bits);
 499                 if (!dev)
 500                         vfree(lc->clean_bits);
 501                 else
 502                         dm_io_client_destroy(lc->io_req.client);
 503                 vfree(lc->disk_header);
 504                 kfree(lc);
 505                 return -ENOMEM;
 506         }
 507         memset(lc->recovering_bits, 0, bitset_size);
 508         lc->sync_search = 0;
 509         log->context = lc;
 510
 511         return 0;
 512 }
 513
 514 static int core_ctr(struct dm_dirty_log *log, struct dm_target *ti,
 515                     unsigned int argc, char **argv)
 516 {
 517         return create_log_context(log, ti, argc, argv, NULL);
 518 }
 519
 520 static void destroy_log_context(struct log_c *lc)
 521 {
 522         vfree(lc->sync_bits);
 523         vfree(lc->recovering_bits);
 524         kfree(lc);
 525 }
 526
 527 static void core_dtr(struct dm_dirty_log *log)
 528 {
 529         struct log_c *lc = (struct log_c *) log->context;
 530
 531         vfree(lc->clean_bits);
 532         destroy_log_context(lc);
 533 }
 534
 535 /*----------------------------------------------------------------
 536  * disk log constructor/destructor
 537  *
 538  * argv contains log_device region_size followed optionally by [no]sync
 539  *--------------------------------------------------------------*/
 540 static int disk_ctr(struct dm_dirty_log *log, struct dm_target *ti,
 541                     unsigned int argc, char **argv)
 542 {
 543         int r;
 544         struct dm_dev *dev;
 545
 546         if (argc < 2 || argc > 3) {
 547                 DMWARN("wrong number of arguments to disk dirty region log");
 548                 return -EINVAL;
 549         }
 550
 551         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &dev);
 552         if (r)
 553                 return r;
 554
 555         r = create_log_context(log, ti, argc - 1, argv + 1, dev);
 556         if (r) {
 557                 dm_put_device(ti, dev);
 558                 return r;
 559         }
 560
 561         return 0;
 562 }
 563
 564 static void disk_dtr(struct dm_dirty_log *log)
 565 {
 566         struct log_c *lc = (struct log_c *) log->context;
 567
 568         dm_put_device(lc->ti, lc->log_dev);
 569         vfree(lc->disk_header);
 570         dm_io_client_destroy(lc->io_req.client);
 571         destroy_log_context(lc);
 572 }
 573
 574 static int count_bits32(uint32_t *addr, unsigned size)
 575 {
 576         int count = 0, i;
 577
 578         for (i = 0; i < size; i++) {
 579                 count += hweight32(*(addr+i));
 580         }
 581         return count;
 582 }
 583
 584 static void fail_log_device(struct log_c *lc)
 585 {
 586         if (lc->log_dev_failed)
 587                 return;
 588
 589         lc->log_dev_failed = 1;
 590         dm_table_event(lc->ti->table);
 591 }
 592
 593 static int disk_resume(struct dm_dirty_log *log)
 594 {
 595         int r;
 596         unsigned i;
 597         struct log_c *lc = (struct log_c *) log->context;
 598         size_t size = lc->bitset_uint32_count * sizeof(uint32_t);
 599
 600         /* read the disk header */
 601         r = read_header(lc);
 602         if (r) {
 603                 DMWARN("%s: Failed to read header on dirty region log device",
 604                        lc->log_dev->name);
 605                 fail_log_device(lc);
 606                 /*
 607                  * If the log device cannot be read, we must assume
 608                  * all regions are out-of-sync.  If we simply return
 609                  * here, the state will be uninitialized and could
 610                  * lead us to return 'in-sync' status for regions
 611                  * that are actually 'out-of-sync'.
 612                  */
 613                 lc->header.nr_regions = 0;
 614         }
 615
 616         /* set or clear any new bits -- device has grown */
 617         if (lc->sync == NOSYNC)
 618                 for (i = lc->header.nr_regions; i < lc->region_count; i++)
 619                         /* FIXME: amazingly inefficient */
 620                         log_set_bit(lc, lc->clean_bits, i);
 621         else
 622                 for (i = lc->header.nr_regions; i < lc->region_count; i++)
 623                         /* FIXME: amazingly inefficient */
 624                         log_clear_bit(lc, lc->clean_bits, i);
 625
 626         /* clear any old bits -- device has shrunk */
 627         for (i = lc->region_count; i % (sizeof(*lc->clean_bits) << BYTE_SHIFT); i++)
 628                 log_clear_bit(lc, lc->clean_bits, i);
 629
 630         /* copy clean across to sync */
 631         memcpy(lc->sync_bits, lc->clean_bits, size);
 632         lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count);
 633         lc->sync_search = 0;
 634
 635         /* set the correct number of regions in the header */
 636         lc->header.nr_regions = lc->region_count;
 637
 638         header_to_disk(&lc->header, lc->disk_header);
 639
 640         /* write the new header */
 641         r = rw_header(lc, WRITE);
 642         if (!r) {
 643                 r = flush_header(lc);
 644                 if (r)
 645                         lc->log_dev_flush_failed = 1;
 646         }
 647         if (r) {
 648                 DMWARN("%s: Failed to write header on dirty region log device",
 649                        lc->log_dev->name);
 650                 fail_log_device(lc);
 651         }
 652
 653         return r;
 654 }
 655
 656 static uint32_t core_get_region_size(struct dm_dirty_log *log)
 657 {
 658         struct log_c *lc = (struct log_c *) log->context;
 659         return lc->region_size;
 660 }
 661
 662 static int core_resume(struct dm_dirty_log *log)
 663 {
 664         struct log_c *lc = (struct log_c *) log->context;
 665         lc->sync_search = 0;
 666         return 0;
 667 }
 668
 669 static int core_is_clean(struct dm_dirty_log *log, region_t region)
 670 {
 671         struct log_c *lc = (struct log_c *) log->context;
 672         return log_test_bit(lc->clean_bits, region);
 673 }
 674
 675 static int core_in_sync(struct dm_dirty_log *log, region_t region, int block)
 676 {
 677         struct log_c *lc = (struct log_c *) log->context;
 678         return log_test_bit(lc->sync_bits, region);
 679 }
 680
 681 static int core_flush(struct dm_dirty_log *log)
 682 {
 683         /* no op */
 684         return 0;
 685 }
 686
 687 static int disk_flush(struct dm_dirty_log *log)
 688 {
 689         int r, i;
 690         struct log_c *lc = log->context;
 691
 692         /* only write if the log has changed */
 693         if (!lc->touched_cleaned && !lc->touched_dirtied)
 694                 return 0;
 695
 696         if (lc->touched_cleaned && log->flush_callback_fn &&
 697             log->flush_callback_fn(lc->ti)) {
 698                 /*
 699                  * At this point it is impossible to determine which
 700                  * regions are clean and which are dirty (without
 701                  * re-reading the log off disk). So mark all of them
 702                  * dirty.
 703                  */
 704                 lc->flush_failed = 1;
 705                 for (i = 0; i < lc->region_count; i++)
 706                         log_clear_bit(lc, lc->clean_bits, i);
 707         }
 708
 709         r = rw_header(lc, WRITE);
 710         if (r)
 711                 fail_log_device(lc);
 712         else {
 713                 if (lc->touched_dirtied) {
 714                         r = flush_header(lc);
 715                         if (r) {
 716                                 lc->log_dev_flush_failed = 1;
 717                                 fail_log_device(lc);
 718                         } else
 719                                 lc->touched_dirtied = 0;
 720                 }
 721                 lc->touched_cleaned = 0;
 722         }
 723
 724         return r;
 725 }
 726
 727 static void core_mark_region(struct dm_dirty_log *log, region_t region)
 728 {
 729         struct log_c *lc = (struct log_c *) log->context;
 730         log_clear_bit(lc, lc->clean_bits, region);
 731 }
 732
 733 static void core_clear_region(struct dm_dirty_log *log, region_t region)
 734 {
 735         struct log_c *lc = (struct log_c *) log->context;
 736         if (likely(!lc->flush_failed))
 737                 log_set_bit(lc, lc->clean_bits, region);
 738 }
 739
 740 static int core_get_resync_work(struct dm_dirty_log *log, region_t *region)
 741 {
 742         struct log_c *lc = (struct log_c *) log->context;
 743
 744         if (lc->sync_search >= lc->region_count)
 745                 return 0;
 746
 747         do {
 748                 *region = find_next_zero_bit_le(
 749                                              (unsigned long *) lc->sync_bits,
 750                                              lc->region_count,
 751                                              lc->sync_search);
 752                 lc->sync_search = *region + 1;
 753
 754                 if (*region >= lc->region_count)
 755                         return 0;
 756
 757         } while (log_test_bit(lc->recovering_bits, *region));
 758
 759         log_set_bit(lc, lc->recovering_bits, *region);
 760         return 1;
 761 }
 762
 763 static void core_set_region_sync(struct dm_dirty_log *log, region_t region,
 764                                  int in_sync)
 765 {
 766         struct log_c *lc = (struct log_c *) log->context;
 767
 768         log_clear_bit(lc, lc->recovering_bits, region);
 769         if (in_sync) {
 770                 log_set_bit(lc, lc->sync_bits, region);
 771                 lc->sync_count++;
 772         } else if (log_test_bit(lc->sync_bits, region)) {
 773                 lc->sync_count--;
 774                 log_clear_bit(lc, lc->sync_bits, region);
 775         }
 776 }
 777
 778 static region_t core_get_sync_count(struct dm_dirty_log *log)
 779 {
 780         struct log_c *lc = (struct log_c *) log->context;
 781
 782         return lc->sync_count;
 783 }
 784
 785 #define DMEMIT_SYNC \
 786         if (lc->sync != DEFAULTSYNC) \
 787                 DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "")
 788
 789 static int core_status(struct dm_dirty_log *log, status_type_t status,
 790                        char *result, unsigned int maxlen)
 791 {
 792         int sz = 0;
 793         struct log_c *lc = log->context;
 794
 795         switch(status) {
 796         case STATUSTYPE_INFO:
 797                 DMEMIT("1 %s", log->type->name);
 798                 break;
 799
 800         case STATUSTYPE_TABLE:
 801                 DMEMIT("%s %u %u ", log->type->name,
 802                        lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size);
 803                 DMEMIT_SYNC;
 804         }
 805
 806         return sz;
 807 }
 808
 809 static int disk_status(struct dm_dirty_log *log, status_type_t status,
 810                        char *result, unsigned int maxlen)
 811 {
 812         int sz = 0;
 813         struct log_c *lc = log->context;
 814
 815         switch(status) {
 816         case STATUSTYPE_INFO:
 817                 DMEMIT("3 %s %s %c", log->type->name, lc->log_dev->name,
 818                        lc->log_dev_flush_failed ? 'F' :
 819                        lc->log_dev_failed ? 'D' :
 820                        'A');
 821                 break;
 822
 823         case STATUSTYPE_TABLE:
 824                 DMEMIT("%s %u %s %u ", log->type->name,
 825                        lc->sync == DEFAULTSYNC ? 2 : 3, lc->log_dev->name,
 826                        lc->region_size);
 827                 DMEMIT_SYNC;
 828         }
 829
 830         return sz;
 831 }
 832
 833 static struct dm_dirty_log_type _core_type = {
 834         .name = "core",
 835         .module = THIS_MODULE,
 836         .ctr = core_ctr,
 837         .dtr = core_dtr,
 838         .resume = core_resume,
 839         .get_region_size = core_get_region_size,
 840         .is_clean = core_is_clean,
 841         .in_sync = core_in_sync,
 842         .flush = core_flush,
 843         .mark_region = core_mark_region,
 844         .clear_region = core_clear_region,
 845         .get_resync_work = core_get_resync_work,
 846         .set_region_sync = core_set_region_sync,
 847         .get_sync_count = core_get_sync_count,
 848         .status = core_status,
 849 };
 850
 851 static struct dm_dirty_log_type _disk_type = {
 852         .name = "disk",
 853         .module = THIS_MODULE,
 854         .ctr = disk_ctr,
 855         .dtr = disk_dtr,
 856         .postsuspend = disk_flush,
 857         .resume = disk_resume,
 858         .get_region_size = core_get_region_size,
 859         .is_clean = core_is_clean,
 860         .in_sync = core_in_sync,
 861         .flush = disk_flush,
 862         .mark_region = core_mark_region,
 863         .clear_region = core_clear_region,
 864         .get_resync_work = core_get_resync_work,
 865         .set_region_sync = core_set_region_sync,
 866         .get_sync_count = core_get_sync_count,
 867         .status = disk_status,
 868 };
 869
 870 static int __init dm_dirty_log_init(void)
 871 {
 872         int r;
 873
 874         r = dm_dirty_log_type_register(&_core_type);
 875         if (r)
 876                 DMWARN("couldn't register core log");
 877
 878         r = dm_dirty_log_type_register(&_disk_type);
 879         if (r) {
 880                 DMWARN("couldn't register disk type");
 881                 dm_dirty_log_type_unregister(&_core_type);
 882         }
 883
 884         return r;
 885 }
 886
 887 static void __exit dm_dirty_log_exit(void)
 888 {
 889         dm_dirty_log_type_unregister(&_disk_type);
 890         dm_dirty_log_type_unregister(&_core_type);
 891 }
 892
 893 module_init(dm_dirty_log_init);
 894 module_exit(dm_dirty_log_exit);
 895
 896 MODULE_DESCRIPTION(DM_NAME " dirty region log");
 897 MODULE_AUTHOR("Joe Thornber, Heinz Mauelshagen <dm-devel@redhat.com>");
 898 MODULE_LICENSE("GPL");