drivers/md/dm-raid.c

   1 /*
   2  * Copyright (C) 2010-2011 Neil Brown
   3  * Copyright (C) 2010-2011 Red Hat, Inc. All rights reserved.
   4  *
   5  * This file is released under the GPL.
   6  */
   7
   8 #include <linux/slab.h>
   9 #include <linux/module.h>
  10
  11 #include "md.h"
  12 #include "raid1.h"
  13 #include "raid5.h"
  14 #include "raid10.h"
  15 #include "bitmap.h"
  16
  17 #include <linux/device-mapper.h>
  18
  19 #define DM_MSG_PREFIX "raid"
  20
  21 /*
  22  * The following flags are used by dm-raid.c to set up the array state.
  23  * They must be cleared before md_run is called.
  24  */
  25 #define FirstUse 10             /* rdev flag */
  26
  27 struct raid_dev {
  28         /*
  29          * Two DM devices, one to hold metadata and one to hold the
  30          * actual data/parity.  The reason for this is to not confuse
  31          * ti->len and give more flexibility in altering size and
  32          * characteristics.
  33          *
  34          * While it is possible for this device to be associated
  35          * with a different physical device than the data_dev, it
  36          * is intended for it to be the same.
  37          *    |--------- Physical Device ---------|
  38          *    |- meta_dev -|------ data_dev ------|
  39          */
  40         struct dm_dev *meta_dev;
  41         struct dm_dev *data_dev;
  42         struct md_rdev rdev;
  43 };
  44
  45 /*
  46  * Flags for rs->print_flags field.
  47  */
  48 #define DMPF_SYNC              0x1
  49 #define DMPF_NOSYNC            0x2
  50 #define DMPF_REBUILD           0x4
  51 #define DMPF_DAEMON_SLEEP      0x8
  52 #define DMPF_MIN_RECOVERY_RATE 0x10
  53 #define DMPF_MAX_RECOVERY_RATE 0x20
  54 #define DMPF_MAX_WRITE_BEHIND  0x40
  55 #define DMPF_STRIPE_CACHE      0x80
  56 #define DMPF_REGION_SIZE       0x100
  57 #define DMPF_RAID10_COPIES     0x200
  58 #define DMPF_RAID10_FORMAT     0x400
  59
  60 struct raid_set {
  61         struct dm_target *ti;
  62
  63         uint32_t bitmap_loaded;
  64         uint32_t print_flags;
  65
  66         struct mddev md;
  67         struct raid_type *raid_type;
  68         struct dm_target_callbacks callbacks;
  69
  70         struct raid_dev dev[0];
  71 };
  72
  73 /* Supported raid types and properties. */
  74 static struct raid_type {
  75         const char *name;               /* RAID algorithm. */
  76         const char *descr;              /* Descriptor text for logging. */
  77         const unsigned parity_devs;     /* # of parity devices. */
  78         const unsigned minimal_devs;    /* minimal # of devices in set. */
  79         const unsigned level;           /* RAID level. */
  80         const unsigned algorithm;       /* RAID algorithm. */
  81 } raid_types[] = {
  82         {"raid1",    "RAID1 (mirroring)",               0, 2, 1, 0 /* NONE */},
  83         {"raid10",   "RAID10 (striped mirrors)",        0, 2, 10, UINT_MAX /* Varies */},
  84         {"raid4",    "RAID4 (dedicated parity disk)",   1, 2, 5, ALGORITHM_PARITY_0},
  85         {"raid5_la", "RAID5 (left asymmetric)",         1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
  86         {"raid5_ra", "RAID5 (right asymmetric)",        1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
  87         {"raid5_ls", "RAID5 (left symmetric)",          1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
  88         {"raid5_rs", "RAID5 (right symmetric)",         1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
  89         {"raid6_zr", "RAID6 (zero restart)",            2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
  90         {"raid6_nr", "RAID6 (N restart)",               2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
  91         {"raid6_nc", "RAID6 (N continue)",              2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE}
  92 };
  93
  94 static unsigned raid10_md_layout_to_copies(int layout)
  95 {
  96         return layout & 0xFF;
  97 }
  98
  99 static int raid10_format_to_md_layout(char *format, unsigned copies)
 100 {
 101         /* 1 "far" copy, and 'copies' "near" copies */
 102         return (1 << 8) | (copies & 0xFF);
 103 }
 104
 105 static struct raid_type *get_raid_type(char *name)
 106 {
 107         int i;
 108
 109         for (i = 0; i < ARRAY_SIZE(raid_types); i++)
 110                 if (!strcmp(raid_types[i].name, name))
 111                         return &raid_types[i];
 112
 113         return NULL;
 114 }
 115
 116 static struct raid_set *context_alloc(struct dm_target *ti, struct raid_type *raid_type, unsigned raid_devs)
 117 {
 118         unsigned i;
 119         struct raid_set *rs;
 120
 121         if (raid_devs <= raid_type->parity_devs) {
 122                 ti->error = "Insufficient number of devices";
 123                 return ERR_PTR(-EINVAL);
 124         }
 125
 126         rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
 127         if (!rs) {
 128                 ti->error = "Cannot allocate raid context";
 129                 return ERR_PTR(-ENOMEM);
 130         }
 131
 132         mddev_init(&rs->md);
 133
 134         rs->ti = ti;
 135         rs->raid_type = raid_type;
 136         rs->md.raid_disks = raid_devs;
 137         rs->md.level = raid_type->level;
 138         rs->md.new_level = rs->md.level;
 139         rs->md.layout = raid_type->algorithm;
 140         rs->md.new_layout = rs->md.layout;
 141         rs->md.delta_disks = 0;
 142         rs->md.recovery_cp = 0;
 143
 144         for (i = 0; i < raid_devs; i++)
 145                 md_rdev_init(&rs->dev[i].rdev);
 146
 147         /*
 148          * Remaining items to be initialized by further RAID params:
 149          *  rs->md.persistent
 150          *  rs->md.external
 151          *  rs->md.chunk_sectors
 152          *  rs->md.new_chunk_sectors
 153          *  rs->md.dev_sectors
 154          */
 155
 156         return rs;
 157 }
 158
 159 static void context_free(struct raid_set *rs)
 160 {
 161         int i;
 162
 163         for (i = 0; i < rs->md.raid_disks; i++) {
 164                 if (rs->dev[i].meta_dev)
 165                         dm_put_device(rs->ti, rs->dev[i].meta_dev);
 166                 md_rdev_clear(&rs->dev[i].rdev);
 167                 if (rs->dev[i].data_dev)
 168                         dm_put_device(rs->ti, rs->dev[i].data_dev);
 169         }
 170
 171         kfree(rs);
 172 }
 173
 174 /*
 175  * For every device we have two words
 176  *  <meta_dev>: meta device name or '-' if missing
 177  *  <data_dev>: data device name or '-' if missing
 178  *
 179  * The following are permitted:
 180  *    - -
 181  *    - <data_dev>
 182  *    <meta_dev> <data_dev>
 183  *
 184  * The following is not allowed:
 185  *    <meta_dev> -
 186  *
 187  * This code parses those words.  If there is a failure,
 188  * the caller must use context_free to unwind the operations.
 189  */
 190 static int dev_parms(struct raid_set *rs, char **argv)
 191 {
 192         int i;
 193         int rebuild = 0;
 194         int metadata_available = 0;
 195         int ret = 0;
 196
 197         for (i = 0; i < rs->md.raid_disks; i++, argv += 2) {
 198                 rs->dev[i].rdev.raid_disk = i;
 199
 200                 rs->dev[i].meta_dev = NULL;
 201                 rs->dev[i].data_dev = NULL;
 202
 203                 /*
 204                  * There are no offsets, since there is a separate device
 205                  * for data and metadata.
 206                  */
 207                 rs->dev[i].rdev.data_offset = 0;
 208                 rs->dev[i].rdev.mddev = &rs->md;
 209
 210                 if (strcmp(argv[0], "-")) {
 211                         ret = dm_get_device(rs->ti, argv[0],
 212                                             dm_table_get_mode(rs->ti->table),
 213                                             &rs->dev[i].meta_dev);
 214                         rs->ti->error = "RAID metadata device lookup failure";
 215                         if (ret)
 216                                 return ret;
 217
 218                         rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
 219                         if (!rs->dev[i].rdev.sb_page)
 220                                 return -ENOMEM;
 221                 }
 222
 223                 if (!strcmp(argv[1], "-")) {
 224                         if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
 225                             (!rs->dev[i].rdev.recovery_offset)) {
 226                                 rs->ti->error = "Drive designated for rebuild not specified";
 227                                 return -EINVAL;
 228                         }
 229
 230                         rs->ti->error = "No data device supplied with metadata device";
 231                         if (rs->dev[i].meta_dev)
 232                                 return -EINVAL;
 233
 234                         continue;
 235                 }
 236
 237                 ret = dm_get_device(rs->ti, argv[1],
 238                                     dm_table_get_mode(rs->ti->table),
 239                                     &rs->dev[i].data_dev);
 240                 if (ret) {
 241                         rs->ti->error = "RAID device lookup failure";
 242                         return ret;
 243                 }
 244
 245                 if (rs->dev[i].meta_dev) {
 246                         metadata_available = 1;
 247                         rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
 248                 }
 249                 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
 250                 list_add(&rs->dev[i].rdev.same_set, &rs->md.disks);
 251                 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
 252                         rebuild++;
 253         }
 254
 255         if (metadata_available) {
 256                 rs->md.external = 0;
 257                 rs->md.persistent = 1;
 258                 rs->md.major_version = 2;
 259         } else if (rebuild && !rs->md.recovery_cp) {
 260                 /*
 261                  * Without metadata, we will not be able to tell if the array
 262                  * is in-sync or not - we must assume it is not.  Therefore,
 263                  * it is impossible to rebuild a drive.
 264                  *
 265                  * Even if there is metadata, the on-disk information may
 266                  * indicate that the array is not in-sync and it will then
 267                  * fail at that time.
 268                  *
 269                  * User could specify 'nosync' option if desperate.
 270                  */
 271                 DMERR("Unable to rebuild drive while array is not in-sync");
 272                 rs->ti->error = "RAID device lookup failure";
 273                 return -EINVAL;
 274         }
 275
 276         return 0;
 277 }
 278
 279 /*
 280  * validate_region_size
 281  * @rs
 282  * @region_size:  region size in sectors.  If 0, pick a size (4MiB default).
 283  *
 284  * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
 285  * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
 286  *
 287  * Returns: 0 on success, -EINVAL on failure.
 288  */
 289 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
 290 {
 291         unsigned long min_region_size = rs->ti->len / (1 << 21);
 292
 293         if (!region_size) {
 294                 /*
 295                  * Choose a reasonable default.  All figures in sectors.
 296                  */
 297                 if (min_region_size > (1 << 13)) {
 298                         /* If not a power of 2, make it the next power of 2 */
 299                         if (min_region_size & (min_region_size - 1))
 300                                 region_size = 1 << fls(region_size);
 301                         DMINFO("Choosing default region size of %lu sectors",
 302                                region_size);
 303                 } else {
 304                         DMINFO("Choosing default region size of 4MiB");
 305                         region_size = 1 << 13; /* sectors */
 306                 }
 307         } else {
 308                 /*
 309                  * Validate user-supplied value.
 310                  */
 311                 if (region_size > rs->ti->len) {
 312                         rs->ti->error = "Supplied region size is too large";
 313                         return -EINVAL;
 314                 }
 315
 316                 if (region_size < min_region_size) {
 317                         DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
 318                               region_size, min_region_size);
 319                         rs->ti->error = "Supplied region size is too small";
 320                         return -EINVAL;
 321                 }
 322
 323                 if (!is_power_of_2(region_size)) {
 324                         rs->ti->error = "Region size is not a power of 2";
 325                         return -EINVAL;
 326                 }
 327
 328                 if (region_size < rs->md.chunk_sectors) {
 329                         rs->ti->error = "Region size is smaller than the chunk size";
 330                         return -EINVAL;
 331                 }
 332         }
 333
 334         /*
 335          * Convert sectors to bytes.
 336          */
 337         rs->md.bitmap_info.chunksize = (region_size << 9);
 338
 339         return 0;
 340 }
 341
 342 /*
 343  * validate_rebuild_devices
 344  * @rs
 345  *
 346  * Determine if the devices specified for rebuild can result in a valid
 347  * usable array that is capable of rebuilding the given devices.
 348  *
 349  * Returns: 0 on success, -EINVAL on failure.
 350  */
 351 static int validate_rebuild_devices(struct raid_set *rs)
 352 {
 353         unsigned i, rebuild_cnt = 0;
 354         unsigned rebuilds_per_group, copies, d;
 355
 356         if (!(rs->print_flags & DMPF_REBUILD))
 357                 return 0;
 358
 359         for (i = 0; i < rs->md.raid_disks; i++)
 360                 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
 361                         rebuild_cnt++;
 362
 363         switch (rs->raid_type->level) {
 364         case 1:
 365                 if (rebuild_cnt >= rs->md.raid_disks)
 366                         goto too_many;
 367                 break;
 368         case 4:
 369         case 5:
 370         case 6:
 371                 if (rebuild_cnt > rs->raid_type->parity_devs)
 372                         goto too_many;
 373                 break;
 374         case 10:
 375                 copies = raid10_md_layout_to_copies(rs->md.layout);
 376                 if (rebuild_cnt < copies)
 377                         break;
 378
 379                 /*
 380                  * It is possible to have a higher rebuild count for RAID10,
 381                  * as long as the failed devices occur in different mirror
 382                  * groups (i.e. different stripes).
 383                  *
 384                  * Right now, we only allow for "near" copies.  When other
 385                  * formats are added, we will have to check those too.
 386                  *
 387                  * When checking "near" format, make sure no adjacent devices
 388                  * have failed beyond what can be handled.  In addition to the
 389                  * simple case where the number of devices is a multiple of the
 390                  * number of copies, we must also handle cases where the number
 391                  * of devices is not a multiple of the number of copies.
 392                  * E.g.    dev1 dev2 dev3 dev4 dev5
 393                  *          A    A    B    B    C
 394                  *          C    D    D    E    E
 395                  */
 396                 rebuilds_per_group = 0;
 397                 for (i = 0; i < rs->md.raid_disks * copies; i++) {
 398                         d = i % rs->md.raid_disks;
 399                         if (!test_bit(In_sync, &rs->dev[d].rdev.flags) &&
 400                             (++rebuilds_per_group >= copies))
 401                                 goto too_many;
 402                         if (!((i + 1) % copies))
 403                                 rebuilds_per_group = 0;
 404                 }
 405                 break;
 406         default:
 407                 DMERR("The rebuild parameter is not supported for %s",
 408                       rs->raid_type->name);
 409                 rs->ti->error = "Rebuild not supported for this RAID type";
 410                 return -EINVAL;
 411         }
 412
 413         return 0;
 414
 415 too_many:
 416         rs->ti->error = "Too many rebuild devices specified";
 417         return -EINVAL;
 418 }
 419
 420 /*
 421  * Possible arguments are...
 422  *      <chunk_size> [optional_args]
 423  *
 424  * Argument definitions
 425  *    <chunk_size>                      The number of sectors per disk that
 426  *                                      will form the "stripe"
 427  *    [[no]sync]                        Force or prevent recovery of the
 428  *                                      entire array
 429  *    [rebuild <idx>]                   Rebuild the drive indicated by the index
 430  *    [daemon_sleep <ms>]               Time between bitmap daemon work to
 431  *                                      clear bits
 432  *    [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
 433  *    [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
 434  *    [write_mostly <idx>]              Indicate a write mostly drive via index
 435  *    [max_write_behind <sectors>]      See '-write-behind=' (man mdadm)
 436  *    [stripe_cache <sectors>]          Stripe cache size for higher RAIDs
 437  *    [region_size <sectors>]           Defines granularity of bitmap
 438  *
 439  * RAID10-only options:
 440  *    [raid10_copies <# copies>]        Number of copies.  (Default: 2)
 441  *    [raid10_format <near>]            Layout algorithm.  (Default: near)
 442  */
 443 static int parse_raid_params(struct raid_set *rs, char **argv,
 444                              unsigned num_raid_params)
 445 {
 446         char *raid10_format = "near";
 447         unsigned raid10_copies = 2;
 448         unsigned i;
 449         unsigned long value, region_size = 0;
 450         sector_t sectors_per_dev = rs->ti->len;
 451         sector_t max_io_len;
 452         char *key;
 453
 454         /*
 455          * First, parse the in-order required arguments
 456          * "chunk_size" is the only argument of this type.
 457          */
 458         if ((strict_strtoul(argv[0], 10, &value) < 0)) {
 459                 rs->ti->error = "Bad chunk size";
 460                 return -EINVAL;
 461         } else if (rs->raid_type->level == 1) {
 462                 if (value)
 463                         DMERR("Ignoring chunk size parameter for RAID 1");
 464                 value = 0;
 465         } else if (!is_power_of_2(value)) {
 466                 rs->ti->error = "Chunk size must be a power of 2";
 467                 return -EINVAL;
 468         } else if (value < 8) {
 469                 rs->ti->error = "Chunk size value is too small";
 470                 return -EINVAL;
 471         }
 472
 473         rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
 474         argv++;
 475         num_raid_params--;
 476
 477         /*
 478          * We set each individual device as In_sync with a completed
 479          * 'recovery_offset'.  If there has been a device failure or
 480          * replacement then one of the following cases applies:
 481          *
 482          *   1) User specifies 'rebuild'.
 483          *      - Device is reset when param is read.
 484          *   2) A new device is supplied.
 485          *      - No matching superblock found, resets device.
 486          *   3) Device failure was transient and returns on reload.
 487          *      - Failure noticed, resets device for bitmap replay.
 488          *   4) Device hadn't completed recovery after previous failure.
 489          *      - Superblock is read and overrides recovery_offset.
 490          *
 491          * What is found in the superblocks of the devices is always
 492          * authoritative, unless 'rebuild' or '[no]sync' was specified.
 493          */
 494         for (i = 0; i < rs->md.raid_disks; i++) {
 495                 set_bit(In_sync, &rs->dev[i].rdev.flags);
 496                 rs->dev[i].rdev.recovery_offset = MaxSector;
 497         }
 498
 499         /*
 500          * Second, parse the unordered optional arguments
 501          */
 502         for (i = 0; i < num_raid_params; i++) {
 503                 if (!strcasecmp(argv[i], "nosync")) {
 504                         rs->md.recovery_cp = MaxSector;
 505                         rs->print_flags |= DMPF_NOSYNC;
 506                         continue;
 507                 }
 508                 if (!strcasecmp(argv[i], "sync")) {
 509                         rs->md.recovery_cp = 0;
 510                         rs->print_flags |= DMPF_SYNC;
 511                         continue;
 512                 }
 513
 514                 /* The rest of the optional arguments come in key/value pairs */
 515                 if ((i + 1) >= num_raid_params) {
 516                         rs->ti->error = "Wrong number of raid parameters given";
 517                         return -EINVAL;
 518                 }
 519
 520                 key = argv[i++];
 521
 522                 /* Parameters that take a string value are checked here. */
 523                 if (!strcasecmp(key, "raid10_format")) {
 524                         if (rs->raid_type->level != 10) {
 525                                 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
 526                                 return -EINVAL;
 527                         }
 528                         if (strcmp("near", argv[i])) {
 529                                 rs->ti->error = "Invalid 'raid10_format' value given";
 530                                 return -EINVAL;
 531                         }
 532                         raid10_format = argv[i];
 533                         rs->print_flags |= DMPF_RAID10_FORMAT;
 534                         continue;
 535                 }
 536
 537                 if (strict_strtoul(argv[i], 10, &value) < 0) {
 538                         rs->ti->error = "Bad numerical argument given in raid params";
 539                         return -EINVAL;
 540                 }
 541
 542                 /* Parameters that take a numeric value are checked here */
 543                 if (!strcasecmp(key, "rebuild")) {
 544                         if (value >= rs->md.raid_disks) {
 545                                 rs->ti->error = "Invalid rebuild index given";
 546                                 return -EINVAL;
 547                         }
 548                         clear_bit(In_sync, &rs->dev[value].rdev.flags);
 549                         rs->dev[value].rdev.recovery_offset = 0;
 550                         rs->print_flags |= DMPF_REBUILD;
 551                 } else if (!strcasecmp(key, "write_mostly")) {
 552                         if (rs->raid_type->level != 1) {
 553                                 rs->ti->error = "write_mostly option is only valid for RAID1";
 554                                 return -EINVAL;
 555                         }
 556                         if (value >= rs->md.raid_disks) {
 557                                 rs->ti->error = "Invalid write_mostly drive index given";
 558                                 return -EINVAL;
 559                         }
 560                         set_bit(WriteMostly, &rs->dev[value].rdev.flags);
 561                 } else if (!strcasecmp(key, "max_write_behind")) {
 562                         if (rs->raid_type->level != 1) {
 563                                 rs->ti->error = "max_write_behind option is only valid for RAID1";
 564                                 return -EINVAL;
 565                         }
 566                         rs->print_flags |= DMPF_MAX_WRITE_BEHIND;
 567
 568                         /*
 569                          * In device-mapper, we specify things in sectors, but
 570                          * MD records this value in kB
 571                          */
 572                         value /= 2;
 573                         if (value > COUNTER_MAX) {
 574                                 rs->ti->error = "Max write-behind limit out of range";
 575                                 return -EINVAL;
 576                         }
 577                         rs->md.bitmap_info.max_write_behind = value;
 578                 } else if (!strcasecmp(key, "daemon_sleep")) {
 579                         rs->print_flags |= DMPF_DAEMON_SLEEP;
 580                         if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
 581                                 rs->ti->error = "daemon sleep period out of range";
 582                                 return -EINVAL;
 583                         }
 584                         rs->md.bitmap_info.daemon_sleep = value;
 585                 } else if (!strcasecmp(key, "stripe_cache")) {
 586                         rs->print_flags |= DMPF_STRIPE_CACHE;
 587
 588                         /*
 589                          * In device-mapper, we specify things in sectors, but
 590                          * MD records this value in kB
 591                          */
 592                         value /= 2;
 593
 594                         if ((rs->raid_type->level != 5) &&
 595                             (rs->raid_type->level != 6)) {
 596                                 rs->ti->error = "Inappropriate argument: stripe_cache";
 597                                 return -EINVAL;
 598                         }
 599                         if (raid5_set_cache_size(&rs->md, (int)value)) {
 600                                 rs->ti->error = "Bad stripe_cache size";
 601                                 return -EINVAL;
 602                         }
 603                 } else if (!strcasecmp(key, "min_recovery_rate")) {
 604                         rs->print_flags |= DMPF_MIN_RECOVERY_RATE;
 605                         if (value > INT_MAX) {
 606                                 rs->ti->error = "min_recovery_rate out of range";
 607                                 return -EINVAL;
 608                         }
 609                         rs->md.sync_speed_min = (int)value;
 610                 } else if (!strcasecmp(key, "max_recovery_rate")) {
 611                         rs->print_flags |= DMPF_MAX_RECOVERY_RATE;
 612                         if (value > INT_MAX) {
 613                                 rs->ti->error = "max_recovery_rate out of range";
 614                                 return -EINVAL;
 615                         }
 616                         rs->md.sync_speed_max = (int)value;
 617                 } else if (!strcasecmp(key, "region_size")) {
 618                         rs->print_flags |= DMPF_REGION_SIZE;
 619                         region_size = value;
 620                 } else if (!strcasecmp(key, "raid10_copies") &&
 621                            (rs->raid_type->level == 10)) {
 622                         if ((value < 2) || (value > 0xFF)) {
 623                                 rs->ti->error = "Bad value for 'raid10_copies'";
 624                                 return -EINVAL;
 625                         }
 626                         rs->print_flags |= DMPF_RAID10_COPIES;
 627                         raid10_copies = value;
 628                 } else {
 629                         DMERR("Unable to parse RAID parameter: %s", key);
 630                         rs->ti->error = "Unable to parse RAID parameters";
 631                         return -EINVAL;
 632                 }
 633         }
 634
 635         if (validate_region_size(rs, region_size))
 636                 return -EINVAL;
 637
 638         if (rs->md.chunk_sectors)
 639                 max_io_len = rs->md.chunk_sectors;
 640         else
 641                 max_io_len = region_size;
 642
 643         if (dm_set_target_max_io_len(rs->ti, max_io_len))
 644                 return -EINVAL;
 645
 646         if (rs->raid_type->level == 10) {
 647                 if (raid10_copies > rs->md.raid_disks) {
 648                         rs->ti->error = "Not enough devices to satisfy specification";
 649                         return -EINVAL;
 650                 }
 651
 652                 /* (Len * #mirrors) / #devices */
 653                 sectors_per_dev = rs->ti->len * raid10_copies;
 654                 sector_div(sectors_per_dev, rs->md.raid_disks);
 655
 656                 rs->md.layout = raid10_format_to_md_layout(raid10_format,
 657                                                            raid10_copies);
 658                 rs->md.new_layout = rs->md.layout;
 659         } else if ((rs->raid_type->level > 1) &&
 660                    sector_div(sectors_per_dev,
 661                               (rs->md.raid_disks - rs->raid_type->parity_devs))) {
 662                 rs->ti->error = "Target length not divisible by number of data devices";
 663                 return -EINVAL;
 664         }
 665         rs->md.dev_sectors = sectors_per_dev;
 666
 667         if (validate_rebuild_devices(rs))
 668                 return -EINVAL;
 669
 670         /* Assume there are no metadata devices until the drives are parsed */
 671         rs->md.persistent = 0;
 672         rs->md.external = 1;
 673
 674         return 0;
 675 }
 676
 677 static void do_table_event(struct work_struct *ws)
 678 {
 679         struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
 680
 681         dm_table_event(rs->ti->table);
 682 }
 683
 684 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
 685 {
 686         struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
 687
 688         if (rs->raid_type->level == 1)
 689                 return md_raid1_congested(&rs->md, bits);
 690
 691         if (rs->raid_type->level == 10)
 692                 return md_raid10_congested(&rs->md, bits);
 693
 694         return md_raid5_congested(&rs->md, bits);
 695 }
 696
 697 /*
 698  * This structure is never routinely used by userspace, unlike md superblocks.
 699  * Devices with this superblock should only ever be accessed via device-mapper.
 700  */
 701 #define DM_RAID_MAGIC 0x64526D44
 702 struct dm_raid_superblock {
 703         __le32 magic;           /* "DmRd" */
 704         __le32 features;        /* Used to indicate possible future changes */
 705
 706         __le32 num_devices;     /* Number of devices in this array. (Max 64) */
 707         __le32 array_position;  /* The position of this drive in the array */
 708
 709         __le64 events;          /* Incremented by md when superblock updated */
 710         __le64 failed_devices;  /* Bit field of devices to indicate failures */
 711
 712         /*
 713          * This offset tracks the progress of the repair or replacement of
 714          * an individual drive.
 715          */
 716         __le64 disk_recovery_offset;
 717
 718         /*
 719          * This offset tracks the progress of the initial array
 720          * synchronisation/parity calculation.
 721          */
 722         __le64 array_resync_offset;
 723
 724         /*
 725          * RAID characteristics
 726          */
 727         __le32 level;
 728         __le32 layout;
 729         __le32 stripe_sectors;
 730
 731         __u8 pad[452];          /* Round struct to 512 bytes. */
 732                                 /* Always set to 0 when writing. */
 733 } __packed;
 734
 735 static int read_disk_sb(struct md_rdev *rdev, int size)
 736 {
 737         BUG_ON(!rdev->sb_page);
 738
 739         if (rdev->sb_loaded)
 740                 return 0;
 741
 742         if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, 1)) {
 743                 DMERR("Failed to read superblock of device at position %d",
 744                       rdev->raid_disk);
 745                 md_error(rdev->mddev, rdev);
 746                 return -EINVAL;
 747         }
 748
 749         rdev->sb_loaded = 1;
 750
 751         return 0;
 752 }
 753
 754 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
 755 {
 756         int i;
 757         uint64_t failed_devices;
 758         struct dm_raid_superblock *sb;
 759         struct raid_set *rs = container_of(mddev, struct raid_set, md);
 760
 761         sb = page_address(rdev->sb_page);
 762         failed_devices = le64_to_cpu(sb->failed_devices);
 763
 764         for (i = 0; i < mddev->raid_disks; i++)
 765                 if (!rs->dev[i].data_dev ||
 766                     test_bit(Faulty, &(rs->dev[i].rdev.flags)))
 767                         failed_devices |= (1ULL << i);
 768
 769         memset(sb, 0, sizeof(*sb));
 770
 771         sb->magic = cpu_to_le32(DM_RAID_MAGIC);
 772         sb->features = cpu_to_le32(0);  /* No features yet */
 773
 774         sb->num_devices = cpu_to_le32(mddev->raid_disks);
 775         sb->array_position = cpu_to_le32(rdev->raid_disk);
 776
 777         sb->events = cpu_to_le64(mddev->events);
 778         sb->failed_devices = cpu_to_le64(failed_devices);
 779
 780         sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
 781         sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
 782
 783         sb->level = cpu_to_le32(mddev->level);
 784         sb->layout = cpu_to_le32(mddev->layout);
 785         sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
 786 }
 787
 788 /*
 789  * super_load
 790  *
 791  * This function creates a superblock if one is not found on the device
 792  * and will decide which superblock to use if there's a choice.
 793  *
 794  * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
 795  */
 796 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
 797 {
 798         int ret;
 799         struct dm_raid_superblock *sb;
 800         struct dm_raid_superblock *refsb;
 801         uint64_t events_sb, events_refsb;
 802
 803         rdev->sb_start = 0;
 804         rdev->sb_size = sizeof(*sb);
 805
 806         ret = read_disk_sb(rdev, rdev->sb_size);
 807         if (ret)
 808                 return ret;
 809
 810         sb = page_address(rdev->sb_page);
 811
 812         /*
 813          * Two cases that we want to write new superblocks and rebuild:
 814          * 1) New device (no matching magic number)
 815          * 2) Device specified for rebuild (!In_sync w/ offset == 0)
 816          */
 817         if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
 818             (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
 819                 super_sync(rdev->mddev, rdev);
 820
 821                 set_bit(FirstUse, &rdev->flags);
 822
 823                 /* Force writing of superblocks to disk */
 824                 set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags);
 825
 826                 /* Any superblock is better than none, choose that if given */
 827                 return refdev ? 0 : 1;
 828         }
 829
 830         if (!refdev)
 831                 return 1;
 832
 833         events_sb = le64_to_cpu(sb->events);
 834
 835         refsb = page_address(refdev->sb_page);
 836         events_refsb = le64_to_cpu(refsb->events);
 837
 838         return (events_sb > events_refsb) ? 1 : 0;
 839 }
 840
 841 static int super_init_validation(struct mddev *mddev, struct md_rdev *rdev)
 842 {
 843         int role;
 844         struct raid_set *rs = container_of(mddev, struct raid_set, md);
 845         uint64_t events_sb;
 846         uint64_t failed_devices;
 847         struct dm_raid_superblock *sb;
 848         uint32_t new_devs = 0;
 849         uint32_t rebuilds = 0;
 850         struct md_rdev *r;
 851         struct dm_raid_superblock *sb2;
 852
 853         sb = page_address(rdev->sb_page);
 854         events_sb = le64_to_cpu(sb->events);
 855         failed_devices = le64_to_cpu(sb->failed_devices);
 856
 857         /*
 858          * Initialise to 1 if this is a new superblock.
 859          */
 860         mddev->events = events_sb ? : 1;
 861
 862         /*
 863          * Reshaping is not currently allowed
 864          */
 865         if ((le32_to_cpu(sb->level) != mddev->level) ||
 866             (le32_to_cpu(sb->layout) != mddev->layout) ||
 867             (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors)) {
 868                 DMERR("Reshaping arrays not yet supported.");
 869                 return -EINVAL;
 870         }
 871
 872         /* We can only change the number of devices in RAID1 right now */
 873         if ((rs->raid_type->level != 1) &&
 874             (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) {
 875                 DMERR("Reshaping arrays not yet supported.");
 876                 return -EINVAL;
 877         }
 878
 879         if (!(rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC)))
 880                 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
 881
 882         /*
 883          * During load, we set FirstUse if a new superblock was written.
 884          * There are two reasons we might not have a superblock:
 885          * 1) The array is brand new - in which case, all of the
 886          *    devices must have their In_sync bit set.  Also,
 887          *    recovery_cp must be 0, unless forced.
 888          * 2) This is a new device being added to an old array
 889          *    and the new device needs to be rebuilt - in which
 890          *    case the In_sync bit will /not/ be set and
 891          *    recovery_cp must be MaxSector.
 892          */
 893         rdev_for_each(r, mddev) {
 894                 if (!test_bit(In_sync, &r->flags)) {
 895                         DMINFO("Device %d specified for rebuild: "
 896                                "Clearing superblock", r->raid_disk);
 897                         rebuilds++;
 898                 } else if (test_bit(FirstUse, &r->flags))
 899                         new_devs++;
 900         }
 901
 902         if (!rebuilds) {
 903                 if (new_devs == mddev->raid_disks) {
 904                         DMINFO("Superblocks created for new array");
 905                         set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
 906                 } else if (new_devs) {
 907                         DMERR("New device injected "
 908                               "into existing array without 'rebuild' "
 909                               "parameter specified");
 910                         return -EINVAL;
 911                 }
 912         } else if (new_devs) {
 913                 DMERR("'rebuild' devices cannot be "
 914                       "injected into an array with other first-time devices");
 915                 return -EINVAL;
 916         } else if (mddev->recovery_cp != MaxSector) {
 917                 DMERR("'rebuild' specified while array is not in-sync");
 918                 return -EINVAL;
 919         }
 920
 921         /*
 922          * Now we set the Faulty bit for those devices that are
 923          * recorded in the superblock as failed.
 924          */
 925         rdev_for_each(r, mddev) {
 926                 if (!r->sb_page)
 927                         continue;
 928                 sb2 = page_address(r->sb_page);
 929                 sb2->failed_devices = 0;
 930
 931                 /*
 932                  * Check for any device re-ordering.
 933                  */
 934                 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
 935                         role = le32_to_cpu(sb2->array_position);
 936                         if (role != r->raid_disk) {
 937                                 if (rs->raid_type->level != 1) {
 938                                         rs->ti->error = "Cannot change device "
 939                                                 "positions in RAID array";
 940                                         return -EINVAL;
 941                                 }
 942                                 DMINFO("RAID1 device #%d now at position #%d",
 943                                        role, r->raid_disk);
 944                         }
 945
 946                         /*
 947                          * Partial recovery is performed on
 948                          * returning failed devices.
 949                          */
 950                         if (failed_devices & (1 << role))
 951                                 set_bit(Faulty, &r->flags);
 952                 }
 953         }
 954
 955         return 0;
 956 }
 957
 958 static int super_validate(struct mddev *mddev, struct md_rdev *rdev)
 959 {
 960         struct dm_raid_superblock *sb = page_address(rdev->sb_page);
 961
 962         /*
 963          * If mddev->events is not set, we know we have not yet initialized
 964          * the array.
 965          */
 966         if (!mddev->events && super_init_validation(mddev, rdev))
 967                 return -EINVAL;
 968
 969         mddev->bitmap_info.offset = 4096 >> 9; /* Enable bitmap creation */
 970         rdev->mddev->bitmap_info.default_offset = 4096 >> 9;
 971         if (!test_bit(FirstUse, &rdev->flags)) {
 972                 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
 973                 if (rdev->recovery_offset != MaxSector)
 974                         clear_bit(In_sync, &rdev->flags);
 975         }
 976
 977         /*
 978          * If a device comes back, set it as not In_sync and no longer faulty.
 979          */
 980         if (test_bit(Faulty, &rdev->flags)) {
 981                 clear_bit(Faulty, &rdev->flags);
 982                 clear_bit(In_sync, &rdev->flags);
 983                 rdev->saved_raid_disk = rdev->raid_disk;
 984                 rdev->recovery_offset = 0;
 985         }
 986
 987         clear_bit(FirstUse, &rdev->flags);
 988
 989         return 0;
 990 }
 991
 992 /*
 993  * Analyse superblocks and select the freshest.
 994  */
 995 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
 996 {
 997         int ret;
 998         unsigned redundancy = 0;
 999         struct raid_dev *dev;
1000         struct md_rdev *rdev, *tmp, *freshest;
1001         struct mddev *mddev = &rs->md;
1002
1003         switch (rs->raid_type->level) {
1004         case 1:
1005                 redundancy = rs->md.raid_disks - 1;
1006                 break;
1007         case 4:
1008         case 5:
1009         case 6:
1010                 redundancy = rs->raid_type->parity_devs;
1011                 break;
1012         case 10:
1013                 redundancy = raid10_md_layout_to_copies(mddev->layout) - 1;
1014                 break;
1015         default:
1016                 ti->error = "Unknown RAID type";
1017                 return -EINVAL;
1018         }
1019
1020         freshest = NULL;
1021         rdev_for_each_safe(rdev, tmp, mddev) {
1022                 /*
1023                  * Skipping super_load due to DMPF_SYNC will cause
1024                  * the array to undergo initialization again as
1025                  * though it were new.  This is the intended effect
1026                  * of the "sync" directive.
1027                  *
1028                  * When reshaping capability is added, we must ensure
1029                  * that the "sync" directive is disallowed during the
1030                  * reshape.
1031                  */
1032                 if (rs->print_flags & DMPF_SYNC)
1033                         continue;
1034
1035                 if (!rdev->meta_bdev)
1036                         continue;
1037
1038                 ret = super_load(rdev, freshest);
1039
1040                 switch (ret) {
1041                 case 1:
1042                         freshest = rdev;
1043                         break;
1044                 case 0:
1045                         break;
1046                 default:
1047                         dev = container_of(rdev, struct raid_dev, rdev);
1048                         if (redundancy--) {
1049                                 if (dev->meta_dev)
1050                                         dm_put_device(ti, dev->meta_dev);
1051
1052                                 dev->meta_dev = NULL;
1053                                 rdev->meta_bdev = NULL;
1054
1055                                 if (rdev->sb_page)
1056                                         put_page(rdev->sb_page);
1057
1058                                 rdev->sb_page = NULL;
1059
1060                                 rdev->sb_loaded = 0;
1061
1062                                 /*
1063                                  * We might be able to salvage the data device
1064                                  * even though the meta device has failed.  For
1065                                  * now, we behave as though '- -' had been
1066                                  * set for this device in the table.
1067                                  */
1068                                 if (dev->data_dev)
1069                                         dm_put_device(ti, dev->data_dev);
1070
1071                                 dev->data_dev = NULL;
1072                                 rdev->bdev = NULL;
1073
1074                                 list_del(&rdev->same_set);
1075
1076                                 continue;
1077                         }
1078                         ti->error = "Failed to load superblock";
1079                         return ret;
1080                 }
1081         }
1082
1083         if (!freshest)
1084                 return 0;
1085
1086         /*
1087          * Validation of the freshest device provides the source of
1088          * validation for the remaining devices.
1089          */
1090         ti->error = "Unable to assemble array: Invalid superblocks";
1091         if (super_validate(mddev, freshest))
1092                 return -EINVAL;
1093
1094         rdev_for_each(rdev, mddev)
1095                 if ((rdev != freshest) && super_validate(mddev, rdev))
1096                         return -EINVAL;
1097
1098         return 0;
1099 }
1100
1101 /*
1102  * Construct a RAID4/5/6 mapping:
1103  * Args:
1104  *      <raid_type> <#raid_params> <raid_params>                \
1105  *      <#raid_devs> { <meta_dev1> <dev1> .. <meta_devN> <devN> }
1106  *
1107  * <raid_params> varies by <raid_type>.  See 'parse_raid_params' for
1108  * details on possible <raid_params>.
1109  */
1110 static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv)
1111 {
1112         int ret;
1113         struct raid_type *rt;
1114         unsigned long num_raid_params, num_raid_devs;
1115         struct raid_set *rs = NULL;
1116
1117         /* Must have at least <raid_type> <#raid_params> */
1118         if (argc < 2) {
1119                 ti->error = "Too few arguments";
1120                 return -EINVAL;
1121         }
1122
1123         /* raid type */
1124         rt = get_raid_type(argv[0]);
1125         if (!rt) {
1126                 ti->error = "Unrecognised raid_type";
1127                 return -EINVAL;
1128         }
1129         argc--;
1130         argv++;
1131
1132         /* number of RAID parameters */
1133         if (strict_strtoul(argv[0], 10, &num_raid_params) < 0) {
1134                 ti->error = "Cannot understand number of RAID parameters";
1135                 return -EINVAL;
1136         }
1137         argc--;
1138         argv++;
1139
1140         /* Skip over RAID params for now and find out # of devices */
1141         if (num_raid_params + 1 > argc) {
1142                 ti->error = "Arguments do not agree with counts given";
1143                 return -EINVAL;
1144         }
1145
1146         if ((strict_strtoul(argv[num_raid_params], 10, &num_raid_devs) < 0) ||
1147             (num_raid_devs >= INT_MAX)) {
1148                 ti->error = "Cannot understand number of raid devices";
1149                 return -EINVAL;
1150         }
1151
1152         rs = context_alloc(ti, rt, (unsigned)num_raid_devs);
1153         if (IS_ERR(rs))
1154                 return PTR_ERR(rs);
1155
1156         ret = parse_raid_params(rs, argv, (unsigned)num_raid_params);
1157         if (ret)
1158                 goto bad;
1159
1160         ret = -EINVAL;
1161
1162         argc -= num_raid_params + 1; /* +1: we already have num_raid_devs */
1163         argv += num_raid_params + 1;
1164
1165         if (argc != (num_raid_devs * 2)) {
1166                 ti->error = "Supplied RAID devices does not match the count given";
1167                 goto bad;
1168         }
1169
1170         ret = dev_parms(rs, argv);
1171         if (ret)
1172                 goto bad;
1173
1174         rs->md.sync_super = super_sync;
1175         ret = analyse_superblocks(ti, rs);
1176         if (ret)
1177                 goto bad;
1178
1179         INIT_WORK(&rs->md.event_work, do_table_event);
1180         ti->private = rs;
1181         ti->num_flush_requests = 1;
1182
1183         mutex_lock(&rs->md.reconfig_mutex);
1184         ret = md_run(&rs->md);
1185         rs->md.in_sync = 0; /* Assume already marked dirty */
1186         mutex_unlock(&rs->md.reconfig_mutex);
1187
1188         if (ret) {
1189                 ti->error = "Fail to run raid array";
1190                 goto bad;
1191         }
1192
1193         if (ti->len != rs->md.array_sectors) {
1194                 ti->error = "Array size does not match requested target length";
1195                 ret = -EINVAL;
1196                 goto size_mismatch;
1197         }
1198         rs->callbacks.congested_fn = raid_is_congested;
1199         dm_table_add_target_callbacks(ti->table, &rs->callbacks);
1200
1201         mddev_suspend(&rs->md);
1202         return 0;
1203
1204 size_mismatch:
1205         md_stop(&rs->md);
1206 bad:
1207         context_free(rs);
1208
1209         return ret;
1210 }
1211
1212 static void raid_dtr(struct dm_target *ti)
1213 {
1214         struct raid_set *rs = ti->private;
1215
1216         list_del_init(&rs->callbacks.list);
1217         md_stop(&rs->md);
1218         context_free(rs);
1219 }
1220
1221 static int raid_map(struct dm_target *ti, struct bio *bio)
1222 {
1223         struct raid_set *rs = ti->private;
1224         struct mddev *mddev = &rs->md;
1225
1226         mddev->pers->make_request(mddev, bio);
1227
1228         return DM_MAPIO_SUBMITTED;
1229 }
1230
1231 static int raid_status(struct dm_target *ti, status_type_t type,
1232                        unsigned status_flags, char *result, unsigned maxlen)
1233 {
1234         struct raid_set *rs = ti->private;
1235         unsigned raid_param_cnt = 1; /* at least 1 for chunksize */
1236         unsigned sz = 0;
1237         int i, array_in_sync = 0;
1238         sector_t sync;
1239
1240         switch (type) {
1241         case STATUSTYPE_INFO:
1242                 DMEMIT("%s %d ", rs->raid_type->name, rs->md.raid_disks);
1243
1244                 if (test_bit(MD_RECOVERY_RUNNING, &rs->md.recovery))
1245                         sync = rs->md.curr_resync_completed;
1246                 else
1247                         sync = rs->md.recovery_cp;
1248
1249                 if (sync >= rs->md.resync_max_sectors) {
1250                         array_in_sync = 1;
1251                         sync = rs->md.resync_max_sectors;
1252                 } else {
1253                         /*
1254                          * The array may be doing an initial sync, or it may
1255                          * be rebuilding individual components.  If all the
1256                          * devices are In_sync, then it is the array that is
1257                          * being initialized.
1258                          */
1259                         for (i = 0; i < rs->md.raid_disks; i++)
1260                                 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
1261                                         array_in_sync = 1;
1262                 }
1263                 /*
1264                  * Status characters:
1265                  *  'D' = Dead/Failed device
1266                  *  'a' = Alive but not in-sync
1267                  *  'A' = Alive and in-sync
1268                  */
1269                 for (i = 0; i < rs->md.raid_disks; i++) {
1270                         if (test_bit(Faulty, &rs->dev[i].rdev.flags))
1271                                 DMEMIT("D");
1272                         else if (!array_in_sync ||
1273                                  !test_bit(In_sync, &rs->dev[i].rdev.flags))
1274                                 DMEMIT("a");
1275                         else
1276                                 DMEMIT("A");
1277                 }
1278
1279                 /*
1280                  * In-sync ratio:
1281                  *  The in-sync ratio shows the progress of:
1282                  *   - Initializing the array
1283                  *   - Rebuilding a subset of devices of the array
1284                  *  The user can distinguish between the two by referring
1285                  *  to the status characters.
1286                  */
1287                 DMEMIT(" %llu/%llu",
1288                        (unsigned long long) sync,
1289                        (unsigned long long) rs->md.resync_max_sectors);
1290
1291                 break;
1292         case STATUSTYPE_TABLE:
1293                 /* The string you would use to construct this array */
1294                 for (i = 0; i < rs->md.raid_disks; i++) {
1295                         if ((rs->print_flags & DMPF_REBUILD) &&
1296                             rs->dev[i].data_dev &&
1297                             !test_bit(In_sync, &rs->dev[i].rdev.flags))
1298                                 raid_param_cnt += 2; /* for rebuilds */
1299                         if (rs->dev[i].data_dev &&
1300                             test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1301                                 raid_param_cnt += 2;
1302                 }
1303
1304                 raid_param_cnt += (hweight32(rs->print_flags & ~DMPF_REBUILD) * 2);
1305                 if (rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC))
1306                         raid_param_cnt--;
1307
1308                 DMEMIT("%s %u %u", rs->raid_type->name,
1309                        raid_param_cnt, rs->md.chunk_sectors);
1310
1311                 if ((rs->print_flags & DMPF_SYNC) &&
1312                     (rs->md.recovery_cp == MaxSector))
1313                         DMEMIT(" sync");
1314                 if (rs->print_flags & DMPF_NOSYNC)
1315                         DMEMIT(" nosync");
1316
1317                 for (i = 0; i < rs->md.raid_disks; i++)
1318                         if ((rs->print_flags & DMPF_REBUILD) &&
1319                             rs->dev[i].data_dev &&
1320                             !test_bit(In_sync, &rs->dev[i].rdev.flags))
1321                                 DMEMIT(" rebuild %u", i);
1322
1323                 if (rs->print_flags & DMPF_DAEMON_SLEEP)
1324                         DMEMIT(" daemon_sleep %lu",
1325                                rs->md.bitmap_info.daemon_sleep);
1326
1327                 if (rs->print_flags & DMPF_MIN_RECOVERY_RATE)
1328                         DMEMIT(" min_recovery_rate %d", rs->md.sync_speed_min);
1329
1330                 if (rs->print_flags & DMPF_MAX_RECOVERY_RATE)
1331                         DMEMIT(" max_recovery_rate %d", rs->md.sync_speed_max);
1332
1333                 for (i = 0; i < rs->md.raid_disks; i++)
1334                         if (rs->dev[i].data_dev &&
1335                             test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1336                                 DMEMIT(" write_mostly %u", i);
1337
1338                 if (rs->print_flags & DMPF_MAX_WRITE_BEHIND)
1339                         DMEMIT(" max_write_behind %lu",
1340                                rs->md.bitmap_info.max_write_behind);
1341
1342                 if (rs->print_flags & DMPF_STRIPE_CACHE) {
1343                         struct r5conf *conf = rs->md.private;
1344
1345                         /* convert from kiB to sectors */
1346                         DMEMIT(" stripe_cache %d",
1347                                conf ? conf->max_nr_stripes * 2 : 0);
1348                 }
1349
1350                 if (rs->print_flags & DMPF_REGION_SIZE)
1351                         DMEMIT(" region_size %lu",
1352                                rs->md.bitmap_info.chunksize >> 9);
1353
1354                 if (rs->print_flags & DMPF_RAID10_COPIES)
1355                         DMEMIT(" raid10_copies %u",
1356                                raid10_md_layout_to_copies(rs->md.layout));
1357
1358                 if (rs->print_flags & DMPF_RAID10_FORMAT)
1359                         DMEMIT(" raid10_format near");
1360
1361                 DMEMIT(" %d", rs->md.raid_disks);
1362                 for (i = 0; i < rs->md.raid_disks; i++) {
1363                         if (rs->dev[i].meta_dev)
1364                                 DMEMIT(" %s", rs->dev[i].meta_dev->name);
1365                         else
1366                                 DMEMIT(" -");
1367
1368                         if (rs->dev[i].data_dev)
1369                                 DMEMIT(" %s", rs->dev[i].data_dev->name);
1370                         else
1371                                 DMEMIT(" -");
1372                 }
1373         }
1374
1375         return 0;
1376 }
1377
1378 static int raid_iterate_devices(struct dm_target *ti, iterate_devices_callout_fn fn, void *data)
1379 {
1380         struct raid_set *rs = ti->private;
1381         unsigned i;
1382         int ret = 0;
1383
1384         for (i = 0; !ret && i < rs->md.raid_disks; i++)
1385                 if (rs->dev[i].data_dev)
1386                         ret = fn(ti,
1387                                  rs->dev[i].data_dev,
1388                                  0, /* No offset on data devs */
1389                                  rs->md.dev_sectors,
1390                                  data);
1391
1392         return ret;
1393 }
1394
1395 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
1396 {
1397         struct raid_set *rs = ti->private;
1398         unsigned chunk_size = rs->md.chunk_sectors << 9;
1399         struct r5conf *conf = rs->md.private;
1400
1401         blk_limits_io_min(limits, chunk_size);
1402         blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded));
1403 }
1404
1405 static void raid_presuspend(struct dm_target *ti)
1406 {
1407         struct raid_set *rs = ti->private;
1408
1409         md_stop_writes(&rs->md);
1410 }
1411
1412 static void raid_postsuspend(struct dm_target *ti)
1413 {
1414         struct raid_set *rs = ti->private;
1415
1416         mddev_suspend(&rs->md);
1417 }
1418
1419 static void raid_resume(struct dm_target *ti)
1420 {
1421         struct raid_set *rs = ti->private;
1422
1423         set_bit(MD_CHANGE_DEVS, &rs->md.flags);
1424         if (!rs->bitmap_loaded) {
1425                 bitmap_load(&rs->md);
1426                 rs->bitmap_loaded = 1;
1427         }
1428
1429         clear_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
1430         mddev_resume(&rs->md);
1431 }
1432
1433 static struct target_type raid_target = {
1434         .name = "raid",
1435         .version = {1, 4, 0},
1436         .module = THIS_MODULE,
1437         .ctr = raid_ctr,
1438         .dtr = raid_dtr,
1439         .map = raid_map,
1440         .status = raid_status,
1441         .iterate_devices = raid_iterate_devices,
1442         .io_hints = raid_io_hints,
1443         .presuspend = raid_presuspend,
1444         .postsuspend = raid_postsuspend,
1445         .resume = raid_resume,
1446 };
1447
1448 static int __init dm_raid_init(void)
1449 {
1450         return dm_register_target(&raid_target);
1451 }
1452
1453 static void __exit dm_raid_exit(void)
1454 {
1455         dm_unregister_target(&raid_target);
1456 }
1457
1458 module_init(dm_raid_init);
1459 module_exit(dm_raid_exit);
1460
1461 MODULE_DESCRIPTION(DM_NAME " raid4/5/6 target");
1462 MODULE_ALIAS("dm-raid1");
1463 MODULE_ALIAS("dm-raid10");
1464 MODULE_ALIAS("dm-raid4");
1465 MODULE_ALIAS("dm-raid5");
1466 MODULE_ALIAS("dm-raid6");
1467 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
1468 MODULE_LICENSE("GPL");