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                         DMINFO("Choosing default region size of %lu sectors",
 299                                region_size);
 300                         region_size = min_region_size;
 301                 } else {
 302                         DMINFO("Choosing default region size of 4MiB");
 303                         region_size = 1 << 13; /* sectors */
 304                 }
 305         } else {
 306                 /*
 307                  * Validate user-supplied value.
 308                  */
 309                 if (region_size > rs->ti->len) {
 310                         rs->ti->error = "Supplied region size is too large";
 311                         return -EINVAL;
 312                 }
 313
 314                 if (region_size < min_region_size) {
 315                         DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
 316                               region_size, min_region_size);
 317                         rs->ti->error = "Supplied region size is too small";
 318                         return -EINVAL;
 319                 }
 320
 321                 if (!is_power_of_2(region_size)) {
 322                         rs->ti->error = "Region size is not a power of 2";
 323                         return -EINVAL;
 324                 }
 325
 326                 if (region_size < rs->md.chunk_sectors) {
 327                         rs->ti->error = "Region size is smaller than the chunk size";
 328                         return -EINVAL;
 329                 }
 330         }
 331
 332         /*
 333          * Convert sectors to bytes.
 334          */
 335         rs->md.bitmap_info.chunksize = (region_size << 9);
 336
 337         return 0;
 338 }
 339
 340 /*
 341  * Possible arguments are...
 342  *      <chunk_size> [optional_args]
 343  *
 344  * Argument definitions
 345  *    <chunk_size>                      The number of sectors per disk that
 346  *                                      will form the "stripe"
 347  *    [[no]sync]                        Force or prevent recovery of the
 348  *                                      entire array
 349  *    [rebuild <idx>]                   Rebuild the drive indicated by the index
 350  *    [daemon_sleep <ms>]               Time between bitmap daemon work to
 351  *                                      clear bits
 352  *    [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
 353  *    [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
 354  *    [write_mostly <idx>]              Indicate a write mostly drive via index
 355  *    [max_write_behind <sectors>]      See '-write-behind=' (man mdadm)
 356  *    [stripe_cache <sectors>]          Stripe cache size for higher RAIDs
 357  *    [region_size <sectors>]           Defines granularity of bitmap
 358  *
 359  * RAID10-only options:
 360  *    [raid10_copies <# copies>]        Number of copies.  (Default: 2)
 361  *    [raid10_format <near>]            Layout algorithm.  (Default: near)
 362  */
 363 static int parse_raid_params(struct raid_set *rs, char **argv,
 364                              unsigned num_raid_params)
 365 {
 366         char *raid10_format = "near";
 367         unsigned raid10_copies = 2;
 368         unsigned i, rebuild_cnt = 0;
 369         unsigned long value, region_size = 0;
 370         sector_t sectors_per_dev = rs->ti->len;
 371         sector_t max_io_len;
 372         char *key;
 373
 374         /*
 375          * First, parse the in-order required arguments
 376          * "chunk_size" is the only argument of this type.
 377          */
 378         if ((strict_strtoul(argv[0], 10, &value) < 0)) {
 379                 rs->ti->error = "Bad chunk size";
 380                 return -EINVAL;
 381         } else if (rs->raid_type->level == 1) {
 382                 if (value)
 383                         DMERR("Ignoring chunk size parameter for RAID 1");
 384                 value = 0;
 385         } else if (!is_power_of_2(value)) {
 386                 rs->ti->error = "Chunk size must be a power of 2";
 387                 return -EINVAL;
 388         } else if (value < 8) {
 389                 rs->ti->error = "Chunk size value is too small";
 390                 return -EINVAL;
 391         }
 392
 393         rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
 394         argv++;
 395         num_raid_params--;
 396
 397         /*
 398          * We set each individual device as In_sync with a completed
 399          * 'recovery_offset'.  If there has been a device failure or
 400          * replacement then one of the following cases applies:
 401          *
 402          *   1) User specifies 'rebuild'.
 403          *      - Device is reset when param is read.
 404          *   2) A new device is supplied.
 405          *      - No matching superblock found, resets device.
 406          *   3) Device failure was transient and returns on reload.
 407          *      - Failure noticed, resets device for bitmap replay.
 408          *   4) Device hadn't completed recovery after previous failure.
 409          *      - Superblock is read and overrides recovery_offset.
 410          *
 411          * What is found in the superblocks of the devices is always
 412          * authoritative, unless 'rebuild' or '[no]sync' was specified.
 413          */
 414         for (i = 0; i < rs->md.raid_disks; i++) {
 415                 set_bit(In_sync, &rs->dev[i].rdev.flags);
 416                 rs->dev[i].rdev.recovery_offset = MaxSector;
 417         }
 418
 419         /*
 420          * Second, parse the unordered optional arguments
 421          */
 422         for (i = 0; i < num_raid_params; i++) {
 423                 if (!strcasecmp(argv[i], "nosync")) {
 424                         rs->md.recovery_cp = MaxSector;
 425                         rs->print_flags |= DMPF_NOSYNC;
 426                         continue;
 427                 }
 428                 if (!strcasecmp(argv[i], "sync")) {
 429                         rs->md.recovery_cp = 0;
 430                         rs->print_flags |= DMPF_SYNC;
 431                         continue;
 432                 }
 433
 434                 /* The rest of the optional arguments come in key/value pairs */
 435                 if ((i + 1) >= num_raid_params) {
 436                         rs->ti->error = "Wrong number of raid parameters given";
 437                         return -EINVAL;
 438                 }
 439
 440                 key = argv[i++];
 441
 442                 /* Parameters that take a string value are checked here. */
 443                 if (!strcasecmp(key, "raid10_format")) {
 444                         if (rs->raid_type->level != 10) {
 445                                 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
 446                                 return -EINVAL;
 447                         }
 448                         if (strcmp("near", argv[i])) {
 449                                 rs->ti->error = "Invalid 'raid10_format' value given";
 450                                 return -EINVAL;
 451                         }
 452                         raid10_format = argv[i];
 453                         rs->print_flags |= DMPF_RAID10_FORMAT;
 454                         continue;
 455                 }
 456
 457                 if (strict_strtoul(argv[i], 10, &value) < 0) {
 458                         rs->ti->error = "Bad numerical argument given in raid params";
 459                         return -EINVAL;
 460                 }
 461
 462                 /* Parameters that take a numeric value are checked here */
 463                 if (!strcasecmp(key, "rebuild")) {
 464                         rebuild_cnt++;
 465
 466                         switch (rs->raid_type->level) {
 467                         case 1:
 468                                 if (rebuild_cnt >= rs->md.raid_disks) {
 469                                         rs->ti->error = "Too many rebuild devices specified";
 470                                         return -EINVAL;
 471                                 }
 472                                 break;
 473                         case 4:
 474                         case 5:
 475                         case 6:
 476                                 if (rebuild_cnt > rs->raid_type->parity_devs) {
 477                                         rs->ti->error = "Too many rebuild devices specified for given RAID type";
 478                                         return -EINVAL;
 479                                 }
 480                                 break;
 481                         case 10:
 482                         default:
 483                                 DMERR("The rebuild parameter is not supported for %s", rs->raid_type->name);
 484                                 rs->ti->error = "Rebuild not supported for this RAID type";
 485                                 return -EINVAL;
 486                         }
 487
 488                         if (value > rs->md.raid_disks) {
 489                                 rs->ti->error = "Invalid rebuild index given";
 490                                 return -EINVAL;
 491                         }
 492                         clear_bit(In_sync, &rs->dev[value].rdev.flags);
 493                         rs->dev[value].rdev.recovery_offset = 0;
 494                         rs->print_flags |= DMPF_REBUILD;
 495                 } else if (!strcasecmp(key, "write_mostly")) {
 496                         if (rs->raid_type->level != 1) {
 497                                 rs->ti->error = "write_mostly option is only valid for RAID1";
 498                                 return -EINVAL;
 499                         }
 500                         if (value >= rs->md.raid_disks) {
 501                                 rs->ti->error = "Invalid write_mostly drive index given";
 502                                 return -EINVAL;
 503                         }
 504                         set_bit(WriteMostly, &rs->dev[value].rdev.flags);
 505                 } else if (!strcasecmp(key, "max_write_behind")) {
 506                         if (rs->raid_type->level != 1) {
 507                                 rs->ti->error = "max_write_behind option is only valid for RAID1";
 508                                 return -EINVAL;
 509                         }
 510                         rs->print_flags |= DMPF_MAX_WRITE_BEHIND;
 511
 512                         /*
 513                          * In device-mapper, we specify things in sectors, but
 514                          * MD records this value in kB
 515                          */
 516                         value /= 2;
 517                         if (value > COUNTER_MAX) {
 518                                 rs->ti->error = "Max write-behind limit out of range";
 519                                 return -EINVAL;
 520                         }
 521                         rs->md.bitmap_info.max_write_behind = value;
 522                 } else if (!strcasecmp(key, "daemon_sleep")) {
 523                         rs->print_flags |= DMPF_DAEMON_SLEEP;
 524                         if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
 525                                 rs->ti->error = "daemon sleep period out of range";
 526                                 return -EINVAL;
 527                         }
 528                         rs->md.bitmap_info.daemon_sleep = value;
 529                 } else if (!strcasecmp(key, "stripe_cache")) {
 530                         rs->print_flags |= DMPF_STRIPE_CACHE;
 531
 532                         /*
 533                          * In device-mapper, we specify things in sectors, but
 534                          * MD records this value in kB
 535                          */
 536                         value /= 2;
 537
 538                         if ((rs->raid_type->level != 5) &&
 539                             (rs->raid_type->level != 6)) {
 540                                 rs->ti->error = "Inappropriate argument: stripe_cache";
 541                                 return -EINVAL;
 542                         }
 543                         if (raid5_set_cache_size(&rs->md, (int)value)) {
 544                                 rs->ti->error = "Bad stripe_cache size";
 545                                 return -EINVAL;
 546                         }
 547                 } else if (!strcasecmp(key, "min_recovery_rate")) {
 548                         rs->print_flags |= DMPF_MIN_RECOVERY_RATE;
 549                         if (value > INT_MAX) {
 550                                 rs->ti->error = "min_recovery_rate out of range";
 551                                 return -EINVAL;
 552                         }
 553                         rs->md.sync_speed_min = (int)value;
 554                 } else if (!strcasecmp(key, "max_recovery_rate")) {
 555                         rs->print_flags |= DMPF_MAX_RECOVERY_RATE;
 556                         if (value > INT_MAX) {
 557                                 rs->ti->error = "max_recovery_rate out of range";
 558                                 return -EINVAL;
 559                         }
 560                         rs->md.sync_speed_max = (int)value;
 561                 } else if (!strcasecmp(key, "region_size")) {
 562                         rs->print_flags |= DMPF_REGION_SIZE;
 563                         region_size = value;
 564                 } else if (!strcasecmp(key, "raid10_copies") &&
 565                            (rs->raid_type->level == 10)) {
 566                         if ((value < 2) || (value > 0xFF)) {
 567                                 rs->ti->error = "Bad value for 'raid10_copies'";
 568                                 return -EINVAL;
 569                         }
 570                         rs->print_flags |= DMPF_RAID10_COPIES;
 571                         raid10_copies = value;
 572                 } else {
 573                         DMERR("Unable to parse RAID parameter: %s", key);
 574                         rs->ti->error = "Unable to parse RAID parameters";
 575                         return -EINVAL;
 576                 }
 577         }
 578
 579         if (validate_region_size(rs, region_size))
 580                 return -EINVAL;
 581
 582         if (rs->md.chunk_sectors)
 583                 max_io_len = rs->md.chunk_sectors;
 584         else
 585                 max_io_len = region_size;
 586
 587         if (dm_set_target_max_io_len(rs->ti, max_io_len))
 588                 return -EINVAL;
 589
 590         if (rs->raid_type->level == 10) {
 591                 if (raid10_copies > rs->md.raid_disks) {
 592                         rs->ti->error = "Not enough devices to satisfy specification";
 593                         return -EINVAL;
 594                 }
 595
 596                 /* (Len * #mirrors) / #devices */
 597                 sectors_per_dev = rs->ti->len * raid10_copies;
 598                 sector_div(sectors_per_dev, rs->md.raid_disks);
 599
 600                 rs->md.layout = raid10_format_to_md_layout(raid10_format,
 601                                                            raid10_copies);
 602                 rs->md.new_layout = rs->md.layout;
 603         } else if ((rs->raid_type->level > 1) &&
 604                    sector_div(sectors_per_dev,
 605                               (rs->md.raid_disks - rs->raid_type->parity_devs))) {
 606                 rs->ti->error = "Target length not divisible by number of data devices";
 607                 return -EINVAL;
 608         }
 609         rs->md.dev_sectors = sectors_per_dev;
 610
 611         /* Assume there are no metadata devices until the drives are parsed */
 612         rs->md.persistent = 0;
 613         rs->md.external = 1;
 614
 615         return 0;
 616 }
 617
 618 static void do_table_event(struct work_struct *ws)
 619 {
 620         struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
 621
 622         dm_table_event(rs->ti->table);
 623 }
 624
 625 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
 626 {
 627         struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
 628
 629         if (rs->raid_type->level == 1)
 630                 return md_raid1_congested(&rs->md, bits);
 631
 632         if (rs->raid_type->level == 10)
 633                 return md_raid10_congested(&rs->md, bits);
 634
 635         return md_raid5_congested(&rs->md, bits);
 636 }
 637
 638 /*
 639  * This structure is never routinely used by userspace, unlike md superblocks.
 640  * Devices with this superblock should only ever be accessed via device-mapper.
 641  */
 642 #define DM_RAID_MAGIC 0x64526D44
 643 struct dm_raid_superblock {
 644         __le32 magic;           /* "DmRd" */
 645         __le32 features;        /* Used to indicate possible future changes */
 646
 647         __le32 num_devices;     /* Number of devices in this array. (Max 64) */
 648         __le32 array_position;  /* The position of this drive in the array */
 649
 650         __le64 events;          /* Incremented by md when superblock updated */
 651         __le64 failed_devices;  /* Bit field of devices to indicate failures */
 652
 653         /*
 654          * This offset tracks the progress of the repair or replacement of
 655          * an individual drive.
 656          */
 657         __le64 disk_recovery_offset;
 658
 659         /*
 660          * This offset tracks the progress of the initial array
 661          * synchronisation/parity calculation.
 662          */
 663         __le64 array_resync_offset;
 664
 665         /*
 666          * RAID characteristics
 667          */
 668         __le32 level;
 669         __le32 layout;
 670         __le32 stripe_sectors;
 671
 672         __u8 pad[452];          /* Round struct to 512 bytes. */
 673                                 /* Always set to 0 when writing. */
 674 } __packed;
 675
 676 static int read_disk_sb(struct md_rdev *rdev, int size)
 677 {
 678         BUG_ON(!rdev->sb_page);
 679
 680         if (rdev->sb_loaded)
 681                 return 0;
 682
 683         if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, 1)) {
 684                 DMERR("Failed to read superblock of device at position %d",
 685                       rdev->raid_disk);
 686                 md_error(rdev->mddev, rdev);
 687                 return -EINVAL;
 688         }
 689
 690         rdev->sb_loaded = 1;
 691
 692         return 0;
 693 }
 694
 695 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
 696 {
 697         int i;
 698         uint64_t failed_devices;
 699         struct dm_raid_superblock *sb;
 700         struct raid_set *rs = container_of(mddev, struct raid_set, md);
 701
 702         sb = page_address(rdev->sb_page);
 703         failed_devices = le64_to_cpu(sb->failed_devices);
 704
 705         for (i = 0; i < mddev->raid_disks; i++)
 706                 if (!rs->dev[i].data_dev ||
 707                     test_bit(Faulty, &(rs->dev[i].rdev.flags)))
 708                         failed_devices |= (1ULL << i);
 709
 710         memset(sb, 0, sizeof(*sb));
 711
 712         sb->magic = cpu_to_le32(DM_RAID_MAGIC);
 713         sb->features = cpu_to_le32(0);  /* No features yet */
 714
 715         sb->num_devices = cpu_to_le32(mddev->raid_disks);
 716         sb->array_position = cpu_to_le32(rdev->raid_disk);
 717
 718         sb->events = cpu_to_le64(mddev->events);
 719         sb->failed_devices = cpu_to_le64(failed_devices);
 720
 721         sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
 722         sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
 723
 724         sb->level = cpu_to_le32(mddev->level);
 725         sb->layout = cpu_to_le32(mddev->layout);
 726         sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
 727 }
 728
 729 /*
 730  * super_load
 731  *
 732  * This function creates a superblock if one is not found on the device
 733  * and will decide which superblock to use if there's a choice.
 734  *
 735  * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
 736  */
 737 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
 738 {
 739         int ret;
 740         struct dm_raid_superblock *sb;
 741         struct dm_raid_superblock *refsb;
 742         uint64_t events_sb, events_refsb;
 743
 744         rdev->sb_start = 0;
 745         rdev->sb_size = sizeof(*sb);
 746
 747         ret = read_disk_sb(rdev, rdev->sb_size);
 748         if (ret)
 749                 return ret;
 750
 751         sb = page_address(rdev->sb_page);
 752
 753         /*
 754          * Two cases that we want to write new superblocks and rebuild:
 755          * 1) New device (no matching magic number)
 756          * 2) Device specified for rebuild (!In_sync w/ offset == 0)
 757          */
 758         if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
 759             (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
 760                 super_sync(rdev->mddev, rdev);
 761
 762                 set_bit(FirstUse, &rdev->flags);
 763
 764                 /* Force writing of superblocks to disk */
 765                 set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags);
 766
 767                 /* Any superblock is better than none, choose that if given */
 768                 return refdev ? 0 : 1;
 769         }
 770
 771         if (!refdev)
 772                 return 1;
 773
 774         events_sb = le64_to_cpu(sb->events);
 775
 776         refsb = page_address(refdev->sb_page);
 777         events_refsb = le64_to_cpu(refsb->events);
 778
 779         return (events_sb > events_refsb) ? 1 : 0;
 780 }
 781
 782 static int super_init_validation(struct mddev *mddev, struct md_rdev *rdev)
 783 {
 784         int role;
 785         struct raid_set *rs = container_of(mddev, struct raid_set, md);
 786         uint64_t events_sb;
 787         uint64_t failed_devices;
 788         struct dm_raid_superblock *sb;
 789         uint32_t new_devs = 0;
 790         uint32_t rebuilds = 0;
 791         struct md_rdev *r;
 792         struct dm_raid_superblock *sb2;
 793
 794         sb = page_address(rdev->sb_page);
 795         events_sb = le64_to_cpu(sb->events);
 796         failed_devices = le64_to_cpu(sb->failed_devices);
 797
 798         /*
 799          * Initialise to 1 if this is a new superblock.
 800          */
 801         mddev->events = events_sb ? : 1;
 802
 803         /*
 804          * Reshaping is not currently allowed
 805          */
 806         if ((le32_to_cpu(sb->level) != mddev->level) ||
 807             (le32_to_cpu(sb->layout) != mddev->layout) ||
 808             (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors)) {
 809                 DMERR("Reshaping arrays not yet supported.");
 810                 return -EINVAL;
 811         }
 812
 813         /* We can only change the number of devices in RAID1 right now */
 814         if ((rs->raid_type->level != 1) &&
 815             (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) {
 816                 DMERR("Reshaping arrays not yet supported.");
 817                 return -EINVAL;
 818         }
 819
 820         if (!(rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC)))
 821                 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
 822
 823         /*
 824          * During load, we set FirstUse if a new superblock was written.
 825          * There are two reasons we might not have a superblock:
 826          * 1) The array is brand new - in which case, all of the
 827          *    devices must have their In_sync bit set.  Also,
 828          *    recovery_cp must be 0, unless forced.
 829          * 2) This is a new device being added to an old array
 830          *    and the new device needs to be rebuilt - in which
 831          *    case the In_sync bit will /not/ be set and
 832          *    recovery_cp must be MaxSector.
 833          */
 834         rdev_for_each(r, mddev) {
 835                 if (!test_bit(In_sync, &r->flags)) {
 836                         DMINFO("Device %d specified for rebuild: "
 837                                "Clearing superblock", r->raid_disk);
 838                         rebuilds++;
 839                 } else if (test_bit(FirstUse, &r->flags))
 840                         new_devs++;
 841         }
 842
 843         if (!rebuilds) {
 844                 if (new_devs == mddev->raid_disks) {
 845                         DMINFO("Superblocks created for new array");
 846                         set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
 847                 } else if (new_devs) {
 848                         DMERR("New device injected "
 849                               "into existing array without 'rebuild' "
 850                               "parameter specified");
 851                         return -EINVAL;
 852                 }
 853         } else if (new_devs) {
 854                 DMERR("'rebuild' devices cannot be "
 855                       "injected into an array with other first-time devices");
 856                 return -EINVAL;
 857         } else if (mddev->recovery_cp != MaxSector) {
 858                 DMERR("'rebuild' specified while array is not in-sync");
 859                 return -EINVAL;
 860         }
 861
 862         /*
 863          * Now we set the Faulty bit for those devices that are
 864          * recorded in the superblock as failed.
 865          */
 866         rdev_for_each(r, mddev) {
 867                 if (!r->sb_page)
 868                         continue;
 869                 sb2 = page_address(r->sb_page);
 870                 sb2->failed_devices = 0;
 871
 872                 /*
 873                  * Check for any device re-ordering.
 874                  */
 875                 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
 876                         role = le32_to_cpu(sb2->array_position);
 877                         if (role != r->raid_disk) {
 878                                 if (rs->raid_type->level != 1) {
 879                                         rs->ti->error = "Cannot change device "
 880                                                 "positions in RAID array";
 881                                         return -EINVAL;
 882                                 }
 883                                 DMINFO("RAID1 device #%d now at position #%d",
 884                                        role, r->raid_disk);
 885                         }
 886
 887                         /*
 888                          * Partial recovery is performed on
 889                          * returning failed devices.
 890                          */
 891                         if (failed_devices & (1 << role))
 892                                 set_bit(Faulty, &r->flags);
 893                 }
 894         }
 895
 896         return 0;
 897 }
 898
 899 static int super_validate(struct mddev *mddev, struct md_rdev *rdev)
 900 {
 901         struct dm_raid_superblock *sb = page_address(rdev->sb_page);
 902
 903         /*
 904          * If mddev->events is not set, we know we have not yet initialized
 905          * the array.
 906          */
 907         if (!mddev->events && super_init_validation(mddev, rdev))
 908                 return -EINVAL;
 909
 910         mddev->bitmap_info.offset = 4096 >> 9; /* Enable bitmap creation */
 911         rdev->mddev->bitmap_info.default_offset = 4096 >> 9;
 912         if (!test_bit(FirstUse, &rdev->flags)) {
 913                 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
 914                 if (rdev->recovery_offset != MaxSector)
 915                         clear_bit(In_sync, &rdev->flags);
 916         }
 917
 918         /*
 919          * If a device comes back, set it as not In_sync and no longer faulty.
 920          */
 921         if (test_bit(Faulty, &rdev->flags)) {
 922                 clear_bit(Faulty, &rdev->flags);
 923                 clear_bit(In_sync, &rdev->flags);
 924                 rdev->saved_raid_disk = rdev->raid_disk;
 925                 rdev->recovery_offset = 0;
 926         }
 927
 928         clear_bit(FirstUse, &rdev->flags);
 929
 930         return 0;
 931 }
 932
 933 /*
 934  * Analyse superblocks and select the freshest.
 935  */
 936 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
 937 {
 938         int ret;
 939         unsigned redundancy = 0;
 940         struct raid_dev *dev;
 941         struct md_rdev *rdev, *tmp, *freshest;
 942         struct mddev *mddev = &rs->md;
 943
 944         switch (rs->raid_type->level) {
 945         case 1:
 946                 redundancy = rs->md.raid_disks - 1;
 947                 break;
 948         case 4:
 949         case 5:
 950         case 6:
 951                 redundancy = rs->raid_type->parity_devs;
 952                 break;
 953         case 10:
 954                 redundancy = raid10_md_layout_to_copies(mddev->layout) - 1;
 955                 break;
 956         default:
 957                 ti->error = "Unknown RAID type";
 958                 return -EINVAL;
 959         }
 960
 961         freshest = NULL;
 962         rdev_for_each_safe(rdev, tmp, mddev) {
 963                 if (!rdev->meta_bdev)
 964                         continue;
 965
 966                 ret = super_load(rdev, freshest);
 967
 968                 switch (ret) {
 969                 case 1:
 970                         freshest = rdev;
 971                         break;
 972                 case 0:
 973                         break;
 974                 default:
 975                         dev = container_of(rdev, struct raid_dev, rdev);
 976                         if (redundancy--) {
 977                                 if (dev->meta_dev)
 978                                         dm_put_device(ti, dev->meta_dev);
 979
 980                                 dev->meta_dev = NULL;
 981                                 rdev->meta_bdev = NULL;
 982
 983                                 if (rdev->sb_page)
 984                                         put_page(rdev->sb_page);
 985
 986                                 rdev->sb_page = NULL;
 987
 988                                 rdev->sb_loaded = 0;
 989
 990                                 /*
 991                                  * We might be able to salvage the data device
 992                                  * even though the meta device has failed.  For
 993                                  * now, we behave as though '- -' had been
 994                                  * set for this device in the table.
 995                                  */
 996                                 if (dev->data_dev)
 997                                         dm_put_device(ti, dev->data_dev);
 998
 999                                 dev->data_dev = NULL;
1000                                 rdev->bdev = NULL;
1001
1002                                 list_del(&rdev->same_set);
1003
1004                                 continue;
1005                         }
1006                         ti->error = "Failed to load superblock";
1007                         return ret;
1008                 }
1009         }
1010
1011         if (!freshest)
1012                 return 0;
1013
1014         /*
1015          * Validation of the freshest device provides the source of
1016          * validation for the remaining devices.
1017          */
1018         ti->error = "Unable to assemble array: Invalid superblocks";
1019         if (super_validate(mddev, freshest))
1020                 return -EINVAL;
1021
1022         rdev_for_each(rdev, mddev)
1023                 if ((rdev != freshest) && super_validate(mddev, rdev))
1024                         return -EINVAL;
1025
1026         return 0;
1027 }
1028
1029 /*
1030  * Construct a RAID4/5/6 mapping:
1031  * Args:
1032  *      <raid_type> <#raid_params> <raid_params>                \
1033  *      <#raid_devs> { <meta_dev1> <dev1> .. <meta_devN> <devN> }
1034  *
1035  * <raid_params> varies by <raid_type>.  See 'parse_raid_params' for
1036  * details on possible <raid_params>.
1037  */
1038 static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv)
1039 {
1040         int ret;
1041         struct raid_type *rt;
1042         unsigned long num_raid_params, num_raid_devs;
1043         struct raid_set *rs = NULL;
1044
1045         /* Must have at least <raid_type> <#raid_params> */
1046         if (argc < 2) {
1047                 ti->error = "Too few arguments";
1048                 return -EINVAL;
1049         }
1050
1051         /* raid type */
1052         rt = get_raid_type(argv[0]);
1053         if (!rt) {
1054                 ti->error = "Unrecognised raid_type";
1055                 return -EINVAL;
1056         }
1057         argc--;
1058         argv++;
1059
1060         /* number of RAID parameters */
1061         if (strict_strtoul(argv[0], 10, &num_raid_params) < 0) {
1062                 ti->error = "Cannot understand number of RAID parameters";
1063                 return -EINVAL;
1064         }
1065         argc--;
1066         argv++;
1067
1068         /* Skip over RAID params for now and find out # of devices */
1069         if (num_raid_params + 1 > argc) {
1070                 ti->error = "Arguments do not agree with counts given";
1071                 return -EINVAL;
1072         }
1073
1074         if ((strict_strtoul(argv[num_raid_params], 10, &num_raid_devs) < 0) ||
1075             (num_raid_devs >= INT_MAX)) {
1076                 ti->error = "Cannot understand number of raid devices";
1077                 return -EINVAL;
1078         }
1079
1080         rs = context_alloc(ti, rt, (unsigned)num_raid_devs);
1081         if (IS_ERR(rs))
1082                 return PTR_ERR(rs);
1083
1084         ret = parse_raid_params(rs, argv, (unsigned)num_raid_params);
1085         if (ret)
1086                 goto bad;
1087
1088         ret = -EINVAL;
1089
1090         argc -= num_raid_params + 1; /* +1: we already have num_raid_devs */
1091         argv += num_raid_params + 1;
1092
1093         if (argc != (num_raid_devs * 2)) {
1094                 ti->error = "Supplied RAID devices does not match the count given";
1095                 goto bad;
1096         }
1097
1098         ret = dev_parms(rs, argv);
1099         if (ret)
1100                 goto bad;
1101
1102         rs->md.sync_super = super_sync;
1103         ret = analyse_superblocks(ti, rs);
1104         if (ret)
1105                 goto bad;
1106
1107         INIT_WORK(&rs->md.event_work, do_table_event);
1108         ti->private = rs;
1109         ti->num_flush_requests = 1;
1110
1111         mutex_lock(&rs->md.reconfig_mutex);
1112         ret = md_run(&rs->md);
1113         rs->md.in_sync = 0; /* Assume already marked dirty */
1114         mutex_unlock(&rs->md.reconfig_mutex);
1115
1116         if (ret) {
1117                 ti->error = "Fail to run raid array";
1118                 goto bad;
1119         }
1120
1121         if (ti->len != rs->md.array_sectors) {
1122                 ti->error = "Array size does not match requested target length";
1123                 ret = -EINVAL;
1124                 goto size_mismatch;
1125         }
1126         rs->callbacks.congested_fn = raid_is_congested;
1127         dm_table_add_target_callbacks(ti->table, &rs->callbacks);
1128
1129         mddev_suspend(&rs->md);
1130         return 0;
1131
1132 size_mismatch:
1133         md_stop(&rs->md);
1134 bad:
1135         context_free(rs);
1136
1137         return ret;
1138 }
1139
1140 static void raid_dtr(struct dm_target *ti)
1141 {
1142         struct raid_set *rs = ti->private;
1143
1144         list_del_init(&rs->callbacks.list);
1145         md_stop(&rs->md);
1146         context_free(rs);
1147 }
1148
1149 static int raid_map(struct dm_target *ti, struct bio *bio, union map_info *map_context)
1150 {
1151         struct raid_set *rs = ti->private;
1152         struct mddev *mddev = &rs->md;
1153
1154         mddev->pers->make_request(mddev, bio);
1155
1156         return DM_MAPIO_SUBMITTED;
1157 }
1158
1159 static int raid_status(struct dm_target *ti, status_type_t type,
1160                        unsigned status_flags, char *result, unsigned maxlen)
1161 {
1162         struct raid_set *rs = ti->private;
1163         unsigned raid_param_cnt = 1; /* at least 1 for chunksize */
1164         unsigned sz = 0;
1165         int i, array_in_sync = 0;
1166         sector_t sync;
1167
1168         switch (type) {
1169         case STATUSTYPE_INFO:
1170                 DMEMIT("%s %d ", rs->raid_type->name, rs->md.raid_disks);
1171
1172                 if (test_bit(MD_RECOVERY_RUNNING, &rs->md.recovery))
1173                         sync = rs->md.curr_resync_completed;
1174                 else
1175                         sync = rs->md.recovery_cp;
1176
1177                 if (sync >= rs->md.resync_max_sectors) {
1178                         array_in_sync = 1;
1179                         sync = rs->md.resync_max_sectors;
1180                 } else {
1181                         /*
1182                          * The array may be doing an initial sync, or it may
1183                          * be rebuilding individual components.  If all the
1184                          * devices are In_sync, then it is the array that is
1185                          * being initialized.
1186                          */
1187                         for (i = 0; i < rs->md.raid_disks; i++)
1188                                 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
1189                                         array_in_sync = 1;
1190                 }
1191                 /*
1192                  * Status characters:
1193                  *  'D' = Dead/Failed device
1194                  *  'a' = Alive but not in-sync
1195                  *  'A' = Alive and in-sync
1196                  */
1197                 for (i = 0; i < rs->md.raid_disks; i++) {
1198                         if (test_bit(Faulty, &rs->dev[i].rdev.flags))
1199                                 DMEMIT("D");
1200                         else if (!array_in_sync ||
1201                                  !test_bit(In_sync, &rs->dev[i].rdev.flags))
1202                                 DMEMIT("a");
1203                         else
1204                                 DMEMIT("A");
1205                 }
1206
1207                 /*
1208                  * In-sync ratio:
1209                  *  The in-sync ratio shows the progress of:
1210                  *   - Initializing the array
1211                  *   - Rebuilding a subset of devices of the array
1212                  *  The user can distinguish between the two by referring
1213                  *  to the status characters.
1214                  */
1215                 DMEMIT(" %llu/%llu",
1216                        (unsigned long long) sync,
1217                        (unsigned long long) rs->md.resync_max_sectors);
1218
1219                 break;
1220         case STATUSTYPE_TABLE:
1221                 /* The string you would use to construct this array */
1222                 for (i = 0; i < rs->md.raid_disks; i++) {
1223                         if ((rs->print_flags & DMPF_REBUILD) &&
1224                             rs->dev[i].data_dev &&
1225                             !test_bit(In_sync, &rs->dev[i].rdev.flags))
1226                                 raid_param_cnt += 2; /* for rebuilds */
1227                         if (rs->dev[i].data_dev &&
1228                             test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1229                                 raid_param_cnt += 2;
1230                 }
1231
1232                 raid_param_cnt += (hweight32(rs->print_flags & ~DMPF_REBUILD) * 2);
1233                 if (rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC))
1234                         raid_param_cnt--;
1235
1236                 DMEMIT("%s %u %u", rs->raid_type->name,
1237                        raid_param_cnt, rs->md.chunk_sectors);
1238
1239                 if ((rs->print_flags & DMPF_SYNC) &&
1240                     (rs->md.recovery_cp == MaxSector))
1241                         DMEMIT(" sync");
1242                 if (rs->print_flags & DMPF_NOSYNC)
1243                         DMEMIT(" nosync");
1244
1245                 for (i = 0; i < rs->md.raid_disks; i++)
1246                         if ((rs->print_flags & DMPF_REBUILD) &&
1247                             rs->dev[i].data_dev &&
1248                             !test_bit(In_sync, &rs->dev[i].rdev.flags))
1249                                 DMEMIT(" rebuild %u", i);
1250
1251                 if (rs->print_flags & DMPF_DAEMON_SLEEP)
1252                         DMEMIT(" daemon_sleep %lu",
1253                                rs->md.bitmap_info.daemon_sleep);
1254
1255                 if (rs->print_flags & DMPF_MIN_RECOVERY_RATE)
1256                         DMEMIT(" min_recovery_rate %d", rs->md.sync_speed_min);
1257
1258                 if (rs->print_flags & DMPF_MAX_RECOVERY_RATE)
1259                         DMEMIT(" max_recovery_rate %d", rs->md.sync_speed_max);
1260
1261                 for (i = 0; i < rs->md.raid_disks; i++)
1262                         if (rs->dev[i].data_dev &&
1263                             test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1264                                 DMEMIT(" write_mostly %u", i);
1265
1266                 if (rs->print_flags & DMPF_MAX_WRITE_BEHIND)
1267                         DMEMIT(" max_write_behind %lu",
1268                                rs->md.bitmap_info.max_write_behind);
1269
1270                 if (rs->print_flags & DMPF_STRIPE_CACHE) {
1271                         struct r5conf *conf = rs->md.private;
1272
1273                         /* convert from kiB to sectors */
1274                         DMEMIT(" stripe_cache %d",
1275                                conf ? conf->max_nr_stripes * 2 : 0);
1276                 }
1277
1278                 if (rs->print_flags & DMPF_REGION_SIZE)
1279                         DMEMIT(" region_size %lu",
1280                                rs->md.bitmap_info.chunksize >> 9);
1281
1282                 if (rs->print_flags & DMPF_RAID10_COPIES)
1283                         DMEMIT(" raid10_copies %u",
1284                                raid10_md_layout_to_copies(rs->md.layout));
1285
1286                 if (rs->print_flags & DMPF_RAID10_FORMAT)
1287                         DMEMIT(" raid10_format near");
1288
1289                 DMEMIT(" %d", rs->md.raid_disks);
1290                 for (i = 0; i < rs->md.raid_disks; i++) {
1291                         if (rs->dev[i].meta_dev)
1292                                 DMEMIT(" %s", rs->dev[i].meta_dev->name);
1293                         else
1294                                 DMEMIT(" -");
1295
1296                         if (rs->dev[i].data_dev)
1297                                 DMEMIT(" %s", rs->dev[i].data_dev->name);
1298                         else
1299                                 DMEMIT(" -");
1300                 }
1301         }
1302
1303         return 0;
1304 }
1305
1306 static int raid_iterate_devices(struct dm_target *ti, iterate_devices_callout_fn fn, void *data)
1307 {
1308         struct raid_set *rs = ti->private;
1309         unsigned i;
1310         int ret = 0;
1311
1312         for (i = 0; !ret && i < rs->md.raid_disks; i++)
1313                 if (rs->dev[i].data_dev)
1314                         ret = fn(ti,
1315                                  rs->dev[i].data_dev,
1316                                  0, /* No offset on data devs */
1317                                  rs->md.dev_sectors,
1318                                  data);
1319
1320         return ret;
1321 }
1322
1323 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
1324 {
1325         struct raid_set *rs = ti->private;
1326         unsigned chunk_size = rs->md.chunk_sectors << 9;
1327         struct r5conf *conf = rs->md.private;
1328
1329         blk_limits_io_min(limits, chunk_size);
1330         blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded));
1331 }
1332
1333 static void raid_presuspend(struct dm_target *ti)
1334 {
1335         struct raid_set *rs = ti->private;
1336
1337         md_stop_writes(&rs->md);
1338 }
1339
1340 static void raid_postsuspend(struct dm_target *ti)
1341 {
1342         struct raid_set *rs = ti->private;
1343
1344         mddev_suspend(&rs->md);
1345 }
1346
1347 static void raid_resume(struct dm_target *ti)
1348 {
1349         struct raid_set *rs = ti->private;
1350
1351         set_bit(MD_CHANGE_DEVS, &rs->md.flags);
1352         if (!rs->bitmap_loaded) {
1353                 bitmap_load(&rs->md);
1354                 rs->bitmap_loaded = 1;
1355         }
1356
1357         clear_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
1358         mddev_resume(&rs->md);
1359 }
1360
1361 static struct target_type raid_target = {
1362         .name = "raid",
1363         .version = {1, 3, 0},
1364         .module = THIS_MODULE,
1365         .ctr = raid_ctr,
1366         .dtr = raid_dtr,
1367         .map = raid_map,
1368         .status = raid_status,
1369         .iterate_devices = raid_iterate_devices,
1370         .io_hints = raid_io_hints,
1371         .presuspend = raid_presuspend,
1372         .postsuspend = raid_postsuspend,
1373         .resume = raid_resume,
1374 };
1375
1376 static int __init dm_raid_init(void)
1377 {
1378         return dm_register_target(&raid_target);
1379 }
1380
1381 static void __exit dm_raid_exit(void)
1382 {
1383         dm_unregister_target(&raid_target);
1384 }
1385
1386 module_init(dm_raid_init);
1387 module_exit(dm_raid_exit);
1388
1389 MODULE_DESCRIPTION(DM_NAME " raid4/5/6 target");
1390 MODULE_ALIAS("dm-raid1");
1391 MODULE_ALIAS("dm-raid10");
1392 MODULE_ALIAS("dm-raid4");
1393 MODULE_ALIAS("dm-raid5");
1394 MODULE_ALIAS("dm-raid6");
1395 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
1396 MODULE_LICENSE("GPL");