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 char *raid10_md_layout_to_format(int layout)
  95 {
  96         /*
  97          * Bit 16 and 17 stand for "offset" and "use_far_sets"
  98          * Refer to MD's raid10.c for details
  99          */
 100         if ((layout & 0x10000) && (layout & 0x20000))
 101                 return "offset";
 102
 103         if ((layout & 0xFF) > 1)
 104                 return "near";
 105
 106         return "far";
 107 }
 108
 109 static unsigned raid10_md_layout_to_copies(int layout)
 110 {
 111         if ((layout & 0xFF) > 1)
 112                 return layout & 0xFF;
 113         return (layout >> 8) & 0xFF;
 114 }
 115
 116 static int raid10_format_to_md_layout(char *format, unsigned copies)
 117 {
 118         unsigned n = 1, f = 1;
 119
 120         if (!strcmp("near", format))
 121                 n = copies;
 122         else
 123                 f = copies;
 124
 125         if (!strcmp("offset", format))
 126                 return 0x30000 | (f << 8) | n;
 127
 128         if (!strcmp("far", format))
 129                 return 0x20000 | (f << 8) | n;
 130
 131         return (f << 8) | n;
 132 }
 133
 134 static struct raid_type *get_raid_type(char *name)
 135 {
 136         int i;
 137
 138         for (i = 0; i < ARRAY_SIZE(raid_types); i++)
 139                 if (!strcmp(raid_types[i].name, name))
 140                         return &raid_types[i];
 141
 142         return NULL;
 143 }
 144
 145 static struct raid_set *context_alloc(struct dm_target *ti, struct raid_type *raid_type, unsigned raid_devs)
 146 {
 147         unsigned i;
 148         struct raid_set *rs;
 149
 150         if (raid_devs <= raid_type->parity_devs) {
 151                 ti->error = "Insufficient number of devices";
 152                 return ERR_PTR(-EINVAL);
 153         }
 154
 155         rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
 156         if (!rs) {
 157                 ti->error = "Cannot allocate raid context";
 158                 return ERR_PTR(-ENOMEM);
 159         }
 160
 161         mddev_init(&rs->md);
 162
 163         rs->ti = ti;
 164         rs->raid_type = raid_type;
 165         rs->md.raid_disks = raid_devs;
 166         rs->md.level = raid_type->level;
 167         rs->md.new_level = rs->md.level;
 168         rs->md.layout = raid_type->algorithm;
 169         rs->md.new_layout = rs->md.layout;
 170         rs->md.delta_disks = 0;
 171         rs->md.recovery_cp = 0;
 172
 173         for (i = 0; i < raid_devs; i++)
 174                 md_rdev_init(&rs->dev[i].rdev);
 175
 176         /*
 177          * Remaining items to be initialized by further RAID params:
 178          *  rs->md.persistent
 179          *  rs->md.external
 180          *  rs->md.chunk_sectors
 181          *  rs->md.new_chunk_sectors
 182          *  rs->md.dev_sectors
 183          */
 184
 185         return rs;
 186 }
 187
 188 static void context_free(struct raid_set *rs)
 189 {
 190         int i;
 191
 192         for (i = 0; i < rs->md.raid_disks; i++) {
 193                 if (rs->dev[i].meta_dev)
 194                         dm_put_device(rs->ti, rs->dev[i].meta_dev);
 195                 md_rdev_clear(&rs->dev[i].rdev);
 196                 if (rs->dev[i].data_dev)
 197                         dm_put_device(rs->ti, rs->dev[i].data_dev);
 198         }
 199
 200         kfree(rs);
 201 }
 202
 203 /*
 204  * For every device we have two words
 205  *  <meta_dev>: meta device name or '-' if missing
 206  *  <data_dev>: data device name or '-' if missing
 207  *
 208  * The following are permitted:
 209  *    - -
 210  *    - <data_dev>
 211  *    <meta_dev> <data_dev>
 212  *
 213  * The following is not allowed:
 214  *    <meta_dev> -
 215  *
 216  * This code parses those words.  If there is a failure,
 217  * the caller must use context_free to unwind the operations.
 218  */
 219 static int dev_parms(struct raid_set *rs, char **argv)
 220 {
 221         int i;
 222         int rebuild = 0;
 223         int metadata_available = 0;
 224         int ret = 0;
 225
 226         for (i = 0; i < rs->md.raid_disks; i++, argv += 2) {
 227                 rs->dev[i].rdev.raid_disk = i;
 228
 229                 rs->dev[i].meta_dev = NULL;
 230                 rs->dev[i].data_dev = NULL;
 231
 232                 /*
 233                  * There are no offsets, since there is a separate device
 234                  * for data and metadata.
 235                  */
 236                 rs->dev[i].rdev.data_offset = 0;
 237                 rs->dev[i].rdev.mddev = &rs->md;
 238
 239                 if (strcmp(argv[0], "-")) {
 240                         ret = dm_get_device(rs->ti, argv[0],
 241                                             dm_table_get_mode(rs->ti->table),
 242                                             &rs->dev[i].meta_dev);
 243                         rs->ti->error = "RAID metadata device lookup failure";
 244                         if (ret)
 245                                 return ret;
 246
 247                         rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
 248                         if (!rs->dev[i].rdev.sb_page)
 249                                 return -ENOMEM;
 250                 }
 251
 252                 if (!strcmp(argv[1], "-")) {
 253                         if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
 254                             (!rs->dev[i].rdev.recovery_offset)) {
 255                                 rs->ti->error = "Drive designated for rebuild not specified";
 256                                 return -EINVAL;
 257                         }
 258
 259                         rs->ti->error = "No data device supplied with metadata device";
 260                         if (rs->dev[i].meta_dev)
 261                                 return -EINVAL;
 262
 263                         continue;
 264                 }
 265
 266                 ret = dm_get_device(rs->ti, argv[1],
 267                                     dm_table_get_mode(rs->ti->table),
 268                                     &rs->dev[i].data_dev);
 269                 if (ret) {
 270                         rs->ti->error = "RAID device lookup failure";
 271                         return ret;
 272                 }
 273
 274                 if (rs->dev[i].meta_dev) {
 275                         metadata_available = 1;
 276                         rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
 277                 }
 278                 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
 279                 list_add(&rs->dev[i].rdev.same_set, &rs->md.disks);
 280                 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
 281                         rebuild++;
 282         }
 283
 284         if (metadata_available) {
 285                 rs->md.external = 0;
 286                 rs->md.persistent = 1;
 287                 rs->md.major_version = 2;
 288         } else if (rebuild && !rs->md.recovery_cp) {
 289                 /*
 290                  * Without metadata, we will not be able to tell if the array
 291                  * is in-sync or not - we must assume it is not.  Therefore,
 292                  * it is impossible to rebuild a drive.
 293                  *
 294                  * Even if there is metadata, the on-disk information may
 295                  * indicate that the array is not in-sync and it will then
 296                  * fail at that time.
 297                  *
 298                  * User could specify 'nosync' option if desperate.
 299                  */
 300                 DMERR("Unable to rebuild drive while array is not in-sync");
 301                 rs->ti->error = "RAID device lookup failure";
 302                 return -EINVAL;
 303         }
 304
 305         return 0;
 306 }
 307
 308 /*
 309  * validate_region_size
 310  * @rs
 311  * @region_size:  region size in sectors.  If 0, pick a size (4MiB default).
 312  *
 313  * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
 314  * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
 315  *
 316  * Returns: 0 on success, -EINVAL on failure.
 317  */
 318 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
 319 {
 320         unsigned long min_region_size = rs->ti->len / (1 << 21);
 321
 322         if (!region_size) {
 323                 /*
 324                  * Choose a reasonable default.  All figures in sectors.
 325                  */
 326                 if (min_region_size > (1 << 13)) {
 327                         /* If not a power of 2, make it the next power of 2 */
 328                         if (min_region_size & (min_region_size - 1))
 329                                 region_size = 1 << fls(region_size);
 330                         DMINFO("Choosing default region size of %lu sectors",
 331                                region_size);
 332                 } else {
 333                         DMINFO("Choosing default region size of 4MiB");
 334                         region_size = 1 << 13; /* sectors */
 335                 }
 336         } else {
 337                 /*
 338                  * Validate user-supplied value.
 339                  */
 340                 if (region_size > rs->ti->len) {
 341                         rs->ti->error = "Supplied region size is too large";
 342                         return -EINVAL;
 343                 }
 344
 345                 if (region_size < min_region_size) {
 346                         DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
 347                               region_size, min_region_size);
 348                         rs->ti->error = "Supplied region size is too small";
 349                         return -EINVAL;
 350                 }
 351
 352                 if (!is_power_of_2(region_size)) {
 353                         rs->ti->error = "Region size is not a power of 2";
 354                         return -EINVAL;
 355                 }
 356
 357                 if (region_size < rs->md.chunk_sectors) {
 358                         rs->ti->error = "Region size is smaller than the chunk size";
 359                         return -EINVAL;
 360                 }
 361         }
 362
 363         /*
 364          * Convert sectors to bytes.
 365          */
 366         rs->md.bitmap_info.chunksize = (region_size << 9);
 367
 368         return 0;
 369 }
 370
 371 /*
 372  * validate_raid_redundancy
 373  * @rs
 374  *
 375  * Determine if there are enough devices in the array that haven't
 376  * failed (or are being rebuilt) to form a usable array.
 377  *
 378  * Returns: 0 on success, -EINVAL on failure.
 379  */
 380 static int validate_raid_redundancy(struct raid_set *rs)
 381 {
 382         unsigned i, rebuild_cnt = 0;
 383         unsigned rebuilds_per_group = 0, copies, d;
 384         unsigned group_size, last_group_start;
 385
 386         for (i = 0; i < rs->md.raid_disks; i++)
 387                 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
 388                     !rs->dev[i].rdev.sb_page)
 389                         rebuild_cnt++;
 390
 391         switch (rs->raid_type->level) {
 392         case 1:
 393                 if (rebuild_cnt >= rs->md.raid_disks)
 394                         goto too_many;
 395                 break;
 396         case 4:
 397         case 5:
 398         case 6:
 399                 if (rebuild_cnt > rs->raid_type->parity_devs)
 400                         goto too_many;
 401                 break;
 402         case 10:
 403                 copies = raid10_md_layout_to_copies(rs->md.layout);
 404                 if (rebuild_cnt < copies)
 405                         break;
 406
 407                 /*
 408                  * It is possible to have a higher rebuild count for RAID10,
 409                  * as long as the failed devices occur in different mirror
 410                  * groups (i.e. different stripes).
 411                  *
 412                  * When checking "near" format, make sure no adjacent devices
 413                  * have failed beyond what can be handled.  In addition to the
 414                  * simple case where the number of devices is a multiple of the
 415                  * number of copies, we must also handle cases where the number
 416                  * of devices is not a multiple of the number of copies.
 417                  * E.g.    dev1 dev2 dev3 dev4 dev5
 418                  *          A    A    B    B    C
 419                  *          C    D    D    E    E
 420                  */
 421                 if (!strcmp("near", raid10_md_layout_to_format(rs->md.layout))) {
 422                         for (i = 0; i < rs->md.raid_disks * copies; i++) {
 423                                 if (!(i % copies))
 424                                         rebuilds_per_group = 0;
 425                                 d = i % rs->md.raid_disks;
 426                                 if ((!rs->dev[d].rdev.sb_page ||
 427                                      !test_bit(In_sync, &rs->dev[d].rdev.flags)) &&
 428                                     (++rebuilds_per_group >= copies))
 429                                         goto too_many;
 430                         }
 431                         break;
 432                 }
 433
 434                 /*
 435                  * When checking "far" and "offset" formats, we need to ensure
 436                  * that the device that holds its copy is not also dead or
 437                  * being rebuilt.  (Note that "far" and "offset" formats only
 438                  * support two copies right now.  These formats also only ever
 439                  * use the 'use_far_sets' variant.)
 440                  *
 441                  * This check is somewhat complicated by the need to account
 442                  * for arrays that are not a multiple of (far) copies.  This
 443                  * results in the need to treat the last (potentially larger)
 444                  * set differently.
 445                  */
 446                 group_size = (rs->md.raid_disks / copies);
 447                 last_group_start = (rs->md.raid_disks / group_size) - 1;
 448                 last_group_start *= group_size;
 449                 for (i = 0; i < rs->md.raid_disks; i++) {
 450                         if (!(i % copies) && !(i > last_group_start))
 451                                 rebuilds_per_group = 0;
 452                         if ((!rs->dev[i].rdev.sb_page ||
 453                              !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
 454                             (++rebuilds_per_group >= copies))
 455                                         goto too_many;
 456                 }
 457                 break;
 458         default:
 459                 if (rebuild_cnt)
 460                         return -EINVAL;
 461         }
 462
 463         return 0;
 464
 465 too_many:
 466         return -EINVAL;
 467 }
 468
 469 /*
 470  * Possible arguments are...
 471  *      <chunk_size> [optional_args]
 472  *
 473  * Argument definitions
 474  *    <chunk_size>                      The number of sectors per disk that
 475  *                                      will form the "stripe"
 476  *    [[no]sync]                        Force or prevent recovery of the
 477  *                                      entire array
 478  *    [rebuild <idx>]                   Rebuild the drive indicated by the index
 479  *    [daemon_sleep <ms>]               Time between bitmap daemon work to
 480  *                                      clear bits
 481  *    [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
 482  *    [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
 483  *    [write_mostly <idx>]              Indicate a write mostly drive via index
 484  *    [max_write_behind <sectors>]      See '-write-behind=' (man mdadm)
 485  *    [stripe_cache <sectors>]          Stripe cache size for higher RAIDs
 486  *    [region_size <sectors>]           Defines granularity of bitmap
 487  *
 488  * RAID10-only options:
 489  *    [raid10_copies <# copies>]        Number of copies.  (Default: 2)
 490  *    [raid10_format <near|far|offset>] Layout algorithm.  (Default: near)
 491  */
 492 static int parse_raid_params(struct raid_set *rs, char **argv,
 493                              unsigned num_raid_params)
 494 {
 495         char *raid10_format = "near";
 496         unsigned raid10_copies = 2;
 497         unsigned i;
 498         unsigned long value, region_size = 0;
 499         sector_t sectors_per_dev = rs->ti->len;
 500         sector_t max_io_len;
 501         char *key;
 502
 503         /*
 504          * First, parse the in-order required arguments
 505          * "chunk_size" is the only argument of this type.
 506          */
 507         if ((kstrtoul(argv[0], 10, &value) < 0)) {
 508                 rs->ti->error = "Bad chunk size";
 509                 return -EINVAL;
 510         } else if (rs->raid_type->level == 1) {
 511                 if (value)
 512                         DMERR("Ignoring chunk size parameter for RAID 1");
 513                 value = 0;
 514         } else if (!is_power_of_2(value)) {
 515                 rs->ti->error = "Chunk size must be a power of 2";
 516                 return -EINVAL;
 517         } else if (value < 8) {
 518                 rs->ti->error = "Chunk size value is too small";
 519                 return -EINVAL;
 520         }
 521
 522         rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
 523         argv++;
 524         num_raid_params--;
 525
 526         /*
 527          * We set each individual device as In_sync with a completed
 528          * 'recovery_offset'.  If there has been a device failure or
 529          * replacement then one of the following cases applies:
 530          *
 531          *   1) User specifies 'rebuild'.
 532          *      - Device is reset when param is read.
 533          *   2) A new device is supplied.
 534          *      - No matching superblock found, resets device.
 535          *   3) Device failure was transient and returns on reload.
 536          *      - Failure noticed, resets device for bitmap replay.
 537          *   4) Device hadn't completed recovery after previous failure.
 538          *      - Superblock is read and overrides recovery_offset.
 539          *
 540          * What is found in the superblocks of the devices is always
 541          * authoritative, unless 'rebuild' or '[no]sync' was specified.
 542          */
 543         for (i = 0; i < rs->md.raid_disks; i++) {
 544                 set_bit(In_sync, &rs->dev[i].rdev.flags);
 545                 rs->dev[i].rdev.recovery_offset = MaxSector;
 546         }
 547
 548         /*
 549          * Second, parse the unordered optional arguments
 550          */
 551         for (i = 0; i < num_raid_params; i++) {
 552                 if (!strcasecmp(argv[i], "nosync")) {
 553                         rs->md.recovery_cp = MaxSector;
 554                         rs->print_flags |= DMPF_NOSYNC;
 555                         continue;
 556                 }
 557                 if (!strcasecmp(argv[i], "sync")) {
 558                         rs->md.recovery_cp = 0;
 559                         rs->print_flags |= DMPF_SYNC;
 560                         continue;
 561                 }
 562
 563                 /* The rest of the optional arguments come in key/value pairs */
 564                 if ((i + 1) >= num_raid_params) {
 565                         rs->ti->error = "Wrong number of raid parameters given";
 566                         return -EINVAL;
 567                 }
 568
 569                 key = argv[i++];
 570
 571                 /* Parameters that take a string value are checked here. */
 572                 if (!strcasecmp(key, "raid10_format")) {
 573                         if (rs->raid_type->level != 10) {
 574                                 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
 575                                 return -EINVAL;
 576                         }
 577                         if (strcmp("near", argv[i]) &&
 578                             strcmp("far", argv[i]) &&
 579                             strcmp("offset", argv[i])) {
 580                                 rs->ti->error = "Invalid 'raid10_format' value given";
 581                                 return -EINVAL;
 582                         }
 583                         raid10_format = argv[i];
 584                         rs->print_flags |= DMPF_RAID10_FORMAT;
 585                         continue;
 586                 }
 587
 588                 if (kstrtoul(argv[i], 10, &value) < 0) {
 589                         rs->ti->error = "Bad numerical argument given in raid params";
 590                         return -EINVAL;
 591                 }
 592
 593                 /* Parameters that take a numeric value are checked here */
 594                 if (!strcasecmp(key, "rebuild")) {
 595                         if (value >= rs->md.raid_disks) {
 596                                 rs->ti->error = "Invalid rebuild index given";
 597                                 return -EINVAL;
 598                         }
 599                         clear_bit(In_sync, &rs->dev[value].rdev.flags);
 600                         rs->dev[value].rdev.recovery_offset = 0;
 601                         rs->print_flags |= DMPF_REBUILD;
 602                 } else if (!strcasecmp(key, "write_mostly")) {
 603                         if (rs->raid_type->level != 1) {
 604                                 rs->ti->error = "write_mostly option is only valid for RAID1";
 605                                 return -EINVAL;
 606                         }
 607                         if (value >= rs->md.raid_disks) {
 608                                 rs->ti->error = "Invalid write_mostly drive index given";
 609                                 return -EINVAL;
 610                         }
 611                         set_bit(WriteMostly, &rs->dev[value].rdev.flags);
 612                 } else if (!strcasecmp(key, "max_write_behind")) {
 613                         if (rs->raid_type->level != 1) {
 614                                 rs->ti->error = "max_write_behind option is only valid for RAID1";
 615                                 return -EINVAL;
 616                         }
 617                         rs->print_flags |= DMPF_MAX_WRITE_BEHIND;
 618
 619                         /*
 620                          * In device-mapper, we specify things in sectors, but
 621                          * MD records this value in kB
 622                          */
 623                         value /= 2;
 624                         if (value > COUNTER_MAX) {
 625                                 rs->ti->error = "Max write-behind limit out of range";
 626                                 return -EINVAL;
 627                         }
 628                         rs->md.bitmap_info.max_write_behind = value;
 629                 } else if (!strcasecmp(key, "daemon_sleep")) {
 630                         rs->print_flags |= DMPF_DAEMON_SLEEP;
 631                         if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
 632                                 rs->ti->error = "daemon sleep period out of range";
 633                                 return -EINVAL;
 634                         }
 635                         rs->md.bitmap_info.daemon_sleep = value;
 636                 } else if (!strcasecmp(key, "stripe_cache")) {
 637                         rs->print_flags |= DMPF_STRIPE_CACHE;
 638
 639                         /*
 640                          * In device-mapper, we specify things in sectors, but
 641                          * MD records this value in kB
 642                          */
 643                         value /= 2;
 644
 645                         if ((rs->raid_type->level != 5) &&
 646                             (rs->raid_type->level != 6)) {
 647                                 rs->ti->error = "Inappropriate argument: stripe_cache";
 648                                 return -EINVAL;
 649                         }
 650                         if (raid5_set_cache_size(&rs->md, (int)value)) {
 651                                 rs->ti->error = "Bad stripe_cache size";
 652                                 return -EINVAL;
 653                         }
 654                 } else if (!strcasecmp(key, "min_recovery_rate")) {
 655                         rs->print_flags |= DMPF_MIN_RECOVERY_RATE;
 656                         if (value > INT_MAX) {
 657                                 rs->ti->error = "min_recovery_rate out of range";
 658                                 return -EINVAL;
 659                         }
 660                         rs->md.sync_speed_min = (int)value;
 661                 } else if (!strcasecmp(key, "max_recovery_rate")) {
 662                         rs->print_flags |= DMPF_MAX_RECOVERY_RATE;
 663                         if (value > INT_MAX) {
 664                                 rs->ti->error = "max_recovery_rate out of range";
 665                                 return -EINVAL;
 666                         }
 667                         rs->md.sync_speed_max = (int)value;
 668                 } else if (!strcasecmp(key, "region_size")) {
 669                         rs->print_flags |= DMPF_REGION_SIZE;
 670                         region_size = value;
 671                 } else if (!strcasecmp(key, "raid10_copies") &&
 672                            (rs->raid_type->level == 10)) {
 673                         if ((value < 2) || (value > 0xFF)) {
 674                                 rs->ti->error = "Bad value for 'raid10_copies'";
 675                                 return -EINVAL;
 676                         }
 677                         rs->print_flags |= DMPF_RAID10_COPIES;
 678                         raid10_copies = value;
 679                 } else {
 680                         DMERR("Unable to parse RAID parameter: %s", key);
 681                         rs->ti->error = "Unable to parse RAID parameters";
 682                         return -EINVAL;
 683                 }
 684         }
 685
 686         if (validate_region_size(rs, region_size))
 687                 return -EINVAL;
 688
 689         if (rs->md.chunk_sectors)
 690                 max_io_len = rs->md.chunk_sectors;
 691         else
 692                 max_io_len = region_size;
 693
 694         if (dm_set_target_max_io_len(rs->ti, max_io_len))
 695                 return -EINVAL;
 696
 697         if (rs->raid_type->level == 10) {
 698                 if (raid10_copies > rs->md.raid_disks) {
 699                         rs->ti->error = "Not enough devices to satisfy specification";
 700                         return -EINVAL;
 701                 }
 702
 703                 /*
 704                  * If the format is not "near", we only support
 705                  * two copies at the moment.
 706                  */
 707                 if (strcmp("near", raid10_format) && (raid10_copies > 2)) {
 708                         rs->ti->error = "Too many copies for given RAID10 format.";
 709                         return -EINVAL;
 710                 }
 711
 712                 /* (Len * #mirrors) / #devices */
 713                 sectors_per_dev = rs->ti->len * raid10_copies;
 714                 sector_div(sectors_per_dev, rs->md.raid_disks);
 715
 716                 rs->md.layout = raid10_format_to_md_layout(raid10_format,
 717                                                            raid10_copies);
 718                 rs->md.new_layout = rs->md.layout;
 719         } else if ((rs->raid_type->level > 1) &&
 720                    sector_div(sectors_per_dev,
 721                               (rs->md.raid_disks - rs->raid_type->parity_devs))) {
 722                 rs->ti->error = "Target length not divisible by number of data devices";
 723                 return -EINVAL;
 724         }
 725         rs->md.dev_sectors = sectors_per_dev;
 726
 727         /* Assume there are no metadata devices until the drives are parsed */
 728         rs->md.persistent = 0;
 729         rs->md.external = 1;
 730
 731         return 0;
 732 }
 733
 734 static void do_table_event(struct work_struct *ws)
 735 {
 736         struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
 737
 738         dm_table_event(rs->ti->table);
 739 }
 740
 741 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
 742 {
 743         struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
 744
 745         if (rs->raid_type->level == 1)
 746                 return md_raid1_congested(&rs->md, bits);
 747
 748         if (rs->raid_type->level == 10)
 749                 return md_raid10_congested(&rs->md, bits);
 750
 751         return md_raid5_congested(&rs->md, bits);
 752 }
 753
 754 /*
 755  * This structure is never routinely used by userspace, unlike md superblocks.
 756  * Devices with this superblock should only ever be accessed via device-mapper.
 757  */
 758 #define DM_RAID_MAGIC 0x64526D44
 759 struct dm_raid_superblock {
 760         __le32 magic;           /* "DmRd" */
 761         __le32 features;        /* Used to indicate possible future changes */
 762
 763         __le32 num_devices;     /* Number of devices in this array. (Max 64) */
 764         __le32 array_position;  /* The position of this drive in the array */
 765
 766         __le64 events;          /* Incremented by md when superblock updated */
 767         __le64 failed_devices;  /* Bit field of devices to indicate failures */
 768
 769         /*
 770          * This offset tracks the progress of the repair or replacement of
 771          * an individual drive.
 772          */
 773         __le64 disk_recovery_offset;
 774
 775         /*
 776          * This offset tracks the progress of the initial array
 777          * synchronisation/parity calculation.
 778          */
 779         __le64 array_resync_offset;
 780
 781         /*
 782          * RAID characteristics
 783          */
 784         __le32 level;
 785         __le32 layout;
 786         __le32 stripe_sectors;
 787
 788         /* Remainder of a logical block is zero-filled when writing (see super_sync()). */
 789 } __packed;
 790
 791 static int read_disk_sb(struct md_rdev *rdev, int size)
 792 {
 793         BUG_ON(!rdev->sb_page);
 794
 795         if (rdev->sb_loaded)
 796                 return 0;
 797
 798         if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, 1)) {
 799                 DMERR("Failed to read superblock of device at position %d",
 800                       rdev->raid_disk);
 801                 md_error(rdev->mddev, rdev);
 802                 return -EINVAL;
 803         }
 804
 805         rdev->sb_loaded = 1;
 806
 807         return 0;
 808 }
 809
 810 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
 811 {
 812         int i;
 813         uint64_t failed_devices;
 814         struct dm_raid_superblock *sb;
 815         struct raid_set *rs = container_of(mddev, struct raid_set, md);
 816
 817         sb = page_address(rdev->sb_page);
 818         failed_devices = le64_to_cpu(sb->failed_devices);
 819
 820         for (i = 0; i < mddev->raid_disks; i++)
 821                 if (!rs->dev[i].data_dev ||
 822                     test_bit(Faulty, &(rs->dev[i].rdev.flags)))
 823                         failed_devices |= (1ULL << i);
 824
 825         memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
 826
 827         sb->magic = cpu_to_le32(DM_RAID_MAGIC);
 828         sb->features = cpu_to_le32(0);  /* No features yet */
 829
 830         sb->num_devices = cpu_to_le32(mddev->raid_disks);
 831         sb->array_position = cpu_to_le32(rdev->raid_disk);
 832
 833         sb->events = cpu_to_le64(mddev->events);
 834         sb->failed_devices = cpu_to_le64(failed_devices);
 835
 836         sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
 837         sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
 838
 839         sb->level = cpu_to_le32(mddev->level);
 840         sb->layout = cpu_to_le32(mddev->layout);
 841         sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
 842 }
 843
 844 /*
 845  * super_load
 846  *
 847  * This function creates a superblock if one is not found on the device
 848  * and will decide which superblock to use if there's a choice.
 849  *
 850  * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
 851  */
 852 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
 853 {
 854         int ret;
 855         struct dm_raid_superblock *sb;
 856         struct dm_raid_superblock *refsb;
 857         uint64_t events_sb, events_refsb;
 858
 859         rdev->sb_start = 0;
 860         rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
 861         if (rdev->sb_size < sizeof(*sb) || rdev->sb_size > PAGE_SIZE) {
 862                 DMERR("superblock size of a logical block is no longer valid");
 863                 return -EINVAL;
 864         }
 865
 866         ret = read_disk_sb(rdev, rdev->sb_size);
 867         if (ret)
 868                 return ret;
 869
 870         sb = page_address(rdev->sb_page);
 871
 872         /*
 873          * Two cases that we want to write new superblocks and rebuild:
 874          * 1) New device (no matching magic number)
 875          * 2) Device specified for rebuild (!In_sync w/ offset == 0)
 876          */
 877         if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
 878             (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
 879                 super_sync(rdev->mddev, rdev);
 880
 881                 set_bit(FirstUse, &rdev->flags);
 882
 883                 /* Force writing of superblocks to disk */
 884                 set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags);
 885
 886                 /* Any superblock is better than none, choose that if given */
 887                 return refdev ? 0 : 1;
 888         }
 889
 890         if (!refdev)
 891                 return 1;
 892
 893         events_sb = le64_to_cpu(sb->events);
 894
 895         refsb = page_address(refdev->sb_page);
 896         events_refsb = le64_to_cpu(refsb->events);
 897
 898         return (events_sb > events_refsb) ? 1 : 0;
 899 }
 900
 901 static int super_init_validation(struct mddev *mddev, struct md_rdev *rdev)
 902 {
 903         int role;
 904         struct raid_set *rs = container_of(mddev, struct raid_set, md);
 905         uint64_t events_sb;
 906         uint64_t failed_devices;
 907         struct dm_raid_superblock *sb;
 908         uint32_t new_devs = 0;
 909         uint32_t rebuilds = 0;
 910         struct md_rdev *r;
 911         struct dm_raid_superblock *sb2;
 912
 913         sb = page_address(rdev->sb_page);
 914         events_sb = le64_to_cpu(sb->events);
 915         failed_devices = le64_to_cpu(sb->failed_devices);
 916
 917         /*
 918          * Initialise to 1 if this is a new superblock.
 919          */
 920         mddev->events = events_sb ? : 1;
 921
 922         /*
 923          * Reshaping is not currently allowed
 924          */
 925         if (le32_to_cpu(sb->level) != mddev->level) {
 926                 DMERR("Reshaping arrays not yet supported. (RAID level change)");
 927                 return -EINVAL;
 928         }
 929         if (le32_to_cpu(sb->layout) != mddev->layout) {
 930                 DMERR("Reshaping arrays not yet supported. (RAID layout change)");
 931                 DMERR("  0x%X vs 0x%X", le32_to_cpu(sb->layout), mddev->layout);
 932                 DMERR("  Old layout: %s w/ %d copies",
 933                       raid10_md_layout_to_format(le32_to_cpu(sb->layout)),
 934                       raid10_md_layout_to_copies(le32_to_cpu(sb->layout)));
 935                 DMERR("  New layout: %s w/ %d copies",
 936                       raid10_md_layout_to_format(mddev->layout),
 937                       raid10_md_layout_to_copies(mddev->layout));
 938                 return -EINVAL;
 939         }
 940         if (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors) {
 941                 DMERR("Reshaping arrays not yet supported. (stripe sectors change)");
 942                 return -EINVAL;
 943         }
 944
 945         /* We can only change the number of devices in RAID1 right now */
 946         if ((rs->raid_type->level != 1) &&
 947             (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) {
 948                 DMERR("Reshaping arrays not yet supported. (device count change)");
 949                 return -EINVAL;
 950         }
 951
 952         if (!(rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC)))
 953                 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
 954
 955         /*
 956          * During load, we set FirstUse if a new superblock was written.
 957          * There are two reasons we might not have a superblock:
 958          * 1) The array is brand new - in which case, all of the
 959          *    devices must have their In_sync bit set.  Also,
 960          *    recovery_cp must be 0, unless forced.
 961          * 2) This is a new device being added to an old array
 962          *    and the new device needs to be rebuilt - in which
 963          *    case the In_sync bit will /not/ be set and
 964          *    recovery_cp must be MaxSector.
 965          */
 966         rdev_for_each(r, mddev) {
 967                 if (!test_bit(In_sync, &r->flags)) {
 968                         DMINFO("Device %d specified for rebuild: "
 969                                "Clearing superblock", r->raid_disk);
 970                         rebuilds++;
 971                 } else if (test_bit(FirstUse, &r->flags))
 972                         new_devs++;
 973         }
 974
 975         if (!rebuilds) {
 976                 if (new_devs == mddev->raid_disks) {
 977                         DMINFO("Superblocks created for new array");
 978                         set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
 979                 } else if (new_devs) {
 980                         DMERR("New device injected "
 981                               "into existing array without 'rebuild' "
 982                               "parameter specified");
 983                         return -EINVAL;
 984                 }
 985         } else if (new_devs) {
 986                 DMERR("'rebuild' devices cannot be "
 987                       "injected into an array with other first-time devices");
 988                 return -EINVAL;
 989         } else if (mddev->recovery_cp != MaxSector) {
 990                 DMERR("'rebuild' specified while array is not in-sync");
 991                 return -EINVAL;
 992         }
 993
 994         /*
 995          * Now we set the Faulty bit for those devices that are
 996          * recorded in the superblock as failed.
 997          */
 998         rdev_for_each(r, mddev) {
 999                 if (!r->sb_page)
1000                         continue;
1001                 sb2 = page_address(r->sb_page);
1002                 sb2->failed_devices = 0;
1003
1004                 /*
1005                  * Check for any device re-ordering.
1006                  */
1007                 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
1008                         role = le32_to_cpu(sb2->array_position);
1009                         if (role != r->raid_disk) {
1010                                 if (rs->raid_type->level != 1) {
1011                                         rs->ti->error = "Cannot change device "
1012                                                 "positions in RAID array";
1013                                         return -EINVAL;
1014                                 }
1015                                 DMINFO("RAID1 device #%d now at position #%d",
1016                                        role, r->raid_disk);
1017                         }
1018
1019                         /*
1020                          * Partial recovery is performed on
1021                          * returning failed devices.
1022                          */
1023                         if (failed_devices & (1 << role))
1024                                 set_bit(Faulty, &r->flags);
1025                 }
1026         }
1027
1028         return 0;
1029 }
1030
1031 static int super_validate(struct mddev *mddev, struct md_rdev *rdev)
1032 {
1033         struct dm_raid_superblock *sb = page_address(rdev->sb_page);
1034
1035         /*
1036          * If mddev->events is not set, we know we have not yet initialized
1037          * the array.
1038          */
1039         if (!mddev->events && super_init_validation(mddev, rdev))
1040                 return -EINVAL;
1041
1042         mddev->bitmap_info.offset = 4096 >> 9; /* Enable bitmap creation */
1043         rdev->mddev->bitmap_info.default_offset = 4096 >> 9;
1044         if (!test_bit(FirstUse, &rdev->flags)) {
1045                 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
1046                 if (rdev->recovery_offset != MaxSector)
1047                         clear_bit(In_sync, &rdev->flags);
1048         }
1049
1050         /*
1051          * If a device comes back, set it as not In_sync and no longer faulty.
1052          */
1053         if (test_bit(Faulty, &rdev->flags)) {
1054                 clear_bit(Faulty, &rdev->flags);
1055                 clear_bit(In_sync, &rdev->flags);
1056                 rdev->saved_raid_disk = rdev->raid_disk;
1057                 rdev->recovery_offset = 0;
1058         }
1059
1060         clear_bit(FirstUse, &rdev->flags);
1061
1062         return 0;
1063 }
1064
1065 /*
1066  * Analyse superblocks and select the freshest.
1067  */
1068 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
1069 {
1070         int ret;
1071         struct raid_dev *dev;
1072         struct md_rdev *rdev, *tmp, *freshest;
1073         struct mddev *mddev = &rs->md;
1074
1075         freshest = NULL;
1076         rdev_for_each_safe(rdev, tmp, mddev) {
1077                 /*
1078                  * Skipping super_load due to DMPF_SYNC will cause
1079                  * the array to undergo initialization again as
1080                  * though it were new.  This is the intended effect
1081                  * of the "sync" directive.
1082                  *
1083                  * When reshaping capability is added, we must ensure
1084                  * that the "sync" directive is disallowed during the
1085                  * reshape.
1086                  */
1087                 if (rs->print_flags & DMPF_SYNC)
1088                         continue;
1089
1090                 if (!rdev->meta_bdev)
1091                         continue;
1092
1093                 ret = super_load(rdev, freshest);
1094
1095                 switch (ret) {
1096                 case 1:
1097                         freshest = rdev;
1098                         break;
1099                 case 0:
1100                         break;
1101                 default:
1102                         dev = container_of(rdev, struct raid_dev, rdev);
1103                         if (dev->meta_dev)
1104                                 dm_put_device(ti, dev->meta_dev);
1105
1106                         dev->meta_dev = NULL;
1107                         rdev->meta_bdev = NULL;
1108
1109                         if (rdev->sb_page)
1110                                 put_page(rdev->sb_page);
1111
1112                         rdev->sb_page = NULL;
1113
1114                         rdev->sb_loaded = 0;
1115
1116                         /*
1117                          * We might be able to salvage the data device
1118                          * even though the meta device has failed.  For
1119                          * now, we behave as though '- -' had been
1120                          * set for this device in the table.
1121                          */
1122                         if (dev->data_dev)
1123                                 dm_put_device(ti, dev->data_dev);
1124
1125                         dev->data_dev = NULL;
1126                         rdev->bdev = NULL;
1127
1128                         list_del(&rdev->same_set);
1129                 }
1130         }
1131
1132         if (!freshest)
1133                 return 0;
1134
1135         if (validate_raid_redundancy(rs)) {
1136                 rs->ti->error = "Insufficient redundancy to activate array";
1137                 return -EINVAL;
1138         }
1139
1140         /*
1141          * Validation of the freshest device provides the source of
1142          * validation for the remaining devices.
1143          */
1144         ti->error = "Unable to assemble array: Invalid superblocks";
1145         if (super_validate(mddev, freshest))
1146                 return -EINVAL;
1147
1148         rdev_for_each(rdev, mddev)
1149                 if ((rdev != freshest) && super_validate(mddev, rdev))
1150                         return -EINVAL;
1151
1152         return 0;
1153 }
1154
1155 /*
1156  * Construct a RAID4/5/6 mapping:
1157  * Args:
1158  *      <raid_type> <#raid_params> <raid_params>                \
1159  *      <#raid_devs> { <meta_dev1> <dev1> .. <meta_devN> <devN> }
1160  *
1161  * <raid_params> varies by <raid_type>.  See 'parse_raid_params' for
1162  * details on possible <raid_params>.
1163  */
1164 static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv)
1165 {
1166         int ret;
1167         struct raid_type *rt;
1168         unsigned long num_raid_params, num_raid_devs;
1169         struct raid_set *rs = NULL;
1170
1171         /* Must have at least <raid_type> <#raid_params> */
1172         if (argc < 2) {
1173                 ti->error = "Too few arguments";
1174                 return -EINVAL;
1175         }
1176
1177         /* raid type */
1178         rt = get_raid_type(argv[0]);
1179         if (!rt) {
1180                 ti->error = "Unrecognised raid_type";
1181                 return -EINVAL;
1182         }
1183         argc--;
1184         argv++;
1185
1186         /* number of RAID parameters */
1187         if (kstrtoul(argv[0], 10, &num_raid_params) < 0) {
1188                 ti->error = "Cannot understand number of RAID parameters";
1189                 return -EINVAL;
1190         }
1191         argc--;
1192         argv++;
1193
1194         /* Skip over RAID params for now and find out # of devices */
1195         if (num_raid_params + 1 > argc) {
1196                 ti->error = "Arguments do not agree with counts given";
1197                 return -EINVAL;
1198         }
1199
1200         if ((kstrtoul(argv[num_raid_params], 10, &num_raid_devs) < 0) ||
1201             (num_raid_devs >= INT_MAX)) {
1202                 ti->error = "Cannot understand number of raid devices";
1203                 return -EINVAL;
1204         }
1205
1206         rs = context_alloc(ti, rt, (unsigned)num_raid_devs);
1207         if (IS_ERR(rs))
1208                 return PTR_ERR(rs);
1209
1210         ret = parse_raid_params(rs, argv, (unsigned)num_raid_params);
1211         if (ret)
1212                 goto bad;
1213
1214         ret = -EINVAL;
1215
1216         argc -= num_raid_params + 1; /* +1: we already have num_raid_devs */
1217         argv += num_raid_params + 1;
1218
1219         if (argc != (num_raid_devs * 2)) {
1220                 ti->error = "Supplied RAID devices does not match the count given";
1221                 goto bad;
1222         }
1223
1224         ret = dev_parms(rs, argv);
1225         if (ret)
1226                 goto bad;
1227
1228         rs->md.sync_super = super_sync;
1229         ret = analyse_superblocks(ti, rs);
1230         if (ret)
1231                 goto bad;
1232
1233         INIT_WORK(&rs->md.event_work, do_table_event);
1234         ti->private = rs;
1235         ti->num_flush_bios = 1;
1236
1237         mutex_lock(&rs->md.reconfig_mutex);
1238         ret = md_run(&rs->md);
1239         rs->md.in_sync = 0; /* Assume already marked dirty */
1240         mutex_unlock(&rs->md.reconfig_mutex);
1241
1242         if (ret) {
1243                 ti->error = "Fail to run raid array";
1244                 goto bad;
1245         }
1246
1247         if (ti->len != rs->md.array_sectors) {
1248                 ti->error = "Array size does not match requested target length";
1249                 ret = -EINVAL;
1250                 goto size_mismatch;
1251         }
1252         rs->callbacks.congested_fn = raid_is_congested;
1253         dm_table_add_target_callbacks(ti->table, &rs->callbacks);
1254
1255         mddev_suspend(&rs->md);
1256         return 0;
1257
1258 size_mismatch:
1259         md_stop(&rs->md);
1260 bad:
1261         context_free(rs);
1262
1263         return ret;
1264 }
1265
1266 static void raid_dtr(struct dm_target *ti)
1267 {
1268         struct raid_set *rs = ti->private;
1269
1270         list_del_init(&rs->callbacks.list);
1271         md_stop(&rs->md);
1272         context_free(rs);
1273 }
1274
1275 static int raid_map(struct dm_target *ti, struct bio *bio)
1276 {
1277         struct raid_set *rs = ti->private;
1278         struct mddev *mddev = &rs->md;
1279
1280         mddev->pers->make_request(mddev, bio);
1281
1282         return DM_MAPIO_SUBMITTED;
1283 }
1284
1285 static const char *decipher_sync_action(struct mddev *mddev)
1286 {
1287         if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
1288                 return "frozen";
1289
1290         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1291             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
1292                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
1293                         return "reshape";
1294
1295                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1296                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1297                                 return "resync";
1298                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1299                                 return "check";
1300                         return "repair";
1301                 }
1302
1303                 if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
1304                         return "recover";
1305         }
1306
1307         return "idle";
1308 }
1309
1310 static void raid_status(struct dm_target *ti, status_type_t type,
1311                         unsigned status_flags, char *result, unsigned maxlen)
1312 {
1313         struct raid_set *rs = ti->private;
1314         unsigned raid_param_cnt = 1; /* at least 1 for chunksize */
1315         unsigned sz = 0;
1316         int i, array_in_sync = 0;
1317         sector_t sync;
1318
1319         switch (type) {
1320         case STATUSTYPE_INFO:
1321                 DMEMIT("%s %d ", rs->raid_type->name, rs->md.raid_disks);
1322
1323                 if (test_bit(MD_RECOVERY_RUNNING, &rs->md.recovery))
1324                         sync = rs->md.curr_resync_completed;
1325                 else
1326                         sync = rs->md.recovery_cp;
1327
1328                 if (sync >= rs->md.resync_max_sectors) {
1329                         /*
1330                          * Sync complete.
1331                          */
1332                         array_in_sync = 1;
1333                         sync = rs->md.resync_max_sectors;
1334                 } else if (test_bit(MD_RECOVERY_REQUESTED, &rs->md.recovery)) {
1335                         /*
1336                          * If "check" or "repair" is occurring, the array has
1337                          * undergone and initial sync and the health characters
1338                          * should not be 'a' anymore.
1339                          */
1340                         array_in_sync = 1;
1341                 } else {
1342                         /*
1343                          * The array may be doing an initial sync, or it may
1344                          * be rebuilding individual components.  If all the
1345                          * devices are In_sync, then it is the array that is
1346                          * being initialized.
1347                          */
1348                         for (i = 0; i < rs->md.raid_disks; i++)
1349                                 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
1350                                         array_in_sync = 1;
1351                 }
1352
1353                 /*
1354                  * Status characters:
1355                  *  'D' = Dead/Failed device
1356                  *  'a' = Alive but not in-sync
1357                  *  'A' = Alive and in-sync
1358                  */
1359                 for (i = 0; i < rs->md.raid_disks; i++) {
1360                         if (test_bit(Faulty, &rs->dev[i].rdev.flags))
1361                                 DMEMIT("D");
1362                         else if (!array_in_sync ||
1363                                  !test_bit(In_sync, &rs->dev[i].rdev.flags))
1364                                 DMEMIT("a");
1365                         else
1366                                 DMEMIT("A");
1367                 }
1368
1369                 /*
1370                  * In-sync ratio:
1371                  *  The in-sync ratio shows the progress of:
1372                  *   - Initializing the array
1373                  *   - Rebuilding a subset of devices of the array
1374                  *  The user can distinguish between the two by referring
1375                  *  to the status characters.
1376                  */
1377                 DMEMIT(" %llu/%llu",
1378                        (unsigned long long) sync,
1379                        (unsigned long long) rs->md.resync_max_sectors);
1380
1381                 /*
1382                  * Sync action:
1383                  *   See Documentation/device-mapper/dm-raid.c for
1384                  *   information on each of these states.
1385                  */
1386                 DMEMIT(" %s", decipher_sync_action(&rs->md));
1387
1388                 /*
1389                  * resync_mismatches/mismatch_cnt
1390                  *   This field shows the number of discrepancies found when
1391                  *   performing a "check" of the array.
1392                  */
1393                 DMEMIT(" %llu",
1394                        (strcmp(rs->md.last_sync_action, "check")) ? 0 :
1395                        (unsigned long long)
1396                        atomic64_read(&rs->md.resync_mismatches));
1397                 break;
1398         case STATUSTYPE_TABLE:
1399                 /* The string you would use to construct this array */
1400                 for (i = 0; i < rs->md.raid_disks; i++) {
1401                         if ((rs->print_flags & DMPF_REBUILD) &&
1402                             rs->dev[i].data_dev &&
1403                             !test_bit(In_sync, &rs->dev[i].rdev.flags))
1404                                 raid_param_cnt += 2; /* for rebuilds */
1405                         if (rs->dev[i].data_dev &&
1406                             test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1407                                 raid_param_cnt += 2;
1408                 }
1409
1410                 raid_param_cnt += (hweight32(rs->print_flags & ~DMPF_REBUILD) * 2);
1411                 if (rs->print_flags & (DMPF_SYNC | DMPF_NOSYNC))
1412                         raid_param_cnt--;
1413
1414                 DMEMIT("%s %u %u", rs->raid_type->name,
1415                        raid_param_cnt, rs->md.chunk_sectors);
1416
1417                 if ((rs->print_flags & DMPF_SYNC) &&
1418                     (rs->md.recovery_cp == MaxSector))
1419                         DMEMIT(" sync");
1420                 if (rs->print_flags & DMPF_NOSYNC)
1421                         DMEMIT(" nosync");
1422
1423                 for (i = 0; i < rs->md.raid_disks; i++)
1424                         if ((rs->print_flags & DMPF_REBUILD) &&
1425                             rs->dev[i].data_dev &&
1426                             !test_bit(In_sync, &rs->dev[i].rdev.flags))
1427                                 DMEMIT(" rebuild %u", i);
1428
1429                 if (rs->print_flags & DMPF_DAEMON_SLEEP)
1430                         DMEMIT(" daemon_sleep %lu",
1431                                rs->md.bitmap_info.daemon_sleep);
1432
1433                 if (rs->print_flags & DMPF_MIN_RECOVERY_RATE)
1434                         DMEMIT(" min_recovery_rate %d", rs->md.sync_speed_min);
1435
1436                 if (rs->print_flags & DMPF_MAX_RECOVERY_RATE)
1437                         DMEMIT(" max_recovery_rate %d", rs->md.sync_speed_max);
1438
1439                 for (i = 0; i < rs->md.raid_disks; i++)
1440                         if (rs->dev[i].data_dev &&
1441                             test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1442                                 DMEMIT(" write_mostly %u", i);
1443
1444                 if (rs->print_flags & DMPF_MAX_WRITE_BEHIND)
1445                         DMEMIT(" max_write_behind %lu",
1446                                rs->md.bitmap_info.max_write_behind);
1447
1448                 if (rs->print_flags & DMPF_STRIPE_CACHE) {
1449                         struct r5conf *conf = rs->md.private;
1450
1451                         /* convert from kiB to sectors */
1452                         DMEMIT(" stripe_cache %d",
1453                                conf ? conf->max_nr_stripes * 2 : 0);
1454                 }
1455
1456                 if (rs->print_flags & DMPF_REGION_SIZE)
1457                         DMEMIT(" region_size %lu",
1458                                rs->md.bitmap_info.chunksize >> 9);
1459
1460                 if (rs->print_flags & DMPF_RAID10_COPIES)
1461                         DMEMIT(" raid10_copies %u",
1462                                raid10_md_layout_to_copies(rs->md.layout));
1463
1464                 if (rs->print_flags & DMPF_RAID10_FORMAT)
1465                         DMEMIT(" raid10_format %s",
1466                                raid10_md_layout_to_format(rs->md.layout));
1467
1468                 DMEMIT(" %d", rs->md.raid_disks);
1469                 for (i = 0; i < rs->md.raid_disks; i++) {
1470                         if (rs->dev[i].meta_dev)
1471                                 DMEMIT(" %s", rs->dev[i].meta_dev->name);
1472                         else
1473                                 DMEMIT(" -");
1474
1475                         if (rs->dev[i].data_dev)
1476                                 DMEMIT(" %s", rs->dev[i].data_dev->name);
1477                         else
1478                                 DMEMIT(" -");
1479                 }
1480         }
1481 }
1482
1483 static int raid_message(struct dm_target *ti, unsigned argc, char **argv)
1484 {
1485         struct raid_set *rs = ti->private;
1486         struct mddev *mddev = &rs->md;
1487
1488         if (!strcasecmp(argv[0], "reshape")) {
1489                 DMERR("Reshape not supported.");
1490                 return -EINVAL;
1491         }
1492
1493         if (!mddev->pers || !mddev->pers->sync_request)
1494                 return -EINVAL;
1495
1496         if (!strcasecmp(argv[0], "frozen"))
1497                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1498         else
1499                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1500
1501         if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
1502                 if (mddev->sync_thread) {
1503                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1504                         md_reap_sync_thread(mddev);
1505                 }
1506         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1507                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1508                 return -EBUSY;
1509         else if (!strcasecmp(argv[0], "resync"))
1510                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1511         else if (!strcasecmp(argv[0], "recover")) {
1512                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
1513                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1514         } else {
1515                 if (!strcasecmp(argv[0], "check"))
1516                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1517                 else if (!!strcasecmp(argv[0], "repair"))
1518                         return -EINVAL;
1519                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1520                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1521         }
1522         if (mddev->ro == 2) {
1523                 /* A write to sync_action is enough to justify
1524                  * canceling read-auto mode
1525                  */
1526                 mddev->ro = 0;
1527                 if (!mddev->suspended)
1528                         md_wakeup_thread(mddev->sync_thread);
1529         }
1530         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1531         if (!mddev->suspended)
1532                 md_wakeup_thread(mddev->thread);
1533
1534         return 0;
1535 }
1536
1537 static int raid_iterate_devices(struct dm_target *ti,
1538                                 iterate_devices_callout_fn fn, void *data)
1539 {
1540         struct raid_set *rs = ti->private;
1541         unsigned i;
1542         int ret = 0;
1543
1544         for (i = 0; !ret && i < rs->md.raid_disks; i++)
1545                 if (rs->dev[i].data_dev)
1546                         ret = fn(ti,
1547                                  rs->dev[i].data_dev,
1548                                  0, /* No offset on data devs */
1549                                  rs->md.dev_sectors,
1550                                  data);
1551
1552         return ret;
1553 }
1554
1555 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
1556 {
1557         struct raid_set *rs = ti->private;
1558         unsigned chunk_size = rs->md.chunk_sectors << 9;
1559         struct r5conf *conf = rs->md.private;
1560
1561         blk_limits_io_min(limits, chunk_size);
1562         blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded));
1563 }
1564
1565 static void raid_presuspend(struct dm_target *ti)
1566 {
1567         struct raid_set *rs = ti->private;
1568
1569         md_stop_writes(&rs->md);
1570 }
1571
1572 static void raid_postsuspend(struct dm_target *ti)
1573 {
1574         struct raid_set *rs = ti->private;
1575
1576         mddev_suspend(&rs->md);
1577 }
1578
1579 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
1580 {
1581         int i;
1582         uint64_t failed_devices, cleared_failed_devices = 0;
1583         unsigned long flags;
1584         struct dm_raid_superblock *sb;
1585         struct md_rdev *r;
1586
1587         for (i = 0; i < rs->md.raid_disks; i++) {
1588                 r = &rs->dev[i].rdev;
1589                 if (test_bit(Faulty, &r->flags) && r->sb_page &&
1590                     sync_page_io(r, 0, r->sb_size, r->sb_page, READ, 1)) {
1591                         DMINFO("Faulty %s device #%d has readable super block."
1592                                "  Attempting to revive it.",
1593                                rs->raid_type->name, i);
1594
1595                         /*
1596                          * Faulty bit may be set, but sometimes the array can
1597                          * be suspended before the personalities can respond
1598                          * by removing the device from the array (i.e. calling
1599                          * 'hot_remove_disk').  If they haven't yet removed
1600                          * the failed device, its 'raid_disk' number will be
1601                          * '>= 0' - meaning we must call this function
1602                          * ourselves.
1603                          */
1604                         if ((r->raid_disk >= 0) &&
1605                             (r->mddev->pers->hot_remove_disk(r->mddev, r) != 0))
1606                                 /* Failed to revive this device, try next */
1607                                 continue;
1608
1609                         r->raid_disk = i;
1610                         r->saved_raid_disk = i;
1611                         flags = r->flags;
1612                         clear_bit(Faulty, &r->flags);
1613                         clear_bit(WriteErrorSeen, &r->flags);
1614                         clear_bit(In_sync, &r->flags);
1615                         if (r->mddev->pers->hot_add_disk(r->mddev, r)) {
1616                                 r->raid_disk = -1;
1617                                 r->saved_raid_disk = -1;
1618                                 r->flags = flags;
1619                         } else {
1620                                 r->recovery_offset = 0;
1621                                 cleared_failed_devices |= 1 << i;
1622                         }
1623                 }
1624         }
1625         if (cleared_failed_devices) {
1626                 rdev_for_each(r, &rs->md) {
1627                         sb = page_address(r->sb_page);
1628                         failed_devices = le64_to_cpu(sb->failed_devices);
1629                         failed_devices &= ~cleared_failed_devices;
1630                         sb->failed_devices = cpu_to_le64(failed_devices);
1631                 }
1632         }
1633 }
1634
1635 static void raid_resume(struct dm_target *ti)
1636 {
1637         struct raid_set *rs = ti->private;
1638
1639         set_bit(MD_CHANGE_DEVS, &rs->md.flags);
1640         if (!rs->bitmap_loaded) {
1641                 bitmap_load(&rs->md);
1642                 rs->bitmap_loaded = 1;
1643         } else {
1644                 /*
1645                  * A secondary resume while the device is active.
1646                  * Take this opportunity to check whether any failed
1647                  * devices are reachable again.
1648                  */
1649                 attempt_restore_of_faulty_devices(rs);
1650         }
1651
1652         clear_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
1653         mddev_resume(&rs->md);
1654 }
1655
1656 static struct target_type raid_target = {
1657         .name = "raid",
1658         .version = {1, 5, 2},
1659         .module = THIS_MODULE,
1660         .ctr = raid_ctr,
1661         .dtr = raid_dtr,
1662         .map = raid_map,
1663         .status = raid_status,
1664         .message = raid_message,
1665         .iterate_devices = raid_iterate_devices,
1666         .io_hints = raid_io_hints,
1667         .presuspend = raid_presuspend,
1668         .postsuspend = raid_postsuspend,
1669         .resume = raid_resume,
1670 };
1671
1672 static int __init dm_raid_init(void)
1673 {
1674         DMINFO("Loading target version %u.%u.%u",
1675                raid_target.version[0],
1676                raid_target.version[1],
1677                raid_target.version[2]);
1678         return dm_register_target(&raid_target);
1679 }
1680
1681 static void __exit dm_raid_exit(void)
1682 {
1683         dm_unregister_target(&raid_target);
1684 }
1685
1686 module_init(dm_raid_init);
1687 module_exit(dm_raid_exit);
1688
1689 MODULE_DESCRIPTION(DM_NAME " raid4/5/6 target");
1690 MODULE_ALIAS("dm-raid1");
1691 MODULE_ALIAS("dm-raid10");
1692 MODULE_ALIAS("dm-raid4");
1693 MODULE_ALIAS("dm-raid5");
1694 MODULE_ALIAS("dm-raid6");
1695 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
1696 MODULE_LICENSE("GPL");