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