drivers/md/dm-raid.c

   1 /*
   2  * Copyright (C) 2010-2011 Neil Brown
   3  * Copyright (C) 2010-2016 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 #define MAX_RAID_DEVICES        253 /* md-raid kernel limit */
  21
  22 static bool devices_handle_discard_safely = false;
  23
  24 /*
  25  * The following flags are used by dm-raid.c to set up the array state.
  26  * They must be cleared before md_run is called.
  27  */
  28 #define FirstUse 10             /* rdev flag */
  29
  30 struct raid_dev {
  31         /*
  32          * Two DM devices, one to hold metadata and one to hold the
  33          * actual data/parity.  The reason for this is to not confuse
  34          * ti->len and give more flexibility in altering size and
  35          * characteristics.
  36          *
  37          * While it is possible for this device to be associated
  38          * with a different physical device than the data_dev, it
  39          * is intended for it to be the same.
  40          *    |--------- Physical Device ---------|
  41          *    |- meta_dev -|------ data_dev ------|
  42          */
  43         struct dm_dev *meta_dev;
  44         struct dm_dev *data_dev;
  45         struct md_rdev rdev;
  46 };
  47
  48 /*
  49  * Flags for rs->ctr_flags field.
  50  *
  51  * 1 = no flag value
  52  * 2 = flag with value
  53  */
  54 #define CTR_FLAG_SYNC              0x1   /* 1 */ /* Not with raid0! */
  55 #define CTR_FLAG_NOSYNC            0x2   /* 1 */ /* Not with raid0! */
  56 #define CTR_FLAG_REBUILD           0x4   /* 2 */ /* Not with raid0! */
  57 #define CTR_FLAG_DAEMON_SLEEP      0x8   /* 2 */ /* Not with raid0! */
  58 #define CTR_FLAG_MIN_RECOVERY_RATE 0x10  /* 2 */ /* Not with raid0! */
  59 #define CTR_FLAG_MAX_RECOVERY_RATE 0x20  /* 2 */ /* Not with raid0! */
  60 #define CTR_FLAG_MAX_WRITE_BEHIND  0x40  /* 2 */ /* Only with raid1! */
  61 #define CTR_FLAG_WRITE_MOSTLY      0x80  /* 2 */ /* Only with raid1! */
  62 #define CTR_FLAG_STRIPE_CACHE      0x100 /* 2 */ /* Only with raid4/5/6! */
  63 #define CTR_FLAG_REGION_SIZE       0x200 /* 2 */ /* Not with raid0! */
  64 #define CTR_FLAG_RAID10_COPIES     0x400 /* 2 */ /* Only with raid10 */
  65 #define CTR_FLAG_RAID10_FORMAT     0x800 /* 2 */ /* Only with raid10 */
  66
  67 /*
  68  * Definitions of various constructor flags to
  69  * be used in checks of valid / invalid flags
  70  * per raid level.
  71  */
  72 /* Define all any sync flags */
  73 #define CTR_FLAGS_ANY_SYNC              (CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
  74
  75 /* Define flags for options without argument (e.g. 'nosync') */
  76 #define CTR_FLAG_OPTIONS_NO_ARGS        CTR_FLAGS_ANY_SYNC
  77
  78 /* Define flags for options with one argument (e.g. 'delta_disks +2') */
  79 #define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
  80                                   CTR_FLAG_WRITE_MOSTLY | \
  81                                   CTR_FLAG_DAEMON_SLEEP | \
  82                                   CTR_FLAG_MIN_RECOVERY_RATE | \
  83                                   CTR_FLAG_MAX_RECOVERY_RATE | \
  84                                   CTR_FLAG_MAX_WRITE_BEHIND | \
  85                                   CTR_FLAG_STRIPE_CACHE | \
  86                                   CTR_FLAG_REGION_SIZE | \
  87                                   CTR_FLAG_RAID10_COPIES | \
  88                                   CTR_FLAG_RAID10_FORMAT)
  89
  90 /* All ctr optional arguments */
  91 #define ALL_CTR_FLAGS           (CTR_FLAG_OPTIONS_NO_ARGS | \
  92                                  CTR_FLAG_OPTIONS_ONE_ARG)
  93
  94 /* Invalid options definitions per raid level... */
  95
  96 /* "raid0" does not accept any options */
  97 #define RAID0_INVALID_FLAGS ALL_CTR_FLAGS
  98
  99 /* "raid1" does not accept stripe cache or any raid10 options */
 100 #define RAID1_INVALID_FLAGS     (CTR_FLAG_STRIPE_CACHE | \
 101                                  CTR_FLAG_RAID10_COPIES | \
 102                                  CTR_FLAG_RAID10_FORMAT)
 103
 104 /* "raid10" does not accept any raid1 or stripe cache options */
 105 #define RAID10_INVALID_FLAGS    (CTR_FLAG_WRITE_MOSTLY | \
 106                                  CTR_FLAG_MAX_WRITE_BEHIND | \
 107                                  CTR_FLAG_STRIPE_CACHE)
 108 /*
 109  * "raid4/5/6" do not accept any raid1 or raid10 specific options
 110  *
 111  * "raid6" does not accept "nosync", because it is not guaranteed
 112  * that both parity and q-syndrome are being written properly with
 113  * any writes
 114  */
 115 #define RAID45_INVALID_FLAGS    (CTR_FLAG_WRITE_MOSTLY | \
 116                                  CTR_FLAG_MAX_WRITE_BEHIND | \
 117                                  CTR_FLAG_RAID10_FORMAT | \
 118                                  CTR_FLAG_RAID10_COPIES)
 119 #define RAID6_INVALID_FLAGS     (CTR_FLAG_NOSYNC | RAID45_INVALID_FLAGS)
 120 /* ...invalid options definitions per raid level */
 121
 122 struct raid_set {
 123         struct dm_target *ti;
 124
 125         uint32_t bitmap_loaded;
 126         uint32_t ctr_flags;
 127
 128         struct mddev md;
 129         struct raid_type *raid_type;
 130         struct dm_target_callbacks callbacks;
 131
 132         struct raid_dev dev[0];
 133 };
 134
 135 /* Supported raid types and properties. */
 136 static struct raid_type {
 137         const char *name;               /* RAID algorithm. */
 138         const char *descr;              /* Descriptor text for logging. */
 139         const unsigned parity_devs;     /* # of parity devices. */
 140         const unsigned minimal_devs;    /* minimal # of devices in set. */
 141         const unsigned level;           /* RAID level. */
 142         const unsigned algorithm;       /* RAID algorithm. */
 143 } raid_types[] = {
 144         {"raid0",    "RAID0 (striping)",                0, 2, 0, 0 /* NONE */},
 145         {"raid1",    "RAID1 (mirroring)",               0, 2, 1, 0 /* NONE */},
 146         {"raid10",   "RAID10 (striped mirrors)",        0, 2, 10, UINT_MAX /* Varies */},
 147         {"raid4",    "RAID4 (dedicated parity disk)",   1, 2, 5, ALGORITHM_PARITY_0},
 148         {"raid5_la", "RAID5 (left asymmetric)",         1, 2, 5, ALGORITHM_LEFT_ASYMMETRIC},
 149         {"raid5_ra", "RAID5 (right asymmetric)",        1, 2, 5, ALGORITHM_RIGHT_ASYMMETRIC},
 150         {"raid5_ls", "RAID5 (left symmetric)",          1, 2, 5, ALGORITHM_LEFT_SYMMETRIC},
 151         {"raid5_rs", "RAID5 (right symmetric)",         1, 2, 5, ALGORITHM_RIGHT_SYMMETRIC},
 152         {"raid6_zr", "RAID6 (zero restart)",            2, 4, 6, ALGORITHM_ROTATING_ZERO_RESTART},
 153         {"raid6_nr", "RAID6 (N restart)",               2, 4, 6, ALGORITHM_ROTATING_N_RESTART},
 154         {"raid6_nc", "RAID6 (N continue)",              2, 4, 6, ALGORITHM_ROTATING_N_CONTINUE}
 155 };
 156
 157 /* True, if @v is in inclusive range [@min, @max] */
 158 static bool _in_range(long v, long min, long max)
 159 {
 160         return v >= min && v <= max;
 161 }
 162
 163 /* ctr flag bit manipulation... */
 164 /* Set single @flag in @flags */
 165 static void _set_flag(uint32_t flag, uint32_t *flags)
 166 {
 167         WARN_ON_ONCE(hweight32(flag) != 1);
 168         *flags |= flag;
 169 }
 170
 171 /* Test single @flag in @flags */
 172 static bool _test_flag(uint32_t flag, uint32_t flags)
 173 {
 174         WARN_ON_ONCE(hweight32(flag) != 1);
 175         return (flag & flags) ? true : false;
 176 }
 177
 178 /* Test multiple @flags in @all_flags */
 179 static bool _test_flags(uint32_t flags, uint32_t all_flags)
 180 {
 181         return (flags & all_flags) ? true : false;
 182 }
 183
 184 /* Return true if single @flag is set in @*flags, else set it and return false */
 185 static bool _test_and_set_flag(uint32_t flag, uint32_t *flags)
 186 {
 187         if (_test_flag(flag, *flags))
 188                 return true;
 189
 190         _set_flag(flag, flags);
 191         return false;
 192 }
 193 /* ...ctr and runtime flag bit manipulation */
 194
 195 /* All table line arguments are defined here */
 196 static struct arg_name_flag {
 197         const uint32_t flag;
 198         const char *name;
 199 } _arg_name_flags[] = {
 200         { CTR_FLAG_SYNC, "sync"},
 201         { CTR_FLAG_NOSYNC, "nosync"},
 202         { CTR_FLAG_REBUILD, "rebuild"},
 203         { CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
 204         { CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
 205         { CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
 206         { CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
 207         { CTR_FLAG_WRITE_MOSTLY, "writemostly"},
 208         { CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
 209         { CTR_FLAG_REGION_SIZE, "region_size"},
 210         { CTR_FLAG_RAID10_COPIES, "raid10_copies"},
 211         { CTR_FLAG_RAID10_FORMAT, "raid10_format"},
 212 };
 213
 214 /* Return argument name string for given @flag */
 215 static const char *_argname_by_flag(const uint32_t flag)
 216 {
 217         if (hweight32(flag) == 1) {
 218                 struct arg_name_flag *anf = _arg_name_flags + ARRAY_SIZE(_arg_name_flags);
 219
 220                 while (anf-- > _arg_name_flags)
 221                         if (_test_flag(flag, anf->flag))
 222                                 return anf->name;
 223
 224         } else
 225                 DMERR("%s called with more than one flag!", __func__);
 226
 227         return NULL;
 228 }
 229
 230 /*
 231  * bool helpers to test for various raid levels of a raid type
 232  */
 233
 234 /* Return true, if raid type in @rt is raid0 */
 235 static bool rt_is_raid0(struct raid_type *rt)
 236 {
 237         return !rt->level;
 238 }
 239
 240 /* Return true, if raid type in @rt is raid1 */
 241 static bool rt_is_raid1(struct raid_type *rt)
 242 {
 243         return rt->level == 1;
 244 }
 245
 246 /* Return true, if raid type in @rt is raid10 */
 247 static bool rt_is_raid10(struct raid_type *rt)
 248 {
 249         return rt->level == 10;
 250 }
 251
 252 /* Return true, if raid type in @rt is raid4/5 */
 253 static bool rt_is_raid45(struct raid_type *rt)
 254 {
 255         return _in_range(rt->level, 4, 5);
 256 }
 257
 258 /* Return true, if raid type in @rt is raid6 */
 259 static bool rt_is_raid6(struct raid_type *rt)
 260 {
 261         return rt->level == 6;
 262 }
 263 /* END: raid level bools */
 264
 265 /*
 266  * Convenience functions to set ti->error to @errmsg and
 267  * return @r in order to shorten code in a lot of places
 268  */
 269 static int ti_error_ret(struct dm_target *ti, const char *errmsg, int r)
 270 {
 271         ti->error = (char *) errmsg;
 272         return r;
 273 }
 274
 275 static int ti_error_einval(struct dm_target *ti, const char *errmsg)
 276 {
 277         return ti_error_ret(ti, errmsg, -EINVAL);
 278 }
 279 /* END: convenience functions to set ti->error to @errmsg... */
 280
 281 /* Return invalid ctr flags for the raid level of @rs */
 282 static uint32_t _invalid_flags(struct raid_set *rs)
 283 {
 284         if (rt_is_raid0(rs->raid_type))
 285                 return RAID0_INVALID_FLAGS;
 286         else if (rt_is_raid1(rs->raid_type))
 287                 return RAID1_INVALID_FLAGS;
 288         else if (rt_is_raid10(rs->raid_type))
 289                 return RAID10_INVALID_FLAGS;
 290         else if (rt_is_raid45(rs->raid_type))
 291                 return RAID45_INVALID_FLAGS;
 292         else if (rt_is_raid6(rs->raid_type))
 293                 return RAID6_INVALID_FLAGS;
 294
 295         return ~0;
 296 }
 297
 298 /*
 299  * Check for any invalid flags set on @rs defined by bitset @invalid_flags
 300  *
 301  * Has to be called after parsing of the ctr flags!
 302  */
 303 static int rs_check_for_invalid_flags(struct raid_set *rs)
 304 {
 305         if (_test_flags(rs->ctr_flags, _invalid_flags(rs)))
 306                 return ti_error_einval(rs->ti, "Invalid flag combined");
 307
 308         return 0;
 309 }
 310
 311 static char *raid10_md_layout_to_format(int layout)
 312 {
 313         /*
 314          * Bit 16 and 17 stand for "offset" and "use_far_sets"
 315          * Refer to MD's raid10.c for details
 316          */
 317         if ((layout & 0x10000) && (layout & 0x20000))
 318                 return "offset";
 319
 320         if ((layout & 0xFF) > 1)
 321                 return "near";
 322
 323         return "far";
 324 }
 325
 326 static unsigned raid10_md_layout_to_copies(int layout)
 327 {
 328         if ((layout & 0xFF) > 1)
 329                 return layout & 0xFF;
 330         return (layout >> 8) & 0xFF;
 331 }
 332
 333 static int raid10_format_to_md_layout(char *format, unsigned copies)
 334 {
 335         unsigned n = 1, f = 1;
 336
 337         if (!strcasecmp("near", format))
 338                 n = copies;
 339         else
 340                 f = copies;
 341
 342         if (!strcasecmp("offset", format))
 343                 return 0x30000 | (f << 8) | n;
 344
 345         if (!strcasecmp("far", format))
 346                 return 0x20000 | (f << 8) | n;
 347
 348         return (f << 8) | n;
 349 }
 350
 351 static struct raid_type *get_raid_type(const char *name)
 352 {
 353         int i;
 354
 355         for (i = 0; i < ARRAY_SIZE(raid_types); i++)
 356                 if (!strcmp(raid_types[i].name, name))
 357                         return &raid_types[i];
 358
 359         return NULL;
 360 }
 361
 362 static struct raid_set *context_alloc(struct dm_target *ti, struct raid_type *raid_type, unsigned raid_devs)
 363 {
 364         unsigned i;
 365         struct raid_set *rs;
 366
 367         if (raid_devs <= raid_type->parity_devs)
 368                 return ERR_PTR(ti_error_einval(ti, "Insufficient number of devices"));
 369
 370         rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
 371         if (!rs)
 372                 return ERR_PTR(ti_error_ret(ti, "Cannot allocate raid context", -ENOMEM));
 373
 374         mddev_init(&rs->md);
 375
 376         rs->ti = ti;
 377         rs->raid_type = raid_type;
 378         rs->md.raid_disks = raid_devs;
 379         rs->md.level = raid_type->level;
 380         rs->md.new_level = rs->md.level;
 381         rs->md.layout = raid_type->algorithm;
 382         rs->md.new_layout = rs->md.layout;
 383         rs->md.delta_disks = 0;
 384         rs->md.recovery_cp = 0;
 385
 386         for (i = 0; i < raid_devs; i++)
 387                 md_rdev_init(&rs->dev[i].rdev);
 388
 389         /*
 390          * Remaining items to be initialized by further RAID params:
 391          *  rs->md.persistent
 392          *  rs->md.external
 393          *  rs->md.chunk_sectors
 394          *  rs->md.new_chunk_sectors
 395          *  rs->md.dev_sectors
 396          */
 397
 398         return rs;
 399 }
 400
 401 static void context_free(struct raid_set *rs)
 402 {
 403         int i;
 404
 405         for (i = 0; i < rs->md.raid_disks; i++) {
 406                 if (rs->dev[i].meta_dev)
 407                         dm_put_device(rs->ti, rs->dev[i].meta_dev);
 408                 md_rdev_clear(&rs->dev[i].rdev);
 409                 if (rs->dev[i].data_dev)
 410                         dm_put_device(rs->ti, rs->dev[i].data_dev);
 411         }
 412
 413         kfree(rs);
 414 }
 415
 416 /*
 417  * For every device we have two words
 418  *  <meta_dev>: meta device name or '-' if missing
 419  *  <data_dev>: data device name or '-' if missing
 420  *
 421  * The following are permitted:
 422  *    - -
 423  *    - <data_dev>
 424  *    <meta_dev> <data_dev>
 425  *
 426  * The following is not allowed:
 427  *    <meta_dev> -
 428  *
 429  * This code parses those words.  If there is a failure,
 430  * the caller must use context_free to unwind the operations.
 431  */
 432 static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
 433 {
 434         int i;
 435         int rebuild = 0;
 436         int metadata_available = 0;
 437         int r = 0;
 438         const char *arg;
 439
 440         /* Put off the number of raid devices argument to get to dev pairs */
 441         arg = dm_shift_arg(as);
 442         if (!arg)
 443                 return -EINVAL;
 444
 445         for (i = 0; i < rs->md.raid_disks; i++) {
 446                 rs->dev[i].rdev.raid_disk = i;
 447
 448                 rs->dev[i].meta_dev = NULL;
 449                 rs->dev[i].data_dev = NULL;
 450
 451                 /*
 452                  * There are no offsets, since there is a separate device
 453                  * for data and metadata.
 454                  */
 455                 rs->dev[i].rdev.data_offset = 0;
 456                 rs->dev[i].rdev.mddev = &rs->md;
 457
 458                 arg = dm_shift_arg(as);
 459                 if (!arg)
 460                         return -EINVAL;
 461
 462                 if (strcmp(arg, "-")) {
 463                         r = dm_get_device(rs->ti, arg,
 464                                             dm_table_get_mode(rs->ti->table),
 465                                             &rs->dev[i].meta_dev);
 466                         if (r)
 467                                 return ti_error_ret(rs->ti, "RAID metadata device lookup failure", r);
 468
 469                         rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
 470                         if (!rs->dev[i].rdev.sb_page)
 471                                 return ti_error_ret(rs->ti, "Failed to allocate superblock page", -ENOMEM);
 472                 }
 473
 474                 arg = dm_shift_arg(as);
 475                 if (!arg)
 476                         return -EINVAL;
 477
 478                 if (!strcmp(arg, "-")) {
 479                         if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
 480                             (!rs->dev[i].rdev.recovery_offset))
 481                                 return ti_error_einval(rs->ti, "Drive designated for rebuild not specified");
 482
 483                         if (rs->dev[i].meta_dev)
 484                                 return ti_error_einval(rs->ti, "No data device supplied with metadata device");
 485
 486                         continue;
 487                 }
 488
 489                 r = dm_get_device(rs->ti, arg,
 490                                     dm_table_get_mode(rs->ti->table),
 491                                     &rs->dev[i].data_dev);
 492                 if (r)
 493                         return ti_error_ret(rs->ti, "RAID device lookup failure", r);
 494
 495                 if (rs->dev[i].meta_dev) {
 496                         metadata_available = 1;
 497                         rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
 498                 }
 499                 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
 500                 list_add(&rs->dev[i].rdev.same_set, &rs->md.disks);
 501                 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
 502                         rebuild++;
 503         }
 504
 505         if (metadata_available) {
 506                 rs->md.external = 0;
 507                 rs->md.persistent = 1;
 508                 rs->md.major_version = 2;
 509         } else if (rebuild && !rs->md.recovery_cp) {
 510                 /*
 511                  * Without metadata, we will not be able to tell if the array
 512                  * is in-sync or not - we must assume it is not.  Therefore,
 513                  * it is impossible to rebuild a drive.
 514                  *
 515                  * Even if there is metadata, the on-disk information may
 516                  * indicate that the array is not in-sync and it will then
 517                  * fail at that time.
 518                  *
 519                  * User could specify 'nosync' option if desperate.
 520                  */
 521                 DMERR("Unable to rebuild drive while array is not in-sync");
 522                 return ti_error_einval(rs->ti, "Unable to rebuild drive while array is not in-sync");
 523         }
 524
 525         return 0;
 526 }
 527
 528 /*
 529  * validate_region_size
 530  * @rs
 531  * @region_size:  region size in sectors.  If 0, pick a size (4MiB default).
 532  *
 533  * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
 534  * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
 535  *
 536  * Returns: 0 on success, -EINVAL on failure.
 537  */
 538 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
 539 {
 540         unsigned long min_region_size = rs->ti->len / (1 << 21);
 541
 542         if (!region_size) {
 543                 /*
 544                  * Choose a reasonable default.  All figures in sectors.
 545                  */
 546                 if (min_region_size > (1 << 13)) {
 547                         /* If not a power of 2, make it the next power of 2 */
 548                         region_size = roundup_pow_of_two(min_region_size);
 549                         DMINFO("Choosing default region size of %lu sectors",
 550                                region_size);
 551                 } else {
 552                         DMINFO("Choosing default region size of 4MiB");
 553                         region_size = 1 << 13; /* sectors */
 554                 }
 555         } else {
 556                 /*
 557                  * Validate user-supplied value.
 558                  */
 559                 if (region_size > rs->ti->len)
 560                         return ti_error_einval(rs->ti, "Supplied region size is too large");
 561
 562                 if (region_size < min_region_size) {
 563                         DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
 564                               region_size, min_region_size);
 565                         return ti_error_einval(rs->ti, "Supplied region size is too small");
 566                 }
 567
 568                 if (!is_power_of_2(region_size))
 569                         return ti_error_einval(rs->ti, "Region size is not a power of 2");
 570
 571                 if (region_size < rs->md.chunk_sectors)
 572                         return ti_error_einval(rs->ti, "Region size is smaller than the chunk size");
 573         }
 574
 575         /*
 576          * Convert sectors to bytes.
 577          */
 578         rs->md.bitmap_info.chunksize = (region_size << 9);
 579
 580         return 0;
 581 }
 582
 583 /*
 584  * validate_raid_redundancy
 585  * @rs
 586  *
 587  * Determine if there are enough devices in the array that haven't
 588  * failed (or are being rebuilt) to form a usable array.
 589  *
 590  * Returns: 0 on success, -EINVAL on failure.
 591  */
 592 static int validate_raid_redundancy(struct raid_set *rs)
 593 {
 594         unsigned i, rebuild_cnt = 0;
 595         unsigned rebuilds_per_group = 0, copies, d;
 596         unsigned group_size, last_group_start;
 597
 598         for (i = 0; i < rs->md.raid_disks; i++)
 599                 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
 600                     !rs->dev[i].rdev.sb_page)
 601                         rebuild_cnt++;
 602
 603         switch (rs->raid_type->level) {
 604         case 1:
 605                 if (rebuild_cnt >= rs->md.raid_disks)
 606                         goto too_many;
 607                 break;
 608         case 4:
 609         case 5:
 610         case 6:
 611                 if (rebuild_cnt > rs->raid_type->parity_devs)
 612                         goto too_many;
 613                 break;
 614         case 10:
 615                 copies = raid10_md_layout_to_copies(rs->md.layout);
 616                 if (rebuild_cnt < copies)
 617                         break;
 618
 619                 /*
 620                  * It is possible to have a higher rebuild count for RAID10,
 621                  * as long as the failed devices occur in different mirror
 622                  * groups (i.e. different stripes).
 623                  *
 624                  * When checking "near" format, make sure no adjacent devices
 625                  * have failed beyond what can be handled.  In addition to the
 626                  * simple case where the number of devices is a multiple of the
 627                  * number of copies, we must also handle cases where the number
 628                  * of devices is not a multiple of the number of copies.
 629                  * E.g.    dev1 dev2 dev3 dev4 dev5
 630                  *          A    A    B    B    C
 631                  *          C    D    D    E    E
 632                  */
 633                 if (!strcmp("near", raid10_md_layout_to_format(rs->md.layout))) {
 634                         for (i = 0; i < rs->md.raid_disks * copies; i++) {
 635                                 if (!(i % copies))
 636                                         rebuilds_per_group = 0;
 637                                 d = i % rs->md.raid_disks;
 638                                 if ((!rs->dev[d].rdev.sb_page ||
 639                                      !test_bit(In_sync, &rs->dev[d].rdev.flags)) &&
 640                                     (++rebuilds_per_group >= copies))
 641                                         goto too_many;
 642                         }
 643                         break;
 644                 }
 645
 646                 /*
 647                  * When checking "far" and "offset" formats, we need to ensure
 648                  * that the device that holds its copy is not also dead or
 649                  * being rebuilt.  (Note that "far" and "offset" formats only
 650                  * support two copies right now.  These formats also only ever
 651                  * use the 'use_far_sets' variant.)
 652                  *
 653                  * This check is somewhat complicated by the need to account
 654                  * for arrays that are not a multiple of (far) copies.  This
 655                  * results in the need to treat the last (potentially larger)
 656                  * set differently.
 657                  */
 658                 group_size = (rs->md.raid_disks / copies);
 659                 last_group_start = (rs->md.raid_disks / group_size) - 1;
 660                 last_group_start *= group_size;
 661                 for (i = 0; i < rs->md.raid_disks; i++) {
 662                         if (!(i % copies) && !(i > last_group_start))
 663                                 rebuilds_per_group = 0;
 664                         if ((!rs->dev[i].rdev.sb_page ||
 665                              !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
 666                             (++rebuilds_per_group >= copies))
 667                                         goto too_many;
 668                 }
 669                 break;
 670         default:
 671                 if (rebuild_cnt)
 672                         return -EINVAL;
 673         }
 674
 675         return 0;
 676
 677 too_many:
 678         return -EINVAL;
 679 }
 680
 681 /*
 682  * Possible arguments are...
 683  *      <chunk_size> [optional_args]
 684  *
 685  * Argument definitions
 686  *    <chunk_size>                      The number of sectors per disk that
 687  *                                      will form the "stripe"
 688  *    [[no]sync]                        Force or prevent recovery of the
 689  *                                      entire array
 690  *    [rebuild <idx>]                   Rebuild the drive indicated by the index
 691  *    [daemon_sleep <ms>]               Time between bitmap daemon work to
 692  *                                      clear bits
 693  *    [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
 694  *    [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
 695  *    [write_mostly <idx>]              Indicate a write mostly drive via index
 696  *    [max_write_behind <sectors>]      See '-write-behind=' (man mdadm)
 697  *    [stripe_cache <sectors>]          Stripe cache size for higher RAIDs
 698  *    [region_size <sectors>]           Defines granularity of bitmap
 699  *
 700  * RAID10-only options:
 701  *    [raid10_copies <# copies>]        Number of copies.  (Default: 2)
 702  *    [raid10_format <near|far|offset>] Layout algorithm.  (Default: near)
 703  */
 704 static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
 705                              unsigned num_raid_params)
 706 {
 707         char *raid10_format = "near";
 708         unsigned raid10_copies = 2;
 709         unsigned i;
 710         unsigned value, region_size = 0;
 711         sector_t sectors_per_dev = rs->ti->len;
 712         sector_t max_io_len;
 713         const char *arg, *key;
 714         struct raid_dev *rd;
 715
 716         arg = dm_shift_arg(as);
 717         num_raid_params--; /* Account for chunk_size argument */
 718
 719         if (kstrtouint(arg, 10, &value) < 0)
 720                 return ti_error_einval(rs->ti, "Bad numerical argument given for chunk_size");
 721
 722         /*
 723          * First, parse the in-order required arguments
 724          * "chunk_size" is the only argument of this type.
 725          */
 726         if (rs->raid_type->level == 1) {
 727                 if (value)
 728                         DMERR("Ignoring chunk size parameter for RAID 1");
 729                 value = 0;
 730         } else if (!is_power_of_2(value))
 731                 return ti_error_einval(rs->ti, "Chunk size must be a power of 2");
 732         else if (value < 8)
 733                 return ti_error_einval(rs->ti, "Chunk size value is too small");
 734
 735         rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
 736
 737         /*
 738          * We set each individual device as In_sync with a completed
 739          * 'recovery_offset'.  If there has been a device failure or
 740          * replacement then one of the following cases applies:
 741          *
 742          *   1) User specifies 'rebuild'.
 743          *      - Device is reset when param is read.
 744          *   2) A new device is supplied.
 745          *      - No matching superblock found, resets device.
 746          *   3) Device failure was transient and returns on reload.
 747          *      - Failure noticed, resets device for bitmap replay.
 748          *   4) Device hadn't completed recovery after previous failure.
 749          *      - Superblock is read and overrides recovery_offset.
 750          *
 751          * What is found in the superblocks of the devices is always
 752          * authoritative, unless 'rebuild' or '[no]sync' was specified.
 753          */
 754         for (i = 0; i < rs->md.raid_disks; i++) {
 755                 set_bit(In_sync, &rs->dev[i].rdev.flags);
 756                 rs->dev[i].rdev.recovery_offset = MaxSector;
 757         }
 758
 759         /*
 760          * Second, parse the unordered optional arguments
 761          */
 762         for (i = 0; i < num_raid_params; i++) {
 763                 arg = dm_shift_arg(as);
 764                 if (!arg)
 765                         return ti_error_einval(rs->ti, "Not enough raid parameters given");
 766
 767                 if (!strcasecmp(arg, "nosync")) {
 768                         rs->md.recovery_cp = MaxSector;
 769                         _set_flag(CTR_FLAG_NOSYNC, &rs->ctr_flags);
 770                         continue;
 771                 }
 772                 if (!strcasecmp(arg, "sync")) {
 773                         rs->md.recovery_cp = 0;
 774                         _set_flag(CTR_FLAG_SYNC, &rs->ctr_flags);
 775                         continue;
 776                 }
 777
 778                 key = arg;
 779                 arg = dm_shift_arg(as);
 780                 i++; /* Account for the argument pairs */
 781                 if (!arg)
 782                         return ti_error_einval(rs->ti, "Wrong number of raid parameters given");
 783
 784                 /*
 785                  * Parameters that take a string value are checked here.
 786                  */
 787
 788                 if (!strcasecmp(key, _argname_by_flag(CTR_FLAG_RAID10_FORMAT))) {
 789                         if (_test_and_set_flag(CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
 790                                 return ti_error_einval(rs->ti, "Only one raid10_format argument pair allowed");
 791                         if (rs->raid_type->level != 10)
 792                                 return ti_error_einval(rs->ti, "'raid10_format' is an invalid parameter for this RAID type");
 793                         if (strcmp("near", arg) &&
 794                             strcmp("far", arg) &&
 795                             strcmp("offset", arg))
 796                                 return ti_error_einval(rs->ti, "Invalid 'raid10_format' value given");
 797
 798                         raid10_format = (char *) arg;
 799                         continue;
 800                 }
 801
 802                 if (kstrtouint(arg, 10, &value) < 0)
 803                         return ti_error_einval(rs->ti, "Bad numerical argument given in raid params");
 804
 805                 if (!strcasecmp(key, _argname_by_flag(CTR_FLAG_REBUILD))) {
 806                         /*
 807                          * "rebuild" is being passed in by userspace to provide
 808                          * indexes of replaced devices and to set up additional
 809                          * devices on raid level takeover.
 810                          */
 811                         if (!_in_range(value, 0, rs->md.raid_disks - 1))
 812                                 return ti_error_einval(rs->ti, "Invalid rebuild index given");
 813
 814                         rd = rs->dev + value;
 815                         clear_bit(In_sync, &rd->rdev.flags);
 816                         clear_bit(Faulty, &rd->rdev.flags);
 817                         rd->rdev.recovery_offset = 0;
 818                         _set_flag(CTR_FLAG_REBUILD, &rs->ctr_flags);
 819                 } else if (!strcasecmp(key, _argname_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
 820                         if (rs->raid_type->level != 1)
 821                                 return ti_error_einval(rs->ti, "write_mostly option is only valid for RAID1");
 822
 823                         if (!_in_range(value, 0, rs->md.raid_disks - 1))
 824                                 return ti_error_einval(rs->ti, "Invalid write_mostly index given");
 825
 826                         set_bit(WriteMostly, &rs->dev[value].rdev.flags);
 827                         _set_flag(CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
 828                 } else if (!strcasecmp(key, _argname_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
 829                         if (rs->raid_type->level != 1)
 830                                 return ti_error_einval(rs->ti, "max_write_behind option is only valid for RAID1");
 831
 832                         if (_test_and_set_flag(CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
 833                                 return ti_error_einval(rs->ti, "Only one max_write_behind argument pair allowed");
 834
 835                         /*
 836                          * In device-mapper, we specify things in sectors, but
 837                          * MD records this value in kB
 838                          */
 839                         value /= 2;
 840                         if (value > COUNTER_MAX)
 841                                 return ti_error_einval(rs->ti, "Max write-behind limit out of range");
 842
 843                         rs->md.bitmap_info.max_write_behind = value;
 844                 } else if (!strcasecmp(key, _argname_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
 845                         if (_test_and_set_flag(CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
 846                                 return ti_error_einval(rs->ti, "Only one daemon_sleep argument pair allowed");
 847                         if (!value || (value > MAX_SCHEDULE_TIMEOUT))
 848                                 return ti_error_einval(rs->ti, "daemon sleep period out of range");
 849                         rs->md.bitmap_info.daemon_sleep = value;
 850                 } else if (!strcasecmp(key, _argname_by_flag(CTR_FLAG_STRIPE_CACHE))) {
 851                         if (_test_and_set_flag(CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
 852                                 return ti_error_einval(rs->ti, "Only one stripe_cache argument pair allowed");
 853                         /*
 854                          * In device-mapper, we specify things in sectors, but
 855                          * MD records this value in kB
 856                          */
 857                         value /= 2;
 858
 859                         if (!_in_range(rs->raid_type->level, 4, 6))
 860                                 return ti_error_einval(rs->ti, "Inappropriate argument: stripe_cache");
 861                         if (raid5_set_cache_size(&rs->md, (int)value))
 862                                 return ti_error_einval(rs->ti, "Bad stripe_cache size");
 863
 864                 } else if (!strcasecmp(key, _argname_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
 865                         if (_test_and_set_flag(CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
 866                                 return ti_error_einval(rs->ti, "Only one min_recovery_rate argument pair allowed");
 867                         if (value > INT_MAX)
 868                                 return ti_error_einval(rs->ti, "min_recovery_rate out of range");
 869                         rs->md.sync_speed_min = (int)value;
 870                 } else if (!strcasecmp(key, _argname_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
 871                         if (_test_and_set_flag(CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
 872                                 return ti_error_einval(rs->ti, "Only one max_recovery_rate argument pair allowed");
 873                         if (value > INT_MAX)
 874                                 return ti_error_einval(rs->ti, "max_recovery_rate out of range");
 875                         rs->md.sync_speed_max = (int)value;
 876                 } else if (!strcasecmp(key, _argname_by_flag(CTR_FLAG_REGION_SIZE))) {
 877                         if (_test_and_set_flag(CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
 878                                 return ti_error_einval(rs->ti, "Only one region_size argument pair allowed");
 879
 880                         region_size = value;
 881                 } else if (!strcasecmp(key, _argname_by_flag(CTR_FLAG_RAID10_COPIES))) {
 882                         if (_test_and_set_flag(CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
 883                                 return ti_error_einval(rs->ti, "Only one raid10_copies argument pair allowed");
 884
 885                         if (!_in_range(value, 2, rs->md.raid_disks))
 886                                 return ti_error_einval(rs->ti, "Bad value for 'raid10_copies'");
 887
 888                         raid10_copies = value;
 889                 } else {
 890                         DMERR("Unable to parse RAID parameter: %s", key);
 891                         return ti_error_einval(rs->ti, "Unable to parse RAID parameters");
 892                 }
 893         }
 894
 895         if (validate_region_size(rs, region_size))
 896                 return -EINVAL;
 897
 898         if (rs->md.chunk_sectors)
 899                 max_io_len = rs->md.chunk_sectors;
 900         else
 901                 max_io_len = region_size;
 902
 903         if (dm_set_target_max_io_len(rs->ti, max_io_len))
 904                 return -EINVAL;
 905
 906         if (rs->raid_type->level == 10) {
 907                 if (raid10_copies > rs->md.raid_disks)
 908                         return ti_error_einval(rs->ti, "Not enough devices to satisfy specification");
 909
 910                 /*
 911                  * If the format is not "near", we only support
 912                  * two copies at the moment.
 913                  */
 914                 if (strcmp("near", raid10_format) && (raid10_copies > 2))
 915                         return ti_error_einval(rs->ti, "Too many copies for given RAID10 format.");
 916
 917                 /* (Len * #mirrors) / #devices */
 918                 sectors_per_dev = rs->ti->len * raid10_copies;
 919                 sector_div(sectors_per_dev, rs->md.raid_disks);
 920
 921                 rs->md.layout = raid10_format_to_md_layout(raid10_format,
 922                                                            raid10_copies);
 923                 rs->md.new_layout = rs->md.layout;
 924         } else if ((!rs->raid_type->level || rs->raid_type->level > 1) &&
 925                    sector_div(sectors_per_dev,
 926                               (rs->md.raid_disks - rs->raid_type->parity_devs)))
 927                 return ti_error_einval(rs->ti, "Target length not divisible by number of data devices");
 928
 929         rs->md.dev_sectors = sectors_per_dev;
 930
 931         /* Assume there are no metadata devices until the drives are parsed */
 932         rs->md.persistent = 0;
 933         rs->md.external = 1;
 934
 935         /* Check, if any invalid ctr arguments have been passed in for the raid level */
 936         return rs_check_for_invalid_flags(rs);
 937 }
 938
 939 static void do_table_event(struct work_struct *ws)
 940 {
 941         struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
 942
 943         dm_table_event(rs->ti->table);
 944 }
 945
 946 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
 947 {
 948         struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
 949
 950         return mddev_congested(&rs->md, bits);
 951 }
 952
 953 /*
 954  * This structure is never routinely used by userspace, unlike md superblocks.
 955  * Devices with this superblock should only ever be accessed via device-mapper.
 956  */
 957 #define DM_RAID_MAGIC 0x64526D44
 958 struct dm_raid_superblock {
 959         __le32 magic;           /* "DmRd" */
 960         __le32 features;        /* Used to indicate possible future changes */
 961
 962         __le32 num_devices;     /* Number of devices in this array. (Max 64) */
 963         __le32 array_position;  /* The position of this drive in the array */
 964
 965         __le64 events;          /* Incremented by md when superblock updated */
 966         __le64 failed_devices;  /* Bit field of devices to indicate failures */
 967
 968         /*
 969          * This offset tracks the progress of the repair or replacement of
 970          * an individual drive.
 971          */
 972         __le64 disk_recovery_offset;
 973
 974         /*
 975          * This offset tracks the progress of the initial array
 976          * synchronisation/parity calculation.
 977          */
 978         __le64 array_resync_offset;
 979
 980         /*
 981          * RAID characteristics
 982          */
 983         __le32 level;
 984         __le32 layout;
 985         __le32 stripe_sectors;
 986
 987         /* Remainder of a logical block is zero-filled when writing (see super_sync()). */
 988 } __packed;
 989
 990 static int read_disk_sb(struct md_rdev *rdev, int size)
 991 {
 992         BUG_ON(!rdev->sb_page);
 993
 994         if (rdev->sb_loaded)
 995                 return 0;
 996
 997         if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, 1)) {
 998                 DMERR("Failed to read superblock of device at position %d",
 999                       rdev->raid_disk);
1000                 md_error(rdev->mddev, rdev);
1001                 return -EINVAL;
1002         }
1003
1004         rdev->sb_loaded = 1;
1005
1006         return 0;
1007 }
1008
1009 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
1010 {
1011         int i;
1012         uint64_t failed_devices;
1013         struct dm_raid_superblock *sb;
1014         struct raid_set *rs = container_of(mddev, struct raid_set, md);
1015
1016         sb = page_address(rdev->sb_page);
1017         failed_devices = le64_to_cpu(sb->failed_devices);
1018
1019         for (i = 0; i < mddev->raid_disks; i++)
1020                 if (!rs->dev[i].data_dev ||
1021                     test_bit(Faulty, &(rs->dev[i].rdev.flags)))
1022                         failed_devices |= (1ULL << i);
1023
1024         memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
1025
1026         sb->magic = cpu_to_le32(DM_RAID_MAGIC);
1027         sb->features = cpu_to_le32(0);  /* No features yet */
1028
1029         sb->num_devices = cpu_to_le32(mddev->raid_disks);
1030         sb->array_position = cpu_to_le32(rdev->raid_disk);
1031
1032         sb->events = cpu_to_le64(mddev->events);
1033         sb->failed_devices = cpu_to_le64(failed_devices);
1034
1035         sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
1036         sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
1037
1038         sb->level = cpu_to_le32(mddev->level);
1039         sb->layout = cpu_to_le32(mddev->layout);
1040         sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
1041 }
1042
1043 /*
1044  * super_load
1045  *
1046  * This function creates a superblock if one is not found on the device
1047  * and will decide which superblock to use if there's a choice.
1048  *
1049  * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
1050  */
1051 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
1052 {
1053         int r;
1054         struct dm_raid_superblock *sb;
1055         struct dm_raid_superblock *refsb;
1056         uint64_t events_sb, events_refsb;
1057
1058         rdev->sb_start = 0;
1059         rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
1060         if (rdev->sb_size < sizeof(*sb) || rdev->sb_size > PAGE_SIZE) {
1061                 DMERR("superblock size of a logical block is no longer valid");
1062                 return -EINVAL;
1063         }
1064
1065         r = read_disk_sb(rdev, rdev->sb_size);
1066         if (r)
1067                 return r;
1068
1069         sb = page_address(rdev->sb_page);
1070
1071         /*
1072          * Two cases that we want to write new superblocks and rebuild:
1073          * 1) New device (no matching magic number)
1074          * 2) Device specified for rebuild (!In_sync w/ offset == 0)
1075          */
1076         if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
1077             (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
1078                 super_sync(rdev->mddev, rdev);
1079
1080                 set_bit(FirstUse, &rdev->flags);
1081
1082                 /* Force writing of superblocks to disk */
1083                 set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags);
1084
1085                 /* Any superblock is better than none, choose that if given */
1086                 return refdev ? 0 : 1;
1087         }
1088
1089         if (!refdev)
1090                 return 1;
1091
1092         events_sb = le64_to_cpu(sb->events);
1093
1094         refsb = page_address(refdev->sb_page);
1095         events_refsb = le64_to_cpu(refsb->events);
1096
1097         return (events_sb > events_refsb) ? 1 : 0;
1098 }
1099
1100 static int super_init_validation(struct mddev *mddev, struct md_rdev *rdev)
1101 {
1102         int role;
1103         struct raid_set *rs = container_of(mddev, struct raid_set, md);
1104         uint64_t events_sb;
1105         uint64_t failed_devices;
1106         struct dm_raid_superblock *sb;
1107         uint32_t new_devs = 0;
1108         uint32_t rebuilds = 0;
1109         struct md_rdev *r;
1110         struct dm_raid_superblock *sb2;
1111
1112         sb = page_address(rdev->sb_page);
1113         events_sb = le64_to_cpu(sb->events);
1114         failed_devices = le64_to_cpu(sb->failed_devices);
1115
1116         /*
1117          * Initialise to 1 if this is a new superblock.
1118          */
1119         mddev->events = events_sb ? : 1;
1120
1121         /*
1122          * Reshaping is not currently allowed
1123          */
1124         if (le32_to_cpu(sb->level) != mddev->level) {
1125                 DMERR("Reshaping arrays not yet supported. (RAID level change)");
1126                 return -EINVAL;
1127         }
1128         if (le32_to_cpu(sb->layout) != mddev->layout) {
1129                 DMERR("Reshaping arrays not yet supported. (RAID layout change)");
1130                 DMERR("  0x%X vs 0x%X", le32_to_cpu(sb->layout), mddev->layout);
1131                 DMERR("  Old layout: %s w/ %d copies",
1132                       raid10_md_layout_to_format(le32_to_cpu(sb->layout)),
1133                       raid10_md_layout_to_copies(le32_to_cpu(sb->layout)));
1134                 DMERR("  New layout: %s w/ %d copies",
1135                       raid10_md_layout_to_format(mddev->layout),
1136                       raid10_md_layout_to_copies(mddev->layout));
1137                 return -EINVAL;
1138         }
1139         if (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors) {
1140                 DMERR("Reshaping arrays not yet supported. (stripe sectors change)");
1141                 return -EINVAL;
1142         }
1143
1144         /* We can only change the number of devices in RAID1 right now */
1145         if ((rs->raid_type->level != 1) &&
1146             (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) {
1147                 DMERR("Reshaping arrays not yet supported. (device count change)");
1148                 return -EINVAL;
1149         }
1150
1151         if (!(_test_flags(CTR_FLAGS_ANY_SYNC, rs->ctr_flags)))
1152                 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
1153
1154         /*
1155          * During load, we set FirstUse if a new superblock was written.
1156          * There are two reasons we might not have a superblock:
1157          * 1) The array is brand new - in which case, all of the
1158          *    devices must have their In_sync bit set.  Also,
1159          *    recovery_cp must be 0, unless forced.
1160          * 2) This is a new device being added to an old array
1161          *    and the new device needs to be rebuilt - in which
1162          *    case the In_sync bit will /not/ be set and
1163          *    recovery_cp must be MaxSector.
1164          */
1165         rdev_for_each(r, mddev) {
1166                 if (!test_bit(In_sync, &r->flags)) {
1167                         DMINFO("Device %d specified for rebuild: "
1168                                "Clearing superblock", r->raid_disk);
1169                         rebuilds++;
1170                 } else if (test_bit(FirstUse, &r->flags))
1171                         new_devs++;
1172         }
1173
1174         if (!rebuilds) {
1175                 if (new_devs == mddev->raid_disks) {
1176                         DMINFO("Superblocks created for new array");
1177                         set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
1178                 } else if (new_devs) {
1179                         DMERR("New device injected "
1180                               "into existing array without 'rebuild' "
1181                               "parameter specified");
1182                         return -EINVAL;
1183                 }
1184         } else if (new_devs) {
1185                 DMERR("'rebuild' devices cannot be "
1186                       "injected into an array with other first-time devices");
1187                 return -EINVAL;
1188         } else if (mddev->recovery_cp != MaxSector) {
1189                 DMERR("'rebuild' specified while array is not in-sync");
1190                 return -EINVAL;
1191         }
1192
1193         /*
1194          * Now we set the Faulty bit for those devices that are
1195          * recorded in the superblock as failed.
1196          */
1197         rdev_for_each(r, mddev) {
1198                 if (!r->sb_page)
1199                         continue;
1200                 sb2 = page_address(r->sb_page);
1201                 sb2->failed_devices = 0;
1202
1203                 /*
1204                  * Check for any device re-ordering.
1205                  */
1206                 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
1207                         role = le32_to_cpu(sb2->array_position);
1208                         if (role != r->raid_disk) {
1209                                 if (rs->raid_type->level != 1)
1210                                         return ti_error_einval(rs->ti, "Cannot change device "
1211                                                                        "positions in RAID array");
1212                                 DMINFO("RAID1 device #%d now at position #%d",
1213                                        role, r->raid_disk);
1214                         }
1215
1216                         /*
1217                          * Partial recovery is performed on
1218                          * returning failed devices.
1219                          */
1220                         if (failed_devices & (1 << role))
1221                                 set_bit(Faulty, &r->flags);
1222                 }
1223         }
1224
1225         return 0;
1226 }
1227
1228 static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
1229 {
1230         struct mddev *mddev = &rs->md;
1231         struct dm_raid_superblock *sb = page_address(rdev->sb_page);
1232
1233         /*
1234          * If mddev->events is not set, we know we have not yet initialized
1235          * the array.
1236          */
1237         if (!mddev->events && super_init_validation(mddev, rdev))
1238                 return -EINVAL;
1239
1240         if (le32_to_cpu(sb->features)) {
1241                 rs->ti->error = "Unable to assemble array: No feature flags supported yet";
1242                 return -EINVAL;
1243         }
1244
1245         /* Enable bitmap creation for RAID levels != 0 */
1246         mddev->bitmap_info.offset = (rs->raid_type->level) ? to_sector(4096) : 0;
1247         rdev->mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
1248
1249         if (!test_bit(FirstUse, &rdev->flags)) {
1250                 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
1251                 if (rdev->recovery_offset != MaxSector)
1252                         clear_bit(In_sync, &rdev->flags);
1253         }
1254
1255         /*
1256          * If a device comes back, set it as not In_sync and no longer faulty.
1257          */
1258         if (test_bit(Faulty, &rdev->flags)) {
1259                 clear_bit(Faulty, &rdev->flags);
1260                 clear_bit(In_sync, &rdev->flags);
1261                 rdev->saved_raid_disk = rdev->raid_disk;
1262                 rdev->recovery_offset = 0;
1263         }
1264
1265         clear_bit(FirstUse, &rdev->flags);
1266
1267         return 0;
1268 }
1269
1270 /*
1271  * Analyse superblocks and select the freshest.
1272  */
1273 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
1274 {
1275         int r;
1276         struct raid_dev *dev;
1277         struct md_rdev *rdev, *tmp, *freshest;
1278         struct mddev *mddev = &rs->md;
1279
1280         freshest = NULL;
1281         rdev_for_each_safe(rdev, tmp, mddev) {
1282                 /*
1283                  * Skipping super_load due to CTR_FLAG_SYNC will cause
1284                  * the array to undergo initialization again as
1285                  * though it were new.  This is the intended effect
1286                  * of the "sync" directive.
1287                  *
1288                  * When reshaping capability is added, we must ensure
1289                  * that the "sync" directive is disallowed during the
1290                  * reshape.
1291                  */
1292                 rdev->sectors = to_sector(i_size_read(rdev->bdev->bd_inode));
1293
1294                 if (_test_flag(CTR_FLAG_SYNC, rs->ctr_flags))
1295                         continue;
1296
1297                 if (!rdev->meta_bdev)
1298                         continue;
1299
1300                 r = super_load(rdev, freshest);
1301
1302                 switch (r) {
1303                 case 1:
1304                         freshest = rdev;
1305                         break;
1306                 case 0:
1307                         break;
1308                 default:
1309                         dev = container_of(rdev, struct raid_dev, rdev);
1310                         if (dev->meta_dev)
1311                                 dm_put_device(ti, dev->meta_dev);
1312
1313                         dev->meta_dev = NULL;
1314                         rdev->meta_bdev = NULL;
1315
1316                         if (rdev->sb_page)
1317                                 put_page(rdev->sb_page);
1318
1319                         rdev->sb_page = NULL;
1320
1321                         rdev->sb_loaded = 0;
1322
1323                         /*
1324                          * We might be able to salvage the data device
1325                          * even though the meta device has failed.  For
1326                          * now, we behave as though '- -' had been
1327                          * set for this device in the table.
1328                          */
1329                         if (dev->data_dev)
1330                                 dm_put_device(ti, dev->data_dev);
1331
1332                         dev->data_dev = NULL;
1333                         rdev->bdev = NULL;
1334
1335                         list_del(&rdev->same_set);
1336                 }
1337         }
1338
1339         if (!freshest)
1340                 return 0;
1341
1342         if (validate_raid_redundancy(rs))
1343                 return ti_error_einval(rs->ti, "Insufficient redundancy to activate array");
1344
1345         /*
1346          * Validation of the freshest device provides the source of
1347          * validation for the remaining devices.
1348          */
1349         if (super_validate(rs, freshest))
1350                 return ti_error_einval(rs->ti, "Unable to assemble array: Invalid superblocks");
1351
1352         rdev_for_each(rdev, mddev)
1353                 if ((rdev != freshest) && super_validate(rs, rdev))
1354                         return -EINVAL;
1355
1356         return 0;
1357 }
1358
1359 /*
1360  * Enable/disable discard support on RAID set depending on
1361  * RAID level and discard properties of underlying RAID members.
1362  */
1363 static void configure_discard_support(struct dm_target *ti, struct raid_set *rs)
1364 {
1365         int i;
1366         bool raid456;
1367
1368         /* Assume discards not supported until after checks below. */
1369         ti->discards_supported = false;
1370
1371         /* RAID level 4,5,6 require discard_zeroes_data for data integrity! */
1372         raid456 = (rs->md.level == 4 || rs->md.level == 5 || rs->md.level == 6);
1373
1374         for (i = 0; i < rs->md.raid_disks; i++) {
1375                 struct request_queue *q;
1376
1377                 if (!rs->dev[i].rdev.bdev)
1378                         continue;
1379
1380                 q = bdev_get_queue(rs->dev[i].rdev.bdev);
1381                 if (!q || !blk_queue_discard(q))
1382                         return;
1383
1384                 if (raid456) {
1385                         if (!q->limits.discard_zeroes_data)
1386                                 return;
1387                         if (!devices_handle_discard_safely) {
1388                                 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
1389                                 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
1390                                 return;
1391                         }
1392                 }
1393         }
1394
1395         /* All RAID members properly support discards */
1396         ti->discards_supported = true;
1397
1398         /*
1399          * RAID1 and RAID10 personalities require bio splitting,
1400          * RAID0/4/5/6 don't and process large discard bios properly.
1401          */
1402         ti->split_discard_bios = !!(rs->md.level == 1 || rs->md.level == 10);
1403         ti->num_discard_bios = 1;
1404 }
1405
1406 /*
1407  * Construct a RAID0/1/10/4/5/6 mapping:
1408  * Args:
1409  *      <raid_type> <#raid_params> <raid_params>{0,}    \
1410  *      <#raid_devs> [<meta_dev1> <dev1>]{1,}
1411  *
1412  * <raid_params> varies by <raid_type>.  See 'parse_raid_params' for
1413  * details on possible <raid_params>.
1414  *
1415  * Userspace is free to initialize the metadata devices, hence the superblocks to
1416  * enforce recreation based on the passed in table parameters.
1417  *
1418  */
1419 static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv)
1420 {
1421         int r;
1422         struct raid_type *rt;
1423         unsigned num_raid_params, num_raid_devs;
1424         struct raid_set *rs = NULL;
1425         const char *arg;
1426         struct dm_arg_set as = { argc, argv }, as_nrd;
1427         struct dm_arg _args[] = {
1428                 { 0, as.argc, "Cannot understand number of raid parameters" },
1429                 { 1, 254, "Cannot understand number of raid devices parameters" }
1430         };
1431
1432         /* Must have <raid_type> */
1433         arg = dm_shift_arg(&as);
1434         if (!arg)
1435                 return ti_error_einval(rs->ti, "No arguments");
1436
1437         rt = get_raid_type(arg);
1438         if (!rt)
1439                 return ti_error_einval(rs->ti, "Unrecognised raid_type");
1440
1441         /* Must have <#raid_params> */
1442         if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
1443                 return -EINVAL;
1444
1445         /* number of raid device tupples <meta_dev data_dev> */
1446         as_nrd = as;
1447         dm_consume_args(&as_nrd, num_raid_params);
1448         _args[1].max = (as_nrd.argc - 1) / 2;
1449         if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
1450                 return -EINVAL;
1451
1452         if (!_in_range(num_raid_devs, 1, MAX_RAID_DEVICES))
1453                 return ti_error_einval(rs->ti, "Invalid number of supplied raid devices");
1454
1455         rs = context_alloc(ti, rt, num_raid_devs);
1456         if (IS_ERR(rs))
1457                 return PTR_ERR(rs);
1458
1459         r = parse_raid_params(rs, &as, num_raid_params);
1460         if (r)
1461                 goto bad;
1462
1463         r = parse_dev_params(rs, &as);
1464         if (r)
1465                 goto bad;
1466
1467         rs->md.sync_super = super_sync;
1468         r = analyse_superblocks(ti, rs);
1469         if (r)
1470                 goto bad;
1471
1472         INIT_WORK(&rs->md.event_work, do_table_event);
1473         ti->private = rs;
1474         ti->num_flush_bios = 1;
1475
1476         /*
1477          * Disable/enable discard support on RAID set.
1478          */
1479         configure_discard_support(ti, rs);
1480
1481         /* Has to be held on running the array */
1482         mddev_lock_nointr(&rs->md);
1483         r = md_run(&rs->md);
1484         rs->md.in_sync = 0; /* Assume already marked dirty */
1485         mddev_unlock(&rs->md);
1486
1487         if (r) {
1488                 ti->error = "Fail to run raid array";
1489                 goto bad;
1490         }
1491
1492         if (ti->len != rs->md.array_sectors) {
1493                 r = ti_error_einval(ti, "Array size does not match requested target length");
1494                 goto size_mismatch;
1495         }
1496         rs->callbacks.congested_fn = raid_is_congested;
1497         dm_table_add_target_callbacks(ti->table, &rs->callbacks);
1498
1499         mddev_suspend(&rs->md);
1500         return 0;
1501
1502 size_mismatch:
1503         md_stop(&rs->md);
1504 bad:
1505         context_free(rs);
1506
1507         return r;
1508 }
1509
1510 static void raid_dtr(struct dm_target *ti)
1511 {
1512         struct raid_set *rs = ti->private;
1513
1514         list_del_init(&rs->callbacks.list);
1515         md_stop(&rs->md);
1516         context_free(rs);
1517 }
1518
1519 static int raid_map(struct dm_target *ti, struct bio *bio)
1520 {
1521         struct raid_set *rs = ti->private;
1522         struct mddev *mddev = &rs->md;
1523
1524         mddev->pers->make_request(mddev, bio);
1525
1526         return DM_MAPIO_SUBMITTED;
1527 }
1528
1529 static const char *decipher_sync_action(struct mddev *mddev)
1530 {
1531         if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
1532                 return "frozen";
1533
1534         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1535             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
1536                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
1537                         return "reshape";
1538
1539                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1540                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1541                                 return "resync";
1542                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1543                                 return "check";
1544                         return "repair";
1545                 }
1546
1547                 if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
1548                         return "recover";
1549         }
1550
1551         return "idle";
1552 }
1553
1554 static void raid_status(struct dm_target *ti, status_type_t type,
1555                         unsigned status_flags, char *result, unsigned maxlen)
1556 {
1557         struct raid_set *rs = ti->private;
1558         unsigned raid_param_cnt = 1; /* at least 1 for chunksize */
1559         unsigned sz = 0;
1560         int i, array_in_sync = 0;
1561         sector_t sync;
1562
1563         switch (type) {
1564         case STATUSTYPE_INFO:
1565                 DMEMIT("%s %d ", rs->raid_type->name, rs->md.raid_disks);
1566
1567                 if (rs->raid_type->level) {
1568                         if (test_bit(MD_RECOVERY_RUNNING, &rs->md.recovery))
1569                                 sync = rs->md.curr_resync_completed;
1570                         else
1571                                 sync = rs->md.recovery_cp;
1572
1573                         if (sync >= rs->md.resync_max_sectors) {
1574                                 /*
1575                                  * Sync complete.
1576                                  */
1577                                 array_in_sync = 1;
1578                                 sync = rs->md.resync_max_sectors;
1579                         } else if (test_bit(MD_RECOVERY_REQUESTED, &rs->md.recovery)) {
1580                                 /*
1581                                  * If "check" or "repair" is occurring, the array has
1582                                  * undergone and initial sync and the health characters
1583                                  * should not be 'a' anymore.
1584                                  */
1585                                 array_in_sync = 1;
1586                         } else {
1587                                 /*
1588                                  * The array may be doing an initial sync, or it may
1589                                  * be rebuilding individual components.  If all the
1590                                  * devices are In_sync, then it is the array that is
1591                                  * being initialized.
1592                                  */
1593                                 for (i = 0; i < rs->md.raid_disks; i++)
1594                                         if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
1595                                                 array_in_sync = 1;
1596                         }
1597                 } else {
1598                         /* RAID0 */
1599                         array_in_sync = 1;
1600                         sync = rs->md.resync_max_sectors;
1601                 }
1602
1603                 /*
1604                  * Status characters:
1605                  *  'D' = Dead/Failed device
1606                  *  'a' = Alive but not in-sync
1607                  *  'A' = Alive and in-sync
1608                  */
1609                 for (i = 0; i < rs->md.raid_disks; i++) {
1610                         if (test_bit(Faulty, &rs->dev[i].rdev.flags))
1611                                 DMEMIT("D");
1612                         else if (!array_in_sync ||
1613                                  !test_bit(In_sync, &rs->dev[i].rdev.flags))
1614                                 DMEMIT("a");
1615                         else
1616                                 DMEMIT("A");
1617                 }
1618
1619                 /*
1620                  * In-sync ratio:
1621                  *  The in-sync ratio shows the progress of:
1622                  *   - Initializing the array
1623                  *   - Rebuilding a subset of devices of the array
1624                  *  The user can distinguish between the two by referring
1625                  *  to the status characters.
1626                  */
1627                 DMEMIT(" %llu/%llu",
1628                        (unsigned long long) sync,
1629                        (unsigned long long) rs->md.resync_max_sectors);
1630
1631                 /*
1632                  * Sync action:
1633                  *   See Documentation/device-mapper/dm-raid.c for
1634                  *   information on each of these states.
1635                  */
1636                 DMEMIT(" %s", decipher_sync_action(&rs->md));
1637
1638                 /*
1639                  * resync_mismatches/mismatch_cnt
1640                  *   This field shows the number of discrepancies found when
1641                  *   performing a "check" of the array.
1642                  */
1643                 DMEMIT(" %llu",
1644                        (strcmp(rs->md.last_sync_action, "check")) ? 0 :
1645                        (unsigned long long)
1646                        atomic64_read(&rs->md.resync_mismatches));
1647                 break;
1648         case STATUSTYPE_TABLE:
1649                 /* The string you would use to construct this array */
1650                 for (i = 0; i < rs->md.raid_disks; i++) {
1651                         if (_test_flag(CTR_FLAG_REBUILD, rs->ctr_flags) &&
1652                             rs->dev[i].data_dev &&
1653                             !test_bit(In_sync, &rs->dev[i].rdev.flags))
1654                                 raid_param_cnt += 2; /* for rebuilds */
1655                         if (rs->dev[i].data_dev &&
1656                             test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1657                                 raid_param_cnt += 2;
1658                 }
1659
1660                 raid_param_cnt += (hweight32(rs->ctr_flags & ~CTR_FLAG_REBUILD) * 2);
1661                 if (rs->ctr_flags & (CTR_FLAG_SYNC | CTR_FLAG_NOSYNC))
1662                         raid_param_cnt--;
1663
1664                 DMEMIT("%s %u %u", rs->raid_type->name,
1665                        raid_param_cnt, rs->md.chunk_sectors);
1666
1667                 if (_test_flag(CTR_FLAG_SYNC, rs->ctr_flags) &&
1668                     rs->md.recovery_cp == MaxSector)
1669                         DMEMIT(" sync");
1670                 if (_test_flag(CTR_FLAG_NOSYNC, rs->ctr_flags))
1671                         DMEMIT(" nosync");
1672
1673                 for (i = 0; i < rs->md.raid_disks; i++)
1674                         if (_test_flag(CTR_FLAG_REBUILD, rs->ctr_flags) &&
1675                             rs->dev[i].data_dev &&
1676                             !test_bit(In_sync, &rs->dev[i].rdev.flags))
1677                                 DMEMIT(" rebuild %u", i);
1678
1679                 if (_test_flag(CTR_FLAG_DAEMON_SLEEP, rs->ctr_flags))
1680                         DMEMIT(" daemon_sleep %lu",
1681                                rs->md.bitmap_info.daemon_sleep);
1682
1683                 if (_test_flag(CTR_FLAG_MIN_RECOVERY_RATE, rs->ctr_flags))
1684                         DMEMIT(" min_recovery_rate %d", rs->md.sync_speed_min);
1685
1686                 if (_test_flag(CTR_FLAG_MAX_RECOVERY_RATE, rs->ctr_flags))
1687                         DMEMIT(" max_recovery_rate %d", rs->md.sync_speed_max);
1688
1689                 for (i = 0; i < rs->md.raid_disks; i++)
1690                         if (rs->dev[i].data_dev &&
1691                             test_bit(WriteMostly, &rs->dev[i].rdev.flags))
1692                                 DMEMIT(" write_mostly %u", i);
1693
1694                 if (_test_flag(CTR_FLAG_MAX_WRITE_BEHIND, rs->ctr_flags))
1695                         DMEMIT(" max_write_behind %lu",
1696                                rs->md.bitmap_info.max_write_behind);
1697
1698                 if (_test_flag(CTR_FLAG_STRIPE_CACHE, rs->ctr_flags)) {
1699                         struct r5conf *conf = rs->md.private;
1700
1701                         /* convert from kiB to sectors */
1702                         DMEMIT(" stripe_cache %d",
1703                                conf ? conf->max_nr_stripes * 2 : 0);
1704                 }
1705
1706                 if (_test_flag(CTR_FLAG_REGION_SIZE, rs->ctr_flags))
1707                         DMEMIT(" region_size %lu",
1708                                rs->md.bitmap_info.chunksize >> 9);
1709
1710                 if (_test_flag(CTR_FLAG_RAID10_COPIES, rs->ctr_flags))
1711                         DMEMIT(" raid10_copies %u",
1712                                raid10_md_layout_to_copies(rs->md.layout));
1713
1714                 if (_test_flag(CTR_FLAG_RAID10_FORMAT, rs->ctr_flags))
1715                         DMEMIT(" raid10_format %s",
1716                                raid10_md_layout_to_format(rs->md.layout));
1717
1718                 DMEMIT(" %d", rs->md.raid_disks);
1719                 for (i = 0; i < rs->md.raid_disks; i++) {
1720                         if (rs->dev[i].meta_dev)
1721                                 DMEMIT(" %s", rs->dev[i].meta_dev->name);
1722                         else
1723                                 DMEMIT(" -");
1724
1725                         if (rs->dev[i].data_dev)
1726                                 DMEMIT(" %s", rs->dev[i].data_dev->name);
1727                         else
1728                                 DMEMIT(" -");
1729                 }
1730         }
1731 }
1732
1733 static int raid_message(struct dm_target *ti, unsigned argc, char **argv)
1734 {
1735         struct raid_set *rs = ti->private;
1736         struct mddev *mddev = &rs->md;
1737
1738         if (!strcasecmp(argv[0], "reshape")) {
1739                 DMERR("Reshape not supported.");
1740                 return -EINVAL;
1741         }
1742
1743         if (!mddev->pers || !mddev->pers->sync_request)
1744                 return -EINVAL;
1745
1746         if (!strcasecmp(argv[0], "frozen"))
1747                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1748         else
1749                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1750
1751         if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
1752                 if (mddev->sync_thread) {
1753                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1754                         md_reap_sync_thread(mddev);
1755                 }
1756         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1757                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1758                 return -EBUSY;
1759         else if (!strcasecmp(argv[0], "resync"))
1760                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1761         else if (!strcasecmp(argv[0], "recover")) {
1762                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
1763                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1764         } else {
1765                 if (!strcasecmp(argv[0], "check"))
1766                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1767                 else if (!!strcasecmp(argv[0], "repair"))
1768                         return -EINVAL;
1769                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1770                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1771         }
1772         if (mddev->ro == 2) {
1773                 /* A write to sync_action is enough to justify
1774                  * canceling read-auto mode
1775                  */
1776                 mddev->ro = 0;
1777                 if (!mddev->suspended)
1778                         md_wakeup_thread(mddev->sync_thread);
1779         }
1780         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1781         if (!mddev->suspended)
1782                 md_wakeup_thread(mddev->thread);
1783
1784         return 0;
1785 }
1786
1787 static int raid_iterate_devices(struct dm_target *ti,
1788                                 iterate_devices_callout_fn fn, void *data)
1789 {
1790         struct raid_set *rs = ti->private;
1791         unsigned i;
1792         int r = 0;
1793
1794         for (i = 0; !r && i < rs->md.raid_disks; i++)
1795                 if (rs->dev[i].data_dev)
1796                         r = fn(ti,
1797                                  rs->dev[i].data_dev,
1798                                  0, /* No offset on data devs */
1799                                  rs->md.dev_sectors,
1800                                  data);
1801
1802         return r;
1803 }
1804
1805 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
1806 {
1807         struct raid_set *rs = ti->private;
1808         unsigned chunk_size = rs->md.chunk_sectors << 9;
1809         struct r5conf *conf = rs->md.private;
1810
1811         blk_limits_io_min(limits, chunk_size);
1812         blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded));
1813 }
1814
1815 static void raid_presuspend(struct dm_target *ti)
1816 {
1817         struct raid_set *rs = ti->private;
1818
1819         md_stop_writes(&rs->md);
1820 }
1821
1822 static void raid_postsuspend(struct dm_target *ti)
1823 {
1824         struct raid_set *rs = ti->private;
1825
1826         mddev_suspend(&rs->md);
1827 }
1828
1829 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
1830 {
1831         int i;
1832         uint64_t failed_devices, cleared_failed_devices = 0;
1833         unsigned long flags;
1834         struct dm_raid_superblock *sb;
1835         struct md_rdev *r;
1836
1837         for (i = 0; i < rs->md.raid_disks; i++) {
1838                 r = &rs->dev[i].rdev;
1839                 if (test_bit(Faulty, &r->flags) && r->sb_page &&
1840                     sync_page_io(r, 0, r->sb_size, r->sb_page, REQ_OP_READ, 0,
1841                                  1)) {
1842                         DMINFO("Faulty %s device #%d has readable super block."
1843                                "  Attempting to revive it.",
1844                                rs->raid_type->name, i);
1845
1846                         /*
1847                          * Faulty bit may be set, but sometimes the array can
1848                          * be suspended before the personalities can respond
1849                          * by removing the device from the array (i.e. calling
1850                          * 'hot_remove_disk').  If they haven't yet removed
1851                          * the failed device, its 'raid_disk' number will be
1852                          * '>= 0' - meaning we must call this function
1853                          * ourselves.
1854                          */
1855                         if ((r->raid_disk >= 0) &&
1856                             (r->mddev->pers->hot_remove_disk(r->mddev, r) != 0))
1857                                 /* Failed to revive this device, try next */
1858                                 continue;
1859
1860                         r->raid_disk = i;
1861                         r->saved_raid_disk = i;
1862                         flags = r->flags;
1863                         clear_bit(Faulty, &r->flags);
1864                         clear_bit(WriteErrorSeen, &r->flags);
1865                         clear_bit(In_sync, &r->flags);
1866                         if (r->mddev->pers->hot_add_disk(r->mddev, r)) {
1867                                 r->raid_disk = -1;
1868                                 r->saved_raid_disk = -1;
1869                                 r->flags = flags;
1870                         } else {
1871                                 r->recovery_offset = 0;
1872                                 cleared_failed_devices |= 1 << i;
1873                         }
1874                 }
1875         }
1876         if (cleared_failed_devices) {
1877                 rdev_for_each(r, &rs->md) {
1878                         sb = page_address(r->sb_page);
1879                         failed_devices = le64_to_cpu(sb->failed_devices);
1880                         failed_devices &= ~cleared_failed_devices;
1881                         sb->failed_devices = cpu_to_le64(failed_devices);
1882                 }
1883         }
1884 }
1885
1886 static void raid_resume(struct dm_target *ti)
1887 {
1888         struct raid_set *rs = ti->private;
1889
1890         if (rs->raid_type->level) {
1891                 set_bit(MD_CHANGE_DEVS, &rs->md.flags);
1892
1893                 if (!rs->bitmap_loaded) {
1894                         bitmap_load(&rs->md);
1895                         rs->bitmap_loaded = 1;
1896                 } else {
1897                         /*
1898                          * A secondary resume while the device is active.
1899                          * Take this opportunity to check whether any failed
1900                          * devices are reachable again.
1901                          */
1902                         attempt_restore_of_faulty_devices(rs);
1903                 }
1904
1905                 clear_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
1906         }
1907
1908         mddev_resume(&rs->md);
1909 }
1910
1911 static struct target_type raid_target = {
1912         .name = "raid",
1913         .version = {1, 8, 1},
1914         .module = THIS_MODULE,
1915         .ctr = raid_ctr,
1916         .dtr = raid_dtr,
1917         .map = raid_map,
1918         .status = raid_status,
1919         .message = raid_message,
1920         .iterate_devices = raid_iterate_devices,
1921         .io_hints = raid_io_hints,
1922         .presuspend = raid_presuspend,
1923         .postsuspend = raid_postsuspend,
1924         .resume = raid_resume,
1925 };
1926
1927 static int __init dm_raid_init(void)
1928 {
1929         DMINFO("Loading target version %u.%u.%u",
1930                raid_target.version[0],
1931                raid_target.version[1],
1932                raid_target.version[2]);
1933         return dm_register_target(&raid_target);
1934 }
1935
1936 static void __exit dm_raid_exit(void)
1937 {
1938         dm_unregister_target(&raid_target);
1939 }
1940
1941 module_init(dm_raid_init);
1942 module_exit(dm_raid_exit);
1943
1944 module_param(devices_handle_discard_safely, bool, 0644);
1945 MODULE_PARM_DESC(devices_handle_discard_safely,
1946                  "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
1947
1948 MODULE_DESCRIPTION(DM_NAME " raid4/5/6 target");
1949 MODULE_ALIAS("dm-raid1");
1950 MODULE_ALIAS("dm-raid10");
1951 MODULE_ALIAS("dm-raid4");
1952 MODULE_ALIAS("dm-raid5");
1953 MODULE_ALIAS("dm-raid6");
1954 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
1955 MODULE_LICENSE("GPL");