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 /*
  23  * Minimum sectors of free reshape space per raid device
  24  */
  25 #define MIN_FREE_RESHAPE_SPACE to_sector(4*4096)
  26
  27 static bool devices_handle_discard_safely = false;
  28
  29 /*
  30  * The following flags are used by dm-raid.c to set up the array state.
  31  * They must be cleared before md_run is called.
  32  */
  33 #define FirstUse 10             /* rdev flag */
  34
  35 struct raid_dev {
  36         /*
  37          * Two DM devices, one to hold metadata and one to hold the
  38          * actual data/parity.  The reason for this is to not confuse
  39          * ti->len and give more flexibility in altering size and
  40          * characteristics.
  41          *
  42          * While it is possible for this device to be associated
  43          * with a different physical device than the data_dev, it
  44          * is intended for it to be the same.
  45          *    |--------- Physical Device ---------|
  46          *    |- meta_dev -|------ data_dev ------|
  47          */
  48         struct dm_dev *meta_dev;
  49         struct dm_dev *data_dev;
  50         struct md_rdev rdev;
  51 };
  52
  53 /*
  54  * Bits for establishing rs->ctr_flags
  55  *
  56  * 1 = no flag value
  57  * 2 = flag with value
  58  */
  59 #define __CTR_FLAG_SYNC                 0  /* 1 */ /* Not with raid0! */
  60 #define __CTR_FLAG_NOSYNC               1  /* 1 */ /* Not with raid0! */
  61 #define __CTR_FLAG_REBUILD              2  /* 2 */ /* Not with raid0! */
  62 #define __CTR_FLAG_DAEMON_SLEEP         3  /* 2 */ /* Not with raid0! */
  63 #define __CTR_FLAG_MIN_RECOVERY_RATE    4  /* 2 */ /* Not with raid0! */
  64 #define __CTR_FLAG_MAX_RECOVERY_RATE    5  /* 2 */ /* Not with raid0! */
  65 #define __CTR_FLAG_MAX_WRITE_BEHIND     6  /* 2 */ /* Only with raid1! */
  66 #define __CTR_FLAG_WRITE_MOSTLY         7  /* 2 */ /* Only with raid1! */
  67 #define __CTR_FLAG_STRIPE_CACHE         8  /* 2 */ /* Only with raid4/5/6! */
  68 #define __CTR_FLAG_REGION_SIZE          9  /* 2 */ /* Not with raid0! */
  69 #define __CTR_FLAG_RAID10_COPIES        10 /* 2 */ /* Only with raid10 */
  70 #define __CTR_FLAG_RAID10_FORMAT        11 /* 2 */ /* Only with raid10 */
  71 /* New for v1.9.0 */
  72 #define __CTR_FLAG_DELTA_DISKS          12 /* 2 */ /* Only with reshapable raid4/5/6/10! */
  73 #define __CTR_FLAG_DATA_OFFSET          13 /* 2 */ /* Only with reshapable raid4/5/6/10! */
  74 #define __CTR_FLAG_RAID10_USE_NEAR_SETS 14 /* 2 */ /* Only with raid10! */
  75
  76 /*
  77  * Flags for rs->ctr_flags field.
  78  */
  79 #define CTR_FLAG_SYNC                   (1 << __CTR_FLAG_SYNC)
  80 #define CTR_FLAG_NOSYNC                 (1 << __CTR_FLAG_NOSYNC)
  81 #define CTR_FLAG_REBUILD                (1 << __CTR_FLAG_REBUILD)
  82 #define CTR_FLAG_DAEMON_SLEEP           (1 << __CTR_FLAG_DAEMON_SLEEP)
  83 #define CTR_FLAG_MIN_RECOVERY_RATE      (1 << __CTR_FLAG_MIN_RECOVERY_RATE)
  84 #define CTR_FLAG_MAX_RECOVERY_RATE      (1 << __CTR_FLAG_MAX_RECOVERY_RATE)
  85 #define CTR_FLAG_MAX_WRITE_BEHIND       (1 << __CTR_FLAG_MAX_WRITE_BEHIND)
  86 #define CTR_FLAG_WRITE_MOSTLY           (1 << __CTR_FLAG_WRITE_MOSTLY)
  87 #define CTR_FLAG_STRIPE_CACHE           (1 << __CTR_FLAG_STRIPE_CACHE)
  88 #define CTR_FLAG_REGION_SIZE            (1 << __CTR_FLAG_REGION_SIZE)
  89 #define CTR_FLAG_RAID10_COPIES          (1 << __CTR_FLAG_RAID10_COPIES)
  90 #define CTR_FLAG_RAID10_FORMAT          (1 << __CTR_FLAG_RAID10_FORMAT)
  91 #define CTR_FLAG_DELTA_DISKS            (1 << __CTR_FLAG_DELTA_DISKS)
  92 #define CTR_FLAG_DATA_OFFSET            (1 << __CTR_FLAG_DATA_OFFSET)
  93 #define CTR_FLAG_RAID10_USE_NEAR_SETS   (1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
  94
  95 /*
  96  * Definitions of various constructor flags to
  97  * be used in checks of valid / invalid flags
  98  * per raid level.
  99  */
 100 /* Define all any sync flags */
 101 #define CTR_FLAGS_ANY_SYNC              (CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
 102
 103 /* Define flags for options without argument (e.g. 'nosync') */
 104 #define CTR_FLAG_OPTIONS_NO_ARGS        (CTR_FLAGS_ANY_SYNC | \
 105                                          CTR_FLAG_RAID10_USE_NEAR_SETS)
 106
 107 /* Define flags for options with one argument (e.g. 'delta_disks +2') */
 108 #define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
 109                                   CTR_FLAG_WRITE_MOSTLY | \
 110                                   CTR_FLAG_DAEMON_SLEEP | \
 111                                   CTR_FLAG_MIN_RECOVERY_RATE | \
 112                                   CTR_FLAG_MAX_RECOVERY_RATE | \
 113                                   CTR_FLAG_MAX_WRITE_BEHIND | \
 114                                   CTR_FLAG_STRIPE_CACHE | \
 115                                   CTR_FLAG_REGION_SIZE | \
 116                                   CTR_FLAG_RAID10_COPIES | \
 117                                   CTR_FLAG_RAID10_FORMAT | \
 118                                   CTR_FLAG_DELTA_DISKS | \
 119                                   CTR_FLAG_DATA_OFFSET)
 120
 121 /* Valid options definitions per raid level... */
 122
 123 /* "raid0" does only accept data offset */
 124 #define RAID0_VALID_FLAGS       (CTR_FLAG_DATA_OFFSET)
 125
 126 /* "raid1" does not accept stripe cache, data offset, delta_disks or any raid10 options */
 127 #define RAID1_VALID_FLAGS       (CTR_FLAGS_ANY_SYNC | \
 128                                  CTR_FLAG_REBUILD | \
 129                                  CTR_FLAG_WRITE_MOSTLY | \
 130                                  CTR_FLAG_DAEMON_SLEEP | \
 131                                  CTR_FLAG_MIN_RECOVERY_RATE | \
 132                                  CTR_FLAG_MAX_RECOVERY_RATE | \
 133                                  CTR_FLAG_MAX_WRITE_BEHIND | \
 134                                  CTR_FLAG_REGION_SIZE | \
 135                                  CTR_FLAG_DATA_OFFSET)
 136
 137 /* "raid10" does not accept any raid1 or stripe cache options */
 138 #define RAID10_VALID_FLAGS      (CTR_FLAGS_ANY_SYNC | \
 139                                  CTR_FLAG_REBUILD | \
 140                                  CTR_FLAG_DAEMON_SLEEP | \
 141                                  CTR_FLAG_MIN_RECOVERY_RATE | \
 142                                  CTR_FLAG_MAX_RECOVERY_RATE | \
 143                                  CTR_FLAG_REGION_SIZE | \
 144                                  CTR_FLAG_RAID10_COPIES | \
 145                                  CTR_FLAG_RAID10_FORMAT | \
 146                                  CTR_FLAG_DELTA_DISKS | \
 147                                  CTR_FLAG_DATA_OFFSET | \
 148                                  CTR_FLAG_RAID10_USE_NEAR_SETS)
 149
 150 /*
 151  * "raid4/5/6" do not accept any raid1 or raid10 specific options
 152  *
 153  * "raid6" does not accept "nosync", because it is not guaranteed
 154  * that both parity and q-syndrome are being written properly with
 155  * any writes
 156  */
 157 #define RAID45_VALID_FLAGS      (CTR_FLAGS_ANY_SYNC | \
 158                                  CTR_FLAG_REBUILD | \
 159                                  CTR_FLAG_DAEMON_SLEEP | \
 160                                  CTR_FLAG_MIN_RECOVERY_RATE | \
 161                                  CTR_FLAG_MAX_RECOVERY_RATE | \
 162                                  CTR_FLAG_MAX_WRITE_BEHIND | \
 163                                  CTR_FLAG_STRIPE_CACHE | \
 164                                  CTR_FLAG_REGION_SIZE | \
 165                                  CTR_FLAG_DELTA_DISKS | \
 166                                  CTR_FLAG_DATA_OFFSET)
 167
 168 #define RAID6_VALID_FLAGS       (CTR_FLAG_SYNC | \
 169                                  CTR_FLAG_REBUILD | \
 170                                  CTR_FLAG_DAEMON_SLEEP | \
 171                                  CTR_FLAG_MIN_RECOVERY_RATE | \
 172                                  CTR_FLAG_MAX_RECOVERY_RATE | \
 173                                  CTR_FLAG_MAX_WRITE_BEHIND | \
 174                                  CTR_FLAG_STRIPE_CACHE | \
 175                                  CTR_FLAG_REGION_SIZE | \
 176                                  CTR_FLAG_DELTA_DISKS | \
 177                                  CTR_FLAG_DATA_OFFSET)
 178 /* ...valid options definitions per raid level */
 179
 180 /*
 181  * Flags for rs->runtime_flags field
 182  * (RT_FLAG prefix meaning "runtime flag")
 183  *
 184  * These are all internal and used to define runtime state,
 185  * e.g. to prevent another resume from preresume processing
 186  * the raid set all over again.
 187  */
 188 #define RT_FLAG_RS_PRERESUMED           0
 189 #define RT_FLAG_RS_RESUMED              1
 190 #define RT_FLAG_RS_BITMAP_LOADED        2
 191 #define RT_FLAG_UPDATE_SBS              3
 192 #define RT_FLAG_RESHAPE_RS              4
 193 #define RT_FLAG_KEEP_RS_FROZEN          5
 194
 195 /* Array elements of 64 bit needed for rebuild/write_mostly bits */
 196 #define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
 197
 198 /*
 199  * raid set level, layout and chunk sectors backup/restore
 200  */
 201 struct rs_layout {
 202         int new_level;
 203         int new_layout;
 204         int new_chunk_sectors;
 205 };
 206
 207 struct raid_set {
 208         struct dm_target *ti;
 209
 210         uint32_t bitmap_loaded;
 211         uint32_t stripe_cache_entries;
 212         unsigned long ctr_flags;
 213         unsigned long runtime_flags;
 214
 215         uint64_t rebuild_disks[DISKS_ARRAY_ELEMS];
 216
 217         int raid_disks;
 218         int delta_disks;
 219         int data_offset;
 220         int raid10_copies;
 221         int requested_bitmap_chunk_sectors;
 222
 223         struct mddev md;
 224         struct raid_type *raid_type;
 225         struct dm_target_callbacks callbacks;
 226
 227         struct raid_dev dev[0];
 228 };
 229
 230 static void rs_config_backup(struct raid_set *rs, struct rs_layout *l)
 231 {
 232         struct mddev *mddev = &rs->md;
 233
 234         l->new_level = mddev->new_level;
 235         l->new_layout = mddev->new_layout;
 236         l->new_chunk_sectors = mddev->new_chunk_sectors;
 237 }
 238
 239 static void rs_config_restore(struct raid_set *rs, struct rs_layout *l)
 240 {
 241         struct mddev *mddev = &rs->md;
 242
 243         mddev->new_level = l->new_level;
 244         mddev->new_layout = l->new_layout;
 245         mddev->new_chunk_sectors = l->new_chunk_sectors;
 246 }
 247
 248 /* raid10 algorithms (i.e. formats) */
 249 #define ALGORITHM_RAID10_DEFAULT        0
 250 #define ALGORITHM_RAID10_NEAR           1
 251 #define ALGORITHM_RAID10_OFFSET         2
 252 #define ALGORITHM_RAID10_FAR            3
 253
 254 /* Supported raid types and properties. */
 255 static struct raid_type {
 256         const char *name;               /* RAID algorithm. */
 257         const char *descr;              /* Descriptor text for logging. */
 258         const unsigned parity_devs;     /* # of parity devices. */
 259         const unsigned minimal_devs;    /* minimal # of devices in set. */
 260         const unsigned level;           /* RAID level. */
 261         const unsigned algorithm;       /* RAID algorithm. */
 262 } raid_types[] = {
 263         {"raid0",         "raid0 (striping)",                       0, 2, 0,  0 /* NONE */},
 264         {"raid1",         "raid1 (mirroring)",                      0, 2, 1,  0 /* NONE */},
 265         {"raid10_far",    "raid10 far (striped mirrors)",           0, 2, 10, ALGORITHM_RAID10_FAR},
 266         {"raid10_offset", "raid10 offset (striped mirrors)",        0, 2, 10, ALGORITHM_RAID10_OFFSET},
 267         {"raid10_near",   "raid10 near (striped mirrors)",          0, 2, 10, ALGORITHM_RAID10_NEAR},
 268         {"raid10",        "raid10 (striped mirrors)",               0, 2, 10, ALGORITHM_RAID10_DEFAULT},
 269         {"raid4",         "raid4 (dedicated last parity disk)",     1, 2, 4,  ALGORITHM_PARITY_N}, /* raid4 layout = raid5_n */
 270         {"raid5_n",       "raid5 (dedicated last parity disk)",     1, 2, 5,  ALGORITHM_PARITY_N},
 271         {"raid5_ls",      "raid5 (left symmetric)",                 1, 2, 5,  ALGORITHM_LEFT_SYMMETRIC},
 272         {"raid5_rs",      "raid5 (right symmetric)",                1, 2, 5,  ALGORITHM_RIGHT_SYMMETRIC},
 273         {"raid5_la",      "raid5 (left asymmetric)",                1, 2, 5,  ALGORITHM_LEFT_ASYMMETRIC},
 274         {"raid5_ra",      "raid5 (right asymmetric)",               1, 2, 5,  ALGORITHM_RIGHT_ASYMMETRIC},
 275         {"raid6_zr",      "raid6 (zero restart)",                   2, 4, 6,  ALGORITHM_ROTATING_ZERO_RESTART},
 276         {"raid6_nr",      "raid6 (N restart)",                      2, 4, 6,  ALGORITHM_ROTATING_N_RESTART},
 277         {"raid6_nc",      "raid6 (N continue)",                     2, 4, 6,  ALGORITHM_ROTATING_N_CONTINUE},
 278         {"raid6_n_6",     "raid6 (dedicated parity/Q n/6)",         2, 4, 6,  ALGORITHM_PARITY_N_6},
 279         {"raid6_ls_6",    "raid6 (left symmetric dedicated Q 6)",   2, 4, 6,  ALGORITHM_LEFT_SYMMETRIC_6},
 280         {"raid6_rs_6",    "raid6 (right symmetric dedicated Q 6)",  2, 4, 6,  ALGORITHM_RIGHT_SYMMETRIC_6},
 281         {"raid6_la_6",    "raid6 (left asymmetric dedicated Q 6)",  2, 4, 6,  ALGORITHM_LEFT_ASYMMETRIC_6},
 282         {"raid6_ra_6",    "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6,  ALGORITHM_RIGHT_ASYMMETRIC_6}
 283 };
 284
 285 /* True, if @v is in inclusive range [@min, @max] */
 286 static bool __within_range(long v, long min, long max)
 287 {
 288         return v >= min && v <= max;
 289 }
 290
 291 /* All table line arguments are defined here */
 292 static struct arg_name_flag {
 293         const unsigned long flag;
 294         const char *name;
 295 } __arg_name_flags[] = {
 296         { CTR_FLAG_SYNC, "sync"},
 297         { CTR_FLAG_NOSYNC, "nosync"},
 298         { CTR_FLAG_REBUILD, "rebuild"},
 299         { CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
 300         { CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
 301         { CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
 302         { CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
 303         { CTR_FLAG_WRITE_MOSTLY, "writemostly"},
 304         { CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
 305         { CTR_FLAG_REGION_SIZE, "region_size"},
 306         { CTR_FLAG_RAID10_COPIES, "raid10_copies"},
 307         { CTR_FLAG_RAID10_FORMAT, "raid10_format"},
 308         { CTR_FLAG_DATA_OFFSET, "data_offset"},
 309         { CTR_FLAG_DELTA_DISKS, "delta_disks"},
 310         { CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
 311 };
 312
 313 /* Return argument name string for given @flag */
 314 static const char *dm_raid_arg_name_by_flag(const uint32_t flag)
 315 {
 316         if (hweight32(flag) == 1) {
 317                 struct arg_name_flag *anf = __arg_name_flags + ARRAY_SIZE(__arg_name_flags);
 318
 319                 while (anf-- > __arg_name_flags)
 320                         if (flag & anf->flag)
 321                                 return anf->name;
 322
 323         } else
 324                 DMERR("%s called with more than one flag!", __func__);
 325
 326         return NULL;
 327 }
 328
 329 /*
 330  * bool helpers to test for various raid levels of a raid set,
 331  * is. it's level as reported by the superblock rather than
 332  * the requested raid_type passed to the constructor.
 333  */
 334 /* Return true, if raid set in @rs is raid0 */
 335 static bool rs_is_raid0(struct raid_set *rs)
 336 {
 337         return !rs->md.level;
 338 }
 339
 340 /* Return true, if raid set in @rs is raid1 */
 341 static bool rs_is_raid1(struct raid_set *rs)
 342 {
 343         return rs->md.level == 1;
 344 }
 345
 346 /* Return true, if raid set in @rs is raid10 */
 347 static bool rs_is_raid10(struct raid_set *rs)
 348 {
 349         return rs->md.level == 10;
 350 }
 351
 352 /* Return true, if raid set in @rs is level 4, 5 or 6 */
 353 static bool rs_is_raid456(struct raid_set *rs)
 354 {
 355         return __within_range(rs->md.level, 4, 6);
 356 }
 357
 358 /* Return true, if raid set in @rs is reshapable */
 359 static unsigned int __is_raid10_far(int layout);
 360 static bool rs_is_reshapable(struct raid_set *rs)
 361 {
 362         return rs_is_raid456(rs) ||
 363                (rs_is_raid10(rs) && !__is_raid10_far(rs->md.new_layout));
 364 }
 365
 366 /* Return true, if raid set in @rs is recovering */
 367 static bool rs_is_recovering(struct raid_set *rs)
 368 {
 369         return rs->md.recovery_cp != MaxSector;
 370 }
 371
 372 /* Return true, if raid set in @rs is reshaping */
 373 static bool rs_is_reshaping(struct raid_set *rs)
 374 {
 375         return rs->md.reshape_position != MaxSector;
 376 }
 377
 378 /*
 379  * bool helpers to test for various raid levels of a raid type
 380  */
 381
 382 /* Return true, if raid type in @rt is raid0 */
 383 static bool rt_is_raid0(struct raid_type *rt)
 384 {
 385         return !rt->level;
 386 }
 387
 388 /* Return true, if raid type in @rt is raid1 */
 389 static bool rt_is_raid1(struct raid_type *rt)
 390 {
 391         return rt->level == 1;
 392 }
 393
 394 /* Return true, if raid type in @rt is raid10 */
 395 static bool rt_is_raid10(struct raid_type *rt)
 396 {
 397         return rt->level == 10;
 398 }
 399
 400 /* Return true, if raid type in @rt is raid4/5 */
 401 static bool rt_is_raid45(struct raid_type *rt)
 402 {
 403         return __within_range(rt->level, 4, 5);
 404 }
 405
 406 /* Return true, if raid type in @rt is raid6 */
 407 static bool rt_is_raid6(struct raid_type *rt)
 408 {
 409         return rt->level == 6;
 410 }
 411
 412 /* Return true, if raid type in @rt is raid4/5/6 */
 413 static bool rt_is_raid456(struct raid_type *rt)
 414 {
 415         return __within_range(rt->level, 4, 6);
 416 }
 417 /* END: raid level bools */
 418
 419 /* Return valid ctr flags for the raid level of @rs */
 420 static unsigned long __valid_flags(struct raid_set *rs)
 421 {
 422         if (rt_is_raid0(rs->raid_type))
 423                 return RAID0_VALID_FLAGS;
 424         else if (rt_is_raid1(rs->raid_type))
 425                 return RAID1_VALID_FLAGS;
 426         else if (rt_is_raid10(rs->raid_type))
 427                 return RAID10_VALID_FLAGS;
 428         else if (rt_is_raid45(rs->raid_type))
 429                 return RAID45_VALID_FLAGS;
 430         else if (rt_is_raid6(rs->raid_type))
 431                 return RAID6_VALID_FLAGS;
 432
 433         return ~0;
 434 }
 435
 436 /*
 437  * Check for valid flags set on @rs
 438  *
 439  * Has to be called after parsing of the ctr flags!
 440  */
 441 static int rs_check_for_valid_flags(struct raid_set *rs)
 442 {
 443         if (rs->ctr_flags & ~__valid_flags(rs)) {
 444                 rs->ti->error = "Invalid flags combination";
 445                 return -EINVAL;
 446         }
 447
 448         return 0;
 449 }
 450
 451 /* MD raid10 bit definitions and helpers */
 452 #define RAID10_OFFSET                   (1 << 16) /* stripes with data copies area adjacent on devices */
 453 #define RAID10_BROCKEN_USE_FAR_SETS     (1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
 454 #define RAID10_USE_FAR_SETS             (1 << 18) /* Use sets instead of whole stripe rotation */
 455 #define RAID10_FAR_COPIES_SHIFT         8         /* raid10 # far copies shift (2nd byte of layout) */
 456
 457 /* Return md raid10 near copies for @layout */
 458 static unsigned int __raid10_near_copies(int layout)
 459 {
 460         return layout & 0xFF;
 461 }
 462
 463 /* Return md raid10 far copies for @layout */
 464 static unsigned int __raid10_far_copies(int layout)
 465 {
 466         return __raid10_near_copies(layout >> RAID10_FAR_COPIES_SHIFT);
 467 }
 468
 469 /* Return true if md raid10 offset for @layout */
 470 static unsigned int __is_raid10_offset(int layout)
 471 {
 472         return layout & RAID10_OFFSET;
 473 }
 474
 475 /* Return true if md raid10 near for @layout */
 476 static unsigned int __is_raid10_near(int layout)
 477 {
 478         return !__is_raid10_offset(layout) && __raid10_near_copies(layout) > 1;
 479 }
 480
 481 /* Return true if md raid10 far for @layout */
 482 static unsigned int __is_raid10_far(int layout)
 483 {
 484         return !__is_raid10_offset(layout) && __raid10_far_copies(layout) > 1;
 485 }
 486
 487 /* Return md raid10 layout string for @layout */
 488 static const char *raid10_md_layout_to_format(int layout)
 489 {
 490         /*
 491          * Bit 16 stands for "offset"
 492          * (i.e. adjacent stripes hold copies)
 493          *
 494          * Refer to MD's raid10.c for details
 495          */
 496         if (__is_raid10_offset(layout))
 497                 return "offset";
 498
 499         if (__raid10_near_copies(layout) > 1)
 500                 return "near";
 501
 502         WARN_ON(__raid10_far_copies(layout) < 2);
 503
 504         return "far";
 505 }
 506
 507 /* Return md raid10 algorithm for @name */
 508 static int raid10_name_to_format(const char *name)
 509 {
 510         if (!strcasecmp(name, "near"))
 511                 return ALGORITHM_RAID10_NEAR;
 512         else if (!strcasecmp(name, "offset"))
 513                 return ALGORITHM_RAID10_OFFSET;
 514         else if (!strcasecmp(name, "far"))
 515                 return ALGORITHM_RAID10_FAR;
 516
 517         return -EINVAL;
 518 }
 519
 520 /* Return md raid10 copies for @layout */
 521 static unsigned int raid10_md_layout_to_copies(int layout)
 522 {
 523         return __raid10_near_copies(layout) > 1 ?
 524                 __raid10_near_copies(layout) : __raid10_far_copies(layout);
 525 }
 526
 527 /* Return md raid10 format id for @format string */
 528 static int raid10_format_to_md_layout(struct raid_set *rs,
 529                                       unsigned int algorithm,
 530                                       unsigned int copies)
 531 {
 532         unsigned int n = 1, f = 1, r = 0;
 533
 534         /*
 535          * MD resilienece flaw:
 536          *
 537          * enabling use_far_sets for far/offset formats causes copies
 538          * to be colocated on the same devs together with their origins!
 539          *
 540          * -> disable it for now in the definition above
 541          */
 542         if (algorithm == ALGORITHM_RAID10_DEFAULT ||
 543             algorithm == ALGORITHM_RAID10_NEAR)
 544                 n = copies;
 545
 546         else if (algorithm == ALGORITHM_RAID10_OFFSET) {
 547                 f = copies;
 548                 r = RAID10_OFFSET;
 549                 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
 550                         r |= RAID10_USE_FAR_SETS;
 551
 552         } else if (algorithm == ALGORITHM_RAID10_FAR) {
 553                 f = copies;
 554                 r = !RAID10_OFFSET;
 555                 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
 556                         r |= RAID10_USE_FAR_SETS;
 557
 558         } else
 559                 return -EINVAL;
 560
 561         return r | (f << RAID10_FAR_COPIES_SHIFT) | n;
 562 }
 563 /* END: MD raid10 bit definitions and helpers */
 564
 565 /* Check for any of the raid10 algorithms */
 566 static int __got_raid10(struct raid_type *rtp, const int layout)
 567 {
 568         if (rtp->level == 10) {
 569                 switch (rtp->algorithm) {
 570                 case ALGORITHM_RAID10_DEFAULT:
 571                 case ALGORITHM_RAID10_NEAR:
 572                         return __is_raid10_near(layout);
 573                 case ALGORITHM_RAID10_OFFSET:
 574                         return __is_raid10_offset(layout);
 575                 case ALGORITHM_RAID10_FAR:
 576                         return __is_raid10_far(layout);
 577                 default:
 578                         break;
 579                 }
 580         }
 581
 582         return 0;
 583 }
 584
 585 /* Return raid_type for @name */
 586 static struct raid_type *get_raid_type(const char *name)
 587 {
 588         struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
 589
 590         while (rtp-- > raid_types)
 591                 if (!strcasecmp(rtp->name, name))
 592                         return rtp;
 593
 594         return NULL;
 595 }
 596
 597 /* Return raid_type for @name based derived from @level and @layout */
 598 static struct raid_type *get_raid_type_by_ll(const int level, const int layout)
 599 {
 600         struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
 601
 602         while (rtp-- > raid_types) {
 603                 /* RAID10 special checks based on @layout flags/properties */
 604                 if (rtp->level == level &&
 605                     (__got_raid10(rtp, layout) || rtp->algorithm == layout))
 606                         return rtp;
 607         }
 608
 609         return NULL;
 610 }
 611
 612 /*
 613  * Conditionally change bdev capacity of @rs
 614  * in case of a disk add/remove reshape
 615  */
 616 static void rs_set_capacity(struct raid_set *rs)
 617 {
 618         struct mddev *mddev = &rs->md;
 619
 620         if (rs->ti->len != mddev->array_sectors) {
 621                 struct gendisk *gendisk = dm_disk(dm_table_get_md(rs->ti->table));
 622
 623                 set_capacity(gendisk, mddev->array_sectors);
 624                 revalidate_disk(gendisk);
 625         }
 626 }
 627
 628 /*
 629  * Set the mddev properties in @rs to the current
 630  * ones retrieved from the freshest superblock
 631  */
 632 static void rs_set_cur(struct raid_set *rs)
 633 {
 634         struct mddev *mddev = &rs->md;
 635
 636         mddev->new_level = mddev->level;
 637         mddev->new_layout = mddev->layout;
 638         mddev->new_chunk_sectors = mddev->chunk_sectors;
 639 }
 640
 641 /*
 642  * Set the mddev properties in @rs to the new
 643  * ones requested by the ctr
 644  */
 645 static void rs_set_new(struct raid_set *rs)
 646 {
 647         struct mddev *mddev = &rs->md;
 648
 649         mddev->level = mddev->new_level;
 650         mddev->layout = mddev->new_layout;
 651         mddev->chunk_sectors = mddev->new_chunk_sectors;
 652         mddev->raid_disks = rs->raid_disks;
 653         mddev->delta_disks = 0;
 654 }
 655
 656 static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
 657                                        unsigned raid_devs)
 658 {
 659         unsigned i;
 660         struct raid_set *rs;
 661
 662         if (raid_devs <= raid_type->parity_devs) {
 663                 ti->error = "Insufficient number of devices";
 664                 return ERR_PTR(-EINVAL);
 665         }
 666
 667         rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
 668         if (!rs) {
 669                 ti->error = "Cannot allocate raid context";
 670                 return ERR_PTR(-ENOMEM);
 671         }
 672
 673         mddev_init(&rs->md);
 674
 675         rs->raid_disks = raid_devs;
 676         rs->delta_disks = 0;
 677
 678         rs->ti = ti;
 679         rs->raid_type = raid_type;
 680         rs->stripe_cache_entries = 256;
 681         rs->md.raid_disks = raid_devs;
 682         rs->md.level = raid_type->level;
 683         rs->md.new_level = rs->md.level;
 684         rs->md.layout = raid_type->algorithm;
 685         rs->md.new_layout = rs->md.layout;
 686         rs->md.delta_disks = 0;
 687         rs->md.recovery_cp = rs_is_raid0(rs) ? MaxSector : 0;
 688
 689         for (i = 0; i < raid_devs; i++)
 690                 md_rdev_init(&rs->dev[i].rdev);
 691
 692         /*
 693          * Remaining items to be initialized by further RAID params:
 694          *  rs->md.persistent
 695          *  rs->md.external
 696          *  rs->md.chunk_sectors
 697          *  rs->md.new_chunk_sectors
 698          *  rs->md.dev_sectors
 699          */
 700
 701         return rs;
 702 }
 703
 704 static void raid_set_free(struct raid_set *rs)
 705 {
 706         int i;
 707
 708         for (i = 0; i < rs->md.raid_disks; i++) {
 709                 if (rs->dev[i].meta_dev)
 710                         dm_put_device(rs->ti, rs->dev[i].meta_dev);
 711                 md_rdev_clear(&rs->dev[i].rdev);
 712                 if (rs->dev[i].data_dev)
 713                         dm_put_device(rs->ti, rs->dev[i].data_dev);
 714         }
 715
 716         kfree(rs);
 717 }
 718
 719 /*
 720  * For every device we have two words
 721  *  <meta_dev>: meta device name or '-' if missing
 722  *  <data_dev>: data device name or '-' if missing
 723  *
 724  * The following are permitted:
 725  *    - -
 726  *    - <data_dev>
 727  *    <meta_dev> <data_dev>
 728  *
 729  * The following is not allowed:
 730  *    <meta_dev> -
 731  *
 732  * This code parses those words.  If there is a failure,
 733  * the caller must use raid_set_free() to unwind the operations.
 734  */
 735 static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
 736 {
 737         int i;
 738         int rebuild = 0;
 739         int metadata_available = 0;
 740         int r = 0;
 741         const char *arg;
 742
 743         /* Put off the number of raid devices argument to get to dev pairs */
 744         arg = dm_shift_arg(as);
 745         if (!arg)
 746                 return -EINVAL;
 747
 748         for (i = 0; i < rs->md.raid_disks; i++) {
 749                 rs->dev[i].rdev.raid_disk = i;
 750
 751                 rs->dev[i].meta_dev = NULL;
 752                 rs->dev[i].data_dev = NULL;
 753
 754                 /*
 755                  * There are no offsets, since there is a separate device
 756                  * for data and metadata.
 757                  */
 758                 rs->dev[i].rdev.data_offset = 0;
 759                 rs->dev[i].rdev.mddev = &rs->md;
 760
 761                 arg = dm_shift_arg(as);
 762                 if (!arg)
 763                         return -EINVAL;
 764
 765                 if (strcmp(arg, "-")) {
 766                         r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
 767                                           &rs->dev[i].meta_dev);
 768                         if (r) {
 769                                 rs->ti->error = "RAID metadata device lookup failure";
 770                                 return r;
 771                         }
 772
 773                         rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
 774                         if (!rs->dev[i].rdev.sb_page) {
 775                                 rs->ti->error = "Failed to allocate superblock page";
 776                                 return -ENOMEM;
 777                         }
 778                 }
 779
 780                 arg = dm_shift_arg(as);
 781                 if (!arg)
 782                         return -EINVAL;
 783
 784                 if (!strcmp(arg, "-")) {
 785                         if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
 786                             (!rs->dev[i].rdev.recovery_offset)) {
 787                                 rs->ti->error = "Drive designated for rebuild not specified";
 788                                 return -EINVAL;
 789                         }
 790
 791                         if (rs->dev[i].meta_dev) {
 792                                 rs->ti->error = "No data device supplied with metadata device";
 793                                 return -EINVAL;
 794                         }
 795
 796                         continue;
 797                 }
 798
 799                 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
 800                                   &rs->dev[i].data_dev);
 801                 if (r) {
 802                         rs->ti->error = "RAID device lookup failure";
 803                         return r;
 804                 }
 805
 806                 if (rs->dev[i].meta_dev) {
 807                         metadata_available = 1;
 808                         rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
 809                 }
 810                 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
 811                 list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
 812                 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
 813                         rebuild++;
 814         }
 815
 816         if (metadata_available) {
 817                 rs->md.external = 0;
 818                 rs->md.persistent = 1;
 819                 rs->md.major_version = 2;
 820         } else if (rebuild && !rs->md.recovery_cp) {
 821                 /*
 822                  * Without metadata, we will not be able to tell if the array
 823                  * is in-sync or not - we must assume it is not.  Therefore,
 824                  * it is impossible to rebuild a drive.
 825                  *
 826                  * Even if there is metadata, the on-disk information may
 827                  * indicate that the array is not in-sync and it will then
 828                  * fail at that time.
 829                  *
 830                  * User could specify 'nosync' option if desperate.
 831                  */
 832                 rs->ti->error = "Unable to rebuild drive while array is not in-sync";
 833                 return -EINVAL;
 834         }
 835
 836         return 0;
 837 }
 838
 839 /*
 840  * validate_region_size
 841  * @rs
 842  * @region_size:  region size in sectors.  If 0, pick a size (4MiB default).
 843  *
 844  * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
 845  * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
 846  *
 847  * Returns: 0 on success, -EINVAL on failure.
 848  */
 849 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
 850 {
 851         unsigned long min_region_size = rs->ti->len / (1 << 21);
 852
 853         if (!region_size) {
 854                 /*
 855                  * Choose a reasonable default.  All figures in sectors.
 856                  */
 857                 if (min_region_size > (1 << 13)) {
 858                         /* If not a power of 2, make it the next power of 2 */
 859                         region_size = roundup_pow_of_two(min_region_size);
 860                         DMINFO("Choosing default region size of %lu sectors",
 861                                region_size);
 862                 } else {
 863                         DMINFO("Choosing default region size of 4MiB");
 864                         region_size = 1 << 13; /* sectors */
 865                 }
 866         } else {
 867                 /*
 868                  * Validate user-supplied value.
 869                  */
 870                 if (region_size > rs->ti->len) {
 871                         rs->ti->error = "Supplied region size is too large";
 872                         return -EINVAL;
 873                 }
 874
 875                 if (region_size < min_region_size) {
 876                         DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
 877                               region_size, min_region_size);
 878                         rs->ti->error = "Supplied region size is too small";
 879                         return -EINVAL;
 880                 }
 881
 882                 if (!is_power_of_2(region_size)) {
 883                         rs->ti->error = "Region size is not a power of 2";
 884                         return -EINVAL;
 885                 }
 886
 887                 if (region_size < rs->md.chunk_sectors) {
 888                         rs->ti->error = "Region size is smaller than the chunk size";
 889                         return -EINVAL;
 890                 }
 891         }
 892
 893         /*
 894          * Convert sectors to bytes.
 895          */
 896         rs->md.bitmap_info.chunksize = (region_size << 9);
 897
 898         return 0;
 899 }
 900
 901 /*
 902  * validate_raid_redundancy
 903  * @rs
 904  *
 905  * Determine if there are enough devices in the array that haven't
 906  * failed (or are being rebuilt) to form a usable array.
 907  *
 908  * Returns: 0 on success, -EINVAL on failure.
 909  */
 910 static int validate_raid_redundancy(struct raid_set *rs)
 911 {
 912         unsigned i, rebuild_cnt = 0;
 913         unsigned rebuilds_per_group = 0, copies;
 914         unsigned group_size, last_group_start;
 915
 916         for (i = 0; i < rs->md.raid_disks; i++)
 917                 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
 918                     !rs->dev[i].rdev.sb_page)
 919                         rebuild_cnt++;
 920
 921         switch (rs->raid_type->level) {
 922         case 1:
 923                 if (rebuild_cnt >= rs->md.raid_disks)
 924                         goto too_many;
 925                 break;
 926         case 4:
 927         case 5:
 928         case 6:
 929                 if (rebuild_cnt > rs->raid_type->parity_devs)
 930                         goto too_many;
 931                 break;
 932         case 10:
 933                 copies = raid10_md_layout_to_copies(rs->md.new_layout);
 934                 if (rebuild_cnt < copies)
 935                         break;
 936
 937                 /*
 938                  * It is possible to have a higher rebuild count for RAID10,
 939                  * as long as the failed devices occur in different mirror
 940                  * groups (i.e. different stripes).
 941                  *
 942                  * When checking "near" format, make sure no adjacent devices
 943                  * have failed beyond what can be handled.  In addition to the
 944                  * simple case where the number of devices is a multiple of the
 945                  * number of copies, we must also handle cases where the number
 946                  * of devices is not a multiple of the number of copies.
 947                  * E.g.    dev1 dev2 dev3 dev4 dev5
 948                  *          A    A    B    B    C
 949                  *          C    D    D    E    E
 950                  */
 951                 if (__is_raid10_near(rs->md.new_layout)) {
 952                         for (i = 0; i < rs->raid_disks; i++) {
 953                                 if (!(i % copies))
 954                                         rebuilds_per_group = 0;
 955                                 if ((!rs->dev[i].rdev.sb_page ||
 956                                     !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
 957                                     (++rebuilds_per_group >= copies))
 958                                         goto too_many;
 959                         }
 960                         break;
 961                 }
 962
 963                 /*
 964                  * When checking "far" and "offset" formats, we need to ensure
 965                  * that the device that holds its copy is not also dead or
 966                  * being rebuilt.  (Note that "far" and "offset" formats only
 967                  * support two copies right now.  These formats also only ever
 968                  * use the 'use_far_sets' variant.)
 969                  *
 970                  * This check is somewhat complicated by the need to account
 971                  * for arrays that are not a multiple of (far) copies.  This
 972                  * results in the need to treat the last (potentially larger)
 973                  * set differently.
 974                  */
 975                 group_size = (rs->md.raid_disks / copies);
 976                 last_group_start = (rs->md.raid_disks / group_size) - 1;
 977                 last_group_start *= group_size;
 978                 for (i = 0; i < rs->md.raid_disks; i++) {
 979                         if (!(i % copies) && !(i > last_group_start))
 980                                 rebuilds_per_group = 0;
 981                         if ((!rs->dev[i].rdev.sb_page ||
 982                              !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
 983                             (++rebuilds_per_group >= copies))
 984                                         goto too_many;
 985                 }
 986                 break;
 987         default:
 988                 if (rebuild_cnt)
 989                         return -EINVAL;
 990         }
 991
 992         return 0;
 993
 994 too_many:
 995         return -EINVAL;
 996 }
 997
 998 /*
 999  * Possible arguments are...
1000  *      <chunk_size> [optional_args]
1001  *
1002  * Argument definitions
1003  *    <chunk_size>                      The number of sectors per disk that
1004  *                                      will form the "stripe"
1005  *    [[no]sync]                        Force or prevent recovery of the
1006  *                                      entire array
1007  *    [rebuild <idx>]                   Rebuild the drive indicated by the index
1008  *    [daemon_sleep <ms>]               Time between bitmap daemon work to
1009  *                                      clear bits
1010  *    [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1011  *    [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1012  *    [write_mostly <idx>]              Indicate a write mostly drive via index
1013  *    [max_write_behind <sectors>]      See '-write-behind=' (man mdadm)
1014  *    [stripe_cache <sectors>]          Stripe cache size for higher RAIDs
1015  *    [region_size <sectors>]           Defines granularity of bitmap
1016  *
1017  * RAID10-only options:
1018  *    [raid10_copies <# copies>]        Number of copies.  (Default: 2)
1019  *    [raid10_format <near|far|offset>] Layout algorithm.  (Default: near)
1020  */
1021 static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1022                              unsigned num_raid_params)
1023 {
1024         int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1025         unsigned raid10_copies = 2;
1026         unsigned i, write_mostly = 0;
1027         unsigned region_size = 0;
1028         sector_t max_io_len;
1029         const char *arg, *key;
1030         struct raid_dev *rd;
1031         struct raid_type *rt = rs->raid_type;
1032
1033         arg = dm_shift_arg(as);
1034         num_raid_params--; /* Account for chunk_size argument */
1035
1036         if (kstrtoint(arg, 10, &value) < 0) {
1037                 rs->ti->error = "Bad numerical argument given for chunk_size";
1038                 return -EINVAL;
1039         }
1040
1041         /*
1042          * First, parse the in-order required arguments
1043          * "chunk_size" is the only argument of this type.
1044          */
1045         if (rt_is_raid1(rt)) {
1046                 if (value)
1047                         DMERR("Ignoring chunk size parameter for RAID 1");
1048                 value = 0;
1049         } else if (!is_power_of_2(value)) {
1050                 rs->ti->error = "Chunk size must be a power of 2";
1051                 return -EINVAL;
1052         } else if (value < 8) {
1053                 rs->ti->error = "Chunk size value is too small";
1054                 return -EINVAL;
1055         }
1056
1057         rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1058
1059         /*
1060          * We set each individual device as In_sync with a completed
1061          * 'recovery_offset'.  If there has been a device failure or
1062          * replacement then one of the following cases applies:
1063          *
1064          *   1) User specifies 'rebuild'.
1065          *      - Device is reset when param is read.
1066          *   2) A new device is supplied.
1067          *      - No matching superblock found, resets device.
1068          *   3) Device failure was transient and returns on reload.
1069          *      - Failure noticed, resets device for bitmap replay.
1070          *   4) Device hadn't completed recovery after previous failure.
1071          *      - Superblock is read and overrides recovery_offset.
1072          *
1073          * What is found in the superblocks of the devices is always
1074          * authoritative, unless 'rebuild' or '[no]sync' was specified.
1075          */
1076         for (i = 0; i < rs->md.raid_disks; i++) {
1077                 set_bit(In_sync, &rs->dev[i].rdev.flags);
1078                 rs->dev[i].rdev.recovery_offset = MaxSector;
1079         }
1080
1081         /*
1082          * Second, parse the unordered optional arguments
1083          */
1084         for (i = 0; i < num_raid_params; i++) {
1085                 key = dm_shift_arg(as);
1086                 if (!key) {
1087                         rs->ti->error = "Not enough raid parameters given";
1088                         return -EINVAL;
1089                 }
1090
1091                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1092                         if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1093                                 rs->ti->error = "Only one 'nosync' argument allowed";
1094                                 return -EINVAL;
1095                         }
1096                         rs->md.recovery_cp = MaxSector;
1097                         continue;
1098                 }
1099                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1100                         if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1101                                 rs->ti->error = "Only one 'sync' argument allowed";
1102                                 return -EINVAL;
1103                         }
1104                         rs->md.recovery_cp = 0;
1105                         continue;
1106                 }
1107                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1108                         if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1109                                 rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1110                                 return -EINVAL;
1111                         }
1112                         continue;
1113                 }
1114
1115                 arg = dm_shift_arg(as);
1116                 i++; /* Account for the argument pairs */
1117                 if (!arg) {
1118                         rs->ti->error = "Wrong number of raid parameters given";
1119                         return -EINVAL;
1120                 }
1121
1122                 /*
1123                  * Parameters that take a string value are checked here.
1124                  */
1125
1126                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1127                         if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1128                                 rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1129                                 return -EINVAL;
1130                         }
1131                         if (!rt_is_raid10(rt)) {
1132                                 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1133                                 return -EINVAL;
1134                         }
1135                         raid10_format = raid10_name_to_format(arg);
1136                         if (raid10_format < 0) {
1137                                 rs->ti->error = "Invalid 'raid10_format' value given";
1138                                 return raid10_format;
1139                         }
1140                         continue;
1141                 }
1142
1143                 if (kstrtoint(arg, 10, &value) < 0) {
1144                         rs->ti->error = "Bad numerical argument given in raid params";
1145                         return -EINVAL;
1146                 }
1147
1148                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1149                         /*
1150                          * "rebuild" is being passed in by userspace to provide
1151                          * indexes of replaced devices and to set up additional
1152                          * devices on raid level takeover.
1153                          */
1154                         if (!__within_range(value, 0, rs->raid_disks - 1)) {
1155                                 rs->ti->error = "Invalid rebuild index given";
1156                                 return -EINVAL;
1157                         }
1158
1159                         if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1160                                 rs->ti->error = "rebuild for this index already given";
1161                                 return -EINVAL;
1162                         }
1163
1164                         rd = rs->dev + value;
1165                         clear_bit(In_sync, &rd->rdev.flags);
1166                         clear_bit(Faulty, &rd->rdev.flags);
1167                         rd->rdev.recovery_offset = 0;
1168                         set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1169                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1170                         if (!rt_is_raid1(rt)) {
1171                                 rs->ti->error = "write_mostly option is only valid for RAID1";
1172                                 return -EINVAL;
1173                         }
1174
1175                         if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1176                                 rs->ti->error = "Invalid write_mostly index given";
1177                                 return -EINVAL;
1178                         }
1179
1180                         write_mostly++;
1181                         set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1182                         set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1183                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1184                         if (!rt_is_raid1(rt)) {
1185                                 rs->ti->error = "max_write_behind option is only valid for RAID1";
1186                                 return -EINVAL;
1187                         }
1188
1189                         if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1190                                 rs->ti->error = "Only one max_write_behind argument pair allowed";
1191                                 return -EINVAL;
1192                         }
1193
1194                         /*
1195                          * In device-mapper, we specify things in sectors, but
1196                          * MD records this value in kB
1197                          */
1198                         value /= 2;
1199                         if (value > COUNTER_MAX) {
1200                                 rs->ti->error = "Max write-behind limit out of range";
1201                                 return -EINVAL;
1202                         }
1203
1204                         rs->md.bitmap_info.max_write_behind = value;
1205                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1206                         if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1207                                 rs->ti->error = "Only one daemon_sleep argument pair allowed";
1208                                 return -EINVAL;
1209                         }
1210                         if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
1211                                 rs->ti->error = "daemon sleep period out of range";
1212                                 return -EINVAL;
1213                         }
1214                         rs->md.bitmap_info.daemon_sleep = value;
1215                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1216                         /* Userspace passes new data_offset after having extended the the data image LV */
1217                         if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1218                                 rs->ti->error = "Only one data_offset argument pair allowed";
1219                                 return -EINVAL;
1220                         }
1221                         /* Ensure sensible data offset */
1222                         if (value < 0 ||
1223                             (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
1224                                 rs->ti->error = "Bogus data_offset value";
1225                                 return -EINVAL;
1226                         }
1227                         rs->data_offset = value;
1228                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1229                         /* Define the +/-# of disks to add to/remove from the given raid set */
1230                         if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1231                                 rs->ti->error = "Only one delta_disks argument pair allowed";
1232                                 return -EINVAL;
1233                         }
1234                         /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1235                         if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1236                                 rs->ti->error = "Too many delta_disk requested";
1237                                 return -EINVAL;
1238                         }
1239
1240                         rs->delta_disks = value;
1241                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1242                         if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1243                                 rs->ti->error = "Only one stripe_cache argument pair allowed";
1244                                 return -EINVAL;
1245                         }
1246
1247                         if (!rt_is_raid456(rt)) {
1248                                 rs->ti->error = "Inappropriate argument: stripe_cache";
1249                                 return -EINVAL;
1250                         }
1251
1252                         rs->stripe_cache_entries = value;
1253                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1254                         if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1255                                 rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1256                                 return -EINVAL;
1257                         }
1258                         if (value > INT_MAX) {
1259                                 rs->ti->error = "min_recovery_rate out of range";
1260                                 return -EINVAL;
1261                         }
1262                         rs->md.sync_speed_min = (int)value;
1263                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1264                         if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1265                                 rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1266                                 return -EINVAL;
1267                         }
1268                         if (value > INT_MAX) {
1269                                 rs->ti->error = "max_recovery_rate out of range";
1270                                 return -EINVAL;
1271                         }
1272                         rs->md.sync_speed_max = (int)value;
1273                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1274                         if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1275                                 rs->ti->error = "Only one region_size argument pair allowed";
1276                                 return -EINVAL;
1277                         }
1278
1279                         region_size = value;
1280                         rs->requested_bitmap_chunk_sectors = value;
1281                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1282                         if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1283                                 rs->ti->error = "Only one raid10_copies argument pair allowed";
1284                                 return -EINVAL;
1285                         }
1286
1287                         if (!__within_range(value, 2, rs->md.raid_disks)) {
1288                                 rs->ti->error = "Bad value for 'raid10_copies'";
1289                                 return -EINVAL;
1290                         }
1291
1292                         raid10_copies = value;
1293                 } else {
1294                         DMERR("Unable to parse RAID parameter: %s", key);
1295                         rs->ti->error = "Unable to parse RAID parameter";
1296                         return -EINVAL;
1297                 }
1298         }
1299
1300         if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1301             test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1302                 rs->ti->error = "sync and nosync are mutually exclusive";
1303                 return -EINVAL;
1304         }
1305
1306         if (write_mostly >= rs->md.raid_disks) {
1307                 rs->ti->error = "Can't set all raid1 devices to write_mostly";
1308                 return -EINVAL;
1309         }
1310
1311         if (validate_region_size(rs, region_size))
1312                 return -EINVAL;
1313
1314         if (rs->md.chunk_sectors)
1315                 max_io_len = rs->md.chunk_sectors;
1316         else
1317                 max_io_len = region_size;
1318
1319         if (dm_set_target_max_io_len(rs->ti, max_io_len))
1320                 return -EINVAL;
1321
1322         if (rt_is_raid10(rt)) {
1323                 if (raid10_copies > rs->md.raid_disks) {
1324                         rs->ti->error = "Not enough devices to satisfy specification";
1325                         return -EINVAL;
1326                 }
1327
1328                 rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1329                 if (rs->md.new_layout < 0) {
1330                         rs->ti->error = "Error getting raid10 format";
1331                         return rs->md.new_layout;
1332                 }
1333
1334                 rt = get_raid_type_by_ll(10, rs->md.new_layout);
1335                 if (!rt) {
1336                         rs->ti->error = "Failed to recognize new raid10 layout";
1337                         return -EINVAL;
1338                 }
1339
1340                 if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1341                      rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1342                     test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1343                         rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1344                         return -EINVAL;
1345                 }
1346         }
1347
1348         rs->raid10_copies = raid10_copies;
1349
1350         /* Assume there are no metadata devices until the drives are parsed */
1351         rs->md.persistent = 0;
1352         rs->md.external = 1;
1353
1354         /* Check, if any invalid ctr arguments have been passed in for the raid level */
1355         return rs_check_for_valid_flags(rs);
1356 }
1357
1358 /* Set raid4/5/6 cache size */
1359 static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1360 {
1361         int r;
1362         struct r5conf *conf;
1363         struct mddev *mddev = &rs->md;
1364         uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1365         uint32_t nr_stripes = rs->stripe_cache_entries;
1366
1367         if (!rt_is_raid456(rs->raid_type)) {
1368                 rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1369                 return -EINVAL;
1370         }
1371
1372         if (nr_stripes < min_stripes) {
1373                 DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1374                        nr_stripes, min_stripes);
1375                 nr_stripes = min_stripes;
1376         }
1377
1378         conf = mddev->private;
1379         if (!conf) {
1380                 rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1381                 return -EINVAL;
1382         }
1383
1384         /* Try setting number of stripes in raid456 stripe cache */
1385         if (conf->min_nr_stripes != nr_stripes) {
1386                 r = raid5_set_cache_size(mddev, nr_stripes);
1387                 if (r) {
1388                         rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1389                         return r;
1390                 }
1391
1392                 DMINFO("%u stripe cache entries", nr_stripes);
1393         }
1394
1395         return 0;
1396 }
1397
1398 /* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1399 static unsigned int mddev_data_stripes(struct raid_set *rs)
1400 {
1401         return rs->md.raid_disks - rs->raid_type->parity_devs;
1402 }
1403
1404 /* Return # of data stripes of @rs (i.e. as of ctr) */
1405 static unsigned int rs_data_stripes(struct raid_set *rs)
1406 {
1407         return rs->raid_disks - rs->raid_type->parity_devs;
1408 }
1409
1410 /* Calculate the sectors per device and per array used for @rs */
1411 static int rs_set_dev_and_array_sectors(struct raid_set *rs, bool use_mddev)
1412 {
1413         int delta_disks;
1414         unsigned int data_stripes;
1415         struct mddev *mddev = &rs->md;
1416         struct md_rdev *rdev;
1417         sector_t array_sectors = rs->ti->len, dev_sectors = rs->ti->len;
1418         sector_t cur_dev_sectors = rs->dev[0].rdev.sectors;
1419
1420         if (use_mddev) {
1421                 delta_disks = mddev->delta_disks;
1422                 data_stripes = mddev_data_stripes(rs);
1423         } else {
1424                 delta_disks = rs->delta_disks;
1425                 data_stripes = rs_data_stripes(rs);
1426         }
1427
1428         /* Special raid1 case w/o delta_disks support (yet) */
1429         if (rt_is_raid1(rs->raid_type))
1430                 ;
1431         else if (rt_is_raid10(rs->raid_type)) {
1432                 if (rs->raid10_copies < 2 ||
1433                     delta_disks < 0) {
1434                         rs->ti->error = "Bogus raid10 data copies or delta disks";
1435                         return EINVAL;
1436                 }
1437
1438                 dev_sectors *= rs->raid10_copies;
1439                 if (sector_div(dev_sectors, data_stripes))
1440                         goto bad;
1441
1442                 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1443                 if (sector_div(array_sectors, rs->raid10_copies))
1444                         goto bad;
1445
1446         } else if (sector_div(dev_sectors, data_stripes))
1447                 goto bad;
1448
1449         else
1450                 /* Striped layouts */
1451                 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1452
1453         rdev_for_each(rdev, mddev)
1454                 rdev->sectors = dev_sectors;
1455
1456         mddev->array_sectors = array_sectors;
1457         mddev->dev_sectors = dev_sectors;
1458
1459         if (!rs_is_raid0(rs) && dev_sectors > cur_dev_sectors)
1460                 mddev->recovery_cp = dev_sectors;
1461
1462         return 0;
1463 bad:
1464         rs->ti->error = "Target length not divisible by number of data devices";
1465         return EINVAL;
1466 }
1467
1468 static void do_table_event(struct work_struct *ws)
1469 {
1470         struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1471
1472         smp_rmb(); /* Make sure we access most actual mddev properties */
1473         if (!rs_is_reshaping(rs))
1474                 rs_set_capacity(rs);
1475         dm_table_event(rs->ti->table);
1476 }
1477
1478 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
1479 {
1480         struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
1481
1482         return mddev_congested(&rs->md, bits);
1483 }
1484
1485 /*
1486  * Make sure a valid takover (level switch) is being requested on @rs
1487  *
1488  * Conversions of raid sets from one MD personality to another
1489  * have to conform to restrictions which are enforced here.
1490  */
1491 static int rs_check_takeover(struct raid_set *rs)
1492 {
1493         struct mddev *mddev = &rs->md;
1494         unsigned int near_copies;
1495
1496         if (rs->md.degraded) {
1497                 rs->ti->error = "Can't takeover degraded raid set";
1498                 return -EPERM;
1499         }
1500
1501         if (rs_is_reshaping(rs)) {
1502                 rs->ti->error = "Can't takeover reshaping raid set";
1503                 return -EPERM;
1504         }
1505
1506         switch (mddev->level) {
1507         case 0:
1508                 /* raid0 -> raid1/5 with one disk */
1509                 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1510                     mddev->raid_disks == 1)
1511                         return 0;
1512
1513                 /* raid0 -> raid10 */
1514                 if (mddev->new_level == 10 &&
1515                     !(rs->raid_disks % mddev->raid_disks))
1516                         return 0;
1517
1518                 /* raid0 with multiple disks -> raid4/5/6 */
1519                 if (__within_range(mddev->new_level, 4, 6) &&
1520                     mddev->new_layout == ALGORITHM_PARITY_N &&
1521                     mddev->raid_disks > 1)
1522                         return 0;
1523
1524                 break;
1525
1526         case 10:
1527                 /* Can't takeover raid10_offset! */
1528                 if (__is_raid10_offset(mddev->layout))
1529                         break;
1530
1531                 near_copies = __raid10_near_copies(mddev->layout);
1532
1533                 /* raid10* -> raid0 */
1534                 if (mddev->new_level == 0) {
1535                         /* Can takeover raid10_near with raid disks divisable by data copies! */
1536                         if (near_copies > 1 &&
1537                             !(mddev->raid_disks % near_copies)) {
1538                                 mddev->raid_disks /= near_copies;
1539                                 mddev->delta_disks = mddev->raid_disks;
1540                                 return 0;
1541                         }
1542
1543                         /* Can takeover raid10_far */
1544                         if (near_copies == 1 &&
1545                             __raid10_far_copies(mddev->layout) > 1)
1546                                 return 0;
1547
1548                         break;
1549                 }
1550
1551                 /* raid10_{near,far} -> raid1 */
1552                 if (mddev->new_level == 1 &&
1553                     max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1554                         return 0;
1555
1556                 /* raid10_{near,far} with 2 disks -> raid4/5 */
1557                 if (__within_range(mddev->new_level, 4, 5) &&
1558                     mddev->raid_disks == 2)
1559                         return 0;
1560                 break;
1561
1562         case 1:
1563                 /* raid1 with 2 disks -> raid4/5 */
1564                 if (__within_range(mddev->new_level, 4, 5) &&
1565                     mddev->raid_disks == 2) {
1566                         mddev->degraded = 1;
1567                         return 0;
1568                 }
1569
1570                 /* raid1 -> raid0 */
1571                 if (mddev->new_level == 0 &&
1572                     mddev->raid_disks == 1)
1573                         return 0;
1574
1575                 /* raid1 -> raid10 */
1576                 if (mddev->new_level == 10)
1577                         return 0;
1578
1579                 break;
1580
1581         case 4:
1582                 /* raid4 -> raid0 */
1583                 if (mddev->new_level == 0)
1584                         return 0;
1585
1586                 /* raid4 -> raid1/5 with 2 disks */
1587                 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1588                     mddev->raid_disks == 2)
1589                         return 0;
1590
1591                 /* raid4 -> raid5/6 with parity N */
1592                 if (__within_range(mddev->new_level, 5, 6) &&
1593                     mddev->layout == ALGORITHM_PARITY_N)
1594                         return 0;
1595                 break;
1596
1597         case 5:
1598                 /* raid5 with parity N -> raid0 */
1599                 if (mddev->new_level == 0 &&
1600                     mddev->layout == ALGORITHM_PARITY_N)
1601                         return 0;
1602
1603                 /* raid5 with parity N -> raid4 */
1604                 if (mddev->new_level == 4 &&
1605                     mddev->layout == ALGORITHM_PARITY_N)
1606                         return 0;
1607
1608                 /* raid5 with 2 disks -> raid1/4/10 */
1609                 if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1610                     mddev->raid_disks == 2)
1611                         return 0;
1612
1613                 /* raid5_* ->  raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1614                 if (mddev->new_level == 6 &&
1615                     ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1616                       __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1617                         return 0;
1618                 break;
1619
1620         case 6:
1621                 /* raid6 with parity N -> raid0 */
1622                 if (mddev->new_level == 0 &&
1623                     mddev->layout == ALGORITHM_PARITY_N)
1624                         return 0;
1625
1626                 /* raid6 with parity N -> raid4 */
1627                 if (mddev->new_level == 4 &&
1628                     mddev->layout == ALGORITHM_PARITY_N)
1629                         return 0;
1630
1631                 /* raid6_*_n with Q-Syndrome N -> raid5_* */
1632                 if (mddev->new_level == 5 &&
1633                     ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1634                      __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1635                         return 0;
1636
1637         default:
1638                 break;
1639         }
1640
1641         rs->ti->error = "takeover not possible";
1642         return -EINVAL;
1643 }
1644
1645 /* True if @rs requested to be taken over */
1646 static bool rs_takeover_requested(struct raid_set *rs)
1647 {
1648         return rs->md.new_level != rs->md.level;
1649 }
1650
1651 /* True if @rs is requested to reshape by ctr */
1652 static bool rs_reshape_requested(struct raid_set *rs)
1653 {
1654         struct mddev *mddev = &rs->md;
1655
1656         if (!mddev->level)
1657                 return false;
1658
1659         return !__is_raid10_far(mddev->new_layout) &&
1660                mddev->new_level == mddev->level &&
1661                (mddev->new_layout != mddev->layout ||
1662                 mddev->new_chunk_sectors != mddev->chunk_sectors ||
1663                 rs->raid_disks + rs->delta_disks != mddev->raid_disks);
1664 }
1665
1666 /*  Features */
1667 #define FEATURE_FLAG_SUPPORTS_V190      0x1 /* Supports extended superblock */
1668
1669 /* State flags for sb->flags */
1670 #define SB_FLAG_RESHAPE_ACTIVE          0x1
1671 #define SB_FLAG_RESHAPE_BACKWARDS       0x2
1672
1673 /*
1674  * This structure is never routinely used by userspace, unlike md superblocks.
1675  * Devices with this superblock should only ever be accessed via device-mapper.
1676  */
1677 #define DM_RAID_MAGIC 0x64526D44
1678 struct dm_raid_superblock {
1679         __le32 magic;           /* "DmRd" */
1680         __le32 compat_features; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1681
1682         __le32 num_devices;     /* Number of devices in this raid set. (Max 64) */
1683         __le32 array_position;  /* The position of this drive in the raid set */
1684
1685         __le64 events;          /* Incremented by md when superblock updated */
1686         __le64 failed_devices;  /* Pre 1.9.0 part of bit field of devices to */
1687                                 /* indicate failures (see extension below) */
1688
1689         /*
1690          * This offset tracks the progress of the repair or replacement of
1691          * an individual drive.
1692          */
1693         __le64 disk_recovery_offset;
1694
1695         /*
1696          * This offset tracks the progress of the initial raid set
1697          * synchronisation/parity calculation.
1698          */
1699         __le64 array_resync_offset;
1700
1701         /*
1702          * raid characteristics
1703          */
1704         __le32 level;
1705         __le32 layout;
1706         __le32 stripe_sectors;
1707
1708         /********************************************************************
1709          * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1710          *
1711          * FEATURE_FLAG_SUPPORTS_V190 in the features member indicates that those exist
1712          */
1713
1714         __le32 flags; /* Flags defining array states for reshaping */
1715
1716         /*
1717          * This offset tracks the progress of a raid
1718          * set reshape in order to be able to restart it
1719          */
1720         __le64 reshape_position;
1721
1722         /*
1723          * These define the properties of the array in case of an interrupted reshape
1724          */
1725         __le32 new_level;
1726         __le32 new_layout;
1727         __le32 new_stripe_sectors;
1728         __le32 delta_disks;
1729
1730         __le64 array_sectors; /* Array size in sectors */
1731
1732         /*
1733          * Sector offsets to data on devices (reshaping).
1734          * Needed to support out of place reshaping, thus
1735          * not writing over any stripes whilst converting
1736          * them from old to new layout
1737          */
1738         __le64 data_offset;
1739         __le64 new_data_offset;
1740
1741         __le64 sectors; /* Used device size in sectors */
1742
1743         /*
1744          * Additonal Bit field of devices indicating failures to support
1745          * up to 256 devices with the 1.9.0 on-disk metadata format
1746          */
1747         __le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
1748
1749         __le32 incompat_features;       /* Used to indicate any incompatible features */
1750
1751         /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
1752 } __packed;
1753
1754 /*
1755  * Check for reshape constraints on raid set @rs:
1756  *
1757  * - reshape function non-existent
1758  * - degraded set
1759  * - ongoing recovery
1760  * - ongoing reshape
1761  *
1762  * Returns 0 if none or -EPERM if given constraint
1763  * and error message reference in @errmsg
1764  */
1765 static int rs_check_reshape(struct raid_set *rs)
1766 {
1767         struct mddev *mddev = &rs->md;
1768
1769         if (!mddev->pers || !mddev->pers->check_reshape)
1770                 rs->ti->error = "Reshape not supported";
1771         else if (mddev->degraded)
1772                 rs->ti->error = "Can't reshape degraded raid set";
1773         else if (rs_is_recovering(rs))
1774                 rs->ti->error = "Convert request on recovering raid set prohibited";
1775         else if (mddev->reshape_position && rs_is_reshaping(rs))
1776                 rs->ti->error = "raid set already reshaping!";
1777         else if (!(rs_is_raid10(rs) || rs_is_raid456(rs)))
1778                 rs->ti->error = "Reshaping only supported for raid4/5/6/10";
1779         else
1780                 return 0;
1781
1782         return -EPERM;
1783 }
1784
1785 static int read_disk_sb(struct md_rdev *rdev, int size)
1786 {
1787         BUG_ON(!rdev->sb_page);
1788
1789         if (rdev->sb_loaded)
1790                 return 0;
1791
1792         if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) {
1793                 DMERR("Failed to read superblock of device at position %d",
1794                       rdev->raid_disk);
1795                 md_error(rdev->mddev, rdev);
1796                 return -EINVAL;
1797         }
1798
1799         rdev->sb_loaded = 1;
1800
1801         return 0;
1802 }
1803
1804 static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
1805 {
1806         failed_devices[0] = le64_to_cpu(sb->failed_devices);
1807         memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
1808
1809         if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
1810                 int i = ARRAY_SIZE(sb->extended_failed_devices);
1811
1812                 while (i--)
1813                         failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
1814         }
1815 }
1816
1817 static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
1818 {
1819         int i = ARRAY_SIZE(sb->extended_failed_devices);
1820
1821         sb->failed_devices = cpu_to_le64(failed_devices[0]);
1822         while (i--)
1823                 sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
1824 }
1825
1826 /*
1827  * Synchronize the superblock members with the raid set properties
1828  *
1829  * All superblock data is little endian.
1830  */
1831 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
1832 {
1833         bool update_failed_devices = false;
1834         unsigned int i;
1835         uint64_t failed_devices[DISKS_ARRAY_ELEMS];
1836         struct dm_raid_superblock *sb;
1837         struct raid_set *rs = container_of(mddev, struct raid_set, md);
1838
1839         /* No metadata device, no superblock */
1840         if (!rdev->meta_bdev)
1841                 return;
1842
1843         BUG_ON(!rdev->sb_page);
1844
1845         sb = page_address(rdev->sb_page);
1846
1847         sb_retrieve_failed_devices(sb, failed_devices);
1848
1849         for (i = 0; i < rs->raid_disks; i++)
1850                 if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
1851                         update_failed_devices = true;
1852                         set_bit(i, (void *) failed_devices);
1853                 }
1854
1855         if (update_failed_devices)
1856                 sb_update_failed_devices(sb, failed_devices);
1857
1858         sb->magic = cpu_to_le32(DM_RAID_MAGIC);
1859         sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
1860
1861         sb->num_devices = cpu_to_le32(mddev->raid_disks);
1862         sb->array_position = cpu_to_le32(rdev->raid_disk);
1863
1864         sb->events = cpu_to_le64(mddev->events);
1865
1866         sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
1867         sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
1868
1869         sb->level = cpu_to_le32(mddev->level);
1870         sb->layout = cpu_to_le32(mddev->layout);
1871         sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
1872
1873         sb->new_level = cpu_to_le32(mddev->new_level);
1874         sb->new_layout = cpu_to_le32(mddev->new_layout);
1875         sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
1876
1877         sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1878
1879         smp_rmb(); /* Make sure we access most recent reshape position */
1880         sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1881         if (le64_to_cpu(sb->reshape_position) != MaxSector) {
1882                 /* Flag ongoing reshape */
1883                 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
1884
1885                 if (mddev->delta_disks < 0 || mddev->reshape_backwards)
1886                         sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
1887         } else {
1888                 /* Clear reshape flags */
1889                 sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
1890         }
1891
1892         sb->array_sectors = cpu_to_le64(mddev->array_sectors);
1893         sb->data_offset = cpu_to_le64(rdev->data_offset);
1894         sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
1895         sb->sectors = cpu_to_le64(rdev->sectors);
1896
1897         /* Zero out the rest of the payload after the size of the superblock */
1898         memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
1899 }
1900
1901 /*
1902  * super_load
1903  *
1904  * This function creates a superblock if one is not found on the device
1905  * and will decide which superblock to use if there's a choice.
1906  *
1907  * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
1908  */
1909 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
1910 {
1911         int r;
1912         struct dm_raid_superblock *sb;
1913         struct dm_raid_superblock *refsb;
1914         uint64_t events_sb, events_refsb;
1915
1916         rdev->sb_start = 0;
1917         rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
1918         if (rdev->sb_size < sizeof(*sb) || rdev->sb_size > PAGE_SIZE) {
1919                 DMERR("superblock size of a logical block is no longer valid");
1920                 return -EINVAL;
1921         }
1922
1923         r = read_disk_sb(rdev, rdev->sb_size);
1924         if (r)
1925                 return r;
1926
1927         sb = page_address(rdev->sb_page);
1928
1929         /*
1930          * Two cases that we want to write new superblocks and rebuild:
1931          * 1) New device (no matching magic number)
1932          * 2) Device specified for rebuild (!In_sync w/ offset == 0)
1933          */
1934         if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
1935             (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
1936                 super_sync(rdev->mddev, rdev);
1937
1938                 set_bit(FirstUse, &rdev->flags);
1939                 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
1940
1941                 /* Force writing of superblocks to disk */
1942                 set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags);
1943
1944                 /* Any superblock is better than none, choose that if given */
1945                 return refdev ? 0 : 1;
1946         }
1947
1948         if (!refdev)
1949                 return 1;
1950
1951         events_sb = le64_to_cpu(sb->events);
1952
1953         refsb = page_address(refdev->sb_page);
1954         events_refsb = le64_to_cpu(refsb->events);
1955
1956         return (events_sb > events_refsb) ? 1 : 0;
1957 }
1958
1959 static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
1960 {
1961         int role;
1962         unsigned int d;
1963         struct mddev *mddev = &rs->md;
1964         uint64_t events_sb;
1965         uint64_t failed_devices[DISKS_ARRAY_ELEMS];
1966         struct dm_raid_superblock *sb;
1967         uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
1968         struct md_rdev *r;
1969         struct dm_raid_superblock *sb2;
1970
1971         sb = page_address(rdev->sb_page);
1972         events_sb = le64_to_cpu(sb->events);
1973
1974         /*
1975          * Initialise to 1 if this is a new superblock.
1976          */
1977         mddev->events = events_sb ? : 1;
1978
1979         mddev->reshape_position = MaxSector;
1980
1981         /*
1982          * Reshaping is supported, e.g. reshape_position is valid
1983          * in superblock and superblock content is authoritative.
1984          */
1985         if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
1986                 /* Superblock is authoritative wrt given raid set layout! */
1987                 mddev->raid_disks = le32_to_cpu(sb->num_devices);
1988                 mddev->level = le32_to_cpu(sb->level);
1989                 mddev->layout = le32_to_cpu(sb->layout);
1990                 mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
1991                 mddev->new_level = le32_to_cpu(sb->new_level);
1992                 mddev->new_layout = le32_to_cpu(sb->new_layout);
1993                 mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
1994                 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1995                 mddev->array_sectors = le64_to_cpu(sb->array_sectors);
1996
1997                 /* raid was reshaping and got interrupted */
1998                 if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
1999                         if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
2000                                 DMERR("Reshape requested but raid set is still reshaping");
2001                                 return -EINVAL;
2002                         }
2003
2004                         if (mddev->delta_disks < 0 ||
2005                             (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
2006                                 mddev->reshape_backwards = 1;
2007                         else
2008                                 mddev->reshape_backwards = 0;
2009
2010                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2011                         rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2012                 }
2013
2014         } else {
2015                 /*
2016                  * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2017                  */
2018                 if (le32_to_cpu(sb->level) != mddev->level) {
2019                         DMERR("Reshaping/takeover raid sets not yet supported. (raid level/stripes/size change)");
2020                         return -EINVAL;
2021                 }
2022                 if (le32_to_cpu(sb->layout) != mddev->layout) {
2023                         DMERR("Reshaping raid sets not yet supported. (raid layout change)");
2024                         DMERR("  0x%X vs 0x%X", le32_to_cpu(sb->layout), mddev->layout);
2025                         DMERR("  Old layout: %s w/ %d copies",
2026                               raid10_md_layout_to_format(le32_to_cpu(sb->layout)),
2027                               raid10_md_layout_to_copies(le32_to_cpu(sb->layout)));
2028                         DMERR("  New layout: %s w/ %d copies",
2029                               raid10_md_layout_to_format(mddev->layout),
2030                               raid10_md_layout_to_copies(mddev->layout));
2031                         return -EINVAL;
2032                 }
2033                 if (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors) {
2034                         DMERR("Reshaping raid sets not yet supported. (stripe sectors change)");
2035                         return -EINVAL;
2036                 }
2037
2038                 /* We can only change the number of devices in raid1 with old (i.e. pre 1.0.7) metadata */
2039                 if (!rt_is_raid1(rs->raid_type) &&
2040                     (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) {
2041                         DMERR("Reshaping raid sets not yet supported. (device count change from %u to %u)",
2042                               sb->num_devices, mddev->raid_disks);
2043                         return -EINVAL;
2044                 }
2045
2046                 /* Table line is checked vs. authoritative superblock */
2047                 rs_set_new(rs);
2048         }
2049
2050         if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2051                 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2052
2053         /*
2054          * During load, we set FirstUse if a new superblock was written.
2055          * There are two reasons we might not have a superblock:
2056          * 1) The raid set is brand new - in which case, all of the
2057          *    devices must have their In_sync bit set.  Also,
2058          *    recovery_cp must be 0, unless forced.
2059          * 2) This is a new device being added to an old raid set
2060          *    and the new device needs to be rebuilt - in which
2061          *    case the In_sync bit will /not/ be set and
2062          *    recovery_cp must be MaxSector.
2063          * 3) This is/are a new device(s) being added to an old
2064          *    raid set during takeover to a higher raid level
2065          *    to provide capacity for redundancy or during reshape
2066          *    to add capacity to grow the raid set.
2067          */
2068         d = 0;
2069         rdev_for_each(r, mddev) {
2070                 if (test_bit(FirstUse, &r->flags))
2071                         new_devs++;
2072
2073                 if (!test_bit(In_sync, &r->flags)) {
2074                         DMINFO("Device %d specified for rebuild; clearing superblock",
2075                                 r->raid_disk);
2076                         rebuilds++;
2077
2078                         if (test_bit(FirstUse, &r->flags))
2079                                 rebuild_and_new++;
2080                 }
2081
2082                 d++;
2083         }
2084
2085         if (new_devs == rs->raid_disks || !rebuilds) {
2086                 /* Replace a broken device */
2087                 if (new_devs == 1 && !rs->delta_disks)
2088                         ;
2089                 if (new_devs == rs->raid_disks) {
2090                         DMINFO("Superblocks created for new raid set");
2091                         set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2092                         mddev->recovery_cp = 0;
2093                 } else if (new_devs != rebuilds &&
2094                            new_devs != rs->delta_disks) {
2095                         DMERR("New device injected into existing raid set without "
2096                               "'delta_disks' or 'rebuild' parameter specified");
2097                         return -EINVAL;
2098                 }
2099         } else if (new_devs && new_devs != rebuilds) {
2100                 DMERR("%u 'rebuild' devices cannot be injected into"
2101                       " a raid set with %u other first-time devices",
2102                       rebuilds, new_devs);
2103                 return -EINVAL;
2104         } else if (rebuilds) {
2105                 if (rebuild_and_new && rebuilds != rebuild_and_new) {
2106                         DMERR("new device%s provided without 'rebuild'",
2107                               new_devs > 1 ? "s" : "");
2108                         return -EINVAL;
2109                 } else if (rs_is_recovering(rs)) {
2110                         DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2111                               (unsigned long long) mddev->recovery_cp);
2112                         return -EINVAL;
2113                 } else if (rs_is_reshaping(rs)) {
2114                         DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2115                               (unsigned long long) mddev->reshape_position);
2116                         return -EINVAL;
2117                 }
2118         }
2119
2120         /*
2121          * Now we set the Faulty bit for those devices that are
2122          * recorded in the superblock as failed.
2123          */
2124         sb_retrieve_failed_devices(sb, failed_devices);
2125         rdev_for_each(r, mddev) {
2126                 if (!r->sb_page)
2127                         continue;
2128                 sb2 = page_address(r->sb_page);
2129                 sb2->failed_devices = 0;
2130                 memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2131
2132                 /*
2133                  * Check for any device re-ordering.
2134                  */
2135                 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2136                         role = le32_to_cpu(sb2->array_position);
2137                         if (role < 0)
2138                                 continue;
2139
2140                         if (role != r->raid_disk) {
2141                                 if (__is_raid10_near(mddev->layout)) {
2142                                         if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2143                                             rs->raid_disks % rs->raid10_copies) {
2144                                                 rs->ti->error =
2145                                                         "Cannot change raid10 near set to odd # of devices!";
2146                                                 return -EINVAL;
2147                                         }
2148
2149                                         sb2->array_position = cpu_to_le32(r->raid_disk);
2150
2151                                 } else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2152                                            !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2153                                            !rt_is_raid1(rs->raid_type)) {
2154                                         rs->ti->error = "Cannot change device positions in raid set";
2155                                         return -EINVAL;
2156                                 }
2157
2158                                 DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2159                         }
2160
2161                         /*
2162                          * Partial recovery is performed on
2163                          * returning failed devices.
2164                          */
2165                         if (test_bit(role, (void *) failed_devices))
2166                                 set_bit(Faulty, &r->flags);
2167                 }
2168         }
2169
2170         return 0;
2171 }
2172
2173 static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2174 {
2175         struct mddev *mddev = &rs->md;
2176         struct dm_raid_superblock *sb;
2177
2178         if (rs_is_raid0(rs) || !rdev->sb_page)
2179                 return 0;
2180
2181         sb = page_address(rdev->sb_page);
2182
2183         /*
2184          * If mddev->events is not set, we know we have not yet initialized
2185          * the array.
2186          */
2187         if (!mddev->events && super_init_validation(rs, rdev))
2188                 return -EINVAL;
2189
2190         if (le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2191                 rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2192                 return -EINVAL;
2193         }
2194
2195         if (sb->incompat_features) {
2196                 rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2197                 return -EINVAL;
2198         }
2199
2200         /* Enable bitmap creation for RAID levels != 0 */
2201         mddev->bitmap_info.offset = rt_is_raid0(rs->raid_type) ? 0 : to_sector(4096);
2202         rdev->mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2203
2204         if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2205                 /* Retrieve device size stored in superblock to be prepared for shrink */
2206                 rdev->sectors = le64_to_cpu(sb->sectors);
2207                 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2208                 if (rdev->recovery_offset == MaxSector)
2209                         set_bit(In_sync, &rdev->flags);
2210                 /*
2211                  * If no reshape in progress -> we're recovering single
2212                  * disk(s) and have to set the device(s) to out-of-sync
2213                  */
2214                 else if (!rs_is_reshaping(rs))
2215                         clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2216         }
2217
2218         /*
2219          * If a device comes back, set it as not In_sync and no longer faulty.
2220          */
2221         if (test_and_clear_bit(Faulty, &rdev->flags)) {
2222                 rdev->recovery_offset = 0;
2223                 clear_bit(In_sync, &rdev->flags);
2224                 rdev->saved_raid_disk = rdev->raid_disk;
2225         }
2226
2227         /* Reshape support -> restore repective data offsets */
2228         rdev->data_offset = le64_to_cpu(sb->data_offset);
2229         rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2230
2231         return 0;
2232 }
2233
2234 /*
2235  * Analyse superblocks and select the freshest.
2236  */
2237 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2238 {
2239         int r;
2240         struct raid_dev *dev;
2241         struct md_rdev *rdev, *tmp, *freshest;
2242         struct mddev *mddev = &rs->md;
2243
2244         freshest = NULL;
2245         rdev_for_each_safe(rdev, tmp, mddev) {
2246                 /*
2247                  * Skipping super_load due to CTR_FLAG_SYNC will cause
2248                  * the array to undergo initialization again as
2249                  * though it were new.  This is the intended effect
2250                  * of the "sync" directive.
2251                  *
2252                  * When reshaping capability is added, we must ensure
2253                  * that the "sync" directive is disallowed during the
2254                  * reshape.
2255                  */
2256                 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2257                         continue;
2258
2259                 if (!rdev->meta_bdev)
2260                         continue;
2261
2262                 r = super_load(rdev, freshest);
2263
2264                 switch (r) {
2265                 case 1:
2266                         freshest = rdev;
2267                         break;
2268                 case 0:
2269                         break;
2270                 default:
2271                         dev = container_of(rdev, struct raid_dev, rdev);
2272                         if (dev->meta_dev)
2273                                 dm_put_device(ti, dev->meta_dev);
2274
2275                         dev->meta_dev = NULL;
2276                         rdev->meta_bdev = NULL;
2277
2278                         if (rdev->sb_page)
2279                                 put_page(rdev->sb_page);
2280
2281                         rdev->sb_page = NULL;
2282
2283                         rdev->sb_loaded = 0;
2284
2285                         /*
2286                          * We might be able to salvage the data device
2287                          * even though the meta device has failed.  For
2288                          * now, we behave as though '- -' had been
2289                          * set for this device in the table.
2290                          */
2291                         if (dev->data_dev)
2292                                 dm_put_device(ti, dev->data_dev);
2293
2294                         dev->data_dev = NULL;
2295                         rdev->bdev = NULL;
2296
2297                         list_del(&rdev->same_set);
2298                 }
2299         }
2300
2301         if (!freshest)
2302                 return 0;
2303
2304         if (validate_raid_redundancy(rs)) {
2305                 rs->ti->error = "Insufficient redundancy to activate array";
2306                 return -EINVAL;
2307         }
2308
2309         /*
2310          * Validation of the freshest device provides the source of
2311          * validation for the remaining devices.
2312          */
2313         rs->ti->error = "Unable to assemble array: Invalid superblocks";
2314         if (super_validate(rs, freshest))
2315                 return -EINVAL;
2316
2317         rdev_for_each(rdev, mddev)
2318                 if ((rdev != freshest) && super_validate(rs, rdev))
2319                         return -EINVAL;
2320         return 0;
2321 }
2322
2323 /*
2324  * Adjust data_offset and new_data_offset on all disk members of @rs
2325  * for out of place reshaping if requested by contructor
2326  *
2327  * We need free space at the beginning of each raid disk for forward
2328  * and at the end for backward reshapes which userspace has to provide
2329  * via remapping/reordering of space.
2330  */
2331 static int rs_adjust_data_offsets(struct raid_set *rs)
2332 {
2333         sector_t data_offset = 0, new_data_offset = 0;
2334         struct md_rdev *rdev;
2335
2336         /* Constructor did not request data offset change */
2337         if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2338                 if (!rs_is_reshapable(rs))
2339                         goto out;
2340
2341                 return 0;
2342         }
2343
2344         /* HM FIXME: get InSync raid_dev? */
2345         rdev = &rs->dev[0].rdev;
2346
2347         if (rs->delta_disks < 0) {
2348                 /*
2349                  * Removing disks (reshaping backwards):
2350                  *
2351                  * - before reshape: data is at offset 0 and free space
2352                  *                   is at end of each component LV
2353                  *
2354                  * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2355                  */
2356                 data_offset = 0;
2357                 new_data_offset = rs->data_offset;
2358
2359         } else if (rs->delta_disks > 0) {
2360                 /*
2361                  * Adding disks (reshaping forwards):
2362                  *
2363                  * - before reshape: data is at offset rs->data_offset != 0 and
2364                  *                   free space is at begin of each component LV
2365                  *
2366                  * - after reshape: data is at offset 0 on each component LV
2367                  */
2368                 data_offset = rs->data_offset;
2369                 new_data_offset = 0;
2370
2371         } else {
2372                 /*
2373                  * User space passes in 0 for data offset after having removed reshape space
2374                  *
2375                  * - or - (data offset != 0)
2376                  *
2377                  * Changing RAID layout or chunk size -> toggle offsets
2378                  *
2379                  * - before reshape: data is at offset rs->data_offset 0 and
2380                  *                   free space is at end of each component LV
2381                  *                   -or-
2382                  *                   data is at offset rs->data_offset != 0 and
2383                  *                   free space is at begin of each component LV
2384                  *
2385                  * - after reshape: data is at offset 0 if i was at offset != 0
2386                  *                  of at offset != 0 if it was at offset 0
2387                  *                  on each component LV
2388                  *
2389                  */
2390                 data_offset = rs->data_offset ? rdev->data_offset : 0;
2391                 new_data_offset = data_offset ? 0 : rs->data_offset;
2392                 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2393         }
2394
2395         /*
2396          * Make sure we got a minimum amount of free sectors per device
2397          */
2398         if (rs->data_offset &&
2399             to_sector(i_size_read(rdev->bdev->bd_inode)) - rdev->sectors < MIN_FREE_RESHAPE_SPACE) {
2400                 rs->ti->error = data_offset ? "No space for forward reshape" :
2401                                               "No space for backward reshape";
2402                 return -ENOSPC;
2403         }
2404 out:
2405         /* Adjust data offsets on all rdevs */
2406         rdev_for_each(rdev, &rs->md) {
2407                 rdev->data_offset = data_offset;
2408                 rdev->new_data_offset = new_data_offset;
2409         }
2410
2411         return 0;
2412 }
2413
2414 /* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2415 static void __reorder_raid_disk_indexes(struct raid_set *rs)
2416 {
2417         int i = 0;
2418         struct md_rdev *rdev;
2419
2420         rdev_for_each(rdev, &rs->md) {
2421                 rdev->raid_disk = i++;
2422                 rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2423         }
2424 }
2425
2426 /*
2427  * Setup @rs for takeover by a different raid level
2428  */
2429 static int rs_setup_takeover(struct raid_set *rs)
2430 {
2431         struct mddev *mddev = &rs->md;
2432         struct md_rdev *rdev;
2433         unsigned int d = mddev->raid_disks = rs->raid_disks;
2434         sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2435
2436         if (rt_is_raid10(rs->raid_type)) {
2437                 if (mddev->level == 0) {
2438                         /* Userpace reordered disks -> adjust raid_disk indexes */
2439                         __reorder_raid_disk_indexes(rs);
2440
2441                         /* raid0 -> raid10_far layout */
2442                         mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2443                                                                    rs->raid10_copies);
2444                 } else if (mddev->level == 1)
2445                         /* raid1 -> raid10_near layout */
2446                         mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2447                                                                    rs->raid_disks);
2448                  else
2449                         return -EINVAL;
2450
2451         }
2452
2453         clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2454         mddev->recovery_cp = MaxSector;
2455
2456         while (d--) {
2457                 rdev = &rs->dev[d].rdev;
2458
2459                 if (test_bit(d, (void *) rs->rebuild_disks)) {
2460                         clear_bit(In_sync, &rdev->flags);
2461                         clear_bit(Faulty, &rdev->flags);
2462                         mddev->recovery_cp = rdev->recovery_offset = 0;
2463                         /* Bitmap has to be created when we do an "up" takeover */
2464                         set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2465                 }
2466
2467                 rdev->new_data_offset = new_data_offset;
2468         }
2469
2470         return 0;
2471 }
2472
2473 /*
2474  *
2475  * - change raid layout
2476  * - change chunk size
2477  * - add disks
2478  * - remove disks
2479  */
2480 static int rs_setup_reshape(struct raid_set *rs)
2481 {
2482         int r = 0;
2483         unsigned int cur_raid_devs, d;
2484         struct mddev *mddev = &rs->md;
2485         struct md_rdev *rdev;
2486
2487         mddev->delta_disks = rs->delta_disks;
2488         cur_raid_devs = mddev->raid_disks;
2489
2490         /* Ignore impossible layout change whilst adding/removing disks */
2491         if (mddev->delta_disks &&
2492             mddev->layout != mddev->new_layout) {
2493                 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2494                 mddev->new_layout = mddev->layout;
2495         }
2496
2497         /*
2498          * Adjust array size:
2499          *
2500          * - in case of adding disks, array size has
2501          *   to grow after the disk adding reshape,
2502          *   which'll hapen in the event handler;
2503          *   reshape will happen forward, so space has to
2504          *   be available at the beginning of each disk
2505          *
2506          * - in case of removing disks, array size
2507          *   has to shrink before starting the reshape,
2508          *   which'll happen here;
2509          *   reshape will happen backward, so space has to
2510          *   be available at the end of each disk
2511          *
2512          * - data_offset and new_data_offset are
2513          *   adjusted for aforementioned out of place
2514          *   reshaping based on userspace passing in
2515          *   the "data_offset <sectors>" key/value
2516          *   pair via the constructor
2517          */
2518
2519         /* Add disk(s) */
2520         if (rs->delta_disks > 0) {
2521                 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2522                 for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2523                         rdev = &rs->dev[d].rdev;
2524                         clear_bit(In_sync, &rdev->flags);
2525
2526                         /*
2527                          * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2528                          * by md, which'll store that erroneously in the superblock on reshape
2529                          */
2530                         rdev->saved_raid_disk = -1;
2531                         rdev->raid_disk = d;
2532
2533                         rdev->sectors = mddev->dev_sectors;
2534                         rdev->recovery_offset = MaxSector;
2535                 }
2536
2537                 mddev->reshape_backwards = 0; /* adding disks -> forward reshape */
2538
2539         /* Remove disk(s) */
2540         } else if (rs->delta_disks < 0) {
2541                 r = rs_set_dev_and_array_sectors(rs, true);
2542                 mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2543
2544         /* Change layout and/or chunk size */
2545         } else {
2546                 /*
2547                  * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2548                  *
2549                  * keeping number of disks and do layout change ->
2550                  *
2551                  * toggle reshape_backward depending on data_offset:
2552                  *
2553                  * - free space upfront -> reshape forward
2554                  *
2555                  * - free space at the end -> reshape backward
2556                  *
2557                  *
2558                  * This utilizes free reshape space avoiding the need
2559                  * for userspace to move (parts of) LV segments in
2560                  * case of layout/chunksize change  (for disk
2561                  * adding/removing reshape space has to be at
2562                  * the proper address (see above with delta_disks):
2563                  *
2564                  * add disk(s)   -> begin
2565                  * remove disk(s)-> end
2566                  */
2567                 mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2568         }
2569
2570         return r;
2571 }
2572
2573 /*
2574  * Enable/disable discard support on RAID set depending on
2575  * RAID level and discard properties of underlying RAID members.
2576  */
2577 static void configure_discard_support(struct raid_set *rs)
2578 {
2579         int i;
2580         bool raid456;
2581         struct dm_target *ti = rs->ti;
2582
2583         /* Assume discards not supported until after checks below. */
2584         ti->discards_supported = false;
2585
2586         /* RAID level 4,5,6 require discard_zeroes_data for data integrity! */
2587         raid456 = (rs->md.level == 4 || rs->md.level == 5 || rs->md.level == 6);
2588
2589         for (i = 0; i < rs->md.raid_disks; i++) {
2590                 struct request_queue *q;
2591
2592                 if (!rs->dev[i].rdev.bdev)
2593                         continue;
2594
2595                 q = bdev_get_queue(rs->dev[i].rdev.bdev);
2596                 if (!q || !blk_queue_discard(q))
2597                         return;
2598
2599                 if (raid456) {
2600                         if (!q->limits.discard_zeroes_data)
2601                                 return;
2602                         if (!devices_handle_discard_safely) {
2603                                 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2604                                 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2605                                 return;
2606                         }
2607                 }
2608         }
2609
2610         /* All RAID members properly support discards */
2611         ti->discards_supported = true;
2612
2613         /*
2614          * RAID1 and RAID10 personalities require bio splitting,
2615          * RAID0/4/5/6 don't and process large discard bios properly.
2616          */
2617         ti->split_discard_bios = !!(rs->md.level == 1 || rs->md.level == 10);
2618         ti->num_discard_bios = 1;
2619 }
2620
2621 /*
2622  * Construct a RAID0/1/10/4/5/6 mapping:
2623  * Args:
2624  *      <raid_type> <#raid_params> <raid_params>{0,}    \
2625  *      <#raid_devs> [<meta_dev1> <dev1>]{1,}
2626  *
2627  * <raid_params> varies by <raid_type>.  See 'parse_raid_params' for
2628  * details on possible <raid_params>.
2629  *
2630  * Userspace is free to initialize the metadata devices, hence the superblocks to
2631  * enforce recreation based on the passed in table parameters.
2632  *
2633  */
2634 static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv)
2635 {
2636         int r;
2637         struct raid_type *rt;
2638         unsigned num_raid_params, num_raid_devs;
2639         struct raid_set *rs = NULL;
2640         const char *arg;
2641         struct rs_layout rs_layout;
2642         struct dm_arg_set as = { argc, argv }, as_nrd;
2643         struct dm_arg _args[] = {
2644                 { 0, as.argc, "Cannot understand number of raid parameters" },
2645                 { 1, 254, "Cannot understand number of raid devices parameters" }
2646         };
2647
2648         /* Must have <raid_type> */
2649         arg = dm_shift_arg(&as);
2650         if (!arg) {
2651                 ti->error = "No arguments";
2652                 return -EINVAL;
2653         }
2654
2655         rt = get_raid_type(arg);
2656         if (!rt) {
2657                 ti->error = "Unrecognised raid_type";
2658                 return -EINVAL;
2659         }
2660
2661         /* Must have <#raid_params> */
2662         if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
2663                 return -EINVAL;
2664
2665         /* number of raid device tupples <meta_dev data_dev> */
2666         as_nrd = as;
2667         dm_consume_args(&as_nrd, num_raid_params);
2668         _args[1].max = (as_nrd.argc - 1) / 2;
2669         if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
2670                 return -EINVAL;
2671
2672         if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
2673                 ti->error = "Invalid number of supplied raid devices";
2674                 return -EINVAL;
2675         }
2676
2677         rs = raid_set_alloc(ti, rt, num_raid_devs);
2678         if (IS_ERR(rs))
2679                 return PTR_ERR(rs);
2680
2681         r = parse_raid_params(rs, &as, num_raid_params);
2682         if (r)
2683                 goto bad;
2684
2685         r = parse_dev_params(rs, &as);
2686         if (r)
2687                 goto bad;
2688
2689         rs->md.sync_super = super_sync;
2690
2691         r = rs_set_dev_and_array_sectors(rs, false);
2692         if (r)
2693                 return r;
2694
2695         /*
2696          * Backup any new raid set level, layout, ...
2697          * requested to be able to compare to superblock
2698          * members for conversion decisions.
2699          */
2700         rs_config_backup(rs, &rs_layout);
2701
2702         r = analyse_superblocks(ti, rs);
2703         if (r)
2704                 goto bad;
2705
2706         INIT_WORK(&rs->md.event_work, do_table_event);
2707         ti->private = rs;
2708         ti->num_flush_bios = 1;
2709
2710         /* Restore any requested new layout for conversion decision */
2711         rs_config_restore(rs, &rs_layout);
2712
2713         if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
2714                 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2715                 rs_set_new(rs);
2716         } else if (rs_is_reshaping(rs))
2717                 ; /* skip rs setup */
2718         else if (rs_takeover_requested(rs)) {
2719                 if (rs_is_reshaping(rs)) {
2720                         ti->error = "Can't takeover a reshaping raid set";
2721                         return -EPERM;
2722                 }
2723
2724                 /*
2725                  * If a takeover is needed, just set the level to
2726                  * the new requested one and allow the raid set to run.
2727                  */
2728                 r = rs_check_takeover(rs);
2729                 if (r)
2730                         return r;
2731
2732                 r = rs_setup_takeover(rs);
2733                 if (r)
2734                         return r;
2735
2736                 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2737                 set_bit(RT_FLAG_KEEP_RS_FROZEN, &rs->runtime_flags);
2738                 rs_set_new(rs);
2739         } else if (rs_reshape_requested(rs)) {
2740                 if (rs_is_reshaping(rs)) {
2741                         ti->error = "raid set already reshaping!";
2742                         return -EPERM;
2743                 }
2744
2745                 if (rs_is_raid10(rs)) {
2746                         if (rs->raid_disks != rs->md.raid_disks &&
2747                             __is_raid10_near(rs->md.layout) &&
2748                             rs->raid10_copies &&
2749                             rs->raid10_copies != __raid10_near_copies(rs->md.layout)) {
2750                                 /*
2751                                  * raid disk have to be multiple of data copies to allow this conversion,
2752                                  *
2753                                  * This is actually not a reshape it is a
2754                                  * rebuild of any additional mirrors per group
2755                                  */
2756                                 if (rs->raid_disks % rs->raid10_copies) {
2757                                         ti->error = "Can't reshape raid10 mirror groups";
2758                                         return -EINVAL;
2759                                 }
2760
2761                                 /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2762                                 __reorder_raid_disk_indexes(rs);
2763                                 rs->md.layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2764                                                                            rs->raid10_copies);
2765                                 rs->md.new_layout = rs->md.layout;
2766
2767                         } else
2768                                 set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2769
2770                 } else if (rs_is_raid456(rs))
2771                         set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2772
2773                 /*
2774                  * HM FIXME: process raid1 via delta_disks as well?
2775                  *           Would cause allocations in raid1->check_reshape
2776                  *           though, thus more issues with potential failures
2777                  */
2778                 else if (rs_is_raid1(rs)) {
2779                         set_bit(RT_FLAG_KEEP_RS_FROZEN, &rs->runtime_flags);
2780                         rs->md.raid_disks = rs->raid_disks;
2781                 }
2782
2783                 if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
2784                         set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2785                         set_bit(RT_FLAG_KEEP_RS_FROZEN, &rs->runtime_flags);
2786                 }
2787
2788                 if (rs->md.raid_disks < rs->raid_disks)
2789                         set_bit(MD_ARRAY_FIRST_USE, &rs->md.flags);
2790
2791                 rs_set_cur(rs);
2792         } else
2793                 rs_set_cur(rs);
2794
2795         /* If constructor requested it, change data and new_data offsets */
2796         r = rs_adjust_data_offsets(rs);
2797         if (r)
2798                 return r;
2799
2800         /* Start raid set read-only and assumed clean to change in raid_resume() */
2801         rs->md.ro = 1;
2802         rs->md.in_sync = 1;
2803         set_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
2804
2805         /* Has to be held on running the array */
2806         mddev_lock_nointr(&rs->md);
2807         r = md_run(&rs->md);
2808         rs->md.in_sync = 0; /* Assume already marked dirty */
2809
2810         if (r) {
2811                 ti->error = "Failed to run raid array";
2812                 mddev_unlock(&rs->md);
2813                 goto bad;
2814         }
2815
2816         rs->callbacks.congested_fn = raid_is_congested;
2817         dm_table_add_target_callbacks(ti->table, &rs->callbacks);
2818
2819         mddev_suspend(&rs->md);
2820
2821         /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
2822         if (rs_is_raid456(rs)) {
2823                 r = rs_set_raid456_stripe_cache(rs);
2824                 if (r)
2825                         goto bad_stripe_cache;
2826         }
2827
2828         /* Now do an early reshape check */
2829         if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
2830                 r = rs_check_reshape(rs);
2831                 if (r)
2832                         return r;
2833
2834                 /* Restore new, ctr requested layout to perform check */
2835                 rs_config_restore(rs, &rs_layout);
2836
2837                 r = rs->md.pers->check_reshape(&rs->md);
2838                 if (r) {
2839                         ti->error = "Reshape check failed";
2840                         goto bad_check_reshape;
2841                 }
2842         }
2843
2844         mddev_unlock(&rs->md);
2845         return 0;
2846
2847 bad_stripe_cache:
2848 bad_check_reshape:
2849         md_stop(&rs->md);
2850 bad:
2851         raid_set_free(rs);
2852
2853         return r;
2854 }
2855
2856 static void raid_dtr(struct dm_target *ti)
2857 {
2858         struct raid_set *rs = ti->private;
2859
2860         list_del_init(&rs->callbacks.list);
2861         md_stop(&rs->md);
2862         raid_set_free(rs);
2863 }
2864
2865 static int raid_map(struct dm_target *ti, struct bio *bio)
2866 {
2867         struct raid_set *rs = ti->private;
2868         struct mddev *mddev = &rs->md;
2869
2870         /*
2871          * If we're reshaping to add disk(s)), ti->len and
2872          * mddev->array_sectors will differ during the process
2873          * (ti->len > mddev->array_sectors), so we have to requeue
2874          * bios with addresses > mddev->array_sectors here or
2875          * or there will occur accesses past EOD of the component
2876          * data images thus erroring the raid set.
2877          */
2878         if (unlikely(bio_end_sector(bio) > mddev->array_sectors))
2879                 return DM_MAPIO_REQUEUE;
2880
2881         mddev->pers->make_request(mddev, bio);
2882
2883         return DM_MAPIO_SUBMITTED;
2884 }
2885
2886 /* Return string describing the current sync action of @mddev */
2887 static const char *decipher_sync_action(struct mddev *mddev)
2888 {
2889         if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
2890                 return "frozen";
2891
2892         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2893             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2894                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2895                         return "reshape";
2896
2897                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2898                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2899                                 return "resync";
2900                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2901                                 return "check";
2902                         return "repair";
2903                 }
2904
2905                 if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
2906                         return "recover";
2907         }
2908
2909         return "idle";
2910 }
2911
2912 /*
2913  * Return status string @rdev
2914  *
2915  * Status characters:
2916  *
2917  *  'D' = Dead/Failed device
2918  *  'a' = Alive but not in-sync
2919  *  'A' = Alive and in-sync
2920  */
2921 static const char *__raid_dev_status(struct md_rdev *rdev, bool array_in_sync)
2922 {
2923         if (test_bit(Faulty, &rdev->flags))
2924                 return "D";
2925         else if (!array_in_sync || !test_bit(In_sync, &rdev->flags))
2926                 return "a";
2927         else
2928                 return "A";
2929 }
2930
2931 /* Helper to return resync/reshape progress for @rs and @array_in_sync */
2932 static sector_t rs_get_progress(struct raid_set *rs,
2933                                 sector_t resync_max_sectors, bool *array_in_sync)
2934 {
2935         sector_t r, recovery_cp, curr_resync_completed;
2936         struct mddev *mddev = &rs->md;
2937
2938         curr_resync_completed = mddev->curr_resync_completed ?: mddev->recovery_cp;
2939         recovery_cp = mddev->recovery_cp;
2940         *array_in_sync = false;
2941
2942         if (rs_is_raid0(rs)) {
2943                 r = resync_max_sectors;
2944                 *array_in_sync = true;
2945
2946         } else {
2947                 r = mddev->reshape_position;
2948
2949                 /* Reshape is relative to the array size */
2950                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
2951                     r != MaxSector) {
2952                         if (r == MaxSector) {
2953                                 *array_in_sync = true;
2954                                 r = resync_max_sectors;
2955                         } else {
2956                                 /* Got to reverse on backward reshape */
2957                                 if (mddev->reshape_backwards)
2958                                         r = mddev->array_sectors - r;
2959
2960                                 /* Devide by # of data stripes */
2961                                 sector_div(r, mddev_data_stripes(rs));
2962                         }
2963
2964                 /* Sync is relative to the component device size */
2965                 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2966                         r = curr_resync_completed;
2967                 else
2968                         r = recovery_cp;
2969
2970                 if (r == MaxSector) {
2971                         /*
2972                          * Sync complete.
2973                          */
2974                         *array_in_sync = true;
2975                         r = resync_max_sectors;
2976                 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2977                         /*
2978                          * If "check" or "repair" is occurring, the raid set has
2979                          * undergone an initial sync and the health characters
2980                          * should not be 'a' anymore.
2981                          */
2982                         *array_in_sync = true;
2983                 } else {
2984                         struct md_rdev *rdev;
2985
2986                         /*
2987                          * The raid set may be doing an initial sync, or it may
2988                          * be rebuilding individual components.  If all the
2989                          * devices are In_sync, then it is the raid set that is
2990                          * being initialized.
2991                          */
2992                         rdev_for_each(rdev, mddev)
2993                                 if (!test_bit(In_sync, &rdev->flags))
2994                                         *array_in_sync = true;
2995 #if 0
2996                         r = 0; /* HM FIXME: TESTME: https://bugzilla.redhat.com/show_bug.cgi?id=1210637 ? */
2997 #endif
2998                 }
2999         }
3000
3001         return r;
3002 }
3003
3004 /* Helper to return @dev name or "-" if !@dev */
3005 static const char *__get_dev_name(struct dm_dev *dev)
3006 {
3007         return dev ? dev->name : "-";
3008 }
3009
3010 static void raid_status(struct dm_target *ti, status_type_t type,
3011                         unsigned int status_flags, char *result, unsigned int maxlen)
3012 {
3013         struct raid_set *rs = ti->private;
3014         struct mddev *mddev = &rs->md;
3015         struct r5conf *conf = mddev->private;
3016         int max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3017         bool array_in_sync;
3018         unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3019         unsigned int sz = 0;
3020         unsigned int write_mostly_params = 0;
3021         sector_t progress, resync_max_sectors, resync_mismatches;
3022         const char *sync_action;
3023         struct raid_type *rt;
3024         struct md_rdev *rdev;
3025
3026         switch (type) {
3027         case STATUSTYPE_INFO:
3028                 /* *Should* always succeed */
3029                 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3030                 if (!rt)
3031                         return;
3032
3033                 DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3034
3035                 /* Access most recent mddev properties for status output */
3036                 smp_rmb();
3037                 /* Get sensible max sectors even if raid set not yet started */
3038                 resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3039                                       mddev->resync_max_sectors : mddev->dev_sectors;
3040                 progress = rs_get_progress(rs, resync_max_sectors, &array_in_sync);
3041                 resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
3042                                     atomic64_read(&mddev->resync_mismatches) : 0;
3043                 sync_action = decipher_sync_action(&rs->md);
3044
3045                 /* HM FIXME: do we want another state char for raid0? It shows 'D' or 'A' now */
3046                 rdev_for_each(rdev, mddev)
3047                         DMEMIT(__raid_dev_status(rdev, array_in_sync));
3048
3049                 /*
3050                  * In-sync/Reshape ratio:
3051                  *  The in-sync ratio shows the progress of:
3052                  *   - Initializing the raid set
3053                  *   - Rebuilding a subset of devices of the raid set
3054                  *  The user can distinguish between the two by referring
3055                  *  to the status characters.
3056                  *
3057                  *  The reshape ratio shows the progress of
3058                  *  changing the raid layout or the number of
3059                  *  disks of a raid set
3060                  */
3061                 DMEMIT(" %llu/%llu", (unsigned long long) progress,
3062                                      (unsigned long long) resync_max_sectors);
3063
3064                 /*
3065                  * v1.5.0+:
3066                  *
3067                  * Sync action:
3068                  *   See Documentation/device-mapper/dm-raid.txt for
3069                  *   information on each of these states.
3070                  */
3071                 DMEMIT(" %s", sync_action);
3072
3073                 /*
3074                  * v1.5.0+:
3075                  *
3076                  * resync_mismatches/mismatch_cnt
3077                  *   This field shows the number of discrepancies found when
3078                  *   performing a "check" of the raid set.
3079                  */
3080                 DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3081
3082                 /*
3083                  * v1.9.0+:
3084                  *
3085                  * data_offset (needed for out of space reshaping)
3086                  *   This field shows the data offset into the data
3087                  *   image LV where the first stripes data starts.
3088                  *
3089                  * We keep data_offset equal on all raid disks of the set,
3090                  * so retrieving it from the first raid disk is sufficient.
3091                  */
3092                 DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3093                 break;
3094
3095         case STATUSTYPE_TABLE:
3096                 /* Report the table line string you would use to construct this raid set */
3097
3098                 /* Calculate raid parameter count */
3099                 rdev_for_each(rdev, mddev)
3100                         if (test_bit(WriteMostly, &rdev->flags))
3101                                 write_mostly_params += 2;
3102                 raid_param_cnt += memweight(rs->rebuild_disks,
3103                                             DISKS_ARRAY_ELEMS * sizeof(*rs->rebuild_disks)) * 2 +
3104                                   write_mostly_params +
3105                                   hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3106                                   hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2;
3107                 /* Emit table line */
3108                 DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3109                 if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3110                         DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3111                                          raid10_md_layout_to_format(mddev->layout));
3112                 if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3113                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3114                                          raid10_md_layout_to_copies(mddev->layout));
3115                 if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3116                         DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3117                 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3118                         DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3119                 if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3120                         DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3121                                            (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3122                 if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3123                         DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3124                                            (unsigned long long) rs->data_offset);
3125                 if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3126                         DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3127                                           mddev->bitmap_info.daemon_sleep);
3128                 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3129                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3130                                          mddev->delta_disks);
3131                 if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3132                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3133                                          max_nr_stripes);
3134                 rdev_for_each(rdev, mddev)
3135                         if (test_bit(rdev->raid_disk, (void *) rs->rebuild_disks))
3136                                 DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD),
3137                                                  rdev->raid_disk);
3138                 rdev_for_each(rdev, mddev)
3139                         if (test_bit(WriteMostly, &rdev->flags))
3140                                 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3141                                                  rdev->raid_disk);
3142                 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3143                         DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3144                                           mddev->bitmap_info.max_write_behind);
3145                 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3146                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3147                                          mddev->sync_speed_max);
3148                 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3149                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3150                                          mddev->sync_speed_min);
3151                 DMEMIT(" %d", rs->raid_disks);
3152                 rdev_for_each(rdev, mddev) {
3153                         struct raid_dev *rd = container_of(rdev, struct raid_dev, rdev);
3154
3155                         DMEMIT(" %s %s", __get_dev_name(rd->meta_dev),
3156                                          __get_dev_name(rd->data_dev));
3157                 }
3158         }
3159 }
3160
3161 static int raid_message(struct dm_target *ti, unsigned argc, char **argv)
3162 {
3163         struct raid_set *rs = ti->private;
3164         struct mddev *mddev = &rs->md;
3165
3166         if (!mddev->pers || !mddev->pers->sync_request)
3167                 return -EINVAL;
3168
3169         if (!strcasecmp(argv[0], "frozen"))
3170                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3171         else
3172                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3173
3174         if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
3175                 if (mddev->sync_thread) {
3176                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3177                         md_reap_sync_thread(mddev);
3178                 }
3179         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3180                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3181                 return -EBUSY;
3182         else if (!strcasecmp(argv[0], "resync"))
3183                 ; /* MD_RECOVERY_NEEDED set below */
3184         else if (!strcasecmp(argv[0], "recover"))
3185                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3186         else {
3187                 if (!strcasecmp(argv[0], "check"))
3188                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3189                 else if (!!strcasecmp(argv[0], "repair"))
3190                         return -EINVAL;
3191                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3192                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3193         }
3194         if (mddev->ro == 2) {
3195                 /* A write to sync_action is enough to justify
3196                  * canceling read-auto mode
3197                  */
3198                 mddev->ro = 0;
3199                 if (!mddev->suspended && mddev->sync_thread)
3200                         md_wakeup_thread(mddev->sync_thread);
3201         }
3202         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3203         if (!mddev->suspended && mddev->thread)
3204                 md_wakeup_thread(mddev->thread);
3205
3206         return 0;
3207 }
3208
3209 static int raid_iterate_devices(struct dm_target *ti,
3210                                 iterate_devices_callout_fn fn, void *data)
3211 {
3212         struct raid_set *rs = ti->private;
3213         unsigned i;
3214         int r = 0;
3215
3216         for (i = 0; !r && i < rs->md.raid_disks; i++)
3217                 if (rs->dev[i].data_dev)
3218                         r = fn(ti,
3219                                  rs->dev[i].data_dev,
3220                                  0, /* No offset on data devs */
3221                                  rs->md.dev_sectors,
3222                                  data);
3223
3224         return r;
3225 }
3226
3227 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3228 {
3229         struct raid_set *rs = ti->private;
3230         unsigned chunk_size = rs->md.chunk_sectors << 9;
3231         struct r5conf *conf = rs->md.private;
3232
3233         blk_limits_io_min(limits, chunk_size);
3234         blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded));
3235 }
3236
3237 static void raid_presuspend(struct dm_target *ti)
3238 {
3239         struct raid_set *rs = ti->private;
3240
3241         md_stop_writes(&rs->md);
3242 }
3243
3244 static void raid_postsuspend(struct dm_target *ti)
3245 {
3246         struct raid_set *rs = ti->private;
3247
3248         if (test_and_clear_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
3249                 if (!rs->md.suspended)
3250                         mddev_suspend(&rs->md);
3251                 rs->md.ro = 1;
3252         }
3253 }
3254
3255 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3256 {
3257         int i;
3258         uint64_t failed_devices, cleared_failed_devices = 0;
3259         unsigned long flags;
3260         struct dm_raid_superblock *sb;
3261         struct md_rdev *r;
3262
3263         for (i = 0; i < rs->md.raid_disks; i++) {
3264                 r = &rs->dev[i].rdev;
3265                 if (test_bit(Faulty, &r->flags) && r->sb_page &&
3266                     sync_page_io(r, 0, r->sb_size, r->sb_page,
3267                                  REQ_OP_READ, 0, true)) {
3268                         DMINFO("Faulty %s device #%d has readable super block."
3269                                "  Attempting to revive it.",
3270                                rs->raid_type->name, i);
3271
3272                         /*
3273                          * Faulty bit may be set, but sometimes the array can
3274                          * be suspended before the personalities can respond
3275                          * by removing the device from the array (i.e. calling
3276                          * 'hot_remove_disk').  If they haven't yet removed
3277                          * the failed device, its 'raid_disk' number will be
3278                          * '>= 0' - meaning we must call this function
3279                          * ourselves.
3280                          */
3281                         if ((r->raid_disk >= 0) &&
3282                             (r->mddev->pers->hot_remove_disk(r->mddev, r) != 0))
3283                                 /* Failed to revive this device, try next */
3284                                 continue;
3285
3286                         r->raid_disk = i;
3287                         r->saved_raid_disk = i;
3288                         flags = r->flags;
3289                         clear_bit(Faulty, &r->flags);
3290                         clear_bit(WriteErrorSeen, &r->flags);
3291                         clear_bit(In_sync, &r->flags);
3292                         if (r->mddev->pers->hot_add_disk(r->mddev, r)) {
3293                                 r->raid_disk = -1;
3294                                 r->saved_raid_disk = -1;
3295                                 r->flags = flags;
3296                         } else {
3297                                 r->recovery_offset = 0;
3298                                 cleared_failed_devices |= 1 << i;
3299                         }
3300                 }
3301         }
3302         if (cleared_failed_devices) {
3303                 rdev_for_each(r, &rs->md) {
3304                         sb = page_address(r->sb_page);
3305                         failed_devices = le64_to_cpu(sb->failed_devices);
3306                         failed_devices &= ~cleared_failed_devices;
3307                         sb->failed_devices = cpu_to_le64(failed_devices);
3308                 }
3309         }
3310 }
3311
3312 static int __load_dirty_region_bitmap(struct raid_set *rs)
3313 {
3314         int r = 0;
3315
3316         /* Try loading the bitmap unless "raid0", which does not have one */
3317         if (!rs_is_raid0(rs) &&
3318             !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3319                 r = bitmap_load(&rs->md);
3320                 if (r)
3321                         DMERR("Failed to load bitmap");
3322         }
3323
3324         return r;
3325 }
3326
3327 /* Enforce updating all superblocks */
3328 static void rs_update_sbs(struct raid_set *rs)
3329 {
3330         struct mddev *mddev = &rs->md;
3331         int ro = mddev->ro;
3332
3333         set_bit(MD_CHANGE_DEVS, &mddev->flags);
3334         mddev->ro = 0;
3335         md_update_sb(mddev, 1);
3336         mddev->ro = ro;
3337 }
3338
3339 /*
3340  * Reshape changes raid algorithm of @rs to new one within personality
3341  * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3342  * disks from a raid set thus growing/shrinking it or resizes the set
3343  *
3344  * Call mddev_lock_nointr() before!
3345  */
3346 static int rs_start_reshape(struct raid_set *rs)
3347 {
3348         int r;
3349         struct mddev *mddev = &rs->md;
3350         struct md_personality *pers = mddev->pers;
3351
3352         r = rs_setup_reshape(rs);
3353         if (r)
3354                 return r;
3355
3356         /* Need to be resumed to be able to start reshape, recovery is frozen until raid_resume() though */
3357         if (mddev->suspended)
3358                 mddev_resume(mddev);
3359
3360         /*
3361          * Check any reshape constraints enforced by the personalility
3362          *
3363          * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3364          */
3365         r = pers->check_reshape(mddev);
3366         if (r) {
3367                 rs->ti->error = "pers->check_reshape() failed";
3368                 return r;
3369         }
3370
3371         /*
3372          * Personality may not provide start reshape method in which
3373          * case check_reshape above has already covered everything
3374          */
3375         if (pers->start_reshape) {
3376                 r = pers->start_reshape(mddev);
3377                 if (r) {
3378                         rs->ti->error = "pers->start_reshape() failed";
3379                         return r;
3380                 }
3381         }
3382
3383         /* Suspend because a resume will happen in raid_resume() */
3384         if (!mddev->suspended)
3385                 mddev_suspend(mddev);
3386
3387         /*
3388          * Now reshape got set up, update superblocks to
3389          * reflect the fact so that a table reload will
3390          * access proper superblock content in the ctr.
3391          */
3392         rs_update_sbs(rs);
3393
3394         return 0;
3395 }
3396
3397 static int raid_preresume(struct dm_target *ti)
3398 {
3399         int r;
3400         struct raid_set *rs = ti->private;
3401         struct mddev *mddev = &rs->md;
3402
3403         /* This is a resume after a suspend of the set -> it's already started */
3404         if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
3405                 return 0;
3406
3407         /*
3408          * The superblocks need to be updated on disk if the
3409          * array is new or new devices got added (thus zeroed
3410          * out by userspace) or __load_dirty_region_bitmap
3411          * will overwrite them in core with old data or fail.
3412          */
3413         if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
3414                 rs_update_sbs(rs);
3415
3416         /*
3417          * Disable/enable discard support on raid set after any
3418          * conversion, because devices can have been added
3419          */
3420         configure_discard_support(rs);
3421
3422         /* Load the bitmap from disk unless raid0 */
3423         r = __load_dirty_region_bitmap(rs);
3424         if (r)
3425                 return r;
3426
3427         /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) */
3428         if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) &&
3429             mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)) {
3430                 r = bitmap_resize(mddev->bitmap, mddev->dev_sectors,
3431                                   to_bytes(rs->requested_bitmap_chunk_sectors), 0);
3432                 if (r)
3433                         DMERR("Failed to resize bitmap");
3434         }
3435
3436         /* Check for any resize/reshape on @rs and adjust/initiate */
3437         /* Be prepared for mddev_resume() in raid_resume() */
3438         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3439         if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
3440                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3441                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3442                 mddev->resync_min = mddev->recovery_cp;
3443         }
3444
3445         rs_set_capacity(rs);
3446
3447         /* Check for any reshape request and region size change unless new raid set */
3448         if (test_and_clear_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3449                 /* Initiate a reshape. */
3450                 mddev_lock_nointr(mddev);
3451                 r = rs_start_reshape(rs);
3452                 mddev_unlock(mddev);
3453                 if (r)
3454                         DMWARN("Failed to check/start reshape, continuing without change");
3455                 r = 0;
3456         }
3457
3458         return r;
3459 }
3460
3461 static void raid_resume(struct dm_target *ti)
3462 {
3463         struct raid_set *rs = ti->private;
3464         struct mddev *mddev = &rs->md;
3465
3466         if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
3467                 /*
3468                  * A secondary resume while the device is active.
3469                  * Take this opportunity to check whether any failed
3470                  * devices are reachable again.
3471                  */
3472                 attempt_restore_of_faulty_devices(rs);
3473         } else {
3474                 mddev->ro = 0;
3475                 mddev->in_sync = 0;
3476
3477                 /*
3478                  * When passing in flags to the ctr, we expect userspace
3479                  * to reset them because they made it to the superblocks
3480                  * and reload the mapping anyway.
3481                  *
3482                  * -> only unfreeze recovery in case of a table reload or
3483                  *    we'll have a bogus recovery/reshape position
3484                  *    retrieved from the superblock by the ctr because
3485                  *    the ongoing recovery/reshape will change it after read.
3486                  */
3487                 if (!test_bit(RT_FLAG_KEEP_RS_FROZEN, &rs->runtime_flags))
3488                         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3489
3490                 if (mddev->suspended)
3491                         mddev_resume(mddev);
3492         }
3493 }
3494
3495 static struct target_type raid_target = {
3496         .name = "raid",
3497         .version = {1, 9, 0},
3498         .module = THIS_MODULE,
3499         .ctr = raid_ctr,
3500         .dtr = raid_dtr,
3501         .map = raid_map,
3502         .status = raid_status,
3503         .message = raid_message,
3504         .iterate_devices = raid_iterate_devices,
3505         .io_hints = raid_io_hints,
3506         .presuspend = raid_presuspend,
3507         .postsuspend = raid_postsuspend,
3508         .preresume = raid_preresume,
3509         .resume = raid_resume,
3510 };
3511
3512 static int __init dm_raid_init(void)
3513 {
3514         DMINFO("Loading target version %u.%u.%u",
3515                raid_target.version[0],
3516                raid_target.version[1],
3517                raid_target.version[2]);
3518         return dm_register_target(&raid_target);
3519 }
3520
3521 static void __exit dm_raid_exit(void)
3522 {
3523         dm_unregister_target(&raid_target);
3524 }
3525
3526 module_init(dm_raid_init);
3527 module_exit(dm_raid_exit);
3528
3529 module_param(devices_handle_discard_safely, bool, 0644);
3530 MODULE_PARM_DESC(devices_handle_discard_safely,
3531                  "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
3532
3533 MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
3534 MODULE_ALIAS("dm-raid0");
3535 MODULE_ALIAS("dm-raid1");
3536 MODULE_ALIAS("dm-raid10");
3537 MODULE_ALIAS("dm-raid4");
3538 MODULE_ALIAS("dm-raid5");
3539 MODULE_ALIAS("dm-raid6");
3540 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
3541 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
3542 MODULE_LICENSE("GPL");