drivers/md/dm-raid.c

   1 /*
   2  * Copyright (C) 2010-2011 Neil Brown
   3  * Copyright (C) 2010-2017 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 /*
  28  * Minimum journal space 4 MiB in sectors.
  29  */
  30 #define MIN_RAID456_JOURNAL_SPACE (4*2048)
  31
  32 static bool devices_handle_discard_safely = false;
  33
  34 /*
  35  * The following flags are used by dm-raid.c to set up the array state.
  36  * They must be cleared before md_run is called.
  37  */
  38 #define FirstUse 10             /* rdev flag */
  39
  40 struct raid_dev {
  41         /*
  42          * Two DM devices, one to hold metadata and one to hold the
  43          * actual data/parity.  The reason for this is to not confuse
  44          * ti->len and give more flexibility in altering size and
  45          * characteristics.
  46          *
  47          * While it is possible for this device to be associated
  48          * with a different physical device than the data_dev, it
  49          * is intended for it to be the same.
  50          *    |--------- Physical Device ---------|
  51          *    |- meta_dev -|------ data_dev ------|
  52          */
  53         struct dm_dev *meta_dev;
  54         struct dm_dev *data_dev;
  55         struct md_rdev rdev;
  56 };
  57
  58 /*
  59  * Bits for establishing rs->ctr_flags
  60  *
  61  * 1 = no flag value
  62  * 2 = flag with value
  63  */
  64 #define __CTR_FLAG_SYNC                 0  /* 1 */ /* Not with raid0! */
  65 #define __CTR_FLAG_NOSYNC               1  /* 1 */ /* Not with raid0! */
  66 #define __CTR_FLAG_REBUILD              2  /* 2 */ /* Not with raid0! */
  67 #define __CTR_FLAG_DAEMON_SLEEP         3  /* 2 */ /* Not with raid0! */
  68 #define __CTR_FLAG_MIN_RECOVERY_RATE    4  /* 2 */ /* Not with raid0! */
  69 #define __CTR_FLAG_MAX_RECOVERY_RATE    5  /* 2 */ /* Not with raid0! */
  70 #define __CTR_FLAG_MAX_WRITE_BEHIND     6  /* 2 */ /* Only with raid1! */
  71 #define __CTR_FLAG_WRITE_MOSTLY         7  /* 2 */ /* Only with raid1! */
  72 #define __CTR_FLAG_STRIPE_CACHE         8  /* 2 */ /* Only with raid4/5/6! */
  73 #define __CTR_FLAG_REGION_SIZE          9  /* 2 */ /* Not with raid0! */
  74 #define __CTR_FLAG_RAID10_COPIES        10 /* 2 */ /* Only with raid10 */
  75 #define __CTR_FLAG_RAID10_FORMAT        11 /* 2 */ /* Only with raid10 */
  76 /* New for v1.9.0 */
  77 #define __CTR_FLAG_DELTA_DISKS          12 /* 2 */ /* Only with reshapable raid1/4/5/6/10! */
  78 #define __CTR_FLAG_DATA_OFFSET          13 /* 2 */ /* Only with reshapable raid4/5/6/10! */
  79 #define __CTR_FLAG_RAID10_USE_NEAR_SETS 14 /* 2 */ /* Only with raid10! */
  80
  81 /* New for v1.10.0 */
  82 #define __CTR_FLAG_JOURNAL_DEV          15 /* 2 */ /* Only with raid4/5/6 (journal device)! */
  83
  84 /* New for v1.11.1 */
  85 #define __CTR_FLAG_JOURNAL_MODE         16 /* 2 */ /* Only with raid4/5/6 (journal mode)! */
  86
  87 /*
  88  * Flags for rs->ctr_flags field.
  89  */
  90 #define CTR_FLAG_SYNC                   (1 << __CTR_FLAG_SYNC)
  91 #define CTR_FLAG_NOSYNC                 (1 << __CTR_FLAG_NOSYNC)
  92 #define CTR_FLAG_REBUILD                (1 << __CTR_FLAG_REBUILD)
  93 #define CTR_FLAG_DAEMON_SLEEP           (1 << __CTR_FLAG_DAEMON_SLEEP)
  94 #define CTR_FLAG_MIN_RECOVERY_RATE      (1 << __CTR_FLAG_MIN_RECOVERY_RATE)
  95 #define CTR_FLAG_MAX_RECOVERY_RATE      (1 << __CTR_FLAG_MAX_RECOVERY_RATE)
  96 #define CTR_FLAG_MAX_WRITE_BEHIND       (1 << __CTR_FLAG_MAX_WRITE_BEHIND)
  97 #define CTR_FLAG_WRITE_MOSTLY           (1 << __CTR_FLAG_WRITE_MOSTLY)
  98 #define CTR_FLAG_STRIPE_CACHE           (1 << __CTR_FLAG_STRIPE_CACHE)
  99 #define CTR_FLAG_REGION_SIZE            (1 << __CTR_FLAG_REGION_SIZE)
 100 #define CTR_FLAG_RAID10_COPIES          (1 << __CTR_FLAG_RAID10_COPIES)
 101 #define CTR_FLAG_RAID10_FORMAT          (1 << __CTR_FLAG_RAID10_FORMAT)
 102 #define CTR_FLAG_DELTA_DISKS            (1 << __CTR_FLAG_DELTA_DISKS)
 103 #define CTR_FLAG_DATA_OFFSET            (1 << __CTR_FLAG_DATA_OFFSET)
 104 #define CTR_FLAG_RAID10_USE_NEAR_SETS   (1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
 105 #define CTR_FLAG_JOURNAL_DEV            (1 << __CTR_FLAG_JOURNAL_DEV)
 106 #define CTR_FLAG_JOURNAL_MODE           (1 << __CTR_FLAG_JOURNAL_MODE)
 107
 108 #define RESUME_STAY_FROZEN_FLAGS (CTR_FLAG_DELTA_DISKS | CTR_FLAG_DATA_OFFSET)
 109
 110 /*
 111  * Definitions of various constructor flags to
 112  * be used in checks of valid / invalid flags
 113  * per raid level.
 114  */
 115 /* Define all any sync flags */
 116 #define CTR_FLAGS_ANY_SYNC              (CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
 117
 118 /* Define flags for options without argument (e.g. 'nosync') */
 119 #define CTR_FLAG_OPTIONS_NO_ARGS        (CTR_FLAGS_ANY_SYNC | \
 120                                          CTR_FLAG_RAID10_USE_NEAR_SETS)
 121
 122 /* Define flags for options with one argument (e.g. 'delta_disks +2') */
 123 #define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
 124                                   CTR_FLAG_WRITE_MOSTLY | \
 125                                   CTR_FLAG_DAEMON_SLEEP | \
 126                                   CTR_FLAG_MIN_RECOVERY_RATE | \
 127                                   CTR_FLAG_MAX_RECOVERY_RATE | \
 128                                   CTR_FLAG_MAX_WRITE_BEHIND | \
 129                                   CTR_FLAG_STRIPE_CACHE | \
 130                                   CTR_FLAG_REGION_SIZE | \
 131                                   CTR_FLAG_RAID10_COPIES | \
 132                                   CTR_FLAG_RAID10_FORMAT | \
 133                                   CTR_FLAG_DELTA_DISKS | \
 134                                   CTR_FLAG_DATA_OFFSET)
 135
 136 /* Valid options definitions per raid level... */
 137
 138 /* "raid0" does only accept data offset */
 139 #define RAID0_VALID_FLAGS       (CTR_FLAG_DATA_OFFSET)
 140
 141 /* "raid1" does not accept stripe cache, data offset, delta_disks or any raid10 options */
 142 #define RAID1_VALID_FLAGS       (CTR_FLAGS_ANY_SYNC | \
 143                                  CTR_FLAG_REBUILD | \
 144                                  CTR_FLAG_WRITE_MOSTLY | \
 145                                  CTR_FLAG_DAEMON_SLEEP | \
 146                                  CTR_FLAG_MIN_RECOVERY_RATE | \
 147                                  CTR_FLAG_MAX_RECOVERY_RATE | \
 148                                  CTR_FLAG_MAX_WRITE_BEHIND | \
 149                                  CTR_FLAG_REGION_SIZE | \
 150                                  CTR_FLAG_DELTA_DISKS | \
 151                                  CTR_FLAG_DATA_OFFSET)
 152
 153 /* "raid10" does not accept any raid1 or stripe cache options */
 154 #define RAID10_VALID_FLAGS      (CTR_FLAGS_ANY_SYNC | \
 155                                  CTR_FLAG_REBUILD | \
 156                                  CTR_FLAG_DAEMON_SLEEP | \
 157                                  CTR_FLAG_MIN_RECOVERY_RATE | \
 158                                  CTR_FLAG_MAX_RECOVERY_RATE | \
 159                                  CTR_FLAG_REGION_SIZE | \
 160                                  CTR_FLAG_RAID10_COPIES | \
 161                                  CTR_FLAG_RAID10_FORMAT | \
 162                                  CTR_FLAG_DELTA_DISKS | \
 163                                  CTR_FLAG_DATA_OFFSET | \
 164                                  CTR_FLAG_RAID10_USE_NEAR_SETS)
 165
 166 /*
 167  * "raid4/5/6" do not accept any raid1 or raid10 specific options
 168  *
 169  * "raid6" does not accept "nosync", because it is not guaranteed
 170  * that both parity and q-syndrome are being written properly with
 171  * any writes
 172  */
 173 #define RAID45_VALID_FLAGS      (CTR_FLAGS_ANY_SYNC | \
 174                                  CTR_FLAG_REBUILD | \
 175                                  CTR_FLAG_DAEMON_SLEEP | \
 176                                  CTR_FLAG_MIN_RECOVERY_RATE | \
 177                                  CTR_FLAG_MAX_RECOVERY_RATE | \
 178                                  CTR_FLAG_STRIPE_CACHE | \
 179                                  CTR_FLAG_REGION_SIZE | \
 180                                  CTR_FLAG_DELTA_DISKS | \
 181                                  CTR_FLAG_DATA_OFFSET | \
 182                                  CTR_FLAG_JOURNAL_DEV | \
 183                                  CTR_FLAG_JOURNAL_MODE)
 184
 185 #define RAID6_VALID_FLAGS       (CTR_FLAG_SYNC | \
 186                                  CTR_FLAG_REBUILD | \
 187                                  CTR_FLAG_DAEMON_SLEEP | \
 188                                  CTR_FLAG_MIN_RECOVERY_RATE | \
 189                                  CTR_FLAG_MAX_RECOVERY_RATE | \
 190                                  CTR_FLAG_STRIPE_CACHE | \
 191                                  CTR_FLAG_REGION_SIZE | \
 192                                  CTR_FLAG_DELTA_DISKS | \
 193                                  CTR_FLAG_DATA_OFFSET | \
 194                                  CTR_FLAG_JOURNAL_DEV | \
 195                                  CTR_FLAG_JOURNAL_MODE)
 196 /* ...valid options definitions per raid level */
 197
 198 /*
 199  * Flags for rs->runtime_flags field
 200  * (RT_FLAG prefix meaning "runtime flag")
 201  *
 202  * These are all internal and used to define runtime state,
 203  * e.g. to prevent another resume from preresume processing
 204  * the raid set all over again.
 205  */
 206 #define RT_FLAG_RS_PRERESUMED           0
 207 #define RT_FLAG_RS_RESUMED              1
 208 #define RT_FLAG_RS_BITMAP_LOADED        2
 209 #define RT_FLAG_UPDATE_SBS              3
 210 #define RT_FLAG_RESHAPE_RS              4
 211 #define RT_FLAG_RS_SUSPENDED            5
 212
 213 /* Array elements of 64 bit needed for rebuild/failed disk bits */
 214 #define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
 215
 216 /*
 217  * raid set level, layout and chunk sectors backup/restore
 218  */
 219 struct rs_layout {
 220         int new_level;
 221         int new_layout;
 222         int new_chunk_sectors;
 223 };
 224
 225 struct raid_set {
 226         struct dm_target *ti;
 227
 228         uint32_t bitmap_loaded;
 229         uint32_t stripe_cache_entries;
 230         unsigned long ctr_flags;
 231         unsigned long runtime_flags;
 232
 233         uint64_t rebuild_disks[DISKS_ARRAY_ELEMS];
 234
 235         int raid_disks;
 236         int delta_disks;
 237         int data_offset;
 238         int raid10_copies;
 239         int requested_bitmap_chunk_sectors;
 240
 241         struct mddev md;
 242         struct raid_type *raid_type;
 243         struct dm_target_callbacks callbacks;
 244
 245         /* Optional raid4/5/6 journal device */
 246         struct journal_dev {
 247                 struct dm_dev *dev;
 248                 struct md_rdev rdev;
 249                 int mode;
 250         } journal_dev;
 251
 252         struct raid_dev dev[0];
 253 };
 254
 255 static void rs_config_backup(struct raid_set *rs, struct rs_layout *l)
 256 {
 257         struct mddev *mddev = &rs->md;
 258
 259         l->new_level = mddev->new_level;
 260         l->new_layout = mddev->new_layout;
 261         l->new_chunk_sectors = mddev->new_chunk_sectors;
 262 }
 263
 264 static void rs_config_restore(struct raid_set *rs, struct rs_layout *l)
 265 {
 266         struct mddev *mddev = &rs->md;
 267
 268         mddev->new_level = l->new_level;
 269         mddev->new_layout = l->new_layout;
 270         mddev->new_chunk_sectors = l->new_chunk_sectors;
 271 }
 272
 273 /* raid10 algorithms (i.e. formats) */
 274 #define ALGORITHM_RAID10_DEFAULT        0
 275 #define ALGORITHM_RAID10_NEAR           1
 276 #define ALGORITHM_RAID10_OFFSET         2
 277 #define ALGORITHM_RAID10_FAR            3
 278
 279 /* Supported raid types and properties. */
 280 static struct raid_type {
 281         const char *name;               /* RAID algorithm. */
 282         const char *descr;              /* Descriptor text for logging. */
 283         const unsigned int parity_devs; /* # of parity devices. */
 284         const unsigned int minimal_devs;/* minimal # of devices in set. */
 285         const unsigned int level;       /* RAID level. */
 286         const unsigned int algorithm;   /* RAID algorithm. */
 287 } raid_types[] = {
 288         {"raid0",         "raid0 (striping)",                       0, 2, 0,  0 /* NONE */},
 289         {"raid1",         "raid1 (mirroring)",                      0, 2, 1,  0 /* NONE */},
 290         {"raid10_far",    "raid10 far (striped mirrors)",           0, 2, 10, ALGORITHM_RAID10_FAR},
 291         {"raid10_offset", "raid10 offset (striped mirrors)",        0, 2, 10, ALGORITHM_RAID10_OFFSET},
 292         {"raid10_near",   "raid10 near (striped mirrors)",          0, 2, 10, ALGORITHM_RAID10_NEAR},
 293         {"raid10",        "raid10 (striped mirrors)",               0, 2, 10, ALGORITHM_RAID10_DEFAULT},
 294         {"raid4",         "raid4 (dedicated first parity disk)",    1, 2, 5,  ALGORITHM_PARITY_0}, /* raid4 layout = raid5_0 */
 295         {"raid5_n",       "raid5 (dedicated last parity disk)",     1, 2, 5,  ALGORITHM_PARITY_N},
 296         {"raid5_ls",      "raid5 (left symmetric)",                 1, 2, 5,  ALGORITHM_LEFT_SYMMETRIC},
 297         {"raid5_rs",      "raid5 (right symmetric)",                1, 2, 5,  ALGORITHM_RIGHT_SYMMETRIC},
 298         {"raid5_la",      "raid5 (left asymmetric)",                1, 2, 5,  ALGORITHM_LEFT_ASYMMETRIC},
 299         {"raid5_ra",      "raid5 (right asymmetric)",               1, 2, 5,  ALGORITHM_RIGHT_ASYMMETRIC},
 300         {"raid6_zr",      "raid6 (zero restart)",                   2, 4, 6,  ALGORITHM_ROTATING_ZERO_RESTART},
 301         {"raid6_nr",      "raid6 (N restart)",                      2, 4, 6,  ALGORITHM_ROTATING_N_RESTART},
 302         {"raid6_nc",      "raid6 (N continue)",                     2, 4, 6,  ALGORITHM_ROTATING_N_CONTINUE},
 303         {"raid6_n_6",     "raid6 (dedicated parity/Q n/6)",         2, 4, 6,  ALGORITHM_PARITY_N_6},
 304         {"raid6_ls_6",    "raid6 (left symmetric dedicated Q 6)",   2, 4, 6,  ALGORITHM_LEFT_SYMMETRIC_6},
 305         {"raid6_rs_6",    "raid6 (right symmetric dedicated Q 6)",  2, 4, 6,  ALGORITHM_RIGHT_SYMMETRIC_6},
 306         {"raid6_la_6",    "raid6 (left asymmetric dedicated Q 6)",  2, 4, 6,  ALGORITHM_LEFT_ASYMMETRIC_6},
 307         {"raid6_ra_6",    "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6,  ALGORITHM_RIGHT_ASYMMETRIC_6}
 308 };
 309
 310 /* True, if @v is in inclusive range [@min, @max] */
 311 static bool __within_range(long v, long min, long max)
 312 {
 313         return v >= min && v <= max;
 314 }
 315
 316 /* All table line arguments are defined here */
 317 static struct arg_name_flag {
 318         const unsigned long flag;
 319         const char *name;
 320 } __arg_name_flags[] = {
 321         { CTR_FLAG_SYNC, "sync"},
 322         { CTR_FLAG_NOSYNC, "nosync"},
 323         { CTR_FLAG_REBUILD, "rebuild"},
 324         { CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
 325         { CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
 326         { CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
 327         { CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
 328         { CTR_FLAG_WRITE_MOSTLY, "write_mostly"},
 329         { CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
 330         { CTR_FLAG_REGION_SIZE, "region_size"},
 331         { CTR_FLAG_RAID10_COPIES, "raid10_copies"},
 332         { CTR_FLAG_RAID10_FORMAT, "raid10_format"},
 333         { CTR_FLAG_DATA_OFFSET, "data_offset"},
 334         { CTR_FLAG_DELTA_DISKS, "delta_disks"},
 335         { CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
 336         { CTR_FLAG_JOURNAL_DEV, "journal_dev" },
 337         { CTR_FLAG_JOURNAL_MODE, "journal_mode" },
 338 };
 339
 340 /* Return argument name string for given @flag */
 341 static const char *dm_raid_arg_name_by_flag(const uint32_t flag)
 342 {
 343         if (hweight32(flag) == 1) {
 344                 struct arg_name_flag *anf = __arg_name_flags + ARRAY_SIZE(__arg_name_flags);
 345
 346                 while (anf-- > __arg_name_flags)
 347                         if (flag & anf->flag)
 348                                 return anf->name;
 349
 350         } else
 351                 DMERR("%s called with more than one flag!", __func__);
 352
 353         return NULL;
 354 }
 355
 356 /* Define correlation of raid456 journal cache modes and dm-raid target line parameters */
 357 static struct {
 358         const int mode;
 359         const char *param;
 360 } _raid456_journal_mode[] = {
 361         { R5C_JOURNAL_MODE_WRITE_THROUGH , "writethrough" },
 362         { R5C_JOURNAL_MODE_WRITE_BACK    , "writeback" }
 363 };
 364
 365 /* Return MD raid4/5/6 journal mode for dm @journal_mode one */
 366 static int dm_raid_journal_mode_to_md(const char *mode)
 367 {
 368         int m = ARRAY_SIZE(_raid456_journal_mode);
 369
 370         while (m--)
 371                 if (!strcasecmp(mode, _raid456_journal_mode[m].param))
 372                         return _raid456_journal_mode[m].mode;
 373
 374         return -EINVAL;
 375 }
 376
 377 /* Return dm-raid raid4/5/6 journal mode string for @mode */
 378 static const char *md_journal_mode_to_dm_raid(const int mode)
 379 {
 380         int m = ARRAY_SIZE(_raid456_journal_mode);
 381
 382         while (m--)
 383                 if (mode == _raid456_journal_mode[m].mode)
 384                         return _raid456_journal_mode[m].param;
 385
 386         return "unknown";
 387 }
 388
 389 /*
 390  * Bool helpers to test for various raid levels of a raid set.
 391  * It's level as reported by the superblock rather than
 392  * the requested raid_type passed to the constructor.
 393  */
 394 /* Return true, if raid set in @rs is raid0 */
 395 static bool rs_is_raid0(struct raid_set *rs)
 396 {
 397         return !rs->md.level;
 398 }
 399
 400 /* Return true, if raid set in @rs is raid1 */
 401 static bool rs_is_raid1(struct raid_set *rs)
 402 {
 403         return rs->md.level == 1;
 404 }
 405
 406 /* Return true, if raid set in @rs is raid10 */
 407 static bool rs_is_raid10(struct raid_set *rs)
 408 {
 409         return rs->md.level == 10;
 410 }
 411
 412 /* Return true, if raid set in @rs is level 6 */
 413 static bool rs_is_raid6(struct raid_set *rs)
 414 {
 415         return rs->md.level == 6;
 416 }
 417
 418 /* Return true, if raid set in @rs is level 4, 5 or 6 */
 419 static bool rs_is_raid456(struct raid_set *rs)
 420 {
 421         return __within_range(rs->md.level, 4, 6);
 422 }
 423
 424 /* Return true, if raid set in @rs is reshapable */
 425 static bool __is_raid10_far(int layout);
 426 static bool rs_is_reshapable(struct raid_set *rs)
 427 {
 428         return rs_is_raid456(rs) ||
 429                (rs_is_raid10(rs) && !__is_raid10_far(rs->md.new_layout));
 430 }
 431
 432 /* Return true, if raid set in @rs is recovering */
 433 static bool rs_is_recovering(struct raid_set *rs)
 434 {
 435         return rs->md.recovery_cp < rs->md.dev_sectors;
 436 }
 437
 438 /* Return true, if raid set in @rs is reshaping */
 439 static bool rs_is_reshaping(struct raid_set *rs)
 440 {
 441         return rs->md.reshape_position != MaxSector;
 442 }
 443
 444 /*
 445  * bool helpers to test for various raid levels of a raid type @rt
 446  */
 447
 448 /* Return true, if raid type in @rt is raid0 */
 449 static bool rt_is_raid0(struct raid_type *rt)
 450 {
 451         return !rt->level;
 452 }
 453
 454 /* Return true, if raid type in @rt is raid1 */
 455 static bool rt_is_raid1(struct raid_type *rt)
 456 {
 457         return rt->level == 1;
 458 }
 459
 460 /* Return true, if raid type in @rt is raid10 */
 461 static bool rt_is_raid10(struct raid_type *rt)
 462 {
 463         return rt->level == 10;
 464 }
 465
 466 /* Return true, if raid type in @rt is raid4/5 */
 467 static bool rt_is_raid45(struct raid_type *rt)
 468 {
 469         return __within_range(rt->level, 4, 5);
 470 }
 471
 472 /* Return true, if raid type in @rt is raid6 */
 473 static bool rt_is_raid6(struct raid_type *rt)
 474 {
 475         return rt->level == 6;
 476 }
 477
 478 /* Return true, if raid type in @rt is raid4/5/6 */
 479 static bool rt_is_raid456(struct raid_type *rt)
 480 {
 481         return __within_range(rt->level, 4, 6);
 482 }
 483 /* END: raid level bools */
 484
 485 /* Return valid ctr flags for the raid level of @rs */
 486 static unsigned long __valid_flags(struct raid_set *rs)
 487 {
 488         if (rt_is_raid0(rs->raid_type))
 489                 return RAID0_VALID_FLAGS;
 490         else if (rt_is_raid1(rs->raid_type))
 491                 return RAID1_VALID_FLAGS;
 492         else if (rt_is_raid10(rs->raid_type))
 493                 return RAID10_VALID_FLAGS;
 494         else if (rt_is_raid45(rs->raid_type))
 495                 return RAID45_VALID_FLAGS;
 496         else if (rt_is_raid6(rs->raid_type))
 497                 return RAID6_VALID_FLAGS;
 498
 499         return 0;
 500 }
 501
 502 /*
 503  * Check for valid flags set on @rs
 504  *
 505  * Has to be called after parsing of the ctr flags!
 506  */
 507 static int rs_check_for_valid_flags(struct raid_set *rs)
 508 {
 509         if (rs->ctr_flags & ~__valid_flags(rs)) {
 510                 rs->ti->error = "Invalid flags combination";
 511                 return -EINVAL;
 512         }
 513
 514         return 0;
 515 }
 516
 517 /* MD raid10 bit definitions and helpers */
 518 #define RAID10_OFFSET                   (1 << 16) /* stripes with data copies area adjacent on devices */
 519 #define RAID10_BROCKEN_USE_FAR_SETS     (1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
 520 #define RAID10_USE_FAR_SETS             (1 << 18) /* Use sets instead of whole stripe rotation */
 521 #define RAID10_FAR_COPIES_SHIFT         8         /* raid10 # far copies shift (2nd byte of layout) */
 522
 523 /* Return md raid10 near copies for @layout */
 524 static unsigned int __raid10_near_copies(int layout)
 525 {
 526         return layout & 0xFF;
 527 }
 528
 529 /* Return md raid10 far copies for @layout */
 530 static unsigned int __raid10_far_copies(int layout)
 531 {
 532         return __raid10_near_copies(layout >> RAID10_FAR_COPIES_SHIFT);
 533 }
 534
 535 /* Return true if md raid10 offset for @layout */
 536 static bool __is_raid10_offset(int layout)
 537 {
 538         return !!(layout & RAID10_OFFSET);
 539 }
 540
 541 /* Return true if md raid10 near for @layout */
 542 static bool __is_raid10_near(int layout)
 543 {
 544         return !__is_raid10_offset(layout) && __raid10_near_copies(layout) > 1;
 545 }
 546
 547 /* Return true if md raid10 far for @layout */
 548 static bool __is_raid10_far(int layout)
 549 {
 550         return !__is_raid10_offset(layout) && __raid10_far_copies(layout) > 1;
 551 }
 552
 553 /* Return md raid10 layout string for @layout */
 554 static const char *raid10_md_layout_to_format(int layout)
 555 {
 556         /*
 557          * Bit 16 stands for "offset"
 558          * (i.e. adjacent stripes hold copies)
 559          *
 560          * Refer to MD's raid10.c for details
 561          */
 562         if (__is_raid10_offset(layout))
 563                 return "offset";
 564
 565         if (__raid10_near_copies(layout) > 1)
 566                 return "near";
 567
 568         if (__raid10_far_copies(layout) > 1)
 569                 return "far";
 570
 571         return "unknown";
 572 }
 573
 574 /* Return md raid10 algorithm for @name */
 575 static int raid10_name_to_format(const char *name)
 576 {
 577         if (!strcasecmp(name, "near"))
 578                 return ALGORITHM_RAID10_NEAR;
 579         else if (!strcasecmp(name, "offset"))
 580                 return ALGORITHM_RAID10_OFFSET;
 581         else if (!strcasecmp(name, "far"))
 582                 return ALGORITHM_RAID10_FAR;
 583
 584         return -EINVAL;
 585 }
 586
 587 /* Return md raid10 copies for @layout */
 588 static unsigned int raid10_md_layout_to_copies(int layout)
 589 {
 590         return max(__raid10_near_copies(layout), __raid10_far_copies(layout));
 591 }
 592
 593 /* Return md raid10 format id for @format string */
 594 static int raid10_format_to_md_layout(struct raid_set *rs,
 595                                       unsigned int algorithm,
 596                                       unsigned int copies)
 597 {
 598         unsigned int n = 1, f = 1, r = 0;
 599
 600         /*
 601          * MD resilienece flaw:
 602          *
 603          * enabling use_far_sets for far/offset formats causes copies
 604          * to be colocated on the same devs together with their origins!
 605          *
 606          * -> disable it for now in the definition above
 607          */
 608         if (algorithm == ALGORITHM_RAID10_DEFAULT ||
 609             algorithm == ALGORITHM_RAID10_NEAR)
 610                 n = copies;
 611
 612         else if (algorithm == ALGORITHM_RAID10_OFFSET) {
 613                 f = copies;
 614                 r = RAID10_OFFSET;
 615                 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
 616                         r |= RAID10_USE_FAR_SETS;
 617
 618         } else if (algorithm == ALGORITHM_RAID10_FAR) {
 619                 f = copies;
 620                 r = !RAID10_OFFSET;
 621                 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
 622                         r |= RAID10_USE_FAR_SETS;
 623
 624         } else
 625                 return -EINVAL;
 626
 627         return r | (f << RAID10_FAR_COPIES_SHIFT) | n;
 628 }
 629 /* END: MD raid10 bit definitions and helpers */
 630
 631 /* Check for any of the raid10 algorithms */
 632 static bool __got_raid10(struct raid_type *rtp, const int layout)
 633 {
 634         if (rtp->level == 10) {
 635                 switch (rtp->algorithm) {
 636                 case ALGORITHM_RAID10_DEFAULT:
 637                 case ALGORITHM_RAID10_NEAR:
 638                         return __is_raid10_near(layout);
 639                 case ALGORITHM_RAID10_OFFSET:
 640                         return __is_raid10_offset(layout);
 641                 case ALGORITHM_RAID10_FAR:
 642                         return __is_raid10_far(layout);
 643                 default:
 644                         break;
 645                 }
 646         }
 647
 648         return false;
 649 }
 650
 651 /* Return raid_type for @name */
 652 static struct raid_type *get_raid_type(const char *name)
 653 {
 654         struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
 655
 656         while (rtp-- > raid_types)
 657                 if (!strcasecmp(rtp->name, name))
 658                         return rtp;
 659
 660         return NULL;
 661 }
 662
 663 /* Return raid_type for @name based derived from @level and @layout */
 664 static struct raid_type *get_raid_type_by_ll(const int level, const int layout)
 665 {
 666         struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
 667
 668         while (rtp-- > raid_types) {
 669                 /* RAID10 special checks based on @layout flags/properties */
 670                 if (rtp->level == level &&
 671                     (__got_raid10(rtp, layout) || rtp->algorithm == layout))
 672                         return rtp;
 673         }
 674
 675         return NULL;
 676 }
 677
 678 /* Adjust rdev sectors */
 679 static void rs_set_rdev_sectors(struct raid_set *rs)
 680 {
 681         struct mddev *mddev = &rs->md;
 682         struct md_rdev *rdev;
 683
 684         /*
 685          * raid10 sets rdev->sector to the device size, which
 686          * is unintended in case of out-of-place reshaping
 687          */
 688         rdev_for_each(rdev, mddev)
 689                 if (!test_bit(Journal, &rdev->flags))
 690                         rdev->sectors = mddev->dev_sectors;
 691 }
 692
 693 /*
 694  * Change bdev capacity of @rs in case of a disk add/remove reshape
 695  */
 696 static void rs_set_capacity(struct raid_set *rs)
 697 {
 698         struct gendisk *gendisk = dm_disk(dm_table_get_md(rs->ti->table));
 699
 700         set_capacity(gendisk, rs->md.array_sectors);
 701         revalidate_disk(gendisk);
 702 }
 703
 704 /*
 705  * Set the mddev properties in @rs to the current
 706  * ones retrieved from the freshest superblock
 707  */
 708 static void rs_set_cur(struct raid_set *rs)
 709 {
 710         struct mddev *mddev = &rs->md;
 711
 712         mddev->new_level = mddev->level;
 713         mddev->new_layout = mddev->layout;
 714         mddev->new_chunk_sectors = mddev->chunk_sectors;
 715 }
 716
 717 /*
 718  * Set the mddev properties in @rs to the new
 719  * ones requested by the ctr
 720  */
 721 static void rs_set_new(struct raid_set *rs)
 722 {
 723         struct mddev *mddev = &rs->md;
 724
 725         mddev->level = mddev->new_level;
 726         mddev->layout = mddev->new_layout;
 727         mddev->chunk_sectors = mddev->new_chunk_sectors;
 728         mddev->raid_disks = rs->raid_disks;
 729         mddev->delta_disks = 0;
 730 }
 731
 732 static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
 733                                        unsigned int raid_devs)
 734 {
 735         unsigned int i;
 736         struct raid_set *rs;
 737
 738         if (raid_devs <= raid_type->parity_devs) {
 739                 ti->error = "Insufficient number of devices";
 740                 return ERR_PTR(-EINVAL);
 741         }
 742
 743         rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
 744         if (!rs) {
 745                 ti->error = "Cannot allocate raid context";
 746                 return ERR_PTR(-ENOMEM);
 747         }
 748
 749         mddev_init(&rs->md);
 750
 751         rs->raid_disks = raid_devs;
 752         rs->delta_disks = 0;
 753
 754         rs->ti = ti;
 755         rs->raid_type = raid_type;
 756         rs->stripe_cache_entries = 256;
 757         rs->md.raid_disks = raid_devs;
 758         rs->md.level = raid_type->level;
 759         rs->md.new_level = rs->md.level;
 760         rs->md.layout = raid_type->algorithm;
 761         rs->md.new_layout = rs->md.layout;
 762         rs->md.delta_disks = 0;
 763         rs->md.recovery_cp = MaxSector;
 764
 765         for (i = 0; i < raid_devs; i++)
 766                 md_rdev_init(&rs->dev[i].rdev);
 767
 768         /*
 769          * Remaining items to be initialized by further RAID params:
 770          *  rs->md.persistent
 771          *  rs->md.external
 772          *  rs->md.chunk_sectors
 773          *  rs->md.new_chunk_sectors
 774          *  rs->md.dev_sectors
 775          */
 776
 777         return rs;
 778 }
 779
 780 static void raid_set_free(struct raid_set *rs)
 781 {
 782         int i;
 783
 784         if (rs->journal_dev.dev) {
 785                 md_rdev_clear(&rs->journal_dev.rdev);
 786                 dm_put_device(rs->ti, rs->journal_dev.dev);
 787         }
 788
 789         for (i = 0; i < rs->raid_disks; i++) {
 790                 if (rs->dev[i].meta_dev)
 791                         dm_put_device(rs->ti, rs->dev[i].meta_dev);
 792                 md_rdev_clear(&rs->dev[i].rdev);
 793                 if (rs->dev[i].data_dev)
 794                         dm_put_device(rs->ti, rs->dev[i].data_dev);
 795         }
 796
 797         kfree(rs);
 798 }
 799
 800 /*
 801  * For every device we have two words
 802  *  <meta_dev>: meta device name or '-' if missing
 803  *  <data_dev>: data device name or '-' if missing
 804  *
 805  * The following are permitted:
 806  *    - -
 807  *    - <data_dev>
 808  *    <meta_dev> <data_dev>
 809  *
 810  * The following is not allowed:
 811  *    <meta_dev> -
 812  *
 813  * This code parses those words.  If there is a failure,
 814  * the caller must use raid_set_free() to unwind the operations.
 815  */
 816 static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
 817 {
 818         int i;
 819         int rebuild = 0;
 820         int metadata_available = 0;
 821         int r = 0;
 822         const char *arg;
 823
 824         /* Put off the number of raid devices argument to get to dev pairs */
 825         arg = dm_shift_arg(as);
 826         if (!arg)
 827                 return -EINVAL;
 828
 829         for (i = 0; i < rs->raid_disks; i++) {
 830                 rs->dev[i].rdev.raid_disk = i;
 831
 832                 rs->dev[i].meta_dev = NULL;
 833                 rs->dev[i].data_dev = NULL;
 834
 835                 /*
 836                  * There are no offsets initially.
 837                  * Out of place reshape will set them accordingly.
 838                  */
 839                 rs->dev[i].rdev.data_offset = 0;
 840                 rs->dev[i].rdev.new_data_offset = 0;
 841                 rs->dev[i].rdev.mddev = &rs->md;
 842
 843                 arg = dm_shift_arg(as);
 844                 if (!arg)
 845                         return -EINVAL;
 846
 847                 if (strcmp(arg, "-")) {
 848                         r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
 849                                           &rs->dev[i].meta_dev);
 850                         if (r) {
 851                                 rs->ti->error = "RAID metadata device lookup failure";
 852                                 return r;
 853                         }
 854
 855                         rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
 856                         if (!rs->dev[i].rdev.sb_page) {
 857                                 rs->ti->error = "Failed to allocate superblock page";
 858                                 return -ENOMEM;
 859                         }
 860                 }
 861
 862                 arg = dm_shift_arg(as);
 863                 if (!arg)
 864                         return -EINVAL;
 865
 866                 if (!strcmp(arg, "-")) {
 867                         if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
 868                             (!rs->dev[i].rdev.recovery_offset)) {
 869                                 rs->ti->error = "Drive designated for rebuild not specified";
 870                                 return -EINVAL;
 871                         }
 872
 873                         if (rs->dev[i].meta_dev) {
 874                                 rs->ti->error = "No data device supplied with metadata device";
 875                                 return -EINVAL;
 876                         }
 877
 878                         continue;
 879                 }
 880
 881                 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
 882                                   &rs->dev[i].data_dev);
 883                 if (r) {
 884                         rs->ti->error = "RAID device lookup failure";
 885                         return r;
 886                 }
 887
 888                 if (rs->dev[i].meta_dev) {
 889                         metadata_available = 1;
 890                         rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
 891                 }
 892                 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
 893                 list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
 894                 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
 895                         rebuild++;
 896         }
 897
 898         if (rs->journal_dev.dev)
 899                 list_add_tail(&rs->journal_dev.rdev.same_set, &rs->md.disks);
 900
 901         if (metadata_available) {
 902                 rs->md.external = 0;
 903                 rs->md.persistent = 1;
 904                 rs->md.major_version = 2;
 905         } else if (rebuild && !rs->md.recovery_cp) {
 906                 /*
 907                  * Without metadata, we will not be able to tell if the array
 908                  * is in-sync or not - we must assume it is not.  Therefore,
 909                  * it is impossible to rebuild a drive.
 910                  *
 911                  * Even if there is metadata, the on-disk information may
 912                  * indicate that the array is not in-sync and it will then
 913                  * fail at that time.
 914                  *
 915                  * User could specify 'nosync' option if desperate.
 916                  */
 917                 rs->ti->error = "Unable to rebuild drive while array is not in-sync";
 918                 return -EINVAL;
 919         }
 920
 921         return 0;
 922 }
 923
 924 /*
 925  * validate_region_size
 926  * @rs
 927  * @region_size:  region size in sectors.  If 0, pick a size (4MiB default).
 928  *
 929  * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
 930  * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
 931  *
 932  * Returns: 0 on success, -EINVAL on failure.
 933  */
 934 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
 935 {
 936         unsigned long min_region_size = rs->ti->len / (1 << 21);
 937
 938         if (rs_is_raid0(rs))
 939                 return 0;
 940
 941         if (!region_size) {
 942                 /*
 943                  * Choose a reasonable default.  All figures in sectors.
 944                  */
 945                 if (min_region_size > (1 << 13)) {
 946                         /* If not a power of 2, make it the next power of 2 */
 947                         region_size = roundup_pow_of_two(min_region_size);
 948                         DMINFO("Choosing default region size of %lu sectors",
 949                                region_size);
 950                 } else {
 951                         DMINFO("Choosing default region size of 4MiB");
 952                         region_size = 1 << 13; /* sectors */
 953                 }
 954         } else {
 955                 /*
 956                  * Validate user-supplied value.
 957                  */
 958                 if (region_size > rs->ti->len) {
 959                         rs->ti->error = "Supplied region size is too large";
 960                         return -EINVAL;
 961                 }
 962
 963                 if (region_size < min_region_size) {
 964                         DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
 965                               region_size, min_region_size);
 966                         rs->ti->error = "Supplied region size is too small";
 967                         return -EINVAL;
 968                 }
 969
 970                 if (!is_power_of_2(region_size)) {
 971                         rs->ti->error = "Region size is not a power of 2";
 972                         return -EINVAL;
 973                 }
 974
 975                 if (region_size < rs->md.chunk_sectors) {
 976                         rs->ti->error = "Region size is smaller than the chunk size";
 977                         return -EINVAL;
 978                 }
 979         }
 980
 981         /*
 982          * Convert sectors to bytes.
 983          */
 984         rs->md.bitmap_info.chunksize = to_bytes(region_size);
 985
 986         return 0;
 987 }
 988
 989 /*
 990  * validate_raid_redundancy
 991  * @rs
 992  *
 993  * Determine if there are enough devices in the array that haven't
 994  * failed (or are being rebuilt) to form a usable array.
 995  *
 996  * Returns: 0 on success, -EINVAL on failure.
 997  */
 998 static int validate_raid_redundancy(struct raid_set *rs)
 999 {
1000         unsigned int i, rebuild_cnt = 0;
1001         unsigned int rebuilds_per_group = 0, copies, raid_disks;
1002         unsigned int group_size, last_group_start;
1003
1004         for (i = 0; i < rs->raid_disks; i++)
1005                 if (!test_bit(FirstUse, &rs->dev[i].rdev.flags) &&
1006                     ((!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
1007                       !rs->dev[i].rdev.sb_page)))
1008                         rebuild_cnt++;
1009
1010         switch (rs->raid_type->level) {
1011         case 0:
1012                 break;
1013         case 1:
1014                 if (rebuild_cnt >= rs->md.raid_disks)
1015                         goto too_many;
1016                 break;
1017         case 4:
1018         case 5:
1019         case 6:
1020                 if (rebuild_cnt > rs->raid_type->parity_devs)
1021                         goto too_many;
1022                 break;
1023         case 10:
1024                 copies = raid10_md_layout_to_copies(rs->md.new_layout);
1025                 if (rebuild_cnt < copies)
1026                         break;
1027
1028                 /*
1029                  * It is possible to have a higher rebuild count for RAID10,
1030                  * as long as the failed devices occur in different mirror
1031                  * groups (i.e. different stripes).
1032                  *
1033                  * When checking "near" format, make sure no adjacent devices
1034                  * have failed beyond what can be handled.  In addition to the
1035                  * simple case where the number of devices is a multiple of the
1036                  * number of copies, we must also handle cases where the number
1037                  * of devices is not a multiple of the number of copies.
1038                  * E.g.    dev1 dev2 dev3 dev4 dev5
1039                  *          A    A    B    B    C
1040                  *          C    D    D    E    E
1041                  */
1042                 raid_disks = min(rs->raid_disks, rs->md.raid_disks);
1043                 if (__is_raid10_near(rs->md.new_layout)) {
1044                         for (i = 0; i < raid_disks; i++) {
1045                                 if (!(i % copies))
1046                                         rebuilds_per_group = 0;
1047                                 if ((!rs->dev[i].rdev.sb_page ||
1048                                     !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1049                                     (++rebuilds_per_group >= copies))
1050                                         goto too_many;
1051                         }
1052                         break;
1053                 }
1054
1055                 /*
1056                  * When checking "far" and "offset" formats, we need to ensure
1057                  * that the device that holds its copy is not also dead or
1058                  * being rebuilt.  (Note that "far" and "offset" formats only
1059                  * support two copies right now.  These formats also only ever
1060                  * use the 'use_far_sets' variant.)
1061                  *
1062                  * This check is somewhat complicated by the need to account
1063                  * for arrays that are not a multiple of (far) copies.  This
1064                  * results in the need to treat the last (potentially larger)
1065                  * set differently.
1066                  */
1067                 group_size = (raid_disks / copies);
1068                 last_group_start = (raid_disks / group_size) - 1;
1069                 last_group_start *= group_size;
1070                 for (i = 0; i < raid_disks; i++) {
1071                         if (!(i % copies) && !(i > last_group_start))
1072                                 rebuilds_per_group = 0;
1073                         if ((!rs->dev[i].rdev.sb_page ||
1074                              !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1075                             (++rebuilds_per_group >= copies))
1076                                         goto too_many;
1077                 }
1078                 break;
1079         default:
1080                 if (rebuild_cnt)
1081                         return -EINVAL;
1082         }
1083
1084         return 0;
1085
1086 too_many:
1087         return -EINVAL;
1088 }
1089
1090 /*
1091  * Possible arguments are...
1092  *      <chunk_size> [optional_args]
1093  *
1094  * Argument definitions
1095  *    <chunk_size>                      The number of sectors per disk that
1096  *                                      will form the "stripe"
1097  *    [[no]sync]                        Force or prevent recovery of the
1098  *                                      entire array
1099  *    [rebuild <idx>]                   Rebuild the drive indicated by the index
1100  *    [daemon_sleep <ms>]               Time between bitmap daemon work to
1101  *                                      clear bits
1102  *    [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1103  *    [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1104  *    [write_mostly <idx>]              Indicate a write mostly drive via index
1105  *    [max_write_behind <sectors>]      See '-write-behind=' (man mdadm)
1106  *    [stripe_cache <sectors>]          Stripe cache size for higher RAIDs
1107  *    [region_size <sectors>]           Defines granularity of bitmap
1108  *    [journal_dev <dev>]               raid4/5/6 journaling deviice
1109  *                                      (i.e. write hole closing log)
1110  *
1111  * RAID10-only options:
1112  *    [raid10_copies <# copies>]        Number of copies.  (Default: 2)
1113  *    [raid10_format <near|far|offset>] Layout algorithm.  (Default: near)
1114  */
1115 static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1116                              unsigned int num_raid_params)
1117 {
1118         int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1119         unsigned int raid10_copies = 2;
1120         unsigned int i, write_mostly = 0;
1121         unsigned int region_size = 0;
1122         sector_t max_io_len;
1123         const char *arg, *key;
1124         struct raid_dev *rd;
1125         struct raid_type *rt = rs->raid_type;
1126
1127         arg = dm_shift_arg(as);
1128         num_raid_params--; /* Account for chunk_size argument */
1129
1130         if (kstrtoint(arg, 10, &value) < 0) {
1131                 rs->ti->error = "Bad numerical argument given for chunk_size";
1132                 return -EINVAL;
1133         }
1134
1135         /*
1136          * First, parse the in-order required arguments
1137          * "chunk_size" is the only argument of this type.
1138          */
1139         if (rt_is_raid1(rt)) {
1140                 if (value)
1141                         DMERR("Ignoring chunk size parameter for RAID 1");
1142                 value = 0;
1143         } else if (!is_power_of_2(value)) {
1144                 rs->ti->error = "Chunk size must be a power of 2";
1145                 return -EINVAL;
1146         } else if (value < 8) {
1147                 rs->ti->error = "Chunk size value is too small";
1148                 return -EINVAL;
1149         }
1150
1151         rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1152
1153         /*
1154          * We set each individual device as In_sync with a completed
1155          * 'recovery_offset'.  If there has been a device failure or
1156          * replacement then one of the following cases applies:
1157          *
1158          *   1) User specifies 'rebuild'.
1159          *      - Device is reset when param is read.
1160          *   2) A new device is supplied.
1161          *      - No matching superblock found, resets device.
1162          *   3) Device failure was transient and returns on reload.
1163          *      - Failure noticed, resets device for bitmap replay.
1164          *   4) Device hadn't completed recovery after previous failure.
1165          *      - Superblock is read and overrides recovery_offset.
1166          *
1167          * What is found in the superblocks of the devices is always
1168          * authoritative, unless 'rebuild' or '[no]sync' was specified.
1169          */
1170         for (i = 0; i < rs->raid_disks; i++) {
1171                 set_bit(In_sync, &rs->dev[i].rdev.flags);
1172                 rs->dev[i].rdev.recovery_offset = MaxSector;
1173         }
1174
1175         /*
1176          * Second, parse the unordered optional arguments
1177          */
1178         for (i = 0; i < num_raid_params; i++) {
1179                 key = dm_shift_arg(as);
1180                 if (!key) {
1181                         rs->ti->error = "Not enough raid parameters given";
1182                         return -EINVAL;
1183                 }
1184
1185                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1186                         if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1187                                 rs->ti->error = "Only one 'nosync' argument allowed";
1188                                 return -EINVAL;
1189                         }
1190                         continue;
1191                 }
1192                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1193                         if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1194                                 rs->ti->error = "Only one 'sync' argument allowed";
1195                                 return -EINVAL;
1196                         }
1197                         continue;
1198                 }
1199                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1200                         if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1201                                 rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1202                                 return -EINVAL;
1203                         }
1204                         continue;
1205                 }
1206
1207                 arg = dm_shift_arg(as);
1208                 i++; /* Account for the argument pairs */
1209                 if (!arg) {
1210                         rs->ti->error = "Wrong number of raid parameters given";
1211                         return -EINVAL;
1212                 }
1213
1214                 /*
1215                  * Parameters that take a string value are checked here.
1216                  */
1217                 /* "raid10_format {near|offset|far} */
1218                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1219                         if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1220                                 rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1221                                 return -EINVAL;
1222                         }
1223                         if (!rt_is_raid10(rt)) {
1224                                 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1225                                 return -EINVAL;
1226                         }
1227                         raid10_format = raid10_name_to_format(arg);
1228                         if (raid10_format < 0) {
1229                                 rs->ti->error = "Invalid 'raid10_format' value given";
1230                                 return raid10_format;
1231                         }
1232                         continue;
1233                 }
1234
1235                 /* "journal_dev <dev>" */
1236                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV))) {
1237                         int r;
1238                         struct md_rdev *jdev;
1239
1240                         if (test_and_set_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1241                                 rs->ti->error = "Only one raid4/5/6 set journaling device allowed";
1242                                 return -EINVAL;
1243                         }
1244                         if (!rt_is_raid456(rt)) {
1245                                 rs->ti->error = "'journal_dev' is an invalid parameter for this RAID type";
1246                                 return -EINVAL;
1247                         }
1248                         r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
1249                                           &rs->journal_dev.dev);
1250                         if (r) {
1251                                 rs->ti->error = "raid4/5/6 journal device lookup failure";
1252                                 return r;
1253                         }
1254                         jdev = &rs->journal_dev.rdev;
1255                         md_rdev_init(jdev);
1256                         jdev->mddev = &rs->md;
1257                         jdev->bdev = rs->journal_dev.dev->bdev;
1258                         jdev->sectors = to_sector(i_size_read(jdev->bdev->bd_inode));
1259                         if (jdev->sectors < MIN_RAID456_JOURNAL_SPACE) {
1260                                 rs->ti->error = "No space for raid4/5/6 journal";
1261                                 return -ENOSPC;
1262                         }
1263                         rs->journal_dev.mode = R5C_JOURNAL_MODE_WRITE_THROUGH;
1264                         set_bit(Journal, &jdev->flags);
1265                         continue;
1266                 }
1267
1268                 /* "journal_mode <mode>" ("journal_dev" mandatory!) */
1269                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE))) {
1270                         int r;
1271
1272                         if (!test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1273                                 rs->ti->error = "raid4/5/6 'journal_mode' is invalid without 'journal_dev'";
1274                                 return -EINVAL;
1275                         }
1276                         if (test_and_set_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
1277                                 rs->ti->error = "Only one raid4/5/6 'journal_mode' argument allowed";
1278                                 return -EINVAL;
1279                         }
1280                         r = dm_raid_journal_mode_to_md(arg);
1281                         if (r < 0) {
1282                                 rs->ti->error = "Invalid 'journal_mode' argument";
1283                                 return r;
1284                         }
1285                         rs->journal_dev.mode = r;
1286                         continue;
1287                 }
1288
1289                 /*
1290                  * Parameters with number values from here on.
1291                  */
1292                 if (kstrtoint(arg, 10, &value) < 0) {
1293                         rs->ti->error = "Bad numerical argument given in raid params";
1294                         return -EINVAL;
1295                 }
1296
1297                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1298                         /*
1299                          * "rebuild" is being passed in by userspace to provide
1300                          * indexes of replaced devices and to set up additional
1301                          * devices on raid level takeover.
1302                          */
1303                         if (!__within_range(value, 0, rs->raid_disks - 1)) {
1304                                 rs->ti->error = "Invalid rebuild index given";
1305                                 return -EINVAL;
1306                         }
1307
1308                         if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1309                                 rs->ti->error = "rebuild for this index already given";
1310                                 return -EINVAL;
1311                         }
1312
1313                         rd = rs->dev + value;
1314                         clear_bit(In_sync, &rd->rdev.flags);
1315                         clear_bit(Faulty, &rd->rdev.flags);
1316                         rd->rdev.recovery_offset = 0;
1317                         set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1318                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1319                         if (!rt_is_raid1(rt)) {
1320                                 rs->ti->error = "write_mostly option is only valid for RAID1";
1321                                 return -EINVAL;
1322                         }
1323
1324                         if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1325                                 rs->ti->error = "Invalid write_mostly index given";
1326                                 return -EINVAL;
1327                         }
1328
1329                         write_mostly++;
1330                         set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1331                         set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1332                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1333                         if (!rt_is_raid1(rt)) {
1334                                 rs->ti->error = "max_write_behind option is only valid for RAID1";
1335                                 return -EINVAL;
1336                         }
1337
1338                         if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1339                                 rs->ti->error = "Only one max_write_behind argument pair allowed";
1340                                 return -EINVAL;
1341                         }
1342
1343                         /*
1344                          * In device-mapper, we specify things in sectors, but
1345                          * MD records this value in kB
1346                          */
1347                         value /= 2;
1348                         if (value > COUNTER_MAX) {
1349                                 rs->ti->error = "Max write-behind limit out of range";
1350                                 return -EINVAL;
1351                         }
1352
1353                         rs->md.bitmap_info.max_write_behind = value;
1354                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1355                         if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1356                                 rs->ti->error = "Only one daemon_sleep argument pair allowed";
1357                                 return -EINVAL;
1358                         }
1359                         if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
1360                                 rs->ti->error = "daemon sleep period out of range";
1361                                 return -EINVAL;
1362                         }
1363                         rs->md.bitmap_info.daemon_sleep = value;
1364                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1365                         /* Userspace passes new data_offset after having extended the the data image LV */
1366                         if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1367                                 rs->ti->error = "Only one data_offset argument pair allowed";
1368                                 return -EINVAL;
1369                         }
1370                         /* Ensure sensible data offset */
1371                         if (value < 0 ||
1372                             (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
1373                                 rs->ti->error = "Bogus data_offset value";
1374                                 return -EINVAL;
1375                         }
1376                         rs->data_offset = value;
1377                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1378                         /* Define the +/-# of disks to add to/remove from the given raid set */
1379                         if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1380                                 rs->ti->error = "Only one delta_disks argument pair allowed";
1381                                 return -EINVAL;
1382                         }
1383                         /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1384                         if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1385                                 rs->ti->error = "Too many delta_disk requested";
1386                                 return -EINVAL;
1387                         }
1388
1389                         rs->delta_disks = value;
1390                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1391                         if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1392                                 rs->ti->error = "Only one stripe_cache argument pair allowed";
1393                                 return -EINVAL;
1394                         }
1395
1396                         if (!rt_is_raid456(rt)) {
1397                                 rs->ti->error = "Inappropriate argument: stripe_cache";
1398                                 return -EINVAL;
1399                         }
1400
1401                         rs->stripe_cache_entries = value;
1402                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1403                         if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1404                                 rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1405                                 return -EINVAL;
1406                         }
1407                         if (value > INT_MAX) {
1408                                 rs->ti->error = "min_recovery_rate out of range";
1409                                 return -EINVAL;
1410                         }
1411                         rs->md.sync_speed_min = (int)value;
1412                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1413                         if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
1414                                 rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1415                                 return -EINVAL;
1416                         }
1417                         if (value > INT_MAX) {
1418                                 rs->ti->error = "max_recovery_rate out of range";
1419                                 return -EINVAL;
1420                         }
1421                         rs->md.sync_speed_max = (int)value;
1422                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1423                         if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1424                                 rs->ti->error = "Only one region_size argument pair allowed";
1425                                 return -EINVAL;
1426                         }
1427
1428                         region_size = value;
1429                         rs->requested_bitmap_chunk_sectors = value;
1430                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1431                         if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1432                                 rs->ti->error = "Only one raid10_copies argument pair allowed";
1433                                 return -EINVAL;
1434                         }
1435
1436                         if (!__within_range(value, 2, rs->md.raid_disks)) {
1437                                 rs->ti->error = "Bad value for 'raid10_copies'";
1438                                 return -EINVAL;
1439                         }
1440
1441                         raid10_copies = value;
1442                 } else {
1443                         DMERR("Unable to parse RAID parameter: %s", key);
1444                         rs->ti->error = "Unable to parse RAID parameter";
1445                         return -EINVAL;
1446                 }
1447         }
1448
1449         if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1450             test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1451                 rs->ti->error = "sync and nosync are mutually exclusive";
1452                 return -EINVAL;
1453         }
1454
1455         if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1456             (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1457              test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1458                 rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1459                 return -EINVAL;
1460         }
1461
1462         if (write_mostly >= rs->md.raid_disks) {
1463                 rs->ti->error = "Can't set all raid1 devices to write_mostly";
1464                 return -EINVAL;
1465         }
1466
1467         if (validate_region_size(rs, region_size))
1468                 return -EINVAL;
1469
1470         if (rs->md.chunk_sectors)
1471                 max_io_len = rs->md.chunk_sectors;
1472         else
1473                 max_io_len = region_size;
1474
1475         if (dm_set_target_max_io_len(rs->ti, max_io_len))
1476                 return -EINVAL;
1477
1478         if (rt_is_raid10(rt)) {
1479                 if (raid10_copies > rs->md.raid_disks) {
1480                         rs->ti->error = "Not enough devices to satisfy specification";
1481                         return -EINVAL;
1482                 }
1483
1484                 rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1485                 if (rs->md.new_layout < 0) {
1486                         rs->ti->error = "Error getting raid10 format";
1487                         return rs->md.new_layout;
1488                 }
1489
1490                 rt = get_raid_type_by_ll(10, rs->md.new_layout);
1491                 if (!rt) {
1492                         rs->ti->error = "Failed to recognize new raid10 layout";
1493                         return -EINVAL;
1494                 }
1495
1496                 if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1497                      rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1498                     test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1499                         rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1500                         return -EINVAL;
1501                 }
1502         }
1503
1504         rs->raid10_copies = raid10_copies;
1505
1506         /* Assume there are no metadata devices until the drives are parsed */
1507         rs->md.persistent = 0;
1508         rs->md.external = 1;
1509
1510         /* Check, if any invalid ctr arguments have been passed in for the raid level */
1511         return rs_check_for_valid_flags(rs);
1512 }
1513
1514 /* Set raid4/5/6 cache size */
1515 static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1516 {
1517         int r;
1518         struct r5conf *conf;
1519         struct mddev *mddev = &rs->md;
1520         uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1521         uint32_t nr_stripes = rs->stripe_cache_entries;
1522
1523         if (!rt_is_raid456(rs->raid_type)) {
1524                 rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1525                 return -EINVAL;
1526         }
1527
1528         if (nr_stripes < min_stripes) {
1529                 DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1530                        nr_stripes, min_stripes);
1531                 nr_stripes = min_stripes;
1532         }
1533
1534         conf = mddev->private;
1535         if (!conf) {
1536                 rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1537                 return -EINVAL;
1538         }
1539
1540         /* Try setting number of stripes in raid456 stripe cache */
1541         if (conf->min_nr_stripes != nr_stripes) {
1542                 r = raid5_set_cache_size(mddev, nr_stripes);
1543                 if (r) {
1544                         rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1545                         return r;
1546                 }
1547
1548                 DMINFO("%u stripe cache entries", nr_stripes);
1549         }
1550
1551         return 0;
1552 }
1553
1554 /* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1555 static unsigned int mddev_data_stripes(struct raid_set *rs)
1556 {
1557         return rs->md.raid_disks - rs->raid_type->parity_devs;
1558 }
1559
1560 /* Return # of data stripes of @rs (i.e. as of ctr) */
1561 static unsigned int rs_data_stripes(struct raid_set *rs)
1562 {
1563         return rs->raid_disks - rs->raid_type->parity_devs;
1564 }
1565
1566 /*
1567  * Retrieve rdev->sectors from any valid raid device of @rs
1568  * to allow userpace to pass in arbitray "- -" device tupples.
1569  */
1570 static sector_t __rdev_sectors(struct raid_set *rs)
1571 {
1572         int i;
1573
1574         for (i = 0; i < rs->raid_disks; i++) {
1575                 struct md_rdev *rdev = &rs->dev[i].rdev;
1576
1577                 if (!test_bit(Journal, &rdev->flags) &&
1578                     rdev->bdev && rdev->sectors)
1579                         return rdev->sectors;
1580         }
1581
1582         return 0;
1583 }
1584
1585 /* Calculate the sectors per device and per array used for @rs */
1586 static int rs_set_dev_and_array_sectors(struct raid_set *rs, bool use_mddev)
1587 {
1588         int delta_disks;
1589         unsigned int data_stripes;
1590         struct mddev *mddev = &rs->md;
1591         struct md_rdev *rdev;
1592         sector_t array_sectors = rs->ti->len, dev_sectors = rs->ti->len;
1593
1594         if (use_mddev) {
1595                 delta_disks = mddev->delta_disks;
1596                 data_stripes = mddev_data_stripes(rs);
1597         } else {
1598                 delta_disks = rs->delta_disks;
1599                 data_stripes = rs_data_stripes(rs);
1600         }
1601
1602         /* Special raid1 case w/o delta_disks support (yet) */
1603         if (rt_is_raid1(rs->raid_type))
1604                 ;
1605         else if (rt_is_raid10(rs->raid_type)) {
1606                 if (rs->raid10_copies < 2 ||
1607                     delta_disks < 0) {
1608                         rs->ti->error = "Bogus raid10 data copies or delta disks";
1609                         return -EINVAL;
1610                 }
1611
1612                 dev_sectors *= rs->raid10_copies;
1613                 if (sector_div(dev_sectors, data_stripes))
1614                         goto bad;
1615
1616                 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1617                 if (sector_div(array_sectors, rs->raid10_copies))
1618                         goto bad;
1619
1620         } else if (sector_div(dev_sectors, data_stripes))
1621                 goto bad;
1622
1623         else
1624                 /* Striped layouts */
1625                 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1626
1627         rdev_for_each(rdev, mddev)
1628                 if (!test_bit(Journal, &rdev->flags))
1629                         rdev->sectors = dev_sectors;
1630
1631         mddev->array_sectors = array_sectors;
1632         mddev->dev_sectors = dev_sectors;
1633
1634         return 0;
1635 bad:
1636         rs->ti->error = "Target length not divisible by number of data devices";
1637         return -EINVAL;
1638 }
1639
1640 /* Setup recovery on @rs */
1641 static void __rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1642 {
1643         /* raid0 does not recover */
1644         if (rs_is_raid0(rs))
1645                 rs->md.recovery_cp = MaxSector;
1646         /*
1647          * A raid6 set has to be recovered either
1648          * completely or for the grown part to
1649          * ensure proper parity and Q-Syndrome
1650          */
1651         else if (rs_is_raid6(rs))
1652                 rs->md.recovery_cp = dev_sectors;
1653         /*
1654          * Other raid set types may skip recovery
1655          * depending on the 'nosync' flag.
1656          */
1657         else
1658                 rs->md.recovery_cp = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1659                                      ? MaxSector : dev_sectors;
1660 }
1661
1662 /* Setup recovery on @rs based on raid type, device size and 'nosync' flag */
1663 static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1664 {
1665         if (!dev_sectors)
1666                 /* New raid set or 'sync' flag provided */
1667                 __rs_setup_recovery(rs, 0);
1668         else if (dev_sectors == MaxSector)
1669                 /* Prevent recovery */
1670                 __rs_setup_recovery(rs, MaxSector);
1671         else if (__rdev_sectors(rs) < dev_sectors)
1672                 /* Grown raid set */
1673                 __rs_setup_recovery(rs, __rdev_sectors(rs));
1674         else
1675                 __rs_setup_recovery(rs, MaxSector);
1676 }
1677
1678 static void do_table_event(struct work_struct *ws)
1679 {
1680         struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1681
1682         smp_rmb(); /* Make sure we access most actual mddev properties */
1683         if (!rs_is_reshaping(rs)) {
1684                 if (rs_is_raid10(rs))
1685                         rs_set_rdev_sectors(rs);
1686                 rs_set_capacity(rs);
1687         }
1688         dm_table_event(rs->ti->table);
1689 }
1690
1691 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
1692 {
1693         struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
1694
1695         return mddev_congested(&rs->md, bits);
1696 }
1697
1698 /*
1699  * Make sure a valid takover (level switch) is being requested on @rs
1700  *
1701  * Conversions of raid sets from one MD personality to another
1702  * have to conform to restrictions which are enforced here.
1703  */
1704 static int rs_check_takeover(struct raid_set *rs)
1705 {
1706         struct mddev *mddev = &rs->md;
1707         unsigned int near_copies;
1708
1709         if (rs->md.degraded) {
1710                 rs->ti->error = "Can't takeover degraded raid set";
1711                 return -EPERM;
1712         }
1713
1714         if (rs_is_reshaping(rs)) {
1715                 rs->ti->error = "Can't takeover reshaping raid set";
1716                 return -EPERM;
1717         }
1718
1719         switch (mddev->level) {
1720         case 0:
1721                 /* raid0 -> raid1/5 with one disk */
1722                 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1723                     mddev->raid_disks == 1)
1724                         return 0;
1725
1726                 /* raid0 -> raid10 */
1727                 if (mddev->new_level == 10 &&
1728                     !(rs->raid_disks % mddev->raid_disks))
1729                         return 0;
1730
1731                 /* raid0 with multiple disks -> raid4/5/6 */
1732                 if (__within_range(mddev->new_level, 4, 6) &&
1733                     mddev->new_layout == ALGORITHM_PARITY_N &&
1734                     mddev->raid_disks > 1)
1735                         return 0;
1736
1737                 break;
1738
1739         case 10:
1740                 /* Can't takeover raid10_offset! */
1741                 if (__is_raid10_offset(mddev->layout))
1742                         break;
1743
1744                 near_copies = __raid10_near_copies(mddev->layout);
1745
1746                 /* raid10* -> raid0 */
1747                 if (mddev->new_level == 0) {
1748                         /* Can takeover raid10_near with raid disks divisable by data copies! */
1749                         if (near_copies > 1 &&
1750                             !(mddev->raid_disks % near_copies)) {
1751                                 mddev->raid_disks /= near_copies;
1752                                 mddev->delta_disks = mddev->raid_disks;
1753                                 return 0;
1754                         }
1755
1756                         /* Can takeover raid10_far */
1757                         if (near_copies == 1 &&
1758                             __raid10_far_copies(mddev->layout) > 1)
1759                                 return 0;
1760
1761                         break;
1762                 }
1763
1764                 /* raid10_{near,far} -> raid1 */
1765                 if (mddev->new_level == 1 &&
1766                     max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1767                         return 0;
1768
1769                 /* raid10_{near,far} with 2 disks -> raid4/5 */
1770                 if (__within_range(mddev->new_level, 4, 5) &&
1771                     mddev->raid_disks == 2)
1772                         return 0;
1773                 break;
1774
1775         case 1:
1776                 /* raid1 with 2 disks -> raid4/5 */
1777                 if (__within_range(mddev->new_level, 4, 5) &&
1778                     mddev->raid_disks == 2) {
1779                         mddev->degraded = 1;
1780                         return 0;
1781                 }
1782
1783                 /* raid1 -> raid0 */
1784                 if (mddev->new_level == 0 &&
1785                     mddev->raid_disks == 1)
1786                         return 0;
1787
1788                 /* raid1 -> raid10 */
1789                 if (mddev->new_level == 10)
1790                         return 0;
1791                 break;
1792
1793         case 4:
1794                 /* raid4 -> raid0 */
1795                 if (mddev->new_level == 0)
1796                         return 0;
1797
1798                 /* raid4 -> raid1/5 with 2 disks */
1799                 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1800                     mddev->raid_disks == 2)
1801                         return 0;
1802
1803                 /* raid4 -> raid5/6 with parity N */
1804                 if (__within_range(mddev->new_level, 5, 6) &&
1805                     mddev->layout == ALGORITHM_PARITY_N)
1806                         return 0;
1807                 break;
1808
1809         case 5:
1810                 /* raid5 with parity N -> raid0 */
1811                 if (mddev->new_level == 0 &&
1812                     mddev->layout == ALGORITHM_PARITY_N)
1813                         return 0;
1814
1815                 /* raid5 with parity N -> raid4 */
1816                 if (mddev->new_level == 4 &&
1817                     mddev->layout == ALGORITHM_PARITY_N)
1818                         return 0;
1819
1820                 /* raid5 with 2 disks -> raid1/4/10 */
1821                 if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1822                     mddev->raid_disks == 2)
1823                         return 0;
1824
1825                 /* raid5_* ->  raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1826                 if (mddev->new_level == 6 &&
1827                     ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1828                       __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1829                         return 0;
1830                 break;
1831
1832         case 6:
1833                 /* raid6 with parity N -> raid0 */
1834                 if (mddev->new_level == 0 &&
1835                     mddev->layout == ALGORITHM_PARITY_N)
1836                         return 0;
1837
1838                 /* raid6 with parity N -> raid4 */
1839                 if (mddev->new_level == 4 &&
1840                     mddev->layout == ALGORITHM_PARITY_N)
1841                         return 0;
1842
1843                 /* raid6_*_n with Q-Syndrome N -> raid5_* */
1844                 if (mddev->new_level == 5 &&
1845                     ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1846                      __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1847                         return 0;
1848
1849         default:
1850                 break;
1851         }
1852
1853         rs->ti->error = "takeover not possible";
1854         return -EINVAL;
1855 }
1856
1857 /* True if @rs requested to be taken over */
1858 static bool rs_takeover_requested(struct raid_set *rs)
1859 {
1860         return rs->md.new_level != rs->md.level;
1861 }
1862
1863 /* True if @rs is requested to reshape by ctr */
1864 static bool rs_reshape_requested(struct raid_set *rs)
1865 {
1866         bool change;
1867         struct mddev *mddev = &rs->md;
1868
1869         if (rs_takeover_requested(rs))
1870                 return false;
1871
1872         if (!mddev->level)
1873                 return false;
1874
1875         change = mddev->new_layout != mddev->layout ||
1876                  mddev->new_chunk_sectors != mddev->chunk_sectors ||
1877                  rs->delta_disks;
1878
1879         /* Historical case to support raid1 reshape without delta disks */
1880         if (mddev->level == 1) {
1881                 if (rs->delta_disks)
1882                         return !!rs->delta_disks;
1883
1884                 return !change &&
1885                        mddev->raid_disks != rs->raid_disks;
1886         }
1887
1888         if (mddev->level == 10)
1889                 return change &&
1890                        !__is_raid10_far(mddev->new_layout) &&
1891                        rs->delta_disks >= 0;
1892
1893         return change;
1894 }
1895
1896 /*  Features */
1897 #define FEATURE_FLAG_SUPPORTS_V190      0x1 /* Supports extended superblock */
1898
1899 /* State flags for sb->flags */
1900 #define SB_FLAG_RESHAPE_ACTIVE          0x1
1901 #define SB_FLAG_RESHAPE_BACKWARDS       0x2
1902
1903 /*
1904  * This structure is never routinely used by userspace, unlike md superblocks.
1905  * Devices with this superblock should only ever be accessed via device-mapper.
1906  */
1907 #define DM_RAID_MAGIC 0x64526D44
1908 struct dm_raid_superblock {
1909         __le32 magic;           /* "DmRd" */
1910         __le32 compat_features; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1911
1912         __le32 num_devices;     /* Number of devices in this raid set. (Max 64) */
1913         __le32 array_position;  /* The position of this drive in the raid set */
1914
1915         __le64 events;          /* Incremented by md when superblock updated */
1916         __le64 failed_devices;  /* Pre 1.9.0 part of bit field of devices to */
1917                                 /* indicate failures (see extension below) */
1918
1919         /*
1920          * This offset tracks the progress of the repair or replacement of
1921          * an individual drive.
1922          */
1923         __le64 disk_recovery_offset;
1924
1925         /*
1926          * This offset tracks the progress of the initial raid set
1927          * synchronisation/parity calculation.
1928          */
1929         __le64 array_resync_offset;
1930
1931         /*
1932          * raid characteristics
1933          */
1934         __le32 level;
1935         __le32 layout;
1936         __le32 stripe_sectors;
1937
1938         /********************************************************************
1939          * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1940          *
1941          * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
1942          */
1943
1944         __le32 flags; /* Flags defining array states for reshaping */
1945
1946         /*
1947          * This offset tracks the progress of a raid
1948          * set reshape in order to be able to restart it
1949          */
1950         __le64 reshape_position;
1951
1952         /*
1953          * These define the properties of the array in case of an interrupted reshape
1954          */
1955         __le32 new_level;
1956         __le32 new_layout;
1957         __le32 new_stripe_sectors;
1958         __le32 delta_disks;
1959
1960         __le64 array_sectors; /* Array size in sectors */
1961
1962         /*
1963          * Sector offsets to data on devices (reshaping).
1964          * Needed to support out of place reshaping, thus
1965          * not writing over any stripes whilst converting
1966          * them from old to new layout
1967          */
1968         __le64 data_offset;
1969         __le64 new_data_offset;
1970
1971         __le64 sectors; /* Used device size in sectors */
1972
1973         /*
1974          * Additonal Bit field of devices indicating failures to support
1975          * up to 256 devices with the 1.9.0 on-disk metadata format
1976          */
1977         __le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
1978
1979         __le32 incompat_features;       /* Used to indicate any incompatible features */
1980
1981         /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
1982 } __packed;
1983
1984 /*
1985  * Check for reshape constraints on raid set @rs:
1986  *
1987  * - reshape function non-existent
1988  * - degraded set
1989  * - ongoing recovery
1990  * - ongoing reshape
1991  *
1992  * Returns 0 if none or -EPERM if given constraint
1993  * and error message reference in @errmsg
1994  */
1995 static int rs_check_reshape(struct raid_set *rs)
1996 {
1997         struct mddev *mddev = &rs->md;
1998
1999         if (!mddev->pers || !mddev->pers->check_reshape)
2000                 rs->ti->error = "Reshape not supported";
2001         else if (mddev->degraded)
2002                 rs->ti->error = "Can't reshape degraded raid set";
2003         else if (rs_is_recovering(rs))
2004                 rs->ti->error = "Convert request on recovering raid set prohibited";
2005         else if (rs_is_reshaping(rs))
2006                 rs->ti->error = "raid set already reshaping!";
2007         else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
2008                 rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
2009         else
2010                 return 0;
2011
2012         return -EPERM;
2013 }
2014
2015 static int read_disk_sb(struct md_rdev *rdev, int size, bool force_reload)
2016 {
2017         BUG_ON(!rdev->sb_page);
2018
2019         if (rdev->sb_loaded && !force_reload)
2020                 return 0;
2021
2022         rdev->sb_loaded = 0;
2023
2024         if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) {
2025                 DMERR("Failed to read superblock of device at position %d",
2026                       rdev->raid_disk);
2027                 md_error(rdev->mddev, rdev);
2028                 set_bit(Faulty, &rdev->flags);
2029                 return -EIO;
2030         }
2031
2032         rdev->sb_loaded = 1;
2033
2034         return 0;
2035 }
2036
2037 static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2038 {
2039         failed_devices[0] = le64_to_cpu(sb->failed_devices);
2040         memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
2041
2042         if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2043                 int i = ARRAY_SIZE(sb->extended_failed_devices);
2044
2045                 while (i--)
2046                         failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
2047         }
2048 }
2049
2050 static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2051 {
2052         int i = ARRAY_SIZE(sb->extended_failed_devices);
2053
2054         sb->failed_devices = cpu_to_le64(failed_devices[0]);
2055         while (i--)
2056                 sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
2057 }
2058
2059 /*
2060  * Synchronize the superblock members with the raid set properties
2061  *
2062  * All superblock data is little endian.
2063  */
2064 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
2065 {
2066         bool update_failed_devices = false;
2067         unsigned int i;
2068         uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2069         struct dm_raid_superblock *sb;
2070         struct raid_set *rs = container_of(mddev, struct raid_set, md);
2071
2072         /* No metadata device, no superblock */
2073         if (!rdev->meta_bdev)
2074                 return;
2075
2076         BUG_ON(!rdev->sb_page);
2077
2078         sb = page_address(rdev->sb_page);
2079
2080         sb_retrieve_failed_devices(sb, failed_devices);
2081
2082         for (i = 0; i < rs->raid_disks; i++)
2083                 if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
2084                         update_failed_devices = true;
2085                         set_bit(i, (void *) failed_devices);
2086                 }
2087
2088         if (update_failed_devices)
2089                 sb_update_failed_devices(sb, failed_devices);
2090
2091         sb->magic = cpu_to_le32(DM_RAID_MAGIC);
2092         sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2093
2094         sb->num_devices = cpu_to_le32(mddev->raid_disks);
2095         sb->array_position = cpu_to_le32(rdev->raid_disk);
2096
2097         sb->events = cpu_to_le64(mddev->events);
2098
2099         sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
2100         sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
2101
2102         sb->level = cpu_to_le32(mddev->level);
2103         sb->layout = cpu_to_le32(mddev->layout);
2104         sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
2105
2106         /********************************************************************
2107          * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
2108          *
2109          * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
2110          */
2111         sb->new_level = cpu_to_le32(mddev->new_level);
2112         sb->new_layout = cpu_to_le32(mddev->new_layout);
2113         sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
2114
2115         sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2116
2117         smp_rmb(); /* Make sure we access most recent reshape position */
2118         sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2119         if (le64_to_cpu(sb->reshape_position) != MaxSector) {
2120                 /* Flag ongoing reshape */
2121                 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
2122
2123                 if (mddev->delta_disks < 0 || mddev->reshape_backwards)
2124                         sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
2125         } else {
2126                 /* Clear reshape flags */
2127                 sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
2128         }
2129
2130         sb->array_sectors = cpu_to_le64(mddev->array_sectors);
2131         sb->data_offset = cpu_to_le64(rdev->data_offset);
2132         sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
2133         sb->sectors = cpu_to_le64(rdev->sectors);
2134         sb->incompat_features = cpu_to_le32(0);
2135
2136         /* Zero out the rest of the payload after the size of the superblock */
2137         memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
2138 }
2139
2140 /*
2141  * super_load
2142  *
2143  * This function creates a superblock if one is not found on the device
2144  * and will decide which superblock to use if there's a choice.
2145  *
2146  * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
2147  */
2148 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
2149 {
2150         int r;
2151         struct dm_raid_superblock *sb;
2152         struct dm_raid_superblock *refsb;
2153         uint64_t events_sb, events_refsb;
2154
2155         r = read_disk_sb(rdev, rdev->sb_size, false);
2156         if (r)
2157                 return r;
2158
2159         sb = page_address(rdev->sb_page);
2160
2161         /*
2162          * Two cases that we want to write new superblocks and rebuild:
2163          * 1) New device (no matching magic number)
2164          * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2165          */
2166         if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2167             (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
2168                 super_sync(rdev->mddev, rdev);
2169
2170                 set_bit(FirstUse, &rdev->flags);
2171                 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2172
2173                 /* Force writing of superblocks to disk */
2174                 set_bit(MD_SB_CHANGE_DEVS, &rdev->mddev->sb_flags);
2175
2176                 /* Any superblock is better than none, choose that if given */
2177                 return refdev ? 0 : 1;
2178         }
2179
2180         if (!refdev)
2181                 return 1;
2182
2183         events_sb = le64_to_cpu(sb->events);
2184
2185         refsb = page_address(refdev->sb_page);
2186         events_refsb = le64_to_cpu(refsb->events);
2187
2188         return (events_sb > events_refsb) ? 1 : 0;
2189 }
2190
2191 static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
2192 {
2193         int role;
2194         unsigned int d;
2195         struct mddev *mddev = &rs->md;
2196         uint64_t events_sb;
2197         uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2198         struct dm_raid_superblock *sb;
2199         uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
2200         struct md_rdev *r;
2201         struct dm_raid_superblock *sb2;
2202
2203         sb = page_address(rdev->sb_page);
2204         events_sb = le64_to_cpu(sb->events);
2205
2206         /*
2207          * Initialise to 1 if this is a new superblock.
2208          */
2209         mddev->events = events_sb ? : 1;
2210
2211         mddev->reshape_position = MaxSector;
2212
2213         mddev->raid_disks = le32_to_cpu(sb->num_devices);
2214         mddev->level = le32_to_cpu(sb->level);
2215         mddev->layout = le32_to_cpu(sb->layout);
2216         mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2217
2218         /*
2219          * Reshaping is supported, e.g. reshape_position is valid
2220          * in superblock and superblock content is authoritative.
2221          */
2222         if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2223                 /* Superblock is authoritative wrt given raid set layout! */
2224                 mddev->new_level = le32_to_cpu(sb->new_level);
2225                 mddev->new_layout = le32_to_cpu(sb->new_layout);
2226                 mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2227                 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2228                 mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2229
2230                 /* raid was reshaping and got interrupted */
2231                 if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2232                         if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
2233                                 DMERR("Reshape requested but raid set is still reshaping");
2234                                 return -EINVAL;
2235                         }
2236
2237                         if (mddev->delta_disks < 0 ||
2238                             (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
2239                                 mddev->reshape_backwards = 1;
2240                         else
2241                                 mddev->reshape_backwards = 0;
2242
2243                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2244                         rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2245                 }
2246
2247         } else {
2248                 /*
2249                  * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2250                  */
2251                 struct raid_type *rt_cur = get_raid_type_by_ll(mddev->level, mddev->layout);
2252                 struct raid_type *rt_new = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
2253
2254                 if (rs_takeover_requested(rs)) {
2255                         if (rt_cur && rt_new)
2256                                 DMERR("Takeover raid sets from %s to %s not yet supported by metadata. (raid level change)",
2257                                       rt_cur->name, rt_new->name);
2258                         else
2259                                 DMERR("Takeover raid sets not yet supported by metadata. (raid level change)");
2260                         return -EINVAL;
2261                 } else if (rs_reshape_requested(rs)) {
2262                         DMERR("Reshaping raid sets not yet supported by metadata. (raid layout change keeping level)");
2263                         if (mddev->layout != mddev->new_layout) {
2264                                 if (rt_cur && rt_new)
2265                                         DMERR("  current layout %s vs new layout %s",
2266                                               rt_cur->name, rt_new->name);
2267                                 else
2268                                         DMERR("  current layout 0x%X vs new layout 0x%X",
2269                                               le32_to_cpu(sb->layout), mddev->new_layout);
2270                         }
2271                         if (mddev->chunk_sectors != mddev->new_chunk_sectors)
2272                                 DMERR("  current stripe sectors %u vs new stripe sectors %u",
2273                                       mddev->chunk_sectors, mddev->new_chunk_sectors);
2274                         if (rs->delta_disks)
2275                                 DMERR("  current %u disks vs new %u disks",
2276                                       mddev->raid_disks, mddev->raid_disks + rs->delta_disks);
2277                         if (rs_is_raid10(rs)) {
2278                                 DMERR("  Old layout: %s w/ %u copies",
2279                                       raid10_md_layout_to_format(mddev->layout),
2280                                       raid10_md_layout_to_copies(mddev->layout));
2281                                 DMERR("  New layout: %s w/ %u copies",
2282                                       raid10_md_layout_to_format(mddev->new_layout),
2283                                       raid10_md_layout_to_copies(mddev->new_layout));
2284                         }
2285                         return -EINVAL;
2286                 }
2287
2288                 DMINFO("Discovered old metadata format; upgrading to extended metadata format");
2289         }
2290
2291         if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2292                 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2293
2294         /*
2295          * During load, we set FirstUse if a new superblock was written.
2296          * There are two reasons we might not have a superblock:
2297          * 1) The raid set is brand new - in which case, all of the
2298          *    devices must have their In_sync bit set.  Also,
2299          *    recovery_cp must be 0, unless forced.
2300          * 2) This is a new device being added to an old raid set
2301          *    and the new device needs to be rebuilt - in which
2302          *    case the In_sync bit will /not/ be set and
2303          *    recovery_cp must be MaxSector.
2304          * 3) This is/are a new device(s) being added to an old
2305          *    raid set during takeover to a higher raid level
2306          *    to provide capacity for redundancy or during reshape
2307          *    to add capacity to grow the raid set.
2308          */
2309         d = 0;
2310         rdev_for_each(r, mddev) {
2311                 if (test_bit(Journal, &rdev->flags))
2312                         continue;
2313
2314                 if (test_bit(FirstUse, &r->flags))
2315                         new_devs++;
2316
2317                 if (!test_bit(In_sync, &r->flags)) {
2318                         DMINFO("Device %d specified for rebuild; clearing superblock",
2319                                 r->raid_disk);
2320                         rebuilds++;
2321
2322                         if (test_bit(FirstUse, &r->flags))
2323                                 rebuild_and_new++;
2324                 }
2325
2326                 d++;
2327         }
2328
2329         if (new_devs == rs->raid_disks || !rebuilds) {
2330                 /* Replace a broken device */
2331                 if (new_devs == 1 && !rs->delta_disks)
2332                         ;
2333                 if (new_devs == rs->raid_disks) {
2334                         DMINFO("Superblocks created for new raid set");
2335                         set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2336                 } else if (new_devs != rebuilds &&
2337                            new_devs != rs->delta_disks) {
2338                         DMERR("New device injected into existing raid set without "
2339                               "'delta_disks' or 'rebuild' parameter specified");
2340                         return -EINVAL;
2341                 }
2342         } else if (new_devs && new_devs != rebuilds) {
2343                 DMERR("%u 'rebuild' devices cannot be injected into"
2344                       " a raid set with %u other first-time devices",
2345                       rebuilds, new_devs);
2346                 return -EINVAL;
2347         } else if (rebuilds) {
2348                 if (rebuild_and_new && rebuilds != rebuild_and_new) {
2349                         DMERR("new device%s provided without 'rebuild'",
2350                               new_devs > 1 ? "s" : "");
2351                         return -EINVAL;
2352                 } else if (rs_is_recovering(rs)) {
2353                         DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2354                               (unsigned long long) mddev->recovery_cp);
2355                         return -EINVAL;
2356                 } else if (rs_is_reshaping(rs)) {
2357                         DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2358                               (unsigned long long) mddev->reshape_position);
2359                         return -EINVAL;
2360                 }
2361         }
2362
2363         /*
2364          * Now we set the Faulty bit for those devices that are
2365          * recorded in the superblock as failed.
2366          */
2367         sb_retrieve_failed_devices(sb, failed_devices);
2368         rdev_for_each(r, mddev) {
2369                 if (test_bit(Journal, &rdev->flags) ||
2370                     !r->sb_page)
2371                         continue;
2372                 sb2 = page_address(r->sb_page);
2373                 sb2->failed_devices = 0;
2374                 memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2375
2376                 /*
2377                  * Check for any device re-ordering.
2378                  */
2379                 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2380                         role = le32_to_cpu(sb2->array_position);
2381                         if (role < 0)
2382                                 continue;
2383
2384                         if (role != r->raid_disk) {
2385                                 if (rs_is_raid10(rs) && __is_raid10_near(mddev->layout)) {
2386                                         if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2387                                             rs->raid_disks % rs->raid10_copies) {
2388                                                 rs->ti->error =
2389                                                         "Cannot change raid10 near set to odd # of devices!";
2390                                                 return -EINVAL;
2391                                         }
2392
2393                                         sb2->array_position = cpu_to_le32(r->raid_disk);
2394
2395                                 } else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2396                                            !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2397                                            !rt_is_raid1(rs->raid_type)) {
2398                                         rs->ti->error = "Cannot change device positions in raid set";
2399                                         return -EINVAL;
2400                                 }
2401
2402                                 DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2403                         }
2404
2405                         /*
2406                          * Partial recovery is performed on
2407                          * returning failed devices.
2408                          */
2409                         if (test_bit(role, (void *) failed_devices))
2410                                 set_bit(Faulty, &r->flags);
2411                 }
2412         }
2413
2414         return 0;
2415 }
2416
2417 static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2418 {
2419         struct mddev *mddev = &rs->md;
2420         struct dm_raid_superblock *sb;
2421
2422         if (rs_is_raid0(rs) || !rdev->sb_page || rdev->raid_disk < 0)
2423                 return 0;
2424
2425         sb = page_address(rdev->sb_page);
2426
2427         /*
2428          * If mddev->events is not set, we know we have not yet initialized
2429          * the array.
2430          */
2431         if (!mddev->events && super_init_validation(rs, rdev))
2432                 return -EINVAL;
2433
2434         if (le32_to_cpu(sb->compat_features) &&
2435             le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2436                 rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2437                 return -EINVAL;
2438         }
2439
2440         if (sb->incompat_features) {
2441                 rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2442                 return -EINVAL;
2443         }
2444
2445         /* Enable bitmap creation for RAID levels != 0 */
2446         mddev->bitmap_info.offset = (rt_is_raid0(rs->raid_type) || rs->journal_dev.dev) ? 0 : to_sector(4096);
2447         mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2448
2449         if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2450                 /*
2451                  * Retrieve rdev size stored in superblock to be prepared for shrink.
2452                  * Check extended superblock members are present otherwise the size
2453                  * will not be set!
2454                  */
2455                 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190)
2456                         rdev->sectors = le64_to_cpu(sb->sectors);
2457
2458                 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2459                 if (rdev->recovery_offset == MaxSector)
2460                         set_bit(In_sync, &rdev->flags);
2461                 /*
2462                  * If no reshape in progress -> we're recovering single
2463                  * disk(s) and have to set the device(s) to out-of-sync
2464                  */
2465                 else if (!rs_is_reshaping(rs))
2466                         clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2467         }
2468
2469         /*
2470          * If a device comes back, set it as not In_sync and no longer faulty.
2471          */
2472         if (test_and_clear_bit(Faulty, &rdev->flags)) {
2473                 rdev->recovery_offset = 0;
2474                 clear_bit(In_sync, &rdev->flags);
2475                 rdev->saved_raid_disk = rdev->raid_disk;
2476         }
2477
2478         /* Reshape support -> restore repective data offsets */
2479         rdev->data_offset = le64_to_cpu(sb->data_offset);
2480         rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2481
2482         return 0;
2483 }
2484
2485 /*
2486  * Analyse superblocks and select the freshest.
2487  */
2488 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2489 {
2490         int r;
2491         struct md_rdev *rdev, *freshest;
2492         struct mddev *mddev = &rs->md;
2493
2494         freshest = NULL;
2495         rdev_for_each(rdev, mddev) {
2496                 if (test_bit(Journal, &rdev->flags))
2497                         continue;
2498
2499                 if (!rdev->meta_bdev)
2500                         continue;
2501
2502                 /* Set superblock offset/size for metadata device. */
2503                 rdev->sb_start = 0;
2504                 rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
2505                 if (rdev->sb_size < sizeof(struct dm_raid_superblock) || rdev->sb_size > PAGE_SIZE) {
2506                         DMERR("superblock size of a logical block is no longer valid");
2507                         return -EINVAL;
2508                 }
2509
2510                 /*
2511                  * Skipping super_load due to CTR_FLAG_SYNC will cause
2512                  * the array to undergo initialization again as
2513                  * though it were new.  This is the intended effect
2514                  * of the "sync" directive.
2515                  *
2516                  * With reshaping capability added, we must ensure that
2517                  * that the "sync" directive is disallowed during the reshape.
2518                  */
2519                 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2520                         continue;
2521
2522                 r = super_load(rdev, freshest);
2523
2524                 switch (r) {
2525                 case 1:
2526                         freshest = rdev;
2527                         break;
2528                 case 0:
2529                         break;
2530                 default:
2531                         /* This is a failure to read the superblock from the metadata device. */
2532                         /*
2533                          * We have to keep any raid0 data/metadata device pairs or
2534                          * the MD raid0 personality will fail to start the array.
2535                          */
2536                         if (rs_is_raid0(rs))
2537                                 continue;
2538
2539                         /*
2540                          * We keep the dm_devs to be able to emit the device tuple
2541                          * properly on the table line in raid_status() (rather than
2542                          * mistakenly acting as if '- -' got passed into the constructor).
2543                          *
2544                          * The rdev has to stay on the same_set list to allow for
2545                          * the attempt to restore faulty devices on second resume.
2546                          */
2547                         rdev->raid_disk = rdev->saved_raid_disk = -1;
2548                         break;
2549                 }
2550         }
2551
2552         if (!freshest)
2553                 return 0;
2554
2555         /*
2556          * Validation of the freshest device provides the source of
2557          * validation for the remaining devices.
2558          */
2559         rs->ti->error = "Unable to assemble array: Invalid superblocks";
2560         if (super_validate(rs, freshest))
2561                 return -EINVAL;
2562
2563         if (validate_raid_redundancy(rs)) {
2564                 rs->ti->error = "Insufficient redundancy to activate array";
2565                 return -EINVAL;
2566         }
2567
2568         rdev_for_each(rdev, mddev)
2569                 if (!test_bit(Journal, &rdev->flags) &&
2570                     rdev != freshest &&
2571                     super_validate(rs, rdev))
2572                         return -EINVAL;
2573         return 0;
2574 }
2575
2576 /*
2577  * Adjust data_offset and new_data_offset on all disk members of @rs
2578  * for out of place reshaping if requested by contructor
2579  *
2580  * We need free space at the beginning of each raid disk for forward
2581  * and at the end for backward reshapes which userspace has to provide
2582  * via remapping/reordering of space.
2583  */
2584 static int rs_adjust_data_offsets(struct raid_set *rs)
2585 {
2586         sector_t data_offset = 0, new_data_offset = 0;
2587         struct md_rdev *rdev;
2588
2589         /* Constructor did not request data offset change */
2590         if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2591                 if (!rs_is_reshapable(rs))
2592                         goto out;
2593
2594                 return 0;
2595         }
2596
2597         /* HM FIXME: get InSync raid_dev? */
2598         rdev = &rs->dev[0].rdev;
2599
2600         if (rs->delta_disks < 0) {
2601                 /*
2602                  * Removing disks (reshaping backwards):
2603                  *
2604                  * - before reshape: data is at offset 0 and free space
2605                  *                   is at end of each component LV
2606                  *
2607                  * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2608                  */
2609                 data_offset = 0;
2610                 new_data_offset = rs->data_offset;
2611
2612         } else if (rs->delta_disks > 0) {
2613                 /*
2614                  * Adding disks (reshaping forwards):
2615                  *
2616                  * - before reshape: data is at offset rs->data_offset != 0 and
2617                  *                   free space is at begin of each component LV
2618                  *
2619                  * - after reshape: data is at offset 0 on each component LV
2620                  */
2621                 data_offset = rs->data_offset;
2622                 new_data_offset = 0;
2623
2624         } else {
2625                 /*
2626                  * User space passes in 0 for data offset after having removed reshape space
2627                  *
2628                  * - or - (data offset != 0)
2629                  *
2630                  * Changing RAID layout or chunk size -> toggle offsets
2631                  *
2632                  * - before reshape: data is at offset rs->data_offset 0 and
2633                  *                   free space is at end of each component LV
2634                  *                   -or-
2635                  *                   data is at offset rs->data_offset != 0 and
2636                  *                   free space is at begin of each component LV
2637                  *
2638                  * - after reshape: data is at offset 0 if it was at offset != 0
2639                  *                  or at offset != 0 if it was at offset 0
2640                  *                  on each component LV
2641                  *
2642                  */
2643                 data_offset = rs->data_offset ? rdev->data_offset : 0;
2644                 new_data_offset = data_offset ? 0 : rs->data_offset;
2645                 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2646         }
2647
2648         /*
2649          * Make sure we got a minimum amount of free sectors per device
2650          */
2651         if (rs->data_offset &&
2652             to_sector(i_size_read(rdev->bdev->bd_inode)) - rdev->sectors < MIN_FREE_RESHAPE_SPACE) {
2653                 rs->ti->error = data_offset ? "No space for forward reshape" :
2654                                               "No space for backward reshape";
2655                 return -ENOSPC;
2656         }
2657 out:
2658         /* Adjust data offsets on all rdevs but on any raid4/5/6 journal device */
2659         rdev_for_each(rdev, &rs->md) {
2660                 if (!test_bit(Journal, &rdev->flags)) {
2661                         rdev->data_offset = data_offset;
2662                         rdev->new_data_offset = new_data_offset;
2663                 }
2664         }
2665
2666         return 0;
2667 }
2668
2669 /* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2670 static void __reorder_raid_disk_indexes(struct raid_set *rs)
2671 {
2672         int i = 0;
2673         struct md_rdev *rdev;
2674
2675         rdev_for_each(rdev, &rs->md) {
2676                 if (!test_bit(Journal, &rdev->flags)) {
2677                         rdev->raid_disk = i++;
2678                         rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2679                 }
2680         }
2681 }
2682
2683 /*
2684  * Setup @rs for takeover by a different raid level
2685  */
2686 static int rs_setup_takeover(struct raid_set *rs)
2687 {
2688         struct mddev *mddev = &rs->md;
2689         struct md_rdev *rdev;
2690         unsigned int d = mddev->raid_disks = rs->raid_disks;
2691         sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2692
2693         if (rt_is_raid10(rs->raid_type)) {
2694                 if (mddev->level == 0) {
2695                         /* Userpace reordered disks -> adjust raid_disk indexes */
2696                         __reorder_raid_disk_indexes(rs);
2697
2698                         /* raid0 -> raid10_far layout */
2699                         mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2700                                                                    rs->raid10_copies);
2701                 } else if (mddev->level == 1)
2702                         /* raid1 -> raid10_near layout */
2703                         mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2704                                                                    rs->raid_disks);
2705                 else
2706                         return -EINVAL;
2707
2708         }
2709
2710         clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2711         mddev->recovery_cp = MaxSector;
2712
2713         while (d--) {
2714                 rdev = &rs->dev[d].rdev;
2715
2716                 if (test_bit(d, (void *) rs->rebuild_disks)) {
2717                         clear_bit(In_sync, &rdev->flags);
2718                         clear_bit(Faulty, &rdev->flags);
2719                         mddev->recovery_cp = rdev->recovery_offset = 0;
2720                         /* Bitmap has to be created when we do an "up" takeover */
2721                         set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2722                 }
2723
2724                 rdev->new_data_offset = new_data_offset;
2725         }
2726
2727         return 0;
2728 }
2729
2730 /* Prepare @rs for reshape */
2731 static int rs_prepare_reshape(struct raid_set *rs)
2732 {
2733         bool reshape;
2734         struct mddev *mddev = &rs->md;
2735
2736         if (rs_is_raid10(rs)) {
2737                 if (rs->raid_disks != mddev->raid_disks &&
2738                     __is_raid10_near(mddev->layout) &&
2739                     rs->raid10_copies &&
2740                     rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2741                         /*
2742                          * raid disk have to be multiple of data copies to allow this conversion,
2743                          *
2744                          * This is actually not a reshape it is a
2745                          * rebuild of any additional mirrors per group
2746                          */
2747                         if (rs->raid_disks % rs->raid10_copies) {
2748                                 rs->ti->error = "Can't reshape raid10 mirror groups";
2749                                 return -EINVAL;
2750                         }
2751
2752                         /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2753                         __reorder_raid_disk_indexes(rs);
2754                         mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2755                                                                    rs->raid10_copies);
2756                         mddev->new_layout = mddev->layout;
2757                         reshape = false;
2758                 } else
2759                         reshape = true;
2760
2761         } else if (rs_is_raid456(rs))
2762                 reshape = true;
2763
2764         else if (rs_is_raid1(rs)) {
2765                 if (rs->delta_disks) {
2766                         /* Process raid1 via delta_disks */
2767                         mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2768                         reshape = true;
2769                 } else {
2770                         /* Process raid1 without delta_disks */
2771                         mddev->raid_disks = rs->raid_disks;
2772                         reshape = false;
2773                 }
2774         } else {
2775                 rs->ti->error = "Called with bogus raid type";
2776                 return -EINVAL;
2777         }
2778
2779         if (reshape) {
2780                 set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2781                 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2782         } else if (mddev->raid_disks < rs->raid_disks)
2783                 /* Create new superblocks and bitmaps, if any new disks */
2784                 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2785
2786         return 0;
2787 }
2788
2789 /*
2790  *
2791  * - change raid layout
2792  * - change chunk size
2793  * - add disks
2794  * - remove disks
2795  */
2796 static int rs_setup_reshape(struct raid_set *rs)
2797 {
2798         int r = 0;
2799         unsigned int cur_raid_devs, d;
2800         struct mddev *mddev = &rs->md;
2801         struct md_rdev *rdev;
2802
2803         mddev->delta_disks = rs->delta_disks;
2804         cur_raid_devs = mddev->raid_disks;
2805
2806         /* Ignore impossible layout change whilst adding/removing disks */
2807         if (mddev->delta_disks &&
2808             mddev->layout != mddev->new_layout) {
2809                 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2810                 mddev->new_layout = mddev->layout;
2811         }
2812
2813         /*
2814          * Adjust array size:
2815          *
2816          * - in case of adding disks, array size has
2817          *   to grow after the disk adding reshape,
2818          *   which'll hapen in the event handler;
2819          *   reshape will happen forward, so space has to
2820          *   be available at the beginning of each disk
2821          *
2822          * - in case of removing disks, array size
2823          *   has to shrink before starting the reshape,
2824          *   which'll happen here;
2825          *   reshape will happen backward, so space has to
2826          *   be available at the end of each disk
2827          *
2828          * - data_offset and new_data_offset are
2829          *   adjusted for aforementioned out of place
2830          *   reshaping based on userspace passing in
2831          *   the "data_offset <sectors>" key/value
2832          *   pair via the constructor
2833          */
2834
2835         /* Add disk(s) */
2836         if (rs->delta_disks > 0) {
2837                 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2838                 for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2839                         rdev = &rs->dev[d].rdev;
2840                         clear_bit(In_sync, &rdev->flags);
2841
2842                         /*
2843                          * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2844                          * by md, which'll store that erroneously in the superblock on reshape
2845                          */
2846                         rdev->saved_raid_disk = -1;
2847                         rdev->raid_disk = d;
2848
2849                         rdev->sectors = mddev->dev_sectors;
2850                         rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
2851                 }
2852
2853                 mddev->reshape_backwards = 0; /* adding disks -> forward reshape */
2854
2855         /* Remove disk(s) */
2856         } else if (rs->delta_disks < 0) {
2857                 r = rs_set_dev_and_array_sectors(rs, true);
2858                 mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2859
2860         /* Change layout and/or chunk size */
2861         } else {
2862                 /*
2863                  * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2864                  *
2865                  * keeping number of disks and do layout change ->
2866                  *
2867                  * toggle reshape_backward depending on data_offset:
2868                  *
2869                  * - free space upfront -> reshape forward
2870                  *
2871                  * - free space at the end -> reshape backward
2872                  *
2873                  *
2874                  * This utilizes free reshape space avoiding the need
2875                  * for userspace to move (parts of) LV segments in
2876                  * case of layout/chunksize change  (for disk
2877                  * adding/removing reshape space has to be at
2878                  * the proper address (see above with delta_disks):
2879                  *
2880                  * add disk(s)   -> begin
2881                  * remove disk(s)-> end
2882                  */
2883                 mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2884         }
2885
2886         return r;
2887 }
2888
2889 /*
2890  * Enable/disable discard support on RAID set depending on
2891  * RAID level and discard properties of underlying RAID members.
2892  */
2893 static void configure_discard_support(struct raid_set *rs)
2894 {
2895         int i;
2896         bool raid456;
2897         struct dm_target *ti = rs->ti;
2898
2899         /* Assume discards not supported until after checks below. */
2900         ti->discards_supported = false;
2901
2902         /*
2903          * XXX: RAID level 4,5,6 require zeroing for safety.
2904          */
2905         raid456 = (rs->md.level == 4 || rs->md.level == 5 || rs->md.level == 6);
2906
2907         for (i = 0; i < rs->raid_disks; i++) {
2908                 struct request_queue *q;
2909
2910                 if (!rs->dev[i].rdev.bdev)
2911                         continue;
2912
2913                 q = bdev_get_queue(rs->dev[i].rdev.bdev);
2914                 if (!q || !blk_queue_discard(q))
2915                         return;
2916
2917                 if (raid456) {
2918                         if (!devices_handle_discard_safely) {
2919                                 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2920                                 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2921                                 return;
2922                         }
2923                 }
2924         }
2925
2926         /* All RAID members properly support discards */
2927         ti->discards_supported = true;
2928
2929         /*
2930          * RAID1 and RAID10 personalities require bio splitting,
2931          * RAID0/4/5/6 don't and process large discard bios properly.
2932          */
2933         ti->split_discard_bios = !!(rs->md.level == 1 || rs->md.level == 10);
2934         ti->num_discard_bios = 1;
2935 }
2936
2937 /*
2938  * Construct a RAID0/1/10/4/5/6 mapping:
2939  * Args:
2940  *      <raid_type> <#raid_params> <raid_params>{0,}    \
2941  *      <#raid_devs> [<meta_dev1> <dev1>]{1,}
2942  *
2943  * <raid_params> varies by <raid_type>.  See 'parse_raid_params' for
2944  * details on possible <raid_params>.
2945  *
2946  * Userspace is free to initialize the metadata devices, hence the superblocks to
2947  * enforce recreation based on the passed in table parameters.
2948  *
2949  */
2950 static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2951 {
2952         int r;
2953         bool resize;
2954         struct raid_type *rt;
2955         unsigned int num_raid_params, num_raid_devs;
2956         sector_t calculated_dev_sectors, rdev_sectors;
2957         struct raid_set *rs = NULL;
2958         const char *arg;
2959         struct rs_layout rs_layout;
2960         struct dm_arg_set as = { argc, argv }, as_nrd;
2961         struct dm_arg _args[] = {
2962                 { 0, as.argc, "Cannot understand number of raid parameters" },
2963                 { 1, 254, "Cannot understand number of raid devices parameters" }
2964         };
2965
2966         /* Must have <raid_type> */
2967         arg = dm_shift_arg(&as);
2968         if (!arg) {
2969                 ti->error = "No arguments";
2970                 return -EINVAL;
2971         }
2972
2973         rt = get_raid_type(arg);
2974         if (!rt) {
2975                 ti->error = "Unrecognised raid_type";
2976                 return -EINVAL;
2977         }
2978
2979         /* Must have <#raid_params> */
2980         if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
2981                 return -EINVAL;
2982
2983         /* number of raid device tupples <meta_dev data_dev> */
2984         as_nrd = as;
2985         dm_consume_args(&as_nrd, num_raid_params);
2986         _args[1].max = (as_nrd.argc - 1) / 2;
2987         if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
2988                 return -EINVAL;
2989
2990         if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
2991                 ti->error = "Invalid number of supplied raid devices";
2992                 return -EINVAL;
2993         }
2994
2995         rs = raid_set_alloc(ti, rt, num_raid_devs);
2996         if (IS_ERR(rs))
2997                 return PTR_ERR(rs);
2998
2999         r = parse_raid_params(rs, &as, num_raid_params);
3000         if (r)
3001                 goto bad;
3002
3003         r = parse_dev_params(rs, &as);
3004         if (r)
3005                 goto bad;
3006
3007         rs->md.sync_super = super_sync;
3008
3009         /*
3010          * Calculate ctr requested array and device sizes to allow
3011          * for superblock analysis needing device sizes defined.
3012          *
3013          * Any existing superblock will overwrite the array and device sizes
3014          */
3015         r = rs_set_dev_and_array_sectors(rs, false);
3016         if (r)
3017                 goto bad;
3018
3019         calculated_dev_sectors = rs->md.dev_sectors;
3020
3021         /*
3022          * Backup any new raid set level, layout, ...
3023          * requested to be able to compare to superblock
3024          * members for conversion decisions.
3025          */
3026         rs_config_backup(rs, &rs_layout);
3027
3028         r = analyse_superblocks(ti, rs);
3029         if (r)
3030                 goto bad;
3031
3032         rdev_sectors = __rdev_sectors(rs);
3033         if (!rdev_sectors) {
3034                 ti->error = "Invalid rdev size";
3035                 r = -EINVAL;
3036                 goto bad;
3037         }
3038
3039         resize = calculated_dev_sectors != rdev_sectors;
3040
3041         INIT_WORK(&rs->md.event_work, do_table_event);
3042         ti->private = rs;
3043         ti->num_flush_bios = 1;
3044
3045         /* Restore any requested new layout for conversion decision */
3046         rs_config_restore(rs, &rs_layout);
3047
3048         /*
3049          * Now that we have any superblock metadata available,
3050          * check for new, recovering, reshaping, to be taken over,
3051          * to be reshaped or an existing, unchanged raid set to
3052          * run in sequence.
3053          */
3054         if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
3055                 /* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
3056                 if (rs_is_raid6(rs) &&
3057                     test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
3058                         ti->error = "'nosync' not allowed for new raid6 set";
3059                         r = -EINVAL;
3060                         goto bad;
3061                 }
3062                 rs_setup_recovery(rs, 0);
3063                 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3064                 rs_set_new(rs);
3065         } else if (rs_is_recovering(rs)) {
3066                 /* Rebuild particular devices */
3067                 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3068                         set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3069                         rs_setup_recovery(rs, MaxSector);
3070                 }
3071                 /* A recovering raid set may be resized */
3072                 ; /* skip setup rs */
3073         } else if (rs_is_reshaping(rs)) {
3074                 /* Have to reject size change request during reshape */
3075                 if (resize) {
3076                         ti->error = "Can't resize a reshaping raid set";
3077                         r = -EPERM;
3078                         goto bad;
3079                 }
3080                 /* skip setup rs */
3081         } else if (rs_takeover_requested(rs)) {
3082                 if (rs_is_reshaping(rs)) {
3083                         ti->error = "Can't takeover a reshaping raid set";
3084                         r = -EPERM;
3085                         goto bad;
3086                 }
3087
3088                 /* We can't takeover a journaled raid4/5/6 */
3089                 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3090                         ti->error = "Can't takeover a journaled raid4/5/6 set";
3091                         r = -EPERM;
3092                         goto bad;
3093                 }
3094
3095                 /*
3096                  * If a takeover is needed, userspace sets any additional
3097                  * devices to rebuild and we can check for a valid request here.
3098                  *
3099                  * If acceptible, set the level to the new requested
3100                  * one, prohibit requesting recovery, allow the raid
3101                  * set to run and store superblocks during resume.
3102                  */
3103                 r = rs_check_takeover(rs);
3104                 if (r)
3105                         goto bad;
3106
3107                 r = rs_setup_takeover(rs);
3108                 if (r)
3109                         goto bad;
3110
3111                 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3112                 /* Takeover ain't recovery, so disable recovery */
3113                 rs_setup_recovery(rs, MaxSector);
3114                 rs_set_new(rs);
3115         } else if (rs_reshape_requested(rs)) {
3116                 /*
3117                  * No need to check for 'ongoing' takeover here, because takeover
3118                  * is an instant operation as oposed to an ongoing reshape.
3119                  */
3120
3121                 /* We can't reshape a journaled raid4/5/6 */
3122                 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3123                         ti->error = "Can't reshape a journaled raid4/5/6 set";
3124                         r = -EPERM;
3125                         goto bad;
3126                 }
3127
3128                 /*
3129                   * We can only prepare for a reshape here, because the
3130                   * raid set needs to run to provide the repective reshape
3131                   * check functions via its MD personality instance.
3132                   *
3133                   * So do the reshape check after md_run() succeeded.
3134                   */
3135                 r = rs_prepare_reshape(rs);
3136                 if (r)
3137                         return r;
3138
3139                 /* Reshaping ain't recovery, so disable recovery */
3140                 rs_setup_recovery(rs, MaxSector);
3141                 rs_set_cur(rs);
3142         } else {
3143                 /* May not set recovery when a device rebuild is requested */
3144                 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3145                         rs_setup_recovery(rs, MaxSector);
3146                         set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3147                 } else
3148                         rs_setup_recovery(rs, test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ?
3149                                               0 : (resize ? calculated_dev_sectors : MaxSector));
3150                 rs_set_cur(rs);
3151         }
3152
3153         /* If constructor requested it, change data and new_data offsets */
3154         r = rs_adjust_data_offsets(rs);
3155         if (r)
3156                 goto bad;
3157
3158         /* Start raid set read-only and assumed clean to change in raid_resume() */
3159         rs->md.ro = 1;
3160         rs->md.in_sync = 1;
3161         set_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
3162
3163         /* Has to be held on running the array */
3164         mddev_lock_nointr(&rs->md);
3165         r = md_run(&rs->md);
3166         rs->md.in_sync = 0; /* Assume already marked dirty */
3167
3168         if (r) {
3169                 ti->error = "Failed to run raid array";
3170                 mddev_unlock(&rs->md);
3171                 goto bad;
3172         }
3173
3174         rs->callbacks.congested_fn = raid_is_congested;
3175         dm_table_add_target_callbacks(ti->table, &rs->callbacks);
3176
3177         /* If raid4/5/6 journal mode explictely requested (only possible with journal dev) -> set it */
3178         if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
3179                 r = r5c_journal_mode_set(&rs->md, rs->journal_dev.mode);
3180                 if (r) {
3181                         ti->error = "Failed to set raid4/5/6 journal mode";
3182                         mddev_unlock(&rs->md);
3183                         goto bad_journal_mode_set;
3184                 }
3185         }
3186
3187         mddev_suspend(&rs->md);
3188         set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags);
3189
3190         /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
3191         if (rs_is_raid456(rs)) {
3192                 r = rs_set_raid456_stripe_cache(rs);
3193                 if (r)
3194                         goto bad_stripe_cache;
3195         }
3196
3197         /* Now do an early reshape check */
3198         if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3199                 r = rs_check_reshape(rs);
3200                 if (r)
3201                         goto bad_check_reshape;
3202
3203                 /* Restore new, ctr requested layout to perform check */
3204                 rs_config_restore(rs, &rs_layout);
3205
3206                 if (rs->md.pers->start_reshape) {
3207                         r = rs->md.pers->check_reshape(&rs->md);
3208                         if (r) {
3209                                 ti->error = "Reshape check failed";
3210                                 goto bad_check_reshape;
3211                         }
3212                 }
3213         }
3214
3215         /* Disable/enable discard support on raid set. */
3216         configure_discard_support(rs);
3217
3218         mddev_unlock(&rs->md);
3219         return 0;
3220
3221 bad_journal_mode_set:
3222 bad_stripe_cache:
3223 bad_check_reshape:
3224         md_stop(&rs->md);
3225 bad:
3226         raid_set_free(rs);
3227
3228         return r;
3229 }
3230
3231 static void raid_dtr(struct dm_target *ti)
3232 {
3233         struct raid_set *rs = ti->private;
3234
3235         list_del_init(&rs->callbacks.list);
3236         md_stop(&rs->md);
3237         raid_set_free(rs);
3238 }
3239
3240 static int raid_map(struct dm_target *ti, struct bio *bio)
3241 {
3242         struct raid_set *rs = ti->private;
3243         struct mddev *mddev = &rs->md;
3244
3245         /*
3246          * If we're reshaping to add disk(s)), ti->len and
3247          * mddev->array_sectors will differ during the process
3248          * (ti->len > mddev->array_sectors), so we have to requeue
3249          * bios with addresses > mddev->array_sectors here or
3250          * there will occur accesses past EOD of the component
3251          * data images thus erroring the raid set.
3252          */
3253         if (unlikely(bio_end_sector(bio) > mddev->array_sectors))
3254                 return DM_MAPIO_REQUEUE;
3255
3256         md_handle_request(mddev, bio);
3257
3258         return DM_MAPIO_SUBMITTED;
3259 }
3260
3261 /* Return string describing the current sync action of @mddev */
3262 static const char *decipher_sync_action(struct mddev *mddev)
3263 {
3264         if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3265                 return "frozen";
3266
3267         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3268             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3269                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3270                         return "reshape";
3271
3272                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3273                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3274                                 return "resync";
3275                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3276                                 return "check";
3277                         return "repair";
3278                 }
3279
3280                 if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3281                         return "recover";
3282         }
3283
3284         return "idle";
3285 }
3286
3287 /*
3288  * Return status string for @rdev
3289  *
3290  * Status characters:
3291  *
3292  *  'D' = Dead/Failed raid set component or raid4/5/6 journal device
3293  *  'a' = Alive but not in-sync raid set component _or_ alive raid4/5/6 'write_back' journal device
3294  *  'A' = Alive and in-sync raid set component _or_ alive raid4/5/6 'write_through' journal device
3295  *  '-' = Non-existing device (i.e. uspace passed '- -' into the ctr)
3296  */
3297 static const char *__raid_dev_status(struct raid_set *rs, struct md_rdev *rdev, bool array_in_sync)
3298 {
3299         if (!rdev->bdev)
3300                 return "-";
3301         else if (test_bit(Faulty, &rdev->flags))
3302                 return "D";
3303         else if (test_bit(Journal, &rdev->flags))
3304                 return (rs->journal_dev.mode == R5C_JOURNAL_MODE_WRITE_THROUGH) ? "A" : "a";
3305         else if (!array_in_sync || !test_bit(In_sync, &rdev->flags))
3306                 return "a";
3307         else
3308                 return "A";
3309 }
3310
3311 /* Helper to return resync/reshape progress for @rs and @array_in_sync */
3312 static sector_t rs_get_progress(struct raid_set *rs,
3313                                 sector_t resync_max_sectors, bool *array_in_sync)
3314 {
3315         sector_t r, curr_resync_completed;
3316         struct mddev *mddev = &rs->md;
3317
3318         curr_resync_completed = mddev->curr_resync_completed ?: mddev->recovery_cp;
3319         *array_in_sync = false;
3320
3321         if (rs_is_raid0(rs)) {
3322                 r = resync_max_sectors;
3323                 *array_in_sync = true;
3324
3325         } else {
3326                 r = mddev->reshape_position;
3327
3328                 /* Reshape is relative to the array size */
3329                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
3330                     r != MaxSector) {
3331                         if (r == MaxSector) {
3332                                 *array_in_sync = true;
3333                                 r = resync_max_sectors;
3334                         } else {
3335                                 /* Got to reverse on backward reshape */
3336                                 if (mddev->reshape_backwards)
3337                                         r = mddev->array_sectors - r;
3338
3339                                 /* Devide by # of data stripes */
3340                                 sector_div(r, mddev_data_stripes(rs));
3341                         }
3342
3343                 /* Sync is relative to the component device size */
3344                 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3345                         r = curr_resync_completed;
3346                 else
3347                         r = mddev->recovery_cp;
3348
3349                 if ((r == MaxSector) ||
3350                     (test_bit(MD_RECOVERY_DONE, &mddev->recovery) &&
3351                      (mddev->curr_resync_completed == resync_max_sectors))) {
3352                         /*
3353                          * Sync complete.
3354                          */
3355                         *array_in_sync = true;
3356                         r = resync_max_sectors;
3357                 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
3358                         /*
3359                          * If "check" or "repair" is occurring, the raid set has
3360                          * undergone an initial sync and the health characters
3361                          * should not be 'a' anymore.
3362                          */
3363                         *array_in_sync = true;
3364                 } else {
3365                         struct md_rdev *rdev;
3366
3367                         /*
3368                          * The raid set may be doing an initial sync, or it may
3369                          * be rebuilding individual components.  If all the
3370                          * devices are In_sync, then it is the raid set that is
3371                          * being initialized.
3372                          */
3373                         rdev_for_each(rdev, mddev)
3374                                 if (!test_bit(Journal, &rdev->flags) &&
3375                                     !test_bit(In_sync, &rdev->flags))
3376                                         *array_in_sync = true;
3377 #if 0
3378                         r = 0; /* HM FIXME: TESTME: https://bugzilla.redhat.com/show_bug.cgi?id=1210637 ? */
3379 #endif
3380                 }
3381         }
3382
3383         return r;
3384 }
3385
3386 /* Helper to return @dev name or "-" if !@dev */
3387 static const char *__get_dev_name(struct dm_dev *dev)
3388 {
3389         return dev ? dev->name : "-";
3390 }
3391
3392 static void raid_status(struct dm_target *ti, status_type_t type,
3393                         unsigned int status_flags, char *result, unsigned int maxlen)
3394 {
3395         struct raid_set *rs = ti->private;
3396         struct mddev *mddev = &rs->md;
3397         struct r5conf *conf = mddev->private;
3398         int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3399         bool array_in_sync;
3400         unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3401         unsigned int sz = 0;
3402         unsigned int rebuild_disks;
3403         unsigned int write_mostly_params = 0;
3404         sector_t progress, resync_max_sectors, resync_mismatches;
3405         const char *sync_action;
3406         struct raid_type *rt;
3407
3408         switch (type) {
3409         case STATUSTYPE_INFO:
3410                 /* *Should* always succeed */
3411                 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3412                 if (!rt)
3413                         return;
3414
3415                 DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3416
3417                 /* Access most recent mddev properties for status output */
3418                 smp_rmb();
3419                 /* Get sensible max sectors even if raid set not yet started */
3420                 resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3421                                       mddev->resync_max_sectors : mddev->dev_sectors;
3422                 progress = rs_get_progress(rs, resync_max_sectors, &array_in_sync);
3423                 resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
3424                                     atomic64_read(&mddev->resync_mismatches) : 0;
3425                 sync_action = decipher_sync_action(&rs->md);
3426
3427                 /* HM FIXME: do we want another state char for raid0? It shows 'D'/'A'/'-' now */
3428                 for (i = 0; i < rs->raid_disks; i++)
3429                         DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev, array_in_sync));
3430
3431                 /*
3432                  * In-sync/Reshape ratio:
3433                  *  The in-sync ratio shows the progress of:
3434                  *   - Initializing the raid set
3435                  *   - Rebuilding a subset of devices of the raid set
3436                  *  The user can distinguish between the two by referring
3437                  *  to the status characters.
3438                  *
3439                  *  The reshape ratio shows the progress of
3440                  *  changing the raid layout or the number of
3441                  *  disks of a raid set
3442                  */
3443                 DMEMIT(" %llu/%llu", (unsigned long long) progress,
3444                                      (unsigned long long) resync_max_sectors);
3445
3446                 /*
3447                  * v1.5.0+:
3448                  *
3449                  * Sync action:
3450                  *   See Documentation/device-mapper/dm-raid.txt for
3451                  *   information on each of these states.
3452                  */
3453                 DMEMIT(" %s", sync_action);
3454
3455                 /*
3456                  * v1.5.0+:
3457                  *
3458                  * resync_mismatches/mismatch_cnt
3459                  *   This field shows the number of discrepancies found when
3460                  *   performing a "check" of the raid set.
3461                  */
3462                 DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3463
3464                 /*
3465                  * v1.9.0+:
3466                  *
3467                  * data_offset (needed for out of space reshaping)
3468                  *   This field shows the data offset into the data
3469                  *   image LV where the first stripes data starts.
3470                  *
3471                  * We keep data_offset equal on all raid disks of the set,
3472                  * so retrieving it from the first raid disk is sufficient.
3473                  */
3474                 DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3475
3476                 /*
3477                  * v1.10.0+:
3478                  */
3479                 DMEMIT(" %s", test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ?
3480                               __raid_dev_status(rs, &rs->journal_dev.rdev, 0) : "-");
3481                 break;
3482
3483         case STATUSTYPE_TABLE:
3484                 /* Report the table line string you would use to construct this raid set */
3485
3486                 /* Calculate raid parameter count */
3487                 for (i = 0; i < rs->raid_disks; i++)
3488                         if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3489                                 write_mostly_params += 2;
3490                 rebuild_disks = memweight(rs->rebuild_disks, DISKS_ARRAY_ELEMS * sizeof(*rs->rebuild_disks));
3491                 raid_param_cnt += rebuild_disks * 2 +
3492                                   write_mostly_params +
3493                                   hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3494                                   hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2 +
3495                                   (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ? 2 : 0) +
3496                                   (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags) ? 2 : 0);
3497
3498                 /* Emit table line */
3499                 /* This has to be in the documented order for userspace! */
3500                 DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3501                 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3502                         DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3503                 if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3504                         DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3505                 if (rebuild_disks)
3506                         for (i = 0; i < rs->raid_disks; i++)
3507                                 if (test_bit(rs->dev[i].rdev.raid_disk, (void *) rs->rebuild_disks))
3508                                         DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD),
3509                                                          rs->dev[i].rdev.raid_disk);
3510                 if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3511                         DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3512                                           mddev->bitmap_info.daemon_sleep);
3513                 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3514                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3515                                          mddev->sync_speed_min);
3516                 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3517                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3518                                          mddev->sync_speed_max);
3519                 if (write_mostly_params)
3520                         for (i = 0; i < rs->raid_disks; i++)
3521                                 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3522                                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3523                                                rs->dev[i].rdev.raid_disk);
3524                 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3525                         DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3526                                           mddev->bitmap_info.max_write_behind);
3527                 if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3528                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3529                                          max_nr_stripes);
3530                 if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3531                         DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3532                                            (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3533                 if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3534                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3535                                          raid10_md_layout_to_copies(mddev->layout));
3536                 if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3537                         DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3538                                          raid10_md_layout_to_format(mddev->layout));
3539                 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3540                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3541                                          max(rs->delta_disks, mddev->delta_disks));
3542                 if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3543                         DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3544                                            (unsigned long long) rs->data_offset);
3545                 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags))
3546                         DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV),
3547                                         __get_dev_name(rs->journal_dev.dev));
3548                 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags))
3549                         DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE),
3550                                          md_journal_mode_to_dm_raid(rs->journal_dev.mode));
3551                 DMEMIT(" %d", rs->raid_disks);
3552                 for (i = 0; i < rs->raid_disks; i++)
3553                         DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3554                                          __get_dev_name(rs->dev[i].data_dev));
3555         }
3556 }
3557
3558 static int raid_message(struct dm_target *ti, unsigned int argc, char **argv)
3559 {
3560         struct raid_set *rs = ti->private;
3561         struct mddev *mddev = &rs->md;
3562
3563         if (!mddev->pers || !mddev->pers->sync_request)
3564                 return -EINVAL;
3565
3566         if (!strcasecmp(argv[0], "frozen"))
3567                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3568         else
3569                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3570
3571         if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
3572                 if (mddev->sync_thread) {
3573                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3574                         md_reap_sync_thread(mddev);
3575                 }
3576         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3577                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3578                 return -EBUSY;
3579         else if (!strcasecmp(argv[0], "resync"))
3580                 ; /* MD_RECOVERY_NEEDED set below */
3581         else if (!strcasecmp(argv[0], "recover"))
3582                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3583         else {
3584                 if (!strcasecmp(argv[0], "check")) {
3585                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3586                         set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3587                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3588                 } else if (!strcasecmp(argv[0], "repair")) {
3589                         set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3590                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3591                 } else
3592                         return -EINVAL;
3593         }
3594         if (mddev->ro == 2) {
3595                 /* A write to sync_action is enough to justify
3596                  * canceling read-auto mode
3597                  */
3598                 mddev->ro = 0;
3599                 if (!mddev->suspended && mddev->sync_thread)
3600                         md_wakeup_thread(mddev->sync_thread);
3601         }
3602         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3603         if (!mddev->suspended && mddev->thread)
3604                 md_wakeup_thread(mddev->thread);
3605
3606         return 0;
3607 }
3608
3609 static int raid_iterate_devices(struct dm_target *ti,
3610                                 iterate_devices_callout_fn fn, void *data)
3611 {
3612         struct raid_set *rs = ti->private;
3613         unsigned int i;
3614         int r = 0;
3615
3616         for (i = 0; !r && i < rs->raid_disks; i++) {
3617                 if (rs->dev[i].data_dev) {
3618                         r = fn(ti, rs->dev[i].data_dev,
3619                                0, /* No offset on data devs */
3620                                rs->md.dev_sectors, data);
3621                 }
3622         }
3623
3624         return r;
3625 }
3626
3627 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3628 {
3629         struct raid_set *rs = ti->private;
3630         unsigned int chunk_size = to_bytes(rs->md.chunk_sectors);
3631
3632         blk_limits_io_min(limits, chunk_size);
3633         blk_limits_io_opt(limits, chunk_size * mddev_data_stripes(rs));
3634 }
3635
3636 static void raid_presuspend(struct dm_target *ti)
3637 {
3638         struct raid_set *rs = ti->private;
3639
3640         md_stop_writes(&rs->md);
3641 }
3642
3643 static void raid_postsuspend(struct dm_target *ti)
3644 {
3645         struct raid_set *rs = ti->private;
3646
3647         if (!test_and_set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
3648                 mddev_lock_nointr(&rs->md);
3649                 mddev_suspend(&rs->md);
3650                 mddev_unlock(&rs->md);
3651         }
3652
3653         rs->md.ro = 1;
3654 }
3655
3656 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3657 {
3658         int i;
3659         uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
3660         unsigned long flags;
3661         bool cleared = false;
3662         struct dm_raid_superblock *sb;
3663         struct mddev *mddev = &rs->md;
3664         struct md_rdev *r;
3665
3666         /* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3667         if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3668                 return;
3669
3670         memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3671
3672         for (i = 0; i < mddev->raid_disks; i++) {
3673                 r = &rs->dev[i].rdev;
3674                 /* HM FIXME: enhance journal device recovery processing */
3675                 if (test_bit(Journal, &r->flags))
3676                         continue;
3677
3678                 if (test_bit(Faulty, &r->flags) &&
3679                     r->meta_bdev && !read_disk_sb(r, r->sb_size, true)) {
3680                         DMINFO("Faulty %s device #%d has readable super block."
3681                                "  Attempting to revive it.",
3682                                rs->raid_type->name, i);
3683
3684                         /*
3685                          * Faulty bit may be set, but sometimes the array can
3686                          * be suspended before the personalities can respond
3687                          * by removing the device from the array (i.e. calling
3688                          * 'hot_remove_disk').  If they haven't yet removed
3689                          * the failed device, its 'raid_disk' number will be
3690                          * '>= 0' - meaning we must call this function
3691                          * ourselves.
3692                          */
3693                         flags = r->flags;
3694                         clear_bit(In_sync, &r->flags); /* Mandatory for hot remove. */
3695                         if (r->raid_disk >= 0) {
3696                                 if (mddev->pers->hot_remove_disk(mddev, r)) {
3697                                         /* Failed to revive this device, try next */
3698                                         r->flags = flags;
3699                                         continue;
3700                                 }
3701                         } else
3702                                 r->raid_disk = r->saved_raid_disk = i;
3703
3704                         clear_bit(Faulty, &r->flags);
3705                         clear_bit(WriteErrorSeen, &r->flags);
3706
3707                         if (mddev->pers->hot_add_disk(mddev, r)) {
3708                                 /* Failed to revive this device, try next */
3709                                 r->raid_disk = r->saved_raid_disk = -1;
3710                                 r->flags = flags;
3711                         } else {
3712                                 clear_bit(In_sync, &r->flags);
3713                                 r->recovery_offset = 0;
3714                                 set_bit(i, (void *) cleared_failed_devices);
3715                                 cleared = true;
3716                         }
3717                 }
3718         }
3719
3720         /* If any failed devices could be cleared, update all sbs failed_devices bits */
3721         if (cleared) {
3722                 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3723
3724                 rdev_for_each(r, &rs->md) {
3725                         if (test_bit(Journal, &r->flags))
3726                                 continue;
3727
3728                         sb = page_address(r->sb_page);
3729                         sb_retrieve_failed_devices(sb, failed_devices);
3730
3731                         for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3732                                 failed_devices[i] &= ~cleared_failed_devices[i];
3733
3734                         sb_update_failed_devices(sb, failed_devices);
3735                 }
3736         }
3737 }
3738
3739 static int __load_dirty_region_bitmap(struct raid_set *rs)
3740 {
3741         int r = 0;
3742
3743         /* Try loading the bitmap unless "raid0", which does not have one */
3744         if (!rs_is_raid0(rs) &&
3745             !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3746                 r = bitmap_load(&rs->md);
3747                 if (r)
3748                         DMERR("Failed to load bitmap");
3749         }
3750
3751         return r;
3752 }
3753
3754 /* Enforce updating all superblocks */
3755 static void rs_update_sbs(struct raid_set *rs)
3756 {
3757         struct mddev *mddev = &rs->md;
3758         int ro = mddev->ro;
3759
3760         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3761         mddev->ro = 0;
3762         md_update_sb(mddev, 1);
3763         mddev->ro = ro;
3764 }
3765
3766 /*
3767  * Reshape changes raid algorithm of @rs to new one within personality
3768  * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3769  * disks from a raid set thus growing/shrinking it or resizes the set
3770  *
3771  * Call mddev_lock_nointr() before!
3772  */
3773 static int rs_start_reshape(struct raid_set *rs)
3774 {
3775         int r;
3776         struct mddev *mddev = &rs->md;
3777         struct md_personality *pers = mddev->pers;
3778
3779         r = rs_setup_reshape(rs);
3780         if (r)
3781                 return r;
3782
3783         /* Need to be resumed to be able to start reshape, recovery is frozen until raid_resume() though */
3784         if (test_and_clear_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags))
3785                 mddev_resume(mddev);
3786
3787         /*
3788          * Check any reshape constraints enforced by the personalility
3789          *
3790          * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3791          */
3792         r = pers->check_reshape(mddev);
3793         if (r) {
3794                 rs->ti->error = "pers->check_reshape() failed";
3795                 return r;
3796         }
3797
3798         /*
3799          * Personality may not provide start reshape method in which
3800          * case check_reshape above has already covered everything
3801          */
3802         if (pers->start_reshape) {
3803                 r = pers->start_reshape(mddev);
3804                 if (r) {
3805                         rs->ti->error = "pers->start_reshape() failed";
3806                         return r;
3807                 }
3808         }
3809
3810         /* Suspend because a resume will happen in raid_resume() */
3811         set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags);
3812         mddev_suspend(mddev);
3813
3814         /*
3815          * Now reshape got set up, update superblocks to
3816          * reflect the fact so that a table reload will
3817          * access proper superblock content in the ctr.
3818          */
3819         rs_update_sbs(rs);
3820
3821         return 0;
3822 }
3823
3824 static int raid_preresume(struct dm_target *ti)
3825 {
3826         int r;
3827         struct raid_set *rs = ti->private;
3828         struct mddev *mddev = &rs->md;
3829
3830         /* This is a resume after a suspend of the set -> it's already started */
3831         if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
3832                 return 0;
3833
3834         /*
3835          * The superblocks need to be updated on disk if the
3836          * array is new or new devices got added (thus zeroed
3837          * out by userspace) or __load_dirty_region_bitmap
3838          * will overwrite them in core with old data or fail.
3839          */
3840         if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
3841                 rs_update_sbs(rs);
3842
3843         /* Load the bitmap from disk unless raid0 */
3844         r = __load_dirty_region_bitmap(rs);
3845         if (r)
3846                 return r;
3847
3848         /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) */
3849         if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) && mddev->bitmap &&
3850             mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)) {
3851                 r = bitmap_resize(mddev->bitmap, mddev->dev_sectors,
3852                                   to_bytes(rs->requested_bitmap_chunk_sectors), 0);
3853                 if (r)
3854                         DMERR("Failed to resize bitmap");
3855         }
3856
3857         /* Check for any resize/reshape on @rs and adjust/initiate */
3858         /* Be prepared for mddev_resume() in raid_resume() */
3859         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3860         if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
3861                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3862                 mddev->resync_min = mddev->recovery_cp;
3863         }
3864
3865         /* Check for any reshape request unless new raid set */
3866         if (test_and_clear_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3867                 /* Initiate a reshape. */
3868                 rs_set_rdev_sectors(rs);
3869                 mddev_lock_nointr(mddev);
3870                 r = rs_start_reshape(rs);
3871                 mddev_unlock(mddev);
3872                 if (r)
3873                         DMWARN("Failed to check/start reshape, continuing without change");
3874                 r = 0;
3875         }
3876
3877         return r;
3878 }
3879
3880 static void raid_resume(struct dm_target *ti)
3881 {
3882         struct raid_set *rs = ti->private;
3883         struct mddev *mddev = &rs->md;
3884
3885         if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
3886                 /*
3887                  * A secondary resume while the device is active.
3888                  * Take this opportunity to check whether any failed
3889                  * devices are reachable again.
3890                  */
3891                 attempt_restore_of_faulty_devices(rs);
3892         }
3893
3894         mddev->ro = 0;
3895         mddev->in_sync = 0;
3896
3897         /* Only reduce raid set size before running a disk removing reshape. */
3898         if (mddev->delta_disks < 0)
3899                 rs_set_capacity(rs);
3900
3901         /*
3902          * Keep the RAID set frozen if reshape/rebuild flags are set.
3903          * The RAID set is unfrozen once the next table load/resume,
3904          * which clears the reshape/rebuild flags, occurs.
3905          * This ensures that the constructor for the inactive table
3906          * retrieves an up-to-date reshape_position.
3907          */
3908         if (!(rs->ctr_flags & RESUME_STAY_FROZEN_FLAGS))
3909                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3910
3911         if (test_and_clear_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
3912                 mddev_lock_nointr(mddev);
3913                 mddev_resume(mddev);
3914                 mddev_unlock(mddev);
3915         }
3916 }
3917
3918 static struct target_type raid_target = {
3919         .name = "raid",
3920         .version = {1, 13, 0},
3921         .module = THIS_MODULE,
3922         .ctr = raid_ctr,
3923         .dtr = raid_dtr,
3924         .map = raid_map,
3925         .status = raid_status,
3926         .message = raid_message,
3927         .iterate_devices = raid_iterate_devices,
3928         .io_hints = raid_io_hints,
3929         .presuspend = raid_presuspend,
3930         .postsuspend = raid_postsuspend,
3931         .preresume = raid_preresume,
3932         .resume = raid_resume,
3933 };
3934
3935 static int __init dm_raid_init(void)
3936 {
3937         DMINFO("Loading target version %u.%u.%u",
3938                raid_target.version[0],
3939                raid_target.version[1],
3940                raid_target.version[2]);
3941         return dm_register_target(&raid_target);
3942 }
3943
3944 static void __exit dm_raid_exit(void)
3945 {
3946         dm_unregister_target(&raid_target);
3947 }
3948
3949 module_init(dm_raid_init);
3950 module_exit(dm_raid_exit);
3951
3952 module_param(devices_handle_discard_safely, bool, 0644);
3953 MODULE_PARM_DESC(devices_handle_discard_safely,
3954                  "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
3955
3956 MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
3957 MODULE_ALIAS("dm-raid0");
3958 MODULE_ALIAS("dm-raid1");
3959 MODULE_ALIAS("dm-raid10");
3960 MODULE_ALIAS("dm-raid4");
3961 MODULE_ALIAS("dm-raid5");
3962 MODULE_ALIAS("dm-raid6");
3963 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
3964 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
3965 MODULE_LICENSE("GPL");