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;
1002         unsigned int group_size, last_group_start;
1003
1004         for (i = 0; i < rs->md.raid_disks; i++)
1005                 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
1006                     !rs->dev[i].rdev.sb_page)
1007                         rebuild_cnt++;
1008
1009         switch (rs->raid_type->level) {
1010         case 0:
1011                 break;
1012         case 1:
1013                 if (rebuild_cnt >= rs->md.raid_disks)
1014                         goto too_many;
1015                 break;
1016         case 4:
1017         case 5:
1018         case 6:
1019                 if (rebuild_cnt > rs->raid_type->parity_devs)
1020                         goto too_many;
1021                 break;
1022         case 10:
1023                 copies = raid10_md_layout_to_copies(rs->md.new_layout);
1024                 if (rebuild_cnt < copies)
1025                         break;
1026
1027                 /*
1028                  * It is possible to have a higher rebuild count for RAID10,
1029                  * as long as the failed devices occur in different mirror
1030                  * groups (i.e. different stripes).
1031                  *
1032                  * When checking "near" format, make sure no adjacent devices
1033                  * have failed beyond what can be handled.  In addition to the
1034                  * simple case where the number of devices is a multiple of the
1035                  * number of copies, we must also handle cases where the number
1036                  * of devices is not a multiple of the number of copies.
1037                  * E.g.    dev1 dev2 dev3 dev4 dev5
1038                  *          A    A    B    B    C
1039                  *          C    D    D    E    E
1040                  */
1041                 if (__is_raid10_near(rs->md.new_layout)) {
1042                         for (i = 0; i < rs->md.raid_disks; i++) {
1043                                 if (!(i % copies))
1044                                         rebuilds_per_group = 0;
1045                                 if ((!rs->dev[i].rdev.sb_page ||
1046                                     !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1047                                     (++rebuilds_per_group >= copies))
1048                                         goto too_many;
1049                         }
1050                         break;
1051                 }
1052
1053                 /*
1054                  * When checking "far" and "offset" formats, we need to ensure
1055                  * that the device that holds its copy is not also dead or
1056                  * being rebuilt.  (Note that "far" and "offset" formats only
1057                  * support two copies right now.  These formats also only ever
1058                  * use the 'use_far_sets' variant.)
1059                  *
1060                  * This check is somewhat complicated by the need to account
1061                  * for arrays that are not a multiple of (far) copies.  This
1062                  * results in the need to treat the last (potentially larger)
1063                  * set differently.
1064                  */
1065                 group_size = (rs->md.raid_disks / copies);
1066                 last_group_start = (rs->md.raid_disks / group_size) - 1;
1067                 last_group_start *= group_size;
1068                 for (i = 0; i < rs->md.raid_disks; i++) {
1069                         if (!(i % copies) && !(i > last_group_start))
1070                                 rebuilds_per_group = 0;
1071                         if ((!rs->dev[i].rdev.sb_page ||
1072                              !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1073                             (++rebuilds_per_group >= copies))
1074                                         goto too_many;
1075                 }
1076                 break;
1077         default:
1078                 if (rebuild_cnt)
1079                         return -EINVAL;
1080         }
1081
1082         return 0;
1083
1084 too_many:
1085         return -EINVAL;
1086 }
1087
1088 /*
1089  * Possible arguments are...
1090  *      <chunk_size> [optional_args]
1091  *
1092  * Argument definitions
1093  *    <chunk_size>                      The number of sectors per disk that
1094  *                                      will form the "stripe"
1095  *    [[no]sync]                        Force or prevent recovery of the
1096  *                                      entire array
1097  *    [rebuild <idx>]                   Rebuild the drive indicated by the index
1098  *    [daemon_sleep <ms>]               Time between bitmap daemon work to
1099  *                                      clear bits
1100  *    [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1101  *    [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1102  *    [write_mostly <idx>]              Indicate a write mostly drive via index
1103  *    [max_write_behind <sectors>]      See '-write-behind=' (man mdadm)
1104  *    [stripe_cache <sectors>]          Stripe cache size for higher RAIDs
1105  *    [region_size <sectors>]           Defines granularity of bitmap
1106  *    [journal_dev <dev>]               raid4/5/6 journaling deviice
1107  *                                      (i.e. write hole closing log)
1108  *
1109  * RAID10-only options:
1110  *    [raid10_copies <# copies>]        Number of copies.  (Default: 2)
1111  *    [raid10_format <near|far|offset>] Layout algorithm.  (Default: near)
1112  */
1113 static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1114                              unsigned int num_raid_params)
1115 {
1116         int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1117         unsigned int raid10_copies = 2;
1118         unsigned int i, write_mostly = 0;
1119         unsigned int region_size = 0;
1120         sector_t max_io_len;
1121         const char *arg, *key;
1122         struct raid_dev *rd;
1123         struct raid_type *rt = rs->raid_type;
1124
1125         arg = dm_shift_arg(as);
1126         num_raid_params--; /* Account for chunk_size argument */
1127
1128         if (kstrtoint(arg, 10, &value) < 0) {
1129                 rs->ti->error = "Bad numerical argument given for chunk_size";
1130                 return -EINVAL;
1131         }
1132
1133         /*
1134          * First, parse the in-order required arguments
1135          * "chunk_size" is the only argument of this type.
1136          */
1137         if (rt_is_raid1(rt)) {
1138                 if (value)
1139                         DMERR("Ignoring chunk size parameter for RAID 1");
1140                 value = 0;
1141         } else if (!is_power_of_2(value)) {
1142                 rs->ti->error = "Chunk size must be a power of 2";
1143                 return -EINVAL;
1144         } else if (value < 8) {
1145                 rs->ti->error = "Chunk size value is too small";
1146                 return -EINVAL;
1147         }
1148
1149         rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1150
1151         /*
1152          * We set each individual device as In_sync with a completed
1153          * 'recovery_offset'.  If there has been a device failure or
1154          * replacement then one of the following cases applies:
1155          *
1156          *   1) User specifies 'rebuild'.
1157          *      - Device is reset when param is read.
1158          *   2) A new device is supplied.
1159          *      - No matching superblock found, resets device.
1160          *   3) Device failure was transient and returns on reload.
1161          *      - Failure noticed, resets device for bitmap replay.
1162          *   4) Device hadn't completed recovery after previous failure.
1163          *      - Superblock is read and overrides recovery_offset.
1164          *
1165          * What is found in the superblocks of the devices is always
1166          * authoritative, unless 'rebuild' or '[no]sync' was specified.
1167          */
1168         for (i = 0; i < rs->raid_disks; i++) {
1169                 set_bit(In_sync, &rs->dev[i].rdev.flags);
1170                 rs->dev[i].rdev.recovery_offset = MaxSector;
1171         }
1172
1173         /*
1174          * Second, parse the unordered optional arguments
1175          */
1176         for (i = 0; i < num_raid_params; i++) {
1177                 key = dm_shift_arg(as);
1178                 if (!key) {
1179                         rs->ti->error = "Not enough raid parameters given";
1180                         return -EINVAL;
1181                 }
1182
1183                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1184                         if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1185                                 rs->ti->error = "Only one 'nosync' argument allowed";
1186                                 return -EINVAL;
1187                         }
1188                         continue;
1189                 }
1190                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1191                         if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1192                                 rs->ti->error = "Only one 'sync' argument allowed";
1193                                 return -EINVAL;
1194                         }
1195                         continue;
1196                 }
1197                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1198                         if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1199                                 rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1200                                 return -EINVAL;
1201                         }
1202                         continue;
1203                 }
1204
1205                 arg = dm_shift_arg(as);
1206                 i++; /* Account for the argument pairs */
1207                 if (!arg) {
1208                         rs->ti->error = "Wrong number of raid parameters given";
1209                         return -EINVAL;
1210                 }
1211
1212                 /*
1213                  * Parameters that take a string value are checked here.
1214                  */
1215                 /* "raid10_format {near|offset|far} */
1216                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1217                         if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1218                                 rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1219                                 return -EINVAL;
1220                         }
1221                         if (!rt_is_raid10(rt)) {
1222                                 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1223                                 return -EINVAL;
1224                         }
1225                         raid10_format = raid10_name_to_format(arg);
1226                         if (raid10_format < 0) {
1227                                 rs->ti->error = "Invalid 'raid10_format' value given";
1228                                 return raid10_format;
1229                         }
1230                         continue;
1231                 }
1232
1233                 /* "journal_dev <dev>" */
1234                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV))) {
1235                         int r;
1236                         struct md_rdev *jdev;
1237
1238                         if (test_and_set_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1239                                 rs->ti->error = "Only one raid4/5/6 set journaling device allowed";
1240                                 return -EINVAL;
1241                         }
1242                         if (!rt_is_raid456(rt)) {
1243                                 rs->ti->error = "'journal_dev' is an invalid parameter for this RAID type";
1244                                 return -EINVAL;
1245                         }
1246                         r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
1247                                           &rs->journal_dev.dev);
1248                         if (r) {
1249                                 rs->ti->error = "raid4/5/6 journal device lookup failure";
1250                                 return r;
1251                         }
1252                         jdev = &rs->journal_dev.rdev;
1253                         md_rdev_init(jdev);
1254                         jdev->mddev = &rs->md;
1255                         jdev->bdev = rs->journal_dev.dev->bdev;
1256                         jdev->sectors = to_sector(i_size_read(jdev->bdev->bd_inode));
1257                         if (jdev->sectors < MIN_RAID456_JOURNAL_SPACE) {
1258                                 rs->ti->error = "No space for raid4/5/6 journal";
1259                                 return -ENOSPC;
1260                         }
1261                         rs->journal_dev.mode = R5C_JOURNAL_MODE_WRITE_THROUGH;
1262                         set_bit(Journal, &jdev->flags);
1263                         continue;
1264                 }
1265
1266                 /* "journal_mode <mode>" ("journal_dev" mandatory!) */
1267                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE))) {
1268                         int r;
1269
1270                         if (!test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1271                                 rs->ti->error = "raid4/5/6 'journal_mode' is invalid without 'journal_dev'";
1272                                 return -EINVAL;
1273                         }
1274                         if (test_and_set_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
1275                                 rs->ti->error = "Only one raid4/5/6 'journal_mode' argument allowed";
1276                                 return -EINVAL;
1277                         }
1278                         r = dm_raid_journal_mode_to_md(arg);
1279                         if (r < 0) {
1280                                 rs->ti->error = "Invalid 'journal_mode' argument";
1281                                 return r;
1282                         }
1283                         rs->journal_dev.mode = r;
1284                         continue;
1285                 }
1286
1287                 /*
1288                  * Parameters with number values from here on.
1289                  */
1290                 if (kstrtoint(arg, 10, &value) < 0) {
1291                         rs->ti->error = "Bad numerical argument given in raid params";
1292                         return -EINVAL;
1293                 }
1294
1295                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1296                         /*
1297                          * "rebuild" is being passed in by userspace to provide
1298                          * indexes of replaced devices and to set up additional
1299                          * devices on raid level takeover.
1300                          */
1301                         if (!__within_range(value, 0, rs->raid_disks - 1)) {
1302                                 rs->ti->error = "Invalid rebuild index given";
1303                                 return -EINVAL;
1304                         }
1305
1306                         if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1307                                 rs->ti->error = "rebuild for this index already given";
1308                                 return -EINVAL;
1309                         }
1310
1311                         rd = rs->dev + value;
1312                         clear_bit(In_sync, &rd->rdev.flags);
1313                         clear_bit(Faulty, &rd->rdev.flags);
1314                         rd->rdev.recovery_offset = 0;
1315                         set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1316                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1317                         if (!rt_is_raid1(rt)) {
1318                                 rs->ti->error = "write_mostly option is only valid for RAID1";
1319                                 return -EINVAL;
1320                         }
1321
1322                         if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1323                                 rs->ti->error = "Invalid write_mostly index given";
1324                                 return -EINVAL;
1325                         }
1326
1327                         write_mostly++;
1328                         set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1329                         set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1330                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1331                         if (!rt_is_raid1(rt)) {
1332                                 rs->ti->error = "max_write_behind option is only valid for RAID1";
1333                                 return -EINVAL;
1334                         }
1335
1336                         if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1337                                 rs->ti->error = "Only one max_write_behind argument pair allowed";
1338                                 return -EINVAL;
1339                         }
1340
1341                         /*
1342                          * In device-mapper, we specify things in sectors, but
1343                          * MD records this value in kB
1344                          */
1345                         value /= 2;
1346                         if (value > COUNTER_MAX) {
1347                                 rs->ti->error = "Max write-behind limit out of range";
1348                                 return -EINVAL;
1349                         }
1350
1351                         rs->md.bitmap_info.max_write_behind = value;
1352                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1353                         if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1354                                 rs->ti->error = "Only one daemon_sleep argument pair allowed";
1355                                 return -EINVAL;
1356                         }
1357                         if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
1358                                 rs->ti->error = "daemon sleep period out of range";
1359                                 return -EINVAL;
1360                         }
1361                         rs->md.bitmap_info.daemon_sleep = value;
1362                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1363                         /* Userspace passes new data_offset after having extended the the data image LV */
1364                         if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1365                                 rs->ti->error = "Only one data_offset argument pair allowed";
1366                                 return -EINVAL;
1367                         }
1368                         /* Ensure sensible data offset */
1369                         if (value < 0 ||
1370                             (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
1371                                 rs->ti->error = "Bogus data_offset value";
1372                                 return -EINVAL;
1373                         }
1374                         rs->data_offset = value;
1375                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1376                         /* Define the +/-# of disks to add to/remove from the given raid set */
1377                         if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1378                                 rs->ti->error = "Only one delta_disks argument pair allowed";
1379                                 return -EINVAL;
1380                         }
1381                         /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1382                         if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1383                                 rs->ti->error = "Too many delta_disk requested";
1384                                 return -EINVAL;
1385                         }
1386
1387                         rs->delta_disks = value;
1388                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1389                         if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1390                                 rs->ti->error = "Only one stripe_cache argument pair allowed";
1391                                 return -EINVAL;
1392                         }
1393
1394                         if (!rt_is_raid456(rt)) {
1395                                 rs->ti->error = "Inappropriate argument: stripe_cache";
1396                                 return -EINVAL;
1397                         }
1398
1399                         rs->stripe_cache_entries = value;
1400                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1401                         if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1402                                 rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1403                                 return -EINVAL;
1404                         }
1405                         if (value > INT_MAX) {
1406                                 rs->ti->error = "min_recovery_rate out of range";
1407                                 return -EINVAL;
1408                         }
1409                         rs->md.sync_speed_min = (int)value;
1410                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1411                         if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
1412                                 rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1413                                 return -EINVAL;
1414                         }
1415                         if (value > INT_MAX) {
1416                                 rs->ti->error = "max_recovery_rate out of range";
1417                                 return -EINVAL;
1418                         }
1419                         rs->md.sync_speed_max = (int)value;
1420                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1421                         if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1422                                 rs->ti->error = "Only one region_size argument pair allowed";
1423                                 return -EINVAL;
1424                         }
1425
1426                         region_size = value;
1427                         rs->requested_bitmap_chunk_sectors = value;
1428                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1429                         if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1430                                 rs->ti->error = "Only one raid10_copies argument pair allowed";
1431                                 return -EINVAL;
1432                         }
1433
1434                         if (!__within_range(value, 2, rs->md.raid_disks)) {
1435                                 rs->ti->error = "Bad value for 'raid10_copies'";
1436                                 return -EINVAL;
1437                         }
1438
1439                         raid10_copies = value;
1440                 } else {
1441                         DMERR("Unable to parse RAID parameter: %s", key);
1442                         rs->ti->error = "Unable to parse RAID parameter";
1443                         return -EINVAL;
1444                 }
1445         }
1446
1447         if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1448             test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1449                 rs->ti->error = "sync and nosync are mutually exclusive";
1450                 return -EINVAL;
1451         }
1452
1453         if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1454             (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1455              test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1456                 rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1457                 return -EINVAL;
1458         }
1459
1460         if (write_mostly >= rs->md.raid_disks) {
1461                 rs->ti->error = "Can't set all raid1 devices to write_mostly";
1462                 return -EINVAL;
1463         }
1464
1465         if (validate_region_size(rs, region_size))
1466                 return -EINVAL;
1467
1468         if (rs->md.chunk_sectors)
1469                 max_io_len = rs->md.chunk_sectors;
1470         else
1471                 max_io_len = region_size;
1472
1473         if (dm_set_target_max_io_len(rs->ti, max_io_len))
1474                 return -EINVAL;
1475
1476         if (rt_is_raid10(rt)) {
1477                 if (raid10_copies > rs->md.raid_disks) {
1478                         rs->ti->error = "Not enough devices to satisfy specification";
1479                         return -EINVAL;
1480                 }
1481
1482                 rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1483                 if (rs->md.new_layout < 0) {
1484                         rs->ti->error = "Error getting raid10 format";
1485                         return rs->md.new_layout;
1486                 }
1487
1488                 rt = get_raid_type_by_ll(10, rs->md.new_layout);
1489                 if (!rt) {
1490                         rs->ti->error = "Failed to recognize new raid10 layout";
1491                         return -EINVAL;
1492                 }
1493
1494                 if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1495                      rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1496                     test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1497                         rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1498                         return -EINVAL;
1499                 }
1500         }
1501
1502         rs->raid10_copies = raid10_copies;
1503
1504         /* Assume there are no metadata devices until the drives are parsed */
1505         rs->md.persistent = 0;
1506         rs->md.external = 1;
1507
1508         /* Check, if any invalid ctr arguments have been passed in for the raid level */
1509         return rs_check_for_valid_flags(rs);
1510 }
1511
1512 /* Set raid4/5/6 cache size */
1513 static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1514 {
1515         int r;
1516         struct r5conf *conf;
1517         struct mddev *mddev = &rs->md;
1518         uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1519         uint32_t nr_stripes = rs->stripe_cache_entries;
1520
1521         if (!rt_is_raid456(rs->raid_type)) {
1522                 rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1523                 return -EINVAL;
1524         }
1525
1526         if (nr_stripes < min_stripes) {
1527                 DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1528                        nr_stripes, min_stripes);
1529                 nr_stripes = min_stripes;
1530         }
1531
1532         conf = mddev->private;
1533         if (!conf) {
1534                 rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1535                 return -EINVAL;
1536         }
1537
1538         /* Try setting number of stripes in raid456 stripe cache */
1539         if (conf->min_nr_stripes != nr_stripes) {
1540                 r = raid5_set_cache_size(mddev, nr_stripes);
1541                 if (r) {
1542                         rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1543                         return r;
1544                 }
1545
1546                 DMINFO("%u stripe cache entries", nr_stripes);
1547         }
1548
1549         return 0;
1550 }
1551
1552 /* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1553 static unsigned int mddev_data_stripes(struct raid_set *rs)
1554 {
1555         return rs->md.raid_disks - rs->raid_type->parity_devs;
1556 }
1557
1558 /* Return # of data stripes of @rs (i.e. as of ctr) */
1559 static unsigned int rs_data_stripes(struct raid_set *rs)
1560 {
1561         return rs->raid_disks - rs->raid_type->parity_devs;
1562 }
1563
1564 /*
1565  * Retrieve rdev->sectors from any valid raid device of @rs
1566  * to allow userpace to pass in arbitray "- -" device tupples.
1567  */
1568 static sector_t __rdev_sectors(struct raid_set *rs)
1569 {
1570         int i;
1571
1572         for (i = 0; i < rs->md.raid_disks; i++) {
1573                 struct md_rdev *rdev = &rs->dev[i].rdev;
1574
1575                 if (!test_bit(Journal, &rdev->flags) &&
1576                     rdev->bdev && rdev->sectors)
1577                         return rdev->sectors;
1578         }
1579
1580         return 0;
1581 }
1582
1583 /* Calculate the sectors per device and per array used for @rs */
1584 static int rs_set_dev_and_array_sectors(struct raid_set *rs, bool use_mddev)
1585 {
1586         int delta_disks;
1587         unsigned int data_stripes;
1588         struct mddev *mddev = &rs->md;
1589         struct md_rdev *rdev;
1590         sector_t array_sectors = rs->ti->len, dev_sectors = rs->ti->len;
1591
1592         if (use_mddev) {
1593                 delta_disks = mddev->delta_disks;
1594                 data_stripes = mddev_data_stripes(rs);
1595         } else {
1596                 delta_disks = rs->delta_disks;
1597                 data_stripes = rs_data_stripes(rs);
1598         }
1599
1600         /* Special raid1 case w/o delta_disks support (yet) */
1601         if (rt_is_raid1(rs->raid_type))
1602                 ;
1603         else if (rt_is_raid10(rs->raid_type)) {
1604                 if (rs->raid10_copies < 2 ||
1605                     delta_disks < 0) {
1606                         rs->ti->error = "Bogus raid10 data copies or delta disks";
1607                         return -EINVAL;
1608                 }
1609
1610                 dev_sectors *= rs->raid10_copies;
1611                 if (sector_div(dev_sectors, data_stripes))
1612                         goto bad;
1613
1614                 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1615                 if (sector_div(array_sectors, rs->raid10_copies))
1616                         goto bad;
1617
1618         } else if (sector_div(dev_sectors, data_stripes))
1619                 goto bad;
1620
1621         else
1622                 /* Striped layouts */
1623                 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1624
1625         rdev_for_each(rdev, mddev)
1626                 if (!test_bit(Journal, &rdev->flags))
1627                         rdev->sectors = dev_sectors;
1628
1629         mddev->array_sectors = array_sectors;
1630         mddev->dev_sectors = dev_sectors;
1631
1632         return 0;
1633 bad:
1634         rs->ti->error = "Target length not divisible by number of data devices";
1635         return -EINVAL;
1636 }
1637
1638 /* Setup recovery on @rs */
1639 static void __rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1640 {
1641         /* raid0 does not recover */
1642         if (rs_is_raid0(rs))
1643                 rs->md.recovery_cp = MaxSector;
1644         /*
1645          * A raid6 set has to be recovered either
1646          * completely or for the grown part to
1647          * ensure proper parity and Q-Syndrome
1648          */
1649         else if (rs_is_raid6(rs))
1650                 rs->md.recovery_cp = dev_sectors;
1651         /*
1652          * Other raid set types may skip recovery
1653          * depending on the 'nosync' flag.
1654          */
1655         else
1656                 rs->md.recovery_cp = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1657                                      ? MaxSector : dev_sectors;
1658 }
1659
1660 /* Setup recovery on @rs based on raid type, device size and 'nosync' flag */
1661 static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1662 {
1663         if (!dev_sectors)
1664                 /* New raid set or 'sync' flag provided */
1665                 __rs_setup_recovery(rs, 0);
1666         else if (dev_sectors == MaxSector)
1667                 /* Prevent recovery */
1668                 __rs_setup_recovery(rs, MaxSector);
1669         else if (__rdev_sectors(rs) < dev_sectors)
1670                 /* Grown raid set */
1671                 __rs_setup_recovery(rs, __rdev_sectors(rs));
1672         else
1673                 __rs_setup_recovery(rs, MaxSector);
1674 }
1675
1676 static void do_table_event(struct work_struct *ws)
1677 {
1678         struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1679
1680         smp_rmb(); /* Make sure we access most actual mddev properties */
1681         if (!rs_is_reshaping(rs)) {
1682                 if (rs_is_raid10(rs))
1683                         rs_set_rdev_sectors(rs);
1684                 rs_set_capacity(rs);
1685         }
1686         dm_table_event(rs->ti->table);
1687 }
1688
1689 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
1690 {
1691         struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
1692
1693         return mddev_congested(&rs->md, bits);
1694 }
1695
1696 /*
1697  * Make sure a valid takover (level switch) is being requested on @rs
1698  *
1699  * Conversions of raid sets from one MD personality to another
1700  * have to conform to restrictions which are enforced here.
1701  */
1702 static int rs_check_takeover(struct raid_set *rs)
1703 {
1704         struct mddev *mddev = &rs->md;
1705         unsigned int near_copies;
1706
1707         if (rs->md.degraded) {
1708                 rs->ti->error = "Can't takeover degraded raid set";
1709                 return -EPERM;
1710         }
1711
1712         if (rs_is_reshaping(rs)) {
1713                 rs->ti->error = "Can't takeover reshaping raid set";
1714                 return -EPERM;
1715         }
1716
1717         switch (mddev->level) {
1718         case 0:
1719                 /* raid0 -> raid1/5 with one disk */
1720                 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1721                     mddev->raid_disks == 1)
1722                         return 0;
1723
1724                 /* raid0 -> raid10 */
1725                 if (mddev->new_level == 10 &&
1726                     !(rs->raid_disks % mddev->raid_disks))
1727                         return 0;
1728
1729                 /* raid0 with multiple disks -> raid4/5/6 */
1730                 if (__within_range(mddev->new_level, 4, 6) &&
1731                     mddev->new_layout == ALGORITHM_PARITY_N &&
1732                     mddev->raid_disks > 1)
1733                         return 0;
1734
1735                 break;
1736
1737         case 10:
1738                 /* Can't takeover raid10_offset! */
1739                 if (__is_raid10_offset(mddev->layout))
1740                         break;
1741
1742                 near_copies = __raid10_near_copies(mddev->layout);
1743
1744                 /* raid10* -> raid0 */
1745                 if (mddev->new_level == 0) {
1746                         /* Can takeover raid10_near with raid disks divisable by data copies! */
1747                         if (near_copies > 1 &&
1748                             !(mddev->raid_disks % near_copies)) {
1749                                 mddev->raid_disks /= near_copies;
1750                                 mddev->delta_disks = mddev->raid_disks;
1751                                 return 0;
1752                         }
1753
1754                         /* Can takeover raid10_far */
1755                         if (near_copies == 1 &&
1756                             __raid10_far_copies(mddev->layout) > 1)
1757                                 return 0;
1758
1759                         break;
1760                 }
1761
1762                 /* raid10_{near,far} -> raid1 */
1763                 if (mddev->new_level == 1 &&
1764                     max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1765                         return 0;
1766
1767                 /* raid10_{near,far} with 2 disks -> raid4/5 */
1768                 if (__within_range(mddev->new_level, 4, 5) &&
1769                     mddev->raid_disks == 2)
1770                         return 0;
1771                 break;
1772
1773         case 1:
1774                 /* raid1 with 2 disks -> raid4/5 */
1775                 if (__within_range(mddev->new_level, 4, 5) &&
1776                     mddev->raid_disks == 2) {
1777                         mddev->degraded = 1;
1778                         return 0;
1779                 }
1780
1781                 /* raid1 -> raid0 */
1782                 if (mddev->new_level == 0 &&
1783                     mddev->raid_disks == 1)
1784                         return 0;
1785
1786                 /* raid1 -> raid10 */
1787                 if (mddev->new_level == 10)
1788                         return 0;
1789                 break;
1790
1791         case 4:
1792                 /* raid4 -> raid0 */
1793                 if (mddev->new_level == 0)
1794                         return 0;
1795
1796                 /* raid4 -> raid1/5 with 2 disks */
1797                 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1798                     mddev->raid_disks == 2)
1799                         return 0;
1800
1801                 /* raid4 -> raid5/6 with parity N */
1802                 if (__within_range(mddev->new_level, 5, 6) &&
1803                     mddev->layout == ALGORITHM_PARITY_N)
1804                         return 0;
1805                 break;
1806
1807         case 5:
1808                 /* raid5 with parity N -> raid0 */
1809                 if (mddev->new_level == 0 &&
1810                     mddev->layout == ALGORITHM_PARITY_N)
1811                         return 0;
1812
1813                 /* raid5 with parity N -> raid4 */
1814                 if (mddev->new_level == 4 &&
1815                     mddev->layout == ALGORITHM_PARITY_N)
1816                         return 0;
1817
1818                 /* raid5 with 2 disks -> raid1/4/10 */
1819                 if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1820                     mddev->raid_disks == 2)
1821                         return 0;
1822
1823                 /* raid5_* ->  raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1824                 if (mddev->new_level == 6 &&
1825                     ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1826                       __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1827                         return 0;
1828                 break;
1829
1830         case 6:
1831                 /* raid6 with parity N -> raid0 */
1832                 if (mddev->new_level == 0 &&
1833                     mddev->layout == ALGORITHM_PARITY_N)
1834                         return 0;
1835
1836                 /* raid6 with parity N -> raid4 */
1837                 if (mddev->new_level == 4 &&
1838                     mddev->layout == ALGORITHM_PARITY_N)
1839                         return 0;
1840
1841                 /* raid6_*_n with Q-Syndrome N -> raid5_* */
1842                 if (mddev->new_level == 5 &&
1843                     ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1844                      __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1845                         return 0;
1846
1847         default:
1848                 break;
1849         }
1850
1851         rs->ti->error = "takeover not possible";
1852         return -EINVAL;
1853 }
1854
1855 /* True if @rs requested to be taken over */
1856 static bool rs_takeover_requested(struct raid_set *rs)
1857 {
1858         return rs->md.new_level != rs->md.level;
1859 }
1860
1861 /* True if @rs is requested to reshape by ctr */
1862 static bool rs_reshape_requested(struct raid_set *rs)
1863 {
1864         bool change;
1865         struct mddev *mddev = &rs->md;
1866
1867         if (rs_takeover_requested(rs))
1868                 return false;
1869
1870         if (!mddev->level)
1871                 return false;
1872
1873         change = mddev->new_layout != mddev->layout ||
1874                  mddev->new_chunk_sectors != mddev->chunk_sectors ||
1875                  rs->delta_disks;
1876
1877         /* Historical case to support raid1 reshape without delta disks */
1878         if (mddev->level == 1) {
1879                 if (rs->delta_disks)
1880                         return !!rs->delta_disks;
1881
1882                 return !change &&
1883                        mddev->raid_disks != rs->raid_disks;
1884         }
1885
1886         if (mddev->level == 10)
1887                 return change &&
1888                        !__is_raid10_far(mddev->new_layout) &&
1889                        rs->delta_disks >= 0;
1890
1891         return change;
1892 }
1893
1894 /*  Features */
1895 #define FEATURE_FLAG_SUPPORTS_V190      0x1 /* Supports extended superblock */
1896
1897 /* State flags for sb->flags */
1898 #define SB_FLAG_RESHAPE_ACTIVE          0x1
1899 #define SB_FLAG_RESHAPE_BACKWARDS       0x2
1900
1901 /*
1902  * This structure is never routinely used by userspace, unlike md superblocks.
1903  * Devices with this superblock should only ever be accessed via device-mapper.
1904  */
1905 #define DM_RAID_MAGIC 0x64526D44
1906 struct dm_raid_superblock {
1907         __le32 magic;           /* "DmRd" */
1908         __le32 compat_features; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1909
1910         __le32 num_devices;     /* Number of devices in this raid set. (Max 64) */
1911         __le32 array_position;  /* The position of this drive in the raid set */
1912
1913         __le64 events;          /* Incremented by md when superblock updated */
1914         __le64 failed_devices;  /* Pre 1.9.0 part of bit field of devices to */
1915                                 /* indicate failures (see extension below) */
1916
1917         /*
1918          * This offset tracks the progress of the repair or replacement of
1919          * an individual drive.
1920          */
1921         __le64 disk_recovery_offset;
1922
1923         /*
1924          * This offset tracks the progress of the initial raid set
1925          * synchronisation/parity calculation.
1926          */
1927         __le64 array_resync_offset;
1928
1929         /*
1930          * raid characteristics
1931          */
1932         __le32 level;
1933         __le32 layout;
1934         __le32 stripe_sectors;
1935
1936         /********************************************************************
1937          * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1938          *
1939          * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
1940          */
1941
1942         __le32 flags; /* Flags defining array states for reshaping */
1943
1944         /*
1945          * This offset tracks the progress of a raid
1946          * set reshape in order to be able to restart it
1947          */
1948         __le64 reshape_position;
1949
1950         /*
1951          * These define the properties of the array in case of an interrupted reshape
1952          */
1953         __le32 new_level;
1954         __le32 new_layout;
1955         __le32 new_stripe_sectors;
1956         __le32 delta_disks;
1957
1958         __le64 array_sectors; /* Array size in sectors */
1959
1960         /*
1961          * Sector offsets to data on devices (reshaping).
1962          * Needed to support out of place reshaping, thus
1963          * not writing over any stripes whilst converting
1964          * them from old to new layout
1965          */
1966         __le64 data_offset;
1967         __le64 new_data_offset;
1968
1969         __le64 sectors; /* Used device size in sectors */
1970
1971         /*
1972          * Additonal Bit field of devices indicating failures to support
1973          * up to 256 devices with the 1.9.0 on-disk metadata format
1974          */
1975         __le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
1976
1977         __le32 incompat_features;       /* Used to indicate any incompatible features */
1978
1979         /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
1980 } __packed;
1981
1982 /*
1983  * Check for reshape constraints on raid set @rs:
1984  *
1985  * - reshape function non-existent
1986  * - degraded set
1987  * - ongoing recovery
1988  * - ongoing reshape
1989  *
1990  * Returns 0 if none or -EPERM if given constraint
1991  * and error message reference in @errmsg
1992  */
1993 static int rs_check_reshape(struct raid_set *rs)
1994 {
1995         struct mddev *mddev = &rs->md;
1996
1997         if (!mddev->pers || !mddev->pers->check_reshape)
1998                 rs->ti->error = "Reshape not supported";
1999         else if (mddev->degraded)
2000                 rs->ti->error = "Can't reshape degraded raid set";
2001         else if (rs_is_recovering(rs))
2002                 rs->ti->error = "Convert request on recovering raid set prohibited";
2003         else if (rs_is_reshaping(rs))
2004                 rs->ti->error = "raid set already reshaping!";
2005         else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
2006                 rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
2007         else
2008                 return 0;
2009
2010         return -EPERM;
2011 }
2012
2013 static int read_disk_sb(struct md_rdev *rdev, int size, bool force_reload)
2014 {
2015         BUG_ON(!rdev->sb_page);
2016
2017         if (rdev->sb_loaded && !force_reload)
2018                 return 0;
2019
2020         rdev->sb_loaded = 0;
2021
2022         if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) {
2023                 DMERR("Failed to read superblock of device at position %d",
2024                       rdev->raid_disk);
2025                 md_error(rdev->mddev, rdev);
2026                 set_bit(Faulty, &rdev->flags);
2027                 return -EIO;
2028         }
2029
2030         rdev->sb_loaded = 1;
2031
2032         return 0;
2033 }
2034
2035 static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2036 {
2037         failed_devices[0] = le64_to_cpu(sb->failed_devices);
2038         memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
2039
2040         if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2041                 int i = ARRAY_SIZE(sb->extended_failed_devices);
2042
2043                 while (i--)
2044                         failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
2045         }
2046 }
2047
2048 static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2049 {
2050         int i = ARRAY_SIZE(sb->extended_failed_devices);
2051
2052         sb->failed_devices = cpu_to_le64(failed_devices[0]);
2053         while (i--)
2054                 sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
2055 }
2056
2057 /*
2058  * Synchronize the superblock members with the raid set properties
2059  *
2060  * All superblock data is little endian.
2061  */
2062 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
2063 {
2064         bool update_failed_devices = false;
2065         unsigned int i;
2066         uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2067         struct dm_raid_superblock *sb;
2068         struct raid_set *rs = container_of(mddev, struct raid_set, md);
2069
2070         /* No metadata device, no superblock */
2071         if (!rdev->meta_bdev)
2072                 return;
2073
2074         BUG_ON(!rdev->sb_page);
2075
2076         sb = page_address(rdev->sb_page);
2077
2078         sb_retrieve_failed_devices(sb, failed_devices);
2079
2080         for (i = 0; i < rs->raid_disks; i++)
2081                 if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
2082                         update_failed_devices = true;
2083                         set_bit(i, (void *) failed_devices);
2084                 }
2085
2086         if (update_failed_devices)
2087                 sb_update_failed_devices(sb, failed_devices);
2088
2089         sb->magic = cpu_to_le32(DM_RAID_MAGIC);
2090         sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2091
2092         sb->num_devices = cpu_to_le32(mddev->raid_disks);
2093         sb->array_position = cpu_to_le32(rdev->raid_disk);
2094
2095         sb->events = cpu_to_le64(mddev->events);
2096
2097         sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
2098         sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
2099
2100         sb->level = cpu_to_le32(mddev->level);
2101         sb->layout = cpu_to_le32(mddev->layout);
2102         sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
2103
2104         /********************************************************************
2105          * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
2106          *
2107          * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
2108          */
2109         sb->new_level = cpu_to_le32(mddev->new_level);
2110         sb->new_layout = cpu_to_le32(mddev->new_layout);
2111         sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
2112
2113         sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2114
2115         smp_rmb(); /* Make sure we access most recent reshape position */
2116         sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2117         if (le64_to_cpu(sb->reshape_position) != MaxSector) {
2118                 /* Flag ongoing reshape */
2119                 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
2120
2121                 if (mddev->delta_disks < 0 || mddev->reshape_backwards)
2122                         sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
2123         } else {
2124                 /* Clear reshape flags */
2125                 sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
2126         }
2127
2128         sb->array_sectors = cpu_to_le64(mddev->array_sectors);
2129         sb->data_offset = cpu_to_le64(rdev->data_offset);
2130         sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
2131         sb->sectors = cpu_to_le64(rdev->sectors);
2132         sb->incompat_features = cpu_to_le32(0);
2133
2134         /* Zero out the rest of the payload after the size of the superblock */
2135         memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
2136 }
2137
2138 /*
2139  * super_load
2140  *
2141  * This function creates a superblock if one is not found on the device
2142  * and will decide which superblock to use if there's a choice.
2143  *
2144  * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
2145  */
2146 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
2147 {
2148         int r;
2149         struct dm_raid_superblock *sb;
2150         struct dm_raid_superblock *refsb;
2151         uint64_t events_sb, events_refsb;
2152
2153         r = read_disk_sb(rdev, rdev->sb_size, false);
2154         if (r)
2155                 return r;
2156
2157         sb = page_address(rdev->sb_page);
2158
2159         /*
2160          * Two cases that we want to write new superblocks and rebuild:
2161          * 1) New device (no matching magic number)
2162          * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2163          */
2164         if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2165             (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
2166                 super_sync(rdev->mddev, rdev);
2167
2168                 set_bit(FirstUse, &rdev->flags);
2169                 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2170
2171                 /* Force writing of superblocks to disk */
2172                 set_bit(MD_SB_CHANGE_DEVS, &rdev->mddev->sb_flags);
2173
2174                 /* Any superblock is better than none, choose that if given */
2175                 return refdev ? 0 : 1;
2176         }
2177
2178         if (!refdev)
2179                 return 1;
2180
2181         events_sb = le64_to_cpu(sb->events);
2182
2183         refsb = page_address(refdev->sb_page);
2184         events_refsb = le64_to_cpu(refsb->events);
2185
2186         return (events_sb > events_refsb) ? 1 : 0;
2187 }
2188
2189 static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
2190 {
2191         int role;
2192         unsigned int d;
2193         struct mddev *mddev = &rs->md;
2194         uint64_t events_sb;
2195         uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2196         struct dm_raid_superblock *sb;
2197         uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
2198         struct md_rdev *r;
2199         struct dm_raid_superblock *sb2;
2200
2201         sb = page_address(rdev->sb_page);
2202         events_sb = le64_to_cpu(sb->events);
2203
2204         /*
2205          * Initialise to 1 if this is a new superblock.
2206          */
2207         mddev->events = events_sb ? : 1;
2208
2209         mddev->reshape_position = MaxSector;
2210
2211         mddev->raid_disks = le32_to_cpu(sb->num_devices);
2212         mddev->level = le32_to_cpu(sb->level);
2213         mddev->layout = le32_to_cpu(sb->layout);
2214         mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2215
2216         /*
2217          * Reshaping is supported, e.g. reshape_position is valid
2218          * in superblock and superblock content is authoritative.
2219          */
2220         if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2221                 /* Superblock is authoritative wrt given raid set layout! */
2222                 mddev->new_level = le32_to_cpu(sb->new_level);
2223                 mddev->new_layout = le32_to_cpu(sb->new_layout);
2224                 mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2225                 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2226                 mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2227
2228                 /* raid was reshaping and got interrupted */
2229                 if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2230                         if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
2231                                 DMERR("Reshape requested but raid set is still reshaping");
2232                                 return -EINVAL;
2233                         }
2234
2235                         if (mddev->delta_disks < 0 ||
2236                             (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
2237                                 mddev->reshape_backwards = 1;
2238                         else
2239                                 mddev->reshape_backwards = 0;
2240
2241                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2242                         rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2243                 }
2244
2245         } else {
2246                 /*
2247                  * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2248                  */
2249                 struct raid_type *rt_cur = get_raid_type_by_ll(mddev->level, mddev->layout);
2250                 struct raid_type *rt_new = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
2251
2252                 if (rs_takeover_requested(rs)) {
2253                         if (rt_cur && rt_new)
2254                                 DMERR("Takeover raid sets from %s to %s not yet supported by metadata. (raid level change)",
2255                                       rt_cur->name, rt_new->name);
2256                         else
2257                                 DMERR("Takeover raid sets not yet supported by metadata. (raid level change)");
2258                         return -EINVAL;
2259                 } else if (rs_reshape_requested(rs)) {
2260                         DMERR("Reshaping raid sets not yet supported by metadata. (raid layout change keeping level)");
2261                         if (mddev->layout != mddev->new_layout) {
2262                                 if (rt_cur && rt_new)
2263                                         DMERR("  current layout %s vs new layout %s",
2264                                               rt_cur->name, rt_new->name);
2265                                 else
2266                                         DMERR("  current layout 0x%X vs new layout 0x%X",
2267                                               le32_to_cpu(sb->layout), mddev->new_layout);
2268                         }
2269                         if (mddev->chunk_sectors != mddev->new_chunk_sectors)
2270                                 DMERR("  current stripe sectors %u vs new stripe sectors %u",
2271                                       mddev->chunk_sectors, mddev->new_chunk_sectors);
2272                         if (rs->delta_disks)
2273                                 DMERR("  current %u disks vs new %u disks",
2274                                       mddev->raid_disks, mddev->raid_disks + rs->delta_disks);
2275                         if (rs_is_raid10(rs)) {
2276                                 DMERR("  Old layout: %s w/ %u copies",
2277                                       raid10_md_layout_to_format(mddev->layout),
2278                                       raid10_md_layout_to_copies(mddev->layout));
2279                                 DMERR("  New layout: %s w/ %u copies",
2280                                       raid10_md_layout_to_format(mddev->new_layout),
2281                                       raid10_md_layout_to_copies(mddev->new_layout));
2282                         }
2283                         return -EINVAL;
2284                 }
2285
2286                 DMINFO("Discovered old metadata format; upgrading to extended metadata format");
2287         }
2288
2289         if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2290                 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2291
2292         /*
2293          * During load, we set FirstUse if a new superblock was written.
2294          * There are two reasons we might not have a superblock:
2295          * 1) The raid set is brand new - in which case, all of the
2296          *    devices must have their In_sync bit set.  Also,
2297          *    recovery_cp must be 0, unless forced.
2298          * 2) This is a new device being added to an old raid set
2299          *    and the new device needs to be rebuilt - in which
2300          *    case the In_sync bit will /not/ be set and
2301          *    recovery_cp must be MaxSector.
2302          * 3) This is/are a new device(s) being added to an old
2303          *    raid set during takeover to a higher raid level
2304          *    to provide capacity for redundancy or during reshape
2305          *    to add capacity to grow the raid set.
2306          */
2307         d = 0;
2308         rdev_for_each(r, mddev) {
2309                 if (test_bit(Journal, &rdev->flags))
2310                         continue;
2311
2312                 if (test_bit(FirstUse, &r->flags))
2313                         new_devs++;
2314
2315                 if (!test_bit(In_sync, &r->flags)) {
2316                         DMINFO("Device %d specified for rebuild; clearing superblock",
2317                                 r->raid_disk);
2318                         rebuilds++;
2319
2320                         if (test_bit(FirstUse, &r->flags))
2321                                 rebuild_and_new++;
2322                 }
2323
2324                 d++;
2325         }
2326
2327         if (new_devs == rs->raid_disks || !rebuilds) {
2328                 /* Replace a broken device */
2329                 if (new_devs == 1 && !rs->delta_disks)
2330                         ;
2331                 if (new_devs == rs->raid_disks) {
2332                         DMINFO("Superblocks created for new raid set");
2333                         set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2334                 } else if (new_devs != rebuilds &&
2335                            new_devs != rs->delta_disks) {
2336                         DMERR("New device injected into existing raid set without "
2337                               "'delta_disks' or 'rebuild' parameter specified");
2338                         return -EINVAL;
2339                 }
2340         } else if (new_devs && new_devs != rebuilds) {
2341                 DMERR("%u 'rebuild' devices cannot be injected into"
2342                       " a raid set with %u other first-time devices",
2343                       rebuilds, new_devs);
2344                 return -EINVAL;
2345         } else if (rebuilds) {
2346                 if (rebuild_and_new && rebuilds != rebuild_and_new) {
2347                         DMERR("new device%s provided without 'rebuild'",
2348                               new_devs > 1 ? "s" : "");
2349                         return -EINVAL;
2350                 } else if (rs_is_recovering(rs)) {
2351                         DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2352                               (unsigned long long) mddev->recovery_cp);
2353                         return -EINVAL;
2354                 } else if (rs_is_reshaping(rs)) {
2355                         DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2356                               (unsigned long long) mddev->reshape_position);
2357                         return -EINVAL;
2358                 }
2359         }
2360
2361         /*
2362          * Now we set the Faulty bit for those devices that are
2363          * recorded in the superblock as failed.
2364          */
2365         sb_retrieve_failed_devices(sb, failed_devices);
2366         rdev_for_each(r, mddev) {
2367                 if (test_bit(Journal, &rdev->flags) ||
2368                     !r->sb_page)
2369                         continue;
2370                 sb2 = page_address(r->sb_page);
2371                 sb2->failed_devices = 0;
2372                 memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2373
2374                 /*
2375                  * Check for any device re-ordering.
2376                  */
2377                 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2378                         role = le32_to_cpu(sb2->array_position);
2379                         if (role < 0)
2380                                 continue;
2381
2382                         if (role != r->raid_disk) {
2383                                 if (rs_is_raid10(rs) && __is_raid10_near(mddev->layout)) {
2384                                         if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2385                                             rs->raid_disks % rs->raid10_copies) {
2386                                                 rs->ti->error =
2387                                                         "Cannot change raid10 near set to odd # of devices!";
2388                                                 return -EINVAL;
2389                                         }
2390
2391                                         sb2->array_position = cpu_to_le32(r->raid_disk);
2392
2393                                 } else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2394                                            !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2395                                            !rt_is_raid1(rs->raid_type)) {
2396                                         rs->ti->error = "Cannot change device positions in raid set";
2397                                         return -EINVAL;
2398                                 }
2399
2400                                 DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2401                         }
2402
2403                         /*
2404                          * Partial recovery is performed on
2405                          * returning failed devices.
2406                          */
2407                         if (test_bit(role, (void *) failed_devices))
2408                                 set_bit(Faulty, &r->flags);
2409                 }
2410         }
2411
2412         return 0;
2413 }
2414
2415 static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2416 {
2417         struct mddev *mddev = &rs->md;
2418         struct dm_raid_superblock *sb;
2419
2420         if (rs_is_raid0(rs) || !rdev->sb_page || rdev->raid_disk < 0)
2421                 return 0;
2422
2423         sb = page_address(rdev->sb_page);
2424
2425         /*
2426          * If mddev->events is not set, we know we have not yet initialized
2427          * the array.
2428          */
2429         if (!mddev->events && super_init_validation(rs, rdev))
2430                 return -EINVAL;
2431
2432         if (le32_to_cpu(sb->compat_features) &&
2433             le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2434                 rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2435                 return -EINVAL;
2436         }
2437
2438         if (sb->incompat_features) {
2439                 rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2440                 return -EINVAL;
2441         }
2442
2443         /* Enable bitmap creation for RAID levels != 0 */
2444         mddev->bitmap_info.offset = (rt_is_raid0(rs->raid_type) || rs->journal_dev.dev) ? 0 : to_sector(4096);
2445         mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2446
2447         if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2448                 /*
2449                  * Retrieve rdev size stored in superblock to be prepared for shrink.
2450                  * Check extended superblock members are present otherwise the size
2451                  * will not be set!
2452                  */
2453                 if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190)
2454                         rdev->sectors = le64_to_cpu(sb->sectors);
2455
2456                 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2457                 if (rdev->recovery_offset == MaxSector)
2458                         set_bit(In_sync, &rdev->flags);
2459                 /*
2460                  * If no reshape in progress -> we're recovering single
2461                  * disk(s) and have to set the device(s) to out-of-sync
2462                  */
2463                 else if (!rs_is_reshaping(rs))
2464                         clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2465         }
2466
2467         /*
2468          * If a device comes back, set it as not In_sync and no longer faulty.
2469          */
2470         if (test_and_clear_bit(Faulty, &rdev->flags)) {
2471                 rdev->recovery_offset = 0;
2472                 clear_bit(In_sync, &rdev->flags);
2473                 rdev->saved_raid_disk = rdev->raid_disk;
2474         }
2475
2476         /* Reshape support -> restore repective data offsets */
2477         rdev->data_offset = le64_to_cpu(sb->data_offset);
2478         rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2479
2480         return 0;
2481 }
2482
2483 /*
2484  * Analyse superblocks and select the freshest.
2485  */
2486 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2487 {
2488         int r;
2489         struct md_rdev *rdev, *freshest;
2490         struct mddev *mddev = &rs->md;
2491
2492         freshest = NULL;
2493         rdev_for_each(rdev, mddev) {
2494                 if (test_bit(Journal, &rdev->flags))
2495                         continue;
2496
2497                 if (!rdev->meta_bdev)
2498                         continue;
2499
2500                 /* Set superblock offset/size for metadata device. */
2501                 rdev->sb_start = 0;
2502                 rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
2503                 if (rdev->sb_size < sizeof(struct dm_raid_superblock) || rdev->sb_size > PAGE_SIZE) {
2504                         DMERR("superblock size of a logical block is no longer valid");
2505                         return -EINVAL;
2506                 }
2507
2508                 /*
2509                  * Skipping super_load due to CTR_FLAG_SYNC will cause
2510                  * the array to undergo initialization again as
2511                  * though it were new.  This is the intended effect
2512                  * of the "sync" directive.
2513                  *
2514                  * With reshaping capability added, we must ensure that
2515                  * that the "sync" directive is disallowed during the reshape.
2516                  */
2517                 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2518                         continue;
2519
2520                 r = super_load(rdev, freshest);
2521
2522                 switch (r) {
2523                 case 1:
2524                         freshest = rdev;
2525                         break;
2526                 case 0:
2527                         break;
2528                 default:
2529                         /* This is a failure to read the superblock from the metadata device. */
2530                         /*
2531                          * We have to keep any raid0 data/metadata device pairs or
2532                          * the MD raid0 personality will fail to start the array.
2533                          */
2534                         if (rs_is_raid0(rs))
2535                                 continue;
2536
2537                         /*
2538                          * We keep the dm_devs to be able to emit the device tuple
2539                          * properly on the table line in raid_status() (rather than
2540                          * mistakenly acting as if '- -' got passed into the constructor).
2541                          *
2542                          * The rdev has to stay on the same_set list to allow for
2543                          * the attempt to restore faulty devices on second resume.
2544                          */
2545                         rdev->raid_disk = rdev->saved_raid_disk = -1;
2546                         break;
2547                 }
2548         }
2549
2550         if (!freshest)
2551                 return 0;
2552
2553         /*
2554          * Validation of the freshest device provides the source of
2555          * validation for the remaining devices.
2556          */
2557         rs->ti->error = "Unable to assemble array: Invalid superblocks";
2558         if (super_validate(rs, freshest))
2559                 return -EINVAL;
2560
2561         if (validate_raid_redundancy(rs)) {
2562                 rs->ti->error = "Insufficient redundancy to activate array";
2563                 return -EINVAL;
2564         }
2565
2566         rdev_for_each(rdev, mddev)
2567                 if (!test_bit(Journal, &rdev->flags) &&
2568                     rdev != freshest &&
2569                     super_validate(rs, rdev))
2570                         return -EINVAL;
2571         return 0;
2572 }
2573
2574 /*
2575  * Adjust data_offset and new_data_offset on all disk members of @rs
2576  * for out of place reshaping if requested by contructor
2577  *
2578  * We need free space at the beginning of each raid disk for forward
2579  * and at the end for backward reshapes which userspace has to provide
2580  * via remapping/reordering of space.
2581  */
2582 static int rs_adjust_data_offsets(struct raid_set *rs)
2583 {
2584         sector_t data_offset = 0, new_data_offset = 0;
2585         struct md_rdev *rdev;
2586
2587         /* Constructor did not request data offset change */
2588         if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2589                 if (!rs_is_reshapable(rs))
2590                         goto out;
2591
2592                 return 0;
2593         }
2594
2595         /* HM FIXME: get InSync raid_dev? */
2596         rdev = &rs->dev[0].rdev;
2597
2598         if (rs->delta_disks < 0) {
2599                 /*
2600                  * Removing disks (reshaping backwards):
2601                  *
2602                  * - before reshape: data is at offset 0 and free space
2603                  *                   is at end of each component LV
2604                  *
2605                  * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2606                  */
2607                 data_offset = 0;
2608                 new_data_offset = rs->data_offset;
2609
2610         } else if (rs->delta_disks > 0) {
2611                 /*
2612                  * Adding disks (reshaping forwards):
2613                  *
2614                  * - before reshape: data is at offset rs->data_offset != 0 and
2615                  *                   free space is at begin of each component LV
2616                  *
2617                  * - after reshape: data is at offset 0 on each component LV
2618                  */
2619                 data_offset = rs->data_offset;
2620                 new_data_offset = 0;
2621
2622         } else {
2623                 /*
2624                  * User space passes in 0 for data offset after having removed reshape space
2625                  *
2626                  * - or - (data offset != 0)
2627                  *
2628                  * Changing RAID layout or chunk size -> toggle offsets
2629                  *
2630                  * - before reshape: data is at offset rs->data_offset 0 and
2631                  *                   free space is at end of each component LV
2632                  *                   -or-
2633                  *                   data is at offset rs->data_offset != 0 and
2634                  *                   free space is at begin of each component LV
2635                  *
2636                  * - after reshape: data is at offset 0 if it was at offset != 0
2637                  *                  or at offset != 0 if it was at offset 0
2638                  *                  on each component LV
2639                  *
2640                  */
2641                 data_offset = rs->data_offset ? rdev->data_offset : 0;
2642                 new_data_offset = data_offset ? 0 : rs->data_offset;
2643                 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2644         }
2645
2646         /*
2647          * Make sure we got a minimum amount of free sectors per device
2648          */
2649         if (rs->data_offset &&
2650             to_sector(i_size_read(rdev->bdev->bd_inode)) - rdev->sectors < MIN_FREE_RESHAPE_SPACE) {
2651                 rs->ti->error = data_offset ? "No space for forward reshape" :
2652                                               "No space for backward reshape";
2653                 return -ENOSPC;
2654         }
2655 out:
2656         /* Adjust data offsets on all rdevs but on any raid4/5/6 journal device */
2657         rdev_for_each(rdev, &rs->md) {
2658                 if (!test_bit(Journal, &rdev->flags)) {
2659                         rdev->data_offset = data_offset;
2660                         rdev->new_data_offset = new_data_offset;
2661                 }
2662         }
2663
2664         return 0;
2665 }
2666
2667 /* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2668 static void __reorder_raid_disk_indexes(struct raid_set *rs)
2669 {
2670         int i = 0;
2671         struct md_rdev *rdev;
2672
2673         rdev_for_each(rdev, &rs->md) {
2674                 if (!test_bit(Journal, &rdev->flags)) {
2675                         rdev->raid_disk = i++;
2676                         rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2677                 }
2678         }
2679 }
2680
2681 /*
2682  * Setup @rs for takeover by a different raid level
2683  */
2684 static int rs_setup_takeover(struct raid_set *rs)
2685 {
2686         struct mddev *mddev = &rs->md;
2687         struct md_rdev *rdev;
2688         unsigned int d = mddev->raid_disks = rs->raid_disks;
2689         sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2690
2691         if (rt_is_raid10(rs->raid_type)) {
2692                 if (mddev->level == 0) {
2693                         /* Userpace reordered disks -> adjust raid_disk indexes */
2694                         __reorder_raid_disk_indexes(rs);
2695
2696                         /* raid0 -> raid10_far layout */
2697                         mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2698                                                                    rs->raid10_copies);
2699                 } else if (mddev->level == 1)
2700                         /* raid1 -> raid10_near layout */
2701                         mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2702                                                                    rs->raid_disks);
2703                 else
2704                         return -EINVAL;
2705
2706         }
2707
2708         clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2709         mddev->recovery_cp = MaxSector;
2710
2711         while (d--) {
2712                 rdev = &rs->dev[d].rdev;
2713
2714                 if (test_bit(d, (void *) rs->rebuild_disks)) {
2715                         clear_bit(In_sync, &rdev->flags);
2716                         clear_bit(Faulty, &rdev->flags);
2717                         mddev->recovery_cp = rdev->recovery_offset = 0;
2718                         /* Bitmap has to be created when we do an "up" takeover */
2719                         set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2720                 }
2721
2722                 rdev->new_data_offset = new_data_offset;
2723         }
2724
2725         return 0;
2726 }
2727
2728 /* Prepare @rs for reshape */
2729 static int rs_prepare_reshape(struct raid_set *rs)
2730 {
2731         bool reshape;
2732         struct mddev *mddev = &rs->md;
2733
2734         if (rs_is_raid10(rs)) {
2735                 if (rs->raid_disks != mddev->raid_disks &&
2736                     __is_raid10_near(mddev->layout) &&
2737                     rs->raid10_copies &&
2738                     rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2739                         /*
2740                          * raid disk have to be multiple of data copies to allow this conversion,
2741                          *
2742                          * This is actually not a reshape it is a
2743                          * rebuild of any additional mirrors per group
2744                          */
2745                         if (rs->raid_disks % rs->raid10_copies) {
2746                                 rs->ti->error = "Can't reshape raid10 mirror groups";
2747                                 return -EINVAL;
2748                         }
2749
2750                         /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2751                         __reorder_raid_disk_indexes(rs);
2752                         mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2753                                                                    rs->raid10_copies);
2754                         mddev->new_layout = mddev->layout;
2755                         reshape = false;
2756                 } else
2757                         reshape = true;
2758
2759         } else if (rs_is_raid456(rs))
2760                 reshape = true;
2761
2762         else if (rs_is_raid1(rs)) {
2763                 if (rs->delta_disks) {
2764                         /* Process raid1 via delta_disks */
2765                         mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2766                         reshape = true;
2767                 } else {
2768                         /* Process raid1 without delta_disks */
2769                         mddev->raid_disks = rs->raid_disks;
2770                         reshape = false;
2771                 }
2772         } else {
2773                 rs->ti->error = "Called with bogus raid type";
2774                 return -EINVAL;
2775         }
2776
2777         if (reshape) {
2778                 set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2779                 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2780         } else if (mddev->raid_disks < rs->raid_disks)
2781                 /* Create new superblocks and bitmaps, if any new disks */
2782                 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2783
2784         return 0;
2785 }
2786
2787 /*
2788  *
2789  * - change raid layout
2790  * - change chunk size
2791  * - add disks
2792  * - remove disks
2793  */
2794 static int rs_setup_reshape(struct raid_set *rs)
2795 {
2796         int r = 0;
2797         unsigned int cur_raid_devs, d;
2798         struct mddev *mddev = &rs->md;
2799         struct md_rdev *rdev;
2800
2801         mddev->delta_disks = rs->delta_disks;
2802         cur_raid_devs = mddev->raid_disks;
2803
2804         /* Ignore impossible layout change whilst adding/removing disks */
2805         if (mddev->delta_disks &&
2806             mddev->layout != mddev->new_layout) {
2807                 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2808                 mddev->new_layout = mddev->layout;
2809         }
2810
2811         /*
2812          * Adjust array size:
2813          *
2814          * - in case of adding disks, array size has
2815          *   to grow after the disk adding reshape,
2816          *   which'll hapen in the event handler;
2817          *   reshape will happen forward, so space has to
2818          *   be available at the beginning of each disk
2819          *
2820          * - in case of removing disks, array size
2821          *   has to shrink before starting the reshape,
2822          *   which'll happen here;
2823          *   reshape will happen backward, so space has to
2824          *   be available at the end of each disk
2825          *
2826          * - data_offset and new_data_offset are
2827          *   adjusted for aforementioned out of place
2828          *   reshaping based on userspace passing in
2829          *   the "data_offset <sectors>" key/value
2830          *   pair via the constructor
2831          */
2832
2833         /* Add disk(s) */
2834         if (rs->delta_disks > 0) {
2835                 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2836                 for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2837                         rdev = &rs->dev[d].rdev;
2838                         clear_bit(In_sync, &rdev->flags);
2839
2840                         /*
2841                          * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2842                          * by md, which'll store that erroneously in the superblock on reshape
2843                          */
2844                         rdev->saved_raid_disk = -1;
2845                         rdev->raid_disk = d;
2846
2847                         rdev->sectors = mddev->dev_sectors;
2848                         rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
2849                 }
2850
2851                 mddev->reshape_backwards = 0; /* adding disks -> forward reshape */
2852
2853         /* Remove disk(s) */
2854         } else if (rs->delta_disks < 0) {
2855                 r = rs_set_dev_and_array_sectors(rs, true);
2856                 mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2857
2858         /* Change layout and/or chunk size */
2859         } else {
2860                 /*
2861                  * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2862                  *
2863                  * keeping number of disks and do layout change ->
2864                  *
2865                  * toggle reshape_backward depending on data_offset:
2866                  *
2867                  * - free space upfront -> reshape forward
2868                  *
2869                  * - free space at the end -> reshape backward
2870                  *
2871                  *
2872                  * This utilizes free reshape space avoiding the need
2873                  * for userspace to move (parts of) LV segments in
2874                  * case of layout/chunksize change  (for disk
2875                  * adding/removing reshape space has to be at
2876                  * the proper address (see above with delta_disks):
2877                  *
2878                  * add disk(s)   -> begin
2879                  * remove disk(s)-> end
2880                  */
2881                 mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2882         }
2883
2884         return r;
2885 }
2886
2887 /*
2888  * Enable/disable discard support on RAID set depending on
2889  * RAID level and discard properties of underlying RAID members.
2890  */
2891 static void configure_discard_support(struct raid_set *rs)
2892 {
2893         int i;
2894         bool raid456;
2895         struct dm_target *ti = rs->ti;
2896
2897         /* Assume discards not supported until after checks below. */
2898         ti->discards_supported = false;
2899
2900         /*
2901          * XXX: RAID level 4,5,6 require zeroing for safety.
2902          */
2903         raid456 = (rs->md.level == 4 || rs->md.level == 5 || rs->md.level == 6);
2904
2905         for (i = 0; i < rs->raid_disks; i++) {
2906                 struct request_queue *q;
2907
2908                 if (!rs->dev[i].rdev.bdev)
2909                         continue;
2910
2911                 q = bdev_get_queue(rs->dev[i].rdev.bdev);
2912                 if (!q || !blk_queue_discard(q))
2913                         return;
2914
2915                 if (raid456) {
2916                         if (!devices_handle_discard_safely) {
2917                                 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2918                                 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2919                                 return;
2920                         }
2921                 }
2922         }
2923
2924         /* All RAID members properly support discards */
2925         ti->discards_supported = true;
2926
2927         /*
2928          * RAID1 and RAID10 personalities require bio splitting,
2929          * RAID0/4/5/6 don't and process large discard bios properly.
2930          */
2931         ti->split_discard_bios = !!(rs->md.level == 1 || rs->md.level == 10);
2932         ti->num_discard_bios = 1;
2933 }
2934
2935 /*
2936  * Construct a RAID0/1/10/4/5/6 mapping:
2937  * Args:
2938  *      <raid_type> <#raid_params> <raid_params>{0,}    \
2939  *      <#raid_devs> [<meta_dev1> <dev1>]{1,}
2940  *
2941  * <raid_params> varies by <raid_type>.  See 'parse_raid_params' for
2942  * details on possible <raid_params>.
2943  *
2944  * Userspace is free to initialize the metadata devices, hence the superblocks to
2945  * enforce recreation based on the passed in table parameters.
2946  *
2947  */
2948 static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2949 {
2950         int r;
2951         bool resize;
2952         struct raid_type *rt;
2953         unsigned int num_raid_params, num_raid_devs;
2954         sector_t calculated_dev_sectors, rdev_sectors;
2955         struct raid_set *rs = NULL;
2956         const char *arg;
2957         struct rs_layout rs_layout;
2958         struct dm_arg_set as = { argc, argv }, as_nrd;
2959         struct dm_arg _args[] = {
2960                 { 0, as.argc, "Cannot understand number of raid parameters" },
2961                 { 1, 254, "Cannot understand number of raid devices parameters" }
2962         };
2963
2964         /* Must have <raid_type> */
2965         arg = dm_shift_arg(&as);
2966         if (!arg) {
2967                 ti->error = "No arguments";
2968                 return -EINVAL;
2969         }
2970
2971         rt = get_raid_type(arg);
2972         if (!rt) {
2973                 ti->error = "Unrecognised raid_type";
2974                 return -EINVAL;
2975         }
2976
2977         /* Must have <#raid_params> */
2978         if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
2979                 return -EINVAL;
2980
2981         /* number of raid device tupples <meta_dev data_dev> */
2982         as_nrd = as;
2983         dm_consume_args(&as_nrd, num_raid_params);
2984         _args[1].max = (as_nrd.argc - 1) / 2;
2985         if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
2986                 return -EINVAL;
2987
2988         if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
2989                 ti->error = "Invalid number of supplied raid devices";
2990                 return -EINVAL;
2991         }
2992
2993         rs = raid_set_alloc(ti, rt, num_raid_devs);
2994         if (IS_ERR(rs))
2995                 return PTR_ERR(rs);
2996
2997         r = parse_raid_params(rs, &as, num_raid_params);
2998         if (r)
2999                 goto bad;
3000
3001         r = parse_dev_params(rs, &as);
3002         if (r)
3003                 goto bad;
3004
3005         rs->md.sync_super = super_sync;
3006
3007         /*
3008          * Calculate ctr requested array and device sizes to allow
3009          * for superblock analysis needing device sizes defined.
3010          *
3011          * Any existing superblock will overwrite the array and device sizes
3012          */
3013         r = rs_set_dev_and_array_sectors(rs, false);
3014         if (r)
3015                 goto bad;
3016
3017         calculated_dev_sectors = rs->md.dev_sectors;
3018
3019         /*
3020          * Backup any new raid set level, layout, ...
3021          * requested to be able to compare to superblock
3022          * members for conversion decisions.
3023          */
3024         rs_config_backup(rs, &rs_layout);
3025
3026         r = analyse_superblocks(ti, rs);
3027         if (r)
3028                 goto bad;
3029
3030         rdev_sectors = __rdev_sectors(rs);
3031         if (!rdev_sectors) {
3032                 ti->error = "Invalid rdev size";
3033                 r = -EINVAL;
3034                 goto bad;
3035         }
3036
3037         resize = calculated_dev_sectors != rdev_sectors;
3038
3039         INIT_WORK(&rs->md.event_work, do_table_event);
3040         ti->private = rs;
3041         ti->num_flush_bios = 1;
3042
3043         /* Restore any requested new layout for conversion decision */
3044         rs_config_restore(rs, &rs_layout);
3045
3046         /*
3047          * Now that we have any superblock metadata available,
3048          * check for new, recovering, reshaping, to be taken over,
3049          * to be reshaped or an existing, unchanged raid set to
3050          * run in sequence.
3051          */
3052         if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
3053                 /* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
3054                 if (rs_is_raid6(rs) &&
3055                     test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
3056                         ti->error = "'nosync' not allowed for new raid6 set";
3057                         r = -EINVAL;
3058                         goto bad;
3059                 }
3060                 rs_setup_recovery(rs, 0);
3061                 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3062                 rs_set_new(rs);
3063         } else if (rs_is_recovering(rs)) {
3064                 /* Rebuild particular devices */
3065                 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3066                         set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3067                         rs_setup_recovery(rs, MaxSector);
3068                 }
3069                 /* A recovering raid set may be resized */
3070                 ; /* skip setup rs */
3071         } else if (rs_is_reshaping(rs)) {
3072                 /* Have to reject size change request during reshape */
3073                 if (resize) {
3074                         ti->error = "Can't resize a reshaping raid set";
3075                         r = -EPERM;
3076                         goto bad;
3077                 }
3078                 /* skip setup rs */
3079         } else if (rs_takeover_requested(rs)) {
3080                 if (rs_is_reshaping(rs)) {
3081                         ti->error = "Can't takeover a reshaping raid set";
3082                         r = -EPERM;
3083                         goto bad;
3084                 }
3085
3086                 /* We can't takeover a journaled raid4/5/6 */
3087                 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3088                         ti->error = "Can't takeover a journaled raid4/5/6 set";
3089                         r = -EPERM;
3090                         goto bad;
3091                 }
3092
3093                 /*
3094                  * If a takeover is needed, userspace sets any additional
3095                  * devices to rebuild and we can check for a valid request here.
3096                  *
3097                  * If acceptible, set the level to the new requested
3098                  * one, prohibit requesting recovery, allow the raid
3099                  * set to run and store superblocks during resume.
3100                  */
3101                 r = rs_check_takeover(rs);
3102                 if (r)
3103                         goto bad;
3104
3105                 r = rs_setup_takeover(rs);
3106                 if (r)
3107                         goto bad;
3108
3109                 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3110                 /* Takeover ain't recovery, so disable recovery */
3111                 rs_setup_recovery(rs, MaxSector);
3112                 rs_set_new(rs);
3113         } else if (rs_reshape_requested(rs)) {
3114                 /*
3115                  * No need to check for 'ongoing' takeover here, because takeover
3116                  * is an instant operation as oposed to an ongoing reshape.
3117                  */
3118
3119                 /* We can't reshape a journaled raid4/5/6 */
3120                 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3121                         ti->error = "Can't reshape a journaled raid4/5/6 set";
3122                         r = -EPERM;
3123                         goto bad;
3124                 }
3125
3126                 /*
3127                   * We can only prepare for a reshape here, because the
3128                   * raid set needs to run to provide the repective reshape
3129                   * check functions via its MD personality instance.
3130                   *
3131                   * So do the reshape check after md_run() succeeded.
3132                   */
3133                 r = rs_prepare_reshape(rs);
3134                 if (r)
3135                         return r;
3136
3137                 /* Reshaping ain't recovery, so disable recovery */
3138                 rs_setup_recovery(rs, MaxSector);
3139                 rs_set_cur(rs);
3140         } else {
3141                 /* May not set recovery when a device rebuild is requested */
3142                 if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3143                         rs_setup_recovery(rs, MaxSector);
3144                         set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3145                 } else
3146                         rs_setup_recovery(rs, test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ?
3147                                               0 : (resize ? calculated_dev_sectors : MaxSector));
3148                 rs_set_cur(rs);
3149         }
3150
3151         /* If constructor requested it, change data and new_data offsets */
3152         r = rs_adjust_data_offsets(rs);
3153         if (r)
3154                 goto bad;
3155
3156         /* Start raid set read-only and assumed clean to change in raid_resume() */
3157         rs->md.ro = 1;
3158         rs->md.in_sync = 1;
3159         set_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
3160
3161         /* Has to be held on running the array */
3162         mddev_lock_nointr(&rs->md);
3163         r = md_run(&rs->md);
3164         rs->md.in_sync = 0; /* Assume already marked dirty */
3165
3166         if (r) {
3167                 ti->error = "Failed to run raid array";
3168                 mddev_unlock(&rs->md);
3169                 goto bad;
3170         }
3171
3172         rs->callbacks.congested_fn = raid_is_congested;
3173         dm_table_add_target_callbacks(ti->table, &rs->callbacks);
3174
3175         /* If raid4/5/6 journal mode explictely requested (only possible with journal dev) -> set it */
3176         if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
3177                 r = r5c_journal_mode_set(&rs->md, rs->journal_dev.mode);
3178                 if (r) {
3179                         ti->error = "Failed to set raid4/5/6 journal mode";
3180                         mddev_unlock(&rs->md);
3181                         goto bad_journal_mode_set;
3182                 }
3183         }
3184
3185         mddev_suspend(&rs->md);
3186         set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags);
3187
3188         /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
3189         if (rs_is_raid456(rs)) {
3190                 r = rs_set_raid456_stripe_cache(rs);
3191                 if (r)
3192                         goto bad_stripe_cache;
3193         }
3194
3195         /* Now do an early reshape check */
3196         if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3197                 r = rs_check_reshape(rs);
3198                 if (r)
3199                         goto bad_check_reshape;
3200
3201                 /* Restore new, ctr requested layout to perform check */
3202                 rs_config_restore(rs, &rs_layout);
3203
3204                 if (rs->md.pers->start_reshape) {
3205                         r = rs->md.pers->check_reshape(&rs->md);
3206                         if (r) {
3207                                 ti->error = "Reshape check failed";
3208                                 goto bad_check_reshape;
3209                         }
3210                 }
3211         }
3212
3213         /* Disable/enable discard support on raid set. */
3214         configure_discard_support(rs);
3215
3216         mddev_unlock(&rs->md);
3217         return 0;
3218
3219 bad_journal_mode_set:
3220 bad_stripe_cache:
3221 bad_check_reshape:
3222         md_stop(&rs->md);
3223 bad:
3224         raid_set_free(rs);
3225
3226         return r;
3227 }
3228
3229 static void raid_dtr(struct dm_target *ti)
3230 {
3231         struct raid_set *rs = ti->private;
3232
3233         list_del_init(&rs->callbacks.list);
3234         md_stop(&rs->md);
3235         raid_set_free(rs);
3236 }
3237
3238 static int raid_map(struct dm_target *ti, struct bio *bio)
3239 {
3240         struct raid_set *rs = ti->private;
3241         struct mddev *mddev = &rs->md;
3242
3243         /*
3244          * If we're reshaping to add disk(s)), ti->len and
3245          * mddev->array_sectors will differ during the process
3246          * (ti->len > mddev->array_sectors), so we have to requeue
3247          * bios with addresses > mddev->array_sectors here or
3248          * there will occur accesses past EOD of the component
3249          * data images thus erroring the raid set.
3250          */
3251         if (unlikely(bio_end_sector(bio) > mddev->array_sectors))
3252                 return DM_MAPIO_REQUEUE;
3253
3254         md_handle_request(mddev, bio);
3255
3256         return DM_MAPIO_SUBMITTED;
3257 }
3258
3259 /* Return string describing the current sync action of @mddev */
3260 static const char *decipher_sync_action(struct mddev *mddev)
3261 {
3262         if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3263                 return "frozen";
3264
3265         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3266             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3267                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3268                         return "reshape";
3269
3270                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3271                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3272                                 return "resync";
3273                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3274                                 return "check";
3275                         return "repair";
3276                 }
3277
3278                 if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3279                         return "recover";
3280         }
3281
3282         return "idle";
3283 }
3284
3285 /*
3286  * Return status string for @rdev
3287  *
3288  * Status characters:
3289  *
3290  *  'D' = Dead/Failed raid set component or raid4/5/6 journal device
3291  *  'a' = Alive but not in-sync raid set component _or_ alive raid4/5/6 'write_back' journal device
3292  *  'A' = Alive and in-sync raid set component _or_ alive raid4/5/6 'write_through' journal device
3293  *  '-' = Non-existing device (i.e. uspace passed '- -' into the ctr)
3294  */
3295 static const char *__raid_dev_status(struct raid_set *rs, struct md_rdev *rdev, bool array_in_sync)
3296 {
3297         if (!rdev->bdev)
3298                 return "-";
3299         else if (test_bit(Faulty, &rdev->flags))
3300                 return "D";
3301         else if (test_bit(Journal, &rdev->flags))
3302                 return (rs->journal_dev.mode == R5C_JOURNAL_MODE_WRITE_THROUGH) ? "A" : "a";
3303         else if (!array_in_sync || !test_bit(In_sync, &rdev->flags))
3304                 return "a";
3305         else
3306                 return "A";
3307 }
3308
3309 /* Helper to return resync/reshape progress for @rs and @array_in_sync */
3310 static sector_t rs_get_progress(struct raid_set *rs,
3311                                 sector_t resync_max_sectors, bool *array_in_sync)
3312 {
3313         sector_t r, curr_resync_completed;
3314         struct mddev *mddev = &rs->md;
3315
3316         curr_resync_completed = mddev->curr_resync_completed ?: mddev->recovery_cp;
3317         *array_in_sync = false;
3318
3319         if (rs_is_raid0(rs)) {
3320                 r = resync_max_sectors;
3321                 *array_in_sync = true;
3322
3323         } else {
3324                 r = mddev->reshape_position;
3325
3326                 /* Reshape is relative to the array size */
3327                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
3328                     r != MaxSector) {
3329                         if (r == MaxSector) {
3330                                 *array_in_sync = true;
3331                                 r = resync_max_sectors;
3332                         } else {
3333                                 /* Got to reverse on backward reshape */
3334                                 if (mddev->reshape_backwards)
3335                                         r = mddev->array_sectors - r;
3336
3337                                 /* Devide by # of data stripes */
3338                                 sector_div(r, mddev_data_stripes(rs));
3339                         }
3340
3341                 /* Sync is relative to the component device size */
3342                 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3343                         r = curr_resync_completed;
3344                 else
3345                         r = mddev->recovery_cp;
3346
3347                 if ((r == MaxSector) ||
3348                     (test_bit(MD_RECOVERY_DONE, &mddev->recovery) &&
3349                      (mddev->curr_resync_completed == resync_max_sectors))) {
3350                         /*
3351                          * Sync complete.
3352                          */
3353                         *array_in_sync = true;
3354                         r = resync_max_sectors;
3355                 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
3356                         /*
3357                          * If "check" or "repair" is occurring, the raid set has
3358                          * undergone an initial sync and the health characters
3359                          * should not be 'a' anymore.
3360                          */
3361                         *array_in_sync = true;
3362                 } else {
3363                         struct md_rdev *rdev;
3364
3365                         /*
3366                          * The raid set may be doing an initial sync, or it may
3367                          * be rebuilding individual components.  If all the
3368                          * devices are In_sync, then it is the raid set that is
3369                          * being initialized.
3370                          */
3371                         rdev_for_each(rdev, mddev)
3372                                 if (!test_bit(Journal, &rdev->flags) &&
3373                                     !test_bit(In_sync, &rdev->flags))
3374                                         *array_in_sync = true;
3375 #if 0
3376                         r = 0; /* HM FIXME: TESTME: https://bugzilla.redhat.com/show_bug.cgi?id=1210637 ? */
3377 #endif
3378                 }
3379         }
3380
3381         return r;
3382 }
3383
3384 /* Helper to return @dev name or "-" if !@dev */
3385 static const char *__get_dev_name(struct dm_dev *dev)
3386 {
3387         return dev ? dev->name : "-";
3388 }
3389
3390 static void raid_status(struct dm_target *ti, status_type_t type,
3391                         unsigned int status_flags, char *result, unsigned int maxlen)
3392 {
3393         struct raid_set *rs = ti->private;
3394         struct mddev *mddev = &rs->md;
3395         struct r5conf *conf = mddev->private;
3396         int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3397         bool array_in_sync;
3398         unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3399         unsigned int sz = 0;
3400         unsigned int rebuild_disks;
3401         unsigned int write_mostly_params = 0;
3402         sector_t progress, resync_max_sectors, resync_mismatches;
3403         const char *sync_action;
3404         struct raid_type *rt;
3405
3406         switch (type) {
3407         case STATUSTYPE_INFO:
3408                 /* *Should* always succeed */
3409                 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3410                 if (!rt)
3411                         return;
3412
3413                 DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3414
3415                 /* Access most recent mddev properties for status output */
3416                 smp_rmb();
3417                 /* Get sensible max sectors even if raid set not yet started */
3418                 resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3419                                       mddev->resync_max_sectors : mddev->dev_sectors;
3420                 progress = rs_get_progress(rs, resync_max_sectors, &array_in_sync);
3421                 resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
3422                                     atomic64_read(&mddev->resync_mismatches) : 0;
3423                 sync_action = decipher_sync_action(&rs->md);
3424
3425                 /* HM FIXME: do we want another state char for raid0? It shows 'D'/'A'/'-' now */
3426                 for (i = 0; i < rs->raid_disks; i++)
3427                         DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev, array_in_sync));
3428
3429                 /*
3430                  * In-sync/Reshape ratio:
3431                  *  The in-sync ratio shows the progress of:
3432                  *   - Initializing the raid set
3433                  *   - Rebuilding a subset of devices of the raid set
3434                  *  The user can distinguish between the two by referring
3435                  *  to the status characters.
3436                  *
3437                  *  The reshape ratio shows the progress of
3438                  *  changing the raid layout or the number of
3439                  *  disks of a raid set
3440                  */
3441                 DMEMIT(" %llu/%llu", (unsigned long long) progress,
3442                                      (unsigned long long) resync_max_sectors);
3443
3444                 /*
3445                  * v1.5.0+:
3446                  *
3447                  * Sync action:
3448                  *   See Documentation/device-mapper/dm-raid.txt for
3449                  *   information on each of these states.
3450                  */
3451                 DMEMIT(" %s", sync_action);
3452
3453                 /*
3454                  * v1.5.0+:
3455                  *
3456                  * resync_mismatches/mismatch_cnt
3457                  *   This field shows the number of discrepancies found when
3458                  *   performing a "check" of the raid set.
3459                  */
3460                 DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3461
3462                 /*
3463                  * v1.9.0+:
3464                  *
3465                  * data_offset (needed for out of space reshaping)
3466                  *   This field shows the data offset into the data
3467                  *   image LV where the first stripes data starts.
3468                  *
3469                  * We keep data_offset equal on all raid disks of the set,
3470                  * so retrieving it from the first raid disk is sufficient.
3471                  */
3472                 DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3473
3474                 /*
3475                  * v1.10.0+:
3476                  */
3477                 DMEMIT(" %s", test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ?
3478                               __raid_dev_status(rs, &rs->journal_dev.rdev, 0) : "-");
3479                 break;
3480
3481         case STATUSTYPE_TABLE:
3482                 /* Report the table line string you would use to construct this raid set */
3483
3484                 /* Calculate raid parameter count */
3485                 for (i = 0; i < rs->raid_disks; i++)
3486                         if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3487                                 write_mostly_params += 2;
3488                 rebuild_disks = memweight(rs->rebuild_disks, DISKS_ARRAY_ELEMS * sizeof(*rs->rebuild_disks));
3489                 raid_param_cnt += rebuild_disks * 2 +
3490                                   write_mostly_params +
3491                                   hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3492                                   hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2 +
3493                                   (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ? 2 : 0) +
3494                                   (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags) ? 2 : 0);
3495
3496                 /* Emit table line */
3497                 /* This has to be in the documented order for userspace! */
3498                 DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3499                 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3500                         DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3501                 if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3502                         DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3503                 if (rebuild_disks)
3504                         for (i = 0; i < rs->raid_disks; i++)
3505                                 if (test_bit(rs->dev[i].rdev.raid_disk, (void *) rs->rebuild_disks))
3506                                         DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD),
3507                                                          rs->dev[i].rdev.raid_disk);
3508                 if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3509                         DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3510                                           mddev->bitmap_info.daemon_sleep);
3511                 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3512                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3513                                          mddev->sync_speed_min);
3514                 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3515                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3516                                          mddev->sync_speed_max);
3517                 if (write_mostly_params)
3518                         for (i = 0; i < rs->raid_disks; i++)
3519                                 if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3520                                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3521                                                rs->dev[i].rdev.raid_disk);
3522                 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3523                         DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3524                                           mddev->bitmap_info.max_write_behind);
3525                 if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3526                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3527                                          max_nr_stripes);
3528                 if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3529                         DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3530                                            (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3531                 if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3532                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3533                                          raid10_md_layout_to_copies(mddev->layout));
3534                 if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3535                         DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3536                                          raid10_md_layout_to_format(mddev->layout));
3537                 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3538                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3539                                          max(rs->delta_disks, mddev->delta_disks));
3540                 if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3541                         DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3542                                            (unsigned long long) rs->data_offset);
3543                 if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags))
3544                         DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV),
3545                                         __get_dev_name(rs->journal_dev.dev));
3546                 if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags))
3547                         DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE),
3548                                          md_journal_mode_to_dm_raid(rs->journal_dev.mode));
3549                 DMEMIT(" %d", rs->raid_disks);
3550                 for (i = 0; i < rs->raid_disks; i++)
3551                         DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3552                                          __get_dev_name(rs->dev[i].data_dev));
3553         }
3554 }
3555
3556 static int raid_message(struct dm_target *ti, unsigned int argc, char **argv)
3557 {
3558         struct raid_set *rs = ti->private;
3559         struct mddev *mddev = &rs->md;
3560
3561         if (!mddev->pers || !mddev->pers->sync_request)
3562                 return -EINVAL;
3563
3564         if (!strcasecmp(argv[0], "frozen"))
3565                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3566         else
3567                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3568
3569         if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
3570                 if (mddev->sync_thread) {
3571                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3572                         md_reap_sync_thread(mddev);
3573                 }
3574         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3575                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3576                 return -EBUSY;
3577         else if (!strcasecmp(argv[0], "resync"))
3578                 ; /* MD_RECOVERY_NEEDED set below */
3579         else if (!strcasecmp(argv[0], "recover"))
3580                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3581         else {
3582                 if (!strcasecmp(argv[0], "check")) {
3583                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3584                         set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3585                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3586                 } else if (!strcasecmp(argv[0], "repair")) {
3587                         set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3588                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3589                 } else
3590                         return -EINVAL;
3591         }
3592         if (mddev->ro == 2) {
3593                 /* A write to sync_action is enough to justify
3594                  * canceling read-auto mode
3595                  */
3596                 mddev->ro = 0;
3597                 if (!mddev->suspended && mddev->sync_thread)
3598                         md_wakeup_thread(mddev->sync_thread);
3599         }
3600         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3601         if (!mddev->suspended && mddev->thread)
3602                 md_wakeup_thread(mddev->thread);
3603
3604         return 0;
3605 }
3606
3607 static int raid_iterate_devices(struct dm_target *ti,
3608                                 iterate_devices_callout_fn fn, void *data)
3609 {
3610         struct raid_set *rs = ti->private;
3611         unsigned int i;
3612         int r = 0;
3613
3614         for (i = 0; !r && i < rs->md.raid_disks; i++)
3615                 if (rs->dev[i].data_dev)
3616                         r = fn(ti,
3617                                  rs->dev[i].data_dev,
3618                                  0, /* No offset on data devs */
3619                                  rs->md.dev_sectors,
3620                                  data);
3621
3622         return r;
3623 }
3624
3625 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3626 {
3627         struct raid_set *rs = ti->private;
3628         unsigned int chunk_size = to_bytes(rs->md.chunk_sectors);
3629
3630         blk_limits_io_min(limits, chunk_size);
3631         blk_limits_io_opt(limits, chunk_size * mddev_data_stripes(rs));
3632 }
3633
3634 static void raid_presuspend(struct dm_target *ti)
3635 {
3636         struct raid_set *rs = ti->private;
3637
3638         md_stop_writes(&rs->md);
3639 }
3640
3641 static void raid_postsuspend(struct dm_target *ti)
3642 {
3643         struct raid_set *rs = ti->private;
3644
3645         if (!test_and_set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
3646                 mddev_lock_nointr(&rs->md);
3647                 mddev_suspend(&rs->md);
3648                 mddev_unlock(&rs->md);
3649         }
3650
3651         rs->md.ro = 1;
3652 }
3653
3654 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3655 {
3656         int i;
3657         uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
3658         unsigned long flags;
3659         bool cleared = false;
3660         struct dm_raid_superblock *sb;
3661         struct mddev *mddev = &rs->md;
3662         struct md_rdev *r;
3663
3664         /* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3665         if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3666                 return;
3667
3668         memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3669
3670         for (i = 0; i < mddev->raid_disks; i++) {
3671                 r = &rs->dev[i].rdev;
3672                 /* HM FIXME: enhance journal device recovery processing */
3673                 if (test_bit(Journal, &r->flags))
3674                         continue;
3675
3676                 if (test_bit(Faulty, &r->flags) &&
3677                     r->meta_bdev && !read_disk_sb(r, r->sb_size, true)) {
3678                         DMINFO("Faulty %s device #%d has readable super block."
3679                                "  Attempting to revive it.",
3680                                rs->raid_type->name, i);
3681
3682                         /*
3683                          * Faulty bit may be set, but sometimes the array can
3684                          * be suspended before the personalities can respond
3685                          * by removing the device from the array (i.e. calling
3686                          * 'hot_remove_disk').  If they haven't yet removed
3687                          * the failed device, its 'raid_disk' number will be
3688                          * '>= 0' - meaning we must call this function
3689                          * ourselves.
3690                          */
3691                         flags = r->flags;
3692                         clear_bit(In_sync, &r->flags); /* Mandatory for hot remove. */
3693                         if (r->raid_disk >= 0) {
3694                                 if (mddev->pers->hot_remove_disk(mddev, r)) {
3695                                         /* Failed to revive this device, try next */
3696                                         r->flags = flags;
3697                                         continue;
3698                                 }
3699                         } else
3700                                 r->raid_disk = r->saved_raid_disk = i;
3701
3702                         clear_bit(Faulty, &r->flags);
3703                         clear_bit(WriteErrorSeen, &r->flags);
3704
3705                         if (mddev->pers->hot_add_disk(mddev, r)) {
3706                                 /* Failed to revive this device, try next */
3707                                 r->raid_disk = r->saved_raid_disk = -1;
3708                                 r->flags = flags;
3709                         } else {
3710                                 clear_bit(In_sync, &r->flags);
3711                                 r->recovery_offset = 0;
3712                                 set_bit(i, (void *) cleared_failed_devices);
3713                                 cleared = true;
3714                         }
3715                 }
3716         }
3717
3718         /* If any failed devices could be cleared, update all sbs failed_devices bits */
3719         if (cleared) {
3720                 uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3721
3722                 rdev_for_each(r, &rs->md) {
3723                         if (test_bit(Journal, &r->flags))
3724                                 continue;
3725
3726                         sb = page_address(r->sb_page);
3727                         sb_retrieve_failed_devices(sb, failed_devices);
3728
3729                         for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3730                                 failed_devices[i] &= ~cleared_failed_devices[i];
3731
3732                         sb_update_failed_devices(sb, failed_devices);
3733                 }
3734         }
3735 }
3736
3737 static int __load_dirty_region_bitmap(struct raid_set *rs)
3738 {
3739         int r = 0;
3740
3741         /* Try loading the bitmap unless "raid0", which does not have one */
3742         if (!rs_is_raid0(rs) &&
3743             !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3744                 r = bitmap_load(&rs->md);
3745                 if (r)
3746                         DMERR("Failed to load bitmap");
3747         }
3748
3749         return r;
3750 }
3751
3752 /* Enforce updating all superblocks */
3753 static void rs_update_sbs(struct raid_set *rs)
3754 {
3755         struct mddev *mddev = &rs->md;
3756         int ro = mddev->ro;
3757
3758         set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3759         mddev->ro = 0;
3760         md_update_sb(mddev, 1);
3761         mddev->ro = ro;
3762 }
3763
3764 /*
3765  * Reshape changes raid algorithm of @rs to new one within personality
3766  * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3767  * disks from a raid set thus growing/shrinking it or resizes the set
3768  *
3769  * Call mddev_lock_nointr() before!
3770  */
3771 static int rs_start_reshape(struct raid_set *rs)
3772 {
3773         int r;
3774         struct mddev *mddev = &rs->md;
3775         struct md_personality *pers = mddev->pers;
3776
3777         r = rs_setup_reshape(rs);
3778         if (r)
3779                 return r;
3780
3781         /* Need to be resumed to be able to start reshape, recovery is frozen until raid_resume() though */
3782         if (test_and_clear_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags))
3783                 mddev_resume(mddev);
3784
3785         /*
3786          * Check any reshape constraints enforced by the personalility
3787          *
3788          * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3789          */
3790         r = pers->check_reshape(mddev);
3791         if (r) {
3792                 rs->ti->error = "pers->check_reshape() failed";
3793                 return r;
3794         }
3795
3796         /*
3797          * Personality may not provide start reshape method in which
3798          * case check_reshape above has already covered everything
3799          */
3800         if (pers->start_reshape) {
3801                 r = pers->start_reshape(mddev);
3802                 if (r) {
3803                         rs->ti->error = "pers->start_reshape() failed";
3804                         return r;
3805                 }
3806         }
3807
3808         /* Suspend because a resume will happen in raid_resume() */
3809         set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags);
3810         mddev_suspend(mddev);
3811
3812         /*
3813          * Now reshape got set up, update superblocks to
3814          * reflect the fact so that a table reload will
3815          * access proper superblock content in the ctr.
3816          */
3817         rs_update_sbs(rs);
3818
3819         return 0;
3820 }
3821
3822 static int raid_preresume(struct dm_target *ti)
3823 {
3824         int r;
3825         struct raid_set *rs = ti->private;
3826         struct mddev *mddev = &rs->md;
3827
3828         /* This is a resume after a suspend of the set -> it's already started */
3829         if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
3830                 return 0;
3831
3832         /*
3833          * The superblocks need to be updated on disk if the
3834          * array is new or new devices got added (thus zeroed
3835          * out by userspace) or __load_dirty_region_bitmap
3836          * will overwrite them in core with old data or fail.
3837          */
3838         if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
3839                 rs_update_sbs(rs);
3840
3841         /* Load the bitmap from disk unless raid0 */
3842         r = __load_dirty_region_bitmap(rs);
3843         if (r)
3844                 return r;
3845
3846         /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) */
3847         if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) && mddev->bitmap &&
3848             mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)) {
3849                 r = bitmap_resize(mddev->bitmap, mddev->dev_sectors,
3850                                   to_bytes(rs->requested_bitmap_chunk_sectors), 0);
3851                 if (r)
3852                         DMERR("Failed to resize bitmap");
3853         }
3854
3855         /* Check for any resize/reshape on @rs and adjust/initiate */
3856         /* Be prepared for mddev_resume() in raid_resume() */
3857         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3858         if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
3859                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3860                 mddev->resync_min = mddev->recovery_cp;
3861         }
3862
3863         /* Check for any reshape request unless new raid set */
3864         if (test_and_clear_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3865                 /* Initiate a reshape. */
3866                 rs_set_rdev_sectors(rs);
3867                 mddev_lock_nointr(mddev);
3868                 r = rs_start_reshape(rs);
3869                 mddev_unlock(mddev);
3870                 if (r)
3871                         DMWARN("Failed to check/start reshape, continuing without change");
3872                 r = 0;
3873         }
3874
3875         return r;
3876 }
3877
3878 static void raid_resume(struct dm_target *ti)
3879 {
3880         struct raid_set *rs = ti->private;
3881         struct mddev *mddev = &rs->md;
3882
3883         if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
3884                 /*
3885                  * A secondary resume while the device is active.
3886                  * Take this opportunity to check whether any failed
3887                  * devices are reachable again.
3888                  */
3889                 attempt_restore_of_faulty_devices(rs);
3890         }
3891
3892         mddev->ro = 0;
3893         mddev->in_sync = 0;
3894
3895         /* Only reduce raid set size before running a disk removing reshape. */
3896         if (mddev->delta_disks < 0)
3897                 rs_set_capacity(rs);
3898
3899         /*
3900          * Keep the RAID set frozen if reshape/rebuild flags are set.
3901          * The RAID set is unfrozen once the next table load/resume,
3902          * which clears the reshape/rebuild flags, occurs.
3903          * This ensures that the constructor for the inactive table
3904          * retrieves an up-to-date reshape_position.
3905          */
3906         if (!(rs->ctr_flags & RESUME_STAY_FROZEN_FLAGS))
3907                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3908
3909         if (test_and_clear_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
3910                 mddev_lock_nointr(mddev);
3911                 mddev_resume(mddev);
3912                 mddev_unlock(mddev);
3913         }
3914 }
3915
3916 static struct target_type raid_target = {
3917         .name = "raid",
3918         .version = {1, 13, 0},
3919         .module = THIS_MODULE,
3920         .ctr = raid_ctr,
3921         .dtr = raid_dtr,
3922         .map = raid_map,
3923         .status = raid_status,
3924         .message = raid_message,
3925         .iterate_devices = raid_iterate_devices,
3926         .io_hints = raid_io_hints,
3927         .presuspend = raid_presuspend,
3928         .postsuspend = raid_postsuspend,
3929         .preresume = raid_preresume,
3930         .resume = raid_resume,
3931 };
3932
3933 static int __init dm_raid_init(void)
3934 {
3935         DMINFO("Loading target version %u.%u.%u",
3936                raid_target.version[0],
3937                raid_target.version[1],
3938                raid_target.version[2]);
3939         return dm_register_target(&raid_target);
3940 }
3941
3942 static void __exit dm_raid_exit(void)
3943 {
3944         dm_unregister_target(&raid_target);
3945 }
3946
3947 module_init(dm_raid_init);
3948 module_exit(dm_raid_exit);
3949
3950 module_param(devices_handle_discard_safely, bool, 0644);
3951 MODULE_PARM_DESC(devices_handle_discard_safely,
3952                  "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
3953
3954 MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
3955 MODULE_ALIAS("dm-raid0");
3956 MODULE_ALIAS("dm-raid1");
3957 MODULE_ALIAS("dm-raid10");
3958 MODULE_ALIAS("dm-raid4");
3959 MODULE_ALIAS("dm-raid5");
3960 MODULE_ALIAS("dm-raid6");
3961 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
3962 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
3963 MODULE_LICENSE("GPL");