drivers/md/dm-raid.c

   1 /*
   2  * Copyright (C) 2010-2011 Neil Brown
   3  * Copyright (C) 2010-2016 Red Hat, Inc. All rights reserved.
   4  *
   5  * This file is released under the GPL.
   6  */
   7
   8 #include <linux/slab.h>
   9 #include <linux/module.h>
  10
  11 #include "md.h"
  12 #include "raid1.h"
  13 #include "raid5.h"
  14 #include "raid10.h"
  15 #include "bitmap.h"
  16
  17 #include <linux/device-mapper.h>
  18
  19 #define DM_MSG_PREFIX "raid"
  20 #define MAX_RAID_DEVICES        253 /* md-raid kernel limit */
  21
  22 /*
  23  * Minimum sectors of free reshape space per raid device
  24  */
  25 #define MIN_FREE_RESHAPE_SPACE to_sector(4*4096)
  26
  27 static bool devices_handle_discard_safely = false;
  28
  29 /*
  30  * The following flags are used by dm-raid.c to set up the array state.
  31  * They must be cleared before md_run is called.
  32  */
  33 #define FirstUse 10             /* rdev flag */
  34
  35 struct raid_dev {
  36         /*
  37          * Two DM devices, one to hold metadata and one to hold the
  38          * actual data/parity.  The reason for this is to not confuse
  39          * ti->len and give more flexibility in altering size and
  40          * characteristics.
  41          *
  42          * While it is possible for this device to be associated
  43          * with a different physical device than the data_dev, it
  44          * is intended for it to be the same.
  45          *    |--------- Physical Device ---------|
  46          *    |- meta_dev -|------ data_dev ------|
  47          */
  48         struct dm_dev *meta_dev;
  49         struct dm_dev *data_dev;
  50         struct md_rdev rdev;
  51 };
  52
  53 /*
  54  * Bits for establishing rs->ctr_flags
  55  *
  56  * 1 = no flag value
  57  * 2 = flag with value
  58  */
  59 #define __CTR_FLAG_SYNC                 0  /* 1 */ /* Not with raid0! */
  60 #define __CTR_FLAG_NOSYNC               1  /* 1 */ /* Not with raid0! */
  61 #define __CTR_FLAG_REBUILD              2  /* 2 */ /* Not with raid0! */
  62 #define __CTR_FLAG_DAEMON_SLEEP         3  /* 2 */ /* Not with raid0! */
  63 #define __CTR_FLAG_MIN_RECOVERY_RATE    4  /* 2 */ /* Not with raid0! */
  64 #define __CTR_FLAG_MAX_RECOVERY_RATE    5  /* 2 */ /* Not with raid0! */
  65 #define __CTR_FLAG_MAX_WRITE_BEHIND     6  /* 2 */ /* Only with raid1! */
  66 #define __CTR_FLAG_WRITE_MOSTLY         7  /* 2 */ /* Only with raid1! */
  67 #define __CTR_FLAG_STRIPE_CACHE         8  /* 2 */ /* Only with raid4/5/6! */
  68 #define __CTR_FLAG_REGION_SIZE          9  /* 2 */ /* Not with raid0! */
  69 #define __CTR_FLAG_RAID10_COPIES        10 /* 2 */ /* Only with raid10 */
  70 #define __CTR_FLAG_RAID10_FORMAT        11 /* 2 */ /* Only with raid10 */
  71 /* New for v1.9.0 */
  72 #define __CTR_FLAG_DELTA_DISKS          12 /* 2 */ /* Only with reshapable raid4/5/6/10! */
  73 #define __CTR_FLAG_DATA_OFFSET          13 /* 2 */ /* Only with reshapable raid4/5/6/10! */
  74 #define __CTR_FLAG_RAID10_USE_NEAR_SETS 14 /* 2 */ /* Only with raid10! */
  75
  76 /*
  77  * Flags for rs->ctr_flags field.
  78  */
  79 #define CTR_FLAG_SYNC                   (1 << __CTR_FLAG_SYNC)
  80 #define CTR_FLAG_NOSYNC                 (1 << __CTR_FLAG_NOSYNC)
  81 #define CTR_FLAG_REBUILD                (1 << __CTR_FLAG_REBUILD)
  82 #define CTR_FLAG_DAEMON_SLEEP           (1 << __CTR_FLAG_DAEMON_SLEEP)
  83 #define CTR_FLAG_MIN_RECOVERY_RATE      (1 << __CTR_FLAG_MIN_RECOVERY_RATE)
  84 #define CTR_FLAG_MAX_RECOVERY_RATE      (1 << __CTR_FLAG_MAX_RECOVERY_RATE)
  85 #define CTR_FLAG_MAX_WRITE_BEHIND       (1 << __CTR_FLAG_MAX_WRITE_BEHIND)
  86 #define CTR_FLAG_WRITE_MOSTLY           (1 << __CTR_FLAG_WRITE_MOSTLY)
  87 #define CTR_FLAG_STRIPE_CACHE           (1 << __CTR_FLAG_STRIPE_CACHE)
  88 #define CTR_FLAG_REGION_SIZE            (1 << __CTR_FLAG_REGION_SIZE)
  89 #define CTR_FLAG_RAID10_COPIES          (1 << __CTR_FLAG_RAID10_COPIES)
  90 #define CTR_FLAG_RAID10_FORMAT          (1 << __CTR_FLAG_RAID10_FORMAT)
  91 #define CTR_FLAG_DELTA_DISKS            (1 << __CTR_FLAG_DELTA_DISKS)
  92 #define CTR_FLAG_DATA_OFFSET            (1 << __CTR_FLAG_DATA_OFFSET)
  93 #define CTR_FLAG_RAID10_USE_NEAR_SETS   (1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
  94
  95 /*
  96  * Definitions of various constructor flags to
  97  * be used in checks of valid / invalid flags
  98  * per raid level.
  99  */
 100 /* Define all any sync flags */
 101 #define CTR_FLAGS_ANY_SYNC              (CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
 102
 103 /* Define flags for options without argument (e.g. 'nosync') */
 104 #define CTR_FLAG_OPTIONS_NO_ARGS        (CTR_FLAGS_ANY_SYNC | \
 105                                          CTR_FLAG_RAID10_USE_NEAR_SETS)
 106
 107 /* Define flags for options with one argument (e.g. 'delta_disks +2') */
 108 #define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
 109                                   CTR_FLAG_WRITE_MOSTLY | \
 110                                   CTR_FLAG_DAEMON_SLEEP | \
 111                                   CTR_FLAG_MIN_RECOVERY_RATE | \
 112                                   CTR_FLAG_MAX_RECOVERY_RATE | \
 113                                   CTR_FLAG_MAX_WRITE_BEHIND | \
 114                                   CTR_FLAG_STRIPE_CACHE | \
 115                                   CTR_FLAG_REGION_SIZE | \
 116                                   CTR_FLAG_RAID10_COPIES | \
 117                                   CTR_FLAG_RAID10_FORMAT | \
 118                                   CTR_FLAG_DELTA_DISKS | \
 119                                   CTR_FLAG_DATA_OFFSET)
 120
 121 /* Valid options definitions per raid level... */
 122
 123 /* "raid0" does only accept data offset */
 124 #define RAID0_VALID_FLAGS       (CTR_FLAG_DATA_OFFSET)
 125
 126 /* "raid1" does not accept stripe cache, data offset, delta_disks or any raid10 options */
 127 #define RAID1_VALID_FLAGS       (CTR_FLAGS_ANY_SYNC | \
 128                                  CTR_FLAG_REBUILD | \
 129                                  CTR_FLAG_WRITE_MOSTLY | \
 130                                  CTR_FLAG_DAEMON_SLEEP | \
 131                                  CTR_FLAG_MIN_RECOVERY_RATE | \
 132                                  CTR_FLAG_MAX_RECOVERY_RATE | \
 133                                  CTR_FLAG_MAX_WRITE_BEHIND | \
 134                                  CTR_FLAG_REGION_SIZE | \
 135                                  CTR_FLAG_DATA_OFFSET)
 136
 137 /* "raid10" does not accept any raid1 or stripe cache options */
 138 #define RAID10_VALID_FLAGS      (CTR_FLAGS_ANY_SYNC | \
 139                                  CTR_FLAG_REBUILD | \
 140                                  CTR_FLAG_DAEMON_SLEEP | \
 141                                  CTR_FLAG_MIN_RECOVERY_RATE | \
 142                                  CTR_FLAG_MAX_RECOVERY_RATE | \
 143                                  CTR_FLAG_REGION_SIZE | \
 144                                  CTR_FLAG_RAID10_COPIES | \
 145                                  CTR_FLAG_RAID10_FORMAT | \
 146                                  CTR_FLAG_DELTA_DISKS | \
 147                                  CTR_FLAG_DATA_OFFSET | \
 148                                  CTR_FLAG_RAID10_USE_NEAR_SETS)
 149
 150 /*
 151  * "raid4/5/6" do not accept any raid1 or raid10 specific options
 152  *
 153  * "raid6" does not accept "nosync", because it is not guaranteed
 154  * that both parity and q-syndrome are being written properly with
 155  * any writes
 156  */
 157 #define RAID45_VALID_FLAGS      (CTR_FLAGS_ANY_SYNC | \
 158                                  CTR_FLAG_REBUILD | \
 159                                  CTR_FLAG_DAEMON_SLEEP | \
 160                                  CTR_FLAG_MIN_RECOVERY_RATE | \
 161                                  CTR_FLAG_MAX_RECOVERY_RATE | \
 162                                  CTR_FLAG_MAX_WRITE_BEHIND | \
 163                                  CTR_FLAG_STRIPE_CACHE | \
 164                                  CTR_FLAG_REGION_SIZE | \
 165                                  CTR_FLAG_DELTA_DISKS | \
 166                                  CTR_FLAG_DATA_OFFSET)
 167
 168 #define RAID6_VALID_FLAGS       (CTR_FLAG_SYNC | \
 169                                  CTR_FLAG_REBUILD | \
 170                                  CTR_FLAG_DAEMON_SLEEP | \
 171                                  CTR_FLAG_MIN_RECOVERY_RATE | \
 172                                  CTR_FLAG_MAX_RECOVERY_RATE | \
 173                                  CTR_FLAG_MAX_WRITE_BEHIND | \
 174                                  CTR_FLAG_STRIPE_CACHE | \
 175                                  CTR_FLAG_REGION_SIZE | \
 176                                  CTR_FLAG_DELTA_DISKS | \
 177                                  CTR_FLAG_DATA_OFFSET)
 178 /* ...valid options definitions per raid level */
 179
 180 /*
 181  * Flags for rs->runtime_flags field
 182  * (RT_FLAG prefix meaning "runtime flag")
 183  *
 184  * These are all internal and used to define runtime state,
 185  * e.g. to prevent another resume from preresume processing
 186  * the raid set all over again.
 187  */
 188 #define RT_FLAG_RS_PRERESUMED           0
 189 #define RT_FLAG_RS_RESUMED              1
 190 #define RT_FLAG_RS_BITMAP_LOADED        2
 191 #define RT_FLAG_UPDATE_SBS              3
 192 #define RT_FLAG_RESHAPE_RS              4
 193
 194 /* Array elements of 64 bit needed for rebuild/write_mostly bits */
 195 #define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
 196
 197 /*
 198  * raid set level, layout and chunk sectors backup/restore
 199  */
 200 struct rs_layout {
 201         int new_level;
 202         int new_layout;
 203         int new_chunk_sectors;
 204 };
 205
 206 struct raid_set {
 207         struct dm_target *ti;
 208
 209         uint32_t bitmap_loaded;
 210         uint32_t stripe_cache_entries;
 211         unsigned long ctr_flags;
 212         unsigned long runtime_flags;
 213
 214         uint64_t rebuild_disks[DISKS_ARRAY_ELEMS];
 215
 216         int raid_disks;
 217         int delta_disks;
 218         int data_offset;
 219         int raid10_copies;
 220         int requested_bitmap_chunk_sectors;
 221
 222         struct mddev md;
 223         struct raid_type *raid_type;
 224         struct dm_target_callbacks callbacks;
 225
 226         struct raid_dev dev[0];
 227 };
 228
 229 static void rs_config_backup(struct raid_set *rs, struct rs_layout *l)
 230 {
 231         struct mddev *mddev = &rs->md;
 232
 233         l->new_level = mddev->new_level;
 234         l->new_layout = mddev->new_layout;
 235         l->new_chunk_sectors = mddev->new_chunk_sectors;
 236 }
 237
 238 static void rs_config_restore(struct raid_set *rs, struct rs_layout *l)
 239 {
 240         struct mddev *mddev = &rs->md;
 241
 242         mddev->new_level = l->new_level;
 243         mddev->new_layout = l->new_layout;
 244         mddev->new_chunk_sectors = l->new_chunk_sectors;
 245 }
 246
 247 /* raid10 algorithms (i.e. formats) */
 248 #define ALGORITHM_RAID10_DEFAULT        0
 249 #define ALGORITHM_RAID10_NEAR           1
 250 #define ALGORITHM_RAID10_OFFSET         2
 251 #define ALGORITHM_RAID10_FAR            3
 252
 253 /* Supported raid types and properties. */
 254 static struct raid_type {
 255         const char *name;               /* RAID algorithm. */
 256         const char *descr;              /* Descriptor text for logging. */
 257         const unsigned parity_devs;     /* # of parity devices. */
 258         const unsigned minimal_devs;    /* minimal # of devices in set. */
 259         const unsigned level;           /* RAID level. */
 260         const unsigned algorithm;       /* RAID algorithm. */
 261 } raid_types[] = {
 262         {"raid0",         "raid0 (striping)",                       0, 2, 0,  0 /* NONE */},
 263         {"raid1",         "raid1 (mirroring)",                      0, 2, 1,  0 /* NONE */},
 264         {"raid10_far",    "raid10 far (striped mirrors)",           0, 2, 10, ALGORITHM_RAID10_FAR},
 265         {"raid10_offset", "raid10 offset (striped mirrors)",        0, 2, 10, ALGORITHM_RAID10_OFFSET},
 266         {"raid10_near",   "raid10 near (striped mirrors)",          0, 2, 10, ALGORITHM_RAID10_NEAR},
 267         {"raid10",        "raid10 (striped mirrors)",               0, 2, 10, ALGORITHM_RAID10_DEFAULT},
 268         {"raid4",         "raid4 (dedicated last parity disk)",     1, 2, 4,  ALGORITHM_PARITY_N}, /* raid4 layout = raid5_n */
 269         {"raid5_n",       "raid5 (dedicated last parity disk)",     1, 2, 5,  ALGORITHM_PARITY_N},
 270         {"raid5_ls",      "raid5 (left symmetric)",                 1, 2, 5,  ALGORITHM_LEFT_SYMMETRIC},
 271         {"raid5_rs",      "raid5 (right symmetric)",                1, 2, 5,  ALGORITHM_RIGHT_SYMMETRIC},
 272         {"raid5_la",      "raid5 (left asymmetric)",                1, 2, 5,  ALGORITHM_LEFT_ASYMMETRIC},
 273         {"raid5_ra",      "raid5 (right asymmetric)",               1, 2, 5,  ALGORITHM_RIGHT_ASYMMETRIC},
 274         {"raid6_zr",      "raid6 (zero restart)",                   2, 4, 6,  ALGORITHM_ROTATING_ZERO_RESTART},
 275         {"raid6_nr",      "raid6 (N restart)",                      2, 4, 6,  ALGORITHM_ROTATING_N_RESTART},
 276         {"raid6_nc",      "raid6 (N continue)",                     2, 4, 6,  ALGORITHM_ROTATING_N_CONTINUE},
 277         {"raid6_n_6",     "raid6 (dedicated parity/Q n/6)",         2, 4, 6,  ALGORITHM_PARITY_N_6},
 278         {"raid6_ls_6",    "raid6 (left symmetric dedicated Q 6)",   2, 4, 6,  ALGORITHM_LEFT_SYMMETRIC_6},
 279         {"raid6_rs_6",    "raid6 (right symmetric dedicated Q 6)",  2, 4, 6,  ALGORITHM_RIGHT_SYMMETRIC_6},
 280         {"raid6_la_6",    "raid6 (left asymmetric dedicated Q 6)",  2, 4, 6,  ALGORITHM_LEFT_ASYMMETRIC_6},
 281         {"raid6_ra_6",    "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6,  ALGORITHM_RIGHT_ASYMMETRIC_6}
 282 };
 283
 284 /* True, if @v is in inclusive range [@min, @max] */
 285 static bool __within_range(long v, long min, long max)
 286 {
 287         return v >= min && v <= max;
 288 }
 289
 290 /* All table line arguments are defined here */
 291 static struct arg_name_flag {
 292         const unsigned long flag;
 293         const char *name;
 294 } __arg_name_flags[] = {
 295         { CTR_FLAG_SYNC, "sync"},
 296         { CTR_FLAG_NOSYNC, "nosync"},
 297         { CTR_FLAG_REBUILD, "rebuild"},
 298         { CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
 299         { CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
 300         { CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
 301         { CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
 302         { CTR_FLAG_WRITE_MOSTLY, "writemostly"},
 303         { CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
 304         { CTR_FLAG_REGION_SIZE, "region_size"},
 305         { CTR_FLAG_RAID10_COPIES, "raid10_copies"},
 306         { CTR_FLAG_RAID10_FORMAT, "raid10_format"},
 307         { CTR_FLAG_DATA_OFFSET, "data_offset"},
 308         { CTR_FLAG_DELTA_DISKS, "delta_disks"},
 309         { CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
 310 };
 311
 312 /* Return argument name string for given @flag */
 313 static const char *dm_raid_arg_name_by_flag(const uint32_t flag)
 314 {
 315         if (hweight32(flag) == 1) {
 316                 struct arg_name_flag *anf = __arg_name_flags + ARRAY_SIZE(__arg_name_flags);
 317
 318                 while (anf-- > __arg_name_flags)
 319                         if (flag & anf->flag)
 320                                 return anf->name;
 321
 322         } else
 323                 DMERR("%s called with more than one flag!", __func__);
 324
 325         return NULL;
 326 }
 327
 328 /*
 329  * bool helpers to test for various raid levels of a raid set,
 330  * is. it's level as reported by the superblock rather than
 331  * the requested raid_type passed to the constructor.
 332  */
 333 /* Return true, if raid set in @rs is raid0 */
 334 static bool rs_is_raid0(struct raid_set *rs)
 335 {
 336         return !rs->md.level;
 337 }
 338
 339 /* Return true, if raid set in @rs is raid1 */
 340 static bool rs_is_raid1(struct raid_set *rs)
 341 {
 342         return rs->md.level == 1;
 343 }
 344
 345 /* Return true, if raid set in @rs is raid10 */
 346 static bool rs_is_raid10(struct raid_set *rs)
 347 {
 348         return rs->md.level == 10;
 349 }
 350
 351 /* Return true, if raid set in @rs is level 4, 5 or 6 */
 352 static bool rs_is_raid456(struct raid_set *rs)
 353 {
 354         return __within_range(rs->md.level, 4, 6);
 355 }
 356
 357 /* Return true, if raid set in @rs is reshapable */
 358 static unsigned int __is_raid10_far(int layout);
 359 static bool rs_is_reshapable(struct raid_set *rs)
 360 {
 361         return rs_is_raid456(rs) ||
 362                (rs_is_raid10(rs) && !__is_raid10_far(rs->md.new_layout));
 363 }
 364
 365 /* Return true, if raid set in @rs is recovering */
 366 static bool rs_is_recovering(struct raid_set *rs)
 367 {
 368         smp_rmb();
 369         return rs->md.recovery_cp != MaxSector;
 370 }
 371
 372 /* Return true, if raid set in @rs is reshaping */
 373 static bool rs_is_reshaping(struct raid_set *rs)
 374 {
 375         smp_rmb();
 376         return rs->md.reshape_position != MaxSector;
 377 }
 378
 379 /*
 380  * bool helpers to test for various raid levels of a raid type
 381  */
 382
 383 /* Return true, if raid type in @rt is raid0 */
 384 static bool rt_is_raid0(struct raid_type *rt)
 385 {
 386         return !rt->level;
 387 }
 388
 389 /* Return true, if raid type in @rt is raid1 */
 390 static bool rt_is_raid1(struct raid_type *rt)
 391 {
 392         return rt->level == 1;
 393 }
 394
 395 /* Return true, if raid type in @rt is raid10 */
 396 static bool rt_is_raid10(struct raid_type *rt)
 397 {
 398         return rt->level == 10;
 399 }
 400
 401 /* Return true, if raid type in @rt is raid4/5 */
 402 static bool rt_is_raid45(struct raid_type *rt)
 403 {
 404         return __within_range(rt->level, 4, 5);
 405 }
 406
 407 /* Return true, if raid type in @rt is raid6 */
 408 static bool rt_is_raid6(struct raid_type *rt)
 409 {
 410         return rt->level == 6;
 411 }
 412
 413 /* Return true, if raid type in @rt is raid4/5/6 */
 414 static bool rt_is_raid456(struct raid_type *rt)
 415 {
 416         return __within_range(rt->level, 4, 6);
 417 }
 418 /* END: raid level bools */
 419
 420 /* Return valid ctr flags for the raid level of @rs */
 421 static unsigned long __valid_flags(struct raid_set *rs)
 422 {
 423         if (rt_is_raid0(rs->raid_type))
 424                 return RAID0_VALID_FLAGS;
 425         else if (rt_is_raid1(rs->raid_type))
 426                 return RAID1_VALID_FLAGS;
 427         else if (rt_is_raid10(rs->raid_type))
 428                 return RAID10_VALID_FLAGS;
 429         else if (rt_is_raid45(rs->raid_type))
 430                 return RAID45_VALID_FLAGS;
 431         else if (rt_is_raid6(rs->raid_type))
 432                 return RAID6_VALID_FLAGS;
 433
 434         return ~0;
 435 }
 436
 437 /*
 438  * Check for valid flags set on @rs
 439  *
 440  * Has to be called after parsing of the ctr flags!
 441  */
 442 static int rs_check_for_valid_flags(struct raid_set *rs)
 443 {
 444         if (rs->ctr_flags & ~__valid_flags(rs)) {
 445                 rs->ti->error = "Invalid flags combination";
 446                 return -EINVAL;
 447         }
 448
 449         return 0;
 450 }
 451
 452 /* MD raid10 bit definitions and helpers */
 453 #define RAID10_OFFSET                   (1 << 16) /* stripes with data copies area adjacent on devices */
 454 #define RAID10_BROCKEN_USE_FAR_SETS     (1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
 455 #define RAID10_USE_FAR_SETS             (1 << 18) /* Use sets instead of whole stripe rotation */
 456 #define RAID10_FAR_COPIES_SHIFT         8         /* raid10 # far copies shift (2nd byte of layout) */
 457
 458 /* Return md raid10 near copies for @layout */
 459 static unsigned int __raid10_near_copies(int layout)
 460 {
 461         return layout & 0xFF;
 462 }
 463
 464 /* Return md raid10 far copies for @layout */
 465 static unsigned int __raid10_far_copies(int layout)
 466 {
 467         return __raid10_near_copies(layout >> RAID10_FAR_COPIES_SHIFT);
 468 }
 469
 470 /* Return true if md raid10 offset for @layout */
 471 static unsigned int __is_raid10_offset(int layout)
 472 {
 473         return layout & RAID10_OFFSET;
 474 }
 475
 476 /* Return true if md raid10 near for @layout */
 477 static unsigned int __is_raid10_near(int layout)
 478 {
 479         return !__is_raid10_offset(layout) && __raid10_near_copies(layout) > 1;
 480 }
 481
 482 /* Return true if md raid10 far for @layout */
 483 static unsigned int __is_raid10_far(int layout)
 484 {
 485         return !__is_raid10_offset(layout) && __raid10_far_copies(layout) > 1;
 486 }
 487
 488 /* Return md raid10 layout string for @layout */
 489 static const char *raid10_md_layout_to_format(int layout)
 490 {
 491         /*
 492          * Bit 16 stands for "offset"
 493          * (i.e. adjacent stripes hold copies)
 494          *
 495          * Refer to MD's raid10.c for details
 496          */
 497         if (__is_raid10_offset(layout))
 498                 return "offset";
 499
 500         if (__raid10_near_copies(layout) > 1)
 501                 return "near";
 502
 503         WARN_ON(__raid10_far_copies(layout) < 2);
 504
 505         return "far";
 506 }
 507
 508 /* Return md raid10 algorithm for @name */
 509 static const int raid10_name_to_format(const char *name)
 510 {
 511         if (!strcasecmp(name, "near"))
 512                 return ALGORITHM_RAID10_NEAR;
 513         else if (!strcasecmp(name, "offset"))
 514                 return ALGORITHM_RAID10_OFFSET;
 515         else if (!strcasecmp(name, "far"))
 516                 return ALGORITHM_RAID10_FAR;
 517
 518         return -EINVAL;
 519 }
 520
 521 /* Return md raid10 copies for @layout */
 522 static unsigned int raid10_md_layout_to_copies(int layout)
 523 {
 524         return __raid10_near_copies(layout) > 1 ?
 525                 __raid10_near_copies(layout) : __raid10_far_copies(layout);
 526 }
 527
 528 /* Return md raid10 format id for @format string */
 529 static int raid10_format_to_md_layout(struct raid_set *rs,
 530                                       unsigned int algorithm,
 531                                       unsigned int copies)
 532 {
 533         unsigned int n = 1, f = 1, r = 0;
 534
 535         /*
 536          * MD resilienece flaw:
 537          *
 538          * enabling use_far_sets for far/offset formats causes copies
 539          * to be colocated on the same devs together with their origins!
 540          *
 541          * -> disable it for now in the definition above
 542          */
 543         if (algorithm == ALGORITHM_RAID10_DEFAULT ||
 544             algorithm == ALGORITHM_RAID10_NEAR)
 545                 n = copies;
 546
 547         else if (algorithm == ALGORITHM_RAID10_OFFSET) {
 548                 f = copies;
 549                 r = RAID10_OFFSET;
 550                 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
 551                         r |= RAID10_USE_FAR_SETS;
 552
 553         } else if (algorithm == ALGORITHM_RAID10_FAR) {
 554                 f = copies;
 555                 r = !RAID10_OFFSET;
 556                 if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
 557                         r |= RAID10_USE_FAR_SETS;
 558
 559         } else
 560                 return -EINVAL;
 561
 562         return r | (f << RAID10_FAR_COPIES_SHIFT) | n;
 563 }
 564 /* END: MD raid10 bit definitions and helpers */
 565
 566 /* Check for any of the raid10 algorithms */
 567 static int __got_raid10(struct raid_type *rtp, const int layout)
 568 {
 569         if (rtp->level == 10) {
 570                 switch (rtp->algorithm) {
 571                 case ALGORITHM_RAID10_DEFAULT:
 572                 case ALGORITHM_RAID10_NEAR:
 573                         return __is_raid10_near(layout);
 574                 case ALGORITHM_RAID10_OFFSET:
 575                         return __is_raid10_offset(layout);
 576                 case ALGORITHM_RAID10_FAR:
 577                         return __is_raid10_far(layout);
 578                 default:
 579                         break;
 580                 }
 581         }
 582
 583         return 0;
 584 }
 585
 586 /* Return raid_type for @name */
 587 static struct raid_type *get_raid_type(const char *name)
 588 {
 589         struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
 590
 591         while (rtp-- > raid_types)
 592                 if (!strcasecmp(rtp->name, name))
 593                         return rtp;
 594
 595         return NULL;
 596 }
 597
 598 /* Return raid_type for @name based derived from @level and @layout */
 599 static struct raid_type *get_raid_type_by_ll(const int level, const int layout)
 600 {
 601         struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
 602
 603         while (rtp-- > raid_types) {
 604                 /* RAID10 special checks based on @layout flags/properties */
 605                 if (rtp->level == level &&
 606                     (__got_raid10(rtp, layout) || rtp->algorithm == layout))
 607                         return rtp;
 608         }
 609
 610         return NULL;
 611 }
 612
 613 /*
 614  * Conditionally change bdev capacity of @rs
 615  * in case of a disk add/remove reshape
 616  */
 617 static void rs_set_capacity(struct raid_set *rs)
 618 {
 619         struct mddev *mddev = &rs->md;
 620
 621         /* Make sure we access most actual mddev properties */
 622         smp_rmb();
 623         if (rs->ti->len != mddev->array_sectors && !rs_is_reshaping(rs)) {
 624                 struct gendisk *gendisk = dm_disk(dm_table_get_md(rs->ti->table));
 625
 626                 set_capacity(gendisk, mddev->array_sectors);
 627                 revalidate_disk(gendisk);
 628         }
 629 }
 630
 631 /*
 632  * Set the mddev properties in @rs to the current
 633  * ones retrieved from the freshest superblock
 634  */
 635 static void rs_set_cur(struct raid_set *rs)
 636 {
 637         struct mddev *mddev = &rs->md;
 638
 639         mddev->new_level = mddev->level;
 640         mddev->new_layout = mddev->layout;
 641         mddev->new_chunk_sectors = mddev->chunk_sectors;
 642 }
 643
 644 /*
 645  * Set the mddev properties in @rs to the new
 646  * ones requested by the ctr
 647  */
 648 static void rs_set_new(struct raid_set *rs)
 649 {
 650         struct mddev *mddev = &rs->md;
 651
 652         mddev->level = mddev->new_level;
 653         mddev->layout = mddev->new_layout;
 654         mddev->chunk_sectors = mddev->new_chunk_sectors;
 655         mddev->raid_disks = rs->raid_disks;
 656         mddev->delta_disks = 0;
 657 }
 658
 659 static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
 660                                        unsigned raid_devs)
 661 {
 662         unsigned i;
 663         struct raid_set *rs;
 664
 665         if (raid_devs <= raid_type->parity_devs) {
 666                 ti->error = "Insufficient number of devices";
 667                 return ERR_PTR(-EINVAL);
 668         }
 669
 670         rs = kzalloc(sizeof(*rs) + raid_devs * sizeof(rs->dev[0]), GFP_KERNEL);
 671         if (!rs) {
 672                 ti->error = "Cannot allocate raid context";
 673                 return ERR_PTR(-ENOMEM);
 674         }
 675
 676         mddev_init(&rs->md);
 677
 678         rs->raid_disks = raid_devs;
 679         rs->delta_disks = 0;
 680
 681         rs->ti = ti;
 682         rs->raid_type = raid_type;
 683         rs->stripe_cache_entries = 256;
 684         rs->md.raid_disks = raid_devs;
 685         rs->md.level = raid_type->level;
 686         rs->md.new_level = rs->md.level;
 687         rs->md.layout = raid_type->algorithm;
 688         rs->md.new_layout = rs->md.layout;
 689         rs->md.delta_disks = 0;
 690         rs->md.recovery_cp = rs_is_raid0(rs) ? MaxSector : 0;
 691
 692         for (i = 0; i < raid_devs; i++)
 693                 md_rdev_init(&rs->dev[i].rdev);
 694
 695         /*
 696          * Remaining items to be initialized by further RAID params:
 697          *  rs->md.persistent
 698          *  rs->md.external
 699          *  rs->md.chunk_sectors
 700          *  rs->md.new_chunk_sectors
 701          *  rs->md.dev_sectors
 702          */
 703
 704         return rs;
 705 }
 706
 707 static void raid_set_free(struct raid_set *rs)
 708 {
 709         int i;
 710
 711         for (i = 0; i < rs->md.raid_disks; i++) {
 712                 if (rs->dev[i].meta_dev)
 713                         dm_put_device(rs->ti, rs->dev[i].meta_dev);
 714                 md_rdev_clear(&rs->dev[i].rdev);
 715                 if (rs->dev[i].data_dev)
 716                         dm_put_device(rs->ti, rs->dev[i].data_dev);
 717         }
 718
 719         kfree(rs);
 720 }
 721
 722 /*
 723  * For every device we have two words
 724  *  <meta_dev>: meta device name or '-' if missing
 725  *  <data_dev>: data device name or '-' if missing
 726  *
 727  * The following are permitted:
 728  *    - -
 729  *    - <data_dev>
 730  *    <meta_dev> <data_dev>
 731  *
 732  * The following is not allowed:
 733  *    <meta_dev> -
 734  *
 735  * This code parses those words.  If there is a failure,
 736  * the caller must use raid_set_free() to unwind the operations.
 737  */
 738 static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
 739 {
 740         int i;
 741         int rebuild = 0;
 742         int metadata_available = 0;
 743         int r = 0;
 744         const char *arg;
 745
 746         /* Put off the number of raid devices argument to get to dev pairs */
 747         arg = dm_shift_arg(as);
 748         if (!arg)
 749                 return -EINVAL;
 750
 751         for (i = 0; i < rs->md.raid_disks; i++) {
 752                 rs->dev[i].rdev.raid_disk = i;
 753
 754                 rs->dev[i].meta_dev = NULL;
 755                 rs->dev[i].data_dev = NULL;
 756
 757                 /*
 758                  * There are no offsets, since there is a separate device
 759                  * for data and metadata.
 760                  */
 761                 rs->dev[i].rdev.data_offset = 0;
 762                 rs->dev[i].rdev.mddev = &rs->md;
 763
 764                 arg = dm_shift_arg(as);
 765                 if (!arg)
 766                         return -EINVAL;
 767
 768                 if (strcmp(arg, "-")) {
 769                         r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
 770                                           &rs->dev[i].meta_dev);
 771                         if (r) {
 772                                 rs->ti->error = "RAID metadata device lookup failure";
 773                                 return r;
 774                         }
 775
 776                         rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
 777                         if (!rs->dev[i].rdev.sb_page) {
 778                                 rs->ti->error = "Failed to allocate superblock page";
 779                                 return -ENOMEM;
 780                         }
 781                 }
 782
 783                 arg = dm_shift_arg(as);
 784                 if (!arg)
 785                         return -EINVAL;
 786
 787                 if (!strcmp(arg, "-")) {
 788                         if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
 789                             (!rs->dev[i].rdev.recovery_offset)) {
 790                                 rs->ti->error = "Drive designated for rebuild not specified";
 791                                 return -EINVAL;
 792                         }
 793
 794                         if (rs->dev[i].meta_dev) {
 795                                 rs->ti->error = "No data device supplied with metadata device";
 796                                 return -EINVAL;
 797                         }
 798
 799                         continue;
 800                 }
 801
 802                 r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
 803                                   &rs->dev[i].data_dev);
 804                 if (r) {
 805                         rs->ti->error = "RAID device lookup failure";
 806                         return r;
 807                 }
 808
 809                 if (rs->dev[i].meta_dev) {
 810                         metadata_available = 1;
 811                         rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
 812                 }
 813                 rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
 814                 list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
 815                 if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
 816                         rebuild++;
 817         }
 818
 819         if (metadata_available) {
 820                 rs->md.external = 0;
 821                 rs->md.persistent = 1;
 822                 rs->md.major_version = 2;
 823         } else if (rebuild && !rs->md.recovery_cp) {
 824                 /*
 825                  * Without metadata, we will not be able to tell if the array
 826                  * is in-sync or not - we must assume it is not.  Therefore,
 827                  * it is impossible to rebuild a drive.
 828                  *
 829                  * Even if there is metadata, the on-disk information may
 830                  * indicate that the array is not in-sync and it will then
 831                  * fail at that time.
 832                  *
 833                  * User could specify 'nosync' option if desperate.
 834                  */
 835                 rs->ti->error = "Unable to rebuild drive while array is not in-sync";
 836                 return -EINVAL;
 837         }
 838
 839         return 0;
 840 }
 841
 842 /*
 843  * validate_region_size
 844  * @rs
 845  * @region_size:  region size in sectors.  If 0, pick a size (4MiB default).
 846  *
 847  * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
 848  * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
 849  *
 850  * Returns: 0 on success, -EINVAL on failure.
 851  */
 852 static int validate_region_size(struct raid_set *rs, unsigned long region_size)
 853 {
 854         unsigned long min_region_size = rs->ti->len / (1 << 21);
 855
 856         if (!region_size) {
 857                 /*
 858                  * Choose a reasonable default.  All figures in sectors.
 859                  */
 860                 if (min_region_size > (1 << 13)) {
 861                         /* If not a power of 2, make it the next power of 2 */
 862                         region_size = roundup_pow_of_two(min_region_size);
 863                         DMINFO("Choosing default region size of %lu sectors",
 864                                region_size);
 865                 } else {
 866                         DMINFO("Choosing default region size of 4MiB");
 867                         region_size = 1 << 13; /* sectors */
 868                 }
 869         } else {
 870                 /*
 871                  * Validate user-supplied value.
 872                  */
 873                 if (region_size > rs->ti->len) {
 874                         rs->ti->error = "Supplied region size is too large";
 875                         return -EINVAL;
 876                 }
 877
 878                 if (region_size < min_region_size) {
 879                         DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
 880                               region_size, min_region_size);
 881                         rs->ti->error = "Supplied region size is too small";
 882                         return -EINVAL;
 883                 }
 884
 885                 if (!is_power_of_2(region_size)) {
 886                         rs->ti->error = "Region size is not a power of 2";
 887                         return -EINVAL;
 888                 }
 889
 890                 if (region_size < rs->md.chunk_sectors) {
 891                         rs->ti->error = "Region size is smaller than the chunk size";
 892                         return -EINVAL;
 893                 }
 894         }
 895
 896         /*
 897          * Convert sectors to bytes.
 898          */
 899         rs->md.bitmap_info.chunksize = (region_size << 9);
 900
 901         return 0;
 902 }
 903
 904 /*
 905  * validate_raid_redundancy
 906  * @rs
 907  *
 908  * Determine if there are enough devices in the array that haven't
 909  * failed (or are being rebuilt) to form a usable array.
 910  *
 911  * Returns: 0 on success, -EINVAL on failure.
 912  */
 913 static int validate_raid_redundancy(struct raid_set *rs)
 914 {
 915         unsigned i, rebuild_cnt = 0;
 916         unsigned rebuilds_per_group = 0, copies;
 917         unsigned group_size, last_group_start;
 918
 919         for (i = 0; i < rs->md.raid_disks; i++)
 920                 if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
 921                     !rs->dev[i].rdev.sb_page)
 922                         rebuild_cnt++;
 923
 924         switch (rs->raid_type->level) {
 925         case 1:
 926                 if (rebuild_cnt >= rs->md.raid_disks)
 927                         goto too_many;
 928                 break;
 929         case 4:
 930         case 5:
 931         case 6:
 932                 if (rebuild_cnt > rs->raid_type->parity_devs)
 933                         goto too_many;
 934                 break;
 935         case 10:
 936                 copies = raid10_md_layout_to_copies(rs->md.new_layout);
 937                 if (rebuild_cnt < copies)
 938                         break;
 939
 940                 /*
 941                  * It is possible to have a higher rebuild count for RAID10,
 942                  * as long as the failed devices occur in different mirror
 943                  * groups (i.e. different stripes).
 944                  *
 945                  * When checking "near" format, make sure no adjacent devices
 946                  * have failed beyond what can be handled.  In addition to the
 947                  * simple case where the number of devices is a multiple of the
 948                  * number of copies, we must also handle cases where the number
 949                  * of devices is not a multiple of the number of copies.
 950                  * E.g.    dev1 dev2 dev3 dev4 dev5
 951                  *          A    A    B    B    C
 952                  *          C    D    D    E    E
 953                  */
 954                 if (__is_raid10_near(rs->md.new_layout)) {
 955                         for (i = 0; i < rs->raid_disks; i++) {
 956                                 if (!(i % copies))
 957                                         rebuilds_per_group = 0;
 958                                 if ((!rs->dev[i].rdev.sb_page ||
 959                                     !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
 960                                     (++rebuilds_per_group >= copies))
 961                                         goto too_many;
 962                         }
 963                         break;
 964                 }
 965
 966                 /*
 967                  * When checking "far" and "offset" formats, we need to ensure
 968                  * that the device that holds its copy is not also dead or
 969                  * being rebuilt.  (Note that "far" and "offset" formats only
 970                  * support two copies right now.  These formats also only ever
 971                  * use the 'use_far_sets' variant.)
 972                  *
 973                  * This check is somewhat complicated by the need to account
 974                  * for arrays that are not a multiple of (far) copies.  This
 975                  * results in the need to treat the last (potentially larger)
 976                  * set differently.
 977                  */
 978                 group_size = (rs->md.raid_disks / copies);
 979                 last_group_start = (rs->md.raid_disks / group_size) - 1;
 980                 last_group_start *= group_size;
 981                 for (i = 0; i < rs->md.raid_disks; i++) {
 982                         if (!(i % copies) && !(i > last_group_start))
 983                                 rebuilds_per_group = 0;
 984                         if ((!rs->dev[i].rdev.sb_page ||
 985                              !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
 986                             (++rebuilds_per_group >= copies))
 987                                         goto too_many;
 988                 }
 989                 break;
 990         default:
 991                 if (rebuild_cnt)
 992                         return -EINVAL;
 993         }
 994
 995         return 0;
 996
 997 too_many:
 998         return -EINVAL;
 999 }
1000
1001 /*
1002  * Possible arguments are...
1003  *      <chunk_size> [optional_args]
1004  *
1005  * Argument definitions
1006  *    <chunk_size>                      The number of sectors per disk that
1007  *                                      will form the "stripe"
1008  *    [[no]sync]                        Force or prevent recovery of the
1009  *                                      entire array
1010  *    [rebuild <idx>]                   Rebuild the drive indicated by the index
1011  *    [daemon_sleep <ms>]               Time between bitmap daemon work to
1012  *                                      clear bits
1013  *    [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1014  *    [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1015  *    [write_mostly <idx>]              Indicate a write mostly drive via index
1016  *    [max_write_behind <sectors>]      See '-write-behind=' (man mdadm)
1017  *    [stripe_cache <sectors>]          Stripe cache size for higher RAIDs
1018  *    [region_size <sectors>]           Defines granularity of bitmap
1019  *
1020  * RAID10-only options:
1021  *    [raid10_copies <# copies>]        Number of copies.  (Default: 2)
1022  *    [raid10_format <near|far|offset>] Layout algorithm.  (Default: near)
1023  */
1024 static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1025                              unsigned num_raid_params)
1026 {
1027         int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1028         unsigned raid10_copies = 2;
1029         unsigned i;
1030         unsigned region_size = 0;
1031         sector_t max_io_len;
1032         const char *arg, *key;
1033         struct raid_dev *rd;
1034         struct raid_type *rt = rs->raid_type;
1035
1036         arg = dm_shift_arg(as);
1037         num_raid_params--; /* Account for chunk_size argument */
1038
1039         if (kstrtoint(arg, 10, &value) < 0) {
1040                 rs->ti->error = "Bad numerical argument given for chunk_size";
1041                 return -EINVAL;
1042         }
1043
1044         /*
1045          * First, parse the in-order required arguments
1046          * "chunk_size" is the only argument of this type.
1047          */
1048         if (rt_is_raid1(rt)) {
1049                 if (value)
1050                         DMERR("Ignoring chunk size parameter for RAID 1");
1051                 value = 0;
1052         } else if (!is_power_of_2(value)) {
1053                 rs->ti->error = "Chunk size must be a power of 2";
1054                 return -EINVAL;
1055         } else if (value < 8) {
1056                 rs->ti->error = "Chunk size value is too small";
1057                 return -EINVAL;
1058         }
1059
1060         rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1061
1062         /*
1063          * We set each individual device as In_sync with a completed
1064          * 'recovery_offset'.  If there has been a device failure or
1065          * replacement then one of the following cases applies:
1066          *
1067          *   1) User specifies 'rebuild'.
1068          *      - Device is reset when param is read.
1069          *   2) A new device is supplied.
1070          *      - No matching superblock found, resets device.
1071          *   3) Device failure was transient and returns on reload.
1072          *      - Failure noticed, resets device for bitmap replay.
1073          *   4) Device hadn't completed recovery after previous failure.
1074          *      - Superblock is read and overrides recovery_offset.
1075          *
1076          * What is found in the superblocks of the devices is always
1077          * authoritative, unless 'rebuild' or '[no]sync' was specified.
1078          */
1079         for (i = 0; i < rs->md.raid_disks; i++) {
1080                 set_bit(In_sync, &rs->dev[i].rdev.flags);
1081                 rs->dev[i].rdev.recovery_offset = MaxSector;
1082         }
1083
1084         /*
1085          * Second, parse the unordered optional arguments
1086          */
1087         for (i = 0; i < num_raid_params; i++) {
1088                 key = dm_shift_arg(as);
1089                 if (!key) {
1090                         rs->ti->error = "Not enough raid parameters given";
1091                         return -EINVAL;
1092                 }
1093
1094                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1095                         if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1096                                 rs->ti->error = "Only one 'nosync' argument allowed";
1097                                 return -EINVAL;
1098                         }
1099                         rs->md.recovery_cp = MaxSector;
1100                         continue;
1101                 }
1102                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1103                         if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1104                                 rs->ti->error = "Only one 'sync' argument allowed";
1105                                 return -EINVAL;
1106                         }
1107                         rs->md.recovery_cp = 0;
1108                         continue;
1109                 }
1110                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1111                         if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1112                                 rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1113                                 return -EINVAL;
1114                         }
1115                         continue;
1116                 }
1117
1118                 arg = dm_shift_arg(as);
1119                 i++; /* Account for the argument pairs */
1120                 if (!arg) {
1121                         rs->ti->error = "Wrong number of raid parameters given";
1122                         return -EINVAL;
1123                 }
1124
1125                 /*
1126                  * Parameters that take a string value are checked here.
1127                  */
1128
1129                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1130                         if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1131                                 rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1132                                 return -EINVAL;
1133                         }
1134                         if (!rt_is_raid10(rt)) {
1135                                 rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1136                                 return -EINVAL;
1137                         }
1138                         raid10_format = raid10_name_to_format(arg);
1139                         if (raid10_format < 0) {
1140                                 rs->ti->error = "Invalid 'raid10_format' value given";
1141                                 return raid10_format;
1142                         }
1143                         continue;
1144                 }
1145
1146                 if (kstrtoint(arg, 10, &value) < 0) {
1147                         rs->ti->error = "Bad numerical argument given in raid params";
1148                         return -EINVAL;
1149                 }
1150
1151                 if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1152                         /*
1153                          * "rebuild" is being passed in by userspace to provide
1154                          * indexes of replaced devices and to set up additional
1155                          * devices on raid level takeover.
1156                          */
1157                         if (!__within_range(value, 0, rs->raid_disks - 1)) {
1158                                 rs->ti->error = "Invalid rebuild index given";
1159                                 return -EINVAL;
1160                         }
1161
1162                         if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1163                                 rs->ti->error = "rebuild for this index already given";
1164                                 return -EINVAL;
1165                         }
1166
1167                         rd = rs->dev + value;
1168                         clear_bit(In_sync, &rd->rdev.flags);
1169                         clear_bit(Faulty, &rd->rdev.flags);
1170                         rd->rdev.recovery_offset = 0;
1171                         set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1172                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1173                         if (!rt_is_raid1(rt)) {
1174                                 rs->ti->error = "write_mostly option is only valid for RAID1";
1175                                 return -EINVAL;
1176                         }
1177
1178                         if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1179                                 rs->ti->error = "Invalid write_mostly index given";
1180                                 return -EINVAL;
1181                         }
1182
1183                         set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1184                         set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1185                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1186                         if (!rt_is_raid1(rt)) {
1187                                 rs->ti->error = "max_write_behind option is only valid for RAID1";
1188                                 return -EINVAL;
1189                         }
1190
1191                         if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1192                                 rs->ti->error = "Only one max_write_behind argument pair allowed";
1193                                 return -EINVAL;
1194                         }
1195
1196                         /*
1197                          * In device-mapper, we specify things in sectors, but
1198                          * MD records this value in kB
1199                          */
1200                         value /= 2;
1201                         if (value > COUNTER_MAX) {
1202                                 rs->ti->error = "Max write-behind limit out of range";
1203                                 return -EINVAL;
1204                         }
1205
1206                         rs->md.bitmap_info.max_write_behind = value;
1207                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1208                         if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1209                                 rs->ti->error = "Only one daemon_sleep argument pair allowed";
1210                                 return -EINVAL;
1211                         }
1212                         if (!value || (value > MAX_SCHEDULE_TIMEOUT)) {
1213                                 rs->ti->error = "daemon sleep period out of range";
1214                                 return -EINVAL;
1215                         }
1216                         rs->md.bitmap_info.daemon_sleep = value;
1217                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1218                         /* Userspace passes new data_offset after having extended the the data image LV */
1219                         if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1220                                 rs->ti->error = "Only one data_offset argument pair allowed";
1221                                 return -EINVAL;
1222                         }
1223                         /* Ensure sensible data offset */
1224                         if (value < 0) {
1225                                 rs->ti->error = "Bogus data_offset value";
1226                                 return -EINVAL;
1227                         }
1228                         rs->data_offset = value;
1229                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1230                         /* Define the +/-# of disks to add to/remove from the given raid set */
1231                         if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1232                                 rs->ti->error = "Only one delta_disks argument pair allowed";
1233                                 return -EINVAL;
1234                         }
1235                         /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1236                         if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1237                                 rs->ti->error = "Too many delta_disk requested";
1238                                 return -EINVAL;
1239                         }
1240
1241                         rs->delta_disks = value;
1242                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1243                         if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1244                                 rs->ti->error = "Only one stripe_cache argument pair allowed";
1245                                 return -EINVAL;
1246                         }
1247
1248                         if (!rt_is_raid456(rt)) {
1249                                 rs->ti->error = "Inappropriate argument: stripe_cache";
1250                                 return -EINVAL;
1251                         }
1252
1253                         rs->stripe_cache_entries = value;
1254                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1255                         if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1256                                 rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1257                                 return -EINVAL;
1258                         }
1259                         if (value > INT_MAX) {
1260                                 rs->ti->error = "min_recovery_rate out of range";
1261                                 return -EINVAL;
1262                         }
1263                         rs->md.sync_speed_min = (int)value;
1264                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1265                         if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1266                                 rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1267                                 return -EINVAL;
1268                         }
1269                         if (value > INT_MAX) {
1270                                 rs->ti->error = "max_recovery_rate out of range";
1271                                 return -EINVAL;
1272                         }
1273                         rs->md.sync_speed_max = (int)value;
1274                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1275                         if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1276                                 rs->ti->error = "Only one region_size argument pair allowed";
1277                                 return -EINVAL;
1278                         }
1279
1280                         region_size = value;
1281                         rs->requested_bitmap_chunk_sectors = value;
1282                 } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1283                         if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1284                                 rs->ti->error = "Only one raid10_copies argument pair allowed";
1285                                 return -EINVAL;
1286                         }
1287
1288                         if (!__within_range(value, 2, rs->md.raid_disks)) {
1289                                 rs->ti->error = "Bad value for 'raid10_copies'";
1290                                 return -EINVAL;
1291                         }
1292
1293                         raid10_copies = value;
1294                 } else {
1295                         DMERR("Unable to parse RAID parameter: %s", key);
1296                         rs->ti->error = "Unable to parse RAID parameter";
1297                         return -EINVAL;
1298                 }
1299         }
1300
1301         if (validate_region_size(rs, region_size))
1302                 return -EINVAL;
1303
1304         if (rs->md.chunk_sectors)
1305                 max_io_len = rs->md.chunk_sectors;
1306         else
1307                 max_io_len = region_size;
1308
1309         if (dm_set_target_max_io_len(rs->ti, max_io_len))
1310                 return -EINVAL;
1311
1312         if (rt_is_raid10(rt)) {
1313                 if (raid10_copies > rs->md.raid_disks) {
1314                         rs->ti->error = "Not enough devices to satisfy specification";
1315                         return -EINVAL;
1316                 }
1317
1318                 rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1319                 if (rs->md.new_layout < 0) {
1320                         rs->ti->error = "Error getting raid10 format";
1321                         return rs->md.new_layout;
1322                 }
1323
1324                 rt = get_raid_type_by_ll(10, rs->md.new_layout);
1325                 if (!rt) {
1326                         rs->ti->error = "Failed to recognize new raid10 layout";
1327                         return -EINVAL;
1328                 }
1329
1330                 if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1331                      rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1332                     test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1333                         rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1334                         return -EINVAL;
1335                 }
1336         }
1337
1338         rs->raid10_copies = raid10_copies;
1339
1340         /* Assume there are no metadata devices until the drives are parsed */
1341         rs->md.persistent = 0;
1342         rs->md.external = 1;
1343
1344         /* Check, if any invalid ctr arguments have been passed in for the raid level */
1345         return rs_check_for_valid_flags(rs);
1346 }
1347
1348 /* Set raid4/5/6 cache size */
1349 static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1350 {
1351         int r;
1352         struct r5conf *conf;
1353         struct mddev *mddev = &rs->md;
1354         uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1355         uint32_t nr_stripes = rs->stripe_cache_entries;
1356
1357         if (!rt_is_raid456(rs->raid_type)) {
1358                 rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1359                 return -EINVAL;
1360         }
1361
1362         if (nr_stripes < min_stripes) {
1363                 DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1364                        nr_stripes, min_stripes);
1365                 nr_stripes = min_stripes;
1366         }
1367
1368         conf = mddev->private;
1369         if (!conf) {
1370                 rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1371                 return -EINVAL;
1372         }
1373
1374         /* Try setting number of stripes in raid456 stripe cache */
1375         if (conf->min_nr_stripes != nr_stripes) {
1376                 r = raid5_set_cache_size(mddev, nr_stripes);
1377                 if (r) {
1378                         rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1379                         return r;
1380                 }
1381
1382                 DMINFO("%u stripe cache entries", nr_stripes);
1383         }
1384
1385         return 0;
1386 }
1387
1388 /* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1389 static unsigned int mddev_data_stripes(struct raid_set *rs)
1390 {
1391         return rs->md.raid_disks - rs->raid_type->parity_devs;
1392 }
1393
1394 /* Return # of data stripes of @rs (i.e. as of ctr) */
1395 static unsigned int rs_data_stripes(struct raid_set *rs)
1396 {
1397         return rs->raid_disks - rs->raid_type->parity_devs;
1398 }
1399
1400 /* Calculate the sectors per device and per array used for @rs */
1401 static int rs_set_dev_and_array_sectors(struct raid_set *rs, bool use_mddev)
1402 {
1403         int delta_disks;
1404         unsigned int data_stripes;
1405         struct mddev *mddev = &rs->md;
1406         struct md_rdev *rdev;
1407         sector_t array_sectors = rs->ti->len, dev_sectors = rs->ti->len;
1408         sector_t cur_dev_sectors = rs->dev[0].rdev.sectors;
1409
1410         if (use_mddev) {
1411                 delta_disks = mddev->delta_disks;
1412                 data_stripes = mddev_data_stripes(rs);
1413         } else {
1414                 delta_disks = rs->delta_disks;
1415                 data_stripes = rs_data_stripes(rs);
1416         }
1417
1418         /* Special raid1 case w/o delta_disks support (yet) */
1419         if (rt_is_raid1(rs->raid_type))
1420                 ;
1421         else if (rt_is_raid10(rs->raid_type)) {
1422                 if (rs->raid10_copies < 2 ||
1423                     delta_disks < 0) {
1424                         rs->ti->error = "Bogus raid10 data copies or delta disks";
1425                         return EINVAL;
1426                 }
1427
1428                 dev_sectors *= rs->raid10_copies;
1429                 if (sector_div(dev_sectors, data_stripes))
1430                         goto bad;
1431
1432                 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1433                 if (sector_div(array_sectors, rs->raid10_copies))
1434                         goto bad;
1435
1436         } else if (sector_div(dev_sectors, data_stripes))
1437                 goto bad;
1438
1439         else
1440                 /* Striped layouts */
1441                 array_sectors = (data_stripes + delta_disks) * dev_sectors;
1442
1443         rdev_for_each(rdev, mddev)
1444                 rdev->sectors = dev_sectors;
1445
1446         mddev->array_sectors = array_sectors;
1447         mddev->dev_sectors = dev_sectors;
1448
1449         if (!rs_is_raid0(rs) && dev_sectors > cur_dev_sectors)
1450                 mddev->recovery_cp = dev_sectors;
1451
1452         return 0;
1453 bad:
1454         rs->ti->error = "Target length not divisible by number of data devices";
1455         return EINVAL;
1456 }
1457
1458 static void do_table_event(struct work_struct *ws)
1459 {
1460         struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1461
1462         rs_set_capacity(rs);
1463         dm_table_event(rs->ti->table);
1464 }
1465
1466 static int raid_is_congested(struct dm_target_callbacks *cb, int bits)
1467 {
1468         struct raid_set *rs = container_of(cb, struct raid_set, callbacks);
1469
1470         return mddev_congested(&rs->md, bits);
1471 }
1472
1473 /*
1474  * Make sure a valid takover (level switch) is being requested on @rs
1475  *
1476  * Conversions of raid sets from one MD personality to another
1477  * have to conform to restrictions which are enforced here.
1478  *
1479  * Degration is already checked for in rs_check_conversion() below.
1480  */
1481 static int rs_check_takeover(struct raid_set *rs)
1482 {
1483         struct mddev *mddev = &rs->md;
1484         unsigned int near_copies;
1485
1486         smp_rmb();
1487         if (rs->md.degraded) {
1488                 rs->ti->error = "Can't takeover degraded raid set";
1489                 return -EPERM;
1490         }
1491
1492         if (rs_is_reshaping(rs)) {
1493                 rs->ti->error = "Can't takeover reshaping raid set";
1494                 return -EPERM;
1495         }
1496
1497         switch (mddev->level) {
1498         case 0:
1499                 /* raid0 -> raid1/5 with one disk */
1500                 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1501                     mddev->raid_disks == 1)
1502                         return 0;
1503
1504                 /* raid0 -> raid10 */
1505                 if (mddev->new_level == 10 &&
1506                     !(rs->raid_disks % mddev->raid_disks))
1507                         return 0;
1508
1509                 /* raid0 with multiple disks -> raid4/5/6 */
1510                 if (__within_range(mddev->new_level, 4, 6) &&
1511                     mddev->new_layout == ALGORITHM_PARITY_N &&
1512                     mddev->raid_disks > 1)
1513                         return 0;
1514
1515                 break;
1516
1517         case 10:
1518                 /* Can't takeover raid10_offset! */
1519                 if (__is_raid10_offset(mddev->layout))
1520                         break;
1521
1522                 near_copies = __raid10_near_copies(mddev->layout);
1523
1524                 /* raid10* -> raid0 */
1525                 if (mddev->new_level == 0) {
1526                         /* Can takeover raid10_near with raid disks divisable by data copies! */
1527                         if (near_copies > 1 &&
1528                             !(mddev->raid_disks % near_copies)) {
1529                                 mddev->raid_disks /= near_copies;
1530                                 mddev->delta_disks = mddev->raid_disks;
1531                                 return 0;
1532                         }
1533
1534                         /* Can takeover raid10_far */
1535                         if (near_copies == 1 &&
1536                             __raid10_far_copies(mddev->layout) > 1)
1537                                 return 0;
1538
1539                         break;
1540                 }
1541
1542                 /* raid10_{near,far} -> raid1 */
1543                 if (mddev->new_level == 1 &&
1544                     max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1545                         return 0;
1546
1547                 /* raid10_{near,far} with 2 disks -> raid4/5 */
1548                 if (__within_range(mddev->new_level, 4, 5) &&
1549                     mddev->raid_disks == 2)
1550                         return 0;
1551                 break;
1552
1553         case 1:
1554                 /* raid1 with 2 disks -> raid4/5 */
1555                 if (__within_range(mddev->new_level, 4, 5) &&
1556                     mddev->raid_disks == 2) {
1557                         mddev->degraded = 1;
1558                         return 0;
1559                 }
1560
1561                 /* raid1 -> raid0 */
1562                 if (mddev->new_level == 0 &&
1563                     mddev->raid_disks == 1)
1564                         return 0;
1565
1566                 /* raid1 -> raid10 */
1567                 if (mddev->new_level == 10)
1568                         return 0;
1569
1570                 break;
1571
1572         case 4:
1573                 /* raid4 -> raid0 */
1574                 if (mddev->new_level == 0)
1575                         return 0;
1576
1577                 /* raid4 -> raid1/5 with 2 disks */
1578                 if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1579                     mddev->raid_disks == 2)
1580                         return 0;
1581
1582                 /* raid4 -> raid5/6 with parity N */
1583                 if (__within_range(mddev->new_level, 5, 6) &&
1584                     mddev->layout == ALGORITHM_PARITY_N)
1585                         return 0;
1586                 break;
1587
1588         case 5:
1589                 /* raid5 with parity N -> raid0 */
1590                 if (mddev->new_level == 0 &&
1591                     mddev->layout == ALGORITHM_PARITY_N)
1592                         return 0;
1593
1594                 /* raid5 with parity N -> raid4 */
1595                 if (mddev->new_level == 4 &&
1596                     mddev->layout == ALGORITHM_PARITY_N)
1597                         return 0;
1598
1599                 /* raid5 with 2 disks -> raid1/4/10 */
1600                 if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1601                     mddev->raid_disks == 2)
1602                         return 0;
1603
1604                 /* raid5 with parity N -> raid6 with parity N */
1605                 if (mddev->new_level == 6 &&
1606                     ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1607                       __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1608                         return 0;
1609                 break;
1610
1611         case 6:
1612                 /* raid6 with parity N -> raid0 */
1613                 if (mddev->new_level == 0 &&
1614                     mddev->layout == ALGORITHM_PARITY_N)
1615                         return 0;
1616
1617                 /* raid6 with parity N -> raid4 */
1618                 if (mddev->new_level == 4 &&
1619                     mddev->layout == ALGORITHM_PARITY_N)
1620                         return 0;
1621
1622                 /* raid6_*_n with parity N -> raid5_* */
1623                 if (mddev->new_level == 5 &&
1624                     ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1625                      __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1626                         return 0;
1627
1628         default:
1629                 break;
1630         }
1631
1632         rs->ti->error = "takeover not possible";
1633         return -EINVAL;
1634 }
1635
1636 /* True if @rs requested to be taken over */
1637 static bool rs_takeover_requested(struct raid_set *rs)
1638 {
1639         return rs->md.new_level != rs->md.level;
1640 }
1641
1642 /* True if @rs is requested to reshape by ctr */
1643 static bool rs_reshape_requested(struct raid_set *rs)
1644 {
1645         struct mddev *mddev = &rs->md;
1646
1647         if (!mddev->level)
1648                 return false;
1649
1650         return !__is_raid10_far(mddev->new_layout) &&
1651                mddev->new_level == mddev->level &&
1652                (mddev->new_layout != mddev->layout ||
1653                 mddev->new_chunk_sectors != mddev->chunk_sectors ||
1654                 rs->raid_disks + rs->delta_disks != mddev->raid_disks);
1655 }
1656
1657 /*  Features */
1658 #define FEATURE_FLAG_SUPPORTS_V190      0x1 /* Supports extended superblock */
1659
1660 /* State flags for sb->flags */
1661 #define SB_FLAG_RESHAPE_ACTIVE          0x1
1662 #define SB_FLAG_RESHAPE_BACKWARDS       0x2
1663
1664 /*
1665  * This structure is never routinely used by userspace, unlike md superblocks.
1666  * Devices with this superblock should only ever be accessed via device-mapper.
1667  */
1668 #define DM_RAID_MAGIC 0x64526D44
1669 struct dm_raid_superblock {
1670         __le32 magic;           /* "DmRd" */
1671         __le32 compat_features; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1672
1673         __le32 num_devices;     /* Number of devices in this raid set. (Max 64) */
1674         __le32 array_position;  /* The position of this drive in the raid set */
1675
1676         __le64 events;          /* Incremented by md when superblock updated */
1677         __le64 failed_devices;  /* Pre 1.9.0 part of bit field of devices to */
1678                                 /* indicate failures (see extension below) */
1679
1680         /*
1681          * This offset tracks the progress of the repair or replacement of
1682          * an individual drive.
1683          */
1684         __le64 disk_recovery_offset;
1685
1686         /*
1687          * This offset tracks the progress of the initial raid set
1688          * synchronisation/parity calculation.
1689          */
1690         __le64 array_resync_offset;
1691
1692         /*
1693          * raid characteristics
1694          */
1695         __le32 level;
1696         __le32 layout;
1697         __le32 stripe_sectors;
1698
1699         /********************************************************************
1700          * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1701          *
1702          * FEATURE_FLAG_SUPPORTS_V190 in the features member indicates that those exist
1703          */
1704
1705         __le32 flags; /* Flags defining array states for reshaping */
1706
1707         /*
1708          * This offset tracks the progress of a raid
1709          * set reshape in order to be able to restart it
1710          */
1711         __le64 reshape_position;
1712
1713         /*
1714          * These define the properties of the array in case of an interrupted reshape
1715          */
1716         __le32 new_level;
1717         __le32 new_layout;
1718         __le32 new_stripe_sectors;
1719         __le32 delta_disks;
1720
1721         __le64 array_sectors; /* Array size in sectors */
1722
1723         /*
1724          * Sector offsets to data on devices (reshaping).
1725          * Needed to support out of place reshaping, thus
1726          * not writing over any stripes whilst converting
1727          * them from old to new layout
1728          */
1729         __le64 data_offset;
1730         __le64 new_data_offset;
1731
1732         __le64 sectors; /* Used device size in sectors */
1733
1734         /*
1735          * Additonal Bit field of devices indicating failures to support
1736          * up to 256 devices with the 1.9.0 on-disk metadata format
1737          */
1738         __le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
1739
1740         __le32 incompat_features;       /* Used to indicate any incompatible features */
1741
1742         /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
1743 } __packed;
1744
1745 /*
1746  * Check for reshape constraints on raid set @rs:
1747  *
1748  * - reshape function non-existent
1749  * - degraded set
1750  * - ongoing recovery
1751  * - ongoing reshape
1752  *
1753  * Returns 0 if none or -EPERM if given constraint
1754  * and error message reference in @errmsg
1755  */
1756 static int rs_check_reshape(struct raid_set *rs)
1757 {
1758         struct mddev *mddev = &rs->md;
1759
1760         smp_rmb(); /* Make sure we access recent reshape position */
1761
1762         if (!mddev->pers || !mddev->pers->check_reshape)
1763                 rs->ti->error = "Reshape not supported";
1764         else if (mddev->degraded)
1765                 rs->ti->error = "Can't reshape degraded raid set";
1766         else if (rs_is_recovering(rs))
1767                 rs->ti->error = "Convert request on recovering raid set prohibited";
1768         else if (mddev->reshape_position && rs_is_reshaping(rs))
1769                 rs->ti->error = "raid set already reshaping!";
1770         else if (!(rs_is_raid10(rs) || rs_is_raid456(rs)))
1771                 rs->ti->error = "Reshaping only supported for raid4/5/6/10";
1772         else
1773                 return 0;
1774
1775         return -EPERM;
1776 }
1777
1778 static int read_disk_sb(struct md_rdev *rdev, int size)
1779 {
1780         BUG_ON(!rdev->sb_page);
1781
1782         if (rdev->sb_loaded)
1783                 return 0;
1784
1785         if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, 1)) {
1786                 DMERR("Failed to read superblock of device at position %d",
1787                       rdev->raid_disk);
1788                 md_error(rdev->mddev, rdev);
1789                 return -EINVAL;
1790         }
1791
1792         rdev->sb_loaded = 1;
1793
1794         return 0;
1795 }
1796
1797 static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
1798 {
1799         failed_devices[0] = le64_to_cpu(sb->failed_devices);
1800         memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
1801
1802         if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
1803                 int i = ARRAY_SIZE(sb->extended_failed_devices);
1804
1805                 while (i--)
1806                         failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
1807         }
1808 }
1809
1810 static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
1811 {
1812         int i = ARRAY_SIZE(sb->extended_failed_devices);
1813
1814         sb->failed_devices = cpu_to_le64(failed_devices[0]);
1815         while (i--)
1816                 sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
1817 }
1818
1819 /*
1820  * Synchronize the superblock members with the raid set properties
1821  *
1822  * All superblock data is little endian.
1823  */
1824 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
1825 {
1826         bool update_failed_devices = false;
1827         unsigned int i;
1828         uint64_t failed_devices[DISKS_ARRAY_ELEMS];
1829         struct dm_raid_superblock *sb;
1830         struct raid_set *rs = container_of(mddev, struct raid_set, md);
1831
1832         /* No metadata device, no superblock */
1833         if (!rdev->meta_bdev)
1834                 return;
1835
1836         BUG_ON(!rdev->sb_page);
1837
1838         sb = page_address(rdev->sb_page);
1839
1840         sb_retrieve_failed_devices(sb, failed_devices);
1841
1842         for (i = 0; i < rs->raid_disks; i++)
1843                 if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
1844                         update_failed_devices = true;
1845                         set_bit(i, (void *) failed_devices);
1846                 }
1847
1848         if (update_failed_devices)
1849                 sb_update_failed_devices(sb, failed_devices);
1850
1851         sb->magic = cpu_to_le32(DM_RAID_MAGIC);
1852         sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
1853
1854         sb->num_devices = cpu_to_le32(mddev->raid_disks);
1855         sb->array_position = cpu_to_le32(rdev->raid_disk);
1856
1857         sb->events = cpu_to_le64(mddev->events);
1858
1859         sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
1860         sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
1861
1862         sb->level = cpu_to_le32(mddev->level);
1863         sb->layout = cpu_to_le32(mddev->layout);
1864         sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
1865
1866         sb->new_level = cpu_to_le32(mddev->new_level);
1867         sb->new_layout = cpu_to_le32(mddev->new_layout);
1868         sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
1869
1870         sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1871
1872         smp_rmb(); /* Make sure we access most recent reshape position */
1873         sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1874         if (le64_to_cpu(sb->reshape_position) != MaxSector) {
1875                 /* Flag ongoing reshape */
1876                 sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
1877
1878                 if (mddev->delta_disks < 0 || mddev->reshape_backwards)
1879                         sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
1880         } else {
1881                 /* Clear reshape flags */
1882                 sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
1883         }
1884
1885         sb->array_sectors = cpu_to_le64(mddev->array_sectors);
1886         sb->data_offset = cpu_to_le64(rdev->data_offset);
1887         sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
1888         sb->sectors = cpu_to_le64(rdev->sectors);
1889
1890         /* Zero out the rest of the payload after the size of the superblock */
1891         memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
1892 }
1893
1894 /*
1895  * super_load
1896  *
1897  * This function creates a superblock if one is not found on the device
1898  * and will decide which superblock to use if there's a choice.
1899  *
1900  * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
1901  */
1902 static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
1903 {
1904         int r;
1905         struct dm_raid_superblock *sb;
1906         struct dm_raid_superblock *refsb;
1907         uint64_t events_sb, events_refsb;
1908
1909         rdev->sb_start = 0;
1910         rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
1911         if (rdev->sb_size < sizeof(*sb) || rdev->sb_size > PAGE_SIZE) {
1912                 DMERR("superblock size of a logical block is no longer valid");
1913                 return -EINVAL;
1914         }
1915
1916         r = read_disk_sb(rdev, rdev->sb_size);
1917         if (r)
1918                 return r;
1919
1920         sb = page_address(rdev->sb_page);
1921
1922         /*
1923          * Two cases that we want to write new superblocks and rebuild:
1924          * 1) New device (no matching magic number)
1925          * 2) Device specified for rebuild (!In_sync w/ offset == 0)
1926          */
1927         if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
1928             (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
1929                 super_sync(rdev->mddev, rdev);
1930
1931                 set_bit(FirstUse, &rdev->flags);
1932                 sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
1933
1934                 /* Force writing of superblocks to disk */
1935                 set_bit(MD_CHANGE_DEVS, &rdev->mddev->flags);
1936
1937                 /* Any superblock is better than none, choose that if given */
1938                 return refdev ? 0 : 1;
1939         }
1940
1941         if (!refdev)
1942                 return 1;
1943
1944         events_sb = le64_to_cpu(sb->events);
1945
1946         refsb = page_address(refdev->sb_page);
1947         events_refsb = le64_to_cpu(refsb->events);
1948
1949         return (events_sb > events_refsb) ? 1 : 0;
1950 }
1951
1952 static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
1953 {
1954         int role;
1955         unsigned int d;
1956         struct mddev *mddev = &rs->md;
1957         uint64_t events_sb;
1958         uint64_t failed_devices[DISKS_ARRAY_ELEMS];
1959         struct dm_raid_superblock *sb;
1960         uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
1961         struct md_rdev *r;
1962         struct dm_raid_superblock *sb2;
1963
1964         sb = page_address(rdev->sb_page);
1965         events_sb = le64_to_cpu(sb->events);
1966
1967         /*
1968          * Initialise to 1 if this is a new superblock.
1969          */
1970         mddev->events = events_sb ? : 1;
1971
1972         mddev->reshape_position = MaxSector;
1973
1974         /*
1975          * Reshaping is supported, e.g. reshape_position is valid
1976          * in superblock and superblock content is authoritative.
1977          */
1978         if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
1979                 /* Superblock is authoritative wrt given raid set layout! */
1980                 mddev->raid_disks = le32_to_cpu(sb->num_devices);
1981                 mddev->level = le32_to_cpu(sb->level);
1982                 mddev->layout = le32_to_cpu(sb->layout);
1983                 mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
1984                 mddev->new_level = le32_to_cpu(sb->new_level);
1985                 mddev->new_layout = le32_to_cpu(sb->new_layout);
1986                 mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
1987                 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1988                 mddev->array_sectors = le64_to_cpu(sb->array_sectors);
1989
1990                 /* raid was reshaping and got interrupted */
1991                 if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
1992                         if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1993                                 DMERR("Reshape requested but raid set is still reshaping");
1994                                 return -EINVAL;
1995                         }
1996
1997                         if (mddev->delta_disks < 0 ||
1998                             (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
1999                                 mddev->reshape_backwards = 1;
2000                         else
2001                                 mddev->reshape_backwards = 0;
2002
2003                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2004                         rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2005                 }
2006
2007         } else {
2008                 /*
2009                  * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2010                  */
2011                 if (le32_to_cpu(sb->level) != mddev->level) {
2012                         DMERR("Reshaping/takeover raid sets not yet supported. (raid level/stripes/size change)");
2013                         return -EINVAL;
2014                 }
2015                 if (le32_to_cpu(sb->layout) != mddev->layout) {
2016                         DMERR("Reshaping raid sets not yet supported. (raid layout change)");
2017                         DMERR("  0x%X vs 0x%X", le32_to_cpu(sb->layout), mddev->layout);
2018                         DMERR("  Old layout: %s w/ %d copies",
2019                               raid10_md_layout_to_format(le32_to_cpu(sb->layout)),
2020                               raid10_md_layout_to_copies(le32_to_cpu(sb->layout)));
2021                         DMERR("  New layout: %s w/ %d copies",
2022                               raid10_md_layout_to_format(mddev->layout),
2023                               raid10_md_layout_to_copies(mddev->layout));
2024                         return -EINVAL;
2025                 }
2026                 if (le32_to_cpu(sb->stripe_sectors) != mddev->chunk_sectors) {
2027                         DMERR("Reshaping raid sets not yet supported. (stripe sectors change)");
2028                         return -EINVAL;
2029                 }
2030
2031                 /* We can only change the number of devices in raid1 with old (i.e. pre 1.0.7) metadata */
2032                 if (!rt_is_raid1(rs->raid_type) &&
2033                     (le32_to_cpu(sb->num_devices) != mddev->raid_disks)) {
2034                         DMERR("Reshaping raid sets not yet supported. (device count change from %u to %u)",
2035                               sb->num_devices, mddev->raid_disks);
2036                         return -EINVAL;
2037                 }
2038
2039                 /* Table line is checked vs. authoritative superblock */
2040                 rs_set_new(rs);
2041         }
2042
2043         if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2044                 mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2045
2046         /*
2047          * During load, we set FirstUse if a new superblock was written.
2048          * There are two reasons we might not have a superblock:
2049          * 1) The raid set is brand new - in which case, all of the
2050          *    devices must have their In_sync bit set.  Also,
2051          *    recovery_cp must be 0, unless forced.
2052          * 2) This is a new device being added to an old raid set
2053          *    and the new device needs to be rebuilt - in which
2054          *    case the In_sync bit will /not/ be set and
2055          *    recovery_cp must be MaxSector.
2056          * 3) This is/are a new device(s) being added to an old
2057          *    raid set during takeover to a higher raid level
2058          *    to provide capacity for redundancy or during reshape
2059          *    to add capacity to grow the raid set.
2060          */
2061         d = 0;
2062         rdev_for_each(r, mddev) {
2063                 if (test_bit(FirstUse, &r->flags))
2064                         new_devs++;
2065
2066                 if (!test_bit(In_sync, &r->flags)) {
2067                         DMINFO("Device %d specified for rebuild; clearing superblock",
2068                                 r->raid_disk);
2069                         rebuilds++;
2070
2071                         if (test_bit(FirstUse, &r->flags))
2072                                 rebuild_and_new++;
2073                 }
2074
2075                 d++;
2076         }
2077
2078         if (new_devs == rs->raid_disks || !rebuilds) {
2079                 /* Replace a broken device */
2080                 if (new_devs == 1 && !rs->delta_disks)
2081                         ;
2082                 if (new_devs == rs->raid_disks) {
2083                         DMINFO("Superblocks created for new raid set");
2084                         set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2085                         mddev->recovery_cp = 0;
2086                 } else if (new_devs != rebuilds &&
2087                            new_devs != rs->delta_disks) {
2088                         DMERR("New device injected into existing raid set without "
2089                               "'delta_disks' or 'rebuild' parameter specified");
2090                         return -EINVAL;
2091                 }
2092         } else if (new_devs && new_devs != rebuilds) {
2093                 DMERR("%u 'rebuild' devices cannot be injected into"
2094                       " a raid set with %u other first-time devices",
2095                       rebuilds, new_devs);
2096                 return -EINVAL;
2097         } else if (rebuilds) {
2098                 if (rebuild_and_new && rebuilds != rebuild_and_new) {
2099                         DMERR("new device%s provided without 'rebuild'",
2100                               new_devs > 1 ? "s" : "");
2101                         return -EINVAL;
2102                 } else if (rs_is_recovering(rs)) {
2103                         DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2104                               (unsigned long long) mddev->recovery_cp);
2105                         return -EINVAL;
2106                 } else if (rs_is_reshaping(rs)) {
2107                         DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2108                               (unsigned long long) mddev->reshape_position);
2109                         return -EINVAL;
2110                 }
2111         }
2112
2113         /*
2114          * Now we set the Faulty bit for those devices that are
2115          * recorded in the superblock as failed.
2116          */
2117         sb_retrieve_failed_devices(sb, failed_devices);
2118         rdev_for_each(r, mddev) {
2119                 if (!r->sb_page)
2120                         continue;
2121                 sb2 = page_address(r->sb_page);
2122                 sb2->failed_devices = 0;
2123                 memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2124
2125                 /*
2126                  * Check for any device re-ordering.
2127                  */
2128                 if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2129                         role = le32_to_cpu(sb2->array_position);
2130                         if (role < 0)
2131                                 continue;
2132
2133                         if (role != r->raid_disk) {
2134                                 if (__is_raid10_near(mddev->layout)) {
2135                                         if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2136                                             rs->raid_disks % rs->raid10_copies) {
2137                                                 rs->ti->error =
2138                                                         "Cannot change raid10 near set to odd # of devices!";
2139                                                 return -EINVAL;
2140                                         }
2141
2142                                         sb2->array_position = cpu_to_le32(r->raid_disk);
2143
2144                                 } else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2145                                            !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2146                                            !rt_is_raid1(rs->raid_type)) {
2147                                         rs->ti->error = "Cannot change device positions in raid set";
2148                                         return -EINVAL;
2149                                 }
2150
2151                                 DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2152                         }
2153
2154                         /*
2155                          * Partial recovery is performed on
2156                          * returning failed devices.
2157                          */
2158                         if (test_bit(role, (void *) failed_devices))
2159                                 set_bit(Faulty, &r->flags);
2160                 }
2161         }
2162
2163         return 0;
2164 }
2165
2166 static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2167 {
2168         struct mddev *mddev = &rs->md;
2169         struct dm_raid_superblock *sb;
2170
2171         if (rs_is_raid0(rs) || !rdev->sb_page)
2172                 return 0;
2173
2174         sb = page_address(rdev->sb_page);
2175
2176         /*
2177          * If mddev->events is not set, we know we have not yet initialized
2178          * the array.
2179          */
2180         if (!mddev->events && super_init_validation(rs, rdev))
2181                 return -EINVAL;
2182
2183         if (le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2184                 rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2185                 return -EINVAL;
2186         }
2187
2188         if (sb->incompat_features) {
2189                 rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2190                 return -EINVAL;
2191         }
2192
2193         /* Enable bitmap creation for RAID levels != 0 */
2194         mddev->bitmap_info.offset = rt_is_raid0(rs->raid_type) ? 0 : to_sector(4096);
2195         rdev->mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2196
2197         if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2198                 /* Retrieve device size stored in superblock to be prepared for shrink */
2199                 rdev->sectors = le64_to_cpu(sb->sectors);
2200                 rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2201                 if (rdev->recovery_offset == MaxSector)
2202                         set_bit(In_sync, &rdev->flags);
2203                 /*
2204                  * If no reshape in progress -> we're recovering single
2205                  * disk(s) and have to set the device(s) to out-of-sync
2206                  */
2207                 else if (!rs_is_reshaping(rs))
2208                         clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2209         }
2210
2211         /*
2212          * If a device comes back, set it as not In_sync and no longer faulty.
2213          */
2214         if (test_and_clear_bit(Faulty, &rdev->flags)) {
2215                 rdev->recovery_offset = 0;
2216                 clear_bit(In_sync, &rdev->flags);
2217                 rdev->saved_raid_disk = rdev->raid_disk;
2218         }
2219
2220         /* Reshape support -> restore repective data offsets */
2221         rdev->data_offset = le64_to_cpu(sb->data_offset);
2222         rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2223
2224         return 0;
2225 }
2226
2227 /*
2228  * Analyse superblocks and select the freshest.
2229  */
2230 static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2231 {
2232         int r;
2233         struct raid_dev *dev;
2234         struct md_rdev *rdev, *tmp, *freshest;
2235         struct mddev *mddev = &rs->md;
2236
2237         freshest = NULL;
2238         rdev_for_each_safe(rdev, tmp, mddev) {
2239                 /*
2240                  * Skipping super_load due to CTR_FLAG_SYNC will cause
2241                  * the array to undergo initialization again as
2242                  * though it were new.  This is the intended effect
2243                  * of the "sync" directive.
2244                  *
2245                  * When reshaping capability is added, we must ensure
2246                  * that the "sync" directive is disallowed during the
2247                  * reshape.
2248                  */
2249                 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2250                         continue;
2251
2252                 if (!rdev->meta_bdev)
2253                         continue;
2254
2255                 r = super_load(rdev, freshest);
2256
2257                 switch (r) {
2258                 case 1:
2259                         freshest = rdev;
2260                         break;
2261                 case 0:
2262                         break;
2263                 default:
2264                         dev = container_of(rdev, struct raid_dev, rdev);
2265                         if (dev->meta_dev)
2266                                 dm_put_device(ti, dev->meta_dev);
2267
2268                         dev->meta_dev = NULL;
2269                         rdev->meta_bdev = NULL;
2270
2271                         if (rdev->sb_page)
2272                                 put_page(rdev->sb_page);
2273
2274                         rdev->sb_page = NULL;
2275
2276                         rdev->sb_loaded = 0;
2277
2278                         /*
2279                          * We might be able to salvage the data device
2280                          * even though the meta device has failed.  For
2281                          * now, we behave as though '- -' had been
2282                          * set for this device in the table.
2283                          */
2284                         if (dev->data_dev)
2285                                 dm_put_device(ti, dev->data_dev);
2286
2287                         dev->data_dev = NULL;
2288                         rdev->bdev = NULL;
2289
2290                         list_del(&rdev->same_set);
2291                 }
2292         }
2293
2294         if (!freshest)
2295                 return 0;
2296
2297         if (validate_raid_redundancy(rs)) {
2298                 rs->ti->error = "Insufficient redundancy to activate array";
2299                 return -EINVAL;
2300         }
2301
2302         /*
2303          * Validation of the freshest device provides the source of
2304          * validation for the remaining devices.
2305          */
2306         rs->ti->error = "Unable to assemble array: Invalid superblocks";
2307         if (super_validate(rs, freshest))
2308                 return -EINVAL;
2309
2310         rdev_for_each(rdev, mddev)
2311                 if ((rdev != freshest) && super_validate(rs, rdev))
2312                         return -EINVAL;
2313         return 0;
2314 }
2315
2316 /*
2317  * Adjust data_offset and new_data_offset on all disk members of @rs
2318  * for out of place reshaping if requested by contructor
2319  *
2320  * We need free space at the beginning of each raid disk for forward
2321  * and at the end for backward reshapes which userspace has to provide
2322  * via remapping/reordering of space.
2323  */
2324 static int rs_adjust_data_offsets(struct raid_set *rs)
2325 {
2326         sector_t data_offset = 0, new_data_offset = 0;
2327         struct md_rdev *rdev;
2328
2329         /* Constructor did not request data offset change */
2330         if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2331                 if (!rs_is_reshapable(rs))
2332                         goto out;
2333
2334                 return 0;
2335         }
2336
2337         /* HM FIXME: get InSync raid_dev? */
2338         rdev = &rs->dev[0].rdev;
2339
2340         if (rs->delta_disks < 0) {
2341                 /*
2342                  * Removing disks (reshaping backwards):
2343                  *
2344                  * - before reshape: data is at offset 0 and free space
2345                  *                   is at end of each component LV
2346                  *
2347                  * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2348                  */
2349                 data_offset = 0;
2350                 new_data_offset = rs->data_offset;
2351
2352         } else if (rs->delta_disks > 0) {
2353                 /*
2354                  * Adding disks (reshaping forwards):
2355                  *
2356                  * - before reshape: data is at offset rs->data_offset != 0 and
2357                  *                   free space is at begin of each component LV
2358                  *
2359                  * - after reshape: data is at offset 0 on each component LV
2360                  */
2361                 data_offset = rs->data_offset;
2362                 new_data_offset = 0;
2363
2364         } else {
2365                 /*
2366                  * User space passes in 0 for data offset after having removed reshape space
2367                  *
2368                  * - or - (data offset != 0)
2369                  *
2370                  * Changing RAID layout or chunk size -> toggle offsets
2371                  *
2372                  * - before reshape: data is at offset rs->data_offset 0 and
2373                  *                   free space is at end of each component LV
2374                  *                   -or-
2375                  *                   data is at offset rs->data_offset != 0 and
2376                  *                   free space is at begin of each component LV
2377                  *
2378                  * - after reshape: data is at offset 0 if i was at offset != 0
2379                  *                  of at offset != 0 if it was at offset 0
2380                  *                  on each component LV
2381                  *
2382                  */
2383                 data_offset = rs->data_offset ? rdev->data_offset : 0;
2384                 new_data_offset = data_offset ? 0 : rs->data_offset;
2385                 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2386         }
2387
2388         /*
2389          * Make sure we got a minimum amount of free sectors per device
2390          */
2391         if (rs->data_offset &&
2392             to_sector(i_size_read(rdev->bdev->bd_inode)) - rdev->sectors < MIN_FREE_RESHAPE_SPACE) {
2393                 rs->ti->error = data_offset ? "No space for forward reshape" :
2394                                               "No space for backward reshape";
2395                 return -ENOSPC;
2396         }
2397 out:
2398         /* Adjust data offsets on all rdevs */
2399         rdev_for_each(rdev, &rs->md) {
2400                 rdev->data_offset = data_offset;
2401                 rdev->new_data_offset = new_data_offset;
2402         }
2403
2404         return 0;
2405 }
2406
2407 /* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2408 static void __reorder_raid_disk_indexes(struct raid_set *rs)
2409 {
2410         int i = 0;
2411         struct md_rdev *rdev;
2412
2413         rdev_for_each(rdev, &rs->md) {
2414                 rdev->raid_disk = i++;
2415                 rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2416         }
2417 }
2418
2419 /*
2420  * Setup @rs for takeover by a different raid level
2421  */
2422 static int rs_setup_takeover(struct raid_set *rs)
2423 {
2424         struct mddev *mddev = &rs->md;
2425         struct md_rdev *rdev;
2426         unsigned int d = mddev->raid_disks = rs->raid_disks;
2427         sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2428
2429         if (rt_is_raid10(rs->raid_type)) {
2430                 if (mddev->level == 0) {
2431                         /* Userpace reordered disks -> adjust raid_disk indexes */
2432                         __reorder_raid_disk_indexes(rs);
2433
2434                         /* raid0 -> raid10_far layout */
2435                         mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2436                                                                    rs->raid10_copies);
2437                 } else if (mddev->level == 1)
2438                         /* raid1 -> raid10_near layout */
2439                         mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2440                                                                    rs->raid_disks);
2441                  else
2442                         return -EINVAL;
2443
2444         }
2445
2446         clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2447         mddev->recovery_cp = MaxSector;
2448
2449         while (d--) {
2450                 rdev = &rs->dev[d].rdev;
2451
2452                 if (test_bit(d, (void *) rs->rebuild_disks)) {
2453                         clear_bit(In_sync, &rdev->flags);
2454                         clear_bit(Faulty, &rdev->flags);
2455                         mddev->recovery_cp = rdev->recovery_offset = 0;
2456                         /* Bitmap has to be created when we do an "up" takeover */
2457                         set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2458                 }
2459
2460                 rdev->new_data_offset = new_data_offset;
2461         }
2462
2463         return 0;
2464 }
2465
2466 /*
2467  *
2468  * - change raid layout
2469  * - change chunk size
2470  * - add disks
2471  * - remove disks
2472  */
2473 static int rs_setup_reshape(struct raid_set *rs)
2474 {
2475         int r = 0;
2476         unsigned int cur_raid_devs, d;
2477         struct mddev *mddev = &rs->md;
2478         struct md_rdev *rdev;
2479
2480         mddev->delta_disks = rs->delta_disks;
2481         cur_raid_devs = mddev->raid_disks;
2482
2483         /* Ignore impossible layout change whilst adding/removing disks */
2484         if (mddev->delta_disks &&
2485             mddev->layout != mddev->new_layout) {
2486                 DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2487                 mddev->new_layout = mddev->layout;
2488         }
2489
2490         /*
2491          * Adjust array size:
2492          *
2493          * - in case of adding disks, array size has
2494          *   to grow after the disk adding reshape,
2495          *   which'll hapen in the event handler;
2496          *   reshape will happen forward, so space has to
2497          *   be available at the beginning of each disk
2498          *
2499          * - in case of removing disks, array size
2500          *   has to shrink before starting the reshape,
2501          *   which'll happen here;
2502          *   reshape will happen backward, so space has to
2503          *   be available at the end of each disk
2504          *
2505          * - data_offset and new_data_offset are
2506          *   adjusted for afreentioned out of place
2507          *   reshaping based on userspace passing in
2508          *   the "data_offset <sectors>" key/value
2509          *   pair via te constructor
2510          */
2511
2512         /* Add disk(s) */
2513         if (rs->delta_disks > 0) {
2514                 /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2515                 for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2516                         rdev = &rs->dev[d].rdev;
2517                         clear_bit(In_sync, &rdev->flags);
2518
2519                         /*
2520                          * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2521                          * by md, which'll store that erroneously in the superblock on reshape
2522                          */
2523                         rdev->saved_raid_disk = -1;
2524                         rdev->raid_disk = d;
2525
2526                         rdev->sectors = mddev->dev_sectors;
2527                         rdev->recovery_offset = MaxSector;
2528                 }
2529
2530                 mddev->reshape_backwards = 0; /* adding disks -> forward reshape */
2531
2532         /* Remove disk(s) */
2533         } else if (rs->delta_disks < 0) {
2534                 r = rs_set_dev_and_array_sectors(rs, true);
2535                 mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2536
2537         /* Change layout and/or chunk size */
2538         } else {
2539                 /*
2540                  * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2541                  *
2542                  * keeping number of disks and do layout change ->
2543                  *
2544                  * toggle reshape_backward depending on data_offset:
2545                  *
2546                  * - free space upfront -> reshape forward
2547                  *
2548                  * - free space at the end -> reshape backward
2549                  *
2550                  *
2551                  * This utilizes free reshape space avoiding the need
2552                  * for userspace to move (parts of) LV segments in
2553                  * case of layout/chunksize change  (for disk
2554                  * adding/removing reshape space has to be at
2555                  * the proper address (see above with delta_disks):
2556                  *
2557                  * add disk(s)   -> begin
2558                  * remove disk(s)-> end
2559                  */
2560                 mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2561         }
2562
2563         return r;
2564 }
2565
2566 /*
2567  * Enable/disable discard support on RAID set depending on
2568  * RAID level and discard properties of underlying RAID members.
2569  */
2570 static void configure_discard_support(struct raid_set *rs)
2571 {
2572         int i;
2573         bool raid456;
2574         struct dm_target *ti = rs->ti;
2575
2576         /* Assume discards not supported until after checks below. */
2577         ti->discards_supported = false;
2578
2579         /* RAID level 4,5,6 require discard_zeroes_data for data integrity! */
2580         raid456 = (rs->md.level == 4 || rs->md.level == 5 || rs->md.level == 6);
2581
2582         for (i = 0; i < rs->md.raid_disks; i++) {
2583                 struct request_queue *q;
2584
2585                 if (!rs->dev[i].rdev.bdev)
2586                         continue;
2587
2588                 q = bdev_get_queue(rs->dev[i].rdev.bdev);
2589                 if (!q || !blk_queue_discard(q))
2590                         return;
2591
2592                 if (raid456) {
2593                         if (!q->limits.discard_zeroes_data)
2594                                 return;
2595                         if (!devices_handle_discard_safely) {
2596                                 DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2597                                 DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2598                                 return;
2599                         }
2600                 }
2601         }
2602
2603         /* All RAID members properly support discards */
2604         ti->discards_supported = true;
2605
2606         /*
2607          * RAID1 and RAID10 personalities require bio splitting,
2608          * RAID0/4/5/6 don't and process large discard bios properly.
2609          */
2610         ti->split_discard_bios = !!(rs->md.level == 1 || rs->md.level == 10);
2611         ti->num_discard_bios = 1;
2612 }
2613
2614 /*
2615  * Construct a RAID0/1/10/4/5/6 mapping:
2616  * Args:
2617  *      <raid_type> <#raid_params> <raid_params>{0,}    \
2618  *      <#raid_devs> [<meta_dev1> <dev1>]{1,}
2619  *
2620  * <raid_params> varies by <raid_type>.  See 'parse_raid_params' for
2621  * details on possible <raid_params>.
2622  *
2623  * Userspace is free to initialize the metadata devices, hence the superblocks to
2624  * enforce recreation based on the passed in table parameters.
2625  *
2626  */
2627 static int raid_ctr(struct dm_target *ti, unsigned argc, char **argv)
2628 {
2629         int r;
2630         struct raid_type *rt;
2631         unsigned num_raid_params, num_raid_devs;
2632         struct raid_set *rs = NULL;
2633         const char *arg;
2634         struct rs_layout rs_layout;
2635         struct dm_arg_set as = { argc, argv }, as_nrd;
2636         struct dm_arg _args[] = {
2637                 { 0, as.argc, "Cannot understand number of raid parameters" },
2638                 { 1, 254, "Cannot understand number of raid devices parameters" }
2639         };
2640
2641         /* Must have <raid_type> */
2642         arg = dm_shift_arg(&as);
2643         if (!arg) {
2644                 ti->error = "No arguments";
2645                 return -EINVAL;
2646         }
2647
2648         rt = get_raid_type(arg);
2649         if (!rt) {
2650                 ti->error = "Unrecognised raid_type";
2651                 return -EINVAL;
2652         }
2653
2654         /* Must have <#raid_params> */
2655         if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
2656                 return -EINVAL;
2657
2658         /* number of raid device tupples <meta_dev data_dev> */
2659         as_nrd = as;
2660         dm_consume_args(&as_nrd, num_raid_params);
2661         _args[1].max = (as_nrd.argc - 1) / 2;
2662         if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
2663                 return -EINVAL;
2664
2665         if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
2666                 ti->error = "Invalid number of supplied raid devices";
2667                 return -EINVAL;
2668         }
2669
2670         rs = raid_set_alloc(ti, rt, num_raid_devs);
2671         if (IS_ERR(rs))
2672                 return PTR_ERR(rs);
2673
2674         r = parse_raid_params(rs, &as, num_raid_params);
2675         if (r)
2676                 goto bad;
2677
2678         r = parse_dev_params(rs, &as);
2679         if (r)
2680                 goto bad;
2681
2682         rs->md.sync_super = super_sync;
2683
2684         r = rs_set_dev_and_array_sectors(rs, false);
2685         if (r)
2686                 return r;
2687
2688         /*
2689          * Backup any new raid set level, layout, ...
2690          * requested to be able to compare to superblock
2691          * members for conversion decisions.
2692          */
2693         rs_config_backup(rs, &rs_layout);
2694
2695         r = analyse_superblocks(ti, rs);
2696         if (r)
2697                 goto bad;
2698
2699         INIT_WORK(&rs->md.event_work, do_table_event);
2700         ti->private = rs;
2701         ti->num_flush_bios = 1;
2702
2703         /* Restore any requested new layout for conversion decision */
2704         rs_config_restore(rs, &rs_layout);
2705
2706         if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
2707                 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2708                 rs_set_new(rs);
2709         } else if (rs_is_reshaping(rs))
2710                 ; /* skip rs setup */
2711         else if (rs_takeover_requested(rs)) {
2712                 if (rs_is_reshaping(rs)) {
2713                         ti->error = "Can't takeover a reshaping raid set";
2714                         return -EPERM;
2715                 }
2716
2717                 /*
2718                  * If a takeover is needed, just set the level to
2719                  * the new requested one and allow the raid set to run.
2720                  */
2721                 r = rs_check_takeover(rs);
2722                 if (r)
2723                         return r;
2724
2725                 r = rs_setup_takeover(rs);
2726                 if (r)
2727                         return r;
2728
2729                 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2730                 rs_set_new(rs);
2731         } else if (rs_reshape_requested(rs)) {
2732                 if (rs_is_reshaping(rs)) {
2733                         ti->error = "raid set already reshaping!";
2734                         return -EPERM;
2735                 }
2736
2737                 if (rs_is_raid10(rs)) {
2738                         if (rs->raid_disks != rs->md.raid_disks &&
2739                             __is_raid10_near(rs->md.layout) &&
2740                             rs->raid10_copies &&
2741                             rs->raid10_copies != __raid10_near_copies(rs->md.layout)) {
2742                                 /*
2743                                  * raid disk have to be multiple of data copies to allow this conversion,
2744                                  *
2745                                  * This is actually not a reshape it is a
2746                                  * rebuild of any additional mirrors per group
2747                                  */
2748                                 if (rs->raid_disks % rs->raid10_copies) {
2749                                         ti->error = "Can't reshape raid10 mirror groups";
2750                                         return -EINVAL;
2751                                 }
2752
2753                                 /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2754                                 __reorder_raid_disk_indexes(rs);
2755                                 rs->md.layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2756                                                                            rs->raid10_copies);
2757                                 rs->md.new_layout = rs->md.layout;
2758
2759                         } else
2760                                 set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2761
2762                 } else if (rs_is_raid456(rs))
2763                         set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2764
2765                 /*
2766                  * HM FIXME: process raid1 via delta_disks as well?
2767                  *           Would cause allocations in raid1->check_reshape
2768                  *           though, thus more issues with potential failures
2769                  */
2770                 else if (rs_is_raid1(rs))
2771                         rs->md.raid_disks = rs->raid_disks;
2772
2773                 if (rs->md.raid_disks < rs->raid_disks)
2774                         set_bit(MD_ARRAY_FIRST_USE, &rs->md.flags);
2775
2776                 set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2777                 rs_set_cur(rs);
2778         } else
2779                 rs_set_cur(rs);
2780
2781         /* If constructor requested it, change data and new_data offsets */
2782         r = rs_adjust_data_offsets(rs);
2783         if (r)
2784                 return r;
2785
2786         /* Start raid set read-only and assumed clean to change in raid_resume() */
2787         rs->md.ro = 1;
2788         rs->md.in_sync = 1;
2789         set_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
2790
2791         /* Has to be held on running the array */
2792         mddev_lock_nointr(&rs->md);
2793         r = md_run(&rs->md);
2794         rs->md.in_sync = 0; /* Assume already marked dirty */
2795
2796         if (r) {
2797                 ti->error = "Failed to run raid array";
2798                 mddev_unlock(&rs->md);
2799                 goto bad;
2800         }
2801
2802         rs->callbacks.congested_fn = raid_is_congested;
2803         dm_table_add_target_callbacks(ti->table, &rs->callbacks);
2804
2805         mddev_suspend(&rs->md);
2806
2807         /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
2808         if (rs_is_raid456(rs)) {
2809                 r = rs_set_raid456_stripe_cache(rs);
2810                 if (r)
2811                         goto bad_stripe_cache;
2812         }
2813
2814         /* Now do an early reshape check */
2815         if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
2816                 r = rs_check_reshape(rs);
2817                 if (r)
2818                         return r;
2819
2820                 /* Restore new, ctr requested layout to perform check */
2821                 rs_config_restore(rs, &rs_layout);
2822
2823                 r = rs->md.pers->check_reshape(&rs->md);
2824                 if (r) {
2825                         ti->error = "Reshape check failed";
2826                         goto bad_check_reshape;
2827                 }
2828         }
2829
2830         mddev_unlock(&rs->md);
2831         return 0;
2832
2833 bad_stripe_cache:
2834 bad_check_reshape:
2835         md_stop(&rs->md);
2836 bad:
2837         raid_set_free(rs);
2838
2839         return r;
2840 }
2841
2842 static void raid_dtr(struct dm_target *ti)
2843 {
2844         struct raid_set *rs = ti->private;
2845
2846         list_del_init(&rs->callbacks.list);
2847         md_stop(&rs->md);
2848         raid_set_free(rs);
2849 }
2850
2851 static int raid_map(struct dm_target *ti, struct bio *bio)
2852 {
2853         struct raid_set *rs = ti->private;
2854         struct mddev *mddev = &rs->md;
2855
2856         /*
2857          * If we're reshaping to add disk(s)), ti->len and
2858          * mddev->array_sectors will differ during the process
2859          * (ti->len > mddev->array_sectors), so we have to requeue
2860          * bios with addresses > mddev->array_sectors here or
2861          * or there will occur accesses past EOD of the component
2862          * data images thus erroring the raid set.
2863          */
2864         if (unlikely(bio_end_sector(bio) > mddev->array_sectors))
2865                 return DM_MAPIO_REQUEUE;
2866
2867         mddev->pers->make_request(mddev, bio);
2868
2869         return DM_MAPIO_SUBMITTED;
2870 }
2871
2872 /* Return string describing the current sync action of @mddev */
2873 static const char *decipher_sync_action(struct mddev *mddev)
2874 {
2875         if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
2876                 return "frozen";
2877
2878         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2879             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2880                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2881                         return "reshape";
2882
2883                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2884                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2885                                 return "resync";
2886                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2887                                 return "check";
2888                         return "repair";
2889                 }
2890
2891                 if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
2892                         return "recover";
2893         }
2894
2895         return "idle";
2896 }
2897
2898 /*
2899  * Return status string @rdev
2900  *
2901  * Status characters:
2902  *
2903  *  'D' = Dead/Failed device
2904  *  'a' = Alive but not in-sync
2905  *  'A' = Alive and in-sync
2906  */
2907 static const char *__raid_dev_status(struct md_rdev *rdev, bool array_in_sync)
2908 {
2909         if (test_bit(Faulty, &rdev->flags))
2910                 return "D";
2911         else if (!array_in_sync || !test_bit(In_sync, &rdev->flags))
2912                 return "a";
2913         else
2914                 return "A";
2915 }
2916
2917 /* Helper to return resync/reshape progress for @rs and @array_in_sync */
2918 static sector_t rs_get_progress(struct raid_set *rs,
2919                                 sector_t resync_max_sectors, bool *array_in_sync)
2920 {
2921         sector_t r, recovery_cp, curr_resync_completed;
2922         struct mddev *mddev = &rs->md;
2923
2924         curr_resync_completed = mddev->curr_resync_completed ?: mddev->recovery_cp;
2925         recovery_cp = mddev->recovery_cp;
2926         *array_in_sync = false;
2927
2928         if (rs_is_raid0(rs)) {
2929                 r = resync_max_sectors;
2930                 *array_in_sync = true;
2931
2932         } else {
2933                 r = mddev->reshape_position;
2934
2935                 /* Reshape is relative to the array size */
2936                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
2937                     r != MaxSector) {
2938                         if (r == MaxSector) {
2939                                 *array_in_sync = true;
2940                                 r = resync_max_sectors;
2941                         } else {
2942                                 /* Got to reverse on backward reshape */
2943                                 if (mddev->reshape_backwards)
2944                                         r = mddev->array_sectors - r;
2945
2946                                 /* Devide by # of data stripes */
2947                                 sector_div(r, mddev_data_stripes(rs));
2948                         }
2949
2950                 /* Sync is relative to the component device size */
2951                 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2952                         r = curr_resync_completed;
2953                 else
2954                         r = recovery_cp;
2955
2956                 if (r == MaxSector) {
2957                         /*
2958                          * Sync complete.
2959                          */
2960                         *array_in_sync = true;
2961                         r = resync_max_sectors;
2962                 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2963                         /*
2964                          * If "check" or "repair" is occurring, the raid set has
2965                          * undergone an initial sync and the health characters
2966                          * should not be 'a' anymore.
2967                          */
2968                         *array_in_sync = true;
2969                 } else {
2970                         struct md_rdev *rdev;
2971
2972                         /*
2973                          * The raid set may be doing an initial sync, or it may
2974                          * be rebuilding individual components.  If all the
2975                          * devices are In_sync, then it is the raid set that is
2976                          * being initialized.
2977                          */
2978                         rdev_for_each(rdev, mddev)
2979                                 if (!test_bit(In_sync, &rdev->flags))
2980                                         *array_in_sync = true;
2981 #if 0
2982                         r = 0; /* HM FIXME: TESTME: https://bugzilla.redhat.com/show_bug.cgi?id=1210637 ? */
2983 #endif
2984                 }
2985         }
2986
2987         return r;
2988 }
2989
2990 /* Helper to return @dev name or "-" if !@dev */
2991 static const char *__get_dev_name(struct dm_dev *dev)
2992 {
2993         return dev ? dev->name : "-";
2994 }
2995
2996 static void raid_status(struct dm_target *ti, status_type_t type,
2997                         unsigned int status_flags, char *result, unsigned int maxlen)
2998 {
2999         struct raid_set *rs = ti->private;
3000         struct mddev *mddev = &rs->md;
3001         struct r5conf *conf = mddev->private;
3002         int max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3003         bool array_in_sync;
3004         unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3005         unsigned int sz = 0;
3006         unsigned int write_mostly_params = 0;
3007         sector_t progress, resync_max_sectors, resync_mismatches;
3008         const char *sync_action;
3009         struct raid_type *rt;
3010         struct md_rdev *rdev;
3011
3012         switch (type) {
3013         case STATUSTYPE_INFO:
3014                 /* *Should* always succeed */
3015                 rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3016                 if (!rt)
3017                         return;
3018
3019                 DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3020
3021                 /* Access most recent mddev properties for status output */
3022                 smp_rmb();
3023                 /* Get sensible max sectors even if raid set not yet started */
3024                 resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3025                                       mddev->resync_max_sectors : mddev->dev_sectors;
3026                 progress = rs_get_progress(rs, resync_max_sectors, &array_in_sync);
3027                 resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
3028                                     atomic64_read(&mddev->resync_mismatches) : 0;
3029                 sync_action = decipher_sync_action(&rs->md);
3030
3031                 /* HM FIXME: do we want another state char for raid0? It shows 'D' or 'A' now */
3032                 rdev_for_each(rdev, mddev)
3033                         DMEMIT(__raid_dev_status(rdev, array_in_sync));
3034
3035                 /*
3036                  * In-sync/Reshape ratio:
3037                  *  The in-sync ratio shows the progress of:
3038                  *   - Initializing the raid set
3039                  *   - Rebuilding a subset of devices of the raid set
3040                  *  The user can distinguish between the two by referring
3041                  *  to the status characters.
3042                  *
3043                  *  The reshape ratio shows the progress of
3044                  *  changing the raid layout or the number of
3045                  *  disks of a raid set
3046                  */
3047                 DMEMIT(" %llu/%llu", (unsigned long long) progress,
3048                                      (unsigned long long) resync_max_sectors);
3049
3050                 /*
3051                  * v1.5.0+:
3052                  *
3053                  * Sync action:
3054                  *   See Documentation/device-mapper/dm-raid.txt for
3055                  *   information on each of these states.
3056                  */
3057                 DMEMIT(" %s", sync_action);
3058
3059                 /*
3060                  * v1.5.0+:
3061                  *
3062                  * resync_mismatches/mismatch_cnt
3063                  *   This field shows the number of discrepancies found when
3064                  *   performing a "check" of the raid set.
3065                  */
3066                 DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3067
3068                 /*
3069                  * v1.9.0+:
3070                  *
3071                  * data_offset (needed for out of space reshaping)
3072                  *   This field shows the data offset into the data
3073                  *   image LV where the first stripes data starts.
3074                  *
3075                  * We keep data_offset equal on all raid disks of the set,
3076                  * so retrieving it from the first raid disk is sufficient.
3077                  */
3078                 DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3079                 break;
3080
3081         case STATUSTYPE_TABLE:
3082                 /* Report the table line string you would use to construct this raid set */
3083
3084                 /* Calculate raid parameter count */
3085                 rdev_for_each(rdev, mddev)
3086                         if (test_bit(WriteMostly, &rdev->flags))
3087                                 write_mostly_params += 2;
3088                 raid_param_cnt += memweight(rs->rebuild_disks,
3089                                             DISKS_ARRAY_ELEMS * sizeof(*rs->rebuild_disks)) * 2 +
3090                                   write_mostly_params +
3091                                   hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3092                                   hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2;
3093                 /* Emit table line */
3094                 DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3095                 if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3096                         DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3097                                          raid10_md_layout_to_format(mddev->layout));
3098                 if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3099                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3100                                          raid10_md_layout_to_copies(mddev->layout));
3101                 if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3102                         DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3103                 if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3104                         DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3105                 if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3106                         DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3107                                            (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3108                 if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3109                         DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3110                                            (unsigned long long) rs->data_offset);
3111                 if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3112                         DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3113                                           mddev->bitmap_info.daemon_sleep);
3114                 if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3115                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3116                                          mddev->delta_disks);
3117                 if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3118                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3119                                          max_nr_stripes);
3120                 rdev_for_each(rdev, mddev)
3121                         if (test_bit(rdev->raid_disk, (void *) rs->rebuild_disks))
3122                                 DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD),
3123                                                  rdev->raid_disk);
3124                 rdev_for_each(rdev, mddev)
3125                         if (test_bit(WriteMostly, &rdev->flags))
3126                                 DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3127                                                  rdev->raid_disk);
3128                 if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3129                         DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3130                                           mddev->bitmap_info.max_write_behind);
3131                 if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3132                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3133                                          mddev->sync_speed_max);
3134                 if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3135                         DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3136                                          mddev->sync_speed_min);
3137                 DMEMIT(" %d", rs->raid_disks);
3138                 rdev_for_each(rdev, mddev) {
3139                         struct raid_dev *rd = container_of(rdev, struct raid_dev, rdev);
3140
3141                         DMEMIT(" %s %s", __get_dev_name(rd->meta_dev),
3142                                          __get_dev_name(rd->data_dev));
3143                 }
3144         }
3145 }
3146
3147 static int raid_message(struct dm_target *ti, unsigned argc, char **argv)
3148 {
3149         struct raid_set *rs = ti->private;
3150         struct mddev *mddev = &rs->md;
3151
3152         if (!mddev->pers || !mddev->pers->sync_request)
3153                 return -EINVAL;
3154
3155         if (!strcasecmp(argv[0], "frozen"))
3156                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3157         else
3158                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3159
3160         if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
3161                 if (mddev->sync_thread) {
3162                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3163                         md_reap_sync_thread(mddev);
3164                 }
3165         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3166                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3167                 return -EBUSY;
3168         else if (!strcasecmp(argv[0], "resync"))
3169                 ; /* MD_RECOVERY_NEEDED set below */
3170         else if (!strcasecmp(argv[0], "recover"))
3171                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3172         else {
3173                 if (!strcasecmp(argv[0], "check"))
3174                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3175                 else if (!!strcasecmp(argv[0], "repair"))
3176                         return -EINVAL;
3177                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3178                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3179         }
3180         if (mddev->ro == 2) {
3181                 /* A write to sync_action is enough to justify
3182                  * canceling read-auto mode
3183                  */
3184                 mddev->ro = 0;
3185                 if (!mddev->suspended && mddev->sync_thread)
3186                         md_wakeup_thread(mddev->sync_thread);
3187         }
3188         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3189         if (!mddev->suspended && mddev->thread)
3190                 md_wakeup_thread(mddev->thread);
3191
3192         return 0;
3193 }
3194
3195 static int raid_iterate_devices(struct dm_target *ti,
3196                                 iterate_devices_callout_fn fn, void *data)
3197 {
3198         struct raid_set *rs = ti->private;
3199         unsigned i;
3200         int r = 0;
3201
3202         for (i = 0; !r && i < rs->md.raid_disks; i++)
3203                 if (rs->dev[i].data_dev)
3204                         r = fn(ti,
3205                                  rs->dev[i].data_dev,
3206                                  0, /* No offset on data devs */
3207                                  rs->md.dev_sectors,
3208                                  data);
3209
3210         return r;
3211 }
3212
3213 static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3214 {
3215         struct raid_set *rs = ti->private;
3216         unsigned chunk_size = rs->md.chunk_sectors << 9;
3217         struct r5conf *conf = rs->md.private;
3218
3219         blk_limits_io_min(limits, chunk_size);
3220         blk_limits_io_opt(limits, chunk_size * (conf->raid_disks - conf->max_degraded));
3221 }
3222
3223 static void raid_presuspend(struct dm_target *ti)
3224 {
3225         struct raid_set *rs = ti->private;
3226
3227         md_stop_writes(&rs->md);
3228 }
3229
3230 static void raid_postsuspend(struct dm_target *ti)
3231 {
3232         struct raid_set *rs = ti->private;
3233
3234         mddev_suspend(&rs->md);
3235         rs->md.ro = 1;
3236         clear_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags);
3237 }
3238
3239 static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3240 {
3241         int i;
3242         uint64_t failed_devices, cleared_failed_devices = 0;
3243         unsigned long flags;
3244         struct dm_raid_superblock *sb;
3245         struct md_rdev *r;
3246
3247         for (i = 0; i < rs->md.raid_disks; i++) {
3248                 r = &rs->dev[i].rdev;
3249                 if (test_bit(Faulty, &r->flags) && r->sb_page &&
3250                     sync_page_io(r, 0, r->sb_size, r->sb_page, REQ_OP_READ, 0,
3251                                  1)) {
3252                         DMINFO("Faulty %s device #%d has readable super block."
3253                                "  Attempting to revive it.",
3254                                rs->raid_type->name, i);
3255
3256                         /*
3257                          * Faulty bit may be set, but sometimes the array can
3258                          * be suspended before the personalities can respond
3259                          * by removing the device from the array (i.e. calling
3260                          * 'hot_remove_disk').  If they haven't yet removed
3261                          * the failed device, its 'raid_disk' number will be
3262                          * '>= 0' - meaning we must call this function
3263                          * ourselves.
3264                          */
3265                         if ((r->raid_disk >= 0) &&
3266                             (r->mddev->pers->hot_remove_disk(r->mddev, r) != 0))
3267                                 /* Failed to revive this device, try next */
3268                                 continue;
3269
3270                         r->raid_disk = i;
3271                         r->saved_raid_disk = i;
3272                         flags = r->flags;
3273                         clear_bit(Faulty, &r->flags);
3274                         clear_bit(WriteErrorSeen, &r->flags);
3275                         clear_bit(In_sync, &r->flags);
3276                         if (r->mddev->pers->hot_add_disk(r->mddev, r)) {
3277                                 r->raid_disk = -1;
3278                                 r->saved_raid_disk = -1;
3279                                 r->flags = flags;
3280                         } else {
3281                                 r->recovery_offset = 0;
3282                                 cleared_failed_devices |= 1 << i;
3283                         }
3284                 }
3285         }
3286         if (cleared_failed_devices) {
3287                 rdev_for_each(r, &rs->md) {
3288                         sb = page_address(r->sb_page);
3289                         failed_devices = le64_to_cpu(sb->failed_devices);
3290                         failed_devices &= ~cleared_failed_devices;
3291                         sb->failed_devices = cpu_to_le64(failed_devices);
3292                 }
3293         }
3294 }
3295
3296 static int __load_dirty_region_bitmap(struct raid_set *rs)
3297 {
3298         int r = 0;
3299
3300         /* Try loading the bitmap unless "raid0", which does not have one */
3301         if (!rs_is_raid0(rs) &&
3302             !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3303                 r = bitmap_load(&rs->md);
3304                 if (r)
3305                         DMERR("Failed to load bitmap");
3306         }
3307
3308         return r;
3309 }
3310
3311 /*
3312  * Reshape changes raid algorithm of @rs to new one within personality
3313  * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3314  * disks from a raid set thus growing/shrinking it or resizes the set
3315  *
3316  * Call mddev_lock_nointr() before!
3317  */
3318 static int rs_start_reshape(struct raid_set *rs)
3319 {
3320         int r;
3321         struct mddev *mddev = &rs->md;
3322         struct md_personality *pers = mddev->pers;
3323
3324         r = rs_setup_reshape(rs);
3325         if (r)
3326                 return r;
3327
3328         /* Need to be resumed to be able to start reshape, recovery is frozen until raid_resume() though */
3329         if (mddev->suspended)
3330                 mddev_resume(mddev);
3331
3332         /*
3333          * Check any reshape constraints enforced by the personalility
3334          *
3335          * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3336          */
3337         r = pers->check_reshape(mddev);
3338         if (r) {
3339                 rs->ti->error = "pers->check_reshape() failed";
3340                 return r;
3341         }
3342
3343         /*
3344          * Personality may not provide start reshape method in which
3345          * case check_reshape above has already covered everything
3346          */
3347         if (pers->start_reshape) {
3348                 r = pers->start_reshape(mddev);
3349                 if (r) {
3350                         rs->ti->error = "pers->start_reshape() failed";
3351                         return r;
3352                 }
3353         }
3354
3355         /* Suspend because a resume will happen in raid_resume() */
3356         if (!mddev->suspended)
3357                 mddev_suspend(mddev);
3358
3359         mddev->ro = 0;
3360         md_update_sb(mddev, 1);
3361         mddev->ro = 1;
3362
3363         return 0;
3364 }
3365
3366 static int raid_preresume(struct dm_target *ti)
3367 {
3368         int r;
3369         struct raid_set *rs = ti->private;
3370         struct mddev *mddev = &rs->md;
3371
3372         /* This is a resume after a suspend of the set -> it's already started */
3373         if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
3374                 return 0;
3375
3376         /*
3377          * The superblocks need to be updated on disk if the
3378          * array is new or __load_dirty_region_bitmap will overwrite them
3379          * in core with old data.
3380          *
3381          * In case the array got modified (takeover/reshape/resize)
3382          * or the data offsets on the component devices changed, they
3383          * have to be updated as well.
3384          *
3385          * Have to switch to readwrite and back in order to
3386          * allow for the superblock updates.
3387          */
3388         if (test_and_clear_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags)) {
3389                 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3390                 mddev->ro = 0;
3391                 md_update_sb(mddev, 1);
3392                 mddev->ro = 1;
3393         }
3394
3395         /*
3396          * Disable/enable discard support on raid set after any
3397          * conversion, because devices can have been added
3398          */
3399         configure_discard_support(rs);
3400
3401         /* Load the bitmap from disk unless raid0 */
3402         r = __load_dirty_region_bitmap(rs);
3403         if (r)
3404                 return r;
3405
3406         /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) */
3407         if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) &&
3408             mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)) {
3409                 r = bitmap_resize(mddev->bitmap, mddev->dev_sectors,
3410                                   to_bytes(rs->requested_bitmap_chunk_sectors), 0);
3411                 if (r)
3412                         DMERR("Failed to resize bitmap");
3413         }
3414
3415         /* Check for any resize/reshape on @rs and adjust/initiate */
3416         /* Be prepared for mddev_resume() in raid_resume() */
3417         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3418         if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
3419                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3420                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3421                 mddev->resync_min = mddev->recovery_cp;
3422         }
3423
3424         rs_set_capacity(rs);
3425
3426         /* Check for any reshape request and region size change unless new raid set */
3427         if (test_and_clear_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3428                 /* Initiate a reshape. */
3429                 mddev_lock_nointr(mddev);
3430                 r = rs_start_reshape(rs);
3431                 mddev_unlock(mddev);
3432                 if (r)
3433                         DMWARN("Failed to check/start reshape, continuing without change");
3434                 r = 0;
3435         }
3436
3437         return r;
3438 }
3439
3440 static void raid_resume(struct dm_target *ti)
3441 {
3442         struct raid_set *rs = ti->private;
3443         struct mddev *mddev = &rs->md;
3444
3445         if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
3446                 /*
3447                  * A secondary resume while the device is active.
3448                  * Take this opportunity to check whether any failed
3449                  * devices are reachable again.
3450                  */
3451                 attempt_restore_of_faulty_devices(rs);
3452         }
3453
3454         mddev->ro = 0;
3455         mddev->in_sync = 0;
3456         clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3457
3458         if (mddev->suspended)
3459                 mddev_resume(mddev);
3460 }
3461
3462 static struct target_type raid_target = {
3463         .name = "raid",
3464         .version = {1, 9, 0},
3465         .module = THIS_MODULE,
3466         .ctr = raid_ctr,
3467         .dtr = raid_dtr,
3468         .map = raid_map,
3469         .status = raid_status,
3470         .message = raid_message,
3471         .iterate_devices = raid_iterate_devices,
3472         .io_hints = raid_io_hints,
3473         .presuspend = raid_presuspend,
3474         .postsuspend = raid_postsuspend,
3475         .preresume = raid_preresume,
3476         .resume = raid_resume,
3477 };
3478
3479 static int __init dm_raid_init(void)
3480 {
3481         DMINFO("Loading target version %u.%u.%u",
3482                raid_target.version[0],
3483                raid_target.version[1],
3484                raid_target.version[2]);
3485         return dm_register_target(&raid_target);
3486 }
3487
3488 static void __exit dm_raid_exit(void)
3489 {
3490         dm_unregister_target(&raid_target);
3491 }
3492
3493 module_init(dm_raid_init);
3494 module_exit(dm_raid_exit);
3495
3496 module_param(devices_handle_discard_safely, bool, 0644);
3497 MODULE_PARM_DESC(devices_handle_discard_safely,
3498                  "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
3499
3500 MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
3501 MODULE_ALIAS("dm-raid0");
3502 MODULE_ALIAS("dm-raid1");
3503 MODULE_ALIAS("dm-raid10");
3504 MODULE_ALIAS("dm-raid4");
3505 MODULE_ALIAS("dm-raid5");
3506 MODULE_ALIAS("dm-raid6");
3507 MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
3508 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
3509 MODULE_LICENSE("GPL");