linux/drivers/md/dm-raid.c
<<
>>
Prefs
   1/*
   2 * Copyright (C) 2010-2011 Neil Brown
   3 * Copyright (C) 2010-2018 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 "md-bitmap.h"
  16
  17#include <linux/device-mapper.h>
  18
  19#define DM_MSG_PREFIX "raid"
  20#define MAX_RAID_DEVICES        253 /* md-raid kernel limit */
  21
  22/*
  23 * Minimum sectors of free reshape space per raid device
  24 */
  25#define MIN_FREE_RESHAPE_SPACE to_sector(4*4096)
  26
  27/*
  28 * Minimum journal space 4 MiB in sectors.
  29 */
  30#define MIN_RAID456_JOURNAL_SPACE (4*2048)
  31
  32static bool devices_handle_discard_safely = false;
  33
  34/*
  35 * The following flags are used by dm-raid.c to set up the array state.
  36 * They must be cleared before md_run is called.
  37 */
  38#define FirstUse 10             /* rdev flag */
  39
  40struct raid_dev {
  41        /*
  42         * Two DM devices, one to hold metadata and one to hold the
  43         * actual data/parity.  The reason for this is to not confuse
  44         * ti->len and give more flexibility in altering size and
  45         * characteristics.
  46         *
  47         * While it is possible for this device to be associated
  48         * with a different physical device than the data_dev, it
  49         * is intended for it to be the same.
  50         *    |--------- Physical Device ---------|
  51         *    |- meta_dev -|------ data_dev ------|
  52         */
  53        struct dm_dev *meta_dev;
  54        struct dm_dev *data_dev;
  55        struct md_rdev rdev;
  56};
  57
  58/*
  59 * Bits for establishing rs->ctr_flags
  60 *
  61 * 1 = no flag value
  62 * 2 = flag with value
  63 */
  64#define __CTR_FLAG_SYNC                 0  /* 1 */ /* Not with raid0! */
  65#define __CTR_FLAG_NOSYNC               1  /* 1 */ /* Not with raid0! */
  66#define __CTR_FLAG_REBUILD              2  /* 2 */ /* Not with raid0! */
  67#define __CTR_FLAG_DAEMON_SLEEP         3  /* 2 */ /* Not with raid0! */
  68#define __CTR_FLAG_MIN_RECOVERY_RATE    4  /* 2 */ /* Not with raid0! */
  69#define __CTR_FLAG_MAX_RECOVERY_RATE    5  /* 2 */ /* Not with raid0! */
  70#define __CTR_FLAG_MAX_WRITE_BEHIND     6  /* 2 */ /* Only with raid1! */
  71#define __CTR_FLAG_WRITE_MOSTLY         7  /* 2 */ /* Only with raid1! */
  72#define __CTR_FLAG_STRIPE_CACHE         8  /* 2 */ /* Only with raid4/5/6! */
  73#define __CTR_FLAG_REGION_SIZE          9  /* 2 */ /* Not with raid0! */
  74#define __CTR_FLAG_RAID10_COPIES        10 /* 2 */ /* Only with raid10 */
  75#define __CTR_FLAG_RAID10_FORMAT        11 /* 2 */ /* Only with raid10 */
  76/* New for v1.9.0 */
  77#define __CTR_FLAG_DELTA_DISKS          12 /* 2 */ /* Only with reshapable raid1/4/5/6/10! */
  78#define __CTR_FLAG_DATA_OFFSET          13 /* 2 */ /* Only with reshapable raid4/5/6/10! */
  79#define __CTR_FLAG_RAID10_USE_NEAR_SETS 14 /* 2 */ /* Only with raid10! */
  80
  81/* New for v1.10.0 */
  82#define __CTR_FLAG_JOURNAL_DEV          15 /* 2 */ /* Only with raid4/5/6 (journal device)! */
  83
  84/* New for v1.11.1 */
  85#define __CTR_FLAG_JOURNAL_MODE         16 /* 2 */ /* Only with raid4/5/6 (journal mode)! */
  86
  87/*
  88 * Flags for rs->ctr_flags field.
  89 */
  90#define CTR_FLAG_SYNC                   (1 << __CTR_FLAG_SYNC)
  91#define CTR_FLAG_NOSYNC                 (1 << __CTR_FLAG_NOSYNC)
  92#define CTR_FLAG_REBUILD                (1 << __CTR_FLAG_REBUILD)
  93#define CTR_FLAG_DAEMON_SLEEP           (1 << __CTR_FLAG_DAEMON_SLEEP)
  94#define CTR_FLAG_MIN_RECOVERY_RATE      (1 << __CTR_FLAG_MIN_RECOVERY_RATE)
  95#define CTR_FLAG_MAX_RECOVERY_RATE      (1 << __CTR_FLAG_MAX_RECOVERY_RATE)
  96#define CTR_FLAG_MAX_WRITE_BEHIND       (1 << __CTR_FLAG_MAX_WRITE_BEHIND)
  97#define CTR_FLAG_WRITE_MOSTLY           (1 << __CTR_FLAG_WRITE_MOSTLY)
  98#define CTR_FLAG_STRIPE_CACHE           (1 << __CTR_FLAG_STRIPE_CACHE)
  99#define CTR_FLAG_REGION_SIZE            (1 << __CTR_FLAG_REGION_SIZE)
 100#define CTR_FLAG_RAID10_COPIES          (1 << __CTR_FLAG_RAID10_COPIES)
 101#define CTR_FLAG_RAID10_FORMAT          (1 << __CTR_FLAG_RAID10_FORMAT)
 102#define CTR_FLAG_DELTA_DISKS            (1 << __CTR_FLAG_DELTA_DISKS)
 103#define CTR_FLAG_DATA_OFFSET            (1 << __CTR_FLAG_DATA_OFFSET)
 104#define CTR_FLAG_RAID10_USE_NEAR_SETS   (1 << __CTR_FLAG_RAID10_USE_NEAR_SETS)
 105#define CTR_FLAG_JOURNAL_DEV            (1 << __CTR_FLAG_JOURNAL_DEV)
 106#define CTR_FLAG_JOURNAL_MODE           (1 << __CTR_FLAG_JOURNAL_MODE)
 107
 108/*
 109 * Definitions of various constructor flags to
 110 * be used in checks of valid / invalid flags
 111 * per raid level.
 112 */
 113/* Define all any sync flags */
 114#define CTR_FLAGS_ANY_SYNC              (CTR_FLAG_SYNC | CTR_FLAG_NOSYNC)
 115
 116/* Define flags for options without argument (e.g. 'nosync') */
 117#define CTR_FLAG_OPTIONS_NO_ARGS        (CTR_FLAGS_ANY_SYNC | \
 118                                         CTR_FLAG_RAID10_USE_NEAR_SETS)
 119
 120/* Define flags for options with one argument (e.g. 'delta_disks +2') */
 121#define CTR_FLAG_OPTIONS_ONE_ARG (CTR_FLAG_REBUILD | \
 122                                  CTR_FLAG_WRITE_MOSTLY | \
 123                                  CTR_FLAG_DAEMON_SLEEP | \
 124                                  CTR_FLAG_MIN_RECOVERY_RATE | \
 125                                  CTR_FLAG_MAX_RECOVERY_RATE | \
 126                                  CTR_FLAG_MAX_WRITE_BEHIND | \
 127                                  CTR_FLAG_STRIPE_CACHE | \
 128                                  CTR_FLAG_REGION_SIZE | \
 129                                  CTR_FLAG_RAID10_COPIES | \
 130                                  CTR_FLAG_RAID10_FORMAT | \
 131                                  CTR_FLAG_DELTA_DISKS | \
 132                                  CTR_FLAG_DATA_OFFSET | \
 133                                  CTR_FLAG_JOURNAL_DEV | \
 134                                  CTR_FLAG_JOURNAL_MODE)
 135
 136/* Valid options definitions per raid level... */
 137
 138/* "raid0" does only accept data offset */
 139#define RAID0_VALID_FLAGS       (CTR_FLAG_DATA_OFFSET)
 140
 141/* "raid1" does not accept stripe cache, data offset, delta_disks or any raid10 options */
 142#define RAID1_VALID_FLAGS       (CTR_FLAGS_ANY_SYNC | \
 143                                 CTR_FLAG_REBUILD | \
 144                                 CTR_FLAG_WRITE_MOSTLY | \
 145                                 CTR_FLAG_DAEMON_SLEEP | \
 146                                 CTR_FLAG_MIN_RECOVERY_RATE | \
 147                                 CTR_FLAG_MAX_RECOVERY_RATE | \
 148                                 CTR_FLAG_MAX_WRITE_BEHIND | \
 149                                 CTR_FLAG_REGION_SIZE | \
 150                                 CTR_FLAG_DELTA_DISKS | \
 151                                 CTR_FLAG_DATA_OFFSET)
 152
 153/* "raid10" does not accept any raid1 or stripe cache options */
 154#define RAID10_VALID_FLAGS      (CTR_FLAGS_ANY_SYNC | \
 155                                 CTR_FLAG_REBUILD | \
 156                                 CTR_FLAG_DAEMON_SLEEP | \
 157                                 CTR_FLAG_MIN_RECOVERY_RATE | \
 158                                 CTR_FLAG_MAX_RECOVERY_RATE | \
 159                                 CTR_FLAG_REGION_SIZE | \
 160                                 CTR_FLAG_RAID10_COPIES | \
 161                                 CTR_FLAG_RAID10_FORMAT | \
 162                                 CTR_FLAG_DELTA_DISKS | \
 163                                 CTR_FLAG_DATA_OFFSET | \
 164                                 CTR_FLAG_RAID10_USE_NEAR_SETS)
 165
 166/*
 167 * "raid4/5/6" do not accept any raid1 or raid10 specific options
 168 *
 169 * "raid6" does not accept "nosync", because it is not guaranteed
 170 * that both parity and q-syndrome are being written properly with
 171 * any writes
 172 */
 173#define RAID45_VALID_FLAGS      (CTR_FLAGS_ANY_SYNC | \
 174                                 CTR_FLAG_REBUILD | \
 175                                 CTR_FLAG_DAEMON_SLEEP | \
 176                                 CTR_FLAG_MIN_RECOVERY_RATE | \
 177                                 CTR_FLAG_MAX_RECOVERY_RATE | \
 178                                 CTR_FLAG_STRIPE_CACHE | \
 179                                 CTR_FLAG_REGION_SIZE | \
 180                                 CTR_FLAG_DELTA_DISKS | \
 181                                 CTR_FLAG_DATA_OFFSET | \
 182                                 CTR_FLAG_JOURNAL_DEV | \
 183                                 CTR_FLAG_JOURNAL_MODE)
 184
 185#define RAID6_VALID_FLAGS       (CTR_FLAG_SYNC | \
 186                                 CTR_FLAG_REBUILD | \
 187                                 CTR_FLAG_DAEMON_SLEEP | \
 188                                 CTR_FLAG_MIN_RECOVERY_RATE | \
 189                                 CTR_FLAG_MAX_RECOVERY_RATE | \
 190                                 CTR_FLAG_STRIPE_CACHE | \
 191                                 CTR_FLAG_REGION_SIZE | \
 192                                 CTR_FLAG_DELTA_DISKS | \
 193                                 CTR_FLAG_DATA_OFFSET | \
 194                                 CTR_FLAG_JOURNAL_DEV | \
 195                                 CTR_FLAG_JOURNAL_MODE)
 196/* ...valid options definitions per raid level */
 197
 198/*
 199 * Flags for rs->runtime_flags field
 200 * (RT_FLAG prefix meaning "runtime flag")
 201 *
 202 * These are all internal and used to define runtime state,
 203 * e.g. to prevent another resume from preresume processing
 204 * the raid set all over again.
 205 */
 206#define RT_FLAG_RS_PRERESUMED           0
 207#define RT_FLAG_RS_RESUMED              1
 208#define RT_FLAG_RS_BITMAP_LOADED        2
 209#define RT_FLAG_UPDATE_SBS              3
 210#define RT_FLAG_RESHAPE_RS              4
 211#define RT_FLAG_RS_SUSPENDED            5
 212#define RT_FLAG_RS_IN_SYNC              6
 213#define RT_FLAG_RS_RESYNCING            7
 214#define RT_FLAG_RS_GROW                 8
 215
 216/* Array elements of 64 bit needed for rebuild/failed disk bits */
 217#define DISKS_ARRAY_ELEMS ((MAX_RAID_DEVICES + (sizeof(uint64_t) * 8 - 1)) / sizeof(uint64_t) / 8)
 218
 219/*
 220 * raid set level, layout and chunk sectors backup/restore
 221 */
 222struct rs_layout {
 223        int new_level;
 224        int new_layout;
 225        int new_chunk_sectors;
 226};
 227
 228struct raid_set {
 229        struct dm_target *ti;
 230
 231        uint32_t stripe_cache_entries;
 232        unsigned long ctr_flags;
 233        unsigned long runtime_flags;
 234
 235        uint64_t rebuild_disks[DISKS_ARRAY_ELEMS];
 236
 237        int raid_disks;
 238        int delta_disks;
 239        int data_offset;
 240        int raid10_copies;
 241        int requested_bitmap_chunk_sectors;
 242
 243        struct mddev md;
 244        struct raid_type *raid_type;
 245
 246        sector_t array_sectors;
 247        sector_t dev_sectors;
 248
 249        /* Optional raid4/5/6 journal device */
 250        struct journal_dev {
 251                struct dm_dev *dev;
 252                struct md_rdev rdev;
 253                int mode;
 254        } journal_dev;
 255
 256        struct raid_dev dev[];
 257};
 258
 259static void rs_config_backup(struct raid_set *rs, struct rs_layout *l)
 260{
 261        struct mddev *mddev = &rs->md;
 262
 263        l->new_level = mddev->new_level;
 264        l->new_layout = mddev->new_layout;
 265        l->new_chunk_sectors = mddev->new_chunk_sectors;
 266}
 267
 268static void rs_config_restore(struct raid_set *rs, struct rs_layout *l)
 269{
 270        struct mddev *mddev = &rs->md;
 271
 272        mddev->new_level = l->new_level;
 273        mddev->new_layout = l->new_layout;
 274        mddev->new_chunk_sectors = l->new_chunk_sectors;
 275}
 276
 277/* raid10 algorithms (i.e. formats) */
 278#define ALGORITHM_RAID10_DEFAULT        0
 279#define ALGORITHM_RAID10_NEAR           1
 280#define ALGORITHM_RAID10_OFFSET         2
 281#define ALGORITHM_RAID10_FAR            3
 282
 283/* Supported raid types and properties. */
 284static struct raid_type {
 285        const char *name;               /* RAID algorithm. */
 286        const char *descr;              /* Descriptor text for logging. */
 287        const unsigned int parity_devs; /* # of parity devices. */
 288        const unsigned int minimal_devs;/* minimal # of devices in set. */
 289        const unsigned int level;       /* RAID level. */
 290        const unsigned int algorithm;   /* RAID algorithm. */
 291} raid_types[] = {
 292        {"raid0",         "raid0 (striping)",                       0, 2, 0,  0 /* NONE */},
 293        {"raid1",         "raid1 (mirroring)",                      0, 2, 1,  0 /* NONE */},
 294        {"raid10_far",    "raid10 far (striped mirrors)",           0, 2, 10, ALGORITHM_RAID10_FAR},
 295        {"raid10_offset", "raid10 offset (striped mirrors)",        0, 2, 10, ALGORITHM_RAID10_OFFSET},
 296        {"raid10_near",   "raid10 near (striped mirrors)",          0, 2, 10, ALGORITHM_RAID10_NEAR},
 297        {"raid10",        "raid10 (striped mirrors)",               0, 2, 10, ALGORITHM_RAID10_DEFAULT},
 298        {"raid4",         "raid4 (dedicated first parity disk)",    1, 2, 5,  ALGORITHM_PARITY_0}, /* raid4 layout = raid5_0 */
 299        {"raid5_n",       "raid5 (dedicated last parity disk)",     1, 2, 5,  ALGORITHM_PARITY_N},
 300        {"raid5_ls",      "raid5 (left symmetric)",                 1, 2, 5,  ALGORITHM_LEFT_SYMMETRIC},
 301        {"raid5_rs",      "raid5 (right symmetric)",                1, 2, 5,  ALGORITHM_RIGHT_SYMMETRIC},
 302        {"raid5_la",      "raid5 (left asymmetric)",                1, 2, 5,  ALGORITHM_LEFT_ASYMMETRIC},
 303        {"raid5_ra",      "raid5 (right asymmetric)",               1, 2, 5,  ALGORITHM_RIGHT_ASYMMETRIC},
 304        {"raid6_zr",      "raid6 (zero restart)",                   2, 4, 6,  ALGORITHM_ROTATING_ZERO_RESTART},
 305        {"raid6_nr",      "raid6 (N restart)",                      2, 4, 6,  ALGORITHM_ROTATING_N_RESTART},
 306        {"raid6_nc",      "raid6 (N continue)",                     2, 4, 6,  ALGORITHM_ROTATING_N_CONTINUE},
 307        {"raid6_n_6",     "raid6 (dedicated parity/Q n/6)",         2, 4, 6,  ALGORITHM_PARITY_N_6},
 308        {"raid6_ls_6",    "raid6 (left symmetric dedicated Q 6)",   2, 4, 6,  ALGORITHM_LEFT_SYMMETRIC_6},
 309        {"raid6_rs_6",    "raid6 (right symmetric dedicated Q 6)",  2, 4, 6,  ALGORITHM_RIGHT_SYMMETRIC_6},
 310        {"raid6_la_6",    "raid6 (left asymmetric dedicated Q 6)",  2, 4, 6,  ALGORITHM_LEFT_ASYMMETRIC_6},
 311        {"raid6_ra_6",    "raid6 (right asymmetric dedicated Q 6)", 2, 4, 6,  ALGORITHM_RIGHT_ASYMMETRIC_6}
 312};
 313
 314/* True, if @v is in inclusive range [@min, @max] */
 315static bool __within_range(long v, long min, long max)
 316{
 317        return v >= min && v <= max;
 318}
 319
 320/* All table line arguments are defined here */
 321static struct arg_name_flag {
 322        const unsigned long flag;
 323        const char *name;
 324} __arg_name_flags[] = {
 325        { CTR_FLAG_SYNC, "sync"},
 326        { CTR_FLAG_NOSYNC, "nosync"},
 327        { CTR_FLAG_REBUILD, "rebuild"},
 328        { CTR_FLAG_DAEMON_SLEEP, "daemon_sleep"},
 329        { CTR_FLAG_MIN_RECOVERY_RATE, "min_recovery_rate"},
 330        { CTR_FLAG_MAX_RECOVERY_RATE, "max_recovery_rate"},
 331        { CTR_FLAG_MAX_WRITE_BEHIND, "max_write_behind"},
 332        { CTR_FLAG_WRITE_MOSTLY, "write_mostly"},
 333        { CTR_FLAG_STRIPE_CACHE, "stripe_cache"},
 334        { CTR_FLAG_REGION_SIZE, "region_size"},
 335        { CTR_FLAG_RAID10_COPIES, "raid10_copies"},
 336        { CTR_FLAG_RAID10_FORMAT, "raid10_format"},
 337        { CTR_FLAG_DATA_OFFSET, "data_offset"},
 338        { CTR_FLAG_DELTA_DISKS, "delta_disks"},
 339        { CTR_FLAG_RAID10_USE_NEAR_SETS, "raid10_use_near_sets"},
 340        { CTR_FLAG_JOURNAL_DEV, "journal_dev" },
 341        { CTR_FLAG_JOURNAL_MODE, "journal_mode" },
 342};
 343
 344/* Return argument name string for given @flag */
 345static const char *dm_raid_arg_name_by_flag(const uint32_t flag)
 346{
 347        if (hweight32(flag) == 1) {
 348                struct arg_name_flag *anf = __arg_name_flags + ARRAY_SIZE(__arg_name_flags);
 349
 350                while (anf-- > __arg_name_flags)
 351                        if (flag & anf->flag)
 352                                return anf->name;
 353
 354        } else
 355                DMERR("%s called with more than one flag!", __func__);
 356
 357        return NULL;
 358}
 359
 360/* Define correlation of raid456 journal cache modes and dm-raid target line parameters */
 361static struct {
 362        const int mode;
 363        const char *param;
 364} _raid456_journal_mode[] = {
 365        { R5C_JOURNAL_MODE_WRITE_THROUGH , "writethrough" },
 366        { R5C_JOURNAL_MODE_WRITE_BACK    , "writeback" }
 367};
 368
 369/* Return MD raid4/5/6 journal mode for dm @journal_mode one */
 370static int dm_raid_journal_mode_to_md(const char *mode)
 371{
 372        int m = ARRAY_SIZE(_raid456_journal_mode);
 373
 374        while (m--)
 375                if (!strcasecmp(mode, _raid456_journal_mode[m].param))
 376                        return _raid456_journal_mode[m].mode;
 377
 378        return -EINVAL;
 379}
 380
 381/* Return dm-raid raid4/5/6 journal mode string for @mode */
 382static const char *md_journal_mode_to_dm_raid(const int mode)
 383{
 384        int m = ARRAY_SIZE(_raid456_journal_mode);
 385
 386        while (m--)
 387                if (mode == _raid456_journal_mode[m].mode)
 388                        return _raid456_journal_mode[m].param;
 389
 390        return "unknown";
 391}
 392
 393/*
 394 * Bool helpers to test for various raid levels of a raid set.
 395 * It's level as reported by the superblock rather than
 396 * the requested raid_type passed to the constructor.
 397 */
 398/* Return true, if raid set in @rs is raid0 */
 399static bool rs_is_raid0(struct raid_set *rs)
 400{
 401        return !rs->md.level;
 402}
 403
 404/* Return true, if raid set in @rs is raid1 */
 405static bool rs_is_raid1(struct raid_set *rs)
 406{
 407        return rs->md.level == 1;
 408}
 409
 410/* Return true, if raid set in @rs is raid10 */
 411static bool rs_is_raid10(struct raid_set *rs)
 412{
 413        return rs->md.level == 10;
 414}
 415
 416/* Return true, if raid set in @rs is level 6 */
 417static bool rs_is_raid6(struct raid_set *rs)
 418{
 419        return rs->md.level == 6;
 420}
 421
 422/* Return true, if raid set in @rs is level 4, 5 or 6 */
 423static bool rs_is_raid456(struct raid_set *rs)
 424{
 425        return __within_range(rs->md.level, 4, 6);
 426}
 427
 428/* Return true, if raid set in @rs is reshapable */
 429static bool __is_raid10_far(int layout);
 430static bool rs_is_reshapable(struct raid_set *rs)
 431{
 432        return rs_is_raid456(rs) ||
 433               (rs_is_raid10(rs) && !__is_raid10_far(rs->md.new_layout));
 434}
 435
 436/* Return true, if raid set in @rs is recovering */
 437static bool rs_is_recovering(struct raid_set *rs)
 438{
 439        return rs->md.recovery_cp < rs->md.dev_sectors;
 440}
 441
 442/* Return true, if raid set in @rs is reshaping */
 443static bool rs_is_reshaping(struct raid_set *rs)
 444{
 445        return rs->md.reshape_position != MaxSector;
 446}
 447
 448/*
 449 * bool helpers to test for various raid levels of a raid type @rt
 450 */
 451
 452/* Return true, if raid type in @rt is raid0 */
 453static bool rt_is_raid0(struct raid_type *rt)
 454{
 455        return !rt->level;
 456}
 457
 458/* Return true, if raid type in @rt is raid1 */
 459static bool rt_is_raid1(struct raid_type *rt)
 460{
 461        return rt->level == 1;
 462}
 463
 464/* Return true, if raid type in @rt is raid10 */
 465static bool rt_is_raid10(struct raid_type *rt)
 466{
 467        return rt->level == 10;
 468}
 469
 470/* Return true, if raid type in @rt is raid4/5 */
 471static bool rt_is_raid45(struct raid_type *rt)
 472{
 473        return __within_range(rt->level, 4, 5);
 474}
 475
 476/* Return true, if raid type in @rt is raid6 */
 477static bool rt_is_raid6(struct raid_type *rt)
 478{
 479        return rt->level == 6;
 480}
 481
 482/* Return true, if raid type in @rt is raid4/5/6 */
 483static bool rt_is_raid456(struct raid_type *rt)
 484{
 485        return __within_range(rt->level, 4, 6);
 486}
 487/* END: raid level bools */
 488
 489/* Return valid ctr flags for the raid level of @rs */
 490static unsigned long __valid_flags(struct raid_set *rs)
 491{
 492        if (rt_is_raid0(rs->raid_type))
 493                return RAID0_VALID_FLAGS;
 494        else if (rt_is_raid1(rs->raid_type))
 495                return RAID1_VALID_FLAGS;
 496        else if (rt_is_raid10(rs->raid_type))
 497                return RAID10_VALID_FLAGS;
 498        else if (rt_is_raid45(rs->raid_type))
 499                return RAID45_VALID_FLAGS;
 500        else if (rt_is_raid6(rs->raid_type))
 501                return RAID6_VALID_FLAGS;
 502
 503        return 0;
 504}
 505
 506/*
 507 * Check for valid flags set on @rs
 508 *
 509 * Has to be called after parsing of the ctr flags!
 510 */
 511static int rs_check_for_valid_flags(struct raid_set *rs)
 512{
 513        if (rs->ctr_flags & ~__valid_flags(rs)) {
 514                rs->ti->error = "Invalid flags combination";
 515                return -EINVAL;
 516        }
 517
 518        return 0;
 519}
 520
 521/* MD raid10 bit definitions and helpers */
 522#define RAID10_OFFSET                   (1 << 16) /* stripes with data copies area adjacent on devices */
 523#define RAID10_BROCKEN_USE_FAR_SETS     (1 << 17) /* Broken in raid10.c: use sets instead of whole stripe rotation */
 524#define RAID10_USE_FAR_SETS             (1 << 18) /* Use sets instead of whole stripe rotation */
 525#define RAID10_FAR_COPIES_SHIFT         8         /* raid10 # far copies shift (2nd byte of layout) */
 526
 527/* Return md raid10 near copies for @layout */
 528static unsigned int __raid10_near_copies(int layout)
 529{
 530        return layout & 0xFF;
 531}
 532
 533/* Return md raid10 far copies for @layout */
 534static unsigned int __raid10_far_copies(int layout)
 535{
 536        return __raid10_near_copies(layout >> RAID10_FAR_COPIES_SHIFT);
 537}
 538
 539/* Return true if md raid10 offset for @layout */
 540static bool __is_raid10_offset(int layout)
 541{
 542        return !!(layout & RAID10_OFFSET);
 543}
 544
 545/* Return true if md raid10 near for @layout */
 546static bool __is_raid10_near(int layout)
 547{
 548        return !__is_raid10_offset(layout) && __raid10_near_copies(layout) > 1;
 549}
 550
 551/* Return true if md raid10 far for @layout */
 552static bool __is_raid10_far(int layout)
 553{
 554        return !__is_raid10_offset(layout) && __raid10_far_copies(layout) > 1;
 555}
 556
 557/* Return md raid10 layout string for @layout */
 558static const char *raid10_md_layout_to_format(int layout)
 559{
 560        /*
 561         * Bit 16 stands for "offset"
 562         * (i.e. adjacent stripes hold copies)
 563         *
 564         * Refer to MD's raid10.c for details
 565         */
 566        if (__is_raid10_offset(layout))
 567                return "offset";
 568
 569        if (__raid10_near_copies(layout) > 1)
 570                return "near";
 571
 572        if (__raid10_far_copies(layout) > 1)
 573                return "far";
 574
 575        return "unknown";
 576}
 577
 578/* Return md raid10 algorithm for @name */
 579static int raid10_name_to_format(const char *name)
 580{
 581        if (!strcasecmp(name, "near"))
 582                return ALGORITHM_RAID10_NEAR;
 583        else if (!strcasecmp(name, "offset"))
 584                return ALGORITHM_RAID10_OFFSET;
 585        else if (!strcasecmp(name, "far"))
 586                return ALGORITHM_RAID10_FAR;
 587
 588        return -EINVAL;
 589}
 590
 591/* Return md raid10 copies for @layout */
 592static unsigned int raid10_md_layout_to_copies(int layout)
 593{
 594        return max(__raid10_near_copies(layout), __raid10_far_copies(layout));
 595}
 596
 597/* Return md raid10 format id for @format string */
 598static int raid10_format_to_md_layout(struct raid_set *rs,
 599                                      unsigned int algorithm,
 600                                      unsigned int copies)
 601{
 602        unsigned int n = 1, f = 1, r = 0;
 603
 604        /*
 605         * MD resilienece flaw:
 606         *
 607         * enabling use_far_sets for far/offset formats causes copies
 608         * to be colocated on the same devs together with their origins!
 609         *
 610         * -> disable it for now in the definition above
 611         */
 612        if (algorithm == ALGORITHM_RAID10_DEFAULT ||
 613            algorithm == ALGORITHM_RAID10_NEAR)
 614                n = copies;
 615
 616        else if (algorithm == ALGORITHM_RAID10_OFFSET) {
 617                f = copies;
 618                r = RAID10_OFFSET;
 619                if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
 620                        r |= RAID10_USE_FAR_SETS;
 621
 622        } else if (algorithm == ALGORITHM_RAID10_FAR) {
 623                f = copies;
 624                if (!test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags))
 625                        r |= RAID10_USE_FAR_SETS;
 626
 627        } else
 628                return -EINVAL;
 629
 630        return r | (f << RAID10_FAR_COPIES_SHIFT) | n;
 631}
 632/* END: MD raid10 bit definitions and helpers */
 633
 634/* Check for any of the raid10 algorithms */
 635static bool __got_raid10(struct raid_type *rtp, const int layout)
 636{
 637        if (rtp->level == 10) {
 638                switch (rtp->algorithm) {
 639                case ALGORITHM_RAID10_DEFAULT:
 640                case ALGORITHM_RAID10_NEAR:
 641                        return __is_raid10_near(layout);
 642                case ALGORITHM_RAID10_OFFSET:
 643                        return __is_raid10_offset(layout);
 644                case ALGORITHM_RAID10_FAR:
 645                        return __is_raid10_far(layout);
 646                default:
 647                        break;
 648                }
 649        }
 650
 651        return false;
 652}
 653
 654/* Return raid_type for @name */
 655static struct raid_type *get_raid_type(const char *name)
 656{
 657        struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
 658
 659        while (rtp-- > raid_types)
 660                if (!strcasecmp(rtp->name, name))
 661                        return rtp;
 662
 663        return NULL;
 664}
 665
 666/* Return raid_type for @name based derived from @level and @layout */
 667static struct raid_type *get_raid_type_by_ll(const int level, const int layout)
 668{
 669        struct raid_type *rtp = raid_types + ARRAY_SIZE(raid_types);
 670
 671        while (rtp-- > raid_types) {
 672                /* RAID10 special checks based on @layout flags/properties */
 673                if (rtp->level == level &&
 674                    (__got_raid10(rtp, layout) || rtp->algorithm == layout))
 675                        return rtp;
 676        }
 677
 678        return NULL;
 679}
 680
 681/* Adjust rdev sectors */
 682static void rs_set_rdev_sectors(struct raid_set *rs)
 683{
 684        struct mddev *mddev = &rs->md;
 685        struct md_rdev *rdev;
 686
 687        /*
 688         * raid10 sets rdev->sector to the device size, which
 689         * is unintended in case of out-of-place reshaping
 690         */
 691        rdev_for_each(rdev, mddev)
 692                if (!test_bit(Journal, &rdev->flags))
 693                        rdev->sectors = mddev->dev_sectors;
 694}
 695
 696/*
 697 * Change bdev capacity of @rs in case of a disk add/remove reshape
 698 */
 699static void rs_set_capacity(struct raid_set *rs)
 700{
 701        struct gendisk *gendisk = dm_disk(dm_table_get_md(rs->ti->table));
 702
 703        set_capacity_and_notify(gendisk, rs->md.array_sectors);
 704}
 705
 706/*
 707 * Set the mddev properties in @rs to the current
 708 * ones retrieved from the freshest superblock
 709 */
 710static void rs_set_cur(struct raid_set *rs)
 711{
 712        struct mddev *mddev = &rs->md;
 713
 714        mddev->new_level = mddev->level;
 715        mddev->new_layout = mddev->layout;
 716        mddev->new_chunk_sectors = mddev->chunk_sectors;
 717}
 718
 719/*
 720 * Set the mddev properties in @rs to the new
 721 * ones requested by the ctr
 722 */
 723static void rs_set_new(struct raid_set *rs)
 724{
 725        struct mddev *mddev = &rs->md;
 726
 727        mddev->level = mddev->new_level;
 728        mddev->layout = mddev->new_layout;
 729        mddev->chunk_sectors = mddev->new_chunk_sectors;
 730        mddev->raid_disks = rs->raid_disks;
 731        mddev->delta_disks = 0;
 732}
 733
 734static struct raid_set *raid_set_alloc(struct dm_target *ti, struct raid_type *raid_type,
 735                                       unsigned int raid_devs)
 736{
 737        unsigned int i;
 738        struct raid_set *rs;
 739
 740        if (raid_devs <= raid_type->parity_devs) {
 741                ti->error = "Insufficient number of devices";
 742                return ERR_PTR(-EINVAL);
 743        }
 744
 745        rs = kzalloc(struct_size(rs, dev, raid_devs), GFP_KERNEL);
 746        if (!rs) {
 747                ti->error = "Cannot allocate raid context";
 748                return ERR_PTR(-ENOMEM);
 749        }
 750
 751        mddev_init(&rs->md);
 752
 753        rs->raid_disks = raid_devs;
 754        rs->delta_disks = 0;
 755
 756        rs->ti = ti;
 757        rs->raid_type = raid_type;
 758        rs->stripe_cache_entries = 256;
 759        rs->md.raid_disks = raid_devs;
 760        rs->md.level = raid_type->level;
 761        rs->md.new_level = rs->md.level;
 762        rs->md.layout = raid_type->algorithm;
 763        rs->md.new_layout = rs->md.layout;
 764        rs->md.delta_disks = 0;
 765        rs->md.recovery_cp = MaxSector;
 766
 767        for (i = 0; i < raid_devs; i++)
 768                md_rdev_init(&rs->dev[i].rdev);
 769
 770        /*
 771         * Remaining items to be initialized by further RAID params:
 772         *  rs->md.persistent
 773         *  rs->md.external
 774         *  rs->md.chunk_sectors
 775         *  rs->md.new_chunk_sectors
 776         *  rs->md.dev_sectors
 777         */
 778
 779        return rs;
 780}
 781
 782/* Free all @rs allocations */
 783static void raid_set_free(struct raid_set *rs)
 784{
 785        int i;
 786
 787        if (rs->journal_dev.dev) {
 788                md_rdev_clear(&rs->journal_dev.rdev);
 789                dm_put_device(rs->ti, rs->journal_dev.dev);
 790        }
 791
 792        for (i = 0; i < rs->raid_disks; i++) {
 793                if (rs->dev[i].meta_dev)
 794                        dm_put_device(rs->ti, rs->dev[i].meta_dev);
 795                md_rdev_clear(&rs->dev[i].rdev);
 796                if (rs->dev[i].data_dev)
 797                        dm_put_device(rs->ti, rs->dev[i].data_dev);
 798        }
 799
 800        kfree(rs);
 801}
 802
 803/*
 804 * For every device we have two words
 805 *  <meta_dev>: meta device name or '-' if missing
 806 *  <data_dev>: data device name or '-' if missing
 807 *
 808 * The following are permitted:
 809 *    - -
 810 *    - <data_dev>
 811 *    <meta_dev> <data_dev>
 812 *
 813 * The following is not allowed:
 814 *    <meta_dev> -
 815 *
 816 * This code parses those words.  If there is a failure,
 817 * the caller must use raid_set_free() to unwind the operations.
 818 */
 819static int parse_dev_params(struct raid_set *rs, struct dm_arg_set *as)
 820{
 821        int i;
 822        int rebuild = 0;
 823        int metadata_available = 0;
 824        int r = 0;
 825        const char *arg;
 826
 827        /* Put off the number of raid devices argument to get to dev pairs */
 828        arg = dm_shift_arg(as);
 829        if (!arg)
 830                return -EINVAL;
 831
 832        for (i = 0; i < rs->raid_disks; i++) {
 833                rs->dev[i].rdev.raid_disk = i;
 834
 835                rs->dev[i].meta_dev = NULL;
 836                rs->dev[i].data_dev = NULL;
 837
 838                /*
 839                 * There are no offsets initially.
 840                 * Out of place reshape will set them accordingly.
 841                 */
 842                rs->dev[i].rdev.data_offset = 0;
 843                rs->dev[i].rdev.new_data_offset = 0;
 844                rs->dev[i].rdev.mddev = &rs->md;
 845
 846                arg = dm_shift_arg(as);
 847                if (!arg)
 848                        return -EINVAL;
 849
 850                if (strcmp(arg, "-")) {
 851                        r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
 852                                          &rs->dev[i].meta_dev);
 853                        if (r) {
 854                                rs->ti->error = "RAID metadata device lookup failure";
 855                                return r;
 856                        }
 857
 858                        rs->dev[i].rdev.sb_page = alloc_page(GFP_KERNEL);
 859                        if (!rs->dev[i].rdev.sb_page) {
 860                                rs->ti->error = "Failed to allocate superblock page";
 861                                return -ENOMEM;
 862                        }
 863                }
 864
 865                arg = dm_shift_arg(as);
 866                if (!arg)
 867                        return -EINVAL;
 868
 869                if (!strcmp(arg, "-")) {
 870                        if (!test_bit(In_sync, &rs->dev[i].rdev.flags) &&
 871                            (!rs->dev[i].rdev.recovery_offset)) {
 872                                rs->ti->error = "Drive designated for rebuild not specified";
 873                                return -EINVAL;
 874                        }
 875
 876                        if (rs->dev[i].meta_dev) {
 877                                rs->ti->error = "No data device supplied with metadata device";
 878                                return -EINVAL;
 879                        }
 880
 881                        continue;
 882                }
 883
 884                r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
 885                                  &rs->dev[i].data_dev);
 886                if (r) {
 887                        rs->ti->error = "RAID device lookup failure";
 888                        return r;
 889                }
 890
 891                if (rs->dev[i].meta_dev) {
 892                        metadata_available = 1;
 893                        rs->dev[i].rdev.meta_bdev = rs->dev[i].meta_dev->bdev;
 894                }
 895                rs->dev[i].rdev.bdev = rs->dev[i].data_dev->bdev;
 896                list_add_tail(&rs->dev[i].rdev.same_set, &rs->md.disks);
 897                if (!test_bit(In_sync, &rs->dev[i].rdev.flags))
 898                        rebuild++;
 899        }
 900
 901        if (rs->journal_dev.dev)
 902                list_add_tail(&rs->journal_dev.rdev.same_set, &rs->md.disks);
 903
 904        if (metadata_available) {
 905                rs->md.external = 0;
 906                rs->md.persistent = 1;
 907                rs->md.major_version = 2;
 908        } else if (rebuild && !rs->md.recovery_cp) {
 909                /*
 910                 * Without metadata, we will not be able to tell if the array
 911                 * is in-sync or not - we must assume it is not.  Therefore,
 912                 * it is impossible to rebuild a drive.
 913                 *
 914                 * Even if there is metadata, the on-disk information may
 915                 * indicate that the array is not in-sync and it will then
 916                 * fail at that time.
 917                 *
 918                 * User could specify 'nosync' option if desperate.
 919                 */
 920                rs->ti->error = "Unable to rebuild drive while array is not in-sync";
 921                return -EINVAL;
 922        }
 923
 924        return 0;
 925}
 926
 927/*
 928 * validate_region_size
 929 * @rs
 930 * @region_size:  region size in sectors.  If 0, pick a size (4MiB default).
 931 *
 932 * Set rs->md.bitmap_info.chunksize (which really refers to 'region size').
 933 * Ensure that (ti->len/region_size < 2^21) - required by MD bitmap.
 934 *
 935 * Returns: 0 on success, -EINVAL on failure.
 936 */
 937static int validate_region_size(struct raid_set *rs, unsigned long region_size)
 938{
 939        unsigned long min_region_size = rs->ti->len / (1 << 21);
 940
 941        if (rs_is_raid0(rs))
 942                return 0;
 943
 944        if (!region_size) {
 945                /*
 946                 * Choose a reasonable default.  All figures in sectors.
 947                 */
 948                if (min_region_size > (1 << 13)) {
 949                        /* If not a power of 2, make it the next power of 2 */
 950                        region_size = roundup_pow_of_two(min_region_size);
 951                        DMINFO("Choosing default region size of %lu sectors",
 952                               region_size);
 953                } else {
 954                        DMINFO("Choosing default region size of 4MiB");
 955                        region_size = 1 << 13; /* sectors */
 956                }
 957        } else {
 958                /*
 959                 * Validate user-supplied value.
 960                 */
 961                if (region_size > rs->ti->len) {
 962                        rs->ti->error = "Supplied region size is too large";
 963                        return -EINVAL;
 964                }
 965
 966                if (region_size < min_region_size) {
 967                        DMERR("Supplied region_size (%lu sectors) below minimum (%lu)",
 968                              region_size, min_region_size);
 969                        rs->ti->error = "Supplied region size is too small";
 970                        return -EINVAL;
 971                }
 972
 973                if (!is_power_of_2(region_size)) {
 974                        rs->ti->error = "Region size is not a power of 2";
 975                        return -EINVAL;
 976                }
 977
 978                if (region_size < rs->md.chunk_sectors) {
 979                        rs->ti->error = "Region size is smaller than the chunk size";
 980                        return -EINVAL;
 981                }
 982        }
 983
 984        /*
 985         * Convert sectors to bytes.
 986         */
 987        rs->md.bitmap_info.chunksize = to_bytes(region_size);
 988
 989        return 0;
 990}
 991
 992/*
 993 * validate_raid_redundancy
 994 * @rs
 995 *
 996 * Determine if there are enough devices in the array that haven't
 997 * failed (or are being rebuilt) to form a usable array.
 998 *
 999 * Returns: 0 on success, -EINVAL on failure.
1000 */
1001static int validate_raid_redundancy(struct raid_set *rs)
1002{
1003        unsigned int i, rebuild_cnt = 0;
1004        unsigned int rebuilds_per_group = 0, copies;
1005        unsigned int group_size, last_group_start;
1006
1007        for (i = 0; i < rs->md.raid_disks; i++)
1008                if (!test_bit(In_sync, &rs->dev[i].rdev.flags) ||
1009                    !rs->dev[i].rdev.sb_page)
1010                        rebuild_cnt++;
1011
1012        switch (rs->md.level) {
1013        case 0:
1014                break;
1015        case 1:
1016                if (rebuild_cnt >= rs->md.raid_disks)
1017                        goto too_many;
1018                break;
1019        case 4:
1020        case 5:
1021        case 6:
1022                if (rebuild_cnt > rs->raid_type->parity_devs)
1023                        goto too_many;
1024                break;
1025        case 10:
1026                copies = raid10_md_layout_to_copies(rs->md.new_layout);
1027                if (copies < 2) {
1028                        DMERR("Bogus raid10 data copies < 2!");
1029                        return -EINVAL;
1030                }
1031
1032                if (rebuild_cnt < copies)
1033                        break;
1034
1035                /*
1036                 * It is possible to have a higher rebuild count for RAID10,
1037                 * as long as the failed devices occur in different mirror
1038                 * groups (i.e. different stripes).
1039                 *
1040                 * When checking "near" format, make sure no adjacent devices
1041                 * have failed beyond what can be handled.  In addition to the
1042                 * simple case where the number of devices is a multiple of the
1043                 * number of copies, we must also handle cases where the number
1044                 * of devices is not a multiple of the number of copies.
1045                 * E.g.    dev1 dev2 dev3 dev4 dev5
1046                 *          A    A    B    B    C
1047                 *          C    D    D    E    E
1048                 */
1049                if (__is_raid10_near(rs->md.new_layout)) {
1050                        for (i = 0; i < rs->md.raid_disks; i++) {
1051                                if (!(i % copies))
1052                                        rebuilds_per_group = 0;
1053                                if ((!rs->dev[i].rdev.sb_page ||
1054                                    !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1055                                    (++rebuilds_per_group >= copies))
1056                                        goto too_many;
1057                        }
1058                        break;
1059                }
1060
1061                /*
1062                 * When checking "far" and "offset" formats, we need to ensure
1063                 * that the device that holds its copy is not also dead or
1064                 * being rebuilt.  (Note that "far" and "offset" formats only
1065                 * support two copies right now.  These formats also only ever
1066                 * use the 'use_far_sets' variant.)
1067                 *
1068                 * This check is somewhat complicated by the need to account
1069                 * for arrays that are not a multiple of (far) copies.  This
1070                 * results in the need to treat the last (potentially larger)
1071                 * set differently.
1072                 */
1073                group_size = (rs->md.raid_disks / copies);
1074                last_group_start = (rs->md.raid_disks / group_size) - 1;
1075                last_group_start *= group_size;
1076                for (i = 0; i < rs->md.raid_disks; i++) {
1077                        if (!(i % copies) && !(i > last_group_start))
1078                                rebuilds_per_group = 0;
1079                        if ((!rs->dev[i].rdev.sb_page ||
1080                             !test_bit(In_sync, &rs->dev[i].rdev.flags)) &&
1081                            (++rebuilds_per_group >= copies))
1082                                        goto too_many;
1083                }
1084                break;
1085        default:
1086                if (rebuild_cnt)
1087                        return -EINVAL;
1088        }
1089
1090        return 0;
1091
1092too_many:
1093        return -EINVAL;
1094}
1095
1096/*
1097 * Possible arguments are...
1098 *      <chunk_size> [optional_args]
1099 *
1100 * Argument definitions
1101 *    <chunk_size>                      The number of sectors per disk that
1102 *                                      will form the "stripe"
1103 *    [[no]sync]                        Force or prevent recovery of the
1104 *                                      entire array
1105 *    [rebuild <idx>]                   Rebuild the drive indicated by the index
1106 *    [daemon_sleep <ms>]               Time between bitmap daemon work to
1107 *                                      clear bits
1108 *    [min_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1109 *    [max_recovery_rate <kB/sec/disk>] Throttle RAID initialization
1110 *    [write_mostly <idx>]              Indicate a write mostly drive via index
1111 *    [max_write_behind <sectors>]      See '-write-behind=' (man mdadm)
1112 *    [stripe_cache <sectors>]          Stripe cache size for higher RAIDs
1113 *    [region_size <sectors>]           Defines granularity of bitmap
1114 *    [journal_dev <dev>]               raid4/5/6 journaling deviice
1115 *                                      (i.e. write hole closing log)
1116 *
1117 * RAID10-only options:
1118 *    [raid10_copies <# copies>]        Number of copies.  (Default: 2)
1119 *    [raid10_format <near|far|offset>] Layout algorithm.  (Default: near)
1120 */
1121static int parse_raid_params(struct raid_set *rs, struct dm_arg_set *as,
1122                             unsigned int num_raid_params)
1123{
1124        int value, raid10_format = ALGORITHM_RAID10_DEFAULT;
1125        unsigned int raid10_copies = 2;
1126        unsigned int i, write_mostly = 0;
1127        unsigned int region_size = 0;
1128        sector_t max_io_len;
1129        const char *arg, *key;
1130        struct raid_dev *rd;
1131        struct raid_type *rt = rs->raid_type;
1132
1133        arg = dm_shift_arg(as);
1134        num_raid_params--; /* Account for chunk_size argument */
1135
1136        if (kstrtoint(arg, 10, &value) < 0) {
1137                rs->ti->error = "Bad numerical argument given for chunk_size";
1138                return -EINVAL;
1139        }
1140
1141        /*
1142         * First, parse the in-order required arguments
1143         * "chunk_size" is the only argument of this type.
1144         */
1145        if (rt_is_raid1(rt)) {
1146                if (value)
1147                        DMERR("Ignoring chunk size parameter for RAID 1");
1148                value = 0;
1149        } else if (!is_power_of_2(value)) {
1150                rs->ti->error = "Chunk size must be a power of 2";
1151                return -EINVAL;
1152        } else if (value < 8) {
1153                rs->ti->error = "Chunk size value is too small";
1154                return -EINVAL;
1155        }
1156
1157        rs->md.new_chunk_sectors = rs->md.chunk_sectors = value;
1158
1159        /*
1160         * We set each individual device as In_sync with a completed
1161         * 'recovery_offset'.  If there has been a device failure or
1162         * replacement then one of the following cases applies:
1163         *
1164         *   1) User specifies 'rebuild'.
1165         *      - Device is reset when param is read.
1166         *   2) A new device is supplied.
1167         *      - No matching superblock found, resets device.
1168         *   3) Device failure was transient and returns on reload.
1169         *      - Failure noticed, resets device for bitmap replay.
1170         *   4) Device hadn't completed recovery after previous failure.
1171         *      - Superblock is read and overrides recovery_offset.
1172         *
1173         * What is found in the superblocks of the devices is always
1174         * authoritative, unless 'rebuild' or '[no]sync' was specified.
1175         */
1176        for (i = 0; i < rs->raid_disks; i++) {
1177                set_bit(In_sync, &rs->dev[i].rdev.flags);
1178                rs->dev[i].rdev.recovery_offset = MaxSector;
1179        }
1180
1181        /*
1182         * Second, parse the unordered optional arguments
1183         */
1184        for (i = 0; i < num_raid_params; i++) {
1185                key = dm_shift_arg(as);
1186                if (!key) {
1187                        rs->ti->error = "Not enough raid parameters given";
1188                        return -EINVAL;
1189                }
1190
1191                if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC))) {
1192                        if (test_and_set_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1193                                rs->ti->error = "Only one 'nosync' argument allowed";
1194                                return -EINVAL;
1195                        }
1196                        continue;
1197                }
1198                if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_SYNC))) {
1199                        if (test_and_set_bit(__CTR_FLAG_SYNC, &rs->ctr_flags)) {
1200                                rs->ti->error = "Only one 'sync' argument allowed";
1201                                return -EINVAL;
1202                        }
1203                        continue;
1204                }
1205                if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_USE_NEAR_SETS))) {
1206                        if (test_and_set_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1207                                rs->ti->error = "Only one 'raid10_use_new_sets' argument allowed";
1208                                return -EINVAL;
1209                        }
1210                        continue;
1211                }
1212
1213                arg = dm_shift_arg(as);
1214                i++; /* Account for the argument pairs */
1215                if (!arg) {
1216                        rs->ti->error = "Wrong number of raid parameters given";
1217                        return -EINVAL;
1218                }
1219
1220                /*
1221                 * Parameters that take a string value are checked here.
1222                 */
1223                /* "raid10_format {near|offset|far} */
1224                if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT))) {
1225                        if (test_and_set_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags)) {
1226                                rs->ti->error = "Only one 'raid10_format' argument pair allowed";
1227                                return -EINVAL;
1228                        }
1229                        if (!rt_is_raid10(rt)) {
1230                                rs->ti->error = "'raid10_format' is an invalid parameter for this RAID type";
1231                                return -EINVAL;
1232                        }
1233                        raid10_format = raid10_name_to_format(arg);
1234                        if (raid10_format < 0) {
1235                                rs->ti->error = "Invalid 'raid10_format' value given";
1236                                return raid10_format;
1237                        }
1238                        continue;
1239                }
1240
1241                /* "journal_dev <dev>" */
1242                if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV))) {
1243                        int r;
1244                        struct md_rdev *jdev;
1245
1246                        if (test_and_set_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1247                                rs->ti->error = "Only one raid4/5/6 set journaling device allowed";
1248                                return -EINVAL;
1249                        }
1250                        if (!rt_is_raid456(rt)) {
1251                                rs->ti->error = "'journal_dev' is an invalid parameter for this RAID type";
1252                                return -EINVAL;
1253                        }
1254                        r = dm_get_device(rs->ti, arg, dm_table_get_mode(rs->ti->table),
1255                                          &rs->journal_dev.dev);
1256                        if (r) {
1257                                rs->ti->error = "raid4/5/6 journal device lookup failure";
1258                                return r;
1259                        }
1260                        jdev = &rs->journal_dev.rdev;
1261                        md_rdev_init(jdev);
1262                        jdev->mddev = &rs->md;
1263                        jdev->bdev = rs->journal_dev.dev->bdev;
1264                        jdev->sectors = to_sector(i_size_read(jdev->bdev->bd_inode));
1265                        if (jdev->sectors < MIN_RAID456_JOURNAL_SPACE) {
1266                                rs->ti->error = "No space for raid4/5/6 journal";
1267                                return -ENOSPC;
1268                        }
1269                        rs->journal_dev.mode = R5C_JOURNAL_MODE_WRITE_THROUGH;
1270                        set_bit(Journal, &jdev->flags);
1271                        continue;
1272                }
1273
1274                /* "journal_mode <mode>" ("journal_dev" mandatory!) */
1275                if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE))) {
1276                        int r;
1277
1278                        if (!test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
1279                                rs->ti->error = "raid4/5/6 'journal_mode' is invalid without 'journal_dev'";
1280                                return -EINVAL;
1281                        }
1282                        if (test_and_set_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
1283                                rs->ti->error = "Only one raid4/5/6 'journal_mode' argument allowed";
1284                                return -EINVAL;
1285                        }
1286                        r = dm_raid_journal_mode_to_md(arg);
1287                        if (r < 0) {
1288                                rs->ti->error = "Invalid 'journal_mode' argument";
1289                                return r;
1290                        }
1291                        rs->journal_dev.mode = r;
1292                        continue;
1293                }
1294
1295                /*
1296                 * Parameters with number values from here on.
1297                 */
1298                if (kstrtoint(arg, 10, &value) < 0) {
1299                        rs->ti->error = "Bad numerical argument given in raid params";
1300                        return -EINVAL;
1301                }
1302
1303                if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD))) {
1304                        /*
1305                         * "rebuild" is being passed in by userspace to provide
1306                         * indexes of replaced devices and to set up additional
1307                         * devices on raid level takeover.
1308                         */
1309                        if (!__within_range(value, 0, rs->raid_disks - 1)) {
1310                                rs->ti->error = "Invalid rebuild index given";
1311                                return -EINVAL;
1312                        }
1313
1314                        if (test_and_set_bit(value, (void *) rs->rebuild_disks)) {
1315                                rs->ti->error = "rebuild for this index already given";
1316                                return -EINVAL;
1317                        }
1318
1319                        rd = rs->dev + value;
1320                        clear_bit(In_sync, &rd->rdev.flags);
1321                        clear_bit(Faulty, &rd->rdev.flags);
1322                        rd->rdev.recovery_offset = 0;
1323                        set_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags);
1324                } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY))) {
1325                        if (!rt_is_raid1(rt)) {
1326                                rs->ti->error = "write_mostly option is only valid for RAID1";
1327                                return -EINVAL;
1328                        }
1329
1330                        if (!__within_range(value, 0, rs->md.raid_disks - 1)) {
1331                                rs->ti->error = "Invalid write_mostly index given";
1332                                return -EINVAL;
1333                        }
1334
1335                        write_mostly++;
1336                        set_bit(WriteMostly, &rs->dev[value].rdev.flags);
1337                        set_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags);
1338                } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND))) {
1339                        if (!rt_is_raid1(rt)) {
1340                                rs->ti->error = "max_write_behind option is only valid for RAID1";
1341                                return -EINVAL;
1342                        }
1343
1344                        if (test_and_set_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags)) {
1345                                rs->ti->error = "Only one max_write_behind argument pair allowed";
1346                                return -EINVAL;
1347                        }
1348
1349                        /*
1350                         * In device-mapper, we specify things in sectors, but
1351                         * MD records this value in kB
1352                         */
1353                        if (value < 0 || value / 2 > COUNTER_MAX) {
1354                                rs->ti->error = "Max write-behind limit out of range";
1355                                return -EINVAL;
1356                        }
1357
1358                        rs->md.bitmap_info.max_write_behind = value / 2;
1359                } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP))) {
1360                        if (test_and_set_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags)) {
1361                                rs->ti->error = "Only one daemon_sleep argument pair allowed";
1362                                return -EINVAL;
1363                        }
1364                        if (value < 0) {
1365                                rs->ti->error = "daemon sleep period out of range";
1366                                return -EINVAL;
1367                        }
1368                        rs->md.bitmap_info.daemon_sleep = value;
1369                } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET))) {
1370                        /* Userspace passes new data_offset after having extended the the data image LV */
1371                        if (test_and_set_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
1372                                rs->ti->error = "Only one data_offset argument pair allowed";
1373                                return -EINVAL;
1374                        }
1375                        /* Ensure sensible data offset */
1376                        if (value < 0 ||
1377                            (value && (value < MIN_FREE_RESHAPE_SPACE || value % to_sector(PAGE_SIZE)))) {
1378                                rs->ti->error = "Bogus data_offset value";
1379                                return -EINVAL;
1380                        }
1381                        rs->data_offset = value;
1382                } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS))) {
1383                        /* Define the +/-# of disks to add to/remove from the given raid set */
1384                        if (test_and_set_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
1385                                rs->ti->error = "Only one delta_disks argument pair allowed";
1386                                return -EINVAL;
1387                        }
1388                        /* Ensure MAX_RAID_DEVICES and raid type minimal_devs! */
1389                        if (!__within_range(abs(value), 1, MAX_RAID_DEVICES - rt->minimal_devs)) {
1390                                rs->ti->error = "Too many delta_disk requested";
1391                                return -EINVAL;
1392                        }
1393
1394                        rs->delta_disks = value;
1395                } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE))) {
1396                        if (test_and_set_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags)) {
1397                                rs->ti->error = "Only one stripe_cache argument pair allowed";
1398                                return -EINVAL;
1399                        }
1400
1401                        if (!rt_is_raid456(rt)) {
1402                                rs->ti->error = "Inappropriate argument: stripe_cache";
1403                                return -EINVAL;
1404                        }
1405
1406                        if (value < 0) {
1407                                rs->ti->error = "Bogus stripe cache entries value";
1408                                return -EINVAL;
1409                        }
1410                        rs->stripe_cache_entries = value;
1411                } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE))) {
1412                        if (test_and_set_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags)) {
1413                                rs->ti->error = "Only one min_recovery_rate argument pair allowed";
1414                                return -EINVAL;
1415                        }
1416
1417                        if (value < 0) {
1418                                rs->ti->error = "min_recovery_rate out of range";
1419                                return -EINVAL;
1420                        }
1421                        rs->md.sync_speed_min = value;
1422                } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE))) {
1423                        if (test_and_set_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags)) {
1424                                rs->ti->error = "Only one max_recovery_rate argument pair allowed";
1425                                return -EINVAL;
1426                        }
1427
1428                        if (value < 0) {
1429                                rs->ti->error = "max_recovery_rate out of range";
1430                                return -EINVAL;
1431                        }
1432                        rs->md.sync_speed_max = value;
1433                } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE))) {
1434                        if (test_and_set_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags)) {
1435                                rs->ti->error = "Only one region_size argument pair allowed";
1436                                return -EINVAL;
1437                        }
1438
1439                        region_size = value;
1440                        rs->requested_bitmap_chunk_sectors = value;
1441                } else if (!strcasecmp(key, dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES))) {
1442                        if (test_and_set_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags)) {
1443                                rs->ti->error = "Only one raid10_copies argument pair allowed";
1444                                return -EINVAL;
1445                        }
1446
1447                        if (!__within_range(value, 2, rs->md.raid_disks)) {
1448                                rs->ti->error = "Bad value for 'raid10_copies'";
1449                                return -EINVAL;
1450                        }
1451
1452                        raid10_copies = value;
1453                } else {
1454                        DMERR("Unable to parse RAID parameter: %s", key);
1455                        rs->ti->error = "Unable to parse RAID parameter";
1456                        return -EINVAL;
1457                }
1458        }
1459
1460        if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) &&
1461            test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
1462                rs->ti->error = "sync and nosync are mutually exclusive";
1463                return -EINVAL;
1464        }
1465
1466        if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) &&
1467            (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags) ||
1468             test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))) {
1469                rs->ti->error = "sync/nosync and rebuild are mutually exclusive";
1470                return -EINVAL;
1471        }
1472
1473        if (write_mostly >= rs->md.raid_disks) {
1474                rs->ti->error = "Can't set all raid1 devices to write_mostly";
1475                return -EINVAL;
1476        }
1477
1478        if (rs->md.sync_speed_max &&
1479            rs->md.sync_speed_min > rs->md.sync_speed_max) {
1480                rs->ti->error = "Bogus recovery rates";
1481                return -EINVAL;
1482        }
1483
1484        if (validate_region_size(rs, region_size))
1485                return -EINVAL;
1486
1487        if (rs->md.chunk_sectors)
1488                max_io_len = rs->md.chunk_sectors;
1489        else
1490                max_io_len = region_size;
1491
1492        if (dm_set_target_max_io_len(rs->ti, max_io_len))
1493                return -EINVAL;
1494
1495        if (rt_is_raid10(rt)) {
1496                if (raid10_copies > rs->md.raid_disks) {
1497                        rs->ti->error = "Not enough devices to satisfy specification";
1498                        return -EINVAL;
1499                }
1500
1501                rs->md.new_layout = raid10_format_to_md_layout(rs, raid10_format, raid10_copies);
1502                if (rs->md.new_layout < 0) {
1503                        rs->ti->error = "Error getting raid10 format";
1504                        return rs->md.new_layout;
1505                }
1506
1507                rt = get_raid_type_by_ll(10, rs->md.new_layout);
1508                if (!rt) {
1509                        rs->ti->error = "Failed to recognize new raid10 layout";
1510                        return -EINVAL;
1511                }
1512
1513                if ((rt->algorithm == ALGORITHM_RAID10_DEFAULT ||
1514                     rt->algorithm == ALGORITHM_RAID10_NEAR) &&
1515                    test_bit(__CTR_FLAG_RAID10_USE_NEAR_SETS, &rs->ctr_flags)) {
1516                        rs->ti->error = "RAID10 format 'near' and 'raid10_use_near_sets' are incompatible";
1517                        return -EINVAL;
1518                }
1519        }
1520
1521        rs->raid10_copies = raid10_copies;
1522
1523        /* Assume there are no metadata devices until the drives are parsed */
1524        rs->md.persistent = 0;
1525        rs->md.external = 1;
1526
1527        /* Check, if any invalid ctr arguments have been passed in for the raid level */
1528        return rs_check_for_valid_flags(rs);
1529}
1530
1531/* Set raid4/5/6 cache size */
1532static int rs_set_raid456_stripe_cache(struct raid_set *rs)
1533{
1534        int r;
1535        struct r5conf *conf;
1536        struct mddev *mddev = &rs->md;
1537        uint32_t min_stripes = max(mddev->chunk_sectors, mddev->new_chunk_sectors) / 2;
1538        uint32_t nr_stripes = rs->stripe_cache_entries;
1539
1540        if (!rt_is_raid456(rs->raid_type)) {
1541                rs->ti->error = "Inappropriate raid level; cannot change stripe_cache size";
1542                return -EINVAL;
1543        }
1544
1545        if (nr_stripes < min_stripes) {
1546                DMINFO("Adjusting requested %u stripe cache entries to %u to suit stripe size",
1547                       nr_stripes, min_stripes);
1548                nr_stripes = min_stripes;
1549        }
1550
1551        conf = mddev->private;
1552        if (!conf) {
1553                rs->ti->error = "Cannot change stripe_cache size on inactive RAID set";
1554                return -EINVAL;
1555        }
1556
1557        /* Try setting number of stripes in raid456 stripe cache */
1558        if (conf->min_nr_stripes != nr_stripes) {
1559                r = raid5_set_cache_size(mddev, nr_stripes);
1560                if (r) {
1561                        rs->ti->error = "Failed to set raid4/5/6 stripe cache size";
1562                        return r;
1563                }
1564
1565                DMINFO("%u stripe cache entries", nr_stripes);
1566        }
1567
1568        return 0;
1569}
1570
1571/* Return # of data stripes as kept in mddev as of @rs (i.e. as of superblock) */
1572static unsigned int mddev_data_stripes(struct raid_set *rs)
1573{
1574        return rs->md.raid_disks - rs->raid_type->parity_devs;
1575}
1576
1577/* Return # of data stripes of @rs (i.e. as of ctr) */
1578static unsigned int rs_data_stripes(struct raid_set *rs)
1579{
1580        return rs->raid_disks - rs->raid_type->parity_devs;
1581}
1582
1583/*
1584 * Retrieve rdev->sectors from any valid raid device of @rs
1585 * to allow userpace to pass in arbitray "- -" device tupples.
1586 */
1587static sector_t __rdev_sectors(struct raid_set *rs)
1588{
1589        int i;
1590
1591        for (i = 0; i < rs->md.raid_disks; i++) {
1592                struct md_rdev *rdev = &rs->dev[i].rdev;
1593
1594                if (!test_bit(Journal, &rdev->flags) &&
1595                    rdev->bdev && rdev->sectors)
1596                        return rdev->sectors;
1597        }
1598
1599        return 0;
1600}
1601
1602/* Check that calculated dev_sectors fits all component devices. */
1603static int _check_data_dev_sectors(struct raid_set *rs)
1604{
1605        sector_t ds = ~0;
1606        struct md_rdev *rdev;
1607
1608        rdev_for_each(rdev, &rs->md)
1609                if (!test_bit(Journal, &rdev->flags) && rdev->bdev) {
1610                        ds = min(ds, to_sector(i_size_read(rdev->bdev->bd_inode)));
1611                        if (ds < rs->md.dev_sectors) {
1612                                rs->ti->error = "Component device(s) too small";
1613                                return -EINVAL;
1614                        }
1615                }
1616
1617        return 0;
1618}
1619
1620/* Calculate the sectors per device and per array used for @rs */
1621static int rs_set_dev_and_array_sectors(struct raid_set *rs, sector_t sectors, bool use_mddev)
1622{
1623        int delta_disks;
1624        unsigned int data_stripes;
1625        sector_t array_sectors = sectors, dev_sectors = sectors;
1626        struct mddev *mddev = &rs->md;
1627
1628        if (use_mddev) {
1629                delta_disks = mddev->delta_disks;
1630                data_stripes = mddev_data_stripes(rs);
1631        } else {
1632                delta_disks = rs->delta_disks;
1633                data_stripes = rs_data_stripes(rs);
1634        }
1635
1636        /* Special raid1 case w/o delta_disks support (yet) */
1637        if (rt_is_raid1(rs->raid_type))
1638                ;
1639        else if (rt_is_raid10(rs->raid_type)) {
1640                if (rs->raid10_copies < 2 ||
1641                    delta_disks < 0) {
1642                        rs->ti->error = "Bogus raid10 data copies or delta disks";
1643                        return -EINVAL;
1644                }
1645
1646                dev_sectors *= rs->raid10_copies;
1647                if (sector_div(dev_sectors, data_stripes))
1648                        goto bad;
1649
1650                array_sectors = (data_stripes + delta_disks) * dev_sectors;
1651                if (sector_div(array_sectors, rs->raid10_copies))
1652                        goto bad;
1653
1654        } else if (sector_div(dev_sectors, data_stripes))
1655                goto bad;
1656
1657        else
1658                /* Striped layouts */
1659                array_sectors = (data_stripes + delta_disks) * dev_sectors;
1660
1661        mddev->array_sectors = array_sectors;
1662        mddev->dev_sectors = dev_sectors;
1663        rs_set_rdev_sectors(rs);
1664
1665        return _check_data_dev_sectors(rs);
1666bad:
1667        rs->ti->error = "Target length not divisible by number of data devices";
1668        return -EINVAL;
1669}
1670
1671/* Setup recovery on @rs */
1672static void rs_setup_recovery(struct raid_set *rs, sector_t dev_sectors)
1673{
1674        /* raid0 does not recover */
1675        if (rs_is_raid0(rs))
1676                rs->md.recovery_cp = MaxSector;
1677        /*
1678         * A raid6 set has to be recovered either
1679         * completely or for the grown part to
1680         * ensure proper parity and Q-Syndrome
1681         */
1682        else if (rs_is_raid6(rs))
1683                rs->md.recovery_cp = dev_sectors;
1684        /*
1685         * Other raid set types may skip recovery
1686         * depending on the 'nosync' flag.
1687         */
1688        else
1689                rs->md.recovery_cp = test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)
1690                                     ? MaxSector : dev_sectors;
1691}
1692
1693static void do_table_event(struct work_struct *ws)
1694{
1695        struct raid_set *rs = container_of(ws, struct raid_set, md.event_work);
1696
1697        smp_rmb(); /* Make sure we access most actual mddev properties */
1698        if (!rs_is_reshaping(rs)) {
1699                if (rs_is_raid10(rs))
1700                        rs_set_rdev_sectors(rs);
1701                rs_set_capacity(rs);
1702        }
1703        dm_table_event(rs->ti->table);
1704}
1705
1706/*
1707 * Make sure a valid takover (level switch) is being requested on @rs
1708 *
1709 * Conversions of raid sets from one MD personality to another
1710 * have to conform to restrictions which are enforced here.
1711 */
1712static int rs_check_takeover(struct raid_set *rs)
1713{
1714        struct mddev *mddev = &rs->md;
1715        unsigned int near_copies;
1716
1717        if (rs->md.degraded) {
1718                rs->ti->error = "Can't takeover degraded raid set";
1719                return -EPERM;
1720        }
1721
1722        if (rs_is_reshaping(rs)) {
1723                rs->ti->error = "Can't takeover reshaping raid set";
1724                return -EPERM;
1725        }
1726
1727        switch (mddev->level) {
1728        case 0:
1729                /* raid0 -> raid1/5 with one disk */
1730                if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1731                    mddev->raid_disks == 1)
1732                        return 0;
1733
1734                /* raid0 -> raid10 */
1735                if (mddev->new_level == 10 &&
1736                    !(rs->raid_disks % mddev->raid_disks))
1737                        return 0;
1738
1739                /* raid0 with multiple disks -> raid4/5/6 */
1740                if (__within_range(mddev->new_level, 4, 6) &&
1741                    mddev->new_layout == ALGORITHM_PARITY_N &&
1742                    mddev->raid_disks > 1)
1743                        return 0;
1744
1745                break;
1746
1747        case 10:
1748                /* Can't takeover raid10_offset! */
1749                if (__is_raid10_offset(mddev->layout))
1750                        break;
1751
1752                near_copies = __raid10_near_copies(mddev->layout);
1753
1754                /* raid10* -> raid0 */
1755                if (mddev->new_level == 0) {
1756                        /* Can takeover raid10_near with raid disks divisable by data copies! */
1757                        if (near_copies > 1 &&
1758                            !(mddev->raid_disks % near_copies)) {
1759                                mddev->raid_disks /= near_copies;
1760                                mddev->delta_disks = mddev->raid_disks;
1761                                return 0;
1762                        }
1763
1764                        /* Can takeover raid10_far */
1765                        if (near_copies == 1 &&
1766                            __raid10_far_copies(mddev->layout) > 1)
1767                                return 0;
1768
1769                        break;
1770                }
1771
1772                /* raid10_{near,far} -> raid1 */
1773                if (mddev->new_level == 1 &&
1774                    max(near_copies, __raid10_far_copies(mddev->layout)) == mddev->raid_disks)
1775                        return 0;
1776
1777                /* raid10_{near,far} with 2 disks -> raid4/5 */
1778                if (__within_range(mddev->new_level, 4, 5) &&
1779                    mddev->raid_disks == 2)
1780                        return 0;
1781                break;
1782
1783        case 1:
1784                /* raid1 with 2 disks -> raid4/5 */
1785                if (__within_range(mddev->new_level, 4, 5) &&
1786                    mddev->raid_disks == 2) {
1787                        mddev->degraded = 1;
1788                        return 0;
1789                }
1790
1791                /* raid1 -> raid0 */
1792                if (mddev->new_level == 0 &&
1793                    mddev->raid_disks == 1)
1794                        return 0;
1795
1796                /* raid1 -> raid10 */
1797                if (mddev->new_level == 10)
1798                        return 0;
1799                break;
1800
1801        case 4:
1802                /* raid4 -> raid0 */
1803                if (mddev->new_level == 0)
1804                        return 0;
1805
1806                /* raid4 -> raid1/5 with 2 disks */
1807                if ((mddev->new_level == 1 || mddev->new_level == 5) &&
1808                    mddev->raid_disks == 2)
1809                        return 0;
1810
1811                /* raid4 -> raid5/6 with parity N */
1812                if (__within_range(mddev->new_level, 5, 6) &&
1813                    mddev->layout == ALGORITHM_PARITY_N)
1814                        return 0;
1815                break;
1816
1817        case 5:
1818                /* raid5 with parity N -> raid0 */
1819                if (mddev->new_level == 0 &&
1820                    mddev->layout == ALGORITHM_PARITY_N)
1821                        return 0;
1822
1823                /* raid5 with parity N -> raid4 */
1824                if (mddev->new_level == 4 &&
1825                    mddev->layout == ALGORITHM_PARITY_N)
1826                        return 0;
1827
1828                /* raid5 with 2 disks -> raid1/4/10 */
1829                if ((mddev->new_level == 1 || mddev->new_level == 4 || mddev->new_level == 10) &&
1830                    mddev->raid_disks == 2)
1831                        return 0;
1832
1833                /* raid5_* ->  raid6_*_6 with Q-Syndrome N (e.g. raid5_ra -> raid6_ra_6 */
1834                if (mddev->new_level == 6 &&
1835                    ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1836                      __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC_6, ALGORITHM_RIGHT_SYMMETRIC_6)))
1837                        return 0;
1838                break;
1839
1840        case 6:
1841                /* raid6 with parity N -> raid0 */
1842                if (mddev->new_level == 0 &&
1843                    mddev->layout == ALGORITHM_PARITY_N)
1844                        return 0;
1845
1846                /* raid6 with parity N -> raid4 */
1847                if (mddev->new_level == 4 &&
1848                    mddev->layout == ALGORITHM_PARITY_N)
1849                        return 0;
1850
1851                /* raid6_*_n with Q-Syndrome N -> raid5_* */
1852                if (mddev->new_level == 5 &&
1853                    ((mddev->layout == ALGORITHM_PARITY_N && mddev->new_layout == ALGORITHM_PARITY_N) ||
1854                     __within_range(mddev->new_layout, ALGORITHM_LEFT_ASYMMETRIC, ALGORITHM_RIGHT_SYMMETRIC)))
1855                        return 0;
1856
1857        default:
1858                break;
1859        }
1860
1861        rs->ti->error = "takeover not possible";
1862        return -EINVAL;
1863}
1864
1865/* True if @rs requested to be taken over */
1866static bool rs_takeover_requested(struct raid_set *rs)
1867{
1868        return rs->md.new_level != rs->md.level;
1869}
1870
1871/* True if @rs is requested to reshape by ctr */
1872static bool rs_reshape_requested(struct raid_set *rs)
1873{
1874        bool change;
1875        struct mddev *mddev = &rs->md;
1876
1877        if (rs_takeover_requested(rs))
1878                return false;
1879
1880        if (rs_is_raid0(rs))
1881                return false;
1882
1883        change = mddev->new_layout != mddev->layout ||
1884                 mddev->new_chunk_sectors != mddev->chunk_sectors ||
1885                 rs->delta_disks;
1886
1887        /* Historical case to support raid1 reshape without delta disks */
1888        if (rs_is_raid1(rs)) {
1889                if (rs->delta_disks)
1890                        return !!rs->delta_disks;
1891
1892                return !change &&
1893                       mddev->raid_disks != rs->raid_disks;
1894        }
1895
1896        if (rs_is_raid10(rs))
1897                return change &&
1898                       !__is_raid10_far(mddev->new_layout) &&
1899                       rs->delta_disks >= 0;
1900
1901        return change;
1902}
1903
1904/*  Features */
1905#define FEATURE_FLAG_SUPPORTS_V190      0x1 /* Supports extended superblock */
1906
1907/* State flags for sb->flags */
1908#define SB_FLAG_RESHAPE_ACTIVE          0x1
1909#define SB_FLAG_RESHAPE_BACKWARDS       0x2
1910
1911/*
1912 * This structure is never routinely used by userspace, unlike md superblocks.
1913 * Devices with this superblock should only ever be accessed via device-mapper.
1914 */
1915#define DM_RAID_MAGIC 0x64526D44
1916struct dm_raid_superblock {
1917        __le32 magic;           /* "DmRd" */
1918        __le32 compat_features; /* Used to indicate compatible features (like 1.9.0 ondisk metadata extension) */
1919
1920        __le32 num_devices;     /* Number of devices in this raid set. (Max 64) */
1921        __le32 array_position;  /* The position of this drive in the raid set */
1922
1923        __le64 events;          /* Incremented by md when superblock updated */
1924        __le64 failed_devices;  /* Pre 1.9.0 part of bit field of devices to */
1925                                /* indicate failures (see extension below) */
1926
1927        /*
1928         * This offset tracks the progress of the repair or replacement of
1929         * an individual drive.
1930         */
1931        __le64 disk_recovery_offset;
1932
1933        /*
1934         * This offset tracks the progress of the initial raid set
1935         * synchronisation/parity calculation.
1936         */
1937        __le64 array_resync_offset;
1938
1939        /*
1940         * raid characteristics
1941         */
1942        __le32 level;
1943        __le32 layout;
1944        __le32 stripe_sectors;
1945
1946        /********************************************************************
1947         * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
1948         *
1949         * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
1950         */
1951
1952        __le32 flags; /* Flags defining array states for reshaping */
1953
1954        /*
1955         * This offset tracks the progress of a raid
1956         * set reshape in order to be able to restart it
1957         */
1958        __le64 reshape_position;
1959
1960        /*
1961         * These define the properties of the array in case of an interrupted reshape
1962         */
1963        __le32 new_level;
1964        __le32 new_layout;
1965        __le32 new_stripe_sectors;
1966        __le32 delta_disks;
1967
1968        __le64 array_sectors; /* Array size in sectors */
1969
1970        /*
1971         * Sector offsets to data on devices (reshaping).
1972         * Needed to support out of place reshaping, thus
1973         * not writing over any stripes whilst converting
1974         * them from old to new layout
1975         */
1976        __le64 data_offset;
1977        __le64 new_data_offset;
1978
1979        __le64 sectors; /* Used device size in sectors */
1980
1981        /*
1982         * Additonal Bit field of devices indicating failures to support
1983         * up to 256 devices with the 1.9.0 on-disk metadata format
1984         */
1985        __le64 extended_failed_devices[DISKS_ARRAY_ELEMS - 1];
1986
1987        __le32 incompat_features;       /* Used to indicate any incompatible features */
1988
1989        /* Always set rest up to logical block size to 0 when writing (see get_metadata_device() below). */
1990} __packed;
1991
1992/*
1993 * Check for reshape constraints on raid set @rs:
1994 *
1995 * - reshape function non-existent
1996 * - degraded set
1997 * - ongoing recovery
1998 * - ongoing reshape
1999 *
2000 * Returns 0 if none or -EPERM if given constraint
2001 * and error message reference in @errmsg
2002 */
2003static int rs_check_reshape(struct raid_set *rs)
2004{
2005        struct mddev *mddev = &rs->md;
2006
2007        if (!mddev->pers || !mddev->pers->check_reshape)
2008                rs->ti->error = "Reshape not supported";
2009        else if (mddev->degraded)
2010                rs->ti->error = "Can't reshape degraded raid set";
2011        else if (rs_is_recovering(rs))
2012                rs->ti->error = "Convert request on recovering raid set prohibited";
2013        else if (rs_is_reshaping(rs))
2014                rs->ti->error = "raid set already reshaping!";
2015        else if (!(rs_is_raid1(rs) || rs_is_raid10(rs) || rs_is_raid456(rs)))
2016                rs->ti->error = "Reshaping only supported for raid1/4/5/6/10";
2017        else
2018                return 0;
2019
2020        return -EPERM;
2021}
2022
2023static int read_disk_sb(struct md_rdev *rdev, int size, bool force_reload)
2024{
2025        BUG_ON(!rdev->sb_page);
2026
2027        if (rdev->sb_loaded && !force_reload)
2028                return 0;
2029
2030        rdev->sb_loaded = 0;
2031
2032        if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true)) {
2033                DMERR("Failed to read superblock of device at position %d",
2034                      rdev->raid_disk);
2035                md_error(rdev->mddev, rdev);
2036                set_bit(Faulty, &rdev->flags);
2037                return -EIO;
2038        }
2039
2040        rdev->sb_loaded = 1;
2041
2042        return 0;
2043}
2044
2045static void sb_retrieve_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2046{
2047        failed_devices[0] = le64_to_cpu(sb->failed_devices);
2048        memset(failed_devices + 1, 0, sizeof(sb->extended_failed_devices));
2049
2050        if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2051                int i = ARRAY_SIZE(sb->extended_failed_devices);
2052
2053                while (i--)
2054                        failed_devices[i+1] = le64_to_cpu(sb->extended_failed_devices[i]);
2055        }
2056}
2057
2058static void sb_update_failed_devices(struct dm_raid_superblock *sb, uint64_t *failed_devices)
2059{
2060        int i = ARRAY_SIZE(sb->extended_failed_devices);
2061
2062        sb->failed_devices = cpu_to_le64(failed_devices[0]);
2063        while (i--)
2064                sb->extended_failed_devices[i] = cpu_to_le64(failed_devices[i+1]);
2065}
2066
2067/*
2068 * Synchronize the superblock members with the raid set properties
2069 *
2070 * All superblock data is little endian.
2071 */
2072static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
2073{
2074        bool update_failed_devices = false;
2075        unsigned int i;
2076        uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2077        struct dm_raid_superblock *sb;
2078        struct raid_set *rs = container_of(mddev, struct raid_set, md);
2079
2080        /* No metadata device, no superblock */
2081        if (!rdev->meta_bdev)
2082                return;
2083
2084        BUG_ON(!rdev->sb_page);
2085
2086        sb = page_address(rdev->sb_page);
2087
2088        sb_retrieve_failed_devices(sb, failed_devices);
2089
2090        for (i = 0; i < rs->raid_disks; i++)
2091                if (!rs->dev[i].data_dev || test_bit(Faulty, &rs->dev[i].rdev.flags)) {
2092                        update_failed_devices = true;
2093                        set_bit(i, (void *) failed_devices);
2094                }
2095
2096        if (update_failed_devices)
2097                sb_update_failed_devices(sb, failed_devices);
2098
2099        sb->magic = cpu_to_le32(DM_RAID_MAGIC);
2100        sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2101
2102        sb->num_devices = cpu_to_le32(mddev->raid_disks);
2103        sb->array_position = cpu_to_le32(rdev->raid_disk);
2104
2105        sb->events = cpu_to_le64(mddev->events);
2106
2107        sb->disk_recovery_offset = cpu_to_le64(rdev->recovery_offset);
2108        sb->array_resync_offset = cpu_to_le64(mddev->recovery_cp);
2109
2110        sb->level = cpu_to_le32(mddev->level);
2111        sb->layout = cpu_to_le32(mddev->layout);
2112        sb->stripe_sectors = cpu_to_le32(mddev->chunk_sectors);
2113
2114        /********************************************************************
2115         * BELOW FOLLOW V1.9.0 EXTENSIONS TO THE PRISTINE SUPERBLOCK FORMAT!!!
2116         *
2117         * FEATURE_FLAG_SUPPORTS_V190 in the compat_features member indicates that those exist
2118         */
2119        sb->new_level = cpu_to_le32(mddev->new_level);
2120        sb->new_layout = cpu_to_le32(mddev->new_layout);
2121        sb->new_stripe_sectors = cpu_to_le32(mddev->new_chunk_sectors);
2122
2123        sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2124
2125        smp_rmb(); /* Make sure we access most recent reshape position */
2126        sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2127        if (le64_to_cpu(sb->reshape_position) != MaxSector) {
2128                /* Flag ongoing reshape */
2129                sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE);
2130
2131                if (mddev->delta_disks < 0 || mddev->reshape_backwards)
2132                        sb->flags |= cpu_to_le32(SB_FLAG_RESHAPE_BACKWARDS);
2133        } else {
2134                /* Clear reshape flags */
2135                sb->flags &= ~(cpu_to_le32(SB_FLAG_RESHAPE_ACTIVE|SB_FLAG_RESHAPE_BACKWARDS));
2136        }
2137
2138        sb->array_sectors = cpu_to_le64(mddev->array_sectors);
2139        sb->data_offset = cpu_to_le64(rdev->data_offset);
2140        sb->new_data_offset = cpu_to_le64(rdev->new_data_offset);
2141        sb->sectors = cpu_to_le64(rdev->sectors);
2142        sb->incompat_features = cpu_to_le32(0);
2143
2144        /* Zero out the rest of the payload after the size of the superblock */
2145        memset(sb + 1, 0, rdev->sb_size - sizeof(*sb));
2146}
2147
2148/*
2149 * super_load
2150 *
2151 * This function creates a superblock if one is not found on the device
2152 * and will decide which superblock to use if there's a choice.
2153 *
2154 * Return: 1 if use rdev, 0 if use refdev, -Exxx otherwise
2155 */
2156static int super_load(struct md_rdev *rdev, struct md_rdev *refdev)
2157{
2158        int r;
2159        struct dm_raid_superblock *sb;
2160        struct dm_raid_superblock *refsb;
2161        uint64_t events_sb, events_refsb;
2162
2163        r = read_disk_sb(rdev, rdev->sb_size, false);
2164        if (r)
2165                return r;
2166
2167        sb = page_address(rdev->sb_page);
2168
2169        /*
2170         * Two cases that we want to write new superblocks and rebuild:
2171         * 1) New device (no matching magic number)
2172         * 2) Device specified for rebuild (!In_sync w/ offset == 0)
2173         */
2174        if ((sb->magic != cpu_to_le32(DM_RAID_MAGIC)) ||
2175            (!test_bit(In_sync, &rdev->flags) && !rdev->recovery_offset)) {
2176                super_sync(rdev->mddev, rdev);
2177
2178                set_bit(FirstUse, &rdev->flags);
2179                sb->compat_features = cpu_to_le32(FEATURE_FLAG_SUPPORTS_V190);
2180
2181                /* Force writing of superblocks to disk */
2182                set_bit(MD_SB_CHANGE_DEVS, &rdev->mddev->sb_flags);
2183
2184                /* Any superblock is better than none, choose that if given */
2185                return refdev ? 0 : 1;
2186        }
2187
2188        if (!refdev)
2189                return 1;
2190
2191        events_sb = le64_to_cpu(sb->events);
2192
2193        refsb = page_address(refdev->sb_page);
2194        events_refsb = le64_to_cpu(refsb->events);
2195
2196        return (events_sb > events_refsb) ? 1 : 0;
2197}
2198
2199static int super_init_validation(struct raid_set *rs, struct md_rdev *rdev)
2200{
2201        int role;
2202        unsigned int d;
2203        struct mddev *mddev = &rs->md;
2204        uint64_t events_sb;
2205        uint64_t failed_devices[DISKS_ARRAY_ELEMS];
2206        struct dm_raid_superblock *sb;
2207        uint32_t new_devs = 0, rebuild_and_new = 0, rebuilds = 0;
2208        struct md_rdev *r;
2209        struct dm_raid_superblock *sb2;
2210
2211        sb = page_address(rdev->sb_page);
2212        events_sb = le64_to_cpu(sb->events);
2213
2214        /*
2215         * Initialise to 1 if this is a new superblock.
2216         */
2217        mddev->events = events_sb ? : 1;
2218
2219        mddev->reshape_position = MaxSector;
2220
2221        mddev->raid_disks = le32_to_cpu(sb->num_devices);
2222        mddev->level = le32_to_cpu(sb->level);
2223        mddev->layout = le32_to_cpu(sb->layout);
2224        mddev->chunk_sectors = le32_to_cpu(sb->stripe_sectors);
2225
2226        /*
2227         * Reshaping is supported, e.g. reshape_position is valid
2228         * in superblock and superblock content is authoritative.
2229         */
2230        if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190) {
2231                /* Superblock is authoritative wrt given raid set layout! */
2232                mddev->new_level = le32_to_cpu(sb->new_level);
2233                mddev->new_layout = le32_to_cpu(sb->new_layout);
2234                mddev->new_chunk_sectors = le32_to_cpu(sb->new_stripe_sectors);
2235                mddev->delta_disks = le32_to_cpu(sb->delta_disks);
2236                mddev->array_sectors = le64_to_cpu(sb->array_sectors);
2237
2238                /* raid was reshaping and got interrupted */
2239                if (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_ACTIVE) {
2240                        if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags)) {
2241                                DMERR("Reshape requested but raid set is still reshaping");
2242                                return -EINVAL;
2243                        }
2244
2245                        if (mddev->delta_disks < 0 ||
2246                            (!mddev->delta_disks && (le32_to_cpu(sb->flags) & SB_FLAG_RESHAPE_BACKWARDS)))
2247                                mddev->reshape_backwards = 1;
2248                        else
2249                                mddev->reshape_backwards = 0;
2250
2251                        mddev->reshape_position = le64_to_cpu(sb->reshape_position);
2252                        rs->raid_type = get_raid_type_by_ll(mddev->level, mddev->layout);
2253                }
2254
2255        } else {
2256                /*
2257                 * No takeover/reshaping, because we don't have the extended v1.9.0 metadata
2258                 */
2259                struct raid_type *rt_cur = get_raid_type_by_ll(mddev->level, mddev->layout);
2260                struct raid_type *rt_new = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
2261
2262                if (rs_takeover_requested(rs)) {
2263                        if (rt_cur && rt_new)
2264                                DMERR("Takeover raid sets from %s to %s not yet supported by metadata. (raid level change)",
2265                                      rt_cur->name, rt_new->name);
2266                        else
2267                                DMERR("Takeover raid sets not yet supported by metadata. (raid level change)");
2268                        return -EINVAL;
2269                } else if (rs_reshape_requested(rs)) {
2270                        DMERR("Reshaping raid sets not yet supported by metadata. (raid layout change keeping level)");
2271                        if (mddev->layout != mddev->new_layout) {
2272                                if (rt_cur && rt_new)
2273                                        DMERR("  current layout %s vs new layout %s",
2274                                              rt_cur->name, rt_new->name);
2275                                else
2276                                        DMERR("  current layout 0x%X vs new layout 0x%X",
2277                                              le32_to_cpu(sb->layout), mddev->new_layout);
2278                        }
2279                        if (mddev->chunk_sectors != mddev->new_chunk_sectors)
2280                                DMERR("  current stripe sectors %u vs new stripe sectors %u",
2281                                      mddev->chunk_sectors, mddev->new_chunk_sectors);
2282                        if (rs->delta_disks)
2283                                DMERR("  current %u disks vs new %u disks",
2284                                      mddev->raid_disks, mddev->raid_disks + rs->delta_disks);
2285                        if (rs_is_raid10(rs)) {
2286                                DMERR("  Old layout: %s w/ %u copies",
2287                                      raid10_md_layout_to_format(mddev->layout),
2288                                      raid10_md_layout_to_copies(mddev->layout));
2289                                DMERR("  New layout: %s w/ %u copies",
2290                                      raid10_md_layout_to_format(mddev->new_layout),
2291                                      raid10_md_layout_to_copies(mddev->new_layout));
2292                        }
2293                        return -EINVAL;
2294                }
2295
2296                DMINFO("Discovered old metadata format; upgrading to extended metadata format");
2297        }
2298
2299        if (!test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
2300                mddev->recovery_cp = le64_to_cpu(sb->array_resync_offset);
2301
2302        /*
2303         * During load, we set FirstUse if a new superblock was written.
2304         * There are two reasons we might not have a superblock:
2305         * 1) The raid set is brand new - in which case, all of the
2306         *    devices must have their In_sync bit set.  Also,
2307         *    recovery_cp must be 0, unless forced.
2308         * 2) This is a new device being added to an old raid set
2309         *    and the new device needs to be rebuilt - in which
2310         *    case the In_sync bit will /not/ be set and
2311         *    recovery_cp must be MaxSector.
2312         * 3) This is/are a new device(s) being added to an old
2313         *    raid set during takeover to a higher raid level
2314         *    to provide capacity for redundancy or during reshape
2315         *    to add capacity to grow the raid set.
2316         */
2317        d = 0;
2318        rdev_for_each(r, mddev) {
2319                if (test_bit(Journal, &rdev->flags))
2320                        continue;
2321
2322                if (test_bit(FirstUse, &r->flags))
2323                        new_devs++;
2324
2325                if (!test_bit(In_sync, &r->flags)) {
2326                        DMINFO("Device %d specified for rebuild; clearing superblock",
2327                                r->raid_disk);
2328                        rebuilds++;
2329
2330                        if (test_bit(FirstUse, &r->flags))
2331                                rebuild_and_new++;
2332                }
2333
2334                d++;
2335        }
2336
2337        if (new_devs == rs->raid_disks || !rebuilds) {
2338                /* Replace a broken device */
2339                if (new_devs == rs->raid_disks) {
2340                        DMINFO("Superblocks created for new raid set");
2341                        set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2342                } else if (new_devs != rebuilds &&
2343                           new_devs != rs->delta_disks) {
2344                        DMERR("New device injected into existing raid set without "
2345                              "'delta_disks' or 'rebuild' parameter specified");
2346                        return -EINVAL;
2347                }
2348        } else if (new_devs && new_devs != rebuilds) {
2349                DMERR("%u 'rebuild' devices cannot be injected into"
2350                      " a raid set with %u other first-time devices",
2351                      rebuilds, new_devs);
2352                return -EINVAL;
2353        } else if (rebuilds) {
2354                if (rebuild_and_new && rebuilds != rebuild_and_new) {
2355                        DMERR("new device%s provided without 'rebuild'",
2356                              new_devs > 1 ? "s" : "");
2357                        return -EINVAL;
2358                } else if (!test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) && rs_is_recovering(rs)) {
2359                        DMERR("'rebuild' specified while raid set is not in-sync (recovery_cp=%llu)",
2360                              (unsigned long long) mddev->recovery_cp);
2361                        return -EINVAL;
2362                } else if (rs_is_reshaping(rs)) {
2363                        DMERR("'rebuild' specified while raid set is being reshaped (reshape_position=%llu)",
2364                              (unsigned long long) mddev->reshape_position);
2365                        return -EINVAL;
2366                }
2367        }
2368
2369        /*
2370         * Now we set the Faulty bit for those devices that are
2371         * recorded in the superblock as failed.
2372         */
2373        sb_retrieve_failed_devices(sb, failed_devices);
2374        rdev_for_each(r, mddev) {
2375                if (test_bit(Journal, &rdev->flags) ||
2376                    !r->sb_page)
2377                        continue;
2378                sb2 = page_address(r->sb_page);
2379                sb2->failed_devices = 0;
2380                memset(sb2->extended_failed_devices, 0, sizeof(sb2->extended_failed_devices));
2381
2382                /*
2383                 * Check for any device re-ordering.
2384                 */
2385                if (!test_bit(FirstUse, &r->flags) && (r->raid_disk >= 0)) {
2386                        role = le32_to_cpu(sb2->array_position);
2387                        if (role < 0)
2388                                continue;
2389
2390                        if (role != r->raid_disk) {
2391                                if (rs_is_raid10(rs) && __is_raid10_near(mddev->layout)) {
2392                                        if (mddev->raid_disks % __raid10_near_copies(mddev->layout) ||
2393                                            rs->raid_disks % rs->raid10_copies) {
2394                                                rs->ti->error =
2395                                                        "Cannot change raid10 near set to odd # of devices!";
2396                                                return -EINVAL;
2397                                        }
2398
2399                                        sb2->array_position = cpu_to_le32(r->raid_disk);
2400
2401                                } else if (!(rs_is_raid10(rs) && rt_is_raid0(rs->raid_type)) &&
2402                                           !(rs_is_raid0(rs) && rt_is_raid10(rs->raid_type)) &&
2403                                           !rt_is_raid1(rs->raid_type)) {
2404                                        rs->ti->error = "Cannot change device positions in raid set";
2405                                        return -EINVAL;
2406                                }
2407
2408                                DMINFO("raid device #%d now at position #%d", role, r->raid_disk);
2409                        }
2410
2411                        /*
2412                         * Partial recovery is performed on
2413                         * returning failed devices.
2414                         */
2415                        if (test_bit(role, (void *) failed_devices))
2416                                set_bit(Faulty, &r->flags);
2417                }
2418        }
2419
2420        return 0;
2421}
2422
2423static int super_validate(struct raid_set *rs, struct md_rdev *rdev)
2424{
2425        struct mddev *mddev = &rs->md;
2426        struct dm_raid_superblock *sb;
2427
2428        if (rs_is_raid0(rs) || !rdev->sb_page || rdev->raid_disk < 0)
2429                return 0;
2430
2431        sb = page_address(rdev->sb_page);
2432
2433        /*
2434         * If mddev->events is not set, we know we have not yet initialized
2435         * the array.
2436         */
2437        if (!mddev->events && super_init_validation(rs, rdev))
2438                return -EINVAL;
2439
2440        if (le32_to_cpu(sb->compat_features) &&
2441            le32_to_cpu(sb->compat_features) != FEATURE_FLAG_SUPPORTS_V190) {
2442                rs->ti->error = "Unable to assemble array: Unknown flag(s) in compatible feature flags";
2443                return -EINVAL;
2444        }
2445
2446        if (sb->incompat_features) {
2447                rs->ti->error = "Unable to assemble array: No incompatible feature flags supported yet";
2448                return -EINVAL;
2449        }
2450
2451        /* Enable bitmap creation on @rs unless no metadevs or raid0 or journaled raid4/5/6 set. */
2452        mddev->bitmap_info.offset = (rt_is_raid0(rs->raid_type) || rs->journal_dev.dev) ? 0 : to_sector(4096);
2453        mddev->bitmap_info.default_offset = mddev->bitmap_info.offset;
2454
2455        if (!test_and_clear_bit(FirstUse, &rdev->flags)) {
2456                /*
2457                 * Retrieve rdev size stored in superblock to be prepared for shrink.
2458                 * Check extended superblock members are present otherwise the size
2459                 * will not be set!
2460                 */
2461                if (le32_to_cpu(sb->compat_features) & FEATURE_FLAG_SUPPORTS_V190)
2462                        rdev->sectors = le64_to_cpu(sb->sectors);
2463
2464                rdev->recovery_offset = le64_to_cpu(sb->disk_recovery_offset);
2465                if (rdev->recovery_offset == MaxSector)
2466                        set_bit(In_sync, &rdev->flags);
2467                /*
2468                 * If no reshape in progress -> we're recovering single
2469                 * disk(s) and have to set the device(s) to out-of-sync
2470                 */
2471                else if (!rs_is_reshaping(rs))
2472                        clear_bit(In_sync, &rdev->flags); /* Mandatory for recovery */
2473        }
2474
2475        /*
2476         * If a device comes back, set it as not In_sync and no longer faulty.
2477         */
2478        if (test_and_clear_bit(Faulty, &rdev->flags)) {
2479                rdev->recovery_offset = 0;
2480                clear_bit(In_sync, &rdev->flags);
2481                rdev->saved_raid_disk = rdev->raid_disk;
2482        }
2483
2484        /* Reshape support -> restore repective data offsets */
2485        rdev->data_offset = le64_to_cpu(sb->data_offset);
2486        rdev->new_data_offset = le64_to_cpu(sb->new_data_offset);
2487
2488        return 0;
2489}
2490
2491/*
2492 * Analyse superblocks and select the freshest.
2493 */
2494static int analyse_superblocks(struct dm_target *ti, struct raid_set *rs)
2495{
2496        int r;
2497        struct md_rdev *rdev, *freshest;
2498        struct mddev *mddev = &rs->md;
2499
2500        freshest = NULL;
2501        rdev_for_each(rdev, mddev) {
2502                if (test_bit(Journal, &rdev->flags))
2503                        continue;
2504
2505                if (!rdev->meta_bdev)
2506                        continue;
2507
2508                /* Set superblock offset/size for metadata device. */
2509                rdev->sb_start = 0;
2510                rdev->sb_size = bdev_logical_block_size(rdev->meta_bdev);
2511                if (rdev->sb_size < sizeof(struct dm_raid_superblock) || rdev->sb_size > PAGE_SIZE) {
2512                        DMERR("superblock size of a logical block is no longer valid");
2513                        return -EINVAL;
2514                }
2515
2516                /*
2517                 * Skipping super_load due to CTR_FLAG_SYNC will cause
2518                 * the array to undergo initialization again as
2519                 * though it were new.  This is the intended effect
2520                 * of the "sync" directive.
2521                 *
2522                 * With reshaping capability added, we must ensure that
2523                 * that the "sync" directive is disallowed during the reshape.
2524                 */
2525                if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
2526                        continue;
2527
2528                r = super_load(rdev, freshest);
2529
2530                switch (r) {
2531                case 1:
2532                        freshest = rdev;
2533                        break;
2534                case 0:
2535                        break;
2536                default:
2537                        /* This is a failure to read the superblock from the metadata device. */
2538                        /*
2539                         * We have to keep any raid0 data/metadata device pairs or
2540                         * the MD raid0 personality will fail to start the array.
2541                         */
2542                        if (rs_is_raid0(rs))
2543                                continue;
2544
2545                        /*
2546                         * We keep the dm_devs to be able to emit the device tuple
2547                         * properly on the table line in raid_status() (rather than
2548                         * mistakenly acting as if '- -' got passed into the constructor).
2549                         *
2550                         * The rdev has to stay on the same_set list to allow for
2551                         * the attempt to restore faulty devices on second resume.
2552                         */
2553                        rdev->raid_disk = rdev->saved_raid_disk = -1;
2554                        break;
2555                }
2556        }
2557
2558        if (!freshest)
2559                return 0;
2560
2561        /*
2562         * Validation of the freshest device provides the source of
2563         * validation for the remaining devices.
2564         */
2565        rs->ti->error = "Unable to assemble array: Invalid superblocks";
2566        if (super_validate(rs, freshest))
2567                return -EINVAL;
2568
2569        if (validate_raid_redundancy(rs)) {
2570                rs->ti->error = "Insufficient redundancy to activate array";
2571                return -EINVAL;
2572        }
2573
2574        rdev_for_each(rdev, mddev)
2575                if (!test_bit(Journal, &rdev->flags) &&
2576                    rdev != freshest &&
2577                    super_validate(rs, rdev))
2578                        return -EINVAL;
2579        return 0;
2580}
2581
2582/*
2583 * Adjust data_offset and new_data_offset on all disk members of @rs
2584 * for out of place reshaping if requested by contructor
2585 *
2586 * We need free space at the beginning of each raid disk for forward
2587 * and at the end for backward reshapes which userspace has to provide
2588 * via remapping/reordering of space.
2589 */
2590static int rs_adjust_data_offsets(struct raid_set *rs)
2591{
2592        sector_t data_offset = 0, new_data_offset = 0;
2593        struct md_rdev *rdev;
2594
2595        /* Constructor did not request data offset change */
2596        if (!test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags)) {
2597                if (!rs_is_reshapable(rs))
2598                        goto out;
2599
2600                return 0;
2601        }
2602
2603        /* HM FIXME: get In_Sync raid_dev? */
2604        rdev = &rs->dev[0].rdev;
2605
2606        if (rs->delta_disks < 0) {
2607                /*
2608                 * Removing disks (reshaping backwards):
2609                 *
2610                 * - before reshape: data is at offset 0 and free space
2611                 *                   is at end of each component LV
2612                 *
2613                 * - after reshape: data is at offset rs->data_offset != 0 on each component LV
2614                 */
2615                data_offset = 0;
2616                new_data_offset = rs->data_offset;
2617
2618        } else if (rs->delta_disks > 0) {
2619                /*
2620                 * Adding disks (reshaping forwards):
2621                 *
2622                 * - before reshape: data is at offset rs->data_offset != 0 and
2623                 *                   free space is at begin of each component LV
2624                 *
2625                 * - after reshape: data is at offset 0 on each component LV
2626                 */
2627                data_offset = rs->data_offset;
2628                new_data_offset = 0;
2629
2630        } else {
2631                /*
2632                 * User space passes in 0 for data offset after having removed reshape space
2633                 *
2634                 * - or - (data offset != 0)
2635                 *
2636                 * Changing RAID layout or chunk size -> toggle offsets
2637                 *
2638                 * - before reshape: data is at offset rs->data_offset 0 and
2639                 *                   free space is at end of each component LV
2640                 *                   -or-
2641                 *                   data is at offset rs->data_offset != 0 and
2642                 *                   free space is at begin of each component LV
2643                 *
2644                 * - after reshape: data is at offset 0 if it was at offset != 0
2645                 *                  or at offset != 0 if it was at offset 0
2646                 *                  on each component LV
2647                 *
2648                 */
2649                data_offset = rs->data_offset ? rdev->data_offset : 0;
2650                new_data_offset = data_offset ? 0 : rs->data_offset;
2651                set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2652        }
2653
2654        /*
2655         * Make sure we got a minimum amount of free sectors per device
2656         */
2657        if (rs->data_offset &&
2658            to_sector(i_size_read(rdev->bdev->bd_inode)) - rs->md.dev_sectors < MIN_FREE_RESHAPE_SPACE) {
2659                rs->ti->error = data_offset ? "No space for forward reshape" :
2660                                              "No space for backward reshape";
2661                return -ENOSPC;
2662        }
2663out:
2664        /*
2665         * Raise recovery_cp in case data_offset != 0 to
2666         * avoid false recovery positives in the constructor.
2667         */
2668        if (rs->md.recovery_cp < rs->md.dev_sectors)
2669                rs->md.recovery_cp += rs->dev[0].rdev.data_offset;
2670
2671        /* Adjust data offsets on all rdevs but on any raid4/5/6 journal device */
2672        rdev_for_each(rdev, &rs->md) {
2673                if (!test_bit(Journal, &rdev->flags)) {
2674                        rdev->data_offset = data_offset;
2675                        rdev->new_data_offset = new_data_offset;
2676                }
2677        }
2678
2679        return 0;
2680}
2681
2682/* Userpace reordered disks -> adjust raid_disk indexes in @rs */
2683static void __reorder_raid_disk_indexes(struct raid_set *rs)
2684{
2685        int i = 0;
2686        struct md_rdev *rdev;
2687
2688        rdev_for_each(rdev, &rs->md) {
2689                if (!test_bit(Journal, &rdev->flags)) {
2690                        rdev->raid_disk = i++;
2691                        rdev->saved_raid_disk = rdev->new_raid_disk = -1;
2692                }
2693        }
2694}
2695
2696/*
2697 * Setup @rs for takeover by a different raid level
2698 */
2699static int rs_setup_takeover(struct raid_set *rs)
2700{
2701        struct mddev *mddev = &rs->md;
2702        struct md_rdev *rdev;
2703        unsigned int d = mddev->raid_disks = rs->raid_disks;
2704        sector_t new_data_offset = rs->dev[0].rdev.data_offset ? 0 : rs->data_offset;
2705
2706        if (rt_is_raid10(rs->raid_type)) {
2707                if (rs_is_raid0(rs)) {
2708                        /* Userpace reordered disks -> adjust raid_disk indexes */
2709                        __reorder_raid_disk_indexes(rs);
2710
2711                        /* raid0 -> raid10_far layout */
2712                        mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_FAR,
2713                                                                   rs->raid10_copies);
2714                } else if (rs_is_raid1(rs))
2715                        /* raid1 -> raid10_near layout */
2716                        mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2717                                                                   rs->raid_disks);
2718                else
2719                        return -EINVAL;
2720
2721        }
2722
2723        clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2724        mddev->recovery_cp = MaxSector;
2725
2726        while (d--) {
2727                rdev = &rs->dev[d].rdev;
2728
2729                if (test_bit(d, (void *) rs->rebuild_disks)) {
2730                        clear_bit(In_sync, &rdev->flags);
2731                        clear_bit(Faulty, &rdev->flags);
2732                        mddev->recovery_cp = rdev->recovery_offset = 0;
2733                        /* Bitmap has to be created when we do an "up" takeover */
2734                        set_bit(MD_ARRAY_FIRST_USE, &mddev->flags);
2735                }
2736
2737                rdev->new_data_offset = new_data_offset;
2738        }
2739
2740        return 0;
2741}
2742
2743/* Prepare @rs for reshape */
2744static int rs_prepare_reshape(struct raid_set *rs)
2745{
2746        bool reshape;
2747        struct mddev *mddev = &rs->md;
2748
2749        if (rs_is_raid10(rs)) {
2750                if (rs->raid_disks != mddev->raid_disks &&
2751                    __is_raid10_near(mddev->layout) &&
2752                    rs->raid10_copies &&
2753                    rs->raid10_copies != __raid10_near_copies(mddev->layout)) {
2754                        /*
2755                         * raid disk have to be multiple of data copies to allow this conversion,
2756                         *
2757                         * This is actually not a reshape it is a
2758                         * rebuild of any additional mirrors per group
2759                         */
2760                        if (rs->raid_disks % rs->raid10_copies) {
2761                                rs->ti->error = "Can't reshape raid10 mirror groups";
2762                                return -EINVAL;
2763                        }
2764
2765                        /* Userpace reordered disks to add/remove mirrors -> adjust raid_disk indexes */
2766                        __reorder_raid_disk_indexes(rs);
2767                        mddev->layout = raid10_format_to_md_layout(rs, ALGORITHM_RAID10_NEAR,
2768                                                                   rs->raid10_copies);
2769                        mddev->new_layout = mddev->layout;
2770                        reshape = false;
2771                } else
2772                        reshape = true;
2773
2774        } else if (rs_is_raid456(rs))
2775                reshape = true;
2776
2777        else if (rs_is_raid1(rs)) {
2778                if (rs->delta_disks) {
2779                        /* Process raid1 via delta_disks */
2780                        mddev->degraded = rs->delta_disks < 0 ? -rs->delta_disks : rs->delta_disks;
2781                        reshape = true;
2782                } else {
2783                        /* Process raid1 without delta_disks */
2784                        mddev->raid_disks = rs->raid_disks;
2785                        reshape = false;
2786                }
2787        } else {
2788                rs->ti->error = "Called with bogus raid type";
2789                return -EINVAL;
2790        }
2791
2792        if (reshape) {
2793                set_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags);
2794                set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2795        } else if (mddev->raid_disks < rs->raid_disks)
2796                /* Create new superblocks and bitmaps, if any new disks */
2797                set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
2798
2799        return 0;
2800}
2801
2802/* Get reshape sectors from data_offsets or raid set */
2803static sector_t _get_reshape_sectors(struct raid_set *rs)
2804{
2805        struct md_rdev *rdev;
2806        sector_t reshape_sectors = 0;
2807
2808        rdev_for_each(rdev, &rs->md)
2809                if (!test_bit(Journal, &rdev->flags)) {
2810                        reshape_sectors = (rdev->data_offset > rdev->new_data_offset) ?
2811                                        rdev->data_offset - rdev->new_data_offset :
2812                                        rdev->new_data_offset - rdev->data_offset;
2813                        break;
2814                }
2815
2816        return max(reshape_sectors, (sector_t) rs->data_offset);
2817}
2818
2819/*
2820 *
2821 * - change raid layout
2822 * - change chunk size
2823 * - add disks
2824 * - remove disks
2825 */
2826static int rs_setup_reshape(struct raid_set *rs)
2827{
2828        int r = 0;
2829        unsigned int cur_raid_devs, d;
2830        sector_t reshape_sectors = _get_reshape_sectors(rs);
2831        struct mddev *mddev = &rs->md;
2832        struct md_rdev *rdev;
2833
2834        mddev->delta_disks = rs->delta_disks;
2835        cur_raid_devs = mddev->raid_disks;
2836
2837        /* Ignore impossible layout change whilst adding/removing disks */
2838        if (mddev->delta_disks &&
2839            mddev->layout != mddev->new_layout) {
2840                DMINFO("Ignoring invalid layout change with delta_disks=%d", rs->delta_disks);
2841                mddev->new_layout = mddev->layout;
2842        }
2843
2844        /*
2845         * Adjust array size:
2846         *
2847         * - in case of adding disk(s), array size has
2848         *   to grow after the disk adding reshape,
2849         *   which'll hapen in the event handler;
2850         *   reshape will happen forward, so space has to
2851         *   be available at the beginning of each disk
2852         *
2853         * - in case of removing disk(s), array size
2854         *   has to shrink before starting the reshape,
2855         *   which'll happen here;
2856         *   reshape will happen backward, so space has to
2857         *   be available at the end of each disk
2858         *
2859         * - data_offset and new_data_offset are
2860         *   adjusted for aforementioned out of place
2861         *   reshaping based on userspace passing in
2862         *   the "data_offset <sectors>" key/value
2863         *   pair via the constructor
2864         */
2865
2866        /* Add disk(s) */
2867        if (rs->delta_disks > 0) {
2868                /* Prepare disks for check in raid4/5/6/10 {check|start}_reshape */
2869                for (d = cur_raid_devs; d < rs->raid_disks; d++) {
2870                        rdev = &rs->dev[d].rdev;
2871                        clear_bit(In_sync, &rdev->flags);
2872
2873                        /*
2874                         * save_raid_disk needs to be -1, or recovery_offset will be set to 0
2875                         * by md, which'll store that erroneously in the superblock on reshape
2876                         */
2877                        rdev->saved_raid_disk = -1;
2878                        rdev->raid_disk = d;
2879
2880                        rdev->sectors = mddev->dev_sectors;
2881                        rdev->recovery_offset = rs_is_raid1(rs) ? 0 : MaxSector;
2882                }
2883
2884                mddev->reshape_backwards = 0; /* adding disk(s) -> forward reshape */
2885
2886        /* Remove disk(s) */
2887        } else if (rs->delta_disks < 0) {
2888                r = rs_set_dev_and_array_sectors(rs, rs->ti->len, true);
2889                mddev->reshape_backwards = 1; /* removing disk(s) -> backward reshape */
2890
2891        /* Change layout and/or chunk size */
2892        } else {
2893                /*
2894                 * Reshape layout (e.g. raid5_ls -> raid5_n) and/or chunk size:
2895                 *
2896                 * keeping number of disks and do layout change ->
2897                 *
2898                 * toggle reshape_backward depending on data_offset:
2899                 *
2900                 * - free space upfront -> reshape forward
2901                 *
2902                 * - free space at the end -> reshape backward
2903                 *
2904                 *
2905                 * This utilizes free reshape space avoiding the need
2906                 * for userspace to move (parts of) LV segments in
2907                 * case of layout/chunksize change  (for disk
2908                 * adding/removing reshape space has to be at
2909                 * the proper address (see above with delta_disks):
2910                 *
2911                 * add disk(s)   -> begin
2912                 * remove disk(s)-> end
2913                 */
2914                mddev->reshape_backwards = rs->dev[0].rdev.data_offset ? 0 : 1;
2915        }
2916
2917        /*
2918         * Adjust device size for forward reshape
2919         * because md_finish_reshape() reduces it.
2920         */
2921        if (!mddev->reshape_backwards)
2922                rdev_for_each(rdev, &rs->md)
2923                        if (!test_bit(Journal, &rdev->flags))
2924                                rdev->sectors += reshape_sectors;
2925
2926        return r;
2927}
2928
2929/*
2930 * Enable/disable discard support on RAID set depending on
2931 * RAID level and discard properties of underlying RAID members.
2932 */
2933static void configure_discard_support(struct raid_set *rs)
2934{
2935        int i;
2936        bool raid456;
2937        struct dm_target *ti = rs->ti;
2938
2939        /*
2940         * XXX: RAID level 4,5,6 require zeroing for safety.
2941         */
2942        raid456 = rs_is_raid456(rs);
2943
2944        for (i = 0; i < rs->raid_disks; i++) {
2945                struct request_queue *q;
2946
2947                if (!rs->dev[i].rdev.bdev)
2948                        continue;
2949
2950                q = bdev_get_queue(rs->dev[i].rdev.bdev);
2951                if (!q || !blk_queue_discard(q))
2952                        return;
2953
2954                if (raid456) {
2955                        if (!devices_handle_discard_safely) {
2956                                DMERR("raid456 discard support disabled due to discard_zeroes_data uncertainty.");
2957                                DMERR("Set dm-raid.devices_handle_discard_safely=Y to override.");
2958                                return;
2959                        }
2960                }
2961        }
2962
2963        ti->num_discard_bios = 1;
2964}
2965
2966/*
2967 * Construct a RAID0/1/10/4/5/6 mapping:
2968 * Args:
2969 *      <raid_type> <#raid_params> <raid_params>{0,}    \
2970 *      <#raid_devs> [<meta_dev1> <dev1>]{1,}
2971 *
2972 * <raid_params> varies by <raid_type>.  See 'parse_raid_params' for
2973 * details on possible <raid_params>.
2974 *
2975 * Userspace is free to initialize the metadata devices, hence the superblocks to
2976 * enforce recreation based on the passed in table parameters.
2977 *
2978 */
2979static int raid_ctr(struct dm_target *ti, unsigned int argc, char **argv)
2980{
2981        int r;
2982        bool resize = false;
2983        struct raid_type *rt;
2984        unsigned int num_raid_params, num_raid_devs;
2985        sector_t sb_array_sectors, rdev_sectors, reshape_sectors;
2986        struct raid_set *rs = NULL;
2987        const char *arg;
2988        struct rs_layout rs_layout;
2989        struct dm_arg_set as = { argc, argv }, as_nrd;
2990        struct dm_arg _args[] = {
2991                { 0, as.argc, "Cannot understand number of raid parameters" },
2992                { 1, 254, "Cannot understand number of raid devices parameters" }
2993        };
2994
2995        arg = dm_shift_arg(&as);
2996        if (!arg) {
2997                ti->error = "No arguments";
2998                return -EINVAL;
2999        }
3000
3001        rt = get_raid_type(arg);
3002        if (!rt) {
3003                ti->error = "Unrecognised raid_type";
3004                return -EINVAL;
3005        }
3006
3007        /* Must have <#raid_params> */
3008        if (dm_read_arg_group(_args, &as, &num_raid_params, &ti->error))
3009                return -EINVAL;
3010
3011        /* number of raid device tupples <meta_dev data_dev> */
3012        as_nrd = as;
3013        dm_consume_args(&as_nrd, num_raid_params);
3014        _args[1].max = (as_nrd.argc - 1) / 2;
3015        if (dm_read_arg(_args + 1, &as_nrd, &num_raid_devs, &ti->error))
3016                return -EINVAL;
3017
3018        if (!__within_range(num_raid_devs, 1, MAX_RAID_DEVICES)) {
3019                ti->error = "Invalid number of supplied raid devices";
3020                return -EINVAL;
3021        }
3022
3023        rs = raid_set_alloc(ti, rt, num_raid_devs);
3024        if (IS_ERR(rs))
3025                return PTR_ERR(rs);
3026
3027        r = parse_raid_params(rs, &as, num_raid_params);
3028        if (r)
3029                goto bad;
3030
3031        r = parse_dev_params(rs, &as);
3032        if (r)
3033                goto bad;
3034
3035        rs->md.sync_super = super_sync;
3036
3037        /*
3038         * Calculate ctr requested array and device sizes to allow
3039         * for superblock analysis needing device sizes defined.
3040         *
3041         * Any existing superblock will overwrite the array and device sizes
3042         */
3043        r = rs_set_dev_and_array_sectors(rs, rs->ti->len, false);
3044        if (r)
3045                goto bad;
3046
3047        /* Memorize just calculated, potentially larger sizes to grow the raid set in preresume */
3048        rs->array_sectors = rs->md.array_sectors;
3049        rs->dev_sectors = rs->md.dev_sectors;
3050
3051        /*
3052         * Backup any new raid set level, layout, ...
3053         * requested to be able to compare to superblock
3054         * members for conversion decisions.
3055         */
3056        rs_config_backup(rs, &rs_layout);
3057
3058        r = analyse_superblocks(ti, rs);
3059        if (r)
3060                goto bad;
3061
3062        /* All in-core metadata now as of current superblocks after calling analyse_superblocks() */
3063        sb_array_sectors = rs->md.array_sectors;
3064        rdev_sectors = __rdev_sectors(rs);
3065        if (!rdev_sectors) {
3066                ti->error = "Invalid rdev size";
3067                r = -EINVAL;
3068                goto bad;
3069        }
3070
3071
3072        reshape_sectors = _get_reshape_sectors(rs);
3073        if (rs->dev_sectors != rdev_sectors) {
3074                resize = (rs->dev_sectors != rdev_sectors - reshape_sectors);
3075                if (rs->dev_sectors > rdev_sectors - reshape_sectors)
3076                        set_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3077        }
3078
3079        INIT_WORK(&rs->md.event_work, do_table_event);
3080        ti->private = rs;
3081        ti->num_flush_bios = 1;
3082
3083        /* Restore any requested new layout for conversion decision */
3084        rs_config_restore(rs, &rs_layout);
3085
3086        /*
3087         * Now that we have any superblock metadata available,
3088         * check for new, recovering, reshaping, to be taken over,
3089         * to be reshaped or an existing, unchanged raid set to
3090         * run in sequence.
3091         */
3092        if (test_bit(MD_ARRAY_FIRST_USE, &rs->md.flags)) {
3093                /* A new raid6 set has to be recovered to ensure proper parity and Q-Syndrome */
3094                if (rs_is_raid6(rs) &&
3095                    test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags)) {
3096                        ti->error = "'nosync' not allowed for new raid6 set";
3097                        r = -EINVAL;
3098                        goto bad;
3099                }
3100                rs_setup_recovery(rs, 0);
3101                set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3102                rs_set_new(rs);
3103        } else if (rs_is_recovering(rs)) {
3104                /* A recovering raid set may be resized */
3105                goto size_check;
3106        } else if (rs_is_reshaping(rs)) {
3107                /* Have to reject size change request during reshape */
3108                if (resize) {
3109                        ti->error = "Can't resize a reshaping raid set";
3110                        r = -EPERM;
3111                        goto bad;
3112                }
3113                /* skip setup rs */
3114        } else if (rs_takeover_requested(rs)) {
3115                if (rs_is_reshaping(rs)) {
3116                        ti->error = "Can't takeover a reshaping raid set";
3117                        r = -EPERM;
3118                        goto bad;
3119                }
3120
3121                /* We can't takeover a journaled raid4/5/6 */
3122                if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3123                        ti->error = "Can't takeover a journaled raid4/5/6 set";
3124                        r = -EPERM;
3125                        goto bad;
3126                }
3127
3128                /*
3129                 * If a takeover is needed, userspace sets any additional
3130                 * devices to rebuild and we can check for a valid request here.
3131                 *
3132                 * If acceptible, set the level to the new requested
3133                 * one, prohibit requesting recovery, allow the raid
3134                 * set to run and store superblocks during resume.
3135                 */
3136                r = rs_check_takeover(rs);
3137                if (r)
3138                        goto bad;
3139
3140                r = rs_setup_takeover(rs);
3141                if (r)
3142                        goto bad;
3143
3144                set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3145                /* Takeover ain't recovery, so disable recovery */
3146                rs_setup_recovery(rs, MaxSector);
3147                rs_set_new(rs);
3148        } else if (rs_reshape_requested(rs)) {
3149                /* Only request grow on raid set size extensions, not on reshapes. */
3150                clear_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3151
3152                /*
3153                 * No need to check for 'ongoing' takeover here, because takeover
3154                 * is an instant operation as oposed to an ongoing reshape.
3155                 */
3156
3157                /* We can't reshape a journaled raid4/5/6 */
3158                if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags)) {
3159                        ti->error = "Can't reshape a journaled raid4/5/6 set";
3160                        r = -EPERM;
3161                        goto bad;
3162                }
3163
3164                /* Out-of-place space has to be available to allow for a reshape unless raid1! */
3165                if (reshape_sectors || rs_is_raid1(rs)) {
3166                        /*
3167                          * We can only prepare for a reshape here, because the
3168                          * raid set needs to run to provide the repective reshape
3169                          * check functions via its MD personality instance.
3170                          *
3171                          * So do the reshape check after md_run() succeeded.
3172                          */
3173                        r = rs_prepare_reshape(rs);
3174                        if (r)
3175                                goto bad;
3176
3177                        /* Reshaping ain't recovery, so disable recovery */
3178                        rs_setup_recovery(rs, MaxSector);
3179                }
3180                rs_set_cur(rs);
3181        } else {
3182size_check:
3183                /* May not set recovery when a device rebuild is requested */
3184                if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags)) {
3185                        clear_bit(RT_FLAG_RS_GROW, &rs->runtime_flags);
3186                        set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3187                        rs_setup_recovery(rs, MaxSector);
3188                } else if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags)) {
3189                        /*
3190                         * Set raid set to current size, i.e. size as of
3191                         * superblocks to grow to larger size in preresume.
3192                         */
3193                        r = rs_set_dev_and_array_sectors(rs, sb_array_sectors, false);
3194                        if (r)
3195                                goto bad;
3196
3197                        rs_setup_recovery(rs, rs->md.recovery_cp < rs->md.dev_sectors ? rs->md.recovery_cp : rs->md.dev_sectors);
3198                } else {
3199                        /* This is no size change or it is shrinking, update size and record in superblocks */
3200                        r = rs_set_dev_and_array_sectors(rs, rs->ti->len, false);
3201                        if (r)
3202                                goto bad;
3203
3204                        if (sb_array_sectors > rs->array_sectors)
3205                                set_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags);
3206                }
3207                rs_set_cur(rs);
3208        }
3209
3210        /* If constructor requested it, change data and new_data offsets */
3211        r = rs_adjust_data_offsets(rs);
3212        if (r)
3213                goto bad;
3214
3215        /* Start raid set read-only and assumed clean to change in raid_resume() */
3216        rs->md.ro = 1;
3217        rs->md.in_sync = 1;
3218
3219        /* Keep array frozen */
3220        set_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
3221
3222        /* Has to be held on running the array */
3223        mddev_lock_nointr(&rs->md);
3224        r = md_run(&rs->md);
3225        rs->md.in_sync = 0; /* Assume already marked dirty */
3226        if (r) {
3227                ti->error = "Failed to run raid array";
3228                mddev_unlock(&rs->md);
3229                goto bad;
3230        }
3231
3232        r = md_start(&rs->md);
3233
3234        if (r) {
3235                ti->error = "Failed to start raid array";
3236                mddev_unlock(&rs->md);
3237                goto bad_md_start;
3238        }
3239
3240        /* If raid4/5/6 journal mode explicitly requested (only possible with journal dev) -> set it */
3241        if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags)) {
3242                r = r5c_journal_mode_set(&rs->md, rs->journal_dev.mode);
3243                if (r) {
3244                        ti->error = "Failed to set raid4/5/6 journal mode";
3245                        mddev_unlock(&rs->md);
3246                        goto bad_journal_mode_set;
3247                }
3248        }
3249
3250        mddev_suspend(&rs->md);
3251        set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags);
3252
3253        /* Try to adjust the raid4/5/6 stripe cache size to the stripe size */
3254        if (rs_is_raid456(rs)) {
3255                r = rs_set_raid456_stripe_cache(rs);
3256                if (r)
3257                        goto bad_stripe_cache;
3258        }
3259
3260        /* Now do an early reshape check */
3261        if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3262                r = rs_check_reshape(rs);
3263                if (r)
3264                        goto bad_check_reshape;
3265
3266                /* Restore new, ctr requested layout to perform check */
3267                rs_config_restore(rs, &rs_layout);
3268
3269                if (rs->md.pers->start_reshape) {
3270                        r = rs->md.pers->check_reshape(&rs->md);
3271                        if (r) {
3272                                ti->error = "Reshape check failed";
3273                                goto bad_check_reshape;
3274                        }
3275                }
3276        }
3277
3278        /* Disable/enable discard support on raid set. */
3279        configure_discard_support(rs);
3280
3281        mddev_unlock(&rs->md);
3282        return 0;
3283
3284bad_md_start:
3285bad_journal_mode_set:
3286bad_stripe_cache:
3287bad_check_reshape:
3288        md_stop(&rs->md);
3289bad:
3290        raid_set_free(rs);
3291
3292        return r;
3293}
3294
3295static void raid_dtr(struct dm_target *ti)
3296{
3297        struct raid_set *rs = ti->private;
3298
3299        md_stop(&rs->md);
3300        raid_set_free(rs);
3301}
3302
3303static int raid_map(struct dm_target *ti, struct bio *bio)
3304{
3305        struct raid_set *rs = ti->private;
3306        struct mddev *mddev = &rs->md;
3307
3308        /*
3309         * If we're reshaping to add disk(s)), ti->len and
3310         * mddev->array_sectors will differ during the process
3311         * (ti->len > mddev->array_sectors), so we have to requeue
3312         * bios with addresses > mddev->array_sectors here or
3313         * there will occur accesses past EOD of the component
3314         * data images thus erroring the raid set.
3315         */
3316        if (unlikely(bio_end_sector(bio) > mddev->array_sectors))
3317                return DM_MAPIO_REQUEUE;
3318
3319        md_handle_request(mddev, bio);
3320
3321        return DM_MAPIO_SUBMITTED;
3322}
3323
3324/* Return sync state string for @state */
3325enum sync_state { st_frozen, st_reshape, st_resync, st_check, st_repair, st_recover, st_idle };
3326static const char *sync_str(enum sync_state state)
3327{
3328        /* Has to be in above sync_state order! */
3329        static const char *sync_strs[] = {
3330                "frozen",
3331                "reshape",
3332                "resync",
3333                "check",
3334                "repair",
3335                "recover",
3336                "idle"
3337        };
3338
3339        return __within_range(state, 0, ARRAY_SIZE(sync_strs) - 1) ? sync_strs[state] : "undef";
3340};
3341
3342/* Return enum sync_state for @mddev derived from @recovery flags */
3343static enum sync_state decipher_sync_action(struct mddev *mddev, unsigned long recovery)
3344{
3345        if (test_bit(MD_RECOVERY_FROZEN, &recovery))
3346                return st_frozen;
3347
3348        /* The MD sync thread can be done with io or be interrupted but still be running */
3349        if (!test_bit(MD_RECOVERY_DONE, &recovery) &&
3350            (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
3351             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery)))) {
3352                if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
3353                        return st_reshape;
3354
3355                if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
3356                        if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
3357                                return st_resync;
3358                        if (test_bit(MD_RECOVERY_CHECK, &recovery))
3359                                return st_check;
3360                        return st_repair;
3361                }
3362
3363                if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3364                        return st_recover;
3365
3366                if (mddev->reshape_position != MaxSector)
3367                        return st_reshape;
3368        }
3369
3370        return st_idle;
3371}
3372
3373/*
3374 * Return status string for @rdev
3375 *
3376 * Status characters:
3377 *
3378 *  'D' = Dead/Failed raid set component or raid4/5/6 journal device
3379 *  'a' = Alive but not in-sync raid set component _or_ alive raid4/5/6 'write_back' journal device
3380 *  'A' = Alive and in-sync raid set component _or_ alive raid4/5/6 'write_through' journal device
3381 *  '-' = Non-existing device (i.e. uspace passed '- -' into the ctr)
3382 */
3383static const char *__raid_dev_status(struct raid_set *rs, struct md_rdev *rdev)
3384{
3385        if (!rdev->bdev)
3386                return "-";
3387        else if (test_bit(Faulty, &rdev->flags))
3388                return "D";
3389        else if (test_bit(Journal, &rdev->flags))
3390                return (rs->journal_dev.mode == R5C_JOURNAL_MODE_WRITE_THROUGH) ? "A" : "a";
3391        else if (test_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags) ||
3392                 (!test_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags) &&
3393                  !test_bit(In_sync, &rdev->flags)))
3394                return "a";
3395        else
3396                return "A";
3397}
3398
3399/* Helper to return resync/reshape progress for @rs and runtime flags for raid set in sync / resynching */
3400static sector_t rs_get_progress(struct raid_set *rs, unsigned long recovery,
3401                                enum sync_state state, sector_t resync_max_sectors)
3402{
3403        sector_t r;
3404        struct mddev *mddev = &rs->md;
3405
3406        clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3407        clear_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3408
3409        if (rs_is_raid0(rs)) {
3410                r = resync_max_sectors;
3411                set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3412
3413        } else {
3414                if (state == st_idle && !test_bit(MD_RECOVERY_INTR, &recovery))
3415                        r = mddev->recovery_cp;
3416                else
3417                        r = mddev->curr_resync_completed;
3418
3419                if (state == st_idle && r >= resync_max_sectors) {
3420                        /*
3421                         * Sync complete.
3422                         */
3423                        /* In case we have finished recovering, the array is in sync. */
3424                        if (test_bit(MD_RECOVERY_RECOVER, &recovery))
3425                                set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3426
3427                } else if (state == st_recover)
3428                        /*
3429                         * In case we are recovering, the array is not in sync
3430                         * and health chars should show the recovering legs.
3431                         *
3432                         * Already retrieved recovery offset from curr_resync_completed above.
3433                         */
3434                        ;
3435
3436                else if (state == st_resync || state == st_reshape)
3437                        /*
3438                         * If "resync/reshape" is occurring, the raid set
3439                         * is or may be out of sync hence the health
3440                         * characters shall be 'a'.
3441                         */
3442                        set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3443
3444                else if (state == st_check || state == st_repair)
3445                        /*
3446                         * If "check" or "repair" is occurring, the raid set has
3447                         * undergone an initial sync and the health characters
3448                         * should not be 'a' anymore.
3449                         */
3450                        set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3451
3452                else if (test_bit(MD_RECOVERY_NEEDED, &recovery))
3453                        /*
3454                         * We are idle and recovery is needed, prevent 'A' chars race
3455                         * caused by components still set to in-sync by constructor.
3456                         */
3457                        set_bit(RT_FLAG_RS_RESYNCING, &rs->runtime_flags);
3458
3459                else {
3460                        /*
3461                         * We are idle and the raid set may be doing an initial
3462                         * sync, or it may be rebuilding individual components.
3463                         * If all the devices are In_sync, then it is the raid set
3464                         * that is being initialized.
3465                         */
3466                        struct md_rdev *rdev;
3467
3468                        set_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3469                        rdev_for_each(rdev, mddev)
3470                                if (!test_bit(Journal, &rdev->flags) &&
3471                                    !test_bit(In_sync, &rdev->flags)) {
3472                                        clear_bit(RT_FLAG_RS_IN_SYNC, &rs->runtime_flags);
3473                                        break;
3474                                }
3475                }
3476        }
3477
3478        return min(r, resync_max_sectors);
3479}
3480
3481/* Helper to return @dev name or "-" if !@dev */
3482static const char *__get_dev_name(struct dm_dev *dev)
3483{
3484        return dev ? dev->name : "-";
3485}
3486
3487static void raid_status(struct dm_target *ti, status_type_t type,
3488                        unsigned int status_flags, char *result, unsigned int maxlen)
3489{
3490        struct raid_set *rs = ti->private;
3491        struct mddev *mddev = &rs->md;
3492        struct r5conf *conf = mddev->private;
3493        int i, max_nr_stripes = conf ? conf->max_nr_stripes : 0;
3494        unsigned long recovery;
3495        unsigned int raid_param_cnt = 1; /* at least 1 for chunksize */
3496        unsigned int sz = 0;
3497        unsigned int rebuild_writemostly_count = 0;
3498        sector_t progress, resync_max_sectors, resync_mismatches;
3499        enum sync_state state;
3500        struct raid_type *rt;
3501
3502        switch (type) {
3503        case STATUSTYPE_INFO:
3504                /* *Should* always succeed */
3505                rt = get_raid_type_by_ll(mddev->new_level, mddev->new_layout);
3506                if (!rt)
3507                        return;
3508
3509                DMEMIT("%s %d ", rt->name, mddev->raid_disks);
3510
3511                /* Access most recent mddev properties for status output */
3512                smp_rmb();
3513                /* Get sensible max sectors even if raid set not yet started */
3514                resync_max_sectors = test_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags) ?
3515                                      mddev->resync_max_sectors : mddev->dev_sectors;
3516                recovery = rs->md.recovery;
3517                state = decipher_sync_action(mddev, recovery);
3518                progress = rs_get_progress(rs, recovery, state, resync_max_sectors);
3519                resync_mismatches = (mddev->last_sync_action && !strcasecmp(mddev->last_sync_action, "check")) ?
3520                                    atomic64_read(&mddev->resync_mismatches) : 0;
3521
3522                /* HM FIXME: do we want another state char for raid0? It shows 'D'/'A'/'-' now */
3523                for (i = 0; i < rs->raid_disks; i++)
3524                        DMEMIT(__raid_dev_status(rs, &rs->dev[i].rdev));
3525
3526                /*
3527                 * In-sync/Reshape ratio:
3528                 *  The in-sync ratio shows the progress of:
3529                 *   - Initializing the raid set
3530                 *   - Rebuilding a subset of devices of the raid set
3531                 *  The user can distinguish between the two by referring
3532                 *  to the status characters.
3533                 *
3534                 *  The reshape ratio shows the progress of
3535                 *  changing the raid layout or the number of
3536                 *  disks of a raid set
3537                 */
3538                DMEMIT(" %llu/%llu", (unsigned long long) progress,
3539                                     (unsigned long long) resync_max_sectors);
3540
3541                /*
3542                 * v1.5.0+:
3543                 *
3544                 * Sync action:
3545                 *   See Documentation/admin-guide/device-mapper/dm-raid.rst for
3546                 *   information on each of these states.
3547                 */
3548                DMEMIT(" %s", sync_str(state));
3549
3550                /*
3551                 * v1.5.0+:
3552                 *
3553                 * resync_mismatches/mismatch_cnt
3554                 *   This field shows the number of discrepancies found when
3555                 *   performing a "check" of the raid set.
3556                 */
3557                DMEMIT(" %llu", (unsigned long long) resync_mismatches);
3558
3559                /*
3560                 * v1.9.0+:
3561                 *
3562                 * data_offset (needed for out of space reshaping)
3563                 *   This field shows the data offset into the data
3564                 *   image LV where the first stripes data starts.
3565                 *
3566                 * We keep data_offset equal on all raid disks of the set,
3567                 * so retrieving it from the first raid disk is sufficient.
3568                 */
3569                DMEMIT(" %llu", (unsigned long long) rs->dev[0].rdev.data_offset);
3570
3571                /*
3572                 * v1.10.0+:
3573                 */
3574                DMEMIT(" %s", test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags) ?
3575                              __raid_dev_status(rs, &rs->journal_dev.rdev) : "-");
3576                break;
3577
3578        case STATUSTYPE_TABLE:
3579                /* Report the table line string you would use to construct this raid set */
3580
3581                /*
3582                 * Count any rebuild or writemostly argument pairs and subtract the
3583                 * hweight count being added below of any rebuild and writemostly ctr flags.
3584                 */
3585                for (i = 0; i < rs->raid_disks; i++) {
3586                        rebuild_writemostly_count += (test_bit(i, (void *) rs->rebuild_disks) ? 2 : 0) +
3587                                                     (test_bit(WriteMostly, &rs->dev[i].rdev.flags) ? 2 : 0);
3588                }
3589                rebuild_writemostly_count -= (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags) ? 2 : 0) +
3590                                             (test_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags) ? 2 : 0);
3591                /* Calculate raid parameter count based on ^ rebuild/writemostly argument counts and ctr flags set. */
3592                raid_param_cnt += rebuild_writemostly_count +
3593                                  hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_NO_ARGS) +
3594                                  hweight32(rs->ctr_flags & CTR_FLAG_OPTIONS_ONE_ARG) * 2;
3595                /* Emit table line */
3596                /* This has to be in the documented order for userspace! */
3597                DMEMIT("%s %u %u", rs->raid_type->name, raid_param_cnt, mddev->new_chunk_sectors);
3598                if (test_bit(__CTR_FLAG_SYNC, &rs->ctr_flags))
3599                        DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_SYNC));
3600                if (test_bit(__CTR_FLAG_NOSYNC, &rs->ctr_flags))
3601                        DMEMIT(" %s", dm_raid_arg_name_by_flag(CTR_FLAG_NOSYNC));
3602                if (test_bit(__CTR_FLAG_REBUILD, &rs->ctr_flags))
3603                        for (i = 0; i < rs->raid_disks; i++)
3604                                if (test_bit(i, (void *) rs->rebuild_disks))
3605                                        DMEMIT(" %s %u", dm_raid_arg_name_by_flag(CTR_FLAG_REBUILD), i);
3606                if (test_bit(__CTR_FLAG_DAEMON_SLEEP, &rs->ctr_flags))
3607                        DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_DAEMON_SLEEP),
3608                                          mddev->bitmap_info.daemon_sleep);
3609                if (test_bit(__CTR_FLAG_MIN_RECOVERY_RATE, &rs->ctr_flags))
3610                        DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MIN_RECOVERY_RATE),
3611                                         mddev->sync_speed_min);
3612                if (test_bit(__CTR_FLAG_MAX_RECOVERY_RATE, &rs->ctr_flags))
3613                        DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_RECOVERY_RATE),
3614                                         mddev->sync_speed_max);
3615                if (test_bit(__CTR_FLAG_WRITE_MOSTLY, &rs->ctr_flags))
3616                        for (i = 0; i < rs->raid_disks; i++)
3617                                if (test_bit(WriteMostly, &rs->dev[i].rdev.flags))
3618                                        DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_WRITE_MOSTLY),
3619                                               rs->dev[i].rdev.raid_disk);
3620                if (test_bit(__CTR_FLAG_MAX_WRITE_BEHIND, &rs->ctr_flags))
3621                        DMEMIT(" %s %lu", dm_raid_arg_name_by_flag(CTR_FLAG_MAX_WRITE_BEHIND),
3622                                          mddev->bitmap_info.max_write_behind);
3623                if (test_bit(__CTR_FLAG_STRIPE_CACHE, &rs->ctr_flags))
3624                        DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_STRIPE_CACHE),
3625                                         max_nr_stripes);
3626                if (test_bit(__CTR_FLAG_REGION_SIZE, &rs->ctr_flags))
3627                        DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_REGION_SIZE),
3628                                           (unsigned long long) to_sector(mddev->bitmap_info.chunksize));
3629                if (test_bit(__CTR_FLAG_RAID10_COPIES, &rs->ctr_flags))
3630                        DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_COPIES),
3631                                         raid10_md_layout_to_copies(mddev->layout));
3632                if (test_bit(__CTR_FLAG_RAID10_FORMAT, &rs->ctr_flags))
3633                        DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_RAID10_FORMAT),
3634                                         raid10_md_layout_to_format(mddev->layout));
3635                if (test_bit(__CTR_FLAG_DELTA_DISKS, &rs->ctr_flags))
3636                        DMEMIT(" %s %d", dm_raid_arg_name_by_flag(CTR_FLAG_DELTA_DISKS),
3637                                         max(rs->delta_disks, mddev->delta_disks));
3638                if (test_bit(__CTR_FLAG_DATA_OFFSET, &rs->ctr_flags))
3639                        DMEMIT(" %s %llu", dm_raid_arg_name_by_flag(CTR_FLAG_DATA_OFFSET),
3640                                           (unsigned long long) rs->data_offset);
3641                if (test_bit(__CTR_FLAG_JOURNAL_DEV, &rs->ctr_flags))
3642                        DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_DEV),
3643                                        __get_dev_name(rs->journal_dev.dev));
3644                if (test_bit(__CTR_FLAG_JOURNAL_MODE, &rs->ctr_flags))
3645                        DMEMIT(" %s %s", dm_raid_arg_name_by_flag(CTR_FLAG_JOURNAL_MODE),
3646                                         md_journal_mode_to_dm_raid(rs->journal_dev.mode));
3647                DMEMIT(" %d", rs->raid_disks);
3648                for (i = 0; i < rs->raid_disks; i++)
3649                        DMEMIT(" %s %s", __get_dev_name(rs->dev[i].meta_dev),
3650                                         __get_dev_name(rs->dev[i].data_dev));
3651        }
3652}
3653
3654static int raid_message(struct dm_target *ti, unsigned int argc, char **argv,
3655                        char *result, unsigned maxlen)
3656{
3657        struct raid_set *rs = ti->private;
3658        struct mddev *mddev = &rs->md;
3659
3660        if (!mddev->pers || !mddev->pers->sync_request)
3661                return -EINVAL;
3662
3663        if (!strcasecmp(argv[0], "frozen"))
3664                set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3665        else
3666                clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3667
3668        if (!strcasecmp(argv[0], "idle") || !strcasecmp(argv[0], "frozen")) {
3669                if (mddev->sync_thread) {
3670                        set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3671                        md_reap_sync_thread(mddev);
3672                }
3673        } else if (decipher_sync_action(mddev, mddev->recovery) != st_idle)
3674                return -EBUSY;
3675        else if (!strcasecmp(argv[0], "resync"))
3676                ; /* MD_RECOVERY_NEEDED set below */
3677        else if (!strcasecmp(argv[0], "recover"))
3678                set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3679        else {
3680                if (!strcasecmp(argv[0], "check")) {
3681                        set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3682                        set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3683                        set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3684                } else if (!strcasecmp(argv[0], "repair")) {
3685                        set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3686                        set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3687                } else
3688                        return -EINVAL;
3689        }
3690        if (mddev->ro == 2) {
3691                /* A write to sync_action is enough to justify
3692                 * canceling read-auto mode
3693                 */
3694                mddev->ro = 0;
3695                if (!mddev->suspended && mddev->sync_thread)
3696                        md_wakeup_thread(mddev->sync_thread);
3697        }
3698        set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3699        if (!mddev->suspended && mddev->thread)
3700                md_wakeup_thread(mddev->thread);
3701
3702        return 0;
3703}
3704
3705static int raid_iterate_devices(struct dm_target *ti,
3706                                iterate_devices_callout_fn fn, void *data)
3707{
3708        struct raid_set *rs = ti->private;
3709        unsigned int i;
3710        int r = 0;
3711
3712        for (i = 0; !r && i < rs->md.raid_disks; i++)
3713                if (rs->dev[i].data_dev)
3714                        r = fn(ti,
3715                                 rs->dev[i].data_dev,
3716                                 0, /* No offset on data devs */
3717                                 rs->md.dev_sectors,
3718                                 data);
3719
3720        return r;
3721}
3722
3723static void raid_io_hints(struct dm_target *ti, struct queue_limits *limits)
3724{
3725        struct raid_set *rs = ti->private;
3726        unsigned int chunk_size_bytes = to_bytes(rs->md.chunk_sectors);
3727
3728        blk_limits_io_min(limits, chunk_size_bytes);
3729        blk_limits_io_opt(limits, chunk_size_bytes * mddev_data_stripes(rs));
3730
3731        /*
3732         * RAID0 and RAID10 personalities require bio splitting,
3733         * RAID1/4/5/6 don't and process large discard bios properly.
3734         */
3735        if (rs_is_raid0(rs) || rs_is_raid10(rs)) {
3736                limits->discard_granularity = chunk_size_bytes;
3737                limits->max_discard_sectors = rs->md.chunk_sectors;
3738        }
3739}
3740
3741static void raid_postsuspend(struct dm_target *ti)
3742{
3743        struct raid_set *rs = ti->private;
3744
3745        if (!test_and_set_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
3746                /* Writes have to be stopped before suspending to avoid deadlocks. */
3747                if (!test_bit(MD_RECOVERY_FROZEN, &rs->md.recovery))
3748                        md_stop_writes(&rs->md);
3749
3750                mddev_lock_nointr(&rs->md);
3751                mddev_suspend(&rs->md);
3752                mddev_unlock(&rs->md);
3753        }
3754}
3755
3756static void attempt_restore_of_faulty_devices(struct raid_set *rs)
3757{
3758        int i;
3759        uint64_t cleared_failed_devices[DISKS_ARRAY_ELEMS];
3760        unsigned long flags;
3761        bool cleared = false;
3762        struct dm_raid_superblock *sb;
3763        struct mddev *mddev = &rs->md;
3764        struct md_rdev *r;
3765
3766        /* RAID personalities have to provide hot add/remove methods or we need to bail out. */
3767        if (!mddev->pers || !mddev->pers->hot_add_disk || !mddev->pers->hot_remove_disk)
3768                return;
3769
3770        memset(cleared_failed_devices, 0, sizeof(cleared_failed_devices));
3771
3772        for (i = 0; i < mddev->raid_disks; i++) {
3773                r = &rs->dev[i].rdev;
3774                /* HM FIXME: enhance journal device recovery processing */
3775                if (test_bit(Journal, &r->flags))
3776                        continue;
3777
3778                if (test_bit(Faulty, &r->flags) &&
3779                    r->meta_bdev && !read_disk_sb(r, r->sb_size, true)) {
3780                        DMINFO("Faulty %s device #%d has readable super block."
3781                               "  Attempting to revive it.",
3782                               rs->raid_type->name, i);
3783
3784                        /*
3785                         * Faulty bit may be set, but sometimes the array can
3786                         * be suspended before the personalities can respond
3787                         * by removing the device from the array (i.e. calling
3788                         * 'hot_remove_disk').  If they haven't yet removed
3789                         * the failed device, its 'raid_disk' number will be
3790                         * '>= 0' - meaning we must call this function
3791                         * ourselves.
3792                         */
3793                        flags = r->flags;
3794                        clear_bit(In_sync, &r->flags); /* Mandatory for hot remove. */
3795                        if (r->raid_disk >= 0) {
3796                                if (mddev->pers->hot_remove_disk(mddev, r)) {
3797                                        /* Failed to revive this device, try next */
3798                                        r->flags = flags;
3799                                        continue;
3800                                }
3801                        } else
3802                                r->raid_disk = r->saved_raid_disk = i;
3803
3804                        clear_bit(Faulty, &r->flags);
3805                        clear_bit(WriteErrorSeen, &r->flags);
3806
3807                        if (mddev->pers->hot_add_disk(mddev, r)) {
3808                                /* Failed to revive this device, try next */
3809                                r->raid_disk = r->saved_raid_disk = -1;
3810                                r->flags = flags;
3811                        } else {
3812                                clear_bit(In_sync, &r->flags);
3813                                r->recovery_offset = 0;
3814                                set_bit(i, (void *) cleared_failed_devices);
3815                                cleared = true;
3816                        }
3817                }
3818        }
3819
3820        /* If any failed devices could be cleared, update all sbs failed_devices bits */
3821        if (cleared) {
3822                uint64_t failed_devices[DISKS_ARRAY_ELEMS];
3823
3824                rdev_for_each(r, &rs->md) {
3825                        if (test_bit(Journal, &r->flags))
3826                                continue;
3827
3828                        sb = page_address(r->sb_page);
3829                        sb_retrieve_failed_devices(sb, failed_devices);
3830
3831                        for (i = 0; i < DISKS_ARRAY_ELEMS; i++)
3832                                failed_devices[i] &= ~cleared_failed_devices[i];
3833
3834                        sb_update_failed_devices(sb, failed_devices);
3835                }
3836        }
3837}
3838
3839static int __load_dirty_region_bitmap(struct raid_set *rs)
3840{
3841        int r = 0;
3842
3843        /* Try loading the bitmap unless "raid0", which does not have one */
3844        if (!rs_is_raid0(rs) &&
3845            !test_and_set_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags)) {
3846                r = md_bitmap_load(&rs->md);
3847                if (r)
3848                        DMERR("Failed to load bitmap");
3849        }
3850
3851        return r;
3852}
3853
3854/* Enforce updating all superblocks */
3855static void rs_update_sbs(struct raid_set *rs)
3856{
3857        struct mddev *mddev = &rs->md;
3858        int ro = mddev->ro;
3859
3860        set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3861        mddev->ro = 0;
3862        md_update_sb(mddev, 1);
3863        mddev->ro = ro;
3864}
3865
3866/*
3867 * Reshape changes raid algorithm of @rs to new one within personality
3868 * (e.g. raid6_zr -> raid6_nc), changes stripe size, adds/removes
3869 * disks from a raid set thus growing/shrinking it or resizes the set
3870 *
3871 * Call mddev_lock_nointr() before!
3872 */
3873static int rs_start_reshape(struct raid_set *rs)
3874{
3875        int r;
3876        struct mddev *mddev = &rs->md;
3877        struct md_personality *pers = mddev->pers;
3878
3879        /* Don't allow the sync thread to work until the table gets reloaded. */
3880        set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
3881
3882        r = rs_setup_reshape(rs);
3883        if (r)
3884                return r;
3885
3886        /*
3887         * Check any reshape constraints enforced by the personalility
3888         *
3889         * May as well already kick the reshape off so that * pers->start_reshape() becomes optional.
3890         */
3891        r = pers->check_reshape(mddev);
3892        if (r) {
3893                rs->ti->error = "pers->check_reshape() failed";
3894                return r;
3895        }
3896
3897        /*
3898         * Personality may not provide start reshape method in which
3899         * case check_reshape above has already covered everything
3900         */
3901        if (pers->start_reshape) {
3902                r = pers->start_reshape(mddev);
3903                if (r) {
3904                        rs->ti->error = "pers->start_reshape() failed";
3905                        return r;
3906                }
3907        }
3908
3909        /*
3910         * Now reshape got set up, update superblocks to
3911         * reflect the fact so that a table reload will
3912         * access proper superblock content in the ctr.
3913         */
3914        rs_update_sbs(rs);
3915
3916        return 0;
3917}
3918
3919static int raid_preresume(struct dm_target *ti)
3920{
3921        int r;
3922        struct raid_set *rs = ti->private;
3923        struct mddev *mddev = &rs->md;
3924
3925        /* This is a resume after a suspend of the set -> it's already started. */
3926        if (test_and_set_bit(RT_FLAG_RS_PRERESUMED, &rs->runtime_flags))
3927                return 0;
3928
3929        /*
3930         * The superblocks need to be updated on disk if the
3931         * array is new or new devices got added (thus zeroed
3932         * out by userspace) or __load_dirty_region_bitmap
3933         * will overwrite them in core with old data or fail.
3934         */
3935        if (test_bit(RT_FLAG_UPDATE_SBS, &rs->runtime_flags))
3936                rs_update_sbs(rs);
3937
3938        /* Load the bitmap from disk unless raid0 */
3939        r = __load_dirty_region_bitmap(rs);
3940        if (r)
3941                return r;
3942
3943        /* We are extending the raid set size, adjust mddev/md_rdev sizes and set capacity. */
3944        if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags)) {
3945                mddev->array_sectors = rs->array_sectors;
3946                mddev->dev_sectors = rs->dev_sectors;
3947                rs_set_rdev_sectors(rs);
3948                rs_set_capacity(rs);
3949        }
3950
3951        /* Resize bitmap to adjust to changed region size (aka MD bitmap chunksize) or grown device size */
3952        if (test_bit(RT_FLAG_RS_BITMAP_LOADED, &rs->runtime_flags) && mddev->bitmap &&
3953            (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags) ||
3954             (rs->requested_bitmap_chunk_sectors &&
3955               mddev->bitmap_info.chunksize != to_bytes(rs->requested_bitmap_chunk_sectors)))) {
3956                int chunksize = to_bytes(rs->requested_bitmap_chunk_sectors) ?: mddev->bitmap_info.chunksize;
3957
3958                r = md_bitmap_resize(mddev->bitmap, mddev->dev_sectors, chunksize, 0);
3959                if (r)
3960                        DMERR("Failed to resize bitmap");
3961        }
3962
3963        /* Check for any resize/reshape on @rs and adjust/initiate */
3964        /* Be prepared for mddev_resume() in raid_resume() */
3965        set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3966        if (mddev->recovery_cp && mddev->recovery_cp < MaxSector) {
3967                set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3968                mddev->resync_min = mddev->recovery_cp;
3969                if (test_bit(RT_FLAG_RS_GROW, &rs->runtime_flags))
3970                        mddev->resync_max_sectors = mddev->dev_sectors;
3971        }
3972
3973        /* Check for any reshape request unless new raid set */
3974        if (test_bit(RT_FLAG_RESHAPE_RS, &rs->runtime_flags)) {
3975                /* Initiate a reshape. */
3976                rs_set_rdev_sectors(rs);
3977                mddev_lock_nointr(mddev);
3978                r = rs_start_reshape(rs);
3979                mddev_unlock(mddev);
3980                if (r)
3981                        DMWARN("Failed to check/start reshape, continuing without change");
3982                r = 0;
3983        }
3984
3985        return r;
3986}
3987
3988static void raid_resume(struct dm_target *ti)
3989{
3990        struct raid_set *rs = ti->private;
3991        struct mddev *mddev = &rs->md;
3992
3993        if (test_and_set_bit(RT_FLAG_RS_RESUMED, &rs->runtime_flags)) {
3994                /*
3995                 * A secondary resume while the device is active.
3996                 * Take this opportunity to check whether any failed
3997                 * devices are reachable again.
3998                 */
3999                attempt_restore_of_faulty_devices(rs);
4000        }
4001
4002        if (test_and_clear_bit(RT_FLAG_RS_SUSPENDED, &rs->runtime_flags)) {
4003                /* Only reduce raid set size before running a disk removing reshape. */
4004                if (mddev->delta_disks < 0)
4005                        rs_set_capacity(rs);
4006
4007                mddev_lock_nointr(mddev);
4008                clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4009                mddev->ro = 0;
4010                mddev->in_sync = 0;
4011                mddev_resume(mddev);
4012                mddev_unlock(mddev);
4013        }
4014}
4015
4016static struct target_type raid_target = {
4017        .name = "raid",
4018        .version = {1, 15, 1},
4019        .module = THIS_MODULE,
4020        .ctr = raid_ctr,
4021        .dtr = raid_dtr,
4022        .map = raid_map,
4023        .status = raid_status,
4024        .message = raid_message,
4025        .iterate_devices = raid_iterate_devices,
4026        .io_hints = raid_io_hints,
4027        .postsuspend = raid_postsuspend,
4028        .preresume = raid_preresume,
4029        .resume = raid_resume,
4030};
4031
4032static int __init dm_raid_init(void)
4033{
4034        DMINFO("Loading target version %u.%u.%u",
4035               raid_target.version[0],
4036               raid_target.version[1],
4037               raid_target.version[2]);
4038        return dm_register_target(&raid_target);
4039}
4040
4041static void __exit dm_raid_exit(void)
4042{
4043        dm_unregister_target(&raid_target);
4044}
4045
4046module_init(dm_raid_init);
4047module_exit(dm_raid_exit);
4048
4049module_param(devices_handle_discard_safely, bool, 0644);
4050MODULE_PARM_DESC(devices_handle_discard_safely,
4051                 "Set to Y if all devices in each array reliably return zeroes on reads from discarded regions");
4052
4053MODULE_DESCRIPTION(DM_NAME " raid0/1/10/4/5/6 target");
4054MODULE_ALIAS("dm-raid0");
4055MODULE_ALIAS("dm-raid1");
4056MODULE_ALIAS("dm-raid10");
4057MODULE_ALIAS("dm-raid4");
4058MODULE_ALIAS("dm-raid5");
4059MODULE_ALIAS("dm-raid6");
4060MODULE_AUTHOR("Neil Brown <dm-devel@redhat.com>");
4061MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
4062MODULE_LICENSE("GPL");
4063