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