md_intel.c revision 220210
1/*- 2 * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27#include <sys/cdefs.h> 28__FBSDID("$FreeBSD: head/sys/geom/raid/md_intel.c 220210 2011-03-31 16:19:53Z mav $"); 29 30#include <sys/param.h> 31#include <sys/bio.h> 32#include <sys/endian.h> 33#include <sys/kernel.h> 34#include <sys/kobj.h> 35#include <sys/limits.h> 36#include <sys/lock.h> 37#include <sys/malloc.h> 38#include <sys/mutex.h> 39#include <sys/systm.h> 40#include <sys/taskqueue.h> 41#include <geom/geom.h> 42#include "geom/raid/g_raid.h" 43#include "g_raid_md_if.h" 44 45static MALLOC_DEFINE(M_MD_INTEL, "md_intel_data", "GEOM_RAID Intel metadata"); 46 47struct intel_raid_map { 48 uint32_t offset; 49 uint32_t disk_sectors; 50 uint32_t stripe_count; 51 uint16_t strip_sectors; 52 uint8_t status; 53#define INTEL_S_READY 0x00 54#define INTEL_S_UNINITIALIZED 0x01 55#define INTEL_S_DEGRADED 0x02 56#define INTEL_S_FAILURE 0x03 57 58 uint8_t type; 59#define INTEL_T_RAID0 0x00 60#define INTEL_T_RAID1 0x01 61#define INTEL_T_RAID5 0x05 62 63 uint8_t total_disks; 64 uint8_t total_domains; 65 uint8_t failed_disk_num; 66 uint8_t ddf; 67 uint32_t filler_2[7]; 68 uint32_t disk_idx[1]; /* total_disks entries. */ 69#define INTEL_DI_IDX 0x00ffffff 70#define INTEL_DI_RBLD 0x01000000 71} __packed; 72 73struct intel_raid_vol { 74 uint8_t name[16]; 75 u_int64_t total_sectors __packed; 76 uint32_t state; 77#define INTEL_ST_BOOTABLE 0x00000001 78#define INTEL_ST_BOOT_DEVICE 0x00000002 79#define INTEL_ST_READ_COALESCING 0x00000004 80#define INTEL_ST_WRITE_COALESCING 0x00000008 81#define INTEL_ST_LAST_SHUTDOWN_DIRTY 0x00000010 82#define INTEL_ST_HIDDEN_AT_BOOT 0x00000020 83#define INTEL_ST_CURRENTLY_HIDDEN 0x00000040 84#define INTEL_ST_VERIFY_AND_FIX 0x00000080 85#define INTEL_ST_MAP_STATE_UNINIT 0x00000100 86#define INTEL_ST_NO_AUTO_RECOVERY 0x00000200 87#define INTEL_ST_CLONE_N_GO 0x00000400 88#define INTEL_ST_CLONE_MAN_SYNC 0x00000800 89#define INTEL_ST_CNG_MASTER_DISK_NUM 0x00001000 90 uint32_t reserved; 91 uint8_t migr_priority; 92 uint8_t num_sub_vols; 93 uint8_t tid; 94 uint8_t cng_master_disk; 95 uint16_t cache_policy; 96 uint8_t cng_state; 97 uint8_t cng_sub_state; 98 uint32_t filler_0[10]; 99 100 uint32_t curr_migr_unit; 101 uint32_t checkpoint_id; 102 uint8_t migr_state; 103 uint8_t migr_type; 104#define INTEL_MT_INIT 0 105#define INTEL_MT_REBUILD 1 106#define INTEL_MT_VERIFY 2 107#define INTEL_MT_GEN_MIGR 3 108#define INTEL_MT_STATE_CHANGE 4 109#define INTEL_MT_REPAIR 5 110 uint8_t dirty; 111 uint8_t fs_state; 112 uint16_t verify_errors; 113 uint16_t bad_blocks; 114 uint32_t filler_1[4]; 115 struct intel_raid_map map[1]; /* 2 entries if migr_state != 0. */ 116} __packed; 117 118struct intel_raid_disk { 119#define INTEL_SERIAL_LEN 16 120 uint8_t serial[INTEL_SERIAL_LEN]; 121 uint32_t sectors; 122 uint32_t id; 123 uint32_t flags; 124#define INTEL_F_SPARE 0x01 125#define INTEL_F_ASSIGNED 0x02 126#define INTEL_F_FAILED 0x04 127#define INTEL_F_ONLINE 0x08 128 129 uint32_t filler[5]; 130} __packed; 131 132struct intel_raid_conf { 133 uint8_t intel_id[24]; 134#define INTEL_MAGIC "Intel Raid ISM Cfg Sig. " 135 136 uint8_t version[6]; 137#define INTEL_VERSION_1000 "1.0.00" /* RAID0 */ 138#define INTEL_VERSION_1100 "1.1.00" /* RAID1 */ 139#define INTEL_VERSION_1200 "1.2.00" /* Many volumes */ 140#define INTEL_VERSION_1201 "1.2.01" /* 3 or 4 disks */ 141#define INTEL_VERSION_1202 "1.2.02" /* RAID5 */ 142#define INTEL_VERSION_1204 "1.2.04" /* 5 or 6 disks */ 143#define INTEL_VERSION_1206 "1.2.06" /* CNG */ 144#define INTEL_VERSION_1300 "1.3.00" /* Attributes */ 145 146 uint8_t dummy_0[2]; 147 uint32_t checksum; 148 uint32_t config_size; 149 uint32_t config_id; 150 uint32_t generation; 151 uint32_t error_log_size; 152 uint32_t attributes; 153#define INTEL_ATTR_RAID0 0x00000001 154#define INTEL_ATTR_RAID1 0x00000002 155#define INTEL_ATTR_RAID10 0x00000004 156#define INTEL_ATTR_RAID1E 0x00000008 157#define INTEL_ATTR_RAID5 0x00000010 158#define INTEL_ATTR_RAIDCNG 0x00000020 159#define INTEL_ATTR_2TB 0x20000000 160#define INTEL_ATTR_PM 0x40000000 161#define INTEL_ATTR_CHECKSUM 0x80000000 162 163 uint8_t total_disks; 164 uint8_t total_volumes; 165 uint8_t dummy_2[2]; 166 uint32_t filler_0[39]; 167 struct intel_raid_disk disk[1]; /* total_disks entries. */ 168 /* Here goes total_volumes of struct intel_raid_vol. */ 169} __packed; 170 171#define INTEL_MAX_MD_SIZE(ndisks) \ 172 (sizeof(struct intel_raid_conf) + \ 173 sizeof(struct intel_raid_disk) * (ndisks - 1) + \ 174 sizeof(struct intel_raid_vol) * 2 + \ 175 sizeof(struct intel_raid_map) * 2 + \ 176 sizeof(uint32_t) * (ndisks - 1) * 4) 177 178struct g_raid_md_intel_perdisk { 179 struct intel_raid_conf *pd_meta; 180 int pd_disk_pos; 181 struct intel_raid_disk pd_disk_meta; 182}; 183 184struct g_raid_md_intel_object { 185 struct g_raid_md_object mdio_base; 186 uint32_t mdio_config_id; 187 uint32_t mdio_generation; 188 struct intel_raid_conf *mdio_meta; 189 struct callout mdio_start_co; /* STARTING state timer. */ 190 int mdio_disks_present; 191 int mdio_started; 192 int mdio_incomplete; 193 struct root_hold_token *mdio_rootmount; /* Root mount delay token. */ 194}; 195 196static g_raid_md_create_t g_raid_md_create_intel; 197static g_raid_md_taste_t g_raid_md_taste_intel; 198static g_raid_md_event_t g_raid_md_event_intel; 199static g_raid_md_ctl_t g_raid_md_ctl_intel; 200static g_raid_md_write_t g_raid_md_write_intel; 201static g_raid_md_fail_disk_t g_raid_md_fail_disk_intel; 202static g_raid_md_free_disk_t g_raid_md_free_disk_intel; 203static g_raid_md_free_t g_raid_md_free_intel; 204 205static kobj_method_t g_raid_md_intel_methods[] = { 206 KOBJMETHOD(g_raid_md_create, g_raid_md_create_intel), 207 KOBJMETHOD(g_raid_md_taste, g_raid_md_taste_intel), 208 KOBJMETHOD(g_raid_md_event, g_raid_md_event_intel), 209 KOBJMETHOD(g_raid_md_ctl, g_raid_md_ctl_intel), 210 KOBJMETHOD(g_raid_md_write, g_raid_md_write_intel), 211 KOBJMETHOD(g_raid_md_fail_disk, g_raid_md_fail_disk_intel), 212 KOBJMETHOD(g_raid_md_free_disk, g_raid_md_free_disk_intel), 213 KOBJMETHOD(g_raid_md_free, g_raid_md_free_intel), 214 { 0, 0 } 215}; 216 217static struct g_raid_md_class g_raid_md_intel_class = { 218 "Intel", 219 g_raid_md_intel_methods, 220 sizeof(struct g_raid_md_intel_object), 221 .mdc_priority = 100 222}; 223 224 225static struct intel_raid_map * 226intel_get_map(struct intel_raid_vol *mvol, int i) 227{ 228 struct intel_raid_map *mmap; 229 230 if (i > (mvol->migr_state ? 1 : 0)) 231 return (NULL); 232 mmap = &mvol->map[0]; 233 for (; i > 0; i--) { 234 mmap = (struct intel_raid_map *) 235 &mmap->disk_idx[mmap->total_disks]; 236 } 237 return ((struct intel_raid_map *)mmap); 238} 239 240static struct intel_raid_vol * 241intel_get_volume(struct intel_raid_conf *meta, int i) 242{ 243 struct intel_raid_vol *mvol; 244 struct intel_raid_map *mmap; 245 246 if (i > 1) 247 return (NULL); 248 mvol = (struct intel_raid_vol *)&meta->disk[meta->total_disks]; 249 for (; i > 0; i--) { 250 mmap = intel_get_map(mvol, mvol->migr_state ? 1 : 0); 251 mvol = (struct intel_raid_vol *) 252 &mmap->disk_idx[mmap->total_disks]; 253 } 254 return (mvol); 255} 256 257static void 258g_raid_md_intel_print(struct intel_raid_conf *meta) 259{ 260 struct intel_raid_vol *mvol; 261 struct intel_raid_map *mmap; 262 int i, j, k; 263 264 if (g_raid_debug < 1) 265 return; 266 267 printf("********* ATA Intel MatrixRAID Metadata *********\n"); 268 printf("intel_id <%.24s>\n", meta->intel_id); 269 printf("version <%.6s>\n", meta->version); 270 printf("checksum 0x%08x\n", meta->checksum); 271 printf("config_size 0x%08x\n", meta->config_size); 272 printf("config_id 0x%08x\n", meta->config_id); 273 printf("generation 0x%08x\n", meta->generation); 274 printf("attributes 0x%08x\n", meta->attributes); 275 printf("total_disks %u\n", meta->total_disks); 276 printf("total_volumes %u\n", meta->total_volumes); 277 printf("DISK# serial disk_sectors disk_id flags\n"); 278 for (i = 0; i < meta->total_disks; i++ ) { 279 printf(" %d <%.16s> %u 0x%08x 0x%08x\n", i, 280 meta->disk[i].serial, meta->disk[i].sectors, 281 meta->disk[i].id, meta->disk[i].flags); 282 } 283 for (i = 0; i < meta->total_volumes; i++) { 284 mvol = intel_get_volume(meta, i); 285 printf(" ****** Volume %d ******\n", i); 286 printf(" name %.16s\n", mvol->name); 287 printf(" total_sectors %ju\n", mvol->total_sectors); 288 printf(" state %u\n", mvol->state); 289 printf(" reserved %u\n", mvol->reserved); 290 printf(" curr_migr_unit %u\n", mvol->curr_migr_unit); 291 printf(" checkpoint_id %u\n", mvol->checkpoint_id); 292 printf(" migr_state %u\n", mvol->migr_state); 293 printf(" migr_type %u\n", mvol->migr_type); 294 printf(" dirty %u\n", mvol->dirty); 295 296 for (j = 0; j < (mvol->migr_state ? 2 : 1); j++) { 297 printf(" *** Map %d ***\n", j); 298 mmap = intel_get_map(mvol, j); 299 printf(" offset %u\n", mmap->offset); 300 printf(" disk_sectors %u\n", mmap->disk_sectors); 301 printf(" stripe_count %u\n", mmap->stripe_count); 302 printf(" strip_sectors %u\n", mmap->strip_sectors); 303 printf(" status %u\n", mmap->status); 304 printf(" type %u\n", mmap->type); 305 printf(" total_disks %u\n", mmap->total_disks); 306 printf(" total_domains %u\n", mmap->total_domains); 307 printf(" failed_disk_num %u\n", mmap->failed_disk_num); 308 printf(" ddf %u\n", mmap->ddf); 309 printf(" disk_idx "); 310 for (k = 0; k < mmap->total_disks; k++) 311 printf(" 0x%08x", mmap->disk_idx[k]); 312 printf("\n"); 313 } 314 } 315 printf("=================================================\n"); 316} 317 318static struct intel_raid_conf * 319intel_meta_copy(struct intel_raid_conf *meta) 320{ 321 struct intel_raid_conf *nmeta; 322 323 nmeta = malloc(meta->config_size, M_MD_INTEL, M_WAITOK); 324 memcpy(nmeta, meta, meta->config_size); 325 return (nmeta); 326} 327 328static int 329intel_meta_find_disk(struct intel_raid_conf *meta, char *serial) 330{ 331 int pos; 332 333 for (pos = 0; pos < meta->total_disks; pos++) { 334 if (strncmp(meta->disk[pos].serial, 335 serial, INTEL_SERIAL_LEN) == 0) 336 return (pos); 337 } 338 return (-1); 339} 340 341static struct intel_raid_conf * 342intel_meta_read(struct g_consumer *cp) 343{ 344 struct g_provider *pp; 345 struct intel_raid_conf *meta; 346 struct intel_raid_vol *mvol; 347 struct intel_raid_map *mmap; 348 char *buf; 349 int error, i, j, k, left, size; 350 uint32_t checksum, *ptr; 351 352 pp = cp->provider; 353 354 /* Read the anchor sector. */ 355 buf = g_read_data(cp, 356 pp->mediasize - pp->sectorsize * 2, pp->sectorsize, &error); 357 if (buf == NULL) { 358 G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).", 359 pp->name, error); 360 return (NULL); 361 } 362 meta = (struct intel_raid_conf *)buf; 363 364 /* Check if this is an Intel RAID struct */ 365 if (strncmp(meta->intel_id, INTEL_MAGIC, strlen(INTEL_MAGIC))) { 366 G_RAID_DEBUG(1, "Intel signature check failed on %s", pp->name); 367 g_free(buf); 368 return (NULL); 369 } 370 if (meta->config_size > 65536 || 371 meta->config_size < sizeof(struct intel_raid_conf)) { 372 G_RAID_DEBUG(1, "Intel metadata size looks wrong: %d", 373 meta->config_size); 374 g_free(buf); 375 return (NULL); 376 } 377 size = meta->config_size; 378 meta = malloc(size, M_MD_INTEL, M_WAITOK); 379 memcpy(meta, buf, min(size, pp->sectorsize)); 380 g_free(buf); 381 382 /* Read all the rest, if needed. */ 383 if (meta->config_size > pp->sectorsize) { 384 left = (meta->config_size - 1) / pp->sectorsize; 385 buf = g_read_data(cp, 386 pp->mediasize - pp->sectorsize * (2 + left), 387 pp->sectorsize * left, &error); 388 if (buf == NULL) { 389 G_RAID_DEBUG(1, "Cannot read remaining metadata" 390 " part from %s (error=%d).", 391 pp->name, error); 392 free(meta, M_MD_INTEL); 393 return (NULL); 394 } 395 memcpy(((char *)meta) + pp->sectorsize, buf, 396 pp->sectorsize * left); 397 g_free(buf); 398 } 399 400 /* Check metadata checksum. */ 401 for (checksum = 0, ptr = (uint32_t *)meta, i = 0; 402 i < (meta->config_size / sizeof(uint32_t)); i++) { 403 checksum += *ptr++; 404 } 405 checksum -= meta->checksum; 406 if (checksum != meta->checksum) { 407 G_RAID_DEBUG(1, "Intel checksum check failed on %s", pp->name); 408 free(meta, M_MD_INTEL); 409 return (NULL); 410 } 411 412 /* Validate metadata size. */ 413 size = sizeof(struct intel_raid_conf) + 414 sizeof(struct intel_raid_disk) * (meta->total_disks - 1) + 415 sizeof(struct intel_raid_vol) * meta->total_volumes; 416 if (size > meta->config_size) { 417badsize: 418 G_RAID_DEBUG(1, "Intel metadata size incorrect %d < %d", 419 meta->config_size, size); 420 free(meta, M_MD_INTEL); 421 return (NULL); 422 } 423 for (i = 0; i < meta->total_volumes; i++) { 424 mvol = intel_get_volume(meta, i); 425 mmap = intel_get_map(mvol, 0); 426 size += 4 * (mmap->total_disks - 1); 427 if (size > meta->config_size) 428 goto badsize; 429 if (mvol->migr_state) { 430 size += sizeof(struct intel_raid_map); 431 if (size > meta->config_size) 432 goto badsize; 433 mmap = intel_get_map(mvol, 1); 434 size += 4 * (mmap->total_disks - 1); 435 if (size > meta->config_size) 436 goto badsize; 437 } 438 } 439 440 /* Validate disk indexes. */ 441 for (i = 0; i < meta->total_volumes; i++) { 442 mvol = intel_get_volume(meta, i); 443 for (j = 0; j < (mvol->migr_state ? 2 : 1); j++) { 444 mmap = intel_get_map(mvol, j); 445 for (k = 0; k < mmap->total_disks; k++) { 446 if ((mmap->disk_idx[k] & INTEL_DI_IDX) > 447 meta->total_disks) { 448 G_RAID_DEBUG(1, "Intel metadata disk" 449 " index %d too big (>%d)", 450 mmap->disk_idx[k] & INTEL_DI_IDX, 451 meta->total_disks); 452 free(meta, M_MD_INTEL); 453 return (NULL); 454 } 455 } 456 } 457 } 458 459 /* Validate migration types. */ 460 for (i = 0; i < meta->total_volumes; i++) { 461 mvol = intel_get_volume(meta, i); 462 if (mvol->migr_state && 463 mvol->migr_type != INTEL_MT_INIT && 464 mvol->migr_type != INTEL_MT_REBUILD && 465 mvol->migr_type != INTEL_MT_VERIFY && 466 mvol->migr_type != INTEL_MT_REPAIR) { 467 G_RAID_DEBUG(1, "Intel metadata has unsupported" 468 " migration type %d", mvol->migr_type); 469 free(meta, M_MD_INTEL); 470 return (NULL); 471 } 472 } 473 474 return (meta); 475} 476 477static int 478intel_meta_write(struct g_consumer *cp, struct intel_raid_conf *meta) 479{ 480 struct g_provider *pp; 481 char *buf; 482 int error, i, sectors; 483 uint32_t checksum, *ptr; 484 485 pp = cp->provider; 486 487 /* Recalculate checksum for case if metadata were changed. */ 488 meta->checksum = 0; 489 for (checksum = 0, ptr = (uint32_t *)meta, i = 0; 490 i < (meta->config_size / sizeof(uint32_t)); i++) { 491 checksum += *ptr++; 492 } 493 meta->checksum = checksum; 494 495 /* Create and fill buffer. */ 496 sectors = (meta->config_size + pp->sectorsize - 1) / pp->sectorsize; 497 buf = malloc(sectors * pp->sectorsize, M_MD_INTEL, M_WAITOK | M_ZERO); 498 if (sectors > 1) { 499 memcpy(buf, ((char *)meta) + pp->sectorsize, 500 (sectors - 1) * pp->sectorsize); 501 } 502 memcpy(buf + (sectors - 1) * pp->sectorsize, meta, pp->sectorsize); 503 504 error = g_write_data(cp, 505 pp->mediasize - pp->sectorsize * (1 + sectors), 506 buf, pp->sectorsize * sectors); 507 if (error != 0) { 508 G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).", 509 pp->name, error); 510 } 511 512 free(buf, M_MD_INTEL); 513 return (error); 514} 515 516static int 517intel_meta_erase(struct g_consumer *cp) 518{ 519 struct g_provider *pp; 520 char *buf; 521 int error; 522 523 pp = cp->provider; 524 buf = malloc(pp->sectorsize, M_MD_INTEL, M_WAITOK | M_ZERO); 525 error = g_write_data(cp, 526 pp->mediasize - 2 * pp->sectorsize, 527 buf, pp->sectorsize); 528 if (error != 0) { 529 G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).", 530 pp->name, error); 531 } 532 free(buf, M_MD_INTEL); 533 return (error); 534} 535 536static int 537intel_meta_write_spare(struct g_consumer *cp, struct intel_raid_disk *d) 538{ 539 struct intel_raid_conf *meta; 540 int error; 541 542 /* Fill anchor and single disk. */ 543 meta = malloc(INTEL_MAX_MD_SIZE(1), M_MD_INTEL, M_WAITOK | M_ZERO); 544 memcpy(&meta->intel_id[0], INTEL_MAGIC, sizeof(INTEL_MAGIC) - 1); 545 memcpy(&meta->version[0], INTEL_VERSION_1000, 546 sizeof(INTEL_VERSION_1000) - 1); 547 meta->config_size = INTEL_MAX_MD_SIZE(1); 548 meta->config_id = arc4random(); 549 meta->generation = 1; 550 meta->total_disks = 1; 551 meta->disk[0] = *d; 552 error = intel_meta_write(cp, meta); 553 free(meta, M_MD_INTEL); 554 return (error); 555} 556 557static struct g_raid_disk * 558g_raid_md_intel_get_disk(struct g_raid_softc *sc, int id) 559{ 560 struct g_raid_disk *disk; 561 struct g_raid_md_intel_perdisk *pd; 562 563 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 564 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 565 if (pd->pd_disk_pos == id) 566 break; 567 } 568 return (disk); 569} 570 571static int 572g_raid_md_intel_supported(int level, int qual, int disks, int force) 573{ 574 575 switch (level) { 576 case G_RAID_VOLUME_RL_RAID0: 577 if (disks < 1) 578 return (0); 579 if (!force && (disks < 2 || disks > 6)) 580 return (0); 581 break; 582 case G_RAID_VOLUME_RL_RAID1: 583 if (disks < 1) 584 return (0); 585 if (!force && (disks != 2)) 586 return (0); 587 break; 588 case G_RAID_VOLUME_RL_RAID1E: 589 if (disks < 2) 590 return (0); 591 if (!force && (disks != 4)) 592 return (0); 593 break; 594 case G_RAID_VOLUME_RL_RAID5: 595 if (disks < 3) 596 return (0); 597 if (!force && disks > 6) 598 return (0); 599 break; 600 default: 601 return (0); 602 } 603 if (qual != G_RAID_VOLUME_RLQ_NONE) 604 return (0); 605 return (1); 606} 607 608static struct g_raid_volume * 609g_raid_md_intel_get_volume(struct g_raid_softc *sc, int id) 610{ 611 struct g_raid_volume *mvol; 612 613 TAILQ_FOREACH(mvol, &sc->sc_volumes, v_next) { 614 if ((intptr_t)(mvol->v_md_data) == id) 615 break; 616 } 617 return (mvol); 618} 619 620static int 621g_raid_md_intel_start_disk(struct g_raid_disk *disk) 622{ 623 struct g_raid_softc *sc; 624 struct g_raid_subdisk *sd, *tmpsd; 625 struct g_raid_disk *olddisk, *tmpdisk; 626 struct g_raid_md_object *md; 627 struct g_raid_md_intel_object *mdi; 628 struct g_raid_md_intel_perdisk *pd, *oldpd; 629 struct intel_raid_conf *meta; 630 struct intel_raid_vol *mvol; 631 struct intel_raid_map *mmap0, *mmap1; 632 int disk_pos, resurrection = 0; 633 634 sc = disk->d_softc; 635 md = sc->sc_md; 636 mdi = (struct g_raid_md_intel_object *)md; 637 meta = mdi->mdio_meta; 638 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 639 olddisk = NULL; 640 641 /* Find disk position in metadata by it's serial. */ 642 disk_pos = intel_meta_find_disk(meta, pd->pd_disk_meta.serial); 643 if (disk_pos < 0) { 644 G_RAID_DEBUG1(1, sc, "Unknown, probably new or stale disk"); 645 /* Failed stale disk is useless for us. */ 646 if (pd->pd_disk_meta.flags & INTEL_F_FAILED) { 647 g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE_FAILED); 648 return (0); 649 } 650 /* If we are in the start process, that's all for now. */ 651 if (!mdi->mdio_started) 652 goto nofit; 653 /* 654 * If we have already started - try to get use of the disk. 655 * Try to replace OFFLINE disks first, then FAILED. 656 */ 657 TAILQ_FOREACH(tmpdisk, &sc->sc_disks, d_next) { 658 if (tmpdisk->d_state != G_RAID_DISK_S_OFFLINE && 659 tmpdisk->d_state != G_RAID_DISK_S_FAILED) 660 continue; 661 /* Make sure this disk is big enough. */ 662 TAILQ_FOREACH(sd, &tmpdisk->d_subdisks, sd_next) { 663 if (sd->sd_offset + sd->sd_size + 4096 > 664 (off_t)pd->pd_disk_meta.sectors * 512) { 665 G_RAID_DEBUG1(1, sc, 666 "Disk too small (%llu < %llu)", 667 ((unsigned long long) 668 pd->pd_disk_meta.sectors) * 512, 669 (unsigned long long) 670 sd->sd_offset + sd->sd_size + 4096); 671 break; 672 } 673 } 674 if (sd != NULL) 675 continue; 676 if (tmpdisk->d_state == G_RAID_DISK_S_OFFLINE) { 677 olddisk = tmpdisk; 678 break; 679 } else if (olddisk == NULL) 680 olddisk = tmpdisk; 681 } 682 if (olddisk == NULL) { 683nofit: 684 if (pd->pd_disk_meta.flags & INTEL_F_SPARE) { 685 g_raid_change_disk_state(disk, 686 G_RAID_DISK_S_SPARE); 687 return (1); 688 } else { 689 g_raid_change_disk_state(disk, 690 G_RAID_DISK_S_STALE); 691 return (0); 692 } 693 } 694 oldpd = (struct g_raid_md_intel_perdisk *)olddisk->d_md_data; 695 disk_pos = oldpd->pd_disk_pos; 696 resurrection = 1; 697 } 698 699 if (olddisk == NULL) { 700 /* Find placeholder by position. */ 701 olddisk = g_raid_md_intel_get_disk(sc, disk_pos); 702 if (olddisk == NULL) 703 panic("No disk at position %d!", disk_pos); 704 if (olddisk->d_state != G_RAID_DISK_S_OFFLINE) { 705 G_RAID_DEBUG1(1, sc, "More then one disk for pos %d", 706 disk_pos); 707 g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE); 708 return (0); 709 } 710 oldpd = (struct g_raid_md_intel_perdisk *)olddisk->d_md_data; 711 } 712 713 /* Replace failed disk or placeholder with new disk. */ 714 TAILQ_FOREACH_SAFE(sd, &olddisk->d_subdisks, sd_next, tmpsd) { 715 TAILQ_REMOVE(&olddisk->d_subdisks, sd, sd_next); 716 TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); 717 sd->sd_disk = disk; 718 } 719 oldpd->pd_disk_pos = -2; 720 pd->pd_disk_pos = disk_pos; 721 722 /* If it was placeholder -- destroy it. */ 723 if (olddisk->d_state == G_RAID_DISK_S_OFFLINE) { 724 g_raid_destroy_disk(olddisk); 725 } else { 726 /* Otherwise, make it STALE_FAILED. */ 727 g_raid_change_disk_state(olddisk, G_RAID_DISK_S_STALE_FAILED); 728 /* Update global metadata just in case. */ 729 memcpy(&meta->disk[disk_pos], &pd->pd_disk_meta, 730 sizeof(struct intel_raid_disk)); 731 } 732 733 /* Welcome the new disk. */ 734 if (resurrection) 735 g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE); 736 else if (meta->disk[disk_pos].flags & INTEL_F_FAILED) 737 g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED); 738 else if (meta->disk[disk_pos].flags & INTEL_F_SPARE) 739 g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE); 740 else 741 g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE); 742 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 743 mvol = intel_get_volume(meta, 744 (uintptr_t)(sd->sd_volume->v_md_data)); 745 mmap0 = intel_get_map(mvol, 0); 746 if (mvol->migr_state) 747 mmap1 = intel_get_map(mvol, 1); 748 else 749 mmap1 = mmap0; 750 751 if (resurrection) { 752 /* Stale disk, almost same as new. */ 753 g_raid_change_subdisk_state(sd, 754 G_RAID_SUBDISK_S_NEW); 755 } else if (meta->disk[disk_pos].flags & INTEL_F_FAILED) { 756 /* Failed disk, almost useless. */ 757 g_raid_change_subdisk_state(sd, 758 G_RAID_SUBDISK_S_FAILED); 759 } else if (mvol->migr_state == 0) { 760 if (mmap0->status == INTEL_S_UNINITIALIZED) { 761 /* Freshly created uninitialized volume. */ 762 g_raid_change_subdisk_state(sd, 763 G_RAID_SUBDISK_S_UNINITIALIZED); 764 } else if (mmap0->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) { 765 /* Freshly inserted disk. */ 766 g_raid_change_subdisk_state(sd, 767 G_RAID_SUBDISK_S_NEW); 768 } else if (mvol->dirty) { 769 /* Dirty volume (unclean shutdown). */ 770 g_raid_change_subdisk_state(sd, 771 G_RAID_SUBDISK_S_STALE); 772 } else { 773 /* Up to date disk. */ 774 g_raid_change_subdisk_state(sd, 775 G_RAID_SUBDISK_S_ACTIVE); 776 } 777 } else if (mvol->migr_type == INTEL_MT_INIT || 778 mvol->migr_type == INTEL_MT_REBUILD) { 779 if (mmap0->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) { 780 /* Freshly inserted disk. */ 781 g_raid_change_subdisk_state(sd, 782 G_RAID_SUBDISK_S_NEW); 783 } else if (mmap1->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) { 784 /* Rebuilding disk. */ 785 g_raid_change_subdisk_state(sd, 786 G_RAID_SUBDISK_S_REBUILD); 787 if (mvol->dirty) { 788 sd->sd_rebuild_pos = 0; 789 } else { 790 sd->sd_rebuild_pos = 791 (off_t)mvol->curr_migr_unit * 792 sd->sd_volume->v_strip_size * 793 mmap0->total_domains; 794 } 795 } else if (mvol->dirty) { 796 /* Dirty volume (unclean shutdown). */ 797 g_raid_change_subdisk_state(sd, 798 G_RAID_SUBDISK_S_STALE); 799 } else { 800 /* Up to date disk. */ 801 g_raid_change_subdisk_state(sd, 802 G_RAID_SUBDISK_S_ACTIVE); 803 } 804 } else if (mvol->migr_type == INTEL_MT_VERIFY || 805 mvol->migr_type == INTEL_MT_REPAIR) { 806 if (mmap0->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) { 807 /* Freshly inserted disk. */ 808 g_raid_change_subdisk_state(sd, 809 G_RAID_SUBDISK_S_NEW); 810 } else if (mmap1->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) { 811 /* Resyncing disk. */ 812 g_raid_change_subdisk_state(sd, 813 G_RAID_SUBDISK_S_RESYNC); 814 if (mvol->dirty) { 815 sd->sd_rebuild_pos = 0; 816 } else { 817 sd->sd_rebuild_pos = 818 (off_t)mvol->curr_migr_unit * 819 sd->sd_volume->v_strip_size * 820 mmap0->total_domains; 821 } 822 } else if (mvol->dirty) { 823 /* Dirty volume (unclean shutdown). */ 824 g_raid_change_subdisk_state(sd, 825 G_RAID_SUBDISK_S_STALE); 826 } else { 827 /* Up to date disk. */ 828 g_raid_change_subdisk_state(sd, 829 G_RAID_SUBDISK_S_ACTIVE); 830 } 831 } 832 g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW, 833 G_RAID_EVENT_SUBDISK); 834 } 835 836 /* Update status of our need for spare. */ 837 if (mdi->mdio_started) { 838 mdi->mdio_incomplete = 839 (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) < 840 meta->total_disks); 841 } 842 843 return (resurrection); 844} 845 846static void 847g_disk_md_intel_retaste(void *arg, int pending) 848{ 849 850 G_RAID_DEBUG(1, "Array is not complete, trying to retaste."); 851 g_retaste(&g_raid_class); 852 free(arg, M_MD_INTEL); 853} 854 855static void 856g_raid_md_intel_refill(struct g_raid_softc *sc) 857{ 858 struct g_raid_md_object *md; 859 struct g_raid_md_intel_object *mdi; 860 struct intel_raid_conf *meta; 861 struct g_raid_disk *disk; 862 struct task *task; 863 int update, na; 864 865 md = sc->sc_md; 866 mdi = (struct g_raid_md_intel_object *)md; 867 meta = mdi->mdio_meta; 868 update = 0; 869 do { 870 /* Make sure we miss anything. */ 871 na = g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE); 872 if (na == meta->total_disks) 873 break; 874 875 G_RAID_DEBUG1(1, md->mdo_softc, 876 "Array is not complete (%d of %d), " 877 "trying to refill.", na, meta->total_disks); 878 879 /* Try to get use some of STALE disks. */ 880 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 881 if (disk->d_state == G_RAID_DISK_S_STALE) { 882 update += g_raid_md_intel_start_disk(disk); 883 if (disk->d_state == G_RAID_DISK_S_ACTIVE) 884 break; 885 } 886 } 887 if (disk != NULL) 888 continue; 889 890 /* Try to get use some of SPARE disks. */ 891 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 892 if (disk->d_state == G_RAID_DISK_S_SPARE) { 893 update += g_raid_md_intel_start_disk(disk); 894 if (disk->d_state == G_RAID_DISK_S_ACTIVE) 895 break; 896 } 897 } 898 } while (disk != NULL); 899 900 /* Write new metadata if we changed something. */ 901 if (update) { 902 g_raid_md_write_intel(md, NULL, NULL, NULL); 903 meta = mdi->mdio_meta; 904 } 905 906 /* Update status of our need for spare. */ 907 mdi->mdio_incomplete = (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) < 908 meta->total_disks); 909 910 /* Request retaste hoping to find spare. */ 911 if (mdi->mdio_incomplete) { 912 task = malloc(sizeof(struct task), 913 M_MD_INTEL, M_WAITOK | M_ZERO); 914 TASK_INIT(task, 0, g_disk_md_intel_retaste, task); 915 taskqueue_enqueue(taskqueue_swi, task); 916 } 917} 918 919static void 920g_raid_md_intel_start(struct g_raid_softc *sc) 921{ 922 struct g_raid_md_object *md; 923 struct g_raid_md_intel_object *mdi; 924 struct g_raid_md_intel_perdisk *pd; 925 struct intel_raid_conf *meta; 926 struct intel_raid_vol *mvol; 927 struct intel_raid_map *mmap; 928 struct g_raid_volume *vol; 929 struct g_raid_subdisk *sd; 930 struct g_raid_disk *disk; 931 int i, j, disk_pos; 932 933 md = sc->sc_md; 934 mdi = (struct g_raid_md_intel_object *)md; 935 meta = mdi->mdio_meta; 936 937 /* Create volumes and subdisks. */ 938 for (i = 0; i < meta->total_volumes; i++) { 939 mvol = intel_get_volume(meta, i); 940 mmap = intel_get_map(mvol, 0); 941 vol = g_raid_create_volume(sc, mvol->name, -1); 942 vol->v_md_data = (void *)(intptr_t)i; 943 if (mmap->type == INTEL_T_RAID0) 944 vol->v_raid_level = G_RAID_VOLUME_RL_RAID0; 945 else if (mmap->type == INTEL_T_RAID1 && 946 mmap->total_domains >= 2 && 947 mmap->total_domains <= mmap->total_disks) { 948 /* Assume total_domains is correct. */ 949 if (mmap->total_domains == mmap->total_disks) 950 vol->v_raid_level = G_RAID_VOLUME_RL_RAID1; 951 else 952 vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E; 953 } else if (mmap->type == INTEL_T_RAID1) { 954 /* total_domains looks wrong. */ 955 if (mmap->total_disks <= 2) 956 vol->v_raid_level = G_RAID_VOLUME_RL_RAID1; 957 else 958 vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E; 959 } else if (mmap->type == INTEL_T_RAID5) 960 vol->v_raid_level = G_RAID_VOLUME_RL_RAID5; 961 else 962 vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; 963 vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE; 964 vol->v_strip_size = (u_int)mmap->strip_sectors * 512; //ZZZ 965 vol->v_disks_count = mmap->total_disks; 966 vol->v_mediasize = (off_t)mvol->total_sectors * 512; //ZZZ 967 vol->v_sectorsize = 512; //ZZZ 968 for (j = 0; j < vol->v_disks_count; j++) { 969 sd = &vol->v_subdisks[j]; 970 sd->sd_offset = (off_t)mmap->offset * 512; //ZZZ 971 sd->sd_size = (off_t)mmap->disk_sectors * 512; //ZZZ 972 } 973 g_raid_start_volume(vol); 974 } 975 976 /* Create disk placeholders to store data for later writing. */ 977 for (disk_pos = 0; disk_pos < meta->total_disks; disk_pos++) { 978 pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO); 979 pd->pd_disk_pos = disk_pos; 980 pd->pd_disk_meta = meta->disk[disk_pos]; 981 disk = g_raid_create_disk(sc); 982 disk->d_md_data = (void *)pd; 983 disk->d_state = G_RAID_DISK_S_OFFLINE; 984 for (i = 0; i < meta->total_volumes; i++) { 985 mvol = intel_get_volume(meta, i); 986 mmap = intel_get_map(mvol, 0); 987 for (j = 0; j < mmap->total_disks; j++) { 988 if ((mmap->disk_idx[j] & INTEL_DI_IDX) == disk_pos) 989 break; 990 } 991 if (j == mmap->total_disks) 992 continue; 993 vol = g_raid_md_intel_get_volume(sc, i); 994 sd = &vol->v_subdisks[j]; 995 sd->sd_disk = disk; 996 TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); 997 } 998 } 999 1000 /* Make all disks found till the moment take their places. */ 1001 do { 1002 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 1003 if (disk->d_state == G_RAID_DISK_S_NONE) { 1004 g_raid_md_intel_start_disk(disk); 1005 break; 1006 } 1007 } 1008 } while (disk != NULL); 1009 1010 mdi->mdio_started = 1; 1011 G_RAID_DEBUG1(0, sc, "Array started."); 1012 g_raid_md_write_intel(md, NULL, NULL, NULL); 1013 1014 /* Pickup any STALE/SPARE disks to refill array if needed. */ 1015 g_raid_md_intel_refill(sc); 1016 1017 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1018 g_raid_event_send(vol, G_RAID_VOLUME_E_START, 1019 G_RAID_EVENT_VOLUME); 1020 } 1021 1022 callout_stop(&mdi->mdio_start_co); 1023 G_RAID_DEBUG1(1, sc, "root_mount_rel %p", mdi->mdio_rootmount); 1024 root_mount_rel(mdi->mdio_rootmount); 1025 mdi->mdio_rootmount = NULL; 1026} 1027 1028static void 1029g_raid_md_intel_new_disk(struct g_raid_disk *disk) 1030{ 1031 struct g_raid_softc *sc; 1032 struct g_raid_md_object *md; 1033 struct g_raid_md_intel_object *mdi; 1034 struct intel_raid_conf *pdmeta; 1035 struct g_raid_md_intel_perdisk *pd; 1036 1037 sc = disk->d_softc; 1038 md = sc->sc_md; 1039 mdi = (struct g_raid_md_intel_object *)md; 1040 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 1041 pdmeta = pd->pd_meta; 1042 1043 if (mdi->mdio_started) { 1044 if (g_raid_md_intel_start_disk(disk)) 1045 g_raid_md_write_intel(md, NULL, NULL, NULL); 1046 } else { 1047 /* If we haven't started yet - check metadata freshness. */ 1048 if (mdi->mdio_meta == NULL || 1049 ((int32_t)(pdmeta->generation - mdi->mdio_generation)) > 0) { 1050 G_RAID_DEBUG1(1, sc, "Newer disk"); 1051 if (mdi->mdio_meta != NULL) 1052 free(mdi->mdio_meta, M_MD_INTEL); 1053 mdi->mdio_meta = intel_meta_copy(pdmeta); 1054 mdi->mdio_generation = mdi->mdio_meta->generation; 1055 mdi->mdio_disks_present = 1; 1056 } else if (pdmeta->generation == mdi->mdio_generation) { 1057 mdi->mdio_disks_present++; 1058 G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)", 1059 mdi->mdio_disks_present, 1060 mdi->mdio_meta->total_disks); 1061 } else { 1062 G_RAID_DEBUG1(1, sc, "Older disk"); 1063 } 1064 /* If we collected all needed disks - start array. */ 1065 if (mdi->mdio_disks_present == mdi->mdio_meta->total_disks) 1066 g_raid_md_intel_start(sc); 1067 } 1068} 1069 1070static void 1071g_raid_intel_go(void *arg) 1072{ 1073 struct g_raid_softc *sc; 1074 struct g_raid_md_object *md; 1075 struct g_raid_md_intel_object *mdi; 1076 1077 sc = arg; 1078 md = sc->sc_md; 1079 mdi = (struct g_raid_md_intel_object *)md; 1080 if (!mdi->mdio_started) { 1081 G_RAID_DEBUG1(0, sc, "Force array start due to timeout."); 1082 g_raid_event_send(sc, G_RAID_NODE_E_START, 0); 1083 } 1084} 1085 1086static int 1087g_raid_md_create_intel(struct g_raid_md_object *md, struct g_class *mp, 1088 struct g_geom **gp) 1089{ 1090 struct g_raid_softc *sc; 1091 struct g_raid_md_intel_object *mdi; 1092 char name[16]; 1093 1094 mdi = (struct g_raid_md_intel_object *)md; 1095 mdi->mdio_config_id = arc4random(); 1096 mdi->mdio_generation = 0; 1097 snprintf(name, sizeof(name), "Intel-%08x", mdi->mdio_config_id); 1098 sc = g_raid_create_node(mp, name, md); 1099 if (sc == NULL) 1100 return (G_RAID_MD_TASTE_FAIL); 1101 md->mdo_softc = sc; 1102 *gp = sc->sc_geom; 1103 return (G_RAID_MD_TASTE_NEW); 1104} 1105 1106/* 1107 * Return the last N characters of the serial label. The Linux and 1108 * ataraid(7) code always uses the last 16 characters of the label to 1109 * store into the Intel meta format. Generalize this to N characters 1110 * since that's easy. Labels can be up to 20 characters for SATA drives 1111 * and up 251 characters for SAS drives. Since intel controllers don't 1112 * support SAS drives, just stick with the SATA limits for stack friendliness. 1113 */ 1114static int 1115g_raid_md_get_label(struct g_consumer *cp, char *serial, int serlen) 1116{ 1117 char serial_buffer[24]; 1118 int len, error; 1119 1120 len = sizeof(serial_buffer); 1121 error = g_io_getattr("GEOM::ident", cp, &len, serial_buffer); 1122 if (error != 0) 1123 return (error); 1124 len = strlen(serial_buffer); 1125 if (len > serlen) 1126 len -= serlen; 1127 else 1128 len = 0; 1129 strncpy(serial, serial_buffer + len, serlen); 1130 return (0); 1131} 1132 1133static int 1134g_raid_md_taste_intel(struct g_raid_md_object *md, struct g_class *mp, 1135 struct g_consumer *cp, struct g_geom **gp) 1136{ 1137 struct g_consumer *rcp; 1138 struct g_provider *pp; 1139 struct g_raid_md_intel_object *mdi, *mdi1; 1140 struct g_raid_softc *sc; 1141 struct g_raid_disk *disk; 1142 struct intel_raid_conf *meta; 1143 struct g_raid_md_intel_perdisk *pd; 1144 struct g_geom *geom; 1145 int error, disk_pos, result, spare, len; 1146 char serial[INTEL_SERIAL_LEN]; 1147 char name[16]; 1148 uint16_t vendor; 1149 1150 G_RAID_DEBUG(1, "Tasting Intel on %s", cp->provider->name); 1151 mdi = (struct g_raid_md_intel_object *)md; 1152 pp = cp->provider; 1153 1154 /* Read metadata from device. */ 1155 meta = NULL; 1156 vendor = 0xffff; 1157 disk_pos = 0; 1158 if (g_access(cp, 1, 0, 0) != 0) 1159 return (G_RAID_MD_TASTE_FAIL); 1160 g_topology_unlock(); 1161 error = g_raid_md_get_label(cp, serial, sizeof(serial)); 1162 if (error != 0) { 1163 G_RAID_DEBUG(1, "Cannot get serial number from %s (error=%d).", 1164 pp->name, error); 1165 goto fail2; 1166 } 1167 len = 2; 1168 if (pp->geom->rank == 1) 1169 g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor); 1170 meta = intel_meta_read(cp); 1171 g_topology_lock(); 1172 g_access(cp, -1, 0, 0); 1173 if (meta == NULL) { 1174 if (g_raid_aggressive_spare) { 1175 if (vendor == 0x8086) { 1176 G_RAID_DEBUG(1, 1177 "No Intel metadata, forcing spare."); 1178 spare = 2; 1179 goto search; 1180 } else { 1181 G_RAID_DEBUG(1, 1182 "Intel vendor mismatch 0x%04x != 0x8086", 1183 vendor); 1184 } 1185 } 1186 return (G_RAID_MD_TASTE_FAIL); 1187 } 1188 1189 /* Check this disk position in obtained metadata. */ 1190 disk_pos = intel_meta_find_disk(meta, serial); 1191 if (disk_pos < 0) { 1192 G_RAID_DEBUG(1, "Intel serial '%s' not found", serial); 1193 goto fail1; 1194 } 1195 if (meta->disk[disk_pos].sectors != 1196 (pp->mediasize / pp->sectorsize)) { 1197 G_RAID_DEBUG(1, "Intel size mismatch %u != %u", 1198 meta->disk[disk_pos].sectors, 1199 (u_int)(pp->mediasize / pp->sectorsize)); 1200 goto fail1; 1201 } 1202 1203 /* Metadata valid. Print it. */ 1204 g_raid_md_intel_print(meta); 1205 G_RAID_DEBUG(1, "Intel disk position %d", disk_pos); 1206 spare = meta->disk[disk_pos].flags & INTEL_F_SPARE; 1207 1208search: 1209 /* Search for matching node. */ 1210 sc = NULL; 1211 mdi1 = NULL; 1212 LIST_FOREACH(geom, &mp->geom, geom) { 1213 sc = geom->softc; 1214 if (sc == NULL) 1215 continue; 1216 if (sc->sc_stopping != 0) 1217 continue; 1218 if (sc->sc_md->mdo_class != md->mdo_class) 1219 continue; 1220 mdi1 = (struct g_raid_md_intel_object *)sc->sc_md; 1221 if (spare) { 1222 if (mdi1->mdio_incomplete) 1223 break; 1224 } else { 1225 if (mdi1->mdio_config_id == meta->config_id) 1226 break; 1227 } 1228 } 1229 1230 /* Found matching node. */ 1231 if (geom != NULL) { 1232 G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name); 1233 result = G_RAID_MD_TASTE_EXISTING; 1234 1235 } else if (spare) { /* Not found needy node -- left for later. */ 1236 G_RAID_DEBUG(1, "Spare is not needed at this time"); 1237 goto fail1; 1238 1239 } else { /* Not found matching node -- create one. */ 1240 result = G_RAID_MD_TASTE_NEW; 1241 mdi->mdio_config_id = meta->config_id; 1242 snprintf(name, sizeof(name), "Intel-%08x", meta->config_id); 1243 sc = g_raid_create_node(mp, name, md); 1244 md->mdo_softc = sc; 1245 geom = sc->sc_geom; 1246 callout_init(&mdi->mdio_start_co, 1); 1247 callout_reset(&mdi->mdio_start_co, g_raid_start_timeout * hz, 1248 g_raid_intel_go, sc); 1249 mdi->mdio_rootmount = root_mount_hold("GRAID-Intel"); 1250 G_RAID_DEBUG1(1, sc, "root_mount_hold %p", mdi->mdio_rootmount); 1251 } 1252 1253 rcp = g_new_consumer(geom); 1254 g_attach(rcp, pp); 1255 if (g_access(rcp, 1, 1, 1) != 0) 1256 ; //goto fail1; 1257 1258 g_topology_unlock(); 1259 sx_xlock(&sc->sc_lock); 1260 1261 pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO); 1262 pd->pd_meta = meta; 1263 pd->pd_disk_pos = -1; 1264 if (spare == 2) { 1265 memcpy(&pd->pd_disk_meta.serial[0], serial, INTEL_SERIAL_LEN); 1266 pd->pd_disk_meta.sectors = pp->mediasize / pp->sectorsize; 1267 pd->pd_disk_meta.id = 0; 1268 pd->pd_disk_meta.flags = INTEL_F_SPARE; 1269 } else { 1270 pd->pd_disk_meta = meta->disk[disk_pos]; 1271 } 1272 disk = g_raid_create_disk(sc); 1273 disk->d_md_data = (void *)pd; 1274 disk->d_consumer = rcp; 1275 rcp->private = disk; 1276 1277 /* Read kernel dumping information. */ 1278 disk->d_kd.offset = 0; 1279 disk->d_kd.length = OFF_MAX; 1280 len = sizeof(disk->d_kd); 1281 error = g_io_getattr("GEOM::kerneldump", rcp, &len, &disk->d_kd); 1282 if (disk->d_kd.di.dumper == NULL) 1283 G_RAID_DEBUG1(2, sc, "Dumping not supported by %s: %d.", 1284 rcp->provider->name, error); 1285 1286 g_raid_md_intel_new_disk(disk); 1287 1288 sx_xunlock(&sc->sc_lock); 1289 g_topology_lock(); 1290 *gp = geom; 1291 return (result); 1292fail2: 1293 g_topology_lock(); 1294 g_access(cp, -1, 0, 0); 1295fail1: 1296 free(meta, M_MD_INTEL); 1297 return (G_RAID_MD_TASTE_FAIL); 1298} 1299 1300static int 1301g_raid_md_event_intel(struct g_raid_md_object *md, 1302 struct g_raid_disk *disk, u_int event) 1303{ 1304 struct g_raid_softc *sc; 1305 struct g_raid_subdisk *sd; 1306 struct g_raid_md_intel_object *mdi; 1307 struct g_raid_md_intel_perdisk *pd; 1308 1309 sc = md->mdo_softc; 1310 mdi = (struct g_raid_md_intel_object *)md; 1311 if (disk == NULL) { 1312 switch (event) { 1313 case G_RAID_NODE_E_START: 1314 if (!mdi->mdio_started) 1315 g_raid_md_intel_start(sc); 1316 return (0); 1317 } 1318 return (-1); 1319 } 1320 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 1321 switch (event) { 1322 case G_RAID_DISK_E_DISCONNECTED: 1323 /* If disk was assigned, just update statuses. */ 1324 if (pd->pd_disk_pos >= 0) { 1325 g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); 1326 if (disk->d_consumer) { 1327 g_raid_kill_consumer(sc, disk->d_consumer); 1328 disk->d_consumer = NULL; 1329 } 1330 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 1331 g_raid_change_subdisk_state(sd, 1332 G_RAID_SUBDISK_S_NONE); 1333 g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, 1334 G_RAID_EVENT_SUBDISK); 1335 } 1336 } else { 1337 /* Otherwise -- delete. */ 1338 g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE); 1339 g_raid_destroy_disk(disk); 1340 } 1341 1342 /* Write updated metadata to all disks. */ 1343 g_raid_md_write_intel(md, NULL, NULL, NULL); 1344 1345 /* Check if anything left except placeholders. */ 1346 if (g_raid_ndisks(sc, -1) == 1347 g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) 1348 g_raid_destroy_node(sc, 0); 1349 else 1350 g_raid_md_intel_refill(sc); 1351 return (0); 1352 } 1353 return (-2); 1354} 1355 1356static int 1357g_raid_md_ctl_intel(struct g_raid_md_object *md, 1358 struct gctl_req *req) 1359{ 1360 struct g_raid_softc *sc; 1361 struct g_raid_volume *vol, *vol1; 1362 struct g_raid_subdisk *sd; 1363 struct g_raid_disk *disk; 1364 struct g_raid_md_intel_object *mdi; 1365 struct g_raid_md_intel_perdisk *pd; 1366 struct g_consumer *cp; 1367 struct g_provider *pp; 1368 char arg[16], serial[INTEL_SERIAL_LEN]; 1369 const char *verb, *volname, *levelname, *diskname; 1370 char *tmp; 1371 int *nargs, *force; 1372 off_t off, size, sectorsize, strip; 1373 intmax_t *sizearg, *striparg; 1374 int numdisks, i, len, level, qual, update; 1375 int error; 1376 1377 sc = md->mdo_softc; 1378 mdi = (struct g_raid_md_intel_object *)md; 1379 verb = gctl_get_param(req, "verb", NULL); 1380 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); 1381 error = 0; 1382 if (strcmp(verb, "label") == 0) { 1383 1384 if (*nargs < 4) { 1385 gctl_error(req, "Invalid number of arguments."); 1386 return (-1); 1387 } 1388 volname = gctl_get_asciiparam(req, "arg1"); 1389 if (volname == NULL) { 1390 gctl_error(req, "No volume name."); 1391 return (-2); 1392 } 1393 levelname = gctl_get_asciiparam(req, "arg2"); 1394 if (levelname == NULL) { 1395 gctl_error(req, "No RAID level."); 1396 return (-3); 1397 } 1398 if (g_raid_volume_str2level(levelname, &level, &qual)) { 1399 gctl_error(req, "Unknown RAID level '%s'.", levelname); 1400 return (-4); 1401 } 1402 numdisks = *nargs - 3; 1403 force = gctl_get_paraml(req, "force", sizeof(*force)); 1404 if (!g_raid_md_intel_supported(level, qual, numdisks, 1405 force ? *force : 0)) { 1406 gctl_error(req, "Unsupported RAID level " 1407 "(0x%02x/0x%02x), or number of disks (%d).", 1408 level, qual, numdisks); 1409 return (-5); 1410 } 1411 1412 /* Search for disks, connect them and probe. */ 1413 size = 0x7fffffffffffffffllu; 1414 sectorsize = 0; 1415 for (i = 0; i < numdisks; i++) { 1416 snprintf(arg, sizeof(arg), "arg%d", i + 3); 1417 diskname = gctl_get_asciiparam(req, arg); 1418 if (diskname == NULL) { 1419 gctl_error(req, "No disk name (%s).", arg); 1420 error = -6; 1421 break; 1422 } 1423 if (strcmp(diskname, "NONE") == 0) { 1424 cp = NULL; 1425 pp = NULL; 1426 } else { 1427 g_topology_lock(); 1428 cp = g_raid_open_consumer(sc, diskname); 1429 if (cp == NULL) { 1430 gctl_error(req, "Can't open disk '%s'.", 1431 diskname); 1432 g_topology_unlock(); 1433 error = -7; 1434 break; 1435 } 1436 pp = cp->provider; 1437 } 1438 pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO); 1439 pd->pd_disk_pos = i; 1440 disk = g_raid_create_disk(sc); 1441 disk->d_md_data = (void *)pd; 1442 disk->d_consumer = cp; 1443 if (cp == NULL) { 1444 strcpy(&pd->pd_disk_meta.serial[0], "NONE"); 1445 pd->pd_disk_meta.id = 0xffffffff; 1446 pd->pd_disk_meta.flags = INTEL_F_ASSIGNED; 1447 continue; 1448 } 1449 cp->private = disk; 1450 g_topology_unlock(); 1451 1452 error = g_raid_md_get_label(cp, 1453 &pd->pd_disk_meta.serial[0], INTEL_SERIAL_LEN); 1454 if (error != 0) { 1455 gctl_error(req, 1456 "Can't get serial for provider '%s'.", 1457 diskname); 1458 error = -8; 1459 break; 1460 } 1461 1462 /* Read kernel dumping information. */ 1463 disk->d_kd.offset = 0; 1464 disk->d_kd.length = OFF_MAX; 1465 len = sizeof(disk->d_kd); 1466 g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); 1467 if (disk->d_kd.di.dumper == NULL) 1468 G_RAID_DEBUG1(2, sc, 1469 "Dumping not supported by %s.", 1470 cp->provider->name); 1471 1472 pd->pd_disk_meta.sectors = pp->mediasize / pp->sectorsize; 1473 if (size > pp->mediasize) 1474 size = pp->mediasize; 1475 if (sectorsize < pp->sectorsize) 1476 sectorsize = pp->sectorsize; 1477 pd->pd_disk_meta.id = 0; 1478 pd->pd_disk_meta.flags = INTEL_F_ASSIGNED | INTEL_F_ONLINE; 1479 } 1480 if (error != 0) 1481 return (error); 1482 1483 if (sectorsize <= 0) { 1484 gctl_error(req, "Can't get sector size."); 1485 return (-8); 1486 } 1487 1488 /* Reserve some space for metadata. */ 1489 size -= ((4096 + sectorsize - 1) / sectorsize) * sectorsize; 1490 1491 /* Handle size argument. */ 1492 len = sizeof(*sizearg); 1493 sizearg = gctl_get_param(req, "size", &len); 1494 if (sizearg != NULL && len == sizeof(*sizearg) && 1495 *sizearg > 0) { 1496 if (*sizearg > size) { 1497 gctl_error(req, "Size too big %lld > %lld.", 1498 (long long)*sizearg, (long long)size); 1499 return (-9); 1500 } 1501 size = *sizearg; 1502 } 1503 1504 /* Handle strip argument. */ 1505 strip = 131072; 1506 len = sizeof(*striparg); 1507 striparg = gctl_get_param(req, "strip", &len); 1508 if (striparg != NULL && len == sizeof(*striparg) && 1509 *striparg > 0) { 1510 if (*striparg < sectorsize) { 1511 gctl_error(req, "Strip size too small."); 1512 return (-10); 1513 } 1514 if (*striparg % sectorsize != 0) { 1515 gctl_error(req, "Incorrect strip size."); 1516 return (-11); 1517 } 1518 if (strip > 65535 * sectorsize) { 1519 gctl_error(req, "Strip size too big."); 1520 return (-12); 1521 } 1522 strip = *striparg; 1523 } 1524 1525 /* Round size down to strip or sector. */ 1526 if (level == G_RAID_VOLUME_RL_RAID1) 1527 size -= (size % sectorsize); 1528 else if (level == G_RAID_VOLUME_RL_RAID1E && 1529 (numdisks & 1) != 0) 1530 size -= (size % (2 * strip)); 1531 else 1532 size -= (size % strip); 1533 if (size <= 0) { 1534 gctl_error(req, "Size too small."); 1535 return (-13); 1536 } 1537 if (size > 0xffffffffllu * sectorsize) { 1538 gctl_error(req, "Size too big."); 1539 return (-14); 1540 } 1541 1542 /* We have all we need, create things: volume, ... */ 1543 mdi->mdio_started = 1; 1544 vol = g_raid_create_volume(sc, volname, -1); 1545 vol->v_md_data = (void *)(intptr_t)0; 1546 vol->v_raid_level = level; 1547 vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE; 1548 vol->v_strip_size = strip; 1549 vol->v_disks_count = numdisks; 1550 if (level == G_RAID_VOLUME_RL_RAID0) 1551 vol->v_mediasize = size * numdisks; 1552 else if (level == G_RAID_VOLUME_RL_RAID1) 1553 vol->v_mediasize = size; 1554 else if (level == G_RAID_VOLUME_RL_RAID5) 1555 vol->v_mediasize = size * (numdisks - 1); 1556 else { /* RAID1E */ 1557 vol->v_mediasize = ((size * numdisks) / strip / 2) * 1558 strip; 1559 } 1560 vol->v_sectorsize = sectorsize; 1561 g_raid_start_volume(vol); 1562 1563 /* , and subdisks. */ 1564 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 1565 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 1566 sd = &vol->v_subdisks[pd->pd_disk_pos]; 1567 sd->sd_disk = disk; 1568 sd->sd_offset = 0; 1569 sd->sd_size = size; 1570 TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); 1571 if (sd->sd_disk->d_consumer != NULL) { 1572 g_raid_change_disk_state(disk, 1573 G_RAID_DISK_S_ACTIVE); 1574 g_raid_change_subdisk_state(sd, 1575 G_RAID_SUBDISK_S_ACTIVE); 1576 g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW, 1577 G_RAID_EVENT_SUBDISK); 1578 } else { 1579 g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); 1580 } 1581 } 1582 1583 /* Write metadata based on created entities. */ 1584 G_RAID_DEBUG1(0, sc, "Array started."); 1585 g_raid_md_write_intel(md, NULL, NULL, NULL); 1586 1587 /* Pickup any STALE/SPARE disks to refill array if needed. */ 1588 g_raid_md_intel_refill(sc); 1589 1590 g_raid_event_send(vol, G_RAID_VOLUME_E_START, 1591 G_RAID_EVENT_VOLUME); 1592 return (0); 1593 } 1594 if (strcmp(verb, "add") == 0) { 1595 1596 if (*nargs != 3) { 1597 gctl_error(req, "Invalid number of arguments."); 1598 return (-1); 1599 } 1600 volname = gctl_get_asciiparam(req, "arg1"); 1601 if (volname == NULL) { 1602 gctl_error(req, "No volume name."); 1603 return (-2); 1604 } 1605 levelname = gctl_get_asciiparam(req, "arg2"); 1606 if (levelname == NULL) { 1607 gctl_error(req, "No RAID level."); 1608 return (-3); 1609 } 1610 if (g_raid_volume_str2level(levelname, &level, &qual)) { 1611 gctl_error(req, "Unknown RAID level '%s'.", levelname); 1612 return (-4); 1613 } 1614 1615 /* Look for existing volumes. */ 1616 i = 0; 1617 vol1 = NULL; 1618 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1619 vol1 = vol; 1620 i++; 1621 } 1622 if (i > 1) { 1623 gctl_error(req, "Maximum two volumes supported."); 1624 return (-6); 1625 } 1626 if (vol1 == NULL) { 1627 gctl_error(req, "At least one volume must exist."); 1628 return (-7); 1629 } 1630 1631 numdisks = vol1->v_disks_count; 1632 force = gctl_get_paraml(req, "force", sizeof(*force)); 1633 if (!g_raid_md_intel_supported(level, qual, numdisks, 1634 force ? *force : 0)) { 1635 gctl_error(req, "Unsupported RAID level " 1636 "(0x%02x/0x%02x), or number of disks (%d).", 1637 level, qual, numdisks); 1638 return (-5); 1639 } 1640 1641 /* Collect info about present disks. */ 1642 size = 0x7fffffffffffffffllu; 1643 sectorsize = 512; 1644 for (i = 0; i < numdisks; i++) { 1645 disk = vol1->v_subdisks[i].sd_disk; 1646 pd = (struct g_raid_md_intel_perdisk *) 1647 disk->d_md_data; 1648 if ((off_t)pd->pd_disk_meta.sectors * 512 < size) 1649 size = (off_t)pd->pd_disk_meta.sectors * 512; 1650 if (disk->d_consumer != NULL && 1651 disk->d_consumer->provider != NULL && 1652 disk->d_consumer->provider->sectorsize > 1653 sectorsize) { 1654 sectorsize = 1655 disk->d_consumer->provider->sectorsize; 1656 } 1657 } 1658 1659 /* Reserve some space for metadata. */ 1660 size -= ((4096 + sectorsize - 1) / sectorsize) * sectorsize; 1661 1662 /* Decide insert before or after. */ 1663 sd = &vol1->v_subdisks[0]; 1664 if (sd->sd_offset > 1665 size - (sd->sd_offset + sd->sd_size)) { 1666 off = 0; 1667 size = sd->sd_offset; 1668 } else { 1669 off = sd->sd_offset + sd->sd_size; 1670 size = size - (sd->sd_offset + sd->sd_size); 1671 } 1672 1673 /* Handle strip argument. */ 1674 strip = 131072; 1675 len = sizeof(*striparg); 1676 striparg = gctl_get_param(req, "strip", &len); 1677 if (striparg != NULL && len == sizeof(*striparg) && 1678 *striparg > 0) { 1679 if (*striparg < sectorsize) { 1680 gctl_error(req, "Strip size too small."); 1681 return (-10); 1682 } 1683 if (*striparg % sectorsize != 0) { 1684 gctl_error(req, "Incorrect strip size."); 1685 return (-11); 1686 } 1687 if (strip > 65535 * sectorsize) { 1688 gctl_error(req, "Strip size too big."); 1689 return (-12); 1690 } 1691 strip = *striparg; 1692 } 1693 1694 /* Round offset up to strip. */ 1695 if (off % strip != 0) { 1696 size -= strip - off % strip; 1697 off += strip - off % strip; 1698 } 1699 1700 /* Handle size argument. */ 1701 len = sizeof(*sizearg); 1702 sizearg = gctl_get_param(req, "size", &len); 1703 if (sizearg != NULL && len == sizeof(*sizearg) && 1704 *sizearg > 0) { 1705 if (*sizearg > size) { 1706 gctl_error(req, "Size too big %lld > %lld.", 1707 (long long)*sizearg, (long long)size); 1708 return (-9); 1709 } 1710 size = *sizearg; 1711 } 1712 1713 /* Round size down to strip or sector. */ 1714 if (level == G_RAID_VOLUME_RL_RAID1) 1715 size -= (size % sectorsize); 1716 else 1717 size -= (size % strip); 1718 if (size <= 0) { 1719 gctl_error(req, "Size too small."); 1720 return (-13); 1721 } 1722 if (size > 0xffffffffllu * sectorsize) { 1723 gctl_error(req, "Size too big."); 1724 return (-14); 1725 } 1726 1727 /* We have all we need, create things: volume, ... */ 1728 vol = g_raid_create_volume(sc, volname, -1); 1729 vol->v_md_data = (void *)(intptr_t)i; 1730 vol->v_raid_level = level; 1731 vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE; 1732 vol->v_strip_size = strip; 1733 vol->v_disks_count = numdisks; 1734 if (level == G_RAID_VOLUME_RL_RAID0) 1735 vol->v_mediasize = size * numdisks; 1736 else if (level == G_RAID_VOLUME_RL_RAID1) 1737 vol->v_mediasize = size; 1738 else if (level == G_RAID_VOLUME_RL_RAID5) 1739 vol->v_mediasize = size * (numdisks - 1); 1740 else { /* RAID1E */ 1741 vol->v_mediasize = ((size * numdisks) / strip / 2) * 1742 strip; 1743 } 1744 vol->v_sectorsize = sectorsize; 1745 g_raid_start_volume(vol); 1746 1747 /* , and subdisks. */ 1748 for (i = 0; i < numdisks; i++) { 1749 disk = vol1->v_subdisks[i].sd_disk; 1750 sd = &vol->v_subdisks[i]; 1751 sd->sd_disk = disk; 1752 sd->sd_offset = off; 1753 sd->sd_size = size; 1754 TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); 1755 if (disk->d_state == G_RAID_DISK_S_ACTIVE) { 1756 g_raid_change_subdisk_state(sd, 1757 G_RAID_SUBDISK_S_ACTIVE); 1758 g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW, 1759 G_RAID_EVENT_SUBDISK); 1760 } 1761 } 1762 1763 /* Write metadata based on created entities. */ 1764 g_raid_md_write_intel(md, NULL, NULL, NULL); 1765 1766 g_raid_event_send(vol, G_RAID_VOLUME_E_START, 1767 G_RAID_EVENT_VOLUME); 1768 return (0); 1769 } 1770 if (strcmp(verb, "delete") == 0) { 1771 1772 /* Full node destruction. */ 1773 if (*nargs == 1) { 1774 /* Check if some volume is still open. */ 1775 force = gctl_get_paraml(req, "force", sizeof(*force)); 1776 if (force != NULL && *force == 0 && 1777 g_raid_nopens(sc) != 0) { 1778 gctl_error(req, "Some volume is still open."); 1779 return (-4); 1780 } 1781 1782 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 1783 if (disk->d_consumer) 1784 intel_meta_erase(disk->d_consumer); 1785 } 1786 g_raid_destroy_node(sc, 0); 1787 return (0); 1788 } 1789 1790 /* Destroy specified volume. If it was last - all node. */ 1791 if (*nargs != 2) { 1792 gctl_error(req, "Invalid number of arguments."); 1793 return (-1); 1794 } 1795 volname = gctl_get_asciiparam(req, "arg1"); 1796 if (volname == NULL) { 1797 gctl_error(req, "No volume name."); 1798 return (-2); 1799 } 1800 1801 /* Search for volume. */ 1802 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1803 if (strcmp(vol->v_name, volname) == 0) 1804 break; 1805 } 1806 if (vol == NULL) { 1807 i = strtol(volname, &tmp, 10); 1808 if (verb != volname && tmp[0] == 0) { 1809 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1810 if (vol->v_global_id == i) 1811 break; 1812 } 1813 } 1814 } 1815 if (vol == NULL) { 1816 gctl_error(req, "Volume '%s' not found.", volname); 1817 return (-3); 1818 } 1819 1820 /* Check if volume is still open. */ 1821 force = gctl_get_paraml(req, "force", sizeof(*force)); 1822 if (force != NULL && *force == 0 && 1823 vol->v_provider_open != 0) { 1824 gctl_error(req, "Volume is still open."); 1825 return (-4); 1826 } 1827 1828 /* Destroy volume and potentially node. */ 1829 i = 0; 1830 TAILQ_FOREACH(vol1, &sc->sc_volumes, v_next) 1831 i++; 1832 if (i >= 2) { 1833 g_raid_destroy_volume(vol); 1834 g_raid_md_write_intel(md, NULL, NULL, NULL); 1835 } else { 1836 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 1837 if (disk->d_consumer) 1838 intel_meta_erase(disk->d_consumer); 1839 } 1840 g_raid_destroy_node(sc, 0); 1841 } 1842 return (0); 1843 } 1844 if (strcmp(verb, "remove") == 0 || 1845 strcmp(verb, "fail") == 0) { 1846 if (*nargs < 2) { 1847 gctl_error(req, "Invalid number of arguments."); 1848 return (-1); 1849 } 1850 for (i = 1; i < *nargs; i++) { 1851 snprintf(arg, sizeof(arg), "arg%d", i); 1852 diskname = gctl_get_asciiparam(req, arg); 1853 if (diskname == NULL) { 1854 gctl_error(req, "No disk name (%s).", arg); 1855 error = -2; 1856 break; 1857 } 1858 if (strncmp(diskname, "/dev/", 5) == 0) 1859 diskname += 5; 1860 1861 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 1862 if (disk->d_consumer != NULL && 1863 disk->d_consumer->provider != NULL && 1864 strcmp(disk->d_consumer->provider->name, 1865 diskname) == 0) 1866 break; 1867 } 1868 if (disk == NULL) { 1869 gctl_error(req, "Disk '%s' not found.", 1870 diskname); 1871 error = -3; 1872 break; 1873 } 1874 1875 if (strcmp(verb, "fail") == 0) { 1876 g_raid_md_fail_disk_intel(md, NULL, disk); 1877 continue; 1878 } 1879 1880 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 1881 1882 /* Erase metadata on deleting disk. */ 1883 intel_meta_erase(disk->d_consumer); 1884 1885 /* If disk was assigned, just update statuses. */ 1886 if (pd->pd_disk_pos >= 0) { 1887 g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); 1888 g_raid_kill_consumer(sc, disk->d_consumer); 1889 disk->d_consumer = NULL; 1890 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 1891 g_raid_change_subdisk_state(sd, 1892 G_RAID_SUBDISK_S_NONE); 1893 g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, 1894 G_RAID_EVENT_SUBDISK); 1895 } 1896 } else { 1897 /* Otherwise -- delete. */ 1898 g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE); 1899 g_raid_destroy_disk(disk); 1900 } 1901 } 1902 1903 /* Write updated metadata to remaining disks. */ 1904 g_raid_md_write_intel(md, NULL, NULL, NULL); 1905 1906 /* Check if anything left except placeholders. */ 1907 if (g_raid_ndisks(sc, -1) == 1908 g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) 1909 g_raid_destroy_node(sc, 0); 1910 else 1911 g_raid_md_intel_refill(sc); 1912 return (error); 1913 } 1914 if (strcmp(verb, "insert") == 0) { 1915 if (*nargs < 2) { 1916 gctl_error(req, "Invalid number of arguments."); 1917 return (-1); 1918 } 1919 update = 0; 1920 for (i = 1; i < *nargs; i++) { 1921 /* Get disk name. */ 1922 snprintf(arg, sizeof(arg), "arg%d", i); 1923 diskname = gctl_get_asciiparam(req, arg); 1924 if (diskname == NULL) { 1925 gctl_error(req, "No disk name (%s).", arg); 1926 error = -3; 1927 break; 1928 } 1929 1930 /* Try to find provider with specified name. */ 1931 g_topology_lock(); 1932 cp = g_raid_open_consumer(sc, diskname); 1933 if (cp == NULL) { 1934 gctl_error(req, "Can't open disk '%s'.", 1935 diskname); 1936 g_topology_unlock(); 1937 error = -4; 1938 break; 1939 } 1940 pp = cp->provider; 1941 g_topology_unlock(); 1942 1943 /* Read disk serial. */ 1944 error = g_raid_md_get_label(cp, 1945 &serial[0], INTEL_SERIAL_LEN); 1946 if (error != 0) { 1947 gctl_error(req, 1948 "Can't get serial for provider '%s'.", 1949 diskname); 1950 g_raid_kill_consumer(sc, cp); 1951 error = -7; 1952 break; 1953 } 1954 1955 pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO); 1956 pd->pd_disk_pos = -1; 1957 1958 disk = g_raid_create_disk(sc); 1959 disk->d_consumer = cp; 1960 disk->d_md_data = (void *)pd; 1961 cp->private = disk; 1962 1963 /* Read kernel dumping information. */ 1964 disk->d_kd.offset = 0; 1965 disk->d_kd.length = OFF_MAX; 1966 len = sizeof(disk->d_kd); 1967 g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); 1968 if (disk->d_kd.di.dumper == NULL) 1969 G_RAID_DEBUG1(2, sc, 1970 "Dumping not supported by %s.", 1971 cp->provider->name); 1972 1973 memcpy(&pd->pd_disk_meta.serial[0], &serial[0], 1974 INTEL_SERIAL_LEN); 1975 pd->pd_disk_meta.sectors = pp->mediasize / pp->sectorsize; 1976 pd->pd_disk_meta.id = 0; 1977 pd->pd_disk_meta.flags = INTEL_F_SPARE; 1978 1979 /* Welcome the "new" disk. */ 1980 update += g_raid_md_intel_start_disk(disk); 1981 if (disk->d_state == G_RAID_DISK_S_SPARE) { 1982 intel_meta_write_spare(cp, &pd->pd_disk_meta); 1983 g_raid_destroy_disk(disk); 1984 } else if (disk->d_state != G_RAID_DISK_S_ACTIVE) { 1985 gctl_error(req, "Disk '%s' doesn't fit.", 1986 diskname); 1987 g_raid_destroy_disk(disk); 1988 error = -8; 1989 break; 1990 } 1991 } 1992 1993 /* Write new metadata if we changed something. */ 1994 if (update) 1995 g_raid_md_write_intel(md, NULL, NULL, NULL); 1996 return (error); 1997 } 1998 return (-100); 1999} 2000 2001static int 2002g_raid_md_write_intel(struct g_raid_md_object *md, struct g_raid_volume *tvol, 2003 struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk) 2004{ 2005 struct g_raid_softc *sc; 2006 struct g_raid_volume *vol; 2007 struct g_raid_subdisk *sd; 2008 struct g_raid_disk *disk; 2009 struct g_raid_md_intel_object *mdi; 2010 struct g_raid_md_intel_perdisk *pd; 2011 struct intel_raid_conf *meta; 2012 struct intel_raid_vol *mvol; 2013 struct intel_raid_map *mmap0, *mmap1; 2014 off_t sectorsize = 512, pos; 2015 const char *version, *cv; 2016 int vi, sdi, numdisks, len, state, stale; 2017 2018 sc = md->mdo_softc; 2019 mdi = (struct g_raid_md_intel_object *)md; 2020 2021 if (sc->sc_stopping == G_RAID_DESTROY_HARD) 2022 return (0); 2023 2024 /* Bump generation. Newly written metadata may differ from previous. */ 2025 mdi->mdio_generation++; 2026 2027 /* Count number of disks. */ 2028 numdisks = 0; 2029 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 2030 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 2031 if (pd->pd_disk_pos < 0) 2032 continue; 2033 numdisks++; 2034 if (disk->d_state == G_RAID_DISK_S_ACTIVE) { 2035 pd->pd_disk_meta.flags = 2036 INTEL_F_ONLINE | INTEL_F_ASSIGNED; 2037 } else if (disk->d_state == G_RAID_DISK_S_FAILED) { 2038 pd->pd_disk_meta.flags = INTEL_F_FAILED | INTEL_F_ASSIGNED; 2039 } else { 2040 pd->pd_disk_meta.flags = INTEL_F_ASSIGNED; 2041 if (pd->pd_disk_meta.id != 0xffffffff) { 2042 pd->pd_disk_meta.id = 0xffffffff; 2043 len = strlen(pd->pd_disk_meta.serial); 2044 len = min(len, INTEL_SERIAL_LEN - 3); 2045 strcpy(pd->pd_disk_meta.serial + len, ":0"); 2046 } 2047 } 2048 } 2049 2050 /* Fill anchor and disks. */ 2051 meta = malloc(INTEL_MAX_MD_SIZE(numdisks), 2052 M_MD_INTEL, M_WAITOK | M_ZERO); 2053 memcpy(&meta->intel_id[0], INTEL_MAGIC, sizeof(INTEL_MAGIC) - 1); 2054 meta->config_size = INTEL_MAX_MD_SIZE(numdisks); 2055 meta->config_id = mdi->mdio_config_id; 2056 meta->generation = mdi->mdio_generation; 2057 meta->attributes = INTEL_ATTR_CHECKSUM; 2058 meta->total_disks = numdisks; 2059 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 2060 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 2061 if (pd->pd_disk_pos < 0) 2062 continue; 2063 meta->disk[pd->pd_disk_pos] = pd->pd_disk_meta; 2064 } 2065 2066 /* Fill volumes and maps. */ 2067 vi = 0; 2068 version = INTEL_VERSION_1000; 2069 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 2070 if (vol->v_stopping) 2071 continue; 2072 mvol = intel_get_volume(meta, vi); 2073 2074 /* New metadata may have different volumes order. */ 2075 vol->v_md_data = (void *)(intptr_t)vi; 2076 2077 for (sdi = 0; sdi < vol->v_disks_count; sdi++) { 2078 sd = &vol->v_subdisks[sdi]; 2079 if (sd->sd_disk != NULL) 2080 break; 2081 } 2082 if (sdi >= vol->v_disks_count) 2083 panic("No any filled subdisk in volume"); 2084 if (vol->v_mediasize >= 0x20000000000llu) 2085 meta->attributes |= INTEL_ATTR_2TB; 2086 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0) 2087 meta->attributes |= INTEL_ATTR_RAID0; 2088 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1) 2089 meta->attributes |= INTEL_ATTR_RAID1; 2090 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5) 2091 meta->attributes |= INTEL_ATTR_RAID5; 2092 else 2093 meta->attributes |= INTEL_ATTR_RAID10; 2094 2095 if (meta->attributes & INTEL_ATTR_2TB) 2096 cv = INTEL_VERSION_1300; 2097// else if (dev->status == DEV_CLONE_N_GO) 2098// cv = INTEL_VERSION_1206; 2099 else if (vol->v_disks_count > 4) 2100 cv = INTEL_VERSION_1204; 2101 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5) 2102 cv = INTEL_VERSION_1202; 2103 else if (vol->v_disks_count > 2) 2104 cv = INTEL_VERSION_1201; 2105 else if (vi > 0) 2106 cv = INTEL_VERSION_1200; 2107 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1) 2108 cv = INTEL_VERSION_1100; 2109 else 2110 cv = INTEL_VERSION_1000; 2111 if (strcmp(cv, version) > 0) 2112 version = cv; 2113 2114 strlcpy(&mvol->name[0], vol->v_name, sizeof(mvol->name)); 2115 mvol->total_sectors = vol->v_mediasize / sectorsize; 2116 2117 /* Check for any recovery in progress. */ 2118 state = G_RAID_SUBDISK_S_ACTIVE; 2119 pos = 0x7fffffffffffffffllu; 2120 stale = 0; 2121 for (sdi = 0; sdi < vol->v_disks_count; sdi++) { 2122 sd = &vol->v_subdisks[sdi]; 2123 if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD) 2124 state = G_RAID_SUBDISK_S_REBUILD; 2125 else if (sd->sd_state == G_RAID_SUBDISK_S_RESYNC && 2126 state != G_RAID_SUBDISK_S_REBUILD) 2127 state = G_RAID_SUBDISK_S_RESYNC; 2128 else if (sd->sd_state == G_RAID_SUBDISK_S_STALE) 2129 stale = 1; 2130 if ((sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2131 sd->sd_state == G_RAID_SUBDISK_S_RESYNC) && 2132 sd->sd_rebuild_pos < pos) 2133 pos = sd->sd_rebuild_pos; 2134 } 2135 if (state == G_RAID_SUBDISK_S_REBUILD) { 2136 mvol->migr_state = 1; 2137 mvol->migr_type = INTEL_MT_REBUILD; 2138 } else if (state == G_RAID_SUBDISK_S_RESYNC) { 2139 mvol->migr_state = 1; 2140 /* mvol->migr_type = INTEL_MT_REPAIR; */ 2141 mvol->migr_type = INTEL_MT_VERIFY; 2142 mvol->state |= INTEL_ST_VERIFY_AND_FIX; 2143 } else 2144 mvol->migr_state = 0; 2145 mvol->dirty = (vol->v_dirty || stale); 2146 2147 mmap0 = intel_get_map(mvol, 0); 2148 2149 /* Write map / common part of two maps. */ 2150 mmap0->offset = sd->sd_offset / sectorsize; 2151 mmap0->disk_sectors = sd->sd_size / sectorsize; 2152 mmap0->strip_sectors = vol->v_strip_size / sectorsize; 2153 if (vol->v_state == G_RAID_VOLUME_S_BROKEN) 2154 mmap0->status = INTEL_S_FAILURE; 2155 else if (vol->v_state == G_RAID_VOLUME_S_DEGRADED) 2156 mmap0->status = INTEL_S_DEGRADED; 2157 else 2158 mmap0->status = INTEL_S_READY; 2159 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0) 2160 mmap0->type = INTEL_T_RAID0; 2161 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 || 2162 vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) 2163 mmap0->type = INTEL_T_RAID1; 2164 else 2165 mmap0->type = INTEL_T_RAID5; 2166 mmap0->total_disks = vol->v_disks_count; 2167 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1) 2168 mmap0->total_domains = vol->v_disks_count; 2169 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) 2170 mmap0->total_domains = 2; 2171 else 2172 mmap0->total_domains = 1; 2173 mmap0->stripe_count = sd->sd_size / vol->v_strip_size / 2174 mmap0->total_domains; 2175 mmap0->failed_disk_num = 0xff; 2176 mmap0->ddf = 1; 2177 2178 /* If there are two maps - copy common and update. */ 2179 if (mvol->migr_state) { 2180 mvol->curr_migr_unit = pos / 2181 vol->v_strip_size / mmap0->total_domains; 2182 mmap1 = intel_get_map(mvol, 1); 2183 memcpy(mmap1, mmap0, sizeof(struct intel_raid_map)); 2184 mmap0->status = INTEL_S_READY; 2185 } else 2186 mmap1 = NULL; 2187 2188 /* Write disk indexes and put rebuild flags. */ 2189 for (sdi = 0; sdi < vol->v_disks_count; sdi++) { 2190 sd = &vol->v_subdisks[sdi]; 2191 pd = (struct g_raid_md_intel_perdisk *) 2192 sd->sd_disk->d_md_data; 2193 mmap0->disk_idx[sdi] = pd->pd_disk_pos; 2194 if (mvol->migr_state) 2195 mmap1->disk_idx[sdi] = pd->pd_disk_pos; 2196 if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2197 sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 2198 mmap1->disk_idx[sdi] |= INTEL_DI_RBLD; 2199 } else if (sd->sd_state != G_RAID_SUBDISK_S_ACTIVE && 2200 sd->sd_state != G_RAID_SUBDISK_S_STALE) { 2201 mmap0->disk_idx[sdi] |= INTEL_DI_RBLD; 2202 if (mvol->migr_state) 2203 mmap1->disk_idx[sdi] |= INTEL_DI_RBLD; 2204 } 2205 if ((sd->sd_state == G_RAID_SUBDISK_S_NONE || 2206 sd->sd_state == G_RAID_SUBDISK_S_FAILED) && 2207 mmap0->failed_disk_num == 0xff) { 2208 mmap0->failed_disk_num = sdi; 2209 if (mvol->migr_state) 2210 mmap1->failed_disk_num = sdi; 2211 } 2212 } 2213 vi++; 2214 } 2215 meta->total_volumes = vi; 2216 if (strcmp(version, INTEL_VERSION_1300) != 0) 2217 meta->attributes &= INTEL_ATTR_CHECKSUM; 2218 memcpy(&meta->version[0], version, sizeof(INTEL_VERSION_1000) - 1); 2219 2220 /* We are done. Print meta data and store them to disks. */ 2221 g_raid_md_intel_print(meta); 2222 if (mdi->mdio_meta != NULL) 2223 free(mdi->mdio_meta, M_MD_INTEL); 2224 mdi->mdio_meta = meta; 2225 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 2226 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 2227 if (disk->d_state != G_RAID_DISK_S_ACTIVE) 2228 continue; 2229 if (pd->pd_meta != NULL) { 2230 free(pd->pd_meta, M_MD_INTEL); 2231 pd->pd_meta = NULL; 2232 } 2233 pd->pd_meta = intel_meta_copy(meta); 2234 intel_meta_write(disk->d_consumer, meta); 2235 } 2236 return (0); 2237} 2238 2239static int 2240g_raid_md_fail_disk_intel(struct g_raid_md_object *md, 2241 struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk) 2242{ 2243 struct g_raid_softc *sc; 2244 struct g_raid_md_intel_object *mdi; 2245 struct g_raid_md_intel_perdisk *pd; 2246 struct g_raid_subdisk *sd; 2247 2248 sc = md->mdo_softc; 2249 mdi = (struct g_raid_md_intel_object *)md; 2250 pd = (struct g_raid_md_intel_perdisk *)tdisk->d_md_data; 2251 2252 /* We can't fail disk that is not a part of array now. */ 2253 if (pd->pd_disk_pos < 0) 2254 return (-1); 2255 2256 /* 2257 * Mark disk as failed in metadata and try to write that metadata 2258 * to the disk itself to prevent it's later resurrection as STALE. 2259 */ 2260 mdi->mdio_meta->disk[pd->pd_disk_pos].flags = INTEL_F_FAILED; 2261 pd->pd_disk_meta.flags = INTEL_F_FAILED; 2262 g_raid_md_intel_print(mdi->mdio_meta); 2263 if (tdisk->d_consumer) 2264 intel_meta_write(tdisk->d_consumer, mdi->mdio_meta); 2265 2266 /* Change states. */ 2267 g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED); 2268 TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) { 2269 g_raid_change_subdisk_state(sd, 2270 G_RAID_SUBDISK_S_FAILED); 2271 g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED, 2272 G_RAID_EVENT_SUBDISK); 2273 } 2274 2275 /* Write updated metadata to remaining disks. */ 2276 g_raid_md_write_intel(md, NULL, NULL, tdisk); 2277 2278 /* Check if anything left except placeholders. */ 2279 if (g_raid_ndisks(sc, -1) == 2280 g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) 2281 g_raid_destroy_node(sc, 0); 2282 else 2283 g_raid_md_intel_refill(sc); 2284 return (0); 2285} 2286 2287static int 2288g_raid_md_free_disk_intel(struct g_raid_md_object *md, 2289 struct g_raid_disk *disk) 2290{ 2291 struct g_raid_md_intel_perdisk *pd; 2292 2293 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 2294 if (pd->pd_meta != NULL) { 2295 free(pd->pd_meta, M_MD_INTEL); 2296 pd->pd_meta = NULL; 2297 } 2298 free(pd, M_MD_INTEL); 2299 disk->d_md_data = NULL; 2300 return (0); 2301} 2302 2303static int 2304g_raid_md_free_intel(struct g_raid_md_object *md) 2305{ 2306 struct g_raid_md_intel_object *mdi; 2307 2308 mdi = (struct g_raid_md_intel_object *)md; 2309 if (!mdi->mdio_started) { 2310 mdi->mdio_started = 0; 2311 callout_stop(&mdi->mdio_start_co); 2312 G_RAID_DEBUG1(1, md->mdo_softc, 2313 "root_mount_rel %p", mdi->mdio_rootmount); 2314 root_mount_rel(mdi->mdio_rootmount); 2315 mdi->mdio_rootmount = NULL; 2316 } 2317 if (mdi->mdio_meta != NULL) { 2318 free(mdi->mdio_meta, M_MD_INTEL); 2319 mdi->mdio_meta = NULL; 2320 } 2321 return (0); 2322} 2323 2324G_RAID_MD_DECLARE(g_raid_md_intel); 2325