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("$FreeBSD$"); 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 45 static MALLOC_DEFINE(M_MD_INTEL, "md_intel_data", "GEOM_RAID Intel metadata"); 46 47 struct 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 offset_hi; 68 uint32_t disk_sectors_hi; 69 uint32_t stripe_count_hi; 70 uint32_t filler_2[4]; 71 uint32_t disk_idx[1]; /* total_disks entries. */ 72 #define INTEL_DI_IDX 0x00ffffff 73 #define INTEL_DI_RBLD 0x01000000 74 } __packed; 75 76 struct intel_raid_vol { 77 uint8_t name[16]; 78 u_int64_t total_sectors __packed; 79 uint32_t state; 80 #define INTEL_ST_BOOTABLE 0x00000001 81 #define INTEL_ST_BOOT_DEVICE 0x00000002 82 #define INTEL_ST_READ_COALESCING 0x00000004 83 #define INTEL_ST_WRITE_COALESCING 0x00000008 84 #define INTEL_ST_LAST_SHUTDOWN_DIRTY 0x00000010 85 #define INTEL_ST_HIDDEN_AT_BOOT 0x00000020 86 #define INTEL_ST_CURRENTLY_HIDDEN 0x00000040 87 #define INTEL_ST_VERIFY_AND_FIX 0x00000080 88 #define INTEL_ST_MAP_STATE_UNINIT 0x00000100 89 #define INTEL_ST_NO_AUTO_RECOVERY 0x00000200 90 #define INTEL_ST_CLONE_N_GO 0x00000400 91 #define INTEL_ST_CLONE_MAN_SYNC 0x00000800 92 #define INTEL_ST_CNG_MASTER_DISK_NUM 0x00001000 93 uint32_t reserved; 94 uint8_t migr_priority; 95 uint8_t num_sub_vols; 96 uint8_t tid; 97 uint8_t cng_master_disk; 98 uint16_t cache_policy; 99 uint8_t cng_state; 100 uint8_t cng_sub_state; 101 uint32_t filler_0[10]; 102 103 uint32_t curr_migr_unit; 104 uint32_t checkpoint_id; 105 uint8_t migr_state; 106 uint8_t migr_type; 107 #define INTEL_MT_INIT 0 108 #define INTEL_MT_REBUILD 1 109 #define INTEL_MT_VERIFY 2 110 #define INTEL_MT_GEN_MIGR 3 111 #define INTEL_MT_STATE_CHANGE 4 112 #define INTEL_MT_REPAIR 5 113 uint8_t dirty; 114 uint8_t fs_state; 115 uint16_t verify_errors; 116 uint16_t bad_blocks; 117 uint32_t curr_migr_unit_hi; 118 uint32_t filler_1[3]; 119 struct intel_raid_map map[1]; /* 2 entries if migr_state != 0. */ 120 } __packed; 121 122 struct intel_raid_disk { 123 #define INTEL_SERIAL_LEN 16 124 uint8_t serial[INTEL_SERIAL_LEN]; 125 uint32_t sectors; 126 uint32_t id; 127 uint32_t flags; 128 #define INTEL_F_SPARE 0x01 129 #define INTEL_F_ASSIGNED 0x02 130 #define INTEL_F_FAILED 0x04 131 #define INTEL_F_ONLINE 0x08 132 uint32_t owner_cfg_num; 133 uint32_t sectors_hi; 134 uint32_t filler[3]; 135 } __packed; 136 137 struct intel_raid_conf { 138 uint8_t intel_id[24]; 139 #define INTEL_MAGIC "Intel Raid ISM Cfg Sig. " 140 141 uint8_t version[6]; 142 #define INTEL_VERSION_1000 "1.0.00" /* RAID0 */ 143 #define INTEL_VERSION_1100 "1.1.00" /* RAID1 */ 144 #define INTEL_VERSION_1200 "1.2.00" /* Many volumes */ 145 #define INTEL_VERSION_1201 "1.2.01" /* 3 or 4 disks */ 146 #define INTEL_VERSION_1202 "1.2.02" /* RAID5 */ 147 #define INTEL_VERSION_1204 "1.2.04" /* 5 or 6 disks */ 148 #define INTEL_VERSION_1206 "1.2.06" /* CNG */ 149 #define INTEL_VERSION_1300 "1.3.00" /* Attributes */ 150 151 uint8_t dummy_0[2]; 152 uint32_t checksum; 153 uint32_t config_size; 154 uint32_t config_id; 155 uint32_t generation; 156 uint32_t error_log_size; 157 uint32_t attributes; 158 #define INTEL_ATTR_RAID0 0x00000001 159 #define INTEL_ATTR_RAID1 0x00000002 160 #define INTEL_ATTR_RAID10 0x00000004 161 #define INTEL_ATTR_RAID1E 0x00000008 162 #define INTEL_ATTR_RAID5 0x00000010 163 #define INTEL_ATTR_RAIDCNG 0x00000020 164 #define INTEL_ATTR_2TB 0x20000000 165 #define INTEL_ATTR_PM 0x40000000 166 #define INTEL_ATTR_CHECKSUM 0x80000000 167 168 uint8_t total_disks; 169 uint8_t total_volumes; 170 uint8_t dummy_2[2]; 171 uint32_t filler_0[39]; 172 struct intel_raid_disk disk[1]; /* total_disks entries. */ 173 /* Here goes total_volumes of struct intel_raid_vol. */ 174 } __packed; 175 176 #define INTEL_MAX_MD_SIZE(ndisks) \ 177 (sizeof(struct intel_raid_conf) + \ 178 sizeof(struct intel_raid_disk) * (ndisks - 1) + \ 179 sizeof(struct intel_raid_vol) * 2 + \ 180 sizeof(struct intel_raid_map) * 2 + \ 181 sizeof(uint32_t) * (ndisks - 1) * 4) 182 183 struct g_raid_md_intel_perdisk { 184 struct intel_raid_conf *pd_meta; 185 int pd_disk_pos; 186 struct intel_raid_disk pd_disk_meta; 187 }; 188 189 struct g_raid_md_intel_object { 190 struct g_raid_md_object mdio_base; 191 uint32_t mdio_config_id; 192 uint32_t mdio_generation; 193 struct intel_raid_conf *mdio_meta; 194 struct callout mdio_start_co; /* STARTING state timer. */ 195 int mdio_disks_present; 196 int mdio_started; 197 int mdio_incomplete; 198 struct root_hold_token *mdio_rootmount; /* Root mount delay token. */ 199 }; 200 201 static g_raid_md_create_t g_raid_md_create_intel; 202 static g_raid_md_taste_t g_raid_md_taste_intel; 203 static g_raid_md_event_t g_raid_md_event_intel; 204 static g_raid_md_ctl_t g_raid_md_ctl_intel; 205 static g_raid_md_write_t g_raid_md_write_intel; 206 static g_raid_md_fail_disk_t g_raid_md_fail_disk_intel; 207 static g_raid_md_free_disk_t g_raid_md_free_disk_intel; 208 static g_raid_md_free_t g_raid_md_free_intel; 209 210 static kobj_method_t g_raid_md_intel_methods[] = { 211 KOBJMETHOD(g_raid_md_create, g_raid_md_create_intel), 212 KOBJMETHOD(g_raid_md_taste, g_raid_md_taste_intel), 213 KOBJMETHOD(g_raid_md_event, g_raid_md_event_intel), 214 KOBJMETHOD(g_raid_md_ctl, g_raid_md_ctl_intel), 215 KOBJMETHOD(g_raid_md_write, g_raid_md_write_intel), 216 KOBJMETHOD(g_raid_md_fail_disk, g_raid_md_fail_disk_intel), 217 KOBJMETHOD(g_raid_md_free_disk, g_raid_md_free_disk_intel), 218 KOBJMETHOD(g_raid_md_free, g_raid_md_free_intel), 219 { 0, 0 } 220 }; 221 222 static struct g_raid_md_class g_raid_md_intel_class = { 223 "Intel", 224 g_raid_md_intel_methods, 225 sizeof(struct g_raid_md_intel_object), 226 .mdc_priority = 100 227 }; 228 229 230 static struct intel_raid_map * 231 intel_get_map(struct intel_raid_vol *mvol, int i) 232 { 233 struct intel_raid_map *mmap; 234 235 if (i > (mvol->migr_state ? 1 : 0)) 236 return (NULL); 237 mmap = &mvol->map[0]; 238 for (; i > 0; i--) { 239 mmap = (struct intel_raid_map *) 240 &mmap->disk_idx[mmap->total_disks]; 241 } 242 return ((struct intel_raid_map *)mmap); 243 } 244 245 static struct intel_raid_vol * 246 intel_get_volume(struct intel_raid_conf *meta, int i) 247 { 248 struct intel_raid_vol *mvol; 249 struct intel_raid_map *mmap; 250 251 if (i > 1) 252 return (NULL); 253 mvol = (struct intel_raid_vol *)&meta->disk[meta->total_disks]; 254 for (; i > 0; i--) { 255 mmap = intel_get_map(mvol, mvol->migr_state ? 1 : 0); 256 mvol = (struct intel_raid_vol *) 257 &mmap->disk_idx[mmap->total_disks]; 258 } 259 return (mvol); 260 } 261 262 static off_t 263 intel_get_map_offset(struct intel_raid_map *mmap) 264 { 265 off_t offset = (off_t)mmap->offset_hi << 32; 266 267 offset += mmap->offset; 268 return (offset); 269 } 270 271 static void 272 intel_set_map_offset(struct intel_raid_map *mmap, off_t offset) 273 { 274 275 mmap->offset = offset & 0xffffffff; 276 mmap->offset_hi = offset >> 32; 277 } 278 279 static off_t 280 intel_get_map_disk_sectors(struct intel_raid_map *mmap) 281 { 282 off_t disk_sectors = (off_t)mmap->disk_sectors_hi << 32; 283 284 disk_sectors += mmap->disk_sectors; 285 return (disk_sectors); 286 } 287 288 static void 289 intel_set_map_disk_sectors(struct intel_raid_map *mmap, off_t disk_sectors) 290 { 291 292 mmap->disk_sectors = disk_sectors & 0xffffffff; 293 mmap->disk_sectors_hi = disk_sectors >> 32; 294 } 295 296 static void 297 intel_set_map_stripe_count(struct intel_raid_map *mmap, off_t stripe_count) 298 { 299 300 mmap->stripe_count = stripe_count & 0xffffffff; 301 mmap->stripe_count_hi = stripe_count >> 32; 302 } 303 304 static off_t 305 intel_get_disk_sectors(struct intel_raid_disk *disk) 306 { 307 off_t sectors = (off_t)disk->sectors_hi << 32; 308 309 sectors += disk->sectors; 310 return (sectors); 311 } 312 313 static void 314 intel_set_disk_sectors(struct intel_raid_disk *disk, off_t sectors) 315 { 316 317 disk->sectors = sectors & 0xffffffff; 318 disk->sectors_hi = sectors >> 32; 319 } 320 321 static off_t 322 intel_get_vol_curr_migr_unit(struct intel_raid_vol *vol) 323 { 324 off_t curr_migr_unit = (off_t)vol->curr_migr_unit_hi << 32; 325 326 curr_migr_unit += vol->curr_migr_unit; 327 return (curr_migr_unit); 328 } 329 330 static void 331 intel_set_vol_curr_migr_unit(struct intel_raid_vol *vol, off_t curr_migr_unit) 332 { 333 334 vol->curr_migr_unit = curr_migr_unit & 0xffffffff; 335 vol->curr_migr_unit_hi = curr_migr_unit >> 32; 336 } 337 338 static void 339 g_raid_md_intel_print(struct intel_raid_conf *meta) 340 { 341 struct intel_raid_vol *mvol; 342 struct intel_raid_map *mmap; 343 int i, j, k; 344 345 if (g_raid_debug < 1) 346 return; 347 348 printf("********* ATA Intel MatrixRAID Metadata *********\n"); 349 printf("intel_id <%.24s>\n", meta->intel_id); 350 printf("version <%.6s>\n", meta->version); 351 printf("checksum 0x%08x\n", meta->checksum); 352 printf("config_size 0x%08x\n", meta->config_size); 353 printf("config_id 0x%08x\n", meta->config_id); 354 printf("generation 0x%08x\n", meta->generation); 355 printf("attributes 0x%08x\n", meta->attributes); 356 printf("total_disks %u\n", meta->total_disks); 357 printf("total_volumes %u\n", meta->total_volumes); 358 printf("DISK# serial disk_sectors disk_sectors_hi disk_id flags\n"); 359 for (i = 0; i < meta->total_disks; i++ ) { 360 printf(" %d <%.16s> %u %u 0x%08x 0x%08x\n", i, 361 meta->disk[i].serial, meta->disk[i].sectors, 362 meta->disk[i].sectors_hi, 363 meta->disk[i].id, meta->disk[i].flags); 364 } 365 for (i = 0; i < meta->total_volumes; i++) { 366 mvol = intel_get_volume(meta, i); 367 printf(" ****** Volume %d ******\n", i); 368 printf(" name %.16s\n", mvol->name); 369 printf(" total_sectors %ju\n", mvol->total_sectors); 370 printf(" state %u\n", mvol->state); 371 printf(" reserved %u\n", mvol->reserved); 372 printf(" curr_migr_unit %u\n", mvol->curr_migr_unit); 373 printf(" curr_migr_unit_hi %u\n", mvol->curr_migr_unit_hi); 374 printf(" checkpoint_id %u\n", mvol->checkpoint_id); 375 printf(" migr_state %u\n", mvol->migr_state); 376 printf(" migr_type %u\n", mvol->migr_type); 377 printf(" dirty %u\n", mvol->dirty); 378 379 for (j = 0; j < (mvol->migr_state ? 2 : 1); j++) { 380 printf(" *** Map %d ***\n", j); 381 mmap = intel_get_map(mvol, j); 382 printf(" offset %u\n", mmap->offset); 383 printf(" offset_hi %u\n", mmap->offset_hi); 384 printf(" disk_sectors %u\n", mmap->disk_sectors); 385 printf(" disk_sectors_hi %u\n", mmap->disk_sectors_hi); 386 printf(" stripe_count %u\n", mmap->stripe_count); 387 printf(" stripe_count_hi %u\n", mmap->stripe_count_hi); 388 printf(" strip_sectors %u\n", mmap->strip_sectors); 389 printf(" status %u\n", mmap->status); 390 printf(" type %u\n", mmap->type); 391 printf(" total_disks %u\n", mmap->total_disks); 392 printf(" total_domains %u\n", mmap->total_domains); 393 printf(" failed_disk_num %u\n", mmap->failed_disk_num); 394 printf(" ddf %u\n", mmap->ddf); 395 printf(" disk_idx "); 396 for (k = 0; k < mmap->total_disks; k++) 397 printf(" 0x%08x", mmap->disk_idx[k]); 398 printf("\n"); 399 } 400 } 401 printf("=================================================\n"); 402 } 403 404 static struct intel_raid_conf * 405 intel_meta_copy(struct intel_raid_conf *meta) 406 { 407 struct intel_raid_conf *nmeta; 408 409 nmeta = malloc(meta->config_size, M_MD_INTEL, M_WAITOK); 410 memcpy(nmeta, meta, meta->config_size); 411 return (nmeta); 412 } 413 414 static int 415 intel_meta_find_disk(struct intel_raid_conf *meta, char *serial) 416 { 417 int pos; 418 419 for (pos = 0; pos < meta->total_disks; pos++) { 420 if (strncmp(meta->disk[pos].serial, 421 serial, INTEL_SERIAL_LEN) == 0) 422 return (pos); 423 } 424 return (-1); 425 } 426 427 static struct intel_raid_conf * 428 intel_meta_read(struct g_consumer *cp) 429 { 430 struct g_provider *pp; 431 struct intel_raid_conf *meta; 432 struct intel_raid_vol *mvol; 433 struct intel_raid_map *mmap; 434 char *buf; 435 int error, i, j, k, left, size; 436 uint32_t checksum, *ptr; 437 438 pp = cp->provider; 439 440 /* Read the anchor sector. */ 441 buf = g_read_data(cp, 442 pp->mediasize - pp->sectorsize * 2, pp->sectorsize, &error); 443 if (buf == NULL) { 444 G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).", 445 pp->name, error); 446 return (NULL); 447 } 448 meta = (struct intel_raid_conf *)buf; 449 450 /* Check if this is an Intel RAID struct */ 451 if (strncmp(meta->intel_id, INTEL_MAGIC, strlen(INTEL_MAGIC))) { 452 G_RAID_DEBUG(1, "Intel signature check failed on %s", pp->name); 453 g_free(buf); 454 return (NULL); 455 } 456 if (meta->config_size > 65536 || 457 meta->config_size < sizeof(struct intel_raid_conf)) { 458 G_RAID_DEBUG(1, "Intel metadata size looks wrong: %d", 459 meta->config_size); 460 g_free(buf); 461 return (NULL); 462 } 463 size = meta->config_size; 464 meta = malloc(size, M_MD_INTEL, M_WAITOK); 465 memcpy(meta, buf, min(size, pp->sectorsize)); 466 g_free(buf); 467 468 /* Read all the rest, if needed. */ 469 if (meta->config_size > pp->sectorsize) { 470 left = (meta->config_size - 1) / pp->sectorsize; 471 buf = g_read_data(cp, 472 pp->mediasize - pp->sectorsize * (2 + left), 473 pp->sectorsize * left, &error); 474 if (buf == NULL) { 475 G_RAID_DEBUG(1, "Cannot read remaining metadata" 476 " part from %s (error=%d).", 477 pp->name, error); 478 free(meta, M_MD_INTEL); 479 return (NULL); 480 } 481 memcpy(((char *)meta) + pp->sectorsize, buf, 482 pp->sectorsize * left); 483 g_free(buf); 484 } 485 486 /* Check metadata checksum. */ 487 for (checksum = 0, ptr = (uint32_t *)meta, i = 0; 488 i < (meta->config_size / sizeof(uint32_t)); i++) { 489 checksum += *ptr++; 490 } 491 checksum -= meta->checksum; 492 if (checksum != meta->checksum) { 493 G_RAID_DEBUG(1, "Intel checksum check failed on %s", pp->name); 494 free(meta, M_MD_INTEL); 495 return (NULL); 496 } 497 498 /* Validate metadata size. */ 499 size = sizeof(struct intel_raid_conf) + 500 sizeof(struct intel_raid_disk) * (meta->total_disks - 1) + 501 sizeof(struct intel_raid_vol) * meta->total_volumes; 502 if (size > meta->config_size) { 503 badsize: 504 G_RAID_DEBUG(1, "Intel metadata size incorrect %d < %d", 505 meta->config_size, size); 506 free(meta, M_MD_INTEL); 507 return (NULL); 508 } 509 for (i = 0; i < meta->total_volumes; i++) { 510 mvol = intel_get_volume(meta, i); 511 mmap = intel_get_map(mvol, 0); 512 size += 4 * (mmap->total_disks - 1); 513 if (size > meta->config_size) 514 goto badsize; 515 if (mvol->migr_state) { 516 size += sizeof(struct intel_raid_map); 517 if (size > meta->config_size) 518 goto badsize; 519 mmap = intel_get_map(mvol, 1); 520 size += 4 * (mmap->total_disks - 1); 521 if (size > meta->config_size) 522 goto badsize; 523 } 524 } 525 526 /* Validate disk indexes. */ 527 for (i = 0; i < meta->total_volumes; i++) { 528 mvol = intel_get_volume(meta, i); 529 for (j = 0; j < (mvol->migr_state ? 2 : 1); j++) { 530 mmap = intel_get_map(mvol, j); 531 for (k = 0; k < mmap->total_disks; k++) { 532 if ((mmap->disk_idx[k] & INTEL_DI_IDX) > 533 meta->total_disks) { 534 G_RAID_DEBUG(1, "Intel metadata disk" 535 " index %d too big (>%d)", 536 mmap->disk_idx[k] & INTEL_DI_IDX, 537 meta->total_disks); 538 free(meta, M_MD_INTEL); 539 return (NULL); 540 } 541 } 542 } 543 } 544 545 /* Validate migration types. */ 546 for (i = 0; i < meta->total_volumes; i++) { 547 mvol = intel_get_volume(meta, i); 548 if (mvol->migr_state && 549 mvol->migr_type != INTEL_MT_INIT && 550 mvol->migr_type != INTEL_MT_REBUILD && 551 mvol->migr_type != INTEL_MT_VERIFY && 552 mvol->migr_type != INTEL_MT_REPAIR) { 553 G_RAID_DEBUG(1, "Intel metadata has unsupported" 554 " migration type %d", mvol->migr_type); 555 free(meta, M_MD_INTEL); 556 return (NULL); 557 } 558 } 559 560 return (meta); 561 } 562 563 static int 564 intel_meta_write(struct g_consumer *cp, struct intel_raid_conf *meta) 565 { 566 struct g_provider *pp; 567 char *buf; 568 int error, i, sectors; 569 uint32_t checksum, *ptr; 570 571 pp = cp->provider; 572 573 /* Recalculate checksum for case if metadata were changed. */ 574 meta->checksum = 0; 575 for (checksum = 0, ptr = (uint32_t *)meta, i = 0; 576 i < (meta->config_size / sizeof(uint32_t)); i++) { 577 checksum += *ptr++; 578 } 579 meta->checksum = checksum; 580 581 /* Create and fill buffer. */ 582 sectors = (meta->config_size + pp->sectorsize - 1) / pp->sectorsize; 583 buf = malloc(sectors * pp->sectorsize, M_MD_INTEL, M_WAITOK | M_ZERO); 584 if (sectors > 1) { 585 memcpy(buf, ((char *)meta) + pp->sectorsize, 586 (sectors - 1) * pp->sectorsize); 587 } 588 memcpy(buf + (sectors - 1) * pp->sectorsize, meta, pp->sectorsize); 589 590 error = g_write_data(cp, 591 pp->mediasize - pp->sectorsize * (1 + sectors), 592 buf, pp->sectorsize * sectors); 593 if (error != 0) { 594 G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).", 595 pp->name, error); 596 } 597 598 free(buf, M_MD_INTEL); 599 return (error); 600 } 601 602 static int 603 intel_meta_erase(struct g_consumer *cp) 604 { 605 struct g_provider *pp; 606 char *buf; 607 int error; 608 609 pp = cp->provider; 610 buf = malloc(pp->sectorsize, M_MD_INTEL, M_WAITOK | M_ZERO); 611 error = g_write_data(cp, 612 pp->mediasize - 2 * pp->sectorsize, 613 buf, pp->sectorsize); 614 if (error != 0) { 615 G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).", 616 pp->name, error); 617 } 618 free(buf, M_MD_INTEL); 619 return (error); 620 } 621 622 static int 623 intel_meta_write_spare(struct g_consumer *cp, struct intel_raid_disk *d) 624 { 625 struct intel_raid_conf *meta; 626 int error; 627 628 /* Fill anchor and single disk. */ 629 meta = malloc(INTEL_MAX_MD_SIZE(1), M_MD_INTEL, M_WAITOK | M_ZERO); 630 memcpy(&meta->intel_id[0], INTEL_MAGIC, sizeof(INTEL_MAGIC) - 1); 631 memcpy(&meta->version[0], INTEL_VERSION_1000, 632 sizeof(INTEL_VERSION_1000) - 1); 633 meta->config_size = INTEL_MAX_MD_SIZE(1); 634 meta->config_id = arc4random(); 635 meta->generation = 1; 636 meta->total_disks = 1; 637 meta->disk[0] = *d; 638 error = intel_meta_write(cp, meta); 639 free(meta, M_MD_INTEL); 640 return (error); 641 } 642 643 static struct g_raid_disk * 644 g_raid_md_intel_get_disk(struct g_raid_softc *sc, int id) 645 { 646 struct g_raid_disk *disk; 647 struct g_raid_md_intel_perdisk *pd; 648 649 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 650 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 651 if (pd->pd_disk_pos == id) 652 break; 653 } 654 return (disk); 655 } 656 657 static int 658 g_raid_md_intel_supported(int level, int qual, int disks, int force) 659 { 660 661 switch (level) { 662 case G_RAID_VOLUME_RL_RAID0: 663 if (disks < 1) 664 return (0); 665 if (!force && (disks < 2 || disks > 6)) 666 return (0); 667 break; 668 case G_RAID_VOLUME_RL_RAID1: 669 if (disks < 1) 670 return (0); 671 if (!force && (disks != 2)) 672 return (0); 673 break; 674 case G_RAID_VOLUME_RL_RAID1E: 675 if (disks < 2) 676 return (0); 677 if (!force && (disks != 4)) 678 return (0); 679 break; 680 case G_RAID_VOLUME_RL_RAID5: 681 if (disks < 3) 682 return (0); 683 if (!force && disks > 6) 684 return (0); 685 break; 686 default: 687 return (0); 688 } 689 if (qual != G_RAID_VOLUME_RLQ_NONE) 690 return (0); 691 return (1); 692 } 693 694 static struct g_raid_volume * 695 g_raid_md_intel_get_volume(struct g_raid_softc *sc, int id) 696 { 697 struct g_raid_volume *mvol; 698 699 TAILQ_FOREACH(mvol, &sc->sc_volumes, v_next) { 700 if ((intptr_t)(mvol->v_md_data) == id) 701 break; 702 } 703 return (mvol); 704 } 705 706 static int 707 g_raid_md_intel_start_disk(struct g_raid_disk *disk) 708 { 709 struct g_raid_softc *sc; 710 struct g_raid_subdisk *sd, *tmpsd; 711 struct g_raid_disk *olddisk, *tmpdisk; 712 struct g_raid_md_object *md; 713 struct g_raid_md_intel_object *mdi; 714 struct g_raid_md_intel_perdisk *pd, *oldpd; 715 struct intel_raid_conf *meta; 716 struct intel_raid_vol *mvol; 717 struct intel_raid_map *mmap0, *mmap1; 718 int disk_pos, resurrection = 0; 719 720 sc = disk->d_softc; 721 md = sc->sc_md; 722 mdi = (struct g_raid_md_intel_object *)md; 723 meta = mdi->mdio_meta; 724 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 725 olddisk = NULL; 726 727 /* Find disk position in metadata by it's serial. */ 728 disk_pos = intel_meta_find_disk(meta, pd->pd_disk_meta.serial); 729 if (disk_pos < 0) { 730 G_RAID_DEBUG1(1, sc, "Unknown, probably new or stale disk"); 731 /* Failed stale disk is useless for us. */ 732 if (pd->pd_disk_meta.flags & INTEL_F_FAILED) { 733 g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE_FAILED); 734 return (0); 735 } 736 /* If we are in the start process, that's all for now. */ 737 if (!mdi->mdio_started) 738 goto nofit; 739 /* 740 * If we have already started - try to get use of the disk. 741 * Try to replace OFFLINE disks first, then FAILED. 742 */ 743 TAILQ_FOREACH(tmpdisk, &sc->sc_disks, d_next) { 744 if (tmpdisk->d_state != G_RAID_DISK_S_OFFLINE && 745 tmpdisk->d_state != G_RAID_DISK_S_FAILED) 746 continue; 747 /* Make sure this disk is big enough. */ 748 TAILQ_FOREACH(sd, &tmpdisk->d_subdisks, sd_next) { 749 off_t disk_sectors = 750 intel_get_disk_sectors(&pd->pd_disk_meta); 751 752 if (sd->sd_offset + sd->sd_size + 4096 > 753 disk_sectors * 512) { 754 G_RAID_DEBUG1(1, sc, 755 "Disk too small (%llu < %llu)", 756 (unsigned long long) 757 disk_sectors * 512, 758 (unsigned long long) 759 sd->sd_offset + sd->sd_size + 4096); 760 break; 761 } 762 } 763 if (sd != NULL) 764 continue; 765 if (tmpdisk->d_state == G_RAID_DISK_S_OFFLINE) { 766 olddisk = tmpdisk; 767 break; 768 } else if (olddisk == NULL) 769 olddisk = tmpdisk; 770 } 771 if (olddisk == NULL) { 772 nofit: 773 if (pd->pd_disk_meta.flags & INTEL_F_SPARE) { 774 g_raid_change_disk_state(disk, 775 G_RAID_DISK_S_SPARE); 776 return (1); 777 } else { 778 g_raid_change_disk_state(disk, 779 G_RAID_DISK_S_STALE); 780 return (0); 781 } 782 } 783 oldpd = (struct g_raid_md_intel_perdisk *)olddisk->d_md_data; 784 disk_pos = oldpd->pd_disk_pos; 785 resurrection = 1; 786 } 787 788 if (olddisk == NULL) { 789 /* Find placeholder by position. */ 790 olddisk = g_raid_md_intel_get_disk(sc, disk_pos); 791 if (olddisk == NULL) 792 panic("No disk at position %d!", disk_pos); 793 if (olddisk->d_state != G_RAID_DISK_S_OFFLINE) { 794 G_RAID_DEBUG1(1, sc, "More then one disk for pos %d", 795 disk_pos); 796 g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE); 797 return (0); 798 } 799 oldpd = (struct g_raid_md_intel_perdisk *)olddisk->d_md_data; 800 } 801 802 /* Replace failed disk or placeholder with new disk. */ 803 TAILQ_FOREACH_SAFE(sd, &olddisk->d_subdisks, sd_next, tmpsd) { 804 TAILQ_REMOVE(&olddisk->d_subdisks, sd, sd_next); 805 TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); 806 sd->sd_disk = disk; 807 } 808 oldpd->pd_disk_pos = -2; 809 pd->pd_disk_pos = disk_pos; 810 811 /* If it was placeholder -- destroy it. */ 812 if (olddisk->d_state == G_RAID_DISK_S_OFFLINE) { 813 g_raid_destroy_disk(olddisk); 814 } else { 815 /* Otherwise, make it STALE_FAILED. */ 816 g_raid_change_disk_state(olddisk, G_RAID_DISK_S_STALE_FAILED); 817 /* Update global metadata just in case. */ 818 memcpy(&meta->disk[disk_pos], &pd->pd_disk_meta, 819 sizeof(struct intel_raid_disk)); 820 } 821 822 /* Welcome the new disk. */ 823 if (resurrection) 824 g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE); 825 else if (meta->disk[disk_pos].flags & INTEL_F_FAILED) 826 g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED); 827 else if (meta->disk[disk_pos].flags & INTEL_F_SPARE) 828 g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE); 829 else 830 g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE); 831 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 832 mvol = intel_get_volume(meta, 833 (uintptr_t)(sd->sd_volume->v_md_data)); 834 mmap0 = intel_get_map(mvol, 0); 835 if (mvol->migr_state) 836 mmap1 = intel_get_map(mvol, 1); 837 else 838 mmap1 = mmap0; 839 840 if (resurrection) { 841 /* Stale disk, almost same as new. */ 842 g_raid_change_subdisk_state(sd, 843 G_RAID_SUBDISK_S_NEW); 844 } else if (meta->disk[disk_pos].flags & INTEL_F_FAILED) { 845 /* Failed disk, almost useless. */ 846 g_raid_change_subdisk_state(sd, 847 G_RAID_SUBDISK_S_FAILED); 848 } else if (mvol->migr_state == 0) { 849 if (mmap0->status == INTEL_S_UNINITIALIZED) { 850 /* Freshly created uninitialized volume. */ 851 g_raid_change_subdisk_state(sd, 852 G_RAID_SUBDISK_S_UNINITIALIZED); 853 } else if (mmap0->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) { 854 /* Freshly inserted disk. */ 855 g_raid_change_subdisk_state(sd, 856 G_RAID_SUBDISK_S_NEW); 857 } else if (mvol->dirty) { 858 /* Dirty volume (unclean shutdown). */ 859 g_raid_change_subdisk_state(sd, 860 G_RAID_SUBDISK_S_STALE); 861 } else { 862 /* Up to date disk. */ 863 g_raid_change_subdisk_state(sd, 864 G_RAID_SUBDISK_S_ACTIVE); 865 } 866 } else if (mvol->migr_type == INTEL_MT_INIT || 867 mvol->migr_type == INTEL_MT_REBUILD) { 868 if (mmap0->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) { 869 /* Freshly inserted disk. */ 870 g_raid_change_subdisk_state(sd, 871 G_RAID_SUBDISK_S_NEW); 872 } else if (mmap1->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) { 873 /* Rebuilding disk. */ 874 g_raid_change_subdisk_state(sd, 875 G_RAID_SUBDISK_S_REBUILD); 876 if (mvol->dirty) { 877 sd->sd_rebuild_pos = 0; 878 } else { 879 sd->sd_rebuild_pos = 880 intel_get_vol_curr_migr_unit(mvol) * 881 sd->sd_volume->v_strip_size * 882 mmap0->total_domains; 883 } 884 } else if (mvol->dirty) { 885 /* Dirty volume (unclean shutdown). */ 886 g_raid_change_subdisk_state(sd, 887 G_RAID_SUBDISK_S_STALE); 888 } else { 889 /* Up to date disk. */ 890 g_raid_change_subdisk_state(sd, 891 G_RAID_SUBDISK_S_ACTIVE); 892 } 893 } else if (mvol->migr_type == INTEL_MT_VERIFY || 894 mvol->migr_type == INTEL_MT_REPAIR) { 895 if (mmap0->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) { 896 /* Freshly inserted disk. */ 897 g_raid_change_subdisk_state(sd, 898 G_RAID_SUBDISK_S_NEW); 899 } else if (mmap1->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) { 900 /* Resyncing disk. */ 901 g_raid_change_subdisk_state(sd, 902 G_RAID_SUBDISK_S_RESYNC); 903 if (mvol->dirty) { 904 sd->sd_rebuild_pos = 0; 905 } else { 906 sd->sd_rebuild_pos = 907 intel_get_vol_curr_migr_unit(mvol) * 908 sd->sd_volume->v_strip_size * 909 mmap0->total_domains; 910 } 911 } else if (mvol->dirty) { 912 /* Dirty volume (unclean shutdown). */ 913 g_raid_change_subdisk_state(sd, 914 G_RAID_SUBDISK_S_STALE); 915 } else { 916 /* Up to date disk. */ 917 g_raid_change_subdisk_state(sd, 918 G_RAID_SUBDISK_S_ACTIVE); 919 } 920 } 921 g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW, 922 G_RAID_EVENT_SUBDISK); 923 } 924 925 /* Update status of our need for spare. */ 926 if (mdi->mdio_started) { 927 mdi->mdio_incomplete = 928 (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) < 929 meta->total_disks); 930 } 931 932 return (resurrection); 933 } 934 935 static void 936 g_disk_md_intel_retaste(void *arg, int pending) 937 { 938 939 G_RAID_DEBUG(1, "Array is not complete, trying to retaste."); 940 g_retaste(&g_raid_class); 941 free(arg, M_MD_INTEL); 942 } 943 944 static void 945 g_raid_md_intel_refill(struct g_raid_softc *sc) 946 { 947 struct g_raid_md_object *md; 948 struct g_raid_md_intel_object *mdi; 949 struct intel_raid_conf *meta; 950 struct g_raid_disk *disk; 951 struct task *task; 952 int update, na; 953 954 md = sc->sc_md; 955 mdi = (struct g_raid_md_intel_object *)md; 956 meta = mdi->mdio_meta; 957 update = 0; 958 do { 959 /* Make sure we miss anything. */ 960 na = g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE); 961 if (na == meta->total_disks) 962 break; 963 964 G_RAID_DEBUG1(1, md->mdo_softc, 965 "Array is not complete (%d of %d), " 966 "trying to refill.", na, meta->total_disks); 967 968 /* Try to get use some of STALE disks. */ 969 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 970 if (disk->d_state == G_RAID_DISK_S_STALE) { 971 update += g_raid_md_intel_start_disk(disk); 972 if (disk->d_state == G_RAID_DISK_S_ACTIVE) 973 break; 974 } 975 } 976 if (disk != NULL) 977 continue; 978 979 /* Try to get use some of SPARE disks. */ 980 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 981 if (disk->d_state == G_RAID_DISK_S_SPARE) { 982 update += g_raid_md_intel_start_disk(disk); 983 if (disk->d_state == G_RAID_DISK_S_ACTIVE) 984 break; 985 } 986 } 987 } while (disk != NULL); 988 989 /* Write new metadata if we changed something. */ 990 if (update) { 991 g_raid_md_write_intel(md, NULL, NULL, NULL); 992 meta = mdi->mdio_meta; 993 } 994 995 /* Update status of our need for spare. */ 996 mdi->mdio_incomplete = (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) < 997 meta->total_disks); 998 999 /* Request retaste hoping to find spare. */ 1000 if (mdi->mdio_incomplete) { 1001 task = malloc(sizeof(struct task), 1002 M_MD_INTEL, M_WAITOK | M_ZERO); 1003 TASK_INIT(task, 0, g_disk_md_intel_retaste, task); 1004 taskqueue_enqueue(taskqueue_swi, task); 1005 } 1006 } 1007 1008 static void 1009 g_raid_md_intel_start(struct g_raid_softc *sc) 1010 { 1011 struct g_raid_md_object *md; 1012 struct g_raid_md_intel_object *mdi; 1013 struct g_raid_md_intel_perdisk *pd; 1014 struct intel_raid_conf *meta; 1015 struct intel_raid_vol *mvol; 1016 struct intel_raid_map *mmap; 1017 struct g_raid_volume *vol; 1018 struct g_raid_subdisk *sd; 1019 struct g_raid_disk *disk; 1020 int i, j, disk_pos; 1021 1022 md = sc->sc_md; 1023 mdi = (struct g_raid_md_intel_object *)md; 1024 meta = mdi->mdio_meta; 1025 1026 /* Create volumes and subdisks. */ 1027 for (i = 0; i < meta->total_volumes; i++) { 1028 mvol = intel_get_volume(meta, i); 1029 mmap = intel_get_map(mvol, 0); 1030 vol = g_raid_create_volume(sc, mvol->name, -1); 1031 vol->v_md_data = (void *)(intptr_t)i; 1032 if (mmap->type == INTEL_T_RAID0) 1033 vol->v_raid_level = G_RAID_VOLUME_RL_RAID0; 1034 else if (mmap->type == INTEL_T_RAID1 && 1035 mmap->total_domains >= 2 && 1036 mmap->total_domains <= mmap->total_disks) { 1037 /* Assume total_domains is correct. */ 1038 if (mmap->total_domains == mmap->total_disks) 1039 vol->v_raid_level = G_RAID_VOLUME_RL_RAID1; 1040 else 1041 vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E; 1042 } else if (mmap->type == INTEL_T_RAID1) { 1043 /* total_domains looks wrong. */ 1044 if (mmap->total_disks <= 2) 1045 vol->v_raid_level = G_RAID_VOLUME_RL_RAID1; 1046 else 1047 vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E; 1048 } else if (mmap->type == INTEL_T_RAID5) 1049 vol->v_raid_level = G_RAID_VOLUME_RL_RAID5; 1050 else 1051 vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; 1052 vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE; 1053 vol->v_strip_size = (u_int)mmap->strip_sectors * 512; //ZZZ 1054 vol->v_disks_count = mmap->total_disks; 1055 vol->v_mediasize = (off_t)mvol->total_sectors * 512; //ZZZ 1056 vol->v_sectorsize = 512; //ZZZ 1057 for (j = 0; j < vol->v_disks_count; j++) { 1058 sd = &vol->v_subdisks[j]; 1059 sd->sd_offset = intel_get_map_offset(mmap) * 512; //ZZZ 1060 sd->sd_size = intel_get_map_disk_sectors(mmap) * 512; //ZZZ 1061 } 1062 g_raid_start_volume(vol); 1063 } 1064 1065 /* Create disk placeholders to store data for later writing. */ 1066 for (disk_pos = 0; disk_pos < meta->total_disks; disk_pos++) { 1067 pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO); 1068 pd->pd_disk_pos = disk_pos; 1069 pd->pd_disk_meta = meta->disk[disk_pos]; 1070 disk = g_raid_create_disk(sc); 1071 disk->d_md_data = (void *)pd; 1072 disk->d_state = G_RAID_DISK_S_OFFLINE; 1073 for (i = 0; i < meta->total_volumes; i++) { 1074 mvol = intel_get_volume(meta, i); 1075 mmap = intel_get_map(mvol, 0); 1076 for (j = 0; j < mmap->total_disks; j++) { 1077 if ((mmap->disk_idx[j] & INTEL_DI_IDX) == disk_pos) 1078 break; 1079 } 1080 if (j == mmap->total_disks) 1081 continue; 1082 vol = g_raid_md_intel_get_volume(sc, i); 1083 sd = &vol->v_subdisks[j]; 1084 sd->sd_disk = disk; 1085 TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); 1086 } 1087 } 1088 1089 /* Make all disks found till the moment take their places. */ 1090 do { 1091 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 1092 if (disk->d_state == G_RAID_DISK_S_NONE) { 1093 g_raid_md_intel_start_disk(disk); 1094 break; 1095 } 1096 } 1097 } while (disk != NULL); 1098 1099 mdi->mdio_started = 1; 1100 G_RAID_DEBUG1(0, sc, "Array started."); 1101 g_raid_md_write_intel(md, NULL, NULL, NULL); 1102 1103 /* Pickup any STALE/SPARE disks to refill array if needed. */ 1104 g_raid_md_intel_refill(sc); 1105 1106 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1107 g_raid_event_send(vol, G_RAID_VOLUME_E_START, 1108 G_RAID_EVENT_VOLUME); 1109 } 1110 1111 callout_stop(&mdi->mdio_start_co); 1112 G_RAID_DEBUG1(1, sc, "root_mount_rel %p", mdi->mdio_rootmount); 1113 root_mount_rel(mdi->mdio_rootmount); 1114 mdi->mdio_rootmount = NULL; 1115 } 1116 1117 static void 1118 g_raid_md_intel_new_disk(struct g_raid_disk *disk) 1119 { 1120 struct g_raid_softc *sc; 1121 struct g_raid_md_object *md; 1122 struct g_raid_md_intel_object *mdi; 1123 struct intel_raid_conf *pdmeta; 1124 struct g_raid_md_intel_perdisk *pd; 1125 1126 sc = disk->d_softc; 1127 md = sc->sc_md; 1128 mdi = (struct g_raid_md_intel_object *)md; 1129 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 1130 pdmeta = pd->pd_meta; 1131 1132 if (mdi->mdio_started) { 1133 if (g_raid_md_intel_start_disk(disk)) 1134 g_raid_md_write_intel(md, NULL, NULL, NULL); 1135 } else { 1136 /* If we haven't started yet - check metadata freshness. */ 1137 if (mdi->mdio_meta == NULL || 1138 ((int32_t)(pdmeta->generation - mdi->mdio_generation)) > 0) { 1139 G_RAID_DEBUG1(1, sc, "Newer disk"); 1140 if (mdi->mdio_meta != NULL) 1141 free(mdi->mdio_meta, M_MD_INTEL); 1142 mdi->mdio_meta = intel_meta_copy(pdmeta); 1143 mdi->mdio_generation = mdi->mdio_meta->generation; 1144 mdi->mdio_disks_present = 1; 1145 } else if (pdmeta->generation == mdi->mdio_generation) { 1146 mdi->mdio_disks_present++; 1147 G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)", 1148 mdi->mdio_disks_present, 1149 mdi->mdio_meta->total_disks); 1150 } else { 1151 G_RAID_DEBUG1(1, sc, "Older disk"); 1152 } 1153 /* If we collected all needed disks - start array. */ 1154 if (mdi->mdio_disks_present == mdi->mdio_meta->total_disks) 1155 g_raid_md_intel_start(sc); 1156 } 1157 } 1158 1159 static void 1160 g_raid_intel_go(void *arg) 1161 { 1162 struct g_raid_softc *sc; 1163 struct g_raid_md_object *md; 1164 struct g_raid_md_intel_object *mdi; 1165 1166 sc = arg; 1167 md = sc->sc_md; 1168 mdi = (struct g_raid_md_intel_object *)md; 1169 if (!mdi->mdio_started) { 1170 G_RAID_DEBUG1(0, sc, "Force array start due to timeout."); 1171 g_raid_event_send(sc, G_RAID_NODE_E_START, 0); 1172 } 1173 } 1174 1175 static int 1176 g_raid_md_create_intel(struct g_raid_md_object *md, struct g_class *mp, 1177 struct g_geom **gp) 1178 { 1179 struct g_raid_softc *sc; 1180 struct g_raid_md_intel_object *mdi; 1181 char name[16]; 1182 1183 mdi = (struct g_raid_md_intel_object *)md; 1184 mdi->mdio_config_id = arc4random(); 1185 mdi->mdio_generation = 0; 1186 snprintf(name, sizeof(name), "Intel-%08x", mdi->mdio_config_id); 1187 sc = g_raid_create_node(mp, name, md); 1188 if (sc == NULL) 1189 return (G_RAID_MD_TASTE_FAIL); 1190 md->mdo_softc = sc; 1191 *gp = sc->sc_geom; 1192 return (G_RAID_MD_TASTE_NEW); 1193 } 1194 1195 /* 1196 * Return the last N characters of the serial label. The Linux and 1197 * ataraid(7) code always uses the last 16 characters of the label to 1198 * store into the Intel meta format. Generalize this to N characters 1199 * since that's easy. Labels can be up to 20 characters for SATA drives 1200 * and up 251 characters for SAS drives. Since intel controllers don't 1201 * support SAS drives, just stick with the SATA limits for stack friendliness. 1202 */ 1203 static int 1204 g_raid_md_get_label(struct g_consumer *cp, char *serial, int serlen) 1205 { 1206 char serial_buffer[24]; 1207 int len, error; 1208 1209 len = sizeof(serial_buffer); 1210 error = g_io_getattr("GEOM::ident", cp, &len, serial_buffer); 1211 if (error != 0) 1212 return (error); 1213 len = strlen(serial_buffer); 1214 if (len > serlen) 1215 len -= serlen; 1216 else 1217 len = 0; 1218 strncpy(serial, serial_buffer + len, serlen); 1219 return (0); 1220 } 1221 1222 static int 1223 g_raid_md_taste_intel(struct g_raid_md_object *md, struct g_class *mp, 1224 struct g_consumer *cp, struct g_geom **gp) 1225 { 1226 struct g_consumer *rcp; 1227 struct g_provider *pp; 1228 struct g_raid_md_intel_object *mdi, *mdi1; 1229 struct g_raid_softc *sc; 1230 struct g_raid_disk *disk; 1231 struct intel_raid_conf *meta; 1232 struct g_raid_md_intel_perdisk *pd; 1233 struct g_geom *geom; 1234 int error, disk_pos, result, spare, len; 1235 char serial[INTEL_SERIAL_LEN]; 1236 char name[16]; 1237 uint16_t vendor; 1238 1239 G_RAID_DEBUG(1, "Tasting Intel on %s", cp->provider->name); 1240 mdi = (struct g_raid_md_intel_object *)md; 1241 pp = cp->provider; 1242 1243 /* Read metadata from device. */ 1244 meta = NULL; 1245 vendor = 0xffff; 1246 disk_pos = 0; 1247 if (g_access(cp, 1, 0, 0) != 0) 1248 return (G_RAID_MD_TASTE_FAIL); 1249 g_topology_unlock(); 1250 error = g_raid_md_get_label(cp, serial, sizeof(serial)); 1251 if (error != 0) { 1252 G_RAID_DEBUG(1, "Cannot get serial number from %s (error=%d).", 1253 pp->name, error); 1254 goto fail2; 1255 } 1256 len = 2; 1257 if (pp->geom->rank == 1) 1258 g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor); 1259 meta = intel_meta_read(cp); 1260 g_topology_lock(); 1261 g_access(cp, -1, 0, 0); 1262 if (meta == NULL) { 1263 if (g_raid_aggressive_spare) { 1264 if (vendor != 0x8086) { 1265 G_RAID_DEBUG(1, 1266 "Intel vendor mismatch 0x%04x != 0x8086", 1267 vendor); 1268 } else { 1269 G_RAID_DEBUG(1, 1270 "No Intel metadata, forcing spare."); 1271 spare = 2; 1272 goto search; 1273 } 1274 } 1275 return (G_RAID_MD_TASTE_FAIL); 1276 } 1277 1278 /* Check this disk position in obtained metadata. */ 1279 disk_pos = intel_meta_find_disk(meta, serial); 1280 if (disk_pos < 0) { 1281 G_RAID_DEBUG(1, "Intel serial '%s' not found", serial); 1282 goto fail1; 1283 } 1284 if (intel_get_disk_sectors(&meta->disk[disk_pos]) != 1285 (pp->mediasize / pp->sectorsize)) { 1286 G_RAID_DEBUG(1, "Intel size mismatch %ju != %ju", 1287 intel_get_disk_sectors(&meta->disk[disk_pos]), 1288 (off_t)(pp->mediasize / pp->sectorsize)); 1289 goto fail1; 1290 } 1291 1292 /* Metadata valid. Print it. */ 1293 g_raid_md_intel_print(meta); 1294 G_RAID_DEBUG(1, "Intel disk position %d", disk_pos); 1295 spare = meta->disk[disk_pos].flags & INTEL_F_SPARE; 1296 1297 search: 1298 /* Search for matching node. */ 1299 sc = NULL; 1300 mdi1 = NULL; 1301 LIST_FOREACH(geom, &mp->geom, geom) { 1302 sc = geom->softc; 1303 if (sc == NULL) 1304 continue; 1305 if (sc->sc_stopping != 0) 1306 continue; 1307 if (sc->sc_md->mdo_class != md->mdo_class) 1308 continue; 1309 mdi1 = (struct g_raid_md_intel_object *)sc->sc_md; 1310 if (spare) { 1311 if (mdi1->mdio_incomplete) 1312 break; 1313 } else { 1314 if (mdi1->mdio_config_id == meta->config_id) 1315 break; 1316 } 1317 } 1318 1319 /* Found matching node. */ 1320 if (geom != NULL) { 1321 G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name); 1322 result = G_RAID_MD_TASTE_EXISTING; 1323 1324 } else if (spare) { /* Not found needy node -- left for later. */ 1325 G_RAID_DEBUG(1, "Spare is not needed at this time"); 1326 goto fail1; 1327 1328 } else { /* Not found matching node -- create one. */ 1329 result = G_RAID_MD_TASTE_NEW; 1330 mdi->mdio_config_id = meta->config_id; 1331 snprintf(name, sizeof(name), "Intel-%08x", meta->config_id); 1332 sc = g_raid_create_node(mp, name, md); 1333 md->mdo_softc = sc; 1334 geom = sc->sc_geom; 1335 callout_init(&mdi->mdio_start_co, 1); 1336 callout_reset(&mdi->mdio_start_co, g_raid_start_timeout * hz, 1337 g_raid_intel_go, sc); 1338 mdi->mdio_rootmount = root_mount_hold("GRAID-Intel"); 1339 G_RAID_DEBUG1(1, sc, "root_mount_hold %p", mdi->mdio_rootmount); 1340 } 1341 1342 rcp = g_new_consumer(geom); 1343 g_attach(rcp, pp); 1344 if (g_access(rcp, 1, 1, 1) != 0) 1345 ; //goto fail1; 1346 1347 g_topology_unlock(); 1348 sx_xlock(&sc->sc_lock); 1349 1350 pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO); 1351 pd->pd_meta = meta; 1352 pd->pd_disk_pos = -1; 1353 if (spare == 2) { 1354 memcpy(&pd->pd_disk_meta.serial[0], serial, INTEL_SERIAL_LEN); 1355 intel_set_disk_sectors(&pd->pd_disk_meta, 1356 pp->mediasize / pp->sectorsize); 1357 pd->pd_disk_meta.id = 0; 1358 pd->pd_disk_meta.flags = INTEL_F_SPARE; 1359 } else { 1360 pd->pd_disk_meta = meta->disk[disk_pos]; 1361 } 1362 disk = g_raid_create_disk(sc); 1363 disk->d_md_data = (void *)pd; 1364 disk->d_consumer = rcp; 1365 rcp->private = disk; 1366 1367 /* Read kernel dumping information. */ 1368 disk->d_kd.offset = 0; 1369 disk->d_kd.length = OFF_MAX; 1370 len = sizeof(disk->d_kd); 1371 error = g_io_getattr("GEOM::kerneldump", rcp, &len, &disk->d_kd); 1372 if (disk->d_kd.di.dumper == NULL) 1373 G_RAID_DEBUG1(2, sc, "Dumping not supported by %s: %d.", 1374 rcp->provider->name, error); 1375 1376 g_raid_md_intel_new_disk(disk); 1377 1378 sx_xunlock(&sc->sc_lock); 1379 g_topology_lock(); 1380 *gp = geom; 1381 return (result); 1382 fail2: 1383 g_topology_lock(); 1384 g_access(cp, -1, 0, 0); 1385 fail1: 1386 free(meta, M_MD_INTEL); 1387 return (G_RAID_MD_TASTE_FAIL); 1388 } 1389 1390 static int 1391 g_raid_md_event_intel(struct g_raid_md_object *md, 1392 struct g_raid_disk *disk, u_int event) 1393 { 1394 struct g_raid_softc *sc; 1395 struct g_raid_subdisk *sd; 1396 struct g_raid_md_intel_object *mdi; 1397 struct g_raid_md_intel_perdisk *pd; 1398 1399 sc = md->mdo_softc; 1400 mdi = (struct g_raid_md_intel_object *)md; 1401 if (disk == NULL) { 1402 switch (event) { 1403 case G_RAID_NODE_E_START: 1404 if (!mdi->mdio_started) 1405 g_raid_md_intel_start(sc); 1406 return (0); 1407 } 1408 return (-1); 1409 } 1410 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 1411 switch (event) { 1412 case G_RAID_DISK_E_DISCONNECTED: 1413 /* If disk was assigned, just update statuses. */ 1414 if (pd->pd_disk_pos >= 0) { 1415 g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); 1416 if (disk->d_consumer) { 1417 g_raid_kill_consumer(sc, disk->d_consumer); 1418 disk->d_consumer = NULL; 1419 } 1420 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 1421 g_raid_change_subdisk_state(sd, 1422 G_RAID_SUBDISK_S_NONE); 1423 g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, 1424 G_RAID_EVENT_SUBDISK); 1425 } 1426 } else { 1427 /* Otherwise -- delete. */ 1428 g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE); 1429 g_raid_destroy_disk(disk); 1430 } 1431 1432 /* Write updated metadata to all disks. */ 1433 g_raid_md_write_intel(md, NULL, NULL, NULL); 1434 1435 /* Check if anything left except placeholders. */ 1436 if (g_raid_ndisks(sc, -1) == 1437 g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) 1438 g_raid_destroy_node(sc, 0); 1439 else 1440 g_raid_md_intel_refill(sc); 1441 return (0); 1442 } 1443 return (-2); 1444 } 1445 1446 static int 1447 g_raid_md_ctl_intel(struct g_raid_md_object *md, 1448 struct gctl_req *req) 1449 { 1450 struct g_raid_softc *sc; 1451 struct g_raid_volume *vol, *vol1; 1452 struct g_raid_subdisk *sd; 1453 struct g_raid_disk *disk; 1454 struct g_raid_md_intel_object *mdi; 1455 struct g_raid_md_intel_perdisk *pd; 1456 struct g_consumer *cp; 1457 struct g_provider *pp; 1458 char arg[16], serial[INTEL_SERIAL_LEN]; 1459 const char *verb, *volname, *levelname, *diskname; 1460 char *tmp; 1461 int *nargs, *force; 1462 off_t off, size, sectorsize, strip, disk_sectors; 1463 intmax_t *sizearg, *striparg; 1464 int numdisks, i, len, level, qual, update; 1465 int error; 1466 1467 sc = md->mdo_softc; 1468 mdi = (struct g_raid_md_intel_object *)md; 1469 verb = gctl_get_param(req, "verb", NULL); 1470 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); 1471 error = 0; 1472 if (strcmp(verb, "label") == 0) { 1473 1474 if (*nargs < 4) { 1475 gctl_error(req, "Invalid number of arguments."); 1476 return (-1); 1477 } 1478 volname = gctl_get_asciiparam(req, "arg1"); 1479 if (volname == NULL) { 1480 gctl_error(req, "No volume name."); 1481 return (-2); 1482 } 1483 levelname = gctl_get_asciiparam(req, "arg2"); 1484 if (levelname == NULL) { 1485 gctl_error(req, "No RAID level."); 1486 return (-3); 1487 } 1488 if (g_raid_volume_str2level(levelname, &level, &qual)) { 1489 gctl_error(req, "Unknown RAID level '%s'.", levelname); 1490 return (-4); 1491 } 1492 numdisks = *nargs - 3; 1493 force = gctl_get_paraml(req, "force", sizeof(*force)); 1494 if (!g_raid_md_intel_supported(level, qual, numdisks, 1495 force ? *force : 0)) { 1496 gctl_error(req, "Unsupported RAID level " 1497 "(0x%02x/0x%02x), or number of disks (%d).", 1498 level, qual, numdisks); 1499 return (-5); 1500 } 1501 1502 /* Search for disks, connect them and probe. */ 1503 size = 0x7fffffffffffffffllu; 1504 sectorsize = 0; 1505 for (i = 0; i < numdisks; i++) { 1506 snprintf(arg, sizeof(arg), "arg%d", i + 3); 1507 diskname = gctl_get_asciiparam(req, arg); 1508 if (diskname == NULL) { 1509 gctl_error(req, "No disk name (%s).", arg); 1510 error = -6; 1511 break; 1512 } 1513 if (strcmp(diskname, "NONE") == 0) { 1514 cp = NULL; 1515 pp = NULL; 1516 } else { 1517 g_topology_lock(); 1518 cp = g_raid_open_consumer(sc, diskname); 1519 if (cp == NULL) { 1520 gctl_error(req, "Can't open disk '%s'.", 1521 diskname); 1522 g_topology_unlock(); 1523 error = -7; 1524 break; 1525 } 1526 pp = cp->provider; 1527 } 1528 pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO); 1529 pd->pd_disk_pos = i; 1530 disk = g_raid_create_disk(sc); 1531 disk->d_md_data = (void *)pd; 1532 disk->d_consumer = cp; 1533 if (cp == NULL) { 1534 strcpy(&pd->pd_disk_meta.serial[0], "NONE"); 1535 pd->pd_disk_meta.id = 0xffffffff; 1536 pd->pd_disk_meta.flags = INTEL_F_ASSIGNED; 1537 continue; 1538 } 1539 cp->private = disk; 1540 g_topology_unlock(); 1541 1542 error = g_raid_md_get_label(cp, 1543 &pd->pd_disk_meta.serial[0], INTEL_SERIAL_LEN); 1544 if (error != 0) { 1545 gctl_error(req, 1546 "Can't get serial for provider '%s'.", 1547 diskname); 1548 error = -8; 1549 break; 1550 } 1551 1552 /* Read kernel dumping information. */ 1553 disk->d_kd.offset = 0; 1554 disk->d_kd.length = OFF_MAX; 1555 len = sizeof(disk->d_kd); 1556 g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); 1557 if (disk->d_kd.di.dumper == NULL) 1558 G_RAID_DEBUG1(2, sc, 1559 "Dumping not supported by %s.", 1560 cp->provider->name); 1561 1562 intel_set_disk_sectors(&pd->pd_disk_meta, 1563 pp->mediasize / pp->sectorsize); 1564 if (size > pp->mediasize) 1565 size = pp->mediasize; 1566 if (sectorsize < pp->sectorsize) 1567 sectorsize = pp->sectorsize; 1568 pd->pd_disk_meta.id = 0; 1569 pd->pd_disk_meta.flags = INTEL_F_ASSIGNED | INTEL_F_ONLINE; 1570 } 1571 if (error != 0) 1572 return (error); 1573 1574 if (sectorsize <= 0) { 1575 gctl_error(req, "Can't get sector size."); 1576 return (-8); 1577 } 1578 1579 /* Reserve some space for metadata. */ 1580 size -= ((4096 + sectorsize - 1) / sectorsize) * sectorsize; 1581 1582 /* Handle size argument. */ 1583 len = sizeof(*sizearg); 1584 sizearg = gctl_get_param(req, "size", &len); 1585 if (sizearg != NULL && len == sizeof(*sizearg) && 1586 *sizearg > 0) { 1587 if (*sizearg > size) { 1588 gctl_error(req, "Size too big %lld > %lld.", 1589 (long long)*sizearg, (long long)size); 1590 return (-9); 1591 } 1592 size = *sizearg; 1593 } 1594 1595 /* Handle strip argument. */ 1596 strip = 131072; 1597 len = sizeof(*striparg); 1598 striparg = gctl_get_param(req, "strip", &len); 1599 if (striparg != NULL && len == sizeof(*striparg) && 1600 *striparg > 0) { 1601 if (*striparg < sectorsize) { 1602 gctl_error(req, "Strip size too small."); 1603 return (-10); 1604 } 1605 if (*striparg % sectorsize != 0) { 1606 gctl_error(req, "Incorrect strip size."); 1607 return (-11); 1608 } 1609 if (strip > 65535 * sectorsize) { 1610 gctl_error(req, "Strip size too big."); 1611 return (-12); 1612 } 1613 strip = *striparg; 1614 } 1615 1616 /* Round size down to strip or sector. */ 1617 if (level == G_RAID_VOLUME_RL_RAID1) 1618 size -= (size % sectorsize); 1619 else if (level == G_RAID_VOLUME_RL_RAID1E && 1620 (numdisks & 1) != 0) 1621 size -= (size % (2 * strip)); 1622 else 1623 size -= (size % strip); 1624 if (size <= 0) { 1625 gctl_error(req, "Size too small."); 1626 return (-13); 1627 } 1628 1629 /* We have all we need, create things: volume, ... */ 1630 mdi->mdio_started = 1; 1631 vol = g_raid_create_volume(sc, volname, -1); 1632 vol->v_md_data = (void *)(intptr_t)0; 1633 vol->v_raid_level = level; 1634 vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE; 1635 vol->v_strip_size = strip; 1636 vol->v_disks_count = numdisks; 1637 if (level == G_RAID_VOLUME_RL_RAID0) 1638 vol->v_mediasize = size * numdisks; 1639 else if (level == G_RAID_VOLUME_RL_RAID1) 1640 vol->v_mediasize = size; 1641 else if (level == G_RAID_VOLUME_RL_RAID5) 1642 vol->v_mediasize = size * (numdisks - 1); 1643 else { /* RAID1E */ 1644 vol->v_mediasize = ((size * numdisks) / strip / 2) * 1645 strip; 1646 } 1647 vol->v_sectorsize = sectorsize; 1648 g_raid_start_volume(vol); 1649 1650 /* , and subdisks. */ 1651 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 1652 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 1653 sd = &vol->v_subdisks[pd->pd_disk_pos]; 1654 sd->sd_disk = disk; 1655 sd->sd_offset = 0; 1656 sd->sd_size = size; 1657 TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); 1658 if (sd->sd_disk->d_consumer != NULL) { 1659 g_raid_change_disk_state(disk, 1660 G_RAID_DISK_S_ACTIVE); 1661 g_raid_change_subdisk_state(sd, 1662 G_RAID_SUBDISK_S_ACTIVE); 1663 g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW, 1664 G_RAID_EVENT_SUBDISK); 1665 } else { 1666 g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); 1667 } 1668 } 1669 1670 /* Write metadata based on created entities. */ 1671 G_RAID_DEBUG1(0, sc, "Array started."); 1672 g_raid_md_write_intel(md, NULL, NULL, NULL); 1673 1674 /* Pickup any STALE/SPARE disks to refill array if needed. */ 1675 g_raid_md_intel_refill(sc); 1676 1677 g_raid_event_send(vol, G_RAID_VOLUME_E_START, 1678 G_RAID_EVENT_VOLUME); 1679 return (0); 1680 } 1681 if (strcmp(verb, "add") == 0) { 1682 1683 if (*nargs != 3) { 1684 gctl_error(req, "Invalid number of arguments."); 1685 return (-1); 1686 } 1687 volname = gctl_get_asciiparam(req, "arg1"); 1688 if (volname == NULL) { 1689 gctl_error(req, "No volume name."); 1690 return (-2); 1691 } 1692 levelname = gctl_get_asciiparam(req, "arg2"); 1693 if (levelname == NULL) { 1694 gctl_error(req, "No RAID level."); 1695 return (-3); 1696 } 1697 if (g_raid_volume_str2level(levelname, &level, &qual)) { 1698 gctl_error(req, "Unknown RAID level '%s'.", levelname); 1699 return (-4); 1700 } 1701 1702 /* Look for existing volumes. */ 1703 i = 0; 1704 vol1 = NULL; 1705 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1706 vol1 = vol; 1707 i++; 1708 } 1709 if (i > 1) { 1710 gctl_error(req, "Maximum two volumes supported."); 1711 return (-6); 1712 } 1713 if (vol1 == NULL) { 1714 gctl_error(req, "At least one volume must exist."); 1715 return (-7); 1716 } 1717 1718 numdisks = vol1->v_disks_count; 1719 force = gctl_get_paraml(req, "force", sizeof(*force)); 1720 if (!g_raid_md_intel_supported(level, qual, numdisks, 1721 force ? *force : 0)) { 1722 gctl_error(req, "Unsupported RAID level " 1723 "(0x%02x/0x%02x), or number of disks (%d).", 1724 level, qual, numdisks); 1725 return (-5); 1726 } 1727 1728 /* Collect info about present disks. */ 1729 size = 0x7fffffffffffffffllu; 1730 sectorsize = 512; 1731 for (i = 0; i < numdisks; i++) { 1732 disk = vol1->v_subdisks[i].sd_disk; 1733 pd = (struct g_raid_md_intel_perdisk *) 1734 disk->d_md_data; 1735 disk_sectors = 1736 intel_get_disk_sectors(&pd->pd_disk_meta); 1737 1738 if (disk_sectors * 512 < size) 1739 size = disk_sectors * 512; 1740 if (disk->d_consumer != NULL && 1741 disk->d_consumer->provider != NULL && 1742 disk->d_consumer->provider->sectorsize > 1743 sectorsize) { 1744 sectorsize = 1745 disk->d_consumer->provider->sectorsize; 1746 } 1747 } 1748 1749 /* Reserve some space for metadata. */ 1750 size -= ((4096 + sectorsize - 1) / sectorsize) * sectorsize; 1751 1752 /* Decide insert before or after. */ 1753 sd = &vol1->v_subdisks[0]; 1754 if (sd->sd_offset > 1755 size - (sd->sd_offset + sd->sd_size)) { 1756 off = 0; 1757 size = sd->sd_offset; 1758 } else { 1759 off = sd->sd_offset + sd->sd_size; 1760 size = size - (sd->sd_offset + sd->sd_size); 1761 } 1762 1763 /* Handle strip argument. */ 1764 strip = 131072; 1765 len = sizeof(*striparg); 1766 striparg = gctl_get_param(req, "strip", &len); 1767 if (striparg != NULL && len == sizeof(*striparg) && 1768 *striparg > 0) { 1769 if (*striparg < sectorsize) { 1770 gctl_error(req, "Strip size too small."); 1771 return (-10); 1772 } 1773 if (*striparg % sectorsize != 0) { 1774 gctl_error(req, "Incorrect strip size."); 1775 return (-11); 1776 } 1777 if (strip > 65535 * sectorsize) { 1778 gctl_error(req, "Strip size too big."); 1779 return (-12); 1780 } 1781 strip = *striparg; 1782 } 1783 1784 /* Round offset up to strip. */ 1785 if (off % strip != 0) { 1786 size -= strip - off % strip; 1787 off += strip - off % strip; 1788 } 1789 1790 /* Handle size argument. */ 1791 len = sizeof(*sizearg); 1792 sizearg = gctl_get_param(req, "size", &len); 1793 if (sizearg != NULL && len == sizeof(*sizearg) && 1794 *sizearg > 0) { 1795 if (*sizearg > size) { 1796 gctl_error(req, "Size too big %lld > %lld.", 1797 (long long)*sizearg, (long long)size); 1798 return (-9); 1799 } 1800 size = *sizearg; 1801 } 1802 1803 /* Round size down to strip or sector. */ 1804 if (level == G_RAID_VOLUME_RL_RAID1) 1805 size -= (size % sectorsize); 1806 else 1807 size -= (size % strip); 1808 if (size <= 0) { 1809 gctl_error(req, "Size too small."); 1810 return (-13); 1811 } 1812 if (size > 0xffffffffllu * sectorsize) { 1813 gctl_error(req, "Size too big."); 1814 return (-14); 1815 } 1816 1817 /* We have all we need, create things: volume, ... */ 1818 vol = g_raid_create_volume(sc, volname, -1); 1819 vol->v_md_data = (void *)(intptr_t)i; 1820 vol->v_raid_level = level; 1821 vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE; 1822 vol->v_strip_size = strip; 1823 vol->v_disks_count = numdisks; 1824 if (level == G_RAID_VOLUME_RL_RAID0) 1825 vol->v_mediasize = size * numdisks; 1826 else if (level == G_RAID_VOLUME_RL_RAID1) 1827 vol->v_mediasize = size; 1828 else if (level == G_RAID_VOLUME_RL_RAID5) 1829 vol->v_mediasize = size * (numdisks - 1); 1830 else { /* RAID1E */ 1831 vol->v_mediasize = ((size * numdisks) / strip / 2) * 1832 strip; 1833 } 1834 vol->v_sectorsize = sectorsize; 1835 g_raid_start_volume(vol); 1836 1837 /* , and subdisks. */ 1838 for (i = 0; i < numdisks; i++) { 1839 disk = vol1->v_subdisks[i].sd_disk; 1840 sd = &vol->v_subdisks[i]; 1841 sd->sd_disk = disk; 1842 sd->sd_offset = off; 1843 sd->sd_size = size; 1844 TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); 1845 if (disk->d_state == G_RAID_DISK_S_ACTIVE) { 1846 g_raid_change_subdisk_state(sd, 1847 G_RAID_SUBDISK_S_ACTIVE); 1848 g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW, 1849 G_RAID_EVENT_SUBDISK); 1850 } 1851 } 1852 1853 /* Write metadata based on created entities. */ 1854 g_raid_md_write_intel(md, NULL, NULL, NULL); 1855 1856 g_raid_event_send(vol, G_RAID_VOLUME_E_START, 1857 G_RAID_EVENT_VOLUME); 1858 return (0); 1859 } 1860 if (strcmp(verb, "delete") == 0) { 1861 1862 /* Full node destruction. */ 1863 if (*nargs == 1) { 1864 /* Check if some volume is still open. */ 1865 force = gctl_get_paraml(req, "force", sizeof(*force)); 1866 if (force != NULL && *force == 0 && 1867 g_raid_nopens(sc) != 0) { 1868 gctl_error(req, "Some volume is still open."); 1869 return (-4); 1870 } 1871 1872 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 1873 if (disk->d_consumer) 1874 intel_meta_erase(disk->d_consumer); 1875 } 1876 g_raid_destroy_node(sc, 0); 1877 return (0); 1878 } 1879 1880 /* Destroy specified volume. If it was last - all node. */ 1881 if (*nargs != 2) { 1882 gctl_error(req, "Invalid number of arguments."); 1883 return (-1); 1884 } 1885 volname = gctl_get_asciiparam(req, "arg1"); 1886 if (volname == NULL) { 1887 gctl_error(req, "No volume name."); 1888 return (-2); 1889 } 1890 1891 /* Search for volume. */ 1892 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1893 if (strcmp(vol->v_name, volname) == 0) 1894 break; 1895 } 1896 if (vol == NULL) { 1897 i = strtol(volname, &tmp, 10); 1898 if (verb != volname && tmp[0] == 0) { 1899 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1900 if (vol->v_global_id == i) 1901 break; 1902 } 1903 } 1904 } 1905 if (vol == NULL) { 1906 gctl_error(req, "Volume '%s' not found.", volname); 1907 return (-3); 1908 } 1909 1910 /* Check if volume is still open. */ 1911 force = gctl_get_paraml(req, "force", sizeof(*force)); 1912 if (force != NULL && *force == 0 && 1913 vol->v_provider_open != 0) { 1914 gctl_error(req, "Volume is still open."); 1915 return (-4); 1916 } 1917 1918 /* Destroy volume and potentially node. */ 1919 i = 0; 1920 TAILQ_FOREACH(vol1, &sc->sc_volumes, v_next) 1921 i++; 1922 if (i >= 2) { 1923 g_raid_destroy_volume(vol); 1924 g_raid_md_write_intel(md, NULL, NULL, NULL); 1925 } else { 1926 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 1927 if (disk->d_consumer) 1928 intel_meta_erase(disk->d_consumer); 1929 } 1930 g_raid_destroy_node(sc, 0); 1931 } 1932 return (0); 1933 } 1934 if (strcmp(verb, "remove") == 0 || 1935 strcmp(verb, "fail") == 0) { 1936 if (*nargs < 2) { 1937 gctl_error(req, "Invalid number of arguments."); 1938 return (-1); 1939 } 1940 for (i = 1; i < *nargs; i++) { 1941 snprintf(arg, sizeof(arg), "arg%d", i); 1942 diskname = gctl_get_asciiparam(req, arg); 1943 if (diskname == NULL) { 1944 gctl_error(req, "No disk name (%s).", arg); 1945 error = -2; 1946 break; 1947 } 1948 if (strncmp(diskname, "/dev/", 5) == 0) 1949 diskname += 5; 1950 1951 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 1952 if (disk->d_consumer != NULL && 1953 disk->d_consumer->provider != NULL && 1954 strcmp(disk->d_consumer->provider->name, 1955 diskname) == 0) 1956 break; 1957 } 1958 if (disk == NULL) { 1959 gctl_error(req, "Disk '%s' not found.", 1960 diskname); 1961 error = -3; 1962 break; 1963 } 1964 1965 if (strcmp(verb, "fail") == 0) { 1966 g_raid_md_fail_disk_intel(md, NULL, disk); 1967 continue; 1968 } 1969 1970 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 1971 1972 /* Erase metadata on deleting disk. */ 1973 intel_meta_erase(disk->d_consumer); 1974 1975 /* If disk was assigned, just update statuses. */ 1976 if (pd->pd_disk_pos >= 0) { 1977 g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); 1978 g_raid_kill_consumer(sc, disk->d_consumer); 1979 disk->d_consumer = NULL; 1980 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 1981 g_raid_change_subdisk_state(sd, 1982 G_RAID_SUBDISK_S_NONE); 1983 g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, 1984 G_RAID_EVENT_SUBDISK); 1985 } 1986 } else { 1987 /* Otherwise -- delete. */ 1988 g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE); 1989 g_raid_destroy_disk(disk); 1990 } 1991 } 1992 1993 /* Write updated metadata to remaining disks. */ 1994 g_raid_md_write_intel(md, NULL, NULL, NULL); 1995 1996 /* Check if anything left except placeholders. */ 1997 if (g_raid_ndisks(sc, -1) == 1998 g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) 1999 g_raid_destroy_node(sc, 0); 2000 else 2001 g_raid_md_intel_refill(sc); 2002 return (error); 2003 } 2004 if (strcmp(verb, "insert") == 0) { 2005 if (*nargs < 2) { 2006 gctl_error(req, "Invalid number of arguments."); 2007 return (-1); 2008 } 2009 update = 0; 2010 for (i = 1; i < *nargs; i++) { 2011 /* Get disk name. */ 2012 snprintf(arg, sizeof(arg), "arg%d", i); 2013 diskname = gctl_get_asciiparam(req, arg); 2014 if (diskname == NULL) { 2015 gctl_error(req, "No disk name (%s).", arg); 2016 error = -3; 2017 break; 2018 } 2019 2020 /* Try to find provider with specified name. */ 2021 g_topology_lock(); 2022 cp = g_raid_open_consumer(sc, diskname); 2023 if (cp == NULL) { 2024 gctl_error(req, "Can't open disk '%s'.", 2025 diskname); 2026 g_topology_unlock(); 2027 error = -4; 2028 break; 2029 } 2030 pp = cp->provider; 2031 g_topology_unlock(); 2032 2033 /* Read disk serial. */ 2034 error = g_raid_md_get_label(cp, 2035 &serial[0], INTEL_SERIAL_LEN); 2036 if (error != 0) { 2037 gctl_error(req, 2038 "Can't get serial for provider '%s'.", 2039 diskname); 2040 g_raid_kill_consumer(sc, cp); 2041 error = -7; 2042 break; 2043 } 2044 2045 pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO); 2046 pd->pd_disk_pos = -1; 2047 2048 disk = g_raid_create_disk(sc); 2049 disk->d_consumer = cp; 2050 disk->d_md_data = (void *)pd; 2051 cp->private = disk; 2052 2053 /* Read kernel dumping information. */ 2054 disk->d_kd.offset = 0; 2055 disk->d_kd.length = OFF_MAX; 2056 len = sizeof(disk->d_kd); 2057 g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); 2058 if (disk->d_kd.di.dumper == NULL) 2059 G_RAID_DEBUG1(2, sc, 2060 "Dumping not supported by %s.", 2061 cp->provider->name); 2062 2063 memcpy(&pd->pd_disk_meta.serial[0], &serial[0], 2064 INTEL_SERIAL_LEN); 2065 intel_set_disk_sectors(&pd->pd_disk_meta, 2066 pp->mediasize / pp->sectorsize); 2067 pd->pd_disk_meta.id = 0; 2068 pd->pd_disk_meta.flags = INTEL_F_SPARE; 2069 2070 /* Welcome the "new" disk. */ 2071 update += g_raid_md_intel_start_disk(disk); 2072 if (disk->d_state == G_RAID_DISK_S_SPARE) { 2073 intel_meta_write_spare(cp, &pd->pd_disk_meta); 2074 g_raid_destroy_disk(disk); 2075 } else if (disk->d_state != G_RAID_DISK_S_ACTIVE) { 2076 gctl_error(req, "Disk '%s' doesn't fit.", 2077 diskname); 2078 g_raid_destroy_disk(disk); 2079 error = -8; 2080 break; 2081 } 2082 } 2083 2084 /* Write new metadata if we changed something. */ 2085 if (update) 2086 g_raid_md_write_intel(md, NULL, NULL, NULL); 2087 return (error); 2088 } 2089 return (-100); 2090 } 2091 2092 static int 2093 g_raid_md_write_intel(struct g_raid_md_object *md, struct g_raid_volume *tvol, 2094 struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk) 2095 { 2096 struct g_raid_softc *sc; 2097 struct g_raid_volume *vol; 2098 struct g_raid_subdisk *sd; 2099 struct g_raid_disk *disk; 2100 struct g_raid_md_intel_object *mdi; 2101 struct g_raid_md_intel_perdisk *pd; 2102 struct intel_raid_conf *meta; 2103 struct intel_raid_vol *mvol; 2104 struct intel_raid_map *mmap0, *mmap1; 2105 off_t sectorsize = 512, pos; 2106 const char *version, *cv; 2107 int vi, sdi, numdisks, len, state, stale; 2108 2109 sc = md->mdo_softc; 2110 mdi = (struct g_raid_md_intel_object *)md; 2111 2112 if (sc->sc_stopping == G_RAID_DESTROY_HARD) 2113 return (0); 2114 2115 /* Bump generation. Newly written metadata may differ from previous. */ 2116 mdi->mdio_generation++; 2117 2118 /* Count number of disks. */ 2119 numdisks = 0; 2120 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 2121 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 2122 if (pd->pd_disk_pos < 0) 2123 continue; 2124 numdisks++; 2125 if (disk->d_state == G_RAID_DISK_S_ACTIVE) { 2126 pd->pd_disk_meta.flags = 2127 INTEL_F_ONLINE | INTEL_F_ASSIGNED; 2128 } else if (disk->d_state == G_RAID_DISK_S_FAILED) { 2129 pd->pd_disk_meta.flags = INTEL_F_FAILED | INTEL_F_ASSIGNED; 2130 } else { 2131 pd->pd_disk_meta.flags = INTEL_F_ASSIGNED; 2132 if (pd->pd_disk_meta.id != 0xffffffff) { 2133 pd->pd_disk_meta.id = 0xffffffff; 2134 len = strlen(pd->pd_disk_meta.serial); 2135 len = min(len, INTEL_SERIAL_LEN - 3); 2136 strcpy(pd->pd_disk_meta.serial + len, ":0"); 2137 } 2138 } 2139 } 2140 2141 /* Fill anchor and disks. */ 2142 meta = malloc(INTEL_MAX_MD_SIZE(numdisks), 2143 M_MD_INTEL, M_WAITOK | M_ZERO); 2144 memcpy(&meta->intel_id[0], INTEL_MAGIC, sizeof(INTEL_MAGIC) - 1); 2145 meta->config_size = INTEL_MAX_MD_SIZE(numdisks); 2146 meta->config_id = mdi->mdio_config_id; 2147 meta->generation = mdi->mdio_generation; 2148 meta->attributes = INTEL_ATTR_CHECKSUM; 2149 meta->total_disks = numdisks; 2150 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 2151 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 2152 if (pd->pd_disk_pos < 0) 2153 continue; 2154 meta->disk[pd->pd_disk_pos] = pd->pd_disk_meta; 2155 } 2156 2157 /* Fill volumes and maps. */ 2158 vi = 0; 2159 version = INTEL_VERSION_1000; 2160 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 2161 if (vol->v_stopping) 2162 continue; 2163 mvol = intel_get_volume(meta, vi); 2164 2165 /* New metadata may have different volumes order. */ 2166 vol->v_md_data = (void *)(intptr_t)vi; 2167 2168 for (sdi = 0; sdi < vol->v_disks_count; sdi++) { 2169 sd = &vol->v_subdisks[sdi]; 2170 if (sd->sd_disk != NULL) 2171 break; 2172 } 2173 if (sdi >= vol->v_disks_count) 2174 panic("No any filled subdisk in volume"); 2175 if (vol->v_mediasize >= 0x20000000000llu) 2176 meta->attributes |= INTEL_ATTR_2TB; 2177 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0) 2178 meta->attributes |= INTEL_ATTR_RAID0; 2179 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1) 2180 meta->attributes |= INTEL_ATTR_RAID1; 2181 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5) 2182 meta->attributes |= INTEL_ATTR_RAID5; 2183 else 2184 meta->attributes |= INTEL_ATTR_RAID10; 2185 2186 if (meta->attributes & INTEL_ATTR_2TB) 2187 cv = INTEL_VERSION_1300; 2188 // else if (dev->status == DEV_CLONE_N_GO) 2189 // cv = INTEL_VERSION_1206; 2190 else if (vol->v_disks_count > 4) 2191 cv = INTEL_VERSION_1204; 2192 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5) 2193 cv = INTEL_VERSION_1202; 2194 else if (vol->v_disks_count > 2) 2195 cv = INTEL_VERSION_1201; 2196 else if (vi > 0) 2197 cv = INTEL_VERSION_1200; 2198 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1) 2199 cv = INTEL_VERSION_1100; 2200 else 2201 cv = INTEL_VERSION_1000; 2202 if (strcmp(cv, version) > 0) 2203 version = cv; 2204 2205 strlcpy(&mvol->name[0], vol->v_name, sizeof(mvol->name)); 2206 mvol->total_sectors = vol->v_mediasize / sectorsize; 2207 2208 /* Check for any recovery in progress. */ 2209 state = G_RAID_SUBDISK_S_ACTIVE; 2210 pos = 0x7fffffffffffffffllu; 2211 stale = 0; 2212 for (sdi = 0; sdi < vol->v_disks_count; sdi++) { 2213 sd = &vol->v_subdisks[sdi]; 2214 if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD) 2215 state = G_RAID_SUBDISK_S_REBUILD; 2216 else if (sd->sd_state == G_RAID_SUBDISK_S_RESYNC && 2217 state != G_RAID_SUBDISK_S_REBUILD) 2218 state = G_RAID_SUBDISK_S_RESYNC; 2219 else if (sd->sd_state == G_RAID_SUBDISK_S_STALE) 2220 stale = 1; 2221 if ((sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2222 sd->sd_state == G_RAID_SUBDISK_S_RESYNC) && 2223 sd->sd_rebuild_pos < pos) 2224 pos = sd->sd_rebuild_pos; 2225 } 2226 if (state == G_RAID_SUBDISK_S_REBUILD) { 2227 mvol->migr_state = 1; 2228 mvol->migr_type = INTEL_MT_REBUILD; 2229 } else if (state == G_RAID_SUBDISK_S_RESYNC) { 2230 mvol->migr_state = 1; 2231 /* mvol->migr_type = INTEL_MT_REPAIR; */ 2232 mvol->migr_type = INTEL_MT_VERIFY; 2233 mvol->state |= INTEL_ST_VERIFY_AND_FIX; 2234 } else 2235 mvol->migr_state = 0; 2236 mvol->dirty = (vol->v_dirty || stale); 2237 2238 mmap0 = intel_get_map(mvol, 0); 2239 2240 /* Write map / common part of two maps. */ 2241 intel_set_map_offset(mmap0, sd->sd_offset / sectorsize); 2242 intel_set_map_disk_sectors(mmap0, sd->sd_size / sectorsize); 2243 mmap0->strip_sectors = vol->v_strip_size / sectorsize; 2244 if (vol->v_state == G_RAID_VOLUME_S_BROKEN) 2245 mmap0->status = INTEL_S_FAILURE; 2246 else if (vol->v_state == G_RAID_VOLUME_S_DEGRADED) 2247 mmap0->status = INTEL_S_DEGRADED; 2248 else 2249 mmap0->status = INTEL_S_READY; 2250 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0) 2251 mmap0->type = INTEL_T_RAID0; 2252 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 || 2253 vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) 2254 mmap0->type = INTEL_T_RAID1; 2255 else 2256 mmap0->type = INTEL_T_RAID5; 2257 mmap0->total_disks = vol->v_disks_count; 2258 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1) 2259 mmap0->total_domains = vol->v_disks_count; 2260 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) 2261 mmap0->total_domains = 2; 2262 else 2263 mmap0->total_domains = 1; 2264 intel_set_map_stripe_count(mmap0, 2265 sd->sd_size / vol->v_strip_size / mmap0->total_domains); 2266 mmap0->failed_disk_num = 0xff; 2267 mmap0->ddf = 1; 2268 2269 /* If there are two maps - copy common and update. */ 2270 if (mvol->migr_state) { 2271 intel_set_vol_curr_migr_unit(mvol, 2272 pos / vol->v_strip_size / mmap0->total_domains); 2273 mmap1 = intel_get_map(mvol, 1); 2274 memcpy(mmap1, mmap0, sizeof(struct intel_raid_map)); 2275 mmap0->status = INTEL_S_READY; 2276 } else 2277 mmap1 = NULL; 2278 2279 /* Write disk indexes and put rebuild flags. */ 2280 for (sdi = 0; sdi < vol->v_disks_count; sdi++) { 2281 sd = &vol->v_subdisks[sdi]; 2282 pd = (struct g_raid_md_intel_perdisk *) 2283 sd->sd_disk->d_md_data; 2284 mmap0->disk_idx[sdi] = pd->pd_disk_pos; 2285 if (mvol->migr_state) 2286 mmap1->disk_idx[sdi] = pd->pd_disk_pos; 2287 if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2288 sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 2289 mmap1->disk_idx[sdi] |= INTEL_DI_RBLD; 2290 } else if (sd->sd_state != G_RAID_SUBDISK_S_ACTIVE && 2291 sd->sd_state != G_RAID_SUBDISK_S_STALE) { 2292 mmap0->disk_idx[sdi] |= INTEL_DI_RBLD; 2293 if (mvol->migr_state) 2294 mmap1->disk_idx[sdi] |= INTEL_DI_RBLD; 2295 } 2296 if ((sd->sd_state == G_RAID_SUBDISK_S_NONE || 2297 sd->sd_state == G_RAID_SUBDISK_S_FAILED) && 2298 mmap0->failed_disk_num == 0xff) { 2299 mmap0->failed_disk_num = sdi; 2300 if (mvol->migr_state) 2301 mmap1->failed_disk_num = sdi; 2302 } 2303 } 2304 vi++; 2305 } 2306 meta->total_volumes = vi; 2307 if (strcmp(version, INTEL_VERSION_1300) != 0) 2308 meta->attributes &= INTEL_ATTR_CHECKSUM; 2309 memcpy(&meta->version[0], version, sizeof(INTEL_VERSION_1000) - 1); 2310 2311 /* We are done. Print meta data and store them to disks. */ 2312 g_raid_md_intel_print(meta); 2313 if (mdi->mdio_meta != NULL) 2314 free(mdi->mdio_meta, M_MD_INTEL); 2315 mdi->mdio_meta = meta; 2316 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 2317 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 2318 if (disk->d_state != G_RAID_DISK_S_ACTIVE) 2319 continue; 2320 if (pd->pd_meta != NULL) { 2321 free(pd->pd_meta, M_MD_INTEL); 2322 pd->pd_meta = NULL; 2323 } 2324 pd->pd_meta = intel_meta_copy(meta); 2325 intel_meta_write(disk->d_consumer, meta); 2326 } 2327 return (0); 2328 } 2329 2330 static int 2331 g_raid_md_fail_disk_intel(struct g_raid_md_object *md, 2332 struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk) 2333 { 2334 struct g_raid_softc *sc; 2335 struct g_raid_md_intel_object *mdi; 2336 struct g_raid_md_intel_perdisk *pd; 2337 struct g_raid_subdisk *sd; 2338 2339 sc = md->mdo_softc; 2340 mdi = (struct g_raid_md_intel_object *)md; 2341 pd = (struct g_raid_md_intel_perdisk *)tdisk->d_md_data; 2342 2343 /* We can't fail disk that is not a part of array now. */ 2344 if (pd->pd_disk_pos < 0) 2345 return (-1); 2346 2347 /* 2348 * Mark disk as failed in metadata and try to write that metadata 2349 * to the disk itself to prevent it's later resurrection as STALE. 2350 */ 2351 mdi->mdio_meta->disk[pd->pd_disk_pos].flags = INTEL_F_FAILED; 2352 pd->pd_disk_meta.flags = INTEL_F_FAILED; 2353 g_raid_md_intel_print(mdi->mdio_meta); 2354 if (tdisk->d_consumer) 2355 intel_meta_write(tdisk->d_consumer, mdi->mdio_meta); 2356 2357 /* Change states. */ 2358 g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED); 2359 TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) { 2360 g_raid_change_subdisk_state(sd, 2361 G_RAID_SUBDISK_S_FAILED); 2362 g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED, 2363 G_RAID_EVENT_SUBDISK); 2364 } 2365 2366 /* Write updated metadata to remaining disks. */ 2367 g_raid_md_write_intel(md, NULL, NULL, tdisk); 2368 2369 /* Check if anything left except placeholders. */ 2370 if (g_raid_ndisks(sc, -1) == 2371 g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) 2372 g_raid_destroy_node(sc, 0); 2373 else 2374 g_raid_md_intel_refill(sc); 2375 return (0); 2376 } 2377 2378 static int 2379 g_raid_md_free_disk_intel(struct g_raid_md_object *md, 2380 struct g_raid_disk *disk) 2381 { 2382 struct g_raid_md_intel_perdisk *pd; 2383 2384 pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; 2385 if (pd->pd_meta != NULL) { 2386 free(pd->pd_meta, M_MD_INTEL); 2387 pd->pd_meta = NULL; 2388 } 2389 free(pd, M_MD_INTEL); 2390 disk->d_md_data = NULL; 2391 return (0); 2392 } 2393 2394 static int 2395 g_raid_md_free_intel(struct g_raid_md_object *md) 2396 { 2397 struct g_raid_md_intel_object *mdi; 2398 2399 mdi = (struct g_raid_md_intel_object *)md; 2400 if (!mdi->mdio_started) { 2401 mdi->mdio_started = 0; 2402 callout_stop(&mdi->mdio_start_co); 2403 G_RAID_DEBUG1(1, md->mdo_softc, 2404 "root_mount_rel %p", mdi->mdio_rootmount); 2405 root_mount_rel(mdi->mdio_rootmount); 2406 mdi->mdio_rootmount = NULL; 2407 } 2408 if (mdi->mdio_meta != NULL) { 2409 free(mdi->mdio_meta, M_MD_INTEL); 2410 mdi->mdio_meta = NULL; 2411 } 2412 return (0); 2413 } 2414 2415 G_RAID_MD_DECLARE(g_raid_md_intel); 2416