1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright 2012 Nexenta Systems, Inc. All rights reserved. 25 * Copyright (c) 2018 by Delphix. All rights reserved. 26 */ 27 28 #include <stdio.h> 29 #include <stdlib.h> 30 #include <errno.h> 31 #include <string.h> 32 #include <strings.h> 33 #include <unistd.h> 34 #include <uuid/uuid.h> 35 #include <zlib.h> 36 #include <libintl.h> 37 #include <sys/types.h> 38 #include <sys/dkio.h> 39 #include <sys/vtoc.h> 40 #include <sys/mhd.h> 41 #include <sys/param.h> 42 #include <sys/dktp/fdisk.h> 43 #include <sys/efi_partition.h> 44 #include <sys/byteorder.h> 45 #include <sys/vdev_disk.h> 46 #include <linux/fs.h> 47 #include <linux/blkpg.h> 48 49 static struct uuid_to_ptag { 50 struct uuid uuid; 51 } conversion_array[] = { 52 { EFI_UNUSED }, 53 { EFI_BOOT }, 54 { EFI_ROOT }, 55 { EFI_SWAP }, 56 { EFI_USR }, 57 { EFI_BACKUP }, 58 { EFI_UNUSED }, /* STAND is never used */ 59 { EFI_VAR }, 60 { EFI_HOME }, 61 { EFI_ALTSCTR }, 62 { EFI_UNUSED }, /* CACHE (cachefs) is never used */ 63 { EFI_RESERVED }, 64 { EFI_SYSTEM }, 65 { EFI_LEGACY_MBR }, 66 { EFI_SYMC_PUB }, 67 { EFI_SYMC_CDS }, 68 { EFI_MSFT_RESV }, 69 { EFI_DELL_BASIC }, 70 { EFI_DELL_RAID }, 71 { EFI_DELL_SWAP }, 72 { EFI_DELL_LVM }, 73 { EFI_DELL_RESV }, 74 { EFI_AAPL_HFS }, 75 { EFI_AAPL_UFS }, 76 { EFI_FREEBSD_BOOT }, 77 { EFI_FREEBSD_SWAP }, 78 { EFI_FREEBSD_UFS }, 79 { EFI_FREEBSD_VINUM }, 80 { EFI_FREEBSD_ZFS }, 81 { EFI_BIOS_BOOT }, 82 { EFI_INTC_RS }, 83 { EFI_SNE_BOOT }, 84 { EFI_LENOVO_BOOT }, 85 { EFI_MSFT_LDMM }, 86 { EFI_MSFT_LDMD }, 87 { EFI_MSFT_RE }, 88 { EFI_IBM_GPFS }, 89 { EFI_MSFT_STORAGESPACES }, 90 { EFI_HPQ_DATA }, 91 { EFI_HPQ_SVC }, 92 { EFI_RHT_DATA }, 93 { EFI_RHT_HOME }, 94 { EFI_RHT_SRV }, 95 { EFI_RHT_DMCRYPT }, 96 { EFI_RHT_LUKS }, 97 { EFI_FREEBSD_DISKLABEL }, 98 { EFI_AAPL_RAID }, 99 { EFI_AAPL_RAIDOFFLINE }, 100 { EFI_AAPL_BOOT }, 101 { EFI_AAPL_LABEL }, 102 { EFI_AAPL_TVRECOVERY }, 103 { EFI_AAPL_CORESTORAGE }, 104 { EFI_NETBSD_SWAP }, 105 { EFI_NETBSD_FFS }, 106 { EFI_NETBSD_LFS }, 107 { EFI_NETBSD_RAID }, 108 { EFI_NETBSD_CAT }, 109 { EFI_NETBSD_CRYPT }, 110 { EFI_GOOG_KERN }, 111 { EFI_GOOG_ROOT }, 112 { EFI_GOOG_RESV }, 113 { EFI_HAIKU_BFS }, 114 { EFI_MIDNIGHTBSD_BOOT }, 115 { EFI_MIDNIGHTBSD_DATA }, 116 { EFI_MIDNIGHTBSD_SWAP }, 117 { EFI_MIDNIGHTBSD_UFS }, 118 { EFI_MIDNIGHTBSD_VINUM }, 119 { EFI_MIDNIGHTBSD_ZFS }, 120 { EFI_CEPH_JOURNAL }, 121 { EFI_CEPH_DMCRYPTJOURNAL }, 122 { EFI_CEPH_OSD }, 123 { EFI_CEPH_DMCRYPTOSD }, 124 { EFI_CEPH_CREATE }, 125 { EFI_CEPH_DMCRYPTCREATE }, 126 { EFI_OPENBSD_DISKLABEL }, 127 { EFI_BBRY_QNX }, 128 { EFI_BELL_PLAN9 }, 129 { EFI_VMW_KCORE }, 130 { EFI_VMW_VMFS }, 131 { EFI_VMW_RESV }, 132 { EFI_RHT_ROOTX86 }, 133 { EFI_RHT_ROOTAMD64 }, 134 { EFI_RHT_ROOTARM }, 135 { EFI_RHT_ROOTARM64 }, 136 { EFI_ACRONIS_SECUREZONE }, 137 { EFI_ONIE_BOOT }, 138 { EFI_ONIE_CONFIG }, 139 { EFI_IBM_PPRPBOOT }, 140 { EFI_FREEDESKTOP_BOOT } 141 }; 142 143 /* 144 * Default vtoc information for non-SVr4 partitions 145 */ 146 struct dk_map2 default_vtoc_map[NDKMAP] = { 147 { V_ROOT, 0 }, /* a - 0 */ 148 { V_SWAP, V_UNMNT }, /* b - 1 */ 149 { V_BACKUP, V_UNMNT }, /* c - 2 */ 150 { V_UNASSIGNED, 0 }, /* d - 3 */ 151 { V_UNASSIGNED, 0 }, /* e - 4 */ 152 { V_UNASSIGNED, 0 }, /* f - 5 */ 153 { V_USR, 0 }, /* g - 6 */ 154 { V_UNASSIGNED, 0 }, /* h - 7 */ 155 156 #if defined(_SUNOS_VTOC_16) 157 158 #if defined(i386) || defined(__amd64) || defined(__arm) || \ 159 defined(__powerpc) || defined(__sparc) || defined(__s390__) || \ 160 defined(__mips__) || defined(__rv64g__) 161 { V_BOOT, V_UNMNT }, /* i - 8 */ 162 { V_ALTSCTR, 0 }, /* j - 9 */ 163 164 #else 165 #error No VTOC format defined. 166 #endif /* defined(i386) */ 167 168 { V_UNASSIGNED, 0 }, /* k - 10 */ 169 { V_UNASSIGNED, 0 }, /* l - 11 */ 170 { V_UNASSIGNED, 0 }, /* m - 12 */ 171 { V_UNASSIGNED, 0 }, /* n - 13 */ 172 { V_UNASSIGNED, 0 }, /* o - 14 */ 173 { V_UNASSIGNED, 0 }, /* p - 15 */ 174 #endif /* defined(_SUNOS_VTOC_16) */ 175 }; 176 177 int efi_debug = 0; 178 179 static int efi_read(int, struct dk_gpt *); 180 181 /* 182 * Return a 32-bit CRC of the contents of the buffer. Pre-and-post 183 * one's conditioning will be handled by crc32() internally. 184 */ 185 static uint32_t 186 efi_crc32(const unsigned char *buf, unsigned int size) 187 { 188 uint32_t crc = crc32(0, Z_NULL, 0); 189 190 crc = crc32(crc, buf, size); 191 192 return (crc); 193 } 194 195 static int 196 read_disk_info(int fd, diskaddr_t *capacity, uint_t *lbsize) 197 { 198 int sector_size; 199 unsigned long long capacity_size; 200 201 if (ioctl(fd, BLKSSZGET, §or_size) < 0) 202 return (-1); 203 204 if (ioctl(fd, BLKGETSIZE64, &capacity_size) < 0) 205 return (-1); 206 207 *lbsize = (uint_t)sector_size; 208 *capacity = (diskaddr_t)(capacity_size / sector_size); 209 210 return (0); 211 } 212 213 /* 214 * Return back the device name associated with the file descriptor. The 215 * caller is responsible for freeing the memory associated with the 216 * returned string. 217 */ 218 static char * 219 efi_get_devname(int fd) 220 { 221 char path[32]; 222 223 /* 224 * The libefi API only provides the open fd and not the file path. 225 * To handle this realpath(3) is used to resolve the block device 226 * name from /proc/self/fd/<fd>. 227 */ 228 (void) snprintf(path, sizeof (path), "/proc/self/fd/%d", fd); 229 return (realpath(path, NULL)); 230 } 231 232 static int 233 efi_get_info(int fd, struct dk_cinfo *dki_info) 234 { 235 char *dev_path; 236 int rval = 0; 237 238 memset(dki_info, 0, sizeof (*dki_info)); 239 240 /* 241 * The simplest way to get the partition number under linux is 242 * to parse it out of the /dev/<disk><partition> block device name. 243 * The kernel creates this using the partition number when it 244 * populates /dev/ so it may be trusted. The tricky bit here is 245 * that the naming convention is based on the block device type. 246 * So we need to take this in to account when parsing out the 247 * partition information. Aside from the partition number we collect 248 * some additional device info. 249 */ 250 dev_path = efi_get_devname(fd); 251 if (dev_path == NULL) 252 goto error; 253 254 if ((strncmp(dev_path, "/dev/sd", 7) == 0)) { 255 strcpy(dki_info->dki_cname, "sd"); 256 dki_info->dki_ctype = DKC_SCSI_CCS; 257 rval = sscanf(dev_path, "/dev/%[a-zA-Z]%hu", 258 dki_info->dki_dname, 259 &dki_info->dki_partition); 260 } else if ((strncmp(dev_path, "/dev/hd", 7) == 0)) { 261 strcpy(dki_info->dki_cname, "hd"); 262 dki_info->dki_ctype = DKC_DIRECT; 263 rval = sscanf(dev_path, "/dev/%[a-zA-Z]%hu", 264 dki_info->dki_dname, 265 &dki_info->dki_partition); 266 } else if ((strncmp(dev_path, "/dev/md", 7) == 0)) { 267 strcpy(dki_info->dki_cname, "pseudo"); 268 dki_info->dki_ctype = DKC_MD; 269 strcpy(dki_info->dki_dname, "md"); 270 rval = sscanf(dev_path, "/dev/md%[0-9]p%hu", 271 dki_info->dki_dname + 2, 272 &dki_info->dki_partition); 273 } else if ((strncmp(dev_path, "/dev/vd", 7) == 0)) { 274 strcpy(dki_info->dki_cname, "vd"); 275 dki_info->dki_ctype = DKC_MD; 276 rval = sscanf(dev_path, "/dev/%[a-zA-Z]%hu", 277 dki_info->dki_dname, 278 &dki_info->dki_partition); 279 } else if ((strncmp(dev_path, "/dev/xvd", 8) == 0)) { 280 strcpy(dki_info->dki_cname, "xvd"); 281 dki_info->dki_ctype = DKC_MD; 282 rval = sscanf(dev_path, "/dev/%[a-zA-Z]%hu", 283 dki_info->dki_dname, 284 &dki_info->dki_partition); 285 } else if ((strncmp(dev_path, "/dev/zd", 7) == 0)) { 286 strcpy(dki_info->dki_cname, "zd"); 287 dki_info->dki_ctype = DKC_MD; 288 strcpy(dki_info->dki_dname, "zd"); 289 rval = sscanf(dev_path, "/dev/zd%[0-9]p%hu", 290 dki_info->dki_dname + 2, 291 &dki_info->dki_partition); 292 } else if ((strncmp(dev_path, "/dev/dm-", 8) == 0)) { 293 strcpy(dki_info->dki_cname, "pseudo"); 294 dki_info->dki_ctype = DKC_VBD; 295 strcpy(dki_info->dki_dname, "dm-"); 296 rval = sscanf(dev_path, "/dev/dm-%[0-9]p%hu", 297 dki_info->dki_dname + 3, 298 &dki_info->dki_partition); 299 } else if ((strncmp(dev_path, "/dev/ram", 8) == 0)) { 300 strcpy(dki_info->dki_cname, "pseudo"); 301 dki_info->dki_ctype = DKC_PCMCIA_MEM; 302 strcpy(dki_info->dki_dname, "ram"); 303 rval = sscanf(dev_path, "/dev/ram%[0-9]p%hu", 304 dki_info->dki_dname + 3, 305 &dki_info->dki_partition); 306 } else if ((strncmp(dev_path, "/dev/loop", 9) == 0)) { 307 strcpy(dki_info->dki_cname, "pseudo"); 308 dki_info->dki_ctype = DKC_VBD; 309 strcpy(dki_info->dki_dname, "loop"); 310 rval = sscanf(dev_path, "/dev/loop%[0-9]p%hu", 311 dki_info->dki_dname + 4, 312 &dki_info->dki_partition); 313 } else if ((strncmp(dev_path, "/dev/nvme", 9) == 0)) { 314 strcpy(dki_info->dki_cname, "nvme"); 315 dki_info->dki_ctype = DKC_SCSI_CCS; 316 strcpy(dki_info->dki_dname, "nvme"); 317 (void) sscanf(dev_path, "/dev/nvme%[0-9]", 318 dki_info->dki_dname + 4); 319 size_t controller_length = strlen( 320 dki_info->dki_dname); 321 strcpy(dki_info->dki_dname + controller_length, 322 "n"); 323 rval = sscanf(dev_path, 324 "/dev/nvme%*[0-9]n%[0-9]p%hu", 325 dki_info->dki_dname + controller_length + 1, 326 &dki_info->dki_partition); 327 } else { 328 strcpy(dki_info->dki_dname, "unknown"); 329 strcpy(dki_info->dki_cname, "unknown"); 330 dki_info->dki_ctype = DKC_UNKNOWN; 331 } 332 333 switch (rval) { 334 case 0: 335 errno = EINVAL; 336 goto error; 337 case 1: 338 dki_info->dki_partition = 0; 339 } 340 341 free(dev_path); 342 343 return (0); 344 error: 345 if (efi_debug) 346 (void) fprintf(stderr, "DKIOCINFO errno 0x%x\n", errno); 347 348 switch (errno) { 349 case EIO: 350 return (VT_EIO); 351 case EINVAL: 352 return (VT_EINVAL); 353 default: 354 return (VT_ERROR); 355 } 356 } 357 358 /* 359 * the number of blocks the EFI label takes up (round up to nearest 360 * block) 361 */ 362 #define NBLOCKS(p, l) (1 + ((((p) * (int)sizeof (efi_gpe_t)) + \ 363 ((l) - 1)) / (l))) 364 /* number of partitions -- limited by what we can malloc */ 365 #define MAX_PARTS ((4294967295UL - sizeof (struct dk_gpt)) / \ 366 sizeof (struct dk_part)) 367 368 int 369 efi_alloc_and_init(int fd, uint32_t nparts, struct dk_gpt **vtoc) 370 { 371 diskaddr_t capacity = 0; 372 uint_t lbsize = 0; 373 uint_t nblocks; 374 size_t length; 375 struct dk_gpt *vptr; 376 struct uuid uuid; 377 struct dk_cinfo dki_info; 378 379 if (read_disk_info(fd, &capacity, &lbsize) != 0) 380 return (-1); 381 382 if (efi_get_info(fd, &dki_info) != 0) 383 return (-1); 384 385 if (dki_info.dki_partition != 0) 386 return (-1); 387 388 if ((dki_info.dki_ctype == DKC_PCMCIA_MEM) || 389 (dki_info.dki_ctype == DKC_VBD) || 390 (dki_info.dki_ctype == DKC_UNKNOWN)) 391 return (-1); 392 393 nblocks = NBLOCKS(nparts, lbsize); 394 if ((nblocks * lbsize) < EFI_MIN_ARRAY_SIZE + lbsize) { 395 /* 16K plus one block for the GPT */ 396 nblocks = EFI_MIN_ARRAY_SIZE / lbsize + 1; 397 } 398 399 if (nparts > MAX_PARTS) { 400 if (efi_debug) { 401 (void) fprintf(stderr, 402 "the maximum number of partitions supported is %lu\n", 403 MAX_PARTS); 404 } 405 return (-1); 406 } 407 408 length = sizeof (struct dk_gpt) + 409 sizeof (struct dk_part) * (nparts - 1); 410 411 vptr = calloc(1, length); 412 if (vptr == NULL) 413 return (-1); 414 415 *vtoc = vptr; 416 417 vptr->efi_version = EFI_VERSION_CURRENT; 418 vptr->efi_lbasize = lbsize; 419 vptr->efi_nparts = nparts; 420 /* 421 * add one block here for the PMBR; on disks with a 512 byte 422 * block size and 128 or fewer partitions, efi_first_u_lba 423 * should work out to "34" 424 */ 425 vptr->efi_first_u_lba = nblocks + 1; 426 vptr->efi_last_lba = capacity - 1; 427 vptr->efi_altern_lba = capacity -1; 428 vptr->efi_last_u_lba = vptr->efi_last_lba - nblocks; 429 430 (void) uuid_generate((uchar_t *)&uuid); 431 UUID_LE_CONVERT(vptr->efi_disk_uguid, uuid); 432 return (0); 433 } 434 435 /* 436 * Read EFI - return partition number upon success. 437 */ 438 int 439 efi_alloc_and_read(int fd, struct dk_gpt **vtoc) 440 { 441 int rval; 442 uint32_t nparts; 443 int length; 444 struct dk_gpt *vptr; 445 446 /* figure out the number of entries that would fit into 16K */ 447 nparts = EFI_MIN_ARRAY_SIZE / sizeof (efi_gpe_t); 448 length = (int) sizeof (struct dk_gpt) + 449 (int) sizeof (struct dk_part) * (nparts - 1); 450 vptr = calloc(1, length); 451 452 if (vptr == NULL) 453 return (VT_ERROR); 454 455 vptr->efi_nparts = nparts; 456 rval = efi_read(fd, vptr); 457 458 if ((rval == VT_EINVAL) && vptr->efi_nparts > nparts) { 459 void *tmp; 460 length = (int) sizeof (struct dk_gpt) + 461 (int) sizeof (struct dk_part) * (vptr->efi_nparts - 1); 462 nparts = vptr->efi_nparts; 463 if ((tmp = realloc(vptr, length)) == NULL) { 464 /* cppcheck-suppress doubleFree */ 465 free(vptr); 466 *vtoc = NULL; 467 return (VT_ERROR); 468 } else { 469 vptr = tmp; 470 rval = efi_read(fd, vptr); 471 } 472 } 473 474 if (rval < 0) { 475 if (efi_debug) { 476 (void) fprintf(stderr, 477 "read of EFI table failed, rval=%d\n", rval); 478 } 479 free(vptr); 480 *vtoc = NULL; 481 } else { 482 *vtoc = vptr; 483 } 484 485 return (rval); 486 } 487 488 static int 489 efi_ioctl(int fd, int cmd, dk_efi_t *dk_ioc) 490 { 491 void *data = dk_ioc->dki_data; 492 int error; 493 diskaddr_t capacity; 494 uint_t lbsize; 495 496 /* 497 * When the IO is not being performed in kernel as an ioctl we need 498 * to know the sector size so we can seek to the proper byte offset. 499 */ 500 if (read_disk_info(fd, &capacity, &lbsize) == -1) { 501 if (efi_debug) 502 fprintf(stderr, "unable to read disk info: %d", errno); 503 504 errno = EIO; 505 return (-1); 506 } 507 508 switch (cmd) { 509 case DKIOCGETEFI: 510 if (lbsize == 0) { 511 if (efi_debug) 512 (void) fprintf(stderr, "DKIOCGETEFI assuming " 513 "LBA %d bytes\n", DEV_BSIZE); 514 515 lbsize = DEV_BSIZE; 516 } 517 518 error = lseek(fd, dk_ioc->dki_lba * lbsize, SEEK_SET); 519 if (error == -1) { 520 if (efi_debug) 521 (void) fprintf(stderr, "DKIOCGETEFI lseek " 522 "error: %d\n", errno); 523 return (error); 524 } 525 526 error = read(fd, data, dk_ioc->dki_length); 527 if (error == -1) { 528 if (efi_debug) 529 (void) fprintf(stderr, "DKIOCGETEFI read " 530 "error: %d\n", errno); 531 return (error); 532 } 533 534 if (error != dk_ioc->dki_length) { 535 if (efi_debug) 536 (void) fprintf(stderr, "DKIOCGETEFI short " 537 "read of %d bytes\n", error); 538 errno = EIO; 539 return (-1); 540 } 541 error = 0; 542 break; 543 544 case DKIOCSETEFI: 545 if (lbsize == 0) { 546 if (efi_debug) 547 (void) fprintf(stderr, "DKIOCSETEFI unknown " 548 "LBA size\n"); 549 errno = EIO; 550 return (-1); 551 } 552 553 error = lseek(fd, dk_ioc->dki_lba * lbsize, SEEK_SET); 554 if (error == -1) { 555 if (efi_debug) 556 (void) fprintf(stderr, "DKIOCSETEFI lseek " 557 "error: %d\n", errno); 558 return (error); 559 } 560 561 error = write(fd, data, dk_ioc->dki_length); 562 if (error == -1) { 563 if (efi_debug) 564 (void) fprintf(stderr, "DKIOCSETEFI write " 565 "error: %d\n", errno); 566 return (error); 567 } 568 569 if (error != dk_ioc->dki_length) { 570 if (efi_debug) 571 (void) fprintf(stderr, "DKIOCSETEFI short " 572 "write of %d bytes\n", error); 573 errno = EIO; 574 return (-1); 575 } 576 577 /* Sync the new EFI table to disk */ 578 error = fsync(fd); 579 if (error == -1) 580 return (error); 581 582 /* Ensure any local disk cache is also flushed */ 583 if (ioctl(fd, BLKFLSBUF, 0) == -1) 584 return (error); 585 586 error = 0; 587 break; 588 589 default: 590 if (efi_debug) 591 (void) fprintf(stderr, "unsupported ioctl()\n"); 592 593 errno = EIO; 594 return (-1); 595 } 596 597 return (error); 598 } 599 600 int 601 efi_rescan(int fd) 602 { 603 int retry = 10; 604 int error; 605 606 /* Notify the kernel a devices partition table has been updated */ 607 while ((error = ioctl(fd, BLKRRPART)) != 0) { 608 if ((--retry == 0) || (errno != EBUSY)) { 609 (void) fprintf(stderr, "the kernel failed to rescan " 610 "the partition table: %d\n", errno); 611 return (-1); 612 } 613 usleep(50000); 614 } 615 616 return (0); 617 } 618 619 static int 620 check_label(int fd, dk_efi_t *dk_ioc) 621 { 622 efi_gpt_t *efi; 623 uint_t crc; 624 625 if (efi_ioctl(fd, DKIOCGETEFI, dk_ioc) == -1) { 626 switch (errno) { 627 case EIO: 628 return (VT_EIO); 629 default: 630 return (VT_ERROR); 631 } 632 } 633 efi = dk_ioc->dki_data; 634 if (efi->efi_gpt_Signature != LE_64(EFI_SIGNATURE)) { 635 if (efi_debug) 636 (void) fprintf(stderr, 637 "Bad EFI signature: 0x%llx != 0x%llx\n", 638 (long long)efi->efi_gpt_Signature, 639 (long long)LE_64(EFI_SIGNATURE)); 640 return (VT_EINVAL); 641 } 642 643 /* 644 * check CRC of the header; the size of the header should 645 * never be larger than one block 646 */ 647 crc = efi->efi_gpt_HeaderCRC32; 648 efi->efi_gpt_HeaderCRC32 = 0; 649 len_t headerSize = (len_t)LE_32(efi->efi_gpt_HeaderSize); 650 651 if (headerSize < EFI_MIN_LABEL_SIZE || headerSize > EFI_LABEL_SIZE) { 652 if (efi_debug) 653 (void) fprintf(stderr, 654 "Invalid EFI HeaderSize %llu. Assuming %d.\n", 655 headerSize, EFI_MIN_LABEL_SIZE); 656 } 657 658 if ((headerSize > dk_ioc->dki_length) || 659 crc != LE_32(efi_crc32((unsigned char *)efi, headerSize))) { 660 if (efi_debug) 661 (void) fprintf(stderr, 662 "Bad EFI CRC: 0x%x != 0x%x\n", 663 crc, LE_32(efi_crc32((unsigned char *)efi, 664 headerSize))); 665 return (VT_EINVAL); 666 } 667 668 return (0); 669 } 670 671 static int 672 efi_read(int fd, struct dk_gpt *vtoc) 673 { 674 int i, j; 675 int label_len; 676 int rval = 0; 677 int md_flag = 0; 678 int vdc_flag = 0; 679 diskaddr_t capacity = 0; 680 uint_t lbsize = 0; 681 struct dk_minfo disk_info; 682 dk_efi_t dk_ioc; 683 efi_gpt_t *efi; 684 efi_gpe_t *efi_parts; 685 struct dk_cinfo dki_info; 686 uint32_t user_length; 687 boolean_t legacy_label = B_FALSE; 688 689 /* 690 * get the partition number for this file descriptor. 691 */ 692 if ((rval = efi_get_info(fd, &dki_info)) != 0) 693 return (rval); 694 695 if ((strncmp(dki_info.dki_cname, "pseudo", 7) == 0) && 696 (strncmp(dki_info.dki_dname, "md", 3) == 0)) { 697 md_flag++; 698 } else if ((strncmp(dki_info.dki_cname, "vdc", 4) == 0) && 699 (strncmp(dki_info.dki_dname, "vdc", 4) == 0)) { 700 /* 701 * The controller and drive name "vdc" (virtual disk client) 702 * indicates a LDoms virtual disk. 703 */ 704 vdc_flag++; 705 } 706 707 /* get the LBA size */ 708 if (read_disk_info(fd, &capacity, &lbsize) == -1) { 709 if (efi_debug) { 710 (void) fprintf(stderr, 711 "unable to read disk info: %d", 712 errno); 713 } 714 return (VT_EINVAL); 715 } 716 717 disk_info.dki_lbsize = lbsize; 718 disk_info.dki_capacity = capacity; 719 720 if (disk_info.dki_lbsize == 0) { 721 if (efi_debug) { 722 (void) fprintf(stderr, 723 "efi_read: assuming LBA 512 bytes\n"); 724 } 725 disk_info.dki_lbsize = DEV_BSIZE; 726 } 727 /* 728 * Read the EFI GPT to figure out how many partitions we need 729 * to deal with. 730 */ 731 dk_ioc.dki_lba = 1; 732 if (NBLOCKS(vtoc->efi_nparts, disk_info.dki_lbsize) < 34) { 733 label_len = EFI_MIN_ARRAY_SIZE + disk_info.dki_lbsize; 734 } else { 735 label_len = vtoc->efi_nparts * (int) sizeof (efi_gpe_t) + 736 disk_info.dki_lbsize; 737 if (label_len % disk_info.dki_lbsize) { 738 /* pad to physical sector size */ 739 label_len += disk_info.dki_lbsize; 740 label_len &= ~(disk_info.dki_lbsize - 1); 741 } 742 } 743 744 if (posix_memalign((void **)&dk_ioc.dki_data, 745 disk_info.dki_lbsize, label_len)) 746 return (VT_ERROR); 747 748 memset(dk_ioc.dki_data, 0, label_len); 749 dk_ioc.dki_length = disk_info.dki_lbsize; 750 user_length = vtoc->efi_nparts; 751 efi = dk_ioc.dki_data; 752 if (md_flag) { 753 dk_ioc.dki_length = label_len; 754 if (efi_ioctl(fd, DKIOCGETEFI, &dk_ioc) == -1) { 755 switch (errno) { 756 case EIO: 757 return (VT_EIO); 758 default: 759 return (VT_ERROR); 760 } 761 } 762 } else if ((rval = check_label(fd, &dk_ioc)) == VT_EINVAL) { 763 /* 764 * No valid label here; try the alternate. Note that here 765 * we just read GPT header and save it into dk_ioc.data, 766 * Later, we will read GUID partition entry array if we 767 * can get valid GPT header. 768 */ 769 770 /* 771 * This is a workaround for legacy systems. In the past, the 772 * last sector of SCSI disk was invisible on x86 platform. At 773 * that time, backup label was saved on the next to the last 774 * sector. It is possible for users to move a disk from previous 775 * solaris system to present system. Here, we attempt to search 776 * legacy backup EFI label first. 777 */ 778 dk_ioc.dki_lba = disk_info.dki_capacity - 2; 779 dk_ioc.dki_length = disk_info.dki_lbsize; 780 rval = check_label(fd, &dk_ioc); 781 if (rval == VT_EINVAL) { 782 /* 783 * we didn't find legacy backup EFI label, try to 784 * search backup EFI label in the last block. 785 */ 786 dk_ioc.dki_lba = disk_info.dki_capacity - 1; 787 dk_ioc.dki_length = disk_info.dki_lbsize; 788 rval = check_label(fd, &dk_ioc); 789 if (rval == 0) { 790 legacy_label = B_TRUE; 791 if (efi_debug) 792 (void) fprintf(stderr, 793 "efi_read: primary label corrupt; " 794 "using EFI backup label located on" 795 " the last block\n"); 796 } 797 } else { 798 if ((efi_debug) && (rval == 0)) 799 (void) fprintf(stderr, "efi_read: primary label" 800 " corrupt; using legacy EFI backup label " 801 " located on the next to last block\n"); 802 } 803 804 if (rval == 0) { 805 dk_ioc.dki_lba = LE_64(efi->efi_gpt_PartitionEntryLBA); 806 vtoc->efi_flags |= EFI_GPT_PRIMARY_CORRUPT; 807 vtoc->efi_nparts = 808 LE_32(efi->efi_gpt_NumberOfPartitionEntries); 809 /* 810 * Partition tables are between backup GPT header 811 * table and ParitionEntryLBA (the starting LBA of 812 * the GUID partition entries array). Now that we 813 * already got valid GPT header and saved it in 814 * dk_ioc.dki_data, we try to get GUID partition 815 * entry array here. 816 */ 817 /* LINTED */ 818 dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data 819 + disk_info.dki_lbsize); 820 if (legacy_label) 821 dk_ioc.dki_length = disk_info.dki_capacity - 1 - 822 dk_ioc.dki_lba; 823 else 824 dk_ioc.dki_length = disk_info.dki_capacity - 2 - 825 dk_ioc.dki_lba; 826 dk_ioc.dki_length *= disk_info.dki_lbsize; 827 if (dk_ioc.dki_length > 828 ((len_t)label_len - sizeof (*dk_ioc.dki_data))) { 829 rval = VT_EINVAL; 830 } else { 831 /* 832 * read GUID partition entry array 833 */ 834 rval = efi_ioctl(fd, DKIOCGETEFI, &dk_ioc); 835 } 836 } 837 838 } else if (rval == 0) { 839 840 dk_ioc.dki_lba = LE_64(efi->efi_gpt_PartitionEntryLBA); 841 /* LINTED */ 842 dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data 843 + disk_info.dki_lbsize); 844 dk_ioc.dki_length = label_len - disk_info.dki_lbsize; 845 rval = efi_ioctl(fd, DKIOCGETEFI, &dk_ioc); 846 847 } else if (vdc_flag && rval == VT_ERROR && errno == EINVAL) { 848 /* 849 * When the device is a LDoms virtual disk, the DKIOCGETEFI 850 * ioctl can fail with EINVAL if the virtual disk backend 851 * is a ZFS volume serviced by a domain running an old version 852 * of Solaris. This is because the DKIOCGETEFI ioctl was 853 * initially incorrectly implemented for a ZFS volume and it 854 * expected the GPT and GPE to be retrieved with a single ioctl. 855 * So we try to read the GPT and the GPE using that old style 856 * ioctl. 857 */ 858 dk_ioc.dki_lba = 1; 859 dk_ioc.dki_length = label_len; 860 rval = check_label(fd, &dk_ioc); 861 } 862 863 if (rval < 0) { 864 free(efi); 865 return (rval); 866 } 867 868 /* LINTED -- always longlong aligned */ 869 efi_parts = (efi_gpe_t *)(((char *)efi) + disk_info.dki_lbsize); 870 871 /* 872 * Assemble this into a "dk_gpt" struct for easier 873 * digestibility by applications. 874 */ 875 vtoc->efi_version = LE_32(efi->efi_gpt_Revision); 876 vtoc->efi_nparts = LE_32(efi->efi_gpt_NumberOfPartitionEntries); 877 vtoc->efi_part_size = LE_32(efi->efi_gpt_SizeOfPartitionEntry); 878 vtoc->efi_lbasize = disk_info.dki_lbsize; 879 vtoc->efi_last_lba = disk_info.dki_capacity - 1; 880 vtoc->efi_first_u_lba = LE_64(efi->efi_gpt_FirstUsableLBA); 881 vtoc->efi_last_u_lba = LE_64(efi->efi_gpt_LastUsableLBA); 882 vtoc->efi_altern_lba = LE_64(efi->efi_gpt_AlternateLBA); 883 UUID_LE_CONVERT(vtoc->efi_disk_uguid, efi->efi_gpt_DiskGUID); 884 885 /* 886 * If the array the user passed in is too small, set the length 887 * to what it needs to be and return 888 */ 889 if (user_length < vtoc->efi_nparts) { 890 return (VT_EINVAL); 891 } 892 893 for (i = 0; i < vtoc->efi_nparts; i++) { 894 895 UUID_LE_CONVERT(vtoc->efi_parts[i].p_guid, 896 efi_parts[i].efi_gpe_PartitionTypeGUID); 897 898 for (j = 0; 899 j < sizeof (conversion_array) 900 / sizeof (struct uuid_to_ptag); j++) { 901 902 if (bcmp(&vtoc->efi_parts[i].p_guid, 903 &conversion_array[j].uuid, 904 sizeof (struct uuid)) == 0) { 905 vtoc->efi_parts[i].p_tag = j; 906 break; 907 } 908 } 909 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED) 910 continue; 911 vtoc->efi_parts[i].p_flag = 912 LE_16(efi_parts[i].efi_gpe_Attributes.PartitionAttrs); 913 vtoc->efi_parts[i].p_start = 914 LE_64(efi_parts[i].efi_gpe_StartingLBA); 915 vtoc->efi_parts[i].p_size = 916 LE_64(efi_parts[i].efi_gpe_EndingLBA) - 917 vtoc->efi_parts[i].p_start + 1; 918 for (j = 0; j < EFI_PART_NAME_LEN; j++) { 919 vtoc->efi_parts[i].p_name[j] = 920 (uchar_t)LE_16( 921 efi_parts[i].efi_gpe_PartitionName[j]); 922 } 923 924 UUID_LE_CONVERT(vtoc->efi_parts[i].p_uguid, 925 efi_parts[i].efi_gpe_UniquePartitionGUID); 926 } 927 free(efi); 928 929 return (dki_info.dki_partition); 930 } 931 932 /* writes a "protective" MBR */ 933 static int 934 write_pmbr(int fd, struct dk_gpt *vtoc) 935 { 936 dk_efi_t dk_ioc; 937 struct mboot mb; 938 uchar_t *cp; 939 diskaddr_t size_in_lba; 940 uchar_t *buf; 941 int len; 942 943 len = (vtoc->efi_lbasize == 0) ? sizeof (mb) : vtoc->efi_lbasize; 944 if (posix_memalign((void **)&buf, len, len)) 945 return (VT_ERROR); 946 947 /* 948 * Preserve any boot code and disk signature if the first block is 949 * already an MBR. 950 */ 951 memset(buf, 0, len); 952 dk_ioc.dki_lba = 0; 953 dk_ioc.dki_length = len; 954 /* LINTED -- always longlong aligned */ 955 dk_ioc.dki_data = (efi_gpt_t *)buf; 956 if (efi_ioctl(fd, DKIOCGETEFI, &dk_ioc) == -1) { 957 (void) memcpy(&mb, buf, sizeof (mb)); 958 bzero(&mb, sizeof (mb)); 959 mb.signature = LE_16(MBB_MAGIC); 960 } else { 961 (void) memcpy(&mb, buf, sizeof (mb)); 962 if (mb.signature != LE_16(MBB_MAGIC)) { 963 bzero(&mb, sizeof (mb)); 964 mb.signature = LE_16(MBB_MAGIC); 965 } 966 } 967 968 bzero(&mb.parts, sizeof (mb.parts)); 969 cp = (uchar_t *)&mb.parts[0]; 970 /* bootable or not */ 971 *cp++ = 0; 972 /* beginning CHS; 0xffffff if not representable */ 973 *cp++ = 0xff; 974 *cp++ = 0xff; 975 *cp++ = 0xff; 976 /* OS type */ 977 *cp++ = EFI_PMBR; 978 /* ending CHS; 0xffffff if not representable */ 979 *cp++ = 0xff; 980 *cp++ = 0xff; 981 *cp++ = 0xff; 982 /* starting LBA: 1 (little endian format) by EFI definition */ 983 *cp++ = 0x01; 984 *cp++ = 0x00; 985 *cp++ = 0x00; 986 *cp++ = 0x00; 987 /* ending LBA: last block on the disk (little endian format) */ 988 size_in_lba = vtoc->efi_last_lba; 989 if (size_in_lba < 0xffffffff) { 990 *cp++ = (size_in_lba & 0x000000ff); 991 *cp++ = (size_in_lba & 0x0000ff00) >> 8; 992 *cp++ = (size_in_lba & 0x00ff0000) >> 16; 993 *cp++ = (size_in_lba & 0xff000000) >> 24; 994 } else { 995 *cp++ = 0xff; 996 *cp++ = 0xff; 997 *cp++ = 0xff; 998 *cp++ = 0xff; 999 } 1000 1001 (void) memcpy(buf, &mb, sizeof (mb)); 1002 /* LINTED -- always longlong aligned */ 1003 dk_ioc.dki_data = (efi_gpt_t *)buf; 1004 dk_ioc.dki_lba = 0; 1005 dk_ioc.dki_length = len; 1006 if (efi_ioctl(fd, DKIOCSETEFI, &dk_ioc) == -1) { 1007 free(buf); 1008 switch (errno) { 1009 case EIO: 1010 return (VT_EIO); 1011 case EINVAL: 1012 return (VT_EINVAL); 1013 default: 1014 return (VT_ERROR); 1015 } 1016 } 1017 free(buf); 1018 return (0); 1019 } 1020 1021 /* make sure the user specified something reasonable */ 1022 static int 1023 check_input(struct dk_gpt *vtoc) 1024 { 1025 int resv_part = -1; 1026 int i, j; 1027 diskaddr_t istart, jstart, isize, jsize, endsect; 1028 1029 /* 1030 * Sanity-check the input (make sure no partitions overlap) 1031 */ 1032 for (i = 0; i < vtoc->efi_nparts; i++) { 1033 /* It can't be unassigned and have an actual size */ 1034 if ((vtoc->efi_parts[i].p_tag == V_UNASSIGNED) && 1035 (vtoc->efi_parts[i].p_size != 0)) { 1036 if (efi_debug) { 1037 (void) fprintf(stderr, "partition %d is " 1038 "\"unassigned\" but has a size of %llu", 1039 i, vtoc->efi_parts[i].p_size); 1040 } 1041 return (VT_EINVAL); 1042 } 1043 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED) { 1044 if (uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid)) 1045 continue; 1046 /* we have encountered an unknown uuid */ 1047 vtoc->efi_parts[i].p_tag = 0xff; 1048 } 1049 if (vtoc->efi_parts[i].p_tag == V_RESERVED) { 1050 if (resv_part != -1) { 1051 if (efi_debug) { 1052 (void) fprintf(stderr, "found " 1053 "duplicate reserved partition " 1054 "at %d\n", i); 1055 } 1056 return (VT_EINVAL); 1057 } 1058 resv_part = i; 1059 } 1060 if ((vtoc->efi_parts[i].p_start < vtoc->efi_first_u_lba) || 1061 (vtoc->efi_parts[i].p_start > vtoc->efi_last_u_lba)) { 1062 if (efi_debug) { 1063 (void) fprintf(stderr, 1064 "Partition %d starts at %llu. ", 1065 i, 1066 vtoc->efi_parts[i].p_start); 1067 (void) fprintf(stderr, 1068 "It must be between %llu and %llu.\n", 1069 vtoc->efi_first_u_lba, 1070 vtoc->efi_last_u_lba); 1071 } 1072 return (VT_EINVAL); 1073 } 1074 if ((vtoc->efi_parts[i].p_start + 1075 vtoc->efi_parts[i].p_size < 1076 vtoc->efi_first_u_lba) || 1077 (vtoc->efi_parts[i].p_start + 1078 vtoc->efi_parts[i].p_size > 1079 vtoc->efi_last_u_lba + 1)) { 1080 if (efi_debug) { 1081 (void) fprintf(stderr, 1082 "Partition %d ends at %llu. ", 1083 i, 1084 vtoc->efi_parts[i].p_start + 1085 vtoc->efi_parts[i].p_size); 1086 (void) fprintf(stderr, 1087 "It must be between %llu and %llu.\n", 1088 vtoc->efi_first_u_lba, 1089 vtoc->efi_last_u_lba); 1090 } 1091 return (VT_EINVAL); 1092 } 1093 1094 for (j = 0; j < vtoc->efi_nparts; j++) { 1095 isize = vtoc->efi_parts[i].p_size; 1096 jsize = vtoc->efi_parts[j].p_size; 1097 istart = vtoc->efi_parts[i].p_start; 1098 jstart = vtoc->efi_parts[j].p_start; 1099 if ((i != j) && (isize != 0) && (jsize != 0)) { 1100 endsect = jstart + jsize -1; 1101 if ((jstart <= istart) && 1102 (istart <= endsect)) { 1103 if (efi_debug) { 1104 (void) fprintf(stderr, 1105 "Partition %d overlaps " 1106 "partition %d.", i, j); 1107 } 1108 return (VT_EINVAL); 1109 } 1110 } 1111 } 1112 } 1113 /* just a warning for now */ 1114 if ((resv_part == -1) && efi_debug) { 1115 (void) fprintf(stderr, 1116 "no reserved partition found\n"); 1117 } 1118 return (0); 1119 } 1120 1121 static int 1122 call_blkpg_ioctl(int fd, int command, diskaddr_t start, 1123 diskaddr_t size, uint_t pno) 1124 { 1125 struct blkpg_ioctl_arg ioctl_arg; 1126 struct blkpg_partition linux_part; 1127 memset(&linux_part, 0, sizeof (linux_part)); 1128 1129 char *path = efi_get_devname(fd); 1130 if (path == NULL) { 1131 (void) fprintf(stderr, "failed to retrieve device name\n"); 1132 return (VT_EINVAL); 1133 } 1134 1135 linux_part.start = start; 1136 linux_part.length = size; 1137 linux_part.pno = pno; 1138 snprintf(linux_part.devname, BLKPG_DEVNAMELTH - 1, "%s%u", path, pno); 1139 linux_part.devname[BLKPG_DEVNAMELTH - 1] = '\0'; 1140 free(path); 1141 1142 ioctl_arg.op = command; 1143 ioctl_arg.flags = 0; 1144 ioctl_arg.datalen = sizeof (struct blkpg_partition); 1145 ioctl_arg.data = &linux_part; 1146 1147 return (ioctl(fd, BLKPG, &ioctl_arg)); 1148 } 1149 1150 /* 1151 * add all the unallocated space to the current label 1152 */ 1153 int 1154 efi_use_whole_disk(int fd) 1155 { 1156 struct dk_gpt *efi_label = NULL; 1157 int rval; 1158 int i; 1159 uint_t resv_index = 0, data_index = 0; 1160 diskaddr_t resv_start = 0, data_start = 0; 1161 diskaddr_t data_size, limit, difference; 1162 boolean_t sync_needed = B_FALSE; 1163 uint_t nblocks; 1164 1165 rval = efi_alloc_and_read(fd, &efi_label); 1166 if (rval < 0) { 1167 if (efi_label != NULL) 1168 efi_free(efi_label); 1169 return (rval); 1170 } 1171 1172 /* 1173 * Find the last physically non-zero partition. 1174 * This should be the reserved partition. 1175 */ 1176 for (i = 0; i < efi_label->efi_nparts; i ++) { 1177 if (resv_start < efi_label->efi_parts[i].p_start) { 1178 resv_start = efi_label->efi_parts[i].p_start; 1179 resv_index = i; 1180 } 1181 } 1182 1183 /* 1184 * Find the last physically non-zero partition before that. 1185 * This is the data partition. 1186 */ 1187 for (i = 0; i < resv_index; i ++) { 1188 if (data_start < efi_label->efi_parts[i].p_start) { 1189 data_start = efi_label->efi_parts[i].p_start; 1190 data_index = i; 1191 } 1192 } 1193 data_size = efi_label->efi_parts[data_index].p_size; 1194 1195 /* 1196 * See the "efi_alloc_and_init" function for more information 1197 * about where this "nblocks" value comes from. 1198 */ 1199 nblocks = efi_label->efi_first_u_lba - 1; 1200 1201 /* 1202 * Determine if the EFI label is out of sync. We check that: 1203 * 1204 * 1. the data partition ends at the limit we set, and 1205 * 2. the reserved partition starts at the limit we set. 1206 * 1207 * If either of these conditions is not met, then we need to 1208 * resync the EFI label. 1209 * 1210 * The limit is the last usable LBA, determined by the last LBA 1211 * and the first usable LBA fields on the EFI label of the disk 1212 * (see the lines directly above). Additionally, we factor in 1213 * EFI_MIN_RESV_SIZE (per its use in "zpool_label_disk") and 1214 * P2ALIGN it to ensure the partition boundaries are aligned 1215 * (for performance reasons). The alignment should match the 1216 * alignment used by the "zpool_label_disk" function. 1217 */ 1218 limit = P2ALIGN(efi_label->efi_last_lba - nblocks - EFI_MIN_RESV_SIZE, 1219 PARTITION_END_ALIGNMENT); 1220 if (data_start + data_size != limit || resv_start != limit) 1221 sync_needed = B_TRUE; 1222 1223 if (efi_debug && sync_needed) 1224 (void) fprintf(stderr, "efi_use_whole_disk: sync needed\n"); 1225 1226 /* 1227 * If alter_lba is 1, we are using the backup label. 1228 * Since we can locate the backup label by disk capacity, 1229 * there must be no unallocated space. 1230 */ 1231 if ((efi_label->efi_altern_lba == 1) || (efi_label->efi_altern_lba 1232 >= efi_label->efi_last_lba && !sync_needed)) { 1233 if (efi_debug) { 1234 (void) fprintf(stderr, 1235 "efi_use_whole_disk: requested space not found\n"); 1236 } 1237 efi_free(efi_label); 1238 return (VT_ENOSPC); 1239 } 1240 1241 /* 1242 * Verify that we've found the reserved partition by checking 1243 * that it looks the way it did when we created it in zpool_label_disk. 1244 * If we've found the incorrect partition, then we know that this 1245 * device was reformatted and no longer is solely used by ZFS. 1246 */ 1247 if ((efi_label->efi_parts[resv_index].p_size != EFI_MIN_RESV_SIZE) || 1248 (efi_label->efi_parts[resv_index].p_tag != V_RESERVED) || 1249 (resv_index != 8)) { 1250 if (efi_debug) { 1251 (void) fprintf(stderr, 1252 "efi_use_whole_disk: wholedisk not available\n"); 1253 } 1254 efi_free(efi_label); 1255 return (VT_ENOSPC); 1256 } 1257 1258 if (data_start + data_size != resv_start) { 1259 if (efi_debug) { 1260 (void) fprintf(stderr, 1261 "efi_use_whole_disk: " 1262 "data_start (%lli) + " 1263 "data_size (%lli) != " 1264 "resv_start (%lli)\n", 1265 data_start, data_size, resv_start); 1266 } 1267 1268 return (VT_EINVAL); 1269 } 1270 1271 if (limit < resv_start) { 1272 if (efi_debug) { 1273 (void) fprintf(stderr, 1274 "efi_use_whole_disk: " 1275 "limit (%lli) < resv_start (%lli)\n", 1276 limit, resv_start); 1277 } 1278 1279 return (VT_EINVAL); 1280 } 1281 1282 difference = limit - resv_start; 1283 1284 if (efi_debug) 1285 (void) fprintf(stderr, 1286 "efi_use_whole_disk: difference is %lli\n", difference); 1287 1288 /* 1289 * Move the reserved partition. There is currently no data in 1290 * here except fabricated devids (which get generated via 1291 * efi_write()). So there is no need to copy data. 1292 */ 1293 efi_label->efi_parts[data_index].p_size += difference; 1294 efi_label->efi_parts[resv_index].p_start += difference; 1295 efi_label->efi_last_u_lba = efi_label->efi_last_lba - nblocks; 1296 1297 /* 1298 * Rescanning the partition table in the kernel can result 1299 * in the device links to be removed (see comment in vdev_disk_open). 1300 * If BLKPG_RESIZE_PARTITION is available, then we can resize 1301 * the partition table online and avoid having to remove the device 1302 * links used by the pool. This provides a very deterministic 1303 * approach to resizing devices and does not require any 1304 * loops waiting for devices to reappear. 1305 */ 1306 #ifdef BLKPG_RESIZE_PARTITION 1307 /* 1308 * Delete the reserved partition since we're about to expand 1309 * the data partition and it would overlap with the reserved 1310 * partition. 1311 * NOTE: The starting index for the ioctl is 1 while for the 1312 * EFI partitions it's 0. For that reason we have to add one 1313 * whenever we make an ioctl call. 1314 */ 1315 rval = call_blkpg_ioctl(fd, BLKPG_DEL_PARTITION, 0, 0, resv_index + 1); 1316 if (rval != 0) 1317 goto out; 1318 1319 /* 1320 * Expand the data partition 1321 */ 1322 rval = call_blkpg_ioctl(fd, BLKPG_RESIZE_PARTITION, 1323 efi_label->efi_parts[data_index].p_start * efi_label->efi_lbasize, 1324 efi_label->efi_parts[data_index].p_size * efi_label->efi_lbasize, 1325 data_index + 1); 1326 if (rval != 0) { 1327 (void) fprintf(stderr, "Unable to resize data " 1328 "partition: %d\n", rval); 1329 /* 1330 * Since we failed to resize, we need to reset the start 1331 * of the reserve partition and re-create it. 1332 */ 1333 efi_label->efi_parts[resv_index].p_start -= difference; 1334 } 1335 1336 /* 1337 * Re-add the reserved partition. If we've expanded the data partition 1338 * then we'll move the reserve partition to the end of the data 1339 * partition. Otherwise, we'll recreate the partition in its original 1340 * location. Note that we do this as best-effort and ignore any 1341 * errors that may arise here. This will ensure that we finish writing 1342 * the EFI label. 1343 */ 1344 (void) call_blkpg_ioctl(fd, BLKPG_ADD_PARTITION, 1345 efi_label->efi_parts[resv_index].p_start * efi_label->efi_lbasize, 1346 efi_label->efi_parts[resv_index].p_size * efi_label->efi_lbasize, 1347 resv_index + 1); 1348 #endif 1349 1350 /* 1351 * We're now ready to write the EFI label. 1352 */ 1353 if (rval == 0) { 1354 rval = efi_write(fd, efi_label); 1355 if (rval < 0 && efi_debug) { 1356 (void) fprintf(stderr, "efi_use_whole_disk:fail " 1357 "to write label, rval=%d\n", rval); 1358 } 1359 } 1360 1361 out: 1362 efi_free(efi_label); 1363 return (rval); 1364 } 1365 1366 /* 1367 * write EFI label and backup label 1368 */ 1369 int 1370 efi_write(int fd, struct dk_gpt *vtoc) 1371 { 1372 dk_efi_t dk_ioc; 1373 efi_gpt_t *efi; 1374 efi_gpe_t *efi_parts; 1375 int i, j; 1376 struct dk_cinfo dki_info; 1377 int rval; 1378 int md_flag = 0; 1379 int nblocks; 1380 diskaddr_t lba_backup_gpt_hdr; 1381 1382 if ((rval = efi_get_info(fd, &dki_info)) != 0) 1383 return (rval); 1384 1385 /* check if we are dealing with a metadevice */ 1386 if ((strncmp(dki_info.dki_cname, "pseudo", 7) == 0) && 1387 (strncmp(dki_info.dki_dname, "md", 3) == 0)) { 1388 md_flag = 1; 1389 } 1390 1391 if (check_input(vtoc)) { 1392 /* 1393 * not valid; if it's a metadevice just pass it down 1394 * because SVM will do its own checking 1395 */ 1396 if (md_flag == 0) { 1397 return (VT_EINVAL); 1398 } 1399 } 1400 1401 dk_ioc.dki_lba = 1; 1402 if (NBLOCKS(vtoc->efi_nparts, vtoc->efi_lbasize) < 34) { 1403 dk_ioc.dki_length = EFI_MIN_ARRAY_SIZE + vtoc->efi_lbasize; 1404 } else { 1405 dk_ioc.dki_length = NBLOCKS(vtoc->efi_nparts, 1406 vtoc->efi_lbasize) * 1407 vtoc->efi_lbasize; 1408 } 1409 1410 /* 1411 * the number of blocks occupied by GUID partition entry array 1412 */ 1413 nblocks = dk_ioc.dki_length / vtoc->efi_lbasize - 1; 1414 1415 /* 1416 * Backup GPT header is located on the block after GUID 1417 * partition entry array. Here, we calculate the address 1418 * for backup GPT header. 1419 */ 1420 lba_backup_gpt_hdr = vtoc->efi_last_u_lba + 1 + nblocks; 1421 if (posix_memalign((void **)&dk_ioc.dki_data, 1422 vtoc->efi_lbasize, dk_ioc.dki_length)) 1423 return (VT_ERROR); 1424 1425 memset(dk_ioc.dki_data, 0, dk_ioc.dki_length); 1426 efi = dk_ioc.dki_data; 1427 1428 /* stuff user's input into EFI struct */ 1429 efi->efi_gpt_Signature = LE_64(EFI_SIGNATURE); 1430 efi->efi_gpt_Revision = LE_32(vtoc->efi_version); /* 0x02000100 */ 1431 efi->efi_gpt_HeaderSize = LE_32(sizeof (struct efi_gpt) - LEN_EFI_PAD); 1432 efi->efi_gpt_Reserved1 = 0; 1433 efi->efi_gpt_MyLBA = LE_64(1ULL); 1434 efi->efi_gpt_AlternateLBA = LE_64(lba_backup_gpt_hdr); 1435 efi->efi_gpt_FirstUsableLBA = LE_64(vtoc->efi_first_u_lba); 1436 efi->efi_gpt_LastUsableLBA = LE_64(vtoc->efi_last_u_lba); 1437 efi->efi_gpt_PartitionEntryLBA = LE_64(2ULL); 1438 efi->efi_gpt_NumberOfPartitionEntries = LE_32(vtoc->efi_nparts); 1439 efi->efi_gpt_SizeOfPartitionEntry = LE_32(sizeof (struct efi_gpe)); 1440 UUID_LE_CONVERT(efi->efi_gpt_DiskGUID, vtoc->efi_disk_uguid); 1441 1442 /* LINTED -- always longlong aligned */ 1443 efi_parts = (efi_gpe_t *)((char *)dk_ioc.dki_data + vtoc->efi_lbasize); 1444 1445 for (i = 0; i < vtoc->efi_nparts; i++) { 1446 for (j = 0; 1447 j < sizeof (conversion_array) / 1448 sizeof (struct uuid_to_ptag); j++) { 1449 1450 if (vtoc->efi_parts[i].p_tag == j) { 1451 UUID_LE_CONVERT( 1452 efi_parts[i].efi_gpe_PartitionTypeGUID, 1453 conversion_array[j].uuid); 1454 break; 1455 } 1456 } 1457 1458 if (j == sizeof (conversion_array) / 1459 sizeof (struct uuid_to_ptag)) { 1460 /* 1461 * If we didn't have a matching uuid match, bail here. 1462 * Don't write a label with unknown uuid. 1463 */ 1464 if (efi_debug) { 1465 (void) fprintf(stderr, 1466 "Unknown uuid for p_tag %d\n", 1467 vtoc->efi_parts[i].p_tag); 1468 } 1469 return (VT_EINVAL); 1470 } 1471 1472 /* Zero's should be written for empty partitions */ 1473 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED) 1474 continue; 1475 1476 efi_parts[i].efi_gpe_StartingLBA = 1477 LE_64(vtoc->efi_parts[i].p_start); 1478 efi_parts[i].efi_gpe_EndingLBA = 1479 LE_64(vtoc->efi_parts[i].p_start + 1480 vtoc->efi_parts[i].p_size - 1); 1481 efi_parts[i].efi_gpe_Attributes.PartitionAttrs = 1482 LE_16(vtoc->efi_parts[i].p_flag); 1483 for (j = 0; j < EFI_PART_NAME_LEN; j++) { 1484 efi_parts[i].efi_gpe_PartitionName[j] = 1485 LE_16((ushort_t)vtoc->efi_parts[i].p_name[j]); 1486 } 1487 if ((vtoc->efi_parts[i].p_tag != V_UNASSIGNED) && 1488 uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_uguid)) { 1489 (void) uuid_generate((uchar_t *) 1490 &vtoc->efi_parts[i].p_uguid); 1491 } 1492 bcopy(&vtoc->efi_parts[i].p_uguid, 1493 &efi_parts[i].efi_gpe_UniquePartitionGUID, 1494 sizeof (uuid_t)); 1495 } 1496 efi->efi_gpt_PartitionEntryArrayCRC32 = 1497 LE_32(efi_crc32((unsigned char *)efi_parts, 1498 vtoc->efi_nparts * (int)sizeof (struct efi_gpe))); 1499 efi->efi_gpt_HeaderCRC32 = 1500 LE_32(efi_crc32((unsigned char *)efi, 1501 LE_32(efi->efi_gpt_HeaderSize))); 1502 1503 if (efi_ioctl(fd, DKIOCSETEFI, &dk_ioc) == -1) { 1504 free(dk_ioc.dki_data); 1505 switch (errno) { 1506 case EIO: 1507 return (VT_EIO); 1508 case EINVAL: 1509 return (VT_EINVAL); 1510 default: 1511 return (VT_ERROR); 1512 } 1513 } 1514 /* if it's a metadevice we're done */ 1515 if (md_flag) { 1516 free(dk_ioc.dki_data); 1517 return (0); 1518 } 1519 1520 /* write backup partition array */ 1521 dk_ioc.dki_lba = vtoc->efi_last_u_lba + 1; 1522 dk_ioc.dki_length -= vtoc->efi_lbasize; 1523 /* LINTED */ 1524 dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data + 1525 vtoc->efi_lbasize); 1526 1527 if (efi_ioctl(fd, DKIOCSETEFI, &dk_ioc) == -1) { 1528 /* 1529 * we wrote the primary label okay, so don't fail 1530 */ 1531 if (efi_debug) { 1532 (void) fprintf(stderr, 1533 "write of backup partitions to block %llu " 1534 "failed, errno %d\n", 1535 vtoc->efi_last_u_lba + 1, 1536 errno); 1537 } 1538 } 1539 /* 1540 * now swap MyLBA and AlternateLBA fields and write backup 1541 * partition table header 1542 */ 1543 dk_ioc.dki_lba = lba_backup_gpt_hdr; 1544 dk_ioc.dki_length = vtoc->efi_lbasize; 1545 /* LINTED */ 1546 dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data - 1547 vtoc->efi_lbasize); 1548 efi->efi_gpt_AlternateLBA = LE_64(1ULL); 1549 efi->efi_gpt_MyLBA = LE_64(lba_backup_gpt_hdr); 1550 efi->efi_gpt_PartitionEntryLBA = LE_64(vtoc->efi_last_u_lba + 1); 1551 efi->efi_gpt_HeaderCRC32 = 0; 1552 efi->efi_gpt_HeaderCRC32 = 1553 LE_32(efi_crc32((unsigned char *)dk_ioc.dki_data, 1554 LE_32(efi->efi_gpt_HeaderSize))); 1555 1556 if (efi_ioctl(fd, DKIOCSETEFI, &dk_ioc) == -1) { 1557 if (efi_debug) { 1558 (void) fprintf(stderr, 1559 "write of backup header to block %llu failed, " 1560 "errno %d\n", 1561 lba_backup_gpt_hdr, 1562 errno); 1563 } 1564 } 1565 /* write the PMBR */ 1566 (void) write_pmbr(fd, vtoc); 1567 free(dk_ioc.dki_data); 1568 1569 return (0); 1570 } 1571 1572 void 1573 efi_free(struct dk_gpt *ptr) 1574 { 1575 free(ptr); 1576 } 1577 1578 /* 1579 * Input: File descriptor 1580 * Output: 1 if disk has an EFI label, or > 2TB with no VTOC or legacy MBR. 1581 * Otherwise 0. 1582 */ 1583 int 1584 efi_type(int fd) 1585 { 1586 #if 0 1587 struct vtoc vtoc; 1588 struct extvtoc extvtoc; 1589 1590 if (ioctl(fd, DKIOCGEXTVTOC, &extvtoc) == -1) { 1591 if (errno == ENOTSUP) 1592 return (1); 1593 else if (errno == ENOTTY) { 1594 if (ioctl(fd, DKIOCGVTOC, &vtoc) == -1) 1595 if (errno == ENOTSUP) 1596 return (1); 1597 } 1598 } 1599 return (0); 1600 #else 1601 return (ENOSYS); 1602 #endif 1603 } 1604 1605 void 1606 efi_err_check(struct dk_gpt *vtoc) 1607 { 1608 int resv_part = -1; 1609 int i, j; 1610 diskaddr_t istart, jstart, isize, jsize, endsect; 1611 int overlap = 0; 1612 1613 /* 1614 * make sure no partitions overlap 1615 */ 1616 for (i = 0; i < vtoc->efi_nparts; i++) { 1617 /* It can't be unassigned and have an actual size */ 1618 if ((vtoc->efi_parts[i].p_tag == V_UNASSIGNED) && 1619 (vtoc->efi_parts[i].p_size != 0)) { 1620 (void) fprintf(stderr, 1621 "partition %d is \"unassigned\" but has a size " 1622 "of %llu\n", i, vtoc->efi_parts[i].p_size); 1623 } 1624 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED) { 1625 continue; 1626 } 1627 if (vtoc->efi_parts[i].p_tag == V_RESERVED) { 1628 if (resv_part != -1) { 1629 (void) fprintf(stderr, 1630 "found duplicate reserved partition at " 1631 "%d\n", i); 1632 } 1633 resv_part = i; 1634 if (vtoc->efi_parts[i].p_size != EFI_MIN_RESV_SIZE) 1635 (void) fprintf(stderr, 1636 "Warning: reserved partition size must " 1637 "be %d sectors\n", EFI_MIN_RESV_SIZE); 1638 } 1639 if ((vtoc->efi_parts[i].p_start < vtoc->efi_first_u_lba) || 1640 (vtoc->efi_parts[i].p_start > vtoc->efi_last_u_lba)) { 1641 (void) fprintf(stderr, 1642 "Partition %d starts at %llu\n", 1643 i, 1644 vtoc->efi_parts[i].p_start); 1645 (void) fprintf(stderr, 1646 "It must be between %llu and %llu.\n", 1647 vtoc->efi_first_u_lba, 1648 vtoc->efi_last_u_lba); 1649 } 1650 if ((vtoc->efi_parts[i].p_start + 1651 vtoc->efi_parts[i].p_size < 1652 vtoc->efi_first_u_lba) || 1653 (vtoc->efi_parts[i].p_start + 1654 vtoc->efi_parts[i].p_size > 1655 vtoc->efi_last_u_lba + 1)) { 1656 (void) fprintf(stderr, 1657 "Partition %d ends at %llu\n", 1658 i, 1659 vtoc->efi_parts[i].p_start + 1660 vtoc->efi_parts[i].p_size); 1661 (void) fprintf(stderr, 1662 "It must be between %llu and %llu.\n", 1663 vtoc->efi_first_u_lba, 1664 vtoc->efi_last_u_lba); 1665 } 1666 1667 for (j = 0; j < vtoc->efi_nparts; j++) { 1668 isize = vtoc->efi_parts[i].p_size; 1669 jsize = vtoc->efi_parts[j].p_size; 1670 istart = vtoc->efi_parts[i].p_start; 1671 jstart = vtoc->efi_parts[j].p_start; 1672 if ((i != j) && (isize != 0) && (jsize != 0)) { 1673 endsect = jstart + jsize -1; 1674 if ((jstart <= istart) && 1675 (istart <= endsect)) { 1676 if (!overlap) { 1677 (void) fprintf(stderr, 1678 "label error: EFI Labels do not " 1679 "support overlapping partitions\n"); 1680 } 1681 (void) fprintf(stderr, 1682 "Partition %d overlaps partition " 1683 "%d.\n", i, j); 1684 overlap = 1; 1685 } 1686 } 1687 } 1688 } 1689 /* make sure there is a reserved partition */ 1690 if (resv_part == -1) { 1691 (void) fprintf(stderr, 1692 "no reserved partition found\n"); 1693 } 1694 } 1695 1696 /* 1697 * We need to get information necessary to construct a *new* efi 1698 * label type 1699 */ 1700 int 1701 efi_auto_sense(int fd, struct dk_gpt **vtoc) 1702 { 1703 1704 int i; 1705 1706 /* 1707 * Now build the default partition table 1708 */ 1709 if (efi_alloc_and_init(fd, EFI_NUMPAR, vtoc) != 0) { 1710 if (efi_debug) { 1711 (void) fprintf(stderr, "efi_alloc_and_init failed.\n"); 1712 } 1713 return (-1); 1714 } 1715 1716 for (i = 0; i < MIN((*vtoc)->efi_nparts, V_NUMPAR); i++) { 1717 (*vtoc)->efi_parts[i].p_tag = default_vtoc_map[i].p_tag; 1718 (*vtoc)->efi_parts[i].p_flag = default_vtoc_map[i].p_flag; 1719 (*vtoc)->efi_parts[i].p_start = 0; 1720 (*vtoc)->efi_parts[i].p_size = 0; 1721 } 1722 /* 1723 * Make constants first 1724 * and variable partitions later 1725 */ 1726 1727 /* root partition - s0 128 MB */ 1728 (*vtoc)->efi_parts[0].p_start = 34; 1729 (*vtoc)->efi_parts[0].p_size = 262144; 1730 1731 /* partition - s1 128 MB */ 1732 (*vtoc)->efi_parts[1].p_start = 262178; 1733 (*vtoc)->efi_parts[1].p_size = 262144; 1734 1735 /* partition -s2 is NOT the Backup disk */ 1736 (*vtoc)->efi_parts[2].p_tag = V_UNASSIGNED; 1737 1738 /* partition -s6 /usr partition - HOG */ 1739 (*vtoc)->efi_parts[6].p_start = 524322; 1740 (*vtoc)->efi_parts[6].p_size = (*vtoc)->efi_last_u_lba - 524322 1741 - (1024 * 16); 1742 1743 /* efi reserved partition - s9 16K */ 1744 (*vtoc)->efi_parts[8].p_start = (*vtoc)->efi_last_u_lba - (1024 * 16); 1745 (*vtoc)->efi_parts[8].p_size = (1024 * 16); 1746 (*vtoc)->efi_parts[8].p_tag = V_RESERVED; 1747 return (0); 1748 } 1749