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