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.
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.
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]
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
31 #include <uuid/uuid.h>
34 #include <sys/types.h>
38 #include <sys/param.h>
39 #include <sys/dktp/fdisk.h>
40 #include <sys/efi_partition.h>
41 #include <sys/byteorder.h>
42 #if defined(__linux__)
46 static struct uuid_to_ptag {
48 } conversion_array[] = {
55 { EFI_UNUSED }, /* STAND is never used */
59 { EFI_UNUSED }, /* CACHE (cachefs) is never used */
76 * Default vtoc information for non-SVr4 partitions
78 struct dk_map2 default_vtoc_map[NDKMAP] = {
79 { V_ROOT, 0 }, /* a - 0 */
80 { V_SWAP, V_UNMNT }, /* b - 1 */
81 { V_BACKUP, V_UNMNT }, /* c - 2 */
82 { V_UNASSIGNED, 0 }, /* d - 3 */
83 { V_UNASSIGNED, 0 }, /* e - 4 */
84 { V_UNASSIGNED, 0 }, /* f - 5 */
85 { V_USR, 0 }, /* g - 6 */
86 { V_UNASSIGNED, 0 }, /* h - 7 */
88 #if defined(_SUNOS_VTOC_16)
90 #if defined(i386) || defined(__amd64)
91 { V_BOOT, V_UNMNT }, /* i - 8 */
92 { V_ALTSCTR, 0 }, /* j - 9 */
95 #error No VTOC format defined.
96 #endif /* defined(i386) */
98 { V_UNASSIGNED, 0 }, /* k - 10 */
99 { V_UNASSIGNED, 0 }, /* l - 11 */
100 { V_UNASSIGNED, 0 }, /* m - 12 */
101 { V_UNASSIGNED, 0 }, /* n - 13 */
102 { V_UNASSIGNED, 0 }, /* o - 14 */
103 { V_UNASSIGNED, 0 }, /* p - 15 */
104 #endif /* defined(_SUNOS_VTOC_16) */
113 static int efi_read(int, struct dk_gpt *);
116 * Return a 32-bit CRC of the contents of the buffer. Pre-and-post
117 * one's conditioning will be handled by crc32() internally.
120 efi_crc32(const unsigned char *buf, unsigned int size)
122 uint32_t crc = crc32(0, Z_NULL, 0);
124 crc = crc32(crc, buf, size);
130 read_disk_info(int fd, diskaddr_t *capacity, uint_t *lbsize)
133 unsigned long long capacity_size;
135 if (ioctl(fd, BLKSSZGET, §or_size) < 0)
138 if (ioctl(fd, BLKGETSIZE64, &capacity_size) < 0)
141 *lbsize = (uint_t)sector_size;
142 *capacity = (diskaddr_t)(capacity_size / sector_size);
148 efi_get_info(int fd, struct dk_cinfo *dki_info)
150 #if defined(__linux__)
155 memset(dki_info, 0, sizeof(*dki_info));
157 path = calloc(PATH_MAX, 1);
162 * The simplest way to get the partition number under linux is
163 * to parse it out of the /dev/<disk><parition> block device name.
164 * The kernel creates this using the partition number when it
165 * populates /dev/ so it may be trusted. The tricky bit here is
166 * that the naming convention is based on the block device type.
167 * So we need to take this in to account when parsing out the
168 * partition information. Another issue is that the libefi API
169 * API only provides the open fd and not the file path. To handle
170 * this realpath(3) is used to resolve the block device name from
171 * /proc/self/fd/<fd>. Aside from the partition number we collect
172 * some additional device info.
174 (void) sprintf(path, "/proc/self/fd/%d", fd);
175 dev_path = realpath(path, NULL);
178 if (dev_path == NULL)
181 if ((strncmp(dev_path, "/dev/sd", 7) == 0)) {
182 strcpy(dki_info->dki_cname, "sd");
183 dki_info->dki_ctype = DKC_SCSI_CCS;
184 rval = sscanf(dev_path, "/dev/%[a-zA-Z]%hu",
186 &dki_info->dki_partition);
187 } else if ((strncmp(dev_path, "/dev/hd", 7) == 0)) {
188 strcpy(dki_info->dki_cname, "hd");
189 dki_info->dki_ctype = DKC_DIRECT;
190 rval = sscanf(dev_path, "/dev/%[a-zA-Z]%hu",
192 &dki_info->dki_partition);
193 } else if ((strncmp(dev_path, "/dev/md", 7) == 0)) {
194 strcpy(dki_info->dki_cname, "pseudo");
195 dki_info->dki_ctype = DKC_MD;
196 rval = sscanf(dev_path, "/dev/%[a-zA-Z0-9]p%hu",
198 &dki_info->dki_partition);
199 } else if ((strncmp(dev_path, "/dev/dm-", 8) == 0)) {
200 strcpy(dki_info->dki_cname, "pseudo");
201 dki_info->dki_ctype = DKC_VBD;
202 rval = sscanf(dev_path, "/dev/%[a-zA-Z0-9-]p%hu",
204 &dki_info->dki_partition);
205 } else if ((strncmp(dev_path, "/dev/ram", 8) == 0)) {
206 strcpy(dki_info->dki_cname, "pseudo");
207 dki_info->dki_ctype = DKC_PCMCIA_MEM;
208 rval = sscanf(dev_path, "/dev/%[a-zA-Z0-9]p%hu",
210 &dki_info->dki_partition);
211 } else if ((strncmp(dev_path, "/dev/loop", 9) == 0)) {
212 strcpy(dki_info->dki_cname, "pseudo");
213 dki_info->dki_ctype = DKC_VBD;
214 rval = sscanf(dev_path, "/dev/%[a-zA-Z0-9]p%hu",
216 &dki_info->dki_partition);
218 strcpy(dki_info->dki_dname, "unknown");
219 strcpy(dki_info->dki_cname, "unknown");
220 dki_info->dki_ctype = DKC_UNKNOWN;
228 dki_info->dki_partition = 0;
233 if (ioctl(fd, DKIOCINFO, (caddr_t)dki_info) == -1)
239 (void) fprintf(stderr, "DKIOCINFO errno 0x%x\n", errno);
252 * the number of blocks the EFI label takes up (round up to nearest
255 #define NBLOCKS(p, l) (1 + ((((p) * (int)sizeof (efi_gpe_t)) + \
257 /* number of partitions -- limited by what we can malloc */
258 #define MAX_PARTS ((4294967295UL - sizeof (struct dk_gpt)) / \
259 sizeof (struct dk_part))
262 efi_alloc_and_init(int fd, uint32_t nparts, struct dk_gpt **vtoc)
264 diskaddr_t capacity = 0;
270 struct dk_cinfo dki_info;
272 if (read_disk_info(fd, &capacity, &lbsize) != 0)
275 #if defined(__linux__)
276 if (efi_get_info(fd, &dki_info) != 0)
279 if (dki_info.dki_partition != 0)
282 if ((dki_info.dki_ctype == DKC_PCMCIA_MEM) ||
283 (dki_info.dki_ctype == DKC_VBD) ||
284 (dki_info.dki_ctype == DKC_UNKNOWN))
288 nblocks = NBLOCKS(nparts, lbsize);
289 if ((nblocks * lbsize) < EFI_MIN_ARRAY_SIZE + lbsize) {
290 /* 16K plus one block for the GPT */
291 nblocks = EFI_MIN_ARRAY_SIZE / lbsize + 1;
294 if (nparts > MAX_PARTS) {
296 (void) fprintf(stderr,
297 "the maximum number of partitions supported is %lu\n",
303 length = sizeof (struct dk_gpt) +
304 sizeof (struct dk_part) * (nparts - 1);
306 if ((*vtoc = calloc(length, 1)) == NULL)
311 vptr->efi_version = EFI_VERSION_CURRENT;
312 vptr->efi_lbasize = lbsize;
313 vptr->efi_nparts = nparts;
315 * add one block here for the PMBR; on disks with a 512 byte
316 * block size and 128 or fewer partitions, efi_first_u_lba
317 * should work out to "34"
319 vptr->efi_first_u_lba = nblocks + 1;
320 vptr->efi_last_lba = capacity - 1;
321 vptr->efi_altern_lba = capacity -1;
322 vptr->efi_last_u_lba = vptr->efi_last_lba - nblocks;
324 (void) uuid_generate((uchar_t *)&uuid);
325 UUID_LE_CONVERT(vptr->efi_disk_uguid, uuid);
330 * Read EFI - return partition number upon success.
333 efi_alloc_and_read(int fd, struct dk_gpt **vtoc)
339 /* figure out the number of entries that would fit into 16K */
340 nparts = EFI_MIN_ARRAY_SIZE / sizeof (efi_gpe_t);
341 length = (int) sizeof (struct dk_gpt) +
342 (int) sizeof (struct dk_part) * (nparts - 1);
343 if ((*vtoc = calloc(length, 1)) == NULL)
346 (*vtoc)->efi_nparts = nparts;
347 rval = efi_read(fd, *vtoc);
349 if ((rval == VT_EINVAL) && (*vtoc)->efi_nparts > nparts) {
351 length = (int) sizeof (struct dk_gpt) +
352 (int) sizeof (struct dk_part) *
353 ((*vtoc)->efi_nparts - 1);
354 nparts = (*vtoc)->efi_nparts;
355 if ((tmp = realloc(*vtoc, length)) == NULL) {
361 rval = efi_read(fd, *vtoc);
367 (void) fprintf(stderr,
368 "read of EFI table failed, rval=%d\n", rval);
378 efi_ioctl(int fd, int cmd, dk_efi_t *dk_ioc)
380 void *data = dk_ioc->dki_data;
382 #if defined(__linux__)
387 * When the IO is not being performed in kernel as an ioctl we need
388 * to know the sector size so we can seek to the proper byte offset.
390 if (read_disk_info(fd, &capacity, &lbsize) == -1) {
392 fprintf(stderr,"unable to read disk info: %d",errno);
402 (void) fprintf(stderr, "DKIOCGETEFI assuming "
403 "LBA %d bytes\n", DEV_BSIZE);
408 error = lseek(fd, dk_ioc->dki_lba * lbsize, SEEK_SET);
411 (void) fprintf(stderr, "DKIOCGETEFI lseek "
412 "error: %d\n", errno);
416 error = read(fd, data, dk_ioc->dki_length);
419 (void) fprintf(stderr, "DKIOCGETEFI read "
420 "error: %d\n", errno);
424 if (error != dk_ioc->dki_length) {
426 (void) fprintf(stderr, "DKIOCGETEFI short "
427 "read of %d bytes\n", error);
437 (void) fprintf(stderr, "DKIOCSETEFI unknown "
443 error = lseek(fd, dk_ioc->dki_lba * lbsize, SEEK_SET);
446 (void) fprintf(stderr, "DKIOCSETEFI lseek "
447 "error: %d\n", errno);
451 error = write(fd, data, dk_ioc->dki_length);
454 (void) fprintf(stderr, "DKIOCSETEFI write "
455 "error: %d\n", errno);
459 if (error != dk_ioc->dki_length) {
461 (void) fprintf(stderr, "DKIOCSETEFI short "
462 "write of %d bytes\n", error);
467 /* Sync the new EFI table to disk */
472 /* Ensure any local disk cache is also flushed */
473 if (ioctl(fd, BLKFLSBUF, 0) == -1)
481 (void) fprintf(stderr, "unsupported ioctl()\n");
487 dk_ioc->dki_data_64 = (uint64_t)(uintptr_t)data;
488 error = ioctl(fd, cmd, (void *)dk_ioc);
489 dk_ioc->dki_data = data;
494 #if defined(__linux__)
501 /* Notify the kernel a devices partition table has been updated */
502 while ((error = ioctl(fd, BLKRRPART)) != 0) {
504 (void) fprintf(stderr, "the kernel failed to rescan "
505 "the partition table: %d\n", errno);
515 check_label(int fd, dk_efi_t *dk_ioc)
520 if (efi_ioctl(fd, DKIOCGETEFI, dk_ioc) == -1) {
528 efi = dk_ioc->dki_data;
529 if (efi->efi_gpt_Signature != LE_64(EFI_SIGNATURE)) {
531 (void) fprintf(stderr,
532 "Bad EFI signature: 0x%llx != 0x%llx\n",
533 (long long)efi->efi_gpt_Signature,
534 (long long)LE_64(EFI_SIGNATURE));
539 * check CRC of the header; the size of the header should
540 * never be larger than one block
542 crc = efi->efi_gpt_HeaderCRC32;
543 efi->efi_gpt_HeaderCRC32 = 0;
545 if (((len_t)LE_32(efi->efi_gpt_HeaderSize) > dk_ioc->dki_length) ||
546 crc != LE_32(efi_crc32((unsigned char *)efi,
547 LE_32(efi->efi_gpt_HeaderSize)))) {
549 (void) fprintf(stderr,
550 "Bad EFI CRC: 0x%x != 0x%x\n",
552 LE_32(efi_crc32((unsigned char *)efi,
553 sizeof (struct efi_gpt))));
561 efi_read(int fd, struct dk_gpt *vtoc)
568 diskaddr_t capacity = 0;
570 struct dk_minfo disk_info;
573 efi_gpe_t *efi_parts;
574 struct dk_cinfo dki_info;
575 uint32_t user_length;
576 boolean_t legacy_label = B_FALSE;
579 * get the partition number for this file descriptor.
581 if ((rval = efi_get_info(fd, &dki_info)) != 0)
584 if ((strncmp(dki_info.dki_cname, "pseudo", 7) == 0) &&
585 (strncmp(dki_info.dki_dname, "md", 3) == 0)) {
587 } else if ((strncmp(dki_info.dki_cname, "vdc", 4) == 0) &&
588 (strncmp(dki_info.dki_dname, "vdc", 4) == 0)) {
590 * The controller and drive name "vdc" (virtual disk client)
591 * indicates a LDoms virtual disk.
596 /* get the LBA size */
597 if (read_disk_info(fd, &capacity, &lbsize) == -1) {
599 (void) fprintf(stderr,
600 "unable to read disk info: %d",
606 disk_info.dki_lbsize = lbsize;
607 disk_info.dki_capacity = capacity;
609 if (disk_info.dki_lbsize == 0) {
611 (void) fprintf(stderr,
612 "efi_read: assuming LBA 512 bytes\n");
614 disk_info.dki_lbsize = DEV_BSIZE;
617 * Read the EFI GPT to figure out how many partitions we need
621 if (NBLOCKS(vtoc->efi_nparts, disk_info.dki_lbsize) < 34) {
622 label_len = EFI_MIN_ARRAY_SIZE + disk_info.dki_lbsize;
624 label_len = vtoc->efi_nparts * (int) sizeof (efi_gpe_t) +
625 disk_info.dki_lbsize;
626 if (label_len % disk_info.dki_lbsize) {
627 /* pad to physical sector size */
628 label_len += disk_info.dki_lbsize;
629 label_len &= ~(disk_info.dki_lbsize - 1);
633 if (posix_memalign((void **)&dk_ioc.dki_data,
634 disk_info.dki_lbsize, label_len))
637 memset(dk_ioc.dki_data, 0, label_len);
638 dk_ioc.dki_length = disk_info.dki_lbsize;
639 user_length = vtoc->efi_nparts;
640 efi = dk_ioc.dki_data;
642 dk_ioc.dki_length = label_len;
643 if (efi_ioctl(fd, DKIOCGETEFI, &dk_ioc) == -1) {
651 } else if ((rval = check_label(fd, &dk_ioc)) == VT_EINVAL) {
653 * No valid label here; try the alternate. Note that here
654 * we just read GPT header and save it into dk_ioc.data,
655 * Later, we will read GUID partition entry array if we
656 * can get valid GPT header.
660 * This is a workaround for legacy systems. In the past, the
661 * last sector of SCSI disk was invisible on x86 platform. At
662 * that time, backup label was saved on the next to the last
663 * sector. It is possible for users to move a disk from previous
664 * solaris system to present system. Here, we attempt to search
665 * legacy backup EFI label first.
667 dk_ioc.dki_lba = disk_info.dki_capacity - 2;
668 dk_ioc.dki_length = disk_info.dki_lbsize;
669 rval = check_label(fd, &dk_ioc);
670 if (rval == VT_EINVAL) {
672 * we didn't find legacy backup EFI label, try to
673 * search backup EFI label in the last block.
675 dk_ioc.dki_lba = disk_info.dki_capacity - 1;
676 dk_ioc.dki_length = disk_info.dki_lbsize;
677 rval = check_label(fd, &dk_ioc);
679 legacy_label = B_TRUE;
681 (void) fprintf(stderr,
682 "efi_read: primary label corrupt; "
683 "using EFI backup label located on"
684 " the last block\n");
687 if ((efi_debug) && (rval == 0))
688 (void) fprintf(stderr, "efi_read: primary label"
689 " corrupt; using legacy EFI backup label "
690 " located on the next to last block\n");
694 dk_ioc.dki_lba = LE_64(efi->efi_gpt_PartitionEntryLBA);
695 vtoc->efi_flags |= EFI_GPT_PRIMARY_CORRUPT;
697 LE_32(efi->efi_gpt_NumberOfPartitionEntries);
699 * Partition tables are between backup GPT header
700 * table and ParitionEntryLBA (the starting LBA of
701 * the GUID partition entries array). Now that we
702 * already got valid GPT header and saved it in
703 * dk_ioc.dki_data, we try to get GUID partition
707 dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data
708 + disk_info.dki_lbsize);
710 dk_ioc.dki_length = disk_info.dki_capacity - 1 -
713 dk_ioc.dki_length = disk_info.dki_capacity - 2 -
715 dk_ioc.dki_length *= disk_info.dki_lbsize;
716 if (dk_ioc.dki_length >
717 ((len_t)label_len - sizeof (*dk_ioc.dki_data))) {
721 * read GUID partition entry array
723 rval = efi_ioctl(fd, DKIOCGETEFI, &dk_ioc);
727 } else if (rval == 0) {
729 dk_ioc.dki_lba = LE_64(efi->efi_gpt_PartitionEntryLBA);
731 dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data
732 + disk_info.dki_lbsize);
733 dk_ioc.dki_length = label_len - disk_info.dki_lbsize;
734 rval = efi_ioctl(fd, DKIOCGETEFI, &dk_ioc);
736 } else if (vdc_flag && rval == VT_ERROR && errno == EINVAL) {
738 * When the device is a LDoms virtual disk, the DKIOCGETEFI
739 * ioctl can fail with EINVAL if the virtual disk backend
740 * is a ZFS volume serviced by a domain running an old version
741 * of Solaris. This is because the DKIOCGETEFI ioctl was
742 * initially incorrectly implemented for a ZFS volume and it
743 * expected the GPT and GPE to be retrieved with a single ioctl.
744 * So we try to read the GPT and the GPE using that old style
748 dk_ioc.dki_length = label_len;
749 rval = check_label(fd, &dk_ioc);
757 /* LINTED -- always longlong aligned */
758 efi_parts = (efi_gpe_t *)(((char *)efi) + disk_info.dki_lbsize);
761 * Assemble this into a "dk_gpt" struct for easier
762 * digestibility by applications.
764 vtoc->efi_version = LE_32(efi->efi_gpt_Revision);
765 vtoc->efi_nparts = LE_32(efi->efi_gpt_NumberOfPartitionEntries);
766 vtoc->efi_part_size = LE_32(efi->efi_gpt_SizeOfPartitionEntry);
767 vtoc->efi_lbasize = disk_info.dki_lbsize;
768 vtoc->efi_last_lba = disk_info.dki_capacity - 1;
769 vtoc->efi_first_u_lba = LE_64(efi->efi_gpt_FirstUsableLBA);
770 vtoc->efi_last_u_lba = LE_64(efi->efi_gpt_LastUsableLBA);
771 vtoc->efi_altern_lba = LE_64(efi->efi_gpt_AlternateLBA);
772 UUID_LE_CONVERT(vtoc->efi_disk_uguid, efi->efi_gpt_DiskGUID);
775 * If the array the user passed in is too small, set the length
776 * to what it needs to be and return
778 if (user_length < vtoc->efi_nparts) {
782 for (i = 0; i < vtoc->efi_nparts; i++) {
784 UUID_LE_CONVERT(vtoc->efi_parts[i].p_guid,
785 efi_parts[i].efi_gpe_PartitionTypeGUID);
788 j < sizeof (conversion_array)
789 / sizeof (struct uuid_to_ptag); j++) {
791 if (bcmp(&vtoc->efi_parts[i].p_guid,
792 &conversion_array[j].uuid,
793 sizeof (struct uuid)) == 0) {
794 vtoc->efi_parts[i].p_tag = j;
798 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED)
800 vtoc->efi_parts[i].p_flag =
801 LE_16(efi_parts[i].efi_gpe_Attributes.PartitionAttrs);
802 vtoc->efi_parts[i].p_start =
803 LE_64(efi_parts[i].efi_gpe_StartingLBA);
804 vtoc->efi_parts[i].p_size =
805 LE_64(efi_parts[i].efi_gpe_EndingLBA) -
806 vtoc->efi_parts[i].p_start + 1;
807 for (j = 0; j < EFI_PART_NAME_LEN; j++) {
808 vtoc->efi_parts[i].p_name[j] =
810 efi_parts[i].efi_gpe_PartitionName[j]);
813 UUID_LE_CONVERT(vtoc->efi_parts[i].p_uguid,
814 efi_parts[i].efi_gpe_UniquePartitionGUID);
818 return (dki_info.dki_partition);
821 /* writes a "protective" MBR */
823 write_pmbr(int fd, struct dk_gpt *vtoc)
828 diskaddr_t size_in_lba;
832 len = (vtoc->efi_lbasize == 0) ? sizeof (mb) : vtoc->efi_lbasize;
833 if (posix_memalign((void **)&buf, len, len))
837 * Preserve any boot code and disk signature if the first block is
842 dk_ioc.dki_length = len;
843 /* LINTED -- always longlong aligned */
844 dk_ioc.dki_data = (efi_gpt_t *)buf;
845 if (efi_ioctl(fd, DKIOCGETEFI, &dk_ioc) == -1) {
846 (void *) memcpy(&mb, buf, sizeof (mb));
847 bzero(&mb, sizeof (mb));
848 mb.signature = LE_16(MBB_MAGIC);
850 (void *) memcpy(&mb, buf, sizeof (mb));
851 if (mb.signature != LE_16(MBB_MAGIC)) {
852 bzero(&mb, sizeof (mb));
853 mb.signature = LE_16(MBB_MAGIC);
857 bzero(&mb.parts, sizeof (mb.parts));
858 cp = (uchar_t *)&mb.parts[0];
859 /* bootable or not */
861 /* beginning CHS; 0xffffff if not representable */
867 /* ending CHS; 0xffffff if not representable */
871 /* starting LBA: 1 (little endian format) by EFI definition */
876 /* ending LBA: last block on the disk (little endian format) */
877 size_in_lba = vtoc->efi_last_lba;
878 if (size_in_lba < 0xffffffff) {
879 *cp++ = (size_in_lba & 0x000000ff);
880 *cp++ = (size_in_lba & 0x0000ff00) >> 8;
881 *cp++ = (size_in_lba & 0x00ff0000) >> 16;
882 *cp++ = (size_in_lba & 0xff000000) >> 24;
890 (void *) memcpy(buf, &mb, sizeof (mb));
891 /* LINTED -- always longlong aligned */
892 dk_ioc.dki_data = (efi_gpt_t *)buf;
894 dk_ioc.dki_length = len;
895 if (efi_ioctl(fd, DKIOCSETEFI, &dk_ioc) == -1) {
910 /* make sure the user specified something reasonable */
912 check_input(struct dk_gpt *vtoc)
916 diskaddr_t istart, jstart, isize, jsize, endsect;
919 * Sanity-check the input (make sure no partitions overlap)
921 for (i = 0; i < vtoc->efi_nparts; i++) {
922 /* It can't be unassigned and have an actual size */
923 if ((vtoc->efi_parts[i].p_tag == V_UNASSIGNED) &&
924 (vtoc->efi_parts[i].p_size != 0)) {
926 (void) fprintf(stderr, "partition %d is "
927 "\"unassigned\" but has a size of %llu",
928 i, vtoc->efi_parts[i].p_size);
932 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED) {
933 if (uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid))
935 /* we have encountered an unknown uuid */
936 vtoc->efi_parts[i].p_tag = 0xff;
938 if (vtoc->efi_parts[i].p_tag == V_RESERVED) {
939 if (resv_part != -1) {
941 (void) fprintf(stderr, "found "
942 "duplicate reserved partition "
949 if ((vtoc->efi_parts[i].p_start < vtoc->efi_first_u_lba) ||
950 (vtoc->efi_parts[i].p_start > vtoc->efi_last_u_lba)) {
952 (void) fprintf(stderr,
953 "Partition %d starts at %llu. ",
955 vtoc->efi_parts[i].p_start);
956 (void) fprintf(stderr,
957 "It must be between %llu and %llu.\n",
958 vtoc->efi_first_u_lba,
959 vtoc->efi_last_u_lba);
963 if ((vtoc->efi_parts[i].p_start +
964 vtoc->efi_parts[i].p_size <
965 vtoc->efi_first_u_lba) ||
966 (vtoc->efi_parts[i].p_start +
967 vtoc->efi_parts[i].p_size >
968 vtoc->efi_last_u_lba + 1)) {
970 (void) fprintf(stderr,
971 "Partition %d ends at %llu. ",
973 vtoc->efi_parts[i].p_start +
974 vtoc->efi_parts[i].p_size);
975 (void) fprintf(stderr,
976 "It must be between %llu and %llu.\n",
977 vtoc->efi_first_u_lba,
978 vtoc->efi_last_u_lba);
983 for (j = 0; j < vtoc->efi_nparts; j++) {
984 isize = vtoc->efi_parts[i].p_size;
985 jsize = vtoc->efi_parts[j].p_size;
986 istart = vtoc->efi_parts[i].p_start;
987 jstart = vtoc->efi_parts[j].p_start;
988 if ((i != j) && (isize != 0) && (jsize != 0)) {
989 endsect = jstart + jsize -1;
990 if ((jstart <= istart) &&
991 (istart <= endsect)) {
993 (void) fprintf(stderr,
994 "Partition %d overlaps "
995 "partition %d.", i, j);
1002 /* just a warning for now */
1003 if ((resv_part == -1) && efi_debug) {
1004 (void) fprintf(stderr,
1005 "no reserved partition found\n");
1011 * add all the unallocated space to the current label
1014 efi_use_whole_disk(int fd)
1016 struct dk_gpt *efi_label;
1019 uint_t phy_last_slice = 0;
1020 diskaddr_t pl_start = 0;
1023 rval = efi_alloc_and_read(fd, &efi_label);
1028 /* find the last physically non-zero partition */
1029 for (i = 0; i < efi_label->efi_nparts - 2; i ++) {
1030 if (pl_start < efi_label->efi_parts[i].p_start) {
1031 pl_start = efi_label->efi_parts[i].p_start;
1035 pl_size = efi_label->efi_parts[phy_last_slice].p_size;
1038 * If alter_lba is 1, we are using the backup label.
1039 * Since we can locate the backup label by disk capacity,
1040 * there must be no unallocated space.
1042 if ((efi_label->efi_altern_lba == 1) || (efi_label->efi_altern_lba
1043 >= efi_label->efi_last_lba)) {
1045 (void) fprintf(stderr,
1046 "efi_use_whole_disk: requested space not found\n");
1048 efi_free(efi_label);
1053 * If there is space between the last physically non-zero partition
1054 * and the reserved partition, just add the unallocated space to this
1055 * area. Otherwise, the unallocated space is added to the last
1056 * physically non-zero partition.
1058 if (pl_start + pl_size - 1 == efi_label->efi_last_u_lba -
1059 EFI_MIN_RESV_SIZE) {
1060 efi_label->efi_parts[phy_last_slice].p_size +=
1061 efi_label->efi_last_lba - efi_label->efi_altern_lba;
1065 * Move the reserved partition. There is currently no data in
1066 * here except fabricated devids (which get generated via
1067 * efi_write()). So there is no need to copy data.
1069 efi_label->efi_parts[efi_label->efi_nparts - 1].p_start +=
1070 efi_label->efi_last_lba - efi_label->efi_altern_lba;
1071 efi_label->efi_last_u_lba += efi_label->efi_last_lba
1072 - efi_label->efi_altern_lba;
1074 rval = efi_write(fd, efi_label);
1077 (void) fprintf(stderr,
1078 "efi_use_whole_disk:fail to write label, rval=%d\n",
1081 efi_free(efi_label);
1085 efi_free(efi_label);
1091 * write EFI label and backup label
1094 efi_write(int fd, struct dk_gpt *vtoc)
1098 efi_gpe_t *efi_parts;
1100 struct dk_cinfo dki_info;
1104 diskaddr_t lba_backup_gpt_hdr;
1106 if ((rval = efi_get_info(fd, &dki_info)) != 0)
1109 /* check if we are dealing wih a metadevice */
1110 if ((strncmp(dki_info.dki_cname, "pseudo", 7) == 0) &&
1111 (strncmp(dki_info.dki_dname, "md", 3) == 0)) {
1115 if (check_input(vtoc)) {
1117 * not valid; if it's a metadevice just pass it down
1118 * because SVM will do its own checking
1126 if (NBLOCKS(vtoc->efi_nparts, vtoc->efi_lbasize) < 34) {
1127 dk_ioc.dki_length = EFI_MIN_ARRAY_SIZE + vtoc->efi_lbasize;
1129 dk_ioc.dki_length = NBLOCKS(vtoc->efi_nparts,
1130 vtoc->efi_lbasize) *
1135 * the number of blocks occupied by GUID partition entry array
1137 nblocks = dk_ioc.dki_length / vtoc->efi_lbasize - 1;
1140 * Backup GPT header is located on the block after GUID
1141 * partition entry array. Here, we calculate the address
1142 * for backup GPT header.
1144 lba_backup_gpt_hdr = vtoc->efi_last_u_lba + 1 + nblocks;
1145 if (posix_memalign((void **)&dk_ioc.dki_data,
1146 vtoc->efi_lbasize, dk_ioc.dki_length))
1149 memset(dk_ioc.dki_data, 0, dk_ioc.dki_length);
1150 efi = dk_ioc.dki_data;
1152 /* stuff user's input into EFI struct */
1153 efi->efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1154 efi->efi_gpt_Revision = LE_32(vtoc->efi_version); /* 0x02000100 */
1155 efi->efi_gpt_HeaderSize = LE_32(sizeof (struct efi_gpt));
1156 efi->efi_gpt_Reserved1 = 0;
1157 efi->efi_gpt_MyLBA = LE_64(1ULL);
1158 efi->efi_gpt_AlternateLBA = LE_64(lba_backup_gpt_hdr);
1159 efi->efi_gpt_FirstUsableLBA = LE_64(vtoc->efi_first_u_lba);
1160 efi->efi_gpt_LastUsableLBA = LE_64(vtoc->efi_last_u_lba);
1161 efi->efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1162 efi->efi_gpt_NumberOfPartitionEntries = LE_32(vtoc->efi_nparts);
1163 efi->efi_gpt_SizeOfPartitionEntry = LE_32(sizeof (struct efi_gpe));
1164 UUID_LE_CONVERT(efi->efi_gpt_DiskGUID, vtoc->efi_disk_uguid);
1166 /* LINTED -- always longlong aligned */
1167 efi_parts = (efi_gpe_t *)((char *)dk_ioc.dki_data + vtoc->efi_lbasize);
1169 for (i = 0; i < vtoc->efi_nparts; i++) {
1171 j < sizeof (conversion_array) /
1172 sizeof (struct uuid_to_ptag); j++) {
1174 if (vtoc->efi_parts[i].p_tag == j) {
1176 efi_parts[i].efi_gpe_PartitionTypeGUID,
1177 conversion_array[j].uuid);
1182 if (j == sizeof (conversion_array) /
1183 sizeof (struct uuid_to_ptag)) {
1185 * If we didn't have a matching uuid match, bail here.
1186 * Don't write a label with unknown uuid.
1189 (void) fprintf(stderr,
1190 "Unknown uuid for p_tag %d\n",
1191 vtoc->efi_parts[i].p_tag);
1196 /* Zero's should be written for empty partitions */
1197 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED)
1200 efi_parts[i].efi_gpe_StartingLBA =
1201 LE_64(vtoc->efi_parts[i].p_start);
1202 efi_parts[i].efi_gpe_EndingLBA =
1203 LE_64(vtoc->efi_parts[i].p_start +
1204 vtoc->efi_parts[i].p_size - 1);
1205 efi_parts[i].efi_gpe_Attributes.PartitionAttrs =
1206 LE_16(vtoc->efi_parts[i].p_flag);
1207 for (j = 0; j < EFI_PART_NAME_LEN; j++) {
1208 efi_parts[i].efi_gpe_PartitionName[j] =
1209 LE_16((ushort_t)vtoc->efi_parts[i].p_name[j]);
1211 if ((vtoc->efi_parts[i].p_tag != V_UNASSIGNED) &&
1212 uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_uguid)) {
1213 (void) uuid_generate((uchar_t *)
1214 &vtoc->efi_parts[i].p_uguid);
1216 bcopy(&vtoc->efi_parts[i].p_uguid,
1217 &efi_parts[i].efi_gpe_UniquePartitionGUID,
1220 efi->efi_gpt_PartitionEntryArrayCRC32 =
1221 LE_32(efi_crc32((unsigned char *)efi_parts,
1222 vtoc->efi_nparts * (int)sizeof (struct efi_gpe)));
1223 efi->efi_gpt_HeaderCRC32 =
1224 LE_32(efi_crc32((unsigned char *)efi, sizeof (struct efi_gpt)));
1226 if (efi_ioctl(fd, DKIOCSETEFI, &dk_ioc) == -1) {
1227 free(dk_ioc.dki_data);
1237 /* if it's a metadevice we're done */
1239 free(dk_ioc.dki_data);
1243 /* write backup partition array */
1244 dk_ioc.dki_lba = vtoc->efi_last_u_lba + 1;
1245 dk_ioc.dki_length -= vtoc->efi_lbasize;
1247 dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data +
1250 if (efi_ioctl(fd, DKIOCSETEFI, &dk_ioc) == -1) {
1252 * we wrote the primary label okay, so don't fail
1255 (void) fprintf(stderr,
1256 "write of backup partitions to block %llu "
1257 "failed, errno %d\n",
1258 vtoc->efi_last_u_lba + 1,
1263 * now swap MyLBA and AlternateLBA fields and write backup
1264 * partition table header
1266 dk_ioc.dki_lba = lba_backup_gpt_hdr;
1267 dk_ioc.dki_length = vtoc->efi_lbasize;
1269 dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data -
1271 efi->efi_gpt_AlternateLBA = LE_64(1ULL);
1272 efi->efi_gpt_MyLBA = LE_64(lba_backup_gpt_hdr);
1273 efi->efi_gpt_PartitionEntryLBA = LE_64(vtoc->efi_last_u_lba + 1);
1274 efi->efi_gpt_HeaderCRC32 = 0;
1275 efi->efi_gpt_HeaderCRC32 =
1276 LE_32(efi_crc32((unsigned char *)dk_ioc.dki_data,
1277 sizeof (struct efi_gpt)));
1279 if (efi_ioctl(fd, DKIOCSETEFI, &dk_ioc) == -1) {
1281 (void) fprintf(stderr,
1282 "write of backup header to block %llu failed, "
1288 /* write the PMBR */
1289 (void) write_pmbr(fd, vtoc);
1290 free(dk_ioc.dki_data);
1292 #if defined(__linux__)
1293 rval = efi_rescan(fd);
1302 efi_free(struct dk_gpt *ptr)
1308 * Input: File descriptor
1309 * Output: 1 if disk has an EFI label, or > 2TB with no VTOC or legacy MBR.
1317 struct extvtoc extvtoc;
1319 if (ioctl(fd, DKIOCGEXTVTOC, &extvtoc) == -1) {
1320 if (errno == ENOTSUP)
1322 else if (errno == ENOTTY) {
1323 if (ioctl(fd, DKIOCGVTOC, &vtoc) == -1)
1324 if (errno == ENOTSUP)
1335 efi_err_check(struct dk_gpt *vtoc)
1339 diskaddr_t istart, jstart, isize, jsize, endsect;
1343 * make sure no partitions overlap
1345 for (i = 0; i < vtoc->efi_nparts; i++) {
1346 /* It can't be unassigned and have an actual size */
1347 if ((vtoc->efi_parts[i].p_tag == V_UNASSIGNED) &&
1348 (vtoc->efi_parts[i].p_size != 0)) {
1349 (void) fprintf(stderr,
1350 "partition %d is \"unassigned\" but has a size "
1351 "of %llu\n", i, vtoc->efi_parts[i].p_size);
1353 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED) {
1356 if (vtoc->efi_parts[i].p_tag == V_RESERVED) {
1357 if (resv_part != -1) {
1358 (void) fprintf(stderr,
1359 "found duplicate reserved partition at "
1363 if (vtoc->efi_parts[i].p_size != EFI_MIN_RESV_SIZE)
1364 (void) fprintf(stderr,
1365 "Warning: reserved partition size must "
1366 "be %d sectors\n", EFI_MIN_RESV_SIZE);
1368 if ((vtoc->efi_parts[i].p_start < vtoc->efi_first_u_lba) ||
1369 (vtoc->efi_parts[i].p_start > vtoc->efi_last_u_lba)) {
1370 (void) fprintf(stderr,
1371 "Partition %d starts at %llu\n",
1373 vtoc->efi_parts[i].p_start);
1374 (void) fprintf(stderr,
1375 "It must be between %llu and %llu.\n",
1376 vtoc->efi_first_u_lba,
1377 vtoc->efi_last_u_lba);
1379 if ((vtoc->efi_parts[i].p_start +
1380 vtoc->efi_parts[i].p_size <
1381 vtoc->efi_first_u_lba) ||
1382 (vtoc->efi_parts[i].p_start +
1383 vtoc->efi_parts[i].p_size >
1384 vtoc->efi_last_u_lba + 1)) {
1385 (void) fprintf(stderr,
1386 "Partition %d ends at %llu\n",
1388 vtoc->efi_parts[i].p_start +
1389 vtoc->efi_parts[i].p_size);
1390 (void) fprintf(stderr,
1391 "It must be between %llu and %llu.\n",
1392 vtoc->efi_first_u_lba,
1393 vtoc->efi_last_u_lba);
1396 for (j = 0; j < vtoc->efi_nparts; j++) {
1397 isize = vtoc->efi_parts[i].p_size;
1398 jsize = vtoc->efi_parts[j].p_size;
1399 istart = vtoc->efi_parts[i].p_start;
1400 jstart = vtoc->efi_parts[j].p_start;
1401 if ((i != j) && (isize != 0) && (jsize != 0)) {
1402 endsect = jstart + jsize -1;
1403 if ((jstart <= istart) &&
1404 (istart <= endsect)) {
1406 (void) fprintf(stderr,
1407 "label error: EFI Labels do not "
1408 "support overlapping partitions\n");
1410 (void) fprintf(stderr,
1411 "Partition %d overlaps partition "
1418 /* make sure there is a reserved partition */
1419 if (resv_part == -1) {
1420 (void) fprintf(stderr,
1421 "no reserved partition found\n");
1426 * We need to get information necessary to construct a *new* efi
1430 efi_auto_sense(int fd, struct dk_gpt **vtoc)
1436 * Now build the default partition table
1438 if (efi_alloc_and_init(fd, EFI_NUMPAR, vtoc) != 0) {
1440 (void) fprintf(stderr, "efi_alloc_and_init failed.\n");
1445 for (i = 0; i < MIN((*vtoc)->efi_nparts, V_NUMPAR); i++) {
1446 (*vtoc)->efi_parts[i].p_tag = default_vtoc_map[i].p_tag;
1447 (*vtoc)->efi_parts[i].p_flag = default_vtoc_map[i].p_flag;
1448 (*vtoc)->efi_parts[i].p_start = 0;
1449 (*vtoc)->efi_parts[i].p_size = 0;
1452 * Make constants first
1453 * and variable partitions later
1456 /* root partition - s0 128 MB */
1457 (*vtoc)->efi_parts[0].p_start = 34;
1458 (*vtoc)->efi_parts[0].p_size = 262144;
1460 /* partition - s1 128 MB */
1461 (*vtoc)->efi_parts[1].p_start = 262178;
1462 (*vtoc)->efi_parts[1].p_size = 262144;
1464 /* partition -s2 is NOT the Backup disk */
1465 (*vtoc)->efi_parts[2].p_tag = V_UNASSIGNED;
1467 /* partition -s6 /usr partition - HOG */
1468 (*vtoc)->efi_parts[6].p_start = 524322;
1469 (*vtoc)->efi_parts[6].p_size = (*vtoc)->efi_last_u_lba - 524322
1472 /* efi reserved partition - s9 16K */
1473 (*vtoc)->efi_parts[8].p_start = (*vtoc)->efi_last_u_lba - (1024 * 16);
1474 (*vtoc)->efi_parts[8].p_size = (1024 * 16);
1475 (*vtoc)->efi_parts[8].p_tag = V_RESERVED;