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) || defined(__arm) || \
91 defined(__powerpc) || defined(__sparc)
92 { V_BOOT
, V_UNMNT
}, /* i - 8 */
93 { V_ALTSCTR
, 0 }, /* j - 9 */
96 #error No VTOC format defined.
97 #endif /* defined(i386) */
99 { V_UNASSIGNED
, 0 }, /* k - 10 */
100 { V_UNASSIGNED
, 0 }, /* l - 11 */
101 { V_UNASSIGNED
, 0 }, /* m - 12 */
102 { V_UNASSIGNED
, 0 }, /* n - 13 */
103 { V_UNASSIGNED
, 0 }, /* o - 14 */
104 { V_UNASSIGNED
, 0 }, /* p - 15 */
105 #endif /* defined(_SUNOS_VTOC_16) */
114 static int efi_read(int, struct dk_gpt
*);
117 * Return a 32-bit CRC of the contents of the buffer. Pre-and-post
118 * one's conditioning will be handled by crc32() internally.
121 efi_crc32(const unsigned char *buf
, unsigned int size
)
123 uint32_t crc
= crc32(0, Z_NULL
, 0);
125 crc
= crc32(crc
, buf
, size
);
131 read_disk_info(int fd
, diskaddr_t
*capacity
, uint_t
*lbsize
)
134 unsigned long long capacity_size
;
136 if (ioctl(fd
, BLKSSZGET
, §or_size
) < 0)
139 if (ioctl(fd
, BLKGETSIZE64
, &capacity_size
) < 0)
142 *lbsize
= (uint_t
)sector_size
;
143 *capacity
= (diskaddr_t
)(capacity_size
/ sector_size
);
149 efi_get_info(int fd
, struct dk_cinfo
*dki_info
)
151 #if defined(__linux__)
156 memset(dki_info
, 0, sizeof (*dki_info
));
158 path
= calloc(PATH_MAX
, 1);
163 * The simplest way to get the partition number under linux is
164 * to parse it out of the /dev/<disk><parition> block device name.
165 * The kernel creates this using the partition number when it
166 * populates /dev/ so it may be trusted. The tricky bit here is
167 * that the naming convention is based on the block device type.
168 * So we need to take this in to account when parsing out the
169 * partition information. Another issue is that the libefi API
170 * API only provides the open fd and not the file path. To handle
171 * this realpath(3) is used to resolve the block device name from
172 * /proc/self/fd/<fd>. Aside from the partition number we collect
173 * some additional device info.
175 (void) sprintf(path
, "/proc/self/fd/%d", fd
);
176 dev_path
= realpath(path
, NULL
);
179 if (dev_path
== NULL
)
182 if ((strncmp(dev_path
, "/dev/sd", 7) == 0)) {
183 strcpy(dki_info
->dki_cname
, "sd");
184 dki_info
->dki_ctype
= DKC_SCSI_CCS
;
185 rval
= sscanf(dev_path
, "/dev/%[a-zA-Z]%hu",
187 &dki_info
->dki_partition
);
188 } else if ((strncmp(dev_path
, "/dev/hd", 7) == 0)) {
189 strcpy(dki_info
->dki_cname
, "hd");
190 dki_info
->dki_ctype
= DKC_DIRECT
;
191 rval
= sscanf(dev_path
, "/dev/%[a-zA-Z]%hu",
193 &dki_info
->dki_partition
);
194 } else if ((strncmp(dev_path
, "/dev/md", 7) == 0)) {
195 strcpy(dki_info
->dki_cname
, "pseudo");
196 dki_info
->dki_ctype
= DKC_MD
;
197 strcpy(dki_info
->dki_dname
, "md");
198 rval
= sscanf(dev_path
, "/dev/md%[0-9]p%hu",
199 dki_info
->dki_dname
+ 2,
200 &dki_info
->dki_partition
);
201 } else if ((strncmp(dev_path
, "/dev/vd", 7) == 0)) {
202 strcpy(dki_info
->dki_cname
, "vd");
203 dki_info
->dki_ctype
= DKC_MD
;
204 rval
= sscanf(dev_path
, "/dev/%[a-zA-Z]%hu",
206 &dki_info
->dki_partition
);
207 } else if ((strncmp(dev_path
, "/dev/dm-", 8) == 0)) {
208 strcpy(dki_info
->dki_cname
, "pseudo");
209 dki_info
->dki_ctype
= DKC_VBD
;
210 strcpy(dki_info
->dki_dname
, "dm-");
211 rval
= sscanf(dev_path
, "/dev/dm-%[0-9]p%hu",
212 dki_info
->dki_dname
+ 3,
213 &dki_info
->dki_partition
);
214 } else if ((strncmp(dev_path
, "/dev/ram", 8) == 0)) {
215 strcpy(dki_info
->dki_cname
, "pseudo");
216 dki_info
->dki_ctype
= DKC_PCMCIA_MEM
;
217 strcpy(dki_info
->dki_dname
, "ram");
218 rval
= sscanf(dev_path
, "/dev/ram%[0-9]p%hu",
219 dki_info
->dki_dname
+ 3,
220 &dki_info
->dki_partition
);
221 } else if ((strncmp(dev_path
, "/dev/loop", 9) == 0)) {
222 strcpy(dki_info
->dki_cname
, "pseudo");
223 dki_info
->dki_ctype
= DKC_VBD
;
224 strcpy(dki_info
->dki_dname
, "loop");
225 rval
= sscanf(dev_path
, "/dev/loop%[0-9]p%hu",
226 dki_info
->dki_dname
+ 4,
227 &dki_info
->dki_partition
);
229 strcpy(dki_info
->dki_dname
, "unknown");
230 strcpy(dki_info
->dki_cname
, "unknown");
231 dki_info
->dki_ctype
= DKC_UNKNOWN
;
239 dki_info
->dki_partition
= 0;
244 if (ioctl(fd
, DKIOCINFO
, (caddr_t
)dki_info
) == -1)
250 (void) fprintf(stderr
, "DKIOCINFO errno 0x%x\n", errno
);
263 * the number of blocks the EFI label takes up (round up to nearest
266 #define NBLOCKS(p, l) (1 + ((((p) * (int)sizeof (efi_gpe_t)) + \
268 /* number of partitions -- limited by what we can malloc */
269 #define MAX_PARTS ((4294967295UL - sizeof (struct dk_gpt)) / \
270 sizeof (struct dk_part))
273 efi_alloc_and_init(int fd
, uint32_t nparts
, struct dk_gpt
**vtoc
)
275 diskaddr_t capacity
= 0;
281 struct dk_cinfo dki_info
;
283 if (read_disk_info(fd
, &capacity
, &lbsize
) != 0)
286 #if defined(__linux__)
287 if (efi_get_info(fd
, &dki_info
) != 0)
290 if (dki_info
.dki_partition
!= 0)
293 if ((dki_info
.dki_ctype
== DKC_PCMCIA_MEM
) ||
294 (dki_info
.dki_ctype
== DKC_VBD
) ||
295 (dki_info
.dki_ctype
== DKC_UNKNOWN
))
299 nblocks
= NBLOCKS(nparts
, lbsize
);
300 if ((nblocks
* lbsize
) < EFI_MIN_ARRAY_SIZE
+ lbsize
) {
301 /* 16K plus one block for the GPT */
302 nblocks
= EFI_MIN_ARRAY_SIZE
/ lbsize
+ 1;
305 if (nparts
> MAX_PARTS
) {
307 (void) fprintf(stderr
,
308 "the maximum number of partitions supported is %lu\n",
314 length
= sizeof (struct dk_gpt
) +
315 sizeof (struct dk_part
) * (nparts
- 1);
317 if ((*vtoc
= calloc(length
, 1)) == NULL
)
322 vptr
->efi_version
= EFI_VERSION_CURRENT
;
323 vptr
->efi_lbasize
= lbsize
;
324 vptr
->efi_nparts
= nparts
;
326 * add one block here for the PMBR; on disks with a 512 byte
327 * block size and 128 or fewer partitions, efi_first_u_lba
328 * should work out to "34"
330 vptr
->efi_first_u_lba
= nblocks
+ 1;
331 vptr
->efi_last_lba
= capacity
- 1;
332 vptr
->efi_altern_lba
= capacity
-1;
333 vptr
->efi_last_u_lba
= vptr
->efi_last_lba
- nblocks
;
335 (void) uuid_generate((uchar_t
*)&uuid
);
336 UUID_LE_CONVERT(vptr
->efi_disk_uguid
, uuid
);
341 * Read EFI - return partition number upon success.
344 efi_alloc_and_read(int fd
, struct dk_gpt
**vtoc
)
350 /* figure out the number of entries that would fit into 16K */
351 nparts
= EFI_MIN_ARRAY_SIZE
/ sizeof (efi_gpe_t
);
352 length
= (int) sizeof (struct dk_gpt
) +
353 (int) sizeof (struct dk_part
) * (nparts
- 1);
354 if ((*vtoc
= calloc(length
, 1)) == NULL
)
357 (*vtoc
)->efi_nparts
= nparts
;
358 rval
= efi_read(fd
, *vtoc
);
360 if ((rval
== VT_EINVAL
) && (*vtoc
)->efi_nparts
> nparts
) {
362 length
= (int) sizeof (struct dk_gpt
) +
363 (int) sizeof (struct dk_part
) *
364 ((*vtoc
)->efi_nparts
- 1);
365 nparts
= (*vtoc
)->efi_nparts
;
366 if ((tmp
= realloc(*vtoc
, length
)) == NULL
) {
372 rval
= efi_read(fd
, *vtoc
);
378 (void) fprintf(stderr
,
379 "read of EFI table failed, rval=%d\n", rval
);
389 efi_ioctl(int fd
, int cmd
, dk_efi_t
*dk_ioc
)
391 void *data
= dk_ioc
->dki_data
;
393 #if defined(__linux__)
398 * When the IO is not being performed in kernel as an ioctl we need
399 * to know the sector size so we can seek to the proper byte offset.
401 if (read_disk_info(fd
, &capacity
, &lbsize
) == -1) {
403 fprintf(stderr
, "unable to read disk info: %d", errno
);
413 (void) fprintf(stderr
, "DKIOCGETEFI assuming "
414 "LBA %d bytes\n", DEV_BSIZE
);
419 error
= lseek(fd
, dk_ioc
->dki_lba
* lbsize
, SEEK_SET
);
422 (void) fprintf(stderr
, "DKIOCGETEFI lseek "
423 "error: %d\n", errno
);
427 error
= read(fd
, data
, dk_ioc
->dki_length
);
430 (void) fprintf(stderr
, "DKIOCGETEFI read "
431 "error: %d\n", errno
);
435 if (error
!= dk_ioc
->dki_length
) {
437 (void) fprintf(stderr
, "DKIOCGETEFI short "
438 "read of %d bytes\n", error
);
448 (void) fprintf(stderr
, "DKIOCSETEFI unknown "
454 error
= lseek(fd
, dk_ioc
->dki_lba
* lbsize
, SEEK_SET
);
457 (void) fprintf(stderr
, "DKIOCSETEFI lseek "
458 "error: %d\n", errno
);
462 error
= write(fd
, data
, dk_ioc
->dki_length
);
465 (void) fprintf(stderr
, "DKIOCSETEFI write "
466 "error: %d\n", errno
);
470 if (error
!= dk_ioc
->dki_length
) {
472 (void) fprintf(stderr
, "DKIOCSETEFI short "
473 "write of %d bytes\n", error
);
478 /* Sync the new EFI table to disk */
483 /* Ensure any local disk cache is also flushed */
484 if (ioctl(fd
, BLKFLSBUF
, 0) == -1)
492 (void) fprintf(stderr
, "unsupported ioctl()\n");
498 dk_ioc
->dki_data_64
= (uint64_t)(uintptr_t)data
;
499 error
= ioctl(fd
, cmd
, (void *)dk_ioc
);
500 dk_ioc
->dki_data
= data
;
508 #if defined(__linux__)
512 /* Notify the kernel a devices partition table has been updated */
513 while ((error
= ioctl(fd
, BLKRRPART
)) != 0) {
514 if ((--retry
== 0) || (errno
!= EBUSY
)) {
515 (void) fprintf(stderr
, "the kernel failed to rescan "
516 "the partition table: %d\n", errno
);
527 check_label(int fd
, dk_efi_t
*dk_ioc
)
532 if (efi_ioctl(fd
, DKIOCGETEFI
, dk_ioc
) == -1) {
540 efi
= dk_ioc
->dki_data
;
541 if (efi
->efi_gpt_Signature
!= LE_64(EFI_SIGNATURE
)) {
543 (void) fprintf(stderr
,
544 "Bad EFI signature: 0x%llx != 0x%llx\n",
545 (long long)efi
->efi_gpt_Signature
,
546 (long long)LE_64(EFI_SIGNATURE
));
551 * check CRC of the header; the size of the header should
552 * never be larger than one block
554 crc
= efi
->efi_gpt_HeaderCRC32
;
555 efi
->efi_gpt_HeaderCRC32
= 0;
556 len_t headerSize
= (len_t
)LE_32(efi
->efi_gpt_HeaderSize
);
558 if (headerSize
< EFI_MIN_LABEL_SIZE
|| headerSize
> EFI_LABEL_SIZE
) {
560 (void) fprintf(stderr
,
561 "Invalid EFI HeaderSize %llu. Assuming %d.\n",
562 headerSize
, EFI_MIN_LABEL_SIZE
);
565 if ((headerSize
> dk_ioc
->dki_length
) ||
566 crc
!= LE_32(efi_crc32((unsigned char *)efi
, headerSize
))) {
568 (void) fprintf(stderr
,
569 "Bad EFI CRC: 0x%x != 0x%x\n",
570 crc
, LE_32(efi_crc32((unsigned char *)efi
,
579 efi_read(int fd
, struct dk_gpt
*vtoc
)
586 diskaddr_t capacity
= 0;
588 struct dk_minfo disk_info
;
591 efi_gpe_t
*efi_parts
;
592 struct dk_cinfo dki_info
;
593 uint32_t user_length
;
594 boolean_t legacy_label
= B_FALSE
;
597 * get the partition number for this file descriptor.
599 if ((rval
= efi_get_info(fd
, &dki_info
)) != 0)
602 if ((strncmp(dki_info
.dki_cname
, "pseudo", 7) == 0) &&
603 (strncmp(dki_info
.dki_dname
, "md", 3) == 0)) {
605 } else if ((strncmp(dki_info
.dki_cname
, "vdc", 4) == 0) &&
606 (strncmp(dki_info
.dki_dname
, "vdc", 4) == 0)) {
608 * The controller and drive name "vdc" (virtual disk client)
609 * indicates a LDoms virtual disk.
614 /* get the LBA size */
615 if (read_disk_info(fd
, &capacity
, &lbsize
) == -1) {
617 (void) fprintf(stderr
,
618 "unable to read disk info: %d",
624 disk_info
.dki_lbsize
= lbsize
;
625 disk_info
.dki_capacity
= capacity
;
627 if (disk_info
.dki_lbsize
== 0) {
629 (void) fprintf(stderr
,
630 "efi_read: assuming LBA 512 bytes\n");
632 disk_info
.dki_lbsize
= DEV_BSIZE
;
635 * Read the EFI GPT to figure out how many partitions we need
639 if (NBLOCKS(vtoc
->efi_nparts
, disk_info
.dki_lbsize
) < 34) {
640 label_len
= EFI_MIN_ARRAY_SIZE
+ disk_info
.dki_lbsize
;
642 label_len
= vtoc
->efi_nparts
* (int) sizeof (efi_gpe_t
) +
643 disk_info
.dki_lbsize
;
644 if (label_len
% disk_info
.dki_lbsize
) {
645 /* pad to physical sector size */
646 label_len
+= disk_info
.dki_lbsize
;
647 label_len
&= ~(disk_info
.dki_lbsize
- 1);
651 if (posix_memalign((void **)&dk_ioc
.dki_data
,
652 disk_info
.dki_lbsize
, label_len
))
655 memset(dk_ioc
.dki_data
, 0, label_len
);
656 dk_ioc
.dki_length
= disk_info
.dki_lbsize
;
657 user_length
= vtoc
->efi_nparts
;
658 efi
= dk_ioc
.dki_data
;
660 dk_ioc
.dki_length
= label_len
;
661 if (efi_ioctl(fd
, DKIOCGETEFI
, &dk_ioc
) == -1) {
669 } else if ((rval
= check_label(fd
, &dk_ioc
)) == VT_EINVAL
) {
671 * No valid label here; try the alternate. Note that here
672 * we just read GPT header and save it into dk_ioc.data,
673 * Later, we will read GUID partition entry array if we
674 * can get valid GPT header.
678 * This is a workaround for legacy systems. In the past, the
679 * last sector of SCSI disk was invisible on x86 platform. At
680 * that time, backup label was saved on the next to the last
681 * sector. It is possible for users to move a disk from previous
682 * solaris system to present system. Here, we attempt to search
683 * legacy backup EFI label first.
685 dk_ioc
.dki_lba
= disk_info
.dki_capacity
- 2;
686 dk_ioc
.dki_length
= disk_info
.dki_lbsize
;
687 rval
= check_label(fd
, &dk_ioc
);
688 if (rval
== VT_EINVAL
) {
690 * we didn't find legacy backup EFI label, try to
691 * search backup EFI label in the last block.
693 dk_ioc
.dki_lba
= disk_info
.dki_capacity
- 1;
694 dk_ioc
.dki_length
= disk_info
.dki_lbsize
;
695 rval
= check_label(fd
, &dk_ioc
);
697 legacy_label
= B_TRUE
;
699 (void) fprintf(stderr
,
700 "efi_read: primary label corrupt; "
701 "using EFI backup label located on"
702 " the last block\n");
705 if ((efi_debug
) && (rval
== 0))
706 (void) fprintf(stderr
, "efi_read: primary label"
707 " corrupt; using legacy EFI backup label "
708 " located on the next to last block\n");
712 dk_ioc
.dki_lba
= LE_64(efi
->efi_gpt_PartitionEntryLBA
);
713 vtoc
->efi_flags
|= EFI_GPT_PRIMARY_CORRUPT
;
715 LE_32(efi
->efi_gpt_NumberOfPartitionEntries
);
717 * Partition tables are between backup GPT header
718 * table and ParitionEntryLBA (the starting LBA of
719 * the GUID partition entries array). Now that we
720 * already got valid GPT header and saved it in
721 * dk_ioc.dki_data, we try to get GUID partition
725 dk_ioc
.dki_data
= (efi_gpt_t
*)((char *)dk_ioc
.dki_data
726 + disk_info
.dki_lbsize
);
728 dk_ioc
.dki_length
= disk_info
.dki_capacity
- 1 -
731 dk_ioc
.dki_length
= disk_info
.dki_capacity
- 2 -
733 dk_ioc
.dki_length
*= disk_info
.dki_lbsize
;
734 if (dk_ioc
.dki_length
>
735 ((len_t
)label_len
- sizeof (*dk_ioc
.dki_data
))) {
739 * read GUID partition entry array
741 rval
= efi_ioctl(fd
, DKIOCGETEFI
, &dk_ioc
);
745 } else if (rval
== 0) {
747 dk_ioc
.dki_lba
= LE_64(efi
->efi_gpt_PartitionEntryLBA
);
749 dk_ioc
.dki_data
= (efi_gpt_t
*)((char *)dk_ioc
.dki_data
750 + disk_info
.dki_lbsize
);
751 dk_ioc
.dki_length
= label_len
- disk_info
.dki_lbsize
;
752 rval
= efi_ioctl(fd
, DKIOCGETEFI
, &dk_ioc
);
754 } else if (vdc_flag
&& rval
== VT_ERROR
&& errno
== EINVAL
) {
756 * When the device is a LDoms virtual disk, the DKIOCGETEFI
757 * ioctl can fail with EINVAL if the virtual disk backend
758 * is a ZFS volume serviced by a domain running an old version
759 * of Solaris. This is because the DKIOCGETEFI ioctl was
760 * initially incorrectly implemented for a ZFS volume and it
761 * expected the GPT and GPE to be retrieved with a single ioctl.
762 * So we try to read the GPT and the GPE using that old style
766 dk_ioc
.dki_length
= label_len
;
767 rval
= check_label(fd
, &dk_ioc
);
775 /* LINTED -- always longlong aligned */
776 efi_parts
= (efi_gpe_t
*)(((char *)efi
) + disk_info
.dki_lbsize
);
779 * Assemble this into a "dk_gpt" struct for easier
780 * digestibility by applications.
782 vtoc
->efi_version
= LE_32(efi
->efi_gpt_Revision
);
783 vtoc
->efi_nparts
= LE_32(efi
->efi_gpt_NumberOfPartitionEntries
);
784 vtoc
->efi_part_size
= LE_32(efi
->efi_gpt_SizeOfPartitionEntry
);
785 vtoc
->efi_lbasize
= disk_info
.dki_lbsize
;
786 vtoc
->efi_last_lba
= disk_info
.dki_capacity
- 1;
787 vtoc
->efi_first_u_lba
= LE_64(efi
->efi_gpt_FirstUsableLBA
);
788 vtoc
->efi_last_u_lba
= LE_64(efi
->efi_gpt_LastUsableLBA
);
789 vtoc
->efi_altern_lba
= LE_64(efi
->efi_gpt_AlternateLBA
);
790 UUID_LE_CONVERT(vtoc
->efi_disk_uguid
, efi
->efi_gpt_DiskGUID
);
793 * If the array the user passed in is too small, set the length
794 * to what it needs to be and return
796 if (user_length
< vtoc
->efi_nparts
) {
800 for (i
= 0; i
< vtoc
->efi_nparts
; i
++) {
802 UUID_LE_CONVERT(vtoc
->efi_parts
[i
].p_guid
,
803 efi_parts
[i
].efi_gpe_PartitionTypeGUID
);
806 j
< sizeof (conversion_array
)
807 / sizeof (struct uuid_to_ptag
); j
++) {
809 if (bcmp(&vtoc
->efi_parts
[i
].p_guid
,
810 &conversion_array
[j
].uuid
,
811 sizeof (struct uuid
)) == 0) {
812 vtoc
->efi_parts
[i
].p_tag
= j
;
816 if (vtoc
->efi_parts
[i
].p_tag
== V_UNASSIGNED
)
818 vtoc
->efi_parts
[i
].p_flag
=
819 LE_16(efi_parts
[i
].efi_gpe_Attributes
.PartitionAttrs
);
820 vtoc
->efi_parts
[i
].p_start
=
821 LE_64(efi_parts
[i
].efi_gpe_StartingLBA
);
822 vtoc
->efi_parts
[i
].p_size
=
823 LE_64(efi_parts
[i
].efi_gpe_EndingLBA
) -
824 vtoc
->efi_parts
[i
].p_start
+ 1;
825 for (j
= 0; j
< EFI_PART_NAME_LEN
; j
++) {
826 vtoc
->efi_parts
[i
].p_name
[j
] =
828 efi_parts
[i
].efi_gpe_PartitionName
[j
]);
831 UUID_LE_CONVERT(vtoc
->efi_parts
[i
].p_uguid
,
832 efi_parts
[i
].efi_gpe_UniquePartitionGUID
);
836 return (dki_info
.dki_partition
);
839 /* writes a "protective" MBR */
841 write_pmbr(int fd
, struct dk_gpt
*vtoc
)
846 diskaddr_t size_in_lba
;
850 len
= (vtoc
->efi_lbasize
== 0) ? sizeof (mb
) : vtoc
->efi_lbasize
;
851 if (posix_memalign((void **)&buf
, len
, len
))
855 * Preserve any boot code and disk signature if the first block is
860 dk_ioc
.dki_length
= len
;
861 /* LINTED -- always longlong aligned */
862 dk_ioc
.dki_data
= (efi_gpt_t
*)buf
;
863 if (efi_ioctl(fd
, DKIOCGETEFI
, &dk_ioc
) == -1) {
864 (void) memcpy(&mb
, buf
, sizeof (mb
));
865 bzero(&mb
, sizeof (mb
));
866 mb
.signature
= LE_16(MBB_MAGIC
);
868 (void) memcpy(&mb
, buf
, sizeof (mb
));
869 if (mb
.signature
!= LE_16(MBB_MAGIC
)) {
870 bzero(&mb
, sizeof (mb
));
871 mb
.signature
= LE_16(MBB_MAGIC
);
875 bzero(&mb
.parts
, sizeof (mb
.parts
));
876 cp
= (uchar_t
*)&mb
.parts
[0];
877 /* bootable or not */
879 /* beginning CHS; 0xffffff if not representable */
885 /* ending CHS; 0xffffff if not representable */
889 /* starting LBA: 1 (little endian format) by EFI definition */
894 /* ending LBA: last block on the disk (little endian format) */
895 size_in_lba
= vtoc
->efi_last_lba
;
896 if (size_in_lba
< 0xffffffff) {
897 *cp
++ = (size_in_lba
& 0x000000ff);
898 *cp
++ = (size_in_lba
& 0x0000ff00) >> 8;
899 *cp
++ = (size_in_lba
& 0x00ff0000) >> 16;
900 *cp
++ = (size_in_lba
& 0xff000000) >> 24;
908 (void) memcpy(buf
, &mb
, sizeof (mb
));
909 /* LINTED -- always longlong aligned */
910 dk_ioc
.dki_data
= (efi_gpt_t
*)buf
;
912 dk_ioc
.dki_length
= len
;
913 if (efi_ioctl(fd
, DKIOCSETEFI
, &dk_ioc
) == -1) {
928 /* make sure the user specified something reasonable */
930 check_input(struct dk_gpt
*vtoc
)
934 diskaddr_t istart
, jstart
, isize
, jsize
, endsect
;
937 * Sanity-check the input (make sure no partitions overlap)
939 for (i
= 0; i
< vtoc
->efi_nparts
; i
++) {
940 /* It can't be unassigned and have an actual size */
941 if ((vtoc
->efi_parts
[i
].p_tag
== V_UNASSIGNED
) &&
942 (vtoc
->efi_parts
[i
].p_size
!= 0)) {
944 (void) fprintf(stderr
, "partition %d is "
945 "\"unassigned\" but has a size of %llu",
946 i
, vtoc
->efi_parts
[i
].p_size
);
950 if (vtoc
->efi_parts
[i
].p_tag
== V_UNASSIGNED
) {
951 if (uuid_is_null((uchar_t
*)&vtoc
->efi_parts
[i
].p_guid
))
953 /* we have encountered an unknown uuid */
954 vtoc
->efi_parts
[i
].p_tag
= 0xff;
956 if (vtoc
->efi_parts
[i
].p_tag
== V_RESERVED
) {
957 if (resv_part
!= -1) {
959 (void) fprintf(stderr
, "found "
960 "duplicate reserved partition "
967 if ((vtoc
->efi_parts
[i
].p_start
< vtoc
->efi_first_u_lba
) ||
968 (vtoc
->efi_parts
[i
].p_start
> vtoc
->efi_last_u_lba
)) {
970 (void) fprintf(stderr
,
971 "Partition %d starts at %llu. ",
973 vtoc
->efi_parts
[i
].p_start
);
974 (void) fprintf(stderr
,
975 "It must be between %llu and %llu.\n",
976 vtoc
->efi_first_u_lba
,
977 vtoc
->efi_last_u_lba
);
981 if ((vtoc
->efi_parts
[i
].p_start
+
982 vtoc
->efi_parts
[i
].p_size
<
983 vtoc
->efi_first_u_lba
) ||
984 (vtoc
->efi_parts
[i
].p_start
+
985 vtoc
->efi_parts
[i
].p_size
>
986 vtoc
->efi_last_u_lba
+ 1)) {
988 (void) fprintf(stderr
,
989 "Partition %d ends at %llu. ",
991 vtoc
->efi_parts
[i
].p_start
+
992 vtoc
->efi_parts
[i
].p_size
);
993 (void) fprintf(stderr
,
994 "It must be between %llu and %llu.\n",
995 vtoc
->efi_first_u_lba
,
996 vtoc
->efi_last_u_lba
);
1001 for (j
= 0; j
< vtoc
->efi_nparts
; j
++) {
1002 isize
= vtoc
->efi_parts
[i
].p_size
;
1003 jsize
= vtoc
->efi_parts
[j
].p_size
;
1004 istart
= vtoc
->efi_parts
[i
].p_start
;
1005 jstart
= vtoc
->efi_parts
[j
].p_start
;
1006 if ((i
!= j
) && (isize
!= 0) && (jsize
!= 0)) {
1007 endsect
= jstart
+ jsize
-1;
1008 if ((jstart
<= istart
) &&
1009 (istart
<= endsect
)) {
1011 (void) fprintf(stderr
,
1012 "Partition %d overlaps "
1013 "partition %d.", i
, j
);
1020 /* just a warning for now */
1021 if ((resv_part
== -1) && efi_debug
) {
1022 (void) fprintf(stderr
,
1023 "no reserved partition found\n");
1029 * add all the unallocated space to the current label
1032 efi_use_whole_disk(int fd
)
1034 struct dk_gpt
*efi_label
;
1037 uint_t resv_index
= 0, data_index
= 0;
1038 diskaddr_t resv_start
= 0, data_start
= 0;
1039 diskaddr_t difference
;
1041 rval
= efi_alloc_and_read(fd
, &efi_label
);
1047 * If alter_lba is 1, we are using the backup label.
1048 * Since we can locate the backup label by disk capacity,
1049 * there must be no unallocated space.
1051 if ((efi_label
->efi_altern_lba
== 1) || (efi_label
->efi_altern_lba
1052 >= efi_label
->efi_last_lba
)) {
1054 (void) fprintf(stderr
,
1055 "efi_use_whole_disk: requested space not found\n");
1057 efi_free(efi_label
);
1061 difference
= efi_label
->efi_last_lba
- efi_label
->efi_altern_lba
;
1064 * Find the last physically non-zero partition.
1065 * This is the reserved partition.
1067 for (i
= 0; i
< efi_label
->efi_nparts
; i
++) {
1068 if (resv_start
< efi_label
->efi_parts
[i
].p_start
) {
1069 resv_start
= efi_label
->efi_parts
[i
].p_start
;
1075 * Find the last physically non-zero partition before that.
1076 * This is the data partition.
1078 for (i
= 0; i
< resv_index
; i
++) {
1079 if (data_start
< efi_label
->efi_parts
[i
].p_start
) {
1080 data_start
= efi_label
->efi_parts
[i
].p_start
;
1086 * Move the reserved partition. There is currently no data in
1087 * here except fabricated devids (which get generated via
1088 * efi_write()). So there is no need to copy data.
1090 efi_label
->efi_parts
[data_index
].p_size
+= difference
;
1091 efi_label
->efi_parts
[resv_index
].p_start
+= difference
;
1092 efi_label
->efi_last_u_lba
+= difference
;
1094 rval
= efi_write(fd
, efi_label
);
1097 (void) fprintf(stderr
,
1098 "efi_use_whole_disk:fail to write label, rval=%d\n",
1101 efi_free(efi_label
);
1105 efi_free(efi_label
);
1111 * write EFI label and backup label
1114 efi_write(int fd
, struct dk_gpt
*vtoc
)
1118 efi_gpe_t
*efi_parts
;
1120 struct dk_cinfo dki_info
;
1124 diskaddr_t lba_backup_gpt_hdr
;
1126 if ((rval
= efi_get_info(fd
, &dki_info
)) != 0)
1129 /* check if we are dealing wih a metadevice */
1130 if ((strncmp(dki_info
.dki_cname
, "pseudo", 7) == 0) &&
1131 (strncmp(dki_info
.dki_dname
, "md", 3) == 0)) {
1135 if (check_input(vtoc
)) {
1137 * not valid; if it's a metadevice just pass it down
1138 * because SVM will do its own checking
1146 if (NBLOCKS(vtoc
->efi_nparts
, vtoc
->efi_lbasize
) < 34) {
1147 dk_ioc
.dki_length
= EFI_MIN_ARRAY_SIZE
+ vtoc
->efi_lbasize
;
1149 dk_ioc
.dki_length
= NBLOCKS(vtoc
->efi_nparts
,
1150 vtoc
->efi_lbasize
) *
1155 * the number of blocks occupied by GUID partition entry array
1157 nblocks
= dk_ioc
.dki_length
/ vtoc
->efi_lbasize
- 1;
1160 * Backup GPT header is located on the block after GUID
1161 * partition entry array. Here, we calculate the address
1162 * for backup GPT header.
1164 lba_backup_gpt_hdr
= vtoc
->efi_last_u_lba
+ 1 + nblocks
;
1165 if (posix_memalign((void **)&dk_ioc
.dki_data
,
1166 vtoc
->efi_lbasize
, dk_ioc
.dki_length
))
1169 memset(dk_ioc
.dki_data
, 0, dk_ioc
.dki_length
);
1170 efi
= dk_ioc
.dki_data
;
1172 /* stuff user's input into EFI struct */
1173 efi
->efi_gpt_Signature
= LE_64(EFI_SIGNATURE
);
1174 efi
->efi_gpt_Revision
= LE_32(vtoc
->efi_version
); /* 0x02000100 */
1175 efi
->efi_gpt_HeaderSize
= LE_32(sizeof (struct efi_gpt
) - LEN_EFI_PAD
);
1176 efi
->efi_gpt_Reserved1
= 0;
1177 efi
->efi_gpt_MyLBA
= LE_64(1ULL);
1178 efi
->efi_gpt_AlternateLBA
= LE_64(lba_backup_gpt_hdr
);
1179 efi
->efi_gpt_FirstUsableLBA
= LE_64(vtoc
->efi_first_u_lba
);
1180 efi
->efi_gpt_LastUsableLBA
= LE_64(vtoc
->efi_last_u_lba
);
1181 efi
->efi_gpt_PartitionEntryLBA
= LE_64(2ULL);
1182 efi
->efi_gpt_NumberOfPartitionEntries
= LE_32(vtoc
->efi_nparts
);
1183 efi
->efi_gpt_SizeOfPartitionEntry
= LE_32(sizeof (struct efi_gpe
));
1184 UUID_LE_CONVERT(efi
->efi_gpt_DiskGUID
, vtoc
->efi_disk_uguid
);
1186 /* LINTED -- always longlong aligned */
1187 efi_parts
= (efi_gpe_t
*)((char *)dk_ioc
.dki_data
+ vtoc
->efi_lbasize
);
1189 for (i
= 0; i
< vtoc
->efi_nparts
; i
++) {
1191 j
< sizeof (conversion_array
) /
1192 sizeof (struct uuid_to_ptag
); j
++) {
1194 if (vtoc
->efi_parts
[i
].p_tag
== j
) {
1196 efi_parts
[i
].efi_gpe_PartitionTypeGUID
,
1197 conversion_array
[j
].uuid
);
1202 if (j
== sizeof (conversion_array
) /
1203 sizeof (struct uuid_to_ptag
)) {
1205 * If we didn't have a matching uuid match, bail here.
1206 * Don't write a label with unknown uuid.
1209 (void) fprintf(stderr
,
1210 "Unknown uuid for p_tag %d\n",
1211 vtoc
->efi_parts
[i
].p_tag
);
1216 /* Zero's should be written for empty partitions */
1217 if (vtoc
->efi_parts
[i
].p_tag
== V_UNASSIGNED
)
1220 efi_parts
[i
].efi_gpe_StartingLBA
=
1221 LE_64(vtoc
->efi_parts
[i
].p_start
);
1222 efi_parts
[i
].efi_gpe_EndingLBA
=
1223 LE_64(vtoc
->efi_parts
[i
].p_start
+
1224 vtoc
->efi_parts
[i
].p_size
- 1);
1225 efi_parts
[i
].efi_gpe_Attributes
.PartitionAttrs
=
1226 LE_16(vtoc
->efi_parts
[i
].p_flag
);
1227 for (j
= 0; j
< EFI_PART_NAME_LEN
; j
++) {
1228 efi_parts
[i
].efi_gpe_PartitionName
[j
] =
1229 LE_16((ushort_t
)vtoc
->efi_parts
[i
].p_name
[j
]);
1231 if ((vtoc
->efi_parts
[i
].p_tag
!= V_UNASSIGNED
) &&
1232 uuid_is_null((uchar_t
*)&vtoc
->efi_parts
[i
].p_uguid
)) {
1233 (void) uuid_generate((uchar_t
*)
1234 &vtoc
->efi_parts
[i
].p_uguid
);
1236 bcopy(&vtoc
->efi_parts
[i
].p_uguid
,
1237 &efi_parts
[i
].efi_gpe_UniquePartitionGUID
,
1240 efi
->efi_gpt_PartitionEntryArrayCRC32
=
1241 LE_32(efi_crc32((unsigned char *)efi_parts
,
1242 vtoc
->efi_nparts
* (int)sizeof (struct efi_gpe
)));
1243 efi
->efi_gpt_HeaderCRC32
=
1244 LE_32(efi_crc32((unsigned char *)efi
,
1245 LE_32(efi
->efi_gpt_HeaderSize
)));
1247 if (efi_ioctl(fd
, DKIOCSETEFI
, &dk_ioc
) == -1) {
1248 free(dk_ioc
.dki_data
);
1258 /* if it's a metadevice we're done */
1260 free(dk_ioc
.dki_data
);
1264 /* write backup partition array */
1265 dk_ioc
.dki_lba
= vtoc
->efi_last_u_lba
+ 1;
1266 dk_ioc
.dki_length
-= vtoc
->efi_lbasize
;
1268 dk_ioc
.dki_data
= (efi_gpt_t
*)((char *)dk_ioc
.dki_data
+
1271 if (efi_ioctl(fd
, DKIOCSETEFI
, &dk_ioc
) == -1) {
1273 * we wrote the primary label okay, so don't fail
1276 (void) fprintf(stderr
,
1277 "write of backup partitions to block %llu "
1278 "failed, errno %d\n",
1279 vtoc
->efi_last_u_lba
+ 1,
1284 * now swap MyLBA and AlternateLBA fields and write backup
1285 * partition table header
1287 dk_ioc
.dki_lba
= lba_backup_gpt_hdr
;
1288 dk_ioc
.dki_length
= vtoc
->efi_lbasize
;
1290 dk_ioc
.dki_data
= (efi_gpt_t
*)((char *)dk_ioc
.dki_data
-
1292 efi
->efi_gpt_AlternateLBA
= LE_64(1ULL);
1293 efi
->efi_gpt_MyLBA
= LE_64(lba_backup_gpt_hdr
);
1294 efi
->efi_gpt_PartitionEntryLBA
= LE_64(vtoc
->efi_last_u_lba
+ 1);
1295 efi
->efi_gpt_HeaderCRC32
= 0;
1296 efi
->efi_gpt_HeaderCRC32
=
1297 LE_32(efi_crc32((unsigned char *)dk_ioc
.dki_data
,
1298 LE_32(efi
->efi_gpt_HeaderSize
)));
1300 if (efi_ioctl(fd
, DKIOCSETEFI
, &dk_ioc
) == -1) {
1302 (void) fprintf(stderr
,
1303 "write of backup header to block %llu failed, "
1309 /* write the PMBR */
1310 (void) write_pmbr(fd
, vtoc
);
1311 free(dk_ioc
.dki_data
);
1317 efi_free(struct dk_gpt
*ptr
)
1323 * Input: File descriptor
1324 * Output: 1 if disk has an EFI label, or > 2TB with no VTOC or legacy MBR.
1332 struct extvtoc extvtoc
;
1334 if (ioctl(fd
, DKIOCGEXTVTOC
, &extvtoc
) == -1) {
1335 if (errno
== ENOTSUP
)
1337 else if (errno
== ENOTTY
) {
1338 if (ioctl(fd
, DKIOCGVTOC
, &vtoc
) == -1)
1339 if (errno
== ENOTSUP
)
1350 efi_err_check(struct dk_gpt
*vtoc
)
1354 diskaddr_t istart
, jstart
, isize
, jsize
, endsect
;
1358 * make sure no partitions overlap
1360 for (i
= 0; i
< vtoc
->efi_nparts
; i
++) {
1361 /* It can't be unassigned and have an actual size */
1362 if ((vtoc
->efi_parts
[i
].p_tag
== V_UNASSIGNED
) &&
1363 (vtoc
->efi_parts
[i
].p_size
!= 0)) {
1364 (void) fprintf(stderr
,
1365 "partition %d is \"unassigned\" but has a size "
1366 "of %llu\n", i
, vtoc
->efi_parts
[i
].p_size
);
1368 if (vtoc
->efi_parts
[i
].p_tag
== V_UNASSIGNED
) {
1371 if (vtoc
->efi_parts
[i
].p_tag
== V_RESERVED
) {
1372 if (resv_part
!= -1) {
1373 (void) fprintf(stderr
,
1374 "found duplicate reserved partition at "
1378 if (vtoc
->efi_parts
[i
].p_size
!= EFI_MIN_RESV_SIZE
)
1379 (void) fprintf(stderr
,
1380 "Warning: reserved partition size must "
1381 "be %d sectors\n", EFI_MIN_RESV_SIZE
);
1383 if ((vtoc
->efi_parts
[i
].p_start
< vtoc
->efi_first_u_lba
) ||
1384 (vtoc
->efi_parts
[i
].p_start
> vtoc
->efi_last_u_lba
)) {
1385 (void) fprintf(stderr
,
1386 "Partition %d starts at %llu\n",
1388 vtoc
->efi_parts
[i
].p_start
);
1389 (void) fprintf(stderr
,
1390 "It must be between %llu and %llu.\n",
1391 vtoc
->efi_first_u_lba
,
1392 vtoc
->efi_last_u_lba
);
1394 if ((vtoc
->efi_parts
[i
].p_start
+
1395 vtoc
->efi_parts
[i
].p_size
<
1396 vtoc
->efi_first_u_lba
) ||
1397 (vtoc
->efi_parts
[i
].p_start
+
1398 vtoc
->efi_parts
[i
].p_size
>
1399 vtoc
->efi_last_u_lba
+ 1)) {
1400 (void) fprintf(stderr
,
1401 "Partition %d ends at %llu\n",
1403 vtoc
->efi_parts
[i
].p_start
+
1404 vtoc
->efi_parts
[i
].p_size
);
1405 (void) fprintf(stderr
,
1406 "It must be between %llu and %llu.\n",
1407 vtoc
->efi_first_u_lba
,
1408 vtoc
->efi_last_u_lba
);
1411 for (j
= 0; j
< vtoc
->efi_nparts
; j
++) {
1412 isize
= vtoc
->efi_parts
[i
].p_size
;
1413 jsize
= vtoc
->efi_parts
[j
].p_size
;
1414 istart
= vtoc
->efi_parts
[i
].p_start
;
1415 jstart
= vtoc
->efi_parts
[j
].p_start
;
1416 if ((i
!= j
) && (isize
!= 0) && (jsize
!= 0)) {
1417 endsect
= jstart
+ jsize
-1;
1418 if ((jstart
<= istart
) &&
1419 (istart
<= endsect
)) {
1421 (void) fprintf(stderr
,
1422 "label error: EFI Labels do not "
1423 "support overlapping partitions\n");
1425 (void) fprintf(stderr
,
1426 "Partition %d overlaps partition "
1433 /* make sure there is a reserved partition */
1434 if (resv_part
== -1) {
1435 (void) fprintf(stderr
,
1436 "no reserved partition found\n");
1441 * We need to get information necessary to construct a *new* efi
1445 efi_auto_sense(int fd
, struct dk_gpt
**vtoc
)
1451 * Now build the default partition table
1453 if (efi_alloc_and_init(fd
, EFI_NUMPAR
, vtoc
) != 0) {
1455 (void) fprintf(stderr
, "efi_alloc_and_init failed.\n");
1460 for (i
= 0; i
< MIN((*vtoc
)->efi_nparts
, V_NUMPAR
); i
++) {
1461 (*vtoc
)->efi_parts
[i
].p_tag
= default_vtoc_map
[i
].p_tag
;
1462 (*vtoc
)->efi_parts
[i
].p_flag
= default_vtoc_map
[i
].p_flag
;
1463 (*vtoc
)->efi_parts
[i
].p_start
= 0;
1464 (*vtoc
)->efi_parts
[i
].p_size
= 0;
1467 * Make constants first
1468 * and variable partitions later
1471 /* root partition - s0 128 MB */
1472 (*vtoc
)->efi_parts
[0].p_start
= 34;
1473 (*vtoc
)->efi_parts
[0].p_size
= 262144;
1475 /* partition - s1 128 MB */
1476 (*vtoc
)->efi_parts
[1].p_start
= 262178;
1477 (*vtoc
)->efi_parts
[1].p_size
= 262144;
1479 /* partition -s2 is NOT the Backup disk */
1480 (*vtoc
)->efi_parts
[2].p_tag
= V_UNASSIGNED
;
1482 /* partition -s6 /usr partition - HOG */
1483 (*vtoc
)->efi_parts
[6].p_start
= 524322;
1484 (*vtoc
)->efi_parts
[6].p_size
= (*vtoc
)->efi_last_u_lba
- 524322
1487 /* efi reserved partition - s9 16K */
1488 (*vtoc
)->efi_parts
[8].p_start
= (*vtoc
)->efi_last_u_lba
- (1024 * 16);
1489 (*vtoc
)->efi_parts
[8].p_size
= (1024 * 16);
1490 (*vtoc
)->efi_parts
[8].p_tag
= V_RESERVED
;