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) {
274 (void) fprintf(stderr
,
275 "couldn't read disk information\n");
278 #if defined(__linux__)
279 if (efi_get_info(fd
, &dki_info
) != 0) {
281 (void) fprintf(stderr
,
282 "couldn't read disk information\n");
286 if (dki_info
.dki_partition
!= 0)
289 if ((dki_info
.dki_ctype
== DKC_PCMCIA_MEM
) ||
290 (dki_info
.dki_ctype
== DKC_VBD
) ||
291 (dki_info
.dki_ctype
== DKC_UNKNOWN
))
295 nblocks
= NBLOCKS(nparts
, lbsize
);
296 if ((nblocks
* lbsize
) < EFI_MIN_ARRAY_SIZE
+ lbsize
) {
297 /* 16K plus one block for the GPT */
298 nblocks
= EFI_MIN_ARRAY_SIZE
/ lbsize
+ 1;
301 if (nparts
> MAX_PARTS
) {
303 (void) fprintf(stderr
,
304 "the maximum number of partitions supported is %lu\n",
310 length
= sizeof (struct dk_gpt
) +
311 sizeof (struct dk_part
) * (nparts
- 1);
313 if ((*vtoc
= calloc(length
, 1)) == NULL
)
318 vptr
->efi_version
= EFI_VERSION_CURRENT
;
319 vptr
->efi_lbasize
= lbsize
;
320 vptr
->efi_nparts
= nparts
;
322 * add one block here for the PMBR; on disks with a 512 byte
323 * block size and 128 or fewer partitions, efi_first_u_lba
324 * should work out to "34"
326 vptr
->efi_first_u_lba
= nblocks
+ 1;
327 vptr
->efi_last_lba
= capacity
- 1;
328 vptr
->efi_altern_lba
= capacity
-1;
329 vptr
->efi_last_u_lba
= vptr
->efi_last_lba
- nblocks
;
331 (void) uuid_generate((uchar_t
*)&uuid
);
332 UUID_LE_CONVERT(vptr
->efi_disk_uguid
, uuid
);
337 * Read EFI - return partition number upon success.
340 efi_alloc_and_read(int fd
, struct dk_gpt
**vtoc
)
346 /* figure out the number of entries that would fit into 16K */
347 nparts
= EFI_MIN_ARRAY_SIZE
/ sizeof (efi_gpe_t
);
348 length
= (int) sizeof (struct dk_gpt
) +
349 (int) sizeof (struct dk_part
) * (nparts
- 1);
350 if ((*vtoc
= calloc(length
, 1)) == NULL
)
353 (*vtoc
)->efi_nparts
= nparts
;
354 rval
= efi_read(fd
, *vtoc
);
356 if ((rval
== VT_EINVAL
) && (*vtoc
)->efi_nparts
> nparts
) {
358 length
= (int) sizeof (struct dk_gpt
) +
359 (int) sizeof (struct dk_part
) *
360 ((*vtoc
)->efi_nparts
- 1);
361 nparts
= (*vtoc
)->efi_nparts
;
362 if ((tmp
= realloc(*vtoc
, length
)) == NULL
) {
368 rval
= efi_read(fd
, *vtoc
);
374 (void) fprintf(stderr
,
375 "read of EFI table failed, rval=%d\n", rval
);
385 efi_ioctl(int fd
, int cmd
, dk_efi_t
*dk_ioc
)
387 void *data
= dk_ioc
->dki_data
;
389 #if defined(__linux__)
394 * When the IO is not being performed in kernel as an ioctl we need
395 * to know the sector size so we can seek to the proper byte offset.
397 if (read_disk_info(fd
, &capacity
, &lbsize
) == -1) {
399 fprintf(stderr
,"unable to read disk info: %d",errno
);
409 (void) fprintf(stderr
, "DKIOCGETEFI assuming "
410 "LBA %d bytes\n", DEV_BSIZE
);
415 error
= lseek(fd
, dk_ioc
->dki_lba
* lbsize
, SEEK_SET
);
418 (void) fprintf(stderr
, "DKIOCGETEFI lseek "
419 "error: %d\n", errno
);
423 error
= read(fd
, data
, dk_ioc
->dki_length
);
426 (void) fprintf(stderr
, "DKIOCGETEFI read "
427 "error: %d\n", errno
);
431 if (error
!= dk_ioc
->dki_length
) {
433 (void) fprintf(stderr
, "DKIOCGETEFI short "
434 "read of %d bytes\n", error
);
444 (void) fprintf(stderr
, "DKIOCSETEFI unknown "
450 error
= lseek(fd
, dk_ioc
->dki_lba
* lbsize
, SEEK_SET
);
453 (void) fprintf(stderr
, "DKIOCSETEFI lseek "
454 "error: %d\n", errno
);
458 error
= write(fd
, data
, dk_ioc
->dki_length
);
461 (void) fprintf(stderr
, "DKIOCSETEFI write "
462 "error: %d\n", errno
);
466 if (error
!= dk_ioc
->dki_length
) {
468 (void) fprintf(stderr
, "DKIOCSETEFI short "
469 "write of %d bytes\n", error
);
474 /* Sync the new EFI table to disk */
479 /* Ensure any local disk cache is also flushed */
480 if (ioctl(fd
, BLKFLSBUF
, 0) == -1)
488 (void) fprintf(stderr
, "unsupported ioctl()\n");
494 dk_ioc
->dki_data_64
= (uint64_t)(uintptr_t)data
;
495 error
= ioctl(fd
, cmd
, (void *)dk_ioc
);
496 dk_ioc
->dki_data
= data
;
501 #if defined(__linux__)
508 /* Notify the kernel a devices partition table has been updated */
509 while ((error
= ioctl(fd
, BLKRRPART
)) != 0) {
511 (void) fprintf(stderr
, "the kernel failed to rescan "
512 "the partition table: %d\n", errno
);
522 check_label(int fd
, dk_efi_t
*dk_ioc
)
527 if (efi_ioctl(fd
, DKIOCGETEFI
, dk_ioc
) == -1) {
535 efi
= dk_ioc
->dki_data
;
536 if (efi
->efi_gpt_Signature
!= LE_64(EFI_SIGNATURE
)) {
538 (void) fprintf(stderr
,
539 "Bad EFI signature: 0x%llx != 0x%llx\n",
540 (long long)efi
->efi_gpt_Signature
,
541 (long long)LE_64(EFI_SIGNATURE
));
546 * check CRC of the header; the size of the header should
547 * never be larger than one block
549 crc
= efi
->efi_gpt_HeaderCRC32
;
550 efi
->efi_gpt_HeaderCRC32
= 0;
552 if (((len_t
)LE_32(efi
->efi_gpt_HeaderSize
) > dk_ioc
->dki_length
) ||
553 crc
!= LE_32(efi_crc32((unsigned char *)efi
,
554 LE_32(efi
->efi_gpt_HeaderSize
)))) {
556 (void) fprintf(stderr
,
557 "Bad EFI CRC: 0x%x != 0x%x\n",
559 LE_32(efi_crc32((unsigned char *)efi
,
560 sizeof (struct efi_gpt
))));
568 efi_read(int fd
, struct dk_gpt
*vtoc
)
575 diskaddr_t capacity
= 0;
577 struct dk_minfo disk_info
;
580 efi_gpe_t
*efi_parts
;
581 struct dk_cinfo dki_info
;
582 uint32_t user_length
;
583 boolean_t legacy_label
= B_FALSE
;
586 * get the partition number for this file descriptor.
588 if ((rval
= efi_get_info(fd
, &dki_info
)) != 0)
591 if ((strncmp(dki_info
.dki_cname
, "pseudo", 7) == 0) &&
592 (strncmp(dki_info
.dki_dname
, "md", 3) == 0)) {
594 } else if ((strncmp(dki_info
.dki_cname
, "vdc", 4) == 0) &&
595 (strncmp(dki_info
.dki_dname
, "vdc", 4) == 0)) {
597 * The controller and drive name "vdc" (virtual disk client)
598 * indicates a LDoms virtual disk.
603 /* get the LBA size */
604 if (read_disk_info(fd
, &capacity
, &lbsize
) == -1) {
606 (void) fprintf(stderr
,
607 "unable to read disk info: %d",
613 disk_info
.dki_lbsize
= lbsize
;
614 disk_info
.dki_capacity
= capacity
;
616 if (disk_info
.dki_lbsize
== 0) {
618 (void) fprintf(stderr
,
619 "efi_read: assuming LBA 512 bytes\n");
621 disk_info
.dki_lbsize
= DEV_BSIZE
;
624 * Read the EFI GPT to figure out how many partitions we need
628 if (NBLOCKS(vtoc
->efi_nparts
, disk_info
.dki_lbsize
) < 34) {
629 label_len
= EFI_MIN_ARRAY_SIZE
+ disk_info
.dki_lbsize
;
631 label_len
= vtoc
->efi_nparts
* (int) sizeof (efi_gpe_t
) +
632 disk_info
.dki_lbsize
;
633 if (label_len
% disk_info
.dki_lbsize
) {
634 /* pad to physical sector size */
635 label_len
+= disk_info
.dki_lbsize
;
636 label_len
&= ~(disk_info
.dki_lbsize
- 1);
640 if (posix_memalign((void **)&dk_ioc
.dki_data
,
641 disk_info
.dki_lbsize
, label_len
))
644 memset(dk_ioc
.dki_data
, 0, label_len
);
645 dk_ioc
.dki_length
= disk_info
.dki_lbsize
;
646 user_length
= vtoc
->efi_nparts
;
647 efi
= dk_ioc
.dki_data
;
649 dk_ioc
.dki_length
= label_len
;
650 if (efi_ioctl(fd
, DKIOCGETEFI
, &dk_ioc
) == -1) {
658 } else if ((rval
= check_label(fd
, &dk_ioc
)) == VT_EINVAL
) {
660 * No valid label here; try the alternate. Note that here
661 * we just read GPT header and save it into dk_ioc.data,
662 * Later, we will read GUID partition entry array if we
663 * can get valid GPT header.
667 * This is a workaround for legacy systems. In the past, the
668 * last sector of SCSI disk was invisible on x86 platform. At
669 * that time, backup label was saved on the next to the last
670 * sector. It is possible for users to move a disk from previous
671 * solaris system to present system. Here, we attempt to search
672 * legacy backup EFI label first.
674 dk_ioc
.dki_lba
= disk_info
.dki_capacity
- 2;
675 dk_ioc
.dki_length
= disk_info
.dki_lbsize
;
676 rval
= check_label(fd
, &dk_ioc
);
677 if (rval
== VT_EINVAL
) {
679 * we didn't find legacy backup EFI label, try to
680 * search backup EFI label in the last block.
682 dk_ioc
.dki_lba
= disk_info
.dki_capacity
- 1;
683 dk_ioc
.dki_length
= disk_info
.dki_lbsize
;
684 rval
= check_label(fd
, &dk_ioc
);
686 legacy_label
= B_TRUE
;
688 (void) fprintf(stderr
,
689 "efi_read: primary label corrupt; "
690 "using EFI backup label located on"
691 " the last block\n");
694 if ((efi_debug
) && (rval
== 0))
695 (void) fprintf(stderr
, "efi_read: primary label"
696 " corrupt; using legacy EFI backup label "
697 " located on the next to last block\n");
701 dk_ioc
.dki_lba
= LE_64(efi
->efi_gpt_PartitionEntryLBA
);
702 vtoc
->efi_flags
|= EFI_GPT_PRIMARY_CORRUPT
;
704 LE_32(efi
->efi_gpt_NumberOfPartitionEntries
);
706 * Partition tables are between backup GPT header
707 * table and ParitionEntryLBA (the starting LBA of
708 * the GUID partition entries array). Now that we
709 * already got valid GPT header and saved it in
710 * dk_ioc.dki_data, we try to get GUID partition
714 dk_ioc
.dki_data
= (efi_gpt_t
*)((char *)dk_ioc
.dki_data
715 + disk_info
.dki_lbsize
);
717 dk_ioc
.dki_length
= disk_info
.dki_capacity
- 1 -
720 dk_ioc
.dki_length
= disk_info
.dki_capacity
- 2 -
722 dk_ioc
.dki_length
*= disk_info
.dki_lbsize
;
723 if (dk_ioc
.dki_length
>
724 ((len_t
)label_len
- sizeof (*dk_ioc
.dki_data
))) {
728 * read GUID partition entry array
730 rval
= efi_ioctl(fd
, DKIOCGETEFI
, &dk_ioc
);
734 } else if (rval
== 0) {
736 dk_ioc
.dki_lba
= LE_64(efi
->efi_gpt_PartitionEntryLBA
);
738 dk_ioc
.dki_data
= (efi_gpt_t
*)((char *)dk_ioc
.dki_data
739 + disk_info
.dki_lbsize
);
740 dk_ioc
.dki_length
= label_len
- disk_info
.dki_lbsize
;
741 rval
= efi_ioctl(fd
, DKIOCGETEFI
, &dk_ioc
);
743 } else if (vdc_flag
&& rval
== VT_ERROR
&& errno
== EINVAL
) {
745 * When the device is a LDoms virtual disk, the DKIOCGETEFI
746 * ioctl can fail with EINVAL if the virtual disk backend
747 * is a ZFS volume serviced by a domain running an old version
748 * of Solaris. This is because the DKIOCGETEFI ioctl was
749 * initially incorrectly implemented for a ZFS volume and it
750 * expected the GPT and GPE to be retrieved with a single ioctl.
751 * So we try to read the GPT and the GPE using that old style
755 dk_ioc
.dki_length
= label_len
;
756 rval
= check_label(fd
, &dk_ioc
);
764 /* LINTED -- always longlong aligned */
765 efi_parts
= (efi_gpe_t
*)(((char *)efi
) + disk_info
.dki_lbsize
);
768 * Assemble this into a "dk_gpt" struct for easier
769 * digestibility by applications.
771 vtoc
->efi_version
= LE_32(efi
->efi_gpt_Revision
);
772 vtoc
->efi_nparts
= LE_32(efi
->efi_gpt_NumberOfPartitionEntries
);
773 vtoc
->efi_part_size
= LE_32(efi
->efi_gpt_SizeOfPartitionEntry
);
774 vtoc
->efi_lbasize
= disk_info
.dki_lbsize
;
775 vtoc
->efi_last_lba
= disk_info
.dki_capacity
- 1;
776 vtoc
->efi_first_u_lba
= LE_64(efi
->efi_gpt_FirstUsableLBA
);
777 vtoc
->efi_last_u_lba
= LE_64(efi
->efi_gpt_LastUsableLBA
);
778 vtoc
->efi_altern_lba
= LE_64(efi
->efi_gpt_AlternateLBA
);
779 UUID_LE_CONVERT(vtoc
->efi_disk_uguid
, efi
->efi_gpt_DiskGUID
);
782 * If the array the user passed in is too small, set the length
783 * to what it needs to be and return
785 if (user_length
< vtoc
->efi_nparts
) {
789 for (i
= 0; i
< vtoc
->efi_nparts
; i
++) {
791 UUID_LE_CONVERT(vtoc
->efi_parts
[i
].p_guid
,
792 efi_parts
[i
].efi_gpe_PartitionTypeGUID
);
795 j
< sizeof (conversion_array
)
796 / sizeof (struct uuid_to_ptag
); j
++) {
798 if (bcmp(&vtoc
->efi_parts
[i
].p_guid
,
799 &conversion_array
[j
].uuid
,
800 sizeof (struct uuid
)) == 0) {
801 vtoc
->efi_parts
[i
].p_tag
= j
;
805 if (vtoc
->efi_parts
[i
].p_tag
== V_UNASSIGNED
)
807 vtoc
->efi_parts
[i
].p_flag
=
808 LE_16(efi_parts
[i
].efi_gpe_Attributes
.PartitionAttrs
);
809 vtoc
->efi_parts
[i
].p_start
=
810 LE_64(efi_parts
[i
].efi_gpe_StartingLBA
);
811 vtoc
->efi_parts
[i
].p_size
=
812 LE_64(efi_parts
[i
].efi_gpe_EndingLBA
) -
813 vtoc
->efi_parts
[i
].p_start
+ 1;
814 for (j
= 0; j
< EFI_PART_NAME_LEN
; j
++) {
815 vtoc
->efi_parts
[i
].p_name
[j
] =
817 efi_parts
[i
].efi_gpe_PartitionName
[j
]);
820 UUID_LE_CONVERT(vtoc
->efi_parts
[i
].p_uguid
,
821 efi_parts
[i
].efi_gpe_UniquePartitionGUID
);
825 return (dki_info
.dki_partition
);
828 /* writes a "protective" MBR */
830 write_pmbr(int fd
, struct dk_gpt
*vtoc
)
835 diskaddr_t size_in_lba
;
839 len
= (vtoc
->efi_lbasize
== 0) ? sizeof (mb
) : vtoc
->efi_lbasize
;
840 if (posix_memalign((void **)&buf
, len
, len
))
844 * Preserve any boot code and disk signature if the first block is
849 dk_ioc
.dki_length
= len
;
850 /* LINTED -- always longlong aligned */
851 dk_ioc
.dki_data
= (efi_gpt_t
*)buf
;
852 if (efi_ioctl(fd
, DKIOCGETEFI
, &dk_ioc
) == -1) {
853 (void *) memcpy(&mb
, buf
, sizeof (mb
));
854 bzero(&mb
, sizeof (mb
));
855 mb
.signature
= LE_16(MBB_MAGIC
);
857 (void *) memcpy(&mb
, buf
, sizeof (mb
));
858 if (mb
.signature
!= LE_16(MBB_MAGIC
)) {
859 bzero(&mb
, sizeof (mb
));
860 mb
.signature
= LE_16(MBB_MAGIC
);
864 bzero(&mb
.parts
, sizeof (mb
.parts
));
865 cp
= (uchar_t
*)&mb
.parts
[0];
866 /* bootable or not */
868 /* beginning CHS; 0xffffff if not representable */
874 /* ending CHS; 0xffffff if not representable */
878 /* starting LBA: 1 (little endian format) by EFI definition */
883 /* ending LBA: last block on the disk (little endian format) */
884 size_in_lba
= vtoc
->efi_last_lba
;
885 if (size_in_lba
< 0xffffffff) {
886 *cp
++ = (size_in_lba
& 0x000000ff);
887 *cp
++ = (size_in_lba
& 0x0000ff00) >> 8;
888 *cp
++ = (size_in_lba
& 0x00ff0000) >> 16;
889 *cp
++ = (size_in_lba
& 0xff000000) >> 24;
897 (void *) memcpy(buf
, &mb
, sizeof (mb
));
898 /* LINTED -- always longlong aligned */
899 dk_ioc
.dki_data
= (efi_gpt_t
*)buf
;
901 dk_ioc
.dki_length
= len
;
902 if (efi_ioctl(fd
, DKIOCSETEFI
, &dk_ioc
) == -1) {
917 /* make sure the user specified something reasonable */
919 check_input(struct dk_gpt
*vtoc
)
923 diskaddr_t istart
, jstart
, isize
, jsize
, endsect
;
926 * Sanity-check the input (make sure no partitions overlap)
928 for (i
= 0; i
< vtoc
->efi_nparts
; i
++) {
929 /* It can't be unassigned and have an actual size */
930 if ((vtoc
->efi_parts
[i
].p_tag
== V_UNASSIGNED
) &&
931 (vtoc
->efi_parts
[i
].p_size
!= 0)) {
933 (void) fprintf(stderr
, "partition %d is "
934 "\"unassigned\" but has a size of %llu",
935 i
, vtoc
->efi_parts
[i
].p_size
);
939 if (vtoc
->efi_parts
[i
].p_tag
== V_UNASSIGNED
) {
940 if (uuid_is_null((uchar_t
*)&vtoc
->efi_parts
[i
].p_guid
))
942 /* we have encountered an unknown uuid */
943 vtoc
->efi_parts
[i
].p_tag
= 0xff;
945 if (vtoc
->efi_parts
[i
].p_tag
== V_RESERVED
) {
946 if (resv_part
!= -1) {
948 (void) fprintf(stderr
, "found "
949 "duplicate reserved partition "
956 if ((vtoc
->efi_parts
[i
].p_start
< vtoc
->efi_first_u_lba
) ||
957 (vtoc
->efi_parts
[i
].p_start
> vtoc
->efi_last_u_lba
)) {
959 (void) fprintf(stderr
,
960 "Partition %d starts at %llu. ",
962 vtoc
->efi_parts
[i
].p_start
);
963 (void) fprintf(stderr
,
964 "It must be between %llu and %llu.\n",
965 vtoc
->efi_first_u_lba
,
966 vtoc
->efi_last_u_lba
);
970 if ((vtoc
->efi_parts
[i
].p_start
+
971 vtoc
->efi_parts
[i
].p_size
<
972 vtoc
->efi_first_u_lba
) ||
973 (vtoc
->efi_parts
[i
].p_start
+
974 vtoc
->efi_parts
[i
].p_size
>
975 vtoc
->efi_last_u_lba
+ 1)) {
977 (void) fprintf(stderr
,
978 "Partition %d ends at %llu. ",
980 vtoc
->efi_parts
[i
].p_start
+
981 vtoc
->efi_parts
[i
].p_size
);
982 (void) fprintf(stderr
,
983 "It must be between %llu and %llu.\n",
984 vtoc
->efi_first_u_lba
,
985 vtoc
->efi_last_u_lba
);
990 for (j
= 0; j
< vtoc
->efi_nparts
; j
++) {
991 isize
= vtoc
->efi_parts
[i
].p_size
;
992 jsize
= vtoc
->efi_parts
[j
].p_size
;
993 istart
= vtoc
->efi_parts
[i
].p_start
;
994 jstart
= vtoc
->efi_parts
[j
].p_start
;
995 if ((i
!= j
) && (isize
!= 0) && (jsize
!= 0)) {
996 endsect
= jstart
+ jsize
-1;
997 if ((jstart
<= istart
) &&
998 (istart
<= endsect
)) {
1000 (void) fprintf(stderr
,
1001 "Partition %d overlaps "
1002 "partition %d.", i
, j
);
1009 /* just a warning for now */
1010 if ((resv_part
== -1) && efi_debug
) {
1011 (void) fprintf(stderr
,
1012 "no reserved partition found\n");
1018 * add all the unallocated space to the current label
1021 efi_use_whole_disk(int fd
)
1023 struct dk_gpt
*efi_label
;
1026 uint_t phy_last_slice
= 0;
1027 diskaddr_t pl_start
= 0;
1030 rval
= efi_alloc_and_read(fd
, &efi_label
);
1035 /* find the last physically non-zero partition */
1036 for (i
= 0; i
< efi_label
->efi_nparts
- 2; i
++) {
1037 if (pl_start
< efi_label
->efi_parts
[i
].p_start
) {
1038 pl_start
= efi_label
->efi_parts
[i
].p_start
;
1042 pl_size
= efi_label
->efi_parts
[phy_last_slice
].p_size
;
1045 * If alter_lba is 1, we are using the backup label.
1046 * Since we can locate the backup label by disk capacity,
1047 * there must be no unallocated space.
1049 if ((efi_label
->efi_altern_lba
== 1) || (efi_label
->efi_altern_lba
1050 >= efi_label
->efi_last_lba
)) {
1052 (void) fprintf(stderr
,
1053 "efi_use_whole_disk: requested space not found\n");
1055 efi_free(efi_label
);
1060 * If there is space between the last physically non-zero partition
1061 * and the reserved partition, just add the unallocated space to this
1062 * area. Otherwise, the unallocated space is added to the last
1063 * physically non-zero partition.
1065 if (pl_start
+ pl_size
- 1 == efi_label
->efi_last_u_lba
-
1066 EFI_MIN_RESV_SIZE
) {
1067 efi_label
->efi_parts
[phy_last_slice
].p_size
+=
1068 efi_label
->efi_last_lba
- efi_label
->efi_altern_lba
;
1072 * Move the reserved partition. There is currently no data in
1073 * here except fabricated devids (which get generated via
1074 * efi_write()). So there is no need to copy data.
1076 efi_label
->efi_parts
[efi_label
->efi_nparts
- 1].p_start
+=
1077 efi_label
->efi_last_lba
- efi_label
->efi_altern_lba
;
1078 efi_label
->efi_last_u_lba
+= efi_label
->efi_last_lba
1079 - efi_label
->efi_altern_lba
;
1081 rval
= efi_write(fd
, efi_label
);
1084 (void) fprintf(stderr
,
1085 "efi_use_whole_disk:fail to write label, rval=%d\n",
1088 efi_free(efi_label
);
1092 efi_free(efi_label
);
1098 * write EFI label and backup label
1101 efi_write(int fd
, struct dk_gpt
*vtoc
)
1105 efi_gpe_t
*efi_parts
;
1107 struct dk_cinfo dki_info
;
1111 diskaddr_t lba_backup_gpt_hdr
;
1113 if ((rval
= efi_get_info(fd
, &dki_info
)) != 0)
1116 /* check if we are dealing wih a metadevice */
1117 if ((strncmp(dki_info
.dki_cname
, "pseudo", 7) == 0) &&
1118 (strncmp(dki_info
.dki_dname
, "md", 3) == 0)) {
1122 if (check_input(vtoc
)) {
1124 * not valid; if it's a metadevice just pass it down
1125 * because SVM will do its own checking
1133 if (NBLOCKS(vtoc
->efi_nparts
, vtoc
->efi_lbasize
) < 34) {
1134 dk_ioc
.dki_length
= EFI_MIN_ARRAY_SIZE
+ vtoc
->efi_lbasize
;
1136 dk_ioc
.dki_length
= NBLOCKS(vtoc
->efi_nparts
,
1137 vtoc
->efi_lbasize
) *
1142 * the number of blocks occupied by GUID partition entry array
1144 nblocks
= dk_ioc
.dki_length
/ vtoc
->efi_lbasize
- 1;
1147 * Backup GPT header is located on the block after GUID
1148 * partition entry array. Here, we calculate the address
1149 * for backup GPT header.
1151 lba_backup_gpt_hdr
= vtoc
->efi_last_u_lba
+ 1 + nblocks
;
1152 if (posix_memalign((void **)&dk_ioc
.dki_data
,
1153 vtoc
->efi_lbasize
, dk_ioc
.dki_length
))
1156 memset(dk_ioc
.dki_data
, 0, dk_ioc
.dki_length
);
1157 efi
= dk_ioc
.dki_data
;
1159 /* stuff user's input into EFI struct */
1160 efi
->efi_gpt_Signature
= LE_64(EFI_SIGNATURE
);
1161 efi
->efi_gpt_Revision
= LE_32(vtoc
->efi_version
); /* 0x02000100 */
1162 efi
->efi_gpt_HeaderSize
= LE_32(sizeof (struct efi_gpt
));
1163 efi
->efi_gpt_Reserved1
= 0;
1164 efi
->efi_gpt_MyLBA
= LE_64(1ULL);
1165 efi
->efi_gpt_AlternateLBA
= LE_64(lba_backup_gpt_hdr
);
1166 efi
->efi_gpt_FirstUsableLBA
= LE_64(vtoc
->efi_first_u_lba
);
1167 efi
->efi_gpt_LastUsableLBA
= LE_64(vtoc
->efi_last_u_lba
);
1168 efi
->efi_gpt_PartitionEntryLBA
= LE_64(2ULL);
1169 efi
->efi_gpt_NumberOfPartitionEntries
= LE_32(vtoc
->efi_nparts
);
1170 efi
->efi_gpt_SizeOfPartitionEntry
= LE_32(sizeof (struct efi_gpe
));
1171 UUID_LE_CONVERT(efi
->efi_gpt_DiskGUID
, vtoc
->efi_disk_uguid
);
1173 /* LINTED -- always longlong aligned */
1174 efi_parts
= (efi_gpe_t
*)((char *)dk_ioc
.dki_data
+ vtoc
->efi_lbasize
);
1176 for (i
= 0; i
< vtoc
->efi_nparts
; i
++) {
1178 j
< sizeof (conversion_array
) /
1179 sizeof (struct uuid_to_ptag
); j
++) {
1181 if (vtoc
->efi_parts
[i
].p_tag
== j
) {
1183 efi_parts
[i
].efi_gpe_PartitionTypeGUID
,
1184 conversion_array
[j
].uuid
);
1189 if (j
== sizeof (conversion_array
) /
1190 sizeof (struct uuid_to_ptag
)) {
1192 * If we didn't have a matching uuid match, bail here.
1193 * Don't write a label with unknown uuid.
1196 (void) fprintf(stderr
,
1197 "Unknown uuid for p_tag %d\n",
1198 vtoc
->efi_parts
[i
].p_tag
);
1203 /* Zero's should be written for empty partitions */
1204 if (vtoc
->efi_parts
[i
].p_tag
== V_UNASSIGNED
)
1207 efi_parts
[i
].efi_gpe_StartingLBA
=
1208 LE_64(vtoc
->efi_parts
[i
].p_start
);
1209 efi_parts
[i
].efi_gpe_EndingLBA
=
1210 LE_64(vtoc
->efi_parts
[i
].p_start
+
1211 vtoc
->efi_parts
[i
].p_size
- 1);
1212 efi_parts
[i
].efi_gpe_Attributes
.PartitionAttrs
=
1213 LE_16(vtoc
->efi_parts
[i
].p_flag
);
1214 for (j
= 0; j
< EFI_PART_NAME_LEN
; j
++) {
1215 efi_parts
[i
].efi_gpe_PartitionName
[j
] =
1216 LE_16((ushort_t
)vtoc
->efi_parts
[i
].p_name
[j
]);
1218 if ((vtoc
->efi_parts
[i
].p_tag
!= V_UNASSIGNED
) &&
1219 uuid_is_null((uchar_t
*)&vtoc
->efi_parts
[i
].p_uguid
)) {
1220 (void) uuid_generate((uchar_t
*)
1221 &vtoc
->efi_parts
[i
].p_uguid
);
1223 bcopy(&vtoc
->efi_parts
[i
].p_uguid
,
1224 &efi_parts
[i
].efi_gpe_UniquePartitionGUID
,
1227 efi
->efi_gpt_PartitionEntryArrayCRC32
=
1228 LE_32(efi_crc32((unsigned char *)efi_parts
,
1229 vtoc
->efi_nparts
* (int)sizeof (struct efi_gpe
)));
1230 efi
->efi_gpt_HeaderCRC32
=
1231 LE_32(efi_crc32((unsigned char *)efi
, sizeof (struct efi_gpt
)));
1233 if (efi_ioctl(fd
, DKIOCSETEFI
, &dk_ioc
) == -1) {
1234 free(dk_ioc
.dki_data
);
1244 /* if it's a metadevice we're done */
1246 free(dk_ioc
.dki_data
);
1250 /* write backup partition array */
1251 dk_ioc
.dki_lba
= vtoc
->efi_last_u_lba
+ 1;
1252 dk_ioc
.dki_length
-= vtoc
->efi_lbasize
;
1254 dk_ioc
.dki_data
= (efi_gpt_t
*)((char *)dk_ioc
.dki_data
+
1257 if (efi_ioctl(fd
, DKIOCSETEFI
, &dk_ioc
) == -1) {
1259 * we wrote the primary label okay, so don't fail
1262 (void) fprintf(stderr
,
1263 "write of backup partitions to block %llu "
1264 "failed, errno %d\n",
1265 vtoc
->efi_last_u_lba
+ 1,
1270 * now swap MyLBA and AlternateLBA fields and write backup
1271 * partition table header
1273 dk_ioc
.dki_lba
= lba_backup_gpt_hdr
;
1274 dk_ioc
.dki_length
= vtoc
->efi_lbasize
;
1276 dk_ioc
.dki_data
= (efi_gpt_t
*)((char *)dk_ioc
.dki_data
-
1278 efi
->efi_gpt_AlternateLBA
= LE_64(1ULL);
1279 efi
->efi_gpt_MyLBA
= LE_64(lba_backup_gpt_hdr
);
1280 efi
->efi_gpt_PartitionEntryLBA
= LE_64(vtoc
->efi_last_u_lba
+ 1);
1281 efi
->efi_gpt_HeaderCRC32
= 0;
1282 efi
->efi_gpt_HeaderCRC32
=
1283 LE_32(efi_crc32((unsigned char *)dk_ioc
.dki_data
,
1284 sizeof (struct efi_gpt
)));
1286 if (efi_ioctl(fd
, DKIOCSETEFI
, &dk_ioc
) == -1) {
1288 (void) fprintf(stderr
,
1289 "write of backup header to block %llu failed, "
1295 /* write the PMBR */
1296 (void) write_pmbr(fd
, vtoc
);
1297 free(dk_ioc
.dki_data
);
1299 #if defined(__linux__)
1300 rval
= efi_rescan(fd
);
1309 efi_free(struct dk_gpt
*ptr
)
1315 * Input: File descriptor
1316 * Output: 1 if disk has an EFI label, or > 2TB with no VTOC or legacy MBR.
1324 struct extvtoc extvtoc
;
1326 if (ioctl(fd
, DKIOCGEXTVTOC
, &extvtoc
) == -1) {
1327 if (errno
== ENOTSUP
)
1329 else if (errno
== ENOTTY
) {
1330 if (ioctl(fd
, DKIOCGVTOC
, &vtoc
) == -1)
1331 if (errno
== ENOTSUP
)
1342 efi_err_check(struct dk_gpt
*vtoc
)
1346 diskaddr_t istart
, jstart
, isize
, jsize
, endsect
;
1350 * make sure no partitions overlap
1352 for (i
= 0; i
< vtoc
->efi_nparts
; i
++) {
1353 /* It can't be unassigned and have an actual size */
1354 if ((vtoc
->efi_parts
[i
].p_tag
== V_UNASSIGNED
) &&
1355 (vtoc
->efi_parts
[i
].p_size
!= 0)) {
1356 (void) fprintf(stderr
,
1357 "partition %d is \"unassigned\" but has a size "
1358 "of %llu\n", i
, vtoc
->efi_parts
[i
].p_size
);
1360 if (vtoc
->efi_parts
[i
].p_tag
== V_UNASSIGNED
) {
1363 if (vtoc
->efi_parts
[i
].p_tag
== V_RESERVED
) {
1364 if (resv_part
!= -1) {
1365 (void) fprintf(stderr
,
1366 "found duplicate reserved partition at "
1370 if (vtoc
->efi_parts
[i
].p_size
!= EFI_MIN_RESV_SIZE
)
1371 (void) fprintf(stderr
,
1372 "Warning: reserved partition size must "
1373 "be %d sectors\n", EFI_MIN_RESV_SIZE
);
1375 if ((vtoc
->efi_parts
[i
].p_start
< vtoc
->efi_first_u_lba
) ||
1376 (vtoc
->efi_parts
[i
].p_start
> vtoc
->efi_last_u_lba
)) {
1377 (void) fprintf(stderr
,
1378 "Partition %d starts at %llu\n",
1380 vtoc
->efi_parts
[i
].p_start
);
1381 (void) fprintf(stderr
,
1382 "It must be between %llu and %llu.\n",
1383 vtoc
->efi_first_u_lba
,
1384 vtoc
->efi_last_u_lba
);
1386 if ((vtoc
->efi_parts
[i
].p_start
+
1387 vtoc
->efi_parts
[i
].p_size
<
1388 vtoc
->efi_first_u_lba
) ||
1389 (vtoc
->efi_parts
[i
].p_start
+
1390 vtoc
->efi_parts
[i
].p_size
>
1391 vtoc
->efi_last_u_lba
+ 1)) {
1392 (void) fprintf(stderr
,
1393 "Partition %d ends at %llu\n",
1395 vtoc
->efi_parts
[i
].p_start
+
1396 vtoc
->efi_parts
[i
].p_size
);
1397 (void) fprintf(stderr
,
1398 "It must be between %llu and %llu.\n",
1399 vtoc
->efi_first_u_lba
,
1400 vtoc
->efi_last_u_lba
);
1403 for (j
= 0; j
< vtoc
->efi_nparts
; j
++) {
1404 isize
= vtoc
->efi_parts
[i
].p_size
;
1405 jsize
= vtoc
->efi_parts
[j
].p_size
;
1406 istart
= vtoc
->efi_parts
[i
].p_start
;
1407 jstart
= vtoc
->efi_parts
[j
].p_start
;
1408 if ((i
!= j
) && (isize
!= 0) && (jsize
!= 0)) {
1409 endsect
= jstart
+ jsize
-1;
1410 if ((jstart
<= istart
) &&
1411 (istart
<= endsect
)) {
1413 (void) fprintf(stderr
,
1414 "label error: EFI Labels do not "
1415 "support overlapping partitions\n");
1417 (void) fprintf(stderr
,
1418 "Partition %d overlaps partition "
1425 /* make sure there is a reserved partition */
1426 if (resv_part
== -1) {
1427 (void) fprintf(stderr
,
1428 "no reserved partition found\n");
1433 * We need to get information necessary to construct a *new* efi
1437 efi_auto_sense(int fd
, struct dk_gpt
**vtoc
)
1443 * Now build the default partition table
1445 if (efi_alloc_and_init(fd
, EFI_NUMPAR
, vtoc
) != 0) {
1447 (void) fprintf(stderr
, "efi_alloc_and_init failed.\n");
1452 for (i
= 0; i
< MIN((*vtoc
)->efi_nparts
, V_NUMPAR
); i
++) {
1453 (*vtoc
)->efi_parts
[i
].p_tag
= default_vtoc_map
[i
].p_tag
;
1454 (*vtoc
)->efi_parts
[i
].p_flag
= default_vtoc_map
[i
].p_flag
;
1455 (*vtoc
)->efi_parts
[i
].p_start
= 0;
1456 (*vtoc
)->efi_parts
[i
].p_size
= 0;
1459 * Make constants first
1460 * and variable partitions later
1463 /* root partition - s0 128 MB */
1464 (*vtoc
)->efi_parts
[0].p_start
= 34;
1465 (*vtoc
)->efi_parts
[0].p_size
= 262144;
1467 /* partition - s1 128 MB */
1468 (*vtoc
)->efi_parts
[1].p_start
= 262178;
1469 (*vtoc
)->efi_parts
[1].p_size
= 262144;
1471 /* partition -s2 is NOT the Backup disk */
1472 (*vtoc
)->efi_parts
[2].p_tag
= V_UNASSIGNED
;
1474 /* partition -s6 /usr partition - HOG */
1475 (*vtoc
)->efi_parts
[6].p_start
= 524322;
1476 (*vtoc
)->efi_parts
[6].p_size
= (*vtoc
)->efi_last_u_lba
- 524322
1479 /* efi reserved partition - s9 16K */
1480 (*vtoc
)->efi_parts
[8].p_start
= (*vtoc
)->efi_last_u_lba
- (1024 * 16);
1481 (*vtoc
)->efi_parts
[8].p_size
= (1024 * 16);
1482 (*vtoc
)->efi_parts
[8].p_tag
= V_RESERVED
;