1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2012 Nexenta Systems, Inc. All rights reserved.
25 * Copyright 2014 Toomas Soome <tsoome@me.com>
26 */
27
28 #include <stdio.h>
29 #include <stdlib.h>
30 #include <errno.h>
31 #include <strings.h>
32 #include <unistd.h>
33 #include <uuid/uuid.h>
34 #include <libintl.h>
35 #include <sys/types.h>
36 #include <sys/dkio.h>
37 #include <sys/vtoc.h>
38 #include <sys/mhd.h>
39 #include <sys/param.h>
40 #include <sys/dktp/fdisk.h>
41 #include <sys/efi_partition.h>
42 #include <sys/byteorder.h>
43 #include <sys/ddi.h>
44
45 static struct uuid_to_ptag {
46 struct uuid uuid;
47 } conversion_array[] = {
48 { EFI_UNUSED },
49 { EFI_BOOT },
50 { EFI_ROOT },
51 { EFI_SWAP },
52 { EFI_USR },
53 { EFI_BACKUP },
54 { 0 }, /* STAND is never used */
55 { EFI_VAR },
56 { EFI_HOME },
57 { EFI_ALTSCTR },
58 { 0 }, /* CACHE (cachefs) is never used */
59 { EFI_RESERVED },
60 { EFI_SYSTEM },
61 { EFI_LEGACY_MBR },
62 { EFI_SYMC_PUB },
63 { EFI_SYMC_CDS },
64 { EFI_MSFT_RESV },
65 { EFI_DELL_BASIC },
66 { EFI_DELL_RAID },
67 { EFI_DELL_SWAP },
68 { EFI_DELL_LVM },
69 { EFI_DELL_RESV },
70 { EFI_AAPL_HFS },
71 { EFI_AAPL_UFS },
72 { EFI_BIOS_BOOT },
73 { EFI_FREEBSD_BOOT },
74 { EFI_FREEBSD_SWAP },
75 { EFI_FREEBSD_UFS },
76 { EFI_FREEBSD_VINUM },
77 { EFI_FREEBSD_ZFS }
78 };
79
80 /*
81 * Default vtoc information for non-SVr4 partitions
82 */
83 struct dk_map2 default_vtoc_map[NDKMAP] = {
84 { V_ROOT, 0 }, /* a - 0 */
85 { V_SWAP, V_UNMNT }, /* b - 1 */
86 { V_BACKUP, V_UNMNT }, /* c - 2 */
87 { V_UNASSIGNED, 0 }, /* d - 3 */
88 { V_UNASSIGNED, 0 }, /* e - 4 */
89 { V_UNASSIGNED, 0 }, /* f - 5 */
90 { V_USR, 0 }, /* g - 6 */
91 { V_UNASSIGNED, 0 }, /* h - 7 */
92
93 #if defined(_SUNOS_VTOC_16)
94
95 #if defined(i386) || defined(__amd64)
96 { V_BOOT, V_UNMNT }, /* i - 8 */
97 { V_ALTSCTR, 0 }, /* j - 9 */
98
99 #else
100 #error No VTOC format defined.
101 #endif /* defined(i386) */
102
103 { V_UNASSIGNED, 0 }, /* k - 10 */
104 { V_UNASSIGNED, 0 }, /* l - 11 */
105 { V_UNASSIGNED, 0 }, /* m - 12 */
106 { V_UNASSIGNED, 0 }, /* n - 13 */
107 { V_UNASSIGNED, 0 }, /* o - 14 */
108 { V_UNASSIGNED, 0 }, /* p - 15 */
109 #endif /* defined(_SUNOS_VTOC_16) */
110 };
111
112 #ifdef DEBUG
113 int efi_debug = 1;
114 #else
115 int efi_debug = 0;
116 #endif
117
118 extern unsigned int efi_crc32(const unsigned char *, unsigned int);
119 static int efi_read(int, struct dk_gpt *);
120
121 static int
122 read_disk_info(int fd, diskaddr_t *capacity, uint_t *lbsize)
123 {
124 struct dk_minfo disk_info;
125
126 if ((ioctl(fd, DKIOCGMEDIAINFO, (caddr_t)&disk_info)) == -1)
127 return (errno);
128 *capacity = disk_info.dki_capacity;
129 *lbsize = disk_info.dki_lbsize;
130 return (0);
131 }
132
133 /*
134 * the number of blocks the EFI label takes up (round up to nearest
135 * block)
136 */
137 #define NBLOCKS(p, l) (1 + ((((p) * (int)sizeof (efi_gpe_t)) + \
138 ((l) - 1)) / (l)))
139 /* number of partitions -- limited by what we can malloc */
140 #define MAX_PARTS ((4294967295UL - sizeof (struct dk_gpt)) / \
141 sizeof (struct dk_part))
142
143 int
144 efi_alloc_and_init(int fd, uint32_t nparts, struct dk_gpt **vtoc)
145 {
146 diskaddr_t capacity;
147 uint_t lbsize;
148 uint_t nblocks;
149 size_t length;
150 struct dk_gpt *vptr;
151 struct uuid uuid;
152
153 if (read_disk_info(fd, &capacity, &lbsize) != 0) {
154 if (efi_debug)
155 (void) fprintf(stderr,
156 "couldn't read disk information\n");
157 return (-1);
158 }
159
160 nblocks = NBLOCKS(nparts, lbsize);
161 if ((nblocks * lbsize) < EFI_MIN_ARRAY_SIZE + lbsize) {
162 /* 16K plus one block for the GPT */
163 nblocks = EFI_MIN_ARRAY_SIZE / lbsize + 1;
164 }
165
166 if (nparts > MAX_PARTS) {
167 if (efi_debug) {
168 (void) fprintf(stderr,
169 "the maximum number of partitions supported is %lu\n",
170 MAX_PARTS);
171 }
172 return (-1);
173 }
174
175 length = sizeof (struct dk_gpt) +
176 sizeof (struct dk_part) * (nparts - 1);
177
178 if ((*vtoc = calloc(length, 1)) == NULL)
179 return (-1);
180
181 vptr = *vtoc;
182
183 vptr->efi_version = EFI_VERSION_CURRENT;
184 vptr->efi_lbasize = lbsize;
185 vptr->efi_nparts = nparts;
186 /*
187 * add one block here for the PMBR; on disks with a 512 byte
188 * block size and 128 or fewer partitions, efi_first_u_lba
189 * should work out to "34"
190 */
191 vptr->efi_first_u_lba = nblocks + 1;
192 vptr->efi_last_lba = capacity - 1;
193 vptr->efi_altern_lba = capacity -1;
194 vptr->efi_last_u_lba = vptr->efi_last_lba - nblocks;
195
196 (void) uuid_generate((uchar_t *)&uuid);
197 UUID_LE_CONVERT(vptr->efi_disk_uguid, uuid);
198 return (0);
199 }
200
201 /*
202 * Read EFI - return partition number upon success.
203 */
204 int
205 efi_alloc_and_read(int fd, struct dk_gpt **vtoc)
206 {
207 int rval;
208 uint32_t nparts;
209 int length;
210
211 /* figure out the number of entries that would fit into 16K */
212 nparts = EFI_MIN_ARRAY_SIZE / sizeof (efi_gpe_t);
213 length = (int) sizeof (struct dk_gpt) +
214 (int) sizeof (struct dk_part) * (nparts - 1);
215 if ((*vtoc = calloc(length, 1)) == NULL)
216 return (VT_ERROR);
217
218 (*vtoc)->efi_nparts = nparts;
219 rval = efi_read(fd, *vtoc);
220
221 if ((rval == VT_EINVAL) && (*vtoc)->efi_nparts > nparts) {
222 void *tmp;
223 length = (int) sizeof (struct dk_gpt) +
224 (int) sizeof (struct dk_part) *
225 ((*vtoc)->efi_nparts - 1);
226 nparts = (*vtoc)->efi_nparts;
227 if ((tmp = realloc(*vtoc, length)) == NULL) {
228 free (*vtoc);
229 *vtoc = NULL;
230 return (VT_ERROR);
231 } else {
232 *vtoc = tmp;
233 rval = efi_read(fd, *vtoc);
234 }
235 }
236
237 if (rval < 0) {
238 if (efi_debug) {
239 (void) fprintf(stderr,
240 "read of EFI table failed, rval=%d\n", rval);
241 }
242 free (*vtoc);
243 *vtoc = NULL;
244 }
245
246 return (rval);
247 }
248
249 static int
250 efi_ioctl(int fd, int cmd, dk_efi_t *dk_ioc)
251 {
252 void *data = dk_ioc->dki_data;
253 int error;
254
255 dk_ioc->dki_data_64 = (uint64_t)(uintptr_t)data;
256 error = ioctl(fd, cmd, (void *)dk_ioc);
257 dk_ioc->dki_data = data;
258
259 return (error);
260 }
261
262 static int
263 check_label(int fd, dk_efi_t *dk_ioc)
264 {
265 efi_gpt_t *efi;
266 uint_t crc;
267
268 if (efi_ioctl(fd, DKIOCGETEFI, dk_ioc) == -1) {
269 switch (errno) {
270 case EIO:
271 return (VT_EIO);
272 default:
273 return (VT_ERROR);
274 }
275 }
276 efi = dk_ioc->dki_data;
277 if (efi->efi_gpt_Signature != LE_64(EFI_SIGNATURE)) {
278 if (efi_debug)
279 (void) fprintf(stderr,
280 "Bad EFI signature: 0x%llx != 0x%llx\n",
281 (long long)efi->efi_gpt_Signature,
282 (long long)LE_64(EFI_SIGNATURE));
283 return (VT_EINVAL);
284 }
285
286 /*
287 * check CRC of the header; the size of the header should
288 * never be larger than one block
289 */
290 crc = efi->efi_gpt_HeaderCRC32;
291 efi->efi_gpt_HeaderCRC32 = 0;
292
293 if (((len_t)LE_32(efi->efi_gpt_HeaderSize) > dk_ioc->dki_length) ||
294 crc != LE_32(efi_crc32((unsigned char *)efi,
295 LE_32(efi->efi_gpt_HeaderSize)))) {
296 if (efi_debug)
297 (void) fprintf(stderr,
298 "Bad EFI CRC: 0x%x != 0x%x\n",
299 crc,
300 LE_32(efi_crc32((unsigned char *)efi,
301 sizeof (struct efi_gpt))));
302 return (VT_EINVAL);
303 }
304
305 return (0);
306 }
307
308 static int
309 efi_read(int fd, struct dk_gpt *vtoc)
310 {
311 int i, j;
312 int label_len;
313 int rval = 0;
314 int md_flag = 0;
315 int vdc_flag = 0;
316 struct dk_minfo disk_info;
317 dk_efi_t dk_ioc;
318 efi_gpt_t *efi;
319 efi_gpe_t *efi_parts;
320 struct dk_cinfo dki_info;
321 uint32_t user_length;
322 boolean_t legacy_label = B_FALSE;
323
324 /*
325 * get the partition number for this file descriptor.
326 */
327 if (ioctl(fd, DKIOCINFO, (caddr_t)&dki_info) == -1) {
328 if (efi_debug) {
329 (void) fprintf(stderr, "DKIOCINFO errno 0x%x\n", errno);
330 }
331 switch (errno) {
332 case EIO:
333 return (VT_EIO);
334 case EINVAL:
335 return (VT_EINVAL);
336 default:
337 return (VT_ERROR);
338 }
339 }
340 if ((strncmp(dki_info.dki_cname, "pseudo", 7) == 0) &&
341 (strncmp(dki_info.dki_dname, "md", 3) == 0)) {
342 md_flag++;
343 } else if ((strncmp(dki_info.dki_cname, "vdc", 4) == 0) &&
344 (strncmp(dki_info.dki_dname, "vdc", 4) == 0)) {
345 /*
346 * The controller and drive name "vdc" (virtual disk client)
347 * indicates a LDoms virtual disk.
348 */
349 vdc_flag++;
350 }
351
352 /* get the LBA size */
353 if (ioctl(fd, DKIOCGMEDIAINFO, (caddr_t)&disk_info) == -1) {
354 if (efi_debug) {
355 (void) fprintf(stderr,
356 "assuming LBA 512 bytes %d\n",
357 errno);
358 }
359 disk_info.dki_lbsize = DEV_BSIZE;
360 }
361 if (disk_info.dki_lbsize == 0) {
362 if (efi_debug) {
363 (void) fprintf(stderr,
364 "efi_read: assuming LBA 512 bytes\n");
365 }
366 disk_info.dki_lbsize = DEV_BSIZE;
367 }
368 /*
369 * Read the EFI GPT to figure out how many partitions we need
370 * to deal with.
371 */
372 dk_ioc.dki_lba = 1;
373 if (NBLOCKS(vtoc->efi_nparts, disk_info.dki_lbsize) < 34) {
374 label_len = EFI_MIN_ARRAY_SIZE + disk_info.dki_lbsize;
375 } else {
376 label_len = vtoc->efi_nparts * (int) sizeof (efi_gpe_t) +
377 disk_info.dki_lbsize;
378 if (label_len % disk_info.dki_lbsize) {
379 /* pad to physical sector size */
380 label_len += disk_info.dki_lbsize;
381 label_len &= ~(disk_info.dki_lbsize - 1);
382 }
383 }
384
385 if ((dk_ioc.dki_data = calloc(label_len, 1)) == NULL)
386 return (VT_ERROR);
387
388 dk_ioc.dki_length = disk_info.dki_lbsize;
389 user_length = vtoc->efi_nparts;
390 efi = dk_ioc.dki_data;
391 if (md_flag) {
392 dk_ioc.dki_length = label_len;
393 if (efi_ioctl(fd, DKIOCGETEFI, &dk_ioc) == -1) {
394 switch (errno) {
395 case EIO:
396 return (VT_EIO);
397 default:
398 return (VT_ERROR);
399 }
400 }
401 } else if ((rval = check_label(fd, &dk_ioc)) == VT_EINVAL) {
402 /*
403 * No valid label here; try the alternate. Note that here
404 * we just read GPT header and save it into dk_ioc.data,
405 * Later, we will read GUID partition entry array if we
406 * can get valid GPT header.
407 */
408
409 /*
410 * This is a workaround for legacy systems. In the past, the
411 * last sector of SCSI disk was invisible on x86 platform. At
412 * that time, backup label was saved on the next to the last
413 * sector. It is possible for users to move a disk from previous
414 * solaris system to present system. Here, we attempt to search
415 * legacy backup EFI label first.
416 */
417 dk_ioc.dki_lba = disk_info.dki_capacity - 2;
418 dk_ioc.dki_length = disk_info.dki_lbsize;
419 rval = check_label(fd, &dk_ioc);
420 if (rval == VT_EINVAL) {
421 /*
422 * we didn't find legacy backup EFI label, try to
423 * search backup EFI label in the last block.
424 */
425 dk_ioc.dki_lba = disk_info.dki_capacity - 1;
426 dk_ioc.dki_length = disk_info.dki_lbsize;
427 rval = check_label(fd, &dk_ioc);
428 if (rval == 0) {
429 legacy_label = B_TRUE;
430 if (efi_debug)
431 (void) fprintf(stderr,
432 "efi_read: primary label corrupt; "
433 "using EFI backup label located on"
434 " the last block\n");
435 }
436 } else {
437 if ((efi_debug) && (rval == 0))
438 (void) fprintf(stderr, "efi_read: primary label"
439 " corrupt; using legacy EFI backup label "
440 " located on the next to last block\n");
441 }
442
443 if (rval == 0) {
444 dk_ioc.dki_lba = LE_64(efi->efi_gpt_PartitionEntryLBA);
445 vtoc->efi_flags |= EFI_GPT_PRIMARY_CORRUPT;
446 vtoc->efi_nparts =
447 LE_32(efi->efi_gpt_NumberOfPartitionEntries);
448 /*
449 * Partition tables are between backup GPT header
450 * table and ParitionEntryLBA (the starting LBA of
451 * the GUID partition entries array). Now that we
452 * already got valid GPT header and saved it in
453 * dk_ioc.dki_data, we try to get GUID partition
454 * entry array here.
455 */
456 /* LINTED */
457 dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data
458 + disk_info.dki_lbsize);
459 if (legacy_label)
460 dk_ioc.dki_length = disk_info.dki_capacity - 1 -
461 dk_ioc.dki_lba;
462 else
463 dk_ioc.dki_length = disk_info.dki_capacity - 2 -
464 dk_ioc.dki_lba;
465 dk_ioc.dki_length *= disk_info.dki_lbsize;
466 if (dk_ioc.dki_length >
467 ((len_t)label_len - sizeof (*dk_ioc.dki_data))) {
468 rval = VT_EINVAL;
469 } else {
470 /*
471 * read GUID partition entry array
472 */
473 rval = efi_ioctl(fd, DKIOCGETEFI, &dk_ioc);
474 }
475 }
476
477 } else if (rval == 0) {
478
479 dk_ioc.dki_lba = LE_64(efi->efi_gpt_PartitionEntryLBA);
480 /* LINTED */
481 dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data
482 + disk_info.dki_lbsize);
483 dk_ioc.dki_length = label_len - disk_info.dki_lbsize;
484 rval = efi_ioctl(fd, DKIOCGETEFI, &dk_ioc);
485
486 } else if (vdc_flag && rval == VT_ERROR && errno == EINVAL) {
487 /*
488 * When the device is a LDoms virtual disk, the DKIOCGETEFI
489 * ioctl can fail with EINVAL if the virtual disk backend
490 * is a ZFS volume serviced by a domain running an old version
491 * of Solaris. This is because the DKIOCGETEFI ioctl was
492 * initially incorrectly implemented for a ZFS volume and it
493 * expected the GPT and GPE to be retrieved with a single ioctl.
494 * So we try to read the GPT and the GPE using that old style
495 * ioctl.
496 */
497 dk_ioc.dki_lba = 1;
498 dk_ioc.dki_length = label_len;
499 rval = check_label(fd, &dk_ioc);
500 }
501
502 if (rval < 0) {
503 free(efi);
504 return (rval);
505 }
506
507 /* LINTED -- always longlong aligned */
508 efi_parts = (efi_gpe_t *)(((char *)efi) + disk_info.dki_lbsize);
509
510 /*
511 * Assemble this into a "dk_gpt" struct for easier
512 * digestibility by applications.
513 */
514 vtoc->efi_version = LE_32(efi->efi_gpt_Revision);
515 vtoc->efi_nparts = LE_32(efi->efi_gpt_NumberOfPartitionEntries);
516 vtoc->efi_part_size = LE_32(efi->efi_gpt_SizeOfPartitionEntry);
517 vtoc->efi_lbasize = disk_info.dki_lbsize;
518 vtoc->efi_last_lba = disk_info.dki_capacity - 1;
519 vtoc->efi_first_u_lba = LE_64(efi->efi_gpt_FirstUsableLBA);
520 vtoc->efi_last_u_lba = LE_64(efi->efi_gpt_LastUsableLBA);
521 vtoc->efi_altern_lba = LE_64(efi->efi_gpt_AlternateLBA);
522 UUID_LE_CONVERT(vtoc->efi_disk_uguid, efi->efi_gpt_DiskGUID);
523
524 /*
525 * If the array the user passed in is too small, set the length
526 * to what it needs to be and return
527 */
528 if (user_length < vtoc->efi_nparts) {
529 return (VT_EINVAL);
530 }
531
532 for (i = 0; i < vtoc->efi_nparts; i++) {
533
534 UUID_LE_CONVERT(vtoc->efi_parts[i].p_guid,
535 efi_parts[i].efi_gpe_PartitionTypeGUID);
536
537 for (j = 0;
538 j < sizeof (conversion_array)
539 / sizeof (struct uuid_to_ptag); j++) {
540
541 if (bcmp(&vtoc->efi_parts[i].p_guid,
542 &conversion_array[j].uuid,
543 sizeof (struct uuid)) == 0) {
544 vtoc->efi_parts[i].p_tag = j;
545 break;
546 }
547 }
548 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED)
549 continue;
550 vtoc->efi_parts[i].p_flag =
551 LE_16(efi_parts[i].efi_gpe_Attributes.PartitionAttrs);
552 vtoc->efi_parts[i].p_start =
553 LE_64(efi_parts[i].efi_gpe_StartingLBA);
554 vtoc->efi_parts[i].p_size =
555 LE_64(efi_parts[i].efi_gpe_EndingLBA) -
556 vtoc->efi_parts[i].p_start + 1;
557 for (j = 0; j < EFI_PART_NAME_LEN; j++) {
558 vtoc->efi_parts[i].p_name[j] =
559 (uchar_t)LE_16(
560 efi_parts[i].efi_gpe_PartitionName[j]);
561 }
562
563 UUID_LE_CONVERT(vtoc->efi_parts[i].p_uguid,
564 efi_parts[i].efi_gpe_UniquePartitionGUID);
565 }
566 free(efi);
567
568 return (dki_info.dki_partition);
569 }
570
571 /* writes a "protective" MBR */
572 static int
573 write_pmbr(int fd, struct dk_gpt *vtoc)
574 {
575 dk_efi_t dk_ioc;
576 struct mboot mb;
577 uchar_t *cp;
578 diskaddr_t size_in_lba;
579 uchar_t *buf;
580 int len;
581
582 len = (vtoc->efi_lbasize == 0) ? sizeof (mb) : vtoc->efi_lbasize;
583 buf = calloc(len, 1);
584
585 /*
586 * Preserve any boot code and disk signature if the first block is
587 * already an MBR.
588 */
589 dk_ioc.dki_lba = 0;
590 dk_ioc.dki_length = len;
591 /* LINTED -- always longlong aligned */
592 dk_ioc.dki_data = (efi_gpt_t *)buf;
593 if (efi_ioctl(fd, DKIOCGETEFI, &dk_ioc) == -1) {
594 (void *) memcpy(&mb, buf, sizeof (mb));
595 bzero(&mb, sizeof (mb));
596 mb.signature = LE_16(MBB_MAGIC);
597 } else {
598 (void *) memcpy(&mb, buf, sizeof (mb));
599 if (mb.signature != LE_16(MBB_MAGIC)) {
600 bzero(&mb, sizeof (mb));
601 mb.signature = LE_16(MBB_MAGIC);
602 }
603 }
604
605 bzero(&mb.parts, sizeof (mb.parts));
606 cp = (uchar_t *)&mb.parts[0];
607 /* bootable or not */
608 *cp++ = 0;
609 /* beginning CHS; 0xffffff if not representable */
610 *cp++ = 0xff;
611 *cp++ = 0xff;
612 *cp++ = 0xff;
613 /* OS type */
614 *cp++ = EFI_PMBR;
615 /* ending CHS; 0xffffff if not representable */
616 *cp++ = 0xff;
617 *cp++ = 0xff;
618 *cp++ = 0xff;
619 /* starting LBA: 1 (little endian format) by EFI definition */
620 *cp++ = 0x01;
621 *cp++ = 0x00;
622 *cp++ = 0x00;
623 *cp++ = 0x00;
624 /* ending LBA: last block on the disk (little endian format) */
625 size_in_lba = vtoc->efi_last_lba;
626 if (size_in_lba < 0xffffffff) {
627 *cp++ = (size_in_lba & 0x000000ff);
628 *cp++ = (size_in_lba & 0x0000ff00) >> 8;
629 *cp++ = (size_in_lba & 0x00ff0000) >> 16;
630 *cp++ = (size_in_lba & 0xff000000) >> 24;
631 } else {
632 *cp++ = 0xff;
633 *cp++ = 0xff;
634 *cp++ = 0xff;
635 *cp++ = 0xff;
636 }
637
638 (void *) memcpy(buf, &mb, sizeof (mb));
639 /* LINTED -- always longlong aligned */
640 dk_ioc.dki_data = (efi_gpt_t *)buf;
641 dk_ioc.dki_lba = 0;
642 dk_ioc.dki_length = len;
643 if (efi_ioctl(fd, DKIOCSETEFI, &dk_ioc) == -1) {
644 free(buf);
645 switch (errno) {
646 case EIO:
647 return (VT_EIO);
648 case EINVAL:
649 return (VT_EINVAL);
650 default:
651 return (VT_ERROR);
652 }
653 }
654 free(buf);
655 return (0);
656 }
657
658 /* make sure the user specified something reasonable */
659 static int
660 check_input(struct dk_gpt *vtoc)
661 {
662 int resv_part = -1;
663 int i, j;
664 diskaddr_t istart, jstart, isize, jsize, endsect;
665
666 /*
667 * Sanity-check the input (make sure no partitions overlap)
668 */
669 for (i = 0; i < vtoc->efi_nparts; i++) {
670 /* It can't be unassigned and have an actual size */
671 if ((vtoc->efi_parts[i].p_tag == V_UNASSIGNED) &&
672 (vtoc->efi_parts[i].p_size != 0)) {
673 if (efi_debug) {
674 (void) fprintf(stderr,
675 "partition %d is \"unassigned\" but has a size of %llu",
676 i,
677 vtoc->efi_parts[i].p_size);
678 }
679 return (VT_EINVAL);
680 }
681 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED) {
682 if (uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid))
683 continue;
684 /* we have encountered an unknown uuid */
685 vtoc->efi_parts[i].p_tag = 0xff;
686 }
687 if (vtoc->efi_parts[i].p_tag == V_RESERVED) {
688 if (resv_part != -1) {
689 if (efi_debug) {
690 (void) fprintf(stderr,
691 "found duplicate reserved partition at %d\n",
692 i);
693 }
694 return (VT_EINVAL);
695 }
696 resv_part = i;
697 }
698 if ((vtoc->efi_parts[i].p_start < vtoc->efi_first_u_lba) ||
699 (vtoc->efi_parts[i].p_start > vtoc->efi_last_u_lba)) {
700 if (efi_debug) {
701 (void) fprintf(stderr,
702 "Partition %d starts at %llu. ",
703 i,
704 vtoc->efi_parts[i].p_start);
705 (void) fprintf(stderr,
706 "It must be between %llu and %llu.\n",
707 vtoc->efi_first_u_lba,
708 vtoc->efi_last_u_lba);
709 }
710 return (VT_EINVAL);
711 }
712 if ((vtoc->efi_parts[i].p_start +
713 vtoc->efi_parts[i].p_size <
714 vtoc->efi_first_u_lba) ||
715 (vtoc->efi_parts[i].p_start +
716 vtoc->efi_parts[i].p_size >
717 vtoc->efi_last_u_lba + 1)) {
718 if (efi_debug) {
719 (void) fprintf(stderr,
720 "Partition %d ends at %llu. ",
721 i,
722 vtoc->efi_parts[i].p_start +
723 vtoc->efi_parts[i].p_size);
724 (void) fprintf(stderr,
725 "It must be between %llu and %llu.\n",
726 vtoc->efi_first_u_lba,
727 vtoc->efi_last_u_lba);
728 }
729 return (VT_EINVAL);
730 }
731
732 for (j = 0; j < vtoc->efi_nparts; j++) {
733 isize = vtoc->efi_parts[i].p_size;
734 jsize = vtoc->efi_parts[j].p_size;
735 istart = vtoc->efi_parts[i].p_start;
736 jstart = vtoc->efi_parts[j].p_start;
737 if ((i != j) && (isize != 0) && (jsize != 0)) {
738 endsect = jstart + jsize -1;
739 if ((jstart <= istart) &&
740 (istart <= endsect)) {
741 if (efi_debug) {
742 (void) fprintf(stderr,
743 "Partition %d overlaps partition %d.",
744 i, j);
745 }
746 return (VT_EINVAL);
747 }
748 }
749 }
750 }
751 /* just a warning for now */
752 if ((resv_part == -1) && efi_debug) {
753 (void) fprintf(stderr,
754 "no reserved partition found\n");
755 }
756 return (0);
757 }
758
759 /*
760 * add all the unallocated space to the current label
761 */
762 int
763 efi_use_whole_disk(int fd)
764 {
765 struct dk_gpt *efi_label;
766 int rval;
767 int i;
768 uint_t phy_last_slice = 0;
769 diskaddr_t pl_start = 0;
770 diskaddr_t pl_size;
771
772 rval = efi_alloc_and_read(fd, &efi_label);
773 if (rval < 0) {
774 return (rval);
775 }
776
777 /* find the last physically non-zero partition */
778 for (i = 0; i < efi_label->efi_nparts - 2; i ++) {
779 if (pl_start < efi_label->efi_parts[i].p_start) {
780 pl_start = efi_label->efi_parts[i].p_start;
781 phy_last_slice = i;
782 }
783 }
784 pl_size = efi_label->efi_parts[phy_last_slice].p_size;
785
786 /*
787 * If alter_lba is 1, we are using the backup label.
788 * Since we can locate the backup label by disk capacity,
789 * there must be no unallocated space.
790 */
791 if ((efi_label->efi_altern_lba == 1) || (efi_label->efi_altern_lba
792 >= efi_label->efi_last_lba)) {
793 if (efi_debug) {
794 (void) fprintf(stderr,
795 "efi_use_whole_disk: requested space not found\n");
796 }
797 efi_free(efi_label);
798 return (VT_ENOSPC);
799 }
800
801 /*
802 * If there is space between the last physically non-zero partition
803 * and the reserved partition, just add the unallocated space to this
804 * area. Otherwise, the unallocated space is added to the last
805 * physically non-zero partition.
806 */
807 if (pl_start + pl_size - 1 == efi_label->efi_last_u_lba -
808 EFI_MIN_RESV_SIZE) {
809 efi_label->efi_parts[phy_last_slice].p_size +=
810 efi_label->efi_last_lba - efi_label->efi_altern_lba;
811 }
812
813 /*
814 * Move the reserved partition. There is currently no data in
815 * here except fabricated devids (which get generated via
816 * efi_write()). So there is no need to copy data.
817 */
818 efi_label->efi_parts[efi_label->efi_nparts - 1].p_start +=
819 efi_label->efi_last_lba - efi_label->efi_altern_lba;
820 efi_label->efi_last_u_lba += efi_label->efi_last_lba
821 - efi_label->efi_altern_lba;
822
823 rval = efi_write(fd, efi_label);
824 if (rval < 0) {
825 if (efi_debug) {
826 (void) fprintf(stderr,
827 "efi_use_whole_disk:fail to write label, rval=%d\n",
828 rval);
829 }
830 efi_free(efi_label);
831 return (rval);
832 }
833
834 efi_free(efi_label);
835 return (0);
836 }
837
838
839 /*
840 * write EFI label and backup label
841 */
842 int
843 efi_write(int fd, struct dk_gpt *vtoc)
844 {
845 dk_efi_t dk_ioc;
846 efi_gpt_t *efi;
847 efi_gpe_t *efi_parts;
848 int i, j;
849 struct dk_cinfo dki_info;
850 int md_flag = 0;
851 int nblocks;
852 diskaddr_t lba_backup_gpt_hdr;
853
854 if (ioctl(fd, DKIOCINFO, (caddr_t)&dki_info) == -1) {
855 if (efi_debug)
856 (void) fprintf(stderr, "DKIOCINFO errno 0x%x\n", errno);
857 switch (errno) {
858 case EIO:
859 return (VT_EIO);
860 case EINVAL:
861 return (VT_EINVAL);
862 default:
863 return (VT_ERROR);
864 }
865 }
866
867 /* check if we are dealing wih a metadevice */
868 if ((strncmp(dki_info.dki_cname, "pseudo", 7) == 0) &&
869 (strncmp(dki_info.dki_dname, "md", 3) == 0)) {
870 md_flag = 1;
871 }
872
873 if (check_input(vtoc)) {
874 /*
875 * not valid; if it's a metadevice just pass it down
876 * because SVM will do its own checking
877 */
878 if (md_flag == 0) {
879 return (VT_EINVAL);
880 }
881 }
882
883 dk_ioc.dki_lba = 1;
884 if (NBLOCKS(vtoc->efi_nparts, vtoc->efi_lbasize) < 34) {
885 dk_ioc.dki_length = EFI_MIN_ARRAY_SIZE + vtoc->efi_lbasize;
886 } else {
887 dk_ioc.dki_length = NBLOCKS(vtoc->efi_nparts,
888 vtoc->efi_lbasize) *
889 vtoc->efi_lbasize;
890 }
891
892 /*
893 * the number of blocks occupied by GUID partition entry array
894 */
895 nblocks = dk_ioc.dki_length / vtoc->efi_lbasize - 1;
896
897 /*
898 * Backup GPT header is located on the block after GUID
899 * partition entry array. Here, we calculate the address
900 * for backup GPT header.
901 */
902 lba_backup_gpt_hdr = vtoc->efi_last_u_lba + 1 + nblocks;
903 if ((dk_ioc.dki_data = calloc(dk_ioc.dki_length, 1)) == NULL)
904 return (VT_ERROR);
905
906 efi = dk_ioc.dki_data;
907
908 /* stuff user's input into EFI struct */
909 efi->efi_gpt_Signature = LE_64(EFI_SIGNATURE);
910 efi->efi_gpt_Revision = LE_32(vtoc->efi_version); /* 0x02000100 */
911 efi->efi_gpt_HeaderSize = LE_32(sizeof (struct efi_gpt));
912 efi->efi_gpt_Reserved1 = 0;
913 efi->efi_gpt_MyLBA = LE_64(1ULL);
914 efi->efi_gpt_AlternateLBA = LE_64(lba_backup_gpt_hdr);
915 efi->efi_gpt_FirstUsableLBA = LE_64(vtoc->efi_first_u_lba);
916 efi->efi_gpt_LastUsableLBA = LE_64(vtoc->efi_last_u_lba);
917 efi->efi_gpt_PartitionEntryLBA = LE_64(2ULL);
918 efi->efi_gpt_NumberOfPartitionEntries = LE_32(vtoc->efi_nparts);
919 efi->efi_gpt_SizeOfPartitionEntry = LE_32(sizeof (struct efi_gpe));
920 UUID_LE_CONVERT(efi->efi_gpt_DiskGUID, vtoc->efi_disk_uguid);
921
922 /* LINTED -- always longlong aligned */
923 efi_parts = (efi_gpe_t *)((char *)dk_ioc.dki_data + vtoc->efi_lbasize);
924
925 for (i = 0; i < vtoc->efi_nparts; i++) {
926 for (j = 0;
927 j < sizeof (conversion_array) /
928 sizeof (struct uuid_to_ptag); j++) {
929
930 if (vtoc->efi_parts[i].p_tag == j) {
931 UUID_LE_CONVERT(
932 efi_parts[i].efi_gpe_PartitionTypeGUID,
933 conversion_array[j].uuid);
934 break;
935 }
936 }
937
938 if (j == sizeof (conversion_array) /
939 sizeof (struct uuid_to_ptag)) {
940 /*
941 * If we didn't have a matching uuid match, bail here.
942 * Don't write a label with unknown uuid.
943 */
944 if (efi_debug) {
945 (void) fprintf(stderr,
946 "Unknown uuid for p_tag %d\n",
947 vtoc->efi_parts[i].p_tag);
948 }
949 return (VT_EINVAL);
950 }
951
952 efi_parts[i].efi_gpe_StartingLBA =
953 LE_64(vtoc->efi_parts[i].p_start);
954 efi_parts[i].efi_gpe_EndingLBA =
955 LE_64(vtoc->efi_parts[i].p_start +
956 vtoc->efi_parts[i].p_size - 1);
957 efi_parts[i].efi_gpe_Attributes.PartitionAttrs =
958 LE_16(vtoc->efi_parts[i].p_flag);
959 for (j = 0; j < EFI_PART_NAME_LEN; j++) {
960 efi_parts[i].efi_gpe_PartitionName[j] =
961 LE_16((ushort_t)vtoc->efi_parts[i].p_name[j]);
962 }
963 if ((vtoc->efi_parts[i].p_tag != V_UNASSIGNED) &&
964 uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_uguid)) {
965 (void) uuid_generate((uchar_t *)
966 &vtoc->efi_parts[i].p_uguid);
967 }
968 bcopy(&vtoc->efi_parts[i].p_uguid,
969 &efi_parts[i].efi_gpe_UniquePartitionGUID,
970 sizeof (uuid_t));
971 }
972 efi->efi_gpt_PartitionEntryArrayCRC32 =
973 LE_32(efi_crc32((unsigned char *)efi_parts,
974 vtoc->efi_nparts * (int)sizeof (struct efi_gpe)));
975 efi->efi_gpt_HeaderCRC32 =
976 LE_32(efi_crc32((unsigned char *)efi, sizeof (struct efi_gpt)));
977
978 if (efi_ioctl(fd, DKIOCSETEFI, &dk_ioc) == -1) {
979 free(dk_ioc.dki_data);
980 switch (errno) {
981 case EIO:
982 return (VT_EIO);
983 case EINVAL:
984 return (VT_EINVAL);
985 default:
986 return (VT_ERROR);
987 }
988 }
989 /* if it's a metadevice we're done */
990 if (md_flag) {
991 free(dk_ioc.dki_data);
992 return (0);
993 }
994
995 /* write backup partition array */
996 dk_ioc.dki_lba = vtoc->efi_last_u_lba + 1;
997 dk_ioc.dki_length -= vtoc->efi_lbasize;
998 /* LINTED */
999 dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data +
1000 vtoc->efi_lbasize);
1001
1002 if (efi_ioctl(fd, DKIOCSETEFI, &dk_ioc) == -1) {
1003 /*
1004 * we wrote the primary label okay, so don't fail
1005 */
1006 if (efi_debug) {
1007 (void) fprintf(stderr,
1008 "write of backup partitions to block %llu "
1009 "failed, errno %d\n",
1010 vtoc->efi_last_u_lba + 1,
1011 errno);
1012 }
1013 }
1014 /*
1015 * now swap MyLBA and AlternateLBA fields and write backup
1016 * partition table header
1017 */
1018 dk_ioc.dki_lba = lba_backup_gpt_hdr;
1019 dk_ioc.dki_length = vtoc->efi_lbasize;
1020 /* LINTED */
1021 dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data -
1022 vtoc->efi_lbasize);
1023 efi->efi_gpt_AlternateLBA = LE_64(1ULL);
1024 efi->efi_gpt_MyLBA = LE_64(lba_backup_gpt_hdr);
1025 efi->efi_gpt_PartitionEntryLBA = LE_64(vtoc->efi_last_u_lba + 1);
1026 efi->efi_gpt_HeaderCRC32 = 0;
1027 efi->efi_gpt_HeaderCRC32 =
1028 LE_32(efi_crc32((unsigned char *)dk_ioc.dki_data,
1029 sizeof (struct efi_gpt)));
1030
1031 if (efi_ioctl(fd, DKIOCSETEFI, &dk_ioc) == -1) {
1032 if (efi_debug) {
1033 (void) fprintf(stderr,
1034 "write of backup header to block %llu failed, "
1035 "errno %d\n",
1036 lba_backup_gpt_hdr,
1037 errno);
1038 }
1039 }
1040 /* write the PMBR */
1041 (void) write_pmbr(fd, vtoc);
1042 free(dk_ioc.dki_data);
1043 return (0);
1044 }
1045
1046 void
1047 efi_free(struct dk_gpt *ptr)
1048 {
1049 free(ptr);
1050 }
1051
1052 /*
1053 * Input: File descriptor
1054 * Output: 1 if disk has an EFI label, or > 2TB with no VTOC or legacy MBR.
1055 * Otherwise 0.
1056 */
1057 int
1058 efi_type(int fd)
1059 {
1060 struct vtoc vtoc;
1061 struct extvtoc extvtoc;
1062
1063 if (ioctl(fd, DKIOCGEXTVTOC, &extvtoc) == -1) {
1064 if (errno == ENOTSUP)
1065 return (1);
1066 else if (errno == ENOTTY) {
1067 if (ioctl(fd, DKIOCGVTOC, &vtoc) == -1)
1068 if (errno == ENOTSUP)
1069 return (1);
1070 }
1071 }
1072 return (0);
1073 }
1074
1075 void
1076 efi_err_check(struct dk_gpt *vtoc)
1077 {
1078 int resv_part = -1;
1079 int i, j;
1080 diskaddr_t istart, jstart, isize, jsize, endsect;
1081 int overlap = 0;
1082
1083 /*
1084 * make sure no partitions overlap
1085 */
1086 for (i = 0; i < vtoc->efi_nparts; i++) {
1087 /* It can't be unassigned and have an actual size */
1088 if ((vtoc->efi_parts[i].p_tag == V_UNASSIGNED) &&
1089 (vtoc->efi_parts[i].p_size != 0)) {
1090 (void) fprintf(stderr,
1091 "partition %d is \"unassigned\" but has a size "
1092 "of %llu\n", i, vtoc->efi_parts[i].p_size);
1093 }
1094 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED) {
1095 continue;
1096 }
1097 if (vtoc->efi_parts[i].p_tag == V_RESERVED) {
1098 if (resv_part != -1) {
1099 (void) fprintf(stderr,
1100 "found duplicate reserved partition at "
1101 "%d\n", i);
1102 }
1103 resv_part = i;
1104 if (vtoc->efi_parts[i].p_size != EFI_MIN_RESV_SIZE)
1105 (void) fprintf(stderr,
1106 "Warning: reserved partition size must "
1107 "be %d sectors\n", EFI_MIN_RESV_SIZE);
1108 }
1109 if ((vtoc->efi_parts[i].p_start < vtoc->efi_first_u_lba) ||
1110 (vtoc->efi_parts[i].p_start > vtoc->efi_last_u_lba)) {
1111 (void) fprintf(stderr,
1112 "Partition %d starts at %llu\n",
1113 i,
1114 vtoc->efi_parts[i].p_start);
1115 (void) fprintf(stderr,
1116 "It must be between %llu and %llu.\n",
1117 vtoc->efi_first_u_lba,
1118 vtoc->efi_last_u_lba);
1119 }
1120 if ((vtoc->efi_parts[i].p_start +
1121 vtoc->efi_parts[i].p_size <
1122 vtoc->efi_first_u_lba) ||
1123 (vtoc->efi_parts[i].p_start +
1124 vtoc->efi_parts[i].p_size >
1125 vtoc->efi_last_u_lba + 1)) {
1126 (void) fprintf(stderr,
1127 "Partition %d ends at %llu\n",
1128 i,
1129 vtoc->efi_parts[i].p_start +
1130 vtoc->efi_parts[i].p_size);
1131 (void) fprintf(stderr,
1132 "It must be between %llu and %llu.\n",
1133 vtoc->efi_first_u_lba,
1134 vtoc->efi_last_u_lba);
1135 }
1136
1137 for (j = 0; j < vtoc->efi_nparts; j++) {
1138 isize = vtoc->efi_parts[i].p_size;
1139 jsize = vtoc->efi_parts[j].p_size;
1140 istart = vtoc->efi_parts[i].p_start;
1141 jstart = vtoc->efi_parts[j].p_start;
1142 if ((i != j) && (isize != 0) && (jsize != 0)) {
1143 endsect = jstart + jsize -1;
1144 if ((jstart <= istart) &&
1145 (istart <= endsect)) {
1146 if (!overlap) {
1147 (void) fprintf(stderr,
1148 "label error: EFI Labels do not "
1149 "support overlapping partitions\n");
1150 }
1151 (void) fprintf(stderr,
1152 "Partition %d overlaps partition "
1153 "%d.\n", i, j);
1154 overlap = 1;
1155 }
1156 }
1157 }
1158 }
1159 /* make sure there is a reserved partition */
1160 if (resv_part == -1) {
1161 (void) fprintf(stderr,
1162 "no reserved partition found\n");
1163 }
1164 }
1165
1166 /*
1167 * We need to get information necessary to construct a *new* efi
1168 * label type
1169 */
1170 int
1171 efi_auto_sense(int fd, struct dk_gpt **vtoc)
1172 {
1173
1174 int i;
1175
1176 /*
1177 * Now build the default partition table
1178 */
1179 if (efi_alloc_and_init(fd, EFI_NUMPAR, vtoc) != 0) {
1180 if (efi_debug) {
1181 (void) fprintf(stderr, "efi_alloc_and_init failed.\n");
1182 }
1183 return (-1);
1184 }
1185
1186 for (i = 0; i < min((*vtoc)->efi_nparts, V_NUMPAR); i++) {
1187 (*vtoc)->efi_parts[i].p_tag = default_vtoc_map[i].p_tag;
1188 (*vtoc)->efi_parts[i].p_flag = default_vtoc_map[i].p_flag;
1189 (*vtoc)->efi_parts[i].p_start = 0;
1190 (*vtoc)->efi_parts[i].p_size = 0;
1191 }
1192 /*
1193 * Make constants first
1194 * and variable partitions later
1195 */
1196
1197 /* root partition - s0 128 MB */
1198 (*vtoc)->efi_parts[0].p_start = 34;
1199 (*vtoc)->efi_parts[0].p_size = 262144;
1200
1201 /* partition - s1 128 MB */
1202 (*vtoc)->efi_parts[1].p_start = 262178;
1203 (*vtoc)->efi_parts[1].p_size = 262144;
1204
1205 /* partition -s2 is NOT the Backup disk */
1206 (*vtoc)->efi_parts[2].p_tag = V_UNASSIGNED;
1207
1208 /* partition -s6 /usr partition - HOG */
1209 (*vtoc)->efi_parts[6].p_start = 524322;
1210 (*vtoc)->efi_parts[6].p_size = (*vtoc)->efi_last_u_lba - 524322
1211 - (1024 * 16);
1212
1213 /* efi reserved partition - s9 16K */
1214 (*vtoc)->efi_parts[8].p_start = (*vtoc)->efi_last_u_lba - (1024 * 16);
1215 (*vtoc)->efi_parts[8].p_size = (1024 * 16);
1216 (*vtoc)->efi_parts[8].p_tag = V_RESERVED;
1217 return (0);
1218 }