1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2012 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2018 by Delphix. All rights reserved.
26 */
27
28 #include <stdio.h>
29 #include <stdlib.h>
30 #include <errno.h>
31 #include <string.h>
32 #include <strings.h>
33 #include <unistd.h>
34 #include <uuid/uuid.h>
35 #include <zlib.h>
36 #include <libintl.h>
37 #include <sys/types.h>
38 #include <sys/dkio.h>
39 #include <sys/vtoc.h>
40 #include <sys/mhd.h>
41 #include <sys/param.h>
42 #include <sys/dktp/fdisk.h>
43 #include <sys/efi_partition.h>
44 #include <sys/byteorder.h>
45 #include <sys/vdev_disk.h>
46 #include <linux/fs.h>
47 #include <linux/blkpg.h>
48
49 static struct uuid_to_ptag {
50 struct uuid uuid;
51 } conversion_array[] = {
52 { EFI_UNUSED },
53 { EFI_BOOT },
54 { EFI_ROOT },
55 { EFI_SWAP },
56 { EFI_USR },
57 { EFI_BACKUP },
58 { EFI_UNUSED }, /* STAND is never used */
59 { EFI_VAR },
60 { EFI_HOME },
61 { EFI_ALTSCTR },
62 { EFI_UNUSED }, /* CACHE (cachefs) is never used */
63 { EFI_RESERVED },
64 { EFI_SYSTEM },
65 { EFI_LEGACY_MBR },
66 { EFI_SYMC_PUB },
67 { EFI_SYMC_CDS },
68 { EFI_MSFT_RESV },
69 { EFI_DELL_BASIC },
70 { EFI_DELL_RAID },
71 { EFI_DELL_SWAP },
72 { EFI_DELL_LVM },
73 { EFI_DELL_RESV },
74 { EFI_AAPL_HFS },
75 { EFI_AAPL_UFS },
76 { EFI_FREEBSD_BOOT },
77 { EFI_FREEBSD_SWAP },
78 { EFI_FREEBSD_UFS },
79 { EFI_FREEBSD_VINUM },
80 { EFI_FREEBSD_ZFS },
81 { EFI_BIOS_BOOT },
82 { EFI_INTC_RS },
83 { EFI_SNE_BOOT },
84 { EFI_LENOVO_BOOT },
85 { EFI_MSFT_LDMM },
86 { EFI_MSFT_LDMD },
87 { EFI_MSFT_RE },
88 { EFI_IBM_GPFS },
89 { EFI_MSFT_STORAGESPACES },
90 { EFI_HPQ_DATA },
91 { EFI_HPQ_SVC },
92 { EFI_RHT_DATA },
93 { EFI_RHT_HOME },
94 { EFI_RHT_SRV },
95 { EFI_RHT_DMCRYPT },
96 { EFI_RHT_LUKS },
97 { EFI_FREEBSD_DISKLABEL },
98 { EFI_AAPL_RAID },
99 { EFI_AAPL_RAIDOFFLINE },
100 { EFI_AAPL_BOOT },
101 { EFI_AAPL_LABEL },
102 { EFI_AAPL_TVRECOVERY },
103 { EFI_AAPL_CORESTORAGE },
104 { EFI_NETBSD_SWAP },
105 { EFI_NETBSD_FFS },
106 { EFI_NETBSD_LFS },
107 { EFI_NETBSD_RAID },
108 { EFI_NETBSD_CAT },
109 { EFI_NETBSD_CRYPT },
110 { EFI_GOOG_KERN },
111 { EFI_GOOG_ROOT },
112 { EFI_GOOG_RESV },
113 { EFI_HAIKU_BFS },
114 { EFI_MIDNIGHTBSD_BOOT },
115 { EFI_MIDNIGHTBSD_DATA },
116 { EFI_MIDNIGHTBSD_SWAP },
117 { EFI_MIDNIGHTBSD_UFS },
118 { EFI_MIDNIGHTBSD_VINUM },
119 { EFI_MIDNIGHTBSD_ZFS },
120 { EFI_CEPH_JOURNAL },
121 { EFI_CEPH_DMCRYPTJOURNAL },
122 { EFI_CEPH_OSD },
123 { EFI_CEPH_DMCRYPTOSD },
124 { EFI_CEPH_CREATE },
125 { EFI_CEPH_DMCRYPTCREATE },
126 { EFI_OPENBSD_DISKLABEL },
127 { EFI_BBRY_QNX },
128 { EFI_BELL_PLAN9 },
129 { EFI_VMW_KCORE },
130 { EFI_VMW_VMFS },
131 { EFI_VMW_RESV },
132 { EFI_RHT_ROOTX86 },
133 { EFI_RHT_ROOTAMD64 },
134 { EFI_RHT_ROOTARM },
135 { EFI_RHT_ROOTARM64 },
136 { EFI_ACRONIS_SECUREZONE },
137 { EFI_ONIE_BOOT },
138 { EFI_ONIE_CONFIG },
139 { EFI_IBM_PPRPBOOT },
140 { EFI_FREEDESKTOP_BOOT }
141 };
142
143 /*
144 * Default vtoc information for non-SVr4 partitions
145 */
146 struct dk_map2 default_vtoc_map[NDKMAP] = {
147 { V_ROOT, 0 }, /* a - 0 */
148 { V_SWAP, V_UNMNT }, /* b - 1 */
149 { V_BACKUP, V_UNMNT }, /* c - 2 */
150 { V_UNASSIGNED, 0 }, /* d - 3 */
151 { V_UNASSIGNED, 0 }, /* e - 4 */
152 { V_UNASSIGNED, 0 }, /* f - 5 */
153 { V_USR, 0 }, /* g - 6 */
154 { V_UNASSIGNED, 0 }, /* h - 7 */
155
156 #if defined(_SUNOS_VTOC_16)
157
158 #if defined(i386) || defined(__amd64) || defined(__arm) || \
159 defined(__powerpc) || defined(__sparc) || defined(__s390__) || \
160 defined(__mips__) || defined(__rv64g__)
161 { V_BOOT, V_UNMNT }, /* i - 8 */
162 { V_ALTSCTR, 0 }, /* j - 9 */
163
164 #else
165 #error No VTOC format defined.
166 #endif /* defined(i386) */
167
168 { V_UNASSIGNED, 0 }, /* k - 10 */
169 { V_UNASSIGNED, 0 }, /* l - 11 */
170 { V_UNASSIGNED, 0 }, /* m - 12 */
171 { V_UNASSIGNED, 0 }, /* n - 13 */
172 { V_UNASSIGNED, 0 }, /* o - 14 */
173 { V_UNASSIGNED, 0 }, /* p - 15 */
174 #endif /* defined(_SUNOS_VTOC_16) */
175 };
176
177 int efi_debug = 0;
178
179 static int efi_read(int, struct dk_gpt *);
180
181 /*
182 * Return a 32-bit CRC of the contents of the buffer. Pre-and-post
183 * one's conditioning will be handled by crc32() internally.
184 */
185 static uint32_t
efi_crc32(const unsigned char * buf,unsigned int size)186 efi_crc32(const unsigned char *buf, unsigned int size)
187 {
188 uint32_t crc = crc32(0, Z_NULL, 0);
189
190 crc = crc32(crc, buf, size);
191
192 return (crc);
193 }
194
195 static int
read_disk_info(int fd,diskaddr_t * capacity,uint_t * lbsize)196 read_disk_info(int fd, diskaddr_t *capacity, uint_t *lbsize)
197 {
198 int sector_size;
199 unsigned long long capacity_size;
200
201 if (ioctl(fd, BLKSSZGET, §or_size) < 0)
202 return (-1);
203
204 if (ioctl(fd, BLKGETSIZE64, &capacity_size) < 0)
205 return (-1);
206
207 *lbsize = (uint_t)sector_size;
208 *capacity = (diskaddr_t)(capacity_size / sector_size);
209
210 return (0);
211 }
212
213 /*
214 * Return back the device name associated with the file descriptor. The
215 * caller is responsible for freeing the memory associated with the
216 * returned string.
217 */
218 static char *
efi_get_devname(int fd)219 efi_get_devname(int fd)
220 {
221 char *path;
222 char *dev_name;
223
224 path = calloc(1, PATH_MAX);
225 if (path == NULL)
226 return (NULL);
227
228 /*
229 * The libefi API only provides the open fd and not the file path.
230 * To handle this realpath(3) is used to resolve the block device
231 * name from /proc/self/fd/<fd>.
232 */
233 (void) sprintf(path, "/proc/self/fd/%d", fd);
234 dev_name = realpath(path, NULL);
235 free(path);
236 return (dev_name);
237 }
238
239 static int
efi_get_info(int fd,struct dk_cinfo * dki_info)240 efi_get_info(int fd, struct dk_cinfo *dki_info)
241 {
242 char *dev_path;
243 int rval = 0;
244
245 memset(dki_info, 0, sizeof (*dki_info));
246
247 /*
248 * The simplest way to get the partition number under linux is
249 * to parse it out of the /dev/<disk><partition> block device name.
250 * The kernel creates this using the partition number when it
251 * populates /dev/ so it may be trusted. The tricky bit here is
252 * that the naming convention is based on the block device type.
253 * So we need to take this in to account when parsing out the
254 * partition information. Aside from the partition number we collect
255 * some additional device info.
256 */
257 dev_path = efi_get_devname(fd);
258 if (dev_path == NULL)
259 goto error;
260
261 if ((strncmp(dev_path, "/dev/sd", 7) == 0)) {
262 strcpy(dki_info->dki_cname, "sd");
263 dki_info->dki_ctype = DKC_SCSI_CCS;
264 rval = sscanf(dev_path, "/dev/%[a-zA-Z]%hu",
265 dki_info->dki_dname,
266 &dki_info->dki_partition);
267 } else if ((strncmp(dev_path, "/dev/hd", 7) == 0)) {
268 strcpy(dki_info->dki_cname, "hd");
269 dki_info->dki_ctype = DKC_DIRECT;
270 rval = sscanf(dev_path, "/dev/%[a-zA-Z]%hu",
271 dki_info->dki_dname,
272 &dki_info->dki_partition);
273 } else if ((strncmp(dev_path, "/dev/md", 7) == 0)) {
274 strcpy(dki_info->dki_cname, "pseudo");
275 dki_info->dki_ctype = DKC_MD;
276 strcpy(dki_info->dki_dname, "md");
277 rval = sscanf(dev_path, "/dev/md%[0-9]p%hu",
278 dki_info->dki_dname + 2,
279 &dki_info->dki_partition);
280 } else if ((strncmp(dev_path, "/dev/vd", 7) == 0)) {
281 strcpy(dki_info->dki_cname, "vd");
282 dki_info->dki_ctype = DKC_MD;
283 rval = sscanf(dev_path, "/dev/%[a-zA-Z]%hu",
284 dki_info->dki_dname,
285 &dki_info->dki_partition);
286 } else if ((strncmp(dev_path, "/dev/xvd", 8) == 0)) {
287 strcpy(dki_info->dki_cname, "xvd");
288 dki_info->dki_ctype = DKC_MD;
289 rval = sscanf(dev_path, "/dev/%[a-zA-Z]%hu",
290 dki_info->dki_dname,
291 &dki_info->dki_partition);
292 } else if ((strncmp(dev_path, "/dev/zd", 7) == 0)) {
293 strcpy(dki_info->dki_cname, "zd");
294 dki_info->dki_ctype = DKC_MD;
295 strcpy(dki_info->dki_dname, "zd");
296 rval = sscanf(dev_path, "/dev/zd%[0-9]p%hu",
297 dki_info->dki_dname + 2,
298 &dki_info->dki_partition);
299 } else if ((strncmp(dev_path, "/dev/dm-", 8) == 0)) {
300 strcpy(dki_info->dki_cname, "pseudo");
301 dki_info->dki_ctype = DKC_VBD;
302 strcpy(dki_info->dki_dname, "dm-");
303 rval = sscanf(dev_path, "/dev/dm-%[0-9]p%hu",
304 dki_info->dki_dname + 3,
305 &dki_info->dki_partition);
306 } else if ((strncmp(dev_path, "/dev/ram", 8) == 0)) {
307 strcpy(dki_info->dki_cname, "pseudo");
308 dki_info->dki_ctype = DKC_PCMCIA_MEM;
309 strcpy(dki_info->dki_dname, "ram");
310 rval = sscanf(dev_path, "/dev/ram%[0-9]p%hu",
311 dki_info->dki_dname + 3,
312 &dki_info->dki_partition);
313 } else if ((strncmp(dev_path, "/dev/loop", 9) == 0)) {
314 strcpy(dki_info->dki_cname, "pseudo");
315 dki_info->dki_ctype = DKC_VBD;
316 strcpy(dki_info->dki_dname, "loop");
317 rval = sscanf(dev_path, "/dev/loop%[0-9]p%hu",
318 dki_info->dki_dname + 4,
319 &dki_info->dki_partition);
320 } else if ((strncmp(dev_path, "/dev/nvme", 9) == 0)) {
321 strcpy(dki_info->dki_cname, "nvme");
322 dki_info->dki_ctype = DKC_SCSI_CCS;
323 strcpy(dki_info->dki_dname, "nvme");
324 (void) sscanf(dev_path, "/dev/nvme%[0-9]",
325 dki_info->dki_dname + 4);
326 size_t controller_length = strlen(
327 dki_info->dki_dname);
328 strcpy(dki_info->dki_dname + controller_length,
329 "n");
330 rval = sscanf(dev_path,
331 "/dev/nvme%*[0-9]n%[0-9]p%hu",
332 dki_info->dki_dname + controller_length + 1,
333 &dki_info->dki_partition);
334 } else {
335 strcpy(dki_info->dki_dname, "unknown");
336 strcpy(dki_info->dki_cname, "unknown");
337 dki_info->dki_ctype = DKC_UNKNOWN;
338 }
339
340 switch (rval) {
341 case 0:
342 errno = EINVAL;
343 goto error;
344 case 1:
345 dki_info->dki_partition = 0;
346 }
347
348 free(dev_path);
349
350 return (0);
351 error:
352 if (efi_debug)
353 (void) fprintf(stderr, "DKIOCINFO errno 0x%x\n", errno);
354
355 switch (errno) {
356 case EIO:
357 return (VT_EIO);
358 case EINVAL:
359 return (VT_EINVAL);
360 default:
361 return (VT_ERROR);
362 }
363 }
364
365 /*
366 * the number of blocks the EFI label takes up (round up to nearest
367 * block)
368 */
369 #define NBLOCKS(p, l) (1 + ((((p) * (int)sizeof (efi_gpe_t)) + \
370 ((l) - 1)) / (l)))
371 /* number of partitions -- limited by what we can malloc */
372 #define MAX_PARTS ((4294967295UL - sizeof (struct dk_gpt)) / \
373 sizeof (struct dk_part))
374
375 int
efi_alloc_and_init(int fd,uint32_t nparts,struct dk_gpt ** vtoc)376 efi_alloc_and_init(int fd, uint32_t nparts, struct dk_gpt **vtoc)
377 {
378 diskaddr_t capacity = 0;
379 uint_t lbsize = 0;
380 uint_t nblocks;
381 size_t length;
382 struct dk_gpt *vptr;
383 struct uuid uuid;
384 struct dk_cinfo dki_info;
385
386 if (read_disk_info(fd, &capacity, &lbsize) != 0)
387 return (-1);
388
389 if (efi_get_info(fd, &dki_info) != 0)
390 return (-1);
391
392 if (dki_info.dki_partition != 0)
393 return (-1);
394
395 if ((dki_info.dki_ctype == DKC_PCMCIA_MEM) ||
396 (dki_info.dki_ctype == DKC_VBD) ||
397 (dki_info.dki_ctype == DKC_UNKNOWN))
398 return (-1);
399
400 nblocks = NBLOCKS(nparts, lbsize);
401 if ((nblocks * lbsize) < EFI_MIN_ARRAY_SIZE + lbsize) {
402 /* 16K plus one block for the GPT */
403 nblocks = EFI_MIN_ARRAY_SIZE / lbsize + 1;
404 }
405
406 if (nparts > MAX_PARTS) {
407 if (efi_debug) {
408 (void) fprintf(stderr,
409 "the maximum number of partitions supported is %lu\n",
410 MAX_PARTS);
411 }
412 return (-1);
413 }
414
415 length = sizeof (struct dk_gpt) +
416 sizeof (struct dk_part) * (nparts - 1);
417
418 vptr = calloc(1, length);
419 if (vptr == NULL)
420 return (-1);
421
422 *vtoc = vptr;
423
424 vptr->efi_version = EFI_VERSION_CURRENT;
425 vptr->efi_lbasize = lbsize;
426 vptr->efi_nparts = nparts;
427 /*
428 * add one block here for the PMBR; on disks with a 512 byte
429 * block size and 128 or fewer partitions, efi_first_u_lba
430 * should work out to "34"
431 */
432 vptr->efi_first_u_lba = nblocks + 1;
433 vptr->efi_last_lba = capacity - 1;
434 vptr->efi_altern_lba = capacity -1;
435 vptr->efi_last_u_lba = vptr->efi_last_lba - nblocks;
436
437 (void) uuid_generate((uchar_t *)&uuid);
438 UUID_LE_CONVERT(vptr->efi_disk_uguid, uuid);
439 return (0);
440 }
441
442 /*
443 * Read EFI - return partition number upon success.
444 */
445 int
efi_alloc_and_read(int fd,struct dk_gpt ** vtoc)446 efi_alloc_and_read(int fd, struct dk_gpt **vtoc)
447 {
448 int rval;
449 uint32_t nparts;
450 int length;
451 struct dk_gpt *vptr;
452
453 /* figure out the number of entries that would fit into 16K */
454 nparts = EFI_MIN_ARRAY_SIZE / sizeof (efi_gpe_t);
455 length = (int) sizeof (struct dk_gpt) +
456 (int) sizeof (struct dk_part) * (nparts - 1);
457 vptr = calloc(1, length);
458
459 if (vptr == NULL)
460 return (VT_ERROR);
461
462 vptr->efi_nparts = nparts;
463 rval = efi_read(fd, vptr);
464
465 if ((rval == VT_EINVAL) && vptr->efi_nparts > nparts) {
466 void *tmp;
467 length = (int) sizeof (struct dk_gpt) +
468 (int) sizeof (struct dk_part) * (vptr->efi_nparts - 1);
469 nparts = vptr->efi_nparts;
470 if ((tmp = realloc(vptr, length)) == NULL) {
471 /* cppcheck-suppress doubleFree */
472 free(vptr);
473 *vtoc = NULL;
474 return (VT_ERROR);
475 } else {
476 vptr = tmp;
477 rval = efi_read(fd, vptr);
478 }
479 }
480
481 if (rval < 0) {
482 if (efi_debug) {
483 (void) fprintf(stderr,
484 "read of EFI table failed, rval=%d\n", rval);
485 }
486 free(vptr);
487 *vtoc = NULL;
488 } else {
489 *vtoc = vptr;
490 }
491
492 return (rval);
493 }
494
495 static int
efi_ioctl(int fd,int cmd,dk_efi_t * dk_ioc)496 efi_ioctl(int fd, int cmd, dk_efi_t *dk_ioc)
497 {
498 void *data = dk_ioc->dki_data;
499 int error;
500 diskaddr_t capacity;
501 uint_t lbsize;
502
503 /*
504 * When the IO is not being performed in kernel as an ioctl we need
505 * to know the sector size so we can seek to the proper byte offset.
506 */
507 if (read_disk_info(fd, &capacity, &lbsize) == -1) {
508 if (efi_debug)
509 fprintf(stderr, "unable to read disk info: %d", errno);
510
511 errno = EIO;
512 return (-1);
513 }
514
515 switch (cmd) {
516 case DKIOCGETEFI:
517 if (lbsize == 0) {
518 if (efi_debug)
519 (void) fprintf(stderr, "DKIOCGETEFI assuming "
520 "LBA %d bytes\n", DEV_BSIZE);
521
522 lbsize = DEV_BSIZE;
523 }
524
525 error = lseek(fd, dk_ioc->dki_lba * lbsize, SEEK_SET);
526 if (error == -1) {
527 if (efi_debug)
528 (void) fprintf(stderr, "DKIOCGETEFI lseek "
529 "error: %d\n", errno);
530 return (error);
531 }
532
533 error = read(fd, data, dk_ioc->dki_length);
534 if (error == -1) {
535 if (efi_debug)
536 (void) fprintf(stderr, "DKIOCGETEFI read "
537 "error: %d\n", errno);
538 return (error);
539 }
540
541 if (error != dk_ioc->dki_length) {
542 if (efi_debug)
543 (void) fprintf(stderr, "DKIOCGETEFI short "
544 "read of %d bytes\n", error);
545 errno = EIO;
546 return (-1);
547 }
548 error = 0;
549 break;
550
551 case DKIOCSETEFI:
552 if (lbsize == 0) {
553 if (efi_debug)
554 (void) fprintf(stderr, "DKIOCSETEFI unknown "
555 "LBA size\n");
556 errno = EIO;
557 return (-1);
558 }
559
560 error = lseek(fd, dk_ioc->dki_lba * lbsize, SEEK_SET);
561 if (error == -1) {
562 if (efi_debug)
563 (void) fprintf(stderr, "DKIOCSETEFI lseek "
564 "error: %d\n", errno);
565 return (error);
566 }
567
568 error = write(fd, data, dk_ioc->dki_length);
569 if (error == -1) {
570 if (efi_debug)
571 (void) fprintf(stderr, "DKIOCSETEFI write "
572 "error: %d\n", errno);
573 return (error);
574 }
575
576 if (error != dk_ioc->dki_length) {
577 if (efi_debug)
578 (void) fprintf(stderr, "DKIOCSETEFI short "
579 "write of %d bytes\n", error);
580 errno = EIO;
581 return (-1);
582 }
583
584 /* Sync the new EFI table to disk */
585 error = fsync(fd);
586 if (error == -1)
587 return (error);
588
589 /* Ensure any local disk cache is also flushed */
590 if (ioctl(fd, BLKFLSBUF, 0) == -1)
591 return (error);
592
593 error = 0;
594 break;
595
596 default:
597 if (efi_debug)
598 (void) fprintf(stderr, "unsupported ioctl()\n");
599
600 errno = EIO;
601 return (-1);
602 }
603
604 return (error);
605 }
606
607 int
efi_rescan(int fd)608 efi_rescan(int fd)
609 {
610 int retry = 10;
611 int error;
612
613 /* Notify the kernel a devices partition table has been updated */
614 while ((error = ioctl(fd, BLKRRPART)) != 0) {
615 if ((--retry == 0) || (errno != EBUSY)) {
616 (void) fprintf(stderr, "the kernel failed to rescan "
617 "the partition table: %d\n", errno);
618 return (-1);
619 }
620 usleep(50000);
621 }
622
623 return (0);
624 }
625
626 static int
check_label(int fd,dk_efi_t * dk_ioc)627 check_label(int fd, dk_efi_t *dk_ioc)
628 {
629 efi_gpt_t *efi;
630 uint_t crc;
631
632 if (efi_ioctl(fd, DKIOCGETEFI, dk_ioc) == -1) {
633 switch (errno) {
634 case EIO:
635 return (VT_EIO);
636 default:
637 return (VT_ERROR);
638 }
639 }
640 efi = dk_ioc->dki_data;
641 if (efi->efi_gpt_Signature != LE_64(EFI_SIGNATURE)) {
642 if (efi_debug)
643 (void) fprintf(stderr,
644 "Bad EFI signature: 0x%llx != 0x%llx\n",
645 (long long)efi->efi_gpt_Signature,
646 (long long)LE_64(EFI_SIGNATURE));
647 return (VT_EINVAL);
648 }
649
650 /*
651 * check CRC of the header; the size of the header should
652 * never be larger than one block
653 */
654 crc = efi->efi_gpt_HeaderCRC32;
655 efi->efi_gpt_HeaderCRC32 = 0;
656 len_t headerSize = (len_t)LE_32(efi->efi_gpt_HeaderSize);
657
658 if (headerSize < EFI_MIN_LABEL_SIZE || headerSize > EFI_LABEL_SIZE) {
659 if (efi_debug)
660 (void) fprintf(stderr,
661 "Invalid EFI HeaderSize %llu. Assuming %d.\n",
662 headerSize, EFI_MIN_LABEL_SIZE);
663 }
664
665 if ((headerSize > dk_ioc->dki_length) ||
666 crc != LE_32(efi_crc32((unsigned char *)efi, headerSize))) {
667 if (efi_debug)
668 (void) fprintf(stderr,
669 "Bad EFI CRC: 0x%x != 0x%x\n",
670 crc, LE_32(efi_crc32((unsigned char *)efi,
671 headerSize)));
672 return (VT_EINVAL);
673 }
674
675 return (0);
676 }
677
678 static int
efi_read(int fd,struct dk_gpt * vtoc)679 efi_read(int fd, struct dk_gpt *vtoc)
680 {
681 int i, j;
682 int label_len;
683 int rval = 0;
684 int md_flag = 0;
685 int vdc_flag = 0;
686 diskaddr_t capacity = 0;
687 uint_t lbsize = 0;
688 struct dk_minfo disk_info;
689 dk_efi_t dk_ioc;
690 efi_gpt_t *efi;
691 efi_gpe_t *efi_parts;
692 struct dk_cinfo dki_info;
693 uint32_t user_length;
694 boolean_t legacy_label = B_FALSE;
695
696 /*
697 * get the partition number for this file descriptor.
698 */
699 if ((rval = efi_get_info(fd, &dki_info)) != 0)
700 return (rval);
701
702 if ((strncmp(dki_info.dki_cname, "pseudo", 7) == 0) &&
703 (strncmp(dki_info.dki_dname, "md", 3) == 0)) {
704 md_flag++;
705 } else if ((strncmp(dki_info.dki_cname, "vdc", 4) == 0) &&
706 (strncmp(dki_info.dki_dname, "vdc", 4) == 0)) {
707 /*
708 * The controller and drive name "vdc" (virtual disk client)
709 * indicates a LDoms virtual disk.
710 */
711 vdc_flag++;
712 }
713
714 /* get the LBA size */
715 if (read_disk_info(fd, &capacity, &lbsize) == -1) {
716 if (efi_debug) {
717 (void) fprintf(stderr,
718 "unable to read disk info: %d",
719 errno);
720 }
721 return (VT_EINVAL);
722 }
723
724 disk_info.dki_lbsize = lbsize;
725 disk_info.dki_capacity = capacity;
726
727 if (disk_info.dki_lbsize == 0) {
728 if (efi_debug) {
729 (void) fprintf(stderr,
730 "efi_read: assuming LBA 512 bytes\n");
731 }
732 disk_info.dki_lbsize = DEV_BSIZE;
733 }
734 /*
735 * Read the EFI GPT to figure out how many partitions we need
736 * to deal with.
737 */
738 dk_ioc.dki_lba = 1;
739 if (NBLOCKS(vtoc->efi_nparts, disk_info.dki_lbsize) < 34) {
740 label_len = EFI_MIN_ARRAY_SIZE + disk_info.dki_lbsize;
741 } else {
742 label_len = vtoc->efi_nparts * (int) sizeof (efi_gpe_t) +
743 disk_info.dki_lbsize;
744 if (label_len % disk_info.dki_lbsize) {
745 /* pad to physical sector size */
746 label_len += disk_info.dki_lbsize;
747 label_len &= ~(disk_info.dki_lbsize - 1);
748 }
749 }
750
751 if (posix_memalign((void **)&dk_ioc.dki_data,
752 disk_info.dki_lbsize, label_len))
753 return (VT_ERROR);
754
755 memset(dk_ioc.dki_data, 0, label_len);
756 dk_ioc.dki_length = disk_info.dki_lbsize;
757 user_length = vtoc->efi_nparts;
758 efi = dk_ioc.dki_data;
759 if (md_flag) {
760 dk_ioc.dki_length = label_len;
761 if (efi_ioctl(fd, DKIOCGETEFI, &dk_ioc) == -1) {
762 switch (errno) {
763 case EIO:
764 return (VT_EIO);
765 default:
766 return (VT_ERROR);
767 }
768 }
769 } else if ((rval = check_label(fd, &dk_ioc)) == VT_EINVAL) {
770 /*
771 * No valid label here; try the alternate. Note that here
772 * we just read GPT header and save it into dk_ioc.data,
773 * Later, we will read GUID partition entry array if we
774 * can get valid GPT header.
775 */
776
777 /*
778 * This is a workaround for legacy systems. In the past, the
779 * last sector of SCSI disk was invisible on x86 platform. At
780 * that time, backup label was saved on the next to the last
781 * sector. It is possible for users to move a disk from previous
782 * solaris system to present system. Here, we attempt to search
783 * legacy backup EFI label first.
784 */
785 dk_ioc.dki_lba = disk_info.dki_capacity - 2;
786 dk_ioc.dki_length = disk_info.dki_lbsize;
787 rval = check_label(fd, &dk_ioc);
788 if (rval == VT_EINVAL) {
789 /*
790 * we didn't find legacy backup EFI label, try to
791 * search backup EFI label in the last block.
792 */
793 dk_ioc.dki_lba = disk_info.dki_capacity - 1;
794 dk_ioc.dki_length = disk_info.dki_lbsize;
795 rval = check_label(fd, &dk_ioc);
796 if (rval == 0) {
797 legacy_label = B_TRUE;
798 if (efi_debug)
799 (void) fprintf(stderr,
800 "efi_read: primary label corrupt; "
801 "using EFI backup label located on"
802 " the last block\n");
803 }
804 } else {
805 if ((efi_debug) && (rval == 0))
806 (void) fprintf(stderr, "efi_read: primary label"
807 " corrupt; using legacy EFI backup label "
808 " located on the next to last block\n");
809 }
810
811 if (rval == 0) {
812 dk_ioc.dki_lba = LE_64(efi->efi_gpt_PartitionEntryLBA);
813 vtoc->efi_flags |= EFI_GPT_PRIMARY_CORRUPT;
814 vtoc->efi_nparts =
815 LE_32(efi->efi_gpt_NumberOfPartitionEntries);
816 /*
817 * Partition tables are between backup GPT header
818 * table and ParitionEntryLBA (the starting LBA of
819 * the GUID partition entries array). Now that we
820 * already got valid GPT header and saved it in
821 * dk_ioc.dki_data, we try to get GUID partition
822 * entry array here.
823 */
824 /* LINTED */
825 dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data
826 + disk_info.dki_lbsize);
827 if (legacy_label)
828 dk_ioc.dki_length = disk_info.dki_capacity - 1 -
829 dk_ioc.dki_lba;
830 else
831 dk_ioc.dki_length = disk_info.dki_capacity - 2 -
832 dk_ioc.dki_lba;
833 dk_ioc.dki_length *= disk_info.dki_lbsize;
834 if (dk_ioc.dki_length >
835 ((len_t)label_len - sizeof (*dk_ioc.dki_data))) {
836 rval = VT_EINVAL;
837 } else {
838 /*
839 * read GUID partition entry array
840 */
841 rval = efi_ioctl(fd, DKIOCGETEFI, &dk_ioc);
842 }
843 }
844
845 } else if (rval == 0) {
846
847 dk_ioc.dki_lba = LE_64(efi->efi_gpt_PartitionEntryLBA);
848 /* LINTED */
849 dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data
850 + disk_info.dki_lbsize);
851 dk_ioc.dki_length = label_len - disk_info.dki_lbsize;
852 rval = efi_ioctl(fd, DKIOCGETEFI, &dk_ioc);
853
854 } else if (vdc_flag && rval == VT_ERROR && errno == EINVAL) {
855 /*
856 * When the device is a LDoms virtual disk, the DKIOCGETEFI
857 * ioctl can fail with EINVAL if the virtual disk backend
858 * is a ZFS volume serviced by a domain running an old version
859 * of Solaris. This is because the DKIOCGETEFI ioctl was
860 * initially incorrectly implemented for a ZFS volume and it
861 * expected the GPT and GPE to be retrieved with a single ioctl.
862 * So we try to read the GPT and the GPE using that old style
863 * ioctl.
864 */
865 dk_ioc.dki_lba = 1;
866 dk_ioc.dki_length = label_len;
867 rval = check_label(fd, &dk_ioc);
868 }
869
870 if (rval < 0) {
871 free(efi);
872 return (rval);
873 }
874
875 /* LINTED -- always longlong aligned */
876 efi_parts = (efi_gpe_t *)(((char *)efi) + disk_info.dki_lbsize);
877
878 /*
879 * Assemble this into a "dk_gpt" struct for easier
880 * digestibility by applications.
881 */
882 vtoc->efi_version = LE_32(efi->efi_gpt_Revision);
883 vtoc->efi_nparts = LE_32(efi->efi_gpt_NumberOfPartitionEntries);
884 vtoc->efi_part_size = LE_32(efi->efi_gpt_SizeOfPartitionEntry);
885 vtoc->efi_lbasize = disk_info.dki_lbsize;
886 vtoc->efi_last_lba = disk_info.dki_capacity - 1;
887 vtoc->efi_first_u_lba = LE_64(efi->efi_gpt_FirstUsableLBA);
888 vtoc->efi_last_u_lba = LE_64(efi->efi_gpt_LastUsableLBA);
889 vtoc->efi_altern_lba = LE_64(efi->efi_gpt_AlternateLBA);
890 UUID_LE_CONVERT(vtoc->efi_disk_uguid, efi->efi_gpt_DiskGUID);
891
892 /*
893 * If the array the user passed in is too small, set the length
894 * to what it needs to be and return
895 */
896 if (user_length < vtoc->efi_nparts) {
897 return (VT_EINVAL);
898 }
899
900 for (i = 0; i < vtoc->efi_nparts; i++) {
901
902 UUID_LE_CONVERT(vtoc->efi_parts[i].p_guid,
903 efi_parts[i].efi_gpe_PartitionTypeGUID);
904
905 for (j = 0;
906 j < sizeof (conversion_array)
907 / sizeof (struct uuid_to_ptag); j++) {
908
909 if (bcmp(&vtoc->efi_parts[i].p_guid,
910 &conversion_array[j].uuid,
911 sizeof (struct uuid)) == 0) {
912 vtoc->efi_parts[i].p_tag = j;
913 break;
914 }
915 }
916 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED)
917 continue;
918 vtoc->efi_parts[i].p_flag =
919 LE_16(efi_parts[i].efi_gpe_Attributes.PartitionAttrs);
920 vtoc->efi_parts[i].p_start =
921 LE_64(efi_parts[i].efi_gpe_StartingLBA);
922 vtoc->efi_parts[i].p_size =
923 LE_64(efi_parts[i].efi_gpe_EndingLBA) -
924 vtoc->efi_parts[i].p_start + 1;
925 for (j = 0; j < EFI_PART_NAME_LEN; j++) {
926 vtoc->efi_parts[i].p_name[j] =
927 (uchar_t)LE_16(
928 efi_parts[i].efi_gpe_PartitionName[j]);
929 }
930
931 UUID_LE_CONVERT(vtoc->efi_parts[i].p_uguid,
932 efi_parts[i].efi_gpe_UniquePartitionGUID);
933 }
934 free(efi);
935
936 return (dki_info.dki_partition);
937 }
938
939 /* writes a "protective" MBR */
940 static int
write_pmbr(int fd,struct dk_gpt * vtoc)941 write_pmbr(int fd, struct dk_gpt *vtoc)
942 {
943 dk_efi_t dk_ioc;
944 struct mboot mb;
945 uchar_t *cp;
946 diskaddr_t size_in_lba;
947 uchar_t *buf;
948 int len;
949
950 len = (vtoc->efi_lbasize == 0) ? sizeof (mb) : vtoc->efi_lbasize;
951 if (posix_memalign((void **)&buf, len, len))
952 return (VT_ERROR);
953
954 /*
955 * Preserve any boot code and disk signature if the first block is
956 * already an MBR.
957 */
958 memset(buf, 0, len);
959 dk_ioc.dki_lba = 0;
960 dk_ioc.dki_length = len;
961 /* LINTED -- always longlong aligned */
962 dk_ioc.dki_data = (efi_gpt_t *)buf;
963 if (efi_ioctl(fd, DKIOCGETEFI, &dk_ioc) == -1) {
964 (void) memcpy(&mb, buf, sizeof (mb));
965 bzero(&mb, sizeof (mb));
966 mb.signature = LE_16(MBB_MAGIC);
967 } else {
968 (void) memcpy(&mb, buf, sizeof (mb));
969 if (mb.signature != LE_16(MBB_MAGIC)) {
970 bzero(&mb, sizeof (mb));
971 mb.signature = LE_16(MBB_MAGIC);
972 }
973 }
974
975 bzero(&mb.parts, sizeof (mb.parts));
976 cp = (uchar_t *)&mb.parts[0];
977 /* bootable or not */
978 *cp++ = 0;
979 /* beginning CHS; 0xffffff if not representable */
980 *cp++ = 0xff;
981 *cp++ = 0xff;
982 *cp++ = 0xff;
983 /* OS type */
984 *cp++ = EFI_PMBR;
985 /* ending CHS; 0xffffff if not representable */
986 *cp++ = 0xff;
987 *cp++ = 0xff;
988 *cp++ = 0xff;
989 /* starting LBA: 1 (little endian format) by EFI definition */
990 *cp++ = 0x01;
991 *cp++ = 0x00;
992 *cp++ = 0x00;
993 *cp++ = 0x00;
994 /* ending LBA: last block on the disk (little endian format) */
995 size_in_lba = vtoc->efi_last_lba;
996 if (size_in_lba < 0xffffffff) {
997 *cp++ = (size_in_lba & 0x000000ff);
998 *cp++ = (size_in_lba & 0x0000ff00) >> 8;
999 *cp++ = (size_in_lba & 0x00ff0000) >> 16;
1000 *cp++ = (size_in_lba & 0xff000000) >> 24;
1001 } else {
1002 *cp++ = 0xff;
1003 *cp++ = 0xff;
1004 *cp++ = 0xff;
1005 *cp++ = 0xff;
1006 }
1007
1008 (void) memcpy(buf, &mb, sizeof (mb));
1009 /* LINTED -- always longlong aligned */
1010 dk_ioc.dki_data = (efi_gpt_t *)buf;
1011 dk_ioc.dki_lba = 0;
1012 dk_ioc.dki_length = len;
1013 if (efi_ioctl(fd, DKIOCSETEFI, &dk_ioc) == -1) {
1014 free(buf);
1015 switch (errno) {
1016 case EIO:
1017 return (VT_EIO);
1018 case EINVAL:
1019 return (VT_EINVAL);
1020 default:
1021 return (VT_ERROR);
1022 }
1023 }
1024 free(buf);
1025 return (0);
1026 }
1027
1028 /* make sure the user specified something reasonable */
1029 static int
check_input(struct dk_gpt * vtoc)1030 check_input(struct dk_gpt *vtoc)
1031 {
1032 int resv_part = -1;
1033 int i, j;
1034 diskaddr_t istart, jstart, isize, jsize, endsect;
1035
1036 /*
1037 * Sanity-check the input (make sure no partitions overlap)
1038 */
1039 for (i = 0; i < vtoc->efi_nparts; i++) {
1040 /* It can't be unassigned and have an actual size */
1041 if ((vtoc->efi_parts[i].p_tag == V_UNASSIGNED) &&
1042 (vtoc->efi_parts[i].p_size != 0)) {
1043 if (efi_debug) {
1044 (void) fprintf(stderr, "partition %d is "
1045 "\"unassigned\" but has a size of %llu",
1046 i, vtoc->efi_parts[i].p_size);
1047 }
1048 return (VT_EINVAL);
1049 }
1050 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED) {
1051 if (uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid))
1052 continue;
1053 /* we have encountered an unknown uuid */
1054 vtoc->efi_parts[i].p_tag = 0xff;
1055 }
1056 if (vtoc->efi_parts[i].p_tag == V_RESERVED) {
1057 if (resv_part != -1) {
1058 if (efi_debug) {
1059 (void) fprintf(stderr, "found "
1060 "duplicate reserved partition "
1061 "at %d\n", i);
1062 }
1063 return (VT_EINVAL);
1064 }
1065 resv_part = i;
1066 }
1067 if ((vtoc->efi_parts[i].p_start < vtoc->efi_first_u_lba) ||
1068 (vtoc->efi_parts[i].p_start > vtoc->efi_last_u_lba)) {
1069 if (efi_debug) {
1070 (void) fprintf(stderr,
1071 "Partition %d starts at %llu. ",
1072 i,
1073 vtoc->efi_parts[i].p_start);
1074 (void) fprintf(stderr,
1075 "It must be between %llu and %llu.\n",
1076 vtoc->efi_first_u_lba,
1077 vtoc->efi_last_u_lba);
1078 }
1079 return (VT_EINVAL);
1080 }
1081 if ((vtoc->efi_parts[i].p_start +
1082 vtoc->efi_parts[i].p_size <
1083 vtoc->efi_first_u_lba) ||
1084 (vtoc->efi_parts[i].p_start +
1085 vtoc->efi_parts[i].p_size >
1086 vtoc->efi_last_u_lba + 1)) {
1087 if (efi_debug) {
1088 (void) fprintf(stderr,
1089 "Partition %d ends at %llu. ",
1090 i,
1091 vtoc->efi_parts[i].p_start +
1092 vtoc->efi_parts[i].p_size);
1093 (void) fprintf(stderr,
1094 "It must be between %llu and %llu.\n",
1095 vtoc->efi_first_u_lba,
1096 vtoc->efi_last_u_lba);
1097 }
1098 return (VT_EINVAL);
1099 }
1100
1101 for (j = 0; j < vtoc->efi_nparts; j++) {
1102 isize = vtoc->efi_parts[i].p_size;
1103 jsize = vtoc->efi_parts[j].p_size;
1104 istart = vtoc->efi_parts[i].p_start;
1105 jstart = vtoc->efi_parts[j].p_start;
1106 if ((i != j) && (isize != 0) && (jsize != 0)) {
1107 endsect = jstart + jsize -1;
1108 if ((jstart <= istart) &&
1109 (istart <= endsect)) {
1110 if (efi_debug) {
1111 (void) fprintf(stderr,
1112 "Partition %d overlaps "
1113 "partition %d.", i, j);
1114 }
1115 return (VT_EINVAL);
1116 }
1117 }
1118 }
1119 }
1120 /* just a warning for now */
1121 if ((resv_part == -1) && efi_debug) {
1122 (void) fprintf(stderr,
1123 "no reserved partition found\n");
1124 }
1125 return (0);
1126 }
1127
1128 static int
call_blkpg_ioctl(int fd,int command,diskaddr_t start,diskaddr_t size,uint_t pno)1129 call_blkpg_ioctl(int fd, int command, diskaddr_t start,
1130 diskaddr_t size, uint_t pno)
1131 {
1132 struct blkpg_ioctl_arg ioctl_arg;
1133 struct blkpg_partition linux_part;
1134 memset(&linux_part, 0, sizeof (linux_part));
1135
1136 char *path = efi_get_devname(fd);
1137 if (path == NULL) {
1138 (void) fprintf(stderr, "failed to retrieve device name\n");
1139 return (VT_EINVAL);
1140 }
1141
1142 linux_part.start = start;
1143 linux_part.length = size;
1144 linux_part.pno = pno;
1145 snprintf(linux_part.devname, BLKPG_DEVNAMELTH - 1, "%s%u", path, pno);
1146 linux_part.devname[BLKPG_DEVNAMELTH - 1] = '\0';
1147 free(path);
1148
1149 ioctl_arg.op = command;
1150 ioctl_arg.flags = 0;
1151 ioctl_arg.datalen = sizeof (struct blkpg_partition);
1152 ioctl_arg.data = &linux_part;
1153
1154 return (ioctl(fd, BLKPG, &ioctl_arg));
1155 }
1156
1157 /*
1158 * add all the unallocated space to the current label
1159 */
1160 int
efi_use_whole_disk(int fd)1161 efi_use_whole_disk(int fd)
1162 {
1163 struct dk_gpt *efi_label = NULL;
1164 int rval;
1165 int i;
1166 uint_t resv_index = 0, data_index = 0;
1167 diskaddr_t resv_start = 0, data_start = 0;
1168 diskaddr_t data_size, limit, difference;
1169 boolean_t sync_needed = B_FALSE;
1170 uint_t nblocks;
1171
1172 rval = efi_alloc_and_read(fd, &efi_label);
1173 if (rval < 0) {
1174 if (efi_label != NULL)
1175 efi_free(efi_label);
1176 return (rval);
1177 }
1178
1179 /*
1180 * Find the last physically non-zero partition.
1181 * This should be the reserved partition.
1182 */
1183 for (i = 0; i < efi_label->efi_nparts; i ++) {
1184 if (resv_start < efi_label->efi_parts[i].p_start) {
1185 resv_start = efi_label->efi_parts[i].p_start;
1186 resv_index = i;
1187 }
1188 }
1189
1190 /*
1191 * Find the last physically non-zero partition before that.
1192 * This is the data partition.
1193 */
1194 for (i = 0; i < resv_index; i ++) {
1195 if (data_start < efi_label->efi_parts[i].p_start) {
1196 data_start = efi_label->efi_parts[i].p_start;
1197 data_index = i;
1198 }
1199 }
1200 data_size = efi_label->efi_parts[data_index].p_size;
1201
1202 /*
1203 * See the "efi_alloc_and_init" function for more information
1204 * about where this "nblocks" value comes from.
1205 */
1206 nblocks = efi_label->efi_first_u_lba - 1;
1207
1208 /*
1209 * Determine if the EFI label is out of sync. We check that:
1210 *
1211 * 1. the data partition ends at the limit we set, and
1212 * 2. the reserved partition starts at the limit we set.
1213 *
1214 * If either of these conditions is not met, then we need to
1215 * resync the EFI label.
1216 *
1217 * The limit is the last usable LBA, determined by the last LBA
1218 * and the first usable LBA fields on the EFI label of the disk
1219 * (see the lines directly above). Additionally, we factor in
1220 * EFI_MIN_RESV_SIZE (per its use in "zpool_label_disk") and
1221 * P2ALIGN it to ensure the partition boundaries are aligned
1222 * (for performance reasons). The alignment should match the
1223 * alignment used by the "zpool_label_disk" function.
1224 */
1225 limit = P2ALIGN(efi_label->efi_last_lba - nblocks - EFI_MIN_RESV_SIZE,
1226 PARTITION_END_ALIGNMENT);
1227 if (data_start + data_size != limit || resv_start != limit)
1228 sync_needed = B_TRUE;
1229
1230 if (efi_debug && sync_needed)
1231 (void) fprintf(stderr, "efi_use_whole_disk: sync needed\n");
1232
1233 /*
1234 * If alter_lba is 1, we are using the backup label.
1235 * Since we can locate the backup label by disk capacity,
1236 * there must be no unallocated space.
1237 */
1238 if ((efi_label->efi_altern_lba == 1) || (efi_label->efi_altern_lba
1239 >= efi_label->efi_last_lba && !sync_needed)) {
1240 if (efi_debug) {
1241 (void) fprintf(stderr,
1242 "efi_use_whole_disk: requested space not found\n");
1243 }
1244 efi_free(efi_label);
1245 return (VT_ENOSPC);
1246 }
1247
1248 /*
1249 * Verify that we've found the reserved partition by checking
1250 * that it looks the way it did when we created it in zpool_label_disk.
1251 * If we've found the incorrect partition, then we know that this
1252 * device was reformatted and no longer is solely used by ZFS.
1253 */
1254 if ((efi_label->efi_parts[resv_index].p_size != EFI_MIN_RESV_SIZE) ||
1255 (efi_label->efi_parts[resv_index].p_tag != V_RESERVED) ||
1256 (resv_index != 8)) {
1257 if (efi_debug) {
1258 (void) fprintf(stderr,
1259 "efi_use_whole_disk: wholedisk not available\n");
1260 }
1261 efi_free(efi_label);
1262 return (VT_ENOSPC);
1263 }
1264
1265 if (data_start + data_size != resv_start) {
1266 if (efi_debug) {
1267 (void) fprintf(stderr,
1268 "efi_use_whole_disk: "
1269 "data_start (%lli) + "
1270 "data_size (%lli) != "
1271 "resv_start (%lli)\n",
1272 data_start, data_size, resv_start);
1273 }
1274
1275 return (VT_EINVAL);
1276 }
1277
1278 if (limit < resv_start) {
1279 if (efi_debug) {
1280 (void) fprintf(stderr,
1281 "efi_use_whole_disk: "
1282 "limit (%lli) < resv_start (%lli)\n",
1283 limit, resv_start);
1284 }
1285
1286 return (VT_EINVAL);
1287 }
1288
1289 difference = limit - resv_start;
1290
1291 if (efi_debug)
1292 (void) fprintf(stderr,
1293 "efi_use_whole_disk: difference is %lli\n", difference);
1294
1295 /*
1296 * Move the reserved partition. There is currently no data in
1297 * here except fabricated devids (which get generated via
1298 * efi_write()). So there is no need to copy data.
1299 */
1300 efi_label->efi_parts[data_index].p_size += difference;
1301 efi_label->efi_parts[resv_index].p_start += difference;
1302 efi_label->efi_last_u_lba = efi_label->efi_last_lba - nblocks;
1303
1304 /*
1305 * Rescanning the partition table in the kernel can result
1306 * in the device links to be removed (see comment in vdev_disk_open).
1307 * If BLKPG_RESIZE_PARTITION is available, then we can resize
1308 * the partition table online and avoid having to remove the device
1309 * links used by the pool. This provides a very deterministic
1310 * approach to resizing devices and does not require any
1311 * loops waiting for devices to reappear.
1312 */
1313 #ifdef BLKPG_RESIZE_PARTITION
1314 /*
1315 * Delete the reserved partition since we're about to expand
1316 * the data partition and it would overlap with the reserved
1317 * partition.
1318 * NOTE: The starting index for the ioctl is 1 while for the
1319 * EFI partitions it's 0. For that reason we have to add one
1320 * whenever we make an ioctl call.
1321 */
1322 rval = call_blkpg_ioctl(fd, BLKPG_DEL_PARTITION, 0, 0, resv_index + 1);
1323 if (rval != 0)
1324 goto out;
1325
1326 /*
1327 * Expand the data partition
1328 */
1329 rval = call_blkpg_ioctl(fd, BLKPG_RESIZE_PARTITION,
1330 efi_label->efi_parts[data_index].p_start * efi_label->efi_lbasize,
1331 efi_label->efi_parts[data_index].p_size * efi_label->efi_lbasize,
1332 data_index + 1);
1333 if (rval != 0) {
1334 (void) fprintf(stderr, "Unable to resize data "
1335 "partition: %d\n", rval);
1336 /*
1337 * Since we failed to resize, we need to reset the start
1338 * of the reserve partition and re-create it.
1339 */
1340 efi_label->efi_parts[resv_index].p_start -= difference;
1341 }
1342
1343 /*
1344 * Re-add the reserved partition. If we've expanded the data partition
1345 * then we'll move the reserve partition to the end of the data
1346 * partition. Otherwise, we'll recreate the partition in its original
1347 * location. Note that we do this as best-effort and ignore any
1348 * errors that may arise here. This will ensure that we finish writing
1349 * the EFI label.
1350 */
1351 (void) call_blkpg_ioctl(fd, BLKPG_ADD_PARTITION,
1352 efi_label->efi_parts[resv_index].p_start * efi_label->efi_lbasize,
1353 efi_label->efi_parts[resv_index].p_size * efi_label->efi_lbasize,
1354 resv_index + 1);
1355 #endif
1356
1357 /*
1358 * We're now ready to write the EFI label.
1359 */
1360 if (rval == 0) {
1361 rval = efi_write(fd, efi_label);
1362 if (rval < 0 && efi_debug) {
1363 (void) fprintf(stderr, "efi_use_whole_disk:fail "
1364 "to write label, rval=%d\n", rval);
1365 }
1366 }
1367
1368 out:
1369 efi_free(efi_label);
1370 return (rval);
1371 }
1372
1373 /*
1374 * write EFI label and backup label
1375 */
1376 int
efi_write(int fd,struct dk_gpt * vtoc)1377 efi_write(int fd, struct dk_gpt *vtoc)
1378 {
1379 dk_efi_t dk_ioc;
1380 efi_gpt_t *efi;
1381 efi_gpe_t *efi_parts;
1382 int i, j;
1383 struct dk_cinfo dki_info;
1384 int rval;
1385 int md_flag = 0;
1386 int nblocks;
1387 diskaddr_t lba_backup_gpt_hdr;
1388
1389 if ((rval = efi_get_info(fd, &dki_info)) != 0)
1390 return (rval);
1391
1392 /* check if we are dealing with a metadevice */
1393 if ((strncmp(dki_info.dki_cname, "pseudo", 7) == 0) &&
1394 (strncmp(dki_info.dki_dname, "md", 3) == 0)) {
1395 md_flag = 1;
1396 }
1397
1398 if (check_input(vtoc)) {
1399 /*
1400 * not valid; if it's a metadevice just pass it down
1401 * because SVM will do its own checking
1402 */
1403 if (md_flag == 0) {
1404 return (VT_EINVAL);
1405 }
1406 }
1407
1408 dk_ioc.dki_lba = 1;
1409 if (NBLOCKS(vtoc->efi_nparts, vtoc->efi_lbasize) < 34) {
1410 dk_ioc.dki_length = EFI_MIN_ARRAY_SIZE + vtoc->efi_lbasize;
1411 } else {
1412 dk_ioc.dki_length = NBLOCKS(vtoc->efi_nparts,
1413 vtoc->efi_lbasize) *
1414 vtoc->efi_lbasize;
1415 }
1416
1417 /*
1418 * the number of blocks occupied by GUID partition entry array
1419 */
1420 nblocks = dk_ioc.dki_length / vtoc->efi_lbasize - 1;
1421
1422 /*
1423 * Backup GPT header is located on the block after GUID
1424 * partition entry array. Here, we calculate the address
1425 * for backup GPT header.
1426 */
1427 lba_backup_gpt_hdr = vtoc->efi_last_u_lba + 1 + nblocks;
1428 if (posix_memalign((void **)&dk_ioc.dki_data,
1429 vtoc->efi_lbasize, dk_ioc.dki_length))
1430 return (VT_ERROR);
1431
1432 memset(dk_ioc.dki_data, 0, dk_ioc.dki_length);
1433 efi = dk_ioc.dki_data;
1434
1435 /* stuff user's input into EFI struct */
1436 efi->efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1437 efi->efi_gpt_Revision = LE_32(vtoc->efi_version); /* 0x02000100 */
1438 efi->efi_gpt_HeaderSize = LE_32(sizeof (struct efi_gpt) - LEN_EFI_PAD);
1439 efi->efi_gpt_Reserved1 = 0;
1440 efi->efi_gpt_MyLBA = LE_64(1ULL);
1441 efi->efi_gpt_AlternateLBA = LE_64(lba_backup_gpt_hdr);
1442 efi->efi_gpt_FirstUsableLBA = LE_64(vtoc->efi_first_u_lba);
1443 efi->efi_gpt_LastUsableLBA = LE_64(vtoc->efi_last_u_lba);
1444 efi->efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1445 efi->efi_gpt_NumberOfPartitionEntries = LE_32(vtoc->efi_nparts);
1446 efi->efi_gpt_SizeOfPartitionEntry = LE_32(sizeof (struct efi_gpe));
1447 UUID_LE_CONVERT(efi->efi_gpt_DiskGUID, vtoc->efi_disk_uguid);
1448
1449 /* LINTED -- always longlong aligned */
1450 efi_parts = (efi_gpe_t *)((char *)dk_ioc.dki_data + vtoc->efi_lbasize);
1451
1452 for (i = 0; i < vtoc->efi_nparts; i++) {
1453 for (j = 0;
1454 j < sizeof (conversion_array) /
1455 sizeof (struct uuid_to_ptag); j++) {
1456
1457 if (vtoc->efi_parts[i].p_tag == j) {
1458 UUID_LE_CONVERT(
1459 efi_parts[i].efi_gpe_PartitionTypeGUID,
1460 conversion_array[j].uuid);
1461 break;
1462 }
1463 }
1464
1465 if (j == sizeof (conversion_array) /
1466 sizeof (struct uuid_to_ptag)) {
1467 /*
1468 * If we didn't have a matching uuid match, bail here.
1469 * Don't write a label with unknown uuid.
1470 */
1471 if (efi_debug) {
1472 (void) fprintf(stderr,
1473 "Unknown uuid for p_tag %d\n",
1474 vtoc->efi_parts[i].p_tag);
1475 }
1476 return (VT_EINVAL);
1477 }
1478
1479 /* Zero's should be written for empty partitions */
1480 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED)
1481 continue;
1482
1483 efi_parts[i].efi_gpe_StartingLBA =
1484 LE_64(vtoc->efi_parts[i].p_start);
1485 efi_parts[i].efi_gpe_EndingLBA =
1486 LE_64(vtoc->efi_parts[i].p_start +
1487 vtoc->efi_parts[i].p_size - 1);
1488 efi_parts[i].efi_gpe_Attributes.PartitionAttrs =
1489 LE_16(vtoc->efi_parts[i].p_flag);
1490 for (j = 0; j < EFI_PART_NAME_LEN; j++) {
1491 efi_parts[i].efi_gpe_PartitionName[j] =
1492 LE_16((ushort_t)vtoc->efi_parts[i].p_name[j]);
1493 }
1494 if ((vtoc->efi_parts[i].p_tag != V_UNASSIGNED) &&
1495 uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_uguid)) {
1496 (void) uuid_generate((uchar_t *)
1497 &vtoc->efi_parts[i].p_uguid);
1498 }
1499 bcopy(&vtoc->efi_parts[i].p_uguid,
1500 &efi_parts[i].efi_gpe_UniquePartitionGUID,
1501 sizeof (uuid_t));
1502 }
1503 efi->efi_gpt_PartitionEntryArrayCRC32 =
1504 LE_32(efi_crc32((unsigned char *)efi_parts,
1505 vtoc->efi_nparts * (int)sizeof (struct efi_gpe)));
1506 efi->efi_gpt_HeaderCRC32 =
1507 LE_32(efi_crc32((unsigned char *)efi,
1508 LE_32(efi->efi_gpt_HeaderSize)));
1509
1510 if (efi_ioctl(fd, DKIOCSETEFI, &dk_ioc) == -1) {
1511 free(dk_ioc.dki_data);
1512 switch (errno) {
1513 case EIO:
1514 return (VT_EIO);
1515 case EINVAL:
1516 return (VT_EINVAL);
1517 default:
1518 return (VT_ERROR);
1519 }
1520 }
1521 /* if it's a metadevice we're done */
1522 if (md_flag) {
1523 free(dk_ioc.dki_data);
1524 return (0);
1525 }
1526
1527 /* write backup partition array */
1528 dk_ioc.dki_lba = vtoc->efi_last_u_lba + 1;
1529 dk_ioc.dki_length -= vtoc->efi_lbasize;
1530 /* LINTED */
1531 dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data +
1532 vtoc->efi_lbasize);
1533
1534 if (efi_ioctl(fd, DKIOCSETEFI, &dk_ioc) == -1) {
1535 /*
1536 * we wrote the primary label okay, so don't fail
1537 */
1538 if (efi_debug) {
1539 (void) fprintf(stderr,
1540 "write of backup partitions to block %llu "
1541 "failed, errno %d\n",
1542 vtoc->efi_last_u_lba + 1,
1543 errno);
1544 }
1545 }
1546 /*
1547 * now swap MyLBA and AlternateLBA fields and write backup
1548 * partition table header
1549 */
1550 dk_ioc.dki_lba = lba_backup_gpt_hdr;
1551 dk_ioc.dki_length = vtoc->efi_lbasize;
1552 /* LINTED */
1553 dk_ioc.dki_data = (efi_gpt_t *)((char *)dk_ioc.dki_data -
1554 vtoc->efi_lbasize);
1555 efi->efi_gpt_AlternateLBA = LE_64(1ULL);
1556 efi->efi_gpt_MyLBA = LE_64(lba_backup_gpt_hdr);
1557 efi->efi_gpt_PartitionEntryLBA = LE_64(vtoc->efi_last_u_lba + 1);
1558 efi->efi_gpt_HeaderCRC32 = 0;
1559 efi->efi_gpt_HeaderCRC32 =
1560 LE_32(efi_crc32((unsigned char *)dk_ioc.dki_data,
1561 LE_32(efi->efi_gpt_HeaderSize)));
1562
1563 if (efi_ioctl(fd, DKIOCSETEFI, &dk_ioc) == -1) {
1564 if (efi_debug) {
1565 (void) fprintf(stderr,
1566 "write of backup header to block %llu failed, "
1567 "errno %d\n",
1568 lba_backup_gpt_hdr,
1569 errno);
1570 }
1571 }
1572 /* write the PMBR */
1573 (void) write_pmbr(fd, vtoc);
1574 free(dk_ioc.dki_data);
1575
1576 return (0);
1577 }
1578
1579 void
efi_free(struct dk_gpt * ptr)1580 efi_free(struct dk_gpt *ptr)
1581 {
1582 free(ptr);
1583 }
1584
1585 /*
1586 * Input: File descriptor
1587 * Output: 1 if disk has an EFI label, or > 2TB with no VTOC or legacy MBR.
1588 * Otherwise 0.
1589 */
1590 int
efi_type(int fd)1591 efi_type(int fd)
1592 {
1593 #if 0
1594 struct vtoc vtoc;
1595 struct extvtoc extvtoc;
1596
1597 if (ioctl(fd, DKIOCGEXTVTOC, &extvtoc) == -1) {
1598 if (errno == ENOTSUP)
1599 return (1);
1600 else if (errno == ENOTTY) {
1601 if (ioctl(fd, DKIOCGVTOC, &vtoc) == -1)
1602 if (errno == ENOTSUP)
1603 return (1);
1604 }
1605 }
1606 return (0);
1607 #else
1608 return (ENOSYS);
1609 #endif
1610 }
1611
1612 void
efi_err_check(struct dk_gpt * vtoc)1613 efi_err_check(struct dk_gpt *vtoc)
1614 {
1615 int resv_part = -1;
1616 int i, j;
1617 diskaddr_t istart, jstart, isize, jsize, endsect;
1618 int overlap = 0;
1619
1620 /*
1621 * make sure no partitions overlap
1622 */
1623 for (i = 0; i < vtoc->efi_nparts; i++) {
1624 /* It can't be unassigned and have an actual size */
1625 if ((vtoc->efi_parts[i].p_tag == V_UNASSIGNED) &&
1626 (vtoc->efi_parts[i].p_size != 0)) {
1627 (void) fprintf(stderr,
1628 "partition %d is \"unassigned\" but has a size "
1629 "of %llu\n", i, vtoc->efi_parts[i].p_size);
1630 }
1631 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED) {
1632 continue;
1633 }
1634 if (vtoc->efi_parts[i].p_tag == V_RESERVED) {
1635 if (resv_part != -1) {
1636 (void) fprintf(stderr,
1637 "found duplicate reserved partition at "
1638 "%d\n", i);
1639 }
1640 resv_part = i;
1641 if (vtoc->efi_parts[i].p_size != EFI_MIN_RESV_SIZE)
1642 (void) fprintf(stderr,
1643 "Warning: reserved partition size must "
1644 "be %d sectors\n", EFI_MIN_RESV_SIZE);
1645 }
1646 if ((vtoc->efi_parts[i].p_start < vtoc->efi_first_u_lba) ||
1647 (vtoc->efi_parts[i].p_start > vtoc->efi_last_u_lba)) {
1648 (void) fprintf(stderr,
1649 "Partition %d starts at %llu\n",
1650 i,
1651 vtoc->efi_parts[i].p_start);
1652 (void) fprintf(stderr,
1653 "It must be between %llu and %llu.\n",
1654 vtoc->efi_first_u_lba,
1655 vtoc->efi_last_u_lba);
1656 }
1657 if ((vtoc->efi_parts[i].p_start +
1658 vtoc->efi_parts[i].p_size <
1659 vtoc->efi_first_u_lba) ||
1660 (vtoc->efi_parts[i].p_start +
1661 vtoc->efi_parts[i].p_size >
1662 vtoc->efi_last_u_lba + 1)) {
1663 (void) fprintf(stderr,
1664 "Partition %d ends at %llu\n",
1665 i,
1666 vtoc->efi_parts[i].p_start +
1667 vtoc->efi_parts[i].p_size);
1668 (void) fprintf(stderr,
1669 "It must be between %llu and %llu.\n",
1670 vtoc->efi_first_u_lba,
1671 vtoc->efi_last_u_lba);
1672 }
1673
1674 for (j = 0; j < vtoc->efi_nparts; j++) {
1675 isize = vtoc->efi_parts[i].p_size;
1676 jsize = vtoc->efi_parts[j].p_size;
1677 istart = vtoc->efi_parts[i].p_start;
1678 jstart = vtoc->efi_parts[j].p_start;
1679 if ((i != j) && (isize != 0) && (jsize != 0)) {
1680 endsect = jstart + jsize -1;
1681 if ((jstart <= istart) &&
1682 (istart <= endsect)) {
1683 if (!overlap) {
1684 (void) fprintf(stderr,
1685 "label error: EFI Labels do not "
1686 "support overlapping partitions\n");
1687 }
1688 (void) fprintf(stderr,
1689 "Partition %d overlaps partition "
1690 "%d.\n", i, j);
1691 overlap = 1;
1692 }
1693 }
1694 }
1695 }
1696 /* make sure there is a reserved partition */
1697 if (resv_part == -1) {
1698 (void) fprintf(stderr,
1699 "no reserved partition found\n");
1700 }
1701 }
1702
1703 /*
1704 * We need to get information necessary to construct a *new* efi
1705 * label type
1706 */
1707 int
efi_auto_sense(int fd,struct dk_gpt ** vtoc)1708 efi_auto_sense(int fd, struct dk_gpt **vtoc)
1709 {
1710
1711 int i;
1712
1713 /*
1714 * Now build the default partition table
1715 */
1716 if (efi_alloc_and_init(fd, EFI_NUMPAR, vtoc) != 0) {
1717 if (efi_debug) {
1718 (void) fprintf(stderr, "efi_alloc_and_init failed.\n");
1719 }
1720 return (-1);
1721 }
1722
1723 for (i = 0; i < MIN((*vtoc)->efi_nparts, V_NUMPAR); i++) {
1724 (*vtoc)->efi_parts[i].p_tag = default_vtoc_map[i].p_tag;
1725 (*vtoc)->efi_parts[i].p_flag = default_vtoc_map[i].p_flag;
1726 (*vtoc)->efi_parts[i].p_start = 0;
1727 (*vtoc)->efi_parts[i].p_size = 0;
1728 }
1729 /*
1730 * Make constants first
1731 * and variable partitions later
1732 */
1733
1734 /* root partition - s0 128 MB */
1735 (*vtoc)->efi_parts[0].p_start = 34;
1736 (*vtoc)->efi_parts[0].p_size = 262144;
1737
1738 /* partition - s1 128 MB */
1739 (*vtoc)->efi_parts[1].p_start = 262178;
1740 (*vtoc)->efi_parts[1].p_size = 262144;
1741
1742 /* partition -s2 is NOT the Backup disk */
1743 (*vtoc)->efi_parts[2].p_tag = V_UNASSIGNED;
1744
1745 /* partition -s6 /usr partition - HOG */
1746 (*vtoc)->efi_parts[6].p_start = 524322;
1747 (*vtoc)->efi_parts[6].p_size = (*vtoc)->efi_last_u_lba - 524322
1748 - (1024 * 16);
1749
1750 /* efi reserved partition - s9 16K */
1751 (*vtoc)->efi_parts[8].p_start = (*vtoc)->efi_last_u_lba - (1024 * 16);
1752 (*vtoc)->efi_parts[8].p_size = (1024 * 16);
1753 (*vtoc)->efi_parts[8].p_tag = V_RESERVED;
1754 return (0);
1755 }
1756