xref: /freebsd/stand/common/part.c (revision b9f654b163bce26de79705e77b872427c9f2afa1)
1 /*-
2  * Copyright (c) 2012 Andrey V. Elsukov <ae@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include <stand.h>
31 #include <sys/param.h>
32 #include <sys/diskmbr.h>
33 #include <sys/disklabel.h>
34 #include <sys/endian.h>
35 #include <sys/gpt.h>
36 #include <sys/stddef.h>
37 #include <sys/queue.h>
38 #include <sys/vtoc.h>
39 
40 #include <fs/cd9660/iso.h>
41 
42 #include <crc32.h>
43 #include <part.h>
44 #include <uuid.h>
45 
46 #ifdef PART_DEBUG
47 #define	DPRINTF(fmt, args...) printf("%s: " fmt "\n", __func__, ## args)
48 #else
49 #define	DPRINTF(fmt, args...)	((void)0)
50 #endif
51 
52 #ifdef LOADER_GPT_SUPPORT
53 #define	MAXTBLSZ	64
54 static const uuid_t gpt_uuid_unused = GPT_ENT_TYPE_UNUSED;
55 static const uuid_t gpt_uuid_ms_basic_data = GPT_ENT_TYPE_MS_BASIC_DATA;
56 static const uuid_t gpt_uuid_freebsd_ufs = GPT_ENT_TYPE_FREEBSD_UFS;
57 static const uuid_t gpt_uuid_efi = GPT_ENT_TYPE_EFI;
58 static const uuid_t gpt_uuid_freebsd = GPT_ENT_TYPE_FREEBSD;
59 static const uuid_t gpt_uuid_freebsd_boot = GPT_ENT_TYPE_FREEBSD_BOOT;
60 static const uuid_t gpt_uuid_freebsd_nandfs = GPT_ENT_TYPE_FREEBSD_NANDFS;
61 static const uuid_t gpt_uuid_freebsd_swap = GPT_ENT_TYPE_FREEBSD_SWAP;
62 static const uuid_t gpt_uuid_freebsd_zfs = GPT_ENT_TYPE_FREEBSD_ZFS;
63 static const uuid_t gpt_uuid_freebsd_vinum = GPT_ENT_TYPE_FREEBSD_VINUM;
64 #endif
65 
66 struct pentry {
67 	struct ptable_entry	part;
68 	uint64_t		flags;
69 	union {
70 		uint8_t bsd;
71 		uint8_t	mbr;
72 		uuid_t	gpt;
73 		uint16_t vtoc8;
74 	} type;
75 	STAILQ_ENTRY(pentry)	entry;
76 };
77 
78 struct ptable {
79 	enum ptable_type	type;
80 	uint16_t		sectorsize;
81 	uint64_t		sectors;
82 
83 	STAILQ_HEAD(, pentry)	entries;
84 };
85 
86 static struct parttypes {
87 	enum partition_type	type;
88 	const char		*desc;
89 } ptypes[] = {
90 	{ PART_UNKNOWN,		"Unknown" },
91 	{ PART_EFI,		"EFI" },
92 	{ PART_FREEBSD,		"FreeBSD" },
93 	{ PART_FREEBSD_BOOT,	"FreeBSD boot" },
94 	{ PART_FREEBSD_NANDFS,	"FreeBSD nandfs" },
95 	{ PART_FREEBSD_UFS,	"FreeBSD UFS" },
96 	{ PART_FREEBSD_ZFS,	"FreeBSD ZFS" },
97 	{ PART_FREEBSD_SWAP,	"FreeBSD swap" },
98 	{ PART_FREEBSD_VINUM,	"FreeBSD vinum" },
99 	{ PART_LINUX,		"Linux" },
100 	{ PART_LINUX_SWAP,	"Linux swap" },
101 	{ PART_DOS,		"DOS/Windows" },
102 	{ PART_ISO9660,		"ISO9660" },
103 };
104 
105 const char *
106 parttype2str(enum partition_type type)
107 {
108 	size_t i;
109 
110 	for (i = 0; i < nitems(ptypes); i++)
111 		if (ptypes[i].type == type)
112 			return (ptypes[i].desc);
113 	return (ptypes[0].desc);
114 }
115 
116 #ifdef LOADER_GPT_SUPPORT
117 static void
118 uuid_letoh(uuid_t *uuid)
119 {
120 
121 	uuid->time_low = le32toh(uuid->time_low);
122 	uuid->time_mid = le16toh(uuid->time_mid);
123 	uuid->time_hi_and_version = le16toh(uuid->time_hi_and_version);
124 }
125 
126 static enum partition_type
127 gpt_parttype(uuid_t type)
128 {
129 
130 	if (uuid_equal(&type, &gpt_uuid_efi, NULL))
131 		return (PART_EFI);
132 	else if (uuid_equal(&type, &gpt_uuid_ms_basic_data, NULL))
133 		return (PART_DOS);
134 	else if (uuid_equal(&type, &gpt_uuid_freebsd_boot, NULL))
135 		return (PART_FREEBSD_BOOT);
136 	else if (uuid_equal(&type, &gpt_uuid_freebsd_ufs, NULL))
137 		return (PART_FREEBSD_UFS);
138 	else if (uuid_equal(&type, &gpt_uuid_freebsd_zfs, NULL))
139 		return (PART_FREEBSD_ZFS);
140 	else if (uuid_equal(&type, &gpt_uuid_freebsd_swap, NULL))
141 		return (PART_FREEBSD_SWAP);
142 	else if (uuid_equal(&type, &gpt_uuid_freebsd_vinum, NULL))
143 		return (PART_FREEBSD_VINUM);
144 	else if (uuid_equal(&type, &gpt_uuid_freebsd_nandfs, NULL))
145 		return (PART_FREEBSD_NANDFS);
146 	else if (uuid_equal(&type, &gpt_uuid_freebsd, NULL))
147 		return (PART_FREEBSD);
148 	return (PART_UNKNOWN);
149 }
150 
151 static struct gpt_hdr *
152 gpt_checkhdr(struct gpt_hdr *hdr, uint64_t lba_self, uint64_t lba_last,
153     uint16_t sectorsize)
154 {
155 	uint32_t sz, crc;
156 
157 	if (memcmp(hdr->hdr_sig, GPT_HDR_SIG, sizeof(hdr->hdr_sig)) != 0) {
158 		DPRINTF("no GPT signature");
159 		return (NULL);
160 	}
161 	sz = le32toh(hdr->hdr_size);
162 	if (sz < 92 || sz > sectorsize) {
163 		DPRINTF("invalid GPT header size: %d", sz);
164 		return (NULL);
165 	}
166 	crc = le32toh(hdr->hdr_crc_self);
167 	hdr->hdr_crc_self = 0;
168 	if (crc32(hdr, sz) != crc) {
169 		DPRINTF("GPT header's CRC doesn't match");
170 		return (NULL);
171 	}
172 	hdr->hdr_crc_self = crc;
173 	hdr->hdr_revision = le32toh(hdr->hdr_revision);
174 	if (hdr->hdr_revision < GPT_HDR_REVISION) {
175 		DPRINTF("unsupported GPT revision %d", hdr->hdr_revision);
176 		return (NULL);
177 	}
178 	hdr->hdr_lba_self = le64toh(hdr->hdr_lba_self);
179 	if (hdr->hdr_lba_self != lba_self) {
180 		DPRINTF("self LBA doesn't match");
181 		return (NULL);
182 	}
183 	hdr->hdr_lba_alt = le64toh(hdr->hdr_lba_alt);
184 	if (hdr->hdr_lba_alt == hdr->hdr_lba_self) {
185 		DPRINTF("invalid alternate LBA");
186 		return (NULL);
187 	}
188 	hdr->hdr_entries = le32toh(hdr->hdr_entries);
189 	hdr->hdr_entsz = le32toh(hdr->hdr_entsz);
190 	if (hdr->hdr_entries == 0 ||
191 	    hdr->hdr_entsz < sizeof(struct gpt_ent) ||
192 	    sectorsize % hdr->hdr_entsz != 0) {
193 		DPRINTF("invalid entry size or number of entries");
194 		return (NULL);
195 	}
196 	hdr->hdr_lba_start = le64toh(hdr->hdr_lba_start);
197 	hdr->hdr_lba_end = le64toh(hdr->hdr_lba_end);
198 	hdr->hdr_lba_table = le64toh(hdr->hdr_lba_table);
199 	hdr->hdr_crc_table = le32toh(hdr->hdr_crc_table);
200 	uuid_letoh(&hdr->hdr_uuid);
201 	return (hdr);
202 }
203 
204 static int
205 gpt_checktbl(const struct gpt_hdr *hdr, uint8_t *tbl, size_t size,
206     uint64_t lba_last)
207 {
208 	struct gpt_ent *ent;
209 	uint32_t i, cnt;
210 
211 	cnt = size / hdr->hdr_entsz;
212 	if (hdr->hdr_entries <= cnt) {
213 		cnt = hdr->hdr_entries;
214 		/* Check CRC only when buffer size is enough for table. */
215 		if (hdr->hdr_crc_table !=
216 		    crc32(tbl, hdr->hdr_entries * hdr->hdr_entsz)) {
217 			DPRINTF("GPT table's CRC doesn't match");
218 			return (-1);
219 		}
220 	}
221 	for (i = 0; i < cnt; i++) {
222 		ent = (struct gpt_ent *)(tbl + i * hdr->hdr_entsz);
223 		uuid_letoh(&ent->ent_type);
224 		if (uuid_equal(&ent->ent_type, &gpt_uuid_unused, NULL))
225 			continue;
226 		ent->ent_lba_start = le64toh(ent->ent_lba_start);
227 		ent->ent_lba_end = le64toh(ent->ent_lba_end);
228 	}
229 	return (0);
230 }
231 
232 static struct ptable *
233 ptable_gptread(struct ptable *table, void *dev, diskread_t dread)
234 {
235 	struct pentry *entry;
236 	struct gpt_hdr *phdr, hdr;
237 	struct gpt_ent *ent;
238 	uint8_t *buf, *tbl;
239 	uint64_t offset;
240 	int pri, sec;
241 	size_t size, i;
242 
243 	buf = malloc(table->sectorsize);
244 	if (buf == NULL)
245 		return (NULL);
246 	tbl = malloc(table->sectorsize * MAXTBLSZ);
247 	if (tbl == NULL) {
248 		free(buf);
249 		return (NULL);
250 	}
251 	/* Read the primary GPT header. */
252 	if (dread(dev, buf, 1, 1) != 0) {
253 		ptable_close(table);
254 		table = NULL;
255 		goto out;
256 	}
257 	pri = sec = 0;
258 	/* Check the primary GPT header. */
259 	phdr = gpt_checkhdr((struct gpt_hdr *)buf, 1, table->sectors - 1,
260 	    table->sectorsize);
261 	if (phdr != NULL) {
262 		/* Read the primary GPT table. */
263 		size = MIN(MAXTBLSZ,
264 		    howmany(phdr->hdr_entries * phdr->hdr_entsz,
265 		        table->sectorsize));
266 		if (dread(dev, tbl, size, phdr->hdr_lba_table) == 0 &&
267 		    gpt_checktbl(phdr, tbl, size * table->sectorsize,
268 		    table->sectors - 1) == 0) {
269 			memcpy(&hdr, phdr, sizeof(hdr));
270 			pri = 1;
271 		}
272 	}
273 	offset = pri ? hdr.hdr_lba_alt: table->sectors - 1;
274 	/* Read the backup GPT header. */
275 	if (dread(dev, buf, 1, offset) != 0)
276 		phdr = NULL;
277 	else
278 		phdr = gpt_checkhdr((struct gpt_hdr *)buf, offset,
279 		    table->sectors - 1, table->sectorsize);
280 	if (phdr != NULL) {
281 		/*
282 		 * Compare primary and backup headers.
283 		 * If they are equal, then we do not need to read backup
284 		 * table. If they are different, then prefer backup header
285 		 * and try to read backup table.
286 		 */
287 		if (pri == 0 ||
288 		    uuid_equal(&hdr.hdr_uuid, &phdr->hdr_uuid, NULL) == 0 ||
289 		    hdr.hdr_revision != phdr->hdr_revision ||
290 		    hdr.hdr_size != phdr->hdr_size ||
291 		    hdr.hdr_lba_start != phdr->hdr_lba_start ||
292 		    hdr.hdr_lba_end != phdr->hdr_lba_end ||
293 		    hdr.hdr_entries != phdr->hdr_entries ||
294 		    hdr.hdr_entsz != phdr->hdr_entsz ||
295 		    hdr.hdr_crc_table != phdr->hdr_crc_table) {
296 			/* Read the backup GPT table. */
297 			size = MIN(MAXTBLSZ,
298 				   howmany(phdr->hdr_entries * phdr->hdr_entsz,
299 				       table->sectorsize));
300 			if (dread(dev, tbl, size, phdr->hdr_lba_table) == 0 &&
301 			    gpt_checktbl(phdr, tbl, size * table->sectorsize,
302 			    table->sectors - 1) == 0) {
303 				memcpy(&hdr, phdr, sizeof(hdr));
304 				sec = 1;
305 			}
306 		}
307 	}
308 	if (pri == 0 && sec == 0) {
309 		/* Both primary and backup tables are invalid. */
310 		table->type = PTABLE_NONE;
311 		goto out;
312 	}
313 	DPRINTF("GPT detected");
314 	size = MIN(hdr.hdr_entries * hdr.hdr_entsz,
315 	    MAXTBLSZ * table->sectorsize);
316 
317 	/*
318 	 * If the disk's sector count is smaller than the sector count recorded
319 	 * in the disk's GPT table header, set the table->sectors to the value
320 	 * recorded in GPT tables. This is done to work around buggy firmware
321 	 * that returns truncated disk sizes.
322 	 *
323 	 * Note, this is still not a foolproof way to get disk's size. For
324 	 * example, an image file can be truncated when copied to smaller media.
325 	 */
326 	table->sectors = hdr.hdr_lba_alt + 1;
327 
328 	for (i = 0; i < size / hdr.hdr_entsz; i++) {
329 		ent = (struct gpt_ent *)(tbl + i * hdr.hdr_entsz);
330 		if (uuid_equal(&ent->ent_type, &gpt_uuid_unused, NULL))
331 			continue;
332 
333 		/* Simple sanity checks. */
334 		if (ent->ent_lba_start < hdr.hdr_lba_start ||
335 		    ent->ent_lba_end > hdr.hdr_lba_end ||
336 		    ent->ent_lba_start > ent->ent_lba_end)
337 			continue;
338 
339 		entry = malloc(sizeof(*entry));
340 		if (entry == NULL)
341 			break;
342 		entry->part.start = ent->ent_lba_start;
343 		entry->part.end = ent->ent_lba_end;
344 		entry->part.index = i + 1;
345 		entry->part.type = gpt_parttype(ent->ent_type);
346 		entry->flags = le64toh(ent->ent_attr);
347 		memcpy(&entry->type.gpt, &ent->ent_type, sizeof(uuid_t));
348 		STAILQ_INSERT_TAIL(&table->entries, entry, entry);
349 		DPRINTF("new GPT partition added");
350 	}
351 out:
352 	free(buf);
353 	free(tbl);
354 	return (table);
355 }
356 #endif /* LOADER_GPT_SUPPORT */
357 
358 #ifdef LOADER_MBR_SUPPORT
359 /* We do not need to support too many EBR partitions in the loader */
360 #define	MAXEBRENTRIES		8
361 static enum partition_type
362 mbr_parttype(uint8_t type)
363 {
364 
365 	switch (type) {
366 	case DOSPTYP_386BSD:
367 		return (PART_FREEBSD);
368 	case DOSPTYP_LINSWP:
369 		return (PART_LINUX_SWAP);
370 	case DOSPTYP_LINUX:
371 		return (PART_LINUX);
372 	case 0x01:
373 	case 0x04:
374 	case 0x06:
375 	case 0x07:
376 	case 0x0b:
377 	case 0x0c:
378 	case 0x0e:
379 		return (PART_DOS);
380 	}
381 	return (PART_UNKNOWN);
382 }
383 
384 static struct ptable *
385 ptable_ebrread(struct ptable *table, void *dev, diskread_t dread)
386 {
387 	struct dos_partition *dp;
388 	struct pentry *e1, *entry;
389 	uint32_t start, end, offset;
390 	u_char *buf;
391 	int i, index;
392 
393 	STAILQ_FOREACH(e1, &table->entries, entry) {
394 		if (e1->type.mbr == DOSPTYP_EXT ||
395 		    e1->type.mbr == DOSPTYP_EXTLBA)
396 			break;
397 	}
398 	if (e1 == NULL)
399 		return (table);
400 	index = 5;
401 	offset = e1->part.start;
402 	buf = malloc(table->sectorsize);
403 	if (buf == NULL)
404 		return (table);
405 	DPRINTF("EBR detected");
406 	for (i = 0; i < MAXEBRENTRIES; i++) {
407 #if 0	/* Some BIOSes return an incorrect number of sectors */
408 		if (offset >= table->sectors)
409 			break;
410 #endif
411 		if (dread(dev, buf, 1, offset) != 0)
412 			break;
413 		dp = (struct dos_partition *)(buf + DOSPARTOFF);
414 		if (dp[0].dp_typ == 0)
415 			break;
416 		start = le32toh(dp[0].dp_start);
417 		if (dp[0].dp_typ == DOSPTYP_EXT &&
418 		    dp[1].dp_typ == 0) {
419 			offset = e1->part.start + start;
420 			continue;
421 		}
422 		end = le32toh(dp[0].dp_size);
423 		entry = malloc(sizeof(*entry));
424 		if (entry == NULL)
425 			break;
426 		entry->part.start = offset + start;
427 		entry->part.end = entry->part.start + end - 1;
428 		entry->part.index = index++;
429 		entry->part.type = mbr_parttype(dp[0].dp_typ);
430 		entry->flags = dp[0].dp_flag;
431 		entry->type.mbr = dp[0].dp_typ;
432 		STAILQ_INSERT_TAIL(&table->entries, entry, entry);
433 		DPRINTF("new EBR partition added");
434 		if (dp[1].dp_typ == 0)
435 			break;
436 		offset = e1->part.start + le32toh(dp[1].dp_start);
437 	}
438 	free(buf);
439 	return (table);
440 }
441 #endif /* LOADER_MBR_SUPPORT */
442 
443 static enum partition_type
444 bsd_parttype(uint8_t type)
445 {
446 
447 	switch (type) {
448 	case FS_NANDFS:
449 		return (PART_FREEBSD_NANDFS);
450 	case FS_SWAP:
451 		return (PART_FREEBSD_SWAP);
452 	case FS_BSDFFS:
453 		return (PART_FREEBSD_UFS);
454 	case FS_VINUM:
455 		return (PART_FREEBSD_VINUM);
456 	case FS_ZFS:
457 		return (PART_FREEBSD_ZFS);
458 	}
459 	return (PART_UNKNOWN);
460 }
461 
462 static struct ptable *
463 ptable_bsdread(struct ptable *table, void *dev, diskread_t dread)
464 {
465 	struct disklabel *dl;
466 	struct partition *part;
467 	struct pentry *entry;
468 	uint8_t *buf;
469 	uint32_t raw_offset;
470 	int i;
471 
472 	if (table->sectorsize < sizeof(struct disklabel)) {
473 		DPRINTF("Too small sectorsize");
474 		return (table);
475 	}
476 	buf = malloc(table->sectorsize);
477 	if (buf == NULL)
478 		return (table);
479 	if (dread(dev, buf, 1, 1) != 0) {
480 		DPRINTF("read failed");
481 		ptable_close(table);
482 		table = NULL;
483 		goto out;
484 	}
485 	dl = (struct disklabel *)buf;
486 	if (le32toh(dl->d_magic) != DISKMAGIC &&
487 	    le32toh(dl->d_magic2) != DISKMAGIC)
488 		goto out;
489 	if (le32toh(dl->d_secsize) != table->sectorsize) {
490 		DPRINTF("unsupported sector size");
491 		goto out;
492 	}
493 	dl->d_npartitions = le16toh(dl->d_npartitions);
494 	if (dl->d_npartitions > 20 || dl->d_npartitions < 8) {
495 		DPRINTF("invalid number of partitions");
496 		goto out;
497 	}
498 	DPRINTF("BSD detected");
499 	part = &dl->d_partitions[0];
500 	raw_offset = le32toh(part[RAW_PART].p_offset);
501 	for (i = 0; i < dl->d_npartitions; i++, part++) {
502 		if (i == RAW_PART)
503 			continue;
504 		if (part->p_size == 0)
505 			continue;
506 		entry = malloc(sizeof(*entry));
507 		if (entry == NULL)
508 			break;
509 		entry->part.start = le32toh(part->p_offset) - raw_offset;
510 		entry->part.end = entry->part.start +
511 		    le32toh(part->p_size) - 1;
512 		entry->part.type = bsd_parttype(part->p_fstype);
513 		entry->part.index = i; /* starts from zero */
514 		entry->type.bsd = part->p_fstype;
515 		STAILQ_INSERT_TAIL(&table->entries, entry, entry);
516 		DPRINTF("new BSD partition added");
517 	}
518 	table->type = PTABLE_BSD;
519 out:
520 	free(buf);
521 	return (table);
522 }
523 
524 #ifdef LOADER_VTOC8_SUPPORT
525 static enum partition_type
526 vtoc8_parttype(uint16_t type)
527 {
528 
529 	switch (type) {
530 	case VTOC_TAG_FREEBSD_NANDFS:
531 		return (PART_FREEBSD_NANDFS);
532 	case VTOC_TAG_FREEBSD_SWAP:
533 		return (PART_FREEBSD_SWAP);
534 	case VTOC_TAG_FREEBSD_UFS:
535 		return (PART_FREEBSD_UFS);
536 	case VTOC_TAG_FREEBSD_VINUM:
537 		return (PART_FREEBSD_VINUM);
538 	case VTOC_TAG_FREEBSD_ZFS:
539 		return (PART_FREEBSD_ZFS);
540 	}
541 	return (PART_UNKNOWN);
542 }
543 
544 static struct ptable *
545 ptable_vtoc8read(struct ptable *table, void *dev, diskread_t dread)
546 {
547 	struct pentry *entry;
548 	struct vtoc8 *dl;
549 	uint8_t *buf;
550 	uint16_t sum, heads, sectors;
551 	int i;
552 
553 	if (table->sectorsize != sizeof(struct vtoc8))
554 		return (table);
555 	buf = malloc(table->sectorsize);
556 	if (buf == NULL)
557 		return (table);
558 	if (dread(dev, buf, 1, 0) != 0) {
559 		DPRINTF("read failed");
560 		ptable_close(table);
561 		table = NULL;
562 		goto out;
563 	}
564 	dl = (struct vtoc8 *)buf;
565 	/* Check the sum */
566 	for (i = sum = 0; i < sizeof(struct vtoc8); i += sizeof(sum))
567 		sum ^= be16dec(buf + i);
568 	if (sum != 0) {
569 		DPRINTF("incorrect checksum");
570 		goto out;
571 	}
572 	if (be16toh(dl->nparts) != VTOC8_NPARTS) {
573 		DPRINTF("invalid number of entries");
574 		goto out;
575 	}
576 	sectors = be16toh(dl->nsecs);
577 	heads = be16toh(dl->nheads);
578 	if (sectors * heads == 0) {
579 		DPRINTF("invalid geometry");
580 		goto out;
581 	}
582 	DPRINTF("VTOC8 detected");
583 	for (i = 0; i < VTOC8_NPARTS; i++) {
584 		dl->part[i].tag = be16toh(dl->part[i].tag);
585 		if (i == VTOC_RAW_PART ||
586 		    dl->part[i].tag == VTOC_TAG_UNASSIGNED)
587 			continue;
588 		entry = malloc(sizeof(*entry));
589 		if (entry == NULL)
590 			break;
591 		entry->part.start = be32toh(dl->map[i].cyl) * heads * sectors;
592 		entry->part.end = be32toh(dl->map[i].nblks) +
593 		    entry->part.start - 1;
594 		entry->part.type = vtoc8_parttype(dl->part[i].tag);
595 		entry->part.index = i; /* starts from zero */
596 		entry->type.vtoc8 = dl->part[i].tag;
597 		STAILQ_INSERT_TAIL(&table->entries, entry, entry);
598 		DPRINTF("new VTOC8 partition added");
599 	}
600 	table->type = PTABLE_VTOC8;
601 out:
602 	free(buf);
603 	return (table);
604 
605 }
606 #endif /* LOADER_VTOC8_SUPPORT */
607 
608 #define cdb2devb(bno)   ((bno) * ISO_DEFAULT_BLOCK_SIZE / table->sectorsize)
609 
610 static struct ptable *
611 ptable_iso9660read(struct ptable *table, void *dev, diskread_t dread)
612 {
613 	uint8_t *buf;
614 	struct iso_primary_descriptor *vd;
615 	struct pentry *entry;
616 
617 	buf = malloc(table->sectorsize);
618 	if (buf == NULL)
619 		return (table);
620 
621 	if (dread(dev, buf, 1, cdb2devb(16)) != 0) {
622 		DPRINTF("read failed");
623 		ptable_close(table);
624 		table = NULL;
625 		goto out;
626 	}
627 	vd = (struct iso_primary_descriptor *)buf;
628 	if (bcmp(vd->id, ISO_STANDARD_ID, sizeof vd->id) != 0)
629 		goto out;
630 
631 	entry = malloc(sizeof(*entry));
632 	if (entry == NULL)
633 		goto out;
634 	entry->part.start = 0;
635 	entry->part.end = table->sectors;
636 	entry->part.type = PART_ISO9660;
637 	entry->part.index = 0;
638 	STAILQ_INSERT_TAIL(&table->entries, entry, entry);
639 
640 	table->type = PTABLE_ISO9660;
641 
642 out:
643 	free(buf);
644 	return (table);
645 }
646 
647 struct ptable *
648 ptable_open(void *dev, uint64_t sectors, uint16_t sectorsize,
649     diskread_t *dread)
650 {
651 	struct dos_partition *dp;
652 	struct ptable *table;
653 	uint8_t *buf;
654 	int i, count;
655 #ifdef LOADER_MBR_SUPPORT
656 	struct pentry *entry;
657 	uint32_t start, end;
658 	int has_ext;
659 #endif
660 	table = NULL;
661 	buf = malloc(sectorsize);
662 	if (buf == NULL)
663 		return (NULL);
664 	/* First, read the MBR. */
665 	if (dread(dev, buf, 1, DOSBBSECTOR) != 0) {
666 		DPRINTF("read failed");
667 		goto out;
668 	}
669 
670 	table = malloc(sizeof(*table));
671 	if (table == NULL)
672 		goto out;
673 	table->sectors = sectors;
674 	table->sectorsize = sectorsize;
675 	table->type = PTABLE_NONE;
676 	STAILQ_INIT(&table->entries);
677 
678 	if (ptable_iso9660read(table, dev, dread) == NULL) {
679 		/* Read error. */
680 		table = NULL;
681 		goto out;
682 	} else if (table->type == PTABLE_ISO9660)
683 		goto out;
684 
685 #ifdef LOADER_VTOC8_SUPPORT
686 	if (be16dec(buf + offsetof(struct vtoc8, magic)) == VTOC_MAGIC) {
687 		if (ptable_vtoc8read(table, dev, dread) == NULL) {
688 			/* Read error. */
689 			table = NULL;
690 			goto out;
691 		} else if (table->type == PTABLE_VTOC8)
692 			goto out;
693 	}
694 #endif
695 	/* Check the BSD label. */
696 	if (ptable_bsdread(table, dev, dread) == NULL) { /* Read error. */
697 		table = NULL;
698 		goto out;
699 	} else if (table->type == PTABLE_BSD)
700 		goto out;
701 
702 #if defined(LOADER_GPT_SUPPORT) || defined(LOADER_MBR_SUPPORT)
703 	/* Check the MBR magic. */
704 	if (buf[DOSMAGICOFFSET] != 0x55 ||
705 	    buf[DOSMAGICOFFSET + 1] != 0xaa) {
706 		DPRINTF("magic sequence not found");
707 #if defined(LOADER_GPT_SUPPORT)
708 		/* There is no PMBR, check that we have backup GPT */
709 		table->type = PTABLE_GPT;
710 		table = ptable_gptread(table, dev, dread);
711 #endif
712 		goto out;
713 	}
714 	/* Check that we have PMBR. Also do some validation. */
715 	dp = (struct dos_partition *)(buf + DOSPARTOFF);
716 	for (i = 0, count = 0; i < NDOSPART; i++) {
717 		if (dp[i].dp_flag != 0 && dp[i].dp_flag != 0x80) {
718 			DPRINTF("invalid partition flag %x", dp[i].dp_flag);
719 			goto out;
720 		}
721 #ifdef LOADER_GPT_SUPPORT
722 		if (dp[i].dp_typ == DOSPTYP_PMBR) {
723 			table->type = PTABLE_GPT;
724 			DPRINTF("PMBR detected");
725 		}
726 #endif
727 		if (dp[i].dp_typ != 0)
728 			count++;
729 	}
730 	/* Do we have some invalid values? */
731 	if (table->type == PTABLE_GPT && count > 1) {
732 		if (dp[1].dp_typ != DOSPTYP_HFS) {
733 			table->type = PTABLE_NONE;
734 			DPRINTF("Incorrect PMBR, ignore it");
735 		} else {
736 			DPRINTF("Bootcamp detected");
737 		}
738 	}
739 #ifdef LOADER_GPT_SUPPORT
740 	if (table->type == PTABLE_GPT) {
741 		table = ptable_gptread(table, dev, dread);
742 		goto out;
743 	}
744 #endif
745 #ifdef LOADER_MBR_SUPPORT
746 	/* Read MBR. */
747 	DPRINTF("MBR detected");
748 	table->type = PTABLE_MBR;
749 	for (i = has_ext = 0; i < NDOSPART; i++) {
750 		if (dp[i].dp_typ == 0)
751 			continue;
752 		start = le32dec(&(dp[i].dp_start));
753 		end = le32dec(&(dp[i].dp_size));
754 		if (start == 0 || end == 0)
755 			continue;
756 #if 0	/* Some BIOSes return an incorrect number of sectors */
757 		if (start + end - 1 >= sectors)
758 			continue;	/* XXX: ignore */
759 #endif
760 		if (dp[i].dp_typ == DOSPTYP_EXT ||
761 		    dp[i].dp_typ == DOSPTYP_EXTLBA)
762 			has_ext = 1;
763 		entry = malloc(sizeof(*entry));
764 		if (entry == NULL)
765 			break;
766 		entry->part.start = start;
767 		entry->part.end = start + end - 1;
768 		entry->part.index = i + 1;
769 		entry->part.type = mbr_parttype(dp[i].dp_typ);
770 		entry->flags = dp[i].dp_flag;
771 		entry->type.mbr = dp[i].dp_typ;
772 		STAILQ_INSERT_TAIL(&table->entries, entry, entry);
773 		DPRINTF("new MBR partition added");
774 	}
775 	if (has_ext) {
776 		table = ptable_ebrread(table, dev, dread);
777 		/* FALLTHROUGH */
778 	}
779 #endif /* LOADER_MBR_SUPPORT */
780 #endif /* LOADER_MBR_SUPPORT || LOADER_GPT_SUPPORT */
781 out:
782 	free(buf);
783 	return (table);
784 }
785 
786 void
787 ptable_close(struct ptable *table)
788 {
789 	struct pentry *entry;
790 
791 	if (table == NULL)
792 		return;
793 
794 	while (!STAILQ_EMPTY(&table->entries)) {
795 		entry = STAILQ_FIRST(&table->entries);
796 		STAILQ_REMOVE_HEAD(&table->entries, entry);
797 		free(entry);
798 	}
799 	free(table);
800 }
801 
802 enum ptable_type
803 ptable_gettype(const struct ptable *table)
804 {
805 
806 	return (table->type);
807 }
808 
809 int
810 ptable_getsize(const struct ptable *table, uint64_t *sizep)
811 {
812 	uint64_t tmp = table->sectors * table->sectorsize;
813 
814 	if (tmp < table->sectors)
815 		return (EOVERFLOW);
816 
817 	if (sizep != NULL)
818 		*sizep = tmp;
819 	return (0);
820 }
821 
822 int
823 ptable_getpart(const struct ptable *table, struct ptable_entry *part, int index)
824 {
825 	struct pentry *entry;
826 
827 	if (part == NULL || table == NULL)
828 		return (EINVAL);
829 
830 	STAILQ_FOREACH(entry, &table->entries, entry) {
831 		if (entry->part.index != index)
832 			continue;
833 		memcpy(part, &entry->part, sizeof(*part));
834 		return (0);
835 	}
836 	return (ENOENT);
837 }
838 
839 /*
840  * Search for a slice with the following preferences:
841  *
842  * 1: Active FreeBSD slice
843  * 2: Non-active FreeBSD slice
844  * 3: Active Linux slice
845  * 4: non-active Linux slice
846  * 5: Active FAT/FAT32 slice
847  * 6: non-active FAT/FAT32 slice
848  */
849 #define	PREF_RAWDISK	0
850 #define	PREF_FBSD_ACT	1
851 #define	PREF_FBSD	2
852 #define	PREF_LINUX_ACT	3
853 #define	PREF_LINUX	4
854 #define	PREF_DOS_ACT	5
855 #define	PREF_DOS	6
856 #define	PREF_NONE	7
857 int
858 ptable_getbestpart(const struct ptable *table, struct ptable_entry *part)
859 {
860 	struct pentry *entry, *best;
861 	int pref, preflevel;
862 
863 	if (part == NULL || table == NULL)
864 		return (EINVAL);
865 
866 	best = NULL;
867 	preflevel = pref = PREF_NONE;
868 	STAILQ_FOREACH(entry, &table->entries, entry) {
869 #ifdef LOADER_MBR_SUPPORT
870 		if (table->type == PTABLE_MBR) {
871 			switch (entry->type.mbr) {
872 			case DOSPTYP_386BSD:
873 				pref = entry->flags & 0x80 ? PREF_FBSD_ACT:
874 				    PREF_FBSD;
875 				break;
876 			case DOSPTYP_LINUX:
877 				pref = entry->flags & 0x80 ? PREF_LINUX_ACT:
878 				    PREF_LINUX;
879 				break;
880 			case 0x01:		/* DOS/Windows */
881 			case 0x04:
882 			case 0x06:
883 			case 0x0c:
884 			case 0x0e:
885 			case DOSPTYP_FAT32:
886 				pref = entry->flags & 0x80 ? PREF_DOS_ACT:
887 				    PREF_DOS;
888 				break;
889 			default:
890 				pref = PREF_NONE;
891 			}
892 		}
893 #endif /* LOADER_MBR_SUPPORT */
894 #ifdef LOADER_GPT_SUPPORT
895 		if (table->type == PTABLE_GPT) {
896 			if (entry->part.type == PART_DOS)
897 				pref = PREF_DOS;
898 			else if (entry->part.type == PART_FREEBSD_UFS ||
899 			    entry->part.type == PART_FREEBSD_ZFS)
900 				pref = PREF_FBSD;
901 			else
902 				pref = PREF_NONE;
903 		}
904 #endif /* LOADER_GPT_SUPPORT */
905 		if (pref < preflevel) {
906 			preflevel = pref;
907 			best = entry;
908 		}
909 	}
910 	if (best != NULL) {
911 		memcpy(part, &best->part, sizeof(*part));
912 		return (0);
913 	}
914 	return (ENOENT);
915 }
916 
917 int
918 ptable_iterate(const struct ptable *table, void *arg, ptable_iterate_t *iter)
919 {
920 	struct pentry *entry;
921 	char name[32];
922 	int ret = 0;
923 
924 	name[0] = '\0';
925 	STAILQ_FOREACH(entry, &table->entries, entry) {
926 #ifdef LOADER_MBR_SUPPORT
927 		if (table->type == PTABLE_MBR)
928 			sprintf(name, "s%d", entry->part.index);
929 		else
930 #endif
931 #ifdef LOADER_GPT_SUPPORT
932 		if (table->type == PTABLE_GPT)
933 			sprintf(name, "p%d", entry->part.index);
934 		else
935 #endif
936 #ifdef LOADER_VTOC8_SUPPORT
937 		if (table->type == PTABLE_VTOC8)
938 			sprintf(name, "%c", (uint8_t) 'a' +
939 			    entry->part.index);
940 		else
941 #endif
942 		if (table->type == PTABLE_BSD)
943 			sprintf(name, "%c", (uint8_t) 'a' +
944 			    entry->part.index);
945 		if ((ret = iter(arg, name, &entry->part)) != 0)
946 			return (ret);
947 	}
948 	return (ret);
949 }
950