xref: /freebsd/stand/common/part.c (revision 7899f917b1c0ea178f1d2be0cfb452086d079d23)
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 <stand.h>
28 #include <sys/param.h>
29 #include <sys/diskmbr.h>
30 #include <sys/disklabel.h>
31 #include <sys/endian.h>
32 #include <sys/gpt.h>
33 #include <sys/stddef.h>
34 #include <sys/queue.h>
35 
36 #include <fs/cd9660/iso.h>
37 
38 #include <zlib.h>
39 #include <part.h>
40 #include <uuid.h>
41 
42 #ifdef PART_DEBUG
43 #define	DPRINTF(fmt, args...) printf("%s: " fmt "\n", __func__, ## args)
44 #else
45 #define	DPRINTF(fmt, args...)	((void)0)
46 #endif
47 
48 #ifdef LOADER_GPT_SUPPORT
49 #define	MAXTBLSZ	64
50 static const uuid_t gpt_uuid_unused = GPT_ENT_TYPE_UNUSED;
51 static const uuid_t gpt_uuid_ms_basic_data = GPT_ENT_TYPE_MS_BASIC_DATA;
52 static const uuid_t gpt_uuid_freebsd_ufs = GPT_ENT_TYPE_FREEBSD_UFS;
53 static const uuid_t gpt_uuid_efi = GPT_ENT_TYPE_EFI;
54 static const uuid_t gpt_uuid_freebsd = GPT_ENT_TYPE_FREEBSD;
55 static const uuid_t gpt_uuid_freebsd_boot = GPT_ENT_TYPE_FREEBSD_BOOT;
56 static const uuid_t gpt_uuid_freebsd_swap = GPT_ENT_TYPE_FREEBSD_SWAP;
57 static const uuid_t gpt_uuid_freebsd_zfs = GPT_ENT_TYPE_FREEBSD_ZFS;
58 static const uuid_t gpt_uuid_freebsd_vinum = GPT_ENT_TYPE_FREEBSD_VINUM;
59 static const uuid_t gpt_uuid_apple_apfs = GPT_ENT_TYPE_APPLE_APFS;
60 #endif
61 
62 struct pentry {
63 	struct ptable_entry	part;
64 	uint64_t		flags;
65 	union {
66 		uint8_t bsd;
67 		uint8_t	mbr;
68 		uuid_t	gpt;
69 	} type;
70 	STAILQ_ENTRY(pentry)	entry;
71 };
72 
73 struct ptable {
74 	enum ptable_type	type;
75 	uint16_t		sectorsize;
76 	uint64_t		sectors;
77 
78 	STAILQ_HEAD(, pentry)	entries;
79 };
80 
81 static struct parttypes {
82 	enum partition_type	type;
83 	const char		*desc;
84 } ptypes[] = {
85 	{ PART_UNKNOWN,		"Unknown" },
86 	{ PART_EFI,		"EFI" },
87 	{ PART_FREEBSD,		"FreeBSD" },
88 	{ PART_FREEBSD_BOOT,	"FreeBSD boot" },
89 	{ PART_FREEBSD_UFS,	"FreeBSD UFS" },
90 	{ PART_FREEBSD_ZFS,	"FreeBSD ZFS" },
91 	{ PART_FREEBSD_SWAP,	"FreeBSD swap" },
92 	{ PART_FREEBSD_VINUM,	"FreeBSD vinum" },
93 	{ PART_LINUX,		"Linux" },
94 	{ PART_LINUX_SWAP,	"Linux swap" },
95 	{ PART_DOS,		"DOS/Windows" },
96 	{ PART_ISO9660,		"ISO9660" },
97 	{ PART_APFS,		"APFS" },
98 };
99 
100 const char *
101 parttype2str(enum partition_type type)
102 {
103 	size_t i;
104 
105 	for (i = 0; i < nitems(ptypes); i++)
106 		if (ptypes[i].type == type)
107 			return (ptypes[i].desc);
108 	return (ptypes[0].desc);
109 }
110 
111 #ifdef LOADER_GPT_SUPPORT
112 static void
113 uuid_letoh(uuid_t *uuid)
114 {
115 
116 	uuid->time_low = le32toh(uuid->time_low);
117 	uuid->time_mid = le16toh(uuid->time_mid);
118 	uuid->time_hi_and_version = le16toh(uuid->time_hi_and_version);
119 }
120 
121 static enum partition_type
122 gpt_parttype(uuid_t type)
123 {
124 
125 	if (uuid_equal(&type, &gpt_uuid_efi, NULL))
126 		return (PART_EFI);
127 	else if (uuid_equal(&type, &gpt_uuid_ms_basic_data, NULL))
128 		return (PART_DOS);
129 	else if (uuid_equal(&type, &gpt_uuid_freebsd_boot, NULL))
130 		return (PART_FREEBSD_BOOT);
131 	else if (uuid_equal(&type, &gpt_uuid_freebsd_ufs, NULL))
132 		return (PART_FREEBSD_UFS);
133 	else if (uuid_equal(&type, &gpt_uuid_freebsd_zfs, NULL))
134 		return (PART_FREEBSD_ZFS);
135 	else if (uuid_equal(&type, &gpt_uuid_freebsd_swap, NULL))
136 		return (PART_FREEBSD_SWAP);
137 	else if (uuid_equal(&type, &gpt_uuid_freebsd_vinum, NULL))
138 		return (PART_FREEBSD_VINUM);
139 	else if (uuid_equal(&type, &gpt_uuid_freebsd, NULL))
140 		return (PART_FREEBSD);
141 	else if (uuid_equal(&type, &gpt_uuid_apple_apfs, NULL))
142 		return (PART_APFS);
143 	return (PART_UNKNOWN);
144 }
145 
146 static struct gpt_hdr *
147 gpt_checkhdr(struct gpt_hdr *hdr, uint64_t lba_self, uint64_t lba_last,
148     uint16_t sectorsize)
149 {
150 	uint32_t sz, crc;
151 
152 	if (memcmp(hdr->hdr_sig, GPT_HDR_SIG, sizeof(hdr->hdr_sig)) != 0) {
153 		DPRINTF("no GPT signature");
154 		return (NULL);
155 	}
156 	sz = le32toh(hdr->hdr_size);
157 	if (sz < 92 || sz > sectorsize) {
158 		DPRINTF("invalid GPT header size: %d", sz);
159 		return (NULL);
160 	}
161 	crc = le32toh(hdr->hdr_crc_self);
162 	hdr->hdr_crc_self = crc32(0, Z_NULL, 0);
163 	if (crc32(hdr->hdr_crc_self, (const Bytef *)hdr, sz) != crc) {
164 		DPRINTF("GPT header's CRC doesn't match");
165 		return (NULL);
166 	}
167 	hdr->hdr_crc_self = crc;
168 	hdr->hdr_revision = le32toh(hdr->hdr_revision);
169 	if (hdr->hdr_revision < GPT_HDR_REVISION) {
170 		DPRINTF("unsupported GPT revision %d", hdr->hdr_revision);
171 		return (NULL);
172 	}
173 	hdr->hdr_lba_self = le64toh(hdr->hdr_lba_self);
174 	if (hdr->hdr_lba_self != lba_self) {
175 		DPRINTF("self LBA doesn't match");
176 		return (NULL);
177 	}
178 	hdr->hdr_lba_alt = le64toh(hdr->hdr_lba_alt);
179 	if (hdr->hdr_lba_alt == hdr->hdr_lba_self) {
180 		DPRINTF("invalid alternate LBA");
181 		return (NULL);
182 	}
183 	hdr->hdr_entries = le32toh(hdr->hdr_entries);
184 	hdr->hdr_entsz = le32toh(hdr->hdr_entsz);
185 	if (hdr->hdr_entries == 0 ||
186 	    hdr->hdr_entsz < sizeof(struct gpt_ent) ||
187 	    sectorsize % hdr->hdr_entsz != 0) {
188 		DPRINTF("invalid entry size or number of entries");
189 		return (NULL);
190 	}
191 	hdr->hdr_lba_start = le64toh(hdr->hdr_lba_start);
192 	hdr->hdr_lba_end = le64toh(hdr->hdr_lba_end);
193 	hdr->hdr_lba_table = le64toh(hdr->hdr_lba_table);
194 	hdr->hdr_crc_table = le32toh(hdr->hdr_crc_table);
195 	uuid_letoh(&hdr->hdr_uuid);
196 	return (hdr);
197 }
198 
199 static int
200 gpt_checktbl(const struct gpt_hdr *hdr, uint8_t *tbl, size_t size,
201     uint64_t lba_last)
202 {
203 	struct gpt_ent *ent;
204 	uint32_t i, cnt;
205 
206 	cnt = size / hdr->hdr_entsz;
207 	if (hdr->hdr_entries <= cnt) {
208 		cnt = hdr->hdr_entries;
209 		/* Check CRC only when buffer size is enough for table. */
210 		if (hdr->hdr_crc_table !=
211 		    crc32(0, tbl, hdr->hdr_entries * hdr->hdr_entsz)) {
212 			DPRINTF("GPT table's CRC doesn't match");
213 			return (-1);
214 		}
215 	}
216 	for (i = 0; i < cnt; i++) {
217 		ent = (struct gpt_ent *)(tbl + i * hdr->hdr_entsz);
218 		uuid_letoh(&ent->ent_type);
219 		if (uuid_equal(&ent->ent_type, &gpt_uuid_unused, NULL))
220 			continue;
221 		ent->ent_lba_start = le64toh(ent->ent_lba_start);
222 		ent->ent_lba_end = le64toh(ent->ent_lba_end);
223 	}
224 	return (0);
225 }
226 
227 static struct ptable *
228 ptable_gptread(struct ptable *table, void *dev, diskread_t dread)
229 {
230 	struct pentry *entry;
231 	struct gpt_hdr *phdr, hdr;
232 	struct gpt_ent *ent;
233 	uint8_t *buf, *tbl;
234 	uint64_t offset;
235 	int pri, sec;
236 	size_t size, i;
237 
238 	buf = malloc(table->sectorsize);
239 	if (buf == NULL)
240 		return (NULL);
241 	tbl = malloc(table->sectorsize * MAXTBLSZ);
242 	if (tbl == NULL) {
243 		free(buf);
244 		return (NULL);
245 	}
246 	/* Read the primary GPT header. */
247 	if (dread(dev, buf, 1, 1) != 0) {
248 		ptable_close(table);
249 		table = NULL;
250 		goto out;
251 	}
252 	pri = sec = 0;
253 	/* Check the primary GPT header. */
254 	phdr = gpt_checkhdr((struct gpt_hdr *)buf, 1, table->sectors - 1,
255 	    table->sectorsize);
256 	if (phdr != NULL) {
257 		/* Read the primary GPT table. */
258 		size = MIN(MAXTBLSZ,
259 		    howmany(phdr->hdr_entries * phdr->hdr_entsz,
260 		        table->sectorsize));
261 		if (dread(dev, tbl, size, phdr->hdr_lba_table) == 0 &&
262 		    gpt_checktbl(phdr, tbl, size * table->sectorsize,
263 		    table->sectors - 1) == 0) {
264 			memcpy(&hdr, phdr, sizeof(hdr));
265 			pri = 1;
266 		}
267 	}
268 	offset = pri ? hdr.hdr_lba_alt: table->sectors - 1;
269 	/* Read the backup GPT header. */
270 	if (dread(dev, buf, 1, offset) != 0)
271 		phdr = NULL;
272 	else
273 		phdr = gpt_checkhdr((struct gpt_hdr *)buf, offset,
274 		    table->sectors - 1, table->sectorsize);
275 	if (phdr != NULL) {
276 		/*
277 		 * Compare primary and backup headers.
278 		 * If they are equal, then we do not need to read backup
279 		 * table. If they are different, then prefer backup header
280 		 * and try to read backup table.
281 		 */
282 		if (pri == 0 ||
283 		    uuid_equal(&hdr.hdr_uuid, &phdr->hdr_uuid, NULL) == 0 ||
284 		    hdr.hdr_revision != phdr->hdr_revision ||
285 		    hdr.hdr_size != phdr->hdr_size ||
286 		    hdr.hdr_lba_start != phdr->hdr_lba_start ||
287 		    hdr.hdr_lba_end != phdr->hdr_lba_end ||
288 		    hdr.hdr_entries != phdr->hdr_entries ||
289 		    hdr.hdr_entsz != phdr->hdr_entsz ||
290 		    hdr.hdr_crc_table != phdr->hdr_crc_table) {
291 			/* Read the backup GPT table. */
292 			size = MIN(MAXTBLSZ,
293 				   howmany(phdr->hdr_entries * phdr->hdr_entsz,
294 				       table->sectorsize));
295 			if (dread(dev, tbl, size, phdr->hdr_lba_table) == 0 &&
296 			    gpt_checktbl(phdr, tbl, size * table->sectorsize,
297 			    table->sectors - 1) == 0) {
298 				memcpy(&hdr, phdr, sizeof(hdr));
299 				sec = 1;
300 			}
301 		}
302 	}
303 	if (pri == 0 && sec == 0) {
304 		/* Both primary and backup tables are invalid. */
305 		table->type = PTABLE_NONE;
306 		goto out;
307 	}
308 	DPRINTF("GPT detected");
309 	size = MIN(hdr.hdr_entries * hdr.hdr_entsz,
310 	    MAXTBLSZ * table->sectorsize);
311 
312 	/*
313 	 * If the disk's sector count is smaller than the sector count recorded
314 	 * in the disk's GPT table header, set the table->sectors to the value
315 	 * recorded in GPT tables. This is done to work around buggy firmware
316 	 * that returns truncated disk sizes.
317 	 *
318 	 * Note, this is still not a foolproof way to get disk's size. For
319 	 * example, an image file can be truncated when copied to smaller media.
320 	 */
321 	table->sectors = hdr.hdr_lba_alt + 1;
322 
323 	for (i = 0; i < size / hdr.hdr_entsz; i++) {
324 		ent = (struct gpt_ent *)(tbl + i * hdr.hdr_entsz);
325 		if (uuid_equal(&ent->ent_type, &gpt_uuid_unused, NULL))
326 			continue;
327 
328 		/* Simple sanity checks. */
329 		if (ent->ent_lba_start < hdr.hdr_lba_start ||
330 		    ent->ent_lba_end > hdr.hdr_lba_end ||
331 		    ent->ent_lba_start > ent->ent_lba_end)
332 			continue;
333 
334 		entry = malloc(sizeof(*entry));
335 		if (entry == NULL)
336 			break;
337 		entry->part.start = ent->ent_lba_start;
338 		entry->part.end = ent->ent_lba_end;
339 		entry->part.index = i + 1;
340 		entry->part.type = gpt_parttype(ent->ent_type);
341 		entry->flags = le64toh(ent->ent_attr);
342 		memcpy(&entry->type.gpt, &ent->ent_type, sizeof(uuid_t));
343 		STAILQ_INSERT_TAIL(&table->entries, entry, entry);
344 		DPRINTF("new GPT partition added");
345 	}
346 out:
347 	free(buf);
348 	free(tbl);
349 	return (table);
350 }
351 #endif /* LOADER_GPT_SUPPORT */
352 
353 #ifdef LOADER_MBR_SUPPORT
354 /* We do not need to support too many EBR partitions in the loader */
355 #define	MAXEBRENTRIES		8
356 static enum partition_type
357 mbr_parttype(uint8_t type)
358 {
359 
360 	switch (type) {
361 	case DOSPTYP_386BSD:
362 		return (PART_FREEBSD);
363 	case DOSPTYP_LINSWP:
364 		return (PART_LINUX_SWAP);
365 	case DOSPTYP_LINUX:
366 		return (PART_LINUX);
367 	case 0x01:
368 	case 0x04:
369 	case 0x06:
370 	case 0x07:
371 	case 0x0b:
372 	case 0x0c:
373 	case 0x0e:
374 		return (PART_DOS);
375 	}
376 	return (PART_UNKNOWN);
377 }
378 
379 static struct ptable *
380 ptable_ebrread(struct ptable *table, void *dev, diskread_t dread)
381 {
382 	struct dos_partition *dp;
383 	struct pentry *e1, *entry;
384 	uint32_t start, end, offset;
385 	u_char *buf;
386 	int i, index;
387 
388 	STAILQ_FOREACH(e1, &table->entries, entry) {
389 		if (e1->type.mbr == DOSPTYP_EXT ||
390 		    e1->type.mbr == DOSPTYP_EXTLBA)
391 			break;
392 	}
393 	if (e1 == NULL)
394 		return (table);
395 	index = 5;
396 	offset = e1->part.start;
397 	buf = malloc(table->sectorsize);
398 	if (buf == NULL)
399 		return (table);
400 	DPRINTF("EBR detected");
401 	for (i = 0; i < MAXEBRENTRIES; i++) {
402 #if 0	/* Some BIOSes return an incorrect number of sectors */
403 		if (offset >= table->sectors)
404 			break;
405 #endif
406 		if (dread(dev, buf, 1, offset) != 0)
407 			break;
408 		dp = (struct dos_partition *)(buf + DOSPARTOFF);
409 		if (dp[0].dp_typ == 0)
410 			break;
411 		start = le32toh(dp[0].dp_start);
412 		if (dp[0].dp_typ == DOSPTYP_EXT &&
413 		    dp[1].dp_typ == 0) {
414 			offset = e1->part.start + start;
415 			continue;
416 		}
417 		end = le32toh(dp[0].dp_size);
418 		entry = malloc(sizeof(*entry));
419 		if (entry == NULL)
420 			break;
421 		entry->part.start = offset + start;
422 		entry->part.end = entry->part.start + end - 1;
423 		entry->part.index = index++;
424 		entry->part.type = mbr_parttype(dp[0].dp_typ);
425 		entry->flags = dp[0].dp_flag;
426 		entry->type.mbr = dp[0].dp_typ;
427 		STAILQ_INSERT_TAIL(&table->entries, entry, entry);
428 		DPRINTF("new EBR partition added");
429 		if (dp[1].dp_typ == 0)
430 			break;
431 		offset = e1->part.start + le32toh(dp[1].dp_start);
432 	}
433 	free(buf);
434 	return (table);
435 }
436 #endif /* LOADER_MBR_SUPPORT */
437 
438 static enum partition_type
439 bsd_parttype(uint8_t type)
440 {
441 
442 	switch (type) {
443 	case FS_SWAP:
444 		return (PART_FREEBSD_SWAP);
445 	case FS_BSDFFS:
446 		return (PART_FREEBSD_UFS);
447 	case FS_VINUM:
448 		return (PART_FREEBSD_VINUM);
449 	case FS_ZFS:
450 		return (PART_FREEBSD_ZFS);
451 	}
452 	return (PART_UNKNOWN);
453 }
454 
455 static struct ptable *
456 ptable_bsdread(struct ptable *table, void *dev, diskread_t dread)
457 {
458 	struct disklabel *dl;
459 	struct partition *part;
460 	struct pentry *entry;
461 	uint8_t *buf;
462 	uint32_t raw_offset;
463 	int i;
464 
465 	if (table->sectorsize < sizeof(struct disklabel)) {
466 		DPRINTF("Too small sectorsize");
467 		return (table);
468 	}
469 	buf = malloc(table->sectorsize);
470 	if (buf == NULL)
471 		return (table);
472 	if (dread(dev, buf, 1, 1) != 0) {
473 		DPRINTF("read failed");
474 		ptable_close(table);
475 		table = NULL;
476 		goto out;
477 	}
478 	dl = (struct disklabel *)buf;
479 	if (le32toh(dl->d_magic) != DISKMAGIC &&
480 	    le32toh(dl->d_magic2) != DISKMAGIC)
481 		goto out;
482 	if (le32toh(dl->d_secsize) != table->sectorsize) {
483 		DPRINTF("unsupported sector size");
484 		goto out;
485 	}
486 	dl->d_npartitions = le16toh(dl->d_npartitions);
487 	if (dl->d_npartitions > 20 || dl->d_npartitions < 8) {
488 		DPRINTF("invalid number of partitions");
489 		goto out;
490 	}
491 	DPRINTF("BSD detected");
492 	part = &dl->d_partitions[0];
493 	raw_offset = le32toh(part[RAW_PART].p_offset);
494 	for (i = 0; i < dl->d_npartitions; i++, part++) {
495 		if (i == RAW_PART)
496 			continue;
497 		if (part->p_size == 0)
498 			continue;
499 		entry = malloc(sizeof(*entry));
500 		if (entry == NULL)
501 			break;
502 		entry->part.start = le32toh(part->p_offset) - raw_offset;
503 		entry->part.end = entry->part.start +
504 		    le32toh(part->p_size) - 1;
505 		entry->part.type = bsd_parttype(part->p_fstype);
506 		entry->part.index = i; /* starts from zero */
507 		entry->type.bsd = part->p_fstype;
508 		STAILQ_INSERT_TAIL(&table->entries, entry, entry);
509 		DPRINTF("new BSD partition added");
510 	}
511 	table->type = PTABLE_BSD;
512 out:
513 	free(buf);
514 	return (table);
515 }
516 
517 #define cdb2devb(bno)   ((bno) * ISO_DEFAULT_BLOCK_SIZE / table->sectorsize)
518 
519 static struct ptable *
520 ptable_iso9660read(struct ptable *table, void *dev, diskread_t dread)
521 {
522 	uint8_t *buf;
523 	struct iso_primary_descriptor *vd;
524 	struct pentry *entry;
525 
526 	buf = malloc(table->sectorsize);
527 	if (buf == NULL)
528 		return (table);
529 
530 	if (dread(dev, buf, 1, cdb2devb(16)) != 0) {
531 		DPRINTF("read failed");
532 		ptable_close(table);
533 		table = NULL;
534 		goto out;
535 	}
536 	vd = (struct iso_primary_descriptor *)buf;
537 	if (bcmp(vd->id, ISO_STANDARD_ID, sizeof vd->id) != 0)
538 		goto out;
539 
540 	entry = malloc(sizeof(*entry));
541 	if (entry == NULL)
542 		goto out;
543 	entry->part.start = 0;
544 	entry->part.end = table->sectors;
545 	entry->part.type = PART_ISO9660;
546 	entry->part.index = 0;
547 	STAILQ_INSERT_TAIL(&table->entries, entry, entry);
548 
549 	table->type = PTABLE_ISO9660;
550 
551 out:
552 	free(buf);
553 	return (table);
554 }
555 
556 struct ptable *
557 ptable_open(void *dev, uint64_t sectors, uint16_t sectorsize,
558     diskread_t *dread)
559 {
560 	struct dos_partition *dp;
561 	struct ptable *table;
562 	uint8_t *buf;
563 #ifdef LOADER_MBR_SUPPORT
564 	struct pentry *entry;
565 	uint32_t start, end;
566 	int has_ext;
567 #endif
568 	table = NULL;
569 	dp = NULL;
570 	buf = malloc(sectorsize);
571 	if (buf == NULL)
572 		return (NULL);
573 	/* First, read the MBR. */
574 	if (dread(dev, buf, 1, DOSBBSECTOR) != 0) {
575 		DPRINTF("read failed");
576 		goto out;
577 	}
578 
579 	table = malloc(sizeof(*table));
580 	if (table == NULL)
581 		goto out;
582 	table->sectors = sectors;
583 	table->sectorsize = sectorsize;
584 	table->type = PTABLE_NONE;
585 	STAILQ_INIT(&table->entries);
586 
587 	if (ptable_iso9660read(table, dev, dread) == NULL) {
588 		/* Read error. */
589 		table = NULL;
590 		goto out;
591 	} else if (table->type == PTABLE_ISO9660)
592 		goto out;
593 
594 	/* Check the BSD label. */
595 	if (ptable_bsdread(table, dev, dread) == NULL) { /* Read error. */
596 		table = NULL;
597 		goto out;
598 	} else if (table->type == PTABLE_BSD)
599 		goto out;
600 
601 #if defined(LOADER_GPT_SUPPORT) || defined(LOADER_MBR_SUPPORT)
602 	/* Check the MBR magic. */
603 	if (buf[DOSMAGICOFFSET] != 0x55 ||
604 	    buf[DOSMAGICOFFSET + 1] != 0xaa) {
605 		DPRINTF("magic sequence not found");
606 #if defined(LOADER_GPT_SUPPORT)
607 		/* There is no PMBR, check that we have backup GPT */
608 		table->type = PTABLE_GPT;
609 		table = ptable_gptread(table, dev, dread);
610 #endif
611 		goto out;
612 	}
613 	/* Check that we have PMBR. Also do some validation. */
614 	dp = malloc(NDOSPART * sizeof(struct dos_partition));
615 	if (dp == NULL)
616 		goto out;
617 	bcopy(buf + DOSPARTOFF, dp, NDOSPART * sizeof(struct dos_partition));
618 
619 	/*
620 	 * In mac we can have PMBR partition in hybrid MBR;
621 	 * that is, MBR partition which has DOSPTYP_PMBR entry defined as
622 	 * start sector 1. After DOSPTYP_PMBR, there may be other partitions.
623 	 * UEFI compliant PMBR has no other partitions.
624 	 */
625 	for (int i = 0; i < NDOSPART; i++) {
626 		if (dp[i].dp_flag != 0 && dp[i].dp_flag != 0x80) {
627 			DPRINTF("invalid partition flag %x", dp[i].dp_flag);
628 			goto out;
629 		}
630 #ifdef LOADER_GPT_SUPPORT
631 		if (dp[i].dp_typ == DOSPTYP_PMBR && dp[i].dp_start == 1) {
632 			table->type = PTABLE_GPT;
633 			DPRINTF("PMBR detected");
634 		}
635 #endif
636 	}
637 #ifdef LOADER_GPT_SUPPORT
638 	if (table->type == PTABLE_GPT) {
639 		table = ptable_gptread(table, dev, dread);
640 		goto out;
641 	}
642 #endif
643 #ifdef LOADER_MBR_SUPPORT
644 	/* Read MBR. */
645 	DPRINTF("MBR detected");
646 	table->type = PTABLE_MBR;
647 	for (int i = has_ext = 0; i < NDOSPART; i++) {
648 		if (dp[i].dp_typ == 0)
649 			continue;
650 		start = le32dec(&(dp[i].dp_start));
651 		end = le32dec(&(dp[i].dp_size));
652 		if (start == 0 || end == 0)
653 			continue;
654 #if 0	/* Some BIOSes return an incorrect number of sectors */
655 		if (start + end - 1 >= sectors)
656 			continue;	/* XXX: ignore */
657 #endif
658 		if (dp[i].dp_typ == DOSPTYP_EXT ||
659 		    dp[i].dp_typ == DOSPTYP_EXTLBA)
660 			has_ext = 1;
661 		entry = malloc(sizeof(*entry));
662 		if (entry == NULL)
663 			break;
664 		entry->part.start = start;
665 		entry->part.end = start + end - 1;
666 		entry->part.index = i + 1;
667 		entry->part.type = mbr_parttype(dp[i].dp_typ);
668 		entry->flags = dp[i].dp_flag;
669 		entry->type.mbr = dp[i].dp_typ;
670 		STAILQ_INSERT_TAIL(&table->entries, entry, entry);
671 		DPRINTF("new MBR partition added");
672 	}
673 	if (has_ext) {
674 		table = ptable_ebrread(table, dev, dread);
675 		/* FALLTHROUGH */
676 	}
677 #endif /* LOADER_MBR_SUPPORT */
678 #endif /* LOADER_MBR_SUPPORT || LOADER_GPT_SUPPORT */
679 out:
680 	free(dp);
681 	free(buf);
682 	return (table);
683 }
684 
685 void
686 ptable_close(struct ptable *table)
687 {
688 	struct pentry *entry;
689 
690 	if (table == NULL)
691 		return;
692 
693 	while (!STAILQ_EMPTY(&table->entries)) {
694 		entry = STAILQ_FIRST(&table->entries);
695 		STAILQ_REMOVE_HEAD(&table->entries, entry);
696 		free(entry);
697 	}
698 	free(table);
699 }
700 
701 enum ptable_type
702 ptable_gettype(const struct ptable *table)
703 {
704 
705 	return (table->type);
706 }
707 
708 int
709 ptable_getsize(const struct ptable *table, uint64_t *sizep)
710 {
711 	uint64_t tmp = table->sectors * table->sectorsize;
712 
713 	if (tmp < table->sectors)
714 		return (EOVERFLOW);
715 
716 	if (sizep != NULL)
717 		*sizep = tmp;
718 	return (0);
719 }
720 
721 int
722 ptable_getpart(const struct ptable *table, struct ptable_entry *part, int index)
723 {
724 	struct pentry *entry;
725 
726 	if (part == NULL || table == NULL)
727 		return (EINVAL);
728 
729 	STAILQ_FOREACH(entry, &table->entries, entry) {
730 		if (entry->part.index != index)
731 			continue;
732 		memcpy(part, &entry->part, sizeof(*part));
733 		return (0);
734 	}
735 	return (ENOENT);
736 }
737 
738 /*
739  * Search for a slice with the following preferences:
740  *
741  * 1: Active FreeBSD slice
742  * 2: Non-active FreeBSD slice
743  * 3: Active Linux slice
744  * 4: non-active Linux slice
745  * 5: Active FAT/FAT32 slice
746  * 6: non-active FAT/FAT32 slice
747  */
748 #define	PREF_RAWDISK	0
749 #define	PREF_FBSD_ACT	1
750 #define	PREF_FBSD	2
751 #define	PREF_LINUX_ACT	3
752 #define	PREF_LINUX	4
753 #define	PREF_DOS_ACT	5
754 #define	PREF_DOS	6
755 #define	PREF_NONE	7
756 int
757 ptable_getbestpart(const struct ptable *table, struct ptable_entry *part)
758 {
759 	struct pentry *entry, *best;
760 	int pref, preflevel;
761 
762 	if (part == NULL || table == NULL)
763 		return (EINVAL);
764 
765 	best = NULL;
766 	preflevel = pref = PREF_NONE;
767 	STAILQ_FOREACH(entry, &table->entries, entry) {
768 #ifdef LOADER_MBR_SUPPORT
769 		if (table->type == PTABLE_MBR) {
770 			switch (entry->type.mbr) {
771 			case DOSPTYP_386BSD:
772 				pref = entry->flags & 0x80 ? PREF_FBSD_ACT:
773 				    PREF_FBSD;
774 				break;
775 			case DOSPTYP_LINUX:
776 				pref = entry->flags & 0x80 ? PREF_LINUX_ACT:
777 				    PREF_LINUX;
778 				break;
779 			case 0x01:		/* DOS/Windows */
780 			case 0x04:
781 			case 0x06:
782 			case 0x0c:
783 			case 0x0e:
784 			case DOSPTYP_FAT32:
785 				pref = entry->flags & 0x80 ? PREF_DOS_ACT:
786 				    PREF_DOS;
787 				break;
788 			default:
789 				pref = PREF_NONE;
790 			}
791 		}
792 #endif /* LOADER_MBR_SUPPORT */
793 #ifdef LOADER_GPT_SUPPORT
794 		if (table->type == PTABLE_GPT) {
795 			if (entry->part.type == PART_DOS)
796 				pref = PREF_DOS;
797 			else if (entry->part.type == PART_FREEBSD_UFS ||
798 			    entry->part.type == PART_FREEBSD_ZFS)
799 				pref = PREF_FBSD;
800 			else
801 				pref = PREF_NONE;
802 		}
803 #endif /* LOADER_GPT_SUPPORT */
804 		if (pref < preflevel) {
805 			preflevel = pref;
806 			best = entry;
807 		}
808 	}
809 	if (best != NULL) {
810 		memcpy(part, &best->part, sizeof(*part));
811 		return (0);
812 	}
813 	return (ENOENT);
814 }
815 
816 int
817 ptable_iterate(const struct ptable *table, void *arg, ptable_iterate_t *iter)
818 {
819 	struct pentry *entry;
820 	char name[32];
821 	int ret = 0;
822 
823 	name[0] = '\0';
824 	STAILQ_FOREACH(entry, &table->entries, entry) {
825 #ifdef LOADER_MBR_SUPPORT
826 		if (table->type == PTABLE_MBR)
827 			sprintf(name, "s%d", entry->part.index);
828 		else
829 #endif
830 #ifdef LOADER_GPT_SUPPORT
831 		if (table->type == PTABLE_GPT)
832 			sprintf(name, "p%d", entry->part.index);
833 		else
834 #endif
835 		if (table->type == PTABLE_BSD)
836 			sprintf(name, "%c", (uint8_t) 'a' +
837 			    entry->part.index);
838 		if ((ret = iter(arg, name, &entry->part)) != 0)
839 			return (ret);
840 	}
841 	return (ret);
842 }
843