xref: /freebsd/sys/geom/part/g_part_gpt.c (revision 28f42739a547ffe0b5dfaaf9f49fb4c4813aa232)
1 /*-
2  * Copyright (c) 2002, 2005-2007, 2011 Marcel Moolenaar
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  *
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include <sys/param.h>
31 #include <sys/bio.h>
32 #include <sys/diskmbr.h>
33 #include <sys/endian.h>
34 #include <sys/gpt.h>
35 #include <sys/kernel.h>
36 #include <sys/kobj.h>
37 #include <sys/limits.h>
38 #include <sys/lock.h>
39 #include <sys/malloc.h>
40 #include <sys/mutex.h>
41 #include <sys/queue.h>
42 #include <sys/sbuf.h>
43 #include <sys/systm.h>
44 #include <sys/sysctl.h>
45 #include <sys/uuid.h>
46 #include <geom/geom.h>
47 #include <geom/geom_int.h>
48 #include <geom/part/g_part.h>
49 
50 #include "g_part_if.h"
51 
52 FEATURE(geom_part_gpt, "GEOM partitioning class for GPT partitions support");
53 
54 CTASSERT(offsetof(struct gpt_hdr, padding) == 92);
55 CTASSERT(sizeof(struct gpt_ent) == 128);
56 
57 #define	EQUUID(a,b)	(memcmp(a, b, sizeof(struct uuid)) == 0)
58 
59 #define	MBRSIZE		512
60 
61 enum gpt_elt {
62 	GPT_ELT_PRIHDR,
63 	GPT_ELT_PRITBL,
64 	GPT_ELT_SECHDR,
65 	GPT_ELT_SECTBL,
66 	GPT_ELT_COUNT
67 };
68 
69 enum gpt_state {
70 	GPT_STATE_UNKNOWN,	/* Not determined. */
71 	GPT_STATE_MISSING,	/* No signature found. */
72 	GPT_STATE_CORRUPT,	/* Checksum mismatch. */
73 	GPT_STATE_INVALID,	/* Nonconformant/invalid. */
74 	GPT_STATE_OK		/* Perfectly fine. */
75 };
76 
77 struct g_part_gpt_table {
78 	struct g_part_table	base;
79 	u_char			mbr[MBRSIZE];
80 	struct gpt_hdr		*hdr;
81 	quad_t			lba[GPT_ELT_COUNT];
82 	enum gpt_state		state[GPT_ELT_COUNT];
83 	int			bootcamp;
84 };
85 
86 struct g_part_gpt_entry {
87 	struct g_part_entry	base;
88 	struct gpt_ent		ent;
89 };
90 
91 static void g_gpt_printf_utf16(struct sbuf *, uint16_t *, size_t);
92 static void g_gpt_utf8_to_utf16(const uint8_t *, uint16_t *, size_t);
93 static void g_gpt_set_defaults(struct g_part_table *, struct g_provider *);
94 
95 static int g_part_gpt_add(struct g_part_table *, struct g_part_entry *,
96     struct g_part_parms *);
97 static int g_part_gpt_bootcode(struct g_part_table *, struct g_part_parms *);
98 static int g_part_gpt_create(struct g_part_table *, struct g_part_parms *);
99 static int g_part_gpt_destroy(struct g_part_table *, struct g_part_parms *);
100 static void g_part_gpt_dumpconf(struct g_part_table *, struct g_part_entry *,
101     struct sbuf *, const char *);
102 static int g_part_gpt_dumpto(struct g_part_table *, struct g_part_entry *);
103 static int g_part_gpt_modify(struct g_part_table *, struct g_part_entry *,
104     struct g_part_parms *);
105 static const char *g_part_gpt_name(struct g_part_table *, struct g_part_entry *,
106     char *, size_t);
107 static int g_part_gpt_probe(struct g_part_table *, struct g_consumer *);
108 static int g_part_gpt_read(struct g_part_table *, struct g_consumer *);
109 static int g_part_gpt_setunset(struct g_part_table *table,
110     struct g_part_entry *baseentry, const char *attrib, unsigned int set);
111 static const char *g_part_gpt_type(struct g_part_table *, struct g_part_entry *,
112     char *, size_t);
113 static int g_part_gpt_write(struct g_part_table *, struct g_consumer *);
114 static int g_part_gpt_resize(struct g_part_table *, struct g_part_entry *,
115     struct g_part_parms *);
116 static int g_part_gpt_recover(struct g_part_table *);
117 
118 static kobj_method_t g_part_gpt_methods[] = {
119 	KOBJMETHOD(g_part_add,		g_part_gpt_add),
120 	KOBJMETHOD(g_part_bootcode,	g_part_gpt_bootcode),
121 	KOBJMETHOD(g_part_create,	g_part_gpt_create),
122 	KOBJMETHOD(g_part_destroy,	g_part_gpt_destroy),
123 	KOBJMETHOD(g_part_dumpconf,	g_part_gpt_dumpconf),
124 	KOBJMETHOD(g_part_dumpto,	g_part_gpt_dumpto),
125 	KOBJMETHOD(g_part_modify,	g_part_gpt_modify),
126 	KOBJMETHOD(g_part_resize,	g_part_gpt_resize),
127 	KOBJMETHOD(g_part_name,		g_part_gpt_name),
128 	KOBJMETHOD(g_part_probe,	g_part_gpt_probe),
129 	KOBJMETHOD(g_part_read,		g_part_gpt_read),
130 	KOBJMETHOD(g_part_recover,	g_part_gpt_recover),
131 	KOBJMETHOD(g_part_setunset,	g_part_gpt_setunset),
132 	KOBJMETHOD(g_part_type,		g_part_gpt_type),
133 	KOBJMETHOD(g_part_write,	g_part_gpt_write),
134 	{ 0, 0 }
135 };
136 
137 static struct g_part_scheme g_part_gpt_scheme = {
138 	"GPT",
139 	g_part_gpt_methods,
140 	sizeof(struct g_part_gpt_table),
141 	.gps_entrysz = sizeof(struct g_part_gpt_entry),
142 	.gps_minent = 128,
143 	.gps_maxent = 4096,
144 	.gps_bootcodesz = MBRSIZE,
145 };
146 G_PART_SCHEME_DECLARE(g_part_gpt);
147 
148 static struct uuid gpt_uuid_apple_boot = GPT_ENT_TYPE_APPLE_BOOT;
149 static struct uuid gpt_uuid_apple_hfs = GPT_ENT_TYPE_APPLE_HFS;
150 static struct uuid gpt_uuid_apple_label = GPT_ENT_TYPE_APPLE_LABEL;
151 static struct uuid gpt_uuid_apple_raid = GPT_ENT_TYPE_APPLE_RAID;
152 static struct uuid gpt_uuid_apple_raid_offline = GPT_ENT_TYPE_APPLE_RAID_OFFLINE;
153 static struct uuid gpt_uuid_apple_tv_recovery = GPT_ENT_TYPE_APPLE_TV_RECOVERY;
154 static struct uuid gpt_uuid_apple_ufs = GPT_ENT_TYPE_APPLE_UFS;
155 static struct uuid gpt_uuid_bios_boot = GPT_ENT_TYPE_BIOS_BOOT;
156 static struct uuid gpt_uuid_efi = GPT_ENT_TYPE_EFI;
157 static struct uuid gpt_uuid_freebsd = GPT_ENT_TYPE_FREEBSD;
158 static struct uuid gpt_uuid_freebsd_boot = GPT_ENT_TYPE_FREEBSD_BOOT;
159 static struct uuid gpt_uuid_freebsd_nandfs = GPT_ENT_TYPE_FREEBSD_NANDFS;
160 static struct uuid gpt_uuid_freebsd_swap = GPT_ENT_TYPE_FREEBSD_SWAP;
161 static struct uuid gpt_uuid_freebsd_ufs = GPT_ENT_TYPE_FREEBSD_UFS;
162 static struct uuid gpt_uuid_freebsd_vinum = GPT_ENT_TYPE_FREEBSD_VINUM;
163 static struct uuid gpt_uuid_freebsd_zfs = GPT_ENT_TYPE_FREEBSD_ZFS;
164 static struct uuid gpt_uuid_linux_data = GPT_ENT_TYPE_LINUX_DATA;
165 static struct uuid gpt_uuid_linux_lvm = GPT_ENT_TYPE_LINUX_LVM;
166 static struct uuid gpt_uuid_linux_raid = GPT_ENT_TYPE_LINUX_RAID;
167 static struct uuid gpt_uuid_linux_swap = GPT_ENT_TYPE_LINUX_SWAP;
168 static struct uuid gpt_uuid_vmfs = GPT_ENT_TYPE_VMFS;
169 static struct uuid gpt_uuid_vmkdiag = GPT_ENT_TYPE_VMKDIAG;
170 static struct uuid gpt_uuid_vmreserved = GPT_ENT_TYPE_VMRESERVED;
171 static struct uuid gpt_uuid_vmvsanhdr = GPT_ENT_TYPE_VMVSANHDR;
172 static struct uuid gpt_uuid_ms_basic_data = GPT_ENT_TYPE_MS_BASIC_DATA;
173 static struct uuid gpt_uuid_ms_reserved = GPT_ENT_TYPE_MS_RESERVED;
174 static struct uuid gpt_uuid_ms_ldm_data = GPT_ENT_TYPE_MS_LDM_DATA;
175 static struct uuid gpt_uuid_ms_ldm_metadata = GPT_ENT_TYPE_MS_LDM_METADATA;
176 static struct uuid gpt_uuid_netbsd_ccd = GPT_ENT_TYPE_NETBSD_CCD;
177 static struct uuid gpt_uuid_netbsd_cgd = GPT_ENT_TYPE_NETBSD_CGD;
178 static struct uuid gpt_uuid_netbsd_ffs = GPT_ENT_TYPE_NETBSD_FFS;
179 static struct uuid gpt_uuid_netbsd_lfs = GPT_ENT_TYPE_NETBSD_LFS;
180 static struct uuid gpt_uuid_netbsd_raid = GPT_ENT_TYPE_NETBSD_RAID;
181 static struct uuid gpt_uuid_netbsd_swap = GPT_ENT_TYPE_NETBSD_SWAP;
182 static struct uuid gpt_uuid_mbr = GPT_ENT_TYPE_MBR;
183 static struct uuid gpt_uuid_unused = GPT_ENT_TYPE_UNUSED;
184 static struct uuid gpt_uuid_dfbsd_swap = GPT_ENT_TYPE_DRAGONFLY_SWAP;
185 static struct uuid gpt_uuid_dfbsd_ufs1 = GPT_ENT_TYPE_DRAGONFLY_UFS1;
186 static struct uuid gpt_uuid_dfbsd_vinum = GPT_ENT_TYPE_DRAGONFLY_VINUM;
187 static struct uuid gpt_uuid_dfbsd_ccd = GPT_ENT_TYPE_DRAGONFLY_CCD;
188 static struct uuid gpt_uuid_dfbsd_legacy = GPT_ENT_TYPE_DRAGONFLY_LEGACY;
189 static struct uuid gpt_uuid_dfbsd_hammer = GPT_ENT_TYPE_DRAGONFLY_HAMMER;
190 static struct uuid gpt_uuid_dfbsd_hammer2 = GPT_ENT_TYPE_DRAGONFLY_HAMMER2;
191 static struct uuid gpt_uuid_dfbsd_label32 = GPT_ENT_TYPE_DRAGONFLY_LABEL32;
192 static struct uuid gpt_uuid_dfbsd_label64 = GPT_ENT_TYPE_DRAGONFLY_LABEL64;
193 static struct uuid gpt_uuid_prep_boot = GPT_ENT_TYPE_PREP_BOOT;
194 
195 static struct g_part_uuid_alias {
196 	struct uuid *uuid;
197 	int alias;
198 	int mbrtype;
199 } gpt_uuid_alias_match[] = {
200 	{ &gpt_uuid_apple_boot,		G_PART_ALIAS_APPLE_BOOT,	 0xab },
201 	{ &gpt_uuid_apple_hfs,		G_PART_ALIAS_APPLE_HFS,		 0xaf },
202 	{ &gpt_uuid_apple_label,	G_PART_ALIAS_APPLE_LABEL,	 0 },
203 	{ &gpt_uuid_apple_raid,		G_PART_ALIAS_APPLE_RAID,	 0 },
204 	{ &gpt_uuid_apple_raid_offline,	G_PART_ALIAS_APPLE_RAID_OFFLINE, 0 },
205 	{ &gpt_uuid_apple_tv_recovery,	G_PART_ALIAS_APPLE_TV_RECOVERY,	 0 },
206 	{ &gpt_uuid_apple_ufs,		G_PART_ALIAS_APPLE_UFS,		 0 },
207 	{ &gpt_uuid_bios_boot,		G_PART_ALIAS_BIOS_BOOT,		 0 },
208 	{ &gpt_uuid_efi, 		G_PART_ALIAS_EFI,		 0xee },
209 	{ &gpt_uuid_freebsd,		G_PART_ALIAS_FREEBSD,		 0xa5 },
210 	{ &gpt_uuid_freebsd_boot, 	G_PART_ALIAS_FREEBSD_BOOT,	 0 },
211 	{ &gpt_uuid_freebsd_nandfs, 	G_PART_ALIAS_FREEBSD_NANDFS,	 0 },
212 	{ &gpt_uuid_freebsd_swap,	G_PART_ALIAS_FREEBSD_SWAP,	 0 },
213 	{ &gpt_uuid_freebsd_ufs,	G_PART_ALIAS_FREEBSD_UFS,	 0 },
214 	{ &gpt_uuid_freebsd_vinum,	G_PART_ALIAS_FREEBSD_VINUM,	 0 },
215 	{ &gpt_uuid_freebsd_zfs,	G_PART_ALIAS_FREEBSD_ZFS,	 0 },
216 	{ &gpt_uuid_linux_data,		G_PART_ALIAS_LINUX_DATA,	 0x0b },
217 	{ &gpt_uuid_linux_lvm,		G_PART_ALIAS_LINUX_LVM,		 0 },
218 	{ &gpt_uuid_linux_raid,		G_PART_ALIAS_LINUX_RAID,	 0 },
219 	{ &gpt_uuid_linux_swap,		G_PART_ALIAS_LINUX_SWAP,	 0 },
220 	{ &gpt_uuid_vmfs,		G_PART_ALIAS_VMFS,		 0 },
221 	{ &gpt_uuid_vmkdiag,		G_PART_ALIAS_VMKDIAG,		 0 },
222 	{ &gpt_uuid_vmreserved,		G_PART_ALIAS_VMRESERVED,	 0 },
223 	{ &gpt_uuid_vmvsanhdr,		G_PART_ALIAS_VMVSANHDR,		 0 },
224 	{ &gpt_uuid_mbr,		G_PART_ALIAS_MBR,		 0 },
225 	{ &gpt_uuid_ms_basic_data,	G_PART_ALIAS_MS_BASIC_DATA,	 0x0b },
226 	{ &gpt_uuid_ms_ldm_data,	G_PART_ALIAS_MS_LDM_DATA,	 0 },
227 	{ &gpt_uuid_ms_ldm_metadata,	G_PART_ALIAS_MS_LDM_METADATA,	 0 },
228 	{ &gpt_uuid_ms_reserved,	G_PART_ALIAS_MS_RESERVED,	 0 },
229 	{ &gpt_uuid_netbsd_ccd,		G_PART_ALIAS_NETBSD_CCD,	 0 },
230 	{ &gpt_uuid_netbsd_cgd,		G_PART_ALIAS_NETBSD_CGD,	 0 },
231 	{ &gpt_uuid_netbsd_ffs,		G_PART_ALIAS_NETBSD_FFS,	 0 },
232 	{ &gpt_uuid_netbsd_lfs,		G_PART_ALIAS_NETBSD_LFS,	 0 },
233 	{ &gpt_uuid_netbsd_raid,	G_PART_ALIAS_NETBSD_RAID,	 0 },
234 	{ &gpt_uuid_netbsd_swap,	G_PART_ALIAS_NETBSD_SWAP,	 0 },
235 	{ &gpt_uuid_dfbsd_swap,		G_PART_ALIAS_DFBSD_SWAP,	 0 },
236 	{ &gpt_uuid_dfbsd_ufs1,		G_PART_ALIAS_DFBSD_UFS,		 0 },
237 	{ &gpt_uuid_dfbsd_vinum,	G_PART_ALIAS_DFBSD_VINUM,	 0 },
238 	{ &gpt_uuid_dfbsd_ccd,		G_PART_ALIAS_DFBSD_CCD,		 0 },
239 	{ &gpt_uuid_dfbsd_legacy,	G_PART_ALIAS_DFBSD_LEGACY,	 0 },
240 	{ &gpt_uuid_dfbsd_hammer,	G_PART_ALIAS_DFBSD_HAMMER,	 0 },
241 	{ &gpt_uuid_dfbsd_hammer2,	G_PART_ALIAS_DFBSD_HAMMER2,	 0 },
242 	{ &gpt_uuid_dfbsd_label32,	G_PART_ALIAS_DFBSD,		 0xa5 },
243 	{ &gpt_uuid_dfbsd_label64,	G_PART_ALIAS_DFBSD64,		 0xa5 },
244 	{ &gpt_uuid_prep_boot,		G_PART_ALIAS_PREP_BOOT,		 0x41 },
245 	{ NULL, 0, 0 }
246 };
247 
248 static int
249 gpt_write_mbr_entry(u_char *mbr, int idx, int typ, quad_t start,
250     quad_t end)
251 {
252 
253 	if (typ == 0 || start > UINT32_MAX || end > UINT32_MAX)
254 		return (EINVAL);
255 
256 	mbr += DOSPARTOFF + idx * DOSPARTSIZE;
257 	mbr[0] = 0;
258 	if (start == 1) {
259 		/*
260 		 * Treat the PMBR partition specially to maximize
261 		 * interoperability with BIOSes.
262 		 */
263 		mbr[1] = mbr[3] = 0;
264 		mbr[2] = 2;
265 	} else
266 		mbr[1] = mbr[2] = mbr[3] = 0xff;
267 	mbr[4] = typ;
268 	mbr[5] = mbr[6] = mbr[7] = 0xff;
269 	le32enc(mbr + 8, (uint32_t)start);
270 	le32enc(mbr + 12, (uint32_t)(end - start + 1));
271 	return (0);
272 }
273 
274 static int
275 gpt_map_type(struct uuid *t)
276 {
277 	struct g_part_uuid_alias *uap;
278 
279 	for (uap = &gpt_uuid_alias_match[0]; uap->uuid; uap++) {
280 		if (EQUUID(t, uap->uuid))
281 			return (uap->mbrtype);
282 	}
283 	return (0);
284 }
285 
286 static void
287 gpt_create_pmbr(struct g_part_gpt_table *table, struct g_provider *pp)
288 {
289 
290 	bzero(table->mbr + DOSPARTOFF, DOSPARTSIZE * NDOSPART);
291 	gpt_write_mbr_entry(table->mbr, 0, 0xee, 1,
292 	    MIN(pp->mediasize / pp->sectorsize - 1, UINT32_MAX));
293 	le16enc(table->mbr + DOSMAGICOFFSET, DOSMAGIC);
294 }
295 
296 /*
297  * Under Boot Camp the PMBR partition (type 0xEE) doesn't cover the
298  * whole disk anymore. Rather, it covers the GPT table and the EFI
299  * system partition only. This way the HFS+ partition and any FAT
300  * partitions can be added to the MBR without creating an overlap.
301  */
302 static int
303 gpt_is_bootcamp(struct g_part_gpt_table *table, const char *provname)
304 {
305 	uint8_t *p;
306 
307 	p = table->mbr + DOSPARTOFF;
308 	if (p[4] != 0xee || le32dec(p + 8) != 1)
309 		return (0);
310 
311 	p += DOSPARTSIZE;
312 	if (p[4] != 0xaf)
313 		return (0);
314 
315 	printf("GEOM: %s: enabling Boot Camp\n", provname);
316 	return (1);
317 }
318 
319 static void
320 gpt_update_bootcamp(struct g_part_table *basetable, struct g_provider *pp)
321 {
322 	struct g_part_entry *baseentry;
323 	struct g_part_gpt_entry *entry;
324 	struct g_part_gpt_table *table;
325 	int bootable, error, index, slices, typ;
326 
327 	table = (struct g_part_gpt_table *)basetable;
328 
329 	bootable = -1;
330 	for (index = 0; index < NDOSPART; index++) {
331 		if (table->mbr[DOSPARTOFF + DOSPARTSIZE * index])
332 			bootable = index;
333 	}
334 
335 	bzero(table->mbr + DOSPARTOFF, DOSPARTSIZE * NDOSPART);
336 	slices = 0;
337 	LIST_FOREACH(baseentry, &basetable->gpt_entry, gpe_entry) {
338 		if (baseentry->gpe_deleted)
339 			continue;
340 		index = baseentry->gpe_index - 1;
341 		if (index >= NDOSPART)
342 			continue;
343 
344 		entry = (struct g_part_gpt_entry *)baseentry;
345 
346 		switch (index) {
347 		case 0:	/* This must be the EFI system partition. */
348 			if (!EQUUID(&entry->ent.ent_type, &gpt_uuid_efi))
349 				goto disable;
350 			error = gpt_write_mbr_entry(table->mbr, index, 0xee,
351 			    1ull, entry->ent.ent_lba_end);
352 			break;
353 		case 1:	/* This must be the HFS+ partition. */
354 			if (!EQUUID(&entry->ent.ent_type, &gpt_uuid_apple_hfs))
355 				goto disable;
356 			error = gpt_write_mbr_entry(table->mbr, index, 0xaf,
357 			    entry->ent.ent_lba_start, entry->ent.ent_lba_end);
358 			break;
359 		default:
360 			typ = gpt_map_type(&entry->ent.ent_type);
361 			error = gpt_write_mbr_entry(table->mbr, index, typ,
362 			    entry->ent.ent_lba_start, entry->ent.ent_lba_end);
363 			break;
364 		}
365 		if (error)
366 			continue;
367 
368 		if (index == bootable)
369 			table->mbr[DOSPARTOFF + DOSPARTSIZE * index] = 0x80;
370 		slices |= 1 << index;
371 	}
372 	if ((slices & 3) == 3)
373 		return;
374 
375  disable:
376 	table->bootcamp = 0;
377 	gpt_create_pmbr(table, pp);
378 }
379 
380 static struct gpt_hdr *
381 gpt_read_hdr(struct g_part_gpt_table *table, struct g_consumer *cp,
382     enum gpt_elt elt)
383 {
384 	struct gpt_hdr *buf, *hdr;
385 	struct g_provider *pp;
386 	quad_t lba, last;
387 	int error;
388 	uint32_t crc, sz;
389 
390 	pp = cp->provider;
391 	last = (pp->mediasize / pp->sectorsize) - 1;
392 	table->state[elt] = GPT_STATE_MISSING;
393 	/*
394 	 * If the primary header is valid look for secondary
395 	 * header in AlternateLBA, otherwise in the last medium's LBA.
396 	 */
397 	if (elt == GPT_ELT_SECHDR) {
398 		if (table->state[GPT_ELT_PRIHDR] != GPT_STATE_OK)
399 			table->lba[elt] = last;
400 	} else
401 		table->lba[elt] = 1;
402 	buf = g_read_data(cp, table->lba[elt] * pp->sectorsize, pp->sectorsize,
403 	    &error);
404 	if (buf == NULL)
405 		return (NULL);
406 	hdr = NULL;
407 	if (memcmp(buf->hdr_sig, GPT_HDR_SIG, sizeof(buf->hdr_sig)) != 0)
408 		goto fail;
409 
410 	table->state[elt] = GPT_STATE_CORRUPT;
411 	sz = le32toh(buf->hdr_size);
412 	if (sz < 92 || sz > pp->sectorsize)
413 		goto fail;
414 
415 	hdr = g_malloc(sz, M_WAITOK | M_ZERO);
416 	bcopy(buf, hdr, sz);
417 	hdr->hdr_size = sz;
418 
419 	crc = le32toh(buf->hdr_crc_self);
420 	buf->hdr_crc_self = 0;
421 	if (crc32(buf, sz) != crc)
422 		goto fail;
423 	hdr->hdr_crc_self = crc;
424 
425 	table->state[elt] = GPT_STATE_INVALID;
426 	hdr->hdr_revision = le32toh(buf->hdr_revision);
427 	if (hdr->hdr_revision < GPT_HDR_REVISION)
428 		goto fail;
429 	hdr->hdr_lba_self = le64toh(buf->hdr_lba_self);
430 	if (hdr->hdr_lba_self != table->lba[elt])
431 		goto fail;
432 	hdr->hdr_lba_alt = le64toh(buf->hdr_lba_alt);
433 	if (hdr->hdr_lba_alt == hdr->hdr_lba_self ||
434 	    hdr->hdr_lba_alt > last)
435 		goto fail;
436 
437 	/* Check the managed area. */
438 	hdr->hdr_lba_start = le64toh(buf->hdr_lba_start);
439 	if (hdr->hdr_lba_start < 2 || hdr->hdr_lba_start >= last)
440 		goto fail;
441 	hdr->hdr_lba_end = le64toh(buf->hdr_lba_end);
442 	if (hdr->hdr_lba_end < hdr->hdr_lba_start || hdr->hdr_lba_end >= last)
443 		goto fail;
444 
445 	/* Check the table location and size of the table. */
446 	hdr->hdr_entries = le32toh(buf->hdr_entries);
447 	hdr->hdr_entsz = le32toh(buf->hdr_entsz);
448 	if (hdr->hdr_entries == 0 || hdr->hdr_entsz < 128 ||
449 	    (hdr->hdr_entsz & 7) != 0)
450 		goto fail;
451 	hdr->hdr_lba_table = le64toh(buf->hdr_lba_table);
452 	if (hdr->hdr_lba_table < 2 || hdr->hdr_lba_table >= last)
453 		goto fail;
454 	if (hdr->hdr_lba_table >= hdr->hdr_lba_start &&
455 	    hdr->hdr_lba_table <= hdr->hdr_lba_end)
456 		goto fail;
457 	lba = hdr->hdr_lba_table +
458 	    (hdr->hdr_entries * hdr->hdr_entsz + pp->sectorsize - 1) /
459 	    pp->sectorsize - 1;
460 	if (lba >= last)
461 		goto fail;
462 	if (lba >= hdr->hdr_lba_start && lba <= hdr->hdr_lba_end)
463 		goto fail;
464 
465 	table->state[elt] = GPT_STATE_OK;
466 	le_uuid_dec(&buf->hdr_uuid, &hdr->hdr_uuid);
467 	hdr->hdr_crc_table = le32toh(buf->hdr_crc_table);
468 
469 	/* save LBA for secondary header */
470 	if (elt == GPT_ELT_PRIHDR)
471 		table->lba[GPT_ELT_SECHDR] = hdr->hdr_lba_alt;
472 
473 	g_free(buf);
474 	return (hdr);
475 
476  fail:
477 	if (hdr != NULL)
478 		g_free(hdr);
479 	g_free(buf);
480 	return (NULL);
481 }
482 
483 static struct gpt_ent *
484 gpt_read_tbl(struct g_part_gpt_table *table, struct g_consumer *cp,
485     enum gpt_elt elt, struct gpt_hdr *hdr)
486 {
487 	struct g_provider *pp;
488 	struct gpt_ent *ent, *tbl;
489 	char *buf, *p;
490 	unsigned int idx, sectors, tblsz, size;
491 	int error;
492 
493 	if (hdr == NULL)
494 		return (NULL);
495 
496 	pp = cp->provider;
497 	table->lba[elt] = hdr->hdr_lba_table;
498 
499 	table->state[elt] = GPT_STATE_MISSING;
500 	tblsz = hdr->hdr_entries * hdr->hdr_entsz;
501 	sectors = (tblsz + pp->sectorsize - 1) / pp->sectorsize;
502 	buf = g_malloc(sectors * pp->sectorsize, M_WAITOK | M_ZERO);
503 	for (idx = 0; idx < sectors; idx += MAXPHYS / pp->sectorsize) {
504 		size = (sectors - idx > MAXPHYS / pp->sectorsize) ?  MAXPHYS:
505 		    (sectors - idx) * pp->sectorsize;
506 		p = g_read_data(cp, (table->lba[elt] + idx) * pp->sectorsize,
507 		    size, &error);
508 		if (p == NULL) {
509 			g_free(buf);
510 			return (NULL);
511 		}
512 		bcopy(p, buf + idx * pp->sectorsize, size);
513 		g_free(p);
514 	}
515 	table->state[elt] = GPT_STATE_CORRUPT;
516 	if (crc32(buf, tblsz) != hdr->hdr_crc_table) {
517 		g_free(buf);
518 		return (NULL);
519 	}
520 
521 	table->state[elt] = GPT_STATE_OK;
522 	tbl = g_malloc(hdr->hdr_entries * sizeof(struct gpt_ent),
523 	    M_WAITOK | M_ZERO);
524 
525 	for (idx = 0, ent = tbl, p = buf;
526 	     idx < hdr->hdr_entries;
527 	     idx++, ent++, p += hdr->hdr_entsz) {
528 		le_uuid_dec(p, &ent->ent_type);
529 		le_uuid_dec(p + 16, &ent->ent_uuid);
530 		ent->ent_lba_start = le64dec(p + 32);
531 		ent->ent_lba_end = le64dec(p + 40);
532 		ent->ent_attr = le64dec(p + 48);
533 		/* Keep UTF-16 in little-endian. */
534 		bcopy(p + 56, ent->ent_name, sizeof(ent->ent_name));
535 	}
536 
537 	g_free(buf);
538 	return (tbl);
539 }
540 
541 static int
542 gpt_matched_hdrs(struct gpt_hdr *pri, struct gpt_hdr *sec)
543 {
544 
545 	if (pri == NULL || sec == NULL)
546 		return (0);
547 
548 	if (!EQUUID(&pri->hdr_uuid, &sec->hdr_uuid))
549 		return (0);
550 	return ((pri->hdr_revision == sec->hdr_revision &&
551 	    pri->hdr_size == sec->hdr_size &&
552 	    pri->hdr_lba_start == sec->hdr_lba_start &&
553 	    pri->hdr_lba_end == sec->hdr_lba_end &&
554 	    pri->hdr_entries == sec->hdr_entries &&
555 	    pri->hdr_entsz == sec->hdr_entsz &&
556 	    pri->hdr_crc_table == sec->hdr_crc_table) ? 1 : 0);
557 }
558 
559 static int
560 gpt_parse_type(const char *type, struct uuid *uuid)
561 {
562 	struct uuid tmp;
563 	const char *alias;
564 	int error;
565 	struct g_part_uuid_alias *uap;
566 
567 	if (type[0] == '!') {
568 		error = parse_uuid(type + 1, &tmp);
569 		if (error)
570 			return (error);
571 		if (EQUUID(&tmp, &gpt_uuid_unused))
572 			return (EINVAL);
573 		*uuid = tmp;
574 		return (0);
575 	}
576 	for (uap = &gpt_uuid_alias_match[0]; uap->uuid; uap++) {
577 		alias = g_part_alias_name(uap->alias);
578 		if (!strcasecmp(type, alias)) {
579 			*uuid = *uap->uuid;
580 			return (0);
581 		}
582 	}
583 	return (EINVAL);
584 }
585 
586 static int
587 g_part_gpt_add(struct g_part_table *basetable, struct g_part_entry *baseentry,
588     struct g_part_parms *gpp)
589 {
590 	struct g_part_gpt_entry *entry;
591 	int error;
592 
593 	entry = (struct g_part_gpt_entry *)baseentry;
594 	error = gpt_parse_type(gpp->gpp_type, &entry->ent.ent_type);
595 	if (error)
596 		return (error);
597 	kern_uuidgen(&entry->ent.ent_uuid, 1);
598 	entry->ent.ent_lba_start = baseentry->gpe_start;
599 	entry->ent.ent_lba_end = baseentry->gpe_end;
600 	if (baseentry->gpe_deleted) {
601 		entry->ent.ent_attr = 0;
602 		bzero(entry->ent.ent_name, sizeof(entry->ent.ent_name));
603 	}
604 	if (gpp->gpp_parms & G_PART_PARM_LABEL)
605 		g_gpt_utf8_to_utf16(gpp->gpp_label, entry->ent.ent_name,
606 		    sizeof(entry->ent.ent_name) /
607 		    sizeof(entry->ent.ent_name[0]));
608 	return (0);
609 }
610 
611 static int
612 g_part_gpt_bootcode(struct g_part_table *basetable, struct g_part_parms *gpp)
613 {
614 	struct g_part_gpt_table *table;
615 	size_t codesz;
616 
617 	codesz = DOSPARTOFF;
618 	table = (struct g_part_gpt_table *)basetable;
619 	bzero(table->mbr, codesz);
620 	codesz = MIN(codesz, gpp->gpp_codesize);
621 	if (codesz > 0)
622 		bcopy(gpp->gpp_codeptr, table->mbr, codesz);
623 	return (0);
624 }
625 
626 static int
627 g_part_gpt_create(struct g_part_table *basetable, struct g_part_parms *gpp)
628 {
629 	struct g_provider *pp;
630 	struct g_part_gpt_table *table;
631 	size_t tblsz;
632 
633 	/* We don't nest, which means that our depth should be 0. */
634 	if (basetable->gpt_depth != 0)
635 		return (ENXIO);
636 
637 	table = (struct g_part_gpt_table *)basetable;
638 	pp = gpp->gpp_provider;
639 	tblsz = (basetable->gpt_entries * sizeof(struct gpt_ent) +
640 	    pp->sectorsize - 1) / pp->sectorsize;
641 	if (pp->sectorsize < MBRSIZE ||
642 	    pp->mediasize < (3 + 2 * tblsz + basetable->gpt_entries) *
643 	    pp->sectorsize)
644 		return (ENOSPC);
645 
646 	gpt_create_pmbr(table, pp);
647 
648 	/* Allocate space for the header */
649 	table->hdr = g_malloc(sizeof(struct gpt_hdr), M_WAITOK | M_ZERO);
650 
651 	bcopy(GPT_HDR_SIG, table->hdr->hdr_sig, sizeof(table->hdr->hdr_sig));
652 	table->hdr->hdr_revision = GPT_HDR_REVISION;
653 	table->hdr->hdr_size = offsetof(struct gpt_hdr, padding);
654 	kern_uuidgen(&table->hdr->hdr_uuid, 1);
655 	table->hdr->hdr_entries = basetable->gpt_entries;
656 	table->hdr->hdr_entsz = sizeof(struct gpt_ent);
657 
658 	g_gpt_set_defaults(basetable, pp);
659 	return (0);
660 }
661 
662 static int
663 g_part_gpt_destroy(struct g_part_table *basetable, struct g_part_parms *gpp)
664 {
665 	struct g_part_gpt_table *table;
666 	struct g_provider *pp;
667 
668 	table = (struct g_part_gpt_table *)basetable;
669 	pp = LIST_FIRST(&basetable->gpt_gp->consumer)->provider;
670 	g_free(table->hdr);
671 	table->hdr = NULL;
672 
673 	/*
674 	 * Wipe the first 2 sectors to clear the partitioning. Wipe the last
675 	 * sector only if it has valid secondary header.
676 	 */
677 	basetable->gpt_smhead |= 3;
678 	if (table->state[GPT_ELT_SECHDR] == GPT_STATE_OK &&
679 	    table->lba[GPT_ELT_SECHDR] == pp->mediasize / pp->sectorsize - 1)
680 		basetable->gpt_smtail |= 1;
681 	return (0);
682 }
683 
684 static void
685 g_part_gpt_dumpconf(struct g_part_table *table, struct g_part_entry *baseentry,
686     struct sbuf *sb, const char *indent)
687 {
688 	struct g_part_gpt_entry *entry;
689 
690 	entry = (struct g_part_gpt_entry *)baseentry;
691 	if (indent == NULL) {
692 		/* conftxt: libdisk compatibility */
693 		sbuf_printf(sb, " xs GPT xt ");
694 		sbuf_printf_uuid(sb, &entry->ent.ent_type);
695 	} else if (entry != NULL) {
696 		/* confxml: partition entry information */
697 		sbuf_printf(sb, "%s<label>", indent);
698 		g_gpt_printf_utf16(sb, entry->ent.ent_name,
699 		    sizeof(entry->ent.ent_name) >> 1);
700 		sbuf_printf(sb, "</label>\n");
701 		if (entry->ent.ent_attr & GPT_ENT_ATTR_BOOTME)
702 			sbuf_printf(sb, "%s<attrib>bootme</attrib>\n", indent);
703 		if (entry->ent.ent_attr & GPT_ENT_ATTR_BOOTONCE) {
704 			sbuf_printf(sb, "%s<attrib>bootonce</attrib>\n",
705 			    indent);
706 		}
707 		if (entry->ent.ent_attr & GPT_ENT_ATTR_BOOTFAILED) {
708 			sbuf_printf(sb, "%s<attrib>bootfailed</attrib>\n",
709 			    indent);
710 		}
711 		sbuf_printf(sb, "%s<rawtype>", indent);
712 		sbuf_printf_uuid(sb, &entry->ent.ent_type);
713 		sbuf_printf(sb, "</rawtype>\n");
714 		sbuf_printf(sb, "%s<rawuuid>", indent);
715 		sbuf_printf_uuid(sb, &entry->ent.ent_uuid);
716 		sbuf_printf(sb, "</rawuuid>\n");
717 	} else {
718 		/* confxml: scheme information */
719 	}
720 }
721 
722 static int
723 g_part_gpt_dumpto(struct g_part_table *table, struct g_part_entry *baseentry)
724 {
725 	struct g_part_gpt_entry *entry;
726 
727 	entry = (struct g_part_gpt_entry *)baseentry;
728 	return ((EQUUID(&entry->ent.ent_type, &gpt_uuid_freebsd_swap) ||
729 	    EQUUID(&entry->ent.ent_type, &gpt_uuid_linux_swap) ||
730 	    EQUUID(&entry->ent.ent_type, &gpt_uuid_dfbsd_swap)) ? 1 : 0);
731 }
732 
733 static int
734 g_part_gpt_modify(struct g_part_table *basetable,
735     struct g_part_entry *baseentry, struct g_part_parms *gpp)
736 {
737 	struct g_part_gpt_entry *entry;
738 	int error;
739 
740 	entry = (struct g_part_gpt_entry *)baseentry;
741 	if (gpp->gpp_parms & G_PART_PARM_TYPE) {
742 		error = gpt_parse_type(gpp->gpp_type, &entry->ent.ent_type);
743 		if (error)
744 			return (error);
745 	}
746 	if (gpp->gpp_parms & G_PART_PARM_LABEL)
747 		g_gpt_utf8_to_utf16(gpp->gpp_label, entry->ent.ent_name,
748 		    sizeof(entry->ent.ent_name) /
749 		    sizeof(entry->ent.ent_name[0]));
750 	return (0);
751 }
752 
753 static int
754 g_part_gpt_resize(struct g_part_table *basetable,
755     struct g_part_entry *baseentry, struct g_part_parms *gpp)
756 {
757 	struct g_part_gpt_entry *entry;
758 
759 	if (baseentry == NULL)
760 		return (EOPNOTSUPP);
761 
762 	entry = (struct g_part_gpt_entry *)baseentry;
763 	baseentry->gpe_end = baseentry->gpe_start + gpp->gpp_size - 1;
764 	entry->ent.ent_lba_end = baseentry->gpe_end;
765 
766 	return (0);
767 }
768 
769 static const char *
770 g_part_gpt_name(struct g_part_table *table, struct g_part_entry *baseentry,
771     char *buf, size_t bufsz)
772 {
773 	struct g_part_gpt_entry *entry;
774 	char c;
775 
776 	entry = (struct g_part_gpt_entry *)baseentry;
777 	c = (EQUUID(&entry->ent.ent_type, &gpt_uuid_freebsd)) ? 's' : 'p';
778 	snprintf(buf, bufsz, "%c%d", c, baseentry->gpe_index);
779 	return (buf);
780 }
781 
782 static int
783 g_part_gpt_probe(struct g_part_table *table, struct g_consumer *cp)
784 {
785 	struct g_provider *pp;
786 	u_char *buf;
787 	int error, index, pri, res;
788 
789 	/* We don't nest, which means that our depth should be 0. */
790 	if (table->gpt_depth != 0)
791 		return (ENXIO);
792 
793 	pp = cp->provider;
794 
795 	/*
796 	 * Sanity-check the provider. Since the first sector on the provider
797 	 * must be a PMBR and a PMBR is 512 bytes large, the sector size
798 	 * must be at least 512 bytes.  Also, since the theoretical minimum
799 	 * number of sectors needed by GPT is 6, any medium that has less
800 	 * than 6 sectors is never going to be able to hold a GPT. The
801 	 * number 6 comes from:
802 	 *	1 sector for the PMBR
803 	 *	2 sectors for the GPT headers (each 1 sector)
804 	 *	2 sectors for the GPT tables (each 1 sector)
805 	 *	1 sector for an actual partition
806 	 * It's better to catch this pathological case early than behaving
807 	 * pathologically later on...
808 	 */
809 	if (pp->sectorsize < MBRSIZE || pp->mediasize < 6 * pp->sectorsize)
810 		return (ENOSPC);
811 
812 	/*
813 	 * Check that there's a MBR or a PMBR. If it's a PMBR, we return
814 	 * as the highest priority on a match, otherwise we assume some
815 	 * GPT-unaware tool has destroyed the GPT by recreating a MBR and
816 	 * we really want the MBR scheme to take precedence.
817 	 */
818 	buf = g_read_data(cp, 0L, pp->sectorsize, &error);
819 	if (buf == NULL)
820 		return (error);
821 	res = le16dec(buf + DOSMAGICOFFSET);
822 	pri = G_PART_PROBE_PRI_LOW;
823 	for (index = 0; index < NDOSPART; index++) {
824 		if (buf[DOSPARTOFF + DOSPARTSIZE * index + 4] == 0xee)
825 			pri = G_PART_PROBE_PRI_HIGH;
826 	}
827 	g_free(buf);
828 	if (res != DOSMAGIC)
829 		return (ENXIO);
830 
831 	/* Check that there's a primary header. */
832 	buf = g_read_data(cp, pp->sectorsize, pp->sectorsize, &error);
833 	if (buf == NULL)
834 		return (error);
835 	res = memcmp(buf, GPT_HDR_SIG, 8);
836 	g_free(buf);
837 	if (res == 0)
838 		return (pri);
839 
840 	/* No primary? Check that there's a secondary. */
841 	buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize,
842 	    &error);
843 	if (buf == NULL)
844 		return (error);
845 	res = memcmp(buf, GPT_HDR_SIG, 8);
846 	g_free(buf);
847 	return ((res == 0) ? pri : ENXIO);
848 }
849 
850 static int
851 g_part_gpt_read(struct g_part_table *basetable, struct g_consumer *cp)
852 {
853 	struct gpt_hdr *prihdr, *sechdr;
854 	struct gpt_ent *tbl, *pritbl, *sectbl;
855 	struct g_provider *pp;
856 	struct g_part_gpt_table *table;
857 	struct g_part_gpt_entry *entry;
858 	u_char *buf;
859 	uint64_t last;
860 	int error, index;
861 
862 	table = (struct g_part_gpt_table *)basetable;
863 	pp = cp->provider;
864 	last = (pp->mediasize / pp->sectorsize) - 1;
865 
866 	/* Read the PMBR */
867 	buf = g_read_data(cp, 0, pp->sectorsize, &error);
868 	if (buf == NULL)
869 		return (error);
870 	bcopy(buf, table->mbr, MBRSIZE);
871 	g_free(buf);
872 
873 	/* Read the primary header and table. */
874 	prihdr = gpt_read_hdr(table, cp, GPT_ELT_PRIHDR);
875 	if (table->state[GPT_ELT_PRIHDR] == GPT_STATE_OK) {
876 		pritbl = gpt_read_tbl(table, cp, GPT_ELT_PRITBL, prihdr);
877 	} else {
878 		table->state[GPT_ELT_PRITBL] = GPT_STATE_MISSING;
879 		pritbl = NULL;
880 	}
881 
882 	/* Read the secondary header and table. */
883 	sechdr = gpt_read_hdr(table, cp, GPT_ELT_SECHDR);
884 	if (table->state[GPT_ELT_SECHDR] == GPT_STATE_OK) {
885 		sectbl = gpt_read_tbl(table, cp, GPT_ELT_SECTBL, sechdr);
886 	} else {
887 		table->state[GPT_ELT_SECTBL] = GPT_STATE_MISSING;
888 		sectbl = NULL;
889 	}
890 
891 	/* Fail if we haven't got any good tables at all. */
892 	if (table->state[GPT_ELT_PRITBL] != GPT_STATE_OK &&
893 	    table->state[GPT_ELT_SECTBL] != GPT_STATE_OK) {
894 		printf("GEOM: %s: corrupt or invalid GPT detected.\n",
895 		    pp->name);
896 		printf("GEOM: %s: GPT rejected -- may not be recoverable.\n",
897 		    pp->name);
898 		return (EINVAL);
899 	}
900 
901 	/*
902 	 * If both headers are good but they disagree with each other,
903 	 * then invalidate one. We prefer to keep the primary header,
904 	 * unless the primary table is corrupt.
905 	 */
906 	if (table->state[GPT_ELT_PRIHDR] == GPT_STATE_OK &&
907 	    table->state[GPT_ELT_SECHDR] == GPT_STATE_OK &&
908 	    !gpt_matched_hdrs(prihdr, sechdr)) {
909 		if (table->state[GPT_ELT_PRITBL] == GPT_STATE_OK) {
910 			table->state[GPT_ELT_SECHDR] = GPT_STATE_INVALID;
911 			table->state[GPT_ELT_SECTBL] = GPT_STATE_MISSING;
912 			g_free(sechdr);
913 			sechdr = NULL;
914 		} else {
915 			table->state[GPT_ELT_PRIHDR] = GPT_STATE_INVALID;
916 			table->state[GPT_ELT_PRITBL] = GPT_STATE_MISSING;
917 			g_free(prihdr);
918 			prihdr = NULL;
919 		}
920 	}
921 
922 	if (table->state[GPT_ELT_PRITBL] != GPT_STATE_OK) {
923 		printf("GEOM: %s: the primary GPT table is corrupt or "
924 		    "invalid.\n", pp->name);
925 		printf("GEOM: %s: using the secondary instead -- recovery "
926 		    "strongly advised.\n", pp->name);
927 		table->hdr = sechdr;
928 		basetable->gpt_corrupt = 1;
929 		if (prihdr != NULL)
930 			g_free(prihdr);
931 		tbl = sectbl;
932 		if (pritbl != NULL)
933 			g_free(pritbl);
934 	} else {
935 		if (table->state[GPT_ELT_SECTBL] != GPT_STATE_OK) {
936 			printf("GEOM: %s: the secondary GPT table is corrupt "
937 			    "or invalid.\n", pp->name);
938 			printf("GEOM: %s: using the primary only -- recovery "
939 			    "suggested.\n", pp->name);
940 			basetable->gpt_corrupt = 1;
941 		} else if (table->lba[GPT_ELT_SECHDR] != last) {
942 			printf( "GEOM: %s: the secondary GPT header is not in "
943 			    "the last LBA.\n", pp->name);
944 			basetable->gpt_corrupt = 1;
945 		}
946 		table->hdr = prihdr;
947 		if (sechdr != NULL)
948 			g_free(sechdr);
949 		tbl = pritbl;
950 		if (sectbl != NULL)
951 			g_free(sectbl);
952 	}
953 
954 	basetable->gpt_first = table->hdr->hdr_lba_start;
955 	basetable->gpt_last = table->hdr->hdr_lba_end;
956 	basetable->gpt_entries = (table->hdr->hdr_lba_start - 2) *
957 	    pp->sectorsize / table->hdr->hdr_entsz;
958 
959 	for (index = table->hdr->hdr_entries - 1; index >= 0; index--) {
960 		if (EQUUID(&tbl[index].ent_type, &gpt_uuid_unused))
961 			continue;
962 		entry = (struct g_part_gpt_entry *)g_part_new_entry(
963 		    basetable, index + 1, tbl[index].ent_lba_start,
964 		    tbl[index].ent_lba_end);
965 		entry->ent = tbl[index];
966 	}
967 
968 	g_free(tbl);
969 
970 	/*
971 	 * Under Mac OS X, the MBR mirrors the first 4 GPT partitions
972 	 * if (and only if) any FAT32 or FAT16 partitions have been
973 	 * created. This happens irrespective of whether Boot Camp is
974 	 * used/enabled, though it's generally understood to be done
975 	 * to support legacy Windows under Boot Camp. We refer to this
976 	 * mirroring simply as Boot Camp. We try to detect Boot Camp
977 	 * so that we can update the MBR if and when GPT changes have
978 	 * been made. Note that we do not enable Boot Camp if not
979 	 * previously enabled because we can't assume that we're on a
980 	 * Mac alongside Mac OS X.
981 	 */
982 	table->bootcamp = gpt_is_bootcamp(table, pp->name);
983 
984 	return (0);
985 }
986 
987 static int
988 g_part_gpt_recover(struct g_part_table *basetable)
989 {
990 	struct g_part_gpt_table *table;
991 	struct g_provider *pp;
992 
993 	table = (struct g_part_gpt_table *)basetable;
994 	pp = LIST_FIRST(&basetable->gpt_gp->consumer)->provider;
995 	gpt_create_pmbr(table, pp);
996 	g_gpt_set_defaults(basetable, pp);
997 	basetable->gpt_corrupt = 0;
998 	return (0);
999 }
1000 
1001 static int
1002 g_part_gpt_setunset(struct g_part_table *basetable,
1003     struct g_part_entry *baseentry, const char *attrib, unsigned int set)
1004 {
1005 	struct g_part_gpt_entry *entry;
1006 	struct g_part_gpt_table *table;
1007 	uint8_t *p;
1008 	uint64_t attr;
1009 	int i;
1010 
1011 	table = (struct g_part_gpt_table *)basetable;
1012 	entry = (struct g_part_gpt_entry *)baseentry;
1013 
1014 	if (strcasecmp(attrib, "active") == 0) {
1015 		if (table->bootcamp) {
1016 			/* The active flag must be set on a valid entry. */
1017 			if (entry == NULL)
1018 				return (ENXIO);
1019 			if (baseentry->gpe_index > NDOSPART)
1020 				return (EINVAL);
1021 			for (i = 0; i < NDOSPART; i++) {
1022 				p = &table->mbr[DOSPARTOFF + i * DOSPARTSIZE];
1023 				p[0] = (i == baseentry->gpe_index - 1)
1024 				    ? ((set) ? 0x80 : 0) : 0;
1025 			}
1026 		} else {
1027 			/* The PMBR is marked as active without an entry. */
1028 			if (entry != NULL)
1029 				return (ENXIO);
1030 			for (i = 0; i < NDOSPART; i++) {
1031 				p = &table->mbr[DOSPARTOFF + i * DOSPARTSIZE];
1032 				p[0] = (p[4] == 0xee) ? ((set) ? 0x80 : 0) : 0;
1033 			}
1034 		}
1035 		return (0);
1036 	}
1037 
1038 	if (entry == NULL)
1039 		return (ENODEV);
1040 
1041 	attr = 0;
1042 	if (strcasecmp(attrib, "bootme") == 0) {
1043 		attr |= GPT_ENT_ATTR_BOOTME;
1044 	} else if (strcasecmp(attrib, "bootonce") == 0) {
1045 		attr |= GPT_ENT_ATTR_BOOTONCE;
1046 		if (set)
1047 			attr |= GPT_ENT_ATTR_BOOTME;
1048 	} else if (strcasecmp(attrib, "bootfailed") == 0) {
1049 		/*
1050 		 * It should only be possible to unset BOOTFAILED, but it might
1051 		 * be useful for test purposes to also be able to set it.
1052 		 */
1053 		attr |= GPT_ENT_ATTR_BOOTFAILED;
1054 	}
1055 	if (attr == 0)
1056 		return (EINVAL);
1057 
1058 	if (set)
1059 		attr = entry->ent.ent_attr | attr;
1060 	else
1061 		attr = entry->ent.ent_attr & ~attr;
1062 	if (attr != entry->ent.ent_attr) {
1063 		entry->ent.ent_attr = attr;
1064 		if (!baseentry->gpe_created)
1065 			baseentry->gpe_modified = 1;
1066 	}
1067 	return (0);
1068 }
1069 
1070 static const char *
1071 g_part_gpt_type(struct g_part_table *basetable, struct g_part_entry *baseentry,
1072     char *buf, size_t bufsz)
1073 {
1074 	struct g_part_gpt_entry *entry;
1075 	struct uuid *type;
1076 	struct g_part_uuid_alias *uap;
1077 
1078 	entry = (struct g_part_gpt_entry *)baseentry;
1079 	type = &entry->ent.ent_type;
1080 	for (uap = &gpt_uuid_alias_match[0]; uap->uuid; uap++)
1081 		if (EQUUID(type, uap->uuid))
1082 			return (g_part_alias_name(uap->alias));
1083 	buf[0] = '!';
1084 	snprintf_uuid(buf + 1, bufsz - 1, type);
1085 
1086 	return (buf);
1087 }
1088 
1089 static int
1090 g_part_gpt_write(struct g_part_table *basetable, struct g_consumer *cp)
1091 {
1092 	unsigned char *buf, *bp;
1093 	struct g_provider *pp;
1094 	struct g_part_entry *baseentry;
1095 	struct g_part_gpt_entry *entry;
1096 	struct g_part_gpt_table *table;
1097 	size_t tblsz;
1098 	uint32_t crc;
1099 	int error, index;
1100 
1101 	pp = cp->provider;
1102 	table = (struct g_part_gpt_table *)basetable;
1103 	tblsz = (table->hdr->hdr_entries * table->hdr->hdr_entsz +
1104 	    pp->sectorsize - 1) / pp->sectorsize;
1105 
1106 	/* Reconstruct the MBR from the GPT if under Boot Camp. */
1107 	if (table->bootcamp)
1108 		gpt_update_bootcamp(basetable, pp);
1109 
1110 	/* Write the PMBR */
1111 	buf = g_malloc(pp->sectorsize, M_WAITOK | M_ZERO);
1112 	bcopy(table->mbr, buf, MBRSIZE);
1113 	error = g_write_data(cp, 0, buf, pp->sectorsize);
1114 	g_free(buf);
1115 	if (error)
1116 		return (error);
1117 
1118 	/* Allocate space for the header and entries. */
1119 	buf = g_malloc((tblsz + 1) * pp->sectorsize, M_WAITOK | M_ZERO);
1120 
1121 	memcpy(buf, table->hdr->hdr_sig, sizeof(table->hdr->hdr_sig));
1122 	le32enc(buf + 8, table->hdr->hdr_revision);
1123 	le32enc(buf + 12, table->hdr->hdr_size);
1124 	le64enc(buf + 40, table->hdr->hdr_lba_start);
1125 	le64enc(buf + 48, table->hdr->hdr_lba_end);
1126 	le_uuid_enc(buf + 56, &table->hdr->hdr_uuid);
1127 	le32enc(buf + 80, table->hdr->hdr_entries);
1128 	le32enc(buf + 84, table->hdr->hdr_entsz);
1129 
1130 	LIST_FOREACH(baseentry, &basetable->gpt_entry, gpe_entry) {
1131 		if (baseentry->gpe_deleted)
1132 			continue;
1133 		entry = (struct g_part_gpt_entry *)baseentry;
1134 		index = baseentry->gpe_index - 1;
1135 		bp = buf + pp->sectorsize + table->hdr->hdr_entsz * index;
1136 		le_uuid_enc(bp, &entry->ent.ent_type);
1137 		le_uuid_enc(bp + 16, &entry->ent.ent_uuid);
1138 		le64enc(bp + 32, entry->ent.ent_lba_start);
1139 		le64enc(bp + 40, entry->ent.ent_lba_end);
1140 		le64enc(bp + 48, entry->ent.ent_attr);
1141 		memcpy(bp + 56, entry->ent.ent_name,
1142 		    sizeof(entry->ent.ent_name));
1143 	}
1144 
1145 	crc = crc32(buf + pp->sectorsize,
1146 	    table->hdr->hdr_entries * table->hdr->hdr_entsz);
1147 	le32enc(buf + 88, crc);
1148 
1149 	/* Write primary meta-data. */
1150 	le32enc(buf + 16, 0);	/* hdr_crc_self. */
1151 	le64enc(buf + 24, table->lba[GPT_ELT_PRIHDR]);	/* hdr_lba_self. */
1152 	le64enc(buf + 32, table->lba[GPT_ELT_SECHDR]);	/* hdr_lba_alt. */
1153 	le64enc(buf + 72, table->lba[GPT_ELT_PRITBL]);	/* hdr_lba_table. */
1154 	crc = crc32(buf, table->hdr->hdr_size);
1155 	le32enc(buf + 16, crc);
1156 
1157 	for (index = 0; index < tblsz; index += MAXPHYS / pp->sectorsize) {
1158 		error = g_write_data(cp,
1159 		    (table->lba[GPT_ELT_PRITBL] + index) * pp->sectorsize,
1160 		    buf + (index + 1) * pp->sectorsize,
1161 		    (tblsz - index > MAXPHYS / pp->sectorsize) ? MAXPHYS:
1162 		    (tblsz - index) * pp->sectorsize);
1163 		if (error)
1164 			goto out;
1165 	}
1166 	error = g_write_data(cp, table->lba[GPT_ELT_PRIHDR] * pp->sectorsize,
1167 	    buf, pp->sectorsize);
1168 	if (error)
1169 		goto out;
1170 
1171 	/* Write secondary meta-data. */
1172 	le32enc(buf + 16, 0);	/* hdr_crc_self. */
1173 	le64enc(buf + 24, table->lba[GPT_ELT_SECHDR]);	/* hdr_lba_self. */
1174 	le64enc(buf + 32, table->lba[GPT_ELT_PRIHDR]);	/* hdr_lba_alt. */
1175 	le64enc(buf + 72, table->lba[GPT_ELT_SECTBL]);	/* hdr_lba_table. */
1176 	crc = crc32(buf, table->hdr->hdr_size);
1177 	le32enc(buf + 16, crc);
1178 
1179 	for (index = 0; index < tblsz; index += MAXPHYS / pp->sectorsize) {
1180 		error = g_write_data(cp,
1181 		    (table->lba[GPT_ELT_SECTBL] + index) * pp->sectorsize,
1182 		    buf + (index + 1) * pp->sectorsize,
1183 		    (tblsz - index > MAXPHYS / pp->sectorsize) ? MAXPHYS:
1184 		    (tblsz - index) * pp->sectorsize);
1185 		if (error)
1186 			goto out;
1187 	}
1188 	error = g_write_data(cp, table->lba[GPT_ELT_SECHDR] * pp->sectorsize,
1189 	    buf, pp->sectorsize);
1190 
1191  out:
1192 	g_free(buf);
1193 	return (error);
1194 }
1195 
1196 static void
1197 g_gpt_set_defaults(struct g_part_table *basetable, struct g_provider *pp)
1198 {
1199 	struct g_part_entry *baseentry;
1200 	struct g_part_gpt_entry *entry;
1201 	struct g_part_gpt_table *table;
1202 	quad_t start, end, min, max;
1203 	quad_t lba, last;
1204 	size_t spb, tblsz;
1205 
1206 	table = (struct g_part_gpt_table *)basetable;
1207 	last = pp->mediasize / pp->sectorsize - 1;
1208 	tblsz = (basetable->gpt_entries * sizeof(struct gpt_ent) +
1209 	    pp->sectorsize - 1) / pp->sectorsize;
1210 
1211 	table->lba[GPT_ELT_PRIHDR] = 1;
1212 	table->lba[GPT_ELT_PRITBL] = 2;
1213 	table->lba[GPT_ELT_SECHDR] = last;
1214 	table->lba[GPT_ELT_SECTBL] = last - tblsz;
1215 	table->state[GPT_ELT_PRIHDR] = GPT_STATE_OK;
1216 	table->state[GPT_ELT_PRITBL] = GPT_STATE_OK;
1217 	table->state[GPT_ELT_SECHDR] = GPT_STATE_OK;
1218 	table->state[GPT_ELT_SECTBL] = GPT_STATE_OK;
1219 
1220 	max = start = 2 + tblsz;
1221 	min = end = last - tblsz - 1;
1222 	LIST_FOREACH(baseentry, &basetable->gpt_entry, gpe_entry) {
1223 		if (baseentry->gpe_deleted)
1224 			continue;
1225 		entry = (struct g_part_gpt_entry *)baseentry;
1226 		if (entry->ent.ent_lba_start < min)
1227 			min = entry->ent.ent_lba_start;
1228 		if (entry->ent.ent_lba_end > max)
1229 			max = entry->ent.ent_lba_end;
1230 	}
1231 	spb = 4096 / pp->sectorsize;
1232 	if (spb > 1) {
1233 		lba = start + ((start % spb) ? spb - start % spb : 0);
1234 		if (lba <= min)
1235 			start = lba;
1236 		lba = end - (end + 1) % spb;
1237 		if (max <= lba)
1238 			end = lba;
1239 	}
1240 	table->hdr->hdr_lba_start = start;
1241 	table->hdr->hdr_lba_end = end;
1242 
1243 	basetable->gpt_first = start;
1244 	basetable->gpt_last = end;
1245 }
1246 
1247 static void
1248 g_gpt_printf_utf16(struct sbuf *sb, uint16_t *str, size_t len)
1249 {
1250 	u_int bo;
1251 	uint32_t ch;
1252 	uint16_t c;
1253 
1254 	bo = LITTLE_ENDIAN;	/* GPT is little-endian */
1255 	while (len > 0 && *str != 0) {
1256 		ch = (bo == BIG_ENDIAN) ? be16toh(*str) : le16toh(*str);
1257 		str++, len--;
1258 		if ((ch & 0xf800) == 0xd800) {
1259 			if (len > 0) {
1260 				c = (bo == BIG_ENDIAN) ? be16toh(*str)
1261 				    : le16toh(*str);
1262 				str++, len--;
1263 			} else
1264 				c = 0xfffd;
1265 			if ((ch & 0x400) == 0 && (c & 0xfc00) == 0xdc00) {
1266 				ch = ((ch & 0x3ff) << 10) + (c & 0x3ff);
1267 				ch += 0x10000;
1268 			} else
1269 				ch = 0xfffd;
1270 		} else if (ch == 0xfffe) { /* BOM (U+FEFF) swapped. */
1271 			bo = (bo == BIG_ENDIAN) ? LITTLE_ENDIAN : BIG_ENDIAN;
1272 			continue;
1273 		} else if (ch == 0xfeff) /* BOM (U+FEFF) unswapped. */
1274 			continue;
1275 
1276 		/* Write the Unicode character in UTF-8 */
1277 		if (ch < 0x80)
1278 			g_conf_printf_escaped(sb, "%c", ch);
1279 		else if (ch < 0x800)
1280 			g_conf_printf_escaped(sb, "%c%c", 0xc0 | (ch >> 6),
1281 			    0x80 | (ch & 0x3f));
1282 		else if (ch < 0x10000)
1283 			g_conf_printf_escaped(sb, "%c%c%c", 0xe0 | (ch >> 12),
1284 			    0x80 | ((ch >> 6) & 0x3f), 0x80 | (ch & 0x3f));
1285 		else if (ch < 0x200000)
1286 			g_conf_printf_escaped(sb, "%c%c%c%c", 0xf0 |
1287 			    (ch >> 18), 0x80 | ((ch >> 12) & 0x3f),
1288 			    0x80 | ((ch >> 6) & 0x3f), 0x80 | (ch & 0x3f));
1289 	}
1290 }
1291 
1292 static void
1293 g_gpt_utf8_to_utf16(const uint8_t *s8, uint16_t *s16, size_t s16len)
1294 {
1295 	size_t s16idx, s8idx;
1296 	uint32_t utfchar;
1297 	unsigned int c, utfbytes;
1298 
1299 	s8idx = s16idx = 0;
1300 	utfchar = 0;
1301 	utfbytes = 0;
1302 	bzero(s16, s16len << 1);
1303 	while (s8[s8idx] != 0 && s16idx < s16len) {
1304 		c = s8[s8idx++];
1305 		if ((c & 0xc0) != 0x80) {
1306 			/* Initial characters. */
1307 			if (utfbytes != 0) {
1308 				/* Incomplete encoding of previous char. */
1309 				s16[s16idx++] = htole16(0xfffd);
1310 			}
1311 			if ((c & 0xf8) == 0xf0) {
1312 				utfchar = c & 0x07;
1313 				utfbytes = 3;
1314 			} else if ((c & 0xf0) == 0xe0) {
1315 				utfchar = c & 0x0f;
1316 				utfbytes = 2;
1317 			} else if ((c & 0xe0) == 0xc0) {
1318 				utfchar = c & 0x1f;
1319 				utfbytes = 1;
1320 			} else {
1321 				utfchar = c & 0x7f;
1322 				utfbytes = 0;
1323 			}
1324 		} else {
1325 			/* Followup characters. */
1326 			if (utfbytes > 0) {
1327 				utfchar = (utfchar << 6) + (c & 0x3f);
1328 				utfbytes--;
1329 			} else if (utfbytes == 0)
1330 				utfbytes = ~0;
1331 		}
1332 		/*
1333 		 * Write the complete Unicode character as UTF-16 when we
1334 		 * have all the UTF-8 charactars collected.
1335 		 */
1336 		if (utfbytes == 0) {
1337 			/*
1338 			 * If we need to write 2 UTF-16 characters, but
1339 			 * we only have room for 1, then we truncate the
1340 			 * string by writing a 0 instead.
1341 			 */
1342 			if (utfchar >= 0x10000 && s16idx < s16len - 1) {
1343 				s16[s16idx++] =
1344 				    htole16(0xd800 | ((utfchar >> 10) - 0x40));
1345 				s16[s16idx++] =
1346 				    htole16(0xdc00 | (utfchar & 0x3ff));
1347 			} else
1348 				s16[s16idx++] = (utfchar >= 0x10000) ? 0 :
1349 				    htole16(utfchar);
1350 		}
1351 	}
1352 	/*
1353 	 * If our input string was truncated, append an invalid encoding
1354 	 * character to the output string.
1355 	 */
1356 	if (utfbytes != 0 && s16idx < s16len)
1357 		s16[s16idx++] = htole16(0xfffd);
1358 }
1359