xref: /freebsd/sys/geom/part/g_part_ldm.c (revision 6dcefcac2b043be030851e03a721607b414b666b)
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  *
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/part/g_part.h>
48 
49 #include "g_part_if.h"
50 
51 FEATURE(geom_part_ldm, "GEOM partitioning class for LDM support");
52 
53 SYSCTL_DECL(_kern_geom_part);
54 static SYSCTL_NODE(_kern_geom_part, OID_AUTO, ldm, CTLFLAG_RW, 0,
55     "GEOM_PART_LDM Logical Disk Manager");
56 
57 static u_int ldm_debug = 0;
58 TUNABLE_INT("kern.geom.part.ldm.debug", &ldm_debug);
59 SYSCTL_UINT(_kern_geom_part_ldm, OID_AUTO, debug,
60     CTLFLAG_RW | CTLFLAG_TUN, &ldm_debug, 0, "Debug level");
61 
62 /*
63  * This allows access to mirrored LDM volumes. Since we do not
64  * doing mirroring here, it is not enabled by default.
65  */
66 static u_int show_mirrors = 0;
67 TUNABLE_INT("kern.geom.part.ldm.show_mirrors", &show_mirrors);
68 SYSCTL_UINT(_kern_geom_part_ldm, OID_AUTO, show_mirrors,
69     CTLFLAG_RW | CTLFLAG_TUN, &show_mirrors, 0, "Show mirrored volumes");
70 
71 #define	LDM_DEBUG(lvl, fmt, ...)	do {				\
72 	if (ldm_debug >= (lvl)) {					\
73 		printf("GEOM_PART: " fmt "\n", __VA_ARGS__);		\
74 	}								\
75 } while (0)
76 #define	LDM_DUMP(buf, size)	do {					\
77 	if (ldm_debug > 1) {						\
78 		hexdump(buf, size, NULL, 0);				\
79 	}								\
80 } while (0)
81 
82 /*
83  * There are internal representations of LDM structures.
84  *
85  * We do not keep all fields of on-disk structures, only most useful.
86  * All numbers in an on-disk structures are in big-endian format.
87  */
88 
89 /*
90  * Private header is 512 bytes long. There are three copies on each disk.
91  * Offset and sizes are in sectors. Location of each copy:
92  * - the first offset is relative to the disk start;
93  * - the second and third offset are relative to the LDM database start.
94  *
95  * On a disk partitioned with GPT, the LDM has not first private header.
96  */
97 #define	LDM_PH_MBRINDEX		0
98 #define	LDM_PH_GPTINDEX		2
99 static const uint64_t	ldm_ph_off[] = {6, 1856, 2047};
100 #define	LDM_VERSION_2K		0x2000b
101 #define	LDM_VERSION_VISTA	0x2000c
102 #define	LDM_PH_VERSION_OFF	0x00c
103 #define	LDM_PH_DISKGUID_OFF	0x030
104 #define	LDM_PH_DGGUID_OFF	0x0b0
105 #define	LDM_PH_DGNAME_OFF	0x0f0
106 #define	LDM_PH_START_OFF	0x11b
107 #define	LDM_PH_SIZE_OFF		0x123
108 #define	LDM_PH_DB_OFF		0x12b
109 #define	LDM_PH_DBSIZE_OFF	0x133
110 #define	LDM_PH_TH1_OFF		0x13b
111 #define	LDM_PH_TH2_OFF		0x143
112 #define	LDM_PH_CONFSIZE_OFF	0x153
113 #define	LDM_PH_LOGSIZE_OFF	0x15b
114 #define	LDM_PH_SIGN		"PRIVHEAD"
115 struct ldm_privhdr {
116 	struct uuid	disk_guid;
117 	struct uuid	dg_guid;
118 	u_char		dg_name[32];
119 	uint64_t	start;		/* logical disk start */
120 	uint64_t	size;		/* logical disk size */
121 	uint64_t	db_offset;	/* LDM database start */
122 #define	LDM_DB_SIZE		2048
123 	uint64_t	db_size;	/* LDM database size */
124 #define	LDM_TH_COUNT		2
125 	uint64_t	th_offset[LDM_TH_COUNT]; /* TOC header offsets */
126 	uint64_t	conf_size;	/* configuration size */
127 	uint64_t	log_size;	/* size of log */
128 };
129 
130 /*
131  * Table of contents header is 512 bytes long.
132  * There are two identical copies at offsets from the private header.
133  * Offsets are relative to the LDM database start.
134  */
135 #define	LDM_TH_SIGN		"TOCBLOCK"
136 #define	LDM_TH_NAME1		"config"
137 #define	LDM_TH_NAME2		"log"
138 #define	LDM_TH_NAME1_OFF	0x024
139 #define	LDM_TH_CONF_OFF		0x02e
140 #define	LDM_TH_CONFSIZE_OFF	0x036
141 #define	LDM_TH_NAME2_OFF	0x046
142 #define	LDM_TH_LOG_OFF		0x050
143 #define	LDM_TH_LOGSIZE_OFF	0x058
144 struct ldm_tochdr {
145 	uint64_t	conf_offset;	/* configuration offset */
146 	uint64_t	log_offset;	/* log offset */
147 };
148 
149 /*
150  * LDM database header is 512 bytes long.
151  */
152 #define	LDM_VMDB_SIGN		"VMDB"
153 #define	LDM_DB_LASTSEQ_OFF	0x004
154 #define	LDM_DB_SIZE_OFF		0x008
155 #define	LDM_DB_STATUS_OFF	0x010
156 #define	LDM_DB_VERSION_OFF	0x012
157 #define	LDM_DB_DGNAME_OFF	0x016
158 #define	LDM_DB_DGGUID_OFF	0x035
159 struct ldm_vmdbhdr {
160 	uint32_t	last_seq;	/* sequence number of last VBLK */
161 	uint32_t	size;		/* size of VBLK */
162 };
163 
164 /*
165  * The LDM database configuration section contains VMDB header and
166  * many VBLKs. Each VBLK represents a disk group, disk partition,
167  * component or volume.
168  *
169  * The most interesting for us are volumes, they are represents
170  * partitions in the GEOM_PART meaning. But volume VBLK does not
171  * contain all information needed to create GEOM provider. And we
172  * should get this information from the related VBLK. This is how
173  * VBLK releated:
174  *	Volumes <- Components <- Partitions -> Disks
175  *
176  * One volume can contain several components. In this case LDM
177  * does mirroring of volume data to each component.
178  *
179  * Also each component can contain several partitions (spanned or
180  * striped volumes).
181  */
182 
183 struct ldm_component {
184 	uint64_t	id;		/* object id */
185 	uint64_t	vol_id;		/* parent volume object id */
186 
187 	int		count;
188 	LIST_HEAD(, ldm_partition) partitions;
189 	LIST_ENTRY(ldm_component) entry;
190 };
191 
192 struct ldm_volume {
193 	uint64_t	id;		/* object id */
194 	uint64_t	size;		/* volume size */
195 	uint8_t		number;		/* used for ordering */
196 	uint8_t		part_type;	/* partition type */
197 
198 	int		count;
199 	LIST_HEAD(, ldm_component) components;
200 	LIST_ENTRY(ldm_volume)	entry;
201 };
202 
203 struct ldm_disk {
204 	uint64_t	id;		/* object id */
205 	struct uuid	guid;		/* disk guid */
206 
207 	LIST_ENTRY(ldm_disk) entry;
208 };
209 
210 #if 0
211 struct ldm_disk_group {
212 	uint64_t	id;		/* object id */
213 	struct uuid	guid;		/* disk group guid */
214 	u_char		name[32];	/* disk group name */
215 
216 	LIST_ENTRY(ldm_disk_group) entry;
217 };
218 #endif
219 
220 struct ldm_partition {
221 	uint64_t	id;		/* object id */
222 	uint64_t	disk_id;	/* disk object id */
223 	uint64_t	comp_id;	/* parent component object id */
224 	uint64_t	start;		/* offset relative to disk start */
225 	uint64_t	offset;		/* offset for spanned volumes */
226 	uint64_t	size;		/* partition size */
227 
228 	LIST_ENTRY(ldm_partition) entry;
229 };
230 
231 /*
232  * Each VBLK is 128 bytes long and has standard 16 bytes header.
233  * Some of VBLK's fields are fixed size, but others has variable size.
234  * Fields with variable size are prefixed with one byte length marker.
235  * Some fields are strings and also can have fixed size and variable.
236  * Strings with fixed size are NULL-terminated, others are not.
237  * All VBLKs have same several first fields:
238  *	Offset		Size		Description
239  *	---------------+---------------+--------------------------
240  *	0x00		16		standard VBLK header
241  *	0x10		2		update status
242  *	0x13		1		VBLK type
243  *	0x18		PS		object id
244  *	0x18+		PN		object name
245  *
246  *  o Offset 0x18+ means '0x18 + length of all variable-width fields'
247  *  o 'P' in size column means 'prefixed' (variable-width),
248  *    'S' - string, 'N' - number.
249  */
250 #define	LDM_VBLK_SIGN		"VBLK"
251 #define	LDM_VBLK_SEQ_OFF	0x04
252 #define	LDM_VBLK_GROUP_OFF	0x08
253 #define	LDM_VBLK_INDEX_OFF	0x0c
254 #define	LDM_VBLK_COUNT_OFF	0x0e
255 #define	LDM_VBLK_TYPE_OFF	0x13
256 #define	LDM_VBLK_OID_OFF	0x18
257 struct ldm_vblkhdr {
258 	uint32_t	seq;		/* sequence number */
259 	uint32_t	group;		/* group number */
260 	uint16_t	index;		/* index in the group */
261 	uint16_t	count;		/* number of entries in the group */
262 };
263 
264 #define	LDM_VBLK_T_COMPONENT	0x32
265 #define	LDM_VBLK_T_PARTITION	0x33
266 #define	LDM_VBLK_T_DISK		0x34
267 #define	LDM_VBLK_T_DISKGROUP	0x35
268 #define	LDM_VBLK_T_DISK4	0x44
269 #define	LDM_VBLK_T_DISKGROUP4	0x45
270 #define	LDM_VBLK_T_VOLUME	0x51
271 struct ldm_vblk {
272 	uint8_t		type;		/* VBLK type */
273 	union {
274 		uint64_t		id;
275 		struct ldm_volume	vol;
276 		struct ldm_component	comp;
277 		struct ldm_disk		disk;
278 		struct ldm_partition	part;
279 #if 0
280 		struct ldm_disk_group	disk_group;
281 #endif
282 	} u;
283 	LIST_ENTRY(ldm_vblk) entry;
284 };
285 
286 /*
287  * Some VBLKs contains a bit more data than can fit into 128 bytes. These
288  * VBLKs are called eXtended VBLK. Before parsing, the data from these VBLK
289  * should be placed into continuous memory buffer. We can determine xVBLK
290  * by the count field in the standard VBLK header (count > 1).
291  */
292 struct ldm_xvblk {
293 	uint32_t	group;		/* xVBLK group number */
294 	uint32_t	size;		/* the total size of xVBLK */
295 	uint8_t		map;		/* bitmask of currently saved VBLKs */
296 	u_char		*data;		/* xVBLK data */
297 
298 	LIST_ENTRY(ldm_xvblk)	entry;
299 };
300 
301 /* The internal representation of LDM database. */
302 struct ldm_db {
303 	struct ldm_privhdr		ph;	/* private header */
304 	struct ldm_tochdr		th;	/* TOC header */
305 	struct ldm_vmdbhdr		dh;	/* VMDB header */
306 
307 	LIST_HEAD(, ldm_volume)		volumes;
308 	LIST_HEAD(, ldm_disk)		disks;
309 	LIST_HEAD(, ldm_vblk)		vblks;
310 	LIST_HEAD(, ldm_xvblk)		xvblks;
311 };
312 
313 static struct uuid gpt_uuid_ms_ldm_metadata = GPT_ENT_TYPE_MS_LDM_METADATA;
314 
315 struct g_part_ldm_table {
316 	struct g_part_table	base;
317 	uint64_t		db_offset;
318 	int			is_gpt;
319 };
320 struct g_part_ldm_entry {
321 	struct g_part_entry	base;
322 	uint8_t			type;
323 };
324 
325 static int g_part_ldm_add(struct g_part_table *, struct g_part_entry *,
326     struct g_part_parms *);
327 static int g_part_ldm_bootcode(struct g_part_table *, struct g_part_parms *);
328 static int g_part_ldm_create(struct g_part_table *, struct g_part_parms *);
329 static int g_part_ldm_destroy(struct g_part_table *, struct g_part_parms *);
330 static void g_part_ldm_dumpconf(struct g_part_table *, struct g_part_entry *,
331     struct sbuf *, const char *);
332 static int g_part_ldm_dumpto(struct g_part_table *, struct g_part_entry *);
333 static int g_part_ldm_modify(struct g_part_table *, struct g_part_entry *,
334     struct g_part_parms *);
335 static const char *g_part_ldm_name(struct g_part_table *, struct g_part_entry *,
336     char *, size_t);
337 static int g_part_ldm_probe(struct g_part_table *, struct g_consumer *);
338 static int g_part_ldm_read(struct g_part_table *, struct g_consumer *);
339 static const char *g_part_ldm_type(struct g_part_table *, struct g_part_entry *,
340     char *, size_t);
341 static int g_part_ldm_write(struct g_part_table *, struct g_consumer *);
342 static int g_part_ldm_resize(struct g_part_table *, struct g_part_entry *,
343     struct g_part_parms *);
344 
345 static kobj_method_t g_part_ldm_methods[] = {
346 	KOBJMETHOD(g_part_add,		g_part_ldm_add),
347 	KOBJMETHOD(g_part_bootcode,	g_part_ldm_bootcode),
348 	KOBJMETHOD(g_part_create,	g_part_ldm_create),
349 	KOBJMETHOD(g_part_destroy,	g_part_ldm_destroy),
350 	KOBJMETHOD(g_part_dumpconf,	g_part_ldm_dumpconf),
351 	KOBJMETHOD(g_part_dumpto,	g_part_ldm_dumpto),
352 	KOBJMETHOD(g_part_modify,	g_part_ldm_modify),
353 	KOBJMETHOD(g_part_resize,	g_part_ldm_resize),
354 	KOBJMETHOD(g_part_name,		g_part_ldm_name),
355 	KOBJMETHOD(g_part_probe,	g_part_ldm_probe),
356 	KOBJMETHOD(g_part_read,		g_part_ldm_read),
357 	KOBJMETHOD(g_part_type,		g_part_ldm_type),
358 	KOBJMETHOD(g_part_write,	g_part_ldm_write),
359 	{ 0, 0 }
360 };
361 
362 static struct g_part_scheme g_part_ldm_scheme = {
363 	"LDM",
364 	g_part_ldm_methods,
365 	sizeof(struct g_part_ldm_table),
366 	.gps_entrysz = sizeof(struct g_part_ldm_entry)
367 };
368 G_PART_SCHEME_DECLARE(g_part_ldm);
369 
370 static struct g_part_ldm_alias {
371 	u_char		typ;
372 	int		alias;
373 } ldm_alias_match[] = {
374 	{ DOSPTYP_NTFS,		G_PART_ALIAS_MS_NTFS },
375 	{ DOSPTYP_FAT32,	G_PART_ALIAS_MS_FAT32 },
376 	{ DOSPTYP_386BSD,	G_PART_ALIAS_FREEBSD },
377 	{ DOSPTYP_LDM,		G_PART_ALIAS_MS_LDM_DATA },
378 	{ DOSPTYP_LINSWP,	G_PART_ALIAS_LINUX_SWAP },
379 	{ DOSPTYP_LINUX,	G_PART_ALIAS_LINUX_DATA },
380 	{ DOSPTYP_LINLVM,	G_PART_ALIAS_LINUX_LVM },
381 	{ DOSPTYP_LINRAID,	G_PART_ALIAS_LINUX_RAID },
382 };
383 
384 static u_char*
385 ldm_privhdr_read(struct g_consumer *cp, uint64_t off, int *error)
386 {
387 	struct g_provider *pp;
388 	u_char *buf;
389 
390 	pp = cp->provider;
391 	buf = g_read_data(cp, off, pp->sectorsize, error);
392 	if (buf == NULL)
393 		return (NULL);
394 
395 	if (memcmp(buf, LDM_PH_SIGN, strlen(LDM_PH_SIGN)) != 0) {
396 		LDM_DEBUG(1, "%s: invalid LDM private header signature",
397 		    pp->name);
398 		g_free(buf);
399 		buf = NULL;
400 		*error = EINVAL;
401 	}
402 	return (buf);
403 }
404 
405 static int
406 ldm_privhdr_parse(struct g_consumer *cp, struct ldm_privhdr *hdr,
407     const u_char *buf)
408 {
409 	uint32_t version;
410 	int error;
411 
412 	memset(hdr, 0, sizeof(*hdr));
413 	version = be32dec(buf + LDM_PH_VERSION_OFF);
414 	if (version != LDM_VERSION_2K &&
415 	    version != LDM_VERSION_VISTA) {
416 		LDM_DEBUG(0, "%s: unsupported LDM version %u.%u",
417 		    cp->provider->name, version >> 16,
418 		    version & 0xFFFF);
419 		return (ENXIO);
420 	}
421 	error = parse_uuid(buf + LDM_PH_DISKGUID_OFF, &hdr->disk_guid);
422 	if (error != 0)
423 		return (error);
424 	error = parse_uuid(buf + LDM_PH_DGGUID_OFF, &hdr->dg_guid);
425 	if (error != 0)
426 		return (error);
427 	strncpy(hdr->dg_name, buf + LDM_PH_DGNAME_OFF, sizeof(hdr->dg_name));
428 	hdr->start = be64dec(buf + LDM_PH_START_OFF);
429 	hdr->size = be64dec(buf + LDM_PH_SIZE_OFF);
430 	hdr->db_offset = be64dec(buf + LDM_PH_DB_OFF);
431 	hdr->db_size = be64dec(buf + LDM_PH_DBSIZE_OFF);
432 	hdr->th_offset[0] = be64dec(buf + LDM_PH_TH1_OFF);
433 	hdr->th_offset[1] = be64dec(buf + LDM_PH_TH2_OFF);
434 	hdr->conf_size = be64dec(buf + LDM_PH_CONFSIZE_OFF);
435 	hdr->log_size = be64dec(buf + LDM_PH_LOGSIZE_OFF);
436 	return (0);
437 }
438 
439 static int
440 ldm_privhdr_check(struct ldm_db *db, struct g_consumer *cp, int is_gpt)
441 {
442 	struct g_consumer *cp2;
443 	struct g_provider *pp;
444 	struct ldm_privhdr hdr;
445 	uint64_t offset, last;
446 	int error, found, i;
447 	u_char *buf;
448 
449 	pp = cp->provider;
450 	if (is_gpt) {
451 		/*
452 		 * The last LBA is used in several checks below, for the
453 		 * GPT case it should be calculated relative to the whole
454 		 * disk.
455 		 */
456 		cp2 = LIST_FIRST(&pp->geom->consumer);
457 		last =
458 		    cp2->provider->mediasize / cp2->provider->sectorsize - 1;
459 	} else
460 		last = pp->mediasize / pp->sectorsize - 1;
461 	for (found = 0, i = is_gpt;
462 	    i < sizeof(ldm_ph_off) / sizeof(ldm_ph_off[0]); i++) {
463 		offset = ldm_ph_off[i];
464 		/*
465 		 * In the GPT case consumer is attached to the LDM metadata
466 		 * partition and we don't need add db_offset.
467 		 */
468 		if (!is_gpt)
469 			offset += db->ph.db_offset;
470 		if (i == LDM_PH_MBRINDEX) {
471 			/*
472 			 * Prepare to errors and setup new base offset
473 			 * to read backup private headers. Assume that LDM
474 			 * database is in the last 1Mbyte area.
475 			 */
476 			db->ph.db_offset = last - LDM_DB_SIZE;
477 		}
478 		buf = ldm_privhdr_read(cp, offset * pp->sectorsize, &error);
479 		if (buf == NULL) {
480 			LDM_DEBUG(1, "%s: failed to read private header "
481 			    "%d at LBA %ju", pp->name, i, (uintmax_t)offset);
482 			continue;
483 		}
484 		error = ldm_privhdr_parse(cp, &hdr, buf);
485 		if (error != 0) {
486 			LDM_DEBUG(1, "%s: failed to parse private "
487 			    "header %d", pp->name, i);
488 			LDM_DUMP(buf, pp->sectorsize);
489 			g_free(buf);
490 			continue;
491 		}
492 		g_free(buf);
493 		if (hdr.start > last ||
494 		    hdr.start + hdr.size - 1 > last ||
495 		    (hdr.start + hdr.size - 1 > hdr.db_offset && !is_gpt) ||
496 		    hdr.db_size != LDM_DB_SIZE ||
497 		    hdr.db_offset + LDM_DB_SIZE - 1 > last ||
498 		    hdr.th_offset[0] >= LDM_DB_SIZE ||
499 		    hdr.th_offset[1] >= LDM_DB_SIZE ||
500 		    hdr.conf_size + hdr.log_size >= LDM_DB_SIZE) {
501 			LDM_DEBUG(1, "%s: invalid values in the "
502 			    "private header %d", pp->name, i);
503 			LDM_DEBUG(2, "%s: start: %jd, size: %jd, "
504 			    "db_offset: %jd, db_size: %jd, th_offset0: %jd, "
505 			    "th_offset1: %jd, conf_size: %jd, log_size: %jd, "
506 			    "last: %jd", pp->name, hdr.start, hdr.size,
507 			    hdr.db_offset, hdr.db_size, hdr.th_offset[0],
508 			    hdr.th_offset[1], hdr.conf_size, hdr.log_size,
509 			    last);
510 			continue;
511 		}
512 		if (found != 0 && memcmp(&db->ph, &hdr, sizeof(hdr)) != 0) {
513 			LDM_DEBUG(0, "%s: private headers are not equal",
514 			    pp->name);
515 			if (i > 1) {
516 				/*
517 				 * We have different headers in the LDM.
518 				 * We can not trust this metadata.
519 				 */
520 				LDM_DEBUG(0, "%s: refuse LDM metadata",
521 				    pp->name);
522 				return (EINVAL);
523 			}
524 			/*
525 			 * We already have read primary private header
526 			 * and it differs from this backup one.
527 			 * Prefer the backup header and save it.
528 			 */
529 			found = 0;
530 		}
531 		if (found == 0)
532 			memcpy(&db->ph, &hdr, sizeof(hdr));
533 		found = 1;
534 	}
535 	if (found == 0) {
536 		LDM_DEBUG(1, "%s: valid LDM private header not found",
537 		    pp->name);
538 		return (ENXIO);
539 	}
540 	return (0);
541 }
542 
543 static int
544 ldm_gpt_check(struct ldm_db *db, struct g_consumer *cp)
545 {
546 	struct g_part_table *gpt;
547 	struct g_part_entry *e;
548 	struct g_consumer *cp2;
549 	int error;
550 
551 	cp2 = LIST_NEXT(cp, consumer);
552 	g_topology_lock();
553 	gpt = cp->provider->geom->softc;
554 	error = 0;
555 	LIST_FOREACH(e, &gpt->gpt_entry, gpe_entry) {
556 		if (cp->provider == e->gpe_pp) {
557 			/* ms-ldm-metadata partition */
558 			if (e->gpe_start != db->ph.db_offset ||
559 			    e->gpe_end != db->ph.db_offset + LDM_DB_SIZE - 1)
560 				error++;
561 		} else if (cp2->provider == e->gpe_pp) {
562 			/* ms-ldm-data partition */
563 			if (e->gpe_start != db->ph.start ||
564 			    e->gpe_end != db->ph.start + db->ph.size - 1)
565 				error++;
566 		}
567 		if (error != 0) {
568 			LDM_DEBUG(0, "%s: GPT partition %d boundaries "
569 			    "do not match with the LDM metadata",
570 			    e->gpe_pp->name, e->gpe_index);
571 			error = ENXIO;
572 			break;
573 		}
574 	}
575 	g_topology_unlock();
576 	return (error);
577 }
578 
579 static int
580 ldm_tochdr_check(struct ldm_db *db, struct g_consumer *cp)
581 {
582 	struct g_provider *pp;
583 	struct ldm_tochdr hdr;
584 	uint64_t offset, conf_size, log_size;
585 	int error, found, i;
586 	u_char *buf;
587 
588 	pp = cp->provider;
589 	for (i = 0, found = 0; i < LDM_TH_COUNT; i++) {
590 		offset = db->ph.db_offset + db->ph.th_offset[i];
591 		buf = g_read_data(cp,
592 		    offset * pp->sectorsize, pp->sectorsize, &error);
593 		if (buf == NULL) {
594 			LDM_DEBUG(1, "%s: failed to read TOC header "
595 			    "at LBA %ju", pp->name, (uintmax_t)offset);
596 			continue;
597 		}
598 		if (memcmp(buf, LDM_TH_SIGN, strlen(LDM_TH_SIGN)) != 0 ||
599 		    memcmp(buf + LDM_TH_NAME1_OFF, LDM_TH_NAME1,
600 		    strlen(LDM_TH_NAME1)) != 0 ||
601 		    memcmp(buf + LDM_TH_NAME2_OFF, LDM_TH_NAME2,
602 		    strlen(LDM_TH_NAME2)) != 0) {
603 			LDM_DEBUG(1, "%s: failed to parse TOC header "
604 			    "at LBA %ju", pp->name, (uintmax_t)offset);
605 			LDM_DUMP(buf, pp->sectorsize);
606 			g_free(buf);
607 			continue;
608 		}
609 		hdr.conf_offset = be64dec(buf + LDM_TH_CONF_OFF);
610 		hdr.log_offset = be64dec(buf + LDM_TH_LOG_OFF);
611 		conf_size = be64dec(buf + LDM_TH_CONFSIZE_OFF);
612 		log_size = be64dec(buf + LDM_TH_LOGSIZE_OFF);
613 		if (conf_size != db->ph.conf_size ||
614 		    hdr.conf_offset + conf_size >= LDM_DB_SIZE ||
615 		    log_size != db->ph.log_size ||
616 		    hdr.log_offset + log_size >= LDM_DB_SIZE) {
617 			LDM_DEBUG(1, "%s: invalid values in the "
618 			    "TOC header at LBA %ju", pp->name,
619 			    (uintmax_t)offset);
620 			LDM_DUMP(buf, pp->sectorsize);
621 			g_free(buf);
622 			continue;
623 		}
624 		g_free(buf);
625 		if (found == 0)
626 			memcpy(&db->th, &hdr, sizeof(hdr));
627 		found = 1;
628 	}
629 	if (found == 0) {
630 		LDM_DEBUG(0, "%s: valid LDM TOC header not found.",
631 		    pp->name);
632 		return (ENXIO);
633 	}
634 	return (0);
635 }
636 
637 static int
638 ldm_vmdbhdr_check(struct ldm_db *db, struct g_consumer *cp)
639 {
640 	struct g_provider *pp;
641 	struct uuid dg_guid;
642 	uint64_t offset;
643 	uint32_t version;
644 	int error;
645 	u_char *buf;
646 
647 	pp = cp->provider;
648 	offset = db->ph.db_offset + db->th.conf_offset;
649 	buf = g_read_data(cp, offset * pp->sectorsize, pp->sectorsize,
650 	    &error);
651 	if (buf == NULL) {
652 		LDM_DEBUG(0, "%s: failed to read VMDB header at "
653 		    "LBA %ju", pp->name, (uintmax_t)offset);
654 		return (error);
655 	}
656 	if (memcmp(buf, LDM_VMDB_SIGN, strlen(LDM_VMDB_SIGN)) != 0) {
657 		g_free(buf);
658 		LDM_DEBUG(0, "%s: failed to parse VMDB header at "
659 		    "LBA %ju", pp->name, (uintmax_t)offset);
660 		return (ENXIO);
661 	}
662 	/* Check version. */
663 	version = be32dec(buf + LDM_DB_VERSION_OFF);
664 	if (version != 0x4000A) {
665 		g_free(buf);
666 		LDM_DEBUG(0, "%s: unsupported VMDB version %u.%u",
667 		    pp->name, version >> 16, version & 0xFFFF);
668 		return (ENXIO);
669 	}
670 	/*
671 	 * Check VMDB update status:
672 	 *	1 - in a consistent state;
673 	 *	2 - in a creation phase;
674 	 *	3 - in a deletion phase;
675 	 */
676 	if (be16dec(buf + LDM_DB_STATUS_OFF) != 1) {
677 		g_free(buf);
678 		LDM_DEBUG(0, "%s: VMDB is not in a consistent state",
679 		    pp->name);
680 		return (ENXIO);
681 	}
682 	db->dh.last_seq = be32dec(buf + LDM_DB_LASTSEQ_OFF);
683 	db->dh.size = be32dec(buf + LDM_DB_SIZE_OFF);
684 	error = parse_uuid(buf + LDM_DB_DGGUID_OFF, &dg_guid);
685 	/* Compare disk group name and guid from VMDB and private headers */
686 	if (error != 0 || db->dh.size == 0 ||
687 	    pp->sectorsize % db->dh.size != 0 ||
688 	    strncmp(buf + LDM_DB_DGNAME_OFF, db->ph.dg_name, 31) != 0 ||
689 	    memcmp(&dg_guid, &db->ph.dg_guid, sizeof(dg_guid)) != 0 ||
690 	    db->dh.size * db->dh.last_seq >
691 	    db->ph.conf_size * pp->sectorsize) {
692 		LDM_DEBUG(0, "%s: invalid values in the VMDB header",
693 		    pp->name);
694 		LDM_DUMP(buf, pp->sectorsize);
695 		g_free(buf);
696 		return (EINVAL);
697 	}
698 	g_free(buf);
699 	return (0);
700 }
701 
702 static int
703 ldm_xvblk_handle(struct ldm_db *db, struct ldm_vblkhdr *vh, const u_char *p)
704 {
705 	struct ldm_xvblk *blk;
706 	size_t size;
707 
708 	size = db->dh.size - 16;
709 	LIST_FOREACH(blk, &db->xvblks, entry)
710 		if (blk->group == vh->group)
711 			break;
712 	if (blk == NULL) {
713 		blk = g_malloc(sizeof(*blk), M_WAITOK | M_ZERO);
714 		blk->group = vh->group;
715 		blk->size = size * vh->count + 16;
716 		blk->data = g_malloc(blk->size, M_WAITOK | M_ZERO);
717 		blk->map = 0xFF << vh->count;
718 		LIST_INSERT_HEAD(&db->xvblks, blk, entry);
719 	}
720 	if ((blk->map & (1 << vh->index)) != 0) {
721 		/* Block with given index has been already saved. */
722 		return (EINVAL);
723 	}
724 	/* Copy the data block to the place related to index. */
725 	memcpy(blk->data + size * vh->index + 16, p + 16, size);
726 	blk->map |= 1 << vh->index;
727 	return (0);
728 }
729 
730 /* Read the variable-width numeric field and return new offset */
731 static int
732 ldm_vnum_get(const u_char *buf, int offset, uint64_t *result, size_t range)
733 {
734 	uint64_t num;
735 	uint8_t len;
736 
737 	len = buf[offset++];
738 	if (len > sizeof(uint64_t) || len + offset >= range)
739 		return (-1);
740 	for (num = 0; len > 0; len--)
741 		num = (num << 8) | buf[offset++];
742 	*result = num;
743 	return (offset);
744 }
745 
746 /* Read the variable-width string and return new offset */
747 static int
748 ldm_vstr_get(const u_char *buf, int offset, u_char *result,
749     size_t maxlen, size_t range)
750 {
751 	uint8_t len;
752 
753 	len = buf[offset++];
754 	if (len >= maxlen || len + offset >= range)
755 		return (-1);
756 	memcpy(result, buf + offset, len);
757 	result[len] = '\0';
758 	return (offset + len);
759 }
760 
761 /* Just skip the variable-width variable and return new offset */
762 static int
763 ldm_vparm_skip(const u_char *buf, int offset, size_t range)
764 {
765 	uint8_t len;
766 
767 	len = buf[offset++];
768 	if (offset + len >= range)
769 		return (-1);
770 
771 	return (offset + len);
772 }
773 
774 static int
775 ldm_vblk_handle(struct ldm_db *db, const u_char *p, size_t size)
776 {
777 	struct ldm_vblk *blk;
778 	struct ldm_volume *volume, *last;
779 	const char *errstr;
780 	u_char vstr[64];
781 	int error, offset;
782 
783 	blk = g_malloc(sizeof(*blk), M_WAITOK | M_ZERO);
784 	blk->type = p[LDM_VBLK_TYPE_OFF];
785 	offset = ldm_vnum_get(p, LDM_VBLK_OID_OFF, &blk->u.id, size);
786 	if (offset < 0) {
787 		errstr = "object id";
788 		goto fail;
789 	}
790 	offset = ldm_vstr_get(p, offset, vstr, sizeof(vstr), size);
791 	if (offset < 0) {
792 		errstr = "object name";
793 		goto fail;
794 	}
795 	switch (blk->type) {
796 	/*
797 	 * Component VBLK fields:
798 	 * Offset	Size	Description
799 	 * ------------+-------+------------------------
800 	 *  0x18+	PS	volume state
801 	 *  0x18+5	PN	component children count
802 	 *  0x1D+16	PN	parent's volume object id
803 	 *  0x2D+1	PN	stripe size
804 	 */
805 	case LDM_VBLK_T_COMPONENT:
806 		offset = ldm_vparm_skip(p, offset, size);
807 		if (offset < 0) {
808 			errstr = "volume state";
809 			goto fail;
810 		}
811 		offset = ldm_vparm_skip(p, offset + 5, size);
812 		if (offset < 0) {
813 			errstr = "children count";
814 			goto fail;
815 		}
816 		offset = ldm_vnum_get(p, offset + 16,
817 		    &blk->u.comp.vol_id, size);
818 		if (offset < 0) {
819 			errstr = "volume id";
820 			goto fail;
821 		}
822 		break;
823 	/*
824 	 * Partition VBLK fields:
825 	 * Offset	Size	Description
826 	 * ------------+-------+------------------------
827 	 *  0x18+12	8	partition start offset
828 	 *  0x18+20	8	volume offset
829 	 *  0x18+28	PN	partition size
830 	 *  0x34+	PN	parent's component object id
831 	 *  0x34+	PN	disk's object id
832 	 */
833 	case LDM_VBLK_T_PARTITION:
834 		if (offset + 28 >= size) {
835 			errstr = "too small buffer";
836 			goto fail;
837 		}
838 		blk->u.part.start = be64dec(p + offset + 12);
839 		blk->u.part.offset = be64dec(p + offset + 20);
840 		offset = ldm_vnum_get(p, offset + 28, &blk->u.part.size, size);
841 		if (offset < 0) {
842 			errstr = "partition size";
843 			goto fail;
844 		}
845 		offset = ldm_vnum_get(p, offset, &blk->u.part.comp_id, size);
846 		if (offset < 0) {
847 			errstr = "component id";
848 			goto fail;
849 		}
850 		offset = ldm_vnum_get(p, offset, &blk->u.part.disk_id, size);
851 		if (offset < 0) {
852 			errstr = "disk id";
853 			goto fail;
854 		}
855 		break;
856 	/*
857 	 * Disk VBLK fields:
858 	 * Offset	Size	Description
859 	 * ------------+-------+------------------------
860 	 *  0x18+	PS	disk GUID
861 	 */
862 	case LDM_VBLK_T_DISK:
863 		errstr = "disk guid";
864 		offset = ldm_vstr_get(p, offset, vstr, sizeof(vstr), size);
865 		if (offset < 0)
866 			goto fail;
867 		error = parse_uuid(vstr, &blk->u.disk.guid);
868 		if (error != 0)
869 			goto fail;
870 		LIST_INSERT_HEAD(&db->disks, &blk->u.disk, entry);
871 		break;
872 	/*
873 	 * Disk group VBLK fields:
874 	 * Offset	Size	Description
875 	 * ------------+-------+------------------------
876 	 *  0x18+	PS	disk group GUID
877 	 */
878 	case LDM_VBLK_T_DISKGROUP:
879 #if 0
880 		strncpy(blk->u.disk_group.name, vstr,
881 		    sizeof(blk->u.disk_group.name));
882 		offset = ldm_vstr_get(p, offset, vstr, sizeof(vstr), size);
883 		if (offset < 0) {
884 			errstr = "disk group guid";
885 			goto fail;
886 		}
887 		error = parse_uuid(name, &blk->u.disk_group.guid);
888 		if (error != 0) {
889 			errstr = "disk group guid";
890 			goto fail;
891 		}
892 		LIST_INSERT_HEAD(&db->groups, &blk->u.disk_group, entry);
893 #endif
894 		break;
895 	/*
896 	 * Disk VBLK fields:
897 	 * Offset	Size	Description
898 	 * ------------+-------+------------------------
899 	 *  0x18+	16	disk GUID
900 	 */
901 	case LDM_VBLK_T_DISK4:
902 		be_uuid_dec(p + offset, &blk->u.disk.guid);
903 		LIST_INSERT_HEAD(&db->disks, &blk->u.disk, entry);
904 		break;
905 	/*
906 	 * Disk group VBLK fields:
907 	 * Offset	Size	Description
908 	 * ------------+-------+------------------------
909 	 *  0x18+	16	disk GUID
910 	 */
911 	case LDM_VBLK_T_DISKGROUP4:
912 #if 0
913 		strncpy(blk->u.disk_group.name, vstr,
914 		    sizeof(blk->u.disk_group.name));
915 		be_uuid_dec(p + offset, &blk->u.disk.guid);
916 		LIST_INSERT_HEAD(&db->groups, &blk->u.disk_group, entry);
917 #endif
918 		break;
919 	/*
920 	 * Volume VBLK fields:
921 	 * Offset	Size	Description
922 	 * ------------+-------+------------------------
923 	 *  0x18+	PS	volume type
924 	 *  0x18+	PS	unknown
925 	 *  0x18+	14(S)	volume state
926 	 *  0x18+16	1	volume number
927 	 *  0x18+21	PN	volume children count
928 	 *  0x2D+16	PN	volume size
929 	 *  0x3D+4	1	partition type
930 	 */
931 	case LDM_VBLK_T_VOLUME:
932 		offset = ldm_vparm_skip(p, offset, size);
933 		if (offset < 0) {
934 			errstr = "volume type";
935 			goto fail;
936 		}
937 		offset = ldm_vparm_skip(p, offset, size);
938 		if (offset < 0) {
939 			errstr = "unknown param";
940 			goto fail;
941 		}
942 		if (offset + 21 >= size) {
943 			errstr = "too small buffer";
944 			goto fail;
945 		}
946 		blk->u.vol.number = p[offset + 16];
947 		offset = ldm_vparm_skip(p, offset + 21, size);
948 		if (offset < 0) {
949 			errstr = "children count";
950 			goto fail;
951 		}
952 		offset = ldm_vnum_get(p, offset + 16, &blk->u.vol.size, size);
953 		if (offset < 0) {
954 			errstr = "volume size";
955 			goto fail;
956 		}
957 		if (offset + 4 >= size) {
958 			errstr = "too small buffer";
959 			goto fail;
960 		}
961 		blk->u.vol.part_type = p[offset + 4];
962 		/* keep volumes ordered by volume number */
963 		last = NULL;
964 		LIST_FOREACH(volume, &db->volumes, entry) {
965 			if (volume->number > blk->u.vol.number)
966 				break;
967 			last = volume;
968 		}
969 		if (last != NULL)
970 			LIST_INSERT_AFTER(last, &blk->u.vol, entry);
971 		else
972 			LIST_INSERT_HEAD(&db->volumes, &blk->u.vol, entry);
973 		break;
974 	default:
975 		LDM_DEBUG(1, "unknown VBLK type 0x%02x\n", blk->type);
976 		LDM_DUMP(p, size);
977 	}
978 	LIST_INSERT_HEAD(&db->vblks, blk, entry);
979 	return (0);
980 fail:
981 	LDM_DEBUG(0, "failed to parse '%s' in VBLK of type 0x%02x\n",
982 	    errstr, blk->type);
983 	LDM_DUMP(p, size);
984 	g_free(blk);
985 	return (EINVAL);
986 }
987 
988 static void
989 ldm_vmdb_free(struct ldm_db *db)
990 {
991 	struct ldm_vblk *vblk;
992 	struct ldm_xvblk *xvblk;
993 
994 	while (!LIST_EMPTY(&db->xvblks)) {
995 		xvblk = LIST_FIRST(&db->xvblks);
996 		LIST_REMOVE(xvblk, entry);
997 		g_free(xvblk->data);
998 		g_free(xvblk);
999 	}
1000 	while (!LIST_EMPTY(&db->vblks)) {
1001 		vblk = LIST_FIRST(&db->vblks);
1002 		LIST_REMOVE(vblk, entry);
1003 		g_free(vblk);
1004 	}
1005 }
1006 
1007 static int
1008 ldm_vmdb_parse(struct ldm_db *db, struct g_consumer *cp)
1009 {
1010 	struct g_provider *pp;
1011 	struct ldm_vblk *vblk;
1012 	struct ldm_xvblk *xvblk;
1013 	struct ldm_volume *volume;
1014 	struct ldm_component *comp;
1015 	struct ldm_vblkhdr vh;
1016 	u_char *buf, *p;
1017 	size_t size, n, sectors;
1018 	uint64_t offset;
1019 	int error;
1020 
1021 	pp = cp->provider;
1022 	size = (db->dh.last_seq * db->dh.size +
1023 	    pp->sectorsize - 1) / pp->sectorsize;
1024 	size -= 1; /* one sector takes vmdb header */
1025 	for (n = 0; n < size; n += MAXPHYS / pp->sectorsize) {
1026 		offset = db->ph.db_offset + db->th.conf_offset + n + 1;
1027 		sectors = (size - n) > (MAXPHYS / pp->sectorsize) ?
1028 		    MAXPHYS / pp->sectorsize: size - n;
1029 		/* read VBLKs */
1030 		buf = g_read_data(cp, offset * pp->sectorsize,
1031 		    sectors * pp->sectorsize, &error);
1032 		if (buf == NULL) {
1033 			LDM_DEBUG(0, "%s: failed to read VBLK\n",
1034 			    pp->name);
1035 			goto fail;
1036 		}
1037 		for (p = buf; p < buf + sectors * pp->sectorsize;
1038 		    p += db->dh.size) {
1039 			if (memcmp(p, LDM_VBLK_SIGN,
1040 			    strlen(LDM_VBLK_SIGN)) != 0) {
1041 				LDM_DEBUG(0, "%s: no VBLK signature\n",
1042 				    pp->name);
1043 				LDM_DUMP(p, db->dh.size);
1044 				goto fail;
1045 			}
1046 			vh.seq = be32dec(p + LDM_VBLK_SEQ_OFF);
1047 			vh.group = be32dec(p + LDM_VBLK_GROUP_OFF);
1048 			/* skip empty blocks */
1049 			if (vh.seq == 0 || vh.group == 0)
1050 				continue;
1051 			vh.index = be16dec(p + LDM_VBLK_INDEX_OFF);
1052 			vh.count = be16dec(p + LDM_VBLK_COUNT_OFF);
1053 			if (vh.count == 0 || vh.count > 4 ||
1054 			    vh.seq > db->dh.last_seq) {
1055 				LDM_DEBUG(0, "%s: invalid values "
1056 				    "in the VBLK header\n", pp->name);
1057 				LDM_DUMP(p, db->dh.size);
1058 				goto fail;
1059 			}
1060 			if (vh.count > 1) {
1061 				error = ldm_xvblk_handle(db, &vh, p);
1062 				if (error != 0) {
1063 					LDM_DEBUG(0, "%s: xVBLK "
1064 					    "is corrupted\n", pp->name);
1065 					LDM_DUMP(p, db->dh.size);
1066 					goto fail;
1067 				}
1068 				continue;
1069 			}
1070 			if (be16dec(p + 16) != 0)
1071 				LDM_DEBUG(1, "%s: VBLK update"
1072 				    " status is %u\n", pp->name,
1073 				    be16dec(p + 16));
1074 			error = ldm_vblk_handle(db, p, db->dh.size);
1075 			if (error != 0)
1076 				goto fail;
1077 		}
1078 		g_free(buf);
1079 		buf = NULL;
1080 	}
1081 	/* Parse xVBLKs */
1082 	while (!LIST_EMPTY(&db->xvblks)) {
1083 		xvblk = LIST_FIRST(&db->xvblks);
1084 		if (xvblk->map == 0xFF) {
1085 			error = ldm_vblk_handle(db, xvblk->data, xvblk->size);
1086 			if (error != 0)
1087 				goto fail;
1088 		} else {
1089 			LDM_DEBUG(0, "%s: incomplete or corrupt "
1090 			    "xVBLK found\n", pp->name);
1091 			goto fail;
1092 		}
1093 		LIST_REMOVE(xvblk, entry);
1094 		g_free(xvblk->data);
1095 		g_free(xvblk);
1096 	}
1097 	/* construct all VBLKs relations */
1098 	LIST_FOREACH(volume, &db->volumes, entry) {
1099 		LIST_FOREACH(vblk, &db->vblks, entry)
1100 			if (vblk->type == LDM_VBLK_T_COMPONENT &&
1101 			    vblk->u.comp.vol_id == volume->id) {
1102 				LIST_INSERT_HEAD(&volume->components,
1103 				    &vblk->u.comp, entry);
1104 				volume->count++;
1105 			}
1106 		LIST_FOREACH(comp, &volume->components, entry)
1107 			LIST_FOREACH(vblk, &db->vblks, entry)
1108 				if (vblk->type == LDM_VBLK_T_PARTITION &&
1109 				    vblk->u.part.comp_id == comp->id) {
1110 					LIST_INSERT_HEAD(&comp->partitions,
1111 					    &vblk->u.part, entry);
1112 					comp->count++;
1113 				}
1114 	}
1115 	return (0);
1116 fail:
1117 	ldm_vmdb_free(db);
1118 	g_free(buf);
1119 	return (ENXIO);
1120 }
1121 
1122 static int
1123 g_part_ldm_add(struct g_part_table *basetable, struct g_part_entry *baseentry,
1124     struct g_part_parms *gpp)
1125 {
1126 
1127 	return (ENOSYS);
1128 }
1129 
1130 static int
1131 g_part_ldm_bootcode(struct g_part_table *basetable, struct g_part_parms *gpp)
1132 {
1133 
1134 	return (ENOSYS);
1135 }
1136 
1137 static int
1138 g_part_ldm_create(struct g_part_table *basetable, struct g_part_parms *gpp)
1139 {
1140 
1141 	return (ENOSYS);
1142 }
1143 
1144 static int
1145 g_part_ldm_destroy(struct g_part_table *basetable, struct g_part_parms *gpp)
1146 {
1147 	struct g_part_ldm_table *table;
1148 	struct g_provider *pp;
1149 
1150 	table = (struct g_part_ldm_table *)basetable;
1151 	/*
1152 	 * To destroy LDM on a disk partitioned with GPT we should delete
1153 	 * ms-ldm-metadata partition, but we can't do this via standard
1154 	 * GEOM_PART method.
1155 	 */
1156 	if (table->is_gpt)
1157 		return (ENOSYS);
1158 	pp = LIST_FIRST(&basetable->gpt_gp->consumer)->provider;
1159 	/*
1160 	 * To destroy LDM we should wipe MBR, first private header and
1161 	 * backup private headers.
1162 	 */
1163 	basetable->gpt_smhead = (1 << ldm_ph_off[0]) | 1;
1164 	/*
1165 	 * Don't touch last backup private header when LDM database is
1166 	 * not located in the last 1MByte area.
1167 	 * XXX: can't remove all blocks.
1168 	 */
1169 	if (table->db_offset + LDM_DB_SIZE ==
1170 	    pp->mediasize / pp->sectorsize)
1171 		basetable->gpt_smtail = 1;
1172 	return (0);
1173 }
1174 
1175 static void
1176 g_part_ldm_dumpconf(struct g_part_table *basetable,
1177     struct g_part_entry *baseentry, struct sbuf *sb, const char *indent)
1178 {
1179 	struct g_part_ldm_entry *entry;
1180 
1181 	entry = (struct g_part_ldm_entry *)baseentry;
1182 	if (indent == NULL) {
1183 		/* conftxt: libdisk compatibility */
1184 		sbuf_printf(sb, " xs LDM xt %u", entry->type);
1185 	} else if (entry != NULL) {
1186 		/* confxml: partition entry information */
1187 		sbuf_printf(sb, "%s<rawtype>%u</rawtype>\n", indent,
1188 		    entry->type);
1189 	} else {
1190 		/* confxml: scheme information */
1191 	}
1192 }
1193 
1194 static int
1195 g_part_ldm_dumpto(struct g_part_table *table, struct g_part_entry *baseentry)
1196 {
1197 
1198 	return (0);
1199 }
1200 
1201 static int
1202 g_part_ldm_modify(struct g_part_table *basetable,
1203     struct g_part_entry *baseentry, struct g_part_parms *gpp)
1204 {
1205 
1206 	return (ENOSYS);
1207 }
1208 
1209 static int
1210 g_part_ldm_resize(struct g_part_table *basetable,
1211     struct g_part_entry *baseentry, struct g_part_parms *gpp)
1212 {
1213 
1214 	return (ENOSYS);
1215 }
1216 
1217 static const char *
1218 g_part_ldm_name(struct g_part_table *table, struct g_part_entry *baseentry,
1219     char *buf, size_t bufsz)
1220 {
1221 
1222 	snprintf(buf, bufsz, "s%d", baseentry->gpe_index);
1223 	return (buf);
1224 }
1225 
1226 static int
1227 ldm_gpt_probe(struct g_part_table *basetable, struct g_consumer *cp)
1228 {
1229 	struct g_part_ldm_table *table;
1230 	struct g_part_table *gpt;
1231 	struct g_part_entry *entry;
1232 	struct g_consumer *cp2;
1233 	struct gpt_ent *part;
1234 	u_char *buf;
1235 	int error;
1236 
1237 	/*
1238 	 * XXX: We use some knowlege about GEOM_PART_GPT internal
1239 	 * structures, but it is easier than parse GPT by himself.
1240 	 */
1241 	g_topology_lock();
1242 	gpt = cp->provider->geom->softc;
1243 	LIST_FOREACH(entry, &gpt->gpt_entry, gpe_entry) {
1244 		part = (struct gpt_ent *)(entry + 1);
1245 		/* Search ms-ldm-metadata partition */
1246 		if (memcmp(&part->ent_type,
1247 		    &gpt_uuid_ms_ldm_metadata, sizeof(struct uuid)) != 0 ||
1248 		    entry->gpe_end - entry->gpe_start < LDM_DB_SIZE - 1)
1249 			continue;
1250 
1251 		/* Create new consumer and attach it to metadata partition */
1252 		cp2 = g_new_consumer(cp->geom);
1253 		error = g_attach(cp2, entry->gpe_pp);
1254 		if (error != 0) {
1255 			g_destroy_consumer(cp2);
1256 			g_topology_unlock();
1257 			return (ENXIO);
1258 		}
1259 		error = g_access(cp2, 1, 0, 0);
1260 		if (error != 0) {
1261 			g_detach(cp2);
1262 			g_destroy_consumer(cp2);
1263 			g_topology_unlock();
1264 			return (ENXIO);
1265 		}
1266 		g_topology_unlock();
1267 
1268 		LDM_DEBUG(2, "%s: LDM metadata partition %s found in the GPT",
1269 		    cp->provider->name, cp2->provider->name);
1270 		/* Read the LDM private header */
1271 		buf = ldm_privhdr_read(cp2,
1272 		    ldm_ph_off[LDM_PH_GPTINDEX] * cp2->provider->sectorsize,
1273 		    &error);
1274 		if (buf != NULL) {
1275 			table = (struct g_part_ldm_table *)basetable;
1276 			table->is_gpt = 1;
1277 			g_free(buf);
1278 			return (G_PART_PROBE_PRI_HIGH);
1279 		}
1280 
1281 		/* second consumer is no longer needed. */
1282 		g_topology_lock();
1283 		g_access(cp2, -1, 0, 0);
1284 		g_detach(cp2);
1285 		g_destroy_consumer(cp2);
1286 		break;
1287 	}
1288 	g_topology_unlock();
1289 	return (ENXIO);
1290 }
1291 
1292 static int
1293 g_part_ldm_probe(struct g_part_table *basetable, struct g_consumer *cp)
1294 {
1295 	struct g_provider *pp;
1296 	u_char *buf, type[64];
1297 	int error, idx;
1298 
1299 
1300 	pp = cp->provider;
1301 	if (pp->sectorsize != 512)
1302 		return (ENXIO);
1303 
1304 	error = g_getattr("PART::scheme", cp, &type);
1305 	if (error == 0 && strcmp(type, "GPT") == 0) {
1306 		if (g_getattr("PART::type", cp, &type) != 0 ||
1307 		    strcmp(type, "ms-ldm-data") != 0)
1308 			return (ENXIO);
1309 		error = ldm_gpt_probe(basetable, cp);
1310 		return (error);
1311 	}
1312 
1313 	if (basetable->gpt_depth != 0)
1314 		return (ENXIO);
1315 
1316 	/* LDM has 1M metadata area */
1317 	if (pp->mediasize <= 1024 * 1024)
1318 		return (ENOSPC);
1319 
1320 	/* Check that there's a MBR */
1321 	buf = g_read_data(cp, 0, pp->sectorsize, &error);
1322 	if (buf == NULL)
1323 		return (error);
1324 
1325 	if (le16dec(buf + DOSMAGICOFFSET) != DOSMAGIC) {
1326 		g_free(buf);
1327 		return (ENXIO);
1328 	}
1329 	error = ENXIO;
1330 	/* Check that we have LDM partitions in the MBR */
1331 	for (idx = 0; idx < NDOSPART && error != 0; idx++) {
1332 		if (buf[DOSPARTOFF + idx * DOSPARTSIZE + 4] == DOSPTYP_LDM)
1333 			error = 0;
1334 	}
1335 	g_free(buf);
1336 	if (error == 0) {
1337 		LDM_DEBUG(2, "%s: LDM data partitions found in MBR",
1338 		    pp->name);
1339 		/* Read the LDM private header */
1340 		buf = ldm_privhdr_read(cp,
1341 		    ldm_ph_off[LDM_PH_MBRINDEX] * pp->sectorsize, &error);
1342 		if (buf == NULL)
1343 			return (error);
1344 		g_free(buf);
1345 		return (G_PART_PROBE_PRI_HIGH);
1346 	}
1347 	return (error);
1348 }
1349 
1350 static int
1351 g_part_ldm_read(struct g_part_table *basetable, struct g_consumer *cp)
1352 {
1353 	struct g_part_ldm_table *table;
1354 	struct g_part_ldm_entry *entry;
1355 	struct g_consumer *cp2;
1356 	struct ldm_component *comp;
1357 	struct ldm_partition *part;
1358 	struct ldm_volume *vol;
1359 	struct ldm_disk *disk;
1360 	struct ldm_db db;
1361 	int error, index, skipped;
1362 
1363 	table = (struct g_part_ldm_table *)basetable;
1364 	memset(&db, 0, sizeof(db));
1365 	cp2 = cp;					/* ms-ldm-data */
1366 	if (table->is_gpt)
1367 		cp = LIST_FIRST(&cp->geom->consumer);	/* ms-ldm-metadata */
1368 	/* Read and parse LDM private headers. */
1369 	error = ldm_privhdr_check(&db, cp, table->is_gpt);
1370 	if (error != 0)
1371 		goto gpt_cleanup;
1372 	basetable->gpt_first = table->is_gpt ? 0: db.ph.start;
1373 	basetable->gpt_last = basetable->gpt_first + db.ph.size - 1;
1374 	table->db_offset = db.ph.db_offset;
1375 	/* Make additional checks for GPT */
1376 	if (table->is_gpt) {
1377 		error = ldm_gpt_check(&db, cp);
1378 		if (error != 0)
1379 			goto gpt_cleanup;
1380 		/*
1381 		 * Now we should reset database offset to zero, because our
1382 		 * consumer cp is attached to the ms-ldm-metadata partition
1383 		 * and we don't need add db_offset to read from it.
1384 		 */
1385 		db.ph.db_offset = 0;
1386 	}
1387 	/* Read and parse LDM TOC headers. */
1388 	error = ldm_tochdr_check(&db, cp);
1389 	if (error != 0)
1390 		goto gpt_cleanup;
1391 	/* Read and parse LDM VMDB header. */
1392 	error = ldm_vmdbhdr_check(&db, cp);
1393 	if (error != 0)
1394 		goto gpt_cleanup;
1395 	error = ldm_vmdb_parse(&db, cp);
1396 	/*
1397 	 * For the GPT case we must detach and destroy
1398 	 * second consumer before return.
1399 	 */
1400 gpt_cleanup:
1401 	if (table->is_gpt) {
1402 		g_topology_lock();
1403 		g_access(cp, -1, 0, 0);
1404 		g_detach(cp);
1405 		g_destroy_consumer(cp);
1406 		g_topology_unlock();
1407 		cp = cp2;
1408 	}
1409 	if (error != 0)
1410 		return (error);
1411 	/* Search current disk in the disk list. */
1412 	LIST_FOREACH(disk, &db.disks, entry)
1413 	    if (memcmp(&disk->guid, &db.ph.disk_guid,
1414 		sizeof(struct uuid)) == 0)
1415 		    break;
1416 	if (disk == NULL) {
1417 		LDM_DEBUG(1, "%s: no LDM volumes on this disk",
1418 		    cp->provider->name);
1419 		ldm_vmdb_free(&db);
1420 		return (ENXIO);
1421 	}
1422 	index = 1;
1423 	LIST_FOREACH(vol, &db.volumes, entry) {
1424 		LIST_FOREACH(comp, &vol->components, entry) {
1425 			/* Skip volumes from different disks. */
1426 			part = LIST_FIRST(&comp->partitions);
1427 			if (part->disk_id != disk->id)
1428 				continue;
1429 			skipped = 0;
1430 			/* We don't support spanned and striped volumes. */
1431 			if (comp->count > 1 || part->offset != 0) {
1432 				LDM_DEBUG(1, "%s: LDM volume component "
1433 				    "%ju has %u partitions. Skipped",
1434 				    cp->provider->name, (uintmax_t)comp->id,
1435 				    comp->count);
1436 				skipped = 1;
1437 			}
1438 			/*
1439 			 * Allow mirrored volumes only when they are explicitly
1440 			 * allowed with kern.geom.part.ldm.show_mirrors=1.
1441 			 */
1442 			if (vol->count > 1 && show_mirrors == 0) {
1443 				LDM_DEBUG(1, "%s: LDM volume %ju has %u "
1444 				    "components. Skipped",
1445 				    cp->provider->name, (uintmax_t)vol->id,
1446 				    vol->count);
1447 				skipped = 1;
1448 			}
1449 			entry = (struct g_part_ldm_entry *)g_part_new_entry(
1450 			    basetable, index++,
1451 			    basetable->gpt_first + part->start,
1452 			    basetable->gpt_first + part->start +
1453 			    part->size - 1);
1454 			/*
1455 			 * Mark skipped partition as ms-ldm-data partition.
1456 			 * We do not support them, but it is better to show
1457 			 * that we have something there, than just show
1458 			 * free space.
1459 			 */
1460 			if (skipped == 0)
1461 				entry->type = vol->part_type;
1462 			else
1463 				entry->type = DOSPTYP_LDM;
1464 			LDM_DEBUG(1, "%s: new volume id: %ju, start: %ju,"
1465 			    " end: %ju, type: 0x%02x\n", cp->provider->name,
1466 			    (uintmax_t)part->id,(uintmax_t)part->start +
1467 			    basetable->gpt_first, (uintmax_t)part->start +
1468 			    part->size + basetable->gpt_first - 1,
1469 			    vol->part_type);
1470 		}
1471 	}
1472 	ldm_vmdb_free(&db);
1473 	return (error);
1474 }
1475 
1476 static const char *
1477 g_part_ldm_type(struct g_part_table *basetable, struct g_part_entry *baseentry,
1478     char *buf, size_t bufsz)
1479 {
1480 	struct g_part_ldm_entry *entry;
1481 	int i;
1482 
1483 	entry = (struct g_part_ldm_entry *)baseentry;
1484 	for (i = 0;
1485 	    i < sizeof(ldm_alias_match) / sizeof(ldm_alias_match[0]); i++) {
1486 		if (ldm_alias_match[i].typ == entry->type)
1487 			return (g_part_alias_name(ldm_alias_match[i].alias));
1488 	}
1489 	snprintf(buf, bufsz, "!%d", entry->type);
1490 	return (buf);
1491 }
1492 
1493 static int
1494 g_part_ldm_write(struct g_part_table *basetable, struct g_consumer *cp)
1495 {
1496 
1497 	return (ENOSYS);
1498 }
1499