xref: /freebsd/sys/geom/part/g_part_ldm.c (revision 7899f917b1c0ea178f1d2be0cfb452086d079d23)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2012 Andrey V. Elsukov <ae@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 #include <sys/param.h>
30 #include <sys/bio.h>
31 #include <sys/diskmbr.h>
32 #include <sys/endian.h>
33 #include <sys/gpt.h>
34 #include <sys/kernel.h>
35 #include <sys/kobj.h>
36 #include <sys/limits.h>
37 #include <sys/lock.h>
38 #include <sys/malloc.h>
39 #include <sys/mutex.h>
40 #include <sys/queue.h>
41 #include <sys/sbuf.h>
42 #include <sys/systm.h>
43 #include <sys/sysctl.h>
44 #include <sys/uuid.h>
45 #include <geom/geom.h>
46 #include <geom/part/g_part.h>
47 
48 #include "g_part_if.h"
49 
50 FEATURE(geom_part_ldm, "GEOM partitioning class for LDM support");
51 
52 SYSCTL_DECL(_kern_geom_part);
53 static SYSCTL_NODE(_kern_geom_part, OID_AUTO, ldm,
54     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
55     "GEOM_PART_LDM Logical Disk Manager");
56 
57 static u_int ldm_debug = 0;
58 SYSCTL_UINT(_kern_geom_part_ldm, OID_AUTO, debug,
59     CTLFLAG_RWTUN, &ldm_debug, 0, "Debug level");
60 
61 /*
62  * This allows access to mirrored LDM volumes. Since we do not
63  * doing mirroring here, it is not enabled by default.
64  */
65 static u_int show_mirrors = 0;
66 SYSCTL_UINT(_kern_geom_part_ldm, OID_AUTO, show_mirrors,
67     CTLFLAG_RWTUN, &show_mirrors, 0, "Show mirrored volumes");
68 
69 #define	LDM_DEBUG(lvl, fmt, ...)	do {				\
70 	if (ldm_debug >= (lvl)) {					\
71 		printf("GEOM_PART: " fmt "\n", __VA_ARGS__);		\
72 	}								\
73 } while (0)
74 #define	LDM_DUMP(buf, size)	do {					\
75 	if (ldm_debug > 1) {						\
76 		hexdump(buf, size, NULL, 0);				\
77 	}								\
78 } while (0)
79 
80 /*
81  * There are internal representations of LDM structures.
82  *
83  * We do not keep all fields of on-disk structures, only most useful.
84  * All numbers in an on-disk structures are in big-endian format.
85  */
86 
87 /*
88  * Private header is 512 bytes long. There are three copies on each disk.
89  * Offset and sizes are in sectors. Location of each copy:
90  * - the first offset is relative to the disk start;
91  * - the second and third offset are relative to the LDM database start.
92  *
93  * On a disk partitioned with GPT, the LDM has not first private header.
94  */
95 #define	LDM_PH_MBRINDEX		0
96 #define	LDM_PH_GPTINDEX		2
97 static const uint64_t	ldm_ph_off[] = {6, 1856, 2047};
98 #define	LDM_VERSION_2K		0x2000b
99 #define	LDM_VERSION_VISTA	0x2000c
100 #define	LDM_PH_VERSION_OFF	0x00c
101 #define	LDM_PH_DISKGUID_OFF	0x030
102 #define	LDM_PH_DGGUID_OFF	0x0b0
103 #define	LDM_PH_DGNAME_OFF	0x0f0
104 #define	LDM_PH_START_OFF	0x11b
105 #define	LDM_PH_SIZE_OFF		0x123
106 #define	LDM_PH_DB_OFF		0x12b
107 #define	LDM_PH_DBSIZE_OFF	0x133
108 #define	LDM_PH_TH1_OFF		0x13b
109 #define	LDM_PH_TH2_OFF		0x143
110 #define	LDM_PH_CONFSIZE_OFF	0x153
111 #define	LDM_PH_LOGSIZE_OFF	0x15b
112 #define	LDM_PH_SIGN		"PRIVHEAD"
113 struct ldm_privhdr {
114 	struct uuid	disk_guid;
115 	struct uuid	dg_guid;
116 	u_char		dg_name[32];
117 	uint64_t	start;		/* logical disk start */
118 	uint64_t	size;		/* logical disk size */
119 	uint64_t	db_offset;	/* LDM database start */
120 #define	LDM_DB_SIZE		2048
121 	uint64_t	db_size;	/* LDM database size */
122 #define	LDM_TH_COUNT		2
123 	uint64_t	th_offset[LDM_TH_COUNT]; /* TOC header offsets */
124 	uint64_t	conf_size;	/* configuration size */
125 	uint64_t	log_size;	/* size of log */
126 };
127 
128 /*
129  * Table of contents header is 512 bytes long.
130  * There are two identical copies at offsets from the private header.
131  * Offsets are relative to the LDM database start.
132  */
133 #define	LDM_TH_SIGN		"TOCBLOCK"
134 #define	LDM_TH_NAME1		"config"
135 #define	LDM_TH_NAME2		"log"
136 #define	LDM_TH_NAME1_OFF	0x024
137 #define	LDM_TH_CONF_OFF		0x02e
138 #define	LDM_TH_CONFSIZE_OFF	0x036
139 #define	LDM_TH_NAME2_OFF	0x046
140 #define	LDM_TH_LOG_OFF		0x050
141 #define	LDM_TH_LOGSIZE_OFF	0x058
142 struct ldm_tochdr {
143 	uint64_t	conf_offset;	/* configuration offset */
144 	uint64_t	log_offset;	/* log offset */
145 };
146 
147 /*
148  * LDM database header is 512 bytes long.
149  */
150 #define	LDM_VMDB_SIGN		"VMDB"
151 #define	LDM_DB_LASTSEQ_OFF	0x004
152 #define	LDM_DB_SIZE_OFF		0x008
153 #define	LDM_DB_STATUS_OFF	0x010
154 #define	LDM_DB_VERSION_OFF	0x012
155 #define	LDM_DB_DGNAME_OFF	0x016
156 #define	LDM_DB_DGGUID_OFF	0x035
157 struct ldm_vmdbhdr {
158 	uint32_t	last_seq;	/* sequence number of last VBLK */
159 	uint32_t	size;		/* size of VBLK */
160 };
161 
162 /*
163  * The LDM database configuration section contains VMDB header and
164  * many VBLKs. Each VBLK represents a disk group, disk partition,
165  * component or volume.
166  *
167  * The most interesting for us are volumes, they are represents
168  * partitions in the GEOM_PART meaning. But volume VBLK does not
169  * contain all information needed to create GEOM provider. And we
170  * should get this information from the related VBLK. This is how
171  * VBLK releated:
172  *	Volumes <- Components <- Partitions -> Disks
173  *
174  * One volume can contain several components. In this case LDM
175  * does mirroring of volume data to each component.
176  *
177  * Also each component can contain several partitions (spanned or
178  * striped volumes).
179  */
180 
181 struct ldm_component {
182 	uint64_t	id;		/* object id */
183 	uint64_t	vol_id;		/* parent volume object id */
184 
185 	int		count;
186 	LIST_HEAD(, ldm_partition) partitions;
187 	LIST_ENTRY(ldm_component) entry;
188 };
189 
190 struct ldm_volume {
191 	uint64_t	id;		/* object id */
192 	uint64_t	size;		/* volume size */
193 	uint8_t		number;		/* used for ordering */
194 	uint8_t		part_type;	/* partition type */
195 
196 	int		count;
197 	LIST_HEAD(, ldm_component) components;
198 	LIST_ENTRY(ldm_volume)	entry;
199 };
200 
201 struct ldm_disk {
202 	uint64_t	id;		/* object id */
203 	struct uuid	guid;		/* disk guid */
204 
205 	LIST_ENTRY(ldm_disk) entry;
206 };
207 
208 #if 0
209 struct ldm_disk_group {
210 	uint64_t	id;		/* object id */
211 	struct uuid	guid;		/* disk group guid */
212 	u_char		name[32];	/* disk group name */
213 
214 	LIST_ENTRY(ldm_disk_group) entry;
215 };
216 #endif
217 
218 struct ldm_partition {
219 	uint64_t	id;		/* object id */
220 	uint64_t	disk_id;	/* disk object id */
221 	uint64_t	comp_id;	/* parent component object id */
222 	uint64_t	start;		/* offset relative to disk start */
223 	uint64_t	offset;		/* offset for spanned volumes */
224 	uint64_t	size;		/* partition size */
225 
226 	LIST_ENTRY(ldm_partition) entry;
227 };
228 
229 /*
230  * Each VBLK is 128 bytes long and has standard 16 bytes header.
231  * Some of VBLK's fields are fixed size, but others has variable size.
232  * Fields with variable size are prefixed with one byte length marker.
233  * Some fields are strings and also can have fixed size and variable.
234  * Strings with fixed size are NULL-terminated, others are not.
235  * All VBLKs have same several first fields:
236  *	Offset		Size		Description
237  *	---------------+---------------+--------------------------
238  *	0x00		16		standard VBLK header
239  *	0x10		2		update status
240  *	0x13		1		VBLK type
241  *	0x18		PS		object id
242  *	0x18+		PN		object name
243  *
244  *  o Offset 0x18+ means '0x18 + length of all variable-width fields'
245  *  o 'P' in size column means 'prefixed' (variable-width),
246  *    'S' - string, 'N' - number.
247  */
248 #define	LDM_VBLK_SIGN		"VBLK"
249 #define	LDM_VBLK_SEQ_OFF	0x04
250 #define	LDM_VBLK_GROUP_OFF	0x08
251 #define	LDM_VBLK_INDEX_OFF	0x0c
252 #define	LDM_VBLK_COUNT_OFF	0x0e
253 #define	LDM_VBLK_TYPE_OFF	0x13
254 #define	LDM_VBLK_OID_OFF	0x18
255 struct ldm_vblkhdr {
256 	uint32_t	seq;		/* sequence number */
257 	uint32_t	group;		/* group number */
258 	uint16_t	index;		/* index in the group */
259 	uint16_t	count;		/* number of entries in the group */
260 };
261 
262 #define	LDM_VBLK_T_COMPONENT	0x32
263 #define	LDM_VBLK_T_PARTITION	0x33
264 #define	LDM_VBLK_T_DISK		0x34
265 #define	LDM_VBLK_T_DISKGROUP	0x35
266 #define	LDM_VBLK_T_DISK4	0x44
267 #define	LDM_VBLK_T_DISKGROUP4	0x45
268 #define	LDM_VBLK_T_VOLUME	0x51
269 struct ldm_vblk {
270 	uint8_t		type;		/* VBLK type */
271 	union {
272 		uint64_t		id;
273 		struct ldm_volume	vol;
274 		struct ldm_component	comp;
275 		struct ldm_disk		disk;
276 		struct ldm_partition	part;
277 #if 0
278 		struct ldm_disk_group	disk_group;
279 #endif
280 	} u;
281 	LIST_ENTRY(ldm_vblk) entry;
282 };
283 
284 /*
285  * Some VBLKs contains a bit more data than can fit into 128 bytes. These
286  * VBLKs are called eXtended VBLK. Before parsing, the data from these VBLK
287  * should be placed into continuous memory buffer. We can determine xVBLK
288  * by the count field in the standard VBLK header (count > 1).
289  */
290 struct ldm_xvblk {
291 	uint32_t	group;		/* xVBLK group number */
292 	uint32_t	size;		/* the total size of xVBLK */
293 	uint8_t		map;		/* bitmask of currently saved VBLKs */
294 	u_char		*data;		/* xVBLK data */
295 
296 	LIST_ENTRY(ldm_xvblk)	entry;
297 };
298 
299 /* The internal representation of LDM database. */
300 struct ldm_db {
301 	struct ldm_privhdr		ph;	/* private header */
302 	struct ldm_tochdr		th;	/* TOC header */
303 	struct ldm_vmdbhdr		dh;	/* VMDB header */
304 
305 	LIST_HEAD(, ldm_volume)		volumes;
306 	LIST_HEAD(, ldm_disk)		disks;
307 	LIST_HEAD(, ldm_vblk)		vblks;
308 	LIST_HEAD(, ldm_xvblk)		xvblks;
309 };
310 
311 static struct uuid gpt_uuid_ms_ldm_metadata = GPT_ENT_TYPE_MS_LDM_METADATA;
312 
313 struct g_part_ldm_table {
314 	struct g_part_table	base;
315 	uint64_t		db_offset;
316 	int			is_gpt;
317 };
318 struct g_part_ldm_entry {
319 	struct g_part_entry	base;
320 	uint8_t			type;
321 };
322 
323 static int g_part_ldm_add(struct g_part_table *, struct g_part_entry *,
324     struct g_part_parms *);
325 static int g_part_ldm_bootcode(struct g_part_table *, struct g_part_parms *);
326 static int g_part_ldm_create(struct g_part_table *, struct g_part_parms *);
327 static int g_part_ldm_destroy(struct g_part_table *, struct g_part_parms *);
328 static void g_part_ldm_dumpconf(struct g_part_table *, struct g_part_entry *,
329     struct sbuf *, const char *);
330 static int g_part_ldm_dumpto(struct g_part_table *, struct g_part_entry *);
331 static int g_part_ldm_modify(struct g_part_table *, struct g_part_entry *,
332     struct g_part_parms *);
333 static const char *g_part_ldm_name(struct g_part_table *, struct g_part_entry *,
334     char *, size_t);
335 static int g_part_ldm_probe(struct g_part_table *, struct g_consumer *);
336 static int g_part_ldm_read(struct g_part_table *, struct g_consumer *);
337 static const char *g_part_ldm_type(struct g_part_table *, struct g_part_entry *,
338     char *, size_t);
339 static int g_part_ldm_write(struct g_part_table *, struct g_consumer *);
340 
341 static kobj_method_t g_part_ldm_methods[] = {
342 	KOBJMETHOD(g_part_add,		g_part_ldm_add),
343 	KOBJMETHOD(g_part_bootcode,	g_part_ldm_bootcode),
344 	KOBJMETHOD(g_part_create,	g_part_ldm_create),
345 	KOBJMETHOD(g_part_destroy,	g_part_ldm_destroy),
346 	KOBJMETHOD(g_part_dumpconf,	g_part_ldm_dumpconf),
347 	KOBJMETHOD(g_part_dumpto,	g_part_ldm_dumpto),
348 	KOBJMETHOD(g_part_modify,	g_part_ldm_modify),
349 	KOBJMETHOD(g_part_name,		g_part_ldm_name),
350 	KOBJMETHOD(g_part_probe,	g_part_ldm_probe),
351 	KOBJMETHOD(g_part_read,		g_part_ldm_read),
352 	KOBJMETHOD(g_part_type,		g_part_ldm_type),
353 	KOBJMETHOD(g_part_write,	g_part_ldm_write),
354 	{ 0, 0 }
355 };
356 
357 static struct g_part_scheme g_part_ldm_scheme = {
358 	"LDM",
359 	g_part_ldm_methods,
360 	sizeof(struct g_part_ldm_table),
361 	.gps_entrysz = sizeof(struct g_part_ldm_entry)
362 };
363 G_PART_SCHEME_DECLARE(g_part_ldm);
364 MODULE_VERSION(geom_part_ldm, 0);
365 
366 static struct g_part_ldm_alias {
367 	u_char		typ;
368 	int		alias;
369 } ldm_alias_match[] = {
370 	{ DOSPTYP_386BSD,	G_PART_ALIAS_FREEBSD },
371 	{ DOSPTYP_FAT32,	G_PART_ALIAS_MS_FAT32 },
372 	{ DOSPTYP_FAT32LBA,	G_PART_ALIAS_MS_FAT32LBA },
373 	{ DOSPTYP_LDM,		G_PART_ALIAS_MS_LDM_DATA },
374 	{ DOSPTYP_LINLVM,	G_PART_ALIAS_LINUX_LVM },
375 	{ DOSPTYP_LINRAID,	G_PART_ALIAS_LINUX_RAID },
376 	{ DOSPTYP_LINSWP,	G_PART_ALIAS_LINUX_SWAP },
377 	{ DOSPTYP_LINUX,	G_PART_ALIAS_LINUX_DATA },
378 	{ DOSPTYP_NTFS,		G_PART_ALIAS_MS_NTFS },
379 };
380 
381 static u_char*
382 ldm_privhdr_read(struct g_consumer *cp, uint64_t off, int *error)
383 {
384 	struct g_provider *pp;
385 	u_char *buf;
386 
387 	pp = cp->provider;
388 	buf = g_read_data(cp, off, pp->sectorsize, error);
389 	if (buf == NULL)
390 		return (NULL);
391 
392 	if (memcmp(buf, LDM_PH_SIGN, strlen(LDM_PH_SIGN)) != 0) {
393 		LDM_DEBUG(1, "%s: invalid LDM private header signature",
394 		    pp->name);
395 		g_free(buf);
396 		buf = NULL;
397 		*error = EINVAL;
398 	}
399 	return (buf);
400 }
401 
402 static int
403 ldm_privhdr_parse(struct g_consumer *cp, struct ldm_privhdr *hdr,
404     const u_char *buf)
405 {
406 	uint32_t version;
407 	int error;
408 
409 	memset(hdr, 0, sizeof(*hdr));
410 	version = be32dec(buf + LDM_PH_VERSION_OFF);
411 	if (version != LDM_VERSION_2K &&
412 	    version != LDM_VERSION_VISTA) {
413 		LDM_DEBUG(0, "%s: unsupported LDM version %u.%u",
414 		    cp->provider->name, version >> 16,
415 		    version & 0xFFFF);
416 		return (ENXIO);
417 	}
418 	error = parse_uuid(buf + LDM_PH_DISKGUID_OFF, &hdr->disk_guid);
419 	if (error != 0)
420 		return (error);
421 	error = parse_uuid(buf + LDM_PH_DGGUID_OFF, &hdr->dg_guid);
422 	if (error != 0)
423 		return (error);
424 	strncpy(hdr->dg_name, buf + LDM_PH_DGNAME_OFF, sizeof(hdr->dg_name));
425 	hdr->start = be64dec(buf + LDM_PH_START_OFF);
426 	hdr->size = be64dec(buf + LDM_PH_SIZE_OFF);
427 	hdr->db_offset = be64dec(buf + LDM_PH_DB_OFF);
428 	hdr->db_size = be64dec(buf + LDM_PH_DBSIZE_OFF);
429 	hdr->th_offset[0] = be64dec(buf + LDM_PH_TH1_OFF);
430 	hdr->th_offset[1] = be64dec(buf + LDM_PH_TH2_OFF);
431 	hdr->conf_size = be64dec(buf + LDM_PH_CONFSIZE_OFF);
432 	hdr->log_size = be64dec(buf + LDM_PH_LOGSIZE_OFF);
433 	return (0);
434 }
435 
436 static int
437 ldm_privhdr_check(struct ldm_db *db, struct g_consumer *cp, int is_gpt)
438 {
439 	struct g_consumer *cp2;
440 	struct g_provider *pp;
441 	struct ldm_privhdr hdr;
442 	uint64_t offset, last;
443 	int error, found, i;
444 	u_char *buf;
445 
446 	pp = cp->provider;
447 	if (is_gpt) {
448 		/*
449 		 * The last LBA is used in several checks below, for the
450 		 * GPT case it should be calculated relative to the whole
451 		 * disk.
452 		 */
453 		cp2 = LIST_FIRST(&pp->geom->consumer);
454 		last =
455 		    cp2->provider->mediasize / cp2->provider->sectorsize - 1;
456 	} else
457 		last = pp->mediasize / pp->sectorsize - 1;
458 	for (found = 0, i = is_gpt; i < nitems(ldm_ph_off); i++) {
459 		offset = ldm_ph_off[i];
460 		/*
461 		 * In the GPT case consumer is attached to the LDM metadata
462 		 * partition and we don't need add db_offset.
463 		 */
464 		if (!is_gpt)
465 			offset += db->ph.db_offset;
466 		if (i == LDM_PH_MBRINDEX) {
467 			/*
468 			 * Prepare to errors and setup new base offset
469 			 * to read backup private headers. Assume that LDM
470 			 * database is in the last 1Mbyte area.
471 			 */
472 			db->ph.db_offset = last - LDM_DB_SIZE;
473 		}
474 		buf = ldm_privhdr_read(cp, offset * pp->sectorsize, &error);
475 		if (buf == NULL) {
476 			LDM_DEBUG(1, "%s: failed to read private header "
477 			    "%d at LBA %ju", pp->name, i, (uintmax_t)offset);
478 			continue;
479 		}
480 		error = ldm_privhdr_parse(cp, &hdr, buf);
481 		if (error != 0) {
482 			LDM_DEBUG(1, "%s: failed to parse private "
483 			    "header %d", pp->name, i);
484 			LDM_DUMP(buf, pp->sectorsize);
485 			g_free(buf);
486 			continue;
487 		}
488 		g_free(buf);
489 		if (hdr.start > last ||
490 		    hdr.start + hdr.size - 1 > last ||
491 		    (hdr.start + hdr.size - 1 > hdr.db_offset && !is_gpt) ||
492 		    hdr.db_size != LDM_DB_SIZE ||
493 		    hdr.db_offset + LDM_DB_SIZE - 1 > last ||
494 		    hdr.th_offset[0] >= LDM_DB_SIZE ||
495 		    hdr.th_offset[1] >= LDM_DB_SIZE ||
496 		    hdr.conf_size + hdr.log_size >= LDM_DB_SIZE) {
497 			LDM_DEBUG(1, "%s: invalid values in the "
498 			    "private header %d", pp->name, i);
499 			LDM_DEBUG(2, "%s: start: %jd, size: %jd, "
500 			    "db_offset: %jd, db_size: %jd, th_offset0: %jd, "
501 			    "th_offset1: %jd, conf_size: %jd, log_size: %jd, "
502 			    "last: %jd", pp->name, hdr.start, hdr.size,
503 			    hdr.db_offset, hdr.db_size, hdr.th_offset[0],
504 			    hdr.th_offset[1], hdr.conf_size, hdr.log_size,
505 			    last);
506 			continue;
507 		}
508 		if (found != 0 && memcmp(&db->ph, &hdr, sizeof(hdr)) != 0) {
509 			LDM_DEBUG(0, "%s: private headers are not equal",
510 			    pp->name);
511 			if (i > 1) {
512 				/*
513 				 * We have different headers in the LDM.
514 				 * We can not trust this metadata.
515 				 */
516 				LDM_DEBUG(0, "%s: refuse LDM metadata",
517 				    pp->name);
518 				return (EINVAL);
519 			}
520 			/*
521 			 * We already have read primary private header
522 			 * and it differs from this backup one.
523 			 * Prefer the backup header and save it.
524 			 */
525 			found = 0;
526 		}
527 		if (found == 0)
528 			memcpy(&db->ph, &hdr, sizeof(hdr));
529 		found = 1;
530 	}
531 	if (found == 0) {
532 		LDM_DEBUG(1, "%s: valid LDM private header not found",
533 		    pp->name);
534 		return (ENXIO);
535 	}
536 	return (0);
537 }
538 
539 static int
540 ldm_gpt_check(struct ldm_db *db, struct g_consumer *cp)
541 {
542 	struct g_part_table *gpt;
543 	struct g_part_entry *e;
544 	struct g_consumer *cp2;
545 	int error;
546 
547 	cp2 = LIST_NEXT(cp, consumer);
548 	g_topology_lock();
549 	gpt = cp->provider->geom->softc;
550 	error = 0;
551 	LIST_FOREACH(e, &gpt->gpt_entry, gpe_entry) {
552 		if (cp->provider == e->gpe_pp) {
553 			/* ms-ldm-metadata partition */
554 			if (e->gpe_start != db->ph.db_offset ||
555 			    e->gpe_end != db->ph.db_offset + LDM_DB_SIZE - 1)
556 				error++;
557 		} else if (cp2->provider == e->gpe_pp) {
558 			/* ms-ldm-data partition */
559 			if (e->gpe_start != db->ph.start ||
560 			    e->gpe_end != db->ph.start + db->ph.size - 1)
561 				error++;
562 		}
563 		if (error != 0) {
564 			LDM_DEBUG(0, "%s: GPT partition %d boundaries "
565 			    "do not match with the LDM metadata",
566 			    e->gpe_pp->name, e->gpe_index);
567 			error = ENXIO;
568 			break;
569 		}
570 	}
571 	g_topology_unlock();
572 	return (error);
573 }
574 
575 static int
576 ldm_tochdr_check(struct ldm_db *db, struct g_consumer *cp)
577 {
578 	struct g_provider *pp;
579 	struct ldm_tochdr hdr;
580 	uint64_t offset, conf_size, log_size;
581 	int error, found, i;
582 	u_char *buf;
583 
584 	pp = cp->provider;
585 	for (i = 0, found = 0; i < LDM_TH_COUNT; i++) {
586 		offset = db->ph.db_offset + db->ph.th_offset[i];
587 		buf = g_read_data(cp,
588 		    offset * pp->sectorsize, pp->sectorsize, &error);
589 		if (buf == NULL) {
590 			LDM_DEBUG(1, "%s: failed to read TOC header "
591 			    "at LBA %ju", pp->name, (uintmax_t)offset);
592 			continue;
593 		}
594 		if (memcmp(buf, LDM_TH_SIGN, strlen(LDM_TH_SIGN)) != 0 ||
595 		    memcmp(buf + LDM_TH_NAME1_OFF, LDM_TH_NAME1,
596 		    strlen(LDM_TH_NAME1)) != 0 ||
597 		    memcmp(buf + LDM_TH_NAME2_OFF, LDM_TH_NAME2,
598 		    strlen(LDM_TH_NAME2)) != 0) {
599 			LDM_DEBUG(1, "%s: failed to parse TOC header "
600 			    "at LBA %ju", pp->name, (uintmax_t)offset);
601 			LDM_DUMP(buf, pp->sectorsize);
602 			g_free(buf);
603 			continue;
604 		}
605 		hdr.conf_offset = be64dec(buf + LDM_TH_CONF_OFF);
606 		hdr.log_offset = be64dec(buf + LDM_TH_LOG_OFF);
607 		conf_size = be64dec(buf + LDM_TH_CONFSIZE_OFF);
608 		log_size = be64dec(buf + LDM_TH_LOGSIZE_OFF);
609 		if (conf_size != db->ph.conf_size ||
610 		    hdr.conf_offset + conf_size >= LDM_DB_SIZE ||
611 		    log_size != db->ph.log_size ||
612 		    hdr.log_offset + log_size >= LDM_DB_SIZE) {
613 			LDM_DEBUG(1, "%s: invalid values in the "
614 			    "TOC header at LBA %ju", pp->name,
615 			    (uintmax_t)offset);
616 			LDM_DUMP(buf, pp->sectorsize);
617 			g_free(buf);
618 			continue;
619 		}
620 		g_free(buf);
621 		if (found == 0)
622 			memcpy(&db->th, &hdr, sizeof(hdr));
623 		found = 1;
624 	}
625 	if (found == 0) {
626 		LDM_DEBUG(0, "%s: valid LDM TOC header not found.",
627 		    pp->name);
628 		return (ENXIO);
629 	}
630 	return (0);
631 }
632 
633 static int
634 ldm_vmdbhdr_check(struct ldm_db *db, struct g_consumer *cp)
635 {
636 	struct g_provider *pp;
637 	struct uuid dg_guid;
638 	uint64_t offset;
639 	uint32_t version;
640 	int error;
641 	u_char *buf;
642 
643 	pp = cp->provider;
644 	offset = db->ph.db_offset + db->th.conf_offset;
645 	buf = g_read_data(cp, offset * pp->sectorsize, pp->sectorsize,
646 	    &error);
647 	if (buf == NULL) {
648 		LDM_DEBUG(0, "%s: failed to read VMDB header at "
649 		    "LBA %ju", pp->name, (uintmax_t)offset);
650 		return (error);
651 	}
652 	if (memcmp(buf, LDM_VMDB_SIGN, strlen(LDM_VMDB_SIGN)) != 0) {
653 		g_free(buf);
654 		LDM_DEBUG(0, "%s: failed to parse VMDB header at "
655 		    "LBA %ju", pp->name, (uintmax_t)offset);
656 		return (ENXIO);
657 	}
658 	/* Check version. */
659 	version = be32dec(buf + LDM_DB_VERSION_OFF);
660 	if (version != 0x4000A) {
661 		g_free(buf);
662 		LDM_DEBUG(0, "%s: unsupported VMDB version %u.%u",
663 		    pp->name, version >> 16, version & 0xFFFF);
664 		return (ENXIO);
665 	}
666 	/*
667 	 * Check VMDB update status:
668 	 *	1 - in a consistent state;
669 	 *	2 - in a creation phase;
670 	 *	3 - in a deletion phase;
671 	 */
672 	if (be16dec(buf + LDM_DB_STATUS_OFF) != 1) {
673 		g_free(buf);
674 		LDM_DEBUG(0, "%s: VMDB is not in a consistent state",
675 		    pp->name);
676 		return (ENXIO);
677 	}
678 	db->dh.last_seq = be32dec(buf + LDM_DB_LASTSEQ_OFF);
679 	db->dh.size = be32dec(buf + LDM_DB_SIZE_OFF);
680 	error = parse_uuid(buf + LDM_DB_DGGUID_OFF, &dg_guid);
681 	/* Compare disk group name and guid from VMDB and private headers */
682 	if (error != 0 || db->dh.size == 0 ||
683 	    pp->sectorsize % db->dh.size != 0 ||
684 	    strncmp(buf + LDM_DB_DGNAME_OFF, db->ph.dg_name, 31) != 0 ||
685 	    memcmp(&dg_guid, &db->ph.dg_guid, sizeof(dg_guid)) != 0 ||
686 	    db->dh.size * db->dh.last_seq >
687 	    db->ph.conf_size * pp->sectorsize) {
688 		LDM_DEBUG(0, "%s: invalid values in the VMDB header",
689 		    pp->name);
690 		LDM_DUMP(buf, pp->sectorsize);
691 		g_free(buf);
692 		return (EINVAL);
693 	}
694 	g_free(buf);
695 	return (0);
696 }
697 
698 static int
699 ldm_xvblk_handle(struct ldm_db *db, struct ldm_vblkhdr *vh, const u_char *p)
700 {
701 	struct ldm_xvblk *blk;
702 	size_t size;
703 
704 	size = db->dh.size - 16;
705 	LIST_FOREACH(blk, &db->xvblks, entry)
706 		if (blk->group == vh->group)
707 			break;
708 	if (blk == NULL) {
709 		blk = g_malloc(sizeof(*blk), M_WAITOK | M_ZERO);
710 		blk->group = vh->group;
711 		blk->size = size * vh->count + 16;
712 		blk->data = g_malloc(blk->size, M_WAITOK | M_ZERO);
713 		blk->map = 0xFF << vh->count;
714 		LIST_INSERT_HEAD(&db->xvblks, blk, entry);
715 	}
716 	if ((blk->map & (1 << vh->index)) != 0) {
717 		/* Block with given index has been already saved. */
718 		return (EINVAL);
719 	}
720 	/* Copy the data block to the place related to index. */
721 	memcpy(blk->data + size * vh->index + 16, p + 16, size);
722 	blk->map |= 1 << vh->index;
723 	return (0);
724 }
725 
726 /* Read the variable-width numeric field and return new offset */
727 static int
728 ldm_vnum_get(const u_char *buf, int offset, uint64_t *result, size_t range)
729 {
730 	uint64_t num;
731 	uint8_t len;
732 
733 	len = buf[offset++];
734 	if (len > sizeof(uint64_t) || len + offset >= range)
735 		return (-1);
736 	for (num = 0; len > 0; len--)
737 		num = (num << 8) | buf[offset++];
738 	*result = num;
739 	return (offset);
740 }
741 
742 /* Read the variable-width string and return new offset */
743 static int
744 ldm_vstr_get(const u_char *buf, int offset, u_char *result,
745     size_t maxlen, size_t range)
746 {
747 	uint8_t len;
748 
749 	len = buf[offset++];
750 	if (len >= maxlen || len + offset >= range)
751 		return (-1);
752 	memcpy(result, buf + offset, len);
753 	result[len] = '\0';
754 	return (offset + len);
755 }
756 
757 /* Just skip the variable-width variable and return new offset */
758 static int
759 ldm_vparm_skip(const u_char *buf, int offset, size_t range)
760 {
761 	uint8_t len;
762 
763 	len = buf[offset++];
764 	if (offset + len >= range)
765 		return (-1);
766 
767 	return (offset + len);
768 }
769 
770 static int
771 ldm_vblk_handle(struct ldm_db *db, const u_char *p, size_t size)
772 {
773 	struct ldm_vblk *blk;
774 	struct ldm_volume *volume, *last;
775 	const char *errstr;
776 	u_char vstr[64];
777 	int error, offset;
778 
779 	blk = g_malloc(sizeof(*blk), M_WAITOK | M_ZERO);
780 	blk->type = p[LDM_VBLK_TYPE_OFF];
781 	offset = ldm_vnum_get(p, LDM_VBLK_OID_OFF, &blk->u.id, size);
782 	if (offset < 0) {
783 		errstr = "object id";
784 		goto fail;
785 	}
786 	offset = ldm_vstr_get(p, offset, vstr, sizeof(vstr), size);
787 	if (offset < 0) {
788 		errstr = "object name";
789 		goto fail;
790 	}
791 	switch (blk->type) {
792 	/*
793 	 * Component VBLK fields:
794 	 * Offset	Size	Description
795 	 * ------------+-------+------------------------
796 	 *  0x18+	PS	volume state
797 	 *  0x18+5	PN	component children count
798 	 *  0x1D+16	PN	parent's volume object id
799 	 *  0x2D+1	PN	stripe size
800 	 */
801 	case LDM_VBLK_T_COMPONENT:
802 		offset = ldm_vparm_skip(p, offset, size);
803 		if (offset < 0) {
804 			errstr = "volume state";
805 			goto fail;
806 		}
807 		offset = ldm_vparm_skip(p, offset + 5, size);
808 		if (offset < 0) {
809 			errstr = "children count";
810 			goto fail;
811 		}
812 		offset = ldm_vnum_get(p, offset + 16,
813 		    &blk->u.comp.vol_id, size);
814 		if (offset < 0) {
815 			errstr = "volume id";
816 			goto fail;
817 		}
818 		break;
819 	/*
820 	 * Partition VBLK fields:
821 	 * Offset	Size	Description
822 	 * ------------+-------+------------------------
823 	 *  0x18+12	8	partition start offset
824 	 *  0x18+20	8	volume offset
825 	 *  0x18+28	PN	partition size
826 	 *  0x34+	PN	parent's component object id
827 	 *  0x34+	PN	disk's object id
828 	 */
829 	case LDM_VBLK_T_PARTITION:
830 		if (offset + 28 >= size) {
831 			errstr = "too small buffer";
832 			goto fail;
833 		}
834 		blk->u.part.start = be64dec(p + offset + 12);
835 		blk->u.part.offset = be64dec(p + offset + 20);
836 		offset = ldm_vnum_get(p, offset + 28, &blk->u.part.size, size);
837 		if (offset < 0) {
838 			errstr = "partition size";
839 			goto fail;
840 		}
841 		offset = ldm_vnum_get(p, offset, &blk->u.part.comp_id, size);
842 		if (offset < 0) {
843 			errstr = "component id";
844 			goto fail;
845 		}
846 		offset = ldm_vnum_get(p, offset, &blk->u.part.disk_id, size);
847 		if (offset < 0) {
848 			errstr = "disk id";
849 			goto fail;
850 		}
851 		break;
852 	/*
853 	 * Disk VBLK fields:
854 	 * Offset	Size	Description
855 	 * ------------+-------+------------------------
856 	 *  0x18+	PS	disk GUID
857 	 */
858 	case LDM_VBLK_T_DISK:
859 		errstr = "disk guid";
860 		offset = ldm_vstr_get(p, offset, vstr, sizeof(vstr), size);
861 		if (offset < 0)
862 			goto fail;
863 		error = parse_uuid(vstr, &blk->u.disk.guid);
864 		if (error != 0)
865 			goto fail;
866 		LIST_INSERT_HEAD(&db->disks, &blk->u.disk, entry);
867 		break;
868 	/*
869 	 * Disk group VBLK fields:
870 	 * Offset	Size	Description
871 	 * ------------+-------+------------------------
872 	 *  0x18+	PS	disk group GUID
873 	 */
874 	case LDM_VBLK_T_DISKGROUP:
875 #if 0
876 		strncpy(blk->u.disk_group.name, vstr,
877 		    sizeof(blk->u.disk_group.name));
878 		offset = ldm_vstr_get(p, offset, vstr, sizeof(vstr), size);
879 		if (offset < 0) {
880 			errstr = "disk group guid";
881 			goto fail;
882 		}
883 		error = parse_uuid(name, &blk->u.disk_group.guid);
884 		if (error != 0) {
885 			errstr = "disk group guid";
886 			goto fail;
887 		}
888 		LIST_INSERT_HEAD(&db->groups, &blk->u.disk_group, entry);
889 #endif
890 		break;
891 	/*
892 	 * Disk VBLK fields:
893 	 * Offset	Size	Description
894 	 * ------------+-------+------------------------
895 	 *  0x18+	16	disk GUID
896 	 */
897 	case LDM_VBLK_T_DISK4:
898 		be_uuid_dec(p + offset, &blk->u.disk.guid);
899 		LIST_INSERT_HEAD(&db->disks, &blk->u.disk, entry);
900 		break;
901 	/*
902 	 * Disk group VBLK fields:
903 	 * Offset	Size	Description
904 	 * ------------+-------+------------------------
905 	 *  0x18+	16	disk GUID
906 	 */
907 	case LDM_VBLK_T_DISKGROUP4:
908 #if 0
909 		strncpy(blk->u.disk_group.name, vstr,
910 		    sizeof(blk->u.disk_group.name));
911 		be_uuid_dec(p + offset, &blk->u.disk.guid);
912 		LIST_INSERT_HEAD(&db->groups, &blk->u.disk_group, entry);
913 #endif
914 		break;
915 	/*
916 	 * Volume VBLK fields:
917 	 * Offset	Size	Description
918 	 * ------------+-------+------------------------
919 	 *  0x18+	PS	volume type
920 	 *  0x18+	PS	unknown
921 	 *  0x18+	14(S)	volume state
922 	 *  0x18+16	1	volume number
923 	 *  0x18+21	PN	volume children count
924 	 *  0x2D+16	PN	volume size
925 	 *  0x3D+4	1	partition type
926 	 */
927 	case LDM_VBLK_T_VOLUME:
928 		offset = ldm_vparm_skip(p, offset, size);
929 		if (offset < 0) {
930 			errstr = "volume type";
931 			goto fail;
932 		}
933 		offset = ldm_vparm_skip(p, offset, size);
934 		if (offset < 0) {
935 			errstr = "unknown param";
936 			goto fail;
937 		}
938 		if (offset + 21 >= size) {
939 			errstr = "too small buffer";
940 			goto fail;
941 		}
942 		blk->u.vol.number = p[offset + 16];
943 		offset = ldm_vparm_skip(p, offset + 21, size);
944 		if (offset < 0) {
945 			errstr = "children count";
946 			goto fail;
947 		}
948 		offset = ldm_vnum_get(p, offset + 16, &blk->u.vol.size, size);
949 		if (offset < 0) {
950 			errstr = "volume size";
951 			goto fail;
952 		}
953 		if (offset + 4 >= size) {
954 			errstr = "too small buffer";
955 			goto fail;
956 		}
957 		blk->u.vol.part_type = p[offset + 4];
958 		/* keep volumes ordered by volume number */
959 		last = NULL;
960 		LIST_FOREACH(volume, &db->volumes, entry) {
961 			if (volume->number > blk->u.vol.number)
962 				break;
963 			last = volume;
964 		}
965 		if (last != NULL)
966 			LIST_INSERT_AFTER(last, &blk->u.vol, entry);
967 		else
968 			LIST_INSERT_HEAD(&db->volumes, &blk->u.vol, entry);
969 		break;
970 	default:
971 		LDM_DEBUG(1, "unknown VBLK type 0x%02x\n", blk->type);
972 		LDM_DUMP(p, size);
973 	}
974 	LIST_INSERT_HEAD(&db->vblks, blk, entry);
975 	return (0);
976 fail:
977 	LDM_DEBUG(0, "failed to parse '%s' in VBLK of type 0x%02x\n",
978 	    errstr, blk->type);
979 	LDM_DUMP(p, size);
980 	g_free(blk);
981 	return (EINVAL);
982 }
983 
984 static void
985 ldm_vmdb_free(struct ldm_db *db)
986 {
987 	struct ldm_vblk *vblk;
988 	struct ldm_xvblk *xvblk;
989 
990 	while (!LIST_EMPTY(&db->xvblks)) {
991 		xvblk = LIST_FIRST(&db->xvblks);
992 		LIST_REMOVE(xvblk, entry);
993 		g_free(xvblk->data);
994 		g_free(xvblk);
995 	}
996 	while (!LIST_EMPTY(&db->vblks)) {
997 		vblk = LIST_FIRST(&db->vblks);
998 		LIST_REMOVE(vblk, entry);
999 		g_free(vblk);
1000 	}
1001 }
1002 
1003 static int
1004 ldm_vmdb_parse(struct ldm_db *db, struct g_consumer *cp)
1005 {
1006 	struct g_provider *pp;
1007 	struct ldm_vblk *vblk;
1008 	struct ldm_xvblk *xvblk;
1009 	struct ldm_volume *volume;
1010 	struct ldm_component *comp;
1011 	struct ldm_vblkhdr vh;
1012 	u_char *buf, *p;
1013 	size_t size, n, sectors;
1014 	uint64_t offset;
1015 	int error;
1016 
1017 	pp = cp->provider;
1018 	size = howmany(db->dh.last_seq * db->dh.size, pp->sectorsize);
1019 	size -= 1; /* one sector takes vmdb header */
1020 	for (n = 0; n < size; n += maxphys / pp->sectorsize) {
1021 		offset = db->ph.db_offset + db->th.conf_offset + n + 1;
1022 		sectors = (size - n) > (maxphys / pp->sectorsize) ?
1023 		    maxphys / pp->sectorsize : size - n;
1024 		/* read VBLKs */
1025 		buf = g_read_data(cp, offset * pp->sectorsize,
1026 		    sectors * pp->sectorsize, &error);
1027 		if (buf == NULL) {
1028 			LDM_DEBUG(0, "%s: failed to read VBLK\n",
1029 			    pp->name);
1030 			goto fail;
1031 		}
1032 		for (p = buf; p < buf + sectors * pp->sectorsize;
1033 		    p += db->dh.size) {
1034 			if (memcmp(p, LDM_VBLK_SIGN,
1035 			    strlen(LDM_VBLK_SIGN)) != 0) {
1036 				LDM_DEBUG(0, "%s: no VBLK signature\n",
1037 				    pp->name);
1038 				LDM_DUMP(p, db->dh.size);
1039 				goto fail;
1040 			}
1041 			vh.seq = be32dec(p + LDM_VBLK_SEQ_OFF);
1042 			vh.group = be32dec(p + LDM_VBLK_GROUP_OFF);
1043 			/* skip empty blocks */
1044 			if (vh.seq == 0 || vh.group == 0)
1045 				continue;
1046 			vh.index = be16dec(p + LDM_VBLK_INDEX_OFF);
1047 			vh.count = be16dec(p + LDM_VBLK_COUNT_OFF);
1048 			if (vh.count == 0 || vh.count > 4 ||
1049 			    vh.seq > db->dh.last_seq) {
1050 				LDM_DEBUG(0, "%s: invalid values "
1051 				    "in the VBLK header\n", pp->name);
1052 				LDM_DUMP(p, db->dh.size);
1053 				goto fail;
1054 			}
1055 			if (vh.count > 1) {
1056 				error = ldm_xvblk_handle(db, &vh, p);
1057 				if (error != 0) {
1058 					LDM_DEBUG(0, "%s: xVBLK "
1059 					    "is corrupted\n", pp->name);
1060 					LDM_DUMP(p, db->dh.size);
1061 					goto fail;
1062 				}
1063 				continue;
1064 			}
1065 			if (be16dec(p + 16) != 0)
1066 				LDM_DEBUG(1, "%s: VBLK update"
1067 				    " status is %u\n", pp->name,
1068 				    be16dec(p + 16));
1069 			error = ldm_vblk_handle(db, p, db->dh.size);
1070 			if (error != 0)
1071 				goto fail;
1072 		}
1073 		g_free(buf);
1074 		buf = NULL;
1075 	}
1076 	/* Parse xVBLKs */
1077 	while (!LIST_EMPTY(&db->xvblks)) {
1078 		xvblk = LIST_FIRST(&db->xvblks);
1079 		if (xvblk->map == 0xFF) {
1080 			error = ldm_vblk_handle(db, xvblk->data, xvblk->size);
1081 			if (error != 0)
1082 				goto fail;
1083 		} else {
1084 			LDM_DEBUG(0, "%s: incomplete or corrupt "
1085 			    "xVBLK found\n", pp->name);
1086 			goto fail;
1087 		}
1088 		LIST_REMOVE(xvblk, entry);
1089 		g_free(xvblk->data);
1090 		g_free(xvblk);
1091 	}
1092 	/* construct all VBLKs relations */
1093 	LIST_FOREACH(volume, &db->volumes, entry) {
1094 		LIST_FOREACH(vblk, &db->vblks, entry)
1095 			if (vblk->type == LDM_VBLK_T_COMPONENT &&
1096 			    vblk->u.comp.vol_id == volume->id) {
1097 				LIST_INSERT_HEAD(&volume->components,
1098 				    &vblk->u.comp, entry);
1099 				volume->count++;
1100 			}
1101 		LIST_FOREACH(comp, &volume->components, entry)
1102 			LIST_FOREACH(vblk, &db->vblks, entry)
1103 				if (vblk->type == LDM_VBLK_T_PARTITION &&
1104 				    vblk->u.part.comp_id == comp->id) {
1105 					LIST_INSERT_HEAD(&comp->partitions,
1106 					    &vblk->u.part, entry);
1107 					comp->count++;
1108 				}
1109 	}
1110 	return (0);
1111 fail:
1112 	ldm_vmdb_free(db);
1113 	g_free(buf);
1114 	return (ENXIO);
1115 }
1116 
1117 static int
1118 g_part_ldm_add(struct g_part_table *basetable, struct g_part_entry *baseentry,
1119     struct g_part_parms *gpp)
1120 {
1121 
1122 	return (ENOSYS);
1123 }
1124 
1125 static int
1126 g_part_ldm_bootcode(struct g_part_table *basetable, struct g_part_parms *gpp)
1127 {
1128 
1129 	return (ENOSYS);
1130 }
1131 
1132 static int
1133 g_part_ldm_create(struct g_part_table *basetable, struct g_part_parms *gpp)
1134 {
1135 
1136 	return (ENOSYS);
1137 }
1138 
1139 static int
1140 g_part_ldm_destroy(struct g_part_table *basetable, struct g_part_parms *gpp)
1141 {
1142 	struct g_part_ldm_table *table;
1143 	struct g_provider *pp;
1144 
1145 	table = (struct g_part_ldm_table *)basetable;
1146 	/*
1147 	 * To destroy LDM on a disk partitioned with GPT we should delete
1148 	 * ms-ldm-metadata partition, but we can't do this via standard
1149 	 * GEOM_PART method.
1150 	 */
1151 	if (table->is_gpt)
1152 		return (ENOSYS);
1153 	pp = LIST_FIRST(&basetable->gpt_gp->consumer)->provider;
1154 	/*
1155 	 * To destroy LDM we should wipe MBR, first private header and
1156 	 * backup private headers.
1157 	 */
1158 	basetable->gpt_smhead = (1 << ldm_ph_off[0]) | 1;
1159 	/*
1160 	 * Don't touch last backup private header when LDM database is
1161 	 * not located in the last 1MByte area.
1162 	 * XXX: can't remove all blocks.
1163 	 */
1164 	if (table->db_offset + LDM_DB_SIZE ==
1165 	    pp->mediasize / pp->sectorsize)
1166 		basetable->gpt_smtail = 1;
1167 	return (0);
1168 }
1169 
1170 static void
1171 g_part_ldm_dumpconf(struct g_part_table *basetable,
1172     struct g_part_entry *baseentry, struct sbuf *sb, const char *indent)
1173 {
1174 	struct g_part_ldm_entry *entry;
1175 
1176 	entry = (struct g_part_ldm_entry *)baseentry;
1177 	if (indent == NULL) {
1178 		/* conftxt: libdisk compatibility */
1179 		sbuf_printf(sb, " xs LDM xt %u", entry->type);
1180 	} else if (entry != NULL) {
1181 		/* confxml: partition entry information */
1182 		sbuf_printf(sb, "%s<rawtype>%u</rawtype>\n", indent,
1183 		    entry->type);
1184 	} else {
1185 		/* confxml: scheme information */
1186 	}
1187 }
1188 
1189 static int
1190 g_part_ldm_dumpto(struct g_part_table *table, struct g_part_entry *baseentry)
1191 {
1192 
1193 	return (0);
1194 }
1195 
1196 static int
1197 g_part_ldm_modify(struct g_part_table *basetable,
1198     struct g_part_entry *baseentry, struct g_part_parms *gpp)
1199 {
1200 
1201 	return (ENOSYS);
1202 }
1203 
1204 static const char *
1205 g_part_ldm_name(struct g_part_table *table, struct g_part_entry *baseentry,
1206     char *buf, size_t bufsz)
1207 {
1208 
1209 	snprintf(buf, bufsz, "s%d", baseentry->gpe_index);
1210 	return (buf);
1211 }
1212 
1213 static int
1214 ldm_gpt_probe(struct g_part_table *basetable, struct g_consumer *cp)
1215 {
1216 	struct g_part_ldm_table *table;
1217 	struct g_part_table *gpt;
1218 	struct g_part_entry *entry;
1219 	struct g_consumer *cp2;
1220 	struct gpt_ent *part;
1221 	u_char *buf;
1222 	int error;
1223 
1224 	/*
1225 	 * XXX: We use some knowledge about GEOM_PART_GPT internal
1226 	 * structures, but it is easier than parse GPT by himself.
1227 	 */
1228 	g_topology_lock();
1229 	gpt = cp->provider->geom->softc;
1230 	LIST_FOREACH(entry, &gpt->gpt_entry, gpe_entry) {
1231 		part = (struct gpt_ent *)(entry + 1);
1232 		/* Search ms-ldm-metadata partition */
1233 		if (memcmp(&part->ent_type,
1234 		    &gpt_uuid_ms_ldm_metadata, sizeof(struct uuid)) != 0 ||
1235 		    entry->gpe_end - entry->gpe_start < LDM_DB_SIZE - 1)
1236 			continue;
1237 
1238 		/* Create new consumer and attach it to metadata partition */
1239 		cp2 = g_new_consumer(cp->geom);
1240 		error = g_attach(cp2, entry->gpe_pp);
1241 		if (error != 0) {
1242 			g_destroy_consumer(cp2);
1243 			g_topology_unlock();
1244 			return (ENXIO);
1245 		}
1246 		error = g_access(cp2, 1, 0, 0);
1247 		if (error != 0) {
1248 			g_detach(cp2);
1249 			g_destroy_consumer(cp2);
1250 			g_topology_unlock();
1251 			return (ENXIO);
1252 		}
1253 		g_topology_unlock();
1254 
1255 		LDM_DEBUG(2, "%s: LDM metadata partition %s found in the GPT",
1256 		    cp->provider->name, cp2->provider->name);
1257 		/* Read the LDM private header */
1258 		buf = ldm_privhdr_read(cp2,
1259 		    ldm_ph_off[LDM_PH_GPTINDEX] * cp2->provider->sectorsize,
1260 		    &error);
1261 		if (buf != NULL) {
1262 			table = (struct g_part_ldm_table *)basetable;
1263 			table->is_gpt = 1;
1264 			g_free(buf);
1265 			return (G_PART_PROBE_PRI_HIGH);
1266 		}
1267 
1268 		/* second consumer is no longer needed. */
1269 		g_topology_lock();
1270 		g_access(cp2, -1, 0, 0);
1271 		g_detach(cp2);
1272 		g_destroy_consumer(cp2);
1273 		break;
1274 	}
1275 	g_topology_unlock();
1276 	return (ENXIO);
1277 }
1278 
1279 static int
1280 g_part_ldm_probe(struct g_part_table *basetable, struct g_consumer *cp)
1281 {
1282 	struct g_provider *pp;
1283 	u_char *buf, type[64];
1284 	int error, idx;
1285 
1286 	pp = cp->provider;
1287 	if (pp->sectorsize != 512)
1288 		return (ENXIO);
1289 
1290 	error = g_getattr("PART::scheme", cp, &type);
1291 	if (error == 0 && strcmp(type, "GPT") == 0) {
1292 		if (g_getattr("PART::type", cp, &type) != 0 ||
1293 		    strcmp(type, "ms-ldm-data") != 0)
1294 			return (ENXIO);
1295 		error = ldm_gpt_probe(basetable, cp);
1296 		return (error);
1297 	}
1298 
1299 	if (basetable->gpt_depth != 0)
1300 		return (ENXIO);
1301 
1302 	/* LDM has 1M metadata area */
1303 	if (pp->mediasize <= 1024 * 1024)
1304 		return (ENOSPC);
1305 
1306 	/* Check that there's a MBR */
1307 	buf = g_read_data(cp, 0, pp->sectorsize, &error);
1308 	if (buf == NULL)
1309 		return (error);
1310 
1311 	if (le16dec(buf + DOSMAGICOFFSET) != DOSMAGIC) {
1312 		g_free(buf);
1313 		return (ENXIO);
1314 	}
1315 	error = ENXIO;
1316 	/* Check that we have LDM partitions in the MBR */
1317 	for (idx = 0; idx < NDOSPART && error != 0; idx++) {
1318 		if (buf[DOSPARTOFF + idx * DOSPARTSIZE + 4] == DOSPTYP_LDM)
1319 			error = 0;
1320 	}
1321 	g_free(buf);
1322 	if (error == 0) {
1323 		LDM_DEBUG(2, "%s: LDM data partitions found in MBR",
1324 		    pp->name);
1325 		/* Read the LDM private header */
1326 		buf = ldm_privhdr_read(cp,
1327 		    ldm_ph_off[LDM_PH_MBRINDEX] * pp->sectorsize, &error);
1328 		if (buf == NULL)
1329 			return (error);
1330 		g_free(buf);
1331 		return (G_PART_PROBE_PRI_HIGH);
1332 	}
1333 	return (error);
1334 }
1335 
1336 static int
1337 g_part_ldm_read(struct g_part_table *basetable, struct g_consumer *cp)
1338 {
1339 	struct g_part_ldm_table *table;
1340 	struct g_part_ldm_entry *entry;
1341 	struct g_consumer *cp2;
1342 	struct ldm_component *comp;
1343 	struct ldm_partition *part;
1344 	struct ldm_volume *vol;
1345 	struct ldm_disk *disk;
1346 	struct ldm_db db;
1347 	int error, index, skipped;
1348 
1349 	table = (struct g_part_ldm_table *)basetable;
1350 	memset(&db, 0, sizeof(db));
1351 	cp2 = cp;					/* ms-ldm-data */
1352 	if (table->is_gpt)
1353 		cp = LIST_FIRST(&cp->geom->consumer);	/* ms-ldm-metadata */
1354 	/* Read and parse LDM private headers. */
1355 	error = ldm_privhdr_check(&db, cp, table->is_gpt);
1356 	if (error != 0)
1357 		goto gpt_cleanup;
1358 	basetable->gpt_first = table->is_gpt ? 0: db.ph.start;
1359 	basetable->gpt_last = basetable->gpt_first + db.ph.size - 1;
1360 	table->db_offset = db.ph.db_offset;
1361 	/* Make additional checks for GPT */
1362 	if (table->is_gpt) {
1363 		error = ldm_gpt_check(&db, cp);
1364 		if (error != 0)
1365 			goto gpt_cleanup;
1366 		/*
1367 		 * Now we should reset database offset to zero, because our
1368 		 * consumer cp is attached to the ms-ldm-metadata partition
1369 		 * and we don't need add db_offset to read from it.
1370 		 */
1371 		db.ph.db_offset = 0;
1372 	}
1373 	/* Read and parse LDM TOC headers. */
1374 	error = ldm_tochdr_check(&db, cp);
1375 	if (error != 0)
1376 		goto gpt_cleanup;
1377 	/* Read and parse LDM VMDB header. */
1378 	error = ldm_vmdbhdr_check(&db, cp);
1379 	if (error != 0)
1380 		goto gpt_cleanup;
1381 	error = ldm_vmdb_parse(&db, cp);
1382 	/*
1383 	 * For the GPT case we must detach and destroy
1384 	 * second consumer before return.
1385 	 */
1386 gpt_cleanup:
1387 	if (table->is_gpt) {
1388 		g_topology_lock();
1389 		g_access(cp, -1, 0, 0);
1390 		g_detach(cp);
1391 		g_destroy_consumer(cp);
1392 		g_topology_unlock();
1393 		cp = cp2;
1394 	}
1395 	if (error != 0)
1396 		return (error);
1397 	/* Search current disk in the disk list. */
1398 	LIST_FOREACH(disk, &db.disks, entry)
1399 	    if (memcmp(&disk->guid, &db.ph.disk_guid,
1400 		sizeof(struct uuid)) == 0)
1401 		    break;
1402 	if (disk == NULL) {
1403 		LDM_DEBUG(1, "%s: no LDM volumes on this disk",
1404 		    cp->provider->name);
1405 		ldm_vmdb_free(&db);
1406 		return (ENXIO);
1407 	}
1408 	index = 1;
1409 	LIST_FOREACH(vol, &db.volumes, entry) {
1410 		LIST_FOREACH(comp, &vol->components, entry) {
1411 			/* Skip volumes from different disks. */
1412 			part = LIST_FIRST(&comp->partitions);
1413 			if (part->disk_id != disk->id)
1414 				continue;
1415 			skipped = 0;
1416 			/* We don't support spanned and striped volumes. */
1417 			if (comp->count > 1 || part->offset != 0) {
1418 				LDM_DEBUG(1, "%s: LDM volume component "
1419 				    "%ju has %u partitions. Skipped",
1420 				    cp->provider->name, (uintmax_t)comp->id,
1421 				    comp->count);
1422 				skipped = 1;
1423 			}
1424 			/*
1425 			 * Allow mirrored volumes only when they are explicitly
1426 			 * allowed with kern.geom.part.ldm.show_mirrors=1.
1427 			 */
1428 			if (vol->count > 1 && show_mirrors == 0) {
1429 				LDM_DEBUG(1, "%s: LDM volume %ju has %u "
1430 				    "components. Skipped",
1431 				    cp->provider->name, (uintmax_t)vol->id,
1432 				    vol->count);
1433 				skipped = 1;
1434 			}
1435 			entry = (struct g_part_ldm_entry *)g_part_new_entry(
1436 			    basetable, index++,
1437 			    basetable->gpt_first + part->start,
1438 			    basetable->gpt_first + part->start +
1439 			    part->size - 1);
1440 			/*
1441 			 * Mark skipped partition as ms-ldm-data partition.
1442 			 * We do not support them, but it is better to show
1443 			 * that we have something there, than just show
1444 			 * free space.
1445 			 */
1446 			if (skipped == 0)
1447 				entry->type = vol->part_type;
1448 			else
1449 				entry->type = DOSPTYP_LDM;
1450 			LDM_DEBUG(1, "%s: new volume id: %ju, start: %ju,"
1451 			    " end: %ju, type: 0x%02x\n", cp->provider->name,
1452 			    (uintmax_t)part->id,(uintmax_t)part->start +
1453 			    basetable->gpt_first, (uintmax_t)part->start +
1454 			    part->size + basetable->gpt_first - 1,
1455 			    vol->part_type);
1456 		}
1457 	}
1458 	ldm_vmdb_free(&db);
1459 	return (error);
1460 }
1461 
1462 static const char *
1463 g_part_ldm_type(struct g_part_table *basetable, struct g_part_entry *baseentry,
1464     char *buf, size_t bufsz)
1465 {
1466 	struct g_part_ldm_entry *entry;
1467 	int i;
1468 
1469 	entry = (struct g_part_ldm_entry *)baseentry;
1470 	for (i = 0; i < nitems(ldm_alias_match); i++) {
1471 		if (ldm_alias_match[i].typ == entry->type)
1472 			return (g_part_alias_name(ldm_alias_match[i].alias));
1473 	}
1474 	snprintf(buf, bufsz, "!%d", entry->type);
1475 	return (buf);
1476 }
1477 
1478 static int
1479 g_part_ldm_write(struct g_part_table *basetable, struct g_consumer *cp)
1480 {
1481 
1482 	return (ENOSYS);
1483 }
1484