xref: /freebsd/sys/geom/part/g_part_ldm.c (revision 4f4b548b1d21ae16955b09f9b717524167a7d97e)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/bio.h>
34 #include <sys/diskmbr.h>
35 #include <sys/endian.h>
36 #include <sys/gpt.h>
37 #include <sys/kernel.h>
38 #include <sys/kobj.h>
39 #include <sys/limits.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/mutex.h>
43 #include <sys/queue.h>
44 #include <sys/sbuf.h>
45 #include <sys/systm.h>
46 #include <sys/sysctl.h>
47 #include <sys/uuid.h>
48 #include <geom/geom.h>
49 #include <geom/part/g_part.h>
50 
51 #include "g_part_if.h"
52 
53 FEATURE(geom_part_ldm, "GEOM partitioning class for LDM support");
54 
55 SYSCTL_DECL(_kern_geom_part);
56 static SYSCTL_NODE(_kern_geom_part, OID_AUTO, ldm, CTLFLAG_RW, 0,
57     "GEOM_PART_LDM Logical Disk Manager");
58 
59 static u_int ldm_debug = 0;
60 SYSCTL_UINT(_kern_geom_part_ldm, OID_AUTO, debug,
61     CTLFLAG_RWTUN, &ldm_debug, 0, "Debug level");
62 
63 /*
64  * This allows access to mirrored LDM volumes. Since we do not
65  * doing mirroring here, it is not enabled by default.
66  */
67 static u_int show_mirrors = 0;
68 SYSCTL_UINT(_kern_geom_part_ldm, OID_AUTO, show_mirrors,
69     CTLFLAG_RWTUN, &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 
343 static kobj_method_t g_part_ldm_methods[] = {
344 	KOBJMETHOD(g_part_add,		g_part_ldm_add),
345 	KOBJMETHOD(g_part_bootcode,	g_part_ldm_bootcode),
346 	KOBJMETHOD(g_part_create,	g_part_ldm_create),
347 	KOBJMETHOD(g_part_destroy,	g_part_ldm_destroy),
348 	KOBJMETHOD(g_part_dumpconf,	g_part_ldm_dumpconf),
349 	KOBJMETHOD(g_part_dumpto,	g_part_ldm_dumpto),
350 	KOBJMETHOD(g_part_modify,	g_part_ldm_modify),
351 	KOBJMETHOD(g_part_name,		g_part_ldm_name),
352 	KOBJMETHOD(g_part_probe,	g_part_ldm_probe),
353 	KOBJMETHOD(g_part_read,		g_part_ldm_read),
354 	KOBJMETHOD(g_part_type,		g_part_ldm_type),
355 	KOBJMETHOD(g_part_write,	g_part_ldm_write),
356 	{ 0, 0 }
357 };
358 
359 static struct g_part_scheme g_part_ldm_scheme = {
360 	"LDM",
361 	g_part_ldm_methods,
362 	sizeof(struct g_part_ldm_table),
363 	.gps_entrysz = sizeof(struct g_part_ldm_entry)
364 };
365 G_PART_SCHEME_DECLARE(g_part_ldm);
366 MODULE_VERSION(geom_part_ldm, 0);
367 
368 static struct g_part_ldm_alias {
369 	u_char		typ;
370 	int		alias;
371 } ldm_alias_match[] = {
372 	{ DOSPTYP_386BSD,	G_PART_ALIAS_FREEBSD },
373 	{ DOSPTYP_FAT32,	G_PART_ALIAS_MS_FAT32 },
374 	{ DOSPTYP_FAT32LBA,	G_PART_ALIAS_MS_FAT32LBA },
375 	{ DOSPTYP_LDM,		G_PART_ALIAS_MS_LDM_DATA },
376 	{ DOSPTYP_LINLVM,	G_PART_ALIAS_LINUX_LVM },
377 	{ DOSPTYP_LINRAID,	G_PART_ALIAS_LINUX_RAID },
378 	{ DOSPTYP_LINSWP,	G_PART_ALIAS_LINUX_SWAP },
379 	{ DOSPTYP_LINUX,	G_PART_ALIAS_LINUX_DATA },
380 	{ DOSPTYP_NTFS,		G_PART_ALIAS_MS_NTFS },
381 };
382 
383 static u_char*
384 ldm_privhdr_read(struct g_consumer *cp, uint64_t off, int *error)
385 {
386 	struct g_provider *pp;
387 	u_char *buf;
388 
389 	pp = cp->provider;
390 	buf = g_read_data(cp, off, pp->sectorsize, error);
391 	if (buf == NULL)
392 		return (NULL);
393 
394 	if (memcmp(buf, LDM_PH_SIGN, strlen(LDM_PH_SIGN)) != 0) {
395 		LDM_DEBUG(1, "%s: invalid LDM private header signature",
396 		    pp->name);
397 		g_free(buf);
398 		buf = NULL;
399 		*error = EINVAL;
400 	}
401 	return (buf);
402 }
403 
404 static int
405 ldm_privhdr_parse(struct g_consumer *cp, struct ldm_privhdr *hdr,
406     const u_char *buf)
407 {
408 	uint32_t version;
409 	int error;
410 
411 	memset(hdr, 0, sizeof(*hdr));
412 	version = be32dec(buf + LDM_PH_VERSION_OFF);
413 	if (version != LDM_VERSION_2K &&
414 	    version != LDM_VERSION_VISTA) {
415 		LDM_DEBUG(0, "%s: unsupported LDM version %u.%u",
416 		    cp->provider->name, version >> 16,
417 		    version & 0xFFFF);
418 		return (ENXIO);
419 	}
420 	error = parse_uuid(buf + LDM_PH_DISKGUID_OFF, &hdr->disk_guid);
421 	if (error != 0)
422 		return (error);
423 	error = parse_uuid(buf + LDM_PH_DGGUID_OFF, &hdr->dg_guid);
424 	if (error != 0)
425 		return (error);
426 	strncpy(hdr->dg_name, buf + LDM_PH_DGNAME_OFF, sizeof(hdr->dg_name));
427 	hdr->start = be64dec(buf + LDM_PH_START_OFF);
428 	hdr->size = be64dec(buf + LDM_PH_SIZE_OFF);
429 	hdr->db_offset = be64dec(buf + LDM_PH_DB_OFF);
430 	hdr->db_size = be64dec(buf + LDM_PH_DBSIZE_OFF);
431 	hdr->th_offset[0] = be64dec(buf + LDM_PH_TH1_OFF);
432 	hdr->th_offset[1] = be64dec(buf + LDM_PH_TH2_OFF);
433 	hdr->conf_size = be64dec(buf + LDM_PH_CONFSIZE_OFF);
434 	hdr->log_size = be64dec(buf + LDM_PH_LOGSIZE_OFF);
435 	return (0);
436 }
437 
438 static int
439 ldm_privhdr_check(struct ldm_db *db, struct g_consumer *cp, int is_gpt)
440 {
441 	struct g_consumer *cp2;
442 	struct g_provider *pp;
443 	struct ldm_privhdr hdr;
444 	uint64_t offset, last;
445 	int error, found, i;
446 	u_char *buf;
447 
448 	pp = cp->provider;
449 	if (is_gpt) {
450 		/*
451 		 * The last LBA is used in several checks below, for the
452 		 * GPT case it should be calculated relative to the whole
453 		 * disk.
454 		 */
455 		cp2 = LIST_FIRST(&pp->geom->consumer);
456 		last =
457 		    cp2->provider->mediasize / cp2->provider->sectorsize - 1;
458 	} else
459 		last = pp->mediasize / pp->sectorsize - 1;
460 	for (found = 0, i = is_gpt; i < nitems(ldm_ph_off); i++) {
461 		offset = ldm_ph_off[i];
462 		/*
463 		 * In the GPT case consumer is attached to the LDM metadata
464 		 * partition and we don't need add db_offset.
465 		 */
466 		if (!is_gpt)
467 			offset += db->ph.db_offset;
468 		if (i == LDM_PH_MBRINDEX) {
469 			/*
470 			 * Prepare to errors and setup new base offset
471 			 * to read backup private headers. Assume that LDM
472 			 * database is in the last 1Mbyte area.
473 			 */
474 			db->ph.db_offset = last - LDM_DB_SIZE;
475 		}
476 		buf = ldm_privhdr_read(cp, offset * pp->sectorsize, &error);
477 		if (buf == NULL) {
478 			LDM_DEBUG(1, "%s: failed to read private header "
479 			    "%d at LBA %ju", pp->name, i, (uintmax_t)offset);
480 			continue;
481 		}
482 		error = ldm_privhdr_parse(cp, &hdr, buf);
483 		if (error != 0) {
484 			LDM_DEBUG(1, "%s: failed to parse private "
485 			    "header %d", pp->name, i);
486 			LDM_DUMP(buf, pp->sectorsize);
487 			g_free(buf);
488 			continue;
489 		}
490 		g_free(buf);
491 		if (hdr.start > last ||
492 		    hdr.start + hdr.size - 1 > last ||
493 		    (hdr.start + hdr.size - 1 > hdr.db_offset && !is_gpt) ||
494 		    hdr.db_size != LDM_DB_SIZE ||
495 		    hdr.db_offset + LDM_DB_SIZE - 1 > last ||
496 		    hdr.th_offset[0] >= LDM_DB_SIZE ||
497 		    hdr.th_offset[1] >= LDM_DB_SIZE ||
498 		    hdr.conf_size + hdr.log_size >= LDM_DB_SIZE) {
499 			LDM_DEBUG(1, "%s: invalid values in the "
500 			    "private header %d", pp->name, i);
501 			LDM_DEBUG(2, "%s: start: %jd, size: %jd, "
502 			    "db_offset: %jd, db_size: %jd, th_offset0: %jd, "
503 			    "th_offset1: %jd, conf_size: %jd, log_size: %jd, "
504 			    "last: %jd", pp->name, hdr.start, hdr.size,
505 			    hdr.db_offset, hdr.db_size, hdr.th_offset[0],
506 			    hdr.th_offset[1], hdr.conf_size, hdr.log_size,
507 			    last);
508 			continue;
509 		}
510 		if (found != 0 && memcmp(&db->ph, &hdr, sizeof(hdr)) != 0) {
511 			LDM_DEBUG(0, "%s: private headers are not equal",
512 			    pp->name);
513 			if (i > 1) {
514 				/*
515 				 * We have different headers in the LDM.
516 				 * We can not trust this metadata.
517 				 */
518 				LDM_DEBUG(0, "%s: refuse LDM metadata",
519 				    pp->name);
520 				return (EINVAL);
521 			}
522 			/*
523 			 * We already have read primary private header
524 			 * and it differs from this backup one.
525 			 * Prefer the backup header and save it.
526 			 */
527 			found = 0;
528 		}
529 		if (found == 0)
530 			memcpy(&db->ph, &hdr, sizeof(hdr));
531 		found = 1;
532 	}
533 	if (found == 0) {
534 		LDM_DEBUG(1, "%s: valid LDM private header not found",
535 		    pp->name);
536 		return (ENXIO);
537 	}
538 	return (0);
539 }
540 
541 static int
542 ldm_gpt_check(struct ldm_db *db, struct g_consumer *cp)
543 {
544 	struct g_part_table *gpt;
545 	struct g_part_entry *e;
546 	struct g_consumer *cp2;
547 	int error;
548 
549 	cp2 = LIST_NEXT(cp, consumer);
550 	g_topology_lock();
551 	gpt = cp->provider->geom->softc;
552 	error = 0;
553 	LIST_FOREACH(e, &gpt->gpt_entry, gpe_entry) {
554 		if (cp->provider == e->gpe_pp) {
555 			/* ms-ldm-metadata partition */
556 			if (e->gpe_start != db->ph.db_offset ||
557 			    e->gpe_end != db->ph.db_offset + LDM_DB_SIZE - 1)
558 				error++;
559 		} else if (cp2->provider == e->gpe_pp) {
560 			/* ms-ldm-data partition */
561 			if (e->gpe_start != db->ph.start ||
562 			    e->gpe_end != db->ph.start + db->ph.size - 1)
563 				error++;
564 		}
565 		if (error != 0) {
566 			LDM_DEBUG(0, "%s: GPT partition %d boundaries "
567 			    "do not match with the LDM metadata",
568 			    e->gpe_pp->name, e->gpe_index);
569 			error = ENXIO;
570 			break;
571 		}
572 	}
573 	g_topology_unlock();
574 	return (error);
575 }
576 
577 static int
578 ldm_tochdr_check(struct ldm_db *db, struct g_consumer *cp)
579 {
580 	struct g_provider *pp;
581 	struct ldm_tochdr hdr;
582 	uint64_t offset, conf_size, log_size;
583 	int error, found, i;
584 	u_char *buf;
585 
586 	pp = cp->provider;
587 	for (i = 0, found = 0; i < LDM_TH_COUNT; i++) {
588 		offset = db->ph.db_offset + db->ph.th_offset[i];
589 		buf = g_read_data(cp,
590 		    offset * pp->sectorsize, pp->sectorsize, &error);
591 		if (buf == NULL) {
592 			LDM_DEBUG(1, "%s: failed to read TOC header "
593 			    "at LBA %ju", pp->name, (uintmax_t)offset);
594 			continue;
595 		}
596 		if (memcmp(buf, LDM_TH_SIGN, strlen(LDM_TH_SIGN)) != 0 ||
597 		    memcmp(buf + LDM_TH_NAME1_OFF, LDM_TH_NAME1,
598 		    strlen(LDM_TH_NAME1)) != 0 ||
599 		    memcmp(buf + LDM_TH_NAME2_OFF, LDM_TH_NAME2,
600 		    strlen(LDM_TH_NAME2)) != 0) {
601 			LDM_DEBUG(1, "%s: failed to parse TOC header "
602 			    "at LBA %ju", pp->name, (uintmax_t)offset);
603 			LDM_DUMP(buf, pp->sectorsize);
604 			g_free(buf);
605 			continue;
606 		}
607 		hdr.conf_offset = be64dec(buf + LDM_TH_CONF_OFF);
608 		hdr.log_offset = be64dec(buf + LDM_TH_LOG_OFF);
609 		conf_size = be64dec(buf + LDM_TH_CONFSIZE_OFF);
610 		log_size = be64dec(buf + LDM_TH_LOGSIZE_OFF);
611 		if (conf_size != db->ph.conf_size ||
612 		    hdr.conf_offset + conf_size >= LDM_DB_SIZE ||
613 		    log_size != db->ph.log_size ||
614 		    hdr.log_offset + log_size >= LDM_DB_SIZE) {
615 			LDM_DEBUG(1, "%s: invalid values in the "
616 			    "TOC header at LBA %ju", pp->name,
617 			    (uintmax_t)offset);
618 			LDM_DUMP(buf, pp->sectorsize);
619 			g_free(buf);
620 			continue;
621 		}
622 		g_free(buf);
623 		if (found == 0)
624 			memcpy(&db->th, &hdr, sizeof(hdr));
625 		found = 1;
626 	}
627 	if (found == 0) {
628 		LDM_DEBUG(0, "%s: valid LDM TOC header not found.",
629 		    pp->name);
630 		return (ENXIO);
631 	}
632 	return (0);
633 }
634 
635 static int
636 ldm_vmdbhdr_check(struct ldm_db *db, struct g_consumer *cp)
637 {
638 	struct g_provider *pp;
639 	struct uuid dg_guid;
640 	uint64_t offset;
641 	uint32_t version;
642 	int error;
643 	u_char *buf;
644 
645 	pp = cp->provider;
646 	offset = db->ph.db_offset + db->th.conf_offset;
647 	buf = g_read_data(cp, offset * pp->sectorsize, pp->sectorsize,
648 	    &error);
649 	if (buf == NULL) {
650 		LDM_DEBUG(0, "%s: failed to read VMDB header at "
651 		    "LBA %ju", pp->name, (uintmax_t)offset);
652 		return (error);
653 	}
654 	if (memcmp(buf, LDM_VMDB_SIGN, strlen(LDM_VMDB_SIGN)) != 0) {
655 		g_free(buf);
656 		LDM_DEBUG(0, "%s: failed to parse VMDB header at "
657 		    "LBA %ju", pp->name, (uintmax_t)offset);
658 		return (ENXIO);
659 	}
660 	/* Check version. */
661 	version = be32dec(buf + LDM_DB_VERSION_OFF);
662 	if (version != 0x4000A) {
663 		g_free(buf);
664 		LDM_DEBUG(0, "%s: unsupported VMDB version %u.%u",
665 		    pp->name, version >> 16, version & 0xFFFF);
666 		return (ENXIO);
667 	}
668 	/*
669 	 * Check VMDB update status:
670 	 *	1 - in a consistent state;
671 	 *	2 - in a creation phase;
672 	 *	3 - in a deletion phase;
673 	 */
674 	if (be16dec(buf + LDM_DB_STATUS_OFF) != 1) {
675 		g_free(buf);
676 		LDM_DEBUG(0, "%s: VMDB is not in a consistent state",
677 		    pp->name);
678 		return (ENXIO);
679 	}
680 	db->dh.last_seq = be32dec(buf + LDM_DB_LASTSEQ_OFF);
681 	db->dh.size = be32dec(buf + LDM_DB_SIZE_OFF);
682 	error = parse_uuid(buf + LDM_DB_DGGUID_OFF, &dg_guid);
683 	/* Compare disk group name and guid from VMDB and private headers */
684 	if (error != 0 || db->dh.size == 0 ||
685 	    pp->sectorsize % db->dh.size != 0 ||
686 	    strncmp(buf + LDM_DB_DGNAME_OFF, db->ph.dg_name, 31) != 0 ||
687 	    memcmp(&dg_guid, &db->ph.dg_guid, sizeof(dg_guid)) != 0 ||
688 	    db->dh.size * db->dh.last_seq >
689 	    db->ph.conf_size * pp->sectorsize) {
690 		LDM_DEBUG(0, "%s: invalid values in the VMDB header",
691 		    pp->name);
692 		LDM_DUMP(buf, pp->sectorsize);
693 		g_free(buf);
694 		return (EINVAL);
695 	}
696 	g_free(buf);
697 	return (0);
698 }
699 
700 static int
701 ldm_xvblk_handle(struct ldm_db *db, struct ldm_vblkhdr *vh, const u_char *p)
702 {
703 	struct ldm_xvblk *blk;
704 	size_t size;
705 
706 	size = db->dh.size - 16;
707 	LIST_FOREACH(blk, &db->xvblks, entry)
708 		if (blk->group == vh->group)
709 			break;
710 	if (blk == NULL) {
711 		blk = g_malloc(sizeof(*blk), M_WAITOK | M_ZERO);
712 		blk->group = vh->group;
713 		blk->size = size * vh->count + 16;
714 		blk->data = g_malloc(blk->size, M_WAITOK | M_ZERO);
715 		blk->map = 0xFF << vh->count;
716 		LIST_INSERT_HEAD(&db->xvblks, blk, entry);
717 	}
718 	if ((blk->map & (1 << vh->index)) != 0) {
719 		/* Block with given index has been already saved. */
720 		return (EINVAL);
721 	}
722 	/* Copy the data block to the place related to index. */
723 	memcpy(blk->data + size * vh->index + 16, p + 16, size);
724 	blk->map |= 1 << vh->index;
725 	return (0);
726 }
727 
728 /* Read the variable-width numeric field and return new offset */
729 static int
730 ldm_vnum_get(const u_char *buf, int offset, uint64_t *result, size_t range)
731 {
732 	uint64_t num;
733 	uint8_t len;
734 
735 	len = buf[offset++];
736 	if (len > sizeof(uint64_t) || len + offset >= range)
737 		return (-1);
738 	for (num = 0; len > 0; len--)
739 		num = (num << 8) | buf[offset++];
740 	*result = num;
741 	return (offset);
742 }
743 
744 /* Read the variable-width string and return new offset */
745 static int
746 ldm_vstr_get(const u_char *buf, int offset, u_char *result,
747     size_t maxlen, size_t range)
748 {
749 	uint8_t len;
750 
751 	len = buf[offset++];
752 	if (len >= maxlen || len + offset >= range)
753 		return (-1);
754 	memcpy(result, buf + offset, len);
755 	result[len] = '\0';
756 	return (offset + len);
757 }
758 
759 /* Just skip the variable-width variable and return new offset */
760 static int
761 ldm_vparm_skip(const u_char *buf, int offset, size_t range)
762 {
763 	uint8_t len;
764 
765 	len = buf[offset++];
766 	if (offset + len >= range)
767 		return (-1);
768 
769 	return (offset + len);
770 }
771 
772 static int
773 ldm_vblk_handle(struct ldm_db *db, const u_char *p, size_t size)
774 {
775 	struct ldm_vblk *blk;
776 	struct ldm_volume *volume, *last;
777 	const char *errstr;
778 	u_char vstr[64];
779 	int error, offset;
780 
781 	blk = g_malloc(sizeof(*blk), M_WAITOK | M_ZERO);
782 	blk->type = p[LDM_VBLK_TYPE_OFF];
783 	offset = ldm_vnum_get(p, LDM_VBLK_OID_OFF, &blk->u.id, size);
784 	if (offset < 0) {
785 		errstr = "object id";
786 		goto fail;
787 	}
788 	offset = ldm_vstr_get(p, offset, vstr, sizeof(vstr), size);
789 	if (offset < 0) {
790 		errstr = "object name";
791 		goto fail;
792 	}
793 	switch (blk->type) {
794 	/*
795 	 * Component VBLK fields:
796 	 * Offset	Size	Description
797 	 * ------------+-------+------------------------
798 	 *  0x18+	PS	volume state
799 	 *  0x18+5	PN	component children count
800 	 *  0x1D+16	PN	parent's volume object id
801 	 *  0x2D+1	PN	stripe size
802 	 */
803 	case LDM_VBLK_T_COMPONENT:
804 		offset = ldm_vparm_skip(p, offset, size);
805 		if (offset < 0) {
806 			errstr = "volume state";
807 			goto fail;
808 		}
809 		offset = ldm_vparm_skip(p, offset + 5, size);
810 		if (offset < 0) {
811 			errstr = "children count";
812 			goto fail;
813 		}
814 		offset = ldm_vnum_get(p, offset + 16,
815 		    &blk->u.comp.vol_id, size);
816 		if (offset < 0) {
817 			errstr = "volume id";
818 			goto fail;
819 		}
820 		break;
821 	/*
822 	 * Partition VBLK fields:
823 	 * Offset	Size	Description
824 	 * ------------+-------+------------------------
825 	 *  0x18+12	8	partition start offset
826 	 *  0x18+20	8	volume offset
827 	 *  0x18+28	PN	partition size
828 	 *  0x34+	PN	parent's component object id
829 	 *  0x34+	PN	disk's object id
830 	 */
831 	case LDM_VBLK_T_PARTITION:
832 		if (offset + 28 >= size) {
833 			errstr = "too small buffer";
834 			goto fail;
835 		}
836 		blk->u.part.start = be64dec(p + offset + 12);
837 		blk->u.part.offset = be64dec(p + offset + 20);
838 		offset = ldm_vnum_get(p, offset + 28, &blk->u.part.size, size);
839 		if (offset < 0) {
840 			errstr = "partition size";
841 			goto fail;
842 		}
843 		offset = ldm_vnum_get(p, offset, &blk->u.part.comp_id, size);
844 		if (offset < 0) {
845 			errstr = "component id";
846 			goto fail;
847 		}
848 		offset = ldm_vnum_get(p, offset, &blk->u.part.disk_id, size);
849 		if (offset < 0) {
850 			errstr = "disk id";
851 			goto fail;
852 		}
853 		break;
854 	/*
855 	 * Disk VBLK fields:
856 	 * Offset	Size	Description
857 	 * ------------+-------+------------------------
858 	 *  0x18+	PS	disk GUID
859 	 */
860 	case LDM_VBLK_T_DISK:
861 		errstr = "disk guid";
862 		offset = ldm_vstr_get(p, offset, vstr, sizeof(vstr), size);
863 		if (offset < 0)
864 			goto fail;
865 		error = parse_uuid(vstr, &blk->u.disk.guid);
866 		if (error != 0)
867 			goto fail;
868 		LIST_INSERT_HEAD(&db->disks, &blk->u.disk, entry);
869 		break;
870 	/*
871 	 * Disk group VBLK fields:
872 	 * Offset	Size	Description
873 	 * ------------+-------+------------------------
874 	 *  0x18+	PS	disk group GUID
875 	 */
876 	case LDM_VBLK_T_DISKGROUP:
877 #if 0
878 		strncpy(blk->u.disk_group.name, vstr,
879 		    sizeof(blk->u.disk_group.name));
880 		offset = ldm_vstr_get(p, offset, vstr, sizeof(vstr), size);
881 		if (offset < 0) {
882 			errstr = "disk group guid";
883 			goto fail;
884 		}
885 		error = parse_uuid(name, &blk->u.disk_group.guid);
886 		if (error != 0) {
887 			errstr = "disk group guid";
888 			goto fail;
889 		}
890 		LIST_INSERT_HEAD(&db->groups, &blk->u.disk_group, entry);
891 #endif
892 		break;
893 	/*
894 	 * Disk VBLK fields:
895 	 * Offset	Size	Description
896 	 * ------------+-------+------------------------
897 	 *  0x18+	16	disk GUID
898 	 */
899 	case LDM_VBLK_T_DISK4:
900 		be_uuid_dec(p + offset, &blk->u.disk.guid);
901 		LIST_INSERT_HEAD(&db->disks, &blk->u.disk, entry);
902 		break;
903 	/*
904 	 * Disk group VBLK fields:
905 	 * Offset	Size	Description
906 	 * ------------+-------+------------------------
907 	 *  0x18+	16	disk GUID
908 	 */
909 	case LDM_VBLK_T_DISKGROUP4:
910 #if 0
911 		strncpy(blk->u.disk_group.name, vstr,
912 		    sizeof(blk->u.disk_group.name));
913 		be_uuid_dec(p + offset, &blk->u.disk.guid);
914 		LIST_INSERT_HEAD(&db->groups, &blk->u.disk_group, entry);
915 #endif
916 		break;
917 	/*
918 	 * Volume VBLK fields:
919 	 * Offset	Size	Description
920 	 * ------------+-------+------------------------
921 	 *  0x18+	PS	volume type
922 	 *  0x18+	PS	unknown
923 	 *  0x18+	14(S)	volume state
924 	 *  0x18+16	1	volume number
925 	 *  0x18+21	PN	volume children count
926 	 *  0x2D+16	PN	volume size
927 	 *  0x3D+4	1	partition type
928 	 */
929 	case LDM_VBLK_T_VOLUME:
930 		offset = ldm_vparm_skip(p, offset, size);
931 		if (offset < 0) {
932 			errstr = "volume type";
933 			goto fail;
934 		}
935 		offset = ldm_vparm_skip(p, offset, size);
936 		if (offset < 0) {
937 			errstr = "unknown param";
938 			goto fail;
939 		}
940 		if (offset + 21 >= size) {
941 			errstr = "too small buffer";
942 			goto fail;
943 		}
944 		blk->u.vol.number = p[offset + 16];
945 		offset = ldm_vparm_skip(p, offset + 21, size);
946 		if (offset < 0) {
947 			errstr = "children count";
948 			goto fail;
949 		}
950 		offset = ldm_vnum_get(p, offset + 16, &blk->u.vol.size, size);
951 		if (offset < 0) {
952 			errstr = "volume size";
953 			goto fail;
954 		}
955 		if (offset + 4 >= size) {
956 			errstr = "too small buffer";
957 			goto fail;
958 		}
959 		blk->u.vol.part_type = p[offset + 4];
960 		/* keep volumes ordered by volume number */
961 		last = NULL;
962 		LIST_FOREACH(volume, &db->volumes, entry) {
963 			if (volume->number > blk->u.vol.number)
964 				break;
965 			last = volume;
966 		}
967 		if (last != NULL)
968 			LIST_INSERT_AFTER(last, &blk->u.vol, entry);
969 		else
970 			LIST_INSERT_HEAD(&db->volumes, &blk->u.vol, entry);
971 		break;
972 	default:
973 		LDM_DEBUG(1, "unknown VBLK type 0x%02x\n", blk->type);
974 		LDM_DUMP(p, size);
975 	}
976 	LIST_INSERT_HEAD(&db->vblks, blk, entry);
977 	return (0);
978 fail:
979 	LDM_DEBUG(0, "failed to parse '%s' in VBLK of type 0x%02x\n",
980 	    errstr, blk->type);
981 	LDM_DUMP(p, size);
982 	g_free(blk);
983 	return (EINVAL);
984 }
985 
986 static void
987 ldm_vmdb_free(struct ldm_db *db)
988 {
989 	struct ldm_vblk *vblk;
990 	struct ldm_xvblk *xvblk;
991 
992 	while (!LIST_EMPTY(&db->xvblks)) {
993 		xvblk = LIST_FIRST(&db->xvblks);
994 		LIST_REMOVE(xvblk, entry);
995 		g_free(xvblk->data);
996 		g_free(xvblk);
997 	}
998 	while (!LIST_EMPTY(&db->vblks)) {
999 		vblk = LIST_FIRST(&db->vblks);
1000 		LIST_REMOVE(vblk, entry);
1001 		g_free(vblk);
1002 	}
1003 }
1004 
1005 static int
1006 ldm_vmdb_parse(struct ldm_db *db, struct g_consumer *cp)
1007 {
1008 	struct g_provider *pp;
1009 	struct ldm_vblk *vblk;
1010 	struct ldm_xvblk *xvblk;
1011 	struct ldm_volume *volume;
1012 	struct ldm_component *comp;
1013 	struct ldm_vblkhdr vh;
1014 	u_char *buf, *p;
1015 	size_t size, n, sectors;
1016 	uint64_t offset;
1017 	int error;
1018 
1019 	pp = cp->provider;
1020 	size = howmany(db->dh.last_seq * db->dh.size, pp->sectorsize);
1021 	size -= 1; /* one sector takes vmdb header */
1022 	for (n = 0; n < size; n += MAXPHYS / pp->sectorsize) {
1023 		offset = db->ph.db_offset + db->th.conf_offset + n + 1;
1024 		sectors = (size - n) > (MAXPHYS / pp->sectorsize) ?
1025 		    MAXPHYS / pp->sectorsize: size - n;
1026 		/* read VBLKs */
1027 		buf = g_read_data(cp, offset * pp->sectorsize,
1028 		    sectors * pp->sectorsize, &error);
1029 		if (buf == NULL) {
1030 			LDM_DEBUG(0, "%s: failed to read VBLK\n",
1031 			    pp->name);
1032 			goto fail;
1033 		}
1034 		for (p = buf; p < buf + sectors * pp->sectorsize;
1035 		    p += db->dh.size) {
1036 			if (memcmp(p, LDM_VBLK_SIGN,
1037 			    strlen(LDM_VBLK_SIGN)) != 0) {
1038 				LDM_DEBUG(0, "%s: no VBLK signature\n",
1039 				    pp->name);
1040 				LDM_DUMP(p, db->dh.size);
1041 				goto fail;
1042 			}
1043 			vh.seq = be32dec(p + LDM_VBLK_SEQ_OFF);
1044 			vh.group = be32dec(p + LDM_VBLK_GROUP_OFF);
1045 			/* skip empty blocks */
1046 			if (vh.seq == 0 || vh.group == 0)
1047 				continue;
1048 			vh.index = be16dec(p + LDM_VBLK_INDEX_OFF);
1049 			vh.count = be16dec(p + LDM_VBLK_COUNT_OFF);
1050 			if (vh.count == 0 || vh.count > 4 ||
1051 			    vh.seq > db->dh.last_seq) {
1052 				LDM_DEBUG(0, "%s: invalid values "
1053 				    "in the VBLK header\n", pp->name);
1054 				LDM_DUMP(p, db->dh.size);
1055 				goto fail;
1056 			}
1057 			if (vh.count > 1) {
1058 				error = ldm_xvblk_handle(db, &vh, p);
1059 				if (error != 0) {
1060 					LDM_DEBUG(0, "%s: xVBLK "
1061 					    "is corrupted\n", pp->name);
1062 					LDM_DUMP(p, db->dh.size);
1063 					goto fail;
1064 				}
1065 				continue;
1066 			}
1067 			if (be16dec(p + 16) != 0)
1068 				LDM_DEBUG(1, "%s: VBLK update"
1069 				    " status is %u\n", pp->name,
1070 				    be16dec(p + 16));
1071 			error = ldm_vblk_handle(db, p, db->dh.size);
1072 			if (error != 0)
1073 				goto fail;
1074 		}
1075 		g_free(buf);
1076 		buf = NULL;
1077 	}
1078 	/* Parse xVBLKs */
1079 	while (!LIST_EMPTY(&db->xvblks)) {
1080 		xvblk = LIST_FIRST(&db->xvblks);
1081 		if (xvblk->map == 0xFF) {
1082 			error = ldm_vblk_handle(db, xvblk->data, xvblk->size);
1083 			if (error != 0)
1084 				goto fail;
1085 		} else {
1086 			LDM_DEBUG(0, "%s: incomplete or corrupt "
1087 			    "xVBLK found\n", pp->name);
1088 			goto fail;
1089 		}
1090 		LIST_REMOVE(xvblk, entry);
1091 		g_free(xvblk->data);
1092 		g_free(xvblk);
1093 	}
1094 	/* construct all VBLKs relations */
1095 	LIST_FOREACH(volume, &db->volumes, entry) {
1096 		LIST_FOREACH(vblk, &db->vblks, entry)
1097 			if (vblk->type == LDM_VBLK_T_COMPONENT &&
1098 			    vblk->u.comp.vol_id == volume->id) {
1099 				LIST_INSERT_HEAD(&volume->components,
1100 				    &vblk->u.comp, entry);
1101 				volume->count++;
1102 			}
1103 		LIST_FOREACH(comp, &volume->components, entry)
1104 			LIST_FOREACH(vblk, &db->vblks, entry)
1105 				if (vblk->type == LDM_VBLK_T_PARTITION &&
1106 				    vblk->u.part.comp_id == comp->id) {
1107 					LIST_INSERT_HEAD(&comp->partitions,
1108 					    &vblk->u.part, entry);
1109 					comp->count++;
1110 				}
1111 	}
1112 	return (0);
1113 fail:
1114 	ldm_vmdb_free(db);
1115 	g_free(buf);
1116 	return (ENXIO);
1117 }
1118 
1119 static int
1120 g_part_ldm_add(struct g_part_table *basetable, struct g_part_entry *baseentry,
1121     struct g_part_parms *gpp)
1122 {
1123 
1124 	return (ENOSYS);
1125 }
1126 
1127 static int
1128 g_part_ldm_bootcode(struct g_part_table *basetable, struct g_part_parms *gpp)
1129 {
1130 
1131 	return (ENOSYS);
1132 }
1133 
1134 static int
1135 g_part_ldm_create(struct g_part_table *basetable, struct g_part_parms *gpp)
1136 {
1137 
1138 	return (ENOSYS);
1139 }
1140 
1141 static int
1142 g_part_ldm_destroy(struct g_part_table *basetable, struct g_part_parms *gpp)
1143 {
1144 	struct g_part_ldm_table *table;
1145 	struct g_provider *pp;
1146 
1147 	table = (struct g_part_ldm_table *)basetable;
1148 	/*
1149 	 * To destroy LDM on a disk partitioned with GPT we should delete
1150 	 * ms-ldm-metadata partition, but we can't do this via standard
1151 	 * GEOM_PART method.
1152 	 */
1153 	if (table->is_gpt)
1154 		return (ENOSYS);
1155 	pp = LIST_FIRST(&basetable->gpt_gp->consumer)->provider;
1156 	/*
1157 	 * To destroy LDM we should wipe MBR, first private header and
1158 	 * backup private headers.
1159 	 */
1160 	basetable->gpt_smhead = (1 << ldm_ph_off[0]) | 1;
1161 	/*
1162 	 * Don't touch last backup private header when LDM database is
1163 	 * not located in the last 1MByte area.
1164 	 * XXX: can't remove all blocks.
1165 	 */
1166 	if (table->db_offset + LDM_DB_SIZE ==
1167 	    pp->mediasize / pp->sectorsize)
1168 		basetable->gpt_smtail = 1;
1169 	return (0);
1170 }
1171 
1172 static void
1173 g_part_ldm_dumpconf(struct g_part_table *basetable,
1174     struct g_part_entry *baseentry, struct sbuf *sb, const char *indent)
1175 {
1176 	struct g_part_ldm_entry *entry;
1177 
1178 	entry = (struct g_part_ldm_entry *)baseentry;
1179 	if (indent == NULL) {
1180 		/* conftxt: libdisk compatibility */
1181 		sbuf_printf(sb, " xs LDM xt %u", entry->type);
1182 	} else if (entry != NULL) {
1183 		/* confxml: partition entry information */
1184 		sbuf_printf(sb, "%s<rawtype>%u</rawtype>\n", indent,
1185 		    entry->type);
1186 	} else {
1187 		/* confxml: scheme information */
1188 	}
1189 }
1190 
1191 static int
1192 g_part_ldm_dumpto(struct g_part_table *table, struct g_part_entry *baseentry)
1193 {
1194 
1195 	return (0);
1196 }
1197 
1198 static int
1199 g_part_ldm_modify(struct g_part_table *basetable,
1200     struct g_part_entry *baseentry, struct g_part_parms *gpp)
1201 {
1202 
1203 	return (ENOSYS);
1204 }
1205 
1206 static const char *
1207 g_part_ldm_name(struct g_part_table *table, struct g_part_entry *baseentry,
1208     char *buf, size_t bufsz)
1209 {
1210 
1211 	snprintf(buf, bufsz, "s%d", baseentry->gpe_index);
1212 	return (buf);
1213 }
1214 
1215 static int
1216 ldm_gpt_probe(struct g_part_table *basetable, struct g_consumer *cp)
1217 {
1218 	struct g_part_ldm_table *table;
1219 	struct g_part_table *gpt;
1220 	struct g_part_entry *entry;
1221 	struct g_consumer *cp2;
1222 	struct gpt_ent *part;
1223 	u_char *buf;
1224 	int error;
1225 
1226 	/*
1227 	 * XXX: We use some knowledge about GEOM_PART_GPT internal
1228 	 * structures, but it is easier than parse GPT by himself.
1229 	 */
1230 	g_topology_lock();
1231 	gpt = cp->provider->geom->softc;
1232 	LIST_FOREACH(entry, &gpt->gpt_entry, gpe_entry) {
1233 		part = (struct gpt_ent *)(entry + 1);
1234 		/* Search ms-ldm-metadata partition */
1235 		if (memcmp(&part->ent_type,
1236 		    &gpt_uuid_ms_ldm_metadata, sizeof(struct uuid)) != 0 ||
1237 		    entry->gpe_end - entry->gpe_start < LDM_DB_SIZE - 1)
1238 			continue;
1239 
1240 		/* Create new consumer and attach it to metadata partition */
1241 		cp2 = g_new_consumer(cp->geom);
1242 		error = g_attach(cp2, entry->gpe_pp);
1243 		if (error != 0) {
1244 			g_destroy_consumer(cp2);
1245 			g_topology_unlock();
1246 			return (ENXIO);
1247 		}
1248 		error = g_access(cp2, 1, 0, 0);
1249 		if (error != 0) {
1250 			g_detach(cp2);
1251 			g_destroy_consumer(cp2);
1252 			g_topology_unlock();
1253 			return (ENXIO);
1254 		}
1255 		g_topology_unlock();
1256 
1257 		LDM_DEBUG(2, "%s: LDM metadata partition %s found in the GPT",
1258 		    cp->provider->name, cp2->provider->name);
1259 		/* Read the LDM private header */
1260 		buf = ldm_privhdr_read(cp2,
1261 		    ldm_ph_off[LDM_PH_GPTINDEX] * cp2->provider->sectorsize,
1262 		    &error);
1263 		if (buf != NULL) {
1264 			table = (struct g_part_ldm_table *)basetable;
1265 			table->is_gpt = 1;
1266 			g_free(buf);
1267 			return (G_PART_PROBE_PRI_HIGH);
1268 		}
1269 
1270 		/* second consumer is no longer needed. */
1271 		g_topology_lock();
1272 		g_access(cp2, -1, 0, 0);
1273 		g_detach(cp2);
1274 		g_destroy_consumer(cp2);
1275 		break;
1276 	}
1277 	g_topology_unlock();
1278 	return (ENXIO);
1279 }
1280 
1281 static int
1282 g_part_ldm_probe(struct g_part_table *basetable, struct g_consumer *cp)
1283 {
1284 	struct g_provider *pp;
1285 	u_char *buf, type[64];
1286 	int error, idx;
1287 
1288 
1289 	pp = cp->provider;
1290 	if (pp->sectorsize != 512)
1291 		return (ENXIO);
1292 
1293 	error = g_getattr("PART::scheme", cp, &type);
1294 	if (error == 0 && strcmp(type, "GPT") == 0) {
1295 		if (g_getattr("PART::type", cp, &type) != 0 ||
1296 		    strcmp(type, "ms-ldm-data") != 0)
1297 			return (ENXIO);
1298 		error = ldm_gpt_probe(basetable, cp);
1299 		return (error);
1300 	}
1301 
1302 	if (basetable->gpt_depth != 0)
1303 		return (ENXIO);
1304 
1305 	/* LDM has 1M metadata area */
1306 	if (pp->mediasize <= 1024 * 1024)
1307 		return (ENOSPC);
1308 
1309 	/* Check that there's a MBR */
1310 	buf = g_read_data(cp, 0, pp->sectorsize, &error);
1311 	if (buf == NULL)
1312 		return (error);
1313 
1314 	if (le16dec(buf + DOSMAGICOFFSET) != DOSMAGIC) {
1315 		g_free(buf);
1316 		return (ENXIO);
1317 	}
1318 	error = ENXIO;
1319 	/* Check that we have LDM partitions in the MBR */
1320 	for (idx = 0; idx < NDOSPART && error != 0; idx++) {
1321 		if (buf[DOSPARTOFF + idx * DOSPARTSIZE + 4] == DOSPTYP_LDM)
1322 			error = 0;
1323 	}
1324 	g_free(buf);
1325 	if (error == 0) {
1326 		LDM_DEBUG(2, "%s: LDM data partitions found in MBR",
1327 		    pp->name);
1328 		/* Read the LDM private header */
1329 		buf = ldm_privhdr_read(cp,
1330 		    ldm_ph_off[LDM_PH_MBRINDEX] * pp->sectorsize, &error);
1331 		if (buf == NULL)
1332 			return (error);
1333 		g_free(buf);
1334 		return (G_PART_PROBE_PRI_HIGH);
1335 	}
1336 	return (error);
1337 }
1338 
1339 static int
1340 g_part_ldm_read(struct g_part_table *basetable, struct g_consumer *cp)
1341 {
1342 	struct g_part_ldm_table *table;
1343 	struct g_part_ldm_entry *entry;
1344 	struct g_consumer *cp2;
1345 	struct ldm_component *comp;
1346 	struct ldm_partition *part;
1347 	struct ldm_volume *vol;
1348 	struct ldm_disk *disk;
1349 	struct ldm_db db;
1350 	int error, index, skipped;
1351 
1352 	table = (struct g_part_ldm_table *)basetable;
1353 	memset(&db, 0, sizeof(db));
1354 	cp2 = cp;					/* ms-ldm-data */
1355 	if (table->is_gpt)
1356 		cp = LIST_FIRST(&cp->geom->consumer);	/* ms-ldm-metadata */
1357 	/* Read and parse LDM private headers. */
1358 	error = ldm_privhdr_check(&db, cp, table->is_gpt);
1359 	if (error != 0)
1360 		goto gpt_cleanup;
1361 	basetable->gpt_first = table->is_gpt ? 0: db.ph.start;
1362 	basetable->gpt_last = basetable->gpt_first + db.ph.size - 1;
1363 	table->db_offset = db.ph.db_offset;
1364 	/* Make additional checks for GPT */
1365 	if (table->is_gpt) {
1366 		error = ldm_gpt_check(&db, cp);
1367 		if (error != 0)
1368 			goto gpt_cleanup;
1369 		/*
1370 		 * Now we should reset database offset to zero, because our
1371 		 * consumer cp is attached to the ms-ldm-metadata partition
1372 		 * and we don't need add db_offset to read from it.
1373 		 */
1374 		db.ph.db_offset = 0;
1375 	}
1376 	/* Read and parse LDM TOC headers. */
1377 	error = ldm_tochdr_check(&db, cp);
1378 	if (error != 0)
1379 		goto gpt_cleanup;
1380 	/* Read and parse LDM VMDB header. */
1381 	error = ldm_vmdbhdr_check(&db, cp);
1382 	if (error != 0)
1383 		goto gpt_cleanup;
1384 	error = ldm_vmdb_parse(&db, cp);
1385 	/*
1386 	 * For the GPT case we must detach and destroy
1387 	 * second consumer before return.
1388 	 */
1389 gpt_cleanup:
1390 	if (table->is_gpt) {
1391 		g_topology_lock();
1392 		g_access(cp, -1, 0, 0);
1393 		g_detach(cp);
1394 		g_destroy_consumer(cp);
1395 		g_topology_unlock();
1396 		cp = cp2;
1397 	}
1398 	if (error != 0)
1399 		return (error);
1400 	/* Search current disk in the disk list. */
1401 	LIST_FOREACH(disk, &db.disks, entry)
1402 	    if (memcmp(&disk->guid, &db.ph.disk_guid,
1403 		sizeof(struct uuid)) == 0)
1404 		    break;
1405 	if (disk == NULL) {
1406 		LDM_DEBUG(1, "%s: no LDM volumes on this disk",
1407 		    cp->provider->name);
1408 		ldm_vmdb_free(&db);
1409 		return (ENXIO);
1410 	}
1411 	index = 1;
1412 	LIST_FOREACH(vol, &db.volumes, entry) {
1413 		LIST_FOREACH(comp, &vol->components, entry) {
1414 			/* Skip volumes from different disks. */
1415 			part = LIST_FIRST(&comp->partitions);
1416 			if (part->disk_id != disk->id)
1417 				continue;
1418 			skipped = 0;
1419 			/* We don't support spanned and striped volumes. */
1420 			if (comp->count > 1 || part->offset != 0) {
1421 				LDM_DEBUG(1, "%s: LDM volume component "
1422 				    "%ju has %u partitions. Skipped",
1423 				    cp->provider->name, (uintmax_t)comp->id,
1424 				    comp->count);
1425 				skipped = 1;
1426 			}
1427 			/*
1428 			 * Allow mirrored volumes only when they are explicitly
1429 			 * allowed with kern.geom.part.ldm.show_mirrors=1.
1430 			 */
1431 			if (vol->count > 1 && show_mirrors == 0) {
1432 				LDM_DEBUG(1, "%s: LDM volume %ju has %u "
1433 				    "components. Skipped",
1434 				    cp->provider->name, (uintmax_t)vol->id,
1435 				    vol->count);
1436 				skipped = 1;
1437 			}
1438 			entry = (struct g_part_ldm_entry *)g_part_new_entry(
1439 			    basetable, index++,
1440 			    basetable->gpt_first + part->start,
1441 			    basetable->gpt_first + part->start +
1442 			    part->size - 1);
1443 			/*
1444 			 * Mark skipped partition as ms-ldm-data partition.
1445 			 * We do not support them, but it is better to show
1446 			 * that we have something there, than just show
1447 			 * free space.
1448 			 */
1449 			if (skipped == 0)
1450 				entry->type = vol->part_type;
1451 			else
1452 				entry->type = DOSPTYP_LDM;
1453 			LDM_DEBUG(1, "%s: new volume id: %ju, start: %ju,"
1454 			    " end: %ju, type: 0x%02x\n", cp->provider->name,
1455 			    (uintmax_t)part->id,(uintmax_t)part->start +
1456 			    basetable->gpt_first, (uintmax_t)part->start +
1457 			    part->size + basetable->gpt_first - 1,
1458 			    vol->part_type);
1459 		}
1460 	}
1461 	ldm_vmdb_free(&db);
1462 	return (error);
1463 }
1464 
1465 static const char *
1466 g_part_ldm_type(struct g_part_table *basetable, struct g_part_entry *baseentry,
1467     char *buf, size_t bufsz)
1468 {
1469 	struct g_part_ldm_entry *entry;
1470 	int i;
1471 
1472 	entry = (struct g_part_ldm_entry *)baseentry;
1473 	for (i = 0; i < nitems(ldm_alias_match); i++) {
1474 		if (ldm_alias_match[i].typ == entry->type)
1475 			return (g_part_alias_name(ldm_alias_match[i].alias));
1476 	}
1477 	snprintf(buf, bufsz, "!%d", entry->type);
1478 	return (buf);
1479 }
1480 
1481 static int
1482 g_part_ldm_write(struct g_part_table *basetable, struct g_consumer *cp)
1483 {
1484 
1485 	return (ENOSYS);
1486 }
1487