2 * Copyright (c) 2012 Andrey V. Elsukov <ae@FreeBSD.org>
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
30 #include <sys/param.h>
32 #include <sys/diskmbr.h>
33 #include <sys/endian.h>
35 #include <sys/kernel.h>
37 #include <sys/limits.h>
39 #include <sys/malloc.h>
40 #include <sys/mutex.h>
41 #include <sys/queue.h>
43 #include <sys/systm.h>
44 #include <sys/sysctl.h>
46 #include <geom/geom.h>
47 #include <geom/part/g_part.h>
49 #include "g_part_if.h"
51 FEATURE(geom_part_ldm, "GEOM partitioning class for LDM support");
53 SYSCTL_DECL(_kern_geom_part);
54 static SYSCTL_NODE(_kern_geom_part, OID_AUTO, ldm, CTLFLAG_RW, 0,
55 "GEOM_PART_LDM Logical Disk Manager");
57 static u_int ldm_debug = 0;
58 TUNABLE_INT("kern.geom.part.ldm.debug", &ldm_debug);
59 SYSCTL_UINT(_kern_geom_part_ldm, OID_AUTO, debug,
60 CTLFLAG_RW | CTLFLAG_TUN, &ldm_debug, 0, "Debug level");
63 * This allows access to mirrored LDM volumes. Since we do not
64 * doing mirroring here, it is not enabled by default.
66 static u_int show_mirrors = 0;
67 TUNABLE_INT("kern.geom.part.ldm.show_mirrors", &show_mirrors);
68 SYSCTL_UINT(_kern_geom_part_ldm, OID_AUTO, show_mirrors,
69 CTLFLAG_RW | CTLFLAG_TUN, &show_mirrors, 0, "Show mirrored volumes");
71 #define LDM_DEBUG(lvl, fmt, ...) do { \
72 if (ldm_debug >= (lvl)) { \
73 printf("GEOM_PART: " fmt "\n", __VA_ARGS__); \
76 #define LDM_DUMP(buf, size) do { \
77 if (ldm_debug > 1) { \
78 hexdump(buf, size, NULL, 0); \
83 * There are internal representations of LDM structures.
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.
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.
95 * On a disk partitioned with GPT, the LDM has not first private header.
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"
116 struct uuid disk_guid;
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 */
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.
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
145 uint64_t conf_offset; /* configuration offset */
146 uint64_t log_offset; /* log offset */
150 * LDM database header is 512 bytes long.
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
160 uint32_t last_seq; /* sequence number of last VBLK */
161 uint32_t size; /* size of VBLK */
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.
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
174 * Volumes <- Components <- Partitions -> Disks
176 * One volume can contain several components. In this case LDM
177 * does mirroring of volume data to each component.
179 * Also each component can contain several partitions (spanned or
183 struct ldm_component {
184 uint64_t id; /* object id */
185 uint64_t vol_id; /* parent volume object id */
188 LIST_HEAD(, ldm_partition) partitions;
189 LIST_ENTRY(ldm_component) entry;
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 */
199 LIST_HEAD(, ldm_component) components;
200 LIST_ENTRY(ldm_volume) entry;
204 uint64_t id; /* object id */
205 struct uuid guid; /* disk guid */
207 LIST_ENTRY(ldm_disk) entry;
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 */
216 LIST_ENTRY(ldm_disk_group) entry;
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 */
228 LIST_ENTRY(ldm_partition) entry;
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
244 * 0x18+ PN object name
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.
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
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 */
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
272 uint8_t type; /* VBLK type */
275 struct ldm_volume vol;
276 struct ldm_component comp;
277 struct ldm_disk disk;
278 struct ldm_partition part;
280 struct ldm_disk_group disk_group;
283 LIST_ENTRY(ldm_vblk) entry;
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).
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 */
298 LIST_ENTRY(ldm_xvblk) entry;
301 /* The internal representation of LDM database. */
303 struct ldm_privhdr ph; /* private header */
304 struct ldm_tochdr th; /* TOC header */
305 struct ldm_vmdbhdr dh; /* VMDB header */
307 LIST_HEAD(, ldm_volume) volumes;
308 LIST_HEAD(, ldm_disk) disks;
309 LIST_HEAD(, ldm_vblk) vblks;
310 LIST_HEAD(, ldm_xvblk) xvblks;
313 static struct uuid gpt_uuid_ms_ldm_metadata = GPT_ENT_TYPE_MS_LDM_METADATA;
315 struct g_part_ldm_table {
316 struct g_part_table base;
320 struct g_part_ldm_entry {
321 struct g_part_entry base;
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 *,
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 *,
341 static int g_part_ldm_write(struct g_part_table *, struct g_consumer *);
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),
359 static struct g_part_scheme g_part_ldm_scheme = {
362 sizeof(struct g_part_ldm_table),
363 .gps_entrysz = sizeof(struct g_part_ldm_entry)
365 G_PART_SCHEME_DECLARE(g_part_ldm);
367 static struct g_part_ldm_alias {
370 } ldm_alias_match[] = {
371 { DOSPTYP_NTFS, G_PART_ALIAS_MS_NTFS },
372 { DOSPTYP_FAT32, G_PART_ALIAS_MS_FAT32 },
373 { DOSPTYP_386BSD, G_PART_ALIAS_FREEBSD },
374 { DOSPTYP_LDM, G_PART_ALIAS_MS_LDM_DATA },
375 { DOSPTYP_LINSWP, G_PART_ALIAS_LINUX_SWAP },
376 { DOSPTYP_LINUX, G_PART_ALIAS_LINUX_DATA },
377 { DOSPTYP_LINLVM, G_PART_ALIAS_LINUX_LVM },
378 { DOSPTYP_LINRAID, G_PART_ALIAS_LINUX_RAID },
382 ldm_privhdr_read(struct g_consumer *cp, uint64_t off, int *error)
384 struct g_provider *pp;
388 buf = g_read_data(cp, off, pp->sectorsize, error);
392 if (memcmp(buf, LDM_PH_SIGN, strlen(LDM_PH_SIGN)) != 0) {
393 LDM_DEBUG(1, "%s: invalid LDM private header signature",
403 ldm_privhdr_parse(struct g_consumer *cp, struct ldm_privhdr *hdr,
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,
418 error = parse_uuid(buf + LDM_PH_DISKGUID_OFF, &hdr->disk_guid);
421 error = parse_uuid(buf + LDM_PH_DGGUID_OFF, &hdr->dg_guid);
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);
437 ldm_privhdr_check(struct ldm_db *db, struct g_consumer *cp, int is_gpt)
439 struct g_consumer *cp2;
440 struct g_provider *pp;
441 struct ldm_privhdr hdr;
442 uint64_t offset, last;
449 * The last LBA is used in several checks below, for the
450 * GPT case it should be calculated relative to the whole
453 cp2 = LIST_FIRST(&pp->geom->consumer);
455 cp2->provider->mediasize / cp2->provider->sectorsize - 1;
457 last = pp->mediasize / pp->sectorsize - 1;
458 for (found = 0, i = is_gpt;
459 i < sizeof(ldm_ph_off) / sizeof(ldm_ph_off[0]); i++) {
460 offset = ldm_ph_off[i];
462 * In the GPT case consumer is attached to the LDM metadata
463 * partition and we don't need add db_offset.
466 offset += db->ph.db_offset;
467 if (i == LDM_PH_MBRINDEX) {
469 * Prepare to errors and setup new base offset
470 * to read backup private headers. Assume that LDM
471 * database is in the last 1Mbyte area.
473 db->ph.db_offset = last - LDM_DB_SIZE;
475 buf = ldm_privhdr_read(cp, offset * pp->sectorsize, &error);
477 LDM_DEBUG(1, "%s: failed to read private header "
478 "%d at LBA %ju", pp->name, i, (uintmax_t)offset);
481 error = ldm_privhdr_parse(cp, &hdr, buf);
483 LDM_DEBUG(1, "%s: failed to parse private "
484 "header %d", pp->name, i);
485 LDM_DUMP(buf, pp->sectorsize);
490 if (hdr.start > last ||
491 hdr.start + hdr.size - 1 > last ||
492 (hdr.start + hdr.size - 1 > hdr.db_offset && !is_gpt) ||
493 hdr.db_size != LDM_DB_SIZE ||
494 hdr.db_offset + LDM_DB_SIZE - 1 > last ||
495 hdr.th_offset[0] >= LDM_DB_SIZE ||
496 hdr.th_offset[1] >= LDM_DB_SIZE ||
497 hdr.conf_size + hdr.log_size >= LDM_DB_SIZE) {
498 LDM_DEBUG(1, "%s: invalid values in the "
499 "private header %d", pp->name, i);
500 LDM_DEBUG(2, "%s: start: %jd, size: %jd, "
501 "db_offset: %jd, db_size: %jd, th_offset0: %jd, "
502 "th_offset1: %jd, conf_size: %jd, log_size: %jd, "
503 "last: %jd", pp->name, hdr.start, hdr.size,
504 hdr.db_offset, hdr.db_size, hdr.th_offset[0],
505 hdr.th_offset[1], hdr.conf_size, hdr.log_size,
509 if (found != 0 && memcmp(&db->ph, &hdr, sizeof(hdr)) != 0) {
510 LDM_DEBUG(0, "%s: private headers are not equal",
514 * We have different headers in the LDM.
515 * We can not trust this metadata.
517 LDM_DEBUG(0, "%s: refuse LDM metadata",
522 * We already have read primary private header
523 * and it differs from this backup one.
524 * Prefer the backup header and save it.
529 memcpy(&db->ph, &hdr, sizeof(hdr));
533 LDM_DEBUG(1, "%s: valid LDM private header not found",
541 ldm_gpt_check(struct ldm_db *db, struct g_consumer *cp)
543 struct g_part_table *gpt;
544 struct g_part_entry *e;
545 struct g_consumer *cp2;
548 cp2 = LIST_NEXT(cp, consumer);
550 gpt = cp->provider->geom->softc;
552 LIST_FOREACH(e, &gpt->gpt_entry, gpe_entry) {
553 if (cp->provider == e->gpe_pp) {
554 /* ms-ldm-metadata partition */
555 if (e->gpe_start != db->ph.db_offset ||
556 e->gpe_end != db->ph.db_offset + LDM_DB_SIZE - 1)
558 } else if (cp2->provider == e->gpe_pp) {
559 /* ms-ldm-data partition */
560 if (e->gpe_start != db->ph.start ||
561 e->gpe_end != db->ph.start + db->ph.size - 1)
565 LDM_DEBUG(0, "%s: GPT partition %d boundaries "
566 "do not match with the LDM metadata",
567 e->gpe_pp->name, e->gpe_index);
577 ldm_tochdr_check(struct ldm_db *db, struct g_consumer *cp)
579 struct g_provider *pp;
580 struct ldm_tochdr hdr;
581 uint64_t offset, conf_size, log_size;
586 for (i = 0, found = 0; i < LDM_TH_COUNT; i++) {
587 offset = db->ph.db_offset + db->ph.th_offset[i];
588 buf = g_read_data(cp,
589 offset * pp->sectorsize, pp->sectorsize, &error);
591 LDM_DEBUG(1, "%s: failed to read TOC header "
592 "at LBA %ju", pp->name, (uintmax_t)offset);
595 if (memcmp(buf, LDM_TH_SIGN, strlen(LDM_TH_SIGN)) != 0 ||
596 memcmp(buf + LDM_TH_NAME1_OFF, LDM_TH_NAME1,
597 strlen(LDM_TH_NAME1)) != 0 ||
598 memcmp(buf + LDM_TH_NAME2_OFF, LDM_TH_NAME2,
599 strlen(LDM_TH_NAME2)) != 0) {
600 LDM_DEBUG(1, "%s: failed to parse TOC header "
601 "at LBA %ju", pp->name, (uintmax_t)offset);
602 LDM_DUMP(buf, pp->sectorsize);
606 hdr.conf_offset = be64dec(buf + LDM_TH_CONF_OFF);
607 hdr.log_offset = be64dec(buf + LDM_TH_LOG_OFF);
608 conf_size = be64dec(buf + LDM_TH_CONFSIZE_OFF);
609 log_size = be64dec(buf + LDM_TH_LOGSIZE_OFF);
610 if (conf_size != db->ph.conf_size ||
611 hdr.conf_offset + conf_size >= LDM_DB_SIZE ||
612 log_size != db->ph.log_size ||
613 hdr.log_offset + log_size >= LDM_DB_SIZE) {
614 LDM_DEBUG(1, "%s: invalid values in the "
615 "TOC header at LBA %ju", pp->name,
617 LDM_DUMP(buf, pp->sectorsize);
623 memcpy(&db->th, &hdr, sizeof(hdr));
627 LDM_DEBUG(0, "%s: valid LDM TOC header not found.",
635 ldm_vmdbhdr_check(struct ldm_db *db, struct g_consumer *cp)
637 struct g_provider *pp;
645 offset = db->ph.db_offset + db->th.conf_offset;
646 buf = g_read_data(cp, offset * pp->sectorsize, pp->sectorsize,
649 LDM_DEBUG(0, "%s: failed to read VMDB header at "
650 "LBA %ju", pp->name, (uintmax_t)offset);
653 if (memcmp(buf, LDM_VMDB_SIGN, strlen(LDM_VMDB_SIGN)) != 0) {
655 LDM_DEBUG(0, "%s: failed to parse VMDB header at "
656 "LBA %ju", pp->name, (uintmax_t)offset);
660 version = be32dec(buf + LDM_DB_VERSION_OFF);
661 if (version != 0x4000A) {
663 LDM_DEBUG(0, "%s: unsupported VMDB version %u.%u",
664 pp->name, version >> 16, version & 0xFFFF);
668 * Check VMDB update status:
669 * 1 - in a consistent state;
670 * 2 - in a creation phase;
671 * 3 - in a deletion phase;
673 if (be16dec(buf + LDM_DB_STATUS_OFF) != 1) {
675 LDM_DEBUG(0, "%s: VMDB is not in a consistent state",
679 db->dh.last_seq = be32dec(buf + LDM_DB_LASTSEQ_OFF);
680 db->dh.size = be32dec(buf + LDM_DB_SIZE_OFF);
681 error = parse_uuid(buf + LDM_DB_DGGUID_OFF, &dg_guid);
682 /* Compare disk group name and guid from VMDB and private headers */
683 if (error != 0 || db->dh.size == 0 ||
684 pp->sectorsize % db->dh.size != 0 ||
685 strncmp(buf + LDM_DB_DGNAME_OFF, db->ph.dg_name, 31) != 0 ||
686 memcmp(&dg_guid, &db->ph.dg_guid, sizeof(dg_guid)) != 0 ||
687 db->dh.size * db->dh.last_seq >
688 db->ph.conf_size * pp->sectorsize) {
689 LDM_DEBUG(0, "%s: invalid values in the VMDB header",
691 LDM_DUMP(buf, pp->sectorsize);
700 ldm_xvblk_handle(struct ldm_db *db, struct ldm_vblkhdr *vh, const u_char *p)
702 struct ldm_xvblk *blk;
705 size = db->dh.size - 16;
706 LIST_FOREACH(blk, &db->xvblks, entry)
707 if (blk->group == vh->group)
710 blk = g_malloc(sizeof(*blk), M_WAITOK | M_ZERO);
711 blk->group = vh->group;
712 blk->size = size * vh->count + 16;
713 blk->data = g_malloc(blk->size, M_WAITOK | M_ZERO);
714 blk->map = 0xFF << vh->count;
715 LIST_INSERT_HEAD(&db->xvblks, blk, entry);
717 if ((blk->map & (1 << vh->index)) != 0) {
718 /* Block with given index has been already saved. */
721 /* Copy the data block to the place related to index. */
722 memcpy(blk->data + size * vh->index + 16, p + 16, size);
723 blk->map |= 1 << vh->index;
727 /* Read the variable-width numeric field and return new offset */
729 ldm_vnum_get(const u_char *buf, int offset, uint64_t *result, size_t range)
735 if (len > sizeof(uint64_t) || len + offset >= range)
737 for (num = 0; len > 0; len--)
738 num = (num << 8) | buf[offset++];
743 /* Read the variable-width string and return new offset */
745 ldm_vstr_get(const u_char *buf, int offset, u_char *result,
746 size_t maxlen, size_t range)
751 if (len >= maxlen || len + offset >= range)
753 memcpy(result, buf + offset, len);
755 return (offset + len);
758 /* Just skip the variable-width variable and return new offset */
760 ldm_vparm_skip(const u_char *buf, int offset, size_t range)
765 if (offset + len >= range)
768 return (offset + len);
772 ldm_vblk_handle(struct ldm_db *db, const u_char *p, size_t size)
774 struct ldm_vblk *blk;
775 struct ldm_volume *volume, *last;
780 blk = g_malloc(sizeof(*blk), M_WAITOK | M_ZERO);
781 blk->type = p[LDM_VBLK_TYPE_OFF];
782 offset = ldm_vnum_get(p, LDM_VBLK_OID_OFF, &blk->u.id, size);
784 errstr = "object id";
787 offset = ldm_vstr_get(p, offset, vstr, sizeof(vstr), size);
789 errstr = "object name";
794 * Component VBLK fields:
795 * Offset Size Description
796 * ------------+-------+------------------------
797 * 0x18+ PS volume state
798 * 0x18+5 PN component children count
799 * 0x1D+16 PN parent's volume object id
800 * 0x2D+1 PN stripe size
802 case LDM_VBLK_T_COMPONENT:
803 offset = ldm_vparm_skip(p, offset, size);
805 errstr = "volume state";
808 offset = ldm_vparm_skip(p, offset + 5, size);
810 errstr = "children count";
813 offset = ldm_vnum_get(p, offset + 16,
814 &blk->u.comp.vol_id, size);
816 errstr = "volume id";
821 * Partition VBLK fields:
822 * Offset Size Description
823 * ------------+-------+------------------------
824 * 0x18+12 8 partition start offset
825 * 0x18+20 8 volume offset
826 * 0x18+28 PN partition size
827 * 0x34+ PN parent's component object id
828 * 0x34+ PN disk's object id
830 case LDM_VBLK_T_PARTITION:
831 if (offset + 28 >= size) {
832 errstr = "too small buffer";
835 blk->u.part.start = be64dec(p + offset + 12);
836 blk->u.part.offset = be64dec(p + offset + 20);
837 offset = ldm_vnum_get(p, offset + 28, &blk->u.part.size, size);
839 errstr = "partition size";
842 offset = ldm_vnum_get(p, offset, &blk->u.part.comp_id, size);
844 errstr = "component id";
847 offset = ldm_vnum_get(p, offset, &blk->u.part.disk_id, size);
855 * Offset Size Description
856 * ------------+-------+------------------------
859 case LDM_VBLK_T_DISK:
860 errstr = "disk guid";
861 offset = ldm_vstr_get(p, offset, vstr, sizeof(vstr), size);
864 error = parse_uuid(vstr, &blk->u.disk.guid);
867 LIST_INSERT_HEAD(&db->disks, &blk->u.disk, entry);
870 * Disk group VBLK fields:
871 * Offset Size Description
872 * ------------+-------+------------------------
873 * 0x18+ PS disk group GUID
875 case LDM_VBLK_T_DISKGROUP:
877 strncpy(blk->u.disk_group.name, vstr,
878 sizeof(blk->u.disk_group.name));
879 offset = ldm_vstr_get(p, offset, vstr, sizeof(vstr), size);
881 errstr = "disk group guid";
884 error = parse_uuid(name, &blk->u.disk_group.guid);
886 errstr = "disk group guid";
889 LIST_INSERT_HEAD(&db->groups, &blk->u.disk_group, entry);
894 * Offset Size Description
895 * ------------+-------+------------------------
898 case LDM_VBLK_T_DISK4:
899 be_uuid_dec(p + offset, &blk->u.disk.guid);
900 LIST_INSERT_HEAD(&db->disks, &blk->u.disk, entry);
903 * Disk group VBLK fields:
904 * Offset Size Description
905 * ------------+-------+------------------------
908 case LDM_VBLK_T_DISKGROUP4:
910 strncpy(blk->u.disk_group.name, vstr,
911 sizeof(blk->u.disk_group.name));
912 be_uuid_dec(p + offset, &blk->u.disk.guid);
913 LIST_INSERT_HEAD(&db->groups, &blk->u.disk_group, entry);
917 * Volume VBLK fields:
918 * Offset Size Description
919 * ------------+-------+------------------------
920 * 0x18+ PS volume type
922 * 0x18+ 14(S) volume state
923 * 0x18+16 1 volume number
924 * 0x18+21 PN volume children count
925 * 0x2D+16 PN volume size
926 * 0x3D+4 1 partition type
928 case LDM_VBLK_T_VOLUME:
929 offset = ldm_vparm_skip(p, offset, size);
931 errstr = "volume type";
934 offset = ldm_vparm_skip(p, offset, size);
936 errstr = "unknown param";
939 if (offset + 21 >= size) {
940 errstr = "too small buffer";
943 blk->u.vol.number = p[offset + 16];
944 offset = ldm_vparm_skip(p, offset + 21, size);
946 errstr = "children count";
949 offset = ldm_vnum_get(p, offset + 16, &blk->u.vol.size, size);
951 errstr = "volume size";
954 if (offset + 4 >= size) {
955 errstr = "too small buffer";
958 blk->u.vol.part_type = p[offset + 4];
959 /* keep volumes ordered by volume number */
961 LIST_FOREACH(volume, &db->volumes, entry) {
962 if (volume->number > blk->u.vol.number)
967 LIST_INSERT_AFTER(last, &blk->u.vol, entry);
969 LIST_INSERT_HEAD(&db->volumes, &blk->u.vol, entry);
972 LDM_DEBUG(1, "unknown VBLK type 0x%02x\n", blk->type);
975 LIST_INSERT_HEAD(&db->vblks, blk, entry);
978 LDM_DEBUG(0, "failed to parse '%s' in VBLK of type 0x%02x\n",
986 ldm_vmdb_free(struct ldm_db *db)
988 struct ldm_vblk *vblk;
989 struct ldm_xvblk *xvblk;
991 while (!LIST_EMPTY(&db->xvblks)) {
992 xvblk = LIST_FIRST(&db->xvblks);
993 LIST_REMOVE(xvblk, entry);
997 while (!LIST_EMPTY(&db->vblks)) {
998 vblk = LIST_FIRST(&db->vblks);
999 LIST_REMOVE(vblk, entry);
1005 ldm_vmdb_parse(struct ldm_db *db, struct g_consumer *cp)
1007 struct g_provider *pp;
1008 struct ldm_vblk *vblk;
1009 struct ldm_xvblk *xvblk;
1010 struct ldm_volume *volume;
1011 struct ldm_component *comp;
1012 struct ldm_vblkhdr vh;
1014 size_t size, n, sectors;
1019 size = (db->dh.last_seq * db->dh.size +
1020 pp->sectorsize - 1) / 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;
1027 buf = g_read_data(cp, offset * pp->sectorsize,
1028 sectors * pp->sectorsize, &error);
1030 LDM_DEBUG(0, "%s: failed to read VBLK\n",
1034 for (p = buf; p < buf + sectors * pp->sectorsize;
1036 if (memcmp(p, LDM_VBLK_SIGN,
1037 strlen(LDM_VBLK_SIGN)) != 0) {
1038 LDM_DEBUG(0, "%s: no VBLK signature\n",
1040 LDM_DUMP(p, db->dh.size);
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)
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);
1058 error = ldm_xvblk_handle(db, &vh, p);
1060 LDM_DEBUG(0, "%s: xVBLK "
1061 "is corrupted\n", pp->name);
1062 LDM_DUMP(p, db->dh.size);
1067 if (be16dec(p + 16) != 0)
1068 LDM_DEBUG(1, "%s: VBLK update"
1069 " status is %u\n", pp->name,
1071 error = ldm_vblk_handle(db, p, db->dh.size);
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);
1086 LDM_DEBUG(0, "%s: incomplete or corrupt "
1087 "xVBLK found\n", pp->name);
1090 LIST_REMOVE(xvblk, entry);
1091 g_free(xvblk->data);
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);
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);
1120 g_part_ldm_add(struct g_part_table *basetable, struct g_part_entry *baseentry,
1121 struct g_part_parms *gpp)
1128 g_part_ldm_bootcode(struct g_part_table *basetable, struct g_part_parms *gpp)
1135 g_part_ldm_create(struct g_part_table *basetable, struct g_part_parms *gpp)
1142 g_part_ldm_destroy(struct g_part_table *basetable, struct g_part_parms *gpp)
1144 struct g_part_ldm_table *table;
1145 struct g_provider *pp;
1147 table = (struct g_part_ldm_table *)basetable;
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
1155 pp = LIST_FIRST(&basetable->gpt_gp->consumer)->provider;
1157 * To destroy LDM we should wipe MBR, first private header and
1158 * backup private headers.
1160 basetable->gpt_smhead = (1 << ldm_ph_off[0]) | 1;
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.
1166 if (table->db_offset + LDM_DB_SIZE ==
1167 pp->mediasize / pp->sectorsize)
1168 basetable->gpt_smtail = 1;
1173 g_part_ldm_dumpconf(struct g_part_table *basetable,
1174 struct g_part_entry *baseentry, struct sbuf *sb, const char *indent)
1176 struct g_part_ldm_entry *entry;
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,
1187 /* confxml: scheme information */
1192 g_part_ldm_dumpto(struct g_part_table *table, struct g_part_entry *baseentry)
1199 g_part_ldm_modify(struct g_part_table *basetable,
1200 struct g_part_entry *baseentry, struct g_part_parms *gpp)
1207 g_part_ldm_name(struct g_part_table *table, struct g_part_entry *baseentry,
1208 char *buf, size_t bufsz)
1211 snprintf(buf, bufsz, "s%d", baseentry->gpe_index);
1216 ldm_gpt_probe(struct g_part_table *basetable, struct g_consumer *cp)
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;
1227 * XXX: We use some knowlege about GEOM_PART_GPT internal
1228 * structures, but it is easier than parse GPT by himself.
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)
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);
1244 g_destroy_consumer(cp2);
1245 g_topology_unlock();
1248 error = g_access(cp2, 1, 0, 0);
1251 g_destroy_consumer(cp2);
1252 g_topology_unlock();
1255 g_topology_unlock();
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,
1264 table = (struct g_part_ldm_table *)basetable;
1267 return (G_PART_PROBE_PRI_HIGH);
1270 /* second consumer is no longer needed. */
1272 g_access(cp2, -1, 0, 0);
1274 g_destroy_consumer(cp2);
1277 g_topology_unlock();
1282 g_part_ldm_probe(struct g_part_table *basetable, struct g_consumer *cp)
1284 struct g_provider *pp;
1285 u_char *buf, type[64];
1290 if (pp->sectorsize != 512)
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)
1298 error = ldm_gpt_probe(basetable, cp);
1302 if (basetable->gpt_depth != 0)
1305 /* LDM has 1M metadata area */
1306 if (pp->mediasize <= 1024 * 1024)
1309 /* Check that there's a MBR */
1310 buf = g_read_data(cp, 0, pp->sectorsize, &error);
1314 if (le16dec(buf + DOSMAGICOFFSET) != DOSMAGIC) {
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)
1326 LDM_DEBUG(2, "%s: LDM data partitions found in MBR",
1328 /* Read the LDM private header */
1329 buf = ldm_privhdr_read(cp,
1330 ldm_ph_off[LDM_PH_MBRINDEX] * pp->sectorsize, &error);
1334 return (G_PART_PROBE_PRI_HIGH);
1340 g_part_ldm_read(struct g_part_table *basetable, struct g_consumer *cp)
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;
1350 int error, index, skipped;
1352 table = (struct g_part_ldm_table *)basetable;
1353 memset(&db, 0, sizeof(db));
1354 cp2 = cp; /* ms-ldm-data */
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);
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);
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.
1374 db.ph.db_offset = 0;
1376 /* Read and parse LDM TOC headers. */
1377 error = ldm_tochdr_check(&db, cp);
1380 /* Read and parse LDM VMDB header. */
1381 error = ldm_vmdbhdr_check(&db, cp);
1384 error = ldm_vmdb_parse(&db, cp);
1386 * For the GPT case we must detach and destroy
1387 * second consumer before return.
1390 if (table->is_gpt) {
1392 g_access(cp, -1, 0, 0);
1394 g_destroy_consumer(cp);
1395 g_topology_unlock();
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)
1406 LDM_DEBUG(1, "%s: no LDM volumes on this disk",
1407 cp->provider->name);
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)
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,
1428 * Allow mirrored volumes only when they are explicitly
1429 * allowed with kern.geom.part.ldm.show_mirrors=1.
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,
1438 entry = (struct g_part_ldm_entry *)g_part_new_entry(
1440 basetable->gpt_first + part->start,
1441 basetable->gpt_first + part->start +
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
1450 entry->type = vol->part_type;
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,
1466 g_part_ldm_type(struct g_part_table *basetable, struct g_part_entry *baseentry,
1467 char *buf, size_t bufsz)
1469 struct g_part_ldm_entry *entry;
1472 entry = (struct g_part_ldm_entry *)baseentry;
1474 i < sizeof(ldm_alias_match) / sizeof(ldm_alias_match[0]); i++) {
1475 if (ldm_alias_match[i].typ == entry->type)
1476 return (g_part_alias_name(ldm_alias_match[i].alias));
1478 snprintf(buf, bufsz, "!%d", entry->type);
1483 g_part_ldm_write(struct g_part_table *basetable, struct g_consumer *cp)