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 SYSCTL_UINT(_kern_geom_part_ldm, OID_AUTO, debug,
59 CTLFLAG_RWTUN, &ldm_debug, 0, "Debug level");
62 * This allows access to mirrored LDM volumes. Since we do not
63 * doing mirroring here, it is not enabled by default.
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");
69 #define LDM_DEBUG(lvl, fmt, ...) do { \
70 if (ldm_debug >= (lvl)) { \
71 printf("GEOM_PART: " fmt "\n", __VA_ARGS__); \
74 #define LDM_DUMP(buf, size) do { \
75 if (ldm_debug > 1) { \
76 hexdump(buf, size, NULL, 0); \
81 * There are internal representations of LDM structures.
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.
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.
93 * On a disk partitioned with GPT, the LDM has not first private header.
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"
114 struct uuid disk_guid;
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 */
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.
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
143 uint64_t conf_offset; /* configuration offset */
144 uint64_t log_offset; /* log offset */
148 * LDM database header is 512 bytes long.
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
158 uint32_t last_seq; /* sequence number of last VBLK */
159 uint32_t size; /* size of VBLK */
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.
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
172 * Volumes <- Components <- Partitions -> Disks
174 * One volume can contain several components. In this case LDM
175 * does mirroring of volume data to each component.
177 * Also each component can contain several partitions (spanned or
181 struct ldm_component {
182 uint64_t id; /* object id */
183 uint64_t vol_id; /* parent volume object id */
186 LIST_HEAD(, ldm_partition) partitions;
187 LIST_ENTRY(ldm_component) entry;
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 */
197 LIST_HEAD(, ldm_component) components;
198 LIST_ENTRY(ldm_volume) entry;
202 uint64_t id; /* object id */
203 struct uuid guid; /* disk guid */
205 LIST_ENTRY(ldm_disk) entry;
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 */
214 LIST_ENTRY(ldm_disk_group) entry;
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 */
226 LIST_ENTRY(ldm_partition) entry;
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
242 * 0x18+ PN object name
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.
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
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 */
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
270 uint8_t type; /* VBLK type */
273 struct ldm_volume vol;
274 struct ldm_component comp;
275 struct ldm_disk disk;
276 struct ldm_partition part;
278 struct ldm_disk_group disk_group;
281 LIST_ENTRY(ldm_vblk) entry;
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).
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 */
296 LIST_ENTRY(ldm_xvblk) entry;
299 /* The internal representation of LDM database. */
301 struct ldm_privhdr ph; /* private header */
302 struct ldm_tochdr th; /* TOC header */
303 struct ldm_vmdbhdr dh; /* VMDB header */
305 LIST_HEAD(, ldm_volume) volumes;
306 LIST_HEAD(, ldm_disk) disks;
307 LIST_HEAD(, ldm_vblk) vblks;
308 LIST_HEAD(, ldm_xvblk) xvblks;
311 static struct uuid gpt_uuid_ms_ldm_metadata = GPT_ENT_TYPE_MS_LDM_METADATA;
313 struct g_part_ldm_table {
314 struct g_part_table base;
318 struct g_part_ldm_entry {
319 struct g_part_entry base;
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 *,
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 *,
339 static int g_part_ldm_write(struct g_part_table *, struct g_consumer *);
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),
357 static struct g_part_scheme g_part_ldm_scheme = {
360 sizeof(struct g_part_ldm_table),
361 .gps_entrysz = sizeof(struct g_part_ldm_entry)
363 G_PART_SCHEME_DECLARE(g_part_ldm);
365 static struct g_part_ldm_alias {
368 } ldm_alias_match[] = {
369 { DOSPTYP_NTFS, G_PART_ALIAS_MS_NTFS },
370 { DOSPTYP_FAT32, G_PART_ALIAS_MS_FAT32 },
371 { DOSPTYP_386BSD, G_PART_ALIAS_FREEBSD },
372 { DOSPTYP_LDM, G_PART_ALIAS_MS_LDM_DATA },
373 { DOSPTYP_LINSWP, G_PART_ALIAS_LINUX_SWAP },
374 { DOSPTYP_LINUX, G_PART_ALIAS_LINUX_DATA },
375 { DOSPTYP_LINLVM, G_PART_ALIAS_LINUX_LVM },
376 { DOSPTYP_LINRAID, G_PART_ALIAS_LINUX_RAID },
380 ldm_privhdr_read(struct g_consumer *cp, uint64_t off, int *error)
382 struct g_provider *pp;
386 buf = g_read_data(cp, off, pp->sectorsize, error);
390 if (memcmp(buf, LDM_PH_SIGN, strlen(LDM_PH_SIGN)) != 0) {
391 LDM_DEBUG(1, "%s: invalid LDM private header signature",
401 ldm_privhdr_parse(struct g_consumer *cp, struct ldm_privhdr *hdr,
407 memset(hdr, 0, sizeof(*hdr));
408 version = be32dec(buf + LDM_PH_VERSION_OFF);
409 if (version != LDM_VERSION_2K &&
410 version != LDM_VERSION_VISTA) {
411 LDM_DEBUG(0, "%s: unsupported LDM version %u.%u",
412 cp->provider->name, version >> 16,
416 error = parse_uuid(buf + LDM_PH_DISKGUID_OFF, &hdr->disk_guid);
419 error = parse_uuid(buf + LDM_PH_DGGUID_OFF, &hdr->dg_guid);
422 strncpy(hdr->dg_name, buf + LDM_PH_DGNAME_OFF, sizeof(hdr->dg_name));
423 hdr->start = be64dec(buf + LDM_PH_START_OFF);
424 hdr->size = be64dec(buf + LDM_PH_SIZE_OFF);
425 hdr->db_offset = be64dec(buf + LDM_PH_DB_OFF);
426 hdr->db_size = be64dec(buf + LDM_PH_DBSIZE_OFF);
427 hdr->th_offset[0] = be64dec(buf + LDM_PH_TH1_OFF);
428 hdr->th_offset[1] = be64dec(buf + LDM_PH_TH2_OFF);
429 hdr->conf_size = be64dec(buf + LDM_PH_CONFSIZE_OFF);
430 hdr->log_size = be64dec(buf + LDM_PH_LOGSIZE_OFF);
435 ldm_privhdr_check(struct ldm_db *db, struct g_consumer *cp, int is_gpt)
437 struct g_consumer *cp2;
438 struct g_provider *pp;
439 struct ldm_privhdr hdr;
440 uint64_t offset, last;
447 * The last LBA is used in several checks below, for the
448 * GPT case it should be calculated relative to the whole
451 cp2 = LIST_FIRST(&pp->geom->consumer);
453 cp2->provider->mediasize / cp2->provider->sectorsize - 1;
455 last = pp->mediasize / pp->sectorsize - 1;
456 for (found = 0, i = is_gpt;
457 i < sizeof(ldm_ph_off) / sizeof(ldm_ph_off[0]); i++) {
458 offset = ldm_ph_off[i];
460 * In the GPT case consumer is attached to the LDM metadata
461 * partition and we don't need add db_offset.
464 offset += db->ph.db_offset;
465 if (i == LDM_PH_MBRINDEX) {
467 * Prepare to errors and setup new base offset
468 * to read backup private headers. Assume that LDM
469 * database is in the last 1Mbyte area.
471 db->ph.db_offset = last - LDM_DB_SIZE;
473 buf = ldm_privhdr_read(cp, offset * pp->sectorsize, &error);
475 LDM_DEBUG(1, "%s: failed to read private header "
476 "%d at LBA %ju", pp->name, i, (uintmax_t)offset);
479 error = ldm_privhdr_parse(cp, &hdr, buf);
481 LDM_DEBUG(1, "%s: failed to parse private "
482 "header %d", pp->name, i);
483 LDM_DUMP(buf, pp->sectorsize);
488 if (hdr.start > last ||
489 hdr.start + hdr.size - 1 > last ||
490 (hdr.start + hdr.size - 1 > hdr.db_offset && !is_gpt) ||
491 hdr.db_size != LDM_DB_SIZE ||
492 hdr.db_offset + LDM_DB_SIZE - 1 > last ||
493 hdr.th_offset[0] >= LDM_DB_SIZE ||
494 hdr.th_offset[1] >= LDM_DB_SIZE ||
495 hdr.conf_size + hdr.log_size >= LDM_DB_SIZE) {
496 LDM_DEBUG(1, "%s: invalid values in the "
497 "private header %d", pp->name, i);
498 LDM_DEBUG(2, "%s: start: %jd, size: %jd, "
499 "db_offset: %jd, db_size: %jd, th_offset0: %jd, "
500 "th_offset1: %jd, conf_size: %jd, log_size: %jd, "
501 "last: %jd", pp->name, hdr.start, hdr.size,
502 hdr.db_offset, hdr.db_size, hdr.th_offset[0],
503 hdr.th_offset[1], hdr.conf_size, hdr.log_size,
507 if (found != 0 && memcmp(&db->ph, &hdr, sizeof(hdr)) != 0) {
508 LDM_DEBUG(0, "%s: private headers are not equal",
512 * We have different headers in the LDM.
513 * We can not trust this metadata.
515 LDM_DEBUG(0, "%s: refuse LDM metadata",
520 * We already have read primary private header
521 * and it differs from this backup one.
522 * Prefer the backup header and save it.
527 memcpy(&db->ph, &hdr, sizeof(hdr));
531 LDM_DEBUG(1, "%s: valid LDM private header not found",
539 ldm_gpt_check(struct ldm_db *db, struct g_consumer *cp)
541 struct g_part_table *gpt;
542 struct g_part_entry *e;
543 struct g_consumer *cp2;
546 cp2 = LIST_NEXT(cp, consumer);
548 gpt = cp->provider->geom->softc;
550 LIST_FOREACH(e, &gpt->gpt_entry, gpe_entry) {
551 if (cp->provider == e->gpe_pp) {
552 /* ms-ldm-metadata partition */
553 if (e->gpe_start != db->ph.db_offset ||
554 e->gpe_end != db->ph.db_offset + LDM_DB_SIZE - 1)
556 } else if (cp2->provider == e->gpe_pp) {
557 /* ms-ldm-data partition */
558 if (e->gpe_start != db->ph.start ||
559 e->gpe_end != db->ph.start + db->ph.size - 1)
563 LDM_DEBUG(0, "%s: GPT partition %d boundaries "
564 "do not match with the LDM metadata",
565 e->gpe_pp->name, e->gpe_index);
575 ldm_tochdr_check(struct ldm_db *db, struct g_consumer *cp)
577 struct g_provider *pp;
578 struct ldm_tochdr hdr;
579 uint64_t offset, conf_size, log_size;
584 for (i = 0, found = 0; i < LDM_TH_COUNT; i++) {
585 offset = db->ph.db_offset + db->ph.th_offset[i];
586 buf = g_read_data(cp,
587 offset * pp->sectorsize, pp->sectorsize, &error);
589 LDM_DEBUG(1, "%s: failed to read TOC header "
590 "at LBA %ju", pp->name, (uintmax_t)offset);
593 if (memcmp(buf, LDM_TH_SIGN, strlen(LDM_TH_SIGN)) != 0 ||
594 memcmp(buf + LDM_TH_NAME1_OFF, LDM_TH_NAME1,
595 strlen(LDM_TH_NAME1)) != 0 ||
596 memcmp(buf + LDM_TH_NAME2_OFF, LDM_TH_NAME2,
597 strlen(LDM_TH_NAME2)) != 0) {
598 LDM_DEBUG(1, "%s: failed to parse TOC header "
599 "at LBA %ju", pp->name, (uintmax_t)offset);
600 LDM_DUMP(buf, pp->sectorsize);
604 hdr.conf_offset = be64dec(buf + LDM_TH_CONF_OFF);
605 hdr.log_offset = be64dec(buf + LDM_TH_LOG_OFF);
606 conf_size = be64dec(buf + LDM_TH_CONFSIZE_OFF);
607 log_size = be64dec(buf + LDM_TH_LOGSIZE_OFF);
608 if (conf_size != db->ph.conf_size ||
609 hdr.conf_offset + conf_size >= LDM_DB_SIZE ||
610 log_size != db->ph.log_size ||
611 hdr.log_offset + log_size >= LDM_DB_SIZE) {
612 LDM_DEBUG(1, "%s: invalid values in the "
613 "TOC header at LBA %ju", pp->name,
615 LDM_DUMP(buf, pp->sectorsize);
621 memcpy(&db->th, &hdr, sizeof(hdr));
625 LDM_DEBUG(0, "%s: valid LDM TOC header not found.",
633 ldm_vmdbhdr_check(struct ldm_db *db, struct g_consumer *cp)
635 struct g_provider *pp;
643 offset = db->ph.db_offset + db->th.conf_offset;
644 buf = g_read_data(cp, offset * pp->sectorsize, pp->sectorsize,
647 LDM_DEBUG(0, "%s: failed to read VMDB header at "
648 "LBA %ju", pp->name, (uintmax_t)offset);
651 if (memcmp(buf, LDM_VMDB_SIGN, strlen(LDM_VMDB_SIGN)) != 0) {
653 LDM_DEBUG(0, "%s: failed to parse VMDB header at "
654 "LBA %ju", pp->name, (uintmax_t)offset);
658 version = be32dec(buf + LDM_DB_VERSION_OFF);
659 if (version != 0x4000A) {
661 LDM_DEBUG(0, "%s: unsupported VMDB version %u.%u",
662 pp->name, version >> 16, version & 0xFFFF);
666 * Check VMDB update status:
667 * 1 - in a consistent state;
668 * 2 - in a creation phase;
669 * 3 - in a deletion phase;
671 if (be16dec(buf + LDM_DB_STATUS_OFF) != 1) {
673 LDM_DEBUG(0, "%s: VMDB is not in a consistent state",
677 db->dh.last_seq = be32dec(buf + LDM_DB_LASTSEQ_OFF);
678 db->dh.size = be32dec(buf + LDM_DB_SIZE_OFF);
679 error = parse_uuid(buf + LDM_DB_DGGUID_OFF, &dg_guid);
680 /* Compare disk group name and guid from VMDB and private headers */
681 if (error != 0 || db->dh.size == 0 ||
682 pp->sectorsize % db->dh.size != 0 ||
683 strncmp(buf + LDM_DB_DGNAME_OFF, db->ph.dg_name, 31) != 0 ||
684 memcmp(&dg_guid, &db->ph.dg_guid, sizeof(dg_guid)) != 0 ||
685 db->dh.size * db->dh.last_seq >
686 db->ph.conf_size * pp->sectorsize) {
687 LDM_DEBUG(0, "%s: invalid values in the VMDB header",
689 LDM_DUMP(buf, pp->sectorsize);
698 ldm_xvblk_handle(struct ldm_db *db, struct ldm_vblkhdr *vh, const u_char *p)
700 struct ldm_xvblk *blk;
703 size = db->dh.size - 16;
704 LIST_FOREACH(blk, &db->xvblks, entry)
705 if (blk->group == vh->group)
708 blk = g_malloc(sizeof(*blk), M_WAITOK | M_ZERO);
709 blk->group = vh->group;
710 blk->size = size * vh->count + 16;
711 blk->data = g_malloc(blk->size, M_WAITOK | M_ZERO);
712 blk->map = 0xFF << vh->count;
713 LIST_INSERT_HEAD(&db->xvblks, blk, entry);
715 if ((blk->map & (1 << vh->index)) != 0) {
716 /* Block with given index has been already saved. */
719 /* Copy the data block to the place related to index. */
720 memcpy(blk->data + size * vh->index + 16, p + 16, size);
721 blk->map |= 1 << vh->index;
725 /* Read the variable-width numeric field and return new offset */
727 ldm_vnum_get(const u_char *buf, int offset, uint64_t *result, size_t range)
733 if (len > sizeof(uint64_t) || len + offset >= range)
735 for (num = 0; len > 0; len--)
736 num = (num << 8) | buf[offset++];
741 /* Read the variable-width string and return new offset */
743 ldm_vstr_get(const u_char *buf, int offset, u_char *result,
744 size_t maxlen, size_t range)
749 if (len >= maxlen || len + offset >= range)
751 memcpy(result, buf + offset, len);
753 return (offset + len);
756 /* Just skip the variable-width variable and return new offset */
758 ldm_vparm_skip(const u_char *buf, int offset, size_t range)
763 if (offset + len >= range)
766 return (offset + len);
770 ldm_vblk_handle(struct ldm_db *db, const u_char *p, size_t size)
772 struct ldm_vblk *blk;
773 struct ldm_volume *volume, *last;
778 blk = g_malloc(sizeof(*blk), M_WAITOK | M_ZERO);
779 blk->type = p[LDM_VBLK_TYPE_OFF];
780 offset = ldm_vnum_get(p, LDM_VBLK_OID_OFF, &blk->u.id, size);
782 errstr = "object id";
785 offset = ldm_vstr_get(p, offset, vstr, sizeof(vstr), size);
787 errstr = "object name";
792 * Component VBLK fields:
793 * Offset Size Description
794 * ------------+-------+------------------------
795 * 0x18+ PS volume state
796 * 0x18+5 PN component children count
797 * 0x1D+16 PN parent's volume object id
798 * 0x2D+1 PN stripe size
800 case LDM_VBLK_T_COMPONENT:
801 offset = ldm_vparm_skip(p, offset, size);
803 errstr = "volume state";
806 offset = ldm_vparm_skip(p, offset + 5, size);
808 errstr = "children count";
811 offset = ldm_vnum_get(p, offset + 16,
812 &blk->u.comp.vol_id, size);
814 errstr = "volume id";
819 * Partition VBLK fields:
820 * Offset Size Description
821 * ------------+-------+------------------------
822 * 0x18+12 8 partition start offset
823 * 0x18+20 8 volume offset
824 * 0x18+28 PN partition size
825 * 0x34+ PN parent's component object id
826 * 0x34+ PN disk's object id
828 case LDM_VBLK_T_PARTITION:
829 if (offset + 28 >= size) {
830 errstr = "too small buffer";
833 blk->u.part.start = be64dec(p + offset + 12);
834 blk->u.part.offset = be64dec(p + offset + 20);
835 offset = ldm_vnum_get(p, offset + 28, &blk->u.part.size, size);
837 errstr = "partition size";
840 offset = ldm_vnum_get(p, offset, &blk->u.part.comp_id, size);
842 errstr = "component id";
845 offset = ldm_vnum_get(p, offset, &blk->u.part.disk_id, size);
853 * Offset Size Description
854 * ------------+-------+------------------------
857 case LDM_VBLK_T_DISK:
858 errstr = "disk guid";
859 offset = ldm_vstr_get(p, offset, vstr, sizeof(vstr), size);
862 error = parse_uuid(vstr, &blk->u.disk.guid);
865 LIST_INSERT_HEAD(&db->disks, &blk->u.disk, entry);
868 * Disk group VBLK fields:
869 * Offset Size Description
870 * ------------+-------+------------------------
871 * 0x18+ PS disk group GUID
873 case LDM_VBLK_T_DISKGROUP:
875 strncpy(blk->u.disk_group.name, vstr,
876 sizeof(blk->u.disk_group.name));
877 offset = ldm_vstr_get(p, offset, vstr, sizeof(vstr), size);
879 errstr = "disk group guid";
882 error = parse_uuid(name, &blk->u.disk_group.guid);
884 errstr = "disk group guid";
887 LIST_INSERT_HEAD(&db->groups, &blk->u.disk_group, entry);
892 * Offset Size Description
893 * ------------+-------+------------------------
896 case LDM_VBLK_T_DISK4:
897 be_uuid_dec(p + offset, &blk->u.disk.guid);
898 LIST_INSERT_HEAD(&db->disks, &blk->u.disk, entry);
901 * Disk group VBLK fields:
902 * Offset Size Description
903 * ------------+-------+------------------------
906 case LDM_VBLK_T_DISKGROUP4:
908 strncpy(blk->u.disk_group.name, vstr,
909 sizeof(blk->u.disk_group.name));
910 be_uuid_dec(p + offset, &blk->u.disk.guid);
911 LIST_INSERT_HEAD(&db->groups, &blk->u.disk_group, entry);
915 * Volume VBLK fields:
916 * Offset Size Description
917 * ------------+-------+------------------------
918 * 0x18+ PS volume type
920 * 0x18+ 14(S) volume state
921 * 0x18+16 1 volume number
922 * 0x18+21 PN volume children count
923 * 0x2D+16 PN volume size
924 * 0x3D+4 1 partition type
926 case LDM_VBLK_T_VOLUME:
927 offset = ldm_vparm_skip(p, offset, size);
929 errstr = "volume type";
932 offset = ldm_vparm_skip(p, offset, size);
934 errstr = "unknown param";
937 if (offset + 21 >= size) {
938 errstr = "too small buffer";
941 blk->u.vol.number = p[offset + 16];
942 offset = ldm_vparm_skip(p, offset + 21, size);
944 errstr = "children count";
947 offset = ldm_vnum_get(p, offset + 16, &blk->u.vol.size, size);
949 errstr = "volume size";
952 if (offset + 4 >= size) {
953 errstr = "too small buffer";
956 blk->u.vol.part_type = p[offset + 4];
957 /* keep volumes ordered by volume number */
959 LIST_FOREACH(volume, &db->volumes, entry) {
960 if (volume->number > blk->u.vol.number)
965 LIST_INSERT_AFTER(last, &blk->u.vol, entry);
967 LIST_INSERT_HEAD(&db->volumes, &blk->u.vol, entry);
970 LDM_DEBUG(1, "unknown VBLK type 0x%02x\n", blk->type);
973 LIST_INSERT_HEAD(&db->vblks, blk, entry);
976 LDM_DEBUG(0, "failed to parse '%s' in VBLK of type 0x%02x\n",
984 ldm_vmdb_free(struct ldm_db *db)
986 struct ldm_vblk *vblk;
987 struct ldm_xvblk *xvblk;
989 while (!LIST_EMPTY(&db->xvblks)) {
990 xvblk = LIST_FIRST(&db->xvblks);
991 LIST_REMOVE(xvblk, entry);
995 while (!LIST_EMPTY(&db->vblks)) {
996 vblk = LIST_FIRST(&db->vblks);
997 LIST_REMOVE(vblk, entry);
1003 ldm_vmdb_parse(struct ldm_db *db, struct g_consumer *cp)
1005 struct g_provider *pp;
1006 struct ldm_vblk *vblk;
1007 struct ldm_xvblk *xvblk;
1008 struct ldm_volume *volume;
1009 struct ldm_component *comp;
1010 struct ldm_vblkhdr vh;
1012 size_t size, n, sectors;
1017 size = (db->dh.last_seq * db->dh.size +
1018 pp->sectorsize - 1) / 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;
1025 buf = g_read_data(cp, offset * pp->sectorsize,
1026 sectors * pp->sectorsize, &error);
1028 LDM_DEBUG(0, "%s: failed to read VBLK\n",
1032 for (p = buf; p < buf + sectors * pp->sectorsize;
1034 if (memcmp(p, LDM_VBLK_SIGN,
1035 strlen(LDM_VBLK_SIGN)) != 0) {
1036 LDM_DEBUG(0, "%s: no VBLK signature\n",
1038 LDM_DUMP(p, db->dh.size);
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)
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);
1056 error = ldm_xvblk_handle(db, &vh, p);
1058 LDM_DEBUG(0, "%s: xVBLK "
1059 "is corrupted\n", pp->name);
1060 LDM_DUMP(p, db->dh.size);
1065 if (be16dec(p + 16) != 0)
1066 LDM_DEBUG(1, "%s: VBLK update"
1067 " status is %u\n", pp->name,
1069 error = ldm_vblk_handle(db, p, db->dh.size);
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);
1084 LDM_DEBUG(0, "%s: incomplete or corrupt "
1085 "xVBLK found\n", pp->name);
1088 LIST_REMOVE(xvblk, entry);
1089 g_free(xvblk->data);
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);
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);
1118 g_part_ldm_add(struct g_part_table *basetable, struct g_part_entry *baseentry,
1119 struct g_part_parms *gpp)
1126 g_part_ldm_bootcode(struct g_part_table *basetable, struct g_part_parms *gpp)
1133 g_part_ldm_create(struct g_part_table *basetable, struct g_part_parms *gpp)
1140 g_part_ldm_destroy(struct g_part_table *basetable, struct g_part_parms *gpp)
1142 struct g_part_ldm_table *table;
1143 struct g_provider *pp;
1145 table = (struct g_part_ldm_table *)basetable;
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
1153 pp = LIST_FIRST(&basetable->gpt_gp->consumer)->provider;
1155 * To destroy LDM we should wipe MBR, first private header and
1156 * backup private headers.
1158 basetable->gpt_smhead = (1 << ldm_ph_off[0]) | 1;
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.
1164 if (table->db_offset + LDM_DB_SIZE ==
1165 pp->mediasize / pp->sectorsize)
1166 basetable->gpt_smtail = 1;
1171 g_part_ldm_dumpconf(struct g_part_table *basetable,
1172 struct g_part_entry *baseentry, struct sbuf *sb, const char *indent)
1174 struct g_part_ldm_entry *entry;
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,
1185 /* confxml: scheme information */
1190 g_part_ldm_dumpto(struct g_part_table *table, struct g_part_entry *baseentry)
1197 g_part_ldm_modify(struct g_part_table *basetable,
1198 struct g_part_entry *baseentry, struct g_part_parms *gpp)
1205 g_part_ldm_name(struct g_part_table *table, struct g_part_entry *baseentry,
1206 char *buf, size_t bufsz)
1209 snprintf(buf, bufsz, "s%d", baseentry->gpe_index);
1214 ldm_gpt_probe(struct g_part_table *basetable, struct g_consumer *cp)
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;
1225 * XXX: We use some knowlege about GEOM_PART_GPT internal
1226 * structures, but it is easier than parse GPT by himself.
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)
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);
1242 g_destroy_consumer(cp2);
1243 g_topology_unlock();
1246 error = g_access(cp2, 1, 0, 0);
1249 g_destroy_consumer(cp2);
1250 g_topology_unlock();
1253 g_topology_unlock();
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,
1262 table = (struct g_part_ldm_table *)basetable;
1265 return (G_PART_PROBE_PRI_HIGH);
1268 /* second consumer is no longer needed. */
1270 g_access(cp2, -1, 0, 0);
1272 g_destroy_consumer(cp2);
1275 g_topology_unlock();
1280 g_part_ldm_probe(struct g_part_table *basetable, struct g_consumer *cp)
1282 struct g_provider *pp;
1283 u_char *buf, type[64];
1288 if (pp->sectorsize != 512)
1291 error = g_getattr("PART::scheme", cp, &type);
1292 if (error == 0 && strcmp(type, "GPT") == 0) {
1293 if (g_getattr("PART::type", cp, &type) != 0 ||
1294 strcmp(type, "ms-ldm-data") != 0)
1296 error = ldm_gpt_probe(basetable, cp);
1300 if (basetable->gpt_depth != 0)
1303 /* LDM has 1M metadata area */
1304 if (pp->mediasize <= 1024 * 1024)
1307 /* Check that there's a MBR */
1308 buf = g_read_data(cp, 0, pp->sectorsize, &error);
1312 if (le16dec(buf + DOSMAGICOFFSET) != DOSMAGIC) {
1317 /* Check that we have LDM partitions in the MBR */
1318 for (idx = 0; idx < NDOSPART && error != 0; idx++) {
1319 if (buf[DOSPARTOFF + idx * DOSPARTSIZE + 4] == DOSPTYP_LDM)
1324 LDM_DEBUG(2, "%s: LDM data partitions found in MBR",
1326 /* Read the LDM private header */
1327 buf = ldm_privhdr_read(cp,
1328 ldm_ph_off[LDM_PH_MBRINDEX] * pp->sectorsize, &error);
1332 return (G_PART_PROBE_PRI_HIGH);
1338 g_part_ldm_read(struct g_part_table *basetable, struct g_consumer *cp)
1340 struct g_part_ldm_table *table;
1341 struct g_part_ldm_entry *entry;
1342 struct g_consumer *cp2;
1343 struct ldm_component *comp;
1344 struct ldm_partition *part;
1345 struct ldm_volume *vol;
1346 struct ldm_disk *disk;
1348 int error, index, skipped;
1350 table = (struct g_part_ldm_table *)basetable;
1351 memset(&db, 0, sizeof(db));
1352 cp2 = cp; /* ms-ldm-data */
1354 cp = LIST_FIRST(&cp->geom->consumer); /* ms-ldm-metadata */
1355 /* Read and parse LDM private headers. */
1356 error = ldm_privhdr_check(&db, cp, table->is_gpt);
1359 basetable->gpt_first = table->is_gpt ? 0: db.ph.start;
1360 basetable->gpt_last = basetable->gpt_first + db.ph.size - 1;
1361 table->db_offset = db.ph.db_offset;
1362 /* Make additional checks for GPT */
1363 if (table->is_gpt) {
1364 error = ldm_gpt_check(&db, cp);
1368 * Now we should reset database offset to zero, because our
1369 * consumer cp is attached to the ms-ldm-metadata partition
1370 * and we don't need add db_offset to read from it.
1372 db.ph.db_offset = 0;
1374 /* Read and parse LDM TOC headers. */
1375 error = ldm_tochdr_check(&db, cp);
1378 /* Read and parse LDM VMDB header. */
1379 error = ldm_vmdbhdr_check(&db, cp);
1382 error = ldm_vmdb_parse(&db, cp);
1384 * For the GPT case we must detach and destroy
1385 * second consumer before return.
1388 if (table->is_gpt) {
1390 g_access(cp, -1, 0, 0);
1392 g_destroy_consumer(cp);
1393 g_topology_unlock();
1398 /* Search current disk in the disk list. */
1399 LIST_FOREACH(disk, &db.disks, entry)
1400 if (memcmp(&disk->guid, &db.ph.disk_guid,
1401 sizeof(struct uuid)) == 0)
1404 LDM_DEBUG(1, "%s: no LDM volumes on this disk",
1405 cp->provider->name);
1410 LIST_FOREACH(vol, &db.volumes, entry) {
1411 LIST_FOREACH(comp, &vol->components, entry) {
1412 /* Skip volumes from different disks. */
1413 part = LIST_FIRST(&comp->partitions);
1414 if (part->disk_id != disk->id)
1417 /* We don't support spanned and striped volumes. */
1418 if (comp->count > 1 || part->offset != 0) {
1419 LDM_DEBUG(1, "%s: LDM volume component "
1420 "%ju has %u partitions. Skipped",
1421 cp->provider->name, (uintmax_t)comp->id,
1426 * Allow mirrored volumes only when they are explicitly
1427 * allowed with kern.geom.part.ldm.show_mirrors=1.
1429 if (vol->count > 1 && show_mirrors == 0) {
1430 LDM_DEBUG(1, "%s: LDM volume %ju has %u "
1431 "components. Skipped",
1432 cp->provider->name, (uintmax_t)vol->id,
1436 entry = (struct g_part_ldm_entry *)g_part_new_entry(
1438 basetable->gpt_first + part->start,
1439 basetable->gpt_first + part->start +
1442 * Mark skipped partition as ms-ldm-data partition.
1443 * We do not support them, but it is better to show
1444 * that we have something there, than just show
1448 entry->type = vol->part_type;
1450 entry->type = DOSPTYP_LDM;
1451 LDM_DEBUG(1, "%s: new volume id: %ju, start: %ju,"
1452 " end: %ju, type: 0x%02x\n", cp->provider->name,
1453 (uintmax_t)part->id,(uintmax_t)part->start +
1454 basetable->gpt_first, (uintmax_t)part->start +
1455 part->size + basetable->gpt_first - 1,
1464 g_part_ldm_type(struct g_part_table *basetable, struct g_part_entry *baseentry,
1465 char *buf, size_t bufsz)
1467 struct g_part_ldm_entry *entry;
1470 entry = (struct g_part_ldm_entry *)baseentry;
1472 i < sizeof(ldm_alias_match) / sizeof(ldm_alias_match[0]); i++) {
1473 if (ldm_alias_match[i].typ == entry->type)
1474 return (g_part_alias_name(ldm_alias_match[i].alias));
1476 snprintf(buf, bufsz, "!%d", entry->type);
1481 g_part_ldm_write(struct g_part_table *basetable, struct g_consumer *cp)