1/*- 2 * Copyright (c) 2012 Andrey V. Elsukov <ae@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 */ 26 27#include <sys/cdefs.h> 28__FBSDID("$FreeBSD$"); 29 30#include <sys/param.h> 31#include <sys/bio.h> 32#include <sys/diskmbr.h> 33#include <sys/endian.h> 34#include <sys/gpt.h> 35#include <sys/kernel.h> 36#include <sys/kobj.h> 37#include <sys/limits.h> 38#include <sys/lock.h> 39#include <sys/malloc.h> 40#include <sys/mutex.h> 41#include <sys/queue.h> 42#include <sys/sbuf.h> 43#include <sys/systm.h> 44#include <sys/sysctl.h> 45#include <sys/uuid.h> 46#include <geom/geom.h> 47#include <geom/part/g_part.h> 48 49#include "g_part_if.h" 50 51FEATURE(geom_part_ldm, "GEOM partitioning class for LDM support"); 52 53SYSCTL_DECL(_kern_geom_part); 54static SYSCTL_NODE(_kern_geom_part, OID_AUTO, ldm, CTLFLAG_RW, 0, 55 "GEOM_PART_LDM Logical Disk Manager"); 56 57static u_int ldm_debug = 0; 58TUNABLE_INT("kern.geom.part.ldm.debug", &ldm_debug); 59SYSCTL_UINT(_kern_geom_part_ldm, OID_AUTO, debug, 60 CTLFLAG_RW | CTLFLAG_TUN, &ldm_debug, 0, "Debug level"); 61 62/* 63 * This allows access to mirrored LDM volumes. Since we do not 64 * doing mirroring here, it is not enabled by default. 65 */ 66static u_int show_mirrors = 0; 67TUNABLE_INT("kern.geom.part.ldm.show_mirrors", &show_mirrors); 68SYSCTL_UINT(_kern_geom_part_ldm, OID_AUTO, show_mirrors, 69 CTLFLAG_RW | CTLFLAG_TUN, &show_mirrors, 0, "Show mirrored volumes"); 70 71#define LDM_DEBUG(lvl, fmt, ...) do { \ 72 if (ldm_debug >= (lvl)) { \ 73 printf("GEOM_PART: " fmt "\n", __VA_ARGS__); \ 74 } \ 75} while (0) 76#define LDM_DUMP(buf, size) do { \ 77 if (ldm_debug > 1) { \ 78 hexdump(buf, size, NULL, 0); \ 79 } \ 80} while (0) 81 82/* 83 * There are internal representations of LDM structures. 84 * 85 * We do not keep all fields of on-disk structures, only most useful. 86 * All numbers in an on-disk structures are in big-endian format. 87 */ 88 89/* 90 * Private header is 512 bytes long. There are three copies on each disk. 91 * Offset and sizes are in sectors. Location of each copy: 92 * - the first offset is relative to the disk start; 93 * - the second and third offset are relative to the LDM database start. 94 * 95 * On a disk partitioned with GPT, the LDM has not first private header. 96 */ 97#define LDM_PH_MBRINDEX 0 98#define LDM_PH_GPTINDEX 2 99static 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" 115struct 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 144struct 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 159struct 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 183struct 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 192struct 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 203struct 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 211struct 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 220struct 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 257struct 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 271struct 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 */ 292struct 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. */ 302struct 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 313static struct uuid gpt_uuid_ms_ldm_metadata = GPT_ENT_TYPE_MS_LDM_METADATA; 314 315struct g_part_ldm_table { 316 struct g_part_table base; 317 uint64_t db_offset; 318 int is_gpt; 319}; 320struct g_part_ldm_entry { 321 struct g_part_entry base; 322 uint8_t type; 323}; 324 325static int g_part_ldm_add(struct g_part_table *, struct g_part_entry *, 326 struct g_part_parms *); 327static int g_part_ldm_bootcode(struct g_part_table *, struct g_part_parms *); 328static int g_part_ldm_create(struct g_part_table *, struct g_part_parms *); 329static int g_part_ldm_destroy(struct g_part_table *, struct g_part_parms *); 330static void g_part_ldm_dumpconf(struct g_part_table *, struct g_part_entry *, 331 struct sbuf *, const char *); 332static int g_part_ldm_dumpto(struct g_part_table *, struct g_part_entry *); 333static int g_part_ldm_modify(struct g_part_table *, struct g_part_entry *, 334 struct g_part_parms *); 335static const char *g_part_ldm_name(struct g_part_table *, struct g_part_entry *, 336 char *, size_t); 337static int g_part_ldm_probe(struct g_part_table *, struct g_consumer *); 338static int g_part_ldm_read(struct g_part_table *, struct g_consumer *); 339static const char *g_part_ldm_type(struct g_part_table *, struct g_part_entry *, 340 char *, size_t); 341static int g_part_ldm_write(struct g_part_table *, struct g_consumer *); 342 343static 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 359static 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}; 365G_PART_SCHEME_DECLARE(g_part_ldm); 366 367static struct g_part_ldm_alias { 368 u_char typ; 369 int 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 }, 379}; 380 381static u_char* 382ldm_privhdr_read(struct g_consumer *cp, uint64_t off, int *error) 383{ 384 struct g_provider *pp; 385 u_char *buf; 386 387 pp = cp->provider; 388 buf = g_read_data(cp, off, pp->sectorsize, error); 389 if (buf == NULL) 390 return (NULL); 391 392 if (memcmp(buf, LDM_PH_SIGN, strlen(LDM_PH_SIGN)) != 0) { 393 LDM_DEBUG(1, "%s: invalid LDM private header signature", 394 pp->name); 395 g_free(buf); 396 buf = NULL; 397 *error = EINVAL; 398 } 399 return (buf); 400} 401 402static int 403ldm_privhdr_parse(struct g_consumer *cp, struct ldm_privhdr *hdr, 404 const u_char *buf) 405{ 406 uint32_t version; 407 int error; 408 409 memset(hdr, 0, sizeof(*hdr)); 410 version = be32dec(buf + LDM_PH_VERSION_OFF); 411 if (version != LDM_VERSION_2K && 412 version != LDM_VERSION_VISTA) { 413 LDM_DEBUG(0, "%s: unsupported LDM version %u.%u", 414 cp->provider->name, version >> 16, 415 version & 0xFFFF); 416 return (ENXIO); 417 } 418 error = parse_uuid(buf + LDM_PH_DISKGUID_OFF, &hdr->disk_guid); 419 if (error != 0) 420 return (error); 421 error = parse_uuid(buf + LDM_PH_DGGUID_OFF, &hdr->dg_guid); 422 if (error != 0) 423 return (error); 424 strncpy(hdr->dg_name, buf + LDM_PH_DGNAME_OFF, sizeof(hdr->dg_name)); 425 hdr->start = be64dec(buf + LDM_PH_START_OFF); 426 hdr->size = be64dec(buf + LDM_PH_SIZE_OFF); 427 hdr->db_offset = be64dec(buf + LDM_PH_DB_OFF); 428 hdr->db_size = be64dec(buf + LDM_PH_DBSIZE_OFF); 429 hdr->th_offset[0] = be64dec(buf + LDM_PH_TH1_OFF); 430 hdr->th_offset[1] = be64dec(buf + LDM_PH_TH2_OFF); 431 hdr->conf_size = be64dec(buf + LDM_PH_CONFSIZE_OFF); 432 hdr->log_size = be64dec(buf + LDM_PH_LOGSIZE_OFF); 433 return (0); 434} 435 436static int 437ldm_privhdr_check(struct ldm_db *db, struct g_consumer *cp, int is_gpt) 438{ 439 struct g_consumer *cp2; 440 struct g_provider *pp; 441 struct ldm_privhdr hdr; 442 uint64_t offset, last; 443 int error, found, i; 444 u_char *buf; 445 446 pp = cp->provider; 447 if (is_gpt) { 448 /* 449 * The last LBA is used in several checks below, for the 450 * GPT case it should be calculated relative to the whole 451 * disk. 452 */ 453 cp2 = LIST_FIRST(&pp->geom->consumer); 454 last = 455 cp2->provider->mediasize / cp2->provider->sectorsize - 1; 456 } else 457 last = pp->mediasize / pp->sectorsize - 1; 458 for (found = 0, i = is_gpt; 459 i < sizeof(ldm_ph_off) / sizeof(ldm_ph_off[0]); i++) { 460 offset = ldm_ph_off[i]; 461 /* 462 * In the GPT case consumer is attached to the LDM metadata 463 * partition and we don't need add db_offset. 464 */ 465 if (!is_gpt) 466 offset += db->ph.db_offset; 467 if (i == LDM_PH_MBRINDEX) { 468 /* 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. 472 */ 473 db->ph.db_offset = last - LDM_DB_SIZE; 474 } 475 buf = ldm_privhdr_read(cp, offset * pp->sectorsize, &error); 476 if (buf == NULL) { 477 LDM_DEBUG(1, "%s: failed to read private header " 478 "%d at LBA %ju", pp->name, i, (uintmax_t)offset); 479 continue; 480 } 481 error = ldm_privhdr_parse(cp, &hdr, buf); 482 if (error != 0) { 483 LDM_DEBUG(1, "%s: failed to parse private " 484 "header %d", pp->name, i); 485 LDM_DUMP(buf, pp->sectorsize); 486 g_free(buf); 487 continue; 488 } 489 g_free(buf); 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, 506 last); 507 continue; 508 } 509 if (found != 0 && memcmp(&db->ph, &hdr, sizeof(hdr)) != 0) { 510 LDM_DEBUG(0, "%s: private headers are not equal", 511 pp->name); 512 if (i > 1) { 513 /* 514 * We have different headers in the LDM. 515 * We can not trust this metadata. 516 */ 517 LDM_DEBUG(0, "%s: refuse LDM metadata", 518 pp->name); 519 return (EINVAL); 520 } 521 /* 522 * We already have read primary private header 523 * and it differs from this backup one. 524 * Prefer the backup header and save it. 525 */ 526 found = 0; 527 } 528 if (found == 0) 529 memcpy(&db->ph, &hdr, sizeof(hdr)); 530 found = 1; 531 } 532 if (found == 0) { 533 LDM_DEBUG(1, "%s: valid LDM private header not found", 534 pp->name); 535 return (ENXIO); 536 } 537 return (0); 538} 539 540static int 541ldm_gpt_check(struct ldm_db *db, struct g_consumer *cp) 542{ 543 struct g_part_table *gpt; 544 struct g_part_entry *e; 545 struct g_consumer *cp2; 546 int error; 547 548 cp2 = LIST_NEXT(cp, consumer); 549 g_topology_lock(); 550 gpt = cp->provider->geom->softc; 551 error = 0; 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) 557 error++; 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) 562 error++; 563 } 564 if (error != 0) { 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); 568 error = ENXIO; 569 break; 570 } 571 } 572 g_topology_unlock(); 573 return (error); 574} 575 576static int 577ldm_tochdr_check(struct ldm_db *db, struct g_consumer *cp) 578{ 579 struct g_provider *pp; 580 struct ldm_tochdr hdr; 581 uint64_t offset, conf_size, log_size; 582 int error, found, i; 583 u_char *buf; 584 585 pp = cp->provider; 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); 590 if (buf == NULL) { 591 LDM_DEBUG(1, "%s: failed to read TOC header " 592 "at LBA %ju", pp->name, (uintmax_t)offset); 593 continue; 594 } 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); 603 g_free(buf); 604 continue; 605 } 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, 616 (uintmax_t)offset); 617 LDM_DUMP(buf, pp->sectorsize); 618 g_free(buf); 619 continue; 620 } 621 g_free(buf); 622 if (found == 0) 623 memcpy(&db->th, &hdr, sizeof(hdr)); 624 found = 1; 625 } 626 if (found == 0) { 627 LDM_DEBUG(0, "%s: valid LDM TOC header not found.", 628 pp->name); 629 return (ENXIO); 630 } 631 return (0); 632} 633 634static int 635ldm_vmdbhdr_check(struct ldm_db *db, struct g_consumer *cp) 636{ 637 struct g_provider *pp; 638 struct uuid dg_guid; 639 uint64_t offset; 640 uint32_t version; 641 int error; 642 u_char *buf; 643 644 pp = cp->provider; 645 offset = db->ph.db_offset + db->th.conf_offset; 646 buf = g_read_data(cp, offset * pp->sectorsize, pp->sectorsize, 647 &error); 648 if (buf == NULL) { 649 LDM_DEBUG(0, "%s: failed to read VMDB header at " 650 "LBA %ju", pp->name, (uintmax_t)offset); 651 return (error); 652 } 653 if (memcmp(buf, LDM_VMDB_SIGN, strlen(LDM_VMDB_SIGN)) != 0) { 654 g_free(buf); 655 LDM_DEBUG(0, "%s: failed to parse VMDB header at " 656 "LBA %ju", pp->name, (uintmax_t)offset); 657 return (ENXIO); 658 } 659 /* Check version. */ 660 version = be32dec(buf + LDM_DB_VERSION_OFF); 661 if (version != 0x4000A) { 662 g_free(buf); 663 LDM_DEBUG(0, "%s: unsupported VMDB version %u.%u", 664 pp->name, version >> 16, version & 0xFFFF); 665 return (ENXIO); 666 } 667 /* 668 * Check VMDB update status: 669 * 1 - in a consistent state; 670 * 2 - in a creation phase; 671 * 3 - in a deletion phase; 672 */ 673 if (be16dec(buf + LDM_DB_STATUS_OFF) != 1) { 674 g_free(buf); 675 LDM_DEBUG(0, "%s: VMDB is not in a consistent state", 676 pp->name); 677 return (ENXIO); 678 } 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", 690 pp->name); 691 LDM_DUMP(buf, pp->sectorsize); 692 g_free(buf); 693 return (EINVAL); 694 } 695 g_free(buf); 696 return (0); 697} 698 699static int 700ldm_xvblk_handle(struct ldm_db *db, struct ldm_vblkhdr *vh, const u_char *p) 701{ 702 struct ldm_xvblk *blk; 703 size_t size; 704 705 size = db->dh.size - 16; 706 LIST_FOREACH(blk, &db->xvblks, entry) 707 if (blk->group == vh->group) 708 break; 709 if (blk == NULL) { 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); 716 } 717 if ((blk->map & (1 << vh->index)) != 0) { 718 /* Block with given index has been already saved. */ 719 return (EINVAL); 720 } 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; 724 return (0); 725} 726 727/* Read the variable-width numeric field and return new offset */ 728static int 729ldm_vnum_get(const u_char *buf, int offset, uint64_t *result, size_t range) 730{ 731 uint64_t num; 732 uint8_t len; 733 734 len = buf[offset++]; 735 if (len > sizeof(uint64_t) || len + offset >= range) 736 return (-1); 737 for (num = 0; len > 0; len--) 738 num = (num << 8) | buf[offset++]; 739 *result = num; 740 return (offset); 741} 742 743/* Read the variable-width string and return new offset */ 744static int 745ldm_vstr_get(const u_char *buf, int offset, u_char *result, 746 size_t maxlen, size_t range) 747{ 748 uint8_t len; 749 750 len = buf[offset++]; 751 if (len >= maxlen || len + offset >= range) 752 return (-1); 753 memcpy(result, buf + offset, len); 754 result[len] = '\0'; 755 return (offset + len); 756} 757 758/* Just skip the variable-width variable and return new offset */ 759static int 760ldm_vparm_skip(const u_char *buf, int offset, size_t range) 761{ 762 uint8_t len; 763 764 len = buf[offset++]; 765 if (offset + len >= range) 766 return (-1); 767 768 return (offset + len); 769} 770 771static int 772ldm_vblk_handle(struct ldm_db *db, const u_char *p, size_t size) 773{ 774 struct ldm_vblk *blk; 775 struct ldm_volume *volume, *last; 776 const char *errstr; 777 u_char vstr[64]; 778 int error, offset; 779 780 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); 783 if (offset < 0) { 784 errstr = "object id"; 785 goto fail; 786 } 787 offset = ldm_vstr_get(p, offset, vstr, sizeof(vstr), size); 788 if (offset < 0) { 789 errstr = "object name"; 790 goto fail; 791 } 792 switch (blk->type) { 793 /* 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 801 */ 802 case LDM_VBLK_T_COMPONENT: 803 offset = ldm_vparm_skip(p, offset, size); 804 if (offset < 0) { 805 errstr = "volume state"; 806 goto fail; 807 } 808 offset = ldm_vparm_skip(p, offset + 5, size); 809 if (offset < 0) { 810 errstr = "children count"; 811 goto fail; 812 } 813 offset = ldm_vnum_get(p, offset + 16, 814 &blk->u.comp.vol_id, size); 815 if (offset < 0) { 816 errstr = "volume id"; 817 goto fail; 818 } 819 break; 820 /* 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 829 */ 830 case LDM_VBLK_T_PARTITION: 831 if (offset + 28 >= size) { 832 errstr = "too small buffer"; 833 goto fail; 834 } 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); 838 if (offset < 0) { 839 errstr = "partition size"; 840 goto fail; 841 } 842 offset = ldm_vnum_get(p, offset, &blk->u.part.comp_id, size); 843 if (offset < 0) { 844 errstr = "component id"; 845 goto fail; 846 } 847 offset = ldm_vnum_get(p, offset, &blk->u.part.disk_id, size); 848 if (offset < 0) { 849 errstr = "disk id"; 850 goto fail; 851 } 852 break; 853 /* 854 * Disk VBLK fields: 855 * Offset Size Description 856 * ------------+-------+------------------------ 857 * 0x18+ PS disk GUID 858 */ 859 case LDM_VBLK_T_DISK: 860 errstr = "disk guid"; 861 offset = ldm_vstr_get(p, offset, vstr, sizeof(vstr), size); 862 if (offset < 0) 863 goto fail; 864 error = parse_uuid(vstr, &blk->u.disk.guid); 865 if (error != 0) 866 goto fail; 867 LIST_INSERT_HEAD(&db->disks, &blk->u.disk, entry); 868 break; 869 /* 870 * Disk group VBLK fields: 871 * Offset Size Description 872 * ------------+-------+------------------------ 873 * 0x18+ PS disk group GUID 874 */ 875 case LDM_VBLK_T_DISKGROUP: 876#if 0 877 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); 880 if (offset < 0) { 881 errstr = "disk group guid"; 882 goto fail; 883 } 884 error = parse_uuid(name, &blk->u.disk_group.guid); 885 if (error != 0) { 886 errstr = "disk group guid"; 887 goto fail; 888 } 889 LIST_INSERT_HEAD(&db->groups, &blk->u.disk_group, entry); 890#endif 891 break; 892 /* 893 * Disk VBLK fields: 894 * Offset Size Description 895 * ------------+-------+------------------------ 896 * 0x18+ 16 disk GUID 897 */ 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); 901 break; 902 /* 903 * Disk group VBLK fields: 904 * Offset Size Description 905 * ------------+-------+------------------------ 906 * 0x18+ 16 disk GUID 907 */ 908 case LDM_VBLK_T_DISKGROUP4: 909#if 0 910 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); 914#endif 915 break; 916 /* 917 * Volume VBLK fields: 918 * Offset Size Description 919 * ------------+-------+------------------------ 920 * 0x18+ PS volume type 921 * 0x18+ PS unknown 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 927 */ 928 case LDM_VBLK_T_VOLUME: 929 offset = ldm_vparm_skip(p, offset, size); 930 if (offset < 0) { 931 errstr = "volume type"; 932 goto fail; 933 } 934 offset = ldm_vparm_skip(p, offset, size); 935 if (offset < 0) { 936 errstr = "unknown param"; 937 goto fail; 938 } 939 if (offset + 21 >= size) { 940 errstr = "too small buffer"; 941 goto fail; 942 } 943 blk->u.vol.number = p[offset + 16]; 944 offset = ldm_vparm_skip(p, offset + 21, size); 945 if (offset < 0) { 946 errstr = "children count"; 947 goto fail; 948 } 949 offset = ldm_vnum_get(p, offset + 16, &blk->u.vol.size, size); 950 if (offset < 0) { 951 errstr = "volume size"; 952 goto fail; 953 } 954 if (offset + 4 >= size) { 955 errstr = "too small buffer"; 956 goto fail; 957 } 958 blk->u.vol.part_type = p[offset + 4]; 959 /* keep volumes ordered by volume number */ 960 last = NULL; 961 LIST_FOREACH(volume, &db->volumes, entry) { 962 if (volume->number > blk->u.vol.number) 963 break; 964 last = volume; 965 } 966 if (last != NULL) 967 LIST_INSERT_AFTER(last, &blk->u.vol, entry); 968 else 969 LIST_INSERT_HEAD(&db->volumes, &blk->u.vol, entry); 970 break; 971 default: 972 LDM_DEBUG(1, "unknown VBLK type 0x%02x\n", blk->type); 973 LDM_DUMP(p, size); 974 } 975 LIST_INSERT_HEAD(&db->vblks, blk, entry); 976 return (0); 977fail: 978 LDM_DEBUG(0, "failed to parse '%s' in VBLK of type 0x%02x\n", 979 errstr, blk->type); 980 LDM_DUMP(p, size); 981 g_free(blk); 982 return (EINVAL); 983} 984 985static void 986ldm_vmdb_free(struct ldm_db *db) 987{ 988 struct ldm_vblk *vblk; 989 struct ldm_xvblk *xvblk; 990 991 while (!LIST_EMPTY(&db->xvblks)) { 992 xvblk = LIST_FIRST(&db->xvblks); 993 LIST_REMOVE(xvblk, entry); 994 g_free(xvblk->data); 995 g_free(xvblk); 996 } 997 while (!LIST_EMPTY(&db->vblks)) { 998 vblk = LIST_FIRST(&db->vblks); 999 LIST_REMOVE(vblk, entry); 1000 g_free(vblk); 1001 } 1002} 1003 1004static int 1005ldm_vmdb_parse(struct ldm_db *db, struct g_consumer *cp) 1006{ 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; 1013 u_char *buf, *p; 1014 size_t size, n, sectors; 1015 uint64_t offset; 1016 int error; 1017 1018 pp = cp->provider; 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; 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); 1113fail: 1114 ldm_vmdb_free(db); 1115 g_free(buf); 1116 return (ENXIO); 1117} 1118 1119static int 1120g_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 1127static int 1128g_part_ldm_bootcode(struct g_part_table *basetable, struct g_part_parms *gpp) 1129{ 1130 1131 return (ENOSYS); 1132} 1133 1134static int 1135g_part_ldm_create(struct g_part_table *basetable, struct g_part_parms *gpp) 1136{ 1137 1138 return (ENOSYS); 1139} 1140 1141static int 1142g_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 1172static void 1173g_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 1191static int 1192g_part_ldm_dumpto(struct g_part_table *table, struct g_part_entry *baseentry) 1193{ 1194 1195 return (0); 1196} 1197 1198static int 1199g_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 1206static const char * 1207g_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 1215static int 1216ldm_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 knowlege 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 1281static int 1282g_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 1339static int 1340g_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 */ 1389gpt_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 1465static const char * 1466g_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; 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)); 1477 } 1478 snprintf(buf, bufsz, "!%d", entry->type); 1479 return (buf); 1480} 1481 1482static int 1483g_part_ldm_write(struct g_part_table *basetable, struct g_consumer *cp) 1484{ 1485 1486 return (ENOSYS); 1487} 1488