1 /*- 2 * Copyright (c) 2004-2006 Pawel Jakub Dawidek <pjd@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 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 * 26 * $FreeBSD$ 27 */ 28 29 #ifndef _G_MIRROR_H_ 30 #define _G_MIRROR_H_ 31 32 #include <sys/endian.h> 33 #include <sys/md5.h> 34 35 #define G_MIRROR_CLASS_NAME "MIRROR" 36 37 #define G_MIRROR_MAGIC "GEOM::MIRROR" 38 /* 39 * Version history: 40 * 0 - Initial version number. 41 * 1 - Added 'prefer' balance algorithm. 42 * 2 - Added md_genid field to metadata. 43 * 3 - Added md_provsize field to metadata. 44 * 4 - Added 'no failure synchronization' flag. 45 */ 46 #define G_MIRROR_VERSION 4 47 48 #define G_MIRROR_BALANCE_NONE 0 49 #define G_MIRROR_BALANCE_ROUND_ROBIN 1 50 #define G_MIRROR_BALANCE_LOAD 2 51 #define G_MIRROR_BALANCE_SPLIT 3 52 #define G_MIRROR_BALANCE_PREFER 4 53 #define G_MIRROR_BALANCE_MIN G_MIRROR_BALANCE_NONE 54 #define G_MIRROR_BALANCE_MAX G_MIRROR_BALANCE_PREFER 55 56 #define G_MIRROR_DISK_FLAG_DIRTY 0x0000000000000001ULL 57 #define G_MIRROR_DISK_FLAG_SYNCHRONIZING 0x0000000000000002ULL 58 #define G_MIRROR_DISK_FLAG_FORCE_SYNC 0x0000000000000004ULL 59 #define G_MIRROR_DISK_FLAG_INACTIVE 0x0000000000000008ULL 60 #define G_MIRROR_DISK_FLAG_HARDCODED 0x0000000000000010ULL 61 #define G_MIRROR_DISK_FLAG_BROKEN 0x0000000000000020ULL 62 #define G_MIRROR_DISK_FLAG_CANDELETE 0x0000000000000040ULL 63 #define G_MIRROR_DISK_FLAG_MASK (G_MIRROR_DISK_FLAG_DIRTY | \ 64 G_MIRROR_DISK_FLAG_SYNCHRONIZING | \ 65 G_MIRROR_DISK_FLAG_FORCE_SYNC | \ 66 G_MIRROR_DISK_FLAG_INACTIVE | \ 67 G_MIRROR_DISK_FLAG_CANDELETE) 68 69 #define G_MIRROR_DEVICE_FLAG_NOAUTOSYNC 0x0000000000000001ULL 70 #define G_MIRROR_DEVICE_FLAG_NOFAILSYNC 0x0000000000000002ULL 71 #define G_MIRROR_DEVICE_FLAG_MASK (G_MIRROR_DEVICE_FLAG_NOAUTOSYNC | \ 72 G_MIRROR_DEVICE_FLAG_NOFAILSYNC) 73 74 #ifdef _KERNEL 75 extern u_int g_mirror_debug; 76 77 #define G_MIRROR_DEBUG(lvl, ...) do { \ 78 if (g_mirror_debug >= (lvl)) { \ 79 printf("GEOM_MIRROR"); \ 80 if (g_mirror_debug > 0) \ 81 printf("[%u]", lvl); \ 82 printf(": "); \ 83 printf(__VA_ARGS__); \ 84 printf("\n"); \ 85 } \ 86 } while (0) 87 #define G_MIRROR_LOGREQ(lvl, bp, ...) do { \ 88 if (g_mirror_debug >= (lvl)) { \ 89 printf("GEOM_MIRROR"); \ 90 if (g_mirror_debug > 0) \ 91 printf("[%u]", lvl); \ 92 printf(": "); \ 93 printf(__VA_ARGS__); \ 94 printf(" "); \ 95 g_print_bio(bp); \ 96 printf("\n"); \ 97 } \ 98 } while (0) 99 100 #define G_MIRROR_BIO_FLAG_REGULAR 0x01 101 #define G_MIRROR_BIO_FLAG_SYNC 0x02 102 103 /* 104 * Informations needed for synchronization. 105 */ 106 struct g_mirror_disk_sync { 107 struct g_consumer *ds_consumer; /* Consumer connected to our mirror. */ 108 off_t ds_offset; /* Offset of next request to send. */ 109 off_t ds_offset_done; /* Offset of already synchronized 110 region. */ 111 u_int ds_syncid; /* Disk's synchronization ID. */ 112 u_int ds_inflight; /* Number of in-flight sync requests. */ 113 struct bio **ds_bios; /* BIOs for synchronization I/O. */ 114 }; 115 116 /* 117 * Informations needed for synchronization. 118 */ 119 struct g_mirror_device_sync { 120 struct g_geom *ds_geom; /* Synchronization geom. */ 121 u_int ds_ndisks; /* Number of disks in SYNCHRONIZING 122 state. */ 123 }; 124 125 #define G_MIRROR_DISK_STATE_NONE 0 126 #define G_MIRROR_DISK_STATE_NEW 1 127 #define G_MIRROR_DISK_STATE_ACTIVE 2 128 #define G_MIRROR_DISK_STATE_STALE 3 129 #define G_MIRROR_DISK_STATE_SYNCHRONIZING 4 130 #define G_MIRROR_DISK_STATE_DISCONNECTED 5 131 #define G_MIRROR_DISK_STATE_DESTROY 6 132 struct g_mirror_disk { 133 uint32_t d_id; /* Disk ID. */ 134 struct g_consumer *d_consumer; /* Consumer. */ 135 struct g_mirror_softc *d_softc; /* Back-pointer to softc. */ 136 int d_state; /* Disk state. */ 137 u_int d_priority; /* Disk priority. */ 138 u_int load; /* Averaged queue length */ 139 off_t d_last_offset; /* Last read offset */ 140 uint64_t d_flags; /* Additional flags. */ 141 u_int d_genid; /* Disk's generation ID. */ 142 struct g_mirror_disk_sync d_sync;/* Sync information. */ 143 LIST_ENTRY(g_mirror_disk) d_next; 144 }; 145 #define d_name d_consumer->provider->name 146 147 #define G_MIRROR_EVENT_DONTWAIT 0x1 148 #define G_MIRROR_EVENT_WAIT 0x2 149 #define G_MIRROR_EVENT_DEVICE 0x4 150 #define G_MIRROR_EVENT_DONE 0x8 151 struct g_mirror_event { 152 struct g_mirror_disk *e_disk; 153 int e_state; 154 int e_flags; 155 int e_error; 156 TAILQ_ENTRY(g_mirror_event) e_next; 157 }; 158 159 #define G_MIRROR_DEVICE_FLAG_DESTROY 0x0100000000000000ULL 160 #define G_MIRROR_DEVICE_FLAG_WAIT 0x0200000000000000ULL 161 #define G_MIRROR_DEVICE_FLAG_DESTROYING 0x0400000000000000ULL 162 #define G_MIRROR_DEVICE_FLAG_TASTING 0x0800000000000000ULL 163 164 #define G_MIRROR_DEVICE_STATE_STARTING 0 165 #define G_MIRROR_DEVICE_STATE_RUNNING 1 166 167 /* Bump syncid on first write. */ 168 #define G_MIRROR_BUMP_SYNCID 0x1 169 /* Bump genid immediately. */ 170 #define G_MIRROR_BUMP_GENID 0x2 171 struct g_mirror_softc { 172 u_int sc_state; /* Device state. */ 173 uint32_t sc_slice; /* Slice size. */ 174 uint8_t sc_balance; /* Balance algorithm. */ 175 uint64_t sc_mediasize; /* Device size. */ 176 uint32_t sc_sectorsize; /* Sector size. */ 177 uint64_t sc_flags; /* Additional flags. */ 178 179 struct g_geom *sc_geom; 180 struct g_provider *sc_provider; 181 182 uint32_t sc_id; /* Mirror unique ID. */ 183 184 struct sx sc_lock; 185 struct bio_queue_head sc_queue; 186 struct mtx sc_queue_mtx; 187 struct proc *sc_worker; 188 struct bio_queue_head sc_regular_delayed; /* Delayed I/O requests due 189 collision with sync 190 requests. */ 191 struct bio_queue_head sc_inflight; /* In-flight regular write 192 requests. */ 193 struct bio_queue_head sc_sync_delayed; /* Delayed sync requests due 194 collision with regular 195 requests. */ 196 197 LIST_HEAD(, g_mirror_disk) sc_disks; 198 u_int sc_ndisks; /* Number of disks. */ 199 struct g_mirror_disk *sc_hint; 200 201 u_int sc_genid; /* Generation ID. */ 202 u_int sc_syncid; /* Synchronization ID. */ 203 int sc_bump_id; 204 struct g_mirror_device_sync sc_sync; 205 int sc_idle; /* DIRTY flags removed. */ 206 time_t sc_last_write; 207 u_int sc_writes; 208 209 TAILQ_HEAD(, g_mirror_event) sc_events; 210 struct mtx sc_events_mtx; 211 212 struct callout sc_callout; 213 214 struct root_hold_token *sc_rootmount; 215 }; 216 #define sc_name sc_geom->name 217 218 u_int g_mirror_ndisks(struct g_mirror_softc *sc, int state); 219 #define G_MIRROR_DESTROY_SOFT 0 220 #define G_MIRROR_DESTROY_DELAYED 1 221 #define G_MIRROR_DESTROY_HARD 2 222 int g_mirror_destroy(struct g_mirror_softc *sc, int how); 223 int g_mirror_event_send(void *arg, int state, int flags); 224 struct g_mirror_metadata; 225 int g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 226 struct g_mirror_metadata *md); 227 int g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md); 228 void g_mirror_fill_metadata(struct g_mirror_softc *sc, 229 struct g_mirror_disk *disk, struct g_mirror_metadata *md); 230 void g_mirror_update_metadata(struct g_mirror_disk *disk); 231 232 g_ctl_req_t g_mirror_config; 233 #endif /* _KERNEL */ 234 235 struct g_mirror_metadata { 236 char md_magic[16]; /* Magic value. */ 237 uint32_t md_version; /* Version number. */ 238 char md_name[16]; /* Mirror name. */ 239 uint32_t md_mid; /* Mirror unique ID. */ 240 uint32_t md_did; /* Disk unique ID. */ 241 uint8_t md_all; /* Number of disks in mirror. */ 242 uint32_t md_genid; /* Generation ID. */ 243 uint32_t md_syncid; /* Synchronization ID. */ 244 uint8_t md_priority; /* Disk priority. */ 245 uint32_t md_slice; /* Slice size. */ 246 uint8_t md_balance; /* Balance type. */ 247 uint64_t md_mediasize; /* Size of the smallest 248 disk in mirror. */ 249 uint32_t md_sectorsize; /* Sector size. */ 250 uint64_t md_sync_offset; /* Synchronized offset. */ 251 uint64_t md_mflags; /* Additional mirror flags. */ 252 uint64_t md_dflags; /* Additional disk flags. */ 253 char md_provider[16]; /* Hardcoded provider. */ 254 uint64_t md_provsize; /* Provider's size. */ 255 u_char md_hash[16]; /* MD5 hash. */ 256 }; 257 static __inline void 258 mirror_metadata_encode(struct g_mirror_metadata *md, u_char *data) 259 { 260 MD5_CTX ctx; 261 262 bcopy(md->md_magic, data, 16); 263 le32enc(data + 16, md->md_version); 264 bcopy(md->md_name, data + 20, 16); 265 le32enc(data + 36, md->md_mid); 266 le32enc(data + 40, md->md_did); 267 *(data + 44) = md->md_all; 268 le32enc(data + 45, md->md_genid); 269 le32enc(data + 49, md->md_syncid); 270 *(data + 53) = md->md_priority; 271 le32enc(data + 54, md->md_slice); 272 *(data + 58) = md->md_balance; 273 le64enc(data + 59, md->md_mediasize); 274 le32enc(data + 67, md->md_sectorsize); 275 le64enc(data + 71, md->md_sync_offset); 276 le64enc(data + 79, md->md_mflags); 277 le64enc(data + 87, md->md_dflags); 278 bcopy(md->md_provider, data + 95, 16); 279 le64enc(data + 111, md->md_provsize); 280 MD5Init(&ctx); 281 MD5Update(&ctx, data, 119); 282 MD5Final(md->md_hash, &ctx); 283 bcopy(md->md_hash, data + 119, 16); 284 } 285 static __inline int 286 mirror_metadata_decode_v0v1(const u_char *data, struct g_mirror_metadata *md) 287 { 288 MD5_CTX ctx; 289 290 bcopy(data + 20, md->md_name, 16); 291 md->md_mid = le32dec(data + 36); 292 md->md_did = le32dec(data + 40); 293 md->md_all = *(data + 44); 294 md->md_syncid = le32dec(data + 45); 295 md->md_priority = *(data + 49); 296 md->md_slice = le32dec(data + 50); 297 md->md_balance = *(data + 54); 298 md->md_mediasize = le64dec(data + 55); 299 md->md_sectorsize = le32dec(data + 63); 300 md->md_sync_offset = le64dec(data + 67); 301 md->md_mflags = le64dec(data + 75); 302 md->md_dflags = le64dec(data + 83); 303 bcopy(data + 91, md->md_provider, 16); 304 bcopy(data + 107, md->md_hash, 16); 305 MD5Init(&ctx); 306 MD5Update(&ctx, data, 107); 307 MD5Final(md->md_hash, &ctx); 308 if (bcmp(md->md_hash, data + 107, 16) != 0) 309 return (EINVAL); 310 311 /* New fields. */ 312 md->md_genid = 0; 313 md->md_provsize = 0; 314 315 return (0); 316 } 317 static __inline int 318 mirror_metadata_decode_v2(const u_char *data, struct g_mirror_metadata *md) 319 { 320 MD5_CTX ctx; 321 322 bcopy(data + 20, md->md_name, 16); 323 md->md_mid = le32dec(data + 36); 324 md->md_did = le32dec(data + 40); 325 md->md_all = *(data + 44); 326 md->md_genid = le32dec(data + 45); 327 md->md_syncid = le32dec(data + 49); 328 md->md_priority = *(data + 53); 329 md->md_slice = le32dec(data + 54); 330 md->md_balance = *(data + 58); 331 md->md_mediasize = le64dec(data + 59); 332 md->md_sectorsize = le32dec(data + 67); 333 md->md_sync_offset = le64dec(data + 71); 334 md->md_mflags = le64dec(data + 79); 335 md->md_dflags = le64dec(data + 87); 336 bcopy(data + 95, md->md_provider, 16); 337 bcopy(data + 111, md->md_hash, 16); 338 MD5Init(&ctx); 339 MD5Update(&ctx, data, 111); 340 MD5Final(md->md_hash, &ctx); 341 if (bcmp(md->md_hash, data + 111, 16) != 0) 342 return (EINVAL); 343 344 /* New fields. */ 345 md->md_provsize = 0; 346 347 return (0); 348 } 349 static __inline int 350 mirror_metadata_decode_v3v4(const u_char *data, struct g_mirror_metadata *md) 351 { 352 MD5_CTX ctx; 353 354 bcopy(data + 20, md->md_name, 16); 355 md->md_mid = le32dec(data + 36); 356 md->md_did = le32dec(data + 40); 357 md->md_all = *(data + 44); 358 md->md_genid = le32dec(data + 45); 359 md->md_syncid = le32dec(data + 49); 360 md->md_priority = *(data + 53); 361 md->md_slice = le32dec(data + 54); 362 md->md_balance = *(data + 58); 363 md->md_mediasize = le64dec(data + 59); 364 md->md_sectorsize = le32dec(data + 67); 365 md->md_sync_offset = le64dec(data + 71); 366 md->md_mflags = le64dec(data + 79); 367 md->md_dflags = le64dec(data + 87); 368 bcopy(data + 95, md->md_provider, 16); 369 md->md_provsize = le64dec(data + 111); 370 bcopy(data + 119, md->md_hash, 16); 371 MD5Init(&ctx); 372 MD5Update(&ctx, data, 119); 373 MD5Final(md->md_hash, &ctx); 374 if (bcmp(md->md_hash, data + 119, 16) != 0) 375 return (EINVAL); 376 return (0); 377 } 378 static __inline int 379 mirror_metadata_decode(const u_char *data, struct g_mirror_metadata *md) 380 { 381 int error; 382 383 bcopy(data, md->md_magic, 16); 384 md->md_version = le32dec(data + 16); 385 switch (md->md_version) { 386 case 0: 387 case 1: 388 error = mirror_metadata_decode_v0v1(data, md); 389 break; 390 case 2: 391 error = mirror_metadata_decode_v2(data, md); 392 break; 393 case 3: 394 case 4: 395 error = mirror_metadata_decode_v3v4(data, md); 396 break; 397 default: 398 error = EINVAL; 399 break; 400 } 401 return (error); 402 } 403 404 static __inline const char * 405 balance_name(u_int balance) 406 { 407 static const char *algorithms[] = { 408 [G_MIRROR_BALANCE_NONE] = "none", 409 [G_MIRROR_BALANCE_ROUND_ROBIN] = "round-robin", 410 [G_MIRROR_BALANCE_LOAD] = "load", 411 [G_MIRROR_BALANCE_SPLIT] = "split", 412 [G_MIRROR_BALANCE_PREFER] = "prefer", 413 [G_MIRROR_BALANCE_MAX + 1] = "unknown" 414 }; 415 416 if (balance > G_MIRROR_BALANCE_MAX) 417 balance = G_MIRROR_BALANCE_MAX + 1; 418 419 return (algorithms[balance]); 420 } 421 422 static __inline int 423 balance_id(const char *name) 424 { 425 static const char *algorithms[] = { 426 [G_MIRROR_BALANCE_NONE] = "none", 427 [G_MIRROR_BALANCE_ROUND_ROBIN] = "round-robin", 428 [G_MIRROR_BALANCE_LOAD] = "load", 429 [G_MIRROR_BALANCE_SPLIT] = "split", 430 [G_MIRROR_BALANCE_PREFER] = "prefer" 431 }; 432 int n; 433 434 for (n = G_MIRROR_BALANCE_MIN; n <= G_MIRROR_BALANCE_MAX; n++) { 435 if (strcmp(name, algorithms[n]) == 0) 436 return (n); 437 } 438 return (-1); 439 } 440 441 static __inline void 442 mirror_metadata_dump(const struct g_mirror_metadata *md) 443 { 444 static const char hex[] = "0123456789abcdef"; 445 char hash[16 * 2 + 1]; 446 u_int i; 447 448 printf(" magic: %s\n", md->md_magic); 449 printf(" version: %u\n", (u_int)md->md_version); 450 printf(" name: %s\n", md->md_name); 451 printf(" mid: %u\n", (u_int)md->md_mid); 452 printf(" did: %u\n", (u_int)md->md_did); 453 printf(" all: %u\n", (u_int)md->md_all); 454 printf(" genid: %u\n", (u_int)md->md_genid); 455 printf(" syncid: %u\n", (u_int)md->md_syncid); 456 printf(" priority: %u\n", (u_int)md->md_priority); 457 printf(" slice: %u\n", (u_int)md->md_slice); 458 printf(" balance: %s\n", balance_name((u_int)md->md_balance)); 459 printf(" mediasize: %jd\n", (intmax_t)md->md_mediasize); 460 printf("sectorsize: %u\n", (u_int)md->md_sectorsize); 461 printf("syncoffset: %jd\n", (intmax_t)md->md_sync_offset); 462 printf(" mflags:"); 463 if (md->md_mflags == 0) 464 printf(" NONE"); 465 else { 466 if ((md->md_mflags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 467 printf(" NOFAILSYNC"); 468 if ((md->md_mflags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0) 469 printf(" NOAUTOSYNC"); 470 } 471 printf("\n"); 472 printf(" dflags:"); 473 if (md->md_dflags == 0) 474 printf(" NONE"); 475 else { 476 if ((md->md_dflags & G_MIRROR_DISK_FLAG_DIRTY) != 0) 477 printf(" DIRTY"); 478 if ((md->md_dflags & G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) 479 printf(" SYNCHRONIZING"); 480 if ((md->md_dflags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) 481 printf(" FORCE_SYNC"); 482 if ((md->md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) 483 printf(" INACTIVE"); 484 } 485 printf("\n"); 486 printf("hcprovider: %s\n", md->md_provider); 487 printf(" provsize: %ju\n", (uintmax_t)md->md_provsize); 488 bzero(hash, sizeof(hash)); 489 for (i = 0; i < 16; i++) { 490 hash[i * 2] = hex[md->md_hash[i] >> 4]; 491 hash[i * 2 + 1] = hex[md->md_hash[i] & 0x0f]; 492 } 493 printf(" MD5 hash: %s\n", hash); 494 } 495 #endif /* !_G_MIRROR_H_ */ 496