1 /*- 2 * Copyright (c) 2004 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 */ 44 #define G_MIRROR_VERSION 2 45 46 #define G_MIRROR_BALANCE_NONE 0 47 #define G_MIRROR_BALANCE_ROUND_ROBIN 1 48 #define G_MIRROR_BALANCE_LOAD 2 49 #define G_MIRROR_BALANCE_SPLIT 3 50 #define G_MIRROR_BALANCE_PREFER 4 51 #define G_MIRROR_BALANCE_MIN G_MIRROR_BALANCE_NONE 52 #define G_MIRROR_BALANCE_MAX G_MIRROR_BALANCE_PREFER 53 54 #define G_MIRROR_DISK_FLAG_DIRTY 0x0000000000000001ULL 55 #define G_MIRROR_DISK_FLAG_SYNCHRONIZING 0x0000000000000002ULL 56 #define G_MIRROR_DISK_FLAG_FORCE_SYNC 0x0000000000000004ULL 57 #define G_MIRROR_DISK_FLAG_INACTIVE 0x0000000000000008ULL 58 #define G_MIRROR_DISK_FLAG_HARDCODED 0x0000000000000010ULL 59 #define G_MIRROR_DISK_FLAG_MASK (G_MIRROR_DISK_FLAG_DIRTY | \ 60 G_MIRROR_DISK_FLAG_SYNCHRONIZING | \ 61 G_MIRROR_DISK_FLAG_FORCE_SYNC | \ 62 G_MIRROR_DISK_FLAG_INACTIVE) 63 64 #define G_MIRROR_DEVICE_FLAG_NOAUTOSYNC 0x0000000000000001ULL 65 #define G_MIRROR_DEVICE_FLAG_MASK (G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) 66 67 #ifdef _KERNEL 68 extern u_int g_mirror_debug; 69 70 #define G_MIRROR_DEBUG(lvl, ...) do { \ 71 if (g_mirror_debug >= (lvl)) { \ 72 printf("GEOM_MIRROR"); \ 73 if (g_mirror_debug > 0) \ 74 printf("[%u]", lvl); \ 75 printf(": "); \ 76 printf(__VA_ARGS__); \ 77 printf("\n"); \ 78 } \ 79 } while (0) 80 #define G_MIRROR_LOGREQ(lvl, bp, ...) do { \ 81 if (g_mirror_debug >= (lvl)) { \ 82 printf("GEOM_MIRROR"); \ 83 if (g_mirror_debug > 0) \ 84 printf("[%u]", lvl); \ 85 printf(": "); \ 86 printf(__VA_ARGS__); \ 87 printf(" "); \ 88 g_print_bio(bp); \ 89 printf("\n"); \ 90 } \ 91 } while (0) 92 93 #define G_MIRROR_BIO_FLAG_REGULAR 0x01 94 #define G_MIRROR_BIO_FLAG_SYNC 0x02 95 96 /* 97 * Informations needed for synchronization. 98 */ 99 struct g_mirror_disk_sync { 100 struct g_consumer *ds_consumer; /* Consumer connected to our mirror. */ 101 off_t ds_offset; /* Offset of next request to send. */ 102 off_t ds_offset_done; /* Offset of already synchronized 103 region. */ 104 off_t ds_resync; /* Resynchronize from this offset. */ 105 u_int ds_syncid; /* Disk's synchronization ID. */ 106 u_char *ds_data; 107 }; 108 109 /* 110 * Informations needed for synchronization. 111 */ 112 struct g_mirror_device_sync { 113 struct g_geom *ds_geom; /* Synchronization geom. */ 114 u_int ds_ndisks; /* Number of disks in SYNCHRONIZING 115 state. */ 116 }; 117 118 #define G_MIRROR_DISK_STATE_NONE 0 119 #define G_MIRROR_DISK_STATE_NEW 1 120 #define G_MIRROR_DISK_STATE_ACTIVE 2 121 #define G_MIRROR_DISK_STATE_STALE 3 122 #define G_MIRROR_DISK_STATE_SYNCHRONIZING 4 123 #define G_MIRROR_DISK_STATE_DISCONNECTED 5 124 #define G_MIRROR_DISK_STATE_DESTROY 6 125 struct g_mirror_disk { 126 uint32_t d_id; /* Disk ID. */ 127 struct g_consumer *d_consumer; /* Consumer. */ 128 struct g_mirror_softc *d_softc; /* Back-pointer to softc. */ 129 int d_state; /* Disk state. */ 130 u_int d_priority; /* Disk priority. */ 131 struct bintime d_delay; /* Disk delay. */ 132 struct bintime d_last_used; /* When disk was last used. */ 133 uint64_t d_flags; /* Additional flags. */ 134 u_int d_genid; /* Disk's generation ID. */ 135 struct g_mirror_disk_sync d_sync;/* Sync information. */ 136 LIST_ENTRY(g_mirror_disk) d_next; 137 }; 138 #define d_name d_consumer->provider->name 139 140 #define G_MIRROR_EVENT_DONTWAIT 0x1 141 #define G_MIRROR_EVENT_WAIT 0x2 142 #define G_MIRROR_EVENT_DEVICE 0x4 143 #define G_MIRROR_EVENT_DONE 0x8 144 struct g_mirror_event { 145 struct g_mirror_disk *e_disk; 146 int e_state; 147 int e_flags; 148 int e_error; 149 TAILQ_ENTRY(g_mirror_event) e_next; 150 }; 151 152 #define G_MIRROR_DEVICE_FLAG_DESTROY 0x0100000000000000ULL 153 #define G_MIRROR_DEVICE_FLAG_WAIT 0x0200000000000000ULL 154 155 #define G_MIRROR_DEVICE_STATE_STARTING 0 156 #define G_MIRROR_DEVICE_STATE_RUNNING 1 157 158 /* Bump syncid on first write. */ 159 #define G_MIRROR_BUMP_SYNCID 0x1 160 /* Bump genid immediately. */ 161 #define G_MIRROR_BUMP_GENID 0x2 162 struct g_mirror_softc { 163 u_int sc_state; /* Device state. */ 164 uint32_t sc_slice; /* Slice size. */ 165 uint8_t sc_balance; /* Balance algorithm. */ 166 uint64_t sc_mediasize; /* Device size. */ 167 uint32_t sc_sectorsize; /* Sector size. */ 168 uint64_t sc_flags; /* Additional flags. */ 169 170 struct g_geom *sc_geom; 171 struct g_provider *sc_provider; 172 173 uint32_t sc_id; /* Mirror unique ID. */ 174 175 struct bio_queue_head sc_queue; 176 struct mtx sc_queue_mtx; 177 struct proc *sc_worker; 178 179 LIST_HEAD(, g_mirror_disk) sc_disks; 180 u_int sc_ndisks; /* Number of disks. */ 181 struct g_mirror_disk *sc_hint; 182 183 u_int sc_genid; /* Generation ID. */ 184 u_int sc_syncid; /* Synchronization ID. */ 185 int sc_bump_id; 186 struct g_mirror_device_sync sc_sync; 187 int sc_idle; /* DIRTY flags removed. */ 188 189 TAILQ_HEAD(, g_mirror_event) sc_events; 190 struct mtx sc_events_mtx; 191 192 struct callout sc_callout; 193 }; 194 #define sc_name sc_geom->name 195 196 u_int g_mirror_ndisks(struct g_mirror_softc *sc, int state); 197 int g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force); 198 int g_mirror_event_send(void *arg, int state, int flags); 199 struct g_mirror_metadata; 200 int g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 201 struct g_mirror_metadata *md); 202 int g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md); 203 void g_mirror_fill_metadata(struct g_mirror_softc *sc, 204 struct g_mirror_disk *disk, struct g_mirror_metadata *md); 205 void g_mirror_update_metadata(struct g_mirror_disk *disk); 206 207 g_ctl_req_t g_mirror_config; 208 #endif /* _KERNEL */ 209 210 struct g_mirror_metadata { 211 char md_magic[16]; /* Magic value. */ 212 uint32_t md_version; /* Version number. */ 213 char md_name[16]; /* Mirror name. */ 214 uint32_t md_mid; /* Mirror unique ID. */ 215 uint32_t md_did; /* Disk unique ID. */ 216 uint8_t md_all; /* Number of disks in mirror. */ 217 uint32_t md_genid; /* Generation ID. */ 218 uint32_t md_syncid; /* Synchronization ID. */ 219 uint8_t md_priority; /* Disk priority. */ 220 uint32_t md_slice; /* Slice size. */ 221 uint8_t md_balance; /* Balance type. */ 222 uint64_t md_mediasize; /* Size of the smallest 223 disk in mirror. */ 224 uint32_t md_sectorsize; /* Sector size. */ 225 uint64_t md_sync_offset; /* Synchronized offset. */ 226 uint64_t md_mflags; /* Additional mirror flags. */ 227 uint64_t md_dflags; /* Additional disk flags. */ 228 char md_provider[16]; /* Hardcoded provider. */ 229 u_char md_hash[16]; /* MD5 hash. */ 230 }; 231 static __inline void 232 mirror_metadata_encode(struct g_mirror_metadata *md, u_char *data) 233 { 234 MD5_CTX ctx; 235 236 bcopy(md->md_magic, data, 16); 237 le32enc(data + 16, md->md_version); 238 bcopy(md->md_name, data + 20, 16); 239 le32enc(data + 36, md->md_mid); 240 le32enc(data + 40, md->md_did); 241 *(data + 44) = md->md_all; 242 le32enc(data + 45, md->md_genid); 243 le32enc(data + 49, md->md_syncid); 244 *(data + 53) = md->md_priority; 245 le32enc(data + 54, md->md_slice); 246 *(data + 58) = md->md_balance; 247 le64enc(data + 59, md->md_mediasize); 248 le32enc(data + 67, md->md_sectorsize); 249 le64enc(data + 71, md->md_sync_offset); 250 le64enc(data + 79, md->md_mflags); 251 le64enc(data + 87, md->md_dflags); 252 bcopy(md->md_provider, data + 95, 16); 253 MD5Init(&ctx); 254 MD5Update(&ctx, data, 111); 255 MD5Final(md->md_hash, &ctx); 256 bcopy(md->md_hash, data + 111, 16); 257 } 258 static __inline int 259 mirror_metadata_decode_v0v1(const u_char *data, struct g_mirror_metadata *md) 260 { 261 MD5_CTX ctx; 262 263 bcopy(data + 20, md->md_name, 16); 264 md->md_mid = le32dec(data + 36); 265 md->md_did = le32dec(data + 40); 266 md->md_all = *(data + 44); 267 md->md_syncid = le32dec(data + 45); 268 md->md_priority = *(data + 49); 269 md->md_slice = le32dec(data + 50); 270 md->md_balance = *(data + 54); 271 md->md_mediasize = le64dec(data + 55); 272 md->md_sectorsize = le32dec(data + 63); 273 md->md_sync_offset = le64dec(data + 67); 274 md->md_mflags = le64dec(data + 75); 275 md->md_dflags = le64dec(data + 83); 276 bcopy(data + 91, md->md_provider, 16); 277 bcopy(data + 107, md->md_hash, 16); 278 MD5Init(&ctx); 279 MD5Update(&ctx, data, 107); 280 MD5Final(md->md_hash, &ctx); 281 if (bcmp(md->md_hash, data + 107, 16) != 0) 282 return (EINVAL); 283 284 /* New fields. */ 285 md->md_genid = 0; 286 287 return (0); 288 } 289 static __inline int 290 mirror_metadata_decode_v2(const u_char *data, struct g_mirror_metadata *md) 291 { 292 MD5_CTX ctx; 293 294 bcopy(data + 20, md->md_name, 16); 295 md->md_mid = le32dec(data + 36); 296 md->md_did = le32dec(data + 40); 297 md->md_all = *(data + 44); 298 md->md_genid = le32dec(data + 45); 299 md->md_syncid = le32dec(data + 49); 300 md->md_priority = *(data + 53); 301 md->md_slice = le32dec(data + 54); 302 md->md_balance = *(data + 58); 303 md->md_mediasize = le64dec(data + 59); 304 md->md_sectorsize = le32dec(data + 67); 305 md->md_sync_offset = le64dec(data + 71); 306 md->md_mflags = le64dec(data + 79); 307 md->md_dflags = le64dec(data + 87); 308 bcopy(data + 95, md->md_provider, 16); 309 bcopy(data + 111, md->md_hash, 16); 310 MD5Init(&ctx); 311 MD5Update(&ctx, data, 111); 312 MD5Final(md->md_hash, &ctx); 313 if (bcmp(md->md_hash, data + 111, 16) != 0) 314 return (EINVAL); 315 return (0); 316 } 317 static __inline int 318 mirror_metadata_decode(const u_char *data, struct g_mirror_metadata *md) 319 { 320 int error; 321 322 bcopy(data, md->md_magic, 16); 323 md->md_version = le32dec(data + 16); 324 switch (md->md_version) { 325 case 0: 326 case 1: 327 error = mirror_metadata_decode_v0v1(data, md); 328 break; 329 case 2: 330 error = mirror_metadata_decode_v2(data, md); 331 break; 332 default: 333 error = EINVAL; 334 break; 335 } 336 return (error); 337 } 338 339 static __inline const char * 340 balance_name(u_int balance) 341 { 342 static const char *algorithms[] = { 343 [G_MIRROR_BALANCE_NONE] = "none", 344 [G_MIRROR_BALANCE_ROUND_ROBIN] = "round-robin", 345 [G_MIRROR_BALANCE_LOAD] = "load", 346 [G_MIRROR_BALANCE_SPLIT] = "split", 347 [G_MIRROR_BALANCE_PREFER] = "prefer", 348 [G_MIRROR_BALANCE_MAX + 1] = "unknown" 349 }; 350 351 if (balance > G_MIRROR_BALANCE_MAX) 352 balance = G_MIRROR_BALANCE_MAX + 1; 353 354 return (algorithms[balance]); 355 } 356 357 static __inline int 358 balance_id(const char *name) 359 { 360 static const char *algorithms[] = { 361 [G_MIRROR_BALANCE_NONE] = "none", 362 [G_MIRROR_BALANCE_ROUND_ROBIN] = "round-robin", 363 [G_MIRROR_BALANCE_LOAD] = "load", 364 [G_MIRROR_BALANCE_SPLIT] = "split", 365 [G_MIRROR_BALANCE_PREFER] = "prefer" 366 }; 367 int n; 368 369 for (n = G_MIRROR_BALANCE_MIN; n <= G_MIRROR_BALANCE_MAX; n++) { 370 if (strcmp(name, algorithms[n]) == 0) 371 return (n); 372 } 373 return (-1); 374 } 375 376 static __inline void 377 mirror_metadata_dump(const struct g_mirror_metadata *md) 378 { 379 static const char hex[] = "0123456789abcdef"; 380 char hash[16 * 2 + 1]; 381 u_int i; 382 383 printf(" magic: %s\n", md->md_magic); 384 printf(" version: %u\n", (u_int)md->md_version); 385 printf(" name: %s\n", md->md_name); 386 printf(" mid: %u\n", (u_int)md->md_mid); 387 printf(" did: %u\n", (u_int)md->md_did); 388 printf(" all: %u\n", (u_int)md->md_all); 389 printf(" genid: %u\n", (u_int)md->md_genid); 390 printf(" syncid: %u\n", (u_int)md->md_syncid); 391 printf(" priority: %u\n", (u_int)md->md_priority); 392 printf(" slice: %u\n", (u_int)md->md_slice); 393 printf(" balance: %s\n", balance_name((u_int)md->md_balance)); 394 printf(" mediasize: %jd\n", (intmax_t)md->md_mediasize); 395 printf("sectorsize: %u\n", (u_int)md->md_sectorsize); 396 printf("syncoffset: %jd\n", (intmax_t)md->md_sync_offset); 397 printf(" mflags:"); 398 if (md->md_mflags == 0) 399 printf(" NONE"); 400 else { 401 if ((md->md_mflags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0) 402 printf(" NOAUTOSYNC"); 403 } 404 printf("\n"); 405 printf(" dflags:"); 406 if (md->md_dflags == 0) 407 printf(" NONE"); 408 else { 409 if ((md->md_dflags & G_MIRROR_DISK_FLAG_DIRTY) != 0) 410 printf(" DIRTY"); 411 if ((md->md_dflags & G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) 412 printf(" SYNCHRONIZING"); 413 if ((md->md_dflags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) 414 printf(" FORCE_SYNC"); 415 if ((md->md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) 416 printf(" INACTIVE"); 417 } 418 printf("\n"); 419 printf("hcprovider: %s\n", md->md_provider); 420 bzero(hash, sizeof(hash)); 421 for (i = 0; i < 16; i++) { 422 hash[i * 2] = hex[md->md_hash[i] >> 4]; 423 hash[i * 2 + 1] = hex[md->md_hash[i] & 0x0f]; 424 } 425 printf(" MD5 hash: %s\n", hash); 426 } 427 #endif /* !_G_MIRROR_H_ */ 428