1 /*- 2 * Copyright (c) 2004-2005 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_RAID3_H_ 30 #define _G_RAID3_H_ 31 32 #include <sys/endian.h> 33 #include <sys/md5.h> 34 35 #define G_RAID3_CLASS_NAME "RAID3" 36 37 #define G_RAID3_MAGIC "GEOM::RAID3" 38 /* 39 * Version history: 40 * 0 - Initial version number. 41 * 1 - Added 'round-robin reading' algorithm. 42 * 2 - Added 'verify reading' algorithm. 43 * 3 - Added md_genid field to metadata. 44 * 4 - Added md_provsize field to metadata. 45 */ 46 #define G_RAID3_VERSION 4 47 48 #define G_RAID3_DISK_FLAG_DIRTY 0x0000000000000001ULL 49 #define G_RAID3_DISK_FLAG_SYNCHRONIZING 0x0000000000000002ULL 50 #define G_RAID3_DISK_FLAG_FORCE_SYNC 0x0000000000000004ULL 51 #define G_RAID3_DISK_FLAG_HARDCODED 0x0000000000000008ULL 52 #define G_RAID3_DISK_FLAG_MASK (G_RAID3_DISK_FLAG_DIRTY | \ 53 G_RAID3_DISK_FLAG_SYNCHRONIZING | \ 54 G_RAID3_DISK_FLAG_FORCE_SYNC) 55 56 #define G_RAID3_DEVICE_FLAG_NOAUTOSYNC 0x0000000000000001ULL 57 #define G_RAID3_DEVICE_FLAG_ROUND_ROBIN 0x0000000000000002ULL 58 #define G_RAID3_DEVICE_FLAG_VERIFY 0x0000000000000004ULL 59 #define G_RAID3_DEVICE_FLAG_MASK (G_RAID3_DEVICE_FLAG_NOAUTOSYNC | \ 60 G_RAID3_DEVICE_FLAG_ROUND_ROBIN | \ 61 G_RAID3_DEVICE_FLAG_VERIFY) 62 63 #ifdef _KERNEL 64 extern u_int g_raid3_debug; 65 66 #define G_RAID3_DEBUG(lvl, ...) do { \ 67 if (g_raid3_debug >= (lvl)) { \ 68 printf("GEOM_RAID3"); \ 69 if (g_raid3_debug > 0) \ 70 printf("[%u]", lvl); \ 71 printf(": "); \ 72 printf(__VA_ARGS__); \ 73 printf("\n"); \ 74 } \ 75 } while (0) 76 #define G_RAID3_LOGREQ(lvl, bp, ...) do { \ 77 if (g_raid3_debug >= (lvl)) { \ 78 printf("GEOM_RAID3"); \ 79 if (g_raid3_debug > 0) \ 80 printf("[%u]", lvl); \ 81 printf(": "); \ 82 printf(__VA_ARGS__); \ 83 printf(" "); \ 84 g_print_bio(bp); \ 85 printf("\n"); \ 86 } \ 87 } while (0) 88 89 #define G_RAID3_BIO_CFLAG_REGULAR 0x01 90 #define G_RAID3_BIO_CFLAG_SYNC 0x02 91 #define G_RAID3_BIO_CFLAG_PARITY 0x04 92 #define G_RAID3_BIO_CFLAG_NODISK 0x08 93 #define G_RAID3_BIO_CFLAG_REGSYNC 0x10 94 #define G_RAID3_BIO_CFLAG_MASK (G_RAID3_BIO_CFLAG_REGULAR | \ 95 G_RAID3_BIO_CFLAG_SYNC | \ 96 G_RAID3_BIO_CFLAG_PARITY | \ 97 G_RAID3_BIO_CFLAG_NODISK | \ 98 G_RAID3_BIO_CFLAG_REGSYNC) 99 100 #define G_RAID3_BIO_PFLAG_DEGRADED 0x01 101 #define G_RAID3_BIO_PFLAG_NOPARITY 0x02 102 #define G_RAID3_BIO_PFLAG_VERIFY 0x04 103 #define G_RAID3_BIO_PFLAG_MASK (G_RAID3_BIO_PFLAG_DEGRADED | \ 104 G_RAID3_BIO_PFLAG_NOPARITY | \ 105 G_RAID3_BIO_PFLAG_VERIFY) 106 107 /* 108 * Informations needed for synchronization. 109 */ 110 struct g_raid3_disk_sync { 111 struct g_consumer *ds_consumer; /* Consumer connected to our device. */ 112 off_t ds_offset; /* Offset of next request to send. */ 113 off_t ds_offset_done; /* Offset of already synchronized 114 region. */ 115 off_t ds_resync; /* Resynchronize from this offset. */ 116 u_int ds_syncid; /* Disk's synchronization ID. */ 117 u_char *ds_data; 118 }; 119 120 /* 121 * Informations needed for synchronization. 122 */ 123 struct g_raid3_device_sync { 124 struct g_geom *ds_geom; /* Synchronization geom. */ 125 }; 126 127 #define G_RAID3_DISK_STATE_NODISK 0 128 #define G_RAID3_DISK_STATE_NONE 1 129 #define G_RAID3_DISK_STATE_NEW 2 130 #define G_RAID3_DISK_STATE_ACTIVE 3 131 #define G_RAID3_DISK_STATE_STALE 4 132 #define G_RAID3_DISK_STATE_SYNCHRONIZING 5 133 #define G_RAID3_DISK_STATE_DISCONNECTED 6 134 #define G_RAID3_DISK_STATE_DESTROY 7 135 struct g_raid3_disk { 136 u_int d_no; /* Disk number. */ 137 struct g_consumer *d_consumer; /* Consumer. */ 138 struct g_raid3_softc *d_softc; /* Back-pointer to softc. */ 139 int d_state; /* Disk state. */ 140 uint64_t d_flags; /* Additional flags. */ 141 u_int d_genid; /* Disk's generation ID. */ 142 struct g_raid3_disk_sync d_sync; /* Sync information. */ 143 LIST_ENTRY(g_raid3_disk) d_next; 144 }; 145 #define d_name d_consumer->provider->name 146 147 #define G_RAID3_EVENT_DONTWAIT 0x1 148 #define G_RAID3_EVENT_WAIT 0x2 149 #define G_RAID3_EVENT_DEVICE 0x4 150 #define G_RAID3_EVENT_DONE 0x8 151 struct g_raid3_event { 152 struct g_raid3_disk *e_disk; 153 int e_state; 154 int e_flags; 155 int e_error; 156 TAILQ_ENTRY(g_raid3_event) e_next; 157 }; 158 159 #define G_RAID3_DEVICE_FLAG_DESTROY 0x0100000000000000ULL 160 #define G_RAID3_DEVICE_FLAG_WAIT 0x0200000000000000ULL 161 162 #define G_RAID3_DEVICE_STATE_STARTING 0 163 #define G_RAID3_DEVICE_STATE_DEGRADED 1 164 #define G_RAID3_DEVICE_STATE_COMPLETE 2 165 166 /* Bump syncid on first write. */ 167 #define G_RAID3_BUMP_SYNCID 0x1 168 /* Bump genid immediately. */ 169 #define G_RAID3_BUMP_GENID 0x2 170 171 struct g_raid3_softc { 172 u_int sc_state; /* Device state. */ 173 uint64_t sc_mediasize; /* Device size. */ 174 uint32_t sc_sectorsize; /* Sector size. */ 175 uint64_t sc_flags; /* Additional flags. */ 176 177 struct g_geom *sc_geom; 178 struct g_provider *sc_provider; 179 180 uint32_t sc_id; /* Device unique ID. */ 181 182 struct bio_queue_head sc_queue; 183 struct mtx sc_queue_mtx; 184 struct proc *sc_worker; 185 186 struct g_raid3_disk *sc_disks; 187 u_int sc_ndisks; /* Number of disks. */ 188 u_int sc_round_robin; 189 struct g_raid3_disk *sc_syncdisk; 190 191 uma_zone_t sc_zone_64k; 192 uma_zone_t sc_zone_16k; 193 uma_zone_t sc_zone_4k; 194 195 u_int sc_genid; /* Generation ID. */ 196 u_int sc_syncid; /* Synchronization ID. */ 197 int sc_bump_id; 198 struct g_raid3_device_sync sc_sync; 199 int sc_idle; /* DIRTY flags removed. */ 200 201 TAILQ_HEAD(, g_raid3_event) sc_events; 202 struct mtx sc_events_mtx; 203 204 struct callout sc_callout; 205 206 struct root_hold_token *sc_rootmount; 207 }; 208 #define sc_name sc_geom->name 209 210 const char *g_raid3_get_diskname(struct g_raid3_disk *disk); 211 u_int g_raid3_ndisks(struct g_raid3_softc *sc, int state); 212 int g_raid3_destroy(struct g_raid3_softc *sc, boolean_t force); 213 int g_raid3_event_send(void *arg, int state, int flags); 214 struct g_raid3_metadata; 215 int g_raid3_add_disk(struct g_raid3_softc *sc, struct g_provider *pp, 216 struct g_raid3_metadata *md); 217 int g_raid3_read_metadata(struct g_consumer *cp, struct g_raid3_metadata *md); 218 void g_raid3_fill_metadata(struct g_raid3_disk *disk, 219 struct g_raid3_metadata *md); 220 int g_raid3_clear_metadata(struct g_raid3_disk *disk); 221 void g_raid3_update_metadata(struct g_raid3_disk *disk); 222 223 g_ctl_req_t g_raid3_config; 224 #endif /* _KERNEL */ 225 226 struct g_raid3_metadata { 227 char md_magic[16]; /* Magic value. */ 228 uint32_t md_version; /* Version number. */ 229 char md_name[16]; /* Device name. */ 230 uint32_t md_id; /* Device unique ID. */ 231 uint16_t md_no; /* Component number. */ 232 uint16_t md_all; /* Number of disks in device. */ 233 uint32_t md_genid; /* Generation ID. */ 234 uint32_t md_syncid; /* Synchronization ID. */ 235 uint64_t md_mediasize; /* Size of whole device. */ 236 uint32_t md_sectorsize; /* Sector size. */ 237 uint64_t md_sync_offset; /* Synchronized offset. */ 238 uint64_t md_mflags; /* Additional device flags. */ 239 uint64_t md_dflags; /* Additional disk flags. */ 240 char md_provider[16]; /* Hardcoded provider. */ 241 uint64_t md_provsize; /* Provider's size. */ 242 u_char md_hash[16]; /* MD5 hash. */ 243 }; 244 static __inline void 245 raid3_metadata_encode(struct g_raid3_metadata *md, u_char *data) 246 { 247 MD5_CTX ctx; 248 249 bcopy(md->md_magic, data, 16); 250 le32enc(data + 16, md->md_version); 251 bcopy(md->md_name, data + 20, 16); 252 le32enc(data + 36, md->md_id); 253 le16enc(data + 40, md->md_no); 254 le16enc(data + 42, md->md_all); 255 le32enc(data + 44, md->md_genid); 256 le32enc(data + 48, md->md_syncid); 257 le64enc(data + 52, md->md_mediasize); 258 le32enc(data + 60, md->md_sectorsize); 259 le64enc(data + 64, md->md_sync_offset); 260 le64enc(data + 72, md->md_mflags); 261 le64enc(data + 80, md->md_dflags); 262 bcopy(md->md_provider, data + 88, 16); 263 le64enc(data + 104, md->md_provsize); 264 MD5Init(&ctx); 265 MD5Update(&ctx, data, 112); 266 MD5Final(md->md_hash, &ctx); 267 bcopy(md->md_hash, data + 112, 16); 268 } 269 static __inline int 270 raid3_metadata_decode_v0v1v2(const u_char *data, struct g_raid3_metadata *md) 271 { 272 MD5_CTX ctx; 273 274 bcopy(data + 20, md->md_name, 16); 275 md->md_id = le32dec(data + 36); 276 md->md_no = le16dec(data + 40); 277 md->md_all = le16dec(data + 42); 278 md->md_syncid = le32dec(data + 44); 279 md->md_mediasize = le64dec(data + 48); 280 md->md_sectorsize = le32dec(data + 56); 281 md->md_sync_offset = le64dec(data + 60); 282 md->md_mflags = le64dec(data + 68); 283 md->md_dflags = le64dec(data + 76); 284 bcopy(data + 84, md->md_provider, 16); 285 bcopy(data + 100, md->md_hash, 16); 286 MD5Init(&ctx); 287 MD5Update(&ctx, data, 100); 288 MD5Final(md->md_hash, &ctx); 289 if (bcmp(md->md_hash, data + 100, 16) != 0) 290 return (EINVAL); 291 292 /* New fields. */ 293 md->md_genid = 0; 294 md->md_provsize = 0; 295 296 return (0); 297 } 298 static __inline int 299 raid3_metadata_decode_v3(const u_char *data, struct g_raid3_metadata *md) 300 { 301 MD5_CTX ctx; 302 303 bcopy(data + 20, md->md_name, 16); 304 md->md_id = le32dec(data + 36); 305 md->md_no = le16dec(data + 40); 306 md->md_all = le16dec(data + 42); 307 md->md_genid = le32dec(data + 44); 308 md->md_syncid = le32dec(data + 48); 309 md->md_mediasize = le64dec(data + 52); 310 md->md_sectorsize = le32dec(data + 60); 311 md->md_sync_offset = le64dec(data + 64); 312 md->md_mflags = le64dec(data + 72); 313 md->md_dflags = le64dec(data + 80); 314 bcopy(data + 88, md->md_provider, 16); 315 bcopy(data + 104, md->md_hash, 16); 316 MD5Init(&ctx); 317 MD5Update(&ctx, data, 104); 318 MD5Final(md->md_hash, &ctx); 319 if (bcmp(md->md_hash, data + 104, 16) != 0) 320 return (EINVAL); 321 322 /* New fields. */ 323 md->md_provsize = 0; 324 325 return (0); 326 } 327 static __inline int 328 raid3_metadata_decode_v4(const u_char *data, struct g_raid3_metadata *md) 329 { 330 MD5_CTX ctx; 331 332 bcopy(data + 20, md->md_name, 16); 333 md->md_id = le32dec(data + 36); 334 md->md_no = le16dec(data + 40); 335 md->md_all = le16dec(data + 42); 336 md->md_genid = le32dec(data + 44); 337 md->md_syncid = le32dec(data + 48); 338 md->md_mediasize = le64dec(data + 52); 339 md->md_sectorsize = le32dec(data + 60); 340 md->md_sync_offset = le64dec(data + 64); 341 md->md_mflags = le64dec(data + 72); 342 md->md_dflags = le64dec(data + 80); 343 bcopy(data + 88, md->md_provider, 16); 344 md->md_provsize = le64dec(data + 104); 345 bcopy(data + 112, md->md_hash, 16); 346 MD5Init(&ctx); 347 MD5Update(&ctx, data, 112); 348 MD5Final(md->md_hash, &ctx); 349 if (bcmp(md->md_hash, data + 112, 16) != 0) 350 return (EINVAL); 351 return (0); 352 } 353 static __inline int 354 raid3_metadata_decode(const u_char *data, struct g_raid3_metadata *md) 355 { 356 int error; 357 358 bcopy(data, md->md_magic, 16); 359 md->md_version = le32dec(data + 16); 360 switch (md->md_version) { 361 case 0: 362 case 1: 363 case 2: 364 error = raid3_metadata_decode_v0v1v2(data, md); 365 break; 366 case 3: 367 error = raid3_metadata_decode_v3(data, md); 368 break; 369 case 4: 370 error = raid3_metadata_decode_v4(data, md); 371 break; 372 default: 373 error = EINVAL; 374 break; 375 } 376 return (error); 377 } 378 379 static __inline void 380 raid3_metadata_dump(const struct g_raid3_metadata *md) 381 { 382 static const char hex[] = "0123456789abcdef"; 383 char hash[16 * 2 + 1]; 384 u_int i; 385 386 printf(" magic: %s\n", md->md_magic); 387 printf(" version: %u\n", (u_int)md->md_version); 388 printf(" name: %s\n", md->md_name); 389 printf(" id: %u\n", (u_int)md->md_id); 390 printf(" no: %u\n", (u_int)md->md_no); 391 printf(" all: %u\n", (u_int)md->md_all); 392 printf(" genid: %u\n", (u_int)md->md_genid); 393 printf(" syncid: %u\n", (u_int)md->md_syncid); 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_RAID3_DEVICE_FLAG_NOAUTOSYNC) != 0) 402 printf(" NOAUTOSYNC"); 403 if ((md->md_mflags & G_RAID3_DEVICE_FLAG_ROUND_ROBIN) != 0) 404 printf(" ROUND-ROBIN"); 405 if ((md->md_mflags & G_RAID3_DEVICE_FLAG_VERIFY) != 0) 406 printf(" VERIFY"); 407 } 408 printf("\n"); 409 printf(" dflags:"); 410 if (md->md_dflags == 0) 411 printf(" NONE"); 412 else { 413 if ((md->md_dflags & G_RAID3_DISK_FLAG_DIRTY) != 0) 414 printf(" DIRTY"); 415 if ((md->md_dflags & G_RAID3_DISK_FLAG_SYNCHRONIZING) != 0) 416 printf(" SYNCHRONIZING"); 417 if ((md->md_dflags & G_RAID3_DISK_FLAG_FORCE_SYNC) != 0) 418 printf(" FORCE_SYNC"); 419 } 420 printf("\n"); 421 printf("hcprovider: %s\n", md->md_provider); 422 printf(" provsize: %ju\n", (uintmax_t)md->md_provsize); 423 bzero(hash, sizeof(hash)); 424 for (i = 0; i < 16; i++) { 425 hash[i * 2] = hex[md->md_hash[i] >> 4]; 426 hash[i * 2 + 1] = hex[md->md_hash[i] & 0x0f]; 427 } 428 printf(" MD5 hash: %s\n", hash); 429 } 430 #endif /* !_G_RAID3_H_ */ 431