1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #include <sys/param.h> 30 #include <sys/systm.h> 31 #include <sys/bio.h> 32 #include <sys/eventhandler.h> 33 #include <sys/kernel.h> 34 #include <sys/kthread.h> 35 #include <sys/limits.h> 36 #include <sys/lock.h> 37 #include <sys/malloc.h> 38 #include <sys/module.h> 39 #include <sys/mutex.h> 40 #include <sys/proc.h> 41 #include <sys/reboot.h> 42 #include <sys/sbuf.h> 43 #include <sys/sched.h> 44 #include <sys/sysctl.h> 45 46 #include <vm/uma.h> 47 48 #include <geom/geom.h> 49 #include <geom/geom_dbg.h> 50 #include <geom/geom_disk.h> 51 #include <geom/raid/g_raid.h> 52 #include "g_raid_md_if.h" 53 #include "g_raid_tr_if.h" 54 55 static MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data"); 56 57 SYSCTL_DECL(_kern_geom); 58 SYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 59 "GEOM_RAID stuff"); 60 int g_raid_enable = 1; 61 SYSCTL_INT(_kern_geom_raid, OID_AUTO, enable, CTLFLAG_RWTUN, 62 &g_raid_enable, 0, "Enable on-disk metadata taste"); 63 u_int g_raid_aggressive_spare = 0; 64 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RWTUN, 65 &g_raid_aggressive_spare, 0, "Use disks without metadata as spare"); 66 u_int g_raid_debug = 0; 67 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RWTUN, &g_raid_debug, 0, 68 "Debug level"); 69 int g_raid_read_err_thresh = 10; 70 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RWTUN, 71 &g_raid_read_err_thresh, 0, 72 "Number of read errors equated to disk failure"); 73 u_int g_raid_start_timeout = 30; 74 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RWTUN, 75 &g_raid_start_timeout, 0, 76 "Time to wait for all array components"); 77 static u_int g_raid_clean_time = 5; 78 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RWTUN, 79 &g_raid_clean_time, 0, "Mark volume as clean when idling"); 80 static u_int g_raid_disconnect_on_failure = 1; 81 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN, 82 &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 83 static u_int g_raid_name_format = 0; 84 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RWTUN, 85 &g_raid_name_format, 0, "Providers name format."); 86 static u_int g_raid_idle_threshold = 1000000; 87 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RWTUN, 88 &g_raid_idle_threshold, 1000000, 89 "Time in microseconds to consider a volume idle."); 90 91 #define MSLEEP(rv, ident, mtx, priority, wmesg, timeout) do { \ 92 G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 93 rv = msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 94 G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 95 } while (0) 96 97 LIST_HEAD(, g_raid_md_class) g_raid_md_classes = 98 LIST_HEAD_INITIALIZER(g_raid_md_classes); 99 100 LIST_HEAD(, g_raid_tr_class) g_raid_tr_classes = 101 LIST_HEAD_INITIALIZER(g_raid_tr_classes); 102 103 LIST_HEAD(, g_raid_volume) g_raid_volumes = 104 LIST_HEAD_INITIALIZER(g_raid_volumes); 105 106 static eventhandler_tag g_raid_post_sync = NULL; 107 static int g_raid_started = 0; 108 static int g_raid_shutdown = 0; 109 110 static int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp, 111 struct g_geom *gp); 112 static g_taste_t g_raid_taste; 113 static void g_raid_init(struct g_class *mp); 114 static void g_raid_fini(struct g_class *mp); 115 116 struct g_class g_raid_class = { 117 .name = G_RAID_CLASS_NAME, 118 .version = G_VERSION, 119 .ctlreq = g_raid_ctl, 120 .taste = g_raid_taste, 121 .destroy_geom = g_raid_destroy_geom, 122 .init = g_raid_init, 123 .fini = g_raid_fini 124 }; 125 126 static void g_raid_destroy_provider(struct g_raid_volume *vol); 127 static int g_raid_update_disk(struct g_raid_disk *disk, u_int event); 128 static int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event); 129 static int g_raid_update_volume(struct g_raid_volume *vol, u_int event); 130 static int g_raid_update_node(struct g_raid_softc *sc, u_int event); 131 static void g_raid_dumpconf(struct sbuf *sb, const char *indent, 132 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 133 static void g_raid_start(struct bio *bp); 134 static void g_raid_start_request(struct bio *bp); 135 static void g_raid_disk_done(struct bio *bp); 136 static void g_raid_poll(struct g_raid_softc *sc); 137 138 static const char * 139 g_raid_node_event2str(int event) 140 { 141 142 switch (event) { 143 case G_RAID_NODE_E_WAKE: 144 return ("WAKE"); 145 case G_RAID_NODE_E_START: 146 return ("START"); 147 default: 148 return ("INVALID"); 149 } 150 } 151 152 const char * 153 g_raid_disk_state2str(int state) 154 { 155 156 switch (state) { 157 case G_RAID_DISK_S_NONE: 158 return ("NONE"); 159 case G_RAID_DISK_S_OFFLINE: 160 return ("OFFLINE"); 161 case G_RAID_DISK_S_DISABLED: 162 return ("DISABLED"); 163 case G_RAID_DISK_S_FAILED: 164 return ("FAILED"); 165 case G_RAID_DISK_S_STALE_FAILED: 166 return ("STALE_FAILED"); 167 case G_RAID_DISK_S_SPARE: 168 return ("SPARE"); 169 case G_RAID_DISK_S_STALE: 170 return ("STALE"); 171 case G_RAID_DISK_S_ACTIVE: 172 return ("ACTIVE"); 173 default: 174 return ("INVALID"); 175 } 176 } 177 178 static const char * 179 g_raid_disk_event2str(int event) 180 { 181 182 switch (event) { 183 case G_RAID_DISK_E_DISCONNECTED: 184 return ("DISCONNECTED"); 185 default: 186 return ("INVALID"); 187 } 188 } 189 190 const char * 191 g_raid_subdisk_state2str(int state) 192 { 193 194 switch (state) { 195 case G_RAID_SUBDISK_S_NONE: 196 return ("NONE"); 197 case G_RAID_SUBDISK_S_FAILED: 198 return ("FAILED"); 199 case G_RAID_SUBDISK_S_NEW: 200 return ("NEW"); 201 case G_RAID_SUBDISK_S_REBUILD: 202 return ("REBUILD"); 203 case G_RAID_SUBDISK_S_UNINITIALIZED: 204 return ("UNINITIALIZED"); 205 case G_RAID_SUBDISK_S_STALE: 206 return ("STALE"); 207 case G_RAID_SUBDISK_S_RESYNC: 208 return ("RESYNC"); 209 case G_RAID_SUBDISK_S_ACTIVE: 210 return ("ACTIVE"); 211 default: 212 return ("INVALID"); 213 } 214 } 215 216 static const char * 217 g_raid_subdisk_event2str(int event) 218 { 219 220 switch (event) { 221 case G_RAID_SUBDISK_E_NEW: 222 return ("NEW"); 223 case G_RAID_SUBDISK_E_FAILED: 224 return ("FAILED"); 225 case G_RAID_SUBDISK_E_DISCONNECTED: 226 return ("DISCONNECTED"); 227 default: 228 return ("INVALID"); 229 } 230 } 231 232 const char * 233 g_raid_volume_state2str(int state) 234 { 235 236 switch (state) { 237 case G_RAID_VOLUME_S_STARTING: 238 return ("STARTING"); 239 case G_RAID_VOLUME_S_BROKEN: 240 return ("BROKEN"); 241 case G_RAID_VOLUME_S_DEGRADED: 242 return ("DEGRADED"); 243 case G_RAID_VOLUME_S_SUBOPTIMAL: 244 return ("SUBOPTIMAL"); 245 case G_RAID_VOLUME_S_OPTIMAL: 246 return ("OPTIMAL"); 247 case G_RAID_VOLUME_S_UNSUPPORTED: 248 return ("UNSUPPORTED"); 249 case G_RAID_VOLUME_S_STOPPED: 250 return ("STOPPED"); 251 default: 252 return ("INVALID"); 253 } 254 } 255 256 static const char * 257 g_raid_volume_event2str(int event) 258 { 259 260 switch (event) { 261 case G_RAID_VOLUME_E_UP: 262 return ("UP"); 263 case G_RAID_VOLUME_E_DOWN: 264 return ("DOWN"); 265 case G_RAID_VOLUME_E_START: 266 return ("START"); 267 case G_RAID_VOLUME_E_STARTMD: 268 return ("STARTMD"); 269 default: 270 return ("INVALID"); 271 } 272 } 273 274 const char * 275 g_raid_volume_level2str(int level, int qual) 276 { 277 278 switch (level) { 279 case G_RAID_VOLUME_RL_RAID0: 280 return ("RAID0"); 281 case G_RAID_VOLUME_RL_RAID1: 282 return ("RAID1"); 283 case G_RAID_VOLUME_RL_RAID3: 284 if (qual == G_RAID_VOLUME_RLQ_R3P0) 285 return ("RAID3-P0"); 286 if (qual == G_RAID_VOLUME_RLQ_R3PN) 287 return ("RAID3-PN"); 288 return ("RAID3"); 289 case G_RAID_VOLUME_RL_RAID4: 290 if (qual == G_RAID_VOLUME_RLQ_R4P0) 291 return ("RAID4-P0"); 292 if (qual == G_RAID_VOLUME_RLQ_R4PN) 293 return ("RAID4-PN"); 294 return ("RAID4"); 295 case G_RAID_VOLUME_RL_RAID5: 296 if (qual == G_RAID_VOLUME_RLQ_R5RA) 297 return ("RAID5-RA"); 298 if (qual == G_RAID_VOLUME_RLQ_R5RS) 299 return ("RAID5-RS"); 300 if (qual == G_RAID_VOLUME_RLQ_R5LA) 301 return ("RAID5-LA"); 302 if (qual == G_RAID_VOLUME_RLQ_R5LS) 303 return ("RAID5-LS"); 304 return ("RAID5"); 305 case G_RAID_VOLUME_RL_RAID6: 306 if (qual == G_RAID_VOLUME_RLQ_R6RA) 307 return ("RAID6-RA"); 308 if (qual == G_RAID_VOLUME_RLQ_R6RS) 309 return ("RAID6-RS"); 310 if (qual == G_RAID_VOLUME_RLQ_R6LA) 311 return ("RAID6-LA"); 312 if (qual == G_RAID_VOLUME_RLQ_R6LS) 313 return ("RAID6-LS"); 314 return ("RAID6"); 315 case G_RAID_VOLUME_RL_RAIDMDF: 316 if (qual == G_RAID_VOLUME_RLQ_RMDFRA) 317 return ("RAIDMDF-RA"); 318 if (qual == G_RAID_VOLUME_RLQ_RMDFRS) 319 return ("RAIDMDF-RS"); 320 if (qual == G_RAID_VOLUME_RLQ_RMDFLA) 321 return ("RAIDMDF-LA"); 322 if (qual == G_RAID_VOLUME_RLQ_RMDFLS) 323 return ("RAIDMDF-LS"); 324 return ("RAIDMDF"); 325 case G_RAID_VOLUME_RL_RAID1E: 326 if (qual == G_RAID_VOLUME_RLQ_R1EA) 327 return ("RAID1E-A"); 328 if (qual == G_RAID_VOLUME_RLQ_R1EO) 329 return ("RAID1E-O"); 330 return ("RAID1E"); 331 case G_RAID_VOLUME_RL_SINGLE: 332 return ("SINGLE"); 333 case G_RAID_VOLUME_RL_CONCAT: 334 return ("CONCAT"); 335 case G_RAID_VOLUME_RL_RAID5E: 336 if (qual == G_RAID_VOLUME_RLQ_R5ERA) 337 return ("RAID5E-RA"); 338 if (qual == G_RAID_VOLUME_RLQ_R5ERS) 339 return ("RAID5E-RS"); 340 if (qual == G_RAID_VOLUME_RLQ_R5ELA) 341 return ("RAID5E-LA"); 342 if (qual == G_RAID_VOLUME_RLQ_R5ELS) 343 return ("RAID5E-LS"); 344 return ("RAID5E"); 345 case G_RAID_VOLUME_RL_RAID5EE: 346 if (qual == G_RAID_VOLUME_RLQ_R5EERA) 347 return ("RAID5EE-RA"); 348 if (qual == G_RAID_VOLUME_RLQ_R5EERS) 349 return ("RAID5EE-RS"); 350 if (qual == G_RAID_VOLUME_RLQ_R5EELA) 351 return ("RAID5EE-LA"); 352 if (qual == G_RAID_VOLUME_RLQ_R5EELS) 353 return ("RAID5EE-LS"); 354 return ("RAID5EE"); 355 case G_RAID_VOLUME_RL_RAID5R: 356 if (qual == G_RAID_VOLUME_RLQ_R5RRA) 357 return ("RAID5R-RA"); 358 if (qual == G_RAID_VOLUME_RLQ_R5RRS) 359 return ("RAID5R-RS"); 360 if (qual == G_RAID_VOLUME_RLQ_R5RLA) 361 return ("RAID5R-LA"); 362 if (qual == G_RAID_VOLUME_RLQ_R5RLS) 363 return ("RAID5R-LS"); 364 return ("RAID5E"); 365 default: 366 return ("UNKNOWN"); 367 } 368 } 369 370 int 371 g_raid_volume_str2level(const char *str, int *level, int *qual) 372 { 373 374 *level = G_RAID_VOLUME_RL_UNKNOWN; 375 *qual = G_RAID_VOLUME_RLQ_NONE; 376 if (strcasecmp(str, "RAID0") == 0) 377 *level = G_RAID_VOLUME_RL_RAID0; 378 else if (strcasecmp(str, "RAID1") == 0) 379 *level = G_RAID_VOLUME_RL_RAID1; 380 else if (strcasecmp(str, "RAID3-P0") == 0) { 381 *level = G_RAID_VOLUME_RL_RAID3; 382 *qual = G_RAID_VOLUME_RLQ_R3P0; 383 } else if (strcasecmp(str, "RAID3-PN") == 0 || 384 strcasecmp(str, "RAID3") == 0) { 385 *level = G_RAID_VOLUME_RL_RAID3; 386 *qual = G_RAID_VOLUME_RLQ_R3PN; 387 } else if (strcasecmp(str, "RAID4-P0") == 0) { 388 *level = G_RAID_VOLUME_RL_RAID4; 389 *qual = G_RAID_VOLUME_RLQ_R4P0; 390 } else if (strcasecmp(str, "RAID4-PN") == 0 || 391 strcasecmp(str, "RAID4") == 0) { 392 *level = G_RAID_VOLUME_RL_RAID4; 393 *qual = G_RAID_VOLUME_RLQ_R4PN; 394 } else if (strcasecmp(str, "RAID5-RA") == 0) { 395 *level = G_RAID_VOLUME_RL_RAID5; 396 *qual = G_RAID_VOLUME_RLQ_R5RA; 397 } else if (strcasecmp(str, "RAID5-RS") == 0) { 398 *level = G_RAID_VOLUME_RL_RAID5; 399 *qual = G_RAID_VOLUME_RLQ_R5RS; 400 } else if (strcasecmp(str, "RAID5") == 0 || 401 strcasecmp(str, "RAID5-LA") == 0) { 402 *level = G_RAID_VOLUME_RL_RAID5; 403 *qual = G_RAID_VOLUME_RLQ_R5LA; 404 } else if (strcasecmp(str, "RAID5-LS") == 0) { 405 *level = G_RAID_VOLUME_RL_RAID5; 406 *qual = G_RAID_VOLUME_RLQ_R5LS; 407 } else if (strcasecmp(str, "RAID6-RA") == 0) { 408 *level = G_RAID_VOLUME_RL_RAID6; 409 *qual = G_RAID_VOLUME_RLQ_R6RA; 410 } else if (strcasecmp(str, "RAID6-RS") == 0) { 411 *level = G_RAID_VOLUME_RL_RAID6; 412 *qual = G_RAID_VOLUME_RLQ_R6RS; 413 } else if (strcasecmp(str, "RAID6") == 0 || 414 strcasecmp(str, "RAID6-LA") == 0) { 415 *level = G_RAID_VOLUME_RL_RAID6; 416 *qual = G_RAID_VOLUME_RLQ_R6LA; 417 } else if (strcasecmp(str, "RAID6-LS") == 0) { 418 *level = G_RAID_VOLUME_RL_RAID6; 419 *qual = G_RAID_VOLUME_RLQ_R6LS; 420 } else if (strcasecmp(str, "RAIDMDF-RA") == 0) { 421 *level = G_RAID_VOLUME_RL_RAIDMDF; 422 *qual = G_RAID_VOLUME_RLQ_RMDFRA; 423 } else if (strcasecmp(str, "RAIDMDF-RS") == 0) { 424 *level = G_RAID_VOLUME_RL_RAIDMDF; 425 *qual = G_RAID_VOLUME_RLQ_RMDFRS; 426 } else if (strcasecmp(str, "RAIDMDF") == 0 || 427 strcasecmp(str, "RAIDMDF-LA") == 0) { 428 *level = G_RAID_VOLUME_RL_RAIDMDF; 429 *qual = G_RAID_VOLUME_RLQ_RMDFLA; 430 } else if (strcasecmp(str, "RAIDMDF-LS") == 0) { 431 *level = G_RAID_VOLUME_RL_RAIDMDF; 432 *qual = G_RAID_VOLUME_RLQ_RMDFLS; 433 } else if (strcasecmp(str, "RAID10") == 0 || 434 strcasecmp(str, "RAID1E") == 0 || 435 strcasecmp(str, "RAID1E-A") == 0) { 436 *level = G_RAID_VOLUME_RL_RAID1E; 437 *qual = G_RAID_VOLUME_RLQ_R1EA; 438 } else if (strcasecmp(str, "RAID1E-O") == 0) { 439 *level = G_RAID_VOLUME_RL_RAID1E; 440 *qual = G_RAID_VOLUME_RLQ_R1EO; 441 } else if (strcasecmp(str, "SINGLE") == 0) 442 *level = G_RAID_VOLUME_RL_SINGLE; 443 else if (strcasecmp(str, "CONCAT") == 0) 444 *level = G_RAID_VOLUME_RL_CONCAT; 445 else if (strcasecmp(str, "RAID5E-RA") == 0) { 446 *level = G_RAID_VOLUME_RL_RAID5E; 447 *qual = G_RAID_VOLUME_RLQ_R5ERA; 448 } else if (strcasecmp(str, "RAID5E-RS") == 0) { 449 *level = G_RAID_VOLUME_RL_RAID5E; 450 *qual = G_RAID_VOLUME_RLQ_R5ERS; 451 } else if (strcasecmp(str, "RAID5E") == 0 || 452 strcasecmp(str, "RAID5E-LA") == 0) { 453 *level = G_RAID_VOLUME_RL_RAID5E; 454 *qual = G_RAID_VOLUME_RLQ_R5ELA; 455 } else if (strcasecmp(str, "RAID5E-LS") == 0) { 456 *level = G_RAID_VOLUME_RL_RAID5E; 457 *qual = G_RAID_VOLUME_RLQ_R5ELS; 458 } else if (strcasecmp(str, "RAID5EE-RA") == 0) { 459 *level = G_RAID_VOLUME_RL_RAID5EE; 460 *qual = G_RAID_VOLUME_RLQ_R5EERA; 461 } else if (strcasecmp(str, "RAID5EE-RS") == 0) { 462 *level = G_RAID_VOLUME_RL_RAID5EE; 463 *qual = G_RAID_VOLUME_RLQ_R5EERS; 464 } else if (strcasecmp(str, "RAID5EE") == 0 || 465 strcasecmp(str, "RAID5EE-LA") == 0) { 466 *level = G_RAID_VOLUME_RL_RAID5EE; 467 *qual = G_RAID_VOLUME_RLQ_R5EELA; 468 } else if (strcasecmp(str, "RAID5EE-LS") == 0) { 469 *level = G_RAID_VOLUME_RL_RAID5EE; 470 *qual = G_RAID_VOLUME_RLQ_R5EELS; 471 } else if (strcasecmp(str, "RAID5R-RA") == 0) { 472 *level = G_RAID_VOLUME_RL_RAID5R; 473 *qual = G_RAID_VOLUME_RLQ_R5RRA; 474 } else if (strcasecmp(str, "RAID5R-RS") == 0) { 475 *level = G_RAID_VOLUME_RL_RAID5R; 476 *qual = G_RAID_VOLUME_RLQ_R5RRS; 477 } else if (strcasecmp(str, "RAID5R") == 0 || 478 strcasecmp(str, "RAID5R-LA") == 0) { 479 *level = G_RAID_VOLUME_RL_RAID5R; 480 *qual = G_RAID_VOLUME_RLQ_R5RLA; 481 } else if (strcasecmp(str, "RAID5R-LS") == 0) { 482 *level = G_RAID_VOLUME_RL_RAID5R; 483 *qual = G_RAID_VOLUME_RLQ_R5RLS; 484 } else 485 return (-1); 486 return (0); 487 } 488 489 const char * 490 g_raid_get_diskname(struct g_raid_disk *disk) 491 { 492 493 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 494 return ("[unknown]"); 495 return (disk->d_consumer->provider->name); 496 } 497 498 void 499 g_raid_get_disk_info(struct g_raid_disk *disk) 500 { 501 struct g_consumer *cp = disk->d_consumer; 502 int error, len; 503 504 /* Read kernel dumping information. */ 505 disk->d_kd.offset = 0; 506 disk->d_kd.length = OFF_MAX; 507 len = sizeof(disk->d_kd); 508 error = g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); 509 if (error) 510 disk->d_kd.di.dumper = NULL; 511 if (disk->d_kd.di.dumper == NULL) 512 G_RAID_DEBUG1(2, disk->d_softc, 513 "Dumping not supported by %s: %d.", 514 cp->provider->name, error); 515 516 /* Read BIO_DELETE support. */ 517 error = g_getattr("GEOM::candelete", cp, &disk->d_candelete); 518 if (error) 519 disk->d_candelete = 0; 520 if (!disk->d_candelete) 521 G_RAID_DEBUG1(2, disk->d_softc, 522 "BIO_DELETE not supported by %s: %d.", 523 cp->provider->name, error); 524 525 /* Read rotation rate. */ 526 error = g_getattr("GEOM::rotation_rate", cp, &disk->d_rotation_rate); 527 if (error) 528 disk->d_rotation_rate = DISK_RR_UNKNOWN; 529 } 530 531 void 532 g_raid_report_disk_state(struct g_raid_disk *disk) 533 { 534 struct g_raid_subdisk *sd; 535 int len, state; 536 uint32_t s; 537 538 if (disk->d_consumer == NULL) 539 return; 540 if (disk->d_state == G_RAID_DISK_S_DISABLED) { 541 s = G_STATE_ACTIVE; /* XXX */ 542 } else if (disk->d_state == G_RAID_DISK_S_FAILED || 543 disk->d_state == G_RAID_DISK_S_STALE_FAILED) { 544 s = G_STATE_FAILED; 545 } else { 546 state = G_RAID_SUBDISK_S_ACTIVE; 547 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 548 if (sd->sd_state < state) 549 state = sd->sd_state; 550 } 551 if (state == G_RAID_SUBDISK_S_FAILED) 552 s = G_STATE_FAILED; 553 else if (state == G_RAID_SUBDISK_S_NEW || 554 state == G_RAID_SUBDISK_S_REBUILD) 555 s = G_STATE_REBUILD; 556 else if (state == G_RAID_SUBDISK_S_STALE || 557 state == G_RAID_SUBDISK_S_RESYNC) 558 s = G_STATE_RESYNC; 559 else 560 s = G_STATE_ACTIVE; 561 } 562 len = sizeof(s); 563 g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s); 564 G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.", 565 g_raid_get_diskname(disk), s); 566 } 567 568 void 569 g_raid_change_disk_state(struct g_raid_disk *disk, int state) 570 { 571 572 G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.", 573 g_raid_get_diskname(disk), 574 g_raid_disk_state2str(disk->d_state), 575 g_raid_disk_state2str(state)); 576 disk->d_state = state; 577 g_raid_report_disk_state(disk); 578 } 579 580 void 581 g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state) 582 { 583 584 G_RAID_DEBUG1(0, sd->sd_softc, 585 "Subdisk %s:%d-%s state changed from %s to %s.", 586 sd->sd_volume->v_name, sd->sd_pos, 587 sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]", 588 g_raid_subdisk_state2str(sd->sd_state), 589 g_raid_subdisk_state2str(state)); 590 sd->sd_state = state; 591 if (sd->sd_disk) 592 g_raid_report_disk_state(sd->sd_disk); 593 } 594 595 void 596 g_raid_change_volume_state(struct g_raid_volume *vol, int state) 597 { 598 599 G_RAID_DEBUG1(0, vol->v_softc, 600 "Volume %s state changed from %s to %s.", 601 vol->v_name, 602 g_raid_volume_state2str(vol->v_state), 603 g_raid_volume_state2str(state)); 604 vol->v_state = state; 605 } 606 607 /* 608 * --- Events handling functions --- 609 * Events in geom_raid are used to maintain subdisks and volumes status 610 * from one thread to simplify locking. 611 */ 612 static void 613 g_raid_event_free(struct g_raid_event *ep) 614 { 615 616 free(ep, M_RAID); 617 } 618 619 int 620 g_raid_event_send(void *arg, int event, int flags) 621 { 622 struct g_raid_softc *sc; 623 struct g_raid_event *ep; 624 int error; 625 626 if ((flags & G_RAID_EVENT_VOLUME) != 0) { 627 sc = ((struct g_raid_volume *)arg)->v_softc; 628 } else if ((flags & G_RAID_EVENT_DISK) != 0) { 629 sc = ((struct g_raid_disk *)arg)->d_softc; 630 } else if ((flags & G_RAID_EVENT_SUBDISK) != 0) { 631 sc = ((struct g_raid_subdisk *)arg)->sd_softc; 632 } else { 633 sc = arg; 634 } 635 ep = malloc(sizeof(*ep), M_RAID, 636 sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT); 637 if (ep == NULL) 638 return (ENOMEM); 639 ep->e_tgt = arg; 640 ep->e_event = event; 641 ep->e_flags = flags; 642 ep->e_error = 0; 643 G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc); 644 mtx_lock(&sc->sc_queue_mtx); 645 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 646 mtx_unlock(&sc->sc_queue_mtx); 647 wakeup(sc); 648 649 if ((flags & G_RAID_EVENT_WAIT) == 0) 650 return (0); 651 652 sx_assert(&sc->sc_lock, SX_XLOCKED); 653 G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep); 654 sx_xunlock(&sc->sc_lock); 655 while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) { 656 mtx_lock(&sc->sc_queue_mtx); 657 MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event", 658 hz * 5); 659 } 660 error = ep->e_error; 661 g_raid_event_free(ep); 662 sx_xlock(&sc->sc_lock); 663 return (error); 664 } 665 666 static void 667 g_raid_event_cancel(struct g_raid_softc *sc, void *tgt) 668 { 669 struct g_raid_event *ep, *tmpep; 670 671 sx_assert(&sc->sc_lock, SX_XLOCKED); 672 673 mtx_lock(&sc->sc_queue_mtx); 674 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 675 if (ep->e_tgt != tgt) 676 continue; 677 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 678 if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) 679 g_raid_event_free(ep); 680 else { 681 ep->e_error = ECANCELED; 682 wakeup(ep); 683 } 684 } 685 mtx_unlock(&sc->sc_queue_mtx); 686 } 687 688 static int 689 g_raid_event_check(struct g_raid_softc *sc, void *tgt) 690 { 691 struct g_raid_event *ep; 692 int res = 0; 693 694 sx_assert(&sc->sc_lock, SX_XLOCKED); 695 696 mtx_lock(&sc->sc_queue_mtx); 697 TAILQ_FOREACH(ep, &sc->sc_events, e_next) { 698 if (ep->e_tgt != tgt) 699 continue; 700 res = 1; 701 break; 702 } 703 mtx_unlock(&sc->sc_queue_mtx); 704 return (res); 705 } 706 707 /* 708 * Return the number of disks in given state. 709 * If state is equal to -1, count all connected disks. 710 */ 711 u_int 712 g_raid_ndisks(struct g_raid_softc *sc, int state) 713 { 714 struct g_raid_disk *disk; 715 u_int n; 716 717 sx_assert(&sc->sc_lock, SX_LOCKED); 718 719 n = 0; 720 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 721 if (disk->d_state == state || state == -1) 722 n++; 723 } 724 return (n); 725 } 726 727 /* 728 * Return the number of subdisks in given state. 729 * If state is equal to -1, count all connected disks. 730 */ 731 u_int 732 g_raid_nsubdisks(struct g_raid_volume *vol, int state) 733 { 734 struct g_raid_subdisk *subdisk; 735 struct g_raid_softc *sc __diagused; 736 u_int i, n ; 737 738 sc = vol->v_softc; 739 sx_assert(&sc->sc_lock, SX_LOCKED); 740 741 n = 0; 742 for (i = 0; i < vol->v_disks_count; i++) { 743 subdisk = &vol->v_subdisks[i]; 744 if ((state == -1 && 745 subdisk->sd_state != G_RAID_SUBDISK_S_NONE) || 746 subdisk->sd_state == state) 747 n++; 748 } 749 return (n); 750 } 751 752 /* 753 * Return the first subdisk in given state. 754 * If state is equal to -1, then the first connected disks. 755 */ 756 struct g_raid_subdisk * 757 g_raid_get_subdisk(struct g_raid_volume *vol, int state) 758 { 759 struct g_raid_subdisk *sd; 760 struct g_raid_softc *sc __diagused; 761 u_int i; 762 763 sc = vol->v_softc; 764 sx_assert(&sc->sc_lock, SX_LOCKED); 765 766 for (i = 0; i < vol->v_disks_count; i++) { 767 sd = &vol->v_subdisks[i]; 768 if ((state == -1 && 769 sd->sd_state != G_RAID_SUBDISK_S_NONE) || 770 sd->sd_state == state) 771 return (sd); 772 } 773 return (NULL); 774 } 775 776 struct g_consumer * 777 g_raid_open_consumer(struct g_raid_softc *sc, const char *name) 778 { 779 struct g_consumer *cp; 780 struct g_provider *pp; 781 782 g_topology_assert(); 783 784 pp = g_provider_by_name(name); 785 if (pp == NULL) 786 return (NULL); 787 cp = g_new_consumer(sc->sc_geom); 788 cp->flags |= G_CF_DIRECT_RECEIVE; 789 if (g_attach(cp, pp) != 0) { 790 g_destroy_consumer(cp); 791 return (NULL); 792 } 793 if (g_access(cp, 1, 1, 1) != 0) { 794 g_detach(cp); 795 g_destroy_consumer(cp); 796 return (NULL); 797 } 798 return (cp); 799 } 800 801 static u_int 802 g_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp) 803 { 804 struct bio *bp; 805 u_int nreqs = 0; 806 807 mtx_lock(&sc->sc_queue_mtx); 808 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 809 if (bp->bio_from == cp) 810 nreqs++; 811 } 812 mtx_unlock(&sc->sc_queue_mtx); 813 return (nreqs); 814 } 815 816 u_int 817 g_raid_nopens(struct g_raid_softc *sc) 818 { 819 struct g_raid_volume *vol; 820 u_int opens; 821 822 opens = 0; 823 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 824 if (vol->v_provider_open != 0) 825 opens++; 826 } 827 return (opens); 828 } 829 830 static int 831 g_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp) 832 { 833 834 if (cp->index > 0) { 835 G_RAID_DEBUG1(2, sc, 836 "I/O requests for %s exist, can't destroy it now.", 837 cp->provider->name); 838 return (1); 839 } 840 if (g_raid_nrequests(sc, cp) > 0) { 841 G_RAID_DEBUG1(2, sc, 842 "I/O requests for %s in queue, can't destroy it now.", 843 cp->provider->name); 844 return (1); 845 } 846 return (0); 847 } 848 849 static void 850 g_raid_destroy_consumer(void *arg, int flags __unused) 851 { 852 struct g_consumer *cp; 853 854 g_topology_assert(); 855 856 cp = arg; 857 G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 858 g_detach(cp); 859 g_destroy_consumer(cp); 860 } 861 862 void 863 g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp) 864 { 865 struct g_provider *pp; 866 int retaste_wait; 867 868 g_topology_assert_not(); 869 870 g_topology_lock(); 871 cp->private = NULL; 872 if (g_raid_consumer_is_busy(sc, cp)) 873 goto out; 874 pp = cp->provider; 875 retaste_wait = 0; 876 if (cp->acw == 1) { 877 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 878 retaste_wait = 1; 879 } 880 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 881 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 882 if (retaste_wait) { 883 /* 884 * After retaste event was send (inside g_access()), we can send 885 * event to detach and destroy consumer. 886 * A class, which has consumer to the given provider connected 887 * will not receive retaste event for the provider. 888 * This is the way how I ignore retaste events when I close 889 * consumers opened for write: I detach and destroy consumer 890 * after retaste event is sent. 891 */ 892 g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL); 893 goto out; 894 } 895 G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name); 896 g_detach(cp); 897 g_destroy_consumer(cp); 898 out: 899 g_topology_unlock(); 900 } 901 902 static void 903 g_raid_orphan(struct g_consumer *cp) 904 { 905 struct g_raid_disk *disk; 906 907 g_topology_assert(); 908 909 disk = cp->private; 910 if (disk == NULL) 911 return; 912 g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED, 913 G_RAID_EVENT_DISK); 914 } 915 916 static void 917 g_raid_clean(struct g_raid_volume *vol, int acw) 918 { 919 struct g_raid_softc *sc; 920 int timeout; 921 922 sc = vol->v_softc; 923 g_topology_assert_not(); 924 sx_assert(&sc->sc_lock, SX_XLOCKED); 925 926 // if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 927 // return; 928 if (!vol->v_dirty) 929 return; 930 if (vol->v_writes > 0) 931 return; 932 if (acw > 0 || (acw == -1 && 933 vol->v_provider != NULL && vol->v_provider->acw > 0)) { 934 timeout = g_raid_clean_time - (time_uptime - vol->v_last_write); 935 if (!g_raid_shutdown && timeout > 0) 936 return; 937 } 938 vol->v_dirty = 0; 939 G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.", 940 vol->v_name); 941 g_raid_write_metadata(sc, vol, NULL, NULL); 942 } 943 944 static void 945 g_raid_dirty(struct g_raid_volume *vol) 946 { 947 struct g_raid_softc *sc; 948 949 sc = vol->v_softc; 950 g_topology_assert_not(); 951 sx_assert(&sc->sc_lock, SX_XLOCKED); 952 953 // if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 954 // return; 955 vol->v_dirty = 1; 956 G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.", 957 vol->v_name); 958 g_raid_write_metadata(sc, vol, NULL, NULL); 959 } 960 961 void 962 g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp) 963 { 964 struct g_raid_volume *vol; 965 struct g_raid_subdisk *sd; 966 struct bio_queue_head queue; 967 struct bio *cbp; 968 int i; 969 970 vol = tr->tro_volume; 971 972 /* 973 * Allocate all bios before sending any request, so we can return 974 * ENOMEM in nice and clean way. 975 */ 976 bioq_init(&queue); 977 for (i = 0; i < vol->v_disks_count; i++) { 978 sd = &vol->v_subdisks[i]; 979 if (sd->sd_state == G_RAID_SUBDISK_S_NONE || 980 sd->sd_state == G_RAID_SUBDISK_S_FAILED) 981 continue; 982 cbp = g_clone_bio(bp); 983 if (cbp == NULL) 984 goto failure; 985 cbp->bio_caller1 = sd; 986 bioq_insert_tail(&queue, cbp); 987 } 988 while ((cbp = bioq_takefirst(&queue)) != NULL) { 989 sd = cbp->bio_caller1; 990 cbp->bio_caller1 = NULL; 991 g_raid_subdisk_iostart(sd, cbp); 992 } 993 return; 994 failure: 995 while ((cbp = bioq_takefirst(&queue)) != NULL) 996 g_destroy_bio(cbp); 997 if (bp->bio_error == 0) 998 bp->bio_error = ENOMEM; 999 g_raid_iodone(bp, bp->bio_error); 1000 } 1001 1002 static void 1003 g_raid_tr_kerneldump_common_done(struct bio *bp) 1004 { 1005 1006 bp->bio_flags |= BIO_DONE; 1007 } 1008 1009 int 1010 g_raid_tr_kerneldump_common(struct g_raid_tr_object *tr, 1011 void *virtual, vm_offset_t physical, off_t offset, size_t length) 1012 { 1013 struct g_raid_softc *sc; 1014 struct g_raid_volume *vol; 1015 struct bio bp; 1016 1017 vol = tr->tro_volume; 1018 sc = vol->v_softc; 1019 1020 g_reset_bio(&bp); 1021 bp.bio_cmd = BIO_WRITE; 1022 bp.bio_done = g_raid_tr_kerneldump_common_done; 1023 bp.bio_attribute = NULL; 1024 bp.bio_offset = offset; 1025 bp.bio_length = length; 1026 bp.bio_data = virtual; 1027 bp.bio_to = vol->v_provider; 1028 1029 g_raid_start(&bp); 1030 while (!(bp.bio_flags & BIO_DONE)) { 1031 G_RAID_DEBUG1(4, sc, "Poll..."); 1032 g_raid_poll(sc); 1033 DELAY(10); 1034 } 1035 1036 return (bp.bio_error != 0 ? EIO : 0); 1037 } 1038 1039 static int 1040 g_raid_dump(void *arg, void *virtual, off_t offset, size_t length) 1041 { 1042 struct g_raid_volume *vol; 1043 int error; 1044 1045 vol = (struct g_raid_volume *)arg; 1046 G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.", 1047 (long long unsigned)offset, (long long unsigned)length); 1048 1049 error = G_RAID_TR_KERNELDUMP(vol->v_tr, virtual, offset, length); 1050 return (error); 1051 } 1052 1053 static void 1054 g_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp) 1055 { 1056 struct g_kerneldump *gkd; 1057 struct g_provider *pp; 1058 struct g_raid_volume *vol; 1059 1060 gkd = (struct g_kerneldump*)bp->bio_data; 1061 pp = bp->bio_to; 1062 vol = pp->private; 1063 g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)", 1064 pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length); 1065 gkd->di.dumper = g_raid_dump; 1066 gkd->di.priv = vol; 1067 gkd->di.blocksize = vol->v_sectorsize; 1068 gkd->di.maxiosize = DFLTPHYS; 1069 gkd->di.mediaoffset = gkd->offset; 1070 if ((gkd->offset + gkd->length) > vol->v_mediasize) 1071 gkd->length = vol->v_mediasize - gkd->offset; 1072 gkd->di.mediasize = gkd->length; 1073 g_io_deliver(bp, 0); 1074 } 1075 1076 static void 1077 g_raid_candelete(struct g_raid_softc *sc, struct bio *bp) 1078 { 1079 struct g_provider *pp; 1080 struct g_raid_volume *vol; 1081 struct g_raid_subdisk *sd; 1082 int i, val; 1083 1084 pp = bp->bio_to; 1085 vol = pp->private; 1086 for (i = 0; i < vol->v_disks_count; i++) { 1087 sd = &vol->v_subdisks[i]; 1088 if (sd->sd_state == G_RAID_SUBDISK_S_NONE) 1089 continue; 1090 if (sd->sd_disk->d_candelete) 1091 break; 1092 } 1093 val = i < vol->v_disks_count; 1094 g_handleattr(bp, "GEOM::candelete", &val, sizeof(val)); 1095 } 1096 1097 static void 1098 g_raid_rotation_rate(struct g_raid_softc *sc, struct bio *bp) 1099 { 1100 struct g_raid_volume *vol; 1101 struct g_raid_subdisk *sd; 1102 bool first = true; 1103 uint16_t rr = DISK_RR_UNKNOWN; 1104 int i; 1105 1106 vol = bp->bio_to->private; 1107 for (i = 0; i < vol->v_disks_count; i++) { 1108 sd = &vol->v_subdisks[i]; 1109 if (sd->sd_state == G_RAID_SUBDISK_S_NONE) 1110 continue; 1111 if (first) 1112 rr = sd->sd_disk->d_rotation_rate; 1113 else if (rr != sd->sd_disk->d_rotation_rate) { 1114 rr = DISK_RR_UNKNOWN; 1115 break; 1116 } 1117 first = false; 1118 } 1119 g_handleattr(bp, "GEOM::rotation_rate", &rr, sizeof(rr)); 1120 } 1121 1122 static void 1123 g_raid_start(struct bio *bp) 1124 { 1125 struct g_raid_softc *sc; 1126 1127 sc = bp->bio_to->geom->softc; 1128 /* 1129 * If sc == NULL or there are no valid disks, provider's error 1130 * should be set and g_raid_start() should not be called at all. 1131 */ 1132 // KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING, 1133 // ("Provider's error should be set (error=%d)(mirror=%s).", 1134 // bp->bio_to->error, bp->bio_to->name)); 1135 G_RAID_LOGREQ(3, bp, "Request received."); 1136 1137 switch (bp->bio_cmd) { 1138 case BIO_READ: 1139 case BIO_WRITE: 1140 case BIO_DELETE: 1141 case BIO_FLUSH: 1142 case BIO_SPEEDUP: 1143 break; 1144 case BIO_GETATTR: 1145 if (!strcmp(bp->bio_attribute, "GEOM::candelete")) 1146 g_raid_candelete(sc, bp); 1147 else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump")) 1148 g_raid_kerneldump(sc, bp); 1149 else if (!strcmp(bp->bio_attribute, "GEOM::rotation_rate")) 1150 g_raid_rotation_rate(sc, bp); 1151 else 1152 g_io_deliver(bp, EOPNOTSUPP); 1153 return; 1154 default: 1155 g_io_deliver(bp, EOPNOTSUPP); 1156 return; 1157 } 1158 mtx_lock(&sc->sc_queue_mtx); 1159 bioq_insert_tail(&sc->sc_queue, bp); 1160 mtx_unlock(&sc->sc_queue_mtx); 1161 if (!dumping) { 1162 G_RAID_DEBUG1(4, sc, "Waking up %p.", sc); 1163 wakeup(sc); 1164 } 1165 } 1166 1167 static int 1168 g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len) 1169 { 1170 /* 1171 * 5 cases: 1172 * (1) bp entirely below NO 1173 * (2) bp entirely above NO 1174 * (3) bp start below, but end in range YES 1175 * (4) bp entirely within YES 1176 * (5) bp starts within, ends above YES 1177 * 1178 * lock range 10-19 (offset 10 length 10) 1179 * (1) 1-5: first if kicks it out 1180 * (2) 30-35: second if kicks it out 1181 * (3) 5-15: passes both ifs 1182 * (4) 12-14: passes both ifs 1183 * (5) 19-20: passes both 1184 */ 1185 off_t lend = lstart + len - 1; 1186 off_t bstart = bp->bio_offset; 1187 off_t bend = bp->bio_offset + bp->bio_length - 1; 1188 1189 if (bend < lstart) 1190 return (0); 1191 if (lend < bstart) 1192 return (0); 1193 return (1); 1194 } 1195 1196 static int 1197 g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp) 1198 { 1199 struct g_raid_lock *lp; 1200 1201 sx_assert(&vol->v_softc->sc_lock, SX_LOCKED); 1202 1203 LIST_FOREACH(lp, &vol->v_locks, l_next) { 1204 if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length)) 1205 return (1); 1206 } 1207 return (0); 1208 } 1209 1210 static void 1211 g_raid_start_request(struct bio *bp) 1212 { 1213 struct g_raid_softc *sc __diagused; 1214 struct g_raid_volume *vol; 1215 1216 sc = bp->bio_to->geom->softc; 1217 sx_assert(&sc->sc_lock, SX_LOCKED); 1218 vol = bp->bio_to->private; 1219 1220 /* 1221 * Check to see if this item is in a locked range. If so, 1222 * queue it to our locked queue and return. We'll requeue 1223 * it when the range is unlocked. Internal I/O for the 1224 * rebuild/rescan/recovery process is excluded from this 1225 * check so we can actually do the recovery. 1226 */ 1227 if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) && 1228 g_raid_is_in_locked_range(vol, bp)) { 1229 G_RAID_LOGREQ(3, bp, "Defer request."); 1230 bioq_insert_tail(&vol->v_locked, bp); 1231 return; 1232 } 1233 1234 /* 1235 * If we're actually going to do the write/delete, then 1236 * update the idle stats for the volume. 1237 */ 1238 if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 1239 if (!vol->v_dirty) 1240 g_raid_dirty(vol); 1241 vol->v_writes++; 1242 } 1243 1244 /* 1245 * Put request onto inflight queue, so we can check if new 1246 * synchronization requests don't collide with it. Then tell 1247 * the transformation layer to start the I/O. 1248 */ 1249 bioq_insert_tail(&vol->v_inflight, bp); 1250 G_RAID_LOGREQ(4, bp, "Request started"); 1251 G_RAID_TR_IOSTART(vol->v_tr, bp); 1252 } 1253 1254 static void 1255 g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp) 1256 { 1257 off_t off, len; 1258 struct bio *nbp; 1259 struct g_raid_lock *lp; 1260 1261 vol->v_pending_lock = 0; 1262 LIST_FOREACH(lp, &vol->v_locks, l_next) { 1263 if (lp->l_pending) { 1264 off = lp->l_offset; 1265 len = lp->l_length; 1266 lp->l_pending = 0; 1267 TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) { 1268 if (g_raid_bio_overlaps(nbp, off, len)) 1269 lp->l_pending++; 1270 } 1271 if (lp->l_pending) { 1272 vol->v_pending_lock = 1; 1273 G_RAID_DEBUG1(4, vol->v_softc, 1274 "Deferred lock(%jd, %jd) has %d pending", 1275 (intmax_t)off, (intmax_t)(off + len), 1276 lp->l_pending); 1277 continue; 1278 } 1279 G_RAID_DEBUG1(4, vol->v_softc, 1280 "Deferred lock of %jd to %jd completed", 1281 (intmax_t)off, (intmax_t)(off + len)); 1282 G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 1283 } 1284 } 1285 } 1286 1287 void 1288 g_raid_iodone(struct bio *bp, int error) 1289 { 1290 struct g_raid_softc *sc __diagused; 1291 struct g_raid_volume *vol; 1292 1293 sc = bp->bio_to->geom->softc; 1294 sx_assert(&sc->sc_lock, SX_LOCKED); 1295 vol = bp->bio_to->private; 1296 G_RAID_LOGREQ(3, bp, "Request done: %d.", error); 1297 1298 /* Update stats if we done write/delete. */ 1299 if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 1300 vol->v_writes--; 1301 vol->v_last_write = time_uptime; 1302 } 1303 1304 bioq_remove(&vol->v_inflight, bp); 1305 if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp)) 1306 g_raid_finish_with_locked_ranges(vol, bp); 1307 getmicrouptime(&vol->v_last_done); 1308 g_io_deliver(bp, error); 1309 } 1310 1311 int 1312 g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len, 1313 struct bio *ignore, void *argp) 1314 { 1315 struct g_raid_softc *sc; 1316 struct g_raid_lock *lp; 1317 struct bio *bp; 1318 1319 sc = vol->v_softc; 1320 lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO); 1321 LIST_INSERT_HEAD(&vol->v_locks, lp, l_next); 1322 lp->l_offset = off; 1323 lp->l_length = len; 1324 lp->l_callback_arg = argp; 1325 1326 lp->l_pending = 0; 1327 TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) { 1328 if (bp != ignore && g_raid_bio_overlaps(bp, off, len)) 1329 lp->l_pending++; 1330 } 1331 1332 /* 1333 * If there are any writes that are pending, we return EBUSY. All 1334 * callers will have to wait until all pending writes clear. 1335 */ 1336 if (lp->l_pending > 0) { 1337 vol->v_pending_lock = 1; 1338 G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend", 1339 (intmax_t)off, (intmax_t)(off+len), lp->l_pending); 1340 return (EBUSY); 1341 } 1342 G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd", 1343 (intmax_t)off, (intmax_t)(off+len)); 1344 G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 1345 return (0); 1346 } 1347 1348 int 1349 g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len) 1350 { 1351 struct g_raid_lock *lp; 1352 struct g_raid_softc *sc; 1353 struct bio *bp; 1354 1355 sc = vol->v_softc; 1356 LIST_FOREACH(lp, &vol->v_locks, l_next) { 1357 if (lp->l_offset == off && lp->l_length == len) { 1358 LIST_REMOVE(lp, l_next); 1359 /* XXX 1360 * Right now we just put them all back on the queue 1361 * and hope for the best. We hope this because any 1362 * locked ranges will go right back on this list 1363 * when the worker thread runs. 1364 * XXX 1365 */ 1366 G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd", 1367 (intmax_t)lp->l_offset, 1368 (intmax_t)(lp->l_offset+lp->l_length)); 1369 mtx_lock(&sc->sc_queue_mtx); 1370 while ((bp = bioq_takefirst(&vol->v_locked)) != NULL) 1371 bioq_insert_tail(&sc->sc_queue, bp); 1372 mtx_unlock(&sc->sc_queue_mtx); 1373 free(lp, M_RAID); 1374 return (0); 1375 } 1376 } 1377 return (EINVAL); 1378 } 1379 1380 void 1381 g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp) 1382 { 1383 struct g_consumer *cp; 1384 struct g_raid_disk *disk, *tdisk; 1385 1386 bp->bio_caller1 = sd; 1387 1388 /* 1389 * Make sure that the disk is present. Generally it is a task of 1390 * transformation layers to not send requests to absent disks, but 1391 * it is better to be safe and report situation then sorry. 1392 */ 1393 if (sd->sd_disk == NULL) { 1394 G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!"); 1395 nodisk: 1396 bp->bio_from = NULL; 1397 bp->bio_to = NULL; 1398 bp->bio_error = ENXIO; 1399 g_raid_disk_done(bp); 1400 return; 1401 } 1402 disk = sd->sd_disk; 1403 if (disk->d_state != G_RAID_DISK_S_ACTIVE && 1404 disk->d_state != G_RAID_DISK_S_FAILED) { 1405 G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a " 1406 "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 1407 goto nodisk; 1408 } 1409 1410 cp = disk->d_consumer; 1411 bp->bio_from = cp; 1412 bp->bio_to = cp->provider; 1413 cp->index++; 1414 1415 /* Update average disks load. */ 1416 TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) { 1417 if (tdisk->d_consumer == NULL) 1418 tdisk->d_load = 0; 1419 else 1420 tdisk->d_load = (tdisk->d_consumer->index * 1421 G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8; 1422 } 1423 1424 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1425 if (dumping) { 1426 G_RAID_LOGREQ(3, bp, "Sending dumping request."); 1427 if (bp->bio_cmd == BIO_WRITE) { 1428 bp->bio_error = g_raid_subdisk_kerneldump(sd, 1429 bp->bio_data, bp->bio_offset, bp->bio_length); 1430 } else 1431 bp->bio_error = EOPNOTSUPP; 1432 g_raid_disk_done(bp); 1433 } else { 1434 bp->bio_done = g_raid_disk_done; 1435 bp->bio_offset += sd->sd_offset; 1436 G_RAID_LOGREQ(3, bp, "Sending request."); 1437 g_io_request(bp, cp); 1438 } 1439 } 1440 1441 int 1442 g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd, void *virtual, 1443 off_t offset, size_t length) 1444 { 1445 1446 if (sd->sd_disk == NULL) 1447 return (ENXIO); 1448 if (sd->sd_disk->d_kd.di.dumper == NULL) 1449 return (EOPNOTSUPP); 1450 return (dump_write(&sd->sd_disk->d_kd.di, virtual, 1451 sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset, length)); 1452 } 1453 1454 static void 1455 g_raid_disk_done(struct bio *bp) 1456 { 1457 struct g_raid_softc *sc; 1458 struct g_raid_subdisk *sd; 1459 1460 sd = bp->bio_caller1; 1461 sc = sd->sd_softc; 1462 mtx_lock(&sc->sc_queue_mtx); 1463 bioq_insert_tail(&sc->sc_queue, bp); 1464 mtx_unlock(&sc->sc_queue_mtx); 1465 if (!dumping) 1466 wakeup(sc); 1467 } 1468 1469 static void 1470 g_raid_disk_done_request(struct bio *bp) 1471 { 1472 struct g_raid_softc *sc; 1473 struct g_raid_disk *disk; 1474 struct g_raid_subdisk *sd; 1475 struct g_raid_volume *vol; 1476 1477 g_topology_assert_not(); 1478 1479 G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error); 1480 sd = bp->bio_caller1; 1481 sc = sd->sd_softc; 1482 vol = sd->sd_volume; 1483 if (bp->bio_from != NULL) { 1484 bp->bio_from->index--; 1485 disk = bp->bio_from->private; 1486 if (disk == NULL) 1487 g_raid_kill_consumer(sc, bp->bio_from); 1488 } 1489 bp->bio_offset -= sd->sd_offset; 1490 1491 G_RAID_TR_IODONE(vol->v_tr, sd, bp); 1492 } 1493 1494 static void 1495 g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep) 1496 { 1497 1498 if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0) 1499 ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event); 1500 else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0) 1501 ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event); 1502 else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0) 1503 ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event); 1504 else 1505 ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event); 1506 if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) { 1507 KASSERT(ep->e_error == 0, 1508 ("Error cannot be handled.")); 1509 g_raid_event_free(ep); 1510 } else { 1511 ep->e_flags |= G_RAID_EVENT_DONE; 1512 G_RAID_DEBUG1(4, sc, "Waking up %p.", ep); 1513 mtx_lock(&sc->sc_queue_mtx); 1514 wakeup(ep); 1515 mtx_unlock(&sc->sc_queue_mtx); 1516 } 1517 } 1518 1519 /* 1520 * Worker thread. 1521 */ 1522 static void 1523 g_raid_worker(void *arg) 1524 { 1525 struct g_raid_softc *sc; 1526 struct g_raid_event *ep; 1527 struct g_raid_volume *vol; 1528 struct bio *bp; 1529 struct timeval now, t; 1530 int timeout, rv; 1531 1532 sc = arg; 1533 thread_lock(curthread); 1534 sched_prio(curthread, PRIBIO); 1535 thread_unlock(curthread); 1536 1537 sx_xlock(&sc->sc_lock); 1538 for (;;) { 1539 mtx_lock(&sc->sc_queue_mtx); 1540 /* 1541 * First take a look at events. 1542 * This is important to handle events before any I/O requests. 1543 */ 1544 bp = NULL; 1545 vol = NULL; 1546 rv = 0; 1547 ep = TAILQ_FIRST(&sc->sc_events); 1548 if (ep != NULL) 1549 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 1550 else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) 1551 ; 1552 else { 1553 getmicrouptime(&now); 1554 t = now; 1555 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1556 if (bioq_first(&vol->v_inflight) == NULL && 1557 vol->v_tr && 1558 timevalcmp(&vol->v_last_done, &t, < )) 1559 t = vol->v_last_done; 1560 } 1561 timevalsub(&t, &now); 1562 timeout = g_raid_idle_threshold + 1563 t.tv_sec * 1000000 + t.tv_usec; 1564 if (timeout > 0) { 1565 /* 1566 * Two steps to avoid overflows at HZ=1000 1567 * and idle timeouts > 2.1s. Some rounding 1568 * errors can occur, but they are < 1tick, 1569 * which is deemed to be close enough for 1570 * this purpose. 1571 */ 1572 int micpertic = 1000000 / hz; 1573 timeout = (timeout + micpertic - 1) / micpertic; 1574 sx_xunlock(&sc->sc_lock); 1575 MSLEEP(rv, sc, &sc->sc_queue_mtx, 1576 PRIBIO | PDROP, "-", timeout); 1577 sx_xlock(&sc->sc_lock); 1578 goto process; 1579 } else 1580 rv = EWOULDBLOCK; 1581 } 1582 mtx_unlock(&sc->sc_queue_mtx); 1583 process: 1584 if (ep != NULL) { 1585 g_raid_handle_event(sc, ep); 1586 } else if (bp != NULL) { 1587 if (bp->bio_to != NULL && 1588 bp->bio_to->geom == sc->sc_geom) 1589 g_raid_start_request(bp); 1590 else 1591 g_raid_disk_done_request(bp); 1592 } else if (rv == EWOULDBLOCK) { 1593 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1594 g_raid_clean(vol, -1); 1595 if (bioq_first(&vol->v_inflight) == NULL && 1596 vol->v_tr) { 1597 t.tv_sec = g_raid_idle_threshold / 1000000; 1598 t.tv_usec = g_raid_idle_threshold % 1000000; 1599 timevaladd(&t, &vol->v_last_done); 1600 getmicrouptime(&now); 1601 if (timevalcmp(&t, &now, <= )) { 1602 G_RAID_TR_IDLE(vol->v_tr); 1603 vol->v_last_done = now; 1604 } 1605 } 1606 } 1607 } 1608 if (sc->sc_stopping == G_RAID_DESTROY_HARD) 1609 g_raid_destroy_node(sc, 1); /* May not return. */ 1610 } 1611 } 1612 1613 static void 1614 g_raid_poll(struct g_raid_softc *sc) 1615 { 1616 struct g_raid_event *ep; 1617 struct bio *bp; 1618 1619 sx_xlock(&sc->sc_lock); 1620 mtx_lock(&sc->sc_queue_mtx); 1621 /* 1622 * First take a look at events. 1623 * This is important to handle events before any I/O requests. 1624 */ 1625 ep = TAILQ_FIRST(&sc->sc_events); 1626 if (ep != NULL) { 1627 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 1628 mtx_unlock(&sc->sc_queue_mtx); 1629 g_raid_handle_event(sc, ep); 1630 goto out; 1631 } 1632 bp = bioq_takefirst(&sc->sc_queue); 1633 if (bp != NULL) { 1634 mtx_unlock(&sc->sc_queue_mtx); 1635 if (bp->bio_from == NULL || 1636 bp->bio_from->geom != sc->sc_geom) 1637 g_raid_start_request(bp); 1638 else 1639 g_raid_disk_done_request(bp); 1640 } 1641 out: 1642 sx_xunlock(&sc->sc_lock); 1643 } 1644 1645 static void 1646 g_raid_launch_provider(struct g_raid_volume *vol) 1647 { 1648 struct g_raid_disk *disk; 1649 struct g_raid_subdisk *sd; 1650 struct g_raid_softc *sc; 1651 struct g_provider *pp; 1652 char name[G_RAID_MAX_VOLUMENAME]; 1653 off_t off; 1654 int i; 1655 1656 sc = vol->v_softc; 1657 sx_assert(&sc->sc_lock, SX_LOCKED); 1658 1659 g_topology_lock(); 1660 /* Try to name provider with volume name. */ 1661 snprintf(name, sizeof(name), "raid/%s", vol->v_name); 1662 if (g_raid_name_format == 0 || vol->v_name[0] == 0 || 1663 g_provider_by_name(name) != NULL) { 1664 /* Otherwise use sequential volume number. */ 1665 snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id); 1666 } 1667 1668 pp = g_new_providerf(sc->sc_geom, "%s", name); 1669 pp->flags |= G_PF_DIRECT_RECEIVE; 1670 if (vol->v_tr->tro_class->trc_accept_unmapped) { 1671 pp->flags |= G_PF_ACCEPT_UNMAPPED; 1672 for (i = 0; i < vol->v_disks_count; i++) { 1673 sd = &vol->v_subdisks[i]; 1674 if (sd->sd_state == G_RAID_SUBDISK_S_NONE) 1675 continue; 1676 if ((sd->sd_disk->d_consumer->provider->flags & 1677 G_PF_ACCEPT_UNMAPPED) == 0) 1678 pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 1679 } 1680 } 1681 pp->private = vol; 1682 pp->mediasize = vol->v_mediasize; 1683 pp->sectorsize = vol->v_sectorsize; 1684 pp->stripesize = 0; 1685 pp->stripeoffset = 0; 1686 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 || 1687 vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 || 1688 vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE || 1689 vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) { 1690 if ((disk = vol->v_subdisks[0].sd_disk) != NULL && 1691 disk->d_consumer != NULL && 1692 disk->d_consumer->provider != NULL) { 1693 pp->stripesize = disk->d_consumer->provider->stripesize; 1694 off = disk->d_consumer->provider->stripeoffset; 1695 pp->stripeoffset = off + vol->v_subdisks[0].sd_offset; 1696 if (off > 0) 1697 pp->stripeoffset %= off; 1698 } 1699 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) { 1700 pp->stripesize *= (vol->v_disks_count - 1); 1701 pp->stripeoffset *= (vol->v_disks_count - 1); 1702 } 1703 } else 1704 pp->stripesize = vol->v_strip_size; 1705 vol->v_provider = pp; 1706 g_error_provider(pp, 0); 1707 g_topology_unlock(); 1708 G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.", 1709 pp->name, vol->v_name); 1710 } 1711 1712 static void 1713 g_raid_destroy_provider(struct g_raid_volume *vol) 1714 { 1715 struct g_raid_softc *sc; 1716 struct g_provider *pp; 1717 struct bio *bp, *tmp; 1718 1719 g_topology_assert_not(); 1720 sc = vol->v_softc; 1721 pp = vol->v_provider; 1722 KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name)); 1723 1724 g_topology_lock(); 1725 g_error_provider(pp, ENXIO); 1726 mtx_lock(&sc->sc_queue_mtx); 1727 TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) { 1728 if (bp->bio_to != pp) 1729 continue; 1730 bioq_remove(&sc->sc_queue, bp); 1731 g_io_deliver(bp, ENXIO); 1732 } 1733 mtx_unlock(&sc->sc_queue_mtx); 1734 G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.", 1735 pp->name, vol->v_name); 1736 g_wither_provider(pp, ENXIO); 1737 g_topology_unlock(); 1738 vol->v_provider = NULL; 1739 } 1740 1741 /* 1742 * Update device state. 1743 */ 1744 static int 1745 g_raid_update_volume(struct g_raid_volume *vol, u_int event) 1746 { 1747 struct g_raid_softc *sc; 1748 1749 sc = vol->v_softc; 1750 sx_assert(&sc->sc_lock, SX_XLOCKED); 1751 1752 G_RAID_DEBUG1(2, sc, "Event %s for volume %s.", 1753 g_raid_volume_event2str(event), 1754 vol->v_name); 1755 switch (event) { 1756 case G_RAID_VOLUME_E_DOWN: 1757 if (vol->v_provider != NULL) 1758 g_raid_destroy_provider(vol); 1759 break; 1760 case G_RAID_VOLUME_E_UP: 1761 if (vol->v_provider == NULL) 1762 g_raid_launch_provider(vol); 1763 break; 1764 case G_RAID_VOLUME_E_START: 1765 if (vol->v_tr) 1766 G_RAID_TR_START(vol->v_tr); 1767 return (0); 1768 default: 1769 if (sc->sc_md) 1770 G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event); 1771 return (0); 1772 } 1773 1774 /* Manage root mount release. */ 1775 if (vol->v_starting) { 1776 vol->v_starting = 0; 1777 G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount); 1778 root_mount_rel(vol->v_rootmount); 1779 vol->v_rootmount = NULL; 1780 } 1781 if (vol->v_stopping && vol->v_provider_open == 0) 1782 g_raid_destroy_volume(vol); 1783 return (0); 1784 } 1785 1786 /* 1787 * Update subdisk state. 1788 */ 1789 static int 1790 g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event) 1791 { 1792 struct g_raid_softc *sc; 1793 struct g_raid_volume *vol; 1794 1795 sc = sd->sd_softc; 1796 vol = sd->sd_volume; 1797 sx_assert(&sc->sc_lock, SX_XLOCKED); 1798 1799 G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.", 1800 g_raid_subdisk_event2str(event), 1801 vol->v_name, sd->sd_pos, 1802 sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]"); 1803 if (vol->v_tr) 1804 G_RAID_TR_EVENT(vol->v_tr, sd, event); 1805 1806 return (0); 1807 } 1808 1809 /* 1810 * Update disk state. 1811 */ 1812 static int 1813 g_raid_update_disk(struct g_raid_disk *disk, u_int event) 1814 { 1815 struct g_raid_softc *sc; 1816 1817 sc = disk->d_softc; 1818 sx_assert(&sc->sc_lock, SX_XLOCKED); 1819 1820 G_RAID_DEBUG1(2, sc, "Event %s for disk %s.", 1821 g_raid_disk_event2str(event), 1822 g_raid_get_diskname(disk)); 1823 1824 if (sc->sc_md) 1825 G_RAID_MD_EVENT(sc->sc_md, disk, event); 1826 return (0); 1827 } 1828 1829 /* 1830 * Node event. 1831 */ 1832 static int 1833 g_raid_update_node(struct g_raid_softc *sc, u_int event) 1834 { 1835 sx_assert(&sc->sc_lock, SX_XLOCKED); 1836 1837 G_RAID_DEBUG1(2, sc, "Event %s for the array.", 1838 g_raid_node_event2str(event)); 1839 1840 if (event == G_RAID_NODE_E_WAKE) 1841 return (0); 1842 if (sc->sc_md) 1843 G_RAID_MD_EVENT(sc->sc_md, NULL, event); 1844 return (0); 1845 } 1846 1847 static int 1848 g_raid_access(struct g_provider *pp, int acr, int acw, int ace) 1849 { 1850 struct g_raid_volume *vol; 1851 struct g_raid_softc *sc; 1852 int dcw, opens, error = 0; 1853 1854 g_topology_assert(); 1855 sc = pp->geom->softc; 1856 vol = pp->private; 1857 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 1858 KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name)); 1859 1860 G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name, 1861 acr, acw, ace); 1862 dcw = pp->acw + acw; 1863 1864 g_topology_unlock(); 1865 sx_xlock(&sc->sc_lock); 1866 /* Deny new opens while dying. */ 1867 if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) { 1868 error = ENXIO; 1869 goto out; 1870 } 1871 /* Deny write opens for read-only volumes. */ 1872 if (vol->v_read_only && acw > 0) { 1873 error = EROFS; 1874 goto out; 1875 } 1876 if (dcw == 0) 1877 g_raid_clean(vol, dcw); 1878 vol->v_provider_open += acr + acw + ace; 1879 /* Handle delayed node destruction. */ 1880 if (sc->sc_stopping == G_RAID_DESTROY_DELAYED && 1881 vol->v_provider_open == 0) { 1882 /* Count open volumes. */ 1883 opens = g_raid_nopens(sc); 1884 if (opens == 0) { 1885 sc->sc_stopping = G_RAID_DESTROY_HARD; 1886 /* Wake up worker to make it selfdestruct. */ 1887 g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 1888 } 1889 } 1890 /* Handle open volume destruction. */ 1891 if (vol->v_stopping && vol->v_provider_open == 0) 1892 g_raid_destroy_volume(vol); 1893 out: 1894 sx_xunlock(&sc->sc_lock); 1895 g_topology_lock(); 1896 return (error); 1897 } 1898 1899 struct g_raid_softc * 1900 g_raid_create_node(struct g_class *mp, 1901 const char *name, struct g_raid_md_object *md) 1902 { 1903 struct g_raid_softc *sc; 1904 struct g_geom *gp; 1905 int error; 1906 1907 g_topology_assert(); 1908 G_RAID_DEBUG(1, "Creating array %s.", name); 1909 1910 gp = g_new_geom(mp, name); 1911 sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO); 1912 gp->start = g_raid_start; 1913 gp->orphan = g_raid_orphan; 1914 gp->access = g_raid_access; 1915 gp->dumpconf = g_raid_dumpconf; 1916 1917 sc->sc_md = md; 1918 sc->sc_geom = gp; 1919 sc->sc_flags = 0; 1920 TAILQ_INIT(&sc->sc_volumes); 1921 TAILQ_INIT(&sc->sc_disks); 1922 sx_init(&sc->sc_lock, "graid:lock"); 1923 mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF); 1924 TAILQ_INIT(&sc->sc_events); 1925 bioq_init(&sc->sc_queue); 1926 gp->softc = sc; 1927 error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0, 1928 "g_raid %s", name); 1929 if (error != 0) { 1930 G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name); 1931 mtx_destroy(&sc->sc_queue_mtx); 1932 sx_destroy(&sc->sc_lock); 1933 g_destroy_geom(sc->sc_geom); 1934 free(sc, M_RAID); 1935 return (NULL); 1936 } 1937 1938 G_RAID_DEBUG1(0, sc, "Array %s created.", name); 1939 return (sc); 1940 } 1941 1942 struct g_raid_volume * 1943 g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id) 1944 { 1945 struct g_raid_volume *vol, *vol1; 1946 int i; 1947 1948 G_RAID_DEBUG1(1, sc, "Creating volume %s.", name); 1949 vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO); 1950 vol->v_softc = sc; 1951 strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME); 1952 vol->v_state = G_RAID_VOLUME_S_STARTING; 1953 vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; 1954 vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN; 1955 vol->v_rotate_parity = 1; 1956 bioq_init(&vol->v_inflight); 1957 bioq_init(&vol->v_locked); 1958 LIST_INIT(&vol->v_locks); 1959 for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 1960 vol->v_subdisks[i].sd_softc = sc; 1961 vol->v_subdisks[i].sd_volume = vol; 1962 vol->v_subdisks[i].sd_pos = i; 1963 vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE; 1964 } 1965 1966 /* Find free ID for this volume. */ 1967 g_topology_lock(); 1968 vol1 = vol; 1969 if (id >= 0) { 1970 LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 1971 if (vol1->v_global_id == id) 1972 break; 1973 } 1974 } 1975 if (vol1 != NULL) { 1976 for (id = 0; ; id++) { 1977 LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 1978 if (vol1->v_global_id == id) 1979 break; 1980 } 1981 if (vol1 == NULL) 1982 break; 1983 } 1984 } 1985 vol->v_global_id = id; 1986 LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next); 1987 g_topology_unlock(); 1988 1989 /* Delay root mounting. */ 1990 vol->v_rootmount = root_mount_hold("GRAID"); 1991 G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount); 1992 vol->v_starting = 1; 1993 TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next); 1994 return (vol); 1995 } 1996 1997 struct g_raid_disk * 1998 g_raid_create_disk(struct g_raid_softc *sc) 1999 { 2000 struct g_raid_disk *disk; 2001 2002 G_RAID_DEBUG1(1, sc, "Creating disk."); 2003 disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO); 2004 disk->d_softc = sc; 2005 disk->d_state = G_RAID_DISK_S_NONE; 2006 TAILQ_INIT(&disk->d_subdisks); 2007 TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next); 2008 return (disk); 2009 } 2010 2011 int g_raid_start_volume(struct g_raid_volume *vol) 2012 { 2013 struct g_raid_tr_class *class; 2014 struct g_raid_tr_object *obj; 2015 int status; 2016 2017 G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name); 2018 LIST_FOREACH(class, &g_raid_tr_classes, trc_list) { 2019 if (!class->trc_enable) 2020 continue; 2021 G_RAID_DEBUG1(2, vol->v_softc, 2022 "Tasting volume %s for %s transformation.", 2023 vol->v_name, class->name); 2024 obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 2025 M_WAITOK); 2026 obj->tro_class = class; 2027 obj->tro_volume = vol; 2028 status = G_RAID_TR_TASTE(obj, vol); 2029 if (status != G_RAID_TR_TASTE_FAIL) 2030 break; 2031 kobj_delete((kobj_t)obj, M_RAID); 2032 } 2033 if (class == NULL) { 2034 G_RAID_DEBUG1(0, vol->v_softc, 2035 "No transformation module found for %s.", 2036 vol->v_name); 2037 vol->v_tr = NULL; 2038 g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED); 2039 g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN, 2040 G_RAID_EVENT_VOLUME); 2041 return (-1); 2042 } 2043 G_RAID_DEBUG1(2, vol->v_softc, 2044 "Transformation module %s chosen for %s.", 2045 class->name, vol->v_name); 2046 vol->v_tr = obj; 2047 return (0); 2048 } 2049 2050 int 2051 g_raid_destroy_node(struct g_raid_softc *sc, int worker) 2052 { 2053 struct g_raid_volume *vol, *tmpv; 2054 struct g_raid_disk *disk, *tmpd; 2055 int error = 0; 2056 2057 sc->sc_stopping = G_RAID_DESTROY_HARD; 2058 TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) { 2059 if (g_raid_destroy_volume(vol)) 2060 error = EBUSY; 2061 } 2062 if (error) 2063 return (error); 2064 TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) { 2065 if (g_raid_destroy_disk(disk)) 2066 error = EBUSY; 2067 } 2068 if (error) 2069 return (error); 2070 if (sc->sc_md) { 2071 G_RAID_MD_FREE(sc->sc_md); 2072 kobj_delete((kobj_t)sc->sc_md, M_RAID); 2073 sc->sc_md = NULL; 2074 } 2075 if (sc->sc_geom != NULL) { 2076 G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name); 2077 g_topology_lock(); 2078 sc->sc_geom->softc = NULL; 2079 g_wither_geom(sc->sc_geom, ENXIO); 2080 g_topology_unlock(); 2081 sc->sc_geom = NULL; 2082 } else 2083 G_RAID_DEBUG(1, "Array destroyed."); 2084 if (worker) { 2085 g_raid_event_cancel(sc, sc); 2086 mtx_destroy(&sc->sc_queue_mtx); 2087 sx_xunlock(&sc->sc_lock); 2088 sx_destroy(&sc->sc_lock); 2089 wakeup(&sc->sc_stopping); 2090 free(sc, M_RAID); 2091 curthread->td_pflags &= ~TDP_GEOM; 2092 G_RAID_DEBUG(1, "Thread exiting."); 2093 kproc_exit(0); 2094 } else { 2095 /* Wake up worker to make it selfdestruct. */ 2096 g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 2097 } 2098 return (0); 2099 } 2100 2101 int 2102 g_raid_destroy_volume(struct g_raid_volume *vol) 2103 { 2104 struct g_raid_softc *sc; 2105 struct g_raid_disk *disk; 2106 int i; 2107 2108 sc = vol->v_softc; 2109 G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name); 2110 vol->v_stopping = 1; 2111 if (vol->v_state != G_RAID_VOLUME_S_STOPPED) { 2112 if (vol->v_tr) { 2113 G_RAID_TR_STOP(vol->v_tr); 2114 return (EBUSY); 2115 } else 2116 vol->v_state = G_RAID_VOLUME_S_STOPPED; 2117 } 2118 if (g_raid_event_check(sc, vol) != 0) 2119 return (EBUSY); 2120 if (vol->v_provider != NULL) 2121 return (EBUSY); 2122 if (vol->v_provider_open != 0) 2123 return (EBUSY); 2124 if (vol->v_tr) { 2125 G_RAID_TR_FREE(vol->v_tr); 2126 kobj_delete((kobj_t)vol->v_tr, M_RAID); 2127 vol->v_tr = NULL; 2128 } 2129 if (vol->v_rootmount) 2130 root_mount_rel(vol->v_rootmount); 2131 g_topology_lock(); 2132 LIST_REMOVE(vol, v_global_next); 2133 g_topology_unlock(); 2134 TAILQ_REMOVE(&sc->sc_volumes, vol, v_next); 2135 for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 2136 g_raid_event_cancel(sc, &vol->v_subdisks[i]); 2137 disk = vol->v_subdisks[i].sd_disk; 2138 if (disk == NULL) 2139 continue; 2140 TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next); 2141 } 2142 G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name); 2143 if (sc->sc_md) 2144 G_RAID_MD_FREE_VOLUME(sc->sc_md, vol); 2145 g_raid_event_cancel(sc, vol); 2146 free(vol, M_RAID); 2147 if (sc->sc_stopping == G_RAID_DESTROY_HARD) { 2148 /* Wake up worker to let it selfdestruct. */ 2149 g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 2150 } 2151 return (0); 2152 } 2153 2154 int 2155 g_raid_destroy_disk(struct g_raid_disk *disk) 2156 { 2157 struct g_raid_softc *sc; 2158 struct g_raid_subdisk *sd, *tmp; 2159 2160 sc = disk->d_softc; 2161 G_RAID_DEBUG1(2, sc, "Destroying disk."); 2162 if (disk->d_consumer) { 2163 g_raid_kill_consumer(sc, disk->d_consumer); 2164 disk->d_consumer = NULL; 2165 } 2166 TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) { 2167 g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE); 2168 g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, 2169 G_RAID_EVENT_SUBDISK); 2170 TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next); 2171 sd->sd_disk = NULL; 2172 } 2173 TAILQ_REMOVE(&sc->sc_disks, disk, d_next); 2174 if (sc->sc_md) 2175 G_RAID_MD_FREE_DISK(sc->sc_md, disk); 2176 g_raid_event_cancel(sc, disk); 2177 free(disk, M_RAID); 2178 return (0); 2179 } 2180 2181 int 2182 g_raid_destroy(struct g_raid_softc *sc, int how) 2183 { 2184 int error, opens; 2185 2186 g_topology_assert_not(); 2187 if (sc == NULL) 2188 return (ENXIO); 2189 sx_assert(&sc->sc_lock, SX_XLOCKED); 2190 2191 /* Count open volumes. */ 2192 opens = g_raid_nopens(sc); 2193 2194 /* React on some opened volumes. */ 2195 if (opens > 0) { 2196 switch (how) { 2197 case G_RAID_DESTROY_SOFT: 2198 G_RAID_DEBUG1(1, sc, 2199 "%d volumes are still open.", 2200 opens); 2201 sx_xunlock(&sc->sc_lock); 2202 return (EBUSY); 2203 case G_RAID_DESTROY_DELAYED: 2204 G_RAID_DEBUG1(1, sc, 2205 "Array will be destroyed on last close."); 2206 sc->sc_stopping = G_RAID_DESTROY_DELAYED; 2207 sx_xunlock(&sc->sc_lock); 2208 return (EBUSY); 2209 case G_RAID_DESTROY_HARD: 2210 G_RAID_DEBUG1(1, sc, 2211 "%d volumes are still open.", 2212 opens); 2213 } 2214 } 2215 2216 /* Mark node for destruction. */ 2217 sc->sc_stopping = G_RAID_DESTROY_HARD; 2218 /* Wake up worker to let it selfdestruct. */ 2219 g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 2220 /* Sleep until node destroyed. */ 2221 error = sx_sleep(&sc->sc_stopping, &sc->sc_lock, 2222 PRIBIO | PDROP, "r:destroy", hz * 3); 2223 return (error == EWOULDBLOCK ? EBUSY : 0); 2224 } 2225 2226 static void 2227 g_raid_taste_orphan(struct g_consumer *cp) 2228 { 2229 2230 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2231 cp->provider->name)); 2232 } 2233 2234 static struct g_geom * 2235 g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2236 { 2237 struct g_consumer *cp; 2238 struct g_geom *gp, *geom; 2239 struct g_raid_md_class *class; 2240 struct g_raid_md_object *obj; 2241 int status; 2242 2243 g_topology_assert(); 2244 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2245 if (!g_raid_enable) 2246 return (NULL); 2247 G_RAID_DEBUG(2, "Tasting provider %s.", pp->name); 2248 2249 geom = NULL; 2250 status = G_RAID_MD_TASTE_FAIL; 2251 gp = g_new_geom(mp, "raid:taste"); 2252 /* 2253 * This orphan function should be never called. 2254 */ 2255 gp->orphan = g_raid_taste_orphan; 2256 cp = g_new_consumer(gp); 2257 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 2258 if (g_attach(cp, pp) != 0) 2259 goto ofail2; 2260 if (g_access(cp, 1, 0, 0) != 0) 2261 goto ofail; 2262 2263 LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 2264 if (!class->mdc_enable) 2265 continue; 2266 G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.", 2267 pp->name, class->name); 2268 obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 2269 M_WAITOK); 2270 obj->mdo_class = class; 2271 status = G_RAID_MD_TASTE(obj, mp, cp, &geom); 2272 if (status != G_RAID_MD_TASTE_NEW) 2273 kobj_delete((kobj_t)obj, M_RAID); 2274 if (status != G_RAID_MD_TASTE_FAIL) 2275 break; 2276 } 2277 2278 if (status == G_RAID_MD_TASTE_FAIL) 2279 (void)g_access(cp, -1, 0, 0); 2280 ofail: 2281 g_detach(cp); 2282 ofail2: 2283 g_destroy_consumer(cp); 2284 g_destroy_geom(gp); 2285 G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name); 2286 return (geom); 2287 } 2288 2289 int 2290 g_raid_create_node_format(const char *format, struct gctl_req *req, 2291 struct g_geom **gp) 2292 { 2293 struct g_raid_md_class *class; 2294 struct g_raid_md_object *obj; 2295 int status; 2296 2297 G_RAID_DEBUG(2, "Creating array for %s metadata.", format); 2298 LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 2299 if (strcasecmp(class->name, format) == 0) 2300 break; 2301 } 2302 if (class == NULL) { 2303 G_RAID_DEBUG(1, "No support for %s metadata.", format); 2304 return (G_RAID_MD_TASTE_FAIL); 2305 } 2306 obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 2307 M_WAITOK); 2308 obj->mdo_class = class; 2309 status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp); 2310 if (status != G_RAID_MD_TASTE_NEW) 2311 kobj_delete((kobj_t)obj, M_RAID); 2312 return (status); 2313 } 2314 2315 static int 2316 g_raid_destroy_geom(struct gctl_req *req __unused, 2317 struct g_class *mp __unused, struct g_geom *gp) 2318 { 2319 struct g_raid_softc *sc; 2320 int error; 2321 2322 g_topology_unlock(); 2323 sc = gp->softc; 2324 sx_xlock(&sc->sc_lock); 2325 g_cancel_event(sc); 2326 error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT); 2327 g_topology_lock(); 2328 return (error); 2329 } 2330 2331 void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol, 2332 struct g_raid_subdisk *sd, struct g_raid_disk *disk) 2333 { 2334 2335 if (sc->sc_stopping == G_RAID_DESTROY_HARD) 2336 return; 2337 if (sc->sc_md) 2338 G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk); 2339 } 2340 2341 void g_raid_fail_disk(struct g_raid_softc *sc, 2342 struct g_raid_subdisk *sd, struct g_raid_disk *disk) 2343 { 2344 2345 if (disk == NULL) 2346 disk = sd->sd_disk; 2347 if (disk == NULL) { 2348 G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!"); 2349 return; 2350 } 2351 if (disk->d_state != G_RAID_DISK_S_ACTIVE) { 2352 G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a " 2353 "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 2354 return; 2355 } 2356 if (sc->sc_md) 2357 G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk); 2358 } 2359 2360 static void 2361 g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2362 struct g_consumer *cp, struct g_provider *pp) 2363 { 2364 struct g_raid_softc *sc; 2365 struct g_raid_volume *vol; 2366 struct g_raid_subdisk *sd; 2367 struct g_raid_disk *disk; 2368 int i, s; 2369 2370 g_topology_assert(); 2371 2372 sc = gp->softc; 2373 if (sc == NULL) 2374 return; 2375 if (pp != NULL) { 2376 vol = pp->private; 2377 g_topology_unlock(); 2378 sx_xlock(&sc->sc_lock); 2379 sbuf_printf(sb, "%s<descr>%s %s volume</descr>\n", indent, 2380 sc->sc_md->mdo_class->name, 2381 g_raid_volume_level2str(vol->v_raid_level, 2382 vol->v_raid_level_qualifier)); 2383 sbuf_printf(sb, "%s<Label>%s</Label>\n", indent, 2384 vol->v_name); 2385 sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent, 2386 g_raid_volume_level2str(vol->v_raid_level, 2387 vol->v_raid_level_qualifier)); 2388 sbuf_printf(sb, 2389 "%s<Transformation>%s</Transformation>\n", indent, 2390 vol->v_tr ? vol->v_tr->tro_class->name : "NONE"); 2391 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2392 vol->v_disks_count); 2393 sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent, 2394 vol->v_strip_size); 2395 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2396 g_raid_volume_state2str(vol->v_state)); 2397 sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent, 2398 vol->v_dirty ? "Yes" : "No"); 2399 sbuf_printf(sb, "%s<Subdisks>", indent); 2400 for (i = 0; i < vol->v_disks_count; i++) { 2401 sd = &vol->v_subdisks[i]; 2402 if (sd->sd_disk != NULL && 2403 sd->sd_disk->d_consumer != NULL) { 2404 sbuf_printf(sb, "%s ", 2405 g_raid_get_diskname(sd->sd_disk)); 2406 } else { 2407 sbuf_cat(sb, "NONE "); 2408 } 2409 sbuf_printf(sb, "(%s", 2410 g_raid_subdisk_state2str(sd->sd_state)); 2411 if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2412 sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 2413 sbuf_printf(sb, " %d%%", 2414 (int)(sd->sd_rebuild_pos * 100 / 2415 sd->sd_size)); 2416 } 2417 sbuf_cat(sb, ")"); 2418 if (i + 1 < vol->v_disks_count) 2419 sbuf_cat(sb, ", "); 2420 } 2421 sbuf_cat(sb, "</Subdisks>\n"); 2422 sx_xunlock(&sc->sc_lock); 2423 g_topology_lock(); 2424 } else if (cp != NULL) { 2425 disk = cp->private; 2426 if (disk == NULL) 2427 return; 2428 g_topology_unlock(); 2429 sx_xlock(&sc->sc_lock); 2430 sbuf_printf(sb, "%s<State>%s", indent, 2431 g_raid_disk_state2str(disk->d_state)); 2432 if (!TAILQ_EMPTY(&disk->d_subdisks)) { 2433 sbuf_cat(sb, " ("); 2434 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 2435 sbuf_printf(sb, "%s", 2436 g_raid_subdisk_state2str(sd->sd_state)); 2437 if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2438 sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 2439 sbuf_printf(sb, " %d%%", 2440 (int)(sd->sd_rebuild_pos * 100 / 2441 sd->sd_size)); 2442 } 2443 if (TAILQ_NEXT(sd, sd_next)) 2444 sbuf_cat(sb, ", "); 2445 } 2446 sbuf_cat(sb, ")"); 2447 } 2448 sbuf_cat(sb, "</State>\n"); 2449 sbuf_printf(sb, "%s<Subdisks>", indent); 2450 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 2451 sbuf_printf(sb, "r%d(%s):%d@%ju", 2452 sd->sd_volume->v_global_id, 2453 sd->sd_volume->v_name, 2454 sd->sd_pos, (uintmax_t)sd->sd_offset); 2455 if (TAILQ_NEXT(sd, sd_next)) 2456 sbuf_cat(sb, ", "); 2457 } 2458 sbuf_cat(sb, "</Subdisks>\n"); 2459 sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent, 2460 disk->d_read_errs); 2461 sx_xunlock(&sc->sc_lock); 2462 g_topology_lock(); 2463 } else { 2464 g_topology_unlock(); 2465 sx_xlock(&sc->sc_lock); 2466 if (sc->sc_md) { 2467 sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent, 2468 sc->sc_md->mdo_class->name); 2469 } 2470 if (!TAILQ_EMPTY(&sc->sc_volumes)) { 2471 s = 0xff; 2472 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 2473 if (vol->v_state < s) 2474 s = vol->v_state; 2475 } 2476 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2477 g_raid_volume_state2str(s)); 2478 } 2479 sx_xunlock(&sc->sc_lock); 2480 g_topology_lock(); 2481 } 2482 } 2483 2484 static void 2485 g_raid_shutdown_post_sync(void *arg, int howto) 2486 { 2487 struct g_class *mp; 2488 struct g_geom *gp, *gp2; 2489 struct g_raid_softc *sc; 2490 struct g_raid_volume *vol; 2491 2492 if ((howto & RB_NOSYNC) != 0) 2493 return; 2494 2495 mp = arg; 2496 g_topology_lock(); 2497 g_raid_shutdown = 1; 2498 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2499 if ((sc = gp->softc) == NULL) 2500 continue; 2501 g_topology_unlock(); 2502 sx_xlock(&sc->sc_lock); 2503 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) 2504 g_raid_clean(vol, -1); 2505 g_cancel_event(sc); 2506 g_raid_destroy(sc, G_RAID_DESTROY_DELAYED); 2507 g_topology_lock(); 2508 } 2509 g_topology_unlock(); 2510 } 2511 2512 static void 2513 g_raid_init(struct g_class *mp) 2514 { 2515 2516 g_raid_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 2517 g_raid_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 2518 if (g_raid_post_sync == NULL) 2519 G_RAID_DEBUG(0, "Warning! Cannot register shutdown event."); 2520 g_raid_started = 1; 2521 } 2522 2523 static void 2524 g_raid_fini(struct g_class *mp) 2525 { 2526 2527 if (g_raid_post_sync != NULL) 2528 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid_post_sync); 2529 g_raid_started = 0; 2530 } 2531 2532 int 2533 g_raid_md_modevent(module_t mod, int type, void *arg) 2534 { 2535 struct g_raid_md_class *class, *c, *nc; 2536 int error; 2537 2538 error = 0; 2539 class = arg; 2540 switch (type) { 2541 case MOD_LOAD: 2542 c = LIST_FIRST(&g_raid_md_classes); 2543 if (c == NULL || c->mdc_priority > class->mdc_priority) 2544 LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list); 2545 else { 2546 while ((nc = LIST_NEXT(c, mdc_list)) != NULL && 2547 nc->mdc_priority < class->mdc_priority) 2548 c = nc; 2549 LIST_INSERT_AFTER(c, class, mdc_list); 2550 } 2551 if (g_raid_started) 2552 g_retaste(&g_raid_class); 2553 break; 2554 case MOD_UNLOAD: 2555 LIST_REMOVE(class, mdc_list); 2556 break; 2557 default: 2558 error = EOPNOTSUPP; 2559 break; 2560 } 2561 2562 return (error); 2563 } 2564 2565 int 2566 g_raid_tr_modevent(module_t mod, int type, void *arg) 2567 { 2568 struct g_raid_tr_class *class, *c, *nc; 2569 int error; 2570 2571 error = 0; 2572 class = arg; 2573 switch (type) { 2574 case MOD_LOAD: 2575 c = LIST_FIRST(&g_raid_tr_classes); 2576 if (c == NULL || c->trc_priority > class->trc_priority) 2577 LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list); 2578 else { 2579 while ((nc = LIST_NEXT(c, trc_list)) != NULL && 2580 nc->trc_priority < class->trc_priority) 2581 c = nc; 2582 LIST_INSERT_AFTER(c, class, trc_list); 2583 } 2584 break; 2585 case MOD_UNLOAD: 2586 LIST_REMOVE(class, trc_list); 2587 break; 2588 default: 2589 error = EOPNOTSUPP; 2590 break; 2591 } 2592 2593 return (error); 2594 } 2595 2596 /* 2597 * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid) 2598 * to reduce module priority, allowing submodules to register them first. 2599 */ 2600 static moduledata_t g_raid_mod = { 2601 "g_raid", 2602 g_modevent, 2603 &g_raid_class 2604 }; 2605 DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_FOURTH); 2606 MODULE_VERSION(geom_raid, 0); 2607