1 /*- 2 * Copyright (c) 2010 Alexander Motin <mav@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 27 #include <sys/cdefs.h> 28 __FBSDID("$FreeBSD$"); 29 30 #include <sys/param.h> 31 #include <sys/systm.h> 32 #include <sys/kernel.h> 33 #include <sys/module.h> 34 #include <sys/limits.h> 35 #include <sys/lock.h> 36 #include <sys/mutex.h> 37 #include <sys/bio.h> 38 #include <sys/sbuf.h> 39 #include <sys/sysctl.h> 40 #include <sys/malloc.h> 41 #include <sys/eventhandler.h> 42 #include <vm/uma.h> 43 #include <geom/geom.h> 44 #include <sys/proc.h> 45 #include <sys/kthread.h> 46 #include <sys/sched.h> 47 #include <geom/raid/g_raid.h> 48 #include "g_raid_md_if.h" 49 #include "g_raid_tr_if.h" 50 51 static MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data"); 52 53 SYSCTL_DECL(_kern_geom); 54 SYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW, 0, "GEOM_RAID stuff"); 55 u_int g_raid_aggressive_spare = 0; 56 TUNABLE_INT("kern.geom.raid.aggressive_spare", &g_raid_aggressive_spare); 57 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RW, 58 &g_raid_aggressive_spare, 0, "Use disks without metadata as spare"); 59 u_int g_raid_debug = 0; 60 TUNABLE_INT("kern.geom.raid.debug", &g_raid_debug); 61 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RW, &g_raid_debug, 0, 62 "Debug level"); 63 int g_raid_read_err_thresh = 10; 64 TUNABLE_INT("kern.geom.raid.read_err_thresh", &g_raid_read_err_thresh); 65 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RW, 66 &g_raid_read_err_thresh, 0, 67 "Number of read errors equated to disk failure"); 68 u_int g_raid_start_timeout = 30; 69 TUNABLE_INT("kern.geom.raid.start_timeout", &g_raid_start_timeout); 70 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RW, 71 &g_raid_start_timeout, 0, 72 "Time to wait for all array components"); 73 static u_int g_raid_clean_time = 5; 74 TUNABLE_INT("kern.geom.raid.clean_time", &g_raid_clean_time); 75 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RW, 76 &g_raid_clean_time, 0, "Mark volume as clean when idling"); 77 static u_int g_raid_disconnect_on_failure = 1; 78 TUNABLE_INT("kern.geom.raid.disconnect_on_failure", 79 &g_raid_disconnect_on_failure); 80 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RW, 81 &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 82 static u_int g_raid_name_format = 0; 83 TUNABLE_INT("kern.geom.raid.name_format", &g_raid_name_format); 84 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RW, 85 &g_raid_name_format, 0, "Providers name format."); 86 static u_int g_raid_idle_threshold = 1000000; 87 TUNABLE_INT("kern.geom.raid.idle_threshold", &g_raid_idle_threshold); 88 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RW, 89 &g_raid_idle_threshold, 1000000, 90 "Time in microseconds to consider a volume idle."); 91 92 #define MSLEEP(rv, ident, mtx, priority, wmesg, timeout) do { \ 93 G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 94 rv = msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 95 G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 96 } while (0) 97 98 LIST_HEAD(, g_raid_md_class) g_raid_md_classes = 99 LIST_HEAD_INITIALIZER(g_raid_md_classes); 100 101 LIST_HEAD(, g_raid_tr_class) g_raid_tr_classes = 102 LIST_HEAD_INITIALIZER(g_raid_tr_classes); 103 104 LIST_HEAD(, g_raid_volume) g_raid_volumes = 105 LIST_HEAD_INITIALIZER(g_raid_volumes); 106 107 static eventhandler_tag g_raid_pre_sync = NULL; 108 static int g_raid_started = 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_FAILED: 162 return ("FAILED"); 163 case G_RAID_DISK_S_STALE_FAILED: 164 return ("STALE_FAILED"); 165 case G_RAID_DISK_S_SPARE: 166 return ("SPARE"); 167 case G_RAID_DISK_S_STALE: 168 return ("STALE"); 169 case G_RAID_DISK_S_ACTIVE: 170 return ("ACTIVE"); 171 default: 172 return ("INVALID"); 173 } 174 } 175 176 static const char * 177 g_raid_disk_event2str(int event) 178 { 179 180 switch (event) { 181 case G_RAID_DISK_E_DISCONNECTED: 182 return ("DISCONNECTED"); 183 default: 184 return ("INVALID"); 185 } 186 } 187 188 const char * 189 g_raid_subdisk_state2str(int state) 190 { 191 192 switch (state) { 193 case G_RAID_SUBDISK_S_NONE: 194 return ("NONE"); 195 case G_RAID_SUBDISK_S_FAILED: 196 return ("FAILED"); 197 case G_RAID_SUBDISK_S_NEW: 198 return ("NEW"); 199 case G_RAID_SUBDISK_S_REBUILD: 200 return ("REBUILD"); 201 case G_RAID_SUBDISK_S_UNINITIALIZED: 202 return ("UNINITIALIZED"); 203 case G_RAID_SUBDISK_S_STALE: 204 return ("STALE"); 205 case G_RAID_SUBDISK_S_RESYNC: 206 return ("RESYNC"); 207 case G_RAID_SUBDISK_S_ACTIVE: 208 return ("ACTIVE"); 209 default: 210 return ("INVALID"); 211 } 212 } 213 214 static const char * 215 g_raid_subdisk_event2str(int event) 216 { 217 218 switch (event) { 219 case G_RAID_SUBDISK_E_NEW: 220 return ("NEW"); 221 case G_RAID_SUBDISK_E_DISCONNECTED: 222 return ("DISCONNECTED"); 223 default: 224 return ("INVALID"); 225 } 226 } 227 228 const char * 229 g_raid_volume_state2str(int state) 230 { 231 232 switch (state) { 233 case G_RAID_VOLUME_S_STARTING: 234 return ("STARTING"); 235 case G_RAID_VOLUME_S_BROKEN: 236 return ("BROKEN"); 237 case G_RAID_VOLUME_S_DEGRADED: 238 return ("DEGRADED"); 239 case G_RAID_VOLUME_S_SUBOPTIMAL: 240 return ("SUBOPTIMAL"); 241 case G_RAID_VOLUME_S_OPTIMAL: 242 return ("OPTIMAL"); 243 case G_RAID_VOLUME_S_UNSUPPORTED: 244 return ("UNSUPPORTED"); 245 case G_RAID_VOLUME_S_STOPPED: 246 return ("STOPPED"); 247 default: 248 return ("INVALID"); 249 } 250 } 251 252 static const char * 253 g_raid_volume_event2str(int event) 254 { 255 256 switch (event) { 257 case G_RAID_VOLUME_E_UP: 258 return ("UP"); 259 case G_RAID_VOLUME_E_DOWN: 260 return ("DOWN"); 261 case G_RAID_VOLUME_E_START: 262 return ("START"); 263 case G_RAID_VOLUME_E_STARTMD: 264 return ("STARTMD"); 265 default: 266 return ("INVALID"); 267 } 268 } 269 270 const char * 271 g_raid_volume_level2str(int level, int qual) 272 { 273 274 switch (level) { 275 case G_RAID_VOLUME_RL_RAID0: 276 return ("RAID0"); 277 case G_RAID_VOLUME_RL_RAID1: 278 return ("RAID1"); 279 case G_RAID_VOLUME_RL_RAID3: 280 return ("RAID3"); 281 case G_RAID_VOLUME_RL_RAID4: 282 return ("RAID4"); 283 case G_RAID_VOLUME_RL_RAID5: 284 if (qual == G_RAID_VOLUME_RLQ_R5RA) 285 return ("RAID5RA"); 286 if (qual == G_RAID_VOLUME_RLQ_R5RS) 287 return ("RAID5RS"); 288 if (qual == G_RAID_VOLUME_RLQ_R5LA) 289 return ("RAID5LA"); 290 if (qual == G_RAID_VOLUME_RLQ_R5LS) 291 return ("RAID5LS"); 292 return ("RAID5"); 293 case G_RAID_VOLUME_RL_RAID6: 294 return ("RAID6"); 295 case G_RAID_VOLUME_RL_RAID1E: 296 return ("RAID1E"); 297 case G_RAID_VOLUME_RL_SINGLE: 298 return ("SINGLE"); 299 case G_RAID_VOLUME_RL_CONCAT: 300 return ("CONCAT"); 301 case G_RAID_VOLUME_RL_RAID5E: 302 return ("RAID5E"); 303 case G_RAID_VOLUME_RL_RAID5EE: 304 return ("RAID5EE"); 305 default: 306 return ("UNKNOWN"); 307 } 308 } 309 310 int 311 g_raid_volume_str2level(const char *str, int *level, int *qual) 312 { 313 314 *level = G_RAID_VOLUME_RL_UNKNOWN; 315 *qual = G_RAID_VOLUME_RLQ_NONE; 316 if (strcasecmp(str, "RAID0") == 0) 317 *level = G_RAID_VOLUME_RL_RAID0; 318 else if (strcasecmp(str, "RAID1") == 0) 319 *level = G_RAID_VOLUME_RL_RAID1; 320 else if (strcasecmp(str, "RAID3") == 0) 321 *level = G_RAID_VOLUME_RL_RAID3; 322 else if (strcasecmp(str, "RAID4") == 0) 323 *level = G_RAID_VOLUME_RL_RAID4; 324 else if (strcasecmp(str, "RAID5RA") == 0) { 325 *level = G_RAID_VOLUME_RL_RAID5; 326 *qual = G_RAID_VOLUME_RLQ_R5RA; 327 } else if (strcasecmp(str, "RAID5RS") == 0) { 328 *level = G_RAID_VOLUME_RL_RAID5; 329 *qual = G_RAID_VOLUME_RLQ_R5RS; 330 } else if (strcasecmp(str, "RAID5") == 0 || 331 strcasecmp(str, "RAID5LA") == 0) { 332 *level = G_RAID_VOLUME_RL_RAID5; 333 *qual = G_RAID_VOLUME_RLQ_R5LA; 334 } else if (strcasecmp(str, "RAID5LS") == 0) { 335 *level = G_RAID_VOLUME_RL_RAID5; 336 *qual = G_RAID_VOLUME_RLQ_R5LS; 337 } else if (strcasecmp(str, "RAID6") == 0) 338 *level = G_RAID_VOLUME_RL_RAID6; 339 else if (strcasecmp(str, "RAID10") == 0 || 340 strcasecmp(str, "RAID1E") == 0) 341 *level = G_RAID_VOLUME_RL_RAID1E; 342 else if (strcasecmp(str, "SINGLE") == 0) 343 *level = G_RAID_VOLUME_RL_SINGLE; 344 else if (strcasecmp(str, "CONCAT") == 0) 345 *level = G_RAID_VOLUME_RL_CONCAT; 346 else if (strcasecmp(str, "RAID5E") == 0) 347 *level = G_RAID_VOLUME_RL_RAID5E; 348 else if (strcasecmp(str, "RAID5EE") == 0) 349 *level = G_RAID_VOLUME_RL_RAID5EE; 350 else 351 return (-1); 352 return (0); 353 } 354 355 const char * 356 g_raid_get_diskname(struct g_raid_disk *disk) 357 { 358 359 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 360 return ("[unknown]"); 361 return (disk->d_consumer->provider->name); 362 } 363 364 void 365 g_raid_report_disk_state(struct g_raid_disk *disk) 366 { 367 struct g_raid_subdisk *sd; 368 int len, state; 369 uint32_t s; 370 371 if (disk->d_consumer == NULL) 372 return; 373 if (disk->d_state == G_RAID_DISK_S_FAILED || 374 disk->d_state == G_RAID_DISK_S_STALE_FAILED) { 375 s = G_STATE_FAILED; 376 } else { 377 state = G_RAID_SUBDISK_S_ACTIVE; 378 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 379 if (sd->sd_state < state) 380 state = sd->sd_state; 381 } 382 if (state == G_RAID_SUBDISK_S_FAILED) 383 s = G_STATE_FAILED; 384 else if (state == G_RAID_SUBDISK_S_NEW || 385 state == G_RAID_SUBDISK_S_REBUILD) 386 s = G_STATE_REBUILD; 387 else if (state == G_RAID_SUBDISK_S_STALE || 388 state == G_RAID_SUBDISK_S_RESYNC) 389 s = G_STATE_RESYNC; 390 else 391 s = G_STATE_ACTIVE; 392 } 393 len = sizeof(s); 394 g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s); 395 G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.", 396 g_raid_get_diskname(disk), s); 397 } 398 399 void 400 g_raid_change_disk_state(struct g_raid_disk *disk, int state) 401 { 402 403 G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.", 404 g_raid_get_diskname(disk), 405 g_raid_disk_state2str(disk->d_state), 406 g_raid_disk_state2str(state)); 407 disk->d_state = state; 408 g_raid_report_disk_state(disk); 409 } 410 411 void 412 g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state) 413 { 414 415 G_RAID_DEBUG1(0, sd->sd_softc, 416 "Subdisk %s:%d-%s state changed from %s to %s.", 417 sd->sd_volume->v_name, sd->sd_pos, 418 sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]", 419 g_raid_subdisk_state2str(sd->sd_state), 420 g_raid_subdisk_state2str(state)); 421 sd->sd_state = state; 422 if (sd->sd_disk) 423 g_raid_report_disk_state(sd->sd_disk); 424 } 425 426 void 427 g_raid_change_volume_state(struct g_raid_volume *vol, int state) 428 { 429 430 G_RAID_DEBUG1(0, vol->v_softc, 431 "Volume %s state changed from %s to %s.", 432 vol->v_name, 433 g_raid_volume_state2str(vol->v_state), 434 g_raid_volume_state2str(state)); 435 vol->v_state = state; 436 } 437 438 /* 439 * --- Events handling functions --- 440 * Events in geom_raid are used to maintain subdisks and volumes status 441 * from one thread to simplify locking. 442 */ 443 static void 444 g_raid_event_free(struct g_raid_event *ep) 445 { 446 447 free(ep, M_RAID); 448 } 449 450 int 451 g_raid_event_send(void *arg, int event, int flags) 452 { 453 struct g_raid_softc *sc; 454 struct g_raid_event *ep; 455 int error; 456 457 if ((flags & G_RAID_EVENT_VOLUME) != 0) { 458 sc = ((struct g_raid_volume *)arg)->v_softc; 459 } else if ((flags & G_RAID_EVENT_DISK) != 0) { 460 sc = ((struct g_raid_disk *)arg)->d_softc; 461 } else if ((flags & G_RAID_EVENT_SUBDISK) != 0) { 462 sc = ((struct g_raid_subdisk *)arg)->sd_softc; 463 } else { 464 sc = arg; 465 } 466 ep = malloc(sizeof(*ep), M_RAID, 467 sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT); 468 if (ep == NULL) 469 return (ENOMEM); 470 ep->e_tgt = arg; 471 ep->e_event = event; 472 ep->e_flags = flags; 473 ep->e_error = 0; 474 G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc); 475 mtx_lock(&sc->sc_queue_mtx); 476 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 477 mtx_unlock(&sc->sc_queue_mtx); 478 wakeup(sc); 479 480 if ((flags & G_RAID_EVENT_WAIT) == 0) 481 return (0); 482 483 sx_assert(&sc->sc_lock, SX_XLOCKED); 484 G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep); 485 sx_xunlock(&sc->sc_lock); 486 while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) { 487 mtx_lock(&sc->sc_queue_mtx); 488 MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event", 489 hz * 5); 490 } 491 error = ep->e_error; 492 g_raid_event_free(ep); 493 sx_xlock(&sc->sc_lock); 494 return (error); 495 } 496 497 static void 498 g_raid_event_cancel(struct g_raid_softc *sc, void *tgt) 499 { 500 struct g_raid_event *ep, *tmpep; 501 502 sx_assert(&sc->sc_lock, SX_XLOCKED); 503 504 mtx_lock(&sc->sc_queue_mtx); 505 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 506 if (ep->e_tgt != tgt) 507 continue; 508 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 509 if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) 510 g_raid_event_free(ep); 511 else { 512 ep->e_error = ECANCELED; 513 wakeup(ep); 514 } 515 } 516 mtx_unlock(&sc->sc_queue_mtx); 517 } 518 519 static int 520 g_raid_event_check(struct g_raid_softc *sc, void *tgt) 521 { 522 struct g_raid_event *ep; 523 int res = 0; 524 525 sx_assert(&sc->sc_lock, SX_XLOCKED); 526 527 mtx_lock(&sc->sc_queue_mtx); 528 TAILQ_FOREACH(ep, &sc->sc_events, e_next) { 529 if (ep->e_tgt != tgt) 530 continue; 531 res = 1; 532 break; 533 } 534 mtx_unlock(&sc->sc_queue_mtx); 535 return (res); 536 } 537 538 /* 539 * Return the number of disks in given state. 540 * If state is equal to -1, count all connected disks. 541 */ 542 u_int 543 g_raid_ndisks(struct g_raid_softc *sc, int state) 544 { 545 struct g_raid_disk *disk; 546 u_int n; 547 548 sx_assert(&sc->sc_lock, SX_LOCKED); 549 550 n = 0; 551 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 552 if (disk->d_state == state || state == -1) 553 n++; 554 } 555 return (n); 556 } 557 558 /* 559 * Return the number of subdisks in given state. 560 * If state is equal to -1, count all connected disks. 561 */ 562 u_int 563 g_raid_nsubdisks(struct g_raid_volume *vol, int state) 564 { 565 struct g_raid_subdisk *subdisk; 566 struct g_raid_softc *sc; 567 u_int i, n ; 568 569 sc = vol->v_softc; 570 sx_assert(&sc->sc_lock, SX_LOCKED); 571 572 n = 0; 573 for (i = 0; i < vol->v_disks_count; i++) { 574 subdisk = &vol->v_subdisks[i]; 575 if ((state == -1 && 576 subdisk->sd_state != G_RAID_SUBDISK_S_NONE) || 577 subdisk->sd_state == state) 578 n++; 579 } 580 return (n); 581 } 582 583 /* 584 * Return the first subdisk in given state. 585 * If state is equal to -1, then the first connected disks. 586 */ 587 struct g_raid_subdisk * 588 g_raid_get_subdisk(struct g_raid_volume *vol, int state) 589 { 590 struct g_raid_subdisk *sd; 591 struct g_raid_softc *sc; 592 u_int i; 593 594 sc = vol->v_softc; 595 sx_assert(&sc->sc_lock, SX_LOCKED); 596 597 for (i = 0; i < vol->v_disks_count; i++) { 598 sd = &vol->v_subdisks[i]; 599 if ((state == -1 && 600 sd->sd_state != G_RAID_SUBDISK_S_NONE) || 601 sd->sd_state == state) 602 return (sd); 603 } 604 return (NULL); 605 } 606 607 struct g_consumer * 608 g_raid_open_consumer(struct g_raid_softc *sc, const char *name) 609 { 610 struct g_consumer *cp; 611 struct g_provider *pp; 612 613 g_topology_assert(); 614 615 if (strncmp(name, "/dev/", 5) == 0) 616 name += 5; 617 pp = g_provider_by_name(name); 618 if (pp == NULL) 619 return (NULL); 620 cp = g_new_consumer(sc->sc_geom); 621 if (g_attach(cp, pp) != 0) { 622 g_destroy_consumer(cp); 623 return (NULL); 624 } 625 if (g_access(cp, 1, 1, 1) != 0) { 626 g_detach(cp); 627 g_destroy_consumer(cp); 628 return (NULL); 629 } 630 return (cp); 631 } 632 633 static u_int 634 g_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp) 635 { 636 struct bio *bp; 637 u_int nreqs = 0; 638 639 mtx_lock(&sc->sc_queue_mtx); 640 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 641 if (bp->bio_from == cp) 642 nreqs++; 643 } 644 mtx_unlock(&sc->sc_queue_mtx); 645 return (nreqs); 646 } 647 648 u_int 649 g_raid_nopens(struct g_raid_softc *sc) 650 { 651 struct g_raid_volume *vol; 652 u_int opens; 653 654 opens = 0; 655 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 656 if (vol->v_provider_open != 0) 657 opens++; 658 } 659 return (opens); 660 } 661 662 static int 663 g_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp) 664 { 665 666 if (cp->index > 0) { 667 G_RAID_DEBUG1(2, sc, 668 "I/O requests for %s exist, can't destroy it now.", 669 cp->provider->name); 670 return (1); 671 } 672 if (g_raid_nrequests(sc, cp) > 0) { 673 G_RAID_DEBUG1(2, sc, 674 "I/O requests for %s in queue, can't destroy it now.", 675 cp->provider->name); 676 return (1); 677 } 678 return (0); 679 } 680 681 static void 682 g_raid_destroy_consumer(void *arg, int flags __unused) 683 { 684 struct g_consumer *cp; 685 686 g_topology_assert(); 687 688 cp = arg; 689 G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 690 g_detach(cp); 691 g_destroy_consumer(cp); 692 } 693 694 void 695 g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp) 696 { 697 struct g_provider *pp; 698 int retaste_wait; 699 700 g_topology_assert_not(); 701 702 g_topology_lock(); 703 cp->private = NULL; 704 if (g_raid_consumer_is_busy(sc, cp)) 705 goto out; 706 pp = cp->provider; 707 retaste_wait = 0; 708 if (cp->acw == 1) { 709 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 710 retaste_wait = 1; 711 } 712 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 713 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 714 if (retaste_wait) { 715 /* 716 * After retaste event was send (inside g_access()), we can send 717 * event to detach and destroy consumer. 718 * A class, which has consumer to the given provider connected 719 * will not receive retaste event for the provider. 720 * This is the way how I ignore retaste events when I close 721 * consumers opened for write: I detach and destroy consumer 722 * after retaste event is sent. 723 */ 724 g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL); 725 goto out; 726 } 727 G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name); 728 g_detach(cp); 729 g_destroy_consumer(cp); 730 out: 731 g_topology_unlock(); 732 } 733 734 static void 735 g_raid_orphan(struct g_consumer *cp) 736 { 737 struct g_raid_disk *disk; 738 739 g_topology_assert(); 740 741 disk = cp->private; 742 if (disk == NULL) 743 return; 744 g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED, 745 G_RAID_EVENT_DISK); 746 } 747 748 static int 749 g_raid_clean(struct g_raid_volume *vol, int acw) 750 { 751 struct g_raid_softc *sc; 752 int timeout; 753 754 sc = vol->v_softc; 755 g_topology_assert_not(); 756 sx_assert(&sc->sc_lock, SX_XLOCKED); 757 758 // if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 759 // return (0); 760 if (!vol->v_dirty) 761 return (0); 762 if (vol->v_writes > 0) 763 return (0); 764 if (acw > 0 || (acw == -1 && 765 vol->v_provider != NULL && vol->v_provider->acw > 0)) { 766 timeout = g_raid_clean_time - (time_uptime - vol->v_last_write); 767 if (timeout > 0) 768 return (timeout); 769 } 770 vol->v_dirty = 0; 771 G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.", 772 vol->v_name); 773 g_raid_write_metadata(sc, vol, NULL, NULL); 774 return (0); 775 } 776 777 static void 778 g_raid_dirty(struct g_raid_volume *vol) 779 { 780 struct g_raid_softc *sc; 781 782 sc = vol->v_softc; 783 g_topology_assert_not(); 784 sx_assert(&sc->sc_lock, SX_XLOCKED); 785 786 // if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 787 // return; 788 vol->v_dirty = 1; 789 G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.", 790 vol->v_name); 791 g_raid_write_metadata(sc, vol, NULL, NULL); 792 } 793 794 void 795 g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp) 796 { 797 struct g_raid_softc *sc; 798 struct g_raid_volume *vol; 799 struct g_raid_subdisk *sd; 800 struct bio_queue_head queue; 801 struct bio *cbp; 802 int i; 803 804 vol = tr->tro_volume; 805 sc = vol->v_softc; 806 807 /* 808 * Allocate all bios before sending any request, so we can return 809 * ENOMEM in nice and clean way. 810 */ 811 bioq_init(&queue); 812 for (i = 0; i < vol->v_disks_count; i++) { 813 sd = &vol->v_subdisks[i]; 814 if (sd->sd_state == G_RAID_SUBDISK_S_NONE || 815 sd->sd_state == G_RAID_SUBDISK_S_FAILED) 816 continue; 817 cbp = g_clone_bio(bp); 818 if (cbp == NULL) 819 goto failure; 820 cbp->bio_caller1 = sd; 821 bioq_insert_tail(&queue, cbp); 822 } 823 for (cbp = bioq_first(&queue); cbp != NULL; 824 cbp = bioq_first(&queue)) { 825 bioq_remove(&queue, cbp); 826 sd = cbp->bio_caller1; 827 cbp->bio_caller1 = NULL; 828 g_raid_subdisk_iostart(sd, cbp); 829 } 830 return; 831 failure: 832 for (cbp = bioq_first(&queue); cbp != NULL; 833 cbp = bioq_first(&queue)) { 834 bioq_remove(&queue, cbp); 835 g_destroy_bio(cbp); 836 } 837 if (bp->bio_error == 0) 838 bp->bio_error = ENOMEM; 839 g_raid_iodone(bp, bp->bio_error); 840 } 841 842 static void 843 g_raid_tr_kerneldump_common_done(struct bio *bp) 844 { 845 846 bp->bio_flags |= BIO_DONE; 847 } 848 849 int 850 g_raid_tr_kerneldump_common(struct g_raid_tr_object *tr, 851 void *virtual, vm_offset_t physical, off_t offset, size_t length) 852 { 853 struct g_raid_softc *sc; 854 struct g_raid_volume *vol; 855 struct bio bp; 856 857 vol = tr->tro_volume; 858 sc = vol->v_softc; 859 860 bzero(&bp, sizeof(bp)); 861 bp.bio_cmd = BIO_WRITE; 862 bp.bio_done = g_raid_tr_kerneldump_common_done; 863 bp.bio_attribute = NULL; 864 bp.bio_offset = offset; 865 bp.bio_length = length; 866 bp.bio_data = virtual; 867 bp.bio_to = vol->v_provider; 868 869 g_raid_start(&bp); 870 while (!(bp.bio_flags & BIO_DONE)) { 871 G_RAID_DEBUG1(4, sc, "Poll..."); 872 g_raid_poll(sc); 873 DELAY(10); 874 } 875 876 return (bp.bio_error != 0 ? EIO : 0); 877 } 878 879 static int 880 g_raid_dump(void *arg, 881 void *virtual, vm_offset_t physical, off_t offset, size_t length) 882 { 883 struct g_raid_volume *vol; 884 int error; 885 886 vol = (struct g_raid_volume *)arg; 887 G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.", 888 (long long unsigned)offset, (long long unsigned)length); 889 890 error = G_RAID_TR_KERNELDUMP(vol->v_tr, 891 virtual, physical, offset, length); 892 return (error); 893 } 894 895 static void 896 g_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp) 897 { 898 struct g_kerneldump *gkd; 899 struct g_provider *pp; 900 struct g_raid_volume *vol; 901 902 gkd = (struct g_kerneldump*)bp->bio_data; 903 pp = bp->bio_to; 904 vol = pp->private; 905 g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)", 906 pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length); 907 gkd->di.dumper = g_raid_dump; 908 gkd->di.priv = vol; 909 gkd->di.blocksize = vol->v_sectorsize; 910 gkd->di.maxiosize = DFLTPHYS; 911 gkd->di.mediaoffset = gkd->offset; 912 if ((gkd->offset + gkd->length) > vol->v_mediasize) 913 gkd->length = vol->v_mediasize - gkd->offset; 914 gkd->di.mediasize = gkd->length; 915 g_io_deliver(bp, 0); 916 } 917 918 static void 919 g_raid_start(struct bio *bp) 920 { 921 struct g_raid_softc *sc; 922 923 sc = bp->bio_to->geom->softc; 924 /* 925 * If sc == NULL or there are no valid disks, provider's error 926 * should be set and g_raid_start() should not be called at all. 927 */ 928 // KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING, 929 // ("Provider's error should be set (error=%d)(mirror=%s).", 930 // bp->bio_to->error, bp->bio_to->name)); 931 G_RAID_LOGREQ(3, bp, "Request received."); 932 933 switch (bp->bio_cmd) { 934 case BIO_READ: 935 case BIO_WRITE: 936 case BIO_DELETE: 937 case BIO_FLUSH: 938 break; 939 case BIO_GETATTR: 940 if (!strcmp(bp->bio_attribute, "GEOM::kerneldump")) 941 g_raid_kerneldump(sc, bp); 942 else 943 g_io_deliver(bp, EOPNOTSUPP); 944 return; 945 default: 946 g_io_deliver(bp, EOPNOTSUPP); 947 return; 948 } 949 mtx_lock(&sc->sc_queue_mtx); 950 bioq_disksort(&sc->sc_queue, bp); 951 mtx_unlock(&sc->sc_queue_mtx); 952 if (!dumping) { 953 G_RAID_DEBUG1(4, sc, "Waking up %p.", sc); 954 wakeup(sc); 955 } 956 } 957 958 static int 959 g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len) 960 { 961 /* 962 * 5 cases: 963 * (1) bp entirely below NO 964 * (2) bp entirely above NO 965 * (3) bp start below, but end in range YES 966 * (4) bp entirely within YES 967 * (5) bp starts within, ends above YES 968 * 969 * lock range 10-19 (offset 10 length 10) 970 * (1) 1-5: first if kicks it out 971 * (2) 30-35: second if kicks it out 972 * (3) 5-15: passes both ifs 973 * (4) 12-14: passes both ifs 974 * (5) 19-20: passes both 975 */ 976 off_t lend = lstart + len - 1; 977 off_t bstart = bp->bio_offset; 978 off_t bend = bp->bio_offset + bp->bio_length - 1; 979 980 if (bend < lstart) 981 return (0); 982 if (lend < bstart) 983 return (0); 984 return (1); 985 } 986 987 static int 988 g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp) 989 { 990 struct g_raid_lock *lp; 991 992 sx_assert(&vol->v_softc->sc_lock, SX_LOCKED); 993 994 LIST_FOREACH(lp, &vol->v_locks, l_next) { 995 if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length)) 996 return (1); 997 } 998 return (0); 999 } 1000 1001 static void 1002 g_raid_start_request(struct bio *bp) 1003 { 1004 struct g_raid_softc *sc; 1005 struct g_raid_volume *vol; 1006 1007 sc = bp->bio_to->geom->softc; 1008 sx_assert(&sc->sc_lock, SX_LOCKED); 1009 vol = bp->bio_to->private; 1010 1011 /* 1012 * Check to see if this item is in a locked range. If so, 1013 * queue it to our locked queue and return. We'll requeue 1014 * it when the range is unlocked. Internal I/O for the 1015 * rebuild/rescan/recovery process is excluded from this 1016 * check so we can actually do the recovery. 1017 */ 1018 if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) && 1019 g_raid_is_in_locked_range(vol, bp)) { 1020 G_RAID_LOGREQ(3, bp, "Defer request."); 1021 bioq_insert_tail(&vol->v_locked, bp); 1022 return; 1023 } 1024 1025 /* 1026 * If we're actually going to do the write/delete, then 1027 * update the idle stats for the volume. 1028 */ 1029 if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 1030 if (!vol->v_dirty) 1031 g_raid_dirty(vol); 1032 vol->v_writes++; 1033 } 1034 1035 /* 1036 * Put request onto inflight queue, so we can check if new 1037 * synchronization requests don't collide with it. Then tell 1038 * the transformation layer to start the I/O. 1039 */ 1040 bioq_insert_tail(&vol->v_inflight, bp); 1041 G_RAID_LOGREQ(4, bp, "Request started"); 1042 G_RAID_TR_IOSTART(vol->v_tr, bp); 1043 } 1044 1045 static void 1046 g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp) 1047 { 1048 off_t off, len; 1049 struct bio *nbp; 1050 struct g_raid_lock *lp; 1051 1052 vol->v_pending_lock = 0; 1053 LIST_FOREACH(lp, &vol->v_locks, l_next) { 1054 if (lp->l_pending) { 1055 off = lp->l_offset; 1056 len = lp->l_length; 1057 lp->l_pending = 0; 1058 TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) { 1059 if (g_raid_bio_overlaps(nbp, off, len)) 1060 lp->l_pending++; 1061 } 1062 if (lp->l_pending) { 1063 vol->v_pending_lock = 1; 1064 G_RAID_DEBUG1(4, vol->v_softc, 1065 "Deferred lock(%jd, %jd) has %d pending", 1066 (intmax_t)off, (intmax_t)(off + len), 1067 lp->l_pending); 1068 continue; 1069 } 1070 G_RAID_DEBUG1(4, vol->v_softc, 1071 "Deferred lock of %jd to %jd completed", 1072 (intmax_t)off, (intmax_t)(off + len)); 1073 G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 1074 } 1075 } 1076 } 1077 1078 void 1079 g_raid_iodone(struct bio *bp, int error) 1080 { 1081 struct g_raid_softc *sc; 1082 struct g_raid_volume *vol; 1083 1084 sc = bp->bio_to->geom->softc; 1085 sx_assert(&sc->sc_lock, SX_LOCKED); 1086 vol = bp->bio_to->private; 1087 G_RAID_LOGREQ(3, bp, "Request done: %d.", error); 1088 1089 /* Update stats if we done write/delete. */ 1090 if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 1091 vol->v_writes--; 1092 vol->v_last_write = time_uptime; 1093 } 1094 1095 bioq_remove(&vol->v_inflight, bp); 1096 if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp)) 1097 g_raid_finish_with_locked_ranges(vol, bp); 1098 getmicrouptime(&vol->v_last_done); 1099 g_io_deliver(bp, error); 1100 } 1101 1102 int 1103 g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len, 1104 struct bio *ignore, void *argp) 1105 { 1106 struct g_raid_softc *sc; 1107 struct g_raid_lock *lp; 1108 struct bio *bp; 1109 1110 sc = vol->v_softc; 1111 lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO); 1112 LIST_INSERT_HEAD(&vol->v_locks, lp, l_next); 1113 lp->l_offset = off; 1114 lp->l_length = len; 1115 lp->l_callback_arg = argp; 1116 1117 lp->l_pending = 0; 1118 TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) { 1119 if (bp != ignore && g_raid_bio_overlaps(bp, off, len)) 1120 lp->l_pending++; 1121 } 1122 1123 /* 1124 * If there are any writes that are pending, we return EBUSY. All 1125 * callers will have to wait until all pending writes clear. 1126 */ 1127 if (lp->l_pending > 0) { 1128 vol->v_pending_lock = 1; 1129 G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend", 1130 (intmax_t)off, (intmax_t)(off+len), lp->l_pending); 1131 return (EBUSY); 1132 } 1133 G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd", 1134 (intmax_t)off, (intmax_t)(off+len)); 1135 G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 1136 return (0); 1137 } 1138 1139 int 1140 g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len) 1141 { 1142 struct g_raid_lock *lp; 1143 struct g_raid_softc *sc; 1144 struct bio *bp; 1145 1146 sc = vol->v_softc; 1147 LIST_FOREACH(lp, &vol->v_locks, l_next) { 1148 if (lp->l_offset == off && lp->l_length == len) { 1149 LIST_REMOVE(lp, l_next); 1150 /* XXX 1151 * Right now we just put them all back on the queue 1152 * and hope for the best. We hope this because any 1153 * locked ranges will go right back on this list 1154 * when the worker thread runs. 1155 * XXX 1156 */ 1157 G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd", 1158 (intmax_t)lp->l_offset, 1159 (intmax_t)(lp->l_offset+lp->l_length)); 1160 mtx_lock(&sc->sc_queue_mtx); 1161 while ((bp = bioq_takefirst(&vol->v_locked)) != NULL) 1162 bioq_disksort(&sc->sc_queue, bp); 1163 mtx_unlock(&sc->sc_queue_mtx); 1164 free(lp, M_RAID); 1165 return (0); 1166 } 1167 } 1168 return (EINVAL); 1169 } 1170 1171 void 1172 g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp) 1173 { 1174 struct g_consumer *cp; 1175 struct g_raid_disk *disk, *tdisk; 1176 1177 bp->bio_caller1 = sd; 1178 1179 /* 1180 * Make sure that the disk is present. Generally it is a task of 1181 * transformation layers to not send requests to absent disks, but 1182 * it is better to be safe and report situation then sorry. 1183 */ 1184 if (sd->sd_disk == NULL) { 1185 G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!"); 1186 nodisk: 1187 bp->bio_from = NULL; 1188 bp->bio_to = NULL; 1189 bp->bio_error = ENXIO; 1190 g_raid_disk_done(bp); 1191 return; 1192 } 1193 disk = sd->sd_disk; 1194 if (disk->d_state != G_RAID_DISK_S_ACTIVE && 1195 disk->d_state != G_RAID_DISK_S_FAILED) { 1196 G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a " 1197 "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 1198 goto nodisk; 1199 } 1200 1201 cp = disk->d_consumer; 1202 bp->bio_from = cp; 1203 bp->bio_to = cp->provider; 1204 cp->index++; 1205 1206 /* Update average disks load. */ 1207 TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) { 1208 if (tdisk->d_consumer == NULL) 1209 tdisk->d_load = 0; 1210 else 1211 tdisk->d_load = (tdisk->d_consumer->index * 1212 G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8; 1213 } 1214 1215 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1216 if (dumping) { 1217 G_RAID_LOGREQ(3, bp, "Sending dumping request."); 1218 if (bp->bio_cmd == BIO_WRITE) { 1219 bp->bio_error = g_raid_subdisk_kerneldump(sd, 1220 bp->bio_data, 0, bp->bio_offset, bp->bio_length); 1221 } else 1222 bp->bio_error = EOPNOTSUPP; 1223 g_raid_disk_done(bp); 1224 } else { 1225 bp->bio_done = g_raid_disk_done; 1226 bp->bio_offset += sd->sd_offset; 1227 G_RAID_LOGREQ(3, bp, "Sending request."); 1228 g_io_request(bp, cp); 1229 } 1230 } 1231 1232 int 1233 g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd, 1234 void *virtual, vm_offset_t physical, off_t offset, size_t length) 1235 { 1236 1237 if (sd->sd_disk == NULL) 1238 return (ENXIO); 1239 if (sd->sd_disk->d_kd.di.dumper == NULL) 1240 return (EOPNOTSUPP); 1241 return (dump_write(&sd->sd_disk->d_kd.di, 1242 virtual, physical, 1243 sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset, 1244 length)); 1245 } 1246 1247 static void 1248 g_raid_disk_done(struct bio *bp) 1249 { 1250 struct g_raid_softc *sc; 1251 struct g_raid_subdisk *sd; 1252 1253 sd = bp->bio_caller1; 1254 sc = sd->sd_softc; 1255 mtx_lock(&sc->sc_queue_mtx); 1256 bioq_disksort(&sc->sc_queue, bp); 1257 mtx_unlock(&sc->sc_queue_mtx); 1258 if (!dumping) 1259 wakeup(sc); 1260 } 1261 1262 static void 1263 g_raid_disk_done_request(struct bio *bp) 1264 { 1265 struct g_raid_softc *sc; 1266 struct g_raid_disk *disk; 1267 struct g_raid_subdisk *sd; 1268 struct g_raid_volume *vol; 1269 1270 g_topology_assert_not(); 1271 1272 G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error); 1273 sd = bp->bio_caller1; 1274 sc = sd->sd_softc; 1275 vol = sd->sd_volume; 1276 if (bp->bio_from != NULL) { 1277 bp->bio_from->index--; 1278 disk = bp->bio_from->private; 1279 if (disk == NULL) 1280 g_raid_kill_consumer(sc, bp->bio_from); 1281 } 1282 bp->bio_offset -= sd->sd_offset; 1283 1284 G_RAID_TR_IODONE(vol->v_tr, sd, bp); 1285 } 1286 1287 static void 1288 g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep) 1289 { 1290 1291 if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0) 1292 ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event); 1293 else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0) 1294 ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event); 1295 else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0) 1296 ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event); 1297 else 1298 ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event); 1299 if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) { 1300 KASSERT(ep->e_error == 0, 1301 ("Error cannot be handled.")); 1302 g_raid_event_free(ep); 1303 } else { 1304 ep->e_flags |= G_RAID_EVENT_DONE; 1305 G_RAID_DEBUG1(4, sc, "Waking up %p.", ep); 1306 mtx_lock(&sc->sc_queue_mtx); 1307 wakeup(ep); 1308 mtx_unlock(&sc->sc_queue_mtx); 1309 } 1310 } 1311 1312 /* 1313 * Worker thread. 1314 */ 1315 static void 1316 g_raid_worker(void *arg) 1317 { 1318 struct g_raid_softc *sc; 1319 struct g_raid_event *ep; 1320 struct g_raid_volume *vol; 1321 struct bio *bp; 1322 struct timeval now, t; 1323 int timeout, rv; 1324 1325 sc = arg; 1326 thread_lock(curthread); 1327 sched_prio(curthread, PRIBIO); 1328 thread_unlock(curthread); 1329 1330 sx_xlock(&sc->sc_lock); 1331 for (;;) { 1332 mtx_lock(&sc->sc_queue_mtx); 1333 /* 1334 * First take a look at events. 1335 * This is important to handle events before any I/O requests. 1336 */ 1337 bp = NULL; 1338 vol = NULL; 1339 rv = 0; 1340 ep = TAILQ_FIRST(&sc->sc_events); 1341 if (ep != NULL) 1342 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 1343 else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) 1344 ; 1345 else { 1346 getmicrouptime(&now); 1347 t = now; 1348 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1349 if (bioq_first(&vol->v_inflight) == NULL && 1350 vol->v_tr && 1351 timevalcmp(&vol->v_last_done, &t, < )) 1352 t = vol->v_last_done; 1353 } 1354 timevalsub(&t, &now); 1355 timeout = g_raid_idle_threshold + 1356 t.tv_sec * 1000000 + t.tv_usec; 1357 if (timeout > 0) { 1358 /* 1359 * Two steps to avoid overflows at HZ=1000 1360 * and idle timeouts > 2.1s. Some rounding 1361 * errors can occur, but they are < 1tick, 1362 * which is deemed to be close enough for 1363 * this purpose. 1364 */ 1365 int micpertic = 1000000 / hz; 1366 timeout = (timeout + micpertic - 1) / micpertic; 1367 sx_xunlock(&sc->sc_lock); 1368 MSLEEP(rv, sc, &sc->sc_queue_mtx, 1369 PRIBIO | PDROP, "-", timeout); 1370 sx_xlock(&sc->sc_lock); 1371 goto process; 1372 } else 1373 rv = EWOULDBLOCK; 1374 } 1375 mtx_unlock(&sc->sc_queue_mtx); 1376 process: 1377 if (ep != NULL) { 1378 g_raid_handle_event(sc, ep); 1379 } else if (bp != NULL) { 1380 if (bp->bio_to != NULL && 1381 bp->bio_to->geom == sc->sc_geom) 1382 g_raid_start_request(bp); 1383 else 1384 g_raid_disk_done_request(bp); 1385 } else if (rv == EWOULDBLOCK) { 1386 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1387 if (vol->v_writes == 0 && vol->v_dirty) 1388 g_raid_clean(vol, -1); 1389 if (bioq_first(&vol->v_inflight) == NULL && 1390 vol->v_tr) { 1391 t.tv_sec = g_raid_idle_threshold / 1000000; 1392 t.tv_usec = g_raid_idle_threshold % 1000000; 1393 timevaladd(&t, &vol->v_last_done); 1394 getmicrouptime(&now); 1395 if (timevalcmp(&t, &now, <= )) { 1396 G_RAID_TR_IDLE(vol->v_tr); 1397 vol->v_last_done = now; 1398 } 1399 } 1400 } 1401 } 1402 if (sc->sc_stopping == G_RAID_DESTROY_HARD) 1403 g_raid_destroy_node(sc, 1); /* May not return. */ 1404 } 1405 } 1406 1407 static void 1408 g_raid_poll(struct g_raid_softc *sc) 1409 { 1410 struct g_raid_event *ep; 1411 struct bio *bp; 1412 1413 sx_xlock(&sc->sc_lock); 1414 mtx_lock(&sc->sc_queue_mtx); 1415 /* 1416 * First take a look at events. 1417 * This is important to handle events before any I/O requests. 1418 */ 1419 ep = TAILQ_FIRST(&sc->sc_events); 1420 if (ep != NULL) { 1421 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 1422 mtx_unlock(&sc->sc_queue_mtx); 1423 g_raid_handle_event(sc, ep); 1424 goto out; 1425 } 1426 bp = bioq_takefirst(&sc->sc_queue); 1427 if (bp != NULL) { 1428 mtx_unlock(&sc->sc_queue_mtx); 1429 if (bp->bio_from == NULL || 1430 bp->bio_from->geom != sc->sc_geom) 1431 g_raid_start_request(bp); 1432 else 1433 g_raid_disk_done_request(bp); 1434 } 1435 out: 1436 sx_xunlock(&sc->sc_lock); 1437 } 1438 1439 static void 1440 g_raid_launch_provider(struct g_raid_volume *vol) 1441 { 1442 struct g_raid_disk *disk; 1443 struct g_raid_softc *sc; 1444 struct g_provider *pp; 1445 char name[G_RAID_MAX_VOLUMENAME]; 1446 off_t off; 1447 1448 sc = vol->v_softc; 1449 sx_assert(&sc->sc_lock, SX_LOCKED); 1450 1451 g_topology_lock(); 1452 /* Try to name provider with volume name. */ 1453 snprintf(name, sizeof(name), "raid/%s", vol->v_name); 1454 if (g_raid_name_format == 0 || vol->v_name[0] == 0 || 1455 g_provider_by_name(name) != NULL) { 1456 /* Otherwise use sequential volume number. */ 1457 snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id); 1458 } 1459 pp = g_new_providerf(sc->sc_geom, "%s", name); 1460 pp->private = vol; 1461 pp->mediasize = vol->v_mediasize; 1462 pp->sectorsize = vol->v_sectorsize; 1463 pp->stripesize = 0; 1464 pp->stripeoffset = 0; 1465 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 || 1466 vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 || 1467 vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE || 1468 vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) { 1469 if ((disk = vol->v_subdisks[0].sd_disk) != NULL && 1470 disk->d_consumer != NULL && 1471 disk->d_consumer->provider != NULL) { 1472 pp->stripesize = disk->d_consumer->provider->stripesize; 1473 off = disk->d_consumer->provider->stripeoffset; 1474 pp->stripeoffset = off + vol->v_subdisks[0].sd_offset; 1475 if (off > 0) 1476 pp->stripeoffset %= off; 1477 } 1478 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) { 1479 pp->stripesize *= (vol->v_disks_count - 1); 1480 pp->stripeoffset *= (vol->v_disks_count - 1); 1481 } 1482 } else 1483 pp->stripesize = vol->v_strip_size; 1484 vol->v_provider = pp; 1485 g_error_provider(pp, 0); 1486 g_topology_unlock(); 1487 G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.", 1488 pp->name, vol->v_name); 1489 } 1490 1491 static void 1492 g_raid_destroy_provider(struct g_raid_volume *vol) 1493 { 1494 struct g_raid_softc *sc; 1495 struct g_provider *pp; 1496 struct bio *bp, *tmp; 1497 1498 g_topology_assert_not(); 1499 sc = vol->v_softc; 1500 pp = vol->v_provider; 1501 KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name)); 1502 1503 g_topology_lock(); 1504 g_error_provider(pp, ENXIO); 1505 mtx_lock(&sc->sc_queue_mtx); 1506 TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) { 1507 if (bp->bio_to != pp) 1508 continue; 1509 bioq_remove(&sc->sc_queue, bp); 1510 g_io_deliver(bp, ENXIO); 1511 } 1512 mtx_unlock(&sc->sc_queue_mtx); 1513 G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.", 1514 pp->name, vol->v_name); 1515 g_wither_provider(pp, ENXIO); 1516 g_topology_unlock(); 1517 vol->v_provider = NULL; 1518 } 1519 1520 /* 1521 * Update device state. 1522 */ 1523 static int 1524 g_raid_update_volume(struct g_raid_volume *vol, u_int event) 1525 { 1526 struct g_raid_softc *sc; 1527 1528 sc = vol->v_softc; 1529 sx_assert(&sc->sc_lock, SX_XLOCKED); 1530 1531 G_RAID_DEBUG1(2, sc, "Event %s for volume %s.", 1532 g_raid_volume_event2str(event), 1533 vol->v_name); 1534 switch (event) { 1535 case G_RAID_VOLUME_E_DOWN: 1536 if (vol->v_provider != NULL) 1537 g_raid_destroy_provider(vol); 1538 break; 1539 case G_RAID_VOLUME_E_UP: 1540 if (vol->v_provider == NULL) 1541 g_raid_launch_provider(vol); 1542 break; 1543 case G_RAID_VOLUME_E_START: 1544 if (vol->v_tr) 1545 G_RAID_TR_START(vol->v_tr); 1546 return (0); 1547 default: 1548 if (sc->sc_md) 1549 G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event); 1550 return (0); 1551 } 1552 1553 /* Manage root mount release. */ 1554 if (vol->v_starting) { 1555 vol->v_starting = 0; 1556 G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount); 1557 root_mount_rel(vol->v_rootmount); 1558 vol->v_rootmount = NULL; 1559 } 1560 if (vol->v_stopping && vol->v_provider_open == 0) 1561 g_raid_destroy_volume(vol); 1562 return (0); 1563 } 1564 1565 /* 1566 * Update subdisk state. 1567 */ 1568 static int 1569 g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event) 1570 { 1571 struct g_raid_softc *sc; 1572 struct g_raid_volume *vol; 1573 1574 sc = sd->sd_softc; 1575 vol = sd->sd_volume; 1576 sx_assert(&sc->sc_lock, SX_XLOCKED); 1577 1578 G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.", 1579 g_raid_subdisk_event2str(event), 1580 vol->v_name, sd->sd_pos, 1581 sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]"); 1582 if (vol->v_tr) 1583 G_RAID_TR_EVENT(vol->v_tr, sd, event); 1584 1585 return (0); 1586 } 1587 1588 /* 1589 * Update disk state. 1590 */ 1591 static int 1592 g_raid_update_disk(struct g_raid_disk *disk, u_int event) 1593 { 1594 struct g_raid_softc *sc; 1595 1596 sc = disk->d_softc; 1597 sx_assert(&sc->sc_lock, SX_XLOCKED); 1598 1599 G_RAID_DEBUG1(2, sc, "Event %s for disk %s.", 1600 g_raid_disk_event2str(event), 1601 g_raid_get_diskname(disk)); 1602 1603 if (sc->sc_md) 1604 G_RAID_MD_EVENT(sc->sc_md, disk, event); 1605 return (0); 1606 } 1607 1608 /* 1609 * Node event. 1610 */ 1611 static int 1612 g_raid_update_node(struct g_raid_softc *sc, u_int event) 1613 { 1614 sx_assert(&sc->sc_lock, SX_XLOCKED); 1615 1616 G_RAID_DEBUG1(2, sc, "Event %s for the array.", 1617 g_raid_node_event2str(event)); 1618 1619 if (event == G_RAID_NODE_E_WAKE) 1620 return (0); 1621 if (sc->sc_md) 1622 G_RAID_MD_EVENT(sc->sc_md, NULL, event); 1623 return (0); 1624 } 1625 1626 static int 1627 g_raid_access(struct g_provider *pp, int acr, int acw, int ace) 1628 { 1629 struct g_raid_volume *vol; 1630 struct g_raid_softc *sc; 1631 int dcw, opens, error = 0; 1632 1633 g_topology_assert(); 1634 sc = pp->geom->softc; 1635 vol = pp->private; 1636 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 1637 KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name)); 1638 1639 G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name, 1640 acr, acw, ace); 1641 dcw = pp->acw + acw; 1642 1643 g_topology_unlock(); 1644 sx_xlock(&sc->sc_lock); 1645 /* Deny new opens while dying. */ 1646 if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) { 1647 error = ENXIO; 1648 goto out; 1649 } 1650 if (dcw == 0 && vol->v_dirty) 1651 g_raid_clean(vol, dcw); 1652 vol->v_provider_open += acr + acw + ace; 1653 /* Handle delayed node destruction. */ 1654 if (sc->sc_stopping == G_RAID_DESTROY_DELAYED && 1655 vol->v_provider_open == 0) { 1656 /* Count open volumes. */ 1657 opens = g_raid_nopens(sc); 1658 if (opens == 0) { 1659 sc->sc_stopping = G_RAID_DESTROY_HARD; 1660 /* Wake up worker to make it selfdestruct. */ 1661 g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 1662 } 1663 } 1664 /* Handle open volume destruction. */ 1665 if (vol->v_stopping && vol->v_provider_open == 0) 1666 g_raid_destroy_volume(vol); 1667 out: 1668 sx_xunlock(&sc->sc_lock); 1669 g_topology_lock(); 1670 return (error); 1671 } 1672 1673 struct g_raid_softc * 1674 g_raid_create_node(struct g_class *mp, 1675 const char *name, struct g_raid_md_object *md) 1676 { 1677 struct g_raid_softc *sc; 1678 struct g_geom *gp; 1679 int error; 1680 1681 g_topology_assert(); 1682 G_RAID_DEBUG(1, "Creating array %s.", name); 1683 1684 gp = g_new_geomf(mp, "%s", name); 1685 sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO); 1686 gp->start = g_raid_start; 1687 gp->orphan = g_raid_orphan; 1688 gp->access = g_raid_access; 1689 gp->dumpconf = g_raid_dumpconf; 1690 1691 sc->sc_md = md; 1692 sc->sc_geom = gp; 1693 sc->sc_flags = 0; 1694 TAILQ_INIT(&sc->sc_volumes); 1695 TAILQ_INIT(&sc->sc_disks); 1696 sx_init(&sc->sc_lock, "gmirror:lock"); 1697 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 1698 TAILQ_INIT(&sc->sc_events); 1699 bioq_init(&sc->sc_queue); 1700 gp->softc = sc; 1701 error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0, 1702 "g_raid %s", name); 1703 if (error != 0) { 1704 G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name); 1705 mtx_destroy(&sc->sc_queue_mtx); 1706 sx_destroy(&sc->sc_lock); 1707 g_destroy_geom(sc->sc_geom); 1708 free(sc, M_RAID); 1709 return (NULL); 1710 } 1711 1712 G_RAID_DEBUG1(0, sc, "Array %s created.", name); 1713 return (sc); 1714 } 1715 1716 struct g_raid_volume * 1717 g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id) 1718 { 1719 struct g_raid_volume *vol, *vol1; 1720 int i; 1721 1722 G_RAID_DEBUG1(1, sc, "Creating volume %s.", name); 1723 vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO); 1724 vol->v_softc = sc; 1725 strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME); 1726 vol->v_state = G_RAID_VOLUME_S_STARTING; 1727 vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; 1728 vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN; 1729 bioq_init(&vol->v_inflight); 1730 bioq_init(&vol->v_locked); 1731 LIST_INIT(&vol->v_locks); 1732 for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 1733 vol->v_subdisks[i].sd_softc = sc; 1734 vol->v_subdisks[i].sd_volume = vol; 1735 vol->v_subdisks[i].sd_pos = i; 1736 vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE; 1737 } 1738 1739 /* Find free ID for this volume. */ 1740 g_topology_lock(); 1741 vol1 = vol; 1742 if (id >= 0) { 1743 LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 1744 if (vol1->v_global_id == id) 1745 break; 1746 } 1747 } 1748 if (vol1 != NULL) { 1749 for (id = 0; ; id++) { 1750 LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 1751 if (vol1->v_global_id == id) 1752 break; 1753 } 1754 if (vol1 == NULL) 1755 break; 1756 } 1757 } 1758 vol->v_global_id = id; 1759 LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next); 1760 g_topology_unlock(); 1761 1762 /* Delay root mounting. */ 1763 vol->v_rootmount = root_mount_hold("GRAID"); 1764 G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount); 1765 vol->v_starting = 1; 1766 TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next); 1767 return (vol); 1768 } 1769 1770 struct g_raid_disk * 1771 g_raid_create_disk(struct g_raid_softc *sc) 1772 { 1773 struct g_raid_disk *disk; 1774 1775 G_RAID_DEBUG1(1, sc, "Creating disk."); 1776 disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO); 1777 disk->d_softc = sc; 1778 disk->d_state = G_RAID_DISK_S_NONE; 1779 TAILQ_INIT(&disk->d_subdisks); 1780 TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next); 1781 return (disk); 1782 } 1783 1784 int g_raid_start_volume(struct g_raid_volume *vol) 1785 { 1786 struct g_raid_tr_class *class; 1787 struct g_raid_tr_object *obj; 1788 int status; 1789 1790 G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name); 1791 LIST_FOREACH(class, &g_raid_tr_classes, trc_list) { 1792 G_RAID_DEBUG1(2, vol->v_softc, 1793 "Tasting volume %s for %s transformation.", 1794 vol->v_name, class->name); 1795 obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 1796 M_WAITOK); 1797 obj->tro_class = class; 1798 obj->tro_volume = vol; 1799 status = G_RAID_TR_TASTE(obj, vol); 1800 if (status != G_RAID_TR_TASTE_FAIL) 1801 break; 1802 kobj_delete((kobj_t)obj, M_RAID); 1803 } 1804 if (class == NULL) { 1805 G_RAID_DEBUG1(0, vol->v_softc, 1806 "No transformation module found for %s.", 1807 vol->v_name); 1808 vol->v_tr = NULL; 1809 g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED); 1810 g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN, 1811 G_RAID_EVENT_VOLUME); 1812 return (-1); 1813 } 1814 G_RAID_DEBUG1(2, vol->v_softc, 1815 "Transformation module %s chosen for %s.", 1816 class->name, vol->v_name); 1817 vol->v_tr = obj; 1818 return (0); 1819 } 1820 1821 int 1822 g_raid_destroy_node(struct g_raid_softc *sc, int worker) 1823 { 1824 struct g_raid_volume *vol, *tmpv; 1825 struct g_raid_disk *disk, *tmpd; 1826 int error = 0; 1827 1828 sc->sc_stopping = G_RAID_DESTROY_HARD; 1829 TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) { 1830 if (g_raid_destroy_volume(vol)) 1831 error = EBUSY; 1832 } 1833 if (error) 1834 return (error); 1835 TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) { 1836 if (g_raid_destroy_disk(disk)) 1837 error = EBUSY; 1838 } 1839 if (error) 1840 return (error); 1841 if (sc->sc_md) { 1842 G_RAID_MD_FREE(sc->sc_md); 1843 kobj_delete((kobj_t)sc->sc_md, M_RAID); 1844 sc->sc_md = NULL; 1845 } 1846 if (sc->sc_geom != NULL) { 1847 G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name); 1848 g_topology_lock(); 1849 sc->sc_geom->softc = NULL; 1850 g_wither_geom(sc->sc_geom, ENXIO); 1851 g_topology_unlock(); 1852 sc->sc_geom = NULL; 1853 } else 1854 G_RAID_DEBUG(1, "Array destroyed."); 1855 if (worker) { 1856 g_raid_event_cancel(sc, sc); 1857 mtx_destroy(&sc->sc_queue_mtx); 1858 sx_xunlock(&sc->sc_lock); 1859 sx_destroy(&sc->sc_lock); 1860 wakeup(&sc->sc_stopping); 1861 free(sc, M_RAID); 1862 curthread->td_pflags &= ~TDP_GEOM; 1863 G_RAID_DEBUG(1, "Thread exiting."); 1864 kproc_exit(0); 1865 } else { 1866 /* Wake up worker to make it selfdestruct. */ 1867 g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 1868 } 1869 return (0); 1870 } 1871 1872 int 1873 g_raid_destroy_volume(struct g_raid_volume *vol) 1874 { 1875 struct g_raid_softc *sc; 1876 struct g_raid_disk *disk; 1877 int i; 1878 1879 sc = vol->v_softc; 1880 G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name); 1881 vol->v_stopping = 1; 1882 if (vol->v_state != G_RAID_VOLUME_S_STOPPED) { 1883 if (vol->v_tr) { 1884 G_RAID_TR_STOP(vol->v_tr); 1885 return (EBUSY); 1886 } else 1887 vol->v_state = G_RAID_VOLUME_S_STOPPED; 1888 } 1889 if (g_raid_event_check(sc, vol) != 0) 1890 return (EBUSY); 1891 if (vol->v_provider != NULL) 1892 return (EBUSY); 1893 if (vol->v_provider_open != 0) 1894 return (EBUSY); 1895 if (vol->v_tr) { 1896 G_RAID_TR_FREE(vol->v_tr); 1897 kobj_delete((kobj_t)vol->v_tr, M_RAID); 1898 vol->v_tr = NULL; 1899 } 1900 if (vol->v_rootmount) 1901 root_mount_rel(vol->v_rootmount); 1902 g_topology_lock(); 1903 LIST_REMOVE(vol, v_global_next); 1904 g_topology_unlock(); 1905 TAILQ_REMOVE(&sc->sc_volumes, vol, v_next); 1906 for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 1907 g_raid_event_cancel(sc, &vol->v_subdisks[i]); 1908 disk = vol->v_subdisks[i].sd_disk; 1909 if (disk == NULL) 1910 continue; 1911 TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next); 1912 } 1913 G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name); 1914 if (sc->sc_md) 1915 G_RAID_MD_FREE_VOLUME(sc->sc_md, vol); 1916 g_raid_event_cancel(sc, vol); 1917 free(vol, M_RAID); 1918 if (sc->sc_stopping == G_RAID_DESTROY_HARD) { 1919 /* Wake up worker to let it selfdestruct. */ 1920 g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 1921 } 1922 return (0); 1923 } 1924 1925 int 1926 g_raid_destroy_disk(struct g_raid_disk *disk) 1927 { 1928 struct g_raid_softc *sc; 1929 struct g_raid_subdisk *sd, *tmp; 1930 1931 sc = disk->d_softc; 1932 G_RAID_DEBUG1(2, sc, "Destroying disk."); 1933 if (disk->d_consumer) { 1934 g_raid_kill_consumer(sc, disk->d_consumer); 1935 disk->d_consumer = NULL; 1936 } 1937 TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) { 1938 g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE); 1939 g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, 1940 G_RAID_EVENT_SUBDISK); 1941 TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next); 1942 sd->sd_disk = NULL; 1943 } 1944 TAILQ_REMOVE(&sc->sc_disks, disk, d_next); 1945 if (sc->sc_md) 1946 G_RAID_MD_FREE_DISK(sc->sc_md, disk); 1947 g_raid_event_cancel(sc, disk); 1948 free(disk, M_RAID); 1949 return (0); 1950 } 1951 1952 int 1953 g_raid_destroy(struct g_raid_softc *sc, int how) 1954 { 1955 int opens; 1956 1957 g_topology_assert_not(); 1958 if (sc == NULL) 1959 return (ENXIO); 1960 sx_assert(&sc->sc_lock, SX_XLOCKED); 1961 1962 /* Count open volumes. */ 1963 opens = g_raid_nopens(sc); 1964 1965 /* React on some opened volumes. */ 1966 if (opens > 0) { 1967 switch (how) { 1968 case G_RAID_DESTROY_SOFT: 1969 G_RAID_DEBUG1(1, sc, 1970 "%d volumes are still open.", 1971 opens); 1972 return (EBUSY); 1973 case G_RAID_DESTROY_DELAYED: 1974 G_RAID_DEBUG1(1, sc, 1975 "Array will be destroyed on last close."); 1976 sc->sc_stopping = G_RAID_DESTROY_DELAYED; 1977 return (EBUSY); 1978 case G_RAID_DESTROY_HARD: 1979 G_RAID_DEBUG1(1, sc, 1980 "%d volumes are still open.", 1981 opens); 1982 } 1983 } 1984 1985 /* Mark node for destruction. */ 1986 sc->sc_stopping = G_RAID_DESTROY_HARD; 1987 /* Wake up worker to let it selfdestruct. */ 1988 g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 1989 /* Sleep until node destroyed. */ 1990 sx_sleep(&sc->sc_stopping, &sc->sc_lock, 1991 PRIBIO | PDROP, "r:destroy", 0); 1992 return (0); 1993 } 1994 1995 static void 1996 g_raid_taste_orphan(struct g_consumer *cp) 1997 { 1998 1999 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2000 cp->provider->name)); 2001 } 2002 2003 static struct g_geom * 2004 g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2005 { 2006 struct g_consumer *cp; 2007 struct g_geom *gp, *geom; 2008 struct g_raid_md_class *class; 2009 struct g_raid_md_object *obj; 2010 int status; 2011 2012 g_topology_assert(); 2013 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2014 G_RAID_DEBUG(2, "Tasting provider %s.", pp->name); 2015 2016 gp = g_new_geomf(mp, "mirror:taste"); 2017 /* 2018 * This orphan function should be never called. 2019 */ 2020 gp->orphan = g_raid_taste_orphan; 2021 cp = g_new_consumer(gp); 2022 g_attach(cp, pp); 2023 2024 geom = NULL; 2025 LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 2026 G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.", 2027 pp->name, class->name); 2028 obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 2029 M_WAITOK); 2030 obj->mdo_class = class; 2031 status = G_RAID_MD_TASTE(obj, mp, cp, &geom); 2032 if (status != G_RAID_MD_TASTE_NEW) 2033 kobj_delete((kobj_t)obj, M_RAID); 2034 if (status != G_RAID_MD_TASTE_FAIL) 2035 break; 2036 } 2037 2038 g_detach(cp); 2039 g_destroy_consumer(cp); 2040 g_destroy_geom(gp); 2041 G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name); 2042 return (geom); 2043 } 2044 2045 int 2046 g_raid_create_node_format(const char *format, struct g_geom **gp) 2047 { 2048 struct g_raid_md_class *class; 2049 struct g_raid_md_object *obj; 2050 int status; 2051 2052 G_RAID_DEBUG(2, "Creating array for %s metadata.", format); 2053 LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 2054 if (strcasecmp(class->name, format) == 0) 2055 break; 2056 } 2057 if (class == NULL) { 2058 G_RAID_DEBUG(1, "No support for %s metadata.", format); 2059 return (G_RAID_MD_TASTE_FAIL); 2060 } 2061 obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 2062 M_WAITOK); 2063 obj->mdo_class = class; 2064 status = G_RAID_MD_CREATE(obj, &g_raid_class, gp); 2065 if (status != G_RAID_MD_TASTE_NEW) 2066 kobj_delete((kobj_t)obj, M_RAID); 2067 return (status); 2068 } 2069 2070 static int 2071 g_raid_destroy_geom(struct gctl_req *req __unused, 2072 struct g_class *mp __unused, struct g_geom *gp) 2073 { 2074 struct g_raid_softc *sc; 2075 int error; 2076 2077 g_topology_unlock(); 2078 sc = gp->softc; 2079 sx_xlock(&sc->sc_lock); 2080 g_cancel_event(sc); 2081 error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT); 2082 if (error != 0) 2083 sx_xunlock(&sc->sc_lock); 2084 g_topology_lock(); 2085 return (error); 2086 } 2087 2088 void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol, 2089 struct g_raid_subdisk *sd, struct g_raid_disk *disk) 2090 { 2091 2092 if (sc->sc_stopping == G_RAID_DESTROY_HARD) 2093 return; 2094 if (sc->sc_md) 2095 G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk); 2096 } 2097 2098 void g_raid_fail_disk(struct g_raid_softc *sc, 2099 struct g_raid_subdisk *sd, struct g_raid_disk *disk) 2100 { 2101 2102 if (disk == NULL) 2103 disk = sd->sd_disk; 2104 if (disk == NULL) { 2105 G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!"); 2106 return; 2107 } 2108 if (disk->d_state != G_RAID_DISK_S_ACTIVE) { 2109 G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a " 2110 "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 2111 return; 2112 } 2113 if (sc->sc_md) 2114 G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk); 2115 } 2116 2117 static void 2118 g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2119 struct g_consumer *cp, struct g_provider *pp) 2120 { 2121 struct g_raid_softc *sc; 2122 struct g_raid_volume *vol; 2123 struct g_raid_subdisk *sd; 2124 struct g_raid_disk *disk; 2125 int i, s; 2126 2127 g_topology_assert(); 2128 2129 sc = gp->softc; 2130 if (sc == NULL) 2131 return; 2132 if (pp != NULL) { 2133 vol = pp->private; 2134 g_topology_unlock(); 2135 sx_xlock(&sc->sc_lock); 2136 sbuf_printf(sb, "%s<Label>%s</Label>\n", indent, 2137 vol->v_name); 2138 sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent, 2139 g_raid_volume_level2str(vol->v_raid_level, 2140 vol->v_raid_level_qualifier)); 2141 sbuf_printf(sb, 2142 "%s<Transformation>%s</Transformation>\n", indent, 2143 vol->v_tr ? vol->v_tr->tro_class->name : "NONE"); 2144 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2145 vol->v_disks_count); 2146 sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent, 2147 vol->v_strip_size); 2148 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2149 g_raid_volume_state2str(vol->v_state)); 2150 sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent, 2151 vol->v_dirty ? "Yes" : "No"); 2152 sbuf_printf(sb, "%s<Subdisks>", indent); 2153 for (i = 0; i < vol->v_disks_count; i++) { 2154 sd = &vol->v_subdisks[i]; 2155 if (sd->sd_disk != NULL && 2156 sd->sd_disk->d_consumer != NULL) { 2157 sbuf_printf(sb, "%s ", 2158 g_raid_get_diskname(sd->sd_disk)); 2159 } else { 2160 sbuf_printf(sb, "NONE "); 2161 } 2162 sbuf_printf(sb, "(%s", 2163 g_raid_subdisk_state2str(sd->sd_state)); 2164 if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2165 sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 2166 sbuf_printf(sb, " %d%%", 2167 (int)(sd->sd_rebuild_pos * 100 / 2168 sd->sd_size)); 2169 } 2170 sbuf_printf(sb, ")"); 2171 if (i + 1 < vol->v_disks_count) 2172 sbuf_printf(sb, ", "); 2173 } 2174 sbuf_printf(sb, "</Subdisks>\n"); 2175 sx_xunlock(&sc->sc_lock); 2176 g_topology_lock(); 2177 } else if (cp != NULL) { 2178 disk = cp->private; 2179 if (disk == NULL) 2180 return; 2181 g_topology_unlock(); 2182 sx_xlock(&sc->sc_lock); 2183 sbuf_printf(sb, "%s<State>%s", indent, 2184 g_raid_disk_state2str(disk->d_state)); 2185 if (!TAILQ_EMPTY(&disk->d_subdisks)) { 2186 sbuf_printf(sb, " ("); 2187 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 2188 sbuf_printf(sb, "%s", 2189 g_raid_subdisk_state2str(sd->sd_state)); 2190 if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2191 sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 2192 sbuf_printf(sb, " %d%%", 2193 (int)(sd->sd_rebuild_pos * 100 / 2194 sd->sd_size)); 2195 } 2196 if (TAILQ_NEXT(sd, sd_next)) 2197 sbuf_printf(sb, ", "); 2198 } 2199 sbuf_printf(sb, ")"); 2200 } 2201 sbuf_printf(sb, "</State>\n"); 2202 sbuf_printf(sb, "%s<Subdisks>", indent); 2203 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 2204 sbuf_printf(sb, "r%d(%s):%d@%ju", 2205 sd->sd_volume->v_global_id, 2206 sd->sd_volume->v_name, 2207 sd->sd_pos, sd->sd_offset); 2208 if (TAILQ_NEXT(sd, sd_next)) 2209 sbuf_printf(sb, ", "); 2210 } 2211 sbuf_printf(sb, "</Subdisks>\n"); 2212 sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent, 2213 disk->d_read_errs); 2214 sx_xunlock(&sc->sc_lock); 2215 g_topology_lock(); 2216 } else { 2217 g_topology_unlock(); 2218 sx_xlock(&sc->sc_lock); 2219 if (sc->sc_md) { 2220 sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent, 2221 sc->sc_md->mdo_class->name); 2222 } 2223 if (!TAILQ_EMPTY(&sc->sc_volumes)) { 2224 s = 0xff; 2225 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 2226 if (vol->v_state < s) 2227 s = vol->v_state; 2228 } 2229 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2230 g_raid_volume_state2str(s)); 2231 } 2232 sx_xunlock(&sc->sc_lock); 2233 g_topology_lock(); 2234 } 2235 } 2236 2237 static void 2238 g_raid_shutdown_pre_sync(void *arg, int howto) 2239 { 2240 struct g_class *mp; 2241 struct g_geom *gp, *gp2; 2242 struct g_raid_softc *sc; 2243 int error; 2244 2245 mp = arg; 2246 DROP_GIANT(); 2247 g_topology_lock(); 2248 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2249 if ((sc = gp->softc) == NULL) 2250 continue; 2251 g_topology_unlock(); 2252 sx_xlock(&sc->sc_lock); 2253 g_cancel_event(sc); 2254 error = g_raid_destroy(sc, G_RAID_DESTROY_DELAYED); 2255 if (error != 0) 2256 sx_xunlock(&sc->sc_lock); 2257 g_topology_lock(); 2258 } 2259 g_topology_unlock(); 2260 PICKUP_GIANT(); 2261 } 2262 2263 static void 2264 g_raid_init(struct g_class *mp) 2265 { 2266 2267 g_raid_pre_sync = EVENTHANDLER_REGISTER(shutdown_pre_sync, 2268 g_raid_shutdown_pre_sync, mp, SHUTDOWN_PRI_FIRST); 2269 if (g_raid_pre_sync == NULL) 2270 G_RAID_DEBUG(0, "Warning! Cannot register shutdown event."); 2271 g_raid_started = 1; 2272 } 2273 2274 static void 2275 g_raid_fini(struct g_class *mp) 2276 { 2277 2278 if (g_raid_pre_sync != NULL) 2279 EVENTHANDLER_DEREGISTER(shutdown_pre_sync, g_raid_pre_sync); 2280 g_raid_started = 0; 2281 } 2282 2283 int 2284 g_raid_md_modevent(module_t mod, int type, void *arg) 2285 { 2286 struct g_raid_md_class *class, *c, *nc; 2287 int error; 2288 2289 error = 0; 2290 class = arg; 2291 switch (type) { 2292 case MOD_LOAD: 2293 c = LIST_FIRST(&g_raid_md_classes); 2294 if (c == NULL || c->mdc_priority > class->mdc_priority) 2295 LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list); 2296 else { 2297 while ((nc = LIST_NEXT(c, mdc_list)) != NULL && 2298 nc->mdc_priority < class->mdc_priority) 2299 c = nc; 2300 LIST_INSERT_AFTER(c, class, mdc_list); 2301 } 2302 if (g_raid_started) 2303 g_retaste(&g_raid_class); 2304 break; 2305 case MOD_UNLOAD: 2306 LIST_REMOVE(class, mdc_list); 2307 break; 2308 default: 2309 error = EOPNOTSUPP; 2310 break; 2311 } 2312 2313 return (error); 2314 } 2315 2316 int 2317 g_raid_tr_modevent(module_t mod, int type, void *arg) 2318 { 2319 struct g_raid_tr_class *class, *c, *nc; 2320 int error; 2321 2322 error = 0; 2323 class = arg; 2324 switch (type) { 2325 case MOD_LOAD: 2326 c = LIST_FIRST(&g_raid_tr_classes); 2327 if (c == NULL || c->trc_priority > class->trc_priority) 2328 LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list); 2329 else { 2330 while ((nc = LIST_NEXT(c, trc_list)) != NULL && 2331 nc->trc_priority < class->trc_priority) 2332 c = nc; 2333 LIST_INSERT_AFTER(c, class, trc_list); 2334 } 2335 break; 2336 case MOD_UNLOAD: 2337 LIST_REMOVE(class, trc_list); 2338 break; 2339 default: 2340 error = EOPNOTSUPP; 2341 break; 2342 } 2343 2344 return (error); 2345 } 2346 2347 /* 2348 * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid) 2349 * to reduce module priority, allowing submodules to register them first. 2350 */ 2351 static moduledata_t g_raid_mod = { 2352 "g_raid", 2353 g_modevent, 2354 &g_raid_class 2355 }; 2356 DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD); 2357 MODULE_VERSION(geom_raid, 0); 2358