1 /*- 2 * Copyright (c) 2004-2006 Pawel Jakub Dawidek <pjd@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 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/raid3/g_raid3.h> 48 49 FEATURE(geom_raid3, "GEOM RAID-3 functionality"); 50 51 static MALLOC_DEFINE(M_RAID3, "raid3_data", "GEOM_RAID3 Data"); 52 53 SYSCTL_DECL(_kern_geom); 54 static SYSCTL_NODE(_kern_geom, OID_AUTO, raid3, CTLFLAG_RW, 0, 55 "GEOM_RAID3 stuff"); 56 u_int g_raid3_debug = 0; 57 TUNABLE_INT("kern.geom.raid3.debug", &g_raid3_debug); 58 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, debug, CTLFLAG_RW, &g_raid3_debug, 0, 59 "Debug level"); 60 static u_int g_raid3_timeout = 4; 61 TUNABLE_INT("kern.geom.raid3.timeout", &g_raid3_timeout); 62 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, timeout, CTLFLAG_RW, &g_raid3_timeout, 63 0, "Time to wait on all raid3 components"); 64 static u_int g_raid3_idletime = 5; 65 TUNABLE_INT("kern.geom.raid3.idletime", &g_raid3_idletime); 66 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, idletime, CTLFLAG_RW, 67 &g_raid3_idletime, 0, "Mark components as clean when idling"); 68 static u_int g_raid3_disconnect_on_failure = 1; 69 TUNABLE_INT("kern.geom.raid3.disconnect_on_failure", 70 &g_raid3_disconnect_on_failure); 71 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, disconnect_on_failure, CTLFLAG_RW, 72 &g_raid3_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 73 static u_int g_raid3_syncreqs = 2; 74 TUNABLE_INT("kern.geom.raid3.sync_requests", &g_raid3_syncreqs); 75 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, sync_requests, CTLFLAG_RDTUN, 76 &g_raid3_syncreqs, 0, "Parallel synchronization I/O requests."); 77 static u_int g_raid3_use_malloc = 0; 78 TUNABLE_INT("kern.geom.raid3.use_malloc", &g_raid3_use_malloc); 79 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, use_malloc, CTLFLAG_RDTUN, 80 &g_raid3_use_malloc, 0, "Use malloc(9) instead of uma(9)."); 81 82 static u_int g_raid3_n64k = 50; 83 TUNABLE_INT("kern.geom.raid3.n64k", &g_raid3_n64k); 84 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n64k, CTLFLAG_RD, &g_raid3_n64k, 0, 85 "Maximum number of 64kB allocations"); 86 static u_int g_raid3_n16k = 200; 87 TUNABLE_INT("kern.geom.raid3.n16k", &g_raid3_n16k); 88 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n16k, CTLFLAG_RD, &g_raid3_n16k, 0, 89 "Maximum number of 16kB allocations"); 90 static u_int g_raid3_n4k = 1200; 91 TUNABLE_INT("kern.geom.raid3.n4k", &g_raid3_n4k); 92 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n4k, CTLFLAG_RD, &g_raid3_n4k, 0, 93 "Maximum number of 4kB allocations"); 94 95 static SYSCTL_NODE(_kern_geom_raid3, OID_AUTO, stat, CTLFLAG_RW, 0, 96 "GEOM_RAID3 statistics"); 97 static u_int g_raid3_parity_mismatch = 0; 98 SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, parity_mismatch, CTLFLAG_RD, 99 &g_raid3_parity_mismatch, 0, "Number of failures in VERIFY mode"); 100 101 #define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 102 G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 103 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 104 G_RAID3_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 105 } while (0) 106 107 static eventhandler_tag g_raid3_post_sync = NULL; 108 static int g_raid3_shutdown = 0; 109 110 static int g_raid3_destroy_geom(struct gctl_req *req, struct g_class *mp, 111 struct g_geom *gp); 112 static g_taste_t g_raid3_taste; 113 static void g_raid3_init(struct g_class *mp); 114 static void g_raid3_fini(struct g_class *mp); 115 116 struct g_class g_raid3_class = { 117 .name = G_RAID3_CLASS_NAME, 118 .version = G_VERSION, 119 .ctlreq = g_raid3_config, 120 .taste = g_raid3_taste, 121 .destroy_geom = g_raid3_destroy_geom, 122 .init = g_raid3_init, 123 .fini = g_raid3_fini 124 }; 125 126 127 static void g_raid3_destroy_provider(struct g_raid3_softc *sc); 128 static int g_raid3_update_disk(struct g_raid3_disk *disk, u_int state); 129 static void g_raid3_update_device(struct g_raid3_softc *sc, boolean_t force); 130 static void g_raid3_dumpconf(struct sbuf *sb, const char *indent, 131 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 132 static void g_raid3_sync_stop(struct g_raid3_softc *sc, int type); 133 static int g_raid3_register_request(struct bio *pbp); 134 static void g_raid3_sync_release(struct g_raid3_softc *sc); 135 136 137 static const char * 138 g_raid3_disk_state2str(int state) 139 { 140 141 switch (state) { 142 case G_RAID3_DISK_STATE_NODISK: 143 return ("NODISK"); 144 case G_RAID3_DISK_STATE_NONE: 145 return ("NONE"); 146 case G_RAID3_DISK_STATE_NEW: 147 return ("NEW"); 148 case G_RAID3_DISK_STATE_ACTIVE: 149 return ("ACTIVE"); 150 case G_RAID3_DISK_STATE_STALE: 151 return ("STALE"); 152 case G_RAID3_DISK_STATE_SYNCHRONIZING: 153 return ("SYNCHRONIZING"); 154 case G_RAID3_DISK_STATE_DISCONNECTED: 155 return ("DISCONNECTED"); 156 default: 157 return ("INVALID"); 158 } 159 } 160 161 static const char * 162 g_raid3_device_state2str(int state) 163 { 164 165 switch (state) { 166 case G_RAID3_DEVICE_STATE_STARTING: 167 return ("STARTING"); 168 case G_RAID3_DEVICE_STATE_DEGRADED: 169 return ("DEGRADED"); 170 case G_RAID3_DEVICE_STATE_COMPLETE: 171 return ("COMPLETE"); 172 default: 173 return ("INVALID"); 174 } 175 } 176 177 const char * 178 g_raid3_get_diskname(struct g_raid3_disk *disk) 179 { 180 181 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 182 return ("[unknown]"); 183 return (disk->d_name); 184 } 185 186 static void * 187 g_raid3_alloc(struct g_raid3_softc *sc, size_t size, int flags) 188 { 189 void *ptr; 190 enum g_raid3_zones zone; 191 192 if (g_raid3_use_malloc || 193 (zone = g_raid3_zone(size)) == G_RAID3_NUM_ZONES) 194 ptr = malloc(size, M_RAID3, flags); 195 else { 196 ptr = uma_zalloc_arg(sc->sc_zones[zone].sz_zone, 197 &sc->sc_zones[zone], flags); 198 sc->sc_zones[zone].sz_requested++; 199 if (ptr == NULL) 200 sc->sc_zones[zone].sz_failed++; 201 } 202 return (ptr); 203 } 204 205 static void 206 g_raid3_free(struct g_raid3_softc *sc, void *ptr, size_t size) 207 { 208 enum g_raid3_zones zone; 209 210 if (g_raid3_use_malloc || 211 (zone = g_raid3_zone(size)) == G_RAID3_NUM_ZONES) 212 free(ptr, M_RAID3); 213 else { 214 uma_zfree_arg(sc->sc_zones[zone].sz_zone, 215 ptr, &sc->sc_zones[zone]); 216 } 217 } 218 219 static int 220 g_raid3_uma_ctor(void *mem, int size, void *arg, int flags) 221 { 222 struct g_raid3_zone *sz = arg; 223 224 if (sz->sz_max > 0 && sz->sz_inuse == sz->sz_max) 225 return (ENOMEM); 226 sz->sz_inuse++; 227 return (0); 228 } 229 230 static void 231 g_raid3_uma_dtor(void *mem, int size, void *arg) 232 { 233 struct g_raid3_zone *sz = arg; 234 235 sz->sz_inuse--; 236 } 237 238 #define g_raid3_xor(src, dst, size) \ 239 _g_raid3_xor((uint64_t *)(src), \ 240 (uint64_t *)(dst), (size_t)size) 241 static void 242 _g_raid3_xor(uint64_t *src, uint64_t *dst, size_t size) 243 { 244 245 KASSERT((size % 128) == 0, ("Invalid size: %zu.", size)); 246 for (; size > 0; size -= 128) { 247 *dst++ ^= (*src++); 248 *dst++ ^= (*src++); 249 *dst++ ^= (*src++); 250 *dst++ ^= (*src++); 251 *dst++ ^= (*src++); 252 *dst++ ^= (*src++); 253 *dst++ ^= (*src++); 254 *dst++ ^= (*src++); 255 *dst++ ^= (*src++); 256 *dst++ ^= (*src++); 257 *dst++ ^= (*src++); 258 *dst++ ^= (*src++); 259 *dst++ ^= (*src++); 260 *dst++ ^= (*src++); 261 *dst++ ^= (*src++); 262 *dst++ ^= (*src++); 263 } 264 } 265 266 static int 267 g_raid3_is_zero(struct bio *bp) 268 { 269 static const uint64_t zeros[] = { 270 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 271 }; 272 u_char *addr; 273 ssize_t size; 274 275 size = bp->bio_length; 276 addr = (u_char *)bp->bio_data; 277 for (; size > 0; size -= sizeof(zeros), addr += sizeof(zeros)) { 278 if (bcmp(addr, zeros, sizeof(zeros)) != 0) 279 return (0); 280 } 281 return (1); 282 } 283 284 /* 285 * --- Events handling functions --- 286 * Events in geom_raid3 are used to maintain disks and device status 287 * from one thread to simplify locking. 288 */ 289 static void 290 g_raid3_event_free(struct g_raid3_event *ep) 291 { 292 293 free(ep, M_RAID3); 294 } 295 296 int 297 g_raid3_event_send(void *arg, int state, int flags) 298 { 299 struct g_raid3_softc *sc; 300 struct g_raid3_disk *disk; 301 struct g_raid3_event *ep; 302 int error; 303 304 ep = malloc(sizeof(*ep), M_RAID3, M_WAITOK); 305 G_RAID3_DEBUG(4, "%s: Sending event %p.", __func__, ep); 306 if ((flags & G_RAID3_EVENT_DEVICE) != 0) { 307 disk = NULL; 308 sc = arg; 309 } else { 310 disk = arg; 311 sc = disk->d_softc; 312 } 313 ep->e_disk = disk; 314 ep->e_state = state; 315 ep->e_flags = flags; 316 ep->e_error = 0; 317 mtx_lock(&sc->sc_events_mtx); 318 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 319 mtx_unlock(&sc->sc_events_mtx); 320 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc); 321 mtx_lock(&sc->sc_queue_mtx); 322 wakeup(sc); 323 wakeup(&sc->sc_queue); 324 mtx_unlock(&sc->sc_queue_mtx); 325 if ((flags & G_RAID3_EVENT_DONTWAIT) != 0) 326 return (0); 327 sx_assert(&sc->sc_lock, SX_XLOCKED); 328 G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 329 sx_xunlock(&sc->sc_lock); 330 while ((ep->e_flags & G_RAID3_EVENT_DONE) == 0) { 331 mtx_lock(&sc->sc_events_mtx); 332 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "r3:event", 333 hz * 5); 334 } 335 error = ep->e_error; 336 g_raid3_event_free(ep); 337 sx_xlock(&sc->sc_lock); 338 return (error); 339 } 340 341 static struct g_raid3_event * 342 g_raid3_event_get(struct g_raid3_softc *sc) 343 { 344 struct g_raid3_event *ep; 345 346 mtx_lock(&sc->sc_events_mtx); 347 ep = TAILQ_FIRST(&sc->sc_events); 348 mtx_unlock(&sc->sc_events_mtx); 349 return (ep); 350 } 351 352 static void 353 g_raid3_event_remove(struct g_raid3_softc *sc, struct g_raid3_event *ep) 354 { 355 356 mtx_lock(&sc->sc_events_mtx); 357 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 358 mtx_unlock(&sc->sc_events_mtx); 359 } 360 361 static void 362 g_raid3_event_cancel(struct g_raid3_disk *disk) 363 { 364 struct g_raid3_softc *sc; 365 struct g_raid3_event *ep, *tmpep; 366 367 sc = disk->d_softc; 368 sx_assert(&sc->sc_lock, SX_XLOCKED); 369 370 mtx_lock(&sc->sc_events_mtx); 371 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 372 if ((ep->e_flags & G_RAID3_EVENT_DEVICE) != 0) 373 continue; 374 if (ep->e_disk != disk) 375 continue; 376 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 377 if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0) 378 g_raid3_event_free(ep); 379 else { 380 ep->e_error = ECANCELED; 381 wakeup(ep); 382 } 383 } 384 mtx_unlock(&sc->sc_events_mtx); 385 } 386 387 /* 388 * Return the number of disks in the given state. 389 * If state is equal to -1, count all connected disks. 390 */ 391 u_int 392 g_raid3_ndisks(struct g_raid3_softc *sc, int state) 393 { 394 struct g_raid3_disk *disk; 395 u_int n, ndisks; 396 397 sx_assert(&sc->sc_lock, SX_LOCKED); 398 399 for (n = ndisks = 0; n < sc->sc_ndisks; n++) { 400 disk = &sc->sc_disks[n]; 401 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 402 continue; 403 if (state == -1 || disk->d_state == state) 404 ndisks++; 405 } 406 return (ndisks); 407 } 408 409 static u_int 410 g_raid3_nrequests(struct g_raid3_softc *sc, struct g_consumer *cp) 411 { 412 struct bio *bp; 413 u_int nreqs = 0; 414 415 mtx_lock(&sc->sc_queue_mtx); 416 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 417 if (bp->bio_from == cp) 418 nreqs++; 419 } 420 mtx_unlock(&sc->sc_queue_mtx); 421 return (nreqs); 422 } 423 424 static int 425 g_raid3_is_busy(struct g_raid3_softc *sc, struct g_consumer *cp) 426 { 427 428 if (cp->index > 0) { 429 G_RAID3_DEBUG(2, 430 "I/O requests for %s exist, can't destroy it now.", 431 cp->provider->name); 432 return (1); 433 } 434 if (g_raid3_nrequests(sc, cp) > 0) { 435 G_RAID3_DEBUG(2, 436 "I/O requests for %s in queue, can't destroy it now.", 437 cp->provider->name); 438 return (1); 439 } 440 return (0); 441 } 442 443 static void 444 g_raid3_destroy_consumer(void *arg, int flags __unused) 445 { 446 struct g_consumer *cp; 447 448 g_topology_assert(); 449 450 cp = arg; 451 G_RAID3_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 452 g_detach(cp); 453 g_destroy_consumer(cp); 454 } 455 456 static void 457 g_raid3_kill_consumer(struct g_raid3_softc *sc, struct g_consumer *cp) 458 { 459 struct g_provider *pp; 460 int retaste_wait; 461 462 g_topology_assert(); 463 464 cp->private = NULL; 465 if (g_raid3_is_busy(sc, cp)) 466 return; 467 G_RAID3_DEBUG(2, "Consumer %s destroyed.", cp->provider->name); 468 pp = cp->provider; 469 retaste_wait = 0; 470 if (cp->acw == 1) { 471 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 472 retaste_wait = 1; 473 } 474 G_RAID3_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 475 -cp->acw, -cp->ace, 0); 476 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 477 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 478 if (retaste_wait) { 479 /* 480 * After retaste event was send (inside g_access()), we can send 481 * event to detach and destroy consumer. 482 * A class, which has consumer to the given provider connected 483 * will not receive retaste event for the provider. 484 * This is the way how I ignore retaste events when I close 485 * consumers opened for write: I detach and destroy consumer 486 * after retaste event is sent. 487 */ 488 g_post_event(g_raid3_destroy_consumer, cp, M_WAITOK, NULL); 489 return; 490 } 491 G_RAID3_DEBUG(1, "Consumer %s destroyed.", pp->name); 492 g_detach(cp); 493 g_destroy_consumer(cp); 494 } 495 496 static int 497 g_raid3_connect_disk(struct g_raid3_disk *disk, struct g_provider *pp) 498 { 499 struct g_consumer *cp; 500 int error; 501 502 g_topology_assert_not(); 503 KASSERT(disk->d_consumer == NULL, 504 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 505 506 g_topology_lock(); 507 cp = g_new_consumer(disk->d_softc->sc_geom); 508 error = g_attach(cp, pp); 509 if (error != 0) { 510 g_destroy_consumer(cp); 511 g_topology_unlock(); 512 return (error); 513 } 514 error = g_access(cp, 1, 1, 1); 515 g_topology_unlock(); 516 if (error != 0) { 517 g_detach(cp); 518 g_destroy_consumer(cp); 519 G_RAID3_DEBUG(0, "Cannot open consumer %s (error=%d).", 520 pp->name, error); 521 return (error); 522 } 523 disk->d_consumer = cp; 524 disk->d_consumer->private = disk; 525 disk->d_consumer->index = 0; 526 G_RAID3_DEBUG(2, "Disk %s connected.", g_raid3_get_diskname(disk)); 527 return (0); 528 } 529 530 static void 531 g_raid3_disconnect_consumer(struct g_raid3_softc *sc, struct g_consumer *cp) 532 { 533 534 g_topology_assert(); 535 536 if (cp == NULL) 537 return; 538 if (cp->provider != NULL) 539 g_raid3_kill_consumer(sc, cp); 540 else 541 g_destroy_consumer(cp); 542 } 543 544 /* 545 * Initialize disk. This means allocate memory, create consumer, attach it 546 * to the provider and open access (r1w1e1) to it. 547 */ 548 static struct g_raid3_disk * 549 g_raid3_init_disk(struct g_raid3_softc *sc, struct g_provider *pp, 550 struct g_raid3_metadata *md, int *errorp) 551 { 552 struct g_raid3_disk *disk; 553 int error; 554 555 disk = &sc->sc_disks[md->md_no]; 556 error = g_raid3_connect_disk(disk, pp); 557 if (error != 0) { 558 if (errorp != NULL) 559 *errorp = error; 560 return (NULL); 561 } 562 disk->d_state = G_RAID3_DISK_STATE_NONE; 563 disk->d_flags = md->md_dflags; 564 if (md->md_provider[0] != '\0') 565 disk->d_flags |= G_RAID3_DISK_FLAG_HARDCODED; 566 disk->d_sync.ds_consumer = NULL; 567 disk->d_sync.ds_offset = md->md_sync_offset; 568 disk->d_sync.ds_offset_done = md->md_sync_offset; 569 disk->d_genid = md->md_genid; 570 disk->d_sync.ds_syncid = md->md_syncid; 571 if (errorp != NULL) 572 *errorp = 0; 573 return (disk); 574 } 575 576 static void 577 g_raid3_destroy_disk(struct g_raid3_disk *disk) 578 { 579 struct g_raid3_softc *sc; 580 581 g_topology_assert_not(); 582 sc = disk->d_softc; 583 sx_assert(&sc->sc_lock, SX_XLOCKED); 584 585 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 586 return; 587 g_raid3_event_cancel(disk); 588 switch (disk->d_state) { 589 case G_RAID3_DISK_STATE_SYNCHRONIZING: 590 if (sc->sc_syncdisk != NULL) 591 g_raid3_sync_stop(sc, 1); 592 /* FALLTHROUGH */ 593 case G_RAID3_DISK_STATE_NEW: 594 case G_RAID3_DISK_STATE_STALE: 595 case G_RAID3_DISK_STATE_ACTIVE: 596 g_topology_lock(); 597 g_raid3_disconnect_consumer(sc, disk->d_consumer); 598 g_topology_unlock(); 599 disk->d_consumer = NULL; 600 break; 601 default: 602 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 603 g_raid3_get_diskname(disk), 604 g_raid3_disk_state2str(disk->d_state))); 605 } 606 disk->d_state = G_RAID3_DISK_STATE_NODISK; 607 } 608 609 static void 610 g_raid3_destroy_device(struct g_raid3_softc *sc) 611 { 612 struct g_raid3_event *ep; 613 struct g_raid3_disk *disk; 614 struct g_geom *gp; 615 struct g_consumer *cp; 616 u_int n; 617 618 g_topology_assert_not(); 619 sx_assert(&sc->sc_lock, SX_XLOCKED); 620 621 gp = sc->sc_geom; 622 if (sc->sc_provider != NULL) 623 g_raid3_destroy_provider(sc); 624 for (n = 0; n < sc->sc_ndisks; n++) { 625 disk = &sc->sc_disks[n]; 626 if (disk->d_state != G_RAID3_DISK_STATE_NODISK) { 627 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 628 g_raid3_update_metadata(disk); 629 g_raid3_destroy_disk(disk); 630 } 631 } 632 while ((ep = g_raid3_event_get(sc)) != NULL) { 633 g_raid3_event_remove(sc, ep); 634 if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0) 635 g_raid3_event_free(ep); 636 else { 637 ep->e_error = ECANCELED; 638 ep->e_flags |= G_RAID3_EVENT_DONE; 639 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, ep); 640 mtx_lock(&sc->sc_events_mtx); 641 wakeup(ep); 642 mtx_unlock(&sc->sc_events_mtx); 643 } 644 } 645 callout_drain(&sc->sc_callout); 646 cp = LIST_FIRST(&sc->sc_sync.ds_geom->consumer); 647 g_topology_lock(); 648 if (cp != NULL) 649 g_raid3_disconnect_consumer(sc, cp); 650 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 651 G_RAID3_DEBUG(0, "Device %s destroyed.", gp->name); 652 g_wither_geom(gp, ENXIO); 653 g_topology_unlock(); 654 if (!g_raid3_use_malloc) { 655 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_64K].sz_zone); 656 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_16K].sz_zone); 657 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_4K].sz_zone); 658 } 659 mtx_destroy(&sc->sc_queue_mtx); 660 mtx_destroy(&sc->sc_events_mtx); 661 sx_xunlock(&sc->sc_lock); 662 sx_destroy(&sc->sc_lock); 663 } 664 665 static void 666 g_raid3_orphan(struct g_consumer *cp) 667 { 668 struct g_raid3_disk *disk; 669 670 g_topology_assert(); 671 672 disk = cp->private; 673 if (disk == NULL) 674 return; 675 disk->d_softc->sc_bump_id = G_RAID3_BUMP_SYNCID; 676 g_raid3_event_send(disk, G_RAID3_DISK_STATE_DISCONNECTED, 677 G_RAID3_EVENT_DONTWAIT); 678 } 679 680 static int 681 g_raid3_write_metadata(struct g_raid3_disk *disk, struct g_raid3_metadata *md) 682 { 683 struct g_raid3_softc *sc; 684 struct g_consumer *cp; 685 off_t offset, length; 686 u_char *sector; 687 int error = 0; 688 689 g_topology_assert_not(); 690 sc = disk->d_softc; 691 sx_assert(&sc->sc_lock, SX_LOCKED); 692 693 cp = disk->d_consumer; 694 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 695 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 696 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 697 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 698 cp->acw, cp->ace)); 699 length = cp->provider->sectorsize; 700 offset = cp->provider->mediasize - length; 701 sector = malloc((size_t)length, M_RAID3, M_WAITOK | M_ZERO); 702 if (md != NULL) 703 raid3_metadata_encode(md, sector); 704 error = g_write_data(cp, offset, sector, length); 705 free(sector, M_RAID3); 706 if (error != 0) { 707 if ((disk->d_flags & G_RAID3_DISK_FLAG_BROKEN) == 0) { 708 G_RAID3_DEBUG(0, "Cannot write metadata on %s " 709 "(device=%s, error=%d).", 710 g_raid3_get_diskname(disk), sc->sc_name, error); 711 disk->d_flags |= G_RAID3_DISK_FLAG_BROKEN; 712 } else { 713 G_RAID3_DEBUG(1, "Cannot write metadata on %s " 714 "(device=%s, error=%d).", 715 g_raid3_get_diskname(disk), sc->sc_name, error); 716 } 717 if (g_raid3_disconnect_on_failure && 718 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 719 sc->sc_bump_id |= G_RAID3_BUMP_GENID; 720 g_raid3_event_send(disk, 721 G_RAID3_DISK_STATE_DISCONNECTED, 722 G_RAID3_EVENT_DONTWAIT); 723 } 724 } 725 return (error); 726 } 727 728 int 729 g_raid3_clear_metadata(struct g_raid3_disk *disk) 730 { 731 int error; 732 733 g_topology_assert_not(); 734 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 735 736 error = g_raid3_write_metadata(disk, NULL); 737 if (error == 0) { 738 G_RAID3_DEBUG(2, "Metadata on %s cleared.", 739 g_raid3_get_diskname(disk)); 740 } else { 741 G_RAID3_DEBUG(0, 742 "Cannot clear metadata on disk %s (error=%d).", 743 g_raid3_get_diskname(disk), error); 744 } 745 return (error); 746 } 747 748 void 749 g_raid3_fill_metadata(struct g_raid3_disk *disk, struct g_raid3_metadata *md) 750 { 751 struct g_raid3_softc *sc; 752 struct g_provider *pp; 753 754 sc = disk->d_softc; 755 strlcpy(md->md_magic, G_RAID3_MAGIC, sizeof(md->md_magic)); 756 md->md_version = G_RAID3_VERSION; 757 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 758 md->md_id = sc->sc_id; 759 md->md_all = sc->sc_ndisks; 760 md->md_genid = sc->sc_genid; 761 md->md_mediasize = sc->sc_mediasize; 762 md->md_sectorsize = sc->sc_sectorsize; 763 md->md_mflags = (sc->sc_flags & G_RAID3_DEVICE_FLAG_MASK); 764 md->md_no = disk->d_no; 765 md->md_syncid = disk->d_sync.ds_syncid; 766 md->md_dflags = (disk->d_flags & G_RAID3_DISK_FLAG_MASK); 767 if (disk->d_state != G_RAID3_DISK_STATE_SYNCHRONIZING) 768 md->md_sync_offset = 0; 769 else { 770 md->md_sync_offset = 771 disk->d_sync.ds_offset_done / (sc->sc_ndisks - 1); 772 } 773 if (disk->d_consumer != NULL && disk->d_consumer->provider != NULL) 774 pp = disk->d_consumer->provider; 775 else 776 pp = NULL; 777 if ((disk->d_flags & G_RAID3_DISK_FLAG_HARDCODED) != 0 && pp != NULL) 778 strlcpy(md->md_provider, pp->name, sizeof(md->md_provider)); 779 else 780 bzero(md->md_provider, sizeof(md->md_provider)); 781 if (pp != NULL) 782 md->md_provsize = pp->mediasize; 783 else 784 md->md_provsize = 0; 785 } 786 787 void 788 g_raid3_update_metadata(struct g_raid3_disk *disk) 789 { 790 struct g_raid3_softc *sc; 791 struct g_raid3_metadata md; 792 int error; 793 794 g_topology_assert_not(); 795 sc = disk->d_softc; 796 sx_assert(&sc->sc_lock, SX_LOCKED); 797 798 g_raid3_fill_metadata(disk, &md); 799 error = g_raid3_write_metadata(disk, &md); 800 if (error == 0) { 801 G_RAID3_DEBUG(2, "Metadata on %s updated.", 802 g_raid3_get_diskname(disk)); 803 } else { 804 G_RAID3_DEBUG(0, 805 "Cannot update metadata on disk %s (error=%d).", 806 g_raid3_get_diskname(disk), error); 807 } 808 } 809 810 static void 811 g_raid3_bump_syncid(struct g_raid3_softc *sc) 812 { 813 struct g_raid3_disk *disk; 814 u_int n; 815 816 g_topology_assert_not(); 817 sx_assert(&sc->sc_lock, SX_XLOCKED); 818 KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) > 0, 819 ("%s called with no active disks (device=%s).", __func__, 820 sc->sc_name)); 821 822 sc->sc_syncid++; 823 G_RAID3_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 824 sc->sc_syncid); 825 for (n = 0; n < sc->sc_ndisks; n++) { 826 disk = &sc->sc_disks[n]; 827 if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 828 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 829 disk->d_sync.ds_syncid = sc->sc_syncid; 830 g_raid3_update_metadata(disk); 831 } 832 } 833 } 834 835 static void 836 g_raid3_bump_genid(struct g_raid3_softc *sc) 837 { 838 struct g_raid3_disk *disk; 839 u_int n; 840 841 g_topology_assert_not(); 842 sx_assert(&sc->sc_lock, SX_XLOCKED); 843 KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) > 0, 844 ("%s called with no active disks (device=%s).", __func__, 845 sc->sc_name)); 846 847 sc->sc_genid++; 848 G_RAID3_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 849 sc->sc_genid); 850 for (n = 0; n < sc->sc_ndisks; n++) { 851 disk = &sc->sc_disks[n]; 852 if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 853 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 854 disk->d_genid = sc->sc_genid; 855 g_raid3_update_metadata(disk); 856 } 857 } 858 } 859 860 static int 861 g_raid3_idle(struct g_raid3_softc *sc, int acw) 862 { 863 struct g_raid3_disk *disk; 864 u_int i; 865 int timeout; 866 867 g_topology_assert_not(); 868 sx_assert(&sc->sc_lock, SX_XLOCKED); 869 870 if (sc->sc_provider == NULL) 871 return (0); 872 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) != 0) 873 return (0); 874 if (sc->sc_idle) 875 return (0); 876 if (sc->sc_writes > 0) 877 return (0); 878 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 879 timeout = g_raid3_idletime - (time_uptime - sc->sc_last_write); 880 if (!g_raid3_shutdown && timeout > 0) 881 return (timeout); 882 } 883 sc->sc_idle = 1; 884 for (i = 0; i < sc->sc_ndisks; i++) { 885 disk = &sc->sc_disks[i]; 886 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) 887 continue; 888 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as clean.", 889 g_raid3_get_diskname(disk), sc->sc_name); 890 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 891 g_raid3_update_metadata(disk); 892 } 893 return (0); 894 } 895 896 static void 897 g_raid3_unidle(struct g_raid3_softc *sc) 898 { 899 struct g_raid3_disk *disk; 900 u_int i; 901 902 g_topology_assert_not(); 903 sx_assert(&sc->sc_lock, SX_XLOCKED); 904 905 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) != 0) 906 return; 907 sc->sc_idle = 0; 908 sc->sc_last_write = time_uptime; 909 for (i = 0; i < sc->sc_ndisks; i++) { 910 disk = &sc->sc_disks[i]; 911 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) 912 continue; 913 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as dirty.", 914 g_raid3_get_diskname(disk), sc->sc_name); 915 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY; 916 g_raid3_update_metadata(disk); 917 } 918 } 919 920 /* 921 * Treat bio_driver1 field in parent bio as list head and field bio_caller1 922 * in child bio as pointer to the next element on the list. 923 */ 924 #define G_RAID3_HEAD_BIO(pbp) (pbp)->bio_driver1 925 926 #define G_RAID3_NEXT_BIO(cbp) (cbp)->bio_caller1 927 928 #define G_RAID3_FOREACH_BIO(pbp, bp) \ 929 for ((bp) = G_RAID3_HEAD_BIO(pbp); (bp) != NULL; \ 930 (bp) = G_RAID3_NEXT_BIO(bp)) 931 932 #define G_RAID3_FOREACH_SAFE_BIO(pbp, bp, tmpbp) \ 933 for ((bp) = G_RAID3_HEAD_BIO(pbp); \ 934 (bp) != NULL && ((tmpbp) = G_RAID3_NEXT_BIO(bp), 1); \ 935 (bp) = (tmpbp)) 936 937 static void 938 g_raid3_init_bio(struct bio *pbp) 939 { 940 941 G_RAID3_HEAD_BIO(pbp) = NULL; 942 } 943 944 static void 945 g_raid3_remove_bio(struct bio *cbp) 946 { 947 struct bio *pbp, *bp; 948 949 pbp = cbp->bio_parent; 950 if (G_RAID3_HEAD_BIO(pbp) == cbp) 951 G_RAID3_HEAD_BIO(pbp) = G_RAID3_NEXT_BIO(cbp); 952 else { 953 G_RAID3_FOREACH_BIO(pbp, bp) { 954 if (G_RAID3_NEXT_BIO(bp) == cbp) { 955 G_RAID3_NEXT_BIO(bp) = G_RAID3_NEXT_BIO(cbp); 956 break; 957 } 958 } 959 } 960 G_RAID3_NEXT_BIO(cbp) = NULL; 961 } 962 963 static void 964 g_raid3_replace_bio(struct bio *sbp, struct bio *dbp) 965 { 966 struct bio *pbp, *bp; 967 968 g_raid3_remove_bio(sbp); 969 pbp = dbp->bio_parent; 970 G_RAID3_NEXT_BIO(sbp) = G_RAID3_NEXT_BIO(dbp); 971 if (G_RAID3_HEAD_BIO(pbp) == dbp) 972 G_RAID3_HEAD_BIO(pbp) = sbp; 973 else { 974 G_RAID3_FOREACH_BIO(pbp, bp) { 975 if (G_RAID3_NEXT_BIO(bp) == dbp) { 976 G_RAID3_NEXT_BIO(bp) = sbp; 977 break; 978 } 979 } 980 } 981 G_RAID3_NEXT_BIO(dbp) = NULL; 982 } 983 984 static void 985 g_raid3_destroy_bio(struct g_raid3_softc *sc, struct bio *cbp) 986 { 987 struct bio *bp, *pbp; 988 size_t size; 989 990 pbp = cbp->bio_parent; 991 pbp->bio_children--; 992 KASSERT(cbp->bio_data != NULL, ("NULL bio_data")); 993 size = pbp->bio_length / (sc->sc_ndisks - 1); 994 g_raid3_free(sc, cbp->bio_data, size); 995 if (G_RAID3_HEAD_BIO(pbp) == cbp) { 996 G_RAID3_HEAD_BIO(pbp) = G_RAID3_NEXT_BIO(cbp); 997 G_RAID3_NEXT_BIO(cbp) = NULL; 998 g_destroy_bio(cbp); 999 } else { 1000 G_RAID3_FOREACH_BIO(pbp, bp) { 1001 if (G_RAID3_NEXT_BIO(bp) == cbp) 1002 break; 1003 } 1004 if (bp != NULL) { 1005 KASSERT(G_RAID3_NEXT_BIO(bp) != NULL, 1006 ("NULL bp->bio_driver1")); 1007 G_RAID3_NEXT_BIO(bp) = G_RAID3_NEXT_BIO(cbp); 1008 G_RAID3_NEXT_BIO(cbp) = NULL; 1009 } 1010 g_destroy_bio(cbp); 1011 } 1012 } 1013 1014 static struct bio * 1015 g_raid3_clone_bio(struct g_raid3_softc *sc, struct bio *pbp) 1016 { 1017 struct bio *bp, *cbp; 1018 size_t size; 1019 int memflag; 1020 1021 cbp = g_clone_bio(pbp); 1022 if (cbp == NULL) 1023 return (NULL); 1024 size = pbp->bio_length / (sc->sc_ndisks - 1); 1025 if ((pbp->bio_cflags & G_RAID3_BIO_CFLAG_REGULAR) != 0) 1026 memflag = M_WAITOK; 1027 else 1028 memflag = M_NOWAIT; 1029 cbp->bio_data = g_raid3_alloc(sc, size, memflag); 1030 if (cbp->bio_data == NULL) { 1031 pbp->bio_children--; 1032 g_destroy_bio(cbp); 1033 return (NULL); 1034 } 1035 G_RAID3_NEXT_BIO(cbp) = NULL; 1036 if (G_RAID3_HEAD_BIO(pbp) == NULL) 1037 G_RAID3_HEAD_BIO(pbp) = cbp; 1038 else { 1039 G_RAID3_FOREACH_BIO(pbp, bp) { 1040 if (G_RAID3_NEXT_BIO(bp) == NULL) { 1041 G_RAID3_NEXT_BIO(bp) = cbp; 1042 break; 1043 } 1044 } 1045 } 1046 return (cbp); 1047 } 1048 1049 static void 1050 g_raid3_scatter(struct bio *pbp) 1051 { 1052 struct g_raid3_softc *sc; 1053 struct g_raid3_disk *disk; 1054 struct bio *bp, *cbp, *tmpbp; 1055 off_t atom, cadd, padd, left; 1056 int first; 1057 1058 sc = pbp->bio_to->geom->softc; 1059 bp = NULL; 1060 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_NOPARITY) == 0) { 1061 /* 1062 * Find bio for which we should calculate data. 1063 */ 1064 G_RAID3_FOREACH_BIO(pbp, cbp) { 1065 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) { 1066 bp = cbp; 1067 break; 1068 } 1069 } 1070 KASSERT(bp != NULL, ("NULL parity bio.")); 1071 } 1072 atom = sc->sc_sectorsize / (sc->sc_ndisks - 1); 1073 cadd = padd = 0; 1074 for (left = pbp->bio_length; left > 0; left -= sc->sc_sectorsize) { 1075 G_RAID3_FOREACH_BIO(pbp, cbp) { 1076 if (cbp == bp) 1077 continue; 1078 bcopy(pbp->bio_data + padd, cbp->bio_data + cadd, atom); 1079 padd += atom; 1080 } 1081 cadd += atom; 1082 } 1083 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_NOPARITY) == 0) { 1084 /* 1085 * Calculate parity. 1086 */ 1087 first = 1; 1088 G_RAID3_FOREACH_SAFE_BIO(pbp, cbp, tmpbp) { 1089 if (cbp == bp) 1090 continue; 1091 if (first) { 1092 bcopy(cbp->bio_data, bp->bio_data, 1093 bp->bio_length); 1094 first = 0; 1095 } else { 1096 g_raid3_xor(cbp->bio_data, bp->bio_data, 1097 bp->bio_length); 1098 } 1099 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_NODISK) != 0) 1100 g_raid3_destroy_bio(sc, cbp); 1101 } 1102 } 1103 G_RAID3_FOREACH_SAFE_BIO(pbp, cbp, tmpbp) { 1104 struct g_consumer *cp; 1105 1106 disk = cbp->bio_caller2; 1107 cp = disk->d_consumer; 1108 cbp->bio_to = cp->provider; 1109 G_RAID3_LOGREQ(3, cbp, "Sending request."); 1110 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1111 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1112 cp->acr, cp->acw, cp->ace)); 1113 cp->index++; 1114 sc->sc_writes++; 1115 g_io_request(cbp, cp); 1116 } 1117 } 1118 1119 static void 1120 g_raid3_gather(struct bio *pbp) 1121 { 1122 struct g_raid3_softc *sc; 1123 struct g_raid3_disk *disk; 1124 struct bio *xbp, *fbp, *cbp; 1125 off_t atom, cadd, padd, left; 1126 1127 sc = pbp->bio_to->geom->softc; 1128 /* 1129 * Find bio for which we have to calculate data. 1130 * While going through this path, check if all requests 1131 * succeeded, if not, deny whole request. 1132 * If we're in COMPLETE mode, we allow one request to fail, 1133 * so if we find one, we're sending it to the parity consumer. 1134 * If there are more failed requests, we deny whole request. 1135 */ 1136 xbp = fbp = NULL; 1137 G_RAID3_FOREACH_BIO(pbp, cbp) { 1138 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) { 1139 KASSERT(xbp == NULL, ("More than one parity bio.")); 1140 xbp = cbp; 1141 } 1142 if (cbp->bio_error == 0) 1143 continue; 1144 /* 1145 * Found failed request. 1146 */ 1147 if (fbp == NULL) { 1148 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_DEGRADED) != 0) { 1149 /* 1150 * We are already in degraded mode, so we can't 1151 * accept any failures. 1152 */ 1153 if (pbp->bio_error == 0) 1154 pbp->bio_error = cbp->bio_error; 1155 } else { 1156 fbp = cbp; 1157 } 1158 } else { 1159 /* 1160 * Next failed request, that's too many. 1161 */ 1162 if (pbp->bio_error == 0) 1163 pbp->bio_error = fbp->bio_error; 1164 } 1165 disk = cbp->bio_caller2; 1166 if (disk == NULL) 1167 continue; 1168 if ((disk->d_flags & G_RAID3_DISK_FLAG_BROKEN) == 0) { 1169 disk->d_flags |= G_RAID3_DISK_FLAG_BROKEN; 1170 G_RAID3_LOGREQ(0, cbp, "Request failed (error=%d).", 1171 cbp->bio_error); 1172 } else { 1173 G_RAID3_LOGREQ(1, cbp, "Request failed (error=%d).", 1174 cbp->bio_error); 1175 } 1176 if (g_raid3_disconnect_on_failure && 1177 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 1178 sc->sc_bump_id |= G_RAID3_BUMP_GENID; 1179 g_raid3_event_send(disk, 1180 G_RAID3_DISK_STATE_DISCONNECTED, 1181 G_RAID3_EVENT_DONTWAIT); 1182 } 1183 } 1184 if (pbp->bio_error != 0) 1185 goto finish; 1186 if (fbp != NULL && (pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) { 1187 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_VERIFY; 1188 if (xbp != fbp) 1189 g_raid3_replace_bio(xbp, fbp); 1190 g_raid3_destroy_bio(sc, fbp); 1191 } else if (fbp != NULL) { 1192 struct g_consumer *cp; 1193 1194 /* 1195 * One request failed, so send the same request to 1196 * the parity consumer. 1197 */ 1198 disk = pbp->bio_driver2; 1199 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) { 1200 pbp->bio_error = fbp->bio_error; 1201 goto finish; 1202 } 1203 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED; 1204 pbp->bio_inbed--; 1205 fbp->bio_flags &= ~(BIO_DONE | BIO_ERROR); 1206 if (disk->d_no == sc->sc_ndisks - 1) 1207 fbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1208 fbp->bio_error = 0; 1209 fbp->bio_completed = 0; 1210 fbp->bio_children = 0; 1211 fbp->bio_inbed = 0; 1212 cp = disk->d_consumer; 1213 fbp->bio_caller2 = disk; 1214 fbp->bio_to = cp->provider; 1215 G_RAID3_LOGREQ(3, fbp, "Sending request (recover)."); 1216 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1217 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1218 cp->acr, cp->acw, cp->ace)); 1219 cp->index++; 1220 g_io_request(fbp, cp); 1221 return; 1222 } 1223 if (xbp != NULL) { 1224 /* 1225 * Calculate parity. 1226 */ 1227 G_RAID3_FOREACH_BIO(pbp, cbp) { 1228 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) 1229 continue; 1230 g_raid3_xor(cbp->bio_data, xbp->bio_data, 1231 xbp->bio_length); 1232 } 1233 xbp->bio_cflags &= ~G_RAID3_BIO_CFLAG_PARITY; 1234 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) { 1235 if (!g_raid3_is_zero(xbp)) { 1236 g_raid3_parity_mismatch++; 1237 pbp->bio_error = EIO; 1238 goto finish; 1239 } 1240 g_raid3_destroy_bio(sc, xbp); 1241 } 1242 } 1243 atom = sc->sc_sectorsize / (sc->sc_ndisks - 1); 1244 cadd = padd = 0; 1245 for (left = pbp->bio_length; left > 0; left -= sc->sc_sectorsize) { 1246 G_RAID3_FOREACH_BIO(pbp, cbp) { 1247 bcopy(cbp->bio_data + cadd, pbp->bio_data + padd, atom); 1248 pbp->bio_completed += atom; 1249 padd += atom; 1250 } 1251 cadd += atom; 1252 } 1253 finish: 1254 if (pbp->bio_error == 0) 1255 G_RAID3_LOGREQ(3, pbp, "Request finished."); 1256 else { 1257 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) 1258 G_RAID3_LOGREQ(1, pbp, "Verification error."); 1259 else 1260 G_RAID3_LOGREQ(0, pbp, "Request failed."); 1261 } 1262 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_MASK; 1263 while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL) 1264 g_raid3_destroy_bio(sc, cbp); 1265 g_io_deliver(pbp, pbp->bio_error); 1266 } 1267 1268 static void 1269 g_raid3_done(struct bio *bp) 1270 { 1271 struct g_raid3_softc *sc; 1272 1273 sc = bp->bio_from->geom->softc; 1274 bp->bio_cflags |= G_RAID3_BIO_CFLAG_REGULAR; 1275 G_RAID3_LOGREQ(3, bp, "Regular request done (error=%d).", bp->bio_error); 1276 mtx_lock(&sc->sc_queue_mtx); 1277 bioq_insert_head(&sc->sc_queue, bp); 1278 mtx_unlock(&sc->sc_queue_mtx); 1279 wakeup(sc); 1280 wakeup(&sc->sc_queue); 1281 } 1282 1283 static void 1284 g_raid3_regular_request(struct bio *cbp) 1285 { 1286 struct g_raid3_softc *sc; 1287 struct g_raid3_disk *disk; 1288 struct bio *pbp; 1289 1290 g_topology_assert_not(); 1291 1292 pbp = cbp->bio_parent; 1293 sc = pbp->bio_to->geom->softc; 1294 cbp->bio_from->index--; 1295 if (cbp->bio_cmd == BIO_WRITE) 1296 sc->sc_writes--; 1297 disk = cbp->bio_from->private; 1298 if (disk == NULL) { 1299 g_topology_lock(); 1300 g_raid3_kill_consumer(sc, cbp->bio_from); 1301 g_topology_unlock(); 1302 } 1303 1304 G_RAID3_LOGREQ(3, cbp, "Request finished."); 1305 pbp->bio_inbed++; 1306 KASSERT(pbp->bio_inbed <= pbp->bio_children, 1307 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 1308 pbp->bio_children)); 1309 if (pbp->bio_inbed != pbp->bio_children) 1310 return; 1311 switch (pbp->bio_cmd) { 1312 case BIO_READ: 1313 g_raid3_gather(pbp); 1314 break; 1315 case BIO_WRITE: 1316 case BIO_DELETE: 1317 { 1318 int error = 0; 1319 1320 pbp->bio_completed = pbp->bio_length; 1321 while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL) { 1322 if (cbp->bio_error == 0) { 1323 g_raid3_destroy_bio(sc, cbp); 1324 continue; 1325 } 1326 1327 if (error == 0) 1328 error = cbp->bio_error; 1329 else if (pbp->bio_error == 0) { 1330 /* 1331 * Next failed request, that's too many. 1332 */ 1333 pbp->bio_error = error; 1334 } 1335 1336 disk = cbp->bio_caller2; 1337 if (disk == NULL) { 1338 g_raid3_destroy_bio(sc, cbp); 1339 continue; 1340 } 1341 1342 if ((disk->d_flags & G_RAID3_DISK_FLAG_BROKEN) == 0) { 1343 disk->d_flags |= G_RAID3_DISK_FLAG_BROKEN; 1344 G_RAID3_LOGREQ(0, cbp, 1345 "Request failed (error=%d).", 1346 cbp->bio_error); 1347 } else { 1348 G_RAID3_LOGREQ(1, cbp, 1349 "Request failed (error=%d).", 1350 cbp->bio_error); 1351 } 1352 if (g_raid3_disconnect_on_failure && 1353 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 1354 sc->sc_bump_id |= G_RAID3_BUMP_GENID; 1355 g_raid3_event_send(disk, 1356 G_RAID3_DISK_STATE_DISCONNECTED, 1357 G_RAID3_EVENT_DONTWAIT); 1358 } 1359 g_raid3_destroy_bio(sc, cbp); 1360 } 1361 if (pbp->bio_error == 0) 1362 G_RAID3_LOGREQ(3, pbp, "Request finished."); 1363 else 1364 G_RAID3_LOGREQ(0, pbp, "Request failed."); 1365 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_DEGRADED; 1366 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_NOPARITY; 1367 bioq_remove(&sc->sc_inflight, pbp); 1368 /* Release delayed sync requests if possible. */ 1369 g_raid3_sync_release(sc); 1370 g_io_deliver(pbp, pbp->bio_error); 1371 break; 1372 } 1373 } 1374 } 1375 1376 static void 1377 g_raid3_sync_done(struct bio *bp) 1378 { 1379 struct g_raid3_softc *sc; 1380 1381 G_RAID3_LOGREQ(3, bp, "Synchronization request delivered."); 1382 sc = bp->bio_from->geom->softc; 1383 bp->bio_cflags |= G_RAID3_BIO_CFLAG_SYNC; 1384 mtx_lock(&sc->sc_queue_mtx); 1385 bioq_insert_head(&sc->sc_queue, bp); 1386 mtx_unlock(&sc->sc_queue_mtx); 1387 wakeup(sc); 1388 wakeup(&sc->sc_queue); 1389 } 1390 1391 static void 1392 g_raid3_flush(struct g_raid3_softc *sc, struct bio *bp) 1393 { 1394 struct bio_queue_head queue; 1395 struct g_raid3_disk *disk; 1396 struct g_consumer *cp; 1397 struct bio *cbp; 1398 u_int i; 1399 1400 bioq_init(&queue); 1401 for (i = 0; i < sc->sc_ndisks; i++) { 1402 disk = &sc->sc_disks[i]; 1403 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) 1404 continue; 1405 cbp = g_clone_bio(bp); 1406 if (cbp == NULL) { 1407 for (cbp = bioq_first(&queue); cbp != NULL; 1408 cbp = bioq_first(&queue)) { 1409 bioq_remove(&queue, cbp); 1410 g_destroy_bio(cbp); 1411 } 1412 if (bp->bio_error == 0) 1413 bp->bio_error = ENOMEM; 1414 g_io_deliver(bp, bp->bio_error); 1415 return; 1416 } 1417 bioq_insert_tail(&queue, cbp); 1418 cbp->bio_done = g_std_done; 1419 cbp->bio_caller1 = disk; 1420 cbp->bio_to = disk->d_consumer->provider; 1421 } 1422 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1423 bioq_remove(&queue, cbp); 1424 G_RAID3_LOGREQ(3, cbp, "Sending request."); 1425 disk = cbp->bio_caller1; 1426 cbp->bio_caller1 = NULL; 1427 cp = disk->d_consumer; 1428 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1429 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1430 cp->acr, cp->acw, cp->ace)); 1431 g_io_request(cbp, disk->d_consumer); 1432 } 1433 } 1434 1435 static void 1436 g_raid3_start(struct bio *bp) 1437 { 1438 struct g_raid3_softc *sc; 1439 1440 sc = bp->bio_to->geom->softc; 1441 /* 1442 * If sc == NULL or there are no valid disks, provider's error 1443 * should be set and g_raid3_start() should not be called at all. 1444 */ 1445 KASSERT(sc != NULL && (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 1446 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE), 1447 ("Provider's error should be set (error=%d)(device=%s).", 1448 bp->bio_to->error, bp->bio_to->name)); 1449 G_RAID3_LOGREQ(3, bp, "Request received."); 1450 1451 switch (bp->bio_cmd) { 1452 case BIO_READ: 1453 case BIO_WRITE: 1454 case BIO_DELETE: 1455 break; 1456 case BIO_FLUSH: 1457 g_raid3_flush(sc, bp); 1458 return; 1459 case BIO_GETATTR: 1460 default: 1461 g_io_deliver(bp, EOPNOTSUPP); 1462 return; 1463 } 1464 mtx_lock(&sc->sc_queue_mtx); 1465 bioq_insert_tail(&sc->sc_queue, bp); 1466 mtx_unlock(&sc->sc_queue_mtx); 1467 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1468 wakeup(sc); 1469 } 1470 1471 /* 1472 * Return TRUE if the given request is colliding with a in-progress 1473 * synchronization request. 1474 */ 1475 static int 1476 g_raid3_sync_collision(struct g_raid3_softc *sc, struct bio *bp) 1477 { 1478 struct g_raid3_disk *disk; 1479 struct bio *sbp; 1480 off_t rstart, rend, sstart, send; 1481 int i; 1482 1483 disk = sc->sc_syncdisk; 1484 if (disk == NULL) 1485 return (0); 1486 rstart = bp->bio_offset; 1487 rend = bp->bio_offset + bp->bio_length; 1488 for (i = 0; i < g_raid3_syncreqs; i++) { 1489 sbp = disk->d_sync.ds_bios[i]; 1490 if (sbp == NULL) 1491 continue; 1492 sstart = sbp->bio_offset; 1493 send = sbp->bio_length; 1494 if (sbp->bio_cmd == BIO_WRITE) { 1495 sstart *= sc->sc_ndisks - 1; 1496 send *= sc->sc_ndisks - 1; 1497 } 1498 send += sstart; 1499 if (rend > sstart && rstart < send) 1500 return (1); 1501 } 1502 return (0); 1503 } 1504 1505 /* 1506 * Return TRUE if the given sync request is colliding with a in-progress regular 1507 * request. 1508 */ 1509 static int 1510 g_raid3_regular_collision(struct g_raid3_softc *sc, struct bio *sbp) 1511 { 1512 off_t rstart, rend, sstart, send; 1513 struct bio *bp; 1514 1515 if (sc->sc_syncdisk == NULL) 1516 return (0); 1517 sstart = sbp->bio_offset; 1518 send = sstart + sbp->bio_length; 1519 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { 1520 rstart = bp->bio_offset; 1521 rend = bp->bio_offset + bp->bio_length; 1522 if (rend > sstart && rstart < send) 1523 return (1); 1524 } 1525 return (0); 1526 } 1527 1528 /* 1529 * Puts request onto delayed queue. 1530 */ 1531 static void 1532 g_raid3_regular_delay(struct g_raid3_softc *sc, struct bio *bp) 1533 { 1534 1535 G_RAID3_LOGREQ(2, bp, "Delaying request."); 1536 bioq_insert_head(&sc->sc_regular_delayed, bp); 1537 } 1538 1539 /* 1540 * Puts synchronization request onto delayed queue. 1541 */ 1542 static void 1543 g_raid3_sync_delay(struct g_raid3_softc *sc, struct bio *bp) 1544 { 1545 1546 G_RAID3_LOGREQ(2, bp, "Delaying synchronization request."); 1547 bioq_insert_tail(&sc->sc_sync_delayed, bp); 1548 } 1549 1550 /* 1551 * Releases delayed regular requests which don't collide anymore with sync 1552 * requests. 1553 */ 1554 static void 1555 g_raid3_regular_release(struct g_raid3_softc *sc) 1556 { 1557 struct bio *bp, *bp2; 1558 1559 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { 1560 if (g_raid3_sync_collision(sc, bp)) 1561 continue; 1562 bioq_remove(&sc->sc_regular_delayed, bp); 1563 G_RAID3_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1564 mtx_lock(&sc->sc_queue_mtx); 1565 bioq_insert_head(&sc->sc_queue, bp); 1566 #if 0 1567 /* 1568 * wakeup() is not needed, because this function is called from 1569 * the worker thread. 1570 */ 1571 wakeup(&sc->sc_queue); 1572 #endif 1573 mtx_unlock(&sc->sc_queue_mtx); 1574 } 1575 } 1576 1577 /* 1578 * Releases delayed sync requests which don't collide anymore with regular 1579 * requests. 1580 */ 1581 static void 1582 g_raid3_sync_release(struct g_raid3_softc *sc) 1583 { 1584 struct bio *bp, *bp2; 1585 1586 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { 1587 if (g_raid3_regular_collision(sc, bp)) 1588 continue; 1589 bioq_remove(&sc->sc_sync_delayed, bp); 1590 G_RAID3_LOGREQ(2, bp, 1591 "Releasing delayed synchronization request."); 1592 g_io_request(bp, bp->bio_from); 1593 } 1594 } 1595 1596 /* 1597 * Handle synchronization requests. 1598 * Every synchronization request is two-steps process: first, READ request is 1599 * send to active provider and then WRITE request (with read data) to the provider 1600 * beeing synchronized. When WRITE is finished, new synchronization request is 1601 * send. 1602 */ 1603 static void 1604 g_raid3_sync_request(struct bio *bp) 1605 { 1606 struct g_raid3_softc *sc; 1607 struct g_raid3_disk *disk; 1608 1609 bp->bio_from->index--; 1610 sc = bp->bio_from->geom->softc; 1611 disk = bp->bio_from->private; 1612 if (disk == NULL) { 1613 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1614 g_topology_lock(); 1615 g_raid3_kill_consumer(sc, bp->bio_from); 1616 g_topology_unlock(); 1617 free(bp->bio_data, M_RAID3); 1618 g_destroy_bio(bp); 1619 sx_xlock(&sc->sc_lock); 1620 return; 1621 } 1622 1623 /* 1624 * Synchronization request. 1625 */ 1626 switch (bp->bio_cmd) { 1627 case BIO_READ: 1628 { 1629 struct g_consumer *cp; 1630 u_char *dst, *src; 1631 off_t left; 1632 u_int atom; 1633 1634 if (bp->bio_error != 0) { 1635 G_RAID3_LOGREQ(0, bp, 1636 "Synchronization request failed (error=%d).", 1637 bp->bio_error); 1638 g_destroy_bio(bp); 1639 return; 1640 } 1641 G_RAID3_LOGREQ(3, bp, "Synchronization request finished."); 1642 atom = sc->sc_sectorsize / (sc->sc_ndisks - 1); 1643 dst = src = bp->bio_data; 1644 if (disk->d_no == sc->sc_ndisks - 1) { 1645 u_int n; 1646 1647 /* Parity component. */ 1648 for (left = bp->bio_length; left > 0; 1649 left -= sc->sc_sectorsize) { 1650 bcopy(src, dst, atom); 1651 src += atom; 1652 for (n = 1; n < sc->sc_ndisks - 1; n++) { 1653 g_raid3_xor(src, dst, atom); 1654 src += atom; 1655 } 1656 dst += atom; 1657 } 1658 } else { 1659 /* Regular component. */ 1660 src += atom * disk->d_no; 1661 for (left = bp->bio_length; left > 0; 1662 left -= sc->sc_sectorsize) { 1663 bcopy(src, dst, atom); 1664 src += sc->sc_sectorsize; 1665 dst += atom; 1666 } 1667 } 1668 bp->bio_driver1 = bp->bio_driver2 = NULL; 1669 bp->bio_pflags = 0; 1670 bp->bio_offset /= sc->sc_ndisks - 1; 1671 bp->bio_length /= sc->sc_ndisks - 1; 1672 bp->bio_cmd = BIO_WRITE; 1673 bp->bio_cflags = 0; 1674 bp->bio_children = bp->bio_inbed = 0; 1675 cp = disk->d_consumer; 1676 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1677 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1678 cp->acr, cp->acw, cp->ace)); 1679 cp->index++; 1680 g_io_request(bp, cp); 1681 return; 1682 } 1683 case BIO_WRITE: 1684 { 1685 struct g_raid3_disk_sync *sync; 1686 off_t boffset, moffset; 1687 void *data; 1688 int i; 1689 1690 if (bp->bio_error != 0) { 1691 G_RAID3_LOGREQ(0, bp, 1692 "Synchronization request failed (error=%d).", 1693 bp->bio_error); 1694 g_destroy_bio(bp); 1695 sc->sc_bump_id |= G_RAID3_BUMP_GENID; 1696 g_raid3_event_send(disk, 1697 G_RAID3_DISK_STATE_DISCONNECTED, 1698 G_RAID3_EVENT_DONTWAIT); 1699 return; 1700 } 1701 G_RAID3_LOGREQ(3, bp, "Synchronization request finished."); 1702 sync = &disk->d_sync; 1703 if (sync->ds_offset == sc->sc_mediasize / (sc->sc_ndisks - 1) || 1704 sync->ds_consumer == NULL || 1705 (sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 1706 /* Don't send more synchronization requests. */ 1707 sync->ds_inflight--; 1708 if (sync->ds_bios != NULL) { 1709 i = (int)(uintptr_t)bp->bio_caller1; 1710 sync->ds_bios[i] = NULL; 1711 } 1712 free(bp->bio_data, M_RAID3); 1713 g_destroy_bio(bp); 1714 if (sync->ds_inflight > 0) 1715 return; 1716 if (sync->ds_consumer == NULL || 1717 (sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 1718 return; 1719 } 1720 /* 1721 * Disk up-to-date, activate it. 1722 */ 1723 g_raid3_event_send(disk, G_RAID3_DISK_STATE_ACTIVE, 1724 G_RAID3_EVENT_DONTWAIT); 1725 return; 1726 } 1727 1728 /* Send next synchronization request. */ 1729 data = bp->bio_data; 1730 bzero(bp, sizeof(*bp)); 1731 bp->bio_cmd = BIO_READ; 1732 bp->bio_offset = sync->ds_offset * (sc->sc_ndisks - 1); 1733 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1734 sync->ds_offset += bp->bio_length / (sc->sc_ndisks - 1); 1735 bp->bio_done = g_raid3_sync_done; 1736 bp->bio_data = data; 1737 bp->bio_from = sync->ds_consumer; 1738 bp->bio_to = sc->sc_provider; 1739 G_RAID3_LOGREQ(3, bp, "Sending synchronization request."); 1740 sync->ds_consumer->index++; 1741 /* 1742 * Delay the request if it is colliding with a regular request. 1743 */ 1744 if (g_raid3_regular_collision(sc, bp)) 1745 g_raid3_sync_delay(sc, bp); 1746 else 1747 g_io_request(bp, sync->ds_consumer); 1748 1749 /* Release delayed requests if possible. */ 1750 g_raid3_regular_release(sc); 1751 1752 /* Find the smallest offset. */ 1753 moffset = sc->sc_mediasize; 1754 for (i = 0; i < g_raid3_syncreqs; i++) { 1755 bp = sync->ds_bios[i]; 1756 boffset = bp->bio_offset; 1757 if (bp->bio_cmd == BIO_WRITE) 1758 boffset *= sc->sc_ndisks - 1; 1759 if (boffset < moffset) 1760 moffset = boffset; 1761 } 1762 if (sync->ds_offset_done + (MAXPHYS * 100) < moffset) { 1763 /* Update offset_done on every 100 blocks. */ 1764 sync->ds_offset_done = moffset; 1765 g_raid3_update_metadata(disk); 1766 } 1767 return; 1768 } 1769 default: 1770 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1771 bp->bio_cmd, sc->sc_name)); 1772 break; 1773 } 1774 } 1775 1776 static int 1777 g_raid3_register_request(struct bio *pbp) 1778 { 1779 struct g_raid3_softc *sc; 1780 struct g_raid3_disk *disk; 1781 struct g_consumer *cp; 1782 struct bio *cbp, *tmpbp; 1783 off_t offset, length; 1784 u_int n, ndisks; 1785 int round_robin, verify; 1786 1787 ndisks = 0; 1788 sc = pbp->bio_to->geom->softc; 1789 if ((pbp->bio_cflags & G_RAID3_BIO_CFLAG_REGSYNC) != 0 && 1790 sc->sc_syncdisk == NULL) { 1791 g_io_deliver(pbp, EIO); 1792 return (0); 1793 } 1794 g_raid3_init_bio(pbp); 1795 length = pbp->bio_length / (sc->sc_ndisks - 1); 1796 offset = pbp->bio_offset / (sc->sc_ndisks - 1); 1797 round_robin = verify = 0; 1798 switch (pbp->bio_cmd) { 1799 case BIO_READ: 1800 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_VERIFY) != 0 && 1801 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 1802 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_VERIFY; 1803 verify = 1; 1804 ndisks = sc->sc_ndisks; 1805 } else { 1806 verify = 0; 1807 ndisks = sc->sc_ndisks - 1; 1808 } 1809 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_ROUND_ROBIN) != 0 && 1810 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 1811 round_robin = 1; 1812 } else { 1813 round_robin = 0; 1814 } 1815 KASSERT(!round_robin || !verify, 1816 ("ROUND-ROBIN and VERIFY are mutually exclusive.")); 1817 pbp->bio_driver2 = &sc->sc_disks[sc->sc_ndisks - 1]; 1818 break; 1819 case BIO_WRITE: 1820 case BIO_DELETE: 1821 /* 1822 * Delay the request if it is colliding with a synchronization 1823 * request. 1824 */ 1825 if (g_raid3_sync_collision(sc, pbp)) { 1826 g_raid3_regular_delay(sc, pbp); 1827 return (0); 1828 } 1829 1830 if (sc->sc_idle) 1831 g_raid3_unidle(sc); 1832 else 1833 sc->sc_last_write = time_uptime; 1834 1835 ndisks = sc->sc_ndisks; 1836 break; 1837 } 1838 for (n = 0; n < ndisks; n++) { 1839 disk = &sc->sc_disks[n]; 1840 cbp = g_raid3_clone_bio(sc, pbp); 1841 if (cbp == NULL) { 1842 while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL) 1843 g_raid3_destroy_bio(sc, cbp); 1844 /* 1845 * To prevent deadlock, we must run back up 1846 * with the ENOMEM for failed requests of any 1847 * of our consumers. Our own sync requests 1848 * can stick around, as they are finite. 1849 */ 1850 if ((pbp->bio_cflags & 1851 G_RAID3_BIO_CFLAG_REGULAR) != 0) { 1852 g_io_deliver(pbp, ENOMEM); 1853 return (0); 1854 } 1855 return (ENOMEM); 1856 } 1857 cbp->bio_offset = offset; 1858 cbp->bio_length = length; 1859 cbp->bio_done = g_raid3_done; 1860 switch (pbp->bio_cmd) { 1861 case BIO_READ: 1862 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) { 1863 /* 1864 * Replace invalid component with the parity 1865 * component. 1866 */ 1867 disk = &sc->sc_disks[sc->sc_ndisks - 1]; 1868 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1869 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED; 1870 } else if (round_robin && 1871 disk->d_no == sc->sc_round_robin) { 1872 /* 1873 * In round-robin mode skip one data component 1874 * and use parity component when reading. 1875 */ 1876 pbp->bio_driver2 = disk; 1877 disk = &sc->sc_disks[sc->sc_ndisks - 1]; 1878 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1879 sc->sc_round_robin++; 1880 round_robin = 0; 1881 } else if (verify && disk->d_no == sc->sc_ndisks - 1) { 1882 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1883 } 1884 break; 1885 case BIO_WRITE: 1886 case BIO_DELETE: 1887 if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 1888 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 1889 if (n == ndisks - 1) { 1890 /* 1891 * Active parity component, mark it as such. 1892 */ 1893 cbp->bio_cflags |= 1894 G_RAID3_BIO_CFLAG_PARITY; 1895 } 1896 } else { 1897 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED; 1898 if (n == ndisks - 1) { 1899 /* 1900 * Parity component is not connected, 1901 * so destroy its request. 1902 */ 1903 pbp->bio_pflags |= 1904 G_RAID3_BIO_PFLAG_NOPARITY; 1905 g_raid3_destroy_bio(sc, cbp); 1906 cbp = NULL; 1907 } else { 1908 cbp->bio_cflags |= 1909 G_RAID3_BIO_CFLAG_NODISK; 1910 disk = NULL; 1911 } 1912 } 1913 break; 1914 } 1915 if (cbp != NULL) 1916 cbp->bio_caller2 = disk; 1917 } 1918 switch (pbp->bio_cmd) { 1919 case BIO_READ: 1920 if (round_robin) { 1921 /* 1922 * If we are in round-robin mode and 'round_robin' is 1923 * still 1, it means, that we skipped parity component 1924 * for this read and must reset sc_round_robin field. 1925 */ 1926 sc->sc_round_robin = 0; 1927 } 1928 G_RAID3_FOREACH_SAFE_BIO(pbp, cbp, tmpbp) { 1929 disk = cbp->bio_caller2; 1930 cp = disk->d_consumer; 1931 cbp->bio_to = cp->provider; 1932 G_RAID3_LOGREQ(3, cbp, "Sending request."); 1933 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1934 ("Consumer %s not opened (r%dw%de%d).", 1935 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1936 cp->index++; 1937 g_io_request(cbp, cp); 1938 } 1939 break; 1940 case BIO_WRITE: 1941 case BIO_DELETE: 1942 /* 1943 * Put request onto inflight queue, so we can check if new 1944 * synchronization requests don't collide with it. 1945 */ 1946 bioq_insert_tail(&sc->sc_inflight, pbp); 1947 1948 /* 1949 * Bump syncid on first write. 1950 */ 1951 if ((sc->sc_bump_id & G_RAID3_BUMP_SYNCID) != 0) { 1952 sc->sc_bump_id &= ~G_RAID3_BUMP_SYNCID; 1953 g_raid3_bump_syncid(sc); 1954 } 1955 g_raid3_scatter(pbp); 1956 break; 1957 } 1958 return (0); 1959 } 1960 1961 static int 1962 g_raid3_can_destroy(struct g_raid3_softc *sc) 1963 { 1964 struct g_geom *gp; 1965 struct g_consumer *cp; 1966 1967 g_topology_assert(); 1968 gp = sc->sc_geom; 1969 if (gp->softc == NULL) 1970 return (1); 1971 LIST_FOREACH(cp, &gp->consumer, consumer) { 1972 if (g_raid3_is_busy(sc, cp)) 1973 return (0); 1974 } 1975 gp = sc->sc_sync.ds_geom; 1976 LIST_FOREACH(cp, &gp->consumer, consumer) { 1977 if (g_raid3_is_busy(sc, cp)) 1978 return (0); 1979 } 1980 G_RAID3_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1981 sc->sc_name); 1982 return (1); 1983 } 1984 1985 static int 1986 g_raid3_try_destroy(struct g_raid3_softc *sc) 1987 { 1988 1989 g_topology_assert_not(); 1990 sx_assert(&sc->sc_lock, SX_XLOCKED); 1991 1992 if (sc->sc_rootmount != NULL) { 1993 G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1994 sc->sc_rootmount); 1995 root_mount_rel(sc->sc_rootmount); 1996 sc->sc_rootmount = NULL; 1997 } 1998 1999 g_topology_lock(); 2000 if (!g_raid3_can_destroy(sc)) { 2001 g_topology_unlock(); 2002 return (0); 2003 } 2004 sc->sc_geom->softc = NULL; 2005 sc->sc_sync.ds_geom->softc = NULL; 2006 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_WAIT) != 0) { 2007 g_topology_unlock(); 2008 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, 2009 &sc->sc_worker); 2010 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 2011 sx_xunlock(&sc->sc_lock); 2012 wakeup(&sc->sc_worker); 2013 sc->sc_worker = NULL; 2014 } else { 2015 g_topology_unlock(); 2016 g_raid3_destroy_device(sc); 2017 free(sc->sc_disks, M_RAID3); 2018 free(sc, M_RAID3); 2019 } 2020 return (1); 2021 } 2022 2023 /* 2024 * Worker thread. 2025 */ 2026 static void 2027 g_raid3_worker(void *arg) 2028 { 2029 struct g_raid3_softc *sc; 2030 struct g_raid3_event *ep; 2031 struct bio *bp; 2032 int timeout; 2033 2034 sc = arg; 2035 thread_lock(curthread); 2036 sched_prio(curthread, PRIBIO); 2037 thread_unlock(curthread); 2038 2039 sx_xlock(&sc->sc_lock); 2040 for (;;) { 2041 G_RAID3_DEBUG(5, "%s: Let's see...", __func__); 2042 /* 2043 * First take a look at events. 2044 * This is important to handle events before any I/O requests. 2045 */ 2046 ep = g_raid3_event_get(sc); 2047 if (ep != NULL) { 2048 g_raid3_event_remove(sc, ep); 2049 if ((ep->e_flags & G_RAID3_EVENT_DEVICE) != 0) { 2050 /* Update only device status. */ 2051 G_RAID3_DEBUG(3, 2052 "Running event for device %s.", 2053 sc->sc_name); 2054 ep->e_error = 0; 2055 g_raid3_update_device(sc, 1); 2056 } else { 2057 /* Update disk status. */ 2058 G_RAID3_DEBUG(3, "Running event for disk %s.", 2059 g_raid3_get_diskname(ep->e_disk)); 2060 ep->e_error = g_raid3_update_disk(ep->e_disk, 2061 ep->e_state); 2062 if (ep->e_error == 0) 2063 g_raid3_update_device(sc, 0); 2064 } 2065 if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0) { 2066 KASSERT(ep->e_error == 0, 2067 ("Error cannot be handled.")); 2068 g_raid3_event_free(ep); 2069 } else { 2070 ep->e_flags |= G_RAID3_EVENT_DONE; 2071 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, 2072 ep); 2073 mtx_lock(&sc->sc_events_mtx); 2074 wakeup(ep); 2075 mtx_unlock(&sc->sc_events_mtx); 2076 } 2077 if ((sc->sc_flags & 2078 G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 2079 if (g_raid3_try_destroy(sc)) { 2080 curthread->td_pflags &= ~TDP_GEOM; 2081 G_RAID3_DEBUG(1, "Thread exiting."); 2082 kproc_exit(0); 2083 } 2084 } 2085 G_RAID3_DEBUG(5, "%s: I'm here 1.", __func__); 2086 continue; 2087 } 2088 /* 2089 * Check if we can mark array as CLEAN and if we can't take 2090 * how much seconds should we wait. 2091 */ 2092 timeout = g_raid3_idle(sc, -1); 2093 /* 2094 * Now I/O requests. 2095 */ 2096 /* Get first request from the queue. */ 2097 mtx_lock(&sc->sc_queue_mtx); 2098 bp = bioq_first(&sc->sc_queue); 2099 if (bp == NULL) { 2100 if ((sc->sc_flags & 2101 G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 2102 mtx_unlock(&sc->sc_queue_mtx); 2103 if (g_raid3_try_destroy(sc)) { 2104 curthread->td_pflags &= ~TDP_GEOM; 2105 G_RAID3_DEBUG(1, "Thread exiting."); 2106 kproc_exit(0); 2107 } 2108 mtx_lock(&sc->sc_queue_mtx); 2109 } 2110 sx_xunlock(&sc->sc_lock); 2111 /* 2112 * XXX: We can miss an event here, because an event 2113 * can be added without sx-device-lock and without 2114 * mtx-queue-lock. Maybe I should just stop using 2115 * dedicated mutex for events synchronization and 2116 * stick with the queue lock? 2117 * The event will hang here until next I/O request 2118 * or next event is received. 2119 */ 2120 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "r3:w1", 2121 timeout * hz); 2122 sx_xlock(&sc->sc_lock); 2123 G_RAID3_DEBUG(5, "%s: I'm here 4.", __func__); 2124 continue; 2125 } 2126 process: 2127 bioq_remove(&sc->sc_queue, bp); 2128 mtx_unlock(&sc->sc_queue_mtx); 2129 2130 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 2131 (bp->bio_cflags & G_RAID3_BIO_CFLAG_SYNC) != 0) { 2132 g_raid3_sync_request(bp); /* READ */ 2133 } else if (bp->bio_to != sc->sc_provider) { 2134 if ((bp->bio_cflags & G_RAID3_BIO_CFLAG_REGULAR) != 0) 2135 g_raid3_regular_request(bp); 2136 else if ((bp->bio_cflags & G_RAID3_BIO_CFLAG_SYNC) != 0) 2137 g_raid3_sync_request(bp); /* WRITE */ 2138 else { 2139 KASSERT(0, 2140 ("Invalid request cflags=0x%hhx to=%s.", 2141 bp->bio_cflags, bp->bio_to->name)); 2142 } 2143 } else if (g_raid3_register_request(bp) != 0) { 2144 mtx_lock(&sc->sc_queue_mtx); 2145 bioq_insert_head(&sc->sc_queue, bp); 2146 /* 2147 * We are short in memory, let see if there are finished 2148 * request we can free. 2149 */ 2150 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 2151 if (bp->bio_cflags & G_RAID3_BIO_CFLAG_REGULAR) 2152 goto process; 2153 } 2154 /* 2155 * No finished regular request, so at least keep 2156 * synchronization running. 2157 */ 2158 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 2159 if (bp->bio_cflags & G_RAID3_BIO_CFLAG_SYNC) 2160 goto process; 2161 } 2162 sx_xunlock(&sc->sc_lock); 2163 MSLEEP(&sc->sc_queue, &sc->sc_queue_mtx, PRIBIO | PDROP, 2164 "r3:lowmem", hz / 10); 2165 sx_xlock(&sc->sc_lock); 2166 } 2167 G_RAID3_DEBUG(5, "%s: I'm here 9.", __func__); 2168 } 2169 } 2170 2171 static void 2172 g_raid3_update_idle(struct g_raid3_softc *sc, struct g_raid3_disk *disk) 2173 { 2174 2175 sx_assert(&sc->sc_lock, SX_LOCKED); 2176 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) != 0) 2177 return; 2178 if (!sc->sc_idle && (disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) == 0) { 2179 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as dirty.", 2180 g_raid3_get_diskname(disk), sc->sc_name); 2181 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY; 2182 } else if (sc->sc_idle && 2183 (disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) != 0) { 2184 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as clean.", 2185 g_raid3_get_diskname(disk), sc->sc_name); 2186 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2187 } 2188 } 2189 2190 static void 2191 g_raid3_sync_start(struct g_raid3_softc *sc) 2192 { 2193 struct g_raid3_disk *disk; 2194 struct g_consumer *cp; 2195 struct bio *bp; 2196 int error; 2197 u_int n; 2198 2199 g_topology_assert_not(); 2200 sx_assert(&sc->sc_lock, SX_XLOCKED); 2201 2202 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED, 2203 ("Device not in DEGRADED state (%s, %u).", sc->sc_name, 2204 sc->sc_state)); 2205 KASSERT(sc->sc_syncdisk == NULL, ("Syncdisk is not NULL (%s, %u).", 2206 sc->sc_name, sc->sc_state)); 2207 disk = NULL; 2208 for (n = 0; n < sc->sc_ndisks; n++) { 2209 if (sc->sc_disks[n].d_state != G_RAID3_DISK_STATE_SYNCHRONIZING) 2210 continue; 2211 disk = &sc->sc_disks[n]; 2212 break; 2213 } 2214 if (disk == NULL) 2215 return; 2216 2217 sx_xunlock(&sc->sc_lock); 2218 g_topology_lock(); 2219 cp = g_new_consumer(sc->sc_sync.ds_geom); 2220 error = g_attach(cp, sc->sc_provider); 2221 KASSERT(error == 0, 2222 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 2223 error = g_access(cp, 1, 0, 0); 2224 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 2225 g_topology_unlock(); 2226 sx_xlock(&sc->sc_lock); 2227 2228 G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 2229 g_raid3_get_diskname(disk)); 2230 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) == 0) 2231 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY; 2232 KASSERT(disk->d_sync.ds_consumer == NULL, 2233 ("Sync consumer already exists (device=%s, disk=%s).", 2234 sc->sc_name, g_raid3_get_diskname(disk))); 2235 2236 disk->d_sync.ds_consumer = cp; 2237 disk->d_sync.ds_consumer->private = disk; 2238 disk->d_sync.ds_consumer->index = 0; 2239 sc->sc_syncdisk = disk; 2240 2241 /* 2242 * Allocate memory for synchronization bios and initialize them. 2243 */ 2244 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_raid3_syncreqs, 2245 M_RAID3, M_WAITOK); 2246 for (n = 0; n < g_raid3_syncreqs; n++) { 2247 bp = g_alloc_bio(); 2248 disk->d_sync.ds_bios[n] = bp; 2249 bp->bio_parent = NULL; 2250 bp->bio_cmd = BIO_READ; 2251 bp->bio_data = malloc(MAXPHYS, M_RAID3, M_WAITOK); 2252 bp->bio_cflags = 0; 2253 bp->bio_offset = disk->d_sync.ds_offset * (sc->sc_ndisks - 1); 2254 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 2255 disk->d_sync.ds_offset += bp->bio_length / (sc->sc_ndisks - 1); 2256 bp->bio_done = g_raid3_sync_done; 2257 bp->bio_from = disk->d_sync.ds_consumer; 2258 bp->bio_to = sc->sc_provider; 2259 bp->bio_caller1 = (void *)(uintptr_t)n; 2260 } 2261 2262 /* Set the number of in-flight synchronization requests. */ 2263 disk->d_sync.ds_inflight = g_raid3_syncreqs; 2264 2265 /* 2266 * Fire off first synchronization requests. 2267 */ 2268 for (n = 0; n < g_raid3_syncreqs; n++) { 2269 bp = disk->d_sync.ds_bios[n]; 2270 G_RAID3_LOGREQ(3, bp, "Sending synchronization request."); 2271 disk->d_sync.ds_consumer->index++; 2272 /* 2273 * Delay the request if it is colliding with a regular request. 2274 */ 2275 if (g_raid3_regular_collision(sc, bp)) 2276 g_raid3_sync_delay(sc, bp); 2277 else 2278 g_io_request(bp, disk->d_sync.ds_consumer); 2279 } 2280 } 2281 2282 /* 2283 * Stop synchronization process. 2284 * type: 0 - synchronization finished 2285 * 1 - synchronization stopped 2286 */ 2287 static void 2288 g_raid3_sync_stop(struct g_raid3_softc *sc, int type) 2289 { 2290 struct g_raid3_disk *disk; 2291 struct g_consumer *cp; 2292 2293 g_topology_assert_not(); 2294 sx_assert(&sc->sc_lock, SX_LOCKED); 2295 2296 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED, 2297 ("Device not in DEGRADED state (%s, %u).", sc->sc_name, 2298 sc->sc_state)); 2299 disk = sc->sc_syncdisk; 2300 sc->sc_syncdisk = NULL; 2301 KASSERT(disk != NULL, ("No disk was synchronized (%s).", sc->sc_name)); 2302 KASSERT(disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING, 2303 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2304 g_raid3_disk_state2str(disk->d_state))); 2305 if (disk->d_sync.ds_consumer == NULL) 2306 return; 2307 2308 if (type == 0) { 2309 G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2310 sc->sc_name, g_raid3_get_diskname(disk)); 2311 } else /* if (type == 1) */ { 2312 G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2313 sc->sc_name, g_raid3_get_diskname(disk)); 2314 } 2315 free(disk->d_sync.ds_bios, M_RAID3); 2316 disk->d_sync.ds_bios = NULL; 2317 cp = disk->d_sync.ds_consumer; 2318 disk->d_sync.ds_consumer = NULL; 2319 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2320 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2321 g_topology_lock(); 2322 g_raid3_kill_consumer(sc, cp); 2323 g_topology_unlock(); 2324 sx_xlock(&sc->sc_lock); 2325 } 2326 2327 static void 2328 g_raid3_launch_provider(struct g_raid3_softc *sc) 2329 { 2330 struct g_provider *pp; 2331 struct g_raid3_disk *disk; 2332 int n; 2333 2334 sx_assert(&sc->sc_lock, SX_LOCKED); 2335 2336 g_topology_lock(); 2337 pp = g_new_providerf(sc->sc_geom, "raid3/%s", sc->sc_name); 2338 pp->mediasize = sc->sc_mediasize; 2339 pp->sectorsize = sc->sc_sectorsize; 2340 pp->stripesize = 0; 2341 pp->stripeoffset = 0; 2342 for (n = 0; n < sc->sc_ndisks; n++) { 2343 disk = &sc->sc_disks[n]; 2344 if (disk->d_consumer && disk->d_consumer->provider && 2345 disk->d_consumer->provider->stripesize > pp->stripesize) { 2346 pp->stripesize = disk->d_consumer->provider->stripesize; 2347 pp->stripeoffset = disk->d_consumer->provider->stripeoffset; 2348 } 2349 } 2350 pp->stripesize *= sc->sc_ndisks - 1; 2351 pp->stripeoffset *= sc->sc_ndisks - 1; 2352 sc->sc_provider = pp; 2353 g_error_provider(pp, 0); 2354 g_topology_unlock(); 2355 G_RAID3_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2356 g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE), sc->sc_ndisks); 2357 2358 if (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED) 2359 g_raid3_sync_start(sc); 2360 } 2361 2362 static void 2363 g_raid3_destroy_provider(struct g_raid3_softc *sc) 2364 { 2365 struct bio *bp; 2366 2367 g_topology_assert_not(); 2368 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2369 sc->sc_name)); 2370 2371 g_topology_lock(); 2372 g_error_provider(sc->sc_provider, ENXIO); 2373 mtx_lock(&sc->sc_queue_mtx); 2374 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 2375 bioq_remove(&sc->sc_queue, bp); 2376 g_io_deliver(bp, ENXIO); 2377 } 2378 mtx_unlock(&sc->sc_queue_mtx); 2379 G_RAID3_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 2380 sc->sc_provider->name); 2381 sc->sc_provider->flags |= G_PF_WITHER; 2382 g_orphan_provider(sc->sc_provider, ENXIO); 2383 g_topology_unlock(); 2384 sc->sc_provider = NULL; 2385 if (sc->sc_syncdisk != NULL) 2386 g_raid3_sync_stop(sc, 1); 2387 } 2388 2389 static void 2390 g_raid3_go(void *arg) 2391 { 2392 struct g_raid3_softc *sc; 2393 2394 sc = arg; 2395 G_RAID3_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2396 g_raid3_event_send(sc, 0, 2397 G_RAID3_EVENT_DONTWAIT | G_RAID3_EVENT_DEVICE); 2398 } 2399 2400 static u_int 2401 g_raid3_determine_state(struct g_raid3_disk *disk) 2402 { 2403 struct g_raid3_softc *sc; 2404 u_int state; 2405 2406 sc = disk->d_softc; 2407 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2408 if ((disk->d_flags & 2409 G_RAID3_DISK_FLAG_SYNCHRONIZING) == 0) { 2410 /* Disk does not need synchronization. */ 2411 state = G_RAID3_DISK_STATE_ACTIVE; 2412 } else { 2413 if ((sc->sc_flags & 2414 G_RAID3_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2415 (disk->d_flags & 2416 G_RAID3_DISK_FLAG_FORCE_SYNC) != 0) { 2417 /* 2418 * We can start synchronization from 2419 * the stored offset. 2420 */ 2421 state = G_RAID3_DISK_STATE_SYNCHRONIZING; 2422 } else { 2423 state = G_RAID3_DISK_STATE_STALE; 2424 } 2425 } 2426 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2427 /* 2428 * Reset all synchronization data for this disk, 2429 * because if it even was synchronized, it was 2430 * synchronized to disks with different syncid. 2431 */ 2432 disk->d_flags |= G_RAID3_DISK_FLAG_SYNCHRONIZING; 2433 disk->d_sync.ds_offset = 0; 2434 disk->d_sync.ds_offset_done = 0; 2435 disk->d_sync.ds_syncid = sc->sc_syncid; 2436 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2437 (disk->d_flags & G_RAID3_DISK_FLAG_FORCE_SYNC) != 0) { 2438 state = G_RAID3_DISK_STATE_SYNCHRONIZING; 2439 } else { 2440 state = G_RAID3_DISK_STATE_STALE; 2441 } 2442 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2443 /* 2444 * Not good, NOT GOOD! 2445 * It means that device was started on stale disks 2446 * and more fresh disk just arrive. 2447 * If there were writes, device is broken, sorry. 2448 * I think the best choice here is don't touch 2449 * this disk and inform the user loudly. 2450 */ 2451 G_RAID3_DEBUG(0, "Device %s was started before the freshest " 2452 "disk (%s) arrives!! It will not be connected to the " 2453 "running device.", sc->sc_name, 2454 g_raid3_get_diskname(disk)); 2455 g_raid3_destroy_disk(disk); 2456 state = G_RAID3_DISK_STATE_NONE; 2457 /* Return immediately, because disk was destroyed. */ 2458 return (state); 2459 } 2460 G_RAID3_DEBUG(3, "State for %s disk: %s.", 2461 g_raid3_get_diskname(disk), g_raid3_disk_state2str(state)); 2462 return (state); 2463 } 2464 2465 /* 2466 * Update device state. 2467 */ 2468 static void 2469 g_raid3_update_device(struct g_raid3_softc *sc, boolean_t force) 2470 { 2471 struct g_raid3_disk *disk; 2472 u_int state; 2473 2474 sx_assert(&sc->sc_lock, SX_XLOCKED); 2475 2476 switch (sc->sc_state) { 2477 case G_RAID3_DEVICE_STATE_STARTING: 2478 { 2479 u_int n, ndirty, ndisks, genid, syncid; 2480 2481 KASSERT(sc->sc_provider == NULL, 2482 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2483 /* 2484 * Are we ready? We are, if all disks are connected or 2485 * one disk is missing and 'force' is true. 2486 */ 2487 if (g_raid3_ndisks(sc, -1) + force == sc->sc_ndisks) { 2488 if (!force) 2489 callout_drain(&sc->sc_callout); 2490 } else { 2491 if (force) { 2492 /* 2493 * Timeout expired, so destroy device. 2494 */ 2495 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 2496 G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", 2497 __LINE__, sc->sc_rootmount); 2498 root_mount_rel(sc->sc_rootmount); 2499 sc->sc_rootmount = NULL; 2500 } 2501 return; 2502 } 2503 2504 /* 2505 * Find the biggest genid. 2506 */ 2507 genid = 0; 2508 for (n = 0; n < sc->sc_ndisks; n++) { 2509 disk = &sc->sc_disks[n]; 2510 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2511 continue; 2512 if (disk->d_genid > genid) 2513 genid = disk->d_genid; 2514 } 2515 sc->sc_genid = genid; 2516 /* 2517 * Remove all disks without the biggest genid. 2518 */ 2519 for (n = 0; n < sc->sc_ndisks; n++) { 2520 disk = &sc->sc_disks[n]; 2521 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2522 continue; 2523 if (disk->d_genid < genid) { 2524 G_RAID3_DEBUG(0, 2525 "Component %s (device %s) broken, skipping.", 2526 g_raid3_get_diskname(disk), sc->sc_name); 2527 g_raid3_destroy_disk(disk); 2528 } 2529 } 2530 2531 /* 2532 * There must be at least 'sc->sc_ndisks - 1' components 2533 * with the same syncid and without SYNCHRONIZING flag. 2534 */ 2535 2536 /* 2537 * Find the biggest syncid, number of valid components and 2538 * number of dirty components. 2539 */ 2540 ndirty = ndisks = syncid = 0; 2541 for (n = 0; n < sc->sc_ndisks; n++) { 2542 disk = &sc->sc_disks[n]; 2543 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2544 continue; 2545 if ((disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) != 0) 2546 ndirty++; 2547 if (disk->d_sync.ds_syncid > syncid) { 2548 syncid = disk->d_sync.ds_syncid; 2549 ndisks = 0; 2550 } else if (disk->d_sync.ds_syncid < syncid) { 2551 continue; 2552 } 2553 if ((disk->d_flags & 2554 G_RAID3_DISK_FLAG_SYNCHRONIZING) != 0) { 2555 continue; 2556 } 2557 ndisks++; 2558 } 2559 /* 2560 * Do we have enough valid components? 2561 */ 2562 if (ndisks + 1 < sc->sc_ndisks) { 2563 G_RAID3_DEBUG(0, 2564 "Device %s is broken, too few valid components.", 2565 sc->sc_name); 2566 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 2567 return; 2568 } 2569 /* 2570 * If there is one DIRTY component and all disks are present, 2571 * mark it for synchronization. If there is more than one DIRTY 2572 * component, mark parity component for synchronization. 2573 */ 2574 if (ndisks == sc->sc_ndisks && ndirty == 1) { 2575 for (n = 0; n < sc->sc_ndisks; n++) { 2576 disk = &sc->sc_disks[n]; 2577 if ((disk->d_flags & 2578 G_RAID3_DISK_FLAG_DIRTY) == 0) { 2579 continue; 2580 } 2581 disk->d_flags |= 2582 G_RAID3_DISK_FLAG_SYNCHRONIZING; 2583 } 2584 } else if (ndisks == sc->sc_ndisks && ndirty > 1) { 2585 disk = &sc->sc_disks[sc->sc_ndisks - 1]; 2586 disk->d_flags |= G_RAID3_DISK_FLAG_SYNCHRONIZING; 2587 } 2588 2589 sc->sc_syncid = syncid; 2590 if (force) { 2591 /* Remember to bump syncid on first write. */ 2592 sc->sc_bump_id |= G_RAID3_BUMP_SYNCID; 2593 } 2594 if (ndisks == sc->sc_ndisks) 2595 state = G_RAID3_DEVICE_STATE_COMPLETE; 2596 else /* if (ndisks == sc->sc_ndisks - 1) */ 2597 state = G_RAID3_DEVICE_STATE_DEGRADED; 2598 G_RAID3_DEBUG(1, "Device %s state changed from %s to %s.", 2599 sc->sc_name, g_raid3_device_state2str(sc->sc_state), 2600 g_raid3_device_state2str(state)); 2601 sc->sc_state = state; 2602 for (n = 0; n < sc->sc_ndisks; n++) { 2603 disk = &sc->sc_disks[n]; 2604 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2605 continue; 2606 state = g_raid3_determine_state(disk); 2607 g_raid3_event_send(disk, state, G_RAID3_EVENT_DONTWAIT); 2608 if (state == G_RAID3_DISK_STATE_STALE) 2609 sc->sc_bump_id |= G_RAID3_BUMP_SYNCID; 2610 } 2611 break; 2612 } 2613 case G_RAID3_DEVICE_STATE_DEGRADED: 2614 /* 2615 * Genid need to be bumped immediately, so do it here. 2616 */ 2617 if ((sc->sc_bump_id & G_RAID3_BUMP_GENID) != 0) { 2618 sc->sc_bump_id &= ~G_RAID3_BUMP_GENID; 2619 g_raid3_bump_genid(sc); 2620 } 2621 2622 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NEW) > 0) 2623 return; 2624 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) < 2625 sc->sc_ndisks - 1) { 2626 if (sc->sc_provider != NULL) 2627 g_raid3_destroy_provider(sc); 2628 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 2629 return; 2630 } 2631 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) == 2632 sc->sc_ndisks) { 2633 state = G_RAID3_DEVICE_STATE_COMPLETE; 2634 G_RAID3_DEBUG(1, 2635 "Device %s state changed from %s to %s.", 2636 sc->sc_name, g_raid3_device_state2str(sc->sc_state), 2637 g_raid3_device_state2str(state)); 2638 sc->sc_state = state; 2639 } 2640 if (sc->sc_provider == NULL) 2641 g_raid3_launch_provider(sc); 2642 if (sc->sc_rootmount != NULL) { 2643 G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2644 sc->sc_rootmount); 2645 root_mount_rel(sc->sc_rootmount); 2646 sc->sc_rootmount = NULL; 2647 } 2648 break; 2649 case G_RAID3_DEVICE_STATE_COMPLETE: 2650 /* 2651 * Genid need to be bumped immediately, so do it here. 2652 */ 2653 if ((sc->sc_bump_id & G_RAID3_BUMP_GENID) != 0) { 2654 sc->sc_bump_id &= ~G_RAID3_BUMP_GENID; 2655 g_raid3_bump_genid(sc); 2656 } 2657 2658 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NEW) > 0) 2659 return; 2660 KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) >= 2661 sc->sc_ndisks - 1, 2662 ("Too few ACTIVE components in COMPLETE state (device %s).", 2663 sc->sc_name)); 2664 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) == 2665 sc->sc_ndisks - 1) { 2666 state = G_RAID3_DEVICE_STATE_DEGRADED; 2667 G_RAID3_DEBUG(1, 2668 "Device %s state changed from %s to %s.", 2669 sc->sc_name, g_raid3_device_state2str(sc->sc_state), 2670 g_raid3_device_state2str(state)); 2671 sc->sc_state = state; 2672 } 2673 if (sc->sc_provider == NULL) 2674 g_raid3_launch_provider(sc); 2675 if (sc->sc_rootmount != NULL) { 2676 G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2677 sc->sc_rootmount); 2678 root_mount_rel(sc->sc_rootmount); 2679 sc->sc_rootmount = NULL; 2680 } 2681 break; 2682 default: 2683 KASSERT(1 == 0, ("Wrong device state (%s, %s).", sc->sc_name, 2684 g_raid3_device_state2str(sc->sc_state))); 2685 break; 2686 } 2687 } 2688 2689 /* 2690 * Update disk state and device state if needed. 2691 */ 2692 #define DISK_STATE_CHANGED() G_RAID3_DEBUG(1, \ 2693 "Disk %s state changed from %s to %s (device %s).", \ 2694 g_raid3_get_diskname(disk), \ 2695 g_raid3_disk_state2str(disk->d_state), \ 2696 g_raid3_disk_state2str(state), sc->sc_name) 2697 static int 2698 g_raid3_update_disk(struct g_raid3_disk *disk, u_int state) 2699 { 2700 struct g_raid3_softc *sc; 2701 2702 sc = disk->d_softc; 2703 sx_assert(&sc->sc_lock, SX_XLOCKED); 2704 2705 again: 2706 G_RAID3_DEBUG(3, "Changing disk %s state from %s to %s.", 2707 g_raid3_get_diskname(disk), g_raid3_disk_state2str(disk->d_state), 2708 g_raid3_disk_state2str(state)); 2709 switch (state) { 2710 case G_RAID3_DISK_STATE_NEW: 2711 /* 2712 * Possible scenarios: 2713 * 1. New disk arrive. 2714 */ 2715 /* Previous state should be NONE. */ 2716 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NONE, 2717 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2718 g_raid3_disk_state2str(disk->d_state))); 2719 DISK_STATE_CHANGED(); 2720 2721 disk->d_state = state; 2722 G_RAID3_DEBUG(1, "Device %s: provider %s detected.", 2723 sc->sc_name, g_raid3_get_diskname(disk)); 2724 if (sc->sc_state == G_RAID3_DEVICE_STATE_STARTING) 2725 break; 2726 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2727 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2728 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2729 g_raid3_device_state2str(sc->sc_state), 2730 g_raid3_get_diskname(disk), 2731 g_raid3_disk_state2str(disk->d_state))); 2732 state = g_raid3_determine_state(disk); 2733 if (state != G_RAID3_DISK_STATE_NONE) 2734 goto again; 2735 break; 2736 case G_RAID3_DISK_STATE_ACTIVE: 2737 /* 2738 * Possible scenarios: 2739 * 1. New disk does not need synchronization. 2740 * 2. Synchronization process finished successfully. 2741 */ 2742 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2743 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2744 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2745 g_raid3_device_state2str(sc->sc_state), 2746 g_raid3_get_diskname(disk), 2747 g_raid3_disk_state2str(disk->d_state))); 2748 /* Previous state should be NEW or SYNCHRONIZING. */ 2749 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW || 2750 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING, 2751 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2752 g_raid3_disk_state2str(disk->d_state))); 2753 DISK_STATE_CHANGED(); 2754 2755 if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 2756 disk->d_flags &= ~G_RAID3_DISK_FLAG_SYNCHRONIZING; 2757 disk->d_flags &= ~G_RAID3_DISK_FLAG_FORCE_SYNC; 2758 g_raid3_sync_stop(sc, 0); 2759 } 2760 disk->d_state = state; 2761 disk->d_sync.ds_offset = 0; 2762 disk->d_sync.ds_offset_done = 0; 2763 g_raid3_update_idle(sc, disk); 2764 g_raid3_update_metadata(disk); 2765 G_RAID3_DEBUG(1, "Device %s: provider %s activated.", 2766 sc->sc_name, g_raid3_get_diskname(disk)); 2767 break; 2768 case G_RAID3_DISK_STATE_STALE: 2769 /* 2770 * Possible scenarios: 2771 * 1. Stale disk was connected. 2772 */ 2773 /* Previous state should be NEW. */ 2774 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW, 2775 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2776 g_raid3_disk_state2str(disk->d_state))); 2777 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2778 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2779 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2780 g_raid3_device_state2str(sc->sc_state), 2781 g_raid3_get_diskname(disk), 2782 g_raid3_disk_state2str(disk->d_state))); 2783 /* 2784 * STALE state is only possible if device is marked 2785 * NOAUTOSYNC. 2786 */ 2787 KASSERT((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOAUTOSYNC) != 0, 2788 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2789 g_raid3_device_state2str(sc->sc_state), 2790 g_raid3_get_diskname(disk), 2791 g_raid3_disk_state2str(disk->d_state))); 2792 DISK_STATE_CHANGED(); 2793 2794 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2795 disk->d_state = state; 2796 g_raid3_update_metadata(disk); 2797 G_RAID3_DEBUG(0, "Device %s: provider %s is stale.", 2798 sc->sc_name, g_raid3_get_diskname(disk)); 2799 break; 2800 case G_RAID3_DISK_STATE_SYNCHRONIZING: 2801 /* 2802 * Possible scenarios: 2803 * 1. Disk which needs synchronization was connected. 2804 */ 2805 /* Previous state should be NEW. */ 2806 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW, 2807 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2808 g_raid3_disk_state2str(disk->d_state))); 2809 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2810 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2811 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2812 g_raid3_device_state2str(sc->sc_state), 2813 g_raid3_get_diskname(disk), 2814 g_raid3_disk_state2str(disk->d_state))); 2815 DISK_STATE_CHANGED(); 2816 2817 if (disk->d_state == G_RAID3_DISK_STATE_NEW) 2818 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2819 disk->d_state = state; 2820 if (sc->sc_provider != NULL) { 2821 g_raid3_sync_start(sc); 2822 g_raid3_update_metadata(disk); 2823 } 2824 break; 2825 case G_RAID3_DISK_STATE_DISCONNECTED: 2826 /* 2827 * Possible scenarios: 2828 * 1. Device wasn't running yet, but disk disappear. 2829 * 2. Disk was active and disapppear. 2830 * 3. Disk disappear during synchronization process. 2831 */ 2832 if (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2833 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 2834 /* 2835 * Previous state should be ACTIVE, STALE or 2836 * SYNCHRONIZING. 2837 */ 2838 KASSERT(disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 2839 disk->d_state == G_RAID3_DISK_STATE_STALE || 2840 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING, 2841 ("Wrong disk state (%s, %s).", 2842 g_raid3_get_diskname(disk), 2843 g_raid3_disk_state2str(disk->d_state))); 2844 } else if (sc->sc_state == G_RAID3_DEVICE_STATE_STARTING) { 2845 /* Previous state should be NEW. */ 2846 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW, 2847 ("Wrong disk state (%s, %s).", 2848 g_raid3_get_diskname(disk), 2849 g_raid3_disk_state2str(disk->d_state))); 2850 /* 2851 * Reset bumping syncid if disk disappeared in STARTING 2852 * state. 2853 */ 2854 if ((sc->sc_bump_id & G_RAID3_BUMP_SYNCID) != 0) 2855 sc->sc_bump_id &= ~G_RAID3_BUMP_SYNCID; 2856 #ifdef INVARIANTS 2857 } else { 2858 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2859 sc->sc_name, 2860 g_raid3_device_state2str(sc->sc_state), 2861 g_raid3_get_diskname(disk), 2862 g_raid3_disk_state2str(disk->d_state))); 2863 #endif 2864 } 2865 DISK_STATE_CHANGED(); 2866 G_RAID3_DEBUG(0, "Device %s: provider %s disconnected.", 2867 sc->sc_name, g_raid3_get_diskname(disk)); 2868 2869 g_raid3_destroy_disk(disk); 2870 break; 2871 default: 2872 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2873 break; 2874 } 2875 return (0); 2876 } 2877 #undef DISK_STATE_CHANGED 2878 2879 int 2880 g_raid3_read_metadata(struct g_consumer *cp, struct g_raid3_metadata *md) 2881 { 2882 struct g_provider *pp; 2883 u_char *buf; 2884 int error; 2885 2886 g_topology_assert(); 2887 2888 error = g_access(cp, 1, 0, 0); 2889 if (error != 0) 2890 return (error); 2891 pp = cp->provider; 2892 g_topology_unlock(); 2893 /* Metadata are stored on last sector. */ 2894 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2895 &error); 2896 g_topology_lock(); 2897 g_access(cp, -1, 0, 0); 2898 if (buf == NULL) { 2899 G_RAID3_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2900 cp->provider->name, error); 2901 return (error); 2902 } 2903 2904 /* Decode metadata. */ 2905 error = raid3_metadata_decode(buf, md); 2906 g_free(buf); 2907 if (strcmp(md->md_magic, G_RAID3_MAGIC) != 0) 2908 return (EINVAL); 2909 if (md->md_version > G_RAID3_VERSION) { 2910 G_RAID3_DEBUG(0, 2911 "Kernel module is too old to handle metadata from %s.", 2912 cp->provider->name); 2913 return (EINVAL); 2914 } 2915 if (error != 0) { 2916 G_RAID3_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2917 cp->provider->name); 2918 return (error); 2919 } 2920 if (md->md_sectorsize > MAXPHYS) { 2921 G_RAID3_DEBUG(0, "The blocksize is too big."); 2922 return (EINVAL); 2923 } 2924 2925 return (0); 2926 } 2927 2928 static int 2929 g_raid3_check_metadata(struct g_raid3_softc *sc, struct g_provider *pp, 2930 struct g_raid3_metadata *md) 2931 { 2932 2933 if (md->md_no >= sc->sc_ndisks) { 2934 G_RAID3_DEBUG(1, "Invalid disk %s number (no=%u), skipping.", 2935 pp->name, md->md_no); 2936 return (EINVAL); 2937 } 2938 if (sc->sc_disks[md->md_no].d_state != G_RAID3_DISK_STATE_NODISK) { 2939 G_RAID3_DEBUG(1, "Disk %s (no=%u) already exists, skipping.", 2940 pp->name, md->md_no); 2941 return (EEXIST); 2942 } 2943 if (md->md_all != sc->sc_ndisks) { 2944 G_RAID3_DEBUG(1, 2945 "Invalid '%s' field on disk %s (device %s), skipping.", 2946 "md_all", pp->name, sc->sc_name); 2947 return (EINVAL); 2948 } 2949 if ((md->md_mediasize % md->md_sectorsize) != 0) { 2950 G_RAID3_DEBUG(1, "Invalid metadata (mediasize %% sectorsize != " 2951 "0) on disk %s (device %s), skipping.", pp->name, 2952 sc->sc_name); 2953 return (EINVAL); 2954 } 2955 if (md->md_mediasize != sc->sc_mediasize) { 2956 G_RAID3_DEBUG(1, 2957 "Invalid '%s' field on disk %s (device %s), skipping.", 2958 "md_mediasize", pp->name, sc->sc_name); 2959 return (EINVAL); 2960 } 2961 if ((md->md_mediasize % (sc->sc_ndisks - 1)) != 0) { 2962 G_RAID3_DEBUG(1, 2963 "Invalid '%s' field on disk %s (device %s), skipping.", 2964 "md_mediasize", pp->name, sc->sc_name); 2965 return (EINVAL); 2966 } 2967 if ((sc->sc_mediasize / (sc->sc_ndisks - 1)) > pp->mediasize) { 2968 G_RAID3_DEBUG(1, 2969 "Invalid size of disk %s (device %s), skipping.", pp->name, 2970 sc->sc_name); 2971 return (EINVAL); 2972 } 2973 if ((md->md_sectorsize / pp->sectorsize) < sc->sc_ndisks - 1) { 2974 G_RAID3_DEBUG(1, 2975 "Invalid '%s' field on disk %s (device %s), skipping.", 2976 "md_sectorsize", pp->name, sc->sc_name); 2977 return (EINVAL); 2978 } 2979 if (md->md_sectorsize != sc->sc_sectorsize) { 2980 G_RAID3_DEBUG(1, 2981 "Invalid '%s' field on disk %s (device %s), skipping.", 2982 "md_sectorsize", pp->name, sc->sc_name); 2983 return (EINVAL); 2984 } 2985 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2986 G_RAID3_DEBUG(1, 2987 "Invalid sector size of disk %s (device %s), skipping.", 2988 pp->name, sc->sc_name); 2989 return (EINVAL); 2990 } 2991 if ((md->md_mflags & ~G_RAID3_DEVICE_FLAG_MASK) != 0) { 2992 G_RAID3_DEBUG(1, 2993 "Invalid device flags on disk %s (device %s), skipping.", 2994 pp->name, sc->sc_name); 2995 return (EINVAL); 2996 } 2997 if ((md->md_mflags & G_RAID3_DEVICE_FLAG_VERIFY) != 0 && 2998 (md->md_mflags & G_RAID3_DEVICE_FLAG_ROUND_ROBIN) != 0) { 2999 /* 3000 * VERIFY and ROUND-ROBIN options are mutally exclusive. 3001 */ 3002 G_RAID3_DEBUG(1, "Both VERIFY and ROUND-ROBIN flags exist on " 3003 "disk %s (device %s), skipping.", pp->name, sc->sc_name); 3004 return (EINVAL); 3005 } 3006 if ((md->md_dflags & ~G_RAID3_DISK_FLAG_MASK) != 0) { 3007 G_RAID3_DEBUG(1, 3008 "Invalid disk flags on disk %s (device %s), skipping.", 3009 pp->name, sc->sc_name); 3010 return (EINVAL); 3011 } 3012 return (0); 3013 } 3014 3015 int 3016 g_raid3_add_disk(struct g_raid3_softc *sc, struct g_provider *pp, 3017 struct g_raid3_metadata *md) 3018 { 3019 struct g_raid3_disk *disk; 3020 int error; 3021 3022 g_topology_assert_not(); 3023 G_RAID3_DEBUG(2, "Adding disk %s.", pp->name); 3024 3025 error = g_raid3_check_metadata(sc, pp, md); 3026 if (error != 0) 3027 return (error); 3028 if (sc->sc_state != G_RAID3_DEVICE_STATE_STARTING && 3029 md->md_genid < sc->sc_genid) { 3030 G_RAID3_DEBUG(0, "Component %s (device %s) broken, skipping.", 3031 pp->name, sc->sc_name); 3032 return (EINVAL); 3033 } 3034 disk = g_raid3_init_disk(sc, pp, md, &error); 3035 if (disk == NULL) 3036 return (error); 3037 error = g_raid3_event_send(disk, G_RAID3_DISK_STATE_NEW, 3038 G_RAID3_EVENT_WAIT); 3039 if (error != 0) 3040 return (error); 3041 if (md->md_version < G_RAID3_VERSION) { 3042 G_RAID3_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 3043 pp->name, md->md_version, G_RAID3_VERSION); 3044 g_raid3_update_metadata(disk); 3045 } 3046 return (0); 3047 } 3048 3049 static void 3050 g_raid3_destroy_delayed(void *arg, int flag) 3051 { 3052 struct g_raid3_softc *sc; 3053 int error; 3054 3055 if (flag == EV_CANCEL) { 3056 G_RAID3_DEBUG(1, "Destroying canceled."); 3057 return; 3058 } 3059 sc = arg; 3060 g_topology_unlock(); 3061 sx_xlock(&sc->sc_lock); 3062 KASSERT((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) == 0, 3063 ("DESTROY flag set on %s.", sc->sc_name)); 3064 KASSERT((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROYING) != 0, 3065 ("DESTROYING flag not set on %s.", sc->sc_name)); 3066 G_RAID3_DEBUG(0, "Destroying %s (delayed).", sc->sc_name); 3067 error = g_raid3_destroy(sc, G_RAID3_DESTROY_SOFT); 3068 if (error != 0) { 3069 G_RAID3_DEBUG(0, "Cannot destroy %s.", sc->sc_name); 3070 sx_xunlock(&sc->sc_lock); 3071 } 3072 g_topology_lock(); 3073 } 3074 3075 static int 3076 g_raid3_access(struct g_provider *pp, int acr, int acw, int ace) 3077 { 3078 struct g_raid3_softc *sc; 3079 int dcr, dcw, dce, error = 0; 3080 3081 g_topology_assert(); 3082 G_RAID3_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 3083 acw, ace); 3084 3085 sc = pp->geom->softc; 3086 if (sc == NULL && acr <= 0 && acw <= 0 && ace <= 0) 3087 return (0); 3088 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 3089 3090 dcr = pp->acr + acr; 3091 dcw = pp->acw + acw; 3092 dce = pp->ace + ace; 3093 3094 g_topology_unlock(); 3095 sx_xlock(&sc->sc_lock); 3096 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) != 0 || 3097 g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) < sc->sc_ndisks - 1) { 3098 if (acr > 0 || acw > 0 || ace > 0) 3099 error = ENXIO; 3100 goto end; 3101 } 3102 if (dcw == 0) 3103 g_raid3_idle(sc, dcw); 3104 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROYING) != 0) { 3105 if (acr > 0 || acw > 0 || ace > 0) { 3106 error = ENXIO; 3107 goto end; 3108 } 3109 if (dcr == 0 && dcw == 0 && dce == 0) { 3110 g_post_event(g_raid3_destroy_delayed, sc, M_WAITOK, 3111 sc, NULL); 3112 } 3113 } 3114 end: 3115 sx_xunlock(&sc->sc_lock); 3116 g_topology_lock(); 3117 return (error); 3118 } 3119 3120 static struct g_geom * 3121 g_raid3_create(struct g_class *mp, const struct g_raid3_metadata *md) 3122 { 3123 struct g_raid3_softc *sc; 3124 struct g_geom *gp; 3125 int error, timeout; 3126 u_int n; 3127 3128 g_topology_assert(); 3129 G_RAID3_DEBUG(1, "Creating device %s (id=%u).", md->md_name, md->md_id); 3130 3131 /* One disk is minimum. */ 3132 if (md->md_all < 1) 3133 return (NULL); 3134 /* 3135 * Action geom. 3136 */ 3137 gp = g_new_geomf(mp, "%s", md->md_name); 3138 sc = malloc(sizeof(*sc), M_RAID3, M_WAITOK | M_ZERO); 3139 sc->sc_disks = malloc(sizeof(struct g_raid3_disk) * md->md_all, M_RAID3, 3140 M_WAITOK | M_ZERO); 3141 gp->start = g_raid3_start; 3142 gp->orphan = g_raid3_orphan; 3143 gp->access = g_raid3_access; 3144 gp->dumpconf = g_raid3_dumpconf; 3145 3146 sc->sc_id = md->md_id; 3147 sc->sc_mediasize = md->md_mediasize; 3148 sc->sc_sectorsize = md->md_sectorsize; 3149 sc->sc_ndisks = md->md_all; 3150 sc->sc_round_robin = 0; 3151 sc->sc_flags = md->md_mflags; 3152 sc->sc_bump_id = 0; 3153 sc->sc_idle = 1; 3154 sc->sc_last_write = time_uptime; 3155 sc->sc_writes = 0; 3156 for (n = 0; n < sc->sc_ndisks; n++) { 3157 sc->sc_disks[n].d_softc = sc; 3158 sc->sc_disks[n].d_no = n; 3159 sc->sc_disks[n].d_state = G_RAID3_DISK_STATE_NODISK; 3160 } 3161 sx_init(&sc->sc_lock, "graid3:lock"); 3162 bioq_init(&sc->sc_queue); 3163 mtx_init(&sc->sc_queue_mtx, "graid3:queue", NULL, MTX_DEF); 3164 bioq_init(&sc->sc_regular_delayed); 3165 bioq_init(&sc->sc_inflight); 3166 bioq_init(&sc->sc_sync_delayed); 3167 TAILQ_INIT(&sc->sc_events); 3168 mtx_init(&sc->sc_events_mtx, "graid3:events", NULL, MTX_DEF); 3169 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 3170 sc->sc_state = G_RAID3_DEVICE_STATE_STARTING; 3171 gp->softc = sc; 3172 sc->sc_geom = gp; 3173 sc->sc_provider = NULL; 3174 /* 3175 * Synchronization geom. 3176 */ 3177 gp = g_new_geomf(mp, "%s.sync", md->md_name); 3178 gp->softc = sc; 3179 gp->orphan = g_raid3_orphan; 3180 sc->sc_sync.ds_geom = gp; 3181 3182 if (!g_raid3_use_malloc) { 3183 sc->sc_zones[G_RAID3_ZONE_64K].sz_zone = uma_zcreate("gr3:64k", 3184 65536, g_raid3_uma_ctor, g_raid3_uma_dtor, NULL, NULL, 3185 UMA_ALIGN_PTR, 0); 3186 sc->sc_zones[G_RAID3_ZONE_64K].sz_inuse = 0; 3187 sc->sc_zones[G_RAID3_ZONE_64K].sz_max = g_raid3_n64k; 3188 sc->sc_zones[G_RAID3_ZONE_64K].sz_requested = 3189 sc->sc_zones[G_RAID3_ZONE_64K].sz_failed = 0; 3190 sc->sc_zones[G_RAID3_ZONE_16K].sz_zone = uma_zcreate("gr3:16k", 3191 16384, g_raid3_uma_ctor, g_raid3_uma_dtor, NULL, NULL, 3192 UMA_ALIGN_PTR, 0); 3193 sc->sc_zones[G_RAID3_ZONE_16K].sz_inuse = 0; 3194 sc->sc_zones[G_RAID3_ZONE_16K].sz_max = g_raid3_n16k; 3195 sc->sc_zones[G_RAID3_ZONE_16K].sz_requested = 3196 sc->sc_zones[G_RAID3_ZONE_16K].sz_failed = 0; 3197 sc->sc_zones[G_RAID3_ZONE_4K].sz_zone = uma_zcreate("gr3:4k", 3198 4096, g_raid3_uma_ctor, g_raid3_uma_dtor, NULL, NULL, 3199 UMA_ALIGN_PTR, 0); 3200 sc->sc_zones[G_RAID3_ZONE_4K].sz_inuse = 0; 3201 sc->sc_zones[G_RAID3_ZONE_4K].sz_max = g_raid3_n4k; 3202 sc->sc_zones[G_RAID3_ZONE_4K].sz_requested = 3203 sc->sc_zones[G_RAID3_ZONE_4K].sz_failed = 0; 3204 } 3205 3206 error = kproc_create(g_raid3_worker, sc, &sc->sc_worker, 0, 0, 3207 "g_raid3 %s", md->md_name); 3208 if (error != 0) { 3209 G_RAID3_DEBUG(1, "Cannot create kernel thread for %s.", 3210 sc->sc_name); 3211 if (!g_raid3_use_malloc) { 3212 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_64K].sz_zone); 3213 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_16K].sz_zone); 3214 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_4K].sz_zone); 3215 } 3216 g_destroy_geom(sc->sc_sync.ds_geom); 3217 mtx_destroy(&sc->sc_events_mtx); 3218 mtx_destroy(&sc->sc_queue_mtx); 3219 sx_destroy(&sc->sc_lock); 3220 g_destroy_geom(sc->sc_geom); 3221 free(sc->sc_disks, M_RAID3); 3222 free(sc, M_RAID3); 3223 return (NULL); 3224 } 3225 3226 G_RAID3_DEBUG(1, "Device %s created (%u components, id=%u).", 3227 sc->sc_name, sc->sc_ndisks, sc->sc_id); 3228 3229 sc->sc_rootmount = root_mount_hold("GRAID3"); 3230 G_RAID3_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 3231 3232 /* 3233 * Run timeout. 3234 */ 3235 timeout = atomic_load_acq_int(&g_raid3_timeout); 3236 callout_reset(&sc->sc_callout, timeout * hz, g_raid3_go, sc); 3237 return (sc->sc_geom); 3238 } 3239 3240 int 3241 g_raid3_destroy(struct g_raid3_softc *sc, int how) 3242 { 3243 struct g_provider *pp; 3244 3245 g_topology_assert_not(); 3246 if (sc == NULL) 3247 return (ENXIO); 3248 sx_assert(&sc->sc_lock, SX_XLOCKED); 3249 3250 pp = sc->sc_provider; 3251 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 3252 switch (how) { 3253 case G_RAID3_DESTROY_SOFT: 3254 G_RAID3_DEBUG(1, 3255 "Device %s is still open (r%dw%de%d).", pp->name, 3256 pp->acr, pp->acw, pp->ace); 3257 return (EBUSY); 3258 case G_RAID3_DESTROY_DELAYED: 3259 G_RAID3_DEBUG(1, 3260 "Device %s will be destroyed on last close.", 3261 pp->name); 3262 if (sc->sc_syncdisk != NULL) 3263 g_raid3_sync_stop(sc, 1); 3264 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROYING; 3265 return (EBUSY); 3266 case G_RAID3_DESTROY_HARD: 3267 G_RAID3_DEBUG(1, "Device %s is still open, so it " 3268 "can't be definitely removed.", pp->name); 3269 break; 3270 } 3271 } 3272 3273 g_topology_lock(); 3274 if (sc->sc_geom->softc == NULL) { 3275 g_topology_unlock(); 3276 return (0); 3277 } 3278 sc->sc_geom->softc = NULL; 3279 sc->sc_sync.ds_geom->softc = NULL; 3280 g_topology_unlock(); 3281 3282 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 3283 sc->sc_flags |= G_RAID3_DEVICE_FLAG_WAIT; 3284 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3285 sx_xunlock(&sc->sc_lock); 3286 mtx_lock(&sc->sc_queue_mtx); 3287 wakeup(sc); 3288 wakeup(&sc->sc_queue); 3289 mtx_unlock(&sc->sc_queue_mtx); 3290 G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3291 while (sc->sc_worker != NULL) 3292 tsleep(&sc->sc_worker, PRIBIO, "r3:destroy", hz / 5); 3293 G_RAID3_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3294 sx_xlock(&sc->sc_lock); 3295 g_raid3_destroy_device(sc); 3296 free(sc->sc_disks, M_RAID3); 3297 free(sc, M_RAID3); 3298 return (0); 3299 } 3300 3301 static void 3302 g_raid3_taste_orphan(struct g_consumer *cp) 3303 { 3304 3305 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3306 cp->provider->name)); 3307 } 3308 3309 static struct g_geom * 3310 g_raid3_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3311 { 3312 struct g_raid3_metadata md; 3313 struct g_raid3_softc *sc; 3314 struct g_consumer *cp; 3315 struct g_geom *gp; 3316 int error; 3317 3318 g_topology_assert(); 3319 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3320 G_RAID3_DEBUG(2, "Tasting %s.", pp->name); 3321 3322 gp = g_new_geomf(mp, "raid3:taste"); 3323 /* This orphan function should be never called. */ 3324 gp->orphan = g_raid3_taste_orphan; 3325 cp = g_new_consumer(gp); 3326 g_attach(cp, pp); 3327 error = g_raid3_read_metadata(cp, &md); 3328 g_detach(cp); 3329 g_destroy_consumer(cp); 3330 g_destroy_geom(gp); 3331 if (error != 0) 3332 return (NULL); 3333 gp = NULL; 3334 3335 if (md.md_provider[0] != '\0' && 3336 !g_compare_names(md.md_provider, pp->name)) 3337 return (NULL); 3338 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3339 return (NULL); 3340 if (g_raid3_debug >= 2) 3341 raid3_metadata_dump(&md); 3342 3343 /* 3344 * Let's check if device already exists. 3345 */ 3346 sc = NULL; 3347 LIST_FOREACH(gp, &mp->geom, geom) { 3348 sc = gp->softc; 3349 if (sc == NULL) 3350 continue; 3351 if (sc->sc_sync.ds_geom == gp) 3352 continue; 3353 if (strcmp(md.md_name, sc->sc_name) != 0) 3354 continue; 3355 if (md.md_id != sc->sc_id) { 3356 G_RAID3_DEBUG(0, "Device %s already configured.", 3357 sc->sc_name); 3358 return (NULL); 3359 } 3360 break; 3361 } 3362 if (gp == NULL) { 3363 gp = g_raid3_create(mp, &md); 3364 if (gp == NULL) { 3365 G_RAID3_DEBUG(0, "Cannot create device %s.", 3366 md.md_name); 3367 return (NULL); 3368 } 3369 sc = gp->softc; 3370 } 3371 G_RAID3_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3372 g_topology_unlock(); 3373 sx_xlock(&sc->sc_lock); 3374 error = g_raid3_add_disk(sc, pp, &md); 3375 if (error != 0) { 3376 G_RAID3_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3377 pp->name, gp->name, error); 3378 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NODISK) == 3379 sc->sc_ndisks) { 3380 g_cancel_event(sc); 3381 g_raid3_destroy(sc, G_RAID3_DESTROY_HARD); 3382 g_topology_lock(); 3383 return (NULL); 3384 } 3385 gp = NULL; 3386 } 3387 sx_xunlock(&sc->sc_lock); 3388 g_topology_lock(); 3389 return (gp); 3390 } 3391 3392 static int 3393 g_raid3_destroy_geom(struct gctl_req *req __unused, struct g_class *mp __unused, 3394 struct g_geom *gp) 3395 { 3396 struct g_raid3_softc *sc; 3397 int error; 3398 3399 g_topology_unlock(); 3400 sc = gp->softc; 3401 sx_xlock(&sc->sc_lock); 3402 g_cancel_event(sc); 3403 error = g_raid3_destroy(gp->softc, G_RAID3_DESTROY_SOFT); 3404 if (error != 0) 3405 sx_xunlock(&sc->sc_lock); 3406 g_topology_lock(); 3407 return (error); 3408 } 3409 3410 static void 3411 g_raid3_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3412 struct g_consumer *cp, struct g_provider *pp) 3413 { 3414 struct g_raid3_softc *sc; 3415 3416 g_topology_assert(); 3417 3418 sc = gp->softc; 3419 if (sc == NULL) 3420 return; 3421 /* Skip synchronization geom. */ 3422 if (gp == sc->sc_sync.ds_geom) 3423 return; 3424 if (pp != NULL) { 3425 /* Nothing here. */ 3426 } else if (cp != NULL) { 3427 struct g_raid3_disk *disk; 3428 3429 disk = cp->private; 3430 if (disk == NULL) 3431 return; 3432 g_topology_unlock(); 3433 sx_xlock(&sc->sc_lock); 3434 sbuf_printf(sb, "%s<Type>", indent); 3435 if (disk->d_no == sc->sc_ndisks - 1) 3436 sbuf_printf(sb, "PARITY"); 3437 else 3438 sbuf_printf(sb, "DATA"); 3439 sbuf_printf(sb, "</Type>\n"); 3440 sbuf_printf(sb, "%s<Number>%u</Number>\n", indent, 3441 (u_int)disk->d_no); 3442 if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 3443 sbuf_printf(sb, "%s<Synchronized>", indent); 3444 if (disk->d_sync.ds_offset == 0) 3445 sbuf_printf(sb, "0%%"); 3446 else { 3447 sbuf_printf(sb, "%u%%", 3448 (u_int)((disk->d_sync.ds_offset * 100) / 3449 (sc->sc_mediasize / (sc->sc_ndisks - 1)))); 3450 } 3451 sbuf_printf(sb, "</Synchronized>\n"); 3452 if (disk->d_sync.ds_offset > 0) { 3453 sbuf_printf(sb, "%s<BytesSynced>%jd" 3454 "</BytesSynced>\n", indent, 3455 (intmax_t)disk->d_sync.ds_offset); 3456 } 3457 } 3458 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3459 disk->d_sync.ds_syncid); 3460 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, disk->d_genid); 3461 sbuf_printf(sb, "%s<Flags>", indent); 3462 if (disk->d_flags == 0) 3463 sbuf_printf(sb, "NONE"); 3464 else { 3465 int first = 1; 3466 3467 #define ADD_FLAG(flag, name) do { \ 3468 if ((disk->d_flags & (flag)) != 0) { \ 3469 if (!first) \ 3470 sbuf_printf(sb, ", "); \ 3471 else \ 3472 first = 0; \ 3473 sbuf_printf(sb, name); \ 3474 } \ 3475 } while (0) 3476 ADD_FLAG(G_RAID3_DISK_FLAG_DIRTY, "DIRTY"); 3477 ADD_FLAG(G_RAID3_DISK_FLAG_HARDCODED, "HARDCODED"); 3478 ADD_FLAG(G_RAID3_DISK_FLAG_SYNCHRONIZING, 3479 "SYNCHRONIZING"); 3480 ADD_FLAG(G_RAID3_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3481 ADD_FLAG(G_RAID3_DISK_FLAG_BROKEN, "BROKEN"); 3482 #undef ADD_FLAG 3483 } 3484 sbuf_printf(sb, "</Flags>\n"); 3485 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3486 g_raid3_disk_state2str(disk->d_state)); 3487 sx_xunlock(&sc->sc_lock); 3488 g_topology_lock(); 3489 } else { 3490 g_topology_unlock(); 3491 sx_xlock(&sc->sc_lock); 3492 if (!g_raid3_use_malloc) { 3493 sbuf_printf(sb, 3494 "%s<Zone4kRequested>%u</Zone4kRequested>\n", indent, 3495 sc->sc_zones[G_RAID3_ZONE_4K].sz_requested); 3496 sbuf_printf(sb, 3497 "%s<Zone4kFailed>%u</Zone4kFailed>\n", indent, 3498 sc->sc_zones[G_RAID3_ZONE_4K].sz_failed); 3499 sbuf_printf(sb, 3500 "%s<Zone16kRequested>%u</Zone16kRequested>\n", indent, 3501 sc->sc_zones[G_RAID3_ZONE_16K].sz_requested); 3502 sbuf_printf(sb, 3503 "%s<Zone16kFailed>%u</Zone16kFailed>\n", indent, 3504 sc->sc_zones[G_RAID3_ZONE_16K].sz_failed); 3505 sbuf_printf(sb, 3506 "%s<Zone64kRequested>%u</Zone64kRequested>\n", indent, 3507 sc->sc_zones[G_RAID3_ZONE_64K].sz_requested); 3508 sbuf_printf(sb, 3509 "%s<Zone64kFailed>%u</Zone64kFailed>\n", indent, 3510 sc->sc_zones[G_RAID3_ZONE_64K].sz_failed); 3511 } 3512 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3513 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3514 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3515 sbuf_printf(sb, "%s<Flags>", indent); 3516 if (sc->sc_flags == 0) 3517 sbuf_printf(sb, "NONE"); 3518 else { 3519 int first = 1; 3520 3521 #define ADD_FLAG(flag, name) do { \ 3522 if ((sc->sc_flags & (flag)) != 0) { \ 3523 if (!first) \ 3524 sbuf_printf(sb, ", "); \ 3525 else \ 3526 first = 0; \ 3527 sbuf_printf(sb, name); \ 3528 } \ 3529 } while (0) 3530 ADD_FLAG(G_RAID3_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3531 ADD_FLAG(G_RAID3_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3532 ADD_FLAG(G_RAID3_DEVICE_FLAG_ROUND_ROBIN, 3533 "ROUND-ROBIN"); 3534 ADD_FLAG(G_RAID3_DEVICE_FLAG_VERIFY, "VERIFY"); 3535 #undef ADD_FLAG 3536 } 3537 sbuf_printf(sb, "</Flags>\n"); 3538 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3539 sc->sc_ndisks); 3540 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3541 g_raid3_device_state2str(sc->sc_state)); 3542 sx_xunlock(&sc->sc_lock); 3543 g_topology_lock(); 3544 } 3545 } 3546 3547 static void 3548 g_raid3_shutdown_post_sync(void *arg, int howto) 3549 { 3550 struct g_class *mp; 3551 struct g_geom *gp, *gp2; 3552 struct g_raid3_softc *sc; 3553 int error; 3554 3555 mp = arg; 3556 DROP_GIANT(); 3557 g_topology_lock(); 3558 g_raid3_shutdown = 1; 3559 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3560 if ((sc = gp->softc) == NULL) 3561 continue; 3562 /* Skip synchronization geom. */ 3563 if (gp == sc->sc_sync.ds_geom) 3564 continue; 3565 g_topology_unlock(); 3566 sx_xlock(&sc->sc_lock); 3567 g_raid3_idle(sc, -1); 3568 g_cancel_event(sc); 3569 error = g_raid3_destroy(sc, G_RAID3_DESTROY_DELAYED); 3570 if (error != 0) 3571 sx_xunlock(&sc->sc_lock); 3572 g_topology_lock(); 3573 } 3574 g_topology_unlock(); 3575 PICKUP_GIANT(); 3576 } 3577 3578 static void 3579 g_raid3_init(struct g_class *mp) 3580 { 3581 3582 g_raid3_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3583 g_raid3_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3584 if (g_raid3_post_sync == NULL) 3585 G_RAID3_DEBUG(0, "Warning! Cannot register shutdown event."); 3586 } 3587 3588 static void 3589 g_raid3_fini(struct g_class *mp) 3590 { 3591 3592 if (g_raid3_post_sync != NULL) 3593 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid3_post_sync); 3594 } 3595 3596 DECLARE_GEOM_CLASS(g_raid3_class, g_raid3); 3597