1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2004-2006 Pawel Jakub Dawidek <pjd@FreeBSD.org> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 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 <geom/geom_dbg.h> 45 #include <sys/proc.h> 46 #include <sys/kthread.h> 47 #include <sys/sched.h> 48 #include <geom/raid3/g_raid3.h> 49 50 FEATURE(geom_raid3, "GEOM RAID-3 functionality"); 51 52 static MALLOC_DEFINE(M_RAID3, "raid3_data", "GEOM_RAID3 Data"); 53 54 SYSCTL_DECL(_kern_geom); 55 static SYSCTL_NODE(_kern_geom, OID_AUTO, raid3, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 56 "GEOM_RAID3 stuff"); 57 u_int g_raid3_debug = 0; 58 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, debug, CTLFLAG_RWTUN, &g_raid3_debug, 0, 59 "Debug level"); 60 static u_int g_raid3_timeout = 4; 61 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, timeout, CTLFLAG_RWTUN, &g_raid3_timeout, 62 0, "Time to wait on all raid3 components"); 63 static u_int g_raid3_idletime = 5; 64 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, idletime, CTLFLAG_RWTUN, 65 &g_raid3_idletime, 0, "Mark components as clean when idling"); 66 static u_int g_raid3_disconnect_on_failure = 1; 67 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN, 68 &g_raid3_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 69 static u_int g_raid3_syncreqs = 2; 70 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, sync_requests, CTLFLAG_RDTUN, 71 &g_raid3_syncreqs, 0, "Parallel synchronization I/O requests."); 72 static u_int g_raid3_use_malloc = 0; 73 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, use_malloc, CTLFLAG_RDTUN, 74 &g_raid3_use_malloc, 0, "Use malloc(9) instead of uma(9)."); 75 76 static u_int g_raid3_n64k = 50; 77 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n64k, CTLFLAG_RDTUN, &g_raid3_n64k, 0, 78 "Maximum number of 64kB allocations"); 79 static u_int g_raid3_n16k = 200; 80 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n16k, CTLFLAG_RDTUN, &g_raid3_n16k, 0, 81 "Maximum number of 16kB allocations"); 82 static u_int g_raid3_n4k = 1200; 83 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n4k, CTLFLAG_RDTUN, &g_raid3_n4k, 0, 84 "Maximum number of 4kB allocations"); 85 86 static SYSCTL_NODE(_kern_geom_raid3, OID_AUTO, stat, 87 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 88 "GEOM_RAID3 statistics"); 89 static u_int g_raid3_parity_mismatch = 0; 90 SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, parity_mismatch, CTLFLAG_RD, 91 &g_raid3_parity_mismatch, 0, "Number of failures in VERIFY mode"); 92 93 #define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 94 G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 95 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 96 G_RAID3_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 97 } while (0) 98 99 static eventhandler_tag g_raid3_post_sync = NULL; 100 static int g_raid3_shutdown = 0; 101 102 static int g_raid3_destroy_geom(struct gctl_req *req, struct g_class *mp, 103 struct g_geom *gp); 104 static g_taste_t g_raid3_taste; 105 static void g_raid3_init(struct g_class *mp); 106 static void g_raid3_fini(struct g_class *mp); 107 108 struct g_class g_raid3_class = { 109 .name = G_RAID3_CLASS_NAME, 110 .version = G_VERSION, 111 .ctlreq = g_raid3_config, 112 .taste = g_raid3_taste, 113 .destroy_geom = g_raid3_destroy_geom, 114 .init = g_raid3_init, 115 .fini = g_raid3_fini 116 }; 117 118 static void g_raid3_destroy_provider(struct g_raid3_softc *sc); 119 static int g_raid3_update_disk(struct g_raid3_disk *disk, u_int state); 120 static void g_raid3_update_device(struct g_raid3_softc *sc, boolean_t force); 121 static void g_raid3_dumpconf(struct sbuf *sb, const char *indent, 122 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 123 static void g_raid3_sync_stop(struct g_raid3_softc *sc, int type); 124 static int g_raid3_register_request(struct bio *pbp); 125 static void g_raid3_sync_release(struct g_raid3_softc *sc); 126 static void g_raid3_timeout_drain(struct g_raid3_softc *sc); 127 128 static const char * 129 g_raid3_disk_state2str(int state) 130 { 131 132 switch (state) { 133 case G_RAID3_DISK_STATE_NODISK: 134 return ("NODISK"); 135 case G_RAID3_DISK_STATE_NONE: 136 return ("NONE"); 137 case G_RAID3_DISK_STATE_NEW: 138 return ("NEW"); 139 case G_RAID3_DISK_STATE_ACTIVE: 140 return ("ACTIVE"); 141 case G_RAID3_DISK_STATE_STALE: 142 return ("STALE"); 143 case G_RAID3_DISK_STATE_SYNCHRONIZING: 144 return ("SYNCHRONIZING"); 145 case G_RAID3_DISK_STATE_DISCONNECTED: 146 return ("DISCONNECTED"); 147 default: 148 return ("INVALID"); 149 } 150 } 151 152 static const char * 153 g_raid3_device_state2str(int state) 154 { 155 156 switch (state) { 157 case G_RAID3_DEVICE_STATE_STARTING: 158 return ("STARTING"); 159 case G_RAID3_DEVICE_STATE_DEGRADED: 160 return ("DEGRADED"); 161 case G_RAID3_DEVICE_STATE_COMPLETE: 162 return ("COMPLETE"); 163 default: 164 return ("INVALID"); 165 } 166 } 167 168 const char * 169 g_raid3_get_diskname(struct g_raid3_disk *disk) 170 { 171 172 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 173 return ("[unknown]"); 174 return (disk->d_name); 175 } 176 177 static void * 178 g_raid3_alloc(struct g_raid3_softc *sc, size_t size, int flags) 179 { 180 void *ptr; 181 enum g_raid3_zones zone; 182 183 if (g_raid3_use_malloc || 184 (zone = g_raid3_zone(size)) == G_RAID3_NUM_ZONES) 185 ptr = malloc(size, M_RAID3, flags); 186 else { 187 ptr = uma_zalloc_arg(sc->sc_zones[zone].sz_zone, 188 &sc->sc_zones[zone], flags); 189 sc->sc_zones[zone].sz_requested++; 190 if (ptr == NULL) 191 sc->sc_zones[zone].sz_failed++; 192 } 193 return (ptr); 194 } 195 196 static void 197 g_raid3_free(struct g_raid3_softc *sc, void *ptr, size_t size) 198 { 199 enum g_raid3_zones zone; 200 201 if (g_raid3_use_malloc || 202 (zone = g_raid3_zone(size)) == G_RAID3_NUM_ZONES) 203 free(ptr, M_RAID3); 204 else { 205 uma_zfree_arg(sc->sc_zones[zone].sz_zone, 206 ptr, &sc->sc_zones[zone]); 207 } 208 } 209 210 static int 211 g_raid3_uma_ctor(void *mem, int size, void *arg, int flags) 212 { 213 struct g_raid3_zone *sz = arg; 214 215 if (sz->sz_max > 0 && sz->sz_inuse == sz->sz_max) 216 return (ENOMEM); 217 sz->sz_inuse++; 218 return (0); 219 } 220 221 static void 222 g_raid3_uma_dtor(void *mem, int size, void *arg) 223 { 224 struct g_raid3_zone *sz = arg; 225 226 sz->sz_inuse--; 227 } 228 229 #define g_raid3_xor(src, dst, size) \ 230 _g_raid3_xor((uint64_t *)(src), \ 231 (uint64_t *)(dst), (size_t)size) 232 static void 233 _g_raid3_xor(uint64_t *src, uint64_t *dst, size_t size) 234 { 235 236 KASSERT((size % 128) == 0, ("Invalid size: %zu.", size)); 237 for (; size > 0; size -= 128) { 238 *dst++ ^= (*src++); 239 *dst++ ^= (*src++); 240 *dst++ ^= (*src++); 241 *dst++ ^= (*src++); 242 *dst++ ^= (*src++); 243 *dst++ ^= (*src++); 244 *dst++ ^= (*src++); 245 *dst++ ^= (*src++); 246 *dst++ ^= (*src++); 247 *dst++ ^= (*src++); 248 *dst++ ^= (*src++); 249 *dst++ ^= (*src++); 250 *dst++ ^= (*src++); 251 *dst++ ^= (*src++); 252 *dst++ ^= (*src++); 253 *dst++ ^= (*src++); 254 } 255 } 256 257 static int 258 g_raid3_is_zero(struct bio *bp) 259 { 260 static const uint64_t zeros[] = { 261 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 262 }; 263 u_char *addr; 264 ssize_t size; 265 266 size = bp->bio_length; 267 addr = (u_char *)bp->bio_data; 268 for (; size > 0; size -= sizeof(zeros), addr += sizeof(zeros)) { 269 if (bcmp(addr, zeros, sizeof(zeros)) != 0) 270 return (0); 271 } 272 return (1); 273 } 274 275 /* 276 * --- Events handling functions --- 277 * Events in geom_raid3 are used to maintain disks and device status 278 * from one thread to simplify locking. 279 */ 280 static void 281 g_raid3_event_free(struct g_raid3_event *ep) 282 { 283 284 free(ep, M_RAID3); 285 } 286 287 static int 288 g_raid3_event_dispatch(struct g_raid3_event *ep, void *arg, int state, 289 int flags) 290 { 291 struct g_raid3_softc *sc; 292 struct g_raid3_disk *disk; 293 int error; 294 295 G_RAID3_DEBUG(4, "%s: Sending event %p.", __func__, ep); 296 if ((flags & G_RAID3_EVENT_DEVICE) != 0) { 297 disk = NULL; 298 sc = arg; 299 } else { 300 disk = arg; 301 sc = disk->d_softc; 302 } 303 ep->e_disk = disk; 304 ep->e_state = state; 305 ep->e_flags = flags; 306 ep->e_error = 0; 307 mtx_lock(&sc->sc_events_mtx); 308 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 309 mtx_unlock(&sc->sc_events_mtx); 310 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc); 311 mtx_lock(&sc->sc_queue_mtx); 312 wakeup(sc); 313 wakeup(&sc->sc_queue); 314 mtx_unlock(&sc->sc_queue_mtx); 315 if ((flags & G_RAID3_EVENT_DONTWAIT) != 0) 316 return (0); 317 sx_assert(&sc->sc_lock, SX_XLOCKED); 318 G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 319 sx_xunlock(&sc->sc_lock); 320 while ((ep->e_flags & G_RAID3_EVENT_DONE) == 0) { 321 mtx_lock(&sc->sc_events_mtx); 322 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "r3:event", 323 hz * 5); 324 } 325 error = ep->e_error; 326 g_raid3_event_free(ep); 327 sx_xlock(&sc->sc_lock); 328 return (error); 329 } 330 331 int 332 g_raid3_event_send(void *arg, int state, int flags) 333 { 334 struct g_raid3_event *ep; 335 336 ep = malloc(sizeof(*ep), M_RAID3, M_WAITOK); 337 return (g_raid3_event_dispatch(ep, arg, state, flags)); 338 } 339 340 static struct g_raid3_event * 341 g_raid3_event_get(struct g_raid3_softc *sc) 342 { 343 struct g_raid3_event *ep; 344 345 mtx_lock(&sc->sc_events_mtx); 346 ep = TAILQ_FIRST(&sc->sc_events); 347 mtx_unlock(&sc->sc_events_mtx); 348 return (ep); 349 } 350 351 static void 352 g_raid3_event_remove(struct g_raid3_softc *sc, struct g_raid3_event *ep) 353 { 354 355 mtx_lock(&sc->sc_events_mtx); 356 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 357 mtx_unlock(&sc->sc_events_mtx); 358 } 359 360 static void 361 g_raid3_event_cancel(struct g_raid3_disk *disk) 362 { 363 struct g_raid3_softc *sc; 364 struct g_raid3_event *ep, *tmpep; 365 366 sc = disk->d_softc; 367 sx_assert(&sc->sc_lock, SX_XLOCKED); 368 369 mtx_lock(&sc->sc_events_mtx); 370 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 371 if ((ep->e_flags & G_RAID3_EVENT_DEVICE) != 0) 372 continue; 373 if (ep->e_disk != disk) 374 continue; 375 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 376 if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0) 377 g_raid3_event_free(ep); 378 else { 379 ep->e_error = ECANCELED; 380 wakeup(ep); 381 } 382 } 383 mtx_unlock(&sc->sc_events_mtx); 384 } 385 386 /* 387 * Return the number of disks in the given state. 388 * If state is equal to -1, count all connected disks. 389 */ 390 u_int 391 g_raid3_ndisks(struct g_raid3_softc *sc, int state) 392 { 393 struct g_raid3_disk *disk; 394 u_int n, ndisks; 395 396 sx_assert(&sc->sc_lock, SX_LOCKED); 397 398 for (n = ndisks = 0; n < sc->sc_ndisks; n++) { 399 disk = &sc->sc_disks[n]; 400 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 401 continue; 402 if (state == -1 || disk->d_state == state) 403 ndisks++; 404 } 405 return (ndisks); 406 } 407 408 static u_int 409 g_raid3_nrequests(struct g_raid3_softc *sc, struct g_consumer *cp) 410 { 411 struct bio *bp; 412 u_int nreqs = 0; 413 414 mtx_lock(&sc->sc_queue_mtx); 415 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 416 if (bp->bio_from == cp) 417 nreqs++; 418 } 419 mtx_unlock(&sc->sc_queue_mtx); 420 return (nreqs); 421 } 422 423 static int 424 g_raid3_is_busy(struct g_raid3_softc *sc, struct g_consumer *cp) 425 { 426 427 if (cp->index > 0) { 428 G_RAID3_DEBUG(2, 429 "I/O requests for %s exist, can't destroy it now.", 430 cp->provider->name); 431 return (1); 432 } 433 if (g_raid3_nrequests(sc, cp) > 0) { 434 G_RAID3_DEBUG(2, 435 "I/O requests for %s in queue, can't destroy it now.", 436 cp->provider->name); 437 return (1); 438 } 439 return (0); 440 } 441 442 static void 443 g_raid3_destroy_consumer(void *arg, int flags __unused) 444 { 445 struct g_consumer *cp; 446 447 g_topology_assert(); 448 449 cp = arg; 450 G_RAID3_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 451 g_detach(cp); 452 g_destroy_consumer(cp); 453 } 454 455 static void 456 g_raid3_kill_consumer(struct g_raid3_softc *sc, struct g_consumer *cp) 457 { 458 struct g_provider *pp; 459 int retaste_wait; 460 461 g_topology_assert(); 462 463 cp->private = NULL; 464 if (g_raid3_is_busy(sc, cp)) 465 return; 466 G_RAID3_DEBUG(2, "Consumer %s destroyed.", cp->provider->name); 467 pp = cp->provider; 468 retaste_wait = 0; 469 if (cp->acw == 1) { 470 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 471 retaste_wait = 1; 472 } 473 G_RAID3_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 474 -cp->acw, -cp->ace, 0); 475 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 476 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 477 if (retaste_wait) { 478 /* 479 * After retaste event was send (inside g_access()), we can send 480 * event to detach and destroy consumer. 481 * A class, which has consumer to the given provider connected 482 * will not receive retaste event for the provider. 483 * This is the way how I ignore retaste events when I close 484 * consumers opened for write: I detach and destroy consumer 485 * after retaste event is sent. 486 */ 487 g_post_event(g_raid3_destroy_consumer, cp, M_WAITOK, NULL); 488 return; 489 } 490 G_RAID3_DEBUG(1, "Consumer %s destroyed.", pp->name); 491 g_detach(cp); 492 g_destroy_consumer(cp); 493 } 494 495 static int 496 g_raid3_connect_disk(struct g_raid3_disk *disk, struct g_provider *pp) 497 { 498 struct g_consumer *cp; 499 int error; 500 501 g_topology_assert_not(); 502 KASSERT(disk->d_consumer == NULL, 503 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 504 505 g_topology_lock(); 506 cp = g_new_consumer(disk->d_softc->sc_geom); 507 error = g_attach(cp, pp); 508 if (error != 0) { 509 g_destroy_consumer(cp); 510 g_topology_unlock(); 511 return (error); 512 } 513 error = g_access(cp, 1, 1, 1); 514 g_topology_unlock(); 515 if (error != 0) { 516 g_detach(cp); 517 g_destroy_consumer(cp); 518 G_RAID3_DEBUG(0, "Cannot open consumer %s (error=%d).", 519 pp->name, error); 520 return (error); 521 } 522 disk->d_consumer = cp; 523 disk->d_consumer->private = disk; 524 disk->d_consumer->index = 0; 525 G_RAID3_DEBUG(2, "Disk %s connected.", g_raid3_get_diskname(disk)); 526 return (0); 527 } 528 529 static void 530 g_raid3_disconnect_consumer(struct g_raid3_softc *sc, struct g_consumer *cp) 531 { 532 533 g_topology_assert(); 534 535 if (cp == NULL) 536 return; 537 if (cp->provider != NULL) 538 g_raid3_kill_consumer(sc, cp); 539 else 540 g_destroy_consumer(cp); 541 } 542 543 /* 544 * Initialize disk. This means allocate memory, create consumer, attach it 545 * to the provider and open access (r1w1e1) to it. 546 */ 547 static struct g_raid3_disk * 548 g_raid3_init_disk(struct g_raid3_softc *sc, struct g_provider *pp, 549 struct g_raid3_metadata *md, int *errorp) 550 { 551 struct g_raid3_disk *disk; 552 int error; 553 554 disk = &sc->sc_disks[md->md_no]; 555 error = g_raid3_connect_disk(disk, pp); 556 if (error != 0) { 557 if (errorp != NULL) 558 *errorp = error; 559 return (NULL); 560 } 561 disk->d_state = G_RAID3_DISK_STATE_NONE; 562 disk->d_flags = md->md_dflags; 563 if (md->md_provider[0] != '\0') 564 disk->d_flags |= G_RAID3_DISK_FLAG_HARDCODED; 565 disk->d_sync.ds_consumer = NULL; 566 disk->d_sync.ds_offset = md->md_sync_offset; 567 disk->d_sync.ds_offset_done = md->md_sync_offset; 568 disk->d_genid = md->md_genid; 569 disk->d_sync.ds_syncid = md->md_syncid; 570 if (errorp != NULL) 571 *errorp = 0; 572 return (disk); 573 } 574 575 static void 576 g_raid3_destroy_disk(struct g_raid3_disk *disk) 577 { 578 struct g_raid3_softc *sc; 579 580 g_topology_assert_not(); 581 sc = disk->d_softc; 582 sx_assert(&sc->sc_lock, SX_XLOCKED); 583 584 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 585 return; 586 g_raid3_event_cancel(disk); 587 switch (disk->d_state) { 588 case G_RAID3_DISK_STATE_SYNCHRONIZING: 589 if (sc->sc_syncdisk != NULL) 590 g_raid3_sync_stop(sc, 1); 591 /* FALLTHROUGH */ 592 case G_RAID3_DISK_STATE_NEW: 593 case G_RAID3_DISK_STATE_STALE: 594 case G_RAID3_DISK_STATE_ACTIVE: 595 g_topology_lock(); 596 g_raid3_disconnect_consumer(sc, disk->d_consumer); 597 g_topology_unlock(); 598 disk->d_consumer = NULL; 599 break; 600 default: 601 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 602 g_raid3_get_diskname(disk), 603 g_raid3_disk_state2str(disk->d_state))); 604 } 605 disk->d_state = G_RAID3_DISK_STATE_NODISK; 606 } 607 608 static void 609 g_raid3_destroy_device(struct g_raid3_softc *sc) 610 { 611 struct g_raid3_event *ep; 612 struct g_raid3_disk *disk; 613 struct g_geom *gp; 614 struct g_consumer *cp; 615 u_int n; 616 617 g_topology_assert_not(); 618 sx_assert(&sc->sc_lock, SX_XLOCKED); 619 620 gp = sc->sc_geom; 621 if (sc->sc_provider != NULL) 622 g_raid3_destroy_provider(sc); 623 for (n = 0; n < sc->sc_ndisks; n++) { 624 disk = &sc->sc_disks[n]; 625 if (disk->d_state != G_RAID3_DISK_STATE_NODISK) { 626 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 627 g_raid3_update_metadata(disk); 628 g_raid3_destroy_disk(disk); 629 } 630 } 631 while ((ep = g_raid3_event_get(sc)) != NULL) { 632 g_raid3_event_remove(sc, ep); 633 if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0) 634 g_raid3_event_free(ep); 635 else { 636 ep->e_error = ECANCELED; 637 ep->e_flags |= G_RAID3_EVENT_DONE; 638 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, ep); 639 mtx_lock(&sc->sc_events_mtx); 640 wakeup(ep); 641 mtx_unlock(&sc->sc_events_mtx); 642 } 643 } 644 g_raid3_timeout_drain(sc); 645 cp = LIST_FIRST(&sc->sc_sync.ds_geom->consumer); 646 g_topology_lock(); 647 if (cp != NULL) 648 g_raid3_disconnect_consumer(sc, cp); 649 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 650 G_RAID3_DEBUG(0, "Device %s destroyed.", gp->name); 651 g_wither_geom(gp, ENXIO); 652 g_topology_unlock(); 653 if (!g_raid3_use_malloc) { 654 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_64K].sz_zone); 655 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_16K].sz_zone); 656 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_4K].sz_zone); 657 } 658 mtx_destroy(&sc->sc_queue_mtx); 659 mtx_destroy(&sc->sc_events_mtx); 660 sx_xunlock(&sc->sc_lock); 661 sx_destroy(&sc->sc_lock); 662 } 663 664 static void 665 g_raid3_orphan(struct g_consumer *cp) 666 { 667 struct g_raid3_disk *disk; 668 669 g_topology_assert(); 670 671 disk = cp->private; 672 if (disk == NULL) 673 return; 674 disk->d_softc->sc_bump_id = G_RAID3_BUMP_SYNCID; 675 g_raid3_event_send(disk, G_RAID3_DISK_STATE_DISCONNECTED, 676 G_RAID3_EVENT_DONTWAIT); 677 } 678 679 static int 680 g_raid3_write_metadata(struct g_raid3_disk *disk, struct g_raid3_metadata *md) 681 { 682 struct g_raid3_softc *sc; 683 struct g_consumer *cp; 684 off_t offset, length; 685 u_char *sector; 686 int error = 0; 687 688 g_topology_assert_not(); 689 sc = disk->d_softc; 690 sx_assert(&sc->sc_lock, SX_LOCKED); 691 692 cp = disk->d_consumer; 693 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 694 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 695 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 696 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 697 cp->acw, cp->ace)); 698 length = cp->provider->sectorsize; 699 offset = cp->provider->mediasize - length; 700 sector = malloc((size_t)length, M_RAID3, M_WAITOK | M_ZERO); 701 if (md != NULL) 702 raid3_metadata_encode(md, sector); 703 error = g_write_data(cp, offset, sector, length); 704 free(sector, M_RAID3); 705 if (error != 0) { 706 if ((disk->d_flags & G_RAID3_DISK_FLAG_BROKEN) == 0) { 707 G_RAID3_DEBUG(0, "Cannot write metadata on %s " 708 "(device=%s, error=%d).", 709 g_raid3_get_diskname(disk), sc->sc_name, error); 710 disk->d_flags |= G_RAID3_DISK_FLAG_BROKEN; 711 } else { 712 G_RAID3_DEBUG(1, "Cannot write metadata on %s " 713 "(device=%s, error=%d).", 714 g_raid3_get_diskname(disk), sc->sc_name, error); 715 } 716 if (g_raid3_disconnect_on_failure && 717 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 718 sc->sc_bump_id |= G_RAID3_BUMP_GENID; 719 g_raid3_event_send(disk, 720 G_RAID3_DISK_STATE_DISCONNECTED, 721 G_RAID3_EVENT_DONTWAIT); 722 } 723 } 724 return (error); 725 } 726 727 int 728 g_raid3_clear_metadata(struct g_raid3_disk *disk) 729 { 730 int error; 731 732 g_topology_assert_not(); 733 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 734 735 error = g_raid3_write_metadata(disk, NULL); 736 if (error == 0) { 737 G_RAID3_DEBUG(2, "Metadata on %s cleared.", 738 g_raid3_get_diskname(disk)); 739 } else { 740 G_RAID3_DEBUG(0, 741 "Cannot clear metadata on disk %s (error=%d).", 742 g_raid3_get_diskname(disk), error); 743 } 744 return (error); 745 } 746 747 void 748 g_raid3_fill_metadata(struct g_raid3_disk *disk, struct g_raid3_metadata *md) 749 { 750 struct g_raid3_softc *sc; 751 struct g_provider *pp; 752 753 bzero(md, sizeof(*md)); 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 = 769 disk->d_sync.ds_offset_done / (sc->sc_ndisks - 1); 770 } 771 if (disk->d_consumer != NULL && disk->d_consumer->provider != NULL) 772 pp = disk->d_consumer->provider; 773 else 774 pp = NULL; 775 if ((disk->d_flags & G_RAID3_DISK_FLAG_HARDCODED) != 0 && pp != NULL) 776 strlcpy(md->md_provider, pp->name, sizeof(md->md_provider)); 777 if (pp != NULL) 778 md->md_provsize = pp->mediasize; 779 } 780 781 void 782 g_raid3_update_metadata(struct g_raid3_disk *disk) 783 { 784 struct g_raid3_softc *sc __diagused; 785 struct g_raid3_metadata md; 786 int error; 787 788 g_topology_assert_not(); 789 sc = disk->d_softc; 790 sx_assert(&sc->sc_lock, SX_LOCKED); 791 792 g_raid3_fill_metadata(disk, &md); 793 error = g_raid3_write_metadata(disk, &md); 794 if (error == 0) { 795 G_RAID3_DEBUG(2, "Metadata on %s updated.", 796 g_raid3_get_diskname(disk)); 797 } else { 798 G_RAID3_DEBUG(0, 799 "Cannot update metadata on disk %s (error=%d).", 800 g_raid3_get_diskname(disk), error); 801 } 802 } 803 804 static void 805 g_raid3_bump_syncid(struct g_raid3_softc *sc) 806 { 807 struct g_raid3_disk *disk; 808 u_int n; 809 810 g_topology_assert_not(); 811 sx_assert(&sc->sc_lock, SX_XLOCKED); 812 KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) > 0, 813 ("%s called with no active disks (device=%s).", __func__, 814 sc->sc_name)); 815 816 sc->sc_syncid++; 817 G_RAID3_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 818 sc->sc_syncid); 819 for (n = 0; n < sc->sc_ndisks; n++) { 820 disk = &sc->sc_disks[n]; 821 if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 822 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 823 disk->d_sync.ds_syncid = sc->sc_syncid; 824 g_raid3_update_metadata(disk); 825 } 826 } 827 } 828 829 static void 830 g_raid3_bump_genid(struct g_raid3_softc *sc) 831 { 832 struct g_raid3_disk *disk; 833 u_int n; 834 835 g_topology_assert_not(); 836 sx_assert(&sc->sc_lock, SX_XLOCKED); 837 KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) > 0, 838 ("%s called with no active disks (device=%s).", __func__, 839 sc->sc_name)); 840 841 sc->sc_genid++; 842 G_RAID3_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 843 sc->sc_genid); 844 for (n = 0; n < sc->sc_ndisks; n++) { 845 disk = &sc->sc_disks[n]; 846 if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 847 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 848 disk->d_genid = sc->sc_genid; 849 g_raid3_update_metadata(disk); 850 } 851 } 852 } 853 854 static int 855 g_raid3_idle(struct g_raid3_softc *sc, int acw) 856 { 857 struct g_raid3_disk *disk; 858 u_int i; 859 int timeout; 860 861 g_topology_assert_not(); 862 sx_assert(&sc->sc_lock, SX_XLOCKED); 863 864 if (sc->sc_provider == NULL) 865 return (0); 866 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) != 0) 867 return (0); 868 if (sc->sc_idle) 869 return (0); 870 if (sc->sc_writes > 0) 871 return (0); 872 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 873 timeout = g_raid3_idletime - (time_uptime - sc->sc_last_write); 874 if (!g_raid3_shutdown && timeout > 0) 875 return (timeout); 876 } 877 sc->sc_idle = 1; 878 for (i = 0; i < sc->sc_ndisks; i++) { 879 disk = &sc->sc_disks[i]; 880 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) 881 continue; 882 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as clean.", 883 g_raid3_get_diskname(disk), sc->sc_name); 884 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 885 g_raid3_update_metadata(disk); 886 } 887 return (0); 888 } 889 890 static void 891 g_raid3_unidle(struct g_raid3_softc *sc) 892 { 893 struct g_raid3_disk *disk; 894 u_int i; 895 896 g_topology_assert_not(); 897 sx_assert(&sc->sc_lock, SX_XLOCKED); 898 899 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) != 0) 900 return; 901 sc->sc_idle = 0; 902 sc->sc_last_write = time_uptime; 903 for (i = 0; i < sc->sc_ndisks; i++) { 904 disk = &sc->sc_disks[i]; 905 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) 906 continue; 907 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as dirty.", 908 g_raid3_get_diskname(disk), sc->sc_name); 909 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY; 910 g_raid3_update_metadata(disk); 911 } 912 } 913 914 /* 915 * Treat bio_driver1 field in parent bio as list head and field bio_caller1 916 * in child bio as pointer to the next element on the list. 917 */ 918 #define G_RAID3_HEAD_BIO(pbp) (pbp)->bio_driver1 919 920 #define G_RAID3_NEXT_BIO(cbp) (cbp)->bio_caller1 921 922 #define G_RAID3_FOREACH_BIO(pbp, bp) \ 923 for ((bp) = G_RAID3_HEAD_BIO(pbp); (bp) != NULL; \ 924 (bp) = G_RAID3_NEXT_BIO(bp)) 925 926 #define G_RAID3_FOREACH_SAFE_BIO(pbp, bp, tmpbp) \ 927 for ((bp) = G_RAID3_HEAD_BIO(pbp); \ 928 (bp) != NULL && ((tmpbp) = G_RAID3_NEXT_BIO(bp), 1); \ 929 (bp) = (tmpbp)) 930 931 static void 932 g_raid3_init_bio(struct bio *pbp) 933 { 934 935 G_RAID3_HEAD_BIO(pbp) = NULL; 936 } 937 938 static void 939 g_raid3_remove_bio(struct bio *cbp) 940 { 941 struct bio *pbp, *bp; 942 943 pbp = cbp->bio_parent; 944 if (G_RAID3_HEAD_BIO(pbp) == cbp) 945 G_RAID3_HEAD_BIO(pbp) = G_RAID3_NEXT_BIO(cbp); 946 else { 947 G_RAID3_FOREACH_BIO(pbp, bp) { 948 if (G_RAID3_NEXT_BIO(bp) == cbp) { 949 G_RAID3_NEXT_BIO(bp) = G_RAID3_NEXT_BIO(cbp); 950 break; 951 } 952 } 953 } 954 G_RAID3_NEXT_BIO(cbp) = NULL; 955 } 956 957 static void 958 g_raid3_replace_bio(struct bio *sbp, struct bio *dbp) 959 { 960 struct bio *pbp, *bp; 961 962 g_raid3_remove_bio(sbp); 963 pbp = dbp->bio_parent; 964 G_RAID3_NEXT_BIO(sbp) = G_RAID3_NEXT_BIO(dbp); 965 if (G_RAID3_HEAD_BIO(pbp) == dbp) 966 G_RAID3_HEAD_BIO(pbp) = sbp; 967 else { 968 G_RAID3_FOREACH_BIO(pbp, bp) { 969 if (G_RAID3_NEXT_BIO(bp) == dbp) { 970 G_RAID3_NEXT_BIO(bp) = sbp; 971 break; 972 } 973 } 974 } 975 G_RAID3_NEXT_BIO(dbp) = NULL; 976 } 977 978 static void 979 g_raid3_destroy_bio(struct g_raid3_softc *sc, struct bio *cbp) 980 { 981 struct bio *bp, *pbp; 982 size_t size; 983 984 pbp = cbp->bio_parent; 985 pbp->bio_children--; 986 KASSERT(cbp->bio_data != NULL, ("NULL bio_data")); 987 size = pbp->bio_length / (sc->sc_ndisks - 1); 988 g_raid3_free(sc, cbp->bio_data, size); 989 if (G_RAID3_HEAD_BIO(pbp) == cbp) { 990 G_RAID3_HEAD_BIO(pbp) = G_RAID3_NEXT_BIO(cbp); 991 G_RAID3_NEXT_BIO(cbp) = NULL; 992 g_destroy_bio(cbp); 993 } else { 994 G_RAID3_FOREACH_BIO(pbp, bp) { 995 if (G_RAID3_NEXT_BIO(bp) == cbp) 996 break; 997 } 998 if (bp != NULL) { 999 KASSERT(G_RAID3_NEXT_BIO(bp) != NULL, 1000 ("NULL bp->bio_driver1")); 1001 G_RAID3_NEXT_BIO(bp) = G_RAID3_NEXT_BIO(cbp); 1002 G_RAID3_NEXT_BIO(cbp) = NULL; 1003 } 1004 g_destroy_bio(cbp); 1005 } 1006 } 1007 1008 static struct bio * 1009 g_raid3_clone_bio(struct g_raid3_softc *sc, struct bio *pbp) 1010 { 1011 struct bio *bp, *cbp; 1012 size_t size; 1013 int memflag; 1014 1015 cbp = g_clone_bio(pbp); 1016 if (cbp == NULL) 1017 return (NULL); 1018 size = pbp->bio_length / (sc->sc_ndisks - 1); 1019 if ((pbp->bio_cflags & G_RAID3_BIO_CFLAG_REGULAR) != 0) 1020 memflag = M_WAITOK; 1021 else 1022 memflag = M_NOWAIT; 1023 cbp->bio_data = g_raid3_alloc(sc, size, memflag); 1024 if (cbp->bio_data == NULL) { 1025 pbp->bio_children--; 1026 g_destroy_bio(cbp); 1027 return (NULL); 1028 } 1029 G_RAID3_NEXT_BIO(cbp) = NULL; 1030 if (G_RAID3_HEAD_BIO(pbp) == NULL) 1031 G_RAID3_HEAD_BIO(pbp) = cbp; 1032 else { 1033 G_RAID3_FOREACH_BIO(pbp, bp) { 1034 if (G_RAID3_NEXT_BIO(bp) == NULL) { 1035 G_RAID3_NEXT_BIO(bp) = cbp; 1036 break; 1037 } 1038 } 1039 } 1040 return (cbp); 1041 } 1042 1043 static void 1044 g_raid3_scatter(struct bio *pbp) 1045 { 1046 struct g_raid3_softc *sc; 1047 struct g_raid3_disk *disk; 1048 struct bio *bp, *cbp, *tmpbp; 1049 off_t atom, cadd, padd, left; 1050 int first; 1051 1052 sc = pbp->bio_to->geom->softc; 1053 bp = NULL; 1054 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_NOPARITY) == 0) { 1055 /* 1056 * Find bio for which we should calculate data. 1057 */ 1058 G_RAID3_FOREACH_BIO(pbp, cbp) { 1059 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) { 1060 bp = cbp; 1061 break; 1062 } 1063 } 1064 KASSERT(bp != NULL, ("NULL parity bio.")); 1065 } 1066 atom = sc->sc_sectorsize / (sc->sc_ndisks - 1); 1067 cadd = padd = 0; 1068 for (left = pbp->bio_length; left > 0; left -= sc->sc_sectorsize) { 1069 G_RAID3_FOREACH_BIO(pbp, cbp) { 1070 if (cbp == bp) 1071 continue; 1072 bcopy(pbp->bio_data + padd, cbp->bio_data + cadd, atom); 1073 padd += atom; 1074 } 1075 cadd += atom; 1076 } 1077 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_NOPARITY) == 0) { 1078 /* 1079 * Calculate parity. 1080 */ 1081 first = 1; 1082 G_RAID3_FOREACH_SAFE_BIO(pbp, cbp, tmpbp) { 1083 if (cbp == bp) 1084 continue; 1085 if (first) { 1086 bcopy(cbp->bio_data, bp->bio_data, 1087 bp->bio_length); 1088 first = 0; 1089 } else { 1090 g_raid3_xor(cbp->bio_data, bp->bio_data, 1091 bp->bio_length); 1092 } 1093 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_NODISK) != 0) 1094 g_raid3_destroy_bio(sc, cbp); 1095 } 1096 } 1097 G_RAID3_FOREACH_SAFE_BIO(pbp, cbp, tmpbp) { 1098 struct g_consumer *cp; 1099 1100 disk = cbp->bio_caller2; 1101 cp = disk->d_consumer; 1102 cbp->bio_to = cp->provider; 1103 G_RAID3_LOGREQ(3, cbp, "Sending request."); 1104 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1105 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1106 cp->acr, cp->acw, cp->ace)); 1107 cp->index++; 1108 sc->sc_writes++; 1109 g_io_request(cbp, cp); 1110 } 1111 } 1112 1113 static void 1114 g_raid3_gather(struct bio *pbp) 1115 { 1116 struct g_raid3_softc *sc; 1117 struct g_raid3_disk *disk; 1118 struct bio *xbp, *fbp, *cbp; 1119 off_t atom, cadd, padd, left; 1120 1121 sc = pbp->bio_to->geom->softc; 1122 /* 1123 * Find bio for which we have to calculate data. 1124 * While going through this path, check if all requests 1125 * succeeded, if not, deny whole request. 1126 * If we're in COMPLETE mode, we allow one request to fail, 1127 * so if we find one, we're sending it to the parity consumer. 1128 * If there are more failed requests, we deny whole request. 1129 */ 1130 xbp = fbp = NULL; 1131 G_RAID3_FOREACH_BIO(pbp, cbp) { 1132 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) { 1133 KASSERT(xbp == NULL, ("More than one parity bio.")); 1134 xbp = cbp; 1135 } 1136 if (cbp->bio_error == 0) 1137 continue; 1138 /* 1139 * Found failed request. 1140 */ 1141 if (fbp == NULL) { 1142 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_DEGRADED) != 0) { 1143 /* 1144 * We are already in degraded mode, so we can't 1145 * accept any failures. 1146 */ 1147 if (pbp->bio_error == 0) 1148 pbp->bio_error = cbp->bio_error; 1149 } else { 1150 fbp = cbp; 1151 } 1152 } else { 1153 /* 1154 * Next failed request, that's too many. 1155 */ 1156 if (pbp->bio_error == 0) 1157 pbp->bio_error = fbp->bio_error; 1158 } 1159 disk = cbp->bio_caller2; 1160 if (disk == NULL) 1161 continue; 1162 if ((disk->d_flags & G_RAID3_DISK_FLAG_BROKEN) == 0) { 1163 disk->d_flags |= G_RAID3_DISK_FLAG_BROKEN; 1164 G_RAID3_LOGREQ(0, cbp, "Request failed (error=%d).", 1165 cbp->bio_error); 1166 } else { 1167 G_RAID3_LOGREQ(1, cbp, "Request failed (error=%d).", 1168 cbp->bio_error); 1169 } 1170 if (g_raid3_disconnect_on_failure && 1171 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 1172 sc->sc_bump_id |= G_RAID3_BUMP_GENID; 1173 g_raid3_event_send(disk, 1174 G_RAID3_DISK_STATE_DISCONNECTED, 1175 G_RAID3_EVENT_DONTWAIT); 1176 } 1177 } 1178 if (pbp->bio_error != 0) 1179 goto finish; 1180 if (fbp != NULL && (pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) { 1181 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_VERIFY; 1182 if (xbp != fbp) 1183 g_raid3_replace_bio(xbp, fbp); 1184 g_raid3_destroy_bio(sc, fbp); 1185 } else if (fbp != NULL) { 1186 struct g_consumer *cp; 1187 1188 /* 1189 * One request failed, so send the same request to 1190 * the parity consumer. 1191 */ 1192 disk = pbp->bio_driver2; 1193 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) { 1194 pbp->bio_error = fbp->bio_error; 1195 goto finish; 1196 } 1197 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED; 1198 pbp->bio_inbed--; 1199 fbp->bio_flags &= ~(BIO_DONE | BIO_ERROR); 1200 if (disk->d_no == sc->sc_ndisks - 1) 1201 fbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1202 fbp->bio_error = 0; 1203 fbp->bio_completed = 0; 1204 fbp->bio_children = 0; 1205 fbp->bio_inbed = 0; 1206 cp = disk->d_consumer; 1207 fbp->bio_caller2 = disk; 1208 fbp->bio_to = cp->provider; 1209 G_RAID3_LOGREQ(3, fbp, "Sending request (recover)."); 1210 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1211 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1212 cp->acr, cp->acw, cp->ace)); 1213 cp->index++; 1214 g_io_request(fbp, cp); 1215 return; 1216 } 1217 if (xbp != NULL) { 1218 /* 1219 * Calculate parity. 1220 */ 1221 G_RAID3_FOREACH_BIO(pbp, cbp) { 1222 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) 1223 continue; 1224 g_raid3_xor(cbp->bio_data, xbp->bio_data, 1225 xbp->bio_length); 1226 } 1227 xbp->bio_cflags &= ~G_RAID3_BIO_CFLAG_PARITY; 1228 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) { 1229 if (!g_raid3_is_zero(xbp)) { 1230 g_raid3_parity_mismatch++; 1231 pbp->bio_error = EIO; 1232 goto finish; 1233 } 1234 g_raid3_destroy_bio(sc, xbp); 1235 } 1236 } 1237 atom = sc->sc_sectorsize / (sc->sc_ndisks - 1); 1238 cadd = padd = 0; 1239 for (left = pbp->bio_length; left > 0; left -= sc->sc_sectorsize) { 1240 G_RAID3_FOREACH_BIO(pbp, cbp) { 1241 bcopy(cbp->bio_data + cadd, pbp->bio_data + padd, atom); 1242 pbp->bio_completed += atom; 1243 padd += atom; 1244 } 1245 cadd += atom; 1246 } 1247 finish: 1248 if (pbp->bio_error == 0) 1249 G_RAID3_LOGREQ(3, pbp, "Request finished."); 1250 else { 1251 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) 1252 G_RAID3_LOGREQ(1, pbp, "Verification error."); 1253 else 1254 G_RAID3_LOGREQ(0, pbp, "Request failed."); 1255 } 1256 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_MASK; 1257 while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL) 1258 g_raid3_destroy_bio(sc, cbp); 1259 g_io_deliver(pbp, pbp->bio_error); 1260 } 1261 1262 static void 1263 g_raid3_done(struct bio *bp) 1264 { 1265 struct g_raid3_softc *sc; 1266 1267 sc = bp->bio_from->geom->softc; 1268 bp->bio_cflags |= G_RAID3_BIO_CFLAG_REGULAR; 1269 G_RAID3_LOGREQ(3, bp, "Regular request done (error=%d).", bp->bio_error); 1270 mtx_lock(&sc->sc_queue_mtx); 1271 bioq_insert_head(&sc->sc_queue, bp); 1272 mtx_unlock(&sc->sc_queue_mtx); 1273 wakeup(sc); 1274 wakeup(&sc->sc_queue); 1275 } 1276 1277 static void 1278 g_raid3_regular_request(struct bio *cbp) 1279 { 1280 struct g_raid3_softc *sc; 1281 struct g_raid3_disk *disk; 1282 struct bio *pbp; 1283 1284 g_topology_assert_not(); 1285 1286 pbp = cbp->bio_parent; 1287 sc = pbp->bio_to->geom->softc; 1288 cbp->bio_from->index--; 1289 if (cbp->bio_cmd == BIO_WRITE) 1290 sc->sc_writes--; 1291 disk = cbp->bio_from->private; 1292 if (disk == NULL) { 1293 g_topology_lock(); 1294 g_raid3_kill_consumer(sc, cbp->bio_from); 1295 g_topology_unlock(); 1296 } 1297 1298 G_RAID3_LOGREQ(3, cbp, "Request finished."); 1299 pbp->bio_inbed++; 1300 KASSERT(pbp->bio_inbed <= pbp->bio_children, 1301 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 1302 pbp->bio_children)); 1303 if (pbp->bio_inbed != pbp->bio_children) 1304 return; 1305 switch (pbp->bio_cmd) { 1306 case BIO_READ: 1307 g_raid3_gather(pbp); 1308 break; 1309 case BIO_WRITE: 1310 case BIO_DELETE: 1311 { 1312 int error = 0; 1313 1314 pbp->bio_completed = pbp->bio_length; 1315 while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL) { 1316 if (cbp->bio_error == 0) { 1317 g_raid3_destroy_bio(sc, cbp); 1318 continue; 1319 } 1320 1321 if (error == 0) 1322 error = cbp->bio_error; 1323 else if (pbp->bio_error == 0) { 1324 /* 1325 * Next failed request, that's too many. 1326 */ 1327 pbp->bio_error = error; 1328 } 1329 1330 disk = cbp->bio_caller2; 1331 if (disk == NULL) { 1332 g_raid3_destroy_bio(sc, cbp); 1333 continue; 1334 } 1335 1336 if ((disk->d_flags & G_RAID3_DISK_FLAG_BROKEN) == 0) { 1337 disk->d_flags |= G_RAID3_DISK_FLAG_BROKEN; 1338 G_RAID3_LOGREQ(0, cbp, 1339 "Request failed (error=%d).", 1340 cbp->bio_error); 1341 } else { 1342 G_RAID3_LOGREQ(1, cbp, 1343 "Request failed (error=%d).", 1344 cbp->bio_error); 1345 } 1346 if (g_raid3_disconnect_on_failure && 1347 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 1348 sc->sc_bump_id |= G_RAID3_BUMP_GENID; 1349 g_raid3_event_send(disk, 1350 G_RAID3_DISK_STATE_DISCONNECTED, 1351 G_RAID3_EVENT_DONTWAIT); 1352 } 1353 g_raid3_destroy_bio(sc, cbp); 1354 } 1355 if (pbp->bio_error == 0) 1356 G_RAID3_LOGREQ(3, pbp, "Request finished."); 1357 else 1358 G_RAID3_LOGREQ(0, pbp, "Request failed."); 1359 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_DEGRADED; 1360 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_NOPARITY; 1361 bioq_remove(&sc->sc_inflight, pbp); 1362 /* Release delayed sync requests if possible. */ 1363 g_raid3_sync_release(sc); 1364 g_io_deliver(pbp, pbp->bio_error); 1365 break; 1366 } 1367 } 1368 } 1369 1370 static void 1371 g_raid3_sync_done(struct bio *bp) 1372 { 1373 struct g_raid3_softc *sc; 1374 1375 G_RAID3_LOGREQ(3, bp, "Synchronization request delivered."); 1376 sc = bp->bio_from->geom->softc; 1377 bp->bio_cflags |= G_RAID3_BIO_CFLAG_SYNC; 1378 mtx_lock(&sc->sc_queue_mtx); 1379 bioq_insert_head(&sc->sc_queue, bp); 1380 mtx_unlock(&sc->sc_queue_mtx); 1381 wakeup(sc); 1382 wakeup(&sc->sc_queue); 1383 } 1384 1385 static void 1386 g_raid3_flush(struct g_raid3_softc *sc, struct bio *bp) 1387 { 1388 struct bio_queue_head queue; 1389 struct g_raid3_disk *disk; 1390 struct g_consumer *cp __diagused; 1391 struct bio *cbp; 1392 u_int i; 1393 1394 bioq_init(&queue); 1395 for (i = 0; i < sc->sc_ndisks; i++) { 1396 disk = &sc->sc_disks[i]; 1397 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) 1398 continue; 1399 cbp = g_clone_bio(bp); 1400 if (cbp == NULL) { 1401 for (cbp = bioq_first(&queue); cbp != NULL; 1402 cbp = bioq_first(&queue)) { 1403 bioq_remove(&queue, cbp); 1404 g_destroy_bio(cbp); 1405 } 1406 if (bp->bio_error == 0) 1407 bp->bio_error = ENOMEM; 1408 g_io_deliver(bp, bp->bio_error); 1409 return; 1410 } 1411 bioq_insert_tail(&queue, cbp); 1412 cbp->bio_done = g_std_done; 1413 cbp->bio_caller1 = disk; 1414 cbp->bio_to = disk->d_consumer->provider; 1415 } 1416 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1417 bioq_remove(&queue, cbp); 1418 G_RAID3_LOGREQ(3, cbp, "Sending request."); 1419 disk = cbp->bio_caller1; 1420 cbp->bio_caller1 = NULL; 1421 cp = disk->d_consumer; 1422 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1423 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1424 cp->acr, cp->acw, cp->ace)); 1425 g_io_request(cbp, disk->d_consumer); 1426 } 1427 } 1428 1429 static void 1430 g_raid3_start(struct bio *bp) 1431 { 1432 struct g_raid3_softc *sc; 1433 1434 sc = bp->bio_to->geom->softc; 1435 /* 1436 * If sc == NULL or there are no valid disks, provider's error 1437 * should be set and g_raid3_start() should not be called at all. 1438 */ 1439 KASSERT(sc != NULL && (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 1440 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE), 1441 ("Provider's error should be set (error=%d)(device=%s).", 1442 bp->bio_to->error, bp->bio_to->name)); 1443 G_RAID3_LOGREQ(3, bp, "Request received."); 1444 1445 switch (bp->bio_cmd) { 1446 case BIO_READ: 1447 case BIO_WRITE: 1448 case BIO_DELETE: 1449 break; 1450 case BIO_SPEEDUP: 1451 case BIO_FLUSH: 1452 g_raid3_flush(sc, bp); 1453 return; 1454 case BIO_GETATTR: 1455 default: 1456 g_io_deliver(bp, EOPNOTSUPP); 1457 return; 1458 } 1459 mtx_lock(&sc->sc_queue_mtx); 1460 bioq_insert_tail(&sc->sc_queue, bp); 1461 mtx_unlock(&sc->sc_queue_mtx); 1462 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1463 wakeup(sc); 1464 } 1465 1466 /* 1467 * Return TRUE if the given request is colliding with a in-progress 1468 * synchronization request. 1469 */ 1470 static int 1471 g_raid3_sync_collision(struct g_raid3_softc *sc, struct bio *bp) 1472 { 1473 struct g_raid3_disk *disk; 1474 struct bio *sbp; 1475 off_t rstart, rend, sstart, send; 1476 int i; 1477 1478 disk = sc->sc_syncdisk; 1479 if (disk == NULL) 1480 return (0); 1481 rstart = bp->bio_offset; 1482 rend = bp->bio_offset + bp->bio_length; 1483 for (i = 0; i < g_raid3_syncreqs; i++) { 1484 sbp = disk->d_sync.ds_bios[i]; 1485 if (sbp == NULL) 1486 continue; 1487 sstart = sbp->bio_offset; 1488 send = sbp->bio_length; 1489 if (sbp->bio_cmd == BIO_WRITE) { 1490 sstart *= sc->sc_ndisks - 1; 1491 send *= sc->sc_ndisks - 1; 1492 } 1493 send += sstart; 1494 if (rend > sstart && rstart < send) 1495 return (1); 1496 } 1497 return (0); 1498 } 1499 1500 /* 1501 * Return TRUE if the given sync request is colliding with a in-progress regular 1502 * request. 1503 */ 1504 static int 1505 g_raid3_regular_collision(struct g_raid3_softc *sc, struct bio *sbp) 1506 { 1507 off_t rstart, rend, sstart, send; 1508 struct bio *bp; 1509 1510 if (sc->sc_syncdisk == NULL) 1511 return (0); 1512 sstart = sbp->bio_offset; 1513 send = sstart + sbp->bio_length; 1514 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { 1515 rstart = bp->bio_offset; 1516 rend = bp->bio_offset + bp->bio_length; 1517 if (rend > sstart && rstart < send) 1518 return (1); 1519 } 1520 return (0); 1521 } 1522 1523 /* 1524 * Puts request onto delayed queue. 1525 */ 1526 static void 1527 g_raid3_regular_delay(struct g_raid3_softc *sc, struct bio *bp) 1528 { 1529 1530 G_RAID3_LOGREQ(2, bp, "Delaying request."); 1531 bioq_insert_head(&sc->sc_regular_delayed, bp); 1532 } 1533 1534 /* 1535 * Puts synchronization request onto delayed queue. 1536 */ 1537 static void 1538 g_raid3_sync_delay(struct g_raid3_softc *sc, struct bio *bp) 1539 { 1540 1541 G_RAID3_LOGREQ(2, bp, "Delaying synchronization request."); 1542 bioq_insert_tail(&sc->sc_sync_delayed, bp); 1543 } 1544 1545 /* 1546 * Releases delayed regular requests which don't collide anymore with sync 1547 * requests. 1548 */ 1549 static void 1550 g_raid3_regular_release(struct g_raid3_softc *sc) 1551 { 1552 struct bio *bp, *bp2; 1553 1554 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { 1555 if (g_raid3_sync_collision(sc, bp)) 1556 continue; 1557 bioq_remove(&sc->sc_regular_delayed, bp); 1558 G_RAID3_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1559 mtx_lock(&sc->sc_queue_mtx); 1560 bioq_insert_head(&sc->sc_queue, bp); 1561 #if 0 1562 /* 1563 * wakeup() is not needed, because this function is called from 1564 * the worker thread. 1565 */ 1566 wakeup(&sc->sc_queue); 1567 #endif 1568 mtx_unlock(&sc->sc_queue_mtx); 1569 } 1570 } 1571 1572 /* 1573 * Releases delayed sync requests which don't collide anymore with regular 1574 * requests. 1575 */ 1576 static void 1577 g_raid3_sync_release(struct g_raid3_softc *sc) 1578 { 1579 struct bio *bp, *bp2; 1580 1581 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { 1582 if (g_raid3_regular_collision(sc, bp)) 1583 continue; 1584 bioq_remove(&sc->sc_sync_delayed, bp); 1585 G_RAID3_LOGREQ(2, bp, 1586 "Releasing delayed synchronization request."); 1587 g_io_request(bp, bp->bio_from); 1588 } 1589 } 1590 1591 /* 1592 * Handle synchronization requests. 1593 * Every synchronization request is two-steps process: first, READ request is 1594 * send to active provider and then WRITE request (with read data) to the provider 1595 * being synchronized. When WRITE is finished, new synchronization request is 1596 * send. 1597 */ 1598 static void 1599 g_raid3_sync_request(struct bio *bp) 1600 { 1601 struct g_raid3_softc *sc; 1602 struct g_raid3_disk *disk; 1603 1604 bp->bio_from->index--; 1605 sc = bp->bio_from->geom->softc; 1606 disk = bp->bio_from->private; 1607 if (disk == NULL) { 1608 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1609 g_topology_lock(); 1610 g_raid3_kill_consumer(sc, bp->bio_from); 1611 g_topology_unlock(); 1612 free(bp->bio_data, M_RAID3); 1613 g_destroy_bio(bp); 1614 sx_xlock(&sc->sc_lock); 1615 return; 1616 } 1617 1618 /* 1619 * Synchronization request. 1620 */ 1621 switch (bp->bio_cmd) { 1622 case BIO_READ: 1623 { 1624 struct g_consumer *cp; 1625 u_char *dst, *src; 1626 off_t left; 1627 u_int atom; 1628 1629 if (bp->bio_error != 0) { 1630 G_RAID3_LOGREQ(0, bp, 1631 "Synchronization request failed (error=%d).", 1632 bp->bio_error); 1633 g_destroy_bio(bp); 1634 return; 1635 } 1636 G_RAID3_LOGREQ(3, bp, "Synchronization request finished."); 1637 atom = sc->sc_sectorsize / (sc->sc_ndisks - 1); 1638 dst = src = bp->bio_data; 1639 if (disk->d_no == sc->sc_ndisks - 1) { 1640 u_int n; 1641 1642 /* Parity component. */ 1643 for (left = bp->bio_length; left > 0; 1644 left -= sc->sc_sectorsize) { 1645 bcopy(src, dst, atom); 1646 src += atom; 1647 for (n = 1; n < sc->sc_ndisks - 1; n++) { 1648 g_raid3_xor(src, dst, atom); 1649 src += atom; 1650 } 1651 dst += atom; 1652 } 1653 } else { 1654 /* Regular component. */ 1655 src += atom * disk->d_no; 1656 for (left = bp->bio_length; left > 0; 1657 left -= sc->sc_sectorsize) { 1658 bcopy(src, dst, atom); 1659 src += sc->sc_sectorsize; 1660 dst += atom; 1661 } 1662 } 1663 bp->bio_driver1 = bp->bio_driver2 = NULL; 1664 bp->bio_pflags = 0; 1665 bp->bio_offset /= sc->sc_ndisks - 1; 1666 bp->bio_length /= sc->sc_ndisks - 1; 1667 bp->bio_cmd = BIO_WRITE; 1668 bp->bio_cflags = 0; 1669 bp->bio_children = bp->bio_inbed = 0; 1670 cp = disk->d_consumer; 1671 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1672 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1673 cp->acr, cp->acw, cp->ace)); 1674 cp->index++; 1675 g_io_request(bp, cp); 1676 return; 1677 } 1678 case BIO_WRITE: 1679 { 1680 struct g_raid3_disk_sync *sync; 1681 off_t boffset, moffset; 1682 void *data; 1683 int i; 1684 1685 if (bp->bio_error != 0) { 1686 G_RAID3_LOGREQ(0, bp, 1687 "Synchronization request failed (error=%d).", 1688 bp->bio_error); 1689 g_destroy_bio(bp); 1690 sc->sc_bump_id |= G_RAID3_BUMP_GENID; 1691 g_raid3_event_send(disk, 1692 G_RAID3_DISK_STATE_DISCONNECTED, 1693 G_RAID3_EVENT_DONTWAIT); 1694 return; 1695 } 1696 G_RAID3_LOGREQ(3, bp, "Synchronization request finished."); 1697 sync = &disk->d_sync; 1698 if (sync->ds_offset == sc->sc_mediasize / (sc->sc_ndisks - 1) || 1699 sync->ds_consumer == NULL || 1700 (sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 1701 /* Don't send more synchronization requests. */ 1702 sync->ds_inflight--; 1703 if (sync->ds_bios != NULL) { 1704 i = (int)(uintptr_t)bp->bio_caller1; 1705 sync->ds_bios[i] = NULL; 1706 } 1707 free(bp->bio_data, M_RAID3); 1708 g_destroy_bio(bp); 1709 if (sync->ds_inflight > 0) 1710 return; 1711 if (sync->ds_consumer == NULL || 1712 (sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 1713 return; 1714 } 1715 /* 1716 * Disk up-to-date, activate it. 1717 */ 1718 g_raid3_event_send(disk, G_RAID3_DISK_STATE_ACTIVE, 1719 G_RAID3_EVENT_DONTWAIT); 1720 return; 1721 } 1722 1723 /* Send next synchronization request. */ 1724 data = bp->bio_data; 1725 g_reset_bio(bp); 1726 bp->bio_cmd = BIO_READ; 1727 bp->bio_offset = sync->ds_offset * (sc->sc_ndisks - 1); 1728 bp->bio_length = MIN(maxphys, sc->sc_mediasize - bp->bio_offset); 1729 sync->ds_offset += bp->bio_length / (sc->sc_ndisks - 1); 1730 bp->bio_done = g_raid3_sync_done; 1731 bp->bio_data = data; 1732 bp->bio_from = sync->ds_consumer; 1733 bp->bio_to = sc->sc_provider; 1734 G_RAID3_LOGREQ(3, bp, "Sending synchronization request."); 1735 sync->ds_consumer->index++; 1736 /* 1737 * Delay the request if it is colliding with a regular request. 1738 */ 1739 if (g_raid3_regular_collision(sc, bp)) 1740 g_raid3_sync_delay(sc, bp); 1741 else 1742 g_io_request(bp, sync->ds_consumer); 1743 1744 /* Release delayed requests if possible. */ 1745 g_raid3_regular_release(sc); 1746 1747 /* Find the smallest offset. */ 1748 moffset = sc->sc_mediasize; 1749 for (i = 0; i < g_raid3_syncreqs; i++) { 1750 bp = sync->ds_bios[i]; 1751 boffset = bp->bio_offset; 1752 if (bp->bio_cmd == BIO_WRITE) 1753 boffset *= sc->sc_ndisks - 1; 1754 if (boffset < moffset) 1755 moffset = boffset; 1756 } 1757 if (sync->ds_offset_done + maxphys * 100 < moffset) { 1758 /* Update offset_done on every 100 blocks. */ 1759 sync->ds_offset_done = moffset; 1760 g_raid3_update_metadata(disk); 1761 } 1762 return; 1763 } 1764 default: 1765 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1766 bp->bio_cmd, sc->sc_name)); 1767 break; 1768 } 1769 } 1770 1771 static int 1772 g_raid3_register_request(struct bio *pbp) 1773 { 1774 struct g_raid3_softc *sc; 1775 struct g_raid3_disk *disk; 1776 struct g_consumer *cp; 1777 struct bio *cbp, *tmpbp; 1778 off_t offset, length; 1779 u_int n, ndisks; 1780 int round_robin, verify; 1781 1782 ndisks = 0; 1783 sc = pbp->bio_to->geom->softc; 1784 if ((pbp->bio_cflags & G_RAID3_BIO_CFLAG_REGSYNC) != 0 && 1785 sc->sc_syncdisk == NULL) { 1786 g_io_deliver(pbp, EIO); 1787 return (0); 1788 } 1789 g_raid3_init_bio(pbp); 1790 length = pbp->bio_length / (sc->sc_ndisks - 1); 1791 offset = pbp->bio_offset / (sc->sc_ndisks - 1); 1792 round_robin = verify = 0; 1793 switch (pbp->bio_cmd) { 1794 case BIO_READ: 1795 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_VERIFY) != 0 && 1796 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 1797 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_VERIFY; 1798 verify = 1; 1799 ndisks = sc->sc_ndisks; 1800 } else { 1801 verify = 0; 1802 ndisks = sc->sc_ndisks - 1; 1803 } 1804 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_ROUND_ROBIN) != 0 && 1805 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 1806 round_robin = 1; 1807 } else { 1808 round_robin = 0; 1809 } 1810 KASSERT(!round_robin || !verify, 1811 ("ROUND-ROBIN and VERIFY are mutually exclusive.")); 1812 pbp->bio_driver2 = &sc->sc_disks[sc->sc_ndisks - 1]; 1813 break; 1814 case BIO_WRITE: 1815 case BIO_DELETE: 1816 /* 1817 * Delay the request if it is colliding with a synchronization 1818 * request. 1819 */ 1820 if (g_raid3_sync_collision(sc, pbp)) { 1821 g_raid3_regular_delay(sc, pbp); 1822 return (0); 1823 } 1824 1825 if (sc->sc_idle) 1826 g_raid3_unidle(sc); 1827 else 1828 sc->sc_last_write = time_uptime; 1829 1830 ndisks = sc->sc_ndisks; 1831 break; 1832 } 1833 for (n = 0; n < ndisks; n++) { 1834 disk = &sc->sc_disks[n]; 1835 cbp = g_raid3_clone_bio(sc, pbp); 1836 if (cbp == NULL) { 1837 while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL) 1838 g_raid3_destroy_bio(sc, cbp); 1839 /* 1840 * To prevent deadlock, we must run back up 1841 * with the ENOMEM for failed requests of any 1842 * of our consumers. Our own sync requests 1843 * can stick around, as they are finite. 1844 */ 1845 if ((pbp->bio_cflags & 1846 G_RAID3_BIO_CFLAG_REGULAR) != 0) { 1847 g_io_deliver(pbp, ENOMEM); 1848 return (0); 1849 } 1850 return (ENOMEM); 1851 } 1852 cbp->bio_offset = offset; 1853 cbp->bio_length = length; 1854 cbp->bio_done = g_raid3_done; 1855 switch (pbp->bio_cmd) { 1856 case BIO_READ: 1857 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) { 1858 /* 1859 * Replace invalid component with the parity 1860 * component. 1861 */ 1862 disk = &sc->sc_disks[sc->sc_ndisks - 1]; 1863 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1864 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED; 1865 } else if (round_robin && 1866 disk->d_no == sc->sc_round_robin) { 1867 /* 1868 * In round-robin mode skip one data component 1869 * and use parity component when reading. 1870 */ 1871 pbp->bio_driver2 = disk; 1872 disk = &sc->sc_disks[sc->sc_ndisks - 1]; 1873 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1874 sc->sc_round_robin++; 1875 round_robin = 0; 1876 } else if (verify && disk->d_no == sc->sc_ndisks - 1) { 1877 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1878 } 1879 break; 1880 case BIO_WRITE: 1881 case BIO_DELETE: 1882 if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 1883 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 1884 if (n == ndisks - 1) { 1885 /* 1886 * Active parity component, mark it as such. 1887 */ 1888 cbp->bio_cflags |= 1889 G_RAID3_BIO_CFLAG_PARITY; 1890 } 1891 } else { 1892 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED; 1893 if (n == ndisks - 1) { 1894 /* 1895 * Parity component is not connected, 1896 * so destroy its request. 1897 */ 1898 pbp->bio_pflags |= 1899 G_RAID3_BIO_PFLAG_NOPARITY; 1900 g_raid3_destroy_bio(sc, cbp); 1901 cbp = NULL; 1902 } else { 1903 cbp->bio_cflags |= 1904 G_RAID3_BIO_CFLAG_NODISK; 1905 disk = NULL; 1906 } 1907 } 1908 break; 1909 } 1910 if (cbp != NULL) 1911 cbp->bio_caller2 = disk; 1912 } 1913 switch (pbp->bio_cmd) { 1914 case BIO_READ: 1915 if (round_robin) { 1916 /* 1917 * If we are in round-robin mode and 'round_robin' is 1918 * still 1, it means, that we skipped parity component 1919 * for this read and must reset sc_round_robin field. 1920 */ 1921 sc->sc_round_robin = 0; 1922 } 1923 G_RAID3_FOREACH_SAFE_BIO(pbp, cbp, tmpbp) { 1924 disk = cbp->bio_caller2; 1925 cp = disk->d_consumer; 1926 cbp->bio_to = cp->provider; 1927 G_RAID3_LOGREQ(3, cbp, "Sending request."); 1928 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1929 ("Consumer %s not opened (r%dw%de%d).", 1930 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1931 cp->index++; 1932 g_io_request(cbp, cp); 1933 } 1934 break; 1935 case BIO_WRITE: 1936 case BIO_DELETE: 1937 /* 1938 * Put request onto inflight queue, so we can check if new 1939 * synchronization requests don't collide with it. 1940 */ 1941 bioq_insert_tail(&sc->sc_inflight, pbp); 1942 1943 /* 1944 * Bump syncid on first write. 1945 */ 1946 if ((sc->sc_bump_id & G_RAID3_BUMP_SYNCID) != 0) { 1947 sc->sc_bump_id &= ~G_RAID3_BUMP_SYNCID; 1948 g_raid3_bump_syncid(sc); 1949 } 1950 g_raid3_scatter(pbp); 1951 break; 1952 } 1953 return (0); 1954 } 1955 1956 static int 1957 g_raid3_can_destroy(struct g_raid3_softc *sc) 1958 { 1959 struct g_geom *gp; 1960 struct g_consumer *cp; 1961 1962 g_topology_assert(); 1963 gp = sc->sc_geom; 1964 if (gp->softc == NULL) 1965 return (1); 1966 LIST_FOREACH(cp, &gp->consumer, consumer) { 1967 if (g_raid3_is_busy(sc, cp)) 1968 return (0); 1969 } 1970 gp = sc->sc_sync.ds_geom; 1971 LIST_FOREACH(cp, &gp->consumer, consumer) { 1972 if (g_raid3_is_busy(sc, cp)) 1973 return (0); 1974 } 1975 G_RAID3_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1976 sc->sc_name); 1977 return (1); 1978 } 1979 1980 static int 1981 g_raid3_try_destroy(struct g_raid3_softc *sc) 1982 { 1983 1984 g_topology_assert_not(); 1985 sx_assert(&sc->sc_lock, SX_XLOCKED); 1986 1987 if (sc->sc_rootmount != NULL) { 1988 G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1989 sc->sc_rootmount); 1990 root_mount_rel(sc->sc_rootmount); 1991 sc->sc_rootmount = NULL; 1992 } 1993 1994 g_topology_lock(); 1995 if (!g_raid3_can_destroy(sc)) { 1996 g_topology_unlock(); 1997 return (0); 1998 } 1999 sc->sc_geom->softc = NULL; 2000 sc->sc_sync.ds_geom->softc = NULL; 2001 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_WAIT) != 0) { 2002 g_topology_unlock(); 2003 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, 2004 &sc->sc_worker); 2005 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 2006 sx_xunlock(&sc->sc_lock); 2007 wakeup(&sc->sc_worker); 2008 sc->sc_worker = NULL; 2009 } else { 2010 g_topology_unlock(); 2011 g_raid3_destroy_device(sc); 2012 free(sc->sc_disks, M_RAID3); 2013 free(sc, M_RAID3); 2014 } 2015 return (1); 2016 } 2017 2018 /* 2019 * Worker thread. 2020 */ 2021 static void 2022 g_raid3_worker(void *arg) 2023 { 2024 struct g_raid3_softc *sc; 2025 struct g_raid3_event *ep; 2026 struct bio *bp; 2027 int timeout; 2028 2029 sc = arg; 2030 thread_lock(curthread); 2031 sched_prio(curthread, PRIBIO); 2032 thread_unlock(curthread); 2033 2034 sx_xlock(&sc->sc_lock); 2035 for (;;) { 2036 G_RAID3_DEBUG(5, "%s: Let's see...", __func__); 2037 /* 2038 * First take a look at events. 2039 * This is important to handle events before any I/O requests. 2040 */ 2041 ep = g_raid3_event_get(sc); 2042 if (ep != NULL) { 2043 g_raid3_event_remove(sc, ep); 2044 if ((ep->e_flags & G_RAID3_EVENT_DEVICE) != 0) { 2045 /* Update only device status. */ 2046 G_RAID3_DEBUG(3, 2047 "Running event for device %s.", 2048 sc->sc_name); 2049 ep->e_error = 0; 2050 g_raid3_update_device(sc, 1); 2051 } else { 2052 /* Update disk status. */ 2053 G_RAID3_DEBUG(3, "Running event for disk %s.", 2054 g_raid3_get_diskname(ep->e_disk)); 2055 ep->e_error = g_raid3_update_disk(ep->e_disk, 2056 ep->e_state); 2057 if (ep->e_error == 0) 2058 g_raid3_update_device(sc, 0); 2059 } 2060 if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0) { 2061 KASSERT(ep->e_error == 0, 2062 ("Error cannot be handled.")); 2063 g_raid3_event_free(ep); 2064 } else { 2065 ep->e_flags |= G_RAID3_EVENT_DONE; 2066 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, 2067 ep); 2068 mtx_lock(&sc->sc_events_mtx); 2069 wakeup(ep); 2070 mtx_unlock(&sc->sc_events_mtx); 2071 } 2072 if ((sc->sc_flags & 2073 G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 2074 if (g_raid3_try_destroy(sc)) { 2075 curthread->td_pflags &= ~TDP_GEOM; 2076 G_RAID3_DEBUG(1, "Thread exiting."); 2077 kproc_exit(0); 2078 } 2079 } 2080 G_RAID3_DEBUG(5, "%s: I'm here 1.", __func__); 2081 continue; 2082 } 2083 /* 2084 * Check if we can mark array as CLEAN and if we can't take 2085 * how much seconds should we wait. 2086 */ 2087 timeout = g_raid3_idle(sc, -1); 2088 /* 2089 * Now I/O requests. 2090 */ 2091 /* Get first request from the queue. */ 2092 mtx_lock(&sc->sc_queue_mtx); 2093 bp = bioq_first(&sc->sc_queue); 2094 if (bp == NULL) { 2095 if ((sc->sc_flags & 2096 G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 2097 mtx_unlock(&sc->sc_queue_mtx); 2098 if (g_raid3_try_destroy(sc)) { 2099 curthread->td_pflags &= ~TDP_GEOM; 2100 G_RAID3_DEBUG(1, "Thread exiting."); 2101 kproc_exit(0); 2102 } 2103 mtx_lock(&sc->sc_queue_mtx); 2104 } 2105 sx_xunlock(&sc->sc_lock); 2106 /* 2107 * XXX: We can miss an event here, because an event 2108 * can be added without sx-device-lock and without 2109 * mtx-queue-lock. Maybe I should just stop using 2110 * dedicated mutex for events synchronization and 2111 * stick with the queue lock? 2112 * The event will hang here until next I/O request 2113 * or next event is received. 2114 */ 2115 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "r3:w1", 2116 timeout * hz); 2117 sx_xlock(&sc->sc_lock); 2118 G_RAID3_DEBUG(5, "%s: I'm here 4.", __func__); 2119 continue; 2120 } 2121 process: 2122 bioq_remove(&sc->sc_queue, bp); 2123 mtx_unlock(&sc->sc_queue_mtx); 2124 2125 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 2126 (bp->bio_cflags & G_RAID3_BIO_CFLAG_SYNC) != 0) { 2127 g_raid3_sync_request(bp); /* READ */ 2128 } else if (bp->bio_to != sc->sc_provider) { 2129 if ((bp->bio_cflags & G_RAID3_BIO_CFLAG_REGULAR) != 0) 2130 g_raid3_regular_request(bp); 2131 else if ((bp->bio_cflags & G_RAID3_BIO_CFLAG_SYNC) != 0) 2132 g_raid3_sync_request(bp); /* WRITE */ 2133 else { 2134 KASSERT(0, 2135 ("Invalid request cflags=0x%hx to=%s.", 2136 bp->bio_cflags, bp->bio_to->name)); 2137 } 2138 } else if (g_raid3_register_request(bp) != 0) { 2139 mtx_lock(&sc->sc_queue_mtx); 2140 bioq_insert_head(&sc->sc_queue, bp); 2141 /* 2142 * We are short in memory, let see if there are finished 2143 * request we can free. 2144 */ 2145 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 2146 if (bp->bio_cflags & G_RAID3_BIO_CFLAG_REGULAR) 2147 goto process; 2148 } 2149 /* 2150 * No finished regular request, so at least keep 2151 * synchronization running. 2152 */ 2153 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 2154 if (bp->bio_cflags & G_RAID3_BIO_CFLAG_SYNC) 2155 goto process; 2156 } 2157 sx_xunlock(&sc->sc_lock); 2158 MSLEEP(&sc->sc_queue, &sc->sc_queue_mtx, PRIBIO | PDROP, 2159 "r3:lowmem", hz / 10); 2160 sx_xlock(&sc->sc_lock); 2161 } 2162 G_RAID3_DEBUG(5, "%s: I'm here 9.", __func__); 2163 } 2164 } 2165 2166 static void 2167 g_raid3_update_idle(struct g_raid3_softc *sc, struct g_raid3_disk *disk) 2168 { 2169 2170 sx_assert(&sc->sc_lock, SX_LOCKED); 2171 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) != 0) 2172 return; 2173 if (!sc->sc_idle && (disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) == 0) { 2174 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as dirty.", 2175 g_raid3_get_diskname(disk), sc->sc_name); 2176 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY; 2177 } else if (sc->sc_idle && 2178 (disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) != 0) { 2179 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as clean.", 2180 g_raid3_get_diskname(disk), sc->sc_name); 2181 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2182 } 2183 } 2184 2185 static void 2186 g_raid3_sync_start(struct g_raid3_softc *sc) 2187 { 2188 struct g_raid3_disk *disk; 2189 struct g_consumer *cp; 2190 struct bio *bp; 2191 int error __diagused; 2192 u_int n; 2193 2194 g_topology_assert_not(); 2195 sx_assert(&sc->sc_lock, SX_XLOCKED); 2196 2197 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED, 2198 ("Device not in DEGRADED state (%s, %u).", sc->sc_name, 2199 sc->sc_state)); 2200 KASSERT(sc->sc_syncdisk == NULL, ("Syncdisk is not NULL (%s, %u).", 2201 sc->sc_name, sc->sc_state)); 2202 disk = NULL; 2203 for (n = 0; n < sc->sc_ndisks; n++) { 2204 if (sc->sc_disks[n].d_state != G_RAID3_DISK_STATE_SYNCHRONIZING) 2205 continue; 2206 disk = &sc->sc_disks[n]; 2207 break; 2208 } 2209 if (disk == NULL) 2210 return; 2211 2212 sx_xunlock(&sc->sc_lock); 2213 g_topology_lock(); 2214 cp = g_new_consumer(sc->sc_sync.ds_geom); 2215 error = g_attach(cp, sc->sc_provider); 2216 KASSERT(error == 0, 2217 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 2218 error = g_access(cp, 1, 0, 0); 2219 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 2220 g_topology_unlock(); 2221 sx_xlock(&sc->sc_lock); 2222 2223 G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 2224 g_raid3_get_diskname(disk)); 2225 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) == 0) 2226 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY; 2227 KASSERT(disk->d_sync.ds_consumer == NULL, 2228 ("Sync consumer already exists (device=%s, disk=%s).", 2229 sc->sc_name, g_raid3_get_diskname(disk))); 2230 2231 disk->d_sync.ds_consumer = cp; 2232 disk->d_sync.ds_consumer->private = disk; 2233 disk->d_sync.ds_consumer->index = 0; 2234 sc->sc_syncdisk = disk; 2235 2236 /* 2237 * Allocate memory for synchronization bios and initialize them. 2238 */ 2239 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_raid3_syncreqs, 2240 M_RAID3, M_WAITOK); 2241 for (n = 0; n < g_raid3_syncreqs; n++) { 2242 bp = g_alloc_bio(); 2243 disk->d_sync.ds_bios[n] = bp; 2244 bp->bio_parent = NULL; 2245 bp->bio_cmd = BIO_READ; 2246 bp->bio_data = malloc(maxphys, M_RAID3, M_WAITOK); 2247 bp->bio_cflags = 0; 2248 bp->bio_offset = disk->d_sync.ds_offset * (sc->sc_ndisks - 1); 2249 bp->bio_length = MIN(maxphys, sc->sc_mediasize - bp->bio_offset); 2250 disk->d_sync.ds_offset += bp->bio_length / (sc->sc_ndisks - 1); 2251 bp->bio_done = g_raid3_sync_done; 2252 bp->bio_from = disk->d_sync.ds_consumer; 2253 bp->bio_to = sc->sc_provider; 2254 bp->bio_caller1 = (void *)(uintptr_t)n; 2255 } 2256 2257 /* Set the number of in-flight synchronization requests. */ 2258 disk->d_sync.ds_inflight = g_raid3_syncreqs; 2259 2260 /* 2261 * Fire off first synchronization requests. 2262 */ 2263 for (n = 0; n < g_raid3_syncreqs; n++) { 2264 bp = disk->d_sync.ds_bios[n]; 2265 G_RAID3_LOGREQ(3, bp, "Sending synchronization request."); 2266 disk->d_sync.ds_consumer->index++; 2267 /* 2268 * Delay the request if it is colliding with a regular request. 2269 */ 2270 if (g_raid3_regular_collision(sc, bp)) 2271 g_raid3_sync_delay(sc, bp); 2272 else 2273 g_io_request(bp, disk->d_sync.ds_consumer); 2274 } 2275 } 2276 2277 /* 2278 * Stop synchronization process. 2279 * type: 0 - synchronization finished 2280 * 1 - synchronization stopped 2281 */ 2282 static void 2283 g_raid3_sync_stop(struct g_raid3_softc *sc, int type) 2284 { 2285 struct g_raid3_disk *disk; 2286 struct g_consumer *cp; 2287 2288 g_topology_assert_not(); 2289 sx_assert(&sc->sc_lock, SX_LOCKED); 2290 2291 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED, 2292 ("Device not in DEGRADED state (%s, %u).", sc->sc_name, 2293 sc->sc_state)); 2294 disk = sc->sc_syncdisk; 2295 sc->sc_syncdisk = NULL; 2296 KASSERT(disk != NULL, ("No disk was synchronized (%s).", sc->sc_name)); 2297 KASSERT(disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING, 2298 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2299 g_raid3_disk_state2str(disk->d_state))); 2300 if (disk->d_sync.ds_consumer == NULL) 2301 return; 2302 2303 if (type == 0) { 2304 G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2305 sc->sc_name, g_raid3_get_diskname(disk)); 2306 } else /* if (type == 1) */ { 2307 G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2308 sc->sc_name, g_raid3_get_diskname(disk)); 2309 } 2310 free(disk->d_sync.ds_bios, M_RAID3); 2311 disk->d_sync.ds_bios = NULL; 2312 cp = disk->d_sync.ds_consumer; 2313 disk->d_sync.ds_consumer = NULL; 2314 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2315 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2316 g_topology_lock(); 2317 g_raid3_kill_consumer(sc, cp); 2318 g_topology_unlock(); 2319 sx_xlock(&sc->sc_lock); 2320 } 2321 2322 static void 2323 g_raid3_launch_provider(struct g_raid3_softc *sc) 2324 { 2325 struct g_provider *pp; 2326 struct g_raid3_disk *disk; 2327 int n; 2328 2329 sx_assert(&sc->sc_lock, SX_LOCKED); 2330 2331 g_topology_lock(); 2332 pp = g_new_providerf(sc->sc_geom, "raid3/%s", sc->sc_name); 2333 pp->mediasize = sc->sc_mediasize; 2334 pp->sectorsize = sc->sc_sectorsize; 2335 pp->stripesize = 0; 2336 pp->stripeoffset = 0; 2337 for (n = 0; n < sc->sc_ndisks; n++) { 2338 disk = &sc->sc_disks[n]; 2339 if (disk->d_consumer && disk->d_consumer->provider && 2340 disk->d_consumer->provider->stripesize > pp->stripesize) { 2341 pp->stripesize = disk->d_consumer->provider->stripesize; 2342 pp->stripeoffset = disk->d_consumer->provider->stripeoffset; 2343 } 2344 } 2345 pp->stripesize *= sc->sc_ndisks - 1; 2346 pp->stripeoffset *= sc->sc_ndisks - 1; 2347 sc->sc_provider = pp; 2348 g_error_provider(pp, 0); 2349 g_topology_unlock(); 2350 G_RAID3_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2351 g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE), sc->sc_ndisks); 2352 2353 if (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED) 2354 g_raid3_sync_start(sc); 2355 } 2356 2357 static void 2358 g_raid3_destroy_provider(struct g_raid3_softc *sc) 2359 { 2360 struct bio *bp; 2361 2362 g_topology_assert_not(); 2363 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2364 sc->sc_name)); 2365 2366 g_topology_lock(); 2367 g_error_provider(sc->sc_provider, ENXIO); 2368 mtx_lock(&sc->sc_queue_mtx); 2369 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 2370 bioq_remove(&sc->sc_queue, bp); 2371 g_io_deliver(bp, ENXIO); 2372 } 2373 mtx_unlock(&sc->sc_queue_mtx); 2374 G_RAID3_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 2375 sc->sc_provider->name); 2376 g_wither_provider(sc->sc_provider, ENXIO); 2377 g_topology_unlock(); 2378 sc->sc_provider = NULL; 2379 if (sc->sc_syncdisk != NULL) 2380 g_raid3_sync_stop(sc, 1); 2381 } 2382 2383 static void 2384 g_raid3_go(void *arg) 2385 { 2386 struct g_raid3_softc *sc; 2387 struct g_raid3_event *ep; 2388 2389 sc = arg; 2390 G_RAID3_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2391 ep = sc->sc_timeout_event; 2392 sc->sc_timeout_event = NULL; 2393 g_raid3_event_dispatch(ep, sc, 0, 2394 G_RAID3_EVENT_DONTWAIT | G_RAID3_EVENT_DEVICE); 2395 } 2396 2397 static void 2398 g_raid3_timeout_drain(struct g_raid3_softc *sc) 2399 { 2400 sx_assert(&sc->sc_lock, SX_XLOCKED); 2401 2402 callout_drain(&sc->sc_callout); 2403 g_raid3_event_free(sc->sc_timeout_event); 2404 sc->sc_timeout_event = NULL; 2405 } 2406 2407 static u_int 2408 g_raid3_determine_state(struct g_raid3_disk *disk) 2409 { 2410 struct g_raid3_softc *sc; 2411 u_int state; 2412 2413 sc = disk->d_softc; 2414 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2415 if ((disk->d_flags & 2416 G_RAID3_DISK_FLAG_SYNCHRONIZING) == 0) { 2417 /* Disk does not need synchronization. */ 2418 state = G_RAID3_DISK_STATE_ACTIVE; 2419 } else { 2420 if ((sc->sc_flags & 2421 G_RAID3_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2422 (disk->d_flags & 2423 G_RAID3_DISK_FLAG_FORCE_SYNC) != 0) { 2424 /* 2425 * We can start synchronization from 2426 * the stored offset. 2427 */ 2428 state = G_RAID3_DISK_STATE_SYNCHRONIZING; 2429 } else { 2430 state = G_RAID3_DISK_STATE_STALE; 2431 } 2432 } 2433 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2434 /* 2435 * Reset all synchronization data for this disk, 2436 * because if it even was synchronized, it was 2437 * synchronized to disks with different syncid. 2438 */ 2439 disk->d_flags |= G_RAID3_DISK_FLAG_SYNCHRONIZING; 2440 disk->d_sync.ds_offset = 0; 2441 disk->d_sync.ds_offset_done = 0; 2442 disk->d_sync.ds_syncid = sc->sc_syncid; 2443 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2444 (disk->d_flags & G_RAID3_DISK_FLAG_FORCE_SYNC) != 0) { 2445 state = G_RAID3_DISK_STATE_SYNCHRONIZING; 2446 } else { 2447 state = G_RAID3_DISK_STATE_STALE; 2448 } 2449 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2450 /* 2451 * Not good, NOT GOOD! 2452 * It means that device was started on stale disks 2453 * and more fresh disk just arrive. 2454 * If there were writes, device is broken, sorry. 2455 * I think the best choice here is don't touch 2456 * this disk and inform the user loudly. 2457 */ 2458 G_RAID3_DEBUG(0, "Device %s was started before the freshest " 2459 "disk (%s) arrives!! It will not be connected to the " 2460 "running device.", sc->sc_name, 2461 g_raid3_get_diskname(disk)); 2462 g_raid3_destroy_disk(disk); 2463 state = G_RAID3_DISK_STATE_NONE; 2464 /* Return immediately, because disk was destroyed. */ 2465 return (state); 2466 } 2467 G_RAID3_DEBUG(3, "State for %s disk: %s.", 2468 g_raid3_get_diskname(disk), g_raid3_disk_state2str(state)); 2469 return (state); 2470 } 2471 2472 /* 2473 * Update device state. 2474 */ 2475 static void 2476 g_raid3_update_device(struct g_raid3_softc *sc, boolean_t force) 2477 { 2478 struct g_raid3_disk *disk; 2479 u_int state; 2480 2481 sx_assert(&sc->sc_lock, SX_XLOCKED); 2482 2483 switch (sc->sc_state) { 2484 case G_RAID3_DEVICE_STATE_STARTING: 2485 { 2486 u_int n, ndirty, ndisks, genid, syncid; 2487 2488 KASSERT(sc->sc_provider == NULL, 2489 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2490 /* 2491 * Are we ready? We are, if all disks are connected or 2492 * one disk is missing and 'force' is true. 2493 */ 2494 if (g_raid3_ndisks(sc, -1) + force == sc->sc_ndisks) { 2495 if (!force) 2496 g_raid3_timeout_drain(sc); 2497 } else { 2498 if (force) { 2499 /* 2500 * Timeout expired, so destroy device. 2501 */ 2502 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 2503 G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", 2504 __LINE__, sc->sc_rootmount); 2505 root_mount_rel(sc->sc_rootmount); 2506 sc->sc_rootmount = NULL; 2507 } 2508 return; 2509 } 2510 2511 /* 2512 * Find the biggest genid. 2513 */ 2514 genid = 0; 2515 for (n = 0; n < sc->sc_ndisks; n++) { 2516 disk = &sc->sc_disks[n]; 2517 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2518 continue; 2519 if (disk->d_genid > genid) 2520 genid = disk->d_genid; 2521 } 2522 sc->sc_genid = genid; 2523 /* 2524 * Remove all disks without the biggest genid. 2525 */ 2526 for (n = 0; n < sc->sc_ndisks; n++) { 2527 disk = &sc->sc_disks[n]; 2528 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2529 continue; 2530 if (disk->d_genid < genid) { 2531 G_RAID3_DEBUG(0, 2532 "Component %s (device %s) broken, skipping.", 2533 g_raid3_get_diskname(disk), sc->sc_name); 2534 g_raid3_destroy_disk(disk); 2535 } 2536 } 2537 2538 /* 2539 * There must be at least 'sc->sc_ndisks - 1' components 2540 * with the same syncid and without SYNCHRONIZING flag. 2541 */ 2542 2543 /* 2544 * Find the biggest syncid, number of valid components and 2545 * number of dirty components. 2546 */ 2547 ndirty = ndisks = syncid = 0; 2548 for (n = 0; n < sc->sc_ndisks; n++) { 2549 disk = &sc->sc_disks[n]; 2550 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2551 continue; 2552 if ((disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) != 0) 2553 ndirty++; 2554 if (disk->d_sync.ds_syncid > syncid) { 2555 syncid = disk->d_sync.ds_syncid; 2556 ndisks = 0; 2557 } else if (disk->d_sync.ds_syncid < syncid) { 2558 continue; 2559 } 2560 if ((disk->d_flags & 2561 G_RAID3_DISK_FLAG_SYNCHRONIZING) != 0) { 2562 continue; 2563 } 2564 ndisks++; 2565 } 2566 /* 2567 * Do we have enough valid components? 2568 */ 2569 if (ndisks + 1 < sc->sc_ndisks) { 2570 G_RAID3_DEBUG(0, 2571 "Device %s is broken, too few valid components.", 2572 sc->sc_name); 2573 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 2574 return; 2575 } 2576 /* 2577 * If there is one DIRTY component and all disks are present, 2578 * mark it for synchronization. If there is more than one DIRTY 2579 * component, mark parity component for synchronization. 2580 */ 2581 if (ndisks == sc->sc_ndisks && ndirty == 1) { 2582 for (n = 0; n < sc->sc_ndisks; n++) { 2583 disk = &sc->sc_disks[n]; 2584 if ((disk->d_flags & 2585 G_RAID3_DISK_FLAG_DIRTY) == 0) { 2586 continue; 2587 } 2588 disk->d_flags |= 2589 G_RAID3_DISK_FLAG_SYNCHRONIZING; 2590 } 2591 } else if (ndisks == sc->sc_ndisks && ndirty > 1) { 2592 disk = &sc->sc_disks[sc->sc_ndisks - 1]; 2593 disk->d_flags |= G_RAID3_DISK_FLAG_SYNCHRONIZING; 2594 } 2595 2596 sc->sc_syncid = syncid; 2597 if (force) { 2598 /* Remember to bump syncid on first write. */ 2599 sc->sc_bump_id |= G_RAID3_BUMP_SYNCID; 2600 } 2601 if (ndisks == sc->sc_ndisks) 2602 state = G_RAID3_DEVICE_STATE_COMPLETE; 2603 else /* if (ndisks == sc->sc_ndisks - 1) */ 2604 state = G_RAID3_DEVICE_STATE_DEGRADED; 2605 G_RAID3_DEBUG(1, "Device %s state changed from %s to %s.", 2606 sc->sc_name, g_raid3_device_state2str(sc->sc_state), 2607 g_raid3_device_state2str(state)); 2608 sc->sc_state = state; 2609 for (n = 0; n < sc->sc_ndisks; n++) { 2610 disk = &sc->sc_disks[n]; 2611 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2612 continue; 2613 state = g_raid3_determine_state(disk); 2614 g_raid3_event_send(disk, state, G_RAID3_EVENT_DONTWAIT); 2615 if (state == G_RAID3_DISK_STATE_STALE) 2616 sc->sc_bump_id |= G_RAID3_BUMP_SYNCID; 2617 } 2618 break; 2619 } 2620 case G_RAID3_DEVICE_STATE_DEGRADED: 2621 /* 2622 * Genid need to be bumped immediately, so do it here. 2623 */ 2624 if ((sc->sc_bump_id & G_RAID3_BUMP_GENID) != 0) { 2625 sc->sc_bump_id &= ~G_RAID3_BUMP_GENID; 2626 g_raid3_bump_genid(sc); 2627 } 2628 2629 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NEW) > 0) 2630 return; 2631 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) < 2632 sc->sc_ndisks - 1) { 2633 if (sc->sc_provider != NULL) 2634 g_raid3_destroy_provider(sc); 2635 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 2636 return; 2637 } 2638 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) == 2639 sc->sc_ndisks) { 2640 state = G_RAID3_DEVICE_STATE_COMPLETE; 2641 G_RAID3_DEBUG(1, 2642 "Device %s state changed from %s to %s.", 2643 sc->sc_name, g_raid3_device_state2str(sc->sc_state), 2644 g_raid3_device_state2str(state)); 2645 sc->sc_state = state; 2646 } 2647 if (sc->sc_provider == NULL) 2648 g_raid3_launch_provider(sc); 2649 if (sc->sc_rootmount != NULL) { 2650 G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2651 sc->sc_rootmount); 2652 root_mount_rel(sc->sc_rootmount); 2653 sc->sc_rootmount = NULL; 2654 } 2655 break; 2656 case G_RAID3_DEVICE_STATE_COMPLETE: 2657 /* 2658 * Genid need to be bumped immediately, so do it here. 2659 */ 2660 if ((sc->sc_bump_id & G_RAID3_BUMP_GENID) != 0) { 2661 sc->sc_bump_id &= ~G_RAID3_BUMP_GENID; 2662 g_raid3_bump_genid(sc); 2663 } 2664 2665 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NEW) > 0) 2666 return; 2667 KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) >= 2668 sc->sc_ndisks - 1, 2669 ("Too few ACTIVE components in COMPLETE state (device %s).", 2670 sc->sc_name)); 2671 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) == 2672 sc->sc_ndisks - 1) { 2673 state = G_RAID3_DEVICE_STATE_DEGRADED; 2674 G_RAID3_DEBUG(1, 2675 "Device %s state changed from %s to %s.", 2676 sc->sc_name, g_raid3_device_state2str(sc->sc_state), 2677 g_raid3_device_state2str(state)); 2678 sc->sc_state = state; 2679 } 2680 if (sc->sc_provider == NULL) 2681 g_raid3_launch_provider(sc); 2682 if (sc->sc_rootmount != NULL) { 2683 G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2684 sc->sc_rootmount); 2685 root_mount_rel(sc->sc_rootmount); 2686 sc->sc_rootmount = NULL; 2687 } 2688 break; 2689 default: 2690 KASSERT(1 == 0, ("Wrong device state (%s, %s).", sc->sc_name, 2691 g_raid3_device_state2str(sc->sc_state))); 2692 break; 2693 } 2694 } 2695 2696 /* 2697 * Update disk state and device state if needed. 2698 */ 2699 #define DISK_STATE_CHANGED() G_RAID3_DEBUG(1, \ 2700 "Disk %s state changed from %s to %s (device %s).", \ 2701 g_raid3_get_diskname(disk), \ 2702 g_raid3_disk_state2str(disk->d_state), \ 2703 g_raid3_disk_state2str(state), sc->sc_name) 2704 static int 2705 g_raid3_update_disk(struct g_raid3_disk *disk, u_int state) 2706 { 2707 struct g_raid3_softc *sc; 2708 2709 sc = disk->d_softc; 2710 sx_assert(&sc->sc_lock, SX_XLOCKED); 2711 2712 again: 2713 G_RAID3_DEBUG(3, "Changing disk %s state from %s to %s.", 2714 g_raid3_get_diskname(disk), g_raid3_disk_state2str(disk->d_state), 2715 g_raid3_disk_state2str(state)); 2716 switch (state) { 2717 case G_RAID3_DISK_STATE_NEW: 2718 /* 2719 * Possible scenarios: 2720 * 1. New disk arrive. 2721 */ 2722 /* Previous state should be NONE. */ 2723 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NONE, 2724 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2725 g_raid3_disk_state2str(disk->d_state))); 2726 DISK_STATE_CHANGED(); 2727 2728 disk->d_state = state; 2729 G_RAID3_DEBUG(1, "Device %s: provider %s detected.", 2730 sc->sc_name, g_raid3_get_diskname(disk)); 2731 if (sc->sc_state == G_RAID3_DEVICE_STATE_STARTING) 2732 break; 2733 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2734 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2735 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2736 g_raid3_device_state2str(sc->sc_state), 2737 g_raid3_get_diskname(disk), 2738 g_raid3_disk_state2str(disk->d_state))); 2739 state = g_raid3_determine_state(disk); 2740 if (state != G_RAID3_DISK_STATE_NONE) 2741 goto again; 2742 break; 2743 case G_RAID3_DISK_STATE_ACTIVE: 2744 /* 2745 * Possible scenarios: 2746 * 1. New disk does not need synchronization. 2747 * 2. Synchronization process finished successfully. 2748 */ 2749 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2750 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2751 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2752 g_raid3_device_state2str(sc->sc_state), 2753 g_raid3_get_diskname(disk), 2754 g_raid3_disk_state2str(disk->d_state))); 2755 /* Previous state should be NEW or SYNCHRONIZING. */ 2756 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW || 2757 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING, 2758 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2759 g_raid3_disk_state2str(disk->d_state))); 2760 DISK_STATE_CHANGED(); 2761 2762 if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 2763 disk->d_flags &= ~G_RAID3_DISK_FLAG_SYNCHRONIZING; 2764 disk->d_flags &= ~G_RAID3_DISK_FLAG_FORCE_SYNC; 2765 g_raid3_sync_stop(sc, 0); 2766 } 2767 disk->d_state = state; 2768 disk->d_sync.ds_offset = 0; 2769 disk->d_sync.ds_offset_done = 0; 2770 g_raid3_update_idle(sc, disk); 2771 g_raid3_update_metadata(disk); 2772 G_RAID3_DEBUG(1, "Device %s: provider %s activated.", 2773 sc->sc_name, g_raid3_get_diskname(disk)); 2774 break; 2775 case G_RAID3_DISK_STATE_STALE: 2776 /* 2777 * Possible scenarios: 2778 * 1. Stale disk was connected. 2779 */ 2780 /* Previous state should be NEW. */ 2781 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW, 2782 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2783 g_raid3_disk_state2str(disk->d_state))); 2784 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2785 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2786 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2787 g_raid3_device_state2str(sc->sc_state), 2788 g_raid3_get_diskname(disk), 2789 g_raid3_disk_state2str(disk->d_state))); 2790 /* 2791 * STALE state is only possible if device is marked 2792 * NOAUTOSYNC. 2793 */ 2794 KASSERT((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOAUTOSYNC) != 0, 2795 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2796 g_raid3_device_state2str(sc->sc_state), 2797 g_raid3_get_diskname(disk), 2798 g_raid3_disk_state2str(disk->d_state))); 2799 DISK_STATE_CHANGED(); 2800 2801 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2802 disk->d_state = state; 2803 g_raid3_update_metadata(disk); 2804 G_RAID3_DEBUG(0, "Device %s: provider %s is stale.", 2805 sc->sc_name, g_raid3_get_diskname(disk)); 2806 break; 2807 case G_RAID3_DISK_STATE_SYNCHRONIZING: 2808 /* 2809 * Possible scenarios: 2810 * 1. Disk which needs synchronization was connected. 2811 */ 2812 /* Previous state should be NEW. */ 2813 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW, 2814 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2815 g_raid3_disk_state2str(disk->d_state))); 2816 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2817 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2818 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2819 g_raid3_device_state2str(sc->sc_state), 2820 g_raid3_get_diskname(disk), 2821 g_raid3_disk_state2str(disk->d_state))); 2822 DISK_STATE_CHANGED(); 2823 2824 if (disk->d_state == G_RAID3_DISK_STATE_NEW) 2825 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2826 disk->d_state = state; 2827 if (sc->sc_provider != NULL) { 2828 g_raid3_sync_start(sc); 2829 g_raid3_update_metadata(disk); 2830 } 2831 break; 2832 case G_RAID3_DISK_STATE_DISCONNECTED: 2833 /* 2834 * Possible scenarios: 2835 * 1. Device wasn't running yet, but disk disappear. 2836 * 2. Disk was active and disapppear. 2837 * 3. Disk disappear during synchronization process. 2838 */ 2839 if (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2840 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 2841 /* 2842 * Previous state should be ACTIVE, STALE or 2843 * SYNCHRONIZING. 2844 */ 2845 KASSERT(disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 2846 disk->d_state == G_RAID3_DISK_STATE_STALE || 2847 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING, 2848 ("Wrong disk state (%s, %s).", 2849 g_raid3_get_diskname(disk), 2850 g_raid3_disk_state2str(disk->d_state))); 2851 } else if (sc->sc_state == G_RAID3_DEVICE_STATE_STARTING) { 2852 /* Previous state should be NEW. */ 2853 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW, 2854 ("Wrong disk state (%s, %s).", 2855 g_raid3_get_diskname(disk), 2856 g_raid3_disk_state2str(disk->d_state))); 2857 /* 2858 * Reset bumping syncid if disk disappeared in STARTING 2859 * state. 2860 */ 2861 if ((sc->sc_bump_id & G_RAID3_BUMP_SYNCID) != 0) 2862 sc->sc_bump_id &= ~G_RAID3_BUMP_SYNCID; 2863 #ifdef INVARIANTS 2864 } else { 2865 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2866 sc->sc_name, 2867 g_raid3_device_state2str(sc->sc_state), 2868 g_raid3_get_diskname(disk), 2869 g_raid3_disk_state2str(disk->d_state))); 2870 #endif 2871 } 2872 DISK_STATE_CHANGED(); 2873 G_RAID3_DEBUG(0, "Device %s: provider %s disconnected.", 2874 sc->sc_name, g_raid3_get_diskname(disk)); 2875 2876 g_raid3_destroy_disk(disk); 2877 break; 2878 default: 2879 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2880 break; 2881 } 2882 return (0); 2883 } 2884 #undef DISK_STATE_CHANGED 2885 2886 int 2887 g_raid3_read_metadata(struct g_consumer *cp, struct g_raid3_metadata *md) 2888 { 2889 struct g_provider *pp; 2890 u_char *buf; 2891 int error; 2892 2893 g_topology_assert(); 2894 2895 error = g_access(cp, 1, 0, 0); 2896 if (error != 0) 2897 return (error); 2898 pp = cp->provider; 2899 g_topology_unlock(); 2900 /* Metadata are stored on last sector. */ 2901 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2902 &error); 2903 g_topology_lock(); 2904 g_access(cp, -1, 0, 0); 2905 if (buf == NULL) { 2906 G_RAID3_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2907 cp->provider->name, error); 2908 return (error); 2909 } 2910 2911 /* Decode metadata. */ 2912 error = raid3_metadata_decode(buf, md); 2913 g_free(buf); 2914 if (strcmp(md->md_magic, G_RAID3_MAGIC) != 0) 2915 return (EINVAL); 2916 if (md->md_version > G_RAID3_VERSION) { 2917 G_RAID3_DEBUG(0, 2918 "Kernel module is too old to handle metadata from %s.", 2919 cp->provider->name); 2920 return (EINVAL); 2921 } 2922 if (error != 0) { 2923 G_RAID3_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2924 cp->provider->name); 2925 return (error); 2926 } 2927 if (md->md_sectorsize > maxphys) { 2928 G_RAID3_DEBUG(0, "The blocksize is too big."); 2929 return (EINVAL); 2930 } 2931 2932 return (0); 2933 } 2934 2935 static int 2936 g_raid3_check_metadata(struct g_raid3_softc *sc, struct g_provider *pp, 2937 struct g_raid3_metadata *md) 2938 { 2939 2940 if (md->md_no >= sc->sc_ndisks) { 2941 G_RAID3_DEBUG(1, "Invalid disk %s number (no=%u), skipping.", 2942 pp->name, md->md_no); 2943 return (EINVAL); 2944 } 2945 if (sc->sc_disks[md->md_no].d_state != G_RAID3_DISK_STATE_NODISK) { 2946 G_RAID3_DEBUG(1, "Disk %s (no=%u) already exists, skipping.", 2947 pp->name, md->md_no); 2948 return (EEXIST); 2949 } 2950 if (md->md_all != sc->sc_ndisks) { 2951 G_RAID3_DEBUG(1, 2952 "Invalid '%s' field on disk %s (device %s), skipping.", 2953 "md_all", pp->name, sc->sc_name); 2954 return (EINVAL); 2955 } 2956 if ((md->md_mediasize % md->md_sectorsize) != 0) { 2957 G_RAID3_DEBUG(1, "Invalid metadata (mediasize %% sectorsize != " 2958 "0) on disk %s (device %s), skipping.", pp->name, 2959 sc->sc_name); 2960 return (EINVAL); 2961 } 2962 if (md->md_mediasize != sc->sc_mediasize) { 2963 G_RAID3_DEBUG(1, 2964 "Invalid '%s' field on disk %s (device %s), skipping.", 2965 "md_mediasize", pp->name, sc->sc_name); 2966 return (EINVAL); 2967 } 2968 if ((md->md_mediasize % (sc->sc_ndisks - 1)) != 0) { 2969 G_RAID3_DEBUG(1, 2970 "Invalid '%s' field on disk %s (device %s), skipping.", 2971 "md_mediasize", pp->name, sc->sc_name); 2972 return (EINVAL); 2973 } 2974 if ((sc->sc_mediasize / (sc->sc_ndisks - 1)) > pp->mediasize) { 2975 G_RAID3_DEBUG(1, 2976 "Invalid size of disk %s (device %s), skipping.", pp->name, 2977 sc->sc_name); 2978 return (EINVAL); 2979 } 2980 if ((md->md_sectorsize / pp->sectorsize) < sc->sc_ndisks - 1) { 2981 G_RAID3_DEBUG(1, 2982 "Invalid '%s' field on disk %s (device %s), skipping.", 2983 "md_sectorsize", pp->name, sc->sc_name); 2984 return (EINVAL); 2985 } 2986 if (md->md_sectorsize != sc->sc_sectorsize) { 2987 G_RAID3_DEBUG(1, 2988 "Invalid '%s' field on disk %s (device %s), skipping.", 2989 "md_sectorsize", pp->name, sc->sc_name); 2990 return (EINVAL); 2991 } 2992 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2993 G_RAID3_DEBUG(1, 2994 "Invalid sector size of disk %s (device %s), skipping.", 2995 pp->name, sc->sc_name); 2996 return (EINVAL); 2997 } 2998 if ((md->md_mflags & ~G_RAID3_DEVICE_FLAG_MASK) != 0) { 2999 G_RAID3_DEBUG(1, 3000 "Invalid device flags on disk %s (device %s), skipping.", 3001 pp->name, sc->sc_name); 3002 return (EINVAL); 3003 } 3004 if ((md->md_mflags & G_RAID3_DEVICE_FLAG_VERIFY) != 0 && 3005 (md->md_mflags & G_RAID3_DEVICE_FLAG_ROUND_ROBIN) != 0) { 3006 /* 3007 * VERIFY and ROUND-ROBIN options are mutally exclusive. 3008 */ 3009 G_RAID3_DEBUG(1, "Both VERIFY and ROUND-ROBIN flags exist on " 3010 "disk %s (device %s), skipping.", pp->name, sc->sc_name); 3011 return (EINVAL); 3012 } 3013 if ((md->md_dflags & ~G_RAID3_DISK_FLAG_MASK) != 0) { 3014 G_RAID3_DEBUG(1, 3015 "Invalid disk flags on disk %s (device %s), skipping.", 3016 pp->name, sc->sc_name); 3017 return (EINVAL); 3018 } 3019 return (0); 3020 } 3021 3022 int 3023 g_raid3_add_disk(struct g_raid3_softc *sc, struct g_provider *pp, 3024 struct g_raid3_metadata *md) 3025 { 3026 struct g_raid3_disk *disk; 3027 int error; 3028 3029 g_topology_assert_not(); 3030 G_RAID3_DEBUG(2, "Adding disk %s.", pp->name); 3031 3032 error = g_raid3_check_metadata(sc, pp, md); 3033 if (error != 0) 3034 return (error); 3035 if (sc->sc_state != G_RAID3_DEVICE_STATE_STARTING && 3036 md->md_genid < sc->sc_genid) { 3037 G_RAID3_DEBUG(0, "Component %s (device %s) broken, skipping.", 3038 pp->name, sc->sc_name); 3039 return (EINVAL); 3040 } 3041 disk = g_raid3_init_disk(sc, pp, md, &error); 3042 if (disk == NULL) 3043 return (error); 3044 error = g_raid3_event_send(disk, G_RAID3_DISK_STATE_NEW, 3045 G_RAID3_EVENT_WAIT); 3046 if (error != 0) 3047 return (error); 3048 if (md->md_version < G_RAID3_VERSION) { 3049 G_RAID3_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 3050 pp->name, md->md_version, G_RAID3_VERSION); 3051 g_raid3_update_metadata(disk); 3052 } 3053 return (0); 3054 } 3055 3056 static void 3057 g_raid3_destroy_delayed(void *arg, int flag) 3058 { 3059 struct g_raid3_softc *sc; 3060 int error; 3061 3062 if (flag == EV_CANCEL) { 3063 G_RAID3_DEBUG(1, "Destroying canceled."); 3064 return; 3065 } 3066 sc = arg; 3067 g_topology_unlock(); 3068 sx_xlock(&sc->sc_lock); 3069 KASSERT((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) == 0, 3070 ("DESTROY flag set on %s.", sc->sc_name)); 3071 KASSERT((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROYING) != 0, 3072 ("DESTROYING flag not set on %s.", sc->sc_name)); 3073 G_RAID3_DEBUG(0, "Destroying %s (delayed).", sc->sc_name); 3074 error = g_raid3_destroy(sc, G_RAID3_DESTROY_SOFT); 3075 if (error != 0) { 3076 G_RAID3_DEBUG(0, "Cannot destroy %s.", sc->sc_name); 3077 sx_xunlock(&sc->sc_lock); 3078 } 3079 g_topology_lock(); 3080 } 3081 3082 static int 3083 g_raid3_access(struct g_provider *pp, int acr, int acw, int ace) 3084 { 3085 struct g_raid3_softc *sc; 3086 int dcr, dcw, dce, error = 0; 3087 3088 g_topology_assert(); 3089 G_RAID3_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 3090 acw, ace); 3091 3092 sc = pp->geom->softc; 3093 if (sc == NULL && acr <= 0 && acw <= 0 && ace <= 0) 3094 return (0); 3095 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 3096 3097 dcr = pp->acr + acr; 3098 dcw = pp->acw + acw; 3099 dce = pp->ace + ace; 3100 3101 g_topology_unlock(); 3102 sx_xlock(&sc->sc_lock); 3103 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) != 0 || 3104 g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) < sc->sc_ndisks - 1) { 3105 if (acr > 0 || acw > 0 || ace > 0) 3106 error = ENXIO; 3107 goto end; 3108 } 3109 if (dcw == 0) 3110 g_raid3_idle(sc, dcw); 3111 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROYING) != 0) { 3112 if (acr > 0 || acw > 0 || ace > 0) { 3113 error = ENXIO; 3114 goto end; 3115 } 3116 if (dcr == 0 && dcw == 0 && dce == 0) { 3117 g_post_event(g_raid3_destroy_delayed, sc, M_WAITOK, 3118 sc, NULL); 3119 } 3120 } 3121 end: 3122 sx_xunlock(&sc->sc_lock); 3123 g_topology_lock(); 3124 return (error); 3125 } 3126 3127 static struct g_geom * 3128 g_raid3_create(struct g_class *mp, const struct g_raid3_metadata *md) 3129 { 3130 struct g_raid3_softc *sc; 3131 struct g_geom *gp; 3132 int error, timeout; 3133 u_int n; 3134 3135 g_topology_assert(); 3136 G_RAID3_DEBUG(1, "Creating device %s (id=%u).", md->md_name, md->md_id); 3137 3138 /* One disk is minimum. */ 3139 if (md->md_all < 1) 3140 return (NULL); 3141 /* 3142 * Action geom. 3143 */ 3144 gp = g_new_geomf(mp, "%s", md->md_name); 3145 sc = malloc(sizeof(*sc), M_RAID3, M_WAITOK | M_ZERO); 3146 sc->sc_disks = malloc(sizeof(struct g_raid3_disk) * md->md_all, M_RAID3, 3147 M_WAITOK | M_ZERO); 3148 gp->start = g_raid3_start; 3149 gp->orphan = g_raid3_orphan; 3150 gp->access = g_raid3_access; 3151 gp->dumpconf = g_raid3_dumpconf; 3152 3153 sc->sc_id = md->md_id; 3154 sc->sc_mediasize = md->md_mediasize; 3155 sc->sc_sectorsize = md->md_sectorsize; 3156 sc->sc_ndisks = md->md_all; 3157 sc->sc_round_robin = 0; 3158 sc->sc_flags = md->md_mflags; 3159 sc->sc_bump_id = 0; 3160 sc->sc_idle = 1; 3161 sc->sc_last_write = time_uptime; 3162 sc->sc_writes = 0; 3163 for (n = 0; n < sc->sc_ndisks; n++) { 3164 sc->sc_disks[n].d_softc = sc; 3165 sc->sc_disks[n].d_no = n; 3166 sc->sc_disks[n].d_state = G_RAID3_DISK_STATE_NODISK; 3167 } 3168 sx_init(&sc->sc_lock, "graid3:lock"); 3169 bioq_init(&sc->sc_queue); 3170 mtx_init(&sc->sc_queue_mtx, "graid3:queue", NULL, MTX_DEF); 3171 bioq_init(&sc->sc_regular_delayed); 3172 bioq_init(&sc->sc_inflight); 3173 bioq_init(&sc->sc_sync_delayed); 3174 TAILQ_INIT(&sc->sc_events); 3175 mtx_init(&sc->sc_events_mtx, "graid3:events", NULL, MTX_DEF); 3176 callout_init(&sc->sc_callout, 1); 3177 sc->sc_state = G_RAID3_DEVICE_STATE_STARTING; 3178 gp->softc = sc; 3179 sc->sc_geom = gp; 3180 sc->sc_provider = NULL; 3181 /* 3182 * Synchronization geom. 3183 */ 3184 gp = g_new_geomf(mp, "%s.sync", md->md_name); 3185 gp->softc = sc; 3186 gp->orphan = g_raid3_orphan; 3187 sc->sc_sync.ds_geom = gp; 3188 3189 if (!g_raid3_use_malloc) { 3190 sc->sc_zones[G_RAID3_ZONE_64K].sz_zone = uma_zcreate("gr3:64k", 3191 65536, g_raid3_uma_ctor, g_raid3_uma_dtor, NULL, NULL, 3192 UMA_ALIGN_PTR, 0); 3193 sc->sc_zones[G_RAID3_ZONE_64K].sz_inuse = 0; 3194 sc->sc_zones[G_RAID3_ZONE_64K].sz_max = g_raid3_n64k; 3195 sc->sc_zones[G_RAID3_ZONE_64K].sz_requested = 3196 sc->sc_zones[G_RAID3_ZONE_64K].sz_failed = 0; 3197 sc->sc_zones[G_RAID3_ZONE_16K].sz_zone = uma_zcreate("gr3:16k", 3198 16384, g_raid3_uma_ctor, g_raid3_uma_dtor, NULL, NULL, 3199 UMA_ALIGN_PTR, 0); 3200 sc->sc_zones[G_RAID3_ZONE_16K].sz_inuse = 0; 3201 sc->sc_zones[G_RAID3_ZONE_16K].sz_max = g_raid3_n16k; 3202 sc->sc_zones[G_RAID3_ZONE_16K].sz_requested = 3203 sc->sc_zones[G_RAID3_ZONE_16K].sz_failed = 0; 3204 sc->sc_zones[G_RAID3_ZONE_4K].sz_zone = uma_zcreate("gr3:4k", 3205 4096, g_raid3_uma_ctor, g_raid3_uma_dtor, NULL, NULL, 3206 UMA_ALIGN_PTR, 0); 3207 sc->sc_zones[G_RAID3_ZONE_4K].sz_inuse = 0; 3208 sc->sc_zones[G_RAID3_ZONE_4K].sz_max = g_raid3_n4k; 3209 sc->sc_zones[G_RAID3_ZONE_4K].sz_requested = 3210 sc->sc_zones[G_RAID3_ZONE_4K].sz_failed = 0; 3211 } 3212 3213 error = kproc_create(g_raid3_worker, sc, &sc->sc_worker, 0, 0, 3214 "g_raid3 %s", md->md_name); 3215 if (error != 0) { 3216 G_RAID3_DEBUG(1, "Cannot create kernel thread for %s.", 3217 sc->sc_name); 3218 if (!g_raid3_use_malloc) { 3219 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_64K].sz_zone); 3220 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_16K].sz_zone); 3221 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_4K].sz_zone); 3222 } 3223 g_destroy_geom(sc->sc_sync.ds_geom); 3224 mtx_destroy(&sc->sc_events_mtx); 3225 mtx_destroy(&sc->sc_queue_mtx); 3226 sx_destroy(&sc->sc_lock); 3227 g_destroy_geom(sc->sc_geom); 3228 free(sc->sc_disks, M_RAID3); 3229 free(sc, M_RAID3); 3230 return (NULL); 3231 } 3232 3233 G_RAID3_DEBUG(1, "Device %s created (%u components, id=%u).", 3234 sc->sc_name, sc->sc_ndisks, sc->sc_id); 3235 3236 sc->sc_rootmount = root_mount_hold("GRAID3"); 3237 G_RAID3_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 3238 3239 /* 3240 * Schedule startup timeout. 3241 */ 3242 timeout = atomic_load_acq_int(&g_raid3_timeout); 3243 sc->sc_timeout_event = malloc(sizeof(struct g_raid3_event), M_RAID3, 3244 M_WAITOK); 3245 callout_reset(&sc->sc_callout, timeout * hz, g_raid3_go, sc); 3246 return (sc->sc_geom); 3247 } 3248 3249 int 3250 g_raid3_destroy(struct g_raid3_softc *sc, int how) 3251 { 3252 struct g_provider *pp; 3253 3254 g_topology_assert_not(); 3255 if (sc == NULL) 3256 return (ENXIO); 3257 sx_assert(&sc->sc_lock, SX_XLOCKED); 3258 3259 pp = sc->sc_provider; 3260 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 3261 switch (how) { 3262 case G_RAID3_DESTROY_SOFT: 3263 G_RAID3_DEBUG(1, 3264 "Device %s is still open (r%dw%de%d).", pp->name, 3265 pp->acr, pp->acw, pp->ace); 3266 return (EBUSY); 3267 case G_RAID3_DESTROY_DELAYED: 3268 G_RAID3_DEBUG(1, 3269 "Device %s will be destroyed on last close.", 3270 pp->name); 3271 if (sc->sc_syncdisk != NULL) 3272 g_raid3_sync_stop(sc, 1); 3273 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROYING; 3274 return (EBUSY); 3275 case G_RAID3_DESTROY_HARD: 3276 G_RAID3_DEBUG(1, "Device %s is still open, so it " 3277 "can't be definitely removed.", pp->name); 3278 break; 3279 } 3280 } 3281 3282 g_topology_lock(); 3283 if (sc->sc_geom->softc == NULL) { 3284 g_topology_unlock(); 3285 return (0); 3286 } 3287 sc->sc_geom->softc = NULL; 3288 sc->sc_sync.ds_geom->softc = NULL; 3289 g_topology_unlock(); 3290 3291 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 3292 sc->sc_flags |= G_RAID3_DEVICE_FLAG_WAIT; 3293 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3294 sx_xunlock(&sc->sc_lock); 3295 mtx_lock(&sc->sc_queue_mtx); 3296 wakeup(sc); 3297 wakeup(&sc->sc_queue); 3298 mtx_unlock(&sc->sc_queue_mtx); 3299 G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3300 while (sc->sc_worker != NULL) 3301 tsleep(&sc->sc_worker, PRIBIO, "r3:destroy", hz / 5); 3302 G_RAID3_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3303 sx_xlock(&sc->sc_lock); 3304 g_raid3_destroy_device(sc); 3305 free(sc->sc_disks, M_RAID3); 3306 free(sc, M_RAID3); 3307 return (0); 3308 } 3309 3310 static void 3311 g_raid3_taste_orphan(struct g_consumer *cp) 3312 { 3313 3314 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3315 cp->provider->name)); 3316 } 3317 3318 static struct g_geom * 3319 g_raid3_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3320 { 3321 struct g_raid3_metadata md; 3322 struct g_raid3_softc *sc; 3323 struct g_consumer *cp; 3324 struct g_geom *gp; 3325 int error; 3326 3327 g_topology_assert(); 3328 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3329 G_RAID3_DEBUG(2, "Tasting %s.", pp->name); 3330 3331 gp = g_new_geomf(mp, "raid3:taste"); 3332 /* This orphan function should be never called. */ 3333 gp->orphan = g_raid3_taste_orphan; 3334 cp = g_new_consumer(gp); 3335 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 3336 error = g_attach(cp, pp); 3337 if (error == 0) { 3338 error = g_raid3_read_metadata(cp, &md); 3339 g_detach(cp); 3340 } 3341 g_destroy_consumer(cp); 3342 g_destroy_geom(gp); 3343 if (error != 0) 3344 return (NULL); 3345 gp = NULL; 3346 3347 if (md.md_provider[0] != '\0' && 3348 !g_compare_names(md.md_provider, pp->name)) 3349 return (NULL); 3350 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3351 return (NULL); 3352 if (g_raid3_debug >= 2) 3353 raid3_metadata_dump(&md); 3354 3355 /* 3356 * Let's check if device already exists. 3357 */ 3358 sc = NULL; 3359 LIST_FOREACH(gp, &mp->geom, geom) { 3360 sc = gp->softc; 3361 if (sc == NULL) 3362 continue; 3363 if (sc->sc_sync.ds_geom == gp) 3364 continue; 3365 if (strcmp(md.md_name, sc->sc_name) != 0) 3366 continue; 3367 if (md.md_id != sc->sc_id) { 3368 G_RAID3_DEBUG(0, "Device %s already configured.", 3369 sc->sc_name); 3370 return (NULL); 3371 } 3372 break; 3373 } 3374 if (gp == NULL) { 3375 gp = g_raid3_create(mp, &md); 3376 if (gp == NULL) { 3377 G_RAID3_DEBUG(0, "Cannot create device %s.", 3378 md.md_name); 3379 return (NULL); 3380 } 3381 sc = gp->softc; 3382 } 3383 G_RAID3_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3384 g_topology_unlock(); 3385 sx_xlock(&sc->sc_lock); 3386 error = g_raid3_add_disk(sc, pp, &md); 3387 if (error != 0) { 3388 G_RAID3_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3389 pp->name, gp->name, error); 3390 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NODISK) == 3391 sc->sc_ndisks) { 3392 g_cancel_event(sc); 3393 g_raid3_destroy(sc, G_RAID3_DESTROY_HARD); 3394 g_topology_lock(); 3395 return (NULL); 3396 } 3397 gp = NULL; 3398 } 3399 sx_xunlock(&sc->sc_lock); 3400 g_topology_lock(); 3401 return (gp); 3402 } 3403 3404 static int 3405 g_raid3_destroy_geom(struct gctl_req *req __unused, struct g_class *mp __unused, 3406 struct g_geom *gp) 3407 { 3408 struct g_raid3_softc *sc; 3409 int error; 3410 3411 g_topology_unlock(); 3412 sc = gp->softc; 3413 sx_xlock(&sc->sc_lock); 3414 g_cancel_event(sc); 3415 error = g_raid3_destroy(gp->softc, G_RAID3_DESTROY_SOFT); 3416 if (error != 0) 3417 sx_xunlock(&sc->sc_lock); 3418 g_topology_lock(); 3419 return (error); 3420 } 3421 3422 static void 3423 g_raid3_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3424 struct g_consumer *cp, struct g_provider *pp) 3425 { 3426 struct g_raid3_softc *sc; 3427 3428 g_topology_assert(); 3429 3430 sc = gp->softc; 3431 if (sc == NULL) 3432 return; 3433 /* Skip synchronization geom. */ 3434 if (gp == sc->sc_sync.ds_geom) 3435 return; 3436 if (pp != NULL) { 3437 /* Nothing here. */ 3438 } else if (cp != NULL) { 3439 struct g_raid3_disk *disk; 3440 3441 disk = cp->private; 3442 if (disk == NULL) 3443 return; 3444 g_topology_unlock(); 3445 sx_xlock(&sc->sc_lock); 3446 sbuf_printf(sb, "%s<Type>", indent); 3447 if (disk->d_no == sc->sc_ndisks - 1) 3448 sbuf_cat(sb, "PARITY"); 3449 else 3450 sbuf_cat(sb, "DATA"); 3451 sbuf_cat(sb, "</Type>\n"); 3452 sbuf_printf(sb, "%s<Number>%u</Number>\n", indent, 3453 (u_int)disk->d_no); 3454 if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 3455 sbuf_printf(sb, "%s<Synchronized>", indent); 3456 if (disk->d_sync.ds_offset == 0) 3457 sbuf_cat(sb, "0%"); 3458 else { 3459 sbuf_printf(sb, "%u%%", 3460 (u_int)((disk->d_sync.ds_offset * 100) / 3461 (sc->sc_mediasize / (sc->sc_ndisks - 1)))); 3462 } 3463 sbuf_cat(sb, "</Synchronized>\n"); 3464 if (disk->d_sync.ds_offset > 0) { 3465 sbuf_printf(sb, "%s<BytesSynced>%jd" 3466 "</BytesSynced>\n", indent, 3467 (intmax_t)disk->d_sync.ds_offset); 3468 } 3469 } 3470 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3471 disk->d_sync.ds_syncid); 3472 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, disk->d_genid); 3473 sbuf_printf(sb, "%s<Flags>", indent); 3474 if (disk->d_flags == 0) 3475 sbuf_cat(sb, "NONE"); 3476 else { 3477 int first = 1; 3478 3479 #define ADD_FLAG(flag, name) do { \ 3480 if ((disk->d_flags & (flag)) != 0) { \ 3481 if (!first) \ 3482 sbuf_cat(sb, ", "); \ 3483 else \ 3484 first = 0; \ 3485 sbuf_cat(sb, name); \ 3486 } \ 3487 } while (0) 3488 ADD_FLAG(G_RAID3_DISK_FLAG_DIRTY, "DIRTY"); 3489 ADD_FLAG(G_RAID3_DISK_FLAG_HARDCODED, "HARDCODED"); 3490 ADD_FLAG(G_RAID3_DISK_FLAG_SYNCHRONIZING, 3491 "SYNCHRONIZING"); 3492 ADD_FLAG(G_RAID3_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3493 ADD_FLAG(G_RAID3_DISK_FLAG_BROKEN, "BROKEN"); 3494 #undef ADD_FLAG 3495 } 3496 sbuf_cat(sb, "</Flags>\n"); 3497 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3498 g_raid3_disk_state2str(disk->d_state)); 3499 sx_xunlock(&sc->sc_lock); 3500 g_topology_lock(); 3501 } else { 3502 g_topology_unlock(); 3503 sx_xlock(&sc->sc_lock); 3504 if (!g_raid3_use_malloc) { 3505 sbuf_printf(sb, 3506 "%s<Zone4kRequested>%u</Zone4kRequested>\n", indent, 3507 sc->sc_zones[G_RAID3_ZONE_4K].sz_requested); 3508 sbuf_printf(sb, 3509 "%s<Zone4kFailed>%u</Zone4kFailed>\n", indent, 3510 sc->sc_zones[G_RAID3_ZONE_4K].sz_failed); 3511 sbuf_printf(sb, 3512 "%s<Zone16kRequested>%u</Zone16kRequested>\n", indent, 3513 sc->sc_zones[G_RAID3_ZONE_16K].sz_requested); 3514 sbuf_printf(sb, 3515 "%s<Zone16kFailed>%u</Zone16kFailed>\n", indent, 3516 sc->sc_zones[G_RAID3_ZONE_16K].sz_failed); 3517 sbuf_printf(sb, 3518 "%s<Zone64kRequested>%u</Zone64kRequested>\n", indent, 3519 sc->sc_zones[G_RAID3_ZONE_64K].sz_requested); 3520 sbuf_printf(sb, 3521 "%s<Zone64kFailed>%u</Zone64kFailed>\n", indent, 3522 sc->sc_zones[G_RAID3_ZONE_64K].sz_failed); 3523 } 3524 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3525 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3526 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3527 sbuf_printf(sb, "%s<Flags>", indent); 3528 if (sc->sc_flags == 0) 3529 sbuf_cat(sb, "NONE"); 3530 else { 3531 int first = 1; 3532 3533 #define ADD_FLAG(flag, name) do { \ 3534 if ((sc->sc_flags & (flag)) != 0) { \ 3535 if (!first) \ 3536 sbuf_cat(sb, ", "); \ 3537 else \ 3538 first = 0; \ 3539 sbuf_cat(sb, name); \ 3540 } \ 3541 } while (0) 3542 ADD_FLAG(G_RAID3_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3543 ADD_FLAG(G_RAID3_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3544 ADD_FLAG(G_RAID3_DEVICE_FLAG_ROUND_ROBIN, 3545 "ROUND-ROBIN"); 3546 ADD_FLAG(G_RAID3_DEVICE_FLAG_VERIFY, "VERIFY"); 3547 #undef ADD_FLAG 3548 } 3549 sbuf_cat(sb, "</Flags>\n"); 3550 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3551 sc->sc_ndisks); 3552 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3553 g_raid3_device_state2str(sc->sc_state)); 3554 sx_xunlock(&sc->sc_lock); 3555 g_topology_lock(); 3556 } 3557 } 3558 3559 static void 3560 g_raid3_shutdown_post_sync(void *arg, int howto) 3561 { 3562 struct g_class *mp; 3563 struct g_geom *gp, *gp2; 3564 struct g_raid3_softc *sc; 3565 int error; 3566 3567 mp = arg; 3568 g_topology_lock(); 3569 g_raid3_shutdown = 1; 3570 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3571 if ((sc = gp->softc) == NULL) 3572 continue; 3573 /* Skip synchronization geom. */ 3574 if (gp == sc->sc_sync.ds_geom) 3575 continue; 3576 g_topology_unlock(); 3577 sx_xlock(&sc->sc_lock); 3578 g_raid3_idle(sc, -1); 3579 g_cancel_event(sc); 3580 error = g_raid3_destroy(sc, G_RAID3_DESTROY_DELAYED); 3581 if (error != 0) 3582 sx_xunlock(&sc->sc_lock); 3583 g_topology_lock(); 3584 } 3585 g_topology_unlock(); 3586 } 3587 3588 static void 3589 g_raid3_init(struct g_class *mp) 3590 { 3591 3592 g_raid3_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3593 g_raid3_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3594 if (g_raid3_post_sync == NULL) 3595 G_RAID3_DEBUG(0, "Warning! Cannot register shutdown event."); 3596 } 3597 3598 static void 3599 g_raid3_fini(struct g_class *mp) 3600 { 3601 3602 if (g_raid3_post_sync != NULL) 3603 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid3_post_sync); 3604 } 3605 3606 DECLARE_GEOM_CLASS(g_raid3_class, g_raid3); 3607 MODULE_VERSION(geom_raid3, 0); 3608