g_raid3.c revision 160081
1/*- 2 * Copyright (c) 2004-2006 Pawel Jakub Dawidek <pjd@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27#include <sys/cdefs.h> 28__FBSDID("$FreeBSD: head/sys/geom/raid3/g_raid3.c 160081 2006-07-03 10:32:38Z pjd $"); 29 30#include <sys/param.h> 31#include <sys/systm.h> 32#include <sys/kernel.h> 33#include <sys/module.h> 34#include <sys/limits.h> 35#include <sys/lock.h> 36#include <sys/mutex.h> 37#include <sys/bio.h> 38#include <sys/sysctl.h> 39#include <sys/malloc.h> 40#include <sys/eventhandler.h> 41#include <vm/uma.h> 42#include <geom/geom.h> 43#include <sys/proc.h> 44#include <sys/kthread.h> 45#include <sys/sched.h> 46#include <geom/raid3/g_raid3.h> 47 48 49static MALLOC_DEFINE(M_RAID3, "raid3_data", "GEOM_RAID3 Data"); 50 51SYSCTL_DECL(_kern_geom); 52SYSCTL_NODE(_kern_geom, OID_AUTO, raid3, CTLFLAG_RW, 0, "GEOM_RAID3 stuff"); 53u_int g_raid3_debug = 0; 54TUNABLE_INT("kern.geom.raid3.debug", &g_raid3_debug); 55SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, debug, CTLFLAG_RW, &g_raid3_debug, 0, 56 "Debug level"); 57static u_int g_raid3_timeout = 4; 58TUNABLE_INT("kern.geom.raid3.timeout", &g_raid3_timeout); 59SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, timeout, CTLFLAG_RW, &g_raid3_timeout, 60 0, "Time to wait on all raid3 components"); 61static u_int g_raid3_idletime = 5; 62TUNABLE_INT("kern.geom.raid3.idletime", &g_raid3_idletime); 63SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, idletime, CTLFLAG_RW, 64 &g_raid3_idletime, 0, "Mark components as clean when idling"); 65static u_int g_raid3_disconnect_on_failure = 1; 66TUNABLE_INT("kern.geom.raid3.disconnect_on_failure", 67 &g_raid3_disconnect_on_failure); 68SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, disconnect_on_failure, CTLFLAG_RW, 69 &g_raid3_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 70static u_int g_raid3_syncreqs = 2; 71TUNABLE_INT("kern.geom.raid3.sync_requests", &g_raid3_syncreqs); 72SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, sync_requests, CTLFLAG_RDTUN, 73 &g_raid3_syncreqs, 0, "Parallel synchronization I/O requests."); 74 75static u_int g_raid3_n64k = 50; 76TUNABLE_INT("kern.geom.raid3.n64k", &g_raid3_n64k); 77SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n64k, CTLFLAG_RD, &g_raid3_n64k, 0, 78 "Maximum number of 64kB allocations"); 79static u_int g_raid3_n16k = 200; 80TUNABLE_INT("kern.geom.raid3.n16k", &g_raid3_n16k); 81SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n16k, CTLFLAG_RD, &g_raid3_n16k, 0, 82 "Maximum number of 16kB allocations"); 83static u_int g_raid3_n4k = 1200; 84TUNABLE_INT("kern.geom.raid3.n4k", &g_raid3_n4k); 85SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n4k, CTLFLAG_RD, &g_raid3_n4k, 0, 86 "Maximum number of 4kB allocations"); 87 88SYSCTL_NODE(_kern_geom_raid3, OID_AUTO, stat, CTLFLAG_RW, 0, 89 "GEOM_RAID3 statistics"); 90static u_int g_raid3_parity_mismatch = 0; 91SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, parity_mismatch, CTLFLAG_RD, 92 &g_raid3_parity_mismatch, 0, "Number of failures in VERIFY mode"); 93 94#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 95 G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 96 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 97 G_RAID3_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 98} while (0) 99 100static eventhandler_tag g_raid3_pre_sync = NULL; 101 102static int g_raid3_destroy_geom(struct gctl_req *req, struct g_class *mp, 103 struct g_geom *gp); 104static g_taste_t g_raid3_taste; 105static void g_raid3_init(struct g_class *mp); 106static void g_raid3_fini(struct g_class *mp); 107 108struct 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 119static void g_raid3_destroy_provider(struct g_raid3_softc *sc); 120static int g_raid3_update_disk(struct g_raid3_disk *disk, u_int state); 121static void g_raid3_update_device(struct g_raid3_softc *sc, boolean_t force); 122static void g_raid3_dumpconf(struct sbuf *sb, const char *indent, 123 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 124static void g_raid3_sync_stop(struct g_raid3_softc *sc, int type); 125static int g_raid3_register_request(struct bio *pbp); 126static void g_raid3_sync_release(struct g_raid3_softc *sc); 127 128 129static const char * 130g_raid3_disk_state2str(int state) 131{ 132 133 switch (state) { 134 case G_RAID3_DISK_STATE_NODISK: 135 return ("NODISK"); 136 case G_RAID3_DISK_STATE_NONE: 137 return ("NONE"); 138 case G_RAID3_DISK_STATE_NEW: 139 return ("NEW"); 140 case G_RAID3_DISK_STATE_ACTIVE: 141 return ("ACTIVE"); 142 case G_RAID3_DISK_STATE_STALE: 143 return ("STALE"); 144 case G_RAID3_DISK_STATE_SYNCHRONIZING: 145 return ("SYNCHRONIZING"); 146 case G_RAID3_DISK_STATE_DISCONNECTED: 147 return ("DISCONNECTED"); 148 default: 149 return ("INVALID"); 150 } 151} 152 153static const char * 154g_raid3_device_state2str(int state) 155{ 156 157 switch (state) { 158 case G_RAID3_DEVICE_STATE_STARTING: 159 return ("STARTING"); 160 case G_RAID3_DEVICE_STATE_DEGRADED: 161 return ("DEGRADED"); 162 case G_RAID3_DEVICE_STATE_COMPLETE: 163 return ("COMPLETE"); 164 default: 165 return ("INVALID"); 166 } 167} 168 169const char * 170g_raid3_get_diskname(struct g_raid3_disk *disk) 171{ 172 173 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 174 return ("[unknown]"); 175 return (disk->d_name); 176} 177 178static int 179g_raid3_uma_ctor(void *mem, int size, void *arg, int flags) 180{ 181 struct g_raid3_zone *sz = arg; 182 183 if (sz->sz_max > 0 && sz->sz_inuse == sz->sz_max) 184 return (ENOMEM); 185 sz->sz_inuse++; 186 return (0); 187} 188 189static void 190g_raid3_uma_dtor(void *mem, int size, void *arg) 191{ 192 struct g_raid3_zone *sz = arg; 193 194 sz->sz_inuse--; 195} 196 197#define g_raid3_xor(src1, src2, dst, size) \ 198 _g_raid3_xor((uint64_t *)(src1), (uint64_t *)(src2), \ 199 (uint64_t *)(dst), (size_t)size) 200static void 201_g_raid3_xor(uint64_t *src1, uint64_t *src2, uint64_t *dst, size_t size) 202{ 203 204 KASSERT((size % 128) == 0, ("Invalid size: %zu.", size)); 205 for (; size > 0; size -= 128) { 206 *dst++ = (*src1++) ^ (*src2++); 207 *dst++ = (*src1++) ^ (*src2++); 208 *dst++ = (*src1++) ^ (*src2++); 209 *dst++ = (*src1++) ^ (*src2++); 210 *dst++ = (*src1++) ^ (*src2++); 211 *dst++ = (*src1++) ^ (*src2++); 212 *dst++ = (*src1++) ^ (*src2++); 213 *dst++ = (*src1++) ^ (*src2++); 214 *dst++ = (*src1++) ^ (*src2++); 215 *dst++ = (*src1++) ^ (*src2++); 216 *dst++ = (*src1++) ^ (*src2++); 217 *dst++ = (*src1++) ^ (*src2++); 218 *dst++ = (*src1++) ^ (*src2++); 219 *dst++ = (*src1++) ^ (*src2++); 220 *dst++ = (*src1++) ^ (*src2++); 221 *dst++ = (*src1++) ^ (*src2++); 222 } 223} 224 225static int 226g_raid3_is_zero(struct bio *bp) 227{ 228 static const uint64_t zeros[] = { 229 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 230 }; 231 u_char *addr; 232 ssize_t size; 233 234 size = bp->bio_length; 235 addr = (u_char *)bp->bio_data; 236 for (; size > 0; size -= sizeof(zeros), addr += sizeof(zeros)) { 237 if (bcmp(addr, zeros, sizeof(zeros)) != 0) 238 return (0); 239 } 240 return (1); 241} 242 243/* 244 * --- Events handling functions --- 245 * Events in geom_raid3 are used to maintain disks and device status 246 * from one thread to simplify locking. 247 */ 248static void 249g_raid3_event_free(struct g_raid3_event *ep) 250{ 251 252 free(ep, M_RAID3); 253} 254 255int 256g_raid3_event_send(void *arg, int state, int flags) 257{ 258 struct g_raid3_softc *sc; 259 struct g_raid3_disk *disk; 260 struct g_raid3_event *ep; 261 int error; 262 263 ep = malloc(sizeof(*ep), M_RAID3, M_WAITOK); 264 G_RAID3_DEBUG(4, "%s: Sending event %p.", __func__, ep); 265 if ((flags & G_RAID3_EVENT_DEVICE) != 0) { 266 disk = NULL; 267 sc = arg; 268 } else { 269 disk = arg; 270 sc = disk->d_softc; 271 } 272 ep->e_disk = disk; 273 ep->e_state = state; 274 ep->e_flags = flags; 275 ep->e_error = 0; 276 mtx_lock(&sc->sc_events_mtx); 277 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 278 mtx_unlock(&sc->sc_events_mtx); 279 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc); 280 mtx_lock(&sc->sc_queue_mtx); 281 wakeup(sc); 282 wakeup(&sc->sc_queue); 283 mtx_unlock(&sc->sc_queue_mtx); 284 if ((flags & G_RAID3_EVENT_DONTWAIT) != 0) 285 return (0); 286 sx_assert(&sc->sc_lock, SX_XLOCKED); 287 G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 288 sx_xunlock(&sc->sc_lock); 289 while ((ep->e_flags & G_RAID3_EVENT_DONE) == 0) { 290 mtx_lock(&sc->sc_events_mtx); 291 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "r3:event", 292 hz * 5); 293 } 294 error = ep->e_error; 295 g_raid3_event_free(ep); 296 sx_xlock(&sc->sc_lock); 297 return (error); 298} 299 300static struct g_raid3_event * 301g_raid3_event_get(struct g_raid3_softc *sc) 302{ 303 struct g_raid3_event *ep; 304 305 mtx_lock(&sc->sc_events_mtx); 306 ep = TAILQ_FIRST(&sc->sc_events); 307 mtx_unlock(&sc->sc_events_mtx); 308 return (ep); 309} 310 311static void 312g_raid3_event_remove(struct g_raid3_softc *sc, struct g_raid3_event *ep) 313{ 314 315 mtx_lock(&sc->sc_events_mtx); 316 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 317 mtx_unlock(&sc->sc_events_mtx); 318} 319 320static void 321g_raid3_event_cancel(struct g_raid3_disk *disk) 322{ 323 struct g_raid3_softc *sc; 324 struct g_raid3_event *ep, *tmpep; 325 326 sc = disk->d_softc; 327 sx_assert(&sc->sc_lock, SX_XLOCKED); 328 329 mtx_lock(&sc->sc_events_mtx); 330 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 331 if ((ep->e_flags & G_RAID3_EVENT_DEVICE) != 0) 332 continue; 333 if (ep->e_disk != disk) 334 continue; 335 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 336 if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0) 337 g_raid3_event_free(ep); 338 else { 339 ep->e_error = ECANCELED; 340 wakeup(ep); 341 } 342 } 343 mtx_unlock(&sc->sc_events_mtx); 344} 345 346/* 347 * Return the number of disks in the given state. 348 * If state is equal to -1, count all connected disks. 349 */ 350u_int 351g_raid3_ndisks(struct g_raid3_softc *sc, int state) 352{ 353 struct g_raid3_disk *disk; 354 u_int n, ndisks; 355 356 sx_assert(&sc->sc_lock, SX_LOCKED); 357 358 for (n = ndisks = 0; n < sc->sc_ndisks; n++) { 359 disk = &sc->sc_disks[n]; 360 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 361 continue; 362 if (state == -1 || disk->d_state == state) 363 ndisks++; 364 } 365 return (ndisks); 366} 367 368static u_int 369g_raid3_nrequests(struct g_raid3_softc *sc, struct g_consumer *cp) 370{ 371 struct bio *bp; 372 u_int nreqs = 0; 373 374 mtx_lock(&sc->sc_queue_mtx); 375 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 376 if (bp->bio_from == cp) 377 nreqs++; 378 } 379 mtx_unlock(&sc->sc_queue_mtx); 380 return (nreqs); 381} 382 383static int 384g_raid3_is_busy(struct g_raid3_softc *sc, struct g_consumer *cp) 385{ 386 387 if (cp->index > 0) { 388 G_RAID3_DEBUG(2, 389 "I/O requests for %s exist, can't destroy it now.", 390 cp->provider->name); 391 return (1); 392 } 393 if (g_raid3_nrequests(sc, cp) > 0) { 394 G_RAID3_DEBUG(2, 395 "I/O requests for %s in queue, can't destroy it now.", 396 cp->provider->name); 397 return (1); 398 } 399 return (0); 400} 401 402static void 403g_raid3_destroy_consumer(void *arg, int flags __unused) 404{ 405 struct g_consumer *cp; 406 407 g_topology_assert(); 408 409 cp = arg; 410 G_RAID3_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 411 g_detach(cp); 412 g_destroy_consumer(cp); 413} 414 415static void 416g_raid3_kill_consumer(struct g_raid3_softc *sc, struct g_consumer *cp) 417{ 418 struct g_provider *pp; 419 int retaste_wait; 420 421 g_topology_assert(); 422 423 cp->private = NULL; 424 if (g_raid3_is_busy(sc, cp)) 425 return; 426 G_RAID3_DEBUG(2, "Consumer %s destroyed.", cp->provider->name); 427 pp = cp->provider; 428 retaste_wait = 0; 429 if (cp->acw == 1) { 430 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 431 retaste_wait = 1; 432 } 433 G_RAID3_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 434 -cp->acw, -cp->ace, 0); 435 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 436 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 437 if (retaste_wait) { 438 /* 439 * After retaste event was send (inside g_access()), we can send 440 * event to detach and destroy consumer. 441 * A class, which has consumer to the given provider connected 442 * will not receive retaste event for the provider. 443 * This is the way how I ignore retaste events when I close 444 * consumers opened for write: I detach and destroy consumer 445 * after retaste event is sent. 446 */ 447 g_post_event(g_raid3_destroy_consumer, cp, M_WAITOK, NULL); 448 return; 449 } 450 G_RAID3_DEBUG(1, "Consumer %s destroyed.", pp->name); 451 g_detach(cp); 452 g_destroy_consumer(cp); 453} 454 455static int 456g_raid3_connect_disk(struct g_raid3_disk *disk, struct g_provider *pp) 457{ 458 struct g_consumer *cp; 459 int error; 460 461 g_topology_assert_not(); 462 KASSERT(disk->d_consumer == NULL, 463 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 464 465 g_topology_lock(); 466 cp = g_new_consumer(disk->d_softc->sc_geom); 467 error = g_attach(cp, pp); 468 if (error != 0) { 469 g_destroy_consumer(cp); 470 g_topology_unlock(); 471 return (error); 472 } 473 error = g_access(cp, 1, 1, 1); 474 g_topology_unlock(); 475 if (error != 0) { 476 g_detach(cp); 477 g_destroy_consumer(cp); 478 G_RAID3_DEBUG(0, "Cannot open consumer %s (error=%d).", 479 pp->name, error); 480 return (error); 481 } 482 disk->d_consumer = cp; 483 disk->d_consumer->private = disk; 484 disk->d_consumer->index = 0; 485 G_RAID3_DEBUG(2, "Disk %s connected.", g_raid3_get_diskname(disk)); 486 return (0); 487} 488 489static void 490g_raid3_disconnect_consumer(struct g_raid3_softc *sc, struct g_consumer *cp) 491{ 492 493 g_topology_assert(); 494 495 if (cp == NULL) 496 return; 497 if (cp->provider != NULL) 498 g_raid3_kill_consumer(sc, cp); 499 else 500 g_destroy_consumer(cp); 501} 502 503/* 504 * Initialize disk. This means allocate memory, create consumer, attach it 505 * to the provider and open access (r1w1e1) to it. 506 */ 507static struct g_raid3_disk * 508g_raid3_init_disk(struct g_raid3_softc *sc, struct g_provider *pp, 509 struct g_raid3_metadata *md, int *errorp) 510{ 511 struct g_raid3_disk *disk; 512 int error; 513 514 disk = &sc->sc_disks[md->md_no]; 515 error = g_raid3_connect_disk(disk, pp); 516 if (error != 0) { 517 if (errorp != NULL) 518 *errorp = error; 519 return (NULL); 520 } 521 disk->d_state = G_RAID3_DISK_STATE_NONE; 522 disk->d_flags = md->md_dflags; 523 if (md->md_provider[0] != '\0') 524 disk->d_flags |= G_RAID3_DISK_FLAG_HARDCODED; 525 disk->d_sync.ds_consumer = NULL; 526 disk->d_sync.ds_offset = md->md_sync_offset; 527 disk->d_sync.ds_offset_done = md->md_sync_offset; 528 disk->d_genid = md->md_genid; 529 disk->d_sync.ds_syncid = md->md_syncid; 530 if (errorp != NULL) 531 *errorp = 0; 532 return (disk); 533} 534 535static void 536g_raid3_destroy_disk(struct g_raid3_disk *disk) 537{ 538 struct g_raid3_softc *sc; 539 540 g_topology_assert_not(); 541 sc = disk->d_softc; 542 sx_assert(&sc->sc_lock, SX_XLOCKED); 543 544 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 545 return; 546 g_raid3_event_cancel(disk); 547 switch (disk->d_state) { 548 case G_RAID3_DISK_STATE_SYNCHRONIZING: 549 if (sc->sc_syncdisk != NULL) 550 g_raid3_sync_stop(sc, 1); 551 /* FALLTHROUGH */ 552 case G_RAID3_DISK_STATE_NEW: 553 case G_RAID3_DISK_STATE_STALE: 554 case G_RAID3_DISK_STATE_ACTIVE: 555 g_topology_lock(); 556 g_raid3_disconnect_consumer(sc, disk->d_consumer); 557 g_topology_unlock(); 558 disk->d_consumer = NULL; 559 break; 560 default: 561 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 562 g_raid3_get_diskname(disk), 563 g_raid3_disk_state2str(disk->d_state))); 564 } 565 disk->d_state = G_RAID3_DISK_STATE_NODISK; 566} 567 568static void 569g_raid3_destroy_device(struct g_raid3_softc *sc) 570{ 571 struct g_raid3_event *ep; 572 struct g_raid3_disk *disk; 573 struct g_geom *gp; 574 struct g_consumer *cp; 575 u_int n; 576 577 g_topology_assert_not(); 578 sx_assert(&sc->sc_lock, SX_XLOCKED); 579 580 gp = sc->sc_geom; 581 if (sc->sc_provider != NULL) 582 g_raid3_destroy_provider(sc); 583 for (n = 0; n < sc->sc_ndisks; n++) { 584 disk = &sc->sc_disks[n]; 585 if (disk->d_state != G_RAID3_DISK_STATE_NODISK) { 586 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 587 g_raid3_update_metadata(disk); 588 g_raid3_destroy_disk(disk); 589 } 590 } 591 while ((ep = g_raid3_event_get(sc)) != NULL) { 592 g_raid3_event_remove(sc, ep); 593 if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0) 594 g_raid3_event_free(ep); 595 else { 596 ep->e_error = ECANCELED; 597 ep->e_flags |= G_RAID3_EVENT_DONE; 598 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, ep); 599 mtx_lock(&sc->sc_events_mtx); 600 wakeup(ep); 601 mtx_unlock(&sc->sc_events_mtx); 602 } 603 } 604 callout_drain(&sc->sc_callout); 605 cp = LIST_FIRST(&sc->sc_sync.ds_geom->consumer); 606 g_topology_lock(); 607 if (cp != NULL) 608 g_raid3_disconnect_consumer(sc, cp); 609 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 610 G_RAID3_DEBUG(0, "Device %s destroyed.", gp->name); 611 g_wither_geom(gp, ENXIO); 612 g_topology_unlock(); 613 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_64K].sz_zone); 614 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_16K].sz_zone); 615 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_4K].sz_zone); 616 mtx_destroy(&sc->sc_queue_mtx); 617 mtx_destroy(&sc->sc_events_mtx); 618 sx_xunlock(&sc->sc_lock); 619 sx_destroy(&sc->sc_lock); 620} 621 622static void 623g_raid3_orphan(struct g_consumer *cp) 624{ 625 struct g_raid3_disk *disk; 626 627 g_topology_assert(); 628 629 disk = cp->private; 630 if (disk == NULL) 631 return; 632 disk->d_softc->sc_bump_id = G_RAID3_BUMP_SYNCID; 633 g_raid3_event_send(disk, G_RAID3_DISK_STATE_DISCONNECTED, 634 G_RAID3_EVENT_DONTWAIT); 635} 636 637static int 638g_raid3_write_metadata(struct g_raid3_disk *disk, struct g_raid3_metadata *md) 639{ 640 struct g_raid3_softc *sc; 641 struct g_consumer *cp; 642 off_t offset, length; 643 u_char *sector; 644 int error = 0; 645 646 g_topology_assert_not(); 647 sc = disk->d_softc; 648 sx_assert(&sc->sc_lock, SX_LOCKED); 649 650 cp = disk->d_consumer; 651 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 652 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 653 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 654 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 655 cp->acw, cp->ace)); 656 length = cp->provider->sectorsize; 657 offset = cp->provider->mediasize - length; 658 sector = malloc((size_t)length, M_RAID3, M_WAITOK | M_ZERO); 659 if (md != NULL) 660 raid3_metadata_encode(md, sector); 661 error = g_write_data(cp, offset, sector, length); 662 free(sector, M_RAID3); 663 if (error != 0) { 664 if ((disk->d_flags & G_RAID3_DISK_FLAG_BROKEN) == 0) { 665 G_RAID3_DEBUG(0, "Cannot write metadata on %s " 666 "(device=%s, error=%d).", 667 g_raid3_get_diskname(disk), sc->sc_name, error); 668 disk->d_flags |= G_RAID3_DISK_FLAG_BROKEN; 669 } else { 670 G_RAID3_DEBUG(1, "Cannot write metadata on %s " 671 "(device=%s, error=%d).", 672 g_raid3_get_diskname(disk), sc->sc_name, error); 673 } 674 if (g_raid3_disconnect_on_failure && 675 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 676 sc->sc_bump_id |= G_RAID3_BUMP_GENID; 677 g_raid3_event_send(disk, 678 G_RAID3_DISK_STATE_DISCONNECTED, 679 G_RAID3_EVENT_DONTWAIT); 680 } 681 } 682 return (error); 683} 684 685int 686g_raid3_clear_metadata(struct g_raid3_disk *disk) 687{ 688 int error; 689 690 g_topology_assert_not(); 691 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 692 693 error = g_raid3_write_metadata(disk, NULL); 694 if (error == 0) { 695 G_RAID3_DEBUG(2, "Metadata on %s cleared.", 696 g_raid3_get_diskname(disk)); 697 } else { 698 G_RAID3_DEBUG(0, 699 "Cannot clear metadata on disk %s (error=%d).", 700 g_raid3_get_diskname(disk), error); 701 } 702 return (error); 703} 704 705void 706g_raid3_fill_metadata(struct g_raid3_disk *disk, struct g_raid3_metadata *md) 707{ 708 struct g_raid3_softc *sc; 709 struct g_provider *pp; 710 711 sc = disk->d_softc; 712 strlcpy(md->md_magic, G_RAID3_MAGIC, sizeof(md->md_magic)); 713 md->md_version = G_RAID3_VERSION; 714 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 715 md->md_id = sc->sc_id; 716 md->md_all = sc->sc_ndisks; 717 md->md_genid = sc->sc_genid; 718 md->md_mediasize = sc->sc_mediasize; 719 md->md_sectorsize = sc->sc_sectorsize; 720 md->md_mflags = (sc->sc_flags & G_RAID3_DEVICE_FLAG_MASK); 721 md->md_no = disk->d_no; 722 md->md_syncid = disk->d_sync.ds_syncid; 723 md->md_dflags = (disk->d_flags & G_RAID3_DISK_FLAG_MASK); 724 if (disk->d_state != G_RAID3_DISK_STATE_SYNCHRONIZING) 725 md->md_sync_offset = 0; 726 else { 727 md->md_sync_offset = 728 disk->d_sync.ds_offset_done / (sc->sc_ndisks - 1); 729 } 730 if (disk->d_consumer != NULL && disk->d_consumer->provider != NULL) 731 pp = disk->d_consumer->provider; 732 else 733 pp = NULL; 734 if ((disk->d_flags & G_RAID3_DISK_FLAG_HARDCODED) != 0 && pp != NULL) 735 strlcpy(md->md_provider, pp->name, sizeof(md->md_provider)); 736 else 737 bzero(md->md_provider, sizeof(md->md_provider)); 738 if (pp != NULL) 739 md->md_provsize = pp->mediasize; 740 else 741 md->md_provsize = 0; 742} 743 744void 745g_raid3_update_metadata(struct g_raid3_disk *disk) 746{ 747 struct g_raid3_softc *sc; 748 struct g_raid3_metadata md; 749 int error; 750 751 g_topology_assert_not(); 752 sc = disk->d_softc; 753 sx_assert(&sc->sc_lock, SX_LOCKED); 754 755 g_raid3_fill_metadata(disk, &md); 756 error = g_raid3_write_metadata(disk, &md); 757 if (error == 0) { 758 G_RAID3_DEBUG(2, "Metadata on %s updated.", 759 g_raid3_get_diskname(disk)); 760 } else { 761 G_RAID3_DEBUG(0, 762 "Cannot update metadata on disk %s (error=%d).", 763 g_raid3_get_diskname(disk), error); 764 } 765} 766 767static void 768g_raid3_bump_syncid(struct g_raid3_softc *sc) 769{ 770 struct g_raid3_disk *disk; 771 u_int n; 772 773 g_topology_assert_not(); 774 sx_assert(&sc->sc_lock, SX_XLOCKED); 775 KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) > 0, 776 ("%s called with no active disks (device=%s).", __func__, 777 sc->sc_name)); 778 779 sc->sc_syncid++; 780 G_RAID3_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 781 sc->sc_syncid); 782 for (n = 0; n < sc->sc_ndisks; n++) { 783 disk = &sc->sc_disks[n]; 784 if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 785 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 786 disk->d_sync.ds_syncid = sc->sc_syncid; 787 g_raid3_update_metadata(disk); 788 } 789 } 790} 791 792static void 793g_raid3_bump_genid(struct g_raid3_softc *sc) 794{ 795 struct g_raid3_disk *disk; 796 u_int n; 797 798 g_topology_assert_not(); 799 sx_assert(&sc->sc_lock, SX_XLOCKED); 800 KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) > 0, 801 ("%s called with no active disks (device=%s).", __func__, 802 sc->sc_name)); 803 804 sc->sc_genid++; 805 G_RAID3_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 806 sc->sc_genid); 807 for (n = 0; n < sc->sc_ndisks; n++) { 808 disk = &sc->sc_disks[n]; 809 if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 810 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 811 disk->d_genid = sc->sc_genid; 812 g_raid3_update_metadata(disk); 813 } 814 } 815} 816 817static int 818g_raid3_idle(struct g_raid3_softc *sc, int acw) 819{ 820 struct g_raid3_disk *disk; 821 u_int i; 822 int timeout; 823 824 g_topology_assert_not(); 825 sx_assert(&sc->sc_lock, SX_XLOCKED); 826 827 if (sc->sc_provider == NULL) 828 return (0); 829 if (sc->sc_idle) 830 return (0); 831 if (sc->sc_writes > 0) 832 return (0); 833 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 834 timeout = g_raid3_idletime - (time_uptime - sc->sc_last_write); 835 if (timeout > 0) 836 return (timeout); 837 } 838 sc->sc_idle = 1; 839 for (i = 0; i < sc->sc_ndisks; i++) { 840 disk = &sc->sc_disks[i]; 841 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) 842 continue; 843 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as clean.", 844 g_raid3_get_diskname(disk), sc->sc_name); 845 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 846 g_raid3_update_metadata(disk); 847 } 848 return (0); 849} 850 851static void 852g_raid3_unidle(struct g_raid3_softc *sc) 853{ 854 struct g_raid3_disk *disk; 855 u_int i; 856 857 g_topology_assert_not(); 858 sx_assert(&sc->sc_lock, SX_XLOCKED); 859 860 sc->sc_idle = 0; 861 sc->sc_last_write = time_uptime; 862 for (i = 0; i < sc->sc_ndisks; i++) { 863 disk = &sc->sc_disks[i]; 864 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) 865 continue; 866 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as dirty.", 867 g_raid3_get_diskname(disk), sc->sc_name); 868 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY; 869 g_raid3_update_metadata(disk); 870 } 871} 872 873/* 874 * Treat bio_driver1 field in parent bio as list head and field bio_caller1 875 * in child bio as pointer to the next element on the list. 876 */ 877#define G_RAID3_HEAD_BIO(pbp) (pbp)->bio_driver1 878 879#define G_RAID3_NEXT_BIO(cbp) (cbp)->bio_caller1 880 881#define G_RAID3_FOREACH_BIO(pbp, bp) \ 882 for ((bp) = G_RAID3_HEAD_BIO(pbp); (bp) != NULL; \ 883 (bp) = G_RAID3_NEXT_BIO(bp)) 884 885#define G_RAID3_FOREACH_SAFE_BIO(pbp, bp, tmpbp) \ 886 for ((bp) = G_RAID3_HEAD_BIO(pbp); \ 887 (bp) != NULL && ((tmpbp) = G_RAID3_NEXT_BIO(bp), 1); \ 888 (bp) = (tmpbp)) 889 890static void 891g_raid3_init_bio(struct bio *pbp) 892{ 893 894 G_RAID3_HEAD_BIO(pbp) = NULL; 895} 896 897static void 898g_raid3_remove_bio(struct bio *cbp) 899{ 900 struct bio *pbp, *bp; 901 902 pbp = cbp->bio_parent; 903 if (G_RAID3_HEAD_BIO(pbp) == cbp) 904 G_RAID3_HEAD_BIO(pbp) = G_RAID3_NEXT_BIO(cbp); 905 else { 906 G_RAID3_FOREACH_BIO(pbp, bp) { 907 if (G_RAID3_NEXT_BIO(bp) == cbp) { 908 G_RAID3_NEXT_BIO(bp) = G_RAID3_NEXT_BIO(cbp); 909 break; 910 } 911 } 912 } 913 G_RAID3_NEXT_BIO(cbp) = NULL; 914} 915 916static void 917g_raid3_replace_bio(struct bio *sbp, struct bio *dbp) 918{ 919 struct bio *pbp, *bp; 920 921 g_raid3_remove_bio(sbp); 922 pbp = dbp->bio_parent; 923 G_RAID3_NEXT_BIO(sbp) = G_RAID3_NEXT_BIO(dbp); 924 if (G_RAID3_HEAD_BIO(pbp) == dbp) 925 G_RAID3_HEAD_BIO(pbp) = sbp; 926 else { 927 G_RAID3_FOREACH_BIO(pbp, bp) { 928 if (G_RAID3_NEXT_BIO(bp) == dbp) { 929 G_RAID3_NEXT_BIO(bp) = sbp; 930 break; 931 } 932 } 933 } 934 G_RAID3_NEXT_BIO(dbp) = NULL; 935} 936 937static void 938g_raid3_destroy_bio(struct g_raid3_softc *sc, struct bio *cbp) 939{ 940 struct bio *bp, *pbp; 941 size_t size; 942 943 pbp = cbp->bio_parent; 944 pbp->bio_children--; 945 KASSERT(cbp->bio_data != NULL, ("NULL bio_data")); 946 size = pbp->bio_length / (sc->sc_ndisks - 1); 947 uma_zfree_arg(sc->sc_zones[g_raid3_zone(size)].sz_zone, 948 cbp->bio_data, 949 &sc->sc_zones[g_raid3_zone(size)]); 950 if (G_RAID3_HEAD_BIO(pbp) == cbp) { 951 G_RAID3_HEAD_BIO(pbp) = G_RAID3_NEXT_BIO(cbp); 952 G_RAID3_NEXT_BIO(cbp) = NULL; 953 g_destroy_bio(cbp); 954 } else { 955 G_RAID3_FOREACH_BIO(pbp, bp) { 956 if (G_RAID3_NEXT_BIO(bp) == cbp) 957 break; 958 } 959 if (bp != NULL) { 960 KASSERT(G_RAID3_NEXT_BIO(bp) != NULL, 961 ("NULL bp->bio_driver1")); 962 G_RAID3_NEXT_BIO(bp) = G_RAID3_NEXT_BIO(cbp); 963 G_RAID3_NEXT_BIO(cbp) = NULL; 964 } 965 g_destroy_bio(cbp); 966 } 967} 968 969static struct bio * 970g_raid3_clone_bio(struct g_raid3_softc *sc, struct bio *pbp) 971{ 972 struct bio *bp, *cbp; 973 size_t size; 974 int memflag; 975 976 cbp = g_clone_bio(pbp); 977 if (cbp == NULL) 978 return (NULL); 979 size = pbp->bio_length / (sc->sc_ndisks - 1); 980 if ((pbp->bio_cflags & G_RAID3_BIO_CFLAG_REGULAR) != 0) 981 memflag = M_WAITOK; 982 else 983 memflag = M_NOWAIT; 984 cbp->bio_data = uma_zalloc_arg(sc->sc_zones[g_raid3_zone(size)].sz_zone, 985 &sc->sc_zones[g_raid3_zone(size)], memflag); 986 sc->sc_zones[g_raid3_zone(size)].sz_requested++; 987 if (cbp->bio_data == NULL) { 988 sc->sc_zones[g_raid3_zone(size)].sz_failed++; 989 pbp->bio_children--; 990 g_destroy_bio(cbp); 991 return (NULL); 992 } 993 G_RAID3_NEXT_BIO(cbp) = NULL; 994 if (G_RAID3_HEAD_BIO(pbp) == NULL) 995 G_RAID3_HEAD_BIO(pbp) = cbp; 996 else { 997 G_RAID3_FOREACH_BIO(pbp, bp) { 998 if (G_RAID3_NEXT_BIO(bp) == NULL) { 999 G_RAID3_NEXT_BIO(bp) = cbp; 1000 break; 1001 } 1002 } 1003 } 1004 return (cbp); 1005} 1006 1007static void 1008g_raid3_scatter(struct bio *pbp) 1009{ 1010 struct g_raid3_softc *sc; 1011 struct g_raid3_disk *disk; 1012 struct bio *bp, *cbp, *tmpbp; 1013 off_t atom, cadd, padd, left; 1014 1015 sc = pbp->bio_to->geom->softc; 1016 bp = NULL; 1017 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_NOPARITY) == 0) { 1018 /* 1019 * Find bio for which we should calculate data. 1020 */ 1021 G_RAID3_FOREACH_BIO(pbp, cbp) { 1022 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) { 1023 bp = cbp; 1024 break; 1025 } 1026 } 1027 KASSERT(bp != NULL, ("NULL parity bio.")); 1028 } 1029 atom = sc->sc_sectorsize / (sc->sc_ndisks - 1); 1030 cadd = padd = 0; 1031 for (left = pbp->bio_length; left > 0; left -= sc->sc_sectorsize) { 1032 G_RAID3_FOREACH_BIO(pbp, cbp) { 1033 if (cbp == bp) 1034 continue; 1035 bcopy(pbp->bio_data + padd, cbp->bio_data + cadd, atom); 1036 padd += atom; 1037 } 1038 cadd += atom; 1039 } 1040 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_NOPARITY) == 0) { 1041 /* 1042 * Calculate parity. 1043 */ 1044 bzero(bp->bio_data, bp->bio_length); 1045 G_RAID3_FOREACH_SAFE_BIO(pbp, cbp, tmpbp) { 1046 if (cbp == bp) 1047 continue; 1048 g_raid3_xor(cbp->bio_data, bp->bio_data, bp->bio_data, 1049 bp->bio_length); 1050 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_NODISK) != 0) 1051 g_raid3_destroy_bio(sc, cbp); 1052 } 1053 } 1054 G_RAID3_FOREACH_SAFE_BIO(pbp, cbp, tmpbp) { 1055 struct g_consumer *cp; 1056 1057 disk = cbp->bio_caller2; 1058 cp = disk->d_consumer; 1059 cbp->bio_to = cp->provider; 1060 G_RAID3_LOGREQ(3, cbp, "Sending request."); 1061 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1062 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1063 cp->acr, cp->acw, cp->ace)); 1064 cp->index++; 1065 sc->sc_writes++; 1066 g_io_request(cbp, cp); 1067 } 1068} 1069 1070static void 1071g_raid3_gather(struct bio *pbp) 1072{ 1073 struct g_raid3_softc *sc; 1074 struct g_raid3_disk *disk; 1075 struct bio *xbp, *fbp, *cbp; 1076 off_t atom, cadd, padd, left; 1077 1078 sc = pbp->bio_to->geom->softc; 1079 /* 1080 * Find bio for which we have to calculate data. 1081 * While going through this path, check if all requests 1082 * succeeded, if not, deny whole request. 1083 * If we're in COMPLETE mode, we allow one request to fail, 1084 * so if we find one, we're sending it to the parity consumer. 1085 * If there are more failed requests, we deny whole request. 1086 */ 1087 xbp = fbp = NULL; 1088 G_RAID3_FOREACH_BIO(pbp, cbp) { 1089 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) { 1090 KASSERT(xbp == NULL, ("More than one parity bio.")); 1091 xbp = cbp; 1092 } 1093 if (cbp->bio_error == 0) 1094 continue; 1095 /* 1096 * Found failed request. 1097 */ 1098 if (fbp == NULL) { 1099 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_DEGRADED) != 0) { 1100 /* 1101 * We are already in degraded mode, so we can't 1102 * accept any failures. 1103 */ 1104 if (pbp->bio_error == 0) 1105 pbp->bio_error = cbp->bio_error; 1106 } else { 1107 fbp = cbp; 1108 } 1109 } else { 1110 /* 1111 * Next failed request, that's too many. 1112 */ 1113 if (pbp->bio_error == 0) 1114 pbp->bio_error = fbp->bio_error; 1115 } 1116 disk = cbp->bio_caller2; 1117 if (disk == NULL) 1118 continue; 1119 if ((disk->d_flags & G_RAID3_DISK_FLAG_BROKEN) == 0) { 1120 disk->d_flags |= G_RAID3_DISK_FLAG_BROKEN; 1121 G_RAID3_LOGREQ(0, cbp, "Request failed (error=%d).", 1122 cbp->bio_error); 1123 } else { 1124 G_RAID3_LOGREQ(1, cbp, "Request failed (error=%d).", 1125 cbp->bio_error); 1126 } 1127 if (g_raid3_disconnect_on_failure && 1128 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 1129 sc->sc_bump_id |= G_RAID3_BUMP_GENID; 1130 g_raid3_event_send(disk, 1131 G_RAID3_DISK_STATE_DISCONNECTED, 1132 G_RAID3_EVENT_DONTWAIT); 1133 } 1134 } 1135 if (pbp->bio_error != 0) 1136 goto finish; 1137 if (fbp != NULL && (pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) { 1138 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_VERIFY; 1139 if (xbp != fbp) 1140 g_raid3_replace_bio(xbp, fbp); 1141 g_raid3_destroy_bio(sc, fbp); 1142 } else if (fbp != NULL) { 1143 struct g_consumer *cp; 1144 1145 /* 1146 * One request failed, so send the same request to 1147 * the parity consumer. 1148 */ 1149 disk = pbp->bio_driver2; 1150 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) { 1151 pbp->bio_error = fbp->bio_error; 1152 goto finish; 1153 } 1154 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED; 1155 pbp->bio_inbed--; 1156 fbp->bio_flags &= ~(BIO_DONE | BIO_ERROR); 1157 if (disk->d_no == sc->sc_ndisks - 1) 1158 fbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1159 fbp->bio_error = 0; 1160 fbp->bio_completed = 0; 1161 fbp->bio_children = 0; 1162 fbp->bio_inbed = 0; 1163 cp = disk->d_consumer; 1164 fbp->bio_caller2 = disk; 1165 fbp->bio_to = cp->provider; 1166 G_RAID3_LOGREQ(3, fbp, "Sending request (recover)."); 1167 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1168 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1169 cp->acr, cp->acw, cp->ace)); 1170 cp->index++; 1171 g_io_request(fbp, cp); 1172 return; 1173 } 1174 if (xbp != NULL) { 1175 /* 1176 * Calculate parity. 1177 */ 1178 G_RAID3_FOREACH_BIO(pbp, cbp) { 1179 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) 1180 continue; 1181 g_raid3_xor(cbp->bio_data, xbp->bio_data, xbp->bio_data, 1182 xbp->bio_length); 1183 } 1184 xbp->bio_cflags &= ~G_RAID3_BIO_CFLAG_PARITY; 1185 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) { 1186 if (!g_raid3_is_zero(xbp)) { 1187 g_raid3_parity_mismatch++; 1188 pbp->bio_error = EIO; 1189 goto finish; 1190 } 1191 g_raid3_destroy_bio(sc, xbp); 1192 } 1193 } 1194 atom = sc->sc_sectorsize / (sc->sc_ndisks - 1); 1195 cadd = padd = 0; 1196 for (left = pbp->bio_length; left > 0; left -= sc->sc_sectorsize) { 1197 G_RAID3_FOREACH_BIO(pbp, cbp) { 1198 bcopy(cbp->bio_data + cadd, pbp->bio_data + padd, atom); 1199 pbp->bio_completed += atom; 1200 padd += atom; 1201 } 1202 cadd += atom; 1203 } 1204finish: 1205 if (pbp->bio_error == 0) 1206 G_RAID3_LOGREQ(3, pbp, "Request finished."); 1207 else { 1208 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) 1209 G_RAID3_LOGREQ(1, pbp, "Verification error."); 1210 else 1211 G_RAID3_LOGREQ(0, pbp, "Request failed."); 1212 } 1213 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_MASK; 1214 while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL) 1215 g_raid3_destroy_bio(sc, cbp); 1216 g_io_deliver(pbp, pbp->bio_error); 1217} 1218 1219static void 1220g_raid3_done(struct bio *bp) 1221{ 1222 struct g_raid3_softc *sc; 1223 1224 sc = bp->bio_from->geom->softc; 1225 bp->bio_cflags |= G_RAID3_BIO_CFLAG_REGULAR; 1226 G_RAID3_LOGREQ(3, bp, "Regular request done (error=%d).", bp->bio_error); 1227 mtx_lock(&sc->sc_queue_mtx); 1228 bioq_insert_head(&sc->sc_queue, bp); 1229 wakeup(sc); 1230 wakeup(&sc->sc_queue); 1231 mtx_unlock(&sc->sc_queue_mtx); 1232} 1233 1234static void 1235g_raid3_regular_request(struct bio *cbp) 1236{ 1237 struct g_raid3_softc *sc; 1238 struct g_raid3_disk *disk; 1239 struct bio *pbp; 1240 1241 g_topology_assert_not(); 1242 1243 pbp = cbp->bio_parent; 1244 sc = pbp->bio_to->geom->softc; 1245 cbp->bio_from->index--; 1246 if (cbp->bio_cmd == BIO_WRITE) 1247 sc->sc_writes--; 1248 disk = cbp->bio_from->private; 1249 if (disk == NULL) { 1250 g_topology_lock(); 1251 g_raid3_kill_consumer(sc, cbp->bio_from); 1252 g_topology_unlock(); 1253 } 1254 1255 G_RAID3_LOGREQ(3, cbp, "Request finished."); 1256 pbp->bio_inbed++; 1257 KASSERT(pbp->bio_inbed <= pbp->bio_children, 1258 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 1259 pbp->bio_children)); 1260 if (pbp->bio_inbed != pbp->bio_children) 1261 return; 1262 switch (pbp->bio_cmd) { 1263 case BIO_READ: 1264 g_raid3_gather(pbp); 1265 break; 1266 case BIO_WRITE: 1267 case BIO_DELETE: 1268 { 1269 int error = 0; 1270 1271 pbp->bio_completed = pbp->bio_length; 1272 while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL) { 1273 if (cbp->bio_error == 0) { 1274 g_raid3_destroy_bio(sc, cbp); 1275 continue; 1276 } 1277 1278 if (error == 0) 1279 error = cbp->bio_error; 1280 else if (pbp->bio_error == 0) { 1281 /* 1282 * Next failed request, that's too many. 1283 */ 1284 pbp->bio_error = error; 1285 } 1286 1287 disk = cbp->bio_caller2; 1288 if (disk == NULL) { 1289 g_raid3_destroy_bio(sc, cbp); 1290 continue; 1291 } 1292 1293 if ((disk->d_flags & G_RAID3_DISK_FLAG_BROKEN) == 0) { 1294 disk->d_flags |= G_RAID3_DISK_FLAG_BROKEN; 1295 G_RAID3_LOGREQ(0, cbp, 1296 "Request failed (error=%d).", 1297 cbp->bio_error); 1298 } else { 1299 G_RAID3_LOGREQ(1, cbp, 1300 "Request failed (error=%d).", 1301 cbp->bio_error); 1302 } 1303 if (g_raid3_disconnect_on_failure && 1304 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 1305 sc->sc_bump_id |= G_RAID3_BUMP_GENID; 1306 g_raid3_event_send(disk, 1307 G_RAID3_DISK_STATE_DISCONNECTED, 1308 G_RAID3_EVENT_DONTWAIT); 1309 } 1310 g_raid3_destroy_bio(sc, cbp); 1311 } 1312 if (pbp->bio_error == 0) 1313 G_RAID3_LOGREQ(3, pbp, "Request finished."); 1314 else 1315 G_RAID3_LOGREQ(0, pbp, "Request failed."); 1316 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_DEGRADED; 1317 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_NOPARITY; 1318 bioq_remove(&sc->sc_inflight, pbp); 1319 /* Release delayed sync requests if possible. */ 1320 g_raid3_sync_release(sc); 1321 g_io_deliver(pbp, pbp->bio_error); 1322 break; 1323 } 1324 } 1325} 1326 1327static void 1328g_raid3_sync_done(struct bio *bp) 1329{ 1330 struct g_raid3_softc *sc; 1331 1332 G_RAID3_LOGREQ(3, bp, "Synchronization request delivered."); 1333 sc = bp->bio_from->geom->softc; 1334 bp->bio_cflags |= G_RAID3_BIO_CFLAG_SYNC; 1335 mtx_lock(&sc->sc_queue_mtx); 1336 bioq_insert_head(&sc->sc_queue, bp); 1337 wakeup(sc); 1338 wakeup(&sc->sc_queue); 1339 mtx_unlock(&sc->sc_queue_mtx); 1340} 1341 1342static void 1343g_raid3_start(struct bio *bp) 1344{ 1345 struct g_raid3_softc *sc; 1346 1347 sc = bp->bio_to->geom->softc; 1348 /* 1349 * If sc == NULL or there are no valid disks, provider's error 1350 * should be set and g_raid3_start() should not be called at all. 1351 */ 1352 KASSERT(sc != NULL && (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 1353 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE), 1354 ("Provider's error should be set (error=%d)(device=%s).", 1355 bp->bio_to->error, bp->bio_to->name)); 1356 G_RAID3_LOGREQ(3, bp, "Request received."); 1357 1358 switch (bp->bio_cmd) { 1359 case BIO_READ: 1360 case BIO_WRITE: 1361 case BIO_DELETE: 1362 break; 1363 case BIO_GETATTR: 1364 default: 1365 g_io_deliver(bp, EOPNOTSUPP); 1366 return; 1367 } 1368 mtx_lock(&sc->sc_queue_mtx); 1369 bioq_insert_tail(&sc->sc_queue, bp); 1370 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1371 wakeup(sc); 1372 mtx_unlock(&sc->sc_queue_mtx); 1373} 1374 1375/* 1376 * Return TRUE if the given request is colliding with a in-progress 1377 * synchronization request. 1378 */ 1379static int 1380g_raid3_sync_collision(struct g_raid3_softc *sc, struct bio *bp) 1381{ 1382 struct g_raid3_disk *disk; 1383 struct bio *sbp; 1384 off_t rstart, rend, sstart, send; 1385 int i; 1386 1387 disk = sc->sc_syncdisk; 1388 if (disk == NULL) 1389 return (0); 1390 rstart = bp->bio_offset; 1391 rend = bp->bio_offset + bp->bio_length; 1392 for (i = 0; i < g_raid3_syncreqs; i++) { 1393 sbp = disk->d_sync.ds_bios[i]; 1394 if (sbp == NULL) 1395 continue; 1396 sstart = sbp->bio_offset; 1397 send = sbp->bio_length; 1398 if (sbp->bio_cmd == BIO_WRITE) { 1399 sstart *= sc->sc_ndisks - 1; 1400 send *= sc->sc_ndisks - 1; 1401 } 1402 send += sstart; 1403 if (rend > sstart && rstart < send) 1404 return (1); 1405 } 1406 return (0); 1407} 1408 1409/* 1410 * Return TRUE if the given sync request is colliding with a in-progress regular 1411 * request. 1412 */ 1413static int 1414g_raid3_regular_collision(struct g_raid3_softc *sc, struct bio *sbp) 1415{ 1416 off_t rstart, rend, sstart, send; 1417 struct bio *bp; 1418 1419 if (sc->sc_syncdisk == NULL) 1420 return (0); 1421 sstart = sbp->bio_offset; 1422 send = sstart + sbp->bio_length; 1423 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { 1424 rstart = bp->bio_offset; 1425 rend = bp->bio_offset + bp->bio_length; 1426 if (rend > sstart && rstart < send) 1427 return (1); 1428 } 1429 return (0); 1430} 1431 1432/* 1433 * Puts request onto delayed queue. 1434 */ 1435static void 1436g_raid3_regular_delay(struct g_raid3_softc *sc, struct bio *bp) 1437{ 1438 1439 G_RAID3_LOGREQ(2, bp, "Delaying request."); 1440 bioq_insert_head(&sc->sc_regular_delayed, bp); 1441} 1442 1443/* 1444 * Puts synchronization request onto delayed queue. 1445 */ 1446static void 1447g_raid3_sync_delay(struct g_raid3_softc *sc, struct bio *bp) 1448{ 1449 1450 G_RAID3_LOGREQ(2, bp, "Delaying synchronization request."); 1451 bioq_insert_tail(&sc->sc_sync_delayed, bp); 1452} 1453 1454/* 1455 * Releases delayed regular requests which don't collide anymore with sync 1456 * requests. 1457 */ 1458static void 1459g_raid3_regular_release(struct g_raid3_softc *sc) 1460{ 1461 struct bio *bp, *bp2; 1462 1463 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { 1464 if (g_raid3_sync_collision(sc, bp)) 1465 continue; 1466 bioq_remove(&sc->sc_regular_delayed, bp); 1467 G_RAID3_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1468 mtx_lock(&sc->sc_queue_mtx); 1469 bioq_insert_head(&sc->sc_queue, bp); 1470#if 0 1471 /* 1472 * wakeup() is not needed, because this function is called from 1473 * the worker thread. 1474 */ 1475 wakeup(&sc->sc_queue); 1476#endif 1477 mtx_unlock(&sc->sc_queue_mtx); 1478 } 1479} 1480 1481/* 1482 * Releases delayed sync requests which don't collide anymore with regular 1483 * requests. 1484 */ 1485static void 1486g_raid3_sync_release(struct g_raid3_softc *sc) 1487{ 1488 struct bio *bp, *bp2; 1489 1490 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { 1491 if (g_raid3_regular_collision(sc, bp)) 1492 continue; 1493 bioq_remove(&sc->sc_sync_delayed, bp); 1494 G_RAID3_LOGREQ(2, bp, 1495 "Releasing delayed synchronization request."); 1496 g_io_request(bp, bp->bio_from); 1497 } 1498} 1499 1500/* 1501 * Handle synchronization requests. 1502 * Every synchronization request is two-steps process: first, READ request is 1503 * send to active provider and then WRITE request (with read data) to the provider 1504 * beeing synchronized. When WRITE is finished, new synchronization request is 1505 * send. 1506 */ 1507static void 1508g_raid3_sync_request(struct bio *bp) 1509{ 1510 struct g_raid3_softc *sc; 1511 struct g_raid3_disk *disk; 1512 1513 bp->bio_from->index--; 1514 sc = bp->bio_from->geom->softc; 1515 disk = bp->bio_from->private; 1516 if (disk == NULL) { 1517 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1518 g_topology_lock(); 1519 g_raid3_kill_consumer(sc, bp->bio_from); 1520 g_topology_unlock(); 1521 free(bp->bio_data, M_RAID3); 1522 g_destroy_bio(bp); 1523 sx_xlock(&sc->sc_lock); 1524 return; 1525 } 1526 1527 /* 1528 * Synchronization request. 1529 */ 1530 switch (bp->bio_cmd) { 1531 case BIO_READ: 1532 { 1533 struct g_consumer *cp; 1534 u_char *dst, *src; 1535 off_t left; 1536 u_int atom; 1537 1538 if (bp->bio_error != 0) { 1539 G_RAID3_LOGREQ(0, bp, 1540 "Synchronization request failed (error=%d).", 1541 bp->bio_error); 1542 g_destroy_bio(bp); 1543 return; 1544 } 1545 G_RAID3_LOGREQ(3, bp, "Synchronization request finished."); 1546 atom = sc->sc_sectorsize / (sc->sc_ndisks - 1); 1547 dst = src = bp->bio_data; 1548 if (disk->d_no == sc->sc_ndisks - 1) { 1549 u_int n; 1550 1551 /* Parity component. */ 1552 for (left = bp->bio_length; left > 0; 1553 left -= sc->sc_sectorsize) { 1554 bcopy(src, dst, atom); 1555 src += atom; 1556 for (n = 1; n < sc->sc_ndisks - 1; n++) { 1557 g_raid3_xor(src, dst, dst, atom); 1558 src += atom; 1559 } 1560 dst += atom; 1561 } 1562 } else { 1563 /* Regular component. */ 1564 src += atom * disk->d_no; 1565 for (left = bp->bio_length; left > 0; 1566 left -= sc->sc_sectorsize) { 1567 bcopy(src, dst, atom); 1568 src += sc->sc_sectorsize; 1569 dst += atom; 1570 } 1571 } 1572 bp->bio_driver1 = bp->bio_driver2 = NULL; 1573 bp->bio_pflags = 0; 1574 bp->bio_offset /= sc->sc_ndisks - 1; 1575 bp->bio_length /= sc->sc_ndisks - 1; 1576 bp->bio_cmd = BIO_WRITE; 1577 bp->bio_cflags = 0; 1578 bp->bio_children = bp->bio_inbed = 0; 1579 cp = disk->d_consumer; 1580 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1581 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1582 cp->acr, cp->acw, cp->ace)); 1583 cp->index++; 1584 g_io_request(bp, cp); 1585 return; 1586 } 1587 case BIO_WRITE: 1588 { 1589 struct g_raid3_disk_sync *sync; 1590 off_t boffset, moffset; 1591 void *data; 1592 int i; 1593 1594 if (bp->bio_error != 0) { 1595 G_RAID3_LOGREQ(0, bp, 1596 "Synchronization request failed (error=%d).", 1597 bp->bio_error); 1598 g_destroy_bio(bp); 1599 sc->sc_bump_id |= G_RAID3_BUMP_GENID; 1600 g_raid3_event_send(disk, 1601 G_RAID3_DISK_STATE_DISCONNECTED, 1602 G_RAID3_EVENT_DONTWAIT); 1603 return; 1604 } 1605 G_RAID3_LOGREQ(3, bp, "Synchronization request finished."); 1606 sync = &disk->d_sync; 1607 if (sync->ds_offset == sc->sc_mediasize / (sc->sc_ndisks - 1) || 1608 sync->ds_consumer == NULL || 1609 (sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 1610 /* Don't send more synchronization requests. */ 1611 sync->ds_inflight--; 1612 if (sync->ds_bios != NULL) { 1613 i = (int)(uintptr_t)bp->bio_caller1; 1614 sync->ds_bios[i] = NULL; 1615 } 1616 free(bp->bio_data, M_RAID3); 1617 g_destroy_bio(bp); 1618 if (sync->ds_inflight > 0) 1619 return; 1620 if (sync->ds_consumer == NULL || 1621 (sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 1622 return; 1623 } 1624 /* 1625 * Disk up-to-date, activate it. 1626 */ 1627 g_raid3_event_send(disk, G_RAID3_DISK_STATE_ACTIVE, 1628 G_RAID3_EVENT_DONTWAIT); 1629 return; 1630 } 1631 1632 /* Send next synchronization request. */ 1633 data = bp->bio_data; 1634 bzero(bp, sizeof(*bp)); 1635 bp->bio_cmd = BIO_READ; 1636 bp->bio_offset = sync->ds_offset * (sc->sc_ndisks - 1); 1637 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1638 sync->ds_offset += bp->bio_length / (sc->sc_ndisks - 1); 1639 bp->bio_done = g_raid3_sync_done; 1640 bp->bio_data = data; 1641 bp->bio_from = sync->ds_consumer; 1642 bp->bio_to = sc->sc_provider; 1643 G_RAID3_LOGREQ(3, bp, "Sending synchronization request."); 1644 sync->ds_consumer->index++; 1645 /* 1646 * Delay the request if it is colliding with a regular request. 1647 */ 1648 if (g_raid3_regular_collision(sc, bp)) 1649 g_raid3_sync_delay(sc, bp); 1650 else 1651 g_io_request(bp, sync->ds_consumer); 1652 1653 /* Release delayed requests if possible. */ 1654 g_raid3_regular_release(sc); 1655 1656 /* Find the smallest offset. */ 1657 moffset = sc->sc_mediasize; 1658 for (i = 0; i < g_raid3_syncreqs; i++) { 1659 bp = sync->ds_bios[i]; 1660 boffset = bp->bio_offset; 1661 if (bp->bio_cmd == BIO_WRITE) 1662 boffset *= sc->sc_ndisks - 1; 1663 if (boffset < moffset) 1664 moffset = boffset; 1665 } 1666 if (sync->ds_offset_done + (MAXPHYS * 100) < moffset) { 1667 /* Update offset_done on every 100 blocks. */ 1668 sync->ds_offset_done = moffset; 1669 g_raid3_update_metadata(disk); 1670 } 1671 return; 1672 } 1673 default: 1674 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1675 bp->bio_cmd, sc->sc_name)); 1676 break; 1677 } 1678} 1679 1680static int 1681g_raid3_register_request(struct bio *pbp) 1682{ 1683 struct g_raid3_softc *sc; 1684 struct g_raid3_disk *disk; 1685 struct g_consumer *cp; 1686 struct bio *cbp, *tmpbp; 1687 off_t offset, length; 1688 u_int n, ndisks; 1689 int round_robin, verify; 1690 1691 ndisks = 0; 1692 sc = pbp->bio_to->geom->softc; 1693 if ((pbp->bio_cflags & G_RAID3_BIO_CFLAG_REGSYNC) != 0 && 1694 sc->sc_syncdisk == NULL) { 1695 g_io_deliver(pbp, EIO); 1696 return (0); 1697 } 1698 g_raid3_init_bio(pbp); 1699 length = pbp->bio_length / (sc->sc_ndisks - 1); 1700 offset = pbp->bio_offset / (sc->sc_ndisks - 1); 1701 round_robin = verify = 0; 1702 switch (pbp->bio_cmd) { 1703 case BIO_READ: 1704 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_VERIFY) != 0 && 1705 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 1706 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_VERIFY; 1707 verify = 1; 1708 ndisks = sc->sc_ndisks; 1709 } else { 1710 verify = 0; 1711 ndisks = sc->sc_ndisks - 1; 1712 } 1713 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_ROUND_ROBIN) != 0 && 1714 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 1715 round_robin = 1; 1716 } else { 1717 round_robin = 0; 1718 } 1719 KASSERT(!round_robin || !verify, 1720 ("ROUND-ROBIN and VERIFY are mutually exclusive.")); 1721 pbp->bio_driver2 = &sc->sc_disks[sc->sc_ndisks - 1]; 1722 break; 1723 case BIO_WRITE: 1724 case BIO_DELETE: 1725 /* 1726 * Delay the request if it is colliding with a synchronization 1727 * request. 1728 */ 1729 if (g_raid3_sync_collision(sc, pbp)) { 1730 g_raid3_regular_delay(sc, pbp); 1731 return (0); 1732 } 1733 1734 if (sc->sc_idle) 1735 g_raid3_unidle(sc); 1736 else 1737 sc->sc_last_write = time_uptime; 1738 1739 ndisks = sc->sc_ndisks; 1740 break; 1741 } 1742 for (n = 0; n < ndisks; n++) { 1743 disk = &sc->sc_disks[n]; 1744 cbp = g_raid3_clone_bio(sc, pbp); 1745 if (cbp == NULL) { 1746 while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL) 1747 g_raid3_destroy_bio(sc, cbp); 1748 /* 1749 * To prevent deadlock, we must run back up 1750 * with the ENOMEM for failed requests of any 1751 * of our consumers. Our own sync requests 1752 * can stick around, as they are finite. 1753 */ 1754 if ((pbp->bio_cflags & 1755 G_RAID3_BIO_CFLAG_REGULAR) != 0) { 1756 g_io_deliver(pbp, ENOMEM); 1757 return (0); 1758 } 1759 return (ENOMEM); 1760 } 1761 cbp->bio_offset = offset; 1762 cbp->bio_length = length; 1763 cbp->bio_done = g_raid3_done; 1764 switch (pbp->bio_cmd) { 1765 case BIO_READ: 1766 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) { 1767 /* 1768 * Replace invalid component with the parity 1769 * component. 1770 */ 1771 disk = &sc->sc_disks[sc->sc_ndisks - 1]; 1772 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1773 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED; 1774 } else if (round_robin && 1775 disk->d_no == sc->sc_round_robin) { 1776 /* 1777 * In round-robin mode skip one data component 1778 * and use parity component when reading. 1779 */ 1780 pbp->bio_driver2 = disk; 1781 disk = &sc->sc_disks[sc->sc_ndisks - 1]; 1782 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1783 sc->sc_round_robin++; 1784 round_robin = 0; 1785 } else if (verify && disk->d_no == sc->sc_ndisks - 1) { 1786 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1787 } 1788 break; 1789 case BIO_WRITE: 1790 case BIO_DELETE: 1791 if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 1792 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 1793 if (n == ndisks - 1) { 1794 /* 1795 * Active parity component, mark it as such. 1796 */ 1797 cbp->bio_cflags |= 1798 G_RAID3_BIO_CFLAG_PARITY; 1799 } 1800 } else { 1801 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED; 1802 if (n == ndisks - 1) { 1803 /* 1804 * Parity component is not connected, 1805 * so destroy its request. 1806 */ 1807 pbp->bio_pflags |= 1808 G_RAID3_BIO_PFLAG_NOPARITY; 1809 g_raid3_destroy_bio(sc, cbp); 1810 cbp = NULL; 1811 } else { 1812 cbp->bio_cflags |= 1813 G_RAID3_BIO_CFLAG_NODISK; 1814 disk = NULL; 1815 } 1816 } 1817 break; 1818 } 1819 if (cbp != NULL) 1820 cbp->bio_caller2 = disk; 1821 } 1822 switch (pbp->bio_cmd) { 1823 case BIO_READ: 1824 if (round_robin) { 1825 /* 1826 * If we are in round-robin mode and 'round_robin' is 1827 * still 1, it means, that we skipped parity component 1828 * for this read and must reset sc_round_robin field. 1829 */ 1830 sc->sc_round_robin = 0; 1831 } 1832 G_RAID3_FOREACH_SAFE_BIO(pbp, cbp, tmpbp) { 1833 disk = cbp->bio_caller2; 1834 cp = disk->d_consumer; 1835 cbp->bio_to = cp->provider; 1836 G_RAID3_LOGREQ(3, cbp, "Sending request."); 1837 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1838 ("Consumer %s not opened (r%dw%de%d).", 1839 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1840 cp->index++; 1841 g_io_request(cbp, cp); 1842 } 1843 break; 1844 case BIO_WRITE: 1845 case BIO_DELETE: 1846 /* 1847 * Put request onto inflight queue, so we can check if new 1848 * synchronization requests don't collide with it. 1849 */ 1850 bioq_insert_tail(&sc->sc_inflight, pbp); 1851 1852 /* 1853 * Bump syncid on first write. 1854 */ 1855 if ((sc->sc_bump_id & G_RAID3_BUMP_SYNCID) != 0) { 1856 sc->sc_bump_id &= ~G_RAID3_BUMP_SYNCID; 1857 g_raid3_bump_syncid(sc); 1858 } 1859 g_raid3_scatter(pbp); 1860 break; 1861 } 1862 return (0); 1863} 1864 1865static int 1866g_raid3_can_destroy(struct g_raid3_softc *sc) 1867{ 1868 struct g_geom *gp; 1869 struct g_consumer *cp; 1870 1871 g_topology_assert(); 1872 gp = sc->sc_geom; 1873 if (gp->softc == NULL) 1874 return (1); 1875 LIST_FOREACH(cp, &gp->consumer, consumer) { 1876 if (g_raid3_is_busy(sc, cp)) 1877 return (0); 1878 } 1879 gp = sc->sc_sync.ds_geom; 1880 LIST_FOREACH(cp, &gp->consumer, consumer) { 1881 if (g_raid3_is_busy(sc, cp)) 1882 return (0); 1883 } 1884 G_RAID3_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1885 sc->sc_name); 1886 return (1); 1887} 1888 1889static int 1890g_raid3_try_destroy(struct g_raid3_softc *sc) 1891{ 1892 1893 g_topology_assert_not(); 1894 sx_assert(&sc->sc_lock, SX_XLOCKED); 1895 1896 if (sc->sc_rootmount != NULL) { 1897 G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1898 sc->sc_rootmount); 1899 root_mount_rel(sc->sc_rootmount); 1900 sc->sc_rootmount = NULL; 1901 } 1902 1903 g_topology_lock(); 1904 if (!g_raid3_can_destroy(sc)) { 1905 g_topology_unlock(); 1906 return (0); 1907 } 1908 sc->sc_geom->softc = NULL; 1909 sc->sc_sync.ds_geom->softc = NULL; 1910 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_WAIT) != 0) { 1911 g_topology_unlock(); 1912 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, 1913 &sc->sc_worker); 1914 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1915 sx_xunlock(&sc->sc_lock); 1916 wakeup(&sc->sc_worker); 1917 sc->sc_worker = NULL; 1918 } else { 1919 g_topology_unlock(); 1920 g_raid3_destroy_device(sc); 1921 free(sc->sc_disks, M_RAID3); 1922 free(sc, M_RAID3); 1923 } 1924 return (1); 1925} 1926 1927/* 1928 * Worker thread. 1929 */ 1930static void 1931g_raid3_worker(void *arg) 1932{ 1933 struct g_raid3_softc *sc; 1934 struct g_raid3_event *ep; 1935 struct bio *bp; 1936 int timeout; 1937 1938 sc = arg; 1939 mtx_lock_spin(&sched_lock); 1940 sched_prio(curthread, PRIBIO); 1941 mtx_unlock_spin(&sched_lock); 1942 1943 sx_xlock(&sc->sc_lock); 1944 for (;;) { 1945 G_RAID3_DEBUG(5, "%s: Let's see...", __func__); 1946 /* 1947 * First take a look at events. 1948 * This is important to handle events before any I/O requests. 1949 */ 1950 ep = g_raid3_event_get(sc); 1951 if (ep != NULL) { 1952 g_raid3_event_remove(sc, ep); 1953 if ((ep->e_flags & G_RAID3_EVENT_DEVICE) != 0) { 1954 /* Update only device status. */ 1955 G_RAID3_DEBUG(3, 1956 "Running event for device %s.", 1957 sc->sc_name); 1958 ep->e_error = 0; 1959 g_raid3_update_device(sc, 1); 1960 } else { 1961 /* Update disk status. */ 1962 G_RAID3_DEBUG(3, "Running event for disk %s.", 1963 g_raid3_get_diskname(ep->e_disk)); 1964 ep->e_error = g_raid3_update_disk(ep->e_disk, 1965 ep->e_state); 1966 if (ep->e_error == 0) 1967 g_raid3_update_device(sc, 0); 1968 } 1969 if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0) { 1970 KASSERT(ep->e_error == 0, 1971 ("Error cannot be handled.")); 1972 g_raid3_event_free(ep); 1973 } else { 1974 ep->e_flags |= G_RAID3_EVENT_DONE; 1975 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, 1976 ep); 1977 mtx_lock(&sc->sc_events_mtx); 1978 wakeup(ep); 1979 mtx_unlock(&sc->sc_events_mtx); 1980 } 1981 if ((sc->sc_flags & 1982 G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 1983 if (g_raid3_try_destroy(sc)) { 1984 curthread->td_pflags &= ~TDP_GEOM; 1985 G_RAID3_DEBUG(1, "Thread exiting."); 1986 kthread_exit(0); 1987 } 1988 } 1989 G_RAID3_DEBUG(5, "%s: I'm here 1.", __func__); 1990 continue; 1991 } 1992 /* 1993 * Check if we can mark array as CLEAN and if we can't take 1994 * how much seconds should we wait. 1995 */ 1996 timeout = g_raid3_idle(sc, -1); 1997 /* 1998 * Now I/O requests. 1999 */ 2000 /* Get first request from the queue. */ 2001 mtx_lock(&sc->sc_queue_mtx); 2002 bp = bioq_first(&sc->sc_queue); 2003 if (bp == NULL) { 2004 if ((sc->sc_flags & 2005 G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 2006 mtx_unlock(&sc->sc_queue_mtx); 2007 if (g_raid3_try_destroy(sc)) { 2008 curthread->td_pflags &= ~TDP_GEOM; 2009 G_RAID3_DEBUG(1, "Thread exiting."); 2010 kthread_exit(0); 2011 } 2012 mtx_lock(&sc->sc_queue_mtx); 2013 } 2014 sx_xunlock(&sc->sc_lock); 2015 /* 2016 * XXX: We can miss an event here, because an event 2017 * can be added without sx-device-lock and without 2018 * mtx-queue-lock. Maybe I should just stop using 2019 * dedicated mutex for events synchronization and 2020 * stick with the queue lock? 2021 * The event will hang here until next I/O request 2022 * or next event is received. 2023 */ 2024 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "r3:w1", 2025 timeout * hz); 2026 sx_xlock(&sc->sc_lock); 2027 G_RAID3_DEBUG(5, "%s: I'm here 4.", __func__); 2028 continue; 2029 } 2030process: 2031 bioq_remove(&sc->sc_queue, bp); 2032 mtx_unlock(&sc->sc_queue_mtx); 2033 2034 if ((bp->bio_cflags & G_RAID3_BIO_CFLAG_REGULAR) != 0) 2035 g_raid3_regular_request(bp); 2036 else if ((bp->bio_cflags & G_RAID3_BIO_CFLAG_SYNC) != 0) 2037 g_raid3_sync_request(bp); 2038 else if (g_raid3_register_request(bp) != 0) { 2039 mtx_lock(&sc->sc_queue_mtx); 2040 bioq_insert_head(&sc->sc_queue, bp); 2041 /* 2042 * We are short in memory, let see if there are finished 2043 * request we can free. 2044 */ 2045 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 2046 if (bp->bio_cflags & G_RAID3_BIO_CFLAG_REGULAR) 2047 goto process; 2048 } 2049 /* 2050 * No finished regular request, so at least keep 2051 * synchronization running. 2052 */ 2053 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 2054 if (bp->bio_cflags & G_RAID3_BIO_CFLAG_SYNC) 2055 goto process; 2056 } 2057 sx_xunlock(&sc->sc_lock); 2058 MSLEEP(&sc->sc_queue, &sc->sc_queue_mtx, PRIBIO | PDROP, 2059 "r3:lowmem", hz / 10); 2060 sx_xlock(&sc->sc_lock); 2061 } 2062 G_RAID3_DEBUG(5, "%s: I'm here 9.", __func__); 2063 } 2064} 2065 2066static void 2067g_raid3_update_idle(struct g_raid3_softc *sc, struct g_raid3_disk *disk) 2068{ 2069 2070 sx_assert(&sc->sc_lock, SX_LOCKED); 2071 if (!sc->sc_idle && (disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) == 0) { 2072 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as dirty.", 2073 g_raid3_get_diskname(disk), sc->sc_name); 2074 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY; 2075 } else if (sc->sc_idle && 2076 (disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) != 0) { 2077 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as clean.", 2078 g_raid3_get_diskname(disk), sc->sc_name); 2079 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2080 } 2081} 2082 2083static void 2084g_raid3_sync_start(struct g_raid3_softc *sc) 2085{ 2086 struct g_raid3_disk *disk; 2087 struct g_consumer *cp; 2088 struct bio *bp; 2089 int error; 2090 u_int n; 2091 2092 g_topology_assert_not(); 2093 sx_assert(&sc->sc_lock, SX_XLOCKED); 2094 2095 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED, 2096 ("Device not in DEGRADED state (%s, %u).", sc->sc_name, 2097 sc->sc_state)); 2098 KASSERT(sc->sc_syncdisk == NULL, ("Syncdisk is not NULL (%s, %u).", 2099 sc->sc_name, sc->sc_state)); 2100 disk = NULL; 2101 for (n = 0; n < sc->sc_ndisks; n++) { 2102 if (sc->sc_disks[n].d_state != G_RAID3_DISK_STATE_SYNCHRONIZING) 2103 continue; 2104 disk = &sc->sc_disks[n]; 2105 break; 2106 } 2107 if (disk == NULL) 2108 return; 2109 2110 sx_xunlock(&sc->sc_lock); 2111 g_topology_lock(); 2112 cp = g_new_consumer(sc->sc_sync.ds_geom); 2113 error = g_attach(cp, sc->sc_provider); 2114 KASSERT(error == 0, 2115 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 2116 error = g_access(cp, 1, 0, 0); 2117 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 2118 g_topology_unlock(); 2119 sx_xlock(&sc->sc_lock); 2120 2121 G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 2122 g_raid3_get_diskname(disk)); 2123 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY; 2124 KASSERT(disk->d_sync.ds_consumer == NULL, 2125 ("Sync consumer already exists (device=%s, disk=%s).", 2126 sc->sc_name, g_raid3_get_diskname(disk))); 2127 2128 disk->d_sync.ds_consumer = cp; 2129 disk->d_sync.ds_consumer->private = disk; 2130 disk->d_sync.ds_consumer->index = 0; 2131 sc->sc_syncdisk = disk; 2132 2133 /* 2134 * Allocate memory for synchronization bios and initialize them. 2135 */ 2136 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_raid3_syncreqs, 2137 M_RAID3, M_WAITOK); 2138 for (n = 0; n < g_raid3_syncreqs; n++) { 2139 bp = g_alloc_bio(); 2140 disk->d_sync.ds_bios[n] = bp; 2141 bp->bio_parent = NULL; 2142 bp->bio_cmd = BIO_READ; 2143 bp->bio_data = malloc(MAXPHYS, M_RAID3, M_WAITOK); 2144 bp->bio_cflags = 0; 2145 bp->bio_offset = disk->d_sync.ds_offset * (sc->sc_ndisks - 1); 2146 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 2147 disk->d_sync.ds_offset += bp->bio_length / (sc->sc_ndisks - 1); 2148 bp->bio_done = g_raid3_sync_done; 2149 bp->bio_from = disk->d_sync.ds_consumer; 2150 bp->bio_to = sc->sc_provider; 2151 bp->bio_caller1 = (void *)(uintptr_t)n; 2152 } 2153 2154 /* Set the number of in-flight synchronization requests. */ 2155 disk->d_sync.ds_inflight = g_raid3_syncreqs; 2156 2157 /* 2158 * Fire off first synchronization requests. 2159 */ 2160 for (n = 0; n < g_raid3_syncreqs; n++) { 2161 bp = disk->d_sync.ds_bios[n]; 2162 G_RAID3_LOGREQ(3, bp, "Sending synchronization request."); 2163 disk->d_sync.ds_consumer->index++; 2164 /* 2165 * Delay the request if it is colliding with a regular request. 2166 */ 2167 if (g_raid3_regular_collision(sc, bp)) 2168 g_raid3_sync_delay(sc, bp); 2169 else 2170 g_io_request(bp, disk->d_sync.ds_consumer); 2171 } 2172} 2173 2174/* 2175 * Stop synchronization process. 2176 * type: 0 - synchronization finished 2177 * 1 - synchronization stopped 2178 */ 2179static void 2180g_raid3_sync_stop(struct g_raid3_softc *sc, int type) 2181{ 2182 struct g_raid3_disk *disk; 2183 struct g_consumer *cp; 2184 2185 g_topology_assert_not(); 2186 sx_assert(&sc->sc_lock, SX_LOCKED); 2187 2188 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED, 2189 ("Device not in DEGRADED state (%s, %u).", sc->sc_name, 2190 sc->sc_state)); 2191 disk = sc->sc_syncdisk; 2192 sc->sc_syncdisk = NULL; 2193 KASSERT(disk != NULL, ("No disk was synchronized (%s).", sc->sc_name)); 2194 KASSERT(disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING, 2195 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2196 g_raid3_disk_state2str(disk->d_state))); 2197 if (disk->d_sync.ds_consumer == NULL) 2198 return; 2199 2200 if (type == 0) { 2201 G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2202 sc->sc_name, g_raid3_get_diskname(disk)); 2203 } else /* if (type == 1) */ { 2204 G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2205 sc->sc_name, g_raid3_get_diskname(disk)); 2206 } 2207 free(disk->d_sync.ds_bios, M_RAID3); 2208 disk->d_sync.ds_bios = NULL; 2209 cp = disk->d_sync.ds_consumer; 2210 disk->d_sync.ds_consumer = NULL; 2211 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2212 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2213 g_topology_lock(); 2214 g_raid3_kill_consumer(sc, cp); 2215 g_topology_unlock(); 2216 sx_xlock(&sc->sc_lock); 2217} 2218 2219static void 2220g_raid3_launch_provider(struct g_raid3_softc *sc) 2221{ 2222 struct g_provider *pp; 2223 2224 sx_assert(&sc->sc_lock, SX_LOCKED); 2225 2226 g_topology_lock(); 2227 pp = g_new_providerf(sc->sc_geom, "raid3/%s", sc->sc_name); 2228 pp->mediasize = sc->sc_mediasize; 2229 pp->sectorsize = sc->sc_sectorsize; 2230 sc->sc_provider = pp; 2231 g_error_provider(pp, 0); 2232 g_topology_unlock(); 2233 G_RAID3_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 2234 pp->name); 2235 if (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED) 2236 g_raid3_sync_start(sc); 2237} 2238 2239static void 2240g_raid3_destroy_provider(struct g_raid3_softc *sc) 2241{ 2242 struct bio *bp; 2243 2244 g_topology_assert_not(); 2245 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2246 sc->sc_name)); 2247 2248 g_topology_lock(); 2249 g_error_provider(sc->sc_provider, ENXIO); 2250 mtx_lock(&sc->sc_queue_mtx); 2251 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 2252 bioq_remove(&sc->sc_queue, bp); 2253 g_io_deliver(bp, ENXIO); 2254 } 2255 mtx_unlock(&sc->sc_queue_mtx); 2256 G_RAID3_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 2257 sc->sc_provider->name); 2258 sc->sc_provider->flags |= G_PF_WITHER; 2259 g_orphan_provider(sc->sc_provider, ENXIO); 2260 g_topology_unlock(); 2261 sc->sc_provider = NULL; 2262 if (sc->sc_syncdisk != NULL) 2263 g_raid3_sync_stop(sc, 1); 2264} 2265 2266static void 2267g_raid3_go(void *arg) 2268{ 2269 struct g_raid3_softc *sc; 2270 2271 sc = arg; 2272 G_RAID3_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2273 g_raid3_event_send(sc, 0, 2274 G_RAID3_EVENT_DONTWAIT | G_RAID3_EVENT_DEVICE); 2275} 2276 2277static u_int 2278g_raid3_determine_state(struct g_raid3_disk *disk) 2279{ 2280 struct g_raid3_softc *sc; 2281 u_int state; 2282 2283 sc = disk->d_softc; 2284 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2285 if ((disk->d_flags & 2286 G_RAID3_DISK_FLAG_SYNCHRONIZING) == 0) { 2287 /* Disk does not need synchronization. */ 2288 state = G_RAID3_DISK_STATE_ACTIVE; 2289 } else { 2290 if ((sc->sc_flags & 2291 G_RAID3_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2292 (disk->d_flags & 2293 G_RAID3_DISK_FLAG_FORCE_SYNC) != 0) { 2294 /* 2295 * We can start synchronization from 2296 * the stored offset. 2297 */ 2298 state = G_RAID3_DISK_STATE_SYNCHRONIZING; 2299 } else { 2300 state = G_RAID3_DISK_STATE_STALE; 2301 } 2302 } 2303 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2304 /* 2305 * Reset all synchronization data for this disk, 2306 * because if it even was synchronized, it was 2307 * synchronized to disks with different syncid. 2308 */ 2309 disk->d_flags |= G_RAID3_DISK_FLAG_SYNCHRONIZING; 2310 disk->d_sync.ds_offset = 0; 2311 disk->d_sync.ds_offset_done = 0; 2312 disk->d_sync.ds_syncid = sc->sc_syncid; 2313 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2314 (disk->d_flags & G_RAID3_DISK_FLAG_FORCE_SYNC) != 0) { 2315 state = G_RAID3_DISK_STATE_SYNCHRONIZING; 2316 } else { 2317 state = G_RAID3_DISK_STATE_STALE; 2318 } 2319 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2320 /* 2321 * Not good, NOT GOOD! 2322 * It means that device was started on stale disks 2323 * and more fresh disk just arrive. 2324 * If there were writes, device is fucked up, sorry. 2325 * I think the best choice here is don't touch 2326 * this disk and inform the user laudly. 2327 */ 2328 G_RAID3_DEBUG(0, "Device %s was started before the freshest " 2329 "disk (%s) arrives!! It will not be connected to the " 2330 "running device.", sc->sc_name, 2331 g_raid3_get_diskname(disk)); 2332 g_raid3_destroy_disk(disk); 2333 state = G_RAID3_DISK_STATE_NONE; 2334 /* Return immediately, because disk was destroyed. */ 2335 return (state); 2336 } 2337 G_RAID3_DEBUG(3, "State for %s disk: %s.", 2338 g_raid3_get_diskname(disk), g_raid3_disk_state2str(state)); 2339 return (state); 2340} 2341 2342/* 2343 * Update device state. 2344 */ 2345static void 2346g_raid3_update_device(struct g_raid3_softc *sc, boolean_t force) 2347{ 2348 struct g_raid3_disk *disk; 2349 u_int state; 2350 2351 sx_assert(&sc->sc_lock, SX_XLOCKED); 2352 2353 switch (sc->sc_state) { 2354 case G_RAID3_DEVICE_STATE_STARTING: 2355 { 2356 u_int n, ndirty, ndisks, genid, syncid; 2357 2358 KASSERT(sc->sc_provider == NULL, 2359 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2360 /* 2361 * Are we ready? We are, if all disks are connected or 2362 * one disk is missing and 'force' is true. 2363 */ 2364 if (g_raid3_ndisks(sc, -1) + force == sc->sc_ndisks) { 2365 if (!force) 2366 callout_drain(&sc->sc_callout); 2367 } else { 2368 if (force) { 2369 /* 2370 * Timeout expired, so destroy device. 2371 */ 2372 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 2373 G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", 2374 __LINE__, sc->sc_rootmount); 2375 root_mount_rel(sc->sc_rootmount); 2376 sc->sc_rootmount = NULL; 2377 } 2378 return; 2379 } 2380 2381 /* 2382 * Find the biggest genid. 2383 */ 2384 genid = 0; 2385 for (n = 0; n < sc->sc_ndisks; n++) { 2386 disk = &sc->sc_disks[n]; 2387 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2388 continue; 2389 if (disk->d_genid > genid) 2390 genid = disk->d_genid; 2391 } 2392 sc->sc_genid = genid; 2393 /* 2394 * Remove all disks without the biggest genid. 2395 */ 2396 for (n = 0; n < sc->sc_ndisks; n++) { 2397 disk = &sc->sc_disks[n]; 2398 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2399 continue; 2400 if (disk->d_genid < genid) { 2401 G_RAID3_DEBUG(0, 2402 "Component %s (device %s) broken, skipping.", 2403 g_raid3_get_diskname(disk), sc->sc_name); 2404 g_raid3_destroy_disk(disk); 2405 } 2406 } 2407 2408 /* 2409 * There must be at least 'sc->sc_ndisks - 1' components 2410 * with the same syncid and without SYNCHRONIZING flag. 2411 */ 2412 2413 /* 2414 * Find the biggest syncid, number of valid components and 2415 * number of dirty components. 2416 */ 2417 ndirty = ndisks = syncid = 0; 2418 for (n = 0; n < sc->sc_ndisks; n++) { 2419 disk = &sc->sc_disks[n]; 2420 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2421 continue; 2422 if ((disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) != 0) 2423 ndirty++; 2424 if (disk->d_sync.ds_syncid > syncid) { 2425 syncid = disk->d_sync.ds_syncid; 2426 ndisks = 0; 2427 } else if (disk->d_sync.ds_syncid < syncid) { 2428 continue; 2429 } 2430 if ((disk->d_flags & 2431 G_RAID3_DISK_FLAG_SYNCHRONIZING) != 0) { 2432 continue; 2433 } 2434 ndisks++; 2435 } 2436 /* 2437 * Do we have enough valid components? 2438 */ 2439 if (ndisks + 1 < sc->sc_ndisks) { 2440 G_RAID3_DEBUG(0, 2441 "Device %s is broken, too few valid components.", 2442 sc->sc_name); 2443 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 2444 return; 2445 } 2446 /* 2447 * If there is one DIRTY component and all disks are present, 2448 * mark it for synchronization. If there is more than one DIRTY 2449 * component, mark parity component for synchronization. 2450 */ 2451 if (ndisks == sc->sc_ndisks && ndirty == 1) { 2452 for (n = 0; n < sc->sc_ndisks; n++) { 2453 disk = &sc->sc_disks[n]; 2454 if ((disk->d_flags & 2455 G_RAID3_DISK_FLAG_DIRTY) == 0) { 2456 continue; 2457 } 2458 disk->d_flags |= 2459 G_RAID3_DISK_FLAG_SYNCHRONIZING; 2460 } 2461 } else if (ndisks == sc->sc_ndisks && ndirty > 1) { 2462 disk = &sc->sc_disks[sc->sc_ndisks - 1]; 2463 disk->d_flags |= G_RAID3_DISK_FLAG_SYNCHRONIZING; 2464 } 2465 2466 sc->sc_syncid = syncid; 2467 if (force) { 2468 /* Remember to bump syncid on first write. */ 2469 sc->sc_bump_id |= G_RAID3_BUMP_SYNCID; 2470 } 2471 if (ndisks == sc->sc_ndisks) 2472 state = G_RAID3_DEVICE_STATE_COMPLETE; 2473 else /* if (ndisks == sc->sc_ndisks - 1) */ 2474 state = G_RAID3_DEVICE_STATE_DEGRADED; 2475 G_RAID3_DEBUG(1, "Device %s state changed from %s to %s.", 2476 sc->sc_name, g_raid3_device_state2str(sc->sc_state), 2477 g_raid3_device_state2str(state)); 2478 sc->sc_state = state; 2479 for (n = 0; n < sc->sc_ndisks; n++) { 2480 disk = &sc->sc_disks[n]; 2481 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2482 continue; 2483 state = g_raid3_determine_state(disk); 2484 g_raid3_event_send(disk, state, G_RAID3_EVENT_DONTWAIT); 2485 if (state == G_RAID3_DISK_STATE_STALE) 2486 sc->sc_bump_id |= G_RAID3_BUMP_SYNCID; 2487 } 2488 break; 2489 } 2490 case G_RAID3_DEVICE_STATE_DEGRADED: 2491 /* 2492 * Genid need to be bumped immediately, so do it here. 2493 */ 2494 if ((sc->sc_bump_id & G_RAID3_BUMP_GENID) != 0) { 2495 sc->sc_bump_id &= ~G_RAID3_BUMP_GENID; 2496 g_raid3_bump_genid(sc); 2497 } 2498 2499 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NEW) > 0) 2500 return; 2501 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) < 2502 sc->sc_ndisks - 1) { 2503 if (sc->sc_provider != NULL) 2504 g_raid3_destroy_provider(sc); 2505 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 2506 return; 2507 } 2508 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) == 2509 sc->sc_ndisks) { 2510 state = G_RAID3_DEVICE_STATE_COMPLETE; 2511 G_RAID3_DEBUG(1, 2512 "Device %s state changed from %s to %s.", 2513 sc->sc_name, g_raid3_device_state2str(sc->sc_state), 2514 g_raid3_device_state2str(state)); 2515 sc->sc_state = state; 2516 } 2517 if (sc->sc_provider == NULL) 2518 g_raid3_launch_provider(sc); 2519 if (sc->sc_rootmount != NULL) { 2520 G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2521 sc->sc_rootmount); 2522 root_mount_rel(sc->sc_rootmount); 2523 sc->sc_rootmount = NULL; 2524 } 2525 break; 2526 case G_RAID3_DEVICE_STATE_COMPLETE: 2527 /* 2528 * Genid need to be bumped immediately, so do it here. 2529 */ 2530 if ((sc->sc_bump_id & G_RAID3_BUMP_GENID) != 0) { 2531 sc->sc_bump_id &= ~G_RAID3_BUMP_GENID; 2532 g_raid3_bump_genid(sc); 2533 } 2534 2535 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NEW) > 0) 2536 return; 2537 KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) >= 2538 sc->sc_ndisks - 1, 2539 ("Too few ACTIVE components in COMPLETE state (device %s).", 2540 sc->sc_name)); 2541 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) == 2542 sc->sc_ndisks - 1) { 2543 state = G_RAID3_DEVICE_STATE_DEGRADED; 2544 G_RAID3_DEBUG(1, 2545 "Device %s state changed from %s to %s.", 2546 sc->sc_name, g_raid3_device_state2str(sc->sc_state), 2547 g_raid3_device_state2str(state)); 2548 sc->sc_state = state; 2549 } 2550 if (sc->sc_provider == NULL) 2551 g_raid3_launch_provider(sc); 2552 if (sc->sc_rootmount != NULL) { 2553 G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2554 sc->sc_rootmount); 2555 root_mount_rel(sc->sc_rootmount); 2556 sc->sc_rootmount = NULL; 2557 } 2558 break; 2559 default: 2560 KASSERT(1 == 0, ("Wrong device state (%s, %s).", sc->sc_name, 2561 g_raid3_device_state2str(sc->sc_state))); 2562 break; 2563 } 2564} 2565 2566/* 2567 * Update disk state and device state if needed. 2568 */ 2569#define DISK_STATE_CHANGED() G_RAID3_DEBUG(1, \ 2570 "Disk %s state changed from %s to %s (device %s).", \ 2571 g_raid3_get_diskname(disk), \ 2572 g_raid3_disk_state2str(disk->d_state), \ 2573 g_raid3_disk_state2str(state), sc->sc_name) 2574static int 2575g_raid3_update_disk(struct g_raid3_disk *disk, u_int state) 2576{ 2577 struct g_raid3_softc *sc; 2578 2579 sc = disk->d_softc; 2580 sx_assert(&sc->sc_lock, SX_XLOCKED); 2581 2582again: 2583 G_RAID3_DEBUG(3, "Changing disk %s state from %s to %s.", 2584 g_raid3_get_diskname(disk), g_raid3_disk_state2str(disk->d_state), 2585 g_raid3_disk_state2str(state)); 2586 switch (state) { 2587 case G_RAID3_DISK_STATE_NEW: 2588 /* 2589 * Possible scenarios: 2590 * 1. New disk arrive. 2591 */ 2592 /* Previous state should be NONE. */ 2593 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NONE, 2594 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2595 g_raid3_disk_state2str(disk->d_state))); 2596 DISK_STATE_CHANGED(); 2597 2598 disk->d_state = state; 2599 G_RAID3_DEBUG(0, "Device %s: provider %s detected.", 2600 sc->sc_name, g_raid3_get_diskname(disk)); 2601 if (sc->sc_state == G_RAID3_DEVICE_STATE_STARTING) 2602 break; 2603 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2604 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2605 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2606 g_raid3_device_state2str(sc->sc_state), 2607 g_raid3_get_diskname(disk), 2608 g_raid3_disk_state2str(disk->d_state))); 2609 state = g_raid3_determine_state(disk); 2610 if (state != G_RAID3_DISK_STATE_NONE) 2611 goto again; 2612 break; 2613 case G_RAID3_DISK_STATE_ACTIVE: 2614 /* 2615 * Possible scenarios: 2616 * 1. New disk does not need synchronization. 2617 * 2. Synchronization process finished successfully. 2618 */ 2619 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2620 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2621 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2622 g_raid3_device_state2str(sc->sc_state), 2623 g_raid3_get_diskname(disk), 2624 g_raid3_disk_state2str(disk->d_state))); 2625 /* Previous state should be NEW or SYNCHRONIZING. */ 2626 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW || 2627 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING, 2628 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2629 g_raid3_disk_state2str(disk->d_state))); 2630 DISK_STATE_CHANGED(); 2631 2632 if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 2633 disk->d_flags &= ~G_RAID3_DISK_FLAG_SYNCHRONIZING; 2634 disk->d_flags &= ~G_RAID3_DISK_FLAG_FORCE_SYNC; 2635 g_raid3_sync_stop(sc, 0); 2636 } 2637 disk->d_state = state; 2638 disk->d_sync.ds_offset = 0; 2639 disk->d_sync.ds_offset_done = 0; 2640 g_raid3_update_idle(sc, disk); 2641 g_raid3_update_metadata(disk); 2642 G_RAID3_DEBUG(0, "Device %s: provider %s activated.", 2643 sc->sc_name, g_raid3_get_diskname(disk)); 2644 break; 2645 case G_RAID3_DISK_STATE_STALE: 2646 /* 2647 * Possible scenarios: 2648 * 1. Stale disk was connected. 2649 */ 2650 /* Previous state should be NEW. */ 2651 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW, 2652 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2653 g_raid3_disk_state2str(disk->d_state))); 2654 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2655 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2656 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2657 g_raid3_device_state2str(sc->sc_state), 2658 g_raid3_get_diskname(disk), 2659 g_raid3_disk_state2str(disk->d_state))); 2660 /* 2661 * STALE state is only possible if device is marked 2662 * NOAUTOSYNC. 2663 */ 2664 KASSERT((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOAUTOSYNC) != 0, 2665 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2666 g_raid3_device_state2str(sc->sc_state), 2667 g_raid3_get_diskname(disk), 2668 g_raid3_disk_state2str(disk->d_state))); 2669 DISK_STATE_CHANGED(); 2670 2671 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2672 disk->d_state = state; 2673 g_raid3_update_metadata(disk); 2674 G_RAID3_DEBUG(0, "Device %s: provider %s is stale.", 2675 sc->sc_name, g_raid3_get_diskname(disk)); 2676 break; 2677 case G_RAID3_DISK_STATE_SYNCHRONIZING: 2678 /* 2679 * Possible scenarios: 2680 * 1. Disk which needs synchronization was connected. 2681 */ 2682 /* Previous state should be NEW. */ 2683 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW, 2684 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2685 g_raid3_disk_state2str(disk->d_state))); 2686 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2687 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2688 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2689 g_raid3_device_state2str(sc->sc_state), 2690 g_raid3_get_diskname(disk), 2691 g_raid3_disk_state2str(disk->d_state))); 2692 DISK_STATE_CHANGED(); 2693 2694 if (disk->d_state == G_RAID3_DISK_STATE_NEW) 2695 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2696 disk->d_state = state; 2697 if (sc->sc_provider != NULL) { 2698 g_raid3_sync_start(sc); 2699 g_raid3_update_metadata(disk); 2700 } 2701 break; 2702 case G_RAID3_DISK_STATE_DISCONNECTED: 2703 /* 2704 * Possible scenarios: 2705 * 1. Device wasn't running yet, but disk disappear. 2706 * 2. Disk was active and disapppear. 2707 * 3. Disk disappear during synchronization process. 2708 */ 2709 if (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2710 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 2711 /* 2712 * Previous state should be ACTIVE, STALE or 2713 * SYNCHRONIZING. 2714 */ 2715 KASSERT(disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 2716 disk->d_state == G_RAID3_DISK_STATE_STALE || 2717 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING, 2718 ("Wrong disk state (%s, %s).", 2719 g_raid3_get_diskname(disk), 2720 g_raid3_disk_state2str(disk->d_state))); 2721 } else if (sc->sc_state == G_RAID3_DEVICE_STATE_STARTING) { 2722 /* Previous state should be NEW. */ 2723 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW, 2724 ("Wrong disk state (%s, %s).", 2725 g_raid3_get_diskname(disk), 2726 g_raid3_disk_state2str(disk->d_state))); 2727 /* 2728 * Reset bumping syncid if disk disappeared in STARTING 2729 * state. 2730 */ 2731 if ((sc->sc_bump_id & G_RAID3_BUMP_SYNCID) != 0) 2732 sc->sc_bump_id &= ~G_RAID3_BUMP_SYNCID; 2733#ifdef INVARIANTS 2734 } else { 2735 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2736 sc->sc_name, 2737 g_raid3_device_state2str(sc->sc_state), 2738 g_raid3_get_diskname(disk), 2739 g_raid3_disk_state2str(disk->d_state))); 2740#endif 2741 } 2742 DISK_STATE_CHANGED(); 2743 G_RAID3_DEBUG(0, "Device %s: provider %s disconnected.", 2744 sc->sc_name, g_raid3_get_diskname(disk)); 2745 2746 g_raid3_destroy_disk(disk); 2747 break; 2748 default: 2749 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2750 break; 2751 } 2752 return (0); 2753} 2754#undef DISK_STATE_CHANGED 2755 2756int 2757g_raid3_read_metadata(struct g_consumer *cp, struct g_raid3_metadata *md) 2758{ 2759 struct g_provider *pp; 2760 u_char *buf; 2761 int error; 2762 2763 g_topology_assert(); 2764 2765 error = g_access(cp, 1, 0, 0); 2766 if (error != 0) 2767 return (error); 2768 pp = cp->provider; 2769 g_topology_unlock(); 2770 /* Metadata are stored on last sector. */ 2771 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2772 &error); 2773 g_topology_lock(); 2774 g_access(cp, -1, 0, 0); 2775 if (buf == NULL) { 2776 G_RAID3_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2777 cp->provider->name, error); 2778 return (error); 2779 } 2780 2781 /* Decode metadata. */ 2782 error = raid3_metadata_decode(buf, md); 2783 g_free(buf); 2784 if (strcmp(md->md_magic, G_RAID3_MAGIC) != 0) 2785 return (EINVAL); 2786 if (md->md_version > G_RAID3_VERSION) { 2787 G_RAID3_DEBUG(0, 2788 "Kernel module is too old to handle metadata from %s.", 2789 cp->provider->name); 2790 return (EINVAL); 2791 } 2792 if (error != 0) { 2793 G_RAID3_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2794 cp->provider->name); 2795 return (error); 2796 } 2797 2798 return (0); 2799} 2800 2801static int 2802g_raid3_check_metadata(struct g_raid3_softc *sc, struct g_provider *pp, 2803 struct g_raid3_metadata *md) 2804{ 2805 2806 if (md->md_no >= sc->sc_ndisks) { 2807 G_RAID3_DEBUG(1, "Invalid disk %s number (no=%u), skipping.", 2808 pp->name, md->md_no); 2809 return (EINVAL); 2810 } 2811 if (sc->sc_disks[md->md_no].d_state != G_RAID3_DISK_STATE_NODISK) { 2812 G_RAID3_DEBUG(1, "Disk %s (no=%u) already exists, skipping.", 2813 pp->name, md->md_no); 2814 return (EEXIST); 2815 } 2816 if (md->md_all != sc->sc_ndisks) { 2817 G_RAID3_DEBUG(1, 2818 "Invalid '%s' field on disk %s (device %s), skipping.", 2819 "md_all", pp->name, sc->sc_name); 2820 return (EINVAL); 2821 } 2822 if (md->md_mediasize != sc->sc_mediasize) { 2823 G_RAID3_DEBUG(1, 2824 "Invalid '%s' field on disk %s (device %s), skipping.", 2825 "md_mediasize", pp->name, sc->sc_name); 2826 return (EINVAL); 2827 } 2828 if ((md->md_mediasize % (sc->sc_ndisks - 1)) != 0) { 2829 G_RAID3_DEBUG(1, 2830 "Invalid '%s' field on disk %s (device %s), skipping.", 2831 "md_mediasize", pp->name, sc->sc_name); 2832 return (EINVAL); 2833 } 2834 if ((sc->sc_mediasize / (sc->sc_ndisks - 1)) > pp->mediasize) { 2835 G_RAID3_DEBUG(1, 2836 "Invalid size of disk %s (device %s), skipping.", pp->name, 2837 sc->sc_name); 2838 return (EINVAL); 2839 } 2840 if ((md->md_sectorsize / pp->sectorsize) < sc->sc_ndisks - 1) { 2841 G_RAID3_DEBUG(1, 2842 "Invalid '%s' field on disk %s (device %s), skipping.", 2843 "md_sectorsize", pp->name, sc->sc_name); 2844 return (EINVAL); 2845 } 2846 if (md->md_sectorsize != sc->sc_sectorsize) { 2847 G_RAID3_DEBUG(1, 2848 "Invalid '%s' field on disk %s (device %s), skipping.", 2849 "md_sectorsize", pp->name, sc->sc_name); 2850 return (EINVAL); 2851 } 2852 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2853 G_RAID3_DEBUG(1, 2854 "Invalid sector size of disk %s (device %s), skipping.", 2855 pp->name, sc->sc_name); 2856 return (EINVAL); 2857 } 2858 if ((md->md_mflags & ~G_RAID3_DEVICE_FLAG_MASK) != 0) { 2859 G_RAID3_DEBUG(1, 2860 "Invalid device flags on disk %s (device %s), skipping.", 2861 pp->name, sc->sc_name); 2862 return (EINVAL); 2863 } 2864 if ((md->md_mflags & G_RAID3_DEVICE_FLAG_VERIFY) != 0 && 2865 (md->md_mflags & G_RAID3_DEVICE_FLAG_ROUND_ROBIN) != 0) { 2866 /* 2867 * VERIFY and ROUND-ROBIN options are mutally exclusive. 2868 */ 2869 G_RAID3_DEBUG(1, "Both VERIFY and ROUND-ROBIN flags exist on " 2870 "disk %s (device %s), skipping.", pp->name, sc->sc_name); 2871 return (EINVAL); 2872 } 2873 if ((md->md_dflags & ~G_RAID3_DISK_FLAG_MASK) != 0) { 2874 G_RAID3_DEBUG(1, 2875 "Invalid disk flags on disk %s (device %s), skipping.", 2876 pp->name, sc->sc_name); 2877 return (EINVAL); 2878 } 2879 return (0); 2880} 2881 2882int 2883g_raid3_add_disk(struct g_raid3_softc *sc, struct g_provider *pp, 2884 struct g_raid3_metadata *md) 2885{ 2886 struct g_raid3_disk *disk; 2887 int error; 2888 2889 g_topology_assert_not(); 2890 G_RAID3_DEBUG(2, "Adding disk %s.", pp->name); 2891 2892 error = g_raid3_check_metadata(sc, pp, md); 2893 if (error != 0) 2894 return (error); 2895 if (sc->sc_state != G_RAID3_DEVICE_STATE_STARTING && 2896 md->md_genid < sc->sc_genid) { 2897 G_RAID3_DEBUG(0, "Component %s (device %s) broken, skipping.", 2898 pp->name, sc->sc_name); 2899 return (EINVAL); 2900 } 2901 disk = g_raid3_init_disk(sc, pp, md, &error); 2902 if (disk == NULL) 2903 return (error); 2904 error = g_raid3_event_send(disk, G_RAID3_DISK_STATE_NEW, 2905 G_RAID3_EVENT_WAIT); 2906 if (error != 0) 2907 return (error); 2908 if (md->md_version < G_RAID3_VERSION) { 2909 G_RAID3_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2910 pp->name, md->md_version, G_RAID3_VERSION); 2911 g_raid3_update_metadata(disk); 2912 } 2913 return (0); 2914} 2915 2916static void 2917g_raid3_destroy_delayed(void *arg, int flag) 2918{ 2919 struct g_raid3_softc *sc; 2920 int error; 2921 2922 if (flag == EV_CANCEL) { 2923 G_RAID3_DEBUG(1, "Destroying canceled."); 2924 return; 2925 } 2926 sc = arg; 2927 g_topology_unlock(); 2928 sx_xlock(&sc->sc_lock); 2929 KASSERT((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) == 0, 2930 ("DESTROY flag set on %s.", sc->sc_name)); 2931 KASSERT((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROYING) != 0, 2932 ("DESTROYING flag not set on %s.", sc->sc_name)); 2933 G_RAID3_DEBUG(0, "Destroying %s (delayed).", sc->sc_name); 2934 error = g_raid3_destroy(sc, G_RAID3_DESTROY_SOFT); 2935 if (error != 0) { 2936 G_RAID3_DEBUG(0, "Cannot destroy %s.", sc->sc_name); 2937 sx_xunlock(&sc->sc_lock); 2938 } 2939 g_topology_lock(); 2940} 2941 2942static int 2943g_raid3_access(struct g_provider *pp, int acr, int acw, int ace) 2944{ 2945 struct g_raid3_softc *sc; 2946 int dcr, dcw, dce, error = 0; 2947 2948 g_topology_assert(); 2949 G_RAID3_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2950 acw, ace); 2951 2952 sc = pp->geom->softc; 2953 if (sc == NULL && acr <= 0 && acw <= 0 && ace <= 0) 2954 return (0); 2955 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 2956 2957 dcr = pp->acr + acr; 2958 dcw = pp->acw + acw; 2959 dce = pp->ace + ace; 2960 2961 g_topology_unlock(); 2962 sx_xlock(&sc->sc_lock); 2963 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) != 0 || 2964 g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) < sc->sc_ndisks - 1) { 2965 if (acr > 0 || acw > 0 || ace > 0) 2966 error = ENXIO; 2967 goto end; 2968 } 2969 if (dcw == 0 && !sc->sc_idle) 2970 g_raid3_idle(sc, dcw); 2971 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROYING) != 0) { 2972 if (acr > 0 || acw > 0 || ace > 0) { 2973 error = ENXIO; 2974 goto end; 2975 } 2976 if (dcr == 0 && dcw == 0 && dce == 0) { 2977 g_post_event(g_raid3_destroy_delayed, sc, M_WAITOK, 2978 sc, NULL); 2979 } 2980 } 2981end: 2982 sx_xunlock(&sc->sc_lock); 2983 g_topology_lock(); 2984 return (error); 2985} 2986 2987static struct g_geom * 2988g_raid3_create(struct g_class *mp, const struct g_raid3_metadata *md) 2989{ 2990 struct g_raid3_softc *sc; 2991 struct g_geom *gp; 2992 int error, timeout; 2993 u_int n; 2994 2995 g_topology_assert(); 2996 G_RAID3_DEBUG(1, "Creating device %s (id=%u).", md->md_name, md->md_id); 2997 2998 /* One disk is minimum. */ 2999 if (md->md_all < 1) 3000 return (NULL); 3001 /* 3002 * Action geom. 3003 */ 3004 gp = g_new_geomf(mp, "%s", md->md_name); 3005 sc = malloc(sizeof(*sc), M_RAID3, M_WAITOK | M_ZERO); 3006 sc->sc_disks = malloc(sizeof(struct g_raid3_disk) * md->md_all, M_RAID3, 3007 M_WAITOK | M_ZERO); 3008 gp->start = g_raid3_start; 3009 gp->orphan = g_raid3_orphan; 3010 gp->access = g_raid3_access; 3011 gp->dumpconf = g_raid3_dumpconf; 3012 3013 sc->sc_id = md->md_id; 3014 sc->sc_mediasize = md->md_mediasize; 3015 sc->sc_sectorsize = md->md_sectorsize; 3016 sc->sc_ndisks = md->md_all; 3017 sc->sc_round_robin = 0; 3018 sc->sc_flags = md->md_mflags; 3019 sc->sc_bump_id = 0; 3020 sc->sc_idle = 1; 3021 sc->sc_last_write = time_uptime; 3022 sc->sc_writes = 0; 3023 for (n = 0; n < sc->sc_ndisks; n++) { 3024 sc->sc_disks[n].d_softc = sc; 3025 sc->sc_disks[n].d_no = n; 3026 sc->sc_disks[n].d_state = G_RAID3_DISK_STATE_NODISK; 3027 } 3028 sx_init(&sc->sc_lock, "graid3:lock"); 3029 bioq_init(&sc->sc_queue); 3030 mtx_init(&sc->sc_queue_mtx, "graid3:queue", NULL, MTX_DEF); 3031 bioq_init(&sc->sc_regular_delayed); 3032 bioq_init(&sc->sc_inflight); 3033 bioq_init(&sc->sc_sync_delayed); 3034 TAILQ_INIT(&sc->sc_events); 3035 mtx_init(&sc->sc_events_mtx, "graid3:events", NULL, MTX_DEF); 3036 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 3037 sc->sc_state = G_RAID3_DEVICE_STATE_STARTING; 3038 gp->softc = sc; 3039 sc->sc_geom = gp; 3040 sc->sc_provider = NULL; 3041 /* 3042 * Synchronization geom. 3043 */ 3044 gp = g_new_geomf(mp, "%s.sync", md->md_name); 3045 gp->softc = sc; 3046 gp->orphan = g_raid3_orphan; 3047 sc->sc_sync.ds_geom = gp; 3048 3049 sc->sc_zones[G_RAID3_ZONE_64K].sz_zone = uma_zcreate("gr3:64k", 65536, 3050 g_raid3_uma_ctor, g_raid3_uma_dtor, NULL, NULL, UMA_ALIGN_PTR, 0); 3051 sc->sc_zones[G_RAID3_ZONE_64K].sz_inuse = 0; 3052 sc->sc_zones[G_RAID3_ZONE_64K].sz_max = g_raid3_n64k; 3053 sc->sc_zones[G_RAID3_ZONE_64K].sz_requested = 3054 sc->sc_zones[G_RAID3_ZONE_64K].sz_failed = 0; 3055 sc->sc_zones[G_RAID3_ZONE_16K].sz_zone = uma_zcreate("gr3:16k", 16384, 3056 g_raid3_uma_ctor, g_raid3_uma_dtor, NULL, NULL, UMA_ALIGN_PTR, 0); 3057 sc->sc_zones[G_RAID3_ZONE_16K].sz_inuse = 0; 3058 sc->sc_zones[G_RAID3_ZONE_16K].sz_max = g_raid3_n16k; 3059 sc->sc_zones[G_RAID3_ZONE_16K].sz_requested = 3060 sc->sc_zones[G_RAID3_ZONE_16K].sz_failed = 0; 3061 sc->sc_zones[G_RAID3_ZONE_4K].sz_zone = uma_zcreate("gr3:4k", 4096, 3062 g_raid3_uma_ctor, g_raid3_uma_dtor, NULL, NULL, UMA_ALIGN_PTR, 0); 3063 sc->sc_zones[G_RAID3_ZONE_4K].sz_inuse = 0; 3064 sc->sc_zones[G_RAID3_ZONE_4K].sz_max = g_raid3_n4k; 3065 sc->sc_zones[G_RAID3_ZONE_4K].sz_requested = 3066 sc->sc_zones[G_RAID3_ZONE_4K].sz_failed = 0; 3067 3068 error = kthread_create(g_raid3_worker, sc, &sc->sc_worker, 0, 0, 3069 "g_raid3 %s", md->md_name); 3070 if (error != 0) { 3071 G_RAID3_DEBUG(1, "Cannot create kernel thread for %s.", 3072 sc->sc_name); 3073 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_64K].sz_zone); 3074 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_16K].sz_zone); 3075 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_4K].sz_zone); 3076 g_destroy_geom(sc->sc_sync.ds_geom); 3077 mtx_destroy(&sc->sc_events_mtx); 3078 mtx_destroy(&sc->sc_queue_mtx); 3079 sx_destroy(&sc->sc_lock); 3080 g_destroy_geom(sc->sc_geom); 3081 free(sc->sc_disks, M_RAID3); 3082 free(sc, M_RAID3); 3083 return (NULL); 3084 } 3085 3086 G_RAID3_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 3087 3088 sc->sc_rootmount = root_mount_hold("GRAID3"); 3089 G_RAID3_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 3090 3091 /* 3092 * Run timeout. 3093 */ 3094 timeout = atomic_load_acq_int(&g_raid3_timeout); 3095 callout_reset(&sc->sc_callout, timeout * hz, g_raid3_go, sc); 3096 return (sc->sc_geom); 3097} 3098 3099int 3100g_raid3_destroy(struct g_raid3_softc *sc, int how) 3101{ 3102 struct g_provider *pp; 3103 3104 g_topology_assert_not(); 3105 if (sc == NULL) 3106 return (ENXIO); 3107 sx_assert(&sc->sc_lock, SX_XLOCKED); 3108 3109 pp = sc->sc_provider; 3110 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 3111 switch (how) { 3112 case G_RAID3_DESTROY_SOFT: 3113 G_RAID3_DEBUG(1, 3114 "Device %s is still open (r%dw%de%d).", pp->name, 3115 pp->acr, pp->acw, pp->ace); 3116 return (EBUSY); 3117 case G_RAID3_DESTROY_DELAYED: 3118 G_RAID3_DEBUG(1, 3119 "Device %s will be destroyed on last close.", 3120 pp->name); 3121 if (sc->sc_syncdisk != NULL) 3122 g_raid3_sync_stop(sc, 1); 3123 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROYING; 3124 return (EBUSY); 3125 case G_RAID3_DESTROY_HARD: 3126 G_RAID3_DEBUG(1, "Device %s is still open, so it " 3127 "can't be definitely removed.", pp->name); 3128 break; 3129 } 3130 } 3131 3132 g_topology_lock(); 3133 if (sc->sc_geom->softc == NULL) { 3134 g_topology_unlock(); 3135 return (0); 3136 } 3137 sc->sc_geom->softc = NULL; 3138 sc->sc_sync.ds_geom->softc = NULL; 3139 g_topology_unlock(); 3140 3141 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 3142 sc->sc_flags |= G_RAID3_DEVICE_FLAG_WAIT; 3143 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3144 sx_xunlock(&sc->sc_lock); 3145 mtx_lock(&sc->sc_queue_mtx); 3146 wakeup(sc); 3147 wakeup(&sc->sc_queue); 3148 mtx_unlock(&sc->sc_queue_mtx); 3149 G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3150 while (sc->sc_worker != NULL) 3151 tsleep(&sc->sc_worker, PRIBIO, "r3:destroy", hz / 5); 3152 G_RAID3_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3153 sx_xlock(&sc->sc_lock); 3154 g_raid3_destroy_device(sc); 3155 free(sc->sc_disks, M_RAID3); 3156 free(sc, M_RAID3); 3157 return (0); 3158} 3159 3160static void 3161g_raid3_taste_orphan(struct g_consumer *cp) 3162{ 3163 3164 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3165 cp->provider->name)); 3166} 3167 3168static struct g_geom * 3169g_raid3_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3170{ 3171 struct g_raid3_metadata md; 3172 struct g_raid3_softc *sc; 3173 struct g_consumer *cp; 3174 struct g_geom *gp; 3175 int error; 3176 3177 g_topology_assert(); 3178 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3179 G_RAID3_DEBUG(2, "Tasting %s.", pp->name); 3180 3181 gp = g_new_geomf(mp, "raid3:taste"); 3182 /* This orphan function should be never called. */ 3183 gp->orphan = g_raid3_taste_orphan; 3184 cp = g_new_consumer(gp); 3185 g_attach(cp, pp); 3186 error = g_raid3_read_metadata(cp, &md); 3187 g_detach(cp); 3188 g_destroy_consumer(cp); 3189 g_destroy_geom(gp); 3190 if (error != 0) 3191 return (NULL); 3192 gp = NULL; 3193 3194 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 3195 return (NULL); 3196 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3197 return (NULL); 3198 if (g_raid3_debug >= 2) 3199 raid3_metadata_dump(&md); 3200 3201 /* 3202 * Let's check if device already exists. 3203 */ 3204 sc = NULL; 3205 LIST_FOREACH(gp, &mp->geom, geom) { 3206 sc = gp->softc; 3207 if (sc == NULL) 3208 continue; 3209 if (sc->sc_sync.ds_geom == gp) 3210 continue; 3211 if (strcmp(md.md_name, sc->sc_name) != 0) 3212 continue; 3213 if (md.md_id != sc->sc_id) { 3214 G_RAID3_DEBUG(0, "Device %s already configured.", 3215 sc->sc_name); 3216 return (NULL); 3217 } 3218 break; 3219 } 3220 if (gp == NULL) { 3221 gp = g_raid3_create(mp, &md); 3222 if (gp == NULL) { 3223 G_RAID3_DEBUG(0, "Cannot create device %s.", 3224 md.md_name); 3225 return (NULL); 3226 } 3227 sc = gp->softc; 3228 } 3229 G_RAID3_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3230 g_topology_unlock(); 3231 sx_xlock(&sc->sc_lock); 3232 error = g_raid3_add_disk(sc, pp, &md); 3233 if (error != 0) { 3234 G_RAID3_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3235 pp->name, gp->name, error); 3236 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NODISK) == 3237 sc->sc_ndisks) { 3238 g_cancel_event(sc); 3239 g_raid3_destroy(sc, 1); 3240 g_topology_lock(); 3241 return (NULL); 3242 } 3243 gp = NULL; 3244 } 3245 sx_xunlock(&sc->sc_lock); 3246 g_topology_lock(); 3247 return (gp); 3248} 3249 3250static int 3251g_raid3_destroy_geom(struct gctl_req *req __unused, struct g_class *mp __unused, 3252 struct g_geom *gp) 3253{ 3254 struct g_raid3_softc *sc; 3255 int error; 3256 3257 g_topology_unlock(); 3258 sc = gp->softc; 3259 sx_xlock(&sc->sc_lock); 3260 g_cancel_event(sc); 3261 error = g_raid3_destroy(gp->softc, 0); 3262 if (error != 0) 3263 sx_xunlock(&sc->sc_lock); 3264 g_topology_lock(); 3265 return (error); 3266} 3267 3268static void 3269g_raid3_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3270 struct g_consumer *cp, struct g_provider *pp) 3271{ 3272 struct g_raid3_softc *sc; 3273 3274 g_topology_assert(); 3275 3276 sc = gp->softc; 3277 if (sc == NULL) 3278 return; 3279 /* Skip synchronization geom. */ 3280 if (gp == sc->sc_sync.ds_geom) 3281 return; 3282 if (pp != NULL) { 3283 /* Nothing here. */ 3284 } else if (cp != NULL) { 3285 struct g_raid3_disk *disk; 3286 3287 disk = cp->private; 3288 if (disk == NULL) 3289 return; 3290 g_topology_unlock(); 3291 sx_xlock(&sc->sc_lock); 3292 sbuf_printf(sb, "%s<Type>", indent); 3293 if (disk->d_no == sc->sc_ndisks - 1) 3294 sbuf_printf(sb, "PARITY"); 3295 else 3296 sbuf_printf(sb, "DATA"); 3297 sbuf_printf(sb, "</Type>\n"); 3298 sbuf_printf(sb, "%s<Number>%u</Number>\n", indent, 3299 (u_int)disk->d_no); 3300 if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 3301 sbuf_printf(sb, "%s<Synchronized>", indent); 3302 if (disk->d_sync.ds_offset == 0) 3303 sbuf_printf(sb, "0%%"); 3304 else { 3305 sbuf_printf(sb, "%u%%", 3306 (u_int)((disk->d_sync.ds_offset * 100) / 3307 (sc->sc_mediasize / (sc->sc_ndisks - 1)))); 3308 } 3309 sbuf_printf(sb, "</Synchronized>\n"); 3310 } 3311 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3312 disk->d_sync.ds_syncid); 3313 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, disk->d_genid); 3314 sbuf_printf(sb, "%s<Flags>", indent); 3315 if (disk->d_flags == 0) 3316 sbuf_printf(sb, "NONE"); 3317 else { 3318 int first = 1; 3319 3320#define ADD_FLAG(flag, name) do { \ 3321 if ((disk->d_flags & (flag)) != 0) { \ 3322 if (!first) \ 3323 sbuf_printf(sb, ", "); \ 3324 else \ 3325 first = 0; \ 3326 sbuf_printf(sb, name); \ 3327 } \ 3328} while (0) 3329 ADD_FLAG(G_RAID3_DISK_FLAG_DIRTY, "DIRTY"); 3330 ADD_FLAG(G_RAID3_DISK_FLAG_HARDCODED, "HARDCODED"); 3331 ADD_FLAG(G_RAID3_DISK_FLAG_SYNCHRONIZING, 3332 "SYNCHRONIZING"); 3333 ADD_FLAG(G_RAID3_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3334 ADD_FLAG(G_RAID3_DISK_FLAG_BROKEN, "BROKEN"); 3335#undef ADD_FLAG 3336 } 3337 sbuf_printf(sb, "</Flags>\n"); 3338 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3339 g_raid3_disk_state2str(disk->d_state)); 3340 sx_xunlock(&sc->sc_lock); 3341 g_topology_lock(); 3342 } else { 3343 g_topology_unlock(); 3344 sx_xlock(&sc->sc_lock); 3345 sbuf_printf(sb, "%s<Zone4kRequested>%u</Zone4kRequested>\n", 3346 indent, sc->sc_zones[G_RAID3_ZONE_4K].sz_requested); 3347 sbuf_printf(sb, "%s<Zone4kFailed>%u</Zone4kFailed>\n", 3348 indent, sc->sc_zones[G_RAID3_ZONE_4K].sz_failed); 3349 sbuf_printf(sb, "%s<Zone16kRequested>%u</Zone16kRequested>\n", 3350 indent, sc->sc_zones[G_RAID3_ZONE_16K].sz_requested); 3351 sbuf_printf(sb, "%s<Zone16kFailed>%u</Zone16kFailed>\n", 3352 indent, sc->sc_zones[G_RAID3_ZONE_16K].sz_failed); 3353 sbuf_printf(sb, "%s<Zone64kRequested>%u</Zone64kRequested>\n", 3354 indent, sc->sc_zones[G_RAID3_ZONE_64K].sz_requested); 3355 sbuf_printf(sb, "%s<Zone64kFailed>%u</Zone64kFailed>\n", 3356 indent, sc->sc_zones[G_RAID3_ZONE_64K].sz_failed); 3357 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3358 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3359 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3360 sbuf_printf(sb, "%s<Flags>", indent); 3361 if (sc->sc_flags == 0) 3362 sbuf_printf(sb, "NONE"); 3363 else { 3364 int first = 1; 3365 3366#define ADD_FLAG(flag, name) do { \ 3367 if ((sc->sc_flags & (flag)) != 0) { \ 3368 if (!first) \ 3369 sbuf_printf(sb, ", "); \ 3370 else \ 3371 first = 0; \ 3372 sbuf_printf(sb, name); \ 3373 } \ 3374} while (0) 3375 ADD_FLAG(G_RAID3_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3376 ADD_FLAG(G_RAID3_DEVICE_FLAG_ROUND_ROBIN, 3377 "ROUND-ROBIN"); 3378 ADD_FLAG(G_RAID3_DEVICE_FLAG_VERIFY, "VERIFY"); 3379#undef ADD_FLAG 3380 } 3381 sbuf_printf(sb, "</Flags>\n"); 3382 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3383 sc->sc_ndisks); 3384 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3385 g_raid3_device_state2str(sc->sc_state)); 3386 sx_xunlock(&sc->sc_lock); 3387 g_topology_lock(); 3388 } 3389} 3390 3391static void 3392g_raid3_shutdown_pre_sync(void *arg, int howto) 3393{ 3394 struct g_class *mp; 3395 struct g_geom *gp, *gp2; 3396 struct g_raid3_softc *sc; 3397 int error; 3398 3399 mp = arg; 3400 DROP_GIANT(); 3401 g_topology_lock(); 3402 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3403 if ((sc = gp->softc) == NULL) 3404 continue; 3405 /* Skip synchronization geom. */ 3406 if (gp == sc->sc_sync.ds_geom) 3407 continue; 3408 g_topology_unlock(); 3409 sx_xlock(&sc->sc_lock); 3410 g_cancel_event(sc); 3411 error = g_raid3_destroy(sc, G_RAID3_DESTROY_DELAYED); 3412 if (error != 0) 3413 sx_xunlock(&sc->sc_lock); 3414 g_topology_lock(); 3415 } 3416 g_topology_unlock(); 3417 PICKUP_GIANT(); 3418} 3419 3420static void 3421g_raid3_init(struct g_class *mp) 3422{ 3423 3424 g_raid3_pre_sync = EVENTHANDLER_REGISTER(shutdown_pre_sync, 3425 g_raid3_shutdown_pre_sync, mp, SHUTDOWN_PRI_FIRST); 3426 if (g_raid3_pre_sync == NULL) 3427 G_RAID3_DEBUG(0, "Warning! Cannot register shutdown event."); 3428} 3429 3430static void 3431g_raid3_fini(struct g_class *mp) 3432{ 3433 3434 if (g_raid3_pre_sync != NULL) 3435 EVENTHANDLER_DEREGISTER(shutdown_pre_sync, g_raid3_pre_sync); 3436} 3437 3438DECLARE_GEOM_CLASS(g_raid3_class, g_raid3); 3439