g_raid3.c revision 139451
1/*- 2 * Copyright (c) 2004 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 139451 2004-12-30 20:29:58Z jhb $"); 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 <machine/atomic.h> 43#include <geom/geom.h> 44#include <sys/proc.h> 45#include <sys/kthread.h> 46#include <sys/sched.h> 47#include <geom/raid3/g_raid3.h> 48 49 50static MALLOC_DEFINE(M_RAID3, "raid3 data", "GEOM_RAID3 Data"); 51 52SYSCTL_DECL(_kern_geom); 53SYSCTL_NODE(_kern_geom, OID_AUTO, raid3, CTLFLAG_RW, 0, "GEOM_RAID3 stuff"); 54u_int g_raid3_debug = 0; 55TUNABLE_INT("kern.geom.raid3.debug", &g_raid3_debug); 56SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, debug, CTLFLAG_RW, &g_raid3_debug, 0, 57 "Debug level"); 58static u_int g_raid3_timeout = 4; 59TUNABLE_INT("kern.geom.raid3.timeout", &g_raid3_timeout); 60SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, timeout, CTLFLAG_RW, &g_raid3_timeout, 61 0, "Time to wait on all raid3 components"); 62static u_int g_raid3_idletime = 5; 63TUNABLE_INT("kern.geom.raid3.idletime", &g_raid3_idletime); 64SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, idletime, CTLFLAG_RW, 65 &g_raid3_idletime, 0, "Mark components as clean when idling"); 66static u_int g_raid3_reqs_per_sync = 5; 67SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, reqs_per_sync, CTLFLAG_RW, 68 &g_raid3_reqs_per_sync, 0, 69 "Number of regular I/O requests per synchronization request"); 70static u_int g_raid3_syncs_per_sec = 100; 71SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, syncs_per_sec, CTLFLAG_RW, 72 &g_raid3_syncs_per_sec, 0, 73 "Number of synchronizations requests per second"); 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"); 93static u_int g_raid3_64k_requested = 0; 94SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, 64k_requested, CTLFLAG_RD, 95 &g_raid3_64k_requested, 0, "Number of requested 64kB allocations"); 96static u_int g_raid3_64k_failed = 0; 97SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, 64k_failed, CTLFLAG_RD, 98 &g_raid3_64k_failed, 0, "Number of failed 64kB allocations"); 99static u_int g_raid3_16k_requested = 0; 100SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, 16k_requested, CTLFLAG_RD, 101 &g_raid3_16k_requested, 0, "Number of requested 16kB allocations"); 102static u_int g_raid3_16k_failed = 0; 103SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, 16k_failed, CTLFLAG_RD, 104 &g_raid3_16k_failed, 0, "Number of failed 16kB allocations"); 105static u_int g_raid3_4k_requested = 0; 106SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, 4k_requested, CTLFLAG_RD, 107 &g_raid3_4k_requested, 0, "Number of requested 4kB allocations"); 108static u_int g_raid3_4k_failed = 0; 109SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, 4k_failed, CTLFLAG_RD, 110 &g_raid3_4k_failed, 0, "Number of failed 4kB allocations"); 111 112#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 113 G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 114 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 115 G_RAID3_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 116} while (0) 117 118static eventhandler_tag g_raid3_ehtag = NULL; 119 120static int g_raid3_destroy_geom(struct gctl_req *req, struct g_class *mp, 121 struct g_geom *gp); 122static g_taste_t g_raid3_taste; 123static void g_raid3_init(struct g_class *mp); 124static void g_raid3_fini(struct g_class *mp); 125 126struct g_class g_raid3_class = { 127 .name = G_RAID3_CLASS_NAME, 128 .version = G_VERSION, 129 .ctlreq = g_raid3_config, 130 .taste = g_raid3_taste, 131 .destroy_geom = g_raid3_destroy_geom, 132 .init = g_raid3_init, 133 .fini = g_raid3_fini 134}; 135 136 137static void g_raid3_destroy_provider(struct g_raid3_softc *sc); 138static int g_raid3_update_disk(struct g_raid3_disk *disk, u_int state); 139static void g_raid3_update_device(struct g_raid3_softc *sc, boolean_t force); 140static void g_raid3_dumpconf(struct sbuf *sb, const char *indent, 141 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 142static void g_raid3_sync_stop(struct g_raid3_softc *sc, int type); 143 144 145static const char * 146g_raid3_disk_state2str(int state) 147{ 148 149 switch (state) { 150 case G_RAID3_DISK_STATE_NODISK: 151 return ("NODISK"); 152 case G_RAID3_DISK_STATE_NONE: 153 return ("NONE"); 154 case G_RAID3_DISK_STATE_NEW: 155 return ("NEW"); 156 case G_RAID3_DISK_STATE_ACTIVE: 157 return ("ACTIVE"); 158 case G_RAID3_DISK_STATE_STALE: 159 return ("STALE"); 160 case G_RAID3_DISK_STATE_SYNCHRONIZING: 161 return ("SYNCHRONIZING"); 162 case G_RAID3_DISK_STATE_DISCONNECTED: 163 return ("DISCONNECTED"); 164 default: 165 return ("INVALID"); 166 } 167} 168 169static const char * 170g_raid3_device_state2str(int state) 171{ 172 173 switch (state) { 174 case G_RAID3_DEVICE_STATE_STARTING: 175 return ("STARTING"); 176 case G_RAID3_DEVICE_STATE_DEGRADED: 177 return ("DEGRADED"); 178 case G_RAID3_DEVICE_STATE_COMPLETE: 179 return ("COMPLETE"); 180 default: 181 return ("INVALID"); 182 } 183} 184 185const char * 186g_raid3_get_diskname(struct g_raid3_disk *disk) 187{ 188 189 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 190 return ("[unknown]"); 191 return (disk->d_name); 192} 193 194#define g_raid3_xor(src1, src2, dst, size) \ 195 _g_raid3_xor((uint64_t *)(src1), (uint64_t *)(src2), \ 196 (uint64_t *)(dst), (size_t)size) 197static void 198_g_raid3_xor(uint64_t *src1, uint64_t *src2, uint64_t *dst, size_t size) 199{ 200 201 KASSERT((size % 128) == 0, ("Invalid size: %zu.", size)); 202 for (; size > 0; size -= 128) { 203 *dst++ = (*src1++) ^ (*src2++); 204 *dst++ = (*src1++) ^ (*src2++); 205 *dst++ = (*src1++) ^ (*src2++); 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 } 220} 221 222static int 223g_raid3_is_zero(struct bio *bp) 224{ 225 static const uint64_t zeros[] = { 226 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 227 }; 228 u_char *addr; 229 ssize_t size; 230 231 size = bp->bio_length; 232 addr = (u_char *)bp->bio_data; 233 for (; size > 0; size -= sizeof(zeros), addr += sizeof(zeros)) { 234 if (bcmp(addr, zeros, sizeof(zeros)) != 0) 235 return (0); 236 } 237 return (1); 238} 239 240/* 241 * --- Events handling functions --- 242 * Events in geom_raid3 are used to maintain disks and device status 243 * from one thread to simplify locking. 244 */ 245static void 246g_raid3_event_free(struct g_raid3_event *ep) 247{ 248 249 free(ep, M_RAID3); 250} 251 252int 253g_raid3_event_send(void *arg, int state, int flags) 254{ 255 struct g_raid3_softc *sc; 256 struct g_raid3_disk *disk; 257 struct g_raid3_event *ep; 258 int error; 259 260 ep = malloc(sizeof(*ep), M_RAID3, M_WAITOK); 261 G_RAID3_DEBUG(4, "%s: Sending event %p.", __func__, ep); 262 if ((flags & G_RAID3_EVENT_DEVICE) != 0) { 263 disk = NULL; 264 sc = arg; 265 } else { 266 disk = arg; 267 sc = disk->d_softc; 268 } 269 ep->e_disk = disk; 270 ep->e_state = state; 271 ep->e_flags = flags; 272 ep->e_error = 0; 273 mtx_lock(&sc->sc_events_mtx); 274 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 275 mtx_unlock(&sc->sc_events_mtx); 276 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc); 277 mtx_lock(&sc->sc_queue_mtx); 278 wakeup(sc); 279 wakeup(&sc->sc_queue); 280 mtx_unlock(&sc->sc_queue_mtx); 281 if ((flags & G_RAID3_EVENT_DONTWAIT) != 0) 282 return (0); 283 g_topology_assert(); 284 G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 285 g_topology_unlock(); 286 while ((ep->e_flags & G_RAID3_EVENT_DONE) == 0) { 287 mtx_lock(&sc->sc_events_mtx); 288 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "r3:event", 289 hz * 5); 290 } 291 /* Don't even try to use 'sc' here, because it could be already dead. */ 292 g_topology_lock(); 293 error = ep->e_error; 294 g_raid3_event_free(ep); 295 return (error); 296} 297 298static struct g_raid3_event * 299g_raid3_event_get(struct g_raid3_softc *sc) 300{ 301 struct g_raid3_event *ep; 302 303 mtx_lock(&sc->sc_events_mtx); 304 ep = TAILQ_FIRST(&sc->sc_events); 305 mtx_unlock(&sc->sc_events_mtx); 306 return (ep); 307} 308 309static void 310g_raid3_event_remove(struct g_raid3_softc *sc, struct g_raid3_event *ep) 311{ 312 313 mtx_lock(&sc->sc_events_mtx); 314 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 315 mtx_unlock(&sc->sc_events_mtx); 316} 317 318static void 319g_raid3_event_cancel(struct g_raid3_disk *disk) 320{ 321 struct g_raid3_softc *sc; 322 struct g_raid3_event *ep, *tmpep; 323 324 g_topology_assert(); 325 326 sc = disk->d_softc; 327 mtx_lock(&sc->sc_events_mtx); 328 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 329 if ((ep->e_flags & G_RAID3_EVENT_DEVICE) != 0) 330 continue; 331 if (ep->e_disk != disk) 332 continue; 333 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 334 if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0) 335 g_raid3_event_free(ep); 336 else { 337 ep->e_error = ECANCELED; 338 wakeup(ep); 339 } 340 } 341 mtx_unlock(&sc->sc_events_mtx); 342} 343 344/* 345 * Return the number of disks in the given state. 346 * If state is equal to -1, count all connected disks. 347 */ 348u_int 349g_raid3_ndisks(struct g_raid3_softc *sc, int state) 350{ 351 struct g_raid3_disk *disk; 352 u_int n, ndisks; 353 354 for (n = ndisks = 0; n < sc->sc_ndisks; n++) { 355 disk = &sc->sc_disks[n]; 356 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 357 continue; 358 if (state == -1 || disk->d_state == state) 359 ndisks++; 360 } 361 return (ndisks); 362} 363 364static u_int 365g_raid3_nrequests(struct g_raid3_softc *sc, struct g_consumer *cp) 366{ 367 struct bio *bp; 368 u_int nreqs = 0; 369 370 mtx_lock(&sc->sc_queue_mtx); 371 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 372 if (bp->bio_from == cp) 373 nreqs++; 374 } 375 mtx_unlock(&sc->sc_queue_mtx); 376 return (nreqs); 377} 378 379static int 380g_raid3_is_busy(struct g_raid3_softc *sc, struct g_consumer *cp) 381{ 382 383 if (cp->index > 0) { 384 G_RAID3_DEBUG(2, 385 "I/O requests for %s exist, can't destroy it now.", 386 cp->provider->name); 387 return (1); 388 } 389 if (g_raid3_nrequests(sc, cp) > 0) { 390 G_RAID3_DEBUG(2, 391 "I/O requests for %s in queue, can't destroy it now.", 392 cp->provider->name); 393 return (1); 394 } 395 return (0); 396} 397 398static void 399g_raid3_destroy_consumer(void *arg, int flags __unused) 400{ 401 struct g_consumer *cp; 402 403 cp = arg; 404 G_RAID3_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 405 g_detach(cp); 406 g_destroy_consumer(cp); 407} 408 409static void 410g_raid3_kill_consumer(struct g_raid3_softc *sc, struct g_consumer *cp) 411{ 412 struct g_provider *pp; 413 int retaste_wait; 414 415 g_topology_assert(); 416 417 cp->private = NULL; 418 if (g_raid3_is_busy(sc, cp)) 419 return; 420 G_RAID3_DEBUG(2, "Consumer %s destroyed.", cp->provider->name); 421 pp = cp->provider; 422 retaste_wait = 0; 423 if (cp->acw == 1) { 424 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 425 retaste_wait = 1; 426 } 427 G_RAID3_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 428 -cp->acw, -cp->ace, 0); 429 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 430 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 431 if (retaste_wait) { 432 /* 433 * After retaste event was send (inside g_access()), we can send 434 * event to detach and destroy consumer. 435 * A class, which has consumer to the given provider connected 436 * will not receive retaste event for the provider. 437 * This is the way how I ignore retaste events when I close 438 * consumers opened for write: I detach and destroy consumer 439 * after retaste event is sent. 440 */ 441 g_post_event(g_raid3_destroy_consumer, cp, M_WAITOK, NULL); 442 return; 443 } 444 G_RAID3_DEBUG(1, "Consumer %s destroyed.", pp->name); 445 g_detach(cp); 446 g_destroy_consumer(cp); 447} 448 449static int 450g_raid3_connect_disk(struct g_raid3_disk *disk, struct g_provider *pp) 451{ 452 int error; 453 454 g_topology_assert(); 455 KASSERT(disk->d_consumer == NULL, 456 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 457 458 disk->d_consumer = g_new_consumer(disk->d_softc->sc_geom); 459 disk->d_consumer->private = disk; 460 disk->d_consumer->index = 0; 461 error = g_attach(disk->d_consumer, pp); 462 if (error != 0) 463 return (error); 464 error = g_access(disk->d_consumer, 1, 1, 1); 465 if (error != 0) { 466 G_RAID3_DEBUG(0, "Cannot open consumer %s (error=%d).", 467 pp->name, error); 468 return (error); 469 } 470 G_RAID3_DEBUG(2, "Disk %s connected.", g_raid3_get_diskname(disk)); 471 return (0); 472} 473 474static void 475g_raid3_disconnect_consumer(struct g_raid3_softc *sc, struct g_consumer *cp) 476{ 477 478 g_topology_assert(); 479 480 if (cp == NULL) 481 return; 482 if (cp->provider != NULL) 483 g_raid3_kill_consumer(sc, cp); 484 else 485 g_destroy_consumer(cp); 486} 487 488/* 489 * Initialize disk. This means allocate memory, create consumer, attach it 490 * to the provider and open access (r1w1e1) to it. 491 */ 492static struct g_raid3_disk * 493g_raid3_init_disk(struct g_raid3_softc *sc, struct g_provider *pp, 494 struct g_raid3_metadata *md, int *errorp) 495{ 496 struct g_raid3_disk *disk; 497 int error; 498 499 disk = &sc->sc_disks[md->md_no]; 500 error = g_raid3_connect_disk(disk, pp); 501 if (error != 0) 502 goto fail; 503 disk->d_state = G_RAID3_DISK_STATE_NONE; 504 disk->d_flags = md->md_dflags; 505 if (md->md_provider[0] != '\0') 506 disk->d_flags |= G_RAID3_DISK_FLAG_HARDCODED; 507 disk->d_sync.ds_consumer = NULL; 508 disk->d_sync.ds_offset = md->md_sync_offset; 509 disk->d_sync.ds_offset_done = md->md_sync_offset; 510 disk->d_sync.ds_resync = -1; 511 disk->d_genid = md->md_genid; 512 disk->d_sync.ds_syncid = md->md_syncid; 513 if (errorp != NULL) 514 *errorp = 0; 515 return (disk); 516fail: 517 if (errorp != NULL) 518 *errorp = error; 519 if (disk != NULL) 520 g_raid3_disconnect_consumer(sc, disk->d_consumer); 521 return (NULL); 522} 523 524static void 525g_raid3_destroy_disk(struct g_raid3_disk *disk) 526{ 527 struct g_raid3_softc *sc; 528 529 g_topology_assert(); 530 531 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 532 return; 533 g_raid3_event_cancel(disk); 534 sc = disk->d_softc; 535 switch (disk->d_state) { 536 case G_RAID3_DISK_STATE_SYNCHRONIZING: 537 if (sc->sc_syncdisk != NULL) 538 g_raid3_sync_stop(sc, 1); 539 /* FALLTHROUGH */ 540 case G_RAID3_DISK_STATE_NEW: 541 case G_RAID3_DISK_STATE_STALE: 542 case G_RAID3_DISK_STATE_ACTIVE: 543 g_raid3_disconnect_consumer(sc, disk->d_consumer); 544 disk->d_consumer = NULL; 545 break; 546 default: 547 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 548 g_raid3_get_diskname(disk), 549 g_raid3_disk_state2str(disk->d_state))); 550 } 551 disk->d_state = G_RAID3_DISK_STATE_NODISK; 552} 553 554static void 555g_raid3_destroy_device(struct g_raid3_softc *sc) 556{ 557 struct g_raid3_event *ep; 558 struct g_raid3_disk *disk; 559 struct g_geom *gp; 560 struct g_consumer *cp; 561 u_int n; 562 563 g_topology_assert(); 564 565 gp = sc->sc_geom; 566 if (sc->sc_provider != NULL) 567 g_raid3_destroy_provider(sc); 568 for (n = 0; n < sc->sc_ndisks; n++) { 569 disk = &sc->sc_disks[n]; 570 if (disk->d_state != G_RAID3_DISK_STATE_NODISK) { 571 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 572 g_raid3_update_metadata(disk); 573 g_raid3_destroy_disk(disk); 574 } 575 } 576 while ((ep = g_raid3_event_get(sc)) != NULL) { 577 g_raid3_event_remove(sc, ep); 578 if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0) 579 g_raid3_event_free(ep); 580 else { 581 ep->e_error = ECANCELED; 582 ep->e_flags |= G_RAID3_EVENT_DONE; 583 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, ep); 584 mtx_lock(&sc->sc_events_mtx); 585 wakeup(ep); 586 mtx_unlock(&sc->sc_events_mtx); 587 } 588 } 589 callout_drain(&sc->sc_callout); 590 gp->softc = NULL; 591 cp = LIST_FIRST(&sc->sc_sync.ds_geom->consumer); 592 if (cp != NULL) 593 g_raid3_disconnect_consumer(sc, cp); 594 sc->sc_sync.ds_geom->softc = NULL; 595 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 596 uma_zdestroy(sc->sc_zone_64k); 597 uma_zdestroy(sc->sc_zone_16k); 598 uma_zdestroy(sc->sc_zone_4k); 599 mtx_destroy(&sc->sc_queue_mtx); 600 mtx_destroy(&sc->sc_events_mtx); 601 G_RAID3_DEBUG(0, "Device %s destroyed.", gp->name); 602 g_wither_geom(gp, ENXIO); 603} 604 605static void 606g_raid3_orphan(struct g_consumer *cp) 607{ 608 struct g_raid3_disk *disk; 609 610 g_topology_assert(); 611 612 disk = cp->private; 613 if (disk == NULL) 614 return; 615 disk->d_softc->sc_bump_id = G_RAID3_BUMP_SYNCID_OFW; 616 g_raid3_event_send(disk, G_RAID3_DISK_STATE_DISCONNECTED, 617 G_RAID3_EVENT_DONTWAIT); 618} 619 620static void 621g_raid3_spoiled(struct g_consumer *cp) 622{ 623 struct g_raid3_disk *disk; 624 625 g_topology_assert(); 626 627 disk = cp->private; 628 if (disk == NULL) 629 return; 630 disk->d_softc->sc_bump_id = G_RAID3_BUMP_SYNCID_IMM; 631 g_raid3_event_send(disk, G_RAID3_DISK_STATE_DISCONNECTED, 632 G_RAID3_EVENT_DONTWAIT); 633} 634 635static int 636g_raid3_write_metadata(struct g_raid3_disk *disk, struct g_raid3_metadata *md) 637{ 638 struct g_raid3_softc *sc; 639 struct g_consumer *cp; 640 off_t offset, length; 641 u_char *sector; 642 int error = 0; 643 644 g_topology_assert(); 645 646 sc = disk->d_softc; 647 cp = disk->d_consumer; 648 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 649 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 650 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 651 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 652 cp->acw, cp->ace)); 653 length = cp->provider->sectorsize; 654 offset = cp->provider->mediasize - length; 655 sector = malloc((size_t)length, M_RAID3, M_WAITOK | M_ZERO); 656 if (md != NULL) 657 raid3_metadata_encode(md, sector); 658 g_topology_unlock(); 659 error = g_write_data(cp, offset, sector, length); 660 g_topology_lock(); 661 free(sector, M_RAID3); 662 if (error != 0) { 663 disk->d_softc->sc_bump_id = G_RAID3_BUMP_GENID_IMM; 664 g_raid3_event_send(disk, G_RAID3_DISK_STATE_DISCONNECTED, 665 G_RAID3_EVENT_DONTWAIT); 666 } 667 return (error); 668} 669 670int 671g_raid3_clear_metadata(struct g_raid3_disk *disk) 672{ 673 int error; 674 675 g_topology_assert(); 676 error = g_raid3_write_metadata(disk, NULL); 677 if (error == 0) { 678 G_RAID3_DEBUG(2, "Metadata on %s cleared.", 679 g_raid3_get_diskname(disk)); 680 } else { 681 G_RAID3_DEBUG(0, 682 "Cannot clear metadata on disk %s (error=%d).", 683 g_raid3_get_diskname(disk), error); 684 } 685 return (error); 686} 687 688void 689g_raid3_fill_metadata(struct g_raid3_disk *disk, struct g_raid3_metadata *md) 690{ 691 struct g_raid3_softc *sc; 692 693 sc = disk->d_softc; 694 strlcpy(md->md_magic, G_RAID3_MAGIC, sizeof(md->md_magic)); 695 md->md_version = G_RAID3_VERSION; 696 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 697 md->md_id = sc->sc_id; 698 md->md_all = sc->sc_ndisks; 699 md->md_genid = sc->sc_genid; 700 md->md_mediasize = sc->sc_mediasize; 701 md->md_sectorsize = sc->sc_sectorsize; 702 md->md_mflags = (sc->sc_flags & G_RAID3_DEVICE_FLAG_MASK); 703 md->md_no = disk->d_no; 704 md->md_syncid = disk->d_sync.ds_syncid; 705 md->md_dflags = (disk->d_flags & G_RAID3_DISK_FLAG_MASK); 706 if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) 707 md->md_sync_offset = disk->d_sync.ds_offset_done; 708 else 709 md->md_sync_offset = 0; 710 if ((disk->d_flags & G_RAID3_DISK_FLAG_HARDCODED) != 0 && 711 disk->d_consumer != NULL && disk->d_consumer->provider != NULL) { 712 strlcpy(md->md_provider, disk->d_consumer->provider->name, 713 sizeof(md->md_provider)); 714 } else { 715 bzero(md->md_provider, sizeof(md->md_provider)); 716 } 717} 718 719void 720g_raid3_update_metadata(struct g_raid3_disk *disk) 721{ 722 struct g_raid3_metadata md; 723 int error; 724 725 g_topology_assert(); 726 g_raid3_fill_metadata(disk, &md); 727 error = g_raid3_write_metadata(disk, &md); 728 if (error == 0) { 729 G_RAID3_DEBUG(2, "Metadata on %s updated.", 730 g_raid3_get_diskname(disk)); 731 } else { 732 G_RAID3_DEBUG(0, 733 "Cannot update metadata on disk %s (error=%d).", 734 g_raid3_get_diskname(disk), error); 735 } 736} 737 738static void 739g_raid3_bump_syncid(struct g_raid3_softc *sc) 740{ 741 struct g_raid3_disk *disk; 742 u_int n; 743 744 g_topology_assert(); 745 KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) > 0, 746 ("%s called with no active disks (device=%s).", __func__, 747 sc->sc_name)); 748 749 sc->sc_syncid++; 750 G_RAID3_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 751 sc->sc_syncid); 752 for (n = 0; n < sc->sc_ndisks; n++) { 753 disk = &sc->sc_disks[n]; 754 if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 755 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 756 disk->d_sync.ds_syncid = sc->sc_syncid; 757 g_raid3_update_metadata(disk); 758 } 759 } 760} 761 762static void 763g_raid3_bump_genid(struct g_raid3_softc *sc) 764{ 765 struct g_raid3_disk *disk; 766 u_int n; 767 768 g_topology_assert(); 769 KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) > 0, 770 ("%s called with no active disks (device=%s).", __func__, 771 sc->sc_name)); 772 773 sc->sc_genid++; 774 G_RAID3_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 775 sc->sc_genid); 776 for (n = 0; n < sc->sc_ndisks; n++) { 777 disk = &sc->sc_disks[n]; 778 if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 779 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 780 disk->d_genid = sc->sc_genid; 781 g_raid3_update_metadata(disk); 782 } 783 } 784} 785 786static void 787g_raid3_idle(struct g_raid3_softc *sc) 788{ 789 struct g_raid3_disk *disk; 790 u_int i; 791 792 if (sc->sc_provider == NULL || sc->sc_provider->acw == 0) 793 return; 794 sc->sc_idle = 1; 795 g_topology_lock(); 796 for (i = 0; i < sc->sc_ndisks; i++) { 797 disk = &sc->sc_disks[i]; 798 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) 799 continue; 800 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as clean.", 801 g_raid3_get_diskname(disk), sc->sc_name); 802 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 803 g_raid3_update_metadata(disk); 804 } 805 g_topology_unlock(); 806} 807 808static void 809g_raid3_unidle(struct g_raid3_softc *sc) 810{ 811 struct g_raid3_disk *disk; 812 u_int i; 813 814 sc->sc_idle = 0; 815 g_topology_lock(); 816 for (i = 0; i < sc->sc_ndisks; i++) { 817 disk = &sc->sc_disks[i]; 818 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) 819 continue; 820 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as dirty.", 821 g_raid3_get_diskname(disk), sc->sc_name); 822 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY; 823 g_raid3_update_metadata(disk); 824 } 825 g_topology_unlock(); 826} 827 828/* 829 * Return 1 if we should check if RAID3 device is idling. 830 */ 831static int 832g_raid3_check_idle(struct g_raid3_softc *sc) 833{ 834 struct g_raid3_disk *disk; 835 u_int i; 836 837 if (sc->sc_idle) 838 return (0); 839 if (sc->sc_provider != NULL && sc->sc_provider->acw == 0) 840 return (0); 841 /* 842 * Check if there are no in-flight requests. 843 */ 844 for (i = 0; i < sc->sc_ndisks; i++) { 845 disk = &sc->sc_disks[i]; 846 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) 847 continue; 848 if (disk->d_consumer->index > 0) 849 return (0); 850 } 851 return (1); 852} 853 854/* 855 * Treat bio_driver1 field in parent bio as list head and field bio_caller1 856 * in child bio as pointer to the next element on the list. 857 */ 858#define G_RAID3_HEAD_BIO(pbp) (pbp)->bio_driver1 859 860#define G_RAID3_NEXT_BIO(cbp) (cbp)->bio_caller1 861 862#define G_RAID3_FOREACH_BIO(pbp, bp) \ 863 for ((bp) = G_RAID3_HEAD_BIO(pbp); (bp) != NULL; \ 864 (bp) = G_RAID3_NEXT_BIO(bp)) 865 866#define G_RAID3_FOREACH_SAFE_BIO(pbp, bp, tmpbp) \ 867 for ((bp) = G_RAID3_HEAD_BIO(pbp); \ 868 (bp) != NULL && ((tmpbp) = G_RAID3_NEXT_BIO(bp), 1); \ 869 (bp) = (tmpbp)) 870 871static void 872g_raid3_init_bio(struct bio *pbp) 873{ 874 875 G_RAID3_HEAD_BIO(pbp) = NULL; 876} 877 878static void 879g_raid3_remove_bio(struct bio *cbp) 880{ 881 struct bio *pbp, *bp; 882 883 pbp = cbp->bio_parent; 884 if (G_RAID3_HEAD_BIO(pbp) == cbp) 885 G_RAID3_HEAD_BIO(pbp) = G_RAID3_NEXT_BIO(cbp); 886 else { 887 G_RAID3_FOREACH_BIO(pbp, bp) { 888 if (G_RAID3_NEXT_BIO(bp) == cbp) { 889 G_RAID3_NEXT_BIO(bp) = G_RAID3_NEXT_BIO(cbp); 890 break; 891 } 892 } 893 } 894 G_RAID3_NEXT_BIO(cbp) = NULL; 895} 896 897static void 898g_raid3_replace_bio(struct bio *sbp, struct bio *dbp) 899{ 900 struct bio *pbp, *bp; 901 902 g_raid3_remove_bio(sbp); 903 pbp = dbp->bio_parent; 904 G_RAID3_NEXT_BIO(sbp) = G_RAID3_NEXT_BIO(dbp); 905 if (G_RAID3_HEAD_BIO(pbp) == dbp) 906 G_RAID3_HEAD_BIO(pbp) = sbp; 907 else { 908 G_RAID3_FOREACH_BIO(pbp, bp) { 909 if (G_RAID3_NEXT_BIO(bp) == dbp) { 910 G_RAID3_NEXT_BIO(bp) = sbp; 911 break; 912 } 913 } 914 } 915 G_RAID3_NEXT_BIO(dbp) = NULL; 916} 917 918static void 919g_raid3_destroy_bio(struct g_raid3_softc *sc, struct bio *cbp) 920{ 921 struct bio *bp, *pbp; 922 size_t size; 923 924 pbp = cbp->bio_parent; 925 pbp->bio_children--; 926 KASSERT(cbp->bio_data != NULL, ("NULL bio_data")); 927 size = pbp->bio_length / (sc->sc_ndisks - 1); 928 if (size > 16384) 929 uma_zfree(sc->sc_zone_64k, cbp->bio_data); 930 else if (size > 4096) 931 uma_zfree(sc->sc_zone_16k, cbp->bio_data); 932 else 933 uma_zfree(sc->sc_zone_4k, cbp->bio_data); 934 if (G_RAID3_HEAD_BIO(pbp) == cbp) { 935 G_RAID3_HEAD_BIO(pbp) = G_RAID3_NEXT_BIO(cbp); 936 G_RAID3_NEXT_BIO(cbp) = NULL; 937 g_destroy_bio(cbp); 938 } else { 939 G_RAID3_FOREACH_BIO(pbp, bp) { 940 if (G_RAID3_NEXT_BIO(bp) == cbp) 941 break; 942 } 943 if (bp != NULL) { 944 KASSERT(G_RAID3_NEXT_BIO(bp) != NULL, 945 ("NULL bp->bio_driver1")); 946 G_RAID3_NEXT_BIO(bp) = G_RAID3_NEXT_BIO(cbp); 947 G_RAID3_NEXT_BIO(cbp) = NULL; 948 } 949 g_destroy_bio(cbp); 950 } 951} 952 953static struct bio * 954g_raid3_clone_bio(struct g_raid3_softc *sc, struct bio *pbp) 955{ 956 struct bio *bp, *cbp; 957 size_t size; 958 959 cbp = g_clone_bio(pbp); 960 if (cbp == NULL) 961 return (NULL); 962 size = pbp->bio_length / (sc->sc_ndisks - 1); 963 if (size > 16384) { 964 cbp->bio_data = uma_zalloc(sc->sc_zone_64k, M_NOWAIT); 965 g_raid3_64k_requested++; 966 } else if (size > 4096) { 967 cbp->bio_data = uma_zalloc(sc->sc_zone_16k, M_NOWAIT); 968 g_raid3_16k_requested++; 969 } else { 970 cbp->bio_data = uma_zalloc(sc->sc_zone_4k, M_NOWAIT); 971 g_raid3_4k_requested++; 972 } 973 if (cbp->bio_data == NULL) { 974 if (size > 16384) 975 g_raid3_64k_failed++; 976 if (size > 4096) 977 g_raid3_16k_failed++; 978 else 979 g_raid3_4k_failed++; 980 pbp->bio_children--; 981 g_destroy_bio(cbp); 982 return (NULL); 983 } 984 G_RAID3_NEXT_BIO(cbp) = NULL; 985 if (G_RAID3_HEAD_BIO(pbp) == NULL) 986 G_RAID3_HEAD_BIO(pbp) = cbp; 987 else { 988 G_RAID3_FOREACH_BIO(pbp, bp) { 989 if (G_RAID3_NEXT_BIO(bp) == NULL) { 990 G_RAID3_NEXT_BIO(bp) = cbp; 991 break; 992 } 993 } 994 } 995 return (cbp); 996} 997 998static void 999g_raid3_scatter(struct bio *pbp) 1000{ 1001 struct g_raid3_softc *sc; 1002 struct g_raid3_disk *disk; 1003 struct bio *bp, *cbp; 1004 off_t atom, cadd, padd, left; 1005 1006 sc = pbp->bio_to->geom->softc; 1007 bp = NULL; 1008 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_NOPARITY) == 0) { 1009 /* 1010 * Find bio for which we should calculate data. 1011 */ 1012 G_RAID3_FOREACH_BIO(pbp, cbp) { 1013 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) { 1014 bp = cbp; 1015 break; 1016 } 1017 } 1018 KASSERT(bp != NULL, ("NULL parity bio.")); 1019 } 1020 atom = sc->sc_sectorsize / (sc->sc_ndisks - 1); 1021 cadd = padd = 0; 1022 for (left = pbp->bio_length; left > 0; left -= sc->sc_sectorsize) { 1023 G_RAID3_FOREACH_BIO(pbp, cbp) { 1024 if (cbp == bp) 1025 continue; 1026 bcopy(pbp->bio_data + padd, cbp->bio_data + cadd, atom); 1027 padd += atom; 1028 } 1029 cadd += atom; 1030 } 1031 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_NOPARITY) == 0) { 1032 struct bio *tmpbp; 1033 1034 /* 1035 * Calculate parity. 1036 */ 1037 bzero(bp->bio_data, bp->bio_length); 1038 G_RAID3_FOREACH_SAFE_BIO(pbp, cbp, tmpbp) { 1039 if (cbp == bp) 1040 continue; 1041 g_raid3_xor(cbp->bio_data, bp->bio_data, bp->bio_data, 1042 bp->bio_length); 1043 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_NODISK) != 0) 1044 g_raid3_destroy_bio(sc, cbp); 1045 } 1046 } 1047 G_RAID3_FOREACH_BIO(pbp, cbp) { 1048 struct g_consumer *cp; 1049 1050 disk = cbp->bio_caller2; 1051 cp = disk->d_consumer; 1052 cbp->bio_to = cp->provider; 1053 G_RAID3_LOGREQ(3, cbp, "Sending request."); 1054 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1055 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1056 cp->acr, cp->acw, cp->ace)); 1057 cp->index++; 1058 g_io_request(cbp, cp); 1059 } 1060} 1061 1062static void 1063g_raid3_gather(struct bio *pbp) 1064{ 1065 struct g_raid3_softc *sc; 1066 struct g_raid3_disk *disk; 1067 struct bio *xbp, *fbp, *cbp; 1068 off_t atom, cadd, padd, left; 1069 1070 sc = pbp->bio_to->geom->softc; 1071 /* 1072 * Find bio for which we have to calculate data. 1073 * While going through this path, check if all requests 1074 * succeeded, if not, deny whole request. 1075 * If we're in COMPLETE mode, we allow one request to fail, 1076 * so if we find one, we're sending it to the parity consumer. 1077 * If there are more failed requests, we deny whole request. 1078 */ 1079 xbp = fbp = NULL; 1080 G_RAID3_FOREACH_BIO(pbp, cbp) { 1081 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) { 1082 KASSERT(xbp == NULL, ("More than one parity bio.")); 1083 xbp = cbp; 1084 } 1085 if (cbp->bio_error == 0) 1086 continue; 1087 /* 1088 * Found failed request. 1089 */ 1090 G_RAID3_LOGREQ(0, cbp, "Request failed."); 1091 disk = cbp->bio_caller2; 1092 if (disk != NULL) { 1093 /* 1094 * Actually this is pointless to bump genid, 1095 * because whole device is fucked up. 1096 */ 1097 sc->sc_bump_id |= G_RAID3_BUMP_GENID_IMM; 1098 g_raid3_event_send(disk, 1099 G_RAID3_DISK_STATE_DISCONNECTED, 1100 G_RAID3_EVENT_DONTWAIT); 1101 } 1102 if (fbp == NULL) { 1103 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_DEGRADED) != 0) { 1104 /* 1105 * We are already in degraded mode, so we can't 1106 * accept any failures. 1107 */ 1108 if (pbp->bio_error == 0) 1109 pbp->bio_error = fbp->bio_error; 1110 } else { 1111 fbp = cbp; 1112 } 1113 } else { 1114 /* 1115 * Next failed request, that's too many. 1116 */ 1117 if (pbp->bio_error == 0) 1118 pbp->bio_error = fbp->bio_error; 1119 } 1120 } 1121 if (pbp->bio_error != 0) 1122 goto finish; 1123 if (fbp != NULL && (pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) { 1124 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_VERIFY; 1125 if (xbp != fbp) 1126 g_raid3_replace_bio(xbp, fbp); 1127 g_raid3_destroy_bio(sc, fbp); 1128 } else if (fbp != NULL) { 1129 struct g_consumer *cp; 1130 1131 /* 1132 * One request failed, so send the same request to 1133 * the parity consumer. 1134 */ 1135 disk = pbp->bio_driver2; 1136 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) { 1137 pbp->bio_error = fbp->bio_error; 1138 goto finish; 1139 } 1140 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED; 1141 pbp->bio_inbed--; 1142 fbp->bio_flags &= ~(BIO_DONE | BIO_ERROR); 1143 if (disk->d_no == sc->sc_ndisks - 1) 1144 fbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1145 fbp->bio_error = 0; 1146 fbp->bio_completed = 0; 1147 fbp->bio_children = 0; 1148 fbp->bio_inbed = 0; 1149 cp = disk->d_consumer; 1150 fbp->bio_caller2 = disk; 1151 fbp->bio_to = cp->provider; 1152 G_RAID3_LOGREQ(3, fbp, "Sending request (recover)."); 1153 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1154 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1155 cp->acr, cp->acw, cp->ace)); 1156 cp->index++; 1157 g_io_request(fbp, cp); 1158 return; 1159 } 1160 if (xbp != NULL) { 1161 /* 1162 * Calculate parity. 1163 */ 1164 G_RAID3_FOREACH_BIO(pbp, cbp) { 1165 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) 1166 continue; 1167 g_raid3_xor(cbp->bio_data, xbp->bio_data, xbp->bio_data, 1168 xbp->bio_length); 1169 } 1170 xbp->bio_cflags &= ~G_RAID3_BIO_CFLAG_PARITY; 1171 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) { 1172 if (!g_raid3_is_zero(xbp)) { 1173 g_raid3_parity_mismatch++; 1174 pbp->bio_error = EIO; 1175 goto finish; 1176 } 1177 g_raid3_destroy_bio(sc, xbp); 1178 } 1179 } 1180 atom = sc->sc_sectorsize / (sc->sc_ndisks - 1); 1181 cadd = padd = 0; 1182 for (left = pbp->bio_length; left > 0; left -= sc->sc_sectorsize) { 1183 G_RAID3_FOREACH_BIO(pbp, cbp) { 1184 bcopy(cbp->bio_data + cadd, pbp->bio_data + padd, atom); 1185 pbp->bio_completed += atom; 1186 padd += atom; 1187 } 1188 cadd += atom; 1189 } 1190finish: 1191 if (pbp->bio_error == 0) 1192 G_RAID3_LOGREQ(3, pbp, "Request finished."); 1193 else { 1194 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) 1195 G_RAID3_LOGREQ(1, pbp, "Verification error."); 1196 else 1197 G_RAID3_LOGREQ(0, pbp, "Request failed."); 1198 } 1199 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_MASK; 1200 g_io_deliver(pbp, pbp->bio_error); 1201 while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL) 1202 g_raid3_destroy_bio(sc, cbp); 1203} 1204 1205static void 1206g_raid3_done(struct bio *bp) 1207{ 1208 struct g_raid3_softc *sc; 1209 1210 sc = bp->bio_from->geom->softc; 1211 bp->bio_cflags |= G_RAID3_BIO_CFLAG_REGULAR; 1212 G_RAID3_LOGREQ(3, bp, "Regular request done (error=%d).", bp->bio_error); 1213 mtx_lock(&sc->sc_queue_mtx); 1214 bioq_insert_head(&sc->sc_queue, bp); 1215 wakeup(sc); 1216 wakeup(&sc->sc_queue); 1217 mtx_unlock(&sc->sc_queue_mtx); 1218} 1219 1220static void 1221g_raid3_regular_request(struct bio *cbp) 1222{ 1223 struct g_raid3_softc *sc; 1224 struct g_raid3_disk *disk; 1225 struct bio *pbp; 1226 1227 g_topology_assert_not(); 1228 1229 cbp->bio_from->index--; 1230 pbp = cbp->bio_parent; 1231 sc = pbp->bio_to->geom->softc; 1232 disk = cbp->bio_from->private; 1233 if (disk == NULL) { 1234 g_topology_lock(); 1235 g_raid3_kill_consumer(sc, cbp->bio_from); 1236 g_topology_unlock(); 1237 } 1238 1239 G_RAID3_LOGREQ(3, cbp, "Request finished."); 1240 pbp->bio_inbed++; 1241 KASSERT(pbp->bio_inbed <= pbp->bio_children, 1242 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 1243 pbp->bio_children)); 1244 if (pbp->bio_inbed != pbp->bio_children) 1245 return; 1246 switch (pbp->bio_cmd) { 1247 case BIO_READ: 1248 g_raid3_gather(pbp); 1249 break; 1250 case BIO_WRITE: 1251 case BIO_DELETE: 1252 { 1253 int error = 0; 1254 1255 pbp->bio_completed = pbp->bio_length; 1256 while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL) { 1257 if (cbp->bio_error != 0) { 1258 disk = cbp->bio_caller2; 1259 if (disk != NULL) { 1260 sc->sc_bump_id |= 1261 G_RAID3_BUMP_GENID_IMM; 1262 g_raid3_event_send(disk, 1263 G_RAID3_DISK_STATE_DISCONNECTED, 1264 G_RAID3_EVENT_DONTWAIT); 1265 } 1266 if (error == 0) 1267 error = cbp->bio_error; 1268 else if (pbp->bio_error == 0) { 1269 /* 1270 * Next failed request, that's too many. 1271 */ 1272 pbp->bio_error = error; 1273 } 1274 } 1275 g_raid3_destroy_bio(sc, cbp); 1276 } 1277 if (pbp->bio_error == 0) 1278 G_RAID3_LOGREQ(3, pbp, "Request finished."); 1279 else 1280 G_RAID3_LOGREQ(0, pbp, "Request failed."); 1281 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_DEGRADED; 1282 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_NOPARITY; 1283 g_io_deliver(pbp, pbp->bio_error); 1284 break; 1285 } 1286 } 1287} 1288 1289static void 1290g_raid3_sync_done(struct bio *bp) 1291{ 1292 struct g_raid3_softc *sc; 1293 1294 G_RAID3_LOGREQ(3, bp, "Synchronization request delivered."); 1295 sc = bp->bio_from->geom->softc; 1296 bp->bio_cflags |= G_RAID3_BIO_CFLAG_SYNC; 1297 mtx_lock(&sc->sc_queue_mtx); 1298 bioq_insert_head(&sc->sc_queue, bp); 1299 wakeup(sc); 1300 wakeup(&sc->sc_queue); 1301 mtx_unlock(&sc->sc_queue_mtx); 1302} 1303 1304static void 1305g_raid3_start(struct bio *bp) 1306{ 1307 struct g_raid3_softc *sc; 1308 1309 sc = bp->bio_to->geom->softc; 1310 /* 1311 * If sc == NULL or there are no valid disks, provider's error 1312 * should be set and g_raid3_start() should not be called at all. 1313 */ 1314 KASSERT(sc != NULL && (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 1315 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE), 1316 ("Provider's error should be set (error=%d)(device=%s).", 1317 bp->bio_to->error, bp->bio_to->name)); 1318 G_RAID3_LOGREQ(3, bp, "Request received."); 1319 1320 switch (bp->bio_cmd) { 1321 case BIO_READ: 1322 case BIO_WRITE: 1323 case BIO_DELETE: 1324 break; 1325 case BIO_GETATTR: 1326 default: 1327 g_io_deliver(bp, EOPNOTSUPP); 1328 return; 1329 } 1330 mtx_lock(&sc->sc_queue_mtx); 1331 bioq_insert_tail(&sc->sc_queue, bp); 1332 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1333 wakeup(sc); 1334 mtx_unlock(&sc->sc_queue_mtx); 1335} 1336 1337/* 1338 * Send one synchronization request. 1339 */ 1340static void 1341g_raid3_sync_one(struct g_raid3_softc *sc) 1342{ 1343 struct g_raid3_disk *disk; 1344 struct bio *bp; 1345 1346 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED, 1347 ("Wrong device state (%s, %s).", sc->sc_name, 1348 g_raid3_device_state2str(sc->sc_state))); 1349 disk = sc->sc_syncdisk; 1350 KASSERT(disk != NULL, ("No sync disk (%s).", sc->sc_name)); 1351 KASSERT(disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING, 1352 ("Disk %s is not marked for synchronization.", 1353 g_raid3_get_diskname(disk))); 1354 1355 bp = g_new_bio(); 1356 if (bp == NULL) 1357 return; 1358 bp->bio_parent = NULL; 1359 bp->bio_cmd = BIO_READ; 1360 bp->bio_offset = disk->d_sync.ds_offset * (sc->sc_ndisks - 1); 1361 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1362 bp->bio_cflags = 0; 1363 bp->bio_done = g_raid3_sync_done; 1364 bp->bio_data = disk->d_sync.ds_data; 1365 if (bp->bio_data == NULL) { 1366 g_destroy_bio(bp); 1367 return; 1368 } 1369 bp->bio_cflags = G_RAID3_BIO_CFLAG_REGSYNC; 1370 disk->d_sync.ds_offset += bp->bio_length / (sc->sc_ndisks - 1); 1371 bp->bio_to = sc->sc_provider; 1372 G_RAID3_LOGREQ(3, bp, "Sending synchronization request."); 1373 disk->d_sync.ds_consumer->index++; 1374 g_io_request(bp, disk->d_sync.ds_consumer); 1375} 1376 1377static void 1378g_raid3_sync_request(struct bio *bp) 1379{ 1380 struct g_raid3_softc *sc; 1381 struct g_raid3_disk *disk; 1382 1383 bp->bio_from->index--; 1384 sc = bp->bio_from->geom->softc; 1385 disk = bp->bio_from->private; 1386 if (disk == NULL) { 1387 g_topology_lock(); 1388 g_raid3_kill_consumer(sc, bp->bio_from); 1389 g_topology_unlock(); 1390 g_destroy_bio(bp); 1391 return; 1392 } 1393 1394 /* 1395 * Synchronization request. 1396 */ 1397 switch (bp->bio_cmd) { 1398 case BIO_READ: 1399 { 1400 struct g_consumer *cp; 1401 u_char *dst, *src; 1402 off_t left; 1403 u_int atom; 1404 1405 if (bp->bio_error != 0) { 1406 G_RAID3_LOGREQ(0, bp, 1407 "Synchronization request failed (error=%d).", 1408 bp->bio_error); 1409 g_destroy_bio(bp); 1410 return; 1411 } 1412 G_RAID3_LOGREQ(3, bp, "Synchronization request finished."); 1413 atom = sc->sc_sectorsize / (sc->sc_ndisks - 1); 1414 dst = src = bp->bio_data; 1415 if (disk->d_no == sc->sc_ndisks - 1) { 1416 u_int n; 1417 1418 /* Parity component. */ 1419 for (left = bp->bio_length; left > 0; 1420 left -= sc->sc_sectorsize) { 1421 bcopy(src, dst, atom); 1422 src += atom; 1423 for (n = 1; n < sc->sc_ndisks - 1; n++) { 1424 g_raid3_xor(src, dst, dst, atom); 1425 src += atom; 1426 } 1427 dst += atom; 1428 } 1429 } else { 1430 /* Regular component. */ 1431 src += atom * disk->d_no; 1432 for (left = bp->bio_length; left > 0; 1433 left -= sc->sc_sectorsize) { 1434 bcopy(src, dst, atom); 1435 src += sc->sc_sectorsize; 1436 dst += atom; 1437 } 1438 } 1439 bp->bio_offset /= sc->sc_ndisks - 1; 1440 bp->bio_length /= sc->sc_ndisks - 1; 1441 bp->bio_cmd = BIO_WRITE; 1442 bp->bio_cflags = 0; 1443 bp->bio_children = bp->bio_inbed = 0; 1444 cp = disk->d_consumer; 1445 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1446 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1447 cp->acr, cp->acw, cp->ace)); 1448 cp->index++; 1449 g_io_request(bp, cp); 1450 return; 1451 } 1452 case BIO_WRITE: 1453 { 1454 struct g_raid3_disk_sync *sync; 1455 1456 if (bp->bio_error != 0) { 1457 G_RAID3_LOGREQ(0, bp, 1458 "Synchronization request failed (error=%d).", 1459 bp->bio_error); 1460 g_destroy_bio(bp); 1461 sc->sc_bump_id |= G_RAID3_BUMP_GENID_IMM; 1462 g_raid3_event_send(disk, 1463 G_RAID3_DISK_STATE_DISCONNECTED, 1464 G_RAID3_EVENT_DONTWAIT); 1465 return; 1466 } 1467 G_RAID3_LOGREQ(3, bp, "Synchronization request finished."); 1468 sync = &disk->d_sync; 1469 sync->ds_offset_done = bp->bio_offset + bp->bio_length; 1470 g_destroy_bio(bp); 1471 if (sync->ds_resync != -1) 1472 return; 1473 if (sync->ds_offset_done == 1474 sc->sc_mediasize / (sc->sc_ndisks - 1)) { 1475 /* 1476 * Disk up-to-date, activate it. 1477 */ 1478 g_raid3_event_send(disk, G_RAID3_DISK_STATE_ACTIVE, 1479 G_RAID3_EVENT_DONTWAIT); 1480 return; 1481 } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) { 1482 /* 1483 * Update offset_done on every 100 blocks. 1484 * XXX: This should be configurable. 1485 */ 1486 g_topology_lock(); 1487 g_raid3_update_metadata(disk); 1488 g_topology_unlock(); 1489 } 1490 return; 1491 } 1492 default: 1493 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1494 bp->bio_cmd, sc->sc_name)); 1495 break; 1496 } 1497} 1498 1499static int 1500g_raid3_register_request(struct bio *pbp) 1501{ 1502 struct g_raid3_softc *sc; 1503 struct g_raid3_disk *disk; 1504 struct g_consumer *cp; 1505 struct bio *cbp; 1506 off_t offset, length; 1507 u_int n, ndisks; 1508 int round_robin, verify; 1509 1510 ndisks = 0; 1511 sc = pbp->bio_to->geom->softc; 1512 if ((pbp->bio_cflags & G_RAID3_BIO_CFLAG_REGSYNC) != 0 && 1513 sc->sc_syncdisk == NULL) { 1514 g_io_deliver(pbp, EIO); 1515 return (0); 1516 } 1517 g_raid3_init_bio(pbp); 1518 length = pbp->bio_length / (sc->sc_ndisks - 1); 1519 offset = pbp->bio_offset / (sc->sc_ndisks - 1); 1520 round_robin = verify = 0; 1521 switch (pbp->bio_cmd) { 1522 case BIO_READ: 1523 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_VERIFY) != 0 && 1524 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 1525 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_VERIFY; 1526 verify = 1; 1527 ndisks = sc->sc_ndisks; 1528 } else { 1529 verify = 0; 1530 ndisks = sc->sc_ndisks - 1; 1531 } 1532 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_ROUND_ROBIN) != 0 && 1533 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 1534 round_robin = 1; 1535 } else { 1536 round_robin = 0; 1537 } 1538 KASSERT(!round_robin || !verify, 1539 ("ROUND-ROBIN and VERIFY are mutually exclusive.")); 1540 pbp->bio_driver2 = &sc->sc_disks[sc->sc_ndisks - 1]; 1541 break; 1542 case BIO_WRITE: 1543 case BIO_DELETE: 1544 { 1545 struct g_raid3_disk_sync *sync; 1546 1547 if (sc->sc_idle) 1548 g_raid3_unidle(sc); 1549 1550 ndisks = sc->sc_ndisks; 1551 1552 if (sc->sc_syncdisk == NULL) 1553 break; 1554 sync = &sc->sc_syncdisk->d_sync; 1555 if (offset >= sync->ds_offset) 1556 break; 1557 if (offset + length <= sync->ds_offset_done) 1558 break; 1559 if (offset >= sync->ds_resync && sync->ds_resync != -1) 1560 break; 1561 sync->ds_resync = offset - (offset % MAXPHYS); 1562 break; 1563 } 1564 } 1565 for (n = 0; n < ndisks; n++) { 1566 disk = &sc->sc_disks[n]; 1567 cbp = g_raid3_clone_bio(sc, pbp); 1568 if (cbp == NULL) { 1569 while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL) 1570 g_raid3_destroy_bio(sc, cbp); 1571 return (ENOMEM); 1572 } 1573 cbp->bio_offset = offset; 1574 cbp->bio_length = length; 1575 cbp->bio_done = g_raid3_done; 1576 switch (pbp->bio_cmd) { 1577 case BIO_READ: 1578 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) { 1579 /* 1580 * Replace invalid component with the parity 1581 * component. 1582 */ 1583 disk = &sc->sc_disks[sc->sc_ndisks - 1]; 1584 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1585 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED; 1586 } else if (round_robin && 1587 disk->d_no == sc->sc_round_robin) { 1588 /* 1589 * In round-robin mode skip one data component 1590 * and use parity component when reading. 1591 */ 1592 pbp->bio_driver2 = disk; 1593 disk = &sc->sc_disks[sc->sc_ndisks - 1]; 1594 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1595 sc->sc_round_robin++; 1596 round_robin = 0; 1597 } else if (verify && disk->d_no == sc->sc_ndisks - 1) { 1598 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1599 } 1600 break; 1601 case BIO_WRITE: 1602 case BIO_DELETE: 1603 if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 1604 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 1605 if (n == ndisks - 1) { 1606 /* 1607 * Active parity component, mark it as such. 1608 */ 1609 cbp->bio_cflags |= 1610 G_RAID3_BIO_CFLAG_PARITY; 1611 } 1612 } else { 1613 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED; 1614 if (n == ndisks - 1) { 1615 /* 1616 * Parity component is not connected, 1617 * so destroy its request. 1618 */ 1619 pbp->bio_pflags |= 1620 G_RAID3_BIO_PFLAG_NOPARITY; 1621 g_raid3_destroy_bio(sc, cbp); 1622 cbp = NULL; 1623 } else { 1624 cbp->bio_cflags |= 1625 G_RAID3_BIO_CFLAG_NODISK; 1626 disk = NULL; 1627 } 1628 } 1629 break; 1630 } 1631 if (cbp != NULL) 1632 cbp->bio_caller2 = disk; 1633 } 1634 switch (pbp->bio_cmd) { 1635 case BIO_READ: 1636 if (round_robin) { 1637 /* 1638 * If we are in round-robin mode and 'round_robin' is 1639 * still 1, it means, that we skipped parity component 1640 * for this read and must reset sc_round_robin field. 1641 */ 1642 sc->sc_round_robin = 0; 1643 } 1644 G_RAID3_FOREACH_BIO(pbp, cbp) { 1645 disk = cbp->bio_caller2; 1646 cp = disk->d_consumer; 1647 cbp->bio_to = cp->provider; 1648 G_RAID3_LOGREQ(3, cbp, "Sending request."); 1649 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1650 ("Consumer %s not opened (r%dw%de%d).", 1651 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1652 cp->index++; 1653 g_io_request(cbp, cp); 1654 } 1655 break; 1656 case BIO_WRITE: 1657 case BIO_DELETE: 1658 /* 1659 * Bump syncid on first write. 1660 */ 1661 if ((sc->sc_bump_id & G_RAID3_BUMP_SYNCID_OFW) != 0) { 1662 sc->sc_bump_id &= ~G_RAID3_BUMP_SYNCID; 1663 g_topology_lock(); 1664 g_raid3_bump_syncid(sc); 1665 g_topology_unlock(); 1666 } 1667 g_raid3_scatter(pbp); 1668 break; 1669 } 1670 return (0); 1671} 1672 1673static int 1674g_raid3_can_destroy(struct g_raid3_softc *sc) 1675{ 1676 struct g_geom *gp; 1677 struct g_consumer *cp; 1678 1679 g_topology_assert(); 1680 gp = sc->sc_geom; 1681 LIST_FOREACH(cp, &gp->consumer, consumer) { 1682 if (g_raid3_is_busy(sc, cp)) 1683 return (0); 1684 } 1685 gp = sc->sc_sync.ds_geom; 1686 LIST_FOREACH(cp, &gp->consumer, consumer) { 1687 if (g_raid3_is_busy(sc, cp)) 1688 return (0); 1689 } 1690 G_RAID3_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1691 sc->sc_name); 1692 return (1); 1693} 1694 1695static int 1696g_raid3_try_destroy(struct g_raid3_softc *sc) 1697{ 1698 1699 g_topology_lock(); 1700 if (!g_raid3_can_destroy(sc)) { 1701 g_topology_unlock(); 1702 return (0); 1703 } 1704 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_WAIT) != 0) { 1705 g_topology_unlock(); 1706 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, 1707 &sc->sc_worker); 1708 wakeup(&sc->sc_worker); 1709 sc->sc_worker = NULL; 1710 } else { 1711 g_raid3_destroy_device(sc); 1712 g_topology_unlock(); 1713 free(sc->sc_disks, M_RAID3); 1714 free(sc, M_RAID3); 1715 } 1716 return (1); 1717} 1718 1719/* 1720 * Worker thread. 1721 */ 1722static void 1723g_raid3_worker(void *arg) 1724{ 1725 struct g_raid3_softc *sc; 1726 struct g_raid3_disk *disk; 1727 struct g_raid3_disk_sync *sync; 1728 struct g_raid3_event *ep; 1729 struct bio *bp; 1730 u_int nreqs; 1731 1732 sc = arg; 1733 mtx_lock_spin(&sched_lock); 1734 sched_prio(curthread, PRIBIO); 1735 mtx_unlock_spin(&sched_lock); 1736 1737 nreqs = 0; 1738 for (;;) { 1739 G_RAID3_DEBUG(5, "%s: Let's see...", __func__); 1740 /* 1741 * First take a look at events. 1742 * This is important to handle events before any I/O requests. 1743 */ 1744 ep = g_raid3_event_get(sc); 1745 if (ep != NULL && g_topology_try_lock()) { 1746 g_raid3_event_remove(sc, ep); 1747 if ((ep->e_flags & G_RAID3_EVENT_DEVICE) != 0) { 1748 /* Update only device status. */ 1749 G_RAID3_DEBUG(3, 1750 "Running event for device %s.", 1751 sc->sc_name); 1752 ep->e_error = 0; 1753 g_raid3_update_device(sc, 1); 1754 } else { 1755 /* Update disk status. */ 1756 G_RAID3_DEBUG(3, "Running event for disk %s.", 1757 g_raid3_get_diskname(ep->e_disk)); 1758 ep->e_error = g_raid3_update_disk(ep->e_disk, 1759 ep->e_state); 1760 if (ep->e_error == 0) 1761 g_raid3_update_device(sc, 0); 1762 } 1763 g_topology_unlock(); 1764 if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0) { 1765 KASSERT(ep->e_error == 0, 1766 ("Error cannot be handled.")); 1767 g_raid3_event_free(ep); 1768 } else { 1769 ep->e_flags |= G_RAID3_EVENT_DONE; 1770 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, 1771 ep); 1772 mtx_lock(&sc->sc_events_mtx); 1773 wakeup(ep); 1774 mtx_unlock(&sc->sc_events_mtx); 1775 } 1776 if ((sc->sc_flags & 1777 G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 1778 if (g_raid3_try_destroy(sc)) 1779 kthread_exit(0); 1780 } 1781 G_RAID3_DEBUG(5, "%s: I'm here 1.", __func__); 1782 continue; 1783 } 1784 /* 1785 * Now I/O requests. 1786 */ 1787 /* Get first request from the queue. */ 1788 mtx_lock(&sc->sc_queue_mtx); 1789 bp = bioq_first(&sc->sc_queue); 1790 if (bp == NULL) { 1791 if (ep != NULL) { 1792 /* 1793 * No I/O requests and topology lock was 1794 * already held? Try again. 1795 */ 1796 mtx_unlock(&sc->sc_queue_mtx); 1797 tsleep(ep, PRIBIO, "r3:top1", hz / 5); 1798 continue; 1799 } 1800 if ((sc->sc_flags & 1801 G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 1802 mtx_unlock(&sc->sc_queue_mtx); 1803 if (g_raid3_try_destroy(sc)) 1804 kthread_exit(0); 1805 mtx_lock(&sc->sc_queue_mtx); 1806 } 1807 } 1808 if (sc->sc_syncdisk != NULL && 1809 (bp == NULL || nreqs > g_raid3_reqs_per_sync)) { 1810 mtx_unlock(&sc->sc_queue_mtx); 1811 /* 1812 * It is time for synchronization... 1813 */ 1814 nreqs = 0; 1815 disk = sc->sc_syncdisk; 1816 sync = &disk->d_sync; 1817 if (sync->ds_offset < 1818 sc->sc_mediasize / (sc->sc_ndisks - 1) && 1819 sync->ds_offset == sync->ds_offset_done) { 1820 if (sync->ds_resync != -1) { 1821 sync->ds_offset = sync->ds_resync; 1822 sync->ds_offset_done = sync->ds_resync; 1823 sync->ds_resync = -1; 1824 } 1825 g_raid3_sync_one(sc); 1826 } 1827 G_RAID3_DEBUG(5, "%s: I'm here 2.", __func__); 1828 goto sleep; 1829 } 1830 if (bp == NULL) { 1831 if (g_raid3_check_idle(sc)) { 1832 u_int idletime; 1833 1834 idletime = g_raid3_idletime; 1835 if (idletime == 0) 1836 idletime = 1; 1837 idletime *= hz; 1838 if (msleep(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1839 "r3:w1", idletime) == EWOULDBLOCK) { 1840 G_RAID3_DEBUG(5, "%s: I'm here 3.", 1841 __func__); 1842 /* 1843 * No I/O requests in 'idletime' 1844 * seconds, so mark components as clean. 1845 */ 1846 g_raid3_idle(sc); 1847 } 1848 G_RAID3_DEBUG(5, "%s: I'm here 4.", __func__); 1849 } else { 1850 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1851 "r3:w2", 0); 1852 G_RAID3_DEBUG(5, "%s: I'm here 5.", __func__); 1853 } 1854 continue; 1855 } 1856 nreqs++; 1857 bioq_remove(&sc->sc_queue, bp); 1858 mtx_unlock(&sc->sc_queue_mtx); 1859 1860 if ((bp->bio_cflags & G_RAID3_BIO_CFLAG_REGULAR) != 0) { 1861 g_raid3_regular_request(bp); 1862 } else if ((bp->bio_cflags & G_RAID3_BIO_CFLAG_SYNC) != 0) { 1863 u_int timeout, sps; 1864 1865 g_raid3_sync_request(bp); 1866sleep: 1867 sps = atomic_load_acq_int(&g_raid3_syncs_per_sec); 1868 if (sps == 0) { 1869 G_RAID3_DEBUG(5, "%s: I'm here 6.", __func__); 1870 continue; 1871 } 1872 if (ep != NULL) { 1873 /* 1874 * We have some pending events, don't sleep now. 1875 */ 1876 G_RAID3_DEBUG(5, "%s: I'm here 7.", __func__); 1877 tsleep(ep, PRIBIO, "r3:top2", hz / 5); 1878 continue; 1879 } 1880 mtx_lock(&sc->sc_queue_mtx); 1881 if (bioq_first(&sc->sc_queue) != NULL) { 1882 mtx_unlock(&sc->sc_queue_mtx); 1883 G_RAID3_DEBUG(5, "%s: I'm here 8.", __func__); 1884 continue; 1885 } 1886 timeout = hz / sps; 1887 if (timeout == 0) 1888 timeout = 1; 1889 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "r3:w2", 1890 timeout); 1891 } else { 1892 if (g_raid3_register_request(bp) != 0) { 1893 mtx_lock(&sc->sc_queue_mtx); 1894 bioq_insert_tail(&sc->sc_queue, bp); 1895 MSLEEP(&sc->sc_queue, &sc->sc_queue_mtx, 1896 PRIBIO | PDROP, "r3:lowmem", hz / 10); 1897 } 1898 } 1899 G_RAID3_DEBUG(5, "%s: I'm here 9.", __func__); 1900 } 1901} 1902 1903/* 1904 * Open disk's consumer if needed. 1905 */ 1906static void 1907g_raid3_update_access(struct g_raid3_disk *disk) 1908{ 1909 struct g_provider *pp; 1910 1911 g_topology_assert(); 1912 1913 pp = disk->d_softc->sc_provider; 1914 if (pp == NULL) 1915 return; 1916 if (pp->acw > 0) { 1917 if ((disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) == 0) { 1918 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1919 g_raid3_get_diskname(disk), disk->d_softc->sc_name); 1920 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY; 1921 } 1922 } else if (pp->acw == 0) { 1923 if ((disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) != 0) { 1924 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as clean.", 1925 g_raid3_get_diskname(disk), disk->d_softc->sc_name); 1926 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 1927 } 1928 } 1929} 1930 1931static void 1932g_raid3_sync_start(struct g_raid3_softc *sc) 1933{ 1934 struct g_raid3_disk *disk; 1935 int error; 1936 u_int n; 1937 1938 g_topology_assert(); 1939 1940 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED, 1941 ("Device not in DEGRADED state (%s, %u).", sc->sc_name, 1942 sc->sc_state)); 1943 KASSERT(sc->sc_syncdisk == NULL, ("Syncdisk is not NULL (%s, %u).", 1944 sc->sc_name, sc->sc_state)); 1945 disk = NULL; 1946 for (n = 0; n < sc->sc_ndisks; n++) { 1947 if (sc->sc_disks[n].d_state != G_RAID3_DISK_STATE_SYNCHRONIZING) 1948 continue; 1949 disk = &sc->sc_disks[n]; 1950 break; 1951 } 1952 if (disk == NULL) 1953 return; 1954 1955 G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1956 g_raid3_get_diskname(disk)); 1957 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY; 1958 KASSERT(disk->d_sync.ds_consumer == NULL, 1959 ("Sync consumer already exists (device=%s, disk=%s).", 1960 sc->sc_name, g_raid3_get_diskname(disk))); 1961 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1962 disk->d_sync.ds_consumer->private = disk; 1963 disk->d_sync.ds_consumer->index = 0; 1964 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1965 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1966 disk->d_softc->sc_name, error)); 1967 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1968 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1969 disk->d_softc->sc_name, error)); 1970 disk->d_sync.ds_data = malloc(MAXPHYS, M_RAID3, M_WAITOK); 1971 sc->sc_syncdisk = disk; 1972} 1973 1974/* 1975 * Stop synchronization process. 1976 * type: 0 - synchronization finished 1977 * 1 - synchronization stopped 1978 */ 1979static void 1980g_raid3_sync_stop(struct g_raid3_softc *sc, int type) 1981{ 1982 struct g_raid3_disk *disk; 1983 1984 g_topology_assert(); 1985 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED, 1986 ("Device not in DEGRADED state (%s, %u).", sc->sc_name, 1987 sc->sc_state)); 1988 disk = sc->sc_syncdisk; 1989 sc->sc_syncdisk = NULL; 1990 KASSERT(disk != NULL, ("No disk was synchronized (%s).", sc->sc_name)); 1991 KASSERT(disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING, 1992 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 1993 g_raid3_disk_state2str(disk->d_state))); 1994 if (disk->d_sync.ds_consumer == NULL) 1995 return; 1996 1997 if (type == 0) { 1998 G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1999 disk->d_softc->sc_name, g_raid3_get_diskname(disk)); 2000 } else /* if (type == 1) */ { 2001 G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2002 disk->d_softc->sc_name, g_raid3_get_diskname(disk)); 2003 } 2004 g_raid3_kill_consumer(disk->d_softc, disk->d_sync.ds_consumer); 2005 free(disk->d_sync.ds_data, M_RAID3); 2006 disk->d_sync.ds_consumer = NULL; 2007 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2008} 2009 2010static void 2011g_raid3_launch_provider(struct g_raid3_softc *sc) 2012{ 2013 struct g_provider *pp; 2014 2015 g_topology_assert(); 2016 2017 pp = g_new_providerf(sc->sc_geom, "raid3/%s", sc->sc_name); 2018 pp->mediasize = sc->sc_mediasize; 2019 pp->sectorsize = sc->sc_sectorsize; 2020 sc->sc_provider = pp; 2021 g_error_provider(pp, 0); 2022 G_RAID3_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 2023 pp->name); 2024 if (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED) 2025 g_raid3_sync_start(sc); 2026} 2027 2028static void 2029g_raid3_destroy_provider(struct g_raid3_softc *sc) 2030{ 2031 struct bio *bp; 2032 2033 g_topology_assert(); 2034 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2035 sc->sc_name)); 2036 2037 g_error_provider(sc->sc_provider, ENXIO); 2038 mtx_lock(&sc->sc_queue_mtx); 2039 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 2040 bioq_remove(&sc->sc_queue, bp); 2041 g_io_deliver(bp, ENXIO); 2042 } 2043 mtx_unlock(&sc->sc_queue_mtx); 2044 G_RAID3_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 2045 sc->sc_provider->name); 2046 sc->sc_provider->flags |= G_PF_WITHER; 2047 g_orphan_provider(sc->sc_provider, ENXIO); 2048 sc->sc_provider = NULL; 2049 if (sc->sc_syncdisk != NULL) 2050 g_raid3_sync_stop(sc, 1); 2051} 2052 2053static void 2054g_raid3_go(void *arg) 2055{ 2056 struct g_raid3_softc *sc; 2057 2058 sc = arg; 2059 G_RAID3_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2060 g_raid3_event_send(sc, 0, 2061 G_RAID3_EVENT_DONTWAIT | G_RAID3_EVENT_DEVICE); 2062} 2063 2064static u_int 2065g_raid3_determine_state(struct g_raid3_disk *disk) 2066{ 2067 struct g_raid3_softc *sc; 2068 u_int state; 2069 2070 sc = disk->d_softc; 2071 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2072 if ((disk->d_flags & 2073 G_RAID3_DISK_FLAG_SYNCHRONIZING) == 0) { 2074 /* Disk does not need synchronization. */ 2075 state = G_RAID3_DISK_STATE_ACTIVE; 2076 } else { 2077 if ((sc->sc_flags & 2078 G_RAID3_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2079 (disk->d_flags & 2080 G_RAID3_DISK_FLAG_FORCE_SYNC) != 0) { 2081 /* 2082 * We can start synchronization from 2083 * the stored offset. 2084 */ 2085 state = G_RAID3_DISK_STATE_SYNCHRONIZING; 2086 } else { 2087 state = G_RAID3_DISK_STATE_STALE; 2088 } 2089 } 2090 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2091 /* 2092 * Reset all synchronization data for this disk, 2093 * because if it even was synchronized, it was 2094 * synchronized to disks with different syncid. 2095 */ 2096 disk->d_flags |= G_RAID3_DISK_FLAG_SYNCHRONIZING; 2097 disk->d_sync.ds_offset = 0; 2098 disk->d_sync.ds_offset_done = 0; 2099 disk->d_sync.ds_syncid = sc->sc_syncid; 2100 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2101 (disk->d_flags & G_RAID3_DISK_FLAG_FORCE_SYNC) != 0) { 2102 state = G_RAID3_DISK_STATE_SYNCHRONIZING; 2103 } else { 2104 state = G_RAID3_DISK_STATE_STALE; 2105 } 2106 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2107 /* 2108 * Not good, NOT GOOD! 2109 * It means that device was started on stale disks 2110 * and more fresh disk just arrive. 2111 * If there were writes, device is fucked up, sorry. 2112 * I think the best choice here is don't touch 2113 * this disk and inform the user laudly. 2114 */ 2115 G_RAID3_DEBUG(0, "Device %s was started before the freshest " 2116 "disk (%s) arrives!! It will not be connected to the " 2117 "running device.", sc->sc_name, 2118 g_raid3_get_diskname(disk)); 2119 g_raid3_destroy_disk(disk); 2120 state = G_RAID3_DISK_STATE_NONE; 2121 /* Return immediately, because disk was destroyed. */ 2122 return (state); 2123 } 2124 G_RAID3_DEBUG(3, "State for %s disk: %s.", 2125 g_raid3_get_diskname(disk), g_raid3_disk_state2str(state)); 2126 return (state); 2127} 2128 2129/* 2130 * Update device state. 2131 */ 2132static void 2133g_raid3_update_device(struct g_raid3_softc *sc, boolean_t force) 2134{ 2135 struct g_raid3_disk *disk; 2136 u_int state; 2137 2138 g_topology_assert(); 2139 2140 switch (sc->sc_state) { 2141 case G_RAID3_DEVICE_STATE_STARTING: 2142 { 2143 u_int n, ndirty, ndisks, genid, syncid; 2144 2145 KASSERT(sc->sc_provider == NULL, 2146 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2147 /* 2148 * Are we ready? We are, if all disks are connected or 2149 * one disk is missing and 'force' is true. 2150 */ 2151 if (g_raid3_ndisks(sc, -1) + force == sc->sc_ndisks) { 2152 if (!force) 2153 callout_drain(&sc->sc_callout); 2154 } else { 2155 if (force) { 2156 /* 2157 * Timeout expired, so destroy device. 2158 */ 2159 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 2160 } 2161 return; 2162 } 2163 2164 /* 2165 * Find the biggest genid. 2166 */ 2167 genid = 0; 2168 for (n = 0; n < sc->sc_ndisks; n++) { 2169 disk = &sc->sc_disks[n]; 2170 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2171 continue; 2172 if (disk->d_genid > genid) 2173 genid = disk->d_genid; 2174 } 2175 sc->sc_genid = genid; 2176 /* 2177 * Remove all disks without the biggest genid. 2178 */ 2179 for (n = 0; n < sc->sc_ndisks; n++) { 2180 disk = &sc->sc_disks[n]; 2181 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2182 continue; 2183 if (disk->d_genid < genid) { 2184 G_RAID3_DEBUG(0, 2185 "Component %s (device %s) broken, skipping.", 2186 g_raid3_get_diskname(disk), sc->sc_name); 2187 g_raid3_destroy_disk(disk); 2188 } 2189 } 2190 2191 /* 2192 * There must be at least 'sc->sc_ndisks - 1' components 2193 * with the same syncid and without SYNCHRONIZING flag. 2194 */ 2195 2196 /* 2197 * Find the biggest syncid, number of valid components and 2198 * number of dirty components. 2199 */ 2200 ndirty = ndisks = syncid = 0; 2201 for (n = 0; n < sc->sc_ndisks; n++) { 2202 disk = &sc->sc_disks[n]; 2203 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2204 continue; 2205 if ((disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) != 0) 2206 ndirty++; 2207 if (disk->d_sync.ds_syncid > syncid) { 2208 syncid = disk->d_sync.ds_syncid; 2209 ndisks = 0; 2210 } else if (disk->d_sync.ds_syncid < syncid) { 2211 continue; 2212 } 2213 if ((disk->d_flags & 2214 G_RAID3_DISK_FLAG_SYNCHRONIZING) != 0) { 2215 continue; 2216 } 2217 ndisks++; 2218 } 2219 /* 2220 * Do we have enough valid components? 2221 */ 2222 if (ndisks + 1 < sc->sc_ndisks) { 2223 G_RAID3_DEBUG(0, 2224 "Device %s is broken, too few valid components.", 2225 sc->sc_name); 2226 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 2227 return; 2228 } 2229 /* 2230 * If there is one DIRTY component and all disks are present, 2231 * mark it for synchronization. If there is more than one DIRTY 2232 * component, mark parity component for synchronization. 2233 */ 2234 if (ndisks == sc->sc_ndisks && ndirty == 1) { 2235 for (n = 0; n < sc->sc_ndisks; n++) { 2236 disk = &sc->sc_disks[n]; 2237 if ((disk->d_flags & 2238 G_RAID3_DISK_FLAG_DIRTY) == 0) { 2239 continue; 2240 } 2241 disk->d_flags |= 2242 G_RAID3_DISK_FLAG_SYNCHRONIZING; 2243 } 2244 } else if (ndisks == sc->sc_ndisks && ndirty > 1) { 2245 disk = &sc->sc_disks[sc->sc_ndisks - 1]; 2246 disk->d_flags |= G_RAID3_DISK_FLAG_SYNCHRONIZING; 2247 } 2248 2249 sc->sc_syncid = syncid; 2250 if (force) { 2251 /* Remember to bump syncid on first write. */ 2252 sc->sc_bump_id |= G_RAID3_BUMP_SYNCID_OFW; 2253 } 2254 if (ndisks == sc->sc_ndisks) 2255 state = G_RAID3_DEVICE_STATE_COMPLETE; 2256 else /* if (ndisks == sc->sc_ndisks - 1) */ 2257 state = G_RAID3_DEVICE_STATE_DEGRADED; 2258 G_RAID3_DEBUG(1, "Device %s state changed from %s to %s.", 2259 sc->sc_name, g_raid3_device_state2str(sc->sc_state), 2260 g_raid3_device_state2str(state)); 2261 sc->sc_state = state; 2262 for (n = 0; n < sc->sc_ndisks; n++) { 2263 disk = &sc->sc_disks[n]; 2264 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2265 continue; 2266 state = g_raid3_determine_state(disk); 2267 g_raid3_event_send(disk, state, G_RAID3_EVENT_DONTWAIT); 2268 if (state == G_RAID3_DISK_STATE_STALE) 2269 sc->sc_bump_id |= G_RAID3_BUMP_SYNCID_OFW; 2270 } 2271 break; 2272 } 2273 case G_RAID3_DEVICE_STATE_DEGRADED: 2274 /* 2275 * Bump syncid and/or genid here, if we need to do it 2276 * immediately. 2277 */ 2278 if ((sc->sc_bump_id & G_RAID3_BUMP_SYNCID_IMM) != 0) { 2279 sc->sc_bump_id &= ~G_RAID3_BUMP_SYNCID; 2280 g_raid3_bump_syncid(sc); 2281 } 2282 if ((sc->sc_bump_id & G_RAID3_BUMP_GENID_IMM) != 0) { 2283 sc->sc_bump_id &= ~G_RAID3_BUMP_GENID; 2284 g_raid3_bump_genid(sc); 2285 } 2286 2287 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NEW) > 0) 2288 return; 2289 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) < 2290 sc->sc_ndisks - 1) { 2291 if (sc->sc_provider != NULL) 2292 g_raid3_destroy_provider(sc); 2293 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 2294 return; 2295 } 2296 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) == 2297 sc->sc_ndisks) { 2298 state = G_RAID3_DEVICE_STATE_COMPLETE; 2299 G_RAID3_DEBUG(1, 2300 "Device %s state changed from %s to %s.", 2301 sc->sc_name, g_raid3_device_state2str(sc->sc_state), 2302 g_raid3_device_state2str(state)); 2303 sc->sc_state = state; 2304 } 2305 if (sc->sc_provider == NULL) 2306 g_raid3_launch_provider(sc); 2307 break; 2308 case G_RAID3_DEVICE_STATE_COMPLETE: 2309 /* 2310 * Bump syncid and/or genid here, if we need to do it 2311 * immediately. 2312 */ 2313 if ((sc->sc_bump_id & G_RAID3_BUMP_SYNCID_IMM) != 0) { 2314 sc->sc_bump_id &= ~G_RAID3_BUMP_SYNCID; 2315 g_raid3_bump_syncid(sc); 2316 } 2317 if ((sc->sc_bump_id & G_RAID3_BUMP_GENID_IMM) != 0) { 2318 sc->sc_bump_id &= ~G_RAID3_BUMP_GENID; 2319 g_raid3_bump_genid(sc); 2320 } 2321 2322 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NEW) > 0) 2323 return; 2324 KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) >= 2325 sc->sc_ndisks - 1, 2326 ("Too few ACTIVE components in COMPLETE state (device %s).", 2327 sc->sc_name)); 2328 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) == 2329 sc->sc_ndisks - 1) { 2330 state = G_RAID3_DEVICE_STATE_DEGRADED; 2331 G_RAID3_DEBUG(1, 2332 "Device %s state changed from %s to %s.", 2333 sc->sc_name, g_raid3_device_state2str(sc->sc_state), 2334 g_raid3_device_state2str(state)); 2335 sc->sc_state = state; 2336 } 2337 if (sc->sc_provider == NULL) 2338 g_raid3_launch_provider(sc); 2339 break; 2340 default: 2341 KASSERT(1 == 0, ("Wrong device state (%s, %s).", sc->sc_name, 2342 g_raid3_device_state2str(sc->sc_state))); 2343 break; 2344 } 2345} 2346 2347/* 2348 * Update disk state and device state if needed. 2349 */ 2350#define DISK_STATE_CHANGED() G_RAID3_DEBUG(1, \ 2351 "Disk %s state changed from %s to %s (device %s).", \ 2352 g_raid3_get_diskname(disk), \ 2353 g_raid3_disk_state2str(disk->d_state), \ 2354 g_raid3_disk_state2str(state), sc->sc_name) 2355static int 2356g_raid3_update_disk(struct g_raid3_disk *disk, u_int state) 2357{ 2358 struct g_raid3_softc *sc; 2359 2360 g_topology_assert(); 2361 2362 sc = disk->d_softc; 2363again: 2364 G_RAID3_DEBUG(3, "Changing disk %s state from %s to %s.", 2365 g_raid3_get_diskname(disk), g_raid3_disk_state2str(disk->d_state), 2366 g_raid3_disk_state2str(state)); 2367 switch (state) { 2368 case G_RAID3_DISK_STATE_NEW: 2369 /* 2370 * Possible scenarios: 2371 * 1. New disk arrive. 2372 */ 2373 /* Previous state should be NONE. */ 2374 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NONE, 2375 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2376 g_raid3_disk_state2str(disk->d_state))); 2377 DISK_STATE_CHANGED(); 2378 2379 disk->d_state = state; 2380 G_RAID3_DEBUG(0, "Device %s: provider %s detected.", 2381 sc->sc_name, g_raid3_get_diskname(disk)); 2382 if (sc->sc_state == G_RAID3_DEVICE_STATE_STARTING) 2383 break; 2384 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2385 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2386 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2387 g_raid3_device_state2str(sc->sc_state), 2388 g_raid3_get_diskname(disk), 2389 g_raid3_disk_state2str(disk->d_state))); 2390 state = g_raid3_determine_state(disk); 2391 if (state != G_RAID3_DISK_STATE_NONE) 2392 goto again; 2393 break; 2394 case G_RAID3_DISK_STATE_ACTIVE: 2395 /* 2396 * Possible scenarios: 2397 * 1. New disk does not need synchronization. 2398 * 2. Synchronization process finished successfully. 2399 */ 2400 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2401 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2402 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2403 g_raid3_device_state2str(sc->sc_state), 2404 g_raid3_get_diskname(disk), 2405 g_raid3_disk_state2str(disk->d_state))); 2406 /* Previous state should be NEW or SYNCHRONIZING. */ 2407 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW || 2408 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING, 2409 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2410 g_raid3_disk_state2str(disk->d_state))); 2411 DISK_STATE_CHANGED(); 2412 2413 if (disk->d_state == G_RAID3_DISK_STATE_NEW) 2414 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2415 else if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 2416 disk->d_flags &= ~G_RAID3_DISK_FLAG_SYNCHRONIZING; 2417 disk->d_flags &= ~G_RAID3_DISK_FLAG_FORCE_SYNC; 2418 g_raid3_sync_stop(sc, 0); 2419 } 2420 disk->d_state = state; 2421 disk->d_sync.ds_offset = 0; 2422 disk->d_sync.ds_offset_done = 0; 2423 g_raid3_update_access(disk); 2424 g_raid3_update_metadata(disk); 2425 G_RAID3_DEBUG(0, "Device %s: provider %s activated.", 2426 sc->sc_name, g_raid3_get_diskname(disk)); 2427 break; 2428 case G_RAID3_DISK_STATE_STALE: 2429 /* 2430 * Possible scenarios: 2431 * 1. Stale disk was connected. 2432 */ 2433 /* Previous state should be NEW. */ 2434 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW, 2435 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2436 g_raid3_disk_state2str(disk->d_state))); 2437 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2438 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2439 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2440 g_raid3_device_state2str(sc->sc_state), 2441 g_raid3_get_diskname(disk), 2442 g_raid3_disk_state2str(disk->d_state))); 2443 /* 2444 * STALE state is only possible if device is marked 2445 * NOAUTOSYNC. 2446 */ 2447 KASSERT((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOAUTOSYNC) != 0, 2448 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2449 g_raid3_device_state2str(sc->sc_state), 2450 g_raid3_get_diskname(disk), 2451 g_raid3_disk_state2str(disk->d_state))); 2452 DISK_STATE_CHANGED(); 2453 2454 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2455 disk->d_state = state; 2456 g_raid3_update_metadata(disk); 2457 G_RAID3_DEBUG(0, "Device %s: provider %s is stale.", 2458 sc->sc_name, g_raid3_get_diskname(disk)); 2459 break; 2460 case G_RAID3_DISK_STATE_SYNCHRONIZING: 2461 /* 2462 * Possible scenarios: 2463 * 1. Disk which needs synchronization was connected. 2464 */ 2465 /* Previous state should be NEW. */ 2466 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW, 2467 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2468 g_raid3_disk_state2str(disk->d_state))); 2469 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2470 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2471 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2472 g_raid3_device_state2str(sc->sc_state), 2473 g_raid3_get_diskname(disk), 2474 g_raid3_disk_state2str(disk->d_state))); 2475 DISK_STATE_CHANGED(); 2476 2477 if (disk->d_state == G_RAID3_DISK_STATE_NEW) 2478 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2479 disk->d_state = state; 2480 if (sc->sc_provider != NULL) { 2481 g_raid3_sync_start(sc); 2482 g_raid3_update_metadata(disk); 2483 } 2484 break; 2485 case G_RAID3_DISK_STATE_DISCONNECTED: 2486 /* 2487 * Possible scenarios: 2488 * 1. Device wasn't running yet, but disk disappear. 2489 * 2. Disk was active and disapppear. 2490 * 3. Disk disappear during synchronization process. 2491 */ 2492 if (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2493 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 2494 /* 2495 * Previous state should be ACTIVE, STALE or 2496 * SYNCHRONIZING. 2497 */ 2498 KASSERT(disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 2499 disk->d_state == G_RAID3_DISK_STATE_STALE || 2500 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING, 2501 ("Wrong disk state (%s, %s).", 2502 g_raid3_get_diskname(disk), 2503 g_raid3_disk_state2str(disk->d_state))); 2504 } else if (sc->sc_state == G_RAID3_DEVICE_STATE_STARTING) { 2505 /* Previous state should be NEW. */ 2506 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW, 2507 ("Wrong disk state (%s, %s).", 2508 g_raid3_get_diskname(disk), 2509 g_raid3_disk_state2str(disk->d_state))); 2510 /* 2511 * Reset bumping syncid if disk disappeared in STARTING 2512 * state. 2513 */ 2514 if ((sc->sc_bump_id & G_RAID3_BUMP_SYNCID_OFW) != 0) 2515 sc->sc_bump_id &= ~G_RAID3_BUMP_SYNCID; 2516#ifdef INVARIANTS 2517 } else { 2518 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2519 sc->sc_name, 2520 g_raid3_device_state2str(sc->sc_state), 2521 g_raid3_get_diskname(disk), 2522 g_raid3_disk_state2str(disk->d_state))); 2523#endif 2524 } 2525 DISK_STATE_CHANGED(); 2526 G_RAID3_DEBUG(0, "Device %s: provider %s disconnected.", 2527 sc->sc_name, g_raid3_get_diskname(disk)); 2528 2529 g_raid3_destroy_disk(disk); 2530 break; 2531 default: 2532 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2533 break; 2534 } 2535 return (0); 2536} 2537#undef DISK_STATE_CHANGED 2538 2539static int 2540g_raid3_read_metadata(struct g_consumer *cp, struct g_raid3_metadata *md) 2541{ 2542 struct g_provider *pp; 2543 u_char *buf; 2544 int error; 2545 2546 g_topology_assert(); 2547 2548 error = g_access(cp, 1, 0, 0); 2549 if (error != 0) 2550 return (error); 2551 pp = cp->provider; 2552 g_topology_unlock(); 2553 /* Metadata are stored on last sector. */ 2554 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2555 &error); 2556 g_topology_lock(); 2557 g_access(cp, -1, 0, 0); 2558 if (error != 0) { 2559 G_RAID3_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2560 cp->provider->name, error); 2561 if (buf != NULL) 2562 g_free(buf); 2563 return (error); 2564 } 2565 2566 /* Decode metadata. */ 2567 error = raid3_metadata_decode(buf, md); 2568 g_free(buf); 2569 if (strcmp(md->md_magic, G_RAID3_MAGIC) != 0) 2570 return (EINVAL); 2571 if (md->md_version > G_RAID3_VERSION) { 2572 G_RAID3_DEBUG(0, 2573 "Kernel module is too old to handle metadata from %s.", 2574 cp->provider->name); 2575 return (EINVAL); 2576 } 2577 if (error != 0) { 2578 G_RAID3_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2579 cp->provider->name); 2580 return (error); 2581 } 2582 2583 return (0); 2584} 2585 2586static int 2587g_raid3_check_metadata(struct g_raid3_softc *sc, struct g_provider *pp, 2588 struct g_raid3_metadata *md) 2589{ 2590 2591 if (md->md_no >= sc->sc_ndisks) { 2592 G_RAID3_DEBUG(1, "Invalid disk %s number (no=%u), skipping.", 2593 pp->name, md->md_no); 2594 return (EINVAL); 2595 } 2596 if (sc->sc_disks[md->md_no].d_state != G_RAID3_DISK_STATE_NODISK) { 2597 G_RAID3_DEBUG(1, "Disk %s (no=%u) already exists, skipping.", 2598 pp->name, md->md_no); 2599 return (EEXIST); 2600 } 2601 if (md->md_all != sc->sc_ndisks) { 2602 G_RAID3_DEBUG(1, 2603 "Invalid '%s' field on disk %s (device %s), skipping.", 2604 "md_all", pp->name, sc->sc_name); 2605 return (EINVAL); 2606 } 2607 if (md->md_mediasize != sc->sc_mediasize) { 2608 G_RAID3_DEBUG(1, 2609 "Invalid '%s' field on disk %s (device %s), skipping.", 2610 "md_mediasize", pp->name, sc->sc_name); 2611 return (EINVAL); 2612 } 2613 if ((md->md_mediasize % (sc->sc_ndisks - 1)) != 0) { 2614 G_RAID3_DEBUG(1, 2615 "Invalid '%s' field on disk %s (device %s), skipping.", 2616 "md_mediasize", pp->name, sc->sc_name); 2617 return (EINVAL); 2618 } 2619 if ((sc->sc_mediasize / (sc->sc_ndisks - 1)) > pp->mediasize) { 2620 G_RAID3_DEBUG(1, 2621 "Invalid size of disk %s (device %s), skipping.", pp->name, 2622 sc->sc_name); 2623 return (EINVAL); 2624 } 2625 if ((md->md_sectorsize / pp->sectorsize) < sc->sc_ndisks - 1) { 2626 G_RAID3_DEBUG(1, 2627 "Invalid '%s' field on disk %s (device %s), skipping.", 2628 "md_sectorsize", pp->name, sc->sc_name); 2629 return (EINVAL); 2630 } 2631 if (md->md_sectorsize != sc->sc_sectorsize) { 2632 G_RAID3_DEBUG(1, 2633 "Invalid '%s' field on disk %s (device %s), skipping.", 2634 "md_sectorsize", pp->name, sc->sc_name); 2635 return (EINVAL); 2636 } 2637 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2638 G_RAID3_DEBUG(1, 2639 "Invalid sector size of disk %s (device %s), skipping.", 2640 pp->name, sc->sc_name); 2641 return (EINVAL); 2642 } 2643 if ((md->md_mflags & ~G_RAID3_DEVICE_FLAG_MASK) != 0) { 2644 G_RAID3_DEBUG(1, 2645 "Invalid device flags on disk %s (device %s), skipping.", 2646 pp->name, sc->sc_name); 2647 return (EINVAL); 2648 } 2649 if ((md->md_mflags & G_RAID3_DEVICE_FLAG_VERIFY) != 0 && 2650 (md->md_mflags & G_RAID3_DEVICE_FLAG_ROUND_ROBIN) != 0) { 2651 /* 2652 * VERIFY and ROUND-ROBIN options are mutally exclusive. 2653 */ 2654 G_RAID3_DEBUG(1, "Both VERIFY and ROUND-ROBIN flags exist on " 2655 "disk %s (device %s), skipping.", pp->name, sc->sc_name); 2656 return (EINVAL); 2657 } 2658 if ((md->md_dflags & ~G_RAID3_DISK_FLAG_MASK) != 0) { 2659 G_RAID3_DEBUG(1, 2660 "Invalid disk flags on disk %s (device %s), skipping.", 2661 pp->name, sc->sc_name); 2662 return (EINVAL); 2663 } 2664 return (0); 2665} 2666 2667static int 2668g_raid3_add_disk(struct g_raid3_softc *sc, struct g_provider *pp, 2669 struct g_raid3_metadata *md) 2670{ 2671 struct g_raid3_disk *disk; 2672 int error; 2673 2674 g_topology_assert(); 2675 G_RAID3_DEBUG(2, "Adding disk %s.", pp->name); 2676 2677 error = g_raid3_check_metadata(sc, pp, md); 2678 if (error != 0) 2679 return (error); 2680 if (sc->sc_state != G_RAID3_DEVICE_STATE_STARTING && 2681 md->md_genid < sc->sc_genid) { 2682 G_RAID3_DEBUG(0, "Component %s (device %s) broken, skipping.", 2683 pp->name, sc->sc_name); 2684 return (EINVAL); 2685 } 2686 disk = g_raid3_init_disk(sc, pp, md, &error); 2687 if (disk == NULL) 2688 return (error); 2689 error = g_raid3_event_send(disk, G_RAID3_DISK_STATE_NEW, 2690 G_RAID3_EVENT_WAIT); 2691 if (error != 0) 2692 return (error); 2693 if (md->md_version < G_RAID3_VERSION) { 2694 G_RAID3_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2695 pp->name, md->md_version, G_RAID3_VERSION); 2696 g_raid3_update_metadata(disk); 2697 } 2698 return (0); 2699} 2700 2701static int 2702g_raid3_access(struct g_provider *pp, int acr, int acw, int ace) 2703{ 2704 struct g_raid3_softc *sc; 2705 struct g_raid3_disk *disk; 2706 int dcr, dcw, dce; 2707 u_int n; 2708 2709 g_topology_assert(); 2710 G_RAID3_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2711 acw, ace); 2712 2713 dcr = pp->acr + acr; 2714 dcw = pp->acw + acw; 2715 dce = pp->ace + ace; 2716 2717 sc = pp->geom->softc; 2718 if (sc == NULL || 2719 g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) < sc->sc_ndisks - 1 || 2720 (sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 2721 if (acr <= 0 && acw <= 0 && ace <= 0) 2722 return (0); 2723 else 2724 return (ENXIO); 2725 } 2726 for (n = 0; n < sc->sc_ndisks; n++) { 2727 disk = &sc->sc_disks[n]; 2728 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) 2729 continue; 2730 /* 2731 * Mark disk as dirty on open and unmark on close. 2732 */ 2733 if (pp->acw == 0 && dcw > 0) { 2734 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as dirty.", 2735 g_raid3_get_diskname(disk), sc->sc_name); 2736 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY; 2737 g_raid3_update_metadata(disk); 2738 } else if (pp->acw > 0 && dcw == 0) { 2739 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as clean.", 2740 g_raid3_get_diskname(disk), sc->sc_name); 2741 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2742 g_raid3_update_metadata(disk); 2743 } 2744 } 2745 return (0); 2746} 2747 2748static struct g_geom * 2749g_raid3_create(struct g_class *mp, const struct g_raid3_metadata *md) 2750{ 2751 struct g_raid3_softc *sc; 2752 struct g_geom *gp; 2753 int error, timeout; 2754 u_int n; 2755 2756 g_topology_assert(); 2757 G_RAID3_DEBUG(1, "Creating device %s (id=%u).", md->md_name, md->md_id); 2758 2759 /* One disk is minimum. */ 2760 if (md->md_all < 1) 2761 return (NULL); 2762 /* 2763 * Action geom. 2764 */ 2765 gp = g_new_geomf(mp, "%s", md->md_name); 2766 sc = malloc(sizeof(*sc), M_RAID3, M_WAITOK | M_ZERO); 2767 sc->sc_disks = malloc(sizeof(struct g_raid3_disk) * md->md_all, M_RAID3, 2768 M_WAITOK | M_ZERO); 2769 gp->start = g_raid3_start; 2770 gp->spoiled = g_raid3_spoiled; 2771 gp->orphan = g_raid3_orphan; 2772 gp->access = g_raid3_access; 2773 gp->dumpconf = g_raid3_dumpconf; 2774 2775 sc->sc_id = md->md_id; 2776 sc->sc_mediasize = md->md_mediasize; 2777 sc->sc_sectorsize = md->md_sectorsize; 2778 sc->sc_ndisks = md->md_all; 2779 sc->sc_round_robin = 0; 2780 sc->sc_flags = md->md_mflags; 2781 sc->sc_bump_id = 0; 2782 sc->sc_idle = 0; 2783 for (n = 0; n < sc->sc_ndisks; n++) { 2784 sc->sc_disks[n].d_softc = sc; 2785 sc->sc_disks[n].d_no = n; 2786 sc->sc_disks[n].d_state = G_RAID3_DISK_STATE_NODISK; 2787 } 2788 bioq_init(&sc->sc_queue); 2789 mtx_init(&sc->sc_queue_mtx, "graid3:queue", NULL, MTX_DEF); 2790 TAILQ_INIT(&sc->sc_events); 2791 mtx_init(&sc->sc_events_mtx, "graid3:events", NULL, MTX_DEF); 2792 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2793 sc->sc_state = G_RAID3_DEVICE_STATE_STARTING; 2794 gp->softc = sc; 2795 sc->sc_geom = gp; 2796 sc->sc_provider = NULL; 2797 /* 2798 * Synchronization geom. 2799 */ 2800 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2801 gp->softc = sc; 2802 gp->orphan = g_raid3_orphan; 2803 sc->sc_sync.ds_geom = gp; 2804 sc->sc_zone_64k = uma_zcreate("gr3:64k", 65536, NULL, NULL, NULL, NULL, 2805 UMA_ALIGN_PTR, 0); 2806 uma_zone_set_max(sc->sc_zone_64k, g_raid3_n64k); 2807 sc->sc_zone_16k = uma_zcreate("gr3:16k", 16384, NULL, NULL, NULL, NULL, 2808 UMA_ALIGN_PTR, 0); 2809 uma_zone_set_max(sc->sc_zone_64k, g_raid3_n16k); 2810 sc->sc_zone_4k = uma_zcreate("gr3:4k", 4096, NULL, NULL, NULL, NULL, 2811 UMA_ALIGN_PTR, 0); 2812 uma_zone_set_max(sc->sc_zone_4k, g_raid3_n4k); 2813 error = kthread_create(g_raid3_worker, sc, &sc->sc_worker, 0, 0, 2814 "g_raid3 %s", md->md_name); 2815 if (error != 0) { 2816 G_RAID3_DEBUG(1, "Cannot create kernel thread for %s.", 2817 sc->sc_name); 2818 uma_zdestroy(sc->sc_zone_64k); 2819 uma_zdestroy(sc->sc_zone_16k); 2820 uma_zdestroy(sc->sc_zone_4k); 2821 g_destroy_geom(sc->sc_sync.ds_geom); 2822 mtx_destroy(&sc->sc_events_mtx); 2823 mtx_destroy(&sc->sc_queue_mtx); 2824 g_destroy_geom(sc->sc_geom); 2825 free(sc->sc_disks, M_RAID3); 2826 free(sc, M_RAID3); 2827 return (NULL); 2828 } 2829 2830 G_RAID3_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2831 2832 /* 2833 * Run timeout. 2834 */ 2835 timeout = atomic_load_acq_int(&g_raid3_timeout); 2836 callout_reset(&sc->sc_callout, timeout * hz, g_raid3_go, sc); 2837 return (sc->sc_geom); 2838} 2839 2840int 2841g_raid3_destroy(struct g_raid3_softc *sc, boolean_t force) 2842{ 2843 struct g_provider *pp; 2844 2845 g_topology_assert(); 2846 2847 if (sc == NULL) 2848 return (ENXIO); 2849 pp = sc->sc_provider; 2850 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2851 if (force) { 2852 G_RAID3_DEBUG(1, "Device %s is still open, so it " 2853 "can't be definitely removed.", pp->name); 2854 } else { 2855 G_RAID3_DEBUG(1, 2856 "Device %s is still open (r%dw%de%d).", pp->name, 2857 pp->acr, pp->acw, pp->ace); 2858 return (EBUSY); 2859 } 2860 } 2861 2862 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 2863 sc->sc_flags |= G_RAID3_DEVICE_FLAG_WAIT; 2864 g_topology_unlock(); 2865 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2866 mtx_lock(&sc->sc_queue_mtx); 2867 wakeup(sc); 2868 wakeup(&sc->sc_queue); 2869 mtx_unlock(&sc->sc_queue_mtx); 2870 G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2871 while (sc->sc_worker != NULL) 2872 tsleep(&sc->sc_worker, PRIBIO, "r3:destroy", hz / 5); 2873 G_RAID3_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2874 g_topology_lock(); 2875 g_raid3_destroy_device(sc); 2876 free(sc->sc_disks, M_RAID3); 2877 free(sc, M_RAID3); 2878 return (0); 2879} 2880 2881static void 2882g_raid3_taste_orphan(struct g_consumer *cp) 2883{ 2884 2885 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2886 cp->provider->name)); 2887} 2888 2889static struct g_geom * 2890g_raid3_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2891{ 2892 struct g_raid3_metadata md; 2893 struct g_raid3_softc *sc; 2894 struct g_consumer *cp; 2895 struct g_geom *gp; 2896 int error; 2897 2898 g_topology_assert(); 2899 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2900 G_RAID3_DEBUG(2, "Tasting %s.", pp->name); 2901 2902 gp = g_new_geomf(mp, "raid3:taste"); 2903 /* This orphan function should be never called. */ 2904 gp->orphan = g_raid3_taste_orphan; 2905 cp = g_new_consumer(gp); 2906 g_attach(cp, pp); 2907 error = g_raid3_read_metadata(cp, &md); 2908 g_detach(cp); 2909 g_destroy_consumer(cp); 2910 g_destroy_geom(gp); 2911 if (error != 0) 2912 return (NULL); 2913 gp = NULL; 2914 2915 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2916 return (NULL); 2917 if (g_raid3_debug >= 2) 2918 raid3_metadata_dump(&md); 2919 2920 /* 2921 * Let's check if device already exists. 2922 */ 2923 sc = NULL; 2924 LIST_FOREACH(gp, &mp->geom, geom) { 2925 sc = gp->softc; 2926 if (sc == NULL) 2927 continue; 2928 if (sc->sc_sync.ds_geom == gp) 2929 continue; 2930 if (strcmp(md.md_name, sc->sc_name) != 0) 2931 continue; 2932 if (md.md_id != sc->sc_id) { 2933 G_RAID3_DEBUG(0, "Device %s already configured.", 2934 sc->sc_name); 2935 return (NULL); 2936 } 2937 break; 2938 } 2939 if (gp == NULL) { 2940 gp = g_raid3_create(mp, &md); 2941 if (gp == NULL) { 2942 G_RAID3_DEBUG(0, "Cannot create device %s.", 2943 md.md_name); 2944 return (NULL); 2945 } 2946 sc = gp->softc; 2947 } 2948 G_RAID3_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2949 error = g_raid3_add_disk(sc, pp, &md); 2950 if (error != 0) { 2951 G_RAID3_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2952 pp->name, gp->name, error); 2953 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NODISK) == 2954 sc->sc_ndisks) { 2955 g_raid3_destroy(sc, 1); 2956 } 2957 return (NULL); 2958 } 2959 return (gp); 2960} 2961 2962static int 2963g_raid3_destroy_geom(struct gctl_req *req __unused, struct g_class *mp __unused, 2964 struct g_geom *gp) 2965{ 2966 2967 return (g_raid3_destroy(gp->softc, 0)); 2968} 2969 2970static void 2971g_raid3_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2972 struct g_consumer *cp, struct g_provider *pp) 2973{ 2974 struct g_raid3_softc *sc; 2975 2976 g_topology_assert(); 2977 2978 sc = gp->softc; 2979 if (sc == NULL) 2980 return; 2981 /* Skip synchronization geom. */ 2982 if (gp == sc->sc_sync.ds_geom) 2983 return; 2984 if (pp != NULL) { 2985 /* Nothing here. */ 2986 } else if (cp != NULL) { 2987 struct g_raid3_disk *disk; 2988 2989 disk = cp->private; 2990 if (disk == NULL) 2991 return; 2992 sbuf_printf(sb, "%s<Type>", indent); 2993 if (disk->d_no == sc->sc_ndisks - 1) 2994 sbuf_printf(sb, "PARITY"); 2995 else 2996 sbuf_printf(sb, "DATA"); 2997 sbuf_printf(sb, "</Type>\n"); 2998 sbuf_printf(sb, "%s<Number>%u</Number>\n", indent, 2999 (u_int)disk->d_no); 3000 if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 3001 sbuf_printf(sb, "%s<Synchronized>", indent); 3002 if (disk->d_sync.ds_offset_done == 0) 3003 sbuf_printf(sb, "0%%"); 3004 else { 3005 sbuf_printf(sb, "%u%%", 3006 (u_int)((disk->d_sync.ds_offset_done * 100) / 3007 (sc->sc_mediasize / (sc->sc_ndisks - 1)))); 3008 } 3009 sbuf_printf(sb, "</Synchronized>\n"); 3010 } 3011 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3012 disk->d_sync.ds_syncid); 3013 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, disk->d_genid); 3014 sbuf_printf(sb, "%s<Flags>", indent); 3015 if (disk->d_flags == 0) 3016 sbuf_printf(sb, "NONE"); 3017 else { 3018 int first = 1; 3019 3020#define ADD_FLAG(flag, name) do { \ 3021 if ((disk->d_flags & (flag)) != 0) { \ 3022 if (!first) \ 3023 sbuf_printf(sb, ", "); \ 3024 else \ 3025 first = 0; \ 3026 sbuf_printf(sb, name); \ 3027 } \ 3028} while (0) 3029 ADD_FLAG(G_RAID3_DISK_FLAG_DIRTY, "DIRTY"); 3030 ADD_FLAG(G_RAID3_DISK_FLAG_HARDCODED, "HARDCODED"); 3031 ADD_FLAG(G_RAID3_DISK_FLAG_SYNCHRONIZING, 3032 "SYNCHRONIZING"); 3033 ADD_FLAG(G_RAID3_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3034#undef ADD_FLAG 3035 } 3036 sbuf_printf(sb, "</Flags>\n"); 3037 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3038 g_raid3_disk_state2str(disk->d_state)); 3039 } else { 3040 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3041 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3042 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3043 sbuf_printf(sb, "%s<Flags>", indent); 3044 if (sc->sc_flags == 0) 3045 sbuf_printf(sb, "NONE"); 3046 else { 3047 int first = 1; 3048 3049#define ADD_FLAG(flag, name) do { \ 3050 if ((sc->sc_flags & (flag)) != 0) { \ 3051 if (!first) \ 3052 sbuf_printf(sb, ", "); \ 3053 else \ 3054 first = 0; \ 3055 sbuf_printf(sb, name); \ 3056 } \ 3057} while (0) 3058 ADD_FLAG(G_RAID3_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3059 ADD_FLAG(G_RAID3_DEVICE_FLAG_ROUND_ROBIN, 3060 "ROUND-ROBIN"); 3061 ADD_FLAG(G_RAID3_DEVICE_FLAG_VERIFY, "VERIFY"); 3062#undef ADD_FLAG 3063 } 3064 sbuf_printf(sb, "</Flags>\n"); 3065 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3066 sc->sc_ndisks); 3067 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3068 g_raid3_device_state2str(sc->sc_state)); 3069 } 3070} 3071 3072static void 3073g_raid3_shutdown(void *arg, int howto) 3074{ 3075 struct g_class *mp; 3076 struct g_geom *gp, *gp2; 3077 3078 mp = arg; 3079 DROP_GIANT(); 3080 g_topology_lock(); 3081 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3082 if (gp->softc == NULL) 3083 continue; 3084 g_raid3_destroy(gp->softc, 1); 3085 } 3086 g_topology_unlock(); 3087 PICKUP_GIANT(); 3088#if 0 3089 tsleep(&gp, PRIBIO, "r3:shutdown", hz * 20); 3090#endif 3091} 3092 3093static void 3094g_raid3_init(struct g_class *mp) 3095{ 3096 3097 g_raid3_ehtag = EVENTHANDLER_REGISTER(shutdown_post_sync, 3098 g_raid3_shutdown, mp, SHUTDOWN_PRI_FIRST); 3099 if (g_raid3_ehtag == NULL) 3100 G_RAID3_DEBUG(0, "Warning! Cannot register shutdown event."); 3101} 3102 3103static void 3104g_raid3_fini(struct g_class *mp) 3105{ 3106 3107 if (g_raid3_ehtag == NULL) 3108 return; 3109 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid3_ehtag); 3110} 3111 3112DECLARE_GEOM_CLASS(g_raid3_class, g_raid3); 3113