g_raid3.c revision 142727
1/*- 2 * Copyright (c) 2004-2005 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 142727 2005-02-27 23:07:47Z pjd $"); 29 30#include <sys/param.h> 31#include <sys/systm.h> 32#include <sys/kernel.h> 33#include <sys/module.h> 34#include <sys/limits.h> 35#include <sys/lock.h> 36#include <sys/mutex.h> 37#include <sys/bio.h> 38#include <sys/sysctl.h> 39#include <sys/malloc.h> 40#include <sys/eventhandler.h> 41#include <vm/uma.h> 42#include <geom/geom.h> 43#include <sys/proc.h> 44#include <sys/kthread.h> 45#include <sys/sched.h> 46#include <geom/raid3/g_raid3.h> 47 48 49static MALLOC_DEFINE(M_RAID3, "raid3 data", "GEOM_RAID3 Data"); 50 51SYSCTL_DECL(_kern_geom); 52SYSCTL_NODE(_kern_geom, OID_AUTO, raid3, CTLFLAG_RW, 0, "GEOM_RAID3 stuff"); 53u_int g_raid3_debug = 0; 54TUNABLE_INT("kern.geom.raid3.debug", &g_raid3_debug); 55SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, debug, CTLFLAG_RW, &g_raid3_debug, 0, 56 "Debug level"); 57static u_int g_raid3_timeout = 4; 58TUNABLE_INT("kern.geom.raid3.timeout", &g_raid3_timeout); 59SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, timeout, CTLFLAG_RW, &g_raid3_timeout, 60 0, "Time to wait on all raid3 components"); 61static u_int g_raid3_idletime = 5; 62TUNABLE_INT("kern.geom.raid3.idletime", &g_raid3_idletime); 63SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, idletime, CTLFLAG_RW, 64 &g_raid3_idletime, 0, "Mark components as clean when idling"); 65static u_int g_raid3_reqs_per_sync = 5; 66SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, reqs_per_sync, CTLFLAG_RW, 67 &g_raid3_reqs_per_sync, 0, 68 "Number of regular I/O requests per synchronization request"); 69static u_int g_raid3_syncs_per_sec = 1000; 70SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, syncs_per_sec, CTLFLAG_RW, 71 &g_raid3_syncs_per_sec, 0, 72 "Number of synchronizations requests per second"); 73 74static u_int g_raid3_n64k = 50; 75TUNABLE_INT("kern.geom.raid3.n64k", &g_raid3_n64k); 76SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n64k, CTLFLAG_RD, &g_raid3_n64k, 0, 77 "Maximum number of 64kB allocations"); 78static u_int g_raid3_n16k = 200; 79TUNABLE_INT("kern.geom.raid3.n16k", &g_raid3_n16k); 80SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n16k, CTLFLAG_RD, &g_raid3_n16k, 0, 81 "Maximum number of 16kB allocations"); 82static u_int g_raid3_n4k = 1200; 83TUNABLE_INT("kern.geom.raid3.n4k", &g_raid3_n4k); 84SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n4k, CTLFLAG_RD, &g_raid3_n4k, 0, 85 "Maximum number of 4kB allocations"); 86 87SYSCTL_NODE(_kern_geom_raid3, OID_AUTO, stat, CTLFLAG_RW, 0, 88 "GEOM_RAID3 statistics"); 89static u_int g_raid3_parity_mismatch = 0; 90SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, parity_mismatch, CTLFLAG_RD, 91 &g_raid3_parity_mismatch, 0, "Number of failures in VERIFY mode"); 92static u_int g_raid3_64k_requested = 0; 93SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, 64k_requested, CTLFLAG_RD, 94 &g_raid3_64k_requested, 0, "Number of requested 64kB allocations"); 95static u_int g_raid3_64k_failed = 0; 96SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, 64k_failed, CTLFLAG_RD, 97 &g_raid3_64k_failed, 0, "Number of failed 64kB allocations"); 98static u_int g_raid3_16k_requested = 0; 99SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, 16k_requested, CTLFLAG_RD, 100 &g_raid3_16k_requested, 0, "Number of requested 16kB allocations"); 101static u_int g_raid3_16k_failed = 0; 102SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, 16k_failed, CTLFLAG_RD, 103 &g_raid3_16k_failed, 0, "Number of failed 16kB allocations"); 104static u_int g_raid3_4k_requested = 0; 105SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, 4k_requested, CTLFLAG_RD, 106 &g_raid3_4k_requested, 0, "Number of requested 4kB allocations"); 107static u_int g_raid3_4k_failed = 0; 108SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, 4k_failed, CTLFLAG_RD, 109 &g_raid3_4k_failed, 0, "Number of failed 4kB allocations"); 110 111#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 112 G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 113 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 114 G_RAID3_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 115} while (0) 116 117static eventhandler_tag g_raid3_ehtag = NULL; 118 119static int g_raid3_destroy_geom(struct gctl_req *req, struct g_class *mp, 120 struct g_geom *gp); 121static g_taste_t g_raid3_taste; 122static void g_raid3_init(struct g_class *mp); 123static void g_raid3_fini(struct g_class *mp); 124 125struct g_class g_raid3_class = { 126 .name = G_RAID3_CLASS_NAME, 127 .version = G_VERSION, 128 .ctlreq = g_raid3_config, 129 .taste = g_raid3_taste, 130 .destroy_geom = g_raid3_destroy_geom, 131 .init = g_raid3_init, 132 .fini = g_raid3_fini 133}; 134 135 136static void g_raid3_destroy_provider(struct g_raid3_softc *sc); 137static int g_raid3_update_disk(struct g_raid3_disk *disk, u_int state); 138static void g_raid3_update_device(struct g_raid3_softc *sc, boolean_t force); 139static void g_raid3_dumpconf(struct sbuf *sb, const char *indent, 140 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 141static void g_raid3_sync_stop(struct g_raid3_softc *sc, int type); 142 143 144static const char * 145g_raid3_disk_state2str(int state) 146{ 147 148 switch (state) { 149 case G_RAID3_DISK_STATE_NODISK: 150 return ("NODISK"); 151 case G_RAID3_DISK_STATE_NONE: 152 return ("NONE"); 153 case G_RAID3_DISK_STATE_NEW: 154 return ("NEW"); 155 case G_RAID3_DISK_STATE_ACTIVE: 156 return ("ACTIVE"); 157 case G_RAID3_DISK_STATE_STALE: 158 return ("STALE"); 159 case G_RAID3_DISK_STATE_SYNCHRONIZING: 160 return ("SYNCHRONIZING"); 161 case G_RAID3_DISK_STATE_DISCONNECTED: 162 return ("DISCONNECTED"); 163 default: 164 return ("INVALID"); 165 } 166} 167 168static const char * 169g_raid3_device_state2str(int state) 170{ 171 172 switch (state) { 173 case G_RAID3_DEVICE_STATE_STARTING: 174 return ("STARTING"); 175 case G_RAID3_DEVICE_STATE_DEGRADED: 176 return ("DEGRADED"); 177 case G_RAID3_DEVICE_STATE_COMPLETE: 178 return ("COMPLETE"); 179 default: 180 return ("INVALID"); 181 } 182} 183 184const char * 185g_raid3_get_diskname(struct g_raid3_disk *disk) 186{ 187 188 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 189 return ("[unknown]"); 190 return (disk->d_name); 191} 192 193#define g_raid3_xor(src1, src2, dst, size) \ 194 _g_raid3_xor((uint64_t *)(src1), (uint64_t *)(src2), \ 195 (uint64_t *)(dst), (size_t)size) 196static void 197_g_raid3_xor(uint64_t *src1, uint64_t *src2, uint64_t *dst, size_t size) 198{ 199 200 KASSERT((size % 128) == 0, ("Invalid size: %zu.", size)); 201 for (; size > 0; size -= 128) { 202 *dst++ = (*src1++) ^ (*src2++); 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 } 219} 220 221static int 222g_raid3_is_zero(struct bio *bp) 223{ 224 static const uint64_t zeros[] = { 225 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 226 }; 227 u_char *addr; 228 ssize_t size; 229 230 size = bp->bio_length; 231 addr = (u_char *)bp->bio_data; 232 for (; size > 0; size -= sizeof(zeros), addr += sizeof(zeros)) { 233 if (bcmp(addr, zeros, sizeof(zeros)) != 0) 234 return (0); 235 } 236 return (1); 237} 238 239/* 240 * --- Events handling functions --- 241 * Events in geom_raid3 are used to maintain disks and device status 242 * from one thread to simplify locking. 243 */ 244static void 245g_raid3_event_free(struct g_raid3_event *ep) 246{ 247 248 free(ep, M_RAID3); 249} 250 251int 252g_raid3_event_send(void *arg, int state, int flags) 253{ 254 struct g_raid3_softc *sc; 255 struct g_raid3_disk *disk; 256 struct g_raid3_event *ep; 257 int error; 258 259 ep = malloc(sizeof(*ep), M_RAID3, M_WAITOK); 260 G_RAID3_DEBUG(4, "%s: Sending event %p.", __func__, ep); 261 if ((flags & G_RAID3_EVENT_DEVICE) != 0) { 262 disk = NULL; 263 sc = arg; 264 } else { 265 disk = arg; 266 sc = disk->d_softc; 267 } 268 ep->e_disk = disk; 269 ep->e_state = state; 270 ep->e_flags = flags; 271 ep->e_error = 0; 272 mtx_lock(&sc->sc_events_mtx); 273 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 274 mtx_unlock(&sc->sc_events_mtx); 275 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc); 276 mtx_lock(&sc->sc_queue_mtx); 277 wakeup(sc); 278 wakeup(&sc->sc_queue); 279 mtx_unlock(&sc->sc_queue_mtx); 280 if ((flags & G_RAID3_EVENT_DONTWAIT) != 0) 281 return (0); 282 g_topology_assert(); 283 G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 284 g_topology_unlock(); 285 while ((ep->e_flags & G_RAID3_EVENT_DONE) == 0) { 286 mtx_lock(&sc->sc_events_mtx); 287 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "r3:event", 288 hz * 5); 289 } 290 /* Don't even try to use 'sc' here, because it could be already dead. */ 291 g_topology_lock(); 292 error = ep->e_error; 293 g_raid3_event_free(ep); 294 return (error); 295} 296 297static struct g_raid3_event * 298g_raid3_event_get(struct g_raid3_softc *sc) 299{ 300 struct g_raid3_event *ep; 301 302 mtx_lock(&sc->sc_events_mtx); 303 ep = TAILQ_FIRST(&sc->sc_events); 304 mtx_unlock(&sc->sc_events_mtx); 305 return (ep); 306} 307 308static void 309g_raid3_event_remove(struct g_raid3_softc *sc, struct g_raid3_event *ep) 310{ 311 312 mtx_lock(&sc->sc_events_mtx); 313 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 314 mtx_unlock(&sc->sc_events_mtx); 315} 316 317static void 318g_raid3_event_cancel(struct g_raid3_disk *disk) 319{ 320 struct g_raid3_softc *sc; 321 struct g_raid3_event *ep, *tmpep; 322 323 g_topology_assert(); 324 325 sc = disk->d_softc; 326 mtx_lock(&sc->sc_events_mtx); 327 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 328 if ((ep->e_flags & G_RAID3_EVENT_DEVICE) != 0) 329 continue; 330 if (ep->e_disk != disk) 331 continue; 332 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 333 if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0) 334 g_raid3_event_free(ep); 335 else { 336 ep->e_error = ECANCELED; 337 wakeup(ep); 338 } 339 } 340 mtx_unlock(&sc->sc_events_mtx); 341} 342 343/* 344 * Return the number of disks in the given state. 345 * If state is equal to -1, count all connected disks. 346 */ 347u_int 348g_raid3_ndisks(struct g_raid3_softc *sc, int state) 349{ 350 struct g_raid3_disk *disk; 351 u_int n, ndisks; 352 353 for (n = ndisks = 0; n < sc->sc_ndisks; n++) { 354 disk = &sc->sc_disks[n]; 355 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 356 continue; 357 if (state == -1 || disk->d_state == state) 358 ndisks++; 359 } 360 return (ndisks); 361} 362 363static u_int 364g_raid3_nrequests(struct g_raid3_softc *sc, struct g_consumer *cp) 365{ 366 struct bio *bp; 367 u_int nreqs = 0; 368 369 mtx_lock(&sc->sc_queue_mtx); 370 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 371 if (bp->bio_from == cp) 372 nreqs++; 373 } 374 mtx_unlock(&sc->sc_queue_mtx); 375 return (nreqs); 376} 377 378static int 379g_raid3_is_busy(struct g_raid3_softc *sc, struct g_consumer *cp) 380{ 381 382 if (cp->index > 0) { 383 G_RAID3_DEBUG(2, 384 "I/O requests for %s exist, can't destroy it now.", 385 cp->provider->name); 386 return (1); 387 } 388 if (g_raid3_nrequests(sc, cp) > 0) { 389 G_RAID3_DEBUG(2, 390 "I/O requests for %s in queue, can't destroy it now.", 391 cp->provider->name); 392 return (1); 393 } 394 return (0); 395} 396 397static void 398g_raid3_destroy_consumer(void *arg, int flags __unused) 399{ 400 struct g_consumer *cp; 401 402 cp = arg; 403 G_RAID3_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 404 g_detach(cp); 405 g_destroy_consumer(cp); 406} 407 408static void 409g_raid3_kill_consumer(struct g_raid3_softc *sc, struct g_consumer *cp) 410{ 411 struct g_provider *pp; 412 int retaste_wait; 413 414 g_topology_assert(); 415 416 cp->private = NULL; 417 if (g_raid3_is_busy(sc, cp)) 418 return; 419 G_RAID3_DEBUG(2, "Consumer %s destroyed.", cp->provider->name); 420 pp = cp->provider; 421 retaste_wait = 0; 422 if (cp->acw == 1) { 423 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 424 retaste_wait = 1; 425 } 426 G_RAID3_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 427 -cp->acw, -cp->ace, 0); 428 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 429 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 430 if (retaste_wait) { 431 /* 432 * After retaste event was send (inside g_access()), we can send 433 * event to detach and destroy consumer. 434 * A class, which has consumer to the given provider connected 435 * will not receive retaste event for the provider. 436 * This is the way how I ignore retaste events when I close 437 * consumers opened for write: I detach and destroy consumer 438 * after retaste event is sent. 439 */ 440 g_post_event(g_raid3_destroy_consumer, cp, M_WAITOK, NULL); 441 return; 442 } 443 G_RAID3_DEBUG(1, "Consumer %s destroyed.", pp->name); 444 g_detach(cp); 445 g_destroy_consumer(cp); 446} 447 448static int 449g_raid3_connect_disk(struct g_raid3_disk *disk, struct g_provider *pp) 450{ 451 int error; 452 453 g_topology_assert(); 454 KASSERT(disk->d_consumer == NULL, 455 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 456 457 disk->d_consumer = g_new_consumer(disk->d_softc->sc_geom); 458 disk->d_consumer->private = disk; 459 disk->d_consumer->index = 0; 460 error = g_attach(disk->d_consumer, pp); 461 if (error != 0) 462 return (error); 463 error = g_access(disk->d_consumer, 1, 1, 1); 464 if (error != 0) { 465 G_RAID3_DEBUG(0, "Cannot open consumer %s (error=%d).", 466 pp->name, error); 467 return (error); 468 } 469 G_RAID3_DEBUG(2, "Disk %s connected.", g_raid3_get_diskname(disk)); 470 return (0); 471} 472 473static void 474g_raid3_disconnect_consumer(struct g_raid3_softc *sc, struct g_consumer *cp) 475{ 476 477 g_topology_assert(); 478 479 if (cp == NULL) 480 return; 481 if (cp->provider != NULL) 482 g_raid3_kill_consumer(sc, cp); 483 else 484 g_destroy_consumer(cp); 485} 486 487/* 488 * Initialize disk. This means allocate memory, create consumer, attach it 489 * to the provider and open access (r1w1e1) to it. 490 */ 491static struct g_raid3_disk * 492g_raid3_init_disk(struct g_raid3_softc *sc, struct g_provider *pp, 493 struct g_raid3_metadata *md, int *errorp) 494{ 495 struct g_raid3_disk *disk; 496 int error; 497 498 disk = &sc->sc_disks[md->md_no]; 499 error = g_raid3_connect_disk(disk, pp); 500 if (error != 0) 501 goto fail; 502 disk->d_state = G_RAID3_DISK_STATE_NONE; 503 disk->d_flags = md->md_dflags; 504 if (md->md_provider[0] != '\0') 505 disk->d_flags |= G_RAID3_DISK_FLAG_HARDCODED; 506 disk->d_sync.ds_consumer = NULL; 507 disk->d_sync.ds_offset = md->md_sync_offset; 508 disk->d_sync.ds_offset_done = md->md_sync_offset; 509 disk->d_sync.ds_resync = -1; 510 disk->d_genid = md->md_genid; 511 disk->d_sync.ds_syncid = md->md_syncid; 512 if (errorp != NULL) 513 *errorp = 0; 514 return (disk); 515fail: 516 if (errorp != NULL) 517 *errorp = error; 518 if (disk != NULL) 519 g_raid3_disconnect_consumer(sc, disk->d_consumer); 520 return (NULL); 521} 522 523static void 524g_raid3_destroy_disk(struct g_raid3_disk *disk) 525{ 526 struct g_raid3_softc *sc; 527 528 g_topology_assert(); 529 530 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 531 return; 532 g_raid3_event_cancel(disk); 533 sc = disk->d_softc; 534 switch (disk->d_state) { 535 case G_RAID3_DISK_STATE_SYNCHRONIZING: 536 if (sc->sc_syncdisk != NULL) 537 g_raid3_sync_stop(sc, 1); 538 /* FALLTHROUGH */ 539 case G_RAID3_DISK_STATE_NEW: 540 case G_RAID3_DISK_STATE_STALE: 541 case G_RAID3_DISK_STATE_ACTIVE: 542 g_raid3_disconnect_consumer(sc, disk->d_consumer); 543 disk->d_consumer = NULL; 544 break; 545 default: 546 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 547 g_raid3_get_diskname(disk), 548 g_raid3_disk_state2str(disk->d_state))); 549 } 550 disk->d_state = G_RAID3_DISK_STATE_NODISK; 551} 552 553static void 554g_raid3_destroy_device(struct g_raid3_softc *sc) 555{ 556 struct g_raid3_event *ep; 557 struct g_raid3_disk *disk; 558 struct g_geom *gp; 559 struct g_consumer *cp; 560 u_int n; 561 562 g_topology_assert(); 563 564 gp = sc->sc_geom; 565 if (sc->sc_provider != NULL) 566 g_raid3_destroy_provider(sc); 567 for (n = 0; n < sc->sc_ndisks; n++) { 568 disk = &sc->sc_disks[n]; 569 if (disk->d_state != G_RAID3_DISK_STATE_NODISK) { 570 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 571 g_raid3_update_metadata(disk); 572 g_raid3_destroy_disk(disk); 573 } 574 } 575 while ((ep = g_raid3_event_get(sc)) != NULL) { 576 g_raid3_event_remove(sc, ep); 577 if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0) 578 g_raid3_event_free(ep); 579 else { 580 ep->e_error = ECANCELED; 581 ep->e_flags |= G_RAID3_EVENT_DONE; 582 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, ep); 583 mtx_lock(&sc->sc_events_mtx); 584 wakeup(ep); 585 mtx_unlock(&sc->sc_events_mtx); 586 } 587 } 588 callout_drain(&sc->sc_callout); 589 gp->softc = NULL; 590 cp = LIST_FIRST(&sc->sc_sync.ds_geom->consumer); 591 if (cp != NULL) 592 g_raid3_disconnect_consumer(sc, cp); 593 sc->sc_sync.ds_geom->softc = NULL; 594 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 595 uma_zdestroy(sc->sc_zone_64k); 596 uma_zdestroy(sc->sc_zone_16k); 597 uma_zdestroy(sc->sc_zone_4k); 598 mtx_destroy(&sc->sc_queue_mtx); 599 mtx_destroy(&sc->sc_events_mtx); 600 G_RAID3_DEBUG(0, "Device %s destroyed.", gp->name); 601 g_wither_geom(gp, ENXIO); 602} 603 604static void 605g_raid3_orphan(struct g_consumer *cp) 606{ 607 struct g_raid3_disk *disk; 608 609 g_topology_assert(); 610 611 disk = cp->private; 612 if (disk == NULL) 613 return; 614 disk->d_softc->sc_bump_id = G_RAID3_BUMP_SYNCID; 615 g_raid3_event_send(disk, G_RAID3_DISK_STATE_DISCONNECTED, 616 G_RAID3_EVENT_DONTWAIT); 617} 618 619static int 620g_raid3_write_metadata(struct g_raid3_disk *disk, struct g_raid3_metadata *md) 621{ 622 struct g_raid3_softc *sc; 623 struct g_consumer *cp; 624 off_t offset, length; 625 u_char *sector; 626 int error = 0; 627 628 g_topology_assert(); 629 630 sc = disk->d_softc; 631 cp = disk->d_consumer; 632 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 633 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 634 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 635 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 636 cp->acw, cp->ace)); 637 length = cp->provider->sectorsize; 638 offset = cp->provider->mediasize - length; 639 sector = malloc((size_t)length, M_RAID3, M_WAITOK | M_ZERO); 640 if (md != NULL) 641 raid3_metadata_encode(md, sector); 642 g_topology_unlock(); 643 error = g_write_data(cp, offset, sector, length); 644 g_topology_lock(); 645 free(sector, M_RAID3); 646 if (error != 0) { 647 disk->d_softc->sc_bump_id = G_RAID3_BUMP_GENID; 648 g_raid3_event_send(disk, G_RAID3_DISK_STATE_DISCONNECTED, 649 G_RAID3_EVENT_DONTWAIT); 650 } 651 return (error); 652} 653 654int 655g_raid3_clear_metadata(struct g_raid3_disk *disk) 656{ 657 int error; 658 659 g_topology_assert(); 660 error = g_raid3_write_metadata(disk, NULL); 661 if (error == 0) { 662 G_RAID3_DEBUG(2, "Metadata on %s cleared.", 663 g_raid3_get_diskname(disk)); 664 } else { 665 G_RAID3_DEBUG(0, 666 "Cannot clear metadata on disk %s (error=%d).", 667 g_raid3_get_diskname(disk), error); 668 } 669 return (error); 670} 671 672void 673g_raid3_fill_metadata(struct g_raid3_disk *disk, struct g_raid3_metadata *md) 674{ 675 struct g_raid3_softc *sc; 676 struct g_provider *pp; 677 678 sc = disk->d_softc; 679 strlcpy(md->md_magic, G_RAID3_MAGIC, sizeof(md->md_magic)); 680 md->md_version = G_RAID3_VERSION; 681 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 682 md->md_id = sc->sc_id; 683 md->md_all = sc->sc_ndisks; 684 md->md_genid = sc->sc_genid; 685 md->md_mediasize = sc->sc_mediasize; 686 md->md_sectorsize = sc->sc_sectorsize; 687 md->md_mflags = (sc->sc_flags & G_RAID3_DEVICE_FLAG_MASK); 688 md->md_no = disk->d_no; 689 md->md_syncid = disk->d_sync.ds_syncid; 690 md->md_dflags = (disk->d_flags & G_RAID3_DISK_FLAG_MASK); 691 if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) 692 md->md_sync_offset = disk->d_sync.ds_offset_done; 693 else 694 md->md_sync_offset = 0; 695 if (disk->d_consumer != NULL && disk->d_consumer->provider != NULL) 696 pp = disk->d_consumer->provider; 697 else 698 pp = NULL; 699 if ((disk->d_flags & G_RAID3_DISK_FLAG_HARDCODED) != 0 && pp != NULL) 700 strlcpy(md->md_provider, pp->name, sizeof(md->md_provider)); 701 else 702 bzero(md->md_provider, sizeof(md->md_provider)); 703 if (pp != NULL) 704 md->md_provsize = pp->mediasize; 705 else 706 md->md_provsize = 0; 707} 708 709void 710g_raid3_update_metadata(struct g_raid3_disk *disk) 711{ 712 struct g_raid3_metadata md; 713 int error; 714 715 g_topology_assert(); 716 g_raid3_fill_metadata(disk, &md); 717 error = g_raid3_write_metadata(disk, &md); 718 if (error == 0) { 719 G_RAID3_DEBUG(2, "Metadata on %s updated.", 720 g_raid3_get_diskname(disk)); 721 } else { 722 G_RAID3_DEBUG(0, 723 "Cannot update metadata on disk %s (error=%d).", 724 g_raid3_get_diskname(disk), error); 725 } 726} 727 728static void 729g_raid3_bump_syncid(struct g_raid3_softc *sc) 730{ 731 struct g_raid3_disk *disk; 732 u_int n; 733 734 g_topology_assert(); 735 KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) > 0, 736 ("%s called with no active disks (device=%s).", __func__, 737 sc->sc_name)); 738 739 sc->sc_syncid++; 740 G_RAID3_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 741 sc->sc_syncid); 742 for (n = 0; n < sc->sc_ndisks; n++) { 743 disk = &sc->sc_disks[n]; 744 if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 745 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 746 disk->d_sync.ds_syncid = sc->sc_syncid; 747 g_raid3_update_metadata(disk); 748 } 749 } 750} 751 752static void 753g_raid3_bump_genid(struct g_raid3_softc *sc) 754{ 755 struct g_raid3_disk *disk; 756 u_int n; 757 758 g_topology_assert(); 759 KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) > 0, 760 ("%s called with no active disks (device=%s).", __func__, 761 sc->sc_name)); 762 763 sc->sc_genid++; 764 G_RAID3_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 765 sc->sc_genid); 766 for (n = 0; n < sc->sc_ndisks; n++) { 767 disk = &sc->sc_disks[n]; 768 if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 769 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 770 disk->d_genid = sc->sc_genid; 771 g_raid3_update_metadata(disk); 772 } 773 } 774} 775 776static void 777g_raid3_idle(struct g_raid3_softc *sc) 778{ 779 struct g_raid3_disk *disk; 780 u_int i; 781 782 if (sc->sc_provider == NULL || sc->sc_provider->acw == 0) 783 return; 784 sc->sc_idle = 1; 785 g_topology_lock(); 786 for (i = 0; i < sc->sc_ndisks; i++) { 787 disk = &sc->sc_disks[i]; 788 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) 789 continue; 790 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as clean.", 791 g_raid3_get_diskname(disk), sc->sc_name); 792 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 793 g_raid3_update_metadata(disk); 794 } 795 g_topology_unlock(); 796} 797 798static void 799g_raid3_unidle(struct g_raid3_softc *sc) 800{ 801 struct g_raid3_disk *disk; 802 u_int i; 803 804 sc->sc_idle = 0; 805 g_topology_lock(); 806 for (i = 0; i < sc->sc_ndisks; i++) { 807 disk = &sc->sc_disks[i]; 808 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) 809 continue; 810 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as dirty.", 811 g_raid3_get_diskname(disk), sc->sc_name); 812 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY; 813 g_raid3_update_metadata(disk); 814 } 815 g_topology_unlock(); 816} 817 818/* 819 * Return 1 if we should check if RAID3 device is idling. 820 */ 821static int 822g_raid3_check_idle(struct g_raid3_softc *sc) 823{ 824 struct g_raid3_disk *disk; 825 u_int i; 826 827 if (sc->sc_idle) 828 return (0); 829 if (sc->sc_provider != NULL && sc->sc_provider->acw == 0) 830 return (0); 831 /* 832 * Check if there are no in-flight requests. 833 */ 834 for (i = 0; i < sc->sc_ndisks; i++) { 835 disk = &sc->sc_disks[i]; 836 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) 837 continue; 838 if (disk->d_consumer->index > 0) 839 return (0); 840 } 841 return (1); 842} 843 844/* 845 * Treat bio_driver1 field in parent bio as list head and field bio_caller1 846 * in child bio as pointer to the next element on the list. 847 */ 848#define G_RAID3_HEAD_BIO(pbp) (pbp)->bio_driver1 849 850#define G_RAID3_NEXT_BIO(cbp) (cbp)->bio_caller1 851 852#define G_RAID3_FOREACH_BIO(pbp, bp) \ 853 for ((bp) = G_RAID3_HEAD_BIO(pbp); (bp) != NULL; \ 854 (bp) = G_RAID3_NEXT_BIO(bp)) 855 856#define G_RAID3_FOREACH_SAFE_BIO(pbp, bp, tmpbp) \ 857 for ((bp) = G_RAID3_HEAD_BIO(pbp); \ 858 (bp) != NULL && ((tmpbp) = G_RAID3_NEXT_BIO(bp), 1); \ 859 (bp) = (tmpbp)) 860 861static void 862g_raid3_init_bio(struct bio *pbp) 863{ 864 865 G_RAID3_HEAD_BIO(pbp) = NULL; 866} 867 868static void 869g_raid3_remove_bio(struct bio *cbp) 870{ 871 struct bio *pbp, *bp; 872 873 pbp = cbp->bio_parent; 874 if (G_RAID3_HEAD_BIO(pbp) == cbp) 875 G_RAID3_HEAD_BIO(pbp) = G_RAID3_NEXT_BIO(cbp); 876 else { 877 G_RAID3_FOREACH_BIO(pbp, bp) { 878 if (G_RAID3_NEXT_BIO(bp) == cbp) { 879 G_RAID3_NEXT_BIO(bp) = G_RAID3_NEXT_BIO(cbp); 880 break; 881 } 882 } 883 } 884 G_RAID3_NEXT_BIO(cbp) = NULL; 885} 886 887static void 888g_raid3_replace_bio(struct bio *sbp, struct bio *dbp) 889{ 890 struct bio *pbp, *bp; 891 892 g_raid3_remove_bio(sbp); 893 pbp = dbp->bio_parent; 894 G_RAID3_NEXT_BIO(sbp) = G_RAID3_NEXT_BIO(dbp); 895 if (G_RAID3_HEAD_BIO(pbp) == dbp) 896 G_RAID3_HEAD_BIO(pbp) = sbp; 897 else { 898 G_RAID3_FOREACH_BIO(pbp, bp) { 899 if (G_RAID3_NEXT_BIO(bp) == dbp) { 900 G_RAID3_NEXT_BIO(bp) = sbp; 901 break; 902 } 903 } 904 } 905 G_RAID3_NEXT_BIO(dbp) = NULL; 906} 907 908static void 909g_raid3_destroy_bio(struct g_raid3_softc *sc, struct bio *cbp) 910{ 911 struct bio *bp, *pbp; 912 size_t size; 913 914 pbp = cbp->bio_parent; 915 pbp->bio_children--; 916 KASSERT(cbp->bio_data != NULL, ("NULL bio_data")); 917 size = pbp->bio_length / (sc->sc_ndisks - 1); 918 if (size > 16384) 919 uma_zfree(sc->sc_zone_64k, cbp->bio_data); 920 else if (size > 4096) 921 uma_zfree(sc->sc_zone_16k, cbp->bio_data); 922 else 923 uma_zfree(sc->sc_zone_4k, cbp->bio_data); 924 if (G_RAID3_HEAD_BIO(pbp) == cbp) { 925 G_RAID3_HEAD_BIO(pbp) = G_RAID3_NEXT_BIO(cbp); 926 G_RAID3_NEXT_BIO(cbp) = NULL; 927 g_destroy_bio(cbp); 928 } else { 929 G_RAID3_FOREACH_BIO(pbp, bp) { 930 if (G_RAID3_NEXT_BIO(bp) == cbp) 931 break; 932 } 933 if (bp != NULL) { 934 KASSERT(G_RAID3_NEXT_BIO(bp) != NULL, 935 ("NULL bp->bio_driver1")); 936 G_RAID3_NEXT_BIO(bp) = G_RAID3_NEXT_BIO(cbp); 937 G_RAID3_NEXT_BIO(cbp) = NULL; 938 } 939 g_destroy_bio(cbp); 940 } 941} 942 943static struct bio * 944g_raid3_clone_bio(struct g_raid3_softc *sc, struct bio *pbp) 945{ 946 struct bio *bp, *cbp; 947 size_t size; 948 949 cbp = g_clone_bio(pbp); 950 if (cbp == NULL) 951 return (NULL); 952 size = pbp->bio_length / (sc->sc_ndisks - 1); 953 if (size > 16384) { 954 cbp->bio_data = uma_zalloc(sc->sc_zone_64k, M_NOWAIT); 955 g_raid3_64k_requested++; 956 } else if (size > 4096) { 957 cbp->bio_data = uma_zalloc(sc->sc_zone_16k, M_NOWAIT); 958 g_raid3_16k_requested++; 959 } else { 960 cbp->bio_data = uma_zalloc(sc->sc_zone_4k, M_NOWAIT); 961 g_raid3_4k_requested++; 962 } 963 if (cbp->bio_data == NULL) { 964 if (size > 16384) 965 g_raid3_64k_failed++; 966 if (size > 4096) 967 g_raid3_16k_failed++; 968 else 969 g_raid3_4k_failed++; 970 pbp->bio_children--; 971 g_destroy_bio(cbp); 972 return (NULL); 973 } 974 G_RAID3_NEXT_BIO(cbp) = NULL; 975 if (G_RAID3_HEAD_BIO(pbp) == NULL) 976 G_RAID3_HEAD_BIO(pbp) = cbp; 977 else { 978 G_RAID3_FOREACH_BIO(pbp, bp) { 979 if (G_RAID3_NEXT_BIO(bp) == NULL) { 980 G_RAID3_NEXT_BIO(bp) = cbp; 981 break; 982 } 983 } 984 } 985 return (cbp); 986} 987 988static void 989g_raid3_scatter(struct bio *pbp) 990{ 991 struct g_raid3_softc *sc; 992 struct g_raid3_disk *disk; 993 struct bio *bp, *cbp; 994 off_t atom, cadd, padd, left; 995 996 sc = pbp->bio_to->geom->softc; 997 bp = NULL; 998 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_NOPARITY) == 0) { 999 /* 1000 * Find bio for which we should calculate data. 1001 */ 1002 G_RAID3_FOREACH_BIO(pbp, cbp) { 1003 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) { 1004 bp = cbp; 1005 break; 1006 } 1007 } 1008 KASSERT(bp != NULL, ("NULL parity bio.")); 1009 } 1010 atom = sc->sc_sectorsize / (sc->sc_ndisks - 1); 1011 cadd = padd = 0; 1012 for (left = pbp->bio_length; left > 0; left -= sc->sc_sectorsize) { 1013 G_RAID3_FOREACH_BIO(pbp, cbp) { 1014 if (cbp == bp) 1015 continue; 1016 bcopy(pbp->bio_data + padd, cbp->bio_data + cadd, atom); 1017 padd += atom; 1018 } 1019 cadd += atom; 1020 } 1021 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_NOPARITY) == 0) { 1022 struct bio *tmpbp; 1023 1024 /* 1025 * Calculate parity. 1026 */ 1027 bzero(bp->bio_data, bp->bio_length); 1028 G_RAID3_FOREACH_SAFE_BIO(pbp, cbp, tmpbp) { 1029 if (cbp == bp) 1030 continue; 1031 g_raid3_xor(cbp->bio_data, bp->bio_data, bp->bio_data, 1032 bp->bio_length); 1033 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_NODISK) != 0) 1034 g_raid3_destroy_bio(sc, cbp); 1035 } 1036 } 1037 G_RAID3_FOREACH_BIO(pbp, cbp) { 1038 struct g_consumer *cp; 1039 1040 disk = cbp->bio_caller2; 1041 cp = disk->d_consumer; 1042 cbp->bio_to = cp->provider; 1043 G_RAID3_LOGREQ(3, cbp, "Sending request."); 1044 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1045 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1046 cp->acr, cp->acw, cp->ace)); 1047 cp->index++; 1048 g_io_request(cbp, cp); 1049 } 1050} 1051 1052static void 1053g_raid3_gather(struct bio *pbp) 1054{ 1055 struct g_raid3_softc *sc; 1056 struct g_raid3_disk *disk; 1057 struct bio *xbp, *fbp, *cbp; 1058 off_t atom, cadd, padd, left; 1059 1060 sc = pbp->bio_to->geom->softc; 1061 /* 1062 * Find bio for which we have to calculate data. 1063 * While going through this path, check if all requests 1064 * succeeded, if not, deny whole request. 1065 * If we're in COMPLETE mode, we allow one request to fail, 1066 * so if we find one, we're sending it to the parity consumer. 1067 * If there are more failed requests, we deny whole request. 1068 */ 1069 xbp = fbp = NULL; 1070 G_RAID3_FOREACH_BIO(pbp, cbp) { 1071 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) { 1072 KASSERT(xbp == NULL, ("More than one parity bio.")); 1073 xbp = cbp; 1074 } 1075 if (cbp->bio_error == 0) 1076 continue; 1077 /* 1078 * Found failed request. 1079 */ 1080 G_RAID3_LOGREQ(0, cbp, "Request failed."); 1081 disk = cbp->bio_caller2; 1082 if (disk != NULL) { 1083 /* 1084 * Actually this is pointless to bump genid, 1085 * because whole device is fucked up. 1086 */ 1087 sc->sc_bump_id |= G_RAID3_BUMP_GENID; 1088 g_raid3_event_send(disk, 1089 G_RAID3_DISK_STATE_DISCONNECTED, 1090 G_RAID3_EVENT_DONTWAIT); 1091 } 1092 if (fbp == NULL) { 1093 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_DEGRADED) != 0) { 1094 /* 1095 * We are already in degraded mode, so we can't 1096 * accept any failures. 1097 */ 1098 if (pbp->bio_error == 0) 1099 pbp->bio_error = fbp->bio_error; 1100 } else { 1101 fbp = cbp; 1102 } 1103 } else { 1104 /* 1105 * Next failed request, that's too many. 1106 */ 1107 if (pbp->bio_error == 0) 1108 pbp->bio_error = fbp->bio_error; 1109 } 1110 } 1111 if (pbp->bio_error != 0) 1112 goto finish; 1113 if (fbp != NULL && (pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) { 1114 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_VERIFY; 1115 if (xbp != fbp) 1116 g_raid3_replace_bio(xbp, fbp); 1117 g_raid3_destroy_bio(sc, fbp); 1118 } else if (fbp != NULL) { 1119 struct g_consumer *cp; 1120 1121 /* 1122 * One request failed, so send the same request to 1123 * the parity consumer. 1124 */ 1125 disk = pbp->bio_driver2; 1126 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) { 1127 pbp->bio_error = fbp->bio_error; 1128 goto finish; 1129 } 1130 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED; 1131 pbp->bio_inbed--; 1132 fbp->bio_flags &= ~(BIO_DONE | BIO_ERROR); 1133 if (disk->d_no == sc->sc_ndisks - 1) 1134 fbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1135 fbp->bio_error = 0; 1136 fbp->bio_completed = 0; 1137 fbp->bio_children = 0; 1138 fbp->bio_inbed = 0; 1139 cp = disk->d_consumer; 1140 fbp->bio_caller2 = disk; 1141 fbp->bio_to = cp->provider; 1142 G_RAID3_LOGREQ(3, fbp, "Sending request (recover)."); 1143 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1144 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1145 cp->acr, cp->acw, cp->ace)); 1146 cp->index++; 1147 g_io_request(fbp, cp); 1148 return; 1149 } 1150 if (xbp != NULL) { 1151 /* 1152 * Calculate parity. 1153 */ 1154 G_RAID3_FOREACH_BIO(pbp, cbp) { 1155 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) 1156 continue; 1157 g_raid3_xor(cbp->bio_data, xbp->bio_data, xbp->bio_data, 1158 xbp->bio_length); 1159 } 1160 xbp->bio_cflags &= ~G_RAID3_BIO_CFLAG_PARITY; 1161 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) { 1162 if (!g_raid3_is_zero(xbp)) { 1163 g_raid3_parity_mismatch++; 1164 pbp->bio_error = EIO; 1165 goto finish; 1166 } 1167 g_raid3_destroy_bio(sc, xbp); 1168 } 1169 } 1170 atom = sc->sc_sectorsize / (sc->sc_ndisks - 1); 1171 cadd = padd = 0; 1172 for (left = pbp->bio_length; left > 0; left -= sc->sc_sectorsize) { 1173 G_RAID3_FOREACH_BIO(pbp, cbp) { 1174 bcopy(cbp->bio_data + cadd, pbp->bio_data + padd, atom); 1175 pbp->bio_completed += atom; 1176 padd += atom; 1177 } 1178 cadd += atom; 1179 } 1180finish: 1181 if (pbp->bio_error == 0) 1182 G_RAID3_LOGREQ(3, pbp, "Request finished."); 1183 else { 1184 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) 1185 G_RAID3_LOGREQ(1, pbp, "Verification error."); 1186 else 1187 G_RAID3_LOGREQ(0, pbp, "Request failed."); 1188 } 1189 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_MASK; 1190 g_io_deliver(pbp, pbp->bio_error); 1191 while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL) 1192 g_raid3_destroy_bio(sc, cbp); 1193} 1194 1195static void 1196g_raid3_done(struct bio *bp) 1197{ 1198 struct g_raid3_softc *sc; 1199 1200 sc = bp->bio_from->geom->softc; 1201 bp->bio_cflags |= G_RAID3_BIO_CFLAG_REGULAR; 1202 G_RAID3_LOGREQ(3, bp, "Regular request done (error=%d).", bp->bio_error); 1203 mtx_lock(&sc->sc_queue_mtx); 1204 bioq_insert_head(&sc->sc_queue, bp); 1205 wakeup(sc); 1206 wakeup(&sc->sc_queue); 1207 mtx_unlock(&sc->sc_queue_mtx); 1208} 1209 1210static void 1211g_raid3_regular_request(struct bio *cbp) 1212{ 1213 struct g_raid3_softc *sc; 1214 struct g_raid3_disk *disk; 1215 struct bio *pbp; 1216 1217 g_topology_assert_not(); 1218 1219 cbp->bio_from->index--; 1220 pbp = cbp->bio_parent; 1221 sc = pbp->bio_to->geom->softc; 1222 disk = cbp->bio_from->private; 1223 if (disk == NULL) { 1224 g_topology_lock(); 1225 g_raid3_kill_consumer(sc, cbp->bio_from); 1226 g_topology_unlock(); 1227 } 1228 1229 G_RAID3_LOGREQ(3, cbp, "Request finished."); 1230 pbp->bio_inbed++; 1231 KASSERT(pbp->bio_inbed <= pbp->bio_children, 1232 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 1233 pbp->bio_children)); 1234 if (pbp->bio_inbed != pbp->bio_children) 1235 return; 1236 switch (pbp->bio_cmd) { 1237 case BIO_READ: 1238 g_raid3_gather(pbp); 1239 break; 1240 case BIO_WRITE: 1241 case BIO_DELETE: 1242 { 1243 int error = 0; 1244 1245 pbp->bio_completed = pbp->bio_length; 1246 while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL) { 1247 if (cbp->bio_error != 0) { 1248 disk = cbp->bio_caller2; 1249 if (disk != NULL) { 1250 sc->sc_bump_id |= G_RAID3_BUMP_GENID; 1251 g_raid3_event_send(disk, 1252 G_RAID3_DISK_STATE_DISCONNECTED, 1253 G_RAID3_EVENT_DONTWAIT); 1254 } 1255 if (error == 0) 1256 error = cbp->bio_error; 1257 else if (pbp->bio_error == 0) { 1258 /* 1259 * Next failed request, that's too many. 1260 */ 1261 pbp->bio_error = error; 1262 } 1263 } 1264 g_raid3_destroy_bio(sc, cbp); 1265 } 1266 if (pbp->bio_error == 0) 1267 G_RAID3_LOGREQ(3, pbp, "Request finished."); 1268 else 1269 G_RAID3_LOGREQ(0, pbp, "Request failed."); 1270 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_DEGRADED; 1271 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_NOPARITY; 1272 g_io_deliver(pbp, pbp->bio_error); 1273 break; 1274 } 1275 } 1276} 1277 1278static void 1279g_raid3_sync_done(struct bio *bp) 1280{ 1281 struct g_raid3_softc *sc; 1282 1283 G_RAID3_LOGREQ(3, bp, "Synchronization request delivered."); 1284 sc = bp->bio_from->geom->softc; 1285 bp->bio_cflags |= G_RAID3_BIO_CFLAG_SYNC; 1286 mtx_lock(&sc->sc_queue_mtx); 1287 bioq_insert_head(&sc->sc_queue, bp); 1288 wakeup(sc); 1289 wakeup(&sc->sc_queue); 1290 mtx_unlock(&sc->sc_queue_mtx); 1291} 1292 1293static void 1294g_raid3_start(struct bio *bp) 1295{ 1296 struct g_raid3_softc *sc; 1297 1298 sc = bp->bio_to->geom->softc; 1299 /* 1300 * If sc == NULL or there are no valid disks, provider's error 1301 * should be set and g_raid3_start() should not be called at all. 1302 */ 1303 KASSERT(sc != NULL && (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 1304 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE), 1305 ("Provider's error should be set (error=%d)(device=%s).", 1306 bp->bio_to->error, bp->bio_to->name)); 1307 G_RAID3_LOGREQ(3, bp, "Request received."); 1308 1309 switch (bp->bio_cmd) { 1310 case BIO_READ: 1311 case BIO_WRITE: 1312 case BIO_DELETE: 1313 break; 1314 case BIO_GETATTR: 1315 default: 1316 g_io_deliver(bp, EOPNOTSUPP); 1317 return; 1318 } 1319 mtx_lock(&sc->sc_queue_mtx); 1320 bioq_insert_tail(&sc->sc_queue, bp); 1321 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1322 wakeup(sc); 1323 mtx_unlock(&sc->sc_queue_mtx); 1324} 1325 1326/* 1327 * Send one synchronization request. 1328 */ 1329static void 1330g_raid3_sync_one(struct g_raid3_softc *sc) 1331{ 1332 struct g_raid3_disk *disk; 1333 struct bio *bp; 1334 1335 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED, 1336 ("Wrong device state (%s, %s).", sc->sc_name, 1337 g_raid3_device_state2str(sc->sc_state))); 1338 disk = sc->sc_syncdisk; 1339 KASSERT(disk != NULL, ("No sync disk (%s).", sc->sc_name)); 1340 KASSERT(disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING, 1341 ("Disk %s is not marked for synchronization.", 1342 g_raid3_get_diskname(disk))); 1343 1344 bp = g_new_bio(); 1345 if (bp == NULL) 1346 return; 1347 bp->bio_parent = NULL; 1348 bp->bio_cmd = BIO_READ; 1349 bp->bio_offset = disk->d_sync.ds_offset * (sc->sc_ndisks - 1); 1350 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1351 bp->bio_cflags = 0; 1352 bp->bio_done = g_raid3_sync_done; 1353 bp->bio_data = disk->d_sync.ds_data; 1354 if (bp->bio_data == NULL) { 1355 g_destroy_bio(bp); 1356 return; 1357 } 1358 bp->bio_cflags = G_RAID3_BIO_CFLAG_REGSYNC; 1359 disk->d_sync.ds_offset += bp->bio_length / (sc->sc_ndisks - 1); 1360 bp->bio_to = sc->sc_provider; 1361 G_RAID3_LOGREQ(3, bp, "Sending synchronization request."); 1362 disk->d_sync.ds_consumer->index++; 1363 g_io_request(bp, disk->d_sync.ds_consumer); 1364} 1365 1366static void 1367g_raid3_sync_request(struct bio *bp) 1368{ 1369 struct g_raid3_softc *sc; 1370 struct g_raid3_disk *disk; 1371 1372 bp->bio_from->index--; 1373 sc = bp->bio_from->geom->softc; 1374 disk = bp->bio_from->private; 1375 if (disk == NULL) { 1376 g_topology_lock(); 1377 g_raid3_kill_consumer(sc, bp->bio_from); 1378 g_topology_unlock(); 1379 g_destroy_bio(bp); 1380 return; 1381 } 1382 1383 /* 1384 * Synchronization request. 1385 */ 1386 switch (bp->bio_cmd) { 1387 case BIO_READ: 1388 { 1389 struct g_consumer *cp; 1390 u_char *dst, *src; 1391 off_t left; 1392 u_int atom; 1393 1394 if (bp->bio_error != 0) { 1395 G_RAID3_LOGREQ(0, bp, 1396 "Synchronization request failed (error=%d).", 1397 bp->bio_error); 1398 g_destroy_bio(bp); 1399 return; 1400 } 1401 G_RAID3_LOGREQ(3, bp, "Synchronization request finished."); 1402 atom = sc->sc_sectorsize / (sc->sc_ndisks - 1); 1403 dst = src = bp->bio_data; 1404 if (disk->d_no == sc->sc_ndisks - 1) { 1405 u_int n; 1406 1407 /* Parity component. */ 1408 for (left = bp->bio_length; left > 0; 1409 left -= sc->sc_sectorsize) { 1410 bcopy(src, dst, atom); 1411 src += atom; 1412 for (n = 1; n < sc->sc_ndisks - 1; n++) { 1413 g_raid3_xor(src, dst, dst, atom); 1414 src += atom; 1415 } 1416 dst += atom; 1417 } 1418 } else { 1419 /* Regular component. */ 1420 src += atom * disk->d_no; 1421 for (left = bp->bio_length; left > 0; 1422 left -= sc->sc_sectorsize) { 1423 bcopy(src, dst, atom); 1424 src += sc->sc_sectorsize; 1425 dst += atom; 1426 } 1427 } 1428 bp->bio_offset /= sc->sc_ndisks - 1; 1429 bp->bio_length /= sc->sc_ndisks - 1; 1430 bp->bio_cmd = BIO_WRITE; 1431 bp->bio_cflags = 0; 1432 bp->bio_children = bp->bio_inbed = 0; 1433 cp = disk->d_consumer; 1434 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1435 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1436 cp->acr, cp->acw, cp->ace)); 1437 cp->index++; 1438 g_io_request(bp, cp); 1439 return; 1440 } 1441 case BIO_WRITE: 1442 { 1443 struct g_raid3_disk_sync *sync; 1444 1445 if (bp->bio_error != 0) { 1446 G_RAID3_LOGREQ(0, bp, 1447 "Synchronization request failed (error=%d).", 1448 bp->bio_error); 1449 g_destroy_bio(bp); 1450 sc->sc_bump_id |= G_RAID3_BUMP_GENID; 1451 g_raid3_event_send(disk, 1452 G_RAID3_DISK_STATE_DISCONNECTED, 1453 G_RAID3_EVENT_DONTWAIT); 1454 return; 1455 } 1456 G_RAID3_LOGREQ(3, bp, "Synchronization request finished."); 1457 sync = &disk->d_sync; 1458 sync->ds_offset_done = bp->bio_offset + bp->bio_length; 1459 g_destroy_bio(bp); 1460 if (sync->ds_resync != -1) 1461 return; 1462 if (sync->ds_offset_done == 1463 sc->sc_mediasize / (sc->sc_ndisks - 1)) { 1464 /* 1465 * Disk up-to-date, activate it. 1466 */ 1467 g_raid3_event_send(disk, G_RAID3_DISK_STATE_ACTIVE, 1468 G_RAID3_EVENT_DONTWAIT); 1469 return; 1470 } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) { 1471 /* 1472 * Update offset_done on every 100 blocks. 1473 * XXX: This should be configurable. 1474 */ 1475 g_topology_lock(); 1476 g_raid3_update_metadata(disk); 1477 g_topology_unlock(); 1478 } 1479 return; 1480 } 1481 default: 1482 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1483 bp->bio_cmd, sc->sc_name)); 1484 break; 1485 } 1486} 1487 1488static int 1489g_raid3_register_request(struct bio *pbp) 1490{ 1491 struct g_raid3_softc *sc; 1492 struct g_raid3_disk *disk; 1493 struct g_consumer *cp; 1494 struct bio *cbp; 1495 off_t offset, length; 1496 u_int n, ndisks; 1497 int round_robin, verify; 1498 1499 ndisks = 0; 1500 sc = pbp->bio_to->geom->softc; 1501 if ((pbp->bio_cflags & G_RAID3_BIO_CFLAG_REGSYNC) != 0 && 1502 sc->sc_syncdisk == NULL) { 1503 g_io_deliver(pbp, EIO); 1504 return (0); 1505 } 1506 g_raid3_init_bio(pbp); 1507 length = pbp->bio_length / (sc->sc_ndisks - 1); 1508 offset = pbp->bio_offset / (sc->sc_ndisks - 1); 1509 round_robin = verify = 0; 1510 switch (pbp->bio_cmd) { 1511 case BIO_READ: 1512 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_VERIFY) != 0 && 1513 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 1514 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_VERIFY; 1515 verify = 1; 1516 ndisks = sc->sc_ndisks; 1517 } else { 1518 verify = 0; 1519 ndisks = sc->sc_ndisks - 1; 1520 } 1521 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_ROUND_ROBIN) != 0 && 1522 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 1523 round_robin = 1; 1524 } else { 1525 round_robin = 0; 1526 } 1527 KASSERT(!round_robin || !verify, 1528 ("ROUND-ROBIN and VERIFY are mutually exclusive.")); 1529 pbp->bio_driver2 = &sc->sc_disks[sc->sc_ndisks - 1]; 1530 break; 1531 case BIO_WRITE: 1532 case BIO_DELETE: 1533 { 1534 struct g_raid3_disk_sync *sync; 1535 1536 if (sc->sc_idle) 1537 g_raid3_unidle(sc); 1538 1539 ndisks = sc->sc_ndisks; 1540 1541 if (sc->sc_syncdisk == NULL) 1542 break; 1543 sync = &sc->sc_syncdisk->d_sync; 1544 if (offset >= sync->ds_offset) 1545 break; 1546 if (offset + length <= sync->ds_offset_done) 1547 break; 1548 if (offset >= sync->ds_resync && sync->ds_resync != -1) 1549 break; 1550 sync->ds_resync = offset - (offset % MAXPHYS); 1551 break; 1552 } 1553 } 1554 for (n = 0; n < ndisks; n++) { 1555 disk = &sc->sc_disks[n]; 1556 cbp = g_raid3_clone_bio(sc, pbp); 1557 if (cbp == NULL) { 1558 while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL) 1559 g_raid3_destroy_bio(sc, cbp); 1560 return (ENOMEM); 1561 } 1562 cbp->bio_offset = offset; 1563 cbp->bio_length = length; 1564 cbp->bio_done = g_raid3_done; 1565 switch (pbp->bio_cmd) { 1566 case BIO_READ: 1567 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) { 1568 /* 1569 * Replace invalid component with the parity 1570 * component. 1571 */ 1572 disk = &sc->sc_disks[sc->sc_ndisks - 1]; 1573 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1574 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED; 1575 } else if (round_robin && 1576 disk->d_no == sc->sc_round_robin) { 1577 /* 1578 * In round-robin mode skip one data component 1579 * and use parity component when reading. 1580 */ 1581 pbp->bio_driver2 = disk; 1582 disk = &sc->sc_disks[sc->sc_ndisks - 1]; 1583 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1584 sc->sc_round_robin++; 1585 round_robin = 0; 1586 } else if (verify && disk->d_no == sc->sc_ndisks - 1) { 1587 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY; 1588 } 1589 break; 1590 case BIO_WRITE: 1591 case BIO_DELETE: 1592 if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 1593 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 1594 if (n == ndisks - 1) { 1595 /* 1596 * Active parity component, mark it as such. 1597 */ 1598 cbp->bio_cflags |= 1599 G_RAID3_BIO_CFLAG_PARITY; 1600 } 1601 } else { 1602 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED; 1603 if (n == ndisks - 1) { 1604 /* 1605 * Parity component is not connected, 1606 * so destroy its request. 1607 */ 1608 pbp->bio_pflags |= 1609 G_RAID3_BIO_PFLAG_NOPARITY; 1610 g_raid3_destroy_bio(sc, cbp); 1611 cbp = NULL; 1612 } else { 1613 cbp->bio_cflags |= 1614 G_RAID3_BIO_CFLAG_NODISK; 1615 disk = NULL; 1616 } 1617 } 1618 break; 1619 } 1620 if (cbp != NULL) 1621 cbp->bio_caller2 = disk; 1622 } 1623 switch (pbp->bio_cmd) { 1624 case BIO_READ: 1625 if (round_robin) { 1626 /* 1627 * If we are in round-robin mode and 'round_robin' is 1628 * still 1, it means, that we skipped parity component 1629 * for this read and must reset sc_round_robin field. 1630 */ 1631 sc->sc_round_robin = 0; 1632 } 1633 G_RAID3_FOREACH_BIO(pbp, cbp) { 1634 disk = cbp->bio_caller2; 1635 cp = disk->d_consumer; 1636 cbp->bio_to = cp->provider; 1637 G_RAID3_LOGREQ(3, cbp, "Sending request."); 1638 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1639 ("Consumer %s not opened (r%dw%de%d).", 1640 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1641 cp->index++; 1642 g_io_request(cbp, cp); 1643 } 1644 break; 1645 case BIO_WRITE: 1646 case BIO_DELETE: 1647 /* 1648 * Bump syncid on first write. 1649 */ 1650 if ((sc->sc_bump_id & G_RAID3_BUMP_SYNCID) != 0) { 1651 sc->sc_bump_id &= ~G_RAID3_BUMP_SYNCID; 1652 g_topology_lock(); 1653 g_raid3_bump_syncid(sc); 1654 g_topology_unlock(); 1655 } 1656 g_raid3_scatter(pbp); 1657 break; 1658 } 1659 return (0); 1660} 1661 1662static int 1663g_raid3_can_destroy(struct g_raid3_softc *sc) 1664{ 1665 struct g_geom *gp; 1666 struct g_consumer *cp; 1667 1668 g_topology_assert(); 1669 gp = sc->sc_geom; 1670 LIST_FOREACH(cp, &gp->consumer, consumer) { 1671 if (g_raid3_is_busy(sc, cp)) 1672 return (0); 1673 } 1674 gp = sc->sc_sync.ds_geom; 1675 LIST_FOREACH(cp, &gp->consumer, consumer) { 1676 if (g_raid3_is_busy(sc, cp)) 1677 return (0); 1678 } 1679 G_RAID3_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1680 sc->sc_name); 1681 return (1); 1682} 1683 1684static int 1685g_raid3_try_destroy(struct g_raid3_softc *sc) 1686{ 1687 1688 g_topology_lock(); 1689 if (!g_raid3_can_destroy(sc)) { 1690 g_topology_unlock(); 1691 return (0); 1692 } 1693 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_WAIT) != 0) { 1694 g_topology_unlock(); 1695 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, 1696 &sc->sc_worker); 1697 wakeup(&sc->sc_worker); 1698 sc->sc_worker = NULL; 1699 } else { 1700 g_raid3_destroy_device(sc); 1701 g_topology_unlock(); 1702 free(sc->sc_disks, M_RAID3); 1703 free(sc, M_RAID3); 1704 } 1705 return (1); 1706} 1707 1708/* 1709 * Worker thread. 1710 */ 1711static void 1712g_raid3_worker(void *arg) 1713{ 1714 struct g_raid3_softc *sc; 1715 struct g_raid3_disk *disk; 1716 struct g_raid3_disk_sync *sync; 1717 struct g_raid3_event *ep; 1718 struct bio *bp; 1719 u_int nreqs; 1720 1721 sc = arg; 1722 mtx_lock_spin(&sched_lock); 1723 sched_prio(curthread, PRIBIO); 1724 mtx_unlock_spin(&sched_lock); 1725 1726 nreqs = 0; 1727 for (;;) { 1728 G_RAID3_DEBUG(5, "%s: Let's see...", __func__); 1729 /* 1730 * First take a look at events. 1731 * This is important to handle events before any I/O requests. 1732 */ 1733 ep = g_raid3_event_get(sc); 1734 if (ep != NULL && g_topology_try_lock()) { 1735 g_raid3_event_remove(sc, ep); 1736 if ((ep->e_flags & G_RAID3_EVENT_DEVICE) != 0) { 1737 /* Update only device status. */ 1738 G_RAID3_DEBUG(3, 1739 "Running event for device %s.", 1740 sc->sc_name); 1741 ep->e_error = 0; 1742 g_raid3_update_device(sc, 1); 1743 } else { 1744 /* Update disk status. */ 1745 G_RAID3_DEBUG(3, "Running event for disk %s.", 1746 g_raid3_get_diskname(ep->e_disk)); 1747 ep->e_error = g_raid3_update_disk(ep->e_disk, 1748 ep->e_state); 1749 if (ep->e_error == 0) 1750 g_raid3_update_device(sc, 0); 1751 } 1752 g_topology_unlock(); 1753 if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0) { 1754 KASSERT(ep->e_error == 0, 1755 ("Error cannot be handled.")); 1756 g_raid3_event_free(ep); 1757 } else { 1758 ep->e_flags |= G_RAID3_EVENT_DONE; 1759 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, 1760 ep); 1761 mtx_lock(&sc->sc_events_mtx); 1762 wakeup(ep); 1763 mtx_unlock(&sc->sc_events_mtx); 1764 } 1765 if ((sc->sc_flags & 1766 G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 1767 if (g_raid3_try_destroy(sc)) 1768 kthread_exit(0); 1769 } 1770 G_RAID3_DEBUG(5, "%s: I'm here 1.", __func__); 1771 continue; 1772 } 1773 /* 1774 * Now I/O requests. 1775 */ 1776 /* Get first request from the queue. */ 1777 mtx_lock(&sc->sc_queue_mtx); 1778 bp = bioq_first(&sc->sc_queue); 1779 if (bp == NULL) { 1780 if (ep != NULL) { 1781 /* 1782 * No I/O requests and topology lock was 1783 * already held? Try again. 1784 */ 1785 mtx_unlock(&sc->sc_queue_mtx); 1786 tsleep(ep, PRIBIO, "r3:top1", hz / 5); 1787 continue; 1788 } 1789 if ((sc->sc_flags & 1790 G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 1791 mtx_unlock(&sc->sc_queue_mtx); 1792 if (g_raid3_try_destroy(sc)) 1793 kthread_exit(0); 1794 mtx_lock(&sc->sc_queue_mtx); 1795 } 1796 } 1797 if (sc->sc_syncdisk != NULL && 1798 (bp == NULL || nreqs > g_raid3_reqs_per_sync)) { 1799 mtx_unlock(&sc->sc_queue_mtx); 1800 /* 1801 * It is time for synchronization... 1802 */ 1803 nreqs = 0; 1804 disk = sc->sc_syncdisk; 1805 sync = &disk->d_sync; 1806 if (sync->ds_offset < 1807 sc->sc_mediasize / (sc->sc_ndisks - 1) && 1808 sync->ds_offset == sync->ds_offset_done) { 1809 if (sync->ds_resync != -1) { 1810 sync->ds_offset = sync->ds_resync; 1811 sync->ds_offset_done = sync->ds_resync; 1812 sync->ds_resync = -1; 1813 } 1814 g_raid3_sync_one(sc); 1815 } 1816 G_RAID3_DEBUG(5, "%s: I'm here 2.", __func__); 1817 goto sleep; 1818 } 1819 if (bp == NULL) { 1820 if (g_raid3_check_idle(sc)) { 1821 u_int idletime; 1822 1823 idletime = g_raid3_idletime; 1824 if (idletime == 0) 1825 idletime = 1; 1826 idletime *= hz; 1827 if (msleep(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1828 "r3:w1", idletime) == EWOULDBLOCK) { 1829 G_RAID3_DEBUG(5, "%s: I'm here 3.", 1830 __func__); 1831 /* 1832 * No I/O requests in 'idletime' 1833 * seconds, so mark components as clean. 1834 */ 1835 g_raid3_idle(sc); 1836 } 1837 G_RAID3_DEBUG(5, "%s: I'm here 4.", __func__); 1838 } else { 1839 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1840 "r3:w2", 0); 1841 G_RAID3_DEBUG(5, "%s: I'm here 5.", __func__); 1842 } 1843 continue; 1844 } 1845 nreqs++; 1846 bioq_remove(&sc->sc_queue, bp); 1847 mtx_unlock(&sc->sc_queue_mtx); 1848 1849 if ((bp->bio_cflags & G_RAID3_BIO_CFLAG_REGULAR) != 0) { 1850 g_raid3_regular_request(bp); 1851 } else if ((bp->bio_cflags & G_RAID3_BIO_CFLAG_SYNC) != 0) { 1852 u_int timeout, sps; 1853 1854 g_raid3_sync_request(bp); 1855sleep: 1856 sps = atomic_load_acq_int(&g_raid3_syncs_per_sec); 1857 if (sps == 0) { 1858 G_RAID3_DEBUG(5, "%s: I'm here 6.", __func__); 1859 continue; 1860 } 1861 if (ep != NULL) { 1862 /* 1863 * We have some pending events, don't sleep now. 1864 */ 1865 G_RAID3_DEBUG(5, "%s: I'm here 7.", __func__); 1866 tsleep(ep, PRIBIO, "r3:top2", hz / 5); 1867 continue; 1868 } 1869 mtx_lock(&sc->sc_queue_mtx); 1870 if (bioq_first(&sc->sc_queue) != NULL) { 1871 mtx_unlock(&sc->sc_queue_mtx); 1872 G_RAID3_DEBUG(5, "%s: I'm here 8.", __func__); 1873 continue; 1874 } 1875 timeout = hz / sps; 1876 if (timeout == 0) 1877 timeout = 1; 1878 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "r3:w2", 1879 timeout); 1880 } else { 1881 if (g_raid3_register_request(bp) != 0) { 1882 mtx_lock(&sc->sc_queue_mtx); 1883 bioq_insert_tail(&sc->sc_queue, bp); 1884 MSLEEP(&sc->sc_queue, &sc->sc_queue_mtx, 1885 PRIBIO | PDROP, "r3:lowmem", hz / 10); 1886 } 1887 } 1888 G_RAID3_DEBUG(5, "%s: I'm here 9.", __func__); 1889 } 1890} 1891 1892/* 1893 * Open disk's consumer if needed. 1894 */ 1895static void 1896g_raid3_update_access(struct g_raid3_disk *disk) 1897{ 1898 struct g_provider *pp; 1899 1900 g_topology_assert(); 1901 1902 pp = disk->d_softc->sc_provider; 1903 if (pp == NULL) 1904 return; 1905 if (pp->acw > 0) { 1906 if ((disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) == 0) { 1907 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1908 g_raid3_get_diskname(disk), disk->d_softc->sc_name); 1909 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY; 1910 } 1911 } else if (pp->acw == 0) { 1912 if ((disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) != 0) { 1913 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as clean.", 1914 g_raid3_get_diskname(disk), disk->d_softc->sc_name); 1915 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 1916 } 1917 } 1918} 1919 1920static void 1921g_raid3_sync_start(struct g_raid3_softc *sc) 1922{ 1923 struct g_raid3_disk *disk; 1924 int error; 1925 u_int n; 1926 1927 g_topology_assert(); 1928 1929 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED, 1930 ("Device not in DEGRADED state (%s, %u).", sc->sc_name, 1931 sc->sc_state)); 1932 KASSERT(sc->sc_syncdisk == NULL, ("Syncdisk is not NULL (%s, %u).", 1933 sc->sc_name, sc->sc_state)); 1934 disk = NULL; 1935 for (n = 0; n < sc->sc_ndisks; n++) { 1936 if (sc->sc_disks[n].d_state != G_RAID3_DISK_STATE_SYNCHRONIZING) 1937 continue; 1938 disk = &sc->sc_disks[n]; 1939 break; 1940 } 1941 if (disk == NULL) 1942 return; 1943 1944 G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1945 g_raid3_get_diskname(disk)); 1946 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY; 1947 KASSERT(disk->d_sync.ds_consumer == NULL, 1948 ("Sync consumer already exists (device=%s, disk=%s).", 1949 sc->sc_name, g_raid3_get_diskname(disk))); 1950 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1951 disk->d_sync.ds_consumer->private = disk; 1952 disk->d_sync.ds_consumer->index = 0; 1953 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1954 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1955 disk->d_softc->sc_name, error)); 1956 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1957 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1958 disk->d_softc->sc_name, error)); 1959 disk->d_sync.ds_data = malloc(MAXPHYS, M_RAID3, M_WAITOK); 1960 sc->sc_syncdisk = disk; 1961} 1962 1963/* 1964 * Stop synchronization process. 1965 * type: 0 - synchronization finished 1966 * 1 - synchronization stopped 1967 */ 1968static void 1969g_raid3_sync_stop(struct g_raid3_softc *sc, int type) 1970{ 1971 struct g_raid3_disk *disk; 1972 1973 g_topology_assert(); 1974 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED, 1975 ("Device not in DEGRADED state (%s, %u).", sc->sc_name, 1976 sc->sc_state)); 1977 disk = sc->sc_syncdisk; 1978 sc->sc_syncdisk = NULL; 1979 KASSERT(disk != NULL, ("No disk was synchronized (%s).", sc->sc_name)); 1980 KASSERT(disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING, 1981 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 1982 g_raid3_disk_state2str(disk->d_state))); 1983 if (disk->d_sync.ds_consumer == NULL) 1984 return; 1985 1986 if (type == 0) { 1987 G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1988 disk->d_softc->sc_name, g_raid3_get_diskname(disk)); 1989 } else /* if (type == 1) */ { 1990 G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1991 disk->d_softc->sc_name, g_raid3_get_diskname(disk)); 1992 } 1993 g_raid3_kill_consumer(disk->d_softc, disk->d_sync.ds_consumer); 1994 free(disk->d_sync.ds_data, M_RAID3); 1995 disk->d_sync.ds_consumer = NULL; 1996 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 1997} 1998 1999static void 2000g_raid3_launch_provider(struct g_raid3_softc *sc) 2001{ 2002 struct g_provider *pp; 2003 2004 g_topology_assert(); 2005 2006 pp = g_new_providerf(sc->sc_geom, "raid3/%s", sc->sc_name); 2007 pp->mediasize = sc->sc_mediasize; 2008 pp->sectorsize = sc->sc_sectorsize; 2009 sc->sc_provider = pp; 2010 g_error_provider(pp, 0); 2011 G_RAID3_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 2012 pp->name); 2013 if (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED) 2014 g_raid3_sync_start(sc); 2015} 2016 2017static void 2018g_raid3_destroy_provider(struct g_raid3_softc *sc) 2019{ 2020 struct bio *bp; 2021 2022 g_topology_assert(); 2023 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2024 sc->sc_name)); 2025 2026 g_error_provider(sc->sc_provider, ENXIO); 2027 mtx_lock(&sc->sc_queue_mtx); 2028 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 2029 bioq_remove(&sc->sc_queue, bp); 2030 g_io_deliver(bp, ENXIO); 2031 } 2032 mtx_unlock(&sc->sc_queue_mtx); 2033 G_RAID3_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 2034 sc->sc_provider->name); 2035 sc->sc_provider->flags |= G_PF_WITHER; 2036 g_orphan_provider(sc->sc_provider, ENXIO); 2037 sc->sc_provider = NULL; 2038 if (sc->sc_syncdisk != NULL) 2039 g_raid3_sync_stop(sc, 1); 2040} 2041 2042static void 2043g_raid3_go(void *arg) 2044{ 2045 struct g_raid3_softc *sc; 2046 2047 sc = arg; 2048 G_RAID3_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2049 g_raid3_event_send(sc, 0, 2050 G_RAID3_EVENT_DONTWAIT | G_RAID3_EVENT_DEVICE); 2051} 2052 2053static u_int 2054g_raid3_determine_state(struct g_raid3_disk *disk) 2055{ 2056 struct g_raid3_softc *sc; 2057 u_int state; 2058 2059 sc = disk->d_softc; 2060 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2061 if ((disk->d_flags & 2062 G_RAID3_DISK_FLAG_SYNCHRONIZING) == 0) { 2063 /* Disk does not need synchronization. */ 2064 state = G_RAID3_DISK_STATE_ACTIVE; 2065 } else { 2066 if ((sc->sc_flags & 2067 G_RAID3_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2068 (disk->d_flags & 2069 G_RAID3_DISK_FLAG_FORCE_SYNC) != 0) { 2070 /* 2071 * We can start synchronization from 2072 * the stored offset. 2073 */ 2074 state = G_RAID3_DISK_STATE_SYNCHRONIZING; 2075 } else { 2076 state = G_RAID3_DISK_STATE_STALE; 2077 } 2078 } 2079 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2080 /* 2081 * Reset all synchronization data for this disk, 2082 * because if it even was synchronized, it was 2083 * synchronized to disks with different syncid. 2084 */ 2085 disk->d_flags |= G_RAID3_DISK_FLAG_SYNCHRONIZING; 2086 disk->d_sync.ds_offset = 0; 2087 disk->d_sync.ds_offset_done = 0; 2088 disk->d_sync.ds_syncid = sc->sc_syncid; 2089 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2090 (disk->d_flags & G_RAID3_DISK_FLAG_FORCE_SYNC) != 0) { 2091 state = G_RAID3_DISK_STATE_SYNCHRONIZING; 2092 } else { 2093 state = G_RAID3_DISK_STATE_STALE; 2094 } 2095 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2096 /* 2097 * Not good, NOT GOOD! 2098 * It means that device was started on stale disks 2099 * and more fresh disk just arrive. 2100 * If there were writes, device is fucked up, sorry. 2101 * I think the best choice here is don't touch 2102 * this disk and inform the user laudly. 2103 */ 2104 G_RAID3_DEBUG(0, "Device %s was started before the freshest " 2105 "disk (%s) arrives!! It will not be connected to the " 2106 "running device.", sc->sc_name, 2107 g_raid3_get_diskname(disk)); 2108 g_raid3_destroy_disk(disk); 2109 state = G_RAID3_DISK_STATE_NONE; 2110 /* Return immediately, because disk was destroyed. */ 2111 return (state); 2112 } 2113 G_RAID3_DEBUG(3, "State for %s disk: %s.", 2114 g_raid3_get_diskname(disk), g_raid3_disk_state2str(state)); 2115 return (state); 2116} 2117 2118/* 2119 * Update device state. 2120 */ 2121static void 2122g_raid3_update_device(struct g_raid3_softc *sc, boolean_t force) 2123{ 2124 struct g_raid3_disk *disk; 2125 u_int state; 2126 2127 g_topology_assert(); 2128 2129 switch (sc->sc_state) { 2130 case G_RAID3_DEVICE_STATE_STARTING: 2131 { 2132 u_int n, ndirty, ndisks, genid, syncid; 2133 2134 KASSERT(sc->sc_provider == NULL, 2135 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2136 /* 2137 * Are we ready? We are, if all disks are connected or 2138 * one disk is missing and 'force' is true. 2139 */ 2140 if (g_raid3_ndisks(sc, -1) + force == sc->sc_ndisks) { 2141 if (!force) 2142 callout_drain(&sc->sc_callout); 2143 } else { 2144 if (force) { 2145 /* 2146 * Timeout expired, so destroy device. 2147 */ 2148 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 2149 } 2150 return; 2151 } 2152 2153 /* 2154 * Find the biggest genid. 2155 */ 2156 genid = 0; 2157 for (n = 0; n < sc->sc_ndisks; n++) { 2158 disk = &sc->sc_disks[n]; 2159 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2160 continue; 2161 if (disk->d_genid > genid) 2162 genid = disk->d_genid; 2163 } 2164 sc->sc_genid = genid; 2165 /* 2166 * Remove all disks without the biggest genid. 2167 */ 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 G_RAID3_DEBUG(0, 2174 "Component %s (device %s) broken, skipping.", 2175 g_raid3_get_diskname(disk), sc->sc_name); 2176 g_raid3_destroy_disk(disk); 2177 } 2178 } 2179 2180 /* 2181 * There must be at least 'sc->sc_ndisks - 1' components 2182 * with the same syncid and without SYNCHRONIZING flag. 2183 */ 2184 2185 /* 2186 * Find the biggest syncid, number of valid components and 2187 * number of dirty components. 2188 */ 2189 ndirty = ndisks = syncid = 0; 2190 for (n = 0; n < sc->sc_ndisks; n++) { 2191 disk = &sc->sc_disks[n]; 2192 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2193 continue; 2194 if ((disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) != 0) 2195 ndirty++; 2196 if (disk->d_sync.ds_syncid > syncid) { 2197 syncid = disk->d_sync.ds_syncid; 2198 ndisks = 0; 2199 } else if (disk->d_sync.ds_syncid < syncid) { 2200 continue; 2201 } 2202 if ((disk->d_flags & 2203 G_RAID3_DISK_FLAG_SYNCHRONIZING) != 0) { 2204 continue; 2205 } 2206 ndisks++; 2207 } 2208 /* 2209 * Do we have enough valid components? 2210 */ 2211 if (ndisks + 1 < sc->sc_ndisks) { 2212 G_RAID3_DEBUG(0, 2213 "Device %s is broken, too few valid components.", 2214 sc->sc_name); 2215 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 2216 return; 2217 } 2218 /* 2219 * If there is one DIRTY component and all disks are present, 2220 * mark it for synchronization. If there is more than one DIRTY 2221 * component, mark parity component for synchronization. 2222 */ 2223 if (ndisks == sc->sc_ndisks && ndirty == 1) { 2224 for (n = 0; n < sc->sc_ndisks; n++) { 2225 disk = &sc->sc_disks[n]; 2226 if ((disk->d_flags & 2227 G_RAID3_DISK_FLAG_DIRTY) == 0) { 2228 continue; 2229 } 2230 disk->d_flags |= 2231 G_RAID3_DISK_FLAG_SYNCHRONIZING; 2232 } 2233 } else if (ndisks == sc->sc_ndisks && ndirty > 1) { 2234 disk = &sc->sc_disks[sc->sc_ndisks - 1]; 2235 disk->d_flags |= G_RAID3_DISK_FLAG_SYNCHRONIZING; 2236 } 2237 2238 sc->sc_syncid = syncid; 2239 if (force) { 2240 /* Remember to bump syncid on first write. */ 2241 sc->sc_bump_id |= G_RAID3_BUMP_SYNCID; 2242 } 2243 if (ndisks == sc->sc_ndisks) 2244 state = G_RAID3_DEVICE_STATE_COMPLETE; 2245 else /* if (ndisks == sc->sc_ndisks - 1) */ 2246 state = G_RAID3_DEVICE_STATE_DEGRADED; 2247 G_RAID3_DEBUG(1, "Device %s state changed from %s to %s.", 2248 sc->sc_name, g_raid3_device_state2str(sc->sc_state), 2249 g_raid3_device_state2str(state)); 2250 sc->sc_state = state; 2251 for (n = 0; n < sc->sc_ndisks; n++) { 2252 disk = &sc->sc_disks[n]; 2253 if (disk->d_state == G_RAID3_DISK_STATE_NODISK) 2254 continue; 2255 state = g_raid3_determine_state(disk); 2256 g_raid3_event_send(disk, state, G_RAID3_EVENT_DONTWAIT); 2257 if (state == G_RAID3_DISK_STATE_STALE) 2258 sc->sc_bump_id |= G_RAID3_BUMP_SYNCID; 2259 } 2260 break; 2261 } 2262 case G_RAID3_DEVICE_STATE_DEGRADED: 2263 /* 2264 * Genid need to be bumped immediately, so do it here. 2265 */ 2266 if ((sc->sc_bump_id & G_RAID3_BUMP_GENID) != 0) { 2267 sc->sc_bump_id &= ~G_RAID3_BUMP_GENID; 2268 g_raid3_bump_genid(sc); 2269 } 2270 2271 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NEW) > 0) 2272 return; 2273 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) < 2274 sc->sc_ndisks - 1) { 2275 if (sc->sc_provider != NULL) 2276 g_raid3_destroy_provider(sc); 2277 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 2278 return; 2279 } 2280 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) == 2281 sc->sc_ndisks) { 2282 state = G_RAID3_DEVICE_STATE_COMPLETE; 2283 G_RAID3_DEBUG(1, 2284 "Device %s state changed from %s to %s.", 2285 sc->sc_name, g_raid3_device_state2str(sc->sc_state), 2286 g_raid3_device_state2str(state)); 2287 sc->sc_state = state; 2288 } 2289 if (sc->sc_provider == NULL) 2290 g_raid3_launch_provider(sc); 2291 break; 2292 case G_RAID3_DEVICE_STATE_COMPLETE: 2293 /* 2294 * Genid need to be bumped immediately, so do it here. 2295 */ 2296 if ((sc->sc_bump_id & G_RAID3_BUMP_GENID) != 0) { 2297 sc->sc_bump_id &= ~G_RAID3_BUMP_GENID; 2298 g_raid3_bump_genid(sc); 2299 } 2300 2301 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NEW) > 0) 2302 return; 2303 KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) >= 2304 sc->sc_ndisks - 1, 2305 ("Too few ACTIVE components in COMPLETE state (device %s).", 2306 sc->sc_name)); 2307 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) == 2308 sc->sc_ndisks - 1) { 2309 state = G_RAID3_DEVICE_STATE_DEGRADED; 2310 G_RAID3_DEBUG(1, 2311 "Device %s state changed from %s to %s.", 2312 sc->sc_name, g_raid3_device_state2str(sc->sc_state), 2313 g_raid3_device_state2str(state)); 2314 sc->sc_state = state; 2315 } 2316 if (sc->sc_provider == NULL) 2317 g_raid3_launch_provider(sc); 2318 break; 2319 default: 2320 KASSERT(1 == 0, ("Wrong device state (%s, %s).", sc->sc_name, 2321 g_raid3_device_state2str(sc->sc_state))); 2322 break; 2323 } 2324} 2325 2326/* 2327 * Update disk state and device state if needed. 2328 */ 2329#define DISK_STATE_CHANGED() G_RAID3_DEBUG(1, \ 2330 "Disk %s state changed from %s to %s (device %s).", \ 2331 g_raid3_get_diskname(disk), \ 2332 g_raid3_disk_state2str(disk->d_state), \ 2333 g_raid3_disk_state2str(state), sc->sc_name) 2334static int 2335g_raid3_update_disk(struct g_raid3_disk *disk, u_int state) 2336{ 2337 struct g_raid3_softc *sc; 2338 2339 g_topology_assert(); 2340 2341 sc = disk->d_softc; 2342again: 2343 G_RAID3_DEBUG(3, "Changing disk %s state from %s to %s.", 2344 g_raid3_get_diskname(disk), g_raid3_disk_state2str(disk->d_state), 2345 g_raid3_disk_state2str(state)); 2346 switch (state) { 2347 case G_RAID3_DISK_STATE_NEW: 2348 /* 2349 * Possible scenarios: 2350 * 1. New disk arrive. 2351 */ 2352 /* Previous state should be NONE. */ 2353 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NONE, 2354 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2355 g_raid3_disk_state2str(disk->d_state))); 2356 DISK_STATE_CHANGED(); 2357 2358 disk->d_state = state; 2359 G_RAID3_DEBUG(0, "Device %s: provider %s detected.", 2360 sc->sc_name, g_raid3_get_diskname(disk)); 2361 if (sc->sc_state == G_RAID3_DEVICE_STATE_STARTING) 2362 break; 2363 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2364 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2365 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2366 g_raid3_device_state2str(sc->sc_state), 2367 g_raid3_get_diskname(disk), 2368 g_raid3_disk_state2str(disk->d_state))); 2369 state = g_raid3_determine_state(disk); 2370 if (state != G_RAID3_DISK_STATE_NONE) 2371 goto again; 2372 break; 2373 case G_RAID3_DISK_STATE_ACTIVE: 2374 /* 2375 * Possible scenarios: 2376 * 1. New disk does not need synchronization. 2377 * 2. Synchronization process finished successfully. 2378 */ 2379 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2380 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2381 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2382 g_raid3_device_state2str(sc->sc_state), 2383 g_raid3_get_diskname(disk), 2384 g_raid3_disk_state2str(disk->d_state))); 2385 /* Previous state should be NEW or SYNCHRONIZING. */ 2386 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW || 2387 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING, 2388 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2389 g_raid3_disk_state2str(disk->d_state))); 2390 DISK_STATE_CHANGED(); 2391 2392 if (disk->d_state == G_RAID3_DISK_STATE_NEW) 2393 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2394 else if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 2395 disk->d_flags &= ~G_RAID3_DISK_FLAG_SYNCHRONIZING; 2396 disk->d_flags &= ~G_RAID3_DISK_FLAG_FORCE_SYNC; 2397 g_raid3_sync_stop(sc, 0); 2398 } 2399 disk->d_state = state; 2400 disk->d_sync.ds_offset = 0; 2401 disk->d_sync.ds_offset_done = 0; 2402 g_raid3_update_access(disk); 2403 g_raid3_update_metadata(disk); 2404 G_RAID3_DEBUG(0, "Device %s: provider %s activated.", 2405 sc->sc_name, g_raid3_get_diskname(disk)); 2406 break; 2407 case G_RAID3_DISK_STATE_STALE: 2408 /* 2409 * Possible scenarios: 2410 * 1. Stale disk was connected. 2411 */ 2412 /* Previous state should be NEW. */ 2413 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW, 2414 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2415 g_raid3_disk_state2str(disk->d_state))); 2416 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2417 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2418 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2419 g_raid3_device_state2str(sc->sc_state), 2420 g_raid3_get_diskname(disk), 2421 g_raid3_disk_state2str(disk->d_state))); 2422 /* 2423 * STALE state is only possible if device is marked 2424 * NOAUTOSYNC. 2425 */ 2426 KASSERT((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOAUTOSYNC) != 0, 2427 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2428 g_raid3_device_state2str(sc->sc_state), 2429 g_raid3_get_diskname(disk), 2430 g_raid3_disk_state2str(disk->d_state))); 2431 DISK_STATE_CHANGED(); 2432 2433 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2434 disk->d_state = state; 2435 g_raid3_update_metadata(disk); 2436 G_RAID3_DEBUG(0, "Device %s: provider %s is stale.", 2437 sc->sc_name, g_raid3_get_diskname(disk)); 2438 break; 2439 case G_RAID3_DISK_STATE_SYNCHRONIZING: 2440 /* 2441 * Possible scenarios: 2442 * 1. Disk which needs synchronization was connected. 2443 */ 2444 /* Previous state should be NEW. */ 2445 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW, 2446 ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk), 2447 g_raid3_disk_state2str(disk->d_state))); 2448 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2449 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE, 2450 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2451 g_raid3_device_state2str(sc->sc_state), 2452 g_raid3_get_diskname(disk), 2453 g_raid3_disk_state2str(disk->d_state))); 2454 DISK_STATE_CHANGED(); 2455 2456 if (disk->d_state == G_RAID3_DISK_STATE_NEW) 2457 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2458 disk->d_state = state; 2459 if (sc->sc_provider != NULL) { 2460 g_raid3_sync_start(sc); 2461 g_raid3_update_metadata(disk); 2462 } 2463 break; 2464 case G_RAID3_DISK_STATE_DISCONNECTED: 2465 /* 2466 * Possible scenarios: 2467 * 1. Device wasn't running yet, but disk disappear. 2468 * 2. Disk was active and disapppear. 2469 * 3. Disk disappear during synchronization process. 2470 */ 2471 if (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED || 2472 sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) { 2473 /* 2474 * Previous state should be ACTIVE, STALE or 2475 * SYNCHRONIZING. 2476 */ 2477 KASSERT(disk->d_state == G_RAID3_DISK_STATE_ACTIVE || 2478 disk->d_state == G_RAID3_DISK_STATE_STALE || 2479 disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING, 2480 ("Wrong disk state (%s, %s).", 2481 g_raid3_get_diskname(disk), 2482 g_raid3_disk_state2str(disk->d_state))); 2483 } else if (sc->sc_state == G_RAID3_DEVICE_STATE_STARTING) { 2484 /* Previous state should be NEW. */ 2485 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW, 2486 ("Wrong disk state (%s, %s).", 2487 g_raid3_get_diskname(disk), 2488 g_raid3_disk_state2str(disk->d_state))); 2489 /* 2490 * Reset bumping syncid if disk disappeared in STARTING 2491 * state. 2492 */ 2493 if ((sc->sc_bump_id & G_RAID3_BUMP_SYNCID) != 0) 2494 sc->sc_bump_id &= ~G_RAID3_BUMP_SYNCID; 2495#ifdef INVARIANTS 2496 } else { 2497 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2498 sc->sc_name, 2499 g_raid3_device_state2str(sc->sc_state), 2500 g_raid3_get_diskname(disk), 2501 g_raid3_disk_state2str(disk->d_state))); 2502#endif 2503 } 2504 DISK_STATE_CHANGED(); 2505 G_RAID3_DEBUG(0, "Device %s: provider %s disconnected.", 2506 sc->sc_name, g_raid3_get_diskname(disk)); 2507 2508 g_raid3_destroy_disk(disk); 2509 break; 2510 default: 2511 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2512 break; 2513 } 2514 return (0); 2515} 2516#undef DISK_STATE_CHANGED 2517 2518int 2519g_raid3_read_metadata(struct g_consumer *cp, struct g_raid3_metadata *md) 2520{ 2521 struct g_provider *pp; 2522 u_char *buf; 2523 int error; 2524 2525 g_topology_assert(); 2526 2527 error = g_access(cp, 1, 0, 0); 2528 if (error != 0) 2529 return (error); 2530 pp = cp->provider; 2531 g_topology_unlock(); 2532 /* Metadata are stored on last sector. */ 2533 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2534 &error); 2535 g_topology_lock(); 2536 g_access(cp, -1, 0, 0); 2537 if (error != 0) { 2538 G_RAID3_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2539 cp->provider->name, error); 2540 if (buf != NULL) 2541 g_free(buf); 2542 return (error); 2543 } 2544 2545 /* Decode metadata. */ 2546 error = raid3_metadata_decode(buf, md); 2547 g_free(buf); 2548 if (strcmp(md->md_magic, G_RAID3_MAGIC) != 0) 2549 return (EINVAL); 2550 if (md->md_version > G_RAID3_VERSION) { 2551 G_RAID3_DEBUG(0, 2552 "Kernel module is too old to handle metadata from %s.", 2553 cp->provider->name); 2554 return (EINVAL); 2555 } 2556 if (error != 0) { 2557 G_RAID3_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2558 cp->provider->name); 2559 return (error); 2560 } 2561 2562 return (0); 2563} 2564 2565static int 2566g_raid3_check_metadata(struct g_raid3_softc *sc, struct g_provider *pp, 2567 struct g_raid3_metadata *md) 2568{ 2569 2570 if (md->md_no >= sc->sc_ndisks) { 2571 G_RAID3_DEBUG(1, "Invalid disk %s number (no=%u), skipping.", 2572 pp->name, md->md_no); 2573 return (EINVAL); 2574 } 2575 if (sc->sc_disks[md->md_no].d_state != G_RAID3_DISK_STATE_NODISK) { 2576 G_RAID3_DEBUG(1, "Disk %s (no=%u) already exists, skipping.", 2577 pp->name, md->md_no); 2578 return (EEXIST); 2579 } 2580 if (md->md_all != sc->sc_ndisks) { 2581 G_RAID3_DEBUG(1, 2582 "Invalid '%s' field on disk %s (device %s), skipping.", 2583 "md_all", pp->name, sc->sc_name); 2584 return (EINVAL); 2585 } 2586 if (md->md_mediasize != sc->sc_mediasize) { 2587 G_RAID3_DEBUG(1, 2588 "Invalid '%s' field on disk %s (device %s), skipping.", 2589 "md_mediasize", pp->name, sc->sc_name); 2590 return (EINVAL); 2591 } 2592 if ((md->md_mediasize % (sc->sc_ndisks - 1)) != 0) { 2593 G_RAID3_DEBUG(1, 2594 "Invalid '%s' field on disk %s (device %s), skipping.", 2595 "md_mediasize", pp->name, sc->sc_name); 2596 return (EINVAL); 2597 } 2598 if ((sc->sc_mediasize / (sc->sc_ndisks - 1)) > pp->mediasize) { 2599 G_RAID3_DEBUG(1, 2600 "Invalid size of disk %s (device %s), skipping.", pp->name, 2601 sc->sc_name); 2602 return (EINVAL); 2603 } 2604 if ((md->md_sectorsize / pp->sectorsize) < sc->sc_ndisks - 1) { 2605 G_RAID3_DEBUG(1, 2606 "Invalid '%s' field on disk %s (device %s), skipping.", 2607 "md_sectorsize", pp->name, sc->sc_name); 2608 return (EINVAL); 2609 } 2610 if (md->md_sectorsize != sc->sc_sectorsize) { 2611 G_RAID3_DEBUG(1, 2612 "Invalid '%s' field on disk %s (device %s), skipping.", 2613 "md_sectorsize", pp->name, sc->sc_name); 2614 return (EINVAL); 2615 } 2616 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2617 G_RAID3_DEBUG(1, 2618 "Invalid sector size of disk %s (device %s), skipping.", 2619 pp->name, sc->sc_name); 2620 return (EINVAL); 2621 } 2622 if ((md->md_mflags & ~G_RAID3_DEVICE_FLAG_MASK) != 0) { 2623 G_RAID3_DEBUG(1, 2624 "Invalid device flags on disk %s (device %s), skipping.", 2625 pp->name, sc->sc_name); 2626 return (EINVAL); 2627 } 2628 if ((md->md_mflags & G_RAID3_DEVICE_FLAG_VERIFY) != 0 && 2629 (md->md_mflags & G_RAID3_DEVICE_FLAG_ROUND_ROBIN) != 0) { 2630 /* 2631 * VERIFY and ROUND-ROBIN options are mutally exclusive. 2632 */ 2633 G_RAID3_DEBUG(1, "Both VERIFY and ROUND-ROBIN flags exist on " 2634 "disk %s (device %s), skipping.", pp->name, sc->sc_name); 2635 return (EINVAL); 2636 } 2637 if ((md->md_dflags & ~G_RAID3_DISK_FLAG_MASK) != 0) { 2638 G_RAID3_DEBUG(1, 2639 "Invalid disk flags on disk %s (device %s), skipping.", 2640 pp->name, sc->sc_name); 2641 return (EINVAL); 2642 } 2643 return (0); 2644} 2645 2646int 2647g_raid3_add_disk(struct g_raid3_softc *sc, struct g_provider *pp, 2648 struct g_raid3_metadata *md) 2649{ 2650 struct g_raid3_disk *disk; 2651 int error; 2652 2653 g_topology_assert(); 2654 G_RAID3_DEBUG(2, "Adding disk %s.", pp->name); 2655 2656 error = g_raid3_check_metadata(sc, pp, md); 2657 if (error != 0) 2658 return (error); 2659 if (sc->sc_state != G_RAID3_DEVICE_STATE_STARTING && 2660 md->md_genid < sc->sc_genid) { 2661 G_RAID3_DEBUG(0, "Component %s (device %s) broken, skipping.", 2662 pp->name, sc->sc_name); 2663 return (EINVAL); 2664 } 2665 disk = g_raid3_init_disk(sc, pp, md, &error); 2666 if (disk == NULL) 2667 return (error); 2668 error = g_raid3_event_send(disk, G_RAID3_DISK_STATE_NEW, 2669 G_RAID3_EVENT_WAIT); 2670 if (error != 0) 2671 return (error); 2672 if (md->md_version < G_RAID3_VERSION) { 2673 G_RAID3_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2674 pp->name, md->md_version, G_RAID3_VERSION); 2675 g_raid3_update_metadata(disk); 2676 } 2677 return (0); 2678} 2679 2680static int 2681g_raid3_access(struct g_provider *pp, int acr, int acw, int ace) 2682{ 2683 struct g_raid3_softc *sc; 2684 struct g_raid3_disk *disk; 2685 int dcr, dcw, dce; 2686 u_int n; 2687 2688 g_topology_assert(); 2689 G_RAID3_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2690 acw, ace); 2691 2692 dcr = pp->acr + acr; 2693 dcw = pp->acw + acw; 2694 dce = pp->ace + ace; 2695 2696 sc = pp->geom->softc; 2697 if (sc == NULL || 2698 g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) < sc->sc_ndisks - 1 || 2699 (sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) != 0) { 2700 if (acr <= 0 && acw <= 0 && ace <= 0) 2701 return (0); 2702 else 2703 return (ENXIO); 2704 } 2705 for (n = 0; n < sc->sc_ndisks; n++) { 2706 disk = &sc->sc_disks[n]; 2707 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) 2708 continue; 2709 /* 2710 * Mark disk as dirty on open and unmark on close. 2711 */ 2712 if (pp->acw == 0 && dcw > 0) { 2713 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as dirty.", 2714 g_raid3_get_diskname(disk), sc->sc_name); 2715 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY; 2716 g_raid3_update_metadata(disk); 2717 } else if (pp->acw > 0 && dcw == 0) { 2718 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as clean.", 2719 g_raid3_get_diskname(disk), sc->sc_name); 2720 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY; 2721 g_raid3_update_metadata(disk); 2722 } 2723 } 2724 return (0); 2725} 2726 2727static struct g_geom * 2728g_raid3_create(struct g_class *mp, const struct g_raid3_metadata *md) 2729{ 2730 struct g_raid3_softc *sc; 2731 struct g_geom *gp; 2732 int error, timeout; 2733 u_int n; 2734 2735 g_topology_assert(); 2736 G_RAID3_DEBUG(1, "Creating device %s (id=%u).", md->md_name, md->md_id); 2737 2738 /* One disk is minimum. */ 2739 if (md->md_all < 1) 2740 return (NULL); 2741 /* 2742 * Action geom. 2743 */ 2744 gp = g_new_geomf(mp, "%s", md->md_name); 2745 sc = malloc(sizeof(*sc), M_RAID3, M_WAITOK | M_ZERO); 2746 sc->sc_disks = malloc(sizeof(struct g_raid3_disk) * md->md_all, M_RAID3, 2747 M_WAITOK | M_ZERO); 2748 gp->start = g_raid3_start; 2749 gp->orphan = g_raid3_orphan; 2750 gp->access = g_raid3_access; 2751 gp->dumpconf = g_raid3_dumpconf; 2752 2753 sc->sc_id = md->md_id; 2754 sc->sc_mediasize = md->md_mediasize; 2755 sc->sc_sectorsize = md->md_sectorsize; 2756 sc->sc_ndisks = md->md_all; 2757 sc->sc_round_robin = 0; 2758 sc->sc_flags = md->md_mflags; 2759 sc->sc_bump_id = 0; 2760 sc->sc_idle = 0; 2761 for (n = 0; n < sc->sc_ndisks; n++) { 2762 sc->sc_disks[n].d_softc = sc; 2763 sc->sc_disks[n].d_no = n; 2764 sc->sc_disks[n].d_state = G_RAID3_DISK_STATE_NODISK; 2765 } 2766 bioq_init(&sc->sc_queue); 2767 mtx_init(&sc->sc_queue_mtx, "graid3:queue", NULL, MTX_DEF); 2768 TAILQ_INIT(&sc->sc_events); 2769 mtx_init(&sc->sc_events_mtx, "graid3:events", NULL, MTX_DEF); 2770 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2771 sc->sc_state = G_RAID3_DEVICE_STATE_STARTING; 2772 gp->softc = sc; 2773 sc->sc_geom = gp; 2774 sc->sc_provider = NULL; 2775 /* 2776 * Synchronization geom. 2777 */ 2778 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2779 gp->softc = sc; 2780 gp->orphan = g_raid3_orphan; 2781 sc->sc_sync.ds_geom = gp; 2782 sc->sc_zone_64k = uma_zcreate("gr3:64k", 65536, NULL, NULL, NULL, NULL, 2783 UMA_ALIGN_PTR, 0); 2784 uma_zone_set_max(sc->sc_zone_64k, g_raid3_n64k); 2785 sc->sc_zone_16k = uma_zcreate("gr3:16k", 16384, NULL, NULL, NULL, NULL, 2786 UMA_ALIGN_PTR, 0); 2787 uma_zone_set_max(sc->sc_zone_64k, g_raid3_n16k); 2788 sc->sc_zone_4k = uma_zcreate("gr3:4k", 4096, NULL, NULL, NULL, NULL, 2789 UMA_ALIGN_PTR, 0); 2790 uma_zone_set_max(sc->sc_zone_4k, g_raid3_n4k); 2791 error = kthread_create(g_raid3_worker, sc, &sc->sc_worker, 0, 0, 2792 "g_raid3 %s", md->md_name); 2793 if (error != 0) { 2794 G_RAID3_DEBUG(1, "Cannot create kernel thread for %s.", 2795 sc->sc_name); 2796 uma_zdestroy(sc->sc_zone_64k); 2797 uma_zdestroy(sc->sc_zone_16k); 2798 uma_zdestroy(sc->sc_zone_4k); 2799 g_destroy_geom(sc->sc_sync.ds_geom); 2800 mtx_destroy(&sc->sc_events_mtx); 2801 mtx_destroy(&sc->sc_queue_mtx); 2802 g_destroy_geom(sc->sc_geom); 2803 free(sc->sc_disks, M_RAID3); 2804 free(sc, M_RAID3); 2805 return (NULL); 2806 } 2807 2808 G_RAID3_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2809 2810 /* 2811 * Run timeout. 2812 */ 2813 timeout = atomic_load_acq_int(&g_raid3_timeout); 2814 callout_reset(&sc->sc_callout, timeout * hz, g_raid3_go, sc); 2815 return (sc->sc_geom); 2816} 2817 2818int 2819g_raid3_destroy(struct g_raid3_softc *sc, boolean_t force) 2820{ 2821 struct g_provider *pp; 2822 2823 g_topology_assert(); 2824 2825 if (sc == NULL) 2826 return (ENXIO); 2827 pp = sc->sc_provider; 2828 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2829 if (force) { 2830 G_RAID3_DEBUG(1, "Device %s is still open, so it " 2831 "can't be definitely removed.", pp->name); 2832 } else { 2833 G_RAID3_DEBUG(1, 2834 "Device %s is still open (r%dw%de%d).", pp->name, 2835 pp->acr, pp->acw, pp->ace); 2836 return (EBUSY); 2837 } 2838 } 2839 2840 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY; 2841 sc->sc_flags |= G_RAID3_DEVICE_FLAG_WAIT; 2842 g_topology_unlock(); 2843 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2844 mtx_lock(&sc->sc_queue_mtx); 2845 wakeup(sc); 2846 wakeup(&sc->sc_queue); 2847 mtx_unlock(&sc->sc_queue_mtx); 2848 G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2849 while (sc->sc_worker != NULL) 2850 tsleep(&sc->sc_worker, PRIBIO, "r3:destroy", hz / 5); 2851 G_RAID3_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2852 g_topology_lock(); 2853 g_raid3_destroy_device(sc); 2854 free(sc->sc_disks, M_RAID3); 2855 free(sc, M_RAID3); 2856 return (0); 2857} 2858 2859static void 2860g_raid3_taste_orphan(struct g_consumer *cp) 2861{ 2862 2863 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2864 cp->provider->name)); 2865} 2866 2867static struct g_geom * 2868g_raid3_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2869{ 2870 struct g_raid3_metadata md; 2871 struct g_raid3_softc *sc; 2872 struct g_consumer *cp; 2873 struct g_geom *gp; 2874 int error; 2875 2876 g_topology_assert(); 2877 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2878 G_RAID3_DEBUG(2, "Tasting %s.", pp->name); 2879 2880 gp = g_new_geomf(mp, "raid3:taste"); 2881 /* This orphan function should be never called. */ 2882 gp->orphan = g_raid3_taste_orphan; 2883 cp = g_new_consumer(gp); 2884 g_attach(cp, pp); 2885 error = g_raid3_read_metadata(cp, &md); 2886 g_detach(cp); 2887 g_destroy_consumer(cp); 2888 g_destroy_geom(gp); 2889 if (error != 0) 2890 return (NULL); 2891 gp = NULL; 2892 2893 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2894 return (NULL); 2895 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 2896 return (NULL); 2897 if (g_raid3_debug >= 2) 2898 raid3_metadata_dump(&md); 2899 2900 /* 2901 * Let's check if device already exists. 2902 */ 2903 sc = NULL; 2904 LIST_FOREACH(gp, &mp->geom, geom) { 2905 sc = gp->softc; 2906 if (sc == NULL) 2907 continue; 2908 if (sc->sc_sync.ds_geom == gp) 2909 continue; 2910 if (strcmp(md.md_name, sc->sc_name) != 0) 2911 continue; 2912 if (md.md_id != sc->sc_id) { 2913 G_RAID3_DEBUG(0, "Device %s already configured.", 2914 sc->sc_name); 2915 return (NULL); 2916 } 2917 break; 2918 } 2919 if (gp == NULL) { 2920 gp = g_raid3_create(mp, &md); 2921 if (gp == NULL) { 2922 G_RAID3_DEBUG(0, "Cannot create device %s.", 2923 md.md_name); 2924 return (NULL); 2925 } 2926 sc = gp->softc; 2927 } 2928 G_RAID3_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2929 error = g_raid3_add_disk(sc, pp, &md); 2930 if (error != 0) { 2931 G_RAID3_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2932 pp->name, gp->name, error); 2933 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NODISK) == 2934 sc->sc_ndisks) { 2935 g_raid3_destroy(sc, 1); 2936 } 2937 return (NULL); 2938 } 2939 return (gp); 2940} 2941 2942static int 2943g_raid3_destroy_geom(struct gctl_req *req __unused, struct g_class *mp __unused, 2944 struct g_geom *gp) 2945{ 2946 2947 return (g_raid3_destroy(gp->softc, 0)); 2948} 2949 2950static void 2951g_raid3_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2952 struct g_consumer *cp, struct g_provider *pp) 2953{ 2954 struct g_raid3_softc *sc; 2955 2956 g_topology_assert(); 2957 2958 sc = gp->softc; 2959 if (sc == NULL) 2960 return; 2961 /* Skip synchronization geom. */ 2962 if (gp == sc->sc_sync.ds_geom) 2963 return; 2964 if (pp != NULL) { 2965 /* Nothing here. */ 2966 } else if (cp != NULL) { 2967 struct g_raid3_disk *disk; 2968 2969 disk = cp->private; 2970 if (disk == NULL) 2971 return; 2972 sbuf_printf(sb, "%s<Type>", indent); 2973 if (disk->d_no == sc->sc_ndisks - 1) 2974 sbuf_printf(sb, "PARITY"); 2975 else 2976 sbuf_printf(sb, "DATA"); 2977 sbuf_printf(sb, "</Type>\n"); 2978 sbuf_printf(sb, "%s<Number>%u</Number>\n", indent, 2979 (u_int)disk->d_no); 2980 if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) { 2981 sbuf_printf(sb, "%s<Synchronized>", indent); 2982 if (disk->d_sync.ds_offset_done == 0) 2983 sbuf_printf(sb, "0%%"); 2984 else { 2985 sbuf_printf(sb, "%u%%", 2986 (u_int)((disk->d_sync.ds_offset_done * 100) / 2987 (sc->sc_mediasize / (sc->sc_ndisks - 1)))); 2988 } 2989 sbuf_printf(sb, "</Synchronized>\n"); 2990 } 2991 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2992 disk->d_sync.ds_syncid); 2993 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, disk->d_genid); 2994 sbuf_printf(sb, "%s<Flags>", indent); 2995 if (disk->d_flags == 0) 2996 sbuf_printf(sb, "NONE"); 2997 else { 2998 int first = 1; 2999 3000#define ADD_FLAG(flag, name) do { \ 3001 if ((disk->d_flags & (flag)) != 0) { \ 3002 if (!first) \ 3003 sbuf_printf(sb, ", "); \ 3004 else \ 3005 first = 0; \ 3006 sbuf_printf(sb, name); \ 3007 } \ 3008} while (0) 3009 ADD_FLAG(G_RAID3_DISK_FLAG_DIRTY, "DIRTY"); 3010 ADD_FLAG(G_RAID3_DISK_FLAG_HARDCODED, "HARDCODED"); 3011 ADD_FLAG(G_RAID3_DISK_FLAG_SYNCHRONIZING, 3012 "SYNCHRONIZING"); 3013 ADD_FLAG(G_RAID3_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3014#undef ADD_FLAG 3015 } 3016 sbuf_printf(sb, "</Flags>\n"); 3017 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3018 g_raid3_disk_state2str(disk->d_state)); 3019 } else { 3020 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3021 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3022 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3023 sbuf_printf(sb, "%s<Flags>", indent); 3024 if (sc->sc_flags == 0) 3025 sbuf_printf(sb, "NONE"); 3026 else { 3027 int first = 1; 3028 3029#define ADD_FLAG(flag, name) do { \ 3030 if ((sc->sc_flags & (flag)) != 0) { \ 3031 if (!first) \ 3032 sbuf_printf(sb, ", "); \ 3033 else \ 3034 first = 0; \ 3035 sbuf_printf(sb, name); \ 3036 } \ 3037} while (0) 3038 ADD_FLAG(G_RAID3_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3039 ADD_FLAG(G_RAID3_DEVICE_FLAG_ROUND_ROBIN, 3040 "ROUND-ROBIN"); 3041 ADD_FLAG(G_RAID3_DEVICE_FLAG_VERIFY, "VERIFY"); 3042#undef ADD_FLAG 3043 } 3044 sbuf_printf(sb, "</Flags>\n"); 3045 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3046 sc->sc_ndisks); 3047 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3048 g_raid3_device_state2str(sc->sc_state)); 3049 } 3050} 3051 3052static void 3053g_raid3_shutdown(void *arg, int howto) 3054{ 3055 struct g_class *mp; 3056 struct g_geom *gp, *gp2; 3057 3058 mp = arg; 3059 DROP_GIANT(); 3060 g_topology_lock(); 3061 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3062 if (gp->softc == NULL) 3063 continue; 3064 g_raid3_destroy(gp->softc, 1); 3065 } 3066 g_topology_unlock(); 3067 PICKUP_GIANT(); 3068#if 0 3069 tsleep(&gp, PRIBIO, "r3:shutdown", hz * 20); 3070#endif 3071} 3072 3073static void 3074g_raid3_init(struct g_class *mp) 3075{ 3076 3077 g_raid3_ehtag = EVENTHANDLER_REGISTER(shutdown_post_sync, 3078 g_raid3_shutdown, mp, SHUTDOWN_PRI_FIRST); 3079 if (g_raid3_ehtag == NULL) 3080 G_RAID3_DEBUG(0, "Warning! Cannot register shutdown event."); 3081} 3082 3083static void 3084g_raid3_fini(struct g_class *mp) 3085{ 3086 3087 if (g_raid3_ehtag == NULL) 3088 return; 3089 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid3_ehtag); 3090} 3091 3092DECLARE_GEOM_CLASS(g_raid3_class, g_raid3); 3093