sys/geom/raid3/g_raid3.c

   1 /*-
   2  * Copyright (c) 2004-2006 Pawel Jakub Dawidek <pjd@FreeBSD.org>
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  *
  14  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
  15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
  18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  24  * SUCH DAMAGE.
  25  */
  26
  27 #include <sys/cdefs.h>
  28 __FBSDID("$FreeBSD$");
  29
  30 #include <sys/param.h>
  31 #include <sys/systm.h>
  32 #include <sys/kernel.h>
  33 #include <sys/module.h>
  34 #include <sys/limits.h>
  35 #include <sys/lock.h>
  36 #include <sys/mutex.h>
  37 #include <sys/bio.h>
  38 #include <sys/sbuf.h>
  39 #include <sys/sysctl.h>
  40 #include <sys/malloc.h>
  41 #include <sys/eventhandler.h>
  42 #include <vm/uma.h>
  43 #include <geom/geom.h>
  44 #include <sys/proc.h>
  45 #include <sys/kthread.h>
  46 #include <sys/sched.h>
  47 #include <geom/raid3/g_raid3.h>
  48
  49 FEATURE(geom_raid3, "GEOM RAID-3 functionality");
  50
  51 static MALLOC_DEFINE(M_RAID3, "raid3_data", "GEOM_RAID3 Data");
  52
  53 SYSCTL_DECL(_kern_geom);
  54 static SYSCTL_NODE(_kern_geom, OID_AUTO, raid3, CTLFLAG_RW, 0,
  55     "GEOM_RAID3 stuff");
  56 u_int g_raid3_debug = 0;
  57 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, debug, CTLFLAG_RWTUN, &g_raid3_debug, 0,
  58     "Debug level");
  59 static u_int g_raid3_timeout = 4;
  60 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, timeout, CTLFLAG_RWTUN, &g_raid3_timeout,
  61     0, "Time to wait on all raid3 components");
  62 static u_int g_raid3_idletime = 5;
  63 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, idletime, CTLFLAG_RWTUN,
  64     &g_raid3_idletime, 0, "Mark components as clean when idling");
  65 static u_int g_raid3_disconnect_on_failure = 1;
  66 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN,
  67     &g_raid3_disconnect_on_failure, 0, "Disconnect component on I/O failure.");
  68 static u_int g_raid3_syncreqs = 2;
  69 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, sync_requests, CTLFLAG_RDTUN,
  70     &g_raid3_syncreqs, 0, "Parallel synchronization I/O requests.");
  71 static u_int g_raid3_use_malloc = 0;
  72 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, use_malloc, CTLFLAG_RDTUN,
  73     &g_raid3_use_malloc, 0, "Use malloc(9) instead of uma(9).");
  74
  75 static u_int g_raid3_n64k = 50;
  76 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n64k, CTLFLAG_RDTUN, &g_raid3_n64k, 0,
  77     "Maximum number of 64kB allocations");
  78 static u_int g_raid3_n16k = 200;
  79 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n16k, CTLFLAG_RDTUN, &g_raid3_n16k, 0,
  80     "Maximum number of 16kB allocations");
  81 static u_int g_raid3_n4k = 1200;
  82 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n4k, CTLFLAG_RDTUN, &g_raid3_n4k, 0,
  83     "Maximum number of 4kB allocations");
  84
  85 static SYSCTL_NODE(_kern_geom_raid3, OID_AUTO, stat, CTLFLAG_RW, 0,
  86     "GEOM_RAID3 statistics");
  87 static u_int g_raid3_parity_mismatch = 0;
  88 SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, parity_mismatch, CTLFLAG_RD,
  89     &g_raid3_parity_mismatch, 0, "Number of failures in VERIFY mode");
  90
  91 #define MSLEEP(ident, mtx, priority, wmesg, timeout)    do {            \
  92         G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, (ident));        \
  93         msleep((ident), (mtx), (priority), (wmesg), (timeout));         \
  94         G_RAID3_DEBUG(4, "%s: Woken up %p.", __func__, (ident));        \
  95 } while (0)
  96
  97 static eventhandler_tag g_raid3_post_sync = NULL;
  98 static int g_raid3_shutdown = 0;
  99
 100 static int g_raid3_destroy_geom(struct gctl_req *req, struct g_class *mp,
 101     struct g_geom *gp);
 102 static g_taste_t g_raid3_taste;
 103 static void g_raid3_init(struct g_class *mp);
 104 static void g_raid3_fini(struct g_class *mp);
 105
 106 struct g_class g_raid3_class = {
 107         .name = G_RAID3_CLASS_NAME,
 108         .version = G_VERSION,
 109         .ctlreq = g_raid3_config,
 110         .taste = g_raid3_taste,
 111         .destroy_geom = g_raid3_destroy_geom,
 112         .init = g_raid3_init,
 113         .fini = g_raid3_fini
 114 };
 115
 116
 117 static void g_raid3_destroy_provider(struct g_raid3_softc *sc);
 118 static int g_raid3_update_disk(struct g_raid3_disk *disk, u_int state);
 119 static void g_raid3_update_device(struct g_raid3_softc *sc, boolean_t force);
 120 static void g_raid3_dumpconf(struct sbuf *sb, const char *indent,
 121     struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp);
 122 static void g_raid3_sync_stop(struct g_raid3_softc *sc, int type);
 123 static int g_raid3_register_request(struct bio *pbp);
 124 static void g_raid3_sync_release(struct g_raid3_softc *sc);
 125
 126
 127 static const char *
 128 g_raid3_disk_state2str(int state)
 129 {
 130
 131         switch (state) {
 132         case G_RAID3_DISK_STATE_NODISK:
 133                 return ("NODISK");
 134         case G_RAID3_DISK_STATE_NONE:
 135                 return ("NONE");
 136         case G_RAID3_DISK_STATE_NEW:
 137                 return ("NEW");
 138         case G_RAID3_DISK_STATE_ACTIVE:
 139                 return ("ACTIVE");
 140         case G_RAID3_DISK_STATE_STALE:
 141                 return ("STALE");
 142         case G_RAID3_DISK_STATE_SYNCHRONIZING:
 143                 return ("SYNCHRONIZING");
 144         case G_RAID3_DISK_STATE_DISCONNECTED:
 145                 return ("DISCONNECTED");
 146         default:
 147                 return ("INVALID");
 148         }
 149 }
 150
 151 static const char *
 152 g_raid3_device_state2str(int state)
 153 {
 154
 155         switch (state) {
 156         case G_RAID3_DEVICE_STATE_STARTING:
 157                 return ("STARTING");
 158         case G_RAID3_DEVICE_STATE_DEGRADED:
 159                 return ("DEGRADED");
 160         case G_RAID3_DEVICE_STATE_COMPLETE:
 161                 return ("COMPLETE");
 162         default:
 163                 return ("INVALID");
 164         }
 165 }
 166
 167 const char *
 168 g_raid3_get_diskname(struct g_raid3_disk *disk)
 169 {
 170
 171         if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL)
 172                 return ("[unknown]");
 173         return (disk->d_name);
 174 }
 175
 176 static void *
 177 g_raid3_alloc(struct g_raid3_softc *sc, size_t size, int flags)
 178 {
 179         void *ptr;
 180         enum g_raid3_zones zone;
 181
 182         if (g_raid3_use_malloc ||
 183             (zone = g_raid3_zone(size)) == G_RAID3_NUM_ZONES)
 184                 ptr = malloc(size, M_RAID3, flags);
 185         else {
 186                 ptr = uma_zalloc_arg(sc->sc_zones[zone].sz_zone,
 187                    &sc->sc_zones[zone], flags);
 188                 sc->sc_zones[zone].sz_requested++;
 189                 if (ptr == NULL)
 190                         sc->sc_zones[zone].sz_failed++;
 191         }
 192         return (ptr);
 193 }
 194
 195 static void
 196 g_raid3_free(struct g_raid3_softc *sc, void *ptr, size_t size)
 197 {
 198         enum g_raid3_zones zone;
 199
 200         if (g_raid3_use_malloc ||
 201             (zone = g_raid3_zone(size)) == G_RAID3_NUM_ZONES)
 202                 free(ptr, M_RAID3);
 203         else {
 204                 uma_zfree_arg(sc->sc_zones[zone].sz_zone,
 205                     ptr, &sc->sc_zones[zone]);
 206         }
 207 }
 208
 209 static int
 210 g_raid3_uma_ctor(void *mem, int size, void *arg, int flags)
 211 {
 212         struct g_raid3_zone *sz = arg;
 213
 214         if (sz->sz_max > 0 && sz->sz_inuse == sz->sz_max)
 215                 return (ENOMEM);
 216         sz->sz_inuse++;
 217         return (0);
 218 }
 219
 220 static void
 221 g_raid3_uma_dtor(void *mem, int size, void *arg)
 222 {
 223         struct g_raid3_zone *sz = arg;
 224
 225         sz->sz_inuse--;
 226 }
 227
 228 #define g_raid3_xor(src, dst, size)                                     \
 229         _g_raid3_xor((uint64_t *)(src),                                 \
 230             (uint64_t *)(dst), (size_t)size)
 231 static void
 232 _g_raid3_xor(uint64_t *src, uint64_t *dst, size_t size)
 233 {
 234
 235         KASSERT((size % 128) == 0, ("Invalid size: %zu.", size));
 236         for (; size > 0; size -= 128) {
 237                 *dst++ ^= (*src++);
 238                 *dst++ ^= (*src++);
 239                 *dst++ ^= (*src++);
 240                 *dst++ ^= (*src++);
 241                 *dst++ ^= (*src++);
 242                 *dst++ ^= (*src++);
 243                 *dst++ ^= (*src++);
 244                 *dst++ ^= (*src++);
 245                 *dst++ ^= (*src++);
 246                 *dst++ ^= (*src++);
 247                 *dst++ ^= (*src++);
 248                 *dst++ ^= (*src++);
 249                 *dst++ ^= (*src++);
 250                 *dst++ ^= (*src++);
 251                 *dst++ ^= (*src++);
 252                 *dst++ ^= (*src++);
 253         }
 254 }
 255
 256 static int
 257 g_raid3_is_zero(struct bio *bp)
 258 {
 259         static const uint64_t zeros[] = {
 260             0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
 261         };
 262         u_char *addr;
 263         ssize_t size;
 264
 265         size = bp->bio_length;
 266         addr = (u_char *)bp->bio_data;
 267         for (; size > 0; size -= sizeof(zeros), addr += sizeof(zeros)) {
 268                 if (bcmp(addr, zeros, sizeof(zeros)) != 0)
 269                         return (0);
 270         }
 271         return (1);
 272 }
 273
 274 /*
 275  * --- Events handling functions ---
 276  * Events in geom_raid3 are used to maintain disks and device status
 277  * from one thread to simplify locking.
 278  */
 279 static void
 280 g_raid3_event_free(struct g_raid3_event *ep)
 281 {
 282
 283         free(ep, M_RAID3);
 284 }
 285
 286 int
 287 g_raid3_event_send(void *arg, int state, int flags)
 288 {
 289         struct g_raid3_softc *sc;
 290         struct g_raid3_disk *disk;
 291         struct g_raid3_event *ep;
 292         int error;
 293
 294         ep = malloc(sizeof(*ep), M_RAID3, M_WAITOK);
 295         G_RAID3_DEBUG(4, "%s: Sending event %p.", __func__, ep);
 296         if ((flags & G_RAID3_EVENT_DEVICE) != 0) {
 297                 disk = NULL;
 298                 sc = arg;
 299         } else {
 300                 disk = arg;
 301                 sc = disk->d_softc;
 302         }
 303         ep->e_disk = disk;
 304         ep->e_state = state;
 305         ep->e_flags = flags;
 306         ep->e_error = 0;
 307         mtx_lock(&sc->sc_events_mtx);
 308         TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next);
 309         mtx_unlock(&sc->sc_events_mtx);
 310         G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc);
 311         mtx_lock(&sc->sc_queue_mtx);
 312         wakeup(sc);
 313         wakeup(&sc->sc_queue);
 314         mtx_unlock(&sc->sc_queue_mtx);
 315         if ((flags & G_RAID3_EVENT_DONTWAIT) != 0)
 316                 return (0);
 317         sx_assert(&sc->sc_lock, SX_XLOCKED);
 318         G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, ep);
 319         sx_xunlock(&sc->sc_lock);
 320         while ((ep->e_flags & G_RAID3_EVENT_DONE) == 0) {
 321                 mtx_lock(&sc->sc_events_mtx);
 322                 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "r3:event",
 323                     hz * 5);
 324         }
 325         error = ep->e_error;
 326         g_raid3_event_free(ep);
 327         sx_xlock(&sc->sc_lock);
 328         return (error);
 329 }
 330
 331 static struct g_raid3_event *
 332 g_raid3_event_get(struct g_raid3_softc *sc)
 333 {
 334         struct g_raid3_event *ep;
 335
 336         mtx_lock(&sc->sc_events_mtx);
 337         ep = TAILQ_FIRST(&sc->sc_events);
 338         mtx_unlock(&sc->sc_events_mtx);
 339         return (ep);
 340 }
 341
 342 static void
 343 g_raid3_event_remove(struct g_raid3_softc *sc, struct g_raid3_event *ep)
 344 {
 345
 346         mtx_lock(&sc->sc_events_mtx);
 347         TAILQ_REMOVE(&sc->sc_events, ep, e_next);
 348         mtx_unlock(&sc->sc_events_mtx);
 349 }
 350
 351 static void
 352 g_raid3_event_cancel(struct g_raid3_disk *disk)
 353 {
 354         struct g_raid3_softc *sc;
 355         struct g_raid3_event *ep, *tmpep;
 356
 357         sc = disk->d_softc;
 358         sx_assert(&sc->sc_lock, SX_XLOCKED);
 359
 360         mtx_lock(&sc->sc_events_mtx);
 361         TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) {
 362                 if ((ep->e_flags & G_RAID3_EVENT_DEVICE) != 0)
 363                         continue;
 364                 if (ep->e_disk != disk)
 365                         continue;
 366                 TAILQ_REMOVE(&sc->sc_events, ep, e_next);
 367                 if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0)
 368                         g_raid3_event_free(ep);
 369                 else {
 370                         ep->e_error = ECANCELED;
 371                         wakeup(ep);
 372                 }
 373         }
 374         mtx_unlock(&sc->sc_events_mtx);
 375 }
 376
 377 /*
 378  * Return the number of disks in the given state.
 379  * If state is equal to -1, count all connected disks.
 380  */
 381 u_int
 382 g_raid3_ndisks(struct g_raid3_softc *sc, int state)
 383 {
 384         struct g_raid3_disk *disk;
 385         u_int n, ndisks;
 386
 387         sx_assert(&sc->sc_lock, SX_LOCKED);
 388
 389         for (n = ndisks = 0; n < sc->sc_ndisks; n++) {
 390                 disk = &sc->sc_disks[n];
 391                 if (disk->d_state == G_RAID3_DISK_STATE_NODISK)
 392                         continue;
 393                 if (state == -1 || disk->d_state == state)
 394                         ndisks++;
 395         }
 396         return (ndisks);
 397 }
 398
 399 static u_int
 400 g_raid3_nrequests(struct g_raid3_softc *sc, struct g_consumer *cp)
 401 {
 402         struct bio *bp;
 403         u_int nreqs = 0;
 404
 405         mtx_lock(&sc->sc_queue_mtx);
 406         TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
 407                 if (bp->bio_from == cp)
 408                         nreqs++;
 409         }
 410         mtx_unlock(&sc->sc_queue_mtx);
 411         return (nreqs);
 412 }
 413
 414 static int
 415 g_raid3_is_busy(struct g_raid3_softc *sc, struct g_consumer *cp)
 416 {
 417
 418         if (cp->index > 0) {
 419                 G_RAID3_DEBUG(2,
 420                     "I/O requests for %s exist, can't destroy it now.",
 421                     cp->provider->name);
 422                 return (1);
 423         }
 424         if (g_raid3_nrequests(sc, cp) > 0) {
 425                 G_RAID3_DEBUG(2,
 426                     "I/O requests for %s in queue, can't destroy it now.",
 427                     cp->provider->name);
 428                 return (1);
 429         }
 430         return (0);
 431 }
 432
 433 static void
 434 g_raid3_destroy_consumer(void *arg, int flags __unused)
 435 {
 436         struct g_consumer *cp;
 437
 438         g_topology_assert();
 439
 440         cp = arg;
 441         G_RAID3_DEBUG(1, "Consumer %s destroyed.", cp->provider->name);
 442         g_detach(cp);
 443         g_destroy_consumer(cp);
 444 }
 445
 446 static void
 447 g_raid3_kill_consumer(struct g_raid3_softc *sc, struct g_consumer *cp)
 448 {
 449         struct g_provider *pp;
 450         int retaste_wait;
 451
 452         g_topology_assert();
 453
 454         cp->private = NULL;
 455         if (g_raid3_is_busy(sc, cp))
 456                 return;
 457         G_RAID3_DEBUG(2, "Consumer %s destroyed.", cp->provider->name);
 458         pp = cp->provider;
 459         retaste_wait = 0;
 460         if (cp->acw == 1) {
 461                 if ((pp->geom->flags & G_GEOM_WITHER) == 0)
 462                         retaste_wait = 1;
 463         }
 464         G_RAID3_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr,
 465             -cp->acw, -cp->ace, 0);
 466         if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
 467                 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
 468         if (retaste_wait) {
 469                 /*
 470                  * After retaste event was send (inside g_access()), we can send
 471                  * event to detach and destroy consumer.
 472                  * A class, which has consumer to the given provider connected
 473                  * will not receive retaste event for the provider.
 474                  * This is the way how I ignore retaste events when I close
 475                  * consumers opened for write: I detach and destroy consumer
 476                  * after retaste event is sent.
 477                  */
 478                 g_post_event(g_raid3_destroy_consumer, cp, M_WAITOK, NULL);
 479                 return;
 480         }
 481         G_RAID3_DEBUG(1, "Consumer %s destroyed.", pp->name);
 482         g_detach(cp);
 483         g_destroy_consumer(cp);
 484 }
 485
 486 static int
 487 g_raid3_connect_disk(struct g_raid3_disk *disk, struct g_provider *pp)
 488 {
 489         struct g_consumer *cp;
 490         int error;
 491
 492         g_topology_assert_not();
 493         KASSERT(disk->d_consumer == NULL,
 494             ("Disk already connected (device %s).", disk->d_softc->sc_name));
 495
 496         g_topology_lock();
 497         cp = g_new_consumer(disk->d_softc->sc_geom);
 498         error = g_attach(cp, pp);
 499         if (error != 0) {
 500                 g_destroy_consumer(cp);
 501                 g_topology_unlock();
 502                 return (error);
 503         }
 504         error = g_access(cp, 1, 1, 1);
 505                 g_topology_unlock();
 506         if (error != 0) {
 507                 g_detach(cp);
 508                 g_destroy_consumer(cp);
 509                 G_RAID3_DEBUG(0, "Cannot open consumer %s (error=%d).",
 510                     pp->name, error);
 511                 return (error);
 512         }
 513         disk->d_consumer = cp;
 514         disk->d_consumer->private = disk;
 515         disk->d_consumer->index = 0;
 516         G_RAID3_DEBUG(2, "Disk %s connected.", g_raid3_get_diskname(disk));
 517         return (0);
 518 }
 519
 520 static void
 521 g_raid3_disconnect_consumer(struct g_raid3_softc *sc, struct g_consumer *cp)
 522 {
 523
 524         g_topology_assert();
 525
 526         if (cp == NULL)
 527                 return;
 528         if (cp->provider != NULL)
 529                 g_raid3_kill_consumer(sc, cp);
 530         else
 531                 g_destroy_consumer(cp);
 532 }
 533
 534 /*
 535  * Initialize disk. This means allocate memory, create consumer, attach it
 536  * to the provider and open access (r1w1e1) to it.
 537  */
 538 static struct g_raid3_disk *
 539 g_raid3_init_disk(struct g_raid3_softc *sc, struct g_provider *pp,
 540     struct g_raid3_metadata *md, int *errorp)
 541 {
 542         struct g_raid3_disk *disk;
 543         int error;
 544
 545         disk = &sc->sc_disks[md->md_no];
 546         error = g_raid3_connect_disk(disk, pp);
 547         if (error != 0) {
 548                 if (errorp != NULL)
 549                         *errorp = error;
 550                 return (NULL);
 551         }
 552         disk->d_state = G_RAID3_DISK_STATE_NONE;
 553         disk->d_flags = md->md_dflags;
 554         if (md->md_provider[0] != '\0')
 555                 disk->d_flags |= G_RAID3_DISK_FLAG_HARDCODED;
 556         disk->d_sync.ds_consumer = NULL;
 557         disk->d_sync.ds_offset = md->md_sync_offset;
 558         disk->d_sync.ds_offset_done = md->md_sync_offset;
 559         disk->d_genid = md->md_genid;
 560         disk->d_sync.ds_syncid = md->md_syncid;
 561         if (errorp != NULL)
 562                 *errorp = 0;
 563         return (disk);
 564 }
 565
 566 static void
 567 g_raid3_destroy_disk(struct g_raid3_disk *disk)
 568 {
 569         struct g_raid3_softc *sc;
 570
 571         g_topology_assert_not();
 572         sc = disk->d_softc;
 573         sx_assert(&sc->sc_lock, SX_XLOCKED);
 574
 575         if (disk->d_state == G_RAID3_DISK_STATE_NODISK)
 576                 return;
 577         g_raid3_event_cancel(disk);
 578         switch (disk->d_state) {
 579         case G_RAID3_DISK_STATE_SYNCHRONIZING:
 580                 if (sc->sc_syncdisk != NULL)
 581                         g_raid3_sync_stop(sc, 1);
 582                 /* FALLTHROUGH */
 583         case G_RAID3_DISK_STATE_NEW:
 584         case G_RAID3_DISK_STATE_STALE:
 585         case G_RAID3_DISK_STATE_ACTIVE:
 586                 g_topology_lock();
 587                 g_raid3_disconnect_consumer(sc, disk->d_consumer);
 588                 g_topology_unlock();
 589                 disk->d_consumer = NULL;
 590                 break;
 591         default:
 592                 KASSERT(0 == 1, ("Wrong disk state (%s, %s).",
 593                     g_raid3_get_diskname(disk),
 594                     g_raid3_disk_state2str(disk->d_state)));
 595         }
 596         disk->d_state = G_RAID3_DISK_STATE_NODISK;
 597 }
 598
 599 static void
 600 g_raid3_destroy_device(struct g_raid3_softc *sc)
 601 {
 602         struct g_raid3_event *ep;
 603         struct g_raid3_disk *disk;
 604         struct g_geom *gp;
 605         struct g_consumer *cp;
 606         u_int n;
 607
 608         g_topology_assert_not();
 609         sx_assert(&sc->sc_lock, SX_XLOCKED);
 610
 611         gp = sc->sc_geom;
 612         if (sc->sc_provider != NULL)
 613                 g_raid3_destroy_provider(sc);
 614         for (n = 0; n < sc->sc_ndisks; n++) {
 615                 disk = &sc->sc_disks[n];
 616                 if (disk->d_state != G_RAID3_DISK_STATE_NODISK) {
 617                         disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY;
 618                         g_raid3_update_metadata(disk);
 619                         g_raid3_destroy_disk(disk);
 620                 }
 621         }
 622         while ((ep = g_raid3_event_get(sc)) != NULL) {
 623                 g_raid3_event_remove(sc, ep);
 624                 if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0)
 625                         g_raid3_event_free(ep);
 626                 else {
 627                         ep->e_error = ECANCELED;
 628                         ep->e_flags |= G_RAID3_EVENT_DONE;
 629                         G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, ep);
 630                         mtx_lock(&sc->sc_events_mtx);
 631                         wakeup(ep);
 632                         mtx_unlock(&sc->sc_events_mtx);
 633                 }
 634         }
 635         callout_drain(&sc->sc_callout);
 636         cp = LIST_FIRST(&sc->sc_sync.ds_geom->consumer);
 637         g_topology_lock();
 638         if (cp != NULL)
 639                 g_raid3_disconnect_consumer(sc, cp);
 640         g_wither_geom(sc->sc_sync.ds_geom, ENXIO);
 641         G_RAID3_DEBUG(0, "Device %s destroyed.", gp->name);
 642         g_wither_geom(gp, ENXIO);
 643         g_topology_unlock();
 644         if (!g_raid3_use_malloc) {
 645                 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_64K].sz_zone);
 646                 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_16K].sz_zone);
 647                 uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_4K].sz_zone);
 648         }
 649         mtx_destroy(&sc->sc_queue_mtx);
 650         mtx_destroy(&sc->sc_events_mtx);
 651         sx_xunlock(&sc->sc_lock);
 652         sx_destroy(&sc->sc_lock);
 653 }
 654
 655 static void
 656 g_raid3_orphan(struct g_consumer *cp)
 657 {
 658         struct g_raid3_disk *disk;
 659
 660         g_topology_assert();
 661
 662         disk = cp->private;
 663         if (disk == NULL)
 664                 return;
 665         disk->d_softc->sc_bump_id = G_RAID3_BUMP_SYNCID;
 666         g_raid3_event_send(disk, G_RAID3_DISK_STATE_DISCONNECTED,
 667             G_RAID3_EVENT_DONTWAIT);
 668 }
 669
 670 static int
 671 g_raid3_write_metadata(struct g_raid3_disk *disk, struct g_raid3_metadata *md)
 672 {
 673         struct g_raid3_softc *sc;
 674         struct g_consumer *cp;
 675         off_t offset, length;
 676         u_char *sector;
 677         int error = 0;
 678
 679         g_topology_assert_not();
 680         sc = disk->d_softc;
 681         sx_assert(&sc->sc_lock, SX_LOCKED);
 682
 683         cp = disk->d_consumer;
 684         KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name));
 685         KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name));
 686         KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
 687             ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr,
 688             cp->acw, cp->ace));
 689         length = cp->provider->sectorsize;
 690         offset = cp->provider->mediasize - length;
 691         sector = malloc((size_t)length, M_RAID3, M_WAITOK | M_ZERO);
 692         if (md != NULL)
 693                 raid3_metadata_encode(md, sector);
 694         error = g_write_data(cp, offset, sector, length);
 695         free(sector, M_RAID3);
 696         if (error != 0) {
 697                 if ((disk->d_flags & G_RAID3_DISK_FLAG_BROKEN) == 0) {
 698                         G_RAID3_DEBUG(0, "Cannot write metadata on %s "
 699                             "(device=%s, error=%d).",
 700                             g_raid3_get_diskname(disk), sc->sc_name, error);
 701                         disk->d_flags |= G_RAID3_DISK_FLAG_BROKEN;
 702                 } else {
 703                         G_RAID3_DEBUG(1, "Cannot write metadata on %s "
 704                             "(device=%s, error=%d).",
 705                             g_raid3_get_diskname(disk), sc->sc_name, error);
 706                 }
 707                 if (g_raid3_disconnect_on_failure &&
 708                     sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) {
 709                         sc->sc_bump_id |= G_RAID3_BUMP_GENID;
 710                         g_raid3_event_send(disk,
 711                             G_RAID3_DISK_STATE_DISCONNECTED,
 712                             G_RAID3_EVENT_DONTWAIT);
 713                 }
 714         }
 715         return (error);
 716 }
 717
 718 int
 719 g_raid3_clear_metadata(struct g_raid3_disk *disk)
 720 {
 721         int error;
 722
 723         g_topology_assert_not();
 724         sx_assert(&disk->d_softc->sc_lock, SX_LOCKED);
 725
 726         error = g_raid3_write_metadata(disk, NULL);
 727         if (error == 0) {
 728                 G_RAID3_DEBUG(2, "Metadata on %s cleared.",
 729                     g_raid3_get_diskname(disk));
 730         } else {
 731                 G_RAID3_DEBUG(0,
 732                     "Cannot clear metadata on disk %s (error=%d).",
 733                     g_raid3_get_diskname(disk), error);
 734         }
 735         return (error);
 736 }
 737
 738 void
 739 g_raid3_fill_metadata(struct g_raid3_disk *disk, struct g_raid3_metadata *md)
 740 {
 741         struct g_raid3_softc *sc;
 742         struct g_provider *pp;
 743
 744         sc = disk->d_softc;
 745         strlcpy(md->md_magic, G_RAID3_MAGIC, sizeof(md->md_magic));
 746         md->md_version = G_RAID3_VERSION;
 747         strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name));
 748         md->md_id = sc->sc_id;
 749         md->md_all = sc->sc_ndisks;
 750         md->md_genid = sc->sc_genid;
 751         md->md_mediasize = sc->sc_mediasize;
 752         md->md_sectorsize = sc->sc_sectorsize;
 753         md->md_mflags = (sc->sc_flags & G_RAID3_DEVICE_FLAG_MASK);
 754         md->md_no = disk->d_no;
 755         md->md_syncid = disk->d_sync.ds_syncid;
 756         md->md_dflags = (disk->d_flags & G_RAID3_DISK_FLAG_MASK);
 757         if (disk->d_state != G_RAID3_DISK_STATE_SYNCHRONIZING)
 758                 md->md_sync_offset = 0;
 759         else {
 760                 md->md_sync_offset =
 761                     disk->d_sync.ds_offset_done / (sc->sc_ndisks - 1);
 762         }
 763         if (disk->d_consumer != NULL && disk->d_consumer->provider != NULL)
 764                 pp = disk->d_consumer->provider;
 765         else
 766                 pp = NULL;
 767         if ((disk->d_flags & G_RAID3_DISK_FLAG_HARDCODED) != 0 && pp != NULL)
 768                 strlcpy(md->md_provider, pp->name, sizeof(md->md_provider));
 769         else
 770                 bzero(md->md_provider, sizeof(md->md_provider));
 771         if (pp != NULL)
 772                 md->md_provsize = pp->mediasize;
 773         else
 774                 md->md_provsize = 0;
 775 }
 776
 777 void
 778 g_raid3_update_metadata(struct g_raid3_disk *disk)
 779 {
 780         struct g_raid3_softc *sc;
 781         struct g_raid3_metadata md;
 782         int error;
 783
 784         g_topology_assert_not();
 785         sc = disk->d_softc;
 786         sx_assert(&sc->sc_lock, SX_LOCKED);
 787
 788         g_raid3_fill_metadata(disk, &md);
 789         error = g_raid3_write_metadata(disk, &md);
 790         if (error == 0) {
 791                 G_RAID3_DEBUG(2, "Metadata on %s updated.",
 792                     g_raid3_get_diskname(disk));
 793         } else {
 794                 G_RAID3_DEBUG(0,
 795                     "Cannot update metadata on disk %s (error=%d).",
 796                     g_raid3_get_diskname(disk), error);
 797         }
 798 }
 799
 800 static void
 801 g_raid3_bump_syncid(struct g_raid3_softc *sc)
 802 {
 803         struct g_raid3_disk *disk;
 804         u_int n;
 805
 806         g_topology_assert_not();
 807         sx_assert(&sc->sc_lock, SX_XLOCKED);
 808         KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) > 0,
 809             ("%s called with no active disks (device=%s).", __func__,
 810             sc->sc_name));
 811
 812         sc->sc_syncid++;
 813         G_RAID3_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name,
 814             sc->sc_syncid);
 815         for (n = 0; n < sc->sc_ndisks; n++) {
 816                 disk = &sc->sc_disks[n];
 817                 if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE ||
 818                     disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) {
 819                         disk->d_sync.ds_syncid = sc->sc_syncid;
 820                         g_raid3_update_metadata(disk);
 821                 }
 822         }
 823 }
 824
 825 static void
 826 g_raid3_bump_genid(struct g_raid3_softc *sc)
 827 {
 828         struct g_raid3_disk *disk;
 829         u_int n;
 830
 831         g_topology_assert_not();
 832         sx_assert(&sc->sc_lock, SX_XLOCKED);
 833         KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) > 0,
 834             ("%s called with no active disks (device=%s).", __func__,
 835             sc->sc_name));
 836
 837         sc->sc_genid++;
 838         G_RAID3_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name,
 839             sc->sc_genid);
 840         for (n = 0; n < sc->sc_ndisks; n++) {
 841                 disk = &sc->sc_disks[n];
 842                 if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE ||
 843                     disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) {
 844                         disk->d_genid = sc->sc_genid;
 845                         g_raid3_update_metadata(disk);
 846                 }
 847         }
 848 }
 849
 850 static int
 851 g_raid3_idle(struct g_raid3_softc *sc, int acw)
 852 {
 853         struct g_raid3_disk *disk;
 854         u_int i;
 855         int timeout;
 856
 857         g_topology_assert_not();
 858         sx_assert(&sc->sc_lock, SX_XLOCKED);
 859
 860         if (sc->sc_provider == NULL)
 861                 return (0);
 862         if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) != 0)
 863                 return (0);
 864         if (sc->sc_idle)
 865                 return (0);
 866         if (sc->sc_writes > 0)
 867                 return (0);
 868         if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) {
 869                 timeout = g_raid3_idletime - (time_uptime - sc->sc_last_write);
 870                 if (!g_raid3_shutdown && timeout > 0)
 871                         return (timeout);
 872         }
 873         sc->sc_idle = 1;
 874         for (i = 0; i < sc->sc_ndisks; i++) {
 875                 disk = &sc->sc_disks[i];
 876                 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE)
 877                         continue;
 878                 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as clean.",
 879                     g_raid3_get_diskname(disk), sc->sc_name);
 880                 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY;
 881                 g_raid3_update_metadata(disk);
 882         }
 883         return (0);
 884 }
 885
 886 static void
 887 g_raid3_unidle(struct g_raid3_softc *sc)
 888 {
 889         struct g_raid3_disk *disk;
 890         u_int i;
 891
 892         g_topology_assert_not();
 893         sx_assert(&sc->sc_lock, SX_XLOCKED);
 894
 895         if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) != 0)
 896                 return;
 897         sc->sc_idle = 0;
 898         sc->sc_last_write = time_uptime;
 899         for (i = 0; i < sc->sc_ndisks; i++) {
 900                 disk = &sc->sc_disks[i];
 901                 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE)
 902                         continue;
 903                 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as dirty.",
 904                     g_raid3_get_diskname(disk), sc->sc_name);
 905                 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY;
 906                 g_raid3_update_metadata(disk);
 907         }
 908 }
 909
 910 /*
 911  * Treat bio_driver1 field in parent bio as list head and field bio_caller1
 912  * in child bio as pointer to the next element on the list.
 913  */
 914 #define G_RAID3_HEAD_BIO(pbp)   (pbp)->bio_driver1
 915
 916 #define G_RAID3_NEXT_BIO(cbp)   (cbp)->bio_caller1
 917
 918 #define G_RAID3_FOREACH_BIO(pbp, bp)                                    \
 919         for ((bp) = G_RAID3_HEAD_BIO(pbp); (bp) != NULL;                \
 920             (bp) = G_RAID3_NEXT_BIO(bp))
 921
 922 #define G_RAID3_FOREACH_SAFE_BIO(pbp, bp, tmpbp)                        \
 923         for ((bp) = G_RAID3_HEAD_BIO(pbp);                              \
 924             (bp) != NULL && ((tmpbp) = G_RAID3_NEXT_BIO(bp), 1);        \
 925             (bp) = (tmpbp))
 926
 927 static void
 928 g_raid3_init_bio(struct bio *pbp)
 929 {
 930
 931         G_RAID3_HEAD_BIO(pbp) = NULL;
 932 }
 933
 934 static void
 935 g_raid3_remove_bio(struct bio *cbp)
 936 {
 937         struct bio *pbp, *bp;
 938
 939         pbp = cbp->bio_parent;
 940         if (G_RAID3_HEAD_BIO(pbp) == cbp)
 941                 G_RAID3_HEAD_BIO(pbp) = G_RAID3_NEXT_BIO(cbp);
 942         else {
 943                 G_RAID3_FOREACH_BIO(pbp, bp) {
 944                         if (G_RAID3_NEXT_BIO(bp) == cbp) {
 945                                 G_RAID3_NEXT_BIO(bp) = G_RAID3_NEXT_BIO(cbp);
 946                                 break;
 947                         }
 948                 }
 949         }
 950         G_RAID3_NEXT_BIO(cbp) = NULL;
 951 }
 952
 953 static void
 954 g_raid3_replace_bio(struct bio *sbp, struct bio *dbp)
 955 {
 956         struct bio *pbp, *bp;
 957
 958         g_raid3_remove_bio(sbp);
 959         pbp = dbp->bio_parent;
 960         G_RAID3_NEXT_BIO(sbp) = G_RAID3_NEXT_BIO(dbp);
 961         if (G_RAID3_HEAD_BIO(pbp) == dbp)
 962                 G_RAID3_HEAD_BIO(pbp) = sbp;
 963         else {
 964                 G_RAID3_FOREACH_BIO(pbp, bp) {
 965                         if (G_RAID3_NEXT_BIO(bp) == dbp) {
 966                                 G_RAID3_NEXT_BIO(bp) = sbp;
 967                                 break;
 968                         }
 969                 }
 970         }
 971         G_RAID3_NEXT_BIO(dbp) = NULL;
 972 }
 973
 974 static void
 975 g_raid3_destroy_bio(struct g_raid3_softc *sc, struct bio *cbp)
 976 {
 977         struct bio *bp, *pbp;
 978         size_t size;
 979
 980         pbp = cbp->bio_parent;
 981         pbp->bio_children--;
 982         KASSERT(cbp->bio_data != NULL, ("NULL bio_data"));
 983         size = pbp->bio_length / (sc->sc_ndisks - 1);
 984         g_raid3_free(sc, cbp->bio_data, size);
 985         if (G_RAID3_HEAD_BIO(pbp) == cbp) {
 986                 G_RAID3_HEAD_BIO(pbp) = G_RAID3_NEXT_BIO(cbp);
 987                 G_RAID3_NEXT_BIO(cbp) = NULL;
 988                 g_destroy_bio(cbp);
 989         } else {
 990                 G_RAID3_FOREACH_BIO(pbp, bp) {
 991                         if (G_RAID3_NEXT_BIO(bp) == cbp)
 992                                 break;
 993                 }
 994                 if (bp != NULL) {
 995                         KASSERT(G_RAID3_NEXT_BIO(bp) != NULL,
 996                             ("NULL bp->bio_driver1"));
 997                         G_RAID3_NEXT_BIO(bp) = G_RAID3_NEXT_BIO(cbp);
 998                         G_RAID3_NEXT_BIO(cbp) = NULL;
 999                 }
1000                 g_destroy_bio(cbp);
1001         }
1002 }
1003
1004 static struct bio *
1005 g_raid3_clone_bio(struct g_raid3_softc *sc, struct bio *pbp)
1006 {
1007         struct bio *bp, *cbp;
1008         size_t size;
1009         int memflag;
1010
1011         cbp = g_clone_bio(pbp);
1012         if (cbp == NULL)
1013                 return (NULL);
1014         size = pbp->bio_length / (sc->sc_ndisks - 1);
1015         if ((pbp->bio_cflags & G_RAID3_BIO_CFLAG_REGULAR) != 0)
1016                 memflag = M_WAITOK;
1017         else
1018                 memflag = M_NOWAIT;
1019         cbp->bio_data = g_raid3_alloc(sc, size, memflag);
1020         if (cbp->bio_data == NULL) {
1021                 pbp->bio_children--;
1022                 g_destroy_bio(cbp);
1023                 return (NULL);
1024         }
1025         G_RAID3_NEXT_BIO(cbp) = NULL;
1026         if (G_RAID3_HEAD_BIO(pbp) == NULL)
1027                 G_RAID3_HEAD_BIO(pbp) = cbp;
1028         else {
1029                 G_RAID3_FOREACH_BIO(pbp, bp) {
1030                         if (G_RAID3_NEXT_BIO(bp) == NULL) {
1031                                 G_RAID3_NEXT_BIO(bp) = cbp;
1032                                 break;
1033                         }
1034                 }
1035         }
1036         return (cbp);
1037 }
1038
1039 static void
1040 g_raid3_scatter(struct bio *pbp)
1041 {
1042         struct g_raid3_softc *sc;
1043         struct g_raid3_disk *disk;
1044         struct bio *bp, *cbp, *tmpbp;
1045         off_t atom, cadd, padd, left;
1046         int first;
1047
1048         sc = pbp->bio_to->geom->softc;
1049         bp = NULL;
1050         if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_NOPARITY) == 0) {
1051                 /*
1052                  * Find bio for which we should calculate data.
1053                  */
1054                 G_RAID3_FOREACH_BIO(pbp, cbp) {
1055                         if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) {
1056                                 bp = cbp;
1057                                 break;
1058                         }
1059                 }
1060                 KASSERT(bp != NULL, ("NULL parity bio."));
1061         }
1062         atom = sc->sc_sectorsize / (sc->sc_ndisks - 1);
1063         cadd = padd = 0;
1064         for (left = pbp->bio_length; left > 0; left -= sc->sc_sectorsize) {
1065                 G_RAID3_FOREACH_BIO(pbp, cbp) {
1066                         if (cbp == bp)
1067                                 continue;
1068                         bcopy(pbp->bio_data + padd, cbp->bio_data + cadd, atom);
1069                         padd += atom;
1070                 }
1071                 cadd += atom;
1072         }
1073         if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_NOPARITY) == 0) {
1074                 /*
1075                  * Calculate parity.
1076                  */
1077                 first = 1;
1078                 G_RAID3_FOREACH_SAFE_BIO(pbp, cbp, tmpbp) {
1079                         if (cbp == bp)
1080                                 continue;
1081                         if (first) {
1082                                 bcopy(cbp->bio_data, bp->bio_data,
1083                                     bp->bio_length);
1084                                 first = 0;
1085                         } else {
1086                                 g_raid3_xor(cbp->bio_data, bp->bio_data,
1087                                     bp->bio_length);
1088                         }
1089                         if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_NODISK) != 0)
1090                                 g_raid3_destroy_bio(sc, cbp);
1091                 }
1092         }
1093         G_RAID3_FOREACH_SAFE_BIO(pbp, cbp, tmpbp) {
1094                 struct g_consumer *cp;
1095
1096                 disk = cbp->bio_caller2;
1097                 cp = disk->d_consumer;
1098                 cbp->bio_to = cp->provider;
1099                 G_RAID3_LOGREQ(3, cbp, "Sending request.");
1100                 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
1101                     ("Consumer %s not opened (r%dw%de%d).", cp->provider->name,
1102                     cp->acr, cp->acw, cp->ace));
1103                 cp->index++;
1104                 sc->sc_writes++;
1105                 g_io_request(cbp, cp);
1106         }
1107 }
1108
1109 static void
1110 g_raid3_gather(struct bio *pbp)
1111 {
1112         struct g_raid3_softc *sc;
1113         struct g_raid3_disk *disk;
1114         struct bio *xbp, *fbp, *cbp;
1115         off_t atom, cadd, padd, left;
1116
1117         sc = pbp->bio_to->geom->softc;
1118         /*
1119          * Find bio for which we have to calculate data.
1120          * While going through this path, check if all requests
1121          * succeeded, if not, deny whole request.
1122          * If we're in COMPLETE mode, we allow one request to fail,
1123          * so if we find one, we're sending it to the parity consumer.
1124          * If there are more failed requests, we deny whole request.
1125          */
1126         xbp = fbp = NULL;
1127         G_RAID3_FOREACH_BIO(pbp, cbp) {
1128                 if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) {
1129                         KASSERT(xbp == NULL, ("More than one parity bio."));
1130                         xbp = cbp;
1131                 }
1132                 if (cbp->bio_error == 0)
1133                         continue;
1134                 /*
1135                  * Found failed request.
1136                  */
1137                 if (fbp == NULL) {
1138                         if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_DEGRADED) != 0) {
1139                                 /*
1140                                  * We are already in degraded mode, so we can't
1141                                  * accept any failures.
1142                                  */
1143                                 if (pbp->bio_error == 0)
1144                                         pbp->bio_error = cbp->bio_error;
1145                         } else {
1146                                 fbp = cbp;
1147                         }
1148                 } else {
1149                         /*
1150                          * Next failed request, that's too many.
1151                          */
1152                         if (pbp->bio_error == 0)
1153                                 pbp->bio_error = fbp->bio_error;
1154                 }
1155                 disk = cbp->bio_caller2;
1156                 if (disk == NULL)
1157                         continue;
1158                 if ((disk->d_flags & G_RAID3_DISK_FLAG_BROKEN) == 0) {
1159                         disk->d_flags |= G_RAID3_DISK_FLAG_BROKEN;
1160                         G_RAID3_LOGREQ(0, cbp, "Request failed (error=%d).",
1161                             cbp->bio_error);
1162                 } else {
1163                         G_RAID3_LOGREQ(1, cbp, "Request failed (error=%d).",
1164                             cbp->bio_error);
1165                 }
1166                 if (g_raid3_disconnect_on_failure &&
1167                     sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) {
1168                         sc->sc_bump_id |= G_RAID3_BUMP_GENID;
1169                         g_raid3_event_send(disk,
1170                             G_RAID3_DISK_STATE_DISCONNECTED,
1171                             G_RAID3_EVENT_DONTWAIT);
1172                 }
1173         }
1174         if (pbp->bio_error != 0)
1175                 goto finish;
1176         if (fbp != NULL && (pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) {
1177                 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_VERIFY;
1178                 if (xbp != fbp)
1179                         g_raid3_replace_bio(xbp, fbp);
1180                 g_raid3_destroy_bio(sc, fbp);
1181         } else if (fbp != NULL) {
1182                 struct g_consumer *cp;
1183
1184                 /*
1185                  * One request failed, so send the same request to
1186                  * the parity consumer.
1187                  */
1188                 disk = pbp->bio_driver2;
1189                 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) {
1190                         pbp->bio_error = fbp->bio_error;
1191                         goto finish;
1192                 }
1193                 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED;
1194                 pbp->bio_inbed--;
1195                 fbp->bio_flags &= ~(BIO_DONE | BIO_ERROR);
1196                 if (disk->d_no == sc->sc_ndisks - 1)
1197                         fbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY;
1198                 fbp->bio_error = 0;
1199                 fbp->bio_completed = 0;
1200                 fbp->bio_children = 0;
1201                 fbp->bio_inbed = 0;
1202                 cp = disk->d_consumer;
1203                 fbp->bio_caller2 = disk;
1204                 fbp->bio_to = cp->provider;
1205                 G_RAID3_LOGREQ(3, fbp, "Sending request (recover).");
1206                 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
1207                     ("Consumer %s not opened (r%dw%de%d).", cp->provider->name,
1208                     cp->acr, cp->acw, cp->ace));
1209                 cp->index++;
1210                 g_io_request(fbp, cp);
1211                 return;
1212         }
1213         if (xbp != NULL) {
1214                 /*
1215                  * Calculate parity.
1216                  */
1217                 G_RAID3_FOREACH_BIO(pbp, cbp) {
1218                         if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0)
1219                                 continue;
1220                         g_raid3_xor(cbp->bio_data, xbp->bio_data,
1221                             xbp->bio_length);
1222                 }
1223                 xbp->bio_cflags &= ~G_RAID3_BIO_CFLAG_PARITY;
1224                 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) {
1225                         if (!g_raid3_is_zero(xbp)) {
1226                                 g_raid3_parity_mismatch++;
1227                                 pbp->bio_error = EIO;
1228                                 goto finish;
1229                         }
1230                         g_raid3_destroy_bio(sc, xbp);
1231                 }
1232         }
1233         atom = sc->sc_sectorsize / (sc->sc_ndisks - 1);
1234         cadd = padd = 0;
1235         for (left = pbp->bio_length; left > 0; left -= sc->sc_sectorsize) {
1236                 G_RAID3_FOREACH_BIO(pbp, cbp) {
1237                         bcopy(cbp->bio_data + cadd, pbp->bio_data + padd, atom);
1238                         pbp->bio_completed += atom;
1239                         padd += atom;
1240                 }
1241                 cadd += atom;
1242         }
1243 finish:
1244         if (pbp->bio_error == 0)
1245                 G_RAID3_LOGREQ(3, pbp, "Request finished.");
1246         else {
1247                 if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0)
1248                         G_RAID3_LOGREQ(1, pbp, "Verification error.");
1249                 else
1250                         G_RAID3_LOGREQ(0, pbp, "Request failed.");
1251         }
1252         pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_MASK;
1253         while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL)
1254                 g_raid3_destroy_bio(sc, cbp);
1255         g_io_deliver(pbp, pbp->bio_error);
1256 }
1257
1258 static void
1259 g_raid3_done(struct bio *bp)
1260 {
1261         struct g_raid3_softc *sc;
1262
1263         sc = bp->bio_from->geom->softc;
1264         bp->bio_cflags |= G_RAID3_BIO_CFLAG_REGULAR;
1265         G_RAID3_LOGREQ(3, bp, "Regular request done (error=%d).", bp->bio_error);
1266         mtx_lock(&sc->sc_queue_mtx);
1267         bioq_insert_head(&sc->sc_queue, bp);
1268         mtx_unlock(&sc->sc_queue_mtx);
1269         wakeup(sc);
1270         wakeup(&sc->sc_queue);
1271 }
1272
1273 static void
1274 g_raid3_regular_request(struct bio *cbp)
1275 {
1276         struct g_raid3_softc *sc;
1277         struct g_raid3_disk *disk;
1278         struct bio *pbp;
1279
1280         g_topology_assert_not();
1281
1282         pbp = cbp->bio_parent;
1283         sc = pbp->bio_to->geom->softc;
1284         cbp->bio_from->index--;
1285         if (cbp->bio_cmd == BIO_WRITE)
1286                 sc->sc_writes--;
1287         disk = cbp->bio_from->private;
1288         if (disk == NULL) {
1289                 g_topology_lock();
1290                 g_raid3_kill_consumer(sc, cbp->bio_from);
1291                 g_topology_unlock();
1292         }
1293
1294         G_RAID3_LOGREQ(3, cbp, "Request finished.");
1295         pbp->bio_inbed++;
1296         KASSERT(pbp->bio_inbed <= pbp->bio_children,
1297             ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed,
1298             pbp->bio_children));
1299         if (pbp->bio_inbed != pbp->bio_children)
1300                 return;
1301         switch (pbp->bio_cmd) {
1302         case BIO_READ:
1303                 g_raid3_gather(pbp);
1304                 break;
1305         case BIO_WRITE:
1306         case BIO_DELETE:
1307             {
1308                 int error = 0;
1309
1310                 pbp->bio_completed = pbp->bio_length;
1311                 while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL) {
1312                         if (cbp->bio_error == 0) {
1313                                 g_raid3_destroy_bio(sc, cbp);
1314                                 continue;
1315                         }
1316
1317                         if (error == 0)
1318                                 error = cbp->bio_error;
1319                         else if (pbp->bio_error == 0) {
1320                                 /*
1321                                  * Next failed request, that's too many.
1322                                  */
1323                                 pbp->bio_error = error;
1324                         }
1325
1326                         disk = cbp->bio_caller2;
1327                         if (disk == NULL) {
1328                                 g_raid3_destroy_bio(sc, cbp);
1329                                 continue;
1330                         }
1331
1332                         if ((disk->d_flags & G_RAID3_DISK_FLAG_BROKEN) == 0) {
1333                                 disk->d_flags |= G_RAID3_DISK_FLAG_BROKEN;
1334                                 G_RAID3_LOGREQ(0, cbp,
1335                                     "Request failed (error=%d).",
1336                                     cbp->bio_error);
1337                         } else {
1338                                 G_RAID3_LOGREQ(1, cbp,
1339                                     "Request failed (error=%d).",
1340                                     cbp->bio_error);
1341                         }
1342                         if (g_raid3_disconnect_on_failure &&
1343                             sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) {
1344                                 sc->sc_bump_id |= G_RAID3_BUMP_GENID;
1345                                 g_raid3_event_send(disk,
1346                                     G_RAID3_DISK_STATE_DISCONNECTED,
1347                                     G_RAID3_EVENT_DONTWAIT);
1348                         }
1349                         g_raid3_destroy_bio(sc, cbp);
1350                 }
1351                 if (pbp->bio_error == 0)
1352                         G_RAID3_LOGREQ(3, pbp, "Request finished.");
1353                 else
1354                         G_RAID3_LOGREQ(0, pbp, "Request failed.");
1355                 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_DEGRADED;
1356                 pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_NOPARITY;
1357                 bioq_remove(&sc->sc_inflight, pbp);
1358                 /* Release delayed sync requests if possible. */
1359                 g_raid3_sync_release(sc);
1360                 g_io_deliver(pbp, pbp->bio_error);
1361                 break;
1362             }
1363         }
1364 }
1365
1366 static void
1367 g_raid3_sync_done(struct bio *bp)
1368 {
1369         struct g_raid3_softc *sc;
1370
1371         G_RAID3_LOGREQ(3, bp, "Synchronization request delivered.");
1372         sc = bp->bio_from->geom->softc;
1373         bp->bio_cflags |= G_RAID3_BIO_CFLAG_SYNC;
1374         mtx_lock(&sc->sc_queue_mtx);
1375         bioq_insert_head(&sc->sc_queue, bp);
1376         mtx_unlock(&sc->sc_queue_mtx);
1377         wakeup(sc);
1378         wakeup(&sc->sc_queue);
1379 }
1380
1381 static void
1382 g_raid3_flush(struct g_raid3_softc *sc, struct bio *bp)
1383 {
1384         struct bio_queue_head queue;
1385         struct g_raid3_disk *disk;
1386         struct g_consumer *cp;
1387         struct bio *cbp;
1388         u_int i;
1389
1390         bioq_init(&queue);
1391         for (i = 0; i < sc->sc_ndisks; i++) {
1392                 disk = &sc->sc_disks[i];
1393                 if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE)
1394                         continue;
1395                 cbp = g_clone_bio(bp);
1396                 if (cbp == NULL) {
1397                         for (cbp = bioq_first(&queue); cbp != NULL;
1398                             cbp = bioq_first(&queue)) {
1399                                 bioq_remove(&queue, cbp);
1400                                 g_destroy_bio(cbp);
1401                         }
1402                         if (bp->bio_error == 0)
1403                                 bp->bio_error = ENOMEM;
1404                         g_io_deliver(bp, bp->bio_error);
1405                         return;
1406                 }
1407                 bioq_insert_tail(&queue, cbp);
1408                 cbp->bio_done = g_std_done;
1409                 cbp->bio_caller1 = disk;
1410                 cbp->bio_to = disk->d_consumer->provider;
1411         }
1412         for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) {
1413                 bioq_remove(&queue, cbp);
1414                 G_RAID3_LOGREQ(3, cbp, "Sending request.");
1415                 disk = cbp->bio_caller1;
1416                 cbp->bio_caller1 = NULL;
1417                 cp = disk->d_consumer;
1418                 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
1419                     ("Consumer %s not opened (r%dw%de%d).", cp->provider->name,
1420                     cp->acr, cp->acw, cp->ace));
1421                 g_io_request(cbp, disk->d_consumer);
1422         }
1423 }
1424
1425 static void
1426 g_raid3_start(struct bio *bp)
1427 {
1428         struct g_raid3_softc *sc;
1429
1430         sc = bp->bio_to->geom->softc;
1431         /*
1432          * If sc == NULL or there are no valid disks, provider's error
1433          * should be set and g_raid3_start() should not be called at all.
1434          */
1435         KASSERT(sc != NULL && (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED ||
1436             sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE),
1437             ("Provider's error should be set (error=%d)(device=%s).",
1438             bp->bio_to->error, bp->bio_to->name));
1439         G_RAID3_LOGREQ(3, bp, "Request received.");
1440
1441         switch (bp->bio_cmd) {
1442         case BIO_READ:
1443         case BIO_WRITE:
1444         case BIO_DELETE:
1445                 break;
1446         case BIO_FLUSH:
1447                 g_raid3_flush(sc, bp);
1448                 return;
1449         case BIO_GETATTR:
1450         default:
1451                 g_io_deliver(bp, EOPNOTSUPP);
1452                 return;
1453         }
1454         mtx_lock(&sc->sc_queue_mtx);
1455         bioq_insert_tail(&sc->sc_queue, bp);
1456         mtx_unlock(&sc->sc_queue_mtx);
1457         G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc);
1458         wakeup(sc);
1459 }
1460
1461 /*
1462  * Return TRUE if the given request is colliding with a in-progress
1463  * synchronization request.
1464  */
1465 static int
1466 g_raid3_sync_collision(struct g_raid3_softc *sc, struct bio *bp)
1467 {
1468         struct g_raid3_disk *disk;
1469         struct bio *sbp;
1470         off_t rstart, rend, sstart, send;
1471         int i;
1472
1473         disk = sc->sc_syncdisk;
1474         if (disk == NULL)
1475                 return (0);
1476         rstart = bp->bio_offset;
1477         rend = bp->bio_offset + bp->bio_length;
1478         for (i = 0; i < g_raid3_syncreqs; i++) {
1479                 sbp = disk->d_sync.ds_bios[i];
1480                 if (sbp == NULL)
1481                         continue;
1482                 sstart = sbp->bio_offset;
1483                 send = sbp->bio_length;
1484                 if (sbp->bio_cmd == BIO_WRITE) {
1485                         sstart *= sc->sc_ndisks - 1;
1486                         send *= sc->sc_ndisks - 1;
1487                 }
1488                 send += sstart;
1489                 if (rend > sstart && rstart < send)
1490                         return (1);
1491         }
1492         return (0);
1493 }
1494
1495 /*
1496  * Return TRUE if the given sync request is colliding with a in-progress regular
1497  * request.
1498  */
1499 static int
1500 g_raid3_regular_collision(struct g_raid3_softc *sc, struct bio *sbp)
1501 {
1502         off_t rstart, rend, sstart, send;
1503         struct bio *bp;
1504
1505         if (sc->sc_syncdisk == NULL)
1506                 return (0);
1507         sstart = sbp->bio_offset;
1508         send = sstart + sbp->bio_length;
1509         TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) {
1510                 rstart = bp->bio_offset;
1511                 rend = bp->bio_offset + bp->bio_length;
1512                 if (rend > sstart && rstart < send)
1513                         return (1);
1514         }
1515         return (0);
1516 }
1517
1518 /*
1519  * Puts request onto delayed queue.
1520  */
1521 static void
1522 g_raid3_regular_delay(struct g_raid3_softc *sc, struct bio *bp)
1523 {
1524
1525         G_RAID3_LOGREQ(2, bp, "Delaying request.");
1526         bioq_insert_head(&sc->sc_regular_delayed, bp);
1527 }
1528
1529 /*
1530  * Puts synchronization request onto delayed queue.
1531  */
1532 static void
1533 g_raid3_sync_delay(struct g_raid3_softc *sc, struct bio *bp)
1534 {
1535
1536         G_RAID3_LOGREQ(2, bp, "Delaying synchronization request.");
1537         bioq_insert_tail(&sc->sc_sync_delayed, bp);
1538 }
1539
1540 /*
1541  * Releases delayed regular requests which don't collide anymore with sync
1542  * requests.
1543  */
1544 static void
1545 g_raid3_regular_release(struct g_raid3_softc *sc)
1546 {
1547         struct bio *bp, *bp2;
1548
1549         TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) {
1550                 if (g_raid3_sync_collision(sc, bp))
1551                         continue;
1552                 bioq_remove(&sc->sc_regular_delayed, bp);
1553                 G_RAID3_LOGREQ(2, bp, "Releasing delayed request (%p).", bp);
1554                 mtx_lock(&sc->sc_queue_mtx);
1555                 bioq_insert_head(&sc->sc_queue, bp);
1556 #if 0
1557                 /*
1558                  * wakeup() is not needed, because this function is called from
1559                  * the worker thread.
1560                  */
1561                 wakeup(&sc->sc_queue);
1562 #endif
1563                 mtx_unlock(&sc->sc_queue_mtx);
1564         }
1565 }
1566
1567 /*
1568  * Releases delayed sync requests which don't collide anymore with regular
1569  * requests.
1570  */
1571 static void
1572 g_raid3_sync_release(struct g_raid3_softc *sc)
1573 {
1574         struct bio *bp, *bp2;
1575
1576         TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) {
1577                 if (g_raid3_regular_collision(sc, bp))
1578                         continue;
1579                 bioq_remove(&sc->sc_sync_delayed, bp);
1580                 G_RAID3_LOGREQ(2, bp,
1581                     "Releasing delayed synchronization request.");
1582                 g_io_request(bp, bp->bio_from);
1583         }
1584 }
1585
1586 /*
1587  * Handle synchronization requests.
1588  * Every synchronization request is two-steps process: first, READ request is
1589  * send to active provider and then WRITE request (with read data) to the provider
1590  * beeing synchronized. When WRITE is finished, new synchronization request is
1591  * send.
1592  */
1593 static void
1594 g_raid3_sync_request(struct bio *bp)
1595 {
1596         struct g_raid3_softc *sc;
1597         struct g_raid3_disk *disk;
1598
1599         bp->bio_from->index--;
1600         sc = bp->bio_from->geom->softc;
1601         disk = bp->bio_from->private;
1602         if (disk == NULL) {
1603                 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */
1604                 g_topology_lock();
1605                 g_raid3_kill_consumer(sc, bp->bio_from);
1606                 g_topology_unlock();
1607                 free(bp->bio_data, M_RAID3);
1608                 g_destroy_bio(bp);
1609                 sx_xlock(&sc->sc_lock);
1610                 return;
1611         }
1612
1613         /*
1614          * Synchronization request.
1615          */
1616         switch (bp->bio_cmd) {
1617         case BIO_READ:
1618             {
1619                 struct g_consumer *cp;
1620                 u_char *dst, *src;
1621                 off_t left;
1622                 u_int atom;
1623
1624                 if (bp->bio_error != 0) {
1625                         G_RAID3_LOGREQ(0, bp,
1626                             "Synchronization request failed (error=%d).",
1627                             bp->bio_error);
1628                         g_destroy_bio(bp);
1629                         return;
1630                 }
1631                 G_RAID3_LOGREQ(3, bp, "Synchronization request finished.");
1632                 atom = sc->sc_sectorsize / (sc->sc_ndisks - 1);
1633                 dst = src = bp->bio_data;
1634                 if (disk->d_no == sc->sc_ndisks - 1) {
1635                         u_int n;
1636
1637                         /* Parity component. */
1638                         for (left = bp->bio_length; left > 0;
1639                             left -= sc->sc_sectorsize) {
1640                                 bcopy(src, dst, atom);
1641                                 src += atom;
1642                                 for (n = 1; n < sc->sc_ndisks - 1; n++) {
1643                                         g_raid3_xor(src, dst, atom);
1644                                         src += atom;
1645                                 }
1646                                 dst += atom;
1647                         }
1648                 } else {
1649                         /* Regular component. */
1650                         src += atom * disk->d_no;
1651                         for (left = bp->bio_length; left > 0;
1652                             left -= sc->sc_sectorsize) {
1653                                 bcopy(src, dst, atom);
1654                                 src += sc->sc_sectorsize;
1655                                 dst += atom;
1656                         }
1657                 }
1658                 bp->bio_driver1 = bp->bio_driver2 = NULL;
1659                 bp->bio_pflags = 0;
1660                 bp->bio_offset /= sc->sc_ndisks - 1;
1661                 bp->bio_length /= sc->sc_ndisks - 1;
1662                 bp->bio_cmd = BIO_WRITE;
1663                 bp->bio_cflags = 0;
1664                 bp->bio_children = bp->bio_inbed = 0;
1665                 cp = disk->d_consumer;
1666                 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
1667                     ("Consumer %s not opened (r%dw%de%d).", cp->provider->name,
1668                     cp->acr, cp->acw, cp->ace));
1669                 cp->index++;
1670                 g_io_request(bp, cp);
1671                 return;
1672             }
1673         case BIO_WRITE:
1674             {
1675                 struct g_raid3_disk_sync *sync;
1676                 off_t boffset, moffset;
1677                 void *data;
1678                 int i;
1679
1680                 if (bp->bio_error != 0) {
1681                         G_RAID3_LOGREQ(0, bp,
1682                             "Synchronization request failed (error=%d).",
1683                             bp->bio_error);
1684                         g_destroy_bio(bp);
1685                         sc->sc_bump_id |= G_RAID3_BUMP_GENID;
1686                         g_raid3_event_send(disk,
1687                             G_RAID3_DISK_STATE_DISCONNECTED,
1688                             G_RAID3_EVENT_DONTWAIT);
1689                         return;
1690                 }
1691                 G_RAID3_LOGREQ(3, bp, "Synchronization request finished.");
1692                 sync = &disk->d_sync;
1693                 if (sync->ds_offset == sc->sc_mediasize / (sc->sc_ndisks - 1) ||
1694                     sync->ds_consumer == NULL ||
1695                     (sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) != 0) {
1696                         /* Don't send more synchronization requests. */
1697                         sync->ds_inflight--;
1698                         if (sync->ds_bios != NULL) {
1699                                 i = (int)(uintptr_t)bp->bio_caller1;
1700                                 sync->ds_bios[i] = NULL;
1701                         }
1702                         free(bp->bio_data, M_RAID3);
1703                         g_destroy_bio(bp);
1704                         if (sync->ds_inflight > 0)
1705                                 return;
1706                         if (sync->ds_consumer == NULL ||
1707                             (sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) != 0) {
1708                                 return;
1709                         }
1710                         /*
1711                          * Disk up-to-date, activate it.
1712                          */
1713                         g_raid3_event_send(disk, G_RAID3_DISK_STATE_ACTIVE,
1714                             G_RAID3_EVENT_DONTWAIT);
1715                         return;
1716                 }
1717
1718                 /* Send next synchronization request. */
1719                 data = bp->bio_data;
1720                 g_reset_bio(bp);
1721                 bp->bio_cmd = BIO_READ;
1722                 bp->bio_offset = sync->ds_offset * (sc->sc_ndisks - 1);
1723                 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset);
1724                 sync->ds_offset += bp->bio_length / (sc->sc_ndisks - 1);
1725                 bp->bio_done = g_raid3_sync_done;
1726                 bp->bio_data = data;
1727                 bp->bio_from = sync->ds_consumer;
1728                 bp->bio_to = sc->sc_provider;
1729                 G_RAID3_LOGREQ(3, bp, "Sending synchronization request.");
1730                 sync->ds_consumer->index++;
1731                 /*
1732                  * Delay the request if it is colliding with a regular request.
1733                  */
1734                 if (g_raid3_regular_collision(sc, bp))
1735                         g_raid3_sync_delay(sc, bp);
1736                 else
1737                         g_io_request(bp, sync->ds_consumer);
1738
1739                 /* Release delayed requests if possible. */
1740                 g_raid3_regular_release(sc);
1741
1742                 /* Find the smallest offset. */
1743                 moffset = sc->sc_mediasize;
1744                 for (i = 0; i < g_raid3_syncreqs; i++) {
1745                         bp = sync->ds_bios[i];
1746                         boffset = bp->bio_offset;
1747                         if (bp->bio_cmd == BIO_WRITE)
1748                                 boffset *= sc->sc_ndisks - 1;
1749                         if (boffset < moffset)
1750                                 moffset = boffset;
1751                 }
1752                 if (sync->ds_offset_done + (MAXPHYS * 100) < moffset) {
1753                         /* Update offset_done on every 100 blocks. */
1754                         sync->ds_offset_done = moffset;
1755                         g_raid3_update_metadata(disk);
1756                 }
1757                 return;
1758             }
1759         default:
1760                 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)",
1761                     bp->bio_cmd, sc->sc_name));
1762                 break;
1763         }
1764 }
1765
1766 static int
1767 g_raid3_register_request(struct bio *pbp)
1768 {
1769         struct g_raid3_softc *sc;
1770         struct g_raid3_disk *disk;
1771         struct g_consumer *cp;
1772         struct bio *cbp, *tmpbp;
1773         off_t offset, length;
1774         u_int n, ndisks;
1775         int round_robin, verify;
1776
1777         ndisks = 0;
1778         sc = pbp->bio_to->geom->softc;
1779         if ((pbp->bio_cflags & G_RAID3_BIO_CFLAG_REGSYNC) != 0 &&
1780             sc->sc_syncdisk == NULL) {
1781                 g_io_deliver(pbp, EIO);
1782                 return (0);
1783         }
1784         g_raid3_init_bio(pbp);
1785         length = pbp->bio_length / (sc->sc_ndisks - 1);
1786         offset = pbp->bio_offset / (sc->sc_ndisks - 1);
1787         round_robin = verify = 0;
1788         switch (pbp->bio_cmd) {
1789         case BIO_READ:
1790                 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_VERIFY) != 0 &&
1791                     sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) {
1792                         pbp->bio_pflags |= G_RAID3_BIO_PFLAG_VERIFY;
1793                         verify = 1;
1794                         ndisks = sc->sc_ndisks;
1795                 } else {
1796                         verify = 0;
1797                         ndisks = sc->sc_ndisks - 1;
1798                 }
1799                 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_ROUND_ROBIN) != 0 &&
1800                     sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) {
1801                         round_robin = 1;
1802                 } else {
1803                         round_robin = 0;
1804                 }
1805                 KASSERT(!round_robin || !verify,
1806                     ("ROUND-ROBIN and VERIFY are mutually exclusive."));
1807                 pbp->bio_driver2 = &sc->sc_disks[sc->sc_ndisks - 1];
1808                 break;
1809         case BIO_WRITE:
1810         case BIO_DELETE:
1811                 /*
1812                  * Delay the request if it is colliding with a synchronization
1813                  * request.
1814                  */
1815                 if (g_raid3_sync_collision(sc, pbp)) {
1816                         g_raid3_regular_delay(sc, pbp);
1817                         return (0);
1818                 }
1819
1820                 if (sc->sc_idle)
1821                         g_raid3_unidle(sc);
1822                 else
1823                         sc->sc_last_write = time_uptime;
1824
1825                 ndisks = sc->sc_ndisks;
1826                 break;
1827         }
1828         for (n = 0; n < ndisks; n++) {
1829                 disk = &sc->sc_disks[n];
1830                 cbp = g_raid3_clone_bio(sc, pbp);
1831                 if (cbp == NULL) {
1832                         while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL)
1833                                 g_raid3_destroy_bio(sc, cbp);
1834                         /*
1835                          * To prevent deadlock, we must run back up
1836                          * with the ENOMEM for failed requests of any
1837                          * of our consumers.  Our own sync requests
1838                          * can stick around, as they are finite.
1839                          */
1840                         if ((pbp->bio_cflags &
1841                             G_RAID3_BIO_CFLAG_REGULAR) != 0) {
1842                                 g_io_deliver(pbp, ENOMEM);
1843                                 return (0);
1844                         }
1845                         return (ENOMEM);
1846                 }
1847                 cbp->bio_offset = offset;
1848                 cbp->bio_length = length;
1849                 cbp->bio_done = g_raid3_done;
1850                 switch (pbp->bio_cmd) {
1851                 case BIO_READ:
1852                         if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) {
1853                                 /*
1854                                  * Replace invalid component with the parity
1855                                  * component.
1856                                  */
1857                                 disk = &sc->sc_disks[sc->sc_ndisks - 1];
1858                                 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY;
1859                                 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED;
1860                         } else if (round_robin &&
1861                             disk->d_no == sc->sc_round_robin) {
1862                                 /*
1863                                  * In round-robin mode skip one data component
1864                                  * and use parity component when reading.
1865                                  */
1866                                 pbp->bio_driver2 = disk;
1867                                 disk = &sc->sc_disks[sc->sc_ndisks - 1];
1868                                 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY;
1869                                 sc->sc_round_robin++;
1870                                 round_robin = 0;
1871                         } else if (verify && disk->d_no == sc->sc_ndisks - 1) {
1872                                 cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY;
1873                         }
1874                         break;
1875                 case BIO_WRITE:
1876                 case BIO_DELETE:
1877                         if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE ||
1878                             disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) {
1879                                 if (n == ndisks - 1) {
1880                                         /*
1881                                          * Active parity component, mark it as such.
1882                                          */
1883                                         cbp->bio_cflags |=
1884                                             G_RAID3_BIO_CFLAG_PARITY;
1885                                 }
1886                         } else {
1887                                 pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED;
1888                                 if (n == ndisks - 1) {
1889                                         /*
1890                                          * Parity component is not connected,
1891                                          * so destroy its request.
1892                                          */
1893                                         pbp->bio_pflags |=
1894                                             G_RAID3_BIO_PFLAG_NOPARITY;
1895                                         g_raid3_destroy_bio(sc, cbp);
1896                                         cbp = NULL;
1897                                 } else {
1898                                         cbp->bio_cflags |=
1899                                             G_RAID3_BIO_CFLAG_NODISK;
1900                                         disk = NULL;
1901                                 }
1902                         }
1903                         break;
1904                 }
1905                 if (cbp != NULL)
1906                         cbp->bio_caller2 = disk;
1907         }
1908         switch (pbp->bio_cmd) {
1909         case BIO_READ:
1910                 if (round_robin) {
1911                         /*
1912                          * If we are in round-robin mode and 'round_robin' is
1913                          * still 1, it means, that we skipped parity component
1914                          * for this read and must reset sc_round_robin field.
1915                          */
1916                         sc->sc_round_robin = 0;
1917                 }
1918                 G_RAID3_FOREACH_SAFE_BIO(pbp, cbp, tmpbp) {
1919                         disk = cbp->bio_caller2;
1920                         cp = disk->d_consumer;
1921                         cbp->bio_to = cp->provider;
1922                         G_RAID3_LOGREQ(3, cbp, "Sending request.");
1923                         KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
1924                             ("Consumer %s not opened (r%dw%de%d).",
1925                             cp->provider->name, cp->acr, cp->acw, cp->ace));
1926                         cp->index++;
1927                         g_io_request(cbp, cp);
1928                 }
1929                 break;
1930         case BIO_WRITE:
1931         case BIO_DELETE:
1932                 /*
1933                  * Put request onto inflight queue, so we can check if new
1934                  * synchronization requests don't collide with it.
1935                  */
1936                 bioq_insert_tail(&sc->sc_inflight, pbp);
1937
1938                 /*
1939                  * Bump syncid on first write.
1940                  */
1941                 if ((sc->sc_bump_id & G_RAID3_BUMP_SYNCID) != 0) {
1942                         sc->sc_bump_id &= ~G_RAID3_BUMP_SYNCID;
1943                         g_raid3_bump_syncid(sc);
1944                 }
1945                 g_raid3_scatter(pbp);
1946                 break;
1947         }
1948         return (0);
1949 }
1950
1951 static int
1952 g_raid3_can_destroy(struct g_raid3_softc *sc)
1953 {
1954         struct g_geom *gp;
1955         struct g_consumer *cp;
1956
1957         g_topology_assert();
1958         gp = sc->sc_geom;
1959         if (gp->softc == NULL)
1960                 return (1);
1961         LIST_FOREACH(cp, &gp->consumer, consumer) {
1962                 if (g_raid3_is_busy(sc, cp))
1963                         return (0);
1964         }
1965         gp = sc->sc_sync.ds_geom;
1966         LIST_FOREACH(cp, &gp->consumer, consumer) {
1967                 if (g_raid3_is_busy(sc, cp))
1968                         return (0);
1969         }
1970         G_RAID3_DEBUG(2, "No I/O requests for %s, it can be destroyed.",
1971             sc->sc_name);
1972         return (1);
1973 }
1974
1975 static int
1976 g_raid3_try_destroy(struct g_raid3_softc *sc)
1977 {
1978
1979         g_topology_assert_not();
1980         sx_assert(&sc->sc_lock, SX_XLOCKED);
1981
1982         if (sc->sc_rootmount != NULL) {
1983                 G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__,
1984                     sc->sc_rootmount);
1985                 root_mount_rel(sc->sc_rootmount);
1986                 sc->sc_rootmount = NULL;
1987         }
1988
1989         g_topology_lock();
1990         if (!g_raid3_can_destroy(sc)) {
1991                 g_topology_unlock();
1992                 return (0);
1993         }
1994         sc->sc_geom->softc = NULL;
1995         sc->sc_sync.ds_geom->softc = NULL;
1996         if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_WAIT) != 0) {
1997                 g_topology_unlock();
1998                 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__,
1999                     &sc->sc_worker);
2000                 /* Unlock sc_lock here, as it can be destroyed after wakeup. */
2001                 sx_xunlock(&sc->sc_lock);
2002                 wakeup(&sc->sc_worker);
2003                 sc->sc_worker = NULL;
2004         } else {
2005                 g_topology_unlock();
2006                 g_raid3_destroy_device(sc);
2007                 free(sc->sc_disks, M_RAID3);
2008                 free(sc, M_RAID3);
2009         }
2010         return (1);
2011 }
2012
2013 /*
2014  * Worker thread.
2015  */
2016 static void
2017 g_raid3_worker(void *arg)
2018 {
2019         struct g_raid3_softc *sc;
2020         struct g_raid3_event *ep;
2021         struct bio *bp;
2022         int timeout;
2023
2024         sc = arg;
2025         thread_lock(curthread);
2026         sched_prio(curthread, PRIBIO);
2027         thread_unlock(curthread);
2028
2029         sx_xlock(&sc->sc_lock);
2030         for (;;) {
2031                 G_RAID3_DEBUG(5, "%s: Let's see...", __func__);
2032                 /*
2033                  * First take a look at events.
2034                  * This is important to handle events before any I/O requests.
2035                  */
2036                 ep = g_raid3_event_get(sc);
2037                 if (ep != NULL) {
2038                         g_raid3_event_remove(sc, ep);
2039                         if ((ep->e_flags & G_RAID3_EVENT_DEVICE) != 0) {
2040                                 /* Update only device status. */
2041                                 G_RAID3_DEBUG(3,
2042                                     "Running event for device %s.",
2043                                     sc->sc_name);
2044                                 ep->e_error = 0;
2045                                 g_raid3_update_device(sc, 1);
2046                         } else {
2047                                 /* Update disk status. */
2048                                 G_RAID3_DEBUG(3, "Running event for disk %s.",
2049                                      g_raid3_get_diskname(ep->e_disk));
2050                                 ep->e_error = g_raid3_update_disk(ep->e_disk,
2051                                     ep->e_state);
2052                                 if (ep->e_error == 0)
2053                                         g_raid3_update_device(sc, 0);
2054                         }
2055                         if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0) {
2056                                 KASSERT(ep->e_error == 0,
2057                                     ("Error cannot be handled."));
2058                                 g_raid3_event_free(ep);
2059                         } else {
2060                                 ep->e_flags |= G_RAID3_EVENT_DONE;
2061                                 G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__,
2062                                     ep);
2063                                 mtx_lock(&sc->sc_events_mtx);
2064                                 wakeup(ep);
2065                                 mtx_unlock(&sc->sc_events_mtx);
2066                         }
2067                         if ((sc->sc_flags &
2068                             G_RAID3_DEVICE_FLAG_DESTROY) != 0) {
2069                                 if (g_raid3_try_destroy(sc)) {
2070                                         curthread->td_pflags &= ~TDP_GEOM;
2071                                         G_RAID3_DEBUG(1, "Thread exiting.");
2072                                         kproc_exit(0);
2073                                 }
2074                         }
2075                         G_RAID3_DEBUG(5, "%s: I'm here 1.", __func__);
2076                         continue;
2077                 }
2078                 /*
2079                  * Check if we can mark array as CLEAN and if we can't take
2080                  * how much seconds should we wait.
2081                  */
2082                 timeout = g_raid3_idle(sc, -1);
2083                 /*
2084                  * Now I/O requests.
2085                  */
2086                 /* Get first request from the queue. */
2087                 mtx_lock(&sc->sc_queue_mtx);
2088                 bp = bioq_first(&sc->sc_queue);
2089                 if (bp == NULL) {
2090                         if ((sc->sc_flags &
2091                             G_RAID3_DEVICE_FLAG_DESTROY) != 0) {
2092                                 mtx_unlock(&sc->sc_queue_mtx);
2093                                 if (g_raid3_try_destroy(sc)) {
2094                                         curthread->td_pflags &= ~TDP_GEOM;
2095                                         G_RAID3_DEBUG(1, "Thread exiting.");
2096                                         kproc_exit(0);
2097                                 }
2098                                 mtx_lock(&sc->sc_queue_mtx);
2099                         }
2100                         sx_xunlock(&sc->sc_lock);
2101                         /*
2102                          * XXX: We can miss an event here, because an event
2103                          *      can be added without sx-device-lock and without
2104                          *      mtx-queue-lock. Maybe I should just stop using
2105                          *      dedicated mutex for events synchronization and
2106                          *      stick with the queue lock?
2107                          *      The event will hang here until next I/O request
2108                          *      or next event is received.
2109                          */
2110                         MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "r3:w1",
2111                             timeout * hz);
2112                         sx_xlock(&sc->sc_lock);
2113                         G_RAID3_DEBUG(5, "%s: I'm here 4.", __func__);
2114                         continue;
2115                 }
2116 process:
2117                 bioq_remove(&sc->sc_queue, bp);
2118                 mtx_unlock(&sc->sc_queue_mtx);
2119
2120                 if (bp->bio_from->geom == sc->sc_sync.ds_geom &&
2121                     (bp->bio_cflags & G_RAID3_BIO_CFLAG_SYNC) != 0) {
2122                         g_raid3_sync_request(bp);       /* READ */
2123                 } else if (bp->bio_to != sc->sc_provider) {
2124                         if ((bp->bio_cflags & G_RAID3_BIO_CFLAG_REGULAR) != 0)
2125                                 g_raid3_regular_request(bp);
2126                         else if ((bp->bio_cflags & G_RAID3_BIO_CFLAG_SYNC) != 0)
2127                                 g_raid3_sync_request(bp);       /* WRITE */
2128                         else {
2129                                 KASSERT(0,
2130                                     ("Invalid request cflags=0x%hx to=%s.",
2131                                     bp->bio_cflags, bp->bio_to->name));
2132                         }
2133                 } else if (g_raid3_register_request(bp) != 0) {
2134                         mtx_lock(&sc->sc_queue_mtx);
2135                         bioq_insert_head(&sc->sc_queue, bp);
2136                         /*
2137                          * We are short in memory, let see if there are finished
2138                          * request we can free.
2139                          */
2140                         TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
2141                                 if (bp->bio_cflags & G_RAID3_BIO_CFLAG_REGULAR)
2142                                         goto process;
2143                         }
2144                         /*
2145                          * No finished regular request, so at least keep
2146                          * synchronization running.
2147                          */
2148                         TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
2149                                 if (bp->bio_cflags & G_RAID3_BIO_CFLAG_SYNC)
2150                                         goto process;
2151                         }
2152                         sx_xunlock(&sc->sc_lock);
2153                         MSLEEP(&sc->sc_queue, &sc->sc_queue_mtx, PRIBIO | PDROP,
2154                             "r3:lowmem", hz / 10);
2155                         sx_xlock(&sc->sc_lock);
2156                 }
2157                 G_RAID3_DEBUG(5, "%s: I'm here 9.", __func__);
2158         }
2159 }
2160
2161 static void
2162 g_raid3_update_idle(struct g_raid3_softc *sc, struct g_raid3_disk *disk)
2163 {
2164
2165         sx_assert(&sc->sc_lock, SX_LOCKED);
2166         if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) != 0)
2167                 return;
2168         if (!sc->sc_idle && (disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) == 0) {
2169                 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as dirty.",
2170                     g_raid3_get_diskname(disk), sc->sc_name);
2171                 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY;
2172         } else if (sc->sc_idle &&
2173             (disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) != 0) {
2174                 G_RAID3_DEBUG(1, "Disk %s (device %s) marked as clean.",
2175                     g_raid3_get_diskname(disk), sc->sc_name);
2176                 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY;
2177         }
2178 }
2179
2180 static void
2181 g_raid3_sync_start(struct g_raid3_softc *sc)
2182 {
2183         struct g_raid3_disk *disk;
2184         struct g_consumer *cp;
2185         struct bio *bp;
2186         int error;
2187         u_int n;
2188
2189         g_topology_assert_not();
2190         sx_assert(&sc->sc_lock, SX_XLOCKED);
2191
2192         KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED,
2193             ("Device not in DEGRADED state (%s, %u).", sc->sc_name,
2194             sc->sc_state));
2195         KASSERT(sc->sc_syncdisk == NULL, ("Syncdisk is not NULL (%s, %u).",
2196             sc->sc_name, sc->sc_state));
2197         disk = NULL;
2198         for (n = 0; n < sc->sc_ndisks; n++) {
2199                 if (sc->sc_disks[n].d_state != G_RAID3_DISK_STATE_SYNCHRONIZING)
2200                         continue;
2201                 disk = &sc->sc_disks[n];
2202                 break;
2203         }
2204         if (disk == NULL)
2205                 return;
2206
2207         sx_xunlock(&sc->sc_lock);
2208         g_topology_lock();
2209         cp = g_new_consumer(sc->sc_sync.ds_geom);
2210         error = g_attach(cp, sc->sc_provider);
2211         KASSERT(error == 0,
2212             ("Cannot attach to %s (error=%d).", sc->sc_name, error));
2213         error = g_access(cp, 1, 0, 0);
2214         KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error));
2215         g_topology_unlock();
2216         sx_xlock(&sc->sc_lock);
2217
2218         G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name,
2219             g_raid3_get_diskname(disk));
2220         if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) == 0)
2221                 disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY;
2222         KASSERT(disk->d_sync.ds_consumer == NULL,
2223             ("Sync consumer already exists (device=%s, disk=%s).",
2224             sc->sc_name, g_raid3_get_diskname(disk)));
2225
2226         disk->d_sync.ds_consumer = cp;
2227         disk->d_sync.ds_consumer->private = disk;
2228         disk->d_sync.ds_consumer->index = 0;
2229         sc->sc_syncdisk = disk;
2230
2231         /*
2232          * Allocate memory for synchronization bios and initialize them.
2233          */
2234         disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_raid3_syncreqs,
2235             M_RAID3, M_WAITOK);
2236         for (n = 0; n < g_raid3_syncreqs; n++) {
2237                 bp = g_alloc_bio();
2238                 disk->d_sync.ds_bios[n] = bp;
2239                 bp->bio_parent = NULL;
2240                 bp->bio_cmd = BIO_READ;
2241                 bp->bio_data = malloc(MAXPHYS, M_RAID3, M_WAITOK);
2242                 bp->bio_cflags = 0;
2243                 bp->bio_offset = disk->d_sync.ds_offset * (sc->sc_ndisks - 1);
2244                 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset);
2245                 disk->d_sync.ds_offset += bp->bio_length / (sc->sc_ndisks - 1);
2246                 bp->bio_done = g_raid3_sync_done;
2247                 bp->bio_from = disk->d_sync.ds_consumer;
2248                 bp->bio_to = sc->sc_provider;
2249                 bp->bio_caller1 = (void *)(uintptr_t)n;
2250         }
2251
2252         /* Set the number of in-flight synchronization requests. */
2253         disk->d_sync.ds_inflight = g_raid3_syncreqs;
2254
2255         /*
2256          * Fire off first synchronization requests.
2257          */
2258         for (n = 0; n < g_raid3_syncreqs; n++) {
2259                 bp = disk->d_sync.ds_bios[n];
2260                 G_RAID3_LOGREQ(3, bp, "Sending synchronization request.");
2261                 disk->d_sync.ds_consumer->index++;
2262                 /*
2263                  * Delay the request if it is colliding with a regular request.
2264                  */
2265                 if (g_raid3_regular_collision(sc, bp))
2266                         g_raid3_sync_delay(sc, bp);
2267                 else
2268                         g_io_request(bp, disk->d_sync.ds_consumer);
2269         }
2270 }
2271
2272 /*
2273  * Stop synchronization process.
2274  * type: 0 - synchronization finished
2275  *       1 - synchronization stopped
2276  */
2277 static void
2278 g_raid3_sync_stop(struct g_raid3_softc *sc, int type)
2279 {
2280         struct g_raid3_disk *disk;
2281         struct g_consumer *cp;
2282
2283         g_topology_assert_not();
2284         sx_assert(&sc->sc_lock, SX_LOCKED);
2285
2286         KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED,
2287             ("Device not in DEGRADED state (%s, %u).", sc->sc_name,
2288             sc->sc_state));
2289         disk = sc->sc_syncdisk;
2290         sc->sc_syncdisk = NULL;
2291         KASSERT(disk != NULL, ("No disk was synchronized (%s).", sc->sc_name));
2292         KASSERT(disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING,
2293             ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk),
2294             g_raid3_disk_state2str(disk->d_state)));
2295         if (disk->d_sync.ds_consumer == NULL)
2296                 return;
2297
2298         if (type == 0) {
2299                 G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s finished.",
2300                     sc->sc_name, g_raid3_get_diskname(disk));
2301         } else /* if (type == 1) */ {
2302                 G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s stopped.",
2303                     sc->sc_name, g_raid3_get_diskname(disk));
2304         }
2305         free(disk->d_sync.ds_bios, M_RAID3);
2306         disk->d_sync.ds_bios = NULL;
2307         cp = disk->d_sync.ds_consumer;
2308         disk->d_sync.ds_consumer = NULL;
2309         disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY;
2310         sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */
2311         g_topology_lock();
2312         g_raid3_kill_consumer(sc, cp);
2313         g_topology_unlock();
2314         sx_xlock(&sc->sc_lock);
2315 }
2316
2317 static void
2318 g_raid3_launch_provider(struct g_raid3_softc *sc)
2319 {
2320         struct g_provider *pp;
2321         struct g_raid3_disk *disk;
2322         int n;
2323
2324         sx_assert(&sc->sc_lock, SX_LOCKED);
2325
2326         g_topology_lock();
2327         pp = g_new_providerf(sc->sc_geom, "raid3/%s", sc->sc_name);
2328         pp->mediasize = sc->sc_mediasize;
2329         pp->sectorsize = sc->sc_sectorsize;
2330         pp->stripesize = 0;
2331         pp->stripeoffset = 0;
2332         for (n = 0; n < sc->sc_ndisks; n++) {
2333                 disk = &sc->sc_disks[n];
2334                 if (disk->d_consumer && disk->d_consumer->provider &&
2335                     disk->d_consumer->provider->stripesize > pp->stripesize) {
2336                         pp->stripesize = disk->d_consumer->provider->stripesize;
2337                         pp->stripeoffset = disk->d_consumer->provider->stripeoffset;
2338                 }
2339         }
2340         pp->stripesize *= sc->sc_ndisks - 1;
2341         pp->stripeoffset *= sc->sc_ndisks - 1;
2342         sc->sc_provider = pp;
2343         g_error_provider(pp, 0);
2344         g_topology_unlock();
2345         G_RAID3_DEBUG(0, "Device %s launched (%u/%u).", pp->name,
2346             g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE), sc->sc_ndisks);
2347
2348         if (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED)
2349                 g_raid3_sync_start(sc);
2350 }
2351
2352 static void
2353 g_raid3_destroy_provider(struct g_raid3_softc *sc)
2354 {
2355         struct bio *bp;
2356
2357         g_topology_assert_not();
2358         KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).",
2359             sc->sc_name));
2360
2361         g_topology_lock();
2362         g_error_provider(sc->sc_provider, ENXIO);
2363         mtx_lock(&sc->sc_queue_mtx);
2364         while ((bp = bioq_first(&sc->sc_queue)) != NULL) {
2365                 bioq_remove(&sc->sc_queue, bp);
2366                 g_io_deliver(bp, ENXIO);
2367         }
2368         mtx_unlock(&sc->sc_queue_mtx);
2369         G_RAID3_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name,
2370             sc->sc_provider->name);
2371         sc->sc_provider->flags |= G_PF_WITHER;
2372         g_orphan_provider(sc->sc_provider, ENXIO);
2373         g_topology_unlock();
2374         sc->sc_provider = NULL;
2375         if (sc->sc_syncdisk != NULL)
2376                 g_raid3_sync_stop(sc, 1);
2377 }
2378
2379 static void
2380 g_raid3_go(void *arg)
2381 {
2382         struct g_raid3_softc *sc;
2383
2384         sc = arg;
2385         G_RAID3_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name);
2386         g_raid3_event_send(sc, 0,
2387             G_RAID3_EVENT_DONTWAIT | G_RAID3_EVENT_DEVICE);
2388 }
2389
2390 static u_int
2391 g_raid3_determine_state(struct g_raid3_disk *disk)
2392 {
2393         struct g_raid3_softc *sc;
2394         u_int state;
2395
2396         sc = disk->d_softc;
2397         if (sc->sc_syncid == disk->d_sync.ds_syncid) {
2398                 if ((disk->d_flags &
2399                     G_RAID3_DISK_FLAG_SYNCHRONIZING) == 0) {
2400                         /* Disk does not need synchronization. */
2401                         state = G_RAID3_DISK_STATE_ACTIVE;
2402                 } else {
2403                         if ((sc->sc_flags &
2404                              G_RAID3_DEVICE_FLAG_NOAUTOSYNC) == 0 ||
2405                             (disk->d_flags &
2406                              G_RAID3_DISK_FLAG_FORCE_SYNC) != 0) {
2407                                 /*
2408                                  * We can start synchronization from
2409                                  * the stored offset.
2410                                  */
2411                                 state = G_RAID3_DISK_STATE_SYNCHRONIZING;
2412                         } else {
2413                                 state = G_RAID3_DISK_STATE_STALE;
2414                         }
2415                 }
2416         } else if (disk->d_sync.ds_syncid < sc->sc_syncid) {
2417                 /*
2418                  * Reset all synchronization data for this disk,
2419                  * because if it even was synchronized, it was
2420                  * synchronized to disks with different syncid.
2421                  */
2422                 disk->d_flags |= G_RAID3_DISK_FLAG_SYNCHRONIZING;
2423                 disk->d_sync.ds_offset = 0;
2424                 disk->d_sync.ds_offset_done = 0;
2425                 disk->d_sync.ds_syncid = sc->sc_syncid;
2426                 if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOAUTOSYNC) == 0 ||
2427                     (disk->d_flags & G_RAID3_DISK_FLAG_FORCE_SYNC) != 0) {
2428                         state = G_RAID3_DISK_STATE_SYNCHRONIZING;
2429                 } else {
2430                         state = G_RAID3_DISK_STATE_STALE;
2431                 }
2432         } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ {
2433                 /*
2434                  * Not good, NOT GOOD!
2435                  * It means that device was started on stale disks
2436                  * and more fresh disk just arrive.
2437                  * If there were writes, device is broken, sorry.
2438                  * I think the best choice here is don't touch
2439                  * this disk and inform the user loudly.
2440                  */
2441                 G_RAID3_DEBUG(0, "Device %s was started before the freshest "
2442                     "disk (%s) arrives!! It will not be connected to the "
2443                     "running device.", sc->sc_name,
2444                     g_raid3_get_diskname(disk));
2445                 g_raid3_destroy_disk(disk);
2446                 state = G_RAID3_DISK_STATE_NONE;
2447                 /* Return immediately, because disk was destroyed. */
2448                 return (state);
2449         }
2450         G_RAID3_DEBUG(3, "State for %s disk: %s.",
2451             g_raid3_get_diskname(disk), g_raid3_disk_state2str(state));
2452         return (state);
2453 }
2454
2455 /*
2456  * Update device state.
2457  */
2458 static void
2459 g_raid3_update_device(struct g_raid3_softc *sc, boolean_t force)
2460 {
2461         struct g_raid3_disk *disk;
2462         u_int state;
2463
2464         sx_assert(&sc->sc_lock, SX_XLOCKED);
2465
2466         switch (sc->sc_state) {
2467         case G_RAID3_DEVICE_STATE_STARTING:
2468             {
2469                 u_int n, ndirty, ndisks, genid, syncid;
2470
2471                 KASSERT(sc->sc_provider == NULL,
2472                     ("Non-NULL provider in STARTING state (%s).", sc->sc_name));
2473                 /*
2474                  * Are we ready? We are, if all disks are connected or
2475                  * one disk is missing and 'force' is true.
2476                  */
2477                 if (g_raid3_ndisks(sc, -1) + force == sc->sc_ndisks) {
2478                         if (!force)
2479                                 callout_drain(&sc->sc_callout);
2480                 } else {
2481                         if (force) {
2482                                 /*
2483                                  * Timeout expired, so destroy device.
2484                                  */
2485                                 sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY;
2486                                 G_RAID3_DEBUG(1, "root_mount_rel[%u] %p",
2487                                     __LINE__, sc->sc_rootmount);
2488                                 root_mount_rel(sc->sc_rootmount);
2489                                 sc->sc_rootmount = NULL;
2490                         }
2491                         return;
2492                 }
2493
2494                 /*
2495                  * Find the biggest genid.
2496                  */
2497                 genid = 0;
2498                 for (n = 0; n < sc->sc_ndisks; n++) {
2499                         disk = &sc->sc_disks[n];
2500                         if (disk->d_state == G_RAID3_DISK_STATE_NODISK)
2501                                 continue;
2502                         if (disk->d_genid > genid)
2503                                 genid = disk->d_genid;
2504                 }
2505                 sc->sc_genid = genid;
2506                 /*
2507                  * Remove all disks without the biggest genid.
2508                  */
2509                 for (n = 0; n < sc->sc_ndisks; n++) {
2510                         disk = &sc->sc_disks[n];
2511                         if (disk->d_state == G_RAID3_DISK_STATE_NODISK)
2512                                 continue;
2513                         if (disk->d_genid < genid) {
2514                                 G_RAID3_DEBUG(0,
2515                                     "Component %s (device %s) broken, skipping.",
2516                                     g_raid3_get_diskname(disk), sc->sc_name);
2517                                 g_raid3_destroy_disk(disk);
2518                         }
2519                 }
2520
2521                 /*
2522                  * There must be at least 'sc->sc_ndisks - 1' components
2523                  * with the same syncid and without SYNCHRONIZING flag.
2524                  */
2525
2526                 /*
2527                  * Find the biggest syncid, number of valid components and
2528                  * number of dirty components.
2529                  */
2530                 ndirty = ndisks = syncid = 0;
2531                 for (n = 0; n < sc->sc_ndisks; n++) {
2532                         disk = &sc->sc_disks[n];
2533                         if (disk->d_state == G_RAID3_DISK_STATE_NODISK)
2534                                 continue;
2535                         if ((disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) != 0)
2536                                 ndirty++;
2537                         if (disk->d_sync.ds_syncid > syncid) {
2538                                 syncid = disk->d_sync.ds_syncid;
2539                                 ndisks = 0;
2540                         } else if (disk->d_sync.ds_syncid < syncid) {
2541                                 continue;
2542                         }
2543                         if ((disk->d_flags &
2544                             G_RAID3_DISK_FLAG_SYNCHRONIZING) != 0) {
2545                                 continue;
2546                         }
2547                         ndisks++;
2548                 }
2549                 /*
2550                  * Do we have enough valid components?
2551                  */
2552                 if (ndisks + 1 < sc->sc_ndisks) {
2553                         G_RAID3_DEBUG(0,
2554                             "Device %s is broken, too few valid components.",
2555                             sc->sc_name);
2556                         sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY;
2557                         return;
2558                 }
2559                 /*
2560                  * If there is one DIRTY component and all disks are present,
2561                  * mark it for synchronization. If there is more than one DIRTY
2562                  * component, mark parity component for synchronization.
2563                  */
2564                 if (ndisks == sc->sc_ndisks && ndirty == 1) {
2565                         for (n = 0; n < sc->sc_ndisks; n++) {
2566                                 disk = &sc->sc_disks[n];
2567                                 if ((disk->d_flags &
2568                                     G_RAID3_DISK_FLAG_DIRTY) == 0) {
2569                                         continue;
2570                                 }
2571                                 disk->d_flags |=
2572                                     G_RAID3_DISK_FLAG_SYNCHRONIZING;
2573                         }
2574                 } else if (ndisks == sc->sc_ndisks && ndirty > 1) {
2575                         disk = &sc->sc_disks[sc->sc_ndisks - 1];
2576                         disk->d_flags |= G_RAID3_DISK_FLAG_SYNCHRONIZING;
2577                 }
2578
2579                 sc->sc_syncid = syncid;
2580                 if (force) {
2581                         /* Remember to bump syncid on first write. */
2582                         sc->sc_bump_id |= G_RAID3_BUMP_SYNCID;
2583                 }
2584                 if (ndisks == sc->sc_ndisks)
2585                         state = G_RAID3_DEVICE_STATE_COMPLETE;
2586                 else /* if (ndisks == sc->sc_ndisks - 1) */
2587                         state = G_RAID3_DEVICE_STATE_DEGRADED;
2588                 G_RAID3_DEBUG(1, "Device %s state changed from %s to %s.",
2589                     sc->sc_name, g_raid3_device_state2str(sc->sc_state),
2590                     g_raid3_device_state2str(state));
2591                 sc->sc_state = state;
2592                 for (n = 0; n < sc->sc_ndisks; n++) {
2593                         disk = &sc->sc_disks[n];
2594                         if (disk->d_state == G_RAID3_DISK_STATE_NODISK)
2595                                 continue;
2596                         state = g_raid3_determine_state(disk);
2597                         g_raid3_event_send(disk, state, G_RAID3_EVENT_DONTWAIT);
2598                         if (state == G_RAID3_DISK_STATE_STALE)
2599                                 sc->sc_bump_id |= G_RAID3_BUMP_SYNCID;
2600                 }
2601                 break;
2602             }
2603         case G_RAID3_DEVICE_STATE_DEGRADED:
2604                 /*
2605                  * Genid need to be bumped immediately, so do it here.
2606                  */
2607                 if ((sc->sc_bump_id & G_RAID3_BUMP_GENID) != 0) {
2608                         sc->sc_bump_id &= ~G_RAID3_BUMP_GENID;
2609                         g_raid3_bump_genid(sc);
2610                 }
2611
2612                 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NEW) > 0)
2613                         return;
2614                 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) <
2615                     sc->sc_ndisks - 1) {
2616                         if (sc->sc_provider != NULL)
2617                                 g_raid3_destroy_provider(sc);
2618                         sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY;
2619                         return;
2620                 }
2621                 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) ==
2622                     sc->sc_ndisks) {
2623                         state = G_RAID3_DEVICE_STATE_COMPLETE;
2624                         G_RAID3_DEBUG(1,
2625                             "Device %s state changed from %s to %s.",
2626                             sc->sc_name, g_raid3_device_state2str(sc->sc_state),
2627                             g_raid3_device_state2str(state));
2628                         sc->sc_state = state;
2629                 }
2630                 if (sc->sc_provider == NULL)
2631                         g_raid3_launch_provider(sc);
2632                 if (sc->sc_rootmount != NULL) {
2633                         G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__,
2634                             sc->sc_rootmount);
2635                         root_mount_rel(sc->sc_rootmount);
2636                         sc->sc_rootmount = NULL;
2637                 }
2638                 break;
2639         case G_RAID3_DEVICE_STATE_COMPLETE:
2640                 /*
2641                  * Genid need to be bumped immediately, so do it here.
2642                  */
2643                 if ((sc->sc_bump_id & G_RAID3_BUMP_GENID) != 0) {
2644                         sc->sc_bump_id &= ~G_RAID3_BUMP_GENID;
2645                         g_raid3_bump_genid(sc);
2646                 }
2647
2648                 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NEW) > 0)
2649                         return;
2650                 KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) >=
2651                     sc->sc_ndisks - 1,
2652                     ("Too few ACTIVE components in COMPLETE state (device %s).",
2653                     sc->sc_name));
2654                 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) ==
2655                     sc->sc_ndisks - 1) {
2656                         state = G_RAID3_DEVICE_STATE_DEGRADED;
2657                         G_RAID3_DEBUG(1,
2658                             "Device %s state changed from %s to %s.",
2659                             sc->sc_name, g_raid3_device_state2str(sc->sc_state),
2660                             g_raid3_device_state2str(state));
2661                         sc->sc_state = state;
2662                 }
2663                 if (sc->sc_provider == NULL)
2664                         g_raid3_launch_provider(sc);
2665                 if (sc->sc_rootmount != NULL) {
2666                         G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__,
2667                             sc->sc_rootmount);
2668                         root_mount_rel(sc->sc_rootmount);
2669                         sc->sc_rootmount = NULL;
2670                 }
2671                 break;
2672         default:
2673                 KASSERT(1 == 0, ("Wrong device state (%s, %s).", sc->sc_name,
2674                     g_raid3_device_state2str(sc->sc_state)));
2675                 break;
2676         }
2677 }
2678
2679 /*
2680  * Update disk state and device state if needed.
2681  */
2682 #define DISK_STATE_CHANGED()    G_RAID3_DEBUG(1,                        \
2683         "Disk %s state changed from %s to %s (device %s).",             \
2684         g_raid3_get_diskname(disk),                                     \
2685         g_raid3_disk_state2str(disk->d_state),                          \
2686         g_raid3_disk_state2str(state), sc->sc_name)
2687 static int
2688 g_raid3_update_disk(struct g_raid3_disk *disk, u_int state)
2689 {
2690         struct g_raid3_softc *sc;
2691
2692         sc = disk->d_softc;
2693         sx_assert(&sc->sc_lock, SX_XLOCKED);
2694
2695 again:
2696         G_RAID3_DEBUG(3, "Changing disk %s state from %s to %s.",
2697             g_raid3_get_diskname(disk), g_raid3_disk_state2str(disk->d_state),
2698             g_raid3_disk_state2str(state));
2699         switch (state) {
2700         case G_RAID3_DISK_STATE_NEW:
2701                 /*
2702                  * Possible scenarios:
2703                  * 1. New disk arrive.
2704                  */
2705                 /* Previous state should be NONE. */
2706                 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NONE,
2707                     ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk),
2708                     g_raid3_disk_state2str(disk->d_state)));
2709                 DISK_STATE_CHANGED();
2710
2711                 disk->d_state = state;
2712                 G_RAID3_DEBUG(1, "Device %s: provider %s detected.",
2713                     sc->sc_name, g_raid3_get_diskname(disk));
2714                 if (sc->sc_state == G_RAID3_DEVICE_STATE_STARTING)
2715                         break;
2716                 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED ||
2717                     sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE,
2718                     ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2719                     g_raid3_device_state2str(sc->sc_state),
2720                     g_raid3_get_diskname(disk),
2721                     g_raid3_disk_state2str(disk->d_state)));
2722                 state = g_raid3_determine_state(disk);
2723                 if (state != G_RAID3_DISK_STATE_NONE)
2724                         goto again;
2725                 break;
2726         case G_RAID3_DISK_STATE_ACTIVE:
2727                 /*
2728                  * Possible scenarios:
2729                  * 1. New disk does not need synchronization.
2730                  * 2. Synchronization process finished successfully.
2731                  */
2732                 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED ||
2733                     sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE,
2734                     ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2735                     g_raid3_device_state2str(sc->sc_state),
2736                     g_raid3_get_diskname(disk),
2737                     g_raid3_disk_state2str(disk->d_state)));
2738                 /* Previous state should be NEW or SYNCHRONIZING. */
2739                 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW ||
2740                     disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING,
2741                     ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk),
2742                     g_raid3_disk_state2str(disk->d_state)));
2743                 DISK_STATE_CHANGED();
2744
2745                 if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) {
2746                         disk->d_flags &= ~G_RAID3_DISK_FLAG_SYNCHRONIZING;
2747                         disk->d_flags &= ~G_RAID3_DISK_FLAG_FORCE_SYNC;
2748                         g_raid3_sync_stop(sc, 0);
2749                 }
2750                 disk->d_state = state;
2751                 disk->d_sync.ds_offset = 0;
2752                 disk->d_sync.ds_offset_done = 0;
2753                 g_raid3_update_idle(sc, disk);
2754                 g_raid3_update_metadata(disk);
2755                 G_RAID3_DEBUG(1, "Device %s: provider %s activated.",
2756                     sc->sc_name, g_raid3_get_diskname(disk));
2757                 break;
2758         case G_RAID3_DISK_STATE_STALE:
2759                 /*
2760                  * Possible scenarios:
2761                  * 1. Stale disk was connected.
2762                  */
2763                 /* Previous state should be NEW. */
2764                 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW,
2765                     ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk),
2766                     g_raid3_disk_state2str(disk->d_state)));
2767                 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED ||
2768                     sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE,
2769                     ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2770                     g_raid3_device_state2str(sc->sc_state),
2771                     g_raid3_get_diskname(disk),
2772                     g_raid3_disk_state2str(disk->d_state)));
2773                 /*
2774                  * STALE state is only possible if device is marked
2775                  * NOAUTOSYNC.
2776                  */
2777                 KASSERT((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOAUTOSYNC) != 0,
2778                     ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2779                     g_raid3_device_state2str(sc->sc_state),
2780                     g_raid3_get_diskname(disk),
2781                     g_raid3_disk_state2str(disk->d_state)));
2782                 DISK_STATE_CHANGED();
2783
2784                 disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY;
2785                 disk->d_state = state;
2786                 g_raid3_update_metadata(disk);
2787                 G_RAID3_DEBUG(0, "Device %s: provider %s is stale.",
2788                     sc->sc_name, g_raid3_get_diskname(disk));
2789                 break;
2790         case G_RAID3_DISK_STATE_SYNCHRONIZING:
2791                 /*
2792                  * Possible scenarios:
2793                  * 1. Disk which needs synchronization was connected.
2794                  */
2795                 /* Previous state should be NEW. */
2796                 KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW,
2797                     ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk),
2798                     g_raid3_disk_state2str(disk->d_state)));
2799                 KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED ||
2800                     sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE,
2801                     ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2802                     g_raid3_device_state2str(sc->sc_state),
2803                     g_raid3_get_diskname(disk),
2804                     g_raid3_disk_state2str(disk->d_state)));
2805                 DISK_STATE_CHANGED();
2806
2807                 if (disk->d_state == G_RAID3_DISK_STATE_NEW)
2808                         disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY;
2809                 disk->d_state = state;
2810                 if (sc->sc_provider != NULL) {
2811                         g_raid3_sync_start(sc);
2812                         g_raid3_update_metadata(disk);
2813                 }
2814                 break;
2815         case G_RAID3_DISK_STATE_DISCONNECTED:
2816                 /*
2817                  * Possible scenarios:
2818                  * 1. Device wasn't running yet, but disk disappear.
2819                  * 2. Disk was active and disapppear.
2820                  * 3. Disk disappear during synchronization process.
2821                  */
2822                 if (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED ||
2823                     sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) {
2824                         /*
2825                          * Previous state should be ACTIVE, STALE or
2826                          * SYNCHRONIZING.
2827                          */
2828                         KASSERT(disk->d_state == G_RAID3_DISK_STATE_ACTIVE ||
2829                             disk->d_state == G_RAID3_DISK_STATE_STALE ||
2830                             disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING,
2831                             ("Wrong disk state (%s, %s).",
2832                             g_raid3_get_diskname(disk),
2833                             g_raid3_disk_state2str(disk->d_state)));
2834                 } else if (sc->sc_state == G_RAID3_DEVICE_STATE_STARTING) {
2835                         /* Previous state should be NEW. */
2836                         KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW,
2837                             ("Wrong disk state (%s, %s).",
2838                             g_raid3_get_diskname(disk),
2839                             g_raid3_disk_state2str(disk->d_state)));
2840                         /*
2841                          * Reset bumping syncid if disk disappeared in STARTING
2842                          * state.
2843                          */
2844                         if ((sc->sc_bump_id & G_RAID3_BUMP_SYNCID) != 0)
2845                                 sc->sc_bump_id &= ~G_RAID3_BUMP_SYNCID;
2846 #ifdef  INVARIANTS
2847                 } else {
2848                         KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).",
2849                             sc->sc_name,
2850                             g_raid3_device_state2str(sc->sc_state),
2851                             g_raid3_get_diskname(disk),
2852                             g_raid3_disk_state2str(disk->d_state)));
2853 #endif
2854                 }
2855                 DISK_STATE_CHANGED();
2856                 G_RAID3_DEBUG(0, "Device %s: provider %s disconnected.",
2857                     sc->sc_name, g_raid3_get_diskname(disk));
2858
2859                 g_raid3_destroy_disk(disk);
2860                 break;
2861         default:
2862                 KASSERT(1 == 0, ("Unknown state (%u).", state));
2863                 break;
2864         }
2865         return (0);
2866 }
2867 #undef  DISK_STATE_CHANGED
2868
2869 int
2870 g_raid3_read_metadata(struct g_consumer *cp, struct g_raid3_metadata *md)
2871 {
2872         struct g_provider *pp;
2873         u_char *buf;
2874         int error;
2875
2876         g_topology_assert();
2877
2878         error = g_access(cp, 1, 0, 0);
2879         if (error != 0)
2880                 return (error);
2881         pp = cp->provider;
2882         g_topology_unlock();
2883         /* Metadata are stored on last sector. */
2884         buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize,
2885             &error);
2886         g_topology_lock();
2887         g_access(cp, -1, 0, 0);
2888         if (buf == NULL) {
2889                 G_RAID3_DEBUG(1, "Cannot read metadata from %s (error=%d).",
2890                     cp->provider->name, error);
2891                 return (error);
2892         }
2893
2894         /* Decode metadata. */
2895         error = raid3_metadata_decode(buf, md);
2896         g_free(buf);
2897         if (strcmp(md->md_magic, G_RAID3_MAGIC) != 0)
2898                 return (EINVAL);
2899         if (md->md_version > G_RAID3_VERSION) {
2900                 G_RAID3_DEBUG(0,
2901                     "Kernel module is too old to handle metadata from %s.",
2902                     cp->provider->name);
2903                 return (EINVAL);
2904         }
2905         if (error != 0) {
2906                 G_RAID3_DEBUG(1, "MD5 metadata hash mismatch for provider %s.",
2907                     cp->provider->name);
2908                 return (error);
2909         }
2910         if (md->md_sectorsize > MAXPHYS) {
2911                 G_RAID3_DEBUG(0, "The blocksize is too big.");
2912                 return (EINVAL);
2913         }
2914
2915         return (0);
2916 }
2917
2918 static int
2919 g_raid3_check_metadata(struct g_raid3_softc *sc, struct g_provider *pp,
2920     struct g_raid3_metadata *md)
2921 {
2922
2923         if (md->md_no >= sc->sc_ndisks) {
2924                 G_RAID3_DEBUG(1, "Invalid disk %s number (no=%u), skipping.",
2925                     pp->name, md->md_no);
2926                 return (EINVAL);
2927         }
2928         if (sc->sc_disks[md->md_no].d_state != G_RAID3_DISK_STATE_NODISK) {
2929                 G_RAID3_DEBUG(1, "Disk %s (no=%u) already exists, skipping.",
2930                     pp->name, md->md_no);
2931                 return (EEXIST);
2932         }
2933         if (md->md_all != sc->sc_ndisks) {
2934                 G_RAID3_DEBUG(1,
2935                     "Invalid '%s' field on disk %s (device %s), skipping.",
2936                     "md_all", pp->name, sc->sc_name);
2937                 return (EINVAL);
2938         }
2939         if ((md->md_mediasize % md->md_sectorsize) != 0) {
2940                 G_RAID3_DEBUG(1, "Invalid metadata (mediasize %% sectorsize != "
2941                     "0) on disk %s (device %s), skipping.", pp->name,
2942                     sc->sc_name);
2943                 return (EINVAL);
2944         }
2945         if (md->md_mediasize != sc->sc_mediasize) {
2946                 G_RAID3_DEBUG(1,
2947                     "Invalid '%s' field on disk %s (device %s), skipping.",
2948                     "md_mediasize", pp->name, sc->sc_name);
2949                 return (EINVAL);
2950         }
2951         if ((md->md_mediasize % (sc->sc_ndisks - 1)) != 0) {
2952                 G_RAID3_DEBUG(1,
2953                     "Invalid '%s' field on disk %s (device %s), skipping.",
2954                     "md_mediasize", pp->name, sc->sc_name);
2955                 return (EINVAL);
2956         }
2957         if ((sc->sc_mediasize / (sc->sc_ndisks - 1)) > pp->mediasize) {
2958                 G_RAID3_DEBUG(1,
2959                     "Invalid size of disk %s (device %s), skipping.", pp->name,
2960                     sc->sc_name);
2961                 return (EINVAL);
2962         }
2963         if ((md->md_sectorsize / pp->sectorsize) < sc->sc_ndisks - 1) {
2964                 G_RAID3_DEBUG(1,
2965                     "Invalid '%s' field on disk %s (device %s), skipping.",
2966                     "md_sectorsize", pp->name, sc->sc_name);
2967                 return (EINVAL);
2968         }
2969         if (md->md_sectorsize != sc->sc_sectorsize) {
2970                 G_RAID3_DEBUG(1,
2971                     "Invalid '%s' field on disk %s (device %s), skipping.",
2972                     "md_sectorsize", pp->name, sc->sc_name);
2973                 return (EINVAL);
2974         }
2975         if ((sc->sc_sectorsize % pp->sectorsize) != 0) {
2976                 G_RAID3_DEBUG(1,
2977                     "Invalid sector size of disk %s (device %s), skipping.",
2978                     pp->name, sc->sc_name);
2979                 return (EINVAL);
2980         }
2981         if ((md->md_mflags & ~G_RAID3_DEVICE_FLAG_MASK) != 0) {
2982                 G_RAID3_DEBUG(1,
2983                     "Invalid device flags on disk %s (device %s), skipping.",
2984                     pp->name, sc->sc_name);
2985                 return (EINVAL);
2986         }
2987         if ((md->md_mflags & G_RAID3_DEVICE_FLAG_VERIFY) != 0 &&
2988             (md->md_mflags & G_RAID3_DEVICE_FLAG_ROUND_ROBIN) != 0) {
2989                 /*
2990                  * VERIFY and ROUND-ROBIN options are mutally exclusive.
2991                  */
2992                 G_RAID3_DEBUG(1, "Both VERIFY and ROUND-ROBIN flags exist on "
2993                     "disk %s (device %s), skipping.", pp->name, sc->sc_name);
2994                 return (EINVAL);
2995         }
2996         if ((md->md_dflags & ~G_RAID3_DISK_FLAG_MASK) != 0) {
2997                 G_RAID3_DEBUG(1,
2998                     "Invalid disk flags on disk %s (device %s), skipping.",
2999                     pp->name, sc->sc_name);
3000                 return (EINVAL);
3001         }
3002         return (0);
3003 }
3004
3005 int
3006 g_raid3_add_disk(struct g_raid3_softc *sc, struct g_provider *pp,
3007     struct g_raid3_metadata *md)
3008 {
3009         struct g_raid3_disk *disk;
3010         int error;
3011
3012         g_topology_assert_not();
3013         G_RAID3_DEBUG(2, "Adding disk %s.", pp->name);
3014
3015         error = g_raid3_check_metadata(sc, pp, md);
3016         if (error != 0)
3017                 return (error);
3018         if (sc->sc_state != G_RAID3_DEVICE_STATE_STARTING &&
3019             md->md_genid < sc->sc_genid) {
3020                 G_RAID3_DEBUG(0, "Component %s (device %s) broken, skipping.",
3021                     pp->name, sc->sc_name);
3022                 return (EINVAL);
3023         }
3024         disk = g_raid3_init_disk(sc, pp, md, &error);
3025         if (disk == NULL)
3026                 return (error);
3027         error = g_raid3_event_send(disk, G_RAID3_DISK_STATE_NEW,
3028             G_RAID3_EVENT_WAIT);
3029         if (error != 0)
3030                 return (error);
3031         if (md->md_version < G_RAID3_VERSION) {
3032                 G_RAID3_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).",
3033                     pp->name, md->md_version, G_RAID3_VERSION);
3034                 g_raid3_update_metadata(disk);
3035         }
3036         return (0);
3037 }
3038
3039 static void
3040 g_raid3_destroy_delayed(void *arg, int flag)
3041 {
3042         struct g_raid3_softc *sc;
3043         int error;
3044
3045         if (flag == EV_CANCEL) {
3046                 G_RAID3_DEBUG(1, "Destroying canceled.");
3047                 return;
3048         }
3049         sc = arg;
3050         g_topology_unlock();
3051         sx_xlock(&sc->sc_lock);
3052         KASSERT((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) == 0,
3053             ("DESTROY flag set on %s.", sc->sc_name));
3054         KASSERT((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROYING) != 0,
3055             ("DESTROYING flag not set on %s.", sc->sc_name));
3056         G_RAID3_DEBUG(0, "Destroying %s (delayed).", sc->sc_name);
3057         error = g_raid3_destroy(sc, G_RAID3_DESTROY_SOFT);
3058         if (error != 0) {
3059                 G_RAID3_DEBUG(0, "Cannot destroy %s.", sc->sc_name);
3060                 sx_xunlock(&sc->sc_lock);
3061         }
3062         g_topology_lock();
3063 }
3064
3065 static int
3066 g_raid3_access(struct g_provider *pp, int acr, int acw, int ace)
3067 {
3068         struct g_raid3_softc *sc;
3069         int dcr, dcw, dce, error = 0;
3070
3071         g_topology_assert();
3072         G_RAID3_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr,
3073             acw, ace);
3074
3075         sc = pp->geom->softc;
3076         if (sc == NULL && acr <= 0 && acw <= 0 && ace <= 0)
3077                 return (0);
3078         KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name));
3079
3080         dcr = pp->acr + acr;
3081         dcw = pp->acw + acw;
3082         dce = pp->ace + ace;
3083
3084         g_topology_unlock();
3085         sx_xlock(&sc->sc_lock);
3086         if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) != 0 ||
3087             g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) < sc->sc_ndisks - 1) {
3088                 if (acr > 0 || acw > 0 || ace > 0)
3089                         error = ENXIO;
3090                 goto end;
3091         }
3092         if (dcw == 0)
3093                 g_raid3_idle(sc, dcw);
3094         if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROYING) != 0) {
3095                 if (acr > 0 || acw > 0 || ace > 0) {
3096                         error = ENXIO;
3097                         goto end;
3098                 }
3099                 if (dcr == 0 && dcw == 0 && dce == 0) {
3100                         g_post_event(g_raid3_destroy_delayed, sc, M_WAITOK,
3101                             sc, NULL);
3102                 }
3103         }
3104 end:
3105         sx_xunlock(&sc->sc_lock);
3106         g_topology_lock();
3107         return (error);
3108 }
3109
3110 static struct g_geom *
3111 g_raid3_create(struct g_class *mp, const struct g_raid3_metadata *md)
3112 {
3113         struct g_raid3_softc *sc;
3114         struct g_geom *gp;
3115         int error, timeout;
3116         u_int n;
3117
3118         g_topology_assert();
3119         G_RAID3_DEBUG(1, "Creating device %s (id=%u).", md->md_name, md->md_id);
3120
3121         /* One disk is minimum. */
3122         if (md->md_all < 1)
3123                 return (NULL);
3124         /*
3125          * Action geom.
3126          */
3127         gp = g_new_geomf(mp, "%s", md->md_name);
3128         sc = malloc(sizeof(*sc), M_RAID3, M_WAITOK | M_ZERO);
3129         sc->sc_disks = malloc(sizeof(struct g_raid3_disk) * md->md_all, M_RAID3,
3130             M_WAITOK | M_ZERO);
3131         gp->start = g_raid3_start;
3132         gp->orphan = g_raid3_orphan;
3133         gp->access = g_raid3_access;
3134         gp->dumpconf = g_raid3_dumpconf;
3135
3136         sc->sc_id = md->md_id;
3137         sc->sc_mediasize = md->md_mediasize;
3138         sc->sc_sectorsize = md->md_sectorsize;
3139         sc->sc_ndisks = md->md_all;
3140         sc->sc_round_robin = 0;
3141         sc->sc_flags = md->md_mflags;
3142         sc->sc_bump_id = 0;
3143         sc->sc_idle = 1;
3144         sc->sc_last_write = time_uptime;
3145         sc->sc_writes = 0;
3146         for (n = 0; n < sc->sc_ndisks; n++) {
3147                 sc->sc_disks[n].d_softc = sc;
3148                 sc->sc_disks[n].d_no = n;
3149                 sc->sc_disks[n].d_state = G_RAID3_DISK_STATE_NODISK;
3150         }
3151         sx_init(&sc->sc_lock, "graid3:lock");
3152         bioq_init(&sc->sc_queue);
3153         mtx_init(&sc->sc_queue_mtx, "graid3:queue", NULL, MTX_DEF);
3154         bioq_init(&sc->sc_regular_delayed);
3155         bioq_init(&sc->sc_inflight);
3156         bioq_init(&sc->sc_sync_delayed);
3157         TAILQ_INIT(&sc->sc_events);
3158         mtx_init(&sc->sc_events_mtx, "graid3:events", NULL, MTX_DEF);
3159         callout_init(&sc->sc_callout, 1);
3160         sc->sc_state = G_RAID3_DEVICE_STATE_STARTING;
3161         gp->softc = sc;
3162         sc->sc_geom = gp;
3163         sc->sc_provider = NULL;
3164         /*
3165          * Synchronization geom.
3166          */
3167         gp = g_new_geomf(mp, "%s.sync", md->md_name);
3168         gp->softc = sc;
3169         gp->orphan = g_raid3_orphan;
3170         sc->sc_sync.ds_geom = gp;
3171
3172         if (!g_raid3_use_malloc) {
3173                 sc->sc_zones[G_RAID3_ZONE_64K].sz_zone = uma_zcreate("gr3:64k",
3174                     65536, g_raid3_uma_ctor, g_raid3_uma_dtor, NULL, NULL,
3175                     UMA_ALIGN_PTR, 0);
3176                 sc->sc_zones[G_RAID3_ZONE_64K].sz_inuse = 0;
3177                 sc->sc_zones[G_RAID3_ZONE_64K].sz_max = g_raid3_n64k;
3178                 sc->sc_zones[G_RAID3_ZONE_64K].sz_requested =
3179                     sc->sc_zones[G_RAID3_ZONE_64K].sz_failed = 0;
3180                 sc->sc_zones[G_RAID3_ZONE_16K].sz_zone = uma_zcreate("gr3:16k",
3181                     16384, g_raid3_uma_ctor, g_raid3_uma_dtor, NULL, NULL,
3182                     UMA_ALIGN_PTR, 0);
3183                 sc->sc_zones[G_RAID3_ZONE_16K].sz_inuse = 0;
3184                 sc->sc_zones[G_RAID3_ZONE_16K].sz_max = g_raid3_n16k;
3185                 sc->sc_zones[G_RAID3_ZONE_16K].sz_requested =
3186                     sc->sc_zones[G_RAID3_ZONE_16K].sz_failed = 0;
3187                 sc->sc_zones[G_RAID3_ZONE_4K].sz_zone = uma_zcreate("gr3:4k",
3188                     4096, g_raid3_uma_ctor, g_raid3_uma_dtor, NULL, NULL,
3189                     UMA_ALIGN_PTR, 0);
3190                 sc->sc_zones[G_RAID3_ZONE_4K].sz_inuse = 0;
3191                 sc->sc_zones[G_RAID3_ZONE_4K].sz_max = g_raid3_n4k;
3192                 sc->sc_zones[G_RAID3_ZONE_4K].sz_requested =
3193                     sc->sc_zones[G_RAID3_ZONE_4K].sz_failed = 0;
3194         }
3195
3196         error = kproc_create(g_raid3_worker, sc, &sc->sc_worker, 0, 0,
3197             "g_raid3 %s", md->md_name);
3198         if (error != 0) {
3199                 G_RAID3_DEBUG(1, "Cannot create kernel thread for %s.",
3200                     sc->sc_name);
3201                 if (!g_raid3_use_malloc) {
3202                         uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_64K].sz_zone);
3203                         uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_16K].sz_zone);
3204                         uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_4K].sz_zone);
3205                 }
3206                 g_destroy_geom(sc->sc_sync.ds_geom);
3207                 mtx_destroy(&sc->sc_events_mtx);
3208                 mtx_destroy(&sc->sc_queue_mtx);
3209                 sx_destroy(&sc->sc_lock);
3210                 g_destroy_geom(sc->sc_geom);
3211                 free(sc->sc_disks, M_RAID3);
3212                 free(sc, M_RAID3);
3213                 return (NULL);
3214         }
3215
3216         G_RAID3_DEBUG(1, "Device %s created (%u components, id=%u).",
3217             sc->sc_name, sc->sc_ndisks, sc->sc_id);
3218
3219         sc->sc_rootmount = root_mount_hold("GRAID3");
3220         G_RAID3_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount);
3221
3222         /*
3223          * Run timeout.
3224          */
3225         timeout = atomic_load_acq_int(&g_raid3_timeout);
3226         callout_reset(&sc->sc_callout, timeout * hz, g_raid3_go, sc);
3227         return (sc->sc_geom);
3228 }
3229
3230 int
3231 g_raid3_destroy(struct g_raid3_softc *sc, int how)
3232 {
3233         struct g_provider *pp;
3234
3235         g_topology_assert_not();
3236         if (sc == NULL)
3237                 return (ENXIO);
3238         sx_assert(&sc->sc_lock, SX_XLOCKED);
3239
3240         pp = sc->sc_provider;
3241         if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
3242                 switch (how) {
3243                 case G_RAID3_DESTROY_SOFT:
3244                         G_RAID3_DEBUG(1,
3245                             "Device %s is still open (r%dw%de%d).", pp->name,
3246                             pp->acr, pp->acw, pp->ace);
3247                         return (EBUSY);
3248                 case G_RAID3_DESTROY_DELAYED:
3249                         G_RAID3_DEBUG(1,
3250                             "Device %s will be destroyed on last close.",
3251                             pp->name);
3252                         if (sc->sc_syncdisk != NULL)
3253                                 g_raid3_sync_stop(sc, 1);
3254                         sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROYING;
3255                         return (EBUSY);
3256                 case G_RAID3_DESTROY_HARD:
3257                         G_RAID3_DEBUG(1, "Device %s is still open, so it "
3258                             "can't be definitely removed.", pp->name);
3259                         break;
3260                 }
3261         }
3262
3263         g_topology_lock();
3264         if (sc->sc_geom->softc == NULL) {
3265                 g_topology_unlock();
3266                 return (0);
3267         }
3268         sc->sc_geom->softc = NULL;
3269         sc->sc_sync.ds_geom->softc = NULL;
3270         g_topology_unlock();
3271
3272         sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY;
3273         sc->sc_flags |= G_RAID3_DEVICE_FLAG_WAIT;
3274         G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc);
3275         sx_xunlock(&sc->sc_lock);
3276         mtx_lock(&sc->sc_queue_mtx);
3277         wakeup(sc);
3278         wakeup(&sc->sc_queue);
3279         mtx_unlock(&sc->sc_queue_mtx);
3280         G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker);
3281         while (sc->sc_worker != NULL)
3282                 tsleep(&sc->sc_worker, PRIBIO, "r3:destroy", hz / 5);
3283         G_RAID3_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker);
3284         sx_xlock(&sc->sc_lock);
3285         g_raid3_destroy_device(sc);
3286         free(sc->sc_disks, M_RAID3);
3287         free(sc, M_RAID3);
3288         return (0);
3289 }
3290
3291 static void
3292 g_raid3_taste_orphan(struct g_consumer *cp)
3293 {
3294
3295         KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
3296             cp->provider->name));
3297 }
3298
3299 static struct g_geom *
3300 g_raid3_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
3301 {
3302         struct g_raid3_metadata md;
3303         struct g_raid3_softc *sc;
3304         struct g_consumer *cp;
3305         struct g_geom *gp;
3306         int error;
3307
3308         g_topology_assert();
3309         g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
3310         G_RAID3_DEBUG(2, "Tasting %s.", pp->name);
3311
3312         gp = g_new_geomf(mp, "raid3:taste");
3313         /* This orphan function should be never called. */
3314         gp->orphan = g_raid3_taste_orphan;
3315         cp = g_new_consumer(gp);
3316         g_attach(cp, pp);
3317         error = g_raid3_read_metadata(cp, &md);
3318         g_detach(cp);
3319         g_destroy_consumer(cp);
3320         g_destroy_geom(gp);
3321         if (error != 0)
3322                 return (NULL);
3323         gp = NULL;
3324
3325         if (md.md_provider[0] != '\0' &&
3326             !g_compare_names(md.md_provider, pp->name))
3327                 return (NULL);
3328         if (md.md_provsize != 0 && md.md_provsize != pp->mediasize)
3329                 return (NULL);
3330         if (g_raid3_debug >= 2)
3331                 raid3_metadata_dump(&md);
3332
3333         /*
3334          * Let's check if device already exists.
3335          */
3336         sc = NULL;
3337         LIST_FOREACH(gp, &mp->geom, geom) {
3338                 sc = gp->softc;
3339                 if (sc == NULL)
3340                         continue;
3341                 if (sc->sc_sync.ds_geom == gp)
3342                         continue;
3343                 if (strcmp(md.md_name, sc->sc_name) != 0)
3344                         continue;
3345                 if (md.md_id != sc->sc_id) {
3346                         G_RAID3_DEBUG(0, "Device %s already configured.",
3347                             sc->sc_name);
3348                         return (NULL);
3349                 }
3350                 break;
3351         }
3352         if (gp == NULL) {
3353                 gp = g_raid3_create(mp, &md);
3354                 if (gp == NULL) {
3355                         G_RAID3_DEBUG(0, "Cannot create device %s.",
3356                             md.md_name);
3357                         return (NULL);
3358                 }
3359                 sc = gp->softc;
3360         }
3361         G_RAID3_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name);
3362         g_topology_unlock();
3363         sx_xlock(&sc->sc_lock);
3364         error = g_raid3_add_disk(sc, pp, &md);
3365         if (error != 0) {
3366                 G_RAID3_DEBUG(0, "Cannot add disk %s to %s (error=%d).",
3367                     pp->name, gp->name, error);
3368                 if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NODISK) ==
3369                     sc->sc_ndisks) {
3370                         g_cancel_event(sc);
3371                         g_raid3_destroy(sc, G_RAID3_DESTROY_HARD);
3372                         g_topology_lock();
3373                         return (NULL);
3374                 }
3375                 gp = NULL;
3376         }
3377         sx_xunlock(&sc->sc_lock);
3378         g_topology_lock();
3379         return (gp);
3380 }
3381
3382 static int
3383 g_raid3_destroy_geom(struct gctl_req *req __unused, struct g_class *mp __unused,
3384     struct g_geom *gp)
3385 {
3386         struct g_raid3_softc *sc;
3387         int error;
3388
3389         g_topology_unlock();
3390         sc = gp->softc;
3391         sx_xlock(&sc->sc_lock);
3392         g_cancel_event(sc);
3393         error = g_raid3_destroy(gp->softc, G_RAID3_DESTROY_SOFT);
3394         if (error != 0)
3395                 sx_xunlock(&sc->sc_lock);
3396         g_topology_lock();
3397         return (error);
3398 }
3399
3400 static void
3401 g_raid3_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
3402     struct g_consumer *cp, struct g_provider *pp)
3403 {
3404         struct g_raid3_softc *sc;
3405
3406         g_topology_assert();
3407
3408         sc = gp->softc;
3409         if (sc == NULL)
3410                 return;
3411         /* Skip synchronization geom. */
3412         if (gp == sc->sc_sync.ds_geom)
3413                 return;
3414         if (pp != NULL) {
3415                 /* Nothing here. */
3416         } else if (cp != NULL) {
3417                 struct g_raid3_disk *disk;
3418
3419                 disk = cp->private;
3420                 if (disk == NULL)
3421                         return;
3422                 g_topology_unlock();
3423                 sx_xlock(&sc->sc_lock);
3424                 sbuf_printf(sb, "%s<Type>", indent);
3425                 if (disk->d_no == sc->sc_ndisks - 1)
3426                         sbuf_printf(sb, "PARITY");
3427                 else
3428                         sbuf_printf(sb, "DATA");
3429                 sbuf_printf(sb, "</Type>\n");
3430                 sbuf_printf(sb, "%s<Number>%u</Number>\n", indent,
3431                     (u_int)disk->d_no);
3432                 if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) {
3433                         sbuf_printf(sb, "%s<Synchronized>", indent);
3434                         if (disk->d_sync.ds_offset == 0)
3435                                 sbuf_printf(sb, "0%%");
3436                         else {
3437                                 sbuf_printf(sb, "%u%%",
3438                                     (u_int)((disk->d_sync.ds_offset * 100) /
3439                                     (sc->sc_mediasize / (sc->sc_ndisks - 1))));
3440                         }
3441                         sbuf_printf(sb, "</Synchronized>\n");
3442                         if (disk->d_sync.ds_offset > 0) {
3443                                 sbuf_printf(sb, "%s<BytesSynced>%jd"
3444                                     "</BytesSynced>\n", indent,
3445                                     (intmax_t)disk->d_sync.ds_offset);
3446                         }
3447                 }
3448                 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent,
3449                     disk->d_sync.ds_syncid);
3450                 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, disk->d_genid);
3451                 sbuf_printf(sb, "%s<Flags>", indent);
3452                 if (disk->d_flags == 0)
3453                         sbuf_printf(sb, "NONE");
3454                 else {
3455                         int first = 1;
3456
3457 #define ADD_FLAG(flag, name)    do {                                    \
3458         if ((disk->d_flags & (flag)) != 0) {                            \
3459                 if (!first)                                             \
3460                         sbuf_printf(sb, ", ");                          \
3461                 else                                                    \
3462                         first = 0;                                      \
3463                 sbuf_printf(sb, name);                                  \
3464         }                                                               \
3465 } while (0)
3466                         ADD_FLAG(G_RAID3_DISK_FLAG_DIRTY, "DIRTY");
3467                         ADD_FLAG(G_RAID3_DISK_FLAG_HARDCODED, "HARDCODED");
3468                         ADD_FLAG(G_RAID3_DISK_FLAG_SYNCHRONIZING,
3469                             "SYNCHRONIZING");
3470                         ADD_FLAG(G_RAID3_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC");
3471                         ADD_FLAG(G_RAID3_DISK_FLAG_BROKEN, "BROKEN");
3472 #undef  ADD_FLAG
3473                 }
3474                 sbuf_printf(sb, "</Flags>\n");
3475                 sbuf_printf(sb, "%s<State>%s</State>\n", indent,
3476                     g_raid3_disk_state2str(disk->d_state));
3477                 sx_xunlock(&sc->sc_lock);
3478                 g_topology_lock();
3479         } else {
3480                 g_topology_unlock();
3481                 sx_xlock(&sc->sc_lock);
3482                 if (!g_raid3_use_malloc) {
3483                         sbuf_printf(sb,
3484                             "%s<Zone4kRequested>%u</Zone4kRequested>\n", indent,
3485                             sc->sc_zones[G_RAID3_ZONE_4K].sz_requested);
3486                         sbuf_printf(sb,
3487                             "%s<Zone4kFailed>%u</Zone4kFailed>\n", indent,
3488                             sc->sc_zones[G_RAID3_ZONE_4K].sz_failed);
3489                         sbuf_printf(sb,
3490                             "%s<Zone16kRequested>%u</Zone16kRequested>\n", indent,
3491                             sc->sc_zones[G_RAID3_ZONE_16K].sz_requested);
3492                         sbuf_printf(sb,
3493                             "%s<Zone16kFailed>%u</Zone16kFailed>\n", indent,
3494                             sc->sc_zones[G_RAID3_ZONE_16K].sz_failed);
3495                         sbuf_printf(sb,
3496                             "%s<Zone64kRequested>%u</Zone64kRequested>\n", indent,
3497                             sc->sc_zones[G_RAID3_ZONE_64K].sz_requested);
3498                         sbuf_printf(sb,
3499                             "%s<Zone64kFailed>%u</Zone64kFailed>\n", indent,
3500                             sc->sc_zones[G_RAID3_ZONE_64K].sz_failed);
3501                 }
3502                 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id);
3503                 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid);
3504                 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid);
3505                 sbuf_printf(sb, "%s<Flags>", indent);
3506                 if (sc->sc_flags == 0)
3507                         sbuf_printf(sb, "NONE");
3508                 else {
3509                         int first = 1;
3510
3511 #define ADD_FLAG(flag, name)    do {                                    \
3512         if ((sc->sc_flags & (flag)) != 0) {                             \
3513                 if (!first)                                             \
3514                         sbuf_printf(sb, ", ");                          \
3515                 else                                                    \
3516                         first = 0;                                      \
3517                 sbuf_printf(sb, name);                                  \
3518         }                                                               \
3519 } while (0)
3520                         ADD_FLAG(G_RAID3_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC");
3521                         ADD_FLAG(G_RAID3_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC");
3522                         ADD_FLAG(G_RAID3_DEVICE_FLAG_ROUND_ROBIN,
3523                             "ROUND-ROBIN");
3524                         ADD_FLAG(G_RAID3_DEVICE_FLAG_VERIFY, "VERIFY");
3525 #undef  ADD_FLAG
3526                 }
3527                 sbuf_printf(sb, "</Flags>\n");
3528                 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent,
3529                     sc->sc_ndisks);
3530                 sbuf_printf(sb, "%s<State>%s</State>\n", indent,
3531                     g_raid3_device_state2str(sc->sc_state));
3532                 sx_xunlock(&sc->sc_lock);
3533                 g_topology_lock();
3534         }
3535 }
3536
3537 static void
3538 g_raid3_shutdown_post_sync(void *arg, int howto)
3539 {
3540         struct g_class *mp;
3541         struct g_geom *gp, *gp2;
3542         struct g_raid3_softc *sc;
3543         int error;
3544
3545         mp = arg;
3546         DROP_GIANT();
3547         g_topology_lock();
3548         g_raid3_shutdown = 1;
3549         LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
3550                 if ((sc = gp->softc) == NULL)
3551                         continue;
3552                 /* Skip synchronization geom. */
3553                 if (gp == sc->sc_sync.ds_geom)
3554                         continue;
3555                 g_topology_unlock();
3556                 sx_xlock(&sc->sc_lock);
3557                 g_raid3_idle(sc, -1);
3558                 g_cancel_event(sc);
3559                 error = g_raid3_destroy(sc, G_RAID3_DESTROY_DELAYED);
3560                 if (error != 0)
3561                         sx_xunlock(&sc->sc_lock);
3562                 g_topology_lock();
3563         }
3564         g_topology_unlock();
3565         PICKUP_GIANT();
3566 }
3567
3568 static void
3569 g_raid3_init(struct g_class *mp)
3570 {
3571
3572         g_raid3_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync,
3573             g_raid3_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST);
3574         if (g_raid3_post_sync == NULL)
3575                 G_RAID3_DEBUG(0, "Warning! Cannot register shutdown event.");
3576 }
3577
3578 static void
3579 g_raid3_fini(struct g_class *mp)
3580 {
3581
3582         if (g_raid3_post_sync != NULL)
3583                 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid3_post_sync);
3584 }
3585
3586 DECLARE_GEOM_CLASS(g_raid3_class, g_raid3);