1 /*- 2 * Copyright (c) 2004 Pawel Jakub Dawidek <pjd@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 #include <sys/cdefs.h> 28 __FBSDID("$FreeBSD$"); 29 30 #include <sys/param.h> 31 #include <sys/systm.h> 32 #include <sys/kernel.h> 33 #include <sys/module.h> 34 #include <sys/limits.h> 35 #include <sys/lock.h> 36 #include <sys/mutex.h> 37 #include <sys/bio.h> 38 #include <sys/sysctl.h> 39 #include <sys/malloc.h> 40 #include <sys/eventhandler.h> 41 #include <vm/uma.h> 42 #include <geom/geom.h> 43 #include <sys/proc.h> 44 #include <sys/kthread.h> 45 #include <sys/sched.h> 46 #include <geom/mirror/g_mirror.h> 47 48 49 static MALLOC_DEFINE(M_MIRROR, "mirror data", "GEOM_MIRROR Data"); 50 51 SYSCTL_DECL(_kern_geom); 52 SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, "GEOM_MIRROR stuff"); 53 u_int g_mirror_debug = 0; 54 TUNABLE_INT("kern.geom.mirror.debug", &g_mirror_debug); 55 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RW, &g_mirror_debug, 0, 56 "Debug level"); 57 static u_int g_mirror_timeout = 4; 58 TUNABLE_INT("kern.geom.mirror.timeout", &g_mirror_timeout); 59 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RW, &g_mirror_timeout, 60 0, "Time to wait on all mirror components"); 61 static u_int g_mirror_idletime = 5; 62 TUNABLE_INT("kern.geom.mirror.idletime", &g_mirror_idletime); 63 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RW, 64 &g_mirror_idletime, 0, "Mark components as clean when idling"); 65 static u_int g_mirror_reqs_per_sync = 5; 66 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, reqs_per_sync, CTLFLAG_RW, 67 &g_mirror_reqs_per_sync, 0, 68 "Number of regular I/O requests per synchronization request"); 69 static u_int g_mirror_syncs_per_sec = 1000; 70 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, syncs_per_sec, CTLFLAG_RW, 71 &g_mirror_syncs_per_sec, 0, 72 "Number of synchronizations requests per second"); 73 74 #define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 75 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 76 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 77 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 78 } while (0) 79 80 static eventhandler_tag g_mirror_ehtag = NULL; 81 82 static int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp, 83 struct g_geom *gp); 84 static g_taste_t g_mirror_taste; 85 static void g_mirror_init(struct g_class *mp); 86 static void g_mirror_fini(struct g_class *mp); 87 88 struct g_class g_mirror_class = { 89 .name = G_MIRROR_CLASS_NAME, 90 .version = G_VERSION, 91 .ctlreq = g_mirror_config, 92 .taste = g_mirror_taste, 93 .destroy_geom = g_mirror_destroy_geom, 94 .init = g_mirror_init, 95 .fini = g_mirror_fini 96 }; 97 98 99 static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 100 static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 101 static void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force); 102 static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 103 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 104 static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 105 106 107 static const char * 108 g_mirror_disk_state2str(int state) 109 { 110 111 switch (state) { 112 case G_MIRROR_DISK_STATE_NONE: 113 return ("NONE"); 114 case G_MIRROR_DISK_STATE_NEW: 115 return ("NEW"); 116 case G_MIRROR_DISK_STATE_ACTIVE: 117 return ("ACTIVE"); 118 case G_MIRROR_DISK_STATE_STALE: 119 return ("STALE"); 120 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 121 return ("SYNCHRONIZING"); 122 case G_MIRROR_DISK_STATE_DISCONNECTED: 123 return ("DISCONNECTED"); 124 case G_MIRROR_DISK_STATE_DESTROY: 125 return ("DESTROY"); 126 default: 127 return ("INVALID"); 128 } 129 } 130 131 static const char * 132 g_mirror_device_state2str(int state) 133 { 134 135 switch (state) { 136 case G_MIRROR_DEVICE_STATE_STARTING: 137 return ("STARTING"); 138 case G_MIRROR_DEVICE_STATE_RUNNING: 139 return ("RUNNING"); 140 default: 141 return ("INVALID"); 142 } 143 } 144 145 static const char * 146 g_mirror_get_diskname(struct g_mirror_disk *disk) 147 { 148 149 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 150 return ("[unknown]"); 151 return (disk->d_name); 152 } 153 154 /* 155 * --- Events handling functions --- 156 * Events in geom_mirror are used to maintain disks and device status 157 * from one thread to simplify locking. 158 */ 159 static void 160 g_mirror_event_free(struct g_mirror_event *ep) 161 { 162 163 free(ep, M_MIRROR); 164 } 165 166 int 167 g_mirror_event_send(void *arg, int state, int flags) 168 { 169 struct g_mirror_softc *sc; 170 struct g_mirror_disk *disk; 171 struct g_mirror_event *ep; 172 int error; 173 174 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 175 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 176 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 177 disk = NULL; 178 sc = arg; 179 } else { 180 disk = arg; 181 sc = disk->d_softc; 182 } 183 ep->e_disk = disk; 184 ep->e_state = state; 185 ep->e_flags = flags; 186 ep->e_error = 0; 187 mtx_lock(&sc->sc_events_mtx); 188 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 189 mtx_unlock(&sc->sc_events_mtx); 190 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 191 mtx_lock(&sc->sc_queue_mtx); 192 wakeup(sc); 193 mtx_unlock(&sc->sc_queue_mtx); 194 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 195 return (0); 196 g_topology_assert(); 197 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 198 g_topology_unlock(); 199 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 200 mtx_lock(&sc->sc_events_mtx); 201 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 202 hz * 5); 203 } 204 /* Don't even try to use 'sc' here, because it could be already dead. */ 205 g_topology_lock(); 206 error = ep->e_error; 207 g_mirror_event_free(ep); 208 return (error); 209 } 210 211 static struct g_mirror_event * 212 g_mirror_event_get(struct g_mirror_softc *sc) 213 { 214 struct g_mirror_event *ep; 215 216 mtx_lock(&sc->sc_events_mtx); 217 ep = TAILQ_FIRST(&sc->sc_events); 218 mtx_unlock(&sc->sc_events_mtx); 219 return (ep); 220 } 221 222 static void 223 g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 224 { 225 226 mtx_lock(&sc->sc_events_mtx); 227 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 228 mtx_unlock(&sc->sc_events_mtx); 229 } 230 231 static void 232 g_mirror_event_cancel(struct g_mirror_disk *disk) 233 { 234 struct g_mirror_softc *sc; 235 struct g_mirror_event *ep, *tmpep; 236 237 g_topology_assert(); 238 239 sc = disk->d_softc; 240 mtx_lock(&sc->sc_events_mtx); 241 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 242 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 243 continue; 244 if (ep->e_disk != disk) 245 continue; 246 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 247 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 248 g_mirror_event_free(ep); 249 else { 250 ep->e_error = ECANCELED; 251 wakeup(ep); 252 } 253 } 254 mtx_unlock(&sc->sc_events_mtx); 255 } 256 257 /* 258 * Return the number of disks in given state. 259 * If state is equal to -1, count all connected disks. 260 */ 261 u_int 262 g_mirror_ndisks(struct g_mirror_softc *sc, int state) 263 { 264 struct g_mirror_disk *disk; 265 u_int n = 0; 266 267 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 268 if (state == -1 || disk->d_state == state) 269 n++; 270 } 271 return (n); 272 } 273 274 /* 275 * Find a disk in mirror by its disk ID. 276 */ 277 static struct g_mirror_disk * 278 g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 279 { 280 struct g_mirror_disk *disk; 281 282 g_topology_assert(); 283 284 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 285 if (disk->d_id == id) 286 return (disk); 287 } 288 return (NULL); 289 } 290 291 static u_int 292 g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 293 { 294 struct bio *bp; 295 u_int nreqs = 0; 296 297 mtx_lock(&sc->sc_queue_mtx); 298 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 299 if (bp->bio_from == cp) 300 nreqs++; 301 } 302 mtx_unlock(&sc->sc_queue_mtx); 303 return (nreqs); 304 } 305 306 static int 307 g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 308 { 309 310 if (cp->index > 0) { 311 G_MIRROR_DEBUG(2, 312 "I/O requests for %s exist, can't destroy it now.", 313 cp->provider->name); 314 return (1); 315 } 316 if (g_mirror_nrequests(sc, cp) > 0) { 317 G_MIRROR_DEBUG(2, 318 "I/O requests for %s in queue, can't destroy it now.", 319 cp->provider->name); 320 return (1); 321 } 322 return (0); 323 } 324 325 static void 326 g_mirror_destroy_consumer(void *arg, int flags __unused) 327 { 328 struct g_consumer *cp; 329 330 cp = arg; 331 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 332 g_detach(cp); 333 g_destroy_consumer(cp); 334 } 335 336 static void 337 g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 338 { 339 struct g_provider *pp; 340 int retaste_wait; 341 342 g_topology_assert(); 343 344 cp->private = NULL; 345 if (g_mirror_is_busy(sc, cp)) 346 return; 347 pp = cp->provider; 348 retaste_wait = 0; 349 if (cp->acw == 1) { 350 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 351 retaste_wait = 1; 352 } 353 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 354 -cp->acw, -cp->ace, 0); 355 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 356 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 357 if (retaste_wait) { 358 /* 359 * After retaste event was send (inside g_access()), we can send 360 * event to detach and destroy consumer. 361 * A class, which has consumer to the given provider connected 362 * will not receive retaste event for the provider. 363 * This is the way how I ignore retaste events when I close 364 * consumers opened for write: I detach and destroy consumer 365 * after retaste event is sent. 366 */ 367 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 368 return; 369 } 370 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 371 g_detach(cp); 372 g_destroy_consumer(cp); 373 } 374 375 static int 376 g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 377 { 378 int error; 379 380 g_topology_assert(); 381 KASSERT(disk->d_consumer == NULL, 382 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 383 384 disk->d_consumer = g_new_consumer(disk->d_softc->sc_geom); 385 disk->d_consumer->private = disk; 386 disk->d_consumer->index = 0; 387 error = g_attach(disk->d_consumer, pp); 388 if (error != 0) 389 return (error); 390 error = g_access(disk->d_consumer, 1, 1, 1); 391 if (error != 0) { 392 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 393 pp->name, error); 394 return (error); 395 } 396 397 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 398 return (0); 399 } 400 401 static void 402 g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 403 { 404 405 g_topology_assert(); 406 407 if (cp == NULL) 408 return; 409 if (cp->provider != NULL) 410 g_mirror_kill_consumer(sc, cp); 411 else 412 g_destroy_consumer(cp); 413 } 414 415 /* 416 * Initialize disk. This means allocate memory, create consumer, attach it 417 * to the provider and open access (r1w1e1) to it. 418 */ 419 static struct g_mirror_disk * 420 g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 421 struct g_mirror_metadata *md, int *errorp) 422 { 423 struct g_mirror_disk *disk; 424 int error; 425 426 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 427 if (disk == NULL) { 428 error = ENOMEM; 429 goto fail; 430 } 431 disk->d_softc = sc; 432 error = g_mirror_connect_disk(disk, pp); 433 if (error != 0) 434 goto fail; 435 disk->d_id = md->md_did; 436 disk->d_state = G_MIRROR_DISK_STATE_NONE; 437 disk->d_priority = md->md_priority; 438 disk->d_delay.sec = 0; 439 disk->d_delay.frac = 0; 440 binuptime(&disk->d_last_used); 441 disk->d_flags = md->md_dflags; 442 if (md->md_provider[0] != '\0') 443 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 444 disk->d_sync.ds_consumer = NULL; 445 disk->d_sync.ds_offset = md->md_sync_offset; 446 disk->d_sync.ds_offset_done = md->md_sync_offset; 447 disk->d_sync.ds_resync = -1; 448 disk->d_genid = md->md_genid; 449 disk->d_sync.ds_syncid = md->md_syncid; 450 if (errorp != NULL) 451 *errorp = 0; 452 return (disk); 453 fail: 454 if (errorp != NULL) 455 *errorp = error; 456 if (disk != NULL) { 457 g_mirror_disconnect_consumer(sc, disk->d_consumer); 458 free(disk, M_MIRROR); 459 } 460 return (NULL); 461 } 462 463 static void 464 g_mirror_destroy_disk(struct g_mirror_disk *disk) 465 { 466 struct g_mirror_softc *sc; 467 468 g_topology_assert(); 469 470 LIST_REMOVE(disk, d_next); 471 g_mirror_event_cancel(disk); 472 sc = disk->d_softc; 473 if (sc->sc_hint == disk) 474 sc->sc_hint = NULL; 475 switch (disk->d_state) { 476 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 477 g_mirror_sync_stop(disk, 1); 478 /* FALLTHROUGH */ 479 case G_MIRROR_DISK_STATE_NEW: 480 case G_MIRROR_DISK_STATE_STALE: 481 case G_MIRROR_DISK_STATE_ACTIVE: 482 g_mirror_disconnect_consumer(sc, disk->d_consumer); 483 free(disk, M_MIRROR); 484 break; 485 default: 486 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 487 g_mirror_get_diskname(disk), 488 g_mirror_disk_state2str(disk->d_state))); 489 } 490 } 491 492 static void 493 g_mirror_destroy_device(struct g_mirror_softc *sc) 494 { 495 struct g_mirror_disk *disk; 496 struct g_mirror_event *ep; 497 struct g_geom *gp; 498 struct g_consumer *cp, *tmpcp; 499 500 g_topology_assert(); 501 502 gp = sc->sc_geom; 503 if (sc->sc_provider != NULL) 504 g_mirror_destroy_provider(sc); 505 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 506 disk = LIST_FIRST(&sc->sc_disks)) { 507 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 508 g_mirror_update_metadata(disk); 509 g_mirror_destroy_disk(disk); 510 } 511 while ((ep = g_mirror_event_get(sc)) != NULL) { 512 g_mirror_event_remove(sc, ep); 513 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 514 g_mirror_event_free(ep); 515 else { 516 ep->e_error = ECANCELED; 517 ep->e_flags |= G_MIRROR_EVENT_DONE; 518 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 519 mtx_lock(&sc->sc_events_mtx); 520 wakeup(ep); 521 mtx_unlock(&sc->sc_events_mtx); 522 } 523 } 524 callout_drain(&sc->sc_callout); 525 gp->softc = NULL; 526 527 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 528 g_mirror_disconnect_consumer(sc, cp); 529 } 530 sc->sc_sync.ds_geom->softc = NULL; 531 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 532 mtx_destroy(&sc->sc_queue_mtx); 533 mtx_destroy(&sc->sc_events_mtx); 534 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 535 g_wither_geom(gp, ENXIO); 536 } 537 538 static void 539 g_mirror_orphan(struct g_consumer *cp) 540 { 541 struct g_mirror_disk *disk; 542 543 g_topology_assert(); 544 545 disk = cp->private; 546 if (disk == NULL) 547 return; 548 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 549 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 550 G_MIRROR_EVENT_DONTWAIT); 551 } 552 553 /* 554 * Function should return the next active disk on the list. 555 * It is possible that it will be the same disk as given. 556 * If there are no active disks on list, NULL is returned. 557 */ 558 static __inline struct g_mirror_disk * 559 g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 560 { 561 struct g_mirror_disk *dp; 562 563 for (dp = LIST_NEXT(disk, d_next); dp != disk; 564 dp = LIST_NEXT(dp, d_next)) { 565 if (dp == NULL) 566 dp = LIST_FIRST(&sc->sc_disks); 567 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 568 break; 569 } 570 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 571 return (NULL); 572 return (dp); 573 } 574 575 static struct g_mirror_disk * 576 g_mirror_get_disk(struct g_mirror_softc *sc) 577 { 578 struct g_mirror_disk *disk; 579 580 if (sc->sc_hint == NULL) { 581 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 582 if (sc->sc_hint == NULL) 583 return (NULL); 584 } 585 disk = sc->sc_hint; 586 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 587 disk = g_mirror_find_next(sc, disk); 588 if (disk == NULL) 589 return (NULL); 590 } 591 sc->sc_hint = g_mirror_find_next(sc, disk); 592 return (disk); 593 } 594 595 static int 596 g_mirror_write_metadata(struct g_mirror_disk *disk, 597 struct g_mirror_metadata *md) 598 { 599 struct g_mirror_softc *sc; 600 struct g_consumer *cp; 601 off_t offset, length; 602 u_char *sector; 603 int error = 0; 604 605 g_topology_assert(); 606 607 sc = disk->d_softc; 608 cp = disk->d_consumer; 609 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 610 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 611 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 612 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 613 cp->acw, cp->ace)); 614 length = cp->provider->sectorsize; 615 offset = cp->provider->mediasize - length; 616 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 617 if (md != NULL) 618 mirror_metadata_encode(md, sector); 619 g_topology_unlock(); 620 error = g_write_data(cp, offset, sector, length); 621 g_topology_lock(); 622 free(sector, M_MIRROR); 623 if (error != 0) { 624 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_GENID; 625 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 626 G_MIRROR_EVENT_DONTWAIT); 627 } 628 return (error); 629 } 630 631 static int 632 g_mirror_clear_metadata(struct g_mirror_disk *disk) 633 { 634 int error; 635 636 g_topology_assert(); 637 error = g_mirror_write_metadata(disk, NULL); 638 if (error == 0) { 639 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 640 g_mirror_get_diskname(disk)); 641 } else { 642 G_MIRROR_DEBUG(0, 643 "Cannot clear metadata on disk %s (error=%d).", 644 g_mirror_get_diskname(disk), error); 645 } 646 return (error); 647 } 648 649 void 650 g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 651 struct g_mirror_metadata *md) 652 { 653 654 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 655 md->md_version = G_MIRROR_VERSION; 656 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 657 md->md_mid = sc->sc_id; 658 md->md_all = sc->sc_ndisks; 659 md->md_slice = sc->sc_slice; 660 md->md_balance = sc->sc_balance; 661 md->md_genid = sc->sc_genid; 662 md->md_mediasize = sc->sc_mediasize; 663 md->md_sectorsize = sc->sc_sectorsize; 664 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 665 bzero(md->md_provider, sizeof(md->md_provider)); 666 if (disk == NULL) { 667 md->md_did = arc4random(); 668 md->md_priority = 0; 669 md->md_syncid = 0; 670 md->md_dflags = 0; 671 md->md_sync_offset = 0; 672 } else { 673 md->md_did = disk->d_id; 674 md->md_priority = disk->d_priority; 675 md->md_syncid = disk->d_sync.ds_syncid; 676 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 677 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 678 md->md_sync_offset = disk->d_sync.ds_offset_done; 679 else 680 md->md_sync_offset = 0; 681 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 682 strlcpy(md->md_provider, 683 disk->d_consumer->provider->name, 684 sizeof(md->md_provider)); 685 } 686 } 687 } 688 689 void 690 g_mirror_update_metadata(struct g_mirror_disk *disk) 691 { 692 struct g_mirror_metadata md; 693 int error; 694 695 g_topology_assert(); 696 g_mirror_fill_metadata(disk->d_softc, disk, &md); 697 error = g_mirror_write_metadata(disk, &md); 698 if (error == 0) { 699 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 700 g_mirror_get_diskname(disk)); 701 } else { 702 G_MIRROR_DEBUG(0, 703 "Cannot update metadata on disk %s (error=%d).", 704 g_mirror_get_diskname(disk), error); 705 } 706 } 707 708 static void 709 g_mirror_bump_syncid(struct g_mirror_softc *sc) 710 { 711 struct g_mirror_disk *disk; 712 713 g_topology_assert(); 714 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 715 ("%s called with no active disks (device=%s).", __func__, 716 sc->sc_name)); 717 718 sc->sc_syncid++; 719 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 720 sc->sc_syncid); 721 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 722 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 723 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 724 disk->d_sync.ds_syncid = sc->sc_syncid; 725 g_mirror_update_metadata(disk); 726 } 727 } 728 } 729 730 static void 731 g_mirror_bump_genid(struct g_mirror_softc *sc) 732 { 733 struct g_mirror_disk *disk; 734 735 g_topology_assert(); 736 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 737 ("%s called with no active disks (device=%s).", __func__, 738 sc->sc_name)); 739 740 sc->sc_genid++; 741 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 742 sc->sc_genid); 743 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 744 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 745 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 746 disk->d_genid = sc->sc_genid; 747 g_mirror_update_metadata(disk); 748 } 749 } 750 } 751 752 static void 753 g_mirror_idle(struct g_mirror_softc *sc) 754 { 755 struct g_mirror_disk *disk; 756 757 if (sc->sc_provider == NULL || sc->sc_provider->acw == 0) 758 return; 759 sc->sc_idle = 1; 760 g_topology_lock(); 761 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 762 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 763 continue; 764 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 765 g_mirror_get_diskname(disk), sc->sc_name); 766 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 767 g_mirror_update_metadata(disk); 768 } 769 g_topology_unlock(); 770 } 771 772 static void 773 g_mirror_unidle(struct g_mirror_softc *sc) 774 { 775 struct g_mirror_disk *disk; 776 777 sc->sc_idle = 0; 778 g_topology_lock(); 779 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 780 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 781 continue; 782 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 783 g_mirror_get_diskname(disk), sc->sc_name); 784 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 785 g_mirror_update_metadata(disk); 786 } 787 g_topology_unlock(); 788 } 789 790 /* 791 * Return 1 if we should check if mirror is idling. 792 */ 793 static int 794 g_mirror_check_idle(struct g_mirror_softc *sc) 795 { 796 struct g_mirror_disk *disk; 797 798 if (sc->sc_idle) 799 return (0); 800 if (sc->sc_provider != NULL && sc->sc_provider->acw == 0) 801 return (0); 802 /* 803 * Check if there are no in-flight requests. 804 */ 805 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 806 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 807 continue; 808 if (disk->d_consumer->index > 0) 809 return (0); 810 } 811 return (1); 812 } 813 814 static __inline int 815 bintime_cmp(struct bintime *bt1, struct bintime *bt2) 816 { 817 818 if (bt1->sec < bt2->sec) 819 return (-1); 820 else if (bt1->sec > bt2->sec) 821 return (1); 822 if (bt1->frac < bt2->frac) 823 return (-1); 824 else if (bt1->frac > bt2->frac) 825 return (1); 826 return (0); 827 } 828 829 static void 830 g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 831 { 832 833 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 834 return; 835 binuptime(&disk->d_delay); 836 bintime_sub(&disk->d_delay, &bp->bio_t0); 837 } 838 839 static void 840 g_mirror_done(struct bio *bp) 841 { 842 struct g_mirror_softc *sc; 843 844 sc = bp->bio_from->geom->softc; 845 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 846 mtx_lock(&sc->sc_queue_mtx); 847 bioq_disksort(&sc->sc_queue, bp); 848 wakeup(sc); 849 mtx_unlock(&sc->sc_queue_mtx); 850 } 851 852 static void 853 g_mirror_regular_request(struct bio *bp) 854 { 855 struct g_mirror_softc *sc; 856 struct g_mirror_disk *disk; 857 struct bio *pbp; 858 859 g_topology_assert_not(); 860 861 bp->bio_from->index--; 862 pbp = bp->bio_parent; 863 sc = pbp->bio_to->geom->softc; 864 disk = bp->bio_from->private; 865 if (disk == NULL) { 866 g_topology_lock(); 867 g_mirror_kill_consumer(sc, bp->bio_from); 868 g_topology_unlock(); 869 } else { 870 g_mirror_update_delay(disk, bp); 871 } 872 873 pbp->bio_inbed++; 874 KASSERT(pbp->bio_inbed <= pbp->bio_children, 875 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 876 pbp->bio_children)); 877 if (bp->bio_error == 0 && pbp->bio_error == 0) { 878 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 879 g_destroy_bio(bp); 880 if (pbp->bio_children == pbp->bio_inbed) { 881 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 882 pbp->bio_completed = pbp->bio_length; 883 g_io_deliver(pbp, pbp->bio_error); 884 } 885 return; 886 } else if (bp->bio_error != 0) { 887 if (pbp->bio_error == 0) 888 pbp->bio_error = bp->bio_error; 889 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 890 bp->bio_error); 891 if (disk != NULL) { 892 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 893 g_mirror_event_send(disk, 894 G_MIRROR_DISK_STATE_DISCONNECTED, 895 G_MIRROR_EVENT_DONTWAIT); 896 } 897 switch (pbp->bio_cmd) { 898 case BIO_DELETE: 899 case BIO_WRITE: 900 pbp->bio_inbed--; 901 pbp->bio_children--; 902 break; 903 } 904 } 905 g_destroy_bio(bp); 906 907 switch (pbp->bio_cmd) { 908 case BIO_READ: 909 if (pbp->bio_children == pbp->bio_inbed) { 910 pbp->bio_error = 0; 911 mtx_lock(&sc->sc_queue_mtx); 912 bioq_disksort(&sc->sc_queue, pbp); 913 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 914 wakeup(sc); 915 mtx_unlock(&sc->sc_queue_mtx); 916 } 917 break; 918 case BIO_DELETE: 919 case BIO_WRITE: 920 if (pbp->bio_children == 0) { 921 /* 922 * All requests failed. 923 */ 924 } else if (pbp->bio_inbed < pbp->bio_children) { 925 /* Do nothing. */ 926 break; 927 } else if (pbp->bio_children == pbp->bio_inbed) { 928 /* Some requests succeeded. */ 929 pbp->bio_error = 0; 930 pbp->bio_completed = pbp->bio_length; 931 } 932 g_io_deliver(pbp, pbp->bio_error); 933 break; 934 default: 935 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 936 break; 937 } 938 } 939 940 static void 941 g_mirror_sync_done(struct bio *bp) 942 { 943 struct g_mirror_softc *sc; 944 945 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 946 sc = bp->bio_from->geom->softc; 947 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 948 mtx_lock(&sc->sc_queue_mtx); 949 bioq_disksort(&sc->sc_queue, bp); 950 wakeup(sc); 951 mtx_unlock(&sc->sc_queue_mtx); 952 } 953 954 static void 955 g_mirror_start(struct bio *bp) 956 { 957 struct g_mirror_softc *sc; 958 959 sc = bp->bio_to->geom->softc; 960 /* 961 * If sc == NULL or there are no valid disks, provider's error 962 * should be set and g_mirror_start() should not be called at all. 963 */ 964 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 965 ("Provider's error should be set (error=%d)(mirror=%s).", 966 bp->bio_to->error, bp->bio_to->name)); 967 G_MIRROR_LOGREQ(3, bp, "Request received."); 968 969 switch (bp->bio_cmd) { 970 case BIO_READ: 971 case BIO_WRITE: 972 case BIO_DELETE: 973 break; 974 case BIO_GETATTR: 975 default: 976 g_io_deliver(bp, EOPNOTSUPP); 977 return; 978 } 979 mtx_lock(&sc->sc_queue_mtx); 980 bioq_disksort(&sc->sc_queue, bp); 981 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 982 wakeup(sc); 983 mtx_unlock(&sc->sc_queue_mtx); 984 } 985 986 /* 987 * Send one synchronization request. 988 */ 989 static void 990 g_mirror_sync_one(struct g_mirror_disk *disk) 991 { 992 struct g_mirror_softc *sc; 993 struct bio *bp; 994 995 sc = disk->d_softc; 996 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 997 ("Disk %s is not marked for synchronization.", 998 g_mirror_get_diskname(disk))); 999 1000 bp = g_new_bio(); 1001 if (bp == NULL) 1002 return; 1003 bp->bio_parent = NULL; 1004 bp->bio_cmd = BIO_READ; 1005 bp->bio_offset = disk->d_sync.ds_offset; 1006 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1007 bp->bio_cflags = 0; 1008 bp->bio_done = g_mirror_sync_done; 1009 bp->bio_data = disk->d_sync.ds_data; 1010 if (bp->bio_data == NULL) { 1011 g_destroy_bio(bp); 1012 return; 1013 } 1014 disk->d_sync.ds_offset += bp->bio_length; 1015 bp->bio_to = sc->sc_provider; 1016 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1017 disk->d_sync.ds_consumer->index++; 1018 g_io_request(bp, disk->d_sync.ds_consumer); 1019 } 1020 1021 static void 1022 g_mirror_sync_request(struct bio *bp) 1023 { 1024 struct g_mirror_softc *sc; 1025 struct g_mirror_disk *disk; 1026 1027 bp->bio_from->index--; 1028 sc = bp->bio_from->geom->softc; 1029 disk = bp->bio_from->private; 1030 if (disk == NULL) { 1031 g_topology_lock(); 1032 g_mirror_kill_consumer(sc, bp->bio_from); 1033 g_topology_unlock(); 1034 g_destroy_bio(bp); 1035 return; 1036 } 1037 1038 /* 1039 * Synchronization request. 1040 */ 1041 switch (bp->bio_cmd) { 1042 case BIO_READ: 1043 { 1044 struct g_consumer *cp; 1045 1046 if (bp->bio_error != 0) { 1047 G_MIRROR_LOGREQ(0, bp, 1048 "Synchronization request failed (error=%d).", 1049 bp->bio_error); 1050 g_destroy_bio(bp); 1051 return; 1052 } 1053 G_MIRROR_LOGREQ(3, bp, 1054 "Synchronization request half-finished."); 1055 bp->bio_cmd = BIO_WRITE; 1056 bp->bio_cflags = 0; 1057 cp = disk->d_consumer; 1058 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1059 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1060 cp->acr, cp->acw, cp->ace)); 1061 cp->index++; 1062 g_io_request(bp, cp); 1063 return; 1064 } 1065 case BIO_WRITE: 1066 { 1067 struct g_mirror_disk_sync *sync; 1068 1069 if (bp->bio_error != 0) { 1070 G_MIRROR_LOGREQ(0, bp, 1071 "Synchronization request failed (error=%d).", 1072 bp->bio_error); 1073 g_destroy_bio(bp); 1074 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1075 g_mirror_event_send(disk, 1076 G_MIRROR_DISK_STATE_DISCONNECTED, 1077 G_MIRROR_EVENT_DONTWAIT); 1078 return; 1079 } 1080 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1081 sync = &disk->d_sync; 1082 sync->ds_offset_done = bp->bio_offset + bp->bio_length; 1083 g_destroy_bio(bp); 1084 if (sync->ds_resync != -1) 1085 break; 1086 if (sync->ds_offset_done == sc->sc_provider->mediasize) { 1087 /* 1088 * Disk up-to-date, activate it. 1089 */ 1090 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1091 G_MIRROR_EVENT_DONTWAIT); 1092 return; 1093 } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) { 1094 /* 1095 * Update offset_done on every 100 blocks. 1096 * XXX: This should be configurable. 1097 */ 1098 g_topology_lock(); 1099 g_mirror_update_metadata(disk); 1100 g_topology_unlock(); 1101 } 1102 return; 1103 } 1104 default: 1105 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1106 bp->bio_cmd, sc->sc_name)); 1107 break; 1108 } 1109 } 1110 1111 static void 1112 g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1113 { 1114 struct g_mirror_disk *disk; 1115 struct g_consumer *cp; 1116 struct bio *cbp; 1117 1118 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1119 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1120 break; 1121 } 1122 if (disk == NULL) { 1123 if (bp->bio_error == 0) 1124 bp->bio_error = ENXIO; 1125 g_io_deliver(bp, bp->bio_error); 1126 return; 1127 } 1128 cbp = g_clone_bio(bp); 1129 if (cbp == NULL) { 1130 if (bp->bio_error == 0) 1131 bp->bio_error = ENOMEM; 1132 g_io_deliver(bp, bp->bio_error); 1133 return; 1134 } 1135 /* 1136 * Fill in the component buf structure. 1137 */ 1138 cp = disk->d_consumer; 1139 cbp->bio_done = g_mirror_done; 1140 cbp->bio_to = cp->provider; 1141 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1142 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1143 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1144 cp->acw, cp->ace)); 1145 cp->index++; 1146 g_io_request(cbp, cp); 1147 } 1148 1149 static void 1150 g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1151 { 1152 struct g_mirror_disk *disk; 1153 struct g_consumer *cp; 1154 struct bio *cbp; 1155 1156 disk = g_mirror_get_disk(sc); 1157 if (disk == NULL) { 1158 if (bp->bio_error == 0) 1159 bp->bio_error = ENXIO; 1160 g_io_deliver(bp, bp->bio_error); 1161 return; 1162 } 1163 cbp = g_clone_bio(bp); 1164 if (cbp == NULL) { 1165 if (bp->bio_error == 0) 1166 bp->bio_error = ENOMEM; 1167 g_io_deliver(bp, bp->bio_error); 1168 return; 1169 } 1170 /* 1171 * Fill in the component buf structure. 1172 */ 1173 cp = disk->d_consumer; 1174 cbp->bio_done = g_mirror_done; 1175 cbp->bio_to = cp->provider; 1176 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1177 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1178 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1179 cp->acw, cp->ace)); 1180 cp->index++; 1181 g_io_request(cbp, cp); 1182 } 1183 1184 static void 1185 g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1186 { 1187 struct g_mirror_disk *disk, *dp; 1188 struct g_consumer *cp; 1189 struct bio *cbp; 1190 struct bintime curtime; 1191 1192 binuptime(&curtime); 1193 /* 1194 * Find a disk which the smallest load. 1195 */ 1196 disk = NULL; 1197 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1198 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1199 continue; 1200 /* If disk wasn't used for more than 2 sec, use it. */ 1201 if (curtime.sec - dp->d_last_used.sec >= 2) { 1202 disk = dp; 1203 break; 1204 } 1205 if (disk == NULL || 1206 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1207 disk = dp; 1208 } 1209 } 1210 cbp = g_clone_bio(bp); 1211 if (cbp == NULL) { 1212 if (bp->bio_error == 0) 1213 bp->bio_error = ENOMEM; 1214 g_io_deliver(bp, bp->bio_error); 1215 return; 1216 } 1217 /* 1218 * Fill in the component buf structure. 1219 */ 1220 cp = disk->d_consumer; 1221 cbp->bio_done = g_mirror_done; 1222 cbp->bio_to = cp->provider; 1223 binuptime(&disk->d_last_used); 1224 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1225 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1226 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1227 cp->acw, cp->ace)); 1228 cp->index++; 1229 g_io_request(cbp, cp); 1230 } 1231 1232 static void 1233 g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1234 { 1235 struct bio_queue_head queue; 1236 struct g_mirror_disk *disk; 1237 struct g_consumer *cp; 1238 struct bio *cbp; 1239 off_t left, mod, offset, slice; 1240 u_char *data; 1241 u_int ndisks; 1242 1243 if (bp->bio_length <= sc->sc_slice) { 1244 g_mirror_request_round_robin(sc, bp); 1245 return; 1246 } 1247 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1248 slice = bp->bio_length / ndisks; 1249 mod = slice % sc->sc_provider->sectorsize; 1250 if (mod != 0) 1251 slice += sc->sc_provider->sectorsize - mod; 1252 /* 1253 * Allocate all bios before sending any request, so we can 1254 * return ENOMEM in nice and clean way. 1255 */ 1256 left = bp->bio_length; 1257 offset = bp->bio_offset; 1258 data = bp->bio_data; 1259 bioq_init(&queue); 1260 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1261 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1262 continue; 1263 cbp = g_clone_bio(bp); 1264 if (cbp == NULL) { 1265 for (cbp = bioq_first(&queue); cbp != NULL; 1266 cbp = bioq_first(&queue)) { 1267 bioq_remove(&queue, cbp); 1268 g_destroy_bio(cbp); 1269 } 1270 if (bp->bio_error == 0) 1271 bp->bio_error = ENOMEM; 1272 g_io_deliver(bp, bp->bio_error); 1273 return; 1274 } 1275 bioq_insert_tail(&queue, cbp); 1276 cbp->bio_done = g_mirror_done; 1277 cbp->bio_caller1 = disk; 1278 cbp->bio_to = disk->d_consumer->provider; 1279 cbp->bio_offset = offset; 1280 cbp->bio_data = data; 1281 cbp->bio_length = MIN(left, slice); 1282 left -= cbp->bio_length; 1283 if (left == 0) 1284 break; 1285 offset += cbp->bio_length; 1286 data += cbp->bio_length; 1287 } 1288 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1289 bioq_remove(&queue, cbp); 1290 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1291 disk = cbp->bio_caller1; 1292 cbp->bio_caller1 = NULL; 1293 cp = disk->d_consumer; 1294 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1295 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1296 cp->acr, cp->acw, cp->ace)); 1297 disk->d_consumer->index++; 1298 g_io_request(cbp, disk->d_consumer); 1299 } 1300 } 1301 1302 static void 1303 g_mirror_register_request(struct bio *bp) 1304 { 1305 struct g_mirror_softc *sc; 1306 1307 sc = bp->bio_to->geom->softc; 1308 switch (bp->bio_cmd) { 1309 case BIO_READ: 1310 switch (sc->sc_balance) { 1311 case G_MIRROR_BALANCE_LOAD: 1312 g_mirror_request_load(sc, bp); 1313 break; 1314 case G_MIRROR_BALANCE_PREFER: 1315 g_mirror_request_prefer(sc, bp); 1316 break; 1317 case G_MIRROR_BALANCE_ROUND_ROBIN: 1318 g_mirror_request_round_robin(sc, bp); 1319 break; 1320 case G_MIRROR_BALANCE_SPLIT: 1321 g_mirror_request_split(sc, bp); 1322 break; 1323 } 1324 return; 1325 case BIO_WRITE: 1326 case BIO_DELETE: 1327 { 1328 struct g_mirror_disk *disk; 1329 struct g_mirror_disk_sync *sync; 1330 struct bio_queue_head queue; 1331 struct g_consumer *cp; 1332 struct bio *cbp; 1333 1334 if (sc->sc_idle) 1335 g_mirror_unidle(sc); 1336 /* 1337 * Allocate all bios before sending any request, so we can 1338 * return ENOMEM in nice and clean way. 1339 */ 1340 bioq_init(&queue); 1341 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1342 sync = &disk->d_sync; 1343 switch (disk->d_state) { 1344 case G_MIRROR_DISK_STATE_ACTIVE: 1345 break; 1346 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1347 if (bp->bio_offset >= sync->ds_offset) 1348 continue; 1349 else if (bp->bio_offset + bp->bio_length > 1350 sync->ds_offset_done && 1351 (bp->bio_offset < sync->ds_resync || 1352 sync->ds_resync == -1)) { 1353 sync->ds_resync = bp->bio_offset - 1354 (bp->bio_offset % MAXPHYS); 1355 } 1356 break; 1357 default: 1358 continue; 1359 } 1360 cbp = g_clone_bio(bp); 1361 if (cbp == NULL) { 1362 for (cbp = bioq_first(&queue); cbp != NULL; 1363 cbp = bioq_first(&queue)) { 1364 bioq_remove(&queue, cbp); 1365 g_destroy_bio(cbp); 1366 } 1367 if (bp->bio_error == 0) 1368 bp->bio_error = ENOMEM; 1369 g_io_deliver(bp, bp->bio_error); 1370 return; 1371 } 1372 bioq_insert_tail(&queue, cbp); 1373 cbp->bio_done = g_mirror_done; 1374 cp = disk->d_consumer; 1375 cbp->bio_caller1 = cp; 1376 cbp->bio_to = cp->provider; 1377 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1378 ("Consumer %s not opened (r%dw%de%d).", 1379 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1380 } 1381 for (cbp = bioq_first(&queue); cbp != NULL; 1382 cbp = bioq_first(&queue)) { 1383 bioq_remove(&queue, cbp); 1384 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1385 cp = cbp->bio_caller1; 1386 cbp->bio_caller1 = NULL; 1387 cp->index++; 1388 g_io_request(cbp, cp); 1389 } 1390 /* 1391 * Bump syncid on first write. 1392 */ 1393 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1394 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1395 g_topology_lock(); 1396 g_mirror_bump_syncid(sc); 1397 g_topology_unlock(); 1398 } 1399 return; 1400 } 1401 default: 1402 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1403 bp->bio_cmd, sc->sc_name)); 1404 break; 1405 } 1406 } 1407 1408 static int 1409 g_mirror_can_destroy(struct g_mirror_softc *sc) 1410 { 1411 struct g_geom *gp; 1412 struct g_consumer *cp; 1413 1414 g_topology_assert(); 1415 gp = sc->sc_geom; 1416 LIST_FOREACH(cp, &gp->consumer, consumer) { 1417 if (g_mirror_is_busy(sc, cp)) 1418 return (0); 1419 } 1420 gp = sc->sc_sync.ds_geom; 1421 LIST_FOREACH(cp, &gp->consumer, consumer) { 1422 if (g_mirror_is_busy(sc, cp)) 1423 return (0); 1424 } 1425 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1426 sc->sc_name); 1427 return (1); 1428 } 1429 1430 static int 1431 g_mirror_try_destroy(struct g_mirror_softc *sc) 1432 { 1433 1434 g_topology_lock(); 1435 if (!g_mirror_can_destroy(sc)) { 1436 g_topology_unlock(); 1437 return (0); 1438 } 1439 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1440 g_topology_unlock(); 1441 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1442 &sc->sc_worker); 1443 wakeup(&sc->sc_worker); 1444 sc->sc_worker = NULL; 1445 } else { 1446 g_mirror_destroy_device(sc); 1447 g_topology_unlock(); 1448 free(sc, M_MIRROR); 1449 } 1450 return (1); 1451 } 1452 1453 /* 1454 * Worker thread. 1455 */ 1456 static void 1457 g_mirror_worker(void *arg) 1458 { 1459 struct g_mirror_softc *sc; 1460 struct g_mirror_disk *disk; 1461 struct g_mirror_disk_sync *sync; 1462 struct g_mirror_event *ep; 1463 struct bio *bp; 1464 u_int nreqs; 1465 1466 sc = arg; 1467 mtx_lock_spin(&sched_lock); 1468 sched_prio(curthread, PRIBIO); 1469 mtx_unlock_spin(&sched_lock); 1470 1471 nreqs = 0; 1472 for (;;) { 1473 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1474 /* 1475 * First take a look at events. 1476 * This is important to handle events before any I/O requests. 1477 */ 1478 ep = g_mirror_event_get(sc); 1479 if (ep != NULL && g_topology_try_lock()) { 1480 g_mirror_event_remove(sc, ep); 1481 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1482 /* Update only device status. */ 1483 G_MIRROR_DEBUG(3, 1484 "Running event for device %s.", 1485 sc->sc_name); 1486 ep->e_error = 0; 1487 g_mirror_update_device(sc, 1); 1488 } else { 1489 /* Update disk status. */ 1490 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1491 g_mirror_get_diskname(ep->e_disk)); 1492 ep->e_error = g_mirror_update_disk(ep->e_disk, 1493 ep->e_state); 1494 if (ep->e_error == 0) 1495 g_mirror_update_device(sc, 0); 1496 } 1497 g_topology_unlock(); 1498 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1499 KASSERT(ep->e_error == 0, 1500 ("Error cannot be handled.")); 1501 g_mirror_event_free(ep); 1502 } else { 1503 ep->e_flags |= G_MIRROR_EVENT_DONE; 1504 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1505 ep); 1506 mtx_lock(&sc->sc_events_mtx); 1507 wakeup(ep); 1508 mtx_unlock(&sc->sc_events_mtx); 1509 } 1510 if ((sc->sc_flags & 1511 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1512 if (g_mirror_try_destroy(sc)) 1513 kthread_exit(0); 1514 } 1515 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1516 continue; 1517 } 1518 /* 1519 * Now I/O requests. 1520 */ 1521 /* Get first request from the queue. */ 1522 mtx_lock(&sc->sc_queue_mtx); 1523 bp = bioq_first(&sc->sc_queue); 1524 if (bp == NULL) { 1525 if (ep != NULL) { 1526 /* 1527 * No I/O requests and topology lock was 1528 * already held? Try again. 1529 */ 1530 mtx_unlock(&sc->sc_queue_mtx); 1531 continue; 1532 } 1533 if ((sc->sc_flags & 1534 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1535 mtx_unlock(&sc->sc_queue_mtx); 1536 if (g_mirror_try_destroy(sc)) 1537 kthread_exit(0); 1538 mtx_lock(&sc->sc_queue_mtx); 1539 } 1540 } 1541 if (sc->sc_sync.ds_ndisks > 0 && 1542 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1543 mtx_unlock(&sc->sc_queue_mtx); 1544 /* 1545 * It is time for synchronization... 1546 */ 1547 nreqs = 0; 1548 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1549 if (disk->d_state != 1550 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1551 continue; 1552 } 1553 sync = &disk->d_sync; 1554 if (sync->ds_offset >= 1555 sc->sc_provider->mediasize) { 1556 continue; 1557 } 1558 if (sync->ds_offset > sync->ds_offset_done) 1559 continue; 1560 if (sync->ds_resync != -1) { 1561 sync->ds_offset = sync->ds_resync; 1562 sync->ds_offset_done = sync->ds_resync; 1563 sync->ds_resync = -1; 1564 } 1565 g_mirror_sync_one(disk); 1566 } 1567 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1568 goto sleep; 1569 } 1570 if (bp == NULL) { 1571 if (g_mirror_check_idle(sc)) { 1572 u_int idletime; 1573 1574 idletime = g_mirror_idletime; 1575 if (idletime == 0) 1576 idletime = 1; 1577 idletime *= hz; 1578 if (msleep(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1579 "m:w1", idletime) == EWOULDBLOCK) { 1580 G_MIRROR_DEBUG(5, "%s: I'm here 3.", 1581 __func__); 1582 /* 1583 * No I/O requests in 'idletime' seconds, 1584 * so mark components as clean. 1585 */ 1586 g_mirror_idle(sc); 1587 } 1588 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1589 } else { 1590 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1591 "m:w2", 0); 1592 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1593 } 1594 continue; 1595 } 1596 nreqs++; 1597 bioq_remove(&sc->sc_queue, bp); 1598 mtx_unlock(&sc->sc_queue_mtx); 1599 1600 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1601 g_mirror_regular_request(bp); 1602 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1603 u_int timeout, sps; 1604 1605 g_mirror_sync_request(bp); 1606 sleep: 1607 sps = g_mirror_syncs_per_sec; 1608 if (sps == 0) { 1609 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1610 continue; 1611 } 1612 if (ep != NULL) { 1613 /* 1614 * We have some pending events, don't sleep now. 1615 */ 1616 G_MIRROR_DEBUG(5, "%s: I'm here 7.", __func__); 1617 continue; 1618 } 1619 mtx_lock(&sc->sc_queue_mtx); 1620 if (bioq_first(&sc->sc_queue) != NULL) { 1621 mtx_unlock(&sc->sc_queue_mtx); 1622 G_MIRROR_DEBUG(5, "%s: I'm here 8.", __func__); 1623 continue; 1624 } 1625 timeout = hz / sps; 1626 if (timeout == 0) 1627 timeout = 1; 1628 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w3", 1629 timeout); 1630 } else { 1631 g_mirror_register_request(bp); 1632 } 1633 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1634 } 1635 } 1636 1637 /* 1638 * Open disk's consumer if needed. 1639 */ 1640 static void 1641 g_mirror_update_access(struct g_mirror_disk *disk) 1642 { 1643 struct g_provider *pp; 1644 1645 g_topology_assert(); 1646 1647 pp = disk->d_softc->sc_provider; 1648 if (pp == NULL) 1649 return; 1650 if (pp->acw > 0) { 1651 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1652 G_MIRROR_DEBUG(1, 1653 "Disk %s (device %s) marked as dirty.", 1654 g_mirror_get_diskname(disk), 1655 disk->d_softc->sc_name); 1656 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1657 } 1658 } else if (pp->acw == 0) { 1659 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1660 G_MIRROR_DEBUG(1, 1661 "Disk %s (device %s) marked as clean.", 1662 g_mirror_get_diskname(disk), 1663 disk->d_softc->sc_name); 1664 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1665 } 1666 } 1667 } 1668 1669 static void 1670 g_mirror_sync_start(struct g_mirror_disk *disk) 1671 { 1672 struct g_mirror_softc *sc; 1673 int error; 1674 1675 g_topology_assert(); 1676 1677 sc = disk->d_softc; 1678 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1679 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1680 sc->sc_state)); 1681 1682 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1683 g_mirror_get_diskname(disk)); 1684 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1685 KASSERT(disk->d_sync.ds_consumer == NULL, 1686 ("Sync consumer already exists (device=%s, disk=%s).", 1687 sc->sc_name, g_mirror_get_diskname(disk))); 1688 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1689 disk->d_sync.ds_consumer->private = disk; 1690 disk->d_sync.ds_consumer->index = 0; 1691 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1692 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1693 disk->d_softc->sc_name, error)); 1694 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1695 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1696 disk->d_softc->sc_name, error)); 1697 disk->d_sync.ds_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1698 sc->sc_sync.ds_ndisks++; 1699 } 1700 1701 /* 1702 * Stop synchronization process. 1703 * type: 0 - synchronization finished 1704 * 1 - synchronization stopped 1705 */ 1706 static void 1707 g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1708 { 1709 1710 g_topology_assert(); 1711 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1712 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1713 g_mirror_disk_state2str(disk->d_state))); 1714 if (disk->d_sync.ds_consumer == NULL) 1715 return; 1716 1717 if (type == 0) { 1718 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1719 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1720 } else /* if (type == 1) */ { 1721 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1722 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1723 } 1724 g_mirror_kill_consumer(disk->d_softc, disk->d_sync.ds_consumer); 1725 free(disk->d_sync.ds_data, M_MIRROR); 1726 disk->d_sync.ds_consumer = NULL; 1727 disk->d_softc->sc_sync.ds_ndisks--; 1728 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1729 } 1730 1731 static void 1732 g_mirror_launch_provider(struct g_mirror_softc *sc) 1733 { 1734 struct g_mirror_disk *disk; 1735 struct g_provider *pp; 1736 1737 g_topology_assert(); 1738 1739 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1740 pp->mediasize = sc->sc_mediasize; 1741 pp->sectorsize = sc->sc_sectorsize; 1742 sc->sc_provider = pp; 1743 g_error_provider(pp, 0); 1744 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1745 pp->name); 1746 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1747 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1748 g_mirror_sync_start(disk); 1749 } 1750 } 1751 1752 static void 1753 g_mirror_destroy_provider(struct g_mirror_softc *sc) 1754 { 1755 struct g_mirror_disk *disk; 1756 struct bio *bp; 1757 1758 g_topology_assert(); 1759 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1760 sc->sc_name)); 1761 1762 g_error_provider(sc->sc_provider, ENXIO); 1763 mtx_lock(&sc->sc_queue_mtx); 1764 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1765 bioq_remove(&sc->sc_queue, bp); 1766 g_io_deliver(bp, ENXIO); 1767 } 1768 mtx_unlock(&sc->sc_queue_mtx); 1769 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1770 sc->sc_provider->name); 1771 sc->sc_provider->flags |= G_PF_WITHER; 1772 g_orphan_provider(sc->sc_provider, ENXIO); 1773 sc->sc_provider = NULL; 1774 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1775 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1776 g_mirror_sync_stop(disk, 1); 1777 } 1778 } 1779 1780 static void 1781 g_mirror_go(void *arg) 1782 { 1783 struct g_mirror_softc *sc; 1784 1785 sc = arg; 1786 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1787 g_mirror_event_send(sc, 0, 1788 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1789 } 1790 1791 static u_int 1792 g_mirror_determine_state(struct g_mirror_disk *disk) 1793 { 1794 struct g_mirror_softc *sc; 1795 u_int state; 1796 1797 sc = disk->d_softc; 1798 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1799 if ((disk->d_flags & 1800 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1801 /* Disk does not need synchronization. */ 1802 state = G_MIRROR_DISK_STATE_ACTIVE; 1803 } else { 1804 if ((sc->sc_flags & 1805 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1806 (disk->d_flags & 1807 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1808 /* 1809 * We can start synchronization from 1810 * the stored offset. 1811 */ 1812 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1813 } else { 1814 state = G_MIRROR_DISK_STATE_STALE; 1815 } 1816 } 1817 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1818 /* 1819 * Reset all synchronization data for this disk, 1820 * because if it even was synchronized, it was 1821 * synchronized to disks with different syncid. 1822 */ 1823 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1824 disk->d_sync.ds_offset = 0; 1825 disk->d_sync.ds_offset_done = 0; 1826 disk->d_sync.ds_syncid = sc->sc_syncid; 1827 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1828 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1829 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1830 } else { 1831 state = G_MIRROR_DISK_STATE_STALE; 1832 } 1833 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1834 /* 1835 * Not good, NOT GOOD! 1836 * It means that mirror was started on stale disks 1837 * and more fresh disk just arrive. 1838 * If there were writes, mirror is fucked up, sorry. 1839 * I think the best choice here is don't touch 1840 * this disk and inform the user laudly. 1841 */ 1842 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1843 "disk (%s) arrives!! It will not be connected to the " 1844 "running device.", sc->sc_name, 1845 g_mirror_get_diskname(disk)); 1846 g_mirror_destroy_disk(disk); 1847 state = G_MIRROR_DISK_STATE_NONE; 1848 /* Return immediately, because disk was destroyed. */ 1849 return (state); 1850 } 1851 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1852 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1853 return (state); 1854 } 1855 1856 /* 1857 * Update device state. 1858 */ 1859 static void 1860 g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1861 { 1862 struct g_mirror_disk *disk; 1863 u_int state; 1864 1865 g_topology_assert(); 1866 1867 switch (sc->sc_state) { 1868 case G_MIRROR_DEVICE_STATE_STARTING: 1869 { 1870 struct g_mirror_disk *pdisk, *tdisk; 1871 u_int dirty, ndisks, genid, syncid; 1872 1873 KASSERT(sc->sc_provider == NULL, 1874 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1875 /* 1876 * Are we ready? We are, if all disks are connected or 1877 * if we have any disks and 'force' is true. 1878 */ 1879 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1880 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1881 ; 1882 } else if (g_mirror_ndisks(sc, -1) == 0) { 1883 /* 1884 * Disks went down in starting phase, so destroy 1885 * device. 1886 */ 1887 callout_drain(&sc->sc_callout); 1888 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1889 return; 1890 } else { 1891 return; 1892 } 1893 1894 /* 1895 * Activate all disks with the biggest syncid. 1896 */ 1897 if (force) { 1898 /* 1899 * If 'force' is true, we have been called due to 1900 * timeout, so don't bother canceling timeout. 1901 */ 1902 ndisks = 0; 1903 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1904 if ((disk->d_flags & 1905 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1906 ndisks++; 1907 } 1908 } 1909 if (ndisks == 0) { 1910 /* No valid disks found, destroy device. */ 1911 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1912 return; 1913 } 1914 } else { 1915 /* Cancel timeout. */ 1916 callout_drain(&sc->sc_callout); 1917 } 1918 1919 /* 1920 * Find the biggest genid. 1921 */ 1922 genid = 0; 1923 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1924 if (disk->d_genid > genid) 1925 genid = disk->d_genid; 1926 } 1927 sc->sc_genid = genid; 1928 /* 1929 * Remove all disks without the biggest genid. 1930 */ 1931 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 1932 if (disk->d_genid < genid) { 1933 G_MIRROR_DEBUG(0, 1934 "Component %s (device %s) broken, skipping.", 1935 g_mirror_get_diskname(disk), sc->sc_name); 1936 g_mirror_destroy_disk(disk); 1937 } 1938 } 1939 1940 /* 1941 * Find the biggest syncid. 1942 */ 1943 syncid = 0; 1944 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1945 if (disk->d_sync.ds_syncid > syncid) 1946 syncid = disk->d_sync.ds_syncid; 1947 } 1948 1949 /* 1950 * Here we need to look for dirty disks and if all disks 1951 * with the biggest syncid are dirty, we have to choose 1952 * one with the biggest priority and rebuild the rest. 1953 */ 1954 /* 1955 * Find the number of dirty disks with the biggest syncid. 1956 * Find the number of disks with the biggest syncid. 1957 * While here, find a disk with the biggest priority. 1958 */ 1959 dirty = ndisks = 0; 1960 pdisk = NULL; 1961 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1962 if (disk->d_sync.ds_syncid != syncid) 1963 continue; 1964 if ((disk->d_flags & 1965 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1966 continue; 1967 } 1968 ndisks++; 1969 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1970 dirty++; 1971 if (pdisk == NULL || 1972 pdisk->d_priority < disk->d_priority) { 1973 pdisk = disk; 1974 } 1975 } 1976 } 1977 if (dirty == 0) { 1978 /* No dirty disks at all, great. */ 1979 } else if (dirty == ndisks) { 1980 /* 1981 * Force synchronization for all dirty disks except one 1982 * with the biggest priority. 1983 */ 1984 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1985 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1986 "master disk for synchronization.", 1987 g_mirror_get_diskname(pdisk), sc->sc_name); 1988 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1989 if (disk->d_sync.ds_syncid != syncid) 1990 continue; 1991 if ((disk->d_flags & 1992 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1993 continue; 1994 } 1995 KASSERT((disk->d_flags & 1996 G_MIRROR_DISK_FLAG_DIRTY) != 0, 1997 ("Disk %s isn't marked as dirty.", 1998 g_mirror_get_diskname(disk))); 1999 /* Skip the disk with the biggest priority. */ 2000 if (disk == pdisk) 2001 continue; 2002 disk->d_sync.ds_syncid = 0; 2003 } 2004 } else if (dirty < ndisks) { 2005 /* 2006 * Force synchronization for all dirty disks. 2007 * We have some non-dirty disks. 2008 */ 2009 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2010 if (disk->d_sync.ds_syncid != syncid) 2011 continue; 2012 if ((disk->d_flags & 2013 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2014 continue; 2015 } 2016 if ((disk->d_flags & 2017 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2018 continue; 2019 } 2020 disk->d_sync.ds_syncid = 0; 2021 } 2022 } 2023 2024 /* Reset hint. */ 2025 sc->sc_hint = NULL; 2026 sc->sc_syncid = syncid; 2027 if (force) { 2028 /* Remember to bump syncid on first write. */ 2029 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2030 } 2031 state = G_MIRROR_DEVICE_STATE_RUNNING; 2032 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2033 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2034 g_mirror_device_state2str(state)); 2035 sc->sc_state = state; 2036 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2037 state = g_mirror_determine_state(disk); 2038 g_mirror_event_send(disk, state, 2039 G_MIRROR_EVENT_DONTWAIT); 2040 if (state == G_MIRROR_DISK_STATE_STALE) 2041 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2042 } 2043 wakeup(&g_mirror_class); 2044 break; 2045 } 2046 case G_MIRROR_DEVICE_STATE_RUNNING: 2047 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2048 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2049 /* 2050 * No active disks or no disks at all, 2051 * so destroy device. 2052 */ 2053 if (sc->sc_provider != NULL) 2054 g_mirror_destroy_provider(sc); 2055 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2056 break; 2057 } else if (g_mirror_ndisks(sc, 2058 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2059 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2060 /* 2061 * We have active disks, launch provider if it doesn't 2062 * exist. 2063 */ 2064 if (sc->sc_provider == NULL) 2065 g_mirror_launch_provider(sc); 2066 } 2067 /* 2068 * Genid should be bumped immediately, so do it here. 2069 */ 2070 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2071 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2072 g_mirror_bump_genid(sc); 2073 } 2074 break; 2075 default: 2076 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2077 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2078 break; 2079 } 2080 } 2081 2082 /* 2083 * Update disk state and device state if needed. 2084 */ 2085 #define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2086 "Disk %s state changed from %s to %s (device %s).", \ 2087 g_mirror_get_diskname(disk), \ 2088 g_mirror_disk_state2str(disk->d_state), \ 2089 g_mirror_disk_state2str(state), sc->sc_name) 2090 static int 2091 g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2092 { 2093 struct g_mirror_softc *sc; 2094 2095 g_topology_assert(); 2096 2097 sc = disk->d_softc; 2098 again: 2099 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2100 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2101 g_mirror_disk_state2str(state)); 2102 switch (state) { 2103 case G_MIRROR_DISK_STATE_NEW: 2104 /* 2105 * Possible scenarios: 2106 * 1. New disk arrive. 2107 */ 2108 /* Previous state should be NONE. */ 2109 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2110 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2111 g_mirror_disk_state2str(disk->d_state))); 2112 DISK_STATE_CHANGED(); 2113 2114 disk->d_state = state; 2115 if (LIST_EMPTY(&sc->sc_disks)) 2116 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2117 else { 2118 struct g_mirror_disk *dp; 2119 2120 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2121 if (disk->d_priority >= dp->d_priority) { 2122 LIST_INSERT_BEFORE(dp, disk, d_next); 2123 dp = NULL; 2124 break; 2125 } 2126 if (LIST_NEXT(dp, d_next) == NULL) 2127 break; 2128 } 2129 if (dp != NULL) 2130 LIST_INSERT_AFTER(dp, disk, d_next); 2131 } 2132 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2133 sc->sc_name, g_mirror_get_diskname(disk)); 2134 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2135 break; 2136 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2137 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2138 g_mirror_device_state2str(sc->sc_state), 2139 g_mirror_get_diskname(disk), 2140 g_mirror_disk_state2str(disk->d_state))); 2141 state = g_mirror_determine_state(disk); 2142 if (state != G_MIRROR_DISK_STATE_NONE) 2143 goto again; 2144 break; 2145 case G_MIRROR_DISK_STATE_ACTIVE: 2146 /* 2147 * Possible scenarios: 2148 * 1. New disk does not need synchronization. 2149 * 2. Synchronization process finished successfully. 2150 */ 2151 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2152 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2153 g_mirror_device_state2str(sc->sc_state), 2154 g_mirror_get_diskname(disk), 2155 g_mirror_disk_state2str(disk->d_state))); 2156 /* Previous state should be NEW or SYNCHRONIZING. */ 2157 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2158 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2159 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2160 g_mirror_disk_state2str(disk->d_state))); 2161 DISK_STATE_CHANGED(); 2162 2163 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2164 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2165 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2166 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2167 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2168 g_mirror_sync_stop(disk, 0); 2169 } 2170 disk->d_state = state; 2171 disk->d_sync.ds_offset = 0; 2172 disk->d_sync.ds_offset_done = 0; 2173 g_mirror_update_access(disk); 2174 g_mirror_update_metadata(disk); 2175 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2176 sc->sc_name, g_mirror_get_diskname(disk)); 2177 break; 2178 case G_MIRROR_DISK_STATE_STALE: 2179 /* 2180 * Possible scenarios: 2181 * 1. Stale disk was connected. 2182 */ 2183 /* Previous state should be NEW. */ 2184 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2185 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2186 g_mirror_disk_state2str(disk->d_state))); 2187 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2188 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2189 g_mirror_device_state2str(sc->sc_state), 2190 g_mirror_get_diskname(disk), 2191 g_mirror_disk_state2str(disk->d_state))); 2192 /* 2193 * STALE state is only possible if device is marked 2194 * NOAUTOSYNC. 2195 */ 2196 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2197 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2198 g_mirror_device_state2str(sc->sc_state), 2199 g_mirror_get_diskname(disk), 2200 g_mirror_disk_state2str(disk->d_state))); 2201 DISK_STATE_CHANGED(); 2202 2203 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2204 disk->d_state = state; 2205 g_mirror_update_metadata(disk); 2206 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2207 sc->sc_name, g_mirror_get_diskname(disk)); 2208 break; 2209 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2210 /* 2211 * Possible scenarios: 2212 * 1. Disk which needs synchronization was connected. 2213 */ 2214 /* Previous state should be NEW. */ 2215 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2216 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2217 g_mirror_disk_state2str(disk->d_state))); 2218 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2219 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2220 g_mirror_device_state2str(sc->sc_state), 2221 g_mirror_get_diskname(disk), 2222 g_mirror_disk_state2str(disk->d_state))); 2223 DISK_STATE_CHANGED(); 2224 2225 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2226 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2227 disk->d_state = state; 2228 if (sc->sc_provider != NULL) { 2229 g_mirror_sync_start(disk); 2230 g_mirror_update_metadata(disk); 2231 } 2232 break; 2233 case G_MIRROR_DISK_STATE_DISCONNECTED: 2234 /* 2235 * Possible scenarios: 2236 * 1. Device wasn't running yet, but disk disappear. 2237 * 2. Disk was active and disapppear. 2238 * 3. Disk disappear during synchronization process. 2239 */ 2240 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2241 /* 2242 * Previous state should be ACTIVE, STALE or 2243 * SYNCHRONIZING. 2244 */ 2245 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2246 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2247 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2248 ("Wrong disk state (%s, %s).", 2249 g_mirror_get_diskname(disk), 2250 g_mirror_disk_state2str(disk->d_state))); 2251 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2252 /* Previous state should be NEW. */ 2253 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2254 ("Wrong disk state (%s, %s).", 2255 g_mirror_get_diskname(disk), 2256 g_mirror_disk_state2str(disk->d_state))); 2257 /* 2258 * Reset bumping syncid if disk disappeared in STARTING 2259 * state. 2260 */ 2261 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2262 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2263 #ifdef INVARIANTS 2264 } else { 2265 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2266 sc->sc_name, 2267 g_mirror_device_state2str(sc->sc_state), 2268 g_mirror_get_diskname(disk), 2269 g_mirror_disk_state2str(disk->d_state))); 2270 #endif 2271 } 2272 DISK_STATE_CHANGED(); 2273 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2274 sc->sc_name, g_mirror_get_diskname(disk)); 2275 2276 g_mirror_destroy_disk(disk); 2277 break; 2278 case G_MIRROR_DISK_STATE_DESTROY: 2279 { 2280 int error; 2281 2282 error = g_mirror_clear_metadata(disk); 2283 if (error != 0) 2284 return (error); 2285 DISK_STATE_CHANGED(); 2286 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2287 sc->sc_name, g_mirror_get_diskname(disk)); 2288 2289 g_mirror_destroy_disk(disk); 2290 sc->sc_ndisks--; 2291 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2292 g_mirror_update_metadata(disk); 2293 } 2294 break; 2295 } 2296 default: 2297 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2298 break; 2299 } 2300 return (0); 2301 } 2302 #undef DISK_STATE_CHANGED 2303 2304 int 2305 g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2306 { 2307 struct g_provider *pp; 2308 u_char *buf; 2309 int error; 2310 2311 g_topology_assert(); 2312 2313 error = g_access(cp, 1, 0, 0); 2314 if (error != 0) 2315 return (error); 2316 pp = cp->provider; 2317 g_topology_unlock(); 2318 /* Metadata are stored on last sector. */ 2319 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2320 &error); 2321 g_topology_lock(); 2322 g_access(cp, -1, 0, 0); 2323 if (error != 0) { 2324 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2325 cp->provider->name, error); 2326 if (buf != NULL) 2327 g_free(buf); 2328 return (error); 2329 } 2330 2331 /* Decode metadata. */ 2332 error = mirror_metadata_decode(buf, md); 2333 g_free(buf); 2334 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2335 return (EINVAL); 2336 if (md->md_version > G_MIRROR_VERSION) { 2337 G_MIRROR_DEBUG(0, 2338 "Kernel module is too old to handle metadata from %s.", 2339 cp->provider->name); 2340 return (EINVAL); 2341 } 2342 if (error != 0) { 2343 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2344 cp->provider->name); 2345 return (error); 2346 } 2347 2348 return (0); 2349 } 2350 2351 static int 2352 g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2353 struct g_mirror_metadata *md) 2354 { 2355 2356 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2357 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2358 pp->name, md->md_did); 2359 return (EEXIST); 2360 } 2361 if (md->md_all != sc->sc_ndisks) { 2362 G_MIRROR_DEBUG(1, 2363 "Invalid '%s' field on disk %s (device %s), skipping.", 2364 "md_all", pp->name, sc->sc_name); 2365 return (EINVAL); 2366 } 2367 if (md->md_slice != sc->sc_slice) { 2368 G_MIRROR_DEBUG(1, 2369 "Invalid '%s' field on disk %s (device %s), skipping.", 2370 "md_slice", pp->name, sc->sc_name); 2371 return (EINVAL); 2372 } 2373 if (md->md_balance != sc->sc_balance) { 2374 G_MIRROR_DEBUG(1, 2375 "Invalid '%s' field on disk %s (device %s), skipping.", 2376 "md_balance", pp->name, sc->sc_name); 2377 return (EINVAL); 2378 } 2379 if (md->md_mediasize != sc->sc_mediasize) { 2380 G_MIRROR_DEBUG(1, 2381 "Invalid '%s' field on disk %s (device %s), skipping.", 2382 "md_mediasize", pp->name, sc->sc_name); 2383 return (EINVAL); 2384 } 2385 if (sc->sc_mediasize > pp->mediasize) { 2386 G_MIRROR_DEBUG(1, 2387 "Invalid size of disk %s (device %s), skipping.", pp->name, 2388 sc->sc_name); 2389 return (EINVAL); 2390 } 2391 if (md->md_sectorsize != sc->sc_sectorsize) { 2392 G_MIRROR_DEBUG(1, 2393 "Invalid '%s' field on disk %s (device %s), skipping.", 2394 "md_sectorsize", pp->name, sc->sc_name); 2395 return (EINVAL); 2396 } 2397 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2398 G_MIRROR_DEBUG(1, 2399 "Invalid sector size of disk %s (device %s), skipping.", 2400 pp->name, sc->sc_name); 2401 return (EINVAL); 2402 } 2403 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2404 G_MIRROR_DEBUG(1, 2405 "Invalid device flags on disk %s (device %s), skipping.", 2406 pp->name, sc->sc_name); 2407 return (EINVAL); 2408 } 2409 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2410 G_MIRROR_DEBUG(1, 2411 "Invalid disk flags on disk %s (device %s), skipping.", 2412 pp->name, sc->sc_name); 2413 return (EINVAL); 2414 } 2415 return (0); 2416 } 2417 2418 int 2419 g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2420 struct g_mirror_metadata *md) 2421 { 2422 struct g_mirror_disk *disk; 2423 int error; 2424 2425 g_topology_assert(); 2426 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2427 2428 error = g_mirror_check_metadata(sc, pp, md); 2429 if (error != 0) 2430 return (error); 2431 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2432 md->md_genid < sc->sc_genid) { 2433 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2434 pp->name, sc->sc_name); 2435 return (EINVAL); 2436 } 2437 disk = g_mirror_init_disk(sc, pp, md, &error); 2438 if (disk == NULL) 2439 return (error); 2440 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2441 G_MIRROR_EVENT_WAIT); 2442 if (error != 0) 2443 return (error); 2444 if (md->md_version < G_MIRROR_VERSION) { 2445 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2446 pp->name, md->md_version, G_MIRROR_VERSION); 2447 g_mirror_update_metadata(disk); 2448 } 2449 return (0); 2450 } 2451 2452 static int 2453 g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2454 { 2455 struct g_mirror_softc *sc; 2456 struct g_mirror_disk *disk; 2457 int dcr, dcw, dce; 2458 2459 g_topology_assert(); 2460 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2461 acw, ace); 2462 2463 dcr = pp->acr + acr; 2464 dcw = pp->acw + acw; 2465 dce = pp->ace + ace; 2466 2467 sc = pp->geom->softc; 2468 if (sc == NULL || LIST_EMPTY(&sc->sc_disks) || 2469 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 2470 if (acr <= 0 && acw <= 0 && ace <= 0) 2471 return (0); 2472 else 2473 return (ENXIO); 2474 } 2475 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2476 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2477 continue; 2478 /* 2479 * Mark disk as dirty on open and unmark on close. 2480 */ 2481 if (pp->acw == 0 && dcw > 0) { 2482 G_MIRROR_DEBUG(1, 2483 "Disk %s (device %s) marked as dirty.", 2484 g_mirror_get_diskname(disk), sc->sc_name); 2485 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2486 g_mirror_update_metadata(disk); 2487 } else if (pp->acw > 0 && dcw == 0) { 2488 G_MIRROR_DEBUG(1, 2489 "Disk %s (device %s) marked as clean.", 2490 g_mirror_get_diskname(disk), sc->sc_name); 2491 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2492 g_mirror_update_metadata(disk); 2493 } 2494 } 2495 return (0); 2496 } 2497 2498 static struct g_geom * 2499 g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2500 { 2501 struct g_mirror_softc *sc; 2502 struct g_geom *gp; 2503 int error, timeout; 2504 2505 g_topology_assert(); 2506 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2507 md->md_mid); 2508 2509 /* One disk is minimum. */ 2510 if (md->md_all < 1) 2511 return (NULL); 2512 /* 2513 * Action geom. 2514 */ 2515 gp = g_new_geomf(mp, "%s", md->md_name); 2516 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2517 gp->start = g_mirror_start; 2518 gp->orphan = g_mirror_orphan; 2519 gp->access = g_mirror_access; 2520 gp->dumpconf = g_mirror_dumpconf; 2521 2522 sc->sc_id = md->md_mid; 2523 sc->sc_slice = md->md_slice; 2524 sc->sc_balance = md->md_balance; 2525 sc->sc_mediasize = md->md_mediasize; 2526 sc->sc_sectorsize = md->md_sectorsize; 2527 sc->sc_ndisks = md->md_all; 2528 sc->sc_flags = md->md_mflags; 2529 sc->sc_bump_id = 0; 2530 sc->sc_idle = 0; 2531 bioq_init(&sc->sc_queue); 2532 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2533 LIST_INIT(&sc->sc_disks); 2534 TAILQ_INIT(&sc->sc_events); 2535 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2536 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2537 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2538 gp->softc = sc; 2539 sc->sc_geom = gp; 2540 sc->sc_provider = NULL; 2541 /* 2542 * Synchronization geom. 2543 */ 2544 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2545 gp->softc = sc; 2546 gp->orphan = g_mirror_orphan; 2547 sc->sc_sync.ds_geom = gp; 2548 sc->sc_sync.ds_ndisks = 0; 2549 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2550 "g_mirror %s", md->md_name); 2551 if (error != 0) { 2552 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2553 sc->sc_name); 2554 g_destroy_geom(sc->sc_sync.ds_geom); 2555 mtx_destroy(&sc->sc_events_mtx); 2556 mtx_destroy(&sc->sc_queue_mtx); 2557 g_destroy_geom(sc->sc_geom); 2558 free(sc, M_MIRROR); 2559 return (NULL); 2560 } 2561 2562 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2563 2564 /* 2565 * Run timeout. 2566 */ 2567 timeout = g_mirror_timeout * hz; 2568 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2569 return (sc->sc_geom); 2570 } 2571 2572 int 2573 g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2574 { 2575 struct g_provider *pp; 2576 2577 g_topology_assert(); 2578 2579 if (sc == NULL) 2580 return (ENXIO); 2581 pp = sc->sc_provider; 2582 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2583 if (force) { 2584 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 2585 "can't be definitely removed.", pp->name); 2586 } else { 2587 G_MIRROR_DEBUG(1, 2588 "Device %s is still open (r%dw%de%d).", pp->name, 2589 pp->acr, pp->acw, pp->ace); 2590 return (EBUSY); 2591 } 2592 } 2593 2594 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2595 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2596 g_topology_unlock(); 2597 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2598 mtx_lock(&sc->sc_queue_mtx); 2599 wakeup(sc); 2600 mtx_unlock(&sc->sc_queue_mtx); 2601 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2602 while (sc->sc_worker != NULL) 2603 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2604 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2605 g_topology_lock(); 2606 g_mirror_destroy_device(sc); 2607 free(sc, M_MIRROR); 2608 return (0); 2609 } 2610 2611 static void 2612 g_mirror_taste_orphan(struct g_consumer *cp) 2613 { 2614 2615 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2616 cp->provider->name)); 2617 } 2618 2619 static struct g_geom * 2620 g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2621 { 2622 struct g_mirror_metadata md; 2623 struct g_mirror_softc *sc; 2624 struct g_consumer *cp; 2625 struct g_geom *gp; 2626 int error; 2627 2628 g_topology_assert(); 2629 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2630 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2631 2632 gp = g_new_geomf(mp, "mirror:taste"); 2633 /* 2634 * This orphan function should be never called. 2635 */ 2636 gp->orphan = g_mirror_taste_orphan; 2637 cp = g_new_consumer(gp); 2638 g_attach(cp, pp); 2639 error = g_mirror_read_metadata(cp, &md); 2640 g_detach(cp); 2641 g_destroy_consumer(cp); 2642 g_destroy_geom(gp); 2643 if (error != 0) 2644 return (NULL); 2645 gp = NULL; 2646 2647 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2648 return (NULL); 2649 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2650 G_MIRROR_DEBUG(0, 2651 "Device %s: provider %s marked as inactive, skipping.", 2652 md.md_name, pp->name); 2653 return (NULL); 2654 } 2655 if (g_mirror_debug >= 2) 2656 mirror_metadata_dump(&md); 2657 2658 /* 2659 * Let's check if device already exists. 2660 */ 2661 sc = NULL; 2662 LIST_FOREACH(gp, &mp->geom, geom) { 2663 sc = gp->softc; 2664 if (sc == NULL) 2665 continue; 2666 if (sc->sc_sync.ds_geom == gp) 2667 continue; 2668 if (strcmp(md.md_name, sc->sc_name) != 0) 2669 continue; 2670 if (md.md_mid != sc->sc_id) { 2671 G_MIRROR_DEBUG(0, "Device %s already configured.", 2672 sc->sc_name); 2673 return (NULL); 2674 } 2675 break; 2676 } 2677 if (gp == NULL) { 2678 gp = g_mirror_create(mp, &md); 2679 if (gp == NULL) { 2680 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2681 md.md_name); 2682 return (NULL); 2683 } 2684 sc = gp->softc; 2685 } 2686 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2687 error = g_mirror_add_disk(sc, pp, &md); 2688 if (error != 0) { 2689 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2690 pp->name, gp->name, error); 2691 if (LIST_EMPTY(&sc->sc_disks)) 2692 g_mirror_destroy(sc, 1); 2693 return (NULL); 2694 } 2695 return (gp); 2696 } 2697 2698 static int 2699 g_mirror_destroy_geom(struct gctl_req *req __unused, 2700 struct g_class *mp __unused, struct g_geom *gp) 2701 { 2702 2703 return (g_mirror_destroy(gp->softc, 0)); 2704 } 2705 2706 static void 2707 g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2708 struct g_consumer *cp, struct g_provider *pp) 2709 { 2710 struct g_mirror_softc *sc; 2711 2712 g_topology_assert(); 2713 2714 sc = gp->softc; 2715 if (sc == NULL) 2716 return; 2717 /* Skip synchronization geom. */ 2718 if (gp == sc->sc_sync.ds_geom) 2719 return; 2720 if (pp != NULL) { 2721 /* Nothing here. */ 2722 } else if (cp != NULL) { 2723 struct g_mirror_disk *disk; 2724 2725 disk = cp->private; 2726 if (disk == NULL) 2727 return; 2728 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2729 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2730 sbuf_printf(sb, "%s<Synchronized>", indent); 2731 if (disk->d_sync.ds_offset_done == 0) 2732 sbuf_printf(sb, "0%%"); 2733 else { 2734 sbuf_printf(sb, "%u%%", 2735 (u_int)((disk->d_sync.ds_offset_done * 100) / 2736 sc->sc_provider->mediasize)); 2737 } 2738 sbuf_printf(sb, "</Synchronized>\n"); 2739 } 2740 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2741 disk->d_sync.ds_syncid); 2742 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 2743 disk->d_genid); 2744 sbuf_printf(sb, "%s<Flags>", indent); 2745 if (disk->d_flags == 0) 2746 sbuf_printf(sb, "NONE"); 2747 else { 2748 int first = 1; 2749 2750 #define ADD_FLAG(flag, name) do { \ 2751 if ((disk->d_flags & (flag)) != 0) { \ 2752 if (!first) \ 2753 sbuf_printf(sb, ", "); \ 2754 else \ 2755 first = 0; \ 2756 sbuf_printf(sb, name); \ 2757 } \ 2758 } while (0) 2759 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2760 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2761 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2762 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2763 "SYNCHRONIZING"); 2764 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2765 #undef ADD_FLAG 2766 } 2767 sbuf_printf(sb, "</Flags>\n"); 2768 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2769 disk->d_priority); 2770 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2771 g_mirror_disk_state2str(disk->d_state)); 2772 } else { 2773 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2774 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2775 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 2776 sbuf_printf(sb, "%s<Flags>", indent); 2777 if (sc->sc_flags == 0) 2778 sbuf_printf(sb, "NONE"); 2779 else { 2780 int first = 1; 2781 2782 #define ADD_FLAG(flag, name) do { \ 2783 if ((sc->sc_flags & (flag)) != 0) { \ 2784 if (!first) \ 2785 sbuf_printf(sb, ", "); \ 2786 else \ 2787 first = 0; \ 2788 sbuf_printf(sb, name); \ 2789 } \ 2790 } while (0) 2791 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2792 #undef ADD_FLAG 2793 } 2794 sbuf_printf(sb, "</Flags>\n"); 2795 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2796 (u_int)sc->sc_slice); 2797 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2798 balance_name(sc->sc_balance)); 2799 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2800 sc->sc_ndisks); 2801 sbuf_printf(sb, "%s<State>", indent); 2802 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2803 sbuf_printf(sb, "%s", "STARTING"); 2804 else if (sc->sc_ndisks == 2805 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 2806 sbuf_printf(sb, "%s", "COMPLETE"); 2807 else 2808 sbuf_printf(sb, "%s", "DEGRADED"); 2809 sbuf_printf(sb, "</State>\n"); 2810 } 2811 } 2812 2813 static void 2814 g_mirror_shutdown(void *arg, int howto) 2815 { 2816 struct g_class *mp; 2817 struct g_geom *gp, *gp2; 2818 2819 mp = arg; 2820 DROP_GIANT(); 2821 g_topology_lock(); 2822 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2823 if (gp->softc == NULL) 2824 continue; 2825 g_mirror_destroy(gp->softc, 1); 2826 } 2827 g_topology_unlock(); 2828 PICKUP_GIANT(); 2829 #if 0 2830 tsleep(&gp, PRIBIO, "m:shutdown", hz * 20); 2831 #endif 2832 } 2833 2834 static void 2835 g_mirror_init(struct g_class *mp) 2836 { 2837 2838 g_mirror_ehtag = EVENTHANDLER_REGISTER(shutdown_post_sync, 2839 g_mirror_shutdown, mp, SHUTDOWN_PRI_FIRST); 2840 if (g_mirror_ehtag == NULL) 2841 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 2842 } 2843 2844 static void 2845 g_mirror_fini(struct g_class *mp) 2846 { 2847 2848 if (g_mirror_ehtag == NULL) 2849 return; 2850 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_ehtag); 2851 } 2852 2853 static int 2854 g_mirror_can_go(void) 2855 { 2856 struct g_mirror_softc *sc; 2857 struct g_geom *gp; 2858 struct g_provider *pp; 2859 int can_go; 2860 2861 DROP_GIANT(); 2862 can_go = 1; 2863 g_topology_lock(); 2864 LIST_FOREACH(gp, &g_mirror_class.geom, geom) { 2865 sc = gp->softc; 2866 if (sc == NULL) { 2867 can_go = 0; 2868 break; 2869 } 2870 pp = sc->sc_provider; 2871 if (pp == NULL || pp->error != 0) { 2872 can_go = 0; 2873 break; 2874 } 2875 } 2876 g_topology_unlock(); 2877 PICKUP_GIANT(); 2878 return (can_go); 2879 } 2880 2881 static void 2882 g_mirror_rootwait(void) 2883 { 2884 2885 /* 2886 * HACK: Wait for GEOM, because g_mirror_rootwait() can be called, 2887 * HACK: before we get providers for tasting. 2888 */ 2889 tsleep(&g_mirror_class, PRIBIO, "mroot", hz * 3); 2890 /* 2891 * Wait for mirrors in degraded state. 2892 */ 2893 for (;;) { 2894 if (g_mirror_can_go()) 2895 break; 2896 tsleep(&g_mirror_class, PRIBIO, "mroot", hz); 2897 } 2898 } 2899 2900 SYSINIT(g_mirror_root, SI_SUB_RAID, SI_ORDER_FIRST, g_mirror_rootwait, NULL) 2901 2902 DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2903