1 /*- 2 * Copyright (c) 2004-2005 Pawel Jakub Dawidek <pjd@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 #include <sys/cdefs.h> 28 __FBSDID("$FreeBSD$"); 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 struct g_consumer *cp; 379 int error; 380 381 g_topology_assert(); 382 KASSERT(disk->d_consumer == NULL, 383 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 384 385 cp = g_new_consumer(disk->d_softc->sc_geom); 386 error = g_attach(cp, pp); 387 if (error != 0) { 388 g_destroy_consumer(cp); 389 return (error); 390 } 391 error = g_access(cp, 1, 1, 1); 392 if (error != 0) { 393 g_detach(cp); 394 g_destroy_consumer(cp); 395 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 396 pp->name, error); 397 return (error); 398 } 399 disk->d_consumer = cp; 400 disk->d_consumer->private = disk; 401 disk->d_consumer->index = 0; 402 403 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 404 return (0); 405 } 406 407 static void 408 g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 409 { 410 411 g_topology_assert(); 412 413 if (cp == NULL) 414 return; 415 if (cp->provider != NULL) 416 g_mirror_kill_consumer(sc, cp); 417 else 418 g_destroy_consumer(cp); 419 } 420 421 /* 422 * Initialize disk. This means allocate memory, create consumer, attach it 423 * to the provider and open access (r1w1e1) to it. 424 */ 425 static struct g_mirror_disk * 426 g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 427 struct g_mirror_metadata *md, int *errorp) 428 { 429 struct g_mirror_disk *disk; 430 int error; 431 432 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 433 if (disk == NULL) { 434 error = ENOMEM; 435 goto fail; 436 } 437 disk->d_softc = sc; 438 error = g_mirror_connect_disk(disk, pp); 439 if (error != 0) 440 goto fail; 441 disk->d_id = md->md_did; 442 disk->d_state = G_MIRROR_DISK_STATE_NONE; 443 disk->d_priority = md->md_priority; 444 disk->d_delay.sec = 0; 445 disk->d_delay.frac = 0; 446 binuptime(&disk->d_last_used); 447 disk->d_flags = md->md_dflags; 448 if (md->md_provider[0] != '\0') 449 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 450 disk->d_sync.ds_consumer = NULL; 451 disk->d_sync.ds_offset = md->md_sync_offset; 452 disk->d_sync.ds_offset_done = md->md_sync_offset; 453 disk->d_sync.ds_resync = -1; 454 disk->d_genid = md->md_genid; 455 disk->d_sync.ds_syncid = md->md_syncid; 456 if (errorp != NULL) 457 *errorp = 0; 458 return (disk); 459 fail: 460 if (errorp != NULL) 461 *errorp = error; 462 if (disk != NULL) 463 free(disk, M_MIRROR); 464 return (NULL); 465 } 466 467 static void 468 g_mirror_destroy_disk(struct g_mirror_disk *disk) 469 { 470 struct g_mirror_softc *sc; 471 472 g_topology_assert(); 473 474 LIST_REMOVE(disk, d_next); 475 g_mirror_event_cancel(disk); 476 sc = disk->d_softc; 477 if (sc->sc_hint == disk) 478 sc->sc_hint = NULL; 479 switch (disk->d_state) { 480 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 481 g_mirror_sync_stop(disk, 1); 482 /* FALLTHROUGH */ 483 case G_MIRROR_DISK_STATE_NEW: 484 case G_MIRROR_DISK_STATE_STALE: 485 case G_MIRROR_DISK_STATE_ACTIVE: 486 g_mirror_disconnect_consumer(sc, disk->d_consumer); 487 free(disk, M_MIRROR); 488 break; 489 default: 490 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 491 g_mirror_get_diskname(disk), 492 g_mirror_disk_state2str(disk->d_state))); 493 } 494 } 495 496 static void 497 g_mirror_destroy_device(struct g_mirror_softc *sc) 498 { 499 struct g_mirror_disk *disk; 500 struct g_mirror_event *ep; 501 struct g_geom *gp; 502 struct g_consumer *cp, *tmpcp; 503 504 g_topology_assert(); 505 506 gp = sc->sc_geom; 507 if (sc->sc_provider != NULL) 508 g_mirror_destroy_provider(sc); 509 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 510 disk = LIST_FIRST(&sc->sc_disks)) { 511 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 512 g_mirror_update_metadata(disk); 513 g_mirror_destroy_disk(disk); 514 } 515 while ((ep = g_mirror_event_get(sc)) != NULL) { 516 g_mirror_event_remove(sc, ep); 517 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 518 g_mirror_event_free(ep); 519 else { 520 ep->e_error = ECANCELED; 521 ep->e_flags |= G_MIRROR_EVENT_DONE; 522 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 523 mtx_lock(&sc->sc_events_mtx); 524 wakeup(ep); 525 mtx_unlock(&sc->sc_events_mtx); 526 } 527 } 528 callout_drain(&sc->sc_callout); 529 gp->softc = NULL; 530 531 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 532 g_mirror_disconnect_consumer(sc, cp); 533 } 534 sc->sc_sync.ds_geom->softc = NULL; 535 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 536 mtx_destroy(&sc->sc_queue_mtx); 537 mtx_destroy(&sc->sc_events_mtx); 538 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 539 g_wither_geom(gp, ENXIO); 540 } 541 542 static void 543 g_mirror_orphan(struct g_consumer *cp) 544 { 545 struct g_mirror_disk *disk; 546 547 g_topology_assert(); 548 549 disk = cp->private; 550 if (disk == NULL) 551 return; 552 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 553 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 554 G_MIRROR_EVENT_DONTWAIT); 555 } 556 557 /* 558 * Function should return the next active disk on the list. 559 * It is possible that it will be the same disk as given. 560 * If there are no active disks on list, NULL is returned. 561 */ 562 static __inline struct g_mirror_disk * 563 g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 564 { 565 struct g_mirror_disk *dp; 566 567 for (dp = LIST_NEXT(disk, d_next); dp != disk; 568 dp = LIST_NEXT(dp, d_next)) { 569 if (dp == NULL) 570 dp = LIST_FIRST(&sc->sc_disks); 571 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 572 break; 573 } 574 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 575 return (NULL); 576 return (dp); 577 } 578 579 static struct g_mirror_disk * 580 g_mirror_get_disk(struct g_mirror_softc *sc) 581 { 582 struct g_mirror_disk *disk; 583 584 if (sc->sc_hint == NULL) { 585 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 586 if (sc->sc_hint == NULL) 587 return (NULL); 588 } 589 disk = sc->sc_hint; 590 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 591 disk = g_mirror_find_next(sc, disk); 592 if (disk == NULL) 593 return (NULL); 594 } 595 sc->sc_hint = g_mirror_find_next(sc, disk); 596 return (disk); 597 } 598 599 static int 600 g_mirror_write_metadata(struct g_mirror_disk *disk, 601 struct g_mirror_metadata *md) 602 { 603 struct g_mirror_softc *sc; 604 struct g_consumer *cp; 605 off_t offset, length; 606 u_char *sector; 607 int error = 0; 608 609 g_topology_assert(); 610 611 sc = disk->d_softc; 612 cp = disk->d_consumer; 613 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 614 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 615 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 616 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 617 cp->acw, cp->ace)); 618 length = cp->provider->sectorsize; 619 offset = cp->provider->mediasize - length; 620 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 621 if (md != NULL) 622 mirror_metadata_encode(md, sector); 623 g_topology_unlock(); 624 error = g_write_data(cp, offset, sector, length); 625 g_topology_lock(); 626 free(sector, M_MIRROR); 627 if (error != 0) { 628 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_GENID; 629 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 630 G_MIRROR_EVENT_DONTWAIT); 631 } 632 return (error); 633 } 634 635 static int 636 g_mirror_clear_metadata(struct g_mirror_disk *disk) 637 { 638 int error; 639 640 g_topology_assert(); 641 error = g_mirror_write_metadata(disk, NULL); 642 if (error == 0) { 643 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 644 g_mirror_get_diskname(disk)); 645 } else { 646 G_MIRROR_DEBUG(0, 647 "Cannot clear metadata on disk %s (error=%d).", 648 g_mirror_get_diskname(disk), error); 649 } 650 return (error); 651 } 652 653 void 654 g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 655 struct g_mirror_metadata *md) 656 { 657 658 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 659 md->md_version = G_MIRROR_VERSION; 660 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 661 md->md_mid = sc->sc_id; 662 md->md_all = sc->sc_ndisks; 663 md->md_slice = sc->sc_slice; 664 md->md_balance = sc->sc_balance; 665 md->md_genid = sc->sc_genid; 666 md->md_mediasize = sc->sc_mediasize; 667 md->md_sectorsize = sc->sc_sectorsize; 668 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 669 bzero(md->md_provider, sizeof(md->md_provider)); 670 if (disk == NULL) { 671 md->md_did = arc4random(); 672 md->md_priority = 0; 673 md->md_syncid = 0; 674 md->md_dflags = 0; 675 md->md_sync_offset = 0; 676 md->md_provsize = 0; 677 } else { 678 md->md_did = disk->d_id; 679 md->md_priority = disk->d_priority; 680 md->md_syncid = disk->d_sync.ds_syncid; 681 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 682 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 683 md->md_sync_offset = disk->d_sync.ds_offset_done; 684 else 685 md->md_sync_offset = 0; 686 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 687 strlcpy(md->md_provider, 688 disk->d_consumer->provider->name, 689 sizeof(md->md_provider)); 690 } 691 md->md_provsize = disk->d_consumer->provider->mediasize; 692 } 693 } 694 695 void 696 g_mirror_update_metadata(struct g_mirror_disk *disk) 697 { 698 struct g_mirror_metadata md; 699 int error; 700 701 g_topology_assert(); 702 g_mirror_fill_metadata(disk->d_softc, disk, &md); 703 error = g_mirror_write_metadata(disk, &md); 704 if (error == 0) { 705 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 706 g_mirror_get_diskname(disk)); 707 } else { 708 G_MIRROR_DEBUG(0, 709 "Cannot update metadata on disk %s (error=%d).", 710 g_mirror_get_diskname(disk), error); 711 } 712 } 713 714 static void 715 g_mirror_bump_syncid(struct g_mirror_softc *sc) 716 { 717 struct g_mirror_disk *disk; 718 719 g_topology_assert(); 720 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 721 ("%s called with no active disks (device=%s).", __func__, 722 sc->sc_name)); 723 724 sc->sc_syncid++; 725 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 726 sc->sc_syncid); 727 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 728 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 729 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 730 disk->d_sync.ds_syncid = sc->sc_syncid; 731 g_mirror_update_metadata(disk); 732 } 733 } 734 } 735 736 static void 737 g_mirror_bump_genid(struct g_mirror_softc *sc) 738 { 739 struct g_mirror_disk *disk; 740 741 g_topology_assert(); 742 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 743 ("%s called with no active disks (device=%s).", __func__, 744 sc->sc_name)); 745 746 sc->sc_genid++; 747 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 748 sc->sc_genid); 749 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 750 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 751 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 752 disk->d_genid = sc->sc_genid; 753 g_mirror_update_metadata(disk); 754 } 755 } 756 } 757 758 static void 759 g_mirror_idle(struct g_mirror_softc *sc) 760 { 761 struct g_mirror_disk *disk; 762 763 if (sc->sc_provider == NULL || sc->sc_provider->acw == 0) 764 return; 765 sc->sc_idle = 1; 766 g_topology_lock(); 767 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 768 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 769 continue; 770 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 771 g_mirror_get_diskname(disk), sc->sc_name); 772 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 773 g_mirror_update_metadata(disk); 774 } 775 g_topology_unlock(); 776 } 777 778 static void 779 g_mirror_unidle(struct g_mirror_softc *sc) 780 { 781 struct g_mirror_disk *disk; 782 783 sc->sc_idle = 0; 784 g_topology_lock(); 785 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 786 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 787 continue; 788 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 789 g_mirror_get_diskname(disk), sc->sc_name); 790 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 791 g_mirror_update_metadata(disk); 792 } 793 g_topology_unlock(); 794 } 795 796 /* 797 * Return 1 if we should check if mirror is idling. 798 */ 799 static int 800 g_mirror_check_idle(struct g_mirror_softc *sc) 801 { 802 struct g_mirror_disk *disk; 803 804 if (sc->sc_idle) 805 return (0); 806 if (sc->sc_provider != NULL && sc->sc_provider->acw == 0) 807 return (0); 808 /* 809 * Check if there are no in-flight requests. 810 */ 811 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 812 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 813 continue; 814 if (disk->d_consumer->index > 0) 815 return (0); 816 } 817 return (1); 818 } 819 820 static __inline int 821 bintime_cmp(struct bintime *bt1, struct bintime *bt2) 822 { 823 824 if (bt1->sec < bt2->sec) 825 return (-1); 826 else if (bt1->sec > bt2->sec) 827 return (1); 828 if (bt1->frac < bt2->frac) 829 return (-1); 830 else if (bt1->frac > bt2->frac) 831 return (1); 832 return (0); 833 } 834 835 static void 836 g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 837 { 838 839 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 840 return; 841 binuptime(&disk->d_delay); 842 bintime_sub(&disk->d_delay, &bp->bio_t0); 843 } 844 845 static void 846 g_mirror_done(struct bio *bp) 847 { 848 struct g_mirror_softc *sc; 849 850 sc = bp->bio_from->geom->softc; 851 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 852 mtx_lock(&sc->sc_queue_mtx); 853 bioq_disksort(&sc->sc_queue, bp); 854 wakeup(sc); 855 mtx_unlock(&sc->sc_queue_mtx); 856 } 857 858 static void 859 g_mirror_regular_request(struct bio *bp) 860 { 861 struct g_mirror_softc *sc; 862 struct g_mirror_disk *disk; 863 struct bio *pbp; 864 865 g_topology_assert_not(); 866 867 bp->bio_from->index--; 868 pbp = bp->bio_parent; 869 sc = pbp->bio_to->geom->softc; 870 disk = bp->bio_from->private; 871 if (disk == NULL) { 872 g_topology_lock(); 873 g_mirror_kill_consumer(sc, bp->bio_from); 874 g_topology_unlock(); 875 } else { 876 g_mirror_update_delay(disk, bp); 877 } 878 879 pbp->bio_inbed++; 880 KASSERT(pbp->bio_inbed <= pbp->bio_children, 881 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 882 pbp->bio_children)); 883 if (bp->bio_error == 0 && pbp->bio_error == 0) { 884 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 885 g_destroy_bio(bp); 886 if (pbp->bio_children == pbp->bio_inbed) { 887 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 888 pbp->bio_completed = pbp->bio_length; 889 g_io_deliver(pbp, pbp->bio_error); 890 } 891 return; 892 } else if (bp->bio_error != 0) { 893 if (pbp->bio_error == 0) 894 pbp->bio_error = bp->bio_error; 895 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 896 bp->bio_error); 897 if (disk != NULL) { 898 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 899 g_mirror_event_send(disk, 900 G_MIRROR_DISK_STATE_DISCONNECTED, 901 G_MIRROR_EVENT_DONTWAIT); 902 } 903 switch (pbp->bio_cmd) { 904 case BIO_DELETE: 905 case BIO_WRITE: 906 pbp->bio_inbed--; 907 pbp->bio_children--; 908 break; 909 } 910 } 911 g_destroy_bio(bp); 912 913 switch (pbp->bio_cmd) { 914 case BIO_READ: 915 if (pbp->bio_children == pbp->bio_inbed) { 916 pbp->bio_error = 0; 917 mtx_lock(&sc->sc_queue_mtx); 918 bioq_disksort(&sc->sc_queue, pbp); 919 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 920 wakeup(sc); 921 mtx_unlock(&sc->sc_queue_mtx); 922 } 923 break; 924 case BIO_DELETE: 925 case BIO_WRITE: 926 if (pbp->bio_children == 0) { 927 /* 928 * All requests failed. 929 */ 930 } else if (pbp->bio_inbed < pbp->bio_children) { 931 /* Do nothing. */ 932 break; 933 } else if (pbp->bio_children == pbp->bio_inbed) { 934 /* Some requests succeeded. */ 935 pbp->bio_error = 0; 936 pbp->bio_completed = pbp->bio_length; 937 } 938 g_io_deliver(pbp, pbp->bio_error); 939 break; 940 default: 941 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 942 break; 943 } 944 } 945 946 static void 947 g_mirror_sync_done(struct bio *bp) 948 { 949 struct g_mirror_softc *sc; 950 951 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 952 sc = bp->bio_from->geom->softc; 953 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 954 mtx_lock(&sc->sc_queue_mtx); 955 bioq_disksort(&sc->sc_queue, bp); 956 wakeup(sc); 957 mtx_unlock(&sc->sc_queue_mtx); 958 } 959 960 static void 961 g_mirror_start(struct bio *bp) 962 { 963 struct g_mirror_softc *sc; 964 965 sc = bp->bio_to->geom->softc; 966 /* 967 * If sc == NULL or there are no valid disks, provider's error 968 * should be set and g_mirror_start() should not be called at all. 969 */ 970 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 971 ("Provider's error should be set (error=%d)(mirror=%s).", 972 bp->bio_to->error, bp->bio_to->name)); 973 G_MIRROR_LOGREQ(3, bp, "Request received."); 974 975 switch (bp->bio_cmd) { 976 case BIO_READ: 977 case BIO_WRITE: 978 case BIO_DELETE: 979 break; 980 case BIO_GETATTR: 981 default: 982 g_io_deliver(bp, EOPNOTSUPP); 983 return; 984 } 985 mtx_lock(&sc->sc_queue_mtx); 986 bioq_disksort(&sc->sc_queue, bp); 987 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 988 wakeup(sc); 989 mtx_unlock(&sc->sc_queue_mtx); 990 } 991 992 /* 993 * Send one synchronization request. 994 */ 995 static void 996 g_mirror_sync_one(struct g_mirror_disk *disk) 997 { 998 struct g_mirror_softc *sc; 999 struct bio *bp; 1000 1001 sc = disk->d_softc; 1002 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1003 ("Disk %s is not marked for synchronization.", 1004 g_mirror_get_diskname(disk))); 1005 1006 bp = g_new_bio(); 1007 if (bp == NULL) 1008 return; 1009 bp->bio_parent = NULL; 1010 bp->bio_cmd = BIO_READ; 1011 bp->bio_offset = disk->d_sync.ds_offset; 1012 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1013 bp->bio_cflags = 0; 1014 bp->bio_done = g_mirror_sync_done; 1015 bp->bio_data = disk->d_sync.ds_data; 1016 if (bp->bio_data == NULL) { 1017 g_destroy_bio(bp); 1018 return; 1019 } 1020 disk->d_sync.ds_offset += bp->bio_length; 1021 bp->bio_to = sc->sc_provider; 1022 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1023 disk->d_sync.ds_consumer->index++; 1024 g_io_request(bp, disk->d_sync.ds_consumer); 1025 } 1026 1027 static void 1028 g_mirror_sync_request(struct bio *bp) 1029 { 1030 struct g_mirror_softc *sc; 1031 struct g_mirror_disk *disk; 1032 1033 bp->bio_from->index--; 1034 sc = bp->bio_from->geom->softc; 1035 disk = bp->bio_from->private; 1036 if (disk == NULL) { 1037 g_topology_lock(); 1038 g_mirror_kill_consumer(sc, bp->bio_from); 1039 g_topology_unlock(); 1040 g_destroy_bio(bp); 1041 return; 1042 } 1043 1044 /* 1045 * Synchronization request. 1046 */ 1047 switch (bp->bio_cmd) { 1048 case BIO_READ: 1049 { 1050 struct g_consumer *cp; 1051 1052 if (bp->bio_error != 0) { 1053 G_MIRROR_LOGREQ(0, bp, 1054 "Synchronization request failed (error=%d).", 1055 bp->bio_error); 1056 g_destroy_bio(bp); 1057 return; 1058 } 1059 G_MIRROR_LOGREQ(3, bp, 1060 "Synchronization request half-finished."); 1061 bp->bio_cmd = BIO_WRITE; 1062 bp->bio_cflags = 0; 1063 cp = disk->d_consumer; 1064 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1065 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1066 cp->acr, cp->acw, cp->ace)); 1067 cp->index++; 1068 g_io_request(bp, cp); 1069 return; 1070 } 1071 case BIO_WRITE: 1072 { 1073 struct g_mirror_disk_sync *sync; 1074 1075 if (bp->bio_error != 0) { 1076 G_MIRROR_LOGREQ(0, bp, 1077 "Synchronization request failed (error=%d).", 1078 bp->bio_error); 1079 g_destroy_bio(bp); 1080 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1081 g_mirror_event_send(disk, 1082 G_MIRROR_DISK_STATE_DISCONNECTED, 1083 G_MIRROR_EVENT_DONTWAIT); 1084 return; 1085 } 1086 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1087 sync = &disk->d_sync; 1088 sync->ds_offset_done = bp->bio_offset + bp->bio_length; 1089 g_destroy_bio(bp); 1090 if (sync->ds_resync != -1) 1091 break; 1092 if (sync->ds_offset_done == sc->sc_provider->mediasize) { 1093 /* 1094 * Disk up-to-date, activate it. 1095 */ 1096 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1097 G_MIRROR_EVENT_DONTWAIT); 1098 return; 1099 } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) { 1100 /* 1101 * Update offset_done on every 100 blocks. 1102 * XXX: This should be configurable. 1103 */ 1104 g_topology_lock(); 1105 g_mirror_update_metadata(disk); 1106 g_topology_unlock(); 1107 } 1108 return; 1109 } 1110 default: 1111 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1112 bp->bio_cmd, sc->sc_name)); 1113 break; 1114 } 1115 } 1116 1117 static void 1118 g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1119 { 1120 struct g_mirror_disk *disk; 1121 struct g_consumer *cp; 1122 struct bio *cbp; 1123 1124 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1125 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1126 break; 1127 } 1128 if (disk == NULL) { 1129 if (bp->bio_error == 0) 1130 bp->bio_error = ENXIO; 1131 g_io_deliver(bp, bp->bio_error); 1132 return; 1133 } 1134 cbp = g_clone_bio(bp); 1135 if (cbp == NULL) { 1136 if (bp->bio_error == 0) 1137 bp->bio_error = ENOMEM; 1138 g_io_deliver(bp, bp->bio_error); 1139 return; 1140 } 1141 /* 1142 * Fill in the component buf structure. 1143 */ 1144 cp = disk->d_consumer; 1145 cbp->bio_done = g_mirror_done; 1146 cbp->bio_to = cp->provider; 1147 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1148 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1149 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1150 cp->acw, cp->ace)); 1151 cp->index++; 1152 g_io_request(cbp, cp); 1153 } 1154 1155 static void 1156 g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1157 { 1158 struct g_mirror_disk *disk; 1159 struct g_consumer *cp; 1160 struct bio *cbp; 1161 1162 disk = g_mirror_get_disk(sc); 1163 if (disk == NULL) { 1164 if (bp->bio_error == 0) 1165 bp->bio_error = ENXIO; 1166 g_io_deliver(bp, bp->bio_error); 1167 return; 1168 } 1169 cbp = g_clone_bio(bp); 1170 if (cbp == NULL) { 1171 if (bp->bio_error == 0) 1172 bp->bio_error = ENOMEM; 1173 g_io_deliver(bp, bp->bio_error); 1174 return; 1175 } 1176 /* 1177 * Fill in the component buf structure. 1178 */ 1179 cp = disk->d_consumer; 1180 cbp->bio_done = g_mirror_done; 1181 cbp->bio_to = cp->provider; 1182 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1183 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1184 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1185 cp->acw, cp->ace)); 1186 cp->index++; 1187 g_io_request(cbp, cp); 1188 } 1189 1190 static void 1191 g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1192 { 1193 struct g_mirror_disk *disk, *dp; 1194 struct g_consumer *cp; 1195 struct bio *cbp; 1196 struct bintime curtime; 1197 1198 binuptime(&curtime); 1199 /* 1200 * Find a disk which the smallest load. 1201 */ 1202 disk = NULL; 1203 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1204 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1205 continue; 1206 /* If disk wasn't used for more than 2 sec, use it. */ 1207 if (curtime.sec - dp->d_last_used.sec >= 2) { 1208 disk = dp; 1209 break; 1210 } 1211 if (disk == NULL || 1212 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1213 disk = dp; 1214 } 1215 } 1216 cbp = g_clone_bio(bp); 1217 if (cbp == NULL) { 1218 if (bp->bio_error == 0) 1219 bp->bio_error = ENOMEM; 1220 g_io_deliver(bp, bp->bio_error); 1221 return; 1222 } 1223 /* 1224 * Fill in the component buf structure. 1225 */ 1226 cp = disk->d_consumer; 1227 cbp->bio_done = g_mirror_done; 1228 cbp->bio_to = cp->provider; 1229 binuptime(&disk->d_last_used); 1230 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1231 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1232 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1233 cp->acw, cp->ace)); 1234 cp->index++; 1235 g_io_request(cbp, cp); 1236 } 1237 1238 static void 1239 g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1240 { 1241 struct bio_queue_head queue; 1242 struct g_mirror_disk *disk; 1243 struct g_consumer *cp; 1244 struct bio *cbp; 1245 off_t left, mod, offset, slice; 1246 u_char *data; 1247 u_int ndisks; 1248 1249 if (bp->bio_length <= sc->sc_slice) { 1250 g_mirror_request_round_robin(sc, bp); 1251 return; 1252 } 1253 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1254 slice = bp->bio_length / ndisks; 1255 mod = slice % sc->sc_provider->sectorsize; 1256 if (mod != 0) 1257 slice += sc->sc_provider->sectorsize - mod; 1258 /* 1259 * Allocate all bios before sending any request, so we can 1260 * return ENOMEM in nice and clean way. 1261 */ 1262 left = bp->bio_length; 1263 offset = bp->bio_offset; 1264 data = bp->bio_data; 1265 bioq_init(&queue); 1266 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1267 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1268 continue; 1269 cbp = g_clone_bio(bp); 1270 if (cbp == NULL) { 1271 for (cbp = bioq_first(&queue); cbp != NULL; 1272 cbp = bioq_first(&queue)) { 1273 bioq_remove(&queue, cbp); 1274 g_destroy_bio(cbp); 1275 } 1276 if (bp->bio_error == 0) 1277 bp->bio_error = ENOMEM; 1278 g_io_deliver(bp, bp->bio_error); 1279 return; 1280 } 1281 bioq_insert_tail(&queue, cbp); 1282 cbp->bio_done = g_mirror_done; 1283 cbp->bio_caller1 = disk; 1284 cbp->bio_to = disk->d_consumer->provider; 1285 cbp->bio_offset = offset; 1286 cbp->bio_data = data; 1287 cbp->bio_length = MIN(left, slice); 1288 left -= cbp->bio_length; 1289 if (left == 0) 1290 break; 1291 offset += cbp->bio_length; 1292 data += cbp->bio_length; 1293 } 1294 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1295 bioq_remove(&queue, cbp); 1296 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1297 disk = cbp->bio_caller1; 1298 cbp->bio_caller1 = NULL; 1299 cp = disk->d_consumer; 1300 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1301 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1302 cp->acr, cp->acw, cp->ace)); 1303 disk->d_consumer->index++; 1304 g_io_request(cbp, disk->d_consumer); 1305 } 1306 } 1307 1308 static void 1309 g_mirror_register_request(struct bio *bp) 1310 { 1311 struct g_mirror_softc *sc; 1312 1313 sc = bp->bio_to->geom->softc; 1314 switch (bp->bio_cmd) { 1315 case BIO_READ: 1316 switch (sc->sc_balance) { 1317 case G_MIRROR_BALANCE_LOAD: 1318 g_mirror_request_load(sc, bp); 1319 break; 1320 case G_MIRROR_BALANCE_PREFER: 1321 g_mirror_request_prefer(sc, bp); 1322 break; 1323 case G_MIRROR_BALANCE_ROUND_ROBIN: 1324 g_mirror_request_round_robin(sc, bp); 1325 break; 1326 case G_MIRROR_BALANCE_SPLIT: 1327 g_mirror_request_split(sc, bp); 1328 break; 1329 } 1330 return; 1331 case BIO_WRITE: 1332 case BIO_DELETE: 1333 { 1334 struct g_mirror_disk *disk; 1335 struct g_mirror_disk_sync *sync; 1336 struct bio_queue_head queue; 1337 struct g_consumer *cp; 1338 struct bio *cbp; 1339 1340 if (sc->sc_idle) 1341 g_mirror_unidle(sc); 1342 /* 1343 * Allocate all bios before sending any request, so we can 1344 * return ENOMEM in nice and clean way. 1345 */ 1346 bioq_init(&queue); 1347 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1348 sync = &disk->d_sync; 1349 switch (disk->d_state) { 1350 case G_MIRROR_DISK_STATE_ACTIVE: 1351 break; 1352 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1353 if (bp->bio_offset >= sync->ds_offset) 1354 continue; 1355 else if (bp->bio_offset + bp->bio_length > 1356 sync->ds_offset_done && 1357 (bp->bio_offset < sync->ds_resync || 1358 sync->ds_resync == -1)) { 1359 sync->ds_resync = bp->bio_offset - 1360 (bp->bio_offset % MAXPHYS); 1361 } 1362 break; 1363 default: 1364 continue; 1365 } 1366 cbp = g_clone_bio(bp); 1367 if (cbp == NULL) { 1368 for (cbp = bioq_first(&queue); cbp != NULL; 1369 cbp = bioq_first(&queue)) { 1370 bioq_remove(&queue, cbp); 1371 g_destroy_bio(cbp); 1372 } 1373 if (bp->bio_error == 0) 1374 bp->bio_error = ENOMEM; 1375 g_io_deliver(bp, bp->bio_error); 1376 return; 1377 } 1378 bioq_insert_tail(&queue, cbp); 1379 cbp->bio_done = g_mirror_done; 1380 cp = disk->d_consumer; 1381 cbp->bio_caller1 = cp; 1382 cbp->bio_to = cp->provider; 1383 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1384 ("Consumer %s not opened (r%dw%de%d).", 1385 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1386 } 1387 for (cbp = bioq_first(&queue); cbp != NULL; 1388 cbp = bioq_first(&queue)) { 1389 bioq_remove(&queue, cbp); 1390 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1391 cp = cbp->bio_caller1; 1392 cbp->bio_caller1 = NULL; 1393 cp->index++; 1394 g_io_request(cbp, cp); 1395 } 1396 /* 1397 * Bump syncid on first write. 1398 */ 1399 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1400 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1401 g_topology_lock(); 1402 g_mirror_bump_syncid(sc); 1403 g_topology_unlock(); 1404 } 1405 return; 1406 } 1407 default: 1408 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1409 bp->bio_cmd, sc->sc_name)); 1410 break; 1411 } 1412 } 1413 1414 static int 1415 g_mirror_can_destroy(struct g_mirror_softc *sc) 1416 { 1417 struct g_geom *gp; 1418 struct g_consumer *cp; 1419 1420 g_topology_assert(); 1421 gp = sc->sc_geom; 1422 LIST_FOREACH(cp, &gp->consumer, consumer) { 1423 if (g_mirror_is_busy(sc, cp)) 1424 return (0); 1425 } 1426 gp = sc->sc_sync.ds_geom; 1427 LIST_FOREACH(cp, &gp->consumer, consumer) { 1428 if (g_mirror_is_busy(sc, cp)) 1429 return (0); 1430 } 1431 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1432 sc->sc_name); 1433 return (1); 1434 } 1435 1436 static int 1437 g_mirror_try_destroy(struct g_mirror_softc *sc) 1438 { 1439 1440 g_topology_lock(); 1441 if (!g_mirror_can_destroy(sc)) { 1442 g_topology_unlock(); 1443 return (0); 1444 } 1445 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1446 g_topology_unlock(); 1447 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1448 &sc->sc_worker); 1449 wakeup(&sc->sc_worker); 1450 sc->sc_worker = NULL; 1451 } else { 1452 g_mirror_destroy_device(sc); 1453 g_topology_unlock(); 1454 free(sc, M_MIRROR); 1455 } 1456 return (1); 1457 } 1458 1459 /* 1460 * Worker thread. 1461 */ 1462 static void 1463 g_mirror_worker(void *arg) 1464 { 1465 struct g_mirror_softc *sc; 1466 struct g_mirror_disk *disk; 1467 struct g_mirror_disk_sync *sync; 1468 struct g_mirror_event *ep; 1469 struct bio *bp; 1470 u_int nreqs; 1471 1472 sc = arg; 1473 mtx_lock_spin(&sched_lock); 1474 sched_prio(curthread, PRIBIO); 1475 mtx_unlock_spin(&sched_lock); 1476 1477 nreqs = 0; 1478 for (;;) { 1479 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1480 /* 1481 * First take a look at events. 1482 * This is important to handle events before any I/O requests. 1483 */ 1484 ep = g_mirror_event_get(sc); 1485 if (ep != NULL && g_topology_try_lock()) { 1486 g_mirror_event_remove(sc, ep); 1487 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1488 /* Update only device status. */ 1489 G_MIRROR_DEBUG(3, 1490 "Running event for device %s.", 1491 sc->sc_name); 1492 ep->e_error = 0; 1493 g_mirror_update_device(sc, 1); 1494 } else { 1495 /* Update disk status. */ 1496 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1497 g_mirror_get_diskname(ep->e_disk)); 1498 ep->e_error = g_mirror_update_disk(ep->e_disk, 1499 ep->e_state); 1500 if (ep->e_error == 0) 1501 g_mirror_update_device(sc, 0); 1502 } 1503 g_topology_unlock(); 1504 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1505 KASSERT(ep->e_error == 0, 1506 ("Error cannot be handled.")); 1507 g_mirror_event_free(ep); 1508 } else { 1509 ep->e_flags |= G_MIRROR_EVENT_DONE; 1510 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1511 ep); 1512 mtx_lock(&sc->sc_events_mtx); 1513 wakeup(ep); 1514 mtx_unlock(&sc->sc_events_mtx); 1515 } 1516 if ((sc->sc_flags & 1517 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1518 if (g_mirror_try_destroy(sc)) 1519 kthread_exit(0); 1520 } 1521 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1522 continue; 1523 } 1524 /* 1525 * Now I/O requests. 1526 */ 1527 /* Get first request from the queue. */ 1528 mtx_lock(&sc->sc_queue_mtx); 1529 bp = bioq_first(&sc->sc_queue); 1530 if (bp == NULL) { 1531 if (ep != NULL) { 1532 /* 1533 * No I/O requests and topology lock was 1534 * already held? Try again. 1535 */ 1536 mtx_unlock(&sc->sc_queue_mtx); 1537 continue; 1538 } 1539 if ((sc->sc_flags & 1540 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1541 mtx_unlock(&sc->sc_queue_mtx); 1542 if (g_mirror_try_destroy(sc)) 1543 kthread_exit(0); 1544 mtx_lock(&sc->sc_queue_mtx); 1545 } 1546 } 1547 if (sc->sc_sync.ds_ndisks > 0 && 1548 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1549 mtx_unlock(&sc->sc_queue_mtx); 1550 /* 1551 * It is time for synchronization... 1552 */ 1553 nreqs = 0; 1554 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1555 if (disk->d_state != 1556 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1557 continue; 1558 } 1559 sync = &disk->d_sync; 1560 if (sync->ds_offset >= 1561 sc->sc_provider->mediasize) { 1562 continue; 1563 } 1564 if (sync->ds_offset > sync->ds_offset_done) 1565 continue; 1566 if (sync->ds_resync != -1) { 1567 sync->ds_offset = sync->ds_resync; 1568 sync->ds_offset_done = sync->ds_resync; 1569 sync->ds_resync = -1; 1570 } 1571 g_mirror_sync_one(disk); 1572 } 1573 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1574 goto sleep; 1575 } 1576 if (bp == NULL) { 1577 if (g_mirror_check_idle(sc)) { 1578 u_int idletime; 1579 1580 idletime = g_mirror_idletime; 1581 if (idletime == 0) 1582 idletime = 1; 1583 idletime *= hz; 1584 if (msleep(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1585 "m:w1", idletime) == EWOULDBLOCK) { 1586 G_MIRROR_DEBUG(5, "%s: I'm here 3.", 1587 __func__); 1588 /* 1589 * No I/O requests in 'idletime' seconds, 1590 * so mark components as clean. 1591 */ 1592 g_mirror_idle(sc); 1593 } 1594 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1595 } else { 1596 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1597 "m:w2", 0); 1598 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1599 } 1600 continue; 1601 } 1602 nreqs++; 1603 bioq_remove(&sc->sc_queue, bp); 1604 mtx_unlock(&sc->sc_queue_mtx); 1605 1606 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1607 g_mirror_regular_request(bp); 1608 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1609 u_int timeout, sps; 1610 1611 g_mirror_sync_request(bp); 1612 sleep: 1613 sps = g_mirror_syncs_per_sec; 1614 if (sps == 0) { 1615 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1616 continue; 1617 } 1618 if (ep != NULL) { 1619 /* 1620 * We have some pending events, don't sleep now. 1621 */ 1622 G_MIRROR_DEBUG(5, "%s: I'm here 7.", __func__); 1623 continue; 1624 } 1625 mtx_lock(&sc->sc_queue_mtx); 1626 if (bioq_first(&sc->sc_queue) != NULL) { 1627 mtx_unlock(&sc->sc_queue_mtx); 1628 G_MIRROR_DEBUG(5, "%s: I'm here 8.", __func__); 1629 continue; 1630 } 1631 timeout = hz / sps; 1632 if (timeout == 0) 1633 timeout = 1; 1634 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w3", 1635 timeout); 1636 } else { 1637 g_mirror_register_request(bp); 1638 } 1639 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1640 } 1641 } 1642 1643 /* 1644 * Open disk's consumer if needed. 1645 */ 1646 static void 1647 g_mirror_update_access(struct g_mirror_disk *disk) 1648 { 1649 struct g_provider *pp; 1650 1651 g_topology_assert(); 1652 1653 pp = disk->d_softc->sc_provider; 1654 if (pp == NULL) 1655 return; 1656 if (pp->acw > 0) { 1657 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1658 G_MIRROR_DEBUG(1, 1659 "Disk %s (device %s) marked as dirty.", 1660 g_mirror_get_diskname(disk), 1661 disk->d_softc->sc_name); 1662 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1663 } 1664 } else if (pp->acw == 0) { 1665 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1666 G_MIRROR_DEBUG(1, 1667 "Disk %s (device %s) marked as clean.", 1668 g_mirror_get_diskname(disk), 1669 disk->d_softc->sc_name); 1670 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1671 } 1672 } 1673 } 1674 1675 static void 1676 g_mirror_sync_start(struct g_mirror_disk *disk) 1677 { 1678 struct g_mirror_softc *sc; 1679 int error; 1680 1681 g_topology_assert(); 1682 1683 sc = disk->d_softc; 1684 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1685 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1686 sc->sc_state)); 1687 1688 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1689 g_mirror_get_diskname(disk)); 1690 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1691 KASSERT(disk->d_sync.ds_consumer == NULL, 1692 ("Sync consumer already exists (device=%s, disk=%s).", 1693 sc->sc_name, g_mirror_get_diskname(disk))); 1694 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1695 disk->d_sync.ds_consumer->private = disk; 1696 disk->d_sync.ds_consumer->index = 0; 1697 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1698 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1699 disk->d_softc->sc_name, error)); 1700 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1701 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1702 disk->d_softc->sc_name, error)); 1703 disk->d_sync.ds_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1704 sc->sc_sync.ds_ndisks++; 1705 } 1706 1707 /* 1708 * Stop synchronization process. 1709 * type: 0 - synchronization finished 1710 * 1 - synchronization stopped 1711 */ 1712 static void 1713 g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1714 { 1715 1716 g_topology_assert(); 1717 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1718 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1719 g_mirror_disk_state2str(disk->d_state))); 1720 if (disk->d_sync.ds_consumer == NULL) 1721 return; 1722 1723 if (type == 0) { 1724 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1725 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1726 } else /* if (type == 1) */ { 1727 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1728 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1729 } 1730 g_mirror_kill_consumer(disk->d_softc, disk->d_sync.ds_consumer); 1731 free(disk->d_sync.ds_data, M_MIRROR); 1732 disk->d_sync.ds_consumer = NULL; 1733 disk->d_softc->sc_sync.ds_ndisks--; 1734 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1735 } 1736 1737 static void 1738 g_mirror_launch_provider(struct g_mirror_softc *sc) 1739 { 1740 struct g_mirror_disk *disk; 1741 struct g_provider *pp; 1742 1743 g_topology_assert(); 1744 1745 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1746 pp->mediasize = sc->sc_mediasize; 1747 pp->sectorsize = sc->sc_sectorsize; 1748 sc->sc_provider = pp; 1749 g_error_provider(pp, 0); 1750 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1751 pp->name); 1752 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1753 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1754 g_mirror_sync_start(disk); 1755 } 1756 } 1757 1758 static void 1759 g_mirror_destroy_provider(struct g_mirror_softc *sc) 1760 { 1761 struct g_mirror_disk *disk; 1762 struct bio *bp; 1763 1764 g_topology_assert(); 1765 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1766 sc->sc_name)); 1767 1768 g_error_provider(sc->sc_provider, ENXIO); 1769 mtx_lock(&sc->sc_queue_mtx); 1770 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1771 bioq_remove(&sc->sc_queue, bp); 1772 g_io_deliver(bp, ENXIO); 1773 } 1774 mtx_unlock(&sc->sc_queue_mtx); 1775 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1776 sc->sc_provider->name); 1777 sc->sc_provider->flags |= G_PF_WITHER; 1778 g_orphan_provider(sc->sc_provider, ENXIO); 1779 sc->sc_provider = NULL; 1780 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1781 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1782 g_mirror_sync_stop(disk, 1); 1783 } 1784 } 1785 1786 static void 1787 g_mirror_go(void *arg) 1788 { 1789 struct g_mirror_softc *sc; 1790 1791 sc = arg; 1792 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1793 g_mirror_event_send(sc, 0, 1794 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1795 } 1796 1797 static u_int 1798 g_mirror_determine_state(struct g_mirror_disk *disk) 1799 { 1800 struct g_mirror_softc *sc; 1801 u_int state; 1802 1803 sc = disk->d_softc; 1804 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1805 if ((disk->d_flags & 1806 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1807 /* Disk does not need synchronization. */ 1808 state = G_MIRROR_DISK_STATE_ACTIVE; 1809 } else { 1810 if ((sc->sc_flags & 1811 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1812 (disk->d_flags & 1813 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1814 /* 1815 * We can start synchronization from 1816 * the stored offset. 1817 */ 1818 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1819 } else { 1820 state = G_MIRROR_DISK_STATE_STALE; 1821 } 1822 } 1823 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1824 /* 1825 * Reset all synchronization data for this disk, 1826 * because if it even was synchronized, it was 1827 * synchronized to disks with different syncid. 1828 */ 1829 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1830 disk->d_sync.ds_offset = 0; 1831 disk->d_sync.ds_offset_done = 0; 1832 disk->d_sync.ds_syncid = sc->sc_syncid; 1833 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1834 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1835 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1836 } else { 1837 state = G_MIRROR_DISK_STATE_STALE; 1838 } 1839 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1840 /* 1841 * Not good, NOT GOOD! 1842 * It means that mirror was started on stale disks 1843 * and more fresh disk just arrive. 1844 * If there were writes, mirror is fucked up, sorry. 1845 * I think the best choice here is don't touch 1846 * this disk and inform the user laudly. 1847 */ 1848 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1849 "disk (%s) arrives!! It will not be connected to the " 1850 "running device.", sc->sc_name, 1851 g_mirror_get_diskname(disk)); 1852 g_mirror_destroy_disk(disk); 1853 state = G_MIRROR_DISK_STATE_NONE; 1854 /* Return immediately, because disk was destroyed. */ 1855 return (state); 1856 } 1857 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1858 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1859 return (state); 1860 } 1861 1862 /* 1863 * Update device state. 1864 */ 1865 static void 1866 g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1867 { 1868 struct g_mirror_disk *disk; 1869 u_int state; 1870 1871 g_topology_assert(); 1872 1873 switch (sc->sc_state) { 1874 case G_MIRROR_DEVICE_STATE_STARTING: 1875 { 1876 struct g_mirror_disk *pdisk, *tdisk; 1877 u_int dirty, ndisks, genid, syncid; 1878 1879 KASSERT(sc->sc_provider == NULL, 1880 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1881 /* 1882 * Are we ready? We are, if all disks are connected or 1883 * if we have any disks and 'force' is true. 1884 */ 1885 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1886 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1887 ; 1888 } else if (g_mirror_ndisks(sc, -1) == 0) { 1889 /* 1890 * Disks went down in starting phase, so destroy 1891 * device. 1892 */ 1893 callout_drain(&sc->sc_callout); 1894 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1895 return; 1896 } else { 1897 return; 1898 } 1899 1900 /* 1901 * Activate all disks with the biggest syncid. 1902 */ 1903 if (force) { 1904 /* 1905 * If 'force' is true, we have been called due to 1906 * timeout, so don't bother canceling timeout. 1907 */ 1908 ndisks = 0; 1909 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1910 if ((disk->d_flags & 1911 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1912 ndisks++; 1913 } 1914 } 1915 if (ndisks == 0) { 1916 /* No valid disks found, destroy device. */ 1917 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1918 return; 1919 } 1920 } else { 1921 /* Cancel timeout. */ 1922 callout_drain(&sc->sc_callout); 1923 } 1924 1925 /* 1926 * Find the biggest genid. 1927 */ 1928 genid = 0; 1929 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1930 if (disk->d_genid > genid) 1931 genid = disk->d_genid; 1932 } 1933 sc->sc_genid = genid; 1934 /* 1935 * Remove all disks without the biggest genid. 1936 */ 1937 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 1938 if (disk->d_genid < genid) { 1939 G_MIRROR_DEBUG(0, 1940 "Component %s (device %s) broken, skipping.", 1941 g_mirror_get_diskname(disk), sc->sc_name); 1942 g_mirror_destroy_disk(disk); 1943 } 1944 } 1945 1946 /* 1947 * Find the biggest syncid. 1948 */ 1949 syncid = 0; 1950 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1951 if (disk->d_sync.ds_syncid > syncid) 1952 syncid = disk->d_sync.ds_syncid; 1953 } 1954 1955 /* 1956 * Here we need to look for dirty disks and if all disks 1957 * with the biggest syncid are dirty, we have to choose 1958 * one with the biggest priority and rebuild the rest. 1959 */ 1960 /* 1961 * Find the number of dirty disks with the biggest syncid. 1962 * Find the number of disks with the biggest syncid. 1963 * While here, find a disk with the biggest priority. 1964 */ 1965 dirty = ndisks = 0; 1966 pdisk = NULL; 1967 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1968 if (disk->d_sync.ds_syncid != syncid) 1969 continue; 1970 if ((disk->d_flags & 1971 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1972 continue; 1973 } 1974 ndisks++; 1975 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1976 dirty++; 1977 if (pdisk == NULL || 1978 pdisk->d_priority < disk->d_priority) { 1979 pdisk = disk; 1980 } 1981 } 1982 } 1983 if (dirty == 0) { 1984 /* No dirty disks at all, great. */ 1985 } else if (dirty == ndisks) { 1986 /* 1987 * Force synchronization for all dirty disks except one 1988 * with the biggest priority. 1989 */ 1990 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1991 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1992 "master disk for synchronization.", 1993 g_mirror_get_diskname(pdisk), sc->sc_name); 1994 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1995 if (disk->d_sync.ds_syncid != syncid) 1996 continue; 1997 if ((disk->d_flags & 1998 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1999 continue; 2000 } 2001 KASSERT((disk->d_flags & 2002 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2003 ("Disk %s isn't marked as dirty.", 2004 g_mirror_get_diskname(disk))); 2005 /* Skip the disk with the biggest priority. */ 2006 if (disk == pdisk) 2007 continue; 2008 disk->d_sync.ds_syncid = 0; 2009 } 2010 } else if (dirty < ndisks) { 2011 /* 2012 * Force synchronization for all dirty disks. 2013 * We have some non-dirty disks. 2014 */ 2015 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2016 if (disk->d_sync.ds_syncid != syncid) 2017 continue; 2018 if ((disk->d_flags & 2019 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2020 continue; 2021 } 2022 if ((disk->d_flags & 2023 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2024 continue; 2025 } 2026 disk->d_sync.ds_syncid = 0; 2027 } 2028 } 2029 2030 /* Reset hint. */ 2031 sc->sc_hint = NULL; 2032 sc->sc_syncid = syncid; 2033 if (force) { 2034 /* Remember to bump syncid on first write. */ 2035 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2036 } 2037 state = G_MIRROR_DEVICE_STATE_RUNNING; 2038 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2039 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2040 g_mirror_device_state2str(state)); 2041 sc->sc_state = state; 2042 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2043 state = g_mirror_determine_state(disk); 2044 g_mirror_event_send(disk, state, 2045 G_MIRROR_EVENT_DONTWAIT); 2046 if (state == G_MIRROR_DISK_STATE_STALE) 2047 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2048 } 2049 wakeup(&g_mirror_class); 2050 break; 2051 } 2052 case G_MIRROR_DEVICE_STATE_RUNNING: 2053 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2054 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2055 /* 2056 * No active disks or no disks at all, 2057 * so destroy device. 2058 */ 2059 if (sc->sc_provider != NULL) 2060 g_mirror_destroy_provider(sc); 2061 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2062 break; 2063 } else if (g_mirror_ndisks(sc, 2064 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2065 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2066 /* 2067 * We have active disks, launch provider if it doesn't 2068 * exist. 2069 */ 2070 if (sc->sc_provider == NULL) 2071 g_mirror_launch_provider(sc); 2072 } 2073 /* 2074 * Genid should be bumped immediately, so do it here. 2075 */ 2076 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2077 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2078 g_mirror_bump_genid(sc); 2079 } 2080 break; 2081 default: 2082 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2083 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2084 break; 2085 } 2086 } 2087 2088 /* 2089 * Update disk state and device state if needed. 2090 */ 2091 #define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2092 "Disk %s state changed from %s to %s (device %s).", \ 2093 g_mirror_get_diskname(disk), \ 2094 g_mirror_disk_state2str(disk->d_state), \ 2095 g_mirror_disk_state2str(state), sc->sc_name) 2096 static int 2097 g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2098 { 2099 struct g_mirror_softc *sc; 2100 2101 g_topology_assert(); 2102 2103 sc = disk->d_softc; 2104 again: 2105 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2106 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2107 g_mirror_disk_state2str(state)); 2108 switch (state) { 2109 case G_MIRROR_DISK_STATE_NEW: 2110 /* 2111 * Possible scenarios: 2112 * 1. New disk arrive. 2113 */ 2114 /* Previous state should be NONE. */ 2115 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2116 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2117 g_mirror_disk_state2str(disk->d_state))); 2118 DISK_STATE_CHANGED(); 2119 2120 disk->d_state = state; 2121 if (LIST_EMPTY(&sc->sc_disks)) 2122 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2123 else { 2124 struct g_mirror_disk *dp; 2125 2126 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2127 if (disk->d_priority >= dp->d_priority) { 2128 LIST_INSERT_BEFORE(dp, disk, d_next); 2129 dp = NULL; 2130 break; 2131 } 2132 if (LIST_NEXT(dp, d_next) == NULL) 2133 break; 2134 } 2135 if (dp != NULL) 2136 LIST_INSERT_AFTER(dp, disk, d_next); 2137 } 2138 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2139 sc->sc_name, g_mirror_get_diskname(disk)); 2140 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2141 break; 2142 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2143 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2144 g_mirror_device_state2str(sc->sc_state), 2145 g_mirror_get_diskname(disk), 2146 g_mirror_disk_state2str(disk->d_state))); 2147 state = g_mirror_determine_state(disk); 2148 if (state != G_MIRROR_DISK_STATE_NONE) 2149 goto again; 2150 break; 2151 case G_MIRROR_DISK_STATE_ACTIVE: 2152 /* 2153 * Possible scenarios: 2154 * 1. New disk does not need synchronization. 2155 * 2. Synchronization process finished successfully. 2156 */ 2157 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2158 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2159 g_mirror_device_state2str(sc->sc_state), 2160 g_mirror_get_diskname(disk), 2161 g_mirror_disk_state2str(disk->d_state))); 2162 /* Previous state should be NEW or SYNCHRONIZING. */ 2163 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2164 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2165 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2166 g_mirror_disk_state2str(disk->d_state))); 2167 DISK_STATE_CHANGED(); 2168 2169 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2170 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2171 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2172 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2173 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2174 g_mirror_sync_stop(disk, 0); 2175 } 2176 disk->d_state = state; 2177 disk->d_sync.ds_offset = 0; 2178 disk->d_sync.ds_offset_done = 0; 2179 g_mirror_update_access(disk); 2180 g_mirror_update_metadata(disk); 2181 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2182 sc->sc_name, g_mirror_get_diskname(disk)); 2183 break; 2184 case G_MIRROR_DISK_STATE_STALE: 2185 /* 2186 * Possible scenarios: 2187 * 1. Stale disk was connected. 2188 */ 2189 /* Previous state should be NEW. */ 2190 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2191 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2192 g_mirror_disk_state2str(disk->d_state))); 2193 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2194 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2195 g_mirror_device_state2str(sc->sc_state), 2196 g_mirror_get_diskname(disk), 2197 g_mirror_disk_state2str(disk->d_state))); 2198 /* 2199 * STALE state is only possible if device is marked 2200 * NOAUTOSYNC. 2201 */ 2202 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2203 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2204 g_mirror_device_state2str(sc->sc_state), 2205 g_mirror_get_diskname(disk), 2206 g_mirror_disk_state2str(disk->d_state))); 2207 DISK_STATE_CHANGED(); 2208 2209 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2210 disk->d_state = state; 2211 g_mirror_update_metadata(disk); 2212 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2213 sc->sc_name, g_mirror_get_diskname(disk)); 2214 break; 2215 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2216 /* 2217 * Possible scenarios: 2218 * 1. Disk which needs synchronization was connected. 2219 */ 2220 /* Previous state should be NEW. */ 2221 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2222 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2223 g_mirror_disk_state2str(disk->d_state))); 2224 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2225 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2226 g_mirror_device_state2str(sc->sc_state), 2227 g_mirror_get_diskname(disk), 2228 g_mirror_disk_state2str(disk->d_state))); 2229 DISK_STATE_CHANGED(); 2230 2231 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2232 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2233 disk->d_state = state; 2234 if (sc->sc_provider != NULL) { 2235 g_mirror_sync_start(disk); 2236 g_mirror_update_metadata(disk); 2237 } 2238 break; 2239 case G_MIRROR_DISK_STATE_DISCONNECTED: 2240 /* 2241 * Possible scenarios: 2242 * 1. Device wasn't running yet, but disk disappear. 2243 * 2. Disk was active and disapppear. 2244 * 3. Disk disappear during synchronization process. 2245 */ 2246 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2247 /* 2248 * Previous state should be ACTIVE, STALE or 2249 * SYNCHRONIZING. 2250 */ 2251 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2252 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2253 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2254 ("Wrong disk state (%s, %s).", 2255 g_mirror_get_diskname(disk), 2256 g_mirror_disk_state2str(disk->d_state))); 2257 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2258 /* Previous state should be NEW. */ 2259 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2260 ("Wrong disk state (%s, %s).", 2261 g_mirror_get_diskname(disk), 2262 g_mirror_disk_state2str(disk->d_state))); 2263 /* 2264 * Reset bumping syncid if disk disappeared in STARTING 2265 * state. 2266 */ 2267 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2268 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2269 #ifdef INVARIANTS 2270 } else { 2271 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2272 sc->sc_name, 2273 g_mirror_device_state2str(sc->sc_state), 2274 g_mirror_get_diskname(disk), 2275 g_mirror_disk_state2str(disk->d_state))); 2276 #endif 2277 } 2278 DISK_STATE_CHANGED(); 2279 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2280 sc->sc_name, g_mirror_get_diskname(disk)); 2281 2282 g_mirror_destroy_disk(disk); 2283 break; 2284 case G_MIRROR_DISK_STATE_DESTROY: 2285 { 2286 int error; 2287 2288 error = g_mirror_clear_metadata(disk); 2289 if (error != 0) 2290 return (error); 2291 DISK_STATE_CHANGED(); 2292 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2293 sc->sc_name, g_mirror_get_diskname(disk)); 2294 2295 g_mirror_destroy_disk(disk); 2296 sc->sc_ndisks--; 2297 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2298 g_mirror_update_metadata(disk); 2299 } 2300 break; 2301 } 2302 default: 2303 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2304 break; 2305 } 2306 return (0); 2307 } 2308 #undef DISK_STATE_CHANGED 2309 2310 int 2311 g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2312 { 2313 struct g_provider *pp; 2314 u_char *buf; 2315 int error; 2316 2317 g_topology_assert(); 2318 2319 error = g_access(cp, 1, 0, 0); 2320 if (error != 0) 2321 return (error); 2322 pp = cp->provider; 2323 g_topology_unlock(); 2324 /* Metadata are stored on last sector. */ 2325 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2326 &error); 2327 g_topology_lock(); 2328 g_access(cp, -1, 0, 0); 2329 if (error != 0) { 2330 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2331 cp->provider->name, error); 2332 if (buf != NULL) 2333 g_free(buf); 2334 return (error); 2335 } 2336 2337 /* Decode metadata. */ 2338 error = mirror_metadata_decode(buf, md); 2339 g_free(buf); 2340 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2341 return (EINVAL); 2342 if (md->md_version > G_MIRROR_VERSION) { 2343 G_MIRROR_DEBUG(0, 2344 "Kernel module is too old to handle metadata from %s.", 2345 cp->provider->name); 2346 return (EINVAL); 2347 } 2348 if (error != 0) { 2349 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2350 cp->provider->name); 2351 return (error); 2352 } 2353 2354 return (0); 2355 } 2356 2357 static int 2358 g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2359 struct g_mirror_metadata *md) 2360 { 2361 2362 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2363 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2364 pp->name, md->md_did); 2365 return (EEXIST); 2366 } 2367 if (md->md_all != sc->sc_ndisks) { 2368 G_MIRROR_DEBUG(1, 2369 "Invalid '%s' field on disk %s (device %s), skipping.", 2370 "md_all", pp->name, sc->sc_name); 2371 return (EINVAL); 2372 } 2373 if (md->md_slice != sc->sc_slice) { 2374 G_MIRROR_DEBUG(1, 2375 "Invalid '%s' field on disk %s (device %s), skipping.", 2376 "md_slice", pp->name, sc->sc_name); 2377 return (EINVAL); 2378 } 2379 if (md->md_balance != sc->sc_balance) { 2380 G_MIRROR_DEBUG(1, 2381 "Invalid '%s' field on disk %s (device %s), skipping.", 2382 "md_balance", pp->name, sc->sc_name); 2383 return (EINVAL); 2384 } 2385 if (md->md_mediasize != sc->sc_mediasize) { 2386 G_MIRROR_DEBUG(1, 2387 "Invalid '%s' field on disk %s (device %s), skipping.", 2388 "md_mediasize", pp->name, sc->sc_name); 2389 return (EINVAL); 2390 } 2391 if (sc->sc_mediasize > pp->mediasize) { 2392 G_MIRROR_DEBUG(1, 2393 "Invalid size of disk %s (device %s), skipping.", pp->name, 2394 sc->sc_name); 2395 return (EINVAL); 2396 } 2397 if (md->md_sectorsize != sc->sc_sectorsize) { 2398 G_MIRROR_DEBUG(1, 2399 "Invalid '%s' field on disk %s (device %s), skipping.", 2400 "md_sectorsize", pp->name, sc->sc_name); 2401 return (EINVAL); 2402 } 2403 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2404 G_MIRROR_DEBUG(1, 2405 "Invalid sector size of disk %s (device %s), skipping.", 2406 pp->name, sc->sc_name); 2407 return (EINVAL); 2408 } 2409 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2410 G_MIRROR_DEBUG(1, 2411 "Invalid device flags on disk %s (device %s), skipping.", 2412 pp->name, sc->sc_name); 2413 return (EINVAL); 2414 } 2415 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2416 G_MIRROR_DEBUG(1, 2417 "Invalid disk flags on disk %s (device %s), skipping.", 2418 pp->name, sc->sc_name); 2419 return (EINVAL); 2420 } 2421 return (0); 2422 } 2423 2424 int 2425 g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2426 struct g_mirror_metadata *md) 2427 { 2428 struct g_mirror_disk *disk; 2429 int error; 2430 2431 g_topology_assert(); 2432 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2433 2434 error = g_mirror_check_metadata(sc, pp, md); 2435 if (error != 0) 2436 return (error); 2437 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2438 md->md_genid < sc->sc_genid) { 2439 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2440 pp->name, sc->sc_name); 2441 return (EINVAL); 2442 } 2443 disk = g_mirror_init_disk(sc, pp, md, &error); 2444 if (disk == NULL) 2445 return (error); 2446 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2447 G_MIRROR_EVENT_WAIT); 2448 if (error != 0) 2449 return (error); 2450 if (md->md_version < G_MIRROR_VERSION) { 2451 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2452 pp->name, md->md_version, G_MIRROR_VERSION); 2453 g_mirror_update_metadata(disk); 2454 } 2455 return (0); 2456 } 2457 2458 static int 2459 g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2460 { 2461 struct g_mirror_softc *sc; 2462 struct g_mirror_disk *disk; 2463 int dcr, dcw, dce; 2464 2465 g_topology_assert(); 2466 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2467 acw, ace); 2468 2469 dcr = pp->acr + acr; 2470 dcw = pp->acw + acw; 2471 dce = pp->ace + ace; 2472 2473 sc = pp->geom->softc; 2474 if (sc == NULL || LIST_EMPTY(&sc->sc_disks) || 2475 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 2476 if (acr <= 0 && acw <= 0 && ace <= 0) 2477 return (0); 2478 else 2479 return (ENXIO); 2480 } 2481 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2482 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2483 continue; 2484 /* 2485 * Mark disk as dirty on open and unmark on close. 2486 */ 2487 if (pp->acw == 0 && dcw > 0) { 2488 G_MIRROR_DEBUG(1, 2489 "Disk %s (device %s) marked as dirty.", 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 } else if (pp->acw > 0 && dcw == 0) { 2494 G_MIRROR_DEBUG(1, 2495 "Disk %s (device %s) marked as clean.", 2496 g_mirror_get_diskname(disk), sc->sc_name); 2497 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2498 g_mirror_update_metadata(disk); 2499 } 2500 } 2501 return (0); 2502 } 2503 2504 static struct g_geom * 2505 g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2506 { 2507 struct g_mirror_softc *sc; 2508 struct g_geom *gp; 2509 int error, timeout; 2510 2511 g_topology_assert(); 2512 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2513 md->md_mid); 2514 2515 /* One disk is minimum. */ 2516 if (md->md_all < 1) 2517 return (NULL); 2518 /* 2519 * Action geom. 2520 */ 2521 gp = g_new_geomf(mp, "%s", md->md_name); 2522 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2523 gp->start = g_mirror_start; 2524 gp->orphan = g_mirror_orphan; 2525 gp->access = g_mirror_access; 2526 gp->dumpconf = g_mirror_dumpconf; 2527 2528 sc->sc_id = md->md_mid; 2529 sc->sc_slice = md->md_slice; 2530 sc->sc_balance = md->md_balance; 2531 sc->sc_mediasize = md->md_mediasize; 2532 sc->sc_sectorsize = md->md_sectorsize; 2533 sc->sc_ndisks = md->md_all; 2534 sc->sc_flags = md->md_mflags; 2535 sc->sc_bump_id = 0; 2536 sc->sc_idle = 0; 2537 bioq_init(&sc->sc_queue); 2538 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2539 LIST_INIT(&sc->sc_disks); 2540 TAILQ_INIT(&sc->sc_events); 2541 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2542 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2543 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2544 gp->softc = sc; 2545 sc->sc_geom = gp; 2546 sc->sc_provider = NULL; 2547 /* 2548 * Synchronization geom. 2549 */ 2550 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2551 gp->softc = sc; 2552 gp->orphan = g_mirror_orphan; 2553 sc->sc_sync.ds_geom = gp; 2554 sc->sc_sync.ds_ndisks = 0; 2555 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2556 "g_mirror %s", md->md_name); 2557 if (error != 0) { 2558 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2559 sc->sc_name); 2560 g_destroy_geom(sc->sc_sync.ds_geom); 2561 mtx_destroy(&sc->sc_events_mtx); 2562 mtx_destroy(&sc->sc_queue_mtx); 2563 g_destroy_geom(sc->sc_geom); 2564 free(sc, M_MIRROR); 2565 return (NULL); 2566 } 2567 2568 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2569 2570 /* 2571 * Run timeout. 2572 */ 2573 timeout = g_mirror_timeout * hz; 2574 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2575 return (sc->sc_geom); 2576 } 2577 2578 int 2579 g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2580 { 2581 struct g_provider *pp; 2582 2583 g_topology_assert(); 2584 2585 if (sc == NULL) 2586 return (ENXIO); 2587 pp = sc->sc_provider; 2588 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2589 if (force) { 2590 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 2591 "can't be definitely removed.", pp->name); 2592 } else { 2593 G_MIRROR_DEBUG(1, 2594 "Device %s is still open (r%dw%de%d).", pp->name, 2595 pp->acr, pp->acw, pp->ace); 2596 return (EBUSY); 2597 } 2598 } 2599 2600 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2601 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2602 g_topology_unlock(); 2603 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2604 mtx_lock(&sc->sc_queue_mtx); 2605 wakeup(sc); 2606 mtx_unlock(&sc->sc_queue_mtx); 2607 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2608 while (sc->sc_worker != NULL) 2609 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2610 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2611 g_topology_lock(); 2612 g_mirror_destroy_device(sc); 2613 free(sc, M_MIRROR); 2614 return (0); 2615 } 2616 2617 static void 2618 g_mirror_taste_orphan(struct g_consumer *cp) 2619 { 2620 2621 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2622 cp->provider->name)); 2623 } 2624 2625 static struct g_geom * 2626 g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2627 { 2628 struct g_mirror_metadata md; 2629 struct g_mirror_softc *sc; 2630 struct g_consumer *cp; 2631 struct g_geom *gp; 2632 int error; 2633 2634 g_topology_assert(); 2635 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2636 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2637 2638 gp = g_new_geomf(mp, "mirror:taste"); 2639 /* 2640 * This orphan function should be never called. 2641 */ 2642 gp->orphan = g_mirror_taste_orphan; 2643 cp = g_new_consumer(gp); 2644 g_attach(cp, pp); 2645 error = g_mirror_read_metadata(cp, &md); 2646 g_detach(cp); 2647 g_destroy_consumer(cp); 2648 g_destroy_geom(gp); 2649 if (error != 0) 2650 return (NULL); 2651 gp = NULL; 2652 2653 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2654 return (NULL); 2655 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 2656 return (NULL); 2657 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2658 G_MIRROR_DEBUG(0, 2659 "Device %s: provider %s marked as inactive, skipping.", 2660 md.md_name, pp->name); 2661 return (NULL); 2662 } 2663 if (g_mirror_debug >= 2) 2664 mirror_metadata_dump(&md); 2665 2666 /* 2667 * Let's check if device already exists. 2668 */ 2669 sc = NULL; 2670 LIST_FOREACH(gp, &mp->geom, geom) { 2671 sc = gp->softc; 2672 if (sc == NULL) 2673 continue; 2674 if (sc->sc_sync.ds_geom == gp) 2675 continue; 2676 if (strcmp(md.md_name, sc->sc_name) != 0) 2677 continue; 2678 if (md.md_mid != sc->sc_id) { 2679 G_MIRROR_DEBUG(0, "Device %s already configured.", 2680 sc->sc_name); 2681 return (NULL); 2682 } 2683 break; 2684 } 2685 if (gp == NULL) { 2686 gp = g_mirror_create(mp, &md); 2687 if (gp == NULL) { 2688 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2689 md.md_name); 2690 return (NULL); 2691 } 2692 sc = gp->softc; 2693 } 2694 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2695 error = g_mirror_add_disk(sc, pp, &md); 2696 if (error != 0) { 2697 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2698 pp->name, gp->name, error); 2699 if (LIST_EMPTY(&sc->sc_disks)) 2700 g_mirror_destroy(sc, 1); 2701 return (NULL); 2702 } 2703 return (gp); 2704 } 2705 2706 static int 2707 g_mirror_destroy_geom(struct gctl_req *req __unused, 2708 struct g_class *mp __unused, struct g_geom *gp) 2709 { 2710 2711 return (g_mirror_destroy(gp->softc, 0)); 2712 } 2713 2714 static void 2715 g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2716 struct g_consumer *cp, struct g_provider *pp) 2717 { 2718 struct g_mirror_softc *sc; 2719 2720 g_topology_assert(); 2721 2722 sc = gp->softc; 2723 if (sc == NULL) 2724 return; 2725 /* Skip synchronization geom. */ 2726 if (gp == sc->sc_sync.ds_geom) 2727 return; 2728 if (pp != NULL) { 2729 /* Nothing here. */ 2730 } else if (cp != NULL) { 2731 struct g_mirror_disk *disk; 2732 2733 disk = cp->private; 2734 if (disk == NULL) 2735 return; 2736 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2737 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2738 sbuf_printf(sb, "%s<Synchronized>", indent); 2739 if (disk->d_sync.ds_offset_done == 0) 2740 sbuf_printf(sb, "0%%"); 2741 else { 2742 sbuf_printf(sb, "%u%%", 2743 (u_int)((disk->d_sync.ds_offset_done * 100) / 2744 sc->sc_provider->mediasize)); 2745 } 2746 sbuf_printf(sb, "</Synchronized>\n"); 2747 } 2748 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2749 disk->d_sync.ds_syncid); 2750 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 2751 disk->d_genid); 2752 sbuf_printf(sb, "%s<Flags>", indent); 2753 if (disk->d_flags == 0) 2754 sbuf_printf(sb, "NONE"); 2755 else { 2756 int first = 1; 2757 2758 #define ADD_FLAG(flag, name) do { \ 2759 if ((disk->d_flags & (flag)) != 0) { \ 2760 if (!first) \ 2761 sbuf_printf(sb, ", "); \ 2762 else \ 2763 first = 0; \ 2764 sbuf_printf(sb, name); \ 2765 } \ 2766 } while (0) 2767 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2768 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2769 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2770 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2771 "SYNCHRONIZING"); 2772 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2773 #undef ADD_FLAG 2774 } 2775 sbuf_printf(sb, "</Flags>\n"); 2776 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2777 disk->d_priority); 2778 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2779 g_mirror_disk_state2str(disk->d_state)); 2780 } else { 2781 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2782 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2783 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 2784 sbuf_printf(sb, "%s<Flags>", indent); 2785 if (sc->sc_flags == 0) 2786 sbuf_printf(sb, "NONE"); 2787 else { 2788 int first = 1; 2789 2790 #define ADD_FLAG(flag, name) do { \ 2791 if ((sc->sc_flags & (flag)) != 0) { \ 2792 if (!first) \ 2793 sbuf_printf(sb, ", "); \ 2794 else \ 2795 first = 0; \ 2796 sbuf_printf(sb, name); \ 2797 } \ 2798 } while (0) 2799 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2800 #undef ADD_FLAG 2801 } 2802 sbuf_printf(sb, "</Flags>\n"); 2803 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2804 (u_int)sc->sc_slice); 2805 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2806 balance_name(sc->sc_balance)); 2807 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2808 sc->sc_ndisks); 2809 sbuf_printf(sb, "%s<State>", indent); 2810 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2811 sbuf_printf(sb, "%s", "STARTING"); 2812 else if (sc->sc_ndisks == 2813 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 2814 sbuf_printf(sb, "%s", "COMPLETE"); 2815 else 2816 sbuf_printf(sb, "%s", "DEGRADED"); 2817 sbuf_printf(sb, "</State>\n"); 2818 } 2819 } 2820 2821 static void 2822 g_mirror_shutdown(void *arg, int howto) 2823 { 2824 struct g_class *mp; 2825 struct g_geom *gp, *gp2; 2826 2827 mp = arg; 2828 DROP_GIANT(); 2829 g_topology_lock(); 2830 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2831 if (gp->softc == NULL) 2832 continue; 2833 g_mirror_destroy(gp->softc, 1); 2834 } 2835 g_topology_unlock(); 2836 PICKUP_GIANT(); 2837 #if 0 2838 tsleep(&gp, PRIBIO, "m:shutdown", hz * 20); 2839 #endif 2840 } 2841 2842 static void 2843 g_mirror_init(struct g_class *mp) 2844 { 2845 2846 g_mirror_ehtag = EVENTHANDLER_REGISTER(shutdown_post_sync, 2847 g_mirror_shutdown, mp, SHUTDOWN_PRI_FIRST); 2848 if (g_mirror_ehtag == NULL) 2849 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 2850 } 2851 2852 static void 2853 g_mirror_fini(struct g_class *mp) 2854 { 2855 2856 if (g_mirror_ehtag == NULL) 2857 return; 2858 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_ehtag); 2859 } 2860 2861 static int 2862 g_mirror_can_go(void) 2863 { 2864 struct g_mirror_softc *sc; 2865 struct g_geom *gp; 2866 struct g_provider *pp; 2867 int can_go; 2868 2869 DROP_GIANT(); 2870 can_go = 1; 2871 g_topology_lock(); 2872 LIST_FOREACH(gp, &g_mirror_class.geom, geom) { 2873 sc = gp->softc; 2874 if (sc == NULL) { 2875 can_go = 0; 2876 break; 2877 } 2878 pp = sc->sc_provider; 2879 if (pp == NULL || pp->error != 0) { 2880 can_go = 0; 2881 break; 2882 } 2883 } 2884 g_topology_unlock(); 2885 PICKUP_GIANT(); 2886 return (can_go); 2887 } 2888 2889 static void 2890 g_mirror_rootwait(void) 2891 { 2892 2893 /* 2894 * HACK: Wait for GEOM, because g_mirror_rootwait() can be called, 2895 * HACK: before we get providers for tasting. 2896 */ 2897 tsleep(&g_mirror_class, PRIBIO, "mroot", hz * 3); 2898 /* 2899 * Wait for mirrors in degraded state. 2900 */ 2901 for (;;) { 2902 if (g_mirror_can_go()) 2903 break; 2904 tsleep(&g_mirror_class, PRIBIO, "mroot", hz); 2905 } 2906 } 2907 2908 SYSINIT(g_mirror_root, SI_SUB_RAID, SI_ORDER_FIRST, g_mirror_rootwait, NULL) 2909 2910 DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2911