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 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1217 cbp = g_clone_bio(bp); 1218 if (cbp == NULL) { 1219 if (bp->bio_error == 0) 1220 bp->bio_error = ENOMEM; 1221 g_io_deliver(bp, bp->bio_error); 1222 return; 1223 } 1224 /* 1225 * Fill in the component buf structure. 1226 */ 1227 cp = disk->d_consumer; 1228 cbp->bio_done = g_mirror_done; 1229 cbp->bio_to = cp->provider; 1230 binuptime(&disk->d_last_used); 1231 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1232 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1233 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1234 cp->acw, cp->ace)); 1235 cp->index++; 1236 g_io_request(cbp, cp); 1237 } 1238 1239 static void 1240 g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1241 { 1242 struct bio_queue_head queue; 1243 struct g_mirror_disk *disk; 1244 struct g_consumer *cp; 1245 struct bio *cbp; 1246 off_t left, mod, offset, slice; 1247 u_char *data; 1248 u_int ndisks; 1249 1250 if (bp->bio_length <= sc->sc_slice) { 1251 g_mirror_request_round_robin(sc, bp); 1252 return; 1253 } 1254 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1255 slice = bp->bio_length / ndisks; 1256 mod = slice % sc->sc_provider->sectorsize; 1257 if (mod != 0) 1258 slice += sc->sc_provider->sectorsize - mod; 1259 /* 1260 * Allocate all bios before sending any request, so we can 1261 * return ENOMEM in nice and clean way. 1262 */ 1263 left = bp->bio_length; 1264 offset = bp->bio_offset; 1265 data = bp->bio_data; 1266 bioq_init(&queue); 1267 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1268 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1269 continue; 1270 cbp = g_clone_bio(bp); 1271 if (cbp == NULL) { 1272 for (cbp = bioq_first(&queue); cbp != NULL; 1273 cbp = bioq_first(&queue)) { 1274 bioq_remove(&queue, cbp); 1275 g_destroy_bio(cbp); 1276 } 1277 if (bp->bio_error == 0) 1278 bp->bio_error = ENOMEM; 1279 g_io_deliver(bp, bp->bio_error); 1280 return; 1281 } 1282 bioq_insert_tail(&queue, cbp); 1283 cbp->bio_done = g_mirror_done; 1284 cbp->bio_caller1 = disk; 1285 cbp->bio_to = disk->d_consumer->provider; 1286 cbp->bio_offset = offset; 1287 cbp->bio_data = data; 1288 cbp->bio_length = MIN(left, slice); 1289 left -= cbp->bio_length; 1290 if (left == 0) 1291 break; 1292 offset += cbp->bio_length; 1293 data += cbp->bio_length; 1294 } 1295 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1296 bioq_remove(&queue, cbp); 1297 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1298 disk = cbp->bio_caller1; 1299 cbp->bio_caller1 = NULL; 1300 cp = disk->d_consumer; 1301 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1302 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1303 cp->acr, cp->acw, cp->ace)); 1304 disk->d_consumer->index++; 1305 g_io_request(cbp, disk->d_consumer); 1306 } 1307 } 1308 1309 static void 1310 g_mirror_register_request(struct bio *bp) 1311 { 1312 struct g_mirror_softc *sc; 1313 1314 sc = bp->bio_to->geom->softc; 1315 switch (bp->bio_cmd) { 1316 case BIO_READ: 1317 switch (sc->sc_balance) { 1318 case G_MIRROR_BALANCE_LOAD: 1319 g_mirror_request_load(sc, bp); 1320 break; 1321 case G_MIRROR_BALANCE_PREFER: 1322 g_mirror_request_prefer(sc, bp); 1323 break; 1324 case G_MIRROR_BALANCE_ROUND_ROBIN: 1325 g_mirror_request_round_robin(sc, bp); 1326 break; 1327 case G_MIRROR_BALANCE_SPLIT: 1328 g_mirror_request_split(sc, bp); 1329 break; 1330 } 1331 return; 1332 case BIO_WRITE: 1333 case BIO_DELETE: 1334 { 1335 struct g_mirror_disk *disk; 1336 struct g_mirror_disk_sync *sync; 1337 struct bio_queue_head queue; 1338 struct g_consumer *cp; 1339 struct bio *cbp; 1340 1341 if (sc->sc_idle) 1342 g_mirror_unidle(sc); 1343 /* 1344 * Allocate all bios before sending any request, so we can 1345 * return ENOMEM in nice and clean way. 1346 */ 1347 bioq_init(&queue); 1348 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1349 sync = &disk->d_sync; 1350 switch (disk->d_state) { 1351 case G_MIRROR_DISK_STATE_ACTIVE: 1352 break; 1353 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1354 if (bp->bio_offset >= sync->ds_offset) 1355 continue; 1356 else if (bp->bio_offset + bp->bio_length > 1357 sync->ds_offset_done && 1358 (bp->bio_offset < sync->ds_resync || 1359 sync->ds_resync == -1)) { 1360 sync->ds_resync = bp->bio_offset - 1361 (bp->bio_offset % MAXPHYS); 1362 } 1363 break; 1364 default: 1365 continue; 1366 } 1367 cbp = g_clone_bio(bp); 1368 if (cbp == NULL) { 1369 for (cbp = bioq_first(&queue); cbp != NULL; 1370 cbp = bioq_first(&queue)) { 1371 bioq_remove(&queue, cbp); 1372 g_destroy_bio(cbp); 1373 } 1374 if (bp->bio_error == 0) 1375 bp->bio_error = ENOMEM; 1376 g_io_deliver(bp, bp->bio_error); 1377 return; 1378 } 1379 bioq_insert_tail(&queue, cbp); 1380 cbp->bio_done = g_mirror_done; 1381 cp = disk->d_consumer; 1382 cbp->bio_caller1 = cp; 1383 cbp->bio_to = cp->provider; 1384 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1385 ("Consumer %s not opened (r%dw%de%d).", 1386 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1387 } 1388 for (cbp = bioq_first(&queue); cbp != NULL; 1389 cbp = bioq_first(&queue)) { 1390 bioq_remove(&queue, cbp); 1391 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1392 cp = cbp->bio_caller1; 1393 cbp->bio_caller1 = NULL; 1394 cp->index++; 1395 g_io_request(cbp, cp); 1396 } 1397 /* 1398 * Bump syncid on first write. 1399 */ 1400 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1401 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1402 g_topology_lock(); 1403 g_mirror_bump_syncid(sc); 1404 g_topology_unlock(); 1405 } 1406 return; 1407 } 1408 default: 1409 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1410 bp->bio_cmd, sc->sc_name)); 1411 break; 1412 } 1413 } 1414 1415 static int 1416 g_mirror_can_destroy(struct g_mirror_softc *sc) 1417 { 1418 struct g_geom *gp; 1419 struct g_consumer *cp; 1420 1421 g_topology_assert(); 1422 gp = sc->sc_geom; 1423 LIST_FOREACH(cp, &gp->consumer, consumer) { 1424 if (g_mirror_is_busy(sc, cp)) 1425 return (0); 1426 } 1427 gp = sc->sc_sync.ds_geom; 1428 LIST_FOREACH(cp, &gp->consumer, consumer) { 1429 if (g_mirror_is_busy(sc, cp)) 1430 return (0); 1431 } 1432 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1433 sc->sc_name); 1434 return (1); 1435 } 1436 1437 static int 1438 g_mirror_try_destroy(struct g_mirror_softc *sc) 1439 { 1440 1441 if (sc->sc_rootmount != NULL) { 1442 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1443 sc->sc_rootmount); 1444 root_mount_rel(sc->sc_rootmount); 1445 sc->sc_rootmount = NULL; 1446 } 1447 g_topology_lock(); 1448 if (!g_mirror_can_destroy(sc)) { 1449 g_topology_unlock(); 1450 return (0); 1451 } 1452 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1453 g_topology_unlock(); 1454 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1455 &sc->sc_worker); 1456 wakeup(&sc->sc_worker); 1457 sc->sc_worker = NULL; 1458 } else { 1459 g_mirror_destroy_device(sc); 1460 g_topology_unlock(); 1461 free(sc, M_MIRROR); 1462 } 1463 return (1); 1464 } 1465 1466 /* 1467 * Worker thread. 1468 */ 1469 static void 1470 g_mirror_worker(void *arg) 1471 { 1472 struct g_mirror_softc *sc; 1473 struct g_mirror_disk *disk; 1474 struct g_mirror_disk_sync *sync; 1475 struct g_mirror_event *ep; 1476 struct bio *bp; 1477 u_int nreqs; 1478 1479 sc = arg; 1480 mtx_lock_spin(&sched_lock); 1481 sched_prio(curthread, PRIBIO); 1482 mtx_unlock_spin(&sched_lock); 1483 1484 nreqs = 0; 1485 for (;;) { 1486 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1487 /* 1488 * First take a look at events. 1489 * This is important to handle events before any I/O requests. 1490 */ 1491 ep = g_mirror_event_get(sc); 1492 if (ep != NULL && g_topology_try_lock()) { 1493 g_mirror_event_remove(sc, ep); 1494 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1495 /* Update only device status. */ 1496 G_MIRROR_DEBUG(3, 1497 "Running event for device %s.", 1498 sc->sc_name); 1499 ep->e_error = 0; 1500 g_mirror_update_device(sc, 1); 1501 } else { 1502 /* Update disk status. */ 1503 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1504 g_mirror_get_diskname(ep->e_disk)); 1505 ep->e_error = g_mirror_update_disk(ep->e_disk, 1506 ep->e_state); 1507 if (ep->e_error == 0) 1508 g_mirror_update_device(sc, 0); 1509 } 1510 g_topology_unlock(); 1511 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1512 KASSERT(ep->e_error == 0, 1513 ("Error cannot be handled.")); 1514 g_mirror_event_free(ep); 1515 } else { 1516 ep->e_flags |= G_MIRROR_EVENT_DONE; 1517 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1518 ep); 1519 mtx_lock(&sc->sc_events_mtx); 1520 wakeup(ep); 1521 mtx_unlock(&sc->sc_events_mtx); 1522 } 1523 if ((sc->sc_flags & 1524 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1525 if (g_mirror_try_destroy(sc)) 1526 kthread_exit(0); 1527 } 1528 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1529 continue; 1530 } 1531 /* 1532 * Now I/O requests. 1533 */ 1534 /* Get first request from the queue. */ 1535 mtx_lock(&sc->sc_queue_mtx); 1536 bp = bioq_first(&sc->sc_queue); 1537 if (bp == NULL) { 1538 if (ep != NULL) { 1539 /* 1540 * No I/O requests and topology lock was 1541 * already held? Try again. 1542 */ 1543 mtx_unlock(&sc->sc_queue_mtx); 1544 continue; 1545 } 1546 if ((sc->sc_flags & 1547 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1548 mtx_unlock(&sc->sc_queue_mtx); 1549 if (g_mirror_try_destroy(sc)) 1550 kthread_exit(0); 1551 mtx_lock(&sc->sc_queue_mtx); 1552 } 1553 } 1554 if (sc->sc_sync.ds_ndisks > 0 && 1555 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1556 mtx_unlock(&sc->sc_queue_mtx); 1557 /* 1558 * It is time for synchronization... 1559 */ 1560 nreqs = 0; 1561 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1562 if (disk->d_state != 1563 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1564 continue; 1565 } 1566 sync = &disk->d_sync; 1567 if (sync->ds_offset >= 1568 sc->sc_provider->mediasize) { 1569 continue; 1570 } 1571 if (sync->ds_offset > sync->ds_offset_done) 1572 continue; 1573 if (sync->ds_resync != -1) { 1574 sync->ds_offset = sync->ds_resync; 1575 sync->ds_offset_done = sync->ds_resync; 1576 sync->ds_resync = -1; 1577 } 1578 g_mirror_sync_one(disk); 1579 } 1580 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1581 goto sleep; 1582 } 1583 if (bp == NULL) { 1584 if (g_mirror_check_idle(sc)) { 1585 u_int idletime; 1586 1587 idletime = g_mirror_idletime; 1588 if (idletime == 0) 1589 idletime = 1; 1590 idletime *= hz; 1591 if (msleep(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1592 "m:w1", idletime) == EWOULDBLOCK) { 1593 G_MIRROR_DEBUG(5, "%s: I'm here 3.", 1594 __func__); 1595 /* 1596 * No I/O requests in 'idletime' seconds, 1597 * so mark components as clean. 1598 */ 1599 g_mirror_idle(sc); 1600 } 1601 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1602 } else { 1603 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1604 "m:w2", 0); 1605 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1606 } 1607 continue; 1608 } 1609 nreqs++; 1610 bioq_remove(&sc->sc_queue, bp); 1611 mtx_unlock(&sc->sc_queue_mtx); 1612 1613 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1614 g_mirror_regular_request(bp); 1615 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1616 u_int timeout, sps; 1617 1618 g_mirror_sync_request(bp); 1619 sleep: 1620 sps = g_mirror_syncs_per_sec; 1621 if (sps == 0) { 1622 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1623 continue; 1624 } 1625 if (ep != NULL) { 1626 /* 1627 * We have some pending events, don't sleep now. 1628 */ 1629 G_MIRROR_DEBUG(5, "%s: I'm here 7.", __func__); 1630 continue; 1631 } 1632 mtx_lock(&sc->sc_queue_mtx); 1633 if (bioq_first(&sc->sc_queue) != NULL) { 1634 mtx_unlock(&sc->sc_queue_mtx); 1635 G_MIRROR_DEBUG(5, "%s: I'm here 8.", __func__); 1636 continue; 1637 } 1638 timeout = hz / sps; 1639 if (timeout == 0) 1640 timeout = 1; 1641 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w3", 1642 timeout); 1643 } else { 1644 g_mirror_register_request(bp); 1645 } 1646 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1647 } 1648 } 1649 1650 /* 1651 * Open disk's consumer if needed. 1652 */ 1653 static void 1654 g_mirror_update_access(struct g_mirror_disk *disk) 1655 { 1656 struct g_provider *pp; 1657 1658 g_topology_assert(); 1659 1660 pp = disk->d_softc->sc_provider; 1661 if (pp == NULL) 1662 return; 1663 if (pp->acw > 0) { 1664 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1665 G_MIRROR_DEBUG(1, 1666 "Disk %s (device %s) marked as dirty.", 1667 g_mirror_get_diskname(disk), 1668 disk->d_softc->sc_name); 1669 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1670 } 1671 } else if (pp->acw == 0) { 1672 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1673 G_MIRROR_DEBUG(1, 1674 "Disk %s (device %s) marked as clean.", 1675 g_mirror_get_diskname(disk), 1676 disk->d_softc->sc_name); 1677 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1678 } 1679 } 1680 } 1681 1682 static void 1683 g_mirror_sync_start(struct g_mirror_disk *disk) 1684 { 1685 struct g_mirror_softc *sc; 1686 int error; 1687 1688 g_topology_assert(); 1689 1690 sc = disk->d_softc; 1691 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1692 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1693 sc->sc_state)); 1694 1695 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1696 g_mirror_get_diskname(disk)); 1697 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1698 KASSERT(disk->d_sync.ds_consumer == NULL, 1699 ("Sync consumer already exists (device=%s, disk=%s).", 1700 sc->sc_name, g_mirror_get_diskname(disk))); 1701 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1702 disk->d_sync.ds_consumer->private = disk; 1703 disk->d_sync.ds_consumer->index = 0; 1704 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1705 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1706 disk->d_softc->sc_name, error)); 1707 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1708 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1709 disk->d_softc->sc_name, error)); 1710 disk->d_sync.ds_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1711 sc->sc_sync.ds_ndisks++; 1712 } 1713 1714 /* 1715 * Stop synchronization process. 1716 * type: 0 - synchronization finished 1717 * 1 - synchronization stopped 1718 */ 1719 static void 1720 g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1721 { 1722 1723 g_topology_assert(); 1724 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1725 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1726 g_mirror_disk_state2str(disk->d_state))); 1727 if (disk->d_sync.ds_consumer == NULL) 1728 return; 1729 1730 if (type == 0) { 1731 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1732 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1733 } else /* if (type == 1) */ { 1734 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1735 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1736 } 1737 g_mirror_kill_consumer(disk->d_softc, disk->d_sync.ds_consumer); 1738 free(disk->d_sync.ds_data, M_MIRROR); 1739 disk->d_sync.ds_consumer = NULL; 1740 disk->d_softc->sc_sync.ds_ndisks--; 1741 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1742 } 1743 1744 static void 1745 g_mirror_launch_provider(struct g_mirror_softc *sc) 1746 { 1747 struct g_mirror_disk *disk; 1748 struct g_provider *pp; 1749 1750 g_topology_assert(); 1751 1752 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1753 pp->mediasize = sc->sc_mediasize; 1754 pp->sectorsize = sc->sc_sectorsize; 1755 sc->sc_provider = pp; 1756 g_error_provider(pp, 0); 1757 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1758 pp->name); 1759 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1760 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1761 g_mirror_sync_start(disk); 1762 } 1763 } 1764 1765 static void 1766 g_mirror_destroy_provider(struct g_mirror_softc *sc) 1767 { 1768 struct g_mirror_disk *disk; 1769 struct bio *bp; 1770 1771 g_topology_assert(); 1772 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1773 sc->sc_name)); 1774 1775 g_error_provider(sc->sc_provider, ENXIO); 1776 mtx_lock(&sc->sc_queue_mtx); 1777 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1778 bioq_remove(&sc->sc_queue, bp); 1779 g_io_deliver(bp, ENXIO); 1780 } 1781 mtx_unlock(&sc->sc_queue_mtx); 1782 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1783 sc->sc_provider->name); 1784 sc->sc_provider->flags |= G_PF_WITHER; 1785 g_orphan_provider(sc->sc_provider, ENXIO); 1786 sc->sc_provider = NULL; 1787 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1788 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1789 g_mirror_sync_stop(disk, 1); 1790 } 1791 } 1792 1793 static void 1794 g_mirror_go(void *arg) 1795 { 1796 struct g_mirror_softc *sc; 1797 1798 sc = arg; 1799 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1800 g_mirror_event_send(sc, 0, 1801 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1802 } 1803 1804 static u_int 1805 g_mirror_determine_state(struct g_mirror_disk *disk) 1806 { 1807 struct g_mirror_softc *sc; 1808 u_int state; 1809 1810 sc = disk->d_softc; 1811 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1812 if ((disk->d_flags & 1813 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1814 /* Disk does not need synchronization. */ 1815 state = G_MIRROR_DISK_STATE_ACTIVE; 1816 } else { 1817 if ((sc->sc_flags & 1818 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1819 (disk->d_flags & 1820 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1821 /* 1822 * We can start synchronization from 1823 * the stored offset. 1824 */ 1825 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1826 } else { 1827 state = G_MIRROR_DISK_STATE_STALE; 1828 } 1829 } 1830 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1831 /* 1832 * Reset all synchronization data for this disk, 1833 * because if it even was synchronized, it was 1834 * synchronized to disks with different syncid. 1835 */ 1836 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1837 disk->d_sync.ds_offset = 0; 1838 disk->d_sync.ds_offset_done = 0; 1839 disk->d_sync.ds_syncid = sc->sc_syncid; 1840 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1841 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1842 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1843 } else { 1844 state = G_MIRROR_DISK_STATE_STALE; 1845 } 1846 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1847 /* 1848 * Not good, NOT GOOD! 1849 * It means that mirror was started on stale disks 1850 * and more fresh disk just arrive. 1851 * If there were writes, mirror is fucked up, sorry. 1852 * I think the best choice here is don't touch 1853 * this disk and inform the user laudly. 1854 */ 1855 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1856 "disk (%s) arrives!! It will not be connected to the " 1857 "running device.", sc->sc_name, 1858 g_mirror_get_diskname(disk)); 1859 g_mirror_destroy_disk(disk); 1860 state = G_MIRROR_DISK_STATE_NONE; 1861 /* Return immediately, because disk was destroyed. */ 1862 return (state); 1863 } 1864 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1865 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1866 return (state); 1867 } 1868 1869 /* 1870 * Update device state. 1871 */ 1872 static void 1873 g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1874 { 1875 struct g_mirror_disk *disk; 1876 u_int state; 1877 1878 g_topology_assert(); 1879 1880 switch (sc->sc_state) { 1881 case G_MIRROR_DEVICE_STATE_STARTING: 1882 { 1883 struct g_mirror_disk *pdisk, *tdisk; 1884 u_int dirty, ndisks, genid, syncid; 1885 1886 KASSERT(sc->sc_provider == NULL, 1887 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1888 /* 1889 * Are we ready? We are, if all disks are connected or 1890 * if we have any disks and 'force' is true. 1891 */ 1892 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1893 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1894 ; 1895 } else if (g_mirror_ndisks(sc, -1) == 0) { 1896 /* 1897 * Disks went down in starting phase, so destroy 1898 * device. 1899 */ 1900 callout_drain(&sc->sc_callout); 1901 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1902 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1903 sc->sc_rootmount); 1904 root_mount_rel(sc->sc_rootmount); 1905 sc->sc_rootmount = NULL; 1906 return; 1907 } else { 1908 return; 1909 } 1910 1911 /* 1912 * Activate all disks with the biggest syncid. 1913 */ 1914 if (force) { 1915 /* 1916 * If 'force' is true, we have been called due to 1917 * timeout, so don't bother canceling timeout. 1918 */ 1919 ndisks = 0; 1920 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1921 if ((disk->d_flags & 1922 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1923 ndisks++; 1924 } 1925 } 1926 if (ndisks == 0) { 1927 /* No valid disks found, destroy device. */ 1928 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1929 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 1930 __LINE__, sc->sc_rootmount); 1931 root_mount_rel(sc->sc_rootmount); 1932 sc->sc_rootmount = NULL; 1933 return; 1934 } 1935 } else { 1936 /* Cancel timeout. */ 1937 callout_drain(&sc->sc_callout); 1938 } 1939 1940 /* 1941 * Find the biggest genid. 1942 */ 1943 genid = 0; 1944 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1945 if (disk->d_genid > genid) 1946 genid = disk->d_genid; 1947 } 1948 sc->sc_genid = genid; 1949 /* 1950 * Remove all disks without the biggest genid. 1951 */ 1952 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 1953 if (disk->d_genid < genid) { 1954 G_MIRROR_DEBUG(0, 1955 "Component %s (device %s) broken, skipping.", 1956 g_mirror_get_diskname(disk), sc->sc_name); 1957 g_mirror_destroy_disk(disk); 1958 } 1959 } 1960 1961 /* 1962 * Find the biggest syncid. 1963 */ 1964 syncid = 0; 1965 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1966 if (disk->d_sync.ds_syncid > syncid) 1967 syncid = disk->d_sync.ds_syncid; 1968 } 1969 1970 /* 1971 * Here we need to look for dirty disks and if all disks 1972 * with the biggest syncid are dirty, we have to choose 1973 * one with the biggest priority and rebuild the rest. 1974 */ 1975 /* 1976 * Find the number of dirty disks with the biggest syncid. 1977 * Find the number of disks with the biggest syncid. 1978 * While here, find a disk with the biggest priority. 1979 */ 1980 dirty = ndisks = 0; 1981 pdisk = NULL; 1982 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1983 if (disk->d_sync.ds_syncid != syncid) 1984 continue; 1985 if ((disk->d_flags & 1986 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1987 continue; 1988 } 1989 ndisks++; 1990 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1991 dirty++; 1992 if (pdisk == NULL || 1993 pdisk->d_priority < disk->d_priority) { 1994 pdisk = disk; 1995 } 1996 } 1997 } 1998 if (dirty == 0) { 1999 /* No dirty disks at all, great. */ 2000 } else if (dirty == ndisks) { 2001 /* 2002 * Force synchronization for all dirty disks except one 2003 * with the biggest priority. 2004 */ 2005 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2006 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2007 "master disk for synchronization.", 2008 g_mirror_get_diskname(pdisk), sc->sc_name); 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 KASSERT((disk->d_flags & 2017 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2018 ("Disk %s isn't marked as dirty.", 2019 g_mirror_get_diskname(disk))); 2020 /* Skip the disk with the biggest priority. */ 2021 if (disk == pdisk) 2022 continue; 2023 disk->d_sync.ds_syncid = 0; 2024 } 2025 } else if (dirty < ndisks) { 2026 /* 2027 * Force synchronization for all dirty disks. 2028 * We have some non-dirty disks. 2029 */ 2030 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2031 if (disk->d_sync.ds_syncid != syncid) 2032 continue; 2033 if ((disk->d_flags & 2034 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2035 continue; 2036 } 2037 if ((disk->d_flags & 2038 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2039 continue; 2040 } 2041 disk->d_sync.ds_syncid = 0; 2042 } 2043 } 2044 2045 /* Reset hint. */ 2046 sc->sc_hint = NULL; 2047 sc->sc_syncid = syncid; 2048 if (force) { 2049 /* Remember to bump syncid on first write. */ 2050 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2051 } 2052 state = G_MIRROR_DEVICE_STATE_RUNNING; 2053 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2054 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2055 g_mirror_device_state2str(state)); 2056 sc->sc_state = state; 2057 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2058 state = g_mirror_determine_state(disk); 2059 g_mirror_event_send(disk, state, 2060 G_MIRROR_EVENT_DONTWAIT); 2061 if (state == G_MIRROR_DISK_STATE_STALE) 2062 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2063 } 2064 break; 2065 } 2066 case G_MIRROR_DEVICE_STATE_RUNNING: 2067 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2068 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2069 /* 2070 * No active disks or no disks at all, 2071 * so destroy device. 2072 */ 2073 if (sc->sc_provider != NULL) 2074 g_mirror_destroy_provider(sc); 2075 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2076 break; 2077 } else if (g_mirror_ndisks(sc, 2078 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2079 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2080 /* 2081 * We have active disks, launch provider if it doesn't 2082 * exist. 2083 */ 2084 if (sc->sc_provider == NULL) 2085 g_mirror_launch_provider(sc); 2086 if (sc->sc_rootmount != NULL) { 2087 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2088 __LINE__, sc->sc_rootmount); 2089 root_mount_rel(sc->sc_rootmount); 2090 sc->sc_rootmount = NULL; 2091 } 2092 } 2093 /* 2094 * Genid should be bumped immediately, so do it here. 2095 */ 2096 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2097 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2098 g_mirror_bump_genid(sc); 2099 } 2100 break; 2101 default: 2102 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2103 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2104 break; 2105 } 2106 } 2107 2108 /* 2109 * Update disk state and device state if needed. 2110 */ 2111 #define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2112 "Disk %s state changed from %s to %s (device %s).", \ 2113 g_mirror_get_diskname(disk), \ 2114 g_mirror_disk_state2str(disk->d_state), \ 2115 g_mirror_disk_state2str(state), sc->sc_name) 2116 static int 2117 g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2118 { 2119 struct g_mirror_softc *sc; 2120 2121 g_topology_assert(); 2122 2123 sc = disk->d_softc; 2124 again: 2125 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2126 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2127 g_mirror_disk_state2str(state)); 2128 switch (state) { 2129 case G_MIRROR_DISK_STATE_NEW: 2130 /* 2131 * Possible scenarios: 2132 * 1. New disk arrive. 2133 */ 2134 /* Previous state should be NONE. */ 2135 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2136 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2137 g_mirror_disk_state2str(disk->d_state))); 2138 DISK_STATE_CHANGED(); 2139 2140 disk->d_state = state; 2141 if (LIST_EMPTY(&sc->sc_disks)) 2142 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2143 else { 2144 struct g_mirror_disk *dp; 2145 2146 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2147 if (disk->d_priority >= dp->d_priority) { 2148 LIST_INSERT_BEFORE(dp, disk, d_next); 2149 dp = NULL; 2150 break; 2151 } 2152 if (LIST_NEXT(dp, d_next) == NULL) 2153 break; 2154 } 2155 if (dp != NULL) 2156 LIST_INSERT_AFTER(dp, disk, d_next); 2157 } 2158 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2159 sc->sc_name, g_mirror_get_diskname(disk)); 2160 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2161 break; 2162 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2163 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2164 g_mirror_device_state2str(sc->sc_state), 2165 g_mirror_get_diskname(disk), 2166 g_mirror_disk_state2str(disk->d_state))); 2167 state = g_mirror_determine_state(disk); 2168 if (state != G_MIRROR_DISK_STATE_NONE) 2169 goto again; 2170 break; 2171 case G_MIRROR_DISK_STATE_ACTIVE: 2172 /* 2173 * Possible scenarios: 2174 * 1. New disk does not need synchronization. 2175 * 2. Synchronization process finished successfully. 2176 */ 2177 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2178 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2179 g_mirror_device_state2str(sc->sc_state), 2180 g_mirror_get_diskname(disk), 2181 g_mirror_disk_state2str(disk->d_state))); 2182 /* Previous state should be NEW or SYNCHRONIZING. */ 2183 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2184 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2185 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2186 g_mirror_disk_state2str(disk->d_state))); 2187 DISK_STATE_CHANGED(); 2188 2189 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2190 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2191 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2192 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2193 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2194 g_mirror_sync_stop(disk, 0); 2195 } 2196 disk->d_state = state; 2197 disk->d_sync.ds_offset = 0; 2198 disk->d_sync.ds_offset_done = 0; 2199 g_mirror_update_access(disk); 2200 g_mirror_update_metadata(disk); 2201 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2202 sc->sc_name, g_mirror_get_diskname(disk)); 2203 break; 2204 case G_MIRROR_DISK_STATE_STALE: 2205 /* 2206 * Possible scenarios: 2207 * 1. Stale disk was connected. 2208 */ 2209 /* Previous state should be NEW. */ 2210 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2211 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2212 g_mirror_disk_state2str(disk->d_state))); 2213 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2214 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2215 g_mirror_device_state2str(sc->sc_state), 2216 g_mirror_get_diskname(disk), 2217 g_mirror_disk_state2str(disk->d_state))); 2218 /* 2219 * STALE state is only possible if device is marked 2220 * NOAUTOSYNC. 2221 */ 2222 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2223 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2224 g_mirror_device_state2str(sc->sc_state), 2225 g_mirror_get_diskname(disk), 2226 g_mirror_disk_state2str(disk->d_state))); 2227 DISK_STATE_CHANGED(); 2228 2229 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2230 disk->d_state = state; 2231 g_mirror_update_metadata(disk); 2232 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2233 sc->sc_name, g_mirror_get_diskname(disk)); 2234 break; 2235 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2236 /* 2237 * Possible scenarios: 2238 * 1. Disk which needs synchronization was connected. 2239 */ 2240 /* Previous state should be NEW. */ 2241 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2242 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2243 g_mirror_disk_state2str(disk->d_state))); 2244 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2245 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2246 g_mirror_device_state2str(sc->sc_state), 2247 g_mirror_get_diskname(disk), 2248 g_mirror_disk_state2str(disk->d_state))); 2249 DISK_STATE_CHANGED(); 2250 2251 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2252 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2253 disk->d_state = state; 2254 if (sc->sc_provider != NULL) { 2255 g_mirror_sync_start(disk); 2256 g_mirror_update_metadata(disk); 2257 } 2258 break; 2259 case G_MIRROR_DISK_STATE_DISCONNECTED: 2260 /* 2261 * Possible scenarios: 2262 * 1. Device wasn't running yet, but disk disappear. 2263 * 2. Disk was active and disapppear. 2264 * 3. Disk disappear during synchronization process. 2265 */ 2266 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2267 /* 2268 * Previous state should be ACTIVE, STALE or 2269 * SYNCHRONIZING. 2270 */ 2271 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2272 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2273 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2274 ("Wrong disk state (%s, %s).", 2275 g_mirror_get_diskname(disk), 2276 g_mirror_disk_state2str(disk->d_state))); 2277 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2278 /* Previous state should be NEW. */ 2279 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2280 ("Wrong disk state (%s, %s).", 2281 g_mirror_get_diskname(disk), 2282 g_mirror_disk_state2str(disk->d_state))); 2283 /* 2284 * Reset bumping syncid if disk disappeared in STARTING 2285 * state. 2286 */ 2287 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2288 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2289 #ifdef INVARIANTS 2290 } else { 2291 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2292 sc->sc_name, 2293 g_mirror_device_state2str(sc->sc_state), 2294 g_mirror_get_diskname(disk), 2295 g_mirror_disk_state2str(disk->d_state))); 2296 #endif 2297 } 2298 DISK_STATE_CHANGED(); 2299 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2300 sc->sc_name, g_mirror_get_diskname(disk)); 2301 2302 g_mirror_destroy_disk(disk); 2303 break; 2304 case G_MIRROR_DISK_STATE_DESTROY: 2305 { 2306 int error; 2307 2308 error = g_mirror_clear_metadata(disk); 2309 if (error != 0) 2310 return (error); 2311 DISK_STATE_CHANGED(); 2312 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2313 sc->sc_name, g_mirror_get_diskname(disk)); 2314 2315 g_mirror_destroy_disk(disk); 2316 sc->sc_ndisks--; 2317 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2318 g_mirror_update_metadata(disk); 2319 } 2320 break; 2321 } 2322 default: 2323 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2324 break; 2325 } 2326 return (0); 2327 } 2328 #undef DISK_STATE_CHANGED 2329 2330 int 2331 g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2332 { 2333 struct g_provider *pp; 2334 u_char *buf; 2335 int error; 2336 2337 g_topology_assert(); 2338 2339 error = g_access(cp, 1, 0, 0); 2340 if (error != 0) 2341 return (error); 2342 pp = cp->provider; 2343 g_topology_unlock(); 2344 /* Metadata are stored on last sector. */ 2345 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2346 &error); 2347 g_topology_lock(); 2348 g_access(cp, -1, 0, 0); 2349 if (buf == NULL) { 2350 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2351 cp->provider->name, error); 2352 if (buf != NULL) 2353 g_free(buf); 2354 return (error); 2355 } 2356 2357 /* Decode metadata. */ 2358 error = mirror_metadata_decode(buf, md); 2359 g_free(buf); 2360 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2361 return (EINVAL); 2362 if (md->md_version > G_MIRROR_VERSION) { 2363 G_MIRROR_DEBUG(0, 2364 "Kernel module is too old to handle metadata from %s.", 2365 cp->provider->name); 2366 return (EINVAL); 2367 } 2368 if (error != 0) { 2369 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2370 cp->provider->name); 2371 return (error); 2372 } 2373 2374 return (0); 2375 } 2376 2377 static int 2378 g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2379 struct g_mirror_metadata *md) 2380 { 2381 2382 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2383 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2384 pp->name, md->md_did); 2385 return (EEXIST); 2386 } 2387 if (md->md_all != sc->sc_ndisks) { 2388 G_MIRROR_DEBUG(1, 2389 "Invalid '%s' field on disk %s (device %s), skipping.", 2390 "md_all", pp->name, sc->sc_name); 2391 return (EINVAL); 2392 } 2393 if (md->md_slice != sc->sc_slice) { 2394 G_MIRROR_DEBUG(1, 2395 "Invalid '%s' field on disk %s (device %s), skipping.", 2396 "md_slice", pp->name, sc->sc_name); 2397 return (EINVAL); 2398 } 2399 if (md->md_balance != sc->sc_balance) { 2400 G_MIRROR_DEBUG(1, 2401 "Invalid '%s' field on disk %s (device %s), skipping.", 2402 "md_balance", pp->name, sc->sc_name); 2403 return (EINVAL); 2404 } 2405 if (md->md_mediasize != sc->sc_mediasize) { 2406 G_MIRROR_DEBUG(1, 2407 "Invalid '%s' field on disk %s (device %s), skipping.", 2408 "md_mediasize", pp->name, sc->sc_name); 2409 return (EINVAL); 2410 } 2411 if (sc->sc_mediasize > pp->mediasize) { 2412 G_MIRROR_DEBUG(1, 2413 "Invalid size of disk %s (device %s), skipping.", pp->name, 2414 sc->sc_name); 2415 return (EINVAL); 2416 } 2417 if (md->md_sectorsize != sc->sc_sectorsize) { 2418 G_MIRROR_DEBUG(1, 2419 "Invalid '%s' field on disk %s (device %s), skipping.", 2420 "md_sectorsize", pp->name, sc->sc_name); 2421 return (EINVAL); 2422 } 2423 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2424 G_MIRROR_DEBUG(1, 2425 "Invalid sector size of disk %s (device %s), skipping.", 2426 pp->name, sc->sc_name); 2427 return (EINVAL); 2428 } 2429 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2430 G_MIRROR_DEBUG(1, 2431 "Invalid device flags on disk %s (device %s), skipping.", 2432 pp->name, sc->sc_name); 2433 return (EINVAL); 2434 } 2435 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2436 G_MIRROR_DEBUG(1, 2437 "Invalid disk flags on disk %s (device %s), skipping.", 2438 pp->name, sc->sc_name); 2439 return (EINVAL); 2440 } 2441 return (0); 2442 } 2443 2444 int 2445 g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2446 struct g_mirror_metadata *md) 2447 { 2448 struct g_mirror_disk *disk; 2449 int error; 2450 2451 g_topology_assert(); 2452 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2453 2454 error = g_mirror_check_metadata(sc, pp, md); 2455 if (error != 0) 2456 return (error); 2457 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2458 md->md_genid < sc->sc_genid) { 2459 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2460 pp->name, sc->sc_name); 2461 return (EINVAL); 2462 } 2463 disk = g_mirror_init_disk(sc, pp, md, &error); 2464 if (disk == NULL) 2465 return (error); 2466 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2467 G_MIRROR_EVENT_WAIT); 2468 if (error != 0) 2469 return (error); 2470 if (md->md_version < G_MIRROR_VERSION) { 2471 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2472 pp->name, md->md_version, G_MIRROR_VERSION); 2473 g_mirror_update_metadata(disk); 2474 } 2475 return (0); 2476 } 2477 2478 static int 2479 g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2480 { 2481 struct g_mirror_softc *sc; 2482 struct g_mirror_disk *disk; 2483 int dcr, dcw, dce; 2484 2485 g_topology_assert(); 2486 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2487 acw, ace); 2488 2489 dcr = pp->acr + acr; 2490 dcw = pp->acw + acw; 2491 dce = pp->ace + ace; 2492 2493 sc = pp->geom->softc; 2494 if (sc == NULL || LIST_EMPTY(&sc->sc_disks) || 2495 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 2496 if (acr <= 0 && acw <= 0 && ace <= 0) 2497 return (0); 2498 else 2499 return (ENXIO); 2500 } 2501 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2502 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2503 continue; 2504 /* 2505 * Mark disk as dirty on open and unmark on close. 2506 */ 2507 if (pp->acw == 0 && dcw > 0) { 2508 G_MIRROR_DEBUG(1, 2509 "Disk %s (device %s) marked as dirty.", 2510 g_mirror_get_diskname(disk), sc->sc_name); 2511 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2512 g_mirror_update_metadata(disk); 2513 } else if (pp->acw > 0 && dcw == 0) { 2514 G_MIRROR_DEBUG(1, 2515 "Disk %s (device %s) marked as clean.", 2516 g_mirror_get_diskname(disk), sc->sc_name); 2517 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2518 g_mirror_update_metadata(disk); 2519 } 2520 } 2521 return (0); 2522 } 2523 2524 static struct g_geom * 2525 g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2526 { 2527 struct g_mirror_softc *sc; 2528 struct g_geom *gp; 2529 int error, timeout; 2530 2531 g_topology_assert(); 2532 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2533 md->md_mid); 2534 2535 /* One disk is minimum. */ 2536 if (md->md_all < 1) 2537 return (NULL); 2538 /* 2539 * Action geom. 2540 */ 2541 gp = g_new_geomf(mp, "%s", md->md_name); 2542 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2543 gp->start = g_mirror_start; 2544 gp->orphan = g_mirror_orphan; 2545 gp->access = g_mirror_access; 2546 gp->dumpconf = g_mirror_dumpconf; 2547 2548 sc->sc_id = md->md_mid; 2549 sc->sc_slice = md->md_slice; 2550 sc->sc_balance = md->md_balance; 2551 sc->sc_mediasize = md->md_mediasize; 2552 sc->sc_sectorsize = md->md_sectorsize; 2553 sc->sc_ndisks = md->md_all; 2554 sc->sc_flags = md->md_mflags; 2555 sc->sc_bump_id = 0; 2556 sc->sc_idle = 0; 2557 bioq_init(&sc->sc_queue); 2558 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2559 LIST_INIT(&sc->sc_disks); 2560 TAILQ_INIT(&sc->sc_events); 2561 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2562 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2563 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2564 gp->softc = sc; 2565 sc->sc_geom = gp; 2566 sc->sc_provider = NULL; 2567 /* 2568 * Synchronization geom. 2569 */ 2570 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2571 gp->softc = sc; 2572 gp->orphan = g_mirror_orphan; 2573 sc->sc_sync.ds_geom = gp; 2574 sc->sc_sync.ds_ndisks = 0; 2575 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2576 "g_mirror %s", md->md_name); 2577 if (error != 0) { 2578 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2579 sc->sc_name); 2580 g_destroy_geom(sc->sc_sync.ds_geom); 2581 mtx_destroy(&sc->sc_events_mtx); 2582 mtx_destroy(&sc->sc_queue_mtx); 2583 g_destroy_geom(sc->sc_geom); 2584 free(sc, M_MIRROR); 2585 return (NULL); 2586 } 2587 2588 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2589 2590 sc->sc_rootmount = root_mount_hold("GMIRROR"); 2591 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 2592 /* 2593 * Run timeout. 2594 */ 2595 timeout = g_mirror_timeout * hz; 2596 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2597 return (sc->sc_geom); 2598 } 2599 2600 int 2601 g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2602 { 2603 struct g_provider *pp; 2604 2605 g_topology_assert(); 2606 2607 if (sc == NULL) 2608 return (ENXIO); 2609 pp = sc->sc_provider; 2610 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2611 if (force) { 2612 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 2613 "can't be definitely removed.", pp->name); 2614 } else { 2615 G_MIRROR_DEBUG(1, 2616 "Device %s is still open (r%dw%de%d).", pp->name, 2617 pp->acr, pp->acw, pp->ace); 2618 return (EBUSY); 2619 } 2620 } 2621 2622 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2623 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2624 g_topology_unlock(); 2625 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2626 mtx_lock(&sc->sc_queue_mtx); 2627 wakeup(sc); 2628 mtx_unlock(&sc->sc_queue_mtx); 2629 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2630 while (sc->sc_worker != NULL) 2631 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2632 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2633 g_topology_lock(); 2634 g_mirror_destroy_device(sc); 2635 free(sc, M_MIRROR); 2636 return (0); 2637 } 2638 2639 static void 2640 g_mirror_taste_orphan(struct g_consumer *cp) 2641 { 2642 2643 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2644 cp->provider->name)); 2645 } 2646 2647 static struct g_geom * 2648 g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2649 { 2650 struct g_mirror_metadata md; 2651 struct g_mirror_softc *sc; 2652 struct g_consumer *cp; 2653 struct g_geom *gp; 2654 int error; 2655 2656 g_topology_assert(); 2657 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2658 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2659 2660 gp = g_new_geomf(mp, "mirror:taste"); 2661 /* 2662 * This orphan function should be never called. 2663 */ 2664 gp->orphan = g_mirror_taste_orphan; 2665 cp = g_new_consumer(gp); 2666 g_attach(cp, pp); 2667 error = g_mirror_read_metadata(cp, &md); 2668 g_detach(cp); 2669 g_destroy_consumer(cp); 2670 g_destroy_geom(gp); 2671 if (error != 0) 2672 return (NULL); 2673 gp = NULL; 2674 2675 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2676 return (NULL); 2677 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 2678 return (NULL); 2679 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2680 G_MIRROR_DEBUG(0, 2681 "Device %s: provider %s marked as inactive, skipping.", 2682 md.md_name, pp->name); 2683 return (NULL); 2684 } 2685 if (g_mirror_debug >= 2) 2686 mirror_metadata_dump(&md); 2687 2688 /* 2689 * Let's check if device already exists. 2690 */ 2691 sc = NULL; 2692 LIST_FOREACH(gp, &mp->geom, geom) { 2693 sc = gp->softc; 2694 if (sc == NULL) 2695 continue; 2696 if (sc->sc_sync.ds_geom == gp) 2697 continue; 2698 if (strcmp(md.md_name, sc->sc_name) != 0) 2699 continue; 2700 if (md.md_mid != sc->sc_id) { 2701 G_MIRROR_DEBUG(0, "Device %s already configured.", 2702 sc->sc_name); 2703 return (NULL); 2704 } 2705 break; 2706 } 2707 if (gp == NULL) { 2708 gp = g_mirror_create(mp, &md); 2709 if (gp == NULL) { 2710 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2711 md.md_name); 2712 return (NULL); 2713 } 2714 sc = gp->softc; 2715 } 2716 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2717 error = g_mirror_add_disk(sc, pp, &md); 2718 if (error != 0) { 2719 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2720 pp->name, gp->name, error); 2721 if (LIST_EMPTY(&sc->sc_disks)) 2722 g_mirror_destroy(sc, 1); 2723 return (NULL); 2724 } 2725 return (gp); 2726 } 2727 2728 static int 2729 g_mirror_destroy_geom(struct gctl_req *req __unused, 2730 struct g_class *mp __unused, struct g_geom *gp) 2731 { 2732 2733 return (g_mirror_destroy(gp->softc, 0)); 2734 } 2735 2736 static void 2737 g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2738 struct g_consumer *cp, struct g_provider *pp) 2739 { 2740 struct g_mirror_softc *sc; 2741 2742 g_topology_assert(); 2743 2744 sc = gp->softc; 2745 if (sc == NULL) 2746 return; 2747 /* Skip synchronization geom. */ 2748 if (gp == sc->sc_sync.ds_geom) 2749 return; 2750 if (pp != NULL) { 2751 /* Nothing here. */ 2752 } else if (cp != NULL) { 2753 struct g_mirror_disk *disk; 2754 2755 disk = cp->private; 2756 if (disk == NULL) 2757 return; 2758 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2759 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2760 sbuf_printf(sb, "%s<Synchronized>", indent); 2761 if (disk->d_sync.ds_offset_done == 0) 2762 sbuf_printf(sb, "0%%"); 2763 else { 2764 sbuf_printf(sb, "%u%%", 2765 (u_int)((disk->d_sync.ds_offset_done * 100) / 2766 sc->sc_provider->mediasize)); 2767 } 2768 sbuf_printf(sb, "</Synchronized>\n"); 2769 } 2770 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2771 disk->d_sync.ds_syncid); 2772 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 2773 disk->d_genid); 2774 sbuf_printf(sb, "%s<Flags>", indent); 2775 if (disk->d_flags == 0) 2776 sbuf_printf(sb, "NONE"); 2777 else { 2778 int first = 1; 2779 2780 #define ADD_FLAG(flag, name) do { \ 2781 if ((disk->d_flags & (flag)) != 0) { \ 2782 if (!first) \ 2783 sbuf_printf(sb, ", "); \ 2784 else \ 2785 first = 0; \ 2786 sbuf_printf(sb, name); \ 2787 } \ 2788 } while (0) 2789 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2790 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2791 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2792 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2793 "SYNCHRONIZING"); 2794 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2795 #undef ADD_FLAG 2796 } 2797 sbuf_printf(sb, "</Flags>\n"); 2798 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2799 disk->d_priority); 2800 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2801 g_mirror_disk_state2str(disk->d_state)); 2802 } else { 2803 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2804 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2805 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 2806 sbuf_printf(sb, "%s<Flags>", indent); 2807 if (sc->sc_flags == 0) 2808 sbuf_printf(sb, "NONE"); 2809 else { 2810 int first = 1; 2811 2812 #define ADD_FLAG(flag, name) do { \ 2813 if ((sc->sc_flags & (flag)) != 0) { \ 2814 if (!first) \ 2815 sbuf_printf(sb, ", "); \ 2816 else \ 2817 first = 0; \ 2818 sbuf_printf(sb, name); \ 2819 } \ 2820 } while (0) 2821 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2822 #undef ADD_FLAG 2823 } 2824 sbuf_printf(sb, "</Flags>\n"); 2825 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2826 (u_int)sc->sc_slice); 2827 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2828 balance_name(sc->sc_balance)); 2829 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2830 sc->sc_ndisks); 2831 sbuf_printf(sb, "%s<State>", indent); 2832 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2833 sbuf_printf(sb, "%s", "STARTING"); 2834 else if (sc->sc_ndisks == 2835 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 2836 sbuf_printf(sb, "%s", "COMPLETE"); 2837 else 2838 sbuf_printf(sb, "%s", "DEGRADED"); 2839 sbuf_printf(sb, "</State>\n"); 2840 } 2841 } 2842 2843 static void 2844 g_mirror_shutdown(void *arg, int howto) 2845 { 2846 struct g_class *mp; 2847 struct g_geom *gp, *gp2; 2848 2849 mp = arg; 2850 DROP_GIANT(); 2851 g_topology_lock(); 2852 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2853 if (gp->softc == NULL) 2854 continue; 2855 g_mirror_destroy(gp->softc, 1); 2856 } 2857 g_topology_unlock(); 2858 PICKUP_GIANT(); 2859 #if 0 2860 tsleep(&gp, PRIBIO, "m:shutdown", hz * 20); 2861 #endif 2862 } 2863 2864 static void 2865 g_mirror_init(struct g_class *mp) 2866 { 2867 2868 g_mirror_ehtag = EVENTHANDLER_REGISTER(shutdown_post_sync, 2869 g_mirror_shutdown, mp, SHUTDOWN_PRI_FIRST); 2870 if (g_mirror_ehtag == NULL) 2871 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 2872 } 2873 2874 static void 2875 g_mirror_fini(struct g_class *mp) 2876 { 2877 2878 if (g_mirror_ehtag == NULL) 2879 return; 2880 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_ehtag); 2881 } 2882 2883 DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2884