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