1 /*- 2 * Copyright (c) 2004-2006 Pawel Jakub Dawidek <pjd@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 #include <sys/cdefs.h> 28 __FBSDID("$FreeBSD$"); 29 30 #include <sys/param.h> 31 #include <sys/systm.h> 32 #include <sys/fail.h> 33 #include <sys/kernel.h> 34 #include <sys/module.h> 35 #include <sys/limits.h> 36 #include <sys/lock.h> 37 #include <sys/mutex.h> 38 #include <sys/bio.h> 39 #include <sys/sbuf.h> 40 #include <sys/sysctl.h> 41 #include <sys/malloc.h> 42 #include <sys/eventhandler.h> 43 #include <vm/uma.h> 44 #include <geom/geom.h> 45 #include <sys/proc.h> 46 #include <sys/kthread.h> 47 #include <sys/sched.h> 48 #include <geom/mirror/g_mirror.h> 49 50 FEATURE(geom_mirror, "GEOM mirroring support"); 51 52 static MALLOC_DEFINE(M_MIRROR, "mirror_data", "GEOM_MIRROR Data"); 53 54 SYSCTL_DECL(_kern_geom); 55 static SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, 56 "GEOM_MIRROR stuff"); 57 u_int g_mirror_debug = 0; 58 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RWTUN, &g_mirror_debug, 0, 59 "Debug level"); 60 static u_int g_mirror_timeout = 4; 61 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RWTUN, &g_mirror_timeout, 62 0, "Time to wait on all mirror components"); 63 static u_int g_mirror_idletime = 5; 64 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RWTUN, 65 &g_mirror_idletime, 0, "Mark components as clean when idling"); 66 static u_int g_mirror_disconnect_on_failure = 1; 67 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN, 68 &g_mirror_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 69 static u_int g_mirror_syncreqs = 2; 70 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_requests, CTLFLAG_RDTUN, 71 &g_mirror_syncreqs, 0, "Parallel synchronization I/O requests."); 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_post_sync = NULL; 80 static int g_mirror_shutdown = 0; 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 g_resize_t g_mirror_resize; 86 static void g_mirror_init(struct g_class *mp); 87 static void g_mirror_fini(struct g_class *mp); 88 89 struct g_class g_mirror_class = { 90 .name = G_MIRROR_CLASS_NAME, 91 .version = G_VERSION, 92 .ctlreq = g_mirror_config, 93 .taste = g_mirror_taste, 94 .destroy_geom = g_mirror_destroy_geom, 95 .init = g_mirror_init, 96 .fini = g_mirror_fini, 97 .resize = g_mirror_resize 98 }; 99 100 101 static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 102 static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 103 static void g_mirror_update_device(struct g_mirror_softc *sc, bool force); 104 static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 105 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 106 static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 107 static void g_mirror_register_request(struct bio *bp); 108 static void g_mirror_sync_release(struct g_mirror_softc *sc); 109 110 111 static const char * 112 g_mirror_disk_state2str(int state) 113 { 114 115 switch (state) { 116 case G_MIRROR_DISK_STATE_NONE: 117 return ("NONE"); 118 case G_MIRROR_DISK_STATE_NEW: 119 return ("NEW"); 120 case G_MIRROR_DISK_STATE_ACTIVE: 121 return ("ACTIVE"); 122 case G_MIRROR_DISK_STATE_STALE: 123 return ("STALE"); 124 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 125 return ("SYNCHRONIZING"); 126 case G_MIRROR_DISK_STATE_DISCONNECTED: 127 return ("DISCONNECTED"); 128 case G_MIRROR_DISK_STATE_DESTROY: 129 return ("DESTROY"); 130 default: 131 return ("INVALID"); 132 } 133 } 134 135 static const char * 136 g_mirror_device_state2str(int state) 137 { 138 139 switch (state) { 140 case G_MIRROR_DEVICE_STATE_STARTING: 141 return ("STARTING"); 142 case G_MIRROR_DEVICE_STATE_RUNNING: 143 return ("RUNNING"); 144 default: 145 return ("INVALID"); 146 } 147 } 148 149 static const char * 150 g_mirror_get_diskname(struct g_mirror_disk *disk) 151 { 152 153 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 154 return ("[unknown]"); 155 return (disk->d_name); 156 } 157 158 /* 159 * --- Events handling functions --- 160 * Events in geom_mirror are used to maintain disks and device status 161 * from one thread to simplify locking. 162 */ 163 static void 164 g_mirror_event_free(struct g_mirror_event *ep) 165 { 166 167 free(ep, M_MIRROR); 168 } 169 170 int 171 g_mirror_event_send(void *arg, int state, int flags) 172 { 173 struct g_mirror_softc *sc; 174 struct g_mirror_disk *disk; 175 struct g_mirror_event *ep; 176 int error; 177 178 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 179 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 180 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 181 disk = NULL; 182 sc = arg; 183 } else { 184 disk = arg; 185 sc = disk->d_softc; 186 } 187 ep->e_disk = disk; 188 ep->e_state = state; 189 ep->e_flags = flags; 190 ep->e_error = 0; 191 mtx_lock(&sc->sc_events_mtx); 192 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 193 mtx_unlock(&sc->sc_events_mtx); 194 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 195 mtx_lock(&sc->sc_queue_mtx); 196 wakeup(sc); 197 mtx_unlock(&sc->sc_queue_mtx); 198 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 199 return (0); 200 sx_assert(&sc->sc_lock, SX_XLOCKED); 201 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 202 sx_xunlock(&sc->sc_lock); 203 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 204 mtx_lock(&sc->sc_events_mtx); 205 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 206 hz * 5); 207 } 208 error = ep->e_error; 209 g_mirror_event_free(ep); 210 sx_xlock(&sc->sc_lock); 211 return (error); 212 } 213 214 static struct g_mirror_event * 215 g_mirror_event_get(struct g_mirror_softc *sc) 216 { 217 struct g_mirror_event *ep; 218 219 mtx_lock(&sc->sc_events_mtx); 220 ep = TAILQ_FIRST(&sc->sc_events); 221 mtx_unlock(&sc->sc_events_mtx); 222 return (ep); 223 } 224 225 static void 226 g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 227 { 228 229 mtx_lock(&sc->sc_events_mtx); 230 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 231 mtx_unlock(&sc->sc_events_mtx); 232 } 233 234 static void 235 g_mirror_event_cancel(struct g_mirror_disk *disk) 236 { 237 struct g_mirror_softc *sc; 238 struct g_mirror_event *ep, *tmpep; 239 240 sc = disk->d_softc; 241 sx_assert(&sc->sc_lock, SX_XLOCKED); 242 243 mtx_lock(&sc->sc_events_mtx); 244 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 245 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 246 continue; 247 if (ep->e_disk != disk) 248 continue; 249 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 250 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 251 g_mirror_event_free(ep); 252 else { 253 ep->e_error = ECANCELED; 254 wakeup(ep); 255 } 256 } 257 mtx_unlock(&sc->sc_events_mtx); 258 } 259 260 /* 261 * Return the number of disks in given state. 262 * If state is equal to -1, count all connected disks. 263 */ 264 u_int 265 g_mirror_ndisks(struct g_mirror_softc *sc, int state) 266 { 267 struct g_mirror_disk *disk; 268 u_int n = 0; 269 270 sx_assert(&sc->sc_lock, SX_LOCKED); 271 272 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 273 if (state == -1 || disk->d_state == state) 274 n++; 275 } 276 return (n); 277 } 278 279 /* 280 * Find a disk in mirror by its disk ID. 281 */ 282 static struct g_mirror_disk * 283 g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 284 { 285 struct g_mirror_disk *disk; 286 287 sx_assert(&sc->sc_lock, SX_XLOCKED); 288 289 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 290 if (disk->d_id == id) 291 return (disk); 292 } 293 return (NULL); 294 } 295 296 static u_int 297 g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 298 { 299 struct bio *bp; 300 u_int nreqs = 0; 301 302 mtx_lock(&sc->sc_queue_mtx); 303 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 304 if (bp->bio_from == cp) 305 nreqs++; 306 } 307 mtx_unlock(&sc->sc_queue_mtx); 308 return (nreqs); 309 } 310 311 static int 312 g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 313 { 314 315 if (cp->index > 0) { 316 G_MIRROR_DEBUG(2, 317 "I/O requests for %s exist, can't destroy it now.", 318 cp->provider->name); 319 return (1); 320 } 321 if (g_mirror_nrequests(sc, cp) > 0) { 322 G_MIRROR_DEBUG(2, 323 "I/O requests for %s in queue, can't destroy it now.", 324 cp->provider->name); 325 return (1); 326 } 327 return (0); 328 } 329 330 static void 331 g_mirror_destroy_consumer(void *arg, int flags __unused) 332 { 333 struct g_consumer *cp; 334 335 g_topology_assert(); 336 337 cp = arg; 338 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 339 g_detach(cp); 340 g_destroy_consumer(cp); 341 } 342 343 static void 344 g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 345 { 346 struct g_provider *pp; 347 int retaste_wait; 348 349 g_topology_assert(); 350 351 cp->private = NULL; 352 if (g_mirror_is_busy(sc, cp)) 353 return; 354 pp = cp->provider; 355 retaste_wait = 0; 356 if (cp->acw == 1) { 357 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 358 retaste_wait = 1; 359 } 360 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 361 -cp->acw, -cp->ace, 0); 362 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 363 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 364 if (retaste_wait) { 365 /* 366 * After retaste event was send (inside g_access()), we can send 367 * event to detach and destroy consumer. 368 * A class, which has consumer to the given provider connected 369 * will not receive retaste event for the provider. 370 * This is the way how I ignore retaste events when I close 371 * consumers opened for write: I detach and destroy consumer 372 * after retaste event is sent. 373 */ 374 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 375 return; 376 } 377 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 378 g_detach(cp); 379 g_destroy_consumer(cp); 380 } 381 382 static int 383 g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 384 { 385 struct g_consumer *cp; 386 int error; 387 388 g_topology_assert_not(); 389 KASSERT(disk->d_consumer == NULL, 390 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 391 392 g_topology_lock(); 393 cp = g_new_consumer(disk->d_softc->sc_geom); 394 cp->flags |= G_CF_DIRECT_RECEIVE; 395 error = g_attach(cp, pp); 396 if (error != 0) { 397 g_destroy_consumer(cp); 398 g_topology_unlock(); 399 return (error); 400 } 401 error = g_access(cp, 1, 1, 1); 402 if (error != 0) { 403 g_detach(cp); 404 g_destroy_consumer(cp); 405 g_topology_unlock(); 406 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 407 pp->name, error); 408 return (error); 409 } 410 g_topology_unlock(); 411 disk->d_consumer = cp; 412 disk->d_consumer->private = disk; 413 disk->d_consumer->index = 0; 414 415 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 416 return (0); 417 } 418 419 static void 420 g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 421 { 422 423 g_topology_assert(); 424 425 if (cp == NULL) 426 return; 427 if (cp->provider != NULL) 428 g_mirror_kill_consumer(sc, cp); 429 else 430 g_destroy_consumer(cp); 431 } 432 433 /* 434 * Initialize disk. This means allocate memory, create consumer, attach it 435 * to the provider and open access (r1w1e1) to it. 436 */ 437 static struct g_mirror_disk * 438 g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 439 struct g_mirror_metadata *md, int *errorp) 440 { 441 struct g_mirror_disk *disk; 442 int i, error; 443 444 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 445 if (disk == NULL) { 446 error = ENOMEM; 447 goto fail; 448 } 449 disk->d_softc = sc; 450 error = g_mirror_connect_disk(disk, pp); 451 if (error != 0) 452 goto fail; 453 disk->d_id = md->md_did; 454 disk->d_state = G_MIRROR_DISK_STATE_NONE; 455 disk->d_priority = md->md_priority; 456 disk->d_flags = md->md_dflags; 457 error = g_getattr("GEOM::candelete", disk->d_consumer, &i); 458 if (error == 0 && i != 0) 459 disk->d_flags |= G_MIRROR_DISK_FLAG_CANDELETE; 460 if (md->md_provider[0] != '\0') 461 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 462 disk->d_sync.ds_consumer = NULL; 463 disk->d_sync.ds_offset = md->md_sync_offset; 464 disk->d_sync.ds_offset_done = md->md_sync_offset; 465 disk->d_genid = md->md_genid; 466 disk->d_sync.ds_syncid = md->md_syncid; 467 if (errorp != NULL) 468 *errorp = 0; 469 return (disk); 470 fail: 471 if (errorp != NULL) 472 *errorp = error; 473 if (disk != NULL) 474 free(disk, M_MIRROR); 475 return (NULL); 476 } 477 478 static void 479 g_mirror_destroy_disk(struct g_mirror_disk *disk) 480 { 481 struct g_mirror_softc *sc; 482 483 g_topology_assert_not(); 484 sc = disk->d_softc; 485 sx_assert(&sc->sc_lock, SX_XLOCKED); 486 487 LIST_REMOVE(disk, d_next); 488 g_mirror_event_cancel(disk); 489 if (sc->sc_hint == disk) 490 sc->sc_hint = NULL; 491 switch (disk->d_state) { 492 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 493 g_mirror_sync_stop(disk, 1); 494 /* FALLTHROUGH */ 495 case G_MIRROR_DISK_STATE_NEW: 496 case G_MIRROR_DISK_STATE_STALE: 497 case G_MIRROR_DISK_STATE_ACTIVE: 498 g_topology_lock(); 499 g_mirror_disconnect_consumer(sc, disk->d_consumer); 500 g_topology_unlock(); 501 free(disk, M_MIRROR); 502 break; 503 default: 504 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 505 g_mirror_get_diskname(disk), 506 g_mirror_disk_state2str(disk->d_state))); 507 } 508 } 509 510 static void 511 g_mirror_destroy_device(struct g_mirror_softc *sc) 512 { 513 struct g_mirror_disk *disk; 514 struct g_mirror_event *ep; 515 struct g_geom *gp; 516 struct g_consumer *cp, *tmpcp; 517 518 g_topology_assert_not(); 519 sx_assert(&sc->sc_lock, SX_XLOCKED); 520 521 gp = sc->sc_geom; 522 if (sc->sc_provider != NULL) 523 g_mirror_destroy_provider(sc); 524 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 525 disk = LIST_FIRST(&sc->sc_disks)) { 526 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 527 g_mirror_update_metadata(disk); 528 g_mirror_destroy_disk(disk); 529 } 530 while ((ep = g_mirror_event_get(sc)) != NULL) { 531 g_mirror_event_remove(sc, ep); 532 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 533 g_mirror_event_free(ep); 534 else { 535 ep->e_error = ECANCELED; 536 ep->e_flags |= G_MIRROR_EVENT_DONE; 537 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 538 mtx_lock(&sc->sc_events_mtx); 539 wakeup(ep); 540 mtx_unlock(&sc->sc_events_mtx); 541 } 542 } 543 callout_drain(&sc->sc_callout); 544 545 g_topology_lock(); 546 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 547 g_mirror_disconnect_consumer(sc, cp); 548 } 549 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 550 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 551 g_wither_geom(gp, ENXIO); 552 g_topology_unlock(); 553 mtx_destroy(&sc->sc_queue_mtx); 554 mtx_destroy(&sc->sc_events_mtx); 555 mtx_destroy(&sc->sc_done_mtx); 556 sx_xunlock(&sc->sc_lock); 557 sx_destroy(&sc->sc_lock); 558 } 559 560 static void 561 g_mirror_orphan(struct g_consumer *cp) 562 { 563 struct g_mirror_disk *disk; 564 565 g_topology_assert(); 566 567 disk = cp->private; 568 if (disk == NULL) 569 return; 570 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 571 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 572 G_MIRROR_EVENT_DONTWAIT); 573 } 574 575 /* 576 * Function should return the next active disk on the list. 577 * It is possible that it will be the same disk as given. 578 * If there are no active disks on list, NULL is returned. 579 */ 580 static __inline struct g_mirror_disk * 581 g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 582 { 583 struct g_mirror_disk *dp; 584 585 for (dp = LIST_NEXT(disk, d_next); dp != disk; 586 dp = LIST_NEXT(dp, d_next)) { 587 if (dp == NULL) 588 dp = LIST_FIRST(&sc->sc_disks); 589 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 590 break; 591 } 592 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 593 return (NULL); 594 return (dp); 595 } 596 597 static struct g_mirror_disk * 598 g_mirror_get_disk(struct g_mirror_softc *sc) 599 { 600 struct g_mirror_disk *disk; 601 602 if (sc->sc_hint == NULL) { 603 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 604 if (sc->sc_hint == NULL) 605 return (NULL); 606 } 607 disk = sc->sc_hint; 608 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 609 disk = g_mirror_find_next(sc, disk); 610 if (disk == NULL) 611 return (NULL); 612 } 613 sc->sc_hint = g_mirror_find_next(sc, disk); 614 return (disk); 615 } 616 617 static int 618 g_mirror_write_metadata(struct g_mirror_disk *disk, 619 struct g_mirror_metadata *md) 620 { 621 struct g_mirror_softc *sc; 622 struct g_consumer *cp; 623 off_t offset, length; 624 u_char *sector; 625 int error = 0; 626 627 g_topology_assert_not(); 628 sc = disk->d_softc; 629 sx_assert(&sc->sc_lock, SX_LOCKED); 630 631 cp = disk->d_consumer; 632 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 633 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 634 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 635 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 636 cp->acw, cp->ace)); 637 length = cp->provider->sectorsize; 638 offset = cp->provider->mediasize - length; 639 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 640 if (md != NULL && 641 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) { 642 /* 643 * Handle the case, when the size of parent provider reduced. 644 */ 645 if (offset < md->md_mediasize) 646 error = ENOSPC; 647 else 648 mirror_metadata_encode(md, sector); 649 } 650 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_metadata_write, error); 651 if (error == 0) 652 error = g_write_data(cp, offset, sector, length); 653 free(sector, M_MIRROR); 654 if (error != 0) { 655 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 656 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 657 G_MIRROR_DEBUG(0, "Cannot write metadata on %s " 658 "(device=%s, error=%d).", 659 g_mirror_get_diskname(disk), sc->sc_name, error); 660 } else { 661 G_MIRROR_DEBUG(1, "Cannot write metadata on %s " 662 "(device=%s, error=%d).", 663 g_mirror_get_diskname(disk), sc->sc_name, error); 664 } 665 if (g_mirror_disconnect_on_failure && 666 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 667 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 668 g_mirror_event_send(disk, 669 G_MIRROR_DISK_STATE_DISCONNECTED, 670 G_MIRROR_EVENT_DONTWAIT); 671 } 672 } 673 return (error); 674 } 675 676 static int 677 g_mirror_clear_metadata(struct g_mirror_disk *disk) 678 { 679 int error; 680 681 g_topology_assert_not(); 682 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 683 684 error = g_mirror_write_metadata(disk, NULL); 685 if (error == 0) { 686 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 687 g_mirror_get_diskname(disk)); 688 } else { 689 G_MIRROR_DEBUG(0, 690 "Cannot clear metadata on disk %s (error=%d).", 691 g_mirror_get_diskname(disk), error); 692 } 693 return (error); 694 } 695 696 void 697 g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 698 struct g_mirror_metadata *md) 699 { 700 701 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 702 md->md_version = G_MIRROR_VERSION; 703 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 704 md->md_mid = sc->sc_id; 705 md->md_all = sc->sc_ndisks; 706 md->md_slice = sc->sc_slice; 707 md->md_balance = sc->sc_balance; 708 md->md_genid = sc->sc_genid; 709 md->md_mediasize = sc->sc_mediasize; 710 md->md_sectorsize = sc->sc_sectorsize; 711 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 712 bzero(md->md_provider, sizeof(md->md_provider)); 713 if (disk == NULL) { 714 md->md_did = arc4random(); 715 md->md_priority = 0; 716 md->md_syncid = 0; 717 md->md_dflags = 0; 718 md->md_sync_offset = 0; 719 md->md_provsize = 0; 720 } else { 721 md->md_did = disk->d_id; 722 md->md_priority = disk->d_priority; 723 md->md_syncid = disk->d_sync.ds_syncid; 724 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 725 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 726 md->md_sync_offset = disk->d_sync.ds_offset_done; 727 else 728 md->md_sync_offset = 0; 729 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 730 strlcpy(md->md_provider, 731 disk->d_consumer->provider->name, 732 sizeof(md->md_provider)); 733 } 734 md->md_provsize = disk->d_consumer->provider->mediasize; 735 } 736 } 737 738 void 739 g_mirror_update_metadata(struct g_mirror_disk *disk) 740 { 741 struct g_mirror_softc *sc; 742 struct g_mirror_metadata md; 743 int error; 744 745 g_topology_assert_not(); 746 sc = disk->d_softc; 747 sx_assert(&sc->sc_lock, SX_LOCKED); 748 749 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) 750 g_mirror_fill_metadata(sc, disk, &md); 751 error = g_mirror_write_metadata(disk, &md); 752 if (error == 0) { 753 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 754 g_mirror_get_diskname(disk)); 755 } else { 756 G_MIRROR_DEBUG(0, 757 "Cannot update metadata on disk %s (error=%d).", 758 g_mirror_get_diskname(disk), error); 759 } 760 } 761 762 static void 763 g_mirror_bump_syncid(struct g_mirror_softc *sc) 764 { 765 struct g_mirror_disk *disk; 766 767 g_topology_assert_not(); 768 sx_assert(&sc->sc_lock, SX_XLOCKED); 769 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 770 ("%s called with no active disks (device=%s).", __func__, 771 sc->sc_name)); 772 773 sc->sc_syncid++; 774 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 775 sc->sc_syncid); 776 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 777 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 778 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 779 disk->d_sync.ds_syncid = sc->sc_syncid; 780 g_mirror_update_metadata(disk); 781 } 782 } 783 } 784 785 static void 786 g_mirror_bump_genid(struct g_mirror_softc *sc) 787 { 788 struct g_mirror_disk *disk; 789 790 g_topology_assert_not(); 791 sx_assert(&sc->sc_lock, SX_XLOCKED); 792 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 793 ("%s called with no active disks (device=%s).", __func__, 794 sc->sc_name)); 795 796 sc->sc_genid++; 797 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 798 sc->sc_genid); 799 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 800 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 801 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 802 disk->d_genid = sc->sc_genid; 803 g_mirror_update_metadata(disk); 804 } 805 } 806 } 807 808 static int 809 g_mirror_idle(struct g_mirror_softc *sc, int acw) 810 { 811 struct g_mirror_disk *disk; 812 int timeout; 813 814 g_topology_assert_not(); 815 sx_assert(&sc->sc_lock, SX_XLOCKED); 816 817 if (sc->sc_provider == NULL) 818 return (0); 819 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 820 return (0); 821 if (sc->sc_idle) 822 return (0); 823 if (sc->sc_writes > 0) 824 return (0); 825 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 826 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); 827 if (!g_mirror_shutdown && timeout > 0) 828 return (timeout); 829 } 830 sc->sc_idle = 1; 831 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 832 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 833 continue; 834 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 835 g_mirror_get_diskname(disk), sc->sc_name); 836 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 837 g_mirror_update_metadata(disk); 838 } 839 return (0); 840 } 841 842 static void 843 g_mirror_unidle(struct g_mirror_softc *sc) 844 { 845 struct g_mirror_disk *disk; 846 847 g_topology_assert_not(); 848 sx_assert(&sc->sc_lock, SX_XLOCKED); 849 850 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 851 return; 852 sc->sc_idle = 0; 853 sc->sc_last_write = time_uptime; 854 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 855 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 856 continue; 857 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 858 g_mirror_get_diskname(disk), sc->sc_name); 859 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 860 g_mirror_update_metadata(disk); 861 } 862 } 863 864 static void 865 g_mirror_flush_done(struct bio *bp) 866 { 867 struct g_mirror_softc *sc; 868 struct bio *pbp; 869 870 pbp = bp->bio_parent; 871 sc = pbp->bio_to->geom->softc; 872 mtx_lock(&sc->sc_done_mtx); 873 if (pbp->bio_error == 0) 874 pbp->bio_error = bp->bio_error; 875 pbp->bio_completed += bp->bio_completed; 876 pbp->bio_inbed++; 877 if (pbp->bio_children == pbp->bio_inbed) { 878 mtx_unlock(&sc->sc_done_mtx); 879 g_io_deliver(pbp, pbp->bio_error); 880 } else 881 mtx_unlock(&sc->sc_done_mtx); 882 g_destroy_bio(bp); 883 } 884 885 static void 886 g_mirror_done(struct bio *bp) 887 { 888 struct g_mirror_softc *sc; 889 890 sc = bp->bio_from->geom->softc; 891 bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR; 892 mtx_lock(&sc->sc_queue_mtx); 893 bioq_insert_tail(&sc->sc_queue, bp); 894 mtx_unlock(&sc->sc_queue_mtx); 895 wakeup(sc); 896 } 897 898 static void 899 g_mirror_regular_request(struct bio *bp) 900 { 901 struct g_mirror_softc *sc; 902 struct g_mirror_disk *disk; 903 struct bio *pbp; 904 905 g_topology_assert_not(); 906 907 pbp = bp->bio_parent; 908 sc = pbp->bio_to->geom->softc; 909 bp->bio_from->index--; 910 if (bp->bio_cmd == BIO_WRITE) 911 sc->sc_writes--; 912 disk = bp->bio_from->private; 913 if (disk == NULL) { 914 g_topology_lock(); 915 g_mirror_kill_consumer(sc, bp->bio_from); 916 g_topology_unlock(); 917 } 918 919 if (bp->bio_cmd == BIO_READ) 920 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_read, 921 bp->bio_error); 922 else if (bp->bio_cmd == BIO_WRITE) 923 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_write, 924 bp->bio_error); 925 926 pbp->bio_inbed++; 927 KASSERT(pbp->bio_inbed <= pbp->bio_children, 928 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 929 pbp->bio_children)); 930 if (bp->bio_error == 0 && pbp->bio_error == 0) { 931 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 932 g_destroy_bio(bp); 933 if (pbp->bio_children == pbp->bio_inbed) { 934 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 935 pbp->bio_completed = pbp->bio_length; 936 if (pbp->bio_cmd == BIO_WRITE || 937 pbp->bio_cmd == BIO_DELETE) { 938 bioq_remove(&sc->sc_inflight, pbp); 939 /* Release delayed sync requests if possible. */ 940 g_mirror_sync_release(sc); 941 } 942 g_io_deliver(pbp, pbp->bio_error); 943 } 944 return; 945 } else if (bp->bio_error != 0) { 946 if (pbp->bio_error == 0) 947 pbp->bio_error = bp->bio_error; 948 if (disk != NULL) { 949 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 950 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 951 G_MIRROR_LOGREQ(0, bp, 952 "Request failed (error=%d).", 953 bp->bio_error); 954 } else { 955 G_MIRROR_LOGREQ(1, bp, 956 "Request failed (error=%d).", 957 bp->bio_error); 958 } 959 if (g_mirror_disconnect_on_failure && 960 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) 961 { 962 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 963 g_mirror_event_send(disk, 964 G_MIRROR_DISK_STATE_DISCONNECTED, 965 G_MIRROR_EVENT_DONTWAIT); 966 } 967 } 968 switch (pbp->bio_cmd) { 969 case BIO_DELETE: 970 case BIO_WRITE: 971 pbp->bio_inbed--; 972 pbp->bio_children--; 973 break; 974 } 975 } 976 g_destroy_bio(bp); 977 978 switch (pbp->bio_cmd) { 979 case BIO_READ: 980 if (pbp->bio_inbed < pbp->bio_children) 981 break; 982 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 983 g_io_deliver(pbp, pbp->bio_error); 984 else { 985 pbp->bio_error = 0; 986 mtx_lock(&sc->sc_queue_mtx); 987 bioq_insert_tail(&sc->sc_queue, pbp); 988 mtx_unlock(&sc->sc_queue_mtx); 989 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 990 wakeup(sc); 991 } 992 break; 993 case BIO_DELETE: 994 case BIO_WRITE: 995 if (pbp->bio_children == 0) { 996 /* 997 * All requests failed. 998 */ 999 } else if (pbp->bio_inbed < pbp->bio_children) { 1000 /* Do nothing. */ 1001 break; 1002 } else if (pbp->bio_children == pbp->bio_inbed) { 1003 /* Some requests succeeded. */ 1004 pbp->bio_error = 0; 1005 pbp->bio_completed = pbp->bio_length; 1006 } 1007 bioq_remove(&sc->sc_inflight, pbp); 1008 /* Release delayed sync requests if possible. */ 1009 g_mirror_sync_release(sc); 1010 g_io_deliver(pbp, pbp->bio_error); 1011 break; 1012 default: 1013 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 1014 break; 1015 } 1016 } 1017 1018 static void 1019 g_mirror_sync_done(struct bio *bp) 1020 { 1021 struct g_mirror_softc *sc; 1022 1023 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1024 sc = bp->bio_from->geom->softc; 1025 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; 1026 mtx_lock(&sc->sc_queue_mtx); 1027 bioq_insert_tail(&sc->sc_queue, bp); 1028 mtx_unlock(&sc->sc_queue_mtx); 1029 wakeup(sc); 1030 } 1031 1032 static void 1033 g_mirror_candelete(struct bio *bp) 1034 { 1035 struct g_mirror_softc *sc; 1036 struct g_mirror_disk *disk; 1037 int *val; 1038 1039 sc = bp->bio_to->geom->softc; 1040 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1041 if (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) 1042 break; 1043 } 1044 val = (int *)bp->bio_data; 1045 *val = (disk != NULL); 1046 g_io_deliver(bp, 0); 1047 } 1048 1049 static void 1050 g_mirror_kernel_dump(struct bio *bp) 1051 { 1052 struct g_mirror_softc *sc; 1053 struct g_mirror_disk *disk; 1054 struct bio *cbp; 1055 struct g_kerneldump *gkd; 1056 1057 /* 1058 * We configure dumping to the first component, because this component 1059 * will be used for reading with 'prefer' balance algorithm. 1060 * If the component with the highest priority is currently disconnected 1061 * we will not be able to read the dump after the reboot if it will be 1062 * connected and synchronized later. Can we do something better? 1063 */ 1064 sc = bp->bio_to->geom->softc; 1065 disk = LIST_FIRST(&sc->sc_disks); 1066 1067 gkd = (struct g_kerneldump *)bp->bio_data; 1068 if (gkd->length > bp->bio_to->mediasize) 1069 gkd->length = bp->bio_to->mediasize; 1070 cbp = g_clone_bio(bp); 1071 if (cbp == NULL) { 1072 g_io_deliver(bp, ENOMEM); 1073 return; 1074 } 1075 cbp->bio_done = g_std_done; 1076 g_io_request(cbp, disk->d_consumer); 1077 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1078 g_mirror_get_diskname(disk)); 1079 } 1080 1081 static void 1082 g_mirror_flush(struct g_mirror_softc *sc, struct bio *bp) 1083 { 1084 struct bio_queue_head queue; 1085 struct g_mirror_disk *disk; 1086 struct g_consumer *cp; 1087 struct bio *cbp; 1088 1089 bioq_init(&queue); 1090 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1091 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1092 continue; 1093 cbp = g_clone_bio(bp); 1094 if (cbp == NULL) { 1095 while ((cbp = bioq_takefirst(&queue)) != NULL) 1096 g_destroy_bio(cbp); 1097 if (bp->bio_error == 0) 1098 bp->bio_error = ENOMEM; 1099 g_io_deliver(bp, bp->bio_error); 1100 return; 1101 } 1102 bioq_insert_tail(&queue, cbp); 1103 cbp->bio_done = g_mirror_flush_done; 1104 cbp->bio_caller1 = disk; 1105 cbp->bio_to = disk->d_consumer->provider; 1106 } 1107 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1108 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1109 disk = cbp->bio_caller1; 1110 cbp->bio_caller1 = NULL; 1111 cp = disk->d_consumer; 1112 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1113 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1114 cp->acr, cp->acw, cp->ace)); 1115 g_io_request(cbp, disk->d_consumer); 1116 } 1117 } 1118 1119 static void 1120 g_mirror_start(struct bio *bp) 1121 { 1122 struct g_mirror_softc *sc; 1123 1124 sc = bp->bio_to->geom->softc; 1125 /* 1126 * If sc == NULL or there are no valid disks, provider's error 1127 * should be set and g_mirror_start() should not be called at all. 1128 */ 1129 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1130 ("Provider's error should be set (error=%d)(mirror=%s).", 1131 bp->bio_to->error, bp->bio_to->name)); 1132 G_MIRROR_LOGREQ(3, bp, "Request received."); 1133 1134 switch (bp->bio_cmd) { 1135 case BIO_READ: 1136 case BIO_WRITE: 1137 case BIO_DELETE: 1138 break; 1139 case BIO_FLUSH: 1140 g_mirror_flush(sc, bp); 1141 return; 1142 case BIO_GETATTR: 1143 if (!strcmp(bp->bio_attribute, "GEOM::candelete")) { 1144 g_mirror_candelete(bp); 1145 return; 1146 } else if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1147 g_mirror_kernel_dump(bp); 1148 return; 1149 } 1150 /* FALLTHROUGH */ 1151 default: 1152 g_io_deliver(bp, EOPNOTSUPP); 1153 return; 1154 } 1155 mtx_lock(&sc->sc_queue_mtx); 1156 bioq_insert_tail(&sc->sc_queue, bp); 1157 mtx_unlock(&sc->sc_queue_mtx); 1158 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1159 wakeup(sc); 1160 } 1161 1162 /* 1163 * Return TRUE if the given request is colliding with a in-progress 1164 * synchronization request. 1165 */ 1166 static int 1167 g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1168 { 1169 struct g_mirror_disk *disk; 1170 struct bio *sbp; 1171 off_t rstart, rend, sstart, send; 1172 u_int i; 1173 1174 if (sc->sc_sync.ds_ndisks == 0) 1175 return (0); 1176 rstart = bp->bio_offset; 1177 rend = bp->bio_offset + bp->bio_length; 1178 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1179 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1180 continue; 1181 for (i = 0; i < g_mirror_syncreqs; i++) { 1182 sbp = disk->d_sync.ds_bios[i]; 1183 if (sbp == NULL) 1184 continue; 1185 sstart = sbp->bio_offset; 1186 send = sbp->bio_offset + sbp->bio_length; 1187 if (rend > sstart && rstart < send) 1188 return (1); 1189 } 1190 } 1191 return (0); 1192 } 1193 1194 /* 1195 * Return TRUE if the given sync request is colliding with a in-progress regular 1196 * request. 1197 */ 1198 static int 1199 g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1200 { 1201 off_t rstart, rend, sstart, send; 1202 struct bio *bp; 1203 1204 if (sc->sc_sync.ds_ndisks == 0) 1205 return (0); 1206 sstart = sbp->bio_offset; 1207 send = sbp->bio_offset + sbp->bio_length; 1208 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { 1209 rstart = bp->bio_offset; 1210 rend = bp->bio_offset + bp->bio_length; 1211 if (rend > sstart && rstart < send) 1212 return (1); 1213 } 1214 return (0); 1215 } 1216 1217 /* 1218 * Puts request onto delayed queue. 1219 */ 1220 static void 1221 g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1222 { 1223 1224 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1225 bioq_insert_head(&sc->sc_regular_delayed, bp); 1226 } 1227 1228 /* 1229 * Puts synchronization request onto delayed queue. 1230 */ 1231 static void 1232 g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1233 { 1234 1235 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1236 bioq_insert_tail(&sc->sc_sync_delayed, bp); 1237 } 1238 1239 /* 1240 * Releases delayed regular requests which don't collide anymore with sync 1241 * requests. 1242 */ 1243 static void 1244 g_mirror_regular_release(struct g_mirror_softc *sc) 1245 { 1246 struct bio *bp, *bp2; 1247 1248 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { 1249 if (g_mirror_sync_collision(sc, bp)) 1250 continue; 1251 bioq_remove(&sc->sc_regular_delayed, bp); 1252 G_MIRROR_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1253 mtx_lock(&sc->sc_queue_mtx); 1254 bioq_insert_head(&sc->sc_queue, bp); 1255 mtx_unlock(&sc->sc_queue_mtx); 1256 } 1257 } 1258 1259 /* 1260 * Releases delayed sync requests which don't collide anymore with regular 1261 * requests. 1262 */ 1263 static void 1264 g_mirror_sync_release(struct g_mirror_softc *sc) 1265 { 1266 struct bio *bp, *bp2; 1267 1268 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { 1269 if (g_mirror_regular_collision(sc, bp)) 1270 continue; 1271 bioq_remove(&sc->sc_sync_delayed, bp); 1272 G_MIRROR_LOGREQ(2, bp, 1273 "Releasing delayed synchronization request."); 1274 g_io_request(bp, bp->bio_from); 1275 } 1276 } 1277 1278 /* 1279 * Free a synchronization request and clear its slot in the array. 1280 */ 1281 static void 1282 g_mirror_sync_request_free(struct g_mirror_disk *disk, struct bio *bp) 1283 { 1284 int i; 1285 1286 if (disk != NULL && disk->d_sync.ds_bios != NULL) { 1287 i = (int)(uintptr_t)bp->bio_caller1; 1288 disk->d_sync.ds_bios[i] = NULL; 1289 } 1290 free(bp->bio_data, M_MIRROR); 1291 g_destroy_bio(bp); 1292 } 1293 1294 /* 1295 * Handle synchronization requests. 1296 * Every synchronization request is two-steps process: first, READ request is 1297 * send to active provider and then WRITE request (with read data) to the provider 1298 * being synchronized. When WRITE is finished, new synchronization request is 1299 * send. 1300 */ 1301 static void 1302 g_mirror_sync_request(struct bio *bp) 1303 { 1304 struct g_mirror_softc *sc; 1305 struct g_mirror_disk *disk; 1306 struct g_mirror_disk_sync *sync; 1307 1308 bp->bio_from->index--; 1309 sc = bp->bio_from->geom->softc; 1310 disk = bp->bio_from->private; 1311 if (disk == NULL) { 1312 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1313 g_topology_lock(); 1314 g_mirror_kill_consumer(sc, bp->bio_from); 1315 g_topology_unlock(); 1316 g_mirror_sync_request_free(NULL, bp); 1317 sx_xlock(&sc->sc_lock); 1318 return; 1319 } 1320 1321 /* 1322 * Synchronization request. 1323 */ 1324 switch (bp->bio_cmd) { 1325 case BIO_READ: 1326 { 1327 struct g_consumer *cp; 1328 1329 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_read, 1330 bp->bio_error); 1331 1332 if (bp->bio_error != 0) { 1333 G_MIRROR_LOGREQ(0, bp, 1334 "Synchronization request failed (error=%d).", 1335 bp->bio_error); 1336 g_mirror_sync_request_free(disk, bp); 1337 return; 1338 } 1339 G_MIRROR_LOGREQ(3, bp, 1340 "Synchronization request half-finished."); 1341 bp->bio_cmd = BIO_WRITE; 1342 bp->bio_cflags = 0; 1343 cp = disk->d_consumer; 1344 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1345 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1346 cp->acr, cp->acw, cp->ace)); 1347 cp->index++; 1348 g_io_request(bp, cp); 1349 return; 1350 } 1351 case BIO_WRITE: 1352 { 1353 off_t offset; 1354 void *data; 1355 int i; 1356 1357 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_write, 1358 bp->bio_error); 1359 1360 if (bp->bio_error != 0) { 1361 G_MIRROR_LOGREQ(0, bp, 1362 "Synchronization request failed (error=%d).", 1363 bp->bio_error); 1364 g_mirror_sync_request_free(disk, bp); 1365 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1366 g_mirror_event_send(disk, 1367 G_MIRROR_DISK_STATE_DISCONNECTED, 1368 G_MIRROR_EVENT_DONTWAIT); 1369 return; 1370 } 1371 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1372 sync = &disk->d_sync; 1373 if (sync->ds_offset >= sc->sc_mediasize || 1374 sync->ds_consumer == NULL || 1375 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1376 /* Don't send more synchronization requests. */ 1377 sync->ds_inflight--; 1378 g_mirror_sync_request_free(disk, bp); 1379 if (sync->ds_inflight > 0) 1380 return; 1381 if (sync->ds_consumer == NULL || 1382 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1383 return; 1384 } 1385 /* Disk up-to-date, activate it. */ 1386 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1387 G_MIRROR_EVENT_DONTWAIT); 1388 return; 1389 } 1390 1391 /* Send next synchronization request. */ 1392 data = bp->bio_data; 1393 g_reset_bio(bp); 1394 bp->bio_cmd = BIO_READ; 1395 bp->bio_offset = sync->ds_offset; 1396 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1397 sync->ds_offset += bp->bio_length; 1398 bp->bio_done = g_mirror_sync_done; 1399 bp->bio_data = data; 1400 bp->bio_from = sync->ds_consumer; 1401 bp->bio_to = sc->sc_provider; 1402 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1403 sync->ds_consumer->index++; 1404 /* 1405 * Delay the request if it is colliding with a regular request. 1406 */ 1407 if (g_mirror_regular_collision(sc, bp)) 1408 g_mirror_sync_delay(sc, bp); 1409 else 1410 g_io_request(bp, sync->ds_consumer); 1411 1412 /* Release delayed requests if possible. */ 1413 g_mirror_regular_release(sc); 1414 1415 /* Find the smallest offset */ 1416 offset = sc->sc_mediasize; 1417 for (i = 0; i < g_mirror_syncreqs; i++) { 1418 bp = sync->ds_bios[i]; 1419 if (bp->bio_offset < offset) 1420 offset = bp->bio_offset; 1421 } 1422 if (sync->ds_offset_done + (MAXPHYS * 100) < offset) { 1423 /* Update offset_done on every 100 blocks. */ 1424 sync->ds_offset_done = offset; 1425 g_mirror_update_metadata(disk); 1426 } 1427 return; 1428 } 1429 default: 1430 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1431 bp->bio_cmd, sc->sc_name)); 1432 break; 1433 } 1434 } 1435 1436 static void 1437 g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1438 { 1439 struct g_mirror_disk *disk; 1440 struct g_consumer *cp; 1441 struct bio *cbp; 1442 1443 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1444 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1445 break; 1446 } 1447 if (disk == NULL) { 1448 if (bp->bio_error == 0) 1449 bp->bio_error = ENXIO; 1450 g_io_deliver(bp, bp->bio_error); 1451 return; 1452 } 1453 cbp = g_clone_bio(bp); 1454 if (cbp == NULL) { 1455 if (bp->bio_error == 0) 1456 bp->bio_error = ENOMEM; 1457 g_io_deliver(bp, bp->bio_error); 1458 return; 1459 } 1460 /* 1461 * Fill in the component buf structure. 1462 */ 1463 cp = disk->d_consumer; 1464 cbp->bio_done = g_mirror_done; 1465 cbp->bio_to = cp->provider; 1466 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1467 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1468 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1469 cp->acw, cp->ace)); 1470 cp->index++; 1471 g_io_request(cbp, cp); 1472 } 1473 1474 static void 1475 g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1476 { 1477 struct g_mirror_disk *disk; 1478 struct g_consumer *cp; 1479 struct bio *cbp; 1480 1481 disk = g_mirror_get_disk(sc); 1482 if (disk == NULL) { 1483 if (bp->bio_error == 0) 1484 bp->bio_error = ENXIO; 1485 g_io_deliver(bp, bp->bio_error); 1486 return; 1487 } 1488 cbp = g_clone_bio(bp); 1489 if (cbp == NULL) { 1490 if (bp->bio_error == 0) 1491 bp->bio_error = ENOMEM; 1492 g_io_deliver(bp, bp->bio_error); 1493 return; 1494 } 1495 /* 1496 * Fill in the component buf structure. 1497 */ 1498 cp = disk->d_consumer; 1499 cbp->bio_done = g_mirror_done; 1500 cbp->bio_to = cp->provider; 1501 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1502 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1503 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1504 cp->acw, cp->ace)); 1505 cp->index++; 1506 g_io_request(cbp, cp); 1507 } 1508 1509 #define TRACK_SIZE (1 * 1024 * 1024) 1510 #define LOAD_SCALE 256 1511 #define ABS(x) (((x) >= 0) ? (x) : (-(x))) 1512 1513 static void 1514 g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1515 { 1516 struct g_mirror_disk *disk, *dp; 1517 struct g_consumer *cp; 1518 struct bio *cbp; 1519 int prio, best; 1520 1521 /* Find a disk with the smallest load. */ 1522 disk = NULL; 1523 best = INT_MAX; 1524 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1525 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1526 continue; 1527 prio = dp->load; 1528 /* If disk head is precisely in position - highly prefer it. */ 1529 if (dp->d_last_offset == bp->bio_offset) 1530 prio -= 2 * LOAD_SCALE; 1531 else 1532 /* If disk head is close to position - prefer it. */ 1533 if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE) 1534 prio -= 1 * LOAD_SCALE; 1535 if (prio <= best) { 1536 disk = dp; 1537 best = prio; 1538 } 1539 } 1540 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1541 cbp = g_clone_bio(bp); 1542 if (cbp == NULL) { 1543 if (bp->bio_error == 0) 1544 bp->bio_error = ENOMEM; 1545 g_io_deliver(bp, bp->bio_error); 1546 return; 1547 } 1548 /* 1549 * Fill in the component buf structure. 1550 */ 1551 cp = disk->d_consumer; 1552 cbp->bio_done = g_mirror_done; 1553 cbp->bio_to = cp->provider; 1554 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1555 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1556 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1557 cp->acw, cp->ace)); 1558 cp->index++; 1559 /* Remember last head position */ 1560 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1561 /* Update loads. */ 1562 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1563 dp->load = (dp->d_consumer->index * LOAD_SCALE + 1564 dp->load * 7) / 8; 1565 } 1566 g_io_request(cbp, cp); 1567 } 1568 1569 static void 1570 g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1571 { 1572 struct bio_queue_head queue; 1573 struct g_mirror_disk *disk; 1574 struct g_consumer *cp; 1575 struct bio *cbp; 1576 off_t left, mod, offset, slice; 1577 u_char *data; 1578 u_int ndisks; 1579 1580 if (bp->bio_length <= sc->sc_slice) { 1581 g_mirror_request_round_robin(sc, bp); 1582 return; 1583 } 1584 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1585 slice = bp->bio_length / ndisks; 1586 mod = slice % sc->sc_provider->sectorsize; 1587 if (mod != 0) 1588 slice += sc->sc_provider->sectorsize - mod; 1589 /* 1590 * Allocate all bios before sending any request, so we can 1591 * return ENOMEM in nice and clean way. 1592 */ 1593 left = bp->bio_length; 1594 offset = bp->bio_offset; 1595 data = bp->bio_data; 1596 bioq_init(&queue); 1597 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1598 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1599 continue; 1600 cbp = g_clone_bio(bp); 1601 if (cbp == NULL) { 1602 while ((cbp = bioq_takefirst(&queue)) != NULL) 1603 g_destroy_bio(cbp); 1604 if (bp->bio_error == 0) 1605 bp->bio_error = ENOMEM; 1606 g_io_deliver(bp, bp->bio_error); 1607 return; 1608 } 1609 bioq_insert_tail(&queue, cbp); 1610 cbp->bio_done = g_mirror_done; 1611 cbp->bio_caller1 = disk; 1612 cbp->bio_to = disk->d_consumer->provider; 1613 cbp->bio_offset = offset; 1614 cbp->bio_data = data; 1615 cbp->bio_length = MIN(left, slice); 1616 left -= cbp->bio_length; 1617 if (left == 0) 1618 break; 1619 offset += cbp->bio_length; 1620 data += cbp->bio_length; 1621 } 1622 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1623 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1624 disk = cbp->bio_caller1; 1625 cbp->bio_caller1 = NULL; 1626 cp = disk->d_consumer; 1627 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1628 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1629 cp->acr, cp->acw, cp->ace)); 1630 disk->d_consumer->index++; 1631 g_io_request(cbp, disk->d_consumer); 1632 } 1633 } 1634 1635 static void 1636 g_mirror_register_request(struct bio *bp) 1637 { 1638 struct g_mirror_softc *sc; 1639 1640 sc = bp->bio_to->geom->softc; 1641 switch (bp->bio_cmd) { 1642 case BIO_READ: 1643 switch (sc->sc_balance) { 1644 case G_MIRROR_BALANCE_LOAD: 1645 g_mirror_request_load(sc, bp); 1646 break; 1647 case G_MIRROR_BALANCE_PREFER: 1648 g_mirror_request_prefer(sc, bp); 1649 break; 1650 case G_MIRROR_BALANCE_ROUND_ROBIN: 1651 g_mirror_request_round_robin(sc, bp); 1652 break; 1653 case G_MIRROR_BALANCE_SPLIT: 1654 g_mirror_request_split(sc, bp); 1655 break; 1656 } 1657 return; 1658 case BIO_WRITE: 1659 case BIO_DELETE: 1660 { 1661 struct g_mirror_disk *disk; 1662 struct g_mirror_disk_sync *sync; 1663 struct bio_queue_head queue; 1664 struct g_consumer *cp; 1665 struct bio *cbp; 1666 1667 /* 1668 * Delay the request if it is colliding with a synchronization 1669 * request. 1670 */ 1671 if (g_mirror_sync_collision(sc, bp)) { 1672 g_mirror_regular_delay(sc, bp); 1673 return; 1674 } 1675 1676 if (sc->sc_idle) 1677 g_mirror_unidle(sc); 1678 else 1679 sc->sc_last_write = time_uptime; 1680 1681 /* 1682 * Bump syncid on first write. 1683 */ 1684 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1685 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1686 g_mirror_bump_syncid(sc); 1687 } 1688 1689 /* 1690 * Allocate all bios before sending any request, so we can 1691 * return ENOMEM in nice and clean way. 1692 */ 1693 bioq_init(&queue); 1694 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1695 sync = &disk->d_sync; 1696 switch (disk->d_state) { 1697 case G_MIRROR_DISK_STATE_ACTIVE: 1698 break; 1699 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1700 if (bp->bio_offset >= sync->ds_offset) 1701 continue; 1702 break; 1703 default: 1704 continue; 1705 } 1706 if (bp->bio_cmd == BIO_DELETE && 1707 (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0) 1708 continue; 1709 cbp = g_clone_bio(bp); 1710 if (cbp == NULL) { 1711 while ((cbp = bioq_takefirst(&queue)) != NULL) 1712 g_destroy_bio(cbp); 1713 if (bp->bio_error == 0) 1714 bp->bio_error = ENOMEM; 1715 g_io_deliver(bp, bp->bio_error); 1716 return; 1717 } 1718 bioq_insert_tail(&queue, cbp); 1719 cbp->bio_done = g_mirror_done; 1720 cp = disk->d_consumer; 1721 cbp->bio_caller1 = cp; 1722 cbp->bio_to = cp->provider; 1723 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1724 ("Consumer %s not opened (r%dw%de%d).", 1725 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1726 } 1727 if (bioq_first(&queue) == NULL) { 1728 g_io_deliver(bp, EOPNOTSUPP); 1729 return; 1730 } 1731 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1732 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1733 cp = cbp->bio_caller1; 1734 cbp->bio_caller1 = NULL; 1735 cp->index++; 1736 sc->sc_writes++; 1737 g_io_request(cbp, cp); 1738 } 1739 /* 1740 * Put request onto inflight queue, so we can check if new 1741 * synchronization requests don't collide with it. 1742 */ 1743 bioq_insert_tail(&sc->sc_inflight, bp); 1744 return; 1745 } 1746 default: 1747 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1748 bp->bio_cmd, sc->sc_name)); 1749 break; 1750 } 1751 } 1752 1753 static int 1754 g_mirror_can_destroy(struct g_mirror_softc *sc) 1755 { 1756 struct g_geom *gp; 1757 struct g_consumer *cp; 1758 1759 g_topology_assert(); 1760 gp = sc->sc_geom; 1761 if (gp->softc == NULL) 1762 return (1); 1763 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0) 1764 return (0); 1765 LIST_FOREACH(cp, &gp->consumer, consumer) { 1766 if (g_mirror_is_busy(sc, cp)) 1767 return (0); 1768 } 1769 gp = sc->sc_sync.ds_geom; 1770 LIST_FOREACH(cp, &gp->consumer, consumer) { 1771 if (g_mirror_is_busy(sc, cp)) 1772 return (0); 1773 } 1774 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1775 sc->sc_name); 1776 return (1); 1777 } 1778 1779 static int 1780 g_mirror_try_destroy(struct g_mirror_softc *sc) 1781 { 1782 1783 if (sc->sc_rootmount != NULL) { 1784 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1785 sc->sc_rootmount); 1786 root_mount_rel(sc->sc_rootmount); 1787 sc->sc_rootmount = NULL; 1788 } 1789 g_topology_lock(); 1790 if (!g_mirror_can_destroy(sc)) { 1791 g_topology_unlock(); 1792 return (0); 1793 } 1794 sc->sc_geom->softc = NULL; 1795 sc->sc_sync.ds_geom->softc = NULL; 1796 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1797 g_topology_unlock(); 1798 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1799 &sc->sc_worker); 1800 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1801 sx_xunlock(&sc->sc_lock); 1802 wakeup(&sc->sc_worker); 1803 sc->sc_worker = NULL; 1804 } else { 1805 g_topology_unlock(); 1806 g_mirror_destroy_device(sc); 1807 free(sc, M_MIRROR); 1808 } 1809 return (1); 1810 } 1811 1812 /* 1813 * Worker thread. 1814 */ 1815 static void 1816 g_mirror_worker(void *arg) 1817 { 1818 struct g_mirror_softc *sc; 1819 struct g_mirror_event *ep; 1820 struct bio *bp; 1821 int timeout; 1822 1823 sc = arg; 1824 thread_lock(curthread); 1825 sched_prio(curthread, PRIBIO); 1826 thread_unlock(curthread); 1827 1828 sx_xlock(&sc->sc_lock); 1829 for (;;) { 1830 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1831 /* 1832 * First take a look at events. 1833 * This is important to handle events before any I/O requests. 1834 */ 1835 ep = g_mirror_event_get(sc); 1836 if (ep != NULL) { 1837 g_mirror_event_remove(sc, ep); 1838 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1839 /* Update only device status. */ 1840 G_MIRROR_DEBUG(3, 1841 "Running event for device %s.", 1842 sc->sc_name); 1843 ep->e_error = 0; 1844 g_mirror_update_device(sc, true); 1845 } else { 1846 /* Update disk status. */ 1847 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1848 g_mirror_get_diskname(ep->e_disk)); 1849 ep->e_error = g_mirror_update_disk(ep->e_disk, 1850 ep->e_state); 1851 if (ep->e_error == 0) 1852 g_mirror_update_device(sc, false); 1853 } 1854 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1855 KASSERT(ep->e_error == 0, 1856 ("Error cannot be handled.")); 1857 g_mirror_event_free(ep); 1858 } else { 1859 ep->e_flags |= G_MIRROR_EVENT_DONE; 1860 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1861 ep); 1862 mtx_lock(&sc->sc_events_mtx); 1863 wakeup(ep); 1864 mtx_unlock(&sc->sc_events_mtx); 1865 } 1866 if ((sc->sc_flags & 1867 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1868 if (g_mirror_try_destroy(sc)) { 1869 curthread->td_pflags &= ~TDP_GEOM; 1870 G_MIRROR_DEBUG(1, "Thread exiting."); 1871 kproc_exit(0); 1872 } 1873 } 1874 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1875 continue; 1876 } 1877 /* 1878 * Check if we can mark array as CLEAN and if we can't take 1879 * how much seconds should we wait. 1880 */ 1881 timeout = g_mirror_idle(sc, -1); 1882 /* 1883 * Now I/O requests. 1884 */ 1885 /* Get first request from the queue. */ 1886 mtx_lock(&sc->sc_queue_mtx); 1887 bp = bioq_takefirst(&sc->sc_queue); 1888 if (bp == NULL) { 1889 if ((sc->sc_flags & 1890 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1891 mtx_unlock(&sc->sc_queue_mtx); 1892 if (g_mirror_try_destroy(sc)) { 1893 curthread->td_pflags &= ~TDP_GEOM; 1894 G_MIRROR_DEBUG(1, "Thread exiting."); 1895 kproc_exit(0); 1896 } 1897 mtx_lock(&sc->sc_queue_mtx); 1898 } 1899 sx_xunlock(&sc->sc_lock); 1900 /* 1901 * XXX: We can miss an event here, because an event 1902 * can be added without sx-device-lock and without 1903 * mtx-queue-lock. Maybe I should just stop using 1904 * dedicated mutex for events synchronization and 1905 * stick with the queue lock? 1906 * The event will hang here until next I/O request 1907 * or next event is received. 1908 */ 1909 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1910 timeout * hz); 1911 sx_xlock(&sc->sc_lock); 1912 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1913 continue; 1914 } 1915 mtx_unlock(&sc->sc_queue_mtx); 1916 1917 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 1918 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1919 g_mirror_sync_request(bp); /* READ */ 1920 } else if (bp->bio_to != sc->sc_provider) { 1921 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 1922 g_mirror_regular_request(bp); 1923 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 1924 g_mirror_sync_request(bp); /* WRITE */ 1925 else { 1926 KASSERT(0, 1927 ("Invalid request cflags=0x%hx to=%s.", 1928 bp->bio_cflags, bp->bio_to->name)); 1929 } 1930 } else { 1931 g_mirror_register_request(bp); 1932 } 1933 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1934 } 1935 } 1936 1937 static void 1938 g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 1939 { 1940 1941 sx_assert(&sc->sc_lock, SX_LOCKED); 1942 1943 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 1944 return; 1945 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1946 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 1947 g_mirror_get_diskname(disk), sc->sc_name); 1948 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1949 } else if (sc->sc_idle && 1950 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1951 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 1952 g_mirror_get_diskname(disk), sc->sc_name); 1953 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1954 } 1955 } 1956 1957 static void 1958 g_mirror_sync_start(struct g_mirror_disk *disk) 1959 { 1960 struct g_mirror_softc *sc; 1961 struct g_consumer *cp; 1962 struct bio *bp; 1963 int error, i; 1964 1965 g_topology_assert_not(); 1966 sc = disk->d_softc; 1967 sx_assert(&sc->sc_lock, SX_LOCKED); 1968 1969 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1970 ("Disk %s is not marked for synchronization.", 1971 g_mirror_get_diskname(disk))); 1972 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1973 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1974 sc->sc_state)); 1975 1976 sx_xunlock(&sc->sc_lock); 1977 g_topology_lock(); 1978 cp = g_new_consumer(sc->sc_sync.ds_geom); 1979 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 1980 error = g_attach(cp, sc->sc_provider); 1981 KASSERT(error == 0, 1982 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 1983 error = g_access(cp, 1, 0, 0); 1984 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 1985 g_topology_unlock(); 1986 sx_xlock(&sc->sc_lock); 1987 1988 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1989 g_mirror_get_diskname(disk)); 1990 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0) 1991 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1992 KASSERT(disk->d_sync.ds_consumer == NULL, 1993 ("Sync consumer already exists (device=%s, disk=%s).", 1994 sc->sc_name, g_mirror_get_diskname(disk))); 1995 1996 disk->d_sync.ds_consumer = cp; 1997 disk->d_sync.ds_consumer->private = disk; 1998 disk->d_sync.ds_consumer->index = 0; 1999 2000 /* 2001 * Allocate memory for synchronization bios and initialize them. 2002 */ 2003 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 2004 M_MIRROR, M_WAITOK); 2005 for (i = 0; i < g_mirror_syncreqs; i++) { 2006 bp = g_alloc_bio(); 2007 disk->d_sync.ds_bios[i] = bp; 2008 bp->bio_parent = NULL; 2009 bp->bio_cmd = BIO_READ; 2010 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 2011 bp->bio_cflags = 0; 2012 bp->bio_offset = disk->d_sync.ds_offset; 2013 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 2014 disk->d_sync.ds_offset += bp->bio_length; 2015 bp->bio_done = g_mirror_sync_done; 2016 bp->bio_from = disk->d_sync.ds_consumer; 2017 bp->bio_to = sc->sc_provider; 2018 bp->bio_caller1 = (void *)(uintptr_t)i; 2019 } 2020 2021 /* Increase the number of disks in SYNCHRONIZING state. */ 2022 sc->sc_sync.ds_ndisks++; 2023 /* Set the number of in-flight synchronization requests. */ 2024 disk->d_sync.ds_inflight = g_mirror_syncreqs; 2025 2026 /* 2027 * Fire off first synchronization requests. 2028 */ 2029 for (i = 0; i < g_mirror_syncreqs; i++) { 2030 bp = disk->d_sync.ds_bios[i]; 2031 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 2032 disk->d_sync.ds_consumer->index++; 2033 /* 2034 * Delay the request if it is colliding with a regular request. 2035 */ 2036 if (g_mirror_regular_collision(sc, bp)) 2037 g_mirror_sync_delay(sc, bp); 2038 else 2039 g_io_request(bp, disk->d_sync.ds_consumer); 2040 } 2041 } 2042 2043 /* 2044 * Stop synchronization process. 2045 * type: 0 - synchronization finished 2046 * 1 - synchronization stopped 2047 */ 2048 static void 2049 g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 2050 { 2051 struct g_mirror_softc *sc; 2052 struct g_consumer *cp; 2053 2054 g_topology_assert_not(); 2055 sc = disk->d_softc; 2056 sx_assert(&sc->sc_lock, SX_LOCKED); 2057 2058 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2059 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2060 g_mirror_disk_state2str(disk->d_state))); 2061 if (disk->d_sync.ds_consumer == NULL) 2062 return; 2063 2064 if (type == 0) { 2065 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2066 sc->sc_name, g_mirror_get_diskname(disk)); 2067 } else /* if (type == 1) */ { 2068 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2069 sc->sc_name, g_mirror_get_diskname(disk)); 2070 } 2071 g_mirror_regular_release(sc); 2072 free(disk->d_sync.ds_bios, M_MIRROR); 2073 disk->d_sync.ds_bios = NULL; 2074 cp = disk->d_sync.ds_consumer; 2075 disk->d_sync.ds_consumer = NULL; 2076 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2077 sc->sc_sync.ds_ndisks--; 2078 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2079 g_topology_lock(); 2080 g_mirror_kill_consumer(sc, cp); 2081 g_topology_unlock(); 2082 sx_xlock(&sc->sc_lock); 2083 } 2084 2085 static void 2086 g_mirror_launch_provider(struct g_mirror_softc *sc) 2087 { 2088 struct g_mirror_disk *disk; 2089 struct g_provider *pp, *dp; 2090 2091 sx_assert(&sc->sc_lock, SX_LOCKED); 2092 2093 g_topology_lock(); 2094 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 2095 pp->flags |= G_PF_DIRECT_RECEIVE; 2096 pp->mediasize = sc->sc_mediasize; 2097 pp->sectorsize = sc->sc_sectorsize; 2098 pp->stripesize = 0; 2099 pp->stripeoffset = 0; 2100 2101 /* Splitting of unmapped BIO's could work but isn't implemented now */ 2102 if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT) 2103 pp->flags |= G_PF_ACCEPT_UNMAPPED; 2104 2105 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2106 if (disk->d_consumer && disk->d_consumer->provider) { 2107 dp = disk->d_consumer->provider; 2108 if (dp->stripesize > pp->stripesize) { 2109 pp->stripesize = dp->stripesize; 2110 pp->stripeoffset = dp->stripeoffset; 2111 } 2112 /* A provider underneath us doesn't support unmapped */ 2113 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { 2114 G_MIRROR_DEBUG(0, "Cancelling unmapped " 2115 "because of %s.", dp->name); 2116 pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 2117 } 2118 } 2119 } 2120 sc->sc_provider = pp; 2121 g_error_provider(pp, 0); 2122 g_topology_unlock(); 2123 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2124 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 2125 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2126 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2127 g_mirror_sync_start(disk); 2128 } 2129 } 2130 2131 static void 2132 g_mirror_destroy_provider(struct g_mirror_softc *sc) 2133 { 2134 struct g_mirror_disk *disk; 2135 struct bio *bp; 2136 2137 g_topology_assert_not(); 2138 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2139 sc->sc_name)); 2140 2141 g_topology_lock(); 2142 g_error_provider(sc->sc_provider, ENXIO); 2143 mtx_lock(&sc->sc_queue_mtx); 2144 while ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) { 2145 /* 2146 * Abort any pending I/O that wasn't generated by us. 2147 * Synchronization requests and requests destined for individual 2148 * mirror components can be destroyed immediately. 2149 */ 2150 if (bp->bio_to == sc->sc_provider && 2151 bp->bio_from->geom != sc->sc_sync.ds_geom) { 2152 g_io_deliver(bp, ENXIO); 2153 } else { 2154 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2155 free(bp->bio_data, M_MIRROR); 2156 g_destroy_bio(bp); 2157 } 2158 } 2159 mtx_unlock(&sc->sc_queue_mtx); 2160 g_wither_provider(sc->sc_provider, ENXIO); 2161 sc->sc_provider = NULL; 2162 G_MIRROR_DEBUG(0, "Device %s: provider destroyed.", sc->sc_name); 2163 g_topology_unlock(); 2164 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2165 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2166 g_mirror_sync_stop(disk, 1); 2167 } 2168 } 2169 2170 static void 2171 g_mirror_go(void *arg) 2172 { 2173 struct g_mirror_softc *sc; 2174 2175 sc = arg; 2176 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2177 g_mirror_event_send(sc, 0, 2178 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2179 } 2180 2181 static u_int 2182 g_mirror_determine_state(struct g_mirror_disk *disk) 2183 { 2184 struct g_mirror_softc *sc; 2185 u_int state; 2186 2187 sc = disk->d_softc; 2188 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2189 if ((disk->d_flags & 2190 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2191 /* Disk does not need synchronization. */ 2192 state = G_MIRROR_DISK_STATE_ACTIVE; 2193 } else { 2194 if ((sc->sc_flags & 2195 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2196 (disk->d_flags & 2197 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2198 /* 2199 * We can start synchronization from 2200 * the stored offset. 2201 */ 2202 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2203 } else { 2204 state = G_MIRROR_DISK_STATE_STALE; 2205 } 2206 } 2207 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2208 /* 2209 * Reset all synchronization data for this disk, 2210 * because if it even was synchronized, it was 2211 * synchronized to disks with different syncid. 2212 */ 2213 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2214 disk->d_sync.ds_offset = 0; 2215 disk->d_sync.ds_offset_done = 0; 2216 disk->d_sync.ds_syncid = sc->sc_syncid; 2217 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2218 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2219 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2220 } else { 2221 state = G_MIRROR_DISK_STATE_STALE; 2222 } 2223 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2224 /* 2225 * Not good, NOT GOOD! 2226 * It means that mirror was started on stale disks 2227 * and more fresh disk just arrive. 2228 * If there were writes, mirror is broken, sorry. 2229 * I think the best choice here is don't touch 2230 * this disk and inform the user loudly. 2231 */ 2232 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2233 "disk (%s) arrives!! It will not be connected to the " 2234 "running device.", sc->sc_name, 2235 g_mirror_get_diskname(disk)); 2236 g_mirror_destroy_disk(disk); 2237 state = G_MIRROR_DISK_STATE_NONE; 2238 /* Return immediately, because disk was destroyed. */ 2239 return (state); 2240 } 2241 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2242 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2243 return (state); 2244 } 2245 2246 /* 2247 * Update device state. 2248 */ 2249 static void 2250 g_mirror_update_device(struct g_mirror_softc *sc, bool force) 2251 { 2252 struct g_mirror_disk *disk; 2253 u_int state; 2254 2255 sx_assert(&sc->sc_lock, SX_XLOCKED); 2256 2257 switch (sc->sc_state) { 2258 case G_MIRROR_DEVICE_STATE_STARTING: 2259 { 2260 struct g_mirror_disk *pdisk, *tdisk; 2261 u_int dirty, ndisks, genid, syncid; 2262 bool broken; 2263 2264 KASSERT(sc->sc_provider == NULL, 2265 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2266 /* 2267 * Are we ready? We are, if all disks are connected or 2268 * if we have any disks and 'force' is true. 2269 */ 2270 ndisks = g_mirror_ndisks(sc, -1); 2271 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2272 ; 2273 } else if (ndisks == 0) { 2274 /* 2275 * Disks went down in starting phase, so destroy 2276 * device. 2277 */ 2278 callout_drain(&sc->sc_callout); 2279 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2280 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2281 sc->sc_rootmount); 2282 root_mount_rel(sc->sc_rootmount); 2283 sc->sc_rootmount = NULL; 2284 return; 2285 } else { 2286 return; 2287 } 2288 2289 /* 2290 * Activate all disks with the biggest syncid. 2291 */ 2292 if (force) { 2293 /* 2294 * If 'force' is true, we have been called due to 2295 * timeout, so don't bother canceling timeout. 2296 */ 2297 ndisks = 0; 2298 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2299 if ((disk->d_flags & 2300 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2301 ndisks++; 2302 } 2303 } 2304 if (ndisks == 0) { 2305 /* No valid disks found, destroy device. */ 2306 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2307 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2308 __LINE__, sc->sc_rootmount); 2309 root_mount_rel(sc->sc_rootmount); 2310 sc->sc_rootmount = NULL; 2311 return; 2312 } 2313 } else { 2314 /* Cancel timeout. */ 2315 callout_drain(&sc->sc_callout); 2316 } 2317 2318 /* 2319 * Find the biggest genid. 2320 */ 2321 genid = 0; 2322 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2323 if (disk->d_genid > genid) 2324 genid = disk->d_genid; 2325 } 2326 sc->sc_genid = genid; 2327 /* 2328 * Remove all disks without the biggest genid. 2329 */ 2330 broken = false; 2331 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2332 if (disk->d_genid < genid) { 2333 G_MIRROR_DEBUG(0, 2334 "Component %s (device %s) broken, skipping.", 2335 g_mirror_get_diskname(disk), sc->sc_name); 2336 g_mirror_destroy_disk(disk); 2337 /* 2338 * Bump the syncid in case we discover a healthy 2339 * replacement disk after starting the mirror. 2340 */ 2341 broken = true; 2342 } 2343 } 2344 2345 /* 2346 * Find the biggest syncid. 2347 */ 2348 syncid = 0; 2349 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2350 if (disk->d_sync.ds_syncid > syncid) 2351 syncid = disk->d_sync.ds_syncid; 2352 } 2353 2354 /* 2355 * Here we need to look for dirty disks and if all disks 2356 * with the biggest syncid are dirty, we have to choose 2357 * one with the biggest priority and rebuild the rest. 2358 */ 2359 /* 2360 * Find the number of dirty disks with the biggest syncid. 2361 * Find the number of disks with the biggest syncid. 2362 * While here, find a disk with the biggest priority. 2363 */ 2364 dirty = ndisks = 0; 2365 pdisk = NULL; 2366 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2367 if (disk->d_sync.ds_syncid != syncid) 2368 continue; 2369 if ((disk->d_flags & 2370 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2371 continue; 2372 } 2373 ndisks++; 2374 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2375 dirty++; 2376 if (pdisk == NULL || 2377 pdisk->d_priority < disk->d_priority) { 2378 pdisk = disk; 2379 } 2380 } 2381 } 2382 if (dirty == 0) { 2383 /* No dirty disks at all, great. */ 2384 } else if (dirty == ndisks) { 2385 /* 2386 * Force synchronization for all dirty disks except one 2387 * with the biggest priority. 2388 */ 2389 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2390 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2391 "master disk for synchronization.", 2392 g_mirror_get_diskname(pdisk), sc->sc_name); 2393 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2394 if (disk->d_sync.ds_syncid != syncid) 2395 continue; 2396 if ((disk->d_flags & 2397 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2398 continue; 2399 } 2400 KASSERT((disk->d_flags & 2401 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2402 ("Disk %s isn't marked as dirty.", 2403 g_mirror_get_diskname(disk))); 2404 /* Skip the disk with the biggest priority. */ 2405 if (disk == pdisk) 2406 continue; 2407 disk->d_sync.ds_syncid = 0; 2408 } 2409 } else if (dirty < ndisks) { 2410 /* 2411 * Force synchronization for all dirty disks. 2412 * We have some non-dirty disks. 2413 */ 2414 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2415 if (disk->d_sync.ds_syncid != syncid) 2416 continue; 2417 if ((disk->d_flags & 2418 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2419 continue; 2420 } 2421 if ((disk->d_flags & 2422 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2423 continue; 2424 } 2425 disk->d_sync.ds_syncid = 0; 2426 } 2427 } 2428 2429 /* Reset hint. */ 2430 sc->sc_hint = NULL; 2431 sc->sc_syncid = syncid; 2432 if (force || broken) { 2433 /* Remember to bump syncid on first write. */ 2434 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2435 } 2436 state = G_MIRROR_DEVICE_STATE_RUNNING; 2437 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2438 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2439 g_mirror_device_state2str(state)); 2440 sc->sc_state = state; 2441 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2442 state = g_mirror_determine_state(disk); 2443 g_mirror_event_send(disk, state, 2444 G_MIRROR_EVENT_DONTWAIT); 2445 if (state == G_MIRROR_DISK_STATE_STALE) 2446 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2447 } 2448 break; 2449 } 2450 case G_MIRROR_DEVICE_STATE_RUNNING: 2451 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2452 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2453 /* 2454 * No active disks or no disks at all, 2455 * so destroy device. 2456 */ 2457 if (sc->sc_provider != NULL) 2458 g_mirror_destroy_provider(sc); 2459 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2460 break; 2461 } else if (g_mirror_ndisks(sc, 2462 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2463 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2464 /* 2465 * We have active disks, launch provider if it doesn't 2466 * exist. 2467 */ 2468 if (sc->sc_provider == NULL) 2469 g_mirror_launch_provider(sc); 2470 if (sc->sc_rootmount != NULL) { 2471 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2472 __LINE__, sc->sc_rootmount); 2473 root_mount_rel(sc->sc_rootmount); 2474 sc->sc_rootmount = NULL; 2475 } 2476 } 2477 /* 2478 * Genid should be bumped immediately, so do it here. 2479 */ 2480 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2481 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2482 g_mirror_bump_genid(sc); 2483 } 2484 break; 2485 default: 2486 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2487 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2488 break; 2489 } 2490 } 2491 2492 /* 2493 * Update disk state and device state if needed. 2494 */ 2495 #define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2496 "Disk %s state changed from %s to %s (device %s).", \ 2497 g_mirror_get_diskname(disk), \ 2498 g_mirror_disk_state2str(disk->d_state), \ 2499 g_mirror_disk_state2str(state), sc->sc_name) 2500 static int 2501 g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2502 { 2503 struct g_mirror_softc *sc; 2504 2505 sc = disk->d_softc; 2506 sx_assert(&sc->sc_lock, SX_XLOCKED); 2507 2508 again: 2509 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2510 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2511 g_mirror_disk_state2str(state)); 2512 switch (state) { 2513 case G_MIRROR_DISK_STATE_NEW: 2514 /* 2515 * Possible scenarios: 2516 * 1. New disk arrive. 2517 */ 2518 /* Previous state should be NONE. */ 2519 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2520 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2521 g_mirror_disk_state2str(disk->d_state))); 2522 DISK_STATE_CHANGED(); 2523 2524 disk->d_state = state; 2525 if (LIST_EMPTY(&sc->sc_disks)) 2526 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2527 else { 2528 struct g_mirror_disk *dp; 2529 2530 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2531 if (disk->d_priority >= dp->d_priority) { 2532 LIST_INSERT_BEFORE(dp, disk, d_next); 2533 dp = NULL; 2534 break; 2535 } 2536 if (LIST_NEXT(dp, d_next) == NULL) 2537 break; 2538 } 2539 if (dp != NULL) 2540 LIST_INSERT_AFTER(dp, disk, d_next); 2541 } 2542 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2543 sc->sc_name, g_mirror_get_diskname(disk)); 2544 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2545 break; 2546 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2547 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2548 g_mirror_device_state2str(sc->sc_state), 2549 g_mirror_get_diskname(disk), 2550 g_mirror_disk_state2str(disk->d_state))); 2551 state = g_mirror_determine_state(disk); 2552 if (state != G_MIRROR_DISK_STATE_NONE) 2553 goto again; 2554 break; 2555 case G_MIRROR_DISK_STATE_ACTIVE: 2556 /* 2557 * Possible scenarios: 2558 * 1. New disk does not need synchronization. 2559 * 2. Synchronization process finished successfully. 2560 */ 2561 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2562 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2563 g_mirror_device_state2str(sc->sc_state), 2564 g_mirror_get_diskname(disk), 2565 g_mirror_disk_state2str(disk->d_state))); 2566 /* Previous state should be NEW or SYNCHRONIZING. */ 2567 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2568 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2569 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2570 g_mirror_disk_state2str(disk->d_state))); 2571 DISK_STATE_CHANGED(); 2572 2573 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2574 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2575 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2576 g_mirror_sync_stop(disk, 0); 2577 } 2578 disk->d_state = state; 2579 disk->d_sync.ds_offset = 0; 2580 disk->d_sync.ds_offset_done = 0; 2581 g_mirror_update_idle(sc, disk); 2582 g_mirror_update_metadata(disk); 2583 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2584 sc->sc_name, g_mirror_get_diskname(disk)); 2585 break; 2586 case G_MIRROR_DISK_STATE_STALE: 2587 /* 2588 * Possible scenarios: 2589 * 1. Stale disk was connected. 2590 */ 2591 /* Previous state should be NEW. */ 2592 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2593 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2594 g_mirror_disk_state2str(disk->d_state))); 2595 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2596 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2597 g_mirror_device_state2str(sc->sc_state), 2598 g_mirror_get_diskname(disk), 2599 g_mirror_disk_state2str(disk->d_state))); 2600 /* 2601 * STALE state is only possible if device is marked 2602 * NOAUTOSYNC. 2603 */ 2604 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2605 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2606 g_mirror_device_state2str(sc->sc_state), 2607 g_mirror_get_diskname(disk), 2608 g_mirror_disk_state2str(disk->d_state))); 2609 DISK_STATE_CHANGED(); 2610 2611 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2612 disk->d_state = state; 2613 g_mirror_update_metadata(disk); 2614 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2615 sc->sc_name, g_mirror_get_diskname(disk)); 2616 break; 2617 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2618 /* 2619 * Possible scenarios: 2620 * 1. Disk which needs synchronization was connected. 2621 */ 2622 /* Previous state should be NEW. */ 2623 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2624 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2625 g_mirror_disk_state2str(disk->d_state))); 2626 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2627 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2628 g_mirror_device_state2str(sc->sc_state), 2629 g_mirror_get_diskname(disk), 2630 g_mirror_disk_state2str(disk->d_state))); 2631 DISK_STATE_CHANGED(); 2632 2633 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2634 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2635 disk->d_state = state; 2636 if (sc->sc_provider != NULL) { 2637 g_mirror_sync_start(disk); 2638 g_mirror_update_metadata(disk); 2639 } 2640 break; 2641 case G_MIRROR_DISK_STATE_DISCONNECTED: 2642 /* 2643 * Possible scenarios: 2644 * 1. Device wasn't running yet, but disk disappear. 2645 * 2. Disk was active and disapppear. 2646 * 3. Disk disappear during synchronization process. 2647 */ 2648 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2649 /* 2650 * Previous state should be ACTIVE, STALE or 2651 * SYNCHRONIZING. 2652 */ 2653 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2654 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2655 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2656 ("Wrong disk state (%s, %s).", 2657 g_mirror_get_diskname(disk), 2658 g_mirror_disk_state2str(disk->d_state))); 2659 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2660 /* Previous state should be NEW. */ 2661 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2662 ("Wrong disk state (%s, %s).", 2663 g_mirror_get_diskname(disk), 2664 g_mirror_disk_state2str(disk->d_state))); 2665 /* 2666 * Reset bumping syncid if disk disappeared in STARTING 2667 * state. 2668 */ 2669 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2670 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2671 #ifdef INVARIANTS 2672 } else { 2673 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2674 sc->sc_name, 2675 g_mirror_device_state2str(sc->sc_state), 2676 g_mirror_get_diskname(disk), 2677 g_mirror_disk_state2str(disk->d_state))); 2678 #endif 2679 } 2680 DISK_STATE_CHANGED(); 2681 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2682 sc->sc_name, g_mirror_get_diskname(disk)); 2683 2684 g_mirror_destroy_disk(disk); 2685 break; 2686 case G_MIRROR_DISK_STATE_DESTROY: 2687 { 2688 int error; 2689 2690 error = g_mirror_clear_metadata(disk); 2691 if (error != 0) { 2692 G_MIRROR_DEBUG(0, 2693 "Device %s: failed to clear metadata on %s: %d.", 2694 sc->sc_name, g_mirror_get_diskname(disk), error); 2695 break; 2696 } 2697 DISK_STATE_CHANGED(); 2698 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2699 sc->sc_name, g_mirror_get_diskname(disk)); 2700 2701 g_mirror_destroy_disk(disk); 2702 sc->sc_ndisks--; 2703 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2704 g_mirror_update_metadata(disk); 2705 } 2706 break; 2707 } 2708 default: 2709 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2710 break; 2711 } 2712 return (0); 2713 } 2714 #undef DISK_STATE_CHANGED 2715 2716 int 2717 g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2718 { 2719 struct g_provider *pp; 2720 u_char *buf; 2721 int error; 2722 2723 g_topology_assert(); 2724 2725 error = g_access(cp, 1, 0, 0); 2726 if (error != 0) 2727 return (error); 2728 pp = cp->provider; 2729 g_topology_unlock(); 2730 /* Metadata are stored on last sector. */ 2731 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2732 &error); 2733 g_topology_lock(); 2734 g_access(cp, -1, 0, 0); 2735 if (buf == NULL) { 2736 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2737 cp->provider->name, error); 2738 return (error); 2739 } 2740 2741 /* Decode metadata. */ 2742 error = mirror_metadata_decode(buf, md); 2743 g_free(buf); 2744 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2745 return (EINVAL); 2746 if (md->md_version > G_MIRROR_VERSION) { 2747 G_MIRROR_DEBUG(0, 2748 "Kernel module is too old to handle metadata from %s.", 2749 cp->provider->name); 2750 return (EINVAL); 2751 } 2752 if (error != 0) { 2753 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2754 cp->provider->name); 2755 return (error); 2756 } 2757 2758 return (0); 2759 } 2760 2761 static int 2762 g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2763 struct g_mirror_metadata *md) 2764 { 2765 2766 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2767 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2768 pp->name, md->md_did); 2769 return (EEXIST); 2770 } 2771 if (md->md_all != sc->sc_ndisks) { 2772 G_MIRROR_DEBUG(1, 2773 "Invalid '%s' field on disk %s (device %s), skipping.", 2774 "md_all", pp->name, sc->sc_name); 2775 return (EINVAL); 2776 } 2777 if (md->md_slice != sc->sc_slice) { 2778 G_MIRROR_DEBUG(1, 2779 "Invalid '%s' field on disk %s (device %s), skipping.", 2780 "md_slice", pp->name, sc->sc_name); 2781 return (EINVAL); 2782 } 2783 if (md->md_balance != sc->sc_balance) { 2784 G_MIRROR_DEBUG(1, 2785 "Invalid '%s' field on disk %s (device %s), skipping.", 2786 "md_balance", pp->name, sc->sc_name); 2787 return (EINVAL); 2788 } 2789 #if 0 2790 if (md->md_mediasize != sc->sc_mediasize) { 2791 G_MIRROR_DEBUG(1, 2792 "Invalid '%s' field on disk %s (device %s), skipping.", 2793 "md_mediasize", pp->name, sc->sc_name); 2794 return (EINVAL); 2795 } 2796 #endif 2797 if (sc->sc_mediasize > pp->mediasize) { 2798 G_MIRROR_DEBUG(1, 2799 "Invalid size of disk %s (device %s), skipping.", pp->name, 2800 sc->sc_name); 2801 return (EINVAL); 2802 } 2803 if (md->md_sectorsize != sc->sc_sectorsize) { 2804 G_MIRROR_DEBUG(1, 2805 "Invalid '%s' field on disk %s (device %s), skipping.", 2806 "md_sectorsize", pp->name, sc->sc_name); 2807 return (EINVAL); 2808 } 2809 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2810 G_MIRROR_DEBUG(1, 2811 "Invalid sector size of disk %s (device %s), skipping.", 2812 pp->name, sc->sc_name); 2813 return (EINVAL); 2814 } 2815 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2816 G_MIRROR_DEBUG(1, 2817 "Invalid device flags on disk %s (device %s), skipping.", 2818 pp->name, sc->sc_name); 2819 return (EINVAL); 2820 } 2821 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2822 G_MIRROR_DEBUG(1, 2823 "Invalid disk flags on disk %s (device %s), skipping.", 2824 pp->name, sc->sc_name); 2825 return (EINVAL); 2826 } 2827 return (0); 2828 } 2829 2830 int 2831 g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2832 struct g_mirror_metadata *md) 2833 { 2834 struct g_mirror_disk *disk; 2835 int error; 2836 2837 g_topology_assert_not(); 2838 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2839 2840 error = g_mirror_check_metadata(sc, pp, md); 2841 if (error != 0) 2842 return (error); 2843 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2844 md->md_genid < sc->sc_genid) { 2845 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2846 pp->name, sc->sc_name); 2847 return (EINVAL); 2848 } 2849 disk = g_mirror_init_disk(sc, pp, md, &error); 2850 if (disk == NULL) 2851 return (error); 2852 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2853 G_MIRROR_EVENT_WAIT); 2854 if (error != 0) 2855 return (error); 2856 if (md->md_version < G_MIRROR_VERSION) { 2857 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2858 pp->name, md->md_version, G_MIRROR_VERSION); 2859 g_mirror_update_metadata(disk); 2860 } 2861 return (0); 2862 } 2863 2864 static void 2865 g_mirror_destroy_delayed(void *arg, int flag) 2866 { 2867 struct g_mirror_softc *sc; 2868 int error; 2869 2870 if (flag == EV_CANCEL) { 2871 G_MIRROR_DEBUG(1, "Destroying canceled."); 2872 return; 2873 } 2874 sc = arg; 2875 g_topology_unlock(); 2876 sx_xlock(&sc->sc_lock); 2877 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2878 ("DESTROY flag set on %s.", sc->sc_name)); 2879 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0, 2880 ("DESTROYING flag not set on %s.", sc->sc_name)); 2881 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 2882 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 2883 if (error != 0) { 2884 G_MIRROR_DEBUG(0, "Cannot destroy %s (error=%d).", 2885 sc->sc_name, error); 2886 sx_xunlock(&sc->sc_lock); 2887 } 2888 g_topology_lock(); 2889 } 2890 2891 static int 2892 g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2893 { 2894 struct g_mirror_softc *sc; 2895 int error = 0; 2896 2897 g_topology_assert(); 2898 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2899 acw, ace); 2900 2901 sc = pp->geom->softc; 2902 if (sc == NULL && acr <= 0 && acw <= 0 && ace <= 0) 2903 return (0); 2904 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 2905 2906 g_topology_unlock(); 2907 sx_xlock(&sc->sc_lock); 2908 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 2909 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0 || 2910 LIST_EMPTY(&sc->sc_disks)) { 2911 if (acr > 0 || acw > 0 || ace > 0) 2912 error = ENXIO; 2913 goto end; 2914 } 2915 sc->sc_provider_open += acr + acw + ace; 2916 if (pp->acw + acw == 0) 2917 g_mirror_idle(sc, 0); 2918 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0 && 2919 sc->sc_provider_open == 0) 2920 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, sc, NULL); 2921 end: 2922 sx_xunlock(&sc->sc_lock); 2923 g_topology_lock(); 2924 return (error); 2925 } 2926 2927 static struct g_geom * 2928 g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2929 { 2930 struct g_mirror_softc *sc; 2931 struct g_geom *gp; 2932 int error, timeout; 2933 2934 g_topology_assert(); 2935 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2936 md->md_mid); 2937 2938 /* One disk is minimum. */ 2939 if (md->md_all < 1) 2940 return (NULL); 2941 /* 2942 * Action geom. 2943 */ 2944 gp = g_new_geomf(mp, "%s", md->md_name); 2945 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2946 gp->start = g_mirror_start; 2947 gp->orphan = g_mirror_orphan; 2948 gp->access = g_mirror_access; 2949 gp->dumpconf = g_mirror_dumpconf; 2950 2951 sc->sc_id = md->md_mid; 2952 sc->sc_slice = md->md_slice; 2953 sc->sc_balance = md->md_balance; 2954 sc->sc_mediasize = md->md_mediasize; 2955 sc->sc_sectorsize = md->md_sectorsize; 2956 sc->sc_ndisks = md->md_all; 2957 sc->sc_flags = md->md_mflags; 2958 sc->sc_bump_id = 0; 2959 sc->sc_idle = 1; 2960 sc->sc_last_write = time_uptime; 2961 sc->sc_writes = 0; 2962 sx_init(&sc->sc_lock, "gmirror:lock"); 2963 bioq_init(&sc->sc_queue); 2964 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2965 bioq_init(&sc->sc_regular_delayed); 2966 bioq_init(&sc->sc_inflight); 2967 bioq_init(&sc->sc_sync_delayed); 2968 LIST_INIT(&sc->sc_disks); 2969 TAILQ_INIT(&sc->sc_events); 2970 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2971 callout_init(&sc->sc_callout, 1); 2972 mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF); 2973 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2974 gp->softc = sc; 2975 sc->sc_geom = gp; 2976 sc->sc_provider = NULL; 2977 sc->sc_provider_open = 0; 2978 /* 2979 * Synchronization geom. 2980 */ 2981 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2982 gp->softc = sc; 2983 gp->orphan = g_mirror_orphan; 2984 sc->sc_sync.ds_geom = gp; 2985 sc->sc_sync.ds_ndisks = 0; 2986 error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2987 "g_mirror %s", md->md_name); 2988 if (error != 0) { 2989 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2990 sc->sc_name); 2991 g_destroy_geom(sc->sc_sync.ds_geom); 2992 mtx_destroy(&sc->sc_done_mtx); 2993 mtx_destroy(&sc->sc_events_mtx); 2994 mtx_destroy(&sc->sc_queue_mtx); 2995 sx_destroy(&sc->sc_lock); 2996 g_destroy_geom(sc->sc_geom); 2997 free(sc, M_MIRROR); 2998 return (NULL); 2999 } 3000 3001 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 3002 sc->sc_name, sc->sc_ndisks, sc->sc_id); 3003 3004 sc->sc_rootmount = root_mount_hold("GMIRROR"); 3005 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 3006 /* 3007 * Run timeout. 3008 */ 3009 timeout = g_mirror_timeout * hz; 3010 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 3011 return (sc->sc_geom); 3012 } 3013 3014 int 3015 g_mirror_destroy(struct g_mirror_softc *sc, int how) 3016 { 3017 struct g_mirror_disk *disk; 3018 3019 g_topology_assert_not(); 3020 if (sc == NULL) 3021 return (ENXIO); 3022 sx_assert(&sc->sc_lock, SX_XLOCKED); 3023 3024 if (sc->sc_provider_open != 0 || SCHEDULER_STOPPED()) { 3025 switch (how) { 3026 case G_MIRROR_DESTROY_SOFT: 3027 G_MIRROR_DEBUG(1, 3028 "Device %s is still open (%d).", sc->sc_name, 3029 sc->sc_provider_open); 3030 return (EBUSY); 3031 case G_MIRROR_DESTROY_DELAYED: 3032 G_MIRROR_DEBUG(1, 3033 "Device %s will be destroyed on last close.", 3034 sc->sc_name); 3035 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 3036 if (disk->d_state == 3037 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3038 g_mirror_sync_stop(disk, 1); 3039 } 3040 } 3041 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROYING; 3042 return (EBUSY); 3043 case G_MIRROR_DESTROY_HARD: 3044 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 3045 "can't be definitely removed.", sc->sc_name); 3046 } 3047 } 3048 3049 g_topology_lock(); 3050 if (sc->sc_geom->softc == NULL) { 3051 g_topology_unlock(); 3052 return (0); 3053 } 3054 sc->sc_geom->softc = NULL; 3055 sc->sc_sync.ds_geom->softc = NULL; 3056 g_topology_unlock(); 3057 3058 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 3059 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 3060 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3061 sx_xunlock(&sc->sc_lock); 3062 mtx_lock(&sc->sc_queue_mtx); 3063 wakeup(sc); 3064 mtx_unlock(&sc->sc_queue_mtx); 3065 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3066 while (sc->sc_worker != NULL) 3067 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 3068 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3069 sx_xlock(&sc->sc_lock); 3070 g_mirror_destroy_device(sc); 3071 free(sc, M_MIRROR); 3072 return (0); 3073 } 3074 3075 static void 3076 g_mirror_taste_orphan(struct g_consumer *cp) 3077 { 3078 3079 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3080 cp->provider->name)); 3081 } 3082 3083 static struct g_geom * 3084 g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3085 { 3086 struct g_mirror_metadata md; 3087 struct g_mirror_softc *sc; 3088 struct g_consumer *cp; 3089 struct g_geom *gp; 3090 int error; 3091 3092 g_topology_assert(); 3093 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3094 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 3095 3096 gp = g_new_geomf(mp, "mirror:taste"); 3097 /* 3098 * This orphan function should be never called. 3099 */ 3100 gp->orphan = g_mirror_taste_orphan; 3101 cp = g_new_consumer(gp); 3102 g_attach(cp, pp); 3103 error = g_mirror_read_metadata(cp, &md); 3104 g_detach(cp); 3105 g_destroy_consumer(cp); 3106 g_destroy_geom(gp); 3107 if (error != 0) 3108 return (NULL); 3109 gp = NULL; 3110 3111 if (md.md_provider[0] != '\0' && 3112 !g_compare_names(md.md_provider, pp->name)) 3113 return (NULL); 3114 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3115 return (NULL); 3116 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 3117 G_MIRROR_DEBUG(0, 3118 "Device %s: provider %s marked as inactive, skipping.", 3119 md.md_name, pp->name); 3120 return (NULL); 3121 } 3122 if (g_mirror_debug >= 2) 3123 mirror_metadata_dump(&md); 3124 3125 /* 3126 * Let's check if device already exists. 3127 */ 3128 sc = NULL; 3129 LIST_FOREACH(gp, &mp->geom, geom) { 3130 sc = gp->softc; 3131 if (sc == NULL) 3132 continue; 3133 if (sc->sc_sync.ds_geom == gp) 3134 continue; 3135 if (strcmp(md.md_name, sc->sc_name) != 0) 3136 continue; 3137 if (md.md_mid != sc->sc_id) { 3138 G_MIRROR_DEBUG(0, "Device %s already configured.", 3139 sc->sc_name); 3140 return (NULL); 3141 } 3142 break; 3143 } 3144 if (gp == NULL) { 3145 gp = g_mirror_create(mp, &md); 3146 if (gp == NULL) { 3147 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3148 md.md_name); 3149 return (NULL); 3150 } 3151 sc = gp->softc; 3152 } 3153 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3154 g_topology_unlock(); 3155 sx_xlock(&sc->sc_lock); 3156 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; 3157 error = g_mirror_add_disk(sc, pp, &md); 3158 if (error != 0) { 3159 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3160 pp->name, gp->name, error); 3161 if (LIST_EMPTY(&sc->sc_disks)) { 3162 g_cancel_event(sc); 3163 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3164 g_topology_lock(); 3165 return (NULL); 3166 } 3167 gp = NULL; 3168 } 3169 sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; 3170 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3171 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3172 g_topology_lock(); 3173 return (NULL); 3174 } 3175 sx_xunlock(&sc->sc_lock); 3176 g_topology_lock(); 3177 return (gp); 3178 } 3179 3180 static void 3181 g_mirror_resize(struct g_consumer *cp) 3182 { 3183 struct g_mirror_disk *disk; 3184 3185 g_topology_assert(); 3186 g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name); 3187 3188 disk = cp->private; 3189 if (disk == NULL) 3190 return; 3191 g_topology_unlock(); 3192 g_mirror_update_metadata(disk); 3193 g_topology_lock(); 3194 } 3195 3196 static int 3197 g_mirror_destroy_geom(struct gctl_req *req __unused, 3198 struct g_class *mp __unused, struct g_geom *gp) 3199 { 3200 struct g_mirror_softc *sc; 3201 int error; 3202 3203 g_topology_unlock(); 3204 sc = gp->softc; 3205 sx_xlock(&sc->sc_lock); 3206 g_cancel_event(sc); 3207 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3208 if (error != 0) 3209 sx_xunlock(&sc->sc_lock); 3210 g_topology_lock(); 3211 return (error); 3212 } 3213 3214 static void 3215 g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3216 struct g_consumer *cp, struct g_provider *pp) 3217 { 3218 struct g_mirror_softc *sc; 3219 3220 g_topology_assert(); 3221 3222 sc = gp->softc; 3223 if (sc == NULL) 3224 return; 3225 /* Skip synchronization geom. */ 3226 if (gp == sc->sc_sync.ds_geom) 3227 return; 3228 if (pp != NULL) { 3229 /* Nothing here. */ 3230 } else if (cp != NULL) { 3231 struct g_mirror_disk *disk; 3232 3233 disk = cp->private; 3234 if (disk == NULL) 3235 return; 3236 g_topology_unlock(); 3237 sx_xlock(&sc->sc_lock); 3238 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3239 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3240 sbuf_printf(sb, "%s<Synchronized>", indent); 3241 if (disk->d_sync.ds_offset == 0) 3242 sbuf_printf(sb, "0%%"); 3243 else { 3244 sbuf_printf(sb, "%u%%", 3245 (u_int)((disk->d_sync.ds_offset * 100) / 3246 sc->sc_provider->mediasize)); 3247 } 3248 sbuf_printf(sb, "</Synchronized>\n"); 3249 if (disk->d_sync.ds_offset > 0) { 3250 sbuf_printf(sb, "%s<BytesSynced>%jd" 3251 "</BytesSynced>\n", indent, 3252 (intmax_t)disk->d_sync.ds_offset); 3253 } 3254 } 3255 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3256 disk->d_sync.ds_syncid); 3257 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3258 disk->d_genid); 3259 sbuf_printf(sb, "%s<Flags>", indent); 3260 if (disk->d_flags == 0) 3261 sbuf_printf(sb, "NONE"); 3262 else { 3263 int first = 1; 3264 3265 #define ADD_FLAG(flag, name) do { \ 3266 if ((disk->d_flags & (flag)) != 0) { \ 3267 if (!first) \ 3268 sbuf_printf(sb, ", "); \ 3269 else \ 3270 first = 0; \ 3271 sbuf_printf(sb, name); \ 3272 } \ 3273 } while (0) 3274 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3275 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3276 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3277 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3278 "SYNCHRONIZING"); 3279 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3280 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3281 #undef ADD_FLAG 3282 } 3283 sbuf_printf(sb, "</Flags>\n"); 3284 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3285 disk->d_priority); 3286 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3287 g_mirror_disk_state2str(disk->d_state)); 3288 sx_xunlock(&sc->sc_lock); 3289 g_topology_lock(); 3290 } else { 3291 g_topology_unlock(); 3292 sx_xlock(&sc->sc_lock); 3293 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3294 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3295 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3296 sbuf_printf(sb, "%s<Flags>", indent); 3297 if (sc->sc_flags == 0) 3298 sbuf_printf(sb, "NONE"); 3299 else { 3300 int first = 1; 3301 3302 #define ADD_FLAG(flag, name) do { \ 3303 if ((sc->sc_flags & (flag)) != 0) { \ 3304 if (!first) \ 3305 sbuf_printf(sb, ", "); \ 3306 else \ 3307 first = 0; \ 3308 sbuf_printf(sb, name); \ 3309 } \ 3310 } while (0) 3311 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3312 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3313 #undef ADD_FLAG 3314 } 3315 sbuf_printf(sb, "</Flags>\n"); 3316 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3317 (u_int)sc->sc_slice); 3318 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3319 balance_name(sc->sc_balance)); 3320 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3321 sc->sc_ndisks); 3322 sbuf_printf(sb, "%s<State>", indent); 3323 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3324 sbuf_printf(sb, "%s", "STARTING"); 3325 else if (sc->sc_ndisks == 3326 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3327 sbuf_printf(sb, "%s", "COMPLETE"); 3328 else 3329 sbuf_printf(sb, "%s", "DEGRADED"); 3330 sbuf_printf(sb, "</State>\n"); 3331 sx_xunlock(&sc->sc_lock); 3332 g_topology_lock(); 3333 } 3334 } 3335 3336 static void 3337 g_mirror_shutdown_post_sync(void *arg, int howto) 3338 { 3339 struct g_class *mp; 3340 struct g_geom *gp, *gp2; 3341 struct g_mirror_softc *sc; 3342 int error; 3343 3344 mp = arg; 3345 g_topology_lock(); 3346 g_mirror_shutdown = 1; 3347 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3348 if ((sc = gp->softc) == NULL) 3349 continue; 3350 /* Skip synchronization geom. */ 3351 if (gp == sc->sc_sync.ds_geom) 3352 continue; 3353 g_topology_unlock(); 3354 sx_xlock(&sc->sc_lock); 3355 g_mirror_idle(sc, -1); 3356 g_cancel_event(sc); 3357 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3358 if (error != 0) 3359 sx_xunlock(&sc->sc_lock); 3360 g_topology_lock(); 3361 } 3362 g_topology_unlock(); 3363 } 3364 3365 static void 3366 g_mirror_init(struct g_class *mp) 3367 { 3368 3369 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3370 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3371 if (g_mirror_post_sync == NULL) 3372 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3373 } 3374 3375 static void 3376 g_mirror_fini(struct g_class *mp) 3377 { 3378 3379 if (g_mirror_post_sync != NULL) 3380 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); 3381 } 3382 3383 DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3384