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 g_ctl_destroy_geom_t g_mirror_destroy_geom; 83 static g_taste_t g_mirror_taste; 84 static g_init_t g_mirror_init; 85 static g_fini_t g_mirror_fini; 86 static g_provgone_t g_mirror_providergone; 87 static g_resize_t g_mirror_resize; 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 .providergone = g_mirror_providergone, 98 .resize = g_mirror_resize 99 }; 100 101 102 static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 103 static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 104 static void g_mirror_update_device(struct g_mirror_softc *sc, bool force); 105 static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 106 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 107 static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 108 static void g_mirror_register_request(struct bio *bp); 109 static void g_mirror_sync_release(struct g_mirror_softc *sc); 110 111 112 static const char * 113 g_mirror_disk_state2str(int state) 114 { 115 116 switch (state) { 117 case G_MIRROR_DISK_STATE_NONE: 118 return ("NONE"); 119 case G_MIRROR_DISK_STATE_NEW: 120 return ("NEW"); 121 case G_MIRROR_DISK_STATE_ACTIVE: 122 return ("ACTIVE"); 123 case G_MIRROR_DISK_STATE_STALE: 124 return ("STALE"); 125 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 126 return ("SYNCHRONIZING"); 127 case G_MIRROR_DISK_STATE_DISCONNECTED: 128 return ("DISCONNECTED"); 129 case G_MIRROR_DISK_STATE_DESTROY: 130 return ("DESTROY"); 131 default: 132 return ("INVALID"); 133 } 134 } 135 136 static const char * 137 g_mirror_device_state2str(int state) 138 { 139 140 switch (state) { 141 case G_MIRROR_DEVICE_STATE_STARTING: 142 return ("STARTING"); 143 case G_MIRROR_DEVICE_STATE_RUNNING: 144 return ("RUNNING"); 145 default: 146 return ("INVALID"); 147 } 148 } 149 150 static const char * 151 g_mirror_get_diskname(struct g_mirror_disk *disk) 152 { 153 154 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 155 return ("[unknown]"); 156 return (disk->d_name); 157 } 158 159 /* 160 * --- Events handling functions --- 161 * Events in geom_mirror are used to maintain disks and device status 162 * from one thread to simplify locking. 163 */ 164 static void 165 g_mirror_event_free(struct g_mirror_event *ep) 166 { 167 168 free(ep, M_MIRROR); 169 } 170 171 int 172 g_mirror_event_send(void *arg, int state, int flags) 173 { 174 struct g_mirror_softc *sc; 175 struct g_mirror_disk *disk; 176 struct g_mirror_event *ep; 177 int error; 178 179 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 180 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 181 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 182 disk = NULL; 183 sc = arg; 184 } else { 185 disk = arg; 186 sc = disk->d_softc; 187 } 188 ep->e_disk = disk; 189 ep->e_state = state; 190 ep->e_flags = flags; 191 ep->e_error = 0; 192 mtx_lock(&sc->sc_events_mtx); 193 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 194 mtx_unlock(&sc->sc_events_mtx); 195 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 196 mtx_lock(&sc->sc_queue_mtx); 197 wakeup(sc); 198 mtx_unlock(&sc->sc_queue_mtx); 199 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 200 return (0); 201 sx_assert(&sc->sc_lock, SX_XLOCKED); 202 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 203 sx_xunlock(&sc->sc_lock); 204 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 205 mtx_lock(&sc->sc_events_mtx); 206 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 207 hz * 5); 208 } 209 error = ep->e_error; 210 g_mirror_event_free(ep); 211 sx_xlock(&sc->sc_lock); 212 return (error); 213 } 214 215 static struct g_mirror_event * 216 g_mirror_event_get(struct g_mirror_softc *sc) 217 { 218 struct g_mirror_event *ep; 219 220 mtx_lock(&sc->sc_events_mtx); 221 ep = TAILQ_FIRST(&sc->sc_events); 222 mtx_unlock(&sc->sc_events_mtx); 223 return (ep); 224 } 225 226 static void 227 g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 228 { 229 230 mtx_lock(&sc->sc_events_mtx); 231 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 232 mtx_unlock(&sc->sc_events_mtx); 233 } 234 235 static void 236 g_mirror_event_cancel(struct g_mirror_disk *disk) 237 { 238 struct g_mirror_softc *sc; 239 struct g_mirror_event *ep, *tmpep; 240 241 sc = disk->d_softc; 242 sx_assert(&sc->sc_lock, SX_XLOCKED); 243 244 mtx_lock(&sc->sc_events_mtx); 245 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 246 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 247 continue; 248 if (ep->e_disk != disk) 249 continue; 250 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 251 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 252 g_mirror_event_free(ep); 253 else { 254 ep->e_error = ECANCELED; 255 wakeup(ep); 256 } 257 } 258 mtx_unlock(&sc->sc_events_mtx); 259 } 260 261 /* 262 * Return the number of disks in given state. 263 * If state is equal to -1, count all connected disks. 264 */ 265 u_int 266 g_mirror_ndisks(struct g_mirror_softc *sc, int state) 267 { 268 struct g_mirror_disk *disk; 269 u_int n = 0; 270 271 sx_assert(&sc->sc_lock, SX_LOCKED); 272 273 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 274 if (state == -1 || disk->d_state == state) 275 n++; 276 } 277 return (n); 278 } 279 280 /* 281 * Find a disk in mirror by its disk ID. 282 */ 283 static struct g_mirror_disk * 284 g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 285 { 286 struct g_mirror_disk *disk; 287 288 sx_assert(&sc->sc_lock, SX_XLOCKED); 289 290 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 291 if (disk->d_id == id) 292 return (disk); 293 } 294 return (NULL); 295 } 296 297 static u_int 298 g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 299 { 300 struct bio *bp; 301 u_int nreqs = 0; 302 303 mtx_lock(&sc->sc_queue_mtx); 304 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 305 if (bp->bio_from == cp) 306 nreqs++; 307 } 308 mtx_unlock(&sc->sc_queue_mtx); 309 return (nreqs); 310 } 311 312 static int 313 g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 314 { 315 316 if (cp->index > 0) { 317 G_MIRROR_DEBUG(2, 318 "I/O requests for %s exist, can't destroy it now.", 319 cp->provider->name); 320 return (1); 321 } 322 if (g_mirror_nrequests(sc, cp) > 0) { 323 G_MIRROR_DEBUG(2, 324 "I/O requests for %s in queue, can't destroy it now.", 325 cp->provider->name); 326 return (1); 327 } 328 return (0); 329 } 330 331 static void 332 g_mirror_destroy_consumer(void *arg, int flags __unused) 333 { 334 struct g_consumer *cp; 335 336 g_topology_assert(); 337 338 cp = arg; 339 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 340 g_detach(cp); 341 g_destroy_consumer(cp); 342 } 343 344 static void 345 g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 346 { 347 struct g_provider *pp; 348 int retaste_wait; 349 350 g_topology_assert(); 351 352 cp->private = NULL; 353 if (g_mirror_is_busy(sc, cp)) 354 return; 355 pp = cp->provider; 356 retaste_wait = 0; 357 if (cp->acw == 1) { 358 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 359 retaste_wait = 1; 360 } 361 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 362 -cp->acw, -cp->ace, 0); 363 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 364 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 365 if (retaste_wait) { 366 /* 367 * After retaste event was send (inside g_access()), we can send 368 * event to detach and destroy consumer. 369 * A class, which has consumer to the given provider connected 370 * will not receive retaste event for the provider. 371 * This is the way how I ignore retaste events when I close 372 * consumers opened for write: I detach and destroy consumer 373 * after retaste event is sent. 374 */ 375 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 376 return; 377 } 378 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 379 g_detach(cp); 380 g_destroy_consumer(cp); 381 } 382 383 static int 384 g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 385 { 386 struct g_consumer *cp; 387 int error; 388 389 g_topology_assert_not(); 390 KASSERT(disk->d_consumer == NULL, 391 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 392 393 g_topology_lock(); 394 cp = g_new_consumer(disk->d_softc->sc_geom); 395 cp->flags |= G_CF_DIRECT_RECEIVE; 396 error = g_attach(cp, pp); 397 if (error != 0) { 398 g_destroy_consumer(cp); 399 g_topology_unlock(); 400 return (error); 401 } 402 error = g_access(cp, 1, 1, 1); 403 if (error != 0) { 404 g_detach(cp); 405 g_destroy_consumer(cp); 406 g_topology_unlock(); 407 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 408 pp->name, error); 409 return (error); 410 } 411 g_topology_unlock(); 412 disk->d_consumer = cp; 413 disk->d_consumer->private = disk; 414 disk->d_consumer->index = 0; 415 416 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 417 return (0); 418 } 419 420 static void 421 g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 422 { 423 424 g_topology_assert(); 425 426 if (cp == NULL) 427 return; 428 if (cp->provider != NULL) 429 g_mirror_kill_consumer(sc, cp); 430 else 431 g_destroy_consumer(cp); 432 } 433 434 /* 435 * Initialize disk. This means allocate memory, create consumer, attach it 436 * to the provider and open access (r1w1e1) to it. 437 */ 438 static struct g_mirror_disk * 439 g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 440 struct g_mirror_metadata *md, int *errorp) 441 { 442 struct g_mirror_disk *disk; 443 int i, error; 444 445 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 446 if (disk == NULL) { 447 error = ENOMEM; 448 goto fail; 449 } 450 disk->d_softc = sc; 451 error = g_mirror_connect_disk(disk, pp); 452 if (error != 0) 453 goto fail; 454 disk->d_id = md->md_did; 455 disk->d_state = G_MIRROR_DISK_STATE_NONE; 456 disk->d_priority = md->md_priority; 457 disk->d_flags = md->md_dflags; 458 error = g_getattr("GEOM::candelete", disk->d_consumer, &i); 459 if (error == 0 && i != 0) 460 disk->d_flags |= G_MIRROR_DISK_FLAG_CANDELETE; 461 if (md->md_provider[0] != '\0') 462 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 463 disk->d_sync.ds_consumer = NULL; 464 disk->d_sync.ds_offset = md->md_sync_offset; 465 disk->d_sync.ds_offset_done = md->md_sync_offset; 466 disk->d_genid = md->md_genid; 467 disk->d_sync.ds_syncid = md->md_syncid; 468 if (errorp != NULL) 469 *errorp = 0; 470 return (disk); 471 fail: 472 if (errorp != NULL) 473 *errorp = error; 474 if (disk != NULL) 475 free(disk, M_MIRROR); 476 return (NULL); 477 } 478 479 static void 480 g_mirror_destroy_disk(struct g_mirror_disk *disk) 481 { 482 struct g_mirror_softc *sc; 483 484 g_topology_assert_not(); 485 sc = disk->d_softc; 486 sx_assert(&sc->sc_lock, SX_XLOCKED); 487 488 LIST_REMOVE(disk, d_next); 489 g_mirror_event_cancel(disk); 490 if (sc->sc_hint == disk) 491 sc->sc_hint = NULL; 492 switch (disk->d_state) { 493 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 494 g_mirror_sync_stop(disk, 1); 495 /* FALLTHROUGH */ 496 case G_MIRROR_DISK_STATE_NEW: 497 case G_MIRROR_DISK_STATE_STALE: 498 case G_MIRROR_DISK_STATE_ACTIVE: 499 g_topology_lock(); 500 g_mirror_disconnect_consumer(sc, disk->d_consumer); 501 g_topology_unlock(); 502 free(disk, M_MIRROR); 503 break; 504 default: 505 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 506 g_mirror_get_diskname(disk), 507 g_mirror_disk_state2str(disk->d_state))); 508 } 509 } 510 511 static void 512 g_mirror_free_device(struct g_mirror_softc *sc) 513 { 514 515 mtx_destroy(&sc->sc_queue_mtx); 516 mtx_destroy(&sc->sc_events_mtx); 517 mtx_destroy(&sc->sc_done_mtx); 518 sx_destroy(&sc->sc_lock); 519 free(sc, M_MIRROR); 520 } 521 522 static void 523 g_mirror_providergone(struct g_provider *pp) 524 { 525 struct g_mirror_softc *sc = pp->private; 526 527 if ((--sc->sc_refcnt) == 0) 528 g_mirror_free_device(sc); 529 } 530 531 static void 532 g_mirror_destroy_device(struct g_mirror_softc *sc) 533 { 534 struct g_mirror_disk *disk; 535 struct g_mirror_event *ep; 536 struct g_geom *gp; 537 struct g_consumer *cp, *tmpcp; 538 539 g_topology_assert_not(); 540 sx_assert(&sc->sc_lock, SX_XLOCKED); 541 542 gp = sc->sc_geom; 543 if (sc->sc_provider != NULL) 544 g_mirror_destroy_provider(sc); 545 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 546 disk = LIST_FIRST(&sc->sc_disks)) { 547 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 548 g_mirror_update_metadata(disk); 549 g_mirror_destroy_disk(disk); 550 } 551 while ((ep = g_mirror_event_get(sc)) != NULL) { 552 g_mirror_event_remove(sc, ep); 553 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 554 g_mirror_event_free(ep); 555 else { 556 ep->e_error = ECANCELED; 557 ep->e_flags |= G_MIRROR_EVENT_DONE; 558 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 559 mtx_lock(&sc->sc_events_mtx); 560 wakeup(ep); 561 mtx_unlock(&sc->sc_events_mtx); 562 } 563 } 564 callout_drain(&sc->sc_callout); 565 566 g_topology_lock(); 567 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 568 g_mirror_disconnect_consumer(sc, cp); 569 } 570 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 571 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 572 g_wither_geom(gp, ENXIO); 573 sx_xunlock(&sc->sc_lock); 574 if ((--sc->sc_refcnt) == 0) 575 g_mirror_free_device(sc); 576 g_topology_unlock(); 577 } 578 579 static void 580 g_mirror_orphan(struct g_consumer *cp) 581 { 582 struct g_mirror_disk *disk; 583 584 g_topology_assert(); 585 586 disk = cp->private; 587 if (disk == NULL) 588 return; 589 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 590 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 591 G_MIRROR_EVENT_DONTWAIT); 592 } 593 594 /* 595 * Function should return the next active disk on the list. 596 * It is possible that it will be the same disk as given. 597 * If there are no active disks on list, NULL is returned. 598 */ 599 static __inline struct g_mirror_disk * 600 g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 601 { 602 struct g_mirror_disk *dp; 603 604 for (dp = LIST_NEXT(disk, d_next); dp != disk; 605 dp = LIST_NEXT(dp, d_next)) { 606 if (dp == NULL) 607 dp = LIST_FIRST(&sc->sc_disks); 608 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 609 break; 610 } 611 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 612 return (NULL); 613 return (dp); 614 } 615 616 static struct g_mirror_disk * 617 g_mirror_get_disk(struct g_mirror_softc *sc) 618 { 619 struct g_mirror_disk *disk; 620 621 if (sc->sc_hint == NULL) { 622 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 623 if (sc->sc_hint == NULL) 624 return (NULL); 625 } 626 disk = sc->sc_hint; 627 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 628 disk = g_mirror_find_next(sc, disk); 629 if (disk == NULL) 630 return (NULL); 631 } 632 sc->sc_hint = g_mirror_find_next(sc, disk); 633 return (disk); 634 } 635 636 static int 637 g_mirror_write_metadata(struct g_mirror_disk *disk, 638 struct g_mirror_metadata *md) 639 { 640 struct g_mirror_softc *sc; 641 struct g_consumer *cp; 642 off_t offset, length; 643 u_char *sector; 644 int error = 0; 645 646 g_topology_assert_not(); 647 sc = disk->d_softc; 648 sx_assert(&sc->sc_lock, SX_LOCKED); 649 650 cp = disk->d_consumer; 651 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 652 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 653 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 654 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 655 cp->acw, cp->ace)); 656 length = cp->provider->sectorsize; 657 offset = cp->provider->mediasize - length; 658 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 659 if (md != NULL && 660 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) { 661 /* 662 * Handle the case, when the size of parent provider reduced. 663 */ 664 if (offset < md->md_mediasize) 665 error = ENOSPC; 666 else 667 mirror_metadata_encode(md, sector); 668 } 669 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_metadata_write, error); 670 if (error == 0) 671 error = g_write_data(cp, offset, sector, length); 672 free(sector, M_MIRROR); 673 if (error != 0) { 674 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 675 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 676 G_MIRROR_DEBUG(0, "Cannot write metadata on %s " 677 "(device=%s, error=%d).", 678 g_mirror_get_diskname(disk), sc->sc_name, error); 679 } else { 680 G_MIRROR_DEBUG(1, "Cannot write metadata on %s " 681 "(device=%s, error=%d).", 682 g_mirror_get_diskname(disk), sc->sc_name, error); 683 } 684 if (g_mirror_disconnect_on_failure && 685 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 686 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 687 g_mirror_event_send(disk, 688 G_MIRROR_DISK_STATE_DISCONNECTED, 689 G_MIRROR_EVENT_DONTWAIT); 690 } 691 } 692 return (error); 693 } 694 695 static int 696 g_mirror_clear_metadata(struct g_mirror_disk *disk) 697 { 698 int error; 699 700 g_topology_assert_not(); 701 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 702 703 if (disk->d_softc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 704 return (0); 705 error = g_mirror_write_metadata(disk, NULL); 706 if (error == 0) { 707 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 708 g_mirror_get_diskname(disk)); 709 } else { 710 G_MIRROR_DEBUG(0, 711 "Cannot clear metadata on disk %s (error=%d).", 712 g_mirror_get_diskname(disk), error); 713 } 714 return (error); 715 } 716 717 void 718 g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 719 struct g_mirror_metadata *md) 720 { 721 722 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 723 md->md_version = G_MIRROR_VERSION; 724 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 725 md->md_mid = sc->sc_id; 726 md->md_all = sc->sc_ndisks; 727 md->md_slice = sc->sc_slice; 728 md->md_balance = sc->sc_balance; 729 md->md_genid = sc->sc_genid; 730 md->md_mediasize = sc->sc_mediasize; 731 md->md_sectorsize = sc->sc_sectorsize; 732 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 733 bzero(md->md_provider, sizeof(md->md_provider)); 734 if (disk == NULL) { 735 md->md_did = arc4random(); 736 md->md_priority = 0; 737 md->md_syncid = 0; 738 md->md_dflags = 0; 739 md->md_sync_offset = 0; 740 md->md_provsize = 0; 741 } else { 742 md->md_did = disk->d_id; 743 md->md_priority = disk->d_priority; 744 md->md_syncid = disk->d_sync.ds_syncid; 745 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 746 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 747 md->md_sync_offset = disk->d_sync.ds_offset_done; 748 else 749 md->md_sync_offset = 0; 750 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 751 strlcpy(md->md_provider, 752 disk->d_consumer->provider->name, 753 sizeof(md->md_provider)); 754 } 755 md->md_provsize = disk->d_consumer->provider->mediasize; 756 } 757 } 758 759 void 760 g_mirror_update_metadata(struct g_mirror_disk *disk) 761 { 762 struct g_mirror_softc *sc; 763 struct g_mirror_metadata md; 764 int error; 765 766 g_topology_assert_not(); 767 sc = disk->d_softc; 768 sx_assert(&sc->sc_lock, SX_LOCKED); 769 770 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 771 return; 772 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) 773 g_mirror_fill_metadata(sc, disk, &md); 774 error = g_mirror_write_metadata(disk, &md); 775 if (error == 0) { 776 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 777 g_mirror_get_diskname(disk)); 778 } else { 779 G_MIRROR_DEBUG(0, 780 "Cannot update metadata on disk %s (error=%d).", 781 g_mirror_get_diskname(disk), error); 782 } 783 } 784 785 static void 786 g_mirror_bump_syncid(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_syncid++; 797 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 798 sc->sc_syncid); 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_sync.ds_syncid = sc->sc_syncid; 803 g_mirror_update_metadata(disk); 804 } 805 } 806 } 807 808 static void 809 g_mirror_bump_genid(struct g_mirror_softc *sc) 810 { 811 struct g_mirror_disk *disk; 812 813 g_topology_assert_not(); 814 sx_assert(&sc->sc_lock, SX_XLOCKED); 815 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 816 ("%s called with no active disks (device=%s).", __func__, 817 sc->sc_name)); 818 819 sc->sc_genid++; 820 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 821 sc->sc_genid); 822 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 823 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 824 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 825 disk->d_genid = sc->sc_genid; 826 g_mirror_update_metadata(disk); 827 } 828 } 829 } 830 831 static int 832 g_mirror_idle(struct g_mirror_softc *sc, int acw) 833 { 834 struct g_mirror_disk *disk; 835 int timeout; 836 837 g_topology_assert_not(); 838 sx_assert(&sc->sc_lock, SX_XLOCKED); 839 840 if (sc->sc_provider == NULL) 841 return (0); 842 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 843 return (0); 844 if (sc->sc_idle) 845 return (0); 846 if (sc->sc_writes > 0) 847 return (0); 848 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 849 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); 850 if (!g_mirror_shutdown && timeout > 0) 851 return (timeout); 852 } 853 sc->sc_idle = 1; 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 clean.", 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 return (0); 863 } 864 865 static void 866 g_mirror_unidle(struct g_mirror_softc *sc) 867 { 868 struct g_mirror_disk *disk; 869 870 g_topology_assert_not(); 871 sx_assert(&sc->sc_lock, SX_XLOCKED); 872 873 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 874 return; 875 sc->sc_idle = 0; 876 sc->sc_last_write = time_uptime; 877 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 878 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 879 continue; 880 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 881 g_mirror_get_diskname(disk), sc->sc_name); 882 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 883 g_mirror_update_metadata(disk); 884 } 885 } 886 887 static void 888 g_mirror_flush_done(struct bio *bp) 889 { 890 struct g_mirror_softc *sc; 891 struct bio *pbp; 892 893 pbp = bp->bio_parent; 894 sc = pbp->bio_to->private; 895 mtx_lock(&sc->sc_done_mtx); 896 if (pbp->bio_error == 0) 897 pbp->bio_error = bp->bio_error; 898 pbp->bio_completed += bp->bio_completed; 899 pbp->bio_inbed++; 900 if (pbp->bio_children == pbp->bio_inbed) { 901 mtx_unlock(&sc->sc_done_mtx); 902 g_io_deliver(pbp, pbp->bio_error); 903 } else 904 mtx_unlock(&sc->sc_done_mtx); 905 g_destroy_bio(bp); 906 } 907 908 static void 909 g_mirror_done(struct bio *bp) 910 { 911 struct g_mirror_softc *sc; 912 913 sc = bp->bio_from->geom->softc; 914 bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR; 915 mtx_lock(&sc->sc_queue_mtx); 916 bioq_insert_tail(&sc->sc_queue, bp); 917 mtx_unlock(&sc->sc_queue_mtx); 918 wakeup(sc); 919 } 920 921 static void 922 g_mirror_regular_request(struct bio *bp) 923 { 924 struct g_mirror_softc *sc; 925 struct g_mirror_disk *disk; 926 struct bio *pbp; 927 928 g_topology_assert_not(); 929 930 pbp = bp->bio_parent; 931 sc = pbp->bio_to->private; 932 bp->bio_from->index--; 933 if (bp->bio_cmd == BIO_WRITE) 934 sc->sc_writes--; 935 disk = bp->bio_from->private; 936 if (disk == NULL) { 937 g_topology_lock(); 938 g_mirror_kill_consumer(sc, bp->bio_from); 939 g_topology_unlock(); 940 } 941 942 if (bp->bio_cmd == BIO_READ) 943 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_read, 944 bp->bio_error); 945 else if (bp->bio_cmd == BIO_WRITE) 946 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_write, 947 bp->bio_error); 948 949 pbp->bio_inbed++; 950 KASSERT(pbp->bio_inbed <= pbp->bio_children, 951 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 952 pbp->bio_children)); 953 if (bp->bio_error == 0 && pbp->bio_error == 0) { 954 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 955 g_destroy_bio(bp); 956 if (pbp->bio_children == pbp->bio_inbed) { 957 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 958 pbp->bio_completed = pbp->bio_length; 959 if (pbp->bio_cmd == BIO_WRITE || 960 pbp->bio_cmd == BIO_DELETE) { 961 bioq_remove(&sc->sc_inflight, pbp); 962 /* Release delayed sync requests if possible. */ 963 g_mirror_sync_release(sc); 964 } 965 g_io_deliver(pbp, pbp->bio_error); 966 } 967 return; 968 } else if (bp->bio_error != 0) { 969 if (pbp->bio_error == 0) 970 pbp->bio_error = bp->bio_error; 971 if (disk != NULL) { 972 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 973 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 974 G_MIRROR_LOGREQ(0, bp, 975 "Request failed (error=%d).", 976 bp->bio_error); 977 } else { 978 G_MIRROR_LOGREQ(1, bp, 979 "Request failed (error=%d).", 980 bp->bio_error); 981 } 982 if (g_mirror_disconnect_on_failure && 983 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) 984 { 985 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 986 g_mirror_event_send(disk, 987 G_MIRROR_DISK_STATE_DISCONNECTED, 988 G_MIRROR_EVENT_DONTWAIT); 989 } 990 } 991 switch (pbp->bio_cmd) { 992 case BIO_DELETE: 993 case BIO_WRITE: 994 pbp->bio_inbed--; 995 pbp->bio_children--; 996 break; 997 } 998 } 999 g_destroy_bio(bp); 1000 1001 switch (pbp->bio_cmd) { 1002 case BIO_READ: 1003 if (pbp->bio_inbed < pbp->bio_children) 1004 break; 1005 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 1006 g_io_deliver(pbp, pbp->bio_error); 1007 else { 1008 pbp->bio_error = 0; 1009 mtx_lock(&sc->sc_queue_mtx); 1010 bioq_insert_tail(&sc->sc_queue, pbp); 1011 mtx_unlock(&sc->sc_queue_mtx); 1012 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1013 wakeup(sc); 1014 } 1015 break; 1016 case BIO_DELETE: 1017 case BIO_WRITE: 1018 if (pbp->bio_children == 0) { 1019 /* 1020 * All requests failed. 1021 */ 1022 } else if (pbp->bio_inbed < pbp->bio_children) { 1023 /* Do nothing. */ 1024 break; 1025 } else if (pbp->bio_children == pbp->bio_inbed) { 1026 /* Some requests succeeded. */ 1027 pbp->bio_error = 0; 1028 pbp->bio_completed = pbp->bio_length; 1029 } 1030 bioq_remove(&sc->sc_inflight, pbp); 1031 /* Release delayed sync requests if possible. */ 1032 g_mirror_sync_release(sc); 1033 g_io_deliver(pbp, pbp->bio_error); 1034 break; 1035 default: 1036 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 1037 break; 1038 } 1039 } 1040 1041 static void 1042 g_mirror_sync_done(struct bio *bp) 1043 { 1044 struct g_mirror_softc *sc; 1045 1046 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1047 sc = bp->bio_from->geom->softc; 1048 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; 1049 mtx_lock(&sc->sc_queue_mtx); 1050 bioq_insert_tail(&sc->sc_queue, bp); 1051 mtx_unlock(&sc->sc_queue_mtx); 1052 wakeup(sc); 1053 } 1054 1055 static void 1056 g_mirror_candelete(struct bio *bp) 1057 { 1058 struct g_mirror_softc *sc; 1059 struct g_mirror_disk *disk; 1060 int *val; 1061 1062 sc = bp->bio_to->private; 1063 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1064 if (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) 1065 break; 1066 } 1067 val = (int *)bp->bio_data; 1068 *val = (disk != NULL); 1069 g_io_deliver(bp, 0); 1070 } 1071 1072 static void 1073 g_mirror_kernel_dump(struct bio *bp) 1074 { 1075 struct g_mirror_softc *sc; 1076 struct g_mirror_disk *disk; 1077 struct bio *cbp; 1078 struct g_kerneldump *gkd; 1079 1080 /* 1081 * We configure dumping to the first component, because this component 1082 * will be used for reading with 'prefer' balance algorithm. 1083 * If the component with the highest priority is currently disconnected 1084 * we will not be able to read the dump after the reboot if it will be 1085 * connected and synchronized later. Can we do something better? 1086 */ 1087 sc = bp->bio_to->private; 1088 disk = LIST_FIRST(&sc->sc_disks); 1089 1090 gkd = (struct g_kerneldump *)bp->bio_data; 1091 if (gkd->length > bp->bio_to->mediasize) 1092 gkd->length = bp->bio_to->mediasize; 1093 cbp = g_clone_bio(bp); 1094 if (cbp == NULL) { 1095 g_io_deliver(bp, ENOMEM); 1096 return; 1097 } 1098 cbp->bio_done = g_std_done; 1099 g_io_request(cbp, disk->d_consumer); 1100 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1101 g_mirror_get_diskname(disk)); 1102 } 1103 1104 static void 1105 g_mirror_flush(struct g_mirror_softc *sc, struct bio *bp) 1106 { 1107 struct bio_queue_head queue; 1108 struct g_mirror_disk *disk; 1109 struct g_consumer *cp; 1110 struct bio *cbp; 1111 1112 bioq_init(&queue); 1113 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1114 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1115 continue; 1116 cbp = g_clone_bio(bp); 1117 if (cbp == NULL) { 1118 while ((cbp = bioq_takefirst(&queue)) != NULL) 1119 g_destroy_bio(cbp); 1120 if (bp->bio_error == 0) 1121 bp->bio_error = ENOMEM; 1122 g_io_deliver(bp, bp->bio_error); 1123 return; 1124 } 1125 bioq_insert_tail(&queue, cbp); 1126 cbp->bio_done = g_mirror_flush_done; 1127 cbp->bio_caller1 = disk; 1128 cbp->bio_to = disk->d_consumer->provider; 1129 } 1130 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1131 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1132 disk = cbp->bio_caller1; 1133 cbp->bio_caller1 = NULL; 1134 cp = disk->d_consumer; 1135 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1136 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1137 cp->acr, cp->acw, cp->ace)); 1138 g_io_request(cbp, disk->d_consumer); 1139 } 1140 } 1141 1142 static void 1143 g_mirror_start(struct bio *bp) 1144 { 1145 struct g_mirror_softc *sc; 1146 1147 sc = bp->bio_to->private; 1148 /* 1149 * If sc == NULL or there are no valid disks, provider's error 1150 * should be set and g_mirror_start() should not be called at all. 1151 */ 1152 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1153 ("Provider's error should be set (error=%d)(mirror=%s).", 1154 bp->bio_to->error, bp->bio_to->name)); 1155 G_MIRROR_LOGREQ(3, bp, "Request received."); 1156 1157 switch (bp->bio_cmd) { 1158 case BIO_READ: 1159 case BIO_WRITE: 1160 case BIO_DELETE: 1161 break; 1162 case BIO_FLUSH: 1163 g_mirror_flush(sc, bp); 1164 return; 1165 case BIO_GETATTR: 1166 if (!strcmp(bp->bio_attribute, "GEOM::candelete")) { 1167 g_mirror_candelete(bp); 1168 return; 1169 } else if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1170 g_mirror_kernel_dump(bp); 1171 return; 1172 } 1173 /* FALLTHROUGH */ 1174 default: 1175 g_io_deliver(bp, EOPNOTSUPP); 1176 return; 1177 } 1178 mtx_lock(&sc->sc_queue_mtx); 1179 bioq_insert_tail(&sc->sc_queue, bp); 1180 mtx_unlock(&sc->sc_queue_mtx); 1181 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1182 wakeup(sc); 1183 } 1184 1185 /* 1186 * Return TRUE if the given request is colliding with a in-progress 1187 * synchronization request. 1188 */ 1189 static int 1190 g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1191 { 1192 struct g_mirror_disk *disk; 1193 struct bio *sbp; 1194 off_t rstart, rend, sstart, send; 1195 u_int i; 1196 1197 if (sc->sc_sync.ds_ndisks == 0) 1198 return (0); 1199 rstart = bp->bio_offset; 1200 rend = bp->bio_offset + bp->bio_length; 1201 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1202 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1203 continue; 1204 for (i = 0; i < g_mirror_syncreqs; i++) { 1205 sbp = disk->d_sync.ds_bios[i]; 1206 if (sbp == NULL) 1207 continue; 1208 sstart = sbp->bio_offset; 1209 send = sbp->bio_offset + sbp->bio_length; 1210 if (rend > sstart && rstart < send) 1211 return (1); 1212 } 1213 } 1214 return (0); 1215 } 1216 1217 /* 1218 * Return TRUE if the given sync request is colliding with a in-progress regular 1219 * request. 1220 */ 1221 static int 1222 g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1223 { 1224 off_t rstart, rend, sstart, send; 1225 struct bio *bp; 1226 1227 if (sc->sc_sync.ds_ndisks == 0) 1228 return (0); 1229 sstart = sbp->bio_offset; 1230 send = sbp->bio_offset + sbp->bio_length; 1231 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { 1232 rstart = bp->bio_offset; 1233 rend = bp->bio_offset + bp->bio_length; 1234 if (rend > sstart && rstart < send) 1235 return (1); 1236 } 1237 return (0); 1238 } 1239 1240 /* 1241 * Puts request onto delayed queue. 1242 */ 1243 static void 1244 g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1245 { 1246 1247 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1248 bioq_insert_head(&sc->sc_regular_delayed, bp); 1249 } 1250 1251 /* 1252 * Puts synchronization request onto delayed queue. 1253 */ 1254 static void 1255 g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1256 { 1257 1258 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1259 bioq_insert_tail(&sc->sc_sync_delayed, bp); 1260 } 1261 1262 /* 1263 * Releases delayed regular requests which don't collide anymore with sync 1264 * requests. 1265 */ 1266 static void 1267 g_mirror_regular_release(struct g_mirror_softc *sc) 1268 { 1269 struct bio *bp, *bp2; 1270 1271 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { 1272 if (g_mirror_sync_collision(sc, bp)) 1273 continue; 1274 bioq_remove(&sc->sc_regular_delayed, bp); 1275 G_MIRROR_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1276 mtx_lock(&sc->sc_queue_mtx); 1277 bioq_insert_head(&sc->sc_queue, bp); 1278 mtx_unlock(&sc->sc_queue_mtx); 1279 } 1280 } 1281 1282 /* 1283 * Releases delayed sync requests which don't collide anymore with regular 1284 * requests. 1285 */ 1286 static void 1287 g_mirror_sync_release(struct g_mirror_softc *sc) 1288 { 1289 struct bio *bp, *bp2; 1290 1291 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { 1292 if (g_mirror_regular_collision(sc, bp)) 1293 continue; 1294 bioq_remove(&sc->sc_sync_delayed, bp); 1295 G_MIRROR_LOGREQ(2, bp, 1296 "Releasing delayed synchronization request."); 1297 g_io_request(bp, bp->bio_from); 1298 } 1299 } 1300 1301 /* 1302 * Free a synchronization request and clear its slot in the array. 1303 */ 1304 static void 1305 g_mirror_sync_request_free(struct g_mirror_disk *disk, struct bio *bp) 1306 { 1307 int i; 1308 1309 if (disk != NULL && disk->d_sync.ds_bios != NULL) { 1310 i = (int)(uintptr_t)bp->bio_caller1; 1311 disk->d_sync.ds_bios[i] = NULL; 1312 } 1313 free(bp->bio_data, M_MIRROR); 1314 g_destroy_bio(bp); 1315 } 1316 1317 /* 1318 * Handle synchronization requests. 1319 * Every synchronization request is two-steps process: first, READ request is 1320 * send to active provider and then WRITE request (with read data) to the provider 1321 * being synchronized. When WRITE is finished, new synchronization request is 1322 * send. 1323 */ 1324 static void 1325 g_mirror_sync_request(struct bio *bp) 1326 { 1327 struct g_mirror_softc *sc; 1328 struct g_mirror_disk *disk; 1329 struct g_mirror_disk_sync *sync; 1330 1331 bp->bio_from->index--; 1332 sc = bp->bio_from->geom->softc; 1333 disk = bp->bio_from->private; 1334 if (disk == NULL) { 1335 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1336 g_topology_lock(); 1337 g_mirror_kill_consumer(sc, bp->bio_from); 1338 g_topology_unlock(); 1339 g_mirror_sync_request_free(NULL, bp); 1340 sx_xlock(&sc->sc_lock); 1341 return; 1342 } 1343 1344 /* 1345 * Synchronization request. 1346 */ 1347 switch (bp->bio_cmd) { 1348 case BIO_READ: 1349 { 1350 struct g_consumer *cp; 1351 1352 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_read, 1353 bp->bio_error); 1354 1355 if (bp->bio_error != 0) { 1356 G_MIRROR_LOGREQ(0, bp, 1357 "Synchronization request failed (error=%d).", 1358 bp->bio_error); 1359 g_mirror_sync_request_free(disk, bp); 1360 return; 1361 } 1362 G_MIRROR_LOGREQ(3, bp, 1363 "Synchronization request half-finished."); 1364 bp->bio_cmd = BIO_WRITE; 1365 bp->bio_cflags = 0; 1366 cp = disk->d_consumer; 1367 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1368 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1369 cp->acr, cp->acw, cp->ace)); 1370 cp->index++; 1371 g_io_request(bp, cp); 1372 return; 1373 } 1374 case BIO_WRITE: 1375 { 1376 off_t offset; 1377 void *data; 1378 int i; 1379 1380 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_write, 1381 bp->bio_error); 1382 1383 if (bp->bio_error != 0) { 1384 G_MIRROR_LOGREQ(0, bp, 1385 "Synchronization request failed (error=%d).", 1386 bp->bio_error); 1387 g_mirror_sync_request_free(disk, bp); 1388 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1389 g_mirror_event_send(disk, 1390 G_MIRROR_DISK_STATE_DISCONNECTED, 1391 G_MIRROR_EVENT_DONTWAIT); 1392 return; 1393 } 1394 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1395 sync = &disk->d_sync; 1396 if (sync->ds_offset >= sc->sc_mediasize || 1397 sync->ds_consumer == NULL || 1398 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1399 /* Don't send more synchronization requests. */ 1400 sync->ds_inflight--; 1401 g_mirror_sync_request_free(disk, bp); 1402 if (sync->ds_inflight > 0) 1403 return; 1404 if (sync->ds_consumer == NULL || 1405 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1406 return; 1407 } 1408 /* Disk up-to-date, activate it. */ 1409 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1410 G_MIRROR_EVENT_DONTWAIT); 1411 return; 1412 } 1413 1414 /* Send next synchronization request. */ 1415 data = bp->bio_data; 1416 g_reset_bio(bp); 1417 bp->bio_cmd = BIO_READ; 1418 bp->bio_offset = sync->ds_offset; 1419 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1420 sync->ds_offset += bp->bio_length; 1421 bp->bio_done = g_mirror_sync_done; 1422 bp->bio_data = data; 1423 bp->bio_from = sync->ds_consumer; 1424 bp->bio_to = sc->sc_provider; 1425 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1426 sync->ds_consumer->index++; 1427 /* 1428 * Delay the request if it is colliding with a regular request. 1429 */ 1430 if (g_mirror_regular_collision(sc, bp)) 1431 g_mirror_sync_delay(sc, bp); 1432 else 1433 g_io_request(bp, sync->ds_consumer); 1434 1435 /* Release delayed requests if possible. */ 1436 g_mirror_regular_release(sc); 1437 1438 /* Find the smallest offset */ 1439 offset = sc->sc_mediasize; 1440 for (i = 0; i < g_mirror_syncreqs; i++) { 1441 bp = sync->ds_bios[i]; 1442 if (bp->bio_offset < offset) 1443 offset = bp->bio_offset; 1444 } 1445 if (sync->ds_offset_done + (MAXPHYS * 100) < offset) { 1446 /* Update offset_done on every 100 blocks. */ 1447 sync->ds_offset_done = offset; 1448 g_mirror_update_metadata(disk); 1449 } 1450 return; 1451 } 1452 default: 1453 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1454 bp->bio_cmd, sc->sc_name)); 1455 break; 1456 } 1457 } 1458 1459 static void 1460 g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1461 { 1462 struct g_mirror_disk *disk; 1463 struct g_consumer *cp; 1464 struct bio *cbp; 1465 1466 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1467 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1468 break; 1469 } 1470 if (disk == NULL) { 1471 if (bp->bio_error == 0) 1472 bp->bio_error = ENXIO; 1473 g_io_deliver(bp, bp->bio_error); 1474 return; 1475 } 1476 cbp = g_clone_bio(bp); 1477 if (cbp == NULL) { 1478 if (bp->bio_error == 0) 1479 bp->bio_error = ENOMEM; 1480 g_io_deliver(bp, bp->bio_error); 1481 return; 1482 } 1483 /* 1484 * Fill in the component buf structure. 1485 */ 1486 cp = disk->d_consumer; 1487 cbp->bio_done = g_mirror_done; 1488 cbp->bio_to = cp->provider; 1489 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1490 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1491 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1492 cp->acw, cp->ace)); 1493 cp->index++; 1494 g_io_request(cbp, cp); 1495 } 1496 1497 static void 1498 g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1499 { 1500 struct g_mirror_disk *disk; 1501 struct g_consumer *cp; 1502 struct bio *cbp; 1503 1504 disk = g_mirror_get_disk(sc); 1505 if (disk == NULL) { 1506 if (bp->bio_error == 0) 1507 bp->bio_error = ENXIO; 1508 g_io_deliver(bp, bp->bio_error); 1509 return; 1510 } 1511 cbp = g_clone_bio(bp); 1512 if (cbp == NULL) { 1513 if (bp->bio_error == 0) 1514 bp->bio_error = ENOMEM; 1515 g_io_deliver(bp, bp->bio_error); 1516 return; 1517 } 1518 /* 1519 * Fill in the component buf structure. 1520 */ 1521 cp = disk->d_consumer; 1522 cbp->bio_done = g_mirror_done; 1523 cbp->bio_to = cp->provider; 1524 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1525 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1526 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1527 cp->acw, cp->ace)); 1528 cp->index++; 1529 g_io_request(cbp, cp); 1530 } 1531 1532 #define TRACK_SIZE (1 * 1024 * 1024) 1533 #define LOAD_SCALE 256 1534 #define ABS(x) (((x) >= 0) ? (x) : (-(x))) 1535 1536 static void 1537 g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1538 { 1539 struct g_mirror_disk *disk, *dp; 1540 struct g_consumer *cp; 1541 struct bio *cbp; 1542 int prio, best; 1543 1544 /* Find a disk with the smallest load. */ 1545 disk = NULL; 1546 best = INT_MAX; 1547 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1548 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1549 continue; 1550 prio = dp->load; 1551 /* If disk head is precisely in position - highly prefer it. */ 1552 if (dp->d_last_offset == bp->bio_offset) 1553 prio -= 2 * LOAD_SCALE; 1554 else 1555 /* If disk head is close to position - prefer it. */ 1556 if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE) 1557 prio -= 1 * LOAD_SCALE; 1558 if (prio <= best) { 1559 disk = dp; 1560 best = prio; 1561 } 1562 } 1563 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1564 cbp = g_clone_bio(bp); 1565 if (cbp == NULL) { 1566 if (bp->bio_error == 0) 1567 bp->bio_error = ENOMEM; 1568 g_io_deliver(bp, bp->bio_error); 1569 return; 1570 } 1571 /* 1572 * Fill in the component buf structure. 1573 */ 1574 cp = disk->d_consumer; 1575 cbp->bio_done = g_mirror_done; 1576 cbp->bio_to = cp->provider; 1577 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1578 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1579 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1580 cp->acw, cp->ace)); 1581 cp->index++; 1582 /* Remember last head position */ 1583 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1584 /* Update loads. */ 1585 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1586 dp->load = (dp->d_consumer->index * LOAD_SCALE + 1587 dp->load * 7) / 8; 1588 } 1589 g_io_request(cbp, cp); 1590 } 1591 1592 static void 1593 g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1594 { 1595 struct bio_queue_head queue; 1596 struct g_mirror_disk *disk; 1597 struct g_consumer *cp; 1598 struct bio *cbp; 1599 off_t left, mod, offset, slice; 1600 u_char *data; 1601 u_int ndisks; 1602 1603 if (bp->bio_length <= sc->sc_slice) { 1604 g_mirror_request_round_robin(sc, bp); 1605 return; 1606 } 1607 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1608 slice = bp->bio_length / ndisks; 1609 mod = slice % sc->sc_provider->sectorsize; 1610 if (mod != 0) 1611 slice += sc->sc_provider->sectorsize - mod; 1612 /* 1613 * Allocate all bios before sending any request, so we can 1614 * return ENOMEM in nice and clean way. 1615 */ 1616 left = bp->bio_length; 1617 offset = bp->bio_offset; 1618 data = bp->bio_data; 1619 bioq_init(&queue); 1620 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1621 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1622 continue; 1623 cbp = g_clone_bio(bp); 1624 if (cbp == NULL) { 1625 while ((cbp = bioq_takefirst(&queue)) != NULL) 1626 g_destroy_bio(cbp); 1627 if (bp->bio_error == 0) 1628 bp->bio_error = ENOMEM; 1629 g_io_deliver(bp, bp->bio_error); 1630 return; 1631 } 1632 bioq_insert_tail(&queue, cbp); 1633 cbp->bio_done = g_mirror_done; 1634 cbp->bio_caller1 = disk; 1635 cbp->bio_to = disk->d_consumer->provider; 1636 cbp->bio_offset = offset; 1637 cbp->bio_data = data; 1638 cbp->bio_length = MIN(left, slice); 1639 left -= cbp->bio_length; 1640 if (left == 0) 1641 break; 1642 offset += cbp->bio_length; 1643 data += cbp->bio_length; 1644 } 1645 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1646 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1647 disk = cbp->bio_caller1; 1648 cbp->bio_caller1 = NULL; 1649 cp = disk->d_consumer; 1650 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1651 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1652 cp->acr, cp->acw, cp->ace)); 1653 disk->d_consumer->index++; 1654 g_io_request(cbp, disk->d_consumer); 1655 } 1656 } 1657 1658 static void 1659 g_mirror_register_request(struct bio *bp) 1660 { 1661 struct g_mirror_softc *sc; 1662 1663 sc = bp->bio_to->private; 1664 switch (bp->bio_cmd) { 1665 case BIO_READ: 1666 switch (sc->sc_balance) { 1667 case G_MIRROR_BALANCE_LOAD: 1668 g_mirror_request_load(sc, bp); 1669 break; 1670 case G_MIRROR_BALANCE_PREFER: 1671 g_mirror_request_prefer(sc, bp); 1672 break; 1673 case G_MIRROR_BALANCE_ROUND_ROBIN: 1674 g_mirror_request_round_robin(sc, bp); 1675 break; 1676 case G_MIRROR_BALANCE_SPLIT: 1677 g_mirror_request_split(sc, bp); 1678 break; 1679 } 1680 return; 1681 case BIO_WRITE: 1682 case BIO_DELETE: 1683 { 1684 struct g_mirror_disk *disk; 1685 struct g_mirror_disk_sync *sync; 1686 struct bio_queue_head queue; 1687 struct g_consumer *cp; 1688 struct bio *cbp; 1689 1690 /* 1691 * Delay the request if it is colliding with a synchronization 1692 * request. 1693 */ 1694 if (g_mirror_sync_collision(sc, bp)) { 1695 g_mirror_regular_delay(sc, bp); 1696 return; 1697 } 1698 1699 if (sc->sc_idle) 1700 g_mirror_unidle(sc); 1701 else 1702 sc->sc_last_write = time_uptime; 1703 1704 /* 1705 * Bump syncid on first write. 1706 */ 1707 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1708 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1709 g_mirror_bump_syncid(sc); 1710 } 1711 1712 /* 1713 * Allocate all bios before sending any request, so we can 1714 * return ENOMEM in nice and clean way. 1715 */ 1716 bioq_init(&queue); 1717 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1718 sync = &disk->d_sync; 1719 switch (disk->d_state) { 1720 case G_MIRROR_DISK_STATE_ACTIVE: 1721 break; 1722 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1723 if (bp->bio_offset >= sync->ds_offset) 1724 continue; 1725 break; 1726 default: 1727 continue; 1728 } 1729 if (bp->bio_cmd == BIO_DELETE && 1730 (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0) 1731 continue; 1732 cbp = g_clone_bio(bp); 1733 if (cbp == NULL) { 1734 while ((cbp = bioq_takefirst(&queue)) != NULL) 1735 g_destroy_bio(cbp); 1736 if (bp->bio_error == 0) 1737 bp->bio_error = ENOMEM; 1738 g_io_deliver(bp, bp->bio_error); 1739 return; 1740 } 1741 bioq_insert_tail(&queue, cbp); 1742 cbp->bio_done = g_mirror_done; 1743 cp = disk->d_consumer; 1744 cbp->bio_caller1 = cp; 1745 cbp->bio_to = cp->provider; 1746 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1747 ("Consumer %s not opened (r%dw%de%d).", 1748 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1749 } 1750 if (bioq_first(&queue) == NULL) { 1751 g_io_deliver(bp, EOPNOTSUPP); 1752 return; 1753 } 1754 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1755 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1756 cp = cbp->bio_caller1; 1757 cbp->bio_caller1 = NULL; 1758 cp->index++; 1759 sc->sc_writes++; 1760 g_io_request(cbp, cp); 1761 } 1762 /* 1763 * Put request onto inflight queue, so we can check if new 1764 * synchronization requests don't collide with it. 1765 */ 1766 bioq_insert_tail(&sc->sc_inflight, bp); 1767 return; 1768 } 1769 default: 1770 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1771 bp->bio_cmd, sc->sc_name)); 1772 break; 1773 } 1774 } 1775 1776 static int 1777 g_mirror_can_destroy(struct g_mirror_softc *sc) 1778 { 1779 struct g_geom *gp; 1780 struct g_consumer *cp; 1781 1782 g_topology_assert(); 1783 gp = sc->sc_geom; 1784 if (gp->softc == NULL) 1785 return (1); 1786 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0) 1787 return (0); 1788 LIST_FOREACH(cp, &gp->consumer, consumer) { 1789 if (g_mirror_is_busy(sc, cp)) 1790 return (0); 1791 } 1792 gp = sc->sc_sync.ds_geom; 1793 LIST_FOREACH(cp, &gp->consumer, consumer) { 1794 if (g_mirror_is_busy(sc, cp)) 1795 return (0); 1796 } 1797 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1798 sc->sc_name); 1799 return (1); 1800 } 1801 1802 static int 1803 g_mirror_try_destroy(struct g_mirror_softc *sc) 1804 { 1805 1806 if (sc->sc_rootmount != NULL) { 1807 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1808 sc->sc_rootmount); 1809 root_mount_rel(sc->sc_rootmount); 1810 sc->sc_rootmount = NULL; 1811 } 1812 g_topology_lock(); 1813 if (!g_mirror_can_destroy(sc)) { 1814 g_topology_unlock(); 1815 return (0); 1816 } 1817 sc->sc_geom->softc = NULL; 1818 sc->sc_sync.ds_geom->softc = NULL; 1819 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1820 g_topology_unlock(); 1821 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1822 &sc->sc_worker); 1823 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1824 sx_xunlock(&sc->sc_lock); 1825 wakeup(&sc->sc_worker); 1826 sc->sc_worker = NULL; 1827 } else { 1828 g_topology_unlock(); 1829 g_mirror_destroy_device(sc); 1830 } 1831 return (1); 1832 } 1833 1834 /* 1835 * Worker thread. 1836 */ 1837 static void 1838 g_mirror_worker(void *arg) 1839 { 1840 struct g_mirror_softc *sc; 1841 struct g_mirror_event *ep; 1842 struct bio *bp; 1843 int timeout; 1844 1845 sc = arg; 1846 thread_lock(curthread); 1847 sched_prio(curthread, PRIBIO); 1848 thread_unlock(curthread); 1849 1850 sx_xlock(&sc->sc_lock); 1851 for (;;) { 1852 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1853 /* 1854 * First take a look at events. 1855 * This is important to handle events before any I/O requests. 1856 */ 1857 ep = g_mirror_event_get(sc); 1858 if (ep != NULL) { 1859 g_mirror_event_remove(sc, ep); 1860 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1861 /* Update only device status. */ 1862 G_MIRROR_DEBUG(3, 1863 "Running event for device %s.", 1864 sc->sc_name); 1865 ep->e_error = 0; 1866 g_mirror_update_device(sc, true); 1867 } else { 1868 /* Update disk status. */ 1869 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1870 g_mirror_get_diskname(ep->e_disk)); 1871 ep->e_error = g_mirror_update_disk(ep->e_disk, 1872 ep->e_state); 1873 if (ep->e_error == 0) 1874 g_mirror_update_device(sc, false); 1875 } 1876 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1877 KASSERT(ep->e_error == 0, 1878 ("Error cannot be handled.")); 1879 g_mirror_event_free(ep); 1880 } else { 1881 ep->e_flags |= G_MIRROR_EVENT_DONE; 1882 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1883 ep); 1884 mtx_lock(&sc->sc_events_mtx); 1885 wakeup(ep); 1886 mtx_unlock(&sc->sc_events_mtx); 1887 } 1888 if ((sc->sc_flags & 1889 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1890 if (g_mirror_try_destroy(sc)) { 1891 curthread->td_pflags &= ~TDP_GEOM; 1892 G_MIRROR_DEBUG(1, "Thread exiting."); 1893 kproc_exit(0); 1894 } 1895 } 1896 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1897 continue; 1898 } 1899 /* 1900 * Check if we can mark array as CLEAN and if we can't take 1901 * how much seconds should we wait. 1902 */ 1903 timeout = g_mirror_idle(sc, -1); 1904 /* 1905 * Now I/O requests. 1906 */ 1907 /* Get first request from the queue. */ 1908 mtx_lock(&sc->sc_queue_mtx); 1909 bp = bioq_takefirst(&sc->sc_queue); 1910 if (bp == NULL) { 1911 if ((sc->sc_flags & 1912 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1913 mtx_unlock(&sc->sc_queue_mtx); 1914 if (g_mirror_try_destroy(sc)) { 1915 curthread->td_pflags &= ~TDP_GEOM; 1916 G_MIRROR_DEBUG(1, "Thread exiting."); 1917 kproc_exit(0); 1918 } 1919 mtx_lock(&sc->sc_queue_mtx); 1920 } 1921 sx_xunlock(&sc->sc_lock); 1922 /* 1923 * XXX: We can miss an event here, because an event 1924 * can be added without sx-device-lock and without 1925 * mtx-queue-lock. Maybe I should just stop using 1926 * dedicated mutex for events synchronization and 1927 * stick with the queue lock? 1928 * The event will hang here until next I/O request 1929 * or next event is received. 1930 */ 1931 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1932 timeout * hz); 1933 sx_xlock(&sc->sc_lock); 1934 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1935 continue; 1936 } 1937 mtx_unlock(&sc->sc_queue_mtx); 1938 1939 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 1940 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1941 g_mirror_sync_request(bp); /* READ */ 1942 } else if (bp->bio_to != sc->sc_provider) { 1943 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 1944 g_mirror_regular_request(bp); 1945 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 1946 g_mirror_sync_request(bp); /* WRITE */ 1947 else { 1948 KASSERT(0, 1949 ("Invalid request cflags=0x%hx to=%s.", 1950 bp->bio_cflags, bp->bio_to->name)); 1951 } 1952 } else { 1953 g_mirror_register_request(bp); 1954 } 1955 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1956 } 1957 } 1958 1959 static void 1960 g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 1961 { 1962 1963 sx_assert(&sc->sc_lock, SX_LOCKED); 1964 1965 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 1966 return; 1967 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1968 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 1969 g_mirror_get_diskname(disk), sc->sc_name); 1970 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1971 } else if (sc->sc_idle && 1972 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1973 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 1974 g_mirror_get_diskname(disk), sc->sc_name); 1975 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1976 } 1977 } 1978 1979 static void 1980 g_mirror_sync_start(struct g_mirror_disk *disk) 1981 { 1982 struct g_mirror_softc *sc; 1983 struct g_consumer *cp; 1984 struct bio *bp; 1985 int error, i; 1986 1987 g_topology_assert_not(); 1988 sc = disk->d_softc; 1989 sx_assert(&sc->sc_lock, SX_LOCKED); 1990 1991 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1992 ("Disk %s is not marked for synchronization.", 1993 g_mirror_get_diskname(disk))); 1994 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1995 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1996 sc->sc_state)); 1997 1998 sx_xunlock(&sc->sc_lock); 1999 g_topology_lock(); 2000 cp = g_new_consumer(sc->sc_sync.ds_geom); 2001 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 2002 error = g_attach(cp, sc->sc_provider); 2003 KASSERT(error == 0, 2004 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 2005 error = g_access(cp, 1, 0, 0); 2006 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 2007 g_topology_unlock(); 2008 sx_xlock(&sc->sc_lock); 2009 2010 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 2011 g_mirror_get_diskname(disk)); 2012 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0) 2013 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2014 KASSERT(disk->d_sync.ds_consumer == NULL, 2015 ("Sync consumer already exists (device=%s, disk=%s).", 2016 sc->sc_name, g_mirror_get_diskname(disk))); 2017 2018 disk->d_sync.ds_consumer = cp; 2019 disk->d_sync.ds_consumer->private = disk; 2020 disk->d_sync.ds_consumer->index = 0; 2021 2022 /* 2023 * Allocate memory for synchronization bios and initialize them. 2024 */ 2025 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 2026 M_MIRROR, M_WAITOK); 2027 for (i = 0; i < g_mirror_syncreqs; i++) { 2028 bp = g_alloc_bio(); 2029 disk->d_sync.ds_bios[i] = bp; 2030 bp->bio_parent = NULL; 2031 bp->bio_cmd = BIO_READ; 2032 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 2033 bp->bio_cflags = 0; 2034 bp->bio_offset = disk->d_sync.ds_offset; 2035 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 2036 disk->d_sync.ds_offset += bp->bio_length; 2037 bp->bio_done = g_mirror_sync_done; 2038 bp->bio_from = disk->d_sync.ds_consumer; 2039 bp->bio_to = sc->sc_provider; 2040 bp->bio_caller1 = (void *)(uintptr_t)i; 2041 } 2042 2043 /* Increase the number of disks in SYNCHRONIZING state. */ 2044 sc->sc_sync.ds_ndisks++; 2045 /* Set the number of in-flight synchronization requests. */ 2046 disk->d_sync.ds_inflight = g_mirror_syncreqs; 2047 2048 /* 2049 * Fire off first synchronization requests. 2050 */ 2051 for (i = 0; i < g_mirror_syncreqs; i++) { 2052 bp = disk->d_sync.ds_bios[i]; 2053 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 2054 disk->d_sync.ds_consumer->index++; 2055 /* 2056 * Delay the request if it is colliding with a regular request. 2057 */ 2058 if (g_mirror_regular_collision(sc, bp)) 2059 g_mirror_sync_delay(sc, bp); 2060 else 2061 g_io_request(bp, disk->d_sync.ds_consumer); 2062 } 2063 } 2064 2065 /* 2066 * Stop synchronization process. 2067 * type: 0 - synchronization finished 2068 * 1 - synchronization stopped 2069 */ 2070 static void 2071 g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 2072 { 2073 struct g_mirror_softc *sc; 2074 struct g_consumer *cp; 2075 2076 g_topology_assert_not(); 2077 sc = disk->d_softc; 2078 sx_assert(&sc->sc_lock, SX_LOCKED); 2079 2080 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2081 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2082 g_mirror_disk_state2str(disk->d_state))); 2083 if (disk->d_sync.ds_consumer == NULL) 2084 return; 2085 2086 if (type == 0) { 2087 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2088 sc->sc_name, g_mirror_get_diskname(disk)); 2089 } else /* if (type == 1) */ { 2090 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2091 sc->sc_name, g_mirror_get_diskname(disk)); 2092 } 2093 g_mirror_regular_release(sc); 2094 free(disk->d_sync.ds_bios, M_MIRROR); 2095 disk->d_sync.ds_bios = NULL; 2096 cp = disk->d_sync.ds_consumer; 2097 disk->d_sync.ds_consumer = NULL; 2098 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2099 sc->sc_sync.ds_ndisks--; 2100 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2101 g_topology_lock(); 2102 g_mirror_kill_consumer(sc, cp); 2103 g_topology_unlock(); 2104 sx_xlock(&sc->sc_lock); 2105 } 2106 2107 static void 2108 g_mirror_launch_provider(struct g_mirror_softc *sc) 2109 { 2110 struct g_mirror_disk *disk; 2111 struct g_provider *pp, *dp; 2112 2113 sx_assert(&sc->sc_lock, SX_LOCKED); 2114 2115 g_topology_lock(); 2116 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 2117 pp->flags |= G_PF_DIRECT_RECEIVE; 2118 pp->mediasize = sc->sc_mediasize; 2119 pp->sectorsize = sc->sc_sectorsize; 2120 pp->stripesize = 0; 2121 pp->stripeoffset = 0; 2122 2123 /* Splitting of unmapped BIO's could work but isn't implemented now */ 2124 if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT) 2125 pp->flags |= G_PF_ACCEPT_UNMAPPED; 2126 2127 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2128 if (disk->d_consumer && disk->d_consumer->provider) { 2129 dp = disk->d_consumer->provider; 2130 if (dp->stripesize > pp->stripesize) { 2131 pp->stripesize = dp->stripesize; 2132 pp->stripeoffset = dp->stripeoffset; 2133 } 2134 /* A provider underneath us doesn't support unmapped */ 2135 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { 2136 G_MIRROR_DEBUG(0, "Cancelling unmapped " 2137 "because of %s.", dp->name); 2138 pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 2139 } 2140 } 2141 } 2142 pp->private = sc; 2143 sc->sc_refcnt++; 2144 sc->sc_provider = pp; 2145 g_error_provider(pp, 0); 2146 g_topology_unlock(); 2147 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2148 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 2149 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2150 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2151 g_mirror_sync_start(disk); 2152 } 2153 } 2154 2155 static void 2156 g_mirror_destroy_provider(struct g_mirror_softc *sc) 2157 { 2158 struct g_mirror_disk *disk; 2159 struct bio *bp; 2160 2161 g_topology_assert_not(); 2162 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2163 sc->sc_name)); 2164 2165 g_topology_lock(); 2166 g_error_provider(sc->sc_provider, ENXIO); 2167 mtx_lock(&sc->sc_queue_mtx); 2168 while ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) { 2169 /* 2170 * Abort any pending I/O that wasn't generated by us. 2171 * Synchronization requests and requests destined for individual 2172 * mirror components can be destroyed immediately. 2173 */ 2174 if (bp->bio_to == sc->sc_provider && 2175 bp->bio_from->geom != sc->sc_sync.ds_geom) { 2176 g_io_deliver(bp, ENXIO); 2177 } else { 2178 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2179 free(bp->bio_data, M_MIRROR); 2180 g_destroy_bio(bp); 2181 } 2182 } 2183 mtx_unlock(&sc->sc_queue_mtx); 2184 g_wither_provider(sc->sc_provider, ENXIO); 2185 sc->sc_provider = NULL; 2186 G_MIRROR_DEBUG(0, "Device %s: provider destroyed.", sc->sc_name); 2187 g_topology_unlock(); 2188 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2189 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2190 g_mirror_sync_stop(disk, 1); 2191 } 2192 } 2193 2194 static void 2195 g_mirror_go(void *arg) 2196 { 2197 struct g_mirror_softc *sc; 2198 2199 sc = arg; 2200 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2201 g_mirror_event_send(sc, 0, 2202 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2203 } 2204 2205 static u_int 2206 g_mirror_determine_state(struct g_mirror_disk *disk) 2207 { 2208 struct g_mirror_softc *sc; 2209 u_int state; 2210 2211 sc = disk->d_softc; 2212 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2213 if ((disk->d_flags & 2214 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2215 /* Disk does not need synchronization. */ 2216 state = G_MIRROR_DISK_STATE_ACTIVE; 2217 } else { 2218 if ((sc->sc_flags & 2219 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2220 (disk->d_flags & 2221 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2222 /* 2223 * We can start synchronization from 2224 * the stored offset. 2225 */ 2226 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2227 } else { 2228 state = G_MIRROR_DISK_STATE_STALE; 2229 } 2230 } 2231 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2232 /* 2233 * Reset all synchronization data for this disk, 2234 * because if it even was synchronized, it was 2235 * synchronized to disks with different syncid. 2236 */ 2237 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2238 disk->d_sync.ds_offset = 0; 2239 disk->d_sync.ds_offset_done = 0; 2240 disk->d_sync.ds_syncid = sc->sc_syncid; 2241 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2242 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2243 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2244 } else { 2245 state = G_MIRROR_DISK_STATE_STALE; 2246 } 2247 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2248 /* 2249 * Not good, NOT GOOD! 2250 * It means that mirror was started on stale disks 2251 * and more fresh disk just arrive. 2252 * If there were writes, mirror is broken, sorry. 2253 * I think the best choice here is don't touch 2254 * this disk and inform the user loudly. 2255 */ 2256 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2257 "disk (%s) arrives!! It will not be connected to the " 2258 "running device.", sc->sc_name, 2259 g_mirror_get_diskname(disk)); 2260 g_mirror_destroy_disk(disk); 2261 state = G_MIRROR_DISK_STATE_NONE; 2262 /* Return immediately, because disk was destroyed. */ 2263 return (state); 2264 } 2265 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2266 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2267 return (state); 2268 } 2269 2270 /* 2271 * Update device state. 2272 */ 2273 static void 2274 g_mirror_update_device(struct g_mirror_softc *sc, bool force) 2275 { 2276 struct g_mirror_disk *disk; 2277 u_int state; 2278 2279 sx_assert(&sc->sc_lock, SX_XLOCKED); 2280 2281 switch (sc->sc_state) { 2282 case G_MIRROR_DEVICE_STATE_STARTING: 2283 { 2284 struct g_mirror_disk *pdisk, *tdisk; 2285 u_int dirty, ndisks, genid, syncid; 2286 bool broken; 2287 2288 KASSERT(sc->sc_provider == NULL, 2289 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2290 /* 2291 * Are we ready? We are, if all disks are connected or 2292 * if we have any disks and 'force' is true. 2293 */ 2294 ndisks = g_mirror_ndisks(sc, -1); 2295 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2296 ; 2297 } else if (ndisks == 0) { 2298 /* 2299 * Disks went down in starting phase, so destroy 2300 * device. 2301 */ 2302 callout_drain(&sc->sc_callout); 2303 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2304 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2305 sc->sc_rootmount); 2306 root_mount_rel(sc->sc_rootmount); 2307 sc->sc_rootmount = NULL; 2308 return; 2309 } else { 2310 return; 2311 } 2312 2313 /* 2314 * Activate all disks with the biggest syncid. 2315 */ 2316 if (force) { 2317 /* 2318 * If 'force' is true, we have been called due to 2319 * timeout, so don't bother canceling timeout. 2320 */ 2321 ndisks = 0; 2322 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2323 if ((disk->d_flags & 2324 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2325 ndisks++; 2326 } 2327 } 2328 if (ndisks == 0) { 2329 /* No valid disks found, destroy device. */ 2330 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2331 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2332 __LINE__, sc->sc_rootmount); 2333 root_mount_rel(sc->sc_rootmount); 2334 sc->sc_rootmount = NULL; 2335 return; 2336 } 2337 } else { 2338 /* Cancel timeout. */ 2339 callout_drain(&sc->sc_callout); 2340 } 2341 2342 /* 2343 * Find the biggest genid. 2344 */ 2345 genid = 0; 2346 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2347 if (disk->d_genid > genid) 2348 genid = disk->d_genid; 2349 } 2350 sc->sc_genid = genid; 2351 /* 2352 * Remove all disks without the biggest genid. 2353 */ 2354 broken = false; 2355 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2356 if (disk->d_genid < genid) { 2357 G_MIRROR_DEBUG(0, 2358 "Component %s (device %s) broken, skipping.", 2359 g_mirror_get_diskname(disk), sc->sc_name); 2360 g_mirror_destroy_disk(disk); 2361 /* 2362 * Bump the syncid in case we discover a healthy 2363 * replacement disk after starting the mirror. 2364 */ 2365 broken = true; 2366 } 2367 } 2368 2369 /* 2370 * Find the biggest syncid. 2371 */ 2372 syncid = 0; 2373 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2374 if (disk->d_sync.ds_syncid > syncid) 2375 syncid = disk->d_sync.ds_syncid; 2376 } 2377 2378 /* 2379 * Here we need to look for dirty disks and if all disks 2380 * with the biggest syncid are dirty, we have to choose 2381 * one with the biggest priority and rebuild the rest. 2382 */ 2383 /* 2384 * Find the number of dirty disks with the biggest syncid. 2385 * Find the number of disks with the biggest syncid. 2386 * While here, find a disk with the biggest priority. 2387 */ 2388 dirty = ndisks = 0; 2389 pdisk = NULL; 2390 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2391 if (disk->d_sync.ds_syncid != syncid) 2392 continue; 2393 if ((disk->d_flags & 2394 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2395 continue; 2396 } 2397 ndisks++; 2398 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2399 dirty++; 2400 if (pdisk == NULL || 2401 pdisk->d_priority < disk->d_priority) { 2402 pdisk = disk; 2403 } 2404 } 2405 } 2406 if (dirty == 0) { 2407 /* No dirty disks at all, great. */ 2408 } else if (dirty == ndisks) { 2409 /* 2410 * Force synchronization for all dirty disks except one 2411 * with the biggest priority. 2412 */ 2413 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2414 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2415 "master disk for synchronization.", 2416 g_mirror_get_diskname(pdisk), sc->sc_name); 2417 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2418 if (disk->d_sync.ds_syncid != syncid) 2419 continue; 2420 if ((disk->d_flags & 2421 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2422 continue; 2423 } 2424 KASSERT((disk->d_flags & 2425 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2426 ("Disk %s isn't marked as dirty.", 2427 g_mirror_get_diskname(disk))); 2428 /* Skip the disk with the biggest priority. */ 2429 if (disk == pdisk) 2430 continue; 2431 disk->d_sync.ds_syncid = 0; 2432 } 2433 } else if (dirty < ndisks) { 2434 /* 2435 * Force synchronization for all dirty disks. 2436 * We have some non-dirty disks. 2437 */ 2438 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2439 if (disk->d_sync.ds_syncid != syncid) 2440 continue; 2441 if ((disk->d_flags & 2442 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2443 continue; 2444 } 2445 if ((disk->d_flags & 2446 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2447 continue; 2448 } 2449 disk->d_sync.ds_syncid = 0; 2450 } 2451 } 2452 2453 /* Reset hint. */ 2454 sc->sc_hint = NULL; 2455 sc->sc_syncid = syncid; 2456 if (force || broken) { 2457 /* Remember to bump syncid on first write. */ 2458 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2459 } 2460 state = G_MIRROR_DEVICE_STATE_RUNNING; 2461 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2462 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2463 g_mirror_device_state2str(state)); 2464 sc->sc_state = state; 2465 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2466 state = g_mirror_determine_state(disk); 2467 g_mirror_event_send(disk, state, 2468 G_MIRROR_EVENT_DONTWAIT); 2469 if (state == G_MIRROR_DISK_STATE_STALE) 2470 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2471 } 2472 break; 2473 } 2474 case G_MIRROR_DEVICE_STATE_RUNNING: 2475 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2476 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2477 /* 2478 * No active disks or no disks at all, 2479 * so destroy device. 2480 */ 2481 if (sc->sc_provider != NULL) 2482 g_mirror_destroy_provider(sc); 2483 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2484 break; 2485 } else if (g_mirror_ndisks(sc, 2486 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2487 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2488 /* 2489 * We have active disks, launch provider if it doesn't 2490 * exist. 2491 */ 2492 if (sc->sc_provider == NULL) 2493 g_mirror_launch_provider(sc); 2494 if (sc->sc_rootmount != NULL) { 2495 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2496 __LINE__, sc->sc_rootmount); 2497 root_mount_rel(sc->sc_rootmount); 2498 sc->sc_rootmount = NULL; 2499 } 2500 } 2501 /* 2502 * Genid should be bumped immediately, so do it here. 2503 */ 2504 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2505 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2506 g_mirror_bump_genid(sc); 2507 } 2508 break; 2509 default: 2510 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2511 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2512 break; 2513 } 2514 } 2515 2516 /* 2517 * Update disk state and device state if needed. 2518 */ 2519 #define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2520 "Disk %s state changed from %s to %s (device %s).", \ 2521 g_mirror_get_diskname(disk), \ 2522 g_mirror_disk_state2str(disk->d_state), \ 2523 g_mirror_disk_state2str(state), sc->sc_name) 2524 static int 2525 g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2526 { 2527 struct g_mirror_softc *sc; 2528 2529 sc = disk->d_softc; 2530 sx_assert(&sc->sc_lock, SX_XLOCKED); 2531 2532 again: 2533 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2534 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2535 g_mirror_disk_state2str(state)); 2536 switch (state) { 2537 case G_MIRROR_DISK_STATE_NEW: 2538 /* 2539 * Possible scenarios: 2540 * 1. New disk arrive. 2541 */ 2542 /* Previous state should be NONE. */ 2543 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2544 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2545 g_mirror_disk_state2str(disk->d_state))); 2546 DISK_STATE_CHANGED(); 2547 2548 disk->d_state = state; 2549 if (LIST_EMPTY(&sc->sc_disks)) 2550 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2551 else { 2552 struct g_mirror_disk *dp; 2553 2554 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2555 if (disk->d_priority >= dp->d_priority) { 2556 LIST_INSERT_BEFORE(dp, disk, d_next); 2557 dp = NULL; 2558 break; 2559 } 2560 if (LIST_NEXT(dp, d_next) == NULL) 2561 break; 2562 } 2563 if (dp != NULL) 2564 LIST_INSERT_AFTER(dp, disk, d_next); 2565 } 2566 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2567 sc->sc_name, g_mirror_get_diskname(disk)); 2568 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2569 break; 2570 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2571 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2572 g_mirror_device_state2str(sc->sc_state), 2573 g_mirror_get_diskname(disk), 2574 g_mirror_disk_state2str(disk->d_state))); 2575 state = g_mirror_determine_state(disk); 2576 if (state != G_MIRROR_DISK_STATE_NONE) 2577 goto again; 2578 break; 2579 case G_MIRROR_DISK_STATE_ACTIVE: 2580 /* 2581 * Possible scenarios: 2582 * 1. New disk does not need synchronization. 2583 * 2. Synchronization process finished successfully. 2584 */ 2585 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2586 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2587 g_mirror_device_state2str(sc->sc_state), 2588 g_mirror_get_diskname(disk), 2589 g_mirror_disk_state2str(disk->d_state))); 2590 /* Previous state should be NEW or SYNCHRONIZING. */ 2591 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2592 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2593 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2594 g_mirror_disk_state2str(disk->d_state))); 2595 DISK_STATE_CHANGED(); 2596 2597 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2598 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2599 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2600 g_mirror_sync_stop(disk, 0); 2601 } 2602 disk->d_state = state; 2603 disk->d_sync.ds_offset = 0; 2604 disk->d_sync.ds_offset_done = 0; 2605 g_mirror_update_idle(sc, disk); 2606 g_mirror_update_metadata(disk); 2607 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2608 sc->sc_name, g_mirror_get_diskname(disk)); 2609 break; 2610 case G_MIRROR_DISK_STATE_STALE: 2611 /* 2612 * Possible scenarios: 2613 * 1. Stale disk was connected. 2614 */ 2615 /* Previous state should be NEW. */ 2616 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2617 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2618 g_mirror_disk_state2str(disk->d_state))); 2619 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2620 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2621 g_mirror_device_state2str(sc->sc_state), 2622 g_mirror_get_diskname(disk), 2623 g_mirror_disk_state2str(disk->d_state))); 2624 /* 2625 * STALE state is only possible if device is marked 2626 * NOAUTOSYNC. 2627 */ 2628 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2629 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2630 g_mirror_device_state2str(sc->sc_state), 2631 g_mirror_get_diskname(disk), 2632 g_mirror_disk_state2str(disk->d_state))); 2633 DISK_STATE_CHANGED(); 2634 2635 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2636 disk->d_state = state; 2637 g_mirror_update_metadata(disk); 2638 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2639 sc->sc_name, g_mirror_get_diskname(disk)); 2640 break; 2641 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2642 /* 2643 * Possible scenarios: 2644 * 1. Disk which needs synchronization was connected. 2645 */ 2646 /* Previous state should be NEW. */ 2647 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2648 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2649 g_mirror_disk_state2str(disk->d_state))); 2650 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2651 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2652 g_mirror_device_state2str(sc->sc_state), 2653 g_mirror_get_diskname(disk), 2654 g_mirror_disk_state2str(disk->d_state))); 2655 DISK_STATE_CHANGED(); 2656 2657 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2658 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2659 disk->d_state = state; 2660 if (sc->sc_provider != NULL) { 2661 g_mirror_sync_start(disk); 2662 g_mirror_update_metadata(disk); 2663 } 2664 break; 2665 case G_MIRROR_DISK_STATE_DISCONNECTED: 2666 /* 2667 * Possible scenarios: 2668 * 1. Device wasn't running yet, but disk disappear. 2669 * 2. Disk was active and disapppear. 2670 * 3. Disk disappear during synchronization process. 2671 */ 2672 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2673 /* 2674 * Previous state should be ACTIVE, STALE or 2675 * SYNCHRONIZING. 2676 */ 2677 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2678 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2679 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2680 ("Wrong disk state (%s, %s).", 2681 g_mirror_get_diskname(disk), 2682 g_mirror_disk_state2str(disk->d_state))); 2683 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2684 /* Previous state should be NEW. */ 2685 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2686 ("Wrong disk state (%s, %s).", 2687 g_mirror_get_diskname(disk), 2688 g_mirror_disk_state2str(disk->d_state))); 2689 /* 2690 * Reset bumping syncid if disk disappeared in STARTING 2691 * state. 2692 */ 2693 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2694 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2695 #ifdef INVARIANTS 2696 } else { 2697 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2698 sc->sc_name, 2699 g_mirror_device_state2str(sc->sc_state), 2700 g_mirror_get_diskname(disk), 2701 g_mirror_disk_state2str(disk->d_state))); 2702 #endif 2703 } 2704 DISK_STATE_CHANGED(); 2705 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2706 sc->sc_name, g_mirror_get_diskname(disk)); 2707 2708 g_mirror_destroy_disk(disk); 2709 break; 2710 case G_MIRROR_DISK_STATE_DESTROY: 2711 { 2712 int error; 2713 2714 error = g_mirror_clear_metadata(disk); 2715 if (error != 0) { 2716 G_MIRROR_DEBUG(0, 2717 "Device %s: failed to clear metadata on %s: %d.", 2718 sc->sc_name, g_mirror_get_diskname(disk), error); 2719 break; 2720 } 2721 DISK_STATE_CHANGED(); 2722 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2723 sc->sc_name, g_mirror_get_diskname(disk)); 2724 2725 g_mirror_destroy_disk(disk); 2726 sc->sc_ndisks--; 2727 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2728 g_mirror_update_metadata(disk); 2729 } 2730 break; 2731 } 2732 default: 2733 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2734 break; 2735 } 2736 return (0); 2737 } 2738 #undef DISK_STATE_CHANGED 2739 2740 int 2741 g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2742 { 2743 struct g_provider *pp; 2744 u_char *buf; 2745 int error; 2746 2747 g_topology_assert(); 2748 2749 error = g_access(cp, 1, 0, 0); 2750 if (error != 0) 2751 return (error); 2752 pp = cp->provider; 2753 g_topology_unlock(); 2754 /* Metadata are stored on last sector. */ 2755 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2756 &error); 2757 g_topology_lock(); 2758 g_access(cp, -1, 0, 0); 2759 if (buf == NULL) { 2760 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2761 cp->provider->name, error); 2762 return (error); 2763 } 2764 2765 /* Decode metadata. */ 2766 error = mirror_metadata_decode(buf, md); 2767 g_free(buf); 2768 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2769 return (EINVAL); 2770 if (md->md_version > G_MIRROR_VERSION) { 2771 G_MIRROR_DEBUG(0, 2772 "Kernel module is too old to handle metadata from %s.", 2773 cp->provider->name); 2774 return (EINVAL); 2775 } 2776 if (error != 0) { 2777 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2778 cp->provider->name); 2779 return (error); 2780 } 2781 2782 return (0); 2783 } 2784 2785 static int 2786 g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2787 struct g_mirror_metadata *md) 2788 { 2789 2790 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2791 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2792 pp->name, md->md_did); 2793 return (EEXIST); 2794 } 2795 if (md->md_all != sc->sc_ndisks) { 2796 G_MIRROR_DEBUG(1, 2797 "Invalid '%s' field on disk %s (device %s), skipping.", 2798 "md_all", pp->name, sc->sc_name); 2799 return (EINVAL); 2800 } 2801 if (md->md_slice != sc->sc_slice) { 2802 G_MIRROR_DEBUG(1, 2803 "Invalid '%s' field on disk %s (device %s), skipping.", 2804 "md_slice", pp->name, sc->sc_name); 2805 return (EINVAL); 2806 } 2807 if (md->md_balance != sc->sc_balance) { 2808 G_MIRROR_DEBUG(1, 2809 "Invalid '%s' field on disk %s (device %s), skipping.", 2810 "md_balance", pp->name, sc->sc_name); 2811 return (EINVAL); 2812 } 2813 #if 0 2814 if (md->md_mediasize != sc->sc_mediasize) { 2815 G_MIRROR_DEBUG(1, 2816 "Invalid '%s' field on disk %s (device %s), skipping.", 2817 "md_mediasize", pp->name, sc->sc_name); 2818 return (EINVAL); 2819 } 2820 #endif 2821 if (sc->sc_mediasize > pp->mediasize) { 2822 G_MIRROR_DEBUG(1, 2823 "Invalid size of disk %s (device %s), skipping.", pp->name, 2824 sc->sc_name); 2825 return (EINVAL); 2826 } 2827 if (md->md_sectorsize != sc->sc_sectorsize) { 2828 G_MIRROR_DEBUG(1, 2829 "Invalid '%s' field on disk %s (device %s), skipping.", 2830 "md_sectorsize", pp->name, sc->sc_name); 2831 return (EINVAL); 2832 } 2833 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2834 G_MIRROR_DEBUG(1, 2835 "Invalid sector size of disk %s (device %s), skipping.", 2836 pp->name, sc->sc_name); 2837 return (EINVAL); 2838 } 2839 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2840 G_MIRROR_DEBUG(1, 2841 "Invalid device flags on disk %s (device %s), skipping.", 2842 pp->name, sc->sc_name); 2843 return (EINVAL); 2844 } 2845 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2846 G_MIRROR_DEBUG(1, 2847 "Invalid disk flags on disk %s (device %s), skipping.", 2848 pp->name, sc->sc_name); 2849 return (EINVAL); 2850 } 2851 return (0); 2852 } 2853 2854 int 2855 g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2856 struct g_mirror_metadata *md) 2857 { 2858 struct g_mirror_disk *disk; 2859 int error; 2860 2861 g_topology_assert_not(); 2862 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2863 2864 error = g_mirror_check_metadata(sc, pp, md); 2865 if (error != 0) 2866 return (error); 2867 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2868 md->md_genid < sc->sc_genid) { 2869 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2870 pp->name, sc->sc_name); 2871 return (EINVAL); 2872 } 2873 disk = g_mirror_init_disk(sc, pp, md, &error); 2874 if (disk == NULL) 2875 return (error); 2876 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2877 G_MIRROR_EVENT_WAIT); 2878 if (error != 0) 2879 return (error); 2880 if (md->md_version < G_MIRROR_VERSION) { 2881 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2882 pp->name, md->md_version, G_MIRROR_VERSION); 2883 g_mirror_update_metadata(disk); 2884 } 2885 return (0); 2886 } 2887 2888 static void 2889 g_mirror_destroy_delayed(void *arg, int flag) 2890 { 2891 struct g_mirror_softc *sc; 2892 int error; 2893 2894 if (flag == EV_CANCEL) { 2895 G_MIRROR_DEBUG(1, "Destroying canceled."); 2896 return; 2897 } 2898 sc = arg; 2899 g_topology_unlock(); 2900 sx_xlock(&sc->sc_lock); 2901 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2902 ("DESTROY flag set on %s.", sc->sc_name)); 2903 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0, 2904 ("DESTROYING flag not set on %s.", sc->sc_name)); 2905 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 2906 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 2907 if (error != 0) { 2908 G_MIRROR_DEBUG(0, "Cannot destroy %s (error=%d).", 2909 sc->sc_name, error); 2910 sx_xunlock(&sc->sc_lock); 2911 } 2912 g_topology_lock(); 2913 } 2914 2915 static int 2916 g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2917 { 2918 struct g_mirror_softc *sc; 2919 int error = 0; 2920 2921 g_topology_assert(); 2922 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2923 acw, ace); 2924 2925 sc = pp->private; 2926 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 2927 2928 g_topology_unlock(); 2929 sx_xlock(&sc->sc_lock); 2930 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 2931 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0 || 2932 LIST_EMPTY(&sc->sc_disks)) { 2933 if (acr > 0 || acw > 0 || ace > 0) 2934 error = ENXIO; 2935 goto end; 2936 } 2937 sc->sc_provider_open += acr + acw + ace; 2938 if (pp->acw + acw == 0) 2939 g_mirror_idle(sc, 0); 2940 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0 && 2941 sc->sc_provider_open == 0) 2942 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, sc, NULL); 2943 end: 2944 sx_xunlock(&sc->sc_lock); 2945 g_topology_lock(); 2946 return (error); 2947 } 2948 2949 struct g_geom * 2950 g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md, 2951 u_int type) 2952 { 2953 struct g_mirror_softc *sc; 2954 struct g_geom *gp; 2955 int error, timeout; 2956 2957 g_topology_assert(); 2958 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2959 md->md_mid); 2960 2961 /* One disk is minimum. */ 2962 if (md->md_all < 1) 2963 return (NULL); 2964 /* 2965 * Action geom. 2966 */ 2967 gp = g_new_geomf(mp, "%s", md->md_name); 2968 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2969 gp->start = g_mirror_start; 2970 gp->orphan = g_mirror_orphan; 2971 gp->access = g_mirror_access; 2972 gp->dumpconf = g_mirror_dumpconf; 2973 2974 sc->sc_type = type; 2975 sc->sc_id = md->md_mid; 2976 sc->sc_slice = md->md_slice; 2977 sc->sc_balance = md->md_balance; 2978 sc->sc_mediasize = md->md_mediasize; 2979 sc->sc_sectorsize = md->md_sectorsize; 2980 sc->sc_ndisks = md->md_all; 2981 sc->sc_flags = md->md_mflags; 2982 sc->sc_bump_id = 0; 2983 sc->sc_idle = 1; 2984 sc->sc_last_write = time_uptime; 2985 sc->sc_writes = 0; 2986 sc->sc_refcnt = 1; 2987 sx_init(&sc->sc_lock, "gmirror:lock"); 2988 bioq_init(&sc->sc_queue); 2989 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2990 bioq_init(&sc->sc_regular_delayed); 2991 bioq_init(&sc->sc_inflight); 2992 bioq_init(&sc->sc_sync_delayed); 2993 LIST_INIT(&sc->sc_disks); 2994 TAILQ_INIT(&sc->sc_events); 2995 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2996 callout_init(&sc->sc_callout, 1); 2997 mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF); 2998 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2999 gp->softc = sc; 3000 sc->sc_geom = gp; 3001 sc->sc_provider = NULL; 3002 sc->sc_provider_open = 0; 3003 /* 3004 * Synchronization geom. 3005 */ 3006 gp = g_new_geomf(mp, "%s.sync", md->md_name); 3007 gp->softc = sc; 3008 gp->orphan = g_mirror_orphan; 3009 sc->sc_sync.ds_geom = gp; 3010 sc->sc_sync.ds_ndisks = 0; 3011 error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 3012 "g_mirror %s", md->md_name); 3013 if (error != 0) { 3014 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 3015 sc->sc_name); 3016 g_destroy_geom(sc->sc_sync.ds_geom); 3017 g_destroy_geom(sc->sc_geom); 3018 g_mirror_free_device(sc); 3019 return (NULL); 3020 } 3021 3022 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 3023 sc->sc_name, sc->sc_ndisks, sc->sc_id); 3024 3025 sc->sc_rootmount = root_mount_hold("GMIRROR"); 3026 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 3027 /* 3028 * Run timeout. 3029 */ 3030 timeout = g_mirror_timeout * hz; 3031 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 3032 return (sc->sc_geom); 3033 } 3034 3035 int 3036 g_mirror_destroy(struct g_mirror_softc *sc, int how) 3037 { 3038 struct g_mirror_disk *disk; 3039 3040 g_topology_assert_not(); 3041 sx_assert(&sc->sc_lock, SX_XLOCKED); 3042 3043 if (sc->sc_provider_open != 0 || SCHEDULER_STOPPED()) { 3044 switch (how) { 3045 case G_MIRROR_DESTROY_SOFT: 3046 G_MIRROR_DEBUG(1, 3047 "Device %s is still open (%d).", sc->sc_name, 3048 sc->sc_provider_open); 3049 return (EBUSY); 3050 case G_MIRROR_DESTROY_DELAYED: 3051 G_MIRROR_DEBUG(1, 3052 "Device %s will be destroyed on last close.", 3053 sc->sc_name); 3054 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 3055 if (disk->d_state == 3056 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3057 g_mirror_sync_stop(disk, 1); 3058 } 3059 } 3060 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROYING; 3061 return (EBUSY); 3062 case G_MIRROR_DESTROY_HARD: 3063 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 3064 "can't be definitely removed.", sc->sc_name); 3065 } 3066 } 3067 3068 g_topology_lock(); 3069 if (sc->sc_geom->softc == NULL) { 3070 g_topology_unlock(); 3071 return (0); 3072 } 3073 sc->sc_geom->softc = NULL; 3074 sc->sc_sync.ds_geom->softc = NULL; 3075 g_topology_unlock(); 3076 3077 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 3078 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 3079 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3080 sx_xunlock(&sc->sc_lock); 3081 mtx_lock(&sc->sc_queue_mtx); 3082 wakeup(sc); 3083 mtx_unlock(&sc->sc_queue_mtx); 3084 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3085 while (sc->sc_worker != NULL) 3086 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 3087 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3088 sx_xlock(&sc->sc_lock); 3089 g_mirror_destroy_device(sc); 3090 return (0); 3091 } 3092 3093 static void 3094 g_mirror_taste_orphan(struct g_consumer *cp) 3095 { 3096 3097 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3098 cp->provider->name)); 3099 } 3100 3101 static struct g_geom * 3102 g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3103 { 3104 struct g_mirror_metadata md; 3105 struct g_mirror_softc *sc; 3106 struct g_consumer *cp; 3107 struct g_geom *gp; 3108 int error; 3109 3110 g_topology_assert(); 3111 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3112 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 3113 3114 gp = g_new_geomf(mp, "mirror:taste"); 3115 /* 3116 * This orphan function should be never called. 3117 */ 3118 gp->orphan = g_mirror_taste_orphan; 3119 cp = g_new_consumer(gp); 3120 g_attach(cp, pp); 3121 error = g_mirror_read_metadata(cp, &md); 3122 g_detach(cp); 3123 g_destroy_consumer(cp); 3124 g_destroy_geom(gp); 3125 if (error != 0) 3126 return (NULL); 3127 gp = NULL; 3128 3129 if (md.md_provider[0] != '\0' && 3130 !g_compare_names(md.md_provider, pp->name)) 3131 return (NULL); 3132 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3133 return (NULL); 3134 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 3135 G_MIRROR_DEBUG(0, 3136 "Device %s: provider %s marked as inactive, skipping.", 3137 md.md_name, pp->name); 3138 return (NULL); 3139 } 3140 if (g_mirror_debug >= 2) 3141 mirror_metadata_dump(&md); 3142 3143 /* 3144 * Let's check if device already exists. 3145 */ 3146 sc = NULL; 3147 LIST_FOREACH(gp, &mp->geom, geom) { 3148 sc = gp->softc; 3149 if (sc == NULL) 3150 continue; 3151 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 3152 continue; 3153 if (sc->sc_sync.ds_geom == gp) 3154 continue; 3155 if (strcmp(md.md_name, sc->sc_name) != 0) 3156 continue; 3157 if (md.md_mid != sc->sc_id) { 3158 G_MIRROR_DEBUG(0, "Device %s already configured.", 3159 sc->sc_name); 3160 return (NULL); 3161 } 3162 break; 3163 } 3164 if (gp == NULL) { 3165 gp = g_mirror_create(mp, &md, G_MIRROR_TYPE_AUTOMATIC); 3166 if (gp == NULL) { 3167 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3168 md.md_name); 3169 return (NULL); 3170 } 3171 sc = gp->softc; 3172 } 3173 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3174 g_topology_unlock(); 3175 sx_xlock(&sc->sc_lock); 3176 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; 3177 error = g_mirror_add_disk(sc, pp, &md); 3178 if (error != 0) { 3179 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3180 pp->name, gp->name, error); 3181 if (LIST_EMPTY(&sc->sc_disks)) { 3182 g_cancel_event(sc); 3183 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3184 g_topology_lock(); 3185 return (NULL); 3186 } 3187 gp = NULL; 3188 } 3189 sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; 3190 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3191 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3192 g_topology_lock(); 3193 return (NULL); 3194 } 3195 sx_xunlock(&sc->sc_lock); 3196 g_topology_lock(); 3197 return (gp); 3198 } 3199 3200 static void 3201 g_mirror_resize(struct g_consumer *cp) 3202 { 3203 struct g_mirror_disk *disk; 3204 3205 g_topology_assert(); 3206 g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name); 3207 3208 disk = cp->private; 3209 if (disk == NULL) 3210 return; 3211 g_topology_unlock(); 3212 g_mirror_update_metadata(disk); 3213 g_topology_lock(); 3214 } 3215 3216 static int 3217 g_mirror_destroy_geom(struct gctl_req *req __unused, 3218 struct g_class *mp __unused, struct g_geom *gp) 3219 { 3220 struct g_mirror_softc *sc; 3221 int error; 3222 3223 g_topology_unlock(); 3224 sc = gp->softc; 3225 sx_xlock(&sc->sc_lock); 3226 g_cancel_event(sc); 3227 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3228 if (error != 0) 3229 sx_xunlock(&sc->sc_lock); 3230 g_topology_lock(); 3231 return (error); 3232 } 3233 3234 static void 3235 g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3236 struct g_consumer *cp, struct g_provider *pp) 3237 { 3238 struct g_mirror_softc *sc; 3239 3240 g_topology_assert(); 3241 3242 sc = gp->softc; 3243 if (sc == NULL) 3244 return; 3245 /* Skip synchronization geom. */ 3246 if (gp == sc->sc_sync.ds_geom) 3247 return; 3248 if (pp != NULL) { 3249 /* Nothing here. */ 3250 } else if (cp != NULL) { 3251 struct g_mirror_disk *disk; 3252 3253 disk = cp->private; 3254 if (disk == NULL) 3255 return; 3256 g_topology_unlock(); 3257 sx_xlock(&sc->sc_lock); 3258 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3259 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3260 sbuf_printf(sb, "%s<Synchronized>", indent); 3261 if (disk->d_sync.ds_offset == 0) 3262 sbuf_printf(sb, "0%%"); 3263 else { 3264 sbuf_printf(sb, "%u%%", 3265 (u_int)((disk->d_sync.ds_offset * 100) / 3266 sc->sc_provider->mediasize)); 3267 } 3268 sbuf_printf(sb, "</Synchronized>\n"); 3269 if (disk->d_sync.ds_offset > 0) { 3270 sbuf_printf(sb, "%s<BytesSynced>%jd" 3271 "</BytesSynced>\n", indent, 3272 (intmax_t)disk->d_sync.ds_offset); 3273 } 3274 } 3275 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3276 disk->d_sync.ds_syncid); 3277 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3278 disk->d_genid); 3279 sbuf_printf(sb, "%s<Flags>", indent); 3280 if (disk->d_flags == 0) 3281 sbuf_printf(sb, "NONE"); 3282 else { 3283 int first = 1; 3284 3285 #define ADD_FLAG(flag, name) do { \ 3286 if ((disk->d_flags & (flag)) != 0) { \ 3287 if (!first) \ 3288 sbuf_printf(sb, ", "); \ 3289 else \ 3290 first = 0; \ 3291 sbuf_printf(sb, name); \ 3292 } \ 3293 } while (0) 3294 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3295 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3296 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3297 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3298 "SYNCHRONIZING"); 3299 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3300 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3301 #undef ADD_FLAG 3302 } 3303 sbuf_printf(sb, "</Flags>\n"); 3304 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3305 disk->d_priority); 3306 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3307 g_mirror_disk_state2str(disk->d_state)); 3308 sx_xunlock(&sc->sc_lock); 3309 g_topology_lock(); 3310 } else { 3311 g_topology_unlock(); 3312 sx_xlock(&sc->sc_lock); 3313 sbuf_printf(sb, "%s<Type>", indent); 3314 switch (sc->sc_type) { 3315 case G_MIRROR_TYPE_AUTOMATIC: 3316 sbuf_printf(sb, "AUTOMATIC"); 3317 break; 3318 case G_MIRROR_TYPE_MANUAL: 3319 sbuf_printf(sb, "MANUAL"); 3320 break; 3321 default: 3322 sbuf_printf(sb, "UNKNOWN"); 3323 break; 3324 } 3325 sbuf_printf(sb, "</Type>\n"); 3326 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3327 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3328 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3329 sbuf_printf(sb, "%s<Flags>", indent); 3330 if (sc->sc_flags == 0) 3331 sbuf_printf(sb, "NONE"); 3332 else { 3333 int first = 1; 3334 3335 #define ADD_FLAG(flag, name) do { \ 3336 if ((sc->sc_flags & (flag)) != 0) { \ 3337 if (!first) \ 3338 sbuf_printf(sb, ", "); \ 3339 else \ 3340 first = 0; \ 3341 sbuf_printf(sb, name); \ 3342 } \ 3343 } while (0) 3344 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3345 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3346 #undef ADD_FLAG 3347 } 3348 sbuf_printf(sb, "</Flags>\n"); 3349 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3350 (u_int)sc->sc_slice); 3351 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3352 balance_name(sc->sc_balance)); 3353 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3354 sc->sc_ndisks); 3355 sbuf_printf(sb, "%s<State>", indent); 3356 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3357 sbuf_printf(sb, "%s", "STARTING"); 3358 else if (sc->sc_ndisks == 3359 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3360 sbuf_printf(sb, "%s", "COMPLETE"); 3361 else 3362 sbuf_printf(sb, "%s", "DEGRADED"); 3363 sbuf_printf(sb, "</State>\n"); 3364 sx_xunlock(&sc->sc_lock); 3365 g_topology_lock(); 3366 } 3367 } 3368 3369 static void 3370 g_mirror_shutdown_post_sync(void *arg, int howto) 3371 { 3372 struct g_class *mp; 3373 struct g_geom *gp, *gp2; 3374 struct g_mirror_softc *sc; 3375 int error; 3376 3377 mp = arg; 3378 g_topology_lock(); 3379 g_mirror_shutdown = 1; 3380 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3381 if ((sc = gp->softc) == NULL) 3382 continue; 3383 /* Skip synchronization geom. */ 3384 if (gp == sc->sc_sync.ds_geom) 3385 continue; 3386 g_topology_unlock(); 3387 sx_xlock(&sc->sc_lock); 3388 g_mirror_idle(sc, -1); 3389 g_cancel_event(sc); 3390 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3391 if (error != 0) 3392 sx_xunlock(&sc->sc_lock); 3393 g_topology_lock(); 3394 } 3395 g_topology_unlock(); 3396 } 3397 3398 static void 3399 g_mirror_init(struct g_class *mp) 3400 { 3401 3402 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3403 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3404 if (g_mirror_post_sync == NULL) 3405 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3406 } 3407 3408 static void 3409 g_mirror_fini(struct g_class *mp) 3410 { 3411 3412 if (g_mirror_post_sync != NULL) 3413 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); 3414 } 3415 3416 DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3417