1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2004-2006 Pawel Jakub Dawidek <pjd@FreeBSD.org> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include <sys/param.h> 33 #include <sys/systm.h> 34 #include <sys/bio.h> 35 #include <sys/eventhandler.h> 36 #include <sys/fail.h> 37 #include <sys/kernel.h> 38 #include <sys/kthread.h> 39 #include <sys/limits.h> 40 #include <sys/lock.h> 41 #include <sys/malloc.h> 42 #include <sys/mutex.h> 43 #include <sys/proc.h> 44 #include <sys/sbuf.h> 45 #include <sys/sched.h> 46 #include <sys/sx.h> 47 #include <sys/sysctl.h> 48 49 #include <geom/geom.h> 50 #include <geom/mirror/g_mirror.h> 51 52 FEATURE(geom_mirror, "GEOM mirroring support"); 53 54 static MALLOC_DEFINE(M_MIRROR, "mirror_data", "GEOM_MIRROR Data"); 55 56 SYSCTL_DECL(_kern_geom); 57 static SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, 58 "GEOM_MIRROR stuff"); 59 int g_mirror_debug = 0; 60 SYSCTL_INT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RWTUN, &g_mirror_debug, 0, 61 "Debug level"); 62 static u_int g_mirror_timeout = 4; 63 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RWTUN, &g_mirror_timeout, 64 0, "Time to wait on all mirror components"); 65 static u_int g_mirror_idletime = 5; 66 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RWTUN, 67 &g_mirror_idletime, 0, "Mark components as clean when idling"); 68 static u_int g_mirror_disconnect_on_failure = 1; 69 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN, 70 &g_mirror_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 71 static u_int g_mirror_syncreqs = 2; 72 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_requests, CTLFLAG_RDTUN, 73 &g_mirror_syncreqs, 0, "Parallel synchronization I/O requests."); 74 static u_int g_mirror_sync_period = 5; 75 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_update_period, CTLFLAG_RWTUN, 76 &g_mirror_sync_period, 0, 77 "Metadata update period during synchronization, in seconds"); 78 79 #define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 80 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 81 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 82 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 83 } while (0) 84 85 static eventhandler_tag g_mirror_post_sync = NULL; 86 static int g_mirror_shutdown = 0; 87 88 static g_ctl_destroy_geom_t g_mirror_destroy_geom; 89 static g_taste_t g_mirror_taste; 90 static g_init_t g_mirror_init; 91 static g_fini_t g_mirror_fini; 92 static g_provgone_t g_mirror_providergone; 93 static g_resize_t g_mirror_resize; 94 95 struct g_class g_mirror_class = { 96 .name = G_MIRROR_CLASS_NAME, 97 .version = G_VERSION, 98 .ctlreq = g_mirror_config, 99 .taste = g_mirror_taste, 100 .destroy_geom = g_mirror_destroy_geom, 101 .init = g_mirror_init, 102 .fini = g_mirror_fini, 103 .providergone = g_mirror_providergone, 104 .resize = g_mirror_resize 105 }; 106 107 108 static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 109 static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 110 static void g_mirror_update_device(struct g_mirror_softc *sc, bool force); 111 static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 112 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 113 static void g_mirror_sync_reinit(const struct g_mirror_disk *disk, 114 struct bio *bp, off_t offset); 115 static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 116 static void g_mirror_register_request(struct g_mirror_softc *sc, 117 struct bio *bp); 118 static void g_mirror_sync_release(struct g_mirror_softc *sc); 119 120 121 static const char * 122 g_mirror_disk_state2str(int state) 123 { 124 125 switch (state) { 126 case G_MIRROR_DISK_STATE_NONE: 127 return ("NONE"); 128 case G_MIRROR_DISK_STATE_NEW: 129 return ("NEW"); 130 case G_MIRROR_DISK_STATE_ACTIVE: 131 return ("ACTIVE"); 132 case G_MIRROR_DISK_STATE_STALE: 133 return ("STALE"); 134 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 135 return ("SYNCHRONIZING"); 136 case G_MIRROR_DISK_STATE_DISCONNECTED: 137 return ("DISCONNECTED"); 138 case G_MIRROR_DISK_STATE_DESTROY: 139 return ("DESTROY"); 140 default: 141 return ("INVALID"); 142 } 143 } 144 145 static const char * 146 g_mirror_device_state2str(int state) 147 { 148 149 switch (state) { 150 case G_MIRROR_DEVICE_STATE_STARTING: 151 return ("STARTING"); 152 case G_MIRROR_DEVICE_STATE_RUNNING: 153 return ("RUNNING"); 154 default: 155 return ("INVALID"); 156 } 157 } 158 159 static const char * 160 g_mirror_get_diskname(struct g_mirror_disk *disk) 161 { 162 163 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 164 return ("[unknown]"); 165 return (disk->d_name); 166 } 167 168 /* 169 * --- Events handling functions --- 170 * Events in geom_mirror are used to maintain disks and device status 171 * from one thread to simplify locking. 172 */ 173 static void 174 g_mirror_event_free(struct g_mirror_event *ep) 175 { 176 177 free(ep, M_MIRROR); 178 } 179 180 int 181 g_mirror_event_send(void *arg, int state, int flags) 182 { 183 struct g_mirror_softc *sc; 184 struct g_mirror_disk *disk; 185 struct g_mirror_event *ep; 186 int error; 187 188 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 189 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 190 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 191 disk = NULL; 192 sc = arg; 193 } else { 194 disk = arg; 195 sc = disk->d_softc; 196 } 197 ep->e_disk = disk; 198 ep->e_state = state; 199 ep->e_flags = flags; 200 ep->e_error = 0; 201 mtx_lock(&sc->sc_events_mtx); 202 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 203 mtx_unlock(&sc->sc_events_mtx); 204 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 205 mtx_lock(&sc->sc_queue_mtx); 206 wakeup(sc); 207 mtx_unlock(&sc->sc_queue_mtx); 208 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 209 return (0); 210 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 211 sx_xunlock(&sc->sc_lock); 212 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 213 mtx_lock(&sc->sc_events_mtx); 214 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 215 hz * 5); 216 } 217 error = ep->e_error; 218 g_mirror_event_free(ep); 219 sx_xlock(&sc->sc_lock); 220 return (error); 221 } 222 223 static struct g_mirror_event * 224 g_mirror_event_first(struct g_mirror_softc *sc) 225 { 226 struct g_mirror_event *ep; 227 228 mtx_lock(&sc->sc_events_mtx); 229 ep = TAILQ_FIRST(&sc->sc_events); 230 mtx_unlock(&sc->sc_events_mtx); 231 return (ep); 232 } 233 234 static void 235 g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 236 { 237 238 mtx_lock(&sc->sc_events_mtx); 239 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 240 mtx_unlock(&sc->sc_events_mtx); 241 } 242 243 static void 244 g_mirror_event_cancel(struct g_mirror_disk *disk) 245 { 246 struct g_mirror_softc *sc; 247 struct g_mirror_event *ep, *tmpep; 248 249 sc = disk->d_softc; 250 sx_assert(&sc->sc_lock, SX_XLOCKED); 251 252 mtx_lock(&sc->sc_events_mtx); 253 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 254 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 255 continue; 256 if (ep->e_disk != disk) 257 continue; 258 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 259 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 260 g_mirror_event_free(ep); 261 else { 262 ep->e_error = ECANCELED; 263 wakeup(ep); 264 } 265 } 266 mtx_unlock(&sc->sc_events_mtx); 267 } 268 269 /* 270 * Return the number of disks in given state. 271 * If state is equal to -1, count all connected disks. 272 */ 273 u_int 274 g_mirror_ndisks(struct g_mirror_softc *sc, int state) 275 { 276 struct g_mirror_disk *disk; 277 u_int n = 0; 278 279 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 280 if (state == -1 || disk->d_state == state) 281 n++; 282 } 283 return (n); 284 } 285 286 /* 287 * Find a disk in mirror by its disk ID. 288 */ 289 static struct g_mirror_disk * 290 g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 291 { 292 struct g_mirror_disk *disk; 293 294 sx_assert(&sc->sc_lock, SX_XLOCKED); 295 296 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 297 if (disk->d_id == id) 298 return (disk); 299 } 300 return (NULL); 301 } 302 303 static u_int 304 g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 305 { 306 struct bio *bp; 307 u_int nreqs = 0; 308 309 mtx_lock(&sc->sc_queue_mtx); 310 TAILQ_FOREACH(bp, &sc->sc_queue, bio_queue) { 311 if (bp->bio_from == cp) 312 nreqs++; 313 } 314 mtx_unlock(&sc->sc_queue_mtx); 315 return (nreqs); 316 } 317 318 static int 319 g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 320 { 321 322 if (cp->index > 0) { 323 G_MIRROR_DEBUG(2, 324 "I/O requests for %s exist, can't destroy it now.", 325 cp->provider->name); 326 return (1); 327 } 328 if (g_mirror_nrequests(sc, cp) > 0) { 329 G_MIRROR_DEBUG(2, 330 "I/O requests for %s in queue, can't destroy it now.", 331 cp->provider->name); 332 return (1); 333 } 334 return (0); 335 } 336 337 static void 338 g_mirror_destroy_consumer(void *arg, int flags __unused) 339 { 340 struct g_consumer *cp; 341 342 g_topology_assert(); 343 344 cp = arg; 345 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 346 g_detach(cp); 347 g_destroy_consumer(cp); 348 } 349 350 static void 351 g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 352 { 353 struct g_provider *pp; 354 int retaste_wait; 355 356 g_topology_assert(); 357 358 cp->private = NULL; 359 if (g_mirror_is_busy(sc, cp)) 360 return; 361 pp = cp->provider; 362 retaste_wait = 0; 363 if (cp->acw == 1) { 364 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 365 retaste_wait = 1; 366 } 367 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 368 -cp->acw, -cp->ace, 0); 369 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 370 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 371 if (retaste_wait) { 372 /* 373 * After retaste event was send (inside g_access()), we can send 374 * event to detach and destroy consumer. 375 * A class, which has consumer to the given provider connected 376 * will not receive retaste event for the provider. 377 * This is the way how I ignore retaste events when I close 378 * consumers opened for write: I detach and destroy consumer 379 * after retaste event is sent. 380 */ 381 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 382 return; 383 } 384 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 385 g_detach(cp); 386 g_destroy_consumer(cp); 387 } 388 389 static int 390 g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 391 { 392 struct g_consumer *cp; 393 int error; 394 395 g_topology_assert_not(); 396 KASSERT(disk->d_consumer == NULL, 397 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 398 399 g_topology_lock(); 400 cp = g_new_consumer(disk->d_softc->sc_geom); 401 cp->flags |= G_CF_DIRECT_RECEIVE; 402 error = g_attach(cp, pp); 403 if (error != 0) { 404 g_destroy_consumer(cp); 405 g_topology_unlock(); 406 return (error); 407 } 408 error = g_access(cp, 1, 1, 1); 409 if (error != 0) { 410 g_detach(cp); 411 g_destroy_consumer(cp); 412 g_topology_unlock(); 413 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 414 pp->name, error); 415 return (error); 416 } 417 g_topology_unlock(); 418 disk->d_consumer = cp; 419 disk->d_consumer->private = disk; 420 disk->d_consumer->index = 0; 421 422 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 423 return (0); 424 } 425 426 static void 427 g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 428 { 429 430 g_topology_assert(); 431 432 if (cp == NULL) 433 return; 434 if (cp->provider != NULL) 435 g_mirror_kill_consumer(sc, cp); 436 else 437 g_destroy_consumer(cp); 438 } 439 440 /* 441 * Initialize disk. This means allocate memory, create consumer, attach it 442 * to the provider and open access (r1w1e1) to it. 443 */ 444 static struct g_mirror_disk * 445 g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 446 struct g_mirror_metadata *md, int *errorp) 447 { 448 struct g_mirror_disk *disk; 449 int i, error; 450 451 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 452 if (disk == NULL) { 453 error = ENOMEM; 454 goto fail; 455 } 456 disk->d_softc = sc; 457 error = g_mirror_connect_disk(disk, pp); 458 if (error != 0) 459 goto fail; 460 disk->d_id = md->md_did; 461 disk->d_state = G_MIRROR_DISK_STATE_NONE; 462 disk->d_priority = md->md_priority; 463 disk->d_flags = md->md_dflags; 464 error = g_getattr("GEOM::candelete", disk->d_consumer, &i); 465 if (error == 0 && i != 0) 466 disk->d_flags |= G_MIRROR_DISK_FLAG_CANDELETE; 467 if (md->md_provider[0] != '\0') 468 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 469 disk->d_sync.ds_consumer = NULL; 470 disk->d_sync.ds_offset = md->md_sync_offset; 471 disk->d_sync.ds_offset_done = md->md_sync_offset; 472 disk->d_sync.ds_update_ts = time_uptime; 473 disk->d_genid = md->md_genid; 474 disk->d_sync.ds_syncid = md->md_syncid; 475 if (errorp != NULL) 476 *errorp = 0; 477 return (disk); 478 fail: 479 if (errorp != NULL) 480 *errorp = error; 481 if (disk != NULL) 482 free(disk, M_MIRROR); 483 return (NULL); 484 } 485 486 static void 487 g_mirror_destroy_disk(struct g_mirror_disk *disk) 488 { 489 struct g_mirror_softc *sc; 490 491 g_topology_assert_not(); 492 sc = disk->d_softc; 493 sx_assert(&sc->sc_lock, SX_XLOCKED); 494 495 g_topology_lock(); 496 LIST_REMOVE(disk, d_next); 497 g_topology_unlock(); 498 g_mirror_event_cancel(disk); 499 if (sc->sc_hint == disk) 500 sc->sc_hint = NULL; 501 switch (disk->d_state) { 502 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 503 g_mirror_sync_stop(disk, 1); 504 /* FALLTHROUGH */ 505 case G_MIRROR_DISK_STATE_NEW: 506 case G_MIRROR_DISK_STATE_STALE: 507 case G_MIRROR_DISK_STATE_ACTIVE: 508 g_topology_lock(); 509 g_mirror_disconnect_consumer(sc, disk->d_consumer); 510 g_topology_unlock(); 511 free(disk, M_MIRROR); 512 break; 513 default: 514 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 515 g_mirror_get_diskname(disk), 516 g_mirror_disk_state2str(disk->d_state))); 517 } 518 } 519 520 static void 521 g_mirror_free_device(struct g_mirror_softc *sc) 522 { 523 524 g_topology_assert(); 525 526 mtx_destroy(&sc->sc_queue_mtx); 527 mtx_destroy(&sc->sc_events_mtx); 528 mtx_destroy(&sc->sc_done_mtx); 529 sx_destroy(&sc->sc_lock); 530 free(sc, M_MIRROR); 531 } 532 533 static void 534 g_mirror_providergone(struct g_provider *pp) 535 { 536 struct g_mirror_softc *sc = pp->private; 537 538 if ((--sc->sc_refcnt) == 0) 539 g_mirror_free_device(sc); 540 } 541 542 static void 543 g_mirror_destroy_device(struct g_mirror_softc *sc) 544 { 545 struct g_mirror_disk *disk; 546 struct g_mirror_event *ep; 547 struct g_geom *gp; 548 struct g_consumer *cp, *tmpcp; 549 550 g_topology_assert_not(); 551 sx_assert(&sc->sc_lock, SX_XLOCKED); 552 553 gp = sc->sc_geom; 554 if (sc->sc_provider != NULL) 555 g_mirror_destroy_provider(sc); 556 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 557 disk = LIST_FIRST(&sc->sc_disks)) { 558 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 559 g_mirror_update_metadata(disk); 560 g_mirror_destroy_disk(disk); 561 } 562 while ((ep = g_mirror_event_first(sc)) != NULL) { 563 g_mirror_event_remove(sc, ep); 564 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 565 g_mirror_event_free(ep); 566 else { 567 ep->e_error = ECANCELED; 568 ep->e_flags |= G_MIRROR_EVENT_DONE; 569 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 570 mtx_lock(&sc->sc_events_mtx); 571 wakeup(ep); 572 mtx_unlock(&sc->sc_events_mtx); 573 } 574 } 575 callout_drain(&sc->sc_callout); 576 577 g_topology_lock(); 578 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 579 g_mirror_disconnect_consumer(sc, cp); 580 } 581 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 582 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 583 g_wither_geom(gp, ENXIO); 584 sx_xunlock(&sc->sc_lock); 585 if ((--sc->sc_refcnt) == 0) 586 g_mirror_free_device(sc); 587 g_topology_unlock(); 588 } 589 590 static void 591 g_mirror_orphan(struct g_consumer *cp) 592 { 593 struct g_mirror_disk *disk; 594 595 g_topology_assert(); 596 597 disk = cp->private; 598 if (disk == NULL) 599 return; 600 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 601 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 602 G_MIRROR_EVENT_DONTWAIT); 603 } 604 605 /* 606 * Function should return the next active disk on the list. 607 * It is possible that it will be the same disk as given. 608 * If there are no active disks on list, NULL is returned. 609 */ 610 static __inline struct g_mirror_disk * 611 g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 612 { 613 struct g_mirror_disk *dp; 614 615 for (dp = LIST_NEXT(disk, d_next); dp != disk; 616 dp = LIST_NEXT(dp, d_next)) { 617 if (dp == NULL) 618 dp = LIST_FIRST(&sc->sc_disks); 619 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 620 break; 621 } 622 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 623 return (NULL); 624 return (dp); 625 } 626 627 static struct g_mirror_disk * 628 g_mirror_get_disk(struct g_mirror_softc *sc) 629 { 630 struct g_mirror_disk *disk; 631 632 if (sc->sc_hint == NULL) { 633 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 634 if (sc->sc_hint == NULL) 635 return (NULL); 636 } 637 disk = sc->sc_hint; 638 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 639 disk = g_mirror_find_next(sc, disk); 640 if (disk == NULL) 641 return (NULL); 642 } 643 sc->sc_hint = g_mirror_find_next(sc, disk); 644 return (disk); 645 } 646 647 static int 648 g_mirror_write_metadata(struct g_mirror_disk *disk, 649 struct g_mirror_metadata *md) 650 { 651 struct g_mirror_softc *sc; 652 struct g_consumer *cp; 653 off_t offset, length; 654 u_char *sector; 655 int error = 0; 656 657 g_topology_assert_not(); 658 sc = disk->d_softc; 659 sx_assert(&sc->sc_lock, SX_LOCKED); 660 661 cp = disk->d_consumer; 662 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 663 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 664 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 665 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 666 cp->acw, cp->ace)); 667 length = cp->provider->sectorsize; 668 offset = cp->provider->mediasize - length; 669 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 670 if (md != NULL && 671 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) { 672 /* 673 * Handle the case, when the size of parent provider reduced. 674 */ 675 if (offset < md->md_mediasize) 676 error = ENOSPC; 677 else 678 mirror_metadata_encode(md, sector); 679 } 680 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_metadata_write, error); 681 if (error == 0) 682 error = g_write_data(cp, offset, sector, length); 683 free(sector, M_MIRROR); 684 if (error != 0) { 685 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 686 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 687 G_MIRROR_DEBUG(0, "Cannot write metadata on %s " 688 "(device=%s, error=%d).", 689 g_mirror_get_diskname(disk), sc->sc_name, error); 690 } else { 691 G_MIRROR_DEBUG(1, "Cannot write metadata on %s " 692 "(device=%s, error=%d).", 693 g_mirror_get_diskname(disk), sc->sc_name, error); 694 } 695 if (g_mirror_disconnect_on_failure && 696 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 697 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 698 g_mirror_event_send(disk, 699 G_MIRROR_DISK_STATE_DISCONNECTED, 700 G_MIRROR_EVENT_DONTWAIT); 701 } 702 } 703 return (error); 704 } 705 706 static int 707 g_mirror_clear_metadata(struct g_mirror_disk *disk) 708 { 709 int error; 710 711 g_topology_assert_not(); 712 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 713 714 if (disk->d_softc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 715 return (0); 716 error = g_mirror_write_metadata(disk, NULL); 717 if (error == 0) { 718 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 719 g_mirror_get_diskname(disk)); 720 } else { 721 G_MIRROR_DEBUG(0, 722 "Cannot clear metadata on disk %s (error=%d).", 723 g_mirror_get_diskname(disk), error); 724 } 725 return (error); 726 } 727 728 void 729 g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 730 struct g_mirror_metadata *md) 731 { 732 733 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 734 md->md_version = G_MIRROR_VERSION; 735 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 736 md->md_mid = sc->sc_id; 737 md->md_all = sc->sc_ndisks; 738 md->md_slice = sc->sc_slice; 739 md->md_balance = sc->sc_balance; 740 md->md_genid = sc->sc_genid; 741 md->md_mediasize = sc->sc_mediasize; 742 md->md_sectorsize = sc->sc_sectorsize; 743 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 744 bzero(md->md_provider, sizeof(md->md_provider)); 745 if (disk == NULL) { 746 md->md_did = arc4random(); 747 md->md_priority = 0; 748 md->md_syncid = 0; 749 md->md_dflags = 0; 750 md->md_sync_offset = 0; 751 md->md_provsize = 0; 752 } else { 753 md->md_did = disk->d_id; 754 md->md_priority = disk->d_priority; 755 md->md_syncid = disk->d_sync.ds_syncid; 756 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 757 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 758 md->md_sync_offset = disk->d_sync.ds_offset_done; 759 else 760 md->md_sync_offset = 0; 761 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 762 strlcpy(md->md_provider, 763 disk->d_consumer->provider->name, 764 sizeof(md->md_provider)); 765 } 766 md->md_provsize = disk->d_consumer->provider->mediasize; 767 } 768 } 769 770 void 771 g_mirror_update_metadata(struct g_mirror_disk *disk) 772 { 773 struct g_mirror_softc *sc; 774 struct g_mirror_metadata md; 775 int error; 776 777 g_topology_assert_not(); 778 sc = disk->d_softc; 779 sx_assert(&sc->sc_lock, SX_LOCKED); 780 781 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 782 return; 783 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) 784 g_mirror_fill_metadata(sc, disk, &md); 785 error = g_mirror_write_metadata(disk, &md); 786 if (error == 0) { 787 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 788 g_mirror_get_diskname(disk)); 789 } else { 790 G_MIRROR_DEBUG(0, 791 "Cannot update metadata on disk %s (error=%d).", 792 g_mirror_get_diskname(disk), error); 793 } 794 } 795 796 static void 797 g_mirror_bump_syncid(struct g_mirror_softc *sc) 798 { 799 struct g_mirror_disk *disk; 800 801 g_topology_assert_not(); 802 sx_assert(&sc->sc_lock, SX_XLOCKED); 803 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 804 ("%s called with no active disks (device=%s).", __func__, 805 sc->sc_name)); 806 807 sc->sc_syncid++; 808 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 809 sc->sc_syncid); 810 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 811 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 812 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 813 disk->d_sync.ds_syncid = sc->sc_syncid; 814 g_mirror_update_metadata(disk); 815 } 816 } 817 } 818 819 static void 820 g_mirror_bump_genid(struct g_mirror_softc *sc) 821 { 822 struct g_mirror_disk *disk; 823 824 g_topology_assert_not(); 825 sx_assert(&sc->sc_lock, SX_XLOCKED); 826 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 827 ("%s called with no active disks (device=%s).", __func__, 828 sc->sc_name)); 829 830 sc->sc_genid++; 831 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 832 sc->sc_genid); 833 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 834 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 835 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 836 disk->d_genid = sc->sc_genid; 837 g_mirror_update_metadata(disk); 838 } 839 } 840 } 841 842 static int 843 g_mirror_idle(struct g_mirror_softc *sc, int acw) 844 { 845 struct g_mirror_disk *disk; 846 int timeout; 847 848 g_topology_assert_not(); 849 sx_assert(&sc->sc_lock, SX_XLOCKED); 850 851 if (sc->sc_provider == NULL) 852 return (0); 853 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 854 return (0); 855 if (sc->sc_idle) 856 return (0); 857 if (sc->sc_writes > 0) 858 return (0); 859 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 860 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); 861 if (!g_mirror_shutdown && timeout > 0) 862 return (timeout); 863 } 864 sc->sc_idle = 1; 865 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 866 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 867 continue; 868 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 869 g_mirror_get_diskname(disk), sc->sc_name); 870 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 871 g_mirror_update_metadata(disk); 872 } 873 return (0); 874 } 875 876 static void 877 g_mirror_unidle(struct g_mirror_softc *sc) 878 { 879 struct g_mirror_disk *disk; 880 881 g_topology_assert_not(); 882 sx_assert(&sc->sc_lock, SX_XLOCKED); 883 884 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 885 return; 886 sc->sc_idle = 0; 887 sc->sc_last_write = time_uptime; 888 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 889 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 890 continue; 891 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 892 g_mirror_get_diskname(disk), sc->sc_name); 893 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 894 g_mirror_update_metadata(disk); 895 } 896 } 897 898 static void 899 g_mirror_done(struct bio *bp) 900 { 901 struct g_mirror_softc *sc; 902 903 sc = bp->bio_from->geom->softc; 904 bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR; 905 mtx_lock(&sc->sc_queue_mtx); 906 TAILQ_INSERT_TAIL(&sc->sc_queue, bp, bio_queue); 907 mtx_unlock(&sc->sc_queue_mtx); 908 wakeup(sc); 909 } 910 911 static void 912 g_mirror_regular_request_error(struct g_mirror_softc *sc, 913 struct g_mirror_disk *disk, struct bio *bp) 914 { 915 916 if (bp->bio_cmd == BIO_FLUSH && bp->bio_error == EOPNOTSUPP) 917 return; 918 919 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 920 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 921 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 922 bp->bio_error); 923 } else { 924 G_MIRROR_LOGREQ(1, bp, "Request failed (error=%d).", 925 bp->bio_error); 926 } 927 if (g_mirror_disconnect_on_failure && 928 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 929 if (bp->bio_error == ENXIO && 930 bp->bio_cmd == BIO_READ) 931 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 932 else if (bp->bio_error == ENXIO) 933 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_NOW; 934 else 935 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 936 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 937 G_MIRROR_EVENT_DONTWAIT); 938 } 939 } 940 941 static void 942 g_mirror_regular_request(struct g_mirror_softc *sc, struct bio *bp) 943 { 944 struct g_mirror_disk *disk; 945 struct bio *pbp; 946 947 g_topology_assert_not(); 948 KASSERT(sc->sc_provider == bp->bio_parent->bio_to, 949 ("regular request %p with unexpected origin", bp)); 950 951 pbp = bp->bio_parent; 952 bp->bio_from->index--; 953 if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) 954 sc->sc_writes--; 955 disk = bp->bio_from->private; 956 if (disk == NULL) { 957 g_topology_lock(); 958 g_mirror_kill_consumer(sc, bp->bio_from); 959 g_topology_unlock(); 960 } 961 962 switch (bp->bio_cmd) { 963 case BIO_READ: 964 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_read, 965 bp->bio_error); 966 break; 967 case BIO_WRITE: 968 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_write, 969 bp->bio_error); 970 break; 971 case BIO_DELETE: 972 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_delete, 973 bp->bio_error); 974 break; 975 case BIO_FLUSH: 976 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_flush, 977 bp->bio_error); 978 break; 979 } 980 981 pbp->bio_inbed++; 982 KASSERT(pbp->bio_inbed <= pbp->bio_children, 983 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 984 pbp->bio_children)); 985 if (bp->bio_error == 0 && pbp->bio_error == 0) { 986 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 987 g_destroy_bio(bp); 988 if (pbp->bio_children == pbp->bio_inbed) { 989 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 990 pbp->bio_completed = pbp->bio_length; 991 if (pbp->bio_cmd == BIO_WRITE || 992 pbp->bio_cmd == BIO_DELETE) { 993 TAILQ_REMOVE(&sc->sc_inflight, pbp, bio_queue); 994 /* Release delayed sync requests if possible. */ 995 g_mirror_sync_release(sc); 996 } 997 g_io_deliver(pbp, pbp->bio_error); 998 } 999 return; 1000 } else if (bp->bio_error != 0) { 1001 if (pbp->bio_error == 0) 1002 pbp->bio_error = bp->bio_error; 1003 if (disk != NULL) 1004 g_mirror_regular_request_error(sc, disk, bp); 1005 switch (pbp->bio_cmd) { 1006 case BIO_DELETE: 1007 case BIO_WRITE: 1008 case BIO_FLUSH: 1009 pbp->bio_inbed--; 1010 pbp->bio_children--; 1011 break; 1012 } 1013 } 1014 g_destroy_bio(bp); 1015 1016 switch (pbp->bio_cmd) { 1017 case BIO_READ: 1018 if (pbp->bio_inbed < pbp->bio_children) 1019 break; 1020 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 1021 g_io_deliver(pbp, pbp->bio_error); 1022 else { 1023 pbp->bio_error = 0; 1024 mtx_lock(&sc->sc_queue_mtx); 1025 TAILQ_INSERT_TAIL(&sc->sc_queue, pbp, bio_queue); 1026 mtx_unlock(&sc->sc_queue_mtx); 1027 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1028 wakeup(sc); 1029 } 1030 break; 1031 case BIO_DELETE: 1032 case BIO_WRITE: 1033 case BIO_FLUSH: 1034 if (pbp->bio_children == 0) { 1035 /* 1036 * All requests failed. 1037 */ 1038 } else if (pbp->bio_inbed < pbp->bio_children) { 1039 /* Do nothing. */ 1040 break; 1041 } else if (pbp->bio_children == pbp->bio_inbed) { 1042 /* Some requests succeeded. */ 1043 pbp->bio_error = 0; 1044 pbp->bio_completed = pbp->bio_length; 1045 } 1046 if (pbp->bio_cmd == BIO_WRITE || pbp->bio_cmd == BIO_DELETE) { 1047 TAILQ_REMOVE(&sc->sc_inflight, pbp, bio_queue); 1048 /* Release delayed sync requests if possible. */ 1049 g_mirror_sync_release(sc); 1050 } 1051 g_io_deliver(pbp, pbp->bio_error); 1052 break; 1053 default: 1054 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 1055 break; 1056 } 1057 } 1058 1059 static void 1060 g_mirror_sync_done(struct bio *bp) 1061 { 1062 struct g_mirror_softc *sc; 1063 1064 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1065 sc = bp->bio_from->geom->softc; 1066 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; 1067 mtx_lock(&sc->sc_queue_mtx); 1068 TAILQ_INSERT_TAIL(&sc->sc_queue, bp, bio_queue); 1069 mtx_unlock(&sc->sc_queue_mtx); 1070 wakeup(sc); 1071 } 1072 1073 static void 1074 g_mirror_candelete(struct bio *bp) 1075 { 1076 struct g_mirror_softc *sc; 1077 struct g_mirror_disk *disk; 1078 int *val; 1079 1080 sc = bp->bio_to->private; 1081 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1082 if (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) 1083 break; 1084 } 1085 val = (int *)bp->bio_data; 1086 *val = (disk != NULL); 1087 g_io_deliver(bp, 0); 1088 } 1089 1090 static void 1091 g_mirror_kernel_dump(struct bio *bp) 1092 { 1093 struct g_mirror_softc *sc; 1094 struct g_mirror_disk *disk; 1095 struct bio *cbp; 1096 struct g_kerneldump *gkd; 1097 1098 /* 1099 * We configure dumping to the first component, because this component 1100 * will be used for reading with 'prefer' balance algorithm. 1101 * If the component with the highest priority is currently disconnected 1102 * we will not be able to read the dump after the reboot if it will be 1103 * connected and synchronized later. Can we do something better? 1104 */ 1105 sc = bp->bio_to->private; 1106 disk = LIST_FIRST(&sc->sc_disks); 1107 1108 gkd = (struct g_kerneldump *)bp->bio_data; 1109 if (gkd->length > bp->bio_to->mediasize) 1110 gkd->length = bp->bio_to->mediasize; 1111 cbp = g_clone_bio(bp); 1112 if (cbp == NULL) { 1113 g_io_deliver(bp, ENOMEM); 1114 return; 1115 } 1116 cbp->bio_done = g_std_done; 1117 g_io_request(cbp, disk->d_consumer); 1118 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1119 g_mirror_get_diskname(disk)); 1120 } 1121 1122 static void 1123 g_mirror_start(struct bio *bp) 1124 { 1125 struct g_mirror_softc *sc; 1126 1127 sc = bp->bio_to->private; 1128 /* 1129 * If sc == NULL or there are no valid disks, provider's error 1130 * should be set and g_mirror_start() should not be called at all. 1131 */ 1132 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1133 ("Provider's error should be set (error=%d)(mirror=%s).", 1134 bp->bio_to->error, bp->bio_to->name)); 1135 G_MIRROR_LOGREQ(3, bp, "Request received."); 1136 1137 switch (bp->bio_cmd) { 1138 case BIO_READ: 1139 case BIO_WRITE: 1140 case BIO_DELETE: 1141 case BIO_FLUSH: 1142 break; 1143 case BIO_GETATTR: 1144 if (!strcmp(bp->bio_attribute, "GEOM::candelete")) { 1145 g_mirror_candelete(bp); 1146 return; 1147 } else if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1148 g_mirror_kernel_dump(bp); 1149 return; 1150 } 1151 /* FALLTHROUGH */ 1152 default: 1153 g_io_deliver(bp, EOPNOTSUPP); 1154 return; 1155 } 1156 mtx_lock(&sc->sc_queue_mtx); 1157 if (bp->bio_to->error != 0) { 1158 mtx_unlock(&sc->sc_queue_mtx); 1159 g_io_deliver(bp, bp->bio_to->error); 1160 return; 1161 } 1162 TAILQ_INSERT_TAIL(&sc->sc_queue, bp, bio_queue); 1163 mtx_unlock(&sc->sc_queue_mtx); 1164 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1165 wakeup(sc); 1166 } 1167 1168 /* 1169 * Return TRUE if the given request is colliding with a in-progress 1170 * synchronization request. 1171 */ 1172 static bool 1173 g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1174 { 1175 struct g_mirror_disk *disk; 1176 struct bio *sbp; 1177 off_t rstart, rend, sstart, send; 1178 u_int i; 1179 1180 if (sc->sc_sync.ds_ndisks == 0) 1181 return (false); 1182 rstart = bp->bio_offset; 1183 rend = bp->bio_offset + bp->bio_length; 1184 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1185 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1186 continue; 1187 for (i = 0; i < g_mirror_syncreqs; i++) { 1188 sbp = disk->d_sync.ds_bios[i]; 1189 if (sbp == NULL) 1190 continue; 1191 sstart = sbp->bio_offset; 1192 send = sbp->bio_offset + sbp->bio_length; 1193 if (rend > sstart && rstart < send) 1194 return (true); 1195 } 1196 } 1197 return (false); 1198 } 1199 1200 /* 1201 * Return TRUE if the given sync request is colliding with a in-progress regular 1202 * request. 1203 */ 1204 static bool 1205 g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1206 { 1207 off_t rstart, rend, sstart, send; 1208 struct bio *bp; 1209 1210 if (sc->sc_sync.ds_ndisks == 0) 1211 return (false); 1212 sstart = sbp->bio_offset; 1213 send = sbp->bio_offset + sbp->bio_length; 1214 TAILQ_FOREACH(bp, &sc->sc_inflight, bio_queue) { 1215 rstart = bp->bio_offset; 1216 rend = bp->bio_offset + bp->bio_length; 1217 if (rend > sstart && rstart < send) 1218 return (true); 1219 } 1220 return (false); 1221 } 1222 1223 /* 1224 * Puts regular request onto delayed queue. 1225 */ 1226 static void 1227 g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1228 { 1229 1230 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1231 TAILQ_INSERT_TAIL(&sc->sc_regular_delayed, bp, bio_queue); 1232 } 1233 1234 /* 1235 * Puts synchronization request onto delayed queue. 1236 */ 1237 static void 1238 g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1239 { 1240 1241 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1242 TAILQ_INSERT_TAIL(&sc->sc_sync_delayed, bp, bio_queue); 1243 } 1244 1245 /* 1246 * Requeue delayed regular requests. 1247 */ 1248 static void 1249 g_mirror_regular_release(struct g_mirror_softc *sc) 1250 { 1251 struct bio *bp; 1252 1253 if ((bp = TAILQ_FIRST(&sc->sc_regular_delayed)) == NULL) 1254 return; 1255 if (g_mirror_sync_collision(sc, bp)) 1256 return; 1257 1258 G_MIRROR_DEBUG(2, "Requeuing regular requests after collision."); 1259 mtx_lock(&sc->sc_queue_mtx); 1260 TAILQ_CONCAT(&sc->sc_regular_delayed, &sc->sc_queue, bio_queue); 1261 TAILQ_SWAP(&sc->sc_regular_delayed, &sc->sc_queue, bio, bio_queue); 1262 mtx_unlock(&sc->sc_queue_mtx); 1263 } 1264 1265 /* 1266 * Releases delayed sync requests which don't collide anymore with regular 1267 * requests. 1268 */ 1269 static void 1270 g_mirror_sync_release(struct g_mirror_softc *sc) 1271 { 1272 struct bio *bp, *bp2; 1273 1274 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed, bio_queue, bp2) { 1275 if (g_mirror_regular_collision(sc, bp)) 1276 continue; 1277 TAILQ_REMOVE(&sc->sc_sync_delayed, bp, bio_queue); 1278 G_MIRROR_LOGREQ(2, bp, 1279 "Releasing delayed synchronization request."); 1280 g_io_request(bp, bp->bio_from); 1281 } 1282 } 1283 1284 /* 1285 * Free a synchronization request and clear its slot in the array. 1286 */ 1287 static void 1288 g_mirror_sync_request_free(struct g_mirror_disk *disk, struct bio *bp) 1289 { 1290 int idx; 1291 1292 if (disk != NULL && disk->d_sync.ds_bios != NULL) { 1293 idx = (int)(uintptr_t)bp->bio_caller1; 1294 KASSERT(disk->d_sync.ds_bios[idx] == bp, 1295 ("unexpected sync BIO at %p:%d", disk, idx)); 1296 disk->d_sync.ds_bios[idx] = NULL; 1297 } 1298 free(bp->bio_data, M_MIRROR); 1299 g_destroy_bio(bp); 1300 } 1301 1302 /* 1303 * Handle synchronization requests. 1304 * Every synchronization request is a two-step process: first, a read request is 1305 * sent to the mirror provider via the sync consumer. If that request completes 1306 * successfully, it is converted to a write and sent to the disk being 1307 * synchronized. If the write also completes successfully, the synchronization 1308 * offset is advanced and a new read request is submitted. 1309 */ 1310 static void 1311 g_mirror_sync_request(struct g_mirror_softc *sc, struct bio *bp) 1312 { 1313 struct g_mirror_disk *disk; 1314 struct g_mirror_disk_sync *sync; 1315 1316 KASSERT((bp->bio_cmd == BIO_READ && 1317 bp->bio_from->geom == sc->sc_sync.ds_geom) || 1318 (bp->bio_cmd == BIO_WRITE && bp->bio_from->geom == sc->sc_geom), 1319 ("Sync BIO %p with unexpected origin", bp)); 1320 1321 bp->bio_from->index--; 1322 disk = bp->bio_from->private; 1323 if (disk == NULL) { 1324 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1325 g_topology_lock(); 1326 g_mirror_kill_consumer(sc, bp->bio_from); 1327 g_topology_unlock(); 1328 g_mirror_sync_request_free(NULL, bp); 1329 sx_xlock(&sc->sc_lock); 1330 return; 1331 } 1332 1333 sync = &disk->d_sync; 1334 1335 /* 1336 * Synchronization request. 1337 */ 1338 switch (bp->bio_cmd) { 1339 case BIO_READ: { 1340 struct g_consumer *cp; 1341 1342 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_read, 1343 bp->bio_error); 1344 1345 if (bp->bio_error != 0) { 1346 G_MIRROR_LOGREQ(0, bp, 1347 "Synchronization request failed (error=%d).", 1348 bp->bio_error); 1349 1350 /* 1351 * The read error will trigger a syncid bump, so there's 1352 * no need to do that here. 1353 * 1354 * The read error handling for regular requests will 1355 * retry the read from all active mirrors before passing 1356 * the error back up, so there's no need to retry here. 1357 */ 1358 g_mirror_sync_request_free(disk, bp); 1359 g_mirror_event_send(disk, 1360 G_MIRROR_DISK_STATE_DISCONNECTED, 1361 G_MIRROR_EVENT_DONTWAIT); 1362 return; 1363 } 1364 G_MIRROR_LOGREQ(3, bp, 1365 "Synchronization request half-finished."); 1366 bp->bio_cmd = BIO_WRITE; 1367 bp->bio_cflags = 0; 1368 cp = disk->d_consumer; 1369 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1370 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1371 cp->acr, cp->acw, cp->ace)); 1372 cp->index++; 1373 g_io_request(bp, cp); 1374 return; 1375 } 1376 case BIO_WRITE: { 1377 off_t offset; 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 if (sync->ds_offset >= sc->sc_mediasize || 1396 sync->ds_consumer == NULL || 1397 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1398 /* Don't send more synchronization requests. */ 1399 sync->ds_inflight--; 1400 g_mirror_sync_request_free(disk, bp); 1401 if (sync->ds_inflight > 0) 1402 return; 1403 if (sync->ds_consumer == NULL || 1404 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1405 return; 1406 } 1407 /* Disk up-to-date, activate it. */ 1408 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1409 G_MIRROR_EVENT_DONTWAIT); 1410 return; 1411 } 1412 1413 /* Send next synchronization request. */ 1414 g_mirror_sync_reinit(disk, bp, sync->ds_offset); 1415 sync->ds_offset += bp->bio_length; 1416 1417 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1418 sync->ds_consumer->index++; 1419 1420 /* 1421 * Delay the request if it is colliding with a regular request. 1422 */ 1423 if (g_mirror_regular_collision(sc, bp)) 1424 g_mirror_sync_delay(sc, bp); 1425 else 1426 g_io_request(bp, sync->ds_consumer); 1427 1428 /* Requeue delayed requests if possible. */ 1429 g_mirror_regular_release(sc); 1430 1431 /* Find the smallest offset */ 1432 offset = sc->sc_mediasize; 1433 for (i = 0; i < g_mirror_syncreqs; i++) { 1434 bp = sync->ds_bios[i]; 1435 if (bp != NULL && bp->bio_offset < offset) 1436 offset = bp->bio_offset; 1437 } 1438 if (g_mirror_sync_period > 0 && 1439 time_uptime - sync->ds_update_ts > g_mirror_sync_period) { 1440 sync->ds_offset_done = offset; 1441 g_mirror_update_metadata(disk); 1442 sync->ds_update_ts = time_uptime; 1443 } 1444 return; 1445 } 1446 default: 1447 panic("Invalid I/O request %p", bp); 1448 } 1449 } 1450 1451 static void 1452 g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1453 { 1454 struct g_mirror_disk *disk; 1455 struct g_consumer *cp; 1456 struct bio *cbp; 1457 1458 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1459 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1460 break; 1461 } 1462 if (disk == NULL) { 1463 if (bp->bio_error == 0) 1464 bp->bio_error = ENXIO; 1465 g_io_deliver(bp, bp->bio_error); 1466 return; 1467 } 1468 cbp = g_clone_bio(bp); 1469 if (cbp == NULL) { 1470 if (bp->bio_error == 0) 1471 bp->bio_error = ENOMEM; 1472 g_io_deliver(bp, bp->bio_error); 1473 return; 1474 } 1475 /* 1476 * Fill in the component buf structure. 1477 */ 1478 cp = disk->d_consumer; 1479 cbp->bio_done = g_mirror_done; 1480 cbp->bio_to = cp->provider; 1481 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1482 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1483 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1484 cp->acw, cp->ace)); 1485 cp->index++; 1486 g_io_request(cbp, cp); 1487 } 1488 1489 static void 1490 g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1491 { 1492 struct g_mirror_disk *disk; 1493 struct g_consumer *cp; 1494 struct bio *cbp; 1495 1496 disk = g_mirror_get_disk(sc); 1497 if (disk == NULL) { 1498 if (bp->bio_error == 0) 1499 bp->bio_error = ENXIO; 1500 g_io_deliver(bp, bp->bio_error); 1501 return; 1502 } 1503 cbp = g_clone_bio(bp); 1504 if (cbp == NULL) { 1505 if (bp->bio_error == 0) 1506 bp->bio_error = ENOMEM; 1507 g_io_deliver(bp, bp->bio_error); 1508 return; 1509 } 1510 /* 1511 * Fill in the component buf structure. 1512 */ 1513 cp = disk->d_consumer; 1514 cbp->bio_done = g_mirror_done; 1515 cbp->bio_to = cp->provider; 1516 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1517 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1518 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1519 cp->acw, cp->ace)); 1520 cp->index++; 1521 g_io_request(cbp, cp); 1522 } 1523 1524 #define TRACK_SIZE (1 * 1024 * 1024) 1525 #define LOAD_SCALE 256 1526 #define ABS(x) (((x) >= 0) ? (x) : (-(x))) 1527 1528 static void 1529 g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1530 { 1531 struct g_mirror_disk *disk, *dp; 1532 struct g_consumer *cp; 1533 struct bio *cbp; 1534 int prio, best; 1535 1536 /* Find a disk with the smallest load. */ 1537 disk = NULL; 1538 best = INT_MAX; 1539 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1540 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1541 continue; 1542 prio = dp->load; 1543 /* If disk head is precisely in position - highly prefer it. */ 1544 if (dp->d_last_offset == bp->bio_offset) 1545 prio -= 2 * LOAD_SCALE; 1546 else 1547 /* If disk head is close to position - prefer it. */ 1548 if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE) 1549 prio -= 1 * LOAD_SCALE; 1550 if (prio <= best) { 1551 disk = dp; 1552 best = prio; 1553 } 1554 } 1555 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1556 cbp = g_clone_bio(bp); 1557 if (cbp == NULL) { 1558 if (bp->bio_error == 0) 1559 bp->bio_error = ENOMEM; 1560 g_io_deliver(bp, bp->bio_error); 1561 return; 1562 } 1563 /* 1564 * Fill in the component buf structure. 1565 */ 1566 cp = disk->d_consumer; 1567 cbp->bio_done = g_mirror_done; 1568 cbp->bio_to = cp->provider; 1569 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1570 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1571 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1572 cp->acw, cp->ace)); 1573 cp->index++; 1574 /* Remember last head position */ 1575 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1576 /* Update loads. */ 1577 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1578 dp->load = (dp->d_consumer->index * LOAD_SCALE + 1579 dp->load * 7) / 8; 1580 } 1581 g_io_request(cbp, cp); 1582 } 1583 1584 static void 1585 g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1586 { 1587 struct bio_queue queue; 1588 struct g_mirror_disk *disk; 1589 struct g_consumer *cp; 1590 struct bio *cbp; 1591 off_t left, mod, offset, slice; 1592 u_char *data; 1593 u_int ndisks; 1594 1595 if (bp->bio_length <= sc->sc_slice) { 1596 g_mirror_request_round_robin(sc, bp); 1597 return; 1598 } 1599 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1600 slice = bp->bio_length / ndisks; 1601 mod = slice % sc->sc_provider->sectorsize; 1602 if (mod != 0) 1603 slice += sc->sc_provider->sectorsize - mod; 1604 /* 1605 * Allocate all bios before sending any request, so we can 1606 * return ENOMEM in nice and clean way. 1607 */ 1608 left = bp->bio_length; 1609 offset = bp->bio_offset; 1610 data = bp->bio_data; 1611 TAILQ_INIT(&queue); 1612 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1613 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1614 continue; 1615 cbp = g_clone_bio(bp); 1616 if (cbp == NULL) { 1617 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1618 TAILQ_REMOVE(&queue, cbp, bio_queue); 1619 g_destroy_bio(cbp); 1620 } 1621 if (bp->bio_error == 0) 1622 bp->bio_error = ENOMEM; 1623 g_io_deliver(bp, bp->bio_error); 1624 return; 1625 } 1626 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 1627 cbp->bio_done = g_mirror_done; 1628 cbp->bio_caller1 = disk; 1629 cbp->bio_to = disk->d_consumer->provider; 1630 cbp->bio_offset = offset; 1631 cbp->bio_data = data; 1632 cbp->bio_length = MIN(left, slice); 1633 left -= cbp->bio_length; 1634 if (left == 0) 1635 break; 1636 offset += cbp->bio_length; 1637 data += cbp->bio_length; 1638 } 1639 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1640 TAILQ_REMOVE(&queue, cbp, bio_queue); 1641 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1642 disk = cbp->bio_caller1; 1643 cbp->bio_caller1 = NULL; 1644 cp = disk->d_consumer; 1645 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1646 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1647 cp->acr, cp->acw, cp->ace)); 1648 disk->d_consumer->index++; 1649 g_io_request(cbp, disk->d_consumer); 1650 } 1651 } 1652 1653 static void 1654 g_mirror_register_request(struct g_mirror_softc *sc, struct bio *bp) 1655 { 1656 struct bio_queue queue; 1657 struct bio *cbp; 1658 struct g_consumer *cp; 1659 struct g_mirror_disk *disk; 1660 1661 sx_assert(&sc->sc_lock, SA_XLOCKED); 1662 1663 /* 1664 * To avoid ordering issues, if a write is deferred because of a 1665 * collision with a sync request, all I/O is deferred until that 1666 * write is initiated. 1667 */ 1668 if (bp->bio_from->geom != sc->sc_sync.ds_geom && 1669 !TAILQ_EMPTY(&sc->sc_regular_delayed)) { 1670 g_mirror_regular_delay(sc, bp); 1671 return; 1672 } 1673 1674 switch (bp->bio_cmd) { 1675 case BIO_READ: 1676 switch (sc->sc_balance) { 1677 case G_MIRROR_BALANCE_LOAD: 1678 g_mirror_request_load(sc, bp); 1679 break; 1680 case G_MIRROR_BALANCE_PREFER: 1681 g_mirror_request_prefer(sc, bp); 1682 break; 1683 case G_MIRROR_BALANCE_ROUND_ROBIN: 1684 g_mirror_request_round_robin(sc, bp); 1685 break; 1686 case G_MIRROR_BALANCE_SPLIT: 1687 g_mirror_request_split(sc, bp); 1688 break; 1689 } 1690 return; 1691 case BIO_WRITE: 1692 case BIO_DELETE: 1693 /* 1694 * Delay the request if it is colliding with a synchronization 1695 * request. 1696 */ 1697 if (g_mirror_sync_collision(sc, bp)) { 1698 g_mirror_regular_delay(sc, bp); 1699 return; 1700 } 1701 1702 if (sc->sc_idle) 1703 g_mirror_unidle(sc); 1704 else 1705 sc->sc_last_write = time_uptime; 1706 1707 /* 1708 * Bump syncid on first write. 1709 */ 1710 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1711 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1712 g_mirror_bump_syncid(sc); 1713 } 1714 1715 /* 1716 * Allocate all bios before sending any request, so we can 1717 * return ENOMEM in nice and clean way. 1718 */ 1719 TAILQ_INIT(&queue); 1720 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1721 switch (disk->d_state) { 1722 case G_MIRROR_DISK_STATE_ACTIVE: 1723 break; 1724 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1725 if (bp->bio_offset >= disk->d_sync.ds_offset) 1726 continue; 1727 break; 1728 default: 1729 continue; 1730 } 1731 if (bp->bio_cmd == BIO_DELETE && 1732 (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0) 1733 continue; 1734 cbp = g_clone_bio(bp); 1735 if (cbp == NULL) { 1736 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1737 TAILQ_REMOVE(&queue, cbp, bio_queue); 1738 g_destroy_bio(cbp); 1739 } 1740 if (bp->bio_error == 0) 1741 bp->bio_error = ENOMEM; 1742 g_io_deliver(bp, bp->bio_error); 1743 return; 1744 } 1745 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 1746 cbp->bio_done = g_mirror_done; 1747 cp = disk->d_consumer; 1748 cbp->bio_caller1 = cp; 1749 cbp->bio_to = cp->provider; 1750 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1751 ("Consumer %s not opened (r%dw%de%d).", 1752 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1753 } 1754 if (TAILQ_EMPTY(&queue)) { 1755 KASSERT(bp->bio_cmd == BIO_DELETE, 1756 ("No consumers for regular request %p", bp)); 1757 g_io_deliver(bp, EOPNOTSUPP); 1758 return; 1759 } 1760 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1761 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1762 TAILQ_REMOVE(&queue, cbp, bio_queue); 1763 cp = cbp->bio_caller1; 1764 cbp->bio_caller1 = NULL; 1765 cp->index++; 1766 sc->sc_writes++; 1767 g_io_request(cbp, cp); 1768 } 1769 /* 1770 * Put request onto inflight queue, so we can check if new 1771 * synchronization requests don't collide with it. 1772 */ 1773 TAILQ_INSERT_TAIL(&sc->sc_inflight, bp, bio_queue); 1774 return; 1775 case BIO_FLUSH: 1776 TAILQ_INIT(&queue); 1777 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1778 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1779 continue; 1780 cbp = g_clone_bio(bp); 1781 if (cbp == NULL) { 1782 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1783 TAILQ_REMOVE(&queue, cbp, bio_queue); 1784 g_destroy_bio(cbp); 1785 } 1786 if (bp->bio_error == 0) 1787 bp->bio_error = ENOMEM; 1788 g_io_deliver(bp, bp->bio_error); 1789 return; 1790 } 1791 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 1792 cbp->bio_done = g_mirror_done; 1793 cbp->bio_caller1 = disk; 1794 cbp->bio_to = disk->d_consumer->provider; 1795 } 1796 KASSERT(!TAILQ_EMPTY(&queue), 1797 ("No consumers for regular request %p", bp)); 1798 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1799 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1800 TAILQ_REMOVE(&queue, cbp, bio_queue); 1801 disk = cbp->bio_caller1; 1802 cbp->bio_caller1 = NULL; 1803 cp = disk->d_consumer; 1804 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1805 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1806 cp->acr, cp->acw, cp->ace)); 1807 cp->index++; 1808 g_io_request(cbp, cp); 1809 } 1810 break; 1811 default: 1812 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1813 bp->bio_cmd, sc->sc_name)); 1814 break; 1815 } 1816 } 1817 1818 static int 1819 g_mirror_can_destroy(struct g_mirror_softc *sc) 1820 { 1821 struct g_geom *gp; 1822 struct g_consumer *cp; 1823 1824 g_topology_assert(); 1825 gp = sc->sc_geom; 1826 if (gp->softc == NULL) 1827 return (1); 1828 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0) 1829 return (0); 1830 LIST_FOREACH(cp, &gp->consumer, consumer) { 1831 if (g_mirror_is_busy(sc, cp)) 1832 return (0); 1833 } 1834 gp = sc->sc_sync.ds_geom; 1835 LIST_FOREACH(cp, &gp->consumer, consumer) { 1836 if (g_mirror_is_busy(sc, cp)) 1837 return (0); 1838 } 1839 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1840 sc->sc_name); 1841 return (1); 1842 } 1843 1844 static int 1845 g_mirror_try_destroy(struct g_mirror_softc *sc) 1846 { 1847 1848 if (sc->sc_rootmount != NULL) { 1849 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1850 sc->sc_rootmount); 1851 root_mount_rel(sc->sc_rootmount); 1852 sc->sc_rootmount = NULL; 1853 } 1854 g_topology_lock(); 1855 if (!g_mirror_can_destroy(sc)) { 1856 g_topology_unlock(); 1857 return (0); 1858 } 1859 sc->sc_geom->softc = NULL; 1860 sc->sc_sync.ds_geom->softc = NULL; 1861 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DRAIN) != 0) { 1862 g_topology_unlock(); 1863 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1864 &sc->sc_worker); 1865 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1866 sx_xunlock(&sc->sc_lock); 1867 wakeup(&sc->sc_worker); 1868 sc->sc_worker = NULL; 1869 } else { 1870 g_topology_unlock(); 1871 g_mirror_destroy_device(sc); 1872 } 1873 return (1); 1874 } 1875 1876 /* 1877 * Worker thread. 1878 */ 1879 static void 1880 g_mirror_worker(void *arg) 1881 { 1882 struct g_mirror_softc *sc; 1883 struct g_mirror_event *ep; 1884 struct bio *bp; 1885 int timeout; 1886 1887 sc = arg; 1888 thread_lock(curthread); 1889 sched_prio(curthread, PRIBIO); 1890 thread_unlock(curthread); 1891 1892 sx_xlock(&sc->sc_lock); 1893 for (;;) { 1894 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1895 /* 1896 * First take a look at events. 1897 * This is important to handle events before any I/O requests. 1898 */ 1899 ep = g_mirror_event_first(sc); 1900 if (ep != NULL) { 1901 g_mirror_event_remove(sc, ep); 1902 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1903 /* Update only device status. */ 1904 G_MIRROR_DEBUG(3, 1905 "Running event for device %s.", 1906 sc->sc_name); 1907 ep->e_error = 0; 1908 g_mirror_update_device(sc, true); 1909 } else { 1910 /* Update disk status. */ 1911 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1912 g_mirror_get_diskname(ep->e_disk)); 1913 ep->e_error = g_mirror_update_disk(ep->e_disk, 1914 ep->e_state); 1915 if (ep->e_error == 0) 1916 g_mirror_update_device(sc, false); 1917 } 1918 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1919 KASSERT(ep->e_error == 0, 1920 ("Error cannot be handled.")); 1921 g_mirror_event_free(ep); 1922 } else { 1923 ep->e_flags |= G_MIRROR_EVENT_DONE; 1924 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1925 ep); 1926 mtx_lock(&sc->sc_events_mtx); 1927 wakeup(ep); 1928 mtx_unlock(&sc->sc_events_mtx); 1929 } 1930 if ((sc->sc_flags & 1931 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1932 if (g_mirror_try_destroy(sc)) { 1933 curthread->td_pflags &= ~TDP_GEOM; 1934 G_MIRROR_DEBUG(1, "Thread exiting."); 1935 kproc_exit(0); 1936 } 1937 } 1938 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1939 continue; 1940 } 1941 1942 /* 1943 * Check if we can mark array as CLEAN and if we can't take 1944 * how much seconds should we wait. 1945 */ 1946 timeout = g_mirror_idle(sc, -1); 1947 1948 /* 1949 * Handle I/O requests. 1950 */ 1951 mtx_lock(&sc->sc_queue_mtx); 1952 bp = TAILQ_FIRST(&sc->sc_queue); 1953 if (bp != NULL) 1954 TAILQ_REMOVE(&sc->sc_queue, bp, bio_queue); 1955 else { 1956 if ((sc->sc_flags & 1957 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1958 mtx_unlock(&sc->sc_queue_mtx); 1959 if (g_mirror_try_destroy(sc)) { 1960 curthread->td_pflags &= ~TDP_GEOM; 1961 G_MIRROR_DEBUG(1, "Thread exiting."); 1962 kproc_exit(0); 1963 } 1964 mtx_lock(&sc->sc_queue_mtx); 1965 if (!TAILQ_EMPTY(&sc->sc_queue)) { 1966 mtx_unlock(&sc->sc_queue_mtx); 1967 continue; 1968 } 1969 } 1970 if (g_mirror_event_first(sc) != NULL) { 1971 mtx_unlock(&sc->sc_queue_mtx); 1972 continue; 1973 } 1974 sx_xunlock(&sc->sc_lock); 1975 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1976 timeout * hz); 1977 sx_xlock(&sc->sc_lock); 1978 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1979 continue; 1980 } 1981 mtx_unlock(&sc->sc_queue_mtx); 1982 1983 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 1984 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1985 /* 1986 * Handle completion of the first half (the read) of a 1987 * block synchronization operation. 1988 */ 1989 g_mirror_sync_request(sc, bp); 1990 } else if (bp->bio_to != sc->sc_provider) { 1991 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 1992 /* 1993 * Handle completion of a regular I/O request. 1994 */ 1995 g_mirror_regular_request(sc, bp); 1996 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 1997 /* 1998 * Handle completion of the second half (the 1999 * write) of a block synchronization operation. 2000 */ 2001 g_mirror_sync_request(sc, bp); 2002 else { 2003 KASSERT(0, 2004 ("Invalid request cflags=0x%hx to=%s.", 2005 bp->bio_cflags, bp->bio_to->name)); 2006 } 2007 } else { 2008 /* 2009 * Initiate an I/O request. 2010 */ 2011 g_mirror_register_request(sc, bp); 2012 } 2013 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 2014 } 2015 } 2016 2017 static void 2018 g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 2019 { 2020 2021 sx_assert(&sc->sc_lock, SX_LOCKED); 2022 2023 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 2024 return; 2025 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2026 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 2027 g_mirror_get_diskname(disk), sc->sc_name); 2028 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2029 } else if (sc->sc_idle && 2030 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2031 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 2032 g_mirror_get_diskname(disk), sc->sc_name); 2033 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2034 } 2035 } 2036 2037 static void 2038 g_mirror_sync_reinit(const struct g_mirror_disk *disk, struct bio *bp, 2039 off_t offset) 2040 { 2041 void *data; 2042 int idx; 2043 2044 data = bp->bio_data; 2045 idx = (int)(uintptr_t)bp->bio_caller1; 2046 g_reset_bio(bp); 2047 2048 bp->bio_cmd = BIO_READ; 2049 bp->bio_data = data; 2050 bp->bio_done = g_mirror_sync_done; 2051 bp->bio_from = disk->d_sync.ds_consumer; 2052 bp->bio_to = disk->d_softc->sc_provider; 2053 bp->bio_caller1 = (void *)(uintptr_t)idx; 2054 bp->bio_offset = offset; 2055 bp->bio_length = MIN(MAXPHYS, 2056 disk->d_softc->sc_mediasize - bp->bio_offset); 2057 } 2058 2059 static void 2060 g_mirror_sync_start(struct g_mirror_disk *disk) 2061 { 2062 struct g_mirror_softc *sc; 2063 struct g_mirror_disk_sync *sync; 2064 struct g_consumer *cp; 2065 struct bio *bp; 2066 int error, i; 2067 2068 g_topology_assert_not(); 2069 sc = disk->d_softc; 2070 sync = &disk->d_sync; 2071 sx_assert(&sc->sc_lock, SX_LOCKED); 2072 2073 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2074 ("Disk %s is not marked for synchronization.", 2075 g_mirror_get_diskname(disk))); 2076 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2077 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 2078 sc->sc_state)); 2079 2080 sx_xunlock(&sc->sc_lock); 2081 g_topology_lock(); 2082 cp = g_new_consumer(sc->sc_sync.ds_geom); 2083 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 2084 error = g_attach(cp, sc->sc_provider); 2085 KASSERT(error == 0, 2086 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 2087 error = g_access(cp, 1, 0, 0); 2088 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 2089 g_topology_unlock(); 2090 sx_xlock(&sc->sc_lock); 2091 2092 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 2093 g_mirror_get_diskname(disk)); 2094 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0) 2095 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2096 KASSERT(sync->ds_consumer == NULL, 2097 ("Sync consumer already exists (device=%s, disk=%s).", 2098 sc->sc_name, g_mirror_get_diskname(disk))); 2099 2100 sync->ds_consumer = cp; 2101 sync->ds_consumer->private = disk; 2102 sync->ds_consumer->index = 0; 2103 2104 /* 2105 * Allocate memory for synchronization bios and initialize them. 2106 */ 2107 sync->ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 2108 M_MIRROR, M_WAITOK); 2109 for (i = 0; i < g_mirror_syncreqs; i++) { 2110 bp = g_alloc_bio(); 2111 sync->ds_bios[i] = bp; 2112 2113 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 2114 bp->bio_caller1 = (void *)(uintptr_t)i; 2115 g_mirror_sync_reinit(disk, bp, sync->ds_offset); 2116 sync->ds_offset += bp->bio_length; 2117 } 2118 2119 /* Increase the number of disks in SYNCHRONIZING state. */ 2120 sc->sc_sync.ds_ndisks++; 2121 /* Set the number of in-flight synchronization requests. */ 2122 sync->ds_inflight = g_mirror_syncreqs; 2123 2124 /* 2125 * Fire off first synchronization requests. 2126 */ 2127 for (i = 0; i < g_mirror_syncreqs; i++) { 2128 bp = sync->ds_bios[i]; 2129 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 2130 sync->ds_consumer->index++; 2131 /* 2132 * Delay the request if it is colliding with a regular request. 2133 */ 2134 if (g_mirror_regular_collision(sc, bp)) 2135 g_mirror_sync_delay(sc, bp); 2136 else 2137 g_io_request(bp, sync->ds_consumer); 2138 } 2139 } 2140 2141 /* 2142 * Stop synchronization process. 2143 * type: 0 - synchronization finished 2144 * 1 - synchronization stopped 2145 */ 2146 static void 2147 g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 2148 { 2149 struct g_mirror_softc *sc; 2150 struct g_consumer *cp; 2151 2152 g_topology_assert_not(); 2153 sc = disk->d_softc; 2154 sx_assert(&sc->sc_lock, SX_LOCKED); 2155 2156 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2157 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2158 g_mirror_disk_state2str(disk->d_state))); 2159 if (disk->d_sync.ds_consumer == NULL) 2160 return; 2161 2162 if (type == 0) { 2163 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2164 sc->sc_name, g_mirror_get_diskname(disk)); 2165 } else /* if (type == 1) */ { 2166 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2167 sc->sc_name, g_mirror_get_diskname(disk)); 2168 } 2169 g_mirror_regular_release(sc); 2170 free(disk->d_sync.ds_bios, M_MIRROR); 2171 disk->d_sync.ds_bios = NULL; 2172 cp = disk->d_sync.ds_consumer; 2173 disk->d_sync.ds_consumer = NULL; 2174 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2175 sc->sc_sync.ds_ndisks--; 2176 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2177 g_topology_lock(); 2178 g_mirror_kill_consumer(sc, cp); 2179 g_topology_unlock(); 2180 sx_xlock(&sc->sc_lock); 2181 } 2182 2183 static void 2184 g_mirror_launch_provider(struct g_mirror_softc *sc) 2185 { 2186 struct g_mirror_disk *disk; 2187 struct g_provider *pp, *dp; 2188 2189 sx_assert(&sc->sc_lock, SX_LOCKED); 2190 2191 g_topology_lock(); 2192 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 2193 pp->flags |= G_PF_DIRECT_RECEIVE; 2194 pp->mediasize = sc->sc_mediasize; 2195 pp->sectorsize = sc->sc_sectorsize; 2196 pp->stripesize = 0; 2197 pp->stripeoffset = 0; 2198 2199 /* Splitting of unmapped BIO's could work but isn't implemented now */ 2200 if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT) 2201 pp->flags |= G_PF_ACCEPT_UNMAPPED; 2202 2203 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2204 if (disk->d_consumer && disk->d_consumer->provider) { 2205 dp = disk->d_consumer->provider; 2206 if (dp->stripesize > pp->stripesize) { 2207 pp->stripesize = dp->stripesize; 2208 pp->stripeoffset = dp->stripeoffset; 2209 } 2210 /* A provider underneath us doesn't support unmapped */ 2211 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { 2212 G_MIRROR_DEBUG(0, "Cancelling unmapped " 2213 "because of %s.", dp->name); 2214 pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 2215 } 2216 } 2217 } 2218 pp->private = sc; 2219 sc->sc_refcnt++; 2220 sc->sc_provider = pp; 2221 g_error_provider(pp, 0); 2222 g_topology_unlock(); 2223 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2224 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 2225 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2226 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2227 g_mirror_sync_start(disk); 2228 } 2229 } 2230 2231 static void 2232 g_mirror_destroy_provider(struct g_mirror_softc *sc) 2233 { 2234 struct g_mirror_disk *disk; 2235 struct bio *bp; 2236 2237 g_topology_assert_not(); 2238 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2239 sc->sc_name)); 2240 2241 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2242 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2243 g_mirror_sync_stop(disk, 1); 2244 } 2245 2246 g_topology_lock(); 2247 g_error_provider(sc->sc_provider, ENXIO); 2248 mtx_lock(&sc->sc_queue_mtx); 2249 while ((bp = TAILQ_FIRST(&sc->sc_queue)) != NULL) { 2250 TAILQ_REMOVE(&sc->sc_queue, bp, bio_queue); 2251 /* 2252 * Abort any pending I/O that wasn't generated by us. 2253 * Synchronization requests and requests destined for individual 2254 * mirror components can be destroyed immediately. 2255 */ 2256 if (bp->bio_to == sc->sc_provider && 2257 bp->bio_from->geom != sc->sc_sync.ds_geom) { 2258 g_io_deliver(bp, ENXIO); 2259 } else { 2260 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2261 free(bp->bio_data, M_MIRROR); 2262 g_destroy_bio(bp); 2263 } 2264 } 2265 mtx_unlock(&sc->sc_queue_mtx); 2266 g_wither_provider(sc->sc_provider, ENXIO); 2267 sc->sc_provider = NULL; 2268 G_MIRROR_DEBUG(0, "Device %s: provider destroyed.", sc->sc_name); 2269 g_topology_unlock(); 2270 } 2271 2272 static void 2273 g_mirror_go(void *arg) 2274 { 2275 struct g_mirror_softc *sc; 2276 2277 sc = arg; 2278 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2279 g_mirror_event_send(sc, 0, 2280 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2281 } 2282 2283 static u_int 2284 g_mirror_determine_state(struct g_mirror_disk *disk) 2285 { 2286 struct g_mirror_softc *sc; 2287 u_int state; 2288 2289 sc = disk->d_softc; 2290 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2291 if ((disk->d_flags & 2292 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0 && 2293 (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 || 2294 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0)) { 2295 /* Disk does not need synchronization. */ 2296 state = G_MIRROR_DISK_STATE_ACTIVE; 2297 } else { 2298 if ((sc->sc_flags & 2299 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2300 (disk->d_flags & 2301 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2302 /* 2303 * We can start synchronization from 2304 * the stored offset. 2305 */ 2306 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2307 } else { 2308 state = G_MIRROR_DISK_STATE_STALE; 2309 } 2310 } 2311 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2312 /* 2313 * Reset all synchronization data for this disk, 2314 * because if it even was synchronized, it was 2315 * synchronized to disks with different syncid. 2316 */ 2317 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2318 disk->d_sync.ds_offset = 0; 2319 disk->d_sync.ds_offset_done = 0; 2320 disk->d_sync.ds_syncid = sc->sc_syncid; 2321 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2322 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2323 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2324 } else { 2325 state = G_MIRROR_DISK_STATE_STALE; 2326 } 2327 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2328 /* 2329 * Not good, NOT GOOD! 2330 * It means that mirror was started on stale disks 2331 * and more fresh disk just arrive. 2332 * If there were writes, mirror is broken, sorry. 2333 * I think the best choice here is don't touch 2334 * this disk and inform the user loudly. 2335 */ 2336 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2337 "disk (%s) arrives!! It will not be connected to the " 2338 "running device.", sc->sc_name, 2339 g_mirror_get_diskname(disk)); 2340 g_mirror_destroy_disk(disk); 2341 state = G_MIRROR_DISK_STATE_NONE; 2342 /* Return immediately, because disk was destroyed. */ 2343 return (state); 2344 } 2345 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2346 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2347 return (state); 2348 } 2349 2350 /* 2351 * Update device state. 2352 */ 2353 static void 2354 g_mirror_update_device(struct g_mirror_softc *sc, bool force) 2355 { 2356 struct g_mirror_disk *disk; 2357 u_int state; 2358 2359 sx_assert(&sc->sc_lock, SX_XLOCKED); 2360 2361 switch (sc->sc_state) { 2362 case G_MIRROR_DEVICE_STATE_STARTING: 2363 { 2364 struct g_mirror_disk *pdisk, *tdisk; 2365 u_int dirty, ndisks, genid, syncid; 2366 bool broken; 2367 2368 KASSERT(sc->sc_provider == NULL, 2369 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2370 /* 2371 * Are we ready? We are, if all disks are connected or 2372 * if we have any disks and 'force' is true. 2373 */ 2374 ndisks = g_mirror_ndisks(sc, -1); 2375 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2376 ; 2377 } else if (ndisks == 0) { 2378 /* 2379 * Disks went down in starting phase, so destroy 2380 * device. 2381 */ 2382 callout_drain(&sc->sc_callout); 2383 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2384 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2385 sc->sc_rootmount); 2386 root_mount_rel(sc->sc_rootmount); 2387 sc->sc_rootmount = NULL; 2388 return; 2389 } else { 2390 return; 2391 } 2392 2393 /* 2394 * Activate all disks with the biggest syncid. 2395 */ 2396 if (force) { 2397 /* 2398 * If 'force' is true, we have been called due to 2399 * timeout, so don't bother canceling timeout. 2400 */ 2401 ndisks = 0; 2402 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2403 if ((disk->d_flags & 2404 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2405 ndisks++; 2406 } 2407 } 2408 if (ndisks == 0) { 2409 /* No valid disks found, destroy device. */ 2410 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2411 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2412 __LINE__, sc->sc_rootmount); 2413 root_mount_rel(sc->sc_rootmount); 2414 sc->sc_rootmount = NULL; 2415 return; 2416 } 2417 } else { 2418 /* Cancel timeout. */ 2419 callout_drain(&sc->sc_callout); 2420 } 2421 2422 /* 2423 * Find the biggest genid. 2424 */ 2425 genid = 0; 2426 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2427 if (disk->d_genid > genid) 2428 genid = disk->d_genid; 2429 } 2430 sc->sc_genid = genid; 2431 /* 2432 * Remove all disks without the biggest genid. 2433 */ 2434 broken = false; 2435 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2436 if (disk->d_genid < genid) { 2437 G_MIRROR_DEBUG(0, 2438 "Component %s (device %s) broken, skipping.", 2439 g_mirror_get_diskname(disk), sc->sc_name); 2440 g_mirror_destroy_disk(disk); 2441 /* 2442 * Bump the syncid in case we discover a healthy 2443 * replacement disk after starting the mirror. 2444 */ 2445 broken = true; 2446 } 2447 } 2448 2449 /* 2450 * Find the biggest syncid. 2451 */ 2452 syncid = 0; 2453 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2454 if (disk->d_sync.ds_syncid > syncid) 2455 syncid = disk->d_sync.ds_syncid; 2456 } 2457 2458 /* 2459 * Here we need to look for dirty disks and if all disks 2460 * with the biggest syncid are dirty, we have to choose 2461 * one with the biggest priority and rebuild the rest. 2462 */ 2463 /* 2464 * Find the number of dirty disks with the biggest syncid. 2465 * Find the number of disks with the biggest syncid. 2466 * While here, find a disk with the biggest priority. 2467 */ 2468 dirty = ndisks = 0; 2469 pdisk = NULL; 2470 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2471 if (disk->d_sync.ds_syncid != syncid) 2472 continue; 2473 if ((disk->d_flags & 2474 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2475 continue; 2476 } 2477 ndisks++; 2478 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2479 dirty++; 2480 if (pdisk == NULL || 2481 pdisk->d_priority < disk->d_priority) { 2482 pdisk = disk; 2483 } 2484 } 2485 } 2486 if (dirty == 0) { 2487 /* No dirty disks at all, great. */ 2488 } else if (dirty == ndisks) { 2489 /* 2490 * Force synchronization for all dirty disks except one 2491 * with the biggest priority. 2492 */ 2493 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2494 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2495 "master disk for synchronization.", 2496 g_mirror_get_diskname(pdisk), sc->sc_name); 2497 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2498 if (disk->d_sync.ds_syncid != syncid) 2499 continue; 2500 if ((disk->d_flags & 2501 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2502 continue; 2503 } 2504 KASSERT((disk->d_flags & 2505 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2506 ("Disk %s isn't marked as dirty.", 2507 g_mirror_get_diskname(disk))); 2508 /* Skip the disk with the biggest priority. */ 2509 if (disk == pdisk) 2510 continue; 2511 disk->d_sync.ds_syncid = 0; 2512 } 2513 } else if (dirty < ndisks) { 2514 /* 2515 * Force synchronization for all dirty disks. 2516 * We have some non-dirty disks. 2517 */ 2518 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2519 if (disk->d_sync.ds_syncid != syncid) 2520 continue; 2521 if ((disk->d_flags & 2522 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2523 continue; 2524 } 2525 if ((disk->d_flags & 2526 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2527 continue; 2528 } 2529 disk->d_sync.ds_syncid = 0; 2530 } 2531 } 2532 2533 /* Reset hint. */ 2534 sc->sc_hint = NULL; 2535 sc->sc_syncid = syncid; 2536 if (force || broken) { 2537 /* Remember to bump syncid on first write. */ 2538 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2539 } 2540 state = G_MIRROR_DEVICE_STATE_RUNNING; 2541 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2542 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2543 g_mirror_device_state2str(state)); 2544 sc->sc_state = state; 2545 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2546 state = g_mirror_determine_state(disk); 2547 g_mirror_event_send(disk, state, 2548 G_MIRROR_EVENT_DONTWAIT); 2549 if (state == G_MIRROR_DISK_STATE_STALE) 2550 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2551 } 2552 break; 2553 } 2554 case G_MIRROR_DEVICE_STATE_RUNNING: 2555 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2556 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2557 /* 2558 * No usable disks, so destroy the device. 2559 */ 2560 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2561 break; 2562 } else if (g_mirror_ndisks(sc, 2563 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2564 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2565 /* 2566 * We have active disks, launch provider if it doesn't 2567 * exist. 2568 */ 2569 if (sc->sc_provider == NULL) 2570 g_mirror_launch_provider(sc); 2571 if (sc->sc_rootmount != NULL) { 2572 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2573 __LINE__, sc->sc_rootmount); 2574 root_mount_rel(sc->sc_rootmount); 2575 sc->sc_rootmount = NULL; 2576 } 2577 } 2578 /* 2579 * Genid should be bumped immediately, so do it here. 2580 */ 2581 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2582 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2583 g_mirror_bump_genid(sc); 2584 } 2585 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_NOW) != 0) { 2586 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID_NOW; 2587 g_mirror_bump_syncid(sc); 2588 } 2589 break; 2590 default: 2591 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2592 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2593 break; 2594 } 2595 } 2596 2597 /* 2598 * Update disk state and device state if needed. 2599 */ 2600 #define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2601 "Disk %s state changed from %s to %s (device %s).", \ 2602 g_mirror_get_diskname(disk), \ 2603 g_mirror_disk_state2str(disk->d_state), \ 2604 g_mirror_disk_state2str(state), sc->sc_name) 2605 static int 2606 g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2607 { 2608 struct g_mirror_softc *sc; 2609 2610 sc = disk->d_softc; 2611 sx_assert(&sc->sc_lock, SX_XLOCKED); 2612 2613 again: 2614 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2615 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2616 g_mirror_disk_state2str(state)); 2617 switch (state) { 2618 case G_MIRROR_DISK_STATE_NEW: 2619 /* 2620 * Possible scenarios: 2621 * 1. New disk arrive. 2622 */ 2623 /* Previous state should be NONE. */ 2624 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2625 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2626 g_mirror_disk_state2str(disk->d_state))); 2627 DISK_STATE_CHANGED(); 2628 2629 disk->d_state = state; 2630 g_topology_lock(); 2631 if (LIST_EMPTY(&sc->sc_disks)) 2632 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2633 else { 2634 struct g_mirror_disk *dp; 2635 2636 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2637 if (disk->d_priority >= dp->d_priority) { 2638 LIST_INSERT_BEFORE(dp, disk, d_next); 2639 dp = NULL; 2640 break; 2641 } 2642 if (LIST_NEXT(dp, d_next) == NULL) 2643 break; 2644 } 2645 if (dp != NULL) 2646 LIST_INSERT_AFTER(dp, disk, d_next); 2647 } 2648 g_topology_unlock(); 2649 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2650 sc->sc_name, g_mirror_get_diskname(disk)); 2651 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2652 break; 2653 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2654 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2655 g_mirror_device_state2str(sc->sc_state), 2656 g_mirror_get_diskname(disk), 2657 g_mirror_disk_state2str(disk->d_state))); 2658 state = g_mirror_determine_state(disk); 2659 if (state != G_MIRROR_DISK_STATE_NONE) 2660 goto again; 2661 break; 2662 case G_MIRROR_DISK_STATE_ACTIVE: 2663 /* 2664 * Possible scenarios: 2665 * 1. New disk does not need synchronization. 2666 * 2. Synchronization process finished successfully. 2667 */ 2668 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2669 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2670 g_mirror_device_state2str(sc->sc_state), 2671 g_mirror_get_diskname(disk), 2672 g_mirror_disk_state2str(disk->d_state))); 2673 /* Previous state should be NEW or SYNCHRONIZING. */ 2674 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2675 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2676 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2677 g_mirror_disk_state2str(disk->d_state))); 2678 DISK_STATE_CHANGED(); 2679 2680 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2681 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2682 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2683 g_mirror_sync_stop(disk, 0); 2684 } 2685 disk->d_state = state; 2686 disk->d_sync.ds_offset = 0; 2687 disk->d_sync.ds_offset_done = 0; 2688 g_mirror_update_idle(sc, disk); 2689 g_mirror_update_metadata(disk); 2690 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2691 sc->sc_name, g_mirror_get_diskname(disk)); 2692 break; 2693 case G_MIRROR_DISK_STATE_STALE: 2694 /* 2695 * Possible scenarios: 2696 * 1. Stale disk was connected. 2697 */ 2698 /* Previous state should be NEW. */ 2699 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2700 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2701 g_mirror_disk_state2str(disk->d_state))); 2702 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2703 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2704 g_mirror_device_state2str(sc->sc_state), 2705 g_mirror_get_diskname(disk), 2706 g_mirror_disk_state2str(disk->d_state))); 2707 /* 2708 * STALE state is only possible if device is marked 2709 * NOAUTOSYNC. 2710 */ 2711 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2712 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2713 g_mirror_device_state2str(sc->sc_state), 2714 g_mirror_get_diskname(disk), 2715 g_mirror_disk_state2str(disk->d_state))); 2716 DISK_STATE_CHANGED(); 2717 2718 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2719 disk->d_state = state; 2720 g_mirror_update_metadata(disk); 2721 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2722 sc->sc_name, g_mirror_get_diskname(disk)); 2723 break; 2724 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2725 /* 2726 * Possible scenarios: 2727 * 1. Disk which needs synchronization was connected. 2728 */ 2729 /* Previous state should be NEW. */ 2730 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2731 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2732 g_mirror_disk_state2str(disk->d_state))); 2733 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2734 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2735 g_mirror_device_state2str(sc->sc_state), 2736 g_mirror_get_diskname(disk), 2737 g_mirror_disk_state2str(disk->d_state))); 2738 DISK_STATE_CHANGED(); 2739 2740 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2741 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2742 disk->d_state = state; 2743 if (sc->sc_provider != NULL) { 2744 g_mirror_sync_start(disk); 2745 g_mirror_update_metadata(disk); 2746 } 2747 break; 2748 case G_MIRROR_DISK_STATE_DISCONNECTED: 2749 /* 2750 * Possible scenarios: 2751 * 1. Device wasn't running yet, but disk disappear. 2752 * 2. Disk was active and disapppear. 2753 * 3. Disk disappear during synchronization process. 2754 */ 2755 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2756 /* 2757 * Previous state should be ACTIVE, STALE or 2758 * SYNCHRONIZING. 2759 */ 2760 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2761 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2762 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2763 ("Wrong disk state (%s, %s).", 2764 g_mirror_get_diskname(disk), 2765 g_mirror_disk_state2str(disk->d_state))); 2766 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2767 /* Previous state should be NEW. */ 2768 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2769 ("Wrong disk state (%s, %s).", 2770 g_mirror_get_diskname(disk), 2771 g_mirror_disk_state2str(disk->d_state))); 2772 /* 2773 * Reset bumping syncid if disk disappeared in STARTING 2774 * state. 2775 */ 2776 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2777 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2778 #ifdef INVARIANTS 2779 } else { 2780 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2781 sc->sc_name, 2782 g_mirror_device_state2str(sc->sc_state), 2783 g_mirror_get_diskname(disk), 2784 g_mirror_disk_state2str(disk->d_state))); 2785 #endif 2786 } 2787 DISK_STATE_CHANGED(); 2788 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2789 sc->sc_name, g_mirror_get_diskname(disk)); 2790 2791 g_mirror_destroy_disk(disk); 2792 break; 2793 case G_MIRROR_DISK_STATE_DESTROY: 2794 { 2795 int error; 2796 2797 error = g_mirror_clear_metadata(disk); 2798 if (error != 0) { 2799 G_MIRROR_DEBUG(0, 2800 "Device %s: failed to clear metadata on %s: %d.", 2801 sc->sc_name, g_mirror_get_diskname(disk), error); 2802 break; 2803 } 2804 DISK_STATE_CHANGED(); 2805 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2806 sc->sc_name, g_mirror_get_diskname(disk)); 2807 2808 g_mirror_destroy_disk(disk); 2809 sc->sc_ndisks--; 2810 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2811 g_mirror_update_metadata(disk); 2812 } 2813 break; 2814 } 2815 default: 2816 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2817 break; 2818 } 2819 return (0); 2820 } 2821 #undef DISK_STATE_CHANGED 2822 2823 int 2824 g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2825 { 2826 struct g_provider *pp; 2827 u_char *buf; 2828 int error; 2829 2830 g_topology_assert(); 2831 2832 error = g_access(cp, 1, 0, 0); 2833 if (error != 0) 2834 return (error); 2835 pp = cp->provider; 2836 g_topology_unlock(); 2837 /* Metadata are stored on last sector. */ 2838 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2839 &error); 2840 g_topology_lock(); 2841 g_access(cp, -1, 0, 0); 2842 if (buf == NULL) { 2843 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2844 cp->provider->name, error); 2845 return (error); 2846 } 2847 2848 /* Decode metadata. */ 2849 error = mirror_metadata_decode(buf, md); 2850 g_free(buf); 2851 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2852 return (EINVAL); 2853 if (md->md_version > G_MIRROR_VERSION) { 2854 G_MIRROR_DEBUG(0, 2855 "Kernel module is too old to handle metadata from %s.", 2856 cp->provider->name); 2857 return (EINVAL); 2858 } 2859 if (error != 0) { 2860 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2861 cp->provider->name); 2862 return (error); 2863 } 2864 2865 return (0); 2866 } 2867 2868 static int 2869 g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2870 struct g_mirror_metadata *md) 2871 { 2872 2873 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2874 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2875 pp->name, md->md_did); 2876 return (EEXIST); 2877 } 2878 if (md->md_all != sc->sc_ndisks) { 2879 G_MIRROR_DEBUG(1, 2880 "Invalid '%s' field on disk %s (device %s), skipping.", 2881 "md_all", pp->name, sc->sc_name); 2882 return (EINVAL); 2883 } 2884 if (md->md_slice != sc->sc_slice) { 2885 G_MIRROR_DEBUG(1, 2886 "Invalid '%s' field on disk %s (device %s), skipping.", 2887 "md_slice", pp->name, sc->sc_name); 2888 return (EINVAL); 2889 } 2890 if (md->md_balance != sc->sc_balance) { 2891 G_MIRROR_DEBUG(1, 2892 "Invalid '%s' field on disk %s (device %s), skipping.", 2893 "md_balance", pp->name, sc->sc_name); 2894 return (EINVAL); 2895 } 2896 #if 0 2897 if (md->md_mediasize != sc->sc_mediasize) { 2898 G_MIRROR_DEBUG(1, 2899 "Invalid '%s' field on disk %s (device %s), skipping.", 2900 "md_mediasize", pp->name, sc->sc_name); 2901 return (EINVAL); 2902 } 2903 #endif 2904 if (sc->sc_mediasize > pp->mediasize) { 2905 G_MIRROR_DEBUG(1, 2906 "Invalid size of disk %s (device %s), skipping.", pp->name, 2907 sc->sc_name); 2908 return (EINVAL); 2909 } 2910 if (md->md_sectorsize != sc->sc_sectorsize) { 2911 G_MIRROR_DEBUG(1, 2912 "Invalid '%s' field on disk %s (device %s), skipping.", 2913 "md_sectorsize", pp->name, sc->sc_name); 2914 return (EINVAL); 2915 } 2916 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2917 G_MIRROR_DEBUG(1, 2918 "Invalid sector size of disk %s (device %s), skipping.", 2919 pp->name, sc->sc_name); 2920 return (EINVAL); 2921 } 2922 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2923 G_MIRROR_DEBUG(1, 2924 "Invalid device flags on disk %s (device %s), skipping.", 2925 pp->name, sc->sc_name); 2926 return (EINVAL); 2927 } 2928 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2929 G_MIRROR_DEBUG(1, 2930 "Invalid disk flags on disk %s (device %s), skipping.", 2931 pp->name, sc->sc_name); 2932 return (EINVAL); 2933 } 2934 return (0); 2935 } 2936 2937 int 2938 g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2939 struct g_mirror_metadata *md) 2940 { 2941 struct g_mirror_disk *disk; 2942 int error; 2943 2944 g_topology_assert_not(); 2945 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2946 2947 error = g_mirror_check_metadata(sc, pp, md); 2948 if (error != 0) 2949 return (error); 2950 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2951 md->md_genid < sc->sc_genid) { 2952 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2953 pp->name, sc->sc_name); 2954 return (EINVAL); 2955 } 2956 disk = g_mirror_init_disk(sc, pp, md, &error); 2957 if (disk == NULL) 2958 return (error); 2959 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2960 G_MIRROR_EVENT_WAIT); 2961 if (error != 0) 2962 return (error); 2963 if (md->md_version < G_MIRROR_VERSION) { 2964 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2965 pp->name, md->md_version, G_MIRROR_VERSION); 2966 g_mirror_update_metadata(disk); 2967 } 2968 return (0); 2969 } 2970 2971 static void 2972 g_mirror_destroy_delayed(void *arg, int flag) 2973 { 2974 struct g_mirror_softc *sc; 2975 int error; 2976 2977 if (flag == EV_CANCEL) { 2978 G_MIRROR_DEBUG(1, "Destroying canceled."); 2979 return; 2980 } 2981 sc = arg; 2982 g_topology_unlock(); 2983 sx_xlock(&sc->sc_lock); 2984 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2985 ("DESTROY flag set on %s.", sc->sc_name)); 2986 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0, 2987 ("CLOSEWAIT flag not set on %s.", sc->sc_name)); 2988 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 2989 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 2990 if (error != 0) { 2991 G_MIRROR_DEBUG(0, "Cannot destroy %s (error=%d).", 2992 sc->sc_name, error); 2993 sx_xunlock(&sc->sc_lock); 2994 } 2995 g_topology_lock(); 2996 } 2997 2998 static int 2999 g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 3000 { 3001 struct g_mirror_softc *sc; 3002 int error = 0; 3003 3004 g_topology_assert(); 3005 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 3006 acw, ace); 3007 3008 sc = pp->private; 3009 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 3010 3011 g_topology_unlock(); 3012 sx_xlock(&sc->sc_lock); 3013 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 3014 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 || 3015 LIST_EMPTY(&sc->sc_disks)) { 3016 if (acr > 0 || acw > 0 || ace > 0) 3017 error = ENXIO; 3018 goto end; 3019 } 3020 sc->sc_provider_open += acr + acw + ace; 3021 if (pp->acw + acw == 0) 3022 g_mirror_idle(sc, 0); 3023 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 && 3024 sc->sc_provider_open == 0) 3025 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, sc, NULL); 3026 end: 3027 sx_xunlock(&sc->sc_lock); 3028 g_topology_lock(); 3029 return (error); 3030 } 3031 3032 struct g_geom * 3033 g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md, 3034 u_int type) 3035 { 3036 struct g_mirror_softc *sc; 3037 struct g_geom *gp; 3038 int error, timeout; 3039 3040 g_topology_assert(); 3041 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 3042 md->md_mid); 3043 3044 /* One disk is minimum. */ 3045 if (md->md_all < 1) 3046 return (NULL); 3047 /* 3048 * Action geom. 3049 */ 3050 gp = g_new_geomf(mp, "%s", md->md_name); 3051 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 3052 gp->start = g_mirror_start; 3053 gp->orphan = g_mirror_orphan; 3054 gp->access = g_mirror_access; 3055 gp->dumpconf = g_mirror_dumpconf; 3056 3057 sc->sc_type = type; 3058 sc->sc_id = md->md_mid; 3059 sc->sc_slice = md->md_slice; 3060 sc->sc_balance = md->md_balance; 3061 sc->sc_mediasize = md->md_mediasize; 3062 sc->sc_sectorsize = md->md_sectorsize; 3063 sc->sc_ndisks = md->md_all; 3064 sc->sc_flags = md->md_mflags; 3065 sc->sc_bump_id = 0; 3066 sc->sc_idle = 1; 3067 sc->sc_last_write = time_uptime; 3068 sc->sc_writes = 0; 3069 sc->sc_refcnt = 1; 3070 sx_init(&sc->sc_lock, "gmirror:lock"); 3071 TAILQ_INIT(&sc->sc_queue); 3072 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 3073 TAILQ_INIT(&sc->sc_regular_delayed); 3074 TAILQ_INIT(&sc->sc_inflight); 3075 TAILQ_INIT(&sc->sc_sync_delayed); 3076 LIST_INIT(&sc->sc_disks); 3077 TAILQ_INIT(&sc->sc_events); 3078 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 3079 callout_init(&sc->sc_callout, 1); 3080 mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF); 3081 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 3082 gp->softc = sc; 3083 sc->sc_geom = gp; 3084 sc->sc_provider = NULL; 3085 sc->sc_provider_open = 0; 3086 /* 3087 * Synchronization geom. 3088 */ 3089 gp = g_new_geomf(mp, "%s.sync", md->md_name); 3090 gp->softc = sc; 3091 gp->orphan = g_mirror_orphan; 3092 sc->sc_sync.ds_geom = gp; 3093 sc->sc_sync.ds_ndisks = 0; 3094 error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 3095 "g_mirror %s", md->md_name); 3096 if (error != 0) { 3097 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 3098 sc->sc_name); 3099 g_destroy_geom(sc->sc_sync.ds_geom); 3100 g_destroy_geom(sc->sc_geom); 3101 g_mirror_free_device(sc); 3102 return (NULL); 3103 } 3104 3105 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 3106 sc->sc_name, sc->sc_ndisks, sc->sc_id); 3107 3108 sc->sc_rootmount = root_mount_hold("GMIRROR"); 3109 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 3110 /* 3111 * Run timeout. 3112 */ 3113 timeout = g_mirror_timeout * hz; 3114 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 3115 return (sc->sc_geom); 3116 } 3117 3118 int 3119 g_mirror_destroy(struct g_mirror_softc *sc, int how) 3120 { 3121 struct g_mirror_disk *disk; 3122 3123 g_topology_assert_not(); 3124 sx_assert(&sc->sc_lock, SX_XLOCKED); 3125 3126 if (sc->sc_provider_open != 0) { 3127 switch (how) { 3128 case G_MIRROR_DESTROY_SOFT: 3129 G_MIRROR_DEBUG(1, 3130 "Device %s is still open (%d).", sc->sc_name, 3131 sc->sc_provider_open); 3132 return (EBUSY); 3133 case G_MIRROR_DESTROY_DELAYED: 3134 G_MIRROR_DEBUG(1, 3135 "Device %s will be destroyed on last close.", 3136 sc->sc_name); 3137 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 3138 if (disk->d_state == 3139 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3140 g_mirror_sync_stop(disk, 1); 3141 } 3142 } 3143 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_CLOSEWAIT; 3144 return (EBUSY); 3145 case G_MIRROR_DESTROY_HARD: 3146 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 3147 "can't be definitely removed.", sc->sc_name); 3148 } 3149 } 3150 3151 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3152 sx_xunlock(&sc->sc_lock); 3153 return (0); 3154 } 3155 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 3156 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DRAIN; 3157 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3158 sx_xunlock(&sc->sc_lock); 3159 mtx_lock(&sc->sc_queue_mtx); 3160 wakeup(sc); 3161 mtx_unlock(&sc->sc_queue_mtx); 3162 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3163 while (sc->sc_worker != NULL) 3164 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 3165 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3166 sx_xlock(&sc->sc_lock); 3167 g_mirror_destroy_device(sc); 3168 return (0); 3169 } 3170 3171 static void 3172 g_mirror_taste_orphan(struct g_consumer *cp) 3173 { 3174 3175 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3176 cp->provider->name)); 3177 } 3178 3179 static struct g_geom * 3180 g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3181 { 3182 struct g_mirror_metadata md; 3183 struct g_mirror_softc *sc; 3184 struct g_consumer *cp; 3185 struct g_geom *gp; 3186 int error; 3187 3188 g_topology_assert(); 3189 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3190 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 3191 3192 gp = g_new_geomf(mp, "mirror:taste"); 3193 /* 3194 * This orphan function should be never called. 3195 */ 3196 gp->orphan = g_mirror_taste_orphan; 3197 cp = g_new_consumer(gp); 3198 g_attach(cp, pp); 3199 error = g_mirror_read_metadata(cp, &md); 3200 g_detach(cp); 3201 g_destroy_consumer(cp); 3202 g_destroy_geom(gp); 3203 if (error != 0) 3204 return (NULL); 3205 gp = NULL; 3206 3207 if (md.md_provider[0] != '\0' && 3208 !g_compare_names(md.md_provider, pp->name)) 3209 return (NULL); 3210 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3211 return (NULL); 3212 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 3213 G_MIRROR_DEBUG(0, 3214 "Device %s: provider %s marked as inactive, skipping.", 3215 md.md_name, pp->name); 3216 return (NULL); 3217 } 3218 if (g_mirror_debug >= 2) 3219 mirror_metadata_dump(&md); 3220 3221 /* 3222 * Let's check if device already exists. 3223 */ 3224 sc = NULL; 3225 LIST_FOREACH(gp, &mp->geom, geom) { 3226 sc = gp->softc; 3227 if (sc == NULL) 3228 continue; 3229 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 3230 continue; 3231 if (sc->sc_sync.ds_geom == gp) 3232 continue; 3233 if (strcmp(md.md_name, sc->sc_name) != 0) 3234 continue; 3235 if (md.md_mid != sc->sc_id) { 3236 G_MIRROR_DEBUG(0, "Device %s already configured.", 3237 sc->sc_name); 3238 return (NULL); 3239 } 3240 break; 3241 } 3242 if (gp == NULL) { 3243 gp = g_mirror_create(mp, &md, G_MIRROR_TYPE_AUTOMATIC); 3244 if (gp == NULL) { 3245 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3246 md.md_name); 3247 return (NULL); 3248 } 3249 sc = gp->softc; 3250 } 3251 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3252 g_topology_unlock(); 3253 sx_xlock(&sc->sc_lock); 3254 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; 3255 error = g_mirror_add_disk(sc, pp, &md); 3256 if (error != 0) { 3257 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3258 pp->name, gp->name, error); 3259 if (LIST_EMPTY(&sc->sc_disks)) { 3260 g_cancel_event(sc); 3261 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3262 g_topology_lock(); 3263 return (NULL); 3264 } 3265 gp = NULL; 3266 } 3267 sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; 3268 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3269 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3270 g_topology_lock(); 3271 return (NULL); 3272 } 3273 sx_xunlock(&sc->sc_lock); 3274 g_topology_lock(); 3275 return (gp); 3276 } 3277 3278 static void 3279 g_mirror_resize(struct g_consumer *cp) 3280 { 3281 struct g_mirror_disk *disk; 3282 3283 g_topology_assert(); 3284 g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name); 3285 3286 disk = cp->private; 3287 if (disk == NULL) 3288 return; 3289 g_topology_unlock(); 3290 g_mirror_update_metadata(disk); 3291 g_topology_lock(); 3292 } 3293 3294 static int 3295 g_mirror_destroy_geom(struct gctl_req *req __unused, 3296 struct g_class *mp __unused, struct g_geom *gp) 3297 { 3298 struct g_mirror_softc *sc; 3299 int error; 3300 3301 g_topology_unlock(); 3302 sc = gp->softc; 3303 sx_xlock(&sc->sc_lock); 3304 g_cancel_event(sc); 3305 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3306 if (error != 0) 3307 sx_xunlock(&sc->sc_lock); 3308 g_topology_lock(); 3309 return (error); 3310 } 3311 3312 static void 3313 g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3314 struct g_consumer *cp, struct g_provider *pp) 3315 { 3316 struct g_mirror_softc *sc; 3317 3318 g_topology_assert(); 3319 3320 sc = gp->softc; 3321 if (sc == NULL) 3322 return; 3323 /* Skip synchronization geom. */ 3324 if (gp == sc->sc_sync.ds_geom) 3325 return; 3326 if (pp != NULL) { 3327 /* Nothing here. */ 3328 } else if (cp != NULL) { 3329 struct g_mirror_disk *disk; 3330 3331 disk = cp->private; 3332 if (disk == NULL) 3333 return; 3334 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3335 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3336 sbuf_printf(sb, "%s<Synchronized>", indent); 3337 if (disk->d_sync.ds_offset == 0) 3338 sbuf_printf(sb, "0%%"); 3339 else 3340 sbuf_printf(sb, "%u%%", 3341 (u_int)((disk->d_sync.ds_offset * 100) / 3342 sc->sc_mediasize)); 3343 sbuf_printf(sb, "</Synchronized>\n"); 3344 if (disk->d_sync.ds_offset > 0) 3345 sbuf_printf(sb, "%s<BytesSynced>%jd" 3346 "</BytesSynced>\n", indent, 3347 (intmax_t)disk->d_sync.ds_offset); 3348 } 3349 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3350 disk->d_sync.ds_syncid); 3351 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3352 disk->d_genid); 3353 sbuf_printf(sb, "%s<Flags>", indent); 3354 if (disk->d_flags == 0) 3355 sbuf_printf(sb, "NONE"); 3356 else { 3357 int first = 1; 3358 3359 #define ADD_FLAG(flag, name) do { \ 3360 if ((disk->d_flags & (flag)) != 0) { \ 3361 if (!first) \ 3362 sbuf_printf(sb, ", "); \ 3363 else \ 3364 first = 0; \ 3365 sbuf_printf(sb, name); \ 3366 } \ 3367 } while (0) 3368 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3369 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3370 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3371 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3372 "SYNCHRONIZING"); 3373 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3374 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3375 #undef ADD_FLAG 3376 } 3377 sbuf_printf(sb, "</Flags>\n"); 3378 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3379 disk->d_priority); 3380 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3381 g_mirror_disk_state2str(disk->d_state)); 3382 } else { 3383 sbuf_printf(sb, "%s<Type>", indent); 3384 switch (sc->sc_type) { 3385 case G_MIRROR_TYPE_AUTOMATIC: 3386 sbuf_printf(sb, "AUTOMATIC"); 3387 break; 3388 case G_MIRROR_TYPE_MANUAL: 3389 sbuf_printf(sb, "MANUAL"); 3390 break; 3391 default: 3392 sbuf_printf(sb, "UNKNOWN"); 3393 break; 3394 } 3395 sbuf_printf(sb, "</Type>\n"); 3396 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3397 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3398 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3399 sbuf_printf(sb, "%s<Flags>", indent); 3400 if (sc->sc_flags == 0) 3401 sbuf_printf(sb, "NONE"); 3402 else { 3403 int first = 1; 3404 3405 #define ADD_FLAG(flag, name) do { \ 3406 if ((sc->sc_flags & (flag)) != 0) { \ 3407 if (!first) \ 3408 sbuf_printf(sb, ", "); \ 3409 else \ 3410 first = 0; \ 3411 sbuf_printf(sb, name); \ 3412 } \ 3413 } while (0) 3414 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3415 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3416 #undef ADD_FLAG 3417 } 3418 sbuf_printf(sb, "</Flags>\n"); 3419 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3420 (u_int)sc->sc_slice); 3421 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3422 balance_name(sc->sc_balance)); 3423 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3424 sc->sc_ndisks); 3425 sbuf_printf(sb, "%s<State>", indent); 3426 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3427 sbuf_printf(sb, "%s", "STARTING"); 3428 else if (sc->sc_ndisks == 3429 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3430 sbuf_printf(sb, "%s", "COMPLETE"); 3431 else 3432 sbuf_printf(sb, "%s", "DEGRADED"); 3433 sbuf_printf(sb, "</State>\n"); 3434 } 3435 } 3436 3437 static void 3438 g_mirror_shutdown_post_sync(void *arg, int howto) 3439 { 3440 struct g_class *mp; 3441 struct g_geom *gp, *gp2; 3442 struct g_mirror_softc *sc; 3443 int error; 3444 3445 if (panicstr != NULL) 3446 return; 3447 3448 mp = arg; 3449 g_topology_lock(); 3450 g_mirror_shutdown = 1; 3451 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3452 if ((sc = gp->softc) == NULL) 3453 continue; 3454 /* Skip synchronization geom. */ 3455 if (gp == sc->sc_sync.ds_geom) 3456 continue; 3457 g_topology_unlock(); 3458 sx_xlock(&sc->sc_lock); 3459 g_mirror_idle(sc, -1); 3460 g_cancel_event(sc); 3461 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3462 if (error != 0) 3463 sx_xunlock(&sc->sc_lock); 3464 g_topology_lock(); 3465 } 3466 g_topology_unlock(); 3467 } 3468 3469 static void 3470 g_mirror_init(struct g_class *mp) 3471 { 3472 3473 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3474 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3475 if (g_mirror_post_sync == NULL) 3476 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3477 } 3478 3479 static void 3480 g_mirror_fini(struct g_class *mp) 3481 { 3482 3483 if (g_mirror_post_sync != NULL) 3484 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); 3485 } 3486 3487 DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3488 MODULE_VERSION(geom_mirror, 0); 3489