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 sx_assert(&sc->sc_lock, SX_XLOCKED); 211 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 212 sx_xunlock(&sc->sc_lock); 213 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 214 mtx_lock(&sc->sc_events_mtx); 215 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 216 hz * 5); 217 } 218 error = ep->e_error; 219 g_mirror_event_free(ep); 220 sx_xlock(&sc->sc_lock); 221 return (error); 222 } 223 224 static struct g_mirror_event * 225 g_mirror_event_first(struct g_mirror_softc *sc) 226 { 227 struct g_mirror_event *ep; 228 229 mtx_lock(&sc->sc_events_mtx); 230 ep = TAILQ_FIRST(&sc->sc_events); 231 mtx_unlock(&sc->sc_events_mtx); 232 return (ep); 233 } 234 235 static void 236 g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 237 { 238 239 mtx_lock(&sc->sc_events_mtx); 240 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 241 mtx_unlock(&sc->sc_events_mtx); 242 } 243 244 static void 245 g_mirror_event_cancel(struct g_mirror_disk *disk) 246 { 247 struct g_mirror_softc *sc; 248 struct g_mirror_event *ep, *tmpep; 249 250 sc = disk->d_softc; 251 sx_assert(&sc->sc_lock, SX_XLOCKED); 252 253 mtx_lock(&sc->sc_events_mtx); 254 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 255 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 256 continue; 257 if (ep->e_disk != disk) 258 continue; 259 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 260 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 261 g_mirror_event_free(ep); 262 else { 263 ep->e_error = ECANCELED; 264 wakeup(ep); 265 } 266 } 267 mtx_unlock(&sc->sc_events_mtx); 268 } 269 270 /* 271 * Return the number of disks in given state. 272 * If state is equal to -1, count all connected disks. 273 */ 274 u_int 275 g_mirror_ndisks(struct g_mirror_softc *sc, int state) 276 { 277 struct g_mirror_disk *disk; 278 u_int n = 0; 279 280 sx_assert(&sc->sc_lock, SX_LOCKED); 281 282 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 283 if (state == -1 || disk->d_state == state) 284 n++; 285 } 286 return (n); 287 } 288 289 /* 290 * Find a disk in mirror by its disk ID. 291 */ 292 static struct g_mirror_disk * 293 g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 294 { 295 struct g_mirror_disk *disk; 296 297 sx_assert(&sc->sc_lock, SX_XLOCKED); 298 299 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 300 if (disk->d_id == id) 301 return (disk); 302 } 303 return (NULL); 304 } 305 306 static u_int 307 g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 308 { 309 struct bio *bp; 310 u_int nreqs = 0; 311 312 mtx_lock(&sc->sc_queue_mtx); 313 TAILQ_FOREACH(bp, &sc->sc_queue, bio_queue) { 314 if (bp->bio_from == cp) 315 nreqs++; 316 } 317 mtx_unlock(&sc->sc_queue_mtx); 318 return (nreqs); 319 } 320 321 static int 322 g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 323 { 324 325 if (cp->index > 0) { 326 G_MIRROR_DEBUG(2, 327 "I/O requests for %s exist, can't destroy it now.", 328 cp->provider->name); 329 return (1); 330 } 331 if (g_mirror_nrequests(sc, cp) > 0) { 332 G_MIRROR_DEBUG(2, 333 "I/O requests for %s in queue, can't destroy it now.", 334 cp->provider->name); 335 return (1); 336 } 337 return (0); 338 } 339 340 static void 341 g_mirror_destroy_consumer(void *arg, int flags __unused) 342 { 343 struct g_consumer *cp; 344 345 g_topology_assert(); 346 347 cp = arg; 348 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 349 g_detach(cp); 350 g_destroy_consumer(cp); 351 } 352 353 static void 354 g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 355 { 356 struct g_provider *pp; 357 int retaste_wait; 358 359 g_topology_assert(); 360 361 cp->private = NULL; 362 if (g_mirror_is_busy(sc, cp)) 363 return; 364 pp = cp->provider; 365 retaste_wait = 0; 366 if (cp->acw == 1) { 367 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 368 retaste_wait = 1; 369 } 370 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 371 -cp->acw, -cp->ace, 0); 372 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 373 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 374 if (retaste_wait) { 375 /* 376 * After retaste event was send (inside g_access()), we can send 377 * event to detach and destroy consumer. 378 * A class, which has consumer to the given provider connected 379 * will not receive retaste event for the provider. 380 * This is the way how I ignore retaste events when I close 381 * consumers opened for write: I detach and destroy consumer 382 * after retaste event is sent. 383 */ 384 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 385 return; 386 } 387 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 388 g_detach(cp); 389 g_destroy_consumer(cp); 390 } 391 392 static int 393 g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 394 { 395 struct g_consumer *cp; 396 int error; 397 398 g_topology_assert_not(); 399 KASSERT(disk->d_consumer == NULL, 400 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 401 402 g_topology_lock(); 403 cp = g_new_consumer(disk->d_softc->sc_geom); 404 cp->flags |= G_CF_DIRECT_RECEIVE; 405 error = g_attach(cp, pp); 406 if (error != 0) { 407 g_destroy_consumer(cp); 408 g_topology_unlock(); 409 return (error); 410 } 411 error = g_access(cp, 1, 1, 1); 412 if (error != 0) { 413 g_detach(cp); 414 g_destroy_consumer(cp); 415 g_topology_unlock(); 416 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 417 pp->name, error); 418 return (error); 419 } 420 g_topology_unlock(); 421 disk->d_consumer = cp; 422 disk->d_consumer->private = disk; 423 disk->d_consumer->index = 0; 424 425 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 426 return (0); 427 } 428 429 static void 430 g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 431 { 432 433 g_topology_assert(); 434 435 if (cp == NULL) 436 return; 437 if (cp->provider != NULL) 438 g_mirror_kill_consumer(sc, cp); 439 else 440 g_destroy_consumer(cp); 441 } 442 443 /* 444 * Initialize disk. This means allocate memory, create consumer, attach it 445 * to the provider and open access (r1w1e1) to it. 446 */ 447 static struct g_mirror_disk * 448 g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 449 struct g_mirror_metadata *md, int *errorp) 450 { 451 struct g_mirror_disk *disk; 452 int i, error; 453 454 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 455 if (disk == NULL) { 456 error = ENOMEM; 457 goto fail; 458 } 459 disk->d_softc = sc; 460 error = g_mirror_connect_disk(disk, pp); 461 if (error != 0) 462 goto fail; 463 disk->d_id = md->md_did; 464 disk->d_state = G_MIRROR_DISK_STATE_NONE; 465 disk->d_priority = md->md_priority; 466 disk->d_flags = md->md_dflags; 467 error = g_getattr("GEOM::candelete", disk->d_consumer, &i); 468 if (error == 0 && i != 0) 469 disk->d_flags |= G_MIRROR_DISK_FLAG_CANDELETE; 470 if (md->md_provider[0] != '\0') 471 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 472 disk->d_sync.ds_consumer = NULL; 473 disk->d_sync.ds_offset = md->md_sync_offset; 474 disk->d_sync.ds_offset_done = md->md_sync_offset; 475 disk->d_sync.ds_update_ts = time_uptime; 476 disk->d_genid = md->md_genid; 477 disk->d_sync.ds_syncid = md->md_syncid; 478 if (errorp != NULL) 479 *errorp = 0; 480 return (disk); 481 fail: 482 if (errorp != NULL) 483 *errorp = error; 484 if (disk != NULL) 485 free(disk, M_MIRROR); 486 return (NULL); 487 } 488 489 static void 490 g_mirror_destroy_disk(struct g_mirror_disk *disk) 491 { 492 struct g_mirror_softc *sc; 493 494 g_topology_assert_not(); 495 sc = disk->d_softc; 496 sx_assert(&sc->sc_lock, SX_XLOCKED); 497 498 LIST_REMOVE(disk, d_next); 499 g_mirror_event_cancel(disk); 500 if (sc->sc_hint == disk) 501 sc->sc_hint = NULL; 502 switch (disk->d_state) { 503 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 504 g_mirror_sync_stop(disk, 1); 505 /* FALLTHROUGH */ 506 case G_MIRROR_DISK_STATE_NEW: 507 case G_MIRROR_DISK_STATE_STALE: 508 case G_MIRROR_DISK_STATE_ACTIVE: 509 g_topology_lock(); 510 g_mirror_disconnect_consumer(sc, disk->d_consumer); 511 g_topology_unlock(); 512 free(disk, M_MIRROR); 513 break; 514 default: 515 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 516 g_mirror_get_diskname(disk), 517 g_mirror_disk_state2str(disk->d_state))); 518 } 519 } 520 521 static void 522 g_mirror_free_device(struct g_mirror_softc *sc) 523 { 524 525 mtx_destroy(&sc->sc_queue_mtx); 526 mtx_destroy(&sc->sc_events_mtx); 527 mtx_destroy(&sc->sc_done_mtx); 528 sx_destroy(&sc->sc_lock); 529 free(sc, M_MIRROR); 530 } 531 532 static void 533 g_mirror_providergone(struct g_provider *pp) 534 { 535 struct g_mirror_softc *sc = pp->private; 536 537 if ((--sc->sc_refcnt) == 0) 538 g_mirror_free_device(sc); 539 } 540 541 static void 542 g_mirror_destroy_device(struct g_mirror_softc *sc) 543 { 544 struct g_mirror_disk *disk; 545 struct g_mirror_event *ep; 546 struct g_geom *gp; 547 struct g_consumer *cp, *tmpcp; 548 549 g_topology_assert_not(); 550 sx_assert(&sc->sc_lock, SX_XLOCKED); 551 552 gp = sc->sc_geom; 553 if (sc->sc_provider != NULL) 554 g_mirror_destroy_provider(sc); 555 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 556 disk = LIST_FIRST(&sc->sc_disks)) { 557 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 558 g_mirror_update_metadata(disk); 559 g_mirror_destroy_disk(disk); 560 } 561 while ((ep = g_mirror_event_first(sc)) != NULL) { 562 g_mirror_event_remove(sc, ep); 563 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 564 g_mirror_event_free(ep); 565 else { 566 ep->e_error = ECANCELED; 567 ep->e_flags |= G_MIRROR_EVENT_DONE; 568 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 569 mtx_lock(&sc->sc_events_mtx); 570 wakeup(ep); 571 mtx_unlock(&sc->sc_events_mtx); 572 } 573 } 574 callout_drain(&sc->sc_callout); 575 576 g_topology_lock(); 577 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 578 g_mirror_disconnect_consumer(sc, cp); 579 } 580 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 581 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 582 g_wither_geom(gp, ENXIO); 583 sx_xunlock(&sc->sc_lock); 584 if ((--sc->sc_refcnt) == 0) 585 g_mirror_free_device(sc); 586 g_topology_unlock(); 587 } 588 589 static void 590 g_mirror_orphan(struct g_consumer *cp) 591 { 592 struct g_mirror_disk *disk; 593 594 g_topology_assert(); 595 596 disk = cp->private; 597 if (disk == NULL) 598 return; 599 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 600 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 601 G_MIRROR_EVENT_DONTWAIT); 602 } 603 604 /* 605 * Function should return the next active disk on the list. 606 * It is possible that it will be the same disk as given. 607 * If there are no active disks on list, NULL is returned. 608 */ 609 static __inline struct g_mirror_disk * 610 g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 611 { 612 struct g_mirror_disk *dp; 613 614 for (dp = LIST_NEXT(disk, d_next); dp != disk; 615 dp = LIST_NEXT(dp, d_next)) { 616 if (dp == NULL) 617 dp = LIST_FIRST(&sc->sc_disks); 618 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 619 break; 620 } 621 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 622 return (NULL); 623 return (dp); 624 } 625 626 static struct g_mirror_disk * 627 g_mirror_get_disk(struct g_mirror_softc *sc) 628 { 629 struct g_mirror_disk *disk; 630 631 if (sc->sc_hint == NULL) { 632 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 633 if (sc->sc_hint == NULL) 634 return (NULL); 635 } 636 disk = sc->sc_hint; 637 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 638 disk = g_mirror_find_next(sc, disk); 639 if (disk == NULL) 640 return (NULL); 641 } 642 sc->sc_hint = g_mirror_find_next(sc, disk); 643 return (disk); 644 } 645 646 static int 647 g_mirror_write_metadata(struct g_mirror_disk *disk, 648 struct g_mirror_metadata *md) 649 { 650 struct g_mirror_softc *sc; 651 struct g_consumer *cp; 652 off_t offset, length; 653 u_char *sector; 654 int error = 0; 655 656 g_topology_assert_not(); 657 sc = disk->d_softc; 658 sx_assert(&sc->sc_lock, SX_LOCKED); 659 660 cp = disk->d_consumer; 661 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 662 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 663 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 664 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 665 cp->acw, cp->ace)); 666 length = cp->provider->sectorsize; 667 offset = cp->provider->mediasize - length; 668 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 669 if (md != NULL && 670 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) { 671 /* 672 * Handle the case, when the size of parent provider reduced. 673 */ 674 if (offset < md->md_mediasize) 675 error = ENOSPC; 676 else 677 mirror_metadata_encode(md, sector); 678 } 679 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_metadata_write, error); 680 if (error == 0) 681 error = g_write_data(cp, offset, sector, length); 682 free(sector, M_MIRROR); 683 if (error != 0) { 684 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 685 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 686 G_MIRROR_DEBUG(0, "Cannot write metadata on %s " 687 "(device=%s, error=%d).", 688 g_mirror_get_diskname(disk), sc->sc_name, error); 689 } else { 690 G_MIRROR_DEBUG(1, "Cannot write metadata on %s " 691 "(device=%s, error=%d).", 692 g_mirror_get_diskname(disk), sc->sc_name, error); 693 } 694 if (g_mirror_disconnect_on_failure && 695 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 696 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 697 g_mirror_event_send(disk, 698 G_MIRROR_DISK_STATE_DISCONNECTED, 699 G_MIRROR_EVENT_DONTWAIT); 700 } 701 } 702 return (error); 703 } 704 705 static int 706 g_mirror_clear_metadata(struct g_mirror_disk *disk) 707 { 708 int error; 709 710 g_topology_assert_not(); 711 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 712 713 if (disk->d_softc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 714 return (0); 715 error = g_mirror_write_metadata(disk, NULL); 716 if (error == 0) { 717 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 718 g_mirror_get_diskname(disk)); 719 } else { 720 G_MIRROR_DEBUG(0, 721 "Cannot clear metadata on disk %s (error=%d).", 722 g_mirror_get_diskname(disk), error); 723 } 724 return (error); 725 } 726 727 void 728 g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 729 struct g_mirror_metadata *md) 730 { 731 732 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 733 md->md_version = G_MIRROR_VERSION; 734 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 735 md->md_mid = sc->sc_id; 736 md->md_all = sc->sc_ndisks; 737 md->md_slice = sc->sc_slice; 738 md->md_balance = sc->sc_balance; 739 md->md_genid = sc->sc_genid; 740 md->md_mediasize = sc->sc_mediasize; 741 md->md_sectorsize = sc->sc_sectorsize; 742 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 743 bzero(md->md_provider, sizeof(md->md_provider)); 744 if (disk == NULL) { 745 md->md_did = arc4random(); 746 md->md_priority = 0; 747 md->md_syncid = 0; 748 md->md_dflags = 0; 749 md->md_sync_offset = 0; 750 md->md_provsize = 0; 751 } else { 752 md->md_did = disk->d_id; 753 md->md_priority = disk->d_priority; 754 md->md_syncid = disk->d_sync.ds_syncid; 755 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 756 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 757 md->md_sync_offset = disk->d_sync.ds_offset_done; 758 else 759 md->md_sync_offset = 0; 760 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 761 strlcpy(md->md_provider, 762 disk->d_consumer->provider->name, 763 sizeof(md->md_provider)); 764 } 765 md->md_provsize = disk->d_consumer->provider->mediasize; 766 } 767 } 768 769 void 770 g_mirror_update_metadata(struct g_mirror_disk *disk) 771 { 772 struct g_mirror_softc *sc; 773 struct g_mirror_metadata md; 774 int error; 775 776 g_topology_assert_not(); 777 sc = disk->d_softc; 778 sx_assert(&sc->sc_lock, SX_LOCKED); 779 780 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 781 return; 782 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) 783 g_mirror_fill_metadata(sc, disk, &md); 784 error = g_mirror_write_metadata(disk, &md); 785 if (error == 0) { 786 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 787 g_mirror_get_diskname(disk)); 788 } else { 789 G_MIRROR_DEBUG(0, 790 "Cannot update metadata on disk %s (error=%d).", 791 g_mirror_get_diskname(disk), error); 792 } 793 } 794 795 static void 796 g_mirror_bump_syncid(struct g_mirror_softc *sc) 797 { 798 struct g_mirror_disk *disk; 799 800 g_topology_assert_not(); 801 sx_assert(&sc->sc_lock, SX_XLOCKED); 802 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 803 ("%s called with no active disks (device=%s).", __func__, 804 sc->sc_name)); 805 806 sc->sc_syncid++; 807 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 808 sc->sc_syncid); 809 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 810 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 811 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 812 disk->d_sync.ds_syncid = sc->sc_syncid; 813 g_mirror_update_metadata(disk); 814 } 815 } 816 } 817 818 static void 819 g_mirror_bump_genid(struct g_mirror_softc *sc) 820 { 821 struct g_mirror_disk *disk; 822 823 g_topology_assert_not(); 824 sx_assert(&sc->sc_lock, SX_XLOCKED); 825 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 826 ("%s called with no active disks (device=%s).", __func__, 827 sc->sc_name)); 828 829 sc->sc_genid++; 830 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 831 sc->sc_genid); 832 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 833 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 834 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 835 disk->d_genid = sc->sc_genid; 836 g_mirror_update_metadata(disk); 837 } 838 } 839 } 840 841 static int 842 g_mirror_idle(struct g_mirror_softc *sc, int acw) 843 { 844 struct g_mirror_disk *disk; 845 int timeout; 846 847 g_topology_assert_not(); 848 sx_assert(&sc->sc_lock, SX_XLOCKED); 849 850 if (sc->sc_provider == NULL) 851 return (0); 852 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 853 return (0); 854 if (sc->sc_idle) 855 return (0); 856 if (sc->sc_writes > 0) 857 return (0); 858 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 859 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); 860 if (!g_mirror_shutdown && timeout > 0) 861 return (timeout); 862 } 863 sc->sc_idle = 1; 864 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 865 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 866 continue; 867 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 868 g_mirror_get_diskname(disk), sc->sc_name); 869 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 870 g_mirror_update_metadata(disk); 871 } 872 return (0); 873 } 874 875 static void 876 g_mirror_unidle(struct g_mirror_softc *sc) 877 { 878 struct g_mirror_disk *disk; 879 880 g_topology_assert_not(); 881 sx_assert(&sc->sc_lock, SX_XLOCKED); 882 883 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 884 return; 885 sc->sc_idle = 0; 886 sc->sc_last_write = time_uptime; 887 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 888 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 889 continue; 890 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 891 g_mirror_get_diskname(disk), sc->sc_name); 892 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 893 g_mirror_update_metadata(disk); 894 } 895 } 896 897 static void 898 g_mirror_done(struct bio *bp) 899 { 900 struct g_mirror_softc *sc; 901 902 sc = bp->bio_from->geom->softc; 903 bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR; 904 mtx_lock(&sc->sc_queue_mtx); 905 TAILQ_INSERT_TAIL(&sc->sc_queue, bp, bio_queue); 906 mtx_unlock(&sc->sc_queue_mtx); 907 wakeup(sc); 908 } 909 910 static void 911 g_mirror_regular_request_error(struct g_mirror_softc *sc, 912 struct g_mirror_disk *disk, struct bio *bp) 913 { 914 915 if (bp->bio_cmd == BIO_FLUSH && bp->bio_error == EOPNOTSUPP) 916 return; 917 918 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 919 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 920 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 921 bp->bio_error); 922 } else { 923 G_MIRROR_LOGREQ(1, bp, "Request failed (error=%d).", 924 bp->bio_error); 925 } 926 if (g_mirror_disconnect_on_failure && 927 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 928 if (bp->bio_error == ENXIO && 929 bp->bio_cmd == BIO_READ) 930 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 931 else if (bp->bio_error == ENXIO) 932 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_NOW; 933 else 934 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 935 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 936 G_MIRROR_EVENT_DONTWAIT); 937 } 938 } 939 940 static void 941 g_mirror_regular_request(struct g_mirror_softc *sc, struct bio *bp) 942 { 943 struct g_mirror_disk *disk; 944 struct bio *pbp; 945 946 g_topology_assert_not(); 947 KASSERT(sc->sc_provider == bp->bio_parent->bio_to, 948 ("regular request %p with unexpected origin", bp)); 949 950 pbp = bp->bio_parent; 951 bp->bio_from->index--; 952 if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) 953 sc->sc_writes--; 954 disk = bp->bio_from->private; 955 if (disk == NULL) { 956 g_topology_lock(); 957 g_mirror_kill_consumer(sc, bp->bio_from); 958 g_topology_unlock(); 959 } 960 961 switch (bp->bio_cmd) { 962 case BIO_READ: 963 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_read, 964 bp->bio_error); 965 break; 966 case BIO_WRITE: 967 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_write, 968 bp->bio_error); 969 break; 970 case BIO_DELETE: 971 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_delete, 972 bp->bio_error); 973 break; 974 case BIO_FLUSH: 975 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_flush, 976 bp->bio_error); 977 break; 978 } 979 980 pbp->bio_inbed++; 981 KASSERT(pbp->bio_inbed <= pbp->bio_children, 982 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 983 pbp->bio_children)); 984 if (bp->bio_error == 0 && pbp->bio_error == 0) { 985 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 986 g_destroy_bio(bp); 987 if (pbp->bio_children == pbp->bio_inbed) { 988 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 989 pbp->bio_completed = pbp->bio_length; 990 if (pbp->bio_cmd == BIO_WRITE || 991 pbp->bio_cmd == BIO_DELETE) { 992 TAILQ_REMOVE(&sc->sc_inflight, pbp, bio_queue); 993 /* Release delayed sync requests if possible. */ 994 g_mirror_sync_release(sc); 995 } 996 g_io_deliver(pbp, pbp->bio_error); 997 } 998 return; 999 } else if (bp->bio_error != 0) { 1000 if (pbp->bio_error == 0) 1001 pbp->bio_error = bp->bio_error; 1002 if (disk != NULL) 1003 g_mirror_regular_request_error(sc, disk, bp); 1004 switch (pbp->bio_cmd) { 1005 case BIO_DELETE: 1006 case BIO_WRITE: 1007 case BIO_FLUSH: 1008 pbp->bio_inbed--; 1009 pbp->bio_children--; 1010 break; 1011 } 1012 } 1013 g_destroy_bio(bp); 1014 1015 switch (pbp->bio_cmd) { 1016 case BIO_READ: 1017 if (pbp->bio_inbed < pbp->bio_children) 1018 break; 1019 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 1020 g_io_deliver(pbp, pbp->bio_error); 1021 else { 1022 pbp->bio_error = 0; 1023 mtx_lock(&sc->sc_queue_mtx); 1024 TAILQ_INSERT_TAIL(&sc->sc_queue, pbp, bio_queue); 1025 mtx_unlock(&sc->sc_queue_mtx); 1026 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1027 wakeup(sc); 1028 } 1029 break; 1030 case BIO_DELETE: 1031 case BIO_WRITE: 1032 case BIO_FLUSH: 1033 if (pbp->bio_children == 0) { 1034 /* 1035 * All requests failed. 1036 */ 1037 } else if (pbp->bio_inbed < pbp->bio_children) { 1038 /* Do nothing. */ 1039 break; 1040 } else if (pbp->bio_children == pbp->bio_inbed) { 1041 /* Some requests succeeded. */ 1042 pbp->bio_error = 0; 1043 pbp->bio_completed = pbp->bio_length; 1044 } 1045 if (pbp->bio_cmd == BIO_WRITE || pbp->bio_cmd == BIO_DELETE) { 1046 TAILQ_REMOVE(&sc->sc_inflight, pbp, bio_queue); 1047 /* Release delayed sync requests if possible. */ 1048 g_mirror_sync_release(sc); 1049 } 1050 g_io_deliver(pbp, pbp->bio_error); 1051 break; 1052 default: 1053 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 1054 break; 1055 } 1056 } 1057 1058 static void 1059 g_mirror_sync_done(struct bio *bp) 1060 { 1061 struct g_mirror_softc *sc; 1062 1063 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1064 sc = bp->bio_from->geom->softc; 1065 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; 1066 mtx_lock(&sc->sc_queue_mtx); 1067 TAILQ_INSERT_TAIL(&sc->sc_queue, bp, bio_queue); 1068 mtx_unlock(&sc->sc_queue_mtx); 1069 wakeup(sc); 1070 } 1071 1072 static void 1073 g_mirror_candelete(struct bio *bp) 1074 { 1075 struct g_mirror_softc *sc; 1076 struct g_mirror_disk *disk; 1077 int *val; 1078 1079 sc = bp->bio_to->private; 1080 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1081 if (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) 1082 break; 1083 } 1084 val = (int *)bp->bio_data; 1085 *val = (disk != NULL); 1086 g_io_deliver(bp, 0); 1087 } 1088 1089 static void 1090 g_mirror_kernel_dump(struct bio *bp) 1091 { 1092 struct g_mirror_softc *sc; 1093 struct g_mirror_disk *disk; 1094 struct bio *cbp; 1095 struct g_kerneldump *gkd; 1096 1097 /* 1098 * We configure dumping to the first component, because this component 1099 * will be used for reading with 'prefer' balance algorithm. 1100 * If the component with the highest priority is currently disconnected 1101 * we will not be able to read the dump after the reboot if it will be 1102 * connected and synchronized later. Can we do something better? 1103 */ 1104 sc = bp->bio_to->private; 1105 disk = LIST_FIRST(&sc->sc_disks); 1106 1107 gkd = (struct g_kerneldump *)bp->bio_data; 1108 if (gkd->length > bp->bio_to->mediasize) 1109 gkd->length = bp->bio_to->mediasize; 1110 cbp = g_clone_bio(bp); 1111 if (cbp == NULL) { 1112 g_io_deliver(bp, ENOMEM); 1113 return; 1114 } 1115 cbp->bio_done = g_std_done; 1116 g_io_request(cbp, disk->d_consumer); 1117 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1118 g_mirror_get_diskname(disk)); 1119 } 1120 1121 static void 1122 g_mirror_start(struct bio *bp) 1123 { 1124 struct g_mirror_softc *sc; 1125 1126 sc = bp->bio_to->private; 1127 /* 1128 * If sc == NULL or there are no valid disks, provider's error 1129 * should be set and g_mirror_start() should not be called at all. 1130 */ 1131 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1132 ("Provider's error should be set (error=%d)(mirror=%s).", 1133 bp->bio_to->error, bp->bio_to->name)); 1134 G_MIRROR_LOGREQ(3, bp, "Request received."); 1135 1136 switch (bp->bio_cmd) { 1137 case BIO_READ: 1138 case BIO_WRITE: 1139 case BIO_DELETE: 1140 case BIO_FLUSH: 1141 break; 1142 case BIO_GETATTR: 1143 if (!strcmp(bp->bio_attribute, "GEOM::candelete")) { 1144 g_mirror_candelete(bp); 1145 return; 1146 } else if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1147 g_mirror_kernel_dump(bp); 1148 return; 1149 } 1150 /* FALLTHROUGH */ 1151 default: 1152 g_io_deliver(bp, EOPNOTSUPP); 1153 return; 1154 } 1155 mtx_lock(&sc->sc_queue_mtx); 1156 if (bp->bio_to->error != 0) { 1157 mtx_unlock(&sc->sc_queue_mtx); 1158 g_io_deliver(bp, bp->bio_to->error); 1159 return; 1160 } 1161 TAILQ_INSERT_TAIL(&sc->sc_queue, bp, bio_queue); 1162 mtx_unlock(&sc->sc_queue_mtx); 1163 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1164 wakeup(sc); 1165 } 1166 1167 /* 1168 * Return TRUE if the given request is colliding with a in-progress 1169 * synchronization request. 1170 */ 1171 static bool 1172 g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1173 { 1174 struct g_mirror_disk *disk; 1175 struct bio *sbp; 1176 off_t rstart, rend, sstart, send; 1177 u_int i; 1178 1179 if (sc->sc_sync.ds_ndisks == 0) 1180 return (false); 1181 rstart = bp->bio_offset; 1182 rend = bp->bio_offset + bp->bio_length; 1183 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1184 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1185 continue; 1186 for (i = 0; i < g_mirror_syncreqs; i++) { 1187 sbp = disk->d_sync.ds_bios[i]; 1188 if (sbp == NULL) 1189 continue; 1190 sstart = sbp->bio_offset; 1191 send = sbp->bio_offset + sbp->bio_length; 1192 if (rend > sstart && rstart < send) 1193 return (true); 1194 } 1195 } 1196 return (false); 1197 } 1198 1199 /* 1200 * Return TRUE if the given sync request is colliding with a in-progress regular 1201 * request. 1202 */ 1203 static bool 1204 g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1205 { 1206 off_t rstart, rend, sstart, send; 1207 struct bio *bp; 1208 1209 if (sc->sc_sync.ds_ndisks == 0) 1210 return (false); 1211 sstart = sbp->bio_offset; 1212 send = sbp->bio_offset + sbp->bio_length; 1213 TAILQ_FOREACH(bp, &sc->sc_inflight, bio_queue) { 1214 rstart = bp->bio_offset; 1215 rend = bp->bio_offset + bp->bio_length; 1216 if (rend > sstart && rstart < send) 1217 return (true); 1218 } 1219 return (false); 1220 } 1221 1222 /* 1223 * Puts regular request onto delayed queue. 1224 */ 1225 static void 1226 g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1227 { 1228 1229 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1230 TAILQ_INSERT_TAIL(&sc->sc_regular_delayed, bp, bio_queue); 1231 } 1232 1233 /* 1234 * Puts synchronization request onto delayed queue. 1235 */ 1236 static void 1237 g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1238 { 1239 1240 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1241 TAILQ_INSERT_TAIL(&sc->sc_sync_delayed, bp, bio_queue); 1242 } 1243 1244 /* 1245 * Requeue delayed regular requests. 1246 */ 1247 static void 1248 g_mirror_regular_release(struct g_mirror_softc *sc) 1249 { 1250 struct bio *bp; 1251 1252 if ((bp = TAILQ_FIRST(&sc->sc_regular_delayed)) == NULL) 1253 return; 1254 if (g_mirror_sync_collision(sc, bp)) 1255 return; 1256 1257 G_MIRROR_DEBUG(2, "Requeuing regular requests after collision."); 1258 mtx_lock(&sc->sc_queue_mtx); 1259 TAILQ_CONCAT(&sc->sc_regular_delayed, &sc->sc_queue, bio_queue); 1260 TAILQ_SWAP(&sc->sc_regular_delayed, &sc->sc_queue, bio, bio_queue); 1261 mtx_unlock(&sc->sc_queue_mtx); 1262 } 1263 1264 /* 1265 * Releases delayed sync requests which don't collide anymore with regular 1266 * requests. 1267 */ 1268 static void 1269 g_mirror_sync_release(struct g_mirror_softc *sc) 1270 { 1271 struct bio *bp, *bp2; 1272 1273 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed, bio_queue, bp2) { 1274 if (g_mirror_regular_collision(sc, bp)) 1275 continue; 1276 TAILQ_REMOVE(&sc->sc_sync_delayed, bp, bio_queue); 1277 G_MIRROR_LOGREQ(2, bp, 1278 "Releasing delayed synchronization request."); 1279 g_io_request(bp, bp->bio_from); 1280 } 1281 } 1282 1283 /* 1284 * Free a synchronization request and clear its slot in the array. 1285 */ 1286 static void 1287 g_mirror_sync_request_free(struct g_mirror_disk *disk, struct bio *bp) 1288 { 1289 int idx; 1290 1291 if (disk != NULL && disk->d_sync.ds_bios != NULL) { 1292 idx = (int)(uintptr_t)bp->bio_caller1; 1293 KASSERT(disk->d_sync.ds_bios[idx] == bp, 1294 ("unexpected sync BIO at %p:%d", disk, idx)); 1295 disk->d_sync.ds_bios[idx] = NULL; 1296 } 1297 free(bp->bio_data, M_MIRROR); 1298 g_destroy_bio(bp); 1299 } 1300 1301 /* 1302 * Handle synchronization requests. 1303 * Every synchronization request is a two-step process: first, a read request is 1304 * sent to the mirror provider via the sync consumer. If that request completes 1305 * successfully, it is converted to a write and sent to the disk being 1306 * synchronized. If the write also completes successfully, the synchronization 1307 * offset is advanced and a new read request is submitted. 1308 */ 1309 static void 1310 g_mirror_sync_request(struct g_mirror_softc *sc, struct bio *bp) 1311 { 1312 struct g_mirror_disk *disk; 1313 struct g_mirror_disk_sync *sync; 1314 1315 KASSERT((bp->bio_cmd == BIO_READ && 1316 bp->bio_from->geom == sc->sc_sync.ds_geom) || 1317 (bp->bio_cmd == BIO_WRITE && bp->bio_from->geom == sc->sc_geom), 1318 ("Sync BIO %p with unexpected origin", bp)); 1319 1320 bp->bio_from->index--; 1321 disk = bp->bio_from->private; 1322 if (disk == NULL) { 1323 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1324 g_topology_lock(); 1325 g_mirror_kill_consumer(sc, bp->bio_from); 1326 g_topology_unlock(); 1327 g_mirror_sync_request_free(NULL, bp); 1328 sx_xlock(&sc->sc_lock); 1329 return; 1330 } 1331 1332 sync = &disk->d_sync; 1333 1334 /* 1335 * Synchronization request. 1336 */ 1337 switch (bp->bio_cmd) { 1338 case BIO_READ: { 1339 struct g_consumer *cp; 1340 1341 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_read, 1342 bp->bio_error); 1343 1344 if (bp->bio_error != 0) { 1345 G_MIRROR_LOGREQ(0, bp, 1346 "Synchronization request failed (error=%d).", 1347 bp->bio_error); 1348 1349 /* 1350 * The read error will trigger a syncid bump, so there's 1351 * no need to do that here. 1352 * 1353 * The read error handling for regular requests will 1354 * retry the read from all active mirrors before passing 1355 * the error back up, so there's no need to retry here. 1356 */ 1357 g_mirror_sync_request_free(disk, bp); 1358 g_mirror_event_send(disk, 1359 G_MIRROR_DISK_STATE_DISCONNECTED, 1360 G_MIRROR_EVENT_DONTWAIT); 1361 return; 1362 } 1363 G_MIRROR_LOGREQ(3, bp, 1364 "Synchronization request half-finished."); 1365 bp->bio_cmd = BIO_WRITE; 1366 bp->bio_cflags = 0; 1367 cp = disk->d_consumer; 1368 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1369 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1370 cp->acr, cp->acw, cp->ace)); 1371 cp->index++; 1372 g_io_request(bp, cp); 1373 return; 1374 } 1375 case BIO_WRITE: { 1376 off_t offset; 1377 int i; 1378 1379 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_write, 1380 bp->bio_error); 1381 1382 if (bp->bio_error != 0) { 1383 G_MIRROR_LOGREQ(0, bp, 1384 "Synchronization request failed (error=%d).", 1385 bp->bio_error); 1386 g_mirror_sync_request_free(disk, bp); 1387 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1388 g_mirror_event_send(disk, 1389 G_MIRROR_DISK_STATE_DISCONNECTED, 1390 G_MIRROR_EVENT_DONTWAIT); 1391 return; 1392 } 1393 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1394 if (sync->ds_offset >= sc->sc_mediasize || 1395 sync->ds_consumer == NULL || 1396 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1397 /* Don't send more synchronization requests. */ 1398 sync->ds_inflight--; 1399 g_mirror_sync_request_free(disk, bp); 1400 if (sync->ds_inflight > 0) 1401 return; 1402 if (sync->ds_consumer == NULL || 1403 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1404 return; 1405 } 1406 /* Disk up-to-date, activate it. */ 1407 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1408 G_MIRROR_EVENT_DONTWAIT); 1409 return; 1410 } 1411 1412 /* Send next synchronization request. */ 1413 g_mirror_sync_reinit(disk, bp, sync->ds_offset); 1414 sync->ds_offset += bp->bio_length; 1415 1416 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1417 sync->ds_consumer->index++; 1418 1419 /* 1420 * Delay the request if it is colliding with a regular request. 1421 */ 1422 if (g_mirror_regular_collision(sc, bp)) 1423 g_mirror_sync_delay(sc, bp); 1424 else 1425 g_io_request(bp, sync->ds_consumer); 1426 1427 /* Requeue delayed requests if possible. */ 1428 g_mirror_regular_release(sc); 1429 1430 /* Find the smallest offset */ 1431 offset = sc->sc_mediasize; 1432 for (i = 0; i < g_mirror_syncreqs; i++) { 1433 bp = sync->ds_bios[i]; 1434 if (bp != NULL && bp->bio_offset < offset) 1435 offset = bp->bio_offset; 1436 } 1437 if (g_mirror_sync_period > 0 && 1438 time_uptime - sync->ds_update_ts > g_mirror_sync_period) { 1439 sync->ds_offset_done = offset; 1440 g_mirror_update_metadata(disk); 1441 sync->ds_update_ts = time_uptime; 1442 } 1443 return; 1444 } 1445 default: 1446 panic("Invalid I/O request %p", bp); 1447 } 1448 } 1449 1450 static void 1451 g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1452 { 1453 struct g_mirror_disk *disk; 1454 struct g_consumer *cp; 1455 struct bio *cbp; 1456 1457 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1458 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1459 break; 1460 } 1461 if (disk == NULL) { 1462 if (bp->bio_error == 0) 1463 bp->bio_error = ENXIO; 1464 g_io_deliver(bp, bp->bio_error); 1465 return; 1466 } 1467 cbp = g_clone_bio(bp); 1468 if (cbp == NULL) { 1469 if (bp->bio_error == 0) 1470 bp->bio_error = ENOMEM; 1471 g_io_deliver(bp, bp->bio_error); 1472 return; 1473 } 1474 /* 1475 * Fill in the component buf structure. 1476 */ 1477 cp = disk->d_consumer; 1478 cbp->bio_done = g_mirror_done; 1479 cbp->bio_to = cp->provider; 1480 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1481 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1482 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1483 cp->acw, cp->ace)); 1484 cp->index++; 1485 g_io_request(cbp, cp); 1486 } 1487 1488 static void 1489 g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1490 { 1491 struct g_mirror_disk *disk; 1492 struct g_consumer *cp; 1493 struct bio *cbp; 1494 1495 disk = g_mirror_get_disk(sc); 1496 if (disk == NULL) { 1497 if (bp->bio_error == 0) 1498 bp->bio_error = ENXIO; 1499 g_io_deliver(bp, bp->bio_error); 1500 return; 1501 } 1502 cbp = g_clone_bio(bp); 1503 if (cbp == NULL) { 1504 if (bp->bio_error == 0) 1505 bp->bio_error = ENOMEM; 1506 g_io_deliver(bp, bp->bio_error); 1507 return; 1508 } 1509 /* 1510 * Fill in the component buf structure. 1511 */ 1512 cp = disk->d_consumer; 1513 cbp->bio_done = g_mirror_done; 1514 cbp->bio_to = cp->provider; 1515 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1516 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1517 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1518 cp->acw, cp->ace)); 1519 cp->index++; 1520 g_io_request(cbp, cp); 1521 } 1522 1523 #define TRACK_SIZE (1 * 1024 * 1024) 1524 #define LOAD_SCALE 256 1525 #define ABS(x) (((x) >= 0) ? (x) : (-(x))) 1526 1527 static void 1528 g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1529 { 1530 struct g_mirror_disk *disk, *dp; 1531 struct g_consumer *cp; 1532 struct bio *cbp; 1533 int prio, best; 1534 1535 /* Find a disk with the smallest load. */ 1536 disk = NULL; 1537 best = INT_MAX; 1538 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1539 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1540 continue; 1541 prio = dp->load; 1542 /* If disk head is precisely in position - highly prefer it. */ 1543 if (dp->d_last_offset == bp->bio_offset) 1544 prio -= 2 * LOAD_SCALE; 1545 else 1546 /* If disk head is close to position - prefer it. */ 1547 if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE) 1548 prio -= 1 * LOAD_SCALE; 1549 if (prio <= best) { 1550 disk = dp; 1551 best = prio; 1552 } 1553 } 1554 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1555 cbp = g_clone_bio(bp); 1556 if (cbp == NULL) { 1557 if (bp->bio_error == 0) 1558 bp->bio_error = ENOMEM; 1559 g_io_deliver(bp, bp->bio_error); 1560 return; 1561 } 1562 /* 1563 * Fill in the component buf structure. 1564 */ 1565 cp = disk->d_consumer; 1566 cbp->bio_done = g_mirror_done; 1567 cbp->bio_to = cp->provider; 1568 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1569 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1570 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1571 cp->acw, cp->ace)); 1572 cp->index++; 1573 /* Remember last head position */ 1574 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1575 /* Update loads. */ 1576 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1577 dp->load = (dp->d_consumer->index * LOAD_SCALE + 1578 dp->load * 7) / 8; 1579 } 1580 g_io_request(cbp, cp); 1581 } 1582 1583 static void 1584 g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1585 { 1586 struct bio_queue queue; 1587 struct g_mirror_disk *disk; 1588 struct g_consumer *cp; 1589 struct bio *cbp; 1590 off_t left, mod, offset, slice; 1591 u_char *data; 1592 u_int ndisks; 1593 1594 if (bp->bio_length <= sc->sc_slice) { 1595 g_mirror_request_round_robin(sc, bp); 1596 return; 1597 } 1598 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1599 slice = bp->bio_length / ndisks; 1600 mod = slice % sc->sc_provider->sectorsize; 1601 if (mod != 0) 1602 slice += sc->sc_provider->sectorsize - mod; 1603 /* 1604 * Allocate all bios before sending any request, so we can 1605 * return ENOMEM in nice and clean way. 1606 */ 1607 left = bp->bio_length; 1608 offset = bp->bio_offset; 1609 data = bp->bio_data; 1610 TAILQ_INIT(&queue); 1611 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1612 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1613 continue; 1614 cbp = g_clone_bio(bp); 1615 if (cbp == NULL) { 1616 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1617 TAILQ_REMOVE(&queue, cbp, bio_queue); 1618 g_destroy_bio(cbp); 1619 } 1620 if (bp->bio_error == 0) 1621 bp->bio_error = ENOMEM; 1622 g_io_deliver(bp, bp->bio_error); 1623 return; 1624 } 1625 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 1626 cbp->bio_done = g_mirror_done; 1627 cbp->bio_caller1 = disk; 1628 cbp->bio_to = disk->d_consumer->provider; 1629 cbp->bio_offset = offset; 1630 cbp->bio_data = data; 1631 cbp->bio_length = MIN(left, slice); 1632 left -= cbp->bio_length; 1633 if (left == 0) 1634 break; 1635 offset += cbp->bio_length; 1636 data += cbp->bio_length; 1637 } 1638 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1639 TAILQ_REMOVE(&queue, cbp, bio_queue); 1640 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1641 disk = cbp->bio_caller1; 1642 cbp->bio_caller1 = NULL; 1643 cp = disk->d_consumer; 1644 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1645 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1646 cp->acr, cp->acw, cp->ace)); 1647 disk->d_consumer->index++; 1648 g_io_request(cbp, disk->d_consumer); 1649 } 1650 } 1651 1652 static void 1653 g_mirror_register_request(struct g_mirror_softc *sc, struct bio *bp) 1654 { 1655 struct bio_queue queue; 1656 struct bio *cbp; 1657 struct g_consumer *cp; 1658 struct g_mirror_disk *disk; 1659 1660 sx_assert(&sc->sc_lock, SA_XLOCKED); 1661 1662 /* 1663 * To avoid ordering issues, if a write is deferred because of a 1664 * collision with a sync request, all I/O is deferred until that 1665 * write is initiated. 1666 */ 1667 if (bp->bio_from->geom != sc->sc_sync.ds_geom && 1668 !TAILQ_EMPTY(&sc->sc_regular_delayed)) { 1669 g_mirror_regular_delay(sc, bp); 1670 return; 1671 } 1672 1673 switch (bp->bio_cmd) { 1674 case BIO_READ: 1675 switch (sc->sc_balance) { 1676 case G_MIRROR_BALANCE_LOAD: 1677 g_mirror_request_load(sc, bp); 1678 break; 1679 case G_MIRROR_BALANCE_PREFER: 1680 g_mirror_request_prefer(sc, bp); 1681 break; 1682 case G_MIRROR_BALANCE_ROUND_ROBIN: 1683 g_mirror_request_round_robin(sc, bp); 1684 break; 1685 case G_MIRROR_BALANCE_SPLIT: 1686 g_mirror_request_split(sc, bp); 1687 break; 1688 } 1689 return; 1690 case BIO_WRITE: 1691 case BIO_DELETE: 1692 /* 1693 * Delay the request if it is colliding with a synchronization 1694 * request. 1695 */ 1696 if (g_mirror_sync_collision(sc, bp)) { 1697 g_mirror_regular_delay(sc, bp); 1698 return; 1699 } 1700 1701 if (sc->sc_idle) 1702 g_mirror_unidle(sc); 1703 else 1704 sc->sc_last_write = time_uptime; 1705 1706 /* 1707 * Bump syncid on first write. 1708 */ 1709 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1710 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1711 g_mirror_bump_syncid(sc); 1712 } 1713 1714 /* 1715 * Allocate all bios before sending any request, so we can 1716 * return ENOMEM in nice and clean way. 1717 */ 1718 TAILQ_INIT(&queue); 1719 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1720 switch (disk->d_state) { 1721 case G_MIRROR_DISK_STATE_ACTIVE: 1722 break; 1723 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1724 if (bp->bio_offset >= disk->d_sync.ds_offset) 1725 continue; 1726 break; 1727 default: 1728 continue; 1729 } 1730 if (bp->bio_cmd == BIO_DELETE && 1731 (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0) 1732 continue; 1733 cbp = g_clone_bio(bp); 1734 if (cbp == NULL) { 1735 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1736 TAILQ_REMOVE(&queue, cbp, bio_queue); 1737 g_destroy_bio(cbp); 1738 } 1739 if (bp->bio_error == 0) 1740 bp->bio_error = ENOMEM; 1741 g_io_deliver(bp, bp->bio_error); 1742 return; 1743 } 1744 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 1745 cbp->bio_done = g_mirror_done; 1746 cp = disk->d_consumer; 1747 cbp->bio_caller1 = cp; 1748 cbp->bio_to = cp->provider; 1749 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1750 ("Consumer %s not opened (r%dw%de%d).", 1751 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1752 } 1753 if (TAILQ_EMPTY(&queue)) { 1754 KASSERT(bp->bio_cmd == BIO_DELETE, 1755 ("No consumers for regular request %p", bp)); 1756 g_io_deliver(bp, EOPNOTSUPP); 1757 return; 1758 } 1759 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1760 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1761 TAILQ_REMOVE(&queue, cbp, bio_queue); 1762 cp = cbp->bio_caller1; 1763 cbp->bio_caller1 = NULL; 1764 cp->index++; 1765 sc->sc_writes++; 1766 g_io_request(cbp, cp); 1767 } 1768 /* 1769 * Put request onto inflight queue, so we can check if new 1770 * synchronization requests don't collide with it. 1771 */ 1772 TAILQ_INSERT_TAIL(&sc->sc_inflight, bp, bio_queue); 1773 return; 1774 case BIO_FLUSH: 1775 TAILQ_INIT(&queue); 1776 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1777 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1778 continue; 1779 cbp = g_clone_bio(bp); 1780 if (cbp == NULL) { 1781 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1782 TAILQ_REMOVE(&queue, cbp, bio_queue); 1783 g_destroy_bio(cbp); 1784 } 1785 if (bp->bio_error == 0) 1786 bp->bio_error = ENOMEM; 1787 g_io_deliver(bp, bp->bio_error); 1788 return; 1789 } 1790 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 1791 cbp->bio_done = g_mirror_done; 1792 cbp->bio_caller1 = disk; 1793 cbp->bio_to = disk->d_consumer->provider; 1794 } 1795 KASSERT(!TAILQ_EMPTY(&queue), 1796 ("No consumers for regular request %p", bp)); 1797 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1798 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1799 TAILQ_REMOVE(&queue, cbp, bio_queue); 1800 disk = cbp->bio_caller1; 1801 cbp->bio_caller1 = NULL; 1802 cp = disk->d_consumer; 1803 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1804 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1805 cp->acr, cp->acw, cp->ace)); 1806 cp->index++; 1807 g_io_request(cbp, cp); 1808 } 1809 break; 1810 default: 1811 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1812 bp->bio_cmd, sc->sc_name)); 1813 break; 1814 } 1815 } 1816 1817 static int 1818 g_mirror_can_destroy(struct g_mirror_softc *sc) 1819 { 1820 struct g_geom *gp; 1821 struct g_consumer *cp; 1822 1823 g_topology_assert(); 1824 gp = sc->sc_geom; 1825 if (gp->softc == NULL) 1826 return (1); 1827 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0) 1828 return (0); 1829 LIST_FOREACH(cp, &gp->consumer, consumer) { 1830 if (g_mirror_is_busy(sc, cp)) 1831 return (0); 1832 } 1833 gp = sc->sc_sync.ds_geom; 1834 LIST_FOREACH(cp, &gp->consumer, consumer) { 1835 if (g_mirror_is_busy(sc, cp)) 1836 return (0); 1837 } 1838 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1839 sc->sc_name); 1840 return (1); 1841 } 1842 1843 static int 1844 g_mirror_try_destroy(struct g_mirror_softc *sc) 1845 { 1846 1847 if (sc->sc_rootmount != NULL) { 1848 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1849 sc->sc_rootmount); 1850 root_mount_rel(sc->sc_rootmount); 1851 sc->sc_rootmount = NULL; 1852 } 1853 g_topology_lock(); 1854 if (!g_mirror_can_destroy(sc)) { 1855 g_topology_unlock(); 1856 return (0); 1857 } 1858 sc->sc_geom->softc = NULL; 1859 sc->sc_sync.ds_geom->softc = NULL; 1860 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DRAIN) != 0) { 1861 g_topology_unlock(); 1862 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1863 &sc->sc_worker); 1864 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1865 sx_xunlock(&sc->sc_lock); 1866 wakeup(&sc->sc_worker); 1867 sc->sc_worker = NULL; 1868 } else { 1869 g_topology_unlock(); 1870 g_mirror_destroy_device(sc); 1871 } 1872 return (1); 1873 } 1874 1875 /* 1876 * Worker thread. 1877 */ 1878 static void 1879 g_mirror_worker(void *arg) 1880 { 1881 struct g_mirror_softc *sc; 1882 struct g_mirror_event *ep; 1883 struct bio *bp; 1884 int timeout; 1885 1886 sc = arg; 1887 thread_lock(curthread); 1888 sched_prio(curthread, PRIBIO); 1889 thread_unlock(curthread); 1890 1891 sx_xlock(&sc->sc_lock); 1892 for (;;) { 1893 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1894 /* 1895 * First take a look at events. 1896 * This is important to handle events before any I/O requests. 1897 */ 1898 ep = g_mirror_event_first(sc); 1899 if (ep != NULL) { 1900 g_mirror_event_remove(sc, ep); 1901 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1902 /* Update only device status. */ 1903 G_MIRROR_DEBUG(3, 1904 "Running event for device %s.", 1905 sc->sc_name); 1906 ep->e_error = 0; 1907 g_mirror_update_device(sc, true); 1908 } else { 1909 /* Update disk status. */ 1910 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1911 g_mirror_get_diskname(ep->e_disk)); 1912 ep->e_error = g_mirror_update_disk(ep->e_disk, 1913 ep->e_state); 1914 if (ep->e_error == 0) 1915 g_mirror_update_device(sc, false); 1916 } 1917 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1918 KASSERT(ep->e_error == 0, 1919 ("Error cannot be handled.")); 1920 g_mirror_event_free(ep); 1921 } else { 1922 ep->e_flags |= G_MIRROR_EVENT_DONE; 1923 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1924 ep); 1925 mtx_lock(&sc->sc_events_mtx); 1926 wakeup(ep); 1927 mtx_unlock(&sc->sc_events_mtx); 1928 } 1929 if ((sc->sc_flags & 1930 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1931 if (g_mirror_try_destroy(sc)) { 1932 curthread->td_pflags &= ~TDP_GEOM; 1933 G_MIRROR_DEBUG(1, "Thread exiting."); 1934 kproc_exit(0); 1935 } 1936 } 1937 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1938 continue; 1939 } 1940 1941 /* 1942 * Check if we can mark array as CLEAN and if we can't take 1943 * how much seconds should we wait. 1944 */ 1945 timeout = g_mirror_idle(sc, -1); 1946 1947 /* 1948 * Handle I/O requests. 1949 */ 1950 mtx_lock(&sc->sc_queue_mtx); 1951 bp = TAILQ_FIRST(&sc->sc_queue); 1952 if (bp != NULL) 1953 TAILQ_REMOVE(&sc->sc_queue, bp, bio_queue); 1954 else { 1955 if ((sc->sc_flags & 1956 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1957 mtx_unlock(&sc->sc_queue_mtx); 1958 if (g_mirror_try_destroy(sc)) { 1959 curthread->td_pflags &= ~TDP_GEOM; 1960 G_MIRROR_DEBUG(1, "Thread exiting."); 1961 kproc_exit(0); 1962 } 1963 mtx_lock(&sc->sc_queue_mtx); 1964 if (!TAILQ_EMPTY(&sc->sc_queue)) { 1965 mtx_unlock(&sc->sc_queue_mtx); 1966 continue; 1967 } 1968 } 1969 if (g_mirror_event_first(sc) != NULL) { 1970 mtx_unlock(&sc->sc_queue_mtx); 1971 continue; 1972 } 1973 sx_xunlock(&sc->sc_lock); 1974 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1975 timeout * hz); 1976 sx_xlock(&sc->sc_lock); 1977 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1978 continue; 1979 } 1980 mtx_unlock(&sc->sc_queue_mtx); 1981 1982 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 1983 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1984 /* 1985 * Handle completion of the first half (the read) of a 1986 * block synchronization operation. 1987 */ 1988 g_mirror_sync_request(sc, bp); 1989 } else if (bp->bio_to != sc->sc_provider) { 1990 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 1991 /* 1992 * Handle completion of a regular I/O request. 1993 */ 1994 g_mirror_regular_request(sc, bp); 1995 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 1996 /* 1997 * Handle completion of the second half (the 1998 * write) of a block synchronization operation. 1999 */ 2000 g_mirror_sync_request(sc, bp); 2001 else { 2002 KASSERT(0, 2003 ("Invalid request cflags=0x%hx to=%s.", 2004 bp->bio_cflags, bp->bio_to->name)); 2005 } 2006 } else { 2007 /* 2008 * Initiate an I/O request. 2009 */ 2010 g_mirror_register_request(sc, bp); 2011 } 2012 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 2013 } 2014 } 2015 2016 static void 2017 g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 2018 { 2019 2020 sx_assert(&sc->sc_lock, SX_LOCKED); 2021 2022 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 2023 return; 2024 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2025 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 2026 g_mirror_get_diskname(disk), sc->sc_name); 2027 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2028 } else if (sc->sc_idle && 2029 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2030 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 2031 g_mirror_get_diskname(disk), sc->sc_name); 2032 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2033 } 2034 } 2035 2036 static void 2037 g_mirror_sync_reinit(const struct g_mirror_disk *disk, struct bio *bp, 2038 off_t offset) 2039 { 2040 void *data; 2041 int idx; 2042 2043 data = bp->bio_data; 2044 idx = (int)(uintptr_t)bp->bio_caller1; 2045 g_reset_bio(bp); 2046 2047 bp->bio_cmd = BIO_READ; 2048 bp->bio_data = data; 2049 bp->bio_done = g_mirror_sync_done; 2050 bp->bio_from = disk->d_sync.ds_consumer; 2051 bp->bio_to = disk->d_softc->sc_provider; 2052 bp->bio_caller1 = (void *)(uintptr_t)idx; 2053 bp->bio_offset = offset; 2054 bp->bio_length = MIN(MAXPHYS, 2055 disk->d_softc->sc_mediasize - bp->bio_offset); 2056 } 2057 2058 static void 2059 g_mirror_sync_start(struct g_mirror_disk *disk) 2060 { 2061 struct g_mirror_softc *sc; 2062 struct g_mirror_disk_sync *sync; 2063 struct g_consumer *cp; 2064 struct bio *bp; 2065 int error, i; 2066 2067 g_topology_assert_not(); 2068 sc = disk->d_softc; 2069 sync = &disk->d_sync; 2070 sx_assert(&sc->sc_lock, SX_LOCKED); 2071 2072 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2073 ("Disk %s is not marked for synchronization.", 2074 g_mirror_get_diskname(disk))); 2075 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2076 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 2077 sc->sc_state)); 2078 2079 sx_xunlock(&sc->sc_lock); 2080 g_topology_lock(); 2081 cp = g_new_consumer(sc->sc_sync.ds_geom); 2082 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 2083 error = g_attach(cp, sc->sc_provider); 2084 KASSERT(error == 0, 2085 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 2086 error = g_access(cp, 1, 0, 0); 2087 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 2088 g_topology_unlock(); 2089 sx_xlock(&sc->sc_lock); 2090 2091 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 2092 g_mirror_get_diskname(disk)); 2093 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0) 2094 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2095 KASSERT(sync->ds_consumer == NULL, 2096 ("Sync consumer already exists (device=%s, disk=%s).", 2097 sc->sc_name, g_mirror_get_diskname(disk))); 2098 2099 sync->ds_consumer = cp; 2100 sync->ds_consumer->private = disk; 2101 sync->ds_consumer->index = 0; 2102 2103 /* 2104 * Allocate memory for synchronization bios and initialize them. 2105 */ 2106 sync->ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 2107 M_MIRROR, M_WAITOK); 2108 for (i = 0; i < g_mirror_syncreqs; i++) { 2109 bp = g_alloc_bio(); 2110 sync->ds_bios[i] = bp; 2111 2112 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 2113 bp->bio_caller1 = (void *)(uintptr_t)i; 2114 g_mirror_sync_reinit(disk, bp, sync->ds_offset); 2115 sync->ds_offset += bp->bio_length; 2116 } 2117 2118 /* Increase the number of disks in SYNCHRONIZING state. */ 2119 sc->sc_sync.ds_ndisks++; 2120 /* Set the number of in-flight synchronization requests. */ 2121 sync->ds_inflight = g_mirror_syncreqs; 2122 2123 /* 2124 * Fire off first synchronization requests. 2125 */ 2126 for (i = 0; i < g_mirror_syncreqs; i++) { 2127 bp = sync->ds_bios[i]; 2128 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 2129 sync->ds_consumer->index++; 2130 /* 2131 * Delay the request if it is colliding with a regular request. 2132 */ 2133 if (g_mirror_regular_collision(sc, bp)) 2134 g_mirror_sync_delay(sc, bp); 2135 else 2136 g_io_request(bp, sync->ds_consumer); 2137 } 2138 } 2139 2140 /* 2141 * Stop synchronization process. 2142 * type: 0 - synchronization finished 2143 * 1 - synchronization stopped 2144 */ 2145 static void 2146 g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 2147 { 2148 struct g_mirror_softc *sc; 2149 struct g_consumer *cp; 2150 2151 g_topology_assert_not(); 2152 sc = disk->d_softc; 2153 sx_assert(&sc->sc_lock, SX_LOCKED); 2154 2155 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2156 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2157 g_mirror_disk_state2str(disk->d_state))); 2158 if (disk->d_sync.ds_consumer == NULL) 2159 return; 2160 2161 if (type == 0) { 2162 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2163 sc->sc_name, g_mirror_get_diskname(disk)); 2164 } else /* if (type == 1) */ { 2165 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2166 sc->sc_name, g_mirror_get_diskname(disk)); 2167 } 2168 g_mirror_regular_release(sc); 2169 free(disk->d_sync.ds_bios, M_MIRROR); 2170 disk->d_sync.ds_bios = NULL; 2171 cp = disk->d_sync.ds_consumer; 2172 disk->d_sync.ds_consumer = NULL; 2173 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2174 sc->sc_sync.ds_ndisks--; 2175 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2176 g_topology_lock(); 2177 g_mirror_kill_consumer(sc, cp); 2178 g_topology_unlock(); 2179 sx_xlock(&sc->sc_lock); 2180 } 2181 2182 static void 2183 g_mirror_launch_provider(struct g_mirror_softc *sc) 2184 { 2185 struct g_mirror_disk *disk; 2186 struct g_provider *pp, *dp; 2187 2188 sx_assert(&sc->sc_lock, SX_LOCKED); 2189 2190 g_topology_lock(); 2191 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 2192 pp->flags |= G_PF_DIRECT_RECEIVE; 2193 pp->mediasize = sc->sc_mediasize; 2194 pp->sectorsize = sc->sc_sectorsize; 2195 pp->stripesize = 0; 2196 pp->stripeoffset = 0; 2197 2198 /* Splitting of unmapped BIO's could work but isn't implemented now */ 2199 if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT) 2200 pp->flags |= G_PF_ACCEPT_UNMAPPED; 2201 2202 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2203 if (disk->d_consumer && disk->d_consumer->provider) { 2204 dp = disk->d_consumer->provider; 2205 if (dp->stripesize > pp->stripesize) { 2206 pp->stripesize = dp->stripesize; 2207 pp->stripeoffset = dp->stripeoffset; 2208 } 2209 /* A provider underneath us doesn't support unmapped */ 2210 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { 2211 G_MIRROR_DEBUG(0, "Cancelling unmapped " 2212 "because of %s.", dp->name); 2213 pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 2214 } 2215 } 2216 } 2217 pp->private = sc; 2218 sc->sc_refcnt++; 2219 sc->sc_provider = pp; 2220 g_error_provider(pp, 0); 2221 g_topology_unlock(); 2222 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2223 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 2224 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2225 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2226 g_mirror_sync_start(disk); 2227 } 2228 } 2229 2230 static void 2231 g_mirror_destroy_provider(struct g_mirror_softc *sc) 2232 { 2233 struct g_mirror_disk *disk; 2234 struct bio *bp; 2235 2236 g_topology_assert_not(); 2237 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2238 sc->sc_name)); 2239 2240 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2241 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2242 g_mirror_sync_stop(disk, 1); 2243 } 2244 2245 g_topology_lock(); 2246 g_error_provider(sc->sc_provider, ENXIO); 2247 mtx_lock(&sc->sc_queue_mtx); 2248 while ((bp = TAILQ_FIRST(&sc->sc_queue)) != NULL) { 2249 TAILQ_REMOVE(&sc->sc_queue, bp, bio_queue); 2250 /* 2251 * Abort any pending I/O that wasn't generated by us. 2252 * Synchronization requests and requests destined for individual 2253 * mirror components can be destroyed immediately. 2254 */ 2255 if (bp->bio_to == sc->sc_provider && 2256 bp->bio_from->geom != sc->sc_sync.ds_geom) { 2257 g_io_deliver(bp, ENXIO); 2258 } else { 2259 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2260 free(bp->bio_data, M_MIRROR); 2261 g_destroy_bio(bp); 2262 } 2263 } 2264 mtx_unlock(&sc->sc_queue_mtx); 2265 g_wither_provider(sc->sc_provider, ENXIO); 2266 sc->sc_provider = NULL; 2267 G_MIRROR_DEBUG(0, "Device %s: provider destroyed.", sc->sc_name); 2268 g_topology_unlock(); 2269 } 2270 2271 static void 2272 g_mirror_go(void *arg) 2273 { 2274 struct g_mirror_softc *sc; 2275 2276 sc = arg; 2277 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2278 g_mirror_event_send(sc, 0, 2279 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2280 } 2281 2282 static u_int 2283 g_mirror_determine_state(struct g_mirror_disk *disk) 2284 { 2285 struct g_mirror_softc *sc; 2286 u_int state; 2287 2288 sc = disk->d_softc; 2289 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2290 if ((disk->d_flags & 2291 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0 && 2292 (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 || 2293 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0)) { 2294 /* Disk does not need synchronization. */ 2295 state = G_MIRROR_DISK_STATE_ACTIVE; 2296 } else { 2297 if ((sc->sc_flags & 2298 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2299 (disk->d_flags & 2300 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2301 /* 2302 * We can start synchronization from 2303 * the stored offset. 2304 */ 2305 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2306 } else { 2307 state = G_MIRROR_DISK_STATE_STALE; 2308 } 2309 } 2310 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2311 /* 2312 * Reset all synchronization data for this disk, 2313 * because if it even was synchronized, it was 2314 * synchronized to disks with different syncid. 2315 */ 2316 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2317 disk->d_sync.ds_offset = 0; 2318 disk->d_sync.ds_offset_done = 0; 2319 disk->d_sync.ds_syncid = sc->sc_syncid; 2320 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2321 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2322 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2323 } else { 2324 state = G_MIRROR_DISK_STATE_STALE; 2325 } 2326 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2327 /* 2328 * Not good, NOT GOOD! 2329 * It means that mirror was started on stale disks 2330 * and more fresh disk just arrive. 2331 * If there were writes, mirror is broken, sorry. 2332 * I think the best choice here is don't touch 2333 * this disk and inform the user loudly. 2334 */ 2335 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2336 "disk (%s) arrives!! It will not be connected to the " 2337 "running device.", sc->sc_name, 2338 g_mirror_get_diskname(disk)); 2339 g_mirror_destroy_disk(disk); 2340 state = G_MIRROR_DISK_STATE_NONE; 2341 /* Return immediately, because disk was destroyed. */ 2342 return (state); 2343 } 2344 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2345 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2346 return (state); 2347 } 2348 2349 /* 2350 * Update device state. 2351 */ 2352 static void 2353 g_mirror_update_device(struct g_mirror_softc *sc, bool force) 2354 { 2355 struct g_mirror_disk *disk; 2356 u_int state; 2357 2358 sx_assert(&sc->sc_lock, SX_XLOCKED); 2359 2360 switch (sc->sc_state) { 2361 case G_MIRROR_DEVICE_STATE_STARTING: 2362 { 2363 struct g_mirror_disk *pdisk, *tdisk; 2364 u_int dirty, ndisks, genid, syncid; 2365 bool broken; 2366 2367 KASSERT(sc->sc_provider == NULL, 2368 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2369 /* 2370 * Are we ready? We are, if all disks are connected or 2371 * if we have any disks and 'force' is true. 2372 */ 2373 ndisks = g_mirror_ndisks(sc, -1); 2374 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2375 ; 2376 } else if (ndisks == 0) { 2377 /* 2378 * Disks went down in starting phase, so destroy 2379 * device. 2380 */ 2381 callout_drain(&sc->sc_callout); 2382 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2383 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2384 sc->sc_rootmount); 2385 root_mount_rel(sc->sc_rootmount); 2386 sc->sc_rootmount = NULL; 2387 return; 2388 } else { 2389 return; 2390 } 2391 2392 /* 2393 * Activate all disks with the biggest syncid. 2394 */ 2395 if (force) { 2396 /* 2397 * If 'force' is true, we have been called due to 2398 * timeout, so don't bother canceling timeout. 2399 */ 2400 ndisks = 0; 2401 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2402 if ((disk->d_flags & 2403 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2404 ndisks++; 2405 } 2406 } 2407 if (ndisks == 0) { 2408 /* No valid disks found, destroy device. */ 2409 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2410 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2411 __LINE__, sc->sc_rootmount); 2412 root_mount_rel(sc->sc_rootmount); 2413 sc->sc_rootmount = NULL; 2414 return; 2415 } 2416 } else { 2417 /* Cancel timeout. */ 2418 callout_drain(&sc->sc_callout); 2419 } 2420 2421 /* 2422 * Find the biggest genid. 2423 */ 2424 genid = 0; 2425 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2426 if (disk->d_genid > genid) 2427 genid = disk->d_genid; 2428 } 2429 sc->sc_genid = genid; 2430 /* 2431 * Remove all disks without the biggest genid. 2432 */ 2433 broken = false; 2434 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2435 if (disk->d_genid < genid) { 2436 G_MIRROR_DEBUG(0, 2437 "Component %s (device %s) broken, skipping.", 2438 g_mirror_get_diskname(disk), sc->sc_name); 2439 g_mirror_destroy_disk(disk); 2440 /* 2441 * Bump the syncid in case we discover a healthy 2442 * replacement disk after starting the mirror. 2443 */ 2444 broken = true; 2445 } 2446 } 2447 2448 /* 2449 * Find the biggest syncid. 2450 */ 2451 syncid = 0; 2452 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2453 if (disk->d_sync.ds_syncid > syncid) 2454 syncid = disk->d_sync.ds_syncid; 2455 } 2456 2457 /* 2458 * Here we need to look for dirty disks and if all disks 2459 * with the biggest syncid are dirty, we have to choose 2460 * one with the biggest priority and rebuild the rest. 2461 */ 2462 /* 2463 * Find the number of dirty disks with the biggest syncid. 2464 * Find the number of disks with the biggest syncid. 2465 * While here, find a disk with the biggest priority. 2466 */ 2467 dirty = ndisks = 0; 2468 pdisk = NULL; 2469 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2470 if (disk->d_sync.ds_syncid != syncid) 2471 continue; 2472 if ((disk->d_flags & 2473 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2474 continue; 2475 } 2476 ndisks++; 2477 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2478 dirty++; 2479 if (pdisk == NULL || 2480 pdisk->d_priority < disk->d_priority) { 2481 pdisk = disk; 2482 } 2483 } 2484 } 2485 if (dirty == 0) { 2486 /* No dirty disks at all, great. */ 2487 } else if (dirty == ndisks) { 2488 /* 2489 * Force synchronization for all dirty disks except one 2490 * with the biggest priority. 2491 */ 2492 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2493 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2494 "master disk for synchronization.", 2495 g_mirror_get_diskname(pdisk), sc->sc_name); 2496 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2497 if (disk->d_sync.ds_syncid != syncid) 2498 continue; 2499 if ((disk->d_flags & 2500 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2501 continue; 2502 } 2503 KASSERT((disk->d_flags & 2504 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2505 ("Disk %s isn't marked as dirty.", 2506 g_mirror_get_diskname(disk))); 2507 /* Skip the disk with the biggest priority. */ 2508 if (disk == pdisk) 2509 continue; 2510 disk->d_sync.ds_syncid = 0; 2511 } 2512 } else if (dirty < ndisks) { 2513 /* 2514 * Force synchronization for all dirty disks. 2515 * We have some non-dirty disks. 2516 */ 2517 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2518 if (disk->d_sync.ds_syncid != syncid) 2519 continue; 2520 if ((disk->d_flags & 2521 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2522 continue; 2523 } 2524 if ((disk->d_flags & 2525 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2526 continue; 2527 } 2528 disk->d_sync.ds_syncid = 0; 2529 } 2530 } 2531 2532 /* Reset hint. */ 2533 sc->sc_hint = NULL; 2534 sc->sc_syncid = syncid; 2535 if (force || broken) { 2536 /* Remember to bump syncid on first write. */ 2537 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2538 } 2539 state = G_MIRROR_DEVICE_STATE_RUNNING; 2540 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2541 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2542 g_mirror_device_state2str(state)); 2543 sc->sc_state = state; 2544 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2545 state = g_mirror_determine_state(disk); 2546 g_mirror_event_send(disk, state, 2547 G_MIRROR_EVENT_DONTWAIT); 2548 if (state == G_MIRROR_DISK_STATE_STALE) 2549 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2550 } 2551 break; 2552 } 2553 case G_MIRROR_DEVICE_STATE_RUNNING: 2554 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2555 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2556 /* 2557 * No usable disks, so destroy the device. 2558 */ 2559 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2560 break; 2561 } else if (g_mirror_ndisks(sc, 2562 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2563 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2564 /* 2565 * We have active disks, launch provider if it doesn't 2566 * exist. 2567 */ 2568 if (sc->sc_provider == NULL) 2569 g_mirror_launch_provider(sc); 2570 if (sc->sc_rootmount != NULL) { 2571 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2572 __LINE__, sc->sc_rootmount); 2573 root_mount_rel(sc->sc_rootmount); 2574 sc->sc_rootmount = NULL; 2575 } 2576 } 2577 /* 2578 * Genid should be bumped immediately, so do it here. 2579 */ 2580 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2581 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2582 g_mirror_bump_genid(sc); 2583 } 2584 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_NOW) != 0) { 2585 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID_NOW; 2586 g_mirror_bump_syncid(sc); 2587 } 2588 break; 2589 default: 2590 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2591 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2592 break; 2593 } 2594 } 2595 2596 /* 2597 * Update disk state and device state if needed. 2598 */ 2599 #define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2600 "Disk %s state changed from %s to %s (device %s).", \ 2601 g_mirror_get_diskname(disk), \ 2602 g_mirror_disk_state2str(disk->d_state), \ 2603 g_mirror_disk_state2str(state), sc->sc_name) 2604 static int 2605 g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2606 { 2607 struct g_mirror_softc *sc; 2608 2609 sc = disk->d_softc; 2610 sx_assert(&sc->sc_lock, SX_XLOCKED); 2611 2612 again: 2613 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2614 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2615 g_mirror_disk_state2str(state)); 2616 switch (state) { 2617 case G_MIRROR_DISK_STATE_NEW: 2618 /* 2619 * Possible scenarios: 2620 * 1. New disk arrive. 2621 */ 2622 /* Previous state should be NONE. */ 2623 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2624 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2625 g_mirror_disk_state2str(disk->d_state))); 2626 DISK_STATE_CHANGED(); 2627 2628 disk->d_state = state; 2629 if (LIST_EMPTY(&sc->sc_disks)) 2630 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2631 else { 2632 struct g_mirror_disk *dp; 2633 2634 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2635 if (disk->d_priority >= dp->d_priority) { 2636 LIST_INSERT_BEFORE(dp, disk, d_next); 2637 dp = NULL; 2638 break; 2639 } 2640 if (LIST_NEXT(dp, d_next) == NULL) 2641 break; 2642 } 2643 if (dp != NULL) 2644 LIST_INSERT_AFTER(dp, disk, d_next); 2645 } 2646 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2647 sc->sc_name, g_mirror_get_diskname(disk)); 2648 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2649 break; 2650 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2651 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2652 g_mirror_device_state2str(sc->sc_state), 2653 g_mirror_get_diskname(disk), 2654 g_mirror_disk_state2str(disk->d_state))); 2655 state = g_mirror_determine_state(disk); 2656 if (state != G_MIRROR_DISK_STATE_NONE) 2657 goto again; 2658 break; 2659 case G_MIRROR_DISK_STATE_ACTIVE: 2660 /* 2661 * Possible scenarios: 2662 * 1. New disk does not need synchronization. 2663 * 2. Synchronization process finished successfully. 2664 */ 2665 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2666 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2667 g_mirror_device_state2str(sc->sc_state), 2668 g_mirror_get_diskname(disk), 2669 g_mirror_disk_state2str(disk->d_state))); 2670 /* Previous state should be NEW or SYNCHRONIZING. */ 2671 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2672 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2673 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2674 g_mirror_disk_state2str(disk->d_state))); 2675 DISK_STATE_CHANGED(); 2676 2677 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2678 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2679 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2680 g_mirror_sync_stop(disk, 0); 2681 } 2682 disk->d_state = state; 2683 disk->d_sync.ds_offset = 0; 2684 disk->d_sync.ds_offset_done = 0; 2685 g_mirror_update_idle(sc, disk); 2686 g_mirror_update_metadata(disk); 2687 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2688 sc->sc_name, g_mirror_get_diskname(disk)); 2689 break; 2690 case G_MIRROR_DISK_STATE_STALE: 2691 /* 2692 * Possible scenarios: 2693 * 1. Stale disk was connected. 2694 */ 2695 /* Previous state should be NEW. */ 2696 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2697 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2698 g_mirror_disk_state2str(disk->d_state))); 2699 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2700 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2701 g_mirror_device_state2str(sc->sc_state), 2702 g_mirror_get_diskname(disk), 2703 g_mirror_disk_state2str(disk->d_state))); 2704 /* 2705 * STALE state is only possible if device is marked 2706 * NOAUTOSYNC. 2707 */ 2708 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2709 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2710 g_mirror_device_state2str(sc->sc_state), 2711 g_mirror_get_diskname(disk), 2712 g_mirror_disk_state2str(disk->d_state))); 2713 DISK_STATE_CHANGED(); 2714 2715 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2716 disk->d_state = state; 2717 g_mirror_update_metadata(disk); 2718 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2719 sc->sc_name, g_mirror_get_diskname(disk)); 2720 break; 2721 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2722 /* 2723 * Possible scenarios: 2724 * 1. Disk which needs synchronization was connected. 2725 */ 2726 /* Previous state should be NEW. */ 2727 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2728 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2729 g_mirror_disk_state2str(disk->d_state))); 2730 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2731 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2732 g_mirror_device_state2str(sc->sc_state), 2733 g_mirror_get_diskname(disk), 2734 g_mirror_disk_state2str(disk->d_state))); 2735 DISK_STATE_CHANGED(); 2736 2737 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2738 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2739 disk->d_state = state; 2740 if (sc->sc_provider != NULL) { 2741 g_mirror_sync_start(disk); 2742 g_mirror_update_metadata(disk); 2743 } 2744 break; 2745 case G_MIRROR_DISK_STATE_DISCONNECTED: 2746 /* 2747 * Possible scenarios: 2748 * 1. Device wasn't running yet, but disk disappear. 2749 * 2. Disk was active and disapppear. 2750 * 3. Disk disappear during synchronization process. 2751 */ 2752 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2753 /* 2754 * Previous state should be ACTIVE, STALE or 2755 * SYNCHRONIZING. 2756 */ 2757 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2758 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2759 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2760 ("Wrong disk state (%s, %s).", 2761 g_mirror_get_diskname(disk), 2762 g_mirror_disk_state2str(disk->d_state))); 2763 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2764 /* Previous state should be NEW. */ 2765 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2766 ("Wrong disk state (%s, %s).", 2767 g_mirror_get_diskname(disk), 2768 g_mirror_disk_state2str(disk->d_state))); 2769 /* 2770 * Reset bumping syncid if disk disappeared in STARTING 2771 * state. 2772 */ 2773 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2774 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2775 #ifdef INVARIANTS 2776 } else { 2777 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2778 sc->sc_name, 2779 g_mirror_device_state2str(sc->sc_state), 2780 g_mirror_get_diskname(disk), 2781 g_mirror_disk_state2str(disk->d_state))); 2782 #endif 2783 } 2784 DISK_STATE_CHANGED(); 2785 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2786 sc->sc_name, g_mirror_get_diskname(disk)); 2787 2788 g_mirror_destroy_disk(disk); 2789 break; 2790 case G_MIRROR_DISK_STATE_DESTROY: 2791 { 2792 int error; 2793 2794 error = g_mirror_clear_metadata(disk); 2795 if (error != 0) { 2796 G_MIRROR_DEBUG(0, 2797 "Device %s: failed to clear metadata on %s: %d.", 2798 sc->sc_name, g_mirror_get_diskname(disk), error); 2799 break; 2800 } 2801 DISK_STATE_CHANGED(); 2802 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2803 sc->sc_name, g_mirror_get_diskname(disk)); 2804 2805 g_mirror_destroy_disk(disk); 2806 sc->sc_ndisks--; 2807 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2808 g_mirror_update_metadata(disk); 2809 } 2810 break; 2811 } 2812 default: 2813 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2814 break; 2815 } 2816 return (0); 2817 } 2818 #undef DISK_STATE_CHANGED 2819 2820 int 2821 g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2822 { 2823 struct g_provider *pp; 2824 u_char *buf; 2825 int error; 2826 2827 g_topology_assert(); 2828 2829 error = g_access(cp, 1, 0, 0); 2830 if (error != 0) 2831 return (error); 2832 pp = cp->provider; 2833 g_topology_unlock(); 2834 /* Metadata are stored on last sector. */ 2835 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2836 &error); 2837 g_topology_lock(); 2838 g_access(cp, -1, 0, 0); 2839 if (buf == NULL) { 2840 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2841 cp->provider->name, error); 2842 return (error); 2843 } 2844 2845 /* Decode metadata. */ 2846 error = mirror_metadata_decode(buf, md); 2847 g_free(buf); 2848 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2849 return (EINVAL); 2850 if (md->md_version > G_MIRROR_VERSION) { 2851 G_MIRROR_DEBUG(0, 2852 "Kernel module is too old to handle metadata from %s.", 2853 cp->provider->name); 2854 return (EINVAL); 2855 } 2856 if (error != 0) { 2857 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2858 cp->provider->name); 2859 return (error); 2860 } 2861 2862 return (0); 2863 } 2864 2865 static int 2866 g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2867 struct g_mirror_metadata *md) 2868 { 2869 2870 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2871 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2872 pp->name, md->md_did); 2873 return (EEXIST); 2874 } 2875 if (md->md_all != sc->sc_ndisks) { 2876 G_MIRROR_DEBUG(1, 2877 "Invalid '%s' field on disk %s (device %s), skipping.", 2878 "md_all", pp->name, sc->sc_name); 2879 return (EINVAL); 2880 } 2881 if (md->md_slice != sc->sc_slice) { 2882 G_MIRROR_DEBUG(1, 2883 "Invalid '%s' field on disk %s (device %s), skipping.", 2884 "md_slice", pp->name, sc->sc_name); 2885 return (EINVAL); 2886 } 2887 if (md->md_balance != sc->sc_balance) { 2888 G_MIRROR_DEBUG(1, 2889 "Invalid '%s' field on disk %s (device %s), skipping.", 2890 "md_balance", pp->name, sc->sc_name); 2891 return (EINVAL); 2892 } 2893 #if 0 2894 if (md->md_mediasize != sc->sc_mediasize) { 2895 G_MIRROR_DEBUG(1, 2896 "Invalid '%s' field on disk %s (device %s), skipping.", 2897 "md_mediasize", pp->name, sc->sc_name); 2898 return (EINVAL); 2899 } 2900 #endif 2901 if (sc->sc_mediasize > pp->mediasize) { 2902 G_MIRROR_DEBUG(1, 2903 "Invalid size of disk %s (device %s), skipping.", pp->name, 2904 sc->sc_name); 2905 return (EINVAL); 2906 } 2907 if (md->md_sectorsize != sc->sc_sectorsize) { 2908 G_MIRROR_DEBUG(1, 2909 "Invalid '%s' field on disk %s (device %s), skipping.", 2910 "md_sectorsize", pp->name, sc->sc_name); 2911 return (EINVAL); 2912 } 2913 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2914 G_MIRROR_DEBUG(1, 2915 "Invalid sector size of disk %s (device %s), skipping.", 2916 pp->name, sc->sc_name); 2917 return (EINVAL); 2918 } 2919 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2920 G_MIRROR_DEBUG(1, 2921 "Invalid device flags on disk %s (device %s), skipping.", 2922 pp->name, sc->sc_name); 2923 return (EINVAL); 2924 } 2925 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2926 G_MIRROR_DEBUG(1, 2927 "Invalid disk flags on disk %s (device %s), skipping.", 2928 pp->name, sc->sc_name); 2929 return (EINVAL); 2930 } 2931 return (0); 2932 } 2933 2934 int 2935 g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2936 struct g_mirror_metadata *md) 2937 { 2938 struct g_mirror_disk *disk; 2939 int error; 2940 2941 g_topology_assert_not(); 2942 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2943 2944 error = g_mirror_check_metadata(sc, pp, md); 2945 if (error != 0) 2946 return (error); 2947 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2948 md->md_genid < sc->sc_genid) { 2949 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2950 pp->name, sc->sc_name); 2951 return (EINVAL); 2952 } 2953 disk = g_mirror_init_disk(sc, pp, md, &error); 2954 if (disk == NULL) 2955 return (error); 2956 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2957 G_MIRROR_EVENT_WAIT); 2958 if (error != 0) 2959 return (error); 2960 if (md->md_version < G_MIRROR_VERSION) { 2961 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2962 pp->name, md->md_version, G_MIRROR_VERSION); 2963 g_mirror_update_metadata(disk); 2964 } 2965 return (0); 2966 } 2967 2968 static void 2969 g_mirror_destroy_delayed(void *arg, int flag) 2970 { 2971 struct g_mirror_softc *sc; 2972 int error; 2973 2974 if (flag == EV_CANCEL) { 2975 G_MIRROR_DEBUG(1, "Destroying canceled."); 2976 return; 2977 } 2978 sc = arg; 2979 g_topology_unlock(); 2980 sx_xlock(&sc->sc_lock); 2981 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2982 ("DESTROY flag set on %s.", sc->sc_name)); 2983 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0, 2984 ("CLOSEWAIT flag not set on %s.", sc->sc_name)); 2985 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 2986 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 2987 if (error != 0) { 2988 G_MIRROR_DEBUG(0, "Cannot destroy %s (error=%d).", 2989 sc->sc_name, error); 2990 sx_xunlock(&sc->sc_lock); 2991 } 2992 g_topology_lock(); 2993 } 2994 2995 static int 2996 g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2997 { 2998 struct g_mirror_softc *sc; 2999 int error = 0; 3000 3001 g_topology_assert(); 3002 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 3003 acw, ace); 3004 3005 sc = pp->private; 3006 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 3007 3008 g_topology_unlock(); 3009 sx_xlock(&sc->sc_lock); 3010 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 3011 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 || 3012 LIST_EMPTY(&sc->sc_disks)) { 3013 if (acr > 0 || acw > 0 || ace > 0) 3014 error = ENXIO; 3015 goto end; 3016 } 3017 sc->sc_provider_open += acr + acw + ace; 3018 if (pp->acw + acw == 0) 3019 g_mirror_idle(sc, 0); 3020 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 && 3021 sc->sc_provider_open == 0) 3022 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, sc, NULL); 3023 end: 3024 sx_xunlock(&sc->sc_lock); 3025 g_topology_lock(); 3026 return (error); 3027 } 3028 3029 struct g_geom * 3030 g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md, 3031 u_int type) 3032 { 3033 struct g_mirror_softc *sc; 3034 struct g_geom *gp; 3035 int error, timeout; 3036 3037 g_topology_assert(); 3038 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 3039 md->md_mid); 3040 3041 /* One disk is minimum. */ 3042 if (md->md_all < 1) 3043 return (NULL); 3044 /* 3045 * Action geom. 3046 */ 3047 gp = g_new_geomf(mp, "%s", md->md_name); 3048 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 3049 gp->start = g_mirror_start; 3050 gp->orphan = g_mirror_orphan; 3051 gp->access = g_mirror_access; 3052 gp->dumpconf = g_mirror_dumpconf; 3053 3054 sc->sc_type = type; 3055 sc->sc_id = md->md_mid; 3056 sc->sc_slice = md->md_slice; 3057 sc->sc_balance = md->md_balance; 3058 sc->sc_mediasize = md->md_mediasize; 3059 sc->sc_sectorsize = md->md_sectorsize; 3060 sc->sc_ndisks = md->md_all; 3061 sc->sc_flags = md->md_mflags; 3062 sc->sc_bump_id = 0; 3063 sc->sc_idle = 1; 3064 sc->sc_last_write = time_uptime; 3065 sc->sc_writes = 0; 3066 sc->sc_refcnt = 1; 3067 sx_init(&sc->sc_lock, "gmirror:lock"); 3068 TAILQ_INIT(&sc->sc_queue); 3069 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 3070 TAILQ_INIT(&sc->sc_regular_delayed); 3071 TAILQ_INIT(&sc->sc_inflight); 3072 TAILQ_INIT(&sc->sc_sync_delayed); 3073 LIST_INIT(&sc->sc_disks); 3074 TAILQ_INIT(&sc->sc_events); 3075 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 3076 callout_init(&sc->sc_callout, 1); 3077 mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF); 3078 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 3079 gp->softc = sc; 3080 sc->sc_geom = gp; 3081 sc->sc_provider = NULL; 3082 sc->sc_provider_open = 0; 3083 /* 3084 * Synchronization geom. 3085 */ 3086 gp = g_new_geomf(mp, "%s.sync", md->md_name); 3087 gp->softc = sc; 3088 gp->orphan = g_mirror_orphan; 3089 sc->sc_sync.ds_geom = gp; 3090 sc->sc_sync.ds_ndisks = 0; 3091 error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 3092 "g_mirror %s", md->md_name); 3093 if (error != 0) { 3094 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 3095 sc->sc_name); 3096 g_destroy_geom(sc->sc_sync.ds_geom); 3097 g_destroy_geom(sc->sc_geom); 3098 g_mirror_free_device(sc); 3099 return (NULL); 3100 } 3101 3102 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 3103 sc->sc_name, sc->sc_ndisks, sc->sc_id); 3104 3105 sc->sc_rootmount = root_mount_hold("GMIRROR"); 3106 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 3107 /* 3108 * Run timeout. 3109 */ 3110 timeout = g_mirror_timeout * hz; 3111 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 3112 return (sc->sc_geom); 3113 } 3114 3115 int 3116 g_mirror_destroy(struct g_mirror_softc *sc, int how) 3117 { 3118 struct g_mirror_disk *disk; 3119 3120 g_topology_assert_not(); 3121 sx_assert(&sc->sc_lock, SX_XLOCKED); 3122 3123 if (sc->sc_provider_open != 0) { 3124 switch (how) { 3125 case G_MIRROR_DESTROY_SOFT: 3126 G_MIRROR_DEBUG(1, 3127 "Device %s is still open (%d).", sc->sc_name, 3128 sc->sc_provider_open); 3129 return (EBUSY); 3130 case G_MIRROR_DESTROY_DELAYED: 3131 G_MIRROR_DEBUG(1, 3132 "Device %s will be destroyed on last close.", 3133 sc->sc_name); 3134 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 3135 if (disk->d_state == 3136 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3137 g_mirror_sync_stop(disk, 1); 3138 } 3139 } 3140 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_CLOSEWAIT; 3141 return (EBUSY); 3142 case G_MIRROR_DESTROY_HARD: 3143 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 3144 "can't be definitely removed.", sc->sc_name); 3145 } 3146 } 3147 3148 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3149 sx_xunlock(&sc->sc_lock); 3150 return (0); 3151 } 3152 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 3153 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DRAIN; 3154 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3155 sx_xunlock(&sc->sc_lock); 3156 mtx_lock(&sc->sc_queue_mtx); 3157 wakeup(sc); 3158 mtx_unlock(&sc->sc_queue_mtx); 3159 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3160 while (sc->sc_worker != NULL) 3161 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 3162 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3163 sx_xlock(&sc->sc_lock); 3164 g_mirror_destroy_device(sc); 3165 return (0); 3166 } 3167 3168 static void 3169 g_mirror_taste_orphan(struct g_consumer *cp) 3170 { 3171 3172 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3173 cp->provider->name)); 3174 } 3175 3176 static struct g_geom * 3177 g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3178 { 3179 struct g_mirror_metadata md; 3180 struct g_mirror_softc *sc; 3181 struct g_consumer *cp; 3182 struct g_geom *gp; 3183 int error; 3184 3185 g_topology_assert(); 3186 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3187 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 3188 3189 gp = g_new_geomf(mp, "mirror:taste"); 3190 /* 3191 * This orphan function should be never called. 3192 */ 3193 gp->orphan = g_mirror_taste_orphan; 3194 cp = g_new_consumer(gp); 3195 g_attach(cp, pp); 3196 error = g_mirror_read_metadata(cp, &md); 3197 g_detach(cp); 3198 g_destroy_consumer(cp); 3199 g_destroy_geom(gp); 3200 if (error != 0) 3201 return (NULL); 3202 gp = NULL; 3203 3204 if (md.md_provider[0] != '\0' && 3205 !g_compare_names(md.md_provider, pp->name)) 3206 return (NULL); 3207 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3208 return (NULL); 3209 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 3210 G_MIRROR_DEBUG(0, 3211 "Device %s: provider %s marked as inactive, skipping.", 3212 md.md_name, pp->name); 3213 return (NULL); 3214 } 3215 if (g_mirror_debug >= 2) 3216 mirror_metadata_dump(&md); 3217 3218 /* 3219 * Let's check if device already exists. 3220 */ 3221 sc = NULL; 3222 LIST_FOREACH(gp, &mp->geom, geom) { 3223 sc = gp->softc; 3224 if (sc == NULL) 3225 continue; 3226 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 3227 continue; 3228 if (sc->sc_sync.ds_geom == gp) 3229 continue; 3230 if (strcmp(md.md_name, sc->sc_name) != 0) 3231 continue; 3232 if (md.md_mid != sc->sc_id) { 3233 G_MIRROR_DEBUG(0, "Device %s already configured.", 3234 sc->sc_name); 3235 return (NULL); 3236 } 3237 break; 3238 } 3239 if (gp == NULL) { 3240 gp = g_mirror_create(mp, &md, G_MIRROR_TYPE_AUTOMATIC); 3241 if (gp == NULL) { 3242 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3243 md.md_name); 3244 return (NULL); 3245 } 3246 sc = gp->softc; 3247 } 3248 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3249 g_topology_unlock(); 3250 sx_xlock(&sc->sc_lock); 3251 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; 3252 error = g_mirror_add_disk(sc, pp, &md); 3253 if (error != 0) { 3254 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3255 pp->name, gp->name, error); 3256 if (LIST_EMPTY(&sc->sc_disks)) { 3257 g_cancel_event(sc); 3258 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3259 g_topology_lock(); 3260 return (NULL); 3261 } 3262 gp = NULL; 3263 } 3264 sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; 3265 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3266 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3267 g_topology_lock(); 3268 return (NULL); 3269 } 3270 sx_xunlock(&sc->sc_lock); 3271 g_topology_lock(); 3272 return (gp); 3273 } 3274 3275 static void 3276 g_mirror_resize(struct g_consumer *cp) 3277 { 3278 struct g_mirror_disk *disk; 3279 3280 g_topology_assert(); 3281 g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name); 3282 3283 disk = cp->private; 3284 if (disk == NULL) 3285 return; 3286 g_topology_unlock(); 3287 g_mirror_update_metadata(disk); 3288 g_topology_lock(); 3289 } 3290 3291 static int 3292 g_mirror_destroy_geom(struct gctl_req *req __unused, 3293 struct g_class *mp __unused, struct g_geom *gp) 3294 { 3295 struct g_mirror_softc *sc; 3296 int error; 3297 3298 g_topology_unlock(); 3299 sc = gp->softc; 3300 sx_xlock(&sc->sc_lock); 3301 g_cancel_event(sc); 3302 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3303 if (error != 0) 3304 sx_xunlock(&sc->sc_lock); 3305 g_topology_lock(); 3306 return (error); 3307 } 3308 3309 static void 3310 g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3311 struct g_consumer *cp, struct g_provider *pp) 3312 { 3313 struct g_mirror_softc *sc; 3314 3315 g_topology_assert(); 3316 3317 sc = gp->softc; 3318 if (sc == NULL) 3319 return; 3320 /* Skip synchronization geom. */ 3321 if (gp == sc->sc_sync.ds_geom) 3322 return; 3323 if (pp != NULL) { 3324 /* Nothing here. */ 3325 } else if (cp != NULL) { 3326 struct g_mirror_disk *disk; 3327 3328 disk = cp->private; 3329 if (disk == NULL) 3330 return; 3331 g_topology_unlock(); 3332 sx_xlock(&sc->sc_lock); 3333 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3334 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3335 sbuf_printf(sb, "%s<Synchronized>", indent); 3336 if (disk->d_sync.ds_offset == 0) 3337 sbuf_printf(sb, "0%%"); 3338 else { 3339 sbuf_printf(sb, "%u%%", 3340 (u_int)((disk->d_sync.ds_offset * 100) / 3341 sc->sc_provider->mediasize)); 3342 } 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 } 3350 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3351 disk->d_sync.ds_syncid); 3352 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3353 disk->d_genid); 3354 sbuf_printf(sb, "%s<Flags>", indent); 3355 if (disk->d_flags == 0) 3356 sbuf_printf(sb, "NONE"); 3357 else { 3358 int first = 1; 3359 3360 #define ADD_FLAG(flag, name) do { \ 3361 if ((disk->d_flags & (flag)) != 0) { \ 3362 if (!first) \ 3363 sbuf_printf(sb, ", "); \ 3364 else \ 3365 first = 0; \ 3366 sbuf_printf(sb, name); \ 3367 } \ 3368 } while (0) 3369 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3370 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3371 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3372 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3373 "SYNCHRONIZING"); 3374 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3375 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3376 #undef ADD_FLAG 3377 } 3378 sbuf_printf(sb, "</Flags>\n"); 3379 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3380 disk->d_priority); 3381 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3382 g_mirror_disk_state2str(disk->d_state)); 3383 sx_xunlock(&sc->sc_lock); 3384 g_topology_lock(); 3385 } else { 3386 g_topology_unlock(); 3387 sx_xlock(&sc->sc_lock); 3388 sbuf_printf(sb, "%s<Type>", indent); 3389 switch (sc->sc_type) { 3390 case G_MIRROR_TYPE_AUTOMATIC: 3391 sbuf_printf(sb, "AUTOMATIC"); 3392 break; 3393 case G_MIRROR_TYPE_MANUAL: 3394 sbuf_printf(sb, "MANUAL"); 3395 break; 3396 default: 3397 sbuf_printf(sb, "UNKNOWN"); 3398 break; 3399 } 3400 sbuf_printf(sb, "</Type>\n"); 3401 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3402 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3403 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3404 sbuf_printf(sb, "%s<Flags>", indent); 3405 if (sc->sc_flags == 0) 3406 sbuf_printf(sb, "NONE"); 3407 else { 3408 int first = 1; 3409 3410 #define ADD_FLAG(flag, name) do { \ 3411 if ((sc->sc_flags & (flag)) != 0) { \ 3412 if (!first) \ 3413 sbuf_printf(sb, ", "); \ 3414 else \ 3415 first = 0; \ 3416 sbuf_printf(sb, name); \ 3417 } \ 3418 } while (0) 3419 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3420 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3421 #undef ADD_FLAG 3422 } 3423 sbuf_printf(sb, "</Flags>\n"); 3424 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3425 (u_int)sc->sc_slice); 3426 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3427 balance_name(sc->sc_balance)); 3428 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3429 sc->sc_ndisks); 3430 sbuf_printf(sb, "%s<State>", indent); 3431 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3432 sbuf_printf(sb, "%s", "STARTING"); 3433 else if (sc->sc_ndisks == 3434 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3435 sbuf_printf(sb, "%s", "COMPLETE"); 3436 else 3437 sbuf_printf(sb, "%s", "DEGRADED"); 3438 sbuf_printf(sb, "</State>\n"); 3439 sx_xunlock(&sc->sc_lock); 3440 g_topology_lock(); 3441 } 3442 } 3443 3444 static void 3445 g_mirror_shutdown_post_sync(void *arg, int howto) 3446 { 3447 struct g_class *mp; 3448 struct g_geom *gp, *gp2; 3449 struct g_mirror_softc *sc; 3450 int error; 3451 3452 if (panicstr != NULL) 3453 return; 3454 3455 mp = arg; 3456 g_topology_lock(); 3457 g_mirror_shutdown = 1; 3458 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3459 if ((sc = gp->softc) == NULL) 3460 continue; 3461 /* Skip synchronization geom. */ 3462 if (gp == sc->sc_sync.ds_geom) 3463 continue; 3464 g_topology_unlock(); 3465 sx_xlock(&sc->sc_lock); 3466 g_mirror_idle(sc, -1); 3467 g_cancel_event(sc); 3468 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3469 if (error != 0) 3470 sx_xunlock(&sc->sc_lock); 3471 g_topology_lock(); 3472 } 3473 g_topology_unlock(); 3474 } 3475 3476 static void 3477 g_mirror_init(struct g_class *mp) 3478 { 3479 3480 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3481 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3482 if (g_mirror_post_sync == NULL) 3483 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3484 } 3485 3486 static void 3487 g_mirror_fini(struct g_class *mp) 3488 { 3489 3490 if (g_mirror_post_sync != NULL) 3491 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); 3492 } 3493 3494 DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3495