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_mirror_disk *d; 1340 struct g_consumer *cp; 1341 int readable; 1342 1343 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_read, 1344 bp->bio_error); 1345 1346 if (bp->bio_error != 0) { 1347 G_MIRROR_LOGREQ(0, bp, 1348 "Synchronization request failed (error=%d).", 1349 bp->bio_error); 1350 1351 /* 1352 * If there's at least one other disk from which we can 1353 * read the block, retry the request. 1354 */ 1355 readable = 0; 1356 LIST_FOREACH(d, &sc->sc_disks, d_next) 1357 if (d->d_state == G_MIRROR_DISK_STATE_ACTIVE && 1358 !(d->d_flags & G_MIRROR_DISK_FLAG_BROKEN)) 1359 readable++; 1360 1361 /* 1362 * The read error will trigger a syncid bump, so there's 1363 * no need to do that here. 1364 * 1365 * If we can retry the read from another disk, do so. 1366 * Otherwise, all we can do is kick out the new disk. 1367 */ 1368 if (readable == 0) { 1369 g_mirror_sync_request_free(disk, bp); 1370 g_mirror_event_send(disk, 1371 G_MIRROR_DISK_STATE_DISCONNECTED, 1372 G_MIRROR_EVENT_DONTWAIT); 1373 } else { 1374 g_mirror_sync_reinit(disk, bp, bp->bio_offset); 1375 goto retry_read; 1376 } 1377 return; 1378 } 1379 G_MIRROR_LOGREQ(3, bp, 1380 "Synchronization request half-finished."); 1381 bp->bio_cmd = BIO_WRITE; 1382 bp->bio_cflags = 0; 1383 cp = disk->d_consumer; 1384 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1385 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1386 cp->acr, cp->acw, cp->ace)); 1387 cp->index++; 1388 g_io_request(bp, cp); 1389 return; 1390 } 1391 case BIO_WRITE: { 1392 off_t offset; 1393 int i; 1394 1395 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_write, 1396 bp->bio_error); 1397 1398 if (bp->bio_error != 0) { 1399 G_MIRROR_LOGREQ(0, bp, 1400 "Synchronization request failed (error=%d).", 1401 bp->bio_error); 1402 g_mirror_sync_request_free(disk, bp); 1403 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1404 g_mirror_event_send(disk, 1405 G_MIRROR_DISK_STATE_DISCONNECTED, 1406 G_MIRROR_EVENT_DONTWAIT); 1407 return; 1408 } 1409 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1410 if (sync->ds_offset >= sc->sc_mediasize || 1411 sync->ds_consumer == NULL || 1412 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1413 /* Don't send more synchronization requests. */ 1414 sync->ds_inflight--; 1415 g_mirror_sync_request_free(disk, bp); 1416 if (sync->ds_inflight > 0) 1417 return; 1418 if (sync->ds_consumer == NULL || 1419 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1420 return; 1421 } 1422 /* Disk up-to-date, activate it. */ 1423 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1424 G_MIRROR_EVENT_DONTWAIT); 1425 return; 1426 } 1427 1428 /* Send next synchronization request. */ 1429 g_mirror_sync_reinit(disk, bp, sync->ds_offset); 1430 sync->ds_offset += bp->bio_length; 1431 1432 retry_read: 1433 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1434 sync->ds_consumer->index++; 1435 1436 /* 1437 * Delay the request if it is colliding with a regular request. 1438 */ 1439 if (g_mirror_regular_collision(sc, bp)) 1440 g_mirror_sync_delay(sc, bp); 1441 else 1442 g_io_request(bp, sync->ds_consumer); 1443 1444 /* Requeue delayed requests if possible. */ 1445 g_mirror_regular_release(sc); 1446 1447 /* Find the smallest offset */ 1448 offset = sc->sc_mediasize; 1449 for (i = 0; i < g_mirror_syncreqs; i++) { 1450 bp = sync->ds_bios[i]; 1451 if (bp != NULL && bp->bio_offset < offset) 1452 offset = bp->bio_offset; 1453 } 1454 if (g_mirror_sync_period > 0 && 1455 time_uptime - sync->ds_update_ts > g_mirror_sync_period) { 1456 sync->ds_offset_done = offset; 1457 g_mirror_update_metadata(disk); 1458 sync->ds_update_ts = time_uptime; 1459 } 1460 return; 1461 } 1462 default: 1463 panic("Invalid I/O request %p", bp); 1464 } 1465 } 1466 1467 static void 1468 g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1469 { 1470 struct g_mirror_disk *disk; 1471 struct g_consumer *cp; 1472 struct bio *cbp; 1473 1474 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1475 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1476 break; 1477 } 1478 if (disk == NULL) { 1479 if (bp->bio_error == 0) 1480 bp->bio_error = ENXIO; 1481 g_io_deliver(bp, bp->bio_error); 1482 return; 1483 } 1484 cbp = g_clone_bio(bp); 1485 if (cbp == NULL) { 1486 if (bp->bio_error == 0) 1487 bp->bio_error = ENOMEM; 1488 g_io_deliver(bp, bp->bio_error); 1489 return; 1490 } 1491 /* 1492 * Fill in the component buf structure. 1493 */ 1494 cp = disk->d_consumer; 1495 cbp->bio_done = g_mirror_done; 1496 cbp->bio_to = cp->provider; 1497 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1498 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1499 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1500 cp->acw, cp->ace)); 1501 cp->index++; 1502 g_io_request(cbp, cp); 1503 } 1504 1505 static void 1506 g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1507 { 1508 struct g_mirror_disk *disk; 1509 struct g_consumer *cp; 1510 struct bio *cbp; 1511 1512 disk = g_mirror_get_disk(sc); 1513 if (disk == NULL) { 1514 if (bp->bio_error == 0) 1515 bp->bio_error = ENXIO; 1516 g_io_deliver(bp, bp->bio_error); 1517 return; 1518 } 1519 cbp = g_clone_bio(bp); 1520 if (cbp == NULL) { 1521 if (bp->bio_error == 0) 1522 bp->bio_error = ENOMEM; 1523 g_io_deliver(bp, bp->bio_error); 1524 return; 1525 } 1526 /* 1527 * Fill in the component buf structure. 1528 */ 1529 cp = disk->d_consumer; 1530 cbp->bio_done = g_mirror_done; 1531 cbp->bio_to = cp->provider; 1532 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1533 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1534 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1535 cp->acw, cp->ace)); 1536 cp->index++; 1537 g_io_request(cbp, cp); 1538 } 1539 1540 #define TRACK_SIZE (1 * 1024 * 1024) 1541 #define LOAD_SCALE 256 1542 #define ABS(x) (((x) >= 0) ? (x) : (-(x))) 1543 1544 static void 1545 g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1546 { 1547 struct g_mirror_disk *disk, *dp; 1548 struct g_consumer *cp; 1549 struct bio *cbp; 1550 int prio, best; 1551 1552 /* Find a disk with the smallest load. */ 1553 disk = NULL; 1554 best = INT_MAX; 1555 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1556 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1557 continue; 1558 prio = dp->load; 1559 /* If disk head is precisely in position - highly prefer it. */ 1560 if (dp->d_last_offset == bp->bio_offset) 1561 prio -= 2 * LOAD_SCALE; 1562 else 1563 /* If disk head is close to position - prefer it. */ 1564 if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE) 1565 prio -= 1 * LOAD_SCALE; 1566 if (prio <= best) { 1567 disk = dp; 1568 best = prio; 1569 } 1570 } 1571 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1572 cbp = g_clone_bio(bp); 1573 if (cbp == NULL) { 1574 if (bp->bio_error == 0) 1575 bp->bio_error = ENOMEM; 1576 g_io_deliver(bp, bp->bio_error); 1577 return; 1578 } 1579 /* 1580 * Fill in the component buf structure. 1581 */ 1582 cp = disk->d_consumer; 1583 cbp->bio_done = g_mirror_done; 1584 cbp->bio_to = cp->provider; 1585 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1586 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1587 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1588 cp->acw, cp->ace)); 1589 cp->index++; 1590 /* Remember last head position */ 1591 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1592 /* Update loads. */ 1593 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1594 dp->load = (dp->d_consumer->index * LOAD_SCALE + 1595 dp->load * 7) / 8; 1596 } 1597 g_io_request(cbp, cp); 1598 } 1599 1600 static void 1601 g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1602 { 1603 struct bio_queue queue; 1604 struct g_mirror_disk *disk; 1605 struct g_consumer *cp; 1606 struct bio *cbp; 1607 off_t left, mod, offset, slice; 1608 u_char *data; 1609 u_int ndisks; 1610 1611 if (bp->bio_length <= sc->sc_slice) { 1612 g_mirror_request_round_robin(sc, bp); 1613 return; 1614 } 1615 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1616 slice = bp->bio_length / ndisks; 1617 mod = slice % sc->sc_provider->sectorsize; 1618 if (mod != 0) 1619 slice += sc->sc_provider->sectorsize - mod; 1620 /* 1621 * Allocate all bios before sending any request, so we can 1622 * return ENOMEM in nice and clean way. 1623 */ 1624 left = bp->bio_length; 1625 offset = bp->bio_offset; 1626 data = bp->bio_data; 1627 TAILQ_INIT(&queue); 1628 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1629 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1630 continue; 1631 cbp = g_clone_bio(bp); 1632 if (cbp == NULL) { 1633 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1634 TAILQ_REMOVE(&queue, cbp, bio_queue); 1635 g_destroy_bio(cbp); 1636 } 1637 if (bp->bio_error == 0) 1638 bp->bio_error = ENOMEM; 1639 g_io_deliver(bp, bp->bio_error); 1640 return; 1641 } 1642 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 1643 cbp->bio_done = g_mirror_done; 1644 cbp->bio_caller1 = disk; 1645 cbp->bio_to = disk->d_consumer->provider; 1646 cbp->bio_offset = offset; 1647 cbp->bio_data = data; 1648 cbp->bio_length = MIN(left, slice); 1649 left -= cbp->bio_length; 1650 if (left == 0) 1651 break; 1652 offset += cbp->bio_length; 1653 data += cbp->bio_length; 1654 } 1655 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1656 TAILQ_REMOVE(&queue, cbp, bio_queue); 1657 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1658 disk = cbp->bio_caller1; 1659 cbp->bio_caller1 = NULL; 1660 cp = disk->d_consumer; 1661 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1662 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1663 cp->acr, cp->acw, cp->ace)); 1664 disk->d_consumer->index++; 1665 g_io_request(cbp, disk->d_consumer); 1666 } 1667 } 1668 1669 static void 1670 g_mirror_register_request(struct g_mirror_softc *sc, struct bio *bp) 1671 { 1672 struct bio_queue queue; 1673 struct bio *cbp; 1674 struct g_consumer *cp; 1675 struct g_mirror_disk *disk; 1676 1677 sx_assert(&sc->sc_lock, SA_XLOCKED); 1678 1679 /* 1680 * To avoid ordering issues, if a write is deferred because of a 1681 * collision with a sync request, all I/O is deferred until that 1682 * write is initiated. 1683 */ 1684 if (bp->bio_from->geom != sc->sc_sync.ds_geom && 1685 !TAILQ_EMPTY(&sc->sc_regular_delayed)) { 1686 g_mirror_regular_delay(sc, bp); 1687 return; 1688 } 1689 1690 switch (bp->bio_cmd) { 1691 case BIO_READ: 1692 switch (sc->sc_balance) { 1693 case G_MIRROR_BALANCE_LOAD: 1694 g_mirror_request_load(sc, bp); 1695 break; 1696 case G_MIRROR_BALANCE_PREFER: 1697 g_mirror_request_prefer(sc, bp); 1698 break; 1699 case G_MIRROR_BALANCE_ROUND_ROBIN: 1700 g_mirror_request_round_robin(sc, bp); 1701 break; 1702 case G_MIRROR_BALANCE_SPLIT: 1703 g_mirror_request_split(sc, bp); 1704 break; 1705 } 1706 return; 1707 case BIO_WRITE: 1708 case BIO_DELETE: 1709 /* 1710 * Delay the request if it is colliding with a synchronization 1711 * request. 1712 */ 1713 if (g_mirror_sync_collision(sc, bp)) { 1714 g_mirror_regular_delay(sc, bp); 1715 return; 1716 } 1717 1718 if (sc->sc_idle) 1719 g_mirror_unidle(sc); 1720 else 1721 sc->sc_last_write = time_uptime; 1722 1723 /* 1724 * Bump syncid on first write. 1725 */ 1726 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1727 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1728 g_mirror_bump_syncid(sc); 1729 } 1730 1731 /* 1732 * Allocate all bios before sending any request, so we can 1733 * return ENOMEM in nice and clean way. 1734 */ 1735 TAILQ_INIT(&queue); 1736 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1737 switch (disk->d_state) { 1738 case G_MIRROR_DISK_STATE_ACTIVE: 1739 break; 1740 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1741 if (bp->bio_offset >= disk->d_sync.ds_offset) 1742 continue; 1743 break; 1744 default: 1745 continue; 1746 } 1747 if (bp->bio_cmd == BIO_DELETE && 1748 (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0) 1749 continue; 1750 cbp = g_clone_bio(bp); 1751 if (cbp == NULL) { 1752 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1753 TAILQ_REMOVE(&queue, cbp, bio_queue); 1754 g_destroy_bio(cbp); 1755 } 1756 if (bp->bio_error == 0) 1757 bp->bio_error = ENOMEM; 1758 g_io_deliver(bp, bp->bio_error); 1759 return; 1760 } 1761 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 1762 cbp->bio_done = g_mirror_done; 1763 cp = disk->d_consumer; 1764 cbp->bio_caller1 = cp; 1765 cbp->bio_to = cp->provider; 1766 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1767 ("Consumer %s not opened (r%dw%de%d).", 1768 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1769 } 1770 if (TAILQ_EMPTY(&queue)) { 1771 KASSERT(bp->bio_cmd == BIO_DELETE, 1772 ("No consumers for regular request %p", bp)); 1773 g_io_deliver(bp, EOPNOTSUPP); 1774 return; 1775 } 1776 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1777 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1778 TAILQ_REMOVE(&queue, cbp, bio_queue); 1779 cp = cbp->bio_caller1; 1780 cbp->bio_caller1 = NULL; 1781 cp->index++; 1782 sc->sc_writes++; 1783 g_io_request(cbp, cp); 1784 } 1785 /* 1786 * Put request onto inflight queue, so we can check if new 1787 * synchronization requests don't collide with it. 1788 */ 1789 TAILQ_INSERT_TAIL(&sc->sc_inflight, bp, bio_queue); 1790 return; 1791 case BIO_FLUSH: 1792 TAILQ_INIT(&queue); 1793 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1794 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1795 continue; 1796 cbp = g_clone_bio(bp); 1797 if (cbp == NULL) { 1798 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1799 TAILQ_REMOVE(&queue, cbp, bio_queue); 1800 g_destroy_bio(cbp); 1801 } 1802 if (bp->bio_error == 0) 1803 bp->bio_error = ENOMEM; 1804 g_io_deliver(bp, bp->bio_error); 1805 return; 1806 } 1807 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 1808 cbp->bio_done = g_mirror_done; 1809 cbp->bio_caller1 = disk; 1810 cbp->bio_to = disk->d_consumer->provider; 1811 } 1812 KASSERT(!TAILQ_EMPTY(&queue), 1813 ("No consumers for regular request %p", bp)); 1814 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1815 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1816 TAILQ_REMOVE(&queue, cbp, bio_queue); 1817 disk = cbp->bio_caller1; 1818 cbp->bio_caller1 = NULL; 1819 cp = disk->d_consumer; 1820 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1821 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1822 cp->acr, cp->acw, cp->ace)); 1823 cp->index++; 1824 g_io_request(cbp, cp); 1825 } 1826 break; 1827 default: 1828 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1829 bp->bio_cmd, sc->sc_name)); 1830 break; 1831 } 1832 } 1833 1834 static int 1835 g_mirror_can_destroy(struct g_mirror_softc *sc) 1836 { 1837 struct g_geom *gp; 1838 struct g_consumer *cp; 1839 1840 g_topology_assert(); 1841 gp = sc->sc_geom; 1842 if (gp->softc == NULL) 1843 return (1); 1844 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0) 1845 return (0); 1846 LIST_FOREACH(cp, &gp->consumer, consumer) { 1847 if (g_mirror_is_busy(sc, cp)) 1848 return (0); 1849 } 1850 gp = sc->sc_sync.ds_geom; 1851 LIST_FOREACH(cp, &gp->consumer, consumer) { 1852 if (g_mirror_is_busy(sc, cp)) 1853 return (0); 1854 } 1855 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1856 sc->sc_name); 1857 return (1); 1858 } 1859 1860 static int 1861 g_mirror_try_destroy(struct g_mirror_softc *sc) 1862 { 1863 1864 if (sc->sc_rootmount != NULL) { 1865 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1866 sc->sc_rootmount); 1867 root_mount_rel(sc->sc_rootmount); 1868 sc->sc_rootmount = NULL; 1869 } 1870 g_topology_lock(); 1871 if (!g_mirror_can_destroy(sc)) { 1872 g_topology_unlock(); 1873 return (0); 1874 } 1875 sc->sc_geom->softc = NULL; 1876 sc->sc_sync.ds_geom->softc = NULL; 1877 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DRAIN) != 0) { 1878 g_topology_unlock(); 1879 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1880 &sc->sc_worker); 1881 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1882 sx_xunlock(&sc->sc_lock); 1883 wakeup(&sc->sc_worker); 1884 sc->sc_worker = NULL; 1885 } else { 1886 g_topology_unlock(); 1887 g_mirror_destroy_device(sc); 1888 } 1889 return (1); 1890 } 1891 1892 /* 1893 * Worker thread. 1894 */ 1895 static void 1896 g_mirror_worker(void *arg) 1897 { 1898 struct g_mirror_softc *sc; 1899 struct g_mirror_event *ep; 1900 struct bio *bp; 1901 int timeout; 1902 1903 sc = arg; 1904 thread_lock(curthread); 1905 sched_prio(curthread, PRIBIO); 1906 thread_unlock(curthread); 1907 1908 sx_xlock(&sc->sc_lock); 1909 for (;;) { 1910 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1911 /* 1912 * First take a look at events. 1913 * This is important to handle events before any I/O requests. 1914 */ 1915 ep = g_mirror_event_first(sc); 1916 if (ep != NULL) { 1917 g_mirror_event_remove(sc, ep); 1918 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1919 /* Update only device status. */ 1920 G_MIRROR_DEBUG(3, 1921 "Running event for device %s.", 1922 sc->sc_name); 1923 ep->e_error = 0; 1924 g_mirror_update_device(sc, true); 1925 } else { 1926 /* Update disk status. */ 1927 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1928 g_mirror_get_diskname(ep->e_disk)); 1929 ep->e_error = g_mirror_update_disk(ep->e_disk, 1930 ep->e_state); 1931 if (ep->e_error == 0) 1932 g_mirror_update_device(sc, false); 1933 } 1934 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1935 KASSERT(ep->e_error == 0, 1936 ("Error cannot be handled.")); 1937 g_mirror_event_free(ep); 1938 } else { 1939 ep->e_flags |= G_MIRROR_EVENT_DONE; 1940 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1941 ep); 1942 mtx_lock(&sc->sc_events_mtx); 1943 wakeup(ep); 1944 mtx_unlock(&sc->sc_events_mtx); 1945 } 1946 if ((sc->sc_flags & 1947 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1948 if (g_mirror_try_destroy(sc)) { 1949 curthread->td_pflags &= ~TDP_GEOM; 1950 G_MIRROR_DEBUG(1, "Thread exiting."); 1951 kproc_exit(0); 1952 } 1953 } 1954 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1955 continue; 1956 } 1957 1958 /* 1959 * Check if we can mark array as CLEAN and if we can't take 1960 * how much seconds should we wait. 1961 */ 1962 timeout = g_mirror_idle(sc, -1); 1963 1964 /* 1965 * Handle I/O requests. 1966 */ 1967 mtx_lock(&sc->sc_queue_mtx); 1968 bp = TAILQ_FIRST(&sc->sc_queue); 1969 if (bp != NULL) 1970 TAILQ_REMOVE(&sc->sc_queue, bp, bio_queue); 1971 else { 1972 if ((sc->sc_flags & 1973 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1974 mtx_unlock(&sc->sc_queue_mtx); 1975 if (g_mirror_try_destroy(sc)) { 1976 curthread->td_pflags &= ~TDP_GEOM; 1977 G_MIRROR_DEBUG(1, "Thread exiting."); 1978 kproc_exit(0); 1979 } 1980 mtx_lock(&sc->sc_queue_mtx); 1981 if (!TAILQ_EMPTY(&sc->sc_queue)) { 1982 mtx_unlock(&sc->sc_queue_mtx); 1983 continue; 1984 } 1985 } 1986 if (g_mirror_event_first(sc) != NULL) { 1987 mtx_unlock(&sc->sc_queue_mtx); 1988 continue; 1989 } 1990 sx_xunlock(&sc->sc_lock); 1991 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1992 timeout * hz); 1993 sx_xlock(&sc->sc_lock); 1994 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1995 continue; 1996 } 1997 mtx_unlock(&sc->sc_queue_mtx); 1998 1999 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 2000 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 2001 /* 2002 * Handle completion of the first half (the read) of a 2003 * block synchronization operation. 2004 */ 2005 g_mirror_sync_request(sc, bp); 2006 } else if (bp->bio_to != sc->sc_provider) { 2007 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 2008 /* 2009 * Handle completion of a regular I/O request. 2010 */ 2011 g_mirror_regular_request(sc, bp); 2012 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2013 /* 2014 * Handle completion of the second half (the 2015 * write) of a block synchronization operation. 2016 */ 2017 g_mirror_sync_request(sc, bp); 2018 else { 2019 KASSERT(0, 2020 ("Invalid request cflags=0x%hx to=%s.", 2021 bp->bio_cflags, bp->bio_to->name)); 2022 } 2023 } else { 2024 /* 2025 * Initiate an I/O request. 2026 */ 2027 g_mirror_register_request(sc, bp); 2028 } 2029 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 2030 } 2031 } 2032 2033 static void 2034 g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 2035 { 2036 2037 sx_assert(&sc->sc_lock, SX_LOCKED); 2038 2039 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 2040 return; 2041 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2042 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 2043 g_mirror_get_diskname(disk), sc->sc_name); 2044 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2045 } else if (sc->sc_idle && 2046 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2047 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 2048 g_mirror_get_diskname(disk), sc->sc_name); 2049 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2050 } 2051 } 2052 2053 static void 2054 g_mirror_sync_reinit(const struct g_mirror_disk *disk, struct bio *bp, 2055 off_t offset) 2056 { 2057 void *data; 2058 int idx; 2059 2060 data = bp->bio_data; 2061 idx = (int)(uintptr_t)bp->bio_caller1; 2062 g_reset_bio(bp); 2063 2064 bp->bio_cmd = BIO_READ; 2065 bp->bio_data = data; 2066 bp->bio_done = g_mirror_sync_done; 2067 bp->bio_from = disk->d_sync.ds_consumer; 2068 bp->bio_to = disk->d_softc->sc_provider; 2069 bp->bio_caller1 = (void *)(uintptr_t)idx; 2070 bp->bio_offset = offset; 2071 bp->bio_length = MIN(MAXPHYS, 2072 disk->d_softc->sc_mediasize - bp->bio_offset); 2073 } 2074 2075 static void 2076 g_mirror_sync_start(struct g_mirror_disk *disk) 2077 { 2078 struct g_mirror_softc *sc; 2079 struct g_mirror_disk_sync *sync; 2080 struct g_consumer *cp; 2081 struct bio *bp; 2082 int error, i; 2083 2084 g_topology_assert_not(); 2085 sc = disk->d_softc; 2086 sync = &disk->d_sync; 2087 sx_assert(&sc->sc_lock, SX_LOCKED); 2088 2089 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2090 ("Disk %s is not marked for synchronization.", 2091 g_mirror_get_diskname(disk))); 2092 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2093 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 2094 sc->sc_state)); 2095 2096 sx_xunlock(&sc->sc_lock); 2097 g_topology_lock(); 2098 cp = g_new_consumer(sc->sc_sync.ds_geom); 2099 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 2100 error = g_attach(cp, sc->sc_provider); 2101 KASSERT(error == 0, 2102 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 2103 error = g_access(cp, 1, 0, 0); 2104 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 2105 g_topology_unlock(); 2106 sx_xlock(&sc->sc_lock); 2107 2108 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 2109 g_mirror_get_diskname(disk)); 2110 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0) 2111 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2112 KASSERT(sync->ds_consumer == NULL, 2113 ("Sync consumer already exists (device=%s, disk=%s).", 2114 sc->sc_name, g_mirror_get_diskname(disk))); 2115 2116 sync->ds_consumer = cp; 2117 sync->ds_consumer->private = disk; 2118 sync->ds_consumer->index = 0; 2119 2120 /* 2121 * Allocate memory for synchronization bios and initialize them. 2122 */ 2123 sync->ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 2124 M_MIRROR, M_WAITOK); 2125 for (i = 0; i < g_mirror_syncreqs; i++) { 2126 bp = g_alloc_bio(); 2127 sync->ds_bios[i] = bp; 2128 2129 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 2130 bp->bio_caller1 = (void *)(uintptr_t)i; 2131 g_mirror_sync_reinit(disk, bp, sync->ds_offset); 2132 sync->ds_offset += bp->bio_length; 2133 } 2134 2135 /* Increase the number of disks in SYNCHRONIZING state. */ 2136 sc->sc_sync.ds_ndisks++; 2137 /* Set the number of in-flight synchronization requests. */ 2138 sync->ds_inflight = g_mirror_syncreqs; 2139 2140 /* 2141 * Fire off first synchronization requests. 2142 */ 2143 for (i = 0; i < g_mirror_syncreqs; i++) { 2144 bp = sync->ds_bios[i]; 2145 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 2146 sync->ds_consumer->index++; 2147 /* 2148 * Delay the request if it is colliding with a regular request. 2149 */ 2150 if (g_mirror_regular_collision(sc, bp)) 2151 g_mirror_sync_delay(sc, bp); 2152 else 2153 g_io_request(bp, sync->ds_consumer); 2154 } 2155 } 2156 2157 /* 2158 * Stop synchronization process. 2159 * type: 0 - synchronization finished 2160 * 1 - synchronization stopped 2161 */ 2162 static void 2163 g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 2164 { 2165 struct g_mirror_softc *sc; 2166 struct g_consumer *cp; 2167 2168 g_topology_assert_not(); 2169 sc = disk->d_softc; 2170 sx_assert(&sc->sc_lock, SX_LOCKED); 2171 2172 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2173 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2174 g_mirror_disk_state2str(disk->d_state))); 2175 if (disk->d_sync.ds_consumer == NULL) 2176 return; 2177 2178 if (type == 0) { 2179 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2180 sc->sc_name, g_mirror_get_diskname(disk)); 2181 } else /* if (type == 1) */ { 2182 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2183 sc->sc_name, g_mirror_get_diskname(disk)); 2184 } 2185 g_mirror_regular_release(sc); 2186 free(disk->d_sync.ds_bios, M_MIRROR); 2187 disk->d_sync.ds_bios = NULL; 2188 cp = disk->d_sync.ds_consumer; 2189 disk->d_sync.ds_consumer = NULL; 2190 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2191 sc->sc_sync.ds_ndisks--; 2192 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2193 g_topology_lock(); 2194 g_mirror_kill_consumer(sc, cp); 2195 g_topology_unlock(); 2196 sx_xlock(&sc->sc_lock); 2197 } 2198 2199 static void 2200 g_mirror_launch_provider(struct g_mirror_softc *sc) 2201 { 2202 struct g_mirror_disk *disk; 2203 struct g_provider *pp, *dp; 2204 2205 sx_assert(&sc->sc_lock, SX_LOCKED); 2206 2207 g_topology_lock(); 2208 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 2209 pp->flags |= G_PF_DIRECT_RECEIVE; 2210 pp->mediasize = sc->sc_mediasize; 2211 pp->sectorsize = sc->sc_sectorsize; 2212 pp->stripesize = 0; 2213 pp->stripeoffset = 0; 2214 2215 /* Splitting of unmapped BIO's could work but isn't implemented now */ 2216 if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT) 2217 pp->flags |= G_PF_ACCEPT_UNMAPPED; 2218 2219 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2220 if (disk->d_consumer && disk->d_consumer->provider) { 2221 dp = disk->d_consumer->provider; 2222 if (dp->stripesize > pp->stripesize) { 2223 pp->stripesize = dp->stripesize; 2224 pp->stripeoffset = dp->stripeoffset; 2225 } 2226 /* A provider underneath us doesn't support unmapped */ 2227 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { 2228 G_MIRROR_DEBUG(0, "Cancelling unmapped " 2229 "because of %s.", dp->name); 2230 pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 2231 } 2232 } 2233 } 2234 pp->private = sc; 2235 sc->sc_refcnt++; 2236 sc->sc_provider = pp; 2237 g_error_provider(pp, 0); 2238 g_topology_unlock(); 2239 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2240 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 2241 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2242 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2243 g_mirror_sync_start(disk); 2244 } 2245 } 2246 2247 static void 2248 g_mirror_destroy_provider(struct g_mirror_softc *sc) 2249 { 2250 struct g_mirror_disk *disk; 2251 struct bio *bp; 2252 2253 g_topology_assert_not(); 2254 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2255 sc->sc_name)); 2256 2257 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2258 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2259 g_mirror_sync_stop(disk, 1); 2260 } 2261 2262 g_topology_lock(); 2263 g_error_provider(sc->sc_provider, ENXIO); 2264 mtx_lock(&sc->sc_queue_mtx); 2265 while ((bp = TAILQ_FIRST(&sc->sc_queue)) != NULL) { 2266 TAILQ_REMOVE(&sc->sc_queue, bp, bio_queue); 2267 /* 2268 * Abort any pending I/O that wasn't generated by us. 2269 * Synchronization requests and requests destined for individual 2270 * mirror components can be destroyed immediately. 2271 */ 2272 if (bp->bio_to == sc->sc_provider && 2273 bp->bio_from->geom != sc->sc_sync.ds_geom) { 2274 g_io_deliver(bp, ENXIO); 2275 } else { 2276 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2277 free(bp->bio_data, M_MIRROR); 2278 g_destroy_bio(bp); 2279 } 2280 } 2281 mtx_unlock(&sc->sc_queue_mtx); 2282 g_wither_provider(sc->sc_provider, ENXIO); 2283 sc->sc_provider = NULL; 2284 G_MIRROR_DEBUG(0, "Device %s: provider destroyed.", sc->sc_name); 2285 g_topology_unlock(); 2286 } 2287 2288 static void 2289 g_mirror_go(void *arg) 2290 { 2291 struct g_mirror_softc *sc; 2292 2293 sc = arg; 2294 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2295 g_mirror_event_send(sc, 0, 2296 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2297 } 2298 2299 static u_int 2300 g_mirror_determine_state(struct g_mirror_disk *disk) 2301 { 2302 struct g_mirror_softc *sc; 2303 u_int state; 2304 2305 sc = disk->d_softc; 2306 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2307 if ((disk->d_flags & 2308 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0 && 2309 (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 || 2310 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0)) { 2311 /* Disk does not need synchronization. */ 2312 state = G_MIRROR_DISK_STATE_ACTIVE; 2313 } else { 2314 if ((sc->sc_flags & 2315 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2316 (disk->d_flags & 2317 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2318 /* 2319 * We can start synchronization from 2320 * the stored offset. 2321 */ 2322 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2323 } else { 2324 state = G_MIRROR_DISK_STATE_STALE; 2325 } 2326 } 2327 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2328 /* 2329 * Reset all synchronization data for this disk, 2330 * because if it even was synchronized, it was 2331 * synchronized to disks with different syncid. 2332 */ 2333 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2334 disk->d_sync.ds_offset = 0; 2335 disk->d_sync.ds_offset_done = 0; 2336 disk->d_sync.ds_syncid = sc->sc_syncid; 2337 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2338 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2339 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2340 } else { 2341 state = G_MIRROR_DISK_STATE_STALE; 2342 } 2343 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2344 /* 2345 * Not good, NOT GOOD! 2346 * It means that mirror was started on stale disks 2347 * and more fresh disk just arrive. 2348 * If there were writes, mirror is broken, sorry. 2349 * I think the best choice here is don't touch 2350 * this disk and inform the user loudly. 2351 */ 2352 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2353 "disk (%s) arrives!! It will not be connected to the " 2354 "running device.", sc->sc_name, 2355 g_mirror_get_diskname(disk)); 2356 g_mirror_destroy_disk(disk); 2357 state = G_MIRROR_DISK_STATE_NONE; 2358 /* Return immediately, because disk was destroyed. */ 2359 return (state); 2360 } 2361 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2362 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2363 return (state); 2364 } 2365 2366 /* 2367 * Update device state. 2368 */ 2369 static void 2370 g_mirror_update_device(struct g_mirror_softc *sc, bool force) 2371 { 2372 struct g_mirror_disk *disk; 2373 u_int state; 2374 2375 sx_assert(&sc->sc_lock, SX_XLOCKED); 2376 2377 switch (sc->sc_state) { 2378 case G_MIRROR_DEVICE_STATE_STARTING: 2379 { 2380 struct g_mirror_disk *pdisk, *tdisk; 2381 u_int dirty, ndisks, genid, syncid; 2382 bool broken; 2383 2384 KASSERT(sc->sc_provider == NULL, 2385 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2386 /* 2387 * Are we ready? We are, if all disks are connected or 2388 * if we have any disks and 'force' is true. 2389 */ 2390 ndisks = g_mirror_ndisks(sc, -1); 2391 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2392 ; 2393 } else if (ndisks == 0) { 2394 /* 2395 * Disks went down in starting phase, so destroy 2396 * device. 2397 */ 2398 callout_drain(&sc->sc_callout); 2399 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2400 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2401 sc->sc_rootmount); 2402 root_mount_rel(sc->sc_rootmount); 2403 sc->sc_rootmount = NULL; 2404 return; 2405 } else { 2406 return; 2407 } 2408 2409 /* 2410 * Activate all disks with the biggest syncid. 2411 */ 2412 if (force) { 2413 /* 2414 * If 'force' is true, we have been called due to 2415 * timeout, so don't bother canceling timeout. 2416 */ 2417 ndisks = 0; 2418 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2419 if ((disk->d_flags & 2420 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2421 ndisks++; 2422 } 2423 } 2424 if (ndisks == 0) { 2425 /* No valid disks found, destroy device. */ 2426 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2427 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2428 __LINE__, sc->sc_rootmount); 2429 root_mount_rel(sc->sc_rootmount); 2430 sc->sc_rootmount = NULL; 2431 return; 2432 } 2433 } else { 2434 /* Cancel timeout. */ 2435 callout_drain(&sc->sc_callout); 2436 } 2437 2438 /* 2439 * Find the biggest genid. 2440 */ 2441 genid = 0; 2442 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2443 if (disk->d_genid > genid) 2444 genid = disk->d_genid; 2445 } 2446 sc->sc_genid = genid; 2447 /* 2448 * Remove all disks without the biggest genid. 2449 */ 2450 broken = false; 2451 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2452 if (disk->d_genid < genid) { 2453 G_MIRROR_DEBUG(0, 2454 "Component %s (device %s) broken, skipping.", 2455 g_mirror_get_diskname(disk), sc->sc_name); 2456 g_mirror_destroy_disk(disk); 2457 /* 2458 * Bump the syncid in case we discover a healthy 2459 * replacement disk after starting the mirror. 2460 */ 2461 broken = true; 2462 } 2463 } 2464 2465 /* 2466 * Find the biggest syncid. 2467 */ 2468 syncid = 0; 2469 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2470 if (disk->d_sync.ds_syncid > syncid) 2471 syncid = disk->d_sync.ds_syncid; 2472 } 2473 2474 /* 2475 * Here we need to look for dirty disks and if all disks 2476 * with the biggest syncid are dirty, we have to choose 2477 * one with the biggest priority and rebuild the rest. 2478 */ 2479 /* 2480 * Find the number of dirty disks with the biggest syncid. 2481 * Find the number of disks with the biggest syncid. 2482 * While here, find a disk with the biggest priority. 2483 */ 2484 dirty = ndisks = 0; 2485 pdisk = NULL; 2486 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2487 if (disk->d_sync.ds_syncid != syncid) 2488 continue; 2489 if ((disk->d_flags & 2490 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2491 continue; 2492 } 2493 ndisks++; 2494 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2495 dirty++; 2496 if (pdisk == NULL || 2497 pdisk->d_priority < disk->d_priority) { 2498 pdisk = disk; 2499 } 2500 } 2501 } 2502 if (dirty == 0) { 2503 /* No dirty disks at all, great. */ 2504 } else if (dirty == ndisks) { 2505 /* 2506 * Force synchronization for all dirty disks except one 2507 * with the biggest priority. 2508 */ 2509 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2510 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2511 "master disk for synchronization.", 2512 g_mirror_get_diskname(pdisk), sc->sc_name); 2513 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2514 if (disk->d_sync.ds_syncid != syncid) 2515 continue; 2516 if ((disk->d_flags & 2517 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2518 continue; 2519 } 2520 KASSERT((disk->d_flags & 2521 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2522 ("Disk %s isn't marked as dirty.", 2523 g_mirror_get_diskname(disk))); 2524 /* Skip the disk with the biggest priority. */ 2525 if (disk == pdisk) 2526 continue; 2527 disk->d_sync.ds_syncid = 0; 2528 } 2529 } else if (dirty < ndisks) { 2530 /* 2531 * Force synchronization for all dirty disks. 2532 * We have some non-dirty disks. 2533 */ 2534 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2535 if (disk->d_sync.ds_syncid != syncid) 2536 continue; 2537 if ((disk->d_flags & 2538 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2539 continue; 2540 } 2541 if ((disk->d_flags & 2542 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2543 continue; 2544 } 2545 disk->d_sync.ds_syncid = 0; 2546 } 2547 } 2548 2549 /* Reset hint. */ 2550 sc->sc_hint = NULL; 2551 sc->sc_syncid = syncid; 2552 if (force || broken) { 2553 /* Remember to bump syncid on first write. */ 2554 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2555 } 2556 state = G_MIRROR_DEVICE_STATE_RUNNING; 2557 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2558 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2559 g_mirror_device_state2str(state)); 2560 sc->sc_state = state; 2561 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2562 state = g_mirror_determine_state(disk); 2563 g_mirror_event_send(disk, state, 2564 G_MIRROR_EVENT_DONTWAIT); 2565 if (state == G_MIRROR_DISK_STATE_STALE) 2566 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2567 } 2568 break; 2569 } 2570 case G_MIRROR_DEVICE_STATE_RUNNING: 2571 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2572 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2573 /* 2574 * No usable disks, so destroy the device. 2575 */ 2576 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2577 break; 2578 } else if (g_mirror_ndisks(sc, 2579 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2580 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2581 /* 2582 * We have active disks, launch provider if it doesn't 2583 * exist. 2584 */ 2585 if (sc->sc_provider == NULL) 2586 g_mirror_launch_provider(sc); 2587 if (sc->sc_rootmount != NULL) { 2588 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2589 __LINE__, sc->sc_rootmount); 2590 root_mount_rel(sc->sc_rootmount); 2591 sc->sc_rootmount = NULL; 2592 } 2593 } 2594 /* 2595 * Genid should be bumped immediately, so do it here. 2596 */ 2597 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2598 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2599 g_mirror_bump_genid(sc); 2600 } 2601 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_NOW) != 0) { 2602 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID_NOW; 2603 g_mirror_bump_syncid(sc); 2604 } 2605 break; 2606 default: 2607 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2608 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2609 break; 2610 } 2611 } 2612 2613 /* 2614 * Update disk state and device state if needed. 2615 */ 2616 #define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2617 "Disk %s state changed from %s to %s (device %s).", \ 2618 g_mirror_get_diskname(disk), \ 2619 g_mirror_disk_state2str(disk->d_state), \ 2620 g_mirror_disk_state2str(state), sc->sc_name) 2621 static int 2622 g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2623 { 2624 struct g_mirror_softc *sc; 2625 2626 sc = disk->d_softc; 2627 sx_assert(&sc->sc_lock, SX_XLOCKED); 2628 2629 again: 2630 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2631 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2632 g_mirror_disk_state2str(state)); 2633 switch (state) { 2634 case G_MIRROR_DISK_STATE_NEW: 2635 /* 2636 * Possible scenarios: 2637 * 1. New disk arrive. 2638 */ 2639 /* Previous state should be NONE. */ 2640 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2641 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2642 g_mirror_disk_state2str(disk->d_state))); 2643 DISK_STATE_CHANGED(); 2644 2645 disk->d_state = state; 2646 if (LIST_EMPTY(&sc->sc_disks)) 2647 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2648 else { 2649 struct g_mirror_disk *dp; 2650 2651 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2652 if (disk->d_priority >= dp->d_priority) { 2653 LIST_INSERT_BEFORE(dp, disk, d_next); 2654 dp = NULL; 2655 break; 2656 } 2657 if (LIST_NEXT(dp, d_next) == NULL) 2658 break; 2659 } 2660 if (dp != NULL) 2661 LIST_INSERT_AFTER(dp, disk, d_next); 2662 } 2663 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2664 sc->sc_name, g_mirror_get_diskname(disk)); 2665 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2666 break; 2667 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2668 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2669 g_mirror_device_state2str(sc->sc_state), 2670 g_mirror_get_diskname(disk), 2671 g_mirror_disk_state2str(disk->d_state))); 2672 state = g_mirror_determine_state(disk); 2673 if (state != G_MIRROR_DISK_STATE_NONE) 2674 goto again; 2675 break; 2676 case G_MIRROR_DISK_STATE_ACTIVE: 2677 /* 2678 * Possible scenarios: 2679 * 1. New disk does not need synchronization. 2680 * 2. Synchronization process finished successfully. 2681 */ 2682 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2683 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2684 g_mirror_device_state2str(sc->sc_state), 2685 g_mirror_get_diskname(disk), 2686 g_mirror_disk_state2str(disk->d_state))); 2687 /* Previous state should be NEW or SYNCHRONIZING. */ 2688 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2689 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2690 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2691 g_mirror_disk_state2str(disk->d_state))); 2692 DISK_STATE_CHANGED(); 2693 2694 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2695 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2696 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2697 g_mirror_sync_stop(disk, 0); 2698 } 2699 disk->d_state = state; 2700 disk->d_sync.ds_offset = 0; 2701 disk->d_sync.ds_offset_done = 0; 2702 g_mirror_update_idle(sc, disk); 2703 g_mirror_update_metadata(disk); 2704 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2705 sc->sc_name, g_mirror_get_diskname(disk)); 2706 break; 2707 case G_MIRROR_DISK_STATE_STALE: 2708 /* 2709 * Possible scenarios: 2710 * 1. Stale disk was connected. 2711 */ 2712 /* Previous state should be NEW. */ 2713 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2714 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2715 g_mirror_disk_state2str(disk->d_state))); 2716 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2717 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2718 g_mirror_device_state2str(sc->sc_state), 2719 g_mirror_get_diskname(disk), 2720 g_mirror_disk_state2str(disk->d_state))); 2721 /* 2722 * STALE state is only possible if device is marked 2723 * NOAUTOSYNC. 2724 */ 2725 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2726 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2727 g_mirror_device_state2str(sc->sc_state), 2728 g_mirror_get_diskname(disk), 2729 g_mirror_disk_state2str(disk->d_state))); 2730 DISK_STATE_CHANGED(); 2731 2732 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2733 disk->d_state = state; 2734 g_mirror_update_metadata(disk); 2735 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2736 sc->sc_name, g_mirror_get_diskname(disk)); 2737 break; 2738 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2739 /* 2740 * Possible scenarios: 2741 * 1. Disk which needs synchronization was connected. 2742 */ 2743 /* Previous state should be NEW. */ 2744 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2745 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2746 g_mirror_disk_state2str(disk->d_state))); 2747 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2748 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2749 g_mirror_device_state2str(sc->sc_state), 2750 g_mirror_get_diskname(disk), 2751 g_mirror_disk_state2str(disk->d_state))); 2752 DISK_STATE_CHANGED(); 2753 2754 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2755 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2756 disk->d_state = state; 2757 if (sc->sc_provider != NULL) { 2758 g_mirror_sync_start(disk); 2759 g_mirror_update_metadata(disk); 2760 } 2761 break; 2762 case G_MIRROR_DISK_STATE_DISCONNECTED: 2763 /* 2764 * Possible scenarios: 2765 * 1. Device wasn't running yet, but disk disappear. 2766 * 2. Disk was active and disapppear. 2767 * 3. Disk disappear during synchronization process. 2768 */ 2769 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2770 /* 2771 * Previous state should be ACTIVE, STALE or 2772 * SYNCHRONIZING. 2773 */ 2774 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2775 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2776 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2777 ("Wrong disk state (%s, %s).", 2778 g_mirror_get_diskname(disk), 2779 g_mirror_disk_state2str(disk->d_state))); 2780 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2781 /* Previous state should be NEW. */ 2782 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2783 ("Wrong disk state (%s, %s).", 2784 g_mirror_get_diskname(disk), 2785 g_mirror_disk_state2str(disk->d_state))); 2786 /* 2787 * Reset bumping syncid if disk disappeared in STARTING 2788 * state. 2789 */ 2790 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2791 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2792 #ifdef INVARIANTS 2793 } else { 2794 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2795 sc->sc_name, 2796 g_mirror_device_state2str(sc->sc_state), 2797 g_mirror_get_diskname(disk), 2798 g_mirror_disk_state2str(disk->d_state))); 2799 #endif 2800 } 2801 DISK_STATE_CHANGED(); 2802 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2803 sc->sc_name, g_mirror_get_diskname(disk)); 2804 2805 g_mirror_destroy_disk(disk); 2806 break; 2807 case G_MIRROR_DISK_STATE_DESTROY: 2808 { 2809 int error; 2810 2811 error = g_mirror_clear_metadata(disk); 2812 if (error != 0) { 2813 G_MIRROR_DEBUG(0, 2814 "Device %s: failed to clear metadata on %s: %d.", 2815 sc->sc_name, g_mirror_get_diskname(disk), error); 2816 break; 2817 } 2818 DISK_STATE_CHANGED(); 2819 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2820 sc->sc_name, g_mirror_get_diskname(disk)); 2821 2822 g_mirror_destroy_disk(disk); 2823 sc->sc_ndisks--; 2824 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2825 g_mirror_update_metadata(disk); 2826 } 2827 break; 2828 } 2829 default: 2830 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2831 break; 2832 } 2833 return (0); 2834 } 2835 #undef DISK_STATE_CHANGED 2836 2837 int 2838 g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2839 { 2840 struct g_provider *pp; 2841 u_char *buf; 2842 int error; 2843 2844 g_topology_assert(); 2845 2846 error = g_access(cp, 1, 0, 0); 2847 if (error != 0) 2848 return (error); 2849 pp = cp->provider; 2850 g_topology_unlock(); 2851 /* Metadata are stored on last sector. */ 2852 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2853 &error); 2854 g_topology_lock(); 2855 g_access(cp, -1, 0, 0); 2856 if (buf == NULL) { 2857 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2858 cp->provider->name, error); 2859 return (error); 2860 } 2861 2862 /* Decode metadata. */ 2863 error = mirror_metadata_decode(buf, md); 2864 g_free(buf); 2865 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2866 return (EINVAL); 2867 if (md->md_version > G_MIRROR_VERSION) { 2868 G_MIRROR_DEBUG(0, 2869 "Kernel module is too old to handle metadata from %s.", 2870 cp->provider->name); 2871 return (EINVAL); 2872 } 2873 if (error != 0) { 2874 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2875 cp->provider->name); 2876 return (error); 2877 } 2878 2879 return (0); 2880 } 2881 2882 static int 2883 g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2884 struct g_mirror_metadata *md) 2885 { 2886 2887 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2888 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2889 pp->name, md->md_did); 2890 return (EEXIST); 2891 } 2892 if (md->md_all != sc->sc_ndisks) { 2893 G_MIRROR_DEBUG(1, 2894 "Invalid '%s' field on disk %s (device %s), skipping.", 2895 "md_all", pp->name, sc->sc_name); 2896 return (EINVAL); 2897 } 2898 if (md->md_slice != sc->sc_slice) { 2899 G_MIRROR_DEBUG(1, 2900 "Invalid '%s' field on disk %s (device %s), skipping.", 2901 "md_slice", pp->name, sc->sc_name); 2902 return (EINVAL); 2903 } 2904 if (md->md_balance != sc->sc_balance) { 2905 G_MIRROR_DEBUG(1, 2906 "Invalid '%s' field on disk %s (device %s), skipping.", 2907 "md_balance", pp->name, sc->sc_name); 2908 return (EINVAL); 2909 } 2910 #if 0 2911 if (md->md_mediasize != sc->sc_mediasize) { 2912 G_MIRROR_DEBUG(1, 2913 "Invalid '%s' field on disk %s (device %s), skipping.", 2914 "md_mediasize", pp->name, sc->sc_name); 2915 return (EINVAL); 2916 } 2917 #endif 2918 if (sc->sc_mediasize > pp->mediasize) { 2919 G_MIRROR_DEBUG(1, 2920 "Invalid size of disk %s (device %s), skipping.", pp->name, 2921 sc->sc_name); 2922 return (EINVAL); 2923 } 2924 if (md->md_sectorsize != sc->sc_sectorsize) { 2925 G_MIRROR_DEBUG(1, 2926 "Invalid '%s' field on disk %s (device %s), skipping.", 2927 "md_sectorsize", pp->name, sc->sc_name); 2928 return (EINVAL); 2929 } 2930 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2931 G_MIRROR_DEBUG(1, 2932 "Invalid sector size of disk %s (device %s), skipping.", 2933 pp->name, sc->sc_name); 2934 return (EINVAL); 2935 } 2936 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2937 G_MIRROR_DEBUG(1, 2938 "Invalid device flags on disk %s (device %s), skipping.", 2939 pp->name, sc->sc_name); 2940 return (EINVAL); 2941 } 2942 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2943 G_MIRROR_DEBUG(1, 2944 "Invalid disk flags on disk %s (device %s), skipping.", 2945 pp->name, sc->sc_name); 2946 return (EINVAL); 2947 } 2948 return (0); 2949 } 2950 2951 int 2952 g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2953 struct g_mirror_metadata *md) 2954 { 2955 struct g_mirror_disk *disk; 2956 int error; 2957 2958 g_topology_assert_not(); 2959 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2960 2961 error = g_mirror_check_metadata(sc, pp, md); 2962 if (error != 0) 2963 return (error); 2964 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2965 md->md_genid < sc->sc_genid) { 2966 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2967 pp->name, sc->sc_name); 2968 return (EINVAL); 2969 } 2970 disk = g_mirror_init_disk(sc, pp, md, &error); 2971 if (disk == NULL) 2972 return (error); 2973 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2974 G_MIRROR_EVENT_WAIT); 2975 if (error != 0) 2976 return (error); 2977 if (md->md_version < G_MIRROR_VERSION) { 2978 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2979 pp->name, md->md_version, G_MIRROR_VERSION); 2980 g_mirror_update_metadata(disk); 2981 } 2982 return (0); 2983 } 2984 2985 static void 2986 g_mirror_destroy_delayed(void *arg, int flag) 2987 { 2988 struct g_mirror_softc *sc; 2989 int error; 2990 2991 if (flag == EV_CANCEL) { 2992 G_MIRROR_DEBUG(1, "Destroying canceled."); 2993 return; 2994 } 2995 sc = arg; 2996 g_topology_unlock(); 2997 sx_xlock(&sc->sc_lock); 2998 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2999 ("DESTROY flag set on %s.", sc->sc_name)); 3000 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0, 3001 ("CLOSEWAIT flag not set on %s.", sc->sc_name)); 3002 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 3003 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 3004 if (error != 0) { 3005 G_MIRROR_DEBUG(0, "Cannot destroy %s (error=%d).", 3006 sc->sc_name, error); 3007 sx_xunlock(&sc->sc_lock); 3008 } 3009 g_topology_lock(); 3010 } 3011 3012 static int 3013 g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 3014 { 3015 struct g_mirror_softc *sc; 3016 int error = 0; 3017 3018 g_topology_assert(); 3019 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 3020 acw, ace); 3021 3022 sc = pp->private; 3023 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 3024 3025 g_topology_unlock(); 3026 sx_xlock(&sc->sc_lock); 3027 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 3028 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 || 3029 LIST_EMPTY(&sc->sc_disks)) { 3030 if (acr > 0 || acw > 0 || ace > 0) 3031 error = ENXIO; 3032 goto end; 3033 } 3034 sc->sc_provider_open += acr + acw + ace; 3035 if (pp->acw + acw == 0) 3036 g_mirror_idle(sc, 0); 3037 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 && 3038 sc->sc_provider_open == 0) 3039 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, sc, NULL); 3040 end: 3041 sx_xunlock(&sc->sc_lock); 3042 g_topology_lock(); 3043 return (error); 3044 } 3045 3046 struct g_geom * 3047 g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md, 3048 u_int type) 3049 { 3050 struct g_mirror_softc *sc; 3051 struct g_geom *gp; 3052 int error, timeout; 3053 3054 g_topology_assert(); 3055 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 3056 md->md_mid); 3057 3058 /* One disk is minimum. */ 3059 if (md->md_all < 1) 3060 return (NULL); 3061 /* 3062 * Action geom. 3063 */ 3064 gp = g_new_geomf(mp, "%s", md->md_name); 3065 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 3066 gp->start = g_mirror_start; 3067 gp->orphan = g_mirror_orphan; 3068 gp->access = g_mirror_access; 3069 gp->dumpconf = g_mirror_dumpconf; 3070 3071 sc->sc_type = type; 3072 sc->sc_id = md->md_mid; 3073 sc->sc_slice = md->md_slice; 3074 sc->sc_balance = md->md_balance; 3075 sc->sc_mediasize = md->md_mediasize; 3076 sc->sc_sectorsize = md->md_sectorsize; 3077 sc->sc_ndisks = md->md_all; 3078 sc->sc_flags = md->md_mflags; 3079 sc->sc_bump_id = 0; 3080 sc->sc_idle = 1; 3081 sc->sc_last_write = time_uptime; 3082 sc->sc_writes = 0; 3083 sc->sc_refcnt = 1; 3084 sx_init(&sc->sc_lock, "gmirror:lock"); 3085 TAILQ_INIT(&sc->sc_queue); 3086 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 3087 TAILQ_INIT(&sc->sc_regular_delayed); 3088 TAILQ_INIT(&sc->sc_inflight); 3089 TAILQ_INIT(&sc->sc_sync_delayed); 3090 LIST_INIT(&sc->sc_disks); 3091 TAILQ_INIT(&sc->sc_events); 3092 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 3093 callout_init(&sc->sc_callout, 1); 3094 mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF); 3095 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 3096 gp->softc = sc; 3097 sc->sc_geom = gp; 3098 sc->sc_provider = NULL; 3099 sc->sc_provider_open = 0; 3100 /* 3101 * Synchronization geom. 3102 */ 3103 gp = g_new_geomf(mp, "%s.sync", md->md_name); 3104 gp->softc = sc; 3105 gp->orphan = g_mirror_orphan; 3106 sc->sc_sync.ds_geom = gp; 3107 sc->sc_sync.ds_ndisks = 0; 3108 error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 3109 "g_mirror %s", md->md_name); 3110 if (error != 0) { 3111 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 3112 sc->sc_name); 3113 g_destroy_geom(sc->sc_sync.ds_geom); 3114 g_destroy_geom(sc->sc_geom); 3115 g_mirror_free_device(sc); 3116 return (NULL); 3117 } 3118 3119 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 3120 sc->sc_name, sc->sc_ndisks, sc->sc_id); 3121 3122 sc->sc_rootmount = root_mount_hold("GMIRROR"); 3123 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 3124 /* 3125 * Run timeout. 3126 */ 3127 timeout = g_mirror_timeout * hz; 3128 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 3129 return (sc->sc_geom); 3130 } 3131 3132 int 3133 g_mirror_destroy(struct g_mirror_softc *sc, int how) 3134 { 3135 struct g_mirror_disk *disk; 3136 3137 g_topology_assert_not(); 3138 sx_assert(&sc->sc_lock, SX_XLOCKED); 3139 3140 if (sc->sc_provider_open != 0) { 3141 switch (how) { 3142 case G_MIRROR_DESTROY_SOFT: 3143 G_MIRROR_DEBUG(1, 3144 "Device %s is still open (%d).", sc->sc_name, 3145 sc->sc_provider_open); 3146 return (EBUSY); 3147 case G_MIRROR_DESTROY_DELAYED: 3148 G_MIRROR_DEBUG(1, 3149 "Device %s will be destroyed on last close.", 3150 sc->sc_name); 3151 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 3152 if (disk->d_state == 3153 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3154 g_mirror_sync_stop(disk, 1); 3155 } 3156 } 3157 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_CLOSEWAIT; 3158 return (EBUSY); 3159 case G_MIRROR_DESTROY_HARD: 3160 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 3161 "can't be definitely removed.", sc->sc_name); 3162 } 3163 } 3164 3165 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3166 sx_xunlock(&sc->sc_lock); 3167 return (0); 3168 } 3169 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 3170 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DRAIN; 3171 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3172 sx_xunlock(&sc->sc_lock); 3173 mtx_lock(&sc->sc_queue_mtx); 3174 wakeup(sc); 3175 mtx_unlock(&sc->sc_queue_mtx); 3176 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3177 while (sc->sc_worker != NULL) 3178 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 3179 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3180 sx_xlock(&sc->sc_lock); 3181 g_mirror_destroy_device(sc); 3182 return (0); 3183 } 3184 3185 static void 3186 g_mirror_taste_orphan(struct g_consumer *cp) 3187 { 3188 3189 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3190 cp->provider->name)); 3191 } 3192 3193 static struct g_geom * 3194 g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3195 { 3196 struct g_mirror_metadata md; 3197 struct g_mirror_softc *sc; 3198 struct g_consumer *cp; 3199 struct g_geom *gp; 3200 int error; 3201 3202 g_topology_assert(); 3203 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3204 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 3205 3206 gp = g_new_geomf(mp, "mirror:taste"); 3207 /* 3208 * This orphan function should be never called. 3209 */ 3210 gp->orphan = g_mirror_taste_orphan; 3211 cp = g_new_consumer(gp); 3212 g_attach(cp, pp); 3213 error = g_mirror_read_metadata(cp, &md); 3214 g_detach(cp); 3215 g_destroy_consumer(cp); 3216 g_destroy_geom(gp); 3217 if (error != 0) 3218 return (NULL); 3219 gp = NULL; 3220 3221 if (md.md_provider[0] != '\0' && 3222 !g_compare_names(md.md_provider, pp->name)) 3223 return (NULL); 3224 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3225 return (NULL); 3226 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 3227 G_MIRROR_DEBUG(0, 3228 "Device %s: provider %s marked as inactive, skipping.", 3229 md.md_name, pp->name); 3230 return (NULL); 3231 } 3232 if (g_mirror_debug >= 2) 3233 mirror_metadata_dump(&md); 3234 3235 /* 3236 * Let's check if device already exists. 3237 */ 3238 sc = NULL; 3239 LIST_FOREACH(gp, &mp->geom, geom) { 3240 sc = gp->softc; 3241 if (sc == NULL) 3242 continue; 3243 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 3244 continue; 3245 if (sc->sc_sync.ds_geom == gp) 3246 continue; 3247 if (strcmp(md.md_name, sc->sc_name) != 0) 3248 continue; 3249 if (md.md_mid != sc->sc_id) { 3250 G_MIRROR_DEBUG(0, "Device %s already configured.", 3251 sc->sc_name); 3252 return (NULL); 3253 } 3254 break; 3255 } 3256 if (gp == NULL) { 3257 gp = g_mirror_create(mp, &md, G_MIRROR_TYPE_AUTOMATIC); 3258 if (gp == NULL) { 3259 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3260 md.md_name); 3261 return (NULL); 3262 } 3263 sc = gp->softc; 3264 } 3265 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3266 g_topology_unlock(); 3267 sx_xlock(&sc->sc_lock); 3268 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; 3269 error = g_mirror_add_disk(sc, pp, &md); 3270 if (error != 0) { 3271 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3272 pp->name, gp->name, error); 3273 if (LIST_EMPTY(&sc->sc_disks)) { 3274 g_cancel_event(sc); 3275 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3276 g_topology_lock(); 3277 return (NULL); 3278 } 3279 gp = NULL; 3280 } 3281 sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; 3282 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3283 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3284 g_topology_lock(); 3285 return (NULL); 3286 } 3287 sx_xunlock(&sc->sc_lock); 3288 g_topology_lock(); 3289 return (gp); 3290 } 3291 3292 static void 3293 g_mirror_resize(struct g_consumer *cp) 3294 { 3295 struct g_mirror_disk *disk; 3296 3297 g_topology_assert(); 3298 g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name); 3299 3300 disk = cp->private; 3301 if (disk == NULL) 3302 return; 3303 g_topology_unlock(); 3304 g_mirror_update_metadata(disk); 3305 g_topology_lock(); 3306 } 3307 3308 static int 3309 g_mirror_destroy_geom(struct gctl_req *req __unused, 3310 struct g_class *mp __unused, struct g_geom *gp) 3311 { 3312 struct g_mirror_softc *sc; 3313 int error; 3314 3315 g_topology_unlock(); 3316 sc = gp->softc; 3317 sx_xlock(&sc->sc_lock); 3318 g_cancel_event(sc); 3319 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3320 if (error != 0) 3321 sx_xunlock(&sc->sc_lock); 3322 g_topology_lock(); 3323 return (error); 3324 } 3325 3326 static void 3327 g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3328 struct g_consumer *cp, struct g_provider *pp) 3329 { 3330 struct g_mirror_softc *sc; 3331 3332 g_topology_assert(); 3333 3334 sc = gp->softc; 3335 if (sc == NULL) 3336 return; 3337 /* Skip synchronization geom. */ 3338 if (gp == sc->sc_sync.ds_geom) 3339 return; 3340 if (pp != NULL) { 3341 /* Nothing here. */ 3342 } else if (cp != NULL) { 3343 struct g_mirror_disk *disk; 3344 3345 disk = cp->private; 3346 if (disk == NULL) 3347 return; 3348 g_topology_unlock(); 3349 sx_xlock(&sc->sc_lock); 3350 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3351 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3352 sbuf_printf(sb, "%s<Synchronized>", indent); 3353 if (disk->d_sync.ds_offset == 0) 3354 sbuf_printf(sb, "0%%"); 3355 else { 3356 sbuf_printf(sb, "%u%%", 3357 (u_int)((disk->d_sync.ds_offset * 100) / 3358 sc->sc_provider->mediasize)); 3359 } 3360 sbuf_printf(sb, "</Synchronized>\n"); 3361 if (disk->d_sync.ds_offset > 0) { 3362 sbuf_printf(sb, "%s<BytesSynced>%jd" 3363 "</BytesSynced>\n", indent, 3364 (intmax_t)disk->d_sync.ds_offset); 3365 } 3366 } 3367 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3368 disk->d_sync.ds_syncid); 3369 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3370 disk->d_genid); 3371 sbuf_printf(sb, "%s<Flags>", indent); 3372 if (disk->d_flags == 0) 3373 sbuf_printf(sb, "NONE"); 3374 else { 3375 int first = 1; 3376 3377 #define ADD_FLAG(flag, name) do { \ 3378 if ((disk->d_flags & (flag)) != 0) { \ 3379 if (!first) \ 3380 sbuf_printf(sb, ", "); \ 3381 else \ 3382 first = 0; \ 3383 sbuf_printf(sb, name); \ 3384 } \ 3385 } while (0) 3386 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3387 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3388 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3389 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3390 "SYNCHRONIZING"); 3391 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3392 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3393 #undef ADD_FLAG 3394 } 3395 sbuf_printf(sb, "</Flags>\n"); 3396 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3397 disk->d_priority); 3398 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3399 g_mirror_disk_state2str(disk->d_state)); 3400 sx_xunlock(&sc->sc_lock); 3401 g_topology_lock(); 3402 } else { 3403 g_topology_unlock(); 3404 sx_xlock(&sc->sc_lock); 3405 sbuf_printf(sb, "%s<Type>", indent); 3406 switch (sc->sc_type) { 3407 case G_MIRROR_TYPE_AUTOMATIC: 3408 sbuf_printf(sb, "AUTOMATIC"); 3409 break; 3410 case G_MIRROR_TYPE_MANUAL: 3411 sbuf_printf(sb, "MANUAL"); 3412 break; 3413 default: 3414 sbuf_printf(sb, "UNKNOWN"); 3415 break; 3416 } 3417 sbuf_printf(sb, "</Type>\n"); 3418 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3419 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3420 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3421 sbuf_printf(sb, "%s<Flags>", indent); 3422 if (sc->sc_flags == 0) 3423 sbuf_printf(sb, "NONE"); 3424 else { 3425 int first = 1; 3426 3427 #define ADD_FLAG(flag, name) do { \ 3428 if ((sc->sc_flags & (flag)) != 0) { \ 3429 if (!first) \ 3430 sbuf_printf(sb, ", "); \ 3431 else \ 3432 first = 0; \ 3433 sbuf_printf(sb, name); \ 3434 } \ 3435 } while (0) 3436 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3437 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3438 #undef ADD_FLAG 3439 } 3440 sbuf_printf(sb, "</Flags>\n"); 3441 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3442 (u_int)sc->sc_slice); 3443 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3444 balance_name(sc->sc_balance)); 3445 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3446 sc->sc_ndisks); 3447 sbuf_printf(sb, "%s<State>", indent); 3448 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3449 sbuf_printf(sb, "%s", "STARTING"); 3450 else if (sc->sc_ndisks == 3451 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3452 sbuf_printf(sb, "%s", "COMPLETE"); 3453 else 3454 sbuf_printf(sb, "%s", "DEGRADED"); 3455 sbuf_printf(sb, "</State>\n"); 3456 sx_xunlock(&sc->sc_lock); 3457 g_topology_lock(); 3458 } 3459 } 3460 3461 static void 3462 g_mirror_shutdown_post_sync(void *arg, int howto) 3463 { 3464 struct g_class *mp; 3465 struct g_geom *gp, *gp2; 3466 struct g_mirror_softc *sc; 3467 int error; 3468 3469 if (panicstr != NULL) 3470 return; 3471 3472 mp = arg; 3473 g_topology_lock(); 3474 g_mirror_shutdown = 1; 3475 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3476 if ((sc = gp->softc) == NULL) 3477 continue; 3478 /* Skip synchronization geom. */ 3479 if (gp == sc->sc_sync.ds_geom) 3480 continue; 3481 g_topology_unlock(); 3482 sx_xlock(&sc->sc_lock); 3483 g_mirror_idle(sc, -1); 3484 g_cancel_event(sc); 3485 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3486 if (error != 0) 3487 sx_xunlock(&sc->sc_lock); 3488 g_topology_lock(); 3489 } 3490 g_topology_unlock(); 3491 } 3492 3493 static void 3494 g_mirror_init(struct g_class *mp) 3495 { 3496 3497 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3498 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3499 if (g_mirror_post_sync == NULL) 3500 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3501 } 3502 3503 static void 3504 g_mirror_fini(struct g_class *mp) 3505 { 3506 3507 if (g_mirror_post_sync != NULL) 3508 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); 3509 } 3510 3511 DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3512