1 /*- 2 * Copyright (c) 2004-2006 Pawel Jakub Dawidek <pjd@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 #include <sys/cdefs.h> 28 __FBSDID("$FreeBSD$"); 29 30 #include <sys/param.h> 31 #include <sys/systm.h> 32 #include <sys/fail.h> 33 #include <sys/kernel.h> 34 #include <sys/module.h> 35 #include <sys/limits.h> 36 #include <sys/lock.h> 37 #include <sys/mutex.h> 38 #include <sys/bio.h> 39 #include <sys/sbuf.h> 40 #include <sys/sysctl.h> 41 #include <sys/malloc.h> 42 #include <sys/eventhandler.h> 43 #include <vm/uma.h> 44 #include <geom/geom.h> 45 #include <sys/proc.h> 46 #include <sys/kthread.h> 47 #include <sys/sched.h> 48 #include <geom/mirror/g_mirror.h> 49 50 FEATURE(geom_mirror, "GEOM mirroring support"); 51 52 static MALLOC_DEFINE(M_MIRROR, "mirror_data", "GEOM_MIRROR Data"); 53 54 SYSCTL_DECL(_kern_geom); 55 static SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, 56 "GEOM_MIRROR stuff"); 57 u_int g_mirror_debug = 0; 58 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RWTUN, &g_mirror_debug, 0, 59 "Debug level"); 60 static u_int g_mirror_timeout = 4; 61 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RWTUN, &g_mirror_timeout, 62 0, "Time to wait on all mirror components"); 63 static u_int g_mirror_idletime = 5; 64 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RWTUN, 65 &g_mirror_idletime, 0, "Mark components as clean when idling"); 66 static u_int g_mirror_disconnect_on_failure = 1; 67 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN, 68 &g_mirror_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 69 static u_int g_mirror_syncreqs = 2; 70 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_requests, CTLFLAG_RDTUN, 71 &g_mirror_syncreqs, 0, "Parallel synchronization I/O requests."); 72 73 #define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 74 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 75 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 76 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 77 } while (0) 78 79 static eventhandler_tag g_mirror_post_sync = NULL; 80 static int g_mirror_shutdown = 0; 81 82 static g_ctl_destroy_geom_t g_mirror_destroy_geom; 83 static g_taste_t g_mirror_taste; 84 static g_init_t g_mirror_init; 85 static g_fini_t g_mirror_fini; 86 static g_provgone_t g_mirror_providergone; 87 static g_resize_t g_mirror_resize; 88 89 struct g_class g_mirror_class = { 90 .name = G_MIRROR_CLASS_NAME, 91 .version = G_VERSION, 92 .ctlreq = g_mirror_config, 93 .taste = g_mirror_taste, 94 .destroy_geom = g_mirror_destroy_geom, 95 .init = g_mirror_init, 96 .fini = g_mirror_fini, 97 .providergone = g_mirror_providergone, 98 .resize = g_mirror_resize 99 }; 100 101 102 static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 103 static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 104 static void g_mirror_update_device(struct g_mirror_softc *sc, bool force); 105 static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 106 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 107 static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 108 static void g_mirror_register_request(struct bio *bp); 109 static void g_mirror_sync_release(struct g_mirror_softc *sc); 110 111 112 static const char * 113 g_mirror_disk_state2str(int state) 114 { 115 116 switch (state) { 117 case G_MIRROR_DISK_STATE_NONE: 118 return ("NONE"); 119 case G_MIRROR_DISK_STATE_NEW: 120 return ("NEW"); 121 case G_MIRROR_DISK_STATE_ACTIVE: 122 return ("ACTIVE"); 123 case G_MIRROR_DISK_STATE_STALE: 124 return ("STALE"); 125 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 126 return ("SYNCHRONIZING"); 127 case G_MIRROR_DISK_STATE_DISCONNECTED: 128 return ("DISCONNECTED"); 129 case G_MIRROR_DISK_STATE_DESTROY: 130 return ("DESTROY"); 131 default: 132 return ("INVALID"); 133 } 134 } 135 136 static const char * 137 g_mirror_device_state2str(int state) 138 { 139 140 switch (state) { 141 case G_MIRROR_DEVICE_STATE_STARTING: 142 return ("STARTING"); 143 case G_MIRROR_DEVICE_STATE_RUNNING: 144 return ("RUNNING"); 145 default: 146 return ("INVALID"); 147 } 148 } 149 150 static const char * 151 g_mirror_get_diskname(struct g_mirror_disk *disk) 152 { 153 154 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 155 return ("[unknown]"); 156 return (disk->d_name); 157 } 158 159 /* 160 * --- Events handling functions --- 161 * Events in geom_mirror are used to maintain disks and device status 162 * from one thread to simplify locking. 163 */ 164 static void 165 g_mirror_event_free(struct g_mirror_event *ep) 166 { 167 168 free(ep, M_MIRROR); 169 } 170 171 int 172 g_mirror_event_send(void *arg, int state, int flags) 173 { 174 struct g_mirror_softc *sc; 175 struct g_mirror_disk *disk; 176 struct g_mirror_event *ep; 177 int error; 178 179 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 180 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 181 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 182 disk = NULL; 183 sc = arg; 184 } else { 185 disk = arg; 186 sc = disk->d_softc; 187 } 188 ep->e_disk = disk; 189 ep->e_state = state; 190 ep->e_flags = flags; 191 ep->e_error = 0; 192 mtx_lock(&sc->sc_events_mtx); 193 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 194 mtx_unlock(&sc->sc_events_mtx); 195 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 196 mtx_lock(&sc->sc_queue_mtx); 197 wakeup(sc); 198 mtx_unlock(&sc->sc_queue_mtx); 199 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 200 return (0); 201 sx_assert(&sc->sc_lock, SX_XLOCKED); 202 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 203 sx_xunlock(&sc->sc_lock); 204 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 205 mtx_lock(&sc->sc_events_mtx); 206 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 207 hz * 5); 208 } 209 error = ep->e_error; 210 g_mirror_event_free(ep); 211 sx_xlock(&sc->sc_lock); 212 return (error); 213 } 214 215 static struct g_mirror_event * 216 g_mirror_event_get(struct g_mirror_softc *sc) 217 { 218 struct g_mirror_event *ep; 219 220 mtx_lock(&sc->sc_events_mtx); 221 ep = TAILQ_FIRST(&sc->sc_events); 222 mtx_unlock(&sc->sc_events_mtx); 223 return (ep); 224 } 225 226 static void 227 g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 228 { 229 230 mtx_lock(&sc->sc_events_mtx); 231 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 232 mtx_unlock(&sc->sc_events_mtx); 233 } 234 235 static void 236 g_mirror_event_cancel(struct g_mirror_disk *disk) 237 { 238 struct g_mirror_softc *sc; 239 struct g_mirror_event *ep, *tmpep; 240 241 sc = disk->d_softc; 242 sx_assert(&sc->sc_lock, SX_XLOCKED); 243 244 mtx_lock(&sc->sc_events_mtx); 245 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 246 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 247 continue; 248 if (ep->e_disk != disk) 249 continue; 250 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 251 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 252 g_mirror_event_free(ep); 253 else { 254 ep->e_error = ECANCELED; 255 wakeup(ep); 256 } 257 } 258 mtx_unlock(&sc->sc_events_mtx); 259 } 260 261 /* 262 * Return the number of disks in given state. 263 * If state is equal to -1, count all connected disks. 264 */ 265 u_int 266 g_mirror_ndisks(struct g_mirror_softc *sc, int state) 267 { 268 struct g_mirror_disk *disk; 269 u_int n = 0; 270 271 sx_assert(&sc->sc_lock, SX_LOCKED); 272 273 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 274 if (state == -1 || disk->d_state == state) 275 n++; 276 } 277 return (n); 278 } 279 280 /* 281 * Find a disk in mirror by its disk ID. 282 */ 283 static struct g_mirror_disk * 284 g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 285 { 286 struct g_mirror_disk *disk; 287 288 sx_assert(&sc->sc_lock, SX_XLOCKED); 289 290 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 291 if (disk->d_id == id) 292 return (disk); 293 } 294 return (NULL); 295 } 296 297 static u_int 298 g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 299 { 300 struct bio *bp; 301 u_int nreqs = 0; 302 303 mtx_lock(&sc->sc_queue_mtx); 304 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 305 if (bp->bio_from == cp) 306 nreqs++; 307 } 308 mtx_unlock(&sc->sc_queue_mtx); 309 return (nreqs); 310 } 311 312 static int 313 g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 314 { 315 316 if (cp->index > 0) { 317 G_MIRROR_DEBUG(2, 318 "I/O requests for %s exist, can't destroy it now.", 319 cp->provider->name); 320 return (1); 321 } 322 if (g_mirror_nrequests(sc, cp) > 0) { 323 G_MIRROR_DEBUG(2, 324 "I/O requests for %s in queue, can't destroy it now.", 325 cp->provider->name); 326 return (1); 327 } 328 return (0); 329 } 330 331 static void 332 g_mirror_destroy_consumer(void *arg, int flags __unused) 333 { 334 struct g_consumer *cp; 335 336 g_topology_assert(); 337 338 cp = arg; 339 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 340 g_detach(cp); 341 g_destroy_consumer(cp); 342 } 343 344 static void 345 g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 346 { 347 struct g_provider *pp; 348 int retaste_wait; 349 350 g_topology_assert(); 351 352 cp->private = NULL; 353 if (g_mirror_is_busy(sc, cp)) 354 return; 355 pp = cp->provider; 356 retaste_wait = 0; 357 if (cp->acw == 1) { 358 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 359 retaste_wait = 1; 360 } 361 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 362 -cp->acw, -cp->ace, 0); 363 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 364 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 365 if (retaste_wait) { 366 /* 367 * After retaste event was send (inside g_access()), we can send 368 * event to detach and destroy consumer. 369 * A class, which has consumer to the given provider connected 370 * will not receive retaste event for the provider. 371 * This is the way how I ignore retaste events when I close 372 * consumers opened for write: I detach and destroy consumer 373 * after retaste event is sent. 374 */ 375 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 376 return; 377 } 378 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 379 g_detach(cp); 380 g_destroy_consumer(cp); 381 } 382 383 static int 384 g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 385 { 386 struct g_consumer *cp; 387 int error; 388 389 g_topology_assert_not(); 390 KASSERT(disk->d_consumer == NULL, 391 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 392 393 g_topology_lock(); 394 cp = g_new_consumer(disk->d_softc->sc_geom); 395 cp->flags |= G_CF_DIRECT_RECEIVE; 396 error = g_attach(cp, pp); 397 if (error != 0) { 398 g_destroy_consumer(cp); 399 g_topology_unlock(); 400 return (error); 401 } 402 error = g_access(cp, 1, 1, 1); 403 if (error != 0) { 404 g_detach(cp); 405 g_destroy_consumer(cp); 406 g_topology_unlock(); 407 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 408 pp->name, error); 409 return (error); 410 } 411 g_topology_unlock(); 412 disk->d_consumer = cp; 413 disk->d_consumer->private = disk; 414 disk->d_consumer->index = 0; 415 416 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 417 return (0); 418 } 419 420 static void 421 g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 422 { 423 424 g_topology_assert(); 425 426 if (cp == NULL) 427 return; 428 if (cp->provider != NULL) 429 g_mirror_kill_consumer(sc, cp); 430 else 431 g_destroy_consumer(cp); 432 } 433 434 /* 435 * Initialize disk. This means allocate memory, create consumer, attach it 436 * to the provider and open access (r1w1e1) to it. 437 */ 438 static struct g_mirror_disk * 439 g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 440 struct g_mirror_metadata *md, int *errorp) 441 { 442 struct g_mirror_disk *disk; 443 int i, error; 444 445 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 446 if (disk == NULL) { 447 error = ENOMEM; 448 goto fail; 449 } 450 disk->d_softc = sc; 451 error = g_mirror_connect_disk(disk, pp); 452 if (error != 0) 453 goto fail; 454 disk->d_id = md->md_did; 455 disk->d_state = G_MIRROR_DISK_STATE_NONE; 456 disk->d_priority = md->md_priority; 457 disk->d_flags = md->md_dflags; 458 error = g_getattr("GEOM::candelete", disk->d_consumer, &i); 459 if (error == 0 && i != 0) 460 disk->d_flags |= G_MIRROR_DISK_FLAG_CANDELETE; 461 if (md->md_provider[0] != '\0') 462 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 463 disk->d_sync.ds_consumer = NULL; 464 disk->d_sync.ds_offset = md->md_sync_offset; 465 disk->d_sync.ds_offset_done = md->md_sync_offset; 466 disk->d_genid = md->md_genid; 467 disk->d_sync.ds_syncid = md->md_syncid; 468 if (errorp != NULL) 469 *errorp = 0; 470 return (disk); 471 fail: 472 if (errorp != NULL) 473 *errorp = error; 474 if (disk != NULL) 475 free(disk, M_MIRROR); 476 return (NULL); 477 } 478 479 static void 480 g_mirror_destroy_disk(struct g_mirror_disk *disk) 481 { 482 struct g_mirror_softc *sc; 483 484 g_topology_assert_not(); 485 sc = disk->d_softc; 486 sx_assert(&sc->sc_lock, SX_XLOCKED); 487 488 LIST_REMOVE(disk, d_next); 489 g_mirror_event_cancel(disk); 490 if (sc->sc_hint == disk) 491 sc->sc_hint = NULL; 492 switch (disk->d_state) { 493 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 494 g_mirror_sync_stop(disk, 1); 495 /* FALLTHROUGH */ 496 case G_MIRROR_DISK_STATE_NEW: 497 case G_MIRROR_DISK_STATE_STALE: 498 case G_MIRROR_DISK_STATE_ACTIVE: 499 g_topology_lock(); 500 g_mirror_disconnect_consumer(sc, disk->d_consumer); 501 g_topology_unlock(); 502 free(disk, M_MIRROR); 503 break; 504 default: 505 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 506 g_mirror_get_diskname(disk), 507 g_mirror_disk_state2str(disk->d_state))); 508 } 509 } 510 511 static void 512 g_mirror_free_device(struct g_mirror_softc *sc) 513 { 514 515 mtx_destroy(&sc->sc_queue_mtx); 516 mtx_destroy(&sc->sc_events_mtx); 517 mtx_destroy(&sc->sc_done_mtx); 518 sx_destroy(&sc->sc_lock); 519 free(sc, M_MIRROR); 520 } 521 522 static void 523 g_mirror_providergone(struct g_provider *pp) 524 { 525 struct g_mirror_softc *sc = pp->private; 526 527 if ((--sc->sc_refcnt) == 0) 528 g_mirror_free_device(sc); 529 } 530 531 static void 532 g_mirror_destroy_device(struct g_mirror_softc *sc) 533 { 534 struct g_mirror_disk *disk; 535 struct g_mirror_event *ep; 536 struct g_geom *gp; 537 struct g_consumer *cp, *tmpcp; 538 539 g_topology_assert_not(); 540 sx_assert(&sc->sc_lock, SX_XLOCKED); 541 542 gp = sc->sc_geom; 543 if (sc->sc_provider != NULL) 544 g_mirror_destroy_provider(sc); 545 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 546 disk = LIST_FIRST(&sc->sc_disks)) { 547 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 548 g_mirror_update_metadata(disk); 549 g_mirror_destroy_disk(disk); 550 } 551 while ((ep = g_mirror_event_get(sc)) != NULL) { 552 g_mirror_event_remove(sc, ep); 553 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 554 g_mirror_event_free(ep); 555 else { 556 ep->e_error = ECANCELED; 557 ep->e_flags |= G_MIRROR_EVENT_DONE; 558 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 559 mtx_lock(&sc->sc_events_mtx); 560 wakeup(ep); 561 mtx_unlock(&sc->sc_events_mtx); 562 } 563 } 564 callout_drain(&sc->sc_callout); 565 566 g_topology_lock(); 567 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 568 g_mirror_disconnect_consumer(sc, cp); 569 } 570 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 571 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 572 g_wither_geom(gp, ENXIO); 573 sx_xunlock(&sc->sc_lock); 574 if ((--sc->sc_refcnt) == 0) 575 g_mirror_free_device(sc); 576 g_topology_unlock(); 577 } 578 579 static void 580 g_mirror_orphan(struct g_consumer *cp) 581 { 582 struct g_mirror_disk *disk; 583 584 g_topology_assert(); 585 586 disk = cp->private; 587 if (disk == NULL) 588 return; 589 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 590 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 591 G_MIRROR_EVENT_DONTWAIT); 592 } 593 594 /* 595 * Function should return the next active disk on the list. 596 * It is possible that it will be the same disk as given. 597 * If there are no active disks on list, NULL is returned. 598 */ 599 static __inline struct g_mirror_disk * 600 g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 601 { 602 struct g_mirror_disk *dp; 603 604 for (dp = LIST_NEXT(disk, d_next); dp != disk; 605 dp = LIST_NEXT(dp, d_next)) { 606 if (dp == NULL) 607 dp = LIST_FIRST(&sc->sc_disks); 608 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 609 break; 610 } 611 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 612 return (NULL); 613 return (dp); 614 } 615 616 static struct g_mirror_disk * 617 g_mirror_get_disk(struct g_mirror_softc *sc) 618 { 619 struct g_mirror_disk *disk; 620 621 if (sc->sc_hint == NULL) { 622 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 623 if (sc->sc_hint == NULL) 624 return (NULL); 625 } 626 disk = sc->sc_hint; 627 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 628 disk = g_mirror_find_next(sc, disk); 629 if (disk == NULL) 630 return (NULL); 631 } 632 sc->sc_hint = g_mirror_find_next(sc, disk); 633 return (disk); 634 } 635 636 static int 637 g_mirror_write_metadata(struct g_mirror_disk *disk, 638 struct g_mirror_metadata *md) 639 { 640 struct g_mirror_softc *sc; 641 struct g_consumer *cp; 642 off_t offset, length; 643 u_char *sector; 644 int error = 0; 645 646 g_topology_assert_not(); 647 sc = disk->d_softc; 648 sx_assert(&sc->sc_lock, SX_LOCKED); 649 650 cp = disk->d_consumer; 651 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 652 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 653 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 654 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 655 cp->acw, cp->ace)); 656 length = cp->provider->sectorsize; 657 offset = cp->provider->mediasize - length; 658 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 659 if (md != NULL && 660 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) { 661 /* 662 * Handle the case, when the size of parent provider reduced. 663 */ 664 if (offset < md->md_mediasize) 665 error = ENOSPC; 666 else 667 mirror_metadata_encode(md, sector); 668 } 669 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_metadata_write, error); 670 if (error == 0) 671 error = g_write_data(cp, offset, sector, length); 672 free(sector, M_MIRROR); 673 if (error != 0) { 674 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 675 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 676 G_MIRROR_DEBUG(0, "Cannot write metadata on %s " 677 "(device=%s, error=%d).", 678 g_mirror_get_diskname(disk), sc->sc_name, error); 679 } else { 680 G_MIRROR_DEBUG(1, "Cannot write metadata on %s " 681 "(device=%s, error=%d).", 682 g_mirror_get_diskname(disk), sc->sc_name, error); 683 } 684 if (g_mirror_disconnect_on_failure && 685 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 686 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 687 g_mirror_event_send(disk, 688 G_MIRROR_DISK_STATE_DISCONNECTED, 689 G_MIRROR_EVENT_DONTWAIT); 690 } 691 } 692 return (error); 693 } 694 695 static int 696 g_mirror_clear_metadata(struct g_mirror_disk *disk) 697 { 698 int error; 699 700 g_topology_assert_not(); 701 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 702 703 if (disk->d_softc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 704 return (0); 705 error = g_mirror_write_metadata(disk, NULL); 706 if (error == 0) { 707 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 708 g_mirror_get_diskname(disk)); 709 } else { 710 G_MIRROR_DEBUG(0, 711 "Cannot clear metadata on disk %s (error=%d).", 712 g_mirror_get_diskname(disk), error); 713 } 714 return (error); 715 } 716 717 void 718 g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 719 struct g_mirror_metadata *md) 720 { 721 722 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 723 md->md_version = G_MIRROR_VERSION; 724 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 725 md->md_mid = sc->sc_id; 726 md->md_all = sc->sc_ndisks; 727 md->md_slice = sc->sc_slice; 728 md->md_balance = sc->sc_balance; 729 md->md_genid = sc->sc_genid; 730 md->md_mediasize = sc->sc_mediasize; 731 md->md_sectorsize = sc->sc_sectorsize; 732 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 733 bzero(md->md_provider, sizeof(md->md_provider)); 734 if (disk == NULL) { 735 md->md_did = arc4random(); 736 md->md_priority = 0; 737 md->md_syncid = 0; 738 md->md_dflags = 0; 739 md->md_sync_offset = 0; 740 md->md_provsize = 0; 741 } else { 742 md->md_did = disk->d_id; 743 md->md_priority = disk->d_priority; 744 md->md_syncid = disk->d_sync.ds_syncid; 745 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 746 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 747 md->md_sync_offset = disk->d_sync.ds_offset_done; 748 else 749 md->md_sync_offset = 0; 750 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 751 strlcpy(md->md_provider, 752 disk->d_consumer->provider->name, 753 sizeof(md->md_provider)); 754 } 755 md->md_provsize = disk->d_consumer->provider->mediasize; 756 } 757 } 758 759 void 760 g_mirror_update_metadata(struct g_mirror_disk *disk) 761 { 762 struct g_mirror_softc *sc; 763 struct g_mirror_metadata md; 764 int error; 765 766 g_topology_assert_not(); 767 sc = disk->d_softc; 768 sx_assert(&sc->sc_lock, SX_LOCKED); 769 770 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 771 return; 772 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) 773 g_mirror_fill_metadata(sc, disk, &md); 774 error = g_mirror_write_metadata(disk, &md); 775 if (error == 0) { 776 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 777 g_mirror_get_diskname(disk)); 778 } else { 779 G_MIRROR_DEBUG(0, 780 "Cannot update metadata on disk %s (error=%d).", 781 g_mirror_get_diskname(disk), error); 782 } 783 } 784 785 static void 786 g_mirror_bump_syncid(struct g_mirror_softc *sc) 787 { 788 struct g_mirror_disk *disk; 789 790 g_topology_assert_not(); 791 sx_assert(&sc->sc_lock, SX_XLOCKED); 792 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 793 ("%s called with no active disks (device=%s).", __func__, 794 sc->sc_name)); 795 796 sc->sc_syncid++; 797 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 798 sc->sc_syncid); 799 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 800 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 801 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 802 disk->d_sync.ds_syncid = sc->sc_syncid; 803 g_mirror_update_metadata(disk); 804 } 805 } 806 } 807 808 static void 809 g_mirror_bump_genid(struct g_mirror_softc *sc) 810 { 811 struct g_mirror_disk *disk; 812 813 g_topology_assert_not(); 814 sx_assert(&sc->sc_lock, SX_XLOCKED); 815 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 816 ("%s called with no active disks (device=%s).", __func__, 817 sc->sc_name)); 818 819 sc->sc_genid++; 820 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 821 sc->sc_genid); 822 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 823 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 824 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 825 disk->d_genid = sc->sc_genid; 826 g_mirror_update_metadata(disk); 827 } 828 } 829 } 830 831 static int 832 g_mirror_idle(struct g_mirror_softc *sc, int acw) 833 { 834 struct g_mirror_disk *disk; 835 int timeout; 836 837 g_topology_assert_not(); 838 sx_assert(&sc->sc_lock, SX_XLOCKED); 839 840 if (sc->sc_provider == NULL) 841 return (0); 842 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 843 return (0); 844 if (sc->sc_idle) 845 return (0); 846 if (sc->sc_writes > 0) 847 return (0); 848 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 849 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); 850 if (!g_mirror_shutdown && timeout > 0) 851 return (timeout); 852 } 853 sc->sc_idle = 1; 854 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 855 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 856 continue; 857 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 858 g_mirror_get_diskname(disk), sc->sc_name); 859 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 860 g_mirror_update_metadata(disk); 861 } 862 return (0); 863 } 864 865 static void 866 g_mirror_unidle(struct g_mirror_softc *sc) 867 { 868 struct g_mirror_disk *disk; 869 870 g_topology_assert_not(); 871 sx_assert(&sc->sc_lock, SX_XLOCKED); 872 873 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 874 return; 875 sc->sc_idle = 0; 876 sc->sc_last_write = time_uptime; 877 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 878 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 879 continue; 880 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 881 g_mirror_get_diskname(disk), sc->sc_name); 882 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 883 g_mirror_update_metadata(disk); 884 } 885 } 886 887 static void 888 g_mirror_flush_done(struct bio *bp) 889 { 890 struct g_mirror_softc *sc; 891 struct bio *pbp; 892 893 pbp = bp->bio_parent; 894 sc = pbp->bio_to->private; 895 mtx_lock(&sc->sc_done_mtx); 896 if (pbp->bio_error == 0) 897 pbp->bio_error = bp->bio_error; 898 pbp->bio_completed += bp->bio_completed; 899 pbp->bio_inbed++; 900 if (pbp->bio_children == pbp->bio_inbed) { 901 mtx_unlock(&sc->sc_done_mtx); 902 g_io_deliver(pbp, pbp->bio_error); 903 } else 904 mtx_unlock(&sc->sc_done_mtx); 905 g_destroy_bio(bp); 906 } 907 908 static void 909 g_mirror_done(struct bio *bp) 910 { 911 struct g_mirror_softc *sc; 912 913 sc = bp->bio_from->geom->softc; 914 bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR; 915 mtx_lock(&sc->sc_queue_mtx); 916 bioq_insert_tail(&sc->sc_queue, bp); 917 mtx_unlock(&sc->sc_queue_mtx); 918 wakeup(sc); 919 } 920 921 static void 922 g_mirror_regular_request(struct bio *bp) 923 { 924 struct g_mirror_softc *sc; 925 struct g_mirror_disk *disk; 926 struct bio *pbp; 927 928 g_topology_assert_not(); 929 930 pbp = bp->bio_parent; 931 sc = pbp->bio_to->private; 932 bp->bio_from->index--; 933 if (bp->bio_cmd == BIO_WRITE) 934 sc->sc_writes--; 935 disk = bp->bio_from->private; 936 if (disk == NULL) { 937 g_topology_lock(); 938 g_mirror_kill_consumer(sc, bp->bio_from); 939 g_topology_unlock(); 940 } 941 942 if (bp->bio_cmd == BIO_READ) 943 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_read, 944 bp->bio_error); 945 else if (bp->bio_cmd == BIO_WRITE) 946 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_write, 947 bp->bio_error); 948 949 pbp->bio_inbed++; 950 KASSERT(pbp->bio_inbed <= pbp->bio_children, 951 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 952 pbp->bio_children)); 953 if (bp->bio_error == 0 && pbp->bio_error == 0) { 954 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 955 g_destroy_bio(bp); 956 if (pbp->bio_children == pbp->bio_inbed) { 957 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 958 pbp->bio_completed = pbp->bio_length; 959 if (pbp->bio_cmd == BIO_WRITE || 960 pbp->bio_cmd == BIO_DELETE) { 961 bioq_remove(&sc->sc_inflight, pbp); 962 /* Release delayed sync requests if possible. */ 963 g_mirror_sync_release(sc); 964 } 965 g_io_deliver(pbp, pbp->bio_error); 966 } 967 return; 968 } else if (bp->bio_error != 0) { 969 if (pbp->bio_error == 0) 970 pbp->bio_error = bp->bio_error; 971 if (disk != NULL) { 972 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 973 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 974 G_MIRROR_LOGREQ(0, bp, 975 "Request failed (error=%d).", 976 bp->bio_error); 977 } else { 978 G_MIRROR_LOGREQ(1, bp, 979 "Request failed (error=%d).", 980 bp->bio_error); 981 } 982 if (g_mirror_disconnect_on_failure && 983 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) 984 { 985 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 986 g_mirror_event_send(disk, 987 G_MIRROR_DISK_STATE_DISCONNECTED, 988 G_MIRROR_EVENT_DONTWAIT); 989 } 990 } 991 switch (pbp->bio_cmd) { 992 case BIO_DELETE: 993 case BIO_WRITE: 994 pbp->bio_inbed--; 995 pbp->bio_children--; 996 break; 997 } 998 } 999 g_destroy_bio(bp); 1000 1001 switch (pbp->bio_cmd) { 1002 case BIO_READ: 1003 if (pbp->bio_inbed < pbp->bio_children) 1004 break; 1005 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 1006 g_io_deliver(pbp, pbp->bio_error); 1007 else { 1008 pbp->bio_error = 0; 1009 mtx_lock(&sc->sc_queue_mtx); 1010 bioq_insert_tail(&sc->sc_queue, pbp); 1011 mtx_unlock(&sc->sc_queue_mtx); 1012 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1013 wakeup(sc); 1014 } 1015 break; 1016 case BIO_DELETE: 1017 case BIO_WRITE: 1018 if (pbp->bio_children == 0) { 1019 /* 1020 * All requests failed. 1021 */ 1022 } else if (pbp->bio_inbed < pbp->bio_children) { 1023 /* Do nothing. */ 1024 break; 1025 } else if (pbp->bio_children == pbp->bio_inbed) { 1026 /* Some requests succeeded. */ 1027 pbp->bio_error = 0; 1028 pbp->bio_completed = pbp->bio_length; 1029 } 1030 bioq_remove(&sc->sc_inflight, pbp); 1031 /* Release delayed sync requests if possible. */ 1032 g_mirror_sync_release(sc); 1033 g_io_deliver(pbp, pbp->bio_error); 1034 break; 1035 default: 1036 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 1037 break; 1038 } 1039 } 1040 1041 static void 1042 g_mirror_sync_done(struct bio *bp) 1043 { 1044 struct g_mirror_softc *sc; 1045 1046 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1047 sc = bp->bio_from->geom->softc; 1048 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; 1049 mtx_lock(&sc->sc_queue_mtx); 1050 bioq_insert_tail(&sc->sc_queue, bp); 1051 mtx_unlock(&sc->sc_queue_mtx); 1052 wakeup(sc); 1053 } 1054 1055 static void 1056 g_mirror_candelete(struct bio *bp) 1057 { 1058 struct g_mirror_softc *sc; 1059 struct g_mirror_disk *disk; 1060 int *val; 1061 1062 sc = bp->bio_to->private; 1063 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1064 if (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) 1065 break; 1066 } 1067 val = (int *)bp->bio_data; 1068 *val = (disk != NULL); 1069 g_io_deliver(bp, 0); 1070 } 1071 1072 static void 1073 g_mirror_kernel_dump(struct bio *bp) 1074 { 1075 struct g_mirror_softc *sc; 1076 struct g_mirror_disk *disk; 1077 struct bio *cbp; 1078 struct g_kerneldump *gkd; 1079 1080 /* 1081 * We configure dumping to the first component, because this component 1082 * will be used for reading with 'prefer' balance algorithm. 1083 * If the component with the highest priority is currently disconnected 1084 * we will not be able to read the dump after the reboot if it will be 1085 * connected and synchronized later. Can we do something better? 1086 */ 1087 sc = bp->bio_to->private; 1088 disk = LIST_FIRST(&sc->sc_disks); 1089 1090 gkd = (struct g_kerneldump *)bp->bio_data; 1091 if (gkd->length > bp->bio_to->mediasize) 1092 gkd->length = bp->bio_to->mediasize; 1093 cbp = g_clone_bio(bp); 1094 if (cbp == NULL) { 1095 g_io_deliver(bp, ENOMEM); 1096 return; 1097 } 1098 cbp->bio_done = g_std_done; 1099 g_io_request(cbp, disk->d_consumer); 1100 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1101 g_mirror_get_diskname(disk)); 1102 } 1103 1104 static void 1105 g_mirror_flush(struct g_mirror_softc *sc, struct bio *bp) 1106 { 1107 struct bio_queue_head queue; 1108 struct g_mirror_disk *disk; 1109 struct g_consumer *cp; 1110 struct bio *cbp; 1111 1112 bioq_init(&queue); 1113 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1114 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1115 continue; 1116 cbp = g_clone_bio(bp); 1117 if (cbp == NULL) { 1118 while ((cbp = bioq_takefirst(&queue)) != NULL) 1119 g_destroy_bio(cbp); 1120 if (bp->bio_error == 0) 1121 bp->bio_error = ENOMEM; 1122 g_io_deliver(bp, bp->bio_error); 1123 return; 1124 } 1125 bioq_insert_tail(&queue, cbp); 1126 cbp->bio_done = g_mirror_flush_done; 1127 cbp->bio_caller1 = disk; 1128 cbp->bio_to = disk->d_consumer->provider; 1129 } 1130 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1131 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1132 disk = cbp->bio_caller1; 1133 cbp->bio_caller1 = NULL; 1134 cp = disk->d_consumer; 1135 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1136 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1137 cp->acr, cp->acw, cp->ace)); 1138 g_io_request(cbp, disk->d_consumer); 1139 } 1140 } 1141 1142 static void 1143 g_mirror_start(struct bio *bp) 1144 { 1145 struct g_mirror_softc *sc; 1146 1147 sc = bp->bio_to->private; 1148 /* 1149 * If sc == NULL or there are no valid disks, provider's error 1150 * should be set and g_mirror_start() should not be called at all. 1151 */ 1152 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1153 ("Provider's error should be set (error=%d)(mirror=%s).", 1154 bp->bio_to->error, bp->bio_to->name)); 1155 G_MIRROR_LOGREQ(3, bp, "Request received."); 1156 1157 switch (bp->bio_cmd) { 1158 case BIO_READ: 1159 case BIO_WRITE: 1160 case BIO_DELETE: 1161 break; 1162 case BIO_FLUSH: 1163 g_mirror_flush(sc, bp); 1164 return; 1165 case BIO_GETATTR: 1166 if (!strcmp(bp->bio_attribute, "GEOM::candelete")) { 1167 g_mirror_candelete(bp); 1168 return; 1169 } else if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1170 g_mirror_kernel_dump(bp); 1171 return; 1172 } 1173 /* FALLTHROUGH */ 1174 default: 1175 g_io_deliver(bp, EOPNOTSUPP); 1176 return; 1177 } 1178 mtx_lock(&sc->sc_queue_mtx); 1179 if (bp->bio_to->error != 0) { 1180 mtx_unlock(&sc->sc_queue_mtx); 1181 g_io_deliver(bp, bp->bio_to->error); 1182 return; 1183 } 1184 bioq_insert_tail(&sc->sc_queue, bp); 1185 mtx_unlock(&sc->sc_queue_mtx); 1186 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1187 wakeup(sc); 1188 } 1189 1190 /* 1191 * Return TRUE if the given request is colliding with a in-progress 1192 * synchronization request. 1193 */ 1194 static int 1195 g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1196 { 1197 struct g_mirror_disk *disk; 1198 struct bio *sbp; 1199 off_t rstart, rend, sstart, send; 1200 u_int i; 1201 1202 if (sc->sc_sync.ds_ndisks == 0) 1203 return (0); 1204 rstart = bp->bio_offset; 1205 rend = bp->bio_offset + bp->bio_length; 1206 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1207 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1208 continue; 1209 for (i = 0; i < g_mirror_syncreqs; i++) { 1210 sbp = disk->d_sync.ds_bios[i]; 1211 if (sbp == NULL) 1212 continue; 1213 sstart = sbp->bio_offset; 1214 send = sbp->bio_offset + sbp->bio_length; 1215 if (rend > sstart && rstart < send) 1216 return (1); 1217 } 1218 } 1219 return (0); 1220 } 1221 1222 /* 1223 * Return TRUE if the given sync request is colliding with a in-progress regular 1224 * request. 1225 */ 1226 static int 1227 g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1228 { 1229 off_t rstart, rend, sstart, send; 1230 struct bio *bp; 1231 1232 if (sc->sc_sync.ds_ndisks == 0) 1233 return (0); 1234 sstart = sbp->bio_offset; 1235 send = sbp->bio_offset + sbp->bio_length; 1236 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { 1237 rstart = bp->bio_offset; 1238 rend = bp->bio_offset + bp->bio_length; 1239 if (rend > sstart && rstart < send) 1240 return (1); 1241 } 1242 return (0); 1243 } 1244 1245 /* 1246 * Puts request onto delayed queue. 1247 */ 1248 static void 1249 g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1250 { 1251 1252 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1253 bioq_insert_head(&sc->sc_regular_delayed, bp); 1254 } 1255 1256 /* 1257 * Puts synchronization request onto delayed queue. 1258 */ 1259 static void 1260 g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1261 { 1262 1263 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1264 bioq_insert_tail(&sc->sc_sync_delayed, bp); 1265 } 1266 1267 /* 1268 * Releases delayed regular requests which don't collide anymore with sync 1269 * requests. 1270 */ 1271 static void 1272 g_mirror_regular_release(struct g_mirror_softc *sc) 1273 { 1274 struct bio *bp, *bp2; 1275 1276 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { 1277 if (g_mirror_sync_collision(sc, bp)) 1278 continue; 1279 bioq_remove(&sc->sc_regular_delayed, bp); 1280 G_MIRROR_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1281 mtx_lock(&sc->sc_queue_mtx); 1282 bioq_insert_head(&sc->sc_queue, bp); 1283 mtx_unlock(&sc->sc_queue_mtx); 1284 } 1285 } 1286 1287 /* 1288 * Releases delayed sync requests which don't collide anymore with regular 1289 * requests. 1290 */ 1291 static void 1292 g_mirror_sync_release(struct g_mirror_softc *sc) 1293 { 1294 struct bio *bp, *bp2; 1295 1296 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { 1297 if (g_mirror_regular_collision(sc, bp)) 1298 continue; 1299 bioq_remove(&sc->sc_sync_delayed, bp); 1300 G_MIRROR_LOGREQ(2, bp, 1301 "Releasing delayed synchronization request."); 1302 g_io_request(bp, bp->bio_from); 1303 } 1304 } 1305 1306 /* 1307 * Free a synchronization request and clear its slot in the array. 1308 */ 1309 static void 1310 g_mirror_sync_request_free(struct g_mirror_disk *disk, struct bio *bp) 1311 { 1312 int idx; 1313 1314 if (disk != NULL && disk->d_sync.ds_bios != NULL) { 1315 idx = (int)(uintptr_t)bp->bio_caller1; 1316 KASSERT(disk->d_sync.ds_bios[idx] == bp, 1317 ("unexpected sync BIO at %p:%d", disk, idx)); 1318 disk->d_sync.ds_bios[idx] = NULL; 1319 } 1320 free(bp->bio_data, M_MIRROR); 1321 g_destroy_bio(bp); 1322 } 1323 1324 /* 1325 * Handle synchronization requests. 1326 * Every synchronization request is two-steps process: first, READ request is 1327 * send to active provider and then WRITE request (with read data) to the provider 1328 * being synchronized. When WRITE is finished, new synchronization request is 1329 * send. 1330 */ 1331 static void 1332 g_mirror_sync_request(struct bio *bp) 1333 { 1334 struct g_mirror_softc *sc; 1335 struct g_mirror_disk *disk; 1336 struct g_mirror_disk_sync *sync; 1337 1338 bp->bio_from->index--; 1339 sc = bp->bio_from->geom->softc; 1340 disk = bp->bio_from->private; 1341 if (disk == NULL) { 1342 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1343 g_topology_lock(); 1344 g_mirror_kill_consumer(sc, bp->bio_from); 1345 g_topology_unlock(); 1346 g_mirror_sync_request_free(NULL, bp); 1347 sx_xlock(&sc->sc_lock); 1348 return; 1349 } 1350 1351 /* 1352 * Synchronization request. 1353 */ 1354 switch (bp->bio_cmd) { 1355 case BIO_READ: 1356 { 1357 struct g_consumer *cp; 1358 1359 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_read, 1360 bp->bio_error); 1361 1362 if (bp->bio_error != 0) { 1363 G_MIRROR_LOGREQ(0, bp, 1364 "Synchronization request failed (error=%d).", 1365 bp->bio_error); 1366 g_mirror_sync_request_free(disk, bp); 1367 return; 1368 } 1369 G_MIRROR_LOGREQ(3, bp, 1370 "Synchronization request half-finished."); 1371 bp->bio_cmd = BIO_WRITE; 1372 bp->bio_cflags = 0; 1373 cp = disk->d_consumer; 1374 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1375 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1376 cp->acr, cp->acw, cp->ace)); 1377 cp->index++; 1378 g_io_request(bp, cp); 1379 return; 1380 } 1381 case BIO_WRITE: 1382 { 1383 off_t offset; 1384 void *data; 1385 int i, idx; 1386 1387 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_write, 1388 bp->bio_error); 1389 1390 if (bp->bio_error != 0) { 1391 G_MIRROR_LOGREQ(0, bp, 1392 "Synchronization request failed (error=%d).", 1393 bp->bio_error); 1394 g_mirror_sync_request_free(disk, bp); 1395 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1396 g_mirror_event_send(disk, 1397 G_MIRROR_DISK_STATE_DISCONNECTED, 1398 G_MIRROR_EVENT_DONTWAIT); 1399 return; 1400 } 1401 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1402 sync = &disk->d_sync; 1403 if (sync->ds_offset >= sc->sc_mediasize || 1404 sync->ds_consumer == NULL || 1405 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1406 /* Don't send more synchronization requests. */ 1407 sync->ds_inflight--; 1408 g_mirror_sync_request_free(disk, bp); 1409 if (sync->ds_inflight > 0) 1410 return; 1411 if (sync->ds_consumer == NULL || 1412 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1413 return; 1414 } 1415 /* Disk up-to-date, activate it. */ 1416 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1417 G_MIRROR_EVENT_DONTWAIT); 1418 return; 1419 } 1420 1421 /* Send next synchronization request. */ 1422 data = bp->bio_data; 1423 idx = (int)(uintptr_t)bp->bio_caller1; 1424 g_reset_bio(bp); 1425 bp->bio_cmd = BIO_READ; 1426 bp->bio_offset = sync->ds_offset; 1427 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1428 sync->ds_offset += bp->bio_length; 1429 bp->bio_done = g_mirror_sync_done; 1430 bp->bio_data = data; 1431 bp->bio_from = sync->ds_consumer; 1432 bp->bio_to = sc->sc_provider; 1433 bp->bio_caller1 = (void *)(uintptr_t)idx; 1434 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1435 sync->ds_consumer->index++; 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 /* Release 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 (sync->ds_offset_done + (MAXPHYS * 100) < offset) { 1455 /* Update offset_done on every 100 blocks. */ 1456 sync->ds_offset_done = offset; 1457 g_mirror_update_metadata(disk); 1458 } 1459 return; 1460 } 1461 default: 1462 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1463 bp->bio_cmd, sc->sc_name)); 1464 break; 1465 } 1466 } 1467 1468 static void 1469 g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1470 { 1471 struct g_mirror_disk *disk; 1472 struct g_consumer *cp; 1473 struct bio *cbp; 1474 1475 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1476 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1477 break; 1478 } 1479 if (disk == NULL) { 1480 if (bp->bio_error == 0) 1481 bp->bio_error = ENXIO; 1482 g_io_deliver(bp, bp->bio_error); 1483 return; 1484 } 1485 cbp = g_clone_bio(bp); 1486 if (cbp == NULL) { 1487 if (bp->bio_error == 0) 1488 bp->bio_error = ENOMEM; 1489 g_io_deliver(bp, bp->bio_error); 1490 return; 1491 } 1492 /* 1493 * Fill in the component buf structure. 1494 */ 1495 cp = disk->d_consumer; 1496 cbp->bio_done = g_mirror_done; 1497 cbp->bio_to = cp->provider; 1498 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1499 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1500 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1501 cp->acw, cp->ace)); 1502 cp->index++; 1503 g_io_request(cbp, cp); 1504 } 1505 1506 static void 1507 g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1508 { 1509 struct g_mirror_disk *disk; 1510 struct g_consumer *cp; 1511 struct bio *cbp; 1512 1513 disk = g_mirror_get_disk(sc); 1514 if (disk == NULL) { 1515 if (bp->bio_error == 0) 1516 bp->bio_error = ENXIO; 1517 g_io_deliver(bp, bp->bio_error); 1518 return; 1519 } 1520 cbp = g_clone_bio(bp); 1521 if (cbp == NULL) { 1522 if (bp->bio_error == 0) 1523 bp->bio_error = ENOMEM; 1524 g_io_deliver(bp, bp->bio_error); 1525 return; 1526 } 1527 /* 1528 * Fill in the component buf structure. 1529 */ 1530 cp = disk->d_consumer; 1531 cbp->bio_done = g_mirror_done; 1532 cbp->bio_to = cp->provider; 1533 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1534 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1535 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1536 cp->acw, cp->ace)); 1537 cp->index++; 1538 g_io_request(cbp, cp); 1539 } 1540 1541 #define TRACK_SIZE (1 * 1024 * 1024) 1542 #define LOAD_SCALE 256 1543 #define ABS(x) (((x) >= 0) ? (x) : (-(x))) 1544 1545 static void 1546 g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1547 { 1548 struct g_mirror_disk *disk, *dp; 1549 struct g_consumer *cp; 1550 struct bio *cbp; 1551 int prio, best; 1552 1553 /* Find a disk with the smallest load. */ 1554 disk = NULL; 1555 best = INT_MAX; 1556 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1557 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1558 continue; 1559 prio = dp->load; 1560 /* If disk head is precisely in position - highly prefer it. */ 1561 if (dp->d_last_offset == bp->bio_offset) 1562 prio -= 2 * LOAD_SCALE; 1563 else 1564 /* If disk head is close to position - prefer it. */ 1565 if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE) 1566 prio -= 1 * LOAD_SCALE; 1567 if (prio <= best) { 1568 disk = dp; 1569 best = prio; 1570 } 1571 } 1572 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1573 cbp = g_clone_bio(bp); 1574 if (cbp == NULL) { 1575 if (bp->bio_error == 0) 1576 bp->bio_error = ENOMEM; 1577 g_io_deliver(bp, bp->bio_error); 1578 return; 1579 } 1580 /* 1581 * Fill in the component buf structure. 1582 */ 1583 cp = disk->d_consumer; 1584 cbp->bio_done = g_mirror_done; 1585 cbp->bio_to = cp->provider; 1586 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1587 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1588 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1589 cp->acw, cp->ace)); 1590 cp->index++; 1591 /* Remember last head position */ 1592 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1593 /* Update loads. */ 1594 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1595 dp->load = (dp->d_consumer->index * LOAD_SCALE + 1596 dp->load * 7) / 8; 1597 } 1598 g_io_request(cbp, cp); 1599 } 1600 1601 static void 1602 g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1603 { 1604 struct bio_queue_head queue; 1605 struct g_mirror_disk *disk; 1606 struct g_consumer *cp; 1607 struct bio *cbp; 1608 off_t left, mod, offset, slice; 1609 u_char *data; 1610 u_int ndisks; 1611 1612 if (bp->bio_length <= sc->sc_slice) { 1613 g_mirror_request_round_robin(sc, bp); 1614 return; 1615 } 1616 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1617 slice = bp->bio_length / ndisks; 1618 mod = slice % sc->sc_provider->sectorsize; 1619 if (mod != 0) 1620 slice += sc->sc_provider->sectorsize - mod; 1621 /* 1622 * Allocate all bios before sending any request, so we can 1623 * return ENOMEM in nice and clean way. 1624 */ 1625 left = bp->bio_length; 1626 offset = bp->bio_offset; 1627 data = bp->bio_data; 1628 bioq_init(&queue); 1629 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1630 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1631 continue; 1632 cbp = g_clone_bio(bp); 1633 if (cbp == NULL) { 1634 while ((cbp = bioq_takefirst(&queue)) != NULL) 1635 g_destroy_bio(cbp); 1636 if (bp->bio_error == 0) 1637 bp->bio_error = ENOMEM; 1638 g_io_deliver(bp, bp->bio_error); 1639 return; 1640 } 1641 bioq_insert_tail(&queue, cbp); 1642 cbp->bio_done = g_mirror_done; 1643 cbp->bio_caller1 = disk; 1644 cbp->bio_to = disk->d_consumer->provider; 1645 cbp->bio_offset = offset; 1646 cbp->bio_data = data; 1647 cbp->bio_length = MIN(left, slice); 1648 left -= cbp->bio_length; 1649 if (left == 0) 1650 break; 1651 offset += cbp->bio_length; 1652 data += cbp->bio_length; 1653 } 1654 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1655 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1656 disk = cbp->bio_caller1; 1657 cbp->bio_caller1 = NULL; 1658 cp = disk->d_consumer; 1659 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1660 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1661 cp->acr, cp->acw, cp->ace)); 1662 disk->d_consumer->index++; 1663 g_io_request(cbp, disk->d_consumer); 1664 } 1665 } 1666 1667 static void 1668 g_mirror_register_request(struct bio *bp) 1669 { 1670 struct g_mirror_softc *sc; 1671 1672 sc = bp->bio_to->private; 1673 switch (bp->bio_cmd) { 1674 case BIO_READ: 1675 switch (sc->sc_balance) { 1676 case G_MIRROR_BALANCE_LOAD: 1677 g_mirror_request_load(sc, bp); 1678 break; 1679 case G_MIRROR_BALANCE_PREFER: 1680 g_mirror_request_prefer(sc, bp); 1681 break; 1682 case G_MIRROR_BALANCE_ROUND_ROBIN: 1683 g_mirror_request_round_robin(sc, bp); 1684 break; 1685 case G_MIRROR_BALANCE_SPLIT: 1686 g_mirror_request_split(sc, bp); 1687 break; 1688 } 1689 return; 1690 case BIO_WRITE: 1691 case BIO_DELETE: 1692 { 1693 struct g_mirror_disk *disk; 1694 struct g_mirror_disk_sync *sync; 1695 struct bio_queue_head queue; 1696 struct g_consumer *cp; 1697 struct bio *cbp; 1698 1699 /* 1700 * Delay the request if it is colliding with a synchronization 1701 * request. 1702 */ 1703 if (g_mirror_sync_collision(sc, bp)) { 1704 g_mirror_regular_delay(sc, bp); 1705 return; 1706 } 1707 1708 if (sc->sc_idle) 1709 g_mirror_unidle(sc); 1710 else 1711 sc->sc_last_write = time_uptime; 1712 1713 /* 1714 * Bump syncid on first write. 1715 */ 1716 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1717 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1718 g_mirror_bump_syncid(sc); 1719 } 1720 1721 /* 1722 * Allocate all bios before sending any request, so we can 1723 * return ENOMEM in nice and clean way. 1724 */ 1725 bioq_init(&queue); 1726 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1727 sync = &disk->d_sync; 1728 switch (disk->d_state) { 1729 case G_MIRROR_DISK_STATE_ACTIVE: 1730 break; 1731 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1732 if (bp->bio_offset >= sync->ds_offset) 1733 continue; 1734 break; 1735 default: 1736 continue; 1737 } 1738 if (bp->bio_cmd == BIO_DELETE && 1739 (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0) 1740 continue; 1741 cbp = g_clone_bio(bp); 1742 if (cbp == NULL) { 1743 while ((cbp = bioq_takefirst(&queue)) != NULL) 1744 g_destroy_bio(cbp); 1745 if (bp->bio_error == 0) 1746 bp->bio_error = ENOMEM; 1747 g_io_deliver(bp, bp->bio_error); 1748 return; 1749 } 1750 bioq_insert_tail(&queue, cbp); 1751 cbp->bio_done = g_mirror_done; 1752 cp = disk->d_consumer; 1753 cbp->bio_caller1 = cp; 1754 cbp->bio_to = cp->provider; 1755 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1756 ("Consumer %s not opened (r%dw%de%d).", 1757 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1758 } 1759 if (bioq_first(&queue) == NULL) { 1760 g_io_deliver(bp, EOPNOTSUPP); 1761 return; 1762 } 1763 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1764 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1765 cp = cbp->bio_caller1; 1766 cbp->bio_caller1 = NULL; 1767 cp->index++; 1768 sc->sc_writes++; 1769 g_io_request(cbp, cp); 1770 } 1771 /* 1772 * Put request onto inflight queue, so we can check if new 1773 * synchronization requests don't collide with it. 1774 */ 1775 bioq_insert_tail(&sc->sc_inflight, bp); 1776 return; 1777 } 1778 default: 1779 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1780 bp->bio_cmd, sc->sc_name)); 1781 break; 1782 } 1783 } 1784 1785 static int 1786 g_mirror_can_destroy(struct g_mirror_softc *sc) 1787 { 1788 struct g_geom *gp; 1789 struct g_consumer *cp; 1790 1791 g_topology_assert(); 1792 gp = sc->sc_geom; 1793 if (gp->softc == NULL) 1794 return (1); 1795 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0) 1796 return (0); 1797 LIST_FOREACH(cp, &gp->consumer, consumer) { 1798 if (g_mirror_is_busy(sc, cp)) 1799 return (0); 1800 } 1801 gp = sc->sc_sync.ds_geom; 1802 LIST_FOREACH(cp, &gp->consumer, consumer) { 1803 if (g_mirror_is_busy(sc, cp)) 1804 return (0); 1805 } 1806 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1807 sc->sc_name); 1808 return (1); 1809 } 1810 1811 static int 1812 g_mirror_try_destroy(struct g_mirror_softc *sc) 1813 { 1814 1815 if (sc->sc_rootmount != NULL) { 1816 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1817 sc->sc_rootmount); 1818 root_mount_rel(sc->sc_rootmount); 1819 sc->sc_rootmount = NULL; 1820 } 1821 g_topology_lock(); 1822 if (!g_mirror_can_destroy(sc)) { 1823 g_topology_unlock(); 1824 return (0); 1825 } 1826 sc->sc_geom->softc = NULL; 1827 sc->sc_sync.ds_geom->softc = NULL; 1828 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DRAIN) != 0) { 1829 g_topology_unlock(); 1830 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1831 &sc->sc_worker); 1832 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1833 sx_xunlock(&sc->sc_lock); 1834 wakeup(&sc->sc_worker); 1835 sc->sc_worker = NULL; 1836 } else { 1837 g_topology_unlock(); 1838 g_mirror_destroy_device(sc); 1839 } 1840 return (1); 1841 } 1842 1843 /* 1844 * Worker thread. 1845 */ 1846 static void 1847 g_mirror_worker(void *arg) 1848 { 1849 struct g_mirror_softc *sc; 1850 struct g_mirror_event *ep; 1851 struct bio *bp; 1852 int timeout; 1853 1854 sc = arg; 1855 thread_lock(curthread); 1856 sched_prio(curthread, PRIBIO); 1857 thread_unlock(curthread); 1858 1859 sx_xlock(&sc->sc_lock); 1860 for (;;) { 1861 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1862 /* 1863 * First take a look at events. 1864 * This is important to handle events before any I/O requests. 1865 */ 1866 ep = g_mirror_event_get(sc); 1867 if (ep != NULL) { 1868 g_mirror_event_remove(sc, ep); 1869 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1870 /* Update only device status. */ 1871 G_MIRROR_DEBUG(3, 1872 "Running event for device %s.", 1873 sc->sc_name); 1874 ep->e_error = 0; 1875 g_mirror_update_device(sc, true); 1876 } else { 1877 /* Update disk status. */ 1878 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1879 g_mirror_get_diskname(ep->e_disk)); 1880 ep->e_error = g_mirror_update_disk(ep->e_disk, 1881 ep->e_state); 1882 if (ep->e_error == 0) 1883 g_mirror_update_device(sc, false); 1884 } 1885 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1886 KASSERT(ep->e_error == 0, 1887 ("Error cannot be handled.")); 1888 g_mirror_event_free(ep); 1889 } else { 1890 ep->e_flags |= G_MIRROR_EVENT_DONE; 1891 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1892 ep); 1893 mtx_lock(&sc->sc_events_mtx); 1894 wakeup(ep); 1895 mtx_unlock(&sc->sc_events_mtx); 1896 } 1897 if ((sc->sc_flags & 1898 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1899 if (g_mirror_try_destroy(sc)) { 1900 curthread->td_pflags &= ~TDP_GEOM; 1901 G_MIRROR_DEBUG(1, "Thread exiting."); 1902 kproc_exit(0); 1903 } 1904 } 1905 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1906 continue; 1907 } 1908 /* 1909 * Check if we can mark array as CLEAN and if we can't take 1910 * how much seconds should we wait. 1911 */ 1912 timeout = g_mirror_idle(sc, -1); 1913 /* 1914 * Now I/O requests. 1915 */ 1916 /* Get first request from the queue. */ 1917 mtx_lock(&sc->sc_queue_mtx); 1918 bp = bioq_takefirst(&sc->sc_queue); 1919 if (bp == NULL) { 1920 if ((sc->sc_flags & 1921 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1922 mtx_unlock(&sc->sc_queue_mtx); 1923 if (g_mirror_try_destroy(sc)) { 1924 curthread->td_pflags &= ~TDP_GEOM; 1925 G_MIRROR_DEBUG(1, "Thread exiting."); 1926 kproc_exit(0); 1927 } 1928 mtx_lock(&sc->sc_queue_mtx); 1929 if (bioq_first(&sc->sc_queue) != NULL) { 1930 mtx_unlock(&sc->sc_queue_mtx); 1931 continue; 1932 } 1933 } 1934 sx_xunlock(&sc->sc_lock); 1935 /* 1936 * XXX: We can miss an event here, because an event 1937 * can be added without sx-device-lock and without 1938 * mtx-queue-lock. Maybe I should just stop using 1939 * dedicated mutex for events synchronization and 1940 * stick with the queue lock? 1941 * The event will hang here until next I/O request 1942 * or next event is received. 1943 */ 1944 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1945 timeout * hz); 1946 sx_xlock(&sc->sc_lock); 1947 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1948 continue; 1949 } 1950 mtx_unlock(&sc->sc_queue_mtx); 1951 1952 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 1953 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1954 g_mirror_sync_request(bp); /* READ */ 1955 } else if (bp->bio_to != sc->sc_provider) { 1956 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 1957 g_mirror_regular_request(bp); 1958 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 1959 g_mirror_sync_request(bp); /* WRITE */ 1960 else { 1961 KASSERT(0, 1962 ("Invalid request cflags=0x%hx to=%s.", 1963 bp->bio_cflags, bp->bio_to->name)); 1964 } 1965 } else { 1966 g_mirror_register_request(bp); 1967 } 1968 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1969 } 1970 } 1971 1972 static void 1973 g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 1974 { 1975 1976 sx_assert(&sc->sc_lock, SX_LOCKED); 1977 1978 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 1979 return; 1980 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1981 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 1982 g_mirror_get_diskname(disk), sc->sc_name); 1983 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1984 } else if (sc->sc_idle && 1985 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1986 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 1987 g_mirror_get_diskname(disk), sc->sc_name); 1988 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1989 } 1990 } 1991 1992 static void 1993 g_mirror_sync_start(struct g_mirror_disk *disk) 1994 { 1995 struct g_mirror_softc *sc; 1996 struct g_consumer *cp; 1997 struct bio *bp; 1998 int error, i; 1999 2000 g_topology_assert_not(); 2001 sc = disk->d_softc; 2002 sx_assert(&sc->sc_lock, SX_LOCKED); 2003 2004 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2005 ("Disk %s is not marked for synchronization.", 2006 g_mirror_get_diskname(disk))); 2007 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2008 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 2009 sc->sc_state)); 2010 2011 sx_xunlock(&sc->sc_lock); 2012 g_topology_lock(); 2013 cp = g_new_consumer(sc->sc_sync.ds_geom); 2014 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 2015 error = g_attach(cp, sc->sc_provider); 2016 KASSERT(error == 0, 2017 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 2018 error = g_access(cp, 1, 0, 0); 2019 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 2020 g_topology_unlock(); 2021 sx_xlock(&sc->sc_lock); 2022 2023 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 2024 g_mirror_get_diskname(disk)); 2025 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0) 2026 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2027 KASSERT(disk->d_sync.ds_consumer == NULL, 2028 ("Sync consumer already exists (device=%s, disk=%s).", 2029 sc->sc_name, g_mirror_get_diskname(disk))); 2030 2031 disk->d_sync.ds_consumer = cp; 2032 disk->d_sync.ds_consumer->private = disk; 2033 disk->d_sync.ds_consumer->index = 0; 2034 2035 /* 2036 * Allocate memory for synchronization bios and initialize them. 2037 */ 2038 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 2039 M_MIRROR, M_WAITOK); 2040 for (i = 0; i < g_mirror_syncreqs; i++) { 2041 bp = g_alloc_bio(); 2042 disk->d_sync.ds_bios[i] = bp; 2043 bp->bio_parent = NULL; 2044 bp->bio_cmd = BIO_READ; 2045 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 2046 bp->bio_cflags = 0; 2047 bp->bio_offset = disk->d_sync.ds_offset; 2048 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 2049 disk->d_sync.ds_offset += bp->bio_length; 2050 bp->bio_done = g_mirror_sync_done; 2051 bp->bio_from = disk->d_sync.ds_consumer; 2052 bp->bio_to = sc->sc_provider; 2053 bp->bio_caller1 = (void *)(uintptr_t)i; 2054 } 2055 2056 /* Increase the number of disks in SYNCHRONIZING state. */ 2057 sc->sc_sync.ds_ndisks++; 2058 /* Set the number of in-flight synchronization requests. */ 2059 disk->d_sync.ds_inflight = g_mirror_syncreqs; 2060 2061 /* 2062 * Fire off first synchronization requests. 2063 */ 2064 for (i = 0; i < g_mirror_syncreqs; i++) { 2065 bp = disk->d_sync.ds_bios[i]; 2066 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 2067 disk->d_sync.ds_consumer->index++; 2068 /* 2069 * Delay the request if it is colliding with a regular request. 2070 */ 2071 if (g_mirror_regular_collision(sc, bp)) 2072 g_mirror_sync_delay(sc, bp); 2073 else 2074 g_io_request(bp, disk->d_sync.ds_consumer); 2075 } 2076 } 2077 2078 /* 2079 * Stop synchronization process. 2080 * type: 0 - synchronization finished 2081 * 1 - synchronization stopped 2082 */ 2083 static void 2084 g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 2085 { 2086 struct g_mirror_softc *sc; 2087 struct g_consumer *cp; 2088 2089 g_topology_assert_not(); 2090 sc = disk->d_softc; 2091 sx_assert(&sc->sc_lock, SX_LOCKED); 2092 2093 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2094 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2095 g_mirror_disk_state2str(disk->d_state))); 2096 if (disk->d_sync.ds_consumer == NULL) 2097 return; 2098 2099 if (type == 0) { 2100 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2101 sc->sc_name, g_mirror_get_diskname(disk)); 2102 } else /* if (type == 1) */ { 2103 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2104 sc->sc_name, g_mirror_get_diskname(disk)); 2105 } 2106 g_mirror_regular_release(sc); 2107 free(disk->d_sync.ds_bios, M_MIRROR); 2108 disk->d_sync.ds_bios = NULL; 2109 cp = disk->d_sync.ds_consumer; 2110 disk->d_sync.ds_consumer = NULL; 2111 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2112 sc->sc_sync.ds_ndisks--; 2113 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2114 g_topology_lock(); 2115 g_mirror_kill_consumer(sc, cp); 2116 g_topology_unlock(); 2117 sx_xlock(&sc->sc_lock); 2118 } 2119 2120 static void 2121 g_mirror_launch_provider(struct g_mirror_softc *sc) 2122 { 2123 struct g_mirror_disk *disk; 2124 struct g_provider *pp, *dp; 2125 2126 sx_assert(&sc->sc_lock, SX_LOCKED); 2127 2128 g_topology_lock(); 2129 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 2130 pp->flags |= G_PF_DIRECT_RECEIVE; 2131 pp->mediasize = sc->sc_mediasize; 2132 pp->sectorsize = sc->sc_sectorsize; 2133 pp->stripesize = 0; 2134 pp->stripeoffset = 0; 2135 2136 /* Splitting of unmapped BIO's could work but isn't implemented now */ 2137 if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT) 2138 pp->flags |= G_PF_ACCEPT_UNMAPPED; 2139 2140 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2141 if (disk->d_consumer && disk->d_consumer->provider) { 2142 dp = disk->d_consumer->provider; 2143 if (dp->stripesize > pp->stripesize) { 2144 pp->stripesize = dp->stripesize; 2145 pp->stripeoffset = dp->stripeoffset; 2146 } 2147 /* A provider underneath us doesn't support unmapped */ 2148 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { 2149 G_MIRROR_DEBUG(0, "Cancelling unmapped " 2150 "because of %s.", dp->name); 2151 pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 2152 } 2153 } 2154 } 2155 pp->private = sc; 2156 sc->sc_refcnt++; 2157 sc->sc_provider = pp; 2158 g_error_provider(pp, 0); 2159 g_topology_unlock(); 2160 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2161 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 2162 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2163 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2164 g_mirror_sync_start(disk); 2165 } 2166 } 2167 2168 static void 2169 g_mirror_destroy_provider(struct g_mirror_softc *sc) 2170 { 2171 struct g_mirror_disk *disk; 2172 struct bio *bp; 2173 2174 g_topology_assert_not(); 2175 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2176 sc->sc_name)); 2177 2178 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2179 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2180 g_mirror_sync_stop(disk, 1); 2181 } 2182 2183 g_topology_lock(); 2184 g_error_provider(sc->sc_provider, ENXIO); 2185 mtx_lock(&sc->sc_queue_mtx); 2186 while ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) { 2187 /* 2188 * Abort any pending I/O that wasn't generated by us. 2189 * Synchronization requests and requests destined for individual 2190 * mirror components can be destroyed immediately. 2191 */ 2192 if (bp->bio_to == sc->sc_provider && 2193 bp->bio_from->geom != sc->sc_sync.ds_geom) { 2194 g_io_deliver(bp, ENXIO); 2195 } else { 2196 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2197 free(bp->bio_data, M_MIRROR); 2198 g_destroy_bio(bp); 2199 } 2200 } 2201 mtx_unlock(&sc->sc_queue_mtx); 2202 g_wither_provider(sc->sc_provider, ENXIO); 2203 sc->sc_provider = NULL; 2204 G_MIRROR_DEBUG(0, "Device %s: provider destroyed.", sc->sc_name); 2205 g_topology_unlock(); 2206 } 2207 2208 static void 2209 g_mirror_go(void *arg) 2210 { 2211 struct g_mirror_softc *sc; 2212 2213 sc = arg; 2214 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2215 g_mirror_event_send(sc, 0, 2216 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2217 } 2218 2219 static u_int 2220 g_mirror_determine_state(struct g_mirror_disk *disk) 2221 { 2222 struct g_mirror_softc *sc; 2223 u_int state; 2224 2225 sc = disk->d_softc; 2226 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2227 if ((disk->d_flags & 2228 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0 && 2229 (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 || 2230 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0)) { 2231 /* Disk does not need synchronization. */ 2232 state = G_MIRROR_DISK_STATE_ACTIVE; 2233 } else { 2234 if ((sc->sc_flags & 2235 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2236 (disk->d_flags & 2237 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2238 /* 2239 * We can start synchronization from 2240 * the stored offset. 2241 */ 2242 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2243 } else { 2244 state = G_MIRROR_DISK_STATE_STALE; 2245 } 2246 } 2247 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2248 /* 2249 * Reset all synchronization data for this disk, 2250 * because if it even was synchronized, it was 2251 * synchronized to disks with different syncid. 2252 */ 2253 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2254 disk->d_sync.ds_offset = 0; 2255 disk->d_sync.ds_offset_done = 0; 2256 disk->d_sync.ds_syncid = sc->sc_syncid; 2257 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2258 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2259 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2260 } else { 2261 state = G_MIRROR_DISK_STATE_STALE; 2262 } 2263 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2264 /* 2265 * Not good, NOT GOOD! 2266 * It means that mirror was started on stale disks 2267 * and more fresh disk just arrive. 2268 * If there were writes, mirror is broken, sorry. 2269 * I think the best choice here is don't touch 2270 * this disk and inform the user loudly. 2271 */ 2272 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2273 "disk (%s) arrives!! It will not be connected to the " 2274 "running device.", sc->sc_name, 2275 g_mirror_get_diskname(disk)); 2276 g_mirror_destroy_disk(disk); 2277 state = G_MIRROR_DISK_STATE_NONE; 2278 /* Return immediately, because disk was destroyed. */ 2279 return (state); 2280 } 2281 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2282 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2283 return (state); 2284 } 2285 2286 /* 2287 * Update device state. 2288 */ 2289 static void 2290 g_mirror_update_device(struct g_mirror_softc *sc, bool force) 2291 { 2292 struct g_mirror_disk *disk; 2293 u_int state; 2294 2295 sx_assert(&sc->sc_lock, SX_XLOCKED); 2296 2297 switch (sc->sc_state) { 2298 case G_MIRROR_DEVICE_STATE_STARTING: 2299 { 2300 struct g_mirror_disk *pdisk, *tdisk; 2301 u_int dirty, ndisks, genid, syncid; 2302 bool broken; 2303 2304 KASSERT(sc->sc_provider == NULL, 2305 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2306 /* 2307 * Are we ready? We are, if all disks are connected or 2308 * if we have any disks and 'force' is true. 2309 */ 2310 ndisks = g_mirror_ndisks(sc, -1); 2311 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2312 ; 2313 } else if (ndisks == 0) { 2314 /* 2315 * Disks went down in starting phase, so destroy 2316 * device. 2317 */ 2318 callout_drain(&sc->sc_callout); 2319 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2320 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2321 sc->sc_rootmount); 2322 root_mount_rel(sc->sc_rootmount); 2323 sc->sc_rootmount = NULL; 2324 return; 2325 } else { 2326 return; 2327 } 2328 2329 /* 2330 * Activate all disks with the biggest syncid. 2331 */ 2332 if (force) { 2333 /* 2334 * If 'force' is true, we have been called due to 2335 * timeout, so don't bother canceling timeout. 2336 */ 2337 ndisks = 0; 2338 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2339 if ((disk->d_flags & 2340 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2341 ndisks++; 2342 } 2343 } 2344 if (ndisks == 0) { 2345 /* No valid disks found, destroy device. */ 2346 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2347 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2348 __LINE__, sc->sc_rootmount); 2349 root_mount_rel(sc->sc_rootmount); 2350 sc->sc_rootmount = NULL; 2351 return; 2352 } 2353 } else { 2354 /* Cancel timeout. */ 2355 callout_drain(&sc->sc_callout); 2356 } 2357 2358 /* 2359 * Find the biggest genid. 2360 */ 2361 genid = 0; 2362 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2363 if (disk->d_genid > genid) 2364 genid = disk->d_genid; 2365 } 2366 sc->sc_genid = genid; 2367 /* 2368 * Remove all disks without the biggest genid. 2369 */ 2370 broken = false; 2371 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2372 if (disk->d_genid < genid) { 2373 G_MIRROR_DEBUG(0, 2374 "Component %s (device %s) broken, skipping.", 2375 g_mirror_get_diskname(disk), sc->sc_name); 2376 g_mirror_destroy_disk(disk); 2377 /* 2378 * Bump the syncid in case we discover a healthy 2379 * replacement disk after starting the mirror. 2380 */ 2381 broken = true; 2382 } 2383 } 2384 2385 /* 2386 * Find the biggest syncid. 2387 */ 2388 syncid = 0; 2389 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2390 if (disk->d_sync.ds_syncid > syncid) 2391 syncid = disk->d_sync.ds_syncid; 2392 } 2393 2394 /* 2395 * Here we need to look for dirty disks and if all disks 2396 * with the biggest syncid are dirty, we have to choose 2397 * one with the biggest priority and rebuild the rest. 2398 */ 2399 /* 2400 * Find the number of dirty disks with the biggest syncid. 2401 * Find the number of disks with the biggest syncid. 2402 * While here, find a disk with the biggest priority. 2403 */ 2404 dirty = ndisks = 0; 2405 pdisk = NULL; 2406 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2407 if (disk->d_sync.ds_syncid != syncid) 2408 continue; 2409 if ((disk->d_flags & 2410 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2411 continue; 2412 } 2413 ndisks++; 2414 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2415 dirty++; 2416 if (pdisk == NULL || 2417 pdisk->d_priority < disk->d_priority) { 2418 pdisk = disk; 2419 } 2420 } 2421 } 2422 if (dirty == 0) { 2423 /* No dirty disks at all, great. */ 2424 } else if (dirty == ndisks) { 2425 /* 2426 * Force synchronization for all dirty disks except one 2427 * with the biggest priority. 2428 */ 2429 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2430 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2431 "master disk for synchronization.", 2432 g_mirror_get_diskname(pdisk), sc->sc_name); 2433 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2434 if (disk->d_sync.ds_syncid != syncid) 2435 continue; 2436 if ((disk->d_flags & 2437 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2438 continue; 2439 } 2440 KASSERT((disk->d_flags & 2441 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2442 ("Disk %s isn't marked as dirty.", 2443 g_mirror_get_diskname(disk))); 2444 /* Skip the disk with the biggest priority. */ 2445 if (disk == pdisk) 2446 continue; 2447 disk->d_sync.ds_syncid = 0; 2448 } 2449 } else if (dirty < ndisks) { 2450 /* 2451 * Force synchronization for all dirty disks. 2452 * We have some non-dirty disks. 2453 */ 2454 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2455 if (disk->d_sync.ds_syncid != syncid) 2456 continue; 2457 if ((disk->d_flags & 2458 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2459 continue; 2460 } 2461 if ((disk->d_flags & 2462 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2463 continue; 2464 } 2465 disk->d_sync.ds_syncid = 0; 2466 } 2467 } 2468 2469 /* Reset hint. */ 2470 sc->sc_hint = NULL; 2471 sc->sc_syncid = syncid; 2472 if (force || broken) { 2473 /* Remember to bump syncid on first write. */ 2474 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2475 } 2476 state = G_MIRROR_DEVICE_STATE_RUNNING; 2477 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2478 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2479 g_mirror_device_state2str(state)); 2480 sc->sc_state = state; 2481 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2482 state = g_mirror_determine_state(disk); 2483 g_mirror_event_send(disk, state, 2484 G_MIRROR_EVENT_DONTWAIT); 2485 if (state == G_MIRROR_DISK_STATE_STALE) 2486 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2487 } 2488 break; 2489 } 2490 case G_MIRROR_DEVICE_STATE_RUNNING: 2491 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2492 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2493 /* 2494 * No usable disks, so destroy the device. 2495 */ 2496 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2497 break; 2498 } else if (g_mirror_ndisks(sc, 2499 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2500 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2501 /* 2502 * We have active disks, launch provider if it doesn't 2503 * exist. 2504 */ 2505 if (sc->sc_provider == NULL) 2506 g_mirror_launch_provider(sc); 2507 if (sc->sc_rootmount != NULL) { 2508 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2509 __LINE__, sc->sc_rootmount); 2510 root_mount_rel(sc->sc_rootmount); 2511 sc->sc_rootmount = NULL; 2512 } 2513 } 2514 /* 2515 * Genid should be bumped immediately, so do it here. 2516 */ 2517 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2518 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2519 g_mirror_bump_genid(sc); 2520 } 2521 break; 2522 default: 2523 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2524 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2525 break; 2526 } 2527 } 2528 2529 /* 2530 * Update disk state and device state if needed. 2531 */ 2532 #define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2533 "Disk %s state changed from %s to %s (device %s).", \ 2534 g_mirror_get_diskname(disk), \ 2535 g_mirror_disk_state2str(disk->d_state), \ 2536 g_mirror_disk_state2str(state), sc->sc_name) 2537 static int 2538 g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2539 { 2540 struct g_mirror_softc *sc; 2541 2542 sc = disk->d_softc; 2543 sx_assert(&sc->sc_lock, SX_XLOCKED); 2544 2545 again: 2546 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2547 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2548 g_mirror_disk_state2str(state)); 2549 switch (state) { 2550 case G_MIRROR_DISK_STATE_NEW: 2551 /* 2552 * Possible scenarios: 2553 * 1. New disk arrive. 2554 */ 2555 /* Previous state should be NONE. */ 2556 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2557 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2558 g_mirror_disk_state2str(disk->d_state))); 2559 DISK_STATE_CHANGED(); 2560 2561 disk->d_state = state; 2562 if (LIST_EMPTY(&sc->sc_disks)) 2563 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2564 else { 2565 struct g_mirror_disk *dp; 2566 2567 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2568 if (disk->d_priority >= dp->d_priority) { 2569 LIST_INSERT_BEFORE(dp, disk, d_next); 2570 dp = NULL; 2571 break; 2572 } 2573 if (LIST_NEXT(dp, d_next) == NULL) 2574 break; 2575 } 2576 if (dp != NULL) 2577 LIST_INSERT_AFTER(dp, disk, d_next); 2578 } 2579 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2580 sc->sc_name, g_mirror_get_diskname(disk)); 2581 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2582 break; 2583 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2584 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2585 g_mirror_device_state2str(sc->sc_state), 2586 g_mirror_get_diskname(disk), 2587 g_mirror_disk_state2str(disk->d_state))); 2588 state = g_mirror_determine_state(disk); 2589 if (state != G_MIRROR_DISK_STATE_NONE) 2590 goto again; 2591 break; 2592 case G_MIRROR_DISK_STATE_ACTIVE: 2593 /* 2594 * Possible scenarios: 2595 * 1. New disk does not need synchronization. 2596 * 2. Synchronization process finished successfully. 2597 */ 2598 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2599 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2600 g_mirror_device_state2str(sc->sc_state), 2601 g_mirror_get_diskname(disk), 2602 g_mirror_disk_state2str(disk->d_state))); 2603 /* Previous state should be NEW or SYNCHRONIZING. */ 2604 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2605 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2606 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2607 g_mirror_disk_state2str(disk->d_state))); 2608 DISK_STATE_CHANGED(); 2609 2610 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2611 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2612 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2613 g_mirror_sync_stop(disk, 0); 2614 } 2615 disk->d_state = state; 2616 disk->d_sync.ds_offset = 0; 2617 disk->d_sync.ds_offset_done = 0; 2618 g_mirror_update_idle(sc, disk); 2619 g_mirror_update_metadata(disk); 2620 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2621 sc->sc_name, g_mirror_get_diskname(disk)); 2622 break; 2623 case G_MIRROR_DISK_STATE_STALE: 2624 /* 2625 * Possible scenarios: 2626 * 1. Stale disk was connected. 2627 */ 2628 /* Previous state should be NEW. */ 2629 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2630 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2631 g_mirror_disk_state2str(disk->d_state))); 2632 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2633 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2634 g_mirror_device_state2str(sc->sc_state), 2635 g_mirror_get_diskname(disk), 2636 g_mirror_disk_state2str(disk->d_state))); 2637 /* 2638 * STALE state is only possible if device is marked 2639 * NOAUTOSYNC. 2640 */ 2641 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2642 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2643 g_mirror_device_state2str(sc->sc_state), 2644 g_mirror_get_diskname(disk), 2645 g_mirror_disk_state2str(disk->d_state))); 2646 DISK_STATE_CHANGED(); 2647 2648 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2649 disk->d_state = state; 2650 g_mirror_update_metadata(disk); 2651 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2652 sc->sc_name, g_mirror_get_diskname(disk)); 2653 break; 2654 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2655 /* 2656 * Possible scenarios: 2657 * 1. Disk which needs synchronization was connected. 2658 */ 2659 /* Previous state should be NEW. */ 2660 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2661 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2662 g_mirror_disk_state2str(disk->d_state))); 2663 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2664 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2665 g_mirror_device_state2str(sc->sc_state), 2666 g_mirror_get_diskname(disk), 2667 g_mirror_disk_state2str(disk->d_state))); 2668 DISK_STATE_CHANGED(); 2669 2670 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2671 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2672 disk->d_state = state; 2673 if (sc->sc_provider != NULL) { 2674 g_mirror_sync_start(disk); 2675 g_mirror_update_metadata(disk); 2676 } 2677 break; 2678 case G_MIRROR_DISK_STATE_DISCONNECTED: 2679 /* 2680 * Possible scenarios: 2681 * 1. Device wasn't running yet, but disk disappear. 2682 * 2. Disk was active and disapppear. 2683 * 3. Disk disappear during synchronization process. 2684 */ 2685 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2686 /* 2687 * Previous state should be ACTIVE, STALE or 2688 * SYNCHRONIZING. 2689 */ 2690 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2691 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2692 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2693 ("Wrong disk state (%s, %s).", 2694 g_mirror_get_diskname(disk), 2695 g_mirror_disk_state2str(disk->d_state))); 2696 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2697 /* Previous state should be NEW. */ 2698 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2699 ("Wrong disk state (%s, %s).", 2700 g_mirror_get_diskname(disk), 2701 g_mirror_disk_state2str(disk->d_state))); 2702 /* 2703 * Reset bumping syncid if disk disappeared in STARTING 2704 * state. 2705 */ 2706 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2707 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2708 #ifdef INVARIANTS 2709 } else { 2710 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2711 sc->sc_name, 2712 g_mirror_device_state2str(sc->sc_state), 2713 g_mirror_get_diskname(disk), 2714 g_mirror_disk_state2str(disk->d_state))); 2715 #endif 2716 } 2717 DISK_STATE_CHANGED(); 2718 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2719 sc->sc_name, g_mirror_get_diskname(disk)); 2720 2721 g_mirror_destroy_disk(disk); 2722 break; 2723 case G_MIRROR_DISK_STATE_DESTROY: 2724 { 2725 int error; 2726 2727 error = g_mirror_clear_metadata(disk); 2728 if (error != 0) { 2729 G_MIRROR_DEBUG(0, 2730 "Device %s: failed to clear metadata on %s: %d.", 2731 sc->sc_name, g_mirror_get_diskname(disk), error); 2732 break; 2733 } 2734 DISK_STATE_CHANGED(); 2735 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2736 sc->sc_name, g_mirror_get_diskname(disk)); 2737 2738 g_mirror_destroy_disk(disk); 2739 sc->sc_ndisks--; 2740 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2741 g_mirror_update_metadata(disk); 2742 } 2743 break; 2744 } 2745 default: 2746 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2747 break; 2748 } 2749 return (0); 2750 } 2751 #undef DISK_STATE_CHANGED 2752 2753 int 2754 g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2755 { 2756 struct g_provider *pp; 2757 u_char *buf; 2758 int error; 2759 2760 g_topology_assert(); 2761 2762 error = g_access(cp, 1, 0, 0); 2763 if (error != 0) 2764 return (error); 2765 pp = cp->provider; 2766 g_topology_unlock(); 2767 /* Metadata are stored on last sector. */ 2768 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2769 &error); 2770 g_topology_lock(); 2771 g_access(cp, -1, 0, 0); 2772 if (buf == NULL) { 2773 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2774 cp->provider->name, error); 2775 return (error); 2776 } 2777 2778 /* Decode metadata. */ 2779 error = mirror_metadata_decode(buf, md); 2780 g_free(buf); 2781 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2782 return (EINVAL); 2783 if (md->md_version > G_MIRROR_VERSION) { 2784 G_MIRROR_DEBUG(0, 2785 "Kernel module is too old to handle metadata from %s.", 2786 cp->provider->name); 2787 return (EINVAL); 2788 } 2789 if (error != 0) { 2790 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2791 cp->provider->name); 2792 return (error); 2793 } 2794 2795 return (0); 2796 } 2797 2798 static int 2799 g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2800 struct g_mirror_metadata *md) 2801 { 2802 2803 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2804 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2805 pp->name, md->md_did); 2806 return (EEXIST); 2807 } 2808 if (md->md_all != sc->sc_ndisks) { 2809 G_MIRROR_DEBUG(1, 2810 "Invalid '%s' field on disk %s (device %s), skipping.", 2811 "md_all", pp->name, sc->sc_name); 2812 return (EINVAL); 2813 } 2814 if (md->md_slice != sc->sc_slice) { 2815 G_MIRROR_DEBUG(1, 2816 "Invalid '%s' field on disk %s (device %s), skipping.", 2817 "md_slice", pp->name, sc->sc_name); 2818 return (EINVAL); 2819 } 2820 if (md->md_balance != sc->sc_balance) { 2821 G_MIRROR_DEBUG(1, 2822 "Invalid '%s' field on disk %s (device %s), skipping.", 2823 "md_balance", pp->name, sc->sc_name); 2824 return (EINVAL); 2825 } 2826 #if 0 2827 if (md->md_mediasize != sc->sc_mediasize) { 2828 G_MIRROR_DEBUG(1, 2829 "Invalid '%s' field on disk %s (device %s), skipping.", 2830 "md_mediasize", pp->name, sc->sc_name); 2831 return (EINVAL); 2832 } 2833 #endif 2834 if (sc->sc_mediasize > pp->mediasize) { 2835 G_MIRROR_DEBUG(1, 2836 "Invalid size of disk %s (device %s), skipping.", pp->name, 2837 sc->sc_name); 2838 return (EINVAL); 2839 } 2840 if (md->md_sectorsize != sc->sc_sectorsize) { 2841 G_MIRROR_DEBUG(1, 2842 "Invalid '%s' field on disk %s (device %s), skipping.", 2843 "md_sectorsize", pp->name, sc->sc_name); 2844 return (EINVAL); 2845 } 2846 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2847 G_MIRROR_DEBUG(1, 2848 "Invalid sector size of disk %s (device %s), skipping.", 2849 pp->name, sc->sc_name); 2850 return (EINVAL); 2851 } 2852 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2853 G_MIRROR_DEBUG(1, 2854 "Invalid device flags on disk %s (device %s), skipping.", 2855 pp->name, sc->sc_name); 2856 return (EINVAL); 2857 } 2858 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2859 G_MIRROR_DEBUG(1, 2860 "Invalid disk flags on disk %s (device %s), skipping.", 2861 pp->name, sc->sc_name); 2862 return (EINVAL); 2863 } 2864 return (0); 2865 } 2866 2867 int 2868 g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2869 struct g_mirror_metadata *md) 2870 { 2871 struct g_mirror_disk *disk; 2872 int error; 2873 2874 g_topology_assert_not(); 2875 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2876 2877 error = g_mirror_check_metadata(sc, pp, md); 2878 if (error != 0) 2879 return (error); 2880 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2881 md->md_genid < sc->sc_genid) { 2882 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2883 pp->name, sc->sc_name); 2884 return (EINVAL); 2885 } 2886 disk = g_mirror_init_disk(sc, pp, md, &error); 2887 if (disk == NULL) 2888 return (error); 2889 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2890 G_MIRROR_EVENT_WAIT); 2891 if (error != 0) 2892 return (error); 2893 if (md->md_version < G_MIRROR_VERSION) { 2894 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2895 pp->name, md->md_version, G_MIRROR_VERSION); 2896 g_mirror_update_metadata(disk); 2897 } 2898 return (0); 2899 } 2900 2901 static void 2902 g_mirror_destroy_delayed(void *arg, int flag) 2903 { 2904 struct g_mirror_softc *sc; 2905 int error; 2906 2907 if (flag == EV_CANCEL) { 2908 G_MIRROR_DEBUG(1, "Destroying canceled."); 2909 return; 2910 } 2911 sc = arg; 2912 g_topology_unlock(); 2913 sx_xlock(&sc->sc_lock); 2914 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2915 ("DESTROY flag set on %s.", sc->sc_name)); 2916 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0, 2917 ("CLOSEWAIT flag not set on %s.", sc->sc_name)); 2918 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 2919 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 2920 if (error != 0) { 2921 G_MIRROR_DEBUG(0, "Cannot destroy %s (error=%d).", 2922 sc->sc_name, error); 2923 sx_xunlock(&sc->sc_lock); 2924 } 2925 g_topology_lock(); 2926 } 2927 2928 static int 2929 g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2930 { 2931 struct g_mirror_softc *sc; 2932 int error = 0; 2933 2934 g_topology_assert(); 2935 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2936 acw, ace); 2937 2938 sc = pp->private; 2939 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 2940 2941 g_topology_unlock(); 2942 sx_xlock(&sc->sc_lock); 2943 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 2944 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 || 2945 LIST_EMPTY(&sc->sc_disks)) { 2946 if (acr > 0 || acw > 0 || ace > 0) 2947 error = ENXIO; 2948 goto end; 2949 } 2950 sc->sc_provider_open += acr + acw + ace; 2951 if (pp->acw + acw == 0) 2952 g_mirror_idle(sc, 0); 2953 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 && 2954 sc->sc_provider_open == 0) 2955 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, sc, NULL); 2956 end: 2957 sx_xunlock(&sc->sc_lock); 2958 g_topology_lock(); 2959 return (error); 2960 } 2961 2962 struct g_geom * 2963 g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md, 2964 u_int type) 2965 { 2966 struct g_mirror_softc *sc; 2967 struct g_geom *gp; 2968 int error, timeout; 2969 2970 g_topology_assert(); 2971 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2972 md->md_mid); 2973 2974 /* One disk is minimum. */ 2975 if (md->md_all < 1) 2976 return (NULL); 2977 /* 2978 * Action geom. 2979 */ 2980 gp = g_new_geomf(mp, "%s", md->md_name); 2981 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2982 gp->start = g_mirror_start; 2983 gp->orphan = g_mirror_orphan; 2984 gp->access = g_mirror_access; 2985 gp->dumpconf = g_mirror_dumpconf; 2986 2987 sc->sc_type = type; 2988 sc->sc_id = md->md_mid; 2989 sc->sc_slice = md->md_slice; 2990 sc->sc_balance = md->md_balance; 2991 sc->sc_mediasize = md->md_mediasize; 2992 sc->sc_sectorsize = md->md_sectorsize; 2993 sc->sc_ndisks = md->md_all; 2994 sc->sc_flags = md->md_mflags; 2995 sc->sc_bump_id = 0; 2996 sc->sc_idle = 1; 2997 sc->sc_last_write = time_uptime; 2998 sc->sc_writes = 0; 2999 sc->sc_refcnt = 1; 3000 sx_init(&sc->sc_lock, "gmirror:lock"); 3001 bioq_init(&sc->sc_queue); 3002 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 3003 bioq_init(&sc->sc_regular_delayed); 3004 bioq_init(&sc->sc_inflight); 3005 bioq_init(&sc->sc_sync_delayed); 3006 LIST_INIT(&sc->sc_disks); 3007 TAILQ_INIT(&sc->sc_events); 3008 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 3009 callout_init(&sc->sc_callout, 1); 3010 mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF); 3011 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 3012 gp->softc = sc; 3013 sc->sc_geom = gp; 3014 sc->sc_provider = NULL; 3015 sc->sc_provider_open = 0; 3016 /* 3017 * Synchronization geom. 3018 */ 3019 gp = g_new_geomf(mp, "%s.sync", md->md_name); 3020 gp->softc = sc; 3021 gp->orphan = g_mirror_orphan; 3022 sc->sc_sync.ds_geom = gp; 3023 sc->sc_sync.ds_ndisks = 0; 3024 error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 3025 "g_mirror %s", md->md_name); 3026 if (error != 0) { 3027 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 3028 sc->sc_name); 3029 g_destroy_geom(sc->sc_sync.ds_geom); 3030 g_destroy_geom(sc->sc_geom); 3031 g_mirror_free_device(sc); 3032 return (NULL); 3033 } 3034 3035 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 3036 sc->sc_name, sc->sc_ndisks, sc->sc_id); 3037 3038 sc->sc_rootmount = root_mount_hold("GMIRROR"); 3039 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 3040 /* 3041 * Run timeout. 3042 */ 3043 timeout = g_mirror_timeout * hz; 3044 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 3045 return (sc->sc_geom); 3046 } 3047 3048 int 3049 g_mirror_destroy(struct g_mirror_softc *sc, int how) 3050 { 3051 struct g_mirror_disk *disk; 3052 3053 g_topology_assert_not(); 3054 sx_assert(&sc->sc_lock, SX_XLOCKED); 3055 3056 if (sc->sc_provider_open != 0) { 3057 switch (how) { 3058 case G_MIRROR_DESTROY_SOFT: 3059 G_MIRROR_DEBUG(1, 3060 "Device %s is still open (%d).", sc->sc_name, 3061 sc->sc_provider_open); 3062 return (EBUSY); 3063 case G_MIRROR_DESTROY_DELAYED: 3064 G_MIRROR_DEBUG(1, 3065 "Device %s will be destroyed on last close.", 3066 sc->sc_name); 3067 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 3068 if (disk->d_state == 3069 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3070 g_mirror_sync_stop(disk, 1); 3071 } 3072 } 3073 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_CLOSEWAIT; 3074 return (EBUSY); 3075 case G_MIRROR_DESTROY_HARD: 3076 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 3077 "can't be definitely removed.", sc->sc_name); 3078 } 3079 } 3080 3081 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) 3082 return (0); 3083 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 3084 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DRAIN; 3085 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3086 sx_xunlock(&sc->sc_lock); 3087 mtx_lock(&sc->sc_queue_mtx); 3088 wakeup(sc); 3089 mtx_unlock(&sc->sc_queue_mtx); 3090 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3091 while (sc->sc_worker != NULL) 3092 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 3093 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3094 sx_xlock(&sc->sc_lock); 3095 g_mirror_destroy_device(sc); 3096 return (0); 3097 } 3098 3099 static void 3100 g_mirror_taste_orphan(struct g_consumer *cp) 3101 { 3102 3103 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3104 cp->provider->name)); 3105 } 3106 3107 static struct g_geom * 3108 g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3109 { 3110 struct g_mirror_metadata md; 3111 struct g_mirror_softc *sc; 3112 struct g_consumer *cp; 3113 struct g_geom *gp; 3114 int error; 3115 3116 g_topology_assert(); 3117 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3118 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 3119 3120 gp = g_new_geomf(mp, "mirror:taste"); 3121 /* 3122 * This orphan function should be never called. 3123 */ 3124 gp->orphan = g_mirror_taste_orphan; 3125 cp = g_new_consumer(gp); 3126 g_attach(cp, pp); 3127 error = g_mirror_read_metadata(cp, &md); 3128 g_detach(cp); 3129 g_destroy_consumer(cp); 3130 g_destroy_geom(gp); 3131 if (error != 0) 3132 return (NULL); 3133 gp = NULL; 3134 3135 if (md.md_provider[0] != '\0' && 3136 !g_compare_names(md.md_provider, pp->name)) 3137 return (NULL); 3138 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3139 return (NULL); 3140 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 3141 G_MIRROR_DEBUG(0, 3142 "Device %s: provider %s marked as inactive, skipping.", 3143 md.md_name, pp->name); 3144 return (NULL); 3145 } 3146 if (g_mirror_debug >= 2) 3147 mirror_metadata_dump(&md); 3148 3149 /* 3150 * Let's check if device already exists. 3151 */ 3152 sc = NULL; 3153 LIST_FOREACH(gp, &mp->geom, geom) { 3154 sc = gp->softc; 3155 if (sc == NULL) 3156 continue; 3157 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 3158 continue; 3159 if (sc->sc_sync.ds_geom == gp) 3160 continue; 3161 if (strcmp(md.md_name, sc->sc_name) != 0) 3162 continue; 3163 if (md.md_mid != sc->sc_id) { 3164 G_MIRROR_DEBUG(0, "Device %s already configured.", 3165 sc->sc_name); 3166 return (NULL); 3167 } 3168 break; 3169 } 3170 if (gp == NULL) { 3171 gp = g_mirror_create(mp, &md, G_MIRROR_TYPE_AUTOMATIC); 3172 if (gp == NULL) { 3173 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3174 md.md_name); 3175 return (NULL); 3176 } 3177 sc = gp->softc; 3178 } 3179 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3180 g_topology_unlock(); 3181 sx_xlock(&sc->sc_lock); 3182 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; 3183 error = g_mirror_add_disk(sc, pp, &md); 3184 if (error != 0) { 3185 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3186 pp->name, gp->name, error); 3187 if (LIST_EMPTY(&sc->sc_disks)) { 3188 g_cancel_event(sc); 3189 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3190 g_topology_lock(); 3191 return (NULL); 3192 } 3193 gp = NULL; 3194 } 3195 sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; 3196 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3197 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3198 g_topology_lock(); 3199 return (NULL); 3200 } 3201 sx_xunlock(&sc->sc_lock); 3202 g_topology_lock(); 3203 return (gp); 3204 } 3205 3206 static void 3207 g_mirror_resize(struct g_consumer *cp) 3208 { 3209 struct g_mirror_disk *disk; 3210 3211 g_topology_assert(); 3212 g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name); 3213 3214 disk = cp->private; 3215 if (disk == NULL) 3216 return; 3217 g_topology_unlock(); 3218 g_mirror_update_metadata(disk); 3219 g_topology_lock(); 3220 } 3221 3222 static int 3223 g_mirror_destroy_geom(struct gctl_req *req __unused, 3224 struct g_class *mp __unused, struct g_geom *gp) 3225 { 3226 struct g_mirror_softc *sc; 3227 int error; 3228 3229 g_topology_unlock(); 3230 sc = gp->softc; 3231 sx_xlock(&sc->sc_lock); 3232 g_cancel_event(sc); 3233 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3234 if (error != 0) 3235 sx_xunlock(&sc->sc_lock); 3236 g_topology_lock(); 3237 return (error); 3238 } 3239 3240 static void 3241 g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3242 struct g_consumer *cp, struct g_provider *pp) 3243 { 3244 struct g_mirror_softc *sc; 3245 3246 g_topology_assert(); 3247 3248 sc = gp->softc; 3249 if (sc == NULL) 3250 return; 3251 /* Skip synchronization geom. */ 3252 if (gp == sc->sc_sync.ds_geom) 3253 return; 3254 if (pp != NULL) { 3255 /* Nothing here. */ 3256 } else if (cp != NULL) { 3257 struct g_mirror_disk *disk; 3258 3259 disk = cp->private; 3260 if (disk == NULL) 3261 return; 3262 g_topology_unlock(); 3263 sx_xlock(&sc->sc_lock); 3264 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3265 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3266 sbuf_printf(sb, "%s<Synchronized>", indent); 3267 if (disk->d_sync.ds_offset == 0) 3268 sbuf_printf(sb, "0%%"); 3269 else { 3270 sbuf_printf(sb, "%u%%", 3271 (u_int)((disk->d_sync.ds_offset * 100) / 3272 sc->sc_provider->mediasize)); 3273 } 3274 sbuf_printf(sb, "</Synchronized>\n"); 3275 if (disk->d_sync.ds_offset > 0) { 3276 sbuf_printf(sb, "%s<BytesSynced>%jd" 3277 "</BytesSynced>\n", indent, 3278 (intmax_t)disk->d_sync.ds_offset); 3279 } 3280 } 3281 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3282 disk->d_sync.ds_syncid); 3283 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3284 disk->d_genid); 3285 sbuf_printf(sb, "%s<Flags>", indent); 3286 if (disk->d_flags == 0) 3287 sbuf_printf(sb, "NONE"); 3288 else { 3289 int first = 1; 3290 3291 #define ADD_FLAG(flag, name) do { \ 3292 if ((disk->d_flags & (flag)) != 0) { \ 3293 if (!first) \ 3294 sbuf_printf(sb, ", "); \ 3295 else \ 3296 first = 0; \ 3297 sbuf_printf(sb, name); \ 3298 } \ 3299 } while (0) 3300 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3301 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3302 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3303 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3304 "SYNCHRONIZING"); 3305 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3306 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3307 #undef ADD_FLAG 3308 } 3309 sbuf_printf(sb, "</Flags>\n"); 3310 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3311 disk->d_priority); 3312 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3313 g_mirror_disk_state2str(disk->d_state)); 3314 sx_xunlock(&sc->sc_lock); 3315 g_topology_lock(); 3316 } else { 3317 g_topology_unlock(); 3318 sx_xlock(&sc->sc_lock); 3319 sbuf_printf(sb, "%s<Type>", indent); 3320 switch (sc->sc_type) { 3321 case G_MIRROR_TYPE_AUTOMATIC: 3322 sbuf_printf(sb, "AUTOMATIC"); 3323 break; 3324 case G_MIRROR_TYPE_MANUAL: 3325 sbuf_printf(sb, "MANUAL"); 3326 break; 3327 default: 3328 sbuf_printf(sb, "UNKNOWN"); 3329 break; 3330 } 3331 sbuf_printf(sb, "</Type>\n"); 3332 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3333 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3334 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3335 sbuf_printf(sb, "%s<Flags>", indent); 3336 if (sc->sc_flags == 0) 3337 sbuf_printf(sb, "NONE"); 3338 else { 3339 int first = 1; 3340 3341 #define ADD_FLAG(flag, name) do { \ 3342 if ((sc->sc_flags & (flag)) != 0) { \ 3343 if (!first) \ 3344 sbuf_printf(sb, ", "); \ 3345 else \ 3346 first = 0; \ 3347 sbuf_printf(sb, name); \ 3348 } \ 3349 } while (0) 3350 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3351 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3352 #undef ADD_FLAG 3353 } 3354 sbuf_printf(sb, "</Flags>\n"); 3355 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3356 (u_int)sc->sc_slice); 3357 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3358 balance_name(sc->sc_balance)); 3359 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3360 sc->sc_ndisks); 3361 sbuf_printf(sb, "%s<State>", indent); 3362 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3363 sbuf_printf(sb, "%s", "STARTING"); 3364 else if (sc->sc_ndisks == 3365 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3366 sbuf_printf(sb, "%s", "COMPLETE"); 3367 else 3368 sbuf_printf(sb, "%s", "DEGRADED"); 3369 sbuf_printf(sb, "</State>\n"); 3370 sx_xunlock(&sc->sc_lock); 3371 g_topology_lock(); 3372 } 3373 } 3374 3375 static void 3376 g_mirror_shutdown_post_sync(void *arg, int howto) 3377 { 3378 struct g_class *mp; 3379 struct g_geom *gp, *gp2; 3380 struct g_mirror_softc *sc; 3381 int error; 3382 3383 if (panicstr != NULL) 3384 return; 3385 3386 mp = arg; 3387 g_topology_lock(); 3388 g_mirror_shutdown = 1; 3389 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3390 if ((sc = gp->softc) == NULL) 3391 continue; 3392 /* Skip synchronization geom. */ 3393 if (gp == sc->sc_sync.ds_geom) 3394 continue; 3395 g_topology_unlock(); 3396 sx_xlock(&sc->sc_lock); 3397 g_mirror_idle(sc, -1); 3398 g_cancel_event(sc); 3399 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3400 if (error != 0) 3401 sx_xunlock(&sc->sc_lock); 3402 g_topology_lock(); 3403 } 3404 g_topology_unlock(); 3405 } 3406 3407 static void 3408 g_mirror_init(struct g_class *mp) 3409 { 3410 3411 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3412 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3413 if (g_mirror_post_sync == NULL) 3414 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3415 } 3416 3417 static void 3418 g_mirror_fini(struct g_class *mp) 3419 { 3420 3421 if (g_mirror_post_sync != NULL) 3422 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); 3423 } 3424 3425 DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3426