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 if (bp->bio_error == ENXIO && 986 bp->bio_cmd == BIO_READ) 987 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 988 else if (bp->bio_error == ENXIO) 989 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_NOW; 990 else 991 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 992 g_mirror_event_send(disk, 993 G_MIRROR_DISK_STATE_DISCONNECTED, 994 G_MIRROR_EVENT_DONTWAIT); 995 } 996 } 997 switch (pbp->bio_cmd) { 998 case BIO_DELETE: 999 case BIO_WRITE: 1000 pbp->bio_inbed--; 1001 pbp->bio_children--; 1002 break; 1003 } 1004 } 1005 g_destroy_bio(bp); 1006 1007 switch (pbp->bio_cmd) { 1008 case BIO_READ: 1009 if (pbp->bio_inbed < pbp->bio_children) 1010 break; 1011 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 1012 g_io_deliver(pbp, pbp->bio_error); 1013 else { 1014 pbp->bio_error = 0; 1015 mtx_lock(&sc->sc_queue_mtx); 1016 bioq_insert_tail(&sc->sc_queue, pbp); 1017 mtx_unlock(&sc->sc_queue_mtx); 1018 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1019 wakeup(sc); 1020 } 1021 break; 1022 case BIO_DELETE: 1023 case BIO_WRITE: 1024 if (pbp->bio_children == 0) { 1025 /* 1026 * All requests failed. 1027 */ 1028 } else if (pbp->bio_inbed < pbp->bio_children) { 1029 /* Do nothing. */ 1030 break; 1031 } else if (pbp->bio_children == pbp->bio_inbed) { 1032 /* Some requests succeeded. */ 1033 pbp->bio_error = 0; 1034 pbp->bio_completed = pbp->bio_length; 1035 } 1036 bioq_remove(&sc->sc_inflight, pbp); 1037 /* Release delayed sync requests if possible. */ 1038 g_mirror_sync_release(sc); 1039 g_io_deliver(pbp, pbp->bio_error); 1040 break; 1041 default: 1042 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 1043 break; 1044 } 1045 } 1046 1047 static void 1048 g_mirror_sync_done(struct bio *bp) 1049 { 1050 struct g_mirror_softc *sc; 1051 1052 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1053 sc = bp->bio_from->geom->softc; 1054 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; 1055 mtx_lock(&sc->sc_queue_mtx); 1056 bioq_insert_tail(&sc->sc_queue, bp); 1057 mtx_unlock(&sc->sc_queue_mtx); 1058 wakeup(sc); 1059 } 1060 1061 static void 1062 g_mirror_candelete(struct bio *bp) 1063 { 1064 struct g_mirror_softc *sc; 1065 struct g_mirror_disk *disk; 1066 int *val; 1067 1068 sc = bp->bio_to->private; 1069 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1070 if (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) 1071 break; 1072 } 1073 val = (int *)bp->bio_data; 1074 *val = (disk != NULL); 1075 g_io_deliver(bp, 0); 1076 } 1077 1078 static void 1079 g_mirror_kernel_dump(struct bio *bp) 1080 { 1081 struct g_mirror_softc *sc; 1082 struct g_mirror_disk *disk; 1083 struct bio *cbp; 1084 struct g_kerneldump *gkd; 1085 1086 /* 1087 * We configure dumping to the first component, because this component 1088 * will be used for reading with 'prefer' balance algorithm. 1089 * If the component with the highest priority is currently disconnected 1090 * we will not be able to read the dump after the reboot if it will be 1091 * connected and synchronized later. Can we do something better? 1092 */ 1093 sc = bp->bio_to->private; 1094 disk = LIST_FIRST(&sc->sc_disks); 1095 1096 gkd = (struct g_kerneldump *)bp->bio_data; 1097 if (gkd->length > bp->bio_to->mediasize) 1098 gkd->length = bp->bio_to->mediasize; 1099 cbp = g_clone_bio(bp); 1100 if (cbp == NULL) { 1101 g_io_deliver(bp, ENOMEM); 1102 return; 1103 } 1104 cbp->bio_done = g_std_done; 1105 g_io_request(cbp, disk->d_consumer); 1106 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1107 g_mirror_get_diskname(disk)); 1108 } 1109 1110 static void 1111 g_mirror_flush(struct g_mirror_softc *sc, struct bio *bp) 1112 { 1113 struct bio_queue_head queue; 1114 struct g_mirror_disk *disk; 1115 struct g_consumer *cp; 1116 struct bio *cbp; 1117 1118 bioq_init(&queue); 1119 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1120 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1121 continue; 1122 cbp = g_clone_bio(bp); 1123 if (cbp == NULL) { 1124 while ((cbp = bioq_takefirst(&queue)) != NULL) 1125 g_destroy_bio(cbp); 1126 if (bp->bio_error == 0) 1127 bp->bio_error = ENOMEM; 1128 g_io_deliver(bp, bp->bio_error); 1129 return; 1130 } 1131 bioq_insert_tail(&queue, cbp); 1132 cbp->bio_done = g_mirror_flush_done; 1133 cbp->bio_caller1 = disk; 1134 cbp->bio_to = disk->d_consumer->provider; 1135 } 1136 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1137 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1138 disk = cbp->bio_caller1; 1139 cbp->bio_caller1 = NULL; 1140 cp = disk->d_consumer; 1141 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1142 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1143 cp->acr, cp->acw, cp->ace)); 1144 g_io_request(cbp, disk->d_consumer); 1145 } 1146 } 1147 1148 static void 1149 g_mirror_start(struct bio *bp) 1150 { 1151 struct g_mirror_softc *sc; 1152 1153 sc = bp->bio_to->private; 1154 /* 1155 * If sc == NULL or there are no valid disks, provider's error 1156 * should be set and g_mirror_start() should not be called at all. 1157 */ 1158 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1159 ("Provider's error should be set (error=%d)(mirror=%s).", 1160 bp->bio_to->error, bp->bio_to->name)); 1161 G_MIRROR_LOGREQ(3, bp, "Request received."); 1162 1163 switch (bp->bio_cmd) { 1164 case BIO_READ: 1165 case BIO_WRITE: 1166 case BIO_DELETE: 1167 break; 1168 case BIO_FLUSH: 1169 g_mirror_flush(sc, bp); 1170 return; 1171 case BIO_GETATTR: 1172 if (!strcmp(bp->bio_attribute, "GEOM::candelete")) { 1173 g_mirror_candelete(bp); 1174 return; 1175 } else if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1176 g_mirror_kernel_dump(bp); 1177 return; 1178 } 1179 /* FALLTHROUGH */ 1180 default: 1181 g_io_deliver(bp, EOPNOTSUPP); 1182 return; 1183 } 1184 mtx_lock(&sc->sc_queue_mtx); 1185 if (bp->bio_to->error != 0) { 1186 mtx_unlock(&sc->sc_queue_mtx); 1187 g_io_deliver(bp, bp->bio_to->error); 1188 return; 1189 } 1190 bioq_insert_tail(&sc->sc_queue, bp); 1191 mtx_unlock(&sc->sc_queue_mtx); 1192 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1193 wakeup(sc); 1194 } 1195 1196 /* 1197 * Return TRUE if the given request is colliding with a in-progress 1198 * synchronization request. 1199 */ 1200 static int 1201 g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1202 { 1203 struct g_mirror_disk *disk; 1204 struct bio *sbp; 1205 off_t rstart, rend, sstart, send; 1206 u_int i; 1207 1208 if (sc->sc_sync.ds_ndisks == 0) 1209 return (0); 1210 rstart = bp->bio_offset; 1211 rend = bp->bio_offset + bp->bio_length; 1212 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1213 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1214 continue; 1215 for (i = 0; i < g_mirror_syncreqs; i++) { 1216 sbp = disk->d_sync.ds_bios[i]; 1217 if (sbp == NULL) 1218 continue; 1219 sstart = sbp->bio_offset; 1220 send = sbp->bio_offset + sbp->bio_length; 1221 if (rend > sstart && rstart < send) 1222 return (1); 1223 } 1224 } 1225 return (0); 1226 } 1227 1228 /* 1229 * Return TRUE if the given sync request is colliding with a in-progress regular 1230 * request. 1231 */ 1232 static int 1233 g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1234 { 1235 off_t rstart, rend, sstart, send; 1236 struct bio *bp; 1237 1238 if (sc->sc_sync.ds_ndisks == 0) 1239 return (0); 1240 sstart = sbp->bio_offset; 1241 send = sbp->bio_offset + sbp->bio_length; 1242 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { 1243 rstart = bp->bio_offset; 1244 rend = bp->bio_offset + bp->bio_length; 1245 if (rend > sstart && rstart < send) 1246 return (1); 1247 } 1248 return (0); 1249 } 1250 1251 /* 1252 * Puts request onto delayed queue. 1253 */ 1254 static void 1255 g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1256 { 1257 1258 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1259 bioq_insert_head(&sc->sc_regular_delayed, bp); 1260 } 1261 1262 /* 1263 * Puts synchronization request onto delayed queue. 1264 */ 1265 static void 1266 g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1267 { 1268 1269 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1270 bioq_insert_tail(&sc->sc_sync_delayed, bp); 1271 } 1272 1273 /* 1274 * Releases delayed regular requests which don't collide anymore with sync 1275 * requests. 1276 */ 1277 static void 1278 g_mirror_regular_release(struct g_mirror_softc *sc) 1279 { 1280 struct bio *bp, *bp2; 1281 1282 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { 1283 if (g_mirror_sync_collision(sc, bp)) 1284 continue; 1285 bioq_remove(&sc->sc_regular_delayed, bp); 1286 G_MIRROR_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1287 mtx_lock(&sc->sc_queue_mtx); 1288 bioq_insert_head(&sc->sc_queue, bp); 1289 mtx_unlock(&sc->sc_queue_mtx); 1290 } 1291 } 1292 1293 /* 1294 * Releases delayed sync requests which don't collide anymore with regular 1295 * requests. 1296 */ 1297 static void 1298 g_mirror_sync_release(struct g_mirror_softc *sc) 1299 { 1300 struct bio *bp, *bp2; 1301 1302 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { 1303 if (g_mirror_regular_collision(sc, bp)) 1304 continue; 1305 bioq_remove(&sc->sc_sync_delayed, bp); 1306 G_MIRROR_LOGREQ(2, bp, 1307 "Releasing delayed synchronization request."); 1308 g_io_request(bp, bp->bio_from); 1309 } 1310 } 1311 1312 /* 1313 * Free a synchronization request and clear its slot in the array. 1314 */ 1315 static void 1316 g_mirror_sync_request_free(struct g_mirror_disk *disk, struct bio *bp) 1317 { 1318 int idx; 1319 1320 if (disk != NULL && disk->d_sync.ds_bios != NULL) { 1321 idx = (int)(uintptr_t)bp->bio_caller1; 1322 KASSERT(disk->d_sync.ds_bios[idx] == bp, 1323 ("unexpected sync BIO at %p:%d", disk, idx)); 1324 disk->d_sync.ds_bios[idx] = NULL; 1325 } 1326 free(bp->bio_data, M_MIRROR); 1327 g_destroy_bio(bp); 1328 } 1329 1330 /* 1331 * Handle synchronization requests. 1332 * Every synchronization request is two-steps process: first, READ request is 1333 * send to active provider and then WRITE request (with read data) to the provider 1334 * being synchronized. When WRITE is finished, new synchronization request is 1335 * send. 1336 */ 1337 static void 1338 g_mirror_sync_request(struct bio *bp) 1339 { 1340 struct g_mirror_softc *sc; 1341 struct g_mirror_disk *disk; 1342 struct g_mirror_disk_sync *sync; 1343 1344 bp->bio_from->index--; 1345 sc = bp->bio_from->geom->softc; 1346 disk = bp->bio_from->private; 1347 if (disk == NULL) { 1348 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1349 g_topology_lock(); 1350 g_mirror_kill_consumer(sc, bp->bio_from); 1351 g_topology_unlock(); 1352 g_mirror_sync_request_free(NULL, bp); 1353 sx_xlock(&sc->sc_lock); 1354 return; 1355 } 1356 1357 /* 1358 * Synchronization request. 1359 */ 1360 switch (bp->bio_cmd) { 1361 case BIO_READ: 1362 { 1363 struct g_consumer *cp; 1364 1365 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_read, 1366 bp->bio_error); 1367 1368 if (bp->bio_error != 0) { 1369 G_MIRROR_LOGREQ(0, bp, 1370 "Synchronization request failed (error=%d).", 1371 bp->bio_error); 1372 g_mirror_sync_request_free(disk, bp); 1373 return; 1374 } 1375 G_MIRROR_LOGREQ(3, bp, 1376 "Synchronization request half-finished."); 1377 bp->bio_cmd = BIO_WRITE; 1378 bp->bio_cflags = 0; 1379 cp = disk->d_consumer; 1380 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1381 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1382 cp->acr, cp->acw, cp->ace)); 1383 cp->index++; 1384 g_io_request(bp, cp); 1385 return; 1386 } 1387 case BIO_WRITE: 1388 { 1389 off_t offset; 1390 void *data; 1391 int i, idx; 1392 1393 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_write, 1394 bp->bio_error); 1395 1396 if (bp->bio_error != 0) { 1397 G_MIRROR_LOGREQ(0, bp, 1398 "Synchronization request failed (error=%d).", 1399 bp->bio_error); 1400 g_mirror_sync_request_free(disk, bp); 1401 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1402 g_mirror_event_send(disk, 1403 G_MIRROR_DISK_STATE_DISCONNECTED, 1404 G_MIRROR_EVENT_DONTWAIT); 1405 return; 1406 } 1407 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1408 sync = &disk->d_sync; 1409 if (sync->ds_offset >= sc->sc_mediasize || 1410 sync->ds_consumer == NULL || 1411 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1412 /* Don't send more synchronization requests. */ 1413 sync->ds_inflight--; 1414 g_mirror_sync_request_free(disk, bp); 1415 if (sync->ds_inflight > 0) 1416 return; 1417 if (sync->ds_consumer == NULL || 1418 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1419 return; 1420 } 1421 /* Disk up-to-date, activate it. */ 1422 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1423 G_MIRROR_EVENT_DONTWAIT); 1424 return; 1425 } 1426 1427 /* Send next synchronization request. */ 1428 data = bp->bio_data; 1429 idx = (int)(uintptr_t)bp->bio_caller1; 1430 g_reset_bio(bp); 1431 bp->bio_cmd = BIO_READ; 1432 bp->bio_offset = sync->ds_offset; 1433 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1434 sync->ds_offset += bp->bio_length; 1435 bp->bio_done = g_mirror_sync_done; 1436 bp->bio_data = data; 1437 bp->bio_from = sync->ds_consumer; 1438 bp->bio_to = sc->sc_provider; 1439 bp->bio_caller1 = (void *)(uintptr_t)idx; 1440 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1441 sync->ds_consumer->index++; 1442 /* 1443 * Delay the request if it is colliding with a regular request. 1444 */ 1445 if (g_mirror_regular_collision(sc, bp)) 1446 g_mirror_sync_delay(sc, bp); 1447 else 1448 g_io_request(bp, sync->ds_consumer); 1449 1450 /* Release delayed requests if possible. */ 1451 g_mirror_regular_release(sc); 1452 1453 /* Find the smallest offset */ 1454 offset = sc->sc_mediasize; 1455 for (i = 0; i < g_mirror_syncreqs; i++) { 1456 bp = sync->ds_bios[i]; 1457 if (bp != NULL && bp->bio_offset < offset) 1458 offset = bp->bio_offset; 1459 } 1460 if (sync->ds_offset_done + (MAXPHYS * 100) < offset) { 1461 /* Update offset_done on every 100 blocks. */ 1462 sync->ds_offset_done = offset; 1463 g_mirror_update_metadata(disk); 1464 } 1465 return; 1466 } 1467 default: 1468 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1469 bp->bio_cmd, sc->sc_name)); 1470 break; 1471 } 1472 } 1473 1474 static void 1475 g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1476 { 1477 struct g_mirror_disk *disk; 1478 struct g_consumer *cp; 1479 struct bio *cbp; 1480 1481 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1482 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1483 break; 1484 } 1485 if (disk == NULL) { 1486 if (bp->bio_error == 0) 1487 bp->bio_error = ENXIO; 1488 g_io_deliver(bp, bp->bio_error); 1489 return; 1490 } 1491 cbp = g_clone_bio(bp); 1492 if (cbp == NULL) { 1493 if (bp->bio_error == 0) 1494 bp->bio_error = ENOMEM; 1495 g_io_deliver(bp, bp->bio_error); 1496 return; 1497 } 1498 /* 1499 * Fill in the component buf structure. 1500 */ 1501 cp = disk->d_consumer; 1502 cbp->bio_done = g_mirror_done; 1503 cbp->bio_to = cp->provider; 1504 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1505 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1506 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1507 cp->acw, cp->ace)); 1508 cp->index++; 1509 g_io_request(cbp, cp); 1510 } 1511 1512 static void 1513 g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1514 { 1515 struct g_mirror_disk *disk; 1516 struct g_consumer *cp; 1517 struct bio *cbp; 1518 1519 disk = g_mirror_get_disk(sc); 1520 if (disk == NULL) { 1521 if (bp->bio_error == 0) 1522 bp->bio_error = ENXIO; 1523 g_io_deliver(bp, bp->bio_error); 1524 return; 1525 } 1526 cbp = g_clone_bio(bp); 1527 if (cbp == NULL) { 1528 if (bp->bio_error == 0) 1529 bp->bio_error = ENOMEM; 1530 g_io_deliver(bp, bp->bio_error); 1531 return; 1532 } 1533 /* 1534 * Fill in the component buf structure. 1535 */ 1536 cp = disk->d_consumer; 1537 cbp->bio_done = g_mirror_done; 1538 cbp->bio_to = cp->provider; 1539 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1540 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1541 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1542 cp->acw, cp->ace)); 1543 cp->index++; 1544 g_io_request(cbp, cp); 1545 } 1546 1547 #define TRACK_SIZE (1 * 1024 * 1024) 1548 #define LOAD_SCALE 256 1549 #define ABS(x) (((x) >= 0) ? (x) : (-(x))) 1550 1551 static void 1552 g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1553 { 1554 struct g_mirror_disk *disk, *dp; 1555 struct g_consumer *cp; 1556 struct bio *cbp; 1557 int prio, best; 1558 1559 /* Find a disk with the smallest load. */ 1560 disk = NULL; 1561 best = INT_MAX; 1562 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1563 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1564 continue; 1565 prio = dp->load; 1566 /* If disk head is precisely in position - highly prefer it. */ 1567 if (dp->d_last_offset == bp->bio_offset) 1568 prio -= 2 * LOAD_SCALE; 1569 else 1570 /* If disk head is close to position - prefer it. */ 1571 if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE) 1572 prio -= 1 * LOAD_SCALE; 1573 if (prio <= best) { 1574 disk = dp; 1575 best = prio; 1576 } 1577 } 1578 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1579 cbp = g_clone_bio(bp); 1580 if (cbp == NULL) { 1581 if (bp->bio_error == 0) 1582 bp->bio_error = ENOMEM; 1583 g_io_deliver(bp, bp->bio_error); 1584 return; 1585 } 1586 /* 1587 * Fill in the component buf structure. 1588 */ 1589 cp = disk->d_consumer; 1590 cbp->bio_done = g_mirror_done; 1591 cbp->bio_to = cp->provider; 1592 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1593 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1594 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1595 cp->acw, cp->ace)); 1596 cp->index++; 1597 /* Remember last head position */ 1598 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1599 /* Update loads. */ 1600 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1601 dp->load = (dp->d_consumer->index * LOAD_SCALE + 1602 dp->load * 7) / 8; 1603 } 1604 g_io_request(cbp, cp); 1605 } 1606 1607 static void 1608 g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1609 { 1610 struct bio_queue_head queue; 1611 struct g_mirror_disk *disk; 1612 struct g_consumer *cp; 1613 struct bio *cbp; 1614 off_t left, mod, offset, slice; 1615 u_char *data; 1616 u_int ndisks; 1617 1618 if (bp->bio_length <= sc->sc_slice) { 1619 g_mirror_request_round_robin(sc, bp); 1620 return; 1621 } 1622 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1623 slice = bp->bio_length / ndisks; 1624 mod = slice % sc->sc_provider->sectorsize; 1625 if (mod != 0) 1626 slice += sc->sc_provider->sectorsize - mod; 1627 /* 1628 * Allocate all bios before sending any request, so we can 1629 * return ENOMEM in nice and clean way. 1630 */ 1631 left = bp->bio_length; 1632 offset = bp->bio_offset; 1633 data = bp->bio_data; 1634 bioq_init(&queue); 1635 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1636 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1637 continue; 1638 cbp = g_clone_bio(bp); 1639 if (cbp == NULL) { 1640 while ((cbp = bioq_takefirst(&queue)) != NULL) 1641 g_destroy_bio(cbp); 1642 if (bp->bio_error == 0) 1643 bp->bio_error = ENOMEM; 1644 g_io_deliver(bp, bp->bio_error); 1645 return; 1646 } 1647 bioq_insert_tail(&queue, cbp); 1648 cbp->bio_done = g_mirror_done; 1649 cbp->bio_caller1 = disk; 1650 cbp->bio_to = disk->d_consumer->provider; 1651 cbp->bio_offset = offset; 1652 cbp->bio_data = data; 1653 cbp->bio_length = MIN(left, slice); 1654 left -= cbp->bio_length; 1655 if (left == 0) 1656 break; 1657 offset += cbp->bio_length; 1658 data += cbp->bio_length; 1659 } 1660 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1661 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1662 disk = cbp->bio_caller1; 1663 cbp->bio_caller1 = NULL; 1664 cp = disk->d_consumer; 1665 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1666 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1667 cp->acr, cp->acw, cp->ace)); 1668 disk->d_consumer->index++; 1669 g_io_request(cbp, disk->d_consumer); 1670 } 1671 } 1672 1673 static void 1674 g_mirror_register_request(struct bio *bp) 1675 { 1676 struct g_mirror_softc *sc; 1677 1678 sc = bp->bio_to->private; 1679 switch (bp->bio_cmd) { 1680 case BIO_READ: 1681 switch (sc->sc_balance) { 1682 case G_MIRROR_BALANCE_LOAD: 1683 g_mirror_request_load(sc, bp); 1684 break; 1685 case G_MIRROR_BALANCE_PREFER: 1686 g_mirror_request_prefer(sc, bp); 1687 break; 1688 case G_MIRROR_BALANCE_ROUND_ROBIN: 1689 g_mirror_request_round_robin(sc, bp); 1690 break; 1691 case G_MIRROR_BALANCE_SPLIT: 1692 g_mirror_request_split(sc, bp); 1693 break; 1694 } 1695 return; 1696 case BIO_WRITE: 1697 case BIO_DELETE: 1698 { 1699 struct g_mirror_disk *disk; 1700 struct g_mirror_disk_sync *sync; 1701 struct bio_queue_head queue; 1702 struct g_consumer *cp; 1703 struct bio *cbp; 1704 1705 /* 1706 * Delay the request if it is colliding with a synchronization 1707 * request. 1708 */ 1709 if (g_mirror_sync_collision(sc, bp)) { 1710 g_mirror_regular_delay(sc, bp); 1711 return; 1712 } 1713 1714 if (sc->sc_idle) 1715 g_mirror_unidle(sc); 1716 else 1717 sc->sc_last_write = time_uptime; 1718 1719 /* 1720 * Bump syncid on first write. 1721 */ 1722 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1723 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1724 g_mirror_bump_syncid(sc); 1725 } 1726 1727 /* 1728 * Allocate all bios before sending any request, so we can 1729 * return ENOMEM in nice and clean way. 1730 */ 1731 bioq_init(&queue); 1732 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1733 sync = &disk->d_sync; 1734 switch (disk->d_state) { 1735 case G_MIRROR_DISK_STATE_ACTIVE: 1736 break; 1737 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1738 if (bp->bio_offset >= sync->ds_offset) 1739 continue; 1740 break; 1741 default: 1742 continue; 1743 } 1744 if (bp->bio_cmd == BIO_DELETE && 1745 (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0) 1746 continue; 1747 cbp = g_clone_bio(bp); 1748 if (cbp == NULL) { 1749 while ((cbp = bioq_takefirst(&queue)) != NULL) 1750 g_destroy_bio(cbp); 1751 if (bp->bio_error == 0) 1752 bp->bio_error = ENOMEM; 1753 g_io_deliver(bp, bp->bio_error); 1754 return; 1755 } 1756 bioq_insert_tail(&queue, cbp); 1757 cbp->bio_done = g_mirror_done; 1758 cp = disk->d_consumer; 1759 cbp->bio_caller1 = cp; 1760 cbp->bio_to = cp->provider; 1761 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1762 ("Consumer %s not opened (r%dw%de%d).", 1763 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1764 } 1765 if (bioq_first(&queue) == NULL) { 1766 g_io_deliver(bp, EOPNOTSUPP); 1767 return; 1768 } 1769 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1770 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1771 cp = cbp->bio_caller1; 1772 cbp->bio_caller1 = NULL; 1773 cp->index++; 1774 sc->sc_writes++; 1775 g_io_request(cbp, cp); 1776 } 1777 /* 1778 * Put request onto inflight queue, so we can check if new 1779 * synchronization requests don't collide with it. 1780 */ 1781 bioq_insert_tail(&sc->sc_inflight, bp); 1782 return; 1783 } 1784 default: 1785 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1786 bp->bio_cmd, sc->sc_name)); 1787 break; 1788 } 1789 } 1790 1791 static int 1792 g_mirror_can_destroy(struct g_mirror_softc *sc) 1793 { 1794 struct g_geom *gp; 1795 struct g_consumer *cp; 1796 1797 g_topology_assert(); 1798 gp = sc->sc_geom; 1799 if (gp->softc == NULL) 1800 return (1); 1801 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0) 1802 return (0); 1803 LIST_FOREACH(cp, &gp->consumer, consumer) { 1804 if (g_mirror_is_busy(sc, cp)) 1805 return (0); 1806 } 1807 gp = sc->sc_sync.ds_geom; 1808 LIST_FOREACH(cp, &gp->consumer, consumer) { 1809 if (g_mirror_is_busy(sc, cp)) 1810 return (0); 1811 } 1812 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1813 sc->sc_name); 1814 return (1); 1815 } 1816 1817 static int 1818 g_mirror_try_destroy(struct g_mirror_softc *sc) 1819 { 1820 1821 if (sc->sc_rootmount != NULL) { 1822 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1823 sc->sc_rootmount); 1824 root_mount_rel(sc->sc_rootmount); 1825 sc->sc_rootmount = NULL; 1826 } 1827 g_topology_lock(); 1828 if (!g_mirror_can_destroy(sc)) { 1829 g_topology_unlock(); 1830 return (0); 1831 } 1832 sc->sc_geom->softc = NULL; 1833 sc->sc_sync.ds_geom->softc = NULL; 1834 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DRAIN) != 0) { 1835 g_topology_unlock(); 1836 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1837 &sc->sc_worker); 1838 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1839 sx_xunlock(&sc->sc_lock); 1840 wakeup(&sc->sc_worker); 1841 sc->sc_worker = NULL; 1842 } else { 1843 g_topology_unlock(); 1844 g_mirror_destroy_device(sc); 1845 } 1846 return (1); 1847 } 1848 1849 /* 1850 * Worker thread. 1851 */ 1852 static void 1853 g_mirror_worker(void *arg) 1854 { 1855 struct g_mirror_softc *sc; 1856 struct g_mirror_event *ep; 1857 struct bio *bp; 1858 int timeout; 1859 1860 sc = arg; 1861 thread_lock(curthread); 1862 sched_prio(curthread, PRIBIO); 1863 thread_unlock(curthread); 1864 1865 sx_xlock(&sc->sc_lock); 1866 for (;;) { 1867 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1868 /* 1869 * First take a look at events. 1870 * This is important to handle events before any I/O requests. 1871 */ 1872 ep = g_mirror_event_get(sc); 1873 if (ep != NULL) { 1874 g_mirror_event_remove(sc, ep); 1875 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1876 /* Update only device status. */ 1877 G_MIRROR_DEBUG(3, 1878 "Running event for device %s.", 1879 sc->sc_name); 1880 ep->e_error = 0; 1881 g_mirror_update_device(sc, true); 1882 } else { 1883 /* Update disk status. */ 1884 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1885 g_mirror_get_diskname(ep->e_disk)); 1886 ep->e_error = g_mirror_update_disk(ep->e_disk, 1887 ep->e_state); 1888 if (ep->e_error == 0) 1889 g_mirror_update_device(sc, false); 1890 } 1891 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1892 KASSERT(ep->e_error == 0, 1893 ("Error cannot be handled.")); 1894 g_mirror_event_free(ep); 1895 } else { 1896 ep->e_flags |= G_MIRROR_EVENT_DONE; 1897 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1898 ep); 1899 mtx_lock(&sc->sc_events_mtx); 1900 wakeup(ep); 1901 mtx_unlock(&sc->sc_events_mtx); 1902 } 1903 if ((sc->sc_flags & 1904 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1905 if (g_mirror_try_destroy(sc)) { 1906 curthread->td_pflags &= ~TDP_GEOM; 1907 G_MIRROR_DEBUG(1, "Thread exiting."); 1908 kproc_exit(0); 1909 } 1910 } 1911 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1912 continue; 1913 } 1914 /* 1915 * Check if we can mark array as CLEAN and if we can't take 1916 * how much seconds should we wait. 1917 */ 1918 timeout = g_mirror_idle(sc, -1); 1919 /* 1920 * Now I/O requests. 1921 */ 1922 /* Get first request from the queue. */ 1923 mtx_lock(&sc->sc_queue_mtx); 1924 bp = bioq_takefirst(&sc->sc_queue); 1925 if (bp == NULL) { 1926 if ((sc->sc_flags & 1927 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1928 mtx_unlock(&sc->sc_queue_mtx); 1929 if (g_mirror_try_destroy(sc)) { 1930 curthread->td_pflags &= ~TDP_GEOM; 1931 G_MIRROR_DEBUG(1, "Thread exiting."); 1932 kproc_exit(0); 1933 } 1934 mtx_lock(&sc->sc_queue_mtx); 1935 if (bioq_first(&sc->sc_queue) != NULL) { 1936 mtx_unlock(&sc->sc_queue_mtx); 1937 continue; 1938 } 1939 } 1940 sx_xunlock(&sc->sc_lock); 1941 /* 1942 * XXX: We can miss an event here, because an event 1943 * can be added without sx-device-lock and without 1944 * mtx-queue-lock. Maybe I should just stop using 1945 * dedicated mutex for events synchronization and 1946 * stick with the queue lock? 1947 * The event will hang here until next I/O request 1948 * or next event is received. 1949 */ 1950 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1951 timeout * hz); 1952 sx_xlock(&sc->sc_lock); 1953 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1954 continue; 1955 } 1956 mtx_unlock(&sc->sc_queue_mtx); 1957 1958 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 1959 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1960 g_mirror_sync_request(bp); /* READ */ 1961 } else if (bp->bio_to != sc->sc_provider) { 1962 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 1963 g_mirror_regular_request(bp); 1964 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 1965 g_mirror_sync_request(bp); /* WRITE */ 1966 else { 1967 KASSERT(0, 1968 ("Invalid request cflags=0x%hx to=%s.", 1969 bp->bio_cflags, bp->bio_to->name)); 1970 } 1971 } else { 1972 g_mirror_register_request(bp); 1973 } 1974 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1975 } 1976 } 1977 1978 static void 1979 g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 1980 { 1981 1982 sx_assert(&sc->sc_lock, SX_LOCKED); 1983 1984 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 1985 return; 1986 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1987 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 1988 g_mirror_get_diskname(disk), sc->sc_name); 1989 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1990 } else if (sc->sc_idle && 1991 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1992 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 1993 g_mirror_get_diskname(disk), sc->sc_name); 1994 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1995 } 1996 } 1997 1998 static void 1999 g_mirror_sync_start(struct g_mirror_disk *disk) 2000 { 2001 struct g_mirror_softc *sc; 2002 struct g_consumer *cp; 2003 struct bio *bp; 2004 int error, i; 2005 2006 g_topology_assert_not(); 2007 sc = disk->d_softc; 2008 sx_assert(&sc->sc_lock, SX_LOCKED); 2009 2010 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2011 ("Disk %s is not marked for synchronization.", 2012 g_mirror_get_diskname(disk))); 2013 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2014 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 2015 sc->sc_state)); 2016 2017 sx_xunlock(&sc->sc_lock); 2018 g_topology_lock(); 2019 cp = g_new_consumer(sc->sc_sync.ds_geom); 2020 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 2021 error = g_attach(cp, sc->sc_provider); 2022 KASSERT(error == 0, 2023 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 2024 error = g_access(cp, 1, 0, 0); 2025 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 2026 g_topology_unlock(); 2027 sx_xlock(&sc->sc_lock); 2028 2029 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 2030 g_mirror_get_diskname(disk)); 2031 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0) 2032 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2033 KASSERT(disk->d_sync.ds_consumer == NULL, 2034 ("Sync consumer already exists (device=%s, disk=%s).", 2035 sc->sc_name, g_mirror_get_diskname(disk))); 2036 2037 disk->d_sync.ds_consumer = cp; 2038 disk->d_sync.ds_consumer->private = disk; 2039 disk->d_sync.ds_consumer->index = 0; 2040 2041 /* 2042 * Allocate memory for synchronization bios and initialize them. 2043 */ 2044 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 2045 M_MIRROR, M_WAITOK); 2046 for (i = 0; i < g_mirror_syncreqs; i++) { 2047 bp = g_alloc_bio(); 2048 disk->d_sync.ds_bios[i] = bp; 2049 bp->bio_parent = NULL; 2050 bp->bio_cmd = BIO_READ; 2051 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 2052 bp->bio_cflags = 0; 2053 bp->bio_offset = disk->d_sync.ds_offset; 2054 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 2055 disk->d_sync.ds_offset += bp->bio_length; 2056 bp->bio_done = g_mirror_sync_done; 2057 bp->bio_from = disk->d_sync.ds_consumer; 2058 bp->bio_to = sc->sc_provider; 2059 bp->bio_caller1 = (void *)(uintptr_t)i; 2060 } 2061 2062 /* Increase the number of disks in SYNCHRONIZING state. */ 2063 sc->sc_sync.ds_ndisks++; 2064 /* Set the number of in-flight synchronization requests. */ 2065 disk->d_sync.ds_inflight = g_mirror_syncreqs; 2066 2067 /* 2068 * Fire off first synchronization requests. 2069 */ 2070 for (i = 0; i < g_mirror_syncreqs; i++) { 2071 bp = disk->d_sync.ds_bios[i]; 2072 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 2073 disk->d_sync.ds_consumer->index++; 2074 /* 2075 * Delay the request if it is colliding with a regular request. 2076 */ 2077 if (g_mirror_regular_collision(sc, bp)) 2078 g_mirror_sync_delay(sc, bp); 2079 else 2080 g_io_request(bp, disk->d_sync.ds_consumer); 2081 } 2082 } 2083 2084 /* 2085 * Stop synchronization process. 2086 * type: 0 - synchronization finished 2087 * 1 - synchronization stopped 2088 */ 2089 static void 2090 g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 2091 { 2092 struct g_mirror_softc *sc; 2093 struct g_consumer *cp; 2094 2095 g_topology_assert_not(); 2096 sc = disk->d_softc; 2097 sx_assert(&sc->sc_lock, SX_LOCKED); 2098 2099 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2100 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2101 g_mirror_disk_state2str(disk->d_state))); 2102 if (disk->d_sync.ds_consumer == NULL) 2103 return; 2104 2105 if (type == 0) { 2106 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2107 sc->sc_name, g_mirror_get_diskname(disk)); 2108 } else /* if (type == 1) */ { 2109 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2110 sc->sc_name, g_mirror_get_diskname(disk)); 2111 } 2112 g_mirror_regular_release(sc); 2113 free(disk->d_sync.ds_bios, M_MIRROR); 2114 disk->d_sync.ds_bios = NULL; 2115 cp = disk->d_sync.ds_consumer; 2116 disk->d_sync.ds_consumer = NULL; 2117 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2118 sc->sc_sync.ds_ndisks--; 2119 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2120 g_topology_lock(); 2121 g_mirror_kill_consumer(sc, cp); 2122 g_topology_unlock(); 2123 sx_xlock(&sc->sc_lock); 2124 } 2125 2126 static void 2127 g_mirror_launch_provider(struct g_mirror_softc *sc) 2128 { 2129 struct g_mirror_disk *disk; 2130 struct g_provider *pp, *dp; 2131 2132 sx_assert(&sc->sc_lock, SX_LOCKED); 2133 2134 g_topology_lock(); 2135 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 2136 pp->flags |= G_PF_DIRECT_RECEIVE; 2137 pp->mediasize = sc->sc_mediasize; 2138 pp->sectorsize = sc->sc_sectorsize; 2139 pp->stripesize = 0; 2140 pp->stripeoffset = 0; 2141 2142 /* Splitting of unmapped BIO's could work but isn't implemented now */ 2143 if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT) 2144 pp->flags |= G_PF_ACCEPT_UNMAPPED; 2145 2146 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2147 if (disk->d_consumer && disk->d_consumer->provider) { 2148 dp = disk->d_consumer->provider; 2149 if (dp->stripesize > pp->stripesize) { 2150 pp->stripesize = dp->stripesize; 2151 pp->stripeoffset = dp->stripeoffset; 2152 } 2153 /* A provider underneath us doesn't support unmapped */ 2154 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { 2155 G_MIRROR_DEBUG(0, "Cancelling unmapped " 2156 "because of %s.", dp->name); 2157 pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 2158 } 2159 } 2160 } 2161 pp->private = sc; 2162 sc->sc_refcnt++; 2163 sc->sc_provider = pp; 2164 g_error_provider(pp, 0); 2165 g_topology_unlock(); 2166 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2167 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 2168 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2169 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2170 g_mirror_sync_start(disk); 2171 } 2172 } 2173 2174 static void 2175 g_mirror_destroy_provider(struct g_mirror_softc *sc) 2176 { 2177 struct g_mirror_disk *disk; 2178 struct bio *bp; 2179 2180 g_topology_assert_not(); 2181 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2182 sc->sc_name)); 2183 2184 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2185 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2186 g_mirror_sync_stop(disk, 1); 2187 } 2188 2189 g_topology_lock(); 2190 g_error_provider(sc->sc_provider, ENXIO); 2191 mtx_lock(&sc->sc_queue_mtx); 2192 while ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) { 2193 /* 2194 * Abort any pending I/O that wasn't generated by us. 2195 * Synchronization requests and requests destined for individual 2196 * mirror components can be destroyed immediately. 2197 */ 2198 if (bp->bio_to == sc->sc_provider && 2199 bp->bio_from->geom != sc->sc_sync.ds_geom) { 2200 g_io_deliver(bp, ENXIO); 2201 } else { 2202 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2203 free(bp->bio_data, M_MIRROR); 2204 g_destroy_bio(bp); 2205 } 2206 } 2207 mtx_unlock(&sc->sc_queue_mtx); 2208 g_wither_provider(sc->sc_provider, ENXIO); 2209 sc->sc_provider = NULL; 2210 G_MIRROR_DEBUG(0, "Device %s: provider destroyed.", sc->sc_name); 2211 g_topology_unlock(); 2212 } 2213 2214 static void 2215 g_mirror_go(void *arg) 2216 { 2217 struct g_mirror_softc *sc; 2218 2219 sc = arg; 2220 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2221 g_mirror_event_send(sc, 0, 2222 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2223 } 2224 2225 static u_int 2226 g_mirror_determine_state(struct g_mirror_disk *disk) 2227 { 2228 struct g_mirror_softc *sc; 2229 u_int state; 2230 2231 sc = disk->d_softc; 2232 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2233 if ((disk->d_flags & 2234 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0 && 2235 (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 || 2236 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0)) { 2237 /* Disk does not need synchronization. */ 2238 state = G_MIRROR_DISK_STATE_ACTIVE; 2239 } else { 2240 if ((sc->sc_flags & 2241 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2242 (disk->d_flags & 2243 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2244 /* 2245 * We can start synchronization from 2246 * the stored offset. 2247 */ 2248 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2249 } else { 2250 state = G_MIRROR_DISK_STATE_STALE; 2251 } 2252 } 2253 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2254 /* 2255 * Reset all synchronization data for this disk, 2256 * because if it even was synchronized, it was 2257 * synchronized to disks with different syncid. 2258 */ 2259 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2260 disk->d_sync.ds_offset = 0; 2261 disk->d_sync.ds_offset_done = 0; 2262 disk->d_sync.ds_syncid = sc->sc_syncid; 2263 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2264 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2265 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2266 } else { 2267 state = G_MIRROR_DISK_STATE_STALE; 2268 } 2269 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2270 /* 2271 * Not good, NOT GOOD! 2272 * It means that mirror was started on stale disks 2273 * and more fresh disk just arrive. 2274 * If there were writes, mirror is broken, sorry. 2275 * I think the best choice here is don't touch 2276 * this disk and inform the user loudly. 2277 */ 2278 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2279 "disk (%s) arrives!! It will not be connected to the " 2280 "running device.", sc->sc_name, 2281 g_mirror_get_diskname(disk)); 2282 g_mirror_destroy_disk(disk); 2283 state = G_MIRROR_DISK_STATE_NONE; 2284 /* Return immediately, because disk was destroyed. */ 2285 return (state); 2286 } 2287 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2288 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2289 return (state); 2290 } 2291 2292 /* 2293 * Update device state. 2294 */ 2295 static void 2296 g_mirror_update_device(struct g_mirror_softc *sc, bool force) 2297 { 2298 struct g_mirror_disk *disk; 2299 u_int state; 2300 2301 sx_assert(&sc->sc_lock, SX_XLOCKED); 2302 2303 switch (sc->sc_state) { 2304 case G_MIRROR_DEVICE_STATE_STARTING: 2305 { 2306 struct g_mirror_disk *pdisk, *tdisk; 2307 u_int dirty, ndisks, genid, syncid; 2308 bool broken; 2309 2310 KASSERT(sc->sc_provider == NULL, 2311 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2312 /* 2313 * Are we ready? We are, if all disks are connected or 2314 * if we have any disks and 'force' is true. 2315 */ 2316 ndisks = g_mirror_ndisks(sc, -1); 2317 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2318 ; 2319 } else if (ndisks == 0) { 2320 /* 2321 * Disks went down in starting phase, so destroy 2322 * device. 2323 */ 2324 callout_drain(&sc->sc_callout); 2325 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2326 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2327 sc->sc_rootmount); 2328 root_mount_rel(sc->sc_rootmount); 2329 sc->sc_rootmount = NULL; 2330 return; 2331 } else { 2332 return; 2333 } 2334 2335 /* 2336 * Activate all disks with the biggest syncid. 2337 */ 2338 if (force) { 2339 /* 2340 * If 'force' is true, we have been called due to 2341 * timeout, so don't bother canceling timeout. 2342 */ 2343 ndisks = 0; 2344 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2345 if ((disk->d_flags & 2346 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2347 ndisks++; 2348 } 2349 } 2350 if (ndisks == 0) { 2351 /* No valid disks found, destroy device. */ 2352 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2353 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2354 __LINE__, sc->sc_rootmount); 2355 root_mount_rel(sc->sc_rootmount); 2356 sc->sc_rootmount = NULL; 2357 return; 2358 } 2359 } else { 2360 /* Cancel timeout. */ 2361 callout_drain(&sc->sc_callout); 2362 } 2363 2364 /* 2365 * Find the biggest genid. 2366 */ 2367 genid = 0; 2368 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2369 if (disk->d_genid > genid) 2370 genid = disk->d_genid; 2371 } 2372 sc->sc_genid = genid; 2373 /* 2374 * Remove all disks without the biggest genid. 2375 */ 2376 broken = false; 2377 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2378 if (disk->d_genid < genid) { 2379 G_MIRROR_DEBUG(0, 2380 "Component %s (device %s) broken, skipping.", 2381 g_mirror_get_diskname(disk), sc->sc_name); 2382 g_mirror_destroy_disk(disk); 2383 /* 2384 * Bump the syncid in case we discover a healthy 2385 * replacement disk after starting the mirror. 2386 */ 2387 broken = true; 2388 } 2389 } 2390 2391 /* 2392 * Find the biggest syncid. 2393 */ 2394 syncid = 0; 2395 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2396 if (disk->d_sync.ds_syncid > syncid) 2397 syncid = disk->d_sync.ds_syncid; 2398 } 2399 2400 /* 2401 * Here we need to look for dirty disks and if all disks 2402 * with the biggest syncid are dirty, we have to choose 2403 * one with the biggest priority and rebuild the rest. 2404 */ 2405 /* 2406 * Find the number of dirty disks with the biggest syncid. 2407 * Find the number of disks with the biggest syncid. 2408 * While here, find a disk with the biggest priority. 2409 */ 2410 dirty = ndisks = 0; 2411 pdisk = NULL; 2412 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2413 if (disk->d_sync.ds_syncid != syncid) 2414 continue; 2415 if ((disk->d_flags & 2416 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2417 continue; 2418 } 2419 ndisks++; 2420 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2421 dirty++; 2422 if (pdisk == NULL || 2423 pdisk->d_priority < disk->d_priority) { 2424 pdisk = disk; 2425 } 2426 } 2427 } 2428 if (dirty == 0) { 2429 /* No dirty disks at all, great. */ 2430 } else if (dirty == ndisks) { 2431 /* 2432 * Force synchronization for all dirty disks except one 2433 * with the biggest priority. 2434 */ 2435 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2436 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2437 "master disk for synchronization.", 2438 g_mirror_get_diskname(pdisk), sc->sc_name); 2439 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2440 if (disk->d_sync.ds_syncid != syncid) 2441 continue; 2442 if ((disk->d_flags & 2443 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2444 continue; 2445 } 2446 KASSERT((disk->d_flags & 2447 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2448 ("Disk %s isn't marked as dirty.", 2449 g_mirror_get_diskname(disk))); 2450 /* Skip the disk with the biggest priority. */ 2451 if (disk == pdisk) 2452 continue; 2453 disk->d_sync.ds_syncid = 0; 2454 } 2455 } else if (dirty < ndisks) { 2456 /* 2457 * Force synchronization for all dirty disks. 2458 * We have some non-dirty disks. 2459 */ 2460 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2461 if (disk->d_sync.ds_syncid != syncid) 2462 continue; 2463 if ((disk->d_flags & 2464 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2465 continue; 2466 } 2467 if ((disk->d_flags & 2468 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2469 continue; 2470 } 2471 disk->d_sync.ds_syncid = 0; 2472 } 2473 } 2474 2475 /* Reset hint. */ 2476 sc->sc_hint = NULL; 2477 sc->sc_syncid = syncid; 2478 if (force || broken) { 2479 /* Remember to bump syncid on first write. */ 2480 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2481 } 2482 state = G_MIRROR_DEVICE_STATE_RUNNING; 2483 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2484 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2485 g_mirror_device_state2str(state)); 2486 sc->sc_state = state; 2487 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2488 state = g_mirror_determine_state(disk); 2489 g_mirror_event_send(disk, state, 2490 G_MIRROR_EVENT_DONTWAIT); 2491 if (state == G_MIRROR_DISK_STATE_STALE) 2492 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2493 } 2494 break; 2495 } 2496 case G_MIRROR_DEVICE_STATE_RUNNING: 2497 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2498 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2499 /* 2500 * No usable disks, so destroy the device. 2501 */ 2502 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2503 break; 2504 } else if (g_mirror_ndisks(sc, 2505 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2506 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2507 /* 2508 * We have active disks, launch provider if it doesn't 2509 * exist. 2510 */ 2511 if (sc->sc_provider == NULL) 2512 g_mirror_launch_provider(sc); 2513 if (sc->sc_rootmount != NULL) { 2514 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2515 __LINE__, sc->sc_rootmount); 2516 root_mount_rel(sc->sc_rootmount); 2517 sc->sc_rootmount = NULL; 2518 } 2519 } 2520 /* 2521 * Genid should be bumped immediately, so do it here. 2522 */ 2523 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2524 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2525 g_mirror_bump_genid(sc); 2526 } 2527 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_NOW) != 0) { 2528 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID_NOW; 2529 g_mirror_bump_syncid(sc); 2530 } 2531 break; 2532 default: 2533 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2534 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2535 break; 2536 } 2537 } 2538 2539 /* 2540 * Update disk state and device state if needed. 2541 */ 2542 #define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2543 "Disk %s state changed from %s to %s (device %s).", \ 2544 g_mirror_get_diskname(disk), \ 2545 g_mirror_disk_state2str(disk->d_state), \ 2546 g_mirror_disk_state2str(state), sc->sc_name) 2547 static int 2548 g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2549 { 2550 struct g_mirror_softc *sc; 2551 2552 sc = disk->d_softc; 2553 sx_assert(&sc->sc_lock, SX_XLOCKED); 2554 2555 again: 2556 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2557 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2558 g_mirror_disk_state2str(state)); 2559 switch (state) { 2560 case G_MIRROR_DISK_STATE_NEW: 2561 /* 2562 * Possible scenarios: 2563 * 1. New disk arrive. 2564 */ 2565 /* Previous state should be NONE. */ 2566 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2567 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2568 g_mirror_disk_state2str(disk->d_state))); 2569 DISK_STATE_CHANGED(); 2570 2571 disk->d_state = state; 2572 if (LIST_EMPTY(&sc->sc_disks)) 2573 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2574 else { 2575 struct g_mirror_disk *dp; 2576 2577 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2578 if (disk->d_priority >= dp->d_priority) { 2579 LIST_INSERT_BEFORE(dp, disk, d_next); 2580 dp = NULL; 2581 break; 2582 } 2583 if (LIST_NEXT(dp, d_next) == NULL) 2584 break; 2585 } 2586 if (dp != NULL) 2587 LIST_INSERT_AFTER(dp, disk, d_next); 2588 } 2589 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2590 sc->sc_name, g_mirror_get_diskname(disk)); 2591 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2592 break; 2593 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2594 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2595 g_mirror_device_state2str(sc->sc_state), 2596 g_mirror_get_diskname(disk), 2597 g_mirror_disk_state2str(disk->d_state))); 2598 state = g_mirror_determine_state(disk); 2599 if (state != G_MIRROR_DISK_STATE_NONE) 2600 goto again; 2601 break; 2602 case G_MIRROR_DISK_STATE_ACTIVE: 2603 /* 2604 * Possible scenarios: 2605 * 1. New disk does not need synchronization. 2606 * 2. Synchronization process finished successfully. 2607 */ 2608 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2609 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2610 g_mirror_device_state2str(sc->sc_state), 2611 g_mirror_get_diskname(disk), 2612 g_mirror_disk_state2str(disk->d_state))); 2613 /* Previous state should be NEW or SYNCHRONIZING. */ 2614 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2615 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2616 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2617 g_mirror_disk_state2str(disk->d_state))); 2618 DISK_STATE_CHANGED(); 2619 2620 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2621 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2622 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2623 g_mirror_sync_stop(disk, 0); 2624 } 2625 disk->d_state = state; 2626 disk->d_sync.ds_offset = 0; 2627 disk->d_sync.ds_offset_done = 0; 2628 g_mirror_update_idle(sc, disk); 2629 g_mirror_update_metadata(disk); 2630 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2631 sc->sc_name, g_mirror_get_diskname(disk)); 2632 break; 2633 case G_MIRROR_DISK_STATE_STALE: 2634 /* 2635 * Possible scenarios: 2636 * 1. Stale disk was connected. 2637 */ 2638 /* Previous state should be NEW. */ 2639 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2640 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2641 g_mirror_disk_state2str(disk->d_state))); 2642 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2643 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2644 g_mirror_device_state2str(sc->sc_state), 2645 g_mirror_get_diskname(disk), 2646 g_mirror_disk_state2str(disk->d_state))); 2647 /* 2648 * STALE state is only possible if device is marked 2649 * NOAUTOSYNC. 2650 */ 2651 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2652 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2653 g_mirror_device_state2str(sc->sc_state), 2654 g_mirror_get_diskname(disk), 2655 g_mirror_disk_state2str(disk->d_state))); 2656 DISK_STATE_CHANGED(); 2657 2658 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2659 disk->d_state = state; 2660 g_mirror_update_metadata(disk); 2661 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2662 sc->sc_name, g_mirror_get_diskname(disk)); 2663 break; 2664 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2665 /* 2666 * Possible scenarios: 2667 * 1. Disk which needs synchronization was connected. 2668 */ 2669 /* Previous state should be NEW. */ 2670 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2671 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2672 g_mirror_disk_state2str(disk->d_state))); 2673 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2674 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2675 g_mirror_device_state2str(sc->sc_state), 2676 g_mirror_get_diskname(disk), 2677 g_mirror_disk_state2str(disk->d_state))); 2678 DISK_STATE_CHANGED(); 2679 2680 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2681 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2682 disk->d_state = state; 2683 if (sc->sc_provider != NULL) { 2684 g_mirror_sync_start(disk); 2685 g_mirror_update_metadata(disk); 2686 } 2687 break; 2688 case G_MIRROR_DISK_STATE_DISCONNECTED: 2689 /* 2690 * Possible scenarios: 2691 * 1. Device wasn't running yet, but disk disappear. 2692 * 2. Disk was active and disapppear. 2693 * 3. Disk disappear during synchronization process. 2694 */ 2695 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2696 /* 2697 * Previous state should be ACTIVE, STALE or 2698 * SYNCHRONIZING. 2699 */ 2700 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2701 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2702 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2703 ("Wrong disk state (%s, %s).", 2704 g_mirror_get_diskname(disk), 2705 g_mirror_disk_state2str(disk->d_state))); 2706 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2707 /* Previous state should be NEW. */ 2708 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2709 ("Wrong disk state (%s, %s).", 2710 g_mirror_get_diskname(disk), 2711 g_mirror_disk_state2str(disk->d_state))); 2712 /* 2713 * Reset bumping syncid if disk disappeared in STARTING 2714 * state. 2715 */ 2716 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2717 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2718 #ifdef INVARIANTS 2719 } else { 2720 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2721 sc->sc_name, 2722 g_mirror_device_state2str(sc->sc_state), 2723 g_mirror_get_diskname(disk), 2724 g_mirror_disk_state2str(disk->d_state))); 2725 #endif 2726 } 2727 DISK_STATE_CHANGED(); 2728 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2729 sc->sc_name, g_mirror_get_diskname(disk)); 2730 2731 g_mirror_destroy_disk(disk); 2732 break; 2733 case G_MIRROR_DISK_STATE_DESTROY: 2734 { 2735 int error; 2736 2737 error = g_mirror_clear_metadata(disk); 2738 if (error != 0) { 2739 G_MIRROR_DEBUG(0, 2740 "Device %s: failed to clear metadata on %s: %d.", 2741 sc->sc_name, g_mirror_get_diskname(disk), error); 2742 break; 2743 } 2744 DISK_STATE_CHANGED(); 2745 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2746 sc->sc_name, g_mirror_get_diskname(disk)); 2747 2748 g_mirror_destroy_disk(disk); 2749 sc->sc_ndisks--; 2750 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2751 g_mirror_update_metadata(disk); 2752 } 2753 break; 2754 } 2755 default: 2756 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2757 break; 2758 } 2759 return (0); 2760 } 2761 #undef DISK_STATE_CHANGED 2762 2763 int 2764 g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2765 { 2766 struct g_provider *pp; 2767 u_char *buf; 2768 int error; 2769 2770 g_topology_assert(); 2771 2772 error = g_access(cp, 1, 0, 0); 2773 if (error != 0) 2774 return (error); 2775 pp = cp->provider; 2776 g_topology_unlock(); 2777 /* Metadata are stored on last sector. */ 2778 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2779 &error); 2780 g_topology_lock(); 2781 g_access(cp, -1, 0, 0); 2782 if (buf == NULL) { 2783 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2784 cp->provider->name, error); 2785 return (error); 2786 } 2787 2788 /* Decode metadata. */ 2789 error = mirror_metadata_decode(buf, md); 2790 g_free(buf); 2791 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2792 return (EINVAL); 2793 if (md->md_version > G_MIRROR_VERSION) { 2794 G_MIRROR_DEBUG(0, 2795 "Kernel module is too old to handle metadata from %s.", 2796 cp->provider->name); 2797 return (EINVAL); 2798 } 2799 if (error != 0) { 2800 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2801 cp->provider->name); 2802 return (error); 2803 } 2804 2805 return (0); 2806 } 2807 2808 static int 2809 g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2810 struct g_mirror_metadata *md) 2811 { 2812 2813 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2814 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2815 pp->name, md->md_did); 2816 return (EEXIST); 2817 } 2818 if (md->md_all != sc->sc_ndisks) { 2819 G_MIRROR_DEBUG(1, 2820 "Invalid '%s' field on disk %s (device %s), skipping.", 2821 "md_all", pp->name, sc->sc_name); 2822 return (EINVAL); 2823 } 2824 if (md->md_slice != sc->sc_slice) { 2825 G_MIRROR_DEBUG(1, 2826 "Invalid '%s' field on disk %s (device %s), skipping.", 2827 "md_slice", pp->name, sc->sc_name); 2828 return (EINVAL); 2829 } 2830 if (md->md_balance != sc->sc_balance) { 2831 G_MIRROR_DEBUG(1, 2832 "Invalid '%s' field on disk %s (device %s), skipping.", 2833 "md_balance", pp->name, sc->sc_name); 2834 return (EINVAL); 2835 } 2836 #if 0 2837 if (md->md_mediasize != sc->sc_mediasize) { 2838 G_MIRROR_DEBUG(1, 2839 "Invalid '%s' field on disk %s (device %s), skipping.", 2840 "md_mediasize", pp->name, sc->sc_name); 2841 return (EINVAL); 2842 } 2843 #endif 2844 if (sc->sc_mediasize > pp->mediasize) { 2845 G_MIRROR_DEBUG(1, 2846 "Invalid size of disk %s (device %s), skipping.", pp->name, 2847 sc->sc_name); 2848 return (EINVAL); 2849 } 2850 if (md->md_sectorsize != sc->sc_sectorsize) { 2851 G_MIRROR_DEBUG(1, 2852 "Invalid '%s' field on disk %s (device %s), skipping.", 2853 "md_sectorsize", pp->name, sc->sc_name); 2854 return (EINVAL); 2855 } 2856 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2857 G_MIRROR_DEBUG(1, 2858 "Invalid sector size of disk %s (device %s), skipping.", 2859 pp->name, sc->sc_name); 2860 return (EINVAL); 2861 } 2862 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2863 G_MIRROR_DEBUG(1, 2864 "Invalid device flags on disk %s (device %s), skipping.", 2865 pp->name, sc->sc_name); 2866 return (EINVAL); 2867 } 2868 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2869 G_MIRROR_DEBUG(1, 2870 "Invalid disk flags on disk %s (device %s), skipping.", 2871 pp->name, sc->sc_name); 2872 return (EINVAL); 2873 } 2874 return (0); 2875 } 2876 2877 int 2878 g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2879 struct g_mirror_metadata *md) 2880 { 2881 struct g_mirror_disk *disk; 2882 int error; 2883 2884 g_topology_assert_not(); 2885 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2886 2887 error = g_mirror_check_metadata(sc, pp, md); 2888 if (error != 0) 2889 return (error); 2890 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2891 md->md_genid < sc->sc_genid) { 2892 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2893 pp->name, sc->sc_name); 2894 return (EINVAL); 2895 } 2896 disk = g_mirror_init_disk(sc, pp, md, &error); 2897 if (disk == NULL) 2898 return (error); 2899 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2900 G_MIRROR_EVENT_WAIT); 2901 if (error != 0) 2902 return (error); 2903 if (md->md_version < G_MIRROR_VERSION) { 2904 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2905 pp->name, md->md_version, G_MIRROR_VERSION); 2906 g_mirror_update_metadata(disk); 2907 } 2908 return (0); 2909 } 2910 2911 static void 2912 g_mirror_destroy_delayed(void *arg, int flag) 2913 { 2914 struct g_mirror_softc *sc; 2915 int error; 2916 2917 if (flag == EV_CANCEL) { 2918 G_MIRROR_DEBUG(1, "Destroying canceled."); 2919 return; 2920 } 2921 sc = arg; 2922 g_topology_unlock(); 2923 sx_xlock(&sc->sc_lock); 2924 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2925 ("DESTROY flag set on %s.", sc->sc_name)); 2926 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0, 2927 ("CLOSEWAIT flag not set on %s.", sc->sc_name)); 2928 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 2929 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 2930 if (error != 0) { 2931 G_MIRROR_DEBUG(0, "Cannot destroy %s (error=%d).", 2932 sc->sc_name, error); 2933 sx_xunlock(&sc->sc_lock); 2934 } 2935 g_topology_lock(); 2936 } 2937 2938 static int 2939 g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2940 { 2941 struct g_mirror_softc *sc; 2942 int error = 0; 2943 2944 g_topology_assert(); 2945 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2946 acw, ace); 2947 2948 sc = pp->private; 2949 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 2950 2951 g_topology_unlock(); 2952 sx_xlock(&sc->sc_lock); 2953 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 2954 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 || 2955 LIST_EMPTY(&sc->sc_disks)) { 2956 if (acr > 0 || acw > 0 || ace > 0) 2957 error = ENXIO; 2958 goto end; 2959 } 2960 sc->sc_provider_open += acr + acw + ace; 2961 if (pp->acw + acw == 0) 2962 g_mirror_idle(sc, 0); 2963 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 && 2964 sc->sc_provider_open == 0) 2965 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, sc, NULL); 2966 end: 2967 sx_xunlock(&sc->sc_lock); 2968 g_topology_lock(); 2969 return (error); 2970 } 2971 2972 struct g_geom * 2973 g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md, 2974 u_int type) 2975 { 2976 struct g_mirror_softc *sc; 2977 struct g_geom *gp; 2978 int error, timeout; 2979 2980 g_topology_assert(); 2981 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2982 md->md_mid); 2983 2984 /* One disk is minimum. */ 2985 if (md->md_all < 1) 2986 return (NULL); 2987 /* 2988 * Action geom. 2989 */ 2990 gp = g_new_geomf(mp, "%s", md->md_name); 2991 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2992 gp->start = g_mirror_start; 2993 gp->orphan = g_mirror_orphan; 2994 gp->access = g_mirror_access; 2995 gp->dumpconf = g_mirror_dumpconf; 2996 2997 sc->sc_type = type; 2998 sc->sc_id = md->md_mid; 2999 sc->sc_slice = md->md_slice; 3000 sc->sc_balance = md->md_balance; 3001 sc->sc_mediasize = md->md_mediasize; 3002 sc->sc_sectorsize = md->md_sectorsize; 3003 sc->sc_ndisks = md->md_all; 3004 sc->sc_flags = md->md_mflags; 3005 sc->sc_bump_id = 0; 3006 sc->sc_idle = 1; 3007 sc->sc_last_write = time_uptime; 3008 sc->sc_writes = 0; 3009 sc->sc_refcnt = 1; 3010 sx_init(&sc->sc_lock, "gmirror:lock"); 3011 bioq_init(&sc->sc_queue); 3012 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 3013 bioq_init(&sc->sc_regular_delayed); 3014 bioq_init(&sc->sc_inflight); 3015 bioq_init(&sc->sc_sync_delayed); 3016 LIST_INIT(&sc->sc_disks); 3017 TAILQ_INIT(&sc->sc_events); 3018 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 3019 callout_init(&sc->sc_callout, 1); 3020 mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF); 3021 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 3022 gp->softc = sc; 3023 sc->sc_geom = gp; 3024 sc->sc_provider = NULL; 3025 sc->sc_provider_open = 0; 3026 /* 3027 * Synchronization geom. 3028 */ 3029 gp = g_new_geomf(mp, "%s.sync", md->md_name); 3030 gp->softc = sc; 3031 gp->orphan = g_mirror_orphan; 3032 sc->sc_sync.ds_geom = gp; 3033 sc->sc_sync.ds_ndisks = 0; 3034 error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 3035 "g_mirror %s", md->md_name); 3036 if (error != 0) { 3037 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 3038 sc->sc_name); 3039 g_destroy_geom(sc->sc_sync.ds_geom); 3040 g_destroy_geom(sc->sc_geom); 3041 g_mirror_free_device(sc); 3042 return (NULL); 3043 } 3044 3045 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 3046 sc->sc_name, sc->sc_ndisks, sc->sc_id); 3047 3048 sc->sc_rootmount = root_mount_hold("GMIRROR"); 3049 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 3050 /* 3051 * Run timeout. 3052 */ 3053 timeout = g_mirror_timeout * hz; 3054 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 3055 return (sc->sc_geom); 3056 } 3057 3058 int 3059 g_mirror_destroy(struct g_mirror_softc *sc, int how) 3060 { 3061 struct g_mirror_disk *disk; 3062 3063 g_topology_assert_not(); 3064 sx_assert(&sc->sc_lock, SX_XLOCKED); 3065 3066 if (sc->sc_provider_open != 0) { 3067 switch (how) { 3068 case G_MIRROR_DESTROY_SOFT: 3069 G_MIRROR_DEBUG(1, 3070 "Device %s is still open (%d).", sc->sc_name, 3071 sc->sc_provider_open); 3072 return (EBUSY); 3073 case G_MIRROR_DESTROY_DELAYED: 3074 G_MIRROR_DEBUG(1, 3075 "Device %s will be destroyed on last close.", 3076 sc->sc_name); 3077 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 3078 if (disk->d_state == 3079 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3080 g_mirror_sync_stop(disk, 1); 3081 } 3082 } 3083 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_CLOSEWAIT; 3084 return (EBUSY); 3085 case G_MIRROR_DESTROY_HARD: 3086 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 3087 "can't be definitely removed.", sc->sc_name); 3088 } 3089 } 3090 3091 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) 3092 return (0); 3093 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 3094 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DRAIN; 3095 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3096 sx_xunlock(&sc->sc_lock); 3097 mtx_lock(&sc->sc_queue_mtx); 3098 wakeup(sc); 3099 mtx_unlock(&sc->sc_queue_mtx); 3100 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3101 while (sc->sc_worker != NULL) 3102 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 3103 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3104 sx_xlock(&sc->sc_lock); 3105 g_mirror_destroy_device(sc); 3106 return (0); 3107 } 3108 3109 static void 3110 g_mirror_taste_orphan(struct g_consumer *cp) 3111 { 3112 3113 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3114 cp->provider->name)); 3115 } 3116 3117 static struct g_geom * 3118 g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3119 { 3120 struct g_mirror_metadata md; 3121 struct g_mirror_softc *sc; 3122 struct g_consumer *cp; 3123 struct g_geom *gp; 3124 int error; 3125 3126 g_topology_assert(); 3127 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3128 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 3129 3130 gp = g_new_geomf(mp, "mirror:taste"); 3131 /* 3132 * This orphan function should be never called. 3133 */ 3134 gp->orphan = g_mirror_taste_orphan; 3135 cp = g_new_consumer(gp); 3136 g_attach(cp, pp); 3137 error = g_mirror_read_metadata(cp, &md); 3138 g_detach(cp); 3139 g_destroy_consumer(cp); 3140 g_destroy_geom(gp); 3141 if (error != 0) 3142 return (NULL); 3143 gp = NULL; 3144 3145 if (md.md_provider[0] != '\0' && 3146 !g_compare_names(md.md_provider, pp->name)) 3147 return (NULL); 3148 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3149 return (NULL); 3150 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 3151 G_MIRROR_DEBUG(0, 3152 "Device %s: provider %s marked as inactive, skipping.", 3153 md.md_name, pp->name); 3154 return (NULL); 3155 } 3156 if (g_mirror_debug >= 2) 3157 mirror_metadata_dump(&md); 3158 3159 /* 3160 * Let's check if device already exists. 3161 */ 3162 sc = NULL; 3163 LIST_FOREACH(gp, &mp->geom, geom) { 3164 sc = gp->softc; 3165 if (sc == NULL) 3166 continue; 3167 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 3168 continue; 3169 if (sc->sc_sync.ds_geom == gp) 3170 continue; 3171 if (strcmp(md.md_name, sc->sc_name) != 0) 3172 continue; 3173 if (md.md_mid != sc->sc_id) { 3174 G_MIRROR_DEBUG(0, "Device %s already configured.", 3175 sc->sc_name); 3176 return (NULL); 3177 } 3178 break; 3179 } 3180 if (gp == NULL) { 3181 gp = g_mirror_create(mp, &md, G_MIRROR_TYPE_AUTOMATIC); 3182 if (gp == NULL) { 3183 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3184 md.md_name); 3185 return (NULL); 3186 } 3187 sc = gp->softc; 3188 } 3189 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3190 g_topology_unlock(); 3191 sx_xlock(&sc->sc_lock); 3192 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; 3193 error = g_mirror_add_disk(sc, pp, &md); 3194 if (error != 0) { 3195 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3196 pp->name, gp->name, error); 3197 if (LIST_EMPTY(&sc->sc_disks)) { 3198 g_cancel_event(sc); 3199 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3200 g_topology_lock(); 3201 return (NULL); 3202 } 3203 gp = NULL; 3204 } 3205 sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; 3206 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3207 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3208 g_topology_lock(); 3209 return (NULL); 3210 } 3211 sx_xunlock(&sc->sc_lock); 3212 g_topology_lock(); 3213 return (gp); 3214 } 3215 3216 static void 3217 g_mirror_resize(struct g_consumer *cp) 3218 { 3219 struct g_mirror_disk *disk; 3220 3221 g_topology_assert(); 3222 g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name); 3223 3224 disk = cp->private; 3225 if (disk == NULL) 3226 return; 3227 g_topology_unlock(); 3228 g_mirror_update_metadata(disk); 3229 g_topology_lock(); 3230 } 3231 3232 static int 3233 g_mirror_destroy_geom(struct gctl_req *req __unused, 3234 struct g_class *mp __unused, struct g_geom *gp) 3235 { 3236 struct g_mirror_softc *sc; 3237 int error; 3238 3239 g_topology_unlock(); 3240 sc = gp->softc; 3241 sx_xlock(&sc->sc_lock); 3242 g_cancel_event(sc); 3243 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3244 if (error != 0) 3245 sx_xunlock(&sc->sc_lock); 3246 g_topology_lock(); 3247 return (error); 3248 } 3249 3250 static void 3251 g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3252 struct g_consumer *cp, struct g_provider *pp) 3253 { 3254 struct g_mirror_softc *sc; 3255 3256 g_topology_assert(); 3257 3258 sc = gp->softc; 3259 if (sc == NULL) 3260 return; 3261 /* Skip synchronization geom. */ 3262 if (gp == sc->sc_sync.ds_geom) 3263 return; 3264 if (pp != NULL) { 3265 /* Nothing here. */ 3266 } else if (cp != NULL) { 3267 struct g_mirror_disk *disk; 3268 3269 disk = cp->private; 3270 if (disk == NULL) 3271 return; 3272 g_topology_unlock(); 3273 sx_xlock(&sc->sc_lock); 3274 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3275 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3276 sbuf_printf(sb, "%s<Synchronized>", indent); 3277 if (disk->d_sync.ds_offset == 0) 3278 sbuf_printf(sb, "0%%"); 3279 else { 3280 sbuf_printf(sb, "%u%%", 3281 (u_int)((disk->d_sync.ds_offset * 100) / 3282 sc->sc_provider->mediasize)); 3283 } 3284 sbuf_printf(sb, "</Synchronized>\n"); 3285 if (disk->d_sync.ds_offset > 0) { 3286 sbuf_printf(sb, "%s<BytesSynced>%jd" 3287 "</BytesSynced>\n", indent, 3288 (intmax_t)disk->d_sync.ds_offset); 3289 } 3290 } 3291 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3292 disk->d_sync.ds_syncid); 3293 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3294 disk->d_genid); 3295 sbuf_printf(sb, "%s<Flags>", indent); 3296 if (disk->d_flags == 0) 3297 sbuf_printf(sb, "NONE"); 3298 else { 3299 int first = 1; 3300 3301 #define ADD_FLAG(flag, name) do { \ 3302 if ((disk->d_flags & (flag)) != 0) { \ 3303 if (!first) \ 3304 sbuf_printf(sb, ", "); \ 3305 else \ 3306 first = 0; \ 3307 sbuf_printf(sb, name); \ 3308 } \ 3309 } while (0) 3310 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3311 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3312 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3313 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3314 "SYNCHRONIZING"); 3315 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3316 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3317 #undef ADD_FLAG 3318 } 3319 sbuf_printf(sb, "</Flags>\n"); 3320 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3321 disk->d_priority); 3322 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3323 g_mirror_disk_state2str(disk->d_state)); 3324 sx_xunlock(&sc->sc_lock); 3325 g_topology_lock(); 3326 } else { 3327 g_topology_unlock(); 3328 sx_xlock(&sc->sc_lock); 3329 sbuf_printf(sb, "%s<Type>", indent); 3330 switch (sc->sc_type) { 3331 case G_MIRROR_TYPE_AUTOMATIC: 3332 sbuf_printf(sb, "AUTOMATIC"); 3333 break; 3334 case G_MIRROR_TYPE_MANUAL: 3335 sbuf_printf(sb, "MANUAL"); 3336 break; 3337 default: 3338 sbuf_printf(sb, "UNKNOWN"); 3339 break; 3340 } 3341 sbuf_printf(sb, "</Type>\n"); 3342 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3343 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3344 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3345 sbuf_printf(sb, "%s<Flags>", indent); 3346 if (sc->sc_flags == 0) 3347 sbuf_printf(sb, "NONE"); 3348 else { 3349 int first = 1; 3350 3351 #define ADD_FLAG(flag, name) do { \ 3352 if ((sc->sc_flags & (flag)) != 0) { \ 3353 if (!first) \ 3354 sbuf_printf(sb, ", "); \ 3355 else \ 3356 first = 0; \ 3357 sbuf_printf(sb, name); \ 3358 } \ 3359 } while (0) 3360 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3361 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3362 #undef ADD_FLAG 3363 } 3364 sbuf_printf(sb, "</Flags>\n"); 3365 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3366 (u_int)sc->sc_slice); 3367 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3368 balance_name(sc->sc_balance)); 3369 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3370 sc->sc_ndisks); 3371 sbuf_printf(sb, "%s<State>", indent); 3372 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3373 sbuf_printf(sb, "%s", "STARTING"); 3374 else if (sc->sc_ndisks == 3375 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3376 sbuf_printf(sb, "%s", "COMPLETE"); 3377 else 3378 sbuf_printf(sb, "%s", "DEGRADED"); 3379 sbuf_printf(sb, "</State>\n"); 3380 sx_xunlock(&sc->sc_lock); 3381 g_topology_lock(); 3382 } 3383 } 3384 3385 static void 3386 g_mirror_shutdown_post_sync(void *arg, int howto) 3387 { 3388 struct g_class *mp; 3389 struct g_geom *gp, *gp2; 3390 struct g_mirror_softc *sc; 3391 int error; 3392 3393 if (panicstr != NULL) 3394 return; 3395 3396 mp = arg; 3397 g_topology_lock(); 3398 g_mirror_shutdown = 1; 3399 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3400 if ((sc = gp->softc) == NULL) 3401 continue; 3402 /* Skip synchronization geom. */ 3403 if (gp == sc->sc_sync.ds_geom) 3404 continue; 3405 g_topology_unlock(); 3406 sx_xlock(&sc->sc_lock); 3407 g_mirror_idle(sc, -1); 3408 g_cancel_event(sc); 3409 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3410 if (error != 0) 3411 sx_xunlock(&sc->sc_lock); 3412 g_topology_lock(); 3413 } 3414 g_topology_unlock(); 3415 } 3416 3417 static void 3418 g_mirror_init(struct g_class *mp) 3419 { 3420 3421 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3422 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3423 if (g_mirror_post_sync == NULL) 3424 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3425 } 3426 3427 static void 3428 g_mirror_fini(struct g_class *mp) 3429 { 3430 3431 if (g_mirror_post_sync != NULL) 3432 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); 3433 } 3434 3435 DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3436