g_mirror.c revision 162282
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: head/sys/geom/mirror/g_mirror.c 162282 2006-09-13 15:46:49Z pjd $"); 29 30#include <sys/param.h> 31#include <sys/systm.h> 32#include <sys/kernel.h> 33#include <sys/module.h> 34#include <sys/limits.h> 35#include <sys/lock.h> 36#include <sys/mutex.h> 37#include <sys/bio.h> 38#include <sys/sysctl.h> 39#include <sys/malloc.h> 40#include <sys/eventhandler.h> 41#include <vm/uma.h> 42#include <geom/geom.h> 43#include <sys/proc.h> 44#include <sys/kthread.h> 45#include <sys/sched.h> 46#include <geom/mirror/g_mirror.h> 47 48 49static MALLOC_DEFINE(M_MIRROR, "mirror_data", "GEOM_MIRROR Data"); 50 51SYSCTL_DECL(_kern_geom); 52SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, "GEOM_MIRROR stuff"); 53u_int g_mirror_debug = 0; 54TUNABLE_INT("kern.geom.mirror.debug", &g_mirror_debug); 55SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RW, &g_mirror_debug, 0, 56 "Debug level"); 57static u_int g_mirror_timeout = 4; 58TUNABLE_INT("kern.geom.mirror.timeout", &g_mirror_timeout); 59SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RW, &g_mirror_timeout, 60 0, "Time to wait on all mirror components"); 61static u_int g_mirror_idletime = 5; 62TUNABLE_INT("kern.geom.mirror.idletime", &g_mirror_idletime); 63SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RW, 64 &g_mirror_idletime, 0, "Mark components as clean when idling"); 65static u_int g_mirror_disconnect_on_failure = 1; 66TUNABLE_INT("kern.geom.mirror.disconnect_on_failure", 67 &g_mirror_disconnect_on_failure); 68SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, disconnect_on_failure, CTLFLAG_RW, 69 &g_mirror_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 70static u_int g_mirror_syncreqs = 2; 71TUNABLE_INT("kern.geom.mirror.sync_requests", &g_mirror_syncreqs); 72SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_requests, CTLFLAG_RDTUN, 73 &g_mirror_syncreqs, 0, "Parallel synchronization I/O requests."); 74 75#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 76 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 77 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 78 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 79} while (0) 80 81static eventhandler_tag g_mirror_pre_sync = NULL; 82 83static int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp, 84 struct g_geom *gp); 85static g_taste_t g_mirror_taste; 86static void g_mirror_init(struct g_class *mp); 87static void g_mirror_fini(struct g_class *mp); 88 89struct 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}; 98 99 100static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 101static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 102static void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force); 103static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 104 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 105static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 106static void g_mirror_register_request(struct bio *bp); 107static void g_mirror_sync_release(struct g_mirror_softc *sc); 108 109 110static const char * 111g_mirror_disk_state2str(int state) 112{ 113 114 switch (state) { 115 case G_MIRROR_DISK_STATE_NONE: 116 return ("NONE"); 117 case G_MIRROR_DISK_STATE_NEW: 118 return ("NEW"); 119 case G_MIRROR_DISK_STATE_ACTIVE: 120 return ("ACTIVE"); 121 case G_MIRROR_DISK_STATE_STALE: 122 return ("STALE"); 123 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 124 return ("SYNCHRONIZING"); 125 case G_MIRROR_DISK_STATE_DISCONNECTED: 126 return ("DISCONNECTED"); 127 case G_MIRROR_DISK_STATE_DESTROY: 128 return ("DESTROY"); 129 default: 130 return ("INVALID"); 131 } 132} 133 134static const char * 135g_mirror_device_state2str(int state) 136{ 137 138 switch (state) { 139 case G_MIRROR_DEVICE_STATE_STARTING: 140 return ("STARTING"); 141 case G_MIRROR_DEVICE_STATE_RUNNING: 142 return ("RUNNING"); 143 default: 144 return ("INVALID"); 145 } 146} 147 148static const char * 149g_mirror_get_diskname(struct g_mirror_disk *disk) 150{ 151 152 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 153 return ("[unknown]"); 154 return (disk->d_name); 155} 156 157/* 158 * --- Events handling functions --- 159 * Events in geom_mirror are used to maintain disks and device status 160 * from one thread to simplify locking. 161 */ 162static void 163g_mirror_event_free(struct g_mirror_event *ep) 164{ 165 166 free(ep, M_MIRROR); 167} 168 169int 170g_mirror_event_send(void *arg, int state, int flags) 171{ 172 struct g_mirror_softc *sc; 173 struct g_mirror_disk *disk; 174 struct g_mirror_event *ep; 175 int error; 176 177 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 178 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 179 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 180 disk = NULL; 181 sc = arg; 182 } else { 183 disk = arg; 184 sc = disk->d_softc; 185 } 186 ep->e_disk = disk; 187 ep->e_state = state; 188 ep->e_flags = flags; 189 ep->e_error = 0; 190 mtx_lock(&sc->sc_events_mtx); 191 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 192 mtx_unlock(&sc->sc_events_mtx); 193 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 194 mtx_lock(&sc->sc_queue_mtx); 195 wakeup(sc); 196 mtx_unlock(&sc->sc_queue_mtx); 197 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 198 return (0); 199 sx_assert(&sc->sc_lock, SX_XLOCKED); 200 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 201 sx_xunlock(&sc->sc_lock); 202 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 203 mtx_lock(&sc->sc_events_mtx); 204 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 205 hz * 5); 206 } 207 error = ep->e_error; 208 g_mirror_event_free(ep); 209 sx_xlock(&sc->sc_lock); 210 return (error); 211} 212 213static struct g_mirror_event * 214g_mirror_event_get(struct g_mirror_softc *sc) 215{ 216 struct g_mirror_event *ep; 217 218 mtx_lock(&sc->sc_events_mtx); 219 ep = TAILQ_FIRST(&sc->sc_events); 220 mtx_unlock(&sc->sc_events_mtx); 221 return (ep); 222} 223 224static void 225g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 226{ 227 228 mtx_lock(&sc->sc_events_mtx); 229 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 230 mtx_unlock(&sc->sc_events_mtx); 231} 232 233static void 234g_mirror_event_cancel(struct g_mirror_disk *disk) 235{ 236 struct g_mirror_softc *sc; 237 struct g_mirror_event *ep, *tmpep; 238 239 sc = disk->d_softc; 240 sx_assert(&sc->sc_lock, SX_XLOCKED); 241 242 mtx_lock(&sc->sc_events_mtx); 243 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 244 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 245 continue; 246 if (ep->e_disk != disk) 247 continue; 248 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 249 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 250 g_mirror_event_free(ep); 251 else { 252 ep->e_error = ECANCELED; 253 wakeup(ep); 254 } 255 } 256 mtx_unlock(&sc->sc_events_mtx); 257} 258 259/* 260 * Return the number of disks in given state. 261 * If state is equal to -1, count all connected disks. 262 */ 263u_int 264g_mirror_ndisks(struct g_mirror_softc *sc, int state) 265{ 266 struct g_mirror_disk *disk; 267 u_int n = 0; 268 269 sx_assert(&sc->sc_lock, SX_LOCKED); 270 271 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 272 if (state == -1 || disk->d_state == state) 273 n++; 274 } 275 return (n); 276} 277 278/* 279 * Find a disk in mirror by its disk ID. 280 */ 281static struct g_mirror_disk * 282g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 283{ 284 struct g_mirror_disk *disk; 285 286 sx_assert(&sc->sc_lock, SX_XLOCKED); 287 288 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 289 if (disk->d_id == id) 290 return (disk); 291 } 292 return (NULL); 293} 294 295static u_int 296g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 297{ 298 struct bio *bp; 299 u_int nreqs = 0; 300 301 mtx_lock(&sc->sc_queue_mtx); 302 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 303 if (bp->bio_from == cp) 304 nreqs++; 305 } 306 mtx_unlock(&sc->sc_queue_mtx); 307 return (nreqs); 308} 309 310static int 311g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 312{ 313 314 if (cp->index > 0) { 315 G_MIRROR_DEBUG(2, 316 "I/O requests for %s exist, can't destroy it now.", 317 cp->provider->name); 318 return (1); 319 } 320 if (g_mirror_nrequests(sc, cp) > 0) { 321 G_MIRROR_DEBUG(2, 322 "I/O requests for %s in queue, can't destroy it now.", 323 cp->provider->name); 324 return (1); 325 } 326 return (0); 327} 328 329static void 330g_mirror_destroy_consumer(void *arg, int flags __unused) 331{ 332 struct g_consumer *cp; 333 334 g_topology_assert(); 335 336 cp = arg; 337 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 338 g_detach(cp); 339 g_destroy_consumer(cp); 340} 341 342static void 343g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 344{ 345 struct g_provider *pp; 346 int retaste_wait; 347 348 g_topology_assert(); 349 350 cp->private = NULL; 351 if (g_mirror_is_busy(sc, cp)) 352 return; 353 pp = cp->provider; 354 retaste_wait = 0; 355 if (cp->acw == 1) { 356 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 357 retaste_wait = 1; 358 } 359 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 360 -cp->acw, -cp->ace, 0); 361 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 362 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 363 if (retaste_wait) { 364 /* 365 * After retaste event was send (inside g_access()), we can send 366 * event to detach and destroy consumer. 367 * A class, which has consumer to the given provider connected 368 * will not receive retaste event for the provider. 369 * This is the way how I ignore retaste events when I close 370 * consumers opened for write: I detach and destroy consumer 371 * after retaste event is sent. 372 */ 373 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 374 return; 375 } 376 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 377 g_detach(cp); 378 g_destroy_consumer(cp); 379} 380 381static int 382g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 383{ 384 struct g_consumer *cp; 385 int error; 386 387 g_topology_assert_not(); 388 KASSERT(disk->d_consumer == NULL, 389 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 390 391 g_topology_lock(); 392 cp = g_new_consumer(disk->d_softc->sc_geom); 393 error = g_attach(cp, pp); 394 if (error != 0) { 395 g_destroy_consumer(cp); 396 g_topology_unlock(); 397 return (error); 398 } 399 error = g_access(cp, 1, 1, 1); 400 if (error != 0) { 401 g_detach(cp); 402 g_destroy_consumer(cp); 403 g_topology_unlock(); 404 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 405 pp->name, error); 406 return (error); 407 } 408 g_topology_unlock(); 409 disk->d_consumer = cp; 410 disk->d_consumer->private = disk; 411 disk->d_consumer->index = 0; 412 413 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 414 return (0); 415} 416 417static void 418g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 419{ 420 421 g_topology_assert(); 422 423 if (cp == NULL) 424 return; 425 if (cp->provider != NULL) 426 g_mirror_kill_consumer(sc, cp); 427 else 428 g_destroy_consumer(cp); 429} 430 431/* 432 * Initialize disk. This means allocate memory, create consumer, attach it 433 * to the provider and open access (r1w1e1) to it. 434 */ 435static struct g_mirror_disk * 436g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 437 struct g_mirror_metadata *md, int *errorp) 438{ 439 struct g_mirror_disk *disk; 440 int error; 441 442 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 443 if (disk == NULL) { 444 error = ENOMEM; 445 goto fail; 446 } 447 disk->d_softc = sc; 448 error = g_mirror_connect_disk(disk, pp); 449 if (error != 0) 450 goto fail; 451 disk->d_id = md->md_did; 452 disk->d_state = G_MIRROR_DISK_STATE_NONE; 453 disk->d_priority = md->md_priority; 454 disk->d_delay.sec = 0; 455 disk->d_delay.frac = 0; 456 binuptime(&disk->d_last_used); 457 disk->d_flags = md->md_dflags; 458 if (md->md_provider[0] != '\0') 459 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 460 disk->d_sync.ds_consumer = NULL; 461 disk->d_sync.ds_offset = md->md_sync_offset; 462 disk->d_sync.ds_offset_done = md->md_sync_offset; 463 disk->d_genid = md->md_genid; 464 disk->d_sync.ds_syncid = md->md_syncid; 465 if (errorp != NULL) 466 *errorp = 0; 467 return (disk); 468fail: 469 if (errorp != NULL) 470 *errorp = error; 471 if (disk != NULL) 472 free(disk, M_MIRROR); 473 return (NULL); 474} 475 476static void 477g_mirror_destroy_disk(struct g_mirror_disk *disk) 478{ 479 struct g_mirror_softc *sc; 480 481 g_topology_assert_not(); 482 sc = disk->d_softc; 483 sx_assert(&sc->sc_lock, SX_XLOCKED); 484 485 LIST_REMOVE(disk, d_next); 486 g_mirror_event_cancel(disk); 487 if (sc->sc_hint == disk) 488 sc->sc_hint = NULL; 489 switch (disk->d_state) { 490 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 491 g_mirror_sync_stop(disk, 1); 492 /* FALLTHROUGH */ 493 case G_MIRROR_DISK_STATE_NEW: 494 case G_MIRROR_DISK_STATE_STALE: 495 case G_MIRROR_DISK_STATE_ACTIVE: 496 g_topology_lock(); 497 g_mirror_disconnect_consumer(sc, disk->d_consumer); 498 g_topology_unlock(); 499 free(disk, M_MIRROR); 500 break; 501 default: 502 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 503 g_mirror_get_diskname(disk), 504 g_mirror_disk_state2str(disk->d_state))); 505 } 506} 507 508static void 509g_mirror_destroy_device(struct g_mirror_softc *sc) 510{ 511 struct g_mirror_disk *disk; 512 struct g_mirror_event *ep; 513 struct g_geom *gp; 514 struct g_consumer *cp, *tmpcp; 515 516 g_topology_assert_not(); 517 sx_assert(&sc->sc_lock, SX_XLOCKED); 518 519 gp = sc->sc_geom; 520 if (sc->sc_provider != NULL) 521 g_mirror_destroy_provider(sc); 522 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 523 disk = LIST_FIRST(&sc->sc_disks)) { 524 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 525 g_mirror_update_metadata(disk); 526 g_mirror_destroy_disk(disk); 527 } 528 while ((ep = g_mirror_event_get(sc)) != NULL) { 529 g_mirror_event_remove(sc, ep); 530 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 531 g_mirror_event_free(ep); 532 else { 533 ep->e_error = ECANCELED; 534 ep->e_flags |= G_MIRROR_EVENT_DONE; 535 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 536 mtx_lock(&sc->sc_events_mtx); 537 wakeup(ep); 538 mtx_unlock(&sc->sc_events_mtx); 539 } 540 } 541 callout_drain(&sc->sc_callout); 542 543 g_topology_lock(); 544 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 545 g_mirror_disconnect_consumer(sc, cp); 546 } 547 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 548 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 549 g_wither_geom(gp, ENXIO); 550 g_topology_unlock(); 551 mtx_destroy(&sc->sc_queue_mtx); 552 mtx_destroy(&sc->sc_events_mtx); 553 sx_xunlock(&sc->sc_lock); 554 sx_destroy(&sc->sc_lock); 555} 556 557static void 558g_mirror_orphan(struct g_consumer *cp) 559{ 560 struct g_mirror_disk *disk; 561 562 g_topology_assert(); 563 564 disk = cp->private; 565 if (disk == NULL) 566 return; 567 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 568 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 569 G_MIRROR_EVENT_DONTWAIT); 570} 571 572/* 573 * Function should return the next active disk on the list. 574 * It is possible that it will be the same disk as given. 575 * If there are no active disks on list, NULL is returned. 576 */ 577static __inline struct g_mirror_disk * 578g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 579{ 580 struct g_mirror_disk *dp; 581 582 for (dp = LIST_NEXT(disk, d_next); dp != disk; 583 dp = LIST_NEXT(dp, d_next)) { 584 if (dp == NULL) 585 dp = LIST_FIRST(&sc->sc_disks); 586 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 587 break; 588 } 589 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 590 return (NULL); 591 return (dp); 592} 593 594static struct g_mirror_disk * 595g_mirror_get_disk(struct g_mirror_softc *sc) 596{ 597 struct g_mirror_disk *disk; 598 599 if (sc->sc_hint == NULL) { 600 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 601 if (sc->sc_hint == NULL) 602 return (NULL); 603 } 604 disk = sc->sc_hint; 605 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 606 disk = g_mirror_find_next(sc, disk); 607 if (disk == NULL) 608 return (NULL); 609 } 610 sc->sc_hint = g_mirror_find_next(sc, disk); 611 return (disk); 612} 613 614static int 615g_mirror_write_metadata(struct g_mirror_disk *disk, 616 struct g_mirror_metadata *md) 617{ 618 struct g_mirror_softc *sc; 619 struct g_consumer *cp; 620 off_t offset, length; 621 u_char *sector; 622 int error = 0; 623 624 g_topology_assert_not(); 625 sc = disk->d_softc; 626 sx_assert(&sc->sc_lock, SX_LOCKED); 627 628 cp = disk->d_consumer; 629 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 630 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 631 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 632 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 633 cp->acw, cp->ace)); 634 length = cp->provider->sectorsize; 635 offset = cp->provider->mediasize - length; 636 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 637 if (md != NULL) 638 mirror_metadata_encode(md, sector); 639 error = g_write_data(cp, offset, sector, length); 640 free(sector, M_MIRROR); 641 if (error != 0) { 642 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 643 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 644 G_MIRROR_DEBUG(0, "Cannot write metadata on %s " 645 "(device=%s, error=%d).", 646 g_mirror_get_diskname(disk), sc->sc_name, error); 647 } else { 648 G_MIRROR_DEBUG(1, "Cannot write metadata on %s " 649 "(device=%s, error=%d).", 650 g_mirror_get_diskname(disk), sc->sc_name, error); 651 } 652 if (g_mirror_disconnect_on_failure && 653 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 654 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 655 g_mirror_event_send(disk, 656 G_MIRROR_DISK_STATE_DISCONNECTED, 657 G_MIRROR_EVENT_DONTWAIT); 658 } 659 } 660 return (error); 661} 662 663static int 664g_mirror_clear_metadata(struct g_mirror_disk *disk) 665{ 666 int error; 667 668 g_topology_assert_not(); 669 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 670 671 error = g_mirror_write_metadata(disk, NULL); 672 if (error == 0) { 673 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 674 g_mirror_get_diskname(disk)); 675 } else { 676 G_MIRROR_DEBUG(0, 677 "Cannot clear metadata on disk %s (error=%d).", 678 g_mirror_get_diskname(disk), error); 679 } 680 return (error); 681} 682 683void 684g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 685 struct g_mirror_metadata *md) 686{ 687 688 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 689 md->md_version = G_MIRROR_VERSION; 690 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 691 md->md_mid = sc->sc_id; 692 md->md_all = sc->sc_ndisks; 693 md->md_slice = sc->sc_slice; 694 md->md_balance = sc->sc_balance; 695 md->md_genid = sc->sc_genid; 696 md->md_mediasize = sc->sc_mediasize; 697 md->md_sectorsize = sc->sc_sectorsize; 698 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 699 bzero(md->md_provider, sizeof(md->md_provider)); 700 if (disk == NULL) { 701 md->md_did = arc4random(); 702 md->md_priority = 0; 703 md->md_syncid = 0; 704 md->md_dflags = 0; 705 md->md_sync_offset = 0; 706 md->md_provsize = 0; 707 } else { 708 md->md_did = disk->d_id; 709 md->md_priority = disk->d_priority; 710 md->md_syncid = disk->d_sync.ds_syncid; 711 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 712 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 713 md->md_sync_offset = disk->d_sync.ds_offset_done; 714 else 715 md->md_sync_offset = 0; 716 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 717 strlcpy(md->md_provider, 718 disk->d_consumer->provider->name, 719 sizeof(md->md_provider)); 720 } 721 md->md_provsize = disk->d_consumer->provider->mediasize; 722 } 723} 724 725void 726g_mirror_update_metadata(struct g_mirror_disk *disk) 727{ 728 struct g_mirror_softc *sc; 729 struct g_mirror_metadata md; 730 int error; 731 732 g_topology_assert_not(); 733 sc = disk->d_softc; 734 sx_assert(&sc->sc_lock, SX_LOCKED); 735 736 g_mirror_fill_metadata(sc, disk, &md); 737 error = g_mirror_write_metadata(disk, &md); 738 if (error == 0) { 739 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 740 g_mirror_get_diskname(disk)); 741 } else { 742 G_MIRROR_DEBUG(0, 743 "Cannot update metadata on disk %s (error=%d).", 744 g_mirror_get_diskname(disk), error); 745 } 746} 747 748static void 749g_mirror_bump_syncid(struct g_mirror_softc *sc) 750{ 751 struct g_mirror_disk *disk; 752 753 g_topology_assert_not(); 754 sx_assert(&sc->sc_lock, SX_XLOCKED); 755 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 756 ("%s called with no active disks (device=%s).", __func__, 757 sc->sc_name)); 758 759 sc->sc_syncid++; 760 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 761 sc->sc_syncid); 762 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 763 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 764 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 765 disk->d_sync.ds_syncid = sc->sc_syncid; 766 g_mirror_update_metadata(disk); 767 } 768 } 769} 770 771static void 772g_mirror_bump_genid(struct g_mirror_softc *sc) 773{ 774 struct g_mirror_disk *disk; 775 776 g_topology_assert_not(); 777 sx_assert(&sc->sc_lock, SX_XLOCKED); 778 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 779 ("%s called with no active disks (device=%s).", __func__, 780 sc->sc_name)); 781 782 sc->sc_genid++; 783 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 784 sc->sc_genid); 785 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 786 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 787 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 788 disk->d_genid = sc->sc_genid; 789 g_mirror_update_metadata(disk); 790 } 791 } 792} 793 794static int 795g_mirror_idle(struct g_mirror_softc *sc, int acw) 796{ 797 struct g_mirror_disk *disk; 798 int timeout; 799 800 g_topology_assert_not(); 801 sx_assert(&sc->sc_lock, SX_XLOCKED); 802 803 if (sc->sc_provider == NULL) 804 return (0); 805 if (sc->sc_idle) 806 return (0); 807 if (sc->sc_writes > 0) 808 return (0); 809 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 810 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); 811 if (timeout > 0) 812 return (timeout); 813 } 814 sc->sc_idle = 1; 815 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 816 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 817 continue; 818 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 819 g_mirror_get_diskname(disk), sc->sc_name); 820 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 821 g_mirror_update_metadata(disk); 822 } 823 return (0); 824} 825 826static void 827g_mirror_unidle(struct g_mirror_softc *sc) 828{ 829 struct g_mirror_disk *disk; 830 831 g_topology_assert_not(); 832 sx_assert(&sc->sc_lock, SX_XLOCKED); 833 834 sc->sc_idle = 0; 835 sc->sc_last_write = time_uptime; 836 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 837 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 838 continue; 839 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 840 g_mirror_get_diskname(disk), sc->sc_name); 841 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 842 g_mirror_update_metadata(disk); 843 } 844} 845 846static __inline int 847bintime_cmp(struct bintime *bt1, struct bintime *bt2) 848{ 849 850 if (bt1->sec < bt2->sec) 851 return (-1); 852 else if (bt1->sec > bt2->sec) 853 return (1); 854 if (bt1->frac < bt2->frac) 855 return (-1); 856 else if (bt1->frac > bt2->frac) 857 return (1); 858 return (0); 859} 860 861static void 862g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 863{ 864 865 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 866 return; 867 binuptime(&disk->d_delay); 868 bintime_sub(&disk->d_delay, &bp->bio_t0); 869} 870 871static void 872g_mirror_done(struct bio *bp) 873{ 874 struct g_mirror_softc *sc; 875 876 sc = bp->bio_from->geom->softc; 877 bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR; 878 mtx_lock(&sc->sc_queue_mtx); 879 bioq_disksort(&sc->sc_queue, bp); 880 wakeup(sc); 881 mtx_unlock(&sc->sc_queue_mtx); 882} 883 884static void 885g_mirror_regular_request(struct bio *bp) 886{ 887 struct g_mirror_softc *sc; 888 struct g_mirror_disk *disk; 889 struct bio *pbp; 890 891 g_topology_assert_not(); 892 893 pbp = bp->bio_parent; 894 sc = pbp->bio_to->geom->softc; 895 bp->bio_from->index--; 896 if (bp->bio_cmd == BIO_WRITE) 897 sc->sc_writes--; 898 disk = bp->bio_from->private; 899 if (disk == NULL) { 900 g_topology_lock(); 901 g_mirror_kill_consumer(sc, bp->bio_from); 902 g_topology_unlock(); 903 } else { 904 g_mirror_update_delay(disk, bp); 905 } 906 907 pbp->bio_inbed++; 908 KASSERT(pbp->bio_inbed <= pbp->bio_children, 909 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 910 pbp->bio_children)); 911 if (bp->bio_error == 0 && pbp->bio_error == 0) { 912 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 913 g_destroy_bio(bp); 914 if (pbp->bio_children == pbp->bio_inbed) { 915 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 916 pbp->bio_completed = pbp->bio_length; 917 if (pbp->bio_cmd == BIO_WRITE) { 918 bioq_remove(&sc->sc_inflight, pbp); 919 /* Release delayed sync requests if possible. */ 920 g_mirror_sync_release(sc); 921 } 922 g_io_deliver(pbp, pbp->bio_error); 923 } 924 return; 925 } else if (bp->bio_error != 0) { 926 if (pbp->bio_error == 0) 927 pbp->bio_error = bp->bio_error; 928 if (disk != NULL) { 929 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 930 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 931 G_MIRROR_LOGREQ(0, bp, 932 "Request failed (error=%d).", 933 bp->bio_error); 934 } else { 935 G_MIRROR_LOGREQ(1, bp, 936 "Request failed (error=%d).", 937 bp->bio_error); 938 } 939 if (g_mirror_disconnect_on_failure && 940 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) 941 { 942 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 943 g_mirror_event_send(disk, 944 G_MIRROR_DISK_STATE_DISCONNECTED, 945 G_MIRROR_EVENT_DONTWAIT); 946 } 947 } 948 switch (pbp->bio_cmd) { 949 case BIO_DELETE: 950 case BIO_WRITE: 951 pbp->bio_inbed--; 952 pbp->bio_children--; 953 break; 954 } 955 } 956 g_destroy_bio(bp); 957 958 switch (pbp->bio_cmd) { 959 case BIO_READ: 960 if (pbp->bio_inbed < pbp->bio_children) 961 break; 962 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 963 g_io_deliver(pbp, pbp->bio_error); 964 else { 965 pbp->bio_error = 0; 966 mtx_lock(&sc->sc_queue_mtx); 967 bioq_disksort(&sc->sc_queue, pbp); 968 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 969 wakeup(sc); 970 mtx_unlock(&sc->sc_queue_mtx); 971 } 972 break; 973 case BIO_DELETE: 974 case BIO_WRITE: 975 if (pbp->bio_children == 0) { 976 /* 977 * All requests failed. 978 */ 979 } else if (pbp->bio_inbed < pbp->bio_children) { 980 /* Do nothing. */ 981 break; 982 } else if (pbp->bio_children == pbp->bio_inbed) { 983 /* Some requests succeeded. */ 984 pbp->bio_error = 0; 985 pbp->bio_completed = pbp->bio_length; 986 } 987 bioq_remove(&sc->sc_inflight, pbp); 988 /* Release delayed sync requests if possible. */ 989 g_mirror_sync_release(sc); 990 g_io_deliver(pbp, pbp->bio_error); 991 break; 992 default: 993 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 994 break; 995 } 996} 997 998static void 999g_mirror_sync_done(struct bio *bp) 1000{ 1001 struct g_mirror_softc *sc; 1002 1003 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1004 sc = bp->bio_from->geom->softc; 1005 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; 1006 mtx_lock(&sc->sc_queue_mtx); 1007 bioq_disksort(&sc->sc_queue, bp); 1008 wakeup(sc); 1009 mtx_unlock(&sc->sc_queue_mtx); 1010} 1011 1012static void 1013g_mirror_kernel_dump(struct bio *bp) 1014{ 1015 struct g_mirror_softc *sc; 1016 struct g_mirror_disk *disk; 1017 struct bio *cbp; 1018 struct g_kerneldump *gkd; 1019 1020 /* 1021 * We configure dumping to the first component, because this component 1022 * will be used for reading with 'prefer' balance algorithm. 1023 * If the component with the higest priority is currently disconnected 1024 * we will not be able to read the dump after the reboot if it will be 1025 * connected and synchronized later. Can we do something better? 1026 */ 1027 sc = bp->bio_to->geom->softc; 1028 disk = LIST_FIRST(&sc->sc_disks); 1029 1030 gkd = (struct g_kerneldump *)bp->bio_data; 1031 if (gkd->length > bp->bio_to->mediasize) 1032 gkd->length = bp->bio_to->mediasize; 1033 cbp = g_clone_bio(bp); 1034 if (cbp == NULL) { 1035 g_io_deliver(bp, ENOMEM); 1036 return; 1037 } 1038 cbp->bio_done = g_std_done; 1039 g_io_request(cbp, disk->d_consumer); 1040 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1041 g_mirror_get_diskname(disk)); 1042} 1043 1044static void 1045g_mirror_start(struct bio *bp) 1046{ 1047 struct g_mirror_softc *sc; 1048 1049 sc = bp->bio_to->geom->softc; 1050 /* 1051 * If sc == NULL or there are no valid disks, provider's error 1052 * should be set and g_mirror_start() should not be called at all. 1053 */ 1054 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1055 ("Provider's error should be set (error=%d)(mirror=%s).", 1056 bp->bio_to->error, bp->bio_to->name)); 1057 G_MIRROR_LOGREQ(3, bp, "Request received."); 1058 1059 switch (bp->bio_cmd) { 1060 case BIO_READ: 1061 case BIO_WRITE: 1062 case BIO_DELETE: 1063 break; 1064 case BIO_GETATTR: 1065 if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1066 g_mirror_kernel_dump(bp); 1067 return; 1068 } 1069 /* FALLTHROUGH */ 1070 default: 1071 g_io_deliver(bp, EOPNOTSUPP); 1072 return; 1073 } 1074 mtx_lock(&sc->sc_queue_mtx); 1075 bioq_disksort(&sc->sc_queue, bp); 1076 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1077 wakeup(sc); 1078 mtx_unlock(&sc->sc_queue_mtx); 1079} 1080 1081/* 1082 * Return TRUE if the given request is colliding with a in-progress 1083 * synchronization request. 1084 */ 1085static int 1086g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1087{ 1088 struct g_mirror_disk *disk; 1089 struct bio *sbp; 1090 off_t rstart, rend, sstart, send; 1091 int i; 1092 1093 if (sc->sc_sync.ds_ndisks == 0) 1094 return (0); 1095 rstart = bp->bio_offset; 1096 rend = bp->bio_offset + bp->bio_length; 1097 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1098 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1099 continue; 1100 for (i = 0; i < g_mirror_syncreqs; i++) { 1101 sbp = disk->d_sync.ds_bios[i]; 1102 if (sbp == NULL) 1103 continue; 1104 sstart = sbp->bio_offset; 1105 send = sbp->bio_offset + sbp->bio_length; 1106 if (rend > sstart && rstart < send) 1107 return (1); 1108 } 1109 } 1110 return (0); 1111} 1112 1113/* 1114 * Return TRUE if the given sync request is colliding with a in-progress regular 1115 * request. 1116 */ 1117static int 1118g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1119{ 1120 off_t rstart, rend, sstart, send; 1121 struct bio *bp; 1122 1123 if (sc->sc_sync.ds_ndisks == 0) 1124 return (0); 1125 sstart = sbp->bio_offset; 1126 send = sbp->bio_offset + sbp->bio_length; 1127 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { 1128 rstart = bp->bio_offset; 1129 rend = bp->bio_offset + bp->bio_length; 1130 if (rend > sstart && rstart < send) 1131 return (1); 1132 } 1133 return (0); 1134} 1135 1136/* 1137 * Puts request onto delayed queue. 1138 */ 1139static void 1140g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1141{ 1142 1143 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1144 bioq_insert_head(&sc->sc_regular_delayed, bp); 1145} 1146 1147/* 1148 * Puts synchronization request onto delayed queue. 1149 */ 1150static void 1151g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1152{ 1153 1154 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1155 bioq_insert_tail(&sc->sc_sync_delayed, bp); 1156} 1157 1158/* 1159 * Releases delayed regular requests which don't collide anymore with sync 1160 * requests. 1161 */ 1162static void 1163g_mirror_regular_release(struct g_mirror_softc *sc) 1164{ 1165 struct bio *bp, *bp2; 1166 1167 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { 1168 if (g_mirror_sync_collision(sc, bp)) 1169 continue; 1170 bioq_remove(&sc->sc_regular_delayed, bp); 1171 G_MIRROR_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1172 mtx_lock(&sc->sc_queue_mtx); 1173 bioq_insert_head(&sc->sc_queue, bp); 1174#if 0 1175 /* 1176 * wakeup() is not needed, because this function is called from 1177 * the worker thread. 1178 */ 1179 wakeup(&sc->sc_queue); 1180#endif 1181 mtx_unlock(&sc->sc_queue_mtx); 1182 } 1183} 1184 1185/* 1186 * Releases delayed sync requests which don't collide anymore with regular 1187 * requests. 1188 */ 1189static void 1190g_mirror_sync_release(struct g_mirror_softc *sc) 1191{ 1192 struct bio *bp, *bp2; 1193 1194 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { 1195 if (g_mirror_regular_collision(sc, bp)) 1196 continue; 1197 bioq_remove(&sc->sc_sync_delayed, bp); 1198 G_MIRROR_LOGREQ(2, bp, 1199 "Releasing delayed synchronization request."); 1200 g_io_request(bp, bp->bio_from); 1201 } 1202} 1203 1204/* 1205 * Handle synchronization requests. 1206 * Every synchronization request is two-steps process: first, READ request is 1207 * send to active provider and then WRITE request (with read data) to the provider 1208 * beeing synchronized. When WRITE is finished, new synchronization request is 1209 * send. 1210 */ 1211static void 1212g_mirror_sync_request(struct bio *bp) 1213{ 1214 struct g_mirror_softc *sc; 1215 struct g_mirror_disk *disk; 1216 1217 bp->bio_from->index--; 1218 sc = bp->bio_from->geom->softc; 1219 disk = bp->bio_from->private; 1220 if (disk == NULL) { 1221 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1222 g_topology_lock(); 1223 g_mirror_kill_consumer(sc, bp->bio_from); 1224 g_topology_unlock(); 1225 free(bp->bio_data, M_MIRROR); 1226 g_destroy_bio(bp); 1227 sx_xlock(&sc->sc_lock); 1228 return; 1229 } 1230 1231 /* 1232 * Synchronization request. 1233 */ 1234 switch (bp->bio_cmd) { 1235 case BIO_READ: 1236 { 1237 struct g_consumer *cp; 1238 1239 if (bp->bio_error != 0) { 1240 G_MIRROR_LOGREQ(0, bp, 1241 "Synchronization request failed (error=%d).", 1242 bp->bio_error); 1243 g_destroy_bio(bp); 1244 return; 1245 } 1246 G_MIRROR_LOGREQ(3, bp, 1247 "Synchronization request half-finished."); 1248 bp->bio_cmd = BIO_WRITE; 1249 bp->bio_cflags = 0; 1250 cp = disk->d_consumer; 1251 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1252 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1253 cp->acr, cp->acw, cp->ace)); 1254 cp->index++; 1255 g_io_request(bp, cp); 1256 return; 1257 } 1258 case BIO_WRITE: 1259 { 1260 struct g_mirror_disk_sync *sync; 1261 off_t offset; 1262 void *data; 1263 int i; 1264 1265 if (bp->bio_error != 0) { 1266 G_MIRROR_LOGREQ(0, bp, 1267 "Synchronization request failed (error=%d).", 1268 bp->bio_error); 1269 g_destroy_bio(bp); 1270 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1271 g_mirror_event_send(disk, 1272 G_MIRROR_DISK_STATE_DISCONNECTED, 1273 G_MIRROR_EVENT_DONTWAIT); 1274 return; 1275 } 1276 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1277 sync = &disk->d_sync; 1278 if (sync->ds_offset == sc->sc_mediasize || 1279 sync->ds_consumer == NULL || 1280 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1281 /* Don't send more synchronization requests. */ 1282 sync->ds_inflight--; 1283 if (sync->ds_bios != NULL) { 1284 i = (int)(uintptr_t)bp->bio_caller1; 1285 sync->ds_bios[i] = NULL; 1286 } 1287 free(bp->bio_data, M_MIRROR); 1288 g_destroy_bio(bp); 1289 if (sync->ds_inflight > 0) 1290 return; 1291 if (sync->ds_consumer == NULL || 1292 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1293 return; 1294 } 1295 /* Disk up-to-date, activate it. */ 1296 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1297 G_MIRROR_EVENT_DONTWAIT); 1298 return; 1299 } 1300 1301 /* Send next synchronization request. */ 1302 data = bp->bio_data; 1303 bzero(bp, sizeof(*bp)); 1304 bp->bio_cmd = BIO_READ; 1305 bp->bio_offset = sync->ds_offset; 1306 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1307 sync->ds_offset += bp->bio_length; 1308 bp->bio_done = g_mirror_sync_done; 1309 bp->bio_data = data; 1310 bp->bio_from = sync->ds_consumer; 1311 bp->bio_to = sc->sc_provider; 1312 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1313 sync->ds_consumer->index++; 1314 /* 1315 * Delay the request if it is colliding with a regular request. 1316 */ 1317 if (g_mirror_regular_collision(sc, bp)) 1318 g_mirror_sync_delay(sc, bp); 1319 else 1320 g_io_request(bp, sync->ds_consumer); 1321 1322 /* Release delayed requests if possible. */ 1323 g_mirror_regular_release(sc); 1324 1325 /* Find the smallest offset */ 1326 offset = sc->sc_mediasize; 1327 for (i = 0; i < g_mirror_syncreqs; i++) { 1328 bp = sync->ds_bios[i]; 1329 if (bp->bio_offset < offset) 1330 offset = bp->bio_offset; 1331 } 1332 if (sync->ds_offset_done + (MAXPHYS * 100) < offset) { 1333 /* Update offset_done on every 100 blocks. */ 1334 sync->ds_offset_done = offset; 1335 g_mirror_update_metadata(disk); 1336 } 1337 return; 1338 } 1339 default: 1340 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1341 bp->bio_cmd, sc->sc_name)); 1342 break; 1343 } 1344} 1345 1346static void 1347g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1348{ 1349 struct g_mirror_disk *disk; 1350 struct g_consumer *cp; 1351 struct bio *cbp; 1352 1353 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1354 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1355 break; 1356 } 1357 if (disk == NULL) { 1358 if (bp->bio_error == 0) 1359 bp->bio_error = ENXIO; 1360 g_io_deliver(bp, bp->bio_error); 1361 return; 1362 } 1363 cbp = g_clone_bio(bp); 1364 if (cbp == NULL) { 1365 if (bp->bio_error == 0) 1366 bp->bio_error = ENOMEM; 1367 g_io_deliver(bp, bp->bio_error); 1368 return; 1369 } 1370 /* 1371 * Fill in the component buf structure. 1372 */ 1373 cp = disk->d_consumer; 1374 cbp->bio_done = g_mirror_done; 1375 cbp->bio_to = cp->provider; 1376 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1377 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1378 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1379 cp->acw, cp->ace)); 1380 cp->index++; 1381 g_io_request(cbp, cp); 1382} 1383 1384static void 1385g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1386{ 1387 struct g_mirror_disk *disk; 1388 struct g_consumer *cp; 1389 struct bio *cbp; 1390 1391 disk = g_mirror_get_disk(sc); 1392 if (disk == NULL) { 1393 if (bp->bio_error == 0) 1394 bp->bio_error = ENXIO; 1395 g_io_deliver(bp, bp->bio_error); 1396 return; 1397 } 1398 cbp = g_clone_bio(bp); 1399 if (cbp == NULL) { 1400 if (bp->bio_error == 0) 1401 bp->bio_error = ENOMEM; 1402 g_io_deliver(bp, bp->bio_error); 1403 return; 1404 } 1405 /* 1406 * Fill in the component buf structure. 1407 */ 1408 cp = disk->d_consumer; 1409 cbp->bio_done = g_mirror_done; 1410 cbp->bio_to = cp->provider; 1411 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1412 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1413 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1414 cp->acw, cp->ace)); 1415 cp->index++; 1416 g_io_request(cbp, cp); 1417} 1418 1419static void 1420g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1421{ 1422 struct g_mirror_disk *disk, *dp; 1423 struct g_consumer *cp; 1424 struct bio *cbp; 1425 struct bintime curtime; 1426 1427 binuptime(&curtime); 1428 /* 1429 * Find a disk which the smallest load. 1430 */ 1431 disk = NULL; 1432 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1433 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1434 continue; 1435 /* If disk wasn't used for more than 2 sec, use it. */ 1436 if (curtime.sec - dp->d_last_used.sec >= 2) { 1437 disk = dp; 1438 break; 1439 } 1440 if (disk == NULL || 1441 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1442 disk = dp; 1443 } 1444 } 1445 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1446 cbp = g_clone_bio(bp); 1447 if (cbp == NULL) { 1448 if (bp->bio_error == 0) 1449 bp->bio_error = ENOMEM; 1450 g_io_deliver(bp, bp->bio_error); 1451 return; 1452 } 1453 /* 1454 * Fill in the component buf structure. 1455 */ 1456 cp = disk->d_consumer; 1457 cbp->bio_done = g_mirror_done; 1458 cbp->bio_to = cp->provider; 1459 binuptime(&disk->d_last_used); 1460 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1461 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1462 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1463 cp->acw, cp->ace)); 1464 cp->index++; 1465 g_io_request(cbp, cp); 1466} 1467 1468static void 1469g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1470{ 1471 struct bio_queue_head queue; 1472 struct g_mirror_disk *disk; 1473 struct g_consumer *cp; 1474 struct bio *cbp; 1475 off_t left, mod, offset, slice; 1476 u_char *data; 1477 u_int ndisks; 1478 1479 if (bp->bio_length <= sc->sc_slice) { 1480 g_mirror_request_round_robin(sc, bp); 1481 return; 1482 } 1483 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1484 slice = bp->bio_length / ndisks; 1485 mod = slice % sc->sc_provider->sectorsize; 1486 if (mod != 0) 1487 slice += sc->sc_provider->sectorsize - mod; 1488 /* 1489 * Allocate all bios before sending any request, so we can 1490 * return ENOMEM in nice and clean way. 1491 */ 1492 left = bp->bio_length; 1493 offset = bp->bio_offset; 1494 data = bp->bio_data; 1495 bioq_init(&queue); 1496 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1497 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1498 continue; 1499 cbp = g_clone_bio(bp); 1500 if (cbp == NULL) { 1501 for (cbp = bioq_first(&queue); cbp != NULL; 1502 cbp = bioq_first(&queue)) { 1503 bioq_remove(&queue, cbp); 1504 g_destroy_bio(cbp); 1505 } 1506 if (bp->bio_error == 0) 1507 bp->bio_error = ENOMEM; 1508 g_io_deliver(bp, bp->bio_error); 1509 return; 1510 } 1511 bioq_insert_tail(&queue, cbp); 1512 cbp->bio_done = g_mirror_done; 1513 cbp->bio_caller1 = disk; 1514 cbp->bio_to = disk->d_consumer->provider; 1515 cbp->bio_offset = offset; 1516 cbp->bio_data = data; 1517 cbp->bio_length = MIN(left, slice); 1518 left -= cbp->bio_length; 1519 if (left == 0) 1520 break; 1521 offset += cbp->bio_length; 1522 data += cbp->bio_length; 1523 } 1524 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1525 bioq_remove(&queue, cbp); 1526 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1527 disk = cbp->bio_caller1; 1528 cbp->bio_caller1 = NULL; 1529 cp = disk->d_consumer; 1530 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1531 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1532 cp->acr, cp->acw, cp->ace)); 1533 disk->d_consumer->index++; 1534 g_io_request(cbp, disk->d_consumer); 1535 } 1536} 1537 1538static void 1539g_mirror_register_request(struct bio *bp) 1540{ 1541 struct g_mirror_softc *sc; 1542 1543 sc = bp->bio_to->geom->softc; 1544 switch (bp->bio_cmd) { 1545 case BIO_READ: 1546 switch (sc->sc_balance) { 1547 case G_MIRROR_BALANCE_LOAD: 1548 g_mirror_request_load(sc, bp); 1549 break; 1550 case G_MIRROR_BALANCE_PREFER: 1551 g_mirror_request_prefer(sc, bp); 1552 break; 1553 case G_MIRROR_BALANCE_ROUND_ROBIN: 1554 g_mirror_request_round_robin(sc, bp); 1555 break; 1556 case G_MIRROR_BALANCE_SPLIT: 1557 g_mirror_request_split(sc, bp); 1558 break; 1559 } 1560 return; 1561 case BIO_WRITE: 1562 case BIO_DELETE: 1563 { 1564 struct g_mirror_disk *disk; 1565 struct g_mirror_disk_sync *sync; 1566 struct bio_queue_head queue; 1567 struct g_consumer *cp; 1568 struct bio *cbp; 1569 1570 /* 1571 * Delay the request if it is colliding with a synchronization 1572 * request. 1573 */ 1574 if (g_mirror_sync_collision(sc, bp)) { 1575 g_mirror_regular_delay(sc, bp); 1576 return; 1577 } 1578 1579 if (sc->sc_idle) 1580 g_mirror_unidle(sc); 1581 else 1582 sc->sc_last_write = time_uptime; 1583 1584 /* 1585 * Allocate all bios before sending any request, so we can 1586 * return ENOMEM in nice and clean way. 1587 */ 1588 bioq_init(&queue); 1589 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1590 sync = &disk->d_sync; 1591 switch (disk->d_state) { 1592 case G_MIRROR_DISK_STATE_ACTIVE: 1593 break; 1594 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1595 if (bp->bio_offset >= sync->ds_offset) 1596 continue; 1597 break; 1598 default: 1599 continue; 1600 } 1601 cbp = g_clone_bio(bp); 1602 if (cbp == NULL) { 1603 for (cbp = bioq_first(&queue); cbp != NULL; 1604 cbp = bioq_first(&queue)) { 1605 bioq_remove(&queue, cbp); 1606 g_destroy_bio(cbp); 1607 } 1608 if (bp->bio_error == 0) 1609 bp->bio_error = ENOMEM; 1610 g_io_deliver(bp, bp->bio_error); 1611 return; 1612 } 1613 bioq_insert_tail(&queue, cbp); 1614 cbp->bio_done = g_mirror_done; 1615 cp = disk->d_consumer; 1616 cbp->bio_caller1 = cp; 1617 cbp->bio_to = cp->provider; 1618 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1619 ("Consumer %s not opened (r%dw%de%d).", 1620 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1621 } 1622 for (cbp = bioq_first(&queue); cbp != NULL; 1623 cbp = bioq_first(&queue)) { 1624 bioq_remove(&queue, cbp); 1625 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1626 cp = cbp->bio_caller1; 1627 cbp->bio_caller1 = NULL; 1628 cp->index++; 1629 sc->sc_writes++; 1630 g_io_request(cbp, cp); 1631 } 1632 /* 1633 * Put request onto inflight queue, so we can check if new 1634 * synchronization requests don't collide with it. 1635 */ 1636 bioq_insert_tail(&sc->sc_inflight, bp); 1637 /* 1638 * Bump syncid on first write. 1639 */ 1640 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1641 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1642 g_mirror_bump_syncid(sc); 1643 } 1644 return; 1645 } 1646 default: 1647 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1648 bp->bio_cmd, sc->sc_name)); 1649 break; 1650 } 1651} 1652 1653static int 1654g_mirror_can_destroy(struct g_mirror_softc *sc) 1655{ 1656 struct g_geom *gp; 1657 struct g_consumer *cp; 1658 1659 g_topology_assert(); 1660 gp = sc->sc_geom; 1661 if (gp->softc == NULL) 1662 return (1); 1663 LIST_FOREACH(cp, &gp->consumer, consumer) { 1664 if (g_mirror_is_busy(sc, cp)) 1665 return (0); 1666 } 1667 gp = sc->sc_sync.ds_geom; 1668 LIST_FOREACH(cp, &gp->consumer, consumer) { 1669 if (g_mirror_is_busy(sc, cp)) 1670 return (0); 1671 } 1672 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1673 sc->sc_name); 1674 return (1); 1675} 1676 1677static int 1678g_mirror_try_destroy(struct g_mirror_softc *sc) 1679{ 1680 1681 if (sc->sc_rootmount != NULL) { 1682 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1683 sc->sc_rootmount); 1684 root_mount_rel(sc->sc_rootmount); 1685 sc->sc_rootmount = NULL; 1686 } 1687 g_topology_lock(); 1688 if (!g_mirror_can_destroy(sc)) { 1689 g_topology_unlock(); 1690 return (0); 1691 } 1692 sc->sc_geom->softc = NULL; 1693 sc->sc_sync.ds_geom->softc = NULL; 1694 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1695 g_topology_unlock(); 1696 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1697 &sc->sc_worker); 1698 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1699 sx_xunlock(&sc->sc_lock); 1700 wakeup(&sc->sc_worker); 1701 sc->sc_worker = NULL; 1702 } else { 1703 g_topology_unlock(); 1704 g_mirror_destroy_device(sc); 1705 free(sc, M_MIRROR); 1706 } 1707 return (1); 1708} 1709 1710/* 1711 * Worker thread. 1712 */ 1713static void 1714g_mirror_worker(void *arg) 1715{ 1716 struct g_mirror_softc *sc; 1717 struct g_mirror_event *ep; 1718 struct bio *bp; 1719 int timeout; 1720 1721 sc = arg; 1722 mtx_lock_spin(&sched_lock); 1723 sched_prio(curthread, PRIBIO); 1724 mtx_unlock_spin(&sched_lock); 1725 1726 sx_xlock(&sc->sc_lock); 1727 for (;;) { 1728 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1729 /* 1730 * First take a look at events. 1731 * This is important to handle events before any I/O requests. 1732 */ 1733 ep = g_mirror_event_get(sc); 1734 if (ep != NULL) { 1735 g_mirror_event_remove(sc, ep); 1736 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1737 /* Update only device status. */ 1738 G_MIRROR_DEBUG(3, 1739 "Running event for device %s.", 1740 sc->sc_name); 1741 ep->e_error = 0; 1742 g_mirror_update_device(sc, 1); 1743 } else { 1744 /* Update disk status. */ 1745 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1746 g_mirror_get_diskname(ep->e_disk)); 1747 ep->e_error = g_mirror_update_disk(ep->e_disk, 1748 ep->e_state); 1749 if (ep->e_error == 0) 1750 g_mirror_update_device(sc, 0); 1751 } 1752 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1753 KASSERT(ep->e_error == 0, 1754 ("Error cannot be handled.")); 1755 g_mirror_event_free(ep); 1756 } else { 1757 ep->e_flags |= G_MIRROR_EVENT_DONE; 1758 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1759 ep); 1760 mtx_lock(&sc->sc_events_mtx); 1761 wakeup(ep); 1762 mtx_unlock(&sc->sc_events_mtx); 1763 } 1764 if ((sc->sc_flags & 1765 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1766 if (g_mirror_try_destroy(sc)) { 1767 curthread->td_pflags &= ~TDP_GEOM; 1768 G_MIRROR_DEBUG(1, "Thread exiting."); 1769 kthread_exit(0); 1770 } 1771 } 1772 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1773 continue; 1774 } 1775 /* 1776 * Check if we can mark array as CLEAN and if we can't take 1777 * how much seconds should we wait. 1778 */ 1779 timeout = g_mirror_idle(sc, -1); 1780 /* 1781 * Now I/O requests. 1782 */ 1783 /* Get first request from the queue. */ 1784 mtx_lock(&sc->sc_queue_mtx); 1785 bp = bioq_first(&sc->sc_queue); 1786 if (bp == NULL) { 1787 if ((sc->sc_flags & 1788 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1789 mtx_unlock(&sc->sc_queue_mtx); 1790 if (g_mirror_try_destroy(sc)) { 1791 curthread->td_pflags &= ~TDP_GEOM; 1792 G_MIRROR_DEBUG(1, "Thread exiting."); 1793 kthread_exit(0); 1794 } 1795 mtx_lock(&sc->sc_queue_mtx); 1796 } 1797 sx_xunlock(&sc->sc_lock); 1798 /* 1799 * XXX: We can miss an event here, because an event 1800 * can be added without sx-device-lock and without 1801 * mtx-queue-lock. Maybe I should just stop using 1802 * dedicated mutex for events synchronization and 1803 * stick with the queue lock? 1804 * The event will hang here until next I/O request 1805 * or next event is received. 1806 */ 1807 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1808 timeout * hz); 1809 sx_xlock(&sc->sc_lock); 1810 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1811 continue; 1812 } 1813 bioq_remove(&sc->sc_queue, bp); 1814 mtx_unlock(&sc->sc_queue_mtx); 1815 1816 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 1817 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1818 g_mirror_sync_request(bp); /* READ */ 1819 } else if (bp->bio_to != sc->sc_provider) { 1820 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 1821 g_mirror_regular_request(bp); 1822 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 1823 g_mirror_sync_request(bp); /* WRITE */ 1824 else { 1825 KASSERT(0, 1826 ("Invalid request cflags=0x%hhx to=%s.", 1827 bp->bio_cflags, bp->bio_to->name)); 1828 } 1829 } else { 1830 g_mirror_register_request(bp); 1831 } 1832 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1833 } 1834} 1835 1836static void 1837g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 1838{ 1839 1840 sx_assert(&sc->sc_lock, SX_LOCKED); 1841 1842 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1843 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1844 g_mirror_get_diskname(disk), sc->sc_name); 1845 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1846 } else if (sc->sc_idle && 1847 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1848 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1849 g_mirror_get_diskname(disk), sc->sc_name); 1850 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1851 } 1852} 1853 1854static void 1855g_mirror_sync_start(struct g_mirror_disk *disk) 1856{ 1857 struct g_mirror_softc *sc; 1858 struct g_consumer *cp; 1859 struct bio *bp; 1860 int error, i; 1861 1862 g_topology_assert_not(); 1863 sc = disk->d_softc; 1864 sx_assert(&sc->sc_lock, SX_LOCKED); 1865 1866 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1867 ("Disk %s is not marked for synchronization.", 1868 g_mirror_get_diskname(disk))); 1869 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1870 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1871 sc->sc_state)); 1872 1873 sx_xunlock(&sc->sc_lock); 1874 g_topology_lock(); 1875 cp = g_new_consumer(sc->sc_sync.ds_geom); 1876 error = g_attach(cp, sc->sc_provider); 1877 KASSERT(error == 0, 1878 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 1879 error = g_access(cp, 1, 0, 0); 1880 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 1881 g_topology_unlock(); 1882 sx_xlock(&sc->sc_lock); 1883 1884 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1885 g_mirror_get_diskname(disk)); 1886 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1887 KASSERT(disk->d_sync.ds_consumer == NULL, 1888 ("Sync consumer already exists (device=%s, disk=%s).", 1889 sc->sc_name, g_mirror_get_diskname(disk))); 1890 1891 disk->d_sync.ds_consumer = cp; 1892 disk->d_sync.ds_consumer->private = disk; 1893 disk->d_sync.ds_consumer->index = 0; 1894 1895 /* 1896 * Allocate memory for synchronization bios and initialize them. 1897 */ 1898 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 1899 M_MIRROR, M_WAITOK); 1900 for (i = 0; i < g_mirror_syncreqs; i++) { 1901 bp = g_alloc_bio(); 1902 disk->d_sync.ds_bios[i] = bp; 1903 bp->bio_parent = NULL; 1904 bp->bio_cmd = BIO_READ; 1905 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1906 bp->bio_cflags = 0; 1907 bp->bio_offset = disk->d_sync.ds_offset; 1908 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1909 disk->d_sync.ds_offset += bp->bio_length; 1910 bp->bio_done = g_mirror_sync_done; 1911 bp->bio_from = disk->d_sync.ds_consumer; 1912 bp->bio_to = sc->sc_provider; 1913 bp->bio_caller1 = (void *)(uintptr_t)i; 1914 } 1915 1916 /* Increase the number of disks in SYNCHRONIZING state. */ 1917 sc->sc_sync.ds_ndisks++; 1918 /* Set the number of in-flight synchronization requests. */ 1919 disk->d_sync.ds_inflight = g_mirror_syncreqs; 1920 1921 /* 1922 * Fire off first synchronization requests. 1923 */ 1924 for (i = 0; i < g_mirror_syncreqs; i++) { 1925 bp = disk->d_sync.ds_bios[i]; 1926 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1927 disk->d_sync.ds_consumer->index++; 1928 /* 1929 * Delay the request if it is colliding with a regular request. 1930 */ 1931 if (g_mirror_regular_collision(sc, bp)) 1932 g_mirror_sync_delay(sc, bp); 1933 else 1934 g_io_request(bp, disk->d_sync.ds_consumer); 1935 } 1936} 1937 1938/* 1939 * Stop synchronization process. 1940 * type: 0 - synchronization finished 1941 * 1 - synchronization stopped 1942 */ 1943static void 1944g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1945{ 1946 struct g_mirror_softc *sc; 1947 struct g_consumer *cp; 1948 1949 g_topology_assert_not(); 1950 sc = disk->d_softc; 1951 sx_assert(&sc->sc_lock, SX_LOCKED); 1952 1953 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1954 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1955 g_mirror_disk_state2str(disk->d_state))); 1956 if (disk->d_sync.ds_consumer == NULL) 1957 return; 1958 1959 if (type == 0) { 1960 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1961 sc->sc_name, g_mirror_get_diskname(disk)); 1962 } else /* if (type == 1) */ { 1963 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1964 sc->sc_name, g_mirror_get_diskname(disk)); 1965 } 1966 free(disk->d_sync.ds_bios, M_MIRROR); 1967 disk->d_sync.ds_bios = NULL; 1968 cp = disk->d_sync.ds_consumer; 1969 disk->d_sync.ds_consumer = NULL; 1970 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1971 sc->sc_sync.ds_ndisks--; 1972 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1973 g_topology_lock(); 1974 g_mirror_kill_consumer(sc, cp); 1975 g_topology_unlock(); 1976 sx_xlock(&sc->sc_lock); 1977} 1978 1979static void 1980g_mirror_launch_provider(struct g_mirror_softc *sc) 1981{ 1982 struct g_mirror_disk *disk; 1983 struct g_provider *pp; 1984 1985 sx_assert(&sc->sc_lock, SX_LOCKED); 1986 1987 g_topology_lock(); 1988 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1989 pp->mediasize = sc->sc_mediasize; 1990 pp->sectorsize = sc->sc_sectorsize; 1991 sc->sc_provider = pp; 1992 g_error_provider(pp, 0); 1993 g_topology_unlock(); 1994 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 1995 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 1996 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1997 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1998 g_mirror_sync_start(disk); 1999 } 2000} 2001 2002static void 2003g_mirror_destroy_provider(struct g_mirror_softc *sc) 2004{ 2005 struct g_mirror_disk *disk; 2006 struct bio *bp; 2007 2008 g_topology_assert_not(); 2009 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2010 sc->sc_name)); 2011 2012 g_topology_lock(); 2013 g_error_provider(sc->sc_provider, ENXIO); 2014 mtx_lock(&sc->sc_queue_mtx); 2015 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 2016 bioq_remove(&sc->sc_queue, bp); 2017 g_io_deliver(bp, ENXIO); 2018 } 2019 mtx_unlock(&sc->sc_queue_mtx); 2020 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 2021 sc->sc_provider->name); 2022 sc->sc_provider->flags |= G_PF_WITHER; 2023 g_orphan_provider(sc->sc_provider, ENXIO); 2024 g_topology_unlock(); 2025 sc->sc_provider = NULL; 2026 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2027 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2028 g_mirror_sync_stop(disk, 1); 2029 } 2030} 2031 2032static void 2033g_mirror_go(void *arg) 2034{ 2035 struct g_mirror_softc *sc; 2036 2037 sc = arg; 2038 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2039 g_mirror_event_send(sc, 0, 2040 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2041} 2042 2043static u_int 2044g_mirror_determine_state(struct g_mirror_disk *disk) 2045{ 2046 struct g_mirror_softc *sc; 2047 u_int state; 2048 2049 sc = disk->d_softc; 2050 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2051 if ((disk->d_flags & 2052 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2053 /* Disk does not need synchronization. */ 2054 state = G_MIRROR_DISK_STATE_ACTIVE; 2055 } else { 2056 if ((sc->sc_flags & 2057 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2058 (disk->d_flags & 2059 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2060 /* 2061 * We can start synchronization from 2062 * the stored offset. 2063 */ 2064 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2065 } else { 2066 state = G_MIRROR_DISK_STATE_STALE; 2067 } 2068 } 2069 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2070 /* 2071 * Reset all synchronization data for this disk, 2072 * because if it even was synchronized, it was 2073 * synchronized to disks with different syncid. 2074 */ 2075 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2076 disk->d_sync.ds_offset = 0; 2077 disk->d_sync.ds_offset_done = 0; 2078 disk->d_sync.ds_syncid = sc->sc_syncid; 2079 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2080 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2081 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2082 } else { 2083 state = G_MIRROR_DISK_STATE_STALE; 2084 } 2085 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2086 /* 2087 * Not good, NOT GOOD! 2088 * It means that mirror was started on stale disks 2089 * and more fresh disk just arrive. 2090 * If there were writes, mirror is broken, sorry. 2091 * I think the best choice here is don't touch 2092 * this disk and inform the user loudly. 2093 */ 2094 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2095 "disk (%s) arrives!! It will not be connected to the " 2096 "running device.", sc->sc_name, 2097 g_mirror_get_diskname(disk)); 2098 g_mirror_destroy_disk(disk); 2099 state = G_MIRROR_DISK_STATE_NONE; 2100 /* Return immediately, because disk was destroyed. */ 2101 return (state); 2102 } 2103 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2104 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2105 return (state); 2106} 2107 2108/* 2109 * Update device state. 2110 */ 2111static void 2112g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 2113{ 2114 struct g_mirror_disk *disk; 2115 u_int state; 2116 2117 sx_assert(&sc->sc_lock, SX_XLOCKED); 2118 2119 switch (sc->sc_state) { 2120 case G_MIRROR_DEVICE_STATE_STARTING: 2121 { 2122 struct g_mirror_disk *pdisk, *tdisk; 2123 u_int dirty, ndisks, genid, syncid; 2124 2125 KASSERT(sc->sc_provider == NULL, 2126 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2127 /* 2128 * Are we ready? We are, if all disks are connected or 2129 * if we have any disks and 'force' is true. 2130 */ 2131 ndisks = g_mirror_ndisks(sc, -1); 2132 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2133 ; 2134 } else if (ndisks == 0) { 2135 /* 2136 * Disks went down in starting phase, so destroy 2137 * device. 2138 */ 2139 callout_drain(&sc->sc_callout); 2140 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2141 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2142 sc->sc_rootmount); 2143 root_mount_rel(sc->sc_rootmount); 2144 sc->sc_rootmount = NULL; 2145 return; 2146 } else { 2147 return; 2148 } 2149 2150 /* 2151 * Activate all disks with the biggest syncid. 2152 */ 2153 if (force) { 2154 /* 2155 * If 'force' is true, we have been called due to 2156 * timeout, so don't bother canceling timeout. 2157 */ 2158 ndisks = 0; 2159 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2160 if ((disk->d_flags & 2161 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2162 ndisks++; 2163 } 2164 } 2165 if (ndisks == 0) { 2166 /* No valid disks found, destroy device. */ 2167 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2168 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2169 __LINE__, sc->sc_rootmount); 2170 root_mount_rel(sc->sc_rootmount); 2171 sc->sc_rootmount = NULL; 2172 return; 2173 } 2174 } else { 2175 /* Cancel timeout. */ 2176 callout_drain(&sc->sc_callout); 2177 } 2178 2179 /* 2180 * Find the biggest genid. 2181 */ 2182 genid = 0; 2183 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2184 if (disk->d_genid > genid) 2185 genid = disk->d_genid; 2186 } 2187 sc->sc_genid = genid; 2188 /* 2189 * Remove all disks without the biggest genid. 2190 */ 2191 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2192 if (disk->d_genid < genid) { 2193 G_MIRROR_DEBUG(0, 2194 "Component %s (device %s) broken, skipping.", 2195 g_mirror_get_diskname(disk), sc->sc_name); 2196 g_mirror_destroy_disk(disk); 2197 } 2198 } 2199 2200 /* 2201 * Find the biggest syncid. 2202 */ 2203 syncid = 0; 2204 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2205 if (disk->d_sync.ds_syncid > syncid) 2206 syncid = disk->d_sync.ds_syncid; 2207 } 2208 2209 /* 2210 * Here we need to look for dirty disks and if all disks 2211 * with the biggest syncid are dirty, we have to choose 2212 * one with the biggest priority and rebuild the rest. 2213 */ 2214 /* 2215 * Find the number of dirty disks with the biggest syncid. 2216 * Find the number of disks with the biggest syncid. 2217 * While here, find a disk with the biggest priority. 2218 */ 2219 dirty = ndisks = 0; 2220 pdisk = NULL; 2221 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2222 if (disk->d_sync.ds_syncid != syncid) 2223 continue; 2224 if ((disk->d_flags & 2225 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2226 continue; 2227 } 2228 ndisks++; 2229 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2230 dirty++; 2231 if (pdisk == NULL || 2232 pdisk->d_priority < disk->d_priority) { 2233 pdisk = disk; 2234 } 2235 } 2236 } 2237 if (dirty == 0) { 2238 /* No dirty disks at all, great. */ 2239 } else if (dirty == ndisks) { 2240 /* 2241 * Force synchronization for all dirty disks except one 2242 * with the biggest priority. 2243 */ 2244 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2245 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2246 "master disk for synchronization.", 2247 g_mirror_get_diskname(pdisk), sc->sc_name); 2248 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2249 if (disk->d_sync.ds_syncid != syncid) 2250 continue; 2251 if ((disk->d_flags & 2252 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2253 continue; 2254 } 2255 KASSERT((disk->d_flags & 2256 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2257 ("Disk %s isn't marked as dirty.", 2258 g_mirror_get_diskname(disk))); 2259 /* Skip the disk with the biggest priority. */ 2260 if (disk == pdisk) 2261 continue; 2262 disk->d_sync.ds_syncid = 0; 2263 } 2264 } else if (dirty < ndisks) { 2265 /* 2266 * Force synchronization for all dirty disks. 2267 * We have some non-dirty disks. 2268 */ 2269 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2270 if (disk->d_sync.ds_syncid != syncid) 2271 continue; 2272 if ((disk->d_flags & 2273 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2274 continue; 2275 } 2276 if ((disk->d_flags & 2277 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2278 continue; 2279 } 2280 disk->d_sync.ds_syncid = 0; 2281 } 2282 } 2283 2284 /* Reset hint. */ 2285 sc->sc_hint = NULL; 2286 sc->sc_syncid = syncid; 2287 if (force) { 2288 /* Remember to bump syncid on first write. */ 2289 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2290 } 2291 state = G_MIRROR_DEVICE_STATE_RUNNING; 2292 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2293 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2294 g_mirror_device_state2str(state)); 2295 sc->sc_state = state; 2296 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2297 state = g_mirror_determine_state(disk); 2298 g_mirror_event_send(disk, state, 2299 G_MIRROR_EVENT_DONTWAIT); 2300 if (state == G_MIRROR_DISK_STATE_STALE) 2301 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2302 } 2303 break; 2304 } 2305 case G_MIRROR_DEVICE_STATE_RUNNING: 2306 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2307 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2308 /* 2309 * No active disks or no disks at all, 2310 * so destroy device. 2311 */ 2312 if (sc->sc_provider != NULL) 2313 g_mirror_destroy_provider(sc); 2314 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2315 break; 2316 } else if (g_mirror_ndisks(sc, 2317 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2318 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2319 /* 2320 * We have active disks, launch provider if it doesn't 2321 * exist. 2322 */ 2323 if (sc->sc_provider == NULL) 2324 g_mirror_launch_provider(sc); 2325 if (sc->sc_rootmount != NULL) { 2326 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2327 __LINE__, sc->sc_rootmount); 2328 root_mount_rel(sc->sc_rootmount); 2329 sc->sc_rootmount = NULL; 2330 } 2331 } 2332 /* 2333 * Genid should be bumped immediately, so do it here. 2334 */ 2335 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2336 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2337 g_mirror_bump_genid(sc); 2338 } 2339 break; 2340 default: 2341 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2342 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2343 break; 2344 } 2345} 2346 2347/* 2348 * Update disk state and device state if needed. 2349 */ 2350#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2351 "Disk %s state changed from %s to %s (device %s).", \ 2352 g_mirror_get_diskname(disk), \ 2353 g_mirror_disk_state2str(disk->d_state), \ 2354 g_mirror_disk_state2str(state), sc->sc_name) 2355static int 2356g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2357{ 2358 struct g_mirror_softc *sc; 2359 2360 sc = disk->d_softc; 2361 sx_assert(&sc->sc_lock, SX_XLOCKED); 2362 2363again: 2364 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2365 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2366 g_mirror_disk_state2str(state)); 2367 switch (state) { 2368 case G_MIRROR_DISK_STATE_NEW: 2369 /* 2370 * Possible scenarios: 2371 * 1. New disk arrive. 2372 */ 2373 /* Previous state should be NONE. */ 2374 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2375 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2376 g_mirror_disk_state2str(disk->d_state))); 2377 DISK_STATE_CHANGED(); 2378 2379 disk->d_state = state; 2380 if (LIST_EMPTY(&sc->sc_disks)) 2381 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2382 else { 2383 struct g_mirror_disk *dp; 2384 2385 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2386 if (disk->d_priority >= dp->d_priority) { 2387 LIST_INSERT_BEFORE(dp, disk, d_next); 2388 dp = NULL; 2389 break; 2390 } 2391 if (LIST_NEXT(dp, d_next) == NULL) 2392 break; 2393 } 2394 if (dp != NULL) 2395 LIST_INSERT_AFTER(dp, disk, d_next); 2396 } 2397 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2398 sc->sc_name, g_mirror_get_diskname(disk)); 2399 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2400 break; 2401 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2402 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2403 g_mirror_device_state2str(sc->sc_state), 2404 g_mirror_get_diskname(disk), 2405 g_mirror_disk_state2str(disk->d_state))); 2406 state = g_mirror_determine_state(disk); 2407 if (state != G_MIRROR_DISK_STATE_NONE) 2408 goto again; 2409 break; 2410 case G_MIRROR_DISK_STATE_ACTIVE: 2411 /* 2412 * Possible scenarios: 2413 * 1. New disk does not need synchronization. 2414 * 2. Synchronization process finished successfully. 2415 */ 2416 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2417 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2418 g_mirror_device_state2str(sc->sc_state), 2419 g_mirror_get_diskname(disk), 2420 g_mirror_disk_state2str(disk->d_state))); 2421 /* Previous state should be NEW or SYNCHRONIZING. */ 2422 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2423 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2424 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2425 g_mirror_disk_state2str(disk->d_state))); 2426 DISK_STATE_CHANGED(); 2427 2428 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2429 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2430 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2431 g_mirror_sync_stop(disk, 0); 2432 } 2433 disk->d_state = state; 2434 disk->d_sync.ds_offset = 0; 2435 disk->d_sync.ds_offset_done = 0; 2436 g_mirror_update_idle(sc, disk); 2437 g_mirror_update_metadata(disk); 2438 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2439 sc->sc_name, g_mirror_get_diskname(disk)); 2440 break; 2441 case G_MIRROR_DISK_STATE_STALE: 2442 /* 2443 * Possible scenarios: 2444 * 1. Stale disk was connected. 2445 */ 2446 /* Previous state should be NEW. */ 2447 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2448 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2449 g_mirror_disk_state2str(disk->d_state))); 2450 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2451 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2452 g_mirror_device_state2str(sc->sc_state), 2453 g_mirror_get_diskname(disk), 2454 g_mirror_disk_state2str(disk->d_state))); 2455 /* 2456 * STALE state is only possible if device is marked 2457 * NOAUTOSYNC. 2458 */ 2459 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2460 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2461 g_mirror_device_state2str(sc->sc_state), 2462 g_mirror_get_diskname(disk), 2463 g_mirror_disk_state2str(disk->d_state))); 2464 DISK_STATE_CHANGED(); 2465 2466 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2467 disk->d_state = state; 2468 g_mirror_update_metadata(disk); 2469 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2470 sc->sc_name, g_mirror_get_diskname(disk)); 2471 break; 2472 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2473 /* 2474 * Possible scenarios: 2475 * 1. Disk which needs synchronization was connected. 2476 */ 2477 /* Previous state should be NEW. */ 2478 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2479 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2480 g_mirror_disk_state2str(disk->d_state))); 2481 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2482 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2483 g_mirror_device_state2str(sc->sc_state), 2484 g_mirror_get_diskname(disk), 2485 g_mirror_disk_state2str(disk->d_state))); 2486 DISK_STATE_CHANGED(); 2487 2488 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2489 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2490 disk->d_state = state; 2491 if (sc->sc_provider != NULL) { 2492 g_mirror_sync_start(disk); 2493 g_mirror_update_metadata(disk); 2494 } 2495 break; 2496 case G_MIRROR_DISK_STATE_DISCONNECTED: 2497 /* 2498 * Possible scenarios: 2499 * 1. Device wasn't running yet, but disk disappear. 2500 * 2. Disk was active and disapppear. 2501 * 3. Disk disappear during synchronization process. 2502 */ 2503 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2504 /* 2505 * Previous state should be ACTIVE, STALE or 2506 * SYNCHRONIZING. 2507 */ 2508 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2509 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2510 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2511 ("Wrong disk state (%s, %s).", 2512 g_mirror_get_diskname(disk), 2513 g_mirror_disk_state2str(disk->d_state))); 2514 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2515 /* Previous state should be NEW. */ 2516 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2517 ("Wrong disk state (%s, %s).", 2518 g_mirror_get_diskname(disk), 2519 g_mirror_disk_state2str(disk->d_state))); 2520 /* 2521 * Reset bumping syncid if disk disappeared in STARTING 2522 * state. 2523 */ 2524 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2525 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2526#ifdef INVARIANTS 2527 } else { 2528 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2529 sc->sc_name, 2530 g_mirror_device_state2str(sc->sc_state), 2531 g_mirror_get_diskname(disk), 2532 g_mirror_disk_state2str(disk->d_state))); 2533#endif 2534 } 2535 DISK_STATE_CHANGED(); 2536 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2537 sc->sc_name, g_mirror_get_diskname(disk)); 2538 2539 g_mirror_destroy_disk(disk); 2540 break; 2541 case G_MIRROR_DISK_STATE_DESTROY: 2542 { 2543 int error; 2544 2545 error = g_mirror_clear_metadata(disk); 2546 if (error != 0) 2547 return (error); 2548 DISK_STATE_CHANGED(); 2549 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2550 sc->sc_name, g_mirror_get_diskname(disk)); 2551 2552 g_mirror_destroy_disk(disk); 2553 sc->sc_ndisks--; 2554 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2555 g_mirror_update_metadata(disk); 2556 } 2557 break; 2558 } 2559 default: 2560 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2561 break; 2562 } 2563 return (0); 2564} 2565#undef DISK_STATE_CHANGED 2566 2567int 2568g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2569{ 2570 struct g_provider *pp; 2571 u_char *buf; 2572 int error; 2573 2574 g_topology_assert(); 2575 2576 error = g_access(cp, 1, 0, 0); 2577 if (error != 0) 2578 return (error); 2579 pp = cp->provider; 2580 g_topology_unlock(); 2581 /* Metadata are stored on last sector. */ 2582 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2583 &error); 2584 g_topology_lock(); 2585 g_access(cp, -1, 0, 0); 2586 if (buf == NULL) { 2587 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2588 cp->provider->name, error); 2589 return (error); 2590 } 2591 2592 /* Decode metadata. */ 2593 error = mirror_metadata_decode(buf, md); 2594 g_free(buf); 2595 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2596 return (EINVAL); 2597 if (md->md_version > G_MIRROR_VERSION) { 2598 G_MIRROR_DEBUG(0, 2599 "Kernel module is too old to handle metadata from %s.", 2600 cp->provider->name); 2601 return (EINVAL); 2602 } 2603 if (error != 0) { 2604 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2605 cp->provider->name); 2606 return (error); 2607 } 2608 2609 return (0); 2610} 2611 2612static int 2613g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2614 struct g_mirror_metadata *md) 2615{ 2616 2617 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2618 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2619 pp->name, md->md_did); 2620 return (EEXIST); 2621 } 2622 if (md->md_all != sc->sc_ndisks) { 2623 G_MIRROR_DEBUG(1, 2624 "Invalid '%s' field on disk %s (device %s), skipping.", 2625 "md_all", pp->name, sc->sc_name); 2626 return (EINVAL); 2627 } 2628 if (md->md_slice != sc->sc_slice) { 2629 G_MIRROR_DEBUG(1, 2630 "Invalid '%s' field on disk %s (device %s), skipping.", 2631 "md_slice", pp->name, sc->sc_name); 2632 return (EINVAL); 2633 } 2634 if (md->md_balance != sc->sc_balance) { 2635 G_MIRROR_DEBUG(1, 2636 "Invalid '%s' field on disk %s (device %s), skipping.", 2637 "md_balance", pp->name, sc->sc_name); 2638 return (EINVAL); 2639 } 2640 if (md->md_mediasize != sc->sc_mediasize) { 2641 G_MIRROR_DEBUG(1, 2642 "Invalid '%s' field on disk %s (device %s), skipping.", 2643 "md_mediasize", pp->name, sc->sc_name); 2644 return (EINVAL); 2645 } 2646 if (sc->sc_mediasize > pp->mediasize) { 2647 G_MIRROR_DEBUG(1, 2648 "Invalid size of disk %s (device %s), skipping.", pp->name, 2649 sc->sc_name); 2650 return (EINVAL); 2651 } 2652 if (md->md_sectorsize != sc->sc_sectorsize) { 2653 G_MIRROR_DEBUG(1, 2654 "Invalid '%s' field on disk %s (device %s), skipping.", 2655 "md_sectorsize", pp->name, sc->sc_name); 2656 return (EINVAL); 2657 } 2658 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2659 G_MIRROR_DEBUG(1, 2660 "Invalid sector size of disk %s (device %s), skipping.", 2661 pp->name, sc->sc_name); 2662 return (EINVAL); 2663 } 2664 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2665 G_MIRROR_DEBUG(1, 2666 "Invalid device flags on disk %s (device %s), skipping.", 2667 pp->name, sc->sc_name); 2668 return (EINVAL); 2669 } 2670 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2671 G_MIRROR_DEBUG(1, 2672 "Invalid disk flags on disk %s (device %s), skipping.", 2673 pp->name, sc->sc_name); 2674 return (EINVAL); 2675 } 2676 return (0); 2677} 2678 2679int 2680g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2681 struct g_mirror_metadata *md) 2682{ 2683 struct g_mirror_disk *disk; 2684 int error; 2685 2686 g_topology_assert_not(); 2687 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2688 2689 error = g_mirror_check_metadata(sc, pp, md); 2690 if (error != 0) 2691 return (error); 2692 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2693 md->md_genid < sc->sc_genid) { 2694 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2695 pp->name, sc->sc_name); 2696 return (EINVAL); 2697 } 2698 disk = g_mirror_init_disk(sc, pp, md, &error); 2699 if (disk == NULL) 2700 return (error); 2701 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2702 G_MIRROR_EVENT_WAIT); 2703 if (error != 0) 2704 return (error); 2705 if (md->md_version < G_MIRROR_VERSION) { 2706 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2707 pp->name, md->md_version, G_MIRROR_VERSION); 2708 g_mirror_update_metadata(disk); 2709 } 2710 return (0); 2711} 2712 2713static void 2714g_mirror_destroy_delayed(void *arg, int flag) 2715{ 2716 struct g_mirror_softc *sc; 2717 int error; 2718 2719 if (flag == EV_CANCEL) { 2720 G_MIRROR_DEBUG(1, "Destroying canceled."); 2721 return; 2722 } 2723 sc = arg; 2724 g_topology_unlock(); 2725 sx_xlock(&sc->sc_lock); 2726 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2727 ("DESTROY flag set on %s.", sc->sc_name)); 2728 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0, 2729 ("DESTROYING flag not set on %s.", sc->sc_name)); 2730 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 2731 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 2732 if (error != 0) { 2733 G_MIRROR_DEBUG(0, "Cannot destroy %s.", sc->sc_name); 2734 sx_xunlock(&sc->sc_lock); 2735 } 2736 g_topology_lock(); 2737} 2738 2739static int 2740g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2741{ 2742 struct g_mirror_softc *sc; 2743 int dcr, dcw, dce, error = 0; 2744 2745 g_topology_assert(); 2746 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2747 acw, ace); 2748 2749 sc = pp->geom->softc; 2750 if (sc == NULL && acr <= 0 && acw <= 0 && ace <= 0) 2751 return (0); 2752 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 2753 2754 dcr = pp->acr + acr; 2755 dcw = pp->acw + acw; 2756 dce = pp->ace + ace; 2757 2758 g_topology_unlock(); 2759 sx_xlock(&sc->sc_lock); 2760 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 2761 LIST_EMPTY(&sc->sc_disks)) { 2762 if (acr > 0 || acw > 0 || ace > 0) 2763 error = ENXIO; 2764 goto end; 2765 } 2766 if (dcw == 0 && !sc->sc_idle) 2767 g_mirror_idle(sc, dcw); 2768 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0) { 2769 if (acr > 0 || acw > 0 || ace > 0) { 2770 error = ENXIO; 2771 goto end; 2772 } 2773 if (dcr == 0 && dcw == 0 && dce == 0) { 2774 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, 2775 sc, NULL); 2776 } 2777 } 2778end: 2779 sx_xunlock(&sc->sc_lock); 2780 g_topology_lock(); 2781 return (error); 2782} 2783 2784static struct g_geom * 2785g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2786{ 2787 struct g_mirror_softc *sc; 2788 struct g_geom *gp; 2789 int error, timeout; 2790 2791 g_topology_assert(); 2792 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2793 md->md_mid); 2794 2795 /* One disk is minimum. */ 2796 if (md->md_all < 1) 2797 return (NULL); 2798 /* 2799 * Action geom. 2800 */ 2801 gp = g_new_geomf(mp, "%s", md->md_name); 2802 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2803 gp->start = g_mirror_start; 2804 gp->orphan = g_mirror_orphan; 2805 gp->access = g_mirror_access; 2806 gp->dumpconf = g_mirror_dumpconf; 2807 2808 sc->sc_id = md->md_mid; 2809 sc->sc_slice = md->md_slice; 2810 sc->sc_balance = md->md_balance; 2811 sc->sc_mediasize = md->md_mediasize; 2812 sc->sc_sectorsize = md->md_sectorsize; 2813 sc->sc_ndisks = md->md_all; 2814 sc->sc_flags = md->md_mflags; 2815 sc->sc_bump_id = 0; 2816 sc->sc_idle = 1; 2817 sc->sc_last_write = time_uptime; 2818 sc->sc_writes = 0; 2819 sx_init(&sc->sc_lock, "gmirror:lock"); 2820 bioq_init(&sc->sc_queue); 2821 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2822 bioq_init(&sc->sc_regular_delayed); 2823 bioq_init(&sc->sc_inflight); 2824 bioq_init(&sc->sc_sync_delayed); 2825 LIST_INIT(&sc->sc_disks); 2826 TAILQ_INIT(&sc->sc_events); 2827 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2828 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2829 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2830 gp->softc = sc; 2831 sc->sc_geom = gp; 2832 sc->sc_provider = NULL; 2833 /* 2834 * Synchronization geom. 2835 */ 2836 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2837 gp->softc = sc; 2838 gp->orphan = g_mirror_orphan; 2839 sc->sc_sync.ds_geom = gp; 2840 sc->sc_sync.ds_ndisks = 0; 2841 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2842 "g_mirror %s", md->md_name); 2843 if (error != 0) { 2844 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2845 sc->sc_name); 2846 g_destroy_geom(sc->sc_sync.ds_geom); 2847 mtx_destroy(&sc->sc_events_mtx); 2848 mtx_destroy(&sc->sc_queue_mtx); 2849 sx_destroy(&sc->sc_lock); 2850 g_destroy_geom(sc->sc_geom); 2851 free(sc, M_MIRROR); 2852 return (NULL); 2853 } 2854 2855 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 2856 sc->sc_name, sc->sc_ndisks, sc->sc_id); 2857 2858 sc->sc_rootmount = root_mount_hold("GMIRROR"); 2859 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 2860 /* 2861 * Run timeout. 2862 */ 2863 timeout = g_mirror_timeout * hz; 2864 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2865 return (sc->sc_geom); 2866} 2867 2868int 2869g_mirror_destroy(struct g_mirror_softc *sc, int how) 2870{ 2871 struct g_mirror_disk *disk; 2872 struct g_provider *pp; 2873 2874 g_topology_assert_not(); 2875 if (sc == NULL) 2876 return (ENXIO); 2877 sx_assert(&sc->sc_lock, SX_XLOCKED); 2878 2879 pp = sc->sc_provider; 2880 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2881 switch (how) { 2882 case G_MIRROR_DESTROY_SOFT: 2883 G_MIRROR_DEBUG(1, 2884 "Device %s is still open (r%dw%de%d).", pp->name, 2885 pp->acr, pp->acw, pp->ace); 2886 return (EBUSY); 2887 case G_MIRROR_DESTROY_DELAYED: 2888 G_MIRROR_DEBUG(1, 2889 "Device %s will be destroyed on last close.", 2890 pp->name); 2891 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2892 if (disk->d_state == 2893 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2894 g_mirror_sync_stop(disk, 1); 2895 } 2896 } 2897 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROYING; 2898 return (EBUSY); 2899 case G_MIRROR_DESTROY_HARD: 2900 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 2901 "can't be definitely removed.", pp->name); 2902 } 2903 } 2904 2905 g_topology_lock(); 2906 if (sc->sc_geom->softc == NULL) { 2907 g_topology_unlock(); 2908 return (0); 2909 } 2910 sc->sc_geom->softc = NULL; 2911 sc->sc_sync.ds_geom->softc = NULL; 2912 g_topology_unlock(); 2913 2914 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2915 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2916 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2917 sx_xunlock(&sc->sc_lock); 2918 mtx_lock(&sc->sc_queue_mtx); 2919 wakeup(sc); 2920 mtx_unlock(&sc->sc_queue_mtx); 2921 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2922 while (sc->sc_worker != NULL) 2923 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2924 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2925 sx_xlock(&sc->sc_lock); 2926 g_mirror_destroy_device(sc); 2927 free(sc, M_MIRROR); 2928 return (0); 2929} 2930 2931static void 2932g_mirror_taste_orphan(struct g_consumer *cp) 2933{ 2934 2935 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2936 cp->provider->name)); 2937} 2938 2939static struct g_geom * 2940g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2941{ 2942 struct g_mirror_metadata md; 2943 struct g_mirror_softc *sc; 2944 struct g_consumer *cp; 2945 struct g_geom *gp; 2946 int error; 2947 2948 g_topology_assert(); 2949 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2950 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2951 2952 gp = g_new_geomf(mp, "mirror:taste"); 2953 /* 2954 * This orphan function should be never called. 2955 */ 2956 gp->orphan = g_mirror_taste_orphan; 2957 cp = g_new_consumer(gp); 2958 g_attach(cp, pp); 2959 error = g_mirror_read_metadata(cp, &md); 2960 g_detach(cp); 2961 g_destroy_consumer(cp); 2962 g_destroy_geom(gp); 2963 if (error != 0) 2964 return (NULL); 2965 gp = NULL; 2966 2967 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2968 return (NULL); 2969 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 2970 return (NULL); 2971 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2972 G_MIRROR_DEBUG(0, 2973 "Device %s: provider %s marked as inactive, skipping.", 2974 md.md_name, pp->name); 2975 return (NULL); 2976 } 2977 if (g_mirror_debug >= 2) 2978 mirror_metadata_dump(&md); 2979 2980 /* 2981 * Let's check if device already exists. 2982 */ 2983 sc = NULL; 2984 LIST_FOREACH(gp, &mp->geom, geom) { 2985 sc = gp->softc; 2986 if (sc == NULL) 2987 continue; 2988 if (sc->sc_sync.ds_geom == gp) 2989 continue; 2990 if (strcmp(md.md_name, sc->sc_name) != 0) 2991 continue; 2992 if (md.md_mid != sc->sc_id) { 2993 G_MIRROR_DEBUG(0, "Device %s already configured.", 2994 sc->sc_name); 2995 return (NULL); 2996 } 2997 break; 2998 } 2999 if (gp == NULL) { 3000 gp = g_mirror_create(mp, &md); 3001 if (gp == NULL) { 3002 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3003 md.md_name); 3004 return (NULL); 3005 } 3006 sc = gp->softc; 3007 } 3008 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3009 g_topology_unlock(); 3010 sx_xlock(&sc->sc_lock); 3011 error = g_mirror_add_disk(sc, pp, &md); 3012 if (error != 0) { 3013 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3014 pp->name, gp->name, error); 3015 if (LIST_EMPTY(&sc->sc_disks)) { 3016 g_cancel_event(sc); 3017 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3018 g_topology_lock(); 3019 return (NULL); 3020 } 3021 gp = NULL; 3022 } 3023 sx_xunlock(&sc->sc_lock); 3024 g_topology_lock(); 3025 return (gp); 3026} 3027 3028static int 3029g_mirror_destroy_geom(struct gctl_req *req __unused, 3030 struct g_class *mp __unused, struct g_geom *gp) 3031{ 3032 struct g_mirror_softc *sc; 3033 int error; 3034 3035 g_topology_unlock(); 3036 sc = gp->softc; 3037 sx_xlock(&sc->sc_lock); 3038 g_cancel_event(sc); 3039 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3040 if (error != 0) 3041 sx_xunlock(&sc->sc_lock); 3042 g_topology_lock(); 3043 return (error); 3044} 3045 3046static void 3047g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3048 struct g_consumer *cp, struct g_provider *pp) 3049{ 3050 struct g_mirror_softc *sc; 3051 3052 g_topology_assert(); 3053 3054 sc = gp->softc; 3055 if (sc == NULL) 3056 return; 3057 /* Skip synchronization geom. */ 3058 if (gp == sc->sc_sync.ds_geom) 3059 return; 3060 if (pp != NULL) { 3061 /* Nothing here. */ 3062 } else if (cp != NULL) { 3063 struct g_mirror_disk *disk; 3064 3065 disk = cp->private; 3066 if (disk == NULL) 3067 return; 3068 g_topology_unlock(); 3069 sx_xlock(&sc->sc_lock); 3070 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3071 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3072 sbuf_printf(sb, "%s<Synchronized>", indent); 3073 if (disk->d_sync.ds_offset == 0) 3074 sbuf_printf(sb, "0%%"); 3075 else { 3076 sbuf_printf(sb, "%u%%", 3077 (u_int)((disk->d_sync.ds_offset * 100) / 3078 sc->sc_provider->mediasize)); 3079 } 3080 sbuf_printf(sb, "</Synchronized>\n"); 3081 } 3082 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3083 disk->d_sync.ds_syncid); 3084 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3085 disk->d_genid); 3086 sbuf_printf(sb, "%s<Flags>", indent); 3087 if (disk->d_flags == 0) 3088 sbuf_printf(sb, "NONE"); 3089 else { 3090 int first = 1; 3091 3092#define ADD_FLAG(flag, name) do { \ 3093 if ((disk->d_flags & (flag)) != 0) { \ 3094 if (!first) \ 3095 sbuf_printf(sb, ", "); \ 3096 else \ 3097 first = 0; \ 3098 sbuf_printf(sb, name); \ 3099 } \ 3100} while (0) 3101 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3102 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3103 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3104 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3105 "SYNCHRONIZING"); 3106 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3107 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3108#undef ADD_FLAG 3109 } 3110 sbuf_printf(sb, "</Flags>\n"); 3111 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3112 disk->d_priority); 3113 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3114 g_mirror_disk_state2str(disk->d_state)); 3115 sx_xunlock(&sc->sc_lock); 3116 g_topology_lock(); 3117 } else { 3118 g_topology_unlock(); 3119 sx_xlock(&sc->sc_lock); 3120 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3121 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3122 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3123 sbuf_printf(sb, "%s<Flags>", indent); 3124 if (sc->sc_flags == 0) 3125 sbuf_printf(sb, "NONE"); 3126 else { 3127 int first = 1; 3128 3129#define ADD_FLAG(flag, name) do { \ 3130 if ((sc->sc_flags & (flag)) != 0) { \ 3131 if (!first) \ 3132 sbuf_printf(sb, ", "); \ 3133 else \ 3134 first = 0; \ 3135 sbuf_printf(sb, name); \ 3136 } \ 3137} while (0) 3138 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3139#undef ADD_FLAG 3140 } 3141 sbuf_printf(sb, "</Flags>\n"); 3142 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3143 (u_int)sc->sc_slice); 3144 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3145 balance_name(sc->sc_balance)); 3146 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3147 sc->sc_ndisks); 3148 sbuf_printf(sb, "%s<State>", indent); 3149 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3150 sbuf_printf(sb, "%s", "STARTING"); 3151 else if (sc->sc_ndisks == 3152 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3153 sbuf_printf(sb, "%s", "COMPLETE"); 3154 else 3155 sbuf_printf(sb, "%s", "DEGRADED"); 3156 sbuf_printf(sb, "</State>\n"); 3157 sx_xunlock(&sc->sc_lock); 3158 g_topology_lock(); 3159 } 3160} 3161 3162static void 3163g_mirror_shutdown_pre_sync(void *arg, int howto) 3164{ 3165 struct g_class *mp; 3166 struct g_geom *gp, *gp2; 3167 struct g_mirror_softc *sc; 3168 int error; 3169 3170 mp = arg; 3171 DROP_GIANT(); 3172 g_topology_lock(); 3173 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3174 if ((sc = gp->softc) == NULL) 3175 continue; 3176 /* Skip synchronization geom. */ 3177 if (gp == sc->sc_sync.ds_geom) 3178 continue; 3179 g_topology_unlock(); 3180 sx_xlock(&sc->sc_lock); 3181 g_cancel_event(sc); 3182 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3183 if (error != 0) 3184 sx_xunlock(&sc->sc_lock); 3185 g_topology_lock(); 3186 } 3187 g_topology_unlock(); 3188 PICKUP_GIANT(); 3189} 3190 3191static void 3192g_mirror_init(struct g_class *mp) 3193{ 3194 3195 g_mirror_pre_sync = EVENTHANDLER_REGISTER(shutdown_pre_sync, 3196 g_mirror_shutdown_pre_sync, mp, SHUTDOWN_PRI_FIRST); 3197 if (g_mirror_pre_sync == NULL) 3198 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3199} 3200 3201static void 3202g_mirror_fini(struct g_class *mp) 3203{ 3204 3205 if (g_mirror_pre_sync != NULL) 3206 EVENTHANDLER_DEREGISTER(shutdown_pre_sync, g_mirror_pre_sync); 3207} 3208 3209DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3210