g_mirror.c revision 156684
1/*- 2 * Copyright (c) 2004-2005 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 156684 2006-03-13 14:48:45Z ru $"); 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 = 4; 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, g_mirror_post_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 gp->softc = NULL; 543 544 g_topology_lock(); 545 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 546 g_mirror_disconnect_consumer(sc, cp); 547 } 548 sc->sc_sync.ds_geom->softc = NULL; 549 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 550 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 551 g_wither_geom(gp, ENXIO); 552 g_topology_unlock(); 553 mtx_destroy(&sc->sc_queue_mtx); 554 mtx_destroy(&sc->sc_events_mtx); 555 sx_xunlock(&sc->sc_lock); 556 sx_destroy(&sc->sc_lock); 557} 558 559static void 560g_mirror_orphan(struct g_consumer *cp) 561{ 562 struct g_mirror_disk *disk; 563 564 g_topology_assert(); 565 566 disk = cp->private; 567 if (disk == NULL) 568 return; 569 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 570 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 571 G_MIRROR_EVENT_DONTWAIT); 572} 573 574/* 575 * Function should return the next active disk on the list. 576 * It is possible that it will be the same disk as given. 577 * If there are no active disks on list, NULL is returned. 578 */ 579static __inline struct g_mirror_disk * 580g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 581{ 582 struct g_mirror_disk *dp; 583 584 for (dp = LIST_NEXT(disk, d_next); dp != disk; 585 dp = LIST_NEXT(dp, d_next)) { 586 if (dp == NULL) 587 dp = LIST_FIRST(&sc->sc_disks); 588 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 589 break; 590 } 591 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 592 return (NULL); 593 return (dp); 594} 595 596static struct g_mirror_disk * 597g_mirror_get_disk(struct g_mirror_softc *sc) 598{ 599 struct g_mirror_disk *disk; 600 601 if (sc->sc_hint == NULL) { 602 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 603 if (sc->sc_hint == NULL) 604 return (NULL); 605 } 606 disk = sc->sc_hint; 607 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 608 disk = g_mirror_find_next(sc, disk); 609 if (disk == NULL) 610 return (NULL); 611 } 612 sc->sc_hint = g_mirror_find_next(sc, disk); 613 return (disk); 614} 615 616static int 617g_mirror_write_metadata(struct g_mirror_disk *disk, 618 struct g_mirror_metadata *md) 619{ 620 struct g_mirror_softc *sc; 621 struct g_consumer *cp; 622 off_t offset, length; 623 u_char *sector; 624 int error = 0; 625 626 g_topology_assert_not(); 627 sc = disk->d_softc; 628 sx_assert(&sc->sc_lock, SX_LOCKED); 629 630 cp = disk->d_consumer; 631 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 632 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 633 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 634 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 635 cp->acw, cp->ace)); 636 length = cp->provider->sectorsize; 637 offset = cp->provider->mediasize - length; 638 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 639 if (md != NULL) 640 mirror_metadata_encode(md, sector); 641 error = g_write_data(cp, offset, sector, length); 642 free(sector, M_MIRROR); 643 if (error != 0) { 644 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 645 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 646 G_MIRROR_DEBUG(0, "Cannot write metadata on %s " 647 "(device=%s, error=%d).", 648 g_mirror_get_diskname(disk), sc->sc_name, error); 649 } else { 650 G_MIRROR_DEBUG(1, "Cannot write metadata on %s " 651 "(device=%s, error=%d).", 652 g_mirror_get_diskname(disk), sc->sc_name, error); 653 } 654 if (g_mirror_disconnect_on_failure && 655 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 656 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 657 g_mirror_event_send(disk, 658 G_MIRROR_DISK_STATE_DISCONNECTED, 659 G_MIRROR_EVENT_DONTWAIT); 660 } 661 } 662 return (error); 663} 664 665static int 666g_mirror_clear_metadata(struct g_mirror_disk *disk) 667{ 668 int error; 669 670 g_topology_assert_not(); 671 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 672 673 error = g_mirror_write_metadata(disk, NULL); 674 if (error == 0) { 675 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 676 g_mirror_get_diskname(disk)); 677 } else { 678 G_MIRROR_DEBUG(0, 679 "Cannot clear metadata on disk %s (error=%d).", 680 g_mirror_get_diskname(disk), error); 681 } 682 return (error); 683} 684 685void 686g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 687 struct g_mirror_metadata *md) 688{ 689 690 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 691 md->md_version = G_MIRROR_VERSION; 692 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 693 md->md_mid = sc->sc_id; 694 md->md_all = sc->sc_ndisks; 695 md->md_slice = sc->sc_slice; 696 md->md_balance = sc->sc_balance; 697 md->md_genid = sc->sc_genid; 698 md->md_mediasize = sc->sc_mediasize; 699 md->md_sectorsize = sc->sc_sectorsize; 700 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 701 bzero(md->md_provider, sizeof(md->md_provider)); 702 if (disk == NULL) { 703 md->md_did = arc4random(); 704 md->md_priority = 0; 705 md->md_syncid = 0; 706 md->md_dflags = 0; 707 md->md_sync_offset = 0; 708 md->md_provsize = 0; 709 } else { 710 md->md_did = disk->d_id; 711 md->md_priority = disk->d_priority; 712 md->md_syncid = disk->d_sync.ds_syncid; 713 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 714 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 715 md->md_sync_offset = disk->d_sync.ds_offset_done; 716 else 717 md->md_sync_offset = 0; 718 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 719 strlcpy(md->md_provider, 720 disk->d_consumer->provider->name, 721 sizeof(md->md_provider)); 722 } 723 md->md_provsize = disk->d_consumer->provider->mediasize; 724 } 725} 726 727void 728g_mirror_update_metadata(struct g_mirror_disk *disk) 729{ 730 struct g_mirror_softc *sc; 731 struct g_mirror_metadata md; 732 int error; 733 734 g_topology_assert_not(); 735 sc = disk->d_softc; 736 sx_assert(&sc->sc_lock, SX_LOCKED); 737 738 g_mirror_fill_metadata(sc, disk, &md); 739 error = g_mirror_write_metadata(disk, &md); 740 if (error == 0) { 741 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 742 g_mirror_get_diskname(disk)); 743 } else { 744 G_MIRROR_DEBUG(0, 745 "Cannot update metadata on disk %s (error=%d).", 746 g_mirror_get_diskname(disk), error); 747 } 748} 749 750static void 751g_mirror_bump_syncid(struct g_mirror_softc *sc) 752{ 753 struct g_mirror_disk *disk; 754 755 g_topology_assert_not(); 756 sx_assert(&sc->sc_lock, SX_XLOCKED); 757 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 758 ("%s called with no active disks (device=%s).", __func__, 759 sc->sc_name)); 760 761 sc->sc_syncid++; 762 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 763 sc->sc_syncid); 764 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 765 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 766 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 767 disk->d_sync.ds_syncid = sc->sc_syncid; 768 g_mirror_update_metadata(disk); 769 } 770 } 771} 772 773static void 774g_mirror_bump_genid(struct g_mirror_softc *sc) 775{ 776 struct g_mirror_disk *disk; 777 778 g_topology_assert_not(); 779 sx_assert(&sc->sc_lock, SX_XLOCKED); 780 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 781 ("%s called with no active disks (device=%s).", __func__, 782 sc->sc_name)); 783 784 sc->sc_genid++; 785 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 786 sc->sc_genid); 787 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 788 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 789 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 790 disk->d_genid = sc->sc_genid; 791 g_mirror_update_metadata(disk); 792 } 793 } 794} 795 796static int 797g_mirror_idle(struct g_mirror_softc *sc, int acw) 798{ 799 struct g_mirror_disk *disk; 800 int timeout; 801 802 g_topology_assert_not(); 803 sx_assert(&sc->sc_lock, SX_XLOCKED); 804 805 if (sc->sc_provider == NULL) 806 return (0); 807 if (sc->sc_idle) 808 return (0); 809 if (sc->sc_writes > 0) 810 return (0); 811 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 812 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); 813 if (timeout > 0) 814 return (timeout); 815 } 816 sc->sc_idle = 1; 817 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 818 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 819 continue; 820 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 821 g_mirror_get_diskname(disk), sc->sc_name); 822 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 823 g_mirror_update_metadata(disk); 824 } 825 return (0); 826} 827 828static void 829g_mirror_unidle(struct g_mirror_softc *sc) 830{ 831 struct g_mirror_disk *disk; 832 833 g_topology_assert_not(); 834 sx_assert(&sc->sc_lock, SX_XLOCKED); 835 836 sc->sc_idle = 0; 837 sc->sc_last_write = time_uptime; 838 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 839 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 840 continue; 841 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 842 g_mirror_get_diskname(disk), sc->sc_name); 843 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 844 g_mirror_update_metadata(disk); 845 } 846} 847 848static __inline int 849bintime_cmp(struct bintime *bt1, struct bintime *bt2) 850{ 851 852 if (bt1->sec < bt2->sec) 853 return (-1); 854 else if (bt1->sec > bt2->sec) 855 return (1); 856 if (bt1->frac < bt2->frac) 857 return (-1); 858 else if (bt1->frac > bt2->frac) 859 return (1); 860 return (0); 861} 862 863static void 864g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 865{ 866 867 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 868 return; 869 binuptime(&disk->d_delay); 870 bintime_sub(&disk->d_delay, &bp->bio_t0); 871} 872 873static void 874g_mirror_done(struct bio *bp) 875{ 876 struct g_mirror_softc *sc; 877 878 sc = bp->bio_from->geom->softc; 879 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 880 mtx_lock(&sc->sc_queue_mtx); 881 bioq_disksort(&sc->sc_queue, bp); 882 wakeup(sc); 883 mtx_unlock(&sc->sc_queue_mtx); 884} 885 886static void 887g_mirror_regular_request(struct bio *bp) 888{ 889 struct g_mirror_softc *sc; 890 struct g_mirror_disk *disk; 891 struct bio *pbp; 892 893 g_topology_assert_not(); 894 895 pbp = bp->bio_parent; 896 sc = pbp->bio_to->geom->softc; 897 bp->bio_from->index--; 898 if (bp->bio_cmd == BIO_WRITE) 899 sc->sc_writes--; 900 disk = bp->bio_from->private; 901 if (disk == NULL) { 902 g_topology_lock(); 903 g_mirror_kill_consumer(sc, bp->bio_from); 904 g_topology_unlock(); 905 } else { 906 g_mirror_update_delay(disk, bp); 907 } 908 909 pbp->bio_inbed++; 910 KASSERT(pbp->bio_inbed <= pbp->bio_children, 911 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 912 pbp->bio_children)); 913 if (bp->bio_error == 0 && pbp->bio_error == 0) { 914 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 915 g_destroy_bio(bp); 916 if (pbp->bio_children == pbp->bio_inbed) { 917 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 918 pbp->bio_completed = pbp->bio_length; 919 if (pbp->bio_cmd == BIO_WRITE) { 920 bioq_remove(&sc->sc_inflight, pbp); 921 /* Release delayed sync requests if possible. */ 922 g_mirror_sync_release(sc); 923 } 924 g_io_deliver(pbp, pbp->bio_error); 925 } 926 return; 927 } else if (bp->bio_error != 0) { 928 if (pbp->bio_error == 0) 929 pbp->bio_error = bp->bio_error; 930 if (disk != NULL) { 931 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 932 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 933 G_MIRROR_LOGREQ(0, bp, 934 "Request failed (error=%d).", 935 bp->bio_error); 936 } else { 937 G_MIRROR_LOGREQ(1, bp, 938 "Request failed (error=%d).", 939 bp->bio_error); 940 } 941 if (g_mirror_disconnect_on_failure && 942 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) 943 { 944 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 945 g_mirror_event_send(disk, 946 G_MIRROR_DISK_STATE_DISCONNECTED, 947 G_MIRROR_EVENT_DONTWAIT); 948 } 949 } 950 switch (pbp->bio_cmd) { 951 case BIO_DELETE: 952 case BIO_WRITE: 953 pbp->bio_inbed--; 954 pbp->bio_children--; 955 break; 956 } 957 } 958 g_destroy_bio(bp); 959 960 switch (pbp->bio_cmd) { 961 case BIO_READ: 962 if (pbp->bio_inbed < pbp->bio_children) 963 break; 964 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 965 g_io_deliver(pbp, pbp->bio_error); 966 else { 967 pbp->bio_error = 0; 968 mtx_lock(&sc->sc_queue_mtx); 969 bioq_disksort(&sc->sc_queue, pbp); 970 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 971 wakeup(sc); 972 mtx_unlock(&sc->sc_queue_mtx); 973 } 974 break; 975 case BIO_DELETE: 976 case BIO_WRITE: 977 if (pbp->bio_children == 0) { 978 /* 979 * All requests failed. 980 */ 981 } else if (pbp->bio_inbed < pbp->bio_children) { 982 /* Do nothing. */ 983 break; 984 } else if (pbp->bio_children == pbp->bio_inbed) { 985 /* Some requests succeeded. */ 986 pbp->bio_error = 0; 987 pbp->bio_completed = pbp->bio_length; 988 } 989 bioq_remove(&sc->sc_inflight, pbp); 990 /* Release delayed sync requests if possible. */ 991 g_mirror_sync_release(sc); 992 g_io_deliver(pbp, pbp->bio_error); 993 break; 994 default: 995 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 996 break; 997 } 998} 999 1000static void 1001g_mirror_sync_done(struct bio *bp) 1002{ 1003 struct g_mirror_softc *sc; 1004 1005 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1006 sc = bp->bio_from->geom->softc; 1007 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 1008 mtx_lock(&sc->sc_queue_mtx); 1009 bioq_disksort(&sc->sc_queue, bp); 1010 wakeup(sc); 1011 mtx_unlock(&sc->sc_queue_mtx); 1012} 1013 1014static void 1015g_mirror_kernel_dump(struct bio *bp) 1016{ 1017 struct g_mirror_softc *sc; 1018 struct g_mirror_disk *disk; 1019 struct bio *cbp; 1020 struct g_kerneldump *gkd; 1021 1022 /* 1023 * We configure dumping to the first component, because this component 1024 * will be used for reading with 'prefer' balance algorithm. 1025 * If the component with the higest priority is currently disconnected 1026 * we will not be able to read the dump after the reboot if it will be 1027 * connected and synchronized later. Can we do something better? 1028 */ 1029 sc = bp->bio_to->geom->softc; 1030 disk = LIST_FIRST(&sc->sc_disks); 1031 1032 gkd = (struct g_kerneldump *)bp->bio_data; 1033 if (gkd->length > bp->bio_to->mediasize) 1034 gkd->length = bp->bio_to->mediasize; 1035 cbp = g_clone_bio(bp); 1036 if (cbp == NULL) { 1037 g_io_deliver(bp, ENOMEM); 1038 return; 1039 } 1040 cbp->bio_done = g_std_done; 1041 g_io_request(cbp, disk->d_consumer); 1042 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1043 g_mirror_get_diskname(disk)); 1044} 1045 1046static void 1047g_mirror_start(struct bio *bp) 1048{ 1049 struct g_mirror_softc *sc; 1050 1051 sc = bp->bio_to->geom->softc; 1052 /* 1053 * If sc == NULL or there are no valid disks, provider's error 1054 * should be set and g_mirror_start() should not be called at all. 1055 */ 1056 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1057 ("Provider's error should be set (error=%d)(mirror=%s).", 1058 bp->bio_to->error, bp->bio_to->name)); 1059 G_MIRROR_LOGREQ(3, bp, "Request received."); 1060 1061 switch (bp->bio_cmd) { 1062 case BIO_READ: 1063 case BIO_WRITE: 1064 case BIO_DELETE: 1065 break; 1066 case BIO_GETATTR: 1067 if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1068 g_mirror_kernel_dump(bp); 1069 return; 1070 } 1071 /* FALLTHROUGH */ 1072 default: 1073 g_io_deliver(bp, EOPNOTSUPP); 1074 return; 1075 } 1076 mtx_lock(&sc->sc_queue_mtx); 1077 bioq_disksort(&sc->sc_queue, bp); 1078 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1079 wakeup(sc); 1080 mtx_unlock(&sc->sc_queue_mtx); 1081} 1082 1083/* 1084 * Return TRUE if the given request is colliding with a in-progress 1085 * synchronization request. 1086 */ 1087static int 1088g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1089{ 1090 struct g_mirror_disk *disk; 1091 struct bio *sbp; 1092 off_t rstart, rend, sstart, send; 1093 int i; 1094 1095 if (sc->sc_sync.ds_ndisks == 0) 1096 return (0); 1097 rstart = bp->bio_offset; 1098 rend = bp->bio_offset + bp->bio_length; 1099 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1100 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1101 continue; 1102 for (i = 0; i < g_mirror_syncreqs; i++) { 1103 sbp = disk->d_sync.ds_bios[i]; 1104 if (sbp == NULL) 1105 continue; 1106 sstart = sbp->bio_offset; 1107 send = sbp->bio_offset + sbp->bio_length; 1108 if (rend > sstart && rstart < send) 1109 return (1); 1110 } 1111 } 1112 return (0); 1113} 1114 1115/* 1116 * Return TRUE if the given sync request is colliding with a in-progress regular 1117 * request. 1118 */ 1119static int 1120g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1121{ 1122 off_t rstart, rend, sstart, send; 1123 struct bio *bp; 1124 1125 if (sc->sc_sync.ds_ndisks == 0) 1126 return (0); 1127 sstart = sbp->bio_offset; 1128 send = sbp->bio_offset + sbp->bio_length; 1129 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { 1130 rstart = bp->bio_offset; 1131 rend = bp->bio_offset + bp->bio_length; 1132 if (rend > sstart && rstart < send) 1133 return (1); 1134 } 1135 return (0); 1136} 1137 1138/* 1139 * Puts request onto delayed queue. 1140 */ 1141static void 1142g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1143{ 1144 1145 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1146 bioq_insert_head(&sc->sc_regular_delayed, bp); 1147} 1148 1149/* 1150 * Puts synchronization request onto delayed queue. 1151 */ 1152static void 1153g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1154{ 1155 1156 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1157 bioq_insert_tail(&sc->sc_sync_delayed, bp); 1158} 1159 1160/* 1161 * Releases delayed regular requests which don't collide anymore with sync 1162 * requests. 1163 */ 1164static void 1165g_mirror_regular_release(struct g_mirror_softc *sc) 1166{ 1167 struct bio *bp, *bp2; 1168 1169 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { 1170 if (g_mirror_sync_collision(sc, bp)) 1171 continue; 1172 bioq_remove(&sc->sc_regular_delayed, bp); 1173 G_MIRROR_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1174 mtx_lock(&sc->sc_queue_mtx); 1175 bioq_insert_head(&sc->sc_queue, bp); 1176#if 0 1177 /* 1178 * wakeup() is not needed, because this function is called from 1179 * the worker thread. 1180 */ 1181 wakeup(&sc->sc_queue); 1182#endif 1183 mtx_unlock(&sc->sc_queue_mtx); 1184 } 1185} 1186 1187/* 1188 * Releases delayed sync requests which don't collide anymore with regular 1189 * requests. 1190 */ 1191static void 1192g_mirror_sync_release(struct g_mirror_softc *sc) 1193{ 1194 struct bio *bp, *bp2; 1195 1196 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { 1197 if (g_mirror_regular_collision(sc, bp)) 1198 continue; 1199 bioq_remove(&sc->sc_sync_delayed, bp); 1200 G_MIRROR_LOGREQ(2, bp, 1201 "Releasing delayed synchronization request."); 1202 g_io_request(bp, bp->bio_from); 1203 } 1204} 1205 1206/* 1207 * Handle synchronization requests. 1208 * Every synchronization request is two-steps process: first, READ request is 1209 * send to active provider and then WRITE request (with read data) to the provider 1210 * beeing synchronized. When WRITE is finished, new synchronization request is 1211 * send. 1212 */ 1213static void 1214g_mirror_sync_request(struct bio *bp) 1215{ 1216 struct g_mirror_softc *sc; 1217 struct g_mirror_disk *disk; 1218 1219 bp->bio_from->index--; 1220 sc = bp->bio_from->geom->softc; 1221 disk = bp->bio_from->private; 1222 if (disk == NULL) { 1223 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1224 g_topology_lock(); 1225 g_mirror_kill_consumer(sc, bp->bio_from); 1226 g_topology_unlock(); 1227 free(bp->bio_data, M_MIRROR); 1228 g_destroy_bio(bp); 1229 sx_xlock(&sc->sc_lock); 1230 return; 1231 } 1232 1233 /* 1234 * Synchronization request. 1235 */ 1236 switch (bp->bio_cmd) { 1237 case BIO_READ: 1238 { 1239 struct g_consumer *cp; 1240 1241 if (bp->bio_error != 0) { 1242 G_MIRROR_LOGREQ(0, bp, 1243 "Synchronization request failed (error=%d).", 1244 bp->bio_error); 1245 g_destroy_bio(bp); 1246 return; 1247 } 1248 G_MIRROR_LOGREQ(3, bp, 1249 "Synchronization request half-finished."); 1250 bp->bio_cmd = BIO_WRITE; 1251 bp->bio_cflags = 0; 1252 cp = disk->d_consumer; 1253 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1254 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1255 cp->acr, cp->acw, cp->ace)); 1256 cp->index++; 1257 g_io_request(bp, cp); 1258 return; 1259 } 1260 case BIO_WRITE: 1261 { 1262 struct g_mirror_disk_sync *sync; 1263 off_t offset; 1264 void *data; 1265 int i; 1266 1267 if (bp->bio_error != 0) { 1268 G_MIRROR_LOGREQ(0, bp, 1269 "Synchronization request failed (error=%d).", 1270 bp->bio_error); 1271 g_destroy_bio(bp); 1272 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1273 g_mirror_event_send(disk, 1274 G_MIRROR_DISK_STATE_DISCONNECTED, 1275 G_MIRROR_EVENT_DONTWAIT); 1276 return; 1277 } 1278 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1279 sync = &disk->d_sync; 1280 if (sync->ds_offset == sc->sc_mediasize || 1281 sync->ds_consumer == NULL || 1282 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1283 /* Don't send more synchronization requests. */ 1284 sync->ds_inflight--; 1285 if (sync->ds_bios != NULL) { 1286 i = (int)(uintptr_t)bp->bio_caller1; 1287 sync->ds_bios[i] = NULL; 1288 } 1289 free(bp->bio_data, M_MIRROR); 1290 g_destroy_bio(bp); 1291 if (sync->ds_inflight > 0) 1292 return; 1293 if (sync->ds_consumer == NULL || 1294 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1295 return; 1296 } 1297 /* Disk up-to-date, activate it. */ 1298 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1299 G_MIRROR_EVENT_DONTWAIT); 1300 return; 1301 } 1302 1303 /* Send next synchronization request. */ 1304 data = bp->bio_data; 1305 bzero(bp, sizeof(*bp)); 1306 bp->bio_cmd = BIO_READ; 1307 bp->bio_offset = sync->ds_offset; 1308 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1309 sync->ds_offset += bp->bio_length; 1310 bp->bio_done = g_mirror_sync_done; 1311 bp->bio_data = data; 1312 bp->bio_from = sync->ds_consumer; 1313 bp->bio_to = sc->sc_provider; 1314 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1315 sync->ds_consumer->index++; 1316 /* 1317 * Delay the request if it is colliding with a regular request. 1318 */ 1319 if (g_mirror_regular_collision(sc, bp)) 1320 g_mirror_sync_delay(sc, bp); 1321 else 1322 g_io_request(bp, sync->ds_consumer); 1323 1324 /* Release delayed requests if possible. */ 1325 g_mirror_regular_release(sc); 1326 1327 /* Find the smallest offset */ 1328 offset = sc->sc_mediasize; 1329 for (i = 0; i < g_mirror_syncreqs; i++) { 1330 bp = sync->ds_bios[i]; 1331 if (bp->bio_offset < offset) 1332 offset = bp->bio_offset; 1333 } 1334 if (sync->ds_offset_done + (MAXPHYS * 100) < offset) { 1335 /* Update offset_done on every 100 blocks. */ 1336 sync->ds_offset_done = offset; 1337 g_mirror_update_metadata(disk); 1338 } 1339 return; 1340 } 1341 default: 1342 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1343 bp->bio_cmd, sc->sc_name)); 1344 break; 1345 } 1346} 1347 1348static void 1349g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1350{ 1351 struct g_mirror_disk *disk; 1352 struct g_consumer *cp; 1353 struct bio *cbp; 1354 1355 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1356 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1357 break; 1358 } 1359 if (disk == NULL) { 1360 if (bp->bio_error == 0) 1361 bp->bio_error = ENXIO; 1362 g_io_deliver(bp, bp->bio_error); 1363 return; 1364 } 1365 cbp = g_clone_bio(bp); 1366 if (cbp == NULL) { 1367 if (bp->bio_error == 0) 1368 bp->bio_error = ENOMEM; 1369 g_io_deliver(bp, bp->bio_error); 1370 return; 1371 } 1372 /* 1373 * Fill in the component buf structure. 1374 */ 1375 cp = disk->d_consumer; 1376 cbp->bio_done = g_mirror_done; 1377 cbp->bio_to = cp->provider; 1378 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1379 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1380 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1381 cp->acw, cp->ace)); 1382 cp->index++; 1383 g_io_request(cbp, cp); 1384} 1385 1386static void 1387g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1388{ 1389 struct g_mirror_disk *disk; 1390 struct g_consumer *cp; 1391 struct bio *cbp; 1392 1393 disk = g_mirror_get_disk(sc); 1394 if (disk == NULL) { 1395 if (bp->bio_error == 0) 1396 bp->bio_error = ENXIO; 1397 g_io_deliver(bp, bp->bio_error); 1398 return; 1399 } 1400 cbp = g_clone_bio(bp); 1401 if (cbp == NULL) { 1402 if (bp->bio_error == 0) 1403 bp->bio_error = ENOMEM; 1404 g_io_deliver(bp, bp->bio_error); 1405 return; 1406 } 1407 /* 1408 * Fill in the component buf structure. 1409 */ 1410 cp = disk->d_consumer; 1411 cbp->bio_done = g_mirror_done; 1412 cbp->bio_to = cp->provider; 1413 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1414 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1415 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1416 cp->acw, cp->ace)); 1417 cp->index++; 1418 g_io_request(cbp, cp); 1419} 1420 1421static void 1422g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1423{ 1424 struct g_mirror_disk *disk, *dp; 1425 struct g_consumer *cp; 1426 struct bio *cbp; 1427 struct bintime curtime; 1428 1429 binuptime(&curtime); 1430 /* 1431 * Find a disk which the smallest load. 1432 */ 1433 disk = NULL; 1434 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1435 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1436 continue; 1437 /* If disk wasn't used for more than 2 sec, use it. */ 1438 if (curtime.sec - dp->d_last_used.sec >= 2) { 1439 disk = dp; 1440 break; 1441 } 1442 if (disk == NULL || 1443 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1444 disk = dp; 1445 } 1446 } 1447 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1448 cbp = g_clone_bio(bp); 1449 if (cbp == NULL) { 1450 if (bp->bio_error == 0) 1451 bp->bio_error = ENOMEM; 1452 g_io_deliver(bp, bp->bio_error); 1453 return; 1454 } 1455 /* 1456 * Fill in the component buf structure. 1457 */ 1458 cp = disk->d_consumer; 1459 cbp->bio_done = g_mirror_done; 1460 cbp->bio_to = cp->provider; 1461 binuptime(&disk->d_last_used); 1462 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1463 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1464 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1465 cp->acw, cp->ace)); 1466 cp->index++; 1467 g_io_request(cbp, cp); 1468} 1469 1470static void 1471g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1472{ 1473 struct bio_queue_head queue; 1474 struct g_mirror_disk *disk; 1475 struct g_consumer *cp; 1476 struct bio *cbp; 1477 off_t left, mod, offset, slice; 1478 u_char *data; 1479 u_int ndisks; 1480 1481 if (bp->bio_length <= sc->sc_slice) { 1482 g_mirror_request_round_robin(sc, bp); 1483 return; 1484 } 1485 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1486 slice = bp->bio_length / ndisks; 1487 mod = slice % sc->sc_provider->sectorsize; 1488 if (mod != 0) 1489 slice += sc->sc_provider->sectorsize - mod; 1490 /* 1491 * Allocate all bios before sending any request, so we can 1492 * return ENOMEM in nice and clean way. 1493 */ 1494 left = bp->bio_length; 1495 offset = bp->bio_offset; 1496 data = bp->bio_data; 1497 bioq_init(&queue); 1498 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1499 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1500 continue; 1501 cbp = g_clone_bio(bp); 1502 if (cbp == NULL) { 1503 for (cbp = bioq_first(&queue); cbp != NULL; 1504 cbp = bioq_first(&queue)) { 1505 bioq_remove(&queue, cbp); 1506 g_destroy_bio(cbp); 1507 } 1508 if (bp->bio_error == 0) 1509 bp->bio_error = ENOMEM; 1510 g_io_deliver(bp, bp->bio_error); 1511 return; 1512 } 1513 bioq_insert_tail(&queue, cbp); 1514 cbp->bio_done = g_mirror_done; 1515 cbp->bio_caller1 = disk; 1516 cbp->bio_to = disk->d_consumer->provider; 1517 cbp->bio_offset = offset; 1518 cbp->bio_data = data; 1519 cbp->bio_length = MIN(left, slice); 1520 left -= cbp->bio_length; 1521 if (left == 0) 1522 break; 1523 offset += cbp->bio_length; 1524 data += cbp->bio_length; 1525 } 1526 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1527 bioq_remove(&queue, cbp); 1528 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1529 disk = cbp->bio_caller1; 1530 cbp->bio_caller1 = NULL; 1531 cp = disk->d_consumer; 1532 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1533 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1534 cp->acr, cp->acw, cp->ace)); 1535 disk->d_consumer->index++; 1536 g_io_request(cbp, disk->d_consumer); 1537 } 1538} 1539 1540static void 1541g_mirror_register_request(struct bio *bp) 1542{ 1543 struct g_mirror_softc *sc; 1544 1545 sc = bp->bio_to->geom->softc; 1546 switch (bp->bio_cmd) { 1547 case BIO_READ: 1548 switch (sc->sc_balance) { 1549 case G_MIRROR_BALANCE_LOAD: 1550 g_mirror_request_load(sc, bp); 1551 break; 1552 case G_MIRROR_BALANCE_PREFER: 1553 g_mirror_request_prefer(sc, bp); 1554 break; 1555 case G_MIRROR_BALANCE_ROUND_ROBIN: 1556 g_mirror_request_round_robin(sc, bp); 1557 break; 1558 case G_MIRROR_BALANCE_SPLIT: 1559 g_mirror_request_split(sc, bp); 1560 break; 1561 } 1562 return; 1563 case BIO_WRITE: 1564 case BIO_DELETE: 1565 { 1566 struct g_mirror_disk *disk; 1567 struct g_mirror_disk_sync *sync; 1568 struct bio_queue_head queue; 1569 struct g_consumer *cp; 1570 struct bio *cbp; 1571 1572 /* 1573 * Delay the request if it is colliding with a synchronization 1574 * request. 1575 */ 1576 if (g_mirror_sync_collision(sc, bp)) { 1577 g_mirror_regular_delay(sc, bp); 1578 return; 1579 } 1580 1581 if (sc->sc_idle) 1582 g_mirror_unidle(sc); 1583 else 1584 sc->sc_last_write = time_uptime; 1585 1586 /* 1587 * Allocate all bios before sending any request, so we can 1588 * return ENOMEM in nice and clean way. 1589 */ 1590 bioq_init(&queue); 1591 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1592 sync = &disk->d_sync; 1593 switch (disk->d_state) { 1594 case G_MIRROR_DISK_STATE_ACTIVE: 1595 break; 1596 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1597 if (bp->bio_offset >= sync->ds_offset) 1598 continue; 1599 break; 1600 default: 1601 continue; 1602 } 1603 cbp = g_clone_bio(bp); 1604 if (cbp == NULL) { 1605 for (cbp = bioq_first(&queue); cbp != NULL; 1606 cbp = bioq_first(&queue)) { 1607 bioq_remove(&queue, cbp); 1608 g_destroy_bio(cbp); 1609 } 1610 if (bp->bio_error == 0) 1611 bp->bio_error = ENOMEM; 1612 g_io_deliver(bp, bp->bio_error); 1613 return; 1614 } 1615 bioq_insert_tail(&queue, cbp); 1616 cbp->bio_done = g_mirror_done; 1617 cp = disk->d_consumer; 1618 cbp->bio_caller1 = cp; 1619 cbp->bio_to = cp->provider; 1620 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1621 ("Consumer %s not opened (r%dw%de%d).", 1622 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1623 } 1624 for (cbp = bioq_first(&queue); cbp != NULL; 1625 cbp = bioq_first(&queue)) { 1626 bioq_remove(&queue, cbp); 1627 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1628 cp = cbp->bio_caller1; 1629 cbp->bio_caller1 = NULL; 1630 cp->index++; 1631 sc->sc_writes++; 1632 g_io_request(cbp, cp); 1633 } 1634 /* 1635 * Put request onto inflight queue, so we can check if new 1636 * synchronization requests don't collide with it. 1637 */ 1638 bioq_insert_tail(&sc->sc_inflight, bp); 1639 /* 1640 * Bump syncid on first write. 1641 */ 1642 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1643 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1644 g_mirror_bump_syncid(sc); 1645 } 1646 return; 1647 } 1648 default: 1649 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1650 bp->bio_cmd, sc->sc_name)); 1651 break; 1652 } 1653} 1654 1655static int 1656g_mirror_can_destroy(struct g_mirror_softc *sc) 1657{ 1658 struct g_geom *gp; 1659 struct g_consumer *cp; 1660 1661 g_topology_assert(); 1662 gp = sc->sc_geom; 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 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1693 g_topology_unlock(); 1694 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1695 &sc->sc_worker); 1696 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1697 sx_xunlock(&sc->sc_lock); 1698 wakeup(&sc->sc_worker); 1699 sc->sc_worker = NULL; 1700 } else { 1701 g_topology_unlock(); 1702 g_mirror_destroy_device(sc); 1703 free(sc, M_MIRROR); 1704 } 1705 return (1); 1706} 1707 1708/* 1709 * Worker thread. 1710 */ 1711static void 1712g_mirror_worker(void *arg) 1713{ 1714 struct g_mirror_softc *sc; 1715 struct g_mirror_event *ep; 1716 struct bio *bp; 1717 int timeout; 1718 1719 sc = arg; 1720 mtx_lock_spin(&sched_lock); 1721 sched_prio(curthread, PRIBIO); 1722 mtx_unlock_spin(&sched_lock); 1723 1724 sx_xlock(&sc->sc_lock); 1725 for (;;) { 1726 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1727 /* 1728 * First take a look at events. 1729 * This is important to handle events before any I/O requests. 1730 */ 1731 ep = g_mirror_event_get(sc); 1732 if (ep != NULL) { 1733 g_mirror_event_remove(sc, ep); 1734 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1735 /* Update only device status. */ 1736 G_MIRROR_DEBUG(3, 1737 "Running event for device %s.", 1738 sc->sc_name); 1739 ep->e_error = 0; 1740 g_mirror_update_device(sc, 1); 1741 } else { 1742 /* Update disk status. */ 1743 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1744 g_mirror_get_diskname(ep->e_disk)); 1745 ep->e_error = g_mirror_update_disk(ep->e_disk, 1746 ep->e_state); 1747 if (ep->e_error == 0) 1748 g_mirror_update_device(sc, 0); 1749 } 1750 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1751 KASSERT(ep->e_error == 0, 1752 ("Error cannot be handled.")); 1753 g_mirror_event_free(ep); 1754 } else { 1755 ep->e_flags |= G_MIRROR_EVENT_DONE; 1756 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1757 ep); 1758 mtx_lock(&sc->sc_events_mtx); 1759 wakeup(ep); 1760 mtx_unlock(&sc->sc_events_mtx); 1761 } 1762 if ((sc->sc_flags & 1763 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1764 if (g_mirror_try_destroy(sc)) { 1765 curthread->td_pflags &= ~TDP_GEOM; 1766 G_MIRROR_DEBUG(1, "Thread exiting."); 1767 kthread_exit(0); 1768 } 1769 } 1770 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1771 continue; 1772 } 1773 /* 1774 * Check if we can mark array as CLEAN and if we can't take 1775 * how much seconds should we wait. 1776 */ 1777 timeout = g_mirror_idle(sc, -1); 1778 /* 1779 * Now I/O requests. 1780 */ 1781 /* Get first request from the queue. */ 1782 mtx_lock(&sc->sc_queue_mtx); 1783 bp = bioq_first(&sc->sc_queue); 1784 if (bp == NULL) { 1785 if (ep != NULL) { 1786 /* 1787 * We have a pending even, try to serve it 1788 * again. 1789 */ 1790 mtx_unlock(&sc->sc_queue_mtx); 1791 continue; 1792 } 1793 if ((sc->sc_flags & 1794 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1795 mtx_unlock(&sc->sc_queue_mtx); 1796 if (g_mirror_try_destroy(sc)) { 1797 curthread->td_pflags &= ~TDP_GEOM; 1798 G_MIRROR_DEBUG(1, "Thread exiting."); 1799 kthread_exit(0); 1800 } 1801 mtx_lock(&sc->sc_queue_mtx); 1802 } 1803 sx_xunlock(&sc->sc_lock); 1804 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1805 timeout * hz); 1806 sx_xlock(&sc->sc_lock); 1807 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1808 continue; 1809 } 1810 bioq_remove(&sc->sc_queue, bp); 1811 mtx_unlock(&sc->sc_queue_mtx); 1812 1813 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 1814 g_mirror_regular_request(bp); 1815 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 1816 g_mirror_sync_request(bp); 1817 else 1818 g_mirror_register_request(bp); 1819 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1820 } 1821} 1822 1823static void 1824g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 1825{ 1826 1827 sx_assert(&sc->sc_lock, SX_LOCKED); 1828 1829 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1830 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1831 g_mirror_get_diskname(disk), sc->sc_name); 1832 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1833 } else if (sc->sc_idle && 1834 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1835 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1836 g_mirror_get_diskname(disk), sc->sc_name); 1837 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1838 } 1839} 1840 1841static void 1842g_mirror_sync_start(struct g_mirror_disk *disk) 1843{ 1844 struct g_mirror_softc *sc; 1845 struct g_consumer *cp; 1846 struct bio *bp; 1847 int error, i; 1848 1849 g_topology_assert_not(); 1850 sc = disk->d_softc; 1851 sx_assert(&sc->sc_lock, SX_LOCKED); 1852 1853 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1854 ("Disk %s is not marked for synchronization.", 1855 g_mirror_get_diskname(disk))); 1856 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1857 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1858 sc->sc_state)); 1859 1860 sx_xunlock(&sc->sc_lock); 1861 g_topology_lock(); 1862 cp = g_new_consumer(sc->sc_sync.ds_geom); 1863 error = g_attach(cp, sc->sc_provider); 1864 KASSERT(error == 0, 1865 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 1866 error = g_access(cp, 1, 0, 0); 1867 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 1868 g_topology_unlock(); 1869 sx_xlock(&sc->sc_lock); 1870 1871 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1872 g_mirror_get_diskname(disk)); 1873 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1874 KASSERT(disk->d_sync.ds_consumer == NULL, 1875 ("Sync consumer already exists (device=%s, disk=%s).", 1876 sc->sc_name, g_mirror_get_diskname(disk))); 1877 1878 disk->d_sync.ds_consumer = cp; 1879 disk->d_sync.ds_consumer->private = disk; 1880 disk->d_sync.ds_consumer->index = 0; 1881 1882 /* 1883 * Allocate memory for synchronization bios and initialize them. 1884 */ 1885 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 1886 M_MIRROR, M_WAITOK); 1887 for (i = 0; i < g_mirror_syncreqs; i++) { 1888 bp = g_alloc_bio(); 1889 disk->d_sync.ds_bios[i] = bp; 1890 bp->bio_parent = NULL; 1891 bp->bio_cmd = BIO_READ; 1892 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1893 bp->bio_cflags = 0; 1894 bp->bio_offset = disk->d_sync.ds_offset; 1895 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1896 disk->d_sync.ds_offset += bp->bio_length; 1897 bp->bio_done = g_mirror_sync_done; 1898 bp->bio_from = disk->d_sync.ds_consumer; 1899 bp->bio_to = sc->sc_provider; 1900 bp->bio_caller1 = (void *)(uintptr_t)i; 1901 } 1902 1903 /* Increase the number of disks in SYNCHRONIZING state. */ 1904 sc->sc_sync.ds_ndisks++; 1905 /* Set the number of in-flight synchronization requests. */ 1906 disk->d_sync.ds_inflight = g_mirror_syncreqs; 1907 1908 /* 1909 * Fire off first synchronization requests. 1910 */ 1911 for (i = 0; i < g_mirror_syncreqs; i++) { 1912 bp = disk->d_sync.ds_bios[i]; 1913 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1914 disk->d_sync.ds_consumer->index++; 1915 /* 1916 * Delay the request if it is colliding with a regular request. 1917 */ 1918 if (g_mirror_regular_collision(sc, bp)) 1919 g_mirror_sync_delay(sc, bp); 1920 else 1921 g_io_request(bp, disk->d_sync.ds_consumer); 1922 } 1923} 1924 1925/* 1926 * Stop synchronization process. 1927 * type: 0 - synchronization finished 1928 * 1 - synchronization stopped 1929 */ 1930static void 1931g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1932{ 1933 struct g_mirror_softc *sc; 1934 struct g_consumer *cp; 1935 1936 g_topology_assert_not(); 1937 sc = disk->d_softc; 1938 sx_assert(&sc->sc_lock, SX_LOCKED); 1939 1940 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1941 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1942 g_mirror_disk_state2str(disk->d_state))); 1943 if (disk->d_sync.ds_consumer == NULL) 1944 return; 1945 1946 if (type == 0) { 1947 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1948 sc->sc_name, g_mirror_get_diskname(disk)); 1949 } else /* if (type == 1) */ { 1950 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1951 sc->sc_name, g_mirror_get_diskname(disk)); 1952 } 1953 free(disk->d_sync.ds_bios, M_MIRROR); 1954 disk->d_sync.ds_bios = NULL; 1955 cp = disk->d_sync.ds_consumer; 1956 disk->d_sync.ds_consumer = NULL; 1957 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1958 sc->sc_sync.ds_ndisks--; 1959 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1960 g_topology_lock(); 1961 g_mirror_kill_consumer(sc, cp); 1962 g_topology_unlock(); 1963 sx_xlock(&sc->sc_lock); 1964} 1965 1966static void 1967g_mirror_launch_provider(struct g_mirror_softc *sc) 1968{ 1969 struct g_mirror_disk *disk; 1970 struct g_provider *pp; 1971 1972 sx_assert(&sc->sc_lock, SX_LOCKED); 1973 1974 g_topology_lock(); 1975 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1976 pp->mediasize = sc->sc_mediasize; 1977 pp->sectorsize = sc->sc_sectorsize; 1978 sc->sc_provider = pp; 1979 g_error_provider(pp, 0); 1980 g_topology_unlock(); 1981 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1982 pp->name); 1983 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1984 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1985 g_mirror_sync_start(disk); 1986 } 1987} 1988 1989static void 1990g_mirror_destroy_provider(struct g_mirror_softc *sc) 1991{ 1992 struct g_mirror_disk *disk; 1993 struct bio *bp; 1994 1995 g_topology_assert_not(); 1996 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1997 sc->sc_name)); 1998 1999 g_topology_lock(); 2000 g_error_provider(sc->sc_provider, ENXIO); 2001 mtx_lock(&sc->sc_queue_mtx); 2002 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 2003 bioq_remove(&sc->sc_queue, bp); 2004 g_io_deliver(bp, ENXIO); 2005 } 2006 mtx_unlock(&sc->sc_queue_mtx); 2007 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 2008 sc->sc_provider->name); 2009 sc->sc_provider->flags |= G_PF_WITHER; 2010 g_orphan_provider(sc->sc_provider, ENXIO); 2011 g_topology_unlock(); 2012 sc->sc_provider = NULL; 2013 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2014 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2015 g_mirror_sync_stop(disk, 1); 2016 } 2017} 2018 2019static void 2020g_mirror_go(void *arg) 2021{ 2022 struct g_mirror_softc *sc; 2023 2024 sc = arg; 2025 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2026 g_mirror_event_send(sc, 0, 2027 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2028} 2029 2030static u_int 2031g_mirror_determine_state(struct g_mirror_disk *disk) 2032{ 2033 struct g_mirror_softc *sc; 2034 u_int state; 2035 2036 sc = disk->d_softc; 2037 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2038 if ((disk->d_flags & 2039 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2040 /* Disk does not need synchronization. */ 2041 state = G_MIRROR_DISK_STATE_ACTIVE; 2042 } else { 2043 if ((sc->sc_flags & 2044 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2045 (disk->d_flags & 2046 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2047 /* 2048 * We can start synchronization from 2049 * the stored offset. 2050 */ 2051 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2052 } else { 2053 state = G_MIRROR_DISK_STATE_STALE; 2054 } 2055 } 2056 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2057 /* 2058 * Reset all synchronization data for this disk, 2059 * because if it even was synchronized, it was 2060 * synchronized to disks with different syncid. 2061 */ 2062 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2063 disk->d_sync.ds_offset = 0; 2064 disk->d_sync.ds_offset_done = 0; 2065 disk->d_sync.ds_syncid = sc->sc_syncid; 2066 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2067 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2068 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2069 } else { 2070 state = G_MIRROR_DISK_STATE_STALE; 2071 } 2072 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2073 /* 2074 * Not good, NOT GOOD! 2075 * It means that mirror was started on stale disks 2076 * and more fresh disk just arrive. 2077 * If there were writes, mirror is fucked up, sorry. 2078 * I think the best choice here is don't touch 2079 * this disk and inform the user laudly. 2080 */ 2081 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2082 "disk (%s) arrives!! It will not be connected to the " 2083 "running device.", sc->sc_name, 2084 g_mirror_get_diskname(disk)); 2085 g_mirror_destroy_disk(disk); 2086 state = G_MIRROR_DISK_STATE_NONE; 2087 /* Return immediately, because disk was destroyed. */ 2088 return (state); 2089 } 2090 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2091 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2092 return (state); 2093} 2094 2095/* 2096 * Update device state. 2097 */ 2098static void 2099g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 2100{ 2101 struct g_mirror_disk *disk; 2102 u_int state; 2103 2104 sx_assert(&sc->sc_lock, SX_XLOCKED); 2105 2106 switch (sc->sc_state) { 2107 case G_MIRROR_DEVICE_STATE_STARTING: 2108 { 2109 struct g_mirror_disk *pdisk, *tdisk; 2110 u_int dirty, ndisks, genid, syncid; 2111 2112 KASSERT(sc->sc_provider == NULL, 2113 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2114 /* 2115 * Are we ready? We are, if all disks are connected or 2116 * if we have any disks and 'force' is true. 2117 */ 2118 ndisks = g_mirror_ndisks(sc, -1); 2119 if ((force && ndisks) || sc->sc_ndisks == ndisks) { 2120 ; 2121 } else if (ndisks == 0) { 2122 /* 2123 * Disks went down in starting phase, so destroy 2124 * device. 2125 */ 2126 callout_drain(&sc->sc_callout); 2127 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2128 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2129 sc->sc_rootmount); 2130 root_mount_rel(sc->sc_rootmount); 2131 sc->sc_rootmount = NULL; 2132 return; 2133 } else { 2134 return; 2135 } 2136 2137 /* 2138 * Activate all disks with the biggest syncid. 2139 */ 2140 if (force) { 2141 /* 2142 * If 'force' is true, we have been called due to 2143 * timeout, so don't bother canceling timeout. 2144 */ 2145 ndisks = 0; 2146 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2147 if ((disk->d_flags & 2148 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2149 ndisks++; 2150 } 2151 } 2152 if (ndisks == 0) { 2153 /* No valid disks found, destroy device. */ 2154 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2155 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2156 __LINE__, sc->sc_rootmount); 2157 root_mount_rel(sc->sc_rootmount); 2158 sc->sc_rootmount = NULL; 2159 return; 2160 } 2161 } else { 2162 /* Cancel timeout. */ 2163 callout_drain(&sc->sc_callout); 2164 } 2165 2166 /* 2167 * Find the biggest genid. 2168 */ 2169 genid = 0; 2170 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2171 if (disk->d_genid > genid) 2172 genid = disk->d_genid; 2173 } 2174 sc->sc_genid = genid; 2175 /* 2176 * Remove all disks without the biggest genid. 2177 */ 2178 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2179 if (disk->d_genid < genid) { 2180 G_MIRROR_DEBUG(0, 2181 "Component %s (device %s) broken, skipping.", 2182 g_mirror_get_diskname(disk), sc->sc_name); 2183 g_mirror_destroy_disk(disk); 2184 } 2185 } 2186 2187 /* 2188 * Find the biggest syncid. 2189 */ 2190 syncid = 0; 2191 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2192 if (disk->d_sync.ds_syncid > syncid) 2193 syncid = disk->d_sync.ds_syncid; 2194 } 2195 2196 /* 2197 * Here we need to look for dirty disks and if all disks 2198 * with the biggest syncid are dirty, we have to choose 2199 * one with the biggest priority and rebuild the rest. 2200 */ 2201 /* 2202 * Find the number of dirty disks with the biggest syncid. 2203 * Find the number of disks with the biggest syncid. 2204 * While here, find a disk with the biggest priority. 2205 */ 2206 dirty = ndisks = 0; 2207 pdisk = NULL; 2208 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2209 if (disk->d_sync.ds_syncid != syncid) 2210 continue; 2211 if ((disk->d_flags & 2212 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2213 continue; 2214 } 2215 ndisks++; 2216 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2217 dirty++; 2218 if (pdisk == NULL || 2219 pdisk->d_priority < disk->d_priority) { 2220 pdisk = disk; 2221 } 2222 } 2223 } 2224 if (dirty == 0) { 2225 /* No dirty disks at all, great. */ 2226 } else if (dirty == ndisks) { 2227 /* 2228 * Force synchronization for all dirty disks except one 2229 * with the biggest priority. 2230 */ 2231 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2232 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2233 "master disk for synchronization.", 2234 g_mirror_get_diskname(pdisk), sc->sc_name); 2235 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2236 if (disk->d_sync.ds_syncid != syncid) 2237 continue; 2238 if ((disk->d_flags & 2239 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2240 continue; 2241 } 2242 KASSERT((disk->d_flags & 2243 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2244 ("Disk %s isn't marked as dirty.", 2245 g_mirror_get_diskname(disk))); 2246 /* Skip the disk with the biggest priority. */ 2247 if (disk == pdisk) 2248 continue; 2249 disk->d_sync.ds_syncid = 0; 2250 } 2251 } else if (dirty < ndisks) { 2252 /* 2253 * Force synchronization for all dirty disks. 2254 * We have some non-dirty disks. 2255 */ 2256 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2257 if (disk->d_sync.ds_syncid != syncid) 2258 continue; 2259 if ((disk->d_flags & 2260 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2261 continue; 2262 } 2263 if ((disk->d_flags & 2264 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2265 continue; 2266 } 2267 disk->d_sync.ds_syncid = 0; 2268 } 2269 } 2270 2271 /* Reset hint. */ 2272 sc->sc_hint = NULL; 2273 sc->sc_syncid = syncid; 2274 if (force) { 2275 /* Remember to bump syncid on first write. */ 2276 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2277 } 2278 state = G_MIRROR_DEVICE_STATE_RUNNING; 2279 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2280 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2281 g_mirror_device_state2str(state)); 2282 sc->sc_state = state; 2283 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2284 state = g_mirror_determine_state(disk); 2285 g_mirror_event_send(disk, state, 2286 G_MIRROR_EVENT_DONTWAIT); 2287 if (state == G_MIRROR_DISK_STATE_STALE) 2288 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2289 } 2290 break; 2291 } 2292 case G_MIRROR_DEVICE_STATE_RUNNING: 2293 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2294 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2295 /* 2296 * No active disks or no disks at all, 2297 * so destroy device. 2298 */ 2299 if (sc->sc_provider != NULL) 2300 g_mirror_destroy_provider(sc); 2301 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2302 break; 2303 } else if (g_mirror_ndisks(sc, 2304 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2305 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2306 /* 2307 * We have active disks, launch provider if it doesn't 2308 * exist. 2309 */ 2310 if (sc->sc_provider == NULL) 2311 g_mirror_launch_provider(sc); 2312 if (sc->sc_rootmount != NULL) { 2313 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2314 __LINE__, sc->sc_rootmount); 2315 root_mount_rel(sc->sc_rootmount); 2316 sc->sc_rootmount = NULL; 2317 } 2318 } 2319 /* 2320 * Genid should be bumped immediately, so do it here. 2321 */ 2322 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2323 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2324 g_mirror_bump_genid(sc); 2325 } 2326 break; 2327 default: 2328 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2329 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2330 break; 2331 } 2332} 2333 2334/* 2335 * Update disk state and device state if needed. 2336 */ 2337#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2338 "Disk %s state changed from %s to %s (device %s).", \ 2339 g_mirror_get_diskname(disk), \ 2340 g_mirror_disk_state2str(disk->d_state), \ 2341 g_mirror_disk_state2str(state), sc->sc_name) 2342static int 2343g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2344{ 2345 struct g_mirror_softc *sc; 2346 2347 sc = disk->d_softc; 2348 sx_assert(&sc->sc_lock, SX_XLOCKED); 2349 2350again: 2351 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2352 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2353 g_mirror_disk_state2str(state)); 2354 switch (state) { 2355 case G_MIRROR_DISK_STATE_NEW: 2356 /* 2357 * Possible scenarios: 2358 * 1. New disk arrive. 2359 */ 2360 /* Previous state should be NONE. */ 2361 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2362 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2363 g_mirror_disk_state2str(disk->d_state))); 2364 DISK_STATE_CHANGED(); 2365 2366 disk->d_state = state; 2367 if (LIST_EMPTY(&sc->sc_disks)) 2368 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2369 else { 2370 struct g_mirror_disk *dp; 2371 2372 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2373 if (disk->d_priority >= dp->d_priority) { 2374 LIST_INSERT_BEFORE(dp, disk, d_next); 2375 dp = NULL; 2376 break; 2377 } 2378 if (LIST_NEXT(dp, d_next) == NULL) 2379 break; 2380 } 2381 if (dp != NULL) 2382 LIST_INSERT_AFTER(dp, disk, d_next); 2383 } 2384 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2385 sc->sc_name, g_mirror_get_diskname(disk)); 2386 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2387 break; 2388 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2389 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2390 g_mirror_device_state2str(sc->sc_state), 2391 g_mirror_get_diskname(disk), 2392 g_mirror_disk_state2str(disk->d_state))); 2393 state = g_mirror_determine_state(disk); 2394 if (state != G_MIRROR_DISK_STATE_NONE) 2395 goto again; 2396 break; 2397 case G_MIRROR_DISK_STATE_ACTIVE: 2398 /* 2399 * Possible scenarios: 2400 * 1. New disk does not need synchronization. 2401 * 2. Synchronization process finished successfully. 2402 */ 2403 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2404 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2405 g_mirror_device_state2str(sc->sc_state), 2406 g_mirror_get_diskname(disk), 2407 g_mirror_disk_state2str(disk->d_state))); 2408 /* Previous state should be NEW or SYNCHRONIZING. */ 2409 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2410 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2411 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2412 g_mirror_disk_state2str(disk->d_state))); 2413 DISK_STATE_CHANGED(); 2414 2415 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2416 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2417 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2418 g_mirror_sync_stop(disk, 0); 2419 } 2420 disk->d_state = state; 2421 disk->d_sync.ds_offset = 0; 2422 disk->d_sync.ds_offset_done = 0; 2423 g_mirror_update_idle(sc, disk); 2424 g_mirror_update_metadata(disk); 2425 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2426 sc->sc_name, g_mirror_get_diskname(disk)); 2427 break; 2428 case G_MIRROR_DISK_STATE_STALE: 2429 /* 2430 * Possible scenarios: 2431 * 1. Stale disk was connected. 2432 */ 2433 /* Previous state should be NEW. */ 2434 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2435 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2436 g_mirror_disk_state2str(disk->d_state))); 2437 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2438 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2439 g_mirror_device_state2str(sc->sc_state), 2440 g_mirror_get_diskname(disk), 2441 g_mirror_disk_state2str(disk->d_state))); 2442 /* 2443 * STALE state is only possible if device is marked 2444 * NOAUTOSYNC. 2445 */ 2446 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2447 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2448 g_mirror_device_state2str(sc->sc_state), 2449 g_mirror_get_diskname(disk), 2450 g_mirror_disk_state2str(disk->d_state))); 2451 DISK_STATE_CHANGED(); 2452 2453 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2454 disk->d_state = state; 2455 g_mirror_update_metadata(disk); 2456 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2457 sc->sc_name, g_mirror_get_diskname(disk)); 2458 break; 2459 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2460 /* 2461 * Possible scenarios: 2462 * 1. Disk which needs synchronization was connected. 2463 */ 2464 /* Previous state should be NEW. */ 2465 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2466 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2467 g_mirror_disk_state2str(disk->d_state))); 2468 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2469 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2470 g_mirror_device_state2str(sc->sc_state), 2471 g_mirror_get_diskname(disk), 2472 g_mirror_disk_state2str(disk->d_state))); 2473 DISK_STATE_CHANGED(); 2474 2475 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2476 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2477 disk->d_state = state; 2478 if (sc->sc_provider != NULL) { 2479 g_mirror_sync_start(disk); 2480 g_mirror_update_metadata(disk); 2481 } 2482 break; 2483 case G_MIRROR_DISK_STATE_DISCONNECTED: 2484 /* 2485 * Possible scenarios: 2486 * 1. Device wasn't running yet, but disk disappear. 2487 * 2. Disk was active and disapppear. 2488 * 3. Disk disappear during synchronization process. 2489 */ 2490 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2491 /* 2492 * Previous state should be ACTIVE, STALE or 2493 * SYNCHRONIZING. 2494 */ 2495 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2496 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2497 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2498 ("Wrong disk state (%s, %s).", 2499 g_mirror_get_diskname(disk), 2500 g_mirror_disk_state2str(disk->d_state))); 2501 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2502 /* Previous state should be NEW. */ 2503 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2504 ("Wrong disk state (%s, %s).", 2505 g_mirror_get_diskname(disk), 2506 g_mirror_disk_state2str(disk->d_state))); 2507 /* 2508 * Reset bumping syncid if disk disappeared in STARTING 2509 * state. 2510 */ 2511 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2512 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2513#ifdef INVARIANTS 2514 } else { 2515 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2516 sc->sc_name, 2517 g_mirror_device_state2str(sc->sc_state), 2518 g_mirror_get_diskname(disk), 2519 g_mirror_disk_state2str(disk->d_state))); 2520#endif 2521 } 2522 DISK_STATE_CHANGED(); 2523 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2524 sc->sc_name, g_mirror_get_diskname(disk)); 2525 2526 g_mirror_destroy_disk(disk); 2527 break; 2528 case G_MIRROR_DISK_STATE_DESTROY: 2529 { 2530 int error; 2531 2532 error = g_mirror_clear_metadata(disk); 2533 if (error != 0) 2534 return (error); 2535 DISK_STATE_CHANGED(); 2536 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2537 sc->sc_name, g_mirror_get_diskname(disk)); 2538 2539 g_mirror_destroy_disk(disk); 2540 sc->sc_ndisks--; 2541 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2542 g_mirror_update_metadata(disk); 2543 } 2544 break; 2545 } 2546 default: 2547 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2548 break; 2549 } 2550 return (0); 2551} 2552#undef DISK_STATE_CHANGED 2553 2554int 2555g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2556{ 2557 struct g_provider *pp; 2558 u_char *buf; 2559 int error; 2560 2561 g_topology_assert(); 2562 2563 error = g_access(cp, 1, 0, 0); 2564 if (error != 0) 2565 return (error); 2566 pp = cp->provider; 2567 g_topology_unlock(); 2568 /* Metadata are stored on last sector. */ 2569 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2570 &error); 2571 g_topology_lock(); 2572 g_access(cp, -1, 0, 0); 2573 if (buf == NULL) { 2574 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2575 cp->provider->name, error); 2576 return (error); 2577 } 2578 2579 /* Decode metadata. */ 2580 error = mirror_metadata_decode(buf, md); 2581 g_free(buf); 2582 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2583 return (EINVAL); 2584 if (md->md_version > G_MIRROR_VERSION) { 2585 G_MIRROR_DEBUG(0, 2586 "Kernel module is too old to handle metadata from %s.", 2587 cp->provider->name); 2588 return (EINVAL); 2589 } 2590 if (error != 0) { 2591 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2592 cp->provider->name); 2593 return (error); 2594 } 2595 2596 return (0); 2597} 2598 2599static int 2600g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2601 struct g_mirror_metadata *md) 2602{ 2603 2604 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2605 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2606 pp->name, md->md_did); 2607 return (EEXIST); 2608 } 2609 if (md->md_all != sc->sc_ndisks) { 2610 G_MIRROR_DEBUG(1, 2611 "Invalid '%s' field on disk %s (device %s), skipping.", 2612 "md_all", pp->name, sc->sc_name); 2613 return (EINVAL); 2614 } 2615 if (md->md_slice != sc->sc_slice) { 2616 G_MIRROR_DEBUG(1, 2617 "Invalid '%s' field on disk %s (device %s), skipping.", 2618 "md_slice", pp->name, sc->sc_name); 2619 return (EINVAL); 2620 } 2621 if (md->md_balance != sc->sc_balance) { 2622 G_MIRROR_DEBUG(1, 2623 "Invalid '%s' field on disk %s (device %s), skipping.", 2624 "md_balance", pp->name, sc->sc_name); 2625 return (EINVAL); 2626 } 2627 if (md->md_mediasize != sc->sc_mediasize) { 2628 G_MIRROR_DEBUG(1, 2629 "Invalid '%s' field on disk %s (device %s), skipping.", 2630 "md_mediasize", pp->name, sc->sc_name); 2631 return (EINVAL); 2632 } 2633 if (sc->sc_mediasize > pp->mediasize) { 2634 G_MIRROR_DEBUG(1, 2635 "Invalid size of disk %s (device %s), skipping.", pp->name, 2636 sc->sc_name); 2637 return (EINVAL); 2638 } 2639 if (md->md_sectorsize != sc->sc_sectorsize) { 2640 G_MIRROR_DEBUG(1, 2641 "Invalid '%s' field on disk %s (device %s), skipping.", 2642 "md_sectorsize", pp->name, sc->sc_name); 2643 return (EINVAL); 2644 } 2645 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2646 G_MIRROR_DEBUG(1, 2647 "Invalid sector size of disk %s (device %s), skipping.", 2648 pp->name, sc->sc_name); 2649 return (EINVAL); 2650 } 2651 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2652 G_MIRROR_DEBUG(1, 2653 "Invalid device flags on disk %s (device %s), skipping.", 2654 pp->name, sc->sc_name); 2655 return (EINVAL); 2656 } 2657 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2658 G_MIRROR_DEBUG(1, 2659 "Invalid disk flags on disk %s (device %s), skipping.", 2660 pp->name, sc->sc_name); 2661 return (EINVAL); 2662 } 2663 return (0); 2664} 2665 2666int 2667g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2668 struct g_mirror_metadata *md) 2669{ 2670 struct g_mirror_disk *disk; 2671 int error; 2672 2673 g_topology_assert_not(); 2674 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2675 2676 error = g_mirror_check_metadata(sc, pp, md); 2677 if (error != 0) 2678 return (error); 2679 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2680 md->md_genid < sc->sc_genid) { 2681 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2682 pp->name, sc->sc_name); 2683 return (EINVAL); 2684 } 2685 disk = g_mirror_init_disk(sc, pp, md, &error); 2686 if (disk == NULL) 2687 return (error); 2688 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2689 G_MIRROR_EVENT_WAIT); 2690 if (error != 0) 2691 return (error); 2692 if (md->md_version < G_MIRROR_VERSION) { 2693 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2694 pp->name, md->md_version, G_MIRROR_VERSION); 2695 g_mirror_update_metadata(disk); 2696 } 2697 return (0); 2698} 2699 2700static int 2701g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2702{ 2703 struct g_mirror_softc *sc; 2704 int dcr, dcw, dce; 2705 2706 g_topology_assert(); 2707 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2708 acw, ace); 2709 2710 dcr = pp->acr + acr; 2711 dcw = pp->acw + acw; 2712 dce = pp->ace + ace; 2713 2714 sc = pp->geom->softc; 2715 if (sc == NULL || LIST_EMPTY(&sc->sc_disks) || 2716 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 2717 if (acr <= 0 && acw <= 0 && ace <= 0) 2718 return (0); 2719 else 2720 return (ENXIO); 2721 } 2722 if (dcw == 0 && !sc->sc_idle) { 2723 g_topology_unlock(); 2724 sx_xlock(&sc->sc_lock); 2725 g_mirror_idle(sc, dcw); 2726 sx_xunlock(&sc->sc_lock); 2727 g_topology_lock(); 2728 } 2729 return (0); 2730} 2731 2732static struct g_geom * 2733g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2734{ 2735 struct g_mirror_softc *sc; 2736 struct g_geom *gp; 2737 int error, timeout; 2738 2739 g_topology_assert(); 2740 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2741 md->md_mid); 2742 2743 /* One disk is minimum. */ 2744 if (md->md_all < 1) 2745 return (NULL); 2746 /* 2747 * Action geom. 2748 */ 2749 gp = g_new_geomf(mp, "%s", md->md_name); 2750 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2751 gp->start = g_mirror_start; 2752 gp->orphan = g_mirror_orphan; 2753 gp->access = g_mirror_access; 2754 gp->dumpconf = g_mirror_dumpconf; 2755 2756 sc->sc_id = md->md_mid; 2757 sc->sc_slice = md->md_slice; 2758 sc->sc_balance = md->md_balance; 2759 sc->sc_mediasize = md->md_mediasize; 2760 sc->sc_sectorsize = md->md_sectorsize; 2761 sc->sc_ndisks = md->md_all; 2762 sc->sc_flags = md->md_mflags; 2763 sc->sc_bump_id = 0; 2764 sc->sc_idle = 1; 2765 sc->sc_last_write = time_uptime; 2766 sc->sc_writes = 0; 2767 sx_init(&sc->sc_lock, "gmirror:lock"); 2768 bioq_init(&sc->sc_queue); 2769 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2770 bioq_init(&sc->sc_regular_delayed); 2771 bioq_init(&sc->sc_inflight); 2772 bioq_init(&sc->sc_sync_delayed); 2773 LIST_INIT(&sc->sc_disks); 2774 TAILQ_INIT(&sc->sc_events); 2775 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2776 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2777 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2778 gp->softc = sc; 2779 sc->sc_geom = gp; 2780 sc->sc_provider = NULL; 2781 /* 2782 * Synchronization geom. 2783 */ 2784 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2785 gp->softc = sc; 2786 gp->orphan = g_mirror_orphan; 2787 sc->sc_sync.ds_geom = gp; 2788 sc->sc_sync.ds_ndisks = 0; 2789 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2790 "g_mirror %s", md->md_name); 2791 if (error != 0) { 2792 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2793 sc->sc_name); 2794 g_destroy_geom(sc->sc_sync.ds_geom); 2795 mtx_destroy(&sc->sc_events_mtx); 2796 mtx_destroy(&sc->sc_queue_mtx); 2797 sx_destroy(&sc->sc_lock); 2798 g_destroy_geom(sc->sc_geom); 2799 free(sc, M_MIRROR); 2800 return (NULL); 2801 } 2802 2803 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2804 2805 sc->sc_rootmount = root_mount_hold("GMIRROR"); 2806 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 2807 /* 2808 * Run timeout. 2809 */ 2810 timeout = g_mirror_timeout * hz; 2811 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2812 return (sc->sc_geom); 2813} 2814 2815int 2816g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2817{ 2818 struct g_provider *pp; 2819 2820 g_topology_assert_not(); 2821 if (sc == NULL) 2822 return (ENXIO); 2823 sx_assert(&sc->sc_lock, SX_XLOCKED); 2824 2825 pp = sc->sc_provider; 2826 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2827 if (force) { 2828 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 2829 "can't be definitely removed.", pp->name); 2830 } else { 2831 G_MIRROR_DEBUG(1, 2832 "Device %s is still open (r%dw%de%d).", pp->name, 2833 pp->acr, pp->acw, pp->ace); 2834 return (EBUSY); 2835 } 2836 } 2837 2838 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2839 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2840 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2841 sx_xunlock(&sc->sc_lock); 2842 mtx_lock(&sc->sc_queue_mtx); 2843 wakeup(sc); 2844 mtx_unlock(&sc->sc_queue_mtx); 2845 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2846 while (sc->sc_worker != NULL) 2847 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2848 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2849 sx_xlock(&sc->sc_lock); 2850 g_mirror_destroy_device(sc); 2851 free(sc, M_MIRROR); 2852 return (0); 2853} 2854 2855static void 2856g_mirror_taste_orphan(struct g_consumer *cp) 2857{ 2858 2859 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2860 cp->provider->name)); 2861} 2862 2863static struct g_geom * 2864g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2865{ 2866 struct g_mirror_metadata md; 2867 struct g_mirror_softc *sc; 2868 struct g_consumer *cp; 2869 struct g_geom *gp; 2870 int error; 2871 2872 g_topology_assert(); 2873 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2874 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2875 2876 gp = g_new_geomf(mp, "mirror:taste"); 2877 /* 2878 * This orphan function should be never called. 2879 */ 2880 gp->orphan = g_mirror_taste_orphan; 2881 cp = g_new_consumer(gp); 2882 g_attach(cp, pp); 2883 error = g_mirror_read_metadata(cp, &md); 2884 g_detach(cp); 2885 g_destroy_consumer(cp); 2886 g_destroy_geom(gp); 2887 if (error != 0) 2888 return (NULL); 2889 gp = NULL; 2890 2891 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2892 return (NULL); 2893 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 2894 return (NULL); 2895 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2896 G_MIRROR_DEBUG(0, 2897 "Device %s: provider %s marked as inactive, skipping.", 2898 md.md_name, pp->name); 2899 return (NULL); 2900 } 2901 if (g_mirror_debug >= 2) 2902 mirror_metadata_dump(&md); 2903 2904 /* 2905 * Let's check if device already exists. 2906 */ 2907 sc = NULL; 2908 LIST_FOREACH(gp, &mp->geom, geom) { 2909 sc = gp->softc; 2910 if (sc == NULL) 2911 continue; 2912 if (sc->sc_sync.ds_geom == gp) 2913 continue; 2914 if (strcmp(md.md_name, sc->sc_name) != 0) 2915 continue; 2916 if (md.md_mid != sc->sc_id) { 2917 G_MIRROR_DEBUG(0, "Device %s already configured.", 2918 sc->sc_name); 2919 return (NULL); 2920 } 2921 break; 2922 } 2923 if (gp == NULL) { 2924 gp = g_mirror_create(mp, &md); 2925 if (gp == NULL) { 2926 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2927 md.md_name); 2928 return (NULL); 2929 } 2930 sc = gp->softc; 2931 } 2932 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2933 g_topology_unlock(); 2934 sx_xlock(&sc->sc_lock); 2935 error = g_mirror_add_disk(sc, pp, &md); 2936 if (error != 0) { 2937 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2938 pp->name, gp->name, error); 2939 if (LIST_EMPTY(&sc->sc_disks)) { 2940 g_mirror_destroy(sc, 1); 2941 g_topology_lock(); 2942 return (NULL); 2943 } 2944 gp = NULL; 2945 } 2946 sx_xunlock(&sc->sc_lock); 2947 g_topology_lock(); 2948 return (gp); 2949} 2950 2951static int 2952g_mirror_destroy_geom(struct gctl_req *req __unused, 2953 struct g_class *mp __unused, struct g_geom *gp) 2954{ 2955 struct g_mirror_softc *sc; 2956 int error; 2957 2958 g_topology_unlock(); 2959 sc = gp->softc; 2960 sx_xlock(&sc->sc_lock); 2961 error = g_mirror_destroy(gp->softc, 0); 2962 if (error != 0) 2963 sx_xunlock(&sc->sc_lock); 2964 g_topology_lock(); 2965 return (error); 2966} 2967 2968static void 2969g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2970 struct g_consumer *cp, struct g_provider *pp) 2971{ 2972 struct g_mirror_softc *sc; 2973 2974 g_topology_assert(); 2975 2976 sc = gp->softc; 2977 if (sc == NULL) 2978 return; 2979 /* Skip synchronization geom. */ 2980 if (gp == sc->sc_sync.ds_geom) 2981 return; 2982 if (pp != NULL) { 2983 /* Nothing here. */ 2984 } else if (cp != NULL) { 2985 struct g_mirror_disk *disk; 2986 2987 disk = cp->private; 2988 if (disk == NULL) 2989 return; 2990 g_topology_unlock(); 2991 sx_xlock(&sc->sc_lock); 2992 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2993 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2994 sbuf_printf(sb, "%s<Synchronized>", indent); 2995 if (disk->d_sync.ds_offset == 0) 2996 sbuf_printf(sb, "0%%"); 2997 else { 2998 sbuf_printf(sb, "%u%%", 2999 (u_int)((disk->d_sync.ds_offset * 100) / 3000 sc->sc_provider->mediasize)); 3001 } 3002 sbuf_printf(sb, "</Synchronized>\n"); 3003 } 3004 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3005 disk->d_sync.ds_syncid); 3006 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3007 disk->d_genid); 3008 sbuf_printf(sb, "%s<Flags>", indent); 3009 if (disk->d_flags == 0) 3010 sbuf_printf(sb, "NONE"); 3011 else { 3012 int first = 1; 3013 3014#define ADD_FLAG(flag, name) do { \ 3015 if ((disk->d_flags & (flag)) != 0) { \ 3016 if (!first) \ 3017 sbuf_printf(sb, ", "); \ 3018 else \ 3019 first = 0; \ 3020 sbuf_printf(sb, name); \ 3021 } \ 3022} while (0) 3023 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3024 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3025 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3026 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3027 "SYNCHRONIZING"); 3028 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3029 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3030#undef ADD_FLAG 3031 } 3032 sbuf_printf(sb, "</Flags>\n"); 3033 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3034 disk->d_priority); 3035 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3036 g_mirror_disk_state2str(disk->d_state)); 3037 sx_xunlock(&sc->sc_lock); 3038 g_topology_lock(); 3039 } else { 3040 g_topology_unlock(); 3041 sx_xlock(&sc->sc_lock); 3042 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3043 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3044 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3045 sbuf_printf(sb, "%s<Flags>", indent); 3046 if (sc->sc_flags == 0) 3047 sbuf_printf(sb, "NONE"); 3048 else { 3049 int first = 1; 3050 3051#define ADD_FLAG(flag, name) do { \ 3052 if ((sc->sc_flags & (flag)) != 0) { \ 3053 if (!first) \ 3054 sbuf_printf(sb, ", "); \ 3055 else \ 3056 first = 0; \ 3057 sbuf_printf(sb, name); \ 3058 } \ 3059} while (0) 3060 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3061#undef ADD_FLAG 3062 } 3063 sbuf_printf(sb, "</Flags>\n"); 3064 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3065 (u_int)sc->sc_slice); 3066 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3067 balance_name(sc->sc_balance)); 3068 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3069 sc->sc_ndisks); 3070 sbuf_printf(sb, "%s<State>", indent); 3071 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3072 sbuf_printf(sb, "%s", "STARTING"); 3073 else if (sc->sc_ndisks == 3074 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3075 sbuf_printf(sb, "%s", "COMPLETE"); 3076 else 3077 sbuf_printf(sb, "%s", "DEGRADED"); 3078 sbuf_printf(sb, "</State>\n"); 3079 sx_xunlock(&sc->sc_lock); 3080 g_topology_lock(); 3081 } 3082} 3083 3084static void 3085g_mirror_shutdown_pre_sync(void *arg, int howto) 3086{ 3087 struct g_class *mp; 3088 struct g_geom *gp, *gp2; 3089 struct g_mirror_softc *sc; 3090 struct g_mirror_disk *disk; 3091 3092 mp = arg; 3093 DROP_GIANT(); 3094 g_topology_lock(); 3095 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3096 if ((sc = gp->softc) == NULL) 3097 continue; 3098 g_topology_unlock(); 3099 sx_xlock(&sc->sc_lock); 3100 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 3101 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 3102 g_mirror_sync_stop(disk, 1); 3103 } 3104 sx_xunlock(&sc->sc_lock); 3105 g_topology_lock(); 3106 } 3107 g_topology_unlock(); 3108 PICKUP_GIANT(); 3109} 3110 3111static void 3112g_mirror_shutdown_post_sync(void *arg, int howto) 3113{ 3114 struct g_class *mp; 3115 struct g_geom *gp, *gp2; 3116 struct g_mirror_softc *sc; 3117 3118 mp = arg; 3119 DROP_GIANT(); 3120 g_topology_lock(); 3121 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3122 if ((sc = gp->softc) == NULL) 3123 continue; 3124 g_topology_unlock(); 3125 sx_xlock(&sc->sc_lock); 3126 g_mirror_destroy(sc, 1); 3127 g_topology_lock(); 3128 } 3129 g_topology_unlock(); 3130 PICKUP_GIANT(); 3131#if 0 3132 tsleep(&gp, PRIBIO, "m:shutdown", hz * 20); 3133#endif 3134} 3135 3136static void 3137g_mirror_init(struct g_class *mp) 3138{ 3139 3140 g_mirror_pre_sync = EVENTHANDLER_REGISTER(shutdown_pre_sync, 3141 g_mirror_shutdown_pre_sync, mp, SHUTDOWN_PRI_FIRST); 3142 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3143 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3144 if (g_mirror_pre_sync == NULL || g_mirror_post_sync == NULL) 3145 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3146} 3147 3148static void 3149g_mirror_fini(struct g_class *mp) 3150{ 3151 3152 if (g_mirror_pre_sync != NULL) 3153 EVENTHANDLER_DEREGISTER(shutdown_pre_sync, g_mirror_pre_sync); 3154 if (g_mirror_post_sync != NULL) 3155 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); 3156} 3157 3158DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3159