g_mirror.c revision 328334
1/*- 2 * Copyright (c) 2004-2006 Pawel Jakub Dawidek <pjd@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27#include <sys/cdefs.h> 28__FBSDID("$FreeBSD: stable/11/sys/geom/mirror/g_mirror.c 328334 2018-01-24 15:16:17Z markj $"); 29 30#include <sys/param.h> 31#include <sys/systm.h> 32#include <sys/bio.h> 33#include <sys/eventhandler.h> 34#include <sys/fail.h> 35#include <sys/kernel.h> 36#include <sys/kthread.h> 37#include <sys/limits.h> 38#include <sys/lock.h> 39#include <sys/malloc.h> 40#include <sys/mutex.h> 41#include <sys/proc.h> 42#include <sys/sbuf.h> 43#include <sys/sched.h> 44#include <sys/sx.h> 45#include <sys/sysctl.h> 46 47#include <geom/geom.h> 48#include <geom/mirror/g_mirror.h> 49 50FEATURE(geom_mirror, "GEOM mirroring support"); 51 52static MALLOC_DEFINE(M_MIRROR, "mirror_data", "GEOM_MIRROR Data"); 53 54SYSCTL_DECL(_kern_geom); 55static SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, 56 "GEOM_MIRROR stuff"); 57int g_mirror_debug = 0; 58SYSCTL_INT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RWTUN, &g_mirror_debug, 0, 59 "Debug level"); 60static u_int g_mirror_timeout = 4; 61SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RWTUN, &g_mirror_timeout, 62 0, "Time to wait on all mirror components"); 63static u_int g_mirror_idletime = 5; 64SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RWTUN, 65 &g_mirror_idletime, 0, "Mark components as clean when idling"); 66static u_int g_mirror_disconnect_on_failure = 1; 67SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN, 68 &g_mirror_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 69static u_int g_mirror_syncreqs = 2; 70SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_requests, CTLFLAG_RDTUN, 71 &g_mirror_syncreqs, 0, "Parallel synchronization I/O requests."); 72static u_int g_mirror_sync_period = 5; 73SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_update_period, CTLFLAG_RWTUN, 74 &g_mirror_sync_period, 0, 75 "Metadata update period during synchronization, in seconds"); 76 77#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 78 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 79 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 80 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 81} while (0) 82 83static eventhandler_tag g_mirror_post_sync = NULL; 84static int g_mirror_shutdown = 0; 85 86static g_ctl_destroy_geom_t g_mirror_destroy_geom; 87static g_taste_t g_mirror_taste; 88static g_init_t g_mirror_init; 89static g_fini_t g_mirror_fini; 90static g_provgone_t g_mirror_providergone; 91static g_resize_t g_mirror_resize; 92 93struct g_class g_mirror_class = { 94 .name = G_MIRROR_CLASS_NAME, 95 .version = G_VERSION, 96 .ctlreq = g_mirror_config, 97 .taste = g_mirror_taste, 98 .destroy_geom = g_mirror_destroy_geom, 99 .init = g_mirror_init, 100 .fini = g_mirror_fini, 101 .providergone = g_mirror_providergone, 102 .resize = g_mirror_resize 103}; 104 105 106static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 107static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 108static void g_mirror_update_device(struct g_mirror_softc *sc, bool force); 109static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 110 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 111static void g_mirror_sync_reinit(const struct g_mirror_disk *disk, 112 struct bio *bp, off_t offset); 113static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 114static void g_mirror_register_request(struct g_mirror_softc *sc, 115 struct bio *bp); 116static void g_mirror_sync_release(struct g_mirror_softc *sc); 117 118 119static const char * 120g_mirror_disk_state2str(int state) 121{ 122 123 switch (state) { 124 case G_MIRROR_DISK_STATE_NONE: 125 return ("NONE"); 126 case G_MIRROR_DISK_STATE_NEW: 127 return ("NEW"); 128 case G_MIRROR_DISK_STATE_ACTIVE: 129 return ("ACTIVE"); 130 case G_MIRROR_DISK_STATE_STALE: 131 return ("STALE"); 132 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 133 return ("SYNCHRONIZING"); 134 case G_MIRROR_DISK_STATE_DISCONNECTED: 135 return ("DISCONNECTED"); 136 case G_MIRROR_DISK_STATE_DESTROY: 137 return ("DESTROY"); 138 default: 139 return ("INVALID"); 140 } 141} 142 143static const char * 144g_mirror_device_state2str(int state) 145{ 146 147 switch (state) { 148 case G_MIRROR_DEVICE_STATE_STARTING: 149 return ("STARTING"); 150 case G_MIRROR_DEVICE_STATE_RUNNING: 151 return ("RUNNING"); 152 default: 153 return ("INVALID"); 154 } 155} 156 157static const char * 158g_mirror_get_diskname(struct g_mirror_disk *disk) 159{ 160 161 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 162 return ("[unknown]"); 163 return (disk->d_name); 164} 165 166/* 167 * --- Events handling functions --- 168 * Events in geom_mirror are used to maintain disks and device status 169 * from one thread to simplify locking. 170 */ 171static void 172g_mirror_event_free(struct g_mirror_event *ep) 173{ 174 175 free(ep, M_MIRROR); 176} 177 178int 179g_mirror_event_send(void *arg, int state, int flags) 180{ 181 struct g_mirror_softc *sc; 182 struct g_mirror_disk *disk; 183 struct g_mirror_event *ep; 184 int error; 185 186 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 187 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 188 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 189 disk = NULL; 190 sc = arg; 191 } else { 192 disk = arg; 193 sc = disk->d_softc; 194 } 195 ep->e_disk = disk; 196 ep->e_state = state; 197 ep->e_flags = flags; 198 ep->e_error = 0; 199 mtx_lock(&sc->sc_events_mtx); 200 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 201 mtx_unlock(&sc->sc_events_mtx); 202 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 203 mtx_lock(&sc->sc_queue_mtx); 204 wakeup(sc); 205 mtx_unlock(&sc->sc_queue_mtx); 206 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 207 return (0); 208 sx_assert(&sc->sc_lock, SX_XLOCKED); 209 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 210 sx_xunlock(&sc->sc_lock); 211 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 212 mtx_lock(&sc->sc_events_mtx); 213 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 214 hz * 5); 215 } 216 error = ep->e_error; 217 g_mirror_event_free(ep); 218 sx_xlock(&sc->sc_lock); 219 return (error); 220} 221 222static struct g_mirror_event * 223g_mirror_event_first(struct g_mirror_softc *sc) 224{ 225 struct g_mirror_event *ep; 226 227 mtx_lock(&sc->sc_events_mtx); 228 ep = TAILQ_FIRST(&sc->sc_events); 229 mtx_unlock(&sc->sc_events_mtx); 230 return (ep); 231} 232 233static void 234g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 235{ 236 237 mtx_lock(&sc->sc_events_mtx); 238 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 239 mtx_unlock(&sc->sc_events_mtx); 240} 241 242static void 243g_mirror_event_cancel(struct g_mirror_disk *disk) 244{ 245 struct g_mirror_softc *sc; 246 struct g_mirror_event *ep, *tmpep; 247 248 sc = disk->d_softc; 249 sx_assert(&sc->sc_lock, SX_XLOCKED); 250 251 mtx_lock(&sc->sc_events_mtx); 252 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 253 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 254 continue; 255 if (ep->e_disk != disk) 256 continue; 257 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 258 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 259 g_mirror_event_free(ep); 260 else { 261 ep->e_error = ECANCELED; 262 wakeup(ep); 263 } 264 } 265 mtx_unlock(&sc->sc_events_mtx); 266} 267 268/* 269 * Return the number of disks in given state. 270 * If state is equal to -1, count all connected disks. 271 */ 272u_int 273g_mirror_ndisks(struct g_mirror_softc *sc, int state) 274{ 275 struct g_mirror_disk *disk; 276 u_int n = 0; 277 278 sx_assert(&sc->sc_lock, SX_LOCKED); 279 280 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 281 if (state == -1 || disk->d_state == state) 282 n++; 283 } 284 return (n); 285} 286 287/* 288 * Find a disk in mirror by its disk ID. 289 */ 290static struct g_mirror_disk * 291g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 292{ 293 struct g_mirror_disk *disk; 294 295 sx_assert(&sc->sc_lock, SX_XLOCKED); 296 297 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 298 if (disk->d_id == id) 299 return (disk); 300 } 301 return (NULL); 302} 303 304static u_int 305g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 306{ 307 struct bio *bp; 308 u_int nreqs = 0; 309 310 mtx_lock(&sc->sc_queue_mtx); 311 TAILQ_FOREACH(bp, &sc->sc_queue, bio_queue) { 312 if (bp->bio_from == cp) 313 nreqs++; 314 } 315 mtx_unlock(&sc->sc_queue_mtx); 316 return (nreqs); 317} 318 319static int 320g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 321{ 322 323 if (cp->index > 0) { 324 G_MIRROR_DEBUG(2, 325 "I/O requests for %s exist, can't destroy it now.", 326 cp->provider->name); 327 return (1); 328 } 329 if (g_mirror_nrequests(sc, cp) > 0) { 330 G_MIRROR_DEBUG(2, 331 "I/O requests for %s in queue, can't destroy it now.", 332 cp->provider->name); 333 return (1); 334 } 335 return (0); 336} 337 338static void 339g_mirror_destroy_consumer(void *arg, int flags __unused) 340{ 341 struct g_consumer *cp; 342 343 g_topology_assert(); 344 345 cp = arg; 346 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 347 g_detach(cp); 348 g_destroy_consumer(cp); 349} 350 351static void 352g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 353{ 354 struct g_provider *pp; 355 int retaste_wait; 356 357 g_topology_assert(); 358 359 cp->private = NULL; 360 if (g_mirror_is_busy(sc, cp)) 361 return; 362 pp = cp->provider; 363 retaste_wait = 0; 364 if (cp->acw == 1) { 365 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 366 retaste_wait = 1; 367 } 368 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 369 -cp->acw, -cp->ace, 0); 370 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 371 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 372 if (retaste_wait) { 373 /* 374 * After retaste event was send (inside g_access()), we can send 375 * event to detach and destroy consumer. 376 * A class, which has consumer to the given provider connected 377 * will not receive retaste event for the provider. 378 * This is the way how I ignore retaste events when I close 379 * consumers opened for write: I detach and destroy consumer 380 * after retaste event is sent. 381 */ 382 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 383 return; 384 } 385 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 386 g_detach(cp); 387 g_destroy_consumer(cp); 388} 389 390static int 391g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 392{ 393 struct g_consumer *cp; 394 int error; 395 396 g_topology_assert_not(); 397 KASSERT(disk->d_consumer == NULL, 398 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 399 400 g_topology_lock(); 401 cp = g_new_consumer(disk->d_softc->sc_geom); 402 cp->flags |= G_CF_DIRECT_RECEIVE; 403 error = g_attach(cp, pp); 404 if (error != 0) { 405 g_destroy_consumer(cp); 406 g_topology_unlock(); 407 return (error); 408 } 409 error = g_access(cp, 1, 1, 1); 410 if (error != 0) { 411 g_detach(cp); 412 g_destroy_consumer(cp); 413 g_topology_unlock(); 414 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 415 pp->name, error); 416 return (error); 417 } 418 g_topology_unlock(); 419 disk->d_consumer = cp; 420 disk->d_consumer->private = disk; 421 disk->d_consumer->index = 0; 422 423 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 424 return (0); 425} 426 427static void 428g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 429{ 430 431 g_topology_assert(); 432 433 if (cp == NULL) 434 return; 435 if (cp->provider != NULL) 436 g_mirror_kill_consumer(sc, cp); 437 else 438 g_destroy_consumer(cp); 439} 440 441/* 442 * Initialize disk. This means allocate memory, create consumer, attach it 443 * to the provider and open access (r1w1e1) to it. 444 */ 445static struct g_mirror_disk * 446g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 447 struct g_mirror_metadata *md, int *errorp) 448{ 449 struct g_mirror_disk *disk; 450 int i, error; 451 452 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 453 if (disk == NULL) { 454 error = ENOMEM; 455 goto fail; 456 } 457 disk->d_softc = sc; 458 error = g_mirror_connect_disk(disk, pp); 459 if (error != 0) 460 goto fail; 461 disk->d_id = md->md_did; 462 disk->d_state = G_MIRROR_DISK_STATE_NONE; 463 disk->d_priority = md->md_priority; 464 disk->d_flags = md->md_dflags; 465 error = g_getattr("GEOM::candelete", disk->d_consumer, &i); 466 if (error == 0 && i != 0) 467 disk->d_flags |= G_MIRROR_DISK_FLAG_CANDELETE; 468 if (md->md_provider[0] != '\0') 469 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 470 disk->d_sync.ds_consumer = NULL; 471 disk->d_sync.ds_offset = md->md_sync_offset; 472 disk->d_sync.ds_offset_done = md->md_sync_offset; 473 disk->d_sync.ds_update_ts = time_uptime; 474 disk->d_genid = md->md_genid; 475 disk->d_sync.ds_syncid = md->md_syncid; 476 if (errorp != NULL) 477 *errorp = 0; 478 return (disk); 479fail: 480 if (errorp != NULL) 481 *errorp = error; 482 if (disk != NULL) 483 free(disk, M_MIRROR); 484 return (NULL); 485} 486 487static void 488g_mirror_destroy_disk(struct g_mirror_disk *disk) 489{ 490 struct g_mirror_softc *sc; 491 492 g_topology_assert_not(); 493 sc = disk->d_softc; 494 sx_assert(&sc->sc_lock, SX_XLOCKED); 495 496 LIST_REMOVE(disk, d_next); 497 g_mirror_event_cancel(disk); 498 if (sc->sc_hint == disk) 499 sc->sc_hint = NULL; 500 switch (disk->d_state) { 501 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 502 g_mirror_sync_stop(disk, 1); 503 /* FALLTHROUGH */ 504 case G_MIRROR_DISK_STATE_NEW: 505 case G_MIRROR_DISK_STATE_STALE: 506 case G_MIRROR_DISK_STATE_ACTIVE: 507 g_topology_lock(); 508 g_mirror_disconnect_consumer(sc, disk->d_consumer); 509 g_topology_unlock(); 510 free(disk, M_MIRROR); 511 break; 512 default: 513 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 514 g_mirror_get_diskname(disk), 515 g_mirror_disk_state2str(disk->d_state))); 516 } 517} 518 519static void 520g_mirror_free_device(struct g_mirror_softc *sc) 521{ 522 523 mtx_destroy(&sc->sc_queue_mtx); 524 mtx_destroy(&sc->sc_events_mtx); 525 mtx_destroy(&sc->sc_done_mtx); 526 sx_destroy(&sc->sc_lock); 527 free(sc, M_MIRROR); 528} 529 530static void 531g_mirror_providergone(struct g_provider *pp) 532{ 533 struct g_mirror_softc *sc = pp->private; 534 535 if ((--sc->sc_refcnt) == 0) 536 g_mirror_free_device(sc); 537} 538 539static void 540g_mirror_destroy_device(struct g_mirror_softc *sc) 541{ 542 struct g_mirror_disk *disk; 543 struct g_mirror_event *ep; 544 struct g_geom *gp; 545 struct g_consumer *cp, *tmpcp; 546 547 g_topology_assert_not(); 548 sx_assert(&sc->sc_lock, SX_XLOCKED); 549 550 gp = sc->sc_geom; 551 if (sc->sc_provider != NULL) 552 g_mirror_destroy_provider(sc); 553 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 554 disk = LIST_FIRST(&sc->sc_disks)) { 555 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 556 g_mirror_update_metadata(disk); 557 g_mirror_destroy_disk(disk); 558 } 559 while ((ep = g_mirror_event_first(sc)) != NULL) { 560 g_mirror_event_remove(sc, ep); 561 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 562 g_mirror_event_free(ep); 563 else { 564 ep->e_error = ECANCELED; 565 ep->e_flags |= G_MIRROR_EVENT_DONE; 566 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 567 mtx_lock(&sc->sc_events_mtx); 568 wakeup(ep); 569 mtx_unlock(&sc->sc_events_mtx); 570 } 571 } 572 callout_drain(&sc->sc_callout); 573 574 g_topology_lock(); 575 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 576 g_mirror_disconnect_consumer(sc, cp); 577 } 578 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 579 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 580 g_wither_geom(gp, ENXIO); 581 sx_xunlock(&sc->sc_lock); 582 if ((--sc->sc_refcnt) == 0) 583 g_mirror_free_device(sc); 584 g_topology_unlock(); 585} 586 587static void 588g_mirror_orphan(struct g_consumer *cp) 589{ 590 struct g_mirror_disk *disk; 591 592 g_topology_assert(); 593 594 disk = cp->private; 595 if (disk == NULL) 596 return; 597 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 598 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 599 G_MIRROR_EVENT_DONTWAIT); 600} 601 602/* 603 * Function should return the next active disk on the list. 604 * It is possible that it will be the same disk as given. 605 * If there are no active disks on list, NULL is returned. 606 */ 607static __inline struct g_mirror_disk * 608g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 609{ 610 struct g_mirror_disk *dp; 611 612 for (dp = LIST_NEXT(disk, d_next); dp != disk; 613 dp = LIST_NEXT(dp, d_next)) { 614 if (dp == NULL) 615 dp = LIST_FIRST(&sc->sc_disks); 616 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 617 break; 618 } 619 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 620 return (NULL); 621 return (dp); 622} 623 624static struct g_mirror_disk * 625g_mirror_get_disk(struct g_mirror_softc *sc) 626{ 627 struct g_mirror_disk *disk; 628 629 if (sc->sc_hint == NULL) { 630 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 631 if (sc->sc_hint == NULL) 632 return (NULL); 633 } 634 disk = sc->sc_hint; 635 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 636 disk = g_mirror_find_next(sc, disk); 637 if (disk == NULL) 638 return (NULL); 639 } 640 sc->sc_hint = g_mirror_find_next(sc, disk); 641 return (disk); 642} 643 644static int 645g_mirror_write_metadata(struct g_mirror_disk *disk, 646 struct g_mirror_metadata *md) 647{ 648 struct g_mirror_softc *sc; 649 struct g_consumer *cp; 650 off_t offset, length; 651 u_char *sector; 652 int error = 0; 653 654 g_topology_assert_not(); 655 sc = disk->d_softc; 656 sx_assert(&sc->sc_lock, SX_LOCKED); 657 658 cp = disk->d_consumer; 659 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 660 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 661 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 662 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 663 cp->acw, cp->ace)); 664 length = cp->provider->sectorsize; 665 offset = cp->provider->mediasize - length; 666 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 667 if (md != NULL && 668 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) { 669 /* 670 * Handle the case, when the size of parent provider reduced. 671 */ 672 if (offset < md->md_mediasize) 673 error = ENOSPC; 674 else 675 mirror_metadata_encode(md, sector); 676 } 677 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_metadata_write, error); 678 if (error == 0) 679 error = g_write_data(cp, offset, sector, length); 680 free(sector, M_MIRROR); 681 if (error != 0) { 682 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 683 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 684 G_MIRROR_DEBUG(0, "Cannot write metadata on %s " 685 "(device=%s, error=%d).", 686 g_mirror_get_diskname(disk), sc->sc_name, error); 687 } else { 688 G_MIRROR_DEBUG(1, "Cannot write metadata on %s " 689 "(device=%s, error=%d).", 690 g_mirror_get_diskname(disk), sc->sc_name, error); 691 } 692 if (g_mirror_disconnect_on_failure && 693 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 694 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 695 g_mirror_event_send(disk, 696 G_MIRROR_DISK_STATE_DISCONNECTED, 697 G_MIRROR_EVENT_DONTWAIT); 698 } 699 } 700 return (error); 701} 702 703static int 704g_mirror_clear_metadata(struct g_mirror_disk *disk) 705{ 706 int error; 707 708 g_topology_assert_not(); 709 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 710 711 if (disk->d_softc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 712 return (0); 713 error = g_mirror_write_metadata(disk, NULL); 714 if (error == 0) { 715 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 716 g_mirror_get_diskname(disk)); 717 } else { 718 G_MIRROR_DEBUG(0, 719 "Cannot clear metadata on disk %s (error=%d).", 720 g_mirror_get_diskname(disk), error); 721 } 722 return (error); 723} 724 725void 726g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 727 struct g_mirror_metadata *md) 728{ 729 730 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 731 md->md_version = G_MIRROR_VERSION; 732 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 733 md->md_mid = sc->sc_id; 734 md->md_all = sc->sc_ndisks; 735 md->md_slice = sc->sc_slice; 736 md->md_balance = sc->sc_balance; 737 md->md_genid = sc->sc_genid; 738 md->md_mediasize = sc->sc_mediasize; 739 md->md_sectorsize = sc->sc_sectorsize; 740 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 741 bzero(md->md_provider, sizeof(md->md_provider)); 742 if (disk == NULL) { 743 md->md_did = arc4random(); 744 md->md_priority = 0; 745 md->md_syncid = 0; 746 md->md_dflags = 0; 747 md->md_sync_offset = 0; 748 md->md_provsize = 0; 749 } else { 750 md->md_did = disk->d_id; 751 md->md_priority = disk->d_priority; 752 md->md_syncid = disk->d_sync.ds_syncid; 753 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 754 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 755 md->md_sync_offset = disk->d_sync.ds_offset_done; 756 else 757 md->md_sync_offset = 0; 758 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 759 strlcpy(md->md_provider, 760 disk->d_consumer->provider->name, 761 sizeof(md->md_provider)); 762 } 763 md->md_provsize = disk->d_consumer->provider->mediasize; 764 } 765} 766 767void 768g_mirror_update_metadata(struct g_mirror_disk *disk) 769{ 770 struct g_mirror_softc *sc; 771 struct g_mirror_metadata md; 772 int error; 773 774 g_topology_assert_not(); 775 sc = disk->d_softc; 776 sx_assert(&sc->sc_lock, SX_LOCKED); 777 778 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 779 return; 780 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) 781 g_mirror_fill_metadata(sc, disk, &md); 782 error = g_mirror_write_metadata(disk, &md); 783 if (error == 0) { 784 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 785 g_mirror_get_diskname(disk)); 786 } else { 787 G_MIRROR_DEBUG(0, 788 "Cannot update metadata on disk %s (error=%d).", 789 g_mirror_get_diskname(disk), error); 790 } 791} 792 793static void 794g_mirror_bump_syncid(struct g_mirror_softc *sc) 795{ 796 struct g_mirror_disk *disk; 797 798 g_topology_assert_not(); 799 sx_assert(&sc->sc_lock, SX_XLOCKED); 800 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 801 ("%s called with no active disks (device=%s).", __func__, 802 sc->sc_name)); 803 804 sc->sc_syncid++; 805 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 806 sc->sc_syncid); 807 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 808 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 809 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 810 disk->d_sync.ds_syncid = sc->sc_syncid; 811 g_mirror_update_metadata(disk); 812 } 813 } 814} 815 816static void 817g_mirror_bump_genid(struct g_mirror_softc *sc) 818{ 819 struct g_mirror_disk *disk; 820 821 g_topology_assert_not(); 822 sx_assert(&sc->sc_lock, SX_XLOCKED); 823 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 824 ("%s called with no active disks (device=%s).", __func__, 825 sc->sc_name)); 826 827 sc->sc_genid++; 828 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 829 sc->sc_genid); 830 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 831 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 832 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 833 disk->d_genid = sc->sc_genid; 834 g_mirror_update_metadata(disk); 835 } 836 } 837} 838 839static int 840g_mirror_idle(struct g_mirror_softc *sc, int acw) 841{ 842 struct g_mirror_disk *disk; 843 int timeout; 844 845 g_topology_assert_not(); 846 sx_assert(&sc->sc_lock, SX_XLOCKED); 847 848 if (sc->sc_provider == NULL) 849 return (0); 850 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 851 return (0); 852 if (sc->sc_idle) 853 return (0); 854 if (sc->sc_writes > 0) 855 return (0); 856 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 857 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); 858 if (!g_mirror_shutdown && timeout > 0) 859 return (timeout); 860 } 861 sc->sc_idle = 1; 862 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 863 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 864 continue; 865 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 866 g_mirror_get_diskname(disk), sc->sc_name); 867 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 868 g_mirror_update_metadata(disk); 869 } 870 return (0); 871} 872 873static void 874g_mirror_unidle(struct g_mirror_softc *sc) 875{ 876 struct g_mirror_disk *disk; 877 878 g_topology_assert_not(); 879 sx_assert(&sc->sc_lock, SX_XLOCKED); 880 881 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 882 return; 883 sc->sc_idle = 0; 884 sc->sc_last_write = time_uptime; 885 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 886 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 887 continue; 888 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 889 g_mirror_get_diskname(disk), sc->sc_name); 890 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 891 g_mirror_update_metadata(disk); 892 } 893} 894 895static void 896g_mirror_done(struct bio *bp) 897{ 898 struct g_mirror_softc *sc; 899 900 sc = bp->bio_from->geom->softc; 901 bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR; 902 mtx_lock(&sc->sc_queue_mtx); 903 TAILQ_INSERT_TAIL(&sc->sc_queue, bp, bio_queue); 904 mtx_unlock(&sc->sc_queue_mtx); 905 wakeup(sc); 906} 907 908static void 909g_mirror_regular_request_error(struct g_mirror_softc *sc, 910 struct g_mirror_disk *disk, struct bio *bp) 911{ 912 913 if (bp->bio_cmd == BIO_FLUSH && bp->bio_error == EOPNOTSUPP) 914 return; 915 916 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 917 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 918 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 919 bp->bio_error); 920 } else { 921 G_MIRROR_LOGREQ(1, bp, "Request failed (error=%d).", 922 bp->bio_error); 923 } 924 if (g_mirror_disconnect_on_failure && 925 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 926 if (bp->bio_error == ENXIO && 927 bp->bio_cmd == BIO_READ) 928 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 929 else if (bp->bio_error == ENXIO) 930 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_NOW; 931 else 932 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 933 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 934 G_MIRROR_EVENT_DONTWAIT); 935 } 936} 937 938static void 939g_mirror_regular_request(struct g_mirror_softc *sc, struct bio *bp) 940{ 941 struct g_mirror_disk *disk; 942 struct bio *pbp; 943 944 g_topology_assert_not(); 945 KASSERT(sc->sc_provider == bp->bio_parent->bio_to, 946 ("regular request %p with unexpected origin", bp)); 947 948 pbp = bp->bio_parent; 949 bp->bio_from->index--; 950 if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) 951 sc->sc_writes--; 952 disk = bp->bio_from->private; 953 if (disk == NULL) { 954 g_topology_lock(); 955 g_mirror_kill_consumer(sc, bp->bio_from); 956 g_topology_unlock(); 957 } 958 959 switch (bp->bio_cmd) { 960 case BIO_READ: 961 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_read, 962 bp->bio_error); 963 break; 964 case BIO_WRITE: 965 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_write, 966 bp->bio_error); 967 break; 968 case BIO_DELETE: 969 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_delete, 970 bp->bio_error); 971 break; 972 case BIO_FLUSH: 973 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_flush, 974 bp->bio_error); 975 break; 976 } 977 978 pbp->bio_inbed++; 979 KASSERT(pbp->bio_inbed <= pbp->bio_children, 980 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 981 pbp->bio_children)); 982 if (bp->bio_error == 0 && pbp->bio_error == 0) { 983 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 984 g_destroy_bio(bp); 985 if (pbp->bio_children == pbp->bio_inbed) { 986 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 987 pbp->bio_completed = pbp->bio_length; 988 if (pbp->bio_cmd == BIO_WRITE || 989 pbp->bio_cmd == BIO_DELETE) { 990 TAILQ_REMOVE(&sc->sc_inflight, pbp, bio_queue); 991 /* Release delayed sync requests if possible. */ 992 g_mirror_sync_release(sc); 993 } 994 g_io_deliver(pbp, pbp->bio_error); 995 } 996 return; 997 } else if (bp->bio_error != 0) { 998 if (pbp->bio_error == 0) 999 pbp->bio_error = bp->bio_error; 1000 if (disk != NULL) 1001 g_mirror_regular_request_error(sc, disk, bp); 1002 switch (pbp->bio_cmd) { 1003 case BIO_DELETE: 1004 case BIO_WRITE: 1005 case BIO_FLUSH: 1006 pbp->bio_inbed--; 1007 pbp->bio_children--; 1008 break; 1009 } 1010 } 1011 g_destroy_bio(bp); 1012 1013 switch (pbp->bio_cmd) { 1014 case BIO_READ: 1015 if (pbp->bio_inbed < pbp->bio_children) 1016 break; 1017 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 1018 g_io_deliver(pbp, pbp->bio_error); 1019 else { 1020 pbp->bio_error = 0; 1021 mtx_lock(&sc->sc_queue_mtx); 1022 TAILQ_INSERT_TAIL(&sc->sc_queue, pbp, bio_queue); 1023 mtx_unlock(&sc->sc_queue_mtx); 1024 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1025 wakeup(sc); 1026 } 1027 break; 1028 case BIO_DELETE: 1029 case BIO_WRITE: 1030 case BIO_FLUSH: 1031 if (pbp->bio_children == 0) { 1032 /* 1033 * All requests failed. 1034 */ 1035 } else if (pbp->bio_inbed < pbp->bio_children) { 1036 /* Do nothing. */ 1037 break; 1038 } else if (pbp->bio_children == pbp->bio_inbed) { 1039 /* Some requests succeeded. */ 1040 pbp->bio_error = 0; 1041 pbp->bio_completed = pbp->bio_length; 1042 } 1043 if (pbp->bio_cmd == BIO_WRITE || pbp->bio_cmd == BIO_DELETE) { 1044 TAILQ_REMOVE(&sc->sc_inflight, pbp, bio_queue); 1045 /* Release delayed sync requests if possible. */ 1046 g_mirror_sync_release(sc); 1047 } 1048 g_io_deliver(pbp, pbp->bio_error); 1049 break; 1050 default: 1051 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 1052 break; 1053 } 1054} 1055 1056static void 1057g_mirror_sync_done(struct bio *bp) 1058{ 1059 struct g_mirror_softc *sc; 1060 1061 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1062 sc = bp->bio_from->geom->softc; 1063 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; 1064 mtx_lock(&sc->sc_queue_mtx); 1065 TAILQ_INSERT_TAIL(&sc->sc_queue, bp, bio_queue); 1066 mtx_unlock(&sc->sc_queue_mtx); 1067 wakeup(sc); 1068} 1069 1070static void 1071g_mirror_candelete(struct bio *bp) 1072{ 1073 struct g_mirror_softc *sc; 1074 struct g_mirror_disk *disk; 1075 int *val; 1076 1077 sc = bp->bio_to->private; 1078 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1079 if (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) 1080 break; 1081 } 1082 val = (int *)bp->bio_data; 1083 *val = (disk != NULL); 1084 g_io_deliver(bp, 0); 1085} 1086 1087static void 1088g_mirror_kernel_dump(struct bio *bp) 1089{ 1090 struct g_mirror_softc *sc; 1091 struct g_mirror_disk *disk; 1092 struct bio *cbp; 1093 struct g_kerneldump *gkd; 1094 1095 /* 1096 * We configure dumping to the first component, because this component 1097 * will be used for reading with 'prefer' balance algorithm. 1098 * If the component with the highest priority is currently disconnected 1099 * we will not be able to read the dump after the reboot if it will be 1100 * connected and synchronized later. Can we do something better? 1101 */ 1102 sc = bp->bio_to->private; 1103 disk = LIST_FIRST(&sc->sc_disks); 1104 1105 gkd = (struct g_kerneldump *)bp->bio_data; 1106 if (gkd->length > bp->bio_to->mediasize) 1107 gkd->length = bp->bio_to->mediasize; 1108 cbp = g_clone_bio(bp); 1109 if (cbp == NULL) { 1110 g_io_deliver(bp, ENOMEM); 1111 return; 1112 } 1113 cbp->bio_done = g_std_done; 1114 g_io_request(cbp, disk->d_consumer); 1115 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1116 g_mirror_get_diskname(disk)); 1117} 1118 1119static void 1120g_mirror_start(struct bio *bp) 1121{ 1122 struct g_mirror_softc *sc; 1123 1124 sc = bp->bio_to->private; 1125 /* 1126 * If sc == NULL or there are no valid disks, provider's error 1127 * should be set and g_mirror_start() should not be called at all. 1128 */ 1129 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1130 ("Provider's error should be set (error=%d)(mirror=%s).", 1131 bp->bio_to->error, bp->bio_to->name)); 1132 G_MIRROR_LOGREQ(3, bp, "Request received."); 1133 1134 switch (bp->bio_cmd) { 1135 case BIO_READ: 1136 case BIO_WRITE: 1137 case BIO_DELETE: 1138 case BIO_FLUSH: 1139 break; 1140 case BIO_GETATTR: 1141 if (!strcmp(bp->bio_attribute, "GEOM::candelete")) { 1142 g_mirror_candelete(bp); 1143 return; 1144 } else if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1145 g_mirror_kernel_dump(bp); 1146 return; 1147 } 1148 /* FALLTHROUGH */ 1149 default: 1150 g_io_deliver(bp, EOPNOTSUPP); 1151 return; 1152 } 1153 mtx_lock(&sc->sc_queue_mtx); 1154 if (bp->bio_to->error != 0) { 1155 mtx_unlock(&sc->sc_queue_mtx); 1156 g_io_deliver(bp, bp->bio_to->error); 1157 return; 1158 } 1159 TAILQ_INSERT_TAIL(&sc->sc_queue, bp, bio_queue); 1160 mtx_unlock(&sc->sc_queue_mtx); 1161 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1162 wakeup(sc); 1163} 1164 1165/* 1166 * Return TRUE if the given request is colliding with a in-progress 1167 * synchronization request. 1168 */ 1169static bool 1170g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1171{ 1172 struct g_mirror_disk *disk; 1173 struct bio *sbp; 1174 off_t rstart, rend, sstart, send; 1175 u_int i; 1176 1177 if (sc->sc_sync.ds_ndisks == 0) 1178 return (false); 1179 rstart = bp->bio_offset; 1180 rend = bp->bio_offset + bp->bio_length; 1181 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1182 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1183 continue; 1184 for (i = 0; i < g_mirror_syncreqs; i++) { 1185 sbp = disk->d_sync.ds_bios[i]; 1186 if (sbp == NULL) 1187 continue; 1188 sstart = sbp->bio_offset; 1189 send = sbp->bio_offset + sbp->bio_length; 1190 if (rend > sstart && rstart < send) 1191 return (true); 1192 } 1193 } 1194 return (false); 1195} 1196 1197/* 1198 * Return TRUE if the given sync request is colliding with a in-progress regular 1199 * request. 1200 */ 1201static bool 1202g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1203{ 1204 off_t rstart, rend, sstart, send; 1205 struct bio *bp; 1206 1207 if (sc->sc_sync.ds_ndisks == 0) 1208 return (false); 1209 sstart = sbp->bio_offset; 1210 send = sbp->bio_offset + sbp->bio_length; 1211 TAILQ_FOREACH(bp, &sc->sc_inflight, bio_queue) { 1212 rstart = bp->bio_offset; 1213 rend = bp->bio_offset + bp->bio_length; 1214 if (rend > sstart && rstart < send) 1215 return (true); 1216 } 1217 return (false); 1218} 1219 1220/* 1221 * Puts regular request onto delayed queue. 1222 */ 1223static void 1224g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1225{ 1226 1227 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1228 TAILQ_INSERT_TAIL(&sc->sc_regular_delayed, bp, bio_queue); 1229} 1230 1231/* 1232 * Puts synchronization request onto delayed queue. 1233 */ 1234static void 1235g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1236{ 1237 1238 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1239 TAILQ_INSERT_TAIL(&sc->sc_sync_delayed, bp, bio_queue); 1240} 1241 1242/* 1243 * Requeue delayed regular requests. 1244 */ 1245static void 1246g_mirror_regular_release(struct g_mirror_softc *sc) 1247{ 1248 struct bio *bp; 1249 1250 if ((bp = TAILQ_FIRST(&sc->sc_regular_delayed)) == NULL) 1251 return; 1252 if (g_mirror_sync_collision(sc, bp)) 1253 return; 1254 1255 G_MIRROR_DEBUG(2, "Requeuing regular requests after collision."); 1256 mtx_lock(&sc->sc_queue_mtx); 1257 TAILQ_CONCAT(&sc->sc_regular_delayed, &sc->sc_queue, bio_queue); 1258 TAILQ_SWAP(&sc->sc_regular_delayed, &sc->sc_queue, bio, bio_queue); 1259 mtx_unlock(&sc->sc_queue_mtx); 1260} 1261 1262/* 1263 * Releases delayed sync requests which don't collide anymore with regular 1264 * requests. 1265 */ 1266static void 1267g_mirror_sync_release(struct g_mirror_softc *sc) 1268{ 1269 struct bio *bp, *bp2; 1270 1271 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed, bio_queue, bp2) { 1272 if (g_mirror_regular_collision(sc, bp)) 1273 continue; 1274 TAILQ_REMOVE(&sc->sc_sync_delayed, bp, bio_queue); 1275 G_MIRROR_LOGREQ(2, bp, 1276 "Releasing delayed synchronization request."); 1277 g_io_request(bp, bp->bio_from); 1278 } 1279} 1280 1281/* 1282 * Free a synchronization request and clear its slot in the array. 1283 */ 1284static void 1285g_mirror_sync_request_free(struct g_mirror_disk *disk, struct bio *bp) 1286{ 1287 int idx; 1288 1289 if (disk != NULL && disk->d_sync.ds_bios != NULL) { 1290 idx = (int)(uintptr_t)bp->bio_caller1; 1291 KASSERT(disk->d_sync.ds_bios[idx] == bp, 1292 ("unexpected sync BIO at %p:%d", disk, idx)); 1293 disk->d_sync.ds_bios[idx] = NULL; 1294 } 1295 free(bp->bio_data, M_MIRROR); 1296 g_destroy_bio(bp); 1297} 1298 1299/* 1300 * Handle synchronization requests. 1301 * Every synchronization request is a two-step process: first, a read request is 1302 * sent to the mirror provider via the sync consumer. If that request completes 1303 * successfully, it is converted to a write and sent to the disk being 1304 * synchronized. If the write also completes successfully, the synchronization 1305 * offset is advanced and a new read request is submitted. 1306 */ 1307static void 1308g_mirror_sync_request(struct g_mirror_softc *sc, struct bio *bp) 1309{ 1310 struct g_mirror_disk *disk; 1311 struct g_mirror_disk_sync *sync; 1312 1313 KASSERT((bp->bio_cmd == BIO_READ && 1314 bp->bio_from->geom == sc->sc_sync.ds_geom) || 1315 (bp->bio_cmd == BIO_WRITE && bp->bio_from->geom == sc->sc_geom), 1316 ("Sync BIO %p with unexpected origin", bp)); 1317 1318 bp->bio_from->index--; 1319 disk = bp->bio_from->private; 1320 if (disk == NULL) { 1321 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1322 g_topology_lock(); 1323 g_mirror_kill_consumer(sc, bp->bio_from); 1324 g_topology_unlock(); 1325 g_mirror_sync_request_free(NULL, bp); 1326 sx_xlock(&sc->sc_lock); 1327 return; 1328 } 1329 1330 sync = &disk->d_sync; 1331 1332 /* 1333 * Synchronization request. 1334 */ 1335 switch (bp->bio_cmd) { 1336 case BIO_READ: { 1337 struct g_mirror_disk *d; 1338 struct g_consumer *cp; 1339 int readable; 1340 1341 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_read, 1342 bp->bio_error); 1343 1344 if (bp->bio_error != 0) { 1345 G_MIRROR_LOGREQ(0, bp, 1346 "Synchronization request failed (error=%d).", 1347 bp->bio_error); 1348 1349 /* 1350 * If there's at least one other disk from which we can 1351 * read the block, retry the request. 1352 */ 1353 readable = 0; 1354 LIST_FOREACH(d, &sc->sc_disks, d_next) 1355 if (d->d_state == G_MIRROR_DISK_STATE_ACTIVE && 1356 !(d->d_flags & G_MIRROR_DISK_FLAG_BROKEN)) 1357 readable++; 1358 1359 /* 1360 * The read error will trigger a syncid bump, so there's 1361 * no need to do that here. 1362 * 1363 * If we can retry the read from another disk, do so. 1364 * Otherwise, all we can do is kick out the new disk. 1365 */ 1366 if (readable == 0) { 1367 g_mirror_sync_request_free(disk, bp); 1368 g_mirror_event_send(disk, 1369 G_MIRROR_DISK_STATE_DISCONNECTED, 1370 G_MIRROR_EVENT_DONTWAIT); 1371 } else { 1372 g_mirror_sync_reinit(disk, bp, bp->bio_offset); 1373 goto retry_read; 1374 } 1375 return; 1376 } 1377 G_MIRROR_LOGREQ(3, bp, 1378 "Synchronization request half-finished."); 1379 bp->bio_cmd = BIO_WRITE; 1380 bp->bio_cflags = 0; 1381 cp = disk->d_consumer; 1382 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1383 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1384 cp->acr, cp->acw, cp->ace)); 1385 cp->index++; 1386 g_io_request(bp, cp); 1387 return; 1388 } 1389 case BIO_WRITE: { 1390 off_t offset; 1391 int i; 1392 1393 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_write, 1394 bp->bio_error); 1395 1396 if (bp->bio_error != 0) { 1397 G_MIRROR_LOGREQ(0, bp, 1398 "Synchronization request failed (error=%d).", 1399 bp->bio_error); 1400 g_mirror_sync_request_free(disk, bp); 1401 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1402 g_mirror_event_send(disk, 1403 G_MIRROR_DISK_STATE_DISCONNECTED, 1404 G_MIRROR_EVENT_DONTWAIT); 1405 return; 1406 } 1407 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1408 if (sync->ds_offset >= sc->sc_mediasize || 1409 sync->ds_consumer == NULL || 1410 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1411 /* Don't send more synchronization requests. */ 1412 sync->ds_inflight--; 1413 g_mirror_sync_request_free(disk, bp); 1414 if (sync->ds_inflight > 0) 1415 return; 1416 if (sync->ds_consumer == NULL || 1417 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1418 return; 1419 } 1420 /* Disk up-to-date, activate it. */ 1421 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1422 G_MIRROR_EVENT_DONTWAIT); 1423 return; 1424 } 1425 1426 /* Send next synchronization request. */ 1427 g_mirror_sync_reinit(disk, bp, sync->ds_offset); 1428 sync->ds_offset += bp->bio_length; 1429 1430retry_read: 1431 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1432 sync->ds_consumer->index++; 1433 1434 /* 1435 * Delay the request if it is colliding with a regular request. 1436 */ 1437 if (g_mirror_regular_collision(sc, bp)) 1438 g_mirror_sync_delay(sc, bp); 1439 else 1440 g_io_request(bp, sync->ds_consumer); 1441 1442 /* Requeue delayed requests if possible. */ 1443 g_mirror_regular_release(sc); 1444 1445 /* Find the smallest offset */ 1446 offset = sc->sc_mediasize; 1447 for (i = 0; i < g_mirror_syncreqs; i++) { 1448 bp = sync->ds_bios[i]; 1449 if (bp != NULL && bp->bio_offset < offset) 1450 offset = bp->bio_offset; 1451 } 1452 if (g_mirror_sync_period > 0 && 1453 time_uptime - sync->ds_update_ts > g_mirror_sync_period) { 1454 sync->ds_offset_done = offset; 1455 g_mirror_update_metadata(disk); 1456 sync->ds_update_ts = time_uptime; 1457 } 1458 return; 1459 } 1460 default: 1461 panic("Invalid I/O request %p", bp); 1462 } 1463} 1464 1465static void 1466g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1467{ 1468 struct g_mirror_disk *disk; 1469 struct g_consumer *cp; 1470 struct bio *cbp; 1471 1472 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1473 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1474 break; 1475 } 1476 if (disk == NULL) { 1477 if (bp->bio_error == 0) 1478 bp->bio_error = ENXIO; 1479 g_io_deliver(bp, bp->bio_error); 1480 return; 1481 } 1482 cbp = g_clone_bio(bp); 1483 if (cbp == NULL) { 1484 if (bp->bio_error == 0) 1485 bp->bio_error = ENOMEM; 1486 g_io_deliver(bp, bp->bio_error); 1487 return; 1488 } 1489 /* 1490 * Fill in the component buf structure. 1491 */ 1492 cp = disk->d_consumer; 1493 cbp->bio_done = g_mirror_done; 1494 cbp->bio_to = cp->provider; 1495 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1496 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1497 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1498 cp->acw, cp->ace)); 1499 cp->index++; 1500 g_io_request(cbp, cp); 1501} 1502 1503static void 1504g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1505{ 1506 struct g_mirror_disk *disk; 1507 struct g_consumer *cp; 1508 struct bio *cbp; 1509 1510 disk = g_mirror_get_disk(sc); 1511 if (disk == NULL) { 1512 if (bp->bio_error == 0) 1513 bp->bio_error = ENXIO; 1514 g_io_deliver(bp, bp->bio_error); 1515 return; 1516 } 1517 cbp = g_clone_bio(bp); 1518 if (cbp == NULL) { 1519 if (bp->bio_error == 0) 1520 bp->bio_error = ENOMEM; 1521 g_io_deliver(bp, bp->bio_error); 1522 return; 1523 } 1524 /* 1525 * Fill in the component buf structure. 1526 */ 1527 cp = disk->d_consumer; 1528 cbp->bio_done = g_mirror_done; 1529 cbp->bio_to = cp->provider; 1530 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1531 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1532 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1533 cp->acw, cp->ace)); 1534 cp->index++; 1535 g_io_request(cbp, cp); 1536} 1537 1538#define TRACK_SIZE (1 * 1024 * 1024) 1539#define LOAD_SCALE 256 1540#define ABS(x) (((x) >= 0) ? (x) : (-(x))) 1541 1542static void 1543g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1544{ 1545 struct g_mirror_disk *disk, *dp; 1546 struct g_consumer *cp; 1547 struct bio *cbp; 1548 int prio, best; 1549 1550 /* Find a disk with the smallest load. */ 1551 disk = NULL; 1552 best = INT_MAX; 1553 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1554 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1555 continue; 1556 prio = dp->load; 1557 /* If disk head is precisely in position - highly prefer it. */ 1558 if (dp->d_last_offset == bp->bio_offset) 1559 prio -= 2 * LOAD_SCALE; 1560 else 1561 /* If disk head is close to position - prefer it. */ 1562 if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE) 1563 prio -= 1 * LOAD_SCALE; 1564 if (prio <= best) { 1565 disk = dp; 1566 best = prio; 1567 } 1568 } 1569 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1570 cbp = g_clone_bio(bp); 1571 if (cbp == NULL) { 1572 if (bp->bio_error == 0) 1573 bp->bio_error = ENOMEM; 1574 g_io_deliver(bp, bp->bio_error); 1575 return; 1576 } 1577 /* 1578 * Fill in the component buf structure. 1579 */ 1580 cp = disk->d_consumer; 1581 cbp->bio_done = g_mirror_done; 1582 cbp->bio_to = cp->provider; 1583 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1584 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1585 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1586 cp->acw, cp->ace)); 1587 cp->index++; 1588 /* Remember last head position */ 1589 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1590 /* Update loads. */ 1591 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1592 dp->load = (dp->d_consumer->index * LOAD_SCALE + 1593 dp->load * 7) / 8; 1594 } 1595 g_io_request(cbp, cp); 1596} 1597 1598static void 1599g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1600{ 1601 struct bio_queue queue; 1602 struct g_mirror_disk *disk; 1603 struct g_consumer *cp; 1604 struct bio *cbp; 1605 off_t left, mod, offset, slice; 1606 u_char *data; 1607 u_int ndisks; 1608 1609 if (bp->bio_length <= sc->sc_slice) { 1610 g_mirror_request_round_robin(sc, bp); 1611 return; 1612 } 1613 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1614 slice = bp->bio_length / ndisks; 1615 mod = slice % sc->sc_provider->sectorsize; 1616 if (mod != 0) 1617 slice += sc->sc_provider->sectorsize - mod; 1618 /* 1619 * Allocate all bios before sending any request, so we can 1620 * return ENOMEM in nice and clean way. 1621 */ 1622 left = bp->bio_length; 1623 offset = bp->bio_offset; 1624 data = bp->bio_data; 1625 TAILQ_INIT(&queue); 1626 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1627 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1628 continue; 1629 cbp = g_clone_bio(bp); 1630 if (cbp == NULL) { 1631 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1632 TAILQ_REMOVE(&queue, cbp, bio_queue); 1633 g_destroy_bio(cbp); 1634 } 1635 if (bp->bio_error == 0) 1636 bp->bio_error = ENOMEM; 1637 g_io_deliver(bp, bp->bio_error); 1638 return; 1639 } 1640 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 1641 cbp->bio_done = g_mirror_done; 1642 cbp->bio_caller1 = disk; 1643 cbp->bio_to = disk->d_consumer->provider; 1644 cbp->bio_offset = offset; 1645 cbp->bio_data = data; 1646 cbp->bio_length = MIN(left, slice); 1647 left -= cbp->bio_length; 1648 if (left == 0) 1649 break; 1650 offset += cbp->bio_length; 1651 data += cbp->bio_length; 1652 } 1653 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1654 TAILQ_REMOVE(&queue, cbp, bio_queue); 1655 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1656 disk = cbp->bio_caller1; 1657 cbp->bio_caller1 = NULL; 1658 cp = disk->d_consumer; 1659 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1660 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1661 cp->acr, cp->acw, cp->ace)); 1662 disk->d_consumer->index++; 1663 g_io_request(cbp, disk->d_consumer); 1664 } 1665} 1666 1667static void 1668g_mirror_register_request(struct g_mirror_softc *sc, struct bio *bp) 1669{ 1670 struct bio_queue queue; 1671 struct bio *cbp; 1672 struct g_consumer *cp; 1673 struct g_mirror_disk *disk; 1674 1675 sx_assert(&sc->sc_lock, SA_XLOCKED); 1676 1677 /* 1678 * To avoid ordering issues, if a write is deferred because of a 1679 * collision with a sync request, all I/O is deferred until that 1680 * write is initiated. 1681 */ 1682 if (bp->bio_from->geom != sc->sc_sync.ds_geom && 1683 !TAILQ_EMPTY(&sc->sc_regular_delayed)) { 1684 g_mirror_regular_delay(sc, bp); 1685 return; 1686 } 1687 1688 switch (bp->bio_cmd) { 1689 case BIO_READ: 1690 switch (sc->sc_balance) { 1691 case G_MIRROR_BALANCE_LOAD: 1692 g_mirror_request_load(sc, bp); 1693 break; 1694 case G_MIRROR_BALANCE_PREFER: 1695 g_mirror_request_prefer(sc, bp); 1696 break; 1697 case G_MIRROR_BALANCE_ROUND_ROBIN: 1698 g_mirror_request_round_robin(sc, bp); 1699 break; 1700 case G_MIRROR_BALANCE_SPLIT: 1701 g_mirror_request_split(sc, bp); 1702 break; 1703 } 1704 return; 1705 case BIO_WRITE: 1706 case BIO_DELETE: 1707 /* 1708 * Delay the request if it is colliding with a synchronization 1709 * request. 1710 */ 1711 if (g_mirror_sync_collision(sc, bp)) { 1712 g_mirror_regular_delay(sc, bp); 1713 return; 1714 } 1715 1716 if (sc->sc_idle) 1717 g_mirror_unidle(sc); 1718 else 1719 sc->sc_last_write = time_uptime; 1720 1721 /* 1722 * Bump syncid on first write. 1723 */ 1724 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1725 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1726 g_mirror_bump_syncid(sc); 1727 } 1728 1729 /* 1730 * Allocate all bios before sending any request, so we can 1731 * return ENOMEM in nice and clean way. 1732 */ 1733 TAILQ_INIT(&queue); 1734 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1735 switch (disk->d_state) { 1736 case G_MIRROR_DISK_STATE_ACTIVE: 1737 break; 1738 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1739 if (bp->bio_offset >= disk->d_sync.ds_offset) 1740 continue; 1741 break; 1742 default: 1743 continue; 1744 } 1745 if (bp->bio_cmd == BIO_DELETE && 1746 (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0) 1747 continue; 1748 cbp = g_clone_bio(bp); 1749 if (cbp == NULL) { 1750 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1751 TAILQ_REMOVE(&queue, cbp, bio_queue); 1752 g_destroy_bio(cbp); 1753 } 1754 if (bp->bio_error == 0) 1755 bp->bio_error = ENOMEM; 1756 g_io_deliver(bp, bp->bio_error); 1757 return; 1758 } 1759 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 1760 cbp->bio_done = g_mirror_done; 1761 cp = disk->d_consumer; 1762 cbp->bio_caller1 = cp; 1763 cbp->bio_to = cp->provider; 1764 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1765 ("Consumer %s not opened (r%dw%de%d).", 1766 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1767 } 1768 if (TAILQ_EMPTY(&queue)) { 1769 KASSERT(bp->bio_cmd == BIO_DELETE, 1770 ("No consumers for regular request %p", bp)); 1771 g_io_deliver(bp, EOPNOTSUPP); 1772 return; 1773 } 1774 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1775 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1776 TAILQ_REMOVE(&queue, cbp, bio_queue); 1777 cp = cbp->bio_caller1; 1778 cbp->bio_caller1 = NULL; 1779 cp->index++; 1780 sc->sc_writes++; 1781 g_io_request(cbp, cp); 1782 } 1783 /* 1784 * Put request onto inflight queue, so we can check if new 1785 * synchronization requests don't collide with it. 1786 */ 1787 TAILQ_INSERT_TAIL(&sc->sc_inflight, bp, bio_queue); 1788 return; 1789 case BIO_FLUSH: 1790 TAILQ_INIT(&queue); 1791 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1792 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1793 continue; 1794 cbp = g_clone_bio(bp); 1795 if (cbp == NULL) { 1796 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1797 TAILQ_REMOVE(&queue, cbp, bio_queue); 1798 g_destroy_bio(cbp); 1799 } 1800 if (bp->bio_error == 0) 1801 bp->bio_error = ENOMEM; 1802 g_io_deliver(bp, bp->bio_error); 1803 return; 1804 } 1805 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 1806 cbp->bio_done = g_mirror_done; 1807 cbp->bio_caller1 = disk; 1808 cbp->bio_to = disk->d_consumer->provider; 1809 } 1810 KASSERT(!TAILQ_EMPTY(&queue), 1811 ("No consumers for regular request %p", bp)); 1812 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1813 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1814 TAILQ_REMOVE(&queue, cbp, bio_queue); 1815 disk = cbp->bio_caller1; 1816 cbp->bio_caller1 = NULL; 1817 cp = disk->d_consumer; 1818 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1819 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1820 cp->acr, cp->acw, cp->ace)); 1821 cp->index++; 1822 g_io_request(cbp, cp); 1823 } 1824 break; 1825 default: 1826 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1827 bp->bio_cmd, sc->sc_name)); 1828 break; 1829 } 1830} 1831 1832static int 1833g_mirror_can_destroy(struct g_mirror_softc *sc) 1834{ 1835 struct g_geom *gp; 1836 struct g_consumer *cp; 1837 1838 g_topology_assert(); 1839 gp = sc->sc_geom; 1840 if (gp->softc == NULL) 1841 return (1); 1842 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0) 1843 return (0); 1844 LIST_FOREACH(cp, &gp->consumer, consumer) { 1845 if (g_mirror_is_busy(sc, cp)) 1846 return (0); 1847 } 1848 gp = sc->sc_sync.ds_geom; 1849 LIST_FOREACH(cp, &gp->consumer, consumer) { 1850 if (g_mirror_is_busy(sc, cp)) 1851 return (0); 1852 } 1853 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1854 sc->sc_name); 1855 return (1); 1856} 1857 1858static int 1859g_mirror_try_destroy(struct g_mirror_softc *sc) 1860{ 1861 1862 if (sc->sc_rootmount != NULL) { 1863 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1864 sc->sc_rootmount); 1865 root_mount_rel(sc->sc_rootmount); 1866 sc->sc_rootmount = NULL; 1867 } 1868 g_topology_lock(); 1869 if (!g_mirror_can_destroy(sc)) { 1870 g_topology_unlock(); 1871 return (0); 1872 } 1873 sc->sc_geom->softc = NULL; 1874 sc->sc_sync.ds_geom->softc = NULL; 1875 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DRAIN) != 0) { 1876 g_topology_unlock(); 1877 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1878 &sc->sc_worker); 1879 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1880 sx_xunlock(&sc->sc_lock); 1881 wakeup(&sc->sc_worker); 1882 sc->sc_worker = NULL; 1883 } else { 1884 g_topology_unlock(); 1885 g_mirror_destroy_device(sc); 1886 } 1887 return (1); 1888} 1889 1890/* 1891 * Worker thread. 1892 */ 1893static void 1894g_mirror_worker(void *arg) 1895{ 1896 struct g_mirror_softc *sc; 1897 struct g_mirror_event *ep; 1898 struct bio *bp; 1899 int timeout; 1900 1901 sc = arg; 1902 thread_lock(curthread); 1903 sched_prio(curthread, PRIBIO); 1904 thread_unlock(curthread); 1905 1906 sx_xlock(&sc->sc_lock); 1907 for (;;) { 1908 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1909 /* 1910 * First take a look at events. 1911 * This is important to handle events before any I/O requests. 1912 */ 1913 ep = g_mirror_event_first(sc); 1914 if (ep != NULL) { 1915 g_mirror_event_remove(sc, ep); 1916 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1917 /* Update only device status. */ 1918 G_MIRROR_DEBUG(3, 1919 "Running event for device %s.", 1920 sc->sc_name); 1921 ep->e_error = 0; 1922 g_mirror_update_device(sc, true); 1923 } else { 1924 /* Update disk status. */ 1925 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1926 g_mirror_get_diskname(ep->e_disk)); 1927 ep->e_error = g_mirror_update_disk(ep->e_disk, 1928 ep->e_state); 1929 if (ep->e_error == 0) 1930 g_mirror_update_device(sc, false); 1931 } 1932 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1933 KASSERT(ep->e_error == 0, 1934 ("Error cannot be handled.")); 1935 g_mirror_event_free(ep); 1936 } else { 1937 ep->e_flags |= G_MIRROR_EVENT_DONE; 1938 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1939 ep); 1940 mtx_lock(&sc->sc_events_mtx); 1941 wakeup(ep); 1942 mtx_unlock(&sc->sc_events_mtx); 1943 } 1944 if ((sc->sc_flags & 1945 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1946 if (g_mirror_try_destroy(sc)) { 1947 curthread->td_pflags &= ~TDP_GEOM; 1948 G_MIRROR_DEBUG(1, "Thread exiting."); 1949 kproc_exit(0); 1950 } 1951 } 1952 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1953 continue; 1954 } 1955 1956 /* 1957 * Check if we can mark array as CLEAN and if we can't take 1958 * how much seconds should we wait. 1959 */ 1960 timeout = g_mirror_idle(sc, -1); 1961 1962 /* 1963 * Handle I/O requests. 1964 */ 1965 mtx_lock(&sc->sc_queue_mtx); 1966 bp = TAILQ_FIRST(&sc->sc_queue); 1967 if (bp != NULL) 1968 TAILQ_REMOVE(&sc->sc_queue, bp, bio_queue); 1969 else { 1970 if ((sc->sc_flags & 1971 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1972 mtx_unlock(&sc->sc_queue_mtx); 1973 if (g_mirror_try_destroy(sc)) { 1974 curthread->td_pflags &= ~TDP_GEOM; 1975 G_MIRROR_DEBUG(1, "Thread exiting."); 1976 kproc_exit(0); 1977 } 1978 mtx_lock(&sc->sc_queue_mtx); 1979 if (!TAILQ_EMPTY(&sc->sc_queue)) { 1980 mtx_unlock(&sc->sc_queue_mtx); 1981 continue; 1982 } 1983 } 1984 if (g_mirror_event_first(sc) != NULL) { 1985 mtx_unlock(&sc->sc_queue_mtx); 1986 continue; 1987 } 1988 sx_xunlock(&sc->sc_lock); 1989 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1990 timeout * hz); 1991 sx_xlock(&sc->sc_lock); 1992 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1993 continue; 1994 } 1995 mtx_unlock(&sc->sc_queue_mtx); 1996 1997 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 1998 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1999 /* 2000 * Handle completion of the first half (the read) of a 2001 * block synchronization operation. 2002 */ 2003 g_mirror_sync_request(sc, bp); 2004 } else if (bp->bio_to != sc->sc_provider) { 2005 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 2006 /* 2007 * Handle completion of a regular I/O request. 2008 */ 2009 g_mirror_regular_request(sc, bp); 2010 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2011 /* 2012 * Handle completion of the second half (the 2013 * write) of a block synchronization operation. 2014 */ 2015 g_mirror_sync_request(sc, bp); 2016 else { 2017 KASSERT(0, 2018 ("Invalid request cflags=0x%hx to=%s.", 2019 bp->bio_cflags, bp->bio_to->name)); 2020 } 2021 } else { 2022 /* 2023 * Initiate an I/O request. 2024 */ 2025 g_mirror_register_request(sc, bp); 2026 } 2027 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 2028 } 2029} 2030 2031static void 2032g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 2033{ 2034 2035 sx_assert(&sc->sc_lock, SX_LOCKED); 2036 2037 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 2038 return; 2039 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2040 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 2041 g_mirror_get_diskname(disk), sc->sc_name); 2042 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2043 } else if (sc->sc_idle && 2044 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2045 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 2046 g_mirror_get_diskname(disk), sc->sc_name); 2047 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2048 } 2049} 2050 2051static void 2052g_mirror_sync_reinit(const struct g_mirror_disk *disk, struct bio *bp, 2053 off_t offset) 2054{ 2055 void *data; 2056 int idx; 2057 2058 data = bp->bio_data; 2059 idx = (int)(uintptr_t)bp->bio_caller1; 2060 g_reset_bio(bp); 2061 2062 bp->bio_cmd = BIO_READ; 2063 bp->bio_data = data; 2064 bp->bio_done = g_mirror_sync_done; 2065 bp->bio_from = disk->d_sync.ds_consumer; 2066 bp->bio_to = disk->d_softc->sc_provider; 2067 bp->bio_caller1 = (void *)(uintptr_t)idx; 2068 bp->bio_offset = offset; 2069 bp->bio_length = MIN(MAXPHYS, 2070 disk->d_softc->sc_mediasize - bp->bio_offset); 2071} 2072 2073static void 2074g_mirror_sync_start(struct g_mirror_disk *disk) 2075{ 2076 struct g_mirror_softc *sc; 2077 struct g_mirror_disk_sync *sync; 2078 struct g_consumer *cp; 2079 struct bio *bp; 2080 int error, i; 2081 2082 g_topology_assert_not(); 2083 sc = disk->d_softc; 2084 sync = &disk->d_sync; 2085 sx_assert(&sc->sc_lock, SX_LOCKED); 2086 2087 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2088 ("Disk %s is not marked for synchronization.", 2089 g_mirror_get_diskname(disk))); 2090 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2091 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 2092 sc->sc_state)); 2093 2094 sx_xunlock(&sc->sc_lock); 2095 g_topology_lock(); 2096 cp = g_new_consumer(sc->sc_sync.ds_geom); 2097 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 2098 error = g_attach(cp, sc->sc_provider); 2099 KASSERT(error == 0, 2100 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 2101 error = g_access(cp, 1, 0, 0); 2102 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 2103 g_topology_unlock(); 2104 sx_xlock(&sc->sc_lock); 2105 2106 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 2107 g_mirror_get_diskname(disk)); 2108 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0) 2109 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2110 KASSERT(sync->ds_consumer == NULL, 2111 ("Sync consumer already exists (device=%s, disk=%s).", 2112 sc->sc_name, g_mirror_get_diskname(disk))); 2113 2114 sync->ds_consumer = cp; 2115 sync->ds_consumer->private = disk; 2116 sync->ds_consumer->index = 0; 2117 2118 /* 2119 * Allocate memory for synchronization bios and initialize them. 2120 */ 2121 sync->ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 2122 M_MIRROR, M_WAITOK); 2123 for (i = 0; i < g_mirror_syncreqs; i++) { 2124 bp = g_alloc_bio(); 2125 sync->ds_bios[i] = bp; 2126 2127 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 2128 bp->bio_caller1 = (void *)(uintptr_t)i; 2129 g_mirror_sync_reinit(disk, bp, sync->ds_offset); 2130 sync->ds_offset += bp->bio_length; 2131 } 2132 2133 /* Increase the number of disks in SYNCHRONIZING state. */ 2134 sc->sc_sync.ds_ndisks++; 2135 /* Set the number of in-flight synchronization requests. */ 2136 sync->ds_inflight = g_mirror_syncreqs; 2137 2138 /* 2139 * Fire off first synchronization requests. 2140 */ 2141 for (i = 0; i < g_mirror_syncreqs; i++) { 2142 bp = sync->ds_bios[i]; 2143 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 2144 sync->ds_consumer->index++; 2145 /* 2146 * Delay the request if it is colliding with a regular request. 2147 */ 2148 if (g_mirror_regular_collision(sc, bp)) 2149 g_mirror_sync_delay(sc, bp); 2150 else 2151 g_io_request(bp, sync->ds_consumer); 2152 } 2153} 2154 2155/* 2156 * Stop synchronization process. 2157 * type: 0 - synchronization finished 2158 * 1 - synchronization stopped 2159 */ 2160static void 2161g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 2162{ 2163 struct g_mirror_softc *sc; 2164 struct g_consumer *cp; 2165 2166 g_topology_assert_not(); 2167 sc = disk->d_softc; 2168 sx_assert(&sc->sc_lock, SX_LOCKED); 2169 2170 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2171 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2172 g_mirror_disk_state2str(disk->d_state))); 2173 if (disk->d_sync.ds_consumer == NULL) 2174 return; 2175 2176 if (type == 0) { 2177 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2178 sc->sc_name, g_mirror_get_diskname(disk)); 2179 } else /* if (type == 1) */ { 2180 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2181 sc->sc_name, g_mirror_get_diskname(disk)); 2182 } 2183 g_mirror_regular_release(sc); 2184 free(disk->d_sync.ds_bios, M_MIRROR); 2185 disk->d_sync.ds_bios = NULL; 2186 cp = disk->d_sync.ds_consumer; 2187 disk->d_sync.ds_consumer = NULL; 2188 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2189 sc->sc_sync.ds_ndisks--; 2190 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2191 g_topology_lock(); 2192 g_mirror_kill_consumer(sc, cp); 2193 g_topology_unlock(); 2194 sx_xlock(&sc->sc_lock); 2195} 2196 2197static void 2198g_mirror_launch_provider(struct g_mirror_softc *sc) 2199{ 2200 struct g_mirror_disk *disk; 2201 struct g_provider *pp, *dp; 2202 2203 sx_assert(&sc->sc_lock, SX_LOCKED); 2204 2205 g_topology_lock(); 2206 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 2207 pp->flags |= G_PF_DIRECT_RECEIVE; 2208 pp->mediasize = sc->sc_mediasize; 2209 pp->sectorsize = sc->sc_sectorsize; 2210 pp->stripesize = 0; 2211 pp->stripeoffset = 0; 2212 2213 /* Splitting of unmapped BIO's could work but isn't implemented now */ 2214 if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT) 2215 pp->flags |= G_PF_ACCEPT_UNMAPPED; 2216 2217 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2218 if (disk->d_consumer && disk->d_consumer->provider) { 2219 dp = disk->d_consumer->provider; 2220 if (dp->stripesize > pp->stripesize) { 2221 pp->stripesize = dp->stripesize; 2222 pp->stripeoffset = dp->stripeoffset; 2223 } 2224 /* A provider underneath us doesn't support unmapped */ 2225 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { 2226 G_MIRROR_DEBUG(0, "Cancelling unmapped " 2227 "because of %s.", dp->name); 2228 pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 2229 } 2230 } 2231 } 2232 pp->private = sc; 2233 sc->sc_refcnt++; 2234 sc->sc_provider = pp; 2235 g_error_provider(pp, 0); 2236 g_topology_unlock(); 2237 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2238 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 2239 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2240 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2241 g_mirror_sync_start(disk); 2242 } 2243} 2244 2245static void 2246g_mirror_destroy_provider(struct g_mirror_softc *sc) 2247{ 2248 struct g_mirror_disk *disk; 2249 struct bio *bp; 2250 2251 g_topology_assert_not(); 2252 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2253 sc->sc_name)); 2254 2255 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2256 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2257 g_mirror_sync_stop(disk, 1); 2258 } 2259 2260 g_topology_lock(); 2261 g_error_provider(sc->sc_provider, ENXIO); 2262 mtx_lock(&sc->sc_queue_mtx); 2263 while ((bp = TAILQ_FIRST(&sc->sc_queue)) != NULL) { 2264 TAILQ_REMOVE(&sc->sc_queue, bp, bio_queue); 2265 /* 2266 * Abort any pending I/O that wasn't generated by us. 2267 * Synchronization requests and requests destined for individual 2268 * mirror components can be destroyed immediately. 2269 */ 2270 if (bp->bio_to == sc->sc_provider && 2271 bp->bio_from->geom != sc->sc_sync.ds_geom) { 2272 g_io_deliver(bp, ENXIO); 2273 } else { 2274 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2275 free(bp->bio_data, M_MIRROR); 2276 g_destroy_bio(bp); 2277 } 2278 } 2279 mtx_unlock(&sc->sc_queue_mtx); 2280 g_wither_provider(sc->sc_provider, ENXIO); 2281 sc->sc_provider = NULL; 2282 G_MIRROR_DEBUG(0, "Device %s: provider destroyed.", sc->sc_name); 2283 g_topology_unlock(); 2284} 2285 2286static void 2287g_mirror_go(void *arg) 2288{ 2289 struct g_mirror_softc *sc; 2290 2291 sc = arg; 2292 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2293 g_mirror_event_send(sc, 0, 2294 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2295} 2296 2297static u_int 2298g_mirror_determine_state(struct g_mirror_disk *disk) 2299{ 2300 struct g_mirror_softc *sc; 2301 u_int state; 2302 2303 sc = disk->d_softc; 2304 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2305 if ((disk->d_flags & 2306 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0 && 2307 (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 || 2308 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0)) { 2309 /* Disk does not need synchronization. */ 2310 state = G_MIRROR_DISK_STATE_ACTIVE; 2311 } else { 2312 if ((sc->sc_flags & 2313 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2314 (disk->d_flags & 2315 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2316 /* 2317 * We can start synchronization from 2318 * the stored offset. 2319 */ 2320 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2321 } else { 2322 state = G_MIRROR_DISK_STATE_STALE; 2323 } 2324 } 2325 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2326 /* 2327 * Reset all synchronization data for this disk, 2328 * because if it even was synchronized, it was 2329 * synchronized to disks with different syncid. 2330 */ 2331 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2332 disk->d_sync.ds_offset = 0; 2333 disk->d_sync.ds_offset_done = 0; 2334 disk->d_sync.ds_syncid = sc->sc_syncid; 2335 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2336 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2337 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2338 } else { 2339 state = G_MIRROR_DISK_STATE_STALE; 2340 } 2341 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2342 /* 2343 * Not good, NOT GOOD! 2344 * It means that mirror was started on stale disks 2345 * and more fresh disk just arrive. 2346 * If there were writes, mirror is broken, sorry. 2347 * I think the best choice here is don't touch 2348 * this disk and inform the user loudly. 2349 */ 2350 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2351 "disk (%s) arrives!! It will not be connected to the " 2352 "running device.", sc->sc_name, 2353 g_mirror_get_diskname(disk)); 2354 g_mirror_destroy_disk(disk); 2355 state = G_MIRROR_DISK_STATE_NONE; 2356 /* Return immediately, because disk was destroyed. */ 2357 return (state); 2358 } 2359 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2360 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2361 return (state); 2362} 2363 2364/* 2365 * Update device state. 2366 */ 2367static void 2368g_mirror_update_device(struct g_mirror_softc *sc, bool force) 2369{ 2370 struct g_mirror_disk *disk; 2371 u_int state; 2372 2373 sx_assert(&sc->sc_lock, SX_XLOCKED); 2374 2375 switch (sc->sc_state) { 2376 case G_MIRROR_DEVICE_STATE_STARTING: 2377 { 2378 struct g_mirror_disk *pdisk, *tdisk; 2379 u_int dirty, ndisks, genid, syncid; 2380 bool broken; 2381 2382 KASSERT(sc->sc_provider == NULL, 2383 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2384 /* 2385 * Are we ready? We are, if all disks are connected or 2386 * if we have any disks and 'force' is true. 2387 */ 2388 ndisks = g_mirror_ndisks(sc, -1); 2389 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2390 ; 2391 } else if (ndisks == 0) { 2392 /* 2393 * Disks went down in starting phase, so destroy 2394 * device. 2395 */ 2396 callout_drain(&sc->sc_callout); 2397 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2398 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2399 sc->sc_rootmount); 2400 root_mount_rel(sc->sc_rootmount); 2401 sc->sc_rootmount = NULL; 2402 return; 2403 } else { 2404 return; 2405 } 2406 2407 /* 2408 * Activate all disks with the biggest syncid. 2409 */ 2410 if (force) { 2411 /* 2412 * If 'force' is true, we have been called due to 2413 * timeout, so don't bother canceling timeout. 2414 */ 2415 ndisks = 0; 2416 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2417 if ((disk->d_flags & 2418 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2419 ndisks++; 2420 } 2421 } 2422 if (ndisks == 0) { 2423 /* No valid disks found, destroy device. */ 2424 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2425 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2426 __LINE__, sc->sc_rootmount); 2427 root_mount_rel(sc->sc_rootmount); 2428 sc->sc_rootmount = NULL; 2429 return; 2430 } 2431 } else { 2432 /* Cancel timeout. */ 2433 callout_drain(&sc->sc_callout); 2434 } 2435 2436 /* 2437 * Find the biggest genid. 2438 */ 2439 genid = 0; 2440 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2441 if (disk->d_genid > genid) 2442 genid = disk->d_genid; 2443 } 2444 sc->sc_genid = genid; 2445 /* 2446 * Remove all disks without the biggest genid. 2447 */ 2448 broken = false; 2449 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2450 if (disk->d_genid < genid) { 2451 G_MIRROR_DEBUG(0, 2452 "Component %s (device %s) broken, skipping.", 2453 g_mirror_get_diskname(disk), sc->sc_name); 2454 g_mirror_destroy_disk(disk); 2455 /* 2456 * Bump the syncid in case we discover a healthy 2457 * replacement disk after starting the mirror. 2458 */ 2459 broken = true; 2460 } 2461 } 2462 2463 /* 2464 * Find the biggest syncid. 2465 */ 2466 syncid = 0; 2467 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2468 if (disk->d_sync.ds_syncid > syncid) 2469 syncid = disk->d_sync.ds_syncid; 2470 } 2471 2472 /* 2473 * Here we need to look for dirty disks and if all disks 2474 * with the biggest syncid are dirty, we have to choose 2475 * one with the biggest priority and rebuild the rest. 2476 */ 2477 /* 2478 * Find the number of dirty disks with the biggest syncid. 2479 * Find the number of disks with the biggest syncid. 2480 * While here, find a disk with the biggest priority. 2481 */ 2482 dirty = ndisks = 0; 2483 pdisk = NULL; 2484 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2485 if (disk->d_sync.ds_syncid != syncid) 2486 continue; 2487 if ((disk->d_flags & 2488 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2489 continue; 2490 } 2491 ndisks++; 2492 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2493 dirty++; 2494 if (pdisk == NULL || 2495 pdisk->d_priority < disk->d_priority) { 2496 pdisk = disk; 2497 } 2498 } 2499 } 2500 if (dirty == 0) { 2501 /* No dirty disks at all, great. */ 2502 } else if (dirty == ndisks) { 2503 /* 2504 * Force synchronization for all dirty disks except one 2505 * with the biggest priority. 2506 */ 2507 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2508 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2509 "master disk for synchronization.", 2510 g_mirror_get_diskname(pdisk), sc->sc_name); 2511 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2512 if (disk->d_sync.ds_syncid != syncid) 2513 continue; 2514 if ((disk->d_flags & 2515 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2516 continue; 2517 } 2518 KASSERT((disk->d_flags & 2519 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2520 ("Disk %s isn't marked as dirty.", 2521 g_mirror_get_diskname(disk))); 2522 /* Skip the disk with the biggest priority. */ 2523 if (disk == pdisk) 2524 continue; 2525 disk->d_sync.ds_syncid = 0; 2526 } 2527 } else if (dirty < ndisks) { 2528 /* 2529 * Force synchronization for all dirty disks. 2530 * We have some non-dirty disks. 2531 */ 2532 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2533 if (disk->d_sync.ds_syncid != syncid) 2534 continue; 2535 if ((disk->d_flags & 2536 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2537 continue; 2538 } 2539 if ((disk->d_flags & 2540 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2541 continue; 2542 } 2543 disk->d_sync.ds_syncid = 0; 2544 } 2545 } 2546 2547 /* Reset hint. */ 2548 sc->sc_hint = NULL; 2549 sc->sc_syncid = syncid; 2550 if (force || broken) { 2551 /* Remember to bump syncid on first write. */ 2552 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2553 } 2554 state = G_MIRROR_DEVICE_STATE_RUNNING; 2555 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2556 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2557 g_mirror_device_state2str(state)); 2558 sc->sc_state = state; 2559 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2560 state = g_mirror_determine_state(disk); 2561 g_mirror_event_send(disk, state, 2562 G_MIRROR_EVENT_DONTWAIT); 2563 if (state == G_MIRROR_DISK_STATE_STALE) 2564 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2565 } 2566 break; 2567 } 2568 case G_MIRROR_DEVICE_STATE_RUNNING: 2569 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2570 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2571 /* 2572 * No usable disks, so destroy the device. 2573 */ 2574 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2575 break; 2576 } else if (g_mirror_ndisks(sc, 2577 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2578 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2579 /* 2580 * We have active disks, launch provider if it doesn't 2581 * exist. 2582 */ 2583 if (sc->sc_provider == NULL) 2584 g_mirror_launch_provider(sc); 2585 if (sc->sc_rootmount != NULL) { 2586 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2587 __LINE__, sc->sc_rootmount); 2588 root_mount_rel(sc->sc_rootmount); 2589 sc->sc_rootmount = NULL; 2590 } 2591 } 2592 /* 2593 * Genid should be bumped immediately, so do it here. 2594 */ 2595 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2596 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2597 g_mirror_bump_genid(sc); 2598 } 2599 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_NOW) != 0) { 2600 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID_NOW; 2601 g_mirror_bump_syncid(sc); 2602 } 2603 break; 2604 default: 2605 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2606 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2607 break; 2608 } 2609} 2610 2611/* 2612 * Update disk state and device state if needed. 2613 */ 2614#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2615 "Disk %s state changed from %s to %s (device %s).", \ 2616 g_mirror_get_diskname(disk), \ 2617 g_mirror_disk_state2str(disk->d_state), \ 2618 g_mirror_disk_state2str(state), sc->sc_name) 2619static int 2620g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2621{ 2622 struct g_mirror_softc *sc; 2623 2624 sc = disk->d_softc; 2625 sx_assert(&sc->sc_lock, SX_XLOCKED); 2626 2627again: 2628 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2629 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2630 g_mirror_disk_state2str(state)); 2631 switch (state) { 2632 case G_MIRROR_DISK_STATE_NEW: 2633 /* 2634 * Possible scenarios: 2635 * 1. New disk arrive. 2636 */ 2637 /* Previous state should be NONE. */ 2638 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2639 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2640 g_mirror_disk_state2str(disk->d_state))); 2641 DISK_STATE_CHANGED(); 2642 2643 disk->d_state = state; 2644 if (LIST_EMPTY(&sc->sc_disks)) 2645 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2646 else { 2647 struct g_mirror_disk *dp; 2648 2649 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2650 if (disk->d_priority >= dp->d_priority) { 2651 LIST_INSERT_BEFORE(dp, disk, d_next); 2652 dp = NULL; 2653 break; 2654 } 2655 if (LIST_NEXT(dp, d_next) == NULL) 2656 break; 2657 } 2658 if (dp != NULL) 2659 LIST_INSERT_AFTER(dp, disk, d_next); 2660 } 2661 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2662 sc->sc_name, g_mirror_get_diskname(disk)); 2663 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2664 break; 2665 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2666 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2667 g_mirror_device_state2str(sc->sc_state), 2668 g_mirror_get_diskname(disk), 2669 g_mirror_disk_state2str(disk->d_state))); 2670 state = g_mirror_determine_state(disk); 2671 if (state != G_MIRROR_DISK_STATE_NONE) 2672 goto again; 2673 break; 2674 case G_MIRROR_DISK_STATE_ACTIVE: 2675 /* 2676 * Possible scenarios: 2677 * 1. New disk does not need synchronization. 2678 * 2. Synchronization process finished successfully. 2679 */ 2680 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2681 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2682 g_mirror_device_state2str(sc->sc_state), 2683 g_mirror_get_diskname(disk), 2684 g_mirror_disk_state2str(disk->d_state))); 2685 /* Previous state should be NEW or SYNCHRONIZING. */ 2686 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2687 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2688 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2689 g_mirror_disk_state2str(disk->d_state))); 2690 DISK_STATE_CHANGED(); 2691 2692 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2693 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2694 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2695 g_mirror_sync_stop(disk, 0); 2696 } 2697 disk->d_state = state; 2698 disk->d_sync.ds_offset = 0; 2699 disk->d_sync.ds_offset_done = 0; 2700 g_mirror_update_idle(sc, disk); 2701 g_mirror_update_metadata(disk); 2702 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2703 sc->sc_name, g_mirror_get_diskname(disk)); 2704 break; 2705 case G_MIRROR_DISK_STATE_STALE: 2706 /* 2707 * Possible scenarios: 2708 * 1. Stale disk was connected. 2709 */ 2710 /* Previous state should be NEW. */ 2711 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2712 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2713 g_mirror_disk_state2str(disk->d_state))); 2714 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2715 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2716 g_mirror_device_state2str(sc->sc_state), 2717 g_mirror_get_diskname(disk), 2718 g_mirror_disk_state2str(disk->d_state))); 2719 /* 2720 * STALE state is only possible if device is marked 2721 * NOAUTOSYNC. 2722 */ 2723 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2724 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2725 g_mirror_device_state2str(sc->sc_state), 2726 g_mirror_get_diskname(disk), 2727 g_mirror_disk_state2str(disk->d_state))); 2728 DISK_STATE_CHANGED(); 2729 2730 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2731 disk->d_state = state; 2732 g_mirror_update_metadata(disk); 2733 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2734 sc->sc_name, g_mirror_get_diskname(disk)); 2735 break; 2736 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2737 /* 2738 * Possible scenarios: 2739 * 1. Disk which needs synchronization was connected. 2740 */ 2741 /* Previous state should be NEW. */ 2742 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2743 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2744 g_mirror_disk_state2str(disk->d_state))); 2745 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2746 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2747 g_mirror_device_state2str(sc->sc_state), 2748 g_mirror_get_diskname(disk), 2749 g_mirror_disk_state2str(disk->d_state))); 2750 DISK_STATE_CHANGED(); 2751 2752 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2753 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2754 disk->d_state = state; 2755 if (sc->sc_provider != NULL) { 2756 g_mirror_sync_start(disk); 2757 g_mirror_update_metadata(disk); 2758 } 2759 break; 2760 case G_MIRROR_DISK_STATE_DISCONNECTED: 2761 /* 2762 * Possible scenarios: 2763 * 1. Device wasn't running yet, but disk disappear. 2764 * 2. Disk was active and disapppear. 2765 * 3. Disk disappear during synchronization process. 2766 */ 2767 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2768 /* 2769 * Previous state should be ACTIVE, STALE or 2770 * SYNCHRONIZING. 2771 */ 2772 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2773 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2774 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2775 ("Wrong disk state (%s, %s).", 2776 g_mirror_get_diskname(disk), 2777 g_mirror_disk_state2str(disk->d_state))); 2778 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2779 /* Previous state should be NEW. */ 2780 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2781 ("Wrong disk state (%s, %s).", 2782 g_mirror_get_diskname(disk), 2783 g_mirror_disk_state2str(disk->d_state))); 2784 /* 2785 * Reset bumping syncid if disk disappeared in STARTING 2786 * state. 2787 */ 2788 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2789 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2790#ifdef INVARIANTS 2791 } else { 2792 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2793 sc->sc_name, 2794 g_mirror_device_state2str(sc->sc_state), 2795 g_mirror_get_diskname(disk), 2796 g_mirror_disk_state2str(disk->d_state))); 2797#endif 2798 } 2799 DISK_STATE_CHANGED(); 2800 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2801 sc->sc_name, g_mirror_get_diskname(disk)); 2802 2803 g_mirror_destroy_disk(disk); 2804 break; 2805 case G_MIRROR_DISK_STATE_DESTROY: 2806 { 2807 int error; 2808 2809 error = g_mirror_clear_metadata(disk); 2810 if (error != 0) { 2811 G_MIRROR_DEBUG(0, 2812 "Device %s: failed to clear metadata on %s: %d.", 2813 sc->sc_name, g_mirror_get_diskname(disk), error); 2814 break; 2815 } 2816 DISK_STATE_CHANGED(); 2817 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2818 sc->sc_name, g_mirror_get_diskname(disk)); 2819 2820 g_mirror_destroy_disk(disk); 2821 sc->sc_ndisks--; 2822 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2823 g_mirror_update_metadata(disk); 2824 } 2825 break; 2826 } 2827 default: 2828 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2829 break; 2830 } 2831 return (0); 2832} 2833#undef DISK_STATE_CHANGED 2834 2835int 2836g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2837{ 2838 struct g_provider *pp; 2839 u_char *buf; 2840 int error; 2841 2842 g_topology_assert(); 2843 2844 error = g_access(cp, 1, 0, 0); 2845 if (error != 0) 2846 return (error); 2847 pp = cp->provider; 2848 g_topology_unlock(); 2849 /* Metadata are stored on last sector. */ 2850 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2851 &error); 2852 g_topology_lock(); 2853 g_access(cp, -1, 0, 0); 2854 if (buf == NULL) { 2855 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2856 cp->provider->name, error); 2857 return (error); 2858 } 2859 2860 /* Decode metadata. */ 2861 error = mirror_metadata_decode(buf, md); 2862 g_free(buf); 2863 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2864 return (EINVAL); 2865 if (md->md_version > G_MIRROR_VERSION) { 2866 G_MIRROR_DEBUG(0, 2867 "Kernel module is too old to handle metadata from %s.", 2868 cp->provider->name); 2869 return (EINVAL); 2870 } 2871 if (error != 0) { 2872 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2873 cp->provider->name); 2874 return (error); 2875 } 2876 2877 return (0); 2878} 2879 2880static int 2881g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2882 struct g_mirror_metadata *md) 2883{ 2884 2885 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2886 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2887 pp->name, md->md_did); 2888 return (EEXIST); 2889 } 2890 if (md->md_all != sc->sc_ndisks) { 2891 G_MIRROR_DEBUG(1, 2892 "Invalid '%s' field on disk %s (device %s), skipping.", 2893 "md_all", pp->name, sc->sc_name); 2894 return (EINVAL); 2895 } 2896 if (md->md_slice != sc->sc_slice) { 2897 G_MIRROR_DEBUG(1, 2898 "Invalid '%s' field on disk %s (device %s), skipping.", 2899 "md_slice", pp->name, sc->sc_name); 2900 return (EINVAL); 2901 } 2902 if (md->md_balance != sc->sc_balance) { 2903 G_MIRROR_DEBUG(1, 2904 "Invalid '%s' field on disk %s (device %s), skipping.", 2905 "md_balance", pp->name, sc->sc_name); 2906 return (EINVAL); 2907 } 2908#if 0 2909 if (md->md_mediasize != sc->sc_mediasize) { 2910 G_MIRROR_DEBUG(1, 2911 "Invalid '%s' field on disk %s (device %s), skipping.", 2912 "md_mediasize", pp->name, sc->sc_name); 2913 return (EINVAL); 2914 } 2915#endif 2916 if (sc->sc_mediasize > pp->mediasize) { 2917 G_MIRROR_DEBUG(1, 2918 "Invalid size of disk %s (device %s), skipping.", pp->name, 2919 sc->sc_name); 2920 return (EINVAL); 2921 } 2922 if (md->md_sectorsize != sc->sc_sectorsize) { 2923 G_MIRROR_DEBUG(1, 2924 "Invalid '%s' field on disk %s (device %s), skipping.", 2925 "md_sectorsize", pp->name, sc->sc_name); 2926 return (EINVAL); 2927 } 2928 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2929 G_MIRROR_DEBUG(1, 2930 "Invalid sector size of disk %s (device %s), skipping.", 2931 pp->name, sc->sc_name); 2932 return (EINVAL); 2933 } 2934 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2935 G_MIRROR_DEBUG(1, 2936 "Invalid device flags on disk %s (device %s), skipping.", 2937 pp->name, sc->sc_name); 2938 return (EINVAL); 2939 } 2940 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2941 G_MIRROR_DEBUG(1, 2942 "Invalid disk flags on disk %s (device %s), skipping.", 2943 pp->name, sc->sc_name); 2944 return (EINVAL); 2945 } 2946 return (0); 2947} 2948 2949int 2950g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2951 struct g_mirror_metadata *md) 2952{ 2953 struct g_mirror_disk *disk; 2954 int error; 2955 2956 g_topology_assert_not(); 2957 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2958 2959 error = g_mirror_check_metadata(sc, pp, md); 2960 if (error != 0) 2961 return (error); 2962 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2963 md->md_genid < sc->sc_genid) { 2964 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2965 pp->name, sc->sc_name); 2966 return (EINVAL); 2967 } 2968 disk = g_mirror_init_disk(sc, pp, md, &error); 2969 if (disk == NULL) 2970 return (error); 2971 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2972 G_MIRROR_EVENT_WAIT); 2973 if (error != 0) 2974 return (error); 2975 if (md->md_version < G_MIRROR_VERSION) { 2976 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2977 pp->name, md->md_version, G_MIRROR_VERSION); 2978 g_mirror_update_metadata(disk); 2979 } 2980 return (0); 2981} 2982 2983static void 2984g_mirror_destroy_delayed(void *arg, int flag) 2985{ 2986 struct g_mirror_softc *sc; 2987 int error; 2988 2989 if (flag == EV_CANCEL) { 2990 G_MIRROR_DEBUG(1, "Destroying canceled."); 2991 return; 2992 } 2993 sc = arg; 2994 g_topology_unlock(); 2995 sx_xlock(&sc->sc_lock); 2996 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2997 ("DESTROY flag set on %s.", sc->sc_name)); 2998 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0, 2999 ("CLOSEWAIT flag not set on %s.", sc->sc_name)); 3000 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 3001 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 3002 if (error != 0) { 3003 G_MIRROR_DEBUG(0, "Cannot destroy %s (error=%d).", 3004 sc->sc_name, error); 3005 sx_xunlock(&sc->sc_lock); 3006 } 3007 g_topology_lock(); 3008} 3009 3010static int 3011g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 3012{ 3013 struct g_mirror_softc *sc; 3014 int error = 0; 3015 3016 g_topology_assert(); 3017 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 3018 acw, ace); 3019 3020 sc = pp->private; 3021 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 3022 3023 g_topology_unlock(); 3024 sx_xlock(&sc->sc_lock); 3025 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 3026 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 || 3027 LIST_EMPTY(&sc->sc_disks)) { 3028 if (acr > 0 || acw > 0 || ace > 0) 3029 error = ENXIO; 3030 goto end; 3031 } 3032 sc->sc_provider_open += acr + acw + ace; 3033 if (pp->acw + acw == 0) 3034 g_mirror_idle(sc, 0); 3035 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 && 3036 sc->sc_provider_open == 0) 3037 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, sc, NULL); 3038end: 3039 sx_xunlock(&sc->sc_lock); 3040 g_topology_lock(); 3041 return (error); 3042} 3043 3044struct g_geom * 3045g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md, 3046 u_int type) 3047{ 3048 struct g_mirror_softc *sc; 3049 struct g_geom *gp; 3050 int error, timeout; 3051 3052 g_topology_assert(); 3053 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 3054 md->md_mid); 3055 3056 /* One disk is minimum. */ 3057 if (md->md_all < 1) 3058 return (NULL); 3059 /* 3060 * Action geom. 3061 */ 3062 gp = g_new_geomf(mp, "%s", md->md_name); 3063 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 3064 gp->start = g_mirror_start; 3065 gp->orphan = g_mirror_orphan; 3066 gp->access = g_mirror_access; 3067 gp->dumpconf = g_mirror_dumpconf; 3068 3069 sc->sc_type = type; 3070 sc->sc_id = md->md_mid; 3071 sc->sc_slice = md->md_slice; 3072 sc->sc_balance = md->md_balance; 3073 sc->sc_mediasize = md->md_mediasize; 3074 sc->sc_sectorsize = md->md_sectorsize; 3075 sc->sc_ndisks = md->md_all; 3076 sc->sc_flags = md->md_mflags; 3077 sc->sc_bump_id = 0; 3078 sc->sc_idle = 1; 3079 sc->sc_last_write = time_uptime; 3080 sc->sc_writes = 0; 3081 sc->sc_refcnt = 1; 3082 sx_init(&sc->sc_lock, "gmirror:lock"); 3083 TAILQ_INIT(&sc->sc_queue); 3084 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 3085 TAILQ_INIT(&sc->sc_regular_delayed); 3086 TAILQ_INIT(&sc->sc_inflight); 3087 TAILQ_INIT(&sc->sc_sync_delayed); 3088 LIST_INIT(&sc->sc_disks); 3089 TAILQ_INIT(&sc->sc_events); 3090 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 3091 callout_init(&sc->sc_callout, 1); 3092 mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF); 3093 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 3094 gp->softc = sc; 3095 sc->sc_geom = gp; 3096 sc->sc_provider = NULL; 3097 sc->sc_provider_open = 0; 3098 /* 3099 * Synchronization geom. 3100 */ 3101 gp = g_new_geomf(mp, "%s.sync", md->md_name); 3102 gp->softc = sc; 3103 gp->orphan = g_mirror_orphan; 3104 sc->sc_sync.ds_geom = gp; 3105 sc->sc_sync.ds_ndisks = 0; 3106 error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 3107 "g_mirror %s", md->md_name); 3108 if (error != 0) { 3109 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 3110 sc->sc_name); 3111 g_destroy_geom(sc->sc_sync.ds_geom); 3112 g_destroy_geom(sc->sc_geom); 3113 g_mirror_free_device(sc); 3114 return (NULL); 3115 } 3116 3117 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 3118 sc->sc_name, sc->sc_ndisks, sc->sc_id); 3119 3120 sc->sc_rootmount = root_mount_hold("GMIRROR"); 3121 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 3122 /* 3123 * Run timeout. 3124 */ 3125 timeout = g_mirror_timeout * hz; 3126 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 3127 return (sc->sc_geom); 3128} 3129 3130int 3131g_mirror_destroy(struct g_mirror_softc *sc, int how) 3132{ 3133 struct g_mirror_disk *disk; 3134 3135 g_topology_assert_not(); 3136 sx_assert(&sc->sc_lock, SX_XLOCKED); 3137 3138 if (sc->sc_provider_open != 0) { 3139 switch (how) { 3140 case G_MIRROR_DESTROY_SOFT: 3141 G_MIRROR_DEBUG(1, 3142 "Device %s is still open (%d).", sc->sc_name, 3143 sc->sc_provider_open); 3144 return (EBUSY); 3145 case G_MIRROR_DESTROY_DELAYED: 3146 G_MIRROR_DEBUG(1, 3147 "Device %s will be destroyed on last close.", 3148 sc->sc_name); 3149 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 3150 if (disk->d_state == 3151 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3152 g_mirror_sync_stop(disk, 1); 3153 } 3154 } 3155 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_CLOSEWAIT; 3156 return (EBUSY); 3157 case G_MIRROR_DESTROY_HARD: 3158 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 3159 "can't be definitely removed.", sc->sc_name); 3160 } 3161 } 3162 3163 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3164 sx_xunlock(&sc->sc_lock); 3165 return (0); 3166 } 3167 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 3168 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DRAIN; 3169 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3170 sx_xunlock(&sc->sc_lock); 3171 mtx_lock(&sc->sc_queue_mtx); 3172 wakeup(sc); 3173 mtx_unlock(&sc->sc_queue_mtx); 3174 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3175 while (sc->sc_worker != NULL) 3176 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 3177 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3178 sx_xlock(&sc->sc_lock); 3179 g_mirror_destroy_device(sc); 3180 return (0); 3181} 3182 3183static void 3184g_mirror_taste_orphan(struct g_consumer *cp) 3185{ 3186 3187 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3188 cp->provider->name)); 3189} 3190 3191static struct g_geom * 3192g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3193{ 3194 struct g_mirror_metadata md; 3195 struct g_mirror_softc *sc; 3196 struct g_consumer *cp; 3197 struct g_geom *gp; 3198 int error; 3199 3200 g_topology_assert(); 3201 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3202 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 3203 3204 gp = g_new_geomf(mp, "mirror:taste"); 3205 /* 3206 * This orphan function should be never called. 3207 */ 3208 gp->orphan = g_mirror_taste_orphan; 3209 cp = g_new_consumer(gp); 3210 g_attach(cp, pp); 3211 error = g_mirror_read_metadata(cp, &md); 3212 g_detach(cp); 3213 g_destroy_consumer(cp); 3214 g_destroy_geom(gp); 3215 if (error != 0) 3216 return (NULL); 3217 gp = NULL; 3218 3219 if (md.md_provider[0] != '\0' && 3220 !g_compare_names(md.md_provider, pp->name)) 3221 return (NULL); 3222 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3223 return (NULL); 3224 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 3225 G_MIRROR_DEBUG(0, 3226 "Device %s: provider %s marked as inactive, skipping.", 3227 md.md_name, pp->name); 3228 return (NULL); 3229 } 3230 if (g_mirror_debug >= 2) 3231 mirror_metadata_dump(&md); 3232 3233 /* 3234 * Let's check if device already exists. 3235 */ 3236 sc = NULL; 3237 LIST_FOREACH(gp, &mp->geom, geom) { 3238 sc = gp->softc; 3239 if (sc == NULL) 3240 continue; 3241 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 3242 continue; 3243 if (sc->sc_sync.ds_geom == gp) 3244 continue; 3245 if (strcmp(md.md_name, sc->sc_name) != 0) 3246 continue; 3247 if (md.md_mid != sc->sc_id) { 3248 G_MIRROR_DEBUG(0, "Device %s already configured.", 3249 sc->sc_name); 3250 return (NULL); 3251 } 3252 break; 3253 } 3254 if (gp == NULL) { 3255 gp = g_mirror_create(mp, &md, G_MIRROR_TYPE_AUTOMATIC); 3256 if (gp == NULL) { 3257 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3258 md.md_name); 3259 return (NULL); 3260 } 3261 sc = gp->softc; 3262 } 3263 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3264 g_topology_unlock(); 3265 sx_xlock(&sc->sc_lock); 3266 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; 3267 error = g_mirror_add_disk(sc, pp, &md); 3268 if (error != 0) { 3269 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3270 pp->name, gp->name, error); 3271 if (LIST_EMPTY(&sc->sc_disks)) { 3272 g_cancel_event(sc); 3273 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3274 g_topology_lock(); 3275 return (NULL); 3276 } 3277 gp = NULL; 3278 } 3279 sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; 3280 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3281 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3282 g_topology_lock(); 3283 return (NULL); 3284 } 3285 sx_xunlock(&sc->sc_lock); 3286 g_topology_lock(); 3287 return (gp); 3288} 3289 3290static void 3291g_mirror_resize(struct g_consumer *cp) 3292{ 3293 struct g_mirror_disk *disk; 3294 3295 g_topology_assert(); 3296 g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name); 3297 3298 disk = cp->private; 3299 if (disk == NULL) 3300 return; 3301 g_topology_unlock(); 3302 g_mirror_update_metadata(disk); 3303 g_topology_lock(); 3304} 3305 3306static int 3307g_mirror_destroy_geom(struct gctl_req *req __unused, 3308 struct g_class *mp __unused, struct g_geom *gp) 3309{ 3310 struct g_mirror_softc *sc; 3311 int error; 3312 3313 g_topology_unlock(); 3314 sc = gp->softc; 3315 sx_xlock(&sc->sc_lock); 3316 g_cancel_event(sc); 3317 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3318 if (error != 0) 3319 sx_xunlock(&sc->sc_lock); 3320 g_topology_lock(); 3321 return (error); 3322} 3323 3324static void 3325g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3326 struct g_consumer *cp, struct g_provider *pp) 3327{ 3328 struct g_mirror_softc *sc; 3329 3330 g_topology_assert(); 3331 3332 sc = gp->softc; 3333 if (sc == NULL) 3334 return; 3335 /* Skip synchronization geom. */ 3336 if (gp == sc->sc_sync.ds_geom) 3337 return; 3338 if (pp != NULL) { 3339 /* Nothing here. */ 3340 } else if (cp != NULL) { 3341 struct g_mirror_disk *disk; 3342 3343 disk = cp->private; 3344 if (disk == NULL) 3345 return; 3346 g_topology_unlock(); 3347 sx_xlock(&sc->sc_lock); 3348 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3349 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3350 sbuf_printf(sb, "%s<Synchronized>", indent); 3351 if (disk->d_sync.ds_offset == 0) 3352 sbuf_printf(sb, "0%%"); 3353 else { 3354 sbuf_printf(sb, "%u%%", 3355 (u_int)((disk->d_sync.ds_offset * 100) / 3356 sc->sc_provider->mediasize)); 3357 } 3358 sbuf_printf(sb, "</Synchronized>\n"); 3359 if (disk->d_sync.ds_offset > 0) { 3360 sbuf_printf(sb, "%s<BytesSynced>%jd" 3361 "</BytesSynced>\n", indent, 3362 (intmax_t)disk->d_sync.ds_offset); 3363 } 3364 } 3365 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3366 disk->d_sync.ds_syncid); 3367 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3368 disk->d_genid); 3369 sbuf_printf(sb, "%s<Flags>", indent); 3370 if (disk->d_flags == 0) 3371 sbuf_printf(sb, "NONE"); 3372 else { 3373 int first = 1; 3374 3375#define ADD_FLAG(flag, name) do { \ 3376 if ((disk->d_flags & (flag)) != 0) { \ 3377 if (!first) \ 3378 sbuf_printf(sb, ", "); \ 3379 else \ 3380 first = 0; \ 3381 sbuf_printf(sb, name); \ 3382 } \ 3383} while (0) 3384 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3385 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3386 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3387 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3388 "SYNCHRONIZING"); 3389 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3390 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3391#undef ADD_FLAG 3392 } 3393 sbuf_printf(sb, "</Flags>\n"); 3394 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3395 disk->d_priority); 3396 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3397 g_mirror_disk_state2str(disk->d_state)); 3398 sx_xunlock(&sc->sc_lock); 3399 g_topology_lock(); 3400 } else { 3401 g_topology_unlock(); 3402 sx_xlock(&sc->sc_lock); 3403 sbuf_printf(sb, "%s<Type>", indent); 3404 switch (sc->sc_type) { 3405 case G_MIRROR_TYPE_AUTOMATIC: 3406 sbuf_printf(sb, "AUTOMATIC"); 3407 break; 3408 case G_MIRROR_TYPE_MANUAL: 3409 sbuf_printf(sb, "MANUAL"); 3410 break; 3411 default: 3412 sbuf_printf(sb, "UNKNOWN"); 3413 break; 3414 } 3415 sbuf_printf(sb, "</Type>\n"); 3416 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3417 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3418 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3419 sbuf_printf(sb, "%s<Flags>", indent); 3420 if (sc->sc_flags == 0) 3421 sbuf_printf(sb, "NONE"); 3422 else { 3423 int first = 1; 3424 3425#define ADD_FLAG(flag, name) do { \ 3426 if ((sc->sc_flags & (flag)) != 0) { \ 3427 if (!first) \ 3428 sbuf_printf(sb, ", "); \ 3429 else \ 3430 first = 0; \ 3431 sbuf_printf(sb, name); \ 3432 } \ 3433} while (0) 3434 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3435 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3436#undef ADD_FLAG 3437 } 3438 sbuf_printf(sb, "</Flags>\n"); 3439 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3440 (u_int)sc->sc_slice); 3441 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3442 balance_name(sc->sc_balance)); 3443 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3444 sc->sc_ndisks); 3445 sbuf_printf(sb, "%s<State>", indent); 3446 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3447 sbuf_printf(sb, "%s", "STARTING"); 3448 else if (sc->sc_ndisks == 3449 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3450 sbuf_printf(sb, "%s", "COMPLETE"); 3451 else 3452 sbuf_printf(sb, "%s", "DEGRADED"); 3453 sbuf_printf(sb, "</State>\n"); 3454 sx_xunlock(&sc->sc_lock); 3455 g_topology_lock(); 3456 } 3457} 3458 3459static void 3460g_mirror_shutdown_post_sync(void *arg, int howto) 3461{ 3462 struct g_class *mp; 3463 struct g_geom *gp, *gp2; 3464 struct g_mirror_softc *sc; 3465 int error; 3466 3467 if (panicstr != NULL) 3468 return; 3469 3470 mp = arg; 3471 g_topology_lock(); 3472 g_mirror_shutdown = 1; 3473 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3474 if ((sc = gp->softc) == NULL) 3475 continue; 3476 /* Skip synchronization geom. */ 3477 if (gp == sc->sc_sync.ds_geom) 3478 continue; 3479 g_topology_unlock(); 3480 sx_xlock(&sc->sc_lock); 3481 g_mirror_idle(sc, -1); 3482 g_cancel_event(sc); 3483 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3484 if (error != 0) 3485 sx_xunlock(&sc->sc_lock); 3486 g_topology_lock(); 3487 } 3488 g_topology_unlock(); 3489} 3490 3491static void 3492g_mirror_init(struct g_class *mp) 3493{ 3494 3495 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3496 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3497 if (g_mirror_post_sync == NULL) 3498 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3499} 3500 3501static void 3502g_mirror_fini(struct g_class *mp) 3503{ 3504 3505 if (g_mirror_post_sync != NULL) 3506 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); 3507} 3508 3509DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3510