g_mirror.c revision 327804
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 327804 2018-01-11 00:54:54Z markj $"); 29 30#include <sys/param.h> 31#include <sys/systm.h> 32#include <sys/fail.h> 33#include <sys/kernel.h> 34#include <sys/module.h> 35#include <sys/limits.h> 36#include <sys/lock.h> 37#include <sys/mutex.h> 38#include <sys/bio.h> 39#include <sys/sbuf.h> 40#include <sys/sysctl.h> 41#include <sys/malloc.h> 42#include <sys/eventhandler.h> 43#include <vm/uma.h> 44#include <geom/geom.h> 45#include <sys/proc.h> 46#include <sys/kthread.h> 47#include <sys/sched.h> 48#include <geom/mirror/g_mirror.h> 49 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_stop(struct g_mirror_disk *disk, int type); 112static void g_mirror_register_request(struct bio *bp); 113static void g_mirror_sync_release(struct g_mirror_softc *sc); 114 115 116static const char * 117g_mirror_disk_state2str(int state) 118{ 119 120 switch (state) { 121 case G_MIRROR_DISK_STATE_NONE: 122 return ("NONE"); 123 case G_MIRROR_DISK_STATE_NEW: 124 return ("NEW"); 125 case G_MIRROR_DISK_STATE_ACTIVE: 126 return ("ACTIVE"); 127 case G_MIRROR_DISK_STATE_STALE: 128 return ("STALE"); 129 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 130 return ("SYNCHRONIZING"); 131 case G_MIRROR_DISK_STATE_DISCONNECTED: 132 return ("DISCONNECTED"); 133 case G_MIRROR_DISK_STATE_DESTROY: 134 return ("DESTROY"); 135 default: 136 return ("INVALID"); 137 } 138} 139 140static const char * 141g_mirror_device_state2str(int state) 142{ 143 144 switch (state) { 145 case G_MIRROR_DEVICE_STATE_STARTING: 146 return ("STARTING"); 147 case G_MIRROR_DEVICE_STATE_RUNNING: 148 return ("RUNNING"); 149 default: 150 return ("INVALID"); 151 } 152} 153 154static const char * 155g_mirror_get_diskname(struct g_mirror_disk *disk) 156{ 157 158 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 159 return ("[unknown]"); 160 return (disk->d_name); 161} 162 163/* 164 * --- Events handling functions --- 165 * Events in geom_mirror are used to maintain disks and device status 166 * from one thread to simplify locking. 167 */ 168static void 169g_mirror_event_free(struct g_mirror_event *ep) 170{ 171 172 free(ep, M_MIRROR); 173} 174 175int 176g_mirror_event_send(void *arg, int state, int flags) 177{ 178 struct g_mirror_softc *sc; 179 struct g_mirror_disk *disk; 180 struct g_mirror_event *ep; 181 int error; 182 183 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 184 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 185 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 186 disk = NULL; 187 sc = arg; 188 } else { 189 disk = arg; 190 sc = disk->d_softc; 191 } 192 ep->e_disk = disk; 193 ep->e_state = state; 194 ep->e_flags = flags; 195 ep->e_error = 0; 196 mtx_lock(&sc->sc_events_mtx); 197 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 198 mtx_unlock(&sc->sc_events_mtx); 199 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 200 mtx_lock(&sc->sc_queue_mtx); 201 wakeup(sc); 202 mtx_unlock(&sc->sc_queue_mtx); 203 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 204 return (0); 205 sx_assert(&sc->sc_lock, SX_XLOCKED); 206 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 207 sx_xunlock(&sc->sc_lock); 208 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 209 mtx_lock(&sc->sc_events_mtx); 210 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 211 hz * 5); 212 } 213 error = ep->e_error; 214 g_mirror_event_free(ep); 215 sx_xlock(&sc->sc_lock); 216 return (error); 217} 218 219static struct g_mirror_event * 220g_mirror_event_first(struct g_mirror_softc *sc) 221{ 222 struct g_mirror_event *ep; 223 224 mtx_lock(&sc->sc_events_mtx); 225 ep = TAILQ_FIRST(&sc->sc_events); 226 mtx_unlock(&sc->sc_events_mtx); 227 return (ep); 228} 229 230static void 231g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 232{ 233 234 mtx_lock(&sc->sc_events_mtx); 235 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 236 mtx_unlock(&sc->sc_events_mtx); 237} 238 239static void 240g_mirror_event_cancel(struct g_mirror_disk *disk) 241{ 242 struct g_mirror_softc *sc; 243 struct g_mirror_event *ep, *tmpep; 244 245 sc = disk->d_softc; 246 sx_assert(&sc->sc_lock, SX_XLOCKED); 247 248 mtx_lock(&sc->sc_events_mtx); 249 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 250 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 251 continue; 252 if (ep->e_disk != disk) 253 continue; 254 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 255 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 256 g_mirror_event_free(ep); 257 else { 258 ep->e_error = ECANCELED; 259 wakeup(ep); 260 } 261 } 262 mtx_unlock(&sc->sc_events_mtx); 263} 264 265/* 266 * Return the number of disks in given state. 267 * If state is equal to -1, count all connected disks. 268 */ 269u_int 270g_mirror_ndisks(struct g_mirror_softc *sc, int state) 271{ 272 struct g_mirror_disk *disk; 273 u_int n = 0; 274 275 sx_assert(&sc->sc_lock, SX_LOCKED); 276 277 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 278 if (state == -1 || disk->d_state == state) 279 n++; 280 } 281 return (n); 282} 283 284/* 285 * Find a disk in mirror by its disk ID. 286 */ 287static struct g_mirror_disk * 288g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 289{ 290 struct g_mirror_disk *disk; 291 292 sx_assert(&sc->sc_lock, SX_XLOCKED); 293 294 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 295 if (disk->d_id == id) 296 return (disk); 297 } 298 return (NULL); 299} 300 301static u_int 302g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 303{ 304 struct bio *bp; 305 u_int nreqs = 0; 306 307 mtx_lock(&sc->sc_queue_mtx); 308 TAILQ_FOREACH(bp, &sc->sc_queue, bio_queue) { 309 if (bp->bio_from == cp) 310 nreqs++; 311 } 312 mtx_unlock(&sc->sc_queue_mtx); 313 return (nreqs); 314} 315 316static int 317g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 318{ 319 320 if (cp->index > 0) { 321 G_MIRROR_DEBUG(2, 322 "I/O requests for %s exist, can't destroy it now.", 323 cp->provider->name); 324 return (1); 325 } 326 if (g_mirror_nrequests(sc, cp) > 0) { 327 G_MIRROR_DEBUG(2, 328 "I/O requests for %s in queue, can't destroy it now.", 329 cp->provider->name); 330 return (1); 331 } 332 return (0); 333} 334 335static void 336g_mirror_destroy_consumer(void *arg, int flags __unused) 337{ 338 struct g_consumer *cp; 339 340 g_topology_assert(); 341 342 cp = arg; 343 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 344 g_detach(cp); 345 g_destroy_consumer(cp); 346} 347 348static void 349g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 350{ 351 struct g_provider *pp; 352 int retaste_wait; 353 354 g_topology_assert(); 355 356 cp->private = NULL; 357 if (g_mirror_is_busy(sc, cp)) 358 return; 359 pp = cp->provider; 360 retaste_wait = 0; 361 if (cp->acw == 1) { 362 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 363 retaste_wait = 1; 364 } 365 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 366 -cp->acw, -cp->ace, 0); 367 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 368 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 369 if (retaste_wait) { 370 /* 371 * After retaste event was send (inside g_access()), we can send 372 * event to detach and destroy consumer. 373 * A class, which has consumer to the given provider connected 374 * will not receive retaste event for the provider. 375 * This is the way how I ignore retaste events when I close 376 * consumers opened for write: I detach and destroy consumer 377 * after retaste event is sent. 378 */ 379 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 380 return; 381 } 382 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 383 g_detach(cp); 384 g_destroy_consumer(cp); 385} 386 387static int 388g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 389{ 390 struct g_consumer *cp; 391 int error; 392 393 g_topology_assert_not(); 394 KASSERT(disk->d_consumer == NULL, 395 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 396 397 g_topology_lock(); 398 cp = g_new_consumer(disk->d_softc->sc_geom); 399 cp->flags |= G_CF_DIRECT_RECEIVE; 400 error = g_attach(cp, pp); 401 if (error != 0) { 402 g_destroy_consumer(cp); 403 g_topology_unlock(); 404 return (error); 405 } 406 error = g_access(cp, 1, 1, 1); 407 if (error != 0) { 408 g_detach(cp); 409 g_destroy_consumer(cp); 410 g_topology_unlock(); 411 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 412 pp->name, error); 413 return (error); 414 } 415 g_topology_unlock(); 416 disk->d_consumer = cp; 417 disk->d_consumer->private = disk; 418 disk->d_consumer->index = 0; 419 420 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 421 return (0); 422} 423 424static void 425g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 426{ 427 428 g_topology_assert(); 429 430 if (cp == NULL) 431 return; 432 if (cp->provider != NULL) 433 g_mirror_kill_consumer(sc, cp); 434 else 435 g_destroy_consumer(cp); 436} 437 438/* 439 * Initialize disk. This means allocate memory, create consumer, attach it 440 * to the provider and open access (r1w1e1) to it. 441 */ 442static struct g_mirror_disk * 443g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 444 struct g_mirror_metadata *md, int *errorp) 445{ 446 struct g_mirror_disk *disk; 447 int i, error; 448 449 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 450 if (disk == NULL) { 451 error = ENOMEM; 452 goto fail; 453 } 454 disk->d_softc = sc; 455 error = g_mirror_connect_disk(disk, pp); 456 if (error != 0) 457 goto fail; 458 disk->d_id = md->md_did; 459 disk->d_state = G_MIRROR_DISK_STATE_NONE; 460 disk->d_priority = md->md_priority; 461 disk->d_flags = md->md_dflags; 462 error = g_getattr("GEOM::candelete", disk->d_consumer, &i); 463 if (error == 0 && i != 0) 464 disk->d_flags |= G_MIRROR_DISK_FLAG_CANDELETE; 465 if (md->md_provider[0] != '\0') 466 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 467 disk->d_sync.ds_consumer = NULL; 468 disk->d_sync.ds_offset = md->md_sync_offset; 469 disk->d_sync.ds_offset_done = md->md_sync_offset; 470 disk->d_sync.ds_update_ts = time_uptime; 471 disk->d_genid = md->md_genid; 472 disk->d_sync.ds_syncid = md->md_syncid; 473 if (errorp != NULL) 474 *errorp = 0; 475 return (disk); 476fail: 477 if (errorp != NULL) 478 *errorp = error; 479 if (disk != NULL) 480 free(disk, M_MIRROR); 481 return (NULL); 482} 483 484static void 485g_mirror_destroy_disk(struct g_mirror_disk *disk) 486{ 487 struct g_mirror_softc *sc; 488 489 g_topology_assert_not(); 490 sc = disk->d_softc; 491 sx_assert(&sc->sc_lock, SX_XLOCKED); 492 493 LIST_REMOVE(disk, d_next); 494 g_mirror_event_cancel(disk); 495 if (sc->sc_hint == disk) 496 sc->sc_hint = NULL; 497 switch (disk->d_state) { 498 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 499 g_mirror_sync_stop(disk, 1); 500 /* FALLTHROUGH */ 501 case G_MIRROR_DISK_STATE_NEW: 502 case G_MIRROR_DISK_STATE_STALE: 503 case G_MIRROR_DISK_STATE_ACTIVE: 504 g_topology_lock(); 505 g_mirror_disconnect_consumer(sc, disk->d_consumer); 506 g_topology_unlock(); 507 free(disk, M_MIRROR); 508 break; 509 default: 510 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 511 g_mirror_get_diskname(disk), 512 g_mirror_disk_state2str(disk->d_state))); 513 } 514} 515 516static void 517g_mirror_free_device(struct g_mirror_softc *sc) 518{ 519 520 mtx_destroy(&sc->sc_queue_mtx); 521 mtx_destroy(&sc->sc_events_mtx); 522 mtx_destroy(&sc->sc_done_mtx); 523 sx_destroy(&sc->sc_lock); 524 free(sc, M_MIRROR); 525} 526 527static void 528g_mirror_providergone(struct g_provider *pp) 529{ 530 struct g_mirror_softc *sc = pp->private; 531 532 if ((--sc->sc_refcnt) == 0) 533 g_mirror_free_device(sc); 534} 535 536static void 537g_mirror_destroy_device(struct g_mirror_softc *sc) 538{ 539 struct g_mirror_disk *disk; 540 struct g_mirror_event *ep; 541 struct g_geom *gp; 542 struct g_consumer *cp, *tmpcp; 543 544 g_topology_assert_not(); 545 sx_assert(&sc->sc_lock, SX_XLOCKED); 546 547 gp = sc->sc_geom; 548 if (sc->sc_provider != NULL) 549 g_mirror_destroy_provider(sc); 550 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 551 disk = LIST_FIRST(&sc->sc_disks)) { 552 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 553 g_mirror_update_metadata(disk); 554 g_mirror_destroy_disk(disk); 555 } 556 while ((ep = g_mirror_event_first(sc)) != NULL) { 557 g_mirror_event_remove(sc, ep); 558 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 559 g_mirror_event_free(ep); 560 else { 561 ep->e_error = ECANCELED; 562 ep->e_flags |= G_MIRROR_EVENT_DONE; 563 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 564 mtx_lock(&sc->sc_events_mtx); 565 wakeup(ep); 566 mtx_unlock(&sc->sc_events_mtx); 567 } 568 } 569 callout_drain(&sc->sc_callout); 570 571 g_topology_lock(); 572 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 573 g_mirror_disconnect_consumer(sc, cp); 574 } 575 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 576 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 577 g_wither_geom(gp, ENXIO); 578 sx_xunlock(&sc->sc_lock); 579 if ((--sc->sc_refcnt) == 0) 580 g_mirror_free_device(sc); 581 g_topology_unlock(); 582} 583 584static void 585g_mirror_orphan(struct g_consumer *cp) 586{ 587 struct g_mirror_disk *disk; 588 589 g_topology_assert(); 590 591 disk = cp->private; 592 if (disk == NULL) 593 return; 594 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 595 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 596 G_MIRROR_EVENT_DONTWAIT); 597} 598 599/* 600 * Function should return the next active disk on the list. 601 * It is possible that it will be the same disk as given. 602 * If there are no active disks on list, NULL is returned. 603 */ 604static __inline struct g_mirror_disk * 605g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 606{ 607 struct g_mirror_disk *dp; 608 609 for (dp = LIST_NEXT(disk, d_next); dp != disk; 610 dp = LIST_NEXT(dp, d_next)) { 611 if (dp == NULL) 612 dp = LIST_FIRST(&sc->sc_disks); 613 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 614 break; 615 } 616 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 617 return (NULL); 618 return (dp); 619} 620 621static struct g_mirror_disk * 622g_mirror_get_disk(struct g_mirror_softc *sc) 623{ 624 struct g_mirror_disk *disk; 625 626 if (sc->sc_hint == NULL) { 627 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 628 if (sc->sc_hint == NULL) 629 return (NULL); 630 } 631 disk = sc->sc_hint; 632 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 633 disk = g_mirror_find_next(sc, disk); 634 if (disk == NULL) 635 return (NULL); 636 } 637 sc->sc_hint = g_mirror_find_next(sc, disk); 638 return (disk); 639} 640 641static int 642g_mirror_write_metadata(struct g_mirror_disk *disk, 643 struct g_mirror_metadata *md) 644{ 645 struct g_mirror_softc *sc; 646 struct g_consumer *cp; 647 off_t offset, length; 648 u_char *sector; 649 int error = 0; 650 651 g_topology_assert_not(); 652 sc = disk->d_softc; 653 sx_assert(&sc->sc_lock, SX_LOCKED); 654 655 cp = disk->d_consumer; 656 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 657 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 658 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 659 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 660 cp->acw, cp->ace)); 661 length = cp->provider->sectorsize; 662 offset = cp->provider->mediasize - length; 663 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 664 if (md != NULL && 665 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) { 666 /* 667 * Handle the case, when the size of parent provider reduced. 668 */ 669 if (offset < md->md_mediasize) 670 error = ENOSPC; 671 else 672 mirror_metadata_encode(md, sector); 673 } 674 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_metadata_write, error); 675 if (error == 0) 676 error = g_write_data(cp, offset, sector, length); 677 free(sector, M_MIRROR); 678 if (error != 0) { 679 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 680 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 681 G_MIRROR_DEBUG(0, "Cannot write metadata on %s " 682 "(device=%s, error=%d).", 683 g_mirror_get_diskname(disk), sc->sc_name, error); 684 } else { 685 G_MIRROR_DEBUG(1, "Cannot write metadata on %s " 686 "(device=%s, error=%d).", 687 g_mirror_get_diskname(disk), sc->sc_name, error); 688 } 689 if (g_mirror_disconnect_on_failure && 690 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 691 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 692 g_mirror_event_send(disk, 693 G_MIRROR_DISK_STATE_DISCONNECTED, 694 G_MIRROR_EVENT_DONTWAIT); 695 } 696 } 697 return (error); 698} 699 700static int 701g_mirror_clear_metadata(struct g_mirror_disk *disk) 702{ 703 int error; 704 705 g_topology_assert_not(); 706 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 707 708 if (disk->d_softc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 709 return (0); 710 error = g_mirror_write_metadata(disk, NULL); 711 if (error == 0) { 712 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 713 g_mirror_get_diskname(disk)); 714 } else { 715 G_MIRROR_DEBUG(0, 716 "Cannot clear metadata on disk %s (error=%d).", 717 g_mirror_get_diskname(disk), error); 718 } 719 return (error); 720} 721 722void 723g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 724 struct g_mirror_metadata *md) 725{ 726 727 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 728 md->md_version = G_MIRROR_VERSION; 729 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 730 md->md_mid = sc->sc_id; 731 md->md_all = sc->sc_ndisks; 732 md->md_slice = sc->sc_slice; 733 md->md_balance = sc->sc_balance; 734 md->md_genid = sc->sc_genid; 735 md->md_mediasize = sc->sc_mediasize; 736 md->md_sectorsize = sc->sc_sectorsize; 737 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 738 bzero(md->md_provider, sizeof(md->md_provider)); 739 if (disk == NULL) { 740 md->md_did = arc4random(); 741 md->md_priority = 0; 742 md->md_syncid = 0; 743 md->md_dflags = 0; 744 md->md_sync_offset = 0; 745 md->md_provsize = 0; 746 } else { 747 md->md_did = disk->d_id; 748 md->md_priority = disk->d_priority; 749 md->md_syncid = disk->d_sync.ds_syncid; 750 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 751 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 752 md->md_sync_offset = disk->d_sync.ds_offset_done; 753 else 754 md->md_sync_offset = 0; 755 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 756 strlcpy(md->md_provider, 757 disk->d_consumer->provider->name, 758 sizeof(md->md_provider)); 759 } 760 md->md_provsize = disk->d_consumer->provider->mediasize; 761 } 762} 763 764void 765g_mirror_update_metadata(struct g_mirror_disk *disk) 766{ 767 struct g_mirror_softc *sc; 768 struct g_mirror_metadata md; 769 int error; 770 771 g_topology_assert_not(); 772 sc = disk->d_softc; 773 sx_assert(&sc->sc_lock, SX_LOCKED); 774 775 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 776 return; 777 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) 778 g_mirror_fill_metadata(sc, disk, &md); 779 error = g_mirror_write_metadata(disk, &md); 780 if (error == 0) { 781 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 782 g_mirror_get_diskname(disk)); 783 } else { 784 G_MIRROR_DEBUG(0, 785 "Cannot update metadata on disk %s (error=%d).", 786 g_mirror_get_diskname(disk), error); 787 } 788} 789 790static void 791g_mirror_bump_syncid(struct g_mirror_softc *sc) 792{ 793 struct g_mirror_disk *disk; 794 795 g_topology_assert_not(); 796 sx_assert(&sc->sc_lock, SX_XLOCKED); 797 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 798 ("%s called with no active disks (device=%s).", __func__, 799 sc->sc_name)); 800 801 sc->sc_syncid++; 802 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 803 sc->sc_syncid); 804 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 805 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 806 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 807 disk->d_sync.ds_syncid = sc->sc_syncid; 808 g_mirror_update_metadata(disk); 809 } 810 } 811} 812 813static void 814g_mirror_bump_genid(struct g_mirror_softc *sc) 815{ 816 struct g_mirror_disk *disk; 817 818 g_topology_assert_not(); 819 sx_assert(&sc->sc_lock, SX_XLOCKED); 820 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 821 ("%s called with no active disks (device=%s).", __func__, 822 sc->sc_name)); 823 824 sc->sc_genid++; 825 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 826 sc->sc_genid); 827 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 828 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 829 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 830 disk->d_genid = sc->sc_genid; 831 g_mirror_update_metadata(disk); 832 } 833 } 834} 835 836static int 837g_mirror_idle(struct g_mirror_softc *sc, int acw) 838{ 839 struct g_mirror_disk *disk; 840 int timeout; 841 842 g_topology_assert_not(); 843 sx_assert(&sc->sc_lock, SX_XLOCKED); 844 845 if (sc->sc_provider == NULL) 846 return (0); 847 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 848 return (0); 849 if (sc->sc_idle) 850 return (0); 851 if (sc->sc_writes > 0) 852 return (0); 853 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 854 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); 855 if (!g_mirror_shutdown && timeout > 0) 856 return (timeout); 857 } 858 sc->sc_idle = 1; 859 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 860 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 861 continue; 862 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 863 g_mirror_get_diskname(disk), sc->sc_name); 864 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 865 g_mirror_update_metadata(disk); 866 } 867 return (0); 868} 869 870static void 871g_mirror_unidle(struct g_mirror_softc *sc) 872{ 873 struct g_mirror_disk *disk; 874 875 g_topology_assert_not(); 876 sx_assert(&sc->sc_lock, SX_XLOCKED); 877 878 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 879 return; 880 sc->sc_idle = 0; 881 sc->sc_last_write = time_uptime; 882 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 883 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 884 continue; 885 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 886 g_mirror_get_diskname(disk), sc->sc_name); 887 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 888 g_mirror_update_metadata(disk); 889 } 890} 891 892static void 893g_mirror_flush_done(struct bio *bp) 894{ 895 struct g_mirror_softc *sc; 896 struct bio *pbp; 897 898 pbp = bp->bio_parent; 899 sc = pbp->bio_to->private; 900 mtx_lock(&sc->sc_done_mtx); 901 if (pbp->bio_error == 0) 902 pbp->bio_error = bp->bio_error; 903 pbp->bio_completed += bp->bio_completed; 904 pbp->bio_inbed++; 905 if (pbp->bio_children == pbp->bio_inbed) { 906 mtx_unlock(&sc->sc_done_mtx); 907 g_io_deliver(pbp, pbp->bio_error); 908 } else 909 mtx_unlock(&sc->sc_done_mtx); 910 g_destroy_bio(bp); 911} 912 913static void 914g_mirror_done(struct bio *bp) 915{ 916 struct g_mirror_softc *sc; 917 918 sc = bp->bio_from->geom->softc; 919 bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR; 920 mtx_lock(&sc->sc_queue_mtx); 921 TAILQ_INSERT_TAIL(&sc->sc_queue, bp, bio_queue); 922 mtx_unlock(&sc->sc_queue_mtx); 923 wakeup(sc); 924} 925 926static void 927g_mirror_regular_request(struct bio *bp) 928{ 929 struct g_mirror_softc *sc; 930 struct g_mirror_disk *disk; 931 struct bio *pbp; 932 933 g_topology_assert_not(); 934 935 pbp = bp->bio_parent; 936 sc = pbp->bio_to->private; 937 bp->bio_from->index--; 938 if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) 939 sc->sc_writes--; 940 disk = bp->bio_from->private; 941 if (disk == NULL) { 942 g_topology_lock(); 943 g_mirror_kill_consumer(sc, bp->bio_from); 944 g_topology_unlock(); 945 } 946 947 if (bp->bio_cmd == BIO_READ) 948 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_read, 949 bp->bio_error); 950 else if (bp->bio_cmd == BIO_WRITE) 951 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_write, 952 bp->bio_error); 953 954 pbp->bio_inbed++; 955 KASSERT(pbp->bio_inbed <= pbp->bio_children, 956 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 957 pbp->bio_children)); 958 if (bp->bio_error == 0 && pbp->bio_error == 0) { 959 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 960 g_destroy_bio(bp); 961 if (pbp->bio_children == pbp->bio_inbed) { 962 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 963 pbp->bio_completed = pbp->bio_length; 964 if (pbp->bio_cmd == BIO_WRITE || 965 pbp->bio_cmd == BIO_DELETE) { 966 TAILQ_REMOVE(&sc->sc_inflight, pbp, bio_queue); 967 /* Release delayed sync requests if possible. */ 968 g_mirror_sync_release(sc); 969 } 970 g_io_deliver(pbp, pbp->bio_error); 971 } 972 return; 973 } else if (bp->bio_error != 0) { 974 if (pbp->bio_error == 0) 975 pbp->bio_error = bp->bio_error; 976 if (disk != NULL) { 977 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 978 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 979 G_MIRROR_LOGREQ(0, bp, 980 "Request failed (error=%d).", 981 bp->bio_error); 982 } else { 983 G_MIRROR_LOGREQ(1, bp, 984 "Request failed (error=%d).", 985 bp->bio_error); 986 } 987 if (g_mirror_disconnect_on_failure && 988 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) 989 { 990 if (bp->bio_error == ENXIO && 991 bp->bio_cmd == BIO_READ) 992 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 993 else if (bp->bio_error == ENXIO) 994 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_NOW; 995 else 996 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 997 g_mirror_event_send(disk, 998 G_MIRROR_DISK_STATE_DISCONNECTED, 999 G_MIRROR_EVENT_DONTWAIT); 1000 } 1001 } 1002 switch (pbp->bio_cmd) { 1003 case BIO_DELETE: 1004 case BIO_WRITE: 1005 pbp->bio_inbed--; 1006 pbp->bio_children--; 1007 break; 1008 } 1009 } 1010 g_destroy_bio(bp); 1011 1012 switch (pbp->bio_cmd) { 1013 case BIO_READ: 1014 if (pbp->bio_inbed < pbp->bio_children) 1015 break; 1016 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 1017 g_io_deliver(pbp, pbp->bio_error); 1018 else { 1019 pbp->bio_error = 0; 1020 mtx_lock(&sc->sc_queue_mtx); 1021 TAILQ_INSERT_TAIL(&sc->sc_queue, pbp, bio_queue); 1022 mtx_unlock(&sc->sc_queue_mtx); 1023 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1024 wakeup(sc); 1025 } 1026 break; 1027 case BIO_DELETE: 1028 case BIO_WRITE: 1029 if (pbp->bio_children == 0) { 1030 /* 1031 * All requests failed. 1032 */ 1033 } else if (pbp->bio_inbed < pbp->bio_children) { 1034 /* Do nothing. */ 1035 break; 1036 } else if (pbp->bio_children == pbp->bio_inbed) { 1037 /* Some requests succeeded. */ 1038 pbp->bio_error = 0; 1039 pbp->bio_completed = pbp->bio_length; 1040 } 1041 TAILQ_REMOVE(&sc->sc_inflight, pbp, bio_queue); 1042 /* Release delayed sync requests if possible. */ 1043 g_mirror_sync_release(sc); 1044 g_io_deliver(pbp, pbp->bio_error); 1045 break; 1046 default: 1047 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 1048 break; 1049 } 1050} 1051 1052static void 1053g_mirror_sync_done(struct bio *bp) 1054{ 1055 struct g_mirror_softc *sc; 1056 1057 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1058 sc = bp->bio_from->geom->softc; 1059 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; 1060 mtx_lock(&sc->sc_queue_mtx); 1061 TAILQ_INSERT_TAIL(&sc->sc_queue, bp, bio_queue); 1062 mtx_unlock(&sc->sc_queue_mtx); 1063 wakeup(sc); 1064} 1065 1066static void 1067g_mirror_candelete(struct bio *bp) 1068{ 1069 struct g_mirror_softc *sc; 1070 struct g_mirror_disk *disk; 1071 int *val; 1072 1073 sc = bp->bio_to->private; 1074 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1075 if (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) 1076 break; 1077 } 1078 val = (int *)bp->bio_data; 1079 *val = (disk != NULL); 1080 g_io_deliver(bp, 0); 1081} 1082 1083static void 1084g_mirror_kernel_dump(struct bio *bp) 1085{ 1086 struct g_mirror_softc *sc; 1087 struct g_mirror_disk *disk; 1088 struct bio *cbp; 1089 struct g_kerneldump *gkd; 1090 1091 /* 1092 * We configure dumping to the first component, because this component 1093 * will be used for reading with 'prefer' balance algorithm. 1094 * If the component with the highest priority is currently disconnected 1095 * we will not be able to read the dump after the reboot if it will be 1096 * connected and synchronized later. Can we do something better? 1097 */ 1098 sc = bp->bio_to->private; 1099 disk = LIST_FIRST(&sc->sc_disks); 1100 1101 gkd = (struct g_kerneldump *)bp->bio_data; 1102 if (gkd->length > bp->bio_to->mediasize) 1103 gkd->length = bp->bio_to->mediasize; 1104 cbp = g_clone_bio(bp); 1105 if (cbp == NULL) { 1106 g_io_deliver(bp, ENOMEM); 1107 return; 1108 } 1109 cbp->bio_done = g_std_done; 1110 g_io_request(cbp, disk->d_consumer); 1111 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1112 g_mirror_get_diskname(disk)); 1113} 1114 1115static void 1116g_mirror_flush(struct g_mirror_softc *sc, struct bio *bp) 1117{ 1118 struct bio_queue queue; 1119 struct g_mirror_disk *disk; 1120 struct g_consumer *cp; 1121 struct bio *cbp; 1122 1123 TAILQ_INIT(&queue); 1124 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1125 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1126 continue; 1127 cbp = g_clone_bio(bp); 1128 if (cbp == NULL) { 1129 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1130 TAILQ_REMOVE(&queue, cbp, bio_queue); 1131 g_destroy_bio(cbp); 1132 } 1133 if (bp->bio_error == 0) 1134 bp->bio_error = ENOMEM; 1135 g_io_deliver(bp, bp->bio_error); 1136 return; 1137 } 1138 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 1139 cbp->bio_done = g_mirror_flush_done; 1140 cbp->bio_caller1 = disk; 1141 cbp->bio_to = disk->d_consumer->provider; 1142 } 1143 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1144 TAILQ_REMOVE(&queue, cbp, bio_queue); 1145 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1146 disk = cbp->bio_caller1; 1147 cbp->bio_caller1 = NULL; 1148 cp = disk->d_consumer; 1149 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1150 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1151 cp->acr, cp->acw, cp->ace)); 1152 g_io_request(cbp, disk->d_consumer); 1153 } 1154} 1155 1156static void 1157g_mirror_start(struct bio *bp) 1158{ 1159 struct g_mirror_softc *sc; 1160 1161 sc = bp->bio_to->private; 1162 /* 1163 * If sc == NULL or there are no valid disks, provider's error 1164 * should be set and g_mirror_start() should not be called at all. 1165 */ 1166 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1167 ("Provider's error should be set (error=%d)(mirror=%s).", 1168 bp->bio_to->error, bp->bio_to->name)); 1169 G_MIRROR_LOGREQ(3, bp, "Request received."); 1170 1171 switch (bp->bio_cmd) { 1172 case BIO_READ: 1173 case BIO_WRITE: 1174 case BIO_DELETE: 1175 break; 1176 case BIO_FLUSH: 1177 g_mirror_flush(sc, bp); 1178 return; 1179 case BIO_GETATTR: 1180 if (!strcmp(bp->bio_attribute, "GEOM::candelete")) { 1181 g_mirror_candelete(bp); 1182 return; 1183 } else if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1184 g_mirror_kernel_dump(bp); 1185 return; 1186 } 1187 /* FALLTHROUGH */ 1188 default: 1189 g_io_deliver(bp, EOPNOTSUPP); 1190 return; 1191 } 1192 mtx_lock(&sc->sc_queue_mtx); 1193 if (bp->bio_to->error != 0) { 1194 mtx_unlock(&sc->sc_queue_mtx); 1195 g_io_deliver(bp, bp->bio_to->error); 1196 return; 1197 } 1198 TAILQ_INSERT_TAIL(&sc->sc_queue, bp, bio_queue); 1199 mtx_unlock(&sc->sc_queue_mtx); 1200 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1201 wakeup(sc); 1202} 1203 1204/* 1205 * Return TRUE if the given request is colliding with a in-progress 1206 * synchronization request. 1207 */ 1208static bool 1209g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1210{ 1211 struct g_mirror_disk *disk; 1212 struct bio *sbp; 1213 off_t rstart, rend, sstart, send; 1214 u_int i; 1215 1216 if (sc->sc_sync.ds_ndisks == 0) 1217 return (false); 1218 rstart = bp->bio_offset; 1219 rend = bp->bio_offset + bp->bio_length; 1220 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1221 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1222 continue; 1223 for (i = 0; i < g_mirror_syncreqs; i++) { 1224 sbp = disk->d_sync.ds_bios[i]; 1225 if (sbp == NULL) 1226 continue; 1227 sstart = sbp->bio_offset; 1228 send = sbp->bio_offset + sbp->bio_length; 1229 if (rend > sstart && rstart < send) 1230 return (true); 1231 } 1232 } 1233 return (false); 1234} 1235 1236/* 1237 * Return TRUE if the given sync request is colliding with a in-progress regular 1238 * request. 1239 */ 1240static bool 1241g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1242{ 1243 off_t rstart, rend, sstart, send; 1244 struct bio *bp; 1245 1246 if (sc->sc_sync.ds_ndisks == 0) 1247 return (false); 1248 sstart = sbp->bio_offset; 1249 send = sbp->bio_offset + sbp->bio_length; 1250 TAILQ_FOREACH(bp, &sc->sc_inflight, bio_queue) { 1251 rstart = bp->bio_offset; 1252 rend = bp->bio_offset + bp->bio_length; 1253 if (rend > sstart && rstart < send) 1254 return (true); 1255 } 1256 return (false); 1257} 1258 1259/* 1260 * Puts request onto delayed queue. 1261 */ 1262static void 1263g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1264{ 1265 1266 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1267 TAILQ_INSERT_HEAD(&sc->sc_regular_delayed, bp, bio_queue); 1268} 1269 1270/* 1271 * Puts synchronization request onto delayed queue. 1272 */ 1273static void 1274g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1275{ 1276 1277 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1278 TAILQ_INSERT_TAIL(&sc->sc_sync_delayed, bp, bio_queue); 1279} 1280 1281/* 1282 * Releases delayed regular requests which don't collide anymore with sync 1283 * requests. 1284 */ 1285static void 1286g_mirror_regular_release(struct g_mirror_softc *sc) 1287{ 1288 struct bio *bp, *bp2; 1289 1290 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed, bio_queue, bp2) { 1291 if (g_mirror_sync_collision(sc, bp)) 1292 continue; 1293 TAILQ_REMOVE(&sc->sc_regular_delayed, bp, bio_queue); 1294 G_MIRROR_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1295 mtx_lock(&sc->sc_queue_mtx); 1296 TAILQ_INSERT_HEAD(&sc->sc_queue, bp, bio_queue); 1297 mtx_unlock(&sc->sc_queue_mtx); 1298 } 1299} 1300 1301/* 1302 * Releases delayed sync requests which don't collide anymore with regular 1303 * requests. 1304 */ 1305static void 1306g_mirror_sync_release(struct g_mirror_softc *sc) 1307{ 1308 struct bio *bp, *bp2; 1309 1310 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed, bio_queue, bp2) { 1311 if (g_mirror_regular_collision(sc, bp)) 1312 continue; 1313 TAILQ_REMOVE(&sc->sc_sync_delayed, bp, bio_queue); 1314 G_MIRROR_LOGREQ(2, bp, 1315 "Releasing delayed synchronization request."); 1316 g_io_request(bp, bp->bio_from); 1317 } 1318} 1319 1320/* 1321 * Free a synchronization request and clear its slot in the array. 1322 */ 1323static void 1324g_mirror_sync_request_free(struct g_mirror_disk *disk, struct bio *bp) 1325{ 1326 int idx; 1327 1328 if (disk != NULL && disk->d_sync.ds_bios != NULL) { 1329 idx = (int)(uintptr_t)bp->bio_caller1; 1330 KASSERT(disk->d_sync.ds_bios[idx] == bp, 1331 ("unexpected sync BIO at %p:%d", disk, idx)); 1332 disk->d_sync.ds_bios[idx] = NULL; 1333 } 1334 free(bp->bio_data, M_MIRROR); 1335 g_destroy_bio(bp); 1336} 1337 1338/* 1339 * Handle synchronization requests. 1340 * Every synchronization request is two-steps process: first, READ request is 1341 * send to active provider and then WRITE request (with read data) to the provider 1342 * being synchronized. When WRITE is finished, new synchronization request is 1343 * send. 1344 */ 1345static void 1346g_mirror_sync_request(struct bio *bp) 1347{ 1348 struct g_mirror_softc *sc; 1349 struct g_mirror_disk *disk; 1350 struct g_mirror_disk_sync *sync; 1351 1352 bp->bio_from->index--; 1353 sc = bp->bio_from->geom->softc; 1354 disk = bp->bio_from->private; 1355 if (disk == NULL) { 1356 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1357 g_topology_lock(); 1358 g_mirror_kill_consumer(sc, bp->bio_from); 1359 g_topology_unlock(); 1360 g_mirror_sync_request_free(NULL, bp); 1361 sx_xlock(&sc->sc_lock); 1362 return; 1363 } 1364 1365 /* 1366 * Synchronization request. 1367 */ 1368 switch (bp->bio_cmd) { 1369 case BIO_READ: 1370 { 1371 struct g_consumer *cp; 1372 1373 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_read, 1374 bp->bio_error); 1375 1376 if (bp->bio_error != 0) { 1377 G_MIRROR_LOGREQ(0, bp, 1378 "Synchronization request failed (error=%d).", 1379 bp->bio_error); 1380 g_mirror_sync_request_free(disk, bp); 1381 return; 1382 } 1383 G_MIRROR_LOGREQ(3, bp, 1384 "Synchronization request half-finished."); 1385 bp->bio_cmd = BIO_WRITE; 1386 bp->bio_cflags = 0; 1387 cp = disk->d_consumer; 1388 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1389 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1390 cp->acr, cp->acw, cp->ace)); 1391 cp->index++; 1392 g_io_request(bp, cp); 1393 return; 1394 } 1395 case BIO_WRITE: 1396 { 1397 off_t offset; 1398 void *data; 1399 int i, idx; 1400 1401 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_write, 1402 bp->bio_error); 1403 1404 if (bp->bio_error != 0) { 1405 G_MIRROR_LOGREQ(0, bp, 1406 "Synchronization request failed (error=%d).", 1407 bp->bio_error); 1408 g_mirror_sync_request_free(disk, bp); 1409 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1410 g_mirror_event_send(disk, 1411 G_MIRROR_DISK_STATE_DISCONNECTED, 1412 G_MIRROR_EVENT_DONTWAIT); 1413 return; 1414 } 1415 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1416 sync = &disk->d_sync; 1417 if (sync->ds_offset >= sc->sc_mediasize || 1418 sync->ds_consumer == NULL || 1419 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1420 /* Don't send more synchronization requests. */ 1421 sync->ds_inflight--; 1422 g_mirror_sync_request_free(disk, bp); 1423 if (sync->ds_inflight > 0) 1424 return; 1425 if (sync->ds_consumer == NULL || 1426 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1427 return; 1428 } 1429 /* Disk up-to-date, activate it. */ 1430 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1431 G_MIRROR_EVENT_DONTWAIT); 1432 return; 1433 } 1434 1435 /* Send next synchronization request. */ 1436 data = bp->bio_data; 1437 idx = (int)(uintptr_t)bp->bio_caller1; 1438 g_reset_bio(bp); 1439 bp->bio_cmd = BIO_READ; 1440 bp->bio_offset = sync->ds_offset; 1441 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1442 sync->ds_offset += bp->bio_length; 1443 bp->bio_done = g_mirror_sync_done; 1444 bp->bio_data = data; 1445 bp->bio_from = sync->ds_consumer; 1446 bp->bio_to = sc->sc_provider; 1447 bp->bio_caller1 = (void *)(uintptr_t)idx; 1448 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1449 sync->ds_consumer->index++; 1450 /* 1451 * Delay the request if it is colliding with a regular request. 1452 */ 1453 if (g_mirror_regular_collision(sc, bp)) 1454 g_mirror_sync_delay(sc, bp); 1455 else 1456 g_io_request(bp, sync->ds_consumer); 1457 1458 /* Release delayed requests if possible. */ 1459 g_mirror_regular_release(sc); 1460 1461 /* Find the smallest offset */ 1462 offset = sc->sc_mediasize; 1463 for (i = 0; i < g_mirror_syncreqs; i++) { 1464 bp = sync->ds_bios[i]; 1465 if (bp != NULL && bp->bio_offset < offset) 1466 offset = bp->bio_offset; 1467 } 1468 if (g_mirror_sync_period > 0 && 1469 time_uptime - sync->ds_update_ts > g_mirror_sync_period) { 1470 sync->ds_offset_done = offset; 1471 g_mirror_update_metadata(disk); 1472 sync->ds_update_ts = time_uptime; 1473 } 1474 return; 1475 } 1476 default: 1477 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1478 bp->bio_cmd, sc->sc_name)); 1479 break; 1480 } 1481} 1482 1483static void 1484g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1485{ 1486 struct g_mirror_disk *disk; 1487 struct g_consumer *cp; 1488 struct bio *cbp; 1489 1490 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1491 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1492 break; 1493 } 1494 if (disk == NULL) { 1495 if (bp->bio_error == 0) 1496 bp->bio_error = ENXIO; 1497 g_io_deliver(bp, bp->bio_error); 1498 return; 1499 } 1500 cbp = g_clone_bio(bp); 1501 if (cbp == NULL) { 1502 if (bp->bio_error == 0) 1503 bp->bio_error = ENOMEM; 1504 g_io_deliver(bp, bp->bio_error); 1505 return; 1506 } 1507 /* 1508 * Fill in the component buf structure. 1509 */ 1510 cp = disk->d_consumer; 1511 cbp->bio_done = g_mirror_done; 1512 cbp->bio_to = cp->provider; 1513 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1514 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1515 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1516 cp->acw, cp->ace)); 1517 cp->index++; 1518 g_io_request(cbp, cp); 1519} 1520 1521static void 1522g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1523{ 1524 struct g_mirror_disk *disk; 1525 struct g_consumer *cp; 1526 struct bio *cbp; 1527 1528 disk = g_mirror_get_disk(sc); 1529 if (disk == NULL) { 1530 if (bp->bio_error == 0) 1531 bp->bio_error = ENXIO; 1532 g_io_deliver(bp, bp->bio_error); 1533 return; 1534 } 1535 cbp = g_clone_bio(bp); 1536 if (cbp == NULL) { 1537 if (bp->bio_error == 0) 1538 bp->bio_error = ENOMEM; 1539 g_io_deliver(bp, bp->bio_error); 1540 return; 1541 } 1542 /* 1543 * Fill in the component buf structure. 1544 */ 1545 cp = disk->d_consumer; 1546 cbp->bio_done = g_mirror_done; 1547 cbp->bio_to = cp->provider; 1548 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1549 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1550 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1551 cp->acw, cp->ace)); 1552 cp->index++; 1553 g_io_request(cbp, cp); 1554} 1555 1556#define TRACK_SIZE (1 * 1024 * 1024) 1557#define LOAD_SCALE 256 1558#define ABS(x) (((x) >= 0) ? (x) : (-(x))) 1559 1560static void 1561g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1562{ 1563 struct g_mirror_disk *disk, *dp; 1564 struct g_consumer *cp; 1565 struct bio *cbp; 1566 int prio, best; 1567 1568 /* Find a disk with the smallest load. */ 1569 disk = NULL; 1570 best = INT_MAX; 1571 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1572 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1573 continue; 1574 prio = dp->load; 1575 /* If disk head is precisely in position - highly prefer it. */ 1576 if (dp->d_last_offset == bp->bio_offset) 1577 prio -= 2 * LOAD_SCALE; 1578 else 1579 /* If disk head is close to position - prefer it. */ 1580 if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE) 1581 prio -= 1 * LOAD_SCALE; 1582 if (prio <= best) { 1583 disk = dp; 1584 best = prio; 1585 } 1586 } 1587 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1588 cbp = g_clone_bio(bp); 1589 if (cbp == NULL) { 1590 if (bp->bio_error == 0) 1591 bp->bio_error = ENOMEM; 1592 g_io_deliver(bp, bp->bio_error); 1593 return; 1594 } 1595 /* 1596 * Fill in the component buf structure. 1597 */ 1598 cp = disk->d_consumer; 1599 cbp->bio_done = g_mirror_done; 1600 cbp->bio_to = cp->provider; 1601 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1602 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1603 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1604 cp->acw, cp->ace)); 1605 cp->index++; 1606 /* Remember last head position */ 1607 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1608 /* Update loads. */ 1609 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1610 dp->load = (dp->d_consumer->index * LOAD_SCALE + 1611 dp->load * 7) / 8; 1612 } 1613 g_io_request(cbp, cp); 1614} 1615 1616static void 1617g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1618{ 1619 struct bio_queue queue; 1620 struct g_mirror_disk *disk; 1621 struct g_consumer *cp; 1622 struct bio *cbp; 1623 off_t left, mod, offset, slice; 1624 u_char *data; 1625 u_int ndisks; 1626 1627 if (bp->bio_length <= sc->sc_slice) { 1628 g_mirror_request_round_robin(sc, bp); 1629 return; 1630 } 1631 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1632 slice = bp->bio_length / ndisks; 1633 mod = slice % sc->sc_provider->sectorsize; 1634 if (mod != 0) 1635 slice += sc->sc_provider->sectorsize - mod; 1636 /* 1637 * Allocate all bios before sending any request, so we can 1638 * return ENOMEM in nice and clean way. 1639 */ 1640 left = bp->bio_length; 1641 offset = bp->bio_offset; 1642 data = bp->bio_data; 1643 TAILQ_INIT(&queue); 1644 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1645 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1646 continue; 1647 cbp = g_clone_bio(bp); 1648 if (cbp == NULL) { 1649 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1650 TAILQ_REMOVE(&queue, cbp, bio_queue); 1651 g_destroy_bio(cbp); 1652 } 1653 if (bp->bio_error == 0) 1654 bp->bio_error = ENOMEM; 1655 g_io_deliver(bp, bp->bio_error); 1656 return; 1657 } 1658 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 1659 cbp->bio_done = g_mirror_done; 1660 cbp->bio_caller1 = disk; 1661 cbp->bio_to = disk->d_consumer->provider; 1662 cbp->bio_offset = offset; 1663 cbp->bio_data = data; 1664 cbp->bio_length = MIN(left, slice); 1665 left -= cbp->bio_length; 1666 if (left == 0) 1667 break; 1668 offset += cbp->bio_length; 1669 data += cbp->bio_length; 1670 } 1671 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1672 TAILQ_REMOVE(&queue, cbp, bio_queue); 1673 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1674 disk = cbp->bio_caller1; 1675 cbp->bio_caller1 = NULL; 1676 cp = disk->d_consumer; 1677 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1678 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1679 cp->acr, cp->acw, cp->ace)); 1680 disk->d_consumer->index++; 1681 g_io_request(cbp, disk->d_consumer); 1682 } 1683} 1684 1685static void 1686g_mirror_register_request(struct bio *bp) 1687{ 1688 struct g_mirror_softc *sc; 1689 1690 sc = bp->bio_to->private; 1691 switch (bp->bio_cmd) { 1692 case BIO_READ: 1693 switch (sc->sc_balance) { 1694 case G_MIRROR_BALANCE_LOAD: 1695 g_mirror_request_load(sc, bp); 1696 break; 1697 case G_MIRROR_BALANCE_PREFER: 1698 g_mirror_request_prefer(sc, bp); 1699 break; 1700 case G_MIRROR_BALANCE_ROUND_ROBIN: 1701 g_mirror_request_round_robin(sc, bp); 1702 break; 1703 case G_MIRROR_BALANCE_SPLIT: 1704 g_mirror_request_split(sc, bp); 1705 break; 1706 } 1707 return; 1708 case BIO_WRITE: 1709 case BIO_DELETE: 1710 { 1711 struct bio_queue queue; 1712 struct g_mirror_disk *disk; 1713 struct g_mirror_disk_sync *sync; 1714 struct g_consumer *cp; 1715 struct bio *cbp; 1716 1717 /* 1718 * Delay the request if it is colliding with a synchronization 1719 * request. 1720 */ 1721 if (g_mirror_sync_collision(sc, bp)) { 1722 g_mirror_regular_delay(sc, bp); 1723 return; 1724 } 1725 1726 if (sc->sc_idle) 1727 g_mirror_unidle(sc); 1728 else 1729 sc->sc_last_write = time_uptime; 1730 1731 /* 1732 * Bump syncid on first write. 1733 */ 1734 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1735 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1736 g_mirror_bump_syncid(sc); 1737 } 1738 1739 /* 1740 * Allocate all bios before sending any request, so we can 1741 * return ENOMEM in nice and clean way. 1742 */ 1743 TAILQ_INIT(&queue); 1744 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1745 sync = &disk->d_sync; 1746 switch (disk->d_state) { 1747 case G_MIRROR_DISK_STATE_ACTIVE: 1748 break; 1749 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1750 if (bp->bio_offset >= sync->ds_offset) 1751 continue; 1752 break; 1753 default: 1754 continue; 1755 } 1756 if (bp->bio_cmd == BIO_DELETE && 1757 (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0) 1758 continue; 1759 cbp = g_clone_bio(bp); 1760 if (cbp == NULL) { 1761 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1762 TAILQ_REMOVE(&queue, cbp, bio_queue); 1763 g_destroy_bio(cbp); 1764 } 1765 if (bp->bio_error == 0) 1766 bp->bio_error = ENOMEM; 1767 g_io_deliver(bp, bp->bio_error); 1768 return; 1769 } 1770 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue); 1771 cbp->bio_done = g_mirror_done; 1772 cp = disk->d_consumer; 1773 cbp->bio_caller1 = cp; 1774 cbp->bio_to = cp->provider; 1775 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1776 ("Consumer %s not opened (r%dw%de%d).", 1777 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1778 } 1779 if (TAILQ_EMPTY(&queue)) { 1780 g_io_deliver(bp, EOPNOTSUPP); 1781 return; 1782 } 1783 while ((cbp = TAILQ_FIRST(&queue)) != NULL) { 1784 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1785 TAILQ_REMOVE(&queue, cbp, bio_queue); 1786 cp = cbp->bio_caller1; 1787 cbp->bio_caller1 = NULL; 1788 cp->index++; 1789 sc->sc_writes++; 1790 g_io_request(cbp, cp); 1791 } 1792 /* 1793 * Put request onto inflight queue, so we can check if new 1794 * synchronization requests don't collide with it. 1795 */ 1796 TAILQ_INSERT_TAIL(&sc->sc_inflight, bp, bio_queue); 1797 return; 1798 } 1799 default: 1800 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1801 bp->bio_cmd, sc->sc_name)); 1802 break; 1803 } 1804} 1805 1806static int 1807g_mirror_can_destroy(struct g_mirror_softc *sc) 1808{ 1809 struct g_geom *gp; 1810 struct g_consumer *cp; 1811 1812 g_topology_assert(); 1813 gp = sc->sc_geom; 1814 if (gp->softc == NULL) 1815 return (1); 1816 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0) 1817 return (0); 1818 LIST_FOREACH(cp, &gp->consumer, consumer) { 1819 if (g_mirror_is_busy(sc, cp)) 1820 return (0); 1821 } 1822 gp = sc->sc_sync.ds_geom; 1823 LIST_FOREACH(cp, &gp->consumer, consumer) { 1824 if (g_mirror_is_busy(sc, cp)) 1825 return (0); 1826 } 1827 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1828 sc->sc_name); 1829 return (1); 1830} 1831 1832static int 1833g_mirror_try_destroy(struct g_mirror_softc *sc) 1834{ 1835 1836 if (sc->sc_rootmount != NULL) { 1837 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1838 sc->sc_rootmount); 1839 root_mount_rel(sc->sc_rootmount); 1840 sc->sc_rootmount = NULL; 1841 } 1842 g_topology_lock(); 1843 if (!g_mirror_can_destroy(sc)) { 1844 g_topology_unlock(); 1845 return (0); 1846 } 1847 sc->sc_geom->softc = NULL; 1848 sc->sc_sync.ds_geom->softc = NULL; 1849 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DRAIN) != 0) { 1850 g_topology_unlock(); 1851 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1852 &sc->sc_worker); 1853 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1854 sx_xunlock(&sc->sc_lock); 1855 wakeup(&sc->sc_worker); 1856 sc->sc_worker = NULL; 1857 } else { 1858 g_topology_unlock(); 1859 g_mirror_destroy_device(sc); 1860 } 1861 return (1); 1862} 1863 1864/* 1865 * Worker thread. 1866 */ 1867static void 1868g_mirror_worker(void *arg) 1869{ 1870 struct g_mirror_softc *sc; 1871 struct g_mirror_event *ep; 1872 struct bio *bp; 1873 int timeout; 1874 1875 sc = arg; 1876 thread_lock(curthread); 1877 sched_prio(curthread, PRIBIO); 1878 thread_unlock(curthread); 1879 1880 sx_xlock(&sc->sc_lock); 1881 for (;;) { 1882 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1883 /* 1884 * First take a look at events. 1885 * This is important to handle events before any I/O requests. 1886 */ 1887 ep = g_mirror_event_first(sc); 1888 if (ep != NULL) { 1889 g_mirror_event_remove(sc, ep); 1890 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1891 /* Update only device status. */ 1892 G_MIRROR_DEBUG(3, 1893 "Running event for device %s.", 1894 sc->sc_name); 1895 ep->e_error = 0; 1896 g_mirror_update_device(sc, true); 1897 } else { 1898 /* Update disk status. */ 1899 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1900 g_mirror_get_diskname(ep->e_disk)); 1901 ep->e_error = g_mirror_update_disk(ep->e_disk, 1902 ep->e_state); 1903 if (ep->e_error == 0) 1904 g_mirror_update_device(sc, false); 1905 } 1906 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1907 KASSERT(ep->e_error == 0, 1908 ("Error cannot be handled.")); 1909 g_mirror_event_free(ep); 1910 } else { 1911 ep->e_flags |= G_MIRROR_EVENT_DONE; 1912 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1913 ep); 1914 mtx_lock(&sc->sc_events_mtx); 1915 wakeup(ep); 1916 mtx_unlock(&sc->sc_events_mtx); 1917 } 1918 if ((sc->sc_flags & 1919 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1920 if (g_mirror_try_destroy(sc)) { 1921 curthread->td_pflags &= ~TDP_GEOM; 1922 G_MIRROR_DEBUG(1, "Thread exiting."); 1923 kproc_exit(0); 1924 } 1925 } 1926 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1927 continue; 1928 } 1929 /* 1930 * Check if we can mark array as CLEAN and if we can't take 1931 * how much seconds should we wait. 1932 */ 1933 timeout = g_mirror_idle(sc, -1); 1934 /* 1935 * Now I/O requests. 1936 */ 1937 /* Get first request from the queue. */ 1938 mtx_lock(&sc->sc_queue_mtx); 1939 bp = TAILQ_FIRST(&sc->sc_queue); 1940 if (bp != NULL) 1941 TAILQ_REMOVE(&sc->sc_queue, bp, bio_queue); 1942 else { 1943 if ((sc->sc_flags & 1944 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1945 mtx_unlock(&sc->sc_queue_mtx); 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 mtx_lock(&sc->sc_queue_mtx); 1952 if (!TAILQ_EMPTY(&sc->sc_queue)) { 1953 mtx_unlock(&sc->sc_queue_mtx); 1954 continue; 1955 } 1956 } 1957 if (g_mirror_event_first(sc) != NULL) { 1958 mtx_unlock(&sc->sc_queue_mtx); 1959 continue; 1960 } 1961 sx_xunlock(&sc->sc_lock); 1962 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1963 timeout * hz); 1964 sx_xlock(&sc->sc_lock); 1965 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1966 continue; 1967 } 1968 mtx_unlock(&sc->sc_queue_mtx); 1969 1970 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 1971 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1972 g_mirror_sync_request(bp); /* READ */ 1973 } else if (bp->bio_to != sc->sc_provider) { 1974 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 1975 g_mirror_regular_request(bp); 1976 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 1977 g_mirror_sync_request(bp); /* WRITE */ 1978 else { 1979 KASSERT(0, 1980 ("Invalid request cflags=0x%hx to=%s.", 1981 bp->bio_cflags, bp->bio_to->name)); 1982 } 1983 } else { 1984 g_mirror_register_request(bp); 1985 } 1986 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1987 } 1988} 1989 1990static void 1991g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 1992{ 1993 1994 sx_assert(&sc->sc_lock, SX_LOCKED); 1995 1996 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 1997 return; 1998 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1999 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 2000 g_mirror_get_diskname(disk), sc->sc_name); 2001 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2002 } else if (sc->sc_idle && 2003 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2004 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 2005 g_mirror_get_diskname(disk), sc->sc_name); 2006 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2007 } 2008} 2009 2010static void 2011g_mirror_sync_start(struct g_mirror_disk *disk) 2012{ 2013 struct g_mirror_softc *sc; 2014 struct g_consumer *cp; 2015 struct bio *bp; 2016 int error, i; 2017 2018 g_topology_assert_not(); 2019 sc = disk->d_softc; 2020 sx_assert(&sc->sc_lock, SX_LOCKED); 2021 2022 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2023 ("Disk %s is not marked for synchronization.", 2024 g_mirror_get_diskname(disk))); 2025 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2026 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 2027 sc->sc_state)); 2028 2029 sx_xunlock(&sc->sc_lock); 2030 g_topology_lock(); 2031 cp = g_new_consumer(sc->sc_sync.ds_geom); 2032 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 2033 error = g_attach(cp, sc->sc_provider); 2034 KASSERT(error == 0, 2035 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 2036 error = g_access(cp, 1, 0, 0); 2037 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 2038 g_topology_unlock(); 2039 sx_xlock(&sc->sc_lock); 2040 2041 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 2042 g_mirror_get_diskname(disk)); 2043 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0) 2044 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2045 KASSERT(disk->d_sync.ds_consumer == NULL, 2046 ("Sync consumer already exists (device=%s, disk=%s).", 2047 sc->sc_name, g_mirror_get_diskname(disk))); 2048 2049 disk->d_sync.ds_consumer = cp; 2050 disk->d_sync.ds_consumer->private = disk; 2051 disk->d_sync.ds_consumer->index = 0; 2052 2053 /* 2054 * Allocate memory for synchronization bios and initialize them. 2055 */ 2056 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 2057 M_MIRROR, M_WAITOK); 2058 for (i = 0; i < g_mirror_syncreqs; i++) { 2059 bp = g_alloc_bio(); 2060 disk->d_sync.ds_bios[i] = bp; 2061 bp->bio_parent = NULL; 2062 bp->bio_cmd = BIO_READ; 2063 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 2064 bp->bio_cflags = 0; 2065 bp->bio_offset = disk->d_sync.ds_offset; 2066 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 2067 disk->d_sync.ds_offset += bp->bio_length; 2068 bp->bio_done = g_mirror_sync_done; 2069 bp->bio_from = disk->d_sync.ds_consumer; 2070 bp->bio_to = sc->sc_provider; 2071 bp->bio_caller1 = (void *)(uintptr_t)i; 2072 } 2073 2074 /* Increase the number of disks in SYNCHRONIZING state. */ 2075 sc->sc_sync.ds_ndisks++; 2076 /* Set the number of in-flight synchronization requests. */ 2077 disk->d_sync.ds_inflight = g_mirror_syncreqs; 2078 2079 /* 2080 * Fire off first synchronization requests. 2081 */ 2082 for (i = 0; i < g_mirror_syncreqs; i++) { 2083 bp = disk->d_sync.ds_bios[i]; 2084 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 2085 disk->d_sync.ds_consumer->index++; 2086 /* 2087 * Delay the request if it is colliding with a regular request. 2088 */ 2089 if (g_mirror_regular_collision(sc, bp)) 2090 g_mirror_sync_delay(sc, bp); 2091 else 2092 g_io_request(bp, disk->d_sync.ds_consumer); 2093 } 2094} 2095 2096/* 2097 * Stop synchronization process. 2098 * type: 0 - synchronization finished 2099 * 1 - synchronization stopped 2100 */ 2101static void 2102g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 2103{ 2104 struct g_mirror_softc *sc; 2105 struct g_consumer *cp; 2106 2107 g_topology_assert_not(); 2108 sc = disk->d_softc; 2109 sx_assert(&sc->sc_lock, SX_LOCKED); 2110 2111 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2112 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2113 g_mirror_disk_state2str(disk->d_state))); 2114 if (disk->d_sync.ds_consumer == NULL) 2115 return; 2116 2117 if (type == 0) { 2118 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2119 sc->sc_name, g_mirror_get_diskname(disk)); 2120 } else /* if (type == 1) */ { 2121 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2122 sc->sc_name, g_mirror_get_diskname(disk)); 2123 } 2124 g_mirror_regular_release(sc); 2125 free(disk->d_sync.ds_bios, M_MIRROR); 2126 disk->d_sync.ds_bios = NULL; 2127 cp = disk->d_sync.ds_consumer; 2128 disk->d_sync.ds_consumer = NULL; 2129 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2130 sc->sc_sync.ds_ndisks--; 2131 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2132 g_topology_lock(); 2133 g_mirror_kill_consumer(sc, cp); 2134 g_topology_unlock(); 2135 sx_xlock(&sc->sc_lock); 2136} 2137 2138static void 2139g_mirror_launch_provider(struct g_mirror_softc *sc) 2140{ 2141 struct g_mirror_disk *disk; 2142 struct g_provider *pp, *dp; 2143 2144 sx_assert(&sc->sc_lock, SX_LOCKED); 2145 2146 g_topology_lock(); 2147 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 2148 pp->flags |= G_PF_DIRECT_RECEIVE; 2149 pp->mediasize = sc->sc_mediasize; 2150 pp->sectorsize = sc->sc_sectorsize; 2151 pp->stripesize = 0; 2152 pp->stripeoffset = 0; 2153 2154 /* Splitting of unmapped BIO's could work but isn't implemented now */ 2155 if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT) 2156 pp->flags |= G_PF_ACCEPT_UNMAPPED; 2157 2158 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2159 if (disk->d_consumer && disk->d_consumer->provider) { 2160 dp = disk->d_consumer->provider; 2161 if (dp->stripesize > pp->stripesize) { 2162 pp->stripesize = dp->stripesize; 2163 pp->stripeoffset = dp->stripeoffset; 2164 } 2165 /* A provider underneath us doesn't support unmapped */ 2166 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { 2167 G_MIRROR_DEBUG(0, "Cancelling unmapped " 2168 "because of %s.", dp->name); 2169 pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 2170 } 2171 } 2172 } 2173 pp->private = sc; 2174 sc->sc_refcnt++; 2175 sc->sc_provider = pp; 2176 g_error_provider(pp, 0); 2177 g_topology_unlock(); 2178 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2179 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 2180 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2181 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2182 g_mirror_sync_start(disk); 2183 } 2184} 2185 2186static void 2187g_mirror_destroy_provider(struct g_mirror_softc *sc) 2188{ 2189 struct g_mirror_disk *disk; 2190 struct bio *bp; 2191 2192 g_topology_assert_not(); 2193 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2194 sc->sc_name)); 2195 2196 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2197 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2198 g_mirror_sync_stop(disk, 1); 2199 } 2200 2201 g_topology_lock(); 2202 g_error_provider(sc->sc_provider, ENXIO); 2203 mtx_lock(&sc->sc_queue_mtx); 2204 while ((bp = TAILQ_FIRST(&sc->sc_queue)) != NULL) { 2205 TAILQ_REMOVE(&sc->sc_queue, bp, bio_queue); 2206 /* 2207 * Abort any pending I/O that wasn't generated by us. 2208 * Synchronization requests and requests destined for individual 2209 * mirror components can be destroyed immediately. 2210 */ 2211 if (bp->bio_to == sc->sc_provider && 2212 bp->bio_from->geom != sc->sc_sync.ds_geom) { 2213 g_io_deliver(bp, ENXIO); 2214 } else { 2215 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2216 free(bp->bio_data, M_MIRROR); 2217 g_destroy_bio(bp); 2218 } 2219 } 2220 mtx_unlock(&sc->sc_queue_mtx); 2221 g_wither_provider(sc->sc_provider, ENXIO); 2222 sc->sc_provider = NULL; 2223 G_MIRROR_DEBUG(0, "Device %s: provider destroyed.", sc->sc_name); 2224 g_topology_unlock(); 2225} 2226 2227static void 2228g_mirror_go(void *arg) 2229{ 2230 struct g_mirror_softc *sc; 2231 2232 sc = arg; 2233 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2234 g_mirror_event_send(sc, 0, 2235 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2236} 2237 2238static u_int 2239g_mirror_determine_state(struct g_mirror_disk *disk) 2240{ 2241 struct g_mirror_softc *sc; 2242 u_int state; 2243 2244 sc = disk->d_softc; 2245 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2246 if ((disk->d_flags & 2247 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0 && 2248 (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 || 2249 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0)) { 2250 /* Disk does not need synchronization. */ 2251 state = G_MIRROR_DISK_STATE_ACTIVE; 2252 } else { 2253 if ((sc->sc_flags & 2254 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2255 (disk->d_flags & 2256 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2257 /* 2258 * We can start synchronization from 2259 * the stored offset. 2260 */ 2261 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2262 } else { 2263 state = G_MIRROR_DISK_STATE_STALE; 2264 } 2265 } 2266 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2267 /* 2268 * Reset all synchronization data for this disk, 2269 * because if it even was synchronized, it was 2270 * synchronized to disks with different syncid. 2271 */ 2272 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2273 disk->d_sync.ds_offset = 0; 2274 disk->d_sync.ds_offset_done = 0; 2275 disk->d_sync.ds_syncid = sc->sc_syncid; 2276 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2277 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2278 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2279 } else { 2280 state = G_MIRROR_DISK_STATE_STALE; 2281 } 2282 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2283 /* 2284 * Not good, NOT GOOD! 2285 * It means that mirror was started on stale disks 2286 * and more fresh disk just arrive. 2287 * If there were writes, mirror is broken, sorry. 2288 * I think the best choice here is don't touch 2289 * this disk and inform the user loudly. 2290 */ 2291 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2292 "disk (%s) arrives!! It will not be connected to the " 2293 "running device.", sc->sc_name, 2294 g_mirror_get_diskname(disk)); 2295 g_mirror_destroy_disk(disk); 2296 state = G_MIRROR_DISK_STATE_NONE; 2297 /* Return immediately, because disk was destroyed. */ 2298 return (state); 2299 } 2300 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2301 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2302 return (state); 2303} 2304 2305/* 2306 * Update device state. 2307 */ 2308static void 2309g_mirror_update_device(struct g_mirror_softc *sc, bool force) 2310{ 2311 struct g_mirror_disk *disk; 2312 u_int state; 2313 2314 sx_assert(&sc->sc_lock, SX_XLOCKED); 2315 2316 switch (sc->sc_state) { 2317 case G_MIRROR_DEVICE_STATE_STARTING: 2318 { 2319 struct g_mirror_disk *pdisk, *tdisk; 2320 u_int dirty, ndisks, genid, syncid; 2321 bool broken; 2322 2323 KASSERT(sc->sc_provider == NULL, 2324 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2325 /* 2326 * Are we ready? We are, if all disks are connected or 2327 * if we have any disks and 'force' is true. 2328 */ 2329 ndisks = g_mirror_ndisks(sc, -1); 2330 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2331 ; 2332 } else if (ndisks == 0) { 2333 /* 2334 * Disks went down in starting phase, so destroy 2335 * device. 2336 */ 2337 callout_drain(&sc->sc_callout); 2338 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2339 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2340 sc->sc_rootmount); 2341 root_mount_rel(sc->sc_rootmount); 2342 sc->sc_rootmount = NULL; 2343 return; 2344 } else { 2345 return; 2346 } 2347 2348 /* 2349 * Activate all disks with the biggest syncid. 2350 */ 2351 if (force) { 2352 /* 2353 * If 'force' is true, we have been called due to 2354 * timeout, so don't bother canceling timeout. 2355 */ 2356 ndisks = 0; 2357 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2358 if ((disk->d_flags & 2359 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2360 ndisks++; 2361 } 2362 } 2363 if (ndisks == 0) { 2364 /* No valid disks found, destroy device. */ 2365 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2366 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2367 __LINE__, sc->sc_rootmount); 2368 root_mount_rel(sc->sc_rootmount); 2369 sc->sc_rootmount = NULL; 2370 return; 2371 } 2372 } else { 2373 /* Cancel timeout. */ 2374 callout_drain(&sc->sc_callout); 2375 } 2376 2377 /* 2378 * Find the biggest genid. 2379 */ 2380 genid = 0; 2381 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2382 if (disk->d_genid > genid) 2383 genid = disk->d_genid; 2384 } 2385 sc->sc_genid = genid; 2386 /* 2387 * Remove all disks without the biggest genid. 2388 */ 2389 broken = false; 2390 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2391 if (disk->d_genid < genid) { 2392 G_MIRROR_DEBUG(0, 2393 "Component %s (device %s) broken, skipping.", 2394 g_mirror_get_diskname(disk), sc->sc_name); 2395 g_mirror_destroy_disk(disk); 2396 /* 2397 * Bump the syncid in case we discover a healthy 2398 * replacement disk after starting the mirror. 2399 */ 2400 broken = true; 2401 } 2402 } 2403 2404 /* 2405 * Find the biggest syncid. 2406 */ 2407 syncid = 0; 2408 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2409 if (disk->d_sync.ds_syncid > syncid) 2410 syncid = disk->d_sync.ds_syncid; 2411 } 2412 2413 /* 2414 * Here we need to look for dirty disks and if all disks 2415 * with the biggest syncid are dirty, we have to choose 2416 * one with the biggest priority and rebuild the rest. 2417 */ 2418 /* 2419 * Find the number of dirty disks with the biggest syncid. 2420 * Find the number of disks with the biggest syncid. 2421 * While here, find a disk with the biggest priority. 2422 */ 2423 dirty = ndisks = 0; 2424 pdisk = NULL; 2425 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2426 if (disk->d_sync.ds_syncid != syncid) 2427 continue; 2428 if ((disk->d_flags & 2429 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2430 continue; 2431 } 2432 ndisks++; 2433 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2434 dirty++; 2435 if (pdisk == NULL || 2436 pdisk->d_priority < disk->d_priority) { 2437 pdisk = disk; 2438 } 2439 } 2440 } 2441 if (dirty == 0) { 2442 /* No dirty disks at all, great. */ 2443 } else if (dirty == ndisks) { 2444 /* 2445 * Force synchronization for all dirty disks except one 2446 * with the biggest priority. 2447 */ 2448 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2449 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2450 "master disk for synchronization.", 2451 g_mirror_get_diskname(pdisk), sc->sc_name); 2452 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2453 if (disk->d_sync.ds_syncid != syncid) 2454 continue; 2455 if ((disk->d_flags & 2456 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2457 continue; 2458 } 2459 KASSERT((disk->d_flags & 2460 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2461 ("Disk %s isn't marked as dirty.", 2462 g_mirror_get_diskname(disk))); 2463 /* Skip the disk with the biggest priority. */ 2464 if (disk == pdisk) 2465 continue; 2466 disk->d_sync.ds_syncid = 0; 2467 } 2468 } else if (dirty < ndisks) { 2469 /* 2470 * Force synchronization for all dirty disks. 2471 * We have some non-dirty disks. 2472 */ 2473 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2474 if (disk->d_sync.ds_syncid != syncid) 2475 continue; 2476 if ((disk->d_flags & 2477 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2478 continue; 2479 } 2480 if ((disk->d_flags & 2481 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2482 continue; 2483 } 2484 disk->d_sync.ds_syncid = 0; 2485 } 2486 } 2487 2488 /* Reset hint. */ 2489 sc->sc_hint = NULL; 2490 sc->sc_syncid = syncid; 2491 if (force || broken) { 2492 /* Remember to bump syncid on first write. */ 2493 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2494 } 2495 state = G_MIRROR_DEVICE_STATE_RUNNING; 2496 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2497 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2498 g_mirror_device_state2str(state)); 2499 sc->sc_state = state; 2500 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2501 state = g_mirror_determine_state(disk); 2502 g_mirror_event_send(disk, state, 2503 G_MIRROR_EVENT_DONTWAIT); 2504 if (state == G_MIRROR_DISK_STATE_STALE) 2505 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2506 } 2507 break; 2508 } 2509 case G_MIRROR_DEVICE_STATE_RUNNING: 2510 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2511 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2512 /* 2513 * No usable disks, so destroy the device. 2514 */ 2515 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2516 break; 2517 } else if (g_mirror_ndisks(sc, 2518 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2519 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2520 /* 2521 * We have active disks, launch provider if it doesn't 2522 * exist. 2523 */ 2524 if (sc->sc_provider == NULL) 2525 g_mirror_launch_provider(sc); 2526 if (sc->sc_rootmount != NULL) { 2527 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2528 __LINE__, sc->sc_rootmount); 2529 root_mount_rel(sc->sc_rootmount); 2530 sc->sc_rootmount = NULL; 2531 } 2532 } 2533 /* 2534 * Genid should be bumped immediately, so do it here. 2535 */ 2536 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2537 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2538 g_mirror_bump_genid(sc); 2539 } 2540 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_NOW) != 0) { 2541 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID_NOW; 2542 g_mirror_bump_syncid(sc); 2543 } 2544 break; 2545 default: 2546 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2547 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2548 break; 2549 } 2550} 2551 2552/* 2553 * Update disk state and device state if needed. 2554 */ 2555#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2556 "Disk %s state changed from %s to %s (device %s).", \ 2557 g_mirror_get_diskname(disk), \ 2558 g_mirror_disk_state2str(disk->d_state), \ 2559 g_mirror_disk_state2str(state), sc->sc_name) 2560static int 2561g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2562{ 2563 struct g_mirror_softc *sc; 2564 2565 sc = disk->d_softc; 2566 sx_assert(&sc->sc_lock, SX_XLOCKED); 2567 2568again: 2569 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2570 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2571 g_mirror_disk_state2str(state)); 2572 switch (state) { 2573 case G_MIRROR_DISK_STATE_NEW: 2574 /* 2575 * Possible scenarios: 2576 * 1. New disk arrive. 2577 */ 2578 /* Previous state should be NONE. */ 2579 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2580 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2581 g_mirror_disk_state2str(disk->d_state))); 2582 DISK_STATE_CHANGED(); 2583 2584 disk->d_state = state; 2585 if (LIST_EMPTY(&sc->sc_disks)) 2586 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2587 else { 2588 struct g_mirror_disk *dp; 2589 2590 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2591 if (disk->d_priority >= dp->d_priority) { 2592 LIST_INSERT_BEFORE(dp, disk, d_next); 2593 dp = NULL; 2594 break; 2595 } 2596 if (LIST_NEXT(dp, d_next) == NULL) 2597 break; 2598 } 2599 if (dp != NULL) 2600 LIST_INSERT_AFTER(dp, disk, d_next); 2601 } 2602 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2603 sc->sc_name, g_mirror_get_diskname(disk)); 2604 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2605 break; 2606 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2607 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2608 g_mirror_device_state2str(sc->sc_state), 2609 g_mirror_get_diskname(disk), 2610 g_mirror_disk_state2str(disk->d_state))); 2611 state = g_mirror_determine_state(disk); 2612 if (state != G_MIRROR_DISK_STATE_NONE) 2613 goto again; 2614 break; 2615 case G_MIRROR_DISK_STATE_ACTIVE: 2616 /* 2617 * Possible scenarios: 2618 * 1. New disk does not need synchronization. 2619 * 2. Synchronization process finished successfully. 2620 */ 2621 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2622 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2623 g_mirror_device_state2str(sc->sc_state), 2624 g_mirror_get_diskname(disk), 2625 g_mirror_disk_state2str(disk->d_state))); 2626 /* Previous state should be NEW or SYNCHRONIZING. */ 2627 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2628 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2629 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2630 g_mirror_disk_state2str(disk->d_state))); 2631 DISK_STATE_CHANGED(); 2632 2633 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2634 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2635 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2636 g_mirror_sync_stop(disk, 0); 2637 } 2638 disk->d_state = state; 2639 disk->d_sync.ds_offset = 0; 2640 disk->d_sync.ds_offset_done = 0; 2641 g_mirror_update_idle(sc, disk); 2642 g_mirror_update_metadata(disk); 2643 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2644 sc->sc_name, g_mirror_get_diskname(disk)); 2645 break; 2646 case G_MIRROR_DISK_STATE_STALE: 2647 /* 2648 * Possible scenarios: 2649 * 1. Stale disk was connected. 2650 */ 2651 /* Previous state should be NEW. */ 2652 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2653 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2654 g_mirror_disk_state2str(disk->d_state))); 2655 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2656 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2657 g_mirror_device_state2str(sc->sc_state), 2658 g_mirror_get_diskname(disk), 2659 g_mirror_disk_state2str(disk->d_state))); 2660 /* 2661 * STALE state is only possible if device is marked 2662 * NOAUTOSYNC. 2663 */ 2664 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2665 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2666 g_mirror_device_state2str(sc->sc_state), 2667 g_mirror_get_diskname(disk), 2668 g_mirror_disk_state2str(disk->d_state))); 2669 DISK_STATE_CHANGED(); 2670 2671 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2672 disk->d_state = state; 2673 g_mirror_update_metadata(disk); 2674 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2675 sc->sc_name, g_mirror_get_diskname(disk)); 2676 break; 2677 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2678 /* 2679 * Possible scenarios: 2680 * 1. Disk which needs synchronization was connected. 2681 */ 2682 /* Previous state should be NEW. */ 2683 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2684 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2685 g_mirror_disk_state2str(disk->d_state))); 2686 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2687 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2688 g_mirror_device_state2str(sc->sc_state), 2689 g_mirror_get_diskname(disk), 2690 g_mirror_disk_state2str(disk->d_state))); 2691 DISK_STATE_CHANGED(); 2692 2693 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2694 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2695 disk->d_state = state; 2696 if (sc->sc_provider != NULL) { 2697 g_mirror_sync_start(disk); 2698 g_mirror_update_metadata(disk); 2699 } 2700 break; 2701 case G_MIRROR_DISK_STATE_DISCONNECTED: 2702 /* 2703 * Possible scenarios: 2704 * 1. Device wasn't running yet, but disk disappear. 2705 * 2. Disk was active and disapppear. 2706 * 3. Disk disappear during synchronization process. 2707 */ 2708 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2709 /* 2710 * Previous state should be ACTIVE, STALE or 2711 * SYNCHRONIZING. 2712 */ 2713 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2714 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2715 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2716 ("Wrong disk state (%s, %s).", 2717 g_mirror_get_diskname(disk), 2718 g_mirror_disk_state2str(disk->d_state))); 2719 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2720 /* Previous state should be NEW. */ 2721 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2722 ("Wrong disk state (%s, %s).", 2723 g_mirror_get_diskname(disk), 2724 g_mirror_disk_state2str(disk->d_state))); 2725 /* 2726 * Reset bumping syncid if disk disappeared in STARTING 2727 * state. 2728 */ 2729 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2730 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2731#ifdef INVARIANTS 2732 } else { 2733 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2734 sc->sc_name, 2735 g_mirror_device_state2str(sc->sc_state), 2736 g_mirror_get_diskname(disk), 2737 g_mirror_disk_state2str(disk->d_state))); 2738#endif 2739 } 2740 DISK_STATE_CHANGED(); 2741 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2742 sc->sc_name, g_mirror_get_diskname(disk)); 2743 2744 g_mirror_destroy_disk(disk); 2745 break; 2746 case G_MIRROR_DISK_STATE_DESTROY: 2747 { 2748 int error; 2749 2750 error = g_mirror_clear_metadata(disk); 2751 if (error != 0) { 2752 G_MIRROR_DEBUG(0, 2753 "Device %s: failed to clear metadata on %s: %d.", 2754 sc->sc_name, g_mirror_get_diskname(disk), error); 2755 break; 2756 } 2757 DISK_STATE_CHANGED(); 2758 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2759 sc->sc_name, g_mirror_get_diskname(disk)); 2760 2761 g_mirror_destroy_disk(disk); 2762 sc->sc_ndisks--; 2763 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2764 g_mirror_update_metadata(disk); 2765 } 2766 break; 2767 } 2768 default: 2769 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2770 break; 2771 } 2772 return (0); 2773} 2774#undef DISK_STATE_CHANGED 2775 2776int 2777g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2778{ 2779 struct g_provider *pp; 2780 u_char *buf; 2781 int error; 2782 2783 g_topology_assert(); 2784 2785 error = g_access(cp, 1, 0, 0); 2786 if (error != 0) 2787 return (error); 2788 pp = cp->provider; 2789 g_topology_unlock(); 2790 /* Metadata are stored on last sector. */ 2791 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2792 &error); 2793 g_topology_lock(); 2794 g_access(cp, -1, 0, 0); 2795 if (buf == NULL) { 2796 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2797 cp->provider->name, error); 2798 return (error); 2799 } 2800 2801 /* Decode metadata. */ 2802 error = mirror_metadata_decode(buf, md); 2803 g_free(buf); 2804 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2805 return (EINVAL); 2806 if (md->md_version > G_MIRROR_VERSION) { 2807 G_MIRROR_DEBUG(0, 2808 "Kernel module is too old to handle metadata from %s.", 2809 cp->provider->name); 2810 return (EINVAL); 2811 } 2812 if (error != 0) { 2813 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2814 cp->provider->name); 2815 return (error); 2816 } 2817 2818 return (0); 2819} 2820 2821static int 2822g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2823 struct g_mirror_metadata *md) 2824{ 2825 2826 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2827 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2828 pp->name, md->md_did); 2829 return (EEXIST); 2830 } 2831 if (md->md_all != sc->sc_ndisks) { 2832 G_MIRROR_DEBUG(1, 2833 "Invalid '%s' field on disk %s (device %s), skipping.", 2834 "md_all", pp->name, sc->sc_name); 2835 return (EINVAL); 2836 } 2837 if (md->md_slice != sc->sc_slice) { 2838 G_MIRROR_DEBUG(1, 2839 "Invalid '%s' field on disk %s (device %s), skipping.", 2840 "md_slice", pp->name, sc->sc_name); 2841 return (EINVAL); 2842 } 2843 if (md->md_balance != sc->sc_balance) { 2844 G_MIRROR_DEBUG(1, 2845 "Invalid '%s' field on disk %s (device %s), skipping.", 2846 "md_balance", pp->name, sc->sc_name); 2847 return (EINVAL); 2848 } 2849#if 0 2850 if (md->md_mediasize != sc->sc_mediasize) { 2851 G_MIRROR_DEBUG(1, 2852 "Invalid '%s' field on disk %s (device %s), skipping.", 2853 "md_mediasize", pp->name, sc->sc_name); 2854 return (EINVAL); 2855 } 2856#endif 2857 if (sc->sc_mediasize > pp->mediasize) { 2858 G_MIRROR_DEBUG(1, 2859 "Invalid size of disk %s (device %s), skipping.", pp->name, 2860 sc->sc_name); 2861 return (EINVAL); 2862 } 2863 if (md->md_sectorsize != sc->sc_sectorsize) { 2864 G_MIRROR_DEBUG(1, 2865 "Invalid '%s' field on disk %s (device %s), skipping.", 2866 "md_sectorsize", pp->name, sc->sc_name); 2867 return (EINVAL); 2868 } 2869 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2870 G_MIRROR_DEBUG(1, 2871 "Invalid sector size of disk %s (device %s), skipping.", 2872 pp->name, sc->sc_name); 2873 return (EINVAL); 2874 } 2875 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2876 G_MIRROR_DEBUG(1, 2877 "Invalid device flags on disk %s (device %s), skipping.", 2878 pp->name, sc->sc_name); 2879 return (EINVAL); 2880 } 2881 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2882 G_MIRROR_DEBUG(1, 2883 "Invalid disk flags on disk %s (device %s), skipping.", 2884 pp->name, sc->sc_name); 2885 return (EINVAL); 2886 } 2887 return (0); 2888} 2889 2890int 2891g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2892 struct g_mirror_metadata *md) 2893{ 2894 struct g_mirror_disk *disk; 2895 int error; 2896 2897 g_topology_assert_not(); 2898 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2899 2900 error = g_mirror_check_metadata(sc, pp, md); 2901 if (error != 0) 2902 return (error); 2903 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2904 md->md_genid < sc->sc_genid) { 2905 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2906 pp->name, sc->sc_name); 2907 return (EINVAL); 2908 } 2909 disk = g_mirror_init_disk(sc, pp, md, &error); 2910 if (disk == NULL) 2911 return (error); 2912 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2913 G_MIRROR_EVENT_WAIT); 2914 if (error != 0) 2915 return (error); 2916 if (md->md_version < G_MIRROR_VERSION) { 2917 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2918 pp->name, md->md_version, G_MIRROR_VERSION); 2919 g_mirror_update_metadata(disk); 2920 } 2921 return (0); 2922} 2923 2924static void 2925g_mirror_destroy_delayed(void *arg, int flag) 2926{ 2927 struct g_mirror_softc *sc; 2928 int error; 2929 2930 if (flag == EV_CANCEL) { 2931 G_MIRROR_DEBUG(1, "Destroying canceled."); 2932 return; 2933 } 2934 sc = arg; 2935 g_topology_unlock(); 2936 sx_xlock(&sc->sc_lock); 2937 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2938 ("DESTROY flag set on %s.", sc->sc_name)); 2939 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0, 2940 ("CLOSEWAIT flag not set on %s.", sc->sc_name)); 2941 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 2942 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 2943 if (error != 0) { 2944 G_MIRROR_DEBUG(0, "Cannot destroy %s (error=%d).", 2945 sc->sc_name, error); 2946 sx_xunlock(&sc->sc_lock); 2947 } 2948 g_topology_lock(); 2949} 2950 2951static int 2952g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2953{ 2954 struct g_mirror_softc *sc; 2955 int error = 0; 2956 2957 g_topology_assert(); 2958 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2959 acw, ace); 2960 2961 sc = pp->private; 2962 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 2963 2964 g_topology_unlock(); 2965 sx_xlock(&sc->sc_lock); 2966 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 2967 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 || 2968 LIST_EMPTY(&sc->sc_disks)) { 2969 if (acr > 0 || acw > 0 || ace > 0) 2970 error = ENXIO; 2971 goto end; 2972 } 2973 sc->sc_provider_open += acr + acw + ace; 2974 if (pp->acw + acw == 0) 2975 g_mirror_idle(sc, 0); 2976 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 && 2977 sc->sc_provider_open == 0) 2978 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, sc, NULL); 2979end: 2980 sx_xunlock(&sc->sc_lock); 2981 g_topology_lock(); 2982 return (error); 2983} 2984 2985struct g_geom * 2986g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md, 2987 u_int type) 2988{ 2989 struct g_mirror_softc *sc; 2990 struct g_geom *gp; 2991 int error, timeout; 2992 2993 g_topology_assert(); 2994 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2995 md->md_mid); 2996 2997 /* One disk is minimum. */ 2998 if (md->md_all < 1) 2999 return (NULL); 3000 /* 3001 * Action geom. 3002 */ 3003 gp = g_new_geomf(mp, "%s", md->md_name); 3004 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 3005 gp->start = g_mirror_start; 3006 gp->orphan = g_mirror_orphan; 3007 gp->access = g_mirror_access; 3008 gp->dumpconf = g_mirror_dumpconf; 3009 3010 sc->sc_type = type; 3011 sc->sc_id = md->md_mid; 3012 sc->sc_slice = md->md_slice; 3013 sc->sc_balance = md->md_balance; 3014 sc->sc_mediasize = md->md_mediasize; 3015 sc->sc_sectorsize = md->md_sectorsize; 3016 sc->sc_ndisks = md->md_all; 3017 sc->sc_flags = md->md_mflags; 3018 sc->sc_bump_id = 0; 3019 sc->sc_idle = 1; 3020 sc->sc_last_write = time_uptime; 3021 sc->sc_writes = 0; 3022 sc->sc_refcnt = 1; 3023 sx_init(&sc->sc_lock, "gmirror:lock"); 3024 TAILQ_INIT(&sc->sc_queue); 3025 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 3026 TAILQ_INIT(&sc->sc_regular_delayed); 3027 TAILQ_INIT(&sc->sc_inflight); 3028 TAILQ_INIT(&sc->sc_sync_delayed); 3029 LIST_INIT(&sc->sc_disks); 3030 TAILQ_INIT(&sc->sc_events); 3031 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 3032 callout_init(&sc->sc_callout, 1); 3033 mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF); 3034 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 3035 gp->softc = sc; 3036 sc->sc_geom = gp; 3037 sc->sc_provider = NULL; 3038 sc->sc_provider_open = 0; 3039 /* 3040 * Synchronization geom. 3041 */ 3042 gp = g_new_geomf(mp, "%s.sync", md->md_name); 3043 gp->softc = sc; 3044 gp->orphan = g_mirror_orphan; 3045 sc->sc_sync.ds_geom = gp; 3046 sc->sc_sync.ds_ndisks = 0; 3047 error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 3048 "g_mirror %s", md->md_name); 3049 if (error != 0) { 3050 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 3051 sc->sc_name); 3052 g_destroy_geom(sc->sc_sync.ds_geom); 3053 g_destroy_geom(sc->sc_geom); 3054 g_mirror_free_device(sc); 3055 return (NULL); 3056 } 3057 3058 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 3059 sc->sc_name, sc->sc_ndisks, sc->sc_id); 3060 3061 sc->sc_rootmount = root_mount_hold("GMIRROR"); 3062 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 3063 /* 3064 * Run timeout. 3065 */ 3066 timeout = g_mirror_timeout * hz; 3067 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 3068 return (sc->sc_geom); 3069} 3070 3071int 3072g_mirror_destroy(struct g_mirror_softc *sc, int how) 3073{ 3074 struct g_mirror_disk *disk; 3075 3076 g_topology_assert_not(); 3077 sx_assert(&sc->sc_lock, SX_XLOCKED); 3078 3079 if (sc->sc_provider_open != 0) { 3080 switch (how) { 3081 case G_MIRROR_DESTROY_SOFT: 3082 G_MIRROR_DEBUG(1, 3083 "Device %s is still open (%d).", sc->sc_name, 3084 sc->sc_provider_open); 3085 return (EBUSY); 3086 case G_MIRROR_DESTROY_DELAYED: 3087 G_MIRROR_DEBUG(1, 3088 "Device %s will be destroyed on last close.", 3089 sc->sc_name); 3090 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 3091 if (disk->d_state == 3092 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3093 g_mirror_sync_stop(disk, 1); 3094 } 3095 } 3096 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_CLOSEWAIT; 3097 return (EBUSY); 3098 case G_MIRROR_DESTROY_HARD: 3099 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 3100 "can't be definitely removed.", sc->sc_name); 3101 } 3102 } 3103 3104 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3105 sx_xunlock(&sc->sc_lock); 3106 return (0); 3107 } 3108 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 3109 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DRAIN; 3110 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3111 sx_xunlock(&sc->sc_lock); 3112 mtx_lock(&sc->sc_queue_mtx); 3113 wakeup(sc); 3114 mtx_unlock(&sc->sc_queue_mtx); 3115 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3116 while (sc->sc_worker != NULL) 3117 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 3118 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3119 sx_xlock(&sc->sc_lock); 3120 g_mirror_destroy_device(sc); 3121 return (0); 3122} 3123 3124static void 3125g_mirror_taste_orphan(struct g_consumer *cp) 3126{ 3127 3128 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3129 cp->provider->name)); 3130} 3131 3132static struct g_geom * 3133g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3134{ 3135 struct g_mirror_metadata md; 3136 struct g_mirror_softc *sc; 3137 struct g_consumer *cp; 3138 struct g_geom *gp; 3139 int error; 3140 3141 g_topology_assert(); 3142 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3143 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 3144 3145 gp = g_new_geomf(mp, "mirror:taste"); 3146 /* 3147 * This orphan function should be never called. 3148 */ 3149 gp->orphan = g_mirror_taste_orphan; 3150 cp = g_new_consumer(gp); 3151 g_attach(cp, pp); 3152 error = g_mirror_read_metadata(cp, &md); 3153 g_detach(cp); 3154 g_destroy_consumer(cp); 3155 g_destroy_geom(gp); 3156 if (error != 0) 3157 return (NULL); 3158 gp = NULL; 3159 3160 if (md.md_provider[0] != '\0' && 3161 !g_compare_names(md.md_provider, pp->name)) 3162 return (NULL); 3163 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3164 return (NULL); 3165 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 3166 G_MIRROR_DEBUG(0, 3167 "Device %s: provider %s marked as inactive, skipping.", 3168 md.md_name, pp->name); 3169 return (NULL); 3170 } 3171 if (g_mirror_debug >= 2) 3172 mirror_metadata_dump(&md); 3173 3174 /* 3175 * Let's check if device already exists. 3176 */ 3177 sc = NULL; 3178 LIST_FOREACH(gp, &mp->geom, geom) { 3179 sc = gp->softc; 3180 if (sc == NULL) 3181 continue; 3182 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 3183 continue; 3184 if (sc->sc_sync.ds_geom == gp) 3185 continue; 3186 if (strcmp(md.md_name, sc->sc_name) != 0) 3187 continue; 3188 if (md.md_mid != sc->sc_id) { 3189 G_MIRROR_DEBUG(0, "Device %s already configured.", 3190 sc->sc_name); 3191 return (NULL); 3192 } 3193 break; 3194 } 3195 if (gp == NULL) { 3196 gp = g_mirror_create(mp, &md, G_MIRROR_TYPE_AUTOMATIC); 3197 if (gp == NULL) { 3198 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3199 md.md_name); 3200 return (NULL); 3201 } 3202 sc = gp->softc; 3203 } 3204 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3205 g_topology_unlock(); 3206 sx_xlock(&sc->sc_lock); 3207 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; 3208 error = g_mirror_add_disk(sc, pp, &md); 3209 if (error != 0) { 3210 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3211 pp->name, gp->name, error); 3212 if (LIST_EMPTY(&sc->sc_disks)) { 3213 g_cancel_event(sc); 3214 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3215 g_topology_lock(); 3216 return (NULL); 3217 } 3218 gp = NULL; 3219 } 3220 sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; 3221 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3222 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3223 g_topology_lock(); 3224 return (NULL); 3225 } 3226 sx_xunlock(&sc->sc_lock); 3227 g_topology_lock(); 3228 return (gp); 3229} 3230 3231static void 3232g_mirror_resize(struct g_consumer *cp) 3233{ 3234 struct g_mirror_disk *disk; 3235 3236 g_topology_assert(); 3237 g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name); 3238 3239 disk = cp->private; 3240 if (disk == NULL) 3241 return; 3242 g_topology_unlock(); 3243 g_mirror_update_metadata(disk); 3244 g_topology_lock(); 3245} 3246 3247static int 3248g_mirror_destroy_geom(struct gctl_req *req __unused, 3249 struct g_class *mp __unused, struct g_geom *gp) 3250{ 3251 struct g_mirror_softc *sc; 3252 int error; 3253 3254 g_topology_unlock(); 3255 sc = gp->softc; 3256 sx_xlock(&sc->sc_lock); 3257 g_cancel_event(sc); 3258 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3259 if (error != 0) 3260 sx_xunlock(&sc->sc_lock); 3261 g_topology_lock(); 3262 return (error); 3263} 3264 3265static void 3266g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3267 struct g_consumer *cp, struct g_provider *pp) 3268{ 3269 struct g_mirror_softc *sc; 3270 3271 g_topology_assert(); 3272 3273 sc = gp->softc; 3274 if (sc == NULL) 3275 return; 3276 /* Skip synchronization geom. */ 3277 if (gp == sc->sc_sync.ds_geom) 3278 return; 3279 if (pp != NULL) { 3280 /* Nothing here. */ 3281 } else if (cp != NULL) { 3282 struct g_mirror_disk *disk; 3283 3284 disk = cp->private; 3285 if (disk == NULL) 3286 return; 3287 g_topology_unlock(); 3288 sx_xlock(&sc->sc_lock); 3289 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3290 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3291 sbuf_printf(sb, "%s<Synchronized>", indent); 3292 if (disk->d_sync.ds_offset == 0) 3293 sbuf_printf(sb, "0%%"); 3294 else { 3295 sbuf_printf(sb, "%u%%", 3296 (u_int)((disk->d_sync.ds_offset * 100) / 3297 sc->sc_provider->mediasize)); 3298 } 3299 sbuf_printf(sb, "</Synchronized>\n"); 3300 if (disk->d_sync.ds_offset > 0) { 3301 sbuf_printf(sb, "%s<BytesSynced>%jd" 3302 "</BytesSynced>\n", indent, 3303 (intmax_t)disk->d_sync.ds_offset); 3304 } 3305 } 3306 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3307 disk->d_sync.ds_syncid); 3308 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3309 disk->d_genid); 3310 sbuf_printf(sb, "%s<Flags>", indent); 3311 if (disk->d_flags == 0) 3312 sbuf_printf(sb, "NONE"); 3313 else { 3314 int first = 1; 3315 3316#define ADD_FLAG(flag, name) do { \ 3317 if ((disk->d_flags & (flag)) != 0) { \ 3318 if (!first) \ 3319 sbuf_printf(sb, ", "); \ 3320 else \ 3321 first = 0; \ 3322 sbuf_printf(sb, name); \ 3323 } \ 3324} while (0) 3325 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3326 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3327 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3328 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3329 "SYNCHRONIZING"); 3330 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3331 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3332#undef ADD_FLAG 3333 } 3334 sbuf_printf(sb, "</Flags>\n"); 3335 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3336 disk->d_priority); 3337 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3338 g_mirror_disk_state2str(disk->d_state)); 3339 sx_xunlock(&sc->sc_lock); 3340 g_topology_lock(); 3341 } else { 3342 g_topology_unlock(); 3343 sx_xlock(&sc->sc_lock); 3344 sbuf_printf(sb, "%s<Type>", indent); 3345 switch (sc->sc_type) { 3346 case G_MIRROR_TYPE_AUTOMATIC: 3347 sbuf_printf(sb, "AUTOMATIC"); 3348 break; 3349 case G_MIRROR_TYPE_MANUAL: 3350 sbuf_printf(sb, "MANUAL"); 3351 break; 3352 default: 3353 sbuf_printf(sb, "UNKNOWN"); 3354 break; 3355 } 3356 sbuf_printf(sb, "</Type>\n"); 3357 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3358 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3359 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3360 sbuf_printf(sb, "%s<Flags>", indent); 3361 if (sc->sc_flags == 0) 3362 sbuf_printf(sb, "NONE"); 3363 else { 3364 int first = 1; 3365 3366#define ADD_FLAG(flag, name) do { \ 3367 if ((sc->sc_flags & (flag)) != 0) { \ 3368 if (!first) \ 3369 sbuf_printf(sb, ", "); \ 3370 else \ 3371 first = 0; \ 3372 sbuf_printf(sb, name); \ 3373 } \ 3374} while (0) 3375 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3376 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3377#undef ADD_FLAG 3378 } 3379 sbuf_printf(sb, "</Flags>\n"); 3380 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3381 (u_int)sc->sc_slice); 3382 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3383 balance_name(sc->sc_balance)); 3384 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3385 sc->sc_ndisks); 3386 sbuf_printf(sb, "%s<State>", indent); 3387 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3388 sbuf_printf(sb, "%s", "STARTING"); 3389 else if (sc->sc_ndisks == 3390 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3391 sbuf_printf(sb, "%s", "COMPLETE"); 3392 else 3393 sbuf_printf(sb, "%s", "DEGRADED"); 3394 sbuf_printf(sb, "</State>\n"); 3395 sx_xunlock(&sc->sc_lock); 3396 g_topology_lock(); 3397 } 3398} 3399 3400static void 3401g_mirror_shutdown_post_sync(void *arg, int howto) 3402{ 3403 struct g_class *mp; 3404 struct g_geom *gp, *gp2; 3405 struct g_mirror_softc *sc; 3406 int error; 3407 3408 if (panicstr != NULL) 3409 return; 3410 3411 mp = arg; 3412 g_topology_lock(); 3413 g_mirror_shutdown = 1; 3414 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3415 if ((sc = gp->softc) == NULL) 3416 continue; 3417 /* Skip synchronization geom. */ 3418 if (gp == sc->sc_sync.ds_geom) 3419 continue; 3420 g_topology_unlock(); 3421 sx_xlock(&sc->sc_lock); 3422 g_mirror_idle(sc, -1); 3423 g_cancel_event(sc); 3424 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3425 if (error != 0) 3426 sx_xunlock(&sc->sc_lock); 3427 g_topology_lock(); 3428 } 3429 g_topology_unlock(); 3430} 3431 3432static void 3433g_mirror_init(struct g_class *mp) 3434{ 3435 3436 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3437 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3438 if (g_mirror_post_sync == NULL) 3439 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3440} 3441 3442static void 3443g_mirror_fini(struct g_class *mp) 3444{ 3445 3446 if (g_mirror_post_sync != NULL) 3447 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); 3448} 3449 3450DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3451