g_mirror.c revision 324588
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 324588 2017-10-13 09:13:08Z avg $"); 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"); 57u_int g_mirror_debug = 0; 58SYSCTL_UINT(_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."); 72 73#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 74 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 75 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 76 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 77} while (0) 78 79static eventhandler_tag g_mirror_post_sync = NULL; 80static int g_mirror_shutdown = 0; 81 82static g_ctl_destroy_geom_t g_mirror_destroy_geom; 83static g_taste_t g_mirror_taste; 84static g_init_t g_mirror_init; 85static g_fini_t g_mirror_fini; 86static g_provgone_t g_mirror_providergone; 87static g_resize_t g_mirror_resize; 88 89struct g_class g_mirror_class = { 90 .name = G_MIRROR_CLASS_NAME, 91 .version = G_VERSION, 92 .ctlreq = g_mirror_config, 93 .taste = g_mirror_taste, 94 .destroy_geom = g_mirror_destroy_geom, 95 .init = g_mirror_init, 96 .fini = g_mirror_fini, 97 .providergone = g_mirror_providergone, 98 .resize = g_mirror_resize 99}; 100 101 102static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 103static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 104static void g_mirror_update_device(struct g_mirror_softc *sc, bool force); 105static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 106 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 107static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 108static void g_mirror_register_request(struct bio *bp); 109static void g_mirror_sync_release(struct g_mirror_softc *sc); 110 111 112static const char * 113g_mirror_disk_state2str(int state) 114{ 115 116 switch (state) { 117 case G_MIRROR_DISK_STATE_NONE: 118 return ("NONE"); 119 case G_MIRROR_DISK_STATE_NEW: 120 return ("NEW"); 121 case G_MIRROR_DISK_STATE_ACTIVE: 122 return ("ACTIVE"); 123 case G_MIRROR_DISK_STATE_STALE: 124 return ("STALE"); 125 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 126 return ("SYNCHRONIZING"); 127 case G_MIRROR_DISK_STATE_DISCONNECTED: 128 return ("DISCONNECTED"); 129 case G_MIRROR_DISK_STATE_DESTROY: 130 return ("DESTROY"); 131 default: 132 return ("INVALID"); 133 } 134} 135 136static const char * 137g_mirror_device_state2str(int state) 138{ 139 140 switch (state) { 141 case G_MIRROR_DEVICE_STATE_STARTING: 142 return ("STARTING"); 143 case G_MIRROR_DEVICE_STATE_RUNNING: 144 return ("RUNNING"); 145 default: 146 return ("INVALID"); 147 } 148} 149 150static const char * 151g_mirror_get_diskname(struct g_mirror_disk *disk) 152{ 153 154 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 155 return ("[unknown]"); 156 return (disk->d_name); 157} 158 159/* 160 * --- Events handling functions --- 161 * Events in geom_mirror are used to maintain disks and device status 162 * from one thread to simplify locking. 163 */ 164static void 165g_mirror_event_free(struct g_mirror_event *ep) 166{ 167 168 free(ep, M_MIRROR); 169} 170 171int 172g_mirror_event_send(void *arg, int state, int flags) 173{ 174 struct g_mirror_softc *sc; 175 struct g_mirror_disk *disk; 176 struct g_mirror_event *ep; 177 int error; 178 179 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 180 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 181 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 182 disk = NULL; 183 sc = arg; 184 } else { 185 disk = arg; 186 sc = disk->d_softc; 187 } 188 ep->e_disk = disk; 189 ep->e_state = state; 190 ep->e_flags = flags; 191 ep->e_error = 0; 192 mtx_lock(&sc->sc_events_mtx); 193 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 194 mtx_unlock(&sc->sc_events_mtx); 195 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 196 mtx_lock(&sc->sc_queue_mtx); 197 wakeup(sc); 198 mtx_unlock(&sc->sc_queue_mtx); 199 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 200 return (0); 201 sx_assert(&sc->sc_lock, SX_XLOCKED); 202 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 203 sx_xunlock(&sc->sc_lock); 204 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 205 mtx_lock(&sc->sc_events_mtx); 206 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 207 hz * 5); 208 } 209 error = ep->e_error; 210 g_mirror_event_free(ep); 211 sx_xlock(&sc->sc_lock); 212 return (error); 213} 214 215static struct g_mirror_event * 216g_mirror_event_get(struct g_mirror_softc *sc) 217{ 218 struct g_mirror_event *ep; 219 220 mtx_lock(&sc->sc_events_mtx); 221 ep = TAILQ_FIRST(&sc->sc_events); 222 mtx_unlock(&sc->sc_events_mtx); 223 return (ep); 224} 225 226static void 227g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 228{ 229 230 mtx_lock(&sc->sc_events_mtx); 231 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 232 mtx_unlock(&sc->sc_events_mtx); 233} 234 235static void 236g_mirror_event_cancel(struct g_mirror_disk *disk) 237{ 238 struct g_mirror_softc *sc; 239 struct g_mirror_event *ep, *tmpep; 240 241 sc = disk->d_softc; 242 sx_assert(&sc->sc_lock, SX_XLOCKED); 243 244 mtx_lock(&sc->sc_events_mtx); 245 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 246 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 247 continue; 248 if (ep->e_disk != disk) 249 continue; 250 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 251 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 252 g_mirror_event_free(ep); 253 else { 254 ep->e_error = ECANCELED; 255 wakeup(ep); 256 } 257 } 258 mtx_unlock(&sc->sc_events_mtx); 259} 260 261/* 262 * Return the number of disks in given state. 263 * If state is equal to -1, count all connected disks. 264 */ 265u_int 266g_mirror_ndisks(struct g_mirror_softc *sc, int state) 267{ 268 struct g_mirror_disk *disk; 269 u_int n = 0; 270 271 sx_assert(&sc->sc_lock, SX_LOCKED); 272 273 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 274 if (state == -1 || disk->d_state == state) 275 n++; 276 } 277 return (n); 278} 279 280/* 281 * Find a disk in mirror by its disk ID. 282 */ 283static struct g_mirror_disk * 284g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 285{ 286 struct g_mirror_disk *disk; 287 288 sx_assert(&sc->sc_lock, SX_XLOCKED); 289 290 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 291 if (disk->d_id == id) 292 return (disk); 293 } 294 return (NULL); 295} 296 297static u_int 298g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 299{ 300 struct bio *bp; 301 u_int nreqs = 0; 302 303 mtx_lock(&sc->sc_queue_mtx); 304 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 305 if (bp->bio_from == cp) 306 nreqs++; 307 } 308 mtx_unlock(&sc->sc_queue_mtx); 309 return (nreqs); 310} 311 312static int 313g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 314{ 315 316 if (cp->index > 0) { 317 G_MIRROR_DEBUG(2, 318 "I/O requests for %s exist, can't destroy it now.", 319 cp->provider->name); 320 return (1); 321 } 322 if (g_mirror_nrequests(sc, cp) > 0) { 323 G_MIRROR_DEBUG(2, 324 "I/O requests for %s in queue, can't destroy it now.", 325 cp->provider->name); 326 return (1); 327 } 328 return (0); 329} 330 331static void 332g_mirror_destroy_consumer(void *arg, int flags __unused) 333{ 334 struct g_consumer *cp; 335 336 g_topology_assert(); 337 338 cp = arg; 339 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 340 g_detach(cp); 341 g_destroy_consumer(cp); 342} 343 344static void 345g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 346{ 347 struct g_provider *pp; 348 int retaste_wait; 349 350 g_topology_assert(); 351 352 cp->private = NULL; 353 if (g_mirror_is_busy(sc, cp)) 354 return; 355 pp = cp->provider; 356 retaste_wait = 0; 357 if (cp->acw == 1) { 358 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 359 retaste_wait = 1; 360 } 361 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 362 -cp->acw, -cp->ace, 0); 363 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 364 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 365 if (retaste_wait) { 366 /* 367 * After retaste event was send (inside g_access()), we can send 368 * event to detach and destroy consumer. 369 * A class, which has consumer to the given provider connected 370 * will not receive retaste event for the provider. 371 * This is the way how I ignore retaste events when I close 372 * consumers opened for write: I detach and destroy consumer 373 * after retaste event is sent. 374 */ 375 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 376 return; 377 } 378 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 379 g_detach(cp); 380 g_destroy_consumer(cp); 381} 382 383static int 384g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 385{ 386 struct g_consumer *cp; 387 int error; 388 389 g_topology_assert_not(); 390 KASSERT(disk->d_consumer == NULL, 391 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 392 393 g_topology_lock(); 394 cp = g_new_consumer(disk->d_softc->sc_geom); 395 cp->flags |= G_CF_DIRECT_RECEIVE; 396 error = g_attach(cp, pp); 397 if (error != 0) { 398 g_destroy_consumer(cp); 399 g_topology_unlock(); 400 return (error); 401 } 402 error = g_access(cp, 1, 1, 1); 403 if (error != 0) { 404 g_detach(cp); 405 g_destroy_consumer(cp); 406 g_topology_unlock(); 407 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 408 pp->name, error); 409 return (error); 410 } 411 g_topology_unlock(); 412 disk->d_consumer = cp; 413 disk->d_consumer->private = disk; 414 disk->d_consumer->index = 0; 415 416 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 417 return (0); 418} 419 420static void 421g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 422{ 423 424 g_topology_assert(); 425 426 if (cp == NULL) 427 return; 428 if (cp->provider != NULL) 429 g_mirror_kill_consumer(sc, cp); 430 else 431 g_destroy_consumer(cp); 432} 433 434/* 435 * Initialize disk. This means allocate memory, create consumer, attach it 436 * to the provider and open access (r1w1e1) to it. 437 */ 438static struct g_mirror_disk * 439g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 440 struct g_mirror_metadata *md, int *errorp) 441{ 442 struct g_mirror_disk *disk; 443 int i, error; 444 445 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 446 if (disk == NULL) { 447 error = ENOMEM; 448 goto fail; 449 } 450 disk->d_softc = sc; 451 error = g_mirror_connect_disk(disk, pp); 452 if (error != 0) 453 goto fail; 454 disk->d_id = md->md_did; 455 disk->d_state = G_MIRROR_DISK_STATE_NONE; 456 disk->d_priority = md->md_priority; 457 disk->d_flags = md->md_dflags; 458 error = g_getattr("GEOM::candelete", disk->d_consumer, &i); 459 if (error == 0 && i != 0) 460 disk->d_flags |= G_MIRROR_DISK_FLAG_CANDELETE; 461 if (md->md_provider[0] != '\0') 462 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 463 disk->d_sync.ds_consumer = NULL; 464 disk->d_sync.ds_offset = md->md_sync_offset; 465 disk->d_sync.ds_offset_done = md->md_sync_offset; 466 disk->d_genid = md->md_genid; 467 disk->d_sync.ds_syncid = md->md_syncid; 468 if (errorp != NULL) 469 *errorp = 0; 470 return (disk); 471fail: 472 if (errorp != NULL) 473 *errorp = error; 474 if (disk != NULL) 475 free(disk, M_MIRROR); 476 return (NULL); 477} 478 479static void 480g_mirror_destroy_disk(struct g_mirror_disk *disk) 481{ 482 struct g_mirror_softc *sc; 483 484 g_topology_assert_not(); 485 sc = disk->d_softc; 486 sx_assert(&sc->sc_lock, SX_XLOCKED); 487 488 LIST_REMOVE(disk, d_next); 489 g_mirror_event_cancel(disk); 490 if (sc->sc_hint == disk) 491 sc->sc_hint = NULL; 492 switch (disk->d_state) { 493 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 494 g_mirror_sync_stop(disk, 1); 495 /* FALLTHROUGH */ 496 case G_MIRROR_DISK_STATE_NEW: 497 case G_MIRROR_DISK_STATE_STALE: 498 case G_MIRROR_DISK_STATE_ACTIVE: 499 g_topology_lock(); 500 g_mirror_disconnect_consumer(sc, disk->d_consumer); 501 g_topology_unlock(); 502 free(disk, M_MIRROR); 503 break; 504 default: 505 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 506 g_mirror_get_diskname(disk), 507 g_mirror_disk_state2str(disk->d_state))); 508 } 509} 510 511static void 512g_mirror_free_device(struct g_mirror_softc *sc) 513{ 514 515 mtx_destroy(&sc->sc_queue_mtx); 516 mtx_destroy(&sc->sc_events_mtx); 517 mtx_destroy(&sc->sc_done_mtx); 518 sx_destroy(&sc->sc_lock); 519 free(sc, M_MIRROR); 520} 521 522static void 523g_mirror_providergone(struct g_provider *pp) 524{ 525 struct g_mirror_softc *sc = pp->private; 526 527 if ((--sc->sc_refcnt) == 0) 528 g_mirror_free_device(sc); 529} 530 531static void 532g_mirror_destroy_device(struct g_mirror_softc *sc) 533{ 534 struct g_mirror_disk *disk; 535 struct g_mirror_event *ep; 536 struct g_geom *gp; 537 struct g_consumer *cp, *tmpcp; 538 539 g_topology_assert_not(); 540 sx_assert(&sc->sc_lock, SX_XLOCKED); 541 542 gp = sc->sc_geom; 543 if (sc->sc_provider != NULL) 544 g_mirror_destroy_provider(sc); 545 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 546 disk = LIST_FIRST(&sc->sc_disks)) { 547 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 548 g_mirror_update_metadata(disk); 549 g_mirror_destroy_disk(disk); 550 } 551 while ((ep = g_mirror_event_get(sc)) != NULL) { 552 g_mirror_event_remove(sc, ep); 553 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 554 g_mirror_event_free(ep); 555 else { 556 ep->e_error = ECANCELED; 557 ep->e_flags |= G_MIRROR_EVENT_DONE; 558 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 559 mtx_lock(&sc->sc_events_mtx); 560 wakeup(ep); 561 mtx_unlock(&sc->sc_events_mtx); 562 } 563 } 564 callout_drain(&sc->sc_callout); 565 566 g_topology_lock(); 567 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 568 g_mirror_disconnect_consumer(sc, cp); 569 } 570 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 571 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 572 g_wither_geom(gp, ENXIO); 573 sx_xunlock(&sc->sc_lock); 574 if ((--sc->sc_refcnt) == 0) 575 g_mirror_free_device(sc); 576 g_topology_unlock(); 577} 578 579static void 580g_mirror_orphan(struct g_consumer *cp) 581{ 582 struct g_mirror_disk *disk; 583 584 g_topology_assert(); 585 586 disk = cp->private; 587 if (disk == NULL) 588 return; 589 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 590 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 591 G_MIRROR_EVENT_DONTWAIT); 592} 593 594/* 595 * Function should return the next active disk on the list. 596 * It is possible that it will be the same disk as given. 597 * If there are no active disks on list, NULL is returned. 598 */ 599static __inline struct g_mirror_disk * 600g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 601{ 602 struct g_mirror_disk *dp; 603 604 for (dp = LIST_NEXT(disk, d_next); dp != disk; 605 dp = LIST_NEXT(dp, d_next)) { 606 if (dp == NULL) 607 dp = LIST_FIRST(&sc->sc_disks); 608 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 609 break; 610 } 611 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 612 return (NULL); 613 return (dp); 614} 615 616static struct g_mirror_disk * 617g_mirror_get_disk(struct g_mirror_softc *sc) 618{ 619 struct g_mirror_disk *disk; 620 621 if (sc->sc_hint == NULL) { 622 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 623 if (sc->sc_hint == NULL) 624 return (NULL); 625 } 626 disk = sc->sc_hint; 627 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 628 disk = g_mirror_find_next(sc, disk); 629 if (disk == NULL) 630 return (NULL); 631 } 632 sc->sc_hint = g_mirror_find_next(sc, disk); 633 return (disk); 634} 635 636static int 637g_mirror_write_metadata(struct g_mirror_disk *disk, 638 struct g_mirror_metadata *md) 639{ 640 struct g_mirror_softc *sc; 641 struct g_consumer *cp; 642 off_t offset, length; 643 u_char *sector; 644 int error = 0; 645 646 g_topology_assert_not(); 647 sc = disk->d_softc; 648 sx_assert(&sc->sc_lock, SX_LOCKED); 649 650 cp = disk->d_consumer; 651 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 652 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 653 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 654 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 655 cp->acw, cp->ace)); 656 length = cp->provider->sectorsize; 657 offset = cp->provider->mediasize - length; 658 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 659 if (md != NULL && 660 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) { 661 /* 662 * Handle the case, when the size of parent provider reduced. 663 */ 664 if (offset < md->md_mediasize) 665 error = ENOSPC; 666 else 667 mirror_metadata_encode(md, sector); 668 } 669 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_metadata_write, error); 670 if (error == 0) 671 error = g_write_data(cp, offset, sector, length); 672 free(sector, M_MIRROR); 673 if (error != 0) { 674 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 675 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 676 G_MIRROR_DEBUG(0, "Cannot write metadata on %s " 677 "(device=%s, error=%d).", 678 g_mirror_get_diskname(disk), sc->sc_name, error); 679 } else { 680 G_MIRROR_DEBUG(1, "Cannot write metadata on %s " 681 "(device=%s, error=%d).", 682 g_mirror_get_diskname(disk), sc->sc_name, error); 683 } 684 if (g_mirror_disconnect_on_failure && 685 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 686 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 687 g_mirror_event_send(disk, 688 G_MIRROR_DISK_STATE_DISCONNECTED, 689 G_MIRROR_EVENT_DONTWAIT); 690 } 691 } 692 return (error); 693} 694 695static int 696g_mirror_clear_metadata(struct g_mirror_disk *disk) 697{ 698 int error; 699 700 g_topology_assert_not(); 701 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 702 703 if (disk->d_softc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 704 return (0); 705 error = g_mirror_write_metadata(disk, NULL); 706 if (error == 0) { 707 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 708 g_mirror_get_diskname(disk)); 709 } else { 710 G_MIRROR_DEBUG(0, 711 "Cannot clear metadata on disk %s (error=%d).", 712 g_mirror_get_diskname(disk), error); 713 } 714 return (error); 715} 716 717void 718g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 719 struct g_mirror_metadata *md) 720{ 721 722 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 723 md->md_version = G_MIRROR_VERSION; 724 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 725 md->md_mid = sc->sc_id; 726 md->md_all = sc->sc_ndisks; 727 md->md_slice = sc->sc_slice; 728 md->md_balance = sc->sc_balance; 729 md->md_genid = sc->sc_genid; 730 md->md_mediasize = sc->sc_mediasize; 731 md->md_sectorsize = sc->sc_sectorsize; 732 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 733 bzero(md->md_provider, sizeof(md->md_provider)); 734 if (disk == NULL) { 735 md->md_did = arc4random(); 736 md->md_priority = 0; 737 md->md_syncid = 0; 738 md->md_dflags = 0; 739 md->md_sync_offset = 0; 740 md->md_provsize = 0; 741 } else { 742 md->md_did = disk->d_id; 743 md->md_priority = disk->d_priority; 744 md->md_syncid = disk->d_sync.ds_syncid; 745 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 746 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 747 md->md_sync_offset = disk->d_sync.ds_offset_done; 748 else 749 md->md_sync_offset = 0; 750 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 751 strlcpy(md->md_provider, 752 disk->d_consumer->provider->name, 753 sizeof(md->md_provider)); 754 } 755 md->md_provsize = disk->d_consumer->provider->mediasize; 756 } 757} 758 759void 760g_mirror_update_metadata(struct g_mirror_disk *disk) 761{ 762 struct g_mirror_softc *sc; 763 struct g_mirror_metadata md; 764 int error; 765 766 g_topology_assert_not(); 767 sc = disk->d_softc; 768 sx_assert(&sc->sc_lock, SX_LOCKED); 769 770 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 771 return; 772 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) 773 g_mirror_fill_metadata(sc, disk, &md); 774 error = g_mirror_write_metadata(disk, &md); 775 if (error == 0) { 776 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 777 g_mirror_get_diskname(disk)); 778 } else { 779 G_MIRROR_DEBUG(0, 780 "Cannot update metadata on disk %s (error=%d).", 781 g_mirror_get_diskname(disk), error); 782 } 783} 784 785static void 786g_mirror_bump_syncid(struct g_mirror_softc *sc) 787{ 788 struct g_mirror_disk *disk; 789 790 g_topology_assert_not(); 791 sx_assert(&sc->sc_lock, SX_XLOCKED); 792 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 793 ("%s called with no active disks (device=%s).", __func__, 794 sc->sc_name)); 795 796 sc->sc_syncid++; 797 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 798 sc->sc_syncid); 799 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 800 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 801 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 802 disk->d_sync.ds_syncid = sc->sc_syncid; 803 g_mirror_update_metadata(disk); 804 } 805 } 806} 807 808static void 809g_mirror_bump_genid(struct g_mirror_softc *sc) 810{ 811 struct g_mirror_disk *disk; 812 813 g_topology_assert_not(); 814 sx_assert(&sc->sc_lock, SX_XLOCKED); 815 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 816 ("%s called with no active disks (device=%s).", __func__, 817 sc->sc_name)); 818 819 sc->sc_genid++; 820 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 821 sc->sc_genid); 822 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 823 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 824 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 825 disk->d_genid = sc->sc_genid; 826 g_mirror_update_metadata(disk); 827 } 828 } 829} 830 831static int 832g_mirror_idle(struct g_mirror_softc *sc, int acw) 833{ 834 struct g_mirror_disk *disk; 835 int timeout; 836 837 g_topology_assert_not(); 838 sx_assert(&sc->sc_lock, SX_XLOCKED); 839 840 if (sc->sc_provider == NULL) 841 return (0); 842 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 843 return (0); 844 if (sc->sc_idle) 845 return (0); 846 if (sc->sc_writes > 0) 847 return (0); 848 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 849 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); 850 if (!g_mirror_shutdown && timeout > 0) 851 return (timeout); 852 } 853 sc->sc_idle = 1; 854 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 855 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 856 continue; 857 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 858 g_mirror_get_diskname(disk), sc->sc_name); 859 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 860 g_mirror_update_metadata(disk); 861 } 862 return (0); 863} 864 865static void 866g_mirror_unidle(struct g_mirror_softc *sc) 867{ 868 struct g_mirror_disk *disk; 869 870 g_topology_assert_not(); 871 sx_assert(&sc->sc_lock, SX_XLOCKED); 872 873 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 874 return; 875 sc->sc_idle = 0; 876 sc->sc_last_write = time_uptime; 877 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 878 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 879 continue; 880 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 881 g_mirror_get_diskname(disk), sc->sc_name); 882 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 883 g_mirror_update_metadata(disk); 884 } 885} 886 887static void 888g_mirror_flush_done(struct bio *bp) 889{ 890 struct g_mirror_softc *sc; 891 struct bio *pbp; 892 893 pbp = bp->bio_parent; 894 sc = pbp->bio_to->private; 895 mtx_lock(&sc->sc_done_mtx); 896 if (pbp->bio_error == 0) 897 pbp->bio_error = bp->bio_error; 898 pbp->bio_completed += bp->bio_completed; 899 pbp->bio_inbed++; 900 if (pbp->bio_children == pbp->bio_inbed) { 901 mtx_unlock(&sc->sc_done_mtx); 902 g_io_deliver(pbp, pbp->bio_error); 903 } else 904 mtx_unlock(&sc->sc_done_mtx); 905 g_destroy_bio(bp); 906} 907 908static void 909g_mirror_done(struct bio *bp) 910{ 911 struct g_mirror_softc *sc; 912 913 sc = bp->bio_from->geom->softc; 914 bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR; 915 mtx_lock(&sc->sc_queue_mtx); 916 bioq_insert_tail(&sc->sc_queue, bp); 917 mtx_unlock(&sc->sc_queue_mtx); 918 wakeup(sc); 919} 920 921static void 922g_mirror_regular_request(struct bio *bp) 923{ 924 struct g_mirror_softc *sc; 925 struct g_mirror_disk *disk; 926 struct bio *pbp; 927 928 g_topology_assert_not(); 929 930 pbp = bp->bio_parent; 931 sc = pbp->bio_to->private; 932 bp->bio_from->index--; 933 if (bp->bio_cmd == BIO_WRITE) 934 sc->sc_writes--; 935 disk = bp->bio_from->private; 936 if (disk == NULL) { 937 g_topology_lock(); 938 g_mirror_kill_consumer(sc, bp->bio_from); 939 g_topology_unlock(); 940 } 941 942 if (bp->bio_cmd == BIO_READ) 943 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_read, 944 bp->bio_error); 945 else if (bp->bio_cmd == BIO_WRITE) 946 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_write, 947 bp->bio_error); 948 949 pbp->bio_inbed++; 950 KASSERT(pbp->bio_inbed <= pbp->bio_children, 951 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 952 pbp->bio_children)); 953 if (bp->bio_error == 0 && pbp->bio_error == 0) { 954 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 955 g_destroy_bio(bp); 956 if (pbp->bio_children == pbp->bio_inbed) { 957 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 958 pbp->bio_completed = pbp->bio_length; 959 if (pbp->bio_cmd == BIO_WRITE || 960 pbp->bio_cmd == BIO_DELETE) { 961 bioq_remove(&sc->sc_inflight, pbp); 962 /* Release delayed sync requests if possible. */ 963 g_mirror_sync_release(sc); 964 } 965 g_io_deliver(pbp, pbp->bio_error); 966 } 967 return; 968 } else if (bp->bio_error != 0) { 969 if (pbp->bio_error == 0) 970 pbp->bio_error = bp->bio_error; 971 if (disk != NULL) { 972 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 973 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 974 G_MIRROR_LOGREQ(0, bp, 975 "Request failed (error=%d).", 976 bp->bio_error); 977 } else { 978 G_MIRROR_LOGREQ(1, bp, 979 "Request failed (error=%d).", 980 bp->bio_error); 981 } 982 if (g_mirror_disconnect_on_failure && 983 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) 984 { 985 if (bp->bio_error == ENXIO && 986 bp->bio_cmd == BIO_READ) 987 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 988 else if (bp->bio_error == ENXIO) 989 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_NOW; 990 else 991 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 992 g_mirror_event_send(disk, 993 G_MIRROR_DISK_STATE_DISCONNECTED, 994 G_MIRROR_EVENT_DONTWAIT); 995 } 996 } 997 switch (pbp->bio_cmd) { 998 case BIO_DELETE: 999 case BIO_WRITE: 1000 pbp->bio_inbed--; 1001 pbp->bio_children--; 1002 break; 1003 } 1004 } 1005 g_destroy_bio(bp); 1006 1007 switch (pbp->bio_cmd) { 1008 case BIO_READ: 1009 if (pbp->bio_inbed < pbp->bio_children) 1010 break; 1011 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 1012 g_io_deliver(pbp, pbp->bio_error); 1013 else { 1014 pbp->bio_error = 0; 1015 mtx_lock(&sc->sc_queue_mtx); 1016 bioq_insert_tail(&sc->sc_queue, pbp); 1017 mtx_unlock(&sc->sc_queue_mtx); 1018 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1019 wakeup(sc); 1020 } 1021 break; 1022 case BIO_DELETE: 1023 case BIO_WRITE: 1024 if (pbp->bio_children == 0) { 1025 /* 1026 * All requests failed. 1027 */ 1028 } else if (pbp->bio_inbed < pbp->bio_children) { 1029 /* Do nothing. */ 1030 break; 1031 } else if (pbp->bio_children == pbp->bio_inbed) { 1032 /* Some requests succeeded. */ 1033 pbp->bio_error = 0; 1034 pbp->bio_completed = pbp->bio_length; 1035 } 1036 bioq_remove(&sc->sc_inflight, pbp); 1037 /* Release delayed sync requests if possible. */ 1038 g_mirror_sync_release(sc); 1039 g_io_deliver(pbp, pbp->bio_error); 1040 break; 1041 default: 1042 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 1043 break; 1044 } 1045} 1046 1047static void 1048g_mirror_sync_done(struct bio *bp) 1049{ 1050 struct g_mirror_softc *sc; 1051 1052 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1053 sc = bp->bio_from->geom->softc; 1054 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; 1055 mtx_lock(&sc->sc_queue_mtx); 1056 bioq_insert_tail(&sc->sc_queue, bp); 1057 mtx_unlock(&sc->sc_queue_mtx); 1058 wakeup(sc); 1059} 1060 1061static void 1062g_mirror_candelete(struct bio *bp) 1063{ 1064 struct g_mirror_softc *sc; 1065 struct g_mirror_disk *disk; 1066 int *val; 1067 1068 sc = bp->bio_to->private; 1069 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1070 if (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) 1071 break; 1072 } 1073 val = (int *)bp->bio_data; 1074 *val = (disk != NULL); 1075 g_io_deliver(bp, 0); 1076} 1077 1078static void 1079g_mirror_kernel_dump(struct bio *bp) 1080{ 1081 struct g_mirror_softc *sc; 1082 struct g_mirror_disk *disk; 1083 struct bio *cbp; 1084 struct g_kerneldump *gkd; 1085 1086 /* 1087 * We configure dumping to the first component, because this component 1088 * will be used for reading with 'prefer' balance algorithm. 1089 * If the component with the highest priority is currently disconnected 1090 * we will not be able to read the dump after the reboot if it will be 1091 * connected and synchronized later. Can we do something better? 1092 */ 1093 sc = bp->bio_to->private; 1094 disk = LIST_FIRST(&sc->sc_disks); 1095 1096 gkd = (struct g_kerneldump *)bp->bio_data; 1097 if (gkd->length > bp->bio_to->mediasize) 1098 gkd->length = bp->bio_to->mediasize; 1099 cbp = g_clone_bio(bp); 1100 if (cbp == NULL) { 1101 g_io_deliver(bp, ENOMEM); 1102 return; 1103 } 1104 cbp->bio_done = g_std_done; 1105 g_io_request(cbp, disk->d_consumer); 1106 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1107 g_mirror_get_diskname(disk)); 1108} 1109 1110static void 1111g_mirror_flush(struct g_mirror_softc *sc, struct bio *bp) 1112{ 1113 struct bio_queue_head queue; 1114 struct g_mirror_disk *disk; 1115 struct g_consumer *cp; 1116 struct bio *cbp; 1117 1118 bioq_init(&queue); 1119 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1120 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1121 continue; 1122 cbp = g_clone_bio(bp); 1123 if (cbp == NULL) { 1124 while ((cbp = bioq_takefirst(&queue)) != NULL) 1125 g_destroy_bio(cbp); 1126 if (bp->bio_error == 0) 1127 bp->bio_error = ENOMEM; 1128 g_io_deliver(bp, bp->bio_error); 1129 return; 1130 } 1131 bioq_insert_tail(&queue, cbp); 1132 cbp->bio_done = g_mirror_flush_done; 1133 cbp->bio_caller1 = disk; 1134 cbp->bio_to = disk->d_consumer->provider; 1135 } 1136 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1137 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1138 disk = cbp->bio_caller1; 1139 cbp->bio_caller1 = NULL; 1140 cp = disk->d_consumer; 1141 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1142 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1143 cp->acr, cp->acw, cp->ace)); 1144 g_io_request(cbp, disk->d_consumer); 1145 } 1146} 1147 1148static void 1149g_mirror_start(struct bio *bp) 1150{ 1151 struct g_mirror_softc *sc; 1152 1153 sc = bp->bio_to->private; 1154 /* 1155 * If sc == NULL or there are no valid disks, provider's error 1156 * should be set and g_mirror_start() should not be called at all. 1157 */ 1158 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1159 ("Provider's error should be set (error=%d)(mirror=%s).", 1160 bp->bio_to->error, bp->bio_to->name)); 1161 G_MIRROR_LOGREQ(3, bp, "Request received."); 1162 1163 switch (bp->bio_cmd) { 1164 case BIO_READ: 1165 case BIO_WRITE: 1166 case BIO_DELETE: 1167 break; 1168 case BIO_FLUSH: 1169 g_mirror_flush(sc, bp); 1170 return; 1171 case BIO_GETATTR: 1172 if (!strcmp(bp->bio_attribute, "GEOM::candelete")) { 1173 g_mirror_candelete(bp); 1174 return; 1175 } else if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1176 g_mirror_kernel_dump(bp); 1177 return; 1178 } 1179 /* FALLTHROUGH */ 1180 default: 1181 g_io_deliver(bp, EOPNOTSUPP); 1182 return; 1183 } 1184 mtx_lock(&sc->sc_queue_mtx); 1185 bioq_insert_tail(&sc->sc_queue, bp); 1186 mtx_unlock(&sc->sc_queue_mtx); 1187 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1188 wakeup(sc); 1189} 1190 1191/* 1192 * Return TRUE if the given request is colliding with a in-progress 1193 * synchronization request. 1194 */ 1195static int 1196g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1197{ 1198 struct g_mirror_disk *disk; 1199 struct bio *sbp; 1200 off_t rstart, rend, sstart, send; 1201 u_int i; 1202 1203 if (sc->sc_sync.ds_ndisks == 0) 1204 return (0); 1205 rstart = bp->bio_offset; 1206 rend = bp->bio_offset + bp->bio_length; 1207 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1208 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1209 continue; 1210 for (i = 0; i < g_mirror_syncreqs; i++) { 1211 sbp = disk->d_sync.ds_bios[i]; 1212 if (sbp == NULL) 1213 continue; 1214 sstart = sbp->bio_offset; 1215 send = sbp->bio_offset + sbp->bio_length; 1216 if (rend > sstart && rstart < send) 1217 return (1); 1218 } 1219 } 1220 return (0); 1221} 1222 1223/* 1224 * Return TRUE if the given sync request is colliding with a in-progress regular 1225 * request. 1226 */ 1227static int 1228g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1229{ 1230 off_t rstart, rend, sstart, send; 1231 struct bio *bp; 1232 1233 if (sc->sc_sync.ds_ndisks == 0) 1234 return (0); 1235 sstart = sbp->bio_offset; 1236 send = sbp->bio_offset + sbp->bio_length; 1237 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { 1238 rstart = bp->bio_offset; 1239 rend = bp->bio_offset + bp->bio_length; 1240 if (rend > sstart && rstart < send) 1241 return (1); 1242 } 1243 return (0); 1244} 1245 1246/* 1247 * Puts request onto delayed queue. 1248 */ 1249static void 1250g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1251{ 1252 1253 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1254 bioq_insert_head(&sc->sc_regular_delayed, bp); 1255} 1256 1257/* 1258 * Puts synchronization request onto delayed queue. 1259 */ 1260static void 1261g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1262{ 1263 1264 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1265 bioq_insert_tail(&sc->sc_sync_delayed, bp); 1266} 1267 1268/* 1269 * Releases delayed regular requests which don't collide anymore with sync 1270 * requests. 1271 */ 1272static void 1273g_mirror_regular_release(struct g_mirror_softc *sc) 1274{ 1275 struct bio *bp, *bp2; 1276 1277 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { 1278 if (g_mirror_sync_collision(sc, bp)) 1279 continue; 1280 bioq_remove(&sc->sc_regular_delayed, bp); 1281 G_MIRROR_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1282 mtx_lock(&sc->sc_queue_mtx); 1283 bioq_insert_head(&sc->sc_queue, bp); 1284#if 0 1285 /* 1286 * wakeup() is not needed, because this function is called from 1287 * the worker thread. 1288 */ 1289 wakeup(&sc->sc_queue); 1290#endif 1291 mtx_unlock(&sc->sc_queue_mtx); 1292 } 1293} 1294 1295/* 1296 * Releases delayed sync requests which don't collide anymore with regular 1297 * requests. 1298 */ 1299static void 1300g_mirror_sync_release(struct g_mirror_softc *sc) 1301{ 1302 struct bio *bp, *bp2; 1303 1304 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { 1305 if (g_mirror_regular_collision(sc, bp)) 1306 continue; 1307 bioq_remove(&sc->sc_sync_delayed, bp); 1308 G_MIRROR_LOGREQ(2, bp, 1309 "Releasing delayed synchronization request."); 1310 g_io_request(bp, bp->bio_from); 1311 } 1312} 1313 1314/* 1315 * Handle synchronization requests. 1316 * Every synchronization request is two-steps process: first, READ request is 1317 * send to active provider and then WRITE request (with read data) to the provider 1318 * being synchronized. When WRITE is finished, new synchronization request is 1319 * send. 1320 */ 1321static void 1322g_mirror_sync_request(struct bio *bp) 1323{ 1324 struct g_mirror_softc *sc; 1325 struct g_mirror_disk *disk; 1326 1327 bp->bio_from->index--; 1328 sc = bp->bio_from->geom->softc; 1329 disk = bp->bio_from->private; 1330 if (disk == NULL) { 1331 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1332 g_topology_lock(); 1333 g_mirror_kill_consumer(sc, bp->bio_from); 1334 g_topology_unlock(); 1335 free(bp->bio_data, M_MIRROR); 1336 g_destroy_bio(bp); 1337 sx_xlock(&sc->sc_lock); 1338 return; 1339 } 1340 1341 /* 1342 * Synchronization request. 1343 */ 1344 switch (bp->bio_cmd) { 1345 case BIO_READ: 1346 { 1347 struct g_consumer *cp; 1348 1349 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_read, 1350 bp->bio_error); 1351 1352 if (bp->bio_error != 0) { 1353 G_MIRROR_LOGREQ(0, bp, 1354 "Synchronization request failed (error=%d).", 1355 bp->bio_error); 1356 g_destroy_bio(bp); 1357 return; 1358 } 1359 G_MIRROR_LOGREQ(3, bp, 1360 "Synchronization request half-finished."); 1361 bp->bio_cmd = BIO_WRITE; 1362 bp->bio_cflags = 0; 1363 cp = disk->d_consumer; 1364 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1365 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1366 cp->acr, cp->acw, cp->ace)); 1367 cp->index++; 1368 g_io_request(bp, cp); 1369 return; 1370 } 1371 case BIO_WRITE: 1372 { 1373 struct g_mirror_disk_sync *sync; 1374 off_t offset; 1375 void *data; 1376 int i; 1377 1378 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_write, 1379 bp->bio_error); 1380 1381 if (bp->bio_error != 0) { 1382 G_MIRROR_LOGREQ(0, bp, 1383 "Synchronization request failed (error=%d).", 1384 bp->bio_error); 1385 g_destroy_bio(bp); 1386 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1387 g_mirror_event_send(disk, 1388 G_MIRROR_DISK_STATE_DISCONNECTED, 1389 G_MIRROR_EVENT_DONTWAIT); 1390 return; 1391 } 1392 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1393 sync = &disk->d_sync; 1394 if (sync->ds_offset >= sc->sc_mediasize || 1395 sync->ds_consumer == NULL || 1396 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1397 /* Don't send more synchronization requests. */ 1398 sync->ds_inflight--; 1399 if (sync->ds_bios != NULL) { 1400 i = (int)(uintptr_t)bp->bio_caller1; 1401 sync->ds_bios[i] = NULL; 1402 } 1403 free(bp->bio_data, M_MIRROR); 1404 g_destroy_bio(bp); 1405 if (sync->ds_inflight > 0) 1406 return; 1407 if (sync->ds_consumer == NULL || 1408 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1409 return; 1410 } 1411 /* Disk up-to-date, activate it. */ 1412 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1413 G_MIRROR_EVENT_DONTWAIT); 1414 return; 1415 } 1416 1417 /* Send next synchronization request. */ 1418 data = bp->bio_data; 1419 g_reset_bio(bp); 1420 bp->bio_cmd = BIO_READ; 1421 bp->bio_offset = sync->ds_offset; 1422 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1423 sync->ds_offset += bp->bio_length; 1424 bp->bio_done = g_mirror_sync_done; 1425 bp->bio_data = data; 1426 bp->bio_from = sync->ds_consumer; 1427 bp->bio_to = sc->sc_provider; 1428 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1429 sync->ds_consumer->index++; 1430 /* 1431 * Delay the request if it is colliding with a regular request. 1432 */ 1433 if (g_mirror_regular_collision(sc, bp)) 1434 g_mirror_sync_delay(sc, bp); 1435 else 1436 g_io_request(bp, sync->ds_consumer); 1437 1438 /* Release delayed requests if possible. */ 1439 g_mirror_regular_release(sc); 1440 1441 /* Find the smallest offset */ 1442 offset = sc->sc_mediasize; 1443 for (i = 0; i < g_mirror_syncreqs; i++) { 1444 bp = sync->ds_bios[i]; 1445 if (bp->bio_offset < offset) 1446 offset = bp->bio_offset; 1447 } 1448 if (sync->ds_offset_done + (MAXPHYS * 100) < offset) { 1449 /* Update offset_done on every 100 blocks. */ 1450 sync->ds_offset_done = offset; 1451 g_mirror_update_metadata(disk); 1452 } 1453 return; 1454 } 1455 default: 1456 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1457 bp->bio_cmd, sc->sc_name)); 1458 break; 1459 } 1460} 1461 1462static void 1463g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1464{ 1465 struct g_mirror_disk *disk; 1466 struct g_consumer *cp; 1467 struct bio *cbp; 1468 1469 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1470 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1471 break; 1472 } 1473 if (disk == NULL) { 1474 if (bp->bio_error == 0) 1475 bp->bio_error = ENXIO; 1476 g_io_deliver(bp, bp->bio_error); 1477 return; 1478 } 1479 cbp = g_clone_bio(bp); 1480 if (cbp == NULL) { 1481 if (bp->bio_error == 0) 1482 bp->bio_error = ENOMEM; 1483 g_io_deliver(bp, bp->bio_error); 1484 return; 1485 } 1486 /* 1487 * Fill in the component buf structure. 1488 */ 1489 cp = disk->d_consumer; 1490 cbp->bio_done = g_mirror_done; 1491 cbp->bio_to = cp->provider; 1492 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1493 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1494 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1495 cp->acw, cp->ace)); 1496 cp->index++; 1497 g_io_request(cbp, cp); 1498} 1499 1500static void 1501g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1502{ 1503 struct g_mirror_disk *disk; 1504 struct g_consumer *cp; 1505 struct bio *cbp; 1506 1507 disk = g_mirror_get_disk(sc); 1508 if (disk == NULL) { 1509 if (bp->bio_error == 0) 1510 bp->bio_error = ENXIO; 1511 g_io_deliver(bp, bp->bio_error); 1512 return; 1513 } 1514 cbp = g_clone_bio(bp); 1515 if (cbp == NULL) { 1516 if (bp->bio_error == 0) 1517 bp->bio_error = ENOMEM; 1518 g_io_deliver(bp, bp->bio_error); 1519 return; 1520 } 1521 /* 1522 * Fill in the component buf structure. 1523 */ 1524 cp = disk->d_consumer; 1525 cbp->bio_done = g_mirror_done; 1526 cbp->bio_to = cp->provider; 1527 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1528 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1529 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1530 cp->acw, cp->ace)); 1531 cp->index++; 1532 g_io_request(cbp, cp); 1533} 1534 1535#define TRACK_SIZE (1 * 1024 * 1024) 1536#define LOAD_SCALE 256 1537#define ABS(x) (((x) >= 0) ? (x) : (-(x))) 1538 1539static void 1540g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1541{ 1542 struct g_mirror_disk *disk, *dp; 1543 struct g_consumer *cp; 1544 struct bio *cbp; 1545 int prio, best; 1546 1547 /* Find a disk with the smallest load. */ 1548 disk = NULL; 1549 best = INT_MAX; 1550 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1551 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1552 continue; 1553 prio = dp->load; 1554 /* If disk head is precisely in position - highly prefer it. */ 1555 if (dp->d_last_offset == bp->bio_offset) 1556 prio -= 2 * LOAD_SCALE; 1557 else 1558 /* If disk head is close to position - prefer it. */ 1559 if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE) 1560 prio -= 1 * LOAD_SCALE; 1561 if (prio <= best) { 1562 disk = dp; 1563 best = prio; 1564 } 1565 } 1566 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1567 cbp = g_clone_bio(bp); 1568 if (cbp == NULL) { 1569 if (bp->bio_error == 0) 1570 bp->bio_error = ENOMEM; 1571 g_io_deliver(bp, bp->bio_error); 1572 return; 1573 } 1574 /* 1575 * Fill in the component buf structure. 1576 */ 1577 cp = disk->d_consumer; 1578 cbp->bio_done = g_mirror_done; 1579 cbp->bio_to = cp->provider; 1580 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1581 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1582 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1583 cp->acw, cp->ace)); 1584 cp->index++; 1585 /* Remember last head position */ 1586 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1587 /* Update loads. */ 1588 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1589 dp->load = (dp->d_consumer->index * LOAD_SCALE + 1590 dp->load * 7) / 8; 1591 } 1592 g_io_request(cbp, cp); 1593} 1594 1595static void 1596g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1597{ 1598 struct bio_queue_head queue; 1599 struct g_mirror_disk *disk; 1600 struct g_consumer *cp; 1601 struct bio *cbp; 1602 off_t left, mod, offset, slice; 1603 u_char *data; 1604 u_int ndisks; 1605 1606 if (bp->bio_length <= sc->sc_slice) { 1607 g_mirror_request_round_robin(sc, bp); 1608 return; 1609 } 1610 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1611 slice = bp->bio_length / ndisks; 1612 mod = slice % sc->sc_provider->sectorsize; 1613 if (mod != 0) 1614 slice += sc->sc_provider->sectorsize - mod; 1615 /* 1616 * Allocate all bios before sending any request, so we can 1617 * return ENOMEM in nice and clean way. 1618 */ 1619 left = bp->bio_length; 1620 offset = bp->bio_offset; 1621 data = bp->bio_data; 1622 bioq_init(&queue); 1623 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1624 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1625 continue; 1626 cbp = g_clone_bio(bp); 1627 if (cbp == NULL) { 1628 while ((cbp = bioq_takefirst(&queue)) != NULL) 1629 g_destroy_bio(cbp); 1630 if (bp->bio_error == 0) 1631 bp->bio_error = ENOMEM; 1632 g_io_deliver(bp, bp->bio_error); 1633 return; 1634 } 1635 bioq_insert_tail(&queue, cbp); 1636 cbp->bio_done = g_mirror_done; 1637 cbp->bio_caller1 = disk; 1638 cbp->bio_to = disk->d_consumer->provider; 1639 cbp->bio_offset = offset; 1640 cbp->bio_data = data; 1641 cbp->bio_length = MIN(left, slice); 1642 left -= cbp->bio_length; 1643 if (left == 0) 1644 break; 1645 offset += cbp->bio_length; 1646 data += cbp->bio_length; 1647 } 1648 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1649 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1650 disk = cbp->bio_caller1; 1651 cbp->bio_caller1 = NULL; 1652 cp = disk->d_consumer; 1653 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1654 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1655 cp->acr, cp->acw, cp->ace)); 1656 disk->d_consumer->index++; 1657 g_io_request(cbp, disk->d_consumer); 1658 } 1659} 1660 1661static void 1662g_mirror_register_request(struct bio *bp) 1663{ 1664 struct g_mirror_softc *sc; 1665 1666 sc = bp->bio_to->private; 1667 switch (bp->bio_cmd) { 1668 case BIO_READ: 1669 switch (sc->sc_balance) { 1670 case G_MIRROR_BALANCE_LOAD: 1671 g_mirror_request_load(sc, bp); 1672 break; 1673 case G_MIRROR_BALANCE_PREFER: 1674 g_mirror_request_prefer(sc, bp); 1675 break; 1676 case G_MIRROR_BALANCE_ROUND_ROBIN: 1677 g_mirror_request_round_robin(sc, bp); 1678 break; 1679 case G_MIRROR_BALANCE_SPLIT: 1680 g_mirror_request_split(sc, bp); 1681 break; 1682 } 1683 return; 1684 case BIO_WRITE: 1685 case BIO_DELETE: 1686 { 1687 struct g_mirror_disk *disk; 1688 struct g_mirror_disk_sync *sync; 1689 struct bio_queue_head queue; 1690 struct g_consumer *cp; 1691 struct bio *cbp; 1692 1693 /* 1694 * Delay the request if it is colliding with a synchronization 1695 * request. 1696 */ 1697 if (g_mirror_sync_collision(sc, bp)) { 1698 g_mirror_regular_delay(sc, bp); 1699 return; 1700 } 1701 1702 if (sc->sc_idle) 1703 g_mirror_unidle(sc); 1704 else 1705 sc->sc_last_write = time_uptime; 1706 1707 /* 1708 * Allocate all bios before sending any request, so we can 1709 * return ENOMEM in nice and clean way. 1710 */ 1711 bioq_init(&queue); 1712 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1713 sync = &disk->d_sync; 1714 switch (disk->d_state) { 1715 case G_MIRROR_DISK_STATE_ACTIVE: 1716 break; 1717 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1718 if (bp->bio_offset >= sync->ds_offset) 1719 continue; 1720 break; 1721 default: 1722 continue; 1723 } 1724 if (bp->bio_cmd == BIO_DELETE && 1725 (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0) 1726 continue; 1727 cbp = g_clone_bio(bp); 1728 if (cbp == NULL) { 1729 while ((cbp = bioq_takefirst(&queue)) != NULL) 1730 g_destroy_bio(cbp); 1731 if (bp->bio_error == 0) 1732 bp->bio_error = ENOMEM; 1733 g_io_deliver(bp, bp->bio_error); 1734 return; 1735 } 1736 bioq_insert_tail(&queue, cbp); 1737 cbp->bio_done = g_mirror_done; 1738 cp = disk->d_consumer; 1739 cbp->bio_caller1 = cp; 1740 cbp->bio_to = cp->provider; 1741 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1742 ("Consumer %s not opened (r%dw%de%d).", 1743 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1744 } 1745 if (bioq_first(&queue) == NULL) { 1746 g_io_deliver(bp, EOPNOTSUPP); 1747 return; 1748 } 1749 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1750 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1751 cp = cbp->bio_caller1; 1752 cbp->bio_caller1 = NULL; 1753 cp->index++; 1754 sc->sc_writes++; 1755 g_io_request(cbp, cp); 1756 } 1757 /* 1758 * Put request onto inflight queue, so we can check if new 1759 * synchronization requests don't collide with it. 1760 */ 1761 bioq_insert_tail(&sc->sc_inflight, bp); 1762 /* 1763 * Bump syncid on first write. 1764 */ 1765 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1766 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1767 g_mirror_bump_syncid(sc); 1768 } 1769 return; 1770 } 1771 default: 1772 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1773 bp->bio_cmd, sc->sc_name)); 1774 break; 1775 } 1776} 1777 1778static int 1779g_mirror_can_destroy(struct g_mirror_softc *sc) 1780{ 1781 struct g_geom *gp; 1782 struct g_consumer *cp; 1783 1784 g_topology_assert(); 1785 gp = sc->sc_geom; 1786 if (gp->softc == NULL) 1787 return (1); 1788 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0) 1789 return (0); 1790 LIST_FOREACH(cp, &gp->consumer, consumer) { 1791 if (g_mirror_is_busy(sc, cp)) 1792 return (0); 1793 } 1794 gp = sc->sc_sync.ds_geom; 1795 LIST_FOREACH(cp, &gp->consumer, consumer) { 1796 if (g_mirror_is_busy(sc, cp)) 1797 return (0); 1798 } 1799 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1800 sc->sc_name); 1801 return (1); 1802} 1803 1804static int 1805g_mirror_try_destroy(struct g_mirror_softc *sc) 1806{ 1807 1808 if (sc->sc_rootmount != NULL) { 1809 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1810 sc->sc_rootmount); 1811 root_mount_rel(sc->sc_rootmount); 1812 sc->sc_rootmount = NULL; 1813 } 1814 g_topology_lock(); 1815 if (!g_mirror_can_destroy(sc)) { 1816 g_topology_unlock(); 1817 return (0); 1818 } 1819 sc->sc_geom->softc = NULL; 1820 sc->sc_sync.ds_geom->softc = NULL; 1821 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1822 g_topology_unlock(); 1823 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1824 &sc->sc_worker); 1825 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1826 sx_xunlock(&sc->sc_lock); 1827 wakeup(&sc->sc_worker); 1828 sc->sc_worker = NULL; 1829 } else { 1830 g_topology_unlock(); 1831 g_mirror_destroy_device(sc); 1832 } 1833 return (1); 1834} 1835 1836/* 1837 * Worker thread. 1838 */ 1839static void 1840g_mirror_worker(void *arg) 1841{ 1842 struct g_mirror_softc *sc; 1843 struct g_mirror_event *ep; 1844 struct bio *bp; 1845 int timeout; 1846 1847 sc = arg; 1848 thread_lock(curthread); 1849 sched_prio(curthread, PRIBIO); 1850 thread_unlock(curthread); 1851 1852 sx_xlock(&sc->sc_lock); 1853 for (;;) { 1854 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1855 /* 1856 * First take a look at events. 1857 * This is important to handle events before any I/O requests. 1858 */ 1859 ep = g_mirror_event_get(sc); 1860 if (ep != NULL) { 1861 g_mirror_event_remove(sc, ep); 1862 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1863 /* Update only device status. */ 1864 G_MIRROR_DEBUG(3, 1865 "Running event for device %s.", 1866 sc->sc_name); 1867 ep->e_error = 0; 1868 g_mirror_update_device(sc, true); 1869 } else { 1870 /* Update disk status. */ 1871 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1872 g_mirror_get_diskname(ep->e_disk)); 1873 ep->e_error = g_mirror_update_disk(ep->e_disk, 1874 ep->e_state); 1875 if (ep->e_error == 0) 1876 g_mirror_update_device(sc, false); 1877 } 1878 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1879 KASSERT(ep->e_error == 0, 1880 ("Error cannot be handled.")); 1881 g_mirror_event_free(ep); 1882 } else { 1883 ep->e_flags |= G_MIRROR_EVENT_DONE; 1884 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1885 ep); 1886 mtx_lock(&sc->sc_events_mtx); 1887 wakeup(ep); 1888 mtx_unlock(&sc->sc_events_mtx); 1889 } 1890 if ((sc->sc_flags & 1891 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1892 if (g_mirror_try_destroy(sc)) { 1893 curthread->td_pflags &= ~TDP_GEOM; 1894 G_MIRROR_DEBUG(1, "Thread exiting."); 1895 kproc_exit(0); 1896 } 1897 } 1898 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1899 continue; 1900 } 1901 /* 1902 * Check if we can mark array as CLEAN and if we can't take 1903 * how much seconds should we wait. 1904 */ 1905 timeout = g_mirror_idle(sc, -1); 1906 /* 1907 * Now I/O requests. 1908 */ 1909 /* Get first request from the queue. */ 1910 mtx_lock(&sc->sc_queue_mtx); 1911 bp = bioq_takefirst(&sc->sc_queue); 1912 if (bp == NULL) { 1913 if ((sc->sc_flags & 1914 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1915 mtx_unlock(&sc->sc_queue_mtx); 1916 if (g_mirror_try_destroy(sc)) { 1917 curthread->td_pflags &= ~TDP_GEOM; 1918 G_MIRROR_DEBUG(1, "Thread exiting."); 1919 kproc_exit(0); 1920 } 1921 mtx_lock(&sc->sc_queue_mtx); 1922 if (bioq_first(&sc->sc_queue) != NULL) { 1923 mtx_unlock(&sc->sc_queue_mtx); 1924 continue; 1925 } 1926 } 1927 sx_xunlock(&sc->sc_lock); 1928 /* 1929 * XXX: We can miss an event here, because an event 1930 * can be added without sx-device-lock and without 1931 * mtx-queue-lock. Maybe I should just stop using 1932 * dedicated mutex for events synchronization and 1933 * stick with the queue lock? 1934 * The event will hang here until next I/O request 1935 * or next event is received. 1936 */ 1937 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1938 timeout * hz); 1939 sx_xlock(&sc->sc_lock); 1940 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1941 continue; 1942 } 1943 mtx_unlock(&sc->sc_queue_mtx); 1944 1945 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 1946 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1947 g_mirror_sync_request(bp); /* READ */ 1948 } else if (bp->bio_to != sc->sc_provider) { 1949 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 1950 g_mirror_regular_request(bp); 1951 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 1952 g_mirror_sync_request(bp); /* WRITE */ 1953 else { 1954 KASSERT(0, 1955 ("Invalid request cflags=0x%hx to=%s.", 1956 bp->bio_cflags, bp->bio_to->name)); 1957 } 1958 } else { 1959 g_mirror_register_request(bp); 1960 } 1961 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1962 } 1963} 1964 1965static void 1966g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 1967{ 1968 1969 sx_assert(&sc->sc_lock, SX_LOCKED); 1970 1971 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 1972 return; 1973 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1974 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1975 g_mirror_get_diskname(disk), sc->sc_name); 1976 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1977 } else if (sc->sc_idle && 1978 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1979 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1980 g_mirror_get_diskname(disk), sc->sc_name); 1981 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1982 } 1983} 1984 1985static void 1986g_mirror_sync_start(struct g_mirror_disk *disk) 1987{ 1988 struct g_mirror_softc *sc; 1989 struct g_consumer *cp; 1990 struct bio *bp; 1991 int error, i; 1992 1993 g_topology_assert_not(); 1994 sc = disk->d_softc; 1995 sx_assert(&sc->sc_lock, SX_LOCKED); 1996 1997 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1998 ("Disk %s is not marked for synchronization.", 1999 g_mirror_get_diskname(disk))); 2000 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2001 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 2002 sc->sc_state)); 2003 2004 sx_xunlock(&sc->sc_lock); 2005 g_topology_lock(); 2006 cp = g_new_consumer(sc->sc_sync.ds_geom); 2007 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 2008 error = g_attach(cp, sc->sc_provider); 2009 KASSERT(error == 0, 2010 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 2011 error = g_access(cp, 1, 0, 0); 2012 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 2013 g_topology_unlock(); 2014 sx_xlock(&sc->sc_lock); 2015 2016 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 2017 g_mirror_get_diskname(disk)); 2018 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0) 2019 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2020 KASSERT(disk->d_sync.ds_consumer == NULL, 2021 ("Sync consumer already exists (device=%s, disk=%s).", 2022 sc->sc_name, g_mirror_get_diskname(disk))); 2023 2024 disk->d_sync.ds_consumer = cp; 2025 disk->d_sync.ds_consumer->private = disk; 2026 disk->d_sync.ds_consumer->index = 0; 2027 2028 /* 2029 * Allocate memory for synchronization bios and initialize them. 2030 */ 2031 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 2032 M_MIRROR, M_WAITOK); 2033 for (i = 0; i < g_mirror_syncreqs; i++) { 2034 bp = g_alloc_bio(); 2035 disk->d_sync.ds_bios[i] = bp; 2036 bp->bio_parent = NULL; 2037 bp->bio_cmd = BIO_READ; 2038 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 2039 bp->bio_cflags = 0; 2040 bp->bio_offset = disk->d_sync.ds_offset; 2041 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 2042 disk->d_sync.ds_offset += bp->bio_length; 2043 bp->bio_done = g_mirror_sync_done; 2044 bp->bio_from = disk->d_sync.ds_consumer; 2045 bp->bio_to = sc->sc_provider; 2046 bp->bio_caller1 = (void *)(uintptr_t)i; 2047 } 2048 2049 /* Increase the number of disks in SYNCHRONIZING state. */ 2050 sc->sc_sync.ds_ndisks++; 2051 /* Set the number of in-flight synchronization requests. */ 2052 disk->d_sync.ds_inflight = g_mirror_syncreqs; 2053 2054 /* 2055 * Fire off first synchronization requests. 2056 */ 2057 for (i = 0; i < g_mirror_syncreqs; i++) { 2058 bp = disk->d_sync.ds_bios[i]; 2059 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 2060 disk->d_sync.ds_consumer->index++; 2061 /* 2062 * Delay the request if it is colliding with a regular request. 2063 */ 2064 if (g_mirror_regular_collision(sc, bp)) 2065 g_mirror_sync_delay(sc, bp); 2066 else 2067 g_io_request(bp, disk->d_sync.ds_consumer); 2068 } 2069} 2070 2071/* 2072 * Stop synchronization process. 2073 * type: 0 - synchronization finished 2074 * 1 - synchronization stopped 2075 */ 2076static void 2077g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 2078{ 2079 struct g_mirror_softc *sc; 2080 struct g_consumer *cp; 2081 2082 g_topology_assert_not(); 2083 sc = disk->d_softc; 2084 sx_assert(&sc->sc_lock, SX_LOCKED); 2085 2086 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2087 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2088 g_mirror_disk_state2str(disk->d_state))); 2089 if (disk->d_sync.ds_consumer == NULL) 2090 return; 2091 2092 if (type == 0) { 2093 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2094 sc->sc_name, g_mirror_get_diskname(disk)); 2095 } else /* if (type == 1) */ { 2096 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2097 sc->sc_name, g_mirror_get_diskname(disk)); 2098 } 2099 free(disk->d_sync.ds_bios, M_MIRROR); 2100 disk->d_sync.ds_bios = NULL; 2101 cp = disk->d_sync.ds_consumer; 2102 disk->d_sync.ds_consumer = NULL; 2103 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2104 sc->sc_sync.ds_ndisks--; 2105 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2106 g_topology_lock(); 2107 g_mirror_kill_consumer(sc, cp); 2108 g_topology_unlock(); 2109 sx_xlock(&sc->sc_lock); 2110} 2111 2112static void 2113g_mirror_launch_provider(struct g_mirror_softc *sc) 2114{ 2115 struct g_mirror_disk *disk; 2116 struct g_provider *pp, *dp; 2117 2118 sx_assert(&sc->sc_lock, SX_LOCKED); 2119 2120 g_topology_lock(); 2121 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 2122 pp->flags |= G_PF_DIRECT_RECEIVE; 2123 pp->mediasize = sc->sc_mediasize; 2124 pp->sectorsize = sc->sc_sectorsize; 2125 pp->stripesize = 0; 2126 pp->stripeoffset = 0; 2127 2128 /* Splitting of unmapped BIO's could work but isn't implemented now */ 2129 if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT) 2130 pp->flags |= G_PF_ACCEPT_UNMAPPED; 2131 2132 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2133 if (disk->d_consumer && disk->d_consumer->provider) { 2134 dp = disk->d_consumer->provider; 2135 if (dp->stripesize > pp->stripesize) { 2136 pp->stripesize = dp->stripesize; 2137 pp->stripeoffset = dp->stripeoffset; 2138 } 2139 /* A provider underneath us doesn't support unmapped */ 2140 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { 2141 G_MIRROR_DEBUG(0, "Cancelling unmapped " 2142 "because of %s.", dp->name); 2143 pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 2144 } 2145 } 2146 } 2147 pp->private = sc; 2148 sc->sc_refcnt++; 2149 sc->sc_provider = pp; 2150 g_error_provider(pp, 0); 2151 g_topology_unlock(); 2152 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2153 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 2154 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2155 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2156 g_mirror_sync_start(disk); 2157 } 2158} 2159 2160static void 2161g_mirror_destroy_provider(struct g_mirror_softc *sc) 2162{ 2163 struct g_mirror_disk *disk; 2164 struct bio *bp; 2165 2166 g_topology_assert_not(); 2167 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2168 sc->sc_name)); 2169 2170 g_topology_lock(); 2171 g_error_provider(sc->sc_provider, ENXIO); 2172 mtx_lock(&sc->sc_queue_mtx); 2173 while ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) { 2174 /* 2175 * Abort any pending I/O that wasn't generated by us. 2176 * Synchronization requests and requests destined for individual 2177 * mirror components can be destroyed immediately. 2178 */ 2179 if (bp->bio_to == sc->sc_provider && 2180 bp->bio_from->geom != sc->sc_sync.ds_geom) { 2181 g_io_deliver(bp, ENXIO); 2182 } else { 2183 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2184 free(bp->bio_data, M_MIRROR); 2185 g_destroy_bio(bp); 2186 } 2187 } 2188 mtx_unlock(&sc->sc_queue_mtx); 2189 g_wither_provider(sc->sc_provider, ENXIO); 2190 sc->sc_provider = NULL; 2191 G_MIRROR_DEBUG(0, "Device %s: provider destroyed.", sc->sc_name); 2192 g_topology_unlock(); 2193 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2194 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2195 g_mirror_sync_stop(disk, 1); 2196 } 2197} 2198 2199static void 2200g_mirror_go(void *arg) 2201{ 2202 struct g_mirror_softc *sc; 2203 2204 sc = arg; 2205 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2206 g_mirror_event_send(sc, 0, 2207 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2208} 2209 2210static u_int 2211g_mirror_determine_state(struct g_mirror_disk *disk) 2212{ 2213 struct g_mirror_softc *sc; 2214 u_int state; 2215 2216 sc = disk->d_softc; 2217 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2218 if ((disk->d_flags & 2219 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0 && 2220 (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 || 2221 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0)) { 2222 /* Disk does not need synchronization. */ 2223 state = G_MIRROR_DISK_STATE_ACTIVE; 2224 } else { 2225 if ((sc->sc_flags & 2226 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2227 (disk->d_flags & 2228 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2229 /* 2230 * We can start synchronization from 2231 * the stored offset. 2232 */ 2233 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2234 } else { 2235 state = G_MIRROR_DISK_STATE_STALE; 2236 } 2237 } 2238 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2239 /* 2240 * Reset all synchronization data for this disk, 2241 * because if it even was synchronized, it was 2242 * synchronized to disks with different syncid. 2243 */ 2244 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2245 disk->d_sync.ds_offset = 0; 2246 disk->d_sync.ds_offset_done = 0; 2247 disk->d_sync.ds_syncid = sc->sc_syncid; 2248 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2249 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2250 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2251 } else { 2252 state = G_MIRROR_DISK_STATE_STALE; 2253 } 2254 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2255 /* 2256 * Not good, NOT GOOD! 2257 * It means that mirror was started on stale disks 2258 * and more fresh disk just arrive. 2259 * If there were writes, mirror is broken, sorry. 2260 * I think the best choice here is don't touch 2261 * this disk and inform the user loudly. 2262 */ 2263 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2264 "disk (%s) arrives!! It will not be connected to the " 2265 "running device.", sc->sc_name, 2266 g_mirror_get_diskname(disk)); 2267 g_mirror_destroy_disk(disk); 2268 state = G_MIRROR_DISK_STATE_NONE; 2269 /* Return immediately, because disk was destroyed. */ 2270 return (state); 2271 } 2272 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2273 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2274 return (state); 2275} 2276 2277/* 2278 * Update device state. 2279 */ 2280static void 2281g_mirror_update_device(struct g_mirror_softc *sc, bool force) 2282{ 2283 struct g_mirror_disk *disk; 2284 u_int state; 2285 2286 sx_assert(&sc->sc_lock, SX_XLOCKED); 2287 2288 switch (sc->sc_state) { 2289 case G_MIRROR_DEVICE_STATE_STARTING: 2290 { 2291 struct g_mirror_disk *pdisk, *tdisk; 2292 u_int dirty, ndisks, genid, syncid; 2293 bool broken; 2294 2295 KASSERT(sc->sc_provider == NULL, 2296 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2297 /* 2298 * Are we ready? We are, if all disks are connected or 2299 * if we have any disks and 'force' is true. 2300 */ 2301 ndisks = g_mirror_ndisks(sc, -1); 2302 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2303 ; 2304 } else if (ndisks == 0) { 2305 /* 2306 * Disks went down in starting phase, so destroy 2307 * device. 2308 */ 2309 callout_drain(&sc->sc_callout); 2310 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2311 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2312 sc->sc_rootmount); 2313 root_mount_rel(sc->sc_rootmount); 2314 sc->sc_rootmount = NULL; 2315 return; 2316 } else { 2317 return; 2318 } 2319 2320 /* 2321 * Activate all disks with the biggest syncid. 2322 */ 2323 if (force) { 2324 /* 2325 * If 'force' is true, we have been called due to 2326 * timeout, so don't bother canceling timeout. 2327 */ 2328 ndisks = 0; 2329 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2330 if ((disk->d_flags & 2331 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2332 ndisks++; 2333 } 2334 } 2335 if (ndisks == 0) { 2336 /* No valid disks found, destroy device. */ 2337 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2338 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2339 __LINE__, sc->sc_rootmount); 2340 root_mount_rel(sc->sc_rootmount); 2341 sc->sc_rootmount = NULL; 2342 return; 2343 } 2344 } else { 2345 /* Cancel timeout. */ 2346 callout_drain(&sc->sc_callout); 2347 } 2348 2349 /* 2350 * Find the biggest genid. 2351 */ 2352 genid = 0; 2353 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2354 if (disk->d_genid > genid) 2355 genid = disk->d_genid; 2356 } 2357 sc->sc_genid = genid; 2358 /* 2359 * Remove all disks without the biggest genid. 2360 */ 2361 broken = false; 2362 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2363 if (disk->d_genid < genid) { 2364 G_MIRROR_DEBUG(0, 2365 "Component %s (device %s) broken, skipping.", 2366 g_mirror_get_diskname(disk), sc->sc_name); 2367 g_mirror_destroy_disk(disk); 2368 /* 2369 * Bump the syncid in case we discover a healthy 2370 * replacement disk after starting the mirror. 2371 */ 2372 broken = true; 2373 } 2374 } 2375 2376 /* 2377 * Find the biggest syncid. 2378 */ 2379 syncid = 0; 2380 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2381 if (disk->d_sync.ds_syncid > syncid) 2382 syncid = disk->d_sync.ds_syncid; 2383 } 2384 2385 /* 2386 * Here we need to look for dirty disks and if all disks 2387 * with the biggest syncid are dirty, we have to choose 2388 * one with the biggest priority and rebuild the rest. 2389 */ 2390 /* 2391 * Find the number of dirty disks with the biggest syncid. 2392 * Find the number of disks with the biggest syncid. 2393 * While here, find a disk with the biggest priority. 2394 */ 2395 dirty = ndisks = 0; 2396 pdisk = NULL; 2397 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2398 if (disk->d_sync.ds_syncid != syncid) 2399 continue; 2400 if ((disk->d_flags & 2401 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2402 continue; 2403 } 2404 ndisks++; 2405 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2406 dirty++; 2407 if (pdisk == NULL || 2408 pdisk->d_priority < disk->d_priority) { 2409 pdisk = disk; 2410 } 2411 } 2412 } 2413 if (dirty == 0) { 2414 /* No dirty disks at all, great. */ 2415 } else if (dirty == ndisks) { 2416 /* 2417 * Force synchronization for all dirty disks except one 2418 * with the biggest priority. 2419 */ 2420 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2421 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2422 "master disk for synchronization.", 2423 g_mirror_get_diskname(pdisk), sc->sc_name); 2424 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2425 if (disk->d_sync.ds_syncid != syncid) 2426 continue; 2427 if ((disk->d_flags & 2428 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2429 continue; 2430 } 2431 KASSERT((disk->d_flags & 2432 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2433 ("Disk %s isn't marked as dirty.", 2434 g_mirror_get_diskname(disk))); 2435 /* Skip the disk with the biggest priority. */ 2436 if (disk == pdisk) 2437 continue; 2438 disk->d_sync.ds_syncid = 0; 2439 } 2440 } else if (dirty < ndisks) { 2441 /* 2442 * Force synchronization for all dirty disks. 2443 * We have some non-dirty disks. 2444 */ 2445 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2446 if (disk->d_sync.ds_syncid != syncid) 2447 continue; 2448 if ((disk->d_flags & 2449 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2450 continue; 2451 } 2452 if ((disk->d_flags & 2453 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2454 continue; 2455 } 2456 disk->d_sync.ds_syncid = 0; 2457 } 2458 } 2459 2460 /* Reset hint. */ 2461 sc->sc_hint = NULL; 2462 sc->sc_syncid = syncid; 2463 if (force || broken) { 2464 /* Remember to bump syncid on first write. */ 2465 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2466 } 2467 state = G_MIRROR_DEVICE_STATE_RUNNING; 2468 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2469 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2470 g_mirror_device_state2str(state)); 2471 sc->sc_state = state; 2472 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2473 state = g_mirror_determine_state(disk); 2474 g_mirror_event_send(disk, state, 2475 G_MIRROR_EVENT_DONTWAIT); 2476 if (state == G_MIRROR_DISK_STATE_STALE) 2477 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2478 } 2479 break; 2480 } 2481 case G_MIRROR_DEVICE_STATE_RUNNING: 2482 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2483 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2484 /* 2485 * No active disks or no disks at all, 2486 * so destroy device. 2487 */ 2488 if (sc->sc_provider != NULL) 2489 g_mirror_destroy_provider(sc); 2490 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2491 break; 2492 } else if (g_mirror_ndisks(sc, 2493 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2494 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2495 /* 2496 * We have active disks, launch provider if it doesn't 2497 * exist. 2498 */ 2499 if (sc->sc_provider == NULL) 2500 g_mirror_launch_provider(sc); 2501 if (sc->sc_rootmount != NULL) { 2502 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2503 __LINE__, sc->sc_rootmount); 2504 root_mount_rel(sc->sc_rootmount); 2505 sc->sc_rootmount = NULL; 2506 } 2507 } 2508 /* 2509 * Genid should be bumped immediately, so do it here. 2510 */ 2511 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2512 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2513 g_mirror_bump_genid(sc); 2514 } 2515 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_NOW) != 0) { 2516 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID_NOW; 2517 g_mirror_bump_syncid(sc); 2518 } 2519 break; 2520 default: 2521 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2522 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2523 break; 2524 } 2525} 2526 2527/* 2528 * Update disk state and device state if needed. 2529 */ 2530#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2531 "Disk %s state changed from %s to %s (device %s).", \ 2532 g_mirror_get_diskname(disk), \ 2533 g_mirror_disk_state2str(disk->d_state), \ 2534 g_mirror_disk_state2str(state), sc->sc_name) 2535static int 2536g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2537{ 2538 struct g_mirror_softc *sc; 2539 2540 sc = disk->d_softc; 2541 sx_assert(&sc->sc_lock, SX_XLOCKED); 2542 2543again: 2544 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2545 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2546 g_mirror_disk_state2str(state)); 2547 switch (state) { 2548 case G_MIRROR_DISK_STATE_NEW: 2549 /* 2550 * Possible scenarios: 2551 * 1. New disk arrive. 2552 */ 2553 /* Previous state should be NONE. */ 2554 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2555 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2556 g_mirror_disk_state2str(disk->d_state))); 2557 DISK_STATE_CHANGED(); 2558 2559 disk->d_state = state; 2560 if (LIST_EMPTY(&sc->sc_disks)) 2561 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2562 else { 2563 struct g_mirror_disk *dp; 2564 2565 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2566 if (disk->d_priority >= dp->d_priority) { 2567 LIST_INSERT_BEFORE(dp, disk, d_next); 2568 dp = NULL; 2569 break; 2570 } 2571 if (LIST_NEXT(dp, d_next) == NULL) 2572 break; 2573 } 2574 if (dp != NULL) 2575 LIST_INSERT_AFTER(dp, disk, d_next); 2576 } 2577 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2578 sc->sc_name, g_mirror_get_diskname(disk)); 2579 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2580 break; 2581 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2582 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2583 g_mirror_device_state2str(sc->sc_state), 2584 g_mirror_get_diskname(disk), 2585 g_mirror_disk_state2str(disk->d_state))); 2586 state = g_mirror_determine_state(disk); 2587 if (state != G_MIRROR_DISK_STATE_NONE) 2588 goto again; 2589 break; 2590 case G_MIRROR_DISK_STATE_ACTIVE: 2591 /* 2592 * Possible scenarios: 2593 * 1. New disk does not need synchronization. 2594 * 2. Synchronization process finished successfully. 2595 */ 2596 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2597 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2598 g_mirror_device_state2str(sc->sc_state), 2599 g_mirror_get_diskname(disk), 2600 g_mirror_disk_state2str(disk->d_state))); 2601 /* Previous state should be NEW or SYNCHRONIZING. */ 2602 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2603 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2604 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2605 g_mirror_disk_state2str(disk->d_state))); 2606 DISK_STATE_CHANGED(); 2607 2608 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2609 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2610 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2611 g_mirror_sync_stop(disk, 0); 2612 } 2613 disk->d_state = state; 2614 disk->d_sync.ds_offset = 0; 2615 disk->d_sync.ds_offset_done = 0; 2616 g_mirror_update_idle(sc, disk); 2617 g_mirror_update_metadata(disk); 2618 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2619 sc->sc_name, g_mirror_get_diskname(disk)); 2620 break; 2621 case G_MIRROR_DISK_STATE_STALE: 2622 /* 2623 * Possible scenarios: 2624 * 1. Stale disk was connected. 2625 */ 2626 /* Previous state should be NEW. */ 2627 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2628 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2629 g_mirror_disk_state2str(disk->d_state))); 2630 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2631 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2632 g_mirror_device_state2str(sc->sc_state), 2633 g_mirror_get_diskname(disk), 2634 g_mirror_disk_state2str(disk->d_state))); 2635 /* 2636 * STALE state is only possible if device is marked 2637 * NOAUTOSYNC. 2638 */ 2639 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2640 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2641 g_mirror_device_state2str(sc->sc_state), 2642 g_mirror_get_diskname(disk), 2643 g_mirror_disk_state2str(disk->d_state))); 2644 DISK_STATE_CHANGED(); 2645 2646 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2647 disk->d_state = state; 2648 g_mirror_update_metadata(disk); 2649 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2650 sc->sc_name, g_mirror_get_diskname(disk)); 2651 break; 2652 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2653 /* 2654 * Possible scenarios: 2655 * 1. Disk which needs synchronization was connected. 2656 */ 2657 /* Previous state should be NEW. */ 2658 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2659 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2660 g_mirror_disk_state2str(disk->d_state))); 2661 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2662 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2663 g_mirror_device_state2str(sc->sc_state), 2664 g_mirror_get_diskname(disk), 2665 g_mirror_disk_state2str(disk->d_state))); 2666 DISK_STATE_CHANGED(); 2667 2668 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2669 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2670 disk->d_state = state; 2671 if (sc->sc_provider != NULL) { 2672 g_mirror_sync_start(disk); 2673 g_mirror_update_metadata(disk); 2674 } 2675 break; 2676 case G_MIRROR_DISK_STATE_DISCONNECTED: 2677 /* 2678 * Possible scenarios: 2679 * 1. Device wasn't running yet, but disk disappear. 2680 * 2. Disk was active and disapppear. 2681 * 3. Disk disappear during synchronization process. 2682 */ 2683 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2684 /* 2685 * Previous state should be ACTIVE, STALE or 2686 * SYNCHRONIZING. 2687 */ 2688 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2689 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2690 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2691 ("Wrong disk state (%s, %s).", 2692 g_mirror_get_diskname(disk), 2693 g_mirror_disk_state2str(disk->d_state))); 2694 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2695 /* Previous state should be NEW. */ 2696 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2697 ("Wrong disk state (%s, %s).", 2698 g_mirror_get_diskname(disk), 2699 g_mirror_disk_state2str(disk->d_state))); 2700 /* 2701 * Reset bumping syncid if disk disappeared in STARTING 2702 * state. 2703 */ 2704 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2705 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2706#ifdef INVARIANTS 2707 } else { 2708 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2709 sc->sc_name, 2710 g_mirror_device_state2str(sc->sc_state), 2711 g_mirror_get_diskname(disk), 2712 g_mirror_disk_state2str(disk->d_state))); 2713#endif 2714 } 2715 DISK_STATE_CHANGED(); 2716 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2717 sc->sc_name, g_mirror_get_diskname(disk)); 2718 2719 g_mirror_destroy_disk(disk); 2720 break; 2721 case G_MIRROR_DISK_STATE_DESTROY: 2722 { 2723 int error; 2724 2725 error = g_mirror_clear_metadata(disk); 2726 if (error != 0) { 2727 G_MIRROR_DEBUG(0, 2728 "Device %s: failed to clear metadata on %s: %d.", 2729 sc->sc_name, g_mirror_get_diskname(disk), error); 2730 break; 2731 } 2732 DISK_STATE_CHANGED(); 2733 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2734 sc->sc_name, g_mirror_get_diskname(disk)); 2735 2736 g_mirror_destroy_disk(disk); 2737 sc->sc_ndisks--; 2738 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2739 g_mirror_update_metadata(disk); 2740 } 2741 break; 2742 } 2743 default: 2744 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2745 break; 2746 } 2747 return (0); 2748} 2749#undef DISK_STATE_CHANGED 2750 2751int 2752g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2753{ 2754 struct g_provider *pp; 2755 u_char *buf; 2756 int error; 2757 2758 g_topology_assert(); 2759 2760 error = g_access(cp, 1, 0, 0); 2761 if (error != 0) 2762 return (error); 2763 pp = cp->provider; 2764 g_topology_unlock(); 2765 /* Metadata are stored on last sector. */ 2766 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2767 &error); 2768 g_topology_lock(); 2769 g_access(cp, -1, 0, 0); 2770 if (buf == NULL) { 2771 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2772 cp->provider->name, error); 2773 return (error); 2774 } 2775 2776 /* Decode metadata. */ 2777 error = mirror_metadata_decode(buf, md); 2778 g_free(buf); 2779 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2780 return (EINVAL); 2781 if (md->md_version > G_MIRROR_VERSION) { 2782 G_MIRROR_DEBUG(0, 2783 "Kernel module is too old to handle metadata from %s.", 2784 cp->provider->name); 2785 return (EINVAL); 2786 } 2787 if (error != 0) { 2788 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2789 cp->provider->name); 2790 return (error); 2791 } 2792 2793 return (0); 2794} 2795 2796static int 2797g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2798 struct g_mirror_metadata *md) 2799{ 2800 2801 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2802 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2803 pp->name, md->md_did); 2804 return (EEXIST); 2805 } 2806 if (md->md_all != sc->sc_ndisks) { 2807 G_MIRROR_DEBUG(1, 2808 "Invalid '%s' field on disk %s (device %s), skipping.", 2809 "md_all", pp->name, sc->sc_name); 2810 return (EINVAL); 2811 } 2812 if (md->md_slice != sc->sc_slice) { 2813 G_MIRROR_DEBUG(1, 2814 "Invalid '%s' field on disk %s (device %s), skipping.", 2815 "md_slice", pp->name, sc->sc_name); 2816 return (EINVAL); 2817 } 2818 if (md->md_balance != sc->sc_balance) { 2819 G_MIRROR_DEBUG(1, 2820 "Invalid '%s' field on disk %s (device %s), skipping.", 2821 "md_balance", pp->name, sc->sc_name); 2822 return (EINVAL); 2823 } 2824#if 0 2825 if (md->md_mediasize != sc->sc_mediasize) { 2826 G_MIRROR_DEBUG(1, 2827 "Invalid '%s' field on disk %s (device %s), skipping.", 2828 "md_mediasize", pp->name, sc->sc_name); 2829 return (EINVAL); 2830 } 2831#endif 2832 if (sc->sc_mediasize > pp->mediasize) { 2833 G_MIRROR_DEBUG(1, 2834 "Invalid size of disk %s (device %s), skipping.", pp->name, 2835 sc->sc_name); 2836 return (EINVAL); 2837 } 2838 if (md->md_sectorsize != sc->sc_sectorsize) { 2839 G_MIRROR_DEBUG(1, 2840 "Invalid '%s' field on disk %s (device %s), skipping.", 2841 "md_sectorsize", pp->name, sc->sc_name); 2842 return (EINVAL); 2843 } 2844 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2845 G_MIRROR_DEBUG(1, 2846 "Invalid sector size of disk %s (device %s), skipping.", 2847 pp->name, sc->sc_name); 2848 return (EINVAL); 2849 } 2850 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2851 G_MIRROR_DEBUG(1, 2852 "Invalid device flags on disk %s (device %s), skipping.", 2853 pp->name, sc->sc_name); 2854 return (EINVAL); 2855 } 2856 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2857 G_MIRROR_DEBUG(1, 2858 "Invalid disk flags on disk %s (device %s), skipping.", 2859 pp->name, sc->sc_name); 2860 return (EINVAL); 2861 } 2862 return (0); 2863} 2864 2865int 2866g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2867 struct g_mirror_metadata *md) 2868{ 2869 struct g_mirror_disk *disk; 2870 int error; 2871 2872 g_topology_assert_not(); 2873 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2874 2875 error = g_mirror_check_metadata(sc, pp, md); 2876 if (error != 0) 2877 return (error); 2878 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2879 md->md_genid < sc->sc_genid) { 2880 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2881 pp->name, sc->sc_name); 2882 return (EINVAL); 2883 } 2884 disk = g_mirror_init_disk(sc, pp, md, &error); 2885 if (disk == NULL) 2886 return (error); 2887 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2888 G_MIRROR_EVENT_WAIT); 2889 if (error != 0) 2890 return (error); 2891 if (md->md_version < G_MIRROR_VERSION) { 2892 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2893 pp->name, md->md_version, G_MIRROR_VERSION); 2894 g_mirror_update_metadata(disk); 2895 } 2896 return (0); 2897} 2898 2899static void 2900g_mirror_destroy_delayed(void *arg, int flag) 2901{ 2902 struct g_mirror_softc *sc; 2903 int error; 2904 2905 if (flag == EV_CANCEL) { 2906 G_MIRROR_DEBUG(1, "Destroying canceled."); 2907 return; 2908 } 2909 sc = arg; 2910 g_topology_unlock(); 2911 sx_xlock(&sc->sc_lock); 2912 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2913 ("DESTROY flag set on %s.", sc->sc_name)); 2914 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0, 2915 ("DESTROYING flag not set on %s.", sc->sc_name)); 2916 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 2917 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 2918 if (error != 0) { 2919 G_MIRROR_DEBUG(0, "Cannot destroy %s (error=%d).", 2920 sc->sc_name, error); 2921 sx_xunlock(&sc->sc_lock); 2922 } 2923 g_topology_lock(); 2924} 2925 2926static int 2927g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2928{ 2929 struct g_mirror_softc *sc; 2930 int error = 0; 2931 2932 g_topology_assert(); 2933 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2934 acw, ace); 2935 2936 sc = pp->private; 2937 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 2938 2939 g_topology_unlock(); 2940 sx_xlock(&sc->sc_lock); 2941 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 2942 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0 || 2943 LIST_EMPTY(&sc->sc_disks)) { 2944 if (acr > 0 || acw > 0 || ace > 0) 2945 error = ENXIO; 2946 goto end; 2947 } 2948 sc->sc_provider_open += acr + acw + ace; 2949 if (pp->acw + acw == 0) 2950 g_mirror_idle(sc, 0); 2951 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0 && 2952 sc->sc_provider_open == 0) 2953 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, sc, NULL); 2954end: 2955 sx_xunlock(&sc->sc_lock); 2956 g_topology_lock(); 2957 return (error); 2958} 2959 2960struct g_geom * 2961g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md, 2962 u_int type) 2963{ 2964 struct g_mirror_softc *sc; 2965 struct g_geom *gp; 2966 int error, timeout; 2967 2968 g_topology_assert(); 2969 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2970 md->md_mid); 2971 2972 /* One disk is minimum. */ 2973 if (md->md_all < 1) 2974 return (NULL); 2975 /* 2976 * Action geom. 2977 */ 2978 gp = g_new_geomf(mp, "%s", md->md_name); 2979 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2980 gp->start = g_mirror_start; 2981 gp->orphan = g_mirror_orphan; 2982 gp->access = g_mirror_access; 2983 gp->dumpconf = g_mirror_dumpconf; 2984 2985 sc->sc_type = type; 2986 sc->sc_id = md->md_mid; 2987 sc->sc_slice = md->md_slice; 2988 sc->sc_balance = md->md_balance; 2989 sc->sc_mediasize = md->md_mediasize; 2990 sc->sc_sectorsize = md->md_sectorsize; 2991 sc->sc_ndisks = md->md_all; 2992 sc->sc_flags = md->md_mflags; 2993 sc->sc_bump_id = 0; 2994 sc->sc_idle = 1; 2995 sc->sc_last_write = time_uptime; 2996 sc->sc_writes = 0; 2997 sc->sc_refcnt = 1; 2998 sx_init(&sc->sc_lock, "gmirror:lock"); 2999 bioq_init(&sc->sc_queue); 3000 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 3001 bioq_init(&sc->sc_regular_delayed); 3002 bioq_init(&sc->sc_inflight); 3003 bioq_init(&sc->sc_sync_delayed); 3004 LIST_INIT(&sc->sc_disks); 3005 TAILQ_INIT(&sc->sc_events); 3006 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 3007 callout_init(&sc->sc_callout, 1); 3008 mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF); 3009 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 3010 gp->softc = sc; 3011 sc->sc_geom = gp; 3012 sc->sc_provider = NULL; 3013 sc->sc_provider_open = 0; 3014 /* 3015 * Synchronization geom. 3016 */ 3017 gp = g_new_geomf(mp, "%s.sync", md->md_name); 3018 gp->softc = sc; 3019 gp->orphan = g_mirror_orphan; 3020 sc->sc_sync.ds_geom = gp; 3021 sc->sc_sync.ds_ndisks = 0; 3022 error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 3023 "g_mirror %s", md->md_name); 3024 if (error != 0) { 3025 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 3026 sc->sc_name); 3027 g_destroy_geom(sc->sc_sync.ds_geom); 3028 g_destroy_geom(sc->sc_geom); 3029 g_mirror_free_device(sc); 3030 return (NULL); 3031 } 3032 3033 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 3034 sc->sc_name, sc->sc_ndisks, sc->sc_id); 3035 3036 sc->sc_rootmount = root_mount_hold("GMIRROR"); 3037 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 3038 /* 3039 * Run timeout. 3040 */ 3041 timeout = g_mirror_timeout * hz; 3042 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 3043 return (sc->sc_geom); 3044} 3045 3046int 3047g_mirror_destroy(struct g_mirror_softc *sc, int how) 3048{ 3049 struct g_mirror_disk *disk; 3050 3051 g_topology_assert_not(); 3052 sx_assert(&sc->sc_lock, SX_XLOCKED); 3053 3054 if (sc->sc_provider_open != 0) { 3055 switch (how) { 3056 case G_MIRROR_DESTROY_SOFT: 3057 G_MIRROR_DEBUG(1, 3058 "Device %s is still open (%d).", sc->sc_name, 3059 sc->sc_provider_open); 3060 return (EBUSY); 3061 case G_MIRROR_DESTROY_DELAYED: 3062 G_MIRROR_DEBUG(1, 3063 "Device %s will be destroyed on last close.", 3064 sc->sc_name); 3065 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 3066 if (disk->d_state == 3067 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3068 g_mirror_sync_stop(disk, 1); 3069 } 3070 } 3071 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROYING; 3072 return (EBUSY); 3073 case G_MIRROR_DESTROY_HARD: 3074 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 3075 "can't be definitely removed.", sc->sc_name); 3076 } 3077 } 3078 3079 g_topology_lock(); 3080 if (sc->sc_geom->softc == NULL) { 3081 g_topology_unlock(); 3082 return (0); 3083 } 3084 sc->sc_geom->softc = NULL; 3085 sc->sc_sync.ds_geom->softc = NULL; 3086 g_topology_unlock(); 3087 3088 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 3089 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 3090 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3091 sx_xunlock(&sc->sc_lock); 3092 mtx_lock(&sc->sc_queue_mtx); 3093 wakeup(sc); 3094 mtx_unlock(&sc->sc_queue_mtx); 3095 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3096 while (sc->sc_worker != NULL) 3097 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 3098 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3099 sx_xlock(&sc->sc_lock); 3100 g_mirror_destroy_device(sc); 3101 return (0); 3102} 3103 3104static void 3105g_mirror_taste_orphan(struct g_consumer *cp) 3106{ 3107 3108 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3109 cp->provider->name)); 3110} 3111 3112static struct g_geom * 3113g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3114{ 3115 struct g_mirror_metadata md; 3116 struct g_mirror_softc *sc; 3117 struct g_consumer *cp; 3118 struct g_geom *gp; 3119 int error; 3120 3121 g_topology_assert(); 3122 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3123 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 3124 3125 gp = g_new_geomf(mp, "mirror:taste"); 3126 /* 3127 * This orphan function should be never called. 3128 */ 3129 gp->orphan = g_mirror_taste_orphan; 3130 cp = g_new_consumer(gp); 3131 g_attach(cp, pp); 3132 error = g_mirror_read_metadata(cp, &md); 3133 g_detach(cp); 3134 g_destroy_consumer(cp); 3135 g_destroy_geom(gp); 3136 if (error != 0) 3137 return (NULL); 3138 gp = NULL; 3139 3140 if (md.md_provider[0] != '\0' && 3141 !g_compare_names(md.md_provider, pp->name)) 3142 return (NULL); 3143 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3144 return (NULL); 3145 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 3146 G_MIRROR_DEBUG(0, 3147 "Device %s: provider %s marked as inactive, skipping.", 3148 md.md_name, pp->name); 3149 return (NULL); 3150 } 3151 if (g_mirror_debug >= 2) 3152 mirror_metadata_dump(&md); 3153 3154 /* 3155 * Let's check if device already exists. 3156 */ 3157 sc = NULL; 3158 LIST_FOREACH(gp, &mp->geom, geom) { 3159 sc = gp->softc; 3160 if (sc == NULL) 3161 continue; 3162 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 3163 continue; 3164 if (sc->sc_sync.ds_geom == gp) 3165 continue; 3166 if (strcmp(md.md_name, sc->sc_name) != 0) 3167 continue; 3168 if (md.md_mid != sc->sc_id) { 3169 G_MIRROR_DEBUG(0, "Device %s already configured.", 3170 sc->sc_name); 3171 return (NULL); 3172 } 3173 break; 3174 } 3175 if (gp == NULL) { 3176 gp = g_mirror_create(mp, &md, G_MIRROR_TYPE_AUTOMATIC); 3177 if (gp == NULL) { 3178 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3179 md.md_name); 3180 return (NULL); 3181 } 3182 sc = gp->softc; 3183 } 3184 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3185 g_topology_unlock(); 3186 sx_xlock(&sc->sc_lock); 3187 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; 3188 error = g_mirror_add_disk(sc, pp, &md); 3189 if (error != 0) { 3190 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3191 pp->name, gp->name, error); 3192 if (LIST_EMPTY(&sc->sc_disks)) { 3193 g_cancel_event(sc); 3194 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3195 g_topology_lock(); 3196 return (NULL); 3197 } 3198 gp = NULL; 3199 } 3200 sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; 3201 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3202 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3203 g_topology_lock(); 3204 return (NULL); 3205 } 3206 sx_xunlock(&sc->sc_lock); 3207 g_topology_lock(); 3208 return (gp); 3209} 3210 3211static void 3212g_mirror_resize(struct g_consumer *cp) 3213{ 3214 struct g_mirror_disk *disk; 3215 3216 g_topology_assert(); 3217 g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name); 3218 3219 disk = cp->private; 3220 if (disk == NULL) 3221 return; 3222 g_topology_unlock(); 3223 g_mirror_update_metadata(disk); 3224 g_topology_lock(); 3225} 3226 3227static int 3228g_mirror_destroy_geom(struct gctl_req *req __unused, 3229 struct g_class *mp __unused, struct g_geom *gp) 3230{ 3231 struct g_mirror_softc *sc; 3232 int error; 3233 3234 g_topology_unlock(); 3235 sc = gp->softc; 3236 sx_xlock(&sc->sc_lock); 3237 g_cancel_event(sc); 3238 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3239 if (error != 0) 3240 sx_xunlock(&sc->sc_lock); 3241 g_topology_lock(); 3242 return (error); 3243} 3244 3245static void 3246g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3247 struct g_consumer *cp, struct g_provider *pp) 3248{ 3249 struct g_mirror_softc *sc; 3250 3251 g_topology_assert(); 3252 3253 sc = gp->softc; 3254 if (sc == NULL) 3255 return; 3256 /* Skip synchronization geom. */ 3257 if (gp == sc->sc_sync.ds_geom) 3258 return; 3259 if (pp != NULL) { 3260 /* Nothing here. */ 3261 } else if (cp != NULL) { 3262 struct g_mirror_disk *disk; 3263 3264 disk = cp->private; 3265 if (disk == NULL) 3266 return; 3267 g_topology_unlock(); 3268 sx_xlock(&sc->sc_lock); 3269 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3270 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3271 sbuf_printf(sb, "%s<Synchronized>", indent); 3272 if (disk->d_sync.ds_offset == 0) 3273 sbuf_printf(sb, "0%%"); 3274 else { 3275 sbuf_printf(sb, "%u%%", 3276 (u_int)((disk->d_sync.ds_offset * 100) / 3277 sc->sc_provider->mediasize)); 3278 } 3279 sbuf_printf(sb, "</Synchronized>\n"); 3280 if (disk->d_sync.ds_offset > 0) { 3281 sbuf_printf(sb, "%s<BytesSynced>%jd" 3282 "</BytesSynced>\n", indent, 3283 (intmax_t)disk->d_sync.ds_offset); 3284 } 3285 } 3286 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3287 disk->d_sync.ds_syncid); 3288 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3289 disk->d_genid); 3290 sbuf_printf(sb, "%s<Flags>", indent); 3291 if (disk->d_flags == 0) 3292 sbuf_printf(sb, "NONE"); 3293 else { 3294 int first = 1; 3295 3296#define ADD_FLAG(flag, name) do { \ 3297 if ((disk->d_flags & (flag)) != 0) { \ 3298 if (!first) \ 3299 sbuf_printf(sb, ", "); \ 3300 else \ 3301 first = 0; \ 3302 sbuf_printf(sb, name); \ 3303 } \ 3304} while (0) 3305 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3306 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3307 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3308 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3309 "SYNCHRONIZING"); 3310 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3311 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3312#undef ADD_FLAG 3313 } 3314 sbuf_printf(sb, "</Flags>\n"); 3315 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3316 disk->d_priority); 3317 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3318 g_mirror_disk_state2str(disk->d_state)); 3319 sx_xunlock(&sc->sc_lock); 3320 g_topology_lock(); 3321 } else { 3322 g_topology_unlock(); 3323 sx_xlock(&sc->sc_lock); 3324 sbuf_printf(sb, "%s<Type>", indent); 3325 switch (sc->sc_type) { 3326 case G_MIRROR_TYPE_AUTOMATIC: 3327 sbuf_printf(sb, "AUTOMATIC"); 3328 break; 3329 case G_MIRROR_TYPE_MANUAL: 3330 sbuf_printf(sb, "MANUAL"); 3331 break; 3332 default: 3333 sbuf_printf(sb, "UNKNOWN"); 3334 break; 3335 } 3336 sbuf_printf(sb, "</Type>\n"); 3337 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3338 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3339 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3340 sbuf_printf(sb, "%s<Flags>", indent); 3341 if (sc->sc_flags == 0) 3342 sbuf_printf(sb, "NONE"); 3343 else { 3344 int first = 1; 3345 3346#define ADD_FLAG(flag, name) do { \ 3347 if ((sc->sc_flags & (flag)) != 0) { \ 3348 if (!first) \ 3349 sbuf_printf(sb, ", "); \ 3350 else \ 3351 first = 0; \ 3352 sbuf_printf(sb, name); \ 3353 } \ 3354} while (0) 3355 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3356 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3357#undef ADD_FLAG 3358 } 3359 sbuf_printf(sb, "</Flags>\n"); 3360 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3361 (u_int)sc->sc_slice); 3362 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3363 balance_name(sc->sc_balance)); 3364 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3365 sc->sc_ndisks); 3366 sbuf_printf(sb, "%s<State>", indent); 3367 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3368 sbuf_printf(sb, "%s", "STARTING"); 3369 else if (sc->sc_ndisks == 3370 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3371 sbuf_printf(sb, "%s", "COMPLETE"); 3372 else 3373 sbuf_printf(sb, "%s", "DEGRADED"); 3374 sbuf_printf(sb, "</State>\n"); 3375 sx_xunlock(&sc->sc_lock); 3376 g_topology_lock(); 3377 } 3378} 3379 3380static void 3381g_mirror_shutdown_post_sync(void *arg, int howto) 3382{ 3383 struct g_class *mp; 3384 struct g_geom *gp, *gp2; 3385 struct g_mirror_softc *sc; 3386 int error; 3387 3388 if (panicstr != NULL) 3389 return; 3390 3391 mp = arg; 3392 g_topology_lock(); 3393 g_mirror_shutdown = 1; 3394 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3395 if ((sc = gp->softc) == NULL) 3396 continue; 3397 /* Skip synchronization geom. */ 3398 if (gp == sc->sc_sync.ds_geom) 3399 continue; 3400 g_topology_unlock(); 3401 sx_xlock(&sc->sc_lock); 3402 g_mirror_idle(sc, -1); 3403 g_cancel_event(sc); 3404 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3405 if (error != 0) 3406 sx_xunlock(&sc->sc_lock); 3407 g_topology_lock(); 3408 } 3409 g_topology_unlock(); 3410} 3411 3412static void 3413g_mirror_init(struct g_class *mp) 3414{ 3415 3416 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3417 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3418 if (g_mirror_post_sync == NULL) 3419 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3420} 3421 3422static void 3423g_mirror_fini(struct g_class *mp) 3424{ 3425 3426 if (g_mirror_post_sync != NULL) 3427 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); 3428} 3429 3430DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3431