g_mirror.c revision 327080
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 327080 2017-12-22 16:14:20Z markj $"); 29 30#include <sys/param.h> 31#include <sys/systm.h> 32#include <sys/fail.h> 33#include <sys/kernel.h> 34#include <sys/module.h> 35#include <sys/limits.h> 36#include <sys/lock.h> 37#include <sys/mutex.h> 38#include <sys/bio.h> 39#include <sys/sbuf.h> 40#include <sys/sysctl.h> 41#include <sys/malloc.h> 42#include <sys/eventhandler.h> 43#include <vm/uma.h> 44#include <geom/geom.h> 45#include <sys/proc.h> 46#include <sys/kthread.h> 47#include <sys/sched.h> 48#include <geom/mirror/g_mirror.h> 49 50FEATURE(geom_mirror, "GEOM mirroring support"); 51 52static MALLOC_DEFINE(M_MIRROR, "mirror_data", "GEOM_MIRROR Data"); 53 54SYSCTL_DECL(_kern_geom); 55static SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, 56 "GEOM_MIRROR stuff"); 57int g_mirror_debug = 0; 58SYSCTL_INT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RWTUN, &g_mirror_debug, 0, 59 "Debug level"); 60static u_int g_mirror_timeout = 4; 61SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RWTUN, &g_mirror_timeout, 62 0, "Time to wait on all mirror components"); 63static u_int g_mirror_idletime = 5; 64SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RWTUN, 65 &g_mirror_idletime, 0, "Mark components as clean when idling"); 66static u_int g_mirror_disconnect_on_failure = 1; 67SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN, 68 &g_mirror_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 69static u_int g_mirror_syncreqs = 2; 70SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_requests, CTLFLAG_RDTUN, 71 &g_mirror_syncreqs, 0, "Parallel synchronization I/O requests."); 72static u_int g_mirror_sync_period = 5; 73SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_update_period, CTLFLAG_RWTUN, 74 &g_mirror_sync_period, 0, 75 "Metadata update period during synchronization, in seconds"); 76 77#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 78 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 79 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 80 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 81} while (0) 82 83static eventhandler_tag g_mirror_post_sync = NULL; 84static int g_mirror_shutdown = 0; 85 86static g_ctl_destroy_geom_t g_mirror_destroy_geom; 87static g_taste_t g_mirror_taste; 88static g_init_t g_mirror_init; 89static g_fini_t g_mirror_fini; 90static g_provgone_t g_mirror_providergone; 91static g_resize_t g_mirror_resize; 92 93struct g_class g_mirror_class = { 94 .name = G_MIRROR_CLASS_NAME, 95 .version = G_VERSION, 96 .ctlreq = g_mirror_config, 97 .taste = g_mirror_taste, 98 .destroy_geom = g_mirror_destroy_geom, 99 .init = g_mirror_init, 100 .fini = g_mirror_fini, 101 .providergone = g_mirror_providergone, 102 .resize = g_mirror_resize 103}; 104 105 106static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 107static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 108static void g_mirror_update_device(struct g_mirror_softc *sc, bool force); 109static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 110 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 111static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 112static void g_mirror_register_request(struct bio *bp); 113static void g_mirror_sync_release(struct g_mirror_softc *sc); 114 115 116static const char * 117g_mirror_disk_state2str(int state) 118{ 119 120 switch (state) { 121 case G_MIRROR_DISK_STATE_NONE: 122 return ("NONE"); 123 case G_MIRROR_DISK_STATE_NEW: 124 return ("NEW"); 125 case G_MIRROR_DISK_STATE_ACTIVE: 126 return ("ACTIVE"); 127 case G_MIRROR_DISK_STATE_STALE: 128 return ("STALE"); 129 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 130 return ("SYNCHRONIZING"); 131 case G_MIRROR_DISK_STATE_DISCONNECTED: 132 return ("DISCONNECTED"); 133 case G_MIRROR_DISK_STATE_DESTROY: 134 return ("DESTROY"); 135 default: 136 return ("INVALID"); 137 } 138} 139 140static const char * 141g_mirror_device_state2str(int state) 142{ 143 144 switch (state) { 145 case G_MIRROR_DEVICE_STATE_STARTING: 146 return ("STARTING"); 147 case G_MIRROR_DEVICE_STATE_RUNNING: 148 return ("RUNNING"); 149 default: 150 return ("INVALID"); 151 } 152} 153 154static const char * 155g_mirror_get_diskname(struct g_mirror_disk *disk) 156{ 157 158 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 159 return ("[unknown]"); 160 return (disk->d_name); 161} 162 163/* 164 * --- Events handling functions --- 165 * Events in geom_mirror are used to maintain disks and device status 166 * from one thread to simplify locking. 167 */ 168static void 169g_mirror_event_free(struct g_mirror_event *ep) 170{ 171 172 free(ep, M_MIRROR); 173} 174 175int 176g_mirror_event_send(void *arg, int state, int flags) 177{ 178 struct g_mirror_softc *sc; 179 struct g_mirror_disk *disk; 180 struct g_mirror_event *ep; 181 int error; 182 183 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 184 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 185 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 186 disk = NULL; 187 sc = arg; 188 } else { 189 disk = arg; 190 sc = disk->d_softc; 191 } 192 ep->e_disk = disk; 193 ep->e_state = state; 194 ep->e_flags = flags; 195 ep->e_error = 0; 196 mtx_lock(&sc->sc_events_mtx); 197 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 198 mtx_unlock(&sc->sc_events_mtx); 199 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 200 mtx_lock(&sc->sc_queue_mtx); 201 wakeup(sc); 202 mtx_unlock(&sc->sc_queue_mtx); 203 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 204 return (0); 205 sx_assert(&sc->sc_lock, SX_XLOCKED); 206 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 207 sx_xunlock(&sc->sc_lock); 208 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 209 mtx_lock(&sc->sc_events_mtx); 210 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 211 hz * 5); 212 } 213 error = ep->e_error; 214 g_mirror_event_free(ep); 215 sx_xlock(&sc->sc_lock); 216 return (error); 217} 218 219static struct g_mirror_event * 220g_mirror_event_first(struct g_mirror_softc *sc) 221{ 222 struct g_mirror_event *ep; 223 224 mtx_lock(&sc->sc_events_mtx); 225 ep = TAILQ_FIRST(&sc->sc_events); 226 mtx_unlock(&sc->sc_events_mtx); 227 return (ep); 228} 229 230static void 231g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 232{ 233 234 mtx_lock(&sc->sc_events_mtx); 235 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 236 mtx_unlock(&sc->sc_events_mtx); 237} 238 239static void 240g_mirror_event_cancel(struct g_mirror_disk *disk) 241{ 242 struct g_mirror_softc *sc; 243 struct g_mirror_event *ep, *tmpep; 244 245 sc = disk->d_softc; 246 sx_assert(&sc->sc_lock, SX_XLOCKED); 247 248 mtx_lock(&sc->sc_events_mtx); 249 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 250 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 251 continue; 252 if (ep->e_disk != disk) 253 continue; 254 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 255 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 256 g_mirror_event_free(ep); 257 else { 258 ep->e_error = ECANCELED; 259 wakeup(ep); 260 } 261 } 262 mtx_unlock(&sc->sc_events_mtx); 263} 264 265/* 266 * Return the number of disks in given state. 267 * If state is equal to -1, count all connected disks. 268 */ 269u_int 270g_mirror_ndisks(struct g_mirror_softc *sc, int state) 271{ 272 struct g_mirror_disk *disk; 273 u_int n = 0; 274 275 sx_assert(&sc->sc_lock, SX_LOCKED); 276 277 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 278 if (state == -1 || disk->d_state == state) 279 n++; 280 } 281 return (n); 282} 283 284/* 285 * Find a disk in mirror by its disk ID. 286 */ 287static struct g_mirror_disk * 288g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 289{ 290 struct g_mirror_disk *disk; 291 292 sx_assert(&sc->sc_lock, SX_XLOCKED); 293 294 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 295 if (disk->d_id == id) 296 return (disk); 297 } 298 return (NULL); 299} 300 301static u_int 302g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 303{ 304 struct bio *bp; 305 u_int nreqs = 0; 306 307 mtx_lock(&sc->sc_queue_mtx); 308 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 309 if (bp->bio_from == cp) 310 nreqs++; 311 } 312 mtx_unlock(&sc->sc_queue_mtx); 313 return (nreqs); 314} 315 316static int 317g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 318{ 319 320 if (cp->index > 0) { 321 G_MIRROR_DEBUG(2, 322 "I/O requests for %s exist, can't destroy it now.", 323 cp->provider->name); 324 return (1); 325 } 326 if (g_mirror_nrequests(sc, cp) > 0) { 327 G_MIRROR_DEBUG(2, 328 "I/O requests for %s in queue, can't destroy it now.", 329 cp->provider->name); 330 return (1); 331 } 332 return (0); 333} 334 335static void 336g_mirror_destroy_consumer(void *arg, int flags __unused) 337{ 338 struct g_consumer *cp; 339 340 g_topology_assert(); 341 342 cp = arg; 343 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 344 g_detach(cp); 345 g_destroy_consumer(cp); 346} 347 348static void 349g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 350{ 351 struct g_provider *pp; 352 int retaste_wait; 353 354 g_topology_assert(); 355 356 cp->private = NULL; 357 if (g_mirror_is_busy(sc, cp)) 358 return; 359 pp = cp->provider; 360 retaste_wait = 0; 361 if (cp->acw == 1) { 362 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 363 retaste_wait = 1; 364 } 365 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 366 -cp->acw, -cp->ace, 0); 367 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 368 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 369 if (retaste_wait) { 370 /* 371 * After retaste event was send (inside g_access()), we can send 372 * event to detach and destroy consumer. 373 * A class, which has consumer to the given provider connected 374 * will not receive retaste event for the provider. 375 * This is the way how I ignore retaste events when I close 376 * consumers opened for write: I detach and destroy consumer 377 * after retaste event is sent. 378 */ 379 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 380 return; 381 } 382 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 383 g_detach(cp); 384 g_destroy_consumer(cp); 385} 386 387static int 388g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 389{ 390 struct g_consumer *cp; 391 int error; 392 393 g_topology_assert_not(); 394 KASSERT(disk->d_consumer == NULL, 395 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 396 397 g_topology_lock(); 398 cp = g_new_consumer(disk->d_softc->sc_geom); 399 cp->flags |= G_CF_DIRECT_RECEIVE; 400 error = g_attach(cp, pp); 401 if (error != 0) { 402 g_destroy_consumer(cp); 403 g_topology_unlock(); 404 return (error); 405 } 406 error = g_access(cp, 1, 1, 1); 407 if (error != 0) { 408 g_detach(cp); 409 g_destroy_consumer(cp); 410 g_topology_unlock(); 411 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 412 pp->name, error); 413 return (error); 414 } 415 g_topology_unlock(); 416 disk->d_consumer = cp; 417 disk->d_consumer->private = disk; 418 disk->d_consumer->index = 0; 419 420 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 421 return (0); 422} 423 424static void 425g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 426{ 427 428 g_topology_assert(); 429 430 if (cp == NULL) 431 return; 432 if (cp->provider != NULL) 433 g_mirror_kill_consumer(sc, cp); 434 else 435 g_destroy_consumer(cp); 436} 437 438/* 439 * Initialize disk. This means allocate memory, create consumer, attach it 440 * to the provider and open access (r1w1e1) to it. 441 */ 442static struct g_mirror_disk * 443g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 444 struct g_mirror_metadata *md, int *errorp) 445{ 446 struct g_mirror_disk *disk; 447 int i, error; 448 449 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 450 if (disk == NULL) { 451 error = ENOMEM; 452 goto fail; 453 } 454 disk->d_softc = sc; 455 error = g_mirror_connect_disk(disk, pp); 456 if (error != 0) 457 goto fail; 458 disk->d_id = md->md_did; 459 disk->d_state = G_MIRROR_DISK_STATE_NONE; 460 disk->d_priority = md->md_priority; 461 disk->d_flags = md->md_dflags; 462 error = g_getattr("GEOM::candelete", disk->d_consumer, &i); 463 if (error == 0 && i != 0) 464 disk->d_flags |= G_MIRROR_DISK_FLAG_CANDELETE; 465 if (md->md_provider[0] != '\0') 466 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 467 disk->d_sync.ds_consumer = NULL; 468 disk->d_sync.ds_offset = md->md_sync_offset; 469 disk->d_sync.ds_offset_done = md->md_sync_offset; 470 disk->d_sync.ds_update_ts = time_uptime; 471 disk->d_genid = md->md_genid; 472 disk->d_sync.ds_syncid = md->md_syncid; 473 if (errorp != NULL) 474 *errorp = 0; 475 return (disk); 476fail: 477 if (errorp != NULL) 478 *errorp = error; 479 if (disk != NULL) 480 free(disk, M_MIRROR); 481 return (NULL); 482} 483 484static void 485g_mirror_destroy_disk(struct g_mirror_disk *disk) 486{ 487 struct g_mirror_softc *sc; 488 489 g_topology_assert_not(); 490 sc = disk->d_softc; 491 sx_assert(&sc->sc_lock, SX_XLOCKED); 492 493 LIST_REMOVE(disk, d_next); 494 g_mirror_event_cancel(disk); 495 if (sc->sc_hint == disk) 496 sc->sc_hint = NULL; 497 switch (disk->d_state) { 498 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 499 g_mirror_sync_stop(disk, 1); 500 /* FALLTHROUGH */ 501 case G_MIRROR_DISK_STATE_NEW: 502 case G_MIRROR_DISK_STATE_STALE: 503 case G_MIRROR_DISK_STATE_ACTIVE: 504 g_topology_lock(); 505 g_mirror_disconnect_consumer(sc, disk->d_consumer); 506 g_topology_unlock(); 507 free(disk, M_MIRROR); 508 break; 509 default: 510 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 511 g_mirror_get_diskname(disk), 512 g_mirror_disk_state2str(disk->d_state))); 513 } 514} 515 516static void 517g_mirror_free_device(struct g_mirror_softc *sc) 518{ 519 520 mtx_destroy(&sc->sc_queue_mtx); 521 mtx_destroy(&sc->sc_events_mtx); 522 mtx_destroy(&sc->sc_done_mtx); 523 sx_destroy(&sc->sc_lock); 524 free(sc, M_MIRROR); 525} 526 527static void 528g_mirror_providergone(struct g_provider *pp) 529{ 530 struct g_mirror_softc *sc = pp->private; 531 532 if ((--sc->sc_refcnt) == 0) 533 g_mirror_free_device(sc); 534} 535 536static void 537g_mirror_destroy_device(struct g_mirror_softc *sc) 538{ 539 struct g_mirror_disk *disk; 540 struct g_mirror_event *ep; 541 struct g_geom *gp; 542 struct g_consumer *cp, *tmpcp; 543 544 g_topology_assert_not(); 545 sx_assert(&sc->sc_lock, SX_XLOCKED); 546 547 gp = sc->sc_geom; 548 if (sc->sc_provider != NULL) 549 g_mirror_destroy_provider(sc); 550 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 551 disk = LIST_FIRST(&sc->sc_disks)) { 552 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 553 g_mirror_update_metadata(disk); 554 g_mirror_destroy_disk(disk); 555 } 556 while ((ep = g_mirror_event_first(sc)) != NULL) { 557 g_mirror_event_remove(sc, ep); 558 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 559 g_mirror_event_free(ep); 560 else { 561 ep->e_error = ECANCELED; 562 ep->e_flags |= G_MIRROR_EVENT_DONE; 563 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 564 mtx_lock(&sc->sc_events_mtx); 565 wakeup(ep); 566 mtx_unlock(&sc->sc_events_mtx); 567 } 568 } 569 callout_drain(&sc->sc_callout); 570 571 g_topology_lock(); 572 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 573 g_mirror_disconnect_consumer(sc, cp); 574 } 575 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 576 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 577 g_wither_geom(gp, ENXIO); 578 sx_xunlock(&sc->sc_lock); 579 if ((--sc->sc_refcnt) == 0) 580 g_mirror_free_device(sc); 581 g_topology_unlock(); 582} 583 584static void 585g_mirror_orphan(struct g_consumer *cp) 586{ 587 struct g_mirror_disk *disk; 588 589 g_topology_assert(); 590 591 disk = cp->private; 592 if (disk == NULL) 593 return; 594 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 595 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 596 G_MIRROR_EVENT_DONTWAIT); 597} 598 599/* 600 * Function should return the next active disk on the list. 601 * It is possible that it will be the same disk as given. 602 * If there are no active disks on list, NULL is returned. 603 */ 604static __inline struct g_mirror_disk * 605g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 606{ 607 struct g_mirror_disk *dp; 608 609 for (dp = LIST_NEXT(disk, d_next); dp != disk; 610 dp = LIST_NEXT(dp, d_next)) { 611 if (dp == NULL) 612 dp = LIST_FIRST(&sc->sc_disks); 613 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 614 break; 615 } 616 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 617 return (NULL); 618 return (dp); 619} 620 621static struct g_mirror_disk * 622g_mirror_get_disk(struct g_mirror_softc *sc) 623{ 624 struct g_mirror_disk *disk; 625 626 if (sc->sc_hint == NULL) { 627 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 628 if (sc->sc_hint == NULL) 629 return (NULL); 630 } 631 disk = sc->sc_hint; 632 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 633 disk = g_mirror_find_next(sc, disk); 634 if (disk == NULL) 635 return (NULL); 636 } 637 sc->sc_hint = g_mirror_find_next(sc, disk); 638 return (disk); 639} 640 641static int 642g_mirror_write_metadata(struct g_mirror_disk *disk, 643 struct g_mirror_metadata *md) 644{ 645 struct g_mirror_softc *sc; 646 struct g_consumer *cp; 647 off_t offset, length; 648 u_char *sector; 649 int error = 0; 650 651 g_topology_assert_not(); 652 sc = disk->d_softc; 653 sx_assert(&sc->sc_lock, SX_LOCKED); 654 655 cp = disk->d_consumer; 656 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 657 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 658 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 659 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 660 cp->acw, cp->ace)); 661 length = cp->provider->sectorsize; 662 offset = cp->provider->mediasize - length; 663 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 664 if (md != NULL && 665 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) { 666 /* 667 * Handle the case, when the size of parent provider reduced. 668 */ 669 if (offset < md->md_mediasize) 670 error = ENOSPC; 671 else 672 mirror_metadata_encode(md, sector); 673 } 674 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_metadata_write, error); 675 if (error == 0) 676 error = g_write_data(cp, offset, sector, length); 677 free(sector, M_MIRROR); 678 if (error != 0) { 679 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 680 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 681 G_MIRROR_DEBUG(0, "Cannot write metadata on %s " 682 "(device=%s, error=%d).", 683 g_mirror_get_diskname(disk), sc->sc_name, error); 684 } else { 685 G_MIRROR_DEBUG(1, "Cannot write metadata on %s " 686 "(device=%s, error=%d).", 687 g_mirror_get_diskname(disk), sc->sc_name, error); 688 } 689 if (g_mirror_disconnect_on_failure && 690 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 691 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 692 g_mirror_event_send(disk, 693 G_MIRROR_DISK_STATE_DISCONNECTED, 694 G_MIRROR_EVENT_DONTWAIT); 695 } 696 } 697 return (error); 698} 699 700static int 701g_mirror_clear_metadata(struct g_mirror_disk *disk) 702{ 703 int error; 704 705 g_topology_assert_not(); 706 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 707 708 if (disk->d_softc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 709 return (0); 710 error = g_mirror_write_metadata(disk, NULL); 711 if (error == 0) { 712 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 713 g_mirror_get_diskname(disk)); 714 } else { 715 G_MIRROR_DEBUG(0, 716 "Cannot clear metadata on disk %s (error=%d).", 717 g_mirror_get_diskname(disk), error); 718 } 719 return (error); 720} 721 722void 723g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 724 struct g_mirror_metadata *md) 725{ 726 727 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 728 md->md_version = G_MIRROR_VERSION; 729 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 730 md->md_mid = sc->sc_id; 731 md->md_all = sc->sc_ndisks; 732 md->md_slice = sc->sc_slice; 733 md->md_balance = sc->sc_balance; 734 md->md_genid = sc->sc_genid; 735 md->md_mediasize = sc->sc_mediasize; 736 md->md_sectorsize = sc->sc_sectorsize; 737 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 738 bzero(md->md_provider, sizeof(md->md_provider)); 739 if (disk == NULL) { 740 md->md_did = arc4random(); 741 md->md_priority = 0; 742 md->md_syncid = 0; 743 md->md_dflags = 0; 744 md->md_sync_offset = 0; 745 md->md_provsize = 0; 746 } else { 747 md->md_did = disk->d_id; 748 md->md_priority = disk->d_priority; 749 md->md_syncid = disk->d_sync.ds_syncid; 750 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 751 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 752 md->md_sync_offset = disk->d_sync.ds_offset_done; 753 else 754 md->md_sync_offset = 0; 755 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 756 strlcpy(md->md_provider, 757 disk->d_consumer->provider->name, 758 sizeof(md->md_provider)); 759 } 760 md->md_provsize = disk->d_consumer->provider->mediasize; 761 } 762} 763 764void 765g_mirror_update_metadata(struct g_mirror_disk *disk) 766{ 767 struct g_mirror_softc *sc; 768 struct g_mirror_metadata md; 769 int error; 770 771 g_topology_assert_not(); 772 sc = disk->d_softc; 773 sx_assert(&sc->sc_lock, SX_LOCKED); 774 775 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 776 return; 777 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) 778 g_mirror_fill_metadata(sc, disk, &md); 779 error = g_mirror_write_metadata(disk, &md); 780 if (error == 0) { 781 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 782 g_mirror_get_diskname(disk)); 783 } else { 784 G_MIRROR_DEBUG(0, 785 "Cannot update metadata on disk %s (error=%d).", 786 g_mirror_get_diskname(disk), error); 787 } 788} 789 790static void 791g_mirror_bump_syncid(struct g_mirror_softc *sc) 792{ 793 struct g_mirror_disk *disk; 794 795 g_topology_assert_not(); 796 sx_assert(&sc->sc_lock, SX_XLOCKED); 797 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 798 ("%s called with no active disks (device=%s).", __func__, 799 sc->sc_name)); 800 801 sc->sc_syncid++; 802 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 803 sc->sc_syncid); 804 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 805 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 806 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 807 disk->d_sync.ds_syncid = sc->sc_syncid; 808 g_mirror_update_metadata(disk); 809 } 810 } 811} 812 813static void 814g_mirror_bump_genid(struct g_mirror_softc *sc) 815{ 816 struct g_mirror_disk *disk; 817 818 g_topology_assert_not(); 819 sx_assert(&sc->sc_lock, SX_XLOCKED); 820 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 821 ("%s called with no active disks (device=%s).", __func__, 822 sc->sc_name)); 823 824 sc->sc_genid++; 825 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 826 sc->sc_genid); 827 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 828 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 829 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 830 disk->d_genid = sc->sc_genid; 831 g_mirror_update_metadata(disk); 832 } 833 } 834} 835 836static int 837g_mirror_idle(struct g_mirror_softc *sc, int acw) 838{ 839 struct g_mirror_disk *disk; 840 int timeout; 841 842 g_topology_assert_not(); 843 sx_assert(&sc->sc_lock, SX_XLOCKED); 844 845 if (sc->sc_provider == NULL) 846 return (0); 847 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 848 return (0); 849 if (sc->sc_idle) 850 return (0); 851 if (sc->sc_writes > 0) 852 return (0); 853 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 854 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); 855 if (!g_mirror_shutdown && timeout > 0) 856 return (timeout); 857 } 858 sc->sc_idle = 1; 859 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 860 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 861 continue; 862 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 863 g_mirror_get_diskname(disk), sc->sc_name); 864 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 865 g_mirror_update_metadata(disk); 866 } 867 return (0); 868} 869 870static void 871g_mirror_unidle(struct g_mirror_softc *sc) 872{ 873 struct g_mirror_disk *disk; 874 875 g_topology_assert_not(); 876 sx_assert(&sc->sc_lock, SX_XLOCKED); 877 878 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 879 return; 880 sc->sc_idle = 0; 881 sc->sc_last_write = time_uptime; 882 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 883 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 884 continue; 885 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 886 g_mirror_get_diskname(disk), sc->sc_name); 887 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 888 g_mirror_update_metadata(disk); 889 } 890} 891 892static void 893g_mirror_flush_done(struct bio *bp) 894{ 895 struct g_mirror_softc *sc; 896 struct bio *pbp; 897 898 pbp = bp->bio_parent; 899 sc = pbp->bio_to->private; 900 mtx_lock(&sc->sc_done_mtx); 901 if (pbp->bio_error == 0) 902 pbp->bio_error = bp->bio_error; 903 pbp->bio_completed += bp->bio_completed; 904 pbp->bio_inbed++; 905 if (pbp->bio_children == pbp->bio_inbed) { 906 mtx_unlock(&sc->sc_done_mtx); 907 g_io_deliver(pbp, pbp->bio_error); 908 } else 909 mtx_unlock(&sc->sc_done_mtx); 910 g_destroy_bio(bp); 911} 912 913static void 914g_mirror_done(struct bio *bp) 915{ 916 struct g_mirror_softc *sc; 917 918 sc = bp->bio_from->geom->softc; 919 bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR; 920 mtx_lock(&sc->sc_queue_mtx); 921 bioq_insert_tail(&sc->sc_queue, bp); 922 mtx_unlock(&sc->sc_queue_mtx); 923 wakeup(sc); 924} 925 926static void 927g_mirror_regular_request(struct bio *bp) 928{ 929 struct g_mirror_softc *sc; 930 struct g_mirror_disk *disk; 931 struct bio *pbp; 932 933 g_topology_assert_not(); 934 935 pbp = bp->bio_parent; 936 sc = pbp->bio_to->private; 937 bp->bio_from->index--; 938 if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) 939 sc->sc_writes--; 940 disk = bp->bio_from->private; 941 if (disk == NULL) { 942 g_topology_lock(); 943 g_mirror_kill_consumer(sc, bp->bio_from); 944 g_topology_unlock(); 945 } 946 947 if (bp->bio_cmd == BIO_READ) 948 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_read, 949 bp->bio_error); 950 else if (bp->bio_cmd == BIO_WRITE) 951 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_write, 952 bp->bio_error); 953 954 pbp->bio_inbed++; 955 KASSERT(pbp->bio_inbed <= pbp->bio_children, 956 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 957 pbp->bio_children)); 958 if (bp->bio_error == 0 && pbp->bio_error == 0) { 959 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 960 g_destroy_bio(bp); 961 if (pbp->bio_children == pbp->bio_inbed) { 962 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 963 pbp->bio_completed = pbp->bio_length; 964 if (pbp->bio_cmd == BIO_WRITE || 965 pbp->bio_cmd == BIO_DELETE) { 966 bioq_remove(&sc->sc_inflight, pbp); 967 /* Release delayed sync requests if possible. */ 968 g_mirror_sync_release(sc); 969 } 970 g_io_deliver(pbp, pbp->bio_error); 971 } 972 return; 973 } else if (bp->bio_error != 0) { 974 if (pbp->bio_error == 0) 975 pbp->bio_error = bp->bio_error; 976 if (disk != NULL) { 977 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 978 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 979 G_MIRROR_LOGREQ(0, bp, 980 "Request failed (error=%d).", 981 bp->bio_error); 982 } else { 983 G_MIRROR_LOGREQ(1, bp, 984 "Request failed (error=%d).", 985 bp->bio_error); 986 } 987 if (g_mirror_disconnect_on_failure && 988 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) 989 { 990 if (bp->bio_error == ENXIO && 991 bp->bio_cmd == BIO_READ) 992 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 993 else if (bp->bio_error == ENXIO) 994 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_NOW; 995 else 996 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 997 g_mirror_event_send(disk, 998 G_MIRROR_DISK_STATE_DISCONNECTED, 999 G_MIRROR_EVENT_DONTWAIT); 1000 } 1001 } 1002 switch (pbp->bio_cmd) { 1003 case BIO_DELETE: 1004 case BIO_WRITE: 1005 pbp->bio_inbed--; 1006 pbp->bio_children--; 1007 break; 1008 } 1009 } 1010 g_destroy_bio(bp); 1011 1012 switch (pbp->bio_cmd) { 1013 case BIO_READ: 1014 if (pbp->bio_inbed < pbp->bio_children) 1015 break; 1016 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 1017 g_io_deliver(pbp, pbp->bio_error); 1018 else { 1019 pbp->bio_error = 0; 1020 mtx_lock(&sc->sc_queue_mtx); 1021 bioq_insert_tail(&sc->sc_queue, pbp); 1022 mtx_unlock(&sc->sc_queue_mtx); 1023 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1024 wakeup(sc); 1025 } 1026 break; 1027 case BIO_DELETE: 1028 case BIO_WRITE: 1029 if (pbp->bio_children == 0) { 1030 /* 1031 * All requests failed. 1032 */ 1033 } else if (pbp->bio_inbed < pbp->bio_children) { 1034 /* Do nothing. */ 1035 break; 1036 } else if (pbp->bio_children == pbp->bio_inbed) { 1037 /* Some requests succeeded. */ 1038 pbp->bio_error = 0; 1039 pbp->bio_completed = pbp->bio_length; 1040 } 1041 bioq_remove(&sc->sc_inflight, pbp); 1042 /* Release delayed sync requests if possible. */ 1043 g_mirror_sync_release(sc); 1044 g_io_deliver(pbp, pbp->bio_error); 1045 break; 1046 default: 1047 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 1048 break; 1049 } 1050} 1051 1052static void 1053g_mirror_sync_done(struct bio *bp) 1054{ 1055 struct g_mirror_softc *sc; 1056 1057 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1058 sc = bp->bio_from->geom->softc; 1059 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; 1060 mtx_lock(&sc->sc_queue_mtx); 1061 bioq_insert_tail(&sc->sc_queue, bp); 1062 mtx_unlock(&sc->sc_queue_mtx); 1063 wakeup(sc); 1064} 1065 1066static void 1067g_mirror_candelete(struct bio *bp) 1068{ 1069 struct g_mirror_softc *sc; 1070 struct g_mirror_disk *disk; 1071 int *val; 1072 1073 sc = bp->bio_to->private; 1074 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1075 if (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) 1076 break; 1077 } 1078 val = (int *)bp->bio_data; 1079 *val = (disk != NULL); 1080 g_io_deliver(bp, 0); 1081} 1082 1083static void 1084g_mirror_kernel_dump(struct bio *bp) 1085{ 1086 struct g_mirror_softc *sc; 1087 struct g_mirror_disk *disk; 1088 struct bio *cbp; 1089 struct g_kerneldump *gkd; 1090 1091 /* 1092 * We configure dumping to the first component, because this component 1093 * will be used for reading with 'prefer' balance algorithm. 1094 * If the component with the highest priority is currently disconnected 1095 * we will not be able to read the dump after the reboot if it will be 1096 * connected and synchronized later. Can we do something better? 1097 */ 1098 sc = bp->bio_to->private; 1099 disk = LIST_FIRST(&sc->sc_disks); 1100 1101 gkd = (struct g_kerneldump *)bp->bio_data; 1102 if (gkd->length > bp->bio_to->mediasize) 1103 gkd->length = bp->bio_to->mediasize; 1104 cbp = g_clone_bio(bp); 1105 if (cbp == NULL) { 1106 g_io_deliver(bp, ENOMEM); 1107 return; 1108 } 1109 cbp->bio_done = g_std_done; 1110 g_io_request(cbp, disk->d_consumer); 1111 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1112 g_mirror_get_diskname(disk)); 1113} 1114 1115static void 1116g_mirror_flush(struct g_mirror_softc *sc, struct bio *bp) 1117{ 1118 struct bio_queue_head queue; 1119 struct g_mirror_disk *disk; 1120 struct g_consumer *cp; 1121 struct bio *cbp; 1122 1123 bioq_init(&queue); 1124 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1125 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1126 continue; 1127 cbp = g_clone_bio(bp); 1128 if (cbp == NULL) { 1129 while ((cbp = bioq_takefirst(&queue)) != NULL) 1130 g_destroy_bio(cbp); 1131 if (bp->bio_error == 0) 1132 bp->bio_error = ENOMEM; 1133 g_io_deliver(bp, bp->bio_error); 1134 return; 1135 } 1136 bioq_insert_tail(&queue, cbp); 1137 cbp->bio_done = g_mirror_flush_done; 1138 cbp->bio_caller1 = disk; 1139 cbp->bio_to = disk->d_consumer->provider; 1140 } 1141 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1142 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1143 disk = cbp->bio_caller1; 1144 cbp->bio_caller1 = NULL; 1145 cp = disk->d_consumer; 1146 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1147 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1148 cp->acr, cp->acw, cp->ace)); 1149 g_io_request(cbp, disk->d_consumer); 1150 } 1151} 1152 1153static void 1154g_mirror_start(struct bio *bp) 1155{ 1156 struct g_mirror_softc *sc; 1157 1158 sc = bp->bio_to->private; 1159 /* 1160 * If sc == NULL or there are no valid disks, provider's error 1161 * should be set and g_mirror_start() should not be called at all. 1162 */ 1163 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1164 ("Provider's error should be set (error=%d)(mirror=%s).", 1165 bp->bio_to->error, bp->bio_to->name)); 1166 G_MIRROR_LOGREQ(3, bp, "Request received."); 1167 1168 switch (bp->bio_cmd) { 1169 case BIO_READ: 1170 case BIO_WRITE: 1171 case BIO_DELETE: 1172 break; 1173 case BIO_FLUSH: 1174 g_mirror_flush(sc, bp); 1175 return; 1176 case BIO_GETATTR: 1177 if (!strcmp(bp->bio_attribute, "GEOM::candelete")) { 1178 g_mirror_candelete(bp); 1179 return; 1180 } else if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1181 g_mirror_kernel_dump(bp); 1182 return; 1183 } 1184 /* FALLTHROUGH */ 1185 default: 1186 g_io_deliver(bp, EOPNOTSUPP); 1187 return; 1188 } 1189 mtx_lock(&sc->sc_queue_mtx); 1190 if (bp->bio_to->error != 0) { 1191 mtx_unlock(&sc->sc_queue_mtx); 1192 g_io_deliver(bp, bp->bio_to->error); 1193 return; 1194 } 1195 bioq_insert_tail(&sc->sc_queue, bp); 1196 mtx_unlock(&sc->sc_queue_mtx); 1197 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1198 wakeup(sc); 1199} 1200 1201/* 1202 * Return TRUE if the given request is colliding with a in-progress 1203 * synchronization request. 1204 */ 1205static bool 1206g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1207{ 1208 struct g_mirror_disk *disk; 1209 struct bio *sbp; 1210 off_t rstart, rend, sstart, send; 1211 u_int i; 1212 1213 if (sc->sc_sync.ds_ndisks == 0) 1214 return (false); 1215 rstart = bp->bio_offset; 1216 rend = bp->bio_offset + bp->bio_length; 1217 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1218 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1219 continue; 1220 for (i = 0; i < g_mirror_syncreqs; i++) { 1221 sbp = disk->d_sync.ds_bios[i]; 1222 if (sbp == NULL) 1223 continue; 1224 sstart = sbp->bio_offset; 1225 send = sbp->bio_offset + sbp->bio_length; 1226 if (rend > sstart && rstart < send) 1227 return (true); 1228 } 1229 } 1230 return (false); 1231} 1232 1233/* 1234 * Return TRUE if the given sync request is colliding with a in-progress regular 1235 * request. 1236 */ 1237static bool 1238g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1239{ 1240 off_t rstart, rend, sstart, send; 1241 struct bio *bp; 1242 1243 if (sc->sc_sync.ds_ndisks == 0) 1244 return (false); 1245 sstart = sbp->bio_offset; 1246 send = sbp->bio_offset + sbp->bio_length; 1247 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { 1248 rstart = bp->bio_offset; 1249 rend = bp->bio_offset + bp->bio_length; 1250 if (rend > sstart && rstart < send) 1251 return (true); 1252 } 1253 return (false); 1254} 1255 1256/* 1257 * Puts request onto delayed queue. 1258 */ 1259static void 1260g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1261{ 1262 1263 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1264 bioq_insert_head(&sc->sc_regular_delayed, bp); 1265} 1266 1267/* 1268 * Puts synchronization request onto delayed queue. 1269 */ 1270static void 1271g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1272{ 1273 1274 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1275 bioq_insert_tail(&sc->sc_sync_delayed, bp); 1276} 1277 1278/* 1279 * Releases delayed regular requests which don't collide anymore with sync 1280 * requests. 1281 */ 1282static void 1283g_mirror_regular_release(struct g_mirror_softc *sc) 1284{ 1285 struct bio *bp, *bp2; 1286 1287 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { 1288 if (g_mirror_sync_collision(sc, bp)) 1289 continue; 1290 bioq_remove(&sc->sc_regular_delayed, bp); 1291 G_MIRROR_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1292 mtx_lock(&sc->sc_queue_mtx); 1293 bioq_insert_head(&sc->sc_queue, bp); 1294 mtx_unlock(&sc->sc_queue_mtx); 1295 } 1296} 1297 1298/* 1299 * Releases delayed sync requests which don't collide anymore with regular 1300 * requests. 1301 */ 1302static void 1303g_mirror_sync_release(struct g_mirror_softc *sc) 1304{ 1305 struct bio *bp, *bp2; 1306 1307 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { 1308 if (g_mirror_regular_collision(sc, bp)) 1309 continue; 1310 bioq_remove(&sc->sc_sync_delayed, bp); 1311 G_MIRROR_LOGREQ(2, bp, 1312 "Releasing delayed synchronization request."); 1313 g_io_request(bp, bp->bio_from); 1314 } 1315} 1316 1317/* 1318 * Free a synchronization request and clear its slot in the array. 1319 */ 1320static void 1321g_mirror_sync_request_free(struct g_mirror_disk *disk, struct bio *bp) 1322{ 1323 int idx; 1324 1325 if (disk != NULL && disk->d_sync.ds_bios != NULL) { 1326 idx = (int)(uintptr_t)bp->bio_caller1; 1327 KASSERT(disk->d_sync.ds_bios[idx] == bp, 1328 ("unexpected sync BIO at %p:%d", disk, idx)); 1329 disk->d_sync.ds_bios[idx] = NULL; 1330 } 1331 free(bp->bio_data, M_MIRROR); 1332 g_destroy_bio(bp); 1333} 1334 1335/* 1336 * Handle synchronization requests. 1337 * Every synchronization request is two-steps process: first, READ request is 1338 * send to active provider and then WRITE request (with read data) to the provider 1339 * being synchronized. When WRITE is finished, new synchronization request is 1340 * send. 1341 */ 1342static void 1343g_mirror_sync_request(struct bio *bp) 1344{ 1345 struct g_mirror_softc *sc; 1346 struct g_mirror_disk *disk; 1347 struct g_mirror_disk_sync *sync; 1348 1349 bp->bio_from->index--; 1350 sc = bp->bio_from->geom->softc; 1351 disk = bp->bio_from->private; 1352 if (disk == NULL) { 1353 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1354 g_topology_lock(); 1355 g_mirror_kill_consumer(sc, bp->bio_from); 1356 g_topology_unlock(); 1357 g_mirror_sync_request_free(NULL, bp); 1358 sx_xlock(&sc->sc_lock); 1359 return; 1360 } 1361 1362 /* 1363 * Synchronization request. 1364 */ 1365 switch (bp->bio_cmd) { 1366 case BIO_READ: 1367 { 1368 struct g_consumer *cp; 1369 1370 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_read, 1371 bp->bio_error); 1372 1373 if (bp->bio_error != 0) { 1374 G_MIRROR_LOGREQ(0, bp, 1375 "Synchronization request failed (error=%d).", 1376 bp->bio_error); 1377 g_mirror_sync_request_free(disk, bp); 1378 return; 1379 } 1380 G_MIRROR_LOGREQ(3, bp, 1381 "Synchronization request half-finished."); 1382 bp->bio_cmd = BIO_WRITE; 1383 bp->bio_cflags = 0; 1384 cp = disk->d_consumer; 1385 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1386 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1387 cp->acr, cp->acw, cp->ace)); 1388 cp->index++; 1389 g_io_request(bp, cp); 1390 return; 1391 } 1392 case BIO_WRITE: 1393 { 1394 off_t offset; 1395 void *data; 1396 int i, idx; 1397 1398 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_write, 1399 bp->bio_error); 1400 1401 if (bp->bio_error != 0) { 1402 G_MIRROR_LOGREQ(0, bp, 1403 "Synchronization request failed (error=%d).", 1404 bp->bio_error); 1405 g_mirror_sync_request_free(disk, bp); 1406 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1407 g_mirror_event_send(disk, 1408 G_MIRROR_DISK_STATE_DISCONNECTED, 1409 G_MIRROR_EVENT_DONTWAIT); 1410 return; 1411 } 1412 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1413 sync = &disk->d_sync; 1414 if (sync->ds_offset >= sc->sc_mediasize || 1415 sync->ds_consumer == NULL || 1416 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1417 /* Don't send more synchronization requests. */ 1418 sync->ds_inflight--; 1419 g_mirror_sync_request_free(disk, bp); 1420 if (sync->ds_inflight > 0) 1421 return; 1422 if (sync->ds_consumer == NULL || 1423 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1424 return; 1425 } 1426 /* Disk up-to-date, activate it. */ 1427 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1428 G_MIRROR_EVENT_DONTWAIT); 1429 return; 1430 } 1431 1432 /* Send next synchronization request. */ 1433 data = bp->bio_data; 1434 idx = (int)(uintptr_t)bp->bio_caller1; 1435 g_reset_bio(bp); 1436 bp->bio_cmd = BIO_READ; 1437 bp->bio_offset = sync->ds_offset; 1438 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1439 sync->ds_offset += bp->bio_length; 1440 bp->bio_done = g_mirror_sync_done; 1441 bp->bio_data = data; 1442 bp->bio_from = sync->ds_consumer; 1443 bp->bio_to = sc->sc_provider; 1444 bp->bio_caller1 = (void *)(uintptr_t)idx; 1445 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1446 sync->ds_consumer->index++; 1447 /* 1448 * Delay the request if it is colliding with a regular request. 1449 */ 1450 if (g_mirror_regular_collision(sc, bp)) 1451 g_mirror_sync_delay(sc, bp); 1452 else 1453 g_io_request(bp, sync->ds_consumer); 1454 1455 /* Release delayed requests if possible. */ 1456 g_mirror_regular_release(sc); 1457 1458 /* Find the smallest offset */ 1459 offset = sc->sc_mediasize; 1460 for (i = 0; i < g_mirror_syncreqs; i++) { 1461 bp = sync->ds_bios[i]; 1462 if (bp != NULL && bp->bio_offset < offset) 1463 offset = bp->bio_offset; 1464 } 1465 if (g_mirror_sync_period > 0 && 1466 time_uptime - sync->ds_update_ts > g_mirror_sync_period) { 1467 sync->ds_offset_done = offset; 1468 g_mirror_update_metadata(disk); 1469 sync->ds_update_ts = time_uptime; 1470 } 1471 return; 1472 } 1473 default: 1474 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1475 bp->bio_cmd, sc->sc_name)); 1476 break; 1477 } 1478} 1479 1480static void 1481g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1482{ 1483 struct g_mirror_disk *disk; 1484 struct g_consumer *cp; 1485 struct bio *cbp; 1486 1487 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1488 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1489 break; 1490 } 1491 if (disk == NULL) { 1492 if (bp->bio_error == 0) 1493 bp->bio_error = ENXIO; 1494 g_io_deliver(bp, bp->bio_error); 1495 return; 1496 } 1497 cbp = g_clone_bio(bp); 1498 if (cbp == NULL) { 1499 if (bp->bio_error == 0) 1500 bp->bio_error = ENOMEM; 1501 g_io_deliver(bp, bp->bio_error); 1502 return; 1503 } 1504 /* 1505 * Fill in the component buf structure. 1506 */ 1507 cp = disk->d_consumer; 1508 cbp->bio_done = g_mirror_done; 1509 cbp->bio_to = cp->provider; 1510 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1511 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1512 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1513 cp->acw, cp->ace)); 1514 cp->index++; 1515 g_io_request(cbp, cp); 1516} 1517 1518static void 1519g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1520{ 1521 struct g_mirror_disk *disk; 1522 struct g_consumer *cp; 1523 struct bio *cbp; 1524 1525 disk = g_mirror_get_disk(sc); 1526 if (disk == NULL) { 1527 if (bp->bio_error == 0) 1528 bp->bio_error = ENXIO; 1529 g_io_deliver(bp, bp->bio_error); 1530 return; 1531 } 1532 cbp = g_clone_bio(bp); 1533 if (cbp == NULL) { 1534 if (bp->bio_error == 0) 1535 bp->bio_error = ENOMEM; 1536 g_io_deliver(bp, bp->bio_error); 1537 return; 1538 } 1539 /* 1540 * Fill in the component buf structure. 1541 */ 1542 cp = disk->d_consumer; 1543 cbp->bio_done = g_mirror_done; 1544 cbp->bio_to = cp->provider; 1545 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1546 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1547 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1548 cp->acw, cp->ace)); 1549 cp->index++; 1550 g_io_request(cbp, cp); 1551} 1552 1553#define TRACK_SIZE (1 * 1024 * 1024) 1554#define LOAD_SCALE 256 1555#define ABS(x) (((x) >= 0) ? (x) : (-(x))) 1556 1557static void 1558g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1559{ 1560 struct g_mirror_disk *disk, *dp; 1561 struct g_consumer *cp; 1562 struct bio *cbp; 1563 int prio, best; 1564 1565 /* Find a disk with the smallest load. */ 1566 disk = NULL; 1567 best = INT_MAX; 1568 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1569 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1570 continue; 1571 prio = dp->load; 1572 /* If disk head is precisely in position - highly prefer it. */ 1573 if (dp->d_last_offset == bp->bio_offset) 1574 prio -= 2 * LOAD_SCALE; 1575 else 1576 /* If disk head is close to position - prefer it. */ 1577 if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE) 1578 prio -= 1 * LOAD_SCALE; 1579 if (prio <= best) { 1580 disk = dp; 1581 best = prio; 1582 } 1583 } 1584 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1585 cbp = g_clone_bio(bp); 1586 if (cbp == NULL) { 1587 if (bp->bio_error == 0) 1588 bp->bio_error = ENOMEM; 1589 g_io_deliver(bp, bp->bio_error); 1590 return; 1591 } 1592 /* 1593 * Fill in the component buf structure. 1594 */ 1595 cp = disk->d_consumer; 1596 cbp->bio_done = g_mirror_done; 1597 cbp->bio_to = cp->provider; 1598 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1599 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1600 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1601 cp->acw, cp->ace)); 1602 cp->index++; 1603 /* Remember last head position */ 1604 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1605 /* Update loads. */ 1606 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1607 dp->load = (dp->d_consumer->index * LOAD_SCALE + 1608 dp->load * 7) / 8; 1609 } 1610 g_io_request(cbp, cp); 1611} 1612 1613static void 1614g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1615{ 1616 struct bio_queue_head queue; 1617 struct g_mirror_disk *disk; 1618 struct g_consumer *cp; 1619 struct bio *cbp; 1620 off_t left, mod, offset, slice; 1621 u_char *data; 1622 u_int ndisks; 1623 1624 if (bp->bio_length <= sc->sc_slice) { 1625 g_mirror_request_round_robin(sc, bp); 1626 return; 1627 } 1628 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1629 slice = bp->bio_length / ndisks; 1630 mod = slice % sc->sc_provider->sectorsize; 1631 if (mod != 0) 1632 slice += sc->sc_provider->sectorsize - mod; 1633 /* 1634 * Allocate all bios before sending any request, so we can 1635 * return ENOMEM in nice and clean way. 1636 */ 1637 left = bp->bio_length; 1638 offset = bp->bio_offset; 1639 data = bp->bio_data; 1640 bioq_init(&queue); 1641 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1642 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1643 continue; 1644 cbp = g_clone_bio(bp); 1645 if (cbp == NULL) { 1646 while ((cbp = bioq_takefirst(&queue)) != NULL) 1647 g_destroy_bio(cbp); 1648 if (bp->bio_error == 0) 1649 bp->bio_error = ENOMEM; 1650 g_io_deliver(bp, bp->bio_error); 1651 return; 1652 } 1653 bioq_insert_tail(&queue, cbp); 1654 cbp->bio_done = g_mirror_done; 1655 cbp->bio_caller1 = disk; 1656 cbp->bio_to = disk->d_consumer->provider; 1657 cbp->bio_offset = offset; 1658 cbp->bio_data = data; 1659 cbp->bio_length = MIN(left, slice); 1660 left -= cbp->bio_length; 1661 if (left == 0) 1662 break; 1663 offset += cbp->bio_length; 1664 data += cbp->bio_length; 1665 } 1666 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1667 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1668 disk = cbp->bio_caller1; 1669 cbp->bio_caller1 = NULL; 1670 cp = disk->d_consumer; 1671 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1672 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1673 cp->acr, cp->acw, cp->ace)); 1674 disk->d_consumer->index++; 1675 g_io_request(cbp, disk->d_consumer); 1676 } 1677} 1678 1679static void 1680g_mirror_register_request(struct bio *bp) 1681{ 1682 struct g_mirror_softc *sc; 1683 1684 sc = bp->bio_to->private; 1685 switch (bp->bio_cmd) { 1686 case BIO_READ: 1687 switch (sc->sc_balance) { 1688 case G_MIRROR_BALANCE_LOAD: 1689 g_mirror_request_load(sc, bp); 1690 break; 1691 case G_MIRROR_BALANCE_PREFER: 1692 g_mirror_request_prefer(sc, bp); 1693 break; 1694 case G_MIRROR_BALANCE_ROUND_ROBIN: 1695 g_mirror_request_round_robin(sc, bp); 1696 break; 1697 case G_MIRROR_BALANCE_SPLIT: 1698 g_mirror_request_split(sc, bp); 1699 break; 1700 } 1701 return; 1702 case BIO_WRITE: 1703 case BIO_DELETE: 1704 { 1705 struct g_mirror_disk *disk; 1706 struct g_mirror_disk_sync *sync; 1707 struct bio_queue_head queue; 1708 struct g_consumer *cp; 1709 struct bio *cbp; 1710 1711 /* 1712 * Delay the request if it is colliding with a synchronization 1713 * request. 1714 */ 1715 if (g_mirror_sync_collision(sc, bp)) { 1716 g_mirror_regular_delay(sc, bp); 1717 return; 1718 } 1719 1720 if (sc->sc_idle) 1721 g_mirror_unidle(sc); 1722 else 1723 sc->sc_last_write = time_uptime; 1724 1725 /* 1726 * Bump syncid on first write. 1727 */ 1728 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1729 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1730 g_mirror_bump_syncid(sc); 1731 } 1732 1733 /* 1734 * Allocate all bios before sending any request, so we can 1735 * return ENOMEM in nice and clean way. 1736 */ 1737 bioq_init(&queue); 1738 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1739 sync = &disk->d_sync; 1740 switch (disk->d_state) { 1741 case G_MIRROR_DISK_STATE_ACTIVE: 1742 break; 1743 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1744 if (bp->bio_offset >= sync->ds_offset) 1745 continue; 1746 break; 1747 default: 1748 continue; 1749 } 1750 if (bp->bio_cmd == BIO_DELETE && 1751 (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0) 1752 continue; 1753 cbp = g_clone_bio(bp); 1754 if (cbp == NULL) { 1755 while ((cbp = bioq_takefirst(&queue)) != NULL) 1756 g_destroy_bio(cbp); 1757 if (bp->bio_error == 0) 1758 bp->bio_error = ENOMEM; 1759 g_io_deliver(bp, bp->bio_error); 1760 return; 1761 } 1762 bioq_insert_tail(&queue, cbp); 1763 cbp->bio_done = g_mirror_done; 1764 cp = disk->d_consumer; 1765 cbp->bio_caller1 = cp; 1766 cbp->bio_to = cp->provider; 1767 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1768 ("Consumer %s not opened (r%dw%de%d).", 1769 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1770 } 1771 if (bioq_first(&queue) == NULL) { 1772 g_io_deliver(bp, EOPNOTSUPP); 1773 return; 1774 } 1775 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1776 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1777 cp = cbp->bio_caller1; 1778 cbp->bio_caller1 = NULL; 1779 cp->index++; 1780 sc->sc_writes++; 1781 g_io_request(cbp, cp); 1782 } 1783 /* 1784 * Put request onto inflight queue, so we can check if new 1785 * synchronization requests don't collide with it. 1786 */ 1787 bioq_insert_tail(&sc->sc_inflight, bp); 1788 return; 1789 } 1790 default: 1791 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1792 bp->bio_cmd, sc->sc_name)); 1793 break; 1794 } 1795} 1796 1797static int 1798g_mirror_can_destroy(struct g_mirror_softc *sc) 1799{ 1800 struct g_geom *gp; 1801 struct g_consumer *cp; 1802 1803 g_topology_assert(); 1804 gp = sc->sc_geom; 1805 if (gp->softc == NULL) 1806 return (1); 1807 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0) 1808 return (0); 1809 LIST_FOREACH(cp, &gp->consumer, consumer) { 1810 if (g_mirror_is_busy(sc, cp)) 1811 return (0); 1812 } 1813 gp = sc->sc_sync.ds_geom; 1814 LIST_FOREACH(cp, &gp->consumer, consumer) { 1815 if (g_mirror_is_busy(sc, cp)) 1816 return (0); 1817 } 1818 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1819 sc->sc_name); 1820 return (1); 1821} 1822 1823static int 1824g_mirror_try_destroy(struct g_mirror_softc *sc) 1825{ 1826 1827 if (sc->sc_rootmount != NULL) { 1828 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1829 sc->sc_rootmount); 1830 root_mount_rel(sc->sc_rootmount); 1831 sc->sc_rootmount = NULL; 1832 } 1833 g_topology_lock(); 1834 if (!g_mirror_can_destroy(sc)) { 1835 g_topology_unlock(); 1836 return (0); 1837 } 1838 sc->sc_geom->softc = NULL; 1839 sc->sc_sync.ds_geom->softc = NULL; 1840 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DRAIN) != 0) { 1841 g_topology_unlock(); 1842 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1843 &sc->sc_worker); 1844 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1845 sx_xunlock(&sc->sc_lock); 1846 wakeup(&sc->sc_worker); 1847 sc->sc_worker = NULL; 1848 } else { 1849 g_topology_unlock(); 1850 g_mirror_destroy_device(sc); 1851 } 1852 return (1); 1853} 1854 1855/* 1856 * Worker thread. 1857 */ 1858static void 1859g_mirror_worker(void *arg) 1860{ 1861 struct g_mirror_softc *sc; 1862 struct g_mirror_event *ep; 1863 struct bio *bp; 1864 int timeout; 1865 1866 sc = arg; 1867 thread_lock(curthread); 1868 sched_prio(curthread, PRIBIO); 1869 thread_unlock(curthread); 1870 1871 sx_xlock(&sc->sc_lock); 1872 for (;;) { 1873 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1874 /* 1875 * First take a look at events. 1876 * This is important to handle events before any I/O requests. 1877 */ 1878 ep = g_mirror_event_first(sc); 1879 if (ep != NULL) { 1880 g_mirror_event_remove(sc, ep); 1881 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1882 /* Update only device status. */ 1883 G_MIRROR_DEBUG(3, 1884 "Running event for device %s.", 1885 sc->sc_name); 1886 ep->e_error = 0; 1887 g_mirror_update_device(sc, true); 1888 } else { 1889 /* Update disk status. */ 1890 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1891 g_mirror_get_diskname(ep->e_disk)); 1892 ep->e_error = g_mirror_update_disk(ep->e_disk, 1893 ep->e_state); 1894 if (ep->e_error == 0) 1895 g_mirror_update_device(sc, false); 1896 } 1897 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1898 KASSERT(ep->e_error == 0, 1899 ("Error cannot be handled.")); 1900 g_mirror_event_free(ep); 1901 } else { 1902 ep->e_flags |= G_MIRROR_EVENT_DONE; 1903 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1904 ep); 1905 mtx_lock(&sc->sc_events_mtx); 1906 wakeup(ep); 1907 mtx_unlock(&sc->sc_events_mtx); 1908 } 1909 if ((sc->sc_flags & 1910 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1911 if (g_mirror_try_destroy(sc)) { 1912 curthread->td_pflags &= ~TDP_GEOM; 1913 G_MIRROR_DEBUG(1, "Thread exiting."); 1914 kproc_exit(0); 1915 } 1916 } 1917 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1918 continue; 1919 } 1920 /* 1921 * Check if we can mark array as CLEAN and if we can't take 1922 * how much seconds should we wait. 1923 */ 1924 timeout = g_mirror_idle(sc, -1); 1925 /* 1926 * Now I/O requests. 1927 */ 1928 /* Get first request from the queue. */ 1929 mtx_lock(&sc->sc_queue_mtx); 1930 bp = bioq_takefirst(&sc->sc_queue); 1931 if (bp == NULL) { 1932 if ((sc->sc_flags & 1933 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1934 mtx_unlock(&sc->sc_queue_mtx); 1935 if (g_mirror_try_destroy(sc)) { 1936 curthread->td_pflags &= ~TDP_GEOM; 1937 G_MIRROR_DEBUG(1, "Thread exiting."); 1938 kproc_exit(0); 1939 } 1940 mtx_lock(&sc->sc_queue_mtx); 1941 if (bioq_first(&sc->sc_queue) != NULL) { 1942 mtx_unlock(&sc->sc_queue_mtx); 1943 continue; 1944 } 1945 } 1946 if (g_mirror_event_first(sc) != NULL) 1947 continue; 1948 sx_xunlock(&sc->sc_lock); 1949 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1950 timeout * hz); 1951 sx_xlock(&sc->sc_lock); 1952 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1953 continue; 1954 } 1955 mtx_unlock(&sc->sc_queue_mtx); 1956 1957 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 1958 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1959 g_mirror_sync_request(bp); /* READ */ 1960 } else if (bp->bio_to != sc->sc_provider) { 1961 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 1962 g_mirror_regular_request(bp); 1963 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 1964 g_mirror_sync_request(bp); /* WRITE */ 1965 else { 1966 KASSERT(0, 1967 ("Invalid request cflags=0x%hx to=%s.", 1968 bp->bio_cflags, bp->bio_to->name)); 1969 } 1970 } else { 1971 g_mirror_register_request(bp); 1972 } 1973 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1974 } 1975} 1976 1977static void 1978g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 1979{ 1980 1981 sx_assert(&sc->sc_lock, SX_LOCKED); 1982 1983 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 1984 return; 1985 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1986 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 1987 g_mirror_get_diskname(disk), sc->sc_name); 1988 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1989 } else if (sc->sc_idle && 1990 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1991 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 1992 g_mirror_get_diskname(disk), sc->sc_name); 1993 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1994 } 1995} 1996 1997static void 1998g_mirror_sync_start(struct g_mirror_disk *disk) 1999{ 2000 struct g_mirror_softc *sc; 2001 struct g_consumer *cp; 2002 struct bio *bp; 2003 int error, i; 2004 2005 g_topology_assert_not(); 2006 sc = disk->d_softc; 2007 sx_assert(&sc->sc_lock, SX_LOCKED); 2008 2009 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2010 ("Disk %s is not marked for synchronization.", 2011 g_mirror_get_diskname(disk))); 2012 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2013 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 2014 sc->sc_state)); 2015 2016 sx_xunlock(&sc->sc_lock); 2017 g_topology_lock(); 2018 cp = g_new_consumer(sc->sc_sync.ds_geom); 2019 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 2020 error = g_attach(cp, sc->sc_provider); 2021 KASSERT(error == 0, 2022 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 2023 error = g_access(cp, 1, 0, 0); 2024 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 2025 g_topology_unlock(); 2026 sx_xlock(&sc->sc_lock); 2027 2028 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 2029 g_mirror_get_diskname(disk)); 2030 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0) 2031 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2032 KASSERT(disk->d_sync.ds_consumer == NULL, 2033 ("Sync consumer already exists (device=%s, disk=%s).", 2034 sc->sc_name, g_mirror_get_diskname(disk))); 2035 2036 disk->d_sync.ds_consumer = cp; 2037 disk->d_sync.ds_consumer->private = disk; 2038 disk->d_sync.ds_consumer->index = 0; 2039 2040 /* 2041 * Allocate memory for synchronization bios and initialize them. 2042 */ 2043 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 2044 M_MIRROR, M_WAITOK); 2045 for (i = 0; i < g_mirror_syncreqs; i++) { 2046 bp = g_alloc_bio(); 2047 disk->d_sync.ds_bios[i] = bp; 2048 bp->bio_parent = NULL; 2049 bp->bio_cmd = BIO_READ; 2050 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 2051 bp->bio_cflags = 0; 2052 bp->bio_offset = disk->d_sync.ds_offset; 2053 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 2054 disk->d_sync.ds_offset += bp->bio_length; 2055 bp->bio_done = g_mirror_sync_done; 2056 bp->bio_from = disk->d_sync.ds_consumer; 2057 bp->bio_to = sc->sc_provider; 2058 bp->bio_caller1 = (void *)(uintptr_t)i; 2059 } 2060 2061 /* Increase the number of disks in SYNCHRONIZING state. */ 2062 sc->sc_sync.ds_ndisks++; 2063 /* Set the number of in-flight synchronization requests. */ 2064 disk->d_sync.ds_inflight = g_mirror_syncreqs; 2065 2066 /* 2067 * Fire off first synchronization requests. 2068 */ 2069 for (i = 0; i < g_mirror_syncreqs; i++) { 2070 bp = disk->d_sync.ds_bios[i]; 2071 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 2072 disk->d_sync.ds_consumer->index++; 2073 /* 2074 * Delay the request if it is colliding with a regular request. 2075 */ 2076 if (g_mirror_regular_collision(sc, bp)) 2077 g_mirror_sync_delay(sc, bp); 2078 else 2079 g_io_request(bp, disk->d_sync.ds_consumer); 2080 } 2081} 2082 2083/* 2084 * Stop synchronization process. 2085 * type: 0 - synchronization finished 2086 * 1 - synchronization stopped 2087 */ 2088static void 2089g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 2090{ 2091 struct g_mirror_softc *sc; 2092 struct g_consumer *cp; 2093 2094 g_topology_assert_not(); 2095 sc = disk->d_softc; 2096 sx_assert(&sc->sc_lock, SX_LOCKED); 2097 2098 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2099 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2100 g_mirror_disk_state2str(disk->d_state))); 2101 if (disk->d_sync.ds_consumer == NULL) 2102 return; 2103 2104 if (type == 0) { 2105 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2106 sc->sc_name, g_mirror_get_diskname(disk)); 2107 } else /* if (type == 1) */ { 2108 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2109 sc->sc_name, g_mirror_get_diskname(disk)); 2110 } 2111 g_mirror_regular_release(sc); 2112 free(disk->d_sync.ds_bios, M_MIRROR); 2113 disk->d_sync.ds_bios = NULL; 2114 cp = disk->d_sync.ds_consumer; 2115 disk->d_sync.ds_consumer = NULL; 2116 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2117 sc->sc_sync.ds_ndisks--; 2118 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2119 g_topology_lock(); 2120 g_mirror_kill_consumer(sc, cp); 2121 g_topology_unlock(); 2122 sx_xlock(&sc->sc_lock); 2123} 2124 2125static void 2126g_mirror_launch_provider(struct g_mirror_softc *sc) 2127{ 2128 struct g_mirror_disk *disk; 2129 struct g_provider *pp, *dp; 2130 2131 sx_assert(&sc->sc_lock, SX_LOCKED); 2132 2133 g_topology_lock(); 2134 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 2135 pp->flags |= G_PF_DIRECT_RECEIVE; 2136 pp->mediasize = sc->sc_mediasize; 2137 pp->sectorsize = sc->sc_sectorsize; 2138 pp->stripesize = 0; 2139 pp->stripeoffset = 0; 2140 2141 /* Splitting of unmapped BIO's could work but isn't implemented now */ 2142 if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT) 2143 pp->flags |= G_PF_ACCEPT_UNMAPPED; 2144 2145 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2146 if (disk->d_consumer && disk->d_consumer->provider) { 2147 dp = disk->d_consumer->provider; 2148 if (dp->stripesize > pp->stripesize) { 2149 pp->stripesize = dp->stripesize; 2150 pp->stripeoffset = dp->stripeoffset; 2151 } 2152 /* A provider underneath us doesn't support unmapped */ 2153 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { 2154 G_MIRROR_DEBUG(0, "Cancelling unmapped " 2155 "because of %s.", dp->name); 2156 pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 2157 } 2158 } 2159 } 2160 pp->private = sc; 2161 sc->sc_refcnt++; 2162 sc->sc_provider = pp; 2163 g_error_provider(pp, 0); 2164 g_topology_unlock(); 2165 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2166 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 2167 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2168 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2169 g_mirror_sync_start(disk); 2170 } 2171} 2172 2173static void 2174g_mirror_destroy_provider(struct g_mirror_softc *sc) 2175{ 2176 struct g_mirror_disk *disk; 2177 struct bio *bp; 2178 2179 g_topology_assert_not(); 2180 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2181 sc->sc_name)); 2182 2183 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2184 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2185 g_mirror_sync_stop(disk, 1); 2186 } 2187 2188 g_topology_lock(); 2189 g_error_provider(sc->sc_provider, ENXIO); 2190 mtx_lock(&sc->sc_queue_mtx); 2191 while ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) { 2192 /* 2193 * Abort any pending I/O that wasn't generated by us. 2194 * Synchronization requests and requests destined for individual 2195 * mirror components can be destroyed immediately. 2196 */ 2197 if (bp->bio_to == sc->sc_provider && 2198 bp->bio_from->geom != sc->sc_sync.ds_geom) { 2199 g_io_deliver(bp, ENXIO); 2200 } else { 2201 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2202 free(bp->bio_data, M_MIRROR); 2203 g_destroy_bio(bp); 2204 } 2205 } 2206 mtx_unlock(&sc->sc_queue_mtx); 2207 g_wither_provider(sc->sc_provider, ENXIO); 2208 sc->sc_provider = NULL; 2209 G_MIRROR_DEBUG(0, "Device %s: provider destroyed.", sc->sc_name); 2210 g_topology_unlock(); 2211} 2212 2213static void 2214g_mirror_go(void *arg) 2215{ 2216 struct g_mirror_softc *sc; 2217 2218 sc = arg; 2219 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2220 g_mirror_event_send(sc, 0, 2221 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2222} 2223 2224static u_int 2225g_mirror_determine_state(struct g_mirror_disk *disk) 2226{ 2227 struct g_mirror_softc *sc; 2228 u_int state; 2229 2230 sc = disk->d_softc; 2231 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2232 if ((disk->d_flags & 2233 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0 && 2234 (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 || 2235 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0)) { 2236 /* Disk does not need synchronization. */ 2237 state = G_MIRROR_DISK_STATE_ACTIVE; 2238 } else { 2239 if ((sc->sc_flags & 2240 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2241 (disk->d_flags & 2242 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2243 /* 2244 * We can start synchronization from 2245 * the stored offset. 2246 */ 2247 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2248 } else { 2249 state = G_MIRROR_DISK_STATE_STALE; 2250 } 2251 } 2252 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2253 /* 2254 * Reset all synchronization data for this disk, 2255 * because if it even was synchronized, it was 2256 * synchronized to disks with different syncid. 2257 */ 2258 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2259 disk->d_sync.ds_offset = 0; 2260 disk->d_sync.ds_offset_done = 0; 2261 disk->d_sync.ds_syncid = sc->sc_syncid; 2262 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2263 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2264 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2265 } else { 2266 state = G_MIRROR_DISK_STATE_STALE; 2267 } 2268 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2269 /* 2270 * Not good, NOT GOOD! 2271 * It means that mirror was started on stale disks 2272 * and more fresh disk just arrive. 2273 * If there were writes, mirror is broken, sorry. 2274 * I think the best choice here is don't touch 2275 * this disk and inform the user loudly. 2276 */ 2277 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2278 "disk (%s) arrives!! It will not be connected to the " 2279 "running device.", sc->sc_name, 2280 g_mirror_get_diskname(disk)); 2281 g_mirror_destroy_disk(disk); 2282 state = G_MIRROR_DISK_STATE_NONE; 2283 /* Return immediately, because disk was destroyed. */ 2284 return (state); 2285 } 2286 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2287 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2288 return (state); 2289} 2290 2291/* 2292 * Update device state. 2293 */ 2294static void 2295g_mirror_update_device(struct g_mirror_softc *sc, bool force) 2296{ 2297 struct g_mirror_disk *disk; 2298 u_int state; 2299 2300 sx_assert(&sc->sc_lock, SX_XLOCKED); 2301 2302 switch (sc->sc_state) { 2303 case G_MIRROR_DEVICE_STATE_STARTING: 2304 { 2305 struct g_mirror_disk *pdisk, *tdisk; 2306 u_int dirty, ndisks, genid, syncid; 2307 bool broken; 2308 2309 KASSERT(sc->sc_provider == NULL, 2310 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2311 /* 2312 * Are we ready? We are, if all disks are connected or 2313 * if we have any disks and 'force' is true. 2314 */ 2315 ndisks = g_mirror_ndisks(sc, -1); 2316 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2317 ; 2318 } else if (ndisks == 0) { 2319 /* 2320 * Disks went down in starting phase, so destroy 2321 * device. 2322 */ 2323 callout_drain(&sc->sc_callout); 2324 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2325 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2326 sc->sc_rootmount); 2327 root_mount_rel(sc->sc_rootmount); 2328 sc->sc_rootmount = NULL; 2329 return; 2330 } else { 2331 return; 2332 } 2333 2334 /* 2335 * Activate all disks with the biggest syncid. 2336 */ 2337 if (force) { 2338 /* 2339 * If 'force' is true, we have been called due to 2340 * timeout, so don't bother canceling timeout. 2341 */ 2342 ndisks = 0; 2343 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2344 if ((disk->d_flags & 2345 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2346 ndisks++; 2347 } 2348 } 2349 if (ndisks == 0) { 2350 /* No valid disks found, destroy device. */ 2351 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2352 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2353 __LINE__, sc->sc_rootmount); 2354 root_mount_rel(sc->sc_rootmount); 2355 sc->sc_rootmount = NULL; 2356 return; 2357 } 2358 } else { 2359 /* Cancel timeout. */ 2360 callout_drain(&sc->sc_callout); 2361 } 2362 2363 /* 2364 * Find the biggest genid. 2365 */ 2366 genid = 0; 2367 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2368 if (disk->d_genid > genid) 2369 genid = disk->d_genid; 2370 } 2371 sc->sc_genid = genid; 2372 /* 2373 * Remove all disks without the biggest genid. 2374 */ 2375 broken = false; 2376 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2377 if (disk->d_genid < genid) { 2378 G_MIRROR_DEBUG(0, 2379 "Component %s (device %s) broken, skipping.", 2380 g_mirror_get_diskname(disk), sc->sc_name); 2381 g_mirror_destroy_disk(disk); 2382 /* 2383 * Bump the syncid in case we discover a healthy 2384 * replacement disk after starting the mirror. 2385 */ 2386 broken = true; 2387 } 2388 } 2389 2390 /* 2391 * Find the biggest syncid. 2392 */ 2393 syncid = 0; 2394 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2395 if (disk->d_sync.ds_syncid > syncid) 2396 syncid = disk->d_sync.ds_syncid; 2397 } 2398 2399 /* 2400 * Here we need to look for dirty disks and if all disks 2401 * with the biggest syncid are dirty, we have to choose 2402 * one with the biggest priority and rebuild the rest. 2403 */ 2404 /* 2405 * Find the number of dirty disks with the biggest syncid. 2406 * Find the number of disks with the biggest syncid. 2407 * While here, find a disk with the biggest priority. 2408 */ 2409 dirty = ndisks = 0; 2410 pdisk = NULL; 2411 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2412 if (disk->d_sync.ds_syncid != syncid) 2413 continue; 2414 if ((disk->d_flags & 2415 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2416 continue; 2417 } 2418 ndisks++; 2419 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2420 dirty++; 2421 if (pdisk == NULL || 2422 pdisk->d_priority < disk->d_priority) { 2423 pdisk = disk; 2424 } 2425 } 2426 } 2427 if (dirty == 0) { 2428 /* No dirty disks at all, great. */ 2429 } else if (dirty == ndisks) { 2430 /* 2431 * Force synchronization for all dirty disks except one 2432 * with the biggest priority. 2433 */ 2434 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2435 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2436 "master disk for synchronization.", 2437 g_mirror_get_diskname(pdisk), sc->sc_name); 2438 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2439 if (disk->d_sync.ds_syncid != syncid) 2440 continue; 2441 if ((disk->d_flags & 2442 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2443 continue; 2444 } 2445 KASSERT((disk->d_flags & 2446 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2447 ("Disk %s isn't marked as dirty.", 2448 g_mirror_get_diskname(disk))); 2449 /* Skip the disk with the biggest priority. */ 2450 if (disk == pdisk) 2451 continue; 2452 disk->d_sync.ds_syncid = 0; 2453 } 2454 } else if (dirty < ndisks) { 2455 /* 2456 * Force synchronization for all dirty disks. 2457 * We have some non-dirty disks. 2458 */ 2459 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2460 if (disk->d_sync.ds_syncid != syncid) 2461 continue; 2462 if ((disk->d_flags & 2463 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2464 continue; 2465 } 2466 if ((disk->d_flags & 2467 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2468 continue; 2469 } 2470 disk->d_sync.ds_syncid = 0; 2471 } 2472 } 2473 2474 /* Reset hint. */ 2475 sc->sc_hint = NULL; 2476 sc->sc_syncid = syncid; 2477 if (force || broken) { 2478 /* Remember to bump syncid on first write. */ 2479 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2480 } 2481 state = G_MIRROR_DEVICE_STATE_RUNNING; 2482 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2483 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2484 g_mirror_device_state2str(state)); 2485 sc->sc_state = state; 2486 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2487 state = g_mirror_determine_state(disk); 2488 g_mirror_event_send(disk, state, 2489 G_MIRROR_EVENT_DONTWAIT); 2490 if (state == G_MIRROR_DISK_STATE_STALE) 2491 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2492 } 2493 break; 2494 } 2495 case G_MIRROR_DEVICE_STATE_RUNNING: 2496 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2497 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2498 /* 2499 * No usable disks, so destroy the device. 2500 */ 2501 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2502 break; 2503 } else if (g_mirror_ndisks(sc, 2504 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2505 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2506 /* 2507 * We have active disks, launch provider if it doesn't 2508 * exist. 2509 */ 2510 if (sc->sc_provider == NULL) 2511 g_mirror_launch_provider(sc); 2512 if (sc->sc_rootmount != NULL) { 2513 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2514 __LINE__, sc->sc_rootmount); 2515 root_mount_rel(sc->sc_rootmount); 2516 sc->sc_rootmount = NULL; 2517 } 2518 } 2519 /* 2520 * Genid should be bumped immediately, so do it here. 2521 */ 2522 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2523 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2524 g_mirror_bump_genid(sc); 2525 } 2526 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_NOW) != 0) { 2527 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID_NOW; 2528 g_mirror_bump_syncid(sc); 2529 } 2530 break; 2531 default: 2532 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2533 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2534 break; 2535 } 2536} 2537 2538/* 2539 * Update disk state and device state if needed. 2540 */ 2541#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2542 "Disk %s state changed from %s to %s (device %s).", \ 2543 g_mirror_get_diskname(disk), \ 2544 g_mirror_disk_state2str(disk->d_state), \ 2545 g_mirror_disk_state2str(state), sc->sc_name) 2546static int 2547g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2548{ 2549 struct g_mirror_softc *sc; 2550 2551 sc = disk->d_softc; 2552 sx_assert(&sc->sc_lock, SX_XLOCKED); 2553 2554again: 2555 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2556 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2557 g_mirror_disk_state2str(state)); 2558 switch (state) { 2559 case G_MIRROR_DISK_STATE_NEW: 2560 /* 2561 * Possible scenarios: 2562 * 1. New disk arrive. 2563 */ 2564 /* Previous state should be NONE. */ 2565 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2566 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2567 g_mirror_disk_state2str(disk->d_state))); 2568 DISK_STATE_CHANGED(); 2569 2570 disk->d_state = state; 2571 if (LIST_EMPTY(&sc->sc_disks)) 2572 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2573 else { 2574 struct g_mirror_disk *dp; 2575 2576 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2577 if (disk->d_priority >= dp->d_priority) { 2578 LIST_INSERT_BEFORE(dp, disk, d_next); 2579 dp = NULL; 2580 break; 2581 } 2582 if (LIST_NEXT(dp, d_next) == NULL) 2583 break; 2584 } 2585 if (dp != NULL) 2586 LIST_INSERT_AFTER(dp, disk, d_next); 2587 } 2588 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2589 sc->sc_name, g_mirror_get_diskname(disk)); 2590 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2591 break; 2592 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2593 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2594 g_mirror_device_state2str(sc->sc_state), 2595 g_mirror_get_diskname(disk), 2596 g_mirror_disk_state2str(disk->d_state))); 2597 state = g_mirror_determine_state(disk); 2598 if (state != G_MIRROR_DISK_STATE_NONE) 2599 goto again; 2600 break; 2601 case G_MIRROR_DISK_STATE_ACTIVE: 2602 /* 2603 * Possible scenarios: 2604 * 1. New disk does not need synchronization. 2605 * 2. Synchronization process finished successfully. 2606 */ 2607 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2608 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2609 g_mirror_device_state2str(sc->sc_state), 2610 g_mirror_get_diskname(disk), 2611 g_mirror_disk_state2str(disk->d_state))); 2612 /* Previous state should be NEW or SYNCHRONIZING. */ 2613 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2614 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2615 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2616 g_mirror_disk_state2str(disk->d_state))); 2617 DISK_STATE_CHANGED(); 2618 2619 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2620 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2621 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2622 g_mirror_sync_stop(disk, 0); 2623 } 2624 disk->d_state = state; 2625 disk->d_sync.ds_offset = 0; 2626 disk->d_sync.ds_offset_done = 0; 2627 g_mirror_update_idle(sc, disk); 2628 g_mirror_update_metadata(disk); 2629 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2630 sc->sc_name, g_mirror_get_diskname(disk)); 2631 break; 2632 case G_MIRROR_DISK_STATE_STALE: 2633 /* 2634 * Possible scenarios: 2635 * 1. Stale disk was connected. 2636 */ 2637 /* Previous state should be NEW. */ 2638 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2639 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2640 g_mirror_disk_state2str(disk->d_state))); 2641 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2642 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2643 g_mirror_device_state2str(sc->sc_state), 2644 g_mirror_get_diskname(disk), 2645 g_mirror_disk_state2str(disk->d_state))); 2646 /* 2647 * STALE state is only possible if device is marked 2648 * NOAUTOSYNC. 2649 */ 2650 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2651 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2652 g_mirror_device_state2str(sc->sc_state), 2653 g_mirror_get_diskname(disk), 2654 g_mirror_disk_state2str(disk->d_state))); 2655 DISK_STATE_CHANGED(); 2656 2657 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2658 disk->d_state = state; 2659 g_mirror_update_metadata(disk); 2660 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2661 sc->sc_name, g_mirror_get_diskname(disk)); 2662 break; 2663 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2664 /* 2665 * Possible scenarios: 2666 * 1. Disk which needs synchronization was connected. 2667 */ 2668 /* Previous state should be NEW. */ 2669 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2670 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2671 g_mirror_disk_state2str(disk->d_state))); 2672 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2673 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2674 g_mirror_device_state2str(sc->sc_state), 2675 g_mirror_get_diskname(disk), 2676 g_mirror_disk_state2str(disk->d_state))); 2677 DISK_STATE_CHANGED(); 2678 2679 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2680 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2681 disk->d_state = state; 2682 if (sc->sc_provider != NULL) { 2683 g_mirror_sync_start(disk); 2684 g_mirror_update_metadata(disk); 2685 } 2686 break; 2687 case G_MIRROR_DISK_STATE_DISCONNECTED: 2688 /* 2689 * Possible scenarios: 2690 * 1. Device wasn't running yet, but disk disappear. 2691 * 2. Disk was active and disapppear. 2692 * 3. Disk disappear during synchronization process. 2693 */ 2694 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2695 /* 2696 * Previous state should be ACTIVE, STALE or 2697 * SYNCHRONIZING. 2698 */ 2699 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2700 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2701 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2702 ("Wrong disk state (%s, %s).", 2703 g_mirror_get_diskname(disk), 2704 g_mirror_disk_state2str(disk->d_state))); 2705 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2706 /* Previous state should be NEW. */ 2707 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2708 ("Wrong disk state (%s, %s).", 2709 g_mirror_get_diskname(disk), 2710 g_mirror_disk_state2str(disk->d_state))); 2711 /* 2712 * Reset bumping syncid if disk disappeared in STARTING 2713 * state. 2714 */ 2715 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2716 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2717#ifdef INVARIANTS 2718 } else { 2719 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2720 sc->sc_name, 2721 g_mirror_device_state2str(sc->sc_state), 2722 g_mirror_get_diskname(disk), 2723 g_mirror_disk_state2str(disk->d_state))); 2724#endif 2725 } 2726 DISK_STATE_CHANGED(); 2727 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2728 sc->sc_name, g_mirror_get_diskname(disk)); 2729 2730 g_mirror_destroy_disk(disk); 2731 break; 2732 case G_MIRROR_DISK_STATE_DESTROY: 2733 { 2734 int error; 2735 2736 error = g_mirror_clear_metadata(disk); 2737 if (error != 0) { 2738 G_MIRROR_DEBUG(0, 2739 "Device %s: failed to clear metadata on %s: %d.", 2740 sc->sc_name, g_mirror_get_diskname(disk), error); 2741 break; 2742 } 2743 DISK_STATE_CHANGED(); 2744 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2745 sc->sc_name, g_mirror_get_diskname(disk)); 2746 2747 g_mirror_destroy_disk(disk); 2748 sc->sc_ndisks--; 2749 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2750 g_mirror_update_metadata(disk); 2751 } 2752 break; 2753 } 2754 default: 2755 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2756 break; 2757 } 2758 return (0); 2759} 2760#undef DISK_STATE_CHANGED 2761 2762int 2763g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2764{ 2765 struct g_provider *pp; 2766 u_char *buf; 2767 int error; 2768 2769 g_topology_assert(); 2770 2771 error = g_access(cp, 1, 0, 0); 2772 if (error != 0) 2773 return (error); 2774 pp = cp->provider; 2775 g_topology_unlock(); 2776 /* Metadata are stored on last sector. */ 2777 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2778 &error); 2779 g_topology_lock(); 2780 g_access(cp, -1, 0, 0); 2781 if (buf == NULL) { 2782 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2783 cp->provider->name, error); 2784 return (error); 2785 } 2786 2787 /* Decode metadata. */ 2788 error = mirror_metadata_decode(buf, md); 2789 g_free(buf); 2790 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2791 return (EINVAL); 2792 if (md->md_version > G_MIRROR_VERSION) { 2793 G_MIRROR_DEBUG(0, 2794 "Kernel module is too old to handle metadata from %s.", 2795 cp->provider->name); 2796 return (EINVAL); 2797 } 2798 if (error != 0) { 2799 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2800 cp->provider->name); 2801 return (error); 2802 } 2803 2804 return (0); 2805} 2806 2807static int 2808g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2809 struct g_mirror_metadata *md) 2810{ 2811 2812 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2813 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2814 pp->name, md->md_did); 2815 return (EEXIST); 2816 } 2817 if (md->md_all != sc->sc_ndisks) { 2818 G_MIRROR_DEBUG(1, 2819 "Invalid '%s' field on disk %s (device %s), skipping.", 2820 "md_all", pp->name, sc->sc_name); 2821 return (EINVAL); 2822 } 2823 if (md->md_slice != sc->sc_slice) { 2824 G_MIRROR_DEBUG(1, 2825 "Invalid '%s' field on disk %s (device %s), skipping.", 2826 "md_slice", pp->name, sc->sc_name); 2827 return (EINVAL); 2828 } 2829 if (md->md_balance != sc->sc_balance) { 2830 G_MIRROR_DEBUG(1, 2831 "Invalid '%s' field on disk %s (device %s), skipping.", 2832 "md_balance", pp->name, sc->sc_name); 2833 return (EINVAL); 2834 } 2835#if 0 2836 if (md->md_mediasize != sc->sc_mediasize) { 2837 G_MIRROR_DEBUG(1, 2838 "Invalid '%s' field on disk %s (device %s), skipping.", 2839 "md_mediasize", pp->name, sc->sc_name); 2840 return (EINVAL); 2841 } 2842#endif 2843 if (sc->sc_mediasize > pp->mediasize) { 2844 G_MIRROR_DEBUG(1, 2845 "Invalid size of disk %s (device %s), skipping.", pp->name, 2846 sc->sc_name); 2847 return (EINVAL); 2848 } 2849 if (md->md_sectorsize != sc->sc_sectorsize) { 2850 G_MIRROR_DEBUG(1, 2851 "Invalid '%s' field on disk %s (device %s), skipping.", 2852 "md_sectorsize", pp->name, sc->sc_name); 2853 return (EINVAL); 2854 } 2855 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2856 G_MIRROR_DEBUG(1, 2857 "Invalid sector size of disk %s (device %s), skipping.", 2858 pp->name, sc->sc_name); 2859 return (EINVAL); 2860 } 2861 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2862 G_MIRROR_DEBUG(1, 2863 "Invalid device flags on disk %s (device %s), skipping.", 2864 pp->name, sc->sc_name); 2865 return (EINVAL); 2866 } 2867 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2868 G_MIRROR_DEBUG(1, 2869 "Invalid disk flags on disk %s (device %s), skipping.", 2870 pp->name, sc->sc_name); 2871 return (EINVAL); 2872 } 2873 return (0); 2874} 2875 2876int 2877g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2878 struct g_mirror_metadata *md) 2879{ 2880 struct g_mirror_disk *disk; 2881 int error; 2882 2883 g_topology_assert_not(); 2884 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2885 2886 error = g_mirror_check_metadata(sc, pp, md); 2887 if (error != 0) 2888 return (error); 2889 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2890 md->md_genid < sc->sc_genid) { 2891 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2892 pp->name, sc->sc_name); 2893 return (EINVAL); 2894 } 2895 disk = g_mirror_init_disk(sc, pp, md, &error); 2896 if (disk == NULL) 2897 return (error); 2898 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2899 G_MIRROR_EVENT_WAIT); 2900 if (error != 0) 2901 return (error); 2902 if (md->md_version < G_MIRROR_VERSION) { 2903 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2904 pp->name, md->md_version, G_MIRROR_VERSION); 2905 g_mirror_update_metadata(disk); 2906 } 2907 return (0); 2908} 2909 2910static void 2911g_mirror_destroy_delayed(void *arg, int flag) 2912{ 2913 struct g_mirror_softc *sc; 2914 int error; 2915 2916 if (flag == EV_CANCEL) { 2917 G_MIRROR_DEBUG(1, "Destroying canceled."); 2918 return; 2919 } 2920 sc = arg; 2921 g_topology_unlock(); 2922 sx_xlock(&sc->sc_lock); 2923 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2924 ("DESTROY flag set on %s.", sc->sc_name)); 2925 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0, 2926 ("CLOSEWAIT flag not set on %s.", sc->sc_name)); 2927 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 2928 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 2929 if (error != 0) { 2930 G_MIRROR_DEBUG(0, "Cannot destroy %s (error=%d).", 2931 sc->sc_name, error); 2932 sx_xunlock(&sc->sc_lock); 2933 } 2934 g_topology_lock(); 2935} 2936 2937static int 2938g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2939{ 2940 struct g_mirror_softc *sc; 2941 int error = 0; 2942 2943 g_topology_assert(); 2944 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2945 acw, ace); 2946 2947 sc = pp->private; 2948 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 2949 2950 g_topology_unlock(); 2951 sx_xlock(&sc->sc_lock); 2952 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 2953 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 || 2954 LIST_EMPTY(&sc->sc_disks)) { 2955 if (acr > 0 || acw > 0 || ace > 0) 2956 error = ENXIO; 2957 goto end; 2958 } 2959 sc->sc_provider_open += acr + acw + ace; 2960 if (pp->acw + acw == 0) 2961 g_mirror_idle(sc, 0); 2962 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 && 2963 sc->sc_provider_open == 0) 2964 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, sc, NULL); 2965end: 2966 sx_xunlock(&sc->sc_lock); 2967 g_topology_lock(); 2968 return (error); 2969} 2970 2971struct g_geom * 2972g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md, 2973 u_int type) 2974{ 2975 struct g_mirror_softc *sc; 2976 struct g_geom *gp; 2977 int error, timeout; 2978 2979 g_topology_assert(); 2980 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2981 md->md_mid); 2982 2983 /* One disk is minimum. */ 2984 if (md->md_all < 1) 2985 return (NULL); 2986 /* 2987 * Action geom. 2988 */ 2989 gp = g_new_geomf(mp, "%s", md->md_name); 2990 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2991 gp->start = g_mirror_start; 2992 gp->orphan = g_mirror_orphan; 2993 gp->access = g_mirror_access; 2994 gp->dumpconf = g_mirror_dumpconf; 2995 2996 sc->sc_type = type; 2997 sc->sc_id = md->md_mid; 2998 sc->sc_slice = md->md_slice; 2999 sc->sc_balance = md->md_balance; 3000 sc->sc_mediasize = md->md_mediasize; 3001 sc->sc_sectorsize = md->md_sectorsize; 3002 sc->sc_ndisks = md->md_all; 3003 sc->sc_flags = md->md_mflags; 3004 sc->sc_bump_id = 0; 3005 sc->sc_idle = 1; 3006 sc->sc_last_write = time_uptime; 3007 sc->sc_writes = 0; 3008 sc->sc_refcnt = 1; 3009 sx_init(&sc->sc_lock, "gmirror:lock"); 3010 bioq_init(&sc->sc_queue); 3011 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 3012 bioq_init(&sc->sc_regular_delayed); 3013 bioq_init(&sc->sc_inflight); 3014 bioq_init(&sc->sc_sync_delayed); 3015 LIST_INIT(&sc->sc_disks); 3016 TAILQ_INIT(&sc->sc_events); 3017 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 3018 callout_init(&sc->sc_callout, 1); 3019 mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF); 3020 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 3021 gp->softc = sc; 3022 sc->sc_geom = gp; 3023 sc->sc_provider = NULL; 3024 sc->sc_provider_open = 0; 3025 /* 3026 * Synchronization geom. 3027 */ 3028 gp = g_new_geomf(mp, "%s.sync", md->md_name); 3029 gp->softc = sc; 3030 gp->orphan = g_mirror_orphan; 3031 sc->sc_sync.ds_geom = gp; 3032 sc->sc_sync.ds_ndisks = 0; 3033 error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 3034 "g_mirror %s", md->md_name); 3035 if (error != 0) { 3036 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 3037 sc->sc_name); 3038 g_destroy_geom(sc->sc_sync.ds_geom); 3039 g_destroy_geom(sc->sc_geom); 3040 g_mirror_free_device(sc); 3041 return (NULL); 3042 } 3043 3044 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 3045 sc->sc_name, sc->sc_ndisks, sc->sc_id); 3046 3047 sc->sc_rootmount = root_mount_hold("GMIRROR"); 3048 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 3049 /* 3050 * Run timeout. 3051 */ 3052 timeout = g_mirror_timeout * hz; 3053 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 3054 return (sc->sc_geom); 3055} 3056 3057int 3058g_mirror_destroy(struct g_mirror_softc *sc, int how) 3059{ 3060 struct g_mirror_disk *disk; 3061 3062 g_topology_assert_not(); 3063 sx_assert(&sc->sc_lock, SX_XLOCKED); 3064 3065 if (sc->sc_provider_open != 0) { 3066 switch (how) { 3067 case G_MIRROR_DESTROY_SOFT: 3068 G_MIRROR_DEBUG(1, 3069 "Device %s is still open (%d).", sc->sc_name, 3070 sc->sc_provider_open); 3071 return (EBUSY); 3072 case G_MIRROR_DESTROY_DELAYED: 3073 G_MIRROR_DEBUG(1, 3074 "Device %s will be destroyed on last close.", 3075 sc->sc_name); 3076 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 3077 if (disk->d_state == 3078 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3079 g_mirror_sync_stop(disk, 1); 3080 } 3081 } 3082 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_CLOSEWAIT; 3083 return (EBUSY); 3084 case G_MIRROR_DESTROY_HARD: 3085 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 3086 "can't be definitely removed.", sc->sc_name); 3087 } 3088 } 3089 3090 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3091 sx_xunlock(&sc->sc_lock); 3092 return (0); 3093 } 3094 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 3095 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DRAIN; 3096 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3097 sx_xunlock(&sc->sc_lock); 3098 mtx_lock(&sc->sc_queue_mtx); 3099 wakeup(sc); 3100 mtx_unlock(&sc->sc_queue_mtx); 3101 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3102 while (sc->sc_worker != NULL) 3103 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 3104 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3105 sx_xlock(&sc->sc_lock); 3106 g_mirror_destroy_device(sc); 3107 return (0); 3108} 3109 3110static void 3111g_mirror_taste_orphan(struct g_consumer *cp) 3112{ 3113 3114 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3115 cp->provider->name)); 3116} 3117 3118static struct g_geom * 3119g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3120{ 3121 struct g_mirror_metadata md; 3122 struct g_mirror_softc *sc; 3123 struct g_consumer *cp; 3124 struct g_geom *gp; 3125 int error; 3126 3127 g_topology_assert(); 3128 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3129 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 3130 3131 gp = g_new_geomf(mp, "mirror:taste"); 3132 /* 3133 * This orphan function should be never called. 3134 */ 3135 gp->orphan = g_mirror_taste_orphan; 3136 cp = g_new_consumer(gp); 3137 g_attach(cp, pp); 3138 error = g_mirror_read_metadata(cp, &md); 3139 g_detach(cp); 3140 g_destroy_consumer(cp); 3141 g_destroy_geom(gp); 3142 if (error != 0) 3143 return (NULL); 3144 gp = NULL; 3145 3146 if (md.md_provider[0] != '\0' && 3147 !g_compare_names(md.md_provider, pp->name)) 3148 return (NULL); 3149 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3150 return (NULL); 3151 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 3152 G_MIRROR_DEBUG(0, 3153 "Device %s: provider %s marked as inactive, skipping.", 3154 md.md_name, pp->name); 3155 return (NULL); 3156 } 3157 if (g_mirror_debug >= 2) 3158 mirror_metadata_dump(&md); 3159 3160 /* 3161 * Let's check if device already exists. 3162 */ 3163 sc = NULL; 3164 LIST_FOREACH(gp, &mp->geom, geom) { 3165 sc = gp->softc; 3166 if (sc == NULL) 3167 continue; 3168 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 3169 continue; 3170 if (sc->sc_sync.ds_geom == gp) 3171 continue; 3172 if (strcmp(md.md_name, sc->sc_name) != 0) 3173 continue; 3174 if (md.md_mid != sc->sc_id) { 3175 G_MIRROR_DEBUG(0, "Device %s already configured.", 3176 sc->sc_name); 3177 return (NULL); 3178 } 3179 break; 3180 } 3181 if (gp == NULL) { 3182 gp = g_mirror_create(mp, &md, G_MIRROR_TYPE_AUTOMATIC); 3183 if (gp == NULL) { 3184 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3185 md.md_name); 3186 return (NULL); 3187 } 3188 sc = gp->softc; 3189 } 3190 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3191 g_topology_unlock(); 3192 sx_xlock(&sc->sc_lock); 3193 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; 3194 error = g_mirror_add_disk(sc, pp, &md); 3195 if (error != 0) { 3196 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3197 pp->name, gp->name, error); 3198 if (LIST_EMPTY(&sc->sc_disks)) { 3199 g_cancel_event(sc); 3200 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3201 g_topology_lock(); 3202 return (NULL); 3203 } 3204 gp = NULL; 3205 } 3206 sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; 3207 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3208 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3209 g_topology_lock(); 3210 return (NULL); 3211 } 3212 sx_xunlock(&sc->sc_lock); 3213 g_topology_lock(); 3214 return (gp); 3215} 3216 3217static void 3218g_mirror_resize(struct g_consumer *cp) 3219{ 3220 struct g_mirror_disk *disk; 3221 3222 g_topology_assert(); 3223 g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name); 3224 3225 disk = cp->private; 3226 if (disk == NULL) 3227 return; 3228 g_topology_unlock(); 3229 g_mirror_update_metadata(disk); 3230 g_topology_lock(); 3231} 3232 3233static int 3234g_mirror_destroy_geom(struct gctl_req *req __unused, 3235 struct g_class *mp __unused, struct g_geom *gp) 3236{ 3237 struct g_mirror_softc *sc; 3238 int error; 3239 3240 g_topology_unlock(); 3241 sc = gp->softc; 3242 sx_xlock(&sc->sc_lock); 3243 g_cancel_event(sc); 3244 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3245 if (error != 0) 3246 sx_xunlock(&sc->sc_lock); 3247 g_topology_lock(); 3248 return (error); 3249} 3250 3251static void 3252g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3253 struct g_consumer *cp, struct g_provider *pp) 3254{ 3255 struct g_mirror_softc *sc; 3256 3257 g_topology_assert(); 3258 3259 sc = gp->softc; 3260 if (sc == NULL) 3261 return; 3262 /* Skip synchronization geom. */ 3263 if (gp == sc->sc_sync.ds_geom) 3264 return; 3265 if (pp != NULL) { 3266 /* Nothing here. */ 3267 } else if (cp != NULL) { 3268 struct g_mirror_disk *disk; 3269 3270 disk = cp->private; 3271 if (disk == NULL) 3272 return; 3273 g_topology_unlock(); 3274 sx_xlock(&sc->sc_lock); 3275 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3276 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3277 sbuf_printf(sb, "%s<Synchronized>", indent); 3278 if (disk->d_sync.ds_offset == 0) 3279 sbuf_printf(sb, "0%%"); 3280 else { 3281 sbuf_printf(sb, "%u%%", 3282 (u_int)((disk->d_sync.ds_offset * 100) / 3283 sc->sc_provider->mediasize)); 3284 } 3285 sbuf_printf(sb, "</Synchronized>\n"); 3286 if (disk->d_sync.ds_offset > 0) { 3287 sbuf_printf(sb, "%s<BytesSynced>%jd" 3288 "</BytesSynced>\n", indent, 3289 (intmax_t)disk->d_sync.ds_offset); 3290 } 3291 } 3292 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3293 disk->d_sync.ds_syncid); 3294 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3295 disk->d_genid); 3296 sbuf_printf(sb, "%s<Flags>", indent); 3297 if (disk->d_flags == 0) 3298 sbuf_printf(sb, "NONE"); 3299 else { 3300 int first = 1; 3301 3302#define ADD_FLAG(flag, name) do { \ 3303 if ((disk->d_flags & (flag)) != 0) { \ 3304 if (!first) \ 3305 sbuf_printf(sb, ", "); \ 3306 else \ 3307 first = 0; \ 3308 sbuf_printf(sb, name); \ 3309 } \ 3310} while (0) 3311 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3312 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3313 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3314 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3315 "SYNCHRONIZING"); 3316 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3317 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3318#undef ADD_FLAG 3319 } 3320 sbuf_printf(sb, "</Flags>\n"); 3321 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3322 disk->d_priority); 3323 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3324 g_mirror_disk_state2str(disk->d_state)); 3325 sx_xunlock(&sc->sc_lock); 3326 g_topology_lock(); 3327 } else { 3328 g_topology_unlock(); 3329 sx_xlock(&sc->sc_lock); 3330 sbuf_printf(sb, "%s<Type>", indent); 3331 switch (sc->sc_type) { 3332 case G_MIRROR_TYPE_AUTOMATIC: 3333 sbuf_printf(sb, "AUTOMATIC"); 3334 break; 3335 case G_MIRROR_TYPE_MANUAL: 3336 sbuf_printf(sb, "MANUAL"); 3337 break; 3338 default: 3339 sbuf_printf(sb, "UNKNOWN"); 3340 break; 3341 } 3342 sbuf_printf(sb, "</Type>\n"); 3343 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3344 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3345 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3346 sbuf_printf(sb, "%s<Flags>", indent); 3347 if (sc->sc_flags == 0) 3348 sbuf_printf(sb, "NONE"); 3349 else { 3350 int first = 1; 3351 3352#define ADD_FLAG(flag, name) do { \ 3353 if ((sc->sc_flags & (flag)) != 0) { \ 3354 if (!first) \ 3355 sbuf_printf(sb, ", "); \ 3356 else \ 3357 first = 0; \ 3358 sbuf_printf(sb, name); \ 3359 } \ 3360} while (0) 3361 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3362 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3363#undef ADD_FLAG 3364 } 3365 sbuf_printf(sb, "</Flags>\n"); 3366 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3367 (u_int)sc->sc_slice); 3368 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3369 balance_name(sc->sc_balance)); 3370 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3371 sc->sc_ndisks); 3372 sbuf_printf(sb, "%s<State>", indent); 3373 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3374 sbuf_printf(sb, "%s", "STARTING"); 3375 else if (sc->sc_ndisks == 3376 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3377 sbuf_printf(sb, "%s", "COMPLETE"); 3378 else 3379 sbuf_printf(sb, "%s", "DEGRADED"); 3380 sbuf_printf(sb, "</State>\n"); 3381 sx_xunlock(&sc->sc_lock); 3382 g_topology_lock(); 3383 } 3384} 3385 3386static void 3387g_mirror_shutdown_post_sync(void *arg, int howto) 3388{ 3389 struct g_class *mp; 3390 struct g_geom *gp, *gp2; 3391 struct g_mirror_softc *sc; 3392 int error; 3393 3394 if (panicstr != NULL) 3395 return; 3396 3397 mp = arg; 3398 g_topology_lock(); 3399 g_mirror_shutdown = 1; 3400 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3401 if ((sc = gp->softc) == NULL) 3402 continue; 3403 /* Skip synchronization geom. */ 3404 if (gp == sc->sc_sync.ds_geom) 3405 continue; 3406 g_topology_unlock(); 3407 sx_xlock(&sc->sc_lock); 3408 g_mirror_idle(sc, -1); 3409 g_cancel_event(sc); 3410 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3411 if (error != 0) 3412 sx_xunlock(&sc->sc_lock); 3413 g_topology_lock(); 3414 } 3415 g_topology_unlock(); 3416} 3417 3418static void 3419g_mirror_init(struct g_class *mp) 3420{ 3421 3422 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3423 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3424 if (g_mirror_post_sync == NULL) 3425 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3426} 3427 3428static void 3429g_mirror_fini(struct g_class *mp) 3430{ 3431 3432 if (g_mirror_post_sync != NULL) 3433 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); 3434} 3435 3436DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3437