g_mirror.c revision 309205
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 309205 2016-11-27 05:58:47Z mav $"); 29 30#include <sys/param.h> 31#include <sys/systm.h> 32#include <sys/kernel.h> 33#include <sys/module.h> 34#include <sys/limits.h> 35#include <sys/lock.h> 36#include <sys/mutex.h> 37#include <sys/bio.h> 38#include <sys/sbuf.h> 39#include <sys/sysctl.h> 40#include <sys/malloc.h> 41#include <sys/eventhandler.h> 42#include <vm/uma.h> 43#include <geom/geom.h> 44#include <sys/proc.h> 45#include <sys/kthread.h> 46#include <sys/sched.h> 47#include <geom/mirror/g_mirror.h> 48 49FEATURE(geom_mirror, "GEOM mirroring support"); 50 51static MALLOC_DEFINE(M_MIRROR, "mirror_data", "GEOM_MIRROR Data"); 52 53SYSCTL_DECL(_kern_geom); 54static SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, 55 "GEOM_MIRROR stuff"); 56u_int g_mirror_debug = 0; 57SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RWTUN, &g_mirror_debug, 0, 58 "Debug level"); 59static u_int g_mirror_timeout = 4; 60SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RWTUN, &g_mirror_timeout, 61 0, "Time to wait on all mirror components"); 62static u_int g_mirror_idletime = 5; 63SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RWTUN, 64 &g_mirror_idletime, 0, "Mark components as clean when idling"); 65static u_int g_mirror_disconnect_on_failure = 1; 66SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN, 67 &g_mirror_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 68static u_int g_mirror_syncreqs = 2; 69SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_requests, CTLFLAG_RDTUN, 70 &g_mirror_syncreqs, 0, "Parallel synchronization I/O requests."); 71 72#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 73 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 74 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 75 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 76} while (0) 77 78static eventhandler_tag g_mirror_post_sync = NULL; 79static int g_mirror_shutdown = 0; 80 81static g_ctl_destroy_geom_t g_mirror_destroy_geom; 82static g_taste_t g_mirror_taste; 83static g_init_t g_mirror_init; 84static g_fini_t g_mirror_fini; 85static g_provgone_t g_mirror_providergone; 86static g_resize_t g_mirror_resize; 87 88struct g_class g_mirror_class = { 89 .name = G_MIRROR_CLASS_NAME, 90 .version = G_VERSION, 91 .ctlreq = g_mirror_config, 92 .taste = g_mirror_taste, 93 .destroy_geom = g_mirror_destroy_geom, 94 .init = g_mirror_init, 95 .fini = g_mirror_fini, 96 .providergone = g_mirror_providergone, 97 .resize = g_mirror_resize 98}; 99 100 101static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 102static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 103static void g_mirror_update_device(struct g_mirror_softc *sc, bool force); 104static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 105 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 106static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 107static void g_mirror_register_request(struct bio *bp); 108static void g_mirror_sync_release(struct g_mirror_softc *sc); 109 110 111static const char * 112g_mirror_disk_state2str(int state) 113{ 114 115 switch (state) { 116 case G_MIRROR_DISK_STATE_NONE: 117 return ("NONE"); 118 case G_MIRROR_DISK_STATE_NEW: 119 return ("NEW"); 120 case G_MIRROR_DISK_STATE_ACTIVE: 121 return ("ACTIVE"); 122 case G_MIRROR_DISK_STATE_STALE: 123 return ("STALE"); 124 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 125 return ("SYNCHRONIZING"); 126 case G_MIRROR_DISK_STATE_DISCONNECTED: 127 return ("DISCONNECTED"); 128 case G_MIRROR_DISK_STATE_DESTROY: 129 return ("DESTROY"); 130 default: 131 return ("INVALID"); 132 } 133} 134 135static const char * 136g_mirror_device_state2str(int state) 137{ 138 139 switch (state) { 140 case G_MIRROR_DEVICE_STATE_STARTING: 141 return ("STARTING"); 142 case G_MIRROR_DEVICE_STATE_RUNNING: 143 return ("RUNNING"); 144 default: 145 return ("INVALID"); 146 } 147} 148 149static const char * 150g_mirror_get_diskname(struct g_mirror_disk *disk) 151{ 152 153 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 154 return ("[unknown]"); 155 return (disk->d_name); 156} 157 158/* 159 * --- Events handling functions --- 160 * Events in geom_mirror are used to maintain disks and device status 161 * from one thread to simplify locking. 162 */ 163static void 164g_mirror_event_free(struct g_mirror_event *ep) 165{ 166 167 free(ep, M_MIRROR); 168} 169 170int 171g_mirror_event_send(void *arg, int state, int flags) 172{ 173 struct g_mirror_softc *sc; 174 struct g_mirror_disk *disk; 175 struct g_mirror_event *ep; 176 int error; 177 178 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 179 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 180 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 181 disk = NULL; 182 sc = arg; 183 } else { 184 disk = arg; 185 sc = disk->d_softc; 186 } 187 ep->e_disk = disk; 188 ep->e_state = state; 189 ep->e_flags = flags; 190 ep->e_error = 0; 191 mtx_lock(&sc->sc_events_mtx); 192 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 193 mtx_unlock(&sc->sc_events_mtx); 194 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 195 mtx_lock(&sc->sc_queue_mtx); 196 wakeup(sc); 197 mtx_unlock(&sc->sc_queue_mtx); 198 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 199 return (0); 200 sx_assert(&sc->sc_lock, SX_XLOCKED); 201 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 202 sx_xunlock(&sc->sc_lock); 203 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 204 mtx_lock(&sc->sc_events_mtx); 205 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 206 hz * 5); 207 } 208 error = ep->e_error; 209 g_mirror_event_free(ep); 210 sx_xlock(&sc->sc_lock); 211 return (error); 212} 213 214static struct g_mirror_event * 215g_mirror_event_get(struct g_mirror_softc *sc) 216{ 217 struct g_mirror_event *ep; 218 219 mtx_lock(&sc->sc_events_mtx); 220 ep = TAILQ_FIRST(&sc->sc_events); 221 mtx_unlock(&sc->sc_events_mtx); 222 return (ep); 223} 224 225static void 226g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 227{ 228 229 mtx_lock(&sc->sc_events_mtx); 230 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 231 mtx_unlock(&sc->sc_events_mtx); 232} 233 234static void 235g_mirror_event_cancel(struct g_mirror_disk *disk) 236{ 237 struct g_mirror_softc *sc; 238 struct g_mirror_event *ep, *tmpep; 239 240 sc = disk->d_softc; 241 sx_assert(&sc->sc_lock, SX_XLOCKED); 242 243 mtx_lock(&sc->sc_events_mtx); 244 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 245 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 246 continue; 247 if (ep->e_disk != disk) 248 continue; 249 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 250 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 251 g_mirror_event_free(ep); 252 else { 253 ep->e_error = ECANCELED; 254 wakeup(ep); 255 } 256 } 257 mtx_unlock(&sc->sc_events_mtx); 258} 259 260/* 261 * Return the number of disks in given state. 262 * If state is equal to -1, count all connected disks. 263 */ 264u_int 265g_mirror_ndisks(struct g_mirror_softc *sc, int state) 266{ 267 struct g_mirror_disk *disk; 268 u_int n = 0; 269 270 sx_assert(&sc->sc_lock, SX_LOCKED); 271 272 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 273 if (state == -1 || disk->d_state == state) 274 n++; 275 } 276 return (n); 277} 278 279/* 280 * Find a disk in mirror by its disk ID. 281 */ 282static struct g_mirror_disk * 283g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 284{ 285 struct g_mirror_disk *disk; 286 287 sx_assert(&sc->sc_lock, SX_XLOCKED); 288 289 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 290 if (disk->d_id == id) 291 return (disk); 292 } 293 return (NULL); 294} 295 296static u_int 297g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 298{ 299 struct bio *bp; 300 u_int nreqs = 0; 301 302 mtx_lock(&sc->sc_queue_mtx); 303 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 304 if (bp->bio_from == cp) 305 nreqs++; 306 } 307 mtx_unlock(&sc->sc_queue_mtx); 308 return (nreqs); 309} 310 311static int 312g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 313{ 314 315 if (cp->index > 0) { 316 G_MIRROR_DEBUG(2, 317 "I/O requests for %s exist, can't destroy it now.", 318 cp->provider->name); 319 return (1); 320 } 321 if (g_mirror_nrequests(sc, cp) > 0) { 322 G_MIRROR_DEBUG(2, 323 "I/O requests for %s in queue, can't destroy it now.", 324 cp->provider->name); 325 return (1); 326 } 327 return (0); 328} 329 330static void 331g_mirror_destroy_consumer(void *arg, int flags __unused) 332{ 333 struct g_consumer *cp; 334 335 g_topology_assert(); 336 337 cp = arg; 338 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 339 g_detach(cp); 340 g_destroy_consumer(cp); 341} 342 343static void 344g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 345{ 346 struct g_provider *pp; 347 int retaste_wait; 348 349 g_topology_assert(); 350 351 cp->private = NULL; 352 if (g_mirror_is_busy(sc, cp)) 353 return; 354 pp = cp->provider; 355 retaste_wait = 0; 356 if (cp->acw == 1) { 357 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 358 retaste_wait = 1; 359 } 360 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 361 -cp->acw, -cp->ace, 0); 362 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 363 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 364 if (retaste_wait) { 365 /* 366 * After retaste event was send (inside g_access()), we can send 367 * event to detach and destroy consumer. 368 * A class, which has consumer to the given provider connected 369 * will not receive retaste event for the provider. 370 * This is the way how I ignore retaste events when I close 371 * consumers opened for write: I detach and destroy consumer 372 * after retaste event is sent. 373 */ 374 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 375 return; 376 } 377 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 378 g_detach(cp); 379 g_destroy_consumer(cp); 380} 381 382static int 383g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 384{ 385 struct g_consumer *cp; 386 int error; 387 388 g_topology_assert_not(); 389 KASSERT(disk->d_consumer == NULL, 390 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 391 392 g_topology_lock(); 393 cp = g_new_consumer(disk->d_softc->sc_geom); 394 cp->flags |= G_CF_DIRECT_RECEIVE; 395 error = g_attach(cp, pp); 396 if (error != 0) { 397 g_destroy_consumer(cp); 398 g_topology_unlock(); 399 return (error); 400 } 401 error = g_access(cp, 1, 1, 1); 402 if (error != 0) { 403 g_detach(cp); 404 g_destroy_consumer(cp); 405 g_topology_unlock(); 406 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 407 pp->name, error); 408 return (error); 409 } 410 g_topology_unlock(); 411 disk->d_consumer = cp; 412 disk->d_consumer->private = disk; 413 disk->d_consumer->index = 0; 414 415 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 416 return (0); 417} 418 419static void 420g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 421{ 422 423 g_topology_assert(); 424 425 if (cp == NULL) 426 return; 427 if (cp->provider != NULL) 428 g_mirror_kill_consumer(sc, cp); 429 else 430 g_destroy_consumer(cp); 431} 432 433/* 434 * Initialize disk. This means allocate memory, create consumer, attach it 435 * to the provider and open access (r1w1e1) to it. 436 */ 437static struct g_mirror_disk * 438g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 439 struct g_mirror_metadata *md, int *errorp) 440{ 441 struct g_mirror_disk *disk; 442 int i, error; 443 444 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 445 if (disk == NULL) { 446 error = ENOMEM; 447 goto fail; 448 } 449 disk->d_softc = sc; 450 error = g_mirror_connect_disk(disk, pp); 451 if (error != 0) 452 goto fail; 453 disk->d_id = md->md_did; 454 disk->d_state = G_MIRROR_DISK_STATE_NONE; 455 disk->d_priority = md->md_priority; 456 disk->d_flags = md->md_dflags; 457 error = g_getattr("GEOM::candelete", disk->d_consumer, &i); 458 if (error == 0 && i != 0) 459 disk->d_flags |= G_MIRROR_DISK_FLAG_CANDELETE; 460 if (md->md_provider[0] != '\0') 461 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 462 disk->d_sync.ds_consumer = NULL; 463 disk->d_sync.ds_offset = md->md_sync_offset; 464 disk->d_sync.ds_offset_done = md->md_sync_offset; 465 disk->d_genid = md->md_genid; 466 disk->d_sync.ds_syncid = md->md_syncid; 467 if (errorp != NULL) 468 *errorp = 0; 469 return (disk); 470fail: 471 if (errorp != NULL) 472 *errorp = error; 473 if (disk != NULL) 474 free(disk, M_MIRROR); 475 return (NULL); 476} 477 478static void 479g_mirror_destroy_disk(struct g_mirror_disk *disk) 480{ 481 struct g_mirror_softc *sc; 482 483 g_topology_assert_not(); 484 sc = disk->d_softc; 485 sx_assert(&sc->sc_lock, SX_XLOCKED); 486 487 LIST_REMOVE(disk, d_next); 488 g_mirror_event_cancel(disk); 489 if (sc->sc_hint == disk) 490 sc->sc_hint = NULL; 491 switch (disk->d_state) { 492 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 493 g_mirror_sync_stop(disk, 1); 494 /* FALLTHROUGH */ 495 case G_MIRROR_DISK_STATE_NEW: 496 case G_MIRROR_DISK_STATE_STALE: 497 case G_MIRROR_DISK_STATE_ACTIVE: 498 g_topology_lock(); 499 g_mirror_disconnect_consumer(sc, disk->d_consumer); 500 g_topology_unlock(); 501 free(disk, M_MIRROR); 502 break; 503 default: 504 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 505 g_mirror_get_diskname(disk), 506 g_mirror_disk_state2str(disk->d_state))); 507 } 508} 509 510static void 511g_mirror_free_device(struct g_mirror_softc *sc) 512{ 513 514 mtx_destroy(&sc->sc_queue_mtx); 515 mtx_destroy(&sc->sc_events_mtx); 516 mtx_destroy(&sc->sc_done_mtx); 517 sx_destroy(&sc->sc_lock); 518 free(sc, M_MIRROR); 519} 520 521static void 522g_mirror_providergone(struct g_provider *pp) 523{ 524 struct g_mirror_softc *sc = pp->private; 525 526 if ((--sc->sc_refcnt) == 0) 527 g_mirror_free_device(sc); 528} 529 530static void 531g_mirror_destroy_device(struct g_mirror_softc *sc) 532{ 533 struct g_mirror_disk *disk; 534 struct g_mirror_event *ep; 535 struct g_geom *gp; 536 struct g_consumer *cp, *tmpcp; 537 538 g_topology_assert_not(); 539 sx_assert(&sc->sc_lock, SX_XLOCKED); 540 541 gp = sc->sc_geom; 542 if (sc->sc_provider != NULL) 543 g_mirror_destroy_provider(sc); 544 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 545 disk = LIST_FIRST(&sc->sc_disks)) { 546 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 547 g_mirror_update_metadata(disk); 548 g_mirror_destroy_disk(disk); 549 } 550 while ((ep = g_mirror_event_get(sc)) != NULL) { 551 g_mirror_event_remove(sc, ep); 552 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 553 g_mirror_event_free(ep); 554 else { 555 ep->e_error = ECANCELED; 556 ep->e_flags |= G_MIRROR_EVENT_DONE; 557 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 558 mtx_lock(&sc->sc_events_mtx); 559 wakeup(ep); 560 mtx_unlock(&sc->sc_events_mtx); 561 } 562 } 563 callout_drain(&sc->sc_callout); 564 565 g_topology_lock(); 566 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 567 g_mirror_disconnect_consumer(sc, cp); 568 } 569 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 570 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 571 g_wither_geom(gp, ENXIO); 572 sx_xunlock(&sc->sc_lock); 573 if ((--sc->sc_refcnt) == 0) 574 g_mirror_free_device(sc); 575 g_topology_unlock(); 576} 577 578static void 579g_mirror_orphan(struct g_consumer *cp) 580{ 581 struct g_mirror_disk *disk; 582 583 g_topology_assert(); 584 585 disk = cp->private; 586 if (disk == NULL) 587 return; 588 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 589 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 590 G_MIRROR_EVENT_DONTWAIT); 591} 592 593/* 594 * Function should return the next active disk on the list. 595 * It is possible that it will be the same disk as given. 596 * If there are no active disks on list, NULL is returned. 597 */ 598static __inline struct g_mirror_disk * 599g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 600{ 601 struct g_mirror_disk *dp; 602 603 for (dp = LIST_NEXT(disk, d_next); dp != disk; 604 dp = LIST_NEXT(dp, d_next)) { 605 if (dp == NULL) 606 dp = LIST_FIRST(&sc->sc_disks); 607 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 608 break; 609 } 610 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 611 return (NULL); 612 return (dp); 613} 614 615static struct g_mirror_disk * 616g_mirror_get_disk(struct g_mirror_softc *sc) 617{ 618 struct g_mirror_disk *disk; 619 620 if (sc->sc_hint == NULL) { 621 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 622 if (sc->sc_hint == NULL) 623 return (NULL); 624 } 625 disk = sc->sc_hint; 626 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 627 disk = g_mirror_find_next(sc, disk); 628 if (disk == NULL) 629 return (NULL); 630 } 631 sc->sc_hint = g_mirror_find_next(sc, disk); 632 return (disk); 633} 634 635static int 636g_mirror_write_metadata(struct g_mirror_disk *disk, 637 struct g_mirror_metadata *md) 638{ 639 struct g_mirror_softc *sc; 640 struct g_consumer *cp; 641 off_t offset, length; 642 u_char *sector; 643 int error = 0; 644 645 g_topology_assert_not(); 646 sc = disk->d_softc; 647 sx_assert(&sc->sc_lock, SX_LOCKED); 648 649 cp = disk->d_consumer; 650 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 651 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 652 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 653 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 654 cp->acw, cp->ace)); 655 length = cp->provider->sectorsize; 656 offset = cp->provider->mediasize - length; 657 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 658 if (md != NULL && 659 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) { 660 /* 661 * Handle the case, when the size of parent provider reduced. 662 */ 663 if (offset < md->md_mediasize) 664 error = ENOSPC; 665 else 666 mirror_metadata_encode(md, sector); 667 } 668 if (error == 0) 669 error = g_write_data(cp, offset, sector, length); 670 free(sector, M_MIRROR); 671 if (error != 0) { 672 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 673 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 674 G_MIRROR_DEBUG(0, "Cannot write metadata on %s " 675 "(device=%s, error=%d).", 676 g_mirror_get_diskname(disk), sc->sc_name, error); 677 } else { 678 G_MIRROR_DEBUG(1, "Cannot write metadata on %s " 679 "(device=%s, error=%d).", 680 g_mirror_get_diskname(disk), sc->sc_name, error); 681 } 682 if (g_mirror_disconnect_on_failure && 683 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 684 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 685 g_mirror_event_send(disk, 686 G_MIRROR_DISK_STATE_DISCONNECTED, 687 G_MIRROR_EVENT_DONTWAIT); 688 } 689 } 690 return (error); 691} 692 693static int 694g_mirror_clear_metadata(struct g_mirror_disk *disk) 695{ 696 int error; 697 698 g_topology_assert_not(); 699 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 700 701 error = g_mirror_write_metadata(disk, NULL); 702 if (error == 0) { 703 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 704 g_mirror_get_diskname(disk)); 705 } else { 706 G_MIRROR_DEBUG(0, 707 "Cannot clear metadata on disk %s (error=%d).", 708 g_mirror_get_diskname(disk), error); 709 } 710 return (error); 711} 712 713void 714g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 715 struct g_mirror_metadata *md) 716{ 717 718 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 719 md->md_version = G_MIRROR_VERSION; 720 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 721 md->md_mid = sc->sc_id; 722 md->md_all = sc->sc_ndisks; 723 md->md_slice = sc->sc_slice; 724 md->md_balance = sc->sc_balance; 725 md->md_genid = sc->sc_genid; 726 md->md_mediasize = sc->sc_mediasize; 727 md->md_sectorsize = sc->sc_sectorsize; 728 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 729 bzero(md->md_provider, sizeof(md->md_provider)); 730 if (disk == NULL) { 731 md->md_did = arc4random(); 732 md->md_priority = 0; 733 md->md_syncid = 0; 734 md->md_dflags = 0; 735 md->md_sync_offset = 0; 736 md->md_provsize = 0; 737 } else { 738 md->md_did = disk->d_id; 739 md->md_priority = disk->d_priority; 740 md->md_syncid = disk->d_sync.ds_syncid; 741 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 742 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 743 md->md_sync_offset = disk->d_sync.ds_offset_done; 744 else 745 md->md_sync_offset = 0; 746 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 747 strlcpy(md->md_provider, 748 disk->d_consumer->provider->name, 749 sizeof(md->md_provider)); 750 } 751 md->md_provsize = disk->d_consumer->provider->mediasize; 752 } 753} 754 755void 756g_mirror_update_metadata(struct g_mirror_disk *disk) 757{ 758 struct g_mirror_softc *sc; 759 struct g_mirror_metadata md; 760 int error; 761 762 g_topology_assert_not(); 763 sc = disk->d_softc; 764 sx_assert(&sc->sc_lock, SX_LOCKED); 765 766 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) 767 g_mirror_fill_metadata(sc, disk, &md); 768 error = g_mirror_write_metadata(disk, &md); 769 if (error == 0) { 770 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 771 g_mirror_get_diskname(disk)); 772 } else { 773 G_MIRROR_DEBUG(0, 774 "Cannot update metadata on disk %s (error=%d).", 775 g_mirror_get_diskname(disk), error); 776 } 777} 778 779static void 780g_mirror_bump_syncid(struct g_mirror_softc *sc) 781{ 782 struct g_mirror_disk *disk; 783 784 g_topology_assert_not(); 785 sx_assert(&sc->sc_lock, SX_XLOCKED); 786 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 787 ("%s called with no active disks (device=%s).", __func__, 788 sc->sc_name)); 789 790 sc->sc_syncid++; 791 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 792 sc->sc_syncid); 793 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 794 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 795 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 796 disk->d_sync.ds_syncid = sc->sc_syncid; 797 g_mirror_update_metadata(disk); 798 } 799 } 800} 801 802static void 803g_mirror_bump_genid(struct g_mirror_softc *sc) 804{ 805 struct g_mirror_disk *disk; 806 807 g_topology_assert_not(); 808 sx_assert(&sc->sc_lock, SX_XLOCKED); 809 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 810 ("%s called with no active disks (device=%s).", __func__, 811 sc->sc_name)); 812 813 sc->sc_genid++; 814 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 815 sc->sc_genid); 816 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 817 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 818 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 819 disk->d_genid = sc->sc_genid; 820 g_mirror_update_metadata(disk); 821 } 822 } 823} 824 825static int 826g_mirror_idle(struct g_mirror_softc *sc, int acw) 827{ 828 struct g_mirror_disk *disk; 829 int timeout; 830 831 g_topology_assert_not(); 832 sx_assert(&sc->sc_lock, SX_XLOCKED); 833 834 if (sc->sc_provider == NULL) 835 return (0); 836 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 837 return (0); 838 if (sc->sc_idle) 839 return (0); 840 if (sc->sc_writes > 0) 841 return (0); 842 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 843 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); 844 if (!g_mirror_shutdown && timeout > 0) 845 return (timeout); 846 } 847 sc->sc_idle = 1; 848 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 849 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 850 continue; 851 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 852 g_mirror_get_diskname(disk), sc->sc_name); 853 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 854 g_mirror_update_metadata(disk); 855 } 856 return (0); 857} 858 859static void 860g_mirror_unidle(struct g_mirror_softc *sc) 861{ 862 struct g_mirror_disk *disk; 863 864 g_topology_assert_not(); 865 sx_assert(&sc->sc_lock, SX_XLOCKED); 866 867 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 868 return; 869 sc->sc_idle = 0; 870 sc->sc_last_write = time_uptime; 871 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 872 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 873 continue; 874 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 875 g_mirror_get_diskname(disk), sc->sc_name); 876 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 877 g_mirror_update_metadata(disk); 878 } 879} 880 881static void 882g_mirror_flush_done(struct bio *bp) 883{ 884 struct g_mirror_softc *sc; 885 struct bio *pbp; 886 887 pbp = bp->bio_parent; 888 sc = pbp->bio_to->private; 889 mtx_lock(&sc->sc_done_mtx); 890 if (pbp->bio_error == 0) 891 pbp->bio_error = bp->bio_error; 892 pbp->bio_completed += bp->bio_completed; 893 pbp->bio_inbed++; 894 if (pbp->bio_children == pbp->bio_inbed) { 895 mtx_unlock(&sc->sc_done_mtx); 896 g_io_deliver(pbp, pbp->bio_error); 897 } else 898 mtx_unlock(&sc->sc_done_mtx); 899 g_destroy_bio(bp); 900} 901 902static void 903g_mirror_done(struct bio *bp) 904{ 905 struct g_mirror_softc *sc; 906 907 sc = bp->bio_from->geom->softc; 908 bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR; 909 mtx_lock(&sc->sc_queue_mtx); 910 bioq_insert_tail(&sc->sc_queue, bp); 911 mtx_unlock(&sc->sc_queue_mtx); 912 wakeup(sc); 913} 914 915static void 916g_mirror_regular_request(struct bio *bp) 917{ 918 struct g_mirror_softc *sc; 919 struct g_mirror_disk *disk; 920 struct bio *pbp; 921 922 g_topology_assert_not(); 923 924 pbp = bp->bio_parent; 925 sc = pbp->bio_to->private; 926 bp->bio_from->index--; 927 if (bp->bio_cmd == BIO_WRITE) 928 sc->sc_writes--; 929 disk = bp->bio_from->private; 930 if (disk == NULL) { 931 g_topology_lock(); 932 g_mirror_kill_consumer(sc, bp->bio_from); 933 g_topology_unlock(); 934 } 935 936 pbp->bio_inbed++; 937 KASSERT(pbp->bio_inbed <= pbp->bio_children, 938 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 939 pbp->bio_children)); 940 if (bp->bio_error == 0 && pbp->bio_error == 0) { 941 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 942 g_destroy_bio(bp); 943 if (pbp->bio_children == pbp->bio_inbed) { 944 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 945 pbp->bio_completed = pbp->bio_length; 946 if (pbp->bio_cmd == BIO_WRITE || 947 pbp->bio_cmd == BIO_DELETE) { 948 bioq_remove(&sc->sc_inflight, pbp); 949 /* Release delayed sync requests if possible. */ 950 g_mirror_sync_release(sc); 951 } 952 g_io_deliver(pbp, pbp->bio_error); 953 } 954 return; 955 } else if (bp->bio_error != 0) { 956 if (pbp->bio_error == 0) 957 pbp->bio_error = bp->bio_error; 958 if (disk != NULL) { 959 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 960 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 961 G_MIRROR_LOGREQ(0, bp, 962 "Request failed (error=%d).", 963 bp->bio_error); 964 } else { 965 G_MIRROR_LOGREQ(1, bp, 966 "Request failed (error=%d).", 967 bp->bio_error); 968 } 969 if (g_mirror_disconnect_on_failure && 970 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) 971 { 972 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 973 g_mirror_event_send(disk, 974 G_MIRROR_DISK_STATE_DISCONNECTED, 975 G_MIRROR_EVENT_DONTWAIT); 976 } 977 } 978 switch (pbp->bio_cmd) { 979 case BIO_DELETE: 980 case BIO_WRITE: 981 pbp->bio_inbed--; 982 pbp->bio_children--; 983 break; 984 } 985 } 986 g_destroy_bio(bp); 987 988 switch (pbp->bio_cmd) { 989 case BIO_READ: 990 if (pbp->bio_inbed < pbp->bio_children) 991 break; 992 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 993 g_io_deliver(pbp, pbp->bio_error); 994 else { 995 pbp->bio_error = 0; 996 mtx_lock(&sc->sc_queue_mtx); 997 bioq_insert_tail(&sc->sc_queue, pbp); 998 mtx_unlock(&sc->sc_queue_mtx); 999 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1000 wakeup(sc); 1001 } 1002 break; 1003 case BIO_DELETE: 1004 case BIO_WRITE: 1005 if (pbp->bio_children == 0) { 1006 /* 1007 * All requests failed. 1008 */ 1009 } else if (pbp->bio_inbed < pbp->bio_children) { 1010 /* Do nothing. */ 1011 break; 1012 } else if (pbp->bio_children == pbp->bio_inbed) { 1013 /* Some requests succeeded. */ 1014 pbp->bio_error = 0; 1015 pbp->bio_completed = pbp->bio_length; 1016 } 1017 bioq_remove(&sc->sc_inflight, pbp); 1018 /* Release delayed sync requests if possible. */ 1019 g_mirror_sync_release(sc); 1020 g_io_deliver(pbp, pbp->bio_error); 1021 break; 1022 default: 1023 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 1024 break; 1025 } 1026} 1027 1028static void 1029g_mirror_sync_done(struct bio *bp) 1030{ 1031 struct g_mirror_softc *sc; 1032 1033 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1034 sc = bp->bio_from->geom->softc; 1035 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; 1036 mtx_lock(&sc->sc_queue_mtx); 1037 bioq_insert_tail(&sc->sc_queue, bp); 1038 mtx_unlock(&sc->sc_queue_mtx); 1039 wakeup(sc); 1040} 1041 1042static void 1043g_mirror_candelete(struct bio *bp) 1044{ 1045 struct g_mirror_softc *sc; 1046 struct g_mirror_disk *disk; 1047 int *val; 1048 1049 sc = bp->bio_to->private; 1050 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1051 if (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) 1052 break; 1053 } 1054 val = (int *)bp->bio_data; 1055 *val = (disk != NULL); 1056 g_io_deliver(bp, 0); 1057} 1058 1059static void 1060g_mirror_kernel_dump(struct bio *bp) 1061{ 1062 struct g_mirror_softc *sc; 1063 struct g_mirror_disk *disk; 1064 struct bio *cbp; 1065 struct g_kerneldump *gkd; 1066 1067 /* 1068 * We configure dumping to the first component, because this component 1069 * will be used for reading with 'prefer' balance algorithm. 1070 * If the component with the highest priority is currently disconnected 1071 * we will not be able to read the dump after the reboot if it will be 1072 * connected and synchronized later. Can we do something better? 1073 */ 1074 sc = bp->bio_to->private; 1075 disk = LIST_FIRST(&sc->sc_disks); 1076 1077 gkd = (struct g_kerneldump *)bp->bio_data; 1078 if (gkd->length > bp->bio_to->mediasize) 1079 gkd->length = bp->bio_to->mediasize; 1080 cbp = g_clone_bio(bp); 1081 if (cbp == NULL) { 1082 g_io_deliver(bp, ENOMEM); 1083 return; 1084 } 1085 cbp->bio_done = g_std_done; 1086 g_io_request(cbp, disk->d_consumer); 1087 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1088 g_mirror_get_diskname(disk)); 1089} 1090 1091static void 1092g_mirror_flush(struct g_mirror_softc *sc, struct bio *bp) 1093{ 1094 struct bio_queue_head queue; 1095 struct g_mirror_disk *disk; 1096 struct g_consumer *cp; 1097 struct bio *cbp; 1098 1099 bioq_init(&queue); 1100 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1101 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1102 continue; 1103 cbp = g_clone_bio(bp); 1104 if (cbp == NULL) { 1105 while ((cbp = bioq_takefirst(&queue)) != NULL) 1106 g_destroy_bio(cbp); 1107 if (bp->bio_error == 0) 1108 bp->bio_error = ENOMEM; 1109 g_io_deliver(bp, bp->bio_error); 1110 return; 1111 } 1112 bioq_insert_tail(&queue, cbp); 1113 cbp->bio_done = g_mirror_flush_done; 1114 cbp->bio_caller1 = disk; 1115 cbp->bio_to = disk->d_consumer->provider; 1116 } 1117 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1118 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1119 disk = cbp->bio_caller1; 1120 cbp->bio_caller1 = NULL; 1121 cp = disk->d_consumer; 1122 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1123 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1124 cp->acr, cp->acw, cp->ace)); 1125 g_io_request(cbp, disk->d_consumer); 1126 } 1127} 1128 1129static void 1130g_mirror_start(struct bio *bp) 1131{ 1132 struct g_mirror_softc *sc; 1133 1134 sc = bp->bio_to->private; 1135 /* 1136 * If sc == NULL or there are no valid disks, provider's error 1137 * should be set and g_mirror_start() should not be called at all. 1138 */ 1139 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1140 ("Provider's error should be set (error=%d)(mirror=%s).", 1141 bp->bio_to->error, bp->bio_to->name)); 1142 G_MIRROR_LOGREQ(3, bp, "Request received."); 1143 1144 switch (bp->bio_cmd) { 1145 case BIO_READ: 1146 case BIO_WRITE: 1147 case BIO_DELETE: 1148 break; 1149 case BIO_FLUSH: 1150 g_mirror_flush(sc, bp); 1151 return; 1152 case BIO_GETATTR: 1153 if (!strcmp(bp->bio_attribute, "GEOM::candelete")) { 1154 g_mirror_candelete(bp); 1155 return; 1156 } else if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1157 g_mirror_kernel_dump(bp); 1158 return; 1159 } 1160 /* FALLTHROUGH */ 1161 default: 1162 g_io_deliver(bp, EOPNOTSUPP); 1163 return; 1164 } 1165 mtx_lock(&sc->sc_queue_mtx); 1166 bioq_insert_tail(&sc->sc_queue, bp); 1167 mtx_unlock(&sc->sc_queue_mtx); 1168 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1169 wakeup(sc); 1170} 1171 1172/* 1173 * Return TRUE if the given request is colliding with a in-progress 1174 * synchronization request. 1175 */ 1176static int 1177g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1178{ 1179 struct g_mirror_disk *disk; 1180 struct bio *sbp; 1181 off_t rstart, rend, sstart, send; 1182 u_int i; 1183 1184 if (sc->sc_sync.ds_ndisks == 0) 1185 return (0); 1186 rstart = bp->bio_offset; 1187 rend = bp->bio_offset + bp->bio_length; 1188 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1189 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1190 continue; 1191 for (i = 0; i < g_mirror_syncreqs; i++) { 1192 sbp = disk->d_sync.ds_bios[i]; 1193 if (sbp == NULL) 1194 continue; 1195 sstart = sbp->bio_offset; 1196 send = sbp->bio_offset + sbp->bio_length; 1197 if (rend > sstart && rstart < send) 1198 return (1); 1199 } 1200 } 1201 return (0); 1202} 1203 1204/* 1205 * Return TRUE if the given sync request is colliding with a in-progress regular 1206 * request. 1207 */ 1208static int 1209g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1210{ 1211 off_t rstart, rend, sstart, send; 1212 struct bio *bp; 1213 1214 if (sc->sc_sync.ds_ndisks == 0) 1215 return (0); 1216 sstart = sbp->bio_offset; 1217 send = sbp->bio_offset + sbp->bio_length; 1218 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { 1219 rstart = bp->bio_offset; 1220 rend = bp->bio_offset + bp->bio_length; 1221 if (rend > sstart && rstart < send) 1222 return (1); 1223 } 1224 return (0); 1225} 1226 1227/* 1228 * Puts request onto delayed queue. 1229 */ 1230static void 1231g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1232{ 1233 1234 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1235 bioq_insert_head(&sc->sc_regular_delayed, bp); 1236} 1237 1238/* 1239 * Puts synchronization request onto delayed queue. 1240 */ 1241static void 1242g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1243{ 1244 1245 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1246 bioq_insert_tail(&sc->sc_sync_delayed, bp); 1247} 1248 1249/* 1250 * Releases delayed regular requests which don't collide anymore with sync 1251 * requests. 1252 */ 1253static void 1254g_mirror_regular_release(struct g_mirror_softc *sc) 1255{ 1256 struct bio *bp, *bp2; 1257 1258 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { 1259 if (g_mirror_sync_collision(sc, bp)) 1260 continue; 1261 bioq_remove(&sc->sc_regular_delayed, bp); 1262 G_MIRROR_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1263 mtx_lock(&sc->sc_queue_mtx); 1264 bioq_insert_head(&sc->sc_queue, bp); 1265#if 0 1266 /* 1267 * wakeup() is not needed, because this function is called from 1268 * the worker thread. 1269 */ 1270 wakeup(&sc->sc_queue); 1271#endif 1272 mtx_unlock(&sc->sc_queue_mtx); 1273 } 1274} 1275 1276/* 1277 * Releases delayed sync requests which don't collide anymore with regular 1278 * requests. 1279 */ 1280static void 1281g_mirror_sync_release(struct g_mirror_softc *sc) 1282{ 1283 struct bio *bp, *bp2; 1284 1285 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { 1286 if (g_mirror_regular_collision(sc, bp)) 1287 continue; 1288 bioq_remove(&sc->sc_sync_delayed, bp); 1289 G_MIRROR_LOGREQ(2, bp, 1290 "Releasing delayed synchronization request."); 1291 g_io_request(bp, bp->bio_from); 1292 } 1293} 1294 1295/* 1296 * Handle synchronization requests. 1297 * Every synchronization request is two-steps process: first, READ request is 1298 * send to active provider and then WRITE request (with read data) to the provider 1299 * being synchronized. When WRITE is finished, new synchronization request is 1300 * send. 1301 */ 1302static void 1303g_mirror_sync_request(struct bio *bp) 1304{ 1305 struct g_mirror_softc *sc; 1306 struct g_mirror_disk *disk; 1307 1308 bp->bio_from->index--; 1309 sc = bp->bio_from->geom->softc; 1310 disk = bp->bio_from->private; 1311 if (disk == NULL) { 1312 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1313 g_topology_lock(); 1314 g_mirror_kill_consumer(sc, bp->bio_from); 1315 g_topology_unlock(); 1316 free(bp->bio_data, M_MIRROR); 1317 g_destroy_bio(bp); 1318 sx_xlock(&sc->sc_lock); 1319 return; 1320 } 1321 1322 /* 1323 * Synchronization request. 1324 */ 1325 switch (bp->bio_cmd) { 1326 case BIO_READ: 1327 { 1328 struct g_consumer *cp; 1329 1330 if (bp->bio_error != 0) { 1331 G_MIRROR_LOGREQ(0, bp, 1332 "Synchronization request failed (error=%d).", 1333 bp->bio_error); 1334 g_destroy_bio(bp); 1335 return; 1336 } 1337 G_MIRROR_LOGREQ(3, bp, 1338 "Synchronization request half-finished."); 1339 bp->bio_cmd = BIO_WRITE; 1340 bp->bio_cflags = 0; 1341 cp = disk->d_consumer; 1342 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1343 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1344 cp->acr, cp->acw, cp->ace)); 1345 cp->index++; 1346 g_io_request(bp, cp); 1347 return; 1348 } 1349 case BIO_WRITE: 1350 { 1351 struct g_mirror_disk_sync *sync; 1352 off_t offset; 1353 void *data; 1354 int i; 1355 1356 if (bp->bio_error != 0) { 1357 G_MIRROR_LOGREQ(0, bp, 1358 "Synchronization request failed (error=%d).", 1359 bp->bio_error); 1360 g_destroy_bio(bp); 1361 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1362 g_mirror_event_send(disk, 1363 G_MIRROR_DISK_STATE_DISCONNECTED, 1364 G_MIRROR_EVENT_DONTWAIT); 1365 return; 1366 } 1367 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1368 sync = &disk->d_sync; 1369 if (sync->ds_offset >= sc->sc_mediasize || 1370 sync->ds_consumer == NULL || 1371 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1372 /* Don't send more synchronization requests. */ 1373 sync->ds_inflight--; 1374 if (sync->ds_bios != NULL) { 1375 i = (int)(uintptr_t)bp->bio_caller1; 1376 sync->ds_bios[i] = NULL; 1377 } 1378 free(bp->bio_data, M_MIRROR); 1379 g_destroy_bio(bp); 1380 if (sync->ds_inflight > 0) 1381 return; 1382 if (sync->ds_consumer == NULL || 1383 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1384 return; 1385 } 1386 /* Disk up-to-date, activate it. */ 1387 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1388 G_MIRROR_EVENT_DONTWAIT); 1389 return; 1390 } 1391 1392 /* Send next synchronization request. */ 1393 data = bp->bio_data; 1394 g_reset_bio(bp); 1395 bp->bio_cmd = BIO_READ; 1396 bp->bio_offset = sync->ds_offset; 1397 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1398 sync->ds_offset += bp->bio_length; 1399 bp->bio_done = g_mirror_sync_done; 1400 bp->bio_data = data; 1401 bp->bio_from = sync->ds_consumer; 1402 bp->bio_to = sc->sc_provider; 1403 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1404 sync->ds_consumer->index++; 1405 /* 1406 * Delay the request if it is colliding with a regular request. 1407 */ 1408 if (g_mirror_regular_collision(sc, bp)) 1409 g_mirror_sync_delay(sc, bp); 1410 else 1411 g_io_request(bp, sync->ds_consumer); 1412 1413 /* Release delayed requests if possible. */ 1414 g_mirror_regular_release(sc); 1415 1416 /* Find the smallest offset */ 1417 offset = sc->sc_mediasize; 1418 for (i = 0; i < g_mirror_syncreqs; i++) { 1419 bp = sync->ds_bios[i]; 1420 if (bp->bio_offset < offset) 1421 offset = bp->bio_offset; 1422 } 1423 if (sync->ds_offset_done + (MAXPHYS * 100) < offset) { 1424 /* Update offset_done on every 100 blocks. */ 1425 sync->ds_offset_done = offset; 1426 g_mirror_update_metadata(disk); 1427 } 1428 return; 1429 } 1430 default: 1431 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1432 bp->bio_cmd, sc->sc_name)); 1433 break; 1434 } 1435} 1436 1437static void 1438g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1439{ 1440 struct g_mirror_disk *disk; 1441 struct g_consumer *cp; 1442 struct bio *cbp; 1443 1444 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1445 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1446 break; 1447 } 1448 if (disk == NULL) { 1449 if (bp->bio_error == 0) 1450 bp->bio_error = ENXIO; 1451 g_io_deliver(bp, bp->bio_error); 1452 return; 1453 } 1454 cbp = g_clone_bio(bp); 1455 if (cbp == NULL) { 1456 if (bp->bio_error == 0) 1457 bp->bio_error = ENOMEM; 1458 g_io_deliver(bp, bp->bio_error); 1459 return; 1460 } 1461 /* 1462 * Fill in the component buf structure. 1463 */ 1464 cp = disk->d_consumer; 1465 cbp->bio_done = g_mirror_done; 1466 cbp->bio_to = cp->provider; 1467 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1468 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1469 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1470 cp->acw, cp->ace)); 1471 cp->index++; 1472 g_io_request(cbp, cp); 1473} 1474 1475static void 1476g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1477{ 1478 struct g_mirror_disk *disk; 1479 struct g_consumer *cp; 1480 struct bio *cbp; 1481 1482 disk = g_mirror_get_disk(sc); 1483 if (disk == NULL) { 1484 if (bp->bio_error == 0) 1485 bp->bio_error = ENXIO; 1486 g_io_deliver(bp, bp->bio_error); 1487 return; 1488 } 1489 cbp = g_clone_bio(bp); 1490 if (cbp == NULL) { 1491 if (bp->bio_error == 0) 1492 bp->bio_error = ENOMEM; 1493 g_io_deliver(bp, bp->bio_error); 1494 return; 1495 } 1496 /* 1497 * Fill in the component buf structure. 1498 */ 1499 cp = disk->d_consumer; 1500 cbp->bio_done = g_mirror_done; 1501 cbp->bio_to = cp->provider; 1502 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1503 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1504 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1505 cp->acw, cp->ace)); 1506 cp->index++; 1507 g_io_request(cbp, cp); 1508} 1509 1510#define TRACK_SIZE (1 * 1024 * 1024) 1511#define LOAD_SCALE 256 1512#define ABS(x) (((x) >= 0) ? (x) : (-(x))) 1513 1514static void 1515g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1516{ 1517 struct g_mirror_disk *disk, *dp; 1518 struct g_consumer *cp; 1519 struct bio *cbp; 1520 int prio, best; 1521 1522 /* Find a disk with the smallest load. */ 1523 disk = NULL; 1524 best = INT_MAX; 1525 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1526 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1527 continue; 1528 prio = dp->load; 1529 /* If disk head is precisely in position - highly prefer it. */ 1530 if (dp->d_last_offset == bp->bio_offset) 1531 prio -= 2 * LOAD_SCALE; 1532 else 1533 /* If disk head is close to position - prefer it. */ 1534 if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE) 1535 prio -= 1 * LOAD_SCALE; 1536 if (prio <= best) { 1537 disk = dp; 1538 best = prio; 1539 } 1540 } 1541 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1542 cbp = g_clone_bio(bp); 1543 if (cbp == NULL) { 1544 if (bp->bio_error == 0) 1545 bp->bio_error = ENOMEM; 1546 g_io_deliver(bp, bp->bio_error); 1547 return; 1548 } 1549 /* 1550 * Fill in the component buf structure. 1551 */ 1552 cp = disk->d_consumer; 1553 cbp->bio_done = g_mirror_done; 1554 cbp->bio_to = cp->provider; 1555 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1556 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1557 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1558 cp->acw, cp->ace)); 1559 cp->index++; 1560 /* Remember last head position */ 1561 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1562 /* Update loads. */ 1563 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1564 dp->load = (dp->d_consumer->index * LOAD_SCALE + 1565 dp->load * 7) / 8; 1566 } 1567 g_io_request(cbp, cp); 1568} 1569 1570static void 1571g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1572{ 1573 struct bio_queue_head queue; 1574 struct g_mirror_disk *disk; 1575 struct g_consumer *cp; 1576 struct bio *cbp; 1577 off_t left, mod, offset, slice; 1578 u_char *data; 1579 u_int ndisks; 1580 1581 if (bp->bio_length <= sc->sc_slice) { 1582 g_mirror_request_round_robin(sc, bp); 1583 return; 1584 } 1585 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1586 slice = bp->bio_length / ndisks; 1587 mod = slice % sc->sc_provider->sectorsize; 1588 if (mod != 0) 1589 slice += sc->sc_provider->sectorsize - mod; 1590 /* 1591 * Allocate all bios before sending any request, so we can 1592 * return ENOMEM in nice and clean way. 1593 */ 1594 left = bp->bio_length; 1595 offset = bp->bio_offset; 1596 data = bp->bio_data; 1597 bioq_init(&queue); 1598 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1599 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1600 continue; 1601 cbp = g_clone_bio(bp); 1602 if (cbp == NULL) { 1603 while ((cbp = bioq_takefirst(&queue)) != NULL) 1604 g_destroy_bio(cbp); 1605 if (bp->bio_error == 0) 1606 bp->bio_error = ENOMEM; 1607 g_io_deliver(bp, bp->bio_error); 1608 return; 1609 } 1610 bioq_insert_tail(&queue, cbp); 1611 cbp->bio_done = g_mirror_done; 1612 cbp->bio_caller1 = disk; 1613 cbp->bio_to = disk->d_consumer->provider; 1614 cbp->bio_offset = offset; 1615 cbp->bio_data = data; 1616 cbp->bio_length = MIN(left, slice); 1617 left -= cbp->bio_length; 1618 if (left == 0) 1619 break; 1620 offset += cbp->bio_length; 1621 data += cbp->bio_length; 1622 } 1623 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1624 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1625 disk = cbp->bio_caller1; 1626 cbp->bio_caller1 = NULL; 1627 cp = disk->d_consumer; 1628 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1629 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1630 cp->acr, cp->acw, cp->ace)); 1631 disk->d_consumer->index++; 1632 g_io_request(cbp, disk->d_consumer); 1633 } 1634} 1635 1636static void 1637g_mirror_register_request(struct bio *bp) 1638{ 1639 struct g_mirror_softc *sc; 1640 1641 sc = bp->bio_to->private; 1642 switch (bp->bio_cmd) { 1643 case BIO_READ: 1644 switch (sc->sc_balance) { 1645 case G_MIRROR_BALANCE_LOAD: 1646 g_mirror_request_load(sc, bp); 1647 break; 1648 case G_MIRROR_BALANCE_PREFER: 1649 g_mirror_request_prefer(sc, bp); 1650 break; 1651 case G_MIRROR_BALANCE_ROUND_ROBIN: 1652 g_mirror_request_round_robin(sc, bp); 1653 break; 1654 case G_MIRROR_BALANCE_SPLIT: 1655 g_mirror_request_split(sc, bp); 1656 break; 1657 } 1658 return; 1659 case BIO_WRITE: 1660 case BIO_DELETE: 1661 { 1662 struct g_mirror_disk *disk; 1663 struct g_mirror_disk_sync *sync; 1664 struct bio_queue_head queue; 1665 struct g_consumer *cp; 1666 struct bio *cbp; 1667 1668 /* 1669 * Delay the request if it is colliding with a synchronization 1670 * request. 1671 */ 1672 if (g_mirror_sync_collision(sc, bp)) { 1673 g_mirror_regular_delay(sc, bp); 1674 return; 1675 } 1676 1677 if (sc->sc_idle) 1678 g_mirror_unidle(sc); 1679 else 1680 sc->sc_last_write = time_uptime; 1681 1682 /* 1683 * Allocate all bios before sending any request, so we can 1684 * return ENOMEM in nice and clean way. 1685 */ 1686 bioq_init(&queue); 1687 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1688 sync = &disk->d_sync; 1689 switch (disk->d_state) { 1690 case G_MIRROR_DISK_STATE_ACTIVE: 1691 break; 1692 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1693 if (bp->bio_offset >= sync->ds_offset) 1694 continue; 1695 break; 1696 default: 1697 continue; 1698 } 1699 if (bp->bio_cmd == BIO_DELETE && 1700 (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0) 1701 continue; 1702 cbp = g_clone_bio(bp); 1703 if (cbp == NULL) { 1704 while ((cbp = bioq_takefirst(&queue)) != NULL) 1705 g_destroy_bio(cbp); 1706 if (bp->bio_error == 0) 1707 bp->bio_error = ENOMEM; 1708 g_io_deliver(bp, bp->bio_error); 1709 return; 1710 } 1711 bioq_insert_tail(&queue, cbp); 1712 cbp->bio_done = g_mirror_done; 1713 cp = disk->d_consumer; 1714 cbp->bio_caller1 = cp; 1715 cbp->bio_to = cp->provider; 1716 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1717 ("Consumer %s not opened (r%dw%de%d).", 1718 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1719 } 1720 if (bioq_first(&queue) == NULL) { 1721 g_io_deliver(bp, EOPNOTSUPP); 1722 return; 1723 } 1724 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1725 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1726 cp = cbp->bio_caller1; 1727 cbp->bio_caller1 = NULL; 1728 cp->index++; 1729 sc->sc_writes++; 1730 g_io_request(cbp, cp); 1731 } 1732 /* 1733 * Put request onto inflight queue, so we can check if new 1734 * synchronization requests don't collide with it. 1735 */ 1736 bioq_insert_tail(&sc->sc_inflight, bp); 1737 /* 1738 * Bump syncid on first write. 1739 */ 1740 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1741 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1742 g_mirror_bump_syncid(sc); 1743 } 1744 return; 1745 } 1746 default: 1747 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1748 bp->bio_cmd, sc->sc_name)); 1749 break; 1750 } 1751} 1752 1753static int 1754g_mirror_can_destroy(struct g_mirror_softc *sc) 1755{ 1756 struct g_geom *gp; 1757 struct g_consumer *cp; 1758 1759 g_topology_assert(); 1760 gp = sc->sc_geom; 1761 if (gp->softc == NULL) 1762 return (1); 1763 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0) 1764 return (0); 1765 LIST_FOREACH(cp, &gp->consumer, consumer) { 1766 if (g_mirror_is_busy(sc, cp)) 1767 return (0); 1768 } 1769 gp = sc->sc_sync.ds_geom; 1770 LIST_FOREACH(cp, &gp->consumer, consumer) { 1771 if (g_mirror_is_busy(sc, cp)) 1772 return (0); 1773 } 1774 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1775 sc->sc_name); 1776 return (1); 1777} 1778 1779static int 1780g_mirror_try_destroy(struct g_mirror_softc *sc) 1781{ 1782 1783 if (sc->sc_rootmount != NULL) { 1784 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1785 sc->sc_rootmount); 1786 root_mount_rel(sc->sc_rootmount); 1787 sc->sc_rootmount = NULL; 1788 } 1789 g_topology_lock(); 1790 if (!g_mirror_can_destroy(sc)) { 1791 g_topology_unlock(); 1792 return (0); 1793 } 1794 sc->sc_geom->softc = NULL; 1795 sc->sc_sync.ds_geom->softc = NULL; 1796 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1797 g_topology_unlock(); 1798 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1799 &sc->sc_worker); 1800 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1801 sx_xunlock(&sc->sc_lock); 1802 wakeup(&sc->sc_worker); 1803 sc->sc_worker = NULL; 1804 } else { 1805 g_topology_unlock(); 1806 g_mirror_destroy_device(sc); 1807 } 1808 return (1); 1809} 1810 1811/* 1812 * Worker thread. 1813 */ 1814static void 1815g_mirror_worker(void *arg) 1816{ 1817 struct g_mirror_softc *sc; 1818 struct g_mirror_event *ep; 1819 struct bio *bp; 1820 int timeout; 1821 1822 sc = arg; 1823 thread_lock(curthread); 1824 sched_prio(curthread, PRIBIO); 1825 thread_unlock(curthread); 1826 1827 sx_xlock(&sc->sc_lock); 1828 for (;;) { 1829 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1830 /* 1831 * First take a look at events. 1832 * This is important to handle events before any I/O requests. 1833 */ 1834 ep = g_mirror_event_get(sc); 1835 if (ep != NULL) { 1836 g_mirror_event_remove(sc, ep); 1837 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1838 /* Update only device status. */ 1839 G_MIRROR_DEBUG(3, 1840 "Running event for device %s.", 1841 sc->sc_name); 1842 ep->e_error = 0; 1843 g_mirror_update_device(sc, true); 1844 } else { 1845 /* Update disk status. */ 1846 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1847 g_mirror_get_diskname(ep->e_disk)); 1848 ep->e_error = g_mirror_update_disk(ep->e_disk, 1849 ep->e_state); 1850 if (ep->e_error == 0) 1851 g_mirror_update_device(sc, false); 1852 } 1853 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1854 KASSERT(ep->e_error == 0, 1855 ("Error cannot be handled.")); 1856 g_mirror_event_free(ep); 1857 } else { 1858 ep->e_flags |= G_MIRROR_EVENT_DONE; 1859 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1860 ep); 1861 mtx_lock(&sc->sc_events_mtx); 1862 wakeup(ep); 1863 mtx_unlock(&sc->sc_events_mtx); 1864 } 1865 if ((sc->sc_flags & 1866 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1867 if (g_mirror_try_destroy(sc)) { 1868 curthread->td_pflags &= ~TDP_GEOM; 1869 G_MIRROR_DEBUG(1, "Thread exiting."); 1870 kproc_exit(0); 1871 } 1872 } 1873 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1874 continue; 1875 } 1876 /* 1877 * Check if we can mark array as CLEAN and if we can't take 1878 * how much seconds should we wait. 1879 */ 1880 timeout = g_mirror_idle(sc, -1); 1881 /* 1882 * Now I/O requests. 1883 */ 1884 /* Get first request from the queue. */ 1885 mtx_lock(&sc->sc_queue_mtx); 1886 bp = bioq_takefirst(&sc->sc_queue); 1887 if (bp == NULL) { 1888 if ((sc->sc_flags & 1889 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1890 mtx_unlock(&sc->sc_queue_mtx); 1891 if (g_mirror_try_destroy(sc)) { 1892 curthread->td_pflags &= ~TDP_GEOM; 1893 G_MIRROR_DEBUG(1, "Thread exiting."); 1894 kproc_exit(0); 1895 } 1896 mtx_lock(&sc->sc_queue_mtx); 1897 } 1898 sx_xunlock(&sc->sc_lock); 1899 /* 1900 * XXX: We can miss an event here, because an event 1901 * can be added without sx-device-lock and without 1902 * mtx-queue-lock. Maybe I should just stop using 1903 * dedicated mutex for events synchronization and 1904 * stick with the queue lock? 1905 * The event will hang here until next I/O request 1906 * or next event is received. 1907 */ 1908 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1909 timeout * hz); 1910 sx_xlock(&sc->sc_lock); 1911 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1912 continue; 1913 } 1914 mtx_unlock(&sc->sc_queue_mtx); 1915 1916 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 1917 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1918 g_mirror_sync_request(bp); /* READ */ 1919 } else if (bp->bio_to != sc->sc_provider) { 1920 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 1921 g_mirror_regular_request(bp); 1922 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 1923 g_mirror_sync_request(bp); /* WRITE */ 1924 else { 1925 KASSERT(0, 1926 ("Invalid request cflags=0x%hx to=%s.", 1927 bp->bio_cflags, bp->bio_to->name)); 1928 } 1929 } else { 1930 g_mirror_register_request(bp); 1931 } 1932 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1933 } 1934} 1935 1936static void 1937g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 1938{ 1939 1940 sx_assert(&sc->sc_lock, SX_LOCKED); 1941 1942 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 1943 return; 1944 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1945 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1946 g_mirror_get_diskname(disk), sc->sc_name); 1947 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1948 } else if (sc->sc_idle && 1949 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1950 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1951 g_mirror_get_diskname(disk), sc->sc_name); 1952 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1953 } 1954} 1955 1956static void 1957g_mirror_sync_start(struct g_mirror_disk *disk) 1958{ 1959 struct g_mirror_softc *sc; 1960 struct g_consumer *cp; 1961 struct bio *bp; 1962 int error, i; 1963 1964 g_topology_assert_not(); 1965 sc = disk->d_softc; 1966 sx_assert(&sc->sc_lock, SX_LOCKED); 1967 1968 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1969 ("Disk %s is not marked for synchronization.", 1970 g_mirror_get_diskname(disk))); 1971 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1972 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1973 sc->sc_state)); 1974 1975 sx_xunlock(&sc->sc_lock); 1976 g_topology_lock(); 1977 cp = g_new_consumer(sc->sc_sync.ds_geom); 1978 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 1979 error = g_attach(cp, sc->sc_provider); 1980 KASSERT(error == 0, 1981 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 1982 error = g_access(cp, 1, 0, 0); 1983 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 1984 g_topology_unlock(); 1985 sx_xlock(&sc->sc_lock); 1986 1987 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1988 g_mirror_get_diskname(disk)); 1989 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0) 1990 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1991 KASSERT(disk->d_sync.ds_consumer == NULL, 1992 ("Sync consumer already exists (device=%s, disk=%s).", 1993 sc->sc_name, g_mirror_get_diskname(disk))); 1994 1995 disk->d_sync.ds_consumer = cp; 1996 disk->d_sync.ds_consumer->private = disk; 1997 disk->d_sync.ds_consumer->index = 0; 1998 1999 /* 2000 * Allocate memory for synchronization bios and initialize them. 2001 */ 2002 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 2003 M_MIRROR, M_WAITOK); 2004 for (i = 0; i < g_mirror_syncreqs; i++) { 2005 bp = g_alloc_bio(); 2006 disk->d_sync.ds_bios[i] = bp; 2007 bp->bio_parent = NULL; 2008 bp->bio_cmd = BIO_READ; 2009 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 2010 bp->bio_cflags = 0; 2011 bp->bio_offset = disk->d_sync.ds_offset; 2012 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 2013 disk->d_sync.ds_offset += bp->bio_length; 2014 bp->bio_done = g_mirror_sync_done; 2015 bp->bio_from = disk->d_sync.ds_consumer; 2016 bp->bio_to = sc->sc_provider; 2017 bp->bio_caller1 = (void *)(uintptr_t)i; 2018 } 2019 2020 /* Increase the number of disks in SYNCHRONIZING state. */ 2021 sc->sc_sync.ds_ndisks++; 2022 /* Set the number of in-flight synchronization requests. */ 2023 disk->d_sync.ds_inflight = g_mirror_syncreqs; 2024 2025 /* 2026 * Fire off first synchronization requests. 2027 */ 2028 for (i = 0; i < g_mirror_syncreqs; i++) { 2029 bp = disk->d_sync.ds_bios[i]; 2030 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 2031 disk->d_sync.ds_consumer->index++; 2032 /* 2033 * Delay the request if it is colliding with a regular request. 2034 */ 2035 if (g_mirror_regular_collision(sc, bp)) 2036 g_mirror_sync_delay(sc, bp); 2037 else 2038 g_io_request(bp, disk->d_sync.ds_consumer); 2039 } 2040} 2041 2042/* 2043 * Stop synchronization process. 2044 * type: 0 - synchronization finished 2045 * 1 - synchronization stopped 2046 */ 2047static void 2048g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 2049{ 2050 struct g_mirror_softc *sc; 2051 struct g_consumer *cp; 2052 2053 g_topology_assert_not(); 2054 sc = disk->d_softc; 2055 sx_assert(&sc->sc_lock, SX_LOCKED); 2056 2057 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2058 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2059 g_mirror_disk_state2str(disk->d_state))); 2060 if (disk->d_sync.ds_consumer == NULL) 2061 return; 2062 2063 if (type == 0) { 2064 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2065 sc->sc_name, g_mirror_get_diskname(disk)); 2066 } else /* if (type == 1) */ { 2067 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2068 sc->sc_name, g_mirror_get_diskname(disk)); 2069 } 2070 free(disk->d_sync.ds_bios, M_MIRROR); 2071 disk->d_sync.ds_bios = NULL; 2072 cp = disk->d_sync.ds_consumer; 2073 disk->d_sync.ds_consumer = NULL; 2074 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2075 sc->sc_sync.ds_ndisks--; 2076 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2077 g_topology_lock(); 2078 g_mirror_kill_consumer(sc, cp); 2079 g_topology_unlock(); 2080 sx_xlock(&sc->sc_lock); 2081} 2082 2083static void 2084g_mirror_launch_provider(struct g_mirror_softc *sc) 2085{ 2086 struct g_mirror_disk *disk; 2087 struct g_provider *pp, *dp; 2088 2089 sx_assert(&sc->sc_lock, SX_LOCKED); 2090 2091 g_topology_lock(); 2092 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 2093 pp->flags |= G_PF_DIRECT_RECEIVE; 2094 pp->mediasize = sc->sc_mediasize; 2095 pp->sectorsize = sc->sc_sectorsize; 2096 pp->stripesize = 0; 2097 pp->stripeoffset = 0; 2098 2099 /* Splitting of unmapped BIO's could work but isn't implemented now */ 2100 if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT) 2101 pp->flags |= G_PF_ACCEPT_UNMAPPED; 2102 2103 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2104 if (disk->d_consumer && disk->d_consumer->provider) { 2105 dp = disk->d_consumer->provider; 2106 if (dp->stripesize > pp->stripesize) { 2107 pp->stripesize = dp->stripesize; 2108 pp->stripeoffset = dp->stripeoffset; 2109 } 2110 /* A provider underneath us doesn't support unmapped */ 2111 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { 2112 G_MIRROR_DEBUG(0, "Cancelling unmapped " 2113 "because of %s.", dp->name); 2114 pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 2115 } 2116 } 2117 } 2118 pp->private = sc; 2119 sc->sc_refcnt++; 2120 sc->sc_provider = pp; 2121 g_error_provider(pp, 0); 2122 g_topology_unlock(); 2123 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2124 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 2125 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2126 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2127 g_mirror_sync_start(disk); 2128 } 2129} 2130 2131static void 2132g_mirror_destroy_provider(struct g_mirror_softc *sc) 2133{ 2134 struct g_mirror_disk *disk; 2135 struct bio *bp; 2136 2137 g_topology_assert_not(); 2138 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2139 sc->sc_name)); 2140 2141 g_topology_lock(); 2142 g_error_provider(sc->sc_provider, ENXIO); 2143 mtx_lock(&sc->sc_queue_mtx); 2144 while ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) { 2145 /* 2146 * Abort any pending I/O that wasn't generated by us. 2147 * Synchronization requests and requests destined for individual 2148 * mirror components can be destroyed immediately. 2149 */ 2150 if (bp->bio_to == sc->sc_provider && 2151 bp->bio_from->geom != sc->sc_sync.ds_geom) { 2152 g_io_deliver(bp, ENXIO); 2153 } else { 2154 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2155 free(bp->bio_data, M_MIRROR); 2156 g_destroy_bio(bp); 2157 } 2158 } 2159 mtx_unlock(&sc->sc_queue_mtx); 2160 g_wither_provider(sc->sc_provider, ENXIO); 2161 sc->sc_provider = NULL; 2162 G_MIRROR_DEBUG(0, "Device %s: provider destroyed.", sc->sc_name); 2163 g_topology_unlock(); 2164 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2165 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2166 g_mirror_sync_stop(disk, 1); 2167 } 2168} 2169 2170static void 2171g_mirror_go(void *arg) 2172{ 2173 struct g_mirror_softc *sc; 2174 2175 sc = arg; 2176 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2177 g_mirror_event_send(sc, 0, 2178 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2179} 2180 2181static u_int 2182g_mirror_determine_state(struct g_mirror_disk *disk) 2183{ 2184 struct g_mirror_softc *sc; 2185 u_int state; 2186 2187 sc = disk->d_softc; 2188 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2189 if ((disk->d_flags & 2190 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2191 /* Disk does not need synchronization. */ 2192 state = G_MIRROR_DISK_STATE_ACTIVE; 2193 } else { 2194 if ((sc->sc_flags & 2195 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2196 (disk->d_flags & 2197 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2198 /* 2199 * We can start synchronization from 2200 * the stored offset. 2201 */ 2202 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2203 } else { 2204 state = G_MIRROR_DISK_STATE_STALE; 2205 } 2206 } 2207 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2208 /* 2209 * Reset all synchronization data for this disk, 2210 * because if it even was synchronized, it was 2211 * synchronized to disks with different syncid. 2212 */ 2213 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2214 disk->d_sync.ds_offset = 0; 2215 disk->d_sync.ds_offset_done = 0; 2216 disk->d_sync.ds_syncid = sc->sc_syncid; 2217 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2218 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2219 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2220 } else { 2221 state = G_MIRROR_DISK_STATE_STALE; 2222 } 2223 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2224 /* 2225 * Not good, NOT GOOD! 2226 * It means that mirror was started on stale disks 2227 * and more fresh disk just arrive. 2228 * If there were writes, mirror is broken, sorry. 2229 * I think the best choice here is don't touch 2230 * this disk and inform the user loudly. 2231 */ 2232 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2233 "disk (%s) arrives!! It will not be connected to the " 2234 "running device.", sc->sc_name, 2235 g_mirror_get_diskname(disk)); 2236 g_mirror_destroy_disk(disk); 2237 state = G_MIRROR_DISK_STATE_NONE; 2238 /* Return immediately, because disk was destroyed. */ 2239 return (state); 2240 } 2241 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2242 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2243 return (state); 2244} 2245 2246/* 2247 * Update device state. 2248 */ 2249static void 2250g_mirror_update_device(struct g_mirror_softc *sc, bool force) 2251{ 2252 struct g_mirror_disk *disk; 2253 u_int state; 2254 2255 sx_assert(&sc->sc_lock, SX_XLOCKED); 2256 2257 switch (sc->sc_state) { 2258 case G_MIRROR_DEVICE_STATE_STARTING: 2259 { 2260 struct g_mirror_disk *pdisk, *tdisk; 2261 u_int dirty, ndisks, genid, syncid; 2262 2263 KASSERT(sc->sc_provider == NULL, 2264 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2265 /* 2266 * Are we ready? We are, if all disks are connected or 2267 * if we have any disks and 'force' is true. 2268 */ 2269 ndisks = g_mirror_ndisks(sc, -1); 2270 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2271 ; 2272 } else if (ndisks == 0) { 2273 /* 2274 * Disks went down in starting phase, so destroy 2275 * device. 2276 */ 2277 callout_drain(&sc->sc_callout); 2278 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2279 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2280 sc->sc_rootmount); 2281 root_mount_rel(sc->sc_rootmount); 2282 sc->sc_rootmount = NULL; 2283 return; 2284 } else { 2285 return; 2286 } 2287 2288 /* 2289 * Activate all disks with the biggest syncid. 2290 */ 2291 if (force) { 2292 /* 2293 * If 'force' is true, we have been called due to 2294 * timeout, so don't bother canceling timeout. 2295 */ 2296 ndisks = 0; 2297 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2298 if ((disk->d_flags & 2299 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2300 ndisks++; 2301 } 2302 } 2303 if (ndisks == 0) { 2304 /* No valid disks found, destroy device. */ 2305 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2306 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2307 __LINE__, sc->sc_rootmount); 2308 root_mount_rel(sc->sc_rootmount); 2309 sc->sc_rootmount = NULL; 2310 return; 2311 } 2312 } else { 2313 /* Cancel timeout. */ 2314 callout_drain(&sc->sc_callout); 2315 } 2316 2317 /* 2318 * Find the biggest genid. 2319 */ 2320 genid = 0; 2321 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2322 if (disk->d_genid > genid) 2323 genid = disk->d_genid; 2324 } 2325 sc->sc_genid = genid; 2326 /* 2327 * Remove all disks without the biggest genid. 2328 */ 2329 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2330 if (disk->d_genid < genid) { 2331 G_MIRROR_DEBUG(0, 2332 "Component %s (device %s) broken, skipping.", 2333 g_mirror_get_diskname(disk), sc->sc_name); 2334 g_mirror_destroy_disk(disk); 2335 } 2336 } 2337 2338 /* 2339 * Find the biggest syncid. 2340 */ 2341 syncid = 0; 2342 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2343 if (disk->d_sync.ds_syncid > syncid) 2344 syncid = disk->d_sync.ds_syncid; 2345 } 2346 2347 /* 2348 * Here we need to look for dirty disks and if all disks 2349 * with the biggest syncid are dirty, we have to choose 2350 * one with the biggest priority and rebuild the rest. 2351 */ 2352 /* 2353 * Find the number of dirty disks with the biggest syncid. 2354 * Find the number of disks with the biggest syncid. 2355 * While here, find a disk with the biggest priority. 2356 */ 2357 dirty = ndisks = 0; 2358 pdisk = NULL; 2359 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2360 if (disk->d_sync.ds_syncid != syncid) 2361 continue; 2362 if ((disk->d_flags & 2363 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2364 continue; 2365 } 2366 ndisks++; 2367 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2368 dirty++; 2369 if (pdisk == NULL || 2370 pdisk->d_priority < disk->d_priority) { 2371 pdisk = disk; 2372 } 2373 } 2374 } 2375 if (dirty == 0) { 2376 /* No dirty disks at all, great. */ 2377 } else if (dirty == ndisks) { 2378 /* 2379 * Force synchronization for all dirty disks except one 2380 * with the biggest priority. 2381 */ 2382 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2383 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2384 "master disk for synchronization.", 2385 g_mirror_get_diskname(pdisk), sc->sc_name); 2386 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2387 if (disk->d_sync.ds_syncid != syncid) 2388 continue; 2389 if ((disk->d_flags & 2390 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2391 continue; 2392 } 2393 KASSERT((disk->d_flags & 2394 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2395 ("Disk %s isn't marked as dirty.", 2396 g_mirror_get_diskname(disk))); 2397 /* Skip the disk with the biggest priority. */ 2398 if (disk == pdisk) 2399 continue; 2400 disk->d_sync.ds_syncid = 0; 2401 } 2402 } else if (dirty < ndisks) { 2403 /* 2404 * Force synchronization for all dirty disks. 2405 * We have some non-dirty disks. 2406 */ 2407 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2408 if (disk->d_sync.ds_syncid != syncid) 2409 continue; 2410 if ((disk->d_flags & 2411 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2412 continue; 2413 } 2414 if ((disk->d_flags & 2415 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2416 continue; 2417 } 2418 disk->d_sync.ds_syncid = 0; 2419 } 2420 } 2421 2422 /* Reset hint. */ 2423 sc->sc_hint = NULL; 2424 sc->sc_syncid = syncid; 2425 if (force) { 2426 /* Remember to bump syncid on first write. */ 2427 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2428 } 2429 state = G_MIRROR_DEVICE_STATE_RUNNING; 2430 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2431 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2432 g_mirror_device_state2str(state)); 2433 sc->sc_state = state; 2434 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2435 state = g_mirror_determine_state(disk); 2436 g_mirror_event_send(disk, state, 2437 G_MIRROR_EVENT_DONTWAIT); 2438 if (state == G_MIRROR_DISK_STATE_STALE) 2439 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2440 } 2441 break; 2442 } 2443 case G_MIRROR_DEVICE_STATE_RUNNING: 2444 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2445 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2446 /* 2447 * No active disks or no disks at all, 2448 * so destroy device. 2449 */ 2450 if (sc->sc_provider != NULL) 2451 g_mirror_destroy_provider(sc); 2452 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2453 break; 2454 } else if (g_mirror_ndisks(sc, 2455 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2456 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2457 /* 2458 * We have active disks, launch provider if it doesn't 2459 * exist. 2460 */ 2461 if (sc->sc_provider == NULL) 2462 g_mirror_launch_provider(sc); 2463 if (sc->sc_rootmount != NULL) { 2464 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2465 __LINE__, sc->sc_rootmount); 2466 root_mount_rel(sc->sc_rootmount); 2467 sc->sc_rootmount = NULL; 2468 } 2469 } 2470 /* 2471 * Genid should be bumped immediately, so do it here. 2472 */ 2473 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2474 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2475 g_mirror_bump_genid(sc); 2476 } 2477 break; 2478 default: 2479 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2480 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2481 break; 2482 } 2483} 2484 2485/* 2486 * Update disk state and device state if needed. 2487 */ 2488#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2489 "Disk %s state changed from %s to %s (device %s).", \ 2490 g_mirror_get_diskname(disk), \ 2491 g_mirror_disk_state2str(disk->d_state), \ 2492 g_mirror_disk_state2str(state), sc->sc_name) 2493static int 2494g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2495{ 2496 struct g_mirror_softc *sc; 2497 2498 sc = disk->d_softc; 2499 sx_assert(&sc->sc_lock, SX_XLOCKED); 2500 2501again: 2502 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2503 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2504 g_mirror_disk_state2str(state)); 2505 switch (state) { 2506 case G_MIRROR_DISK_STATE_NEW: 2507 /* 2508 * Possible scenarios: 2509 * 1. New disk arrive. 2510 */ 2511 /* Previous state should be NONE. */ 2512 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2513 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2514 g_mirror_disk_state2str(disk->d_state))); 2515 DISK_STATE_CHANGED(); 2516 2517 disk->d_state = state; 2518 if (LIST_EMPTY(&sc->sc_disks)) 2519 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2520 else { 2521 struct g_mirror_disk *dp; 2522 2523 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2524 if (disk->d_priority >= dp->d_priority) { 2525 LIST_INSERT_BEFORE(dp, disk, d_next); 2526 dp = NULL; 2527 break; 2528 } 2529 if (LIST_NEXT(dp, d_next) == NULL) 2530 break; 2531 } 2532 if (dp != NULL) 2533 LIST_INSERT_AFTER(dp, disk, d_next); 2534 } 2535 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2536 sc->sc_name, g_mirror_get_diskname(disk)); 2537 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2538 break; 2539 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2540 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2541 g_mirror_device_state2str(sc->sc_state), 2542 g_mirror_get_diskname(disk), 2543 g_mirror_disk_state2str(disk->d_state))); 2544 state = g_mirror_determine_state(disk); 2545 if (state != G_MIRROR_DISK_STATE_NONE) 2546 goto again; 2547 break; 2548 case G_MIRROR_DISK_STATE_ACTIVE: 2549 /* 2550 * Possible scenarios: 2551 * 1. New disk does not need synchronization. 2552 * 2. Synchronization process finished successfully. 2553 */ 2554 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2555 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2556 g_mirror_device_state2str(sc->sc_state), 2557 g_mirror_get_diskname(disk), 2558 g_mirror_disk_state2str(disk->d_state))); 2559 /* Previous state should be NEW or SYNCHRONIZING. */ 2560 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2561 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2562 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2563 g_mirror_disk_state2str(disk->d_state))); 2564 DISK_STATE_CHANGED(); 2565 2566 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2567 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2568 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2569 g_mirror_sync_stop(disk, 0); 2570 } 2571 disk->d_state = state; 2572 disk->d_sync.ds_offset = 0; 2573 disk->d_sync.ds_offset_done = 0; 2574 g_mirror_update_idle(sc, disk); 2575 g_mirror_update_metadata(disk); 2576 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2577 sc->sc_name, g_mirror_get_diskname(disk)); 2578 break; 2579 case G_MIRROR_DISK_STATE_STALE: 2580 /* 2581 * Possible scenarios: 2582 * 1. Stale disk was connected. 2583 */ 2584 /* Previous state should be NEW. */ 2585 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2586 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2587 g_mirror_disk_state2str(disk->d_state))); 2588 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2589 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2590 g_mirror_device_state2str(sc->sc_state), 2591 g_mirror_get_diskname(disk), 2592 g_mirror_disk_state2str(disk->d_state))); 2593 /* 2594 * STALE state is only possible if device is marked 2595 * NOAUTOSYNC. 2596 */ 2597 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2598 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2599 g_mirror_device_state2str(sc->sc_state), 2600 g_mirror_get_diskname(disk), 2601 g_mirror_disk_state2str(disk->d_state))); 2602 DISK_STATE_CHANGED(); 2603 2604 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2605 disk->d_state = state; 2606 g_mirror_update_metadata(disk); 2607 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2608 sc->sc_name, g_mirror_get_diskname(disk)); 2609 break; 2610 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2611 /* 2612 * Possible scenarios: 2613 * 1. Disk which needs synchronization was connected. 2614 */ 2615 /* Previous state should be NEW. */ 2616 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2617 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2618 g_mirror_disk_state2str(disk->d_state))); 2619 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2620 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2621 g_mirror_device_state2str(sc->sc_state), 2622 g_mirror_get_diskname(disk), 2623 g_mirror_disk_state2str(disk->d_state))); 2624 DISK_STATE_CHANGED(); 2625 2626 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2627 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2628 disk->d_state = state; 2629 if (sc->sc_provider != NULL) { 2630 g_mirror_sync_start(disk); 2631 g_mirror_update_metadata(disk); 2632 } 2633 break; 2634 case G_MIRROR_DISK_STATE_DISCONNECTED: 2635 /* 2636 * Possible scenarios: 2637 * 1. Device wasn't running yet, but disk disappear. 2638 * 2. Disk was active and disapppear. 2639 * 3. Disk disappear during synchronization process. 2640 */ 2641 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2642 /* 2643 * Previous state should be ACTIVE, STALE or 2644 * SYNCHRONIZING. 2645 */ 2646 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2647 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2648 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2649 ("Wrong disk state (%s, %s).", 2650 g_mirror_get_diskname(disk), 2651 g_mirror_disk_state2str(disk->d_state))); 2652 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2653 /* Previous state should be NEW. */ 2654 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2655 ("Wrong disk state (%s, %s).", 2656 g_mirror_get_diskname(disk), 2657 g_mirror_disk_state2str(disk->d_state))); 2658 /* 2659 * Reset bumping syncid if disk disappeared in STARTING 2660 * state. 2661 */ 2662 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2663 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2664#ifdef INVARIANTS 2665 } else { 2666 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2667 sc->sc_name, 2668 g_mirror_device_state2str(sc->sc_state), 2669 g_mirror_get_diskname(disk), 2670 g_mirror_disk_state2str(disk->d_state))); 2671#endif 2672 } 2673 DISK_STATE_CHANGED(); 2674 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2675 sc->sc_name, g_mirror_get_diskname(disk)); 2676 2677 g_mirror_destroy_disk(disk); 2678 break; 2679 case G_MIRROR_DISK_STATE_DESTROY: 2680 { 2681 int error; 2682 2683 error = g_mirror_clear_metadata(disk); 2684 if (error != 0) { 2685 G_MIRROR_DEBUG(0, 2686 "Device %s: failed to clear metadata on %s: %d.", 2687 sc->sc_name, g_mirror_get_diskname(disk), error); 2688 break; 2689 } 2690 DISK_STATE_CHANGED(); 2691 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2692 sc->sc_name, g_mirror_get_diskname(disk)); 2693 2694 g_mirror_destroy_disk(disk); 2695 sc->sc_ndisks--; 2696 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2697 g_mirror_update_metadata(disk); 2698 } 2699 break; 2700 } 2701 default: 2702 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2703 break; 2704 } 2705 return (0); 2706} 2707#undef DISK_STATE_CHANGED 2708 2709int 2710g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2711{ 2712 struct g_provider *pp; 2713 u_char *buf; 2714 int error; 2715 2716 g_topology_assert(); 2717 2718 error = g_access(cp, 1, 0, 0); 2719 if (error != 0) 2720 return (error); 2721 pp = cp->provider; 2722 g_topology_unlock(); 2723 /* Metadata are stored on last sector. */ 2724 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2725 &error); 2726 g_topology_lock(); 2727 g_access(cp, -1, 0, 0); 2728 if (buf == NULL) { 2729 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2730 cp->provider->name, error); 2731 return (error); 2732 } 2733 2734 /* Decode metadata. */ 2735 error = mirror_metadata_decode(buf, md); 2736 g_free(buf); 2737 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2738 return (EINVAL); 2739 if (md->md_version > G_MIRROR_VERSION) { 2740 G_MIRROR_DEBUG(0, 2741 "Kernel module is too old to handle metadata from %s.", 2742 cp->provider->name); 2743 return (EINVAL); 2744 } 2745 if (error != 0) { 2746 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2747 cp->provider->name); 2748 return (error); 2749 } 2750 2751 return (0); 2752} 2753 2754static int 2755g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2756 struct g_mirror_metadata *md) 2757{ 2758 2759 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2760 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2761 pp->name, md->md_did); 2762 return (EEXIST); 2763 } 2764 if (md->md_all != sc->sc_ndisks) { 2765 G_MIRROR_DEBUG(1, 2766 "Invalid '%s' field on disk %s (device %s), skipping.", 2767 "md_all", pp->name, sc->sc_name); 2768 return (EINVAL); 2769 } 2770 if (md->md_slice != sc->sc_slice) { 2771 G_MIRROR_DEBUG(1, 2772 "Invalid '%s' field on disk %s (device %s), skipping.", 2773 "md_slice", pp->name, sc->sc_name); 2774 return (EINVAL); 2775 } 2776 if (md->md_balance != sc->sc_balance) { 2777 G_MIRROR_DEBUG(1, 2778 "Invalid '%s' field on disk %s (device %s), skipping.", 2779 "md_balance", pp->name, sc->sc_name); 2780 return (EINVAL); 2781 } 2782#if 0 2783 if (md->md_mediasize != sc->sc_mediasize) { 2784 G_MIRROR_DEBUG(1, 2785 "Invalid '%s' field on disk %s (device %s), skipping.", 2786 "md_mediasize", pp->name, sc->sc_name); 2787 return (EINVAL); 2788 } 2789#endif 2790 if (sc->sc_mediasize > pp->mediasize) { 2791 G_MIRROR_DEBUG(1, 2792 "Invalid size of disk %s (device %s), skipping.", pp->name, 2793 sc->sc_name); 2794 return (EINVAL); 2795 } 2796 if (md->md_sectorsize != sc->sc_sectorsize) { 2797 G_MIRROR_DEBUG(1, 2798 "Invalid '%s' field on disk %s (device %s), skipping.", 2799 "md_sectorsize", pp->name, sc->sc_name); 2800 return (EINVAL); 2801 } 2802 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2803 G_MIRROR_DEBUG(1, 2804 "Invalid sector size of disk %s (device %s), skipping.", 2805 pp->name, sc->sc_name); 2806 return (EINVAL); 2807 } 2808 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2809 G_MIRROR_DEBUG(1, 2810 "Invalid device flags on disk %s (device %s), skipping.", 2811 pp->name, sc->sc_name); 2812 return (EINVAL); 2813 } 2814 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2815 G_MIRROR_DEBUG(1, 2816 "Invalid disk flags on disk %s (device %s), skipping.", 2817 pp->name, sc->sc_name); 2818 return (EINVAL); 2819 } 2820 return (0); 2821} 2822 2823int 2824g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2825 struct g_mirror_metadata *md) 2826{ 2827 struct g_mirror_disk *disk; 2828 int error; 2829 2830 g_topology_assert_not(); 2831 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2832 2833 error = g_mirror_check_metadata(sc, pp, md); 2834 if (error != 0) 2835 return (error); 2836 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2837 md->md_genid < sc->sc_genid) { 2838 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2839 pp->name, sc->sc_name); 2840 return (EINVAL); 2841 } 2842 disk = g_mirror_init_disk(sc, pp, md, &error); 2843 if (disk == NULL) 2844 return (error); 2845 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2846 G_MIRROR_EVENT_WAIT); 2847 if (error != 0) 2848 return (error); 2849 if (md->md_version < G_MIRROR_VERSION) { 2850 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2851 pp->name, md->md_version, G_MIRROR_VERSION); 2852 g_mirror_update_metadata(disk); 2853 } 2854 return (0); 2855} 2856 2857static void 2858g_mirror_destroy_delayed(void *arg, int flag) 2859{ 2860 struct g_mirror_softc *sc; 2861 int error; 2862 2863 if (flag == EV_CANCEL) { 2864 G_MIRROR_DEBUG(1, "Destroying canceled."); 2865 return; 2866 } 2867 sc = arg; 2868 g_topology_unlock(); 2869 sx_xlock(&sc->sc_lock); 2870 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2871 ("DESTROY flag set on %s.", sc->sc_name)); 2872 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0, 2873 ("DESTROYING flag not set on %s.", sc->sc_name)); 2874 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 2875 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 2876 if (error != 0) { 2877 G_MIRROR_DEBUG(0, "Cannot destroy %s (error=%d).", 2878 sc->sc_name, error); 2879 sx_xunlock(&sc->sc_lock); 2880 } 2881 g_topology_lock(); 2882} 2883 2884static int 2885g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2886{ 2887 struct g_mirror_softc *sc; 2888 int error = 0; 2889 2890 g_topology_assert(); 2891 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2892 acw, ace); 2893 2894 sc = pp->private; 2895 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 2896 2897 g_topology_unlock(); 2898 sx_xlock(&sc->sc_lock); 2899 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 2900 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0 || 2901 LIST_EMPTY(&sc->sc_disks)) { 2902 if (acr > 0 || acw > 0 || ace > 0) 2903 error = ENXIO; 2904 goto end; 2905 } 2906 sc->sc_provider_open += acr + acw + ace; 2907 if (pp->acw + acw == 0) 2908 g_mirror_idle(sc, 0); 2909 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0 && 2910 sc->sc_provider_open == 0) 2911 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, sc, NULL); 2912end: 2913 sx_xunlock(&sc->sc_lock); 2914 g_topology_lock(); 2915 return (error); 2916} 2917 2918static struct g_geom * 2919g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2920{ 2921 struct g_mirror_softc *sc; 2922 struct g_geom *gp; 2923 int error, timeout; 2924 2925 g_topology_assert(); 2926 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2927 md->md_mid); 2928 2929 /* One disk is minimum. */ 2930 if (md->md_all < 1) 2931 return (NULL); 2932 /* 2933 * Action geom. 2934 */ 2935 gp = g_new_geomf(mp, "%s", md->md_name); 2936 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2937 gp->start = g_mirror_start; 2938 gp->orphan = g_mirror_orphan; 2939 gp->access = g_mirror_access; 2940 gp->dumpconf = g_mirror_dumpconf; 2941 2942 sc->sc_id = md->md_mid; 2943 sc->sc_slice = md->md_slice; 2944 sc->sc_balance = md->md_balance; 2945 sc->sc_mediasize = md->md_mediasize; 2946 sc->sc_sectorsize = md->md_sectorsize; 2947 sc->sc_ndisks = md->md_all; 2948 sc->sc_flags = md->md_mflags; 2949 sc->sc_bump_id = 0; 2950 sc->sc_idle = 1; 2951 sc->sc_last_write = time_uptime; 2952 sc->sc_writes = 0; 2953 sc->sc_refcnt = 1; 2954 sx_init(&sc->sc_lock, "gmirror:lock"); 2955 bioq_init(&sc->sc_queue); 2956 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2957 bioq_init(&sc->sc_regular_delayed); 2958 bioq_init(&sc->sc_inflight); 2959 bioq_init(&sc->sc_sync_delayed); 2960 LIST_INIT(&sc->sc_disks); 2961 TAILQ_INIT(&sc->sc_events); 2962 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2963 callout_init(&sc->sc_callout, 1); 2964 mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF); 2965 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2966 gp->softc = sc; 2967 sc->sc_geom = gp; 2968 sc->sc_provider = NULL; 2969 sc->sc_provider_open = 0; 2970 /* 2971 * Synchronization geom. 2972 */ 2973 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2974 gp->softc = sc; 2975 gp->orphan = g_mirror_orphan; 2976 sc->sc_sync.ds_geom = gp; 2977 sc->sc_sync.ds_ndisks = 0; 2978 error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2979 "g_mirror %s", md->md_name); 2980 if (error != 0) { 2981 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2982 sc->sc_name); 2983 g_destroy_geom(sc->sc_sync.ds_geom); 2984 g_destroy_geom(sc->sc_geom); 2985 g_mirror_free_device(sc); 2986 return (NULL); 2987 } 2988 2989 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 2990 sc->sc_name, sc->sc_ndisks, sc->sc_id); 2991 2992 sc->sc_rootmount = root_mount_hold("GMIRROR"); 2993 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 2994 /* 2995 * Run timeout. 2996 */ 2997 timeout = g_mirror_timeout * hz; 2998 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2999 return (sc->sc_geom); 3000} 3001 3002int 3003g_mirror_destroy(struct g_mirror_softc *sc, int how) 3004{ 3005 struct g_mirror_disk *disk; 3006 3007 g_topology_assert_not(); 3008 sx_assert(&sc->sc_lock, SX_XLOCKED); 3009 3010 if (sc->sc_provider_open != 0 || SCHEDULER_STOPPED()) { 3011 switch (how) { 3012 case G_MIRROR_DESTROY_SOFT: 3013 G_MIRROR_DEBUG(1, 3014 "Device %s is still open (%d).", sc->sc_name, 3015 sc->sc_provider_open); 3016 return (EBUSY); 3017 case G_MIRROR_DESTROY_DELAYED: 3018 G_MIRROR_DEBUG(1, 3019 "Device %s will be destroyed on last close.", 3020 sc->sc_name); 3021 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 3022 if (disk->d_state == 3023 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3024 g_mirror_sync_stop(disk, 1); 3025 } 3026 } 3027 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROYING; 3028 return (EBUSY); 3029 case G_MIRROR_DESTROY_HARD: 3030 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 3031 "can't be definitely removed.", sc->sc_name); 3032 } 3033 } 3034 3035 g_topology_lock(); 3036 if (sc->sc_geom->softc == NULL) { 3037 g_topology_unlock(); 3038 return (0); 3039 } 3040 sc->sc_geom->softc = NULL; 3041 sc->sc_sync.ds_geom->softc = NULL; 3042 g_topology_unlock(); 3043 3044 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 3045 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 3046 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3047 sx_xunlock(&sc->sc_lock); 3048 mtx_lock(&sc->sc_queue_mtx); 3049 wakeup(sc); 3050 mtx_unlock(&sc->sc_queue_mtx); 3051 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3052 while (sc->sc_worker != NULL) 3053 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 3054 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3055 sx_xlock(&sc->sc_lock); 3056 g_mirror_destroy_device(sc); 3057 return (0); 3058} 3059 3060static void 3061g_mirror_taste_orphan(struct g_consumer *cp) 3062{ 3063 3064 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3065 cp->provider->name)); 3066} 3067 3068static struct g_geom * 3069g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3070{ 3071 struct g_mirror_metadata md; 3072 struct g_mirror_softc *sc; 3073 struct g_consumer *cp; 3074 struct g_geom *gp; 3075 int error; 3076 3077 g_topology_assert(); 3078 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3079 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 3080 3081 gp = g_new_geomf(mp, "mirror:taste"); 3082 /* 3083 * This orphan function should be never called. 3084 */ 3085 gp->orphan = g_mirror_taste_orphan; 3086 cp = g_new_consumer(gp); 3087 g_attach(cp, pp); 3088 error = g_mirror_read_metadata(cp, &md); 3089 g_detach(cp); 3090 g_destroy_consumer(cp); 3091 g_destroy_geom(gp); 3092 if (error != 0) 3093 return (NULL); 3094 gp = NULL; 3095 3096 if (md.md_provider[0] != '\0' && 3097 !g_compare_names(md.md_provider, pp->name)) 3098 return (NULL); 3099 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3100 return (NULL); 3101 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 3102 G_MIRROR_DEBUG(0, 3103 "Device %s: provider %s marked as inactive, skipping.", 3104 md.md_name, pp->name); 3105 return (NULL); 3106 } 3107 if (g_mirror_debug >= 2) 3108 mirror_metadata_dump(&md); 3109 3110 /* 3111 * Let's check if device already exists. 3112 */ 3113 sc = NULL; 3114 LIST_FOREACH(gp, &mp->geom, geom) { 3115 sc = gp->softc; 3116 if (sc == NULL) 3117 continue; 3118 if (sc->sc_sync.ds_geom == gp) 3119 continue; 3120 if (strcmp(md.md_name, sc->sc_name) != 0) 3121 continue; 3122 if (md.md_mid != sc->sc_id) { 3123 G_MIRROR_DEBUG(0, "Device %s already configured.", 3124 sc->sc_name); 3125 return (NULL); 3126 } 3127 break; 3128 } 3129 if (gp == NULL) { 3130 gp = g_mirror_create(mp, &md); 3131 if (gp == NULL) { 3132 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3133 md.md_name); 3134 return (NULL); 3135 } 3136 sc = gp->softc; 3137 } 3138 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3139 g_topology_unlock(); 3140 sx_xlock(&sc->sc_lock); 3141 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; 3142 error = g_mirror_add_disk(sc, pp, &md); 3143 if (error != 0) { 3144 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3145 pp->name, gp->name, error); 3146 if (LIST_EMPTY(&sc->sc_disks)) { 3147 g_cancel_event(sc); 3148 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3149 g_topology_lock(); 3150 return (NULL); 3151 } 3152 gp = NULL; 3153 } 3154 sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; 3155 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3156 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3157 g_topology_lock(); 3158 return (NULL); 3159 } 3160 sx_xunlock(&sc->sc_lock); 3161 g_topology_lock(); 3162 return (gp); 3163} 3164 3165static void 3166g_mirror_resize(struct g_consumer *cp) 3167{ 3168 struct g_mirror_disk *disk; 3169 3170 g_topology_assert(); 3171 g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name); 3172 3173 disk = cp->private; 3174 if (disk == NULL) 3175 return; 3176 g_topology_unlock(); 3177 g_mirror_update_metadata(disk); 3178 g_topology_lock(); 3179} 3180 3181static int 3182g_mirror_destroy_geom(struct gctl_req *req __unused, 3183 struct g_class *mp __unused, struct g_geom *gp) 3184{ 3185 struct g_mirror_softc *sc; 3186 int error; 3187 3188 g_topology_unlock(); 3189 sc = gp->softc; 3190 sx_xlock(&sc->sc_lock); 3191 g_cancel_event(sc); 3192 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3193 if (error != 0) 3194 sx_xunlock(&sc->sc_lock); 3195 g_topology_lock(); 3196 return (error); 3197} 3198 3199static void 3200g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3201 struct g_consumer *cp, struct g_provider *pp) 3202{ 3203 struct g_mirror_softc *sc; 3204 3205 g_topology_assert(); 3206 3207 sc = gp->softc; 3208 if (sc == NULL) 3209 return; 3210 /* Skip synchronization geom. */ 3211 if (gp == sc->sc_sync.ds_geom) 3212 return; 3213 if (pp != NULL) { 3214 /* Nothing here. */ 3215 } else if (cp != NULL) { 3216 struct g_mirror_disk *disk; 3217 3218 disk = cp->private; 3219 if (disk == NULL) 3220 return; 3221 g_topology_unlock(); 3222 sx_xlock(&sc->sc_lock); 3223 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3224 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3225 sbuf_printf(sb, "%s<Synchronized>", indent); 3226 if (disk->d_sync.ds_offset == 0) 3227 sbuf_printf(sb, "0%%"); 3228 else { 3229 sbuf_printf(sb, "%u%%", 3230 (u_int)((disk->d_sync.ds_offset * 100) / 3231 sc->sc_provider->mediasize)); 3232 } 3233 sbuf_printf(sb, "</Synchronized>\n"); 3234 if (disk->d_sync.ds_offset > 0) { 3235 sbuf_printf(sb, "%s<BytesSynced>%jd" 3236 "</BytesSynced>\n", indent, 3237 (intmax_t)disk->d_sync.ds_offset); 3238 } 3239 } 3240 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3241 disk->d_sync.ds_syncid); 3242 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3243 disk->d_genid); 3244 sbuf_printf(sb, "%s<Flags>", indent); 3245 if (disk->d_flags == 0) 3246 sbuf_printf(sb, "NONE"); 3247 else { 3248 int first = 1; 3249 3250#define ADD_FLAG(flag, name) do { \ 3251 if ((disk->d_flags & (flag)) != 0) { \ 3252 if (!first) \ 3253 sbuf_printf(sb, ", "); \ 3254 else \ 3255 first = 0; \ 3256 sbuf_printf(sb, name); \ 3257 } \ 3258} while (0) 3259 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3260 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3261 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3262 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3263 "SYNCHRONIZING"); 3264 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3265 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3266#undef ADD_FLAG 3267 } 3268 sbuf_printf(sb, "</Flags>\n"); 3269 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3270 disk->d_priority); 3271 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3272 g_mirror_disk_state2str(disk->d_state)); 3273 sx_xunlock(&sc->sc_lock); 3274 g_topology_lock(); 3275 } else { 3276 g_topology_unlock(); 3277 sx_xlock(&sc->sc_lock); 3278 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3279 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3280 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3281 sbuf_printf(sb, "%s<Flags>", indent); 3282 if (sc->sc_flags == 0) 3283 sbuf_printf(sb, "NONE"); 3284 else { 3285 int first = 1; 3286 3287#define ADD_FLAG(flag, name) do { \ 3288 if ((sc->sc_flags & (flag)) != 0) { \ 3289 if (!first) \ 3290 sbuf_printf(sb, ", "); \ 3291 else \ 3292 first = 0; \ 3293 sbuf_printf(sb, name); \ 3294 } \ 3295} while (0) 3296 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3297 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3298#undef ADD_FLAG 3299 } 3300 sbuf_printf(sb, "</Flags>\n"); 3301 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3302 (u_int)sc->sc_slice); 3303 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3304 balance_name(sc->sc_balance)); 3305 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3306 sc->sc_ndisks); 3307 sbuf_printf(sb, "%s<State>", indent); 3308 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3309 sbuf_printf(sb, "%s", "STARTING"); 3310 else if (sc->sc_ndisks == 3311 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3312 sbuf_printf(sb, "%s", "COMPLETE"); 3313 else 3314 sbuf_printf(sb, "%s", "DEGRADED"); 3315 sbuf_printf(sb, "</State>\n"); 3316 sx_xunlock(&sc->sc_lock); 3317 g_topology_lock(); 3318 } 3319} 3320 3321static void 3322g_mirror_shutdown_post_sync(void *arg, int howto) 3323{ 3324 struct g_class *mp; 3325 struct g_geom *gp, *gp2; 3326 struct g_mirror_softc *sc; 3327 int error; 3328 3329 mp = arg; 3330 g_topology_lock(); 3331 g_mirror_shutdown = 1; 3332 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3333 if ((sc = gp->softc) == NULL) 3334 continue; 3335 /* Skip synchronization geom. */ 3336 if (gp == sc->sc_sync.ds_geom) 3337 continue; 3338 g_topology_unlock(); 3339 sx_xlock(&sc->sc_lock); 3340 g_mirror_idle(sc, -1); 3341 g_cancel_event(sc); 3342 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3343 if (error != 0) 3344 sx_xunlock(&sc->sc_lock); 3345 g_topology_lock(); 3346 } 3347 g_topology_unlock(); 3348} 3349 3350static void 3351g_mirror_init(struct g_class *mp) 3352{ 3353 3354 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3355 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3356 if (g_mirror_post_sync == NULL) 3357 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3358} 3359 3360static void 3361g_mirror_fini(struct g_class *mp) 3362{ 3363 3364 if (g_mirror_post_sync != NULL) 3365 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); 3366} 3367 3368DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3369