g_mirror.c revision 324401
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 324401 2017-10-07 22:59:09Z ngie $"); 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 if (disk->d_softc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 702 return (0); 703 error = g_mirror_write_metadata(disk, NULL); 704 if (error == 0) { 705 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 706 g_mirror_get_diskname(disk)); 707 } else { 708 G_MIRROR_DEBUG(0, 709 "Cannot clear metadata on disk %s (error=%d).", 710 g_mirror_get_diskname(disk), error); 711 } 712 return (error); 713} 714 715void 716g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 717 struct g_mirror_metadata *md) 718{ 719 720 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 721 md->md_version = G_MIRROR_VERSION; 722 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 723 md->md_mid = sc->sc_id; 724 md->md_all = sc->sc_ndisks; 725 md->md_slice = sc->sc_slice; 726 md->md_balance = sc->sc_balance; 727 md->md_genid = sc->sc_genid; 728 md->md_mediasize = sc->sc_mediasize; 729 md->md_sectorsize = sc->sc_sectorsize; 730 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 731 bzero(md->md_provider, sizeof(md->md_provider)); 732 if (disk == NULL) { 733 md->md_did = arc4random(); 734 md->md_priority = 0; 735 md->md_syncid = 0; 736 md->md_dflags = 0; 737 md->md_sync_offset = 0; 738 md->md_provsize = 0; 739 } else { 740 md->md_did = disk->d_id; 741 md->md_priority = disk->d_priority; 742 md->md_syncid = disk->d_sync.ds_syncid; 743 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 744 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 745 md->md_sync_offset = disk->d_sync.ds_offset_done; 746 else 747 md->md_sync_offset = 0; 748 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 749 strlcpy(md->md_provider, 750 disk->d_consumer->provider->name, 751 sizeof(md->md_provider)); 752 } 753 md->md_provsize = disk->d_consumer->provider->mediasize; 754 } 755} 756 757void 758g_mirror_update_metadata(struct g_mirror_disk *disk) 759{ 760 struct g_mirror_softc *sc; 761 struct g_mirror_metadata md; 762 int error; 763 764 g_topology_assert_not(); 765 sc = disk->d_softc; 766 sx_assert(&sc->sc_lock, SX_LOCKED); 767 768 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 769 return; 770 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) 771 g_mirror_fill_metadata(sc, disk, &md); 772 error = g_mirror_write_metadata(disk, &md); 773 if (error == 0) { 774 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 775 g_mirror_get_diskname(disk)); 776 } else { 777 G_MIRROR_DEBUG(0, 778 "Cannot update metadata on disk %s (error=%d).", 779 g_mirror_get_diskname(disk), error); 780 } 781} 782 783static void 784g_mirror_bump_syncid(struct g_mirror_softc *sc) 785{ 786 struct g_mirror_disk *disk; 787 788 g_topology_assert_not(); 789 sx_assert(&sc->sc_lock, SX_XLOCKED); 790 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 791 ("%s called with no active disks (device=%s).", __func__, 792 sc->sc_name)); 793 794 sc->sc_syncid++; 795 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 796 sc->sc_syncid); 797 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 798 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 799 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 800 disk->d_sync.ds_syncid = sc->sc_syncid; 801 g_mirror_update_metadata(disk); 802 } 803 } 804} 805 806static void 807g_mirror_bump_genid(struct g_mirror_softc *sc) 808{ 809 struct g_mirror_disk *disk; 810 811 g_topology_assert_not(); 812 sx_assert(&sc->sc_lock, SX_XLOCKED); 813 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 814 ("%s called with no active disks (device=%s).", __func__, 815 sc->sc_name)); 816 817 sc->sc_genid++; 818 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 819 sc->sc_genid); 820 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 821 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 822 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 823 disk->d_genid = sc->sc_genid; 824 g_mirror_update_metadata(disk); 825 } 826 } 827} 828 829static int 830g_mirror_idle(struct g_mirror_softc *sc, int acw) 831{ 832 struct g_mirror_disk *disk; 833 int timeout; 834 835 g_topology_assert_not(); 836 sx_assert(&sc->sc_lock, SX_XLOCKED); 837 838 if (sc->sc_provider == NULL) 839 return (0); 840 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 841 return (0); 842 if (sc->sc_idle) 843 return (0); 844 if (sc->sc_writes > 0) 845 return (0); 846 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 847 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); 848 if (!g_mirror_shutdown && timeout > 0) 849 return (timeout); 850 } 851 sc->sc_idle = 1; 852 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 853 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 854 continue; 855 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 856 g_mirror_get_diskname(disk), sc->sc_name); 857 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 858 g_mirror_update_metadata(disk); 859 } 860 return (0); 861} 862 863static void 864g_mirror_unidle(struct g_mirror_softc *sc) 865{ 866 struct g_mirror_disk *disk; 867 868 g_topology_assert_not(); 869 sx_assert(&sc->sc_lock, SX_XLOCKED); 870 871 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 872 return; 873 sc->sc_idle = 0; 874 sc->sc_last_write = time_uptime; 875 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 876 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 877 continue; 878 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 879 g_mirror_get_diskname(disk), sc->sc_name); 880 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 881 g_mirror_update_metadata(disk); 882 } 883} 884 885static void 886g_mirror_flush_done(struct bio *bp) 887{ 888 struct g_mirror_softc *sc; 889 struct bio *pbp; 890 891 pbp = bp->bio_parent; 892 sc = pbp->bio_to->private; 893 mtx_lock(&sc->sc_done_mtx); 894 if (pbp->bio_error == 0) 895 pbp->bio_error = bp->bio_error; 896 pbp->bio_completed += bp->bio_completed; 897 pbp->bio_inbed++; 898 if (pbp->bio_children == pbp->bio_inbed) { 899 mtx_unlock(&sc->sc_done_mtx); 900 g_io_deliver(pbp, pbp->bio_error); 901 } else 902 mtx_unlock(&sc->sc_done_mtx); 903 g_destroy_bio(bp); 904} 905 906static void 907g_mirror_done(struct bio *bp) 908{ 909 struct g_mirror_softc *sc; 910 911 sc = bp->bio_from->geom->softc; 912 bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR; 913 mtx_lock(&sc->sc_queue_mtx); 914 bioq_insert_tail(&sc->sc_queue, bp); 915 mtx_unlock(&sc->sc_queue_mtx); 916 wakeup(sc); 917} 918 919static void 920g_mirror_regular_request(struct bio *bp) 921{ 922 struct g_mirror_softc *sc; 923 struct g_mirror_disk *disk; 924 struct bio *pbp; 925 926 g_topology_assert_not(); 927 928 pbp = bp->bio_parent; 929 sc = pbp->bio_to->private; 930 bp->bio_from->index--; 931 if (bp->bio_cmd == BIO_WRITE) 932 sc->sc_writes--; 933 disk = bp->bio_from->private; 934 if (disk == NULL) { 935 g_topology_lock(); 936 g_mirror_kill_consumer(sc, bp->bio_from); 937 g_topology_unlock(); 938 } 939 940 pbp->bio_inbed++; 941 KASSERT(pbp->bio_inbed <= pbp->bio_children, 942 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 943 pbp->bio_children)); 944 if (bp->bio_error == 0 && pbp->bio_error == 0) { 945 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 946 g_destroy_bio(bp); 947 if (pbp->bio_children == pbp->bio_inbed) { 948 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 949 pbp->bio_completed = pbp->bio_length; 950 if (pbp->bio_cmd == BIO_WRITE || 951 pbp->bio_cmd == BIO_DELETE) { 952 bioq_remove(&sc->sc_inflight, pbp); 953 /* Release delayed sync requests if possible. */ 954 g_mirror_sync_release(sc); 955 } 956 g_io_deliver(pbp, pbp->bio_error); 957 } 958 return; 959 } else if (bp->bio_error != 0) { 960 if (pbp->bio_error == 0) 961 pbp->bio_error = bp->bio_error; 962 if (disk != NULL) { 963 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 964 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 965 G_MIRROR_LOGREQ(0, bp, 966 "Request failed (error=%d).", 967 bp->bio_error); 968 } else { 969 G_MIRROR_LOGREQ(1, bp, 970 "Request failed (error=%d).", 971 bp->bio_error); 972 } 973 if (g_mirror_disconnect_on_failure && 974 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) 975 { 976 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 977 g_mirror_event_send(disk, 978 G_MIRROR_DISK_STATE_DISCONNECTED, 979 G_MIRROR_EVENT_DONTWAIT); 980 } 981 } 982 switch (pbp->bio_cmd) { 983 case BIO_DELETE: 984 case BIO_WRITE: 985 pbp->bio_inbed--; 986 pbp->bio_children--; 987 break; 988 } 989 } 990 g_destroy_bio(bp); 991 992 switch (pbp->bio_cmd) { 993 case BIO_READ: 994 if (pbp->bio_inbed < pbp->bio_children) 995 break; 996 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 997 g_io_deliver(pbp, pbp->bio_error); 998 else { 999 pbp->bio_error = 0; 1000 mtx_lock(&sc->sc_queue_mtx); 1001 bioq_insert_tail(&sc->sc_queue, pbp); 1002 mtx_unlock(&sc->sc_queue_mtx); 1003 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1004 wakeup(sc); 1005 } 1006 break; 1007 case BIO_DELETE: 1008 case BIO_WRITE: 1009 if (pbp->bio_children == 0) { 1010 /* 1011 * All requests failed. 1012 */ 1013 } else if (pbp->bio_inbed < pbp->bio_children) { 1014 /* Do nothing. */ 1015 break; 1016 } else if (pbp->bio_children == pbp->bio_inbed) { 1017 /* Some requests succeeded. */ 1018 pbp->bio_error = 0; 1019 pbp->bio_completed = pbp->bio_length; 1020 } 1021 bioq_remove(&sc->sc_inflight, pbp); 1022 /* Release delayed sync requests if possible. */ 1023 g_mirror_sync_release(sc); 1024 g_io_deliver(pbp, pbp->bio_error); 1025 break; 1026 default: 1027 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 1028 break; 1029 } 1030} 1031 1032static void 1033g_mirror_sync_done(struct bio *bp) 1034{ 1035 struct g_mirror_softc *sc; 1036 1037 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1038 sc = bp->bio_from->geom->softc; 1039 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; 1040 mtx_lock(&sc->sc_queue_mtx); 1041 bioq_insert_tail(&sc->sc_queue, bp); 1042 mtx_unlock(&sc->sc_queue_mtx); 1043 wakeup(sc); 1044} 1045 1046static void 1047g_mirror_candelete(struct bio *bp) 1048{ 1049 struct g_mirror_softc *sc; 1050 struct g_mirror_disk *disk; 1051 int *val; 1052 1053 sc = bp->bio_to->private; 1054 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1055 if (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) 1056 break; 1057 } 1058 val = (int *)bp->bio_data; 1059 *val = (disk != NULL); 1060 g_io_deliver(bp, 0); 1061} 1062 1063static void 1064g_mirror_kernel_dump(struct bio *bp) 1065{ 1066 struct g_mirror_softc *sc; 1067 struct g_mirror_disk *disk; 1068 struct bio *cbp; 1069 struct g_kerneldump *gkd; 1070 1071 /* 1072 * We configure dumping to the first component, because this component 1073 * will be used for reading with 'prefer' balance algorithm. 1074 * If the component with the highest priority is currently disconnected 1075 * we will not be able to read the dump after the reboot if it will be 1076 * connected and synchronized later. Can we do something better? 1077 */ 1078 sc = bp->bio_to->private; 1079 disk = LIST_FIRST(&sc->sc_disks); 1080 1081 gkd = (struct g_kerneldump *)bp->bio_data; 1082 if (gkd->length > bp->bio_to->mediasize) 1083 gkd->length = bp->bio_to->mediasize; 1084 cbp = g_clone_bio(bp); 1085 if (cbp == NULL) { 1086 g_io_deliver(bp, ENOMEM); 1087 return; 1088 } 1089 cbp->bio_done = g_std_done; 1090 g_io_request(cbp, disk->d_consumer); 1091 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1092 g_mirror_get_diskname(disk)); 1093} 1094 1095static void 1096g_mirror_flush(struct g_mirror_softc *sc, struct bio *bp) 1097{ 1098 struct bio_queue_head queue; 1099 struct g_mirror_disk *disk; 1100 struct g_consumer *cp; 1101 struct bio *cbp; 1102 1103 bioq_init(&queue); 1104 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1105 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1106 continue; 1107 cbp = g_clone_bio(bp); 1108 if (cbp == NULL) { 1109 while ((cbp = bioq_takefirst(&queue)) != NULL) 1110 g_destroy_bio(cbp); 1111 if (bp->bio_error == 0) 1112 bp->bio_error = ENOMEM; 1113 g_io_deliver(bp, bp->bio_error); 1114 return; 1115 } 1116 bioq_insert_tail(&queue, cbp); 1117 cbp->bio_done = g_mirror_flush_done; 1118 cbp->bio_caller1 = disk; 1119 cbp->bio_to = disk->d_consumer->provider; 1120 } 1121 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1122 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1123 disk = cbp->bio_caller1; 1124 cbp->bio_caller1 = NULL; 1125 cp = disk->d_consumer; 1126 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1127 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1128 cp->acr, cp->acw, cp->ace)); 1129 g_io_request(cbp, disk->d_consumer); 1130 } 1131} 1132 1133static void 1134g_mirror_start(struct bio *bp) 1135{ 1136 struct g_mirror_softc *sc; 1137 1138 sc = bp->bio_to->private; 1139 /* 1140 * If sc == NULL or there are no valid disks, provider's error 1141 * should be set and g_mirror_start() should not be called at all. 1142 */ 1143 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1144 ("Provider's error should be set (error=%d)(mirror=%s).", 1145 bp->bio_to->error, bp->bio_to->name)); 1146 G_MIRROR_LOGREQ(3, bp, "Request received."); 1147 1148 switch (bp->bio_cmd) { 1149 case BIO_READ: 1150 case BIO_WRITE: 1151 case BIO_DELETE: 1152 break; 1153 case BIO_FLUSH: 1154 g_mirror_flush(sc, bp); 1155 return; 1156 case BIO_GETATTR: 1157 if (!strcmp(bp->bio_attribute, "GEOM::candelete")) { 1158 g_mirror_candelete(bp); 1159 return; 1160 } else if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1161 g_mirror_kernel_dump(bp); 1162 return; 1163 } 1164 /* FALLTHROUGH */ 1165 default: 1166 g_io_deliver(bp, EOPNOTSUPP); 1167 return; 1168 } 1169 mtx_lock(&sc->sc_queue_mtx); 1170 bioq_insert_tail(&sc->sc_queue, bp); 1171 mtx_unlock(&sc->sc_queue_mtx); 1172 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1173 wakeup(sc); 1174} 1175 1176/* 1177 * Return TRUE if the given request is colliding with a in-progress 1178 * synchronization request. 1179 */ 1180static int 1181g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1182{ 1183 struct g_mirror_disk *disk; 1184 struct bio *sbp; 1185 off_t rstart, rend, sstart, send; 1186 u_int i; 1187 1188 if (sc->sc_sync.ds_ndisks == 0) 1189 return (0); 1190 rstart = bp->bio_offset; 1191 rend = bp->bio_offset + bp->bio_length; 1192 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1193 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1194 continue; 1195 for (i = 0; i < g_mirror_syncreqs; i++) { 1196 sbp = disk->d_sync.ds_bios[i]; 1197 if (sbp == NULL) 1198 continue; 1199 sstart = sbp->bio_offset; 1200 send = sbp->bio_offset + sbp->bio_length; 1201 if (rend > sstart && rstart < send) 1202 return (1); 1203 } 1204 } 1205 return (0); 1206} 1207 1208/* 1209 * Return TRUE if the given sync request is colliding with a in-progress regular 1210 * request. 1211 */ 1212static int 1213g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1214{ 1215 off_t rstart, rend, sstart, send; 1216 struct bio *bp; 1217 1218 if (sc->sc_sync.ds_ndisks == 0) 1219 return (0); 1220 sstart = sbp->bio_offset; 1221 send = sbp->bio_offset + sbp->bio_length; 1222 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { 1223 rstart = bp->bio_offset; 1224 rend = bp->bio_offset + bp->bio_length; 1225 if (rend > sstart && rstart < send) 1226 return (1); 1227 } 1228 return (0); 1229} 1230 1231/* 1232 * Puts request onto delayed queue. 1233 */ 1234static void 1235g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1236{ 1237 1238 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1239 bioq_insert_head(&sc->sc_regular_delayed, bp); 1240} 1241 1242/* 1243 * Puts synchronization request onto delayed queue. 1244 */ 1245static void 1246g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1247{ 1248 1249 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1250 bioq_insert_tail(&sc->sc_sync_delayed, bp); 1251} 1252 1253/* 1254 * Releases delayed regular requests which don't collide anymore with sync 1255 * requests. 1256 */ 1257static void 1258g_mirror_regular_release(struct g_mirror_softc *sc) 1259{ 1260 struct bio *bp, *bp2; 1261 1262 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { 1263 if (g_mirror_sync_collision(sc, bp)) 1264 continue; 1265 bioq_remove(&sc->sc_regular_delayed, bp); 1266 G_MIRROR_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1267 mtx_lock(&sc->sc_queue_mtx); 1268 bioq_insert_head(&sc->sc_queue, bp); 1269#if 0 1270 /* 1271 * wakeup() is not needed, because this function is called from 1272 * the worker thread. 1273 */ 1274 wakeup(&sc->sc_queue); 1275#endif 1276 mtx_unlock(&sc->sc_queue_mtx); 1277 } 1278} 1279 1280/* 1281 * Releases delayed sync requests which don't collide anymore with regular 1282 * requests. 1283 */ 1284static void 1285g_mirror_sync_release(struct g_mirror_softc *sc) 1286{ 1287 struct bio *bp, *bp2; 1288 1289 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { 1290 if (g_mirror_regular_collision(sc, bp)) 1291 continue; 1292 bioq_remove(&sc->sc_sync_delayed, bp); 1293 G_MIRROR_LOGREQ(2, bp, 1294 "Releasing delayed synchronization request."); 1295 g_io_request(bp, bp->bio_from); 1296 } 1297} 1298 1299/* 1300 * Handle synchronization requests. 1301 * Every synchronization request is two-steps process: first, READ request is 1302 * send to active provider and then WRITE request (with read data) to the provider 1303 * being synchronized. When WRITE is finished, new synchronization request is 1304 * send. 1305 */ 1306static void 1307g_mirror_sync_request(struct bio *bp) 1308{ 1309 struct g_mirror_softc *sc; 1310 struct g_mirror_disk *disk; 1311 1312 bp->bio_from->index--; 1313 sc = bp->bio_from->geom->softc; 1314 disk = bp->bio_from->private; 1315 if (disk == NULL) { 1316 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1317 g_topology_lock(); 1318 g_mirror_kill_consumer(sc, bp->bio_from); 1319 g_topology_unlock(); 1320 free(bp->bio_data, M_MIRROR); 1321 g_destroy_bio(bp); 1322 sx_xlock(&sc->sc_lock); 1323 return; 1324 } 1325 1326 /* 1327 * Synchronization request. 1328 */ 1329 switch (bp->bio_cmd) { 1330 case BIO_READ: 1331 { 1332 struct g_consumer *cp; 1333 1334 if (bp->bio_error != 0) { 1335 G_MIRROR_LOGREQ(0, bp, 1336 "Synchronization request failed (error=%d).", 1337 bp->bio_error); 1338 g_destroy_bio(bp); 1339 return; 1340 } 1341 G_MIRROR_LOGREQ(3, bp, 1342 "Synchronization request half-finished."); 1343 bp->bio_cmd = BIO_WRITE; 1344 bp->bio_cflags = 0; 1345 cp = disk->d_consumer; 1346 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1347 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1348 cp->acr, cp->acw, cp->ace)); 1349 cp->index++; 1350 g_io_request(bp, cp); 1351 return; 1352 } 1353 case BIO_WRITE: 1354 { 1355 struct g_mirror_disk_sync *sync; 1356 off_t offset; 1357 void *data; 1358 int i; 1359 1360 if (bp->bio_error != 0) { 1361 G_MIRROR_LOGREQ(0, bp, 1362 "Synchronization request failed (error=%d).", 1363 bp->bio_error); 1364 g_destroy_bio(bp); 1365 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1366 g_mirror_event_send(disk, 1367 G_MIRROR_DISK_STATE_DISCONNECTED, 1368 G_MIRROR_EVENT_DONTWAIT); 1369 return; 1370 } 1371 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1372 sync = &disk->d_sync; 1373 if (sync->ds_offset >= sc->sc_mediasize || 1374 sync->ds_consumer == NULL || 1375 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1376 /* Don't send more synchronization requests. */ 1377 sync->ds_inflight--; 1378 if (sync->ds_bios != NULL) { 1379 i = (int)(uintptr_t)bp->bio_caller1; 1380 sync->ds_bios[i] = NULL; 1381 } 1382 free(bp->bio_data, M_MIRROR); 1383 g_destroy_bio(bp); 1384 if (sync->ds_inflight > 0) 1385 return; 1386 if (sync->ds_consumer == NULL || 1387 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1388 return; 1389 } 1390 /* Disk up-to-date, activate it. */ 1391 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1392 G_MIRROR_EVENT_DONTWAIT); 1393 return; 1394 } 1395 1396 /* Send next synchronization request. */ 1397 data = bp->bio_data; 1398 g_reset_bio(bp); 1399 bp->bio_cmd = BIO_READ; 1400 bp->bio_offset = sync->ds_offset; 1401 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1402 sync->ds_offset += bp->bio_length; 1403 bp->bio_done = g_mirror_sync_done; 1404 bp->bio_data = data; 1405 bp->bio_from = sync->ds_consumer; 1406 bp->bio_to = sc->sc_provider; 1407 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1408 sync->ds_consumer->index++; 1409 /* 1410 * Delay the request if it is colliding with a regular request. 1411 */ 1412 if (g_mirror_regular_collision(sc, bp)) 1413 g_mirror_sync_delay(sc, bp); 1414 else 1415 g_io_request(bp, sync->ds_consumer); 1416 1417 /* Release delayed requests if possible. */ 1418 g_mirror_regular_release(sc); 1419 1420 /* Find the smallest offset */ 1421 offset = sc->sc_mediasize; 1422 for (i = 0; i < g_mirror_syncreqs; i++) { 1423 bp = sync->ds_bios[i]; 1424 if (bp->bio_offset < offset) 1425 offset = bp->bio_offset; 1426 } 1427 if (sync->ds_offset_done + (MAXPHYS * 100) < offset) { 1428 /* Update offset_done on every 100 blocks. */ 1429 sync->ds_offset_done = offset; 1430 g_mirror_update_metadata(disk); 1431 } 1432 return; 1433 } 1434 default: 1435 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1436 bp->bio_cmd, sc->sc_name)); 1437 break; 1438 } 1439} 1440 1441static void 1442g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1443{ 1444 struct g_mirror_disk *disk; 1445 struct g_consumer *cp; 1446 struct bio *cbp; 1447 1448 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1449 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1450 break; 1451 } 1452 if (disk == NULL) { 1453 if (bp->bio_error == 0) 1454 bp->bio_error = ENXIO; 1455 g_io_deliver(bp, bp->bio_error); 1456 return; 1457 } 1458 cbp = g_clone_bio(bp); 1459 if (cbp == NULL) { 1460 if (bp->bio_error == 0) 1461 bp->bio_error = ENOMEM; 1462 g_io_deliver(bp, bp->bio_error); 1463 return; 1464 } 1465 /* 1466 * Fill in the component buf structure. 1467 */ 1468 cp = disk->d_consumer; 1469 cbp->bio_done = g_mirror_done; 1470 cbp->bio_to = cp->provider; 1471 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1472 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1473 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1474 cp->acw, cp->ace)); 1475 cp->index++; 1476 g_io_request(cbp, cp); 1477} 1478 1479static void 1480g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1481{ 1482 struct g_mirror_disk *disk; 1483 struct g_consumer *cp; 1484 struct bio *cbp; 1485 1486 disk = g_mirror_get_disk(sc); 1487 if (disk == NULL) { 1488 if (bp->bio_error == 0) 1489 bp->bio_error = ENXIO; 1490 g_io_deliver(bp, bp->bio_error); 1491 return; 1492 } 1493 cbp = g_clone_bio(bp); 1494 if (cbp == NULL) { 1495 if (bp->bio_error == 0) 1496 bp->bio_error = ENOMEM; 1497 g_io_deliver(bp, bp->bio_error); 1498 return; 1499 } 1500 /* 1501 * Fill in the component buf structure. 1502 */ 1503 cp = disk->d_consumer; 1504 cbp->bio_done = g_mirror_done; 1505 cbp->bio_to = cp->provider; 1506 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1507 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1508 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1509 cp->acw, cp->ace)); 1510 cp->index++; 1511 g_io_request(cbp, cp); 1512} 1513 1514#define TRACK_SIZE (1 * 1024 * 1024) 1515#define LOAD_SCALE 256 1516#define ABS(x) (((x) >= 0) ? (x) : (-(x))) 1517 1518static void 1519g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1520{ 1521 struct g_mirror_disk *disk, *dp; 1522 struct g_consumer *cp; 1523 struct bio *cbp; 1524 int prio, best; 1525 1526 /* Find a disk with the smallest load. */ 1527 disk = NULL; 1528 best = INT_MAX; 1529 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1530 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1531 continue; 1532 prio = dp->load; 1533 /* If disk head is precisely in position - highly prefer it. */ 1534 if (dp->d_last_offset == bp->bio_offset) 1535 prio -= 2 * LOAD_SCALE; 1536 else 1537 /* If disk head is close to position - prefer it. */ 1538 if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE) 1539 prio -= 1 * LOAD_SCALE; 1540 if (prio <= best) { 1541 disk = dp; 1542 best = prio; 1543 } 1544 } 1545 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1546 cbp = g_clone_bio(bp); 1547 if (cbp == NULL) { 1548 if (bp->bio_error == 0) 1549 bp->bio_error = ENOMEM; 1550 g_io_deliver(bp, bp->bio_error); 1551 return; 1552 } 1553 /* 1554 * Fill in the component buf structure. 1555 */ 1556 cp = disk->d_consumer; 1557 cbp->bio_done = g_mirror_done; 1558 cbp->bio_to = cp->provider; 1559 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1560 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1561 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1562 cp->acw, cp->ace)); 1563 cp->index++; 1564 /* Remember last head position */ 1565 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1566 /* Update loads. */ 1567 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1568 dp->load = (dp->d_consumer->index * LOAD_SCALE + 1569 dp->load * 7) / 8; 1570 } 1571 g_io_request(cbp, cp); 1572} 1573 1574static void 1575g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1576{ 1577 struct bio_queue_head queue; 1578 struct g_mirror_disk *disk; 1579 struct g_consumer *cp; 1580 struct bio *cbp; 1581 off_t left, mod, offset, slice; 1582 u_char *data; 1583 u_int ndisks; 1584 1585 if (bp->bio_length <= sc->sc_slice) { 1586 g_mirror_request_round_robin(sc, bp); 1587 return; 1588 } 1589 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1590 slice = bp->bio_length / ndisks; 1591 mod = slice % sc->sc_provider->sectorsize; 1592 if (mod != 0) 1593 slice += sc->sc_provider->sectorsize - mod; 1594 /* 1595 * Allocate all bios before sending any request, so we can 1596 * return ENOMEM in nice and clean way. 1597 */ 1598 left = bp->bio_length; 1599 offset = bp->bio_offset; 1600 data = bp->bio_data; 1601 bioq_init(&queue); 1602 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1603 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1604 continue; 1605 cbp = g_clone_bio(bp); 1606 if (cbp == NULL) { 1607 while ((cbp = bioq_takefirst(&queue)) != NULL) 1608 g_destroy_bio(cbp); 1609 if (bp->bio_error == 0) 1610 bp->bio_error = ENOMEM; 1611 g_io_deliver(bp, bp->bio_error); 1612 return; 1613 } 1614 bioq_insert_tail(&queue, cbp); 1615 cbp->bio_done = g_mirror_done; 1616 cbp->bio_caller1 = disk; 1617 cbp->bio_to = disk->d_consumer->provider; 1618 cbp->bio_offset = offset; 1619 cbp->bio_data = data; 1620 cbp->bio_length = MIN(left, slice); 1621 left -= cbp->bio_length; 1622 if (left == 0) 1623 break; 1624 offset += cbp->bio_length; 1625 data += cbp->bio_length; 1626 } 1627 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1628 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1629 disk = cbp->bio_caller1; 1630 cbp->bio_caller1 = NULL; 1631 cp = disk->d_consumer; 1632 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1633 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1634 cp->acr, cp->acw, cp->ace)); 1635 disk->d_consumer->index++; 1636 g_io_request(cbp, disk->d_consumer); 1637 } 1638} 1639 1640static void 1641g_mirror_register_request(struct bio *bp) 1642{ 1643 struct g_mirror_softc *sc; 1644 1645 sc = bp->bio_to->private; 1646 switch (bp->bio_cmd) { 1647 case BIO_READ: 1648 switch (sc->sc_balance) { 1649 case G_MIRROR_BALANCE_LOAD: 1650 g_mirror_request_load(sc, bp); 1651 break; 1652 case G_MIRROR_BALANCE_PREFER: 1653 g_mirror_request_prefer(sc, bp); 1654 break; 1655 case G_MIRROR_BALANCE_ROUND_ROBIN: 1656 g_mirror_request_round_robin(sc, bp); 1657 break; 1658 case G_MIRROR_BALANCE_SPLIT: 1659 g_mirror_request_split(sc, bp); 1660 break; 1661 } 1662 return; 1663 case BIO_WRITE: 1664 case BIO_DELETE: 1665 { 1666 struct g_mirror_disk *disk; 1667 struct g_mirror_disk_sync *sync; 1668 struct bio_queue_head queue; 1669 struct g_consumer *cp; 1670 struct bio *cbp; 1671 1672 /* 1673 * Delay the request if it is colliding with a synchronization 1674 * request. 1675 */ 1676 if (g_mirror_sync_collision(sc, bp)) { 1677 g_mirror_regular_delay(sc, bp); 1678 return; 1679 } 1680 1681 if (sc->sc_idle) 1682 g_mirror_unidle(sc); 1683 else 1684 sc->sc_last_write = time_uptime; 1685 1686 /* 1687 * Allocate all bios before sending any request, so we can 1688 * return ENOMEM in nice and clean way. 1689 */ 1690 bioq_init(&queue); 1691 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1692 sync = &disk->d_sync; 1693 switch (disk->d_state) { 1694 case G_MIRROR_DISK_STATE_ACTIVE: 1695 break; 1696 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1697 if (bp->bio_offset >= sync->ds_offset) 1698 continue; 1699 break; 1700 default: 1701 continue; 1702 } 1703 if (bp->bio_cmd == BIO_DELETE && 1704 (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0) 1705 continue; 1706 cbp = g_clone_bio(bp); 1707 if (cbp == NULL) { 1708 while ((cbp = bioq_takefirst(&queue)) != NULL) 1709 g_destroy_bio(cbp); 1710 if (bp->bio_error == 0) 1711 bp->bio_error = ENOMEM; 1712 g_io_deliver(bp, bp->bio_error); 1713 return; 1714 } 1715 bioq_insert_tail(&queue, cbp); 1716 cbp->bio_done = g_mirror_done; 1717 cp = disk->d_consumer; 1718 cbp->bio_caller1 = cp; 1719 cbp->bio_to = cp->provider; 1720 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1721 ("Consumer %s not opened (r%dw%de%d).", 1722 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1723 } 1724 if (bioq_first(&queue) == NULL) { 1725 g_io_deliver(bp, EOPNOTSUPP); 1726 return; 1727 } 1728 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1729 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1730 cp = cbp->bio_caller1; 1731 cbp->bio_caller1 = NULL; 1732 cp->index++; 1733 sc->sc_writes++; 1734 g_io_request(cbp, cp); 1735 } 1736 /* 1737 * Put request onto inflight queue, so we can check if new 1738 * synchronization requests don't collide with it. 1739 */ 1740 bioq_insert_tail(&sc->sc_inflight, bp); 1741 /* 1742 * Bump syncid on first write. 1743 */ 1744 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1745 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1746 g_mirror_bump_syncid(sc); 1747 } 1748 return; 1749 } 1750 default: 1751 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1752 bp->bio_cmd, sc->sc_name)); 1753 break; 1754 } 1755} 1756 1757static int 1758g_mirror_can_destroy(struct g_mirror_softc *sc) 1759{ 1760 struct g_geom *gp; 1761 struct g_consumer *cp; 1762 1763 g_topology_assert(); 1764 gp = sc->sc_geom; 1765 if (gp->softc == NULL) 1766 return (1); 1767 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0) 1768 return (0); 1769 LIST_FOREACH(cp, &gp->consumer, consumer) { 1770 if (g_mirror_is_busy(sc, cp)) 1771 return (0); 1772 } 1773 gp = sc->sc_sync.ds_geom; 1774 LIST_FOREACH(cp, &gp->consumer, consumer) { 1775 if (g_mirror_is_busy(sc, cp)) 1776 return (0); 1777 } 1778 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1779 sc->sc_name); 1780 return (1); 1781} 1782 1783static int 1784g_mirror_try_destroy(struct g_mirror_softc *sc) 1785{ 1786 1787 if (sc->sc_rootmount != NULL) { 1788 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1789 sc->sc_rootmount); 1790 root_mount_rel(sc->sc_rootmount); 1791 sc->sc_rootmount = NULL; 1792 } 1793 g_topology_lock(); 1794 if (!g_mirror_can_destroy(sc)) { 1795 g_topology_unlock(); 1796 return (0); 1797 } 1798 sc->sc_geom->softc = NULL; 1799 sc->sc_sync.ds_geom->softc = NULL; 1800 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1801 g_topology_unlock(); 1802 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1803 &sc->sc_worker); 1804 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1805 sx_xunlock(&sc->sc_lock); 1806 wakeup(&sc->sc_worker); 1807 sc->sc_worker = NULL; 1808 } else { 1809 g_topology_unlock(); 1810 g_mirror_destroy_device(sc); 1811 } 1812 return (1); 1813} 1814 1815/* 1816 * Worker thread. 1817 */ 1818static void 1819g_mirror_worker(void *arg) 1820{ 1821 struct g_mirror_softc *sc; 1822 struct g_mirror_event *ep; 1823 struct bio *bp; 1824 int timeout; 1825 1826 sc = arg; 1827 thread_lock(curthread); 1828 sched_prio(curthread, PRIBIO); 1829 thread_unlock(curthread); 1830 1831 sx_xlock(&sc->sc_lock); 1832 for (;;) { 1833 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1834 /* 1835 * First take a look at events. 1836 * This is important to handle events before any I/O requests. 1837 */ 1838 ep = g_mirror_event_get(sc); 1839 if (ep != NULL) { 1840 g_mirror_event_remove(sc, ep); 1841 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1842 /* Update only device status. */ 1843 G_MIRROR_DEBUG(3, 1844 "Running event for device %s.", 1845 sc->sc_name); 1846 ep->e_error = 0; 1847 g_mirror_update_device(sc, true); 1848 } else { 1849 /* Update disk status. */ 1850 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1851 g_mirror_get_diskname(ep->e_disk)); 1852 ep->e_error = g_mirror_update_disk(ep->e_disk, 1853 ep->e_state); 1854 if (ep->e_error == 0) 1855 g_mirror_update_device(sc, false); 1856 } 1857 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1858 KASSERT(ep->e_error == 0, 1859 ("Error cannot be handled.")); 1860 g_mirror_event_free(ep); 1861 } else { 1862 ep->e_flags |= G_MIRROR_EVENT_DONE; 1863 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1864 ep); 1865 mtx_lock(&sc->sc_events_mtx); 1866 wakeup(ep); 1867 mtx_unlock(&sc->sc_events_mtx); 1868 } 1869 if ((sc->sc_flags & 1870 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1871 if (g_mirror_try_destroy(sc)) { 1872 curthread->td_pflags &= ~TDP_GEOM; 1873 G_MIRROR_DEBUG(1, "Thread exiting."); 1874 kproc_exit(0); 1875 } 1876 } 1877 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1878 continue; 1879 } 1880 /* 1881 * Check if we can mark array as CLEAN and if we can't take 1882 * how much seconds should we wait. 1883 */ 1884 timeout = g_mirror_idle(sc, -1); 1885 /* 1886 * Now I/O requests. 1887 */ 1888 /* Get first request from the queue. */ 1889 mtx_lock(&sc->sc_queue_mtx); 1890 bp = bioq_takefirst(&sc->sc_queue); 1891 if (bp == NULL) { 1892 if ((sc->sc_flags & 1893 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1894 mtx_unlock(&sc->sc_queue_mtx); 1895 if (g_mirror_try_destroy(sc)) { 1896 curthread->td_pflags &= ~TDP_GEOM; 1897 G_MIRROR_DEBUG(1, "Thread exiting."); 1898 kproc_exit(0); 1899 } 1900 mtx_lock(&sc->sc_queue_mtx); 1901 if (bioq_first(&sc->sc_queue) != NULL) { 1902 mtx_unlock(&sc->sc_queue_mtx); 1903 continue; 1904 } 1905 } 1906 sx_xunlock(&sc->sc_lock); 1907 /* 1908 * XXX: We can miss an event here, because an event 1909 * can be added without sx-device-lock and without 1910 * mtx-queue-lock. Maybe I should just stop using 1911 * dedicated mutex for events synchronization and 1912 * stick with the queue lock? 1913 * The event will hang here until next I/O request 1914 * or next event is received. 1915 */ 1916 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1917 timeout * hz); 1918 sx_xlock(&sc->sc_lock); 1919 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1920 continue; 1921 } 1922 mtx_unlock(&sc->sc_queue_mtx); 1923 1924 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 1925 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1926 g_mirror_sync_request(bp); /* READ */ 1927 } else if (bp->bio_to != sc->sc_provider) { 1928 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 1929 g_mirror_regular_request(bp); 1930 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 1931 g_mirror_sync_request(bp); /* WRITE */ 1932 else { 1933 KASSERT(0, 1934 ("Invalid request cflags=0x%hx to=%s.", 1935 bp->bio_cflags, bp->bio_to->name)); 1936 } 1937 } else { 1938 g_mirror_register_request(bp); 1939 } 1940 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1941 } 1942} 1943 1944static void 1945g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 1946{ 1947 1948 sx_assert(&sc->sc_lock, SX_LOCKED); 1949 1950 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 1951 return; 1952 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1953 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1954 g_mirror_get_diskname(disk), sc->sc_name); 1955 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1956 } else if (sc->sc_idle && 1957 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1958 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1959 g_mirror_get_diskname(disk), sc->sc_name); 1960 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1961 } 1962} 1963 1964static void 1965g_mirror_sync_start(struct g_mirror_disk *disk) 1966{ 1967 struct g_mirror_softc *sc; 1968 struct g_consumer *cp; 1969 struct bio *bp; 1970 int error, i; 1971 1972 g_topology_assert_not(); 1973 sc = disk->d_softc; 1974 sx_assert(&sc->sc_lock, SX_LOCKED); 1975 1976 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1977 ("Disk %s is not marked for synchronization.", 1978 g_mirror_get_diskname(disk))); 1979 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1980 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1981 sc->sc_state)); 1982 1983 sx_xunlock(&sc->sc_lock); 1984 g_topology_lock(); 1985 cp = g_new_consumer(sc->sc_sync.ds_geom); 1986 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 1987 error = g_attach(cp, sc->sc_provider); 1988 KASSERT(error == 0, 1989 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 1990 error = g_access(cp, 1, 0, 0); 1991 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 1992 g_topology_unlock(); 1993 sx_xlock(&sc->sc_lock); 1994 1995 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1996 g_mirror_get_diskname(disk)); 1997 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0) 1998 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1999 KASSERT(disk->d_sync.ds_consumer == NULL, 2000 ("Sync consumer already exists (device=%s, disk=%s).", 2001 sc->sc_name, g_mirror_get_diskname(disk))); 2002 2003 disk->d_sync.ds_consumer = cp; 2004 disk->d_sync.ds_consumer->private = disk; 2005 disk->d_sync.ds_consumer->index = 0; 2006 2007 /* 2008 * Allocate memory for synchronization bios and initialize them. 2009 */ 2010 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 2011 M_MIRROR, M_WAITOK); 2012 for (i = 0; i < g_mirror_syncreqs; i++) { 2013 bp = g_alloc_bio(); 2014 disk->d_sync.ds_bios[i] = bp; 2015 bp->bio_parent = NULL; 2016 bp->bio_cmd = BIO_READ; 2017 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 2018 bp->bio_cflags = 0; 2019 bp->bio_offset = disk->d_sync.ds_offset; 2020 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 2021 disk->d_sync.ds_offset += bp->bio_length; 2022 bp->bio_done = g_mirror_sync_done; 2023 bp->bio_from = disk->d_sync.ds_consumer; 2024 bp->bio_to = sc->sc_provider; 2025 bp->bio_caller1 = (void *)(uintptr_t)i; 2026 } 2027 2028 /* Increase the number of disks in SYNCHRONIZING state. */ 2029 sc->sc_sync.ds_ndisks++; 2030 /* Set the number of in-flight synchronization requests. */ 2031 disk->d_sync.ds_inflight = g_mirror_syncreqs; 2032 2033 /* 2034 * Fire off first synchronization requests. 2035 */ 2036 for (i = 0; i < g_mirror_syncreqs; i++) { 2037 bp = disk->d_sync.ds_bios[i]; 2038 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 2039 disk->d_sync.ds_consumer->index++; 2040 /* 2041 * Delay the request if it is colliding with a regular request. 2042 */ 2043 if (g_mirror_regular_collision(sc, bp)) 2044 g_mirror_sync_delay(sc, bp); 2045 else 2046 g_io_request(bp, disk->d_sync.ds_consumer); 2047 } 2048} 2049 2050/* 2051 * Stop synchronization process. 2052 * type: 0 - synchronization finished 2053 * 1 - synchronization stopped 2054 */ 2055static void 2056g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 2057{ 2058 struct g_mirror_softc *sc; 2059 struct g_consumer *cp; 2060 2061 g_topology_assert_not(); 2062 sc = disk->d_softc; 2063 sx_assert(&sc->sc_lock, SX_LOCKED); 2064 2065 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2066 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2067 g_mirror_disk_state2str(disk->d_state))); 2068 if (disk->d_sync.ds_consumer == NULL) 2069 return; 2070 2071 if (type == 0) { 2072 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2073 sc->sc_name, g_mirror_get_diskname(disk)); 2074 } else /* if (type == 1) */ { 2075 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2076 sc->sc_name, g_mirror_get_diskname(disk)); 2077 } 2078 free(disk->d_sync.ds_bios, M_MIRROR); 2079 disk->d_sync.ds_bios = NULL; 2080 cp = disk->d_sync.ds_consumer; 2081 disk->d_sync.ds_consumer = NULL; 2082 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2083 sc->sc_sync.ds_ndisks--; 2084 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2085 g_topology_lock(); 2086 g_mirror_kill_consumer(sc, cp); 2087 g_topology_unlock(); 2088 sx_xlock(&sc->sc_lock); 2089} 2090 2091static void 2092g_mirror_launch_provider(struct g_mirror_softc *sc) 2093{ 2094 struct g_mirror_disk *disk; 2095 struct g_provider *pp, *dp; 2096 2097 sx_assert(&sc->sc_lock, SX_LOCKED); 2098 2099 g_topology_lock(); 2100 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 2101 pp->flags |= G_PF_DIRECT_RECEIVE; 2102 pp->mediasize = sc->sc_mediasize; 2103 pp->sectorsize = sc->sc_sectorsize; 2104 pp->stripesize = 0; 2105 pp->stripeoffset = 0; 2106 2107 /* Splitting of unmapped BIO's could work but isn't implemented now */ 2108 if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT) 2109 pp->flags |= G_PF_ACCEPT_UNMAPPED; 2110 2111 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2112 if (disk->d_consumer && disk->d_consumer->provider) { 2113 dp = disk->d_consumer->provider; 2114 if (dp->stripesize > pp->stripesize) { 2115 pp->stripesize = dp->stripesize; 2116 pp->stripeoffset = dp->stripeoffset; 2117 } 2118 /* A provider underneath us doesn't support unmapped */ 2119 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { 2120 G_MIRROR_DEBUG(0, "Cancelling unmapped " 2121 "because of %s.", dp->name); 2122 pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 2123 } 2124 } 2125 } 2126 pp->private = sc; 2127 sc->sc_refcnt++; 2128 sc->sc_provider = pp; 2129 g_error_provider(pp, 0); 2130 g_topology_unlock(); 2131 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2132 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 2133 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2134 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2135 g_mirror_sync_start(disk); 2136 } 2137} 2138 2139static void 2140g_mirror_destroy_provider(struct g_mirror_softc *sc) 2141{ 2142 struct g_mirror_disk *disk; 2143 struct bio *bp; 2144 2145 g_topology_assert_not(); 2146 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2147 sc->sc_name)); 2148 2149 g_topology_lock(); 2150 g_error_provider(sc->sc_provider, ENXIO); 2151 mtx_lock(&sc->sc_queue_mtx); 2152 while ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) { 2153 /* 2154 * Abort any pending I/O that wasn't generated by us. 2155 * Synchronization requests and requests destined for individual 2156 * mirror components can be destroyed immediately. 2157 */ 2158 if (bp->bio_to == sc->sc_provider && 2159 bp->bio_from->geom != sc->sc_sync.ds_geom) { 2160 g_io_deliver(bp, ENXIO); 2161 } else { 2162 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2163 free(bp->bio_data, M_MIRROR); 2164 g_destroy_bio(bp); 2165 } 2166 } 2167 mtx_unlock(&sc->sc_queue_mtx); 2168 g_wither_provider(sc->sc_provider, ENXIO); 2169 sc->sc_provider = NULL; 2170 G_MIRROR_DEBUG(0, "Device %s: provider destroyed.", sc->sc_name); 2171 g_topology_unlock(); 2172 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2173 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2174 g_mirror_sync_stop(disk, 1); 2175 } 2176} 2177 2178static void 2179g_mirror_go(void *arg) 2180{ 2181 struct g_mirror_softc *sc; 2182 2183 sc = arg; 2184 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2185 g_mirror_event_send(sc, 0, 2186 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2187} 2188 2189static u_int 2190g_mirror_determine_state(struct g_mirror_disk *disk) 2191{ 2192 struct g_mirror_softc *sc; 2193 u_int state; 2194 2195 sc = disk->d_softc; 2196 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2197 if ((disk->d_flags & 2198 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0 && 2199 (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 || 2200 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0)) { 2201 /* Disk does not need synchronization. */ 2202 state = G_MIRROR_DISK_STATE_ACTIVE; 2203 } else { 2204 if ((sc->sc_flags & 2205 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2206 (disk->d_flags & 2207 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2208 /* 2209 * We can start synchronization from 2210 * the stored offset. 2211 */ 2212 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2213 } else { 2214 state = G_MIRROR_DISK_STATE_STALE; 2215 } 2216 } 2217 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2218 /* 2219 * Reset all synchronization data for this disk, 2220 * because if it even was synchronized, it was 2221 * synchronized to disks with different syncid. 2222 */ 2223 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2224 disk->d_sync.ds_offset = 0; 2225 disk->d_sync.ds_offset_done = 0; 2226 disk->d_sync.ds_syncid = sc->sc_syncid; 2227 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2228 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2229 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2230 } else { 2231 state = G_MIRROR_DISK_STATE_STALE; 2232 } 2233 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2234 /* 2235 * Not good, NOT GOOD! 2236 * It means that mirror was started on stale disks 2237 * and more fresh disk just arrive. 2238 * If there were writes, mirror is broken, sorry. 2239 * I think the best choice here is don't touch 2240 * this disk and inform the user loudly. 2241 */ 2242 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2243 "disk (%s) arrives!! It will not be connected to the " 2244 "running device.", sc->sc_name, 2245 g_mirror_get_diskname(disk)); 2246 g_mirror_destroy_disk(disk); 2247 state = G_MIRROR_DISK_STATE_NONE; 2248 /* Return immediately, because disk was destroyed. */ 2249 return (state); 2250 } 2251 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2252 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2253 return (state); 2254} 2255 2256/* 2257 * Update device state. 2258 */ 2259static void 2260g_mirror_update_device(struct g_mirror_softc *sc, bool force) 2261{ 2262 struct g_mirror_disk *disk; 2263 u_int state; 2264 2265 sx_assert(&sc->sc_lock, SX_XLOCKED); 2266 2267 switch (sc->sc_state) { 2268 case G_MIRROR_DEVICE_STATE_STARTING: 2269 { 2270 struct g_mirror_disk *pdisk, *tdisk; 2271 u_int dirty, ndisks, genid, syncid; 2272 bool broken; 2273 2274 KASSERT(sc->sc_provider == NULL, 2275 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2276 /* 2277 * Are we ready? We are, if all disks are connected or 2278 * if we have any disks and 'force' is true. 2279 */ 2280 ndisks = g_mirror_ndisks(sc, -1); 2281 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2282 ; 2283 } else if (ndisks == 0) { 2284 /* 2285 * Disks went down in starting phase, so destroy 2286 * device. 2287 */ 2288 callout_drain(&sc->sc_callout); 2289 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2290 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2291 sc->sc_rootmount); 2292 root_mount_rel(sc->sc_rootmount); 2293 sc->sc_rootmount = NULL; 2294 return; 2295 } else { 2296 return; 2297 } 2298 2299 /* 2300 * Activate all disks with the biggest syncid. 2301 */ 2302 if (force) { 2303 /* 2304 * If 'force' is true, we have been called due to 2305 * timeout, so don't bother canceling timeout. 2306 */ 2307 ndisks = 0; 2308 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2309 if ((disk->d_flags & 2310 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2311 ndisks++; 2312 } 2313 } 2314 if (ndisks == 0) { 2315 /* No valid disks found, destroy device. */ 2316 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2317 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2318 __LINE__, sc->sc_rootmount); 2319 root_mount_rel(sc->sc_rootmount); 2320 sc->sc_rootmount = NULL; 2321 return; 2322 } 2323 } else { 2324 /* Cancel timeout. */ 2325 callout_drain(&sc->sc_callout); 2326 } 2327 2328 /* 2329 * Find the biggest genid. 2330 */ 2331 genid = 0; 2332 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2333 if (disk->d_genid > genid) 2334 genid = disk->d_genid; 2335 } 2336 sc->sc_genid = genid; 2337 /* 2338 * Remove all disks without the biggest genid. 2339 */ 2340 broken = false; 2341 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2342 if (disk->d_genid < genid) { 2343 G_MIRROR_DEBUG(0, 2344 "Component %s (device %s) broken, skipping.", 2345 g_mirror_get_diskname(disk), sc->sc_name); 2346 g_mirror_destroy_disk(disk); 2347 /* 2348 * Bump the syncid in case we discover a healthy 2349 * replacement disk after starting the mirror. 2350 */ 2351 broken = true; 2352 } 2353 } 2354 2355 /* 2356 * Find the biggest syncid. 2357 */ 2358 syncid = 0; 2359 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2360 if (disk->d_sync.ds_syncid > syncid) 2361 syncid = disk->d_sync.ds_syncid; 2362 } 2363 2364 /* 2365 * Here we need to look for dirty disks and if all disks 2366 * with the biggest syncid are dirty, we have to choose 2367 * one with the biggest priority and rebuild the rest. 2368 */ 2369 /* 2370 * Find the number of dirty disks with the biggest syncid. 2371 * Find the number of disks with the biggest syncid. 2372 * While here, find a disk with the biggest priority. 2373 */ 2374 dirty = ndisks = 0; 2375 pdisk = NULL; 2376 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2377 if (disk->d_sync.ds_syncid != syncid) 2378 continue; 2379 if ((disk->d_flags & 2380 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2381 continue; 2382 } 2383 ndisks++; 2384 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2385 dirty++; 2386 if (pdisk == NULL || 2387 pdisk->d_priority < disk->d_priority) { 2388 pdisk = disk; 2389 } 2390 } 2391 } 2392 if (dirty == 0) { 2393 /* No dirty disks at all, great. */ 2394 } else if (dirty == ndisks) { 2395 /* 2396 * Force synchronization for all dirty disks except one 2397 * with the biggest priority. 2398 */ 2399 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2400 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2401 "master disk for synchronization.", 2402 g_mirror_get_diskname(pdisk), sc->sc_name); 2403 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2404 if (disk->d_sync.ds_syncid != syncid) 2405 continue; 2406 if ((disk->d_flags & 2407 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2408 continue; 2409 } 2410 KASSERT((disk->d_flags & 2411 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2412 ("Disk %s isn't marked as dirty.", 2413 g_mirror_get_diskname(disk))); 2414 /* Skip the disk with the biggest priority. */ 2415 if (disk == pdisk) 2416 continue; 2417 disk->d_sync.ds_syncid = 0; 2418 } 2419 } else if (dirty < ndisks) { 2420 /* 2421 * Force synchronization for all dirty disks. 2422 * We have some non-dirty disks. 2423 */ 2424 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2425 if (disk->d_sync.ds_syncid != syncid) 2426 continue; 2427 if ((disk->d_flags & 2428 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2429 continue; 2430 } 2431 if ((disk->d_flags & 2432 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2433 continue; 2434 } 2435 disk->d_sync.ds_syncid = 0; 2436 } 2437 } 2438 2439 /* Reset hint. */ 2440 sc->sc_hint = NULL; 2441 sc->sc_syncid = syncid; 2442 if (force || broken) { 2443 /* Remember to bump syncid on first write. */ 2444 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2445 } 2446 state = G_MIRROR_DEVICE_STATE_RUNNING; 2447 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2448 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2449 g_mirror_device_state2str(state)); 2450 sc->sc_state = state; 2451 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2452 state = g_mirror_determine_state(disk); 2453 g_mirror_event_send(disk, state, 2454 G_MIRROR_EVENT_DONTWAIT); 2455 if (state == G_MIRROR_DISK_STATE_STALE) 2456 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2457 } 2458 break; 2459 } 2460 case G_MIRROR_DEVICE_STATE_RUNNING: 2461 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2462 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2463 /* 2464 * No active disks or no disks at all, 2465 * so destroy device. 2466 */ 2467 if (sc->sc_provider != NULL) 2468 g_mirror_destroy_provider(sc); 2469 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2470 break; 2471 } else if (g_mirror_ndisks(sc, 2472 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2473 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2474 /* 2475 * We have active disks, launch provider if it doesn't 2476 * exist. 2477 */ 2478 if (sc->sc_provider == NULL) 2479 g_mirror_launch_provider(sc); 2480 if (sc->sc_rootmount != NULL) { 2481 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2482 __LINE__, sc->sc_rootmount); 2483 root_mount_rel(sc->sc_rootmount); 2484 sc->sc_rootmount = NULL; 2485 } 2486 } 2487 /* 2488 * Genid should be bumped immediately, so do it here. 2489 */ 2490 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2491 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2492 g_mirror_bump_genid(sc); 2493 } 2494 break; 2495 default: 2496 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2497 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2498 break; 2499 } 2500} 2501 2502/* 2503 * Update disk state and device state if needed. 2504 */ 2505#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2506 "Disk %s state changed from %s to %s (device %s).", \ 2507 g_mirror_get_diskname(disk), \ 2508 g_mirror_disk_state2str(disk->d_state), \ 2509 g_mirror_disk_state2str(state), sc->sc_name) 2510static int 2511g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2512{ 2513 struct g_mirror_softc *sc; 2514 2515 sc = disk->d_softc; 2516 sx_assert(&sc->sc_lock, SX_XLOCKED); 2517 2518again: 2519 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2520 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2521 g_mirror_disk_state2str(state)); 2522 switch (state) { 2523 case G_MIRROR_DISK_STATE_NEW: 2524 /* 2525 * Possible scenarios: 2526 * 1. New disk arrive. 2527 */ 2528 /* Previous state should be NONE. */ 2529 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2530 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2531 g_mirror_disk_state2str(disk->d_state))); 2532 DISK_STATE_CHANGED(); 2533 2534 disk->d_state = state; 2535 if (LIST_EMPTY(&sc->sc_disks)) 2536 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2537 else { 2538 struct g_mirror_disk *dp; 2539 2540 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2541 if (disk->d_priority >= dp->d_priority) { 2542 LIST_INSERT_BEFORE(dp, disk, d_next); 2543 dp = NULL; 2544 break; 2545 } 2546 if (LIST_NEXT(dp, d_next) == NULL) 2547 break; 2548 } 2549 if (dp != NULL) 2550 LIST_INSERT_AFTER(dp, disk, d_next); 2551 } 2552 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2553 sc->sc_name, g_mirror_get_diskname(disk)); 2554 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2555 break; 2556 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2557 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2558 g_mirror_device_state2str(sc->sc_state), 2559 g_mirror_get_diskname(disk), 2560 g_mirror_disk_state2str(disk->d_state))); 2561 state = g_mirror_determine_state(disk); 2562 if (state != G_MIRROR_DISK_STATE_NONE) 2563 goto again; 2564 break; 2565 case G_MIRROR_DISK_STATE_ACTIVE: 2566 /* 2567 * Possible scenarios: 2568 * 1. New disk does not need synchronization. 2569 * 2. Synchronization process finished successfully. 2570 */ 2571 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2572 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2573 g_mirror_device_state2str(sc->sc_state), 2574 g_mirror_get_diskname(disk), 2575 g_mirror_disk_state2str(disk->d_state))); 2576 /* Previous state should be NEW or SYNCHRONIZING. */ 2577 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2578 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2579 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2580 g_mirror_disk_state2str(disk->d_state))); 2581 DISK_STATE_CHANGED(); 2582 2583 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2584 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2585 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2586 g_mirror_sync_stop(disk, 0); 2587 } 2588 disk->d_state = state; 2589 disk->d_sync.ds_offset = 0; 2590 disk->d_sync.ds_offset_done = 0; 2591 g_mirror_update_idle(sc, disk); 2592 g_mirror_update_metadata(disk); 2593 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2594 sc->sc_name, g_mirror_get_diskname(disk)); 2595 break; 2596 case G_MIRROR_DISK_STATE_STALE: 2597 /* 2598 * Possible scenarios: 2599 * 1. Stale disk was connected. 2600 */ 2601 /* Previous state should be NEW. */ 2602 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2603 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2604 g_mirror_disk_state2str(disk->d_state))); 2605 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2606 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2607 g_mirror_device_state2str(sc->sc_state), 2608 g_mirror_get_diskname(disk), 2609 g_mirror_disk_state2str(disk->d_state))); 2610 /* 2611 * STALE state is only possible if device is marked 2612 * NOAUTOSYNC. 2613 */ 2614 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2615 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2616 g_mirror_device_state2str(sc->sc_state), 2617 g_mirror_get_diskname(disk), 2618 g_mirror_disk_state2str(disk->d_state))); 2619 DISK_STATE_CHANGED(); 2620 2621 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2622 disk->d_state = state; 2623 g_mirror_update_metadata(disk); 2624 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2625 sc->sc_name, g_mirror_get_diskname(disk)); 2626 break; 2627 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2628 /* 2629 * Possible scenarios: 2630 * 1. Disk which needs synchronization was connected. 2631 */ 2632 /* Previous state should be NEW. */ 2633 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2634 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2635 g_mirror_disk_state2str(disk->d_state))); 2636 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2637 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2638 g_mirror_device_state2str(sc->sc_state), 2639 g_mirror_get_diskname(disk), 2640 g_mirror_disk_state2str(disk->d_state))); 2641 DISK_STATE_CHANGED(); 2642 2643 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2644 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2645 disk->d_state = state; 2646 if (sc->sc_provider != NULL) { 2647 g_mirror_sync_start(disk); 2648 g_mirror_update_metadata(disk); 2649 } 2650 break; 2651 case G_MIRROR_DISK_STATE_DISCONNECTED: 2652 /* 2653 * Possible scenarios: 2654 * 1. Device wasn't running yet, but disk disappear. 2655 * 2. Disk was active and disapppear. 2656 * 3. Disk disappear during synchronization process. 2657 */ 2658 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2659 /* 2660 * Previous state should be ACTIVE, STALE or 2661 * SYNCHRONIZING. 2662 */ 2663 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2664 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2665 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2666 ("Wrong disk state (%s, %s).", 2667 g_mirror_get_diskname(disk), 2668 g_mirror_disk_state2str(disk->d_state))); 2669 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2670 /* Previous state should be NEW. */ 2671 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2672 ("Wrong disk state (%s, %s).", 2673 g_mirror_get_diskname(disk), 2674 g_mirror_disk_state2str(disk->d_state))); 2675 /* 2676 * Reset bumping syncid if disk disappeared in STARTING 2677 * state. 2678 */ 2679 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2680 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2681#ifdef INVARIANTS 2682 } else { 2683 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2684 sc->sc_name, 2685 g_mirror_device_state2str(sc->sc_state), 2686 g_mirror_get_diskname(disk), 2687 g_mirror_disk_state2str(disk->d_state))); 2688#endif 2689 } 2690 DISK_STATE_CHANGED(); 2691 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2692 sc->sc_name, g_mirror_get_diskname(disk)); 2693 2694 g_mirror_destroy_disk(disk); 2695 break; 2696 case G_MIRROR_DISK_STATE_DESTROY: 2697 { 2698 int error; 2699 2700 error = g_mirror_clear_metadata(disk); 2701 if (error != 0) { 2702 G_MIRROR_DEBUG(0, 2703 "Device %s: failed to clear metadata on %s: %d.", 2704 sc->sc_name, g_mirror_get_diskname(disk), error); 2705 break; 2706 } 2707 DISK_STATE_CHANGED(); 2708 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2709 sc->sc_name, g_mirror_get_diskname(disk)); 2710 2711 g_mirror_destroy_disk(disk); 2712 sc->sc_ndisks--; 2713 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2714 g_mirror_update_metadata(disk); 2715 } 2716 break; 2717 } 2718 default: 2719 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2720 break; 2721 } 2722 return (0); 2723} 2724#undef DISK_STATE_CHANGED 2725 2726int 2727g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2728{ 2729 struct g_provider *pp; 2730 u_char *buf; 2731 int error; 2732 2733 g_topology_assert(); 2734 2735 error = g_access(cp, 1, 0, 0); 2736 if (error != 0) 2737 return (error); 2738 pp = cp->provider; 2739 g_topology_unlock(); 2740 /* Metadata are stored on last sector. */ 2741 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2742 &error); 2743 g_topology_lock(); 2744 g_access(cp, -1, 0, 0); 2745 if (buf == NULL) { 2746 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2747 cp->provider->name, error); 2748 return (error); 2749 } 2750 2751 /* Decode metadata. */ 2752 error = mirror_metadata_decode(buf, md); 2753 g_free(buf); 2754 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2755 return (EINVAL); 2756 if (md->md_version > G_MIRROR_VERSION) { 2757 G_MIRROR_DEBUG(0, 2758 "Kernel module is too old to handle metadata from %s.", 2759 cp->provider->name); 2760 return (EINVAL); 2761 } 2762 if (error != 0) { 2763 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2764 cp->provider->name); 2765 return (error); 2766 } 2767 2768 return (0); 2769} 2770 2771static int 2772g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2773 struct g_mirror_metadata *md) 2774{ 2775 2776 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2777 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2778 pp->name, md->md_did); 2779 return (EEXIST); 2780 } 2781 if (md->md_all != sc->sc_ndisks) { 2782 G_MIRROR_DEBUG(1, 2783 "Invalid '%s' field on disk %s (device %s), skipping.", 2784 "md_all", pp->name, sc->sc_name); 2785 return (EINVAL); 2786 } 2787 if (md->md_slice != sc->sc_slice) { 2788 G_MIRROR_DEBUG(1, 2789 "Invalid '%s' field on disk %s (device %s), skipping.", 2790 "md_slice", pp->name, sc->sc_name); 2791 return (EINVAL); 2792 } 2793 if (md->md_balance != sc->sc_balance) { 2794 G_MIRROR_DEBUG(1, 2795 "Invalid '%s' field on disk %s (device %s), skipping.", 2796 "md_balance", pp->name, sc->sc_name); 2797 return (EINVAL); 2798 } 2799#if 0 2800 if (md->md_mediasize != sc->sc_mediasize) { 2801 G_MIRROR_DEBUG(1, 2802 "Invalid '%s' field on disk %s (device %s), skipping.", 2803 "md_mediasize", pp->name, sc->sc_name); 2804 return (EINVAL); 2805 } 2806#endif 2807 if (sc->sc_mediasize > pp->mediasize) { 2808 G_MIRROR_DEBUG(1, 2809 "Invalid size of disk %s (device %s), skipping.", pp->name, 2810 sc->sc_name); 2811 return (EINVAL); 2812 } 2813 if (md->md_sectorsize != sc->sc_sectorsize) { 2814 G_MIRROR_DEBUG(1, 2815 "Invalid '%s' field on disk %s (device %s), skipping.", 2816 "md_sectorsize", pp->name, sc->sc_name); 2817 return (EINVAL); 2818 } 2819 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2820 G_MIRROR_DEBUG(1, 2821 "Invalid sector size of disk %s (device %s), skipping.", 2822 pp->name, sc->sc_name); 2823 return (EINVAL); 2824 } 2825 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2826 G_MIRROR_DEBUG(1, 2827 "Invalid device flags on disk %s (device %s), skipping.", 2828 pp->name, sc->sc_name); 2829 return (EINVAL); 2830 } 2831 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2832 G_MIRROR_DEBUG(1, 2833 "Invalid disk flags on disk %s (device %s), skipping.", 2834 pp->name, sc->sc_name); 2835 return (EINVAL); 2836 } 2837 return (0); 2838} 2839 2840int 2841g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2842 struct g_mirror_metadata *md) 2843{ 2844 struct g_mirror_disk *disk; 2845 int error; 2846 2847 g_topology_assert_not(); 2848 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2849 2850 error = g_mirror_check_metadata(sc, pp, md); 2851 if (error != 0) 2852 return (error); 2853 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2854 md->md_genid < sc->sc_genid) { 2855 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2856 pp->name, sc->sc_name); 2857 return (EINVAL); 2858 } 2859 disk = g_mirror_init_disk(sc, pp, md, &error); 2860 if (disk == NULL) 2861 return (error); 2862 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2863 G_MIRROR_EVENT_WAIT); 2864 if (error != 0) 2865 return (error); 2866 if (md->md_version < G_MIRROR_VERSION) { 2867 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2868 pp->name, md->md_version, G_MIRROR_VERSION); 2869 g_mirror_update_metadata(disk); 2870 } 2871 return (0); 2872} 2873 2874static void 2875g_mirror_destroy_delayed(void *arg, int flag) 2876{ 2877 struct g_mirror_softc *sc; 2878 int error; 2879 2880 if (flag == EV_CANCEL) { 2881 G_MIRROR_DEBUG(1, "Destroying canceled."); 2882 return; 2883 } 2884 sc = arg; 2885 g_topology_unlock(); 2886 sx_xlock(&sc->sc_lock); 2887 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2888 ("DESTROY flag set on %s.", sc->sc_name)); 2889 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0, 2890 ("DESTROYING flag not set on %s.", sc->sc_name)); 2891 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 2892 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 2893 if (error != 0) { 2894 G_MIRROR_DEBUG(0, "Cannot destroy %s (error=%d).", 2895 sc->sc_name, error); 2896 sx_xunlock(&sc->sc_lock); 2897 } 2898 g_topology_lock(); 2899} 2900 2901static int 2902g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2903{ 2904 struct g_mirror_softc *sc; 2905 int error = 0; 2906 2907 g_topology_assert(); 2908 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2909 acw, ace); 2910 2911 sc = pp->private; 2912 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 2913 2914 g_topology_unlock(); 2915 sx_xlock(&sc->sc_lock); 2916 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 2917 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0 || 2918 LIST_EMPTY(&sc->sc_disks)) { 2919 if (acr > 0 || acw > 0 || ace > 0) 2920 error = ENXIO; 2921 goto end; 2922 } 2923 sc->sc_provider_open += acr + acw + ace; 2924 if (pp->acw + acw == 0) 2925 g_mirror_idle(sc, 0); 2926 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0 && 2927 sc->sc_provider_open == 0) 2928 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, sc, NULL); 2929end: 2930 sx_xunlock(&sc->sc_lock); 2931 g_topology_lock(); 2932 return (error); 2933} 2934 2935struct g_geom * 2936g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md, 2937 u_int type) 2938{ 2939 struct g_mirror_softc *sc; 2940 struct g_geom *gp; 2941 int error, timeout; 2942 2943 g_topology_assert(); 2944 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2945 md->md_mid); 2946 2947 /* One disk is minimum. */ 2948 if (md->md_all < 1) 2949 return (NULL); 2950 /* 2951 * Action geom. 2952 */ 2953 gp = g_new_geomf(mp, "%s", md->md_name); 2954 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2955 gp->start = g_mirror_start; 2956 gp->orphan = g_mirror_orphan; 2957 gp->access = g_mirror_access; 2958 gp->dumpconf = g_mirror_dumpconf; 2959 2960 sc->sc_type = type; 2961 sc->sc_id = md->md_mid; 2962 sc->sc_slice = md->md_slice; 2963 sc->sc_balance = md->md_balance; 2964 sc->sc_mediasize = md->md_mediasize; 2965 sc->sc_sectorsize = md->md_sectorsize; 2966 sc->sc_ndisks = md->md_all; 2967 sc->sc_flags = md->md_mflags; 2968 sc->sc_bump_id = 0; 2969 sc->sc_idle = 1; 2970 sc->sc_last_write = time_uptime; 2971 sc->sc_writes = 0; 2972 sc->sc_refcnt = 1; 2973 sx_init(&sc->sc_lock, "gmirror:lock"); 2974 bioq_init(&sc->sc_queue); 2975 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2976 bioq_init(&sc->sc_regular_delayed); 2977 bioq_init(&sc->sc_inflight); 2978 bioq_init(&sc->sc_sync_delayed); 2979 LIST_INIT(&sc->sc_disks); 2980 TAILQ_INIT(&sc->sc_events); 2981 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2982 callout_init(&sc->sc_callout, 1); 2983 mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF); 2984 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2985 gp->softc = sc; 2986 sc->sc_geom = gp; 2987 sc->sc_provider = NULL; 2988 sc->sc_provider_open = 0; 2989 /* 2990 * Synchronization geom. 2991 */ 2992 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2993 gp->softc = sc; 2994 gp->orphan = g_mirror_orphan; 2995 sc->sc_sync.ds_geom = gp; 2996 sc->sc_sync.ds_ndisks = 0; 2997 error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2998 "g_mirror %s", md->md_name); 2999 if (error != 0) { 3000 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 3001 sc->sc_name); 3002 g_destroy_geom(sc->sc_sync.ds_geom); 3003 g_destroy_geom(sc->sc_geom); 3004 g_mirror_free_device(sc); 3005 return (NULL); 3006 } 3007 3008 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 3009 sc->sc_name, sc->sc_ndisks, sc->sc_id); 3010 3011 sc->sc_rootmount = root_mount_hold("GMIRROR"); 3012 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 3013 /* 3014 * Run timeout. 3015 */ 3016 timeout = g_mirror_timeout * hz; 3017 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 3018 return (sc->sc_geom); 3019} 3020 3021int 3022g_mirror_destroy(struct g_mirror_softc *sc, int how) 3023{ 3024 struct g_mirror_disk *disk; 3025 3026 g_topology_assert_not(); 3027 sx_assert(&sc->sc_lock, SX_XLOCKED); 3028 3029 if (sc->sc_provider_open != 0) { 3030 switch (how) { 3031 case G_MIRROR_DESTROY_SOFT: 3032 G_MIRROR_DEBUG(1, 3033 "Device %s is still open (%d).", sc->sc_name, 3034 sc->sc_provider_open); 3035 return (EBUSY); 3036 case G_MIRROR_DESTROY_DELAYED: 3037 G_MIRROR_DEBUG(1, 3038 "Device %s will be destroyed on last close.", 3039 sc->sc_name); 3040 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 3041 if (disk->d_state == 3042 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3043 g_mirror_sync_stop(disk, 1); 3044 } 3045 } 3046 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROYING; 3047 return (EBUSY); 3048 case G_MIRROR_DESTROY_HARD: 3049 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 3050 "can't be definitely removed.", sc->sc_name); 3051 } 3052 } 3053 3054 g_topology_lock(); 3055 if (sc->sc_geom->softc == NULL) { 3056 g_topology_unlock(); 3057 return (0); 3058 } 3059 sc->sc_geom->softc = NULL; 3060 sc->sc_sync.ds_geom->softc = NULL; 3061 g_topology_unlock(); 3062 3063 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 3064 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 3065 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3066 sx_xunlock(&sc->sc_lock); 3067 mtx_lock(&sc->sc_queue_mtx); 3068 wakeup(sc); 3069 mtx_unlock(&sc->sc_queue_mtx); 3070 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3071 while (sc->sc_worker != NULL) 3072 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 3073 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3074 sx_xlock(&sc->sc_lock); 3075 g_mirror_destroy_device(sc); 3076 return (0); 3077} 3078 3079static void 3080g_mirror_taste_orphan(struct g_consumer *cp) 3081{ 3082 3083 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3084 cp->provider->name)); 3085} 3086 3087static struct g_geom * 3088g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3089{ 3090 struct g_mirror_metadata md; 3091 struct g_mirror_softc *sc; 3092 struct g_consumer *cp; 3093 struct g_geom *gp; 3094 int error; 3095 3096 g_topology_assert(); 3097 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3098 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 3099 3100 gp = g_new_geomf(mp, "mirror:taste"); 3101 /* 3102 * This orphan function should be never called. 3103 */ 3104 gp->orphan = g_mirror_taste_orphan; 3105 cp = g_new_consumer(gp); 3106 g_attach(cp, pp); 3107 error = g_mirror_read_metadata(cp, &md); 3108 g_detach(cp); 3109 g_destroy_consumer(cp); 3110 g_destroy_geom(gp); 3111 if (error != 0) 3112 return (NULL); 3113 gp = NULL; 3114 3115 if (md.md_provider[0] != '\0' && 3116 !g_compare_names(md.md_provider, pp->name)) 3117 return (NULL); 3118 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3119 return (NULL); 3120 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 3121 G_MIRROR_DEBUG(0, 3122 "Device %s: provider %s marked as inactive, skipping.", 3123 md.md_name, pp->name); 3124 return (NULL); 3125 } 3126 if (g_mirror_debug >= 2) 3127 mirror_metadata_dump(&md); 3128 3129 /* 3130 * Let's check if device already exists. 3131 */ 3132 sc = NULL; 3133 LIST_FOREACH(gp, &mp->geom, geom) { 3134 sc = gp->softc; 3135 if (sc == NULL) 3136 continue; 3137 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 3138 continue; 3139 if (sc->sc_sync.ds_geom == gp) 3140 continue; 3141 if (strcmp(md.md_name, sc->sc_name) != 0) 3142 continue; 3143 if (md.md_mid != sc->sc_id) { 3144 G_MIRROR_DEBUG(0, "Device %s already configured.", 3145 sc->sc_name); 3146 return (NULL); 3147 } 3148 break; 3149 } 3150 if (gp == NULL) { 3151 gp = g_mirror_create(mp, &md, G_MIRROR_TYPE_AUTOMATIC); 3152 if (gp == NULL) { 3153 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3154 md.md_name); 3155 return (NULL); 3156 } 3157 sc = gp->softc; 3158 } 3159 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3160 g_topology_unlock(); 3161 sx_xlock(&sc->sc_lock); 3162 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; 3163 error = g_mirror_add_disk(sc, pp, &md); 3164 if (error != 0) { 3165 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3166 pp->name, gp->name, error); 3167 if (LIST_EMPTY(&sc->sc_disks)) { 3168 g_cancel_event(sc); 3169 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3170 g_topology_lock(); 3171 return (NULL); 3172 } 3173 gp = NULL; 3174 } 3175 sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; 3176 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3177 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3178 g_topology_lock(); 3179 return (NULL); 3180 } 3181 sx_xunlock(&sc->sc_lock); 3182 g_topology_lock(); 3183 return (gp); 3184} 3185 3186static void 3187g_mirror_resize(struct g_consumer *cp) 3188{ 3189 struct g_mirror_disk *disk; 3190 3191 g_topology_assert(); 3192 g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name); 3193 3194 disk = cp->private; 3195 if (disk == NULL) 3196 return; 3197 g_topology_unlock(); 3198 g_mirror_update_metadata(disk); 3199 g_topology_lock(); 3200} 3201 3202static int 3203g_mirror_destroy_geom(struct gctl_req *req __unused, 3204 struct g_class *mp __unused, struct g_geom *gp) 3205{ 3206 struct g_mirror_softc *sc; 3207 int error; 3208 3209 g_topology_unlock(); 3210 sc = gp->softc; 3211 sx_xlock(&sc->sc_lock); 3212 g_cancel_event(sc); 3213 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3214 if (error != 0) 3215 sx_xunlock(&sc->sc_lock); 3216 g_topology_lock(); 3217 return (error); 3218} 3219 3220static void 3221g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3222 struct g_consumer *cp, struct g_provider *pp) 3223{ 3224 struct g_mirror_softc *sc; 3225 3226 g_topology_assert(); 3227 3228 sc = gp->softc; 3229 if (sc == NULL) 3230 return; 3231 /* Skip synchronization geom. */ 3232 if (gp == sc->sc_sync.ds_geom) 3233 return; 3234 if (pp != NULL) { 3235 /* Nothing here. */ 3236 } else if (cp != NULL) { 3237 struct g_mirror_disk *disk; 3238 3239 disk = cp->private; 3240 if (disk == NULL) 3241 return; 3242 g_topology_unlock(); 3243 sx_xlock(&sc->sc_lock); 3244 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3245 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3246 sbuf_printf(sb, "%s<Synchronized>", indent); 3247 if (disk->d_sync.ds_offset == 0) 3248 sbuf_printf(sb, "0%%"); 3249 else { 3250 sbuf_printf(sb, "%u%%", 3251 (u_int)((disk->d_sync.ds_offset * 100) / 3252 sc->sc_provider->mediasize)); 3253 } 3254 sbuf_printf(sb, "</Synchronized>\n"); 3255 if (disk->d_sync.ds_offset > 0) { 3256 sbuf_printf(sb, "%s<BytesSynced>%jd" 3257 "</BytesSynced>\n", indent, 3258 (intmax_t)disk->d_sync.ds_offset); 3259 } 3260 } 3261 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3262 disk->d_sync.ds_syncid); 3263 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3264 disk->d_genid); 3265 sbuf_printf(sb, "%s<Flags>", indent); 3266 if (disk->d_flags == 0) 3267 sbuf_printf(sb, "NONE"); 3268 else { 3269 int first = 1; 3270 3271#define ADD_FLAG(flag, name) do { \ 3272 if ((disk->d_flags & (flag)) != 0) { \ 3273 if (!first) \ 3274 sbuf_printf(sb, ", "); \ 3275 else \ 3276 first = 0; \ 3277 sbuf_printf(sb, name); \ 3278 } \ 3279} while (0) 3280 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3281 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3282 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3283 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3284 "SYNCHRONIZING"); 3285 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3286 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3287#undef ADD_FLAG 3288 } 3289 sbuf_printf(sb, "</Flags>\n"); 3290 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3291 disk->d_priority); 3292 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3293 g_mirror_disk_state2str(disk->d_state)); 3294 sx_xunlock(&sc->sc_lock); 3295 g_topology_lock(); 3296 } else { 3297 g_topology_unlock(); 3298 sx_xlock(&sc->sc_lock); 3299 sbuf_printf(sb, "%s<Type>", indent); 3300 switch (sc->sc_type) { 3301 case G_MIRROR_TYPE_AUTOMATIC: 3302 sbuf_printf(sb, "AUTOMATIC"); 3303 break; 3304 case G_MIRROR_TYPE_MANUAL: 3305 sbuf_printf(sb, "MANUAL"); 3306 break; 3307 default: 3308 sbuf_printf(sb, "UNKNOWN"); 3309 break; 3310 } 3311 sbuf_printf(sb, "</Type>\n"); 3312 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3313 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3314 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3315 sbuf_printf(sb, "%s<Flags>", indent); 3316 if (sc->sc_flags == 0) 3317 sbuf_printf(sb, "NONE"); 3318 else { 3319 int first = 1; 3320 3321#define ADD_FLAG(flag, name) do { \ 3322 if ((sc->sc_flags & (flag)) != 0) { \ 3323 if (!first) \ 3324 sbuf_printf(sb, ", "); \ 3325 else \ 3326 first = 0; \ 3327 sbuf_printf(sb, name); \ 3328 } \ 3329} while (0) 3330 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3331 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3332#undef ADD_FLAG 3333 } 3334 sbuf_printf(sb, "</Flags>\n"); 3335 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3336 (u_int)sc->sc_slice); 3337 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3338 balance_name(sc->sc_balance)); 3339 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3340 sc->sc_ndisks); 3341 sbuf_printf(sb, "%s<State>", indent); 3342 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3343 sbuf_printf(sb, "%s", "STARTING"); 3344 else if (sc->sc_ndisks == 3345 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3346 sbuf_printf(sb, "%s", "COMPLETE"); 3347 else 3348 sbuf_printf(sb, "%s", "DEGRADED"); 3349 sbuf_printf(sb, "</State>\n"); 3350 sx_xunlock(&sc->sc_lock); 3351 g_topology_lock(); 3352 } 3353} 3354 3355static void 3356g_mirror_shutdown_post_sync(void *arg, int howto) 3357{ 3358 struct g_class *mp; 3359 struct g_geom *gp, *gp2; 3360 struct g_mirror_softc *sc; 3361 int error; 3362 3363 if (panicstr != NULL) 3364 return; 3365 3366 mp = arg; 3367 g_topology_lock(); 3368 g_mirror_shutdown = 1; 3369 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3370 if ((sc = gp->softc) == NULL) 3371 continue; 3372 /* Skip synchronization geom. */ 3373 if (gp == sc->sc_sync.ds_geom) 3374 continue; 3375 g_topology_unlock(); 3376 sx_xlock(&sc->sc_lock); 3377 g_mirror_idle(sc, -1); 3378 g_cancel_event(sc); 3379 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3380 if (error != 0) 3381 sx_xunlock(&sc->sc_lock); 3382 g_topology_lock(); 3383 } 3384 g_topology_unlock(); 3385} 3386 3387static void 3388g_mirror_init(struct g_class *mp) 3389{ 3390 3391 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3392 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3393 if (g_mirror_post_sync == NULL) 3394 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3395} 3396 3397static void 3398g_mirror_fini(struct g_class *mp) 3399{ 3400 3401 if (g_mirror_post_sync != NULL) 3402 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); 3403} 3404 3405DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3406