g_mirror.c revision 306765
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/10/sys/geom/mirror/g_mirror.c 306765 2016-10-06 15:36:13Z mav $"); 29 30#include <sys/param.h> 31#include <sys/systm.h> 32#include <sys/kernel.h> 33#include <sys/module.h> 34#include <sys/limits.h> 35#include <sys/lock.h> 36#include <sys/mutex.h> 37#include <sys/bio.h> 38#include <sys/sbuf.h> 39#include <sys/sysctl.h> 40#include <sys/malloc.h> 41#include <sys/eventhandler.h> 42#include <vm/uma.h> 43#include <geom/geom.h> 44#include <sys/proc.h> 45#include <sys/kthread.h> 46#include <sys/sched.h> 47#include <geom/mirror/g_mirror.h> 48 49FEATURE(geom_mirror, "GEOM mirroring support"); 50 51static MALLOC_DEFINE(M_MIRROR, "mirror_data", "GEOM_MIRROR Data"); 52 53SYSCTL_DECL(_kern_geom); 54static SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, 55 "GEOM_MIRROR stuff"); 56u_int g_mirror_debug = 0; 57TUNABLE_INT("kern.geom.mirror.debug", &g_mirror_debug); 58SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RW, &g_mirror_debug, 0, 59 "Debug level"); 60static u_int g_mirror_timeout = 4; 61TUNABLE_INT("kern.geom.mirror.timeout", &g_mirror_timeout); 62SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RW, &g_mirror_timeout, 63 0, "Time to wait on all mirror components"); 64static u_int g_mirror_idletime = 5; 65TUNABLE_INT("kern.geom.mirror.idletime", &g_mirror_idletime); 66SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RW, 67 &g_mirror_idletime, 0, "Mark components as clean when idling"); 68static u_int g_mirror_disconnect_on_failure = 1; 69TUNABLE_INT("kern.geom.mirror.disconnect_on_failure", 70 &g_mirror_disconnect_on_failure); 71SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, disconnect_on_failure, CTLFLAG_RW, 72 &g_mirror_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 73static u_int g_mirror_syncreqs = 2; 74TUNABLE_INT("kern.geom.mirror.sync_requests", &g_mirror_syncreqs); 75SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_requests, CTLFLAG_RDTUN, 76 &g_mirror_syncreqs, 0, "Parallel synchronization I/O requests."); 77 78#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 79 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 80 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 81 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 82} while (0) 83 84static eventhandler_tag g_mirror_post_sync = NULL; 85static int g_mirror_shutdown = 0; 86 87static int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp, 88 struct g_geom *gp); 89static g_taste_t g_mirror_taste; 90static g_resize_t g_mirror_resize; 91static void g_mirror_init(struct g_class *mp); 92static void g_mirror_fini(struct g_class *mp); 93 94struct g_class g_mirror_class = { 95 .name = G_MIRROR_CLASS_NAME, 96 .version = G_VERSION, 97 .ctlreq = g_mirror_config, 98 .taste = g_mirror_taste, 99 .destroy_geom = g_mirror_destroy_geom, 100 .init = g_mirror_init, 101 .fini = g_mirror_fini, 102 .resize = g_mirror_resize 103}; 104 105 106static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 107static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 108static void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force); 109static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 110 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 111static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 112static void g_mirror_register_request(struct bio *bp); 113static void g_mirror_sync_release(struct g_mirror_softc *sc); 114 115 116static const char * 117g_mirror_disk_state2str(int state) 118{ 119 120 switch (state) { 121 case G_MIRROR_DISK_STATE_NONE: 122 return ("NONE"); 123 case G_MIRROR_DISK_STATE_NEW: 124 return ("NEW"); 125 case G_MIRROR_DISK_STATE_ACTIVE: 126 return ("ACTIVE"); 127 case G_MIRROR_DISK_STATE_STALE: 128 return ("STALE"); 129 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 130 return ("SYNCHRONIZING"); 131 case G_MIRROR_DISK_STATE_DISCONNECTED: 132 return ("DISCONNECTED"); 133 case G_MIRROR_DISK_STATE_DESTROY: 134 return ("DESTROY"); 135 default: 136 return ("INVALID"); 137 } 138} 139 140static const char * 141g_mirror_device_state2str(int state) 142{ 143 144 switch (state) { 145 case G_MIRROR_DEVICE_STATE_STARTING: 146 return ("STARTING"); 147 case G_MIRROR_DEVICE_STATE_RUNNING: 148 return ("RUNNING"); 149 default: 150 return ("INVALID"); 151 } 152} 153 154static const char * 155g_mirror_get_diskname(struct g_mirror_disk *disk) 156{ 157 158 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 159 return ("[unknown]"); 160 return (disk->d_name); 161} 162 163/* 164 * --- Events handling functions --- 165 * Events in geom_mirror are used to maintain disks and device status 166 * from one thread to simplify locking. 167 */ 168static void 169g_mirror_event_free(struct g_mirror_event *ep) 170{ 171 172 free(ep, M_MIRROR); 173} 174 175int 176g_mirror_event_send(void *arg, int state, int flags) 177{ 178 struct g_mirror_softc *sc; 179 struct g_mirror_disk *disk; 180 struct g_mirror_event *ep; 181 int error; 182 183 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 184 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 185 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 186 disk = NULL; 187 sc = arg; 188 } else { 189 disk = arg; 190 sc = disk->d_softc; 191 } 192 ep->e_disk = disk; 193 ep->e_state = state; 194 ep->e_flags = flags; 195 ep->e_error = 0; 196 mtx_lock(&sc->sc_events_mtx); 197 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 198 mtx_unlock(&sc->sc_events_mtx); 199 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 200 mtx_lock(&sc->sc_queue_mtx); 201 wakeup(sc); 202 mtx_unlock(&sc->sc_queue_mtx); 203 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 204 return (0); 205 sx_assert(&sc->sc_lock, SX_XLOCKED); 206 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 207 sx_xunlock(&sc->sc_lock); 208 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 209 mtx_lock(&sc->sc_events_mtx); 210 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 211 hz * 5); 212 } 213 error = ep->e_error; 214 g_mirror_event_free(ep); 215 sx_xlock(&sc->sc_lock); 216 return (error); 217} 218 219static struct g_mirror_event * 220g_mirror_event_get(struct g_mirror_softc *sc) 221{ 222 struct g_mirror_event *ep; 223 224 mtx_lock(&sc->sc_events_mtx); 225 ep = TAILQ_FIRST(&sc->sc_events); 226 mtx_unlock(&sc->sc_events_mtx); 227 return (ep); 228} 229 230static void 231g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 232{ 233 234 mtx_lock(&sc->sc_events_mtx); 235 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 236 mtx_unlock(&sc->sc_events_mtx); 237} 238 239static void 240g_mirror_event_cancel(struct g_mirror_disk *disk) 241{ 242 struct g_mirror_softc *sc; 243 struct g_mirror_event *ep, *tmpep; 244 245 sc = disk->d_softc; 246 sx_assert(&sc->sc_lock, SX_XLOCKED); 247 248 mtx_lock(&sc->sc_events_mtx); 249 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 250 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 251 continue; 252 if (ep->e_disk != disk) 253 continue; 254 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 255 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 256 g_mirror_event_free(ep); 257 else { 258 ep->e_error = ECANCELED; 259 wakeup(ep); 260 } 261 } 262 mtx_unlock(&sc->sc_events_mtx); 263} 264 265/* 266 * Return the number of disks in given state. 267 * If state is equal to -1, count all connected disks. 268 */ 269u_int 270g_mirror_ndisks(struct g_mirror_softc *sc, int state) 271{ 272 struct g_mirror_disk *disk; 273 u_int n = 0; 274 275 sx_assert(&sc->sc_lock, SX_LOCKED); 276 277 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 278 if (state == -1 || disk->d_state == state) 279 n++; 280 } 281 return (n); 282} 283 284/* 285 * Find a disk in mirror by its disk ID. 286 */ 287static struct g_mirror_disk * 288g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 289{ 290 struct g_mirror_disk *disk; 291 292 sx_assert(&sc->sc_lock, SX_XLOCKED); 293 294 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 295 if (disk->d_id == id) 296 return (disk); 297 } 298 return (NULL); 299} 300 301static u_int 302g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 303{ 304 struct bio *bp; 305 u_int nreqs = 0; 306 307 mtx_lock(&sc->sc_queue_mtx); 308 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 309 if (bp->bio_from == cp) 310 nreqs++; 311 } 312 mtx_unlock(&sc->sc_queue_mtx); 313 return (nreqs); 314} 315 316static int 317g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 318{ 319 320 if (cp->index > 0) { 321 G_MIRROR_DEBUG(2, 322 "I/O requests for %s exist, can't destroy it now.", 323 cp->provider->name); 324 return (1); 325 } 326 if (g_mirror_nrequests(sc, cp) > 0) { 327 G_MIRROR_DEBUG(2, 328 "I/O requests for %s in queue, can't destroy it now.", 329 cp->provider->name); 330 return (1); 331 } 332 return (0); 333} 334 335static void 336g_mirror_destroy_consumer(void *arg, int flags __unused) 337{ 338 struct g_consumer *cp; 339 340 g_topology_assert(); 341 342 cp = arg; 343 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 344 g_detach(cp); 345 g_destroy_consumer(cp); 346} 347 348static void 349g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 350{ 351 struct g_provider *pp; 352 int retaste_wait; 353 354 g_topology_assert(); 355 356 cp->private = NULL; 357 if (g_mirror_is_busy(sc, cp)) 358 return; 359 pp = cp->provider; 360 retaste_wait = 0; 361 if (cp->acw == 1) { 362 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 363 retaste_wait = 1; 364 } 365 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 366 -cp->acw, -cp->ace, 0); 367 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 368 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 369 if (retaste_wait) { 370 /* 371 * After retaste event was send (inside g_access()), we can send 372 * event to detach and destroy consumer. 373 * A class, which has consumer to the given provider connected 374 * will not receive retaste event for the provider. 375 * This is the way how I ignore retaste events when I close 376 * consumers opened for write: I detach and destroy consumer 377 * after retaste event is sent. 378 */ 379 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 380 return; 381 } 382 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 383 g_detach(cp); 384 g_destroy_consumer(cp); 385} 386 387static int 388g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 389{ 390 struct g_consumer *cp; 391 int error; 392 393 g_topology_assert_not(); 394 KASSERT(disk->d_consumer == NULL, 395 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 396 397 g_topology_lock(); 398 cp = g_new_consumer(disk->d_softc->sc_geom); 399 cp->flags |= G_CF_DIRECT_RECEIVE; 400 error = g_attach(cp, pp); 401 if (error != 0) { 402 g_destroy_consumer(cp); 403 g_topology_unlock(); 404 return (error); 405 } 406 error = g_access(cp, 1, 1, 1); 407 if (error != 0) { 408 g_detach(cp); 409 g_destroy_consumer(cp); 410 g_topology_unlock(); 411 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 412 pp->name, error); 413 return (error); 414 } 415 g_topology_unlock(); 416 disk->d_consumer = cp; 417 disk->d_consumer->private = disk; 418 disk->d_consumer->index = 0; 419 420 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 421 return (0); 422} 423 424static void 425g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 426{ 427 428 g_topology_assert(); 429 430 if (cp == NULL) 431 return; 432 if (cp->provider != NULL) 433 g_mirror_kill_consumer(sc, cp); 434 else 435 g_destroy_consumer(cp); 436} 437 438/* 439 * Initialize disk. This means allocate memory, create consumer, attach it 440 * to the provider and open access (r1w1e1) to it. 441 */ 442static struct g_mirror_disk * 443g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 444 struct g_mirror_metadata *md, int *errorp) 445{ 446 struct g_mirror_disk *disk; 447 int i, error; 448 449 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 450 if (disk == NULL) { 451 error = ENOMEM; 452 goto fail; 453 } 454 disk->d_softc = sc; 455 error = g_mirror_connect_disk(disk, pp); 456 if (error != 0) 457 goto fail; 458 disk->d_id = md->md_did; 459 disk->d_state = G_MIRROR_DISK_STATE_NONE; 460 disk->d_priority = md->md_priority; 461 disk->d_flags = md->md_dflags; 462 error = g_getattr("GEOM::candelete", disk->d_consumer, &i); 463 if (error == 0 && i != 0) 464 disk->d_flags |= G_MIRROR_DISK_FLAG_CANDELETE; 465 if (md->md_provider[0] != '\0') 466 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 467 disk->d_sync.ds_consumer = NULL; 468 disk->d_sync.ds_offset = md->md_sync_offset; 469 disk->d_sync.ds_offset_done = md->md_sync_offset; 470 disk->d_genid = md->md_genid; 471 disk->d_sync.ds_syncid = md->md_syncid; 472 if (errorp != NULL) 473 *errorp = 0; 474 return (disk); 475fail: 476 if (errorp != NULL) 477 *errorp = error; 478 if (disk != NULL) 479 free(disk, M_MIRROR); 480 return (NULL); 481} 482 483static void 484g_mirror_destroy_disk(struct g_mirror_disk *disk) 485{ 486 struct g_mirror_softc *sc; 487 488 g_topology_assert_not(); 489 sc = disk->d_softc; 490 sx_assert(&sc->sc_lock, SX_XLOCKED); 491 492 LIST_REMOVE(disk, d_next); 493 g_mirror_event_cancel(disk); 494 if (sc->sc_hint == disk) 495 sc->sc_hint = NULL; 496 switch (disk->d_state) { 497 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 498 g_mirror_sync_stop(disk, 1); 499 /* FALLTHROUGH */ 500 case G_MIRROR_DISK_STATE_NEW: 501 case G_MIRROR_DISK_STATE_STALE: 502 case G_MIRROR_DISK_STATE_ACTIVE: 503 g_topology_lock(); 504 g_mirror_disconnect_consumer(sc, disk->d_consumer); 505 g_topology_unlock(); 506 free(disk, M_MIRROR); 507 break; 508 default: 509 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 510 g_mirror_get_diskname(disk), 511 g_mirror_disk_state2str(disk->d_state))); 512 } 513} 514 515static void 516g_mirror_destroy_device(struct g_mirror_softc *sc) 517{ 518 struct g_mirror_disk *disk; 519 struct g_mirror_event *ep; 520 struct g_geom *gp; 521 struct g_consumer *cp, *tmpcp; 522 523 g_topology_assert_not(); 524 sx_assert(&sc->sc_lock, SX_XLOCKED); 525 526 gp = sc->sc_geom; 527 if (sc->sc_provider != NULL) 528 g_mirror_destroy_provider(sc); 529 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 530 disk = LIST_FIRST(&sc->sc_disks)) { 531 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 532 g_mirror_update_metadata(disk); 533 g_mirror_destroy_disk(disk); 534 } 535 while ((ep = g_mirror_event_get(sc)) != NULL) { 536 g_mirror_event_remove(sc, ep); 537 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 538 g_mirror_event_free(ep); 539 else { 540 ep->e_error = ECANCELED; 541 ep->e_flags |= G_MIRROR_EVENT_DONE; 542 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 543 mtx_lock(&sc->sc_events_mtx); 544 wakeup(ep); 545 mtx_unlock(&sc->sc_events_mtx); 546 } 547 } 548 callout_drain(&sc->sc_callout); 549 550 g_topology_lock(); 551 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 552 g_mirror_disconnect_consumer(sc, cp); 553 } 554 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 555 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 556 g_wither_geom(gp, ENXIO); 557 g_topology_unlock(); 558 mtx_destroy(&sc->sc_queue_mtx); 559 mtx_destroy(&sc->sc_events_mtx); 560 mtx_destroy(&sc->sc_done_mtx); 561 sx_xunlock(&sc->sc_lock); 562 sx_destroy(&sc->sc_lock); 563} 564 565static void 566g_mirror_orphan(struct g_consumer *cp) 567{ 568 struct g_mirror_disk *disk; 569 570 g_topology_assert(); 571 572 disk = cp->private; 573 if (disk == NULL) 574 return; 575 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 576 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 577 G_MIRROR_EVENT_DONTWAIT); 578} 579 580/* 581 * Function should return the next active disk on the list. 582 * It is possible that it will be the same disk as given. 583 * If there are no active disks on list, NULL is returned. 584 */ 585static __inline struct g_mirror_disk * 586g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 587{ 588 struct g_mirror_disk *dp; 589 590 for (dp = LIST_NEXT(disk, d_next); dp != disk; 591 dp = LIST_NEXT(dp, d_next)) { 592 if (dp == NULL) 593 dp = LIST_FIRST(&sc->sc_disks); 594 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 595 break; 596 } 597 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 598 return (NULL); 599 return (dp); 600} 601 602static struct g_mirror_disk * 603g_mirror_get_disk(struct g_mirror_softc *sc) 604{ 605 struct g_mirror_disk *disk; 606 607 if (sc->sc_hint == NULL) { 608 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 609 if (sc->sc_hint == NULL) 610 return (NULL); 611 } 612 disk = sc->sc_hint; 613 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 614 disk = g_mirror_find_next(sc, disk); 615 if (disk == NULL) 616 return (NULL); 617 } 618 sc->sc_hint = g_mirror_find_next(sc, disk); 619 return (disk); 620} 621 622static int 623g_mirror_write_metadata(struct g_mirror_disk *disk, 624 struct g_mirror_metadata *md) 625{ 626 struct g_mirror_softc *sc; 627 struct g_consumer *cp; 628 off_t offset, length; 629 u_char *sector; 630 int error = 0; 631 632 g_topology_assert_not(); 633 sc = disk->d_softc; 634 sx_assert(&sc->sc_lock, SX_LOCKED); 635 636 cp = disk->d_consumer; 637 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 638 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 639 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 640 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 641 cp->acw, cp->ace)); 642 length = cp->provider->sectorsize; 643 offset = cp->provider->mediasize - length; 644 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 645 if (md != NULL && 646 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) { 647 /* 648 * Handle the case, when the size of parent provider reduced. 649 */ 650 if (offset < md->md_mediasize) 651 error = ENOSPC; 652 else 653 mirror_metadata_encode(md, sector); 654 } 655 if (error == 0) 656 error = g_write_data(cp, offset, sector, length); 657 free(sector, M_MIRROR); 658 if (error != 0) { 659 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 660 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 661 G_MIRROR_DEBUG(0, "Cannot write metadata on %s " 662 "(device=%s, error=%d).", 663 g_mirror_get_diskname(disk), sc->sc_name, error); 664 } else { 665 G_MIRROR_DEBUG(1, "Cannot write metadata on %s " 666 "(device=%s, error=%d).", 667 g_mirror_get_diskname(disk), sc->sc_name, error); 668 } 669 if (g_mirror_disconnect_on_failure && 670 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 671 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 672 g_mirror_event_send(disk, 673 G_MIRROR_DISK_STATE_DISCONNECTED, 674 G_MIRROR_EVENT_DONTWAIT); 675 } 676 } 677 return (error); 678} 679 680static int 681g_mirror_clear_metadata(struct g_mirror_disk *disk) 682{ 683 int error; 684 685 g_topology_assert_not(); 686 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 687 688 error = g_mirror_write_metadata(disk, NULL); 689 if (error == 0) { 690 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 691 g_mirror_get_diskname(disk)); 692 } else { 693 G_MIRROR_DEBUG(0, 694 "Cannot clear metadata on disk %s (error=%d).", 695 g_mirror_get_diskname(disk), error); 696 } 697 return (error); 698} 699 700void 701g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 702 struct g_mirror_metadata *md) 703{ 704 705 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 706 md->md_version = G_MIRROR_VERSION; 707 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 708 md->md_mid = sc->sc_id; 709 md->md_all = sc->sc_ndisks; 710 md->md_slice = sc->sc_slice; 711 md->md_balance = sc->sc_balance; 712 md->md_genid = sc->sc_genid; 713 md->md_mediasize = sc->sc_mediasize; 714 md->md_sectorsize = sc->sc_sectorsize; 715 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 716 bzero(md->md_provider, sizeof(md->md_provider)); 717 if (disk == NULL) { 718 md->md_did = arc4random(); 719 md->md_priority = 0; 720 md->md_syncid = 0; 721 md->md_dflags = 0; 722 md->md_sync_offset = 0; 723 md->md_provsize = 0; 724 } else { 725 md->md_did = disk->d_id; 726 md->md_priority = disk->d_priority; 727 md->md_syncid = disk->d_sync.ds_syncid; 728 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 729 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 730 md->md_sync_offset = disk->d_sync.ds_offset_done; 731 else 732 md->md_sync_offset = 0; 733 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 734 strlcpy(md->md_provider, 735 disk->d_consumer->provider->name, 736 sizeof(md->md_provider)); 737 } 738 md->md_provsize = disk->d_consumer->provider->mediasize; 739 } 740} 741 742void 743g_mirror_update_metadata(struct g_mirror_disk *disk) 744{ 745 struct g_mirror_softc *sc; 746 struct g_mirror_metadata md; 747 int error; 748 749 g_topology_assert_not(); 750 sc = disk->d_softc; 751 sx_assert(&sc->sc_lock, SX_LOCKED); 752 753 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) 754 g_mirror_fill_metadata(sc, disk, &md); 755 error = g_mirror_write_metadata(disk, &md); 756 if (error == 0) { 757 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 758 g_mirror_get_diskname(disk)); 759 } else { 760 G_MIRROR_DEBUG(0, 761 "Cannot update metadata on disk %s (error=%d).", 762 g_mirror_get_diskname(disk), error); 763 } 764} 765 766static void 767g_mirror_bump_syncid(struct g_mirror_softc *sc) 768{ 769 struct g_mirror_disk *disk; 770 771 g_topology_assert_not(); 772 sx_assert(&sc->sc_lock, SX_XLOCKED); 773 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 774 ("%s called with no active disks (device=%s).", __func__, 775 sc->sc_name)); 776 777 sc->sc_syncid++; 778 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 779 sc->sc_syncid); 780 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 781 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 782 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 783 disk->d_sync.ds_syncid = sc->sc_syncid; 784 g_mirror_update_metadata(disk); 785 } 786 } 787} 788 789static void 790g_mirror_bump_genid(struct g_mirror_softc *sc) 791{ 792 struct g_mirror_disk *disk; 793 794 g_topology_assert_not(); 795 sx_assert(&sc->sc_lock, SX_XLOCKED); 796 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 797 ("%s called with no active disks (device=%s).", __func__, 798 sc->sc_name)); 799 800 sc->sc_genid++; 801 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 802 sc->sc_genid); 803 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 804 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 805 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 806 disk->d_genid = sc->sc_genid; 807 g_mirror_update_metadata(disk); 808 } 809 } 810} 811 812static int 813g_mirror_idle(struct g_mirror_softc *sc, int acw) 814{ 815 struct g_mirror_disk *disk; 816 int timeout; 817 818 g_topology_assert_not(); 819 sx_assert(&sc->sc_lock, SX_XLOCKED); 820 821 if (sc->sc_provider == NULL) 822 return (0); 823 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 824 return (0); 825 if (sc->sc_idle) 826 return (0); 827 if (sc->sc_writes > 0) 828 return (0); 829 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 830 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); 831 if (!g_mirror_shutdown && timeout > 0) 832 return (timeout); 833 } 834 sc->sc_idle = 1; 835 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 836 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 837 continue; 838 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 839 g_mirror_get_diskname(disk), sc->sc_name); 840 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 841 g_mirror_update_metadata(disk); 842 } 843 return (0); 844} 845 846static void 847g_mirror_unidle(struct g_mirror_softc *sc) 848{ 849 struct g_mirror_disk *disk; 850 851 g_topology_assert_not(); 852 sx_assert(&sc->sc_lock, SX_XLOCKED); 853 854 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 855 return; 856 sc->sc_idle = 0; 857 sc->sc_last_write = time_uptime; 858 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 859 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 860 continue; 861 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 862 g_mirror_get_diskname(disk), sc->sc_name); 863 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 864 g_mirror_update_metadata(disk); 865 } 866} 867 868static void 869g_mirror_flush_done(struct bio *bp) 870{ 871 struct g_mirror_softc *sc; 872 struct bio *pbp; 873 874 pbp = bp->bio_parent; 875 sc = pbp->bio_to->geom->softc; 876 mtx_lock(&sc->sc_done_mtx); 877 if (pbp->bio_error == 0) 878 pbp->bio_error = bp->bio_error; 879 pbp->bio_completed += bp->bio_completed; 880 pbp->bio_inbed++; 881 if (pbp->bio_children == pbp->bio_inbed) { 882 mtx_unlock(&sc->sc_done_mtx); 883 g_io_deliver(pbp, pbp->bio_error); 884 } else 885 mtx_unlock(&sc->sc_done_mtx); 886 g_destroy_bio(bp); 887} 888 889static void 890g_mirror_done(struct bio *bp) 891{ 892 struct g_mirror_softc *sc; 893 894 sc = bp->bio_from->geom->softc; 895 bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR; 896 mtx_lock(&sc->sc_queue_mtx); 897 bioq_insert_tail(&sc->sc_queue, bp); 898 mtx_unlock(&sc->sc_queue_mtx); 899 wakeup(sc); 900} 901 902static void 903g_mirror_regular_request(struct bio *bp) 904{ 905 struct g_mirror_softc *sc; 906 struct g_mirror_disk *disk; 907 struct bio *pbp; 908 909 g_topology_assert_not(); 910 911 pbp = bp->bio_parent; 912 sc = pbp->bio_to->geom->softc; 913 bp->bio_from->index--; 914 if (bp->bio_cmd == BIO_WRITE) 915 sc->sc_writes--; 916 disk = bp->bio_from->private; 917 if (disk == NULL) { 918 g_topology_lock(); 919 g_mirror_kill_consumer(sc, bp->bio_from); 920 g_topology_unlock(); 921 } 922 923 pbp->bio_inbed++; 924 KASSERT(pbp->bio_inbed <= pbp->bio_children, 925 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 926 pbp->bio_children)); 927 if (bp->bio_error == 0 && pbp->bio_error == 0) { 928 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 929 g_destroy_bio(bp); 930 if (pbp->bio_children == pbp->bio_inbed) { 931 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 932 pbp->bio_completed = pbp->bio_length; 933 if (pbp->bio_cmd == BIO_WRITE || 934 pbp->bio_cmd == BIO_DELETE) { 935 bioq_remove(&sc->sc_inflight, pbp); 936 /* Release delayed sync requests if possible. */ 937 g_mirror_sync_release(sc); 938 } 939 g_io_deliver(pbp, pbp->bio_error); 940 } 941 return; 942 } else if (bp->bio_error != 0) { 943 if (pbp->bio_error == 0) 944 pbp->bio_error = bp->bio_error; 945 if (disk != NULL) { 946 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 947 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 948 G_MIRROR_LOGREQ(0, bp, 949 "Request failed (error=%d).", 950 bp->bio_error); 951 } else { 952 G_MIRROR_LOGREQ(1, bp, 953 "Request failed (error=%d).", 954 bp->bio_error); 955 } 956 if (g_mirror_disconnect_on_failure && 957 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) 958 { 959 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 960 g_mirror_event_send(disk, 961 G_MIRROR_DISK_STATE_DISCONNECTED, 962 G_MIRROR_EVENT_DONTWAIT); 963 } 964 } 965 switch (pbp->bio_cmd) { 966 case BIO_DELETE: 967 case BIO_WRITE: 968 pbp->bio_inbed--; 969 pbp->bio_children--; 970 break; 971 } 972 } 973 g_destroy_bio(bp); 974 975 switch (pbp->bio_cmd) { 976 case BIO_READ: 977 if (pbp->bio_inbed < pbp->bio_children) 978 break; 979 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 980 g_io_deliver(pbp, pbp->bio_error); 981 else { 982 pbp->bio_error = 0; 983 mtx_lock(&sc->sc_queue_mtx); 984 bioq_insert_tail(&sc->sc_queue, pbp); 985 mtx_unlock(&sc->sc_queue_mtx); 986 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 987 wakeup(sc); 988 } 989 break; 990 case BIO_DELETE: 991 case BIO_WRITE: 992 if (pbp->bio_children == 0) { 993 /* 994 * All requests failed. 995 */ 996 } else if (pbp->bio_inbed < pbp->bio_children) { 997 /* Do nothing. */ 998 break; 999 } else if (pbp->bio_children == pbp->bio_inbed) { 1000 /* Some requests succeeded. */ 1001 pbp->bio_error = 0; 1002 pbp->bio_completed = pbp->bio_length; 1003 } 1004 bioq_remove(&sc->sc_inflight, pbp); 1005 /* Release delayed sync requests if possible. */ 1006 g_mirror_sync_release(sc); 1007 g_io_deliver(pbp, pbp->bio_error); 1008 break; 1009 default: 1010 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 1011 break; 1012 } 1013} 1014 1015static void 1016g_mirror_sync_done(struct bio *bp) 1017{ 1018 struct g_mirror_softc *sc; 1019 1020 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1021 sc = bp->bio_from->geom->softc; 1022 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; 1023 mtx_lock(&sc->sc_queue_mtx); 1024 bioq_insert_tail(&sc->sc_queue, bp); 1025 mtx_unlock(&sc->sc_queue_mtx); 1026 wakeup(sc); 1027} 1028 1029static void 1030g_mirror_candelete(struct bio *bp) 1031{ 1032 struct g_mirror_softc *sc; 1033 struct g_mirror_disk *disk; 1034 int *val; 1035 1036 sc = bp->bio_to->geom->softc; 1037 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1038 if (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) 1039 break; 1040 } 1041 val = (int *)bp->bio_data; 1042 *val = (disk != NULL); 1043 g_io_deliver(bp, 0); 1044} 1045 1046static void 1047g_mirror_kernel_dump(struct bio *bp) 1048{ 1049 struct g_mirror_softc *sc; 1050 struct g_mirror_disk *disk; 1051 struct bio *cbp; 1052 struct g_kerneldump *gkd; 1053 1054 /* 1055 * We configure dumping to the first component, because this component 1056 * will be used for reading with 'prefer' balance algorithm. 1057 * If the component with the higest priority is currently disconnected 1058 * we will not be able to read the dump after the reboot if it will be 1059 * connected and synchronized later. Can we do something better? 1060 */ 1061 sc = bp->bio_to->geom->softc; 1062 disk = LIST_FIRST(&sc->sc_disks); 1063 1064 gkd = (struct g_kerneldump *)bp->bio_data; 1065 if (gkd->length > bp->bio_to->mediasize) 1066 gkd->length = bp->bio_to->mediasize; 1067 cbp = g_clone_bio(bp); 1068 if (cbp == NULL) { 1069 g_io_deliver(bp, ENOMEM); 1070 return; 1071 } 1072 cbp->bio_done = g_std_done; 1073 g_io_request(cbp, disk->d_consumer); 1074 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1075 g_mirror_get_diskname(disk)); 1076} 1077 1078static void 1079g_mirror_flush(struct g_mirror_softc *sc, struct bio *bp) 1080{ 1081 struct bio_queue_head queue; 1082 struct g_mirror_disk *disk; 1083 struct g_consumer *cp; 1084 struct bio *cbp; 1085 1086 bioq_init(&queue); 1087 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1088 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1089 continue; 1090 cbp = g_clone_bio(bp); 1091 if (cbp == NULL) { 1092 while ((cbp = bioq_takefirst(&queue)) != NULL) 1093 g_destroy_bio(cbp); 1094 if (bp->bio_error == 0) 1095 bp->bio_error = ENOMEM; 1096 g_io_deliver(bp, bp->bio_error); 1097 return; 1098 } 1099 bioq_insert_tail(&queue, cbp); 1100 cbp->bio_done = g_mirror_flush_done; 1101 cbp->bio_caller1 = disk; 1102 cbp->bio_to = disk->d_consumer->provider; 1103 } 1104 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1105 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1106 disk = cbp->bio_caller1; 1107 cbp->bio_caller1 = NULL; 1108 cp = disk->d_consumer; 1109 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1110 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1111 cp->acr, cp->acw, cp->ace)); 1112 g_io_request(cbp, disk->d_consumer); 1113 } 1114} 1115 1116static void 1117g_mirror_start(struct bio *bp) 1118{ 1119 struct g_mirror_softc *sc; 1120 1121 sc = bp->bio_to->geom->softc; 1122 /* 1123 * If sc == NULL or there are no valid disks, provider's error 1124 * should be set and g_mirror_start() should not be called at all. 1125 */ 1126 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1127 ("Provider's error should be set (error=%d)(mirror=%s).", 1128 bp->bio_to->error, bp->bio_to->name)); 1129 G_MIRROR_LOGREQ(3, bp, "Request received."); 1130 1131 switch (bp->bio_cmd) { 1132 case BIO_READ: 1133 case BIO_WRITE: 1134 case BIO_DELETE: 1135 break; 1136 case BIO_FLUSH: 1137 g_mirror_flush(sc, bp); 1138 return; 1139 case BIO_GETATTR: 1140 if (!strcmp(bp->bio_attribute, "GEOM::candelete")) { 1141 g_mirror_candelete(bp); 1142 return; 1143 } else if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1144 g_mirror_kernel_dump(bp); 1145 return; 1146 } 1147 /* FALLTHROUGH */ 1148 default: 1149 g_io_deliver(bp, EOPNOTSUPP); 1150 return; 1151 } 1152 mtx_lock(&sc->sc_queue_mtx); 1153 bioq_insert_tail(&sc->sc_queue, bp); 1154 mtx_unlock(&sc->sc_queue_mtx); 1155 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1156 wakeup(sc); 1157} 1158 1159/* 1160 * Return TRUE if the given request is colliding with a in-progress 1161 * synchronization request. 1162 */ 1163static int 1164g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1165{ 1166 struct g_mirror_disk *disk; 1167 struct bio *sbp; 1168 off_t rstart, rend, sstart, send; 1169 u_int i; 1170 1171 if (sc->sc_sync.ds_ndisks == 0) 1172 return (0); 1173 rstart = bp->bio_offset; 1174 rend = bp->bio_offset + bp->bio_length; 1175 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1176 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1177 continue; 1178 for (i = 0; i < g_mirror_syncreqs; i++) { 1179 sbp = disk->d_sync.ds_bios[i]; 1180 if (sbp == NULL) 1181 continue; 1182 sstart = sbp->bio_offset; 1183 send = sbp->bio_offset + sbp->bio_length; 1184 if (rend > sstart && rstart < send) 1185 return (1); 1186 } 1187 } 1188 return (0); 1189} 1190 1191/* 1192 * Return TRUE if the given sync request is colliding with a in-progress regular 1193 * request. 1194 */ 1195static int 1196g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1197{ 1198 off_t rstart, rend, sstart, send; 1199 struct bio *bp; 1200 1201 if (sc->sc_sync.ds_ndisks == 0) 1202 return (0); 1203 sstart = sbp->bio_offset; 1204 send = sbp->bio_offset + sbp->bio_length; 1205 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { 1206 rstart = bp->bio_offset; 1207 rend = bp->bio_offset + bp->bio_length; 1208 if (rend > sstart && rstart < send) 1209 return (1); 1210 } 1211 return (0); 1212} 1213 1214/* 1215 * Puts request onto delayed queue. 1216 */ 1217static void 1218g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1219{ 1220 1221 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1222 bioq_insert_head(&sc->sc_regular_delayed, bp); 1223} 1224 1225/* 1226 * Puts synchronization request onto delayed queue. 1227 */ 1228static void 1229g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1230{ 1231 1232 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1233 bioq_insert_tail(&sc->sc_sync_delayed, bp); 1234} 1235 1236/* 1237 * Releases delayed regular requests which don't collide anymore with sync 1238 * requests. 1239 */ 1240static void 1241g_mirror_regular_release(struct g_mirror_softc *sc) 1242{ 1243 struct bio *bp, *bp2; 1244 1245 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { 1246 if (g_mirror_sync_collision(sc, bp)) 1247 continue; 1248 bioq_remove(&sc->sc_regular_delayed, bp); 1249 G_MIRROR_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1250 mtx_lock(&sc->sc_queue_mtx); 1251 bioq_insert_head(&sc->sc_queue, bp); 1252#if 0 1253 /* 1254 * wakeup() is not needed, because this function is called from 1255 * the worker thread. 1256 */ 1257 wakeup(&sc->sc_queue); 1258#endif 1259 mtx_unlock(&sc->sc_queue_mtx); 1260 } 1261} 1262 1263/* 1264 * Releases delayed sync requests which don't collide anymore with regular 1265 * requests. 1266 */ 1267static void 1268g_mirror_sync_release(struct g_mirror_softc *sc) 1269{ 1270 struct bio *bp, *bp2; 1271 1272 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { 1273 if (g_mirror_regular_collision(sc, bp)) 1274 continue; 1275 bioq_remove(&sc->sc_sync_delayed, bp); 1276 G_MIRROR_LOGREQ(2, bp, 1277 "Releasing delayed synchronization request."); 1278 g_io_request(bp, bp->bio_from); 1279 } 1280} 1281 1282/* 1283 * Handle synchronization requests. 1284 * Every synchronization request is two-steps process: first, READ request is 1285 * send to active provider and then WRITE request (with read data) to the provider 1286 * beeing synchronized. When WRITE is finished, new synchronization request is 1287 * send. 1288 */ 1289static void 1290g_mirror_sync_request(struct bio *bp) 1291{ 1292 struct g_mirror_softc *sc; 1293 struct g_mirror_disk *disk; 1294 1295 bp->bio_from->index--; 1296 sc = bp->bio_from->geom->softc; 1297 disk = bp->bio_from->private; 1298 if (disk == NULL) { 1299 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1300 g_topology_lock(); 1301 g_mirror_kill_consumer(sc, bp->bio_from); 1302 g_topology_unlock(); 1303 free(bp->bio_data, M_MIRROR); 1304 g_destroy_bio(bp); 1305 sx_xlock(&sc->sc_lock); 1306 return; 1307 } 1308 1309 /* 1310 * Synchronization request. 1311 */ 1312 switch (bp->bio_cmd) { 1313 case BIO_READ: 1314 { 1315 struct g_consumer *cp; 1316 1317 if (bp->bio_error != 0) { 1318 G_MIRROR_LOGREQ(0, bp, 1319 "Synchronization request failed (error=%d).", 1320 bp->bio_error); 1321 g_destroy_bio(bp); 1322 return; 1323 } 1324 G_MIRROR_LOGREQ(3, bp, 1325 "Synchronization request half-finished."); 1326 bp->bio_cmd = BIO_WRITE; 1327 bp->bio_cflags = 0; 1328 cp = disk->d_consumer; 1329 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1330 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1331 cp->acr, cp->acw, cp->ace)); 1332 cp->index++; 1333 g_io_request(bp, cp); 1334 return; 1335 } 1336 case BIO_WRITE: 1337 { 1338 struct g_mirror_disk_sync *sync; 1339 off_t offset; 1340 void *data; 1341 int i; 1342 1343 if (bp->bio_error != 0) { 1344 G_MIRROR_LOGREQ(0, bp, 1345 "Synchronization request failed (error=%d).", 1346 bp->bio_error); 1347 g_destroy_bio(bp); 1348 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1349 g_mirror_event_send(disk, 1350 G_MIRROR_DISK_STATE_DISCONNECTED, 1351 G_MIRROR_EVENT_DONTWAIT); 1352 return; 1353 } 1354 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1355 sync = &disk->d_sync; 1356 if (sync->ds_offset >= sc->sc_mediasize || 1357 sync->ds_consumer == NULL || 1358 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1359 /* Don't send more synchronization requests. */ 1360 sync->ds_inflight--; 1361 if (sync->ds_bios != NULL) { 1362 i = (int)(uintptr_t)bp->bio_caller1; 1363 sync->ds_bios[i] = NULL; 1364 } 1365 free(bp->bio_data, M_MIRROR); 1366 g_destroy_bio(bp); 1367 if (sync->ds_inflight > 0) 1368 return; 1369 if (sync->ds_consumer == NULL || 1370 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1371 return; 1372 } 1373 /* Disk up-to-date, activate it. */ 1374 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1375 G_MIRROR_EVENT_DONTWAIT); 1376 return; 1377 } 1378 1379 /* Send next synchronization request. */ 1380 data = bp->bio_data; 1381 bzero(bp, sizeof(*bp)); 1382 bp->bio_cmd = BIO_READ; 1383 bp->bio_offset = sync->ds_offset; 1384 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1385 sync->ds_offset += bp->bio_length; 1386 bp->bio_done = g_mirror_sync_done; 1387 bp->bio_data = data; 1388 bp->bio_from = sync->ds_consumer; 1389 bp->bio_to = sc->sc_provider; 1390 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1391 sync->ds_consumer->index++; 1392 /* 1393 * Delay the request if it is colliding with a regular request. 1394 */ 1395 if (g_mirror_regular_collision(sc, bp)) 1396 g_mirror_sync_delay(sc, bp); 1397 else 1398 g_io_request(bp, sync->ds_consumer); 1399 1400 /* Release delayed requests if possible. */ 1401 g_mirror_regular_release(sc); 1402 1403 /* Find the smallest offset */ 1404 offset = sc->sc_mediasize; 1405 for (i = 0; i < g_mirror_syncreqs; i++) { 1406 bp = sync->ds_bios[i]; 1407 if (bp->bio_offset < offset) 1408 offset = bp->bio_offset; 1409 } 1410 if (sync->ds_offset_done + (MAXPHYS * 100) < offset) { 1411 /* Update offset_done on every 100 blocks. */ 1412 sync->ds_offset_done = offset; 1413 g_mirror_update_metadata(disk); 1414 } 1415 return; 1416 } 1417 default: 1418 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1419 bp->bio_cmd, sc->sc_name)); 1420 break; 1421 } 1422} 1423 1424static void 1425g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1426{ 1427 struct g_mirror_disk *disk; 1428 struct g_consumer *cp; 1429 struct bio *cbp; 1430 1431 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1432 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1433 break; 1434 } 1435 if (disk == NULL) { 1436 if (bp->bio_error == 0) 1437 bp->bio_error = ENXIO; 1438 g_io_deliver(bp, bp->bio_error); 1439 return; 1440 } 1441 cbp = g_clone_bio(bp); 1442 if (cbp == NULL) { 1443 if (bp->bio_error == 0) 1444 bp->bio_error = ENOMEM; 1445 g_io_deliver(bp, bp->bio_error); 1446 return; 1447 } 1448 /* 1449 * Fill in the component buf structure. 1450 */ 1451 cp = disk->d_consumer; 1452 cbp->bio_done = g_mirror_done; 1453 cbp->bio_to = cp->provider; 1454 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1455 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1456 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1457 cp->acw, cp->ace)); 1458 cp->index++; 1459 g_io_request(cbp, cp); 1460} 1461 1462static void 1463g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1464{ 1465 struct g_mirror_disk *disk; 1466 struct g_consumer *cp; 1467 struct bio *cbp; 1468 1469 disk = g_mirror_get_disk(sc); 1470 if (disk == NULL) { 1471 if (bp->bio_error == 0) 1472 bp->bio_error = ENXIO; 1473 g_io_deliver(bp, bp->bio_error); 1474 return; 1475 } 1476 cbp = g_clone_bio(bp); 1477 if (cbp == NULL) { 1478 if (bp->bio_error == 0) 1479 bp->bio_error = ENOMEM; 1480 g_io_deliver(bp, bp->bio_error); 1481 return; 1482 } 1483 /* 1484 * Fill in the component buf structure. 1485 */ 1486 cp = disk->d_consumer; 1487 cbp->bio_done = g_mirror_done; 1488 cbp->bio_to = cp->provider; 1489 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1490 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1491 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1492 cp->acw, cp->ace)); 1493 cp->index++; 1494 g_io_request(cbp, cp); 1495} 1496 1497#define TRACK_SIZE (1 * 1024 * 1024) 1498#define LOAD_SCALE 256 1499#define ABS(x) (((x) >= 0) ? (x) : (-(x))) 1500 1501static void 1502g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1503{ 1504 struct g_mirror_disk *disk, *dp; 1505 struct g_consumer *cp; 1506 struct bio *cbp; 1507 int prio, best; 1508 1509 /* Find a disk with the smallest load. */ 1510 disk = NULL; 1511 best = INT_MAX; 1512 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1513 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1514 continue; 1515 prio = dp->load; 1516 /* If disk head is precisely in position - highly prefer it. */ 1517 if (dp->d_last_offset == bp->bio_offset) 1518 prio -= 2 * LOAD_SCALE; 1519 else 1520 /* If disk head is close to position - prefer it. */ 1521 if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE) 1522 prio -= 1 * LOAD_SCALE; 1523 if (prio <= best) { 1524 disk = dp; 1525 best = prio; 1526 } 1527 } 1528 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1529 cbp = g_clone_bio(bp); 1530 if (cbp == NULL) { 1531 if (bp->bio_error == 0) 1532 bp->bio_error = ENOMEM; 1533 g_io_deliver(bp, bp->bio_error); 1534 return; 1535 } 1536 /* 1537 * Fill in the component buf structure. 1538 */ 1539 cp = disk->d_consumer; 1540 cbp->bio_done = g_mirror_done; 1541 cbp->bio_to = cp->provider; 1542 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1543 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1544 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1545 cp->acw, cp->ace)); 1546 cp->index++; 1547 /* Remember last head position */ 1548 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1549 /* Update loads. */ 1550 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1551 dp->load = (dp->d_consumer->index * LOAD_SCALE + 1552 dp->load * 7) / 8; 1553 } 1554 g_io_request(cbp, cp); 1555} 1556 1557static void 1558g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1559{ 1560 struct bio_queue_head queue; 1561 struct g_mirror_disk *disk; 1562 struct g_consumer *cp; 1563 struct bio *cbp; 1564 off_t left, mod, offset, slice; 1565 u_char *data; 1566 u_int ndisks; 1567 1568 if (bp->bio_length <= sc->sc_slice) { 1569 g_mirror_request_round_robin(sc, bp); 1570 return; 1571 } 1572 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1573 slice = bp->bio_length / ndisks; 1574 mod = slice % sc->sc_provider->sectorsize; 1575 if (mod != 0) 1576 slice += sc->sc_provider->sectorsize - mod; 1577 /* 1578 * Allocate all bios before sending any request, so we can 1579 * return ENOMEM in nice and clean way. 1580 */ 1581 left = bp->bio_length; 1582 offset = bp->bio_offset; 1583 data = bp->bio_data; 1584 bioq_init(&queue); 1585 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1586 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1587 continue; 1588 cbp = g_clone_bio(bp); 1589 if (cbp == NULL) { 1590 while ((cbp = bioq_takefirst(&queue)) != NULL) 1591 g_destroy_bio(cbp); 1592 if (bp->bio_error == 0) 1593 bp->bio_error = ENOMEM; 1594 g_io_deliver(bp, bp->bio_error); 1595 return; 1596 } 1597 bioq_insert_tail(&queue, cbp); 1598 cbp->bio_done = g_mirror_done; 1599 cbp->bio_caller1 = disk; 1600 cbp->bio_to = disk->d_consumer->provider; 1601 cbp->bio_offset = offset; 1602 cbp->bio_data = data; 1603 cbp->bio_length = MIN(left, slice); 1604 left -= cbp->bio_length; 1605 if (left == 0) 1606 break; 1607 offset += cbp->bio_length; 1608 data += cbp->bio_length; 1609 } 1610 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1611 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1612 disk = cbp->bio_caller1; 1613 cbp->bio_caller1 = NULL; 1614 cp = disk->d_consumer; 1615 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1616 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1617 cp->acr, cp->acw, cp->ace)); 1618 disk->d_consumer->index++; 1619 g_io_request(cbp, disk->d_consumer); 1620 } 1621} 1622 1623static void 1624g_mirror_register_request(struct bio *bp) 1625{ 1626 struct g_mirror_softc *sc; 1627 1628 sc = bp->bio_to->geom->softc; 1629 switch (bp->bio_cmd) { 1630 case BIO_READ: 1631 switch (sc->sc_balance) { 1632 case G_MIRROR_BALANCE_LOAD: 1633 g_mirror_request_load(sc, bp); 1634 break; 1635 case G_MIRROR_BALANCE_PREFER: 1636 g_mirror_request_prefer(sc, bp); 1637 break; 1638 case G_MIRROR_BALANCE_ROUND_ROBIN: 1639 g_mirror_request_round_robin(sc, bp); 1640 break; 1641 case G_MIRROR_BALANCE_SPLIT: 1642 g_mirror_request_split(sc, bp); 1643 break; 1644 } 1645 return; 1646 case BIO_WRITE: 1647 case BIO_DELETE: 1648 { 1649 struct g_mirror_disk *disk; 1650 struct g_mirror_disk_sync *sync; 1651 struct bio_queue_head queue; 1652 struct g_consumer *cp; 1653 struct bio *cbp; 1654 1655 /* 1656 * Delay the request if it is colliding with a synchronization 1657 * request. 1658 */ 1659 if (g_mirror_sync_collision(sc, bp)) { 1660 g_mirror_regular_delay(sc, bp); 1661 return; 1662 } 1663 1664 if (sc->sc_idle) 1665 g_mirror_unidle(sc); 1666 else 1667 sc->sc_last_write = time_uptime; 1668 1669 /* 1670 * Allocate all bios before sending any request, so we can 1671 * return ENOMEM in nice and clean way. 1672 */ 1673 bioq_init(&queue); 1674 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1675 sync = &disk->d_sync; 1676 switch (disk->d_state) { 1677 case G_MIRROR_DISK_STATE_ACTIVE: 1678 break; 1679 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1680 if (bp->bio_offset >= sync->ds_offset) 1681 continue; 1682 break; 1683 default: 1684 continue; 1685 } 1686 if (bp->bio_cmd == BIO_DELETE && 1687 (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0) 1688 continue; 1689 cbp = g_clone_bio(bp); 1690 if (cbp == NULL) { 1691 while ((cbp = bioq_takefirst(&queue)) != NULL) 1692 g_destroy_bio(cbp); 1693 if (bp->bio_error == 0) 1694 bp->bio_error = ENOMEM; 1695 g_io_deliver(bp, bp->bio_error); 1696 return; 1697 } 1698 bioq_insert_tail(&queue, cbp); 1699 cbp->bio_done = g_mirror_done; 1700 cp = disk->d_consumer; 1701 cbp->bio_caller1 = cp; 1702 cbp->bio_to = cp->provider; 1703 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1704 ("Consumer %s not opened (r%dw%de%d).", 1705 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1706 } 1707 if (bioq_first(&queue) == NULL) { 1708 g_io_deliver(bp, EOPNOTSUPP); 1709 return; 1710 } 1711 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1712 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1713 cp = cbp->bio_caller1; 1714 cbp->bio_caller1 = NULL; 1715 cp->index++; 1716 sc->sc_writes++; 1717 g_io_request(cbp, cp); 1718 } 1719 /* 1720 * Put request onto inflight queue, so we can check if new 1721 * synchronization requests don't collide with it. 1722 */ 1723 bioq_insert_tail(&sc->sc_inflight, bp); 1724 /* 1725 * Bump syncid on first write. 1726 */ 1727 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1728 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1729 g_mirror_bump_syncid(sc); 1730 } 1731 return; 1732 } 1733 default: 1734 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1735 bp->bio_cmd, sc->sc_name)); 1736 break; 1737 } 1738} 1739 1740static int 1741g_mirror_can_destroy(struct g_mirror_softc *sc) 1742{ 1743 struct g_geom *gp; 1744 struct g_consumer *cp; 1745 1746 g_topology_assert(); 1747 gp = sc->sc_geom; 1748 if (gp->softc == NULL) 1749 return (1); 1750 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0) 1751 return (0); 1752 LIST_FOREACH(cp, &gp->consumer, consumer) { 1753 if (g_mirror_is_busy(sc, cp)) 1754 return (0); 1755 } 1756 gp = sc->sc_sync.ds_geom; 1757 LIST_FOREACH(cp, &gp->consumer, consumer) { 1758 if (g_mirror_is_busy(sc, cp)) 1759 return (0); 1760 } 1761 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1762 sc->sc_name); 1763 return (1); 1764} 1765 1766static int 1767g_mirror_try_destroy(struct g_mirror_softc *sc) 1768{ 1769 1770 if (sc->sc_rootmount != NULL) { 1771 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1772 sc->sc_rootmount); 1773 root_mount_rel(sc->sc_rootmount); 1774 sc->sc_rootmount = NULL; 1775 } 1776 g_topology_lock(); 1777 if (!g_mirror_can_destroy(sc)) { 1778 g_topology_unlock(); 1779 return (0); 1780 } 1781 sc->sc_geom->softc = NULL; 1782 sc->sc_sync.ds_geom->softc = NULL; 1783 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1784 g_topology_unlock(); 1785 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1786 &sc->sc_worker); 1787 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1788 sx_xunlock(&sc->sc_lock); 1789 wakeup(&sc->sc_worker); 1790 sc->sc_worker = NULL; 1791 } else { 1792 g_topology_unlock(); 1793 g_mirror_destroy_device(sc); 1794 free(sc, M_MIRROR); 1795 } 1796 return (1); 1797} 1798 1799/* 1800 * Worker thread. 1801 */ 1802static void 1803g_mirror_worker(void *arg) 1804{ 1805 struct g_mirror_softc *sc; 1806 struct g_mirror_event *ep; 1807 struct bio *bp; 1808 int timeout; 1809 1810 sc = arg; 1811 thread_lock(curthread); 1812 sched_prio(curthread, PRIBIO); 1813 thread_unlock(curthread); 1814 1815 sx_xlock(&sc->sc_lock); 1816 for (;;) { 1817 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1818 /* 1819 * First take a look at events. 1820 * This is important to handle events before any I/O requests. 1821 */ 1822 ep = g_mirror_event_get(sc); 1823 if (ep != NULL) { 1824 g_mirror_event_remove(sc, ep); 1825 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1826 /* Update only device status. */ 1827 G_MIRROR_DEBUG(3, 1828 "Running event for device %s.", 1829 sc->sc_name); 1830 ep->e_error = 0; 1831 g_mirror_update_device(sc, 1); 1832 } else { 1833 /* Update disk status. */ 1834 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1835 g_mirror_get_diskname(ep->e_disk)); 1836 ep->e_error = g_mirror_update_disk(ep->e_disk, 1837 ep->e_state); 1838 if (ep->e_error == 0) 1839 g_mirror_update_device(sc, 0); 1840 } 1841 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1842 KASSERT(ep->e_error == 0, 1843 ("Error cannot be handled.")); 1844 g_mirror_event_free(ep); 1845 } else { 1846 ep->e_flags |= G_MIRROR_EVENT_DONE; 1847 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1848 ep); 1849 mtx_lock(&sc->sc_events_mtx); 1850 wakeup(ep); 1851 mtx_unlock(&sc->sc_events_mtx); 1852 } 1853 if ((sc->sc_flags & 1854 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1855 if (g_mirror_try_destroy(sc)) { 1856 curthread->td_pflags &= ~TDP_GEOM; 1857 G_MIRROR_DEBUG(1, "Thread exiting."); 1858 kproc_exit(0); 1859 } 1860 } 1861 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1862 continue; 1863 } 1864 /* 1865 * Check if we can mark array as CLEAN and if we can't take 1866 * how much seconds should we wait. 1867 */ 1868 timeout = g_mirror_idle(sc, -1); 1869 /* 1870 * Now I/O requests. 1871 */ 1872 /* Get first request from the queue. */ 1873 mtx_lock(&sc->sc_queue_mtx); 1874 bp = bioq_takefirst(&sc->sc_queue); 1875 if (bp == NULL) { 1876 if ((sc->sc_flags & 1877 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1878 mtx_unlock(&sc->sc_queue_mtx); 1879 if (g_mirror_try_destroy(sc)) { 1880 curthread->td_pflags &= ~TDP_GEOM; 1881 G_MIRROR_DEBUG(1, "Thread exiting."); 1882 kproc_exit(0); 1883 } 1884 mtx_lock(&sc->sc_queue_mtx); 1885 } 1886 sx_xunlock(&sc->sc_lock); 1887 /* 1888 * XXX: We can miss an event here, because an event 1889 * can be added without sx-device-lock and without 1890 * mtx-queue-lock. Maybe I should just stop using 1891 * dedicated mutex for events synchronization and 1892 * stick with the queue lock? 1893 * The event will hang here until next I/O request 1894 * or next event is received. 1895 */ 1896 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1897 timeout * hz); 1898 sx_xlock(&sc->sc_lock); 1899 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1900 continue; 1901 } 1902 mtx_unlock(&sc->sc_queue_mtx); 1903 1904 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 1905 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1906 g_mirror_sync_request(bp); /* READ */ 1907 } else if (bp->bio_to != sc->sc_provider) { 1908 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 1909 g_mirror_regular_request(bp); 1910 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 1911 g_mirror_sync_request(bp); /* WRITE */ 1912 else { 1913 KASSERT(0, 1914 ("Invalid request cflags=0x%hhx to=%s.", 1915 bp->bio_cflags, bp->bio_to->name)); 1916 } 1917 } else { 1918 g_mirror_register_request(bp); 1919 } 1920 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1921 } 1922} 1923 1924static void 1925g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 1926{ 1927 1928 sx_assert(&sc->sc_lock, SX_LOCKED); 1929 1930 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 1931 return; 1932 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1933 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1934 g_mirror_get_diskname(disk), sc->sc_name); 1935 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1936 } else if (sc->sc_idle && 1937 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1938 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1939 g_mirror_get_diskname(disk), sc->sc_name); 1940 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1941 } 1942} 1943 1944static void 1945g_mirror_sync_start(struct g_mirror_disk *disk) 1946{ 1947 struct g_mirror_softc *sc; 1948 struct g_consumer *cp; 1949 struct bio *bp; 1950 int error, i; 1951 1952 g_topology_assert_not(); 1953 sc = disk->d_softc; 1954 sx_assert(&sc->sc_lock, SX_LOCKED); 1955 1956 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1957 ("Disk %s is not marked for synchronization.", 1958 g_mirror_get_diskname(disk))); 1959 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1960 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1961 sc->sc_state)); 1962 1963 sx_xunlock(&sc->sc_lock); 1964 g_topology_lock(); 1965 cp = g_new_consumer(sc->sc_sync.ds_geom); 1966 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 1967 error = g_attach(cp, sc->sc_provider); 1968 KASSERT(error == 0, 1969 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 1970 error = g_access(cp, 1, 0, 0); 1971 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 1972 g_topology_unlock(); 1973 sx_xlock(&sc->sc_lock); 1974 1975 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1976 g_mirror_get_diskname(disk)); 1977 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0) 1978 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1979 KASSERT(disk->d_sync.ds_consumer == NULL, 1980 ("Sync consumer already exists (device=%s, disk=%s).", 1981 sc->sc_name, g_mirror_get_diskname(disk))); 1982 1983 disk->d_sync.ds_consumer = cp; 1984 disk->d_sync.ds_consumer->private = disk; 1985 disk->d_sync.ds_consumer->index = 0; 1986 1987 /* 1988 * Allocate memory for synchronization bios and initialize them. 1989 */ 1990 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 1991 M_MIRROR, M_WAITOK); 1992 for (i = 0; i < g_mirror_syncreqs; i++) { 1993 bp = g_alloc_bio(); 1994 disk->d_sync.ds_bios[i] = bp; 1995 bp->bio_parent = NULL; 1996 bp->bio_cmd = BIO_READ; 1997 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1998 bp->bio_cflags = 0; 1999 bp->bio_offset = disk->d_sync.ds_offset; 2000 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 2001 disk->d_sync.ds_offset += bp->bio_length; 2002 bp->bio_done = g_mirror_sync_done; 2003 bp->bio_from = disk->d_sync.ds_consumer; 2004 bp->bio_to = sc->sc_provider; 2005 bp->bio_caller1 = (void *)(uintptr_t)i; 2006 } 2007 2008 /* Increase the number of disks in SYNCHRONIZING state. */ 2009 sc->sc_sync.ds_ndisks++; 2010 /* Set the number of in-flight synchronization requests. */ 2011 disk->d_sync.ds_inflight = g_mirror_syncreqs; 2012 2013 /* 2014 * Fire off first synchronization requests. 2015 */ 2016 for (i = 0; i < g_mirror_syncreqs; i++) { 2017 bp = disk->d_sync.ds_bios[i]; 2018 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 2019 disk->d_sync.ds_consumer->index++; 2020 /* 2021 * Delay the request if it is colliding with a regular request. 2022 */ 2023 if (g_mirror_regular_collision(sc, bp)) 2024 g_mirror_sync_delay(sc, bp); 2025 else 2026 g_io_request(bp, disk->d_sync.ds_consumer); 2027 } 2028} 2029 2030/* 2031 * Stop synchronization process. 2032 * type: 0 - synchronization finished 2033 * 1 - synchronization stopped 2034 */ 2035static void 2036g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 2037{ 2038 struct g_mirror_softc *sc; 2039 struct g_consumer *cp; 2040 2041 g_topology_assert_not(); 2042 sc = disk->d_softc; 2043 sx_assert(&sc->sc_lock, SX_LOCKED); 2044 2045 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2046 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2047 g_mirror_disk_state2str(disk->d_state))); 2048 if (disk->d_sync.ds_consumer == NULL) 2049 return; 2050 2051 if (type == 0) { 2052 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2053 sc->sc_name, g_mirror_get_diskname(disk)); 2054 } else /* if (type == 1) */ { 2055 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2056 sc->sc_name, g_mirror_get_diskname(disk)); 2057 } 2058 free(disk->d_sync.ds_bios, M_MIRROR); 2059 disk->d_sync.ds_bios = NULL; 2060 cp = disk->d_sync.ds_consumer; 2061 disk->d_sync.ds_consumer = NULL; 2062 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2063 sc->sc_sync.ds_ndisks--; 2064 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2065 g_topology_lock(); 2066 g_mirror_kill_consumer(sc, cp); 2067 g_topology_unlock(); 2068 sx_xlock(&sc->sc_lock); 2069} 2070 2071static void 2072g_mirror_launch_provider(struct g_mirror_softc *sc) 2073{ 2074 struct g_mirror_disk *disk; 2075 struct g_provider *pp, *dp; 2076 2077 sx_assert(&sc->sc_lock, SX_LOCKED); 2078 2079 g_topology_lock(); 2080 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 2081 pp->flags |= G_PF_DIRECT_RECEIVE; 2082 pp->mediasize = sc->sc_mediasize; 2083 pp->sectorsize = sc->sc_sectorsize; 2084 pp->stripesize = 0; 2085 pp->stripeoffset = 0; 2086 2087 /* Splitting of unmapped BIO's could work but isn't implemented now */ 2088 if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT) 2089 pp->flags |= G_PF_ACCEPT_UNMAPPED; 2090 2091 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2092 if (disk->d_consumer && disk->d_consumer->provider) { 2093 dp = disk->d_consumer->provider; 2094 if (dp->stripesize > pp->stripesize) { 2095 pp->stripesize = dp->stripesize; 2096 pp->stripeoffset = dp->stripeoffset; 2097 } 2098 /* A provider underneath us doesn't support unmapped */ 2099 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { 2100 G_MIRROR_DEBUG(0, "Cancelling unmapped " 2101 "because of %s.", dp->name); 2102 pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 2103 } 2104 } 2105 } 2106 sc->sc_provider = pp; 2107 g_error_provider(pp, 0); 2108 g_topology_unlock(); 2109 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2110 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 2111 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2112 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2113 g_mirror_sync_start(disk); 2114 } 2115} 2116 2117static void 2118g_mirror_destroy_provider(struct g_mirror_softc *sc) 2119{ 2120 struct g_mirror_disk *disk; 2121 struct bio *bp; 2122 2123 g_topology_assert_not(); 2124 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2125 sc->sc_name)); 2126 2127 g_topology_lock(); 2128 g_error_provider(sc->sc_provider, ENXIO); 2129 mtx_lock(&sc->sc_queue_mtx); 2130 while ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) { 2131 /* 2132 * Abort any pending I/O that wasn't generated by us. 2133 * Synchronization requests and requests destined for individual 2134 * mirror components can be destroyed immediately. 2135 */ 2136 if (bp->bio_to == sc->sc_provider && 2137 bp->bio_from->geom != sc->sc_sync.ds_geom) { 2138 g_io_deliver(bp, ENXIO); 2139 } else { 2140 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2141 free(bp->bio_data, M_MIRROR); 2142 g_destroy_bio(bp); 2143 } 2144 } 2145 mtx_unlock(&sc->sc_queue_mtx); 2146 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 2147 sc->sc_provider->name); 2148 g_wither_provider(sc->sc_provider, ENXIO); 2149 sc->sc_provider = NULL; 2150 g_topology_unlock(); 2151 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2152 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2153 g_mirror_sync_stop(disk, 1); 2154 } 2155} 2156 2157static void 2158g_mirror_go(void *arg) 2159{ 2160 struct g_mirror_softc *sc; 2161 2162 sc = arg; 2163 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2164 g_mirror_event_send(sc, 0, 2165 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2166} 2167 2168static u_int 2169g_mirror_determine_state(struct g_mirror_disk *disk) 2170{ 2171 struct g_mirror_softc *sc; 2172 u_int state; 2173 2174 sc = disk->d_softc; 2175 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2176 if ((disk->d_flags & 2177 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2178 /* Disk does not need synchronization. */ 2179 state = G_MIRROR_DISK_STATE_ACTIVE; 2180 } else { 2181 if ((sc->sc_flags & 2182 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2183 (disk->d_flags & 2184 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2185 /* 2186 * We can start synchronization from 2187 * the stored offset. 2188 */ 2189 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2190 } else { 2191 state = G_MIRROR_DISK_STATE_STALE; 2192 } 2193 } 2194 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2195 /* 2196 * Reset all synchronization data for this disk, 2197 * because if it even was synchronized, it was 2198 * synchronized to disks with different syncid. 2199 */ 2200 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2201 disk->d_sync.ds_offset = 0; 2202 disk->d_sync.ds_offset_done = 0; 2203 disk->d_sync.ds_syncid = sc->sc_syncid; 2204 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2205 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2206 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2207 } else { 2208 state = G_MIRROR_DISK_STATE_STALE; 2209 } 2210 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2211 /* 2212 * Not good, NOT GOOD! 2213 * It means that mirror was started on stale disks 2214 * and more fresh disk just arrive. 2215 * If there were writes, mirror is broken, sorry. 2216 * I think the best choice here is don't touch 2217 * this disk and inform the user loudly. 2218 */ 2219 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2220 "disk (%s) arrives!! It will not be connected to the " 2221 "running device.", sc->sc_name, 2222 g_mirror_get_diskname(disk)); 2223 g_mirror_destroy_disk(disk); 2224 state = G_MIRROR_DISK_STATE_NONE; 2225 /* Return immediately, because disk was destroyed. */ 2226 return (state); 2227 } 2228 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2229 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2230 return (state); 2231} 2232 2233/* 2234 * Update device state. 2235 */ 2236static void 2237g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 2238{ 2239 struct g_mirror_disk *disk; 2240 u_int state; 2241 2242 sx_assert(&sc->sc_lock, SX_XLOCKED); 2243 2244 switch (sc->sc_state) { 2245 case G_MIRROR_DEVICE_STATE_STARTING: 2246 { 2247 struct g_mirror_disk *pdisk, *tdisk; 2248 u_int dirty, ndisks, genid, syncid; 2249 2250 KASSERT(sc->sc_provider == NULL, 2251 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2252 /* 2253 * Are we ready? We are, if all disks are connected or 2254 * if we have any disks and 'force' is true. 2255 */ 2256 ndisks = g_mirror_ndisks(sc, -1); 2257 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2258 ; 2259 } else if (ndisks == 0) { 2260 /* 2261 * Disks went down in starting phase, so destroy 2262 * device. 2263 */ 2264 callout_drain(&sc->sc_callout); 2265 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2266 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2267 sc->sc_rootmount); 2268 root_mount_rel(sc->sc_rootmount); 2269 sc->sc_rootmount = NULL; 2270 return; 2271 } else { 2272 return; 2273 } 2274 2275 /* 2276 * Activate all disks with the biggest syncid. 2277 */ 2278 if (force) { 2279 /* 2280 * If 'force' is true, we have been called due to 2281 * timeout, so don't bother canceling timeout. 2282 */ 2283 ndisks = 0; 2284 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2285 if ((disk->d_flags & 2286 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2287 ndisks++; 2288 } 2289 } 2290 if (ndisks == 0) { 2291 /* No valid disks found, destroy device. */ 2292 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2293 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2294 __LINE__, sc->sc_rootmount); 2295 root_mount_rel(sc->sc_rootmount); 2296 sc->sc_rootmount = NULL; 2297 return; 2298 } 2299 } else { 2300 /* Cancel timeout. */ 2301 callout_drain(&sc->sc_callout); 2302 } 2303 2304 /* 2305 * Find the biggest genid. 2306 */ 2307 genid = 0; 2308 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2309 if (disk->d_genid > genid) 2310 genid = disk->d_genid; 2311 } 2312 sc->sc_genid = genid; 2313 /* 2314 * Remove all disks without the biggest genid. 2315 */ 2316 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2317 if (disk->d_genid < genid) { 2318 G_MIRROR_DEBUG(0, 2319 "Component %s (device %s) broken, skipping.", 2320 g_mirror_get_diskname(disk), sc->sc_name); 2321 g_mirror_destroy_disk(disk); 2322 } 2323 } 2324 2325 /* 2326 * Find the biggest syncid. 2327 */ 2328 syncid = 0; 2329 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2330 if (disk->d_sync.ds_syncid > syncid) 2331 syncid = disk->d_sync.ds_syncid; 2332 } 2333 2334 /* 2335 * Here we need to look for dirty disks and if all disks 2336 * with the biggest syncid are dirty, we have to choose 2337 * one with the biggest priority and rebuild the rest. 2338 */ 2339 /* 2340 * Find the number of dirty disks with the biggest syncid. 2341 * Find the number of disks with the biggest syncid. 2342 * While here, find a disk with the biggest priority. 2343 */ 2344 dirty = ndisks = 0; 2345 pdisk = NULL; 2346 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2347 if (disk->d_sync.ds_syncid != syncid) 2348 continue; 2349 if ((disk->d_flags & 2350 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2351 continue; 2352 } 2353 ndisks++; 2354 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2355 dirty++; 2356 if (pdisk == NULL || 2357 pdisk->d_priority < disk->d_priority) { 2358 pdisk = disk; 2359 } 2360 } 2361 } 2362 if (dirty == 0) { 2363 /* No dirty disks at all, great. */ 2364 } else if (dirty == ndisks) { 2365 /* 2366 * Force synchronization for all dirty disks except one 2367 * with the biggest priority. 2368 */ 2369 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2370 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2371 "master disk for synchronization.", 2372 g_mirror_get_diskname(pdisk), sc->sc_name); 2373 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2374 if (disk->d_sync.ds_syncid != syncid) 2375 continue; 2376 if ((disk->d_flags & 2377 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2378 continue; 2379 } 2380 KASSERT((disk->d_flags & 2381 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2382 ("Disk %s isn't marked as dirty.", 2383 g_mirror_get_diskname(disk))); 2384 /* Skip the disk with the biggest priority. */ 2385 if (disk == pdisk) 2386 continue; 2387 disk->d_sync.ds_syncid = 0; 2388 } 2389 } else if (dirty < ndisks) { 2390 /* 2391 * Force synchronization for all dirty disks. 2392 * We have some non-dirty disks. 2393 */ 2394 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2395 if (disk->d_sync.ds_syncid != syncid) 2396 continue; 2397 if ((disk->d_flags & 2398 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2399 continue; 2400 } 2401 if ((disk->d_flags & 2402 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2403 continue; 2404 } 2405 disk->d_sync.ds_syncid = 0; 2406 } 2407 } 2408 2409 /* Reset hint. */ 2410 sc->sc_hint = NULL; 2411 sc->sc_syncid = syncid; 2412 if (force) { 2413 /* Remember to bump syncid on first write. */ 2414 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2415 } 2416 state = G_MIRROR_DEVICE_STATE_RUNNING; 2417 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2418 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2419 g_mirror_device_state2str(state)); 2420 sc->sc_state = state; 2421 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2422 state = g_mirror_determine_state(disk); 2423 g_mirror_event_send(disk, state, 2424 G_MIRROR_EVENT_DONTWAIT); 2425 if (state == G_MIRROR_DISK_STATE_STALE) 2426 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2427 } 2428 break; 2429 } 2430 case G_MIRROR_DEVICE_STATE_RUNNING: 2431 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2432 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2433 /* 2434 * No active disks or no disks at all, 2435 * so destroy device. 2436 */ 2437 if (sc->sc_provider != NULL) 2438 g_mirror_destroy_provider(sc); 2439 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2440 break; 2441 } else if (g_mirror_ndisks(sc, 2442 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2443 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2444 /* 2445 * We have active disks, launch provider if it doesn't 2446 * exist. 2447 */ 2448 if (sc->sc_provider == NULL) 2449 g_mirror_launch_provider(sc); 2450 if (sc->sc_rootmount != NULL) { 2451 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2452 __LINE__, sc->sc_rootmount); 2453 root_mount_rel(sc->sc_rootmount); 2454 sc->sc_rootmount = NULL; 2455 } 2456 } 2457 /* 2458 * Genid should be bumped immediately, so do it here. 2459 */ 2460 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2461 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2462 g_mirror_bump_genid(sc); 2463 } 2464 break; 2465 default: 2466 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2467 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2468 break; 2469 } 2470} 2471 2472/* 2473 * Update disk state and device state if needed. 2474 */ 2475#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2476 "Disk %s state changed from %s to %s (device %s).", \ 2477 g_mirror_get_diskname(disk), \ 2478 g_mirror_disk_state2str(disk->d_state), \ 2479 g_mirror_disk_state2str(state), sc->sc_name) 2480static int 2481g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2482{ 2483 struct g_mirror_softc *sc; 2484 2485 sc = disk->d_softc; 2486 sx_assert(&sc->sc_lock, SX_XLOCKED); 2487 2488again: 2489 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2490 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2491 g_mirror_disk_state2str(state)); 2492 switch (state) { 2493 case G_MIRROR_DISK_STATE_NEW: 2494 /* 2495 * Possible scenarios: 2496 * 1. New disk arrive. 2497 */ 2498 /* Previous state should be NONE. */ 2499 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2500 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2501 g_mirror_disk_state2str(disk->d_state))); 2502 DISK_STATE_CHANGED(); 2503 2504 disk->d_state = state; 2505 if (LIST_EMPTY(&sc->sc_disks)) 2506 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2507 else { 2508 struct g_mirror_disk *dp; 2509 2510 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2511 if (disk->d_priority >= dp->d_priority) { 2512 LIST_INSERT_BEFORE(dp, disk, d_next); 2513 dp = NULL; 2514 break; 2515 } 2516 if (LIST_NEXT(dp, d_next) == NULL) 2517 break; 2518 } 2519 if (dp != NULL) 2520 LIST_INSERT_AFTER(dp, disk, d_next); 2521 } 2522 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2523 sc->sc_name, g_mirror_get_diskname(disk)); 2524 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2525 break; 2526 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2527 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2528 g_mirror_device_state2str(sc->sc_state), 2529 g_mirror_get_diskname(disk), 2530 g_mirror_disk_state2str(disk->d_state))); 2531 state = g_mirror_determine_state(disk); 2532 if (state != G_MIRROR_DISK_STATE_NONE) 2533 goto again; 2534 break; 2535 case G_MIRROR_DISK_STATE_ACTIVE: 2536 /* 2537 * Possible scenarios: 2538 * 1. New disk does not need synchronization. 2539 * 2. Synchronization process finished successfully. 2540 */ 2541 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2542 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2543 g_mirror_device_state2str(sc->sc_state), 2544 g_mirror_get_diskname(disk), 2545 g_mirror_disk_state2str(disk->d_state))); 2546 /* Previous state should be NEW or SYNCHRONIZING. */ 2547 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2548 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2549 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2550 g_mirror_disk_state2str(disk->d_state))); 2551 DISK_STATE_CHANGED(); 2552 2553 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2554 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2555 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2556 g_mirror_sync_stop(disk, 0); 2557 } 2558 disk->d_state = state; 2559 disk->d_sync.ds_offset = 0; 2560 disk->d_sync.ds_offset_done = 0; 2561 g_mirror_update_idle(sc, disk); 2562 g_mirror_update_metadata(disk); 2563 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2564 sc->sc_name, g_mirror_get_diskname(disk)); 2565 break; 2566 case G_MIRROR_DISK_STATE_STALE: 2567 /* 2568 * Possible scenarios: 2569 * 1. Stale disk was connected. 2570 */ 2571 /* Previous state should be NEW. */ 2572 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2573 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2574 g_mirror_disk_state2str(disk->d_state))); 2575 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2576 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2577 g_mirror_device_state2str(sc->sc_state), 2578 g_mirror_get_diskname(disk), 2579 g_mirror_disk_state2str(disk->d_state))); 2580 /* 2581 * STALE state is only possible if device is marked 2582 * NOAUTOSYNC. 2583 */ 2584 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2585 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2586 g_mirror_device_state2str(sc->sc_state), 2587 g_mirror_get_diskname(disk), 2588 g_mirror_disk_state2str(disk->d_state))); 2589 DISK_STATE_CHANGED(); 2590 2591 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2592 disk->d_state = state; 2593 g_mirror_update_metadata(disk); 2594 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2595 sc->sc_name, g_mirror_get_diskname(disk)); 2596 break; 2597 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2598 /* 2599 * Possible scenarios: 2600 * 1. Disk which needs synchronization was connected. 2601 */ 2602 /* Previous state should be NEW. */ 2603 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2604 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2605 g_mirror_disk_state2str(disk->d_state))); 2606 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2607 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2608 g_mirror_device_state2str(sc->sc_state), 2609 g_mirror_get_diskname(disk), 2610 g_mirror_disk_state2str(disk->d_state))); 2611 DISK_STATE_CHANGED(); 2612 2613 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2614 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2615 disk->d_state = state; 2616 if (sc->sc_provider != NULL) { 2617 g_mirror_sync_start(disk); 2618 g_mirror_update_metadata(disk); 2619 } 2620 break; 2621 case G_MIRROR_DISK_STATE_DISCONNECTED: 2622 /* 2623 * Possible scenarios: 2624 * 1. Device wasn't running yet, but disk disappear. 2625 * 2. Disk was active and disapppear. 2626 * 3. Disk disappear during synchronization process. 2627 */ 2628 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2629 /* 2630 * Previous state should be ACTIVE, STALE or 2631 * SYNCHRONIZING. 2632 */ 2633 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2634 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2635 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2636 ("Wrong disk state (%s, %s).", 2637 g_mirror_get_diskname(disk), 2638 g_mirror_disk_state2str(disk->d_state))); 2639 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2640 /* Previous state should be NEW. */ 2641 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2642 ("Wrong disk state (%s, %s).", 2643 g_mirror_get_diskname(disk), 2644 g_mirror_disk_state2str(disk->d_state))); 2645 /* 2646 * Reset bumping syncid if disk disappeared in STARTING 2647 * state. 2648 */ 2649 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2650 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2651#ifdef INVARIANTS 2652 } else { 2653 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2654 sc->sc_name, 2655 g_mirror_device_state2str(sc->sc_state), 2656 g_mirror_get_diskname(disk), 2657 g_mirror_disk_state2str(disk->d_state))); 2658#endif 2659 } 2660 DISK_STATE_CHANGED(); 2661 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2662 sc->sc_name, g_mirror_get_diskname(disk)); 2663 2664 g_mirror_destroy_disk(disk); 2665 break; 2666 case G_MIRROR_DISK_STATE_DESTROY: 2667 { 2668 int error; 2669 2670 error = g_mirror_clear_metadata(disk); 2671 if (error != 0) { 2672 G_MIRROR_DEBUG(0, 2673 "Device %s: failed to clear metadata on %s: %d.", 2674 sc->sc_name, g_mirror_get_diskname(disk), error); 2675 break; 2676 } 2677 DISK_STATE_CHANGED(); 2678 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2679 sc->sc_name, g_mirror_get_diskname(disk)); 2680 2681 g_mirror_destroy_disk(disk); 2682 sc->sc_ndisks--; 2683 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2684 g_mirror_update_metadata(disk); 2685 } 2686 break; 2687 } 2688 default: 2689 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2690 break; 2691 } 2692 return (0); 2693} 2694#undef DISK_STATE_CHANGED 2695 2696int 2697g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2698{ 2699 struct g_provider *pp; 2700 u_char *buf; 2701 int error; 2702 2703 g_topology_assert(); 2704 2705 error = g_access(cp, 1, 0, 0); 2706 if (error != 0) 2707 return (error); 2708 pp = cp->provider; 2709 g_topology_unlock(); 2710 /* Metadata are stored on last sector. */ 2711 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2712 &error); 2713 g_topology_lock(); 2714 g_access(cp, -1, 0, 0); 2715 if (buf == NULL) { 2716 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2717 cp->provider->name, error); 2718 return (error); 2719 } 2720 2721 /* Decode metadata. */ 2722 error = mirror_metadata_decode(buf, md); 2723 g_free(buf); 2724 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2725 return (EINVAL); 2726 if (md->md_version > G_MIRROR_VERSION) { 2727 G_MIRROR_DEBUG(0, 2728 "Kernel module is too old to handle metadata from %s.", 2729 cp->provider->name); 2730 return (EINVAL); 2731 } 2732 if (error != 0) { 2733 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2734 cp->provider->name); 2735 return (error); 2736 } 2737 2738 return (0); 2739} 2740 2741static int 2742g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2743 struct g_mirror_metadata *md) 2744{ 2745 2746 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2747 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2748 pp->name, md->md_did); 2749 return (EEXIST); 2750 } 2751 if (md->md_all != sc->sc_ndisks) { 2752 G_MIRROR_DEBUG(1, 2753 "Invalid '%s' field on disk %s (device %s), skipping.", 2754 "md_all", pp->name, sc->sc_name); 2755 return (EINVAL); 2756 } 2757 if (md->md_slice != sc->sc_slice) { 2758 G_MIRROR_DEBUG(1, 2759 "Invalid '%s' field on disk %s (device %s), skipping.", 2760 "md_slice", pp->name, sc->sc_name); 2761 return (EINVAL); 2762 } 2763 if (md->md_balance != sc->sc_balance) { 2764 G_MIRROR_DEBUG(1, 2765 "Invalid '%s' field on disk %s (device %s), skipping.", 2766 "md_balance", pp->name, sc->sc_name); 2767 return (EINVAL); 2768 } 2769#if 0 2770 if (md->md_mediasize != sc->sc_mediasize) { 2771 G_MIRROR_DEBUG(1, 2772 "Invalid '%s' field on disk %s (device %s), skipping.", 2773 "md_mediasize", pp->name, sc->sc_name); 2774 return (EINVAL); 2775 } 2776#endif 2777 if (sc->sc_mediasize > pp->mediasize) { 2778 G_MIRROR_DEBUG(1, 2779 "Invalid size of disk %s (device %s), skipping.", pp->name, 2780 sc->sc_name); 2781 return (EINVAL); 2782 } 2783 if (md->md_sectorsize != sc->sc_sectorsize) { 2784 G_MIRROR_DEBUG(1, 2785 "Invalid '%s' field on disk %s (device %s), skipping.", 2786 "md_sectorsize", pp->name, sc->sc_name); 2787 return (EINVAL); 2788 } 2789 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2790 G_MIRROR_DEBUG(1, 2791 "Invalid sector size of disk %s (device %s), skipping.", 2792 pp->name, sc->sc_name); 2793 return (EINVAL); 2794 } 2795 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2796 G_MIRROR_DEBUG(1, 2797 "Invalid device flags on disk %s (device %s), skipping.", 2798 pp->name, sc->sc_name); 2799 return (EINVAL); 2800 } 2801 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2802 G_MIRROR_DEBUG(1, 2803 "Invalid disk flags on disk %s (device %s), skipping.", 2804 pp->name, sc->sc_name); 2805 return (EINVAL); 2806 } 2807 return (0); 2808} 2809 2810int 2811g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2812 struct g_mirror_metadata *md) 2813{ 2814 struct g_mirror_disk *disk; 2815 int error; 2816 2817 g_topology_assert_not(); 2818 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2819 2820 error = g_mirror_check_metadata(sc, pp, md); 2821 if (error != 0) 2822 return (error); 2823 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2824 md->md_genid < sc->sc_genid) { 2825 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2826 pp->name, sc->sc_name); 2827 return (EINVAL); 2828 } 2829 disk = g_mirror_init_disk(sc, pp, md, &error); 2830 if (disk == NULL) 2831 return (error); 2832 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2833 G_MIRROR_EVENT_WAIT); 2834 if (error != 0) 2835 return (error); 2836 if (md->md_version < G_MIRROR_VERSION) { 2837 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2838 pp->name, md->md_version, G_MIRROR_VERSION); 2839 g_mirror_update_metadata(disk); 2840 } 2841 return (0); 2842} 2843 2844static void 2845g_mirror_destroy_delayed(void *arg, int flag) 2846{ 2847 struct g_mirror_softc *sc; 2848 int error; 2849 2850 if (flag == EV_CANCEL) { 2851 G_MIRROR_DEBUG(1, "Destroying canceled."); 2852 return; 2853 } 2854 sc = arg; 2855 g_topology_unlock(); 2856 sx_xlock(&sc->sc_lock); 2857 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2858 ("DESTROY flag set on %s.", sc->sc_name)); 2859 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0, 2860 ("DESTROYING flag not set on %s.", sc->sc_name)); 2861 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 2862 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 2863 if (error != 0) { 2864 G_MIRROR_DEBUG(0, "Cannot destroy %s (error=%d).", 2865 sc->sc_name, error); 2866 sx_xunlock(&sc->sc_lock); 2867 } 2868 g_topology_lock(); 2869} 2870 2871static int 2872g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2873{ 2874 struct g_mirror_softc *sc; 2875 int dcr, dcw, dce, error = 0; 2876 2877 g_topology_assert(); 2878 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2879 acw, ace); 2880 2881 sc = pp->geom->softc; 2882 if (sc == NULL && acr <= 0 && acw <= 0 && ace <= 0) 2883 return (0); 2884 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 2885 2886 dcr = pp->acr + acr; 2887 dcw = pp->acw + acw; 2888 dce = pp->ace + ace; 2889 2890 g_topology_unlock(); 2891 sx_xlock(&sc->sc_lock); 2892 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 2893 LIST_EMPTY(&sc->sc_disks)) { 2894 if (acr > 0 || acw > 0 || ace > 0) 2895 error = ENXIO; 2896 goto end; 2897 } 2898 if (dcw == 0) 2899 g_mirror_idle(sc, dcw); 2900 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0) { 2901 if (acr > 0 || acw > 0 || ace > 0) { 2902 error = ENXIO; 2903 goto end; 2904 } 2905 if (dcr == 0 && dcw == 0 && dce == 0) { 2906 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, 2907 sc, NULL); 2908 } 2909 } 2910end: 2911 sx_xunlock(&sc->sc_lock); 2912 g_topology_lock(); 2913 return (error); 2914} 2915 2916static struct g_geom * 2917g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2918{ 2919 struct g_mirror_softc *sc; 2920 struct g_geom *gp; 2921 int error, timeout; 2922 2923 g_topology_assert(); 2924 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2925 md->md_mid); 2926 2927 /* One disk is minimum. */ 2928 if (md->md_all < 1) 2929 return (NULL); 2930 /* 2931 * Action geom. 2932 */ 2933 gp = g_new_geomf(mp, "%s", md->md_name); 2934 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2935 gp->start = g_mirror_start; 2936 gp->orphan = g_mirror_orphan; 2937 gp->access = g_mirror_access; 2938 gp->dumpconf = g_mirror_dumpconf; 2939 2940 sc->sc_id = md->md_mid; 2941 sc->sc_slice = md->md_slice; 2942 sc->sc_balance = md->md_balance; 2943 sc->sc_mediasize = md->md_mediasize; 2944 sc->sc_sectorsize = md->md_sectorsize; 2945 sc->sc_ndisks = md->md_all; 2946 sc->sc_flags = md->md_mflags; 2947 sc->sc_bump_id = 0; 2948 sc->sc_idle = 1; 2949 sc->sc_last_write = time_uptime; 2950 sc->sc_writes = 0; 2951 sx_init(&sc->sc_lock, "gmirror:lock"); 2952 bioq_init(&sc->sc_queue); 2953 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2954 bioq_init(&sc->sc_regular_delayed); 2955 bioq_init(&sc->sc_inflight); 2956 bioq_init(&sc->sc_sync_delayed); 2957 LIST_INIT(&sc->sc_disks); 2958 TAILQ_INIT(&sc->sc_events); 2959 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2960 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2961 mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF); 2962 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2963 gp->softc = sc; 2964 sc->sc_geom = gp; 2965 sc->sc_provider = NULL; 2966 /* 2967 * Synchronization geom. 2968 */ 2969 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2970 gp->softc = sc; 2971 gp->orphan = g_mirror_orphan; 2972 sc->sc_sync.ds_geom = gp; 2973 sc->sc_sync.ds_ndisks = 0; 2974 error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2975 "g_mirror %s", md->md_name); 2976 if (error != 0) { 2977 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2978 sc->sc_name); 2979 g_destroy_geom(sc->sc_sync.ds_geom); 2980 mtx_destroy(&sc->sc_done_mtx); 2981 mtx_destroy(&sc->sc_events_mtx); 2982 mtx_destroy(&sc->sc_queue_mtx); 2983 sx_destroy(&sc->sc_lock); 2984 g_destroy_geom(sc->sc_geom); 2985 free(sc, M_MIRROR); 2986 return (NULL); 2987 } 2988 2989 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 2990 sc->sc_name, sc->sc_ndisks, sc->sc_id); 2991 2992 sc->sc_rootmount = root_mount_hold("GMIRROR"); 2993 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 2994 /* 2995 * Run timeout. 2996 */ 2997 timeout = g_mirror_timeout * hz; 2998 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2999 return (sc->sc_geom); 3000} 3001 3002int 3003g_mirror_destroy(struct g_mirror_softc *sc, int how) 3004{ 3005 struct g_mirror_disk *disk; 3006 struct g_provider *pp; 3007 3008 g_topology_assert_not(); 3009 if (sc == NULL) 3010 return (ENXIO); 3011 sx_assert(&sc->sc_lock, SX_XLOCKED); 3012 3013 pp = sc->sc_provider; 3014 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 3015 switch (how) { 3016 case G_MIRROR_DESTROY_SOFT: 3017 G_MIRROR_DEBUG(1, 3018 "Device %s is still open (r%dw%de%d).", pp->name, 3019 pp->acr, pp->acw, pp->ace); 3020 return (EBUSY); 3021 case G_MIRROR_DESTROY_DELAYED: 3022 G_MIRROR_DEBUG(1, 3023 "Device %s will be destroyed on last close.", 3024 pp->name); 3025 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 3026 if (disk->d_state == 3027 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3028 g_mirror_sync_stop(disk, 1); 3029 } 3030 } 3031 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROYING; 3032 return (EBUSY); 3033 case G_MIRROR_DESTROY_HARD: 3034 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 3035 "can't be definitely removed.", pp->name); 3036 } 3037 } 3038 3039 g_topology_lock(); 3040 if (sc->sc_geom->softc == NULL) { 3041 g_topology_unlock(); 3042 return (0); 3043 } 3044 sc->sc_geom->softc = NULL; 3045 sc->sc_sync.ds_geom->softc = NULL; 3046 g_topology_unlock(); 3047 3048 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 3049 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 3050 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3051 sx_xunlock(&sc->sc_lock); 3052 mtx_lock(&sc->sc_queue_mtx); 3053 wakeup(sc); 3054 mtx_unlock(&sc->sc_queue_mtx); 3055 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3056 while (sc->sc_worker != NULL) 3057 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 3058 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3059 sx_xlock(&sc->sc_lock); 3060 g_mirror_destroy_device(sc); 3061 free(sc, M_MIRROR); 3062 return (0); 3063} 3064 3065static void 3066g_mirror_taste_orphan(struct g_consumer *cp) 3067{ 3068 3069 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3070 cp->provider->name)); 3071} 3072 3073static struct g_geom * 3074g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3075{ 3076 struct g_mirror_metadata md; 3077 struct g_mirror_softc *sc; 3078 struct g_consumer *cp; 3079 struct g_geom *gp; 3080 int error; 3081 3082 g_topology_assert(); 3083 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3084 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 3085 3086 gp = g_new_geomf(mp, "mirror:taste"); 3087 /* 3088 * This orphan function should be never called. 3089 */ 3090 gp->orphan = g_mirror_taste_orphan; 3091 cp = g_new_consumer(gp); 3092 g_attach(cp, pp); 3093 error = g_mirror_read_metadata(cp, &md); 3094 g_detach(cp); 3095 g_destroy_consumer(cp); 3096 g_destroy_geom(gp); 3097 if (error != 0) 3098 return (NULL); 3099 gp = NULL; 3100 3101 if (md.md_provider[0] != '\0' && 3102 !g_compare_names(md.md_provider, pp->name)) 3103 return (NULL); 3104 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3105 return (NULL); 3106 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 3107 G_MIRROR_DEBUG(0, 3108 "Device %s: provider %s marked as inactive, skipping.", 3109 md.md_name, pp->name); 3110 return (NULL); 3111 } 3112 if (g_mirror_debug >= 2) 3113 mirror_metadata_dump(&md); 3114 3115 /* 3116 * Let's check if device already exists. 3117 */ 3118 sc = NULL; 3119 LIST_FOREACH(gp, &mp->geom, geom) { 3120 sc = gp->softc; 3121 if (sc == NULL) 3122 continue; 3123 if (sc->sc_sync.ds_geom == gp) 3124 continue; 3125 if (strcmp(md.md_name, sc->sc_name) != 0) 3126 continue; 3127 if (md.md_mid != sc->sc_id) { 3128 G_MIRROR_DEBUG(0, "Device %s already configured.", 3129 sc->sc_name); 3130 return (NULL); 3131 } 3132 break; 3133 } 3134 if (gp == NULL) { 3135 gp = g_mirror_create(mp, &md); 3136 if (gp == NULL) { 3137 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3138 md.md_name); 3139 return (NULL); 3140 } 3141 sc = gp->softc; 3142 } 3143 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3144 g_topology_unlock(); 3145 sx_xlock(&sc->sc_lock); 3146 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; 3147 error = g_mirror_add_disk(sc, pp, &md); 3148 if (error != 0) { 3149 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3150 pp->name, gp->name, error); 3151 if (LIST_EMPTY(&sc->sc_disks)) { 3152 g_cancel_event(sc); 3153 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3154 g_topology_lock(); 3155 return (NULL); 3156 } 3157 gp = NULL; 3158 } 3159 sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; 3160 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3161 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3162 g_topology_lock(); 3163 return (NULL); 3164 } 3165 sx_xunlock(&sc->sc_lock); 3166 g_topology_lock(); 3167 return (gp); 3168} 3169 3170static void 3171g_mirror_resize(struct g_consumer *cp) 3172{ 3173 struct g_mirror_disk *disk; 3174 3175 g_topology_assert(); 3176 g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name); 3177 3178 disk = cp->private; 3179 if (disk == NULL) 3180 return; 3181 g_topology_unlock(); 3182 g_mirror_update_metadata(disk); 3183 g_topology_lock(); 3184} 3185 3186static int 3187g_mirror_destroy_geom(struct gctl_req *req __unused, 3188 struct g_class *mp __unused, struct g_geom *gp) 3189{ 3190 struct g_mirror_softc *sc; 3191 int error; 3192 3193 g_topology_unlock(); 3194 sc = gp->softc; 3195 sx_xlock(&sc->sc_lock); 3196 g_cancel_event(sc); 3197 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3198 if (error != 0) 3199 sx_xunlock(&sc->sc_lock); 3200 g_topology_lock(); 3201 return (error); 3202} 3203 3204static void 3205g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3206 struct g_consumer *cp, struct g_provider *pp) 3207{ 3208 struct g_mirror_softc *sc; 3209 3210 g_topology_assert(); 3211 3212 sc = gp->softc; 3213 if (sc == NULL) 3214 return; 3215 /* Skip synchronization geom. */ 3216 if (gp == sc->sc_sync.ds_geom) 3217 return; 3218 if (pp != NULL) { 3219 /* Nothing here. */ 3220 } else if (cp != NULL) { 3221 struct g_mirror_disk *disk; 3222 3223 disk = cp->private; 3224 if (disk == NULL) 3225 return; 3226 g_topology_unlock(); 3227 sx_xlock(&sc->sc_lock); 3228 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3229 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3230 sbuf_printf(sb, "%s<Synchronized>", indent); 3231 if (disk->d_sync.ds_offset == 0) 3232 sbuf_printf(sb, "0%%"); 3233 else { 3234 sbuf_printf(sb, "%u%%", 3235 (u_int)((disk->d_sync.ds_offset * 100) / 3236 sc->sc_provider->mediasize)); 3237 } 3238 sbuf_printf(sb, "</Synchronized>\n"); 3239 if (disk->d_sync.ds_offset > 0) { 3240 sbuf_printf(sb, "%s<BytesSynced>%jd" 3241 "</BytesSynced>\n", indent, 3242 (intmax_t)disk->d_sync.ds_offset); 3243 } 3244 } 3245 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3246 disk->d_sync.ds_syncid); 3247 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3248 disk->d_genid); 3249 sbuf_printf(sb, "%s<Flags>", indent); 3250 if (disk->d_flags == 0) 3251 sbuf_printf(sb, "NONE"); 3252 else { 3253 int first = 1; 3254 3255#define ADD_FLAG(flag, name) do { \ 3256 if ((disk->d_flags & (flag)) != 0) { \ 3257 if (!first) \ 3258 sbuf_printf(sb, ", "); \ 3259 else \ 3260 first = 0; \ 3261 sbuf_printf(sb, name); \ 3262 } \ 3263} while (0) 3264 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3265 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3266 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3267 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3268 "SYNCHRONIZING"); 3269 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3270 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3271#undef ADD_FLAG 3272 } 3273 sbuf_printf(sb, "</Flags>\n"); 3274 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3275 disk->d_priority); 3276 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3277 g_mirror_disk_state2str(disk->d_state)); 3278 sx_xunlock(&sc->sc_lock); 3279 g_topology_lock(); 3280 } else { 3281 g_topology_unlock(); 3282 sx_xlock(&sc->sc_lock); 3283 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3284 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3285 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3286 sbuf_printf(sb, "%s<Flags>", indent); 3287 if (sc->sc_flags == 0) 3288 sbuf_printf(sb, "NONE"); 3289 else { 3290 int first = 1; 3291 3292#define ADD_FLAG(flag, name) do { \ 3293 if ((sc->sc_flags & (flag)) != 0) { \ 3294 if (!first) \ 3295 sbuf_printf(sb, ", "); \ 3296 else \ 3297 first = 0; \ 3298 sbuf_printf(sb, name); \ 3299 } \ 3300} while (0) 3301 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3302 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3303#undef ADD_FLAG 3304 } 3305 sbuf_printf(sb, "</Flags>\n"); 3306 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3307 (u_int)sc->sc_slice); 3308 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3309 balance_name(sc->sc_balance)); 3310 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3311 sc->sc_ndisks); 3312 sbuf_printf(sb, "%s<State>", indent); 3313 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3314 sbuf_printf(sb, "%s", "STARTING"); 3315 else if (sc->sc_ndisks == 3316 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3317 sbuf_printf(sb, "%s", "COMPLETE"); 3318 else 3319 sbuf_printf(sb, "%s", "DEGRADED"); 3320 sbuf_printf(sb, "</State>\n"); 3321 sx_xunlock(&sc->sc_lock); 3322 g_topology_lock(); 3323 } 3324} 3325 3326static void 3327g_mirror_shutdown_post_sync(void *arg, int howto) 3328{ 3329 struct g_class *mp; 3330 struct g_geom *gp, *gp2; 3331 struct g_mirror_softc *sc; 3332 int error; 3333 3334 mp = arg; 3335 DROP_GIANT(); 3336 g_topology_lock(); 3337 g_mirror_shutdown = 1; 3338 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3339 if ((sc = gp->softc) == NULL) 3340 continue; 3341 /* Skip synchronization geom. */ 3342 if (gp == sc->sc_sync.ds_geom) 3343 continue; 3344 g_topology_unlock(); 3345 sx_xlock(&sc->sc_lock); 3346 g_mirror_idle(sc, -1); 3347 g_cancel_event(sc); 3348 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3349 if (error != 0) 3350 sx_xunlock(&sc->sc_lock); 3351 g_topology_lock(); 3352 } 3353 g_topology_unlock(); 3354 PICKUP_GIANT(); 3355} 3356 3357static void 3358g_mirror_init(struct g_class *mp) 3359{ 3360 3361 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3362 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3363 if (g_mirror_post_sync == NULL) 3364 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3365} 3366 3367static void 3368g_mirror_fini(struct g_class *mp) 3369{ 3370 3371 if (g_mirror_post_sync != NULL) 3372 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); 3373} 3374 3375DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3376