g_mirror.c revision 139054
1/*- 2 * Copyright (c) 2004 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: head/sys/geom/mirror/g_mirror.c 139054 2004-12-19 23:55:49Z pjd $"); 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/sysctl.h> 39#include <sys/malloc.h> 40#include <sys/eventhandler.h> 41#include <vm/uma.h> 42#include <geom/geom.h> 43#include <sys/proc.h> 44#include <sys/kthread.h> 45#include <geom/mirror/g_mirror.h> 46 47 48static MALLOC_DEFINE(M_MIRROR, "mirror data", "GEOM_MIRROR Data"); 49 50SYSCTL_DECL(_kern_geom); 51SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, "GEOM_MIRROR stuff"); 52u_int g_mirror_debug = 0; 53TUNABLE_INT("kern.geom.mirror.debug", &g_mirror_debug); 54SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RW, &g_mirror_debug, 0, 55 "Debug level"); 56static u_int g_mirror_timeout = 4; 57TUNABLE_INT("kern.geom.mirror.timeout", &g_mirror_timeout); 58SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RW, &g_mirror_timeout, 59 0, "Time to wait on all mirror components"); 60static u_int g_mirror_idletime = 5; 61TUNABLE_INT("kern.geom.mirror.idletime", &g_mirror_idletime); 62SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RW, 63 &g_mirror_idletime, 0, "Mark components as clean when idling"); 64static u_int g_mirror_reqs_per_sync = 5; 65SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, reqs_per_sync, CTLFLAG_RW, 66 &g_mirror_reqs_per_sync, 0, 67 "Number of regular I/O requests per synchronization request"); 68static u_int g_mirror_syncs_per_sec = 100; 69SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, syncs_per_sec, CTLFLAG_RW, 70 &g_mirror_syncs_per_sec, 0, 71 "Number of synchronizations requests per second"); 72 73#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 74 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 75 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 76 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 77} while (0) 78 79static eventhandler_tag g_mirror_ehtag = NULL; 80 81static int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp, 82 struct g_geom *gp); 83static g_taste_t g_mirror_taste; 84static void g_mirror_init(struct g_class *mp); 85static void g_mirror_fini(struct g_class *mp); 86 87struct g_class g_mirror_class = { 88 .name = G_MIRROR_CLASS_NAME, 89 .version = G_VERSION, 90 .ctlreq = g_mirror_config, 91 .taste = g_mirror_taste, 92 .destroy_geom = g_mirror_destroy_geom, 93 .init = g_mirror_init, 94 .fini = g_mirror_fini 95}; 96 97 98static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 99static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 100static void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force); 101static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 102 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 103static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 104 105 106static const char * 107g_mirror_disk_state2str(int state) 108{ 109 110 switch (state) { 111 case G_MIRROR_DISK_STATE_NONE: 112 return ("NONE"); 113 case G_MIRROR_DISK_STATE_NEW: 114 return ("NEW"); 115 case G_MIRROR_DISK_STATE_ACTIVE: 116 return ("ACTIVE"); 117 case G_MIRROR_DISK_STATE_STALE: 118 return ("STALE"); 119 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 120 return ("SYNCHRONIZING"); 121 case G_MIRROR_DISK_STATE_DISCONNECTED: 122 return ("DISCONNECTED"); 123 case G_MIRROR_DISK_STATE_DESTROY: 124 return ("DESTROY"); 125 default: 126 return ("INVALID"); 127 } 128} 129 130static const char * 131g_mirror_device_state2str(int state) 132{ 133 134 switch (state) { 135 case G_MIRROR_DEVICE_STATE_STARTING: 136 return ("STARTING"); 137 case G_MIRROR_DEVICE_STATE_RUNNING: 138 return ("RUNNING"); 139 default: 140 return ("INVALID"); 141 } 142} 143 144static const char * 145g_mirror_get_diskname(struct g_mirror_disk *disk) 146{ 147 148 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 149 return ("[unknown]"); 150 return (disk->d_name); 151} 152 153/* 154 * --- Events handling functions --- 155 * Events in geom_mirror are used to maintain disks and device status 156 * from one thread to simplify locking. 157 */ 158static void 159g_mirror_event_free(struct g_mirror_event *ep) 160{ 161 162 free(ep, M_MIRROR); 163} 164 165int 166g_mirror_event_send(void *arg, int state, int flags) 167{ 168 struct g_mirror_softc *sc; 169 struct g_mirror_disk *disk; 170 struct g_mirror_event *ep; 171 int error; 172 173 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 174 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 175 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 176 disk = NULL; 177 sc = arg; 178 } else { 179 disk = arg; 180 sc = disk->d_softc; 181 } 182 ep->e_disk = disk; 183 ep->e_state = state; 184 ep->e_flags = flags; 185 ep->e_error = 0; 186 mtx_lock(&sc->sc_events_mtx); 187 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 188 mtx_unlock(&sc->sc_events_mtx); 189 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 190 mtx_lock(&sc->sc_queue_mtx); 191 wakeup(sc); 192 mtx_unlock(&sc->sc_queue_mtx); 193 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 194 return (0); 195 g_topology_assert(); 196 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 197 g_topology_unlock(); 198 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 199 mtx_lock(&sc->sc_events_mtx); 200 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 201 hz * 5); 202 } 203 /* Don't even try to use 'sc' here, because it could be already dead. */ 204 g_topology_lock(); 205 error = ep->e_error; 206 g_mirror_event_free(ep); 207 return (error); 208} 209 210static struct g_mirror_event * 211g_mirror_event_get(struct g_mirror_softc *sc) 212{ 213 struct g_mirror_event *ep; 214 215 mtx_lock(&sc->sc_events_mtx); 216 ep = TAILQ_FIRST(&sc->sc_events); 217 if (ep != NULL) 218 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 219 mtx_unlock(&sc->sc_events_mtx); 220 return (ep); 221} 222 223static void 224g_mirror_event_cancel(struct g_mirror_disk *disk) 225{ 226 struct g_mirror_softc *sc; 227 struct g_mirror_event *ep, *tmpep; 228 229 g_topology_assert(); 230 231 sc = disk->d_softc; 232 mtx_lock(&sc->sc_events_mtx); 233 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 234 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 235 continue; 236 if (ep->e_disk != disk) 237 continue; 238 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 239 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 240 g_mirror_event_free(ep); 241 else { 242 ep->e_error = ECANCELED; 243 wakeup(ep); 244 } 245 } 246 mtx_unlock(&sc->sc_events_mtx); 247} 248 249/* 250 * Return the number of disks in given state. 251 * If state is equal to -1, count all connected disks. 252 */ 253u_int 254g_mirror_ndisks(struct g_mirror_softc *sc, int state) 255{ 256 struct g_mirror_disk *disk; 257 u_int n = 0; 258 259 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 260 if (state == -1 || disk->d_state == state) 261 n++; 262 } 263 return (n); 264} 265 266/* 267 * Find a disk in mirror by its disk ID. 268 */ 269static struct g_mirror_disk * 270g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 271{ 272 struct g_mirror_disk *disk; 273 274 g_topology_assert(); 275 276 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 277 if (disk->d_id == id) 278 return (disk); 279 } 280 return (NULL); 281} 282 283static u_int 284g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 285{ 286 struct bio *bp; 287 u_int nreqs = 0; 288 289 mtx_lock(&sc->sc_queue_mtx); 290 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 291 if (bp->bio_from == cp) 292 nreqs++; 293 } 294 mtx_unlock(&sc->sc_queue_mtx); 295 return (nreqs); 296} 297 298static int 299g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 300{ 301 302 if (cp->index > 0) { 303 G_MIRROR_DEBUG(2, 304 "I/O requests for %s exist, can't destroy it now.", 305 cp->provider->name); 306 return (1); 307 } 308 if (g_mirror_nrequests(sc, cp) > 0) { 309 G_MIRROR_DEBUG(2, 310 "I/O requests for %s in queue, can't destroy it now.", 311 cp->provider->name); 312 return (1); 313 } 314 return (0); 315} 316 317static void 318g_mirror_destroy_consumer(void *arg, int flags __unused) 319{ 320 struct g_consumer *cp; 321 322 cp = arg; 323 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 324 g_detach(cp); 325 g_destroy_consumer(cp); 326} 327 328static void 329g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 330{ 331 struct g_provider *pp; 332 int retaste_wait; 333 334 g_topology_assert(); 335 336 cp->private = NULL; 337 if (g_mirror_is_busy(sc, cp)) 338 return; 339 pp = cp->provider; 340 retaste_wait = 0; 341 if (cp->acw == 1) { 342 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 343 retaste_wait = 1; 344 } 345 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 346 -cp->acw, -cp->ace, 0); 347 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 348 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 349 if (retaste_wait) { 350 /* 351 * After retaste event was send (inside g_access()), we can send 352 * event to detach and destroy consumer. 353 * A class, which has consumer to the given provider connected 354 * will not receive retaste event for the provider. 355 * This is the way how I ignore retaste events when I close 356 * consumers opened for write: I detach and destroy consumer 357 * after retaste event is sent. 358 */ 359 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 360 return; 361 } 362 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 363 g_detach(cp); 364 g_destroy_consumer(cp); 365} 366 367static int 368g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 369{ 370 int error; 371 372 g_topology_assert(); 373 KASSERT(disk->d_consumer == NULL, 374 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 375 376 disk->d_consumer = g_new_consumer(disk->d_softc->sc_geom); 377 disk->d_consumer->private = disk; 378 disk->d_consumer->index = 0; 379 error = g_attach(disk->d_consumer, pp); 380 if (error != 0) 381 return (error); 382 error = g_access(disk->d_consumer, 1, 1, 1); 383 if (error != 0) { 384 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 385 pp->name, error); 386 return (error); 387 } 388 389 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 390 return (0); 391} 392 393static void 394g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 395{ 396 397 g_topology_assert(); 398 399 if (cp == NULL) 400 return; 401 if (cp->provider != NULL) 402 g_mirror_kill_consumer(sc, cp); 403 else 404 g_destroy_consumer(cp); 405} 406 407/* 408 * Initialize disk. This means allocate memory, create consumer, attach it 409 * to the provider and open access (r1w1e1) to it. 410 */ 411static struct g_mirror_disk * 412g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 413 struct g_mirror_metadata *md, int *errorp) 414{ 415 struct g_mirror_disk *disk; 416 int error; 417 418 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 419 if (disk == NULL) { 420 error = ENOMEM; 421 goto fail; 422 } 423 disk->d_softc = sc; 424 error = g_mirror_connect_disk(disk, pp); 425 if (error != 0) 426 goto fail; 427 disk->d_id = md->md_did; 428 disk->d_state = G_MIRROR_DISK_STATE_NONE; 429 disk->d_priority = md->md_priority; 430 disk->d_delay.sec = 0; 431 disk->d_delay.frac = 0; 432 binuptime(&disk->d_last_used); 433 disk->d_flags = md->md_dflags; 434 if (md->md_provider[0] != '\0') 435 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 436 disk->d_sync.ds_consumer = NULL; 437 disk->d_sync.ds_offset = md->md_sync_offset; 438 disk->d_sync.ds_offset_done = md->md_sync_offset; 439 disk->d_sync.ds_resync = -1; 440 disk->d_sync.ds_syncid = md->md_syncid; 441 if (errorp != NULL) 442 *errorp = 0; 443 return (disk); 444fail: 445 if (errorp != NULL) 446 *errorp = error; 447 if (disk != NULL) { 448 g_mirror_disconnect_consumer(sc, disk->d_consumer); 449 free(disk, M_MIRROR); 450 } 451 return (NULL); 452} 453 454static void 455g_mirror_destroy_disk(struct g_mirror_disk *disk) 456{ 457 struct g_mirror_softc *sc; 458 459 g_topology_assert(); 460 461 LIST_REMOVE(disk, d_next); 462 g_mirror_event_cancel(disk); 463 sc = disk->d_softc; 464 if (sc->sc_hint == disk) 465 sc->sc_hint = NULL; 466 switch (disk->d_state) { 467 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 468 g_mirror_sync_stop(disk, 1); 469 /* FALLTHROUGH */ 470 case G_MIRROR_DISK_STATE_NEW: 471 case G_MIRROR_DISK_STATE_STALE: 472 case G_MIRROR_DISK_STATE_ACTIVE: 473 g_mirror_disconnect_consumer(sc, disk->d_consumer); 474 free(disk, M_MIRROR); 475 break; 476 default: 477 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 478 g_mirror_get_diskname(disk), 479 g_mirror_disk_state2str(disk->d_state))); 480 } 481} 482 483static void 484g_mirror_destroy_device(struct g_mirror_softc *sc) 485{ 486 struct g_mirror_disk *disk; 487 struct g_mirror_event *ep; 488 struct g_geom *gp; 489 struct g_consumer *cp, *tmpcp; 490 491 g_topology_assert(); 492 493 gp = sc->sc_geom; 494 if (sc->sc_provider != NULL) 495 g_mirror_destroy_provider(sc); 496 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 497 disk = LIST_FIRST(&sc->sc_disks)) { 498 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 499 g_mirror_update_metadata(disk); 500 g_mirror_destroy_disk(disk); 501 } 502 while ((ep = g_mirror_event_get(sc)) != NULL) { 503 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 504 g_mirror_event_free(ep); 505 else { 506 ep->e_error = ECANCELED; 507 ep->e_flags |= G_MIRROR_EVENT_DONE; 508 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 509 mtx_lock(&sc->sc_events_mtx); 510 wakeup(ep); 511 mtx_unlock(&sc->sc_events_mtx); 512 } 513 } 514 callout_drain(&sc->sc_callout); 515 gp->softc = NULL; 516 517 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 518 g_mirror_disconnect_consumer(sc, cp); 519 } 520 sc->sc_sync.ds_geom->softc = NULL; 521 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 522 mtx_destroy(&sc->sc_queue_mtx); 523 mtx_destroy(&sc->sc_events_mtx); 524 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 525 g_wither_geom(gp, ENXIO); 526} 527 528static void 529g_mirror_orphan(struct g_consumer *cp) 530{ 531 struct g_mirror_disk *disk; 532 533 g_topology_assert(); 534 535 disk = cp->private; 536 if (disk == NULL) 537 return; 538 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 539 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 540 G_MIRROR_EVENT_DONTWAIT); 541} 542 543static void 544g_mirror_spoiled(struct g_consumer *cp) 545{ 546 struct g_mirror_disk *disk; 547 548 g_topology_assert(); 549 550 disk = cp->private; 551 if (disk == NULL) 552 return; 553 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 554 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 555 G_MIRROR_EVENT_DONTWAIT); 556} 557 558/* 559 * Function should return the next active disk on the list. 560 * It is possible that it will be the same disk as given. 561 * If there are no active disks on list, NULL is returned. 562 */ 563static __inline struct g_mirror_disk * 564g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 565{ 566 struct g_mirror_disk *dp; 567 568 for (dp = LIST_NEXT(disk, d_next); dp != disk; 569 dp = LIST_NEXT(dp, d_next)) { 570 if (dp == NULL) 571 dp = LIST_FIRST(&sc->sc_disks); 572 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 573 break; 574 } 575 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 576 return (NULL); 577 return (dp); 578} 579 580static struct g_mirror_disk * 581g_mirror_get_disk(struct g_mirror_softc *sc) 582{ 583 struct g_mirror_disk *disk; 584 585 if (sc->sc_hint == NULL) { 586 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 587 if (sc->sc_hint == NULL) 588 return (NULL); 589 } 590 disk = sc->sc_hint; 591 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 592 disk = g_mirror_find_next(sc, disk); 593 if (disk == NULL) 594 return (NULL); 595 } 596 sc->sc_hint = g_mirror_find_next(sc, disk); 597 return (disk); 598} 599 600static int 601g_mirror_write_metadata(struct g_mirror_disk *disk, 602 struct g_mirror_metadata *md) 603{ 604 struct g_mirror_softc *sc; 605 struct g_consumer *cp; 606 off_t offset, length; 607 u_char *sector; 608 int error = 0; 609 610 g_topology_assert(); 611 612 sc = disk->d_softc; 613 cp = disk->d_consumer; 614 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 615 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 616 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 617 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 618 cp->acw, cp->ace)); 619 length = cp->provider->sectorsize; 620 offset = cp->provider->mediasize - length; 621 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 622 if (md != NULL) 623 mirror_metadata_encode(md, sector); 624 g_topology_unlock(); 625 error = g_write_data(cp, offset, sector, length); 626 g_topology_lock(); 627 free(sector, M_MIRROR); 628 if (error != 0) { 629 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 630 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 631 G_MIRROR_EVENT_DONTWAIT); 632 } 633 return (error); 634} 635 636static int 637g_mirror_clear_metadata(struct g_mirror_disk *disk) 638{ 639 int error; 640 641 g_topology_assert(); 642 error = g_mirror_write_metadata(disk, NULL); 643 if (error == 0) { 644 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 645 g_mirror_get_diskname(disk)); 646 } else { 647 G_MIRROR_DEBUG(0, 648 "Cannot clear metadata on disk %s (error=%d).", 649 g_mirror_get_diskname(disk), error); 650 } 651 return (error); 652} 653 654void 655g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 656 struct g_mirror_metadata *md) 657{ 658 659 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 660 md->md_version = G_MIRROR_VERSION; 661 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 662 md->md_mid = sc->sc_id; 663 md->md_all = sc->sc_ndisks; 664 md->md_slice = sc->sc_slice; 665 md->md_balance = sc->sc_balance; 666 md->md_mediasize = sc->sc_mediasize; 667 md->md_sectorsize = sc->sc_sectorsize; 668 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 669 bzero(md->md_provider, sizeof(md->md_provider)); 670 if (disk == NULL) { 671 md->md_did = arc4random(); 672 md->md_priority = 0; 673 md->md_syncid = 0; 674 md->md_dflags = 0; 675 md->md_sync_offset = 0; 676 } else { 677 md->md_did = disk->d_id; 678 md->md_priority = disk->d_priority; 679 md->md_syncid = disk->d_sync.ds_syncid; 680 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 681 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 682 md->md_sync_offset = disk->d_sync.ds_offset_done; 683 else 684 md->md_sync_offset = 0; 685 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 686 strlcpy(md->md_provider, 687 disk->d_consumer->provider->name, 688 sizeof(md->md_provider)); 689 } 690 } 691} 692 693void 694g_mirror_update_metadata(struct g_mirror_disk *disk) 695{ 696 struct g_mirror_metadata md; 697 int error; 698 699 g_topology_assert(); 700 g_mirror_fill_metadata(disk->d_softc, disk, &md); 701 error = g_mirror_write_metadata(disk, &md); 702 if (error == 0) { 703 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 704 g_mirror_get_diskname(disk)); 705 } else { 706 G_MIRROR_DEBUG(0, 707 "Cannot update metadata on disk %s (error=%d).", 708 g_mirror_get_diskname(disk), error); 709 } 710} 711 712static void 713g_mirror_bump_syncid(struct g_mirror_softc *sc) 714{ 715 struct g_mirror_disk *disk; 716 717 g_topology_assert(); 718 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 719 ("%s called with no active disks (device=%s).", __func__, 720 sc->sc_name)); 721 722 sc->sc_syncid++; 723 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 724 sc->sc_syncid); 725 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 726 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 727 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 728 disk->d_sync.ds_syncid = sc->sc_syncid; 729 g_mirror_update_metadata(disk); 730 } 731 } 732} 733 734static void 735g_mirror_idle(struct g_mirror_softc *sc) 736{ 737 struct g_mirror_disk *disk; 738 739 if (sc->sc_provider == NULL || sc->sc_provider->acw == 0) 740 return; 741 sc->sc_idle = 1; 742 g_topology_lock(); 743 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 744 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 745 continue; 746 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 747 g_mirror_get_diskname(disk), sc->sc_name); 748 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 749 g_mirror_update_metadata(disk); 750 } 751 g_topology_unlock(); 752} 753 754static void 755g_mirror_unidle(struct g_mirror_softc *sc) 756{ 757 struct g_mirror_disk *disk; 758 759 sc->sc_idle = 0; 760 g_topology_lock(); 761 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 762 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 763 continue; 764 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 765 g_mirror_get_diskname(disk), sc->sc_name); 766 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 767 g_mirror_update_metadata(disk); 768 } 769 g_topology_unlock(); 770} 771 772/* 773 * Return 1 if we should check if mirror is idling. 774 */ 775static int 776g_mirror_check_idle(struct g_mirror_softc *sc) 777{ 778 struct g_mirror_disk *disk; 779 780 if (sc->sc_idle) 781 return (0); 782 if (sc->sc_provider != NULL && sc->sc_provider->acw == 0) 783 return (0); 784 /* 785 * Check if there are no in-flight requests. 786 */ 787 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 788 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 789 continue; 790 if (disk->d_consumer->index > 0) 791 return (0); 792 } 793 return (1); 794} 795 796static __inline int 797bintime_cmp(struct bintime *bt1, struct bintime *bt2) 798{ 799 800 if (bt1->sec < bt2->sec) 801 return (-1); 802 else if (bt1->sec > bt2->sec) 803 return (1); 804 if (bt1->frac < bt2->frac) 805 return (-1); 806 else if (bt1->frac > bt2->frac) 807 return (1); 808 return (0); 809} 810 811static void 812g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 813{ 814 815 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 816 return; 817 binuptime(&disk->d_delay); 818 bintime_sub(&disk->d_delay, &bp->bio_t0); 819} 820 821static void 822g_mirror_done(struct bio *bp) 823{ 824 struct g_mirror_softc *sc; 825 826 sc = bp->bio_from->geom->softc; 827 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 828 mtx_lock(&sc->sc_queue_mtx); 829 bioq_disksort(&sc->sc_queue, bp); 830 wakeup(sc); 831 mtx_unlock(&sc->sc_queue_mtx); 832} 833 834static void 835g_mirror_regular_request(struct bio *bp) 836{ 837 struct g_mirror_softc *sc; 838 struct g_mirror_disk *disk; 839 struct bio *pbp; 840 841 g_topology_assert_not(); 842 843 bp->bio_from->index--; 844 pbp = bp->bio_parent; 845 sc = pbp->bio_to->geom->softc; 846 disk = bp->bio_from->private; 847 if (disk == NULL) { 848 g_topology_lock(); 849 g_mirror_kill_consumer(sc, bp->bio_from); 850 g_topology_unlock(); 851 } else { 852 g_mirror_update_delay(disk, bp); 853 } 854 855 pbp->bio_inbed++; 856 KASSERT(pbp->bio_inbed <= pbp->bio_children, 857 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 858 pbp->bio_children)); 859 if (bp->bio_error == 0 && pbp->bio_error == 0) { 860 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 861 g_destroy_bio(bp); 862 if (pbp->bio_children == pbp->bio_inbed) { 863 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 864 pbp->bio_completed = pbp->bio_length; 865 g_io_deliver(pbp, pbp->bio_error); 866 } 867 return; 868 } else if (bp->bio_error != 0) { 869 if (pbp->bio_error == 0) 870 pbp->bio_error = bp->bio_error; 871 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 872 bp->bio_error); 873 if (disk != NULL) { 874 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 875 g_mirror_event_send(disk, 876 G_MIRROR_DISK_STATE_DISCONNECTED, 877 G_MIRROR_EVENT_DONTWAIT); 878 } 879 switch (pbp->bio_cmd) { 880 case BIO_DELETE: 881 case BIO_WRITE: 882 pbp->bio_inbed--; 883 pbp->bio_children--; 884 break; 885 } 886 } 887 g_destroy_bio(bp); 888 889 switch (pbp->bio_cmd) { 890 case BIO_READ: 891 if (pbp->bio_children == pbp->bio_inbed) { 892 pbp->bio_error = 0; 893 mtx_lock(&sc->sc_queue_mtx); 894 bioq_disksort(&sc->sc_queue, pbp); 895 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 896 wakeup(sc); 897 mtx_unlock(&sc->sc_queue_mtx); 898 } 899 break; 900 case BIO_DELETE: 901 case BIO_WRITE: 902 if (pbp->bio_children == 0) { 903 /* 904 * All requests failed. 905 */ 906 } else if (pbp->bio_inbed < pbp->bio_children) { 907 /* Do nothing. */ 908 break; 909 } else if (pbp->bio_children == pbp->bio_inbed) { 910 /* Some requests succeeded. */ 911 pbp->bio_error = 0; 912 pbp->bio_completed = pbp->bio_length; 913 } 914 g_io_deliver(pbp, pbp->bio_error); 915 break; 916 default: 917 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 918 break; 919 } 920} 921 922static void 923g_mirror_sync_done(struct bio *bp) 924{ 925 struct g_mirror_softc *sc; 926 927 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 928 sc = bp->bio_from->geom->softc; 929 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 930 mtx_lock(&sc->sc_queue_mtx); 931 bioq_disksort(&sc->sc_queue, bp); 932 wakeup(sc); 933 mtx_unlock(&sc->sc_queue_mtx); 934} 935 936static void 937g_mirror_start(struct bio *bp) 938{ 939 struct g_mirror_softc *sc; 940 941 sc = bp->bio_to->geom->softc; 942 /* 943 * If sc == NULL or there are no valid disks, provider's error 944 * should be set and g_mirror_start() should not be called at all. 945 */ 946 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 947 ("Provider's error should be set (error=%d)(mirror=%s).", 948 bp->bio_to->error, bp->bio_to->name)); 949 G_MIRROR_LOGREQ(3, bp, "Request received."); 950 951 switch (bp->bio_cmd) { 952 case BIO_READ: 953 case BIO_WRITE: 954 case BIO_DELETE: 955 break; 956 case BIO_GETATTR: 957 default: 958 g_io_deliver(bp, EOPNOTSUPP); 959 return; 960 } 961 mtx_lock(&sc->sc_queue_mtx); 962 bioq_disksort(&sc->sc_queue, bp); 963 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 964 wakeup(sc); 965 mtx_unlock(&sc->sc_queue_mtx); 966} 967 968/* 969 * Send one synchronization request. 970 */ 971static void 972g_mirror_sync_one(struct g_mirror_disk *disk) 973{ 974 struct g_mirror_softc *sc; 975 struct bio *bp; 976 977 sc = disk->d_softc; 978 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 979 ("Disk %s is not marked for synchronization.", 980 g_mirror_get_diskname(disk))); 981 982 bp = g_new_bio(); 983 if (bp == NULL) 984 return; 985 bp->bio_parent = NULL; 986 bp->bio_cmd = BIO_READ; 987 bp->bio_offset = disk->d_sync.ds_offset; 988 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 989 bp->bio_cflags = 0; 990 bp->bio_done = g_mirror_sync_done; 991 bp->bio_data = disk->d_sync.ds_data; 992 if (bp->bio_data == NULL) { 993 g_destroy_bio(bp); 994 return; 995 } 996 disk->d_sync.ds_offset += bp->bio_length; 997 bp->bio_to = sc->sc_provider; 998 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 999 disk->d_sync.ds_consumer->index++; 1000 g_io_request(bp, disk->d_sync.ds_consumer); 1001} 1002 1003static void 1004g_mirror_sync_request(struct bio *bp) 1005{ 1006 struct g_mirror_softc *sc; 1007 struct g_mirror_disk *disk; 1008 1009 bp->bio_from->index--; 1010 sc = bp->bio_from->geom->softc; 1011 disk = bp->bio_from->private; 1012 if (disk == NULL) { 1013 g_topology_lock(); 1014 g_mirror_kill_consumer(sc, bp->bio_from); 1015 g_topology_unlock(); 1016 g_destroy_bio(bp); 1017 return; 1018 } 1019 1020 /* 1021 * Synchronization request. 1022 */ 1023 switch (bp->bio_cmd) { 1024 case BIO_READ: 1025 { 1026 struct g_consumer *cp; 1027 1028 if (bp->bio_error != 0) { 1029 G_MIRROR_LOGREQ(0, bp, 1030 "Synchronization request failed (error=%d).", 1031 bp->bio_error); 1032 g_destroy_bio(bp); 1033 return; 1034 } 1035 G_MIRROR_LOGREQ(3, bp, 1036 "Synchronization request half-finished."); 1037 bp->bio_cmd = BIO_WRITE; 1038 bp->bio_cflags = 0; 1039 cp = disk->d_consumer; 1040 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1041 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1042 cp->acr, cp->acw, cp->ace)); 1043 cp->index++; 1044 g_io_request(bp, cp); 1045 return; 1046 } 1047 case BIO_WRITE: 1048 { 1049 struct g_mirror_disk_sync *sync; 1050 1051 if (bp->bio_error != 0) { 1052 G_MIRROR_LOGREQ(0, bp, 1053 "Synchronization request failed (error=%d).", 1054 bp->bio_error); 1055 g_destroy_bio(bp); 1056 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 1057 g_mirror_event_send(disk, 1058 G_MIRROR_DISK_STATE_DISCONNECTED, 1059 G_MIRROR_EVENT_DONTWAIT); 1060 return; 1061 } 1062 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1063 sync = &disk->d_sync; 1064 sync->ds_offset_done = bp->bio_offset + bp->bio_length; 1065 g_destroy_bio(bp); 1066 if (sync->ds_resync != -1) 1067 break; 1068 if (sync->ds_offset_done == sc->sc_provider->mediasize) { 1069 /* 1070 * Disk up-to-date, activate it. 1071 */ 1072 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1073 G_MIRROR_EVENT_DONTWAIT); 1074 return; 1075 } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) { 1076 /* 1077 * Update offset_done on every 100 blocks. 1078 * XXX: This should be configurable. 1079 */ 1080 g_topology_lock(); 1081 g_mirror_update_metadata(disk); 1082 g_topology_unlock(); 1083 } 1084 return; 1085 } 1086 default: 1087 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1088 bp->bio_cmd, sc->sc_name)); 1089 break; 1090 } 1091} 1092 1093static void 1094g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1095{ 1096 struct g_mirror_disk *disk; 1097 struct g_consumer *cp; 1098 struct bio *cbp; 1099 1100 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1101 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1102 break; 1103 } 1104 if (disk == NULL) { 1105 if (bp->bio_error == 0) 1106 bp->bio_error = ENXIO; 1107 g_io_deliver(bp, bp->bio_error); 1108 return; 1109 } 1110 cbp = g_clone_bio(bp); 1111 if (cbp == NULL) { 1112 if (bp->bio_error == 0) 1113 bp->bio_error = ENOMEM; 1114 g_io_deliver(bp, bp->bio_error); 1115 return; 1116 } 1117 /* 1118 * Fill in the component buf structure. 1119 */ 1120 cp = disk->d_consumer; 1121 cbp->bio_done = g_mirror_done; 1122 cbp->bio_to = cp->provider; 1123 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1124 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1125 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1126 cp->acw, cp->ace)); 1127 cp->index++; 1128 g_io_request(cbp, cp); 1129} 1130 1131static void 1132g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1133{ 1134 struct g_mirror_disk *disk; 1135 struct g_consumer *cp; 1136 struct bio *cbp; 1137 1138 disk = g_mirror_get_disk(sc); 1139 if (disk == NULL) { 1140 if (bp->bio_error == 0) 1141 bp->bio_error = ENXIO; 1142 g_io_deliver(bp, bp->bio_error); 1143 return; 1144 } 1145 cbp = g_clone_bio(bp); 1146 if (cbp == NULL) { 1147 if (bp->bio_error == 0) 1148 bp->bio_error = ENOMEM; 1149 g_io_deliver(bp, bp->bio_error); 1150 return; 1151 } 1152 /* 1153 * Fill in the component buf structure. 1154 */ 1155 cp = disk->d_consumer; 1156 cbp->bio_done = g_mirror_done; 1157 cbp->bio_to = cp->provider; 1158 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1159 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1160 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1161 cp->acw, cp->ace)); 1162 cp->index++; 1163 g_io_request(cbp, cp); 1164} 1165 1166static void 1167g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1168{ 1169 struct g_mirror_disk *disk, *dp; 1170 struct g_consumer *cp; 1171 struct bio *cbp; 1172 struct bintime curtime; 1173 1174 binuptime(&curtime); 1175 /* 1176 * Find a disk which the smallest load. 1177 */ 1178 disk = NULL; 1179 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1180 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1181 continue; 1182 /* If disk wasn't used for more than 2 sec, use it. */ 1183 if (curtime.sec - dp->d_last_used.sec >= 2) { 1184 disk = dp; 1185 break; 1186 } 1187 if (disk == NULL || 1188 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1189 disk = dp; 1190 } 1191 } 1192 cbp = g_clone_bio(bp); 1193 if (cbp == NULL) { 1194 if (bp->bio_error == 0) 1195 bp->bio_error = ENOMEM; 1196 g_io_deliver(bp, bp->bio_error); 1197 return; 1198 } 1199 /* 1200 * Fill in the component buf structure. 1201 */ 1202 cp = disk->d_consumer; 1203 cbp->bio_done = g_mirror_done; 1204 cbp->bio_to = cp->provider; 1205 binuptime(&disk->d_last_used); 1206 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1207 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1208 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1209 cp->acw, cp->ace)); 1210 cp->index++; 1211 g_io_request(cbp, cp); 1212} 1213 1214static void 1215g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1216{ 1217 struct bio_queue_head queue; 1218 struct g_mirror_disk *disk; 1219 struct g_consumer *cp; 1220 struct bio *cbp; 1221 off_t left, mod, offset, slice; 1222 u_char *data; 1223 u_int ndisks; 1224 1225 if (bp->bio_length <= sc->sc_slice) { 1226 g_mirror_request_round_robin(sc, bp); 1227 return; 1228 } 1229 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1230 slice = bp->bio_length / ndisks; 1231 mod = slice % sc->sc_provider->sectorsize; 1232 if (mod != 0) 1233 slice += sc->sc_provider->sectorsize - mod; 1234 /* 1235 * Allocate all bios before sending any request, so we can 1236 * return ENOMEM in nice and clean way. 1237 */ 1238 left = bp->bio_length; 1239 offset = bp->bio_offset; 1240 data = bp->bio_data; 1241 bioq_init(&queue); 1242 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1243 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1244 continue; 1245 cbp = g_clone_bio(bp); 1246 if (cbp == NULL) { 1247 for (cbp = bioq_first(&queue); cbp != NULL; 1248 cbp = bioq_first(&queue)) { 1249 bioq_remove(&queue, cbp); 1250 g_destroy_bio(cbp); 1251 } 1252 if (bp->bio_error == 0) 1253 bp->bio_error = ENOMEM; 1254 g_io_deliver(bp, bp->bio_error); 1255 return; 1256 } 1257 bioq_insert_tail(&queue, cbp); 1258 cbp->bio_done = g_mirror_done; 1259 cbp->bio_caller1 = disk; 1260 cbp->bio_to = disk->d_consumer->provider; 1261 cbp->bio_offset = offset; 1262 cbp->bio_data = data; 1263 cbp->bio_length = MIN(left, slice); 1264 left -= cbp->bio_length; 1265 if (left == 0) 1266 break; 1267 offset += cbp->bio_length; 1268 data += cbp->bio_length; 1269 } 1270 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1271 bioq_remove(&queue, cbp); 1272 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1273 disk = cbp->bio_caller1; 1274 cbp->bio_caller1 = NULL; 1275 cp = disk->d_consumer; 1276 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1277 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1278 cp->acr, cp->acw, cp->ace)); 1279 disk->d_consumer->index++; 1280 g_io_request(cbp, disk->d_consumer); 1281 } 1282} 1283 1284static void 1285g_mirror_register_request(struct bio *bp) 1286{ 1287 struct g_mirror_softc *sc; 1288 1289 sc = bp->bio_to->geom->softc; 1290 switch (bp->bio_cmd) { 1291 case BIO_READ: 1292 switch (sc->sc_balance) { 1293 case G_MIRROR_BALANCE_LOAD: 1294 g_mirror_request_load(sc, bp); 1295 break; 1296 case G_MIRROR_BALANCE_PREFER: 1297 g_mirror_request_prefer(sc, bp); 1298 break; 1299 case G_MIRROR_BALANCE_ROUND_ROBIN: 1300 g_mirror_request_round_robin(sc, bp); 1301 break; 1302 case G_MIRROR_BALANCE_SPLIT: 1303 g_mirror_request_split(sc, bp); 1304 break; 1305 } 1306 return; 1307 case BIO_WRITE: 1308 case BIO_DELETE: 1309 { 1310 struct g_mirror_disk *disk; 1311 struct g_mirror_disk_sync *sync; 1312 struct bio_queue_head queue; 1313 struct g_consumer *cp; 1314 struct bio *cbp; 1315 1316 if (sc->sc_idle) 1317 g_mirror_unidle(sc); 1318 /* 1319 * Allocate all bios before sending any request, so we can 1320 * return ENOMEM in nice and clean way. 1321 */ 1322 bioq_init(&queue); 1323 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1324 sync = &disk->d_sync; 1325 switch (disk->d_state) { 1326 case G_MIRROR_DISK_STATE_ACTIVE: 1327 break; 1328 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1329 if (bp->bio_offset >= sync->ds_offset) 1330 continue; 1331 else if (bp->bio_offset + bp->bio_length > 1332 sync->ds_offset_done && 1333 (bp->bio_offset < sync->ds_resync || 1334 sync->ds_resync == -1)) { 1335 sync->ds_resync = bp->bio_offset - 1336 (bp->bio_offset % MAXPHYS); 1337 } 1338 break; 1339 default: 1340 continue; 1341 } 1342 cbp = g_clone_bio(bp); 1343 if (cbp == NULL) { 1344 for (cbp = bioq_first(&queue); cbp != NULL; 1345 cbp = bioq_first(&queue)) { 1346 bioq_remove(&queue, cbp); 1347 g_destroy_bio(cbp); 1348 } 1349 if (bp->bio_error == 0) 1350 bp->bio_error = ENOMEM; 1351 g_io_deliver(bp, bp->bio_error); 1352 return; 1353 } 1354 bioq_insert_tail(&queue, cbp); 1355 cbp->bio_done = g_mirror_done; 1356 cp = disk->d_consumer; 1357 cbp->bio_caller1 = cp; 1358 cbp->bio_to = cp->provider; 1359 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1360 ("Consumer %s not opened (r%dw%de%d).", 1361 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1362 } 1363 for (cbp = bioq_first(&queue); cbp != NULL; 1364 cbp = bioq_first(&queue)) { 1365 bioq_remove(&queue, cbp); 1366 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1367 cp = cbp->bio_caller1; 1368 cbp->bio_caller1 = NULL; 1369 cp->index++; 1370 g_io_request(cbp, cp); 1371 } 1372 /* 1373 * Bump syncid on first write. 1374 */ 1375 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) { 1376 sc->sc_bump_syncid = 0; 1377 g_topology_lock(); 1378 g_mirror_bump_syncid(sc); 1379 g_topology_unlock(); 1380 } 1381 return; 1382 } 1383 default: 1384 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1385 bp->bio_cmd, sc->sc_name)); 1386 break; 1387 } 1388} 1389 1390static int 1391g_mirror_can_destroy(struct g_mirror_softc *sc) 1392{ 1393 struct g_geom *gp; 1394 struct g_consumer *cp; 1395 1396 g_topology_assert(); 1397 gp = sc->sc_geom; 1398 LIST_FOREACH(cp, &gp->consumer, consumer) { 1399 if (g_mirror_is_busy(sc, cp)) 1400 return (0); 1401 } 1402 gp = sc->sc_sync.ds_geom; 1403 LIST_FOREACH(cp, &gp->consumer, consumer) { 1404 if (g_mirror_is_busy(sc, cp)) 1405 return (0); 1406 } 1407 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1408 sc->sc_name); 1409 return (1); 1410} 1411 1412static int 1413g_mirror_try_destroy(struct g_mirror_softc *sc) 1414{ 1415 1416 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1417 g_topology_lock(); 1418 if (!g_mirror_can_destroy(sc)) { 1419 g_topology_unlock(); 1420 return (0); 1421 } 1422 g_topology_unlock(); 1423 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1424 &sc->sc_worker); 1425 wakeup(&sc->sc_worker); 1426 sc->sc_worker = NULL; 1427 } else { 1428 g_topology_lock(); 1429 if (!g_mirror_can_destroy(sc)) { 1430 g_topology_unlock(); 1431 return (0); 1432 } 1433 g_mirror_destroy_device(sc); 1434 g_topology_unlock(); 1435 free(sc, M_MIRROR); 1436 } 1437 return (1); 1438} 1439 1440/* 1441 * Worker thread. 1442 */ 1443static void 1444g_mirror_worker(void *arg) 1445{ 1446 struct g_mirror_softc *sc; 1447 struct g_mirror_disk *disk; 1448 struct g_mirror_disk_sync *sync; 1449 struct g_mirror_event *ep; 1450 struct bio *bp; 1451 u_int nreqs; 1452 1453 sc = arg; 1454 curthread->td_base_pri = PRIBIO; 1455 1456 nreqs = 0; 1457 for (;;) { 1458 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1459 /* 1460 * First take a look at events. 1461 * This is important to handle events before any I/O requests. 1462 */ 1463 ep = g_mirror_event_get(sc); 1464 if (ep != NULL) { 1465 g_topology_lock(); 1466 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1467 /* Update only device status. */ 1468 G_MIRROR_DEBUG(3, 1469 "Running event for device %s.", 1470 sc->sc_name); 1471 ep->e_error = 0; 1472 g_mirror_update_device(sc, 1); 1473 } else { 1474 /* Update disk status. */ 1475 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1476 g_mirror_get_diskname(ep->e_disk)); 1477 ep->e_error = g_mirror_update_disk(ep->e_disk, 1478 ep->e_state); 1479 if (ep->e_error == 0) 1480 g_mirror_update_device(sc, 0); 1481 } 1482 g_topology_unlock(); 1483 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1484 KASSERT(ep->e_error == 0, 1485 ("Error cannot be handled.")); 1486 g_mirror_event_free(ep); 1487 } else { 1488 ep->e_flags |= G_MIRROR_EVENT_DONE; 1489 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1490 ep); 1491 mtx_lock(&sc->sc_events_mtx); 1492 wakeup(ep); 1493 mtx_unlock(&sc->sc_events_mtx); 1494 } 1495 if ((sc->sc_flags & 1496 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1497 if (g_mirror_try_destroy(sc)) 1498 kthread_exit(0); 1499 } 1500 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1501 continue; 1502 } 1503 /* 1504 * Now I/O requests. 1505 */ 1506 /* Get first request from the queue. */ 1507 mtx_lock(&sc->sc_queue_mtx); 1508 bp = bioq_first(&sc->sc_queue); 1509 if (bp == NULL) { 1510 if ((sc->sc_flags & 1511 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1512 mtx_unlock(&sc->sc_queue_mtx); 1513 if (g_mirror_try_destroy(sc)) 1514 kthread_exit(0); 1515 mtx_lock(&sc->sc_queue_mtx); 1516 } 1517 } 1518 if (sc->sc_sync.ds_ndisks > 0 && 1519 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1520 mtx_unlock(&sc->sc_queue_mtx); 1521 /* 1522 * It is time for synchronization... 1523 */ 1524 nreqs = 0; 1525 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1526 if (disk->d_state != 1527 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1528 continue; 1529 } 1530 sync = &disk->d_sync; 1531 if (sync->ds_offset >= 1532 sc->sc_provider->mediasize) { 1533 continue; 1534 } 1535 if (sync->ds_offset > sync->ds_offset_done) 1536 continue; 1537 if (sync->ds_resync != -1) { 1538 sync->ds_offset = sync->ds_resync; 1539 sync->ds_offset_done = sync->ds_resync; 1540 sync->ds_resync = -1; 1541 } 1542 g_mirror_sync_one(disk); 1543 } 1544 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1545 goto sleep; 1546 } 1547 if (bp == NULL) { 1548 if (g_mirror_check_idle(sc)) { 1549 u_int idletime; 1550 1551 idletime = g_mirror_idletime; 1552 if (idletime == 0) 1553 idletime = 1; 1554 idletime *= hz; 1555 if (msleep(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1556 "m:w1", idletime) == EWOULDBLOCK) { 1557 G_MIRROR_DEBUG(5, "%s: I'm here 3.", 1558 __func__); 1559 /* 1560 * No I/O requests in 'idletime' seconds, 1561 * so mark components as clean. 1562 */ 1563 g_mirror_idle(sc); 1564 } 1565 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1566 } else { 1567 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1568 "m:w2", 0); 1569 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1570 } 1571 continue; 1572 } 1573 nreqs++; 1574 bioq_remove(&sc->sc_queue, bp); 1575 mtx_unlock(&sc->sc_queue_mtx); 1576 1577 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1578 g_mirror_regular_request(bp); 1579 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1580 u_int timeout, sps; 1581 1582 g_mirror_sync_request(bp); 1583sleep: 1584 sps = g_mirror_syncs_per_sec; 1585 if (sps == 0) { 1586 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1587 continue; 1588 } 1589 mtx_lock(&sc->sc_queue_mtx); 1590 if (bioq_first(&sc->sc_queue) != NULL) { 1591 mtx_unlock(&sc->sc_queue_mtx); 1592 G_MIRROR_DEBUG(5, "%s: I'm here 7.", __func__); 1593 continue; 1594 } 1595 timeout = hz / sps; 1596 if (timeout == 0) 1597 timeout = 1; 1598 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w3", 1599 timeout); 1600 } else { 1601 g_mirror_register_request(bp); 1602 } 1603 G_MIRROR_DEBUG(5, "%s: I'm here 8.", __func__); 1604 } 1605} 1606 1607/* 1608 * Open disk's consumer if needed. 1609 */ 1610static void 1611g_mirror_update_access(struct g_mirror_disk *disk) 1612{ 1613 struct g_provider *pp; 1614 1615 g_topology_assert(); 1616 1617 pp = disk->d_softc->sc_provider; 1618 if (pp == NULL) 1619 return; 1620 if (pp->acw > 0) { 1621 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1622 G_MIRROR_DEBUG(1, 1623 "Disk %s (device %s) marked as dirty.", 1624 g_mirror_get_diskname(disk), 1625 disk->d_softc->sc_name); 1626 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1627 } 1628 } else if (pp->acw == 0) { 1629 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1630 G_MIRROR_DEBUG(1, 1631 "Disk %s (device %s) marked as clean.", 1632 g_mirror_get_diskname(disk), 1633 disk->d_softc->sc_name); 1634 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1635 } 1636 } 1637} 1638 1639static void 1640g_mirror_sync_start(struct g_mirror_disk *disk) 1641{ 1642 struct g_mirror_softc *sc; 1643 int error; 1644 1645 g_topology_assert(); 1646 1647 sc = disk->d_softc; 1648 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1649 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1650 sc->sc_state)); 1651 1652 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1653 g_mirror_get_diskname(disk)); 1654 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1655 KASSERT(disk->d_sync.ds_consumer == NULL, 1656 ("Sync consumer already exists (device=%s, disk=%s).", 1657 sc->sc_name, g_mirror_get_diskname(disk))); 1658 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1659 disk->d_sync.ds_consumer->private = disk; 1660 disk->d_sync.ds_consumer->index = 0; 1661 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1662 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1663 disk->d_softc->sc_name, error)); 1664 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1665 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1666 disk->d_softc->sc_name, error)); 1667 disk->d_sync.ds_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1668 sc->sc_sync.ds_ndisks++; 1669} 1670 1671/* 1672 * Stop synchronization process. 1673 * type: 0 - synchronization finished 1674 * 1 - synchronization stopped 1675 */ 1676static void 1677g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1678{ 1679 1680 g_topology_assert(); 1681 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1682 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1683 g_mirror_disk_state2str(disk->d_state))); 1684 if (disk->d_sync.ds_consumer == NULL) 1685 return; 1686 1687 if (type == 0) { 1688 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1689 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1690 } else /* if (type == 1) */ { 1691 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1692 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1693 } 1694 g_mirror_kill_consumer(disk->d_softc, disk->d_sync.ds_consumer); 1695 free(disk->d_sync.ds_data, M_MIRROR); 1696 disk->d_sync.ds_consumer = NULL; 1697 disk->d_softc->sc_sync.ds_ndisks--; 1698 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1699} 1700 1701static void 1702g_mirror_launch_provider(struct g_mirror_softc *sc) 1703{ 1704 struct g_mirror_disk *disk; 1705 struct g_provider *pp; 1706 1707 g_topology_assert(); 1708 1709 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1710 pp->mediasize = sc->sc_mediasize; 1711 pp->sectorsize = sc->sc_sectorsize; 1712 sc->sc_provider = pp; 1713 g_error_provider(pp, 0); 1714 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1715 pp->name); 1716 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1717 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1718 g_mirror_sync_start(disk); 1719 } 1720} 1721 1722static void 1723g_mirror_destroy_provider(struct g_mirror_softc *sc) 1724{ 1725 struct g_mirror_disk *disk; 1726 struct bio *bp; 1727 1728 g_topology_assert(); 1729 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1730 sc->sc_name)); 1731 1732 g_error_provider(sc->sc_provider, ENXIO); 1733 mtx_lock(&sc->sc_queue_mtx); 1734 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1735 bioq_remove(&sc->sc_queue, bp); 1736 g_io_deliver(bp, ENXIO); 1737 } 1738 mtx_unlock(&sc->sc_queue_mtx); 1739 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1740 sc->sc_provider->name); 1741 sc->sc_provider->flags |= G_PF_WITHER; 1742 g_orphan_provider(sc->sc_provider, ENXIO); 1743 sc->sc_provider = NULL; 1744 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1745 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1746 g_mirror_sync_stop(disk, 1); 1747 } 1748} 1749 1750static void 1751g_mirror_go(void *arg) 1752{ 1753 struct g_mirror_softc *sc; 1754 1755 sc = arg; 1756 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1757 g_mirror_event_send(sc, 0, 1758 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1759} 1760 1761static u_int 1762g_mirror_determine_state(struct g_mirror_disk *disk) 1763{ 1764 struct g_mirror_softc *sc; 1765 u_int state; 1766 1767 sc = disk->d_softc; 1768 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1769 if ((disk->d_flags & 1770 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1771 /* Disk does not need synchronization. */ 1772 state = G_MIRROR_DISK_STATE_ACTIVE; 1773 } else { 1774 if ((sc->sc_flags & 1775 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1776 (disk->d_flags & 1777 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1778 /* 1779 * We can start synchronization from 1780 * the stored offset. 1781 */ 1782 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1783 } else { 1784 state = G_MIRROR_DISK_STATE_STALE; 1785 } 1786 } 1787 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1788 /* 1789 * Reset all synchronization data for this disk, 1790 * because if it even was synchronized, it was 1791 * synchronized to disks with different syncid. 1792 */ 1793 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1794 disk->d_sync.ds_offset = 0; 1795 disk->d_sync.ds_offset_done = 0; 1796 disk->d_sync.ds_syncid = sc->sc_syncid; 1797 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1798 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1799 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1800 } else { 1801 state = G_MIRROR_DISK_STATE_STALE; 1802 } 1803 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1804 /* 1805 * Not good, NOT GOOD! 1806 * It means that mirror was started on stale disks 1807 * and more fresh disk just arrive. 1808 * If there were writes, mirror is fucked up, sorry. 1809 * I think the best choice here is don't touch 1810 * this disk and inform the user laudly. 1811 */ 1812 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1813 "disk (%s) arrives!! It will not be connected to the " 1814 "running device.", sc->sc_name, 1815 g_mirror_get_diskname(disk)); 1816 g_mirror_destroy_disk(disk); 1817 state = G_MIRROR_DISK_STATE_NONE; 1818 /* Return immediately, because disk was destroyed. */ 1819 return (state); 1820 } 1821 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1822 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1823 return (state); 1824} 1825 1826/* 1827 * Update device state. 1828 */ 1829static void 1830g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1831{ 1832 struct g_mirror_disk *disk; 1833 u_int state; 1834 1835 g_topology_assert(); 1836 1837 switch (sc->sc_state) { 1838 case G_MIRROR_DEVICE_STATE_STARTING: 1839 { 1840 struct g_mirror_disk *pdisk; 1841 u_int dirty, ndisks, syncid; 1842 1843 KASSERT(sc->sc_provider == NULL, 1844 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1845 /* 1846 * Are we ready? We are, if all disks are connected or 1847 * if we have any disks and 'force' is true. 1848 */ 1849 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1850 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1851 ; 1852 } else if (g_mirror_ndisks(sc, -1) == 0) { 1853 /* 1854 * Disks went down in starting phase, so destroy 1855 * device. 1856 */ 1857 callout_drain(&sc->sc_callout); 1858 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1859 return; 1860 } else { 1861 return; 1862 } 1863 1864 /* 1865 * Activate all disks with the biggest syncid. 1866 */ 1867 if (force) { 1868 /* 1869 * If 'force' is true, we have been called due to 1870 * timeout, so don't bother canceling timeout. 1871 */ 1872 ndisks = 0; 1873 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1874 if ((disk->d_flags & 1875 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1876 ndisks++; 1877 } 1878 } 1879 if (ndisks == 0) { 1880 /* No valid disks found, destroy device. */ 1881 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1882 return; 1883 } 1884 } else { 1885 /* Cancel timeout. */ 1886 callout_drain(&sc->sc_callout); 1887 } 1888 1889 /* 1890 * Find disk with the biggest syncid. 1891 */ 1892 syncid = 0; 1893 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1894 if (disk->d_sync.ds_syncid > syncid) 1895 syncid = disk->d_sync.ds_syncid; 1896 } 1897 1898 /* 1899 * Here we need to look for dirty disks and if all disks 1900 * with the biggest syncid are dirty, we have to choose 1901 * one with the biggest priority and rebuild the rest. 1902 */ 1903 /* 1904 * Find the number of dirty disks with the biggest syncid. 1905 * Find the number of disks with the biggest syncid. 1906 * While here, find a disk with the biggest priority. 1907 */ 1908 dirty = ndisks = 0; 1909 pdisk = NULL; 1910 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1911 if (disk->d_sync.ds_syncid != syncid) 1912 continue; 1913 if ((disk->d_flags & 1914 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1915 continue; 1916 } 1917 ndisks++; 1918 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1919 dirty++; 1920 if (pdisk == NULL || 1921 pdisk->d_priority < disk->d_priority) { 1922 pdisk = disk; 1923 } 1924 } 1925 } 1926 if (dirty == 0) { 1927 /* No dirty disks at all, great. */ 1928 } else if (dirty == ndisks) { 1929 /* 1930 * Force synchronization for all dirty disks except one 1931 * with the biggest priority. 1932 */ 1933 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1934 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1935 "master disk for synchronization.", 1936 g_mirror_get_diskname(pdisk), sc->sc_name); 1937 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1938 if (disk->d_sync.ds_syncid != syncid) 1939 continue; 1940 if ((disk->d_flags & 1941 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1942 continue; 1943 } 1944 KASSERT((disk->d_flags & 1945 G_MIRROR_DISK_FLAG_DIRTY) != 0, 1946 ("Disk %s isn't marked as dirty.", 1947 g_mirror_get_diskname(disk))); 1948 /* Skip the disk with the biggest priority. */ 1949 if (disk == pdisk) 1950 continue; 1951 disk->d_sync.ds_syncid = 0; 1952 } 1953 } else if (dirty < ndisks) { 1954 /* 1955 * Force synchronization for all dirty disks. 1956 * We have some non-dirty disks. 1957 */ 1958 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1959 if (disk->d_sync.ds_syncid != syncid) 1960 continue; 1961 if ((disk->d_flags & 1962 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1963 continue; 1964 } 1965 if ((disk->d_flags & 1966 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1967 continue; 1968 } 1969 disk->d_sync.ds_syncid = 0; 1970 } 1971 } 1972 1973 /* Reset hint. */ 1974 sc->sc_hint = NULL; 1975 sc->sc_syncid = syncid; 1976 if (force) { 1977 /* Remember to bump syncid on first write. */ 1978 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1979 } 1980 state = G_MIRROR_DEVICE_STATE_RUNNING; 1981 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 1982 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 1983 g_mirror_device_state2str(state)); 1984 sc->sc_state = state; 1985 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1986 state = g_mirror_determine_state(disk); 1987 g_mirror_event_send(disk, state, 1988 G_MIRROR_EVENT_DONTWAIT); 1989 if (state == G_MIRROR_DISK_STATE_STALE) { 1990 sc->sc_bump_syncid = 1991 G_MIRROR_BUMP_ON_FIRST_WRITE; 1992 } 1993 } 1994 wakeup(&g_mirror_class); 1995 break; 1996 } 1997 case G_MIRROR_DEVICE_STATE_RUNNING: 1998 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 1999 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2000 /* 2001 * No active disks or no disks at all, 2002 * so destroy device. 2003 */ 2004 if (sc->sc_provider != NULL) 2005 g_mirror_destroy_provider(sc); 2006 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2007 break; 2008 } else if (g_mirror_ndisks(sc, 2009 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2010 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2011 /* 2012 * We have active disks, launch provider if it doesn't 2013 * exist. 2014 */ 2015 if (sc->sc_provider == NULL) 2016 g_mirror_launch_provider(sc); 2017 } 2018 /* 2019 * Bump syncid here, if we need to do it immediately. 2020 */ 2021 if (sc->sc_bump_syncid == G_MIRROR_BUMP_IMMEDIATELY) { 2022 sc->sc_bump_syncid = 0; 2023 g_mirror_bump_syncid(sc); 2024 } 2025 break; 2026 default: 2027 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2028 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2029 break; 2030 } 2031} 2032 2033/* 2034 * Update disk state and device state if needed. 2035 */ 2036#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2037 "Disk %s state changed from %s to %s (device %s).", \ 2038 g_mirror_get_diskname(disk), \ 2039 g_mirror_disk_state2str(disk->d_state), \ 2040 g_mirror_disk_state2str(state), sc->sc_name) 2041static int 2042g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2043{ 2044 struct g_mirror_softc *sc; 2045 2046 g_topology_assert(); 2047 2048 sc = disk->d_softc; 2049again: 2050 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2051 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2052 g_mirror_disk_state2str(state)); 2053 switch (state) { 2054 case G_MIRROR_DISK_STATE_NEW: 2055 /* 2056 * Possible scenarios: 2057 * 1. New disk arrive. 2058 */ 2059 /* Previous state should be NONE. */ 2060 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2061 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2062 g_mirror_disk_state2str(disk->d_state))); 2063 DISK_STATE_CHANGED(); 2064 2065 disk->d_state = state; 2066 if (LIST_EMPTY(&sc->sc_disks)) 2067 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2068 else { 2069 struct g_mirror_disk *dp; 2070 2071 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2072 if (disk->d_priority >= dp->d_priority) { 2073 LIST_INSERT_BEFORE(dp, disk, d_next); 2074 dp = NULL; 2075 break; 2076 } 2077 if (LIST_NEXT(dp, d_next) == NULL) 2078 break; 2079 } 2080 if (dp != NULL) 2081 LIST_INSERT_AFTER(dp, disk, d_next); 2082 } 2083 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2084 sc->sc_name, g_mirror_get_diskname(disk)); 2085 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2086 break; 2087 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2088 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2089 g_mirror_device_state2str(sc->sc_state), 2090 g_mirror_get_diskname(disk), 2091 g_mirror_disk_state2str(disk->d_state))); 2092 state = g_mirror_determine_state(disk); 2093 if (state != G_MIRROR_DISK_STATE_NONE) 2094 goto again; 2095 break; 2096 case G_MIRROR_DISK_STATE_ACTIVE: 2097 /* 2098 * Possible scenarios: 2099 * 1. New disk does not need synchronization. 2100 * 2. Synchronization process finished successfully. 2101 */ 2102 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2103 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2104 g_mirror_device_state2str(sc->sc_state), 2105 g_mirror_get_diskname(disk), 2106 g_mirror_disk_state2str(disk->d_state))); 2107 /* Previous state should be NEW or SYNCHRONIZING. */ 2108 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2109 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2110 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2111 g_mirror_disk_state2str(disk->d_state))); 2112 DISK_STATE_CHANGED(); 2113 2114 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2115 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2116 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2117 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2118 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2119 g_mirror_sync_stop(disk, 0); 2120 } 2121 disk->d_state = state; 2122 disk->d_sync.ds_offset = 0; 2123 disk->d_sync.ds_offset_done = 0; 2124 g_mirror_update_access(disk); 2125 g_mirror_update_metadata(disk); 2126 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2127 sc->sc_name, g_mirror_get_diskname(disk)); 2128 break; 2129 case G_MIRROR_DISK_STATE_STALE: 2130 /* 2131 * Possible scenarios: 2132 * 1. Stale disk was connected. 2133 */ 2134 /* Previous state should be NEW. */ 2135 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2136 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2137 g_mirror_disk_state2str(disk->d_state))); 2138 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2139 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2140 g_mirror_device_state2str(sc->sc_state), 2141 g_mirror_get_diskname(disk), 2142 g_mirror_disk_state2str(disk->d_state))); 2143 /* 2144 * STALE state is only possible if device is marked 2145 * NOAUTOSYNC. 2146 */ 2147 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2148 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2149 g_mirror_device_state2str(sc->sc_state), 2150 g_mirror_get_diskname(disk), 2151 g_mirror_disk_state2str(disk->d_state))); 2152 DISK_STATE_CHANGED(); 2153 2154 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2155 disk->d_state = state; 2156 g_mirror_update_metadata(disk); 2157 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2158 sc->sc_name, g_mirror_get_diskname(disk)); 2159 break; 2160 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2161 /* 2162 * Possible scenarios: 2163 * 1. Disk which needs synchronization was connected. 2164 */ 2165 /* Previous state should be NEW. */ 2166 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2167 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2168 g_mirror_disk_state2str(disk->d_state))); 2169 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2170 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2171 g_mirror_device_state2str(sc->sc_state), 2172 g_mirror_get_diskname(disk), 2173 g_mirror_disk_state2str(disk->d_state))); 2174 DISK_STATE_CHANGED(); 2175 2176 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2177 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2178 disk->d_state = state; 2179 if (sc->sc_provider != NULL) { 2180 g_mirror_sync_start(disk); 2181 g_mirror_update_metadata(disk); 2182 } 2183 break; 2184 case G_MIRROR_DISK_STATE_DISCONNECTED: 2185 /* 2186 * Possible scenarios: 2187 * 1. Device wasn't running yet, but disk disappear. 2188 * 2. Disk was active and disapppear. 2189 * 3. Disk disappear during synchronization process. 2190 */ 2191 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2192 /* 2193 * Previous state should be ACTIVE, STALE or 2194 * SYNCHRONIZING. 2195 */ 2196 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2197 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2198 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2199 ("Wrong disk state (%s, %s).", 2200 g_mirror_get_diskname(disk), 2201 g_mirror_disk_state2str(disk->d_state))); 2202 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2203 /* Previous state should be NEW. */ 2204 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2205 ("Wrong disk state (%s, %s).", 2206 g_mirror_get_diskname(disk), 2207 g_mirror_disk_state2str(disk->d_state))); 2208 /* 2209 * Reset bumping syncid if disk disappeared in STARTING 2210 * state. 2211 */ 2212 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) 2213 sc->sc_bump_syncid = 0; 2214#ifdef INVARIANTS 2215 } else { 2216 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2217 sc->sc_name, 2218 g_mirror_device_state2str(sc->sc_state), 2219 g_mirror_get_diskname(disk), 2220 g_mirror_disk_state2str(disk->d_state))); 2221#endif 2222 } 2223 DISK_STATE_CHANGED(); 2224 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2225 sc->sc_name, g_mirror_get_diskname(disk)); 2226 2227 g_mirror_destroy_disk(disk); 2228 break; 2229 case G_MIRROR_DISK_STATE_DESTROY: 2230 { 2231 int error; 2232 2233 error = g_mirror_clear_metadata(disk); 2234 if (error != 0) 2235 return (error); 2236 DISK_STATE_CHANGED(); 2237 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2238 sc->sc_name, g_mirror_get_diskname(disk)); 2239 2240 g_mirror_destroy_disk(disk); 2241 sc->sc_ndisks--; 2242 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2243 g_mirror_update_metadata(disk); 2244 } 2245 break; 2246 } 2247 default: 2248 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2249 break; 2250 } 2251 return (0); 2252} 2253#undef DISK_STATE_CHANGED 2254 2255static int 2256g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2257{ 2258 struct g_provider *pp; 2259 u_char *buf; 2260 int error; 2261 2262 g_topology_assert(); 2263 2264 error = g_access(cp, 1, 0, 0); 2265 if (error != 0) 2266 return (error); 2267 pp = cp->provider; 2268 g_topology_unlock(); 2269 /* Metadata are stored on last sector. */ 2270 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2271 &error); 2272 g_topology_lock(); 2273 g_access(cp, -1, 0, 0); 2274 if (error != 0) { 2275 if (buf != NULL) 2276 g_free(buf); 2277 return (error); 2278 } 2279 2280 /* Decode metadata. */ 2281 error = mirror_metadata_decode(buf, md); 2282 g_free(buf); 2283 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2284 return (EINVAL); 2285 if (error != 0) { 2286 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2287 cp->provider->name); 2288 return (error); 2289 } 2290 2291 return (0); 2292} 2293 2294static int 2295g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2296 struct g_mirror_metadata *md) 2297{ 2298 2299 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2300 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2301 pp->name, md->md_did); 2302 return (EEXIST); 2303 } 2304 if (md->md_all != sc->sc_ndisks) { 2305 G_MIRROR_DEBUG(1, 2306 "Invalid '%s' field on disk %s (device %s), skipping.", 2307 "md_all", pp->name, sc->sc_name); 2308 return (EINVAL); 2309 } 2310 if (md->md_slice != sc->sc_slice) { 2311 G_MIRROR_DEBUG(1, 2312 "Invalid '%s' field on disk %s (device %s), skipping.", 2313 "md_slice", pp->name, sc->sc_name); 2314 return (EINVAL); 2315 } 2316 if (md->md_balance != sc->sc_balance) { 2317 G_MIRROR_DEBUG(1, 2318 "Invalid '%s' field on disk %s (device %s), skipping.", 2319 "md_balance", pp->name, sc->sc_name); 2320 return (EINVAL); 2321 } 2322 if (md->md_mediasize != sc->sc_mediasize) { 2323 G_MIRROR_DEBUG(1, 2324 "Invalid '%s' field on disk %s (device %s), skipping.", 2325 "md_mediasize", pp->name, sc->sc_name); 2326 return (EINVAL); 2327 } 2328 if (sc->sc_mediasize > pp->mediasize) { 2329 G_MIRROR_DEBUG(1, 2330 "Invalid size of disk %s (device %s), skipping.", pp->name, 2331 sc->sc_name); 2332 return (EINVAL); 2333 } 2334 if (md->md_sectorsize != sc->sc_sectorsize) { 2335 G_MIRROR_DEBUG(1, 2336 "Invalid '%s' field on disk %s (device %s), skipping.", 2337 "md_sectorsize", pp->name, sc->sc_name); 2338 return (EINVAL); 2339 } 2340 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2341 G_MIRROR_DEBUG(1, 2342 "Invalid sector size of disk %s (device %s), skipping.", 2343 pp->name, sc->sc_name); 2344 return (EINVAL); 2345 } 2346 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2347 G_MIRROR_DEBUG(1, 2348 "Invalid device flags on disk %s (device %s), skipping.", 2349 pp->name, sc->sc_name); 2350 return (EINVAL); 2351 } 2352 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2353 G_MIRROR_DEBUG(1, 2354 "Invalid disk flags on disk %s (device %s), skipping.", 2355 pp->name, sc->sc_name); 2356 return (EINVAL); 2357 } 2358 return (0); 2359} 2360 2361static int 2362g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2363 struct g_mirror_metadata *md) 2364{ 2365 struct g_mirror_disk *disk; 2366 int error; 2367 2368 g_topology_assert(); 2369 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2370 2371 error = g_mirror_check_metadata(sc, pp, md); 2372 if (error != 0) 2373 return (error); 2374 disk = g_mirror_init_disk(sc, pp, md, &error); 2375 if (disk == NULL) 2376 return (error); 2377 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2378 G_MIRROR_EVENT_WAIT); 2379 return (error); 2380} 2381 2382static int 2383g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2384{ 2385 struct g_mirror_softc *sc; 2386 struct g_mirror_disk *disk; 2387 int dcr, dcw, dce; 2388 2389 g_topology_assert(); 2390 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2391 acw, ace); 2392 2393 dcr = pp->acr + acr; 2394 dcw = pp->acw + acw; 2395 dce = pp->ace + ace; 2396 2397 sc = pp->geom->softc; 2398 if (sc == NULL || LIST_EMPTY(&sc->sc_disks) || 2399 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 2400 if (acr <= 0 && acw <= 0 && ace <= 0) 2401 return (0); 2402 else 2403 return (ENXIO); 2404 } 2405 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2406 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2407 continue; 2408 /* 2409 * Mark disk as dirty on open and unmark on close. 2410 */ 2411 if (pp->acw == 0 && dcw > 0) { 2412 G_MIRROR_DEBUG(1, 2413 "Disk %s (device %s) marked as dirty.", 2414 g_mirror_get_diskname(disk), sc->sc_name); 2415 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2416 g_mirror_update_metadata(disk); 2417 } else if (pp->acw > 0 && dcw == 0) { 2418 G_MIRROR_DEBUG(1, 2419 "Disk %s (device %s) marked as clean.", 2420 g_mirror_get_diskname(disk), sc->sc_name); 2421 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2422 g_mirror_update_metadata(disk); 2423 } 2424 } 2425 return (0); 2426} 2427 2428static struct g_geom * 2429g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2430{ 2431 struct g_mirror_softc *sc; 2432 struct g_geom *gp; 2433 int error, timeout; 2434 2435 g_topology_assert(); 2436 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2437 md->md_mid); 2438 2439 /* One disk is minimum. */ 2440 if (md->md_all < 1) 2441 return (NULL); 2442 /* 2443 * Action geom. 2444 */ 2445 gp = g_new_geomf(mp, "%s", md->md_name); 2446 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2447 gp->start = g_mirror_start; 2448 gp->spoiled = g_mirror_spoiled; 2449 gp->orphan = g_mirror_orphan; 2450 gp->access = g_mirror_access; 2451 gp->dumpconf = g_mirror_dumpconf; 2452 2453 sc->sc_id = md->md_mid; 2454 sc->sc_slice = md->md_slice; 2455 sc->sc_balance = md->md_balance; 2456 sc->sc_mediasize = md->md_mediasize; 2457 sc->sc_sectorsize = md->md_sectorsize; 2458 sc->sc_ndisks = md->md_all; 2459 sc->sc_flags = md->md_mflags; 2460 sc->sc_bump_syncid = 0; 2461 sc->sc_idle = 0; 2462 bioq_init(&sc->sc_queue); 2463 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2464 LIST_INIT(&sc->sc_disks); 2465 TAILQ_INIT(&sc->sc_events); 2466 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2467 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2468 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2469 gp->softc = sc; 2470 sc->sc_geom = gp; 2471 sc->sc_provider = NULL; 2472 /* 2473 * Synchronization geom. 2474 */ 2475 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2476 gp->softc = sc; 2477 gp->orphan = g_mirror_orphan; 2478 sc->sc_sync.ds_geom = gp; 2479 sc->sc_sync.ds_ndisks = 0; 2480 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2481 "g_mirror %s", md->md_name); 2482 if (error != 0) { 2483 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2484 sc->sc_name); 2485 g_destroy_geom(sc->sc_sync.ds_geom); 2486 mtx_destroy(&sc->sc_events_mtx); 2487 mtx_destroy(&sc->sc_queue_mtx); 2488 g_destroy_geom(sc->sc_geom); 2489 free(sc, M_MIRROR); 2490 return (NULL); 2491 } 2492 2493 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2494 2495 /* 2496 * Run timeout. 2497 */ 2498 timeout = g_mirror_timeout * hz; 2499 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2500 return (sc->sc_geom); 2501} 2502 2503int 2504g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2505{ 2506 struct g_provider *pp; 2507 2508 g_topology_assert(); 2509 2510 if (sc == NULL) 2511 return (ENXIO); 2512 pp = sc->sc_provider; 2513 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2514 if (force) { 2515 G_MIRROR_DEBUG(0, "Device %s is still open, so it " 2516 "can't be definitely removed.", pp->name); 2517 } else { 2518 G_MIRROR_DEBUG(1, 2519 "Device %s is still open (r%dw%de%d).", pp->name, 2520 pp->acr, pp->acw, pp->ace); 2521 return (EBUSY); 2522 } 2523 } 2524 2525 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2526 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2527 g_topology_unlock(); 2528 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2529 mtx_lock(&sc->sc_queue_mtx); 2530 wakeup(sc); 2531 mtx_unlock(&sc->sc_queue_mtx); 2532 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2533 while (sc->sc_worker != NULL) 2534 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2535 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2536 g_topology_lock(); 2537 g_mirror_destroy_device(sc); 2538 free(sc, M_MIRROR); 2539 return (0); 2540} 2541 2542static void 2543g_mirror_taste_orphan(struct g_consumer *cp) 2544{ 2545 2546 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2547 cp->provider->name)); 2548} 2549 2550static struct g_geom * 2551g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2552{ 2553 struct g_mirror_metadata md; 2554 struct g_mirror_softc *sc; 2555 struct g_consumer *cp; 2556 struct g_geom *gp; 2557 int error; 2558 2559 g_topology_assert(); 2560 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2561 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2562 2563 gp = g_new_geomf(mp, "mirror:taste"); 2564 /* 2565 * This orphan function should be never called. 2566 */ 2567 gp->orphan = g_mirror_taste_orphan; 2568 cp = g_new_consumer(gp); 2569 g_attach(cp, pp); 2570 error = g_mirror_read_metadata(cp, &md); 2571 g_detach(cp); 2572 g_destroy_consumer(cp); 2573 g_destroy_geom(gp); 2574 if (error != 0) 2575 return (NULL); 2576 gp = NULL; 2577 2578 if (md.md_version > G_MIRROR_VERSION) { 2579 printf("geom_mirror.ko module is too old to handle %s.\n", 2580 pp->name); 2581 return (NULL); 2582 } 2583 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2584 return (NULL); 2585 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2586 G_MIRROR_DEBUG(0, 2587 "Device %s: provider %s marked as inactive, skipping.", 2588 md.md_name, pp->name); 2589 return (NULL); 2590 } 2591 if (g_mirror_debug >= 2) 2592 mirror_metadata_dump(&md); 2593 2594 /* 2595 * Let's check if device already exists. 2596 */ 2597 sc = NULL; 2598 LIST_FOREACH(gp, &mp->geom, geom) { 2599 sc = gp->softc; 2600 if (sc == NULL) 2601 continue; 2602 if (sc->sc_sync.ds_geom == gp) 2603 continue; 2604 if (strcmp(md.md_name, sc->sc_name) != 0) 2605 continue; 2606 if (md.md_mid != sc->sc_id) { 2607 G_MIRROR_DEBUG(0, "Device %s already configured.", 2608 sc->sc_name); 2609 return (NULL); 2610 } 2611 break; 2612 } 2613 if (gp == NULL) { 2614 gp = g_mirror_create(mp, &md); 2615 if (gp == NULL) { 2616 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2617 md.md_name); 2618 return (NULL); 2619 } 2620 sc = gp->softc; 2621 } 2622 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2623 error = g_mirror_add_disk(sc, pp, &md); 2624 if (error != 0) { 2625 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2626 pp->name, gp->name, error); 2627 if (LIST_EMPTY(&sc->sc_disks)) 2628 g_mirror_destroy(sc, 1); 2629 return (NULL); 2630 } 2631 return (gp); 2632} 2633 2634static int 2635g_mirror_destroy_geom(struct gctl_req *req __unused, 2636 struct g_class *mp __unused, struct g_geom *gp) 2637{ 2638 2639 return (g_mirror_destroy(gp->softc, 0)); 2640} 2641 2642static void 2643g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2644 struct g_consumer *cp, struct g_provider *pp) 2645{ 2646 struct g_mirror_softc *sc; 2647 2648 g_topology_assert(); 2649 2650 sc = gp->softc; 2651 if (sc == NULL) 2652 return; 2653 /* Skip synchronization geom. */ 2654 if (gp == sc->sc_sync.ds_geom) 2655 return; 2656 if (pp != NULL) { 2657 /* Nothing here. */ 2658 } else if (cp != NULL) { 2659 struct g_mirror_disk *disk; 2660 2661 disk = cp->private; 2662 if (disk == NULL) 2663 return; 2664 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2665 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2666 sbuf_printf(sb, "%s<Synchronized>", indent); 2667 if (disk->d_sync.ds_offset_done == 0) 2668 sbuf_printf(sb, "0%%"); 2669 else { 2670 sbuf_printf(sb, "%u%%", 2671 (u_int)((disk->d_sync.ds_offset_done * 100) / 2672 sc->sc_provider->mediasize)); 2673 } 2674 sbuf_printf(sb, "</Synchronized>\n"); 2675 } 2676 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2677 disk->d_sync.ds_syncid); 2678 sbuf_printf(sb, "%s<Flags>", indent); 2679 if (disk->d_flags == 0) 2680 sbuf_printf(sb, "NONE"); 2681 else { 2682 int first = 1; 2683 2684#define ADD_FLAG(flag, name) do { \ 2685 if ((disk->d_flags & (flag)) != 0) { \ 2686 if (!first) \ 2687 sbuf_printf(sb, ", "); \ 2688 else \ 2689 first = 0; \ 2690 sbuf_printf(sb, name); \ 2691 } \ 2692} while (0) 2693 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2694 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2695 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2696 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2697 "SYNCHRONIZING"); 2698 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2699#undef ADD_FLAG 2700 } 2701 sbuf_printf(sb, "</Flags>\n"); 2702 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2703 disk->d_priority); 2704 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2705 g_mirror_disk_state2str(disk->d_state)); 2706 } else { 2707 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2708 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2709 sbuf_printf(sb, "%s<Flags>", indent); 2710 if (sc->sc_flags == 0) 2711 sbuf_printf(sb, "NONE"); 2712 else { 2713 int first = 1; 2714 2715#define ADD_FLAG(flag, name) do { \ 2716 if ((sc->sc_flags & (flag)) != 0) { \ 2717 if (!first) \ 2718 sbuf_printf(sb, ", "); \ 2719 else \ 2720 first = 0; \ 2721 sbuf_printf(sb, name); \ 2722 } \ 2723} while (0) 2724 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2725#undef ADD_FLAG 2726 } 2727 sbuf_printf(sb, "</Flags>\n"); 2728 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2729 (u_int)sc->sc_slice); 2730 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2731 balance_name(sc->sc_balance)); 2732 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2733 sc->sc_ndisks); 2734 sbuf_printf(sb, "%s<State>", indent); 2735 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2736 sbuf_printf(sb, "%s", "STARTING"); 2737 else if (sc->sc_ndisks == 2738 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 2739 sbuf_printf(sb, "%s", "COMPLETE"); 2740 else 2741 sbuf_printf(sb, "%s", "DEGRADED"); 2742 sbuf_printf(sb, "</State>\n"); 2743 } 2744} 2745 2746static void 2747g_mirror_shutdown(void *arg, int howto) 2748{ 2749 struct g_class *mp; 2750 struct g_geom *gp, *gp2; 2751 2752 mp = arg; 2753 DROP_GIANT(); 2754 g_topology_lock(); 2755 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2756 if (gp->softc == NULL) 2757 continue; 2758 g_mirror_destroy(gp->softc, 1); 2759 } 2760 g_topology_unlock(); 2761 PICKUP_GIANT(); 2762#if 0 2763 tsleep(&gp, PRIBIO, "m:shutdown", hz * 20); 2764#endif 2765} 2766 2767static void 2768g_mirror_init(struct g_class *mp) 2769{ 2770 2771 g_mirror_ehtag = EVENTHANDLER_REGISTER(shutdown_post_sync, 2772 g_mirror_shutdown, mp, SHUTDOWN_PRI_FIRST); 2773 if (g_mirror_ehtag == NULL) 2774 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 2775} 2776 2777static void 2778g_mirror_fini(struct g_class *mp) 2779{ 2780 2781 if (g_mirror_ehtag == NULL) 2782 return; 2783 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_ehtag); 2784} 2785 2786static int 2787g_mirror_can_go(void) 2788{ 2789 struct g_mirror_softc *sc; 2790 struct g_geom *gp; 2791 struct g_provider *pp; 2792 int can_go; 2793 2794 DROP_GIANT(); 2795 can_go = 1; 2796 g_topology_lock(); 2797 LIST_FOREACH(gp, &g_mirror_class.geom, geom) { 2798 sc = gp->softc; 2799 if (sc == NULL) { 2800 can_go = 0; 2801 break; 2802 } 2803 pp = sc->sc_provider; 2804 if (pp == NULL || pp->error != 0) { 2805 can_go = 0; 2806 break; 2807 } 2808 } 2809 g_topology_unlock(); 2810 PICKUP_GIANT(); 2811 return (can_go); 2812} 2813 2814static void 2815g_mirror_rootwait(void) 2816{ 2817 2818 /* 2819 * HACK: Wait for GEOM, because g_mirror_rootwait() can be called, 2820 * HACK: before we get providers for tasting. 2821 */ 2822 tsleep(&g_mirror_class, PRIBIO, "mroot", hz * 3); 2823 /* 2824 * Wait for mirrors in degraded state. 2825 */ 2826 for (;;) { 2827 if (g_mirror_can_go()) 2828 break; 2829 tsleep(&g_mirror_class, PRIBIO, "mroot", hz); 2830 } 2831} 2832 2833SYSINIT(g_mirror_root, SI_SUB_RAID, SI_ORDER_FIRST, g_mirror_rootwait, NULL) 2834 2835DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2836