g_mirror.c revision 135522
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 135522 2004-09-20 17:26:25Z 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/bitstring.h> 41#include <vm/uma.h> 42#include <machine/atomic.h> 43#include <geom/geom.h> 44#include <sys/proc.h> 45#include <sys/kthread.h> 46#include <geom/mirror/g_mirror.h> 47 48 49static MALLOC_DEFINE(M_MIRROR, "mirror data", "GEOM_MIRROR Data"); 50 51SYSCTL_DECL(_kern_geom); 52SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, "GEOM_MIRROR stuff"); 53u_int g_mirror_debug = 0; 54TUNABLE_INT("kern.geom.mirror.debug", &g_mirror_debug); 55SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RW, &g_mirror_debug, 0, 56 "Debug level"); 57static u_int g_mirror_timeout = 8; 58TUNABLE_INT("kern.geom.mirror.timeout", &g_mirror_timeout); 59SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RW, &g_mirror_timeout, 60 0, "Time to wait on all mirror components"); 61static u_int g_mirror_reqs_per_sync = 5; 62SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, reqs_per_sync, CTLFLAG_RW, 63 &g_mirror_reqs_per_sync, 0, 64 "Number of regular I/O requests per synchronization request"); 65static u_int g_mirror_syncs_per_sec = 100; 66SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, syncs_per_sec, CTLFLAG_RW, 67 &g_mirror_syncs_per_sec, 0, 68 "Number of synchronizations requests per second"); 69 70#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 71 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 72 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 73 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 74} while (0) 75 76 77static int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp, 78 struct g_geom *gp); 79static g_taste_t g_mirror_taste; 80 81struct g_class g_mirror_class = { 82 .name = G_MIRROR_CLASS_NAME, 83 .version = G_VERSION, 84 .ctlreq = g_mirror_config, 85 .taste = g_mirror_taste, 86 .destroy_geom = g_mirror_destroy_geom 87}; 88 89 90static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 91static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 92static void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force); 93static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 94 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 95static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 96 97 98static const char * 99g_mirror_disk_state2str(int state) 100{ 101 102 switch (state) { 103 case G_MIRROR_DISK_STATE_NONE: 104 return ("NONE"); 105 case G_MIRROR_DISK_STATE_NEW: 106 return ("NEW"); 107 case G_MIRROR_DISK_STATE_ACTIVE: 108 return ("ACTIVE"); 109 case G_MIRROR_DISK_STATE_STALE: 110 return ("STALE"); 111 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 112 return ("SYNCHRONIZING"); 113 case G_MIRROR_DISK_STATE_DISCONNECTED: 114 return ("DISCONNECTED"); 115 case G_MIRROR_DISK_STATE_DESTROY: 116 return ("DESTROY"); 117 default: 118 return ("INVALID"); 119 } 120} 121 122static const char * 123g_mirror_device_state2str(int state) 124{ 125 126 switch (state) { 127 case G_MIRROR_DEVICE_STATE_STARTING: 128 return ("STARTING"); 129 case G_MIRROR_DEVICE_STATE_RUNNING: 130 return ("RUNNING"); 131 default: 132 return ("INVALID"); 133 } 134} 135 136static const char * 137g_mirror_get_diskname(struct g_mirror_disk *disk) 138{ 139 140 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 141 return ("[unknown]"); 142 return (disk->d_name); 143} 144 145/* 146 * --- Events handling functions --- 147 * Events in geom_mirror are used to maintain disks and device status 148 * from one thread to simplify locking. 149 */ 150static void 151g_mirror_event_free(struct g_mirror_event *ep) 152{ 153 154 free(ep, M_MIRROR); 155} 156 157int 158g_mirror_event_send(void *arg, int state, int flags) 159{ 160 struct g_mirror_softc *sc; 161 struct g_mirror_disk *disk; 162 struct g_mirror_event *ep; 163 int error; 164 165 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 166 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 167 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 168 disk = NULL; 169 sc = arg; 170 } else { 171 disk = arg; 172 sc = disk->d_softc; 173 } 174 ep->e_disk = disk; 175 ep->e_state = state; 176 ep->e_flags = flags; 177 ep->e_error = 0; 178 mtx_lock(&sc->sc_events_mtx); 179 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 180 mtx_unlock(&sc->sc_events_mtx); 181 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 182 mtx_lock(&sc->sc_queue_mtx); 183 wakeup(sc); 184 mtx_unlock(&sc->sc_queue_mtx); 185 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 186 return (0); 187 g_topology_assert(); 188 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 189 g_topology_unlock(); 190 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 191 mtx_lock(&sc->sc_events_mtx); 192 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 193 hz * 5); 194 } 195 /* Don't even try to use 'sc' here, because it could be already dead. */ 196 g_topology_lock(); 197 error = ep->e_error; 198 g_mirror_event_free(ep); 199 return (error); 200} 201 202static struct g_mirror_event * 203g_mirror_event_get(struct g_mirror_softc *sc) 204{ 205 struct g_mirror_event *ep; 206 207 mtx_lock(&sc->sc_events_mtx); 208 ep = TAILQ_FIRST(&sc->sc_events); 209 if (ep != NULL) 210 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 211 mtx_unlock(&sc->sc_events_mtx); 212 return (ep); 213} 214 215static void 216g_mirror_event_cancel(struct g_mirror_disk *disk) 217{ 218 struct g_mirror_softc *sc; 219 struct g_mirror_event *ep, *tmpep; 220 221 g_topology_assert(); 222 223 sc = disk->d_softc; 224 mtx_lock(&sc->sc_events_mtx); 225 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 226 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 227 continue; 228 if (ep->e_disk != disk) 229 continue; 230 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 231 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 232 g_mirror_event_free(ep); 233 else { 234 ep->e_error = ECANCELED; 235 wakeup(ep); 236 } 237 } 238 mtx_unlock(&sc->sc_events_mtx); 239} 240 241/* 242 * Return the number of disks in given state. 243 * If state is equal to -1, count all connected disks. 244 */ 245u_int 246g_mirror_ndisks(struct g_mirror_softc *sc, int state) 247{ 248 struct g_mirror_disk *disk; 249 u_int n = 0; 250 251 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 252 if (state == -1 || disk->d_state == state) 253 n++; 254 } 255 return (n); 256} 257 258/* 259 * Find a disk in mirror by its disk ID. 260 */ 261static struct g_mirror_disk * 262g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 263{ 264 struct g_mirror_disk *disk; 265 266 g_topology_assert(); 267 268 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 269 if (disk->d_id == id) 270 return (disk); 271 } 272 return (NULL); 273} 274 275static u_int 276g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 277{ 278 struct bio *bp; 279 u_int nreqs = 0; 280 281 mtx_lock(&sc->sc_queue_mtx); 282 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 283 if (bp->bio_from == cp) 284 nreqs++; 285 } 286 mtx_unlock(&sc->sc_queue_mtx); 287 return (nreqs); 288} 289 290static int 291g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 292{ 293 294 if (cp->nstart != cp->nend) { 295 G_MIRROR_DEBUG(2, 296 "I/O requests for %s exist, can't destroy it now.", 297 cp->provider->name); 298 return (1); 299 } 300 if (g_mirror_nrequests(sc, cp) > 0) { 301 G_MIRROR_DEBUG(2, 302 "I/O requests for %s in queue, can't destroy it now.", 303 cp->provider->name); 304 return (1); 305 } 306 return (0); 307} 308 309static void 310g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 311{ 312 313 g_topology_assert(); 314 315 cp->private = NULL; 316 if (g_mirror_is_busy(sc, cp)) 317 return; 318 G_MIRROR_DEBUG(2, "Consumer %s destroyed.", cp->provider->name); 319 g_detach(cp); 320 g_destroy_consumer(cp); 321} 322 323static int 324g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 325{ 326 int error; 327 328 g_topology_assert(); 329 KASSERT(disk->d_consumer == NULL, 330 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 331 332 disk->d_consumer = g_new_consumer(disk->d_softc->sc_geom); 333 disk->d_consumer->private = disk; 334 error = g_attach(disk->d_consumer, pp); 335 if (error != 0) 336 return (error); 337 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 338 return (0); 339} 340 341static void 342g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 343{ 344 345 g_topology_assert(); 346 347 if (cp == NULL) 348 return; 349 if (cp->provider != NULL) { 350 G_MIRROR_DEBUG(2, "Disk %s disconnected.", cp->provider->name); 351 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) { 352 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 353 cp->provider->name, -cp->acr, -cp->acw, -cp->ace, 354 0); 355 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 356 } 357 g_mirror_kill_consumer(sc, cp); 358 } else { 359 g_destroy_consumer(cp); 360 } 361} 362 363/* 364 * Initialize disk. This means allocate memory, create consumer, attach it 365 * to the provider and open access (r1w1e1) to it. 366 */ 367static struct g_mirror_disk * 368g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 369 struct g_mirror_metadata *md, int *errorp) 370{ 371 struct g_mirror_disk *disk; 372 int error; 373 374 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 375 if (disk == NULL) { 376 error = ENOMEM; 377 goto fail; 378 } 379 disk->d_softc = sc; 380 error = g_mirror_connect_disk(disk, pp); 381 if (error != 0) 382 goto fail; 383 disk->d_id = md->md_did; 384 disk->d_state = G_MIRROR_DISK_STATE_NONE; 385 disk->d_priority = md->md_priority; 386 disk->d_delay.sec = 0; 387 disk->d_delay.frac = 0; 388 binuptime(&disk->d_last_used); 389 disk->d_flags = md->md_dflags; 390 if (md->md_provider[0] != '\0') 391 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 392 disk->d_sync.ds_consumer = NULL; 393 disk->d_sync.ds_offset = md->md_sync_offset; 394 disk->d_sync.ds_offset_done = md->md_sync_offset; 395 disk->d_sync.ds_syncid = md->md_syncid; 396 if (errorp != NULL) 397 *errorp = 0; 398 return (disk); 399fail: 400 if (errorp != NULL) 401 *errorp = error; 402 if (disk != NULL) { 403 g_mirror_disconnect_consumer(sc, disk->d_consumer); 404 free(disk, M_MIRROR); 405 } 406 return (NULL); 407} 408 409static void 410g_mirror_destroy_disk(struct g_mirror_disk *disk) 411{ 412 struct g_mirror_softc *sc; 413 414 g_topology_assert(); 415 416 LIST_REMOVE(disk, d_next); 417 g_mirror_event_cancel(disk); 418 sc = disk->d_softc; 419 if (sc->sc_hint == disk) 420 sc->sc_hint = NULL; 421 switch (disk->d_state) { 422 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 423 g_mirror_sync_stop(disk, 1); 424 /* FALLTHROUGH */ 425 case G_MIRROR_DISK_STATE_NEW: 426 case G_MIRROR_DISK_STATE_STALE: 427 case G_MIRROR_DISK_STATE_ACTIVE: 428 g_mirror_disconnect_consumer(sc, disk->d_consumer); 429 free(disk, M_MIRROR); 430 break; 431 default: 432 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 433 g_mirror_get_diskname(disk), 434 g_mirror_disk_state2str(disk->d_state))); 435 } 436} 437 438static void 439g_mirror_destroy_device(struct g_mirror_softc *sc) 440{ 441 struct g_mirror_disk *disk; 442 struct g_mirror_event *ep; 443 struct g_geom *gp; 444 struct g_consumer *cp, *tmpcp; 445 446 g_topology_assert(); 447 448 gp = sc->sc_geom; 449 if (sc->sc_provider != NULL) 450 g_mirror_destroy_provider(sc); 451 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 452 disk = LIST_FIRST(&sc->sc_disks)) { 453 g_mirror_destroy_disk(disk); 454 } 455 while ((ep = g_mirror_event_get(sc)) != NULL) { 456 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 457 g_mirror_event_free(ep); 458 else { 459 ep->e_error = ECANCELED; 460 ep->e_flags |= G_MIRROR_EVENT_DONE; 461 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 462 mtx_lock(&sc->sc_events_mtx); 463 wakeup(ep); 464 mtx_unlock(&sc->sc_events_mtx); 465 } 466 } 467 callout_drain(&sc->sc_callout); 468 gp->softc = NULL; 469 470 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 471 g_mirror_disconnect_consumer(sc, cp); 472 } 473 sc->sc_sync.ds_geom->softc = NULL; 474 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 475 mtx_destroy(&sc->sc_queue_mtx); 476 mtx_destroy(&sc->sc_events_mtx); 477 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 478 g_wither_geom(gp, ENXIO); 479} 480 481static void 482g_mirror_orphan(struct g_consumer *cp) 483{ 484 struct g_mirror_disk *disk; 485 486 g_topology_assert(); 487 488 disk = cp->private; 489 if (disk == NULL) 490 return; 491 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 492 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 493 G_MIRROR_EVENT_DONTWAIT); 494} 495 496static void 497g_mirror_spoiled(struct g_consumer *cp) 498{ 499 struct g_mirror_disk *disk; 500 501 g_topology_assert(); 502 503 disk = cp->private; 504 if (disk == NULL) 505 return; 506 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 507 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 508 G_MIRROR_EVENT_DONTWAIT); 509} 510 511/* 512 * Function should return the next active disk on the list. 513 * It is possible that it will be the same disk as given. 514 * If there are no active disks on list, NULL is returned. 515 */ 516static __inline struct g_mirror_disk * 517g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 518{ 519 struct g_mirror_disk *dp; 520 521 for (dp = LIST_NEXT(disk, d_next); dp != disk; 522 dp = LIST_NEXT(dp, d_next)) { 523 if (dp == NULL) 524 dp = LIST_FIRST(&sc->sc_disks); 525 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 526 break; 527 } 528 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 529 return (NULL); 530 return (dp); 531} 532 533static struct g_mirror_disk * 534g_mirror_get_disk(struct g_mirror_softc *sc) 535{ 536 struct g_mirror_disk *disk; 537 538 if (sc->sc_hint == NULL) { 539 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 540 if (sc->sc_hint == NULL) 541 return (NULL); 542 } 543 disk = sc->sc_hint; 544 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 545 disk = g_mirror_find_next(sc, disk); 546 if (disk == NULL) 547 return (NULL); 548 } 549 sc->sc_hint = g_mirror_find_next(sc, disk); 550 return (disk); 551} 552 553static int 554g_mirror_write_metadata(struct g_mirror_disk *disk, 555 struct g_mirror_metadata *md) 556{ 557 struct g_mirror_softc *sc; 558 struct g_consumer *cp; 559 off_t offset, length; 560 u_char *sector; 561 int close = 0, error = 0; 562 563 g_topology_assert(); 564 565 sc = disk->d_softc; 566 cp = disk->d_consumer; 567 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 568 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 569 length = cp->provider->sectorsize; 570 offset = cp->provider->mediasize - length; 571 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 572 /* 573 * Open consumer if it wasn't opened and remember to close it. 574 */ 575 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 576 error = g_access(cp, 0, 1, 1); 577 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 578 cp->provider->name, 0, 1, 1, error); 579 if (error == 0) 580 close = 1; 581#ifdef INVARIANTS 582 } else { 583 KASSERT(cp->acw > 0 && cp->ace > 0, 584 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 585 cp->acr, cp->acw, cp->ace)); 586#endif 587 } 588 if (error == 0) { 589 if (md != NULL) 590 mirror_metadata_encode(md, sector); 591 g_topology_unlock(); 592 error = g_write_data(cp, offset, sector, length); 593 g_topology_lock(); 594 } 595 free(sector, M_MIRROR); 596 if (close) { 597 g_access(cp, 0, -1, -1); 598 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 599 cp->provider->name, 0, -1, -1, 0); 600 } 601 if (error != 0) { 602 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 603 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 604 G_MIRROR_EVENT_DONTWAIT); 605 } 606 return (error); 607} 608 609static int 610g_mirror_clear_metadata(struct g_mirror_disk *disk) 611{ 612 int error; 613 614 g_topology_assert(); 615 error = g_mirror_write_metadata(disk, NULL); 616 if (error == 0) { 617 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 618 g_mirror_get_diskname(disk)); 619 } else { 620 G_MIRROR_DEBUG(0, 621 "Cannot clear metadata on disk %s (error=%d).", 622 g_mirror_get_diskname(disk), error); 623 } 624 return (error); 625} 626 627void 628g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 629 struct g_mirror_metadata *md) 630{ 631 632 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 633 md->md_version = G_MIRROR_VERSION; 634 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 635 md->md_mid = sc->sc_id; 636 md->md_all = sc->sc_ndisks; 637 md->md_slice = sc->sc_slice; 638 md->md_balance = sc->sc_balance; 639 md->md_mediasize = sc->sc_mediasize; 640 md->md_sectorsize = sc->sc_sectorsize; 641 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 642 bzero(md->md_provider, sizeof(md->md_provider)); 643 if (disk == NULL) { 644 md->md_did = arc4random(); 645 md->md_priority = 0; 646 md->md_syncid = 0; 647 md->md_dflags = 0; 648 md->md_sync_offset = 0; 649 } else { 650 md->md_did = disk->d_id; 651 md->md_priority = disk->d_priority; 652 md->md_syncid = disk->d_sync.ds_syncid; 653 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 654 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 655 md->md_sync_offset = disk->d_sync.ds_offset_done; 656 else 657 md->md_sync_offset = 0; 658 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 659 strlcpy(md->md_provider, 660 disk->d_consumer->provider->name, 661 sizeof(md->md_provider)); 662 } 663 } 664} 665 666void 667g_mirror_update_metadata(struct g_mirror_disk *disk) 668{ 669 struct g_mirror_metadata md; 670 int error; 671 672 g_topology_assert(); 673 g_mirror_fill_metadata(disk->d_softc, disk, &md); 674 error = g_mirror_write_metadata(disk, &md); 675 if (error == 0) { 676 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 677 g_mirror_get_diskname(disk)); 678 } else { 679 G_MIRROR_DEBUG(0, 680 "Cannot update metadata on disk %s (error=%d).", 681 g_mirror_get_diskname(disk), error); 682 } 683} 684 685static void 686g_mirror_bump_syncid(struct g_mirror_softc *sc) 687{ 688 struct g_mirror_disk *disk; 689 690 g_topology_assert(); 691 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 692 ("%s called with no active disks (device=%s).", __func__, 693 sc->sc_name)); 694 695 sc->sc_syncid++; 696 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 697 sc->sc_syncid); 698 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 699 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 700 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 701 disk->d_sync.ds_syncid = sc->sc_syncid; 702 g_mirror_update_metadata(disk); 703 } 704 } 705} 706 707static __inline int 708bintime_cmp(struct bintime *bt1, struct bintime *bt2) 709{ 710 711 if (bt1->sec < bt2->sec) 712 return (-1); 713 else if (bt1->sec > bt2->sec) 714 return (1); 715 if (bt1->frac < bt2->frac) 716 return (-1); 717 else if (bt1->frac > bt2->frac) 718 return (1); 719 return (0); 720} 721 722static void 723g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 724{ 725 726 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 727 return; 728 binuptime(&disk->d_delay); 729 bintime_sub(&disk->d_delay, &bp->bio_t0); 730} 731 732static void 733g_mirror_done(struct bio *bp) 734{ 735 struct g_mirror_softc *sc; 736 737 sc = bp->bio_from->geom->softc; 738 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 739 mtx_lock(&sc->sc_queue_mtx); 740 bioq_disksort(&sc->sc_queue, bp); 741 wakeup(sc); 742 mtx_unlock(&sc->sc_queue_mtx); 743} 744 745static void 746g_mirror_regular_request(struct bio *bp) 747{ 748 struct g_mirror_softc *sc; 749 struct g_mirror_disk *disk; 750 struct bio *pbp; 751 752 g_topology_assert_not(); 753 754 pbp = bp->bio_parent; 755 sc = pbp->bio_to->geom->softc; 756 disk = bp->bio_from->private; 757 if (disk == NULL) { 758 g_topology_lock(); 759 g_mirror_kill_consumer(sc, bp->bio_from); 760 g_topology_unlock(); 761 } else { 762 g_mirror_update_delay(disk, bp); 763 } 764 765 pbp->bio_inbed++; 766 KASSERT(pbp->bio_inbed <= pbp->bio_children, 767 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 768 pbp->bio_children)); 769 if (bp->bio_error == 0 && pbp->bio_error == 0) { 770 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 771 g_destroy_bio(bp); 772 if (pbp->bio_children == pbp->bio_inbed) { 773 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 774 pbp->bio_completed = pbp->bio_length; 775 g_io_deliver(pbp, pbp->bio_error); 776 } 777 return; 778 } else if (bp->bio_error != 0) { 779 if (pbp->bio_error == 0) 780 pbp->bio_error = bp->bio_error; 781 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 782 bp->bio_error); 783 if (disk != NULL) { 784 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 785 g_mirror_event_send(disk, 786 G_MIRROR_DISK_STATE_DISCONNECTED, 787 G_MIRROR_EVENT_DONTWAIT); 788 } 789 switch (pbp->bio_cmd) { 790 case BIO_DELETE: 791 case BIO_WRITE: 792 pbp->bio_inbed--; 793 pbp->bio_children--; 794 break; 795 } 796 } 797 g_destroy_bio(bp); 798 799 switch (pbp->bio_cmd) { 800 case BIO_READ: 801 if (pbp->bio_children == pbp->bio_inbed) { 802 pbp->bio_error = 0; 803 mtx_lock(&sc->sc_queue_mtx); 804 bioq_disksort(&sc->sc_queue, pbp); 805 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 806 wakeup(sc); 807 mtx_unlock(&sc->sc_queue_mtx); 808 } 809 break; 810 case BIO_DELETE: 811 case BIO_WRITE: 812 if (pbp->bio_children == 0) { 813 /* 814 * All requests failed. 815 */ 816 } else if (pbp->bio_inbed < pbp->bio_children) { 817 /* Do nothing. */ 818 break; 819 } else if (pbp->bio_children == pbp->bio_inbed) { 820 /* Some requests succeeded. */ 821 pbp->bio_error = 0; 822 pbp->bio_completed = pbp->bio_length; 823 } 824 g_io_deliver(pbp, pbp->bio_error); 825 break; 826 default: 827 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 828 break; 829 } 830} 831 832static void 833g_mirror_sync_done(struct bio *bp) 834{ 835 struct g_mirror_softc *sc; 836 837 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 838 sc = bp->bio_from->geom->softc; 839 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 840 mtx_lock(&sc->sc_queue_mtx); 841 bioq_disksort(&sc->sc_queue, bp); 842 wakeup(sc); 843 mtx_unlock(&sc->sc_queue_mtx); 844} 845 846static void 847g_mirror_start(struct bio *bp) 848{ 849 struct g_mirror_softc *sc; 850 851 sc = bp->bio_to->geom->softc; 852 /* 853 * If sc == NULL or there are no valid disks, provider's error 854 * should be set and g_mirror_start() should not be called at all. 855 */ 856 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 857 ("Provider's error should be set (error=%d)(mirror=%s).", 858 bp->bio_to->error, bp->bio_to->name)); 859 G_MIRROR_LOGREQ(3, bp, "Request received."); 860 861 switch (bp->bio_cmd) { 862 case BIO_READ: 863 case BIO_WRITE: 864 case BIO_DELETE: 865 break; 866 case BIO_GETATTR: 867 default: 868 g_io_deliver(bp, EOPNOTSUPP); 869 return; 870 } 871 mtx_lock(&sc->sc_queue_mtx); 872 bioq_disksort(&sc->sc_queue, bp); 873 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 874 wakeup(sc); 875 mtx_unlock(&sc->sc_queue_mtx); 876} 877 878/* 879 * Send one synchronization request. 880 */ 881static void 882g_mirror_sync_one(struct g_mirror_disk *disk) 883{ 884 struct g_mirror_softc *sc; 885 struct bio *bp; 886 887 sc = disk->d_softc; 888 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 889 ("Disk %s is not marked for synchronization.", 890 g_mirror_get_diskname(disk))); 891 892 bp = g_new_bio(); 893 if (bp == NULL) 894 return; 895 bp->bio_parent = NULL; 896 bp->bio_cmd = BIO_READ; 897 bp->bio_offset = disk->d_sync.ds_offset; 898 bp->bio_length = MIN(G_MIRROR_SYNC_BLOCK_SIZE, 899 sc->sc_mediasize - bp->bio_offset); 900 bp->bio_cflags = 0; 901 bp->bio_done = g_mirror_sync_done; 902 bp->bio_data = disk->d_sync.ds_data; 903 if (bp->bio_data == NULL) { 904 g_destroy_bio(bp); 905 return; 906 } 907 disk->d_sync.ds_offset += bp->bio_length; 908 bp->bio_to = sc->sc_provider; 909 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 910 g_io_request(bp, disk->d_sync.ds_consumer); 911} 912 913static void 914g_mirror_sync_request(struct bio *bp) 915{ 916 struct g_mirror_softc *sc; 917 struct g_mirror_disk *disk; 918 919 sc = bp->bio_from->geom->softc; 920 disk = bp->bio_from->private; 921 if (disk == NULL) { 922 g_topology_lock(); 923 g_mirror_kill_consumer(sc, bp->bio_from); 924 g_topology_unlock(); 925 g_destroy_bio(bp); 926 return; 927 } 928 929 /* 930 * Synchronization request. 931 */ 932 switch (bp->bio_cmd) { 933 case BIO_READ: 934 { 935 struct g_consumer *cp; 936 937 if (bp->bio_error != 0) { 938 G_MIRROR_LOGREQ(0, bp, 939 "Synchronization request failed (error=%d).", 940 bp->bio_error); 941 g_destroy_bio(bp); 942 return; 943 } 944 bp->bio_cmd = BIO_WRITE; 945 bp->bio_cflags = 0; 946 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 947 cp = disk->d_consumer; 948 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 949 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 950 cp->acr, cp->acw, cp->ace)); 951 g_io_request(bp, cp); 952 return; 953 } 954 case BIO_WRITE: 955 if (bp->bio_error != 0) { 956 G_MIRROR_LOGREQ(0, bp, 957 "Synchronization request failed (error=%d).", 958 bp->bio_error); 959 g_destroy_bio(bp); 960 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 961 g_mirror_event_send(disk, 962 G_MIRROR_DISK_STATE_DISCONNECTED, 963 G_MIRROR_EVENT_DONTWAIT); 964 return; 965 } 966 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 967 disk->d_sync.ds_offset_done = bp->bio_offset + bp->bio_length; 968 g_destroy_bio(bp); 969 if (disk->d_sync.ds_offset_done == sc->sc_provider->mediasize) { 970 /* 971 * Disk up-to-date, activate it. 972 */ 973 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 974 G_MIRROR_EVENT_DONTWAIT); 975 return; 976 } else if ((disk->d_sync.ds_offset_done % 977 (G_MIRROR_SYNC_BLOCK_SIZE * 100)) == 0) { 978 /* 979 * Update offset_done on every 100 blocks. 980 * XXX: This should be configurable. 981 */ 982 g_topology_lock(); 983 g_mirror_update_metadata(disk); 984 g_topology_unlock(); 985 } 986 return; 987 default: 988 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 989 bp->bio_cmd, sc->sc_name)); 990 break; 991 } 992} 993 994static void 995g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 996{ 997 struct g_mirror_disk *disk; 998 struct g_consumer *cp; 999 struct bio *cbp; 1000 1001 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1002 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1003 break; 1004 } 1005 if (disk == NULL) { 1006 if (bp->bio_error == 0) 1007 bp->bio_error = ENXIO; 1008 g_io_deliver(bp, bp->bio_error); 1009 return; 1010 } 1011 cbp = g_clone_bio(bp); 1012 if (cbp == NULL) { 1013 if (bp->bio_error == 0) 1014 bp->bio_error = ENOMEM; 1015 g_io_deliver(bp, bp->bio_error); 1016 return; 1017 } 1018 /* 1019 * Fill in the component buf structure. 1020 */ 1021 cp = disk->d_consumer; 1022 cbp->bio_done = g_mirror_done; 1023 cbp->bio_to = cp->provider; 1024 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1025 KASSERT(cp->acr > 0 && cp->ace > 0, 1026 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1027 cp->acw, cp->ace)); 1028 g_io_request(cbp, cp); 1029} 1030 1031static void 1032g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1033{ 1034 struct g_mirror_disk *disk; 1035 struct g_consumer *cp; 1036 struct bio *cbp; 1037 1038 disk = g_mirror_get_disk(sc); 1039 if (disk == NULL) { 1040 if (bp->bio_error == 0) 1041 bp->bio_error = ENXIO; 1042 g_io_deliver(bp, bp->bio_error); 1043 return; 1044 } 1045 cbp = g_clone_bio(bp); 1046 if (cbp == NULL) { 1047 if (bp->bio_error == 0) 1048 bp->bio_error = ENOMEM; 1049 g_io_deliver(bp, bp->bio_error); 1050 return; 1051 } 1052 /* 1053 * Fill in the component buf structure. 1054 */ 1055 cp = disk->d_consumer; 1056 cbp->bio_done = g_mirror_done; 1057 cbp->bio_to = cp->provider; 1058 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1059 KASSERT(cp->acr > 0 && cp->ace > 0, 1060 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1061 cp->acw, cp->ace)); 1062 g_io_request(cbp, cp); 1063} 1064 1065static void 1066g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1067{ 1068 struct g_mirror_disk *disk, *dp; 1069 struct g_consumer *cp; 1070 struct bio *cbp; 1071 struct bintime curtime; 1072 1073 binuptime(&curtime); 1074 /* 1075 * Find a disk which the smallest load. 1076 */ 1077 disk = NULL; 1078 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1079 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1080 continue; 1081 /* If disk wasn't used for more than 2 sec, use it. */ 1082 if (curtime.sec - dp->d_last_used.sec >= 2) { 1083 disk = dp; 1084 break; 1085 } 1086 if (disk == NULL || 1087 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1088 disk = dp; 1089 } 1090 } 1091 cbp = g_clone_bio(bp); 1092 if (cbp == NULL) { 1093 if (bp->bio_error == 0) 1094 bp->bio_error = ENOMEM; 1095 g_io_deliver(bp, bp->bio_error); 1096 return; 1097 } 1098 /* 1099 * Fill in the component buf structure. 1100 */ 1101 cp = disk->d_consumer; 1102 cbp->bio_done = g_mirror_done; 1103 cbp->bio_to = cp->provider; 1104 binuptime(&disk->d_last_used); 1105 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1106 KASSERT(cp->acr > 0 && cp->ace > 0, 1107 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1108 cp->acw, cp->ace)); 1109 g_io_request(cbp, cp); 1110} 1111 1112static void 1113g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1114{ 1115 struct bio_queue_head queue; 1116 struct g_mirror_disk *disk; 1117 struct g_consumer *cp; 1118 struct bio *cbp; 1119 off_t left, mod, offset, slice; 1120 u_char *data; 1121 u_int ndisks; 1122 1123 if (bp->bio_length <= sc->sc_slice) { 1124 g_mirror_request_round_robin(sc, bp); 1125 return; 1126 } 1127 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1128 slice = bp->bio_length / ndisks; 1129 mod = slice % sc->sc_provider->sectorsize; 1130 if (mod != 0) 1131 slice += sc->sc_provider->sectorsize - mod; 1132 /* 1133 * Allocate all bios before sending any request, so we can 1134 * return ENOMEM in nice and clean way. 1135 */ 1136 left = bp->bio_length; 1137 offset = bp->bio_offset; 1138 data = bp->bio_data; 1139 bioq_init(&queue); 1140 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1141 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1142 continue; 1143 cbp = g_clone_bio(bp); 1144 if (cbp == NULL) { 1145 for (cbp = bioq_first(&queue); cbp != NULL; 1146 cbp = bioq_first(&queue)) { 1147 bioq_remove(&queue, cbp); 1148 g_destroy_bio(cbp); 1149 } 1150 if (bp->bio_error == 0) 1151 bp->bio_error = ENOMEM; 1152 g_io_deliver(bp, bp->bio_error); 1153 return; 1154 } 1155 bioq_insert_tail(&queue, cbp); 1156 cbp->bio_done = g_mirror_done; 1157 cbp->bio_caller1 = disk; 1158 cbp->bio_to = disk->d_consumer->provider; 1159 cbp->bio_offset = offset; 1160 cbp->bio_data = data; 1161 cbp->bio_length = MIN(left, slice); 1162 left -= cbp->bio_length; 1163 if (left == 0) 1164 break; 1165 offset += cbp->bio_length; 1166 data += cbp->bio_length; 1167 } 1168 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1169 bioq_remove(&queue, cbp); 1170 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1171 disk = cbp->bio_caller1; 1172 cbp->bio_caller1 = NULL; 1173 cp = disk->d_consumer; 1174 KASSERT(cp->acr > 0 && cp->ace > 0, 1175 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1176 cp->acr, cp->acw, cp->ace)); 1177 g_io_request(cbp, disk->d_consumer); 1178 } 1179} 1180 1181static void 1182g_mirror_register_request(struct bio *bp) 1183{ 1184 struct g_mirror_softc *sc; 1185 1186 sc = bp->bio_to->geom->softc; 1187 switch (bp->bio_cmd) { 1188 case BIO_READ: 1189 switch (sc->sc_balance) { 1190 case G_MIRROR_BALANCE_LOAD: 1191 g_mirror_request_load(sc, bp); 1192 break; 1193 case G_MIRROR_BALANCE_PREFER: 1194 g_mirror_request_prefer(sc, bp); 1195 break; 1196 case G_MIRROR_BALANCE_ROUND_ROBIN: 1197 g_mirror_request_round_robin(sc, bp); 1198 break; 1199 case G_MIRROR_BALANCE_SPLIT: 1200 g_mirror_request_split(sc, bp); 1201 break; 1202 } 1203 return; 1204 case BIO_WRITE: 1205 case BIO_DELETE: 1206 { 1207 struct g_mirror_disk *disk; 1208 struct bio_queue_head queue; 1209 struct g_consumer *cp; 1210 struct bio *cbp; 1211 1212 /* 1213 * Allocate all bios before sending any request, so we can 1214 * return ENOMEM in nice and clean way. 1215 */ 1216 bioq_init(&queue); 1217 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1218 switch (disk->d_state) { 1219 case G_MIRROR_DISK_STATE_ACTIVE: 1220 break; 1221 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1222 if (bp->bio_offset >= disk->d_sync.ds_offset) 1223 continue; 1224 break; 1225 default: 1226 continue; 1227 } 1228 cbp = g_clone_bio(bp); 1229 if (cbp == NULL) { 1230 for (cbp = bioq_first(&queue); cbp != NULL; 1231 cbp = bioq_first(&queue)) { 1232 bioq_remove(&queue, cbp); 1233 g_destroy_bio(cbp); 1234 } 1235 if (bp->bio_error == 0) 1236 bp->bio_error = ENOMEM; 1237 g_io_deliver(bp, bp->bio_error); 1238 return; 1239 } 1240 bioq_insert_tail(&queue, cbp); 1241 } 1242 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1243 switch (disk->d_state) { 1244 case G_MIRROR_DISK_STATE_ACTIVE: 1245 break; 1246 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1247 if (bp->bio_offset >= disk->d_sync.ds_offset) 1248 continue; 1249 break; 1250 default: 1251 continue; 1252 } 1253 cbp = bioq_first(&queue); 1254 KASSERT(cbp != NULL, ("NULL cbp! (device %s).", 1255 sc->sc_name)); 1256 bioq_remove(&queue, cbp); 1257 cp = disk->d_consumer; 1258 cbp->bio_done = g_mirror_done; 1259 cbp->bio_to = cp->provider; 1260 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1261 KASSERT(cp->acw > 0 && cp->ace > 0, 1262 ("Consumer %s not opened (r%dw%de%d).", 1263 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1264 g_io_request(cbp, cp); 1265 } 1266 /* 1267 * Bump syncid on first write. 1268 */ 1269 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) { 1270 sc->sc_bump_syncid = 0; 1271 g_topology_lock(); 1272 g_mirror_bump_syncid(sc); 1273 g_topology_unlock(); 1274 } 1275 return; 1276 } 1277 default: 1278 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1279 bp->bio_cmd, sc->sc_name)); 1280 break; 1281 } 1282} 1283 1284static int 1285g_mirror_can_destroy(struct g_mirror_softc *sc) 1286{ 1287 struct g_geom *gp; 1288 struct g_consumer *cp; 1289 1290 g_topology_assert(); 1291 gp = sc->sc_geom; 1292 LIST_FOREACH(cp, &gp->consumer, consumer) { 1293 if (g_mirror_is_busy(sc, cp)) 1294 return (0); 1295 } 1296 gp = sc->sc_sync.ds_geom; 1297 LIST_FOREACH(cp, &gp->consumer, consumer) { 1298 if (g_mirror_is_busy(sc, cp)) 1299 return (0); 1300 } 1301 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1302 sc->sc_name); 1303 return (1); 1304} 1305 1306static int 1307g_mirror_try_destroy(struct g_mirror_softc *sc) 1308{ 1309 1310 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1311 g_topology_lock(); 1312 if (!g_mirror_can_destroy(sc)) { 1313 g_topology_unlock(); 1314 return (0); 1315 } 1316 g_topology_unlock(); 1317 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1318 &sc->sc_worker); 1319 wakeup(&sc->sc_worker); 1320 sc->sc_worker = NULL; 1321 } else { 1322 g_topology_lock(); 1323 if (!g_mirror_can_destroy(sc)) { 1324 g_topology_unlock(); 1325 return (0); 1326 } 1327 g_mirror_destroy_device(sc); 1328 g_topology_unlock(); 1329 free(sc, M_MIRROR); 1330 } 1331 return (1); 1332} 1333 1334/* 1335 * Worker thread. 1336 */ 1337static void 1338g_mirror_worker(void *arg) 1339{ 1340 struct g_mirror_softc *sc; 1341 struct g_mirror_disk *disk; 1342 struct g_mirror_event *ep; 1343 struct bio *bp; 1344 u_int nreqs; 1345 1346 sc = arg; 1347 curthread->td_base_pri = PRIBIO; 1348 1349 nreqs = 0; 1350 for (;;) { 1351 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1352 /* 1353 * First take a look at events. 1354 * This is important to handle events before any I/O requests. 1355 */ 1356 ep = g_mirror_event_get(sc); 1357 if (ep != NULL) { 1358 g_topology_lock(); 1359 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1360 /* Update only device status. */ 1361 G_MIRROR_DEBUG(3, 1362 "Running event for device %s.", 1363 sc->sc_name); 1364 ep->e_error = 0; 1365 g_mirror_update_device(sc, 1); 1366 } else { 1367 /* Update disk status. */ 1368 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1369 g_mirror_get_diskname(ep->e_disk)); 1370 ep->e_error = g_mirror_update_disk(ep->e_disk, 1371 ep->e_state); 1372 if (ep->e_error == 0) 1373 g_mirror_update_device(sc, 0); 1374 } 1375 g_topology_unlock(); 1376 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1377 KASSERT(ep->e_error == 0, 1378 ("Error cannot be handled.")); 1379 g_mirror_event_free(ep); 1380 } else { 1381 ep->e_flags |= G_MIRROR_EVENT_DONE; 1382 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1383 ep); 1384 mtx_lock(&sc->sc_events_mtx); 1385 wakeup(ep); 1386 mtx_unlock(&sc->sc_events_mtx); 1387 } 1388 if ((sc->sc_flags & 1389 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1390 if (g_mirror_try_destroy(sc)) 1391 kthread_exit(0); 1392 } 1393 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1394 continue; 1395 } 1396 /* 1397 * Now I/O requests. 1398 */ 1399 /* Get first request from the queue. */ 1400 mtx_lock(&sc->sc_queue_mtx); 1401 bp = bioq_first(&sc->sc_queue); 1402 if (bp == NULL) { 1403 if ((sc->sc_flags & 1404 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1405 mtx_unlock(&sc->sc_queue_mtx); 1406 if (g_mirror_try_destroy(sc)) 1407 kthread_exit(0); 1408 mtx_lock(&sc->sc_queue_mtx); 1409 } 1410 } 1411 if (sc->sc_sync.ds_ndisks > 0 && 1412 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1413 mtx_unlock(&sc->sc_queue_mtx); 1414 /* 1415 * It is time for synchronization... 1416 */ 1417 nreqs = 0; 1418 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1419 if (disk->d_state != 1420 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1421 continue; 1422 } 1423 if (disk->d_sync.ds_offset >= 1424 sc->sc_provider->mediasize) { 1425 continue; 1426 } 1427 if (disk->d_sync.ds_offset > 1428 disk->d_sync.ds_offset_done) { 1429 continue; 1430 } 1431 g_mirror_sync_one(disk); 1432 } 1433 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1434 goto sleep; 1435 } 1436 if (bp == NULL) { 1437 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 0); 1438 G_MIRROR_DEBUG(5, "%s: I'm here 3.", __func__); 1439 continue; 1440 } 1441 nreqs++; 1442 bioq_remove(&sc->sc_queue, bp); 1443 mtx_unlock(&sc->sc_queue_mtx); 1444 1445 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1446 g_mirror_regular_request(bp); 1447 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1448 u_int timeout, sps; 1449 1450 g_mirror_sync_request(bp); 1451sleep: 1452 sps = atomic_load_acq_int(&g_mirror_syncs_per_sec); 1453 if (sps == 0) { 1454 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1455 continue; 1456 } 1457 mtx_lock(&sc->sc_queue_mtx); 1458 if (bioq_first(&sc->sc_queue) != NULL) { 1459 mtx_unlock(&sc->sc_queue_mtx); 1460 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1461 continue; 1462 } 1463 timeout = hz / sps; 1464 if (timeout == 0) 1465 timeout = 1; 1466 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w2", 1467 timeout); 1468 } else { 1469 g_mirror_register_request(bp); 1470 } 1471 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1472 } 1473} 1474 1475/* 1476 * Open disk's consumer if needed. 1477 */ 1478static void 1479g_mirror_update_access(struct g_mirror_disk *disk) 1480{ 1481 struct g_provider *pp; 1482 struct g_consumer *cp; 1483 int acr, acw, ace, cpw, error; 1484 1485 g_topology_assert(); 1486 1487 cp = disk->d_consumer; 1488 pp = disk->d_softc->sc_provider; 1489 if (pp == NULL) { 1490 acr = -cp->acr; 1491 acw = -cp->acw; 1492 ace = -cp->ace; 1493 } else { 1494 acr = pp->acr - cp->acr; 1495 acw = pp->acw - cp->acw; 1496 ace = pp->ace - cp->ace; 1497 /* Grab an extra "exclusive" bit. */ 1498 if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) 1499 ace++; 1500 } 1501 if (acr == 0 && acw == 0 && ace == 0) 1502 return; 1503 cpw = cp->acw; 1504 error = g_access(cp, acr, acw, ace); 1505 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, acr, 1506 acw, ace, error); 1507 if (error != 0) { 1508 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1509 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1510 G_MIRROR_EVENT_DONTWAIT); 1511 return; 1512 } 1513 if (cpw == 0 && cp->acw > 0) { 1514 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1515 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1516 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1517 } else if (cpw > 0 && cp->acw == 0) { 1518 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1519 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1520 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1521 } 1522} 1523 1524static void 1525g_mirror_sync_start(struct g_mirror_disk *disk) 1526{ 1527 struct g_mirror_softc *sc; 1528 struct g_consumer *cp; 1529 int error; 1530 1531 g_topology_assert(); 1532 1533 sc = disk->d_softc; 1534 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1535 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1536 sc->sc_state)); 1537 cp = disk->d_consumer; 1538 KASSERT(cp->acr == 0 && cp->acw == 0 && cp->ace == 0, 1539 ("Consumer %s already opened.", cp->provider->name)); 1540 1541 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1542 g_mirror_get_diskname(disk)); 1543 error = g_access(cp, 0, 1, 1); 1544 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, 1, 1545 1, error); 1546 if (error != 0) { 1547 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1548 G_MIRROR_EVENT_DONTWAIT); 1549 return; 1550 } 1551 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1552 KASSERT(disk->d_sync.ds_consumer == NULL, 1553 ("Sync consumer already exists (device=%s, disk=%s).", 1554 sc->sc_name, g_mirror_get_diskname(disk))); 1555 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1556 disk->d_sync.ds_consumer->private = disk; 1557 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1558 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1559 disk->d_softc->sc_name, error)); 1560 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1561 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1562 disk->d_softc->sc_name, error)); 1563 disk->d_sync.ds_data = malloc(G_MIRROR_SYNC_BLOCK_SIZE, M_MIRROR, 1564 M_WAITOK); 1565 sc->sc_sync.ds_ndisks++; 1566} 1567 1568/* 1569 * Stop synchronization process. 1570 * type: 0 - synchronization finished 1571 * 1 - synchronization stopped 1572 */ 1573static void 1574g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1575{ 1576 struct g_consumer *cp; 1577 1578 g_topology_assert(); 1579 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1580 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1581 g_mirror_disk_state2str(disk->d_state))); 1582 if (disk->d_sync.ds_consumer == NULL) 1583 return; 1584 1585 if (type == 0) { 1586 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1587 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1588 } else /* if (type == 1) */ { 1589 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1590 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1591 } 1592 cp = disk->d_sync.ds_consumer; 1593 g_access(cp, -1, 0, 0); 1594 g_mirror_kill_consumer(disk->d_softc, cp); 1595 free(disk->d_sync.ds_data, M_MIRROR); 1596 disk->d_sync.ds_consumer = NULL; 1597 disk->d_softc->sc_sync.ds_ndisks--; 1598 cp = disk->d_consumer; 1599 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 1600 ("Consumer %s not opened.", cp->provider->name)); 1601 g_access(cp, 0, -1, -1); 1602 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, -1, 1603 -1, 0); 1604 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1605} 1606 1607static void 1608g_mirror_launch_provider(struct g_mirror_softc *sc) 1609{ 1610 struct g_mirror_disk *disk; 1611 struct g_provider *pp; 1612 1613 g_topology_assert(); 1614 1615 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1616 pp->mediasize = sc->sc_mediasize; 1617 pp->sectorsize = sc->sc_sectorsize; 1618 sc->sc_provider = pp; 1619 g_error_provider(pp, 0); 1620 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1621 pp->name); 1622 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1623 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1624 g_mirror_sync_start(disk); 1625 } 1626} 1627 1628static void 1629g_mirror_destroy_provider(struct g_mirror_softc *sc) 1630{ 1631 struct g_mirror_disk *disk; 1632 struct bio *bp; 1633 1634 g_topology_assert(); 1635 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1636 sc->sc_name)); 1637 1638 g_error_provider(sc->sc_provider, ENXIO); 1639 mtx_lock(&sc->sc_queue_mtx); 1640 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1641 bioq_remove(&sc->sc_queue, bp); 1642 g_io_deliver(bp, ENXIO); 1643 } 1644 mtx_unlock(&sc->sc_queue_mtx); 1645 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1646 sc->sc_provider->name); 1647 sc->sc_provider->flags |= G_PF_WITHER; 1648 g_orphan_provider(sc->sc_provider, ENXIO); 1649 sc->sc_provider = NULL; 1650 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1651 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1652 g_mirror_sync_stop(disk, 1); 1653 } 1654} 1655 1656static void 1657g_mirror_go(void *arg) 1658{ 1659 struct g_mirror_softc *sc; 1660 1661 sc = arg; 1662 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1663 g_mirror_event_send(sc, 0, 1664 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1665} 1666 1667static u_int 1668g_mirror_determine_state(struct g_mirror_disk *disk) 1669{ 1670 struct g_mirror_softc *sc; 1671 u_int state; 1672 1673 sc = disk->d_softc; 1674 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1675 if ((disk->d_flags & 1676 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1677 /* Disk does not need synchronization. */ 1678 state = G_MIRROR_DISK_STATE_ACTIVE; 1679 } else { 1680 if ((sc->sc_flags & 1681 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1682 (disk->d_flags & 1683 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1684 /* 1685 * We can start synchronization from 1686 * the stored offset. 1687 */ 1688 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1689 } else { 1690 state = G_MIRROR_DISK_STATE_STALE; 1691 } 1692 } 1693 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1694 /* 1695 * Reset all synchronization data for this disk, 1696 * because if it even was synchronized, it was 1697 * synchronized to disks with different syncid. 1698 */ 1699 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1700 disk->d_sync.ds_offset = 0; 1701 disk->d_sync.ds_offset_done = 0; 1702 disk->d_sync.ds_syncid = sc->sc_syncid; 1703 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1704 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1705 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1706 } else { 1707 state = G_MIRROR_DISK_STATE_STALE; 1708 } 1709 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1710 /* 1711 * Not good, NOT GOOD! 1712 * It means that mirror was started on stale disks 1713 * and more fresh disk just arrive. 1714 * If there were writes, mirror is fucked up, sorry. 1715 * I think the best choice here is don't touch 1716 * this disk and inform the user laudly. 1717 */ 1718 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1719 "disk (%s) arrives!! It will not be connected to the " 1720 "running device.", sc->sc_name, 1721 g_mirror_get_diskname(disk)); 1722 g_mirror_destroy_disk(disk); 1723 state = G_MIRROR_DISK_STATE_NONE; 1724 /* Return immediately, because disk was destroyed. */ 1725 return (state); 1726 } 1727 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1728 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1729 return (state); 1730} 1731 1732/* 1733 * Update device state. 1734 */ 1735static void 1736g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1737{ 1738 struct g_mirror_disk *disk; 1739 u_int state; 1740 1741 g_topology_assert(); 1742 1743 switch (sc->sc_state) { 1744 case G_MIRROR_DEVICE_STATE_STARTING: 1745 { 1746 struct g_mirror_disk *pdisk; 1747 u_int dirty, ndisks, syncid; 1748 1749 KASSERT(sc->sc_provider == NULL, 1750 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1751 /* 1752 * Are we ready? We are, if all disks are connected or 1753 * if we have any disks and 'force' is true. 1754 */ 1755 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1756 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1757 ; 1758 } else if (g_mirror_ndisks(sc, -1) == 0) { 1759 /* 1760 * Disks went down in starting phase, so destroy 1761 * device. 1762 */ 1763 callout_drain(&sc->sc_callout); 1764 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1765 return; 1766 } else { 1767 return; 1768 } 1769 1770 /* 1771 * Activate all disks with the biggest syncid. 1772 */ 1773 if (force) { 1774 /* 1775 * If 'force' is true, we have been called due to 1776 * timeout, so don't bother canceling timeout. 1777 */ 1778 ndisks = 0; 1779 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1780 if ((disk->d_flags & 1781 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1782 ndisks++; 1783 } 1784 } 1785 if (ndisks == 0) { 1786 /* No valid disks found, destroy device. */ 1787 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1788 return; 1789 } 1790 } else { 1791 /* Cancel timeout. */ 1792 callout_drain(&sc->sc_callout); 1793 } 1794 1795 /* 1796 * Find disk with the biggest syncid. 1797 */ 1798 syncid = 0; 1799 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1800 if (disk->d_sync.ds_syncid > syncid) 1801 syncid = disk->d_sync.ds_syncid; 1802 } 1803 1804 /* 1805 * Here we need to look for dirty disks and if all disks 1806 * with the biggest syncid are dirty, we have to choose 1807 * one with the biggest priority and rebuild the rest. 1808 */ 1809 /* 1810 * Find the number of dirty disks with the biggest syncid. 1811 * Find the number of disks with the biggest syncid. 1812 * While here, find a disk with the biggest priority. 1813 */ 1814 dirty = ndisks = 0; 1815 pdisk = NULL; 1816 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1817 if (disk->d_sync.ds_syncid != syncid) 1818 continue; 1819 if ((disk->d_flags & 1820 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1821 continue; 1822 } 1823 ndisks++; 1824 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1825 dirty++; 1826 if (pdisk == NULL || 1827 pdisk->d_priority < disk->d_priority) { 1828 pdisk = disk; 1829 } 1830 } 1831 } 1832 if (dirty == 0) { 1833 /* No dirty disks at all, great. */ 1834 } else if (dirty == ndisks) { 1835 /* 1836 * Force synchronization for all dirty disks except one 1837 * with the biggest priority. 1838 */ 1839 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1840 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1841 "master disk for synchronization.", 1842 g_mirror_get_diskname(pdisk), sc->sc_name); 1843 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1844 if (disk->d_sync.ds_syncid != syncid) 1845 continue; 1846 if ((disk->d_flags & 1847 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1848 continue; 1849 } 1850 KASSERT((disk->d_flags & 1851 G_MIRROR_DISK_FLAG_DIRTY) != 0, 1852 ("Disk %s isn't marked as dirty.", 1853 g_mirror_get_diskname(disk))); 1854 /* Skip the disk with the biggest priority. */ 1855 if (disk == pdisk) 1856 continue; 1857 disk->d_sync.ds_syncid = 0; 1858 } 1859 } else if (dirty < ndisks) { 1860 /* 1861 * Force synchronization for all dirty disks. 1862 * We have some non-dirty disks. 1863 */ 1864 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1865 if (disk->d_sync.ds_syncid != syncid) 1866 continue; 1867 if ((disk->d_flags & 1868 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1869 continue; 1870 } 1871 if ((disk->d_flags & 1872 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1873 continue; 1874 } 1875 disk->d_sync.ds_syncid = 0; 1876 } 1877 } 1878 1879 /* Reset hint. */ 1880 sc->sc_hint = NULL; 1881 sc->sc_syncid = syncid; 1882 if (force) { 1883 /* Remember to bump syncid on first write. */ 1884 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1885 } 1886 state = G_MIRROR_DEVICE_STATE_RUNNING; 1887 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 1888 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 1889 g_mirror_device_state2str(state)); 1890 sc->sc_state = state; 1891 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1892 state = g_mirror_determine_state(disk); 1893 g_mirror_event_send(disk, state, 1894 G_MIRROR_EVENT_DONTWAIT); 1895 if (state == G_MIRROR_DISK_STATE_STALE) { 1896 sc->sc_bump_syncid = 1897 G_MIRROR_BUMP_ON_FIRST_WRITE; 1898 } 1899 } 1900 break; 1901 } 1902 case G_MIRROR_DEVICE_STATE_RUNNING: 1903 if (g_mirror_ndisks(sc, -1) == 0) { 1904 /* 1905 * No disks at all, we need to destroy device. 1906 */ 1907 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1908 break; 1909 } else if (g_mirror_ndisks(sc, 1910 G_MIRROR_DISK_STATE_ACTIVE) == 0 && 1911 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 1912 /* 1913 * No active disks, destroy provider. 1914 */ 1915 if (sc->sc_provider != NULL) 1916 g_mirror_destroy_provider(sc); 1917 break; 1918 } else if (g_mirror_ndisks(sc, 1919 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 1920 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 1921 /* 1922 * We have active disks, launch provider if it doesn't 1923 * exist. 1924 */ 1925 if (sc->sc_provider == NULL) 1926 g_mirror_launch_provider(sc); 1927 } 1928 /* 1929 * Bump syncid here, if we need to do it immediately. 1930 */ 1931 if (sc->sc_bump_syncid == G_MIRROR_BUMP_IMMEDIATELY) { 1932 sc->sc_bump_syncid = 0; 1933 g_mirror_bump_syncid(sc); 1934 } 1935 break; 1936 default: 1937 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 1938 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 1939 break; 1940 } 1941} 1942 1943/* 1944 * Update disk state and device state if needed. 1945 */ 1946#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 1947 "Disk %s state changed from %s to %s (device %s).", \ 1948 g_mirror_get_diskname(disk), \ 1949 g_mirror_disk_state2str(disk->d_state), \ 1950 g_mirror_disk_state2str(state), sc->sc_name) 1951static int 1952g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 1953{ 1954 struct g_mirror_softc *sc; 1955 1956 g_topology_assert(); 1957 1958 sc = disk->d_softc; 1959again: 1960 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 1961 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 1962 g_mirror_disk_state2str(state)); 1963 switch (state) { 1964 case G_MIRROR_DISK_STATE_NEW: 1965 /* 1966 * Possible scenarios: 1967 * 1. New disk arrive. 1968 */ 1969 /* Previous state should be NONE. */ 1970 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 1971 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1972 g_mirror_disk_state2str(disk->d_state))); 1973 DISK_STATE_CHANGED(); 1974 1975 disk->d_state = state; 1976 if (LIST_EMPTY(&sc->sc_disks)) 1977 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 1978 else { 1979 struct g_mirror_disk *dp; 1980 1981 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1982 if (disk->d_priority >= dp->d_priority) { 1983 LIST_INSERT_BEFORE(dp, disk, d_next); 1984 dp = NULL; 1985 break; 1986 } 1987 if (LIST_NEXT(dp, d_next) == NULL) 1988 break; 1989 } 1990 if (dp != NULL) 1991 LIST_INSERT_AFTER(dp, disk, d_next); 1992 } 1993 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 1994 sc->sc_name, g_mirror_get_diskname(disk)); 1995 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 1996 break; 1997 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1998 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 1999 g_mirror_device_state2str(sc->sc_state), 2000 g_mirror_get_diskname(disk), 2001 g_mirror_disk_state2str(disk->d_state))); 2002 state = g_mirror_determine_state(disk); 2003 if (state != G_MIRROR_DISK_STATE_NONE) 2004 goto again; 2005 break; 2006 case G_MIRROR_DISK_STATE_ACTIVE: 2007 /* 2008 * Possible scenarios: 2009 * 1. New disk does not need synchronization. 2010 * 2. Synchronization process finished successfully. 2011 */ 2012 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2013 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2014 g_mirror_device_state2str(sc->sc_state), 2015 g_mirror_get_diskname(disk), 2016 g_mirror_disk_state2str(disk->d_state))); 2017 /* Previous state should be NEW or SYNCHRONIZING. */ 2018 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2019 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2020 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2021 g_mirror_disk_state2str(disk->d_state))); 2022 DISK_STATE_CHANGED(); 2023 2024 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2025 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2026 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2027 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2028 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2029 g_mirror_sync_stop(disk, 0); 2030 } 2031 disk->d_state = state; 2032 disk->d_sync.ds_offset = 0; 2033 disk->d_sync.ds_offset_done = 0; 2034 g_mirror_update_access(disk); 2035 g_mirror_update_metadata(disk); 2036 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2037 sc->sc_name, g_mirror_get_diskname(disk)); 2038 break; 2039 case G_MIRROR_DISK_STATE_STALE: 2040 /* 2041 * Possible scenarios: 2042 * 1. Stale disk was connected. 2043 */ 2044 /* Previous state should be NEW. */ 2045 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2046 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2047 g_mirror_disk_state2str(disk->d_state))); 2048 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2049 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2050 g_mirror_device_state2str(sc->sc_state), 2051 g_mirror_get_diskname(disk), 2052 g_mirror_disk_state2str(disk->d_state))); 2053 /* 2054 * STALE state is only possible if device is marked 2055 * NOAUTOSYNC. 2056 */ 2057 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2058 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2059 g_mirror_device_state2str(sc->sc_state), 2060 g_mirror_get_diskname(disk), 2061 g_mirror_disk_state2str(disk->d_state))); 2062 DISK_STATE_CHANGED(); 2063 2064 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2065 disk->d_state = state; 2066 g_mirror_update_metadata(disk); 2067 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2068 sc->sc_name, g_mirror_get_diskname(disk)); 2069 break; 2070 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2071 /* 2072 * Possible scenarios: 2073 * 1. Disk which needs synchronization was connected. 2074 */ 2075 /* Previous state should be NEW. */ 2076 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2077 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2078 g_mirror_disk_state2str(disk->d_state))); 2079 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2080 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2081 g_mirror_device_state2str(sc->sc_state), 2082 g_mirror_get_diskname(disk), 2083 g_mirror_disk_state2str(disk->d_state))); 2084 DISK_STATE_CHANGED(); 2085 2086 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2087 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2088 disk->d_state = state; 2089 if (sc->sc_provider != NULL) { 2090 g_mirror_sync_start(disk); 2091 g_mirror_update_metadata(disk); 2092 } 2093 break; 2094 case G_MIRROR_DISK_STATE_DISCONNECTED: 2095 /* 2096 * Possible scenarios: 2097 * 1. Device wasn't running yet, but disk disappear. 2098 * 2. Disk was active and disapppear. 2099 * 3. Disk disappear during synchronization process. 2100 */ 2101 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2102 /* 2103 * Previous state should be ACTIVE, STALE or 2104 * SYNCHRONIZING. 2105 */ 2106 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2107 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2108 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2109 ("Wrong disk state (%s, %s).", 2110 g_mirror_get_diskname(disk), 2111 g_mirror_disk_state2str(disk->d_state))); 2112 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2113 /* Previous state should be NEW. */ 2114 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2115 ("Wrong disk state (%s, %s).", 2116 g_mirror_get_diskname(disk), 2117 g_mirror_disk_state2str(disk->d_state))); 2118 /* 2119 * Reset bumping syncid if disk disappeared in STARTING 2120 * state. 2121 */ 2122 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) 2123 sc->sc_bump_syncid = 0; 2124#ifdef INVARIANTS 2125 } else { 2126 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2127 sc->sc_name, 2128 g_mirror_device_state2str(sc->sc_state), 2129 g_mirror_get_diskname(disk), 2130 g_mirror_disk_state2str(disk->d_state))); 2131#endif 2132 } 2133 DISK_STATE_CHANGED(); 2134 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2135 sc->sc_name, g_mirror_get_diskname(disk)); 2136 2137 g_mirror_destroy_disk(disk); 2138 break; 2139 case G_MIRROR_DISK_STATE_DESTROY: 2140 { 2141 int error; 2142 2143 error = g_mirror_clear_metadata(disk); 2144 if (error != 0) 2145 return (error); 2146 DISK_STATE_CHANGED(); 2147 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2148 sc->sc_name, g_mirror_get_diskname(disk)); 2149 2150 g_mirror_destroy_disk(disk); 2151 sc->sc_ndisks--; 2152 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2153 g_mirror_update_metadata(disk); 2154 } 2155 break; 2156 } 2157 default: 2158 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2159 break; 2160 } 2161 return (0); 2162} 2163#undef DISK_STATE_CHANGED 2164 2165static int 2166g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2167{ 2168 struct g_provider *pp; 2169 u_char *buf; 2170 int error; 2171 2172 g_topology_assert(); 2173 2174 error = g_access(cp, 1, 0, 0); 2175 if (error != 0) 2176 return (error); 2177 pp = cp->provider; 2178 g_topology_unlock(); 2179 /* Metadata are stored on last sector. */ 2180 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2181 &error); 2182 g_topology_lock(); 2183 if (buf == NULL) { 2184 g_access(cp, -1, 0, 0); 2185 return (error); 2186 } 2187 if (error != 0) { 2188 g_access(cp, -1, 0, 0); 2189 g_free(buf); 2190 return (error); 2191 } 2192 error = g_access(cp, -1, 0, 0); 2193 KASSERT(error == 0, ("Cannot decrease access count for %s.", pp->name)); 2194 2195 /* Decode metadata. */ 2196 error = mirror_metadata_decode(buf, md); 2197 g_free(buf); 2198 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2199 return (EINVAL); 2200 if (error != 0) { 2201 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2202 cp->provider->name); 2203 return (error); 2204 } 2205 2206 return (0); 2207} 2208 2209static int 2210g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2211 struct g_mirror_metadata *md) 2212{ 2213 2214 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2215 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2216 pp->name, md->md_did); 2217 return (EEXIST); 2218 } 2219 if (md->md_all != sc->sc_ndisks) { 2220 G_MIRROR_DEBUG(1, 2221 "Invalid '%s' field on disk %s (device %s), skipping.", 2222 "md_all", pp->name, sc->sc_name); 2223 return (EINVAL); 2224 } 2225 if (md->md_slice != sc->sc_slice) { 2226 G_MIRROR_DEBUG(1, 2227 "Invalid '%s' field on disk %s (device %s), skipping.", 2228 "md_slice", pp->name, sc->sc_name); 2229 return (EINVAL); 2230 } 2231 if (md->md_balance != sc->sc_balance) { 2232 G_MIRROR_DEBUG(1, 2233 "Invalid '%s' field on disk %s (device %s), skipping.", 2234 "md_balance", pp->name, sc->sc_name); 2235 return (EINVAL); 2236 } 2237 if (md->md_mediasize != sc->sc_mediasize) { 2238 G_MIRROR_DEBUG(1, 2239 "Invalid '%s' field on disk %s (device %s), skipping.", 2240 "md_mediasize", pp->name, sc->sc_name); 2241 return (EINVAL); 2242 } 2243 if (sc->sc_mediasize > pp->mediasize) { 2244 G_MIRROR_DEBUG(1, 2245 "Invalid size of disk %s (device %s), skipping.", pp->name, 2246 sc->sc_name); 2247 return (EINVAL); 2248 } 2249 if (md->md_sectorsize != sc->sc_sectorsize) { 2250 G_MIRROR_DEBUG(1, 2251 "Invalid '%s' field on disk %s (device %s), skipping.", 2252 "md_sectorsize", pp->name, sc->sc_name); 2253 return (EINVAL); 2254 } 2255 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2256 G_MIRROR_DEBUG(1, 2257 "Invalid sector size of disk %s (device %s), skipping.", 2258 pp->name, sc->sc_name); 2259 return (EINVAL); 2260 } 2261 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2262 G_MIRROR_DEBUG(1, 2263 "Invalid device flags on disk %s (device %s), skipping.", 2264 pp->name, sc->sc_name); 2265 return (EINVAL); 2266 } 2267 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2268 G_MIRROR_DEBUG(1, 2269 "Invalid disk flags on disk %s (device %s), skipping.", 2270 pp->name, sc->sc_name); 2271 return (EINVAL); 2272 } 2273 return (0); 2274} 2275 2276static int 2277g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2278 struct g_mirror_metadata *md) 2279{ 2280 struct g_mirror_disk *disk; 2281 int error; 2282 2283 g_topology_assert(); 2284 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2285 2286 error = g_mirror_check_metadata(sc, pp, md); 2287 if (error != 0) 2288 return (error); 2289 disk = g_mirror_init_disk(sc, pp, md, &error); 2290 if (disk == NULL) 2291 return (error); 2292 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2293 G_MIRROR_EVENT_WAIT); 2294 return (error); 2295} 2296 2297static int 2298g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2299{ 2300 struct g_mirror_softc *sc; 2301 struct g_mirror_disk *disk; 2302 int dcr, dcw, dce, err, error; 2303 2304 g_topology_assert(); 2305 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2306 acw, ace); 2307 2308 dcr = pp->acr + acr; 2309 dcw = pp->acw + acw; 2310 dce = pp->ace + ace; 2311 2312 /* On first open, grab an extra "exclusive" bit */ 2313 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0) 2314 ace++; 2315 /* ... and let go of it on last close */ 2316 if (dcr == 0 && dcw == 0 && dce == 0) 2317 ace--; 2318 2319 sc = pp->geom->softc; 2320 if (sc == NULL || LIST_EMPTY(&sc->sc_disks)) { 2321 if (acr <= 0 && acw <= 0 && ace <= 0) 2322 return (0); 2323 else 2324 return (ENXIO); 2325 } 2326 error = ENXIO; 2327 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2328 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2329 continue; 2330 err = g_access(disk->d_consumer, acr, acw, ace); 2331 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 2332 g_mirror_get_diskname(disk), acr, acw, ace, err); 2333 if (err == 0) { 2334 /* 2335 * Mark disk as dirty on open and unmark on close. 2336 */ 2337 if (pp->acw == 0 && dcw > 0) { 2338 G_MIRROR_DEBUG(1, 2339 "Disk %s (device %s) marked as dirty.", 2340 g_mirror_get_diskname(disk), sc->sc_name); 2341 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2342 g_mirror_update_metadata(disk); 2343 } else if (pp->acw > 0 && dcw == 0) { 2344 G_MIRROR_DEBUG(1, 2345 "Disk %s (device %s) marked as clean.", 2346 g_mirror_get_diskname(disk), sc->sc_name); 2347 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2348 g_mirror_update_metadata(disk); 2349 } 2350 error = 0; 2351 } else { 2352 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 2353 g_mirror_event_send(disk, 2354 G_MIRROR_DISK_STATE_DISCONNECTED, 2355 G_MIRROR_EVENT_DONTWAIT); 2356 } 2357 } 2358 return (error); 2359} 2360 2361static struct g_geom * 2362g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2363{ 2364 struct g_mirror_softc *sc; 2365 struct g_geom *gp; 2366 int error, timeout; 2367 2368 g_topology_assert(); 2369 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2370 md->md_mid); 2371 2372 /* One disk is minimum. */ 2373 if (md->md_all < 1) 2374 return (NULL); 2375 /* 2376 * Action geom. 2377 */ 2378 gp = g_new_geomf(mp, "%s", md->md_name); 2379 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2380 gp->start = g_mirror_start; 2381 gp->spoiled = g_mirror_spoiled; 2382 gp->orphan = g_mirror_orphan; 2383 gp->access = g_mirror_access; 2384 gp->dumpconf = g_mirror_dumpconf; 2385 2386 sc->sc_id = md->md_mid; 2387 sc->sc_slice = md->md_slice; 2388 sc->sc_balance = md->md_balance; 2389 sc->sc_mediasize = md->md_mediasize; 2390 sc->sc_sectorsize = md->md_sectorsize; 2391 sc->sc_ndisks = md->md_all; 2392 sc->sc_flags = md->md_mflags; 2393 sc->sc_bump_syncid = 0; 2394 bioq_init(&sc->sc_queue); 2395 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2396 LIST_INIT(&sc->sc_disks); 2397 TAILQ_INIT(&sc->sc_events); 2398 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2399 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2400 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2401 gp->softc = sc; 2402 sc->sc_geom = gp; 2403 sc->sc_provider = NULL; 2404 /* 2405 * Synchronization geom. 2406 */ 2407 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2408 gp->softc = sc; 2409 gp->orphan = g_mirror_orphan; 2410 sc->sc_sync.ds_geom = gp; 2411 sc->sc_sync.ds_ndisks = 0; 2412 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2413 "g_mirror %s", md->md_name); 2414 if (error != 0) { 2415 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2416 sc->sc_name); 2417 g_destroy_geom(sc->sc_sync.ds_geom); 2418 mtx_destroy(&sc->sc_events_mtx); 2419 mtx_destroy(&sc->sc_queue_mtx); 2420 g_destroy_geom(sc->sc_geom); 2421 free(sc, M_MIRROR); 2422 return (NULL); 2423 } 2424 2425 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2426 2427 /* 2428 * Run timeout. 2429 */ 2430 timeout = atomic_load_acq_int(&g_mirror_timeout); 2431 callout_reset(&sc->sc_callout, timeout * hz, g_mirror_go, sc); 2432 return (sc->sc_geom); 2433} 2434 2435int 2436g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2437{ 2438 struct g_provider *pp; 2439 2440 g_topology_assert(); 2441 2442 if (sc == NULL) 2443 return (ENXIO); 2444 pp = sc->sc_provider; 2445 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2446 if (force) { 2447 G_MIRROR_DEBUG(0, "Device %s is still open, so it " 2448 "can't be definitely removed.", pp->name); 2449 } else { 2450 G_MIRROR_DEBUG(1, 2451 "Device %s is still open (r%dw%de%d).", pp->name, 2452 pp->acr, pp->acw, pp->ace); 2453 return (EBUSY); 2454 } 2455 } 2456 2457 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2458 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2459 g_topology_unlock(); 2460 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2461 mtx_lock(&sc->sc_queue_mtx); 2462 wakeup(sc); 2463 mtx_unlock(&sc->sc_queue_mtx); 2464 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2465 while (sc->sc_worker != NULL) 2466 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2467 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2468 g_topology_lock(); 2469 g_mirror_destroy_device(sc); 2470 free(sc, M_MIRROR); 2471 return (0); 2472} 2473 2474static void 2475g_mirror_taste_orphan(struct g_consumer *cp) 2476{ 2477 2478 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2479 cp->provider->name)); 2480} 2481 2482static struct g_geom * 2483g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2484{ 2485 struct g_mirror_metadata md; 2486 struct g_mirror_softc *sc; 2487 struct g_consumer *cp; 2488 struct g_geom *gp; 2489 int error; 2490 2491 g_topology_assert(); 2492 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2493 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2494 2495 gp = g_new_geomf(mp, "mirror:taste"); 2496 /* 2497 * This orphan function should be never called. 2498 */ 2499 gp->orphan = g_mirror_taste_orphan; 2500 cp = g_new_consumer(gp); 2501 g_attach(cp, pp); 2502 error = g_mirror_read_metadata(cp, &md); 2503 g_detach(cp); 2504 g_destroy_consumer(cp); 2505 g_destroy_geom(gp); 2506 if (error != 0) 2507 return (NULL); 2508 gp = NULL; 2509 2510 if (md.md_version > G_MIRROR_VERSION) { 2511 printf("geom_mirror.ko module is too old to handle %s.\n", 2512 pp->name); 2513 return (NULL); 2514 } 2515 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2516 return (NULL); 2517 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2518 G_MIRROR_DEBUG(0, 2519 "Device %s: provider %s marked as inactive, skipping.", 2520 md.md_name, pp->name); 2521 return (NULL); 2522 } 2523 if (g_mirror_debug >= 2) 2524 mirror_metadata_dump(&md); 2525 2526 /* 2527 * Let's check if device already exists. 2528 */ 2529 sc = NULL; 2530 LIST_FOREACH(gp, &mp->geom, geom) { 2531 sc = gp->softc; 2532 if (sc == NULL) 2533 continue; 2534 if (sc->sc_sync.ds_geom == gp) 2535 continue; 2536 if (strcmp(md.md_name, sc->sc_name) != 0) 2537 continue; 2538 if (md.md_mid != sc->sc_id) { 2539 G_MIRROR_DEBUG(0, "Device %s already configured.", 2540 sc->sc_name); 2541 return (NULL); 2542 } 2543 break; 2544 } 2545 if (gp == NULL) { 2546 gp = g_mirror_create(mp, &md); 2547 if (gp == NULL) { 2548 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2549 md.md_name); 2550 return (NULL); 2551 } 2552 sc = gp->softc; 2553 } 2554 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2555 error = g_mirror_add_disk(sc, pp, &md); 2556 if (error != 0) { 2557 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2558 pp->name, gp->name, error); 2559 if (LIST_EMPTY(&sc->sc_disks)) 2560 g_mirror_destroy(sc, 1); 2561 return (NULL); 2562 } 2563 return (gp); 2564} 2565 2566static int 2567g_mirror_destroy_geom(struct gctl_req *req __unused, 2568 struct g_class *mp __unused, struct g_geom *gp) 2569{ 2570 2571 return (g_mirror_destroy(gp->softc, 0)); 2572} 2573 2574static void 2575g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2576 struct g_consumer *cp, struct g_provider *pp) 2577{ 2578 struct g_mirror_softc *sc; 2579 2580 g_topology_assert(); 2581 2582 sc = gp->softc; 2583 if (sc == NULL) 2584 return; 2585 /* Skip synchronization geom. */ 2586 if (gp == sc->sc_sync.ds_geom) 2587 return; 2588 if (pp != NULL) { 2589 /* Nothing here. */ 2590 } else if (cp != NULL) { 2591 struct g_mirror_disk *disk; 2592 2593 disk = cp->private; 2594 if (disk == NULL) 2595 return; 2596 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2597 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2598 sbuf_printf(sb, "%s<Synchronized>", indent); 2599 if (disk->d_sync.ds_offset_done == 0) 2600 sbuf_printf(sb, "0%%"); 2601 else { 2602 sbuf_printf(sb, "%u%%", 2603 (u_int)((disk->d_sync.ds_offset_done * 100) / 2604 sc->sc_provider->mediasize)); 2605 } 2606 sbuf_printf(sb, "</Synchronized>\n"); 2607 } 2608 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2609 disk->d_sync.ds_syncid); 2610 sbuf_printf(sb, "%s<Flags>", indent); 2611 if (disk->d_flags == 0) 2612 sbuf_printf(sb, "NONE"); 2613 else { 2614 int first = 1; 2615 2616#define ADD_FLAG(flag, name) do { \ 2617 if ((disk->d_flags & (flag)) != 0) { \ 2618 if (!first) \ 2619 sbuf_printf(sb, ", "); \ 2620 else \ 2621 first = 0; \ 2622 sbuf_printf(sb, name); \ 2623 } \ 2624} while (0) 2625 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2626 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2627 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2628 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2629 "SYNCHRONIZING"); 2630 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2631#undef ADD_FLAG 2632 } 2633 sbuf_printf(sb, "</Flags>\n"); 2634 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2635 disk->d_priority); 2636 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2637 g_mirror_disk_state2str(disk->d_state)); 2638 } else { 2639 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2640 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2641 sbuf_printf(sb, "%s<Flags>", indent); 2642 if (sc->sc_flags == 0) 2643 sbuf_printf(sb, "NONE"); 2644 else { 2645 int first = 1; 2646 2647#define ADD_FLAG(flag, name) do { \ 2648 if ((sc->sc_flags & (flag)) != 0) { \ 2649 if (!first) \ 2650 sbuf_printf(sb, ", "); \ 2651 else \ 2652 first = 0; \ 2653 sbuf_printf(sb, name); \ 2654 } \ 2655} while (0) 2656 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2657#undef ADD_FLAG 2658 } 2659 sbuf_printf(sb, "</Flags>\n"); 2660 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2661 (u_int)sc->sc_slice); 2662 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2663 balance_name(sc->sc_balance)); 2664 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2665 sc->sc_ndisks); 2666 sbuf_printf(sb, "%s<State>", indent); 2667 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2668 sbuf_printf(sb, "%s", "STARTING"); 2669 else if (sc->sc_ndisks == 2670 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 2671 sbuf_printf(sb, "%s", "COMPLETE"); 2672 else 2673 sbuf_printf(sb, "%s", "DEGRADED"); 2674 sbuf_printf(sb, "</State>\n"); 2675 } 2676} 2677 2678DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2679