g_mirror.c revision 139650
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 139650 2005-01-03 19:42:37Z 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 <sys/sched.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 = 4; 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_idletime = 5; 62TUNABLE_INT("kern.geom.mirror.idletime", &g_mirror_idletime); 63SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RW, 64 &g_mirror_idletime, 0, "Mark components as clean when idling"); 65static u_int g_mirror_reqs_per_sync = 5; 66SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, reqs_per_sync, CTLFLAG_RW, 67 &g_mirror_reqs_per_sync, 0, 68 "Number of regular I/O requests per synchronization request"); 69static u_int g_mirror_syncs_per_sec = 100; 70SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, syncs_per_sec, CTLFLAG_RW, 71 &g_mirror_syncs_per_sec, 0, 72 "Number of synchronizations requests per second"); 73 74#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 75 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 76 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 77 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 78} while (0) 79 80static eventhandler_tag g_mirror_ehtag = NULL; 81 82static int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp, 83 struct g_geom *gp); 84static g_taste_t g_mirror_taste; 85static void g_mirror_init(struct g_class *mp); 86static void g_mirror_fini(struct g_class *mp); 87 88struct g_class g_mirror_class = { 89 .name = G_MIRROR_CLASS_NAME, 90 .version = G_VERSION, 91 .ctlreq = g_mirror_config, 92 .taste = g_mirror_taste, 93 .destroy_geom = g_mirror_destroy_geom, 94 .init = g_mirror_init, 95 .fini = g_mirror_fini 96}; 97 98 99static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 100static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 101static void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force); 102static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 103 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 104static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 105 106 107static const char * 108g_mirror_disk_state2str(int state) 109{ 110 111 switch (state) { 112 case G_MIRROR_DISK_STATE_NONE: 113 return ("NONE"); 114 case G_MIRROR_DISK_STATE_NEW: 115 return ("NEW"); 116 case G_MIRROR_DISK_STATE_ACTIVE: 117 return ("ACTIVE"); 118 case G_MIRROR_DISK_STATE_STALE: 119 return ("STALE"); 120 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 121 return ("SYNCHRONIZING"); 122 case G_MIRROR_DISK_STATE_DISCONNECTED: 123 return ("DISCONNECTED"); 124 case G_MIRROR_DISK_STATE_DESTROY: 125 return ("DESTROY"); 126 default: 127 return ("INVALID"); 128 } 129} 130 131static const char * 132g_mirror_device_state2str(int state) 133{ 134 135 switch (state) { 136 case G_MIRROR_DEVICE_STATE_STARTING: 137 return ("STARTING"); 138 case G_MIRROR_DEVICE_STATE_RUNNING: 139 return ("RUNNING"); 140 default: 141 return ("INVALID"); 142 } 143} 144 145static const char * 146g_mirror_get_diskname(struct g_mirror_disk *disk) 147{ 148 149 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 150 return ("[unknown]"); 151 return (disk->d_name); 152} 153 154/* 155 * --- Events handling functions --- 156 * Events in geom_mirror are used to maintain disks and device status 157 * from one thread to simplify locking. 158 */ 159static void 160g_mirror_event_free(struct g_mirror_event *ep) 161{ 162 163 free(ep, M_MIRROR); 164} 165 166int 167g_mirror_event_send(void *arg, int state, int flags) 168{ 169 struct g_mirror_softc *sc; 170 struct g_mirror_disk *disk; 171 struct g_mirror_event *ep; 172 int error; 173 174 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 175 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 176 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 177 disk = NULL; 178 sc = arg; 179 } else { 180 disk = arg; 181 sc = disk->d_softc; 182 } 183 ep->e_disk = disk; 184 ep->e_state = state; 185 ep->e_flags = flags; 186 ep->e_error = 0; 187 mtx_lock(&sc->sc_events_mtx); 188 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 189 mtx_unlock(&sc->sc_events_mtx); 190 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 191 mtx_lock(&sc->sc_queue_mtx); 192 wakeup(sc); 193 mtx_unlock(&sc->sc_queue_mtx); 194 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 195 return (0); 196 g_topology_assert(); 197 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 198 g_topology_unlock(); 199 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 200 mtx_lock(&sc->sc_events_mtx); 201 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 202 hz * 5); 203 } 204 /* Don't even try to use 'sc' here, because it could be already dead. */ 205 g_topology_lock(); 206 error = ep->e_error; 207 g_mirror_event_free(ep); 208 return (error); 209} 210 211static struct g_mirror_event * 212g_mirror_event_get(struct g_mirror_softc *sc) 213{ 214 struct g_mirror_event *ep; 215 216 mtx_lock(&sc->sc_events_mtx); 217 ep = TAILQ_FIRST(&sc->sc_events); 218 mtx_unlock(&sc->sc_events_mtx); 219 return (ep); 220} 221 222static void 223g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 224{ 225 226 mtx_lock(&sc->sc_events_mtx); 227 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 228 mtx_unlock(&sc->sc_events_mtx); 229} 230 231static void 232g_mirror_event_cancel(struct g_mirror_disk *disk) 233{ 234 struct g_mirror_softc *sc; 235 struct g_mirror_event *ep, *tmpep; 236 237 g_topology_assert(); 238 239 sc = disk->d_softc; 240 mtx_lock(&sc->sc_events_mtx); 241 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 242 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 243 continue; 244 if (ep->e_disk != disk) 245 continue; 246 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 247 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 248 g_mirror_event_free(ep); 249 else { 250 ep->e_error = ECANCELED; 251 wakeup(ep); 252 } 253 } 254 mtx_unlock(&sc->sc_events_mtx); 255} 256 257/* 258 * Return the number of disks in given state. 259 * If state is equal to -1, count all connected disks. 260 */ 261u_int 262g_mirror_ndisks(struct g_mirror_softc *sc, int state) 263{ 264 struct g_mirror_disk *disk; 265 u_int n = 0; 266 267 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 268 if (state == -1 || disk->d_state == state) 269 n++; 270 } 271 return (n); 272} 273 274/* 275 * Find a disk in mirror by its disk ID. 276 */ 277static struct g_mirror_disk * 278g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 279{ 280 struct g_mirror_disk *disk; 281 282 g_topology_assert(); 283 284 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 285 if (disk->d_id == id) 286 return (disk); 287 } 288 return (NULL); 289} 290 291static u_int 292g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 293{ 294 struct bio *bp; 295 u_int nreqs = 0; 296 297 mtx_lock(&sc->sc_queue_mtx); 298 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 299 if (bp->bio_from == cp) 300 nreqs++; 301 } 302 mtx_unlock(&sc->sc_queue_mtx); 303 return (nreqs); 304} 305 306static int 307g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 308{ 309 310 if (cp->index > 0) { 311 G_MIRROR_DEBUG(2, 312 "I/O requests for %s exist, can't destroy it now.", 313 cp->provider->name); 314 return (1); 315 } 316 if (g_mirror_nrequests(sc, cp) > 0) { 317 G_MIRROR_DEBUG(2, 318 "I/O requests for %s in queue, can't destroy it now.", 319 cp->provider->name); 320 return (1); 321 } 322 return (0); 323} 324 325static void 326g_mirror_destroy_consumer(void *arg, int flags __unused) 327{ 328 struct g_consumer *cp; 329 330 cp = arg; 331 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 332 g_detach(cp); 333 g_destroy_consumer(cp); 334} 335 336static void 337g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 338{ 339 struct g_provider *pp; 340 int retaste_wait; 341 342 g_topology_assert(); 343 344 cp->private = NULL; 345 if (g_mirror_is_busy(sc, cp)) 346 return; 347 pp = cp->provider; 348 retaste_wait = 0; 349 if (cp->acw == 1) { 350 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 351 retaste_wait = 1; 352 } 353 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 354 -cp->acw, -cp->ace, 0); 355 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 356 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 357 if (retaste_wait) { 358 /* 359 * After retaste event was send (inside g_access()), we can send 360 * event to detach and destroy consumer. 361 * A class, which has consumer to the given provider connected 362 * will not receive retaste event for the provider. 363 * This is the way how I ignore retaste events when I close 364 * consumers opened for write: I detach and destroy consumer 365 * after retaste event is sent. 366 */ 367 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 368 return; 369 } 370 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 371 g_detach(cp); 372 g_destroy_consumer(cp); 373} 374 375static int 376g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 377{ 378 int error; 379 380 g_topology_assert(); 381 KASSERT(disk->d_consumer == NULL, 382 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 383 384 disk->d_consumer = g_new_consumer(disk->d_softc->sc_geom); 385 disk->d_consumer->private = disk; 386 disk->d_consumer->index = 0; 387 error = g_attach(disk->d_consumer, pp); 388 if (error != 0) 389 return (error); 390 error = g_access(disk->d_consumer, 1, 1, 1); 391 if (error != 0) { 392 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 393 pp->name, error); 394 return (error); 395 } 396 397 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 398 return (0); 399} 400 401static void 402g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 403{ 404 405 g_topology_assert(); 406 407 if (cp == NULL) 408 return; 409 if (cp->provider != NULL) 410 g_mirror_kill_consumer(sc, cp); 411 else 412 g_destroy_consumer(cp); 413} 414 415/* 416 * Initialize disk. This means allocate memory, create consumer, attach it 417 * to the provider and open access (r1w1e1) to it. 418 */ 419static struct g_mirror_disk * 420g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 421 struct g_mirror_metadata *md, int *errorp) 422{ 423 struct g_mirror_disk *disk; 424 int error; 425 426 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 427 if (disk == NULL) { 428 error = ENOMEM; 429 goto fail; 430 } 431 disk->d_softc = sc; 432 error = g_mirror_connect_disk(disk, pp); 433 if (error != 0) 434 goto fail; 435 disk->d_id = md->md_did; 436 disk->d_state = G_MIRROR_DISK_STATE_NONE; 437 disk->d_priority = md->md_priority; 438 disk->d_delay.sec = 0; 439 disk->d_delay.frac = 0; 440 binuptime(&disk->d_last_used); 441 disk->d_flags = md->md_dflags; 442 if (md->md_provider[0] != '\0') 443 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 444 disk->d_sync.ds_consumer = NULL; 445 disk->d_sync.ds_offset = md->md_sync_offset; 446 disk->d_sync.ds_offset_done = md->md_sync_offset; 447 disk->d_sync.ds_resync = -1; 448 disk->d_genid = md->md_genid; 449 disk->d_sync.ds_syncid = md->md_syncid; 450 if (errorp != NULL) 451 *errorp = 0; 452 return (disk); 453fail: 454 if (errorp != NULL) 455 *errorp = error; 456 if (disk != NULL) { 457 g_mirror_disconnect_consumer(sc, disk->d_consumer); 458 free(disk, M_MIRROR); 459 } 460 return (NULL); 461} 462 463static void 464g_mirror_destroy_disk(struct g_mirror_disk *disk) 465{ 466 struct g_mirror_softc *sc; 467 468 g_topology_assert(); 469 470 LIST_REMOVE(disk, d_next); 471 g_mirror_event_cancel(disk); 472 sc = disk->d_softc; 473 if (sc->sc_hint == disk) 474 sc->sc_hint = NULL; 475 switch (disk->d_state) { 476 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 477 g_mirror_sync_stop(disk, 1); 478 /* FALLTHROUGH */ 479 case G_MIRROR_DISK_STATE_NEW: 480 case G_MIRROR_DISK_STATE_STALE: 481 case G_MIRROR_DISK_STATE_ACTIVE: 482 g_mirror_disconnect_consumer(sc, disk->d_consumer); 483 free(disk, M_MIRROR); 484 break; 485 default: 486 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 487 g_mirror_get_diskname(disk), 488 g_mirror_disk_state2str(disk->d_state))); 489 } 490} 491 492static void 493g_mirror_destroy_device(struct g_mirror_softc *sc) 494{ 495 struct g_mirror_disk *disk; 496 struct g_mirror_event *ep; 497 struct g_geom *gp; 498 struct g_consumer *cp, *tmpcp; 499 500 g_topology_assert(); 501 502 gp = sc->sc_geom; 503 if (sc->sc_provider != NULL) 504 g_mirror_destroy_provider(sc); 505 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 506 disk = LIST_FIRST(&sc->sc_disks)) { 507 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 508 g_mirror_update_metadata(disk); 509 g_mirror_destroy_disk(disk); 510 } 511 while ((ep = g_mirror_event_get(sc)) != NULL) { 512 g_mirror_event_remove(sc, ep); 513 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 514 g_mirror_event_free(ep); 515 else { 516 ep->e_error = ECANCELED; 517 ep->e_flags |= G_MIRROR_EVENT_DONE; 518 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 519 mtx_lock(&sc->sc_events_mtx); 520 wakeup(ep); 521 mtx_unlock(&sc->sc_events_mtx); 522 } 523 } 524 callout_drain(&sc->sc_callout); 525 gp->softc = NULL; 526 527 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 528 g_mirror_disconnect_consumer(sc, cp); 529 } 530 sc->sc_sync.ds_geom->softc = NULL; 531 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 532 mtx_destroy(&sc->sc_queue_mtx); 533 mtx_destroy(&sc->sc_events_mtx); 534 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 535 g_wither_geom(gp, ENXIO); 536} 537 538static void 539g_mirror_orphan(struct g_consumer *cp) 540{ 541 struct g_mirror_disk *disk; 542 543 g_topology_assert(); 544 545 disk = cp->private; 546 if (disk == NULL) 547 return; 548 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_OFW; 549 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 550 G_MIRROR_EVENT_DONTWAIT); 551} 552 553static void 554g_mirror_spoiled(struct g_consumer *cp) 555{ 556 struct g_mirror_disk *disk; 557 558 g_topology_assert(); 559 560 disk = cp->private; 561 if (disk == NULL) 562 return; 563 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_IMM; 564 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 565 G_MIRROR_EVENT_DONTWAIT); 566} 567 568/* 569 * Function should return the next active disk on the list. 570 * It is possible that it will be the same disk as given. 571 * If there are no active disks on list, NULL is returned. 572 */ 573static __inline struct g_mirror_disk * 574g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 575{ 576 struct g_mirror_disk *dp; 577 578 for (dp = LIST_NEXT(disk, d_next); dp != disk; 579 dp = LIST_NEXT(dp, d_next)) { 580 if (dp == NULL) 581 dp = LIST_FIRST(&sc->sc_disks); 582 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 583 break; 584 } 585 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 586 return (NULL); 587 return (dp); 588} 589 590static struct g_mirror_disk * 591g_mirror_get_disk(struct g_mirror_softc *sc) 592{ 593 struct g_mirror_disk *disk; 594 595 if (sc->sc_hint == NULL) { 596 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 597 if (sc->sc_hint == NULL) 598 return (NULL); 599 } 600 disk = sc->sc_hint; 601 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 602 disk = g_mirror_find_next(sc, disk); 603 if (disk == NULL) 604 return (NULL); 605 } 606 sc->sc_hint = g_mirror_find_next(sc, disk); 607 return (disk); 608} 609 610static int 611g_mirror_write_metadata(struct g_mirror_disk *disk, 612 struct g_mirror_metadata *md) 613{ 614 struct g_mirror_softc *sc; 615 struct g_consumer *cp; 616 off_t offset, length; 617 u_char *sector; 618 int error = 0; 619 620 g_topology_assert(); 621 622 sc = disk->d_softc; 623 cp = disk->d_consumer; 624 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 625 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 626 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 627 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 628 cp->acw, cp->ace)); 629 length = cp->provider->sectorsize; 630 offset = cp->provider->mediasize - length; 631 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 632 if (md != NULL) 633 mirror_metadata_encode(md, sector); 634 g_topology_unlock(); 635 error = g_write_data(cp, offset, sector, length); 636 g_topology_lock(); 637 free(sector, M_MIRROR); 638 if (error != 0) { 639 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_GENID_IMM; 640 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 641 G_MIRROR_EVENT_DONTWAIT); 642 } 643 return (error); 644} 645 646static int 647g_mirror_clear_metadata(struct g_mirror_disk *disk) 648{ 649 int error; 650 651 g_topology_assert(); 652 error = g_mirror_write_metadata(disk, NULL); 653 if (error == 0) { 654 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 655 g_mirror_get_diskname(disk)); 656 } else { 657 G_MIRROR_DEBUG(0, 658 "Cannot clear metadata on disk %s (error=%d).", 659 g_mirror_get_diskname(disk), error); 660 } 661 return (error); 662} 663 664void 665g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 666 struct g_mirror_metadata *md) 667{ 668 669 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 670 md->md_version = G_MIRROR_VERSION; 671 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 672 md->md_mid = sc->sc_id; 673 md->md_all = sc->sc_ndisks; 674 md->md_slice = sc->sc_slice; 675 md->md_balance = sc->sc_balance; 676 md->md_genid = sc->sc_genid; 677 md->md_mediasize = sc->sc_mediasize; 678 md->md_sectorsize = sc->sc_sectorsize; 679 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 680 bzero(md->md_provider, sizeof(md->md_provider)); 681 if (disk == NULL) { 682 md->md_did = arc4random(); 683 md->md_priority = 0; 684 md->md_syncid = 0; 685 md->md_dflags = 0; 686 md->md_sync_offset = 0; 687 } else { 688 md->md_did = disk->d_id; 689 md->md_priority = disk->d_priority; 690 md->md_syncid = disk->d_sync.ds_syncid; 691 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 692 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 693 md->md_sync_offset = disk->d_sync.ds_offset_done; 694 else 695 md->md_sync_offset = 0; 696 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 697 strlcpy(md->md_provider, 698 disk->d_consumer->provider->name, 699 sizeof(md->md_provider)); 700 } 701 } 702} 703 704void 705g_mirror_update_metadata(struct g_mirror_disk *disk) 706{ 707 struct g_mirror_metadata md; 708 int error; 709 710 g_topology_assert(); 711 g_mirror_fill_metadata(disk->d_softc, disk, &md); 712 error = g_mirror_write_metadata(disk, &md); 713 if (error == 0) { 714 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 715 g_mirror_get_diskname(disk)); 716 } else { 717 G_MIRROR_DEBUG(0, 718 "Cannot update metadata on disk %s (error=%d).", 719 g_mirror_get_diskname(disk), error); 720 } 721} 722 723static void 724g_mirror_bump_syncid(struct g_mirror_softc *sc) 725{ 726 struct g_mirror_disk *disk; 727 728 g_topology_assert(); 729 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 730 ("%s called with no active disks (device=%s).", __func__, 731 sc->sc_name)); 732 733 sc->sc_syncid++; 734 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 735 sc->sc_syncid); 736 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 737 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 738 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 739 disk->d_sync.ds_syncid = sc->sc_syncid; 740 g_mirror_update_metadata(disk); 741 } 742 } 743} 744 745static void 746g_mirror_bump_genid(struct g_mirror_softc *sc) 747{ 748 struct g_mirror_disk *disk; 749 750 g_topology_assert(); 751 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 752 ("%s called with no active disks (device=%s).", __func__, 753 sc->sc_name)); 754 755 sc->sc_genid++; 756 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 757 sc->sc_genid); 758 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 759 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 760 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 761 disk->d_genid = sc->sc_genid; 762 g_mirror_update_metadata(disk); 763 } 764 } 765} 766 767static void 768g_mirror_idle(struct g_mirror_softc *sc) 769{ 770 struct g_mirror_disk *disk; 771 772 if (sc->sc_provider == NULL || sc->sc_provider->acw == 0) 773 return; 774 sc->sc_idle = 1; 775 g_topology_lock(); 776 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 777 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 778 continue; 779 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 780 g_mirror_get_diskname(disk), sc->sc_name); 781 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 782 g_mirror_update_metadata(disk); 783 } 784 g_topology_unlock(); 785} 786 787static void 788g_mirror_unidle(struct g_mirror_softc *sc) 789{ 790 struct g_mirror_disk *disk; 791 792 sc->sc_idle = 0; 793 g_topology_lock(); 794 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 795 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 796 continue; 797 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 798 g_mirror_get_diskname(disk), sc->sc_name); 799 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 800 g_mirror_update_metadata(disk); 801 } 802 g_topology_unlock(); 803} 804 805/* 806 * Return 1 if we should check if mirror is idling. 807 */ 808static int 809g_mirror_check_idle(struct g_mirror_softc *sc) 810{ 811 struct g_mirror_disk *disk; 812 813 if (sc->sc_idle) 814 return (0); 815 if (sc->sc_provider != NULL && sc->sc_provider->acw == 0) 816 return (0); 817 /* 818 * Check if there are no in-flight requests. 819 */ 820 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 821 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 822 continue; 823 if (disk->d_consumer->index > 0) 824 return (0); 825 } 826 return (1); 827} 828 829static __inline int 830bintime_cmp(struct bintime *bt1, struct bintime *bt2) 831{ 832 833 if (bt1->sec < bt2->sec) 834 return (-1); 835 else if (bt1->sec > bt2->sec) 836 return (1); 837 if (bt1->frac < bt2->frac) 838 return (-1); 839 else if (bt1->frac > bt2->frac) 840 return (1); 841 return (0); 842} 843 844static void 845g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 846{ 847 848 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 849 return; 850 binuptime(&disk->d_delay); 851 bintime_sub(&disk->d_delay, &bp->bio_t0); 852} 853 854static void 855g_mirror_done(struct bio *bp) 856{ 857 struct g_mirror_softc *sc; 858 859 sc = bp->bio_from->geom->softc; 860 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 861 mtx_lock(&sc->sc_queue_mtx); 862 bioq_disksort(&sc->sc_queue, bp); 863 wakeup(sc); 864 mtx_unlock(&sc->sc_queue_mtx); 865} 866 867static void 868g_mirror_regular_request(struct bio *bp) 869{ 870 struct g_mirror_softc *sc; 871 struct g_mirror_disk *disk; 872 struct bio *pbp; 873 874 g_topology_assert_not(); 875 876 bp->bio_from->index--; 877 pbp = bp->bio_parent; 878 sc = pbp->bio_to->geom->softc; 879 disk = bp->bio_from->private; 880 if (disk == NULL) { 881 g_topology_lock(); 882 g_mirror_kill_consumer(sc, bp->bio_from); 883 g_topology_unlock(); 884 } else { 885 g_mirror_update_delay(disk, bp); 886 } 887 888 pbp->bio_inbed++; 889 KASSERT(pbp->bio_inbed <= pbp->bio_children, 890 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 891 pbp->bio_children)); 892 if (bp->bio_error == 0 && pbp->bio_error == 0) { 893 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 894 g_destroy_bio(bp); 895 if (pbp->bio_children == pbp->bio_inbed) { 896 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 897 pbp->bio_completed = pbp->bio_length; 898 g_io_deliver(pbp, pbp->bio_error); 899 } 900 return; 901 } else if (bp->bio_error != 0) { 902 if (pbp->bio_error == 0) 903 pbp->bio_error = bp->bio_error; 904 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 905 bp->bio_error); 906 if (disk != NULL) { 907 sc->sc_bump_id |= G_MIRROR_BUMP_GENID_IMM; 908 g_mirror_event_send(disk, 909 G_MIRROR_DISK_STATE_DISCONNECTED, 910 G_MIRROR_EVENT_DONTWAIT); 911 } 912 switch (pbp->bio_cmd) { 913 case BIO_DELETE: 914 case BIO_WRITE: 915 pbp->bio_inbed--; 916 pbp->bio_children--; 917 break; 918 } 919 } 920 g_destroy_bio(bp); 921 922 switch (pbp->bio_cmd) { 923 case BIO_READ: 924 if (pbp->bio_children == pbp->bio_inbed) { 925 pbp->bio_error = 0; 926 mtx_lock(&sc->sc_queue_mtx); 927 bioq_disksort(&sc->sc_queue, pbp); 928 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 929 wakeup(sc); 930 mtx_unlock(&sc->sc_queue_mtx); 931 } 932 break; 933 case BIO_DELETE: 934 case BIO_WRITE: 935 if (pbp->bio_children == 0) { 936 /* 937 * All requests failed. 938 */ 939 } else if (pbp->bio_inbed < pbp->bio_children) { 940 /* Do nothing. */ 941 break; 942 } else if (pbp->bio_children == pbp->bio_inbed) { 943 /* Some requests succeeded. */ 944 pbp->bio_error = 0; 945 pbp->bio_completed = pbp->bio_length; 946 } 947 g_io_deliver(pbp, pbp->bio_error); 948 break; 949 default: 950 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 951 break; 952 } 953} 954 955static void 956g_mirror_sync_done(struct bio *bp) 957{ 958 struct g_mirror_softc *sc; 959 960 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 961 sc = bp->bio_from->geom->softc; 962 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 963 mtx_lock(&sc->sc_queue_mtx); 964 bioq_disksort(&sc->sc_queue, bp); 965 wakeup(sc); 966 mtx_unlock(&sc->sc_queue_mtx); 967} 968 969static void 970g_mirror_start(struct bio *bp) 971{ 972 struct g_mirror_softc *sc; 973 974 sc = bp->bio_to->geom->softc; 975 /* 976 * If sc == NULL or there are no valid disks, provider's error 977 * should be set and g_mirror_start() should not be called at all. 978 */ 979 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 980 ("Provider's error should be set (error=%d)(mirror=%s).", 981 bp->bio_to->error, bp->bio_to->name)); 982 G_MIRROR_LOGREQ(3, bp, "Request received."); 983 984 switch (bp->bio_cmd) { 985 case BIO_READ: 986 case BIO_WRITE: 987 case BIO_DELETE: 988 break; 989 case BIO_GETATTR: 990 default: 991 g_io_deliver(bp, EOPNOTSUPP); 992 return; 993 } 994 mtx_lock(&sc->sc_queue_mtx); 995 bioq_disksort(&sc->sc_queue, bp); 996 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 997 wakeup(sc); 998 mtx_unlock(&sc->sc_queue_mtx); 999} 1000 1001/* 1002 * Send one synchronization request. 1003 */ 1004static void 1005g_mirror_sync_one(struct g_mirror_disk *disk) 1006{ 1007 struct g_mirror_softc *sc; 1008 struct bio *bp; 1009 1010 sc = disk->d_softc; 1011 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1012 ("Disk %s is not marked for synchronization.", 1013 g_mirror_get_diskname(disk))); 1014 1015 bp = g_new_bio(); 1016 if (bp == NULL) 1017 return; 1018 bp->bio_parent = NULL; 1019 bp->bio_cmd = BIO_READ; 1020 bp->bio_offset = disk->d_sync.ds_offset; 1021 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1022 bp->bio_cflags = 0; 1023 bp->bio_done = g_mirror_sync_done; 1024 bp->bio_data = disk->d_sync.ds_data; 1025 if (bp->bio_data == NULL) { 1026 g_destroy_bio(bp); 1027 return; 1028 } 1029 disk->d_sync.ds_offset += bp->bio_length; 1030 bp->bio_to = sc->sc_provider; 1031 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1032 disk->d_sync.ds_consumer->index++; 1033 g_io_request(bp, disk->d_sync.ds_consumer); 1034} 1035 1036static void 1037g_mirror_sync_request(struct bio *bp) 1038{ 1039 struct g_mirror_softc *sc; 1040 struct g_mirror_disk *disk; 1041 1042 bp->bio_from->index--; 1043 sc = bp->bio_from->geom->softc; 1044 disk = bp->bio_from->private; 1045 if (disk == NULL) { 1046 g_topology_lock(); 1047 g_mirror_kill_consumer(sc, bp->bio_from); 1048 g_topology_unlock(); 1049 g_destroy_bio(bp); 1050 return; 1051 } 1052 1053 /* 1054 * Synchronization request. 1055 */ 1056 switch (bp->bio_cmd) { 1057 case BIO_READ: 1058 { 1059 struct g_consumer *cp; 1060 1061 if (bp->bio_error != 0) { 1062 G_MIRROR_LOGREQ(0, bp, 1063 "Synchronization request failed (error=%d).", 1064 bp->bio_error); 1065 g_destroy_bio(bp); 1066 return; 1067 } 1068 G_MIRROR_LOGREQ(3, bp, 1069 "Synchronization request half-finished."); 1070 bp->bio_cmd = BIO_WRITE; 1071 bp->bio_cflags = 0; 1072 cp = disk->d_consumer; 1073 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1074 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1075 cp->acr, cp->acw, cp->ace)); 1076 cp->index++; 1077 g_io_request(bp, cp); 1078 return; 1079 } 1080 case BIO_WRITE: 1081 { 1082 struct g_mirror_disk_sync *sync; 1083 1084 if (bp->bio_error != 0) { 1085 G_MIRROR_LOGREQ(0, bp, 1086 "Synchronization request failed (error=%d).", 1087 bp->bio_error); 1088 g_destroy_bio(bp); 1089 sc->sc_bump_id |= G_MIRROR_BUMP_GENID_IMM; 1090 g_mirror_event_send(disk, 1091 G_MIRROR_DISK_STATE_DISCONNECTED, 1092 G_MIRROR_EVENT_DONTWAIT); 1093 return; 1094 } 1095 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1096 sync = &disk->d_sync; 1097 sync->ds_offset_done = bp->bio_offset + bp->bio_length; 1098 g_destroy_bio(bp); 1099 if (sync->ds_resync != -1) 1100 break; 1101 if (sync->ds_offset_done == sc->sc_provider->mediasize) { 1102 /* 1103 * Disk up-to-date, activate it. 1104 */ 1105 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1106 G_MIRROR_EVENT_DONTWAIT); 1107 return; 1108 } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) { 1109 /* 1110 * Update offset_done on every 100 blocks. 1111 * XXX: This should be configurable. 1112 */ 1113 g_topology_lock(); 1114 g_mirror_update_metadata(disk); 1115 g_topology_unlock(); 1116 } 1117 return; 1118 } 1119 default: 1120 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1121 bp->bio_cmd, sc->sc_name)); 1122 break; 1123 } 1124} 1125 1126static void 1127g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1128{ 1129 struct g_mirror_disk *disk; 1130 struct g_consumer *cp; 1131 struct bio *cbp; 1132 1133 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1134 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1135 break; 1136 } 1137 if (disk == NULL) { 1138 if (bp->bio_error == 0) 1139 bp->bio_error = ENXIO; 1140 g_io_deliver(bp, bp->bio_error); 1141 return; 1142 } 1143 cbp = g_clone_bio(bp); 1144 if (cbp == NULL) { 1145 if (bp->bio_error == 0) 1146 bp->bio_error = ENOMEM; 1147 g_io_deliver(bp, bp->bio_error); 1148 return; 1149 } 1150 /* 1151 * Fill in the component buf structure. 1152 */ 1153 cp = disk->d_consumer; 1154 cbp->bio_done = g_mirror_done; 1155 cbp->bio_to = cp->provider; 1156 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1157 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1158 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1159 cp->acw, cp->ace)); 1160 cp->index++; 1161 g_io_request(cbp, cp); 1162} 1163 1164static void 1165g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1166{ 1167 struct g_mirror_disk *disk; 1168 struct g_consumer *cp; 1169 struct bio *cbp; 1170 1171 disk = g_mirror_get_disk(sc); 1172 if (disk == NULL) { 1173 if (bp->bio_error == 0) 1174 bp->bio_error = ENXIO; 1175 g_io_deliver(bp, bp->bio_error); 1176 return; 1177 } 1178 cbp = g_clone_bio(bp); 1179 if (cbp == NULL) { 1180 if (bp->bio_error == 0) 1181 bp->bio_error = ENOMEM; 1182 g_io_deliver(bp, bp->bio_error); 1183 return; 1184 } 1185 /* 1186 * Fill in the component buf structure. 1187 */ 1188 cp = disk->d_consumer; 1189 cbp->bio_done = g_mirror_done; 1190 cbp->bio_to = cp->provider; 1191 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1192 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1193 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1194 cp->acw, cp->ace)); 1195 cp->index++; 1196 g_io_request(cbp, cp); 1197} 1198 1199static void 1200g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1201{ 1202 struct g_mirror_disk *disk, *dp; 1203 struct g_consumer *cp; 1204 struct bio *cbp; 1205 struct bintime curtime; 1206 1207 binuptime(&curtime); 1208 /* 1209 * Find a disk which the smallest load. 1210 */ 1211 disk = NULL; 1212 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1213 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1214 continue; 1215 /* If disk wasn't used for more than 2 sec, use it. */ 1216 if (curtime.sec - dp->d_last_used.sec >= 2) { 1217 disk = dp; 1218 break; 1219 } 1220 if (disk == NULL || 1221 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1222 disk = dp; 1223 } 1224 } 1225 cbp = g_clone_bio(bp); 1226 if (cbp == NULL) { 1227 if (bp->bio_error == 0) 1228 bp->bio_error = ENOMEM; 1229 g_io_deliver(bp, bp->bio_error); 1230 return; 1231 } 1232 /* 1233 * Fill in the component buf structure. 1234 */ 1235 cp = disk->d_consumer; 1236 cbp->bio_done = g_mirror_done; 1237 cbp->bio_to = cp->provider; 1238 binuptime(&disk->d_last_used); 1239 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1240 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1241 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1242 cp->acw, cp->ace)); 1243 cp->index++; 1244 g_io_request(cbp, cp); 1245} 1246 1247static void 1248g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1249{ 1250 struct bio_queue_head queue; 1251 struct g_mirror_disk *disk; 1252 struct g_consumer *cp; 1253 struct bio *cbp; 1254 off_t left, mod, offset, slice; 1255 u_char *data; 1256 u_int ndisks; 1257 1258 if (bp->bio_length <= sc->sc_slice) { 1259 g_mirror_request_round_robin(sc, bp); 1260 return; 1261 } 1262 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1263 slice = bp->bio_length / ndisks; 1264 mod = slice % sc->sc_provider->sectorsize; 1265 if (mod != 0) 1266 slice += sc->sc_provider->sectorsize - mod; 1267 /* 1268 * Allocate all bios before sending any request, so we can 1269 * return ENOMEM in nice and clean way. 1270 */ 1271 left = bp->bio_length; 1272 offset = bp->bio_offset; 1273 data = bp->bio_data; 1274 bioq_init(&queue); 1275 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1276 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1277 continue; 1278 cbp = g_clone_bio(bp); 1279 if (cbp == NULL) { 1280 for (cbp = bioq_first(&queue); cbp != NULL; 1281 cbp = bioq_first(&queue)) { 1282 bioq_remove(&queue, cbp); 1283 g_destroy_bio(cbp); 1284 } 1285 if (bp->bio_error == 0) 1286 bp->bio_error = ENOMEM; 1287 g_io_deliver(bp, bp->bio_error); 1288 return; 1289 } 1290 bioq_insert_tail(&queue, cbp); 1291 cbp->bio_done = g_mirror_done; 1292 cbp->bio_caller1 = disk; 1293 cbp->bio_to = disk->d_consumer->provider; 1294 cbp->bio_offset = offset; 1295 cbp->bio_data = data; 1296 cbp->bio_length = MIN(left, slice); 1297 left -= cbp->bio_length; 1298 if (left == 0) 1299 break; 1300 offset += cbp->bio_length; 1301 data += cbp->bio_length; 1302 } 1303 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1304 bioq_remove(&queue, cbp); 1305 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1306 disk = cbp->bio_caller1; 1307 cbp->bio_caller1 = NULL; 1308 cp = disk->d_consumer; 1309 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1310 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1311 cp->acr, cp->acw, cp->ace)); 1312 disk->d_consumer->index++; 1313 g_io_request(cbp, disk->d_consumer); 1314 } 1315} 1316 1317static void 1318g_mirror_register_request(struct bio *bp) 1319{ 1320 struct g_mirror_softc *sc; 1321 1322 sc = bp->bio_to->geom->softc; 1323 switch (bp->bio_cmd) { 1324 case BIO_READ: 1325 switch (sc->sc_balance) { 1326 case G_MIRROR_BALANCE_LOAD: 1327 g_mirror_request_load(sc, bp); 1328 break; 1329 case G_MIRROR_BALANCE_PREFER: 1330 g_mirror_request_prefer(sc, bp); 1331 break; 1332 case G_MIRROR_BALANCE_ROUND_ROBIN: 1333 g_mirror_request_round_robin(sc, bp); 1334 break; 1335 case G_MIRROR_BALANCE_SPLIT: 1336 g_mirror_request_split(sc, bp); 1337 break; 1338 } 1339 return; 1340 case BIO_WRITE: 1341 case BIO_DELETE: 1342 { 1343 struct g_mirror_disk *disk; 1344 struct g_mirror_disk_sync *sync; 1345 struct bio_queue_head queue; 1346 struct g_consumer *cp; 1347 struct bio *cbp; 1348 1349 if (sc->sc_idle) 1350 g_mirror_unidle(sc); 1351 /* 1352 * Allocate all bios before sending any request, so we can 1353 * return ENOMEM in nice and clean way. 1354 */ 1355 bioq_init(&queue); 1356 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1357 sync = &disk->d_sync; 1358 switch (disk->d_state) { 1359 case G_MIRROR_DISK_STATE_ACTIVE: 1360 break; 1361 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1362 if (bp->bio_offset >= sync->ds_offset) 1363 continue; 1364 else if (bp->bio_offset + bp->bio_length > 1365 sync->ds_offset_done && 1366 (bp->bio_offset < sync->ds_resync || 1367 sync->ds_resync == -1)) { 1368 sync->ds_resync = bp->bio_offset - 1369 (bp->bio_offset % MAXPHYS); 1370 } 1371 break; 1372 default: 1373 continue; 1374 } 1375 cbp = g_clone_bio(bp); 1376 if (cbp == NULL) { 1377 for (cbp = bioq_first(&queue); cbp != NULL; 1378 cbp = bioq_first(&queue)) { 1379 bioq_remove(&queue, cbp); 1380 g_destroy_bio(cbp); 1381 } 1382 if (bp->bio_error == 0) 1383 bp->bio_error = ENOMEM; 1384 g_io_deliver(bp, bp->bio_error); 1385 return; 1386 } 1387 bioq_insert_tail(&queue, cbp); 1388 cbp->bio_done = g_mirror_done; 1389 cp = disk->d_consumer; 1390 cbp->bio_caller1 = cp; 1391 cbp->bio_to = cp->provider; 1392 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1393 ("Consumer %s not opened (r%dw%de%d).", 1394 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1395 } 1396 for (cbp = bioq_first(&queue); cbp != NULL; 1397 cbp = bioq_first(&queue)) { 1398 bioq_remove(&queue, cbp); 1399 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1400 cp = cbp->bio_caller1; 1401 cbp->bio_caller1 = NULL; 1402 cp->index++; 1403 g_io_request(cbp, cp); 1404 } 1405 /* 1406 * Bump syncid on first write. 1407 */ 1408 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_OFW) != 0) { 1409 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1410 g_topology_lock(); 1411 g_mirror_bump_syncid(sc); 1412 g_topology_unlock(); 1413 } 1414 return; 1415 } 1416 default: 1417 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1418 bp->bio_cmd, sc->sc_name)); 1419 break; 1420 } 1421} 1422 1423static int 1424g_mirror_can_destroy(struct g_mirror_softc *sc) 1425{ 1426 struct g_geom *gp; 1427 struct g_consumer *cp; 1428 1429 g_topology_assert(); 1430 gp = sc->sc_geom; 1431 LIST_FOREACH(cp, &gp->consumer, consumer) { 1432 if (g_mirror_is_busy(sc, cp)) 1433 return (0); 1434 } 1435 gp = sc->sc_sync.ds_geom; 1436 LIST_FOREACH(cp, &gp->consumer, consumer) { 1437 if (g_mirror_is_busy(sc, cp)) 1438 return (0); 1439 } 1440 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1441 sc->sc_name); 1442 return (1); 1443} 1444 1445static int 1446g_mirror_try_destroy(struct g_mirror_softc *sc) 1447{ 1448 1449 g_topology_lock(); 1450 if (!g_mirror_can_destroy(sc)) { 1451 g_topology_unlock(); 1452 return (0); 1453 } 1454 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1455 g_topology_unlock(); 1456 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1457 &sc->sc_worker); 1458 wakeup(&sc->sc_worker); 1459 sc->sc_worker = NULL; 1460 } else { 1461 g_mirror_destroy_device(sc); 1462 g_topology_unlock(); 1463 free(sc, M_MIRROR); 1464 } 1465 return (1); 1466} 1467 1468/* 1469 * Worker thread. 1470 */ 1471static void 1472g_mirror_worker(void *arg) 1473{ 1474 struct g_mirror_softc *sc; 1475 struct g_mirror_disk *disk; 1476 struct g_mirror_disk_sync *sync; 1477 struct g_mirror_event *ep; 1478 struct bio *bp; 1479 u_int nreqs; 1480 1481 sc = arg; 1482 mtx_lock_spin(&sched_lock); 1483 sched_prio(curthread, PRIBIO); 1484 mtx_unlock_spin(&sched_lock); 1485 1486 nreqs = 0; 1487 for (;;) { 1488 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1489 /* 1490 * First take a look at events. 1491 * This is important to handle events before any I/O requests. 1492 */ 1493 ep = g_mirror_event_get(sc); 1494 if (ep != NULL && g_topology_try_lock()) { 1495 g_mirror_event_remove(sc, ep); 1496 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1497 /* Update only device status. */ 1498 G_MIRROR_DEBUG(3, 1499 "Running event for device %s.", 1500 sc->sc_name); 1501 ep->e_error = 0; 1502 g_mirror_update_device(sc, 1); 1503 } else { 1504 /* Update disk status. */ 1505 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1506 g_mirror_get_diskname(ep->e_disk)); 1507 ep->e_error = g_mirror_update_disk(ep->e_disk, 1508 ep->e_state); 1509 if (ep->e_error == 0) 1510 g_mirror_update_device(sc, 0); 1511 } 1512 g_topology_unlock(); 1513 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1514 KASSERT(ep->e_error == 0, 1515 ("Error cannot be handled.")); 1516 g_mirror_event_free(ep); 1517 } else { 1518 ep->e_flags |= G_MIRROR_EVENT_DONE; 1519 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1520 ep); 1521 mtx_lock(&sc->sc_events_mtx); 1522 wakeup(ep); 1523 mtx_unlock(&sc->sc_events_mtx); 1524 } 1525 if ((sc->sc_flags & 1526 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1527 if (g_mirror_try_destroy(sc)) 1528 kthread_exit(0); 1529 } 1530 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1531 continue; 1532 } 1533 /* 1534 * Now I/O requests. 1535 */ 1536 /* Get first request from the queue. */ 1537 mtx_lock(&sc->sc_queue_mtx); 1538 bp = bioq_first(&sc->sc_queue); 1539 if (bp == NULL) { 1540 if (ep != NULL) { 1541 /* 1542 * No I/O requests and topology lock was 1543 * already held? Try again. 1544 */ 1545 mtx_unlock(&sc->sc_queue_mtx); 1546 continue; 1547 } 1548 if ((sc->sc_flags & 1549 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1550 mtx_unlock(&sc->sc_queue_mtx); 1551 if (g_mirror_try_destroy(sc)) 1552 kthread_exit(0); 1553 mtx_lock(&sc->sc_queue_mtx); 1554 } 1555 } 1556 if (sc->sc_sync.ds_ndisks > 0 && 1557 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1558 mtx_unlock(&sc->sc_queue_mtx); 1559 /* 1560 * It is time for synchronization... 1561 */ 1562 nreqs = 0; 1563 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1564 if (disk->d_state != 1565 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1566 continue; 1567 } 1568 sync = &disk->d_sync; 1569 if (sync->ds_offset >= 1570 sc->sc_provider->mediasize) { 1571 continue; 1572 } 1573 if (sync->ds_offset > sync->ds_offset_done) 1574 continue; 1575 if (sync->ds_resync != -1) { 1576 sync->ds_offset = sync->ds_resync; 1577 sync->ds_offset_done = sync->ds_resync; 1578 sync->ds_resync = -1; 1579 } 1580 g_mirror_sync_one(disk); 1581 } 1582 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1583 goto sleep; 1584 } 1585 if (bp == NULL) { 1586 if (g_mirror_check_idle(sc)) { 1587 u_int idletime; 1588 1589 idletime = g_mirror_idletime; 1590 if (idletime == 0) 1591 idletime = 1; 1592 idletime *= hz; 1593 if (msleep(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1594 "m:w1", idletime) == EWOULDBLOCK) { 1595 G_MIRROR_DEBUG(5, "%s: I'm here 3.", 1596 __func__); 1597 /* 1598 * No I/O requests in 'idletime' seconds, 1599 * so mark components as clean. 1600 */ 1601 g_mirror_idle(sc); 1602 } 1603 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1604 } else { 1605 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1606 "m:w2", 0); 1607 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1608 } 1609 continue; 1610 } 1611 nreqs++; 1612 bioq_remove(&sc->sc_queue, bp); 1613 mtx_unlock(&sc->sc_queue_mtx); 1614 1615 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1616 g_mirror_regular_request(bp); 1617 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1618 u_int timeout, sps; 1619 1620 g_mirror_sync_request(bp); 1621sleep: 1622 sps = g_mirror_syncs_per_sec; 1623 if (sps == 0) { 1624 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1625 continue; 1626 } 1627 if (ep != NULL) { 1628 /* 1629 * We have some pending events, don't sleep now. 1630 */ 1631 G_MIRROR_DEBUG(5, "%s: I'm here 7.", __func__); 1632 continue; 1633 } 1634 mtx_lock(&sc->sc_queue_mtx); 1635 if (bioq_first(&sc->sc_queue) != NULL) { 1636 mtx_unlock(&sc->sc_queue_mtx); 1637 G_MIRROR_DEBUG(5, "%s: I'm here 8.", __func__); 1638 continue; 1639 } 1640 timeout = hz / sps; 1641 if (timeout == 0) 1642 timeout = 1; 1643 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w3", 1644 timeout); 1645 } else { 1646 g_mirror_register_request(bp); 1647 } 1648 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1649 } 1650} 1651 1652/* 1653 * Open disk's consumer if needed. 1654 */ 1655static void 1656g_mirror_update_access(struct g_mirror_disk *disk) 1657{ 1658 struct g_provider *pp; 1659 1660 g_topology_assert(); 1661 1662 pp = disk->d_softc->sc_provider; 1663 if (pp == NULL) 1664 return; 1665 if (pp->acw > 0) { 1666 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1667 G_MIRROR_DEBUG(1, 1668 "Disk %s (device %s) marked as dirty.", 1669 g_mirror_get_diskname(disk), 1670 disk->d_softc->sc_name); 1671 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1672 } 1673 } else if (pp->acw == 0) { 1674 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1675 G_MIRROR_DEBUG(1, 1676 "Disk %s (device %s) marked as clean.", 1677 g_mirror_get_diskname(disk), 1678 disk->d_softc->sc_name); 1679 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1680 } 1681 } 1682} 1683 1684static void 1685g_mirror_sync_start(struct g_mirror_disk *disk) 1686{ 1687 struct g_mirror_softc *sc; 1688 int error; 1689 1690 g_topology_assert(); 1691 1692 sc = disk->d_softc; 1693 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1694 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1695 sc->sc_state)); 1696 1697 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1698 g_mirror_get_diskname(disk)); 1699 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1700 KASSERT(disk->d_sync.ds_consumer == NULL, 1701 ("Sync consumer already exists (device=%s, disk=%s).", 1702 sc->sc_name, g_mirror_get_diskname(disk))); 1703 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1704 disk->d_sync.ds_consumer->private = disk; 1705 disk->d_sync.ds_consumer->index = 0; 1706 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1707 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1708 disk->d_softc->sc_name, error)); 1709 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1710 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1711 disk->d_softc->sc_name, error)); 1712 disk->d_sync.ds_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1713 sc->sc_sync.ds_ndisks++; 1714} 1715 1716/* 1717 * Stop synchronization process. 1718 * type: 0 - synchronization finished 1719 * 1 - synchronization stopped 1720 */ 1721static void 1722g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1723{ 1724 1725 g_topology_assert(); 1726 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1727 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1728 g_mirror_disk_state2str(disk->d_state))); 1729 if (disk->d_sync.ds_consumer == NULL) 1730 return; 1731 1732 if (type == 0) { 1733 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1734 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1735 } else /* if (type == 1) */ { 1736 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1737 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1738 } 1739 g_mirror_kill_consumer(disk->d_softc, disk->d_sync.ds_consumer); 1740 free(disk->d_sync.ds_data, M_MIRROR); 1741 disk->d_sync.ds_consumer = NULL; 1742 disk->d_softc->sc_sync.ds_ndisks--; 1743 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1744} 1745 1746static void 1747g_mirror_launch_provider(struct g_mirror_softc *sc) 1748{ 1749 struct g_mirror_disk *disk; 1750 struct g_provider *pp; 1751 1752 g_topology_assert(); 1753 1754 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1755 pp->mediasize = sc->sc_mediasize; 1756 pp->sectorsize = sc->sc_sectorsize; 1757 sc->sc_provider = pp; 1758 g_error_provider(pp, 0); 1759 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1760 pp->name); 1761 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1762 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1763 g_mirror_sync_start(disk); 1764 } 1765} 1766 1767static void 1768g_mirror_destroy_provider(struct g_mirror_softc *sc) 1769{ 1770 struct g_mirror_disk *disk; 1771 struct bio *bp; 1772 1773 g_topology_assert(); 1774 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1775 sc->sc_name)); 1776 1777 g_error_provider(sc->sc_provider, ENXIO); 1778 mtx_lock(&sc->sc_queue_mtx); 1779 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1780 bioq_remove(&sc->sc_queue, bp); 1781 g_io_deliver(bp, ENXIO); 1782 } 1783 mtx_unlock(&sc->sc_queue_mtx); 1784 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1785 sc->sc_provider->name); 1786 sc->sc_provider->flags |= G_PF_WITHER; 1787 g_orphan_provider(sc->sc_provider, ENXIO); 1788 sc->sc_provider = NULL; 1789 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1790 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1791 g_mirror_sync_stop(disk, 1); 1792 } 1793} 1794 1795static void 1796g_mirror_go(void *arg) 1797{ 1798 struct g_mirror_softc *sc; 1799 1800 sc = arg; 1801 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1802 g_mirror_event_send(sc, 0, 1803 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1804} 1805 1806static u_int 1807g_mirror_determine_state(struct g_mirror_disk *disk) 1808{ 1809 struct g_mirror_softc *sc; 1810 u_int state; 1811 1812 sc = disk->d_softc; 1813 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1814 if ((disk->d_flags & 1815 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1816 /* Disk does not need synchronization. */ 1817 state = G_MIRROR_DISK_STATE_ACTIVE; 1818 } else { 1819 if ((sc->sc_flags & 1820 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1821 (disk->d_flags & 1822 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1823 /* 1824 * We can start synchronization from 1825 * the stored offset. 1826 */ 1827 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1828 } else { 1829 state = G_MIRROR_DISK_STATE_STALE; 1830 } 1831 } 1832 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1833 /* 1834 * Reset all synchronization data for this disk, 1835 * because if it even was synchronized, it was 1836 * synchronized to disks with different syncid. 1837 */ 1838 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1839 disk->d_sync.ds_offset = 0; 1840 disk->d_sync.ds_offset_done = 0; 1841 disk->d_sync.ds_syncid = sc->sc_syncid; 1842 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1843 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1844 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1845 } else { 1846 state = G_MIRROR_DISK_STATE_STALE; 1847 } 1848 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1849 /* 1850 * Not good, NOT GOOD! 1851 * It means that mirror was started on stale disks 1852 * and more fresh disk just arrive. 1853 * If there were writes, mirror is fucked up, sorry. 1854 * I think the best choice here is don't touch 1855 * this disk and inform the user laudly. 1856 */ 1857 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1858 "disk (%s) arrives!! It will not be connected to the " 1859 "running device.", sc->sc_name, 1860 g_mirror_get_diskname(disk)); 1861 g_mirror_destroy_disk(disk); 1862 state = G_MIRROR_DISK_STATE_NONE; 1863 /* Return immediately, because disk was destroyed. */ 1864 return (state); 1865 } 1866 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1867 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1868 return (state); 1869} 1870 1871/* 1872 * Update device state. 1873 */ 1874static void 1875g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1876{ 1877 struct g_mirror_disk *disk; 1878 u_int state; 1879 1880 g_topology_assert(); 1881 1882 switch (sc->sc_state) { 1883 case G_MIRROR_DEVICE_STATE_STARTING: 1884 { 1885 struct g_mirror_disk *pdisk, *tdisk; 1886 u_int dirty, ndisks, genid, syncid; 1887 1888 KASSERT(sc->sc_provider == NULL, 1889 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1890 /* 1891 * Are we ready? We are, if all disks are connected or 1892 * if we have any disks and 'force' is true. 1893 */ 1894 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1895 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1896 ; 1897 } else if (g_mirror_ndisks(sc, -1) == 0) { 1898 /* 1899 * Disks went down in starting phase, so destroy 1900 * device. 1901 */ 1902 callout_drain(&sc->sc_callout); 1903 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1904 return; 1905 } else { 1906 return; 1907 } 1908 1909 /* 1910 * Activate all disks with the biggest syncid. 1911 */ 1912 if (force) { 1913 /* 1914 * If 'force' is true, we have been called due to 1915 * timeout, so don't bother canceling timeout. 1916 */ 1917 ndisks = 0; 1918 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1919 if ((disk->d_flags & 1920 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1921 ndisks++; 1922 } 1923 } 1924 if (ndisks == 0) { 1925 /* No valid disks found, destroy device. */ 1926 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1927 return; 1928 } 1929 } else { 1930 /* Cancel timeout. */ 1931 callout_drain(&sc->sc_callout); 1932 } 1933 1934 /* 1935 * Find the biggest genid. 1936 */ 1937 genid = 0; 1938 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1939 if (disk->d_genid > genid) 1940 genid = disk->d_genid; 1941 } 1942 sc->sc_genid = genid; 1943 /* 1944 * Remove all disks without the biggest genid. 1945 */ 1946 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 1947 if (disk->d_genid < genid) { 1948 G_MIRROR_DEBUG(0, 1949 "Component %s (device %s) broken, skipping.", 1950 g_mirror_get_diskname(disk), sc->sc_name); 1951 g_mirror_destroy_disk(disk); 1952 } 1953 } 1954 1955 /* 1956 * Find the biggest syncid. 1957 */ 1958 syncid = 0; 1959 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1960 if (disk->d_sync.ds_syncid > syncid) 1961 syncid = disk->d_sync.ds_syncid; 1962 } 1963 1964 /* 1965 * Here we need to look for dirty disks and if all disks 1966 * with the biggest syncid are dirty, we have to choose 1967 * one with the biggest priority and rebuild the rest. 1968 */ 1969 /* 1970 * Find the number of dirty disks with the biggest syncid. 1971 * Find the number of disks with the biggest syncid. 1972 * While here, find a disk with the biggest priority. 1973 */ 1974 dirty = ndisks = 0; 1975 pdisk = NULL; 1976 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1977 if (disk->d_sync.ds_syncid != syncid) 1978 continue; 1979 if ((disk->d_flags & 1980 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1981 continue; 1982 } 1983 ndisks++; 1984 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1985 dirty++; 1986 if (pdisk == NULL || 1987 pdisk->d_priority < disk->d_priority) { 1988 pdisk = disk; 1989 } 1990 } 1991 } 1992 if (dirty == 0) { 1993 /* No dirty disks at all, great. */ 1994 } else if (dirty == ndisks) { 1995 /* 1996 * Force synchronization for all dirty disks except one 1997 * with the biggest priority. 1998 */ 1999 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2000 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2001 "master disk for synchronization.", 2002 g_mirror_get_diskname(pdisk), sc->sc_name); 2003 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2004 if (disk->d_sync.ds_syncid != syncid) 2005 continue; 2006 if ((disk->d_flags & 2007 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2008 continue; 2009 } 2010 KASSERT((disk->d_flags & 2011 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2012 ("Disk %s isn't marked as dirty.", 2013 g_mirror_get_diskname(disk))); 2014 /* Skip the disk with the biggest priority. */ 2015 if (disk == pdisk) 2016 continue; 2017 disk->d_sync.ds_syncid = 0; 2018 } 2019 } else if (dirty < ndisks) { 2020 /* 2021 * Force synchronization for all dirty disks. 2022 * We have some non-dirty disks. 2023 */ 2024 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2025 if (disk->d_sync.ds_syncid != syncid) 2026 continue; 2027 if ((disk->d_flags & 2028 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2029 continue; 2030 } 2031 if ((disk->d_flags & 2032 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2033 continue; 2034 } 2035 disk->d_sync.ds_syncid = 0; 2036 } 2037 } 2038 2039 /* Reset hint. */ 2040 sc->sc_hint = NULL; 2041 sc->sc_syncid = syncid; 2042 if (force) { 2043 /* Remember to bump syncid on first write. */ 2044 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_OFW; 2045 } 2046 state = G_MIRROR_DEVICE_STATE_RUNNING; 2047 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2048 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2049 g_mirror_device_state2str(state)); 2050 sc->sc_state = state; 2051 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2052 state = g_mirror_determine_state(disk); 2053 g_mirror_event_send(disk, state, 2054 G_MIRROR_EVENT_DONTWAIT); 2055 if (state == G_MIRROR_DISK_STATE_STALE) 2056 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_OFW; 2057 } 2058 wakeup(&g_mirror_class); 2059 break; 2060 } 2061 case G_MIRROR_DEVICE_STATE_RUNNING: 2062 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2063 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2064 /* 2065 * No active disks or no disks at all, 2066 * so destroy device. 2067 */ 2068 if (sc->sc_provider != NULL) 2069 g_mirror_destroy_provider(sc); 2070 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2071 break; 2072 } else if (g_mirror_ndisks(sc, 2073 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2074 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2075 /* 2076 * We have active disks, launch provider if it doesn't 2077 * exist. 2078 */ 2079 if (sc->sc_provider == NULL) 2080 g_mirror_launch_provider(sc); 2081 } 2082 /* 2083 * Bump syncid here, if we need to do it immediately. 2084 */ 2085 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_IMM) != 0) { 2086 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2087 g_mirror_bump_syncid(sc); 2088 } 2089 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID_IMM) != 0) { 2090 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2091 g_mirror_bump_genid(sc); 2092 } 2093 break; 2094 default: 2095 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2096 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2097 break; 2098 } 2099} 2100 2101/* 2102 * Update disk state and device state if needed. 2103 */ 2104#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2105 "Disk %s state changed from %s to %s (device %s).", \ 2106 g_mirror_get_diskname(disk), \ 2107 g_mirror_disk_state2str(disk->d_state), \ 2108 g_mirror_disk_state2str(state), sc->sc_name) 2109static int 2110g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2111{ 2112 struct g_mirror_softc *sc; 2113 2114 g_topology_assert(); 2115 2116 sc = disk->d_softc; 2117again: 2118 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2119 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2120 g_mirror_disk_state2str(state)); 2121 switch (state) { 2122 case G_MIRROR_DISK_STATE_NEW: 2123 /* 2124 * Possible scenarios: 2125 * 1. New disk arrive. 2126 */ 2127 /* Previous state should be NONE. */ 2128 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2129 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2130 g_mirror_disk_state2str(disk->d_state))); 2131 DISK_STATE_CHANGED(); 2132 2133 disk->d_state = state; 2134 if (LIST_EMPTY(&sc->sc_disks)) 2135 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2136 else { 2137 struct g_mirror_disk *dp; 2138 2139 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2140 if (disk->d_priority >= dp->d_priority) { 2141 LIST_INSERT_BEFORE(dp, disk, d_next); 2142 dp = NULL; 2143 break; 2144 } 2145 if (LIST_NEXT(dp, d_next) == NULL) 2146 break; 2147 } 2148 if (dp != NULL) 2149 LIST_INSERT_AFTER(dp, disk, d_next); 2150 } 2151 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2152 sc->sc_name, g_mirror_get_diskname(disk)); 2153 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2154 break; 2155 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2156 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2157 g_mirror_device_state2str(sc->sc_state), 2158 g_mirror_get_diskname(disk), 2159 g_mirror_disk_state2str(disk->d_state))); 2160 state = g_mirror_determine_state(disk); 2161 if (state != G_MIRROR_DISK_STATE_NONE) 2162 goto again; 2163 break; 2164 case G_MIRROR_DISK_STATE_ACTIVE: 2165 /* 2166 * Possible scenarios: 2167 * 1. New disk does not need synchronization. 2168 * 2. Synchronization process finished successfully. 2169 */ 2170 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2171 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2172 g_mirror_device_state2str(sc->sc_state), 2173 g_mirror_get_diskname(disk), 2174 g_mirror_disk_state2str(disk->d_state))); 2175 /* Previous state should be NEW or SYNCHRONIZING. */ 2176 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2177 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2178 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2179 g_mirror_disk_state2str(disk->d_state))); 2180 DISK_STATE_CHANGED(); 2181 2182 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2183 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2184 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2185 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2186 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2187 g_mirror_sync_stop(disk, 0); 2188 } 2189 disk->d_state = state; 2190 disk->d_sync.ds_offset = 0; 2191 disk->d_sync.ds_offset_done = 0; 2192 g_mirror_update_access(disk); 2193 g_mirror_update_metadata(disk); 2194 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2195 sc->sc_name, g_mirror_get_diskname(disk)); 2196 break; 2197 case G_MIRROR_DISK_STATE_STALE: 2198 /* 2199 * Possible scenarios: 2200 * 1. Stale disk was connected. 2201 */ 2202 /* Previous state should be NEW. */ 2203 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2204 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2205 g_mirror_disk_state2str(disk->d_state))); 2206 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2207 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2208 g_mirror_device_state2str(sc->sc_state), 2209 g_mirror_get_diskname(disk), 2210 g_mirror_disk_state2str(disk->d_state))); 2211 /* 2212 * STALE state is only possible if device is marked 2213 * NOAUTOSYNC. 2214 */ 2215 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2216 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2217 g_mirror_device_state2str(sc->sc_state), 2218 g_mirror_get_diskname(disk), 2219 g_mirror_disk_state2str(disk->d_state))); 2220 DISK_STATE_CHANGED(); 2221 2222 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2223 disk->d_state = state; 2224 g_mirror_update_metadata(disk); 2225 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2226 sc->sc_name, g_mirror_get_diskname(disk)); 2227 break; 2228 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2229 /* 2230 * Possible scenarios: 2231 * 1. Disk which needs synchronization was connected. 2232 */ 2233 /* Previous state should be NEW. */ 2234 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2235 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2236 g_mirror_disk_state2str(disk->d_state))); 2237 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2238 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2239 g_mirror_device_state2str(sc->sc_state), 2240 g_mirror_get_diskname(disk), 2241 g_mirror_disk_state2str(disk->d_state))); 2242 DISK_STATE_CHANGED(); 2243 2244 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2245 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2246 disk->d_state = state; 2247 if (sc->sc_provider != NULL) { 2248 g_mirror_sync_start(disk); 2249 g_mirror_update_metadata(disk); 2250 } 2251 break; 2252 case G_MIRROR_DISK_STATE_DISCONNECTED: 2253 /* 2254 * Possible scenarios: 2255 * 1. Device wasn't running yet, but disk disappear. 2256 * 2. Disk was active and disapppear. 2257 * 3. Disk disappear during synchronization process. 2258 */ 2259 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2260 /* 2261 * Previous state should be ACTIVE, STALE or 2262 * SYNCHRONIZING. 2263 */ 2264 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2265 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2266 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2267 ("Wrong disk state (%s, %s).", 2268 g_mirror_get_diskname(disk), 2269 g_mirror_disk_state2str(disk->d_state))); 2270 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2271 /* Previous state should be NEW. */ 2272 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2273 ("Wrong disk state (%s, %s).", 2274 g_mirror_get_diskname(disk), 2275 g_mirror_disk_state2str(disk->d_state))); 2276 /* 2277 * Reset bumping syncid if disk disappeared in STARTING 2278 * state. 2279 */ 2280 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_OFW) != 0) 2281 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2282#ifdef INVARIANTS 2283 } else { 2284 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2285 sc->sc_name, 2286 g_mirror_device_state2str(sc->sc_state), 2287 g_mirror_get_diskname(disk), 2288 g_mirror_disk_state2str(disk->d_state))); 2289#endif 2290 } 2291 DISK_STATE_CHANGED(); 2292 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2293 sc->sc_name, g_mirror_get_diskname(disk)); 2294 2295 g_mirror_destroy_disk(disk); 2296 break; 2297 case G_MIRROR_DISK_STATE_DESTROY: 2298 { 2299 int error; 2300 2301 error = g_mirror_clear_metadata(disk); 2302 if (error != 0) 2303 return (error); 2304 DISK_STATE_CHANGED(); 2305 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2306 sc->sc_name, g_mirror_get_diskname(disk)); 2307 2308 g_mirror_destroy_disk(disk); 2309 sc->sc_ndisks--; 2310 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2311 g_mirror_update_metadata(disk); 2312 } 2313 break; 2314 } 2315 default: 2316 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2317 break; 2318 } 2319 return (0); 2320} 2321#undef DISK_STATE_CHANGED 2322 2323int 2324g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2325{ 2326 struct g_provider *pp; 2327 u_char *buf; 2328 int error; 2329 2330 g_topology_assert(); 2331 2332 error = g_access(cp, 1, 0, 0); 2333 if (error != 0) 2334 return (error); 2335 pp = cp->provider; 2336 g_topology_unlock(); 2337 /* Metadata are stored on last sector. */ 2338 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2339 &error); 2340 g_topology_lock(); 2341 g_access(cp, -1, 0, 0); 2342 if (error != 0) { 2343 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2344 cp->provider->name, error); 2345 if (buf != NULL) 2346 g_free(buf); 2347 return (error); 2348 } 2349 2350 /* Decode metadata. */ 2351 error = mirror_metadata_decode(buf, md); 2352 g_free(buf); 2353 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2354 return (EINVAL); 2355 if (md->md_version > G_MIRROR_VERSION) { 2356 G_MIRROR_DEBUG(0, 2357 "Kernel module is too old to handle metadata from %s.", 2358 cp->provider->name); 2359 return (EINVAL); 2360 } 2361 if (error != 0) { 2362 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2363 cp->provider->name); 2364 return (error); 2365 } 2366 2367 return (0); 2368} 2369 2370static int 2371g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2372 struct g_mirror_metadata *md) 2373{ 2374 2375 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2376 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2377 pp->name, md->md_did); 2378 return (EEXIST); 2379 } 2380 if (md->md_all != sc->sc_ndisks) { 2381 G_MIRROR_DEBUG(1, 2382 "Invalid '%s' field on disk %s (device %s), skipping.", 2383 "md_all", pp->name, sc->sc_name); 2384 return (EINVAL); 2385 } 2386 if (md->md_slice != sc->sc_slice) { 2387 G_MIRROR_DEBUG(1, 2388 "Invalid '%s' field on disk %s (device %s), skipping.", 2389 "md_slice", pp->name, sc->sc_name); 2390 return (EINVAL); 2391 } 2392 if (md->md_balance != sc->sc_balance) { 2393 G_MIRROR_DEBUG(1, 2394 "Invalid '%s' field on disk %s (device %s), skipping.", 2395 "md_balance", pp->name, sc->sc_name); 2396 return (EINVAL); 2397 } 2398 if (md->md_mediasize != sc->sc_mediasize) { 2399 G_MIRROR_DEBUG(1, 2400 "Invalid '%s' field on disk %s (device %s), skipping.", 2401 "md_mediasize", pp->name, sc->sc_name); 2402 return (EINVAL); 2403 } 2404 if (sc->sc_mediasize > pp->mediasize) { 2405 G_MIRROR_DEBUG(1, 2406 "Invalid size of disk %s (device %s), skipping.", pp->name, 2407 sc->sc_name); 2408 return (EINVAL); 2409 } 2410 if (md->md_sectorsize != sc->sc_sectorsize) { 2411 G_MIRROR_DEBUG(1, 2412 "Invalid '%s' field on disk %s (device %s), skipping.", 2413 "md_sectorsize", pp->name, sc->sc_name); 2414 return (EINVAL); 2415 } 2416 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2417 G_MIRROR_DEBUG(1, 2418 "Invalid sector size of disk %s (device %s), skipping.", 2419 pp->name, sc->sc_name); 2420 return (EINVAL); 2421 } 2422 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2423 G_MIRROR_DEBUG(1, 2424 "Invalid device flags on disk %s (device %s), skipping.", 2425 pp->name, sc->sc_name); 2426 return (EINVAL); 2427 } 2428 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2429 G_MIRROR_DEBUG(1, 2430 "Invalid disk flags on disk %s (device %s), skipping.", 2431 pp->name, sc->sc_name); 2432 return (EINVAL); 2433 } 2434 return (0); 2435} 2436 2437int 2438g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2439 struct g_mirror_metadata *md) 2440{ 2441 struct g_mirror_disk *disk; 2442 int error; 2443 2444 g_topology_assert(); 2445 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2446 2447 error = g_mirror_check_metadata(sc, pp, md); 2448 if (error != 0) 2449 return (error); 2450 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2451 md->md_genid < sc->sc_genid) { 2452 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2453 pp->name, sc->sc_name); 2454 return (EINVAL); 2455 } 2456 disk = g_mirror_init_disk(sc, pp, md, &error); 2457 if (disk == NULL) 2458 return (error); 2459 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2460 G_MIRROR_EVENT_WAIT); 2461 if (error != 0) 2462 return (error); 2463 if (md->md_version < G_MIRROR_VERSION) { 2464 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2465 pp->name, md->md_version, G_MIRROR_VERSION); 2466 g_mirror_update_metadata(disk); 2467 } 2468 return (0); 2469} 2470 2471static int 2472g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2473{ 2474 struct g_mirror_softc *sc; 2475 struct g_mirror_disk *disk; 2476 int dcr, dcw, dce; 2477 2478 g_topology_assert(); 2479 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2480 acw, ace); 2481 2482 dcr = pp->acr + acr; 2483 dcw = pp->acw + acw; 2484 dce = pp->ace + ace; 2485 2486 sc = pp->geom->softc; 2487 if (sc == NULL || LIST_EMPTY(&sc->sc_disks) || 2488 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 2489 if (acr <= 0 && acw <= 0 && ace <= 0) 2490 return (0); 2491 else 2492 return (ENXIO); 2493 } 2494 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2495 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2496 continue; 2497 /* 2498 * Mark disk as dirty on open and unmark on close. 2499 */ 2500 if (pp->acw == 0 && dcw > 0) { 2501 G_MIRROR_DEBUG(1, 2502 "Disk %s (device %s) marked as dirty.", 2503 g_mirror_get_diskname(disk), sc->sc_name); 2504 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2505 g_mirror_update_metadata(disk); 2506 } else if (pp->acw > 0 && dcw == 0) { 2507 G_MIRROR_DEBUG(1, 2508 "Disk %s (device %s) marked as clean.", 2509 g_mirror_get_diskname(disk), sc->sc_name); 2510 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2511 g_mirror_update_metadata(disk); 2512 } 2513 } 2514 return (0); 2515} 2516 2517static struct g_geom * 2518g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2519{ 2520 struct g_mirror_softc *sc; 2521 struct g_geom *gp; 2522 int error, timeout; 2523 2524 g_topology_assert(); 2525 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2526 md->md_mid); 2527 2528 /* One disk is minimum. */ 2529 if (md->md_all < 1) 2530 return (NULL); 2531 /* 2532 * Action geom. 2533 */ 2534 gp = g_new_geomf(mp, "%s", md->md_name); 2535 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2536 gp->start = g_mirror_start; 2537 gp->spoiled = g_mirror_spoiled; 2538 gp->orphan = g_mirror_orphan; 2539 gp->access = g_mirror_access; 2540 gp->dumpconf = g_mirror_dumpconf; 2541 2542 sc->sc_id = md->md_mid; 2543 sc->sc_slice = md->md_slice; 2544 sc->sc_balance = md->md_balance; 2545 sc->sc_mediasize = md->md_mediasize; 2546 sc->sc_sectorsize = md->md_sectorsize; 2547 sc->sc_ndisks = md->md_all; 2548 sc->sc_flags = md->md_mflags; 2549 sc->sc_bump_id = 0; 2550 sc->sc_idle = 0; 2551 bioq_init(&sc->sc_queue); 2552 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2553 LIST_INIT(&sc->sc_disks); 2554 TAILQ_INIT(&sc->sc_events); 2555 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2556 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2557 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2558 gp->softc = sc; 2559 sc->sc_geom = gp; 2560 sc->sc_provider = NULL; 2561 /* 2562 * Synchronization geom. 2563 */ 2564 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2565 gp->softc = sc; 2566 gp->orphan = g_mirror_orphan; 2567 sc->sc_sync.ds_geom = gp; 2568 sc->sc_sync.ds_ndisks = 0; 2569 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2570 "g_mirror %s", md->md_name); 2571 if (error != 0) { 2572 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2573 sc->sc_name); 2574 g_destroy_geom(sc->sc_sync.ds_geom); 2575 mtx_destroy(&sc->sc_events_mtx); 2576 mtx_destroy(&sc->sc_queue_mtx); 2577 g_destroy_geom(sc->sc_geom); 2578 free(sc, M_MIRROR); 2579 return (NULL); 2580 } 2581 2582 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2583 2584 /* 2585 * Run timeout. 2586 */ 2587 timeout = g_mirror_timeout * hz; 2588 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2589 return (sc->sc_geom); 2590} 2591 2592int 2593g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2594{ 2595 struct g_provider *pp; 2596 2597 g_topology_assert(); 2598 2599 if (sc == NULL) 2600 return (ENXIO); 2601 pp = sc->sc_provider; 2602 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2603 if (force) { 2604 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 2605 "can't be definitely removed.", pp->name); 2606 } else { 2607 G_MIRROR_DEBUG(1, 2608 "Device %s is still open (r%dw%de%d).", pp->name, 2609 pp->acr, pp->acw, pp->ace); 2610 return (EBUSY); 2611 } 2612 } 2613 2614 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2615 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2616 g_topology_unlock(); 2617 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2618 mtx_lock(&sc->sc_queue_mtx); 2619 wakeup(sc); 2620 mtx_unlock(&sc->sc_queue_mtx); 2621 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2622 while (sc->sc_worker != NULL) 2623 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2624 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2625 g_topology_lock(); 2626 g_mirror_destroy_device(sc); 2627 free(sc, M_MIRROR); 2628 return (0); 2629} 2630 2631static void 2632g_mirror_taste_orphan(struct g_consumer *cp) 2633{ 2634 2635 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2636 cp->provider->name)); 2637} 2638 2639static struct g_geom * 2640g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2641{ 2642 struct g_mirror_metadata md; 2643 struct g_mirror_softc *sc; 2644 struct g_consumer *cp; 2645 struct g_geom *gp; 2646 int error; 2647 2648 g_topology_assert(); 2649 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2650 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2651 2652 gp = g_new_geomf(mp, "mirror:taste"); 2653 /* 2654 * This orphan function should be never called. 2655 */ 2656 gp->orphan = g_mirror_taste_orphan; 2657 cp = g_new_consumer(gp); 2658 g_attach(cp, pp); 2659 error = g_mirror_read_metadata(cp, &md); 2660 g_detach(cp); 2661 g_destroy_consumer(cp); 2662 g_destroy_geom(gp); 2663 if (error != 0) 2664 return (NULL); 2665 gp = NULL; 2666 2667 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2668 return (NULL); 2669 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2670 G_MIRROR_DEBUG(0, 2671 "Device %s: provider %s marked as inactive, skipping.", 2672 md.md_name, pp->name); 2673 return (NULL); 2674 } 2675 if (g_mirror_debug >= 2) 2676 mirror_metadata_dump(&md); 2677 2678 /* 2679 * Let's check if device already exists. 2680 */ 2681 sc = NULL; 2682 LIST_FOREACH(gp, &mp->geom, geom) { 2683 sc = gp->softc; 2684 if (sc == NULL) 2685 continue; 2686 if (sc->sc_sync.ds_geom == gp) 2687 continue; 2688 if (strcmp(md.md_name, sc->sc_name) != 0) 2689 continue; 2690 if (md.md_mid != sc->sc_id) { 2691 G_MIRROR_DEBUG(0, "Device %s already configured.", 2692 sc->sc_name); 2693 return (NULL); 2694 } 2695 break; 2696 } 2697 if (gp == NULL) { 2698 gp = g_mirror_create(mp, &md); 2699 if (gp == NULL) { 2700 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2701 md.md_name); 2702 return (NULL); 2703 } 2704 sc = gp->softc; 2705 } 2706 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2707 error = g_mirror_add_disk(sc, pp, &md); 2708 if (error != 0) { 2709 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2710 pp->name, gp->name, error); 2711 if (LIST_EMPTY(&sc->sc_disks)) 2712 g_mirror_destroy(sc, 1); 2713 return (NULL); 2714 } 2715 return (gp); 2716} 2717 2718static int 2719g_mirror_destroy_geom(struct gctl_req *req __unused, 2720 struct g_class *mp __unused, struct g_geom *gp) 2721{ 2722 2723 return (g_mirror_destroy(gp->softc, 0)); 2724} 2725 2726static void 2727g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2728 struct g_consumer *cp, struct g_provider *pp) 2729{ 2730 struct g_mirror_softc *sc; 2731 2732 g_topology_assert(); 2733 2734 sc = gp->softc; 2735 if (sc == NULL) 2736 return; 2737 /* Skip synchronization geom. */ 2738 if (gp == sc->sc_sync.ds_geom) 2739 return; 2740 if (pp != NULL) { 2741 /* Nothing here. */ 2742 } else if (cp != NULL) { 2743 struct g_mirror_disk *disk; 2744 2745 disk = cp->private; 2746 if (disk == NULL) 2747 return; 2748 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2749 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2750 sbuf_printf(sb, "%s<Synchronized>", indent); 2751 if (disk->d_sync.ds_offset_done == 0) 2752 sbuf_printf(sb, "0%%"); 2753 else { 2754 sbuf_printf(sb, "%u%%", 2755 (u_int)((disk->d_sync.ds_offset_done * 100) / 2756 sc->sc_provider->mediasize)); 2757 } 2758 sbuf_printf(sb, "</Synchronized>\n"); 2759 } 2760 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2761 disk->d_sync.ds_syncid); 2762 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 2763 disk->d_genid); 2764 sbuf_printf(sb, "%s<Flags>", indent); 2765 if (disk->d_flags == 0) 2766 sbuf_printf(sb, "NONE"); 2767 else { 2768 int first = 1; 2769 2770#define ADD_FLAG(flag, name) do { \ 2771 if ((disk->d_flags & (flag)) != 0) { \ 2772 if (!first) \ 2773 sbuf_printf(sb, ", "); \ 2774 else \ 2775 first = 0; \ 2776 sbuf_printf(sb, name); \ 2777 } \ 2778} while (0) 2779 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2780 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2781 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2782 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2783 "SYNCHRONIZING"); 2784 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2785#undef ADD_FLAG 2786 } 2787 sbuf_printf(sb, "</Flags>\n"); 2788 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2789 disk->d_priority); 2790 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2791 g_mirror_disk_state2str(disk->d_state)); 2792 } else { 2793 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2794 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2795 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 2796 sbuf_printf(sb, "%s<Flags>", indent); 2797 if (sc->sc_flags == 0) 2798 sbuf_printf(sb, "NONE"); 2799 else { 2800 int first = 1; 2801 2802#define ADD_FLAG(flag, name) do { \ 2803 if ((sc->sc_flags & (flag)) != 0) { \ 2804 if (!first) \ 2805 sbuf_printf(sb, ", "); \ 2806 else \ 2807 first = 0; \ 2808 sbuf_printf(sb, name); \ 2809 } \ 2810} while (0) 2811 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2812#undef ADD_FLAG 2813 } 2814 sbuf_printf(sb, "</Flags>\n"); 2815 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2816 (u_int)sc->sc_slice); 2817 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2818 balance_name(sc->sc_balance)); 2819 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2820 sc->sc_ndisks); 2821 sbuf_printf(sb, "%s<State>", indent); 2822 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2823 sbuf_printf(sb, "%s", "STARTING"); 2824 else if (sc->sc_ndisks == 2825 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 2826 sbuf_printf(sb, "%s", "COMPLETE"); 2827 else 2828 sbuf_printf(sb, "%s", "DEGRADED"); 2829 sbuf_printf(sb, "</State>\n"); 2830 } 2831} 2832 2833static void 2834g_mirror_shutdown(void *arg, int howto) 2835{ 2836 struct g_class *mp; 2837 struct g_geom *gp, *gp2; 2838 2839 mp = arg; 2840 DROP_GIANT(); 2841 g_topology_lock(); 2842 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2843 if (gp->softc == NULL) 2844 continue; 2845 g_mirror_destroy(gp->softc, 1); 2846 } 2847 g_topology_unlock(); 2848 PICKUP_GIANT(); 2849#if 0 2850 tsleep(&gp, PRIBIO, "m:shutdown", hz * 20); 2851#endif 2852} 2853 2854static void 2855g_mirror_init(struct g_class *mp) 2856{ 2857 2858 g_mirror_ehtag = EVENTHANDLER_REGISTER(shutdown_post_sync, 2859 g_mirror_shutdown, mp, SHUTDOWN_PRI_FIRST); 2860 if (g_mirror_ehtag == NULL) 2861 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 2862} 2863 2864static void 2865g_mirror_fini(struct g_class *mp) 2866{ 2867 2868 if (g_mirror_ehtag == NULL) 2869 return; 2870 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_ehtag); 2871} 2872 2873static int 2874g_mirror_can_go(void) 2875{ 2876 struct g_mirror_softc *sc; 2877 struct g_geom *gp; 2878 struct g_provider *pp; 2879 int can_go; 2880 2881 DROP_GIANT(); 2882 can_go = 1; 2883 g_topology_lock(); 2884 LIST_FOREACH(gp, &g_mirror_class.geom, geom) { 2885 sc = gp->softc; 2886 if (sc == NULL) { 2887 can_go = 0; 2888 break; 2889 } 2890 pp = sc->sc_provider; 2891 if (pp == NULL || pp->error != 0) { 2892 can_go = 0; 2893 break; 2894 } 2895 } 2896 g_topology_unlock(); 2897 PICKUP_GIANT(); 2898 return (can_go); 2899} 2900 2901static void 2902g_mirror_rootwait(void) 2903{ 2904 2905 /* 2906 * HACK: Wait for GEOM, because g_mirror_rootwait() can be called, 2907 * HACK: before we get providers for tasting. 2908 */ 2909 tsleep(&g_mirror_class, PRIBIO, "mroot", hz * 3); 2910 /* 2911 * Wait for mirrors in degraded state. 2912 */ 2913 for (;;) { 2914 if (g_mirror_can_go()) 2915 break; 2916 tsleep(&g_mirror_class, PRIBIO, "mroot", hz); 2917 } 2918} 2919 2920SYSINIT(g_mirror_root, SI_SUB_RAID, SI_ORDER_FIRST, g_mirror_rootwait, NULL) 2921 2922DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2923