g_mirror.c revision 139140
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 139140 2004-12-21 18:42:51Z pjd $"); 29 30#include <sys/param.h> 31#include <sys/systm.h> 32#include <sys/kernel.h> 33#include <sys/module.h> 34#include <sys/limits.h> 35#include <sys/lock.h> 36#include <sys/mutex.h> 37#include <sys/bio.h> 38#include <sys/sysctl.h> 39#include <sys/malloc.h> 40#include <sys/eventhandler.h> 41#include <vm/uma.h> 42#include <geom/geom.h> 43#include <sys/proc.h> 44#include <sys/kthread.h> 45#include <geom/mirror/g_mirror.h> 46 47 48static MALLOC_DEFINE(M_MIRROR, "mirror data", "GEOM_MIRROR Data"); 49 50SYSCTL_DECL(_kern_geom); 51SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, "GEOM_MIRROR stuff"); 52u_int g_mirror_debug = 0; 53TUNABLE_INT("kern.geom.mirror.debug", &g_mirror_debug); 54SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RW, &g_mirror_debug, 0, 55 "Debug level"); 56static u_int g_mirror_timeout = 4; 57TUNABLE_INT("kern.geom.mirror.timeout", &g_mirror_timeout); 58SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RW, &g_mirror_timeout, 59 0, "Time to wait on all mirror components"); 60static u_int g_mirror_idletime = 5; 61TUNABLE_INT("kern.geom.mirror.idletime", &g_mirror_idletime); 62SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RW, 63 &g_mirror_idletime, 0, "Mark components as clean when idling"); 64static u_int g_mirror_reqs_per_sync = 5; 65SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, reqs_per_sync, CTLFLAG_RW, 66 &g_mirror_reqs_per_sync, 0, 67 "Number of regular I/O requests per synchronization request"); 68static u_int g_mirror_syncs_per_sec = 100; 69SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, syncs_per_sec, CTLFLAG_RW, 70 &g_mirror_syncs_per_sec, 0, 71 "Number of synchronizations requests per second"); 72 73#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 74 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 75 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 76 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 77} while (0) 78 79static eventhandler_tag g_mirror_ehtag = NULL; 80 81static int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp, 82 struct g_geom *gp); 83static g_taste_t g_mirror_taste; 84static void g_mirror_init(struct g_class *mp); 85static void g_mirror_fini(struct g_class *mp); 86 87struct g_class g_mirror_class = { 88 .name = G_MIRROR_CLASS_NAME, 89 .version = G_VERSION, 90 .ctlreq = g_mirror_config, 91 .taste = g_mirror_taste, 92 .destroy_geom = g_mirror_destroy_geom, 93 .init = g_mirror_init, 94 .fini = g_mirror_fini 95}; 96 97 98static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 99static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 100static void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force); 101static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 102 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 103static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 104 105 106static const char * 107g_mirror_disk_state2str(int state) 108{ 109 110 switch (state) { 111 case G_MIRROR_DISK_STATE_NONE: 112 return ("NONE"); 113 case G_MIRROR_DISK_STATE_NEW: 114 return ("NEW"); 115 case G_MIRROR_DISK_STATE_ACTIVE: 116 return ("ACTIVE"); 117 case G_MIRROR_DISK_STATE_STALE: 118 return ("STALE"); 119 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 120 return ("SYNCHRONIZING"); 121 case G_MIRROR_DISK_STATE_DISCONNECTED: 122 return ("DISCONNECTED"); 123 case G_MIRROR_DISK_STATE_DESTROY: 124 return ("DESTROY"); 125 default: 126 return ("INVALID"); 127 } 128} 129 130static const char * 131g_mirror_device_state2str(int state) 132{ 133 134 switch (state) { 135 case G_MIRROR_DEVICE_STATE_STARTING: 136 return ("STARTING"); 137 case G_MIRROR_DEVICE_STATE_RUNNING: 138 return ("RUNNING"); 139 default: 140 return ("INVALID"); 141 } 142} 143 144static const char * 145g_mirror_get_diskname(struct g_mirror_disk *disk) 146{ 147 148 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 149 return ("[unknown]"); 150 return (disk->d_name); 151} 152 153/* 154 * --- Events handling functions --- 155 * Events in geom_mirror are used to maintain disks and device status 156 * from one thread to simplify locking. 157 */ 158static void 159g_mirror_event_free(struct g_mirror_event *ep) 160{ 161 162 free(ep, M_MIRROR); 163} 164 165int 166g_mirror_event_send(void *arg, int state, int flags) 167{ 168 struct g_mirror_softc *sc; 169 struct g_mirror_disk *disk; 170 struct g_mirror_event *ep; 171 int error; 172 173 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 174 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 175 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 176 disk = NULL; 177 sc = arg; 178 } else { 179 disk = arg; 180 sc = disk->d_softc; 181 } 182 ep->e_disk = disk; 183 ep->e_state = state; 184 ep->e_flags = flags; 185 ep->e_error = 0; 186 mtx_lock(&sc->sc_events_mtx); 187 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 188 mtx_unlock(&sc->sc_events_mtx); 189 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 190 mtx_lock(&sc->sc_queue_mtx); 191 wakeup(sc); 192 mtx_unlock(&sc->sc_queue_mtx); 193 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 194 return (0); 195 g_topology_assert(); 196 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 197 g_topology_unlock(); 198 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 199 mtx_lock(&sc->sc_events_mtx); 200 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 201 hz * 5); 202 } 203 /* Don't even try to use 'sc' here, because it could be already dead. */ 204 g_topology_lock(); 205 error = ep->e_error; 206 g_mirror_event_free(ep); 207 return (error); 208} 209 210static struct g_mirror_event * 211g_mirror_event_get(struct g_mirror_softc *sc) 212{ 213 struct g_mirror_event *ep; 214 215 mtx_lock(&sc->sc_events_mtx); 216 ep = TAILQ_FIRST(&sc->sc_events); 217 mtx_unlock(&sc->sc_events_mtx); 218 return (ep); 219} 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_sync.ds_syncid = md->md_syncid; 449 if (errorp != NULL) 450 *errorp = 0; 451 return (disk); 452fail: 453 if (errorp != NULL) 454 *errorp = error; 455 if (disk != NULL) { 456 g_mirror_disconnect_consumer(sc, disk->d_consumer); 457 free(disk, M_MIRROR); 458 } 459 return (NULL); 460} 461 462static void 463g_mirror_destroy_disk(struct g_mirror_disk *disk) 464{ 465 struct g_mirror_softc *sc; 466 467 g_topology_assert(); 468 469 LIST_REMOVE(disk, d_next); 470 g_mirror_event_cancel(disk); 471 sc = disk->d_softc; 472 if (sc->sc_hint == disk) 473 sc->sc_hint = NULL; 474 switch (disk->d_state) { 475 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 476 g_mirror_sync_stop(disk, 1); 477 /* FALLTHROUGH */ 478 case G_MIRROR_DISK_STATE_NEW: 479 case G_MIRROR_DISK_STATE_STALE: 480 case G_MIRROR_DISK_STATE_ACTIVE: 481 g_mirror_disconnect_consumer(sc, disk->d_consumer); 482 free(disk, M_MIRROR); 483 break; 484 default: 485 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 486 g_mirror_get_diskname(disk), 487 g_mirror_disk_state2str(disk->d_state))); 488 } 489} 490 491static void 492g_mirror_destroy_device(struct g_mirror_softc *sc) 493{ 494 struct g_mirror_disk *disk; 495 struct g_mirror_event *ep; 496 struct g_geom *gp; 497 struct g_consumer *cp, *tmpcp; 498 499 g_topology_assert(); 500 501 gp = sc->sc_geom; 502 if (sc->sc_provider != NULL) 503 g_mirror_destroy_provider(sc); 504 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 505 disk = LIST_FIRST(&sc->sc_disks)) { 506 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 507 g_mirror_update_metadata(disk); 508 g_mirror_destroy_disk(disk); 509 } 510 while ((ep = g_mirror_event_get(sc)) != NULL) { 511 g_mirror_event_remove(sc, ep); 512 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 513 g_mirror_event_free(ep); 514 else { 515 ep->e_error = ECANCELED; 516 ep->e_flags |= G_MIRROR_EVENT_DONE; 517 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 518 mtx_lock(&sc->sc_events_mtx); 519 wakeup(ep); 520 mtx_unlock(&sc->sc_events_mtx); 521 } 522 } 523 callout_drain(&sc->sc_callout); 524 gp->softc = NULL; 525 526 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 527 g_mirror_disconnect_consumer(sc, cp); 528 } 529 sc->sc_sync.ds_geom->softc = NULL; 530 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 531 mtx_destroy(&sc->sc_queue_mtx); 532 mtx_destroy(&sc->sc_events_mtx); 533 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 534 g_wither_geom(gp, ENXIO); 535} 536 537static void 538g_mirror_orphan(struct g_consumer *cp) 539{ 540 struct g_mirror_disk *disk; 541 542 g_topology_assert(); 543 544 disk = cp->private; 545 if (disk == NULL) 546 return; 547 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 548 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 549 G_MIRROR_EVENT_DONTWAIT); 550} 551 552static void 553g_mirror_spoiled(struct g_consumer *cp) 554{ 555 struct g_mirror_disk *disk; 556 557 g_topology_assert(); 558 559 disk = cp->private; 560 if (disk == NULL) 561 return; 562 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 563 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 564 G_MIRROR_EVENT_DONTWAIT); 565} 566 567/* 568 * Function should return the next active disk on the list. 569 * It is possible that it will be the same disk as given. 570 * If there are no active disks on list, NULL is returned. 571 */ 572static __inline struct g_mirror_disk * 573g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 574{ 575 struct g_mirror_disk *dp; 576 577 for (dp = LIST_NEXT(disk, d_next); dp != disk; 578 dp = LIST_NEXT(dp, d_next)) { 579 if (dp == NULL) 580 dp = LIST_FIRST(&sc->sc_disks); 581 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 582 break; 583 } 584 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 585 return (NULL); 586 return (dp); 587} 588 589static struct g_mirror_disk * 590g_mirror_get_disk(struct g_mirror_softc *sc) 591{ 592 struct g_mirror_disk *disk; 593 594 if (sc->sc_hint == NULL) { 595 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 596 if (sc->sc_hint == NULL) 597 return (NULL); 598 } 599 disk = sc->sc_hint; 600 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 601 disk = g_mirror_find_next(sc, disk); 602 if (disk == NULL) 603 return (NULL); 604 } 605 sc->sc_hint = g_mirror_find_next(sc, disk); 606 return (disk); 607} 608 609static int 610g_mirror_write_metadata(struct g_mirror_disk *disk, 611 struct g_mirror_metadata *md) 612{ 613 struct g_mirror_softc *sc; 614 struct g_consumer *cp; 615 off_t offset, length; 616 u_char *sector; 617 int error = 0; 618 619 g_topology_assert(); 620 621 sc = disk->d_softc; 622 cp = disk->d_consumer; 623 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 624 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 625 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 626 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 627 cp->acw, cp->ace)); 628 length = cp->provider->sectorsize; 629 offset = cp->provider->mediasize - length; 630 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 631 if (md != NULL) 632 mirror_metadata_encode(md, sector); 633 g_topology_unlock(); 634 error = g_write_data(cp, offset, sector, length); 635 g_topology_lock(); 636 free(sector, M_MIRROR); 637 if (error != 0) { 638 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 639 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 640 G_MIRROR_EVENT_DONTWAIT); 641 } 642 return (error); 643} 644 645static int 646g_mirror_clear_metadata(struct g_mirror_disk *disk) 647{ 648 int error; 649 650 g_topology_assert(); 651 error = g_mirror_write_metadata(disk, NULL); 652 if (error == 0) { 653 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 654 g_mirror_get_diskname(disk)); 655 } else { 656 G_MIRROR_DEBUG(0, 657 "Cannot clear metadata on disk %s (error=%d).", 658 g_mirror_get_diskname(disk), error); 659 } 660 return (error); 661} 662 663void 664g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 665 struct g_mirror_metadata *md) 666{ 667 668 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 669 md->md_version = G_MIRROR_VERSION; 670 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 671 md->md_mid = sc->sc_id; 672 md->md_all = sc->sc_ndisks; 673 md->md_slice = sc->sc_slice; 674 md->md_balance = sc->sc_balance; 675 md->md_mediasize = sc->sc_mediasize; 676 md->md_sectorsize = sc->sc_sectorsize; 677 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 678 bzero(md->md_provider, sizeof(md->md_provider)); 679 if (disk == NULL) { 680 md->md_did = arc4random(); 681 md->md_priority = 0; 682 md->md_syncid = 0; 683 md->md_dflags = 0; 684 md->md_sync_offset = 0; 685 } else { 686 md->md_did = disk->d_id; 687 md->md_priority = disk->d_priority; 688 md->md_syncid = disk->d_sync.ds_syncid; 689 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 690 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 691 md->md_sync_offset = disk->d_sync.ds_offset_done; 692 else 693 md->md_sync_offset = 0; 694 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 695 strlcpy(md->md_provider, 696 disk->d_consumer->provider->name, 697 sizeof(md->md_provider)); 698 } 699 } 700} 701 702void 703g_mirror_update_metadata(struct g_mirror_disk *disk) 704{ 705 struct g_mirror_metadata md; 706 int error; 707 708 g_topology_assert(); 709 g_mirror_fill_metadata(disk->d_softc, disk, &md); 710 error = g_mirror_write_metadata(disk, &md); 711 if (error == 0) { 712 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 713 g_mirror_get_diskname(disk)); 714 } else { 715 G_MIRROR_DEBUG(0, 716 "Cannot update metadata on disk %s (error=%d).", 717 g_mirror_get_diskname(disk), error); 718 } 719} 720 721static void 722g_mirror_bump_syncid(struct g_mirror_softc *sc) 723{ 724 struct g_mirror_disk *disk; 725 726 g_topology_assert(); 727 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 728 ("%s called with no active disks (device=%s).", __func__, 729 sc->sc_name)); 730 731 sc->sc_syncid++; 732 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 733 sc->sc_syncid); 734 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 735 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 736 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 737 disk->d_sync.ds_syncid = sc->sc_syncid; 738 g_mirror_update_metadata(disk); 739 } 740 } 741} 742 743static void 744g_mirror_idle(struct g_mirror_softc *sc) 745{ 746 struct g_mirror_disk *disk; 747 748 if (sc->sc_provider == NULL || sc->sc_provider->acw == 0) 749 return; 750 sc->sc_idle = 1; 751 g_topology_lock(); 752 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 753 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 754 continue; 755 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 756 g_mirror_get_diskname(disk), sc->sc_name); 757 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 758 g_mirror_update_metadata(disk); 759 } 760 g_topology_unlock(); 761} 762 763static void 764g_mirror_unidle(struct g_mirror_softc *sc) 765{ 766 struct g_mirror_disk *disk; 767 768 sc->sc_idle = 0; 769 g_topology_lock(); 770 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 771 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 772 continue; 773 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 774 g_mirror_get_diskname(disk), sc->sc_name); 775 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 776 g_mirror_update_metadata(disk); 777 } 778 g_topology_unlock(); 779} 780 781/* 782 * Return 1 if we should check if mirror is idling. 783 */ 784static int 785g_mirror_check_idle(struct g_mirror_softc *sc) 786{ 787 struct g_mirror_disk *disk; 788 789 if (sc->sc_idle) 790 return (0); 791 if (sc->sc_provider != NULL && sc->sc_provider->acw == 0) 792 return (0); 793 /* 794 * Check if there are no in-flight requests. 795 */ 796 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 797 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 798 continue; 799 if (disk->d_consumer->index > 0) 800 return (0); 801 } 802 return (1); 803} 804 805static __inline int 806bintime_cmp(struct bintime *bt1, struct bintime *bt2) 807{ 808 809 if (bt1->sec < bt2->sec) 810 return (-1); 811 else if (bt1->sec > bt2->sec) 812 return (1); 813 if (bt1->frac < bt2->frac) 814 return (-1); 815 else if (bt1->frac > bt2->frac) 816 return (1); 817 return (0); 818} 819 820static void 821g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 822{ 823 824 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 825 return; 826 binuptime(&disk->d_delay); 827 bintime_sub(&disk->d_delay, &bp->bio_t0); 828} 829 830static void 831g_mirror_done(struct bio *bp) 832{ 833 struct g_mirror_softc *sc; 834 835 sc = bp->bio_from->geom->softc; 836 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 837 mtx_lock(&sc->sc_queue_mtx); 838 bioq_disksort(&sc->sc_queue, bp); 839 wakeup(sc); 840 mtx_unlock(&sc->sc_queue_mtx); 841} 842 843static void 844g_mirror_regular_request(struct bio *bp) 845{ 846 struct g_mirror_softc *sc; 847 struct g_mirror_disk *disk; 848 struct bio *pbp; 849 850 g_topology_assert_not(); 851 852 bp->bio_from->index--; 853 pbp = bp->bio_parent; 854 sc = pbp->bio_to->geom->softc; 855 disk = bp->bio_from->private; 856 if (disk == NULL) { 857 g_topology_lock(); 858 g_mirror_kill_consumer(sc, bp->bio_from); 859 g_topology_unlock(); 860 } else { 861 g_mirror_update_delay(disk, bp); 862 } 863 864 pbp->bio_inbed++; 865 KASSERT(pbp->bio_inbed <= pbp->bio_children, 866 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 867 pbp->bio_children)); 868 if (bp->bio_error == 0 && pbp->bio_error == 0) { 869 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 870 g_destroy_bio(bp); 871 if (pbp->bio_children == pbp->bio_inbed) { 872 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 873 pbp->bio_completed = pbp->bio_length; 874 g_io_deliver(pbp, pbp->bio_error); 875 } 876 return; 877 } else if (bp->bio_error != 0) { 878 if (pbp->bio_error == 0) 879 pbp->bio_error = bp->bio_error; 880 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 881 bp->bio_error); 882 if (disk != NULL) { 883 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 884 g_mirror_event_send(disk, 885 G_MIRROR_DISK_STATE_DISCONNECTED, 886 G_MIRROR_EVENT_DONTWAIT); 887 } 888 switch (pbp->bio_cmd) { 889 case BIO_DELETE: 890 case BIO_WRITE: 891 pbp->bio_inbed--; 892 pbp->bio_children--; 893 break; 894 } 895 } 896 g_destroy_bio(bp); 897 898 switch (pbp->bio_cmd) { 899 case BIO_READ: 900 if (pbp->bio_children == pbp->bio_inbed) { 901 pbp->bio_error = 0; 902 mtx_lock(&sc->sc_queue_mtx); 903 bioq_disksort(&sc->sc_queue, pbp); 904 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 905 wakeup(sc); 906 mtx_unlock(&sc->sc_queue_mtx); 907 } 908 break; 909 case BIO_DELETE: 910 case BIO_WRITE: 911 if (pbp->bio_children == 0) { 912 /* 913 * All requests failed. 914 */ 915 } else if (pbp->bio_inbed < pbp->bio_children) { 916 /* Do nothing. */ 917 break; 918 } else if (pbp->bio_children == pbp->bio_inbed) { 919 /* Some requests succeeded. */ 920 pbp->bio_error = 0; 921 pbp->bio_completed = pbp->bio_length; 922 } 923 g_io_deliver(pbp, pbp->bio_error); 924 break; 925 default: 926 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 927 break; 928 } 929} 930 931static void 932g_mirror_sync_done(struct bio *bp) 933{ 934 struct g_mirror_softc *sc; 935 936 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 937 sc = bp->bio_from->geom->softc; 938 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 939 mtx_lock(&sc->sc_queue_mtx); 940 bioq_disksort(&sc->sc_queue, bp); 941 wakeup(sc); 942 mtx_unlock(&sc->sc_queue_mtx); 943} 944 945static void 946g_mirror_start(struct bio *bp) 947{ 948 struct g_mirror_softc *sc; 949 950 sc = bp->bio_to->geom->softc; 951 /* 952 * If sc == NULL or there are no valid disks, provider's error 953 * should be set and g_mirror_start() should not be called at all. 954 */ 955 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 956 ("Provider's error should be set (error=%d)(mirror=%s).", 957 bp->bio_to->error, bp->bio_to->name)); 958 G_MIRROR_LOGREQ(3, bp, "Request received."); 959 960 switch (bp->bio_cmd) { 961 case BIO_READ: 962 case BIO_WRITE: 963 case BIO_DELETE: 964 break; 965 case BIO_GETATTR: 966 default: 967 g_io_deliver(bp, EOPNOTSUPP); 968 return; 969 } 970 mtx_lock(&sc->sc_queue_mtx); 971 bioq_disksort(&sc->sc_queue, bp); 972 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 973 wakeup(sc); 974 mtx_unlock(&sc->sc_queue_mtx); 975} 976 977/* 978 * Send one synchronization request. 979 */ 980static void 981g_mirror_sync_one(struct g_mirror_disk *disk) 982{ 983 struct g_mirror_softc *sc; 984 struct bio *bp; 985 986 sc = disk->d_softc; 987 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 988 ("Disk %s is not marked for synchronization.", 989 g_mirror_get_diskname(disk))); 990 991 bp = g_new_bio(); 992 if (bp == NULL) 993 return; 994 bp->bio_parent = NULL; 995 bp->bio_cmd = BIO_READ; 996 bp->bio_offset = disk->d_sync.ds_offset; 997 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 998 bp->bio_cflags = 0; 999 bp->bio_done = g_mirror_sync_done; 1000 bp->bio_data = disk->d_sync.ds_data; 1001 if (bp->bio_data == NULL) { 1002 g_destroy_bio(bp); 1003 return; 1004 } 1005 disk->d_sync.ds_offset += bp->bio_length; 1006 bp->bio_to = sc->sc_provider; 1007 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1008 disk->d_sync.ds_consumer->index++; 1009 g_io_request(bp, disk->d_sync.ds_consumer); 1010} 1011 1012static void 1013g_mirror_sync_request(struct bio *bp) 1014{ 1015 struct g_mirror_softc *sc; 1016 struct g_mirror_disk *disk; 1017 1018 bp->bio_from->index--; 1019 sc = bp->bio_from->geom->softc; 1020 disk = bp->bio_from->private; 1021 if (disk == NULL) { 1022 g_topology_lock(); 1023 g_mirror_kill_consumer(sc, bp->bio_from); 1024 g_topology_unlock(); 1025 g_destroy_bio(bp); 1026 return; 1027 } 1028 1029 /* 1030 * Synchronization request. 1031 */ 1032 switch (bp->bio_cmd) { 1033 case BIO_READ: 1034 { 1035 struct g_consumer *cp; 1036 1037 if (bp->bio_error != 0) { 1038 G_MIRROR_LOGREQ(0, bp, 1039 "Synchronization request failed (error=%d).", 1040 bp->bio_error); 1041 g_destroy_bio(bp); 1042 return; 1043 } 1044 G_MIRROR_LOGREQ(3, bp, 1045 "Synchronization request half-finished."); 1046 bp->bio_cmd = BIO_WRITE; 1047 bp->bio_cflags = 0; 1048 cp = disk->d_consumer; 1049 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1050 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1051 cp->acr, cp->acw, cp->ace)); 1052 cp->index++; 1053 g_io_request(bp, cp); 1054 return; 1055 } 1056 case BIO_WRITE: 1057 { 1058 struct g_mirror_disk_sync *sync; 1059 1060 if (bp->bio_error != 0) { 1061 G_MIRROR_LOGREQ(0, bp, 1062 "Synchronization request failed (error=%d).", 1063 bp->bio_error); 1064 g_destroy_bio(bp); 1065 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 1066 g_mirror_event_send(disk, 1067 G_MIRROR_DISK_STATE_DISCONNECTED, 1068 G_MIRROR_EVENT_DONTWAIT); 1069 return; 1070 } 1071 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1072 sync = &disk->d_sync; 1073 sync->ds_offset_done = bp->bio_offset + bp->bio_length; 1074 g_destroy_bio(bp); 1075 if (sync->ds_resync != -1) 1076 break; 1077 if (sync->ds_offset_done == sc->sc_provider->mediasize) { 1078 /* 1079 * Disk up-to-date, activate it. 1080 */ 1081 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1082 G_MIRROR_EVENT_DONTWAIT); 1083 return; 1084 } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) { 1085 /* 1086 * Update offset_done on every 100 blocks. 1087 * XXX: This should be configurable. 1088 */ 1089 g_topology_lock(); 1090 g_mirror_update_metadata(disk); 1091 g_topology_unlock(); 1092 } 1093 return; 1094 } 1095 default: 1096 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1097 bp->bio_cmd, sc->sc_name)); 1098 break; 1099 } 1100} 1101 1102static void 1103g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1104{ 1105 struct g_mirror_disk *disk; 1106 struct g_consumer *cp; 1107 struct bio *cbp; 1108 1109 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1110 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1111 break; 1112 } 1113 if (disk == NULL) { 1114 if (bp->bio_error == 0) 1115 bp->bio_error = ENXIO; 1116 g_io_deliver(bp, bp->bio_error); 1117 return; 1118 } 1119 cbp = g_clone_bio(bp); 1120 if (cbp == NULL) { 1121 if (bp->bio_error == 0) 1122 bp->bio_error = ENOMEM; 1123 g_io_deliver(bp, bp->bio_error); 1124 return; 1125 } 1126 /* 1127 * Fill in the component buf structure. 1128 */ 1129 cp = disk->d_consumer; 1130 cbp->bio_done = g_mirror_done; 1131 cbp->bio_to = cp->provider; 1132 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1133 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1134 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1135 cp->acw, cp->ace)); 1136 cp->index++; 1137 g_io_request(cbp, cp); 1138} 1139 1140static void 1141g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1142{ 1143 struct g_mirror_disk *disk; 1144 struct g_consumer *cp; 1145 struct bio *cbp; 1146 1147 disk = g_mirror_get_disk(sc); 1148 if (disk == NULL) { 1149 if (bp->bio_error == 0) 1150 bp->bio_error = ENXIO; 1151 g_io_deliver(bp, bp->bio_error); 1152 return; 1153 } 1154 cbp = g_clone_bio(bp); 1155 if (cbp == NULL) { 1156 if (bp->bio_error == 0) 1157 bp->bio_error = ENOMEM; 1158 g_io_deliver(bp, bp->bio_error); 1159 return; 1160 } 1161 /* 1162 * Fill in the component buf structure. 1163 */ 1164 cp = disk->d_consumer; 1165 cbp->bio_done = g_mirror_done; 1166 cbp->bio_to = cp->provider; 1167 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1168 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1169 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1170 cp->acw, cp->ace)); 1171 cp->index++; 1172 g_io_request(cbp, cp); 1173} 1174 1175static void 1176g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1177{ 1178 struct g_mirror_disk *disk, *dp; 1179 struct g_consumer *cp; 1180 struct bio *cbp; 1181 struct bintime curtime; 1182 1183 binuptime(&curtime); 1184 /* 1185 * Find a disk which the smallest load. 1186 */ 1187 disk = NULL; 1188 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1189 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1190 continue; 1191 /* If disk wasn't used for more than 2 sec, use it. */ 1192 if (curtime.sec - dp->d_last_used.sec >= 2) { 1193 disk = dp; 1194 break; 1195 } 1196 if (disk == NULL || 1197 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1198 disk = dp; 1199 } 1200 } 1201 cbp = g_clone_bio(bp); 1202 if (cbp == NULL) { 1203 if (bp->bio_error == 0) 1204 bp->bio_error = ENOMEM; 1205 g_io_deliver(bp, bp->bio_error); 1206 return; 1207 } 1208 /* 1209 * Fill in the component buf structure. 1210 */ 1211 cp = disk->d_consumer; 1212 cbp->bio_done = g_mirror_done; 1213 cbp->bio_to = cp->provider; 1214 binuptime(&disk->d_last_used); 1215 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1216 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1217 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1218 cp->acw, cp->ace)); 1219 cp->index++; 1220 g_io_request(cbp, cp); 1221} 1222 1223static void 1224g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1225{ 1226 struct bio_queue_head queue; 1227 struct g_mirror_disk *disk; 1228 struct g_consumer *cp; 1229 struct bio *cbp; 1230 off_t left, mod, offset, slice; 1231 u_char *data; 1232 u_int ndisks; 1233 1234 if (bp->bio_length <= sc->sc_slice) { 1235 g_mirror_request_round_robin(sc, bp); 1236 return; 1237 } 1238 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1239 slice = bp->bio_length / ndisks; 1240 mod = slice % sc->sc_provider->sectorsize; 1241 if (mod != 0) 1242 slice += sc->sc_provider->sectorsize - mod; 1243 /* 1244 * Allocate all bios before sending any request, so we can 1245 * return ENOMEM in nice and clean way. 1246 */ 1247 left = bp->bio_length; 1248 offset = bp->bio_offset; 1249 data = bp->bio_data; 1250 bioq_init(&queue); 1251 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1252 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1253 continue; 1254 cbp = g_clone_bio(bp); 1255 if (cbp == NULL) { 1256 for (cbp = bioq_first(&queue); cbp != NULL; 1257 cbp = bioq_first(&queue)) { 1258 bioq_remove(&queue, cbp); 1259 g_destroy_bio(cbp); 1260 } 1261 if (bp->bio_error == 0) 1262 bp->bio_error = ENOMEM; 1263 g_io_deliver(bp, bp->bio_error); 1264 return; 1265 } 1266 bioq_insert_tail(&queue, cbp); 1267 cbp->bio_done = g_mirror_done; 1268 cbp->bio_caller1 = disk; 1269 cbp->bio_to = disk->d_consumer->provider; 1270 cbp->bio_offset = offset; 1271 cbp->bio_data = data; 1272 cbp->bio_length = MIN(left, slice); 1273 left -= cbp->bio_length; 1274 if (left == 0) 1275 break; 1276 offset += cbp->bio_length; 1277 data += cbp->bio_length; 1278 } 1279 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1280 bioq_remove(&queue, cbp); 1281 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1282 disk = cbp->bio_caller1; 1283 cbp->bio_caller1 = NULL; 1284 cp = disk->d_consumer; 1285 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1286 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1287 cp->acr, cp->acw, cp->ace)); 1288 disk->d_consumer->index++; 1289 g_io_request(cbp, disk->d_consumer); 1290 } 1291} 1292 1293static void 1294g_mirror_register_request(struct bio *bp) 1295{ 1296 struct g_mirror_softc *sc; 1297 1298 sc = bp->bio_to->geom->softc; 1299 switch (bp->bio_cmd) { 1300 case BIO_READ: 1301 switch (sc->sc_balance) { 1302 case G_MIRROR_BALANCE_LOAD: 1303 g_mirror_request_load(sc, bp); 1304 break; 1305 case G_MIRROR_BALANCE_PREFER: 1306 g_mirror_request_prefer(sc, bp); 1307 break; 1308 case G_MIRROR_BALANCE_ROUND_ROBIN: 1309 g_mirror_request_round_robin(sc, bp); 1310 break; 1311 case G_MIRROR_BALANCE_SPLIT: 1312 g_mirror_request_split(sc, bp); 1313 break; 1314 } 1315 return; 1316 case BIO_WRITE: 1317 case BIO_DELETE: 1318 { 1319 struct g_mirror_disk *disk; 1320 struct g_mirror_disk_sync *sync; 1321 struct bio_queue_head queue; 1322 struct g_consumer *cp; 1323 struct bio *cbp; 1324 1325 if (sc->sc_idle) 1326 g_mirror_unidle(sc); 1327 /* 1328 * Allocate all bios before sending any request, so we can 1329 * return ENOMEM in nice and clean way. 1330 */ 1331 bioq_init(&queue); 1332 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1333 sync = &disk->d_sync; 1334 switch (disk->d_state) { 1335 case G_MIRROR_DISK_STATE_ACTIVE: 1336 break; 1337 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1338 if (bp->bio_offset >= sync->ds_offset) 1339 continue; 1340 else if (bp->bio_offset + bp->bio_length > 1341 sync->ds_offset_done && 1342 (bp->bio_offset < sync->ds_resync || 1343 sync->ds_resync == -1)) { 1344 sync->ds_resync = bp->bio_offset - 1345 (bp->bio_offset % MAXPHYS); 1346 } 1347 break; 1348 default: 1349 continue; 1350 } 1351 cbp = g_clone_bio(bp); 1352 if (cbp == NULL) { 1353 for (cbp = bioq_first(&queue); cbp != NULL; 1354 cbp = bioq_first(&queue)) { 1355 bioq_remove(&queue, cbp); 1356 g_destroy_bio(cbp); 1357 } 1358 if (bp->bio_error == 0) 1359 bp->bio_error = ENOMEM; 1360 g_io_deliver(bp, bp->bio_error); 1361 return; 1362 } 1363 bioq_insert_tail(&queue, cbp); 1364 cbp->bio_done = g_mirror_done; 1365 cp = disk->d_consumer; 1366 cbp->bio_caller1 = cp; 1367 cbp->bio_to = cp->provider; 1368 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1369 ("Consumer %s not opened (r%dw%de%d).", 1370 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1371 } 1372 for (cbp = bioq_first(&queue); cbp != NULL; 1373 cbp = bioq_first(&queue)) { 1374 bioq_remove(&queue, cbp); 1375 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1376 cp = cbp->bio_caller1; 1377 cbp->bio_caller1 = NULL; 1378 cp->index++; 1379 g_io_request(cbp, cp); 1380 } 1381 /* 1382 * Bump syncid on first write. 1383 */ 1384 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) { 1385 sc->sc_bump_syncid = 0; 1386 g_topology_lock(); 1387 g_mirror_bump_syncid(sc); 1388 g_topology_unlock(); 1389 } 1390 return; 1391 } 1392 default: 1393 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1394 bp->bio_cmd, sc->sc_name)); 1395 break; 1396 } 1397} 1398 1399static int 1400g_mirror_can_destroy(struct g_mirror_softc *sc) 1401{ 1402 struct g_geom *gp; 1403 struct g_consumer *cp; 1404 1405 g_topology_assert(); 1406 gp = sc->sc_geom; 1407 LIST_FOREACH(cp, &gp->consumer, consumer) { 1408 if (g_mirror_is_busy(sc, cp)) 1409 return (0); 1410 } 1411 gp = sc->sc_sync.ds_geom; 1412 LIST_FOREACH(cp, &gp->consumer, consumer) { 1413 if (g_mirror_is_busy(sc, cp)) 1414 return (0); 1415 } 1416 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1417 sc->sc_name); 1418 return (1); 1419} 1420 1421static int 1422g_mirror_try_destroy(struct g_mirror_softc *sc) 1423{ 1424 1425 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1426 g_topology_lock(); 1427 if (!g_mirror_can_destroy(sc)) { 1428 g_topology_unlock(); 1429 return (0); 1430 } 1431 g_topology_unlock(); 1432 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1433 &sc->sc_worker); 1434 wakeup(&sc->sc_worker); 1435 sc->sc_worker = NULL; 1436 } else { 1437 g_topology_lock(); 1438 if (!g_mirror_can_destroy(sc)) { 1439 g_topology_unlock(); 1440 return (0); 1441 } 1442 g_mirror_destroy_device(sc); 1443 g_topology_unlock(); 1444 free(sc, M_MIRROR); 1445 } 1446 return (1); 1447} 1448 1449/* 1450 * Worker thread. 1451 */ 1452static void 1453g_mirror_worker(void *arg) 1454{ 1455 struct g_mirror_softc *sc; 1456 struct g_mirror_disk *disk; 1457 struct g_mirror_disk_sync *sync; 1458 struct g_mirror_event *ep; 1459 struct bio *bp; 1460 u_int nreqs; 1461 1462 sc = arg; 1463 curthread->td_base_pri = PRIBIO; 1464 1465 nreqs = 0; 1466 for (;;) { 1467 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1468 /* 1469 * First take a look at events. 1470 * This is important to handle events before any I/O requests. 1471 */ 1472 ep = g_mirror_event_get(sc); 1473 if (ep != NULL && g_topology_try_lock()) { 1474 g_mirror_event_remove(sc, ep); 1475 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1476 /* Update only device status. */ 1477 G_MIRROR_DEBUG(3, 1478 "Running event for device %s.", 1479 sc->sc_name); 1480 ep->e_error = 0; 1481 g_mirror_update_device(sc, 1); 1482 } else { 1483 /* Update disk status. */ 1484 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1485 g_mirror_get_diskname(ep->e_disk)); 1486 ep->e_error = g_mirror_update_disk(ep->e_disk, 1487 ep->e_state); 1488 if (ep->e_error == 0) 1489 g_mirror_update_device(sc, 0); 1490 } 1491 g_topology_unlock(); 1492 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1493 KASSERT(ep->e_error == 0, 1494 ("Error cannot be handled.")); 1495 g_mirror_event_free(ep); 1496 } else { 1497 ep->e_flags |= G_MIRROR_EVENT_DONE; 1498 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1499 ep); 1500 mtx_lock(&sc->sc_events_mtx); 1501 wakeup(ep); 1502 mtx_unlock(&sc->sc_events_mtx); 1503 } 1504 if ((sc->sc_flags & 1505 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1506 if (g_mirror_try_destroy(sc)) 1507 kthread_exit(0); 1508 } 1509 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1510 continue; 1511 } 1512 /* 1513 * Now I/O requests. 1514 */ 1515 /* Get first request from the queue. */ 1516 mtx_lock(&sc->sc_queue_mtx); 1517 bp = bioq_first(&sc->sc_queue); 1518 if (bp == NULL) { 1519 if (ep != NULL) { 1520 /* 1521 * No I/O requests and topology lock was 1522 * already held? Try again. 1523 */ 1524 mtx_unlock(&sc->sc_queue_mtx); 1525 continue; 1526 } 1527 if ((sc->sc_flags & 1528 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1529 mtx_unlock(&sc->sc_queue_mtx); 1530 if (g_mirror_try_destroy(sc)) 1531 kthread_exit(0); 1532 mtx_lock(&sc->sc_queue_mtx); 1533 } 1534 } 1535 if (sc->sc_sync.ds_ndisks > 0 && 1536 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1537 mtx_unlock(&sc->sc_queue_mtx); 1538 /* 1539 * It is time for synchronization... 1540 */ 1541 nreqs = 0; 1542 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1543 if (disk->d_state != 1544 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1545 continue; 1546 } 1547 sync = &disk->d_sync; 1548 if (sync->ds_offset >= 1549 sc->sc_provider->mediasize) { 1550 continue; 1551 } 1552 if (sync->ds_offset > sync->ds_offset_done) 1553 continue; 1554 if (sync->ds_resync != -1) { 1555 sync->ds_offset = sync->ds_resync; 1556 sync->ds_offset_done = sync->ds_resync; 1557 sync->ds_resync = -1; 1558 } 1559 g_mirror_sync_one(disk); 1560 } 1561 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1562 goto sleep; 1563 } 1564 if (bp == NULL) { 1565 if (g_mirror_check_idle(sc)) { 1566 u_int idletime; 1567 1568 idletime = g_mirror_idletime; 1569 if (idletime == 0) 1570 idletime = 1; 1571 idletime *= hz; 1572 if (msleep(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1573 "m:w1", idletime) == EWOULDBLOCK) { 1574 G_MIRROR_DEBUG(5, "%s: I'm here 3.", 1575 __func__); 1576 /* 1577 * No I/O requests in 'idletime' seconds, 1578 * so mark components as clean. 1579 */ 1580 g_mirror_idle(sc); 1581 } 1582 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1583 } else { 1584 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1585 "m:w2", 0); 1586 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1587 } 1588 continue; 1589 } 1590 nreqs++; 1591 bioq_remove(&sc->sc_queue, bp); 1592 mtx_unlock(&sc->sc_queue_mtx); 1593 1594 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1595 g_mirror_regular_request(bp); 1596 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1597 u_int timeout, sps; 1598 1599 g_mirror_sync_request(bp); 1600sleep: 1601 sps = g_mirror_syncs_per_sec; 1602 if (sps == 0) { 1603 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1604 continue; 1605 } 1606 if (ep != NULL) { 1607 /* 1608 * We have some pending events, don't sleep now. 1609 */ 1610 G_MIRROR_DEBUG(5, "%s: I'm here 7.", __func__); 1611 continue; 1612 } 1613 mtx_lock(&sc->sc_queue_mtx); 1614 if (bioq_first(&sc->sc_queue) != NULL) { 1615 mtx_unlock(&sc->sc_queue_mtx); 1616 G_MIRROR_DEBUG(5, "%s: I'm here 8.", __func__); 1617 continue; 1618 } 1619 timeout = hz / sps; 1620 if (timeout == 0) 1621 timeout = 1; 1622 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w3", 1623 timeout); 1624 } else { 1625 g_mirror_register_request(bp); 1626 } 1627 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1628 } 1629} 1630 1631/* 1632 * Open disk's consumer if needed. 1633 */ 1634static void 1635g_mirror_update_access(struct g_mirror_disk *disk) 1636{ 1637 struct g_provider *pp; 1638 1639 g_topology_assert(); 1640 1641 pp = disk->d_softc->sc_provider; 1642 if (pp == NULL) 1643 return; 1644 if (pp->acw > 0) { 1645 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1646 G_MIRROR_DEBUG(1, 1647 "Disk %s (device %s) marked as dirty.", 1648 g_mirror_get_diskname(disk), 1649 disk->d_softc->sc_name); 1650 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1651 } 1652 } else if (pp->acw == 0) { 1653 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1654 G_MIRROR_DEBUG(1, 1655 "Disk %s (device %s) marked as clean.", 1656 g_mirror_get_diskname(disk), 1657 disk->d_softc->sc_name); 1658 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1659 } 1660 } 1661} 1662 1663static void 1664g_mirror_sync_start(struct g_mirror_disk *disk) 1665{ 1666 struct g_mirror_softc *sc; 1667 int error; 1668 1669 g_topology_assert(); 1670 1671 sc = disk->d_softc; 1672 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1673 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1674 sc->sc_state)); 1675 1676 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1677 g_mirror_get_diskname(disk)); 1678 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1679 KASSERT(disk->d_sync.ds_consumer == NULL, 1680 ("Sync consumer already exists (device=%s, disk=%s).", 1681 sc->sc_name, g_mirror_get_diskname(disk))); 1682 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1683 disk->d_sync.ds_consumer->private = disk; 1684 disk->d_sync.ds_consumer->index = 0; 1685 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1686 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1687 disk->d_softc->sc_name, error)); 1688 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1689 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1690 disk->d_softc->sc_name, error)); 1691 disk->d_sync.ds_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1692 sc->sc_sync.ds_ndisks++; 1693} 1694 1695/* 1696 * Stop synchronization process. 1697 * type: 0 - synchronization finished 1698 * 1 - synchronization stopped 1699 */ 1700static void 1701g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1702{ 1703 1704 g_topology_assert(); 1705 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1706 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1707 g_mirror_disk_state2str(disk->d_state))); 1708 if (disk->d_sync.ds_consumer == NULL) 1709 return; 1710 1711 if (type == 0) { 1712 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1713 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1714 } else /* if (type == 1) */ { 1715 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1716 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1717 } 1718 g_mirror_kill_consumer(disk->d_softc, disk->d_sync.ds_consumer); 1719 free(disk->d_sync.ds_data, M_MIRROR); 1720 disk->d_sync.ds_consumer = NULL; 1721 disk->d_softc->sc_sync.ds_ndisks--; 1722 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1723} 1724 1725static void 1726g_mirror_launch_provider(struct g_mirror_softc *sc) 1727{ 1728 struct g_mirror_disk *disk; 1729 struct g_provider *pp; 1730 1731 g_topology_assert(); 1732 1733 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1734 pp->mediasize = sc->sc_mediasize; 1735 pp->sectorsize = sc->sc_sectorsize; 1736 sc->sc_provider = pp; 1737 g_error_provider(pp, 0); 1738 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1739 pp->name); 1740 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1741 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1742 g_mirror_sync_start(disk); 1743 } 1744} 1745 1746static void 1747g_mirror_destroy_provider(struct g_mirror_softc *sc) 1748{ 1749 struct g_mirror_disk *disk; 1750 struct bio *bp; 1751 1752 g_topology_assert(); 1753 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1754 sc->sc_name)); 1755 1756 g_error_provider(sc->sc_provider, ENXIO); 1757 mtx_lock(&sc->sc_queue_mtx); 1758 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1759 bioq_remove(&sc->sc_queue, bp); 1760 g_io_deliver(bp, ENXIO); 1761 } 1762 mtx_unlock(&sc->sc_queue_mtx); 1763 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1764 sc->sc_provider->name); 1765 sc->sc_provider->flags |= G_PF_WITHER; 1766 g_orphan_provider(sc->sc_provider, ENXIO); 1767 sc->sc_provider = NULL; 1768 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1769 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1770 g_mirror_sync_stop(disk, 1); 1771 } 1772} 1773 1774static void 1775g_mirror_go(void *arg) 1776{ 1777 struct g_mirror_softc *sc; 1778 1779 sc = arg; 1780 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1781 g_mirror_event_send(sc, 0, 1782 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1783} 1784 1785static u_int 1786g_mirror_determine_state(struct g_mirror_disk *disk) 1787{ 1788 struct g_mirror_softc *sc; 1789 u_int state; 1790 1791 sc = disk->d_softc; 1792 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1793 if ((disk->d_flags & 1794 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1795 /* Disk does not need synchronization. */ 1796 state = G_MIRROR_DISK_STATE_ACTIVE; 1797 } else { 1798 if ((sc->sc_flags & 1799 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1800 (disk->d_flags & 1801 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1802 /* 1803 * We can start synchronization from 1804 * the stored offset. 1805 */ 1806 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1807 } else { 1808 state = G_MIRROR_DISK_STATE_STALE; 1809 } 1810 } 1811 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1812 /* 1813 * Reset all synchronization data for this disk, 1814 * because if it even was synchronized, it was 1815 * synchronized to disks with different syncid. 1816 */ 1817 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1818 disk->d_sync.ds_offset = 0; 1819 disk->d_sync.ds_offset_done = 0; 1820 disk->d_sync.ds_syncid = sc->sc_syncid; 1821 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1822 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1823 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1824 } else { 1825 state = G_MIRROR_DISK_STATE_STALE; 1826 } 1827 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1828 /* 1829 * Not good, NOT GOOD! 1830 * It means that mirror was started on stale disks 1831 * and more fresh disk just arrive. 1832 * If there were writes, mirror is fucked up, sorry. 1833 * I think the best choice here is don't touch 1834 * this disk and inform the user laudly. 1835 */ 1836 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1837 "disk (%s) arrives!! It will not be connected to the " 1838 "running device.", sc->sc_name, 1839 g_mirror_get_diskname(disk)); 1840 g_mirror_destroy_disk(disk); 1841 state = G_MIRROR_DISK_STATE_NONE; 1842 /* Return immediately, because disk was destroyed. */ 1843 return (state); 1844 } 1845 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1846 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1847 return (state); 1848} 1849 1850/* 1851 * Update device state. 1852 */ 1853static void 1854g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1855{ 1856 struct g_mirror_disk *disk; 1857 u_int state; 1858 1859 g_topology_assert(); 1860 1861 switch (sc->sc_state) { 1862 case G_MIRROR_DEVICE_STATE_STARTING: 1863 { 1864 struct g_mirror_disk *pdisk; 1865 u_int dirty, ndisks, syncid; 1866 1867 KASSERT(sc->sc_provider == NULL, 1868 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1869 /* 1870 * Are we ready? We are, if all disks are connected or 1871 * if we have any disks and 'force' is true. 1872 */ 1873 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1874 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1875 ; 1876 } else if (g_mirror_ndisks(sc, -1) == 0) { 1877 /* 1878 * Disks went down in starting phase, so destroy 1879 * device. 1880 */ 1881 callout_drain(&sc->sc_callout); 1882 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1883 return; 1884 } else { 1885 return; 1886 } 1887 1888 /* 1889 * Activate all disks with the biggest syncid. 1890 */ 1891 if (force) { 1892 /* 1893 * If 'force' is true, we have been called due to 1894 * timeout, so don't bother canceling timeout. 1895 */ 1896 ndisks = 0; 1897 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1898 if ((disk->d_flags & 1899 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1900 ndisks++; 1901 } 1902 } 1903 if (ndisks == 0) { 1904 /* No valid disks found, destroy device. */ 1905 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1906 return; 1907 } 1908 } else { 1909 /* Cancel timeout. */ 1910 callout_drain(&sc->sc_callout); 1911 } 1912 1913 /* 1914 * Find disk with the biggest syncid. 1915 */ 1916 syncid = 0; 1917 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1918 if (disk->d_sync.ds_syncid > syncid) 1919 syncid = disk->d_sync.ds_syncid; 1920 } 1921 1922 /* 1923 * Here we need to look for dirty disks and if all disks 1924 * with the biggest syncid are dirty, we have to choose 1925 * one with the biggest priority and rebuild the rest. 1926 */ 1927 /* 1928 * Find the number of dirty disks with the biggest syncid. 1929 * Find the number of disks with the biggest syncid. 1930 * While here, find a disk with the biggest priority. 1931 */ 1932 dirty = ndisks = 0; 1933 pdisk = NULL; 1934 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1935 if (disk->d_sync.ds_syncid != syncid) 1936 continue; 1937 if ((disk->d_flags & 1938 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1939 continue; 1940 } 1941 ndisks++; 1942 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1943 dirty++; 1944 if (pdisk == NULL || 1945 pdisk->d_priority < disk->d_priority) { 1946 pdisk = disk; 1947 } 1948 } 1949 } 1950 if (dirty == 0) { 1951 /* No dirty disks at all, great. */ 1952 } else if (dirty == ndisks) { 1953 /* 1954 * Force synchronization for all dirty disks except one 1955 * with the biggest priority. 1956 */ 1957 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1958 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1959 "master disk for synchronization.", 1960 g_mirror_get_diskname(pdisk), sc->sc_name); 1961 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1962 if (disk->d_sync.ds_syncid != syncid) 1963 continue; 1964 if ((disk->d_flags & 1965 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1966 continue; 1967 } 1968 KASSERT((disk->d_flags & 1969 G_MIRROR_DISK_FLAG_DIRTY) != 0, 1970 ("Disk %s isn't marked as dirty.", 1971 g_mirror_get_diskname(disk))); 1972 /* Skip the disk with the biggest priority. */ 1973 if (disk == pdisk) 1974 continue; 1975 disk->d_sync.ds_syncid = 0; 1976 } 1977 } else if (dirty < ndisks) { 1978 /* 1979 * Force synchronization for all dirty disks. 1980 * We have some non-dirty disks. 1981 */ 1982 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1983 if (disk->d_sync.ds_syncid != syncid) 1984 continue; 1985 if ((disk->d_flags & 1986 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1987 continue; 1988 } 1989 if ((disk->d_flags & 1990 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1991 continue; 1992 } 1993 disk->d_sync.ds_syncid = 0; 1994 } 1995 } 1996 1997 /* Reset hint. */ 1998 sc->sc_hint = NULL; 1999 sc->sc_syncid = syncid; 2000 if (force) { 2001 /* Remember to bump syncid on first write. */ 2002 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 2003 } 2004 state = G_MIRROR_DEVICE_STATE_RUNNING; 2005 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2006 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2007 g_mirror_device_state2str(state)); 2008 sc->sc_state = state; 2009 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2010 state = g_mirror_determine_state(disk); 2011 g_mirror_event_send(disk, state, 2012 G_MIRROR_EVENT_DONTWAIT); 2013 if (state == G_MIRROR_DISK_STATE_STALE) { 2014 sc->sc_bump_syncid = 2015 G_MIRROR_BUMP_ON_FIRST_WRITE; 2016 } 2017 } 2018 wakeup(&g_mirror_class); 2019 break; 2020 } 2021 case G_MIRROR_DEVICE_STATE_RUNNING: 2022 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2023 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2024 /* 2025 * No active disks or no disks at all, 2026 * so destroy device. 2027 */ 2028 if (sc->sc_provider != NULL) 2029 g_mirror_destroy_provider(sc); 2030 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2031 break; 2032 } else if (g_mirror_ndisks(sc, 2033 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2034 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2035 /* 2036 * We have active disks, launch provider if it doesn't 2037 * exist. 2038 */ 2039 if (sc->sc_provider == NULL) 2040 g_mirror_launch_provider(sc); 2041 } 2042 /* 2043 * Bump syncid here, if we need to do it immediately. 2044 */ 2045 if (sc->sc_bump_syncid == G_MIRROR_BUMP_IMMEDIATELY) { 2046 sc->sc_bump_syncid = 0; 2047 g_mirror_bump_syncid(sc); 2048 } 2049 break; 2050 default: 2051 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2052 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2053 break; 2054 } 2055} 2056 2057/* 2058 * Update disk state and device state if needed. 2059 */ 2060#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2061 "Disk %s state changed from %s to %s (device %s).", \ 2062 g_mirror_get_diskname(disk), \ 2063 g_mirror_disk_state2str(disk->d_state), \ 2064 g_mirror_disk_state2str(state), sc->sc_name) 2065static int 2066g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2067{ 2068 struct g_mirror_softc *sc; 2069 2070 g_topology_assert(); 2071 2072 sc = disk->d_softc; 2073again: 2074 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2075 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2076 g_mirror_disk_state2str(state)); 2077 switch (state) { 2078 case G_MIRROR_DISK_STATE_NEW: 2079 /* 2080 * Possible scenarios: 2081 * 1. New disk arrive. 2082 */ 2083 /* Previous state should be NONE. */ 2084 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2085 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2086 g_mirror_disk_state2str(disk->d_state))); 2087 DISK_STATE_CHANGED(); 2088 2089 disk->d_state = state; 2090 if (LIST_EMPTY(&sc->sc_disks)) 2091 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2092 else { 2093 struct g_mirror_disk *dp; 2094 2095 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2096 if (disk->d_priority >= dp->d_priority) { 2097 LIST_INSERT_BEFORE(dp, disk, d_next); 2098 dp = NULL; 2099 break; 2100 } 2101 if (LIST_NEXT(dp, d_next) == NULL) 2102 break; 2103 } 2104 if (dp != NULL) 2105 LIST_INSERT_AFTER(dp, disk, d_next); 2106 } 2107 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2108 sc->sc_name, g_mirror_get_diskname(disk)); 2109 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2110 break; 2111 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2112 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2113 g_mirror_device_state2str(sc->sc_state), 2114 g_mirror_get_diskname(disk), 2115 g_mirror_disk_state2str(disk->d_state))); 2116 state = g_mirror_determine_state(disk); 2117 if (state != G_MIRROR_DISK_STATE_NONE) 2118 goto again; 2119 break; 2120 case G_MIRROR_DISK_STATE_ACTIVE: 2121 /* 2122 * Possible scenarios: 2123 * 1. New disk does not need synchronization. 2124 * 2. Synchronization process finished successfully. 2125 */ 2126 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2127 ("Wrong device state (%s, %s, %s, %s).", 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 /* Previous state should be NEW or SYNCHRONIZING. */ 2132 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2133 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2134 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2135 g_mirror_disk_state2str(disk->d_state))); 2136 DISK_STATE_CHANGED(); 2137 2138 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2139 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2140 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2141 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2142 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2143 g_mirror_sync_stop(disk, 0); 2144 } 2145 disk->d_state = state; 2146 disk->d_sync.ds_offset = 0; 2147 disk->d_sync.ds_offset_done = 0; 2148 g_mirror_update_access(disk); 2149 g_mirror_update_metadata(disk); 2150 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2151 sc->sc_name, g_mirror_get_diskname(disk)); 2152 break; 2153 case G_MIRROR_DISK_STATE_STALE: 2154 /* 2155 * Possible scenarios: 2156 * 1. Stale disk was connected. 2157 */ 2158 /* Previous state should be NEW. */ 2159 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2160 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2161 g_mirror_disk_state2str(disk->d_state))); 2162 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2163 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2164 g_mirror_device_state2str(sc->sc_state), 2165 g_mirror_get_diskname(disk), 2166 g_mirror_disk_state2str(disk->d_state))); 2167 /* 2168 * STALE state is only possible if device is marked 2169 * NOAUTOSYNC. 2170 */ 2171 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2172 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2173 g_mirror_device_state2str(sc->sc_state), 2174 g_mirror_get_diskname(disk), 2175 g_mirror_disk_state2str(disk->d_state))); 2176 DISK_STATE_CHANGED(); 2177 2178 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2179 disk->d_state = state; 2180 g_mirror_update_metadata(disk); 2181 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2182 sc->sc_name, g_mirror_get_diskname(disk)); 2183 break; 2184 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2185 /* 2186 * Possible scenarios: 2187 * 1. Disk which needs synchronization was connected. 2188 */ 2189 /* Previous state should be NEW. */ 2190 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2191 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2192 g_mirror_disk_state2str(disk->d_state))); 2193 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2194 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2195 g_mirror_device_state2str(sc->sc_state), 2196 g_mirror_get_diskname(disk), 2197 g_mirror_disk_state2str(disk->d_state))); 2198 DISK_STATE_CHANGED(); 2199 2200 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2201 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2202 disk->d_state = state; 2203 if (sc->sc_provider != NULL) { 2204 g_mirror_sync_start(disk); 2205 g_mirror_update_metadata(disk); 2206 } 2207 break; 2208 case G_MIRROR_DISK_STATE_DISCONNECTED: 2209 /* 2210 * Possible scenarios: 2211 * 1. Device wasn't running yet, but disk disappear. 2212 * 2. Disk was active and disapppear. 2213 * 3. Disk disappear during synchronization process. 2214 */ 2215 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2216 /* 2217 * Previous state should be ACTIVE, STALE or 2218 * SYNCHRONIZING. 2219 */ 2220 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2221 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2222 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2223 ("Wrong disk state (%s, %s).", 2224 g_mirror_get_diskname(disk), 2225 g_mirror_disk_state2str(disk->d_state))); 2226 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2227 /* Previous state should be NEW. */ 2228 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2229 ("Wrong disk state (%s, %s).", 2230 g_mirror_get_diskname(disk), 2231 g_mirror_disk_state2str(disk->d_state))); 2232 /* 2233 * Reset bumping syncid if disk disappeared in STARTING 2234 * state. 2235 */ 2236 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) 2237 sc->sc_bump_syncid = 0; 2238#ifdef INVARIANTS 2239 } else { 2240 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2241 sc->sc_name, 2242 g_mirror_device_state2str(sc->sc_state), 2243 g_mirror_get_diskname(disk), 2244 g_mirror_disk_state2str(disk->d_state))); 2245#endif 2246 } 2247 DISK_STATE_CHANGED(); 2248 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2249 sc->sc_name, g_mirror_get_diskname(disk)); 2250 2251 g_mirror_destroy_disk(disk); 2252 break; 2253 case G_MIRROR_DISK_STATE_DESTROY: 2254 { 2255 int error; 2256 2257 error = g_mirror_clear_metadata(disk); 2258 if (error != 0) 2259 return (error); 2260 DISK_STATE_CHANGED(); 2261 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2262 sc->sc_name, g_mirror_get_diskname(disk)); 2263 2264 g_mirror_destroy_disk(disk); 2265 sc->sc_ndisks--; 2266 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2267 g_mirror_update_metadata(disk); 2268 } 2269 break; 2270 } 2271 default: 2272 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2273 break; 2274 } 2275 return (0); 2276} 2277#undef DISK_STATE_CHANGED 2278 2279static int 2280g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2281{ 2282 struct g_provider *pp; 2283 u_char *buf; 2284 int error; 2285 2286 g_topology_assert(); 2287 2288 error = g_access(cp, 1, 0, 0); 2289 if (error != 0) 2290 return (error); 2291 pp = cp->provider; 2292 g_topology_unlock(); 2293 /* Metadata are stored on last sector. */ 2294 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2295 &error); 2296 g_topology_lock(); 2297 g_access(cp, -1, 0, 0); 2298 if (error != 0) { 2299 if (buf != NULL) 2300 g_free(buf); 2301 return (error); 2302 } 2303 2304 /* Decode metadata. */ 2305 error = mirror_metadata_decode(buf, md); 2306 g_free(buf); 2307 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2308 return (EINVAL); 2309 if (error != 0) { 2310 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2311 cp->provider->name); 2312 return (error); 2313 } 2314 2315 return (0); 2316} 2317 2318static int 2319g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2320 struct g_mirror_metadata *md) 2321{ 2322 2323 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2324 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2325 pp->name, md->md_did); 2326 return (EEXIST); 2327 } 2328 if (md->md_all != sc->sc_ndisks) { 2329 G_MIRROR_DEBUG(1, 2330 "Invalid '%s' field on disk %s (device %s), skipping.", 2331 "md_all", pp->name, sc->sc_name); 2332 return (EINVAL); 2333 } 2334 if (md->md_slice != sc->sc_slice) { 2335 G_MIRROR_DEBUG(1, 2336 "Invalid '%s' field on disk %s (device %s), skipping.", 2337 "md_slice", pp->name, sc->sc_name); 2338 return (EINVAL); 2339 } 2340 if (md->md_balance != sc->sc_balance) { 2341 G_MIRROR_DEBUG(1, 2342 "Invalid '%s' field on disk %s (device %s), skipping.", 2343 "md_balance", pp->name, sc->sc_name); 2344 return (EINVAL); 2345 } 2346 if (md->md_mediasize != sc->sc_mediasize) { 2347 G_MIRROR_DEBUG(1, 2348 "Invalid '%s' field on disk %s (device %s), skipping.", 2349 "md_mediasize", pp->name, sc->sc_name); 2350 return (EINVAL); 2351 } 2352 if (sc->sc_mediasize > pp->mediasize) { 2353 G_MIRROR_DEBUG(1, 2354 "Invalid size of disk %s (device %s), skipping.", pp->name, 2355 sc->sc_name); 2356 return (EINVAL); 2357 } 2358 if (md->md_sectorsize != sc->sc_sectorsize) { 2359 G_MIRROR_DEBUG(1, 2360 "Invalid '%s' field on disk %s (device %s), skipping.", 2361 "md_sectorsize", pp->name, sc->sc_name); 2362 return (EINVAL); 2363 } 2364 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2365 G_MIRROR_DEBUG(1, 2366 "Invalid sector size of disk %s (device %s), skipping.", 2367 pp->name, sc->sc_name); 2368 return (EINVAL); 2369 } 2370 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2371 G_MIRROR_DEBUG(1, 2372 "Invalid device flags on disk %s (device %s), skipping.", 2373 pp->name, sc->sc_name); 2374 return (EINVAL); 2375 } 2376 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2377 G_MIRROR_DEBUG(1, 2378 "Invalid disk flags on disk %s (device %s), skipping.", 2379 pp->name, sc->sc_name); 2380 return (EINVAL); 2381 } 2382 return (0); 2383} 2384 2385static int 2386g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2387 struct g_mirror_metadata *md) 2388{ 2389 struct g_mirror_disk *disk; 2390 int error; 2391 2392 g_topology_assert(); 2393 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2394 2395 error = g_mirror_check_metadata(sc, pp, md); 2396 if (error != 0) 2397 return (error); 2398 disk = g_mirror_init_disk(sc, pp, md, &error); 2399 if (disk == NULL) 2400 return (error); 2401 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2402 G_MIRROR_EVENT_WAIT); 2403 return (error); 2404} 2405 2406static int 2407g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2408{ 2409 struct g_mirror_softc *sc; 2410 struct g_mirror_disk *disk; 2411 int dcr, dcw, dce; 2412 2413 g_topology_assert(); 2414 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2415 acw, ace); 2416 2417 dcr = pp->acr + acr; 2418 dcw = pp->acw + acw; 2419 dce = pp->ace + ace; 2420 2421 sc = pp->geom->softc; 2422 if (sc == NULL || LIST_EMPTY(&sc->sc_disks) || 2423 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 2424 if (acr <= 0 && acw <= 0 && ace <= 0) 2425 return (0); 2426 else 2427 return (ENXIO); 2428 } 2429 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2430 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2431 continue; 2432 /* 2433 * Mark disk as dirty on open and unmark on close. 2434 */ 2435 if (pp->acw == 0 && dcw > 0) { 2436 G_MIRROR_DEBUG(1, 2437 "Disk %s (device %s) marked as dirty.", 2438 g_mirror_get_diskname(disk), sc->sc_name); 2439 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2440 g_mirror_update_metadata(disk); 2441 } else if (pp->acw > 0 && dcw == 0) { 2442 G_MIRROR_DEBUG(1, 2443 "Disk %s (device %s) marked as clean.", 2444 g_mirror_get_diskname(disk), sc->sc_name); 2445 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2446 g_mirror_update_metadata(disk); 2447 } 2448 } 2449 return (0); 2450} 2451 2452static struct g_geom * 2453g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2454{ 2455 struct g_mirror_softc *sc; 2456 struct g_geom *gp; 2457 int error, timeout; 2458 2459 g_topology_assert(); 2460 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2461 md->md_mid); 2462 2463 /* One disk is minimum. */ 2464 if (md->md_all < 1) 2465 return (NULL); 2466 /* 2467 * Action geom. 2468 */ 2469 gp = g_new_geomf(mp, "%s", md->md_name); 2470 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2471 gp->start = g_mirror_start; 2472 gp->spoiled = g_mirror_spoiled; 2473 gp->orphan = g_mirror_orphan; 2474 gp->access = g_mirror_access; 2475 gp->dumpconf = g_mirror_dumpconf; 2476 2477 sc->sc_id = md->md_mid; 2478 sc->sc_slice = md->md_slice; 2479 sc->sc_balance = md->md_balance; 2480 sc->sc_mediasize = md->md_mediasize; 2481 sc->sc_sectorsize = md->md_sectorsize; 2482 sc->sc_ndisks = md->md_all; 2483 sc->sc_flags = md->md_mflags; 2484 sc->sc_bump_syncid = 0; 2485 sc->sc_idle = 0; 2486 bioq_init(&sc->sc_queue); 2487 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2488 LIST_INIT(&sc->sc_disks); 2489 TAILQ_INIT(&sc->sc_events); 2490 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2491 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2492 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2493 gp->softc = sc; 2494 sc->sc_geom = gp; 2495 sc->sc_provider = NULL; 2496 /* 2497 * Synchronization geom. 2498 */ 2499 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2500 gp->softc = sc; 2501 gp->orphan = g_mirror_orphan; 2502 sc->sc_sync.ds_geom = gp; 2503 sc->sc_sync.ds_ndisks = 0; 2504 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2505 "g_mirror %s", md->md_name); 2506 if (error != 0) { 2507 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2508 sc->sc_name); 2509 g_destroy_geom(sc->sc_sync.ds_geom); 2510 mtx_destroy(&sc->sc_events_mtx); 2511 mtx_destroy(&sc->sc_queue_mtx); 2512 g_destroy_geom(sc->sc_geom); 2513 free(sc, M_MIRROR); 2514 return (NULL); 2515 } 2516 2517 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2518 2519 /* 2520 * Run timeout. 2521 */ 2522 timeout = g_mirror_timeout * hz; 2523 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2524 return (sc->sc_geom); 2525} 2526 2527int 2528g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2529{ 2530 struct g_provider *pp; 2531 2532 g_topology_assert(); 2533 2534 if (sc == NULL) 2535 return (ENXIO); 2536 pp = sc->sc_provider; 2537 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2538 if (force) { 2539 G_MIRROR_DEBUG(0, "Device %s is still open, so it " 2540 "can't be definitely removed.", pp->name); 2541 } else { 2542 G_MIRROR_DEBUG(1, 2543 "Device %s is still open (r%dw%de%d).", pp->name, 2544 pp->acr, pp->acw, pp->ace); 2545 return (EBUSY); 2546 } 2547 } 2548 2549 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2550 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2551 g_topology_unlock(); 2552 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2553 mtx_lock(&sc->sc_queue_mtx); 2554 wakeup(sc); 2555 mtx_unlock(&sc->sc_queue_mtx); 2556 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2557 while (sc->sc_worker != NULL) 2558 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2559 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2560 g_topology_lock(); 2561 g_mirror_destroy_device(sc); 2562 free(sc, M_MIRROR); 2563 return (0); 2564} 2565 2566static void 2567g_mirror_taste_orphan(struct g_consumer *cp) 2568{ 2569 2570 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2571 cp->provider->name)); 2572} 2573 2574static struct g_geom * 2575g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2576{ 2577 struct g_mirror_metadata md; 2578 struct g_mirror_softc *sc; 2579 struct g_consumer *cp; 2580 struct g_geom *gp; 2581 int error; 2582 2583 g_topology_assert(); 2584 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2585 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2586 2587 gp = g_new_geomf(mp, "mirror:taste"); 2588 /* 2589 * This orphan function should be never called. 2590 */ 2591 gp->orphan = g_mirror_taste_orphan; 2592 cp = g_new_consumer(gp); 2593 g_attach(cp, pp); 2594 error = g_mirror_read_metadata(cp, &md); 2595 g_detach(cp); 2596 g_destroy_consumer(cp); 2597 g_destroy_geom(gp); 2598 if (error != 0) 2599 return (NULL); 2600 gp = NULL; 2601 2602 if (md.md_version > G_MIRROR_VERSION) { 2603 printf("geom_mirror.ko module is too old to handle %s.\n", 2604 pp->name); 2605 return (NULL); 2606 } 2607 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2608 return (NULL); 2609 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2610 G_MIRROR_DEBUG(0, 2611 "Device %s: provider %s marked as inactive, skipping.", 2612 md.md_name, pp->name); 2613 return (NULL); 2614 } 2615 if (g_mirror_debug >= 2) 2616 mirror_metadata_dump(&md); 2617 2618 /* 2619 * Let's check if device already exists. 2620 */ 2621 sc = NULL; 2622 LIST_FOREACH(gp, &mp->geom, geom) { 2623 sc = gp->softc; 2624 if (sc == NULL) 2625 continue; 2626 if (sc->sc_sync.ds_geom == gp) 2627 continue; 2628 if (strcmp(md.md_name, sc->sc_name) != 0) 2629 continue; 2630 if (md.md_mid != sc->sc_id) { 2631 G_MIRROR_DEBUG(0, "Device %s already configured.", 2632 sc->sc_name); 2633 return (NULL); 2634 } 2635 break; 2636 } 2637 if (gp == NULL) { 2638 gp = g_mirror_create(mp, &md); 2639 if (gp == NULL) { 2640 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2641 md.md_name); 2642 return (NULL); 2643 } 2644 sc = gp->softc; 2645 } 2646 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2647 error = g_mirror_add_disk(sc, pp, &md); 2648 if (error != 0) { 2649 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2650 pp->name, gp->name, error); 2651 if (LIST_EMPTY(&sc->sc_disks)) 2652 g_mirror_destroy(sc, 1); 2653 return (NULL); 2654 } 2655 return (gp); 2656} 2657 2658static int 2659g_mirror_destroy_geom(struct gctl_req *req __unused, 2660 struct g_class *mp __unused, struct g_geom *gp) 2661{ 2662 2663 return (g_mirror_destroy(gp->softc, 0)); 2664} 2665 2666static void 2667g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2668 struct g_consumer *cp, struct g_provider *pp) 2669{ 2670 struct g_mirror_softc *sc; 2671 2672 g_topology_assert(); 2673 2674 sc = gp->softc; 2675 if (sc == NULL) 2676 return; 2677 /* Skip synchronization geom. */ 2678 if (gp == sc->sc_sync.ds_geom) 2679 return; 2680 if (pp != NULL) { 2681 /* Nothing here. */ 2682 } else if (cp != NULL) { 2683 struct g_mirror_disk *disk; 2684 2685 disk = cp->private; 2686 if (disk == NULL) 2687 return; 2688 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2689 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2690 sbuf_printf(sb, "%s<Synchronized>", indent); 2691 if (disk->d_sync.ds_offset_done == 0) 2692 sbuf_printf(sb, "0%%"); 2693 else { 2694 sbuf_printf(sb, "%u%%", 2695 (u_int)((disk->d_sync.ds_offset_done * 100) / 2696 sc->sc_provider->mediasize)); 2697 } 2698 sbuf_printf(sb, "</Synchronized>\n"); 2699 } 2700 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2701 disk->d_sync.ds_syncid); 2702 sbuf_printf(sb, "%s<Flags>", indent); 2703 if (disk->d_flags == 0) 2704 sbuf_printf(sb, "NONE"); 2705 else { 2706 int first = 1; 2707 2708#define ADD_FLAG(flag, name) do { \ 2709 if ((disk->d_flags & (flag)) != 0) { \ 2710 if (!first) \ 2711 sbuf_printf(sb, ", "); \ 2712 else \ 2713 first = 0; \ 2714 sbuf_printf(sb, name); \ 2715 } \ 2716} while (0) 2717 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2718 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2719 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2720 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2721 "SYNCHRONIZING"); 2722 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2723#undef ADD_FLAG 2724 } 2725 sbuf_printf(sb, "</Flags>\n"); 2726 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2727 disk->d_priority); 2728 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2729 g_mirror_disk_state2str(disk->d_state)); 2730 } else { 2731 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2732 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2733 sbuf_printf(sb, "%s<Flags>", indent); 2734 if (sc->sc_flags == 0) 2735 sbuf_printf(sb, "NONE"); 2736 else { 2737 int first = 1; 2738 2739#define ADD_FLAG(flag, name) do { \ 2740 if ((sc->sc_flags & (flag)) != 0) { \ 2741 if (!first) \ 2742 sbuf_printf(sb, ", "); \ 2743 else \ 2744 first = 0; \ 2745 sbuf_printf(sb, name); \ 2746 } \ 2747} while (0) 2748 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2749#undef ADD_FLAG 2750 } 2751 sbuf_printf(sb, "</Flags>\n"); 2752 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2753 (u_int)sc->sc_slice); 2754 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2755 balance_name(sc->sc_balance)); 2756 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2757 sc->sc_ndisks); 2758 sbuf_printf(sb, "%s<State>", indent); 2759 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2760 sbuf_printf(sb, "%s", "STARTING"); 2761 else if (sc->sc_ndisks == 2762 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 2763 sbuf_printf(sb, "%s", "COMPLETE"); 2764 else 2765 sbuf_printf(sb, "%s", "DEGRADED"); 2766 sbuf_printf(sb, "</State>\n"); 2767 } 2768} 2769 2770static void 2771g_mirror_shutdown(void *arg, int howto) 2772{ 2773 struct g_class *mp; 2774 struct g_geom *gp, *gp2; 2775 2776 mp = arg; 2777 DROP_GIANT(); 2778 g_topology_lock(); 2779 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2780 if (gp->softc == NULL) 2781 continue; 2782 g_mirror_destroy(gp->softc, 1); 2783 } 2784 g_topology_unlock(); 2785 PICKUP_GIANT(); 2786#if 0 2787 tsleep(&gp, PRIBIO, "m:shutdown", hz * 20); 2788#endif 2789} 2790 2791static void 2792g_mirror_init(struct g_class *mp) 2793{ 2794 2795 g_mirror_ehtag = EVENTHANDLER_REGISTER(shutdown_post_sync, 2796 g_mirror_shutdown, mp, SHUTDOWN_PRI_FIRST); 2797 if (g_mirror_ehtag == NULL) 2798 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 2799} 2800 2801static void 2802g_mirror_fini(struct g_class *mp) 2803{ 2804 2805 if (g_mirror_ehtag == NULL) 2806 return; 2807 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_ehtag); 2808} 2809 2810static int 2811g_mirror_can_go(void) 2812{ 2813 struct g_mirror_softc *sc; 2814 struct g_geom *gp; 2815 struct g_provider *pp; 2816 int can_go; 2817 2818 DROP_GIANT(); 2819 can_go = 1; 2820 g_topology_lock(); 2821 LIST_FOREACH(gp, &g_mirror_class.geom, geom) { 2822 sc = gp->softc; 2823 if (sc == NULL) { 2824 can_go = 0; 2825 break; 2826 } 2827 pp = sc->sc_provider; 2828 if (pp == NULL || pp->error != 0) { 2829 can_go = 0; 2830 break; 2831 } 2832 } 2833 g_topology_unlock(); 2834 PICKUP_GIANT(); 2835 return (can_go); 2836} 2837 2838static void 2839g_mirror_rootwait(void) 2840{ 2841 2842 /* 2843 * HACK: Wait for GEOM, because g_mirror_rootwait() can be called, 2844 * HACK: before we get providers for tasting. 2845 */ 2846 tsleep(&g_mirror_class, PRIBIO, "mroot", hz * 3); 2847 /* 2848 * Wait for mirrors in degraded state. 2849 */ 2850 for (;;) { 2851 if (g_mirror_can_go()) 2852 break; 2853 tsleep(&g_mirror_class, PRIBIO, "mroot", hz); 2854 } 2855} 2856 2857SYSINIT(g_mirror_root, SI_SUB_RAID, SI_ORDER_FIRST, g_mirror_rootwait, NULL) 2858 2859DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2860