g_mirror.c revision 142727
1/*- 2 * Copyright (c) 2004-2005 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 142727 2005-02-27 23:07:47Z 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 = 1000; 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; 549 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 550 G_MIRROR_EVENT_DONTWAIT); 551} 552 553/* 554 * Function should return the next active disk on the list. 555 * It is possible that it will be the same disk as given. 556 * If there are no active disks on list, NULL is returned. 557 */ 558static __inline struct g_mirror_disk * 559g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 560{ 561 struct g_mirror_disk *dp; 562 563 for (dp = LIST_NEXT(disk, d_next); dp != disk; 564 dp = LIST_NEXT(dp, d_next)) { 565 if (dp == NULL) 566 dp = LIST_FIRST(&sc->sc_disks); 567 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 568 break; 569 } 570 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 571 return (NULL); 572 return (dp); 573} 574 575static struct g_mirror_disk * 576g_mirror_get_disk(struct g_mirror_softc *sc) 577{ 578 struct g_mirror_disk *disk; 579 580 if (sc->sc_hint == NULL) { 581 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 582 if (sc->sc_hint == NULL) 583 return (NULL); 584 } 585 disk = sc->sc_hint; 586 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 587 disk = g_mirror_find_next(sc, disk); 588 if (disk == NULL) 589 return (NULL); 590 } 591 sc->sc_hint = g_mirror_find_next(sc, disk); 592 return (disk); 593} 594 595static int 596g_mirror_write_metadata(struct g_mirror_disk *disk, 597 struct g_mirror_metadata *md) 598{ 599 struct g_mirror_softc *sc; 600 struct g_consumer *cp; 601 off_t offset, length; 602 u_char *sector; 603 int error = 0; 604 605 g_topology_assert(); 606 607 sc = disk->d_softc; 608 cp = disk->d_consumer; 609 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 610 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 611 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 612 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 613 cp->acw, cp->ace)); 614 length = cp->provider->sectorsize; 615 offset = cp->provider->mediasize - length; 616 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 617 if (md != NULL) 618 mirror_metadata_encode(md, sector); 619 g_topology_unlock(); 620 error = g_write_data(cp, offset, sector, length); 621 g_topology_lock(); 622 free(sector, M_MIRROR); 623 if (error != 0) { 624 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_GENID; 625 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 626 G_MIRROR_EVENT_DONTWAIT); 627 } 628 return (error); 629} 630 631static int 632g_mirror_clear_metadata(struct g_mirror_disk *disk) 633{ 634 int error; 635 636 g_topology_assert(); 637 error = g_mirror_write_metadata(disk, NULL); 638 if (error == 0) { 639 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 640 g_mirror_get_diskname(disk)); 641 } else { 642 G_MIRROR_DEBUG(0, 643 "Cannot clear metadata on disk %s (error=%d).", 644 g_mirror_get_diskname(disk), error); 645 } 646 return (error); 647} 648 649void 650g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 651 struct g_mirror_metadata *md) 652{ 653 654 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 655 md->md_version = G_MIRROR_VERSION; 656 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 657 md->md_mid = sc->sc_id; 658 md->md_all = sc->sc_ndisks; 659 md->md_slice = sc->sc_slice; 660 md->md_balance = sc->sc_balance; 661 md->md_genid = sc->sc_genid; 662 md->md_mediasize = sc->sc_mediasize; 663 md->md_sectorsize = sc->sc_sectorsize; 664 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 665 bzero(md->md_provider, sizeof(md->md_provider)); 666 if (disk == NULL) { 667 md->md_did = arc4random(); 668 md->md_priority = 0; 669 md->md_syncid = 0; 670 md->md_dflags = 0; 671 md->md_sync_offset = 0; 672 md->md_provsize = 0; 673 } else { 674 md->md_did = disk->d_id; 675 md->md_priority = disk->d_priority; 676 md->md_syncid = disk->d_sync.ds_syncid; 677 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 678 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 679 md->md_sync_offset = disk->d_sync.ds_offset_done; 680 else 681 md->md_sync_offset = 0; 682 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 683 strlcpy(md->md_provider, 684 disk->d_consumer->provider->name, 685 sizeof(md->md_provider)); 686 } 687 md->md_provsize = disk->d_consumer->provider->mediasize; 688 } 689} 690 691void 692g_mirror_update_metadata(struct g_mirror_disk *disk) 693{ 694 struct g_mirror_metadata md; 695 int error; 696 697 g_topology_assert(); 698 g_mirror_fill_metadata(disk->d_softc, disk, &md); 699 error = g_mirror_write_metadata(disk, &md); 700 if (error == 0) { 701 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 702 g_mirror_get_diskname(disk)); 703 } else { 704 G_MIRROR_DEBUG(0, 705 "Cannot update metadata on disk %s (error=%d).", 706 g_mirror_get_diskname(disk), error); 707 } 708} 709 710static void 711g_mirror_bump_syncid(struct g_mirror_softc *sc) 712{ 713 struct g_mirror_disk *disk; 714 715 g_topology_assert(); 716 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 717 ("%s called with no active disks (device=%s).", __func__, 718 sc->sc_name)); 719 720 sc->sc_syncid++; 721 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 722 sc->sc_syncid); 723 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 724 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 725 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 726 disk->d_sync.ds_syncid = sc->sc_syncid; 727 g_mirror_update_metadata(disk); 728 } 729 } 730} 731 732static void 733g_mirror_bump_genid(struct g_mirror_softc *sc) 734{ 735 struct g_mirror_disk *disk; 736 737 g_topology_assert(); 738 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 739 ("%s called with no active disks (device=%s).", __func__, 740 sc->sc_name)); 741 742 sc->sc_genid++; 743 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 744 sc->sc_genid); 745 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 746 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 747 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 748 disk->d_genid = sc->sc_genid; 749 g_mirror_update_metadata(disk); 750 } 751 } 752} 753 754static void 755g_mirror_idle(struct g_mirror_softc *sc) 756{ 757 struct g_mirror_disk *disk; 758 759 if (sc->sc_provider == NULL || sc->sc_provider->acw == 0) 760 return; 761 sc->sc_idle = 1; 762 g_topology_lock(); 763 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 764 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 765 continue; 766 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 767 g_mirror_get_diskname(disk), sc->sc_name); 768 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 769 g_mirror_update_metadata(disk); 770 } 771 g_topology_unlock(); 772} 773 774static void 775g_mirror_unidle(struct g_mirror_softc *sc) 776{ 777 struct g_mirror_disk *disk; 778 779 sc->sc_idle = 0; 780 g_topology_lock(); 781 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 782 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 783 continue; 784 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 785 g_mirror_get_diskname(disk), sc->sc_name); 786 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 787 g_mirror_update_metadata(disk); 788 } 789 g_topology_unlock(); 790} 791 792/* 793 * Return 1 if we should check if mirror is idling. 794 */ 795static int 796g_mirror_check_idle(struct g_mirror_softc *sc) 797{ 798 struct g_mirror_disk *disk; 799 800 if (sc->sc_idle) 801 return (0); 802 if (sc->sc_provider != NULL && sc->sc_provider->acw == 0) 803 return (0); 804 /* 805 * Check if there are no in-flight requests. 806 */ 807 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 808 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 809 continue; 810 if (disk->d_consumer->index > 0) 811 return (0); 812 } 813 return (1); 814} 815 816static __inline int 817bintime_cmp(struct bintime *bt1, struct bintime *bt2) 818{ 819 820 if (bt1->sec < bt2->sec) 821 return (-1); 822 else if (bt1->sec > bt2->sec) 823 return (1); 824 if (bt1->frac < bt2->frac) 825 return (-1); 826 else if (bt1->frac > bt2->frac) 827 return (1); 828 return (0); 829} 830 831static void 832g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 833{ 834 835 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 836 return; 837 binuptime(&disk->d_delay); 838 bintime_sub(&disk->d_delay, &bp->bio_t0); 839} 840 841static void 842g_mirror_done(struct bio *bp) 843{ 844 struct g_mirror_softc *sc; 845 846 sc = bp->bio_from->geom->softc; 847 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 848 mtx_lock(&sc->sc_queue_mtx); 849 bioq_disksort(&sc->sc_queue, bp); 850 wakeup(sc); 851 mtx_unlock(&sc->sc_queue_mtx); 852} 853 854static void 855g_mirror_regular_request(struct bio *bp) 856{ 857 struct g_mirror_softc *sc; 858 struct g_mirror_disk *disk; 859 struct bio *pbp; 860 861 g_topology_assert_not(); 862 863 bp->bio_from->index--; 864 pbp = bp->bio_parent; 865 sc = pbp->bio_to->geom->softc; 866 disk = bp->bio_from->private; 867 if (disk == NULL) { 868 g_topology_lock(); 869 g_mirror_kill_consumer(sc, bp->bio_from); 870 g_topology_unlock(); 871 } else { 872 g_mirror_update_delay(disk, bp); 873 } 874 875 pbp->bio_inbed++; 876 KASSERT(pbp->bio_inbed <= pbp->bio_children, 877 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 878 pbp->bio_children)); 879 if (bp->bio_error == 0 && pbp->bio_error == 0) { 880 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 881 g_destroy_bio(bp); 882 if (pbp->bio_children == pbp->bio_inbed) { 883 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 884 pbp->bio_completed = pbp->bio_length; 885 g_io_deliver(pbp, pbp->bio_error); 886 } 887 return; 888 } else if (bp->bio_error != 0) { 889 if (pbp->bio_error == 0) 890 pbp->bio_error = bp->bio_error; 891 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 892 bp->bio_error); 893 if (disk != NULL) { 894 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 895 g_mirror_event_send(disk, 896 G_MIRROR_DISK_STATE_DISCONNECTED, 897 G_MIRROR_EVENT_DONTWAIT); 898 } 899 switch (pbp->bio_cmd) { 900 case BIO_DELETE: 901 case BIO_WRITE: 902 pbp->bio_inbed--; 903 pbp->bio_children--; 904 break; 905 } 906 } 907 g_destroy_bio(bp); 908 909 switch (pbp->bio_cmd) { 910 case BIO_READ: 911 if (pbp->bio_children == pbp->bio_inbed) { 912 pbp->bio_error = 0; 913 mtx_lock(&sc->sc_queue_mtx); 914 bioq_disksort(&sc->sc_queue, pbp); 915 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 916 wakeup(sc); 917 mtx_unlock(&sc->sc_queue_mtx); 918 } 919 break; 920 case BIO_DELETE: 921 case BIO_WRITE: 922 if (pbp->bio_children == 0) { 923 /* 924 * All requests failed. 925 */ 926 } else if (pbp->bio_inbed < pbp->bio_children) { 927 /* Do nothing. */ 928 break; 929 } else if (pbp->bio_children == pbp->bio_inbed) { 930 /* Some requests succeeded. */ 931 pbp->bio_error = 0; 932 pbp->bio_completed = pbp->bio_length; 933 } 934 g_io_deliver(pbp, pbp->bio_error); 935 break; 936 default: 937 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 938 break; 939 } 940} 941 942static void 943g_mirror_sync_done(struct bio *bp) 944{ 945 struct g_mirror_softc *sc; 946 947 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 948 sc = bp->bio_from->geom->softc; 949 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 950 mtx_lock(&sc->sc_queue_mtx); 951 bioq_disksort(&sc->sc_queue, bp); 952 wakeup(sc); 953 mtx_unlock(&sc->sc_queue_mtx); 954} 955 956static void 957g_mirror_start(struct bio *bp) 958{ 959 struct g_mirror_softc *sc; 960 961 sc = bp->bio_to->geom->softc; 962 /* 963 * If sc == NULL or there are no valid disks, provider's error 964 * should be set and g_mirror_start() should not be called at all. 965 */ 966 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 967 ("Provider's error should be set (error=%d)(mirror=%s).", 968 bp->bio_to->error, bp->bio_to->name)); 969 G_MIRROR_LOGREQ(3, bp, "Request received."); 970 971 switch (bp->bio_cmd) { 972 case BIO_READ: 973 case BIO_WRITE: 974 case BIO_DELETE: 975 break; 976 case BIO_GETATTR: 977 default: 978 g_io_deliver(bp, EOPNOTSUPP); 979 return; 980 } 981 mtx_lock(&sc->sc_queue_mtx); 982 bioq_disksort(&sc->sc_queue, bp); 983 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 984 wakeup(sc); 985 mtx_unlock(&sc->sc_queue_mtx); 986} 987 988/* 989 * Send one synchronization request. 990 */ 991static void 992g_mirror_sync_one(struct g_mirror_disk *disk) 993{ 994 struct g_mirror_softc *sc; 995 struct bio *bp; 996 997 sc = disk->d_softc; 998 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 999 ("Disk %s is not marked for synchronization.", 1000 g_mirror_get_diskname(disk))); 1001 1002 bp = g_new_bio(); 1003 if (bp == NULL) 1004 return; 1005 bp->bio_parent = NULL; 1006 bp->bio_cmd = BIO_READ; 1007 bp->bio_offset = disk->d_sync.ds_offset; 1008 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1009 bp->bio_cflags = 0; 1010 bp->bio_done = g_mirror_sync_done; 1011 bp->bio_data = disk->d_sync.ds_data; 1012 if (bp->bio_data == NULL) { 1013 g_destroy_bio(bp); 1014 return; 1015 } 1016 disk->d_sync.ds_offset += bp->bio_length; 1017 bp->bio_to = sc->sc_provider; 1018 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1019 disk->d_sync.ds_consumer->index++; 1020 g_io_request(bp, disk->d_sync.ds_consumer); 1021} 1022 1023static void 1024g_mirror_sync_request(struct bio *bp) 1025{ 1026 struct g_mirror_softc *sc; 1027 struct g_mirror_disk *disk; 1028 1029 bp->bio_from->index--; 1030 sc = bp->bio_from->geom->softc; 1031 disk = bp->bio_from->private; 1032 if (disk == NULL) { 1033 g_topology_lock(); 1034 g_mirror_kill_consumer(sc, bp->bio_from); 1035 g_topology_unlock(); 1036 g_destroy_bio(bp); 1037 return; 1038 } 1039 1040 /* 1041 * Synchronization request. 1042 */ 1043 switch (bp->bio_cmd) { 1044 case BIO_READ: 1045 { 1046 struct g_consumer *cp; 1047 1048 if (bp->bio_error != 0) { 1049 G_MIRROR_LOGREQ(0, bp, 1050 "Synchronization request failed (error=%d).", 1051 bp->bio_error); 1052 g_destroy_bio(bp); 1053 return; 1054 } 1055 G_MIRROR_LOGREQ(3, bp, 1056 "Synchronization request half-finished."); 1057 bp->bio_cmd = BIO_WRITE; 1058 bp->bio_cflags = 0; 1059 cp = disk->d_consumer; 1060 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1061 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1062 cp->acr, cp->acw, cp->ace)); 1063 cp->index++; 1064 g_io_request(bp, cp); 1065 return; 1066 } 1067 case BIO_WRITE: 1068 { 1069 struct g_mirror_disk_sync *sync; 1070 1071 if (bp->bio_error != 0) { 1072 G_MIRROR_LOGREQ(0, bp, 1073 "Synchronization request failed (error=%d).", 1074 bp->bio_error); 1075 g_destroy_bio(bp); 1076 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1077 g_mirror_event_send(disk, 1078 G_MIRROR_DISK_STATE_DISCONNECTED, 1079 G_MIRROR_EVENT_DONTWAIT); 1080 return; 1081 } 1082 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1083 sync = &disk->d_sync; 1084 sync->ds_offset_done = bp->bio_offset + bp->bio_length; 1085 g_destroy_bio(bp); 1086 if (sync->ds_resync != -1) 1087 break; 1088 if (sync->ds_offset_done == sc->sc_provider->mediasize) { 1089 /* 1090 * Disk up-to-date, activate it. 1091 */ 1092 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1093 G_MIRROR_EVENT_DONTWAIT); 1094 return; 1095 } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) { 1096 /* 1097 * Update offset_done on every 100 blocks. 1098 * XXX: This should be configurable. 1099 */ 1100 g_topology_lock(); 1101 g_mirror_update_metadata(disk); 1102 g_topology_unlock(); 1103 } 1104 return; 1105 } 1106 default: 1107 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1108 bp->bio_cmd, sc->sc_name)); 1109 break; 1110 } 1111} 1112 1113static void 1114g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1115{ 1116 struct g_mirror_disk *disk; 1117 struct g_consumer *cp; 1118 struct bio *cbp; 1119 1120 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1121 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1122 break; 1123 } 1124 if (disk == NULL) { 1125 if (bp->bio_error == 0) 1126 bp->bio_error = ENXIO; 1127 g_io_deliver(bp, bp->bio_error); 1128 return; 1129 } 1130 cbp = g_clone_bio(bp); 1131 if (cbp == NULL) { 1132 if (bp->bio_error == 0) 1133 bp->bio_error = ENOMEM; 1134 g_io_deliver(bp, bp->bio_error); 1135 return; 1136 } 1137 /* 1138 * Fill in the component buf structure. 1139 */ 1140 cp = disk->d_consumer; 1141 cbp->bio_done = g_mirror_done; 1142 cbp->bio_to = cp->provider; 1143 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1144 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1145 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1146 cp->acw, cp->ace)); 1147 cp->index++; 1148 g_io_request(cbp, cp); 1149} 1150 1151static void 1152g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1153{ 1154 struct g_mirror_disk *disk; 1155 struct g_consumer *cp; 1156 struct bio *cbp; 1157 1158 disk = g_mirror_get_disk(sc); 1159 if (disk == NULL) { 1160 if (bp->bio_error == 0) 1161 bp->bio_error = ENXIO; 1162 g_io_deliver(bp, bp->bio_error); 1163 return; 1164 } 1165 cbp = g_clone_bio(bp); 1166 if (cbp == NULL) { 1167 if (bp->bio_error == 0) 1168 bp->bio_error = ENOMEM; 1169 g_io_deliver(bp, bp->bio_error); 1170 return; 1171 } 1172 /* 1173 * Fill in the component buf structure. 1174 */ 1175 cp = disk->d_consumer; 1176 cbp->bio_done = g_mirror_done; 1177 cbp->bio_to = cp->provider; 1178 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1179 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1180 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1181 cp->acw, cp->ace)); 1182 cp->index++; 1183 g_io_request(cbp, cp); 1184} 1185 1186static void 1187g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1188{ 1189 struct g_mirror_disk *disk, *dp; 1190 struct g_consumer *cp; 1191 struct bio *cbp; 1192 struct bintime curtime; 1193 1194 binuptime(&curtime); 1195 /* 1196 * Find a disk which the smallest load. 1197 */ 1198 disk = NULL; 1199 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1200 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1201 continue; 1202 /* If disk wasn't used for more than 2 sec, use it. */ 1203 if (curtime.sec - dp->d_last_used.sec >= 2) { 1204 disk = dp; 1205 break; 1206 } 1207 if (disk == NULL || 1208 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1209 disk = dp; 1210 } 1211 } 1212 cbp = g_clone_bio(bp); 1213 if (cbp == NULL) { 1214 if (bp->bio_error == 0) 1215 bp->bio_error = ENOMEM; 1216 g_io_deliver(bp, bp->bio_error); 1217 return; 1218 } 1219 /* 1220 * Fill in the component buf structure. 1221 */ 1222 cp = disk->d_consumer; 1223 cbp->bio_done = g_mirror_done; 1224 cbp->bio_to = cp->provider; 1225 binuptime(&disk->d_last_used); 1226 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1227 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1228 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1229 cp->acw, cp->ace)); 1230 cp->index++; 1231 g_io_request(cbp, cp); 1232} 1233 1234static void 1235g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1236{ 1237 struct bio_queue_head queue; 1238 struct g_mirror_disk *disk; 1239 struct g_consumer *cp; 1240 struct bio *cbp; 1241 off_t left, mod, offset, slice; 1242 u_char *data; 1243 u_int ndisks; 1244 1245 if (bp->bio_length <= sc->sc_slice) { 1246 g_mirror_request_round_robin(sc, bp); 1247 return; 1248 } 1249 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1250 slice = bp->bio_length / ndisks; 1251 mod = slice % sc->sc_provider->sectorsize; 1252 if (mod != 0) 1253 slice += sc->sc_provider->sectorsize - mod; 1254 /* 1255 * Allocate all bios before sending any request, so we can 1256 * return ENOMEM in nice and clean way. 1257 */ 1258 left = bp->bio_length; 1259 offset = bp->bio_offset; 1260 data = bp->bio_data; 1261 bioq_init(&queue); 1262 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1263 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1264 continue; 1265 cbp = g_clone_bio(bp); 1266 if (cbp == NULL) { 1267 for (cbp = bioq_first(&queue); cbp != NULL; 1268 cbp = bioq_first(&queue)) { 1269 bioq_remove(&queue, cbp); 1270 g_destroy_bio(cbp); 1271 } 1272 if (bp->bio_error == 0) 1273 bp->bio_error = ENOMEM; 1274 g_io_deliver(bp, bp->bio_error); 1275 return; 1276 } 1277 bioq_insert_tail(&queue, cbp); 1278 cbp->bio_done = g_mirror_done; 1279 cbp->bio_caller1 = disk; 1280 cbp->bio_to = disk->d_consumer->provider; 1281 cbp->bio_offset = offset; 1282 cbp->bio_data = data; 1283 cbp->bio_length = MIN(left, slice); 1284 left -= cbp->bio_length; 1285 if (left == 0) 1286 break; 1287 offset += cbp->bio_length; 1288 data += cbp->bio_length; 1289 } 1290 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1291 bioq_remove(&queue, cbp); 1292 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1293 disk = cbp->bio_caller1; 1294 cbp->bio_caller1 = NULL; 1295 cp = disk->d_consumer; 1296 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1297 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1298 cp->acr, cp->acw, cp->ace)); 1299 disk->d_consumer->index++; 1300 g_io_request(cbp, disk->d_consumer); 1301 } 1302} 1303 1304static void 1305g_mirror_register_request(struct bio *bp) 1306{ 1307 struct g_mirror_softc *sc; 1308 1309 sc = bp->bio_to->geom->softc; 1310 switch (bp->bio_cmd) { 1311 case BIO_READ: 1312 switch (sc->sc_balance) { 1313 case G_MIRROR_BALANCE_LOAD: 1314 g_mirror_request_load(sc, bp); 1315 break; 1316 case G_MIRROR_BALANCE_PREFER: 1317 g_mirror_request_prefer(sc, bp); 1318 break; 1319 case G_MIRROR_BALANCE_ROUND_ROBIN: 1320 g_mirror_request_round_robin(sc, bp); 1321 break; 1322 case G_MIRROR_BALANCE_SPLIT: 1323 g_mirror_request_split(sc, bp); 1324 break; 1325 } 1326 return; 1327 case BIO_WRITE: 1328 case BIO_DELETE: 1329 { 1330 struct g_mirror_disk *disk; 1331 struct g_mirror_disk_sync *sync; 1332 struct bio_queue_head queue; 1333 struct g_consumer *cp; 1334 struct bio *cbp; 1335 1336 if (sc->sc_idle) 1337 g_mirror_unidle(sc); 1338 /* 1339 * Allocate all bios before sending any request, so we can 1340 * return ENOMEM in nice and clean way. 1341 */ 1342 bioq_init(&queue); 1343 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1344 sync = &disk->d_sync; 1345 switch (disk->d_state) { 1346 case G_MIRROR_DISK_STATE_ACTIVE: 1347 break; 1348 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1349 if (bp->bio_offset >= sync->ds_offset) 1350 continue; 1351 else if (bp->bio_offset + bp->bio_length > 1352 sync->ds_offset_done && 1353 (bp->bio_offset < sync->ds_resync || 1354 sync->ds_resync == -1)) { 1355 sync->ds_resync = bp->bio_offset - 1356 (bp->bio_offset % MAXPHYS); 1357 } 1358 break; 1359 default: 1360 continue; 1361 } 1362 cbp = g_clone_bio(bp); 1363 if (cbp == NULL) { 1364 for (cbp = bioq_first(&queue); cbp != NULL; 1365 cbp = bioq_first(&queue)) { 1366 bioq_remove(&queue, cbp); 1367 g_destroy_bio(cbp); 1368 } 1369 if (bp->bio_error == 0) 1370 bp->bio_error = ENOMEM; 1371 g_io_deliver(bp, bp->bio_error); 1372 return; 1373 } 1374 bioq_insert_tail(&queue, cbp); 1375 cbp->bio_done = g_mirror_done; 1376 cp = disk->d_consumer; 1377 cbp->bio_caller1 = cp; 1378 cbp->bio_to = cp->provider; 1379 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1380 ("Consumer %s not opened (r%dw%de%d).", 1381 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1382 } 1383 for (cbp = bioq_first(&queue); cbp != NULL; 1384 cbp = bioq_first(&queue)) { 1385 bioq_remove(&queue, cbp); 1386 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1387 cp = cbp->bio_caller1; 1388 cbp->bio_caller1 = NULL; 1389 cp->index++; 1390 g_io_request(cbp, cp); 1391 } 1392 /* 1393 * Bump syncid on first write. 1394 */ 1395 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1396 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1397 g_topology_lock(); 1398 g_mirror_bump_syncid(sc); 1399 g_topology_unlock(); 1400 } 1401 return; 1402 } 1403 default: 1404 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1405 bp->bio_cmd, sc->sc_name)); 1406 break; 1407 } 1408} 1409 1410static int 1411g_mirror_can_destroy(struct g_mirror_softc *sc) 1412{ 1413 struct g_geom *gp; 1414 struct g_consumer *cp; 1415 1416 g_topology_assert(); 1417 gp = sc->sc_geom; 1418 LIST_FOREACH(cp, &gp->consumer, consumer) { 1419 if (g_mirror_is_busy(sc, cp)) 1420 return (0); 1421 } 1422 gp = sc->sc_sync.ds_geom; 1423 LIST_FOREACH(cp, &gp->consumer, consumer) { 1424 if (g_mirror_is_busy(sc, cp)) 1425 return (0); 1426 } 1427 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1428 sc->sc_name); 1429 return (1); 1430} 1431 1432static int 1433g_mirror_try_destroy(struct g_mirror_softc *sc) 1434{ 1435 1436 g_topology_lock(); 1437 if (!g_mirror_can_destroy(sc)) { 1438 g_topology_unlock(); 1439 return (0); 1440 } 1441 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1442 g_topology_unlock(); 1443 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1444 &sc->sc_worker); 1445 wakeup(&sc->sc_worker); 1446 sc->sc_worker = NULL; 1447 } else { 1448 g_mirror_destroy_device(sc); 1449 g_topology_unlock(); 1450 free(sc, M_MIRROR); 1451 } 1452 return (1); 1453} 1454 1455/* 1456 * Worker thread. 1457 */ 1458static void 1459g_mirror_worker(void *arg) 1460{ 1461 struct g_mirror_softc *sc; 1462 struct g_mirror_disk *disk; 1463 struct g_mirror_disk_sync *sync; 1464 struct g_mirror_event *ep; 1465 struct bio *bp; 1466 u_int nreqs; 1467 1468 sc = arg; 1469 mtx_lock_spin(&sched_lock); 1470 sched_prio(curthread, PRIBIO); 1471 mtx_unlock_spin(&sched_lock); 1472 1473 nreqs = 0; 1474 for (;;) { 1475 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1476 /* 1477 * First take a look at events. 1478 * This is important to handle events before any I/O requests. 1479 */ 1480 ep = g_mirror_event_get(sc); 1481 if (ep != NULL && g_topology_try_lock()) { 1482 g_mirror_event_remove(sc, ep); 1483 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1484 /* Update only device status. */ 1485 G_MIRROR_DEBUG(3, 1486 "Running event for device %s.", 1487 sc->sc_name); 1488 ep->e_error = 0; 1489 g_mirror_update_device(sc, 1); 1490 } else { 1491 /* Update disk status. */ 1492 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1493 g_mirror_get_diskname(ep->e_disk)); 1494 ep->e_error = g_mirror_update_disk(ep->e_disk, 1495 ep->e_state); 1496 if (ep->e_error == 0) 1497 g_mirror_update_device(sc, 0); 1498 } 1499 g_topology_unlock(); 1500 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1501 KASSERT(ep->e_error == 0, 1502 ("Error cannot be handled.")); 1503 g_mirror_event_free(ep); 1504 } else { 1505 ep->e_flags |= G_MIRROR_EVENT_DONE; 1506 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1507 ep); 1508 mtx_lock(&sc->sc_events_mtx); 1509 wakeup(ep); 1510 mtx_unlock(&sc->sc_events_mtx); 1511 } 1512 if ((sc->sc_flags & 1513 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1514 if (g_mirror_try_destroy(sc)) 1515 kthread_exit(0); 1516 } 1517 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1518 continue; 1519 } 1520 /* 1521 * Now I/O requests. 1522 */ 1523 /* Get first request from the queue. */ 1524 mtx_lock(&sc->sc_queue_mtx); 1525 bp = bioq_first(&sc->sc_queue); 1526 if (bp == NULL) { 1527 if (ep != NULL) { 1528 /* 1529 * No I/O requests and topology lock was 1530 * already held? Try again. 1531 */ 1532 mtx_unlock(&sc->sc_queue_mtx); 1533 continue; 1534 } 1535 if ((sc->sc_flags & 1536 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1537 mtx_unlock(&sc->sc_queue_mtx); 1538 if (g_mirror_try_destroy(sc)) 1539 kthread_exit(0); 1540 mtx_lock(&sc->sc_queue_mtx); 1541 } 1542 } 1543 if (sc->sc_sync.ds_ndisks > 0 && 1544 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1545 mtx_unlock(&sc->sc_queue_mtx); 1546 /* 1547 * It is time for synchronization... 1548 */ 1549 nreqs = 0; 1550 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1551 if (disk->d_state != 1552 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1553 continue; 1554 } 1555 sync = &disk->d_sync; 1556 if (sync->ds_offset >= 1557 sc->sc_provider->mediasize) { 1558 continue; 1559 } 1560 if (sync->ds_offset > sync->ds_offset_done) 1561 continue; 1562 if (sync->ds_resync != -1) { 1563 sync->ds_offset = sync->ds_resync; 1564 sync->ds_offset_done = sync->ds_resync; 1565 sync->ds_resync = -1; 1566 } 1567 g_mirror_sync_one(disk); 1568 } 1569 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1570 goto sleep; 1571 } 1572 if (bp == NULL) { 1573 if (g_mirror_check_idle(sc)) { 1574 u_int idletime; 1575 1576 idletime = g_mirror_idletime; 1577 if (idletime == 0) 1578 idletime = 1; 1579 idletime *= hz; 1580 if (msleep(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1581 "m:w1", idletime) == EWOULDBLOCK) { 1582 G_MIRROR_DEBUG(5, "%s: I'm here 3.", 1583 __func__); 1584 /* 1585 * No I/O requests in 'idletime' seconds, 1586 * so mark components as clean. 1587 */ 1588 g_mirror_idle(sc); 1589 } 1590 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1591 } else { 1592 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1593 "m:w2", 0); 1594 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1595 } 1596 continue; 1597 } 1598 nreqs++; 1599 bioq_remove(&sc->sc_queue, bp); 1600 mtx_unlock(&sc->sc_queue_mtx); 1601 1602 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1603 g_mirror_regular_request(bp); 1604 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1605 u_int timeout, sps; 1606 1607 g_mirror_sync_request(bp); 1608sleep: 1609 sps = g_mirror_syncs_per_sec; 1610 if (sps == 0) { 1611 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1612 continue; 1613 } 1614 if (ep != NULL) { 1615 /* 1616 * We have some pending events, don't sleep now. 1617 */ 1618 G_MIRROR_DEBUG(5, "%s: I'm here 7.", __func__); 1619 continue; 1620 } 1621 mtx_lock(&sc->sc_queue_mtx); 1622 if (bioq_first(&sc->sc_queue) != NULL) { 1623 mtx_unlock(&sc->sc_queue_mtx); 1624 G_MIRROR_DEBUG(5, "%s: I'm here 8.", __func__); 1625 continue; 1626 } 1627 timeout = hz / sps; 1628 if (timeout == 0) 1629 timeout = 1; 1630 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w3", 1631 timeout); 1632 } else { 1633 g_mirror_register_request(bp); 1634 } 1635 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1636 } 1637} 1638 1639/* 1640 * Open disk's consumer if needed. 1641 */ 1642static void 1643g_mirror_update_access(struct g_mirror_disk *disk) 1644{ 1645 struct g_provider *pp; 1646 1647 g_topology_assert(); 1648 1649 pp = disk->d_softc->sc_provider; 1650 if (pp == NULL) 1651 return; 1652 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 dirty.", 1656 g_mirror_get_diskname(disk), 1657 disk->d_softc->sc_name); 1658 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1659 } 1660 } else if (pp->acw == 0) { 1661 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1662 G_MIRROR_DEBUG(1, 1663 "Disk %s (device %s) marked as clean.", 1664 g_mirror_get_diskname(disk), 1665 disk->d_softc->sc_name); 1666 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1667 } 1668 } 1669} 1670 1671static void 1672g_mirror_sync_start(struct g_mirror_disk *disk) 1673{ 1674 struct g_mirror_softc *sc; 1675 int error; 1676 1677 g_topology_assert(); 1678 1679 sc = disk->d_softc; 1680 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1681 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1682 sc->sc_state)); 1683 1684 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1685 g_mirror_get_diskname(disk)); 1686 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1687 KASSERT(disk->d_sync.ds_consumer == NULL, 1688 ("Sync consumer already exists (device=%s, disk=%s).", 1689 sc->sc_name, g_mirror_get_diskname(disk))); 1690 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1691 disk->d_sync.ds_consumer->private = disk; 1692 disk->d_sync.ds_consumer->index = 0; 1693 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1694 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1695 disk->d_softc->sc_name, error)); 1696 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1697 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1698 disk->d_softc->sc_name, error)); 1699 disk->d_sync.ds_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1700 sc->sc_sync.ds_ndisks++; 1701} 1702 1703/* 1704 * Stop synchronization process. 1705 * type: 0 - synchronization finished 1706 * 1 - synchronization stopped 1707 */ 1708static void 1709g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1710{ 1711 1712 g_topology_assert(); 1713 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1714 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1715 g_mirror_disk_state2str(disk->d_state))); 1716 if (disk->d_sync.ds_consumer == NULL) 1717 return; 1718 1719 if (type == 0) { 1720 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1721 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1722 } else /* if (type == 1) */ { 1723 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1724 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1725 } 1726 g_mirror_kill_consumer(disk->d_softc, disk->d_sync.ds_consumer); 1727 free(disk->d_sync.ds_data, M_MIRROR); 1728 disk->d_sync.ds_consumer = NULL; 1729 disk->d_softc->sc_sync.ds_ndisks--; 1730 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1731} 1732 1733static void 1734g_mirror_launch_provider(struct g_mirror_softc *sc) 1735{ 1736 struct g_mirror_disk *disk; 1737 struct g_provider *pp; 1738 1739 g_topology_assert(); 1740 1741 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1742 pp->mediasize = sc->sc_mediasize; 1743 pp->sectorsize = sc->sc_sectorsize; 1744 sc->sc_provider = pp; 1745 g_error_provider(pp, 0); 1746 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1747 pp->name); 1748 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1749 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1750 g_mirror_sync_start(disk); 1751 } 1752} 1753 1754static void 1755g_mirror_destroy_provider(struct g_mirror_softc *sc) 1756{ 1757 struct g_mirror_disk *disk; 1758 struct bio *bp; 1759 1760 g_topology_assert(); 1761 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1762 sc->sc_name)); 1763 1764 g_error_provider(sc->sc_provider, ENXIO); 1765 mtx_lock(&sc->sc_queue_mtx); 1766 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1767 bioq_remove(&sc->sc_queue, bp); 1768 g_io_deliver(bp, ENXIO); 1769 } 1770 mtx_unlock(&sc->sc_queue_mtx); 1771 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1772 sc->sc_provider->name); 1773 sc->sc_provider->flags |= G_PF_WITHER; 1774 g_orphan_provider(sc->sc_provider, ENXIO); 1775 sc->sc_provider = NULL; 1776 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1777 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1778 g_mirror_sync_stop(disk, 1); 1779 } 1780} 1781 1782static void 1783g_mirror_go(void *arg) 1784{ 1785 struct g_mirror_softc *sc; 1786 1787 sc = arg; 1788 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1789 g_mirror_event_send(sc, 0, 1790 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1791} 1792 1793static u_int 1794g_mirror_determine_state(struct g_mirror_disk *disk) 1795{ 1796 struct g_mirror_softc *sc; 1797 u_int state; 1798 1799 sc = disk->d_softc; 1800 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1801 if ((disk->d_flags & 1802 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1803 /* Disk does not need synchronization. */ 1804 state = G_MIRROR_DISK_STATE_ACTIVE; 1805 } else { 1806 if ((sc->sc_flags & 1807 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1808 (disk->d_flags & 1809 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1810 /* 1811 * We can start synchronization from 1812 * the stored offset. 1813 */ 1814 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1815 } else { 1816 state = G_MIRROR_DISK_STATE_STALE; 1817 } 1818 } 1819 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1820 /* 1821 * Reset all synchronization data for this disk, 1822 * because if it even was synchronized, it was 1823 * synchronized to disks with different syncid. 1824 */ 1825 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1826 disk->d_sync.ds_offset = 0; 1827 disk->d_sync.ds_offset_done = 0; 1828 disk->d_sync.ds_syncid = sc->sc_syncid; 1829 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1830 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1831 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1832 } else { 1833 state = G_MIRROR_DISK_STATE_STALE; 1834 } 1835 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1836 /* 1837 * Not good, NOT GOOD! 1838 * It means that mirror was started on stale disks 1839 * and more fresh disk just arrive. 1840 * If there were writes, mirror is fucked up, sorry. 1841 * I think the best choice here is don't touch 1842 * this disk and inform the user laudly. 1843 */ 1844 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1845 "disk (%s) arrives!! It will not be connected to the " 1846 "running device.", sc->sc_name, 1847 g_mirror_get_diskname(disk)); 1848 g_mirror_destroy_disk(disk); 1849 state = G_MIRROR_DISK_STATE_NONE; 1850 /* Return immediately, because disk was destroyed. */ 1851 return (state); 1852 } 1853 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1854 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1855 return (state); 1856} 1857 1858/* 1859 * Update device state. 1860 */ 1861static void 1862g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1863{ 1864 struct g_mirror_disk *disk; 1865 u_int state; 1866 1867 g_topology_assert(); 1868 1869 switch (sc->sc_state) { 1870 case G_MIRROR_DEVICE_STATE_STARTING: 1871 { 1872 struct g_mirror_disk *pdisk, *tdisk; 1873 u_int dirty, ndisks, genid, syncid; 1874 1875 KASSERT(sc->sc_provider == NULL, 1876 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1877 /* 1878 * Are we ready? We are, if all disks are connected or 1879 * if we have any disks and 'force' is true. 1880 */ 1881 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1882 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1883 ; 1884 } else if (g_mirror_ndisks(sc, -1) == 0) { 1885 /* 1886 * Disks went down in starting phase, so destroy 1887 * device. 1888 */ 1889 callout_drain(&sc->sc_callout); 1890 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1891 return; 1892 } else { 1893 return; 1894 } 1895 1896 /* 1897 * Activate all disks with the biggest syncid. 1898 */ 1899 if (force) { 1900 /* 1901 * If 'force' is true, we have been called due to 1902 * timeout, so don't bother canceling timeout. 1903 */ 1904 ndisks = 0; 1905 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1906 if ((disk->d_flags & 1907 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1908 ndisks++; 1909 } 1910 } 1911 if (ndisks == 0) { 1912 /* No valid disks found, destroy device. */ 1913 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1914 return; 1915 } 1916 } else { 1917 /* Cancel timeout. */ 1918 callout_drain(&sc->sc_callout); 1919 } 1920 1921 /* 1922 * Find the biggest genid. 1923 */ 1924 genid = 0; 1925 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1926 if (disk->d_genid > genid) 1927 genid = disk->d_genid; 1928 } 1929 sc->sc_genid = genid; 1930 /* 1931 * Remove all disks without the biggest genid. 1932 */ 1933 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 1934 if (disk->d_genid < genid) { 1935 G_MIRROR_DEBUG(0, 1936 "Component %s (device %s) broken, skipping.", 1937 g_mirror_get_diskname(disk), sc->sc_name); 1938 g_mirror_destroy_disk(disk); 1939 } 1940 } 1941 1942 /* 1943 * Find the biggest syncid. 1944 */ 1945 syncid = 0; 1946 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1947 if (disk->d_sync.ds_syncid > syncid) 1948 syncid = disk->d_sync.ds_syncid; 1949 } 1950 1951 /* 1952 * Here we need to look for dirty disks and if all disks 1953 * with the biggest syncid are dirty, we have to choose 1954 * one with the biggest priority and rebuild the rest. 1955 */ 1956 /* 1957 * Find the number of dirty disks with the biggest syncid. 1958 * Find the number of disks with the biggest syncid. 1959 * While here, find a disk with the biggest priority. 1960 */ 1961 dirty = ndisks = 0; 1962 pdisk = NULL; 1963 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1964 if (disk->d_sync.ds_syncid != syncid) 1965 continue; 1966 if ((disk->d_flags & 1967 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1968 continue; 1969 } 1970 ndisks++; 1971 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1972 dirty++; 1973 if (pdisk == NULL || 1974 pdisk->d_priority < disk->d_priority) { 1975 pdisk = disk; 1976 } 1977 } 1978 } 1979 if (dirty == 0) { 1980 /* No dirty disks at all, great. */ 1981 } else if (dirty == ndisks) { 1982 /* 1983 * Force synchronization for all dirty disks except one 1984 * with the biggest priority. 1985 */ 1986 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1987 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1988 "master disk for synchronization.", 1989 g_mirror_get_diskname(pdisk), sc->sc_name); 1990 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1991 if (disk->d_sync.ds_syncid != syncid) 1992 continue; 1993 if ((disk->d_flags & 1994 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1995 continue; 1996 } 1997 KASSERT((disk->d_flags & 1998 G_MIRROR_DISK_FLAG_DIRTY) != 0, 1999 ("Disk %s isn't marked as dirty.", 2000 g_mirror_get_diskname(disk))); 2001 /* Skip the disk with the biggest priority. */ 2002 if (disk == pdisk) 2003 continue; 2004 disk->d_sync.ds_syncid = 0; 2005 } 2006 } else if (dirty < ndisks) { 2007 /* 2008 * Force synchronization for all dirty disks. 2009 * We have some non-dirty disks. 2010 */ 2011 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2012 if (disk->d_sync.ds_syncid != syncid) 2013 continue; 2014 if ((disk->d_flags & 2015 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2016 continue; 2017 } 2018 if ((disk->d_flags & 2019 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2020 continue; 2021 } 2022 disk->d_sync.ds_syncid = 0; 2023 } 2024 } 2025 2026 /* Reset hint. */ 2027 sc->sc_hint = NULL; 2028 sc->sc_syncid = syncid; 2029 if (force) { 2030 /* Remember to bump syncid on first write. */ 2031 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2032 } 2033 state = G_MIRROR_DEVICE_STATE_RUNNING; 2034 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2035 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2036 g_mirror_device_state2str(state)); 2037 sc->sc_state = state; 2038 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2039 state = g_mirror_determine_state(disk); 2040 g_mirror_event_send(disk, state, 2041 G_MIRROR_EVENT_DONTWAIT); 2042 if (state == G_MIRROR_DISK_STATE_STALE) 2043 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2044 } 2045 wakeup(&g_mirror_class); 2046 break; 2047 } 2048 case G_MIRROR_DEVICE_STATE_RUNNING: 2049 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2050 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2051 /* 2052 * No active disks or no disks at all, 2053 * so destroy device. 2054 */ 2055 if (sc->sc_provider != NULL) 2056 g_mirror_destroy_provider(sc); 2057 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2058 break; 2059 } else if (g_mirror_ndisks(sc, 2060 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2061 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2062 /* 2063 * We have active disks, launch provider if it doesn't 2064 * exist. 2065 */ 2066 if (sc->sc_provider == NULL) 2067 g_mirror_launch_provider(sc); 2068 } 2069 /* 2070 * Genid should be bumped immediately, so do it here. 2071 */ 2072 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2073 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2074 g_mirror_bump_genid(sc); 2075 } 2076 break; 2077 default: 2078 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2079 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2080 break; 2081 } 2082} 2083 2084/* 2085 * Update disk state and device state if needed. 2086 */ 2087#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2088 "Disk %s state changed from %s to %s (device %s).", \ 2089 g_mirror_get_diskname(disk), \ 2090 g_mirror_disk_state2str(disk->d_state), \ 2091 g_mirror_disk_state2str(state), sc->sc_name) 2092static int 2093g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2094{ 2095 struct g_mirror_softc *sc; 2096 2097 g_topology_assert(); 2098 2099 sc = disk->d_softc; 2100again: 2101 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2102 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2103 g_mirror_disk_state2str(state)); 2104 switch (state) { 2105 case G_MIRROR_DISK_STATE_NEW: 2106 /* 2107 * Possible scenarios: 2108 * 1. New disk arrive. 2109 */ 2110 /* Previous state should be NONE. */ 2111 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2112 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2113 g_mirror_disk_state2str(disk->d_state))); 2114 DISK_STATE_CHANGED(); 2115 2116 disk->d_state = state; 2117 if (LIST_EMPTY(&sc->sc_disks)) 2118 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2119 else { 2120 struct g_mirror_disk *dp; 2121 2122 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2123 if (disk->d_priority >= dp->d_priority) { 2124 LIST_INSERT_BEFORE(dp, disk, d_next); 2125 dp = NULL; 2126 break; 2127 } 2128 if (LIST_NEXT(dp, d_next) == NULL) 2129 break; 2130 } 2131 if (dp != NULL) 2132 LIST_INSERT_AFTER(dp, disk, d_next); 2133 } 2134 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2135 sc->sc_name, g_mirror_get_diskname(disk)); 2136 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2137 break; 2138 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2139 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2140 g_mirror_device_state2str(sc->sc_state), 2141 g_mirror_get_diskname(disk), 2142 g_mirror_disk_state2str(disk->d_state))); 2143 state = g_mirror_determine_state(disk); 2144 if (state != G_MIRROR_DISK_STATE_NONE) 2145 goto again; 2146 break; 2147 case G_MIRROR_DISK_STATE_ACTIVE: 2148 /* 2149 * Possible scenarios: 2150 * 1. New disk does not need synchronization. 2151 * 2. Synchronization process finished successfully. 2152 */ 2153 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2154 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2155 g_mirror_device_state2str(sc->sc_state), 2156 g_mirror_get_diskname(disk), 2157 g_mirror_disk_state2str(disk->d_state))); 2158 /* Previous state should be NEW or SYNCHRONIZING. */ 2159 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2160 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2161 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2162 g_mirror_disk_state2str(disk->d_state))); 2163 DISK_STATE_CHANGED(); 2164 2165 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2166 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2167 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2168 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2169 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2170 g_mirror_sync_stop(disk, 0); 2171 } 2172 disk->d_state = state; 2173 disk->d_sync.ds_offset = 0; 2174 disk->d_sync.ds_offset_done = 0; 2175 g_mirror_update_access(disk); 2176 g_mirror_update_metadata(disk); 2177 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2178 sc->sc_name, g_mirror_get_diskname(disk)); 2179 break; 2180 case G_MIRROR_DISK_STATE_STALE: 2181 /* 2182 * Possible scenarios: 2183 * 1. Stale disk was connected. 2184 */ 2185 /* Previous state should be NEW. */ 2186 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2187 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2188 g_mirror_disk_state2str(disk->d_state))); 2189 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2190 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2191 g_mirror_device_state2str(sc->sc_state), 2192 g_mirror_get_diskname(disk), 2193 g_mirror_disk_state2str(disk->d_state))); 2194 /* 2195 * STALE state is only possible if device is marked 2196 * NOAUTOSYNC. 2197 */ 2198 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2199 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2200 g_mirror_device_state2str(sc->sc_state), 2201 g_mirror_get_diskname(disk), 2202 g_mirror_disk_state2str(disk->d_state))); 2203 DISK_STATE_CHANGED(); 2204 2205 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2206 disk->d_state = state; 2207 g_mirror_update_metadata(disk); 2208 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2209 sc->sc_name, g_mirror_get_diskname(disk)); 2210 break; 2211 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2212 /* 2213 * Possible scenarios: 2214 * 1. Disk which needs synchronization was connected. 2215 */ 2216 /* Previous state should be NEW. */ 2217 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2218 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2219 g_mirror_disk_state2str(disk->d_state))); 2220 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2221 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2222 g_mirror_device_state2str(sc->sc_state), 2223 g_mirror_get_diskname(disk), 2224 g_mirror_disk_state2str(disk->d_state))); 2225 DISK_STATE_CHANGED(); 2226 2227 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2228 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2229 disk->d_state = state; 2230 if (sc->sc_provider != NULL) { 2231 g_mirror_sync_start(disk); 2232 g_mirror_update_metadata(disk); 2233 } 2234 break; 2235 case G_MIRROR_DISK_STATE_DISCONNECTED: 2236 /* 2237 * Possible scenarios: 2238 * 1. Device wasn't running yet, but disk disappear. 2239 * 2. Disk was active and disapppear. 2240 * 3. Disk disappear during synchronization process. 2241 */ 2242 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2243 /* 2244 * Previous state should be ACTIVE, STALE or 2245 * SYNCHRONIZING. 2246 */ 2247 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2248 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2249 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2250 ("Wrong disk state (%s, %s).", 2251 g_mirror_get_diskname(disk), 2252 g_mirror_disk_state2str(disk->d_state))); 2253 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2254 /* Previous state should be NEW. */ 2255 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2256 ("Wrong disk state (%s, %s).", 2257 g_mirror_get_diskname(disk), 2258 g_mirror_disk_state2str(disk->d_state))); 2259 /* 2260 * Reset bumping syncid if disk disappeared in STARTING 2261 * state. 2262 */ 2263 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2264 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2265#ifdef INVARIANTS 2266 } else { 2267 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2268 sc->sc_name, 2269 g_mirror_device_state2str(sc->sc_state), 2270 g_mirror_get_diskname(disk), 2271 g_mirror_disk_state2str(disk->d_state))); 2272#endif 2273 } 2274 DISK_STATE_CHANGED(); 2275 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2276 sc->sc_name, g_mirror_get_diskname(disk)); 2277 2278 g_mirror_destroy_disk(disk); 2279 break; 2280 case G_MIRROR_DISK_STATE_DESTROY: 2281 { 2282 int error; 2283 2284 error = g_mirror_clear_metadata(disk); 2285 if (error != 0) 2286 return (error); 2287 DISK_STATE_CHANGED(); 2288 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2289 sc->sc_name, g_mirror_get_diskname(disk)); 2290 2291 g_mirror_destroy_disk(disk); 2292 sc->sc_ndisks--; 2293 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2294 g_mirror_update_metadata(disk); 2295 } 2296 break; 2297 } 2298 default: 2299 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2300 break; 2301 } 2302 return (0); 2303} 2304#undef DISK_STATE_CHANGED 2305 2306int 2307g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2308{ 2309 struct g_provider *pp; 2310 u_char *buf; 2311 int error; 2312 2313 g_topology_assert(); 2314 2315 error = g_access(cp, 1, 0, 0); 2316 if (error != 0) 2317 return (error); 2318 pp = cp->provider; 2319 g_topology_unlock(); 2320 /* Metadata are stored on last sector. */ 2321 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2322 &error); 2323 g_topology_lock(); 2324 g_access(cp, -1, 0, 0); 2325 if (error != 0) { 2326 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2327 cp->provider->name, error); 2328 if (buf != NULL) 2329 g_free(buf); 2330 return (error); 2331 } 2332 2333 /* Decode metadata. */ 2334 error = mirror_metadata_decode(buf, md); 2335 g_free(buf); 2336 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2337 return (EINVAL); 2338 if (md->md_version > G_MIRROR_VERSION) { 2339 G_MIRROR_DEBUG(0, 2340 "Kernel module is too old to handle metadata from %s.", 2341 cp->provider->name); 2342 return (EINVAL); 2343 } 2344 if (error != 0) { 2345 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2346 cp->provider->name); 2347 return (error); 2348 } 2349 2350 return (0); 2351} 2352 2353static int 2354g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2355 struct g_mirror_metadata *md) 2356{ 2357 2358 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2359 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2360 pp->name, md->md_did); 2361 return (EEXIST); 2362 } 2363 if (md->md_all != sc->sc_ndisks) { 2364 G_MIRROR_DEBUG(1, 2365 "Invalid '%s' field on disk %s (device %s), skipping.", 2366 "md_all", pp->name, sc->sc_name); 2367 return (EINVAL); 2368 } 2369 if (md->md_slice != sc->sc_slice) { 2370 G_MIRROR_DEBUG(1, 2371 "Invalid '%s' field on disk %s (device %s), skipping.", 2372 "md_slice", pp->name, sc->sc_name); 2373 return (EINVAL); 2374 } 2375 if (md->md_balance != sc->sc_balance) { 2376 G_MIRROR_DEBUG(1, 2377 "Invalid '%s' field on disk %s (device %s), skipping.", 2378 "md_balance", pp->name, sc->sc_name); 2379 return (EINVAL); 2380 } 2381 if (md->md_mediasize != sc->sc_mediasize) { 2382 G_MIRROR_DEBUG(1, 2383 "Invalid '%s' field on disk %s (device %s), skipping.", 2384 "md_mediasize", pp->name, sc->sc_name); 2385 return (EINVAL); 2386 } 2387 if (sc->sc_mediasize > pp->mediasize) { 2388 G_MIRROR_DEBUG(1, 2389 "Invalid size of disk %s (device %s), skipping.", pp->name, 2390 sc->sc_name); 2391 return (EINVAL); 2392 } 2393 if (md->md_sectorsize != sc->sc_sectorsize) { 2394 G_MIRROR_DEBUG(1, 2395 "Invalid '%s' field on disk %s (device %s), skipping.", 2396 "md_sectorsize", pp->name, sc->sc_name); 2397 return (EINVAL); 2398 } 2399 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2400 G_MIRROR_DEBUG(1, 2401 "Invalid sector size of disk %s (device %s), skipping.", 2402 pp->name, sc->sc_name); 2403 return (EINVAL); 2404 } 2405 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2406 G_MIRROR_DEBUG(1, 2407 "Invalid device flags on disk %s (device %s), skipping.", 2408 pp->name, sc->sc_name); 2409 return (EINVAL); 2410 } 2411 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2412 G_MIRROR_DEBUG(1, 2413 "Invalid disk flags on disk %s (device %s), skipping.", 2414 pp->name, sc->sc_name); 2415 return (EINVAL); 2416 } 2417 return (0); 2418} 2419 2420int 2421g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2422 struct g_mirror_metadata *md) 2423{ 2424 struct g_mirror_disk *disk; 2425 int error; 2426 2427 g_topology_assert(); 2428 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2429 2430 error = g_mirror_check_metadata(sc, pp, md); 2431 if (error != 0) 2432 return (error); 2433 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2434 md->md_genid < sc->sc_genid) { 2435 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2436 pp->name, sc->sc_name); 2437 return (EINVAL); 2438 } 2439 disk = g_mirror_init_disk(sc, pp, md, &error); 2440 if (disk == NULL) 2441 return (error); 2442 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2443 G_MIRROR_EVENT_WAIT); 2444 if (error != 0) 2445 return (error); 2446 if (md->md_version < G_MIRROR_VERSION) { 2447 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2448 pp->name, md->md_version, G_MIRROR_VERSION); 2449 g_mirror_update_metadata(disk); 2450 } 2451 return (0); 2452} 2453 2454static int 2455g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2456{ 2457 struct g_mirror_softc *sc; 2458 struct g_mirror_disk *disk; 2459 int dcr, dcw, dce; 2460 2461 g_topology_assert(); 2462 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2463 acw, ace); 2464 2465 dcr = pp->acr + acr; 2466 dcw = pp->acw + acw; 2467 dce = pp->ace + ace; 2468 2469 sc = pp->geom->softc; 2470 if (sc == NULL || LIST_EMPTY(&sc->sc_disks) || 2471 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 2472 if (acr <= 0 && acw <= 0 && ace <= 0) 2473 return (0); 2474 else 2475 return (ENXIO); 2476 } 2477 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2478 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2479 continue; 2480 /* 2481 * Mark disk as dirty on open and unmark on close. 2482 */ 2483 if (pp->acw == 0 && dcw > 0) { 2484 G_MIRROR_DEBUG(1, 2485 "Disk %s (device %s) marked as dirty.", 2486 g_mirror_get_diskname(disk), sc->sc_name); 2487 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2488 g_mirror_update_metadata(disk); 2489 } else if (pp->acw > 0 && dcw == 0) { 2490 G_MIRROR_DEBUG(1, 2491 "Disk %s (device %s) marked as clean.", 2492 g_mirror_get_diskname(disk), sc->sc_name); 2493 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2494 g_mirror_update_metadata(disk); 2495 } 2496 } 2497 return (0); 2498} 2499 2500static struct g_geom * 2501g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2502{ 2503 struct g_mirror_softc *sc; 2504 struct g_geom *gp; 2505 int error, timeout; 2506 2507 g_topology_assert(); 2508 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2509 md->md_mid); 2510 2511 /* One disk is minimum. */ 2512 if (md->md_all < 1) 2513 return (NULL); 2514 /* 2515 * Action geom. 2516 */ 2517 gp = g_new_geomf(mp, "%s", md->md_name); 2518 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2519 gp->start = g_mirror_start; 2520 gp->orphan = g_mirror_orphan; 2521 gp->access = g_mirror_access; 2522 gp->dumpconf = g_mirror_dumpconf; 2523 2524 sc->sc_id = md->md_mid; 2525 sc->sc_slice = md->md_slice; 2526 sc->sc_balance = md->md_balance; 2527 sc->sc_mediasize = md->md_mediasize; 2528 sc->sc_sectorsize = md->md_sectorsize; 2529 sc->sc_ndisks = md->md_all; 2530 sc->sc_flags = md->md_mflags; 2531 sc->sc_bump_id = 0; 2532 sc->sc_idle = 0; 2533 bioq_init(&sc->sc_queue); 2534 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2535 LIST_INIT(&sc->sc_disks); 2536 TAILQ_INIT(&sc->sc_events); 2537 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2538 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2539 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2540 gp->softc = sc; 2541 sc->sc_geom = gp; 2542 sc->sc_provider = NULL; 2543 /* 2544 * Synchronization geom. 2545 */ 2546 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2547 gp->softc = sc; 2548 gp->orphan = g_mirror_orphan; 2549 sc->sc_sync.ds_geom = gp; 2550 sc->sc_sync.ds_ndisks = 0; 2551 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2552 "g_mirror %s", md->md_name); 2553 if (error != 0) { 2554 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2555 sc->sc_name); 2556 g_destroy_geom(sc->sc_sync.ds_geom); 2557 mtx_destroy(&sc->sc_events_mtx); 2558 mtx_destroy(&sc->sc_queue_mtx); 2559 g_destroy_geom(sc->sc_geom); 2560 free(sc, M_MIRROR); 2561 return (NULL); 2562 } 2563 2564 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2565 2566 /* 2567 * Run timeout. 2568 */ 2569 timeout = g_mirror_timeout * hz; 2570 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2571 return (sc->sc_geom); 2572} 2573 2574int 2575g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2576{ 2577 struct g_provider *pp; 2578 2579 g_topology_assert(); 2580 2581 if (sc == NULL) 2582 return (ENXIO); 2583 pp = sc->sc_provider; 2584 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2585 if (force) { 2586 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 2587 "can't be definitely removed.", pp->name); 2588 } else { 2589 G_MIRROR_DEBUG(1, 2590 "Device %s is still open (r%dw%de%d).", pp->name, 2591 pp->acr, pp->acw, pp->ace); 2592 return (EBUSY); 2593 } 2594 } 2595 2596 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2597 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2598 g_topology_unlock(); 2599 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2600 mtx_lock(&sc->sc_queue_mtx); 2601 wakeup(sc); 2602 mtx_unlock(&sc->sc_queue_mtx); 2603 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2604 while (sc->sc_worker != NULL) 2605 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2606 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2607 g_topology_lock(); 2608 g_mirror_destroy_device(sc); 2609 free(sc, M_MIRROR); 2610 return (0); 2611} 2612 2613static void 2614g_mirror_taste_orphan(struct g_consumer *cp) 2615{ 2616 2617 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2618 cp->provider->name)); 2619} 2620 2621static struct g_geom * 2622g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2623{ 2624 struct g_mirror_metadata md; 2625 struct g_mirror_softc *sc; 2626 struct g_consumer *cp; 2627 struct g_geom *gp; 2628 int error; 2629 2630 g_topology_assert(); 2631 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2632 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2633 2634 gp = g_new_geomf(mp, "mirror:taste"); 2635 /* 2636 * This orphan function should be never called. 2637 */ 2638 gp->orphan = g_mirror_taste_orphan; 2639 cp = g_new_consumer(gp); 2640 g_attach(cp, pp); 2641 error = g_mirror_read_metadata(cp, &md); 2642 g_detach(cp); 2643 g_destroy_consumer(cp); 2644 g_destroy_geom(gp); 2645 if (error != 0) 2646 return (NULL); 2647 gp = NULL; 2648 2649 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2650 return (NULL); 2651 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 2652 return (NULL); 2653 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2654 G_MIRROR_DEBUG(0, 2655 "Device %s: provider %s marked as inactive, skipping.", 2656 md.md_name, pp->name); 2657 return (NULL); 2658 } 2659 if (g_mirror_debug >= 2) 2660 mirror_metadata_dump(&md); 2661 2662 /* 2663 * Let's check if device already exists. 2664 */ 2665 sc = NULL; 2666 LIST_FOREACH(gp, &mp->geom, geom) { 2667 sc = gp->softc; 2668 if (sc == NULL) 2669 continue; 2670 if (sc->sc_sync.ds_geom == gp) 2671 continue; 2672 if (strcmp(md.md_name, sc->sc_name) != 0) 2673 continue; 2674 if (md.md_mid != sc->sc_id) { 2675 G_MIRROR_DEBUG(0, "Device %s already configured.", 2676 sc->sc_name); 2677 return (NULL); 2678 } 2679 break; 2680 } 2681 if (gp == NULL) { 2682 gp = g_mirror_create(mp, &md); 2683 if (gp == NULL) { 2684 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2685 md.md_name); 2686 return (NULL); 2687 } 2688 sc = gp->softc; 2689 } 2690 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2691 error = g_mirror_add_disk(sc, pp, &md); 2692 if (error != 0) { 2693 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2694 pp->name, gp->name, error); 2695 if (LIST_EMPTY(&sc->sc_disks)) 2696 g_mirror_destroy(sc, 1); 2697 return (NULL); 2698 } 2699 return (gp); 2700} 2701 2702static int 2703g_mirror_destroy_geom(struct gctl_req *req __unused, 2704 struct g_class *mp __unused, struct g_geom *gp) 2705{ 2706 2707 return (g_mirror_destroy(gp->softc, 0)); 2708} 2709 2710static void 2711g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2712 struct g_consumer *cp, struct g_provider *pp) 2713{ 2714 struct g_mirror_softc *sc; 2715 2716 g_topology_assert(); 2717 2718 sc = gp->softc; 2719 if (sc == NULL) 2720 return; 2721 /* Skip synchronization geom. */ 2722 if (gp == sc->sc_sync.ds_geom) 2723 return; 2724 if (pp != NULL) { 2725 /* Nothing here. */ 2726 } else if (cp != NULL) { 2727 struct g_mirror_disk *disk; 2728 2729 disk = cp->private; 2730 if (disk == NULL) 2731 return; 2732 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2733 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2734 sbuf_printf(sb, "%s<Synchronized>", indent); 2735 if (disk->d_sync.ds_offset_done == 0) 2736 sbuf_printf(sb, "0%%"); 2737 else { 2738 sbuf_printf(sb, "%u%%", 2739 (u_int)((disk->d_sync.ds_offset_done * 100) / 2740 sc->sc_provider->mediasize)); 2741 } 2742 sbuf_printf(sb, "</Synchronized>\n"); 2743 } 2744 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2745 disk->d_sync.ds_syncid); 2746 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 2747 disk->d_genid); 2748 sbuf_printf(sb, "%s<Flags>", indent); 2749 if (disk->d_flags == 0) 2750 sbuf_printf(sb, "NONE"); 2751 else { 2752 int first = 1; 2753 2754#define ADD_FLAG(flag, name) do { \ 2755 if ((disk->d_flags & (flag)) != 0) { \ 2756 if (!first) \ 2757 sbuf_printf(sb, ", "); \ 2758 else \ 2759 first = 0; \ 2760 sbuf_printf(sb, name); \ 2761 } \ 2762} while (0) 2763 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2764 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2765 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2766 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2767 "SYNCHRONIZING"); 2768 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2769#undef ADD_FLAG 2770 } 2771 sbuf_printf(sb, "</Flags>\n"); 2772 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2773 disk->d_priority); 2774 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2775 g_mirror_disk_state2str(disk->d_state)); 2776 } else { 2777 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2778 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2779 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 2780 sbuf_printf(sb, "%s<Flags>", indent); 2781 if (sc->sc_flags == 0) 2782 sbuf_printf(sb, "NONE"); 2783 else { 2784 int first = 1; 2785 2786#define ADD_FLAG(flag, name) do { \ 2787 if ((sc->sc_flags & (flag)) != 0) { \ 2788 if (!first) \ 2789 sbuf_printf(sb, ", "); \ 2790 else \ 2791 first = 0; \ 2792 sbuf_printf(sb, name); \ 2793 } \ 2794} while (0) 2795 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2796#undef ADD_FLAG 2797 } 2798 sbuf_printf(sb, "</Flags>\n"); 2799 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2800 (u_int)sc->sc_slice); 2801 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2802 balance_name(sc->sc_balance)); 2803 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2804 sc->sc_ndisks); 2805 sbuf_printf(sb, "%s<State>", indent); 2806 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2807 sbuf_printf(sb, "%s", "STARTING"); 2808 else if (sc->sc_ndisks == 2809 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 2810 sbuf_printf(sb, "%s", "COMPLETE"); 2811 else 2812 sbuf_printf(sb, "%s", "DEGRADED"); 2813 sbuf_printf(sb, "</State>\n"); 2814 } 2815} 2816 2817static void 2818g_mirror_shutdown(void *arg, int howto) 2819{ 2820 struct g_class *mp; 2821 struct g_geom *gp, *gp2; 2822 2823 mp = arg; 2824 DROP_GIANT(); 2825 g_topology_lock(); 2826 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2827 if (gp->softc == NULL) 2828 continue; 2829 g_mirror_destroy(gp->softc, 1); 2830 } 2831 g_topology_unlock(); 2832 PICKUP_GIANT(); 2833#if 0 2834 tsleep(&gp, PRIBIO, "m:shutdown", hz * 20); 2835#endif 2836} 2837 2838static void 2839g_mirror_init(struct g_class *mp) 2840{ 2841 2842 g_mirror_ehtag = EVENTHANDLER_REGISTER(shutdown_post_sync, 2843 g_mirror_shutdown, mp, SHUTDOWN_PRI_FIRST); 2844 if (g_mirror_ehtag == NULL) 2845 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 2846} 2847 2848static void 2849g_mirror_fini(struct g_class *mp) 2850{ 2851 2852 if (g_mirror_ehtag == NULL) 2853 return; 2854 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_ehtag); 2855} 2856 2857static int 2858g_mirror_can_go(void) 2859{ 2860 struct g_mirror_softc *sc; 2861 struct g_geom *gp; 2862 struct g_provider *pp; 2863 int can_go; 2864 2865 DROP_GIANT(); 2866 can_go = 1; 2867 g_topology_lock(); 2868 LIST_FOREACH(gp, &g_mirror_class.geom, geom) { 2869 sc = gp->softc; 2870 if (sc == NULL) { 2871 can_go = 0; 2872 break; 2873 } 2874 pp = sc->sc_provider; 2875 if (pp == NULL || pp->error != 0) { 2876 can_go = 0; 2877 break; 2878 } 2879 } 2880 g_topology_unlock(); 2881 PICKUP_GIANT(); 2882 return (can_go); 2883} 2884 2885static void 2886g_mirror_rootwait(void) 2887{ 2888 2889 /* 2890 * HACK: Wait for GEOM, because g_mirror_rootwait() can be called, 2891 * HACK: before we get providers for tasting. 2892 */ 2893 tsleep(&g_mirror_class, PRIBIO, "mroot", hz * 3); 2894 /* 2895 * Wait for mirrors in degraded state. 2896 */ 2897 for (;;) { 2898 if (g_mirror_can_go()) 2899 break; 2900 tsleep(&g_mirror_class, PRIBIO, "mroot", hz); 2901 } 2902} 2903 2904SYSINIT(g_mirror_root, SI_SUB_RAID, SI_ORDER_FIRST, g_mirror_rootwait, NULL) 2905 2906DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2907