g_mirror.c revision 146110
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 146110 2005-05-11 18:13:51Z 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 <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 struct g_consumer *cp; 379 int error; 380 381 g_topology_assert(); 382 KASSERT(disk->d_consumer == NULL, 383 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 384 385 cp = g_new_consumer(disk->d_softc->sc_geom); 386 error = g_attach(cp, pp); 387 if (error != 0) { 388 g_destroy_consumer(cp); 389 return (error); 390 } 391 error = g_access(cp, 1, 1, 1); 392 if (error != 0) { 393 g_detach(cp); 394 g_destroy_consumer(cp); 395 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 396 pp->name, error); 397 return (error); 398 } 399 disk->d_consumer = cp; 400 disk->d_consumer->private = disk; 401 disk->d_consumer->index = 0; 402 403 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 404 return (0); 405} 406 407static void 408g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 409{ 410 411 g_topology_assert(); 412 413 if (cp == NULL) 414 return; 415 if (cp->provider != NULL) 416 g_mirror_kill_consumer(sc, cp); 417 else 418 g_destroy_consumer(cp); 419} 420 421/* 422 * Initialize disk. This means allocate memory, create consumer, attach it 423 * to the provider and open access (r1w1e1) to it. 424 */ 425static struct g_mirror_disk * 426g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 427 struct g_mirror_metadata *md, int *errorp) 428{ 429 struct g_mirror_disk *disk; 430 int error; 431 432 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 433 if (disk == NULL) { 434 error = ENOMEM; 435 goto fail; 436 } 437 disk->d_softc = sc; 438 error = g_mirror_connect_disk(disk, pp); 439 if (error != 0) 440 goto fail; 441 disk->d_id = md->md_did; 442 disk->d_state = G_MIRROR_DISK_STATE_NONE; 443 disk->d_priority = md->md_priority; 444 disk->d_delay.sec = 0; 445 disk->d_delay.frac = 0; 446 binuptime(&disk->d_last_used); 447 disk->d_flags = md->md_dflags; 448 if (md->md_provider[0] != '\0') 449 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 450 disk->d_sync.ds_consumer = NULL; 451 disk->d_sync.ds_offset = md->md_sync_offset; 452 disk->d_sync.ds_offset_done = md->md_sync_offset; 453 disk->d_sync.ds_resync = -1; 454 disk->d_genid = md->md_genid; 455 disk->d_sync.ds_syncid = md->md_syncid; 456 if (errorp != NULL) 457 *errorp = 0; 458 return (disk); 459fail: 460 if (errorp != NULL) 461 *errorp = error; 462 if (disk != NULL) 463 free(disk, M_MIRROR); 464 return (NULL); 465} 466 467static void 468g_mirror_destroy_disk(struct g_mirror_disk *disk) 469{ 470 struct g_mirror_softc *sc; 471 472 g_topology_assert(); 473 474 LIST_REMOVE(disk, d_next); 475 g_mirror_event_cancel(disk); 476 sc = disk->d_softc; 477 if (sc->sc_hint == disk) 478 sc->sc_hint = NULL; 479 switch (disk->d_state) { 480 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 481 g_mirror_sync_stop(disk, 1); 482 /* FALLTHROUGH */ 483 case G_MIRROR_DISK_STATE_NEW: 484 case G_MIRROR_DISK_STATE_STALE: 485 case G_MIRROR_DISK_STATE_ACTIVE: 486 g_mirror_disconnect_consumer(sc, disk->d_consumer); 487 free(disk, M_MIRROR); 488 break; 489 default: 490 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 491 g_mirror_get_diskname(disk), 492 g_mirror_disk_state2str(disk->d_state))); 493 } 494} 495 496static void 497g_mirror_destroy_device(struct g_mirror_softc *sc) 498{ 499 struct g_mirror_disk *disk; 500 struct g_mirror_event *ep; 501 struct g_geom *gp; 502 struct g_consumer *cp, *tmpcp; 503 504 g_topology_assert(); 505 506 gp = sc->sc_geom; 507 if (sc->sc_provider != NULL) 508 g_mirror_destroy_provider(sc); 509 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 510 disk = LIST_FIRST(&sc->sc_disks)) { 511 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 512 g_mirror_update_metadata(disk); 513 g_mirror_destroy_disk(disk); 514 } 515 while ((ep = g_mirror_event_get(sc)) != NULL) { 516 g_mirror_event_remove(sc, ep); 517 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 518 g_mirror_event_free(ep); 519 else { 520 ep->e_error = ECANCELED; 521 ep->e_flags |= G_MIRROR_EVENT_DONE; 522 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 523 mtx_lock(&sc->sc_events_mtx); 524 wakeup(ep); 525 mtx_unlock(&sc->sc_events_mtx); 526 } 527 } 528 callout_drain(&sc->sc_callout); 529 gp->softc = NULL; 530 531 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 532 g_mirror_disconnect_consumer(sc, cp); 533 } 534 sc->sc_sync.ds_geom->softc = NULL; 535 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 536 mtx_destroy(&sc->sc_queue_mtx); 537 mtx_destroy(&sc->sc_events_mtx); 538 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 539 g_wither_geom(gp, ENXIO); 540} 541 542static void 543g_mirror_orphan(struct g_consumer *cp) 544{ 545 struct g_mirror_disk *disk; 546 547 g_topology_assert(); 548 549 disk = cp->private; 550 if (disk == NULL) 551 return; 552 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 553 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 554 G_MIRROR_EVENT_DONTWAIT); 555} 556 557/* 558 * Function should return the next active disk on the list. 559 * It is possible that it will be the same disk as given. 560 * If there are no active disks on list, NULL is returned. 561 */ 562static __inline struct g_mirror_disk * 563g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 564{ 565 struct g_mirror_disk *dp; 566 567 for (dp = LIST_NEXT(disk, d_next); dp != disk; 568 dp = LIST_NEXT(dp, d_next)) { 569 if (dp == NULL) 570 dp = LIST_FIRST(&sc->sc_disks); 571 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 572 break; 573 } 574 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 575 return (NULL); 576 return (dp); 577} 578 579static struct g_mirror_disk * 580g_mirror_get_disk(struct g_mirror_softc *sc) 581{ 582 struct g_mirror_disk *disk; 583 584 if (sc->sc_hint == NULL) { 585 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 586 if (sc->sc_hint == NULL) 587 return (NULL); 588 } 589 disk = sc->sc_hint; 590 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 591 disk = g_mirror_find_next(sc, disk); 592 if (disk == NULL) 593 return (NULL); 594 } 595 sc->sc_hint = g_mirror_find_next(sc, disk); 596 return (disk); 597} 598 599static int 600g_mirror_write_metadata(struct g_mirror_disk *disk, 601 struct g_mirror_metadata *md) 602{ 603 struct g_mirror_softc *sc; 604 struct g_consumer *cp; 605 off_t offset, length; 606 u_char *sector; 607 int error = 0; 608 609 g_topology_assert(); 610 611 sc = disk->d_softc; 612 cp = disk->d_consumer; 613 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 614 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 615 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 616 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 617 cp->acw, cp->ace)); 618 length = cp->provider->sectorsize; 619 offset = cp->provider->mediasize - length; 620 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 621 if (md != NULL) 622 mirror_metadata_encode(md, sector); 623 g_topology_unlock(); 624 error = g_write_data(cp, offset, sector, length); 625 g_topology_lock(); 626 free(sector, M_MIRROR); 627 if (error != 0) { 628 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_GENID; 629 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 630 G_MIRROR_EVENT_DONTWAIT); 631 } 632 return (error); 633} 634 635static int 636g_mirror_clear_metadata(struct g_mirror_disk *disk) 637{ 638 int error; 639 640 g_topology_assert(); 641 error = g_mirror_write_metadata(disk, NULL); 642 if (error == 0) { 643 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 644 g_mirror_get_diskname(disk)); 645 } else { 646 G_MIRROR_DEBUG(0, 647 "Cannot clear metadata on disk %s (error=%d).", 648 g_mirror_get_diskname(disk), error); 649 } 650 return (error); 651} 652 653void 654g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 655 struct g_mirror_metadata *md) 656{ 657 658 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 659 md->md_version = G_MIRROR_VERSION; 660 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 661 md->md_mid = sc->sc_id; 662 md->md_all = sc->sc_ndisks; 663 md->md_slice = sc->sc_slice; 664 md->md_balance = sc->sc_balance; 665 md->md_genid = sc->sc_genid; 666 md->md_mediasize = sc->sc_mediasize; 667 md->md_sectorsize = sc->sc_sectorsize; 668 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 669 bzero(md->md_provider, sizeof(md->md_provider)); 670 if (disk == NULL) { 671 md->md_did = arc4random(); 672 md->md_priority = 0; 673 md->md_syncid = 0; 674 md->md_dflags = 0; 675 md->md_sync_offset = 0; 676 md->md_provsize = 0; 677 } else { 678 md->md_did = disk->d_id; 679 md->md_priority = disk->d_priority; 680 md->md_syncid = disk->d_sync.ds_syncid; 681 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 682 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 683 md->md_sync_offset = disk->d_sync.ds_offset_done; 684 else 685 md->md_sync_offset = 0; 686 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 687 strlcpy(md->md_provider, 688 disk->d_consumer->provider->name, 689 sizeof(md->md_provider)); 690 } 691 md->md_provsize = disk->d_consumer->provider->mediasize; 692 } 693} 694 695void 696g_mirror_update_metadata(struct g_mirror_disk *disk) 697{ 698 struct g_mirror_metadata md; 699 int error; 700 701 g_topology_assert(); 702 g_mirror_fill_metadata(disk->d_softc, disk, &md); 703 error = g_mirror_write_metadata(disk, &md); 704 if (error == 0) { 705 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 706 g_mirror_get_diskname(disk)); 707 } else { 708 G_MIRROR_DEBUG(0, 709 "Cannot update metadata on disk %s (error=%d).", 710 g_mirror_get_diskname(disk), error); 711 } 712} 713 714static void 715g_mirror_bump_syncid(struct g_mirror_softc *sc) 716{ 717 struct g_mirror_disk *disk; 718 719 g_topology_assert(); 720 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 721 ("%s called with no active disks (device=%s).", __func__, 722 sc->sc_name)); 723 724 sc->sc_syncid++; 725 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 726 sc->sc_syncid); 727 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 728 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 729 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 730 disk->d_sync.ds_syncid = sc->sc_syncid; 731 g_mirror_update_metadata(disk); 732 } 733 } 734} 735 736static void 737g_mirror_bump_genid(struct g_mirror_softc *sc) 738{ 739 struct g_mirror_disk *disk; 740 741 g_topology_assert(); 742 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 743 ("%s called with no active disks (device=%s).", __func__, 744 sc->sc_name)); 745 746 sc->sc_genid++; 747 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 748 sc->sc_genid); 749 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 750 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 751 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 752 disk->d_genid = sc->sc_genid; 753 g_mirror_update_metadata(disk); 754 } 755 } 756} 757 758static void 759g_mirror_idle(struct g_mirror_softc *sc) 760{ 761 struct g_mirror_disk *disk; 762 763 if (sc->sc_provider == NULL || sc->sc_provider->acw == 0) 764 return; 765 sc->sc_idle = 1; 766 g_topology_lock(); 767 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 768 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 769 continue; 770 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 771 g_mirror_get_diskname(disk), sc->sc_name); 772 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 773 g_mirror_update_metadata(disk); 774 } 775 g_topology_unlock(); 776} 777 778static void 779g_mirror_unidle(struct g_mirror_softc *sc) 780{ 781 struct g_mirror_disk *disk; 782 783 sc->sc_idle = 0; 784 g_topology_lock(); 785 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 786 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 787 continue; 788 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 789 g_mirror_get_diskname(disk), sc->sc_name); 790 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 791 g_mirror_update_metadata(disk); 792 } 793 g_topology_unlock(); 794} 795 796/* 797 * Return 1 if we should check if mirror is idling. 798 */ 799static int 800g_mirror_check_idle(struct g_mirror_softc *sc) 801{ 802 struct g_mirror_disk *disk; 803 804 if (sc->sc_idle) 805 return (0); 806 if (sc->sc_provider != NULL && sc->sc_provider->acw == 0) 807 return (0); 808 /* 809 * Check if there are no in-flight requests. 810 */ 811 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 812 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 813 continue; 814 if (disk->d_consumer->index > 0) 815 return (0); 816 } 817 return (1); 818} 819 820static __inline int 821bintime_cmp(struct bintime *bt1, struct bintime *bt2) 822{ 823 824 if (bt1->sec < bt2->sec) 825 return (-1); 826 else if (bt1->sec > bt2->sec) 827 return (1); 828 if (bt1->frac < bt2->frac) 829 return (-1); 830 else if (bt1->frac > bt2->frac) 831 return (1); 832 return (0); 833} 834 835static void 836g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 837{ 838 839 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 840 return; 841 binuptime(&disk->d_delay); 842 bintime_sub(&disk->d_delay, &bp->bio_t0); 843} 844 845static void 846g_mirror_done(struct bio *bp) 847{ 848 struct g_mirror_softc *sc; 849 850 sc = bp->bio_from->geom->softc; 851 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 852 mtx_lock(&sc->sc_queue_mtx); 853 bioq_disksort(&sc->sc_queue, bp); 854 wakeup(sc); 855 mtx_unlock(&sc->sc_queue_mtx); 856} 857 858static void 859g_mirror_regular_request(struct bio *bp) 860{ 861 struct g_mirror_softc *sc; 862 struct g_mirror_disk *disk; 863 struct bio *pbp; 864 865 g_topology_assert_not(); 866 867 bp->bio_from->index--; 868 pbp = bp->bio_parent; 869 sc = pbp->bio_to->geom->softc; 870 disk = bp->bio_from->private; 871 if (disk == NULL) { 872 g_topology_lock(); 873 g_mirror_kill_consumer(sc, bp->bio_from); 874 g_topology_unlock(); 875 } else { 876 g_mirror_update_delay(disk, bp); 877 } 878 879 pbp->bio_inbed++; 880 KASSERT(pbp->bio_inbed <= pbp->bio_children, 881 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 882 pbp->bio_children)); 883 if (bp->bio_error == 0 && pbp->bio_error == 0) { 884 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 885 g_destroy_bio(bp); 886 if (pbp->bio_children == pbp->bio_inbed) { 887 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 888 pbp->bio_completed = pbp->bio_length; 889 g_io_deliver(pbp, pbp->bio_error); 890 } 891 return; 892 } else if (bp->bio_error != 0) { 893 if (pbp->bio_error == 0) 894 pbp->bio_error = bp->bio_error; 895 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 896 bp->bio_error); 897 if (disk != NULL) { 898 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 899 g_mirror_event_send(disk, 900 G_MIRROR_DISK_STATE_DISCONNECTED, 901 G_MIRROR_EVENT_DONTWAIT); 902 } 903 switch (pbp->bio_cmd) { 904 case BIO_DELETE: 905 case BIO_WRITE: 906 pbp->bio_inbed--; 907 pbp->bio_children--; 908 break; 909 } 910 } 911 g_destroy_bio(bp); 912 913 switch (pbp->bio_cmd) { 914 case BIO_READ: 915 if (pbp->bio_children == pbp->bio_inbed) { 916 pbp->bio_error = 0; 917 mtx_lock(&sc->sc_queue_mtx); 918 bioq_disksort(&sc->sc_queue, pbp); 919 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 920 wakeup(sc); 921 mtx_unlock(&sc->sc_queue_mtx); 922 } 923 break; 924 case BIO_DELETE: 925 case BIO_WRITE: 926 if (pbp->bio_children == 0) { 927 /* 928 * All requests failed. 929 */ 930 } else if (pbp->bio_inbed < pbp->bio_children) { 931 /* Do nothing. */ 932 break; 933 } else if (pbp->bio_children == pbp->bio_inbed) { 934 /* Some requests succeeded. */ 935 pbp->bio_error = 0; 936 pbp->bio_completed = pbp->bio_length; 937 } 938 g_io_deliver(pbp, pbp->bio_error); 939 break; 940 default: 941 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 942 break; 943 } 944} 945 946static void 947g_mirror_sync_done(struct bio *bp) 948{ 949 struct g_mirror_softc *sc; 950 951 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 952 sc = bp->bio_from->geom->softc; 953 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 954 mtx_lock(&sc->sc_queue_mtx); 955 bioq_disksort(&sc->sc_queue, bp); 956 wakeup(sc); 957 mtx_unlock(&sc->sc_queue_mtx); 958} 959 960static void 961g_mirror_start(struct bio *bp) 962{ 963 struct g_mirror_softc *sc; 964 965 sc = bp->bio_to->geom->softc; 966 /* 967 * If sc == NULL or there are no valid disks, provider's error 968 * should be set and g_mirror_start() should not be called at all. 969 */ 970 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 971 ("Provider's error should be set (error=%d)(mirror=%s).", 972 bp->bio_to->error, bp->bio_to->name)); 973 G_MIRROR_LOGREQ(3, bp, "Request received."); 974 975 switch (bp->bio_cmd) { 976 case BIO_READ: 977 case BIO_WRITE: 978 case BIO_DELETE: 979 break; 980 case BIO_GETATTR: 981 default: 982 g_io_deliver(bp, EOPNOTSUPP); 983 return; 984 } 985 mtx_lock(&sc->sc_queue_mtx); 986 bioq_disksort(&sc->sc_queue, bp); 987 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 988 wakeup(sc); 989 mtx_unlock(&sc->sc_queue_mtx); 990} 991 992/* 993 * Send one synchronization request. 994 */ 995static void 996g_mirror_sync_one(struct g_mirror_disk *disk) 997{ 998 struct g_mirror_softc *sc; 999 struct bio *bp; 1000 1001 sc = disk->d_softc; 1002 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1003 ("Disk %s is not marked for synchronization.", 1004 g_mirror_get_diskname(disk))); 1005 1006 bp = g_new_bio(); 1007 if (bp == NULL) 1008 return; 1009 bp->bio_parent = NULL; 1010 bp->bio_cmd = BIO_READ; 1011 bp->bio_offset = disk->d_sync.ds_offset; 1012 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1013 bp->bio_cflags = 0; 1014 bp->bio_done = g_mirror_sync_done; 1015 bp->bio_data = disk->d_sync.ds_data; 1016 if (bp->bio_data == NULL) { 1017 g_destroy_bio(bp); 1018 return; 1019 } 1020 disk->d_sync.ds_offset += bp->bio_length; 1021 bp->bio_to = sc->sc_provider; 1022 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1023 disk->d_sync.ds_consumer->index++; 1024 g_io_request(bp, disk->d_sync.ds_consumer); 1025} 1026 1027static void 1028g_mirror_sync_request(struct bio *bp) 1029{ 1030 struct g_mirror_softc *sc; 1031 struct g_mirror_disk *disk; 1032 1033 bp->bio_from->index--; 1034 sc = bp->bio_from->geom->softc; 1035 disk = bp->bio_from->private; 1036 if (disk == NULL) { 1037 g_topology_lock(); 1038 g_mirror_kill_consumer(sc, bp->bio_from); 1039 g_topology_unlock(); 1040 g_destroy_bio(bp); 1041 return; 1042 } 1043 1044 /* 1045 * Synchronization request. 1046 */ 1047 switch (bp->bio_cmd) { 1048 case BIO_READ: 1049 { 1050 struct g_consumer *cp; 1051 1052 if (bp->bio_error != 0) { 1053 G_MIRROR_LOGREQ(0, bp, 1054 "Synchronization request failed (error=%d).", 1055 bp->bio_error); 1056 g_destroy_bio(bp); 1057 return; 1058 } 1059 G_MIRROR_LOGREQ(3, bp, 1060 "Synchronization request half-finished."); 1061 bp->bio_cmd = BIO_WRITE; 1062 bp->bio_cflags = 0; 1063 cp = disk->d_consumer; 1064 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1065 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1066 cp->acr, cp->acw, cp->ace)); 1067 cp->index++; 1068 g_io_request(bp, cp); 1069 return; 1070 } 1071 case BIO_WRITE: 1072 { 1073 struct g_mirror_disk_sync *sync; 1074 1075 if (bp->bio_error != 0) { 1076 G_MIRROR_LOGREQ(0, bp, 1077 "Synchronization request failed (error=%d).", 1078 bp->bio_error); 1079 g_destroy_bio(bp); 1080 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1081 g_mirror_event_send(disk, 1082 G_MIRROR_DISK_STATE_DISCONNECTED, 1083 G_MIRROR_EVENT_DONTWAIT); 1084 return; 1085 } 1086 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1087 sync = &disk->d_sync; 1088 sync->ds_offset_done = bp->bio_offset + bp->bio_length; 1089 g_destroy_bio(bp); 1090 if (sync->ds_resync != -1) 1091 break; 1092 if (sync->ds_offset_done == sc->sc_provider->mediasize) { 1093 /* 1094 * Disk up-to-date, activate it. 1095 */ 1096 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1097 G_MIRROR_EVENT_DONTWAIT); 1098 return; 1099 } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) { 1100 /* 1101 * Update offset_done on every 100 blocks. 1102 * XXX: This should be configurable. 1103 */ 1104 g_topology_lock(); 1105 g_mirror_update_metadata(disk); 1106 g_topology_unlock(); 1107 } 1108 return; 1109 } 1110 default: 1111 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1112 bp->bio_cmd, sc->sc_name)); 1113 break; 1114 } 1115} 1116 1117static void 1118g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1119{ 1120 struct g_mirror_disk *disk; 1121 struct g_consumer *cp; 1122 struct bio *cbp; 1123 1124 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1125 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1126 break; 1127 } 1128 if (disk == NULL) { 1129 if (bp->bio_error == 0) 1130 bp->bio_error = ENXIO; 1131 g_io_deliver(bp, bp->bio_error); 1132 return; 1133 } 1134 cbp = g_clone_bio(bp); 1135 if (cbp == NULL) { 1136 if (bp->bio_error == 0) 1137 bp->bio_error = ENOMEM; 1138 g_io_deliver(bp, bp->bio_error); 1139 return; 1140 } 1141 /* 1142 * Fill in the component buf structure. 1143 */ 1144 cp = disk->d_consumer; 1145 cbp->bio_done = g_mirror_done; 1146 cbp->bio_to = cp->provider; 1147 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1148 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1149 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1150 cp->acw, cp->ace)); 1151 cp->index++; 1152 g_io_request(cbp, cp); 1153} 1154 1155static void 1156g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1157{ 1158 struct g_mirror_disk *disk; 1159 struct g_consumer *cp; 1160 struct bio *cbp; 1161 1162 disk = g_mirror_get_disk(sc); 1163 if (disk == NULL) { 1164 if (bp->bio_error == 0) 1165 bp->bio_error = ENXIO; 1166 g_io_deliver(bp, bp->bio_error); 1167 return; 1168 } 1169 cbp = g_clone_bio(bp); 1170 if (cbp == NULL) { 1171 if (bp->bio_error == 0) 1172 bp->bio_error = ENOMEM; 1173 g_io_deliver(bp, bp->bio_error); 1174 return; 1175 } 1176 /* 1177 * Fill in the component buf structure. 1178 */ 1179 cp = disk->d_consumer; 1180 cbp->bio_done = g_mirror_done; 1181 cbp->bio_to = cp->provider; 1182 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1183 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1184 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1185 cp->acw, cp->ace)); 1186 cp->index++; 1187 g_io_request(cbp, cp); 1188} 1189 1190static void 1191g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1192{ 1193 struct g_mirror_disk *disk, *dp; 1194 struct g_consumer *cp; 1195 struct bio *cbp; 1196 struct bintime curtime; 1197 1198 binuptime(&curtime); 1199 /* 1200 * Find a disk which the smallest load. 1201 */ 1202 disk = NULL; 1203 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1204 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1205 continue; 1206 /* If disk wasn't used for more than 2 sec, use it. */ 1207 if (curtime.sec - dp->d_last_used.sec >= 2) { 1208 disk = dp; 1209 break; 1210 } 1211 if (disk == NULL || 1212 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1213 disk = dp; 1214 } 1215 } 1216 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1217 cbp = g_clone_bio(bp); 1218 if (cbp == NULL) { 1219 if (bp->bio_error == 0) 1220 bp->bio_error = ENOMEM; 1221 g_io_deliver(bp, bp->bio_error); 1222 return; 1223 } 1224 /* 1225 * Fill in the component buf structure. 1226 */ 1227 cp = disk->d_consumer; 1228 cbp->bio_done = g_mirror_done; 1229 cbp->bio_to = cp->provider; 1230 binuptime(&disk->d_last_used); 1231 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1232 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1233 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1234 cp->acw, cp->ace)); 1235 cp->index++; 1236 g_io_request(cbp, cp); 1237} 1238 1239static void 1240g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1241{ 1242 struct bio_queue_head queue; 1243 struct g_mirror_disk *disk; 1244 struct g_consumer *cp; 1245 struct bio *cbp; 1246 off_t left, mod, offset, slice; 1247 u_char *data; 1248 u_int ndisks; 1249 1250 if (bp->bio_length <= sc->sc_slice) { 1251 g_mirror_request_round_robin(sc, bp); 1252 return; 1253 } 1254 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1255 slice = bp->bio_length / ndisks; 1256 mod = slice % sc->sc_provider->sectorsize; 1257 if (mod != 0) 1258 slice += sc->sc_provider->sectorsize - mod; 1259 /* 1260 * Allocate all bios before sending any request, so we can 1261 * return ENOMEM in nice and clean way. 1262 */ 1263 left = bp->bio_length; 1264 offset = bp->bio_offset; 1265 data = bp->bio_data; 1266 bioq_init(&queue); 1267 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1268 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1269 continue; 1270 cbp = g_clone_bio(bp); 1271 if (cbp == NULL) { 1272 for (cbp = bioq_first(&queue); cbp != NULL; 1273 cbp = bioq_first(&queue)) { 1274 bioq_remove(&queue, cbp); 1275 g_destroy_bio(cbp); 1276 } 1277 if (bp->bio_error == 0) 1278 bp->bio_error = ENOMEM; 1279 g_io_deliver(bp, bp->bio_error); 1280 return; 1281 } 1282 bioq_insert_tail(&queue, cbp); 1283 cbp->bio_done = g_mirror_done; 1284 cbp->bio_caller1 = disk; 1285 cbp->bio_to = disk->d_consumer->provider; 1286 cbp->bio_offset = offset; 1287 cbp->bio_data = data; 1288 cbp->bio_length = MIN(left, slice); 1289 left -= cbp->bio_length; 1290 if (left == 0) 1291 break; 1292 offset += cbp->bio_length; 1293 data += cbp->bio_length; 1294 } 1295 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1296 bioq_remove(&queue, cbp); 1297 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1298 disk = cbp->bio_caller1; 1299 cbp->bio_caller1 = NULL; 1300 cp = disk->d_consumer; 1301 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1302 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1303 cp->acr, cp->acw, cp->ace)); 1304 disk->d_consumer->index++; 1305 g_io_request(cbp, disk->d_consumer); 1306 } 1307} 1308 1309static void 1310g_mirror_register_request(struct bio *bp) 1311{ 1312 struct g_mirror_softc *sc; 1313 1314 sc = bp->bio_to->geom->softc; 1315 switch (bp->bio_cmd) { 1316 case BIO_READ: 1317 switch (sc->sc_balance) { 1318 case G_MIRROR_BALANCE_LOAD: 1319 g_mirror_request_load(sc, bp); 1320 break; 1321 case G_MIRROR_BALANCE_PREFER: 1322 g_mirror_request_prefer(sc, bp); 1323 break; 1324 case G_MIRROR_BALANCE_ROUND_ROBIN: 1325 g_mirror_request_round_robin(sc, bp); 1326 break; 1327 case G_MIRROR_BALANCE_SPLIT: 1328 g_mirror_request_split(sc, bp); 1329 break; 1330 } 1331 return; 1332 case BIO_WRITE: 1333 case BIO_DELETE: 1334 { 1335 struct g_mirror_disk *disk; 1336 struct g_mirror_disk_sync *sync; 1337 struct bio_queue_head queue; 1338 struct g_consumer *cp; 1339 struct bio *cbp; 1340 1341 if (sc->sc_idle) 1342 g_mirror_unidle(sc); 1343 /* 1344 * Allocate all bios before sending any request, so we can 1345 * return ENOMEM in nice and clean way. 1346 */ 1347 bioq_init(&queue); 1348 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1349 sync = &disk->d_sync; 1350 switch (disk->d_state) { 1351 case G_MIRROR_DISK_STATE_ACTIVE: 1352 break; 1353 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1354 if (bp->bio_offset >= sync->ds_offset) 1355 continue; 1356 else if (bp->bio_offset + bp->bio_length > 1357 sync->ds_offset_done && 1358 (bp->bio_offset < sync->ds_resync || 1359 sync->ds_resync == -1)) { 1360 sync->ds_resync = bp->bio_offset - 1361 (bp->bio_offset % MAXPHYS); 1362 } 1363 break; 1364 default: 1365 continue; 1366 } 1367 cbp = g_clone_bio(bp); 1368 if (cbp == NULL) { 1369 for (cbp = bioq_first(&queue); cbp != NULL; 1370 cbp = bioq_first(&queue)) { 1371 bioq_remove(&queue, cbp); 1372 g_destroy_bio(cbp); 1373 } 1374 if (bp->bio_error == 0) 1375 bp->bio_error = ENOMEM; 1376 g_io_deliver(bp, bp->bio_error); 1377 return; 1378 } 1379 bioq_insert_tail(&queue, cbp); 1380 cbp->bio_done = g_mirror_done; 1381 cp = disk->d_consumer; 1382 cbp->bio_caller1 = cp; 1383 cbp->bio_to = cp->provider; 1384 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1385 ("Consumer %s not opened (r%dw%de%d).", 1386 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1387 } 1388 for (cbp = bioq_first(&queue); cbp != NULL; 1389 cbp = bioq_first(&queue)) { 1390 bioq_remove(&queue, cbp); 1391 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1392 cp = cbp->bio_caller1; 1393 cbp->bio_caller1 = NULL; 1394 cp->index++; 1395 g_io_request(cbp, cp); 1396 } 1397 /* 1398 * Bump syncid on first write. 1399 */ 1400 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1401 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1402 g_topology_lock(); 1403 g_mirror_bump_syncid(sc); 1404 g_topology_unlock(); 1405 } 1406 return; 1407 } 1408 default: 1409 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1410 bp->bio_cmd, sc->sc_name)); 1411 break; 1412 } 1413} 1414 1415static int 1416g_mirror_can_destroy(struct g_mirror_softc *sc) 1417{ 1418 struct g_geom *gp; 1419 struct g_consumer *cp; 1420 1421 g_topology_assert(); 1422 gp = sc->sc_geom; 1423 LIST_FOREACH(cp, &gp->consumer, consumer) { 1424 if (g_mirror_is_busy(sc, cp)) 1425 return (0); 1426 } 1427 gp = sc->sc_sync.ds_geom; 1428 LIST_FOREACH(cp, &gp->consumer, consumer) { 1429 if (g_mirror_is_busy(sc, cp)) 1430 return (0); 1431 } 1432 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1433 sc->sc_name); 1434 return (1); 1435} 1436 1437static int 1438g_mirror_try_destroy(struct g_mirror_softc *sc) 1439{ 1440 1441 g_topology_lock(); 1442 if (!g_mirror_can_destroy(sc)) { 1443 g_topology_unlock(); 1444 return (0); 1445 } 1446 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1447 g_topology_unlock(); 1448 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1449 &sc->sc_worker); 1450 wakeup(&sc->sc_worker); 1451 sc->sc_worker = NULL; 1452 } else { 1453 g_mirror_destroy_device(sc); 1454 g_topology_unlock(); 1455 free(sc, M_MIRROR); 1456 } 1457 return (1); 1458} 1459 1460/* 1461 * Worker thread. 1462 */ 1463static void 1464g_mirror_worker(void *arg) 1465{ 1466 struct g_mirror_softc *sc; 1467 struct g_mirror_disk *disk; 1468 struct g_mirror_disk_sync *sync; 1469 struct g_mirror_event *ep; 1470 struct bio *bp; 1471 u_int nreqs; 1472 1473 sc = arg; 1474 mtx_lock_spin(&sched_lock); 1475 sched_prio(curthread, PRIBIO); 1476 mtx_unlock_spin(&sched_lock); 1477 1478 nreqs = 0; 1479 for (;;) { 1480 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1481 /* 1482 * First take a look at events. 1483 * This is important to handle events before any I/O requests. 1484 */ 1485 ep = g_mirror_event_get(sc); 1486 if (ep != NULL && g_topology_try_lock()) { 1487 g_mirror_event_remove(sc, ep); 1488 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1489 /* Update only device status. */ 1490 G_MIRROR_DEBUG(3, 1491 "Running event for device %s.", 1492 sc->sc_name); 1493 ep->e_error = 0; 1494 g_mirror_update_device(sc, 1); 1495 } else { 1496 /* Update disk status. */ 1497 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1498 g_mirror_get_diskname(ep->e_disk)); 1499 ep->e_error = g_mirror_update_disk(ep->e_disk, 1500 ep->e_state); 1501 if (ep->e_error == 0) 1502 g_mirror_update_device(sc, 0); 1503 } 1504 g_topology_unlock(); 1505 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1506 KASSERT(ep->e_error == 0, 1507 ("Error cannot be handled.")); 1508 g_mirror_event_free(ep); 1509 } else { 1510 ep->e_flags |= G_MIRROR_EVENT_DONE; 1511 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1512 ep); 1513 mtx_lock(&sc->sc_events_mtx); 1514 wakeup(ep); 1515 mtx_unlock(&sc->sc_events_mtx); 1516 } 1517 if ((sc->sc_flags & 1518 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1519 if (g_mirror_try_destroy(sc)) 1520 kthread_exit(0); 1521 } 1522 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1523 continue; 1524 } 1525 /* 1526 * Now I/O requests. 1527 */ 1528 /* Get first request from the queue. */ 1529 mtx_lock(&sc->sc_queue_mtx); 1530 bp = bioq_first(&sc->sc_queue); 1531 if (bp == NULL) { 1532 if (ep != NULL) { 1533 /* 1534 * No I/O requests and topology lock was 1535 * already held? Try again. 1536 */ 1537 mtx_unlock(&sc->sc_queue_mtx); 1538 continue; 1539 } 1540 if ((sc->sc_flags & 1541 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1542 mtx_unlock(&sc->sc_queue_mtx); 1543 if (g_mirror_try_destroy(sc)) 1544 kthread_exit(0); 1545 mtx_lock(&sc->sc_queue_mtx); 1546 } 1547 } 1548 if (sc->sc_sync.ds_ndisks > 0 && 1549 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1550 mtx_unlock(&sc->sc_queue_mtx); 1551 /* 1552 * It is time for synchronization... 1553 */ 1554 nreqs = 0; 1555 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1556 if (disk->d_state != 1557 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1558 continue; 1559 } 1560 sync = &disk->d_sync; 1561 if (sync->ds_offset >= 1562 sc->sc_provider->mediasize) { 1563 continue; 1564 } 1565 if (sync->ds_offset > sync->ds_offset_done) 1566 continue; 1567 if (sync->ds_resync != -1) { 1568 sync->ds_offset = sync->ds_resync; 1569 sync->ds_offset_done = sync->ds_resync; 1570 sync->ds_resync = -1; 1571 } 1572 g_mirror_sync_one(disk); 1573 } 1574 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1575 goto sleep; 1576 } 1577 if (bp == NULL) { 1578 if (g_mirror_check_idle(sc)) { 1579 u_int idletime; 1580 1581 idletime = g_mirror_idletime; 1582 if (idletime == 0) 1583 idletime = 1; 1584 idletime *= hz; 1585 if (msleep(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1586 "m:w1", idletime) == EWOULDBLOCK) { 1587 G_MIRROR_DEBUG(5, "%s: I'm here 3.", 1588 __func__); 1589 /* 1590 * No I/O requests in 'idletime' seconds, 1591 * so mark components as clean. 1592 */ 1593 g_mirror_idle(sc); 1594 } 1595 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1596 } else { 1597 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1598 "m:w2", 0); 1599 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1600 } 1601 continue; 1602 } 1603 nreqs++; 1604 bioq_remove(&sc->sc_queue, bp); 1605 mtx_unlock(&sc->sc_queue_mtx); 1606 1607 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1608 g_mirror_regular_request(bp); 1609 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1610 u_int timeout, sps; 1611 1612 g_mirror_sync_request(bp); 1613sleep: 1614 sps = g_mirror_syncs_per_sec; 1615 if (sps == 0) { 1616 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1617 continue; 1618 } 1619 if (ep != NULL) { 1620 /* 1621 * We have some pending events, don't sleep now. 1622 */ 1623 G_MIRROR_DEBUG(5, "%s: I'm here 7.", __func__); 1624 continue; 1625 } 1626 mtx_lock(&sc->sc_queue_mtx); 1627 if (bioq_first(&sc->sc_queue) != NULL) { 1628 mtx_unlock(&sc->sc_queue_mtx); 1629 G_MIRROR_DEBUG(5, "%s: I'm here 8.", __func__); 1630 continue; 1631 } 1632 timeout = hz / sps; 1633 if (timeout == 0) 1634 timeout = 1; 1635 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w3", 1636 timeout); 1637 } else { 1638 g_mirror_register_request(bp); 1639 } 1640 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1641 } 1642} 1643 1644/* 1645 * Open disk's consumer if needed. 1646 */ 1647static void 1648g_mirror_update_access(struct g_mirror_disk *disk) 1649{ 1650 struct g_provider *pp; 1651 1652 g_topology_assert(); 1653 1654 pp = disk->d_softc->sc_provider; 1655 if (pp == NULL) 1656 return; 1657 if (pp->acw > 0) { 1658 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1659 G_MIRROR_DEBUG(1, 1660 "Disk %s (device %s) marked as dirty.", 1661 g_mirror_get_diskname(disk), 1662 disk->d_softc->sc_name); 1663 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1664 } 1665 } else if (pp->acw == 0) { 1666 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1667 G_MIRROR_DEBUG(1, 1668 "Disk %s (device %s) marked as clean.", 1669 g_mirror_get_diskname(disk), 1670 disk->d_softc->sc_name); 1671 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1672 } 1673 } 1674} 1675 1676static void 1677g_mirror_sync_start(struct g_mirror_disk *disk) 1678{ 1679 struct g_mirror_softc *sc; 1680 int error; 1681 1682 g_topology_assert(); 1683 1684 sc = disk->d_softc; 1685 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1686 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1687 sc->sc_state)); 1688 1689 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1690 g_mirror_get_diskname(disk)); 1691 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1692 KASSERT(disk->d_sync.ds_consumer == NULL, 1693 ("Sync consumer already exists (device=%s, disk=%s).", 1694 sc->sc_name, g_mirror_get_diskname(disk))); 1695 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1696 disk->d_sync.ds_consumer->private = disk; 1697 disk->d_sync.ds_consumer->index = 0; 1698 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1699 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1700 disk->d_softc->sc_name, error)); 1701 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1702 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1703 disk->d_softc->sc_name, error)); 1704 disk->d_sync.ds_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1705 sc->sc_sync.ds_ndisks++; 1706} 1707 1708/* 1709 * Stop synchronization process. 1710 * type: 0 - synchronization finished 1711 * 1 - synchronization stopped 1712 */ 1713static void 1714g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1715{ 1716 1717 g_topology_assert(); 1718 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1719 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1720 g_mirror_disk_state2str(disk->d_state))); 1721 if (disk->d_sync.ds_consumer == NULL) 1722 return; 1723 1724 if (type == 0) { 1725 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1726 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1727 } else /* if (type == 1) */ { 1728 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1729 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1730 } 1731 g_mirror_kill_consumer(disk->d_softc, disk->d_sync.ds_consumer); 1732 free(disk->d_sync.ds_data, M_MIRROR); 1733 disk->d_sync.ds_consumer = NULL; 1734 disk->d_softc->sc_sync.ds_ndisks--; 1735 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1736} 1737 1738static void 1739g_mirror_launch_provider(struct g_mirror_softc *sc) 1740{ 1741 struct g_mirror_disk *disk; 1742 struct g_provider *pp; 1743 1744 g_topology_assert(); 1745 1746 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1747 pp->mediasize = sc->sc_mediasize; 1748 pp->sectorsize = sc->sc_sectorsize; 1749 sc->sc_provider = pp; 1750 g_error_provider(pp, 0); 1751 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1752 pp->name); 1753 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1754 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1755 g_mirror_sync_start(disk); 1756 } 1757} 1758 1759static void 1760g_mirror_destroy_provider(struct g_mirror_softc *sc) 1761{ 1762 struct g_mirror_disk *disk; 1763 struct bio *bp; 1764 1765 g_topology_assert(); 1766 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1767 sc->sc_name)); 1768 1769 g_error_provider(sc->sc_provider, ENXIO); 1770 mtx_lock(&sc->sc_queue_mtx); 1771 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1772 bioq_remove(&sc->sc_queue, bp); 1773 g_io_deliver(bp, ENXIO); 1774 } 1775 mtx_unlock(&sc->sc_queue_mtx); 1776 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1777 sc->sc_provider->name); 1778 sc->sc_provider->flags |= G_PF_WITHER; 1779 g_orphan_provider(sc->sc_provider, ENXIO); 1780 sc->sc_provider = NULL; 1781 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1782 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1783 g_mirror_sync_stop(disk, 1); 1784 } 1785} 1786 1787static void 1788g_mirror_go(void *arg) 1789{ 1790 struct g_mirror_softc *sc; 1791 1792 sc = arg; 1793 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1794 g_mirror_event_send(sc, 0, 1795 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1796} 1797 1798static u_int 1799g_mirror_determine_state(struct g_mirror_disk *disk) 1800{ 1801 struct g_mirror_softc *sc; 1802 u_int state; 1803 1804 sc = disk->d_softc; 1805 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1806 if ((disk->d_flags & 1807 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1808 /* Disk does not need synchronization. */ 1809 state = G_MIRROR_DISK_STATE_ACTIVE; 1810 } else { 1811 if ((sc->sc_flags & 1812 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1813 (disk->d_flags & 1814 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1815 /* 1816 * We can start synchronization from 1817 * the stored offset. 1818 */ 1819 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1820 } else { 1821 state = G_MIRROR_DISK_STATE_STALE; 1822 } 1823 } 1824 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1825 /* 1826 * Reset all synchronization data for this disk, 1827 * because if it even was synchronized, it was 1828 * synchronized to disks with different syncid. 1829 */ 1830 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1831 disk->d_sync.ds_offset = 0; 1832 disk->d_sync.ds_offset_done = 0; 1833 disk->d_sync.ds_syncid = sc->sc_syncid; 1834 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1835 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1836 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1837 } else { 1838 state = G_MIRROR_DISK_STATE_STALE; 1839 } 1840 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1841 /* 1842 * Not good, NOT GOOD! 1843 * It means that mirror was started on stale disks 1844 * and more fresh disk just arrive. 1845 * If there were writes, mirror is fucked up, sorry. 1846 * I think the best choice here is don't touch 1847 * this disk and inform the user laudly. 1848 */ 1849 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1850 "disk (%s) arrives!! It will not be connected to the " 1851 "running device.", sc->sc_name, 1852 g_mirror_get_diskname(disk)); 1853 g_mirror_destroy_disk(disk); 1854 state = G_MIRROR_DISK_STATE_NONE; 1855 /* Return immediately, because disk was destroyed. */ 1856 return (state); 1857 } 1858 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1859 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1860 return (state); 1861} 1862 1863/* 1864 * Update device state. 1865 */ 1866static void 1867g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1868{ 1869 struct g_mirror_disk *disk; 1870 u_int state; 1871 1872 g_topology_assert(); 1873 1874 switch (sc->sc_state) { 1875 case G_MIRROR_DEVICE_STATE_STARTING: 1876 { 1877 struct g_mirror_disk *pdisk, *tdisk; 1878 u_int dirty, ndisks, genid, syncid; 1879 1880 KASSERT(sc->sc_provider == NULL, 1881 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1882 /* 1883 * Are we ready? We are, if all disks are connected or 1884 * if we have any disks and 'force' is true. 1885 */ 1886 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1887 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1888 ; 1889 } else if (g_mirror_ndisks(sc, -1) == 0) { 1890 /* 1891 * Disks went down in starting phase, so destroy 1892 * device. 1893 */ 1894 root_mount_rel(sc->sc_rootmount); 1895 sc->sc_rootmount = NULL; 1896 callout_drain(&sc->sc_callout); 1897 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1898 return; 1899 } else { 1900 return; 1901 } 1902 1903 root_mount_rel(sc->sc_rootmount); 1904 sc->sc_rootmount = NULL; 1905 1906 /* 1907 * Activate all disks with the biggest syncid. 1908 */ 1909 if (force) { 1910 /* 1911 * If 'force' is true, we have been called due to 1912 * timeout, so don't bother canceling timeout. 1913 */ 1914 ndisks = 0; 1915 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1916 if ((disk->d_flags & 1917 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1918 ndisks++; 1919 } 1920 } 1921 if (ndisks == 0) { 1922 /* No valid disks found, destroy device. */ 1923 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1924 return; 1925 } 1926 } else { 1927 /* Cancel timeout. */ 1928 callout_drain(&sc->sc_callout); 1929 } 1930 1931 /* 1932 * Find the biggest genid. 1933 */ 1934 genid = 0; 1935 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1936 if (disk->d_genid > genid) 1937 genid = disk->d_genid; 1938 } 1939 sc->sc_genid = genid; 1940 /* 1941 * Remove all disks without the biggest genid. 1942 */ 1943 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 1944 if (disk->d_genid < genid) { 1945 G_MIRROR_DEBUG(0, 1946 "Component %s (device %s) broken, skipping.", 1947 g_mirror_get_diskname(disk), sc->sc_name); 1948 g_mirror_destroy_disk(disk); 1949 } 1950 } 1951 1952 /* 1953 * Find the biggest syncid. 1954 */ 1955 syncid = 0; 1956 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1957 if (disk->d_sync.ds_syncid > syncid) 1958 syncid = disk->d_sync.ds_syncid; 1959 } 1960 1961 /* 1962 * Here we need to look for dirty disks and if all disks 1963 * with the biggest syncid are dirty, we have to choose 1964 * one with the biggest priority and rebuild the rest. 1965 */ 1966 /* 1967 * Find the number of dirty disks with the biggest syncid. 1968 * Find the number of disks with the biggest syncid. 1969 * While here, find a disk with the biggest priority. 1970 */ 1971 dirty = ndisks = 0; 1972 pdisk = NULL; 1973 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1974 if (disk->d_sync.ds_syncid != syncid) 1975 continue; 1976 if ((disk->d_flags & 1977 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1978 continue; 1979 } 1980 ndisks++; 1981 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1982 dirty++; 1983 if (pdisk == NULL || 1984 pdisk->d_priority < disk->d_priority) { 1985 pdisk = disk; 1986 } 1987 } 1988 } 1989 if (dirty == 0) { 1990 /* No dirty disks at all, great. */ 1991 } else if (dirty == ndisks) { 1992 /* 1993 * Force synchronization for all dirty disks except one 1994 * with the biggest priority. 1995 */ 1996 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1997 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1998 "master disk for synchronization.", 1999 g_mirror_get_diskname(pdisk), sc->sc_name); 2000 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2001 if (disk->d_sync.ds_syncid != syncid) 2002 continue; 2003 if ((disk->d_flags & 2004 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2005 continue; 2006 } 2007 KASSERT((disk->d_flags & 2008 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2009 ("Disk %s isn't marked as dirty.", 2010 g_mirror_get_diskname(disk))); 2011 /* Skip the disk with the biggest priority. */ 2012 if (disk == pdisk) 2013 continue; 2014 disk->d_sync.ds_syncid = 0; 2015 } 2016 } else if (dirty < ndisks) { 2017 /* 2018 * Force synchronization for all dirty disks. 2019 * We have some non-dirty disks. 2020 */ 2021 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2022 if (disk->d_sync.ds_syncid != syncid) 2023 continue; 2024 if ((disk->d_flags & 2025 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2026 continue; 2027 } 2028 if ((disk->d_flags & 2029 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2030 continue; 2031 } 2032 disk->d_sync.ds_syncid = 0; 2033 } 2034 } 2035 2036 /* Reset hint. */ 2037 sc->sc_hint = NULL; 2038 sc->sc_syncid = syncid; 2039 if (force) { 2040 /* Remember to bump syncid on first write. */ 2041 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2042 } 2043 state = G_MIRROR_DEVICE_STATE_RUNNING; 2044 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2045 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2046 g_mirror_device_state2str(state)); 2047 sc->sc_state = state; 2048 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2049 state = g_mirror_determine_state(disk); 2050 g_mirror_event_send(disk, state, 2051 G_MIRROR_EVENT_DONTWAIT); 2052 if (state == G_MIRROR_DISK_STATE_STALE) 2053 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2054 } 2055 wakeup(&g_mirror_class); 2056 break; 2057 } 2058 case G_MIRROR_DEVICE_STATE_RUNNING: 2059 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2060 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2061 /* 2062 * No active disks or no disks at all, 2063 * so destroy device. 2064 */ 2065 if (sc->sc_provider != NULL) 2066 g_mirror_destroy_provider(sc); 2067 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2068 break; 2069 } else if (g_mirror_ndisks(sc, 2070 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2071 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2072 /* 2073 * We have active disks, launch provider if it doesn't 2074 * exist. 2075 */ 2076 if (sc->sc_provider == NULL) 2077 g_mirror_launch_provider(sc); 2078 } 2079 /* 2080 * Genid should be bumped immediately, so do it here. 2081 */ 2082 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2083 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2084 g_mirror_bump_genid(sc); 2085 } 2086 break; 2087 default: 2088 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2089 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2090 break; 2091 } 2092} 2093 2094/* 2095 * Update disk state and device state if needed. 2096 */ 2097#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2098 "Disk %s state changed from %s to %s (device %s).", \ 2099 g_mirror_get_diskname(disk), \ 2100 g_mirror_disk_state2str(disk->d_state), \ 2101 g_mirror_disk_state2str(state), sc->sc_name) 2102static int 2103g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2104{ 2105 struct g_mirror_softc *sc; 2106 2107 g_topology_assert(); 2108 2109 sc = disk->d_softc; 2110again: 2111 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2112 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2113 g_mirror_disk_state2str(state)); 2114 switch (state) { 2115 case G_MIRROR_DISK_STATE_NEW: 2116 /* 2117 * Possible scenarios: 2118 * 1. New disk arrive. 2119 */ 2120 /* Previous state should be NONE. */ 2121 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2122 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2123 g_mirror_disk_state2str(disk->d_state))); 2124 DISK_STATE_CHANGED(); 2125 2126 disk->d_state = state; 2127 if (LIST_EMPTY(&sc->sc_disks)) 2128 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2129 else { 2130 struct g_mirror_disk *dp; 2131 2132 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2133 if (disk->d_priority >= dp->d_priority) { 2134 LIST_INSERT_BEFORE(dp, disk, d_next); 2135 dp = NULL; 2136 break; 2137 } 2138 if (LIST_NEXT(dp, d_next) == NULL) 2139 break; 2140 } 2141 if (dp != NULL) 2142 LIST_INSERT_AFTER(dp, disk, d_next); 2143 } 2144 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2145 sc->sc_name, g_mirror_get_diskname(disk)); 2146 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2147 break; 2148 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2149 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2150 g_mirror_device_state2str(sc->sc_state), 2151 g_mirror_get_diskname(disk), 2152 g_mirror_disk_state2str(disk->d_state))); 2153 state = g_mirror_determine_state(disk); 2154 if (state != G_MIRROR_DISK_STATE_NONE) 2155 goto again; 2156 break; 2157 case G_MIRROR_DISK_STATE_ACTIVE: 2158 /* 2159 * Possible scenarios: 2160 * 1. New disk does not need synchronization. 2161 * 2. Synchronization process finished successfully. 2162 */ 2163 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2164 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2165 g_mirror_device_state2str(sc->sc_state), 2166 g_mirror_get_diskname(disk), 2167 g_mirror_disk_state2str(disk->d_state))); 2168 /* Previous state should be NEW or SYNCHRONIZING. */ 2169 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2170 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2171 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2172 g_mirror_disk_state2str(disk->d_state))); 2173 DISK_STATE_CHANGED(); 2174 2175 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2176 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2177 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2178 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2179 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2180 g_mirror_sync_stop(disk, 0); 2181 } 2182 disk->d_state = state; 2183 disk->d_sync.ds_offset = 0; 2184 disk->d_sync.ds_offset_done = 0; 2185 g_mirror_update_access(disk); 2186 g_mirror_update_metadata(disk); 2187 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2188 sc->sc_name, g_mirror_get_diskname(disk)); 2189 break; 2190 case G_MIRROR_DISK_STATE_STALE: 2191 /* 2192 * Possible scenarios: 2193 * 1. Stale disk was connected. 2194 */ 2195 /* Previous state should be NEW. */ 2196 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2197 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2198 g_mirror_disk_state2str(disk->d_state))); 2199 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2200 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2201 g_mirror_device_state2str(sc->sc_state), 2202 g_mirror_get_diskname(disk), 2203 g_mirror_disk_state2str(disk->d_state))); 2204 /* 2205 * STALE state is only possible if device is marked 2206 * NOAUTOSYNC. 2207 */ 2208 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2209 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2210 g_mirror_device_state2str(sc->sc_state), 2211 g_mirror_get_diskname(disk), 2212 g_mirror_disk_state2str(disk->d_state))); 2213 DISK_STATE_CHANGED(); 2214 2215 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2216 disk->d_state = state; 2217 g_mirror_update_metadata(disk); 2218 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2219 sc->sc_name, g_mirror_get_diskname(disk)); 2220 break; 2221 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2222 /* 2223 * Possible scenarios: 2224 * 1. Disk which needs synchronization was connected. 2225 */ 2226 /* Previous state should be NEW. */ 2227 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2228 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2229 g_mirror_disk_state2str(disk->d_state))); 2230 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2231 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2232 g_mirror_device_state2str(sc->sc_state), 2233 g_mirror_get_diskname(disk), 2234 g_mirror_disk_state2str(disk->d_state))); 2235 DISK_STATE_CHANGED(); 2236 2237 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2238 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2239 disk->d_state = state; 2240 if (sc->sc_provider != NULL) { 2241 g_mirror_sync_start(disk); 2242 g_mirror_update_metadata(disk); 2243 } 2244 break; 2245 case G_MIRROR_DISK_STATE_DISCONNECTED: 2246 /* 2247 * Possible scenarios: 2248 * 1. Device wasn't running yet, but disk disappear. 2249 * 2. Disk was active and disapppear. 2250 * 3. Disk disappear during synchronization process. 2251 */ 2252 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2253 /* 2254 * Previous state should be ACTIVE, STALE or 2255 * SYNCHRONIZING. 2256 */ 2257 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2258 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2259 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2260 ("Wrong disk state (%s, %s).", 2261 g_mirror_get_diskname(disk), 2262 g_mirror_disk_state2str(disk->d_state))); 2263 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2264 /* Previous state should be NEW. */ 2265 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2266 ("Wrong disk state (%s, %s).", 2267 g_mirror_get_diskname(disk), 2268 g_mirror_disk_state2str(disk->d_state))); 2269 /* 2270 * Reset bumping syncid if disk disappeared in STARTING 2271 * state. 2272 */ 2273 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2274 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2275#ifdef INVARIANTS 2276 } else { 2277 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2278 sc->sc_name, 2279 g_mirror_device_state2str(sc->sc_state), 2280 g_mirror_get_diskname(disk), 2281 g_mirror_disk_state2str(disk->d_state))); 2282#endif 2283 } 2284 DISK_STATE_CHANGED(); 2285 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2286 sc->sc_name, g_mirror_get_diskname(disk)); 2287 2288 g_mirror_destroy_disk(disk); 2289 break; 2290 case G_MIRROR_DISK_STATE_DESTROY: 2291 { 2292 int error; 2293 2294 error = g_mirror_clear_metadata(disk); 2295 if (error != 0) 2296 return (error); 2297 DISK_STATE_CHANGED(); 2298 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2299 sc->sc_name, g_mirror_get_diskname(disk)); 2300 2301 g_mirror_destroy_disk(disk); 2302 sc->sc_ndisks--; 2303 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2304 g_mirror_update_metadata(disk); 2305 } 2306 break; 2307 } 2308 default: 2309 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2310 break; 2311 } 2312 return (0); 2313} 2314#undef DISK_STATE_CHANGED 2315 2316int 2317g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2318{ 2319 struct g_provider *pp; 2320 u_char *buf; 2321 int error; 2322 2323 g_topology_assert(); 2324 2325 error = g_access(cp, 1, 0, 0); 2326 if (error != 0) 2327 return (error); 2328 pp = cp->provider; 2329 g_topology_unlock(); 2330 /* Metadata are stored on last sector. */ 2331 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2332 &error); 2333 g_topology_lock(); 2334 g_access(cp, -1, 0, 0); 2335 if (error != 0) { 2336 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2337 cp->provider->name, error); 2338 if (buf != NULL) 2339 g_free(buf); 2340 return (error); 2341 } 2342 2343 /* Decode metadata. */ 2344 error = mirror_metadata_decode(buf, md); 2345 g_free(buf); 2346 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2347 return (EINVAL); 2348 if (md->md_version > G_MIRROR_VERSION) { 2349 G_MIRROR_DEBUG(0, 2350 "Kernel module is too old to handle metadata from %s.", 2351 cp->provider->name); 2352 return (EINVAL); 2353 } 2354 if (error != 0) { 2355 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2356 cp->provider->name); 2357 return (error); 2358 } 2359 2360 return (0); 2361} 2362 2363static int 2364g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2365 struct g_mirror_metadata *md) 2366{ 2367 2368 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2369 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2370 pp->name, md->md_did); 2371 return (EEXIST); 2372 } 2373 if (md->md_all != sc->sc_ndisks) { 2374 G_MIRROR_DEBUG(1, 2375 "Invalid '%s' field on disk %s (device %s), skipping.", 2376 "md_all", pp->name, sc->sc_name); 2377 return (EINVAL); 2378 } 2379 if (md->md_slice != sc->sc_slice) { 2380 G_MIRROR_DEBUG(1, 2381 "Invalid '%s' field on disk %s (device %s), skipping.", 2382 "md_slice", pp->name, sc->sc_name); 2383 return (EINVAL); 2384 } 2385 if (md->md_balance != sc->sc_balance) { 2386 G_MIRROR_DEBUG(1, 2387 "Invalid '%s' field on disk %s (device %s), skipping.", 2388 "md_balance", pp->name, sc->sc_name); 2389 return (EINVAL); 2390 } 2391 if (md->md_mediasize != sc->sc_mediasize) { 2392 G_MIRROR_DEBUG(1, 2393 "Invalid '%s' field on disk %s (device %s), skipping.", 2394 "md_mediasize", pp->name, sc->sc_name); 2395 return (EINVAL); 2396 } 2397 if (sc->sc_mediasize > pp->mediasize) { 2398 G_MIRROR_DEBUG(1, 2399 "Invalid size of disk %s (device %s), skipping.", pp->name, 2400 sc->sc_name); 2401 return (EINVAL); 2402 } 2403 if (md->md_sectorsize != sc->sc_sectorsize) { 2404 G_MIRROR_DEBUG(1, 2405 "Invalid '%s' field on disk %s (device %s), skipping.", 2406 "md_sectorsize", pp->name, sc->sc_name); 2407 return (EINVAL); 2408 } 2409 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2410 G_MIRROR_DEBUG(1, 2411 "Invalid sector size of disk %s (device %s), skipping.", 2412 pp->name, sc->sc_name); 2413 return (EINVAL); 2414 } 2415 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2416 G_MIRROR_DEBUG(1, 2417 "Invalid device flags on disk %s (device %s), skipping.", 2418 pp->name, sc->sc_name); 2419 return (EINVAL); 2420 } 2421 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2422 G_MIRROR_DEBUG(1, 2423 "Invalid disk flags on disk %s (device %s), skipping.", 2424 pp->name, sc->sc_name); 2425 return (EINVAL); 2426 } 2427 return (0); 2428} 2429 2430int 2431g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2432 struct g_mirror_metadata *md) 2433{ 2434 struct g_mirror_disk *disk; 2435 int error; 2436 2437 g_topology_assert(); 2438 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2439 2440 error = g_mirror_check_metadata(sc, pp, md); 2441 if (error != 0) 2442 return (error); 2443 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2444 md->md_genid < sc->sc_genid) { 2445 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2446 pp->name, sc->sc_name); 2447 return (EINVAL); 2448 } 2449 disk = g_mirror_init_disk(sc, pp, md, &error); 2450 if (disk == NULL) 2451 return (error); 2452 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2453 G_MIRROR_EVENT_WAIT); 2454 if (error != 0) 2455 return (error); 2456 if (md->md_version < G_MIRROR_VERSION) { 2457 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2458 pp->name, md->md_version, G_MIRROR_VERSION); 2459 g_mirror_update_metadata(disk); 2460 } 2461 return (0); 2462} 2463 2464static int 2465g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2466{ 2467 struct g_mirror_softc *sc; 2468 struct g_mirror_disk *disk; 2469 int dcr, dcw, dce; 2470 2471 g_topology_assert(); 2472 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2473 acw, ace); 2474 2475 dcr = pp->acr + acr; 2476 dcw = pp->acw + acw; 2477 dce = pp->ace + ace; 2478 2479 sc = pp->geom->softc; 2480 if (sc == NULL || LIST_EMPTY(&sc->sc_disks) || 2481 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 2482 if (acr <= 0 && acw <= 0 && ace <= 0) 2483 return (0); 2484 else 2485 return (ENXIO); 2486 } 2487 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2488 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2489 continue; 2490 /* 2491 * Mark disk as dirty on open and unmark on close. 2492 */ 2493 if (pp->acw == 0 && dcw > 0) { 2494 G_MIRROR_DEBUG(1, 2495 "Disk %s (device %s) marked as dirty.", 2496 g_mirror_get_diskname(disk), sc->sc_name); 2497 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2498 g_mirror_update_metadata(disk); 2499 } else if (pp->acw > 0 && dcw == 0) { 2500 G_MIRROR_DEBUG(1, 2501 "Disk %s (device %s) marked as clean.", 2502 g_mirror_get_diskname(disk), sc->sc_name); 2503 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2504 g_mirror_update_metadata(disk); 2505 } 2506 } 2507 return (0); 2508} 2509 2510static struct g_geom * 2511g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2512{ 2513 struct g_mirror_softc *sc; 2514 struct g_geom *gp; 2515 int error, timeout; 2516 2517 g_topology_assert(); 2518 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2519 md->md_mid); 2520 2521 /* One disk is minimum. */ 2522 if (md->md_all < 1) 2523 return (NULL); 2524 /* 2525 * Action geom. 2526 */ 2527 gp = g_new_geomf(mp, "%s", md->md_name); 2528 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2529 gp->start = g_mirror_start; 2530 gp->orphan = g_mirror_orphan; 2531 gp->access = g_mirror_access; 2532 gp->dumpconf = g_mirror_dumpconf; 2533 2534 sc->sc_id = md->md_mid; 2535 sc->sc_slice = md->md_slice; 2536 sc->sc_balance = md->md_balance; 2537 sc->sc_mediasize = md->md_mediasize; 2538 sc->sc_sectorsize = md->md_sectorsize; 2539 sc->sc_ndisks = md->md_all; 2540 sc->sc_flags = md->md_mflags; 2541 sc->sc_bump_id = 0; 2542 sc->sc_idle = 0; 2543 bioq_init(&sc->sc_queue); 2544 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2545 LIST_INIT(&sc->sc_disks); 2546 TAILQ_INIT(&sc->sc_events); 2547 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2548 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2549 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2550 gp->softc = sc; 2551 sc->sc_geom = gp; 2552 sc->sc_provider = NULL; 2553 /* 2554 * Synchronization geom. 2555 */ 2556 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2557 gp->softc = sc; 2558 gp->orphan = g_mirror_orphan; 2559 sc->sc_sync.ds_geom = gp; 2560 sc->sc_sync.ds_ndisks = 0; 2561 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2562 "g_mirror %s", md->md_name); 2563 if (error != 0) { 2564 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2565 sc->sc_name); 2566 g_destroy_geom(sc->sc_sync.ds_geom); 2567 mtx_destroy(&sc->sc_events_mtx); 2568 mtx_destroy(&sc->sc_queue_mtx); 2569 g_destroy_geom(sc->sc_geom); 2570 free(sc, M_MIRROR); 2571 return (NULL); 2572 } 2573 2574 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2575 2576 sc->sc_rootmount = root_mount_hold("GMIRROR"); 2577 /* 2578 * Run timeout. 2579 */ 2580 timeout = g_mirror_timeout * hz; 2581 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2582 return (sc->sc_geom); 2583} 2584 2585int 2586g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2587{ 2588 struct g_provider *pp; 2589 2590 g_topology_assert(); 2591 2592 if (sc == NULL) 2593 return (ENXIO); 2594 pp = sc->sc_provider; 2595 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2596 if (force) { 2597 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 2598 "can't be definitely removed.", pp->name); 2599 } else { 2600 G_MIRROR_DEBUG(1, 2601 "Device %s is still open (r%dw%de%d).", pp->name, 2602 pp->acr, pp->acw, pp->ace); 2603 return (EBUSY); 2604 } 2605 } 2606 2607 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2608 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2609 g_topology_unlock(); 2610 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2611 mtx_lock(&sc->sc_queue_mtx); 2612 wakeup(sc); 2613 mtx_unlock(&sc->sc_queue_mtx); 2614 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2615 while (sc->sc_worker != NULL) 2616 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2617 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2618 g_topology_lock(); 2619 g_mirror_destroy_device(sc); 2620 free(sc, M_MIRROR); 2621 return (0); 2622} 2623 2624static void 2625g_mirror_taste_orphan(struct g_consumer *cp) 2626{ 2627 2628 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2629 cp->provider->name)); 2630} 2631 2632static struct g_geom * 2633g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2634{ 2635 struct g_mirror_metadata md; 2636 struct g_mirror_softc *sc; 2637 struct g_consumer *cp; 2638 struct g_geom *gp; 2639 int error; 2640 2641 g_topology_assert(); 2642 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2643 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2644 2645 gp = g_new_geomf(mp, "mirror:taste"); 2646 /* 2647 * This orphan function should be never called. 2648 */ 2649 gp->orphan = g_mirror_taste_orphan; 2650 cp = g_new_consumer(gp); 2651 g_attach(cp, pp); 2652 error = g_mirror_read_metadata(cp, &md); 2653 g_detach(cp); 2654 g_destroy_consumer(cp); 2655 g_destroy_geom(gp); 2656 if (error != 0) 2657 return (NULL); 2658 gp = NULL; 2659 2660 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2661 return (NULL); 2662 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 2663 return (NULL); 2664 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2665 G_MIRROR_DEBUG(0, 2666 "Device %s: provider %s marked as inactive, skipping.", 2667 md.md_name, pp->name); 2668 return (NULL); 2669 } 2670 if (g_mirror_debug >= 2) 2671 mirror_metadata_dump(&md); 2672 2673 /* 2674 * Let's check if device already exists. 2675 */ 2676 sc = NULL; 2677 LIST_FOREACH(gp, &mp->geom, geom) { 2678 sc = gp->softc; 2679 if (sc == NULL) 2680 continue; 2681 if (sc->sc_sync.ds_geom == gp) 2682 continue; 2683 if (strcmp(md.md_name, sc->sc_name) != 0) 2684 continue; 2685 if (md.md_mid != sc->sc_id) { 2686 G_MIRROR_DEBUG(0, "Device %s already configured.", 2687 sc->sc_name); 2688 return (NULL); 2689 } 2690 break; 2691 } 2692 if (gp == NULL) { 2693 gp = g_mirror_create(mp, &md); 2694 if (gp == NULL) { 2695 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2696 md.md_name); 2697 return (NULL); 2698 } 2699 sc = gp->softc; 2700 } 2701 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2702 error = g_mirror_add_disk(sc, pp, &md); 2703 if (error != 0) { 2704 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2705 pp->name, gp->name, error); 2706 if (LIST_EMPTY(&sc->sc_disks)) 2707 g_mirror_destroy(sc, 1); 2708 return (NULL); 2709 } 2710 return (gp); 2711} 2712 2713static int 2714g_mirror_destroy_geom(struct gctl_req *req __unused, 2715 struct g_class *mp __unused, struct g_geom *gp) 2716{ 2717 2718 return (g_mirror_destroy(gp->softc, 0)); 2719} 2720 2721static void 2722g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2723 struct g_consumer *cp, struct g_provider *pp) 2724{ 2725 struct g_mirror_softc *sc; 2726 2727 g_topology_assert(); 2728 2729 sc = gp->softc; 2730 if (sc == NULL) 2731 return; 2732 /* Skip synchronization geom. */ 2733 if (gp == sc->sc_sync.ds_geom) 2734 return; 2735 if (pp != NULL) { 2736 /* Nothing here. */ 2737 } else if (cp != NULL) { 2738 struct g_mirror_disk *disk; 2739 2740 disk = cp->private; 2741 if (disk == NULL) 2742 return; 2743 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2744 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2745 sbuf_printf(sb, "%s<Synchronized>", indent); 2746 if (disk->d_sync.ds_offset_done == 0) 2747 sbuf_printf(sb, "0%%"); 2748 else { 2749 sbuf_printf(sb, "%u%%", 2750 (u_int)((disk->d_sync.ds_offset_done * 100) / 2751 sc->sc_provider->mediasize)); 2752 } 2753 sbuf_printf(sb, "</Synchronized>\n"); 2754 } 2755 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2756 disk->d_sync.ds_syncid); 2757 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 2758 disk->d_genid); 2759 sbuf_printf(sb, "%s<Flags>", indent); 2760 if (disk->d_flags == 0) 2761 sbuf_printf(sb, "NONE"); 2762 else { 2763 int first = 1; 2764 2765#define ADD_FLAG(flag, name) do { \ 2766 if ((disk->d_flags & (flag)) != 0) { \ 2767 if (!first) \ 2768 sbuf_printf(sb, ", "); \ 2769 else \ 2770 first = 0; \ 2771 sbuf_printf(sb, name); \ 2772 } \ 2773} while (0) 2774 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2775 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2776 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2777 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2778 "SYNCHRONIZING"); 2779 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2780#undef ADD_FLAG 2781 } 2782 sbuf_printf(sb, "</Flags>\n"); 2783 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2784 disk->d_priority); 2785 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2786 g_mirror_disk_state2str(disk->d_state)); 2787 } else { 2788 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2789 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2790 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 2791 sbuf_printf(sb, "%s<Flags>", indent); 2792 if (sc->sc_flags == 0) 2793 sbuf_printf(sb, "NONE"); 2794 else { 2795 int first = 1; 2796 2797#define ADD_FLAG(flag, name) do { \ 2798 if ((sc->sc_flags & (flag)) != 0) { \ 2799 if (!first) \ 2800 sbuf_printf(sb, ", "); \ 2801 else \ 2802 first = 0; \ 2803 sbuf_printf(sb, name); \ 2804 } \ 2805} while (0) 2806 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2807#undef ADD_FLAG 2808 } 2809 sbuf_printf(sb, "</Flags>\n"); 2810 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2811 (u_int)sc->sc_slice); 2812 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2813 balance_name(sc->sc_balance)); 2814 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2815 sc->sc_ndisks); 2816 sbuf_printf(sb, "%s<State>", indent); 2817 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2818 sbuf_printf(sb, "%s", "STARTING"); 2819 else if (sc->sc_ndisks == 2820 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 2821 sbuf_printf(sb, "%s", "COMPLETE"); 2822 else 2823 sbuf_printf(sb, "%s", "DEGRADED"); 2824 sbuf_printf(sb, "</State>\n"); 2825 } 2826} 2827 2828static void 2829g_mirror_shutdown(void *arg, int howto) 2830{ 2831 struct g_class *mp; 2832 struct g_geom *gp, *gp2; 2833 2834 mp = arg; 2835 DROP_GIANT(); 2836 g_topology_lock(); 2837 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2838 if (gp->softc == NULL) 2839 continue; 2840 g_mirror_destroy(gp->softc, 1); 2841 } 2842 g_topology_unlock(); 2843 PICKUP_GIANT(); 2844#if 0 2845 tsleep(&gp, PRIBIO, "m:shutdown", hz * 20); 2846#endif 2847} 2848 2849static void 2850g_mirror_init(struct g_class *mp) 2851{ 2852 2853 g_mirror_ehtag = EVENTHANDLER_REGISTER(shutdown_post_sync, 2854 g_mirror_shutdown, mp, SHUTDOWN_PRI_FIRST); 2855 if (g_mirror_ehtag == NULL) 2856 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 2857} 2858 2859static void 2860g_mirror_fini(struct g_class *mp) 2861{ 2862 2863 if (g_mirror_ehtag == NULL) 2864 return; 2865 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_ehtag); 2866} 2867 2868DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2869