g_mirror.c revision 137248
1/*- 2 * Copyright (c) 2004 Pawel Jakub Dawidek <pjd@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27#include <sys/cdefs.h> 28__FBSDID("$FreeBSD: head/sys/geom/mirror/g_mirror.c 137248 2004-11-05 09:05:15Z pjd $"); 29 30#include <sys/param.h> 31#include <sys/systm.h> 32#include <sys/kernel.h> 33#include <sys/module.h> 34#include <sys/limits.h> 35#include <sys/lock.h> 36#include <sys/mutex.h> 37#include <sys/bio.h> 38#include <sys/sysctl.h> 39#include <sys/malloc.h> 40#include <sys/bitstring.h> 41#include <vm/uma.h> 42#include <machine/atomic.h> 43#include <geom/geom.h> 44#include <sys/proc.h> 45#include <sys/kthread.h> 46#include <geom/mirror/g_mirror.h> 47 48 49static MALLOC_DEFINE(M_MIRROR, "mirror data", "GEOM_MIRROR Data"); 50 51SYSCTL_DECL(_kern_geom); 52SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, "GEOM_MIRROR stuff"); 53u_int g_mirror_debug = 0; 54TUNABLE_INT("kern.geom.mirror.debug", &g_mirror_debug); 55SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RW, &g_mirror_debug, 0, 56 "Debug level"); 57static u_int g_mirror_timeout = 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_reqs_per_sync = 5; 62SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, reqs_per_sync, CTLFLAG_RW, 63 &g_mirror_reqs_per_sync, 0, 64 "Number of regular I/O requests per synchronization request"); 65static u_int g_mirror_syncs_per_sec = 100; 66SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, syncs_per_sec, CTLFLAG_RW, 67 &g_mirror_syncs_per_sec, 0, 68 "Number of synchronizations requests per second"); 69 70#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 71 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 72 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 73 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 74} while (0) 75 76 77static int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp, 78 struct g_geom *gp); 79static g_taste_t g_mirror_taste; 80 81struct g_class g_mirror_class = { 82 .name = G_MIRROR_CLASS_NAME, 83 .version = G_VERSION, 84 .ctlreq = g_mirror_config, 85 .taste = g_mirror_taste, 86 .destroy_geom = g_mirror_destroy_geom 87}; 88 89 90static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 91static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 92static void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force); 93static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 94 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 95static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 96 97 98static const char * 99g_mirror_disk_state2str(int state) 100{ 101 102 switch (state) { 103 case G_MIRROR_DISK_STATE_NONE: 104 return ("NONE"); 105 case G_MIRROR_DISK_STATE_NEW: 106 return ("NEW"); 107 case G_MIRROR_DISK_STATE_ACTIVE: 108 return ("ACTIVE"); 109 case G_MIRROR_DISK_STATE_STALE: 110 return ("STALE"); 111 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 112 return ("SYNCHRONIZING"); 113 case G_MIRROR_DISK_STATE_DISCONNECTED: 114 return ("DISCONNECTED"); 115 case G_MIRROR_DISK_STATE_DESTROY: 116 return ("DESTROY"); 117 default: 118 return ("INVALID"); 119 } 120} 121 122static const char * 123g_mirror_device_state2str(int state) 124{ 125 126 switch (state) { 127 case G_MIRROR_DEVICE_STATE_STARTING: 128 return ("STARTING"); 129 case G_MIRROR_DEVICE_STATE_RUNNING: 130 return ("RUNNING"); 131 default: 132 return ("INVALID"); 133 } 134} 135 136static const char * 137g_mirror_get_diskname(struct g_mirror_disk *disk) 138{ 139 140 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 141 return ("[unknown]"); 142 return (disk->d_name); 143} 144 145/* 146 * --- Events handling functions --- 147 * Events in geom_mirror are used to maintain disks and device status 148 * from one thread to simplify locking. 149 */ 150static void 151g_mirror_event_free(struct g_mirror_event *ep) 152{ 153 154 free(ep, M_MIRROR); 155} 156 157int 158g_mirror_event_send(void *arg, int state, int flags) 159{ 160 struct g_mirror_softc *sc; 161 struct g_mirror_disk *disk; 162 struct g_mirror_event *ep; 163 int error; 164 165 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 166 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 167 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 168 disk = NULL; 169 sc = arg; 170 } else { 171 disk = arg; 172 sc = disk->d_softc; 173 } 174 ep->e_disk = disk; 175 ep->e_state = state; 176 ep->e_flags = flags; 177 ep->e_error = 0; 178 mtx_lock(&sc->sc_events_mtx); 179 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 180 mtx_unlock(&sc->sc_events_mtx); 181 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 182 mtx_lock(&sc->sc_queue_mtx); 183 wakeup(sc); 184 mtx_unlock(&sc->sc_queue_mtx); 185 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 186 return (0); 187 g_topology_assert(); 188 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 189 g_topology_unlock(); 190 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 191 mtx_lock(&sc->sc_events_mtx); 192 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 193 hz * 5); 194 } 195 /* Don't even try to use 'sc' here, because it could be already dead. */ 196 g_topology_lock(); 197 error = ep->e_error; 198 g_mirror_event_free(ep); 199 return (error); 200} 201 202static struct g_mirror_event * 203g_mirror_event_get(struct g_mirror_softc *sc) 204{ 205 struct g_mirror_event *ep; 206 207 mtx_lock(&sc->sc_events_mtx); 208 ep = TAILQ_FIRST(&sc->sc_events); 209 if (ep != NULL) 210 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 211 mtx_unlock(&sc->sc_events_mtx); 212 return (ep); 213} 214 215static void 216g_mirror_event_cancel(struct g_mirror_disk *disk) 217{ 218 struct g_mirror_softc *sc; 219 struct g_mirror_event *ep, *tmpep; 220 221 g_topology_assert(); 222 223 sc = disk->d_softc; 224 mtx_lock(&sc->sc_events_mtx); 225 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 226 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 227 continue; 228 if (ep->e_disk != disk) 229 continue; 230 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 231 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 232 g_mirror_event_free(ep); 233 else { 234 ep->e_error = ECANCELED; 235 wakeup(ep); 236 } 237 } 238 mtx_unlock(&sc->sc_events_mtx); 239} 240 241/* 242 * Return the number of disks in given state. 243 * If state is equal to -1, count all connected disks. 244 */ 245u_int 246g_mirror_ndisks(struct g_mirror_softc *sc, int state) 247{ 248 struct g_mirror_disk *disk; 249 u_int n = 0; 250 251 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 252 if (state == -1 || disk->d_state == state) 253 n++; 254 } 255 return (n); 256} 257 258/* 259 * Find a disk in mirror by its disk ID. 260 */ 261static struct g_mirror_disk * 262g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 263{ 264 struct g_mirror_disk *disk; 265 266 g_topology_assert(); 267 268 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 269 if (disk->d_id == id) 270 return (disk); 271 } 272 return (NULL); 273} 274 275static u_int 276g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 277{ 278 struct bio *bp; 279 u_int nreqs = 0; 280 281 mtx_lock(&sc->sc_queue_mtx); 282 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 283 if (bp->bio_from == cp) 284 nreqs++; 285 } 286 mtx_unlock(&sc->sc_queue_mtx); 287 return (nreqs); 288} 289 290static int 291g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 292{ 293 294 if (cp->index > 0) { 295 G_MIRROR_DEBUG(2, 296 "I/O requests for %s exist, can't destroy it now.", 297 cp->provider->name); 298 return (1); 299 } 300 if (g_mirror_nrequests(sc, cp) > 0) { 301 G_MIRROR_DEBUG(2, 302 "I/O requests for %s in queue, can't destroy it now.", 303 cp->provider->name); 304 return (1); 305 } 306 return (0); 307} 308 309static void 310g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 311{ 312 313 g_topology_assert(); 314 315 cp->private = NULL; 316 if (g_mirror_is_busy(sc, cp)) 317 return; 318 G_MIRROR_DEBUG(2, "Consumer %s destroyed.", cp->provider->name); 319 g_detach(cp); 320 g_destroy_consumer(cp); 321} 322 323static int 324g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 325{ 326 int error; 327 328 g_topology_assert(); 329 KASSERT(disk->d_consumer == NULL, 330 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 331 332 disk->d_consumer = g_new_consumer(disk->d_softc->sc_geom); 333 disk->d_consumer->private = disk; 334 disk->d_consumer->index = 0; 335 error = g_attach(disk->d_consumer, pp); 336 if (error != 0) 337 return (error); 338 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 339 return (0); 340} 341 342static void 343g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 344{ 345 346 g_topology_assert(); 347 348 if (cp == NULL) 349 return; 350 if (cp->provider != NULL) { 351 G_MIRROR_DEBUG(2, "Disk %s disconnected.", cp->provider->name); 352 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) { 353 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 354 cp->provider->name, -cp->acr, -cp->acw, -cp->ace, 355 0); 356 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 357 } 358 g_mirror_kill_consumer(sc, cp); 359 } else { 360 g_destroy_consumer(cp); 361 } 362} 363 364/* 365 * Initialize disk. This means allocate memory, create consumer, attach it 366 * to the provider and open access (r1w1e1) to it. 367 */ 368static struct g_mirror_disk * 369g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 370 struct g_mirror_metadata *md, int *errorp) 371{ 372 struct g_mirror_disk *disk; 373 int error; 374 375 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 376 if (disk == NULL) { 377 error = ENOMEM; 378 goto fail; 379 } 380 disk->d_softc = sc; 381 error = g_mirror_connect_disk(disk, pp); 382 if (error != 0) 383 goto fail; 384 disk->d_id = md->md_did; 385 disk->d_state = G_MIRROR_DISK_STATE_NONE; 386 disk->d_priority = md->md_priority; 387 disk->d_delay.sec = 0; 388 disk->d_delay.frac = 0; 389 binuptime(&disk->d_last_used); 390 disk->d_flags = md->md_dflags; 391 if (md->md_provider[0] != '\0') 392 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 393 disk->d_sync.ds_consumer = NULL; 394 disk->d_sync.ds_offset = md->md_sync_offset; 395 disk->d_sync.ds_offset_done = md->md_sync_offset; 396 disk->d_sync.ds_resync = -1; 397 disk->d_sync.ds_syncid = md->md_syncid; 398 if (errorp != NULL) 399 *errorp = 0; 400 return (disk); 401fail: 402 if (errorp != NULL) 403 *errorp = error; 404 if (disk != NULL) { 405 g_mirror_disconnect_consumer(sc, disk->d_consumer); 406 free(disk, M_MIRROR); 407 } 408 return (NULL); 409} 410 411static void 412g_mirror_destroy_disk(struct g_mirror_disk *disk) 413{ 414 struct g_mirror_softc *sc; 415 416 g_topology_assert(); 417 418 LIST_REMOVE(disk, d_next); 419 g_mirror_event_cancel(disk); 420 sc = disk->d_softc; 421 if (sc->sc_hint == disk) 422 sc->sc_hint = NULL; 423 switch (disk->d_state) { 424 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 425 g_mirror_sync_stop(disk, 1); 426 /* FALLTHROUGH */ 427 case G_MIRROR_DISK_STATE_NEW: 428 case G_MIRROR_DISK_STATE_STALE: 429 case G_MIRROR_DISK_STATE_ACTIVE: 430 g_mirror_disconnect_consumer(sc, disk->d_consumer); 431 free(disk, M_MIRROR); 432 break; 433 default: 434 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 435 g_mirror_get_diskname(disk), 436 g_mirror_disk_state2str(disk->d_state))); 437 } 438} 439 440static void 441g_mirror_destroy_device(struct g_mirror_softc *sc) 442{ 443 struct g_mirror_disk *disk; 444 struct g_mirror_event *ep; 445 struct g_geom *gp; 446 struct g_consumer *cp, *tmpcp; 447 448 g_topology_assert(); 449 450 gp = sc->sc_geom; 451 if (sc->sc_provider != NULL) 452 g_mirror_destroy_provider(sc); 453 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 454 disk = LIST_FIRST(&sc->sc_disks)) { 455 g_mirror_destroy_disk(disk); 456 } 457 while ((ep = g_mirror_event_get(sc)) != NULL) { 458 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 459 g_mirror_event_free(ep); 460 else { 461 ep->e_error = ECANCELED; 462 ep->e_flags |= G_MIRROR_EVENT_DONE; 463 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 464 mtx_lock(&sc->sc_events_mtx); 465 wakeup(ep); 466 mtx_unlock(&sc->sc_events_mtx); 467 } 468 } 469 callout_drain(&sc->sc_callout); 470 gp->softc = NULL; 471 472 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 473 g_mirror_disconnect_consumer(sc, cp); 474 } 475 sc->sc_sync.ds_geom->softc = NULL; 476 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 477 mtx_destroy(&sc->sc_queue_mtx); 478 mtx_destroy(&sc->sc_events_mtx); 479 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 480 g_wither_geom(gp, ENXIO); 481} 482 483static void 484g_mirror_orphan(struct g_consumer *cp) 485{ 486 struct g_mirror_disk *disk; 487 488 g_topology_assert(); 489 490 disk = cp->private; 491 if (disk == NULL) 492 return; 493 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 494 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 495 G_MIRROR_EVENT_DONTWAIT); 496} 497 498static void 499g_mirror_spoiled(struct g_consumer *cp) 500{ 501 struct g_mirror_disk *disk; 502 503 g_topology_assert(); 504 505 disk = cp->private; 506 if (disk == NULL) 507 return; 508 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 509 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 510 G_MIRROR_EVENT_DONTWAIT); 511} 512 513/* 514 * Function should return the next active disk on the list. 515 * It is possible that it will be the same disk as given. 516 * If there are no active disks on list, NULL is returned. 517 */ 518static __inline struct g_mirror_disk * 519g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 520{ 521 struct g_mirror_disk *dp; 522 523 for (dp = LIST_NEXT(disk, d_next); dp != disk; 524 dp = LIST_NEXT(dp, d_next)) { 525 if (dp == NULL) 526 dp = LIST_FIRST(&sc->sc_disks); 527 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 528 break; 529 } 530 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 531 return (NULL); 532 return (dp); 533} 534 535static struct g_mirror_disk * 536g_mirror_get_disk(struct g_mirror_softc *sc) 537{ 538 struct g_mirror_disk *disk; 539 540 if (sc->sc_hint == NULL) { 541 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 542 if (sc->sc_hint == NULL) 543 return (NULL); 544 } 545 disk = sc->sc_hint; 546 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 547 disk = g_mirror_find_next(sc, disk); 548 if (disk == NULL) 549 return (NULL); 550 } 551 sc->sc_hint = g_mirror_find_next(sc, disk); 552 return (disk); 553} 554 555static int 556g_mirror_write_metadata(struct g_mirror_disk *disk, 557 struct g_mirror_metadata *md) 558{ 559 struct g_mirror_softc *sc; 560 struct g_consumer *cp; 561 off_t offset, length; 562 u_char *sector; 563 int close = 0, error = 0; 564 565 g_topology_assert(); 566 567 sc = disk->d_softc; 568 cp = disk->d_consumer; 569 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 570 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 571 length = cp->provider->sectorsize; 572 offset = cp->provider->mediasize - length; 573 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 574 /* 575 * Open consumer if it wasn't opened and remember to close it. 576 */ 577 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 578 error = g_access(cp, 0, 1, 1); 579 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 580 cp->provider->name, 0, 1, 1, error); 581 if (error == 0) 582 close = 1; 583#ifdef INVARIANTS 584 } else { 585 KASSERT(cp->acw > 0 && cp->ace > 0, 586 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 587 cp->acr, cp->acw, cp->ace)); 588#endif 589 } 590 if (error == 0) { 591 if (md != NULL) 592 mirror_metadata_encode(md, sector); 593 g_topology_unlock(); 594 error = g_write_data(cp, offset, sector, length); 595 g_topology_lock(); 596 } 597 free(sector, M_MIRROR); 598 if (close) { 599 g_access(cp, 0, -1, -1); 600 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 601 cp->provider->name, 0, -1, -1, 0); 602 } 603 if (error != 0) { 604 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 605 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 606 G_MIRROR_EVENT_DONTWAIT); 607 } 608 return (error); 609} 610 611static int 612g_mirror_clear_metadata(struct g_mirror_disk *disk) 613{ 614 int error; 615 616 g_topology_assert(); 617 error = g_mirror_write_metadata(disk, NULL); 618 if (error == 0) { 619 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 620 g_mirror_get_diskname(disk)); 621 } else { 622 G_MIRROR_DEBUG(0, 623 "Cannot clear metadata on disk %s (error=%d).", 624 g_mirror_get_diskname(disk), error); 625 } 626 return (error); 627} 628 629void 630g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 631 struct g_mirror_metadata *md) 632{ 633 634 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 635 md->md_version = G_MIRROR_VERSION; 636 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 637 md->md_mid = sc->sc_id; 638 md->md_all = sc->sc_ndisks; 639 md->md_slice = sc->sc_slice; 640 md->md_balance = sc->sc_balance; 641 md->md_mediasize = sc->sc_mediasize; 642 md->md_sectorsize = sc->sc_sectorsize; 643 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 644 bzero(md->md_provider, sizeof(md->md_provider)); 645 if (disk == NULL) { 646 md->md_did = arc4random(); 647 md->md_priority = 0; 648 md->md_syncid = 0; 649 md->md_dflags = 0; 650 md->md_sync_offset = 0; 651 } else { 652 md->md_did = disk->d_id; 653 md->md_priority = disk->d_priority; 654 md->md_syncid = disk->d_sync.ds_syncid; 655 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 656 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 657 md->md_sync_offset = disk->d_sync.ds_offset_done; 658 else 659 md->md_sync_offset = 0; 660 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 661 strlcpy(md->md_provider, 662 disk->d_consumer->provider->name, 663 sizeof(md->md_provider)); 664 } 665 } 666} 667 668void 669g_mirror_update_metadata(struct g_mirror_disk *disk) 670{ 671 struct g_mirror_metadata md; 672 int error; 673 674 g_topology_assert(); 675 g_mirror_fill_metadata(disk->d_softc, disk, &md); 676 error = g_mirror_write_metadata(disk, &md); 677 if (error == 0) { 678 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 679 g_mirror_get_diskname(disk)); 680 } else { 681 G_MIRROR_DEBUG(0, 682 "Cannot update metadata on disk %s (error=%d).", 683 g_mirror_get_diskname(disk), error); 684 } 685} 686 687static void 688g_mirror_bump_syncid(struct g_mirror_softc *sc) 689{ 690 struct g_mirror_disk *disk; 691 692 g_topology_assert(); 693 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 694 ("%s called with no active disks (device=%s).", __func__, 695 sc->sc_name)); 696 697 sc->sc_syncid++; 698 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 699 sc->sc_syncid); 700 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 701 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 702 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 703 disk->d_sync.ds_syncid = sc->sc_syncid; 704 g_mirror_update_metadata(disk); 705 } 706 } 707} 708 709static void 710g_mirror_idle(struct g_mirror_softc *sc) 711{ 712 struct g_mirror_disk *disk; 713 714 if (sc->sc_provider == NULL || sc->sc_provider->acw == 0) 715 return; 716 sc->sc_idle = 1; 717 g_topology_lock(); 718 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 719 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 720 continue; 721 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 722 g_mirror_get_diskname(disk), sc->sc_name); 723 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 724 g_mirror_update_metadata(disk); 725 } 726 g_topology_unlock(); 727} 728 729static void 730g_mirror_unidle(struct g_mirror_softc *sc) 731{ 732 struct g_mirror_disk *disk; 733 734 sc->sc_idle = 0; 735 g_topology_lock(); 736 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 737 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 738 continue; 739 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 740 g_mirror_get_diskname(disk), sc->sc_name); 741 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 742 g_mirror_update_metadata(disk); 743 } 744 g_topology_unlock(); 745} 746 747static __inline int 748bintime_cmp(struct bintime *bt1, struct bintime *bt2) 749{ 750 751 if (bt1->sec < bt2->sec) 752 return (-1); 753 else if (bt1->sec > bt2->sec) 754 return (1); 755 if (bt1->frac < bt2->frac) 756 return (-1); 757 else if (bt1->frac > bt2->frac) 758 return (1); 759 return (0); 760} 761 762static void 763g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 764{ 765 766 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 767 return; 768 binuptime(&disk->d_delay); 769 bintime_sub(&disk->d_delay, &bp->bio_t0); 770} 771 772static void 773g_mirror_done(struct bio *bp) 774{ 775 struct g_mirror_softc *sc; 776 777 sc = bp->bio_from->geom->softc; 778 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 779 mtx_lock(&sc->sc_queue_mtx); 780 bioq_disksort(&sc->sc_queue, bp); 781 wakeup(sc); 782 mtx_unlock(&sc->sc_queue_mtx); 783} 784 785static void 786g_mirror_regular_request(struct bio *bp) 787{ 788 struct g_mirror_softc *sc; 789 struct g_mirror_disk *disk; 790 struct bio *pbp; 791 792 g_topology_assert_not(); 793 794 bp->bio_from->index--; 795 pbp = bp->bio_parent; 796 sc = pbp->bio_to->geom->softc; 797 disk = bp->bio_from->private; 798 if (disk == NULL) { 799 g_topology_lock(); 800 g_mirror_kill_consumer(sc, bp->bio_from); 801 g_topology_unlock(); 802 } else { 803 g_mirror_update_delay(disk, bp); 804 } 805 806 pbp->bio_inbed++; 807 KASSERT(pbp->bio_inbed <= pbp->bio_children, 808 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 809 pbp->bio_children)); 810 if (bp->bio_error == 0 && pbp->bio_error == 0) { 811 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 812 g_destroy_bio(bp); 813 if (pbp->bio_children == pbp->bio_inbed) { 814 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 815 pbp->bio_completed = pbp->bio_length; 816 g_io_deliver(pbp, pbp->bio_error); 817 } 818 return; 819 } else if (bp->bio_error != 0) { 820 if (pbp->bio_error == 0) 821 pbp->bio_error = bp->bio_error; 822 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 823 bp->bio_error); 824 if (disk != NULL) { 825 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 826 g_mirror_event_send(disk, 827 G_MIRROR_DISK_STATE_DISCONNECTED, 828 G_MIRROR_EVENT_DONTWAIT); 829 } 830 switch (pbp->bio_cmd) { 831 case BIO_DELETE: 832 case BIO_WRITE: 833 pbp->bio_inbed--; 834 pbp->bio_children--; 835 break; 836 } 837 } 838 g_destroy_bio(bp); 839 840 switch (pbp->bio_cmd) { 841 case BIO_READ: 842 if (pbp->bio_children == pbp->bio_inbed) { 843 pbp->bio_error = 0; 844 mtx_lock(&sc->sc_queue_mtx); 845 bioq_disksort(&sc->sc_queue, pbp); 846 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 847 wakeup(sc); 848 mtx_unlock(&sc->sc_queue_mtx); 849 } 850 break; 851 case BIO_DELETE: 852 case BIO_WRITE: 853 if (pbp->bio_children == 0) { 854 /* 855 * All requests failed. 856 */ 857 } else if (pbp->bio_inbed < pbp->bio_children) { 858 /* Do nothing. */ 859 break; 860 } else if (pbp->bio_children == pbp->bio_inbed) { 861 /* Some requests succeeded. */ 862 pbp->bio_error = 0; 863 pbp->bio_completed = pbp->bio_length; 864 } 865 g_io_deliver(pbp, pbp->bio_error); 866 break; 867 default: 868 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 869 break; 870 } 871} 872 873static void 874g_mirror_sync_done(struct bio *bp) 875{ 876 struct g_mirror_softc *sc; 877 878 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 879 sc = bp->bio_from->geom->softc; 880 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 881 mtx_lock(&sc->sc_queue_mtx); 882 bioq_disksort(&sc->sc_queue, bp); 883 wakeup(sc); 884 mtx_unlock(&sc->sc_queue_mtx); 885} 886 887static void 888g_mirror_start(struct bio *bp) 889{ 890 struct g_mirror_softc *sc; 891 892 sc = bp->bio_to->geom->softc; 893 /* 894 * If sc == NULL or there are no valid disks, provider's error 895 * should be set and g_mirror_start() should not be called at all. 896 */ 897 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 898 ("Provider's error should be set (error=%d)(mirror=%s).", 899 bp->bio_to->error, bp->bio_to->name)); 900 G_MIRROR_LOGREQ(3, bp, "Request received."); 901 902 switch (bp->bio_cmd) { 903 case BIO_READ: 904 case BIO_WRITE: 905 case BIO_DELETE: 906 break; 907 case BIO_GETATTR: 908 default: 909 g_io_deliver(bp, EOPNOTSUPP); 910 return; 911 } 912 mtx_lock(&sc->sc_queue_mtx); 913 bioq_disksort(&sc->sc_queue, bp); 914 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 915 wakeup(sc); 916 mtx_unlock(&sc->sc_queue_mtx); 917} 918 919/* 920 * Send one synchronization request. 921 */ 922static void 923g_mirror_sync_one(struct g_mirror_disk *disk) 924{ 925 struct g_mirror_softc *sc; 926 struct bio *bp; 927 928 sc = disk->d_softc; 929 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 930 ("Disk %s is not marked for synchronization.", 931 g_mirror_get_diskname(disk))); 932 933 bp = g_new_bio(); 934 if (bp == NULL) 935 return; 936 bp->bio_parent = NULL; 937 bp->bio_cmd = BIO_READ; 938 bp->bio_offset = disk->d_sync.ds_offset; 939 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 940 bp->bio_cflags = 0; 941 bp->bio_done = g_mirror_sync_done; 942 bp->bio_data = disk->d_sync.ds_data; 943 if (bp->bio_data == NULL) { 944 g_destroy_bio(bp); 945 return; 946 } 947 disk->d_sync.ds_offset += bp->bio_length; 948 bp->bio_to = sc->sc_provider; 949 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 950 disk->d_sync.ds_consumer->index++; 951 g_io_request(bp, disk->d_sync.ds_consumer); 952} 953 954static void 955g_mirror_sync_request(struct bio *bp) 956{ 957 struct g_mirror_softc *sc; 958 struct g_mirror_disk *disk; 959 960 bp->bio_from->index--; 961 sc = bp->bio_from->geom->softc; 962 disk = bp->bio_from->private; 963 if (disk == NULL) { 964 g_topology_lock(); 965 g_mirror_kill_consumer(sc, bp->bio_from); 966 g_topology_unlock(); 967 g_destroy_bio(bp); 968 return; 969 } 970 971 /* 972 * Synchronization request. 973 */ 974 switch (bp->bio_cmd) { 975 case BIO_READ: 976 { 977 struct g_consumer *cp; 978 979 if (bp->bio_error != 0) { 980 G_MIRROR_LOGREQ(0, bp, 981 "Synchronization request failed (error=%d).", 982 bp->bio_error); 983 g_destroy_bio(bp); 984 return; 985 } 986 G_MIRROR_LOGREQ(3, bp, 987 "Synchronization request half-finished."); 988 bp->bio_cmd = BIO_WRITE; 989 bp->bio_cflags = 0; 990 cp = disk->d_consumer; 991 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 992 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 993 cp->acr, cp->acw, cp->ace)); 994 cp->index++; 995 g_io_request(bp, cp); 996 return; 997 } 998 case BIO_WRITE: 999 { 1000 struct g_mirror_disk_sync *sync; 1001 1002 if (bp->bio_error != 0) { 1003 G_MIRROR_LOGREQ(0, bp, 1004 "Synchronization request failed (error=%d).", 1005 bp->bio_error); 1006 g_destroy_bio(bp); 1007 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 1008 g_mirror_event_send(disk, 1009 G_MIRROR_DISK_STATE_DISCONNECTED, 1010 G_MIRROR_EVENT_DONTWAIT); 1011 return; 1012 } 1013 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1014 sync = &disk->d_sync; 1015 sync->ds_offset_done = bp->bio_offset + bp->bio_length; 1016 g_destroy_bio(bp); 1017 if (sync->ds_resync != -1) 1018 break; 1019 if (sync->ds_offset_done == sc->sc_provider->mediasize) { 1020 /* 1021 * Disk up-to-date, activate it. 1022 */ 1023 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1024 G_MIRROR_EVENT_DONTWAIT); 1025 return; 1026 } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) { 1027 /* 1028 * Update offset_done on every 100 blocks. 1029 * XXX: This should be configurable. 1030 */ 1031 g_topology_lock(); 1032 g_mirror_update_metadata(disk); 1033 g_topology_unlock(); 1034 } 1035 return; 1036 } 1037 default: 1038 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1039 bp->bio_cmd, sc->sc_name)); 1040 break; 1041 } 1042} 1043 1044static void 1045g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1046{ 1047 struct g_mirror_disk *disk; 1048 struct g_consumer *cp; 1049 struct bio *cbp; 1050 1051 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1052 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1053 break; 1054 } 1055 if (disk == NULL) { 1056 if (bp->bio_error == 0) 1057 bp->bio_error = ENXIO; 1058 g_io_deliver(bp, bp->bio_error); 1059 return; 1060 } 1061 cbp = g_clone_bio(bp); 1062 if (cbp == NULL) { 1063 if (bp->bio_error == 0) 1064 bp->bio_error = ENOMEM; 1065 g_io_deliver(bp, bp->bio_error); 1066 return; 1067 } 1068 /* 1069 * Fill in the component buf structure. 1070 */ 1071 cp = disk->d_consumer; 1072 cbp->bio_done = g_mirror_done; 1073 cbp->bio_to = cp->provider; 1074 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1075 KASSERT(cp->acr > 0 && cp->ace > 0, 1076 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1077 cp->acw, cp->ace)); 1078 cp->index++; 1079 g_io_request(cbp, cp); 1080} 1081 1082static void 1083g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1084{ 1085 struct g_mirror_disk *disk; 1086 struct g_consumer *cp; 1087 struct bio *cbp; 1088 1089 disk = g_mirror_get_disk(sc); 1090 if (disk == NULL) { 1091 if (bp->bio_error == 0) 1092 bp->bio_error = ENXIO; 1093 g_io_deliver(bp, bp->bio_error); 1094 return; 1095 } 1096 cbp = g_clone_bio(bp); 1097 if (cbp == NULL) { 1098 if (bp->bio_error == 0) 1099 bp->bio_error = ENOMEM; 1100 g_io_deliver(bp, bp->bio_error); 1101 return; 1102 } 1103 /* 1104 * Fill in the component buf structure. 1105 */ 1106 cp = disk->d_consumer; 1107 cbp->bio_done = g_mirror_done; 1108 cbp->bio_to = cp->provider; 1109 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1110 KASSERT(cp->acr > 0 && cp->ace > 0, 1111 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1112 cp->acw, cp->ace)); 1113 cp->index++; 1114 g_io_request(cbp, cp); 1115} 1116 1117static void 1118g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1119{ 1120 struct g_mirror_disk *disk, *dp; 1121 struct g_consumer *cp; 1122 struct bio *cbp; 1123 struct bintime curtime; 1124 1125 binuptime(&curtime); 1126 /* 1127 * Find a disk which the smallest load. 1128 */ 1129 disk = NULL; 1130 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1131 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1132 continue; 1133 /* If disk wasn't used for more than 2 sec, use it. */ 1134 if (curtime.sec - dp->d_last_used.sec >= 2) { 1135 disk = dp; 1136 break; 1137 } 1138 if (disk == NULL || 1139 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1140 disk = dp; 1141 } 1142 } 1143 cbp = g_clone_bio(bp); 1144 if (cbp == NULL) { 1145 if (bp->bio_error == 0) 1146 bp->bio_error = ENOMEM; 1147 g_io_deliver(bp, bp->bio_error); 1148 return; 1149 } 1150 /* 1151 * Fill in the component buf structure. 1152 */ 1153 cp = disk->d_consumer; 1154 cbp->bio_done = g_mirror_done; 1155 cbp->bio_to = cp->provider; 1156 binuptime(&disk->d_last_used); 1157 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1158 KASSERT(cp->acr > 0 && cp->ace > 0, 1159 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1160 cp->acw, cp->ace)); 1161 cp->index++; 1162 g_io_request(cbp, cp); 1163} 1164 1165static void 1166g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1167{ 1168 struct bio_queue_head queue; 1169 struct g_mirror_disk *disk; 1170 struct g_consumer *cp; 1171 struct bio *cbp; 1172 off_t left, mod, offset, slice; 1173 u_char *data; 1174 u_int ndisks; 1175 1176 if (bp->bio_length <= sc->sc_slice) { 1177 g_mirror_request_round_robin(sc, bp); 1178 return; 1179 } 1180 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1181 slice = bp->bio_length / ndisks; 1182 mod = slice % sc->sc_provider->sectorsize; 1183 if (mod != 0) 1184 slice += sc->sc_provider->sectorsize - mod; 1185 /* 1186 * Allocate all bios before sending any request, so we can 1187 * return ENOMEM in nice and clean way. 1188 */ 1189 left = bp->bio_length; 1190 offset = bp->bio_offset; 1191 data = bp->bio_data; 1192 bioq_init(&queue); 1193 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1194 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1195 continue; 1196 cbp = g_clone_bio(bp); 1197 if (cbp == NULL) { 1198 for (cbp = bioq_first(&queue); cbp != NULL; 1199 cbp = bioq_first(&queue)) { 1200 bioq_remove(&queue, cbp); 1201 g_destroy_bio(cbp); 1202 } 1203 if (bp->bio_error == 0) 1204 bp->bio_error = ENOMEM; 1205 g_io_deliver(bp, bp->bio_error); 1206 return; 1207 } 1208 bioq_insert_tail(&queue, cbp); 1209 cbp->bio_done = g_mirror_done; 1210 cbp->bio_caller1 = disk; 1211 cbp->bio_to = disk->d_consumer->provider; 1212 cbp->bio_offset = offset; 1213 cbp->bio_data = data; 1214 cbp->bio_length = MIN(left, slice); 1215 left -= cbp->bio_length; 1216 if (left == 0) 1217 break; 1218 offset += cbp->bio_length; 1219 data += cbp->bio_length; 1220 } 1221 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1222 bioq_remove(&queue, cbp); 1223 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1224 disk = cbp->bio_caller1; 1225 cbp->bio_caller1 = NULL; 1226 cp = disk->d_consumer; 1227 KASSERT(cp->acr > 0 && cp->ace > 0, 1228 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1229 cp->acr, cp->acw, cp->ace)); 1230 disk->d_consumer->index++; 1231 g_io_request(cbp, disk->d_consumer); 1232 } 1233} 1234 1235static void 1236g_mirror_register_request(struct bio *bp) 1237{ 1238 struct g_mirror_softc *sc; 1239 1240 sc = bp->bio_to->geom->softc; 1241 switch (bp->bio_cmd) { 1242 case BIO_READ: 1243 switch (sc->sc_balance) { 1244 case G_MIRROR_BALANCE_LOAD: 1245 g_mirror_request_load(sc, bp); 1246 break; 1247 case G_MIRROR_BALANCE_PREFER: 1248 g_mirror_request_prefer(sc, bp); 1249 break; 1250 case G_MIRROR_BALANCE_ROUND_ROBIN: 1251 g_mirror_request_round_robin(sc, bp); 1252 break; 1253 case G_MIRROR_BALANCE_SPLIT: 1254 g_mirror_request_split(sc, bp); 1255 break; 1256 } 1257 return; 1258 case BIO_WRITE: 1259 case BIO_DELETE: 1260 { 1261 struct g_mirror_disk *disk; 1262 struct g_mirror_disk_sync *sync; 1263 struct bio_queue_head queue; 1264 struct g_consumer *cp; 1265 struct bio *cbp; 1266 1267 if (sc->sc_idle) 1268 g_mirror_unidle(sc); 1269 /* 1270 * Allocate all bios before sending any request, so we can 1271 * return ENOMEM in nice and clean way. 1272 */ 1273 bioq_init(&queue); 1274 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1275 sync = &disk->d_sync; 1276 switch (disk->d_state) { 1277 case G_MIRROR_DISK_STATE_ACTIVE: 1278 break; 1279 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1280 if (bp->bio_offset >= sync->ds_offset) 1281 continue; 1282 else if (bp->bio_offset + bp->bio_length > 1283 sync->ds_offset_done && 1284 (bp->bio_offset < sync->ds_resync || 1285 sync->ds_resync == -1)) { 1286 sync->ds_resync = bp->bio_offset - 1287 (bp->bio_offset % MAXPHYS); 1288 } 1289 break; 1290 default: 1291 continue; 1292 } 1293 cbp = g_clone_bio(bp); 1294 if (cbp == NULL) { 1295 for (cbp = bioq_first(&queue); cbp != NULL; 1296 cbp = bioq_first(&queue)) { 1297 bioq_remove(&queue, cbp); 1298 g_destroy_bio(cbp); 1299 } 1300 if (bp->bio_error == 0) 1301 bp->bio_error = ENOMEM; 1302 g_io_deliver(bp, bp->bio_error); 1303 return; 1304 } 1305 bioq_insert_tail(&queue, cbp); 1306 cbp->bio_done = g_mirror_done; 1307 cp = disk->d_consumer; 1308 cbp->bio_caller1 = cp; 1309 cbp->bio_to = cp->provider; 1310 KASSERT(cp->acw > 0 && cp->ace > 0, 1311 ("Consumer %s not opened (r%dw%de%d).", 1312 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1313 } 1314 for (cbp = bioq_first(&queue); cbp != NULL; 1315 cbp = bioq_first(&queue)) { 1316 bioq_remove(&queue, cbp); 1317 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1318 cp = cbp->bio_caller1; 1319 cbp->bio_caller1 = NULL; 1320 cp->index++; 1321 g_io_request(cbp, cp); 1322 } 1323 /* 1324 * Bump syncid on first write. 1325 */ 1326 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) { 1327 sc->sc_bump_syncid = 0; 1328 g_topology_lock(); 1329 g_mirror_bump_syncid(sc); 1330 g_topology_unlock(); 1331 } 1332 return; 1333 } 1334 default: 1335 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1336 bp->bio_cmd, sc->sc_name)); 1337 break; 1338 } 1339} 1340 1341static int 1342g_mirror_can_destroy(struct g_mirror_softc *sc) 1343{ 1344 struct g_geom *gp; 1345 struct g_consumer *cp; 1346 1347 g_topology_assert(); 1348 gp = sc->sc_geom; 1349 LIST_FOREACH(cp, &gp->consumer, consumer) { 1350 if (g_mirror_is_busy(sc, cp)) 1351 return (0); 1352 } 1353 gp = sc->sc_sync.ds_geom; 1354 LIST_FOREACH(cp, &gp->consumer, consumer) { 1355 if (g_mirror_is_busy(sc, cp)) 1356 return (0); 1357 } 1358 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1359 sc->sc_name); 1360 return (1); 1361} 1362 1363static int 1364g_mirror_try_destroy(struct g_mirror_softc *sc) 1365{ 1366 1367 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1368 g_topology_lock(); 1369 if (!g_mirror_can_destroy(sc)) { 1370 g_topology_unlock(); 1371 return (0); 1372 } 1373 g_topology_unlock(); 1374 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1375 &sc->sc_worker); 1376 wakeup(&sc->sc_worker); 1377 sc->sc_worker = NULL; 1378 } else { 1379 g_topology_lock(); 1380 if (!g_mirror_can_destroy(sc)) { 1381 g_topology_unlock(); 1382 return (0); 1383 } 1384 g_mirror_destroy_device(sc); 1385 g_topology_unlock(); 1386 free(sc, M_MIRROR); 1387 } 1388 return (1); 1389} 1390 1391/* 1392 * Worker thread. 1393 */ 1394static void 1395g_mirror_worker(void *arg) 1396{ 1397 struct g_mirror_softc *sc; 1398 struct g_mirror_disk *disk; 1399 struct g_mirror_disk_sync *sync; 1400 struct g_mirror_event *ep; 1401 struct bio *bp; 1402 u_int nreqs; 1403 1404 sc = arg; 1405 curthread->td_base_pri = PRIBIO; 1406 1407 nreqs = 0; 1408 for (;;) { 1409 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1410 /* 1411 * First take a look at events. 1412 * This is important to handle events before any I/O requests. 1413 */ 1414 ep = g_mirror_event_get(sc); 1415 if (ep != NULL) { 1416 g_topology_lock(); 1417 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1418 /* Update only device status. */ 1419 G_MIRROR_DEBUG(3, 1420 "Running event for device %s.", 1421 sc->sc_name); 1422 ep->e_error = 0; 1423 g_mirror_update_device(sc, 1); 1424 } else { 1425 /* Update disk status. */ 1426 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1427 g_mirror_get_diskname(ep->e_disk)); 1428 ep->e_error = g_mirror_update_disk(ep->e_disk, 1429 ep->e_state); 1430 if (ep->e_error == 0) 1431 g_mirror_update_device(sc, 0); 1432 } 1433 g_topology_unlock(); 1434 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1435 KASSERT(ep->e_error == 0, 1436 ("Error cannot be handled.")); 1437 g_mirror_event_free(ep); 1438 } else { 1439 ep->e_flags |= G_MIRROR_EVENT_DONE; 1440 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1441 ep); 1442 mtx_lock(&sc->sc_events_mtx); 1443 wakeup(ep); 1444 mtx_unlock(&sc->sc_events_mtx); 1445 } 1446 if ((sc->sc_flags & 1447 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1448 if (g_mirror_try_destroy(sc)) 1449 kthread_exit(0); 1450 } 1451 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1452 continue; 1453 } 1454 /* 1455 * Now I/O requests. 1456 */ 1457 /* Get first request from the queue. */ 1458 mtx_lock(&sc->sc_queue_mtx); 1459 bp = bioq_first(&sc->sc_queue); 1460 if (bp == NULL) { 1461 if ((sc->sc_flags & 1462 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1463 mtx_unlock(&sc->sc_queue_mtx); 1464 if (g_mirror_try_destroy(sc)) 1465 kthread_exit(0); 1466 mtx_lock(&sc->sc_queue_mtx); 1467 } 1468 } 1469 if (sc->sc_sync.ds_ndisks > 0 && 1470 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1471 mtx_unlock(&sc->sc_queue_mtx); 1472 /* 1473 * It is time for synchronization... 1474 */ 1475 nreqs = 0; 1476 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1477 if (disk->d_state != 1478 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1479 continue; 1480 } 1481 sync = &disk->d_sync; 1482 if (sync->ds_offset >= 1483 sc->sc_provider->mediasize) { 1484 continue; 1485 } 1486 if (sync->ds_offset > sync->ds_offset_done) 1487 continue; 1488 if (sync->ds_resync != -1) { 1489 sync->ds_offset = sync->ds_resync; 1490 sync->ds_offset_done = sync->ds_resync; 1491 sync->ds_resync = -1; 1492 } 1493 g_mirror_sync_one(disk); 1494 } 1495 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1496 goto sleep; 1497 } 1498 if (bp == NULL) { 1499 if (msleep(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1500 "m:w1", hz * 5) == EWOULDBLOCK) { 1501 /* 1502 * No I/O requests in 5 seconds, so mark 1503 * components as clean. 1504 */ 1505 if (!sc->sc_idle) 1506 g_mirror_idle(sc); 1507 } 1508 continue; 1509 } 1510 nreqs++; 1511 bioq_remove(&sc->sc_queue, bp); 1512 mtx_unlock(&sc->sc_queue_mtx); 1513 1514 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1515 g_mirror_regular_request(bp); 1516 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1517 u_int timeout, sps; 1518 1519 g_mirror_sync_request(bp); 1520sleep: 1521 sps = atomic_load_acq_int(&g_mirror_syncs_per_sec); 1522 if (sps == 0) { 1523 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1524 continue; 1525 } 1526 mtx_lock(&sc->sc_queue_mtx); 1527 if (bioq_first(&sc->sc_queue) != NULL) { 1528 mtx_unlock(&sc->sc_queue_mtx); 1529 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1530 continue; 1531 } 1532 timeout = hz / sps; 1533 if (timeout == 0) 1534 timeout = 1; 1535 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w2", 1536 timeout); 1537 } else { 1538 g_mirror_register_request(bp); 1539 } 1540 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1541 } 1542} 1543 1544/* 1545 * Open disk's consumer if needed. 1546 */ 1547static void 1548g_mirror_update_access(struct g_mirror_disk *disk) 1549{ 1550 struct g_provider *pp; 1551 struct g_consumer *cp; 1552 int acr, acw, ace, cpw, error; 1553 1554 g_topology_assert(); 1555 1556 cp = disk->d_consumer; 1557 pp = disk->d_softc->sc_provider; 1558 if (pp == NULL) { 1559 acr = -cp->acr; 1560 acw = -cp->acw; 1561 ace = -cp->ace; 1562 } else { 1563 acr = pp->acr - cp->acr; 1564 acw = pp->acw - cp->acw; 1565 ace = pp->ace - cp->ace; 1566 /* Grab an extra "exclusive" bit. */ 1567 if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) 1568 ace++; 1569 } 1570 if (acr == 0 && acw == 0 && ace == 0) 1571 return; 1572 cpw = cp->acw; 1573 error = g_access(cp, acr, acw, ace); 1574 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, acr, 1575 acw, ace, error); 1576 if (error != 0) { 1577 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1578 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1579 G_MIRROR_EVENT_DONTWAIT); 1580 return; 1581 } 1582 if (cpw == 0 && cp->acw > 0) { 1583 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1584 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1585 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1586 } else if (cpw > 0 && cp->acw == 0) { 1587 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1588 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1589 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1590 } 1591} 1592 1593static void 1594g_mirror_sync_start(struct g_mirror_disk *disk) 1595{ 1596 struct g_mirror_softc *sc; 1597 struct g_consumer *cp; 1598 int error; 1599 1600 g_topology_assert(); 1601 1602 sc = disk->d_softc; 1603 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1604 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1605 sc->sc_state)); 1606 cp = disk->d_consumer; 1607 KASSERT(cp->acr == 0 && cp->acw == 0 && cp->ace == 0, 1608 ("Consumer %s already opened.", cp->provider->name)); 1609 1610 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1611 g_mirror_get_diskname(disk)); 1612 error = g_access(cp, 0, 1, 1); 1613 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, 1, 1614 1, error); 1615 if (error != 0) { 1616 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1617 G_MIRROR_EVENT_DONTWAIT); 1618 return; 1619 } 1620 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1621 KASSERT(disk->d_sync.ds_consumer == NULL, 1622 ("Sync consumer already exists (device=%s, disk=%s).", 1623 sc->sc_name, g_mirror_get_diskname(disk))); 1624 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1625 disk->d_sync.ds_consumer->private = disk; 1626 disk->d_sync.ds_consumer->index = 0; 1627 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1628 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1629 disk->d_softc->sc_name, error)); 1630 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1631 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1632 disk->d_softc->sc_name, error)); 1633 disk->d_sync.ds_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1634 sc->sc_sync.ds_ndisks++; 1635} 1636 1637/* 1638 * Stop synchronization process. 1639 * type: 0 - synchronization finished 1640 * 1 - synchronization stopped 1641 */ 1642static void 1643g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1644{ 1645 struct g_consumer *cp; 1646 1647 g_topology_assert(); 1648 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1649 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1650 g_mirror_disk_state2str(disk->d_state))); 1651 if (disk->d_sync.ds_consumer == NULL) 1652 return; 1653 1654 if (type == 0) { 1655 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1656 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1657 } else /* if (type == 1) */ { 1658 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1659 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1660 } 1661 cp = disk->d_sync.ds_consumer; 1662 g_access(cp, -1, 0, 0); 1663 g_mirror_kill_consumer(disk->d_softc, cp); 1664 free(disk->d_sync.ds_data, M_MIRROR); 1665 disk->d_sync.ds_consumer = NULL; 1666 disk->d_softc->sc_sync.ds_ndisks--; 1667 cp = disk->d_consumer; 1668 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 1669 ("Consumer %s not opened.", cp->provider->name)); 1670 g_access(cp, 0, -1, -1); 1671 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, -1, 1672 -1, 0); 1673 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1674} 1675 1676static void 1677g_mirror_launch_provider(struct g_mirror_softc *sc) 1678{ 1679 struct g_mirror_disk *disk; 1680 struct g_provider *pp; 1681 1682 g_topology_assert(); 1683 1684 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1685 pp->mediasize = sc->sc_mediasize; 1686 pp->sectorsize = sc->sc_sectorsize; 1687 sc->sc_provider = pp; 1688 g_error_provider(pp, 0); 1689 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1690 pp->name); 1691 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1692 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1693 g_mirror_sync_start(disk); 1694 } 1695} 1696 1697static void 1698g_mirror_destroy_provider(struct g_mirror_softc *sc) 1699{ 1700 struct g_mirror_disk *disk; 1701 struct bio *bp; 1702 1703 g_topology_assert(); 1704 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1705 sc->sc_name)); 1706 1707 g_error_provider(sc->sc_provider, ENXIO); 1708 mtx_lock(&sc->sc_queue_mtx); 1709 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1710 bioq_remove(&sc->sc_queue, bp); 1711 g_io_deliver(bp, ENXIO); 1712 } 1713 mtx_unlock(&sc->sc_queue_mtx); 1714 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1715 sc->sc_provider->name); 1716 sc->sc_provider->flags |= G_PF_WITHER; 1717 g_orphan_provider(sc->sc_provider, ENXIO); 1718 sc->sc_provider = NULL; 1719 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1720 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1721 g_mirror_sync_stop(disk, 1); 1722 } 1723} 1724 1725static void 1726g_mirror_go(void *arg) 1727{ 1728 struct g_mirror_softc *sc; 1729 1730 sc = arg; 1731 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1732 g_mirror_event_send(sc, 0, 1733 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1734} 1735 1736static u_int 1737g_mirror_determine_state(struct g_mirror_disk *disk) 1738{ 1739 struct g_mirror_softc *sc; 1740 u_int state; 1741 1742 sc = disk->d_softc; 1743 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1744 if ((disk->d_flags & 1745 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1746 /* Disk does not need synchronization. */ 1747 state = G_MIRROR_DISK_STATE_ACTIVE; 1748 } else { 1749 if ((sc->sc_flags & 1750 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1751 (disk->d_flags & 1752 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1753 /* 1754 * We can start synchronization from 1755 * the stored offset. 1756 */ 1757 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1758 } else { 1759 state = G_MIRROR_DISK_STATE_STALE; 1760 } 1761 } 1762 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1763 /* 1764 * Reset all synchronization data for this disk, 1765 * because if it even was synchronized, it was 1766 * synchronized to disks with different syncid. 1767 */ 1768 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1769 disk->d_sync.ds_offset = 0; 1770 disk->d_sync.ds_offset_done = 0; 1771 disk->d_sync.ds_syncid = sc->sc_syncid; 1772 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1773 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1774 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1775 } else { 1776 state = G_MIRROR_DISK_STATE_STALE; 1777 } 1778 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1779 /* 1780 * Not good, NOT GOOD! 1781 * It means that mirror was started on stale disks 1782 * and more fresh disk just arrive. 1783 * If there were writes, mirror is fucked up, sorry. 1784 * I think the best choice here is don't touch 1785 * this disk and inform the user laudly. 1786 */ 1787 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1788 "disk (%s) arrives!! It will not be connected to the " 1789 "running device.", sc->sc_name, 1790 g_mirror_get_diskname(disk)); 1791 g_mirror_destroy_disk(disk); 1792 state = G_MIRROR_DISK_STATE_NONE; 1793 /* Return immediately, because disk was destroyed. */ 1794 return (state); 1795 } 1796 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1797 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1798 return (state); 1799} 1800 1801/* 1802 * Update device state. 1803 */ 1804static void 1805g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1806{ 1807 struct g_mirror_disk *disk; 1808 u_int state; 1809 1810 g_topology_assert(); 1811 1812 switch (sc->sc_state) { 1813 case G_MIRROR_DEVICE_STATE_STARTING: 1814 { 1815 struct g_mirror_disk *pdisk; 1816 u_int dirty, ndisks, syncid; 1817 1818 KASSERT(sc->sc_provider == NULL, 1819 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1820 /* 1821 * Are we ready? We are, if all disks are connected or 1822 * if we have any disks and 'force' is true. 1823 */ 1824 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1825 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1826 ; 1827 } else if (g_mirror_ndisks(sc, -1) == 0) { 1828 /* 1829 * Disks went down in starting phase, so destroy 1830 * device. 1831 */ 1832 callout_drain(&sc->sc_callout); 1833 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1834 return; 1835 } else { 1836 return; 1837 } 1838 1839 /* 1840 * Activate all disks with the biggest syncid. 1841 */ 1842 if (force) { 1843 /* 1844 * If 'force' is true, we have been called due to 1845 * timeout, so don't bother canceling timeout. 1846 */ 1847 ndisks = 0; 1848 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1849 if ((disk->d_flags & 1850 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1851 ndisks++; 1852 } 1853 } 1854 if (ndisks == 0) { 1855 /* No valid disks found, destroy device. */ 1856 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1857 return; 1858 } 1859 } else { 1860 /* Cancel timeout. */ 1861 callout_drain(&sc->sc_callout); 1862 } 1863 1864 /* 1865 * Find disk with the biggest syncid. 1866 */ 1867 syncid = 0; 1868 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1869 if (disk->d_sync.ds_syncid > syncid) 1870 syncid = disk->d_sync.ds_syncid; 1871 } 1872 1873 /* 1874 * Here we need to look for dirty disks and if all disks 1875 * with the biggest syncid are dirty, we have to choose 1876 * one with the biggest priority and rebuild the rest. 1877 */ 1878 /* 1879 * Find the number of dirty disks with the biggest syncid. 1880 * Find the number of disks with the biggest syncid. 1881 * While here, find a disk with the biggest priority. 1882 */ 1883 dirty = ndisks = 0; 1884 pdisk = NULL; 1885 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1886 if (disk->d_sync.ds_syncid != syncid) 1887 continue; 1888 if ((disk->d_flags & 1889 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1890 continue; 1891 } 1892 ndisks++; 1893 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1894 dirty++; 1895 if (pdisk == NULL || 1896 pdisk->d_priority < disk->d_priority) { 1897 pdisk = disk; 1898 } 1899 } 1900 } 1901 if (dirty == 0) { 1902 /* No dirty disks at all, great. */ 1903 } else if (dirty == ndisks) { 1904 /* 1905 * Force synchronization for all dirty disks except one 1906 * with the biggest priority. 1907 */ 1908 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1909 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1910 "master disk for synchronization.", 1911 g_mirror_get_diskname(pdisk), sc->sc_name); 1912 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1913 if (disk->d_sync.ds_syncid != syncid) 1914 continue; 1915 if ((disk->d_flags & 1916 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1917 continue; 1918 } 1919 KASSERT((disk->d_flags & 1920 G_MIRROR_DISK_FLAG_DIRTY) != 0, 1921 ("Disk %s isn't marked as dirty.", 1922 g_mirror_get_diskname(disk))); 1923 /* Skip the disk with the biggest priority. */ 1924 if (disk == pdisk) 1925 continue; 1926 disk->d_sync.ds_syncid = 0; 1927 } 1928 } else if (dirty < ndisks) { 1929 /* 1930 * Force synchronization for all dirty disks. 1931 * We have some non-dirty disks. 1932 */ 1933 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1934 if (disk->d_sync.ds_syncid != syncid) 1935 continue; 1936 if ((disk->d_flags & 1937 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1938 continue; 1939 } 1940 if ((disk->d_flags & 1941 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1942 continue; 1943 } 1944 disk->d_sync.ds_syncid = 0; 1945 } 1946 } 1947 1948 /* Reset hint. */ 1949 sc->sc_hint = NULL; 1950 sc->sc_syncid = syncid; 1951 if (force) { 1952 /* Remember to bump syncid on first write. */ 1953 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1954 } 1955 state = G_MIRROR_DEVICE_STATE_RUNNING; 1956 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 1957 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 1958 g_mirror_device_state2str(state)); 1959 sc->sc_state = state; 1960 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1961 state = g_mirror_determine_state(disk); 1962 g_mirror_event_send(disk, state, 1963 G_MIRROR_EVENT_DONTWAIT); 1964 if (state == G_MIRROR_DISK_STATE_STALE) { 1965 sc->sc_bump_syncid = 1966 G_MIRROR_BUMP_ON_FIRST_WRITE; 1967 } 1968 } 1969 wakeup(&g_mirror_class); 1970 break; 1971 } 1972 case G_MIRROR_DEVICE_STATE_RUNNING: 1973 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 1974 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 1975 /* 1976 * No active disks or no disks at all, 1977 * so destroy device. 1978 */ 1979 if (sc->sc_provider != NULL) 1980 g_mirror_destroy_provider(sc); 1981 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1982 break; 1983 } else if (g_mirror_ndisks(sc, 1984 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 1985 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 1986 /* 1987 * We have active disks, launch provider if it doesn't 1988 * exist. 1989 */ 1990 if (sc->sc_provider == NULL) 1991 g_mirror_launch_provider(sc); 1992 } 1993 /* 1994 * Bump syncid here, if we need to do it immediately. 1995 */ 1996 if (sc->sc_bump_syncid == G_MIRROR_BUMP_IMMEDIATELY) { 1997 sc->sc_bump_syncid = 0; 1998 g_mirror_bump_syncid(sc); 1999 } 2000 break; 2001 default: 2002 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2003 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2004 break; 2005 } 2006} 2007 2008/* 2009 * Update disk state and device state if needed. 2010 */ 2011#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2012 "Disk %s state changed from %s to %s (device %s).", \ 2013 g_mirror_get_diskname(disk), \ 2014 g_mirror_disk_state2str(disk->d_state), \ 2015 g_mirror_disk_state2str(state), sc->sc_name) 2016static int 2017g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2018{ 2019 struct g_mirror_softc *sc; 2020 2021 g_topology_assert(); 2022 2023 sc = disk->d_softc; 2024again: 2025 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2026 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2027 g_mirror_disk_state2str(state)); 2028 switch (state) { 2029 case G_MIRROR_DISK_STATE_NEW: 2030 /* 2031 * Possible scenarios: 2032 * 1. New disk arrive. 2033 */ 2034 /* Previous state should be NONE. */ 2035 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2036 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2037 g_mirror_disk_state2str(disk->d_state))); 2038 DISK_STATE_CHANGED(); 2039 2040 disk->d_state = state; 2041 if (LIST_EMPTY(&sc->sc_disks)) 2042 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2043 else { 2044 struct g_mirror_disk *dp; 2045 2046 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2047 if (disk->d_priority >= dp->d_priority) { 2048 LIST_INSERT_BEFORE(dp, disk, d_next); 2049 dp = NULL; 2050 break; 2051 } 2052 if (LIST_NEXT(dp, d_next) == NULL) 2053 break; 2054 } 2055 if (dp != NULL) 2056 LIST_INSERT_AFTER(dp, disk, d_next); 2057 } 2058 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2059 sc->sc_name, g_mirror_get_diskname(disk)); 2060 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2061 break; 2062 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2063 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2064 g_mirror_device_state2str(sc->sc_state), 2065 g_mirror_get_diskname(disk), 2066 g_mirror_disk_state2str(disk->d_state))); 2067 state = g_mirror_determine_state(disk); 2068 if (state != G_MIRROR_DISK_STATE_NONE) 2069 goto again; 2070 break; 2071 case G_MIRROR_DISK_STATE_ACTIVE: 2072 /* 2073 * Possible scenarios: 2074 * 1. New disk does not need synchronization. 2075 * 2. Synchronization process finished successfully. 2076 */ 2077 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2078 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2079 g_mirror_device_state2str(sc->sc_state), 2080 g_mirror_get_diskname(disk), 2081 g_mirror_disk_state2str(disk->d_state))); 2082 /* Previous state should be NEW or SYNCHRONIZING. */ 2083 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2084 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2085 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2086 g_mirror_disk_state2str(disk->d_state))); 2087 DISK_STATE_CHANGED(); 2088 2089 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2090 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2091 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2092 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2093 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2094 g_mirror_sync_stop(disk, 0); 2095 } 2096 disk->d_state = state; 2097 disk->d_sync.ds_offset = 0; 2098 disk->d_sync.ds_offset_done = 0; 2099 g_mirror_update_access(disk); 2100 g_mirror_update_metadata(disk); 2101 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2102 sc->sc_name, g_mirror_get_diskname(disk)); 2103 break; 2104 case G_MIRROR_DISK_STATE_STALE: 2105 /* 2106 * Possible scenarios: 2107 * 1. Stale disk was connected. 2108 */ 2109 /* Previous state should be NEW. */ 2110 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2111 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2112 g_mirror_disk_state2str(disk->d_state))); 2113 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2114 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2115 g_mirror_device_state2str(sc->sc_state), 2116 g_mirror_get_diskname(disk), 2117 g_mirror_disk_state2str(disk->d_state))); 2118 /* 2119 * STALE state is only possible if device is marked 2120 * NOAUTOSYNC. 2121 */ 2122 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2123 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2124 g_mirror_device_state2str(sc->sc_state), 2125 g_mirror_get_diskname(disk), 2126 g_mirror_disk_state2str(disk->d_state))); 2127 DISK_STATE_CHANGED(); 2128 2129 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2130 disk->d_state = state; 2131 g_mirror_update_metadata(disk); 2132 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2133 sc->sc_name, g_mirror_get_diskname(disk)); 2134 break; 2135 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2136 /* 2137 * Possible scenarios: 2138 * 1. Disk which needs synchronization was connected. 2139 */ 2140 /* Previous state should be NEW. */ 2141 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2142 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2143 g_mirror_disk_state2str(disk->d_state))); 2144 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2145 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2146 g_mirror_device_state2str(sc->sc_state), 2147 g_mirror_get_diskname(disk), 2148 g_mirror_disk_state2str(disk->d_state))); 2149 DISK_STATE_CHANGED(); 2150 2151 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2152 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2153 disk->d_state = state; 2154 if (sc->sc_provider != NULL) { 2155 g_mirror_sync_start(disk); 2156 g_mirror_update_metadata(disk); 2157 } 2158 break; 2159 case G_MIRROR_DISK_STATE_DISCONNECTED: 2160 /* 2161 * Possible scenarios: 2162 * 1. Device wasn't running yet, but disk disappear. 2163 * 2. Disk was active and disapppear. 2164 * 3. Disk disappear during synchronization process. 2165 */ 2166 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2167 /* 2168 * Previous state should be ACTIVE, STALE or 2169 * SYNCHRONIZING. 2170 */ 2171 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2172 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2173 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2174 ("Wrong disk state (%s, %s).", 2175 g_mirror_get_diskname(disk), 2176 g_mirror_disk_state2str(disk->d_state))); 2177 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2178 /* Previous state should be NEW. */ 2179 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2180 ("Wrong disk state (%s, %s).", 2181 g_mirror_get_diskname(disk), 2182 g_mirror_disk_state2str(disk->d_state))); 2183 /* 2184 * Reset bumping syncid if disk disappeared in STARTING 2185 * state. 2186 */ 2187 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) 2188 sc->sc_bump_syncid = 0; 2189#ifdef INVARIANTS 2190 } else { 2191 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2192 sc->sc_name, 2193 g_mirror_device_state2str(sc->sc_state), 2194 g_mirror_get_diskname(disk), 2195 g_mirror_disk_state2str(disk->d_state))); 2196#endif 2197 } 2198 DISK_STATE_CHANGED(); 2199 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2200 sc->sc_name, g_mirror_get_diskname(disk)); 2201 2202 g_mirror_destroy_disk(disk); 2203 break; 2204 case G_MIRROR_DISK_STATE_DESTROY: 2205 { 2206 int error; 2207 2208 error = g_mirror_clear_metadata(disk); 2209 if (error != 0) 2210 return (error); 2211 DISK_STATE_CHANGED(); 2212 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2213 sc->sc_name, g_mirror_get_diskname(disk)); 2214 2215 g_mirror_destroy_disk(disk); 2216 sc->sc_ndisks--; 2217 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2218 g_mirror_update_metadata(disk); 2219 } 2220 break; 2221 } 2222 default: 2223 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2224 break; 2225 } 2226 return (0); 2227} 2228#undef DISK_STATE_CHANGED 2229 2230static int 2231g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2232{ 2233 struct g_provider *pp; 2234 u_char *buf; 2235 int error; 2236 2237 g_topology_assert(); 2238 2239 error = g_access(cp, 1, 0, 0); 2240 if (error != 0) 2241 return (error); 2242 pp = cp->provider; 2243 g_topology_unlock(); 2244 /* Metadata are stored on last sector. */ 2245 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2246 &error); 2247 g_topology_lock(); 2248 if (buf == NULL) { 2249 g_access(cp, -1, 0, 0); 2250 return (error); 2251 } 2252 if (error != 0) { 2253 g_access(cp, -1, 0, 0); 2254 g_free(buf); 2255 return (error); 2256 } 2257 error = g_access(cp, -1, 0, 0); 2258 KASSERT(error == 0, ("Cannot decrease access count for %s.", pp->name)); 2259 2260 /* Decode metadata. */ 2261 error = mirror_metadata_decode(buf, md); 2262 g_free(buf); 2263 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2264 return (EINVAL); 2265 if (error != 0) { 2266 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2267 cp->provider->name); 2268 return (error); 2269 } 2270 2271 return (0); 2272} 2273 2274static int 2275g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2276 struct g_mirror_metadata *md) 2277{ 2278 2279 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2280 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2281 pp->name, md->md_did); 2282 return (EEXIST); 2283 } 2284 if (md->md_all != sc->sc_ndisks) { 2285 G_MIRROR_DEBUG(1, 2286 "Invalid '%s' field on disk %s (device %s), skipping.", 2287 "md_all", pp->name, sc->sc_name); 2288 return (EINVAL); 2289 } 2290 if (md->md_slice != sc->sc_slice) { 2291 G_MIRROR_DEBUG(1, 2292 "Invalid '%s' field on disk %s (device %s), skipping.", 2293 "md_slice", pp->name, sc->sc_name); 2294 return (EINVAL); 2295 } 2296 if (md->md_balance != sc->sc_balance) { 2297 G_MIRROR_DEBUG(1, 2298 "Invalid '%s' field on disk %s (device %s), skipping.", 2299 "md_balance", pp->name, sc->sc_name); 2300 return (EINVAL); 2301 } 2302 if (md->md_mediasize != sc->sc_mediasize) { 2303 G_MIRROR_DEBUG(1, 2304 "Invalid '%s' field on disk %s (device %s), skipping.", 2305 "md_mediasize", pp->name, sc->sc_name); 2306 return (EINVAL); 2307 } 2308 if (sc->sc_mediasize > pp->mediasize) { 2309 G_MIRROR_DEBUG(1, 2310 "Invalid size of disk %s (device %s), skipping.", pp->name, 2311 sc->sc_name); 2312 return (EINVAL); 2313 } 2314 if (md->md_sectorsize != sc->sc_sectorsize) { 2315 G_MIRROR_DEBUG(1, 2316 "Invalid '%s' field on disk %s (device %s), skipping.", 2317 "md_sectorsize", pp->name, sc->sc_name); 2318 return (EINVAL); 2319 } 2320 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2321 G_MIRROR_DEBUG(1, 2322 "Invalid sector size of disk %s (device %s), skipping.", 2323 pp->name, sc->sc_name); 2324 return (EINVAL); 2325 } 2326 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2327 G_MIRROR_DEBUG(1, 2328 "Invalid device flags on disk %s (device %s), skipping.", 2329 pp->name, sc->sc_name); 2330 return (EINVAL); 2331 } 2332 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2333 G_MIRROR_DEBUG(1, 2334 "Invalid disk flags on disk %s (device %s), skipping.", 2335 pp->name, sc->sc_name); 2336 return (EINVAL); 2337 } 2338 return (0); 2339} 2340 2341static int 2342g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2343 struct g_mirror_metadata *md) 2344{ 2345 struct g_mirror_disk *disk; 2346 int error; 2347 2348 g_topology_assert(); 2349 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2350 2351 error = g_mirror_check_metadata(sc, pp, md); 2352 if (error != 0) 2353 return (error); 2354 disk = g_mirror_init_disk(sc, pp, md, &error); 2355 if (disk == NULL) 2356 return (error); 2357 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2358 G_MIRROR_EVENT_WAIT); 2359 return (error); 2360} 2361 2362static int 2363g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2364{ 2365 struct g_mirror_softc *sc; 2366 struct g_mirror_disk *disk; 2367 int dcr, dcw, dce, err, error; 2368 2369 g_topology_assert(); 2370 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2371 acw, ace); 2372 2373 dcr = pp->acr + acr; 2374 dcw = pp->acw + acw; 2375 dce = pp->ace + ace; 2376 2377 /* On first open, grab an extra "exclusive" bit */ 2378 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0) 2379 ace++; 2380 /* ... and let go of it on last close */ 2381 if (dcr == 0 && dcw == 0 && dce == 0) 2382 ace--; 2383 2384 sc = pp->geom->softc; 2385 if (sc == NULL || LIST_EMPTY(&sc->sc_disks)) { 2386 if (acr <= 0 && acw <= 0 && ace <= 0) 2387 return (0); 2388 else 2389 return (ENXIO); 2390 } 2391 error = ENXIO; 2392 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2393 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2394 continue; 2395 err = g_access(disk->d_consumer, acr, acw, ace); 2396 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 2397 g_mirror_get_diskname(disk), acr, acw, ace, err); 2398 if (err == 0) { 2399 /* 2400 * Mark disk as dirty on open and unmark on close. 2401 */ 2402 if (pp->acw == 0 && dcw > 0) { 2403 G_MIRROR_DEBUG(1, 2404 "Disk %s (device %s) marked as dirty.", 2405 g_mirror_get_diskname(disk), sc->sc_name); 2406 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2407 g_mirror_update_metadata(disk); 2408 } else if (pp->acw > 0 && dcw == 0) { 2409 G_MIRROR_DEBUG(1, 2410 "Disk %s (device %s) marked as clean.", 2411 g_mirror_get_diskname(disk), sc->sc_name); 2412 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2413 g_mirror_update_metadata(disk); 2414 } 2415 error = 0; 2416 } else { 2417 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 2418 g_mirror_event_send(disk, 2419 G_MIRROR_DISK_STATE_DISCONNECTED, 2420 G_MIRROR_EVENT_DONTWAIT); 2421 } 2422 } 2423 /* 2424 * Be sure to return 0 for negativate access requests. 2425 * In case of some HW problems, it is possible that we don't have 2426 * any active disk here, so loop above will be no-op and error will 2427 * be ENXIO. 2428 */ 2429 if (error != 0 && acr <= 0 && acw <= 0 && ace <= 0) 2430 error = 0; 2431 return (error); 2432} 2433 2434static struct g_geom * 2435g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2436{ 2437 struct g_mirror_softc *sc; 2438 struct g_geom *gp; 2439 int error, timeout; 2440 2441 g_topology_assert(); 2442 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2443 md->md_mid); 2444 2445 /* One disk is minimum. */ 2446 if (md->md_all < 1) 2447 return (NULL); 2448 /* 2449 * Action geom. 2450 */ 2451 gp = g_new_geomf(mp, "%s", md->md_name); 2452 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2453 gp->start = g_mirror_start; 2454 gp->spoiled = g_mirror_spoiled; 2455 gp->orphan = g_mirror_orphan; 2456 gp->access = g_mirror_access; 2457 gp->dumpconf = g_mirror_dumpconf; 2458 2459 sc->sc_id = md->md_mid; 2460 sc->sc_slice = md->md_slice; 2461 sc->sc_balance = md->md_balance; 2462 sc->sc_mediasize = md->md_mediasize; 2463 sc->sc_sectorsize = md->md_sectorsize; 2464 sc->sc_ndisks = md->md_all; 2465 sc->sc_flags = md->md_mflags; 2466 sc->sc_bump_syncid = 0; 2467 sc->sc_idle = 0; 2468 bioq_init(&sc->sc_queue); 2469 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2470 LIST_INIT(&sc->sc_disks); 2471 TAILQ_INIT(&sc->sc_events); 2472 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2473 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2474 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2475 gp->softc = sc; 2476 sc->sc_geom = gp; 2477 sc->sc_provider = NULL; 2478 /* 2479 * Synchronization geom. 2480 */ 2481 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2482 gp->softc = sc; 2483 gp->orphan = g_mirror_orphan; 2484 sc->sc_sync.ds_geom = gp; 2485 sc->sc_sync.ds_ndisks = 0; 2486 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2487 "g_mirror %s", md->md_name); 2488 if (error != 0) { 2489 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2490 sc->sc_name); 2491 g_destroy_geom(sc->sc_sync.ds_geom); 2492 mtx_destroy(&sc->sc_events_mtx); 2493 mtx_destroy(&sc->sc_queue_mtx); 2494 g_destroy_geom(sc->sc_geom); 2495 free(sc, M_MIRROR); 2496 return (NULL); 2497 } 2498 2499 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2500 2501 /* 2502 * Run timeout. 2503 */ 2504 timeout = atomic_load_acq_int(&g_mirror_timeout); 2505 callout_reset(&sc->sc_callout, timeout * hz, g_mirror_go, sc); 2506 return (sc->sc_geom); 2507} 2508 2509int 2510g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2511{ 2512 struct g_provider *pp; 2513 2514 g_topology_assert(); 2515 2516 if (sc == NULL) 2517 return (ENXIO); 2518 pp = sc->sc_provider; 2519 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2520 if (force) { 2521 G_MIRROR_DEBUG(0, "Device %s is still open, so it " 2522 "can't be definitely removed.", pp->name); 2523 } else { 2524 G_MIRROR_DEBUG(1, 2525 "Device %s is still open (r%dw%de%d).", pp->name, 2526 pp->acr, pp->acw, pp->ace); 2527 return (EBUSY); 2528 } 2529 } 2530 2531 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2532 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2533 g_topology_unlock(); 2534 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2535 mtx_lock(&sc->sc_queue_mtx); 2536 wakeup(sc); 2537 mtx_unlock(&sc->sc_queue_mtx); 2538 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2539 while (sc->sc_worker != NULL) 2540 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2541 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2542 g_topology_lock(); 2543 g_mirror_destroy_device(sc); 2544 free(sc, M_MIRROR); 2545 return (0); 2546} 2547 2548static void 2549g_mirror_taste_orphan(struct g_consumer *cp) 2550{ 2551 2552 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2553 cp->provider->name)); 2554} 2555 2556static struct g_geom * 2557g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2558{ 2559 struct g_mirror_metadata md; 2560 struct g_mirror_softc *sc; 2561 struct g_consumer *cp; 2562 struct g_geom *gp; 2563 int error; 2564 2565 g_topology_assert(); 2566 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2567 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2568 2569 gp = g_new_geomf(mp, "mirror:taste"); 2570 /* 2571 * This orphan function should be never called. 2572 */ 2573 gp->orphan = g_mirror_taste_orphan; 2574 cp = g_new_consumer(gp); 2575 g_attach(cp, pp); 2576 error = g_mirror_read_metadata(cp, &md); 2577 g_detach(cp); 2578 g_destroy_consumer(cp); 2579 g_destroy_geom(gp); 2580 if (error != 0) 2581 return (NULL); 2582 gp = NULL; 2583 2584 if (md.md_version > G_MIRROR_VERSION) { 2585 printf("geom_mirror.ko module is too old to handle %s.\n", 2586 pp->name); 2587 return (NULL); 2588 } 2589 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2590 return (NULL); 2591 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2592 G_MIRROR_DEBUG(0, 2593 "Device %s: provider %s marked as inactive, skipping.", 2594 md.md_name, pp->name); 2595 return (NULL); 2596 } 2597 if (g_mirror_debug >= 2) 2598 mirror_metadata_dump(&md); 2599 2600 /* 2601 * Let's check if device already exists. 2602 */ 2603 sc = NULL; 2604 LIST_FOREACH(gp, &mp->geom, geom) { 2605 sc = gp->softc; 2606 if (sc == NULL) 2607 continue; 2608 if (sc->sc_sync.ds_geom == gp) 2609 continue; 2610 if (strcmp(md.md_name, sc->sc_name) != 0) 2611 continue; 2612 if (md.md_mid != sc->sc_id) { 2613 G_MIRROR_DEBUG(0, "Device %s already configured.", 2614 sc->sc_name); 2615 return (NULL); 2616 } 2617 break; 2618 } 2619 if (gp == NULL) { 2620 gp = g_mirror_create(mp, &md); 2621 if (gp == NULL) { 2622 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2623 md.md_name); 2624 return (NULL); 2625 } 2626 sc = gp->softc; 2627 } 2628 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2629 error = g_mirror_add_disk(sc, pp, &md); 2630 if (error != 0) { 2631 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2632 pp->name, gp->name, error); 2633 if (LIST_EMPTY(&sc->sc_disks)) 2634 g_mirror_destroy(sc, 1); 2635 return (NULL); 2636 } 2637 return (gp); 2638} 2639 2640static int 2641g_mirror_destroy_geom(struct gctl_req *req __unused, 2642 struct g_class *mp __unused, struct g_geom *gp) 2643{ 2644 2645 return (g_mirror_destroy(gp->softc, 0)); 2646} 2647 2648static void 2649g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2650 struct g_consumer *cp, struct g_provider *pp) 2651{ 2652 struct g_mirror_softc *sc; 2653 2654 g_topology_assert(); 2655 2656 sc = gp->softc; 2657 if (sc == NULL) 2658 return; 2659 /* Skip synchronization geom. */ 2660 if (gp == sc->sc_sync.ds_geom) 2661 return; 2662 if (pp != NULL) { 2663 /* Nothing here. */ 2664 } else if (cp != NULL) { 2665 struct g_mirror_disk *disk; 2666 2667 disk = cp->private; 2668 if (disk == NULL) 2669 return; 2670 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2671 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2672 sbuf_printf(sb, "%s<Synchronized>", indent); 2673 if (disk->d_sync.ds_offset_done == 0) 2674 sbuf_printf(sb, "0%%"); 2675 else { 2676 sbuf_printf(sb, "%u%%", 2677 (u_int)((disk->d_sync.ds_offset_done * 100) / 2678 sc->sc_provider->mediasize)); 2679 } 2680 sbuf_printf(sb, "</Synchronized>\n"); 2681 } 2682 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2683 disk->d_sync.ds_syncid); 2684 sbuf_printf(sb, "%s<Flags>", indent); 2685 if (disk->d_flags == 0) 2686 sbuf_printf(sb, "NONE"); 2687 else { 2688 int first = 1; 2689 2690#define ADD_FLAG(flag, name) do { \ 2691 if ((disk->d_flags & (flag)) != 0) { \ 2692 if (!first) \ 2693 sbuf_printf(sb, ", "); \ 2694 else \ 2695 first = 0; \ 2696 sbuf_printf(sb, name); \ 2697 } \ 2698} while (0) 2699 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2700 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2701 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2702 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2703 "SYNCHRONIZING"); 2704 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2705#undef ADD_FLAG 2706 } 2707 sbuf_printf(sb, "</Flags>\n"); 2708 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2709 disk->d_priority); 2710 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2711 g_mirror_disk_state2str(disk->d_state)); 2712 } else { 2713 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2714 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2715 sbuf_printf(sb, "%s<Flags>", indent); 2716 if (sc->sc_flags == 0) 2717 sbuf_printf(sb, "NONE"); 2718 else { 2719 int first = 1; 2720 2721#define ADD_FLAG(flag, name) do { \ 2722 if ((sc->sc_flags & (flag)) != 0) { \ 2723 if (!first) \ 2724 sbuf_printf(sb, ", "); \ 2725 else \ 2726 first = 0; \ 2727 sbuf_printf(sb, name); \ 2728 } \ 2729} while (0) 2730 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2731#undef ADD_FLAG 2732 } 2733 sbuf_printf(sb, "</Flags>\n"); 2734 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2735 (u_int)sc->sc_slice); 2736 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2737 balance_name(sc->sc_balance)); 2738 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2739 sc->sc_ndisks); 2740 sbuf_printf(sb, "%s<State>", indent); 2741 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2742 sbuf_printf(sb, "%s", "STARTING"); 2743 else if (sc->sc_ndisks == 2744 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 2745 sbuf_printf(sb, "%s", "COMPLETE"); 2746 else 2747 sbuf_printf(sb, "%s", "DEGRADED"); 2748 sbuf_printf(sb, "</State>\n"); 2749 } 2750} 2751 2752static int 2753g_mirror_can_go(void) 2754{ 2755 struct g_mirror_softc *sc; 2756 struct g_geom *gp; 2757 struct g_provider *pp; 2758 int can_go; 2759 2760 DROP_GIANT(); 2761 can_go = 1; 2762 g_topology_lock(); 2763 LIST_FOREACH(gp, &g_mirror_class.geom, geom) { 2764 sc = gp->softc; 2765 if (sc == NULL) { 2766 can_go = 0; 2767 break; 2768 } 2769 pp = sc->sc_provider; 2770 if (pp == NULL || pp->error != 0) { 2771 can_go = 0; 2772 break; 2773 } 2774 } 2775 g_topology_unlock(); 2776 PICKUP_GIANT(); 2777 return (can_go); 2778} 2779 2780static void 2781g_mirror_rootwait(void) 2782{ 2783 2784 /* 2785 * HACK: Wait for GEOM, because g_mirror_rootwait() can be called, 2786 * HACK: before we get providers for tasting. 2787 */ 2788 tsleep(&g_mirror_class, PRIBIO, "mroot", hz * 3); 2789 /* 2790 * Wait for mirrors in degraded state. 2791 */ 2792 for (;;) { 2793 if (g_mirror_can_go()) 2794 break; 2795 tsleep(&g_mirror_class, PRIBIO, "mroot", hz); 2796 } 2797} 2798 2799SYSINIT(g_mirror_root, SI_SUB_RAID, SI_ORDER_FIRST, g_mirror_rootwait, NULL) 2800 2801DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2802