g_mirror.c revision 133170
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 133170 2004-08-05 13:22:17Z 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; 54SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RW, &g_mirror_debug, 0, 55 "Debug level"); 56static u_int g_mirror_sync_block_size = 131072; 57SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_block_size, CTLFLAG_RW, 58 &g_mirror_sync_block_size, 0, "Synchronization block size"); 59static u_int g_mirror_timeout = 8; 60SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RW, &g_mirror_timeout, 61 0, "Time to wait on all mirror components"); 62static u_int g_mirror_reqs_per_sync = 5; 63SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, reqs_per_sync, CTLFLAG_RW, 64 &g_mirror_reqs_per_sync, 0, 65 "Number of regular I/O requests per synchronization request"); 66static u_int g_mirror_syncs_per_sec = 100; 67SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, syncs_per_sec, CTLFLAG_RW, 68 &g_mirror_syncs_per_sec, 0, 69 "Number of synchronizations requests per second"); 70 71#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 72 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 73 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 74 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 75} while (0) 76 77 78static int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp, 79 struct g_geom *gp); 80static g_taste_t g_mirror_taste; 81 82struct g_class g_mirror_class = { 83 .name = G_MIRROR_CLASS_NAME, 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 void 291g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 292{ 293 294 g_topology_assert(); 295 296 cp->private = NULL; 297 if (cp->nstart != cp->nend) { 298 G_MIRROR_DEBUG(2, 299 "I/O requests for %s exist, can't destroy it now.", 300 cp->provider->name); 301 return; 302 } 303 if (g_mirror_nrequests(sc, cp) > 0) { 304 G_MIRROR_DEBUG(2, 305 "I/O requests for %s in queue, can't destroy it now.", 306 cp->provider->name); 307 return; 308 } 309 G_MIRROR_DEBUG(2, "Consumer %s destroyed.", cp->provider->name); 310 g_detach(cp); 311 g_destroy_consumer(cp); 312} 313 314static int 315g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 316{ 317 int error; 318 319 g_topology_assert(); 320 KASSERT(disk->d_consumer == NULL, 321 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 322 323 disk->d_consumer = g_new_consumer(disk->d_softc->sc_geom); 324 disk->d_consumer->private = disk; 325 error = g_attach(disk->d_consumer, pp); 326 if (error != 0) 327 return (error); 328 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 329 return (0); 330} 331 332static void 333g_mirror_disconnect_disk(struct g_mirror_disk *disk) 334{ 335 struct g_consumer *cp; 336 337 g_topology_assert(); 338 339 cp = disk->d_consumer; 340 if (cp == NULL) 341 return; 342 if (cp->provider != NULL) { 343 G_MIRROR_DEBUG(2, "Disk %s disconnected.", 344 g_mirror_get_diskname(disk)); 345 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) { 346 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 347 cp->provider->name, -cp->acr, -cp->acw, -cp->ace, 348 0); 349 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 350 } 351 g_mirror_kill_consumer(disk->d_softc, cp); 352 } else { 353 g_destroy_consumer(cp); 354 } 355} 356 357/* 358 * Initialize disk. This means allocate memory, create consumer, attach it 359 * to the provider and open access (r1w1e1) to it. 360 */ 361static struct g_mirror_disk * 362g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 363 struct g_mirror_metadata *md, int *errorp) 364{ 365 struct g_mirror_disk *disk; 366 int error; 367 368 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 369 if (disk == NULL) { 370 error = ENOMEM; 371 goto fail; 372 } 373 disk->d_softc = sc; 374 error = g_mirror_connect_disk(disk, pp); 375 if (error != 0) 376 goto fail; 377 disk->d_id = md->md_did; 378 disk->d_state = G_MIRROR_DISK_STATE_NONE; 379 disk->d_priority = md->md_priority; 380 disk->d_delay.sec = 0; 381 disk->d_delay.frac = 0; 382 binuptime(&disk->d_last_used); 383 disk->d_flags = md->md_dflags; 384 disk->d_sync.ds_consumer = NULL; 385 disk->d_sync.ds_offset = md->md_sync_offset; 386 disk->d_sync.ds_offset_done = md->md_sync_offset; 387 disk->d_sync.ds_syncid = md->md_syncid; 388 if (errorp != NULL) 389 *errorp = 0; 390 return (disk); 391fail: 392 if (errorp != NULL) 393 *errorp = error; 394 if (disk != NULL) { 395 g_mirror_disconnect_disk(disk); 396 free(disk, M_MIRROR); 397 } 398 return (NULL); 399} 400 401static void 402g_mirror_destroy_disk(struct g_mirror_disk *disk) 403{ 404 struct g_mirror_softc *sc; 405 406 g_topology_assert(); 407 408 LIST_REMOVE(disk, d_next); 409 g_mirror_event_cancel(disk); 410 sc = disk->d_softc; 411 if (sc->sc_hint == disk) 412 sc->sc_hint = NULL; 413 switch (disk->d_state) { 414 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 415 g_mirror_sync_stop(disk, 1); 416 /* FALLTHROUGH */ 417 case G_MIRROR_DISK_STATE_NEW: 418 case G_MIRROR_DISK_STATE_STALE: 419 case G_MIRROR_DISK_STATE_ACTIVE: 420 g_mirror_disconnect_disk(disk); 421 free(disk, M_MIRROR); 422 break; 423 default: 424 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 425 g_mirror_get_diskname(disk), 426 g_mirror_disk_state2str(disk->d_state))); 427 } 428} 429 430static void 431g_mirror_destroy_device(struct g_mirror_softc *sc) 432{ 433 struct g_mirror_disk *disk; 434 struct g_mirror_event *ep; 435 struct g_geom *gp; 436 struct g_consumer *cp; 437 438 g_topology_assert(); 439 440 gp = sc->sc_geom; 441 if (sc->sc_provider != NULL) 442 g_mirror_destroy_provider(sc); 443 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 444 disk = LIST_FIRST(&sc->sc_disks)) { 445 g_mirror_destroy_disk(disk); 446 } 447 while ((ep = g_mirror_event_get(sc)) != NULL) { 448 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 449 g_mirror_event_free(ep); 450 else { 451 ep->e_error = ECANCELED; 452 ep->e_flags |= G_MIRROR_EVENT_DONE; 453 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 454 mtx_lock(&sc->sc_events_mtx); 455 wakeup(ep); 456 mtx_unlock(&sc->sc_events_mtx); 457 } 458 } 459 callout_drain(&sc->sc_callout); 460 gp->softc = NULL; 461 uma_zdestroy(sc->sc_sync.ds_zone); 462 while ((cp = LIST_FIRST(&sc->sc_sync.ds_geom->consumer)) != NULL) { 463 if (cp->provider != NULL) { 464 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 465 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 466 g_detach(cp); 467 } 468 g_destroy_consumer(cp); 469 } 470 sc->sc_sync.ds_geom->softc = NULL; 471 g_destroy_geom(sc->sc_sync.ds_geom); 472 mtx_destroy(&sc->sc_queue_mtx); 473 mtx_destroy(&sc->sc_events_mtx); 474 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 475 g_wither_geom(gp, ENXIO); 476} 477 478static void 479g_mirror_orphan(struct g_consumer *cp) 480{ 481 struct g_mirror_disk *disk; 482 483 g_topology_assert(); 484 485 disk = cp->private; 486 if (disk == NULL) 487 return; 488 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 489 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 490 G_MIRROR_EVENT_DONTWAIT); 491} 492 493static void 494g_mirror_spoiled(struct g_consumer *cp) 495{ 496 struct g_mirror_disk *disk; 497 498 g_topology_assert(); 499 500 disk = cp->private; 501 if (disk == NULL) 502 return; 503 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 504 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 505 G_MIRROR_EVENT_DONTWAIT); 506} 507 508/* 509 * Function should return the next active disk on the list. 510 * It is possible that it will be the same disk as given. 511 * If there are no active disks on list, NULL is returned. 512 */ 513static __inline struct g_mirror_disk * 514g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 515{ 516 struct g_mirror_disk *dp; 517 518 for (dp = LIST_NEXT(disk, d_next); dp != disk; 519 dp = LIST_NEXT(dp, d_next)) { 520 if (dp == NULL) 521 dp = LIST_FIRST(&sc->sc_disks); 522 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 523 break; 524 } 525 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 526 return (NULL); 527 return (dp); 528} 529 530static struct g_mirror_disk * 531g_mirror_get_disk(struct g_mirror_softc *sc) 532{ 533 struct g_mirror_disk *disk; 534 535 if (sc->sc_hint == NULL) { 536 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 537 if (sc->sc_hint == NULL) 538 return (NULL); 539 } 540 disk = sc->sc_hint; 541 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 542 disk = g_mirror_find_next(sc, disk); 543 if (disk == NULL) 544 return (NULL); 545 } 546 sc->sc_hint = g_mirror_find_next(sc, disk); 547 return (disk); 548} 549 550static int 551g_mirror_clear_metadata(struct g_mirror_disk *disk) 552{ 553 struct g_mirror_softc *sc; 554 struct g_consumer *cp; 555 off_t offset, length; 556 u_char *sector; 557 int close = 0, error = 0; 558 559 g_topology_assert(); 560 561 sc = disk->d_softc; 562 cp = disk->d_consumer; 563 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 564 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 565 length = cp->provider->sectorsize; 566 offset = cp->provider->mediasize - length; 567 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 568 /* 569 * Open consumer if it wasn't opened and remember to close it. 570 */ 571 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 572 error = g_access(cp, 0, 1, 1); 573 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 574 cp->provider->name, 0, 1, 1, error); 575 if (error == 0) 576 close = 1; 577#ifdef INVARIANTS 578 } else { 579 KASSERT(cp->acw > 0 && cp->ace > 0, 580 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 581 cp->acr, cp->acw, cp->ace)); 582#endif 583 } 584 if (error == 0) { 585 g_topology_unlock(); 586 error = g_write_data(cp, offset, sector, length); 587 g_topology_lock(); 588 } 589 free(sector, M_MIRROR); 590 if (close) { 591 g_access(cp, 0, -1, -1); 592 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 593 cp->provider->name, 0, -1, -1, 0); 594 } 595 if (error != 0) { 596 G_MIRROR_DEBUG(0, "Cannot clear metadata on disk %s.", 597 g_mirror_get_diskname(disk)); 598 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 599 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 600 G_MIRROR_EVENT_DONTWAIT); 601 return (error); 602 } 603 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 604 g_mirror_get_diskname(disk)); 605 return (0); 606} 607 608void 609g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 610 struct g_mirror_metadata *md) 611{ 612 613 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 614 md->md_version = G_MIRROR_VERSION; 615 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 616 md->md_mid = sc->sc_id; 617 md->md_all = sc->sc_ndisks; 618 md->md_slice = sc->sc_slice; 619 md->md_balance = sc->sc_balance; 620 md->md_mediasize = sc->sc_mediasize; 621 md->md_sectorsize = sc->sc_sectorsize; 622 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 623 if (disk == NULL) { 624 md->md_did = arc4random(); 625 md->md_priority = 0; 626 md->md_syncid = 0; 627 md->md_dflags = 0; 628 md->md_sync_offset = 0; 629 } else { 630 md->md_did = disk->d_id; 631 md->md_priority = disk->d_priority; 632 md->md_syncid = disk->d_sync.ds_syncid; 633 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 634 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 635 md->md_sync_offset = disk->d_sync.ds_offset_done; 636 else 637 md->md_sync_offset = 0; 638 } 639} 640 641void 642g_mirror_update_metadata(struct g_mirror_disk *disk) 643{ 644 struct g_mirror_softc *sc; 645 struct g_mirror_metadata md; 646 struct g_consumer *cp; 647 off_t offset, length; 648 u_char *sector; 649 int close = 0, error = 0; 650 651 g_topology_assert(); 652 653 sc = disk->d_softc; 654 cp = disk->d_consumer; 655 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 656 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 657 length = cp->provider->sectorsize; 658 offset = cp->provider->mediasize - length; 659 sector = malloc((size_t)length, M_MIRROR, M_WAITOK); 660 /* 661 * Open consumer if it wasn't opened and remember to close it. 662 */ 663 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 664 error = g_access(cp, 0, 1, 1); 665 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 666 cp->provider->name, 0, 1, 1, error); 667 if (error == 0) 668 close = 1; 669#ifdef INVARIANTS 670 } else { 671 KASSERT(cp->acw > 0 && cp->ace > 0, 672 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 673 cp->acr, cp->acw, cp->ace)); 674#endif 675 } 676 if (error == 0) { 677 g_mirror_fill_metadata(sc, disk, &md); 678 mirror_metadata_encode(&md, sector); 679 g_topology_unlock(); 680 error = g_write_data(cp, offset, sector, length); 681 g_topology_lock(); 682 } 683 free(sector, M_MIRROR); 684 if (close) { 685 g_access(cp, 0, -1, -1); 686 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 687 cp->provider->name, 0, -1, -1, 0); 688 } 689 if (error != 0) { 690 G_MIRROR_DEBUG(0, 691 "Cannot update metadata on disk %s (error=%d).", 692 g_mirror_get_diskname(disk), error); 693 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 694 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 695 G_MIRROR_EVENT_DONTWAIT); 696 return; 697 } 698 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 699 g_mirror_get_diskname(disk)); 700} 701 702static void 703g_mirror_bump_syncid(struct g_mirror_softc *sc) 704{ 705 struct g_mirror_disk *disk; 706 707 g_topology_assert(); 708 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 709 ("%s called with no active disks (device=%s).", __func__, 710 sc->sc_name)); 711 712 sc->sc_syncid++; 713 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 714 sc->sc_syncid); 715 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 716 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 717 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 718 disk->d_sync.ds_syncid = sc->sc_syncid; 719 g_mirror_update_metadata(disk); 720 } 721 } 722} 723 724static __inline int 725bintime_cmp(struct bintime *bt1, struct bintime *bt2) 726{ 727 728 if (bt1->sec < bt2->sec) 729 return (-1); 730 else if (bt1->sec > bt2->sec) 731 return (1); 732 if (bt1->frac < bt2->frac) 733 return (-1); 734 else if (bt1->frac > bt2->frac) 735 return (1); 736 return (0); 737} 738 739static void 740g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 741{ 742 743 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 744 return; 745 binuptime(&disk->d_delay); 746 bintime_sub(&disk->d_delay, &bp->bio_t0); 747} 748 749static void 750g_mirror_done(struct bio *bp) 751{ 752 struct g_mirror_softc *sc; 753 754 sc = bp->bio_from->geom->softc; 755 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 756 mtx_lock(&sc->sc_queue_mtx); 757 bioq_disksort(&sc->sc_queue, bp); 758 wakeup(sc); 759 mtx_unlock(&sc->sc_queue_mtx); 760} 761 762static void 763g_mirror_regular_request(struct bio *bp) 764{ 765 struct g_mirror_softc *sc; 766 struct g_mirror_disk *disk; 767 struct bio *pbp; 768 769 g_topology_assert_not(); 770 771 pbp = bp->bio_parent; 772 sc = pbp->bio_to->geom->softc; 773 disk = bp->bio_from->private; 774 if (disk == NULL) { 775 g_topology_lock(); 776 g_mirror_kill_consumer(sc, bp->bio_from); 777 g_topology_unlock(); 778 } else { 779 g_mirror_update_delay(disk, bp); 780 } 781 782 pbp->bio_inbed++; 783 KASSERT(pbp->bio_inbed <= pbp->bio_children, 784 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 785 pbp->bio_children)); 786 if (bp->bio_error == 0 && pbp->bio_error == 0) { 787 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 788 g_destroy_bio(bp); 789 if (pbp->bio_children == pbp->bio_inbed) { 790 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 791 pbp->bio_completed = pbp->bio_length; 792 g_io_deliver(pbp, pbp->bio_error); 793 } 794 return; 795 } else if (bp->bio_error != 0) { 796 if (pbp->bio_error == 0) 797 pbp->bio_error = bp->bio_error; 798 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 799 bp->bio_error); 800 if (disk != NULL) { 801 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 802 g_mirror_event_send(disk, 803 G_MIRROR_DISK_STATE_DISCONNECTED, 804 G_MIRROR_EVENT_DONTWAIT); 805 } 806 switch (pbp->bio_cmd) { 807 case BIO_DELETE: 808 case BIO_WRITE: 809 pbp->bio_inbed--; 810 pbp->bio_children--; 811 break; 812 } 813 } 814 g_destroy_bio(bp); 815 816 switch (pbp->bio_cmd) { 817 case BIO_READ: 818 if (pbp->bio_children == pbp->bio_inbed) { 819 pbp->bio_error = 0; 820 mtx_lock(&sc->sc_queue_mtx); 821 bioq_disksort(&sc->sc_queue, pbp); 822 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 823 wakeup(sc); 824 mtx_unlock(&sc->sc_queue_mtx); 825 } 826 break; 827 case BIO_DELETE: 828 case BIO_WRITE: 829 if (pbp->bio_children == 0) { 830 /* 831 * All requests failed. 832 */ 833 } else if (pbp->bio_inbed < pbp->bio_children) { 834 /* Do nothing. */ 835 break; 836 } else if (pbp->bio_children == pbp->bio_inbed) { 837 /* Some requests succeeded. */ 838 pbp->bio_error = 0; 839 pbp->bio_completed = pbp->bio_length; 840 } 841 g_io_deliver(pbp, pbp->bio_error); 842 break; 843 default: 844 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 845 break; 846 } 847} 848 849static void 850g_mirror_sync_done(struct bio *bp) 851{ 852 struct g_mirror_softc *sc; 853 854 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 855 sc = bp->bio_from->geom->softc; 856 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 857 mtx_lock(&sc->sc_queue_mtx); 858 bioq_disksort(&sc->sc_queue, bp); 859 wakeup(sc); 860 mtx_unlock(&sc->sc_queue_mtx); 861} 862 863static void 864g_mirror_start(struct bio *bp) 865{ 866 struct g_mirror_softc *sc; 867 868 sc = bp->bio_to->geom->softc; 869 /* 870 * If sc == NULL or there are no valid disks, provider's error 871 * should be set and g_mirror_start() should not be called at all. 872 */ 873 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 874 ("Provider's error should be set (error=%d)(mirror=%s).", 875 bp->bio_to->error, bp->bio_to->name)); 876 G_MIRROR_LOGREQ(3, bp, "Request received."); 877 878 switch (bp->bio_cmd) { 879 case BIO_READ: 880 case BIO_WRITE: 881 case BIO_DELETE: 882 break; 883 case BIO_GETATTR: 884 default: 885 g_io_deliver(bp, EOPNOTSUPP); 886 return; 887 } 888 mtx_lock(&sc->sc_queue_mtx); 889 bioq_disksort(&sc->sc_queue, bp); 890 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 891 wakeup(sc); 892 mtx_unlock(&sc->sc_queue_mtx); 893} 894 895/* 896 * Send one synchronization request. 897 */ 898static void 899g_mirror_sync_one(struct g_mirror_disk *disk) 900{ 901 struct g_mirror_softc *sc; 902 struct bio *bp; 903 904 sc = disk->d_softc; 905 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 906 ("Disk %s is not marked for synchronization.", 907 g_mirror_get_diskname(disk))); 908 909 bp = g_new_bio(); 910 if (bp == NULL) 911 return; 912 bp->bio_parent = NULL; 913 bp->bio_cmd = BIO_READ; 914 bp->bio_offset = disk->d_sync.ds_offset; 915 bp->bio_length = MIN(sc->sc_sync.ds_block, 916 sc->sc_mediasize - bp->bio_offset); 917 bp->bio_cflags = 0; 918 bp->bio_done = g_mirror_sync_done; 919 bp->bio_data = uma_zalloc(sc->sc_sync.ds_zone, M_NOWAIT | M_ZERO); 920 if (bp->bio_data == NULL) { 921 g_destroy_bio(bp); 922 return; 923 } 924 disk->d_sync.ds_offset += bp->bio_length; 925 bp->bio_to = sc->sc_provider; 926 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 927 g_io_request(bp, disk->d_sync.ds_consumer); 928} 929 930static void 931g_mirror_sync_request(struct bio *bp) 932{ 933 struct g_mirror_softc *sc; 934 struct g_mirror_disk *disk; 935 936 sc = bp->bio_from->geom->softc; 937 disk = bp->bio_from->private; 938 if (disk == NULL) { 939 g_topology_lock(); 940 g_mirror_kill_consumer(sc, bp->bio_from); 941 g_topology_unlock(); 942 uma_zfree(sc->sc_sync.ds_zone, bp->bio_data); 943 g_destroy_bio(bp); 944 return; 945 } 946 947 /* 948 * Synchronization request. 949 */ 950 switch (bp->bio_cmd) { 951 case BIO_READ: 952 { 953 struct g_consumer *cp; 954 955 if (bp->bio_error != 0) { 956 G_MIRROR_LOGREQ(0, bp, 957 "Synchronization request failed (error=%d).", 958 bp->bio_error); 959 uma_zfree(sc->sc_sync.ds_zone, bp->bio_data); 960 g_destroy_bio(bp); 961 return; 962 } 963 bp->bio_cmd = BIO_WRITE; 964 bp->bio_cflags = 0; 965 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 966 cp = disk->d_consumer; 967 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 968 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 969 cp->acr, cp->acw, cp->ace)); 970 g_io_request(bp, cp); 971 return; 972 } 973 case BIO_WRITE: 974 uma_zfree(sc->sc_sync.ds_zone, bp->bio_data); 975 if (bp->bio_error != 0) { 976 G_MIRROR_LOGREQ(0, bp, 977 "Synchronization request failed (error=%d).", 978 bp->bio_error); 979 g_destroy_bio(bp); 980 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 981 g_mirror_event_send(disk, 982 G_MIRROR_DISK_STATE_DISCONNECTED, 983 G_MIRROR_EVENT_DONTWAIT); 984 return; 985 } 986 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 987 g_destroy_bio(bp); 988 disk->d_sync.ds_offset_done = bp->bio_offset + bp->bio_length; 989 if (bp->bio_offset + bp->bio_length == 990 sc->sc_provider->mediasize) { 991 /* 992 * Disk up-to-date, activate it. 993 */ 994 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 995 G_MIRROR_EVENT_DONTWAIT); 996 return; 997 } else if ((disk->d_sync.ds_offset_done % 998 (sc->sc_sync.ds_block * 100)) == 0) { 999 /* 1000 * Update offset_done on every 100 blocks. 1001 * XXX: This should be configurable. 1002 */ 1003 g_topology_lock(); 1004 g_mirror_update_metadata(disk); 1005 g_topology_unlock(); 1006 } 1007 return; 1008 default: 1009 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1010 bp->bio_cmd, sc->sc_name)); 1011 break; 1012 } 1013} 1014 1015static void 1016g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1017{ 1018 struct g_mirror_disk *disk; 1019 struct g_consumer *cp; 1020 struct bio *cbp; 1021 1022 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1023 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1024 break; 1025 } 1026 if (disk == NULL) { 1027 if (bp->bio_error == 0) 1028 bp->bio_error = ENXIO; 1029 g_io_deliver(bp, bp->bio_error); 1030 return; 1031 } 1032 cbp = g_clone_bio(bp); 1033 if (cbp == NULL) { 1034 if (bp->bio_error == 0) 1035 bp->bio_error = ENOMEM; 1036 g_io_deliver(bp, bp->bio_error); 1037 return; 1038 } 1039 /* 1040 * Fill in the component buf structure. 1041 */ 1042 cp = disk->d_consumer; 1043 cbp->bio_done = g_mirror_done; 1044 cbp->bio_to = cp->provider; 1045 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1046 KASSERT(cp->acr > 0 && cp->ace > 0, 1047 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1048 cp->acw, cp->ace)); 1049 g_io_request(cbp, cp); 1050} 1051 1052static void 1053g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1054{ 1055 struct g_mirror_disk *disk; 1056 struct g_consumer *cp; 1057 struct bio *cbp; 1058 1059 disk = g_mirror_get_disk(sc); 1060 if (disk == NULL) { 1061 if (bp->bio_error == 0) 1062 bp->bio_error = ENXIO; 1063 g_io_deliver(bp, bp->bio_error); 1064 return; 1065 } 1066 cbp = g_clone_bio(bp); 1067 if (cbp == NULL) { 1068 if (bp->bio_error == 0) 1069 bp->bio_error = ENOMEM; 1070 g_io_deliver(bp, bp->bio_error); 1071 return; 1072 } 1073 /* 1074 * Fill in the component buf structure. 1075 */ 1076 cp = disk->d_consumer; 1077 cbp->bio_done = g_mirror_done; 1078 cbp->bio_to = cp->provider; 1079 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1080 KASSERT(cp->acr > 0 && cp->ace > 0, 1081 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1082 cp->acw, cp->ace)); 1083 g_io_request(cbp, cp); 1084} 1085 1086static void 1087g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1088{ 1089 struct g_mirror_disk *disk, *dp; 1090 struct g_consumer *cp; 1091 struct bio *cbp; 1092 struct bintime curtime; 1093 1094 binuptime(&curtime); 1095 /* 1096 * Find a disk which the smallest load. 1097 */ 1098 disk = NULL; 1099 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1100 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1101 continue; 1102 /* If disk wasn't used for more than 2 sec, use it. */ 1103 if (curtime.sec - dp->d_last_used.sec >= 2) { 1104 disk = dp; 1105 break; 1106 } 1107 if (disk == NULL || 1108 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1109 disk = dp; 1110 } 1111 } 1112 cbp = g_clone_bio(bp); 1113 if (cbp == NULL) { 1114 if (bp->bio_error == 0) 1115 bp->bio_error = ENOMEM; 1116 g_io_deliver(bp, bp->bio_error); 1117 return; 1118 } 1119 /* 1120 * Fill in the component buf structure. 1121 */ 1122 cp = disk->d_consumer; 1123 cbp->bio_done = g_mirror_done; 1124 cbp->bio_to = cp->provider; 1125 binuptime(&disk->d_last_used); 1126 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1127 KASSERT(cp->acr > 0 && cp->ace > 0, 1128 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1129 cp->acw, cp->ace)); 1130 g_io_request(cbp, cp); 1131} 1132 1133static void 1134g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1135{ 1136 struct bio_queue_head queue; 1137 struct g_mirror_disk *disk; 1138 struct g_consumer *cp; 1139 struct bio *cbp; 1140 off_t left, mod, offset, slice; 1141 u_char *data; 1142 u_int ndisks; 1143 1144 if (bp->bio_length <= sc->sc_slice) { 1145 g_mirror_request_round_robin(sc, bp); 1146 return; 1147 } 1148 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1149 slice = bp->bio_length / ndisks; 1150 mod = slice % sc->sc_provider->sectorsize; 1151 if (mod != 0) 1152 slice += sc->sc_provider->sectorsize - mod; 1153 /* 1154 * Allocate all bios before sending any request, so we can 1155 * return ENOMEM in nice and clean way. 1156 */ 1157 left = bp->bio_length; 1158 offset = bp->bio_offset; 1159 data = bp->bio_data; 1160 bioq_init(&queue); 1161 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1162 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1163 continue; 1164 cbp = g_clone_bio(bp); 1165 if (cbp == NULL) { 1166 for (cbp = bioq_first(&queue); cbp != NULL; 1167 cbp = bioq_first(&queue)) { 1168 bioq_remove(&queue, cbp); 1169 g_destroy_bio(cbp); 1170 } 1171 if (bp->bio_error == 0) 1172 bp->bio_error = ENOMEM; 1173 g_io_deliver(bp, bp->bio_error); 1174 return; 1175 } 1176 bioq_insert_tail(&queue, cbp); 1177 cbp->bio_done = g_mirror_done; 1178 cbp->bio_caller1 = disk; 1179 cbp->bio_to = disk->d_consumer->provider; 1180 cbp->bio_offset = offset; 1181 cbp->bio_data = data; 1182 cbp->bio_length = MIN(left, slice); 1183 left -= cbp->bio_length; 1184 if (left == 0) 1185 break; 1186 offset += cbp->bio_length; 1187 data += cbp->bio_length; 1188 } 1189 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1190 bioq_remove(&queue, cbp); 1191 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1192 disk = cbp->bio_caller1; 1193 cbp->bio_caller1 = NULL; 1194 cp = disk->d_consumer; 1195 KASSERT(cp->acr > 0 && cp->ace > 0, 1196 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1197 cp->acr, cp->acw, cp->ace)); 1198 g_io_request(cbp, disk->d_consumer); 1199 } 1200} 1201 1202static void 1203g_mirror_register_request(struct bio *bp) 1204{ 1205 struct g_mirror_softc *sc; 1206 1207 sc = bp->bio_to->geom->softc; 1208 switch (bp->bio_cmd) { 1209 case BIO_READ: 1210 switch (sc->sc_balance) { 1211 case G_MIRROR_BALANCE_LOAD: 1212 g_mirror_request_load(sc, bp); 1213 break; 1214 case G_MIRROR_BALANCE_PREFER: 1215 g_mirror_request_prefer(sc, bp); 1216 break; 1217 case G_MIRROR_BALANCE_ROUND_ROBIN: 1218 g_mirror_request_round_robin(sc, bp); 1219 break; 1220 case G_MIRROR_BALANCE_SPLIT: 1221 g_mirror_request_split(sc, bp); 1222 break; 1223 } 1224 return; 1225 case BIO_WRITE: 1226 case BIO_DELETE: 1227 { 1228 struct g_mirror_disk *disk; 1229 struct bio_queue_head queue; 1230 struct g_consumer *cp; 1231 struct bio *cbp; 1232 1233 /* 1234 * Allocate all bios before sending any request, so we can 1235 * return ENOMEM in nice and clean way. 1236 */ 1237 bioq_init(&queue); 1238 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1239 switch (disk->d_state) { 1240 case G_MIRROR_DISK_STATE_ACTIVE: 1241 break; 1242 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1243 if (bp->bio_offset >= disk->d_sync.ds_offset) 1244 continue; 1245 break; 1246 default: 1247 continue; 1248 } 1249 cbp = g_clone_bio(bp); 1250 if (cbp == NULL) { 1251 for (cbp = bioq_first(&queue); cbp != NULL; 1252 cbp = bioq_first(&queue)) { 1253 bioq_remove(&queue, cbp); 1254 g_destroy_bio(cbp); 1255 } 1256 if (bp->bio_error == 0) 1257 bp->bio_error = ENOMEM; 1258 g_io_deliver(bp, bp->bio_error); 1259 return; 1260 } 1261 bioq_insert_tail(&queue, cbp); 1262 } 1263 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1264 switch (disk->d_state) { 1265 case G_MIRROR_DISK_STATE_ACTIVE: 1266 break; 1267 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1268 if (bp->bio_offset >= disk->d_sync.ds_offset) 1269 continue; 1270 break; 1271 default: 1272 continue; 1273 } 1274 cbp = bioq_first(&queue); 1275 KASSERT(cbp != NULL, ("NULL cbp! (device %s).", 1276 sc->sc_name)); 1277 bioq_remove(&queue, cbp); 1278 cp = disk->d_consumer; 1279 cbp->bio_done = g_mirror_done; 1280 cbp->bio_to = cp->provider; 1281 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1282 KASSERT(cp->acw > 0 && cp->ace > 0, 1283 ("Consumer %s not opened (r%dw%de%d).", 1284 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1285 g_io_request(cbp, cp); 1286 } 1287 /* 1288 * Bump syncid on first write. 1289 */ 1290 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) { 1291 sc->sc_bump_syncid = 0; 1292 g_topology_lock(); 1293 g_mirror_bump_syncid(sc); 1294 g_topology_unlock(); 1295 } 1296 return; 1297 } 1298 default: 1299 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1300 bp->bio_cmd, sc->sc_name)); 1301 break; 1302 } 1303} 1304 1305/* 1306 * Worker thread. 1307 */ 1308static void 1309g_mirror_worker(void *arg) 1310{ 1311 struct g_mirror_softc *sc; 1312 struct g_mirror_disk *disk; 1313 struct g_mirror_event *ep; 1314 struct bio *bp; 1315 u_int nreqs; 1316 1317 sc = arg; 1318 curthread->td_base_pri = PRIBIO; 1319 1320 nreqs = 0; 1321 for (;;) { 1322 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1323 /* 1324 * First take a look at events. 1325 * This is important to handle events before any I/O requests. 1326 */ 1327 ep = g_mirror_event_get(sc); 1328 if (ep != NULL) { 1329 g_topology_lock(); 1330 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1331 /* Update only device status. */ 1332 G_MIRROR_DEBUG(3, 1333 "Running event for device %s.", 1334 sc->sc_name); 1335 ep->e_error = 0; 1336 g_mirror_update_device(sc, 1); 1337 } else { 1338 /* Update disk status. */ 1339 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1340 g_mirror_get_diskname(ep->e_disk)); 1341 ep->e_error = g_mirror_update_disk(ep->e_disk, 1342 ep->e_state); 1343 if (ep->e_error == 0) 1344 g_mirror_update_device(sc, 0); 1345 } 1346 g_topology_unlock(); 1347 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1348 KASSERT(ep->e_error == 0, 1349 ("Error cannot be handled.")); 1350 g_mirror_event_free(ep); 1351 } else { 1352 ep->e_flags |= G_MIRROR_EVENT_DONE; 1353 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1354 ep); 1355 mtx_lock(&sc->sc_events_mtx); 1356 wakeup(ep); 1357 mtx_unlock(&sc->sc_events_mtx); 1358 } 1359 if ((sc->sc_flags & 1360 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1361end: 1362 if ((sc->sc_flags & 1363 G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1364 G_MIRROR_DEBUG(4, "%s: Waking up %p.", 1365 __func__, &sc->sc_worker); 1366 wakeup(&sc->sc_worker); 1367 sc->sc_worker = NULL; 1368 } else { 1369 g_topology_lock(); 1370 g_mirror_destroy_device(sc); 1371 g_topology_unlock(); 1372 free(sc, M_MIRROR); 1373 } 1374 kthread_exit(0); 1375 } 1376 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1377 continue; 1378 } 1379 /* 1380 * Now I/O requests. 1381 */ 1382 /* Get first request from the queue. */ 1383 mtx_lock(&sc->sc_queue_mtx); 1384 bp = bioq_first(&sc->sc_queue); 1385 if (bp == NULL) { 1386 if ((sc->sc_flags & 1387 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1388 mtx_unlock(&sc->sc_queue_mtx); 1389 goto end; 1390 } 1391 } 1392 if (sc->sc_sync.ds_ndisks > 0 && 1393 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1394 mtx_unlock(&sc->sc_queue_mtx); 1395 /* 1396 * It is time for synchronization... 1397 */ 1398 nreqs = 0; 1399 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1400 if (disk->d_state != 1401 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1402 continue; 1403 } 1404 if (disk->d_sync.ds_offset >= 1405 sc->sc_provider->mediasize) { 1406 continue; 1407 } 1408 if (disk->d_sync.ds_offset > 1409 disk->d_sync.ds_offset_done) { 1410 continue; 1411 } 1412 g_mirror_sync_one(disk); 1413 } 1414 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1415 goto sleep; 1416 } 1417 if (bp == NULL) { 1418 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 0); 1419 G_MIRROR_DEBUG(5, "%s: I'm here 3.", __func__); 1420 continue; 1421 } 1422 nreqs++; 1423 bioq_remove(&sc->sc_queue, bp); 1424 mtx_unlock(&sc->sc_queue_mtx); 1425 1426 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1427 g_mirror_regular_request(bp); 1428 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1429 u_int timeout, sps; 1430 1431 g_mirror_sync_request(bp); 1432sleep: 1433 sps = atomic_load_acq_int(&g_mirror_syncs_per_sec); 1434 if (sps == 0) { 1435 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1436 continue; 1437 } 1438 mtx_lock(&sc->sc_queue_mtx); 1439 if (bioq_first(&sc->sc_queue) != NULL) { 1440 mtx_unlock(&sc->sc_queue_mtx); 1441 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1442 continue; 1443 } 1444 timeout = hz / sps; 1445 if (timeout == 0) 1446 timeout = 1; 1447 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w2", 1448 timeout); 1449 } else { 1450 g_mirror_register_request(bp); 1451 } 1452 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1453 } 1454} 1455 1456/* 1457 * Open disk's consumer if needed. 1458 */ 1459static void 1460g_mirror_update_access(struct g_mirror_disk *disk) 1461{ 1462 struct g_provider *pp; 1463 struct g_consumer *cp; 1464 int acr, acw, ace, cpw, error; 1465 1466 g_topology_assert(); 1467 1468 cp = disk->d_consumer; 1469 pp = disk->d_softc->sc_provider; 1470 if (pp == NULL) { 1471 acr = -cp->acr; 1472 acw = -cp->acw; 1473 ace = -cp->ace; 1474 } else { 1475 acr = pp->acr - cp->acr; 1476 acw = pp->acw - cp->acw; 1477 ace = pp->ace - cp->ace; 1478 /* Grab an extra "exclusive" bit. */ 1479 if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) 1480 ace++; 1481 } 1482 if (acr == 0 && acw == 0 && ace == 0) 1483 return; 1484 cpw = cp->acw; 1485 error = g_access(cp, acr, acw, ace); 1486 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, acr, 1487 acw, ace, error); 1488 if (error != 0) { 1489 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1490 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1491 G_MIRROR_EVENT_DONTWAIT); 1492 return; 1493 } 1494 if (cpw == 0 && cp->acw > 0) { 1495 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1496 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1497 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1498 } else if (cpw > 0 && cp->acw == 0) { 1499 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1500 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1501 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1502 } 1503} 1504 1505static void 1506g_mirror_sync_start(struct g_mirror_disk *disk) 1507{ 1508 struct g_mirror_softc *sc; 1509 struct g_consumer *cp; 1510 int error; 1511 1512 g_topology_assert(); 1513 1514 sc = disk->d_softc; 1515 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1516 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1517 sc->sc_state)); 1518 cp = disk->d_consumer; 1519 KASSERT(cp->acr == 0 && cp->acw == 0 && cp->ace == 0, 1520 ("Consumer %s already opened.", cp->provider->name)); 1521 1522 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1523 g_mirror_get_diskname(disk)); 1524 error = g_access(cp, 0, 1, 1); 1525 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, 1, 1526 1, error); 1527 if (error != 0) { 1528 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1529 G_MIRROR_EVENT_DONTWAIT); 1530 return; 1531 } 1532 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1533 KASSERT(disk->d_sync.ds_consumer == NULL, 1534 ("Sync consumer already exists (device=%s, disk=%s).", 1535 sc->sc_name, g_mirror_get_diskname(disk))); 1536 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1537 disk->d_sync.ds_consumer->private = disk; 1538 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1539 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1540 disk->d_softc->sc_name, error)); 1541 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1542 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1543 disk->d_softc->sc_name, error)); 1544 sc->sc_sync.ds_ndisks++; 1545} 1546 1547/* 1548 * Stop synchronization process. 1549 * type: 0 - synchronization finished 1550 * 1 - synchronization stopped 1551 */ 1552static void 1553g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1554{ 1555 struct g_consumer *cp; 1556 1557 g_topology_assert(); 1558 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1559 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1560 g_mirror_disk_state2str(disk->d_state))); 1561 if (disk->d_sync.ds_consumer == NULL) 1562 return; 1563 1564 if (type == 0) { 1565 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1566 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1567 } else /* if (type == 1) */ { 1568 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1569 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1570 } 1571 cp = disk->d_sync.ds_consumer; 1572 g_access(cp, -1, 0, 0); 1573 g_mirror_kill_consumer(disk->d_softc, cp); 1574 disk->d_sync.ds_consumer = NULL; 1575 disk->d_softc->sc_sync.ds_ndisks--; 1576 cp = disk->d_consumer; 1577 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 1578 ("Consumer %s not opened.", cp->provider->name)); 1579 g_access(cp, 0, -1, -1); 1580 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, -1, 1581 -1, 0); 1582 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1583} 1584 1585static void 1586g_mirror_launch_provider(struct g_mirror_softc *sc) 1587{ 1588 struct g_mirror_disk *disk; 1589 struct g_provider *pp; 1590 1591 g_topology_assert(); 1592 1593 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1594 pp->mediasize = sc->sc_mediasize; 1595 pp->sectorsize = sc->sc_sectorsize; 1596 sc->sc_provider = pp; 1597 g_error_provider(pp, 0); 1598 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1599 pp->name); 1600 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1601 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1602 g_mirror_sync_start(disk); 1603 } 1604} 1605 1606static void 1607g_mirror_destroy_provider(struct g_mirror_softc *sc) 1608{ 1609 struct g_mirror_disk *disk; 1610 struct bio *bp; 1611 1612 g_topology_assert(); 1613 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1614 sc->sc_name)); 1615 1616 g_error_provider(sc->sc_provider, ENXIO); 1617 mtx_lock(&sc->sc_queue_mtx); 1618 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1619 bioq_remove(&sc->sc_queue, bp); 1620 g_io_deliver(bp, ENXIO); 1621 } 1622 mtx_unlock(&sc->sc_queue_mtx); 1623 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1624 sc->sc_provider->name); 1625 sc->sc_provider->flags |= G_PF_WITHER; 1626 g_orphan_provider(sc->sc_provider, ENXIO); 1627 sc->sc_provider = NULL; 1628 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1629 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1630 g_mirror_sync_stop(disk, 1); 1631 } 1632} 1633 1634static void 1635g_mirror_go(void *arg) 1636{ 1637 struct g_mirror_softc *sc; 1638 1639 sc = arg; 1640 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1641 g_mirror_event_send(sc, 0, 1642 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1643} 1644 1645static u_int 1646g_mirror_determine_state(struct g_mirror_disk *disk) 1647{ 1648 struct g_mirror_softc *sc; 1649 u_int state; 1650 1651 sc = disk->d_softc; 1652 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1653 if ((disk->d_flags & 1654 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1655 /* Disk does not need synchronization. */ 1656 state = G_MIRROR_DISK_STATE_ACTIVE; 1657 } else { 1658 if ((sc->sc_flags & 1659 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1660 (disk->d_flags & 1661 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1662 /* 1663 * We can start synchronization from 1664 * the stored offset. 1665 */ 1666 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1667 } else { 1668 state = G_MIRROR_DISK_STATE_STALE; 1669 } 1670 } 1671 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1672 /* 1673 * Reset all synchronization data for this disk, 1674 * because if it even was synchronized, it was 1675 * synchronized to disks with different syncid. 1676 */ 1677 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1678 disk->d_sync.ds_offset = 0; 1679 disk->d_sync.ds_offset_done = 0; 1680 disk->d_sync.ds_syncid = sc->sc_syncid; 1681 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1682 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1683 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1684 } else { 1685 state = G_MIRROR_DISK_STATE_STALE; 1686 } 1687 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1688 /* 1689 * Not good, NOT GOOD! 1690 * It means that mirror was started on stale disks 1691 * and more fresh disk just arrive. 1692 * If there were writes, mirror is fucked up, sorry. 1693 * I think the best choice here is don't touch 1694 * this disk and inform the user laudly. 1695 */ 1696 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1697 "disk (%s) arrives!! It will not be connected to the " 1698 "running device.", sc->sc_name, 1699 g_mirror_get_diskname(disk)); 1700 g_mirror_destroy_disk(disk); 1701 state = G_MIRROR_DISK_STATE_NONE; 1702 /* Return immediately, because disk was destroyed. */ 1703 return (state); 1704 } 1705 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1706 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1707 return (state); 1708} 1709 1710/* 1711 * Update device state. 1712 */ 1713static void 1714g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1715{ 1716 struct g_mirror_disk *disk; 1717 u_int state; 1718 1719 g_topology_assert(); 1720 1721 switch (sc->sc_state) { 1722 case G_MIRROR_DEVICE_STATE_STARTING: 1723 { 1724 struct g_mirror_disk *pdisk; 1725 u_int dirty, ndisks, syncid; 1726 1727 KASSERT(sc->sc_provider == NULL, 1728 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1729 /* 1730 * Are we ready? We are, if all disks are connected or 1731 * if we have any disks and 'force' is true. 1732 */ 1733 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1734 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1735 ; 1736 } else if (g_mirror_ndisks(sc, -1) == 0) { 1737 /* 1738 * Disks went down in starting phase, so destroy 1739 * device. 1740 */ 1741 callout_drain(&sc->sc_callout); 1742 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1743 return; 1744 } else { 1745 return; 1746 } 1747 1748 /* 1749 * Activate all disks with the biggest syncid. 1750 */ 1751 if (force) { 1752 /* 1753 * If 'force' is true, we have been called due to 1754 * timeout, so don't bother canceling timeout. 1755 */ 1756 ndisks = 0; 1757 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1758 if ((disk->d_flags & 1759 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1760 ndisks++; 1761 } 1762 } 1763 if (ndisks == 0) { 1764 /* No valid disks found, destroy device. */ 1765 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1766 return; 1767 } 1768 } else { 1769 /* Cancel timeout. */ 1770 callout_drain(&sc->sc_callout); 1771 } 1772 1773 /* 1774 * Find disk with the biggest syncid. 1775 */ 1776 syncid = 0; 1777 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1778 if (disk->d_sync.ds_syncid > syncid) 1779 syncid = disk->d_sync.ds_syncid; 1780 } 1781 1782 /* 1783 * Here we need to look for dirty disks and if all disks 1784 * with the biggest syncid are dirty, we have to choose 1785 * one with the biggest priority and rebuild the rest. 1786 */ 1787 /* 1788 * Find the number of dirty disks with the biggest syncid. 1789 * Find the number of disks with the biggest syncid. 1790 * While here, find a disk with the biggest priority. 1791 */ 1792 dirty = ndisks = 0; 1793 pdisk = NULL; 1794 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1795 if (disk->d_sync.ds_syncid != syncid) 1796 continue; 1797 if ((disk->d_flags & 1798 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1799 continue; 1800 } 1801 ndisks++; 1802 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1803 dirty++; 1804 if (pdisk == NULL || 1805 pdisk->d_priority < disk->d_priority) { 1806 pdisk = disk; 1807 } 1808 } 1809 } 1810 if (dirty == 0) { 1811 /* No dirty disks at all, great. */ 1812 } else if (dirty == ndisks) { 1813 /* 1814 * Force synchronization for all dirty disks except one 1815 * with the biggest priority. 1816 */ 1817 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1818 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1819 "master disk for synchronization.", 1820 g_mirror_get_diskname(pdisk), sc->sc_name); 1821 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1822 if (disk->d_sync.ds_syncid != syncid) 1823 continue; 1824 if ((disk->d_flags & 1825 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1826 continue; 1827 } 1828 KASSERT((disk->d_flags & 1829 G_MIRROR_DISK_FLAG_DIRTY) != 0, 1830 ("Disk %s isn't marked as dirty.", 1831 g_mirror_get_diskname(disk))); 1832 /* Skip the disk with the biggest priority. */ 1833 if (disk == pdisk) 1834 continue; 1835 disk->d_sync.ds_syncid = 0; 1836 } 1837 } else if (dirty < ndisks) { 1838 /* 1839 * Force synchronization for all dirty disks. 1840 * We have some non-dirty disks. 1841 */ 1842 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1843 if (disk->d_sync.ds_syncid != syncid) 1844 continue; 1845 if ((disk->d_flags & 1846 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1847 continue; 1848 } 1849 if ((disk->d_flags & 1850 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1851 continue; 1852 } 1853 disk->d_sync.ds_syncid = 0; 1854 } 1855 } 1856 1857 /* Reset hint. */ 1858 sc->sc_hint = NULL; 1859 sc->sc_syncid = syncid; 1860 if (force) { 1861 /* Remember to bump syncid on first write. */ 1862 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1863 } 1864 state = G_MIRROR_DEVICE_STATE_RUNNING; 1865 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 1866 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 1867 g_mirror_device_state2str(state)); 1868 sc->sc_state = state; 1869 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1870 state = g_mirror_determine_state(disk); 1871 g_mirror_event_send(disk, state, 1872 G_MIRROR_EVENT_DONTWAIT); 1873 if (state == G_MIRROR_DISK_STATE_STALE) { 1874 sc->sc_bump_syncid = 1875 G_MIRROR_BUMP_ON_FIRST_WRITE; 1876 } 1877 } 1878 break; 1879 } 1880 case G_MIRROR_DEVICE_STATE_RUNNING: 1881 /* 1882 * Bump syncid here, if we need to do it immediately. 1883 */ 1884 if (sc->sc_bump_syncid == G_MIRROR_BUMP_IMMEDIATELY) { 1885 sc->sc_bump_syncid = 0; 1886 g_mirror_bump_syncid(sc); 1887 } 1888 if (g_mirror_ndisks(sc, -1) == 0) { 1889 /* 1890 * No disks at all, we need to destroy device. 1891 */ 1892 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1893 } else if (g_mirror_ndisks(sc, 1894 G_MIRROR_DISK_STATE_ACTIVE) == 0 && 1895 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 1896 /* 1897 * No active disks, destroy provider. 1898 */ 1899 if (sc->sc_provider != NULL) 1900 g_mirror_destroy_provider(sc); 1901 } else if (g_mirror_ndisks(sc, 1902 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 1903 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 1904 /* 1905 * We have active disks, launch provider if it doesn't 1906 * exist. 1907 */ 1908 if (sc->sc_provider == NULL) 1909 g_mirror_launch_provider(sc); 1910 } 1911 break; 1912 default: 1913 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 1914 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 1915 break; 1916 } 1917} 1918 1919/* 1920 * Update disk state and device state if needed. 1921 */ 1922#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 1923 "Disk %s state changed from %s to %s (device %s).", \ 1924 g_mirror_get_diskname(disk), \ 1925 g_mirror_disk_state2str(disk->d_state), \ 1926 g_mirror_disk_state2str(state), sc->sc_name) 1927static int 1928g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 1929{ 1930 struct g_mirror_softc *sc; 1931 1932 g_topology_assert(); 1933 1934 sc = disk->d_softc; 1935again: 1936 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 1937 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 1938 g_mirror_disk_state2str(state)); 1939 switch (state) { 1940 case G_MIRROR_DISK_STATE_NEW: 1941 /* 1942 * Possible scenarios: 1943 * 1. New disk arrive. 1944 */ 1945 /* Previous state should be NONE. */ 1946 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 1947 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1948 g_mirror_disk_state2str(disk->d_state))); 1949 DISK_STATE_CHANGED(); 1950 1951 disk->d_state = state; 1952 if (LIST_EMPTY(&sc->sc_disks)) 1953 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 1954 else { 1955 struct g_mirror_disk *dp; 1956 1957 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1958 if (disk->d_priority >= dp->d_priority) { 1959 LIST_INSERT_BEFORE(dp, disk, d_next); 1960 dp = NULL; 1961 break; 1962 } 1963 if (LIST_NEXT(dp, d_next) == NULL) 1964 break; 1965 } 1966 if (dp != NULL) 1967 LIST_INSERT_AFTER(dp, disk, d_next); 1968 } 1969 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 1970 sc->sc_name, g_mirror_get_diskname(disk)); 1971 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 1972 break; 1973 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1974 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 1975 g_mirror_device_state2str(sc->sc_state), 1976 g_mirror_get_diskname(disk), 1977 g_mirror_disk_state2str(disk->d_state))); 1978 state = g_mirror_determine_state(disk); 1979 if (state != G_MIRROR_DISK_STATE_NONE) 1980 goto again; 1981 break; 1982 case G_MIRROR_DISK_STATE_ACTIVE: 1983 /* 1984 * Possible scenarios: 1985 * 1. New disk does not need synchronization. 1986 * 2. Synchronization process finished successfully. 1987 */ 1988 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1989 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 1990 g_mirror_device_state2str(sc->sc_state), 1991 g_mirror_get_diskname(disk), 1992 g_mirror_disk_state2str(disk->d_state))); 1993 /* Previous state should be NEW or SYNCHRONIZING. */ 1994 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 1995 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1996 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1997 g_mirror_disk_state2str(disk->d_state))); 1998 DISK_STATE_CHANGED(); 1999 2000 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2001 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2002 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2003 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2004 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2005 g_mirror_sync_stop(disk, 0); 2006 } 2007 disk->d_state = state; 2008 disk->d_sync.ds_offset = 0; 2009 disk->d_sync.ds_offset_done = 0; 2010 g_mirror_update_access(disk); 2011 g_mirror_update_metadata(disk); 2012 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2013 sc->sc_name, g_mirror_get_diskname(disk)); 2014 break; 2015 case G_MIRROR_DISK_STATE_STALE: 2016 /* 2017 * Possible scenarios: 2018 * 1. Stale disk was connected. 2019 */ 2020 /* Previous state should be NEW. */ 2021 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2022 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2023 g_mirror_disk_state2str(disk->d_state))); 2024 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2025 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2026 g_mirror_device_state2str(sc->sc_state), 2027 g_mirror_get_diskname(disk), 2028 g_mirror_disk_state2str(disk->d_state))); 2029 /* 2030 * STALE state is only possible if device is marked 2031 * NOAUTOSYNC. 2032 */ 2033 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2034 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2035 g_mirror_device_state2str(sc->sc_state), 2036 g_mirror_get_diskname(disk), 2037 g_mirror_disk_state2str(disk->d_state))); 2038 DISK_STATE_CHANGED(); 2039 2040 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2041 disk->d_state = state; 2042 g_mirror_update_metadata(disk); 2043 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2044 sc->sc_name, g_mirror_get_diskname(disk)); 2045 break; 2046 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2047 /* 2048 * Possible scenarios: 2049 * 1. Disk which needs synchronization was connected. 2050 */ 2051 /* Previous state should be NEW. */ 2052 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2053 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2054 g_mirror_disk_state2str(disk->d_state))); 2055 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2056 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2057 g_mirror_device_state2str(sc->sc_state), 2058 g_mirror_get_diskname(disk), 2059 g_mirror_disk_state2str(disk->d_state))); 2060 DISK_STATE_CHANGED(); 2061 2062 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2063 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2064 disk->d_state = state; 2065 if (sc->sc_provider != NULL) { 2066 g_mirror_sync_start(disk); 2067 g_mirror_update_metadata(disk); 2068 } 2069 break; 2070 case G_MIRROR_DISK_STATE_DISCONNECTED: 2071 /* 2072 * Possible scenarios: 2073 * 1. Device wasn't running yet, but disk disappear. 2074 * 2. Disk was active and disapppear. 2075 * 3. Disk disappear during synchronization process. 2076 */ 2077 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2078 /* 2079 * Previous state should be ACTIVE, STALE or 2080 * SYNCHRONIZING. 2081 */ 2082 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2083 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2084 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2085 ("Wrong disk state (%s, %s).", 2086 g_mirror_get_diskname(disk), 2087 g_mirror_disk_state2str(disk->d_state))); 2088 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2089 /* Previous state should be NEW. */ 2090 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2091 ("Wrong disk state (%s, %s).", 2092 g_mirror_get_diskname(disk), 2093 g_mirror_disk_state2str(disk->d_state))); 2094 /* 2095 * Reset bumping syncid if disk disappeared in STARTING 2096 * state. 2097 */ 2098 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) 2099 sc->sc_bump_syncid = 0; 2100#ifdef INVARIANTS 2101 } else { 2102 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2103 sc->sc_name, 2104 g_mirror_device_state2str(sc->sc_state), 2105 g_mirror_get_diskname(disk), 2106 g_mirror_disk_state2str(disk->d_state))); 2107#endif 2108 } 2109 DISK_STATE_CHANGED(); 2110 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2111 sc->sc_name, g_mirror_get_diskname(disk)); 2112 2113 g_mirror_destroy_disk(disk); 2114 break; 2115 case G_MIRROR_DISK_STATE_DESTROY: 2116 { 2117 int error; 2118 2119 error = g_mirror_clear_metadata(disk); 2120 if (error != 0) 2121 return (error); 2122 DISK_STATE_CHANGED(); 2123 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2124 sc->sc_name, g_mirror_get_diskname(disk)); 2125 2126 g_mirror_destroy_disk(disk); 2127 sc->sc_ndisks--; 2128 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2129 g_mirror_update_metadata(disk); 2130 } 2131 break; 2132 } 2133 default: 2134 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2135 break; 2136 } 2137 return (0); 2138} 2139#undef DISK_STATE_CHANGED 2140 2141static int 2142g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2143{ 2144 struct g_provider *pp; 2145 u_char *buf; 2146 int error; 2147 2148 g_topology_assert(); 2149 2150 error = g_access(cp, 1, 0, 0); 2151 if (error != 0) 2152 return (error); 2153 pp = cp->provider; 2154 g_topology_unlock(); 2155 /* Metadata are stored on last sector. */ 2156 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2157 &error); 2158 g_topology_lock(); 2159 if (buf == NULL) { 2160 g_access(cp, -1, 0, 0); 2161 return (error); 2162 } 2163 if (error != 0) { 2164 g_access(cp, -1, 0, 0); 2165 g_free(buf); 2166 return (error); 2167 } 2168 error = g_access(cp, -1, 0, 0); 2169 KASSERT(error == 0, ("Cannot decrease access count for %s.", pp->name)); 2170 2171 /* Decode metadata. */ 2172 error = mirror_metadata_decode(buf, md); 2173 g_free(buf); 2174 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2175 return (EINVAL); 2176 if (error != 0) { 2177 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2178 cp->provider->name); 2179 return (error); 2180 } 2181 2182 return (0); 2183} 2184 2185static int 2186g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2187 struct g_mirror_metadata *md) 2188{ 2189 2190 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2191 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2192 pp->name, md->md_did); 2193 return (EEXIST); 2194 } 2195 if (md->md_all != sc->sc_ndisks) { 2196 G_MIRROR_DEBUG(1, 2197 "Invalid '%s' field on disk %s (device %s), skipping.", 2198 "md_all", pp->name, sc->sc_name); 2199 return (EINVAL); 2200 } 2201 if (md->md_slice != sc->sc_slice) { 2202 G_MIRROR_DEBUG(1, 2203 "Invalid '%s' field on disk %s (device %s), skipping.", 2204 "md_slice", pp->name, sc->sc_name); 2205 return (EINVAL); 2206 } 2207 if (md->md_balance != sc->sc_balance) { 2208 G_MIRROR_DEBUG(1, 2209 "Invalid '%s' field on disk %s (device %s), skipping.", 2210 "md_balance", pp->name, sc->sc_name); 2211 return (EINVAL); 2212 } 2213 if (md->md_mediasize != sc->sc_mediasize) { 2214 G_MIRROR_DEBUG(1, 2215 "Invalid '%s' field on disk %s (device %s), skipping.", 2216 "md_mediasize", pp->name, sc->sc_name); 2217 return (EINVAL); 2218 } 2219 if (sc->sc_mediasize > pp->mediasize) { 2220 G_MIRROR_DEBUG(1, 2221 "Invalid size of disk %s (device %s), skipping.", pp->name, 2222 sc->sc_name); 2223 return (EINVAL); 2224 } 2225 if (md->md_sectorsize != sc->sc_sectorsize) { 2226 G_MIRROR_DEBUG(1, 2227 "Invalid '%s' field on disk %s (device %s), skipping.", 2228 "md_sectorsize", pp->name, sc->sc_name); 2229 return (EINVAL); 2230 } 2231 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2232 G_MIRROR_DEBUG(1, 2233 "Invalid sector size of disk %s (device %s), skipping.", 2234 pp->name, sc->sc_name); 2235 return (EINVAL); 2236 } 2237 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2238 G_MIRROR_DEBUG(1, 2239 "Invalid device flags on disk %s (device %s), skipping.", 2240 pp->name, sc->sc_name); 2241 return (EINVAL); 2242 } 2243 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2244 G_MIRROR_DEBUG(1, 2245 "Invalid disk flags on disk %s (device %s), skipping.", 2246 pp->name, sc->sc_name); 2247 return (EINVAL); 2248 } 2249 return (0); 2250} 2251 2252static int 2253g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2254 struct g_mirror_metadata *md) 2255{ 2256 struct g_mirror_disk *disk; 2257 int error; 2258 2259 g_topology_assert(); 2260 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2261 2262 error = g_mirror_check_metadata(sc, pp, md); 2263 if (error != 0) 2264 return (error); 2265 disk = g_mirror_init_disk(sc, pp, md, &error); 2266 if (disk == NULL) 2267 return (error); 2268 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2269 G_MIRROR_EVENT_WAIT); 2270 return (error); 2271} 2272 2273static int 2274g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2275{ 2276 struct g_mirror_softc *sc; 2277 struct g_mirror_disk *disk; 2278 int dcr, dcw, dce, err, error; 2279 2280 g_topology_assert(); 2281 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2282 acw, ace); 2283 2284 dcr = pp->acr + acr; 2285 dcw = pp->acw + acw; 2286 dce = pp->ace + ace; 2287 2288 /* On first open, grab an extra "exclusive" bit */ 2289 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0) 2290 ace++; 2291 /* ... and let go of it on last close */ 2292 if (dcr == 0 && dcw == 0 && dce == 0) 2293 ace--; 2294 2295 sc = pp->geom->softc; 2296 if (sc == NULL || LIST_EMPTY(&sc->sc_disks)) { 2297 if (acr <= 0 && acw <= 0 && ace <= 0) 2298 return (0); 2299 else 2300 return (ENXIO); 2301 } 2302 error = ENXIO; 2303 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2304 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2305 continue; 2306 err = g_access(disk->d_consumer, acr, acw, ace); 2307 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 2308 g_mirror_get_diskname(disk), acr, acw, ace, err); 2309 if (err == 0) { 2310 /* 2311 * Mark disk as dirty on open and unmark on close. 2312 */ 2313 if (pp->acw == 0 && dcw > 0) { 2314 G_MIRROR_DEBUG(1, 2315 "Disk %s (device %s) marked as dirty.", 2316 g_mirror_get_diskname(disk), sc->sc_name); 2317 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2318 g_mirror_update_metadata(disk); 2319 } else if (pp->acw > 0 && dcw == 0) { 2320 G_MIRROR_DEBUG(1, 2321 "Disk %s (device %s) marked as clean.", 2322 g_mirror_get_diskname(disk), sc->sc_name); 2323 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2324 g_mirror_update_metadata(disk); 2325 } 2326 error = 0; 2327 } else { 2328 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 2329 g_mirror_event_send(disk, 2330 G_MIRROR_DISK_STATE_DISCONNECTED, 2331 G_MIRROR_EVENT_DONTWAIT); 2332 } 2333 } 2334 return (error); 2335} 2336 2337static struct g_geom * 2338g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2339{ 2340 struct g_mirror_softc *sc; 2341 struct g_geom *gp; 2342 int error, timeout; 2343 2344 g_topology_assert(); 2345 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2346 md->md_mid); 2347 2348 /* One disk is minimum. */ 2349 if (md->md_all < 1) 2350 return (NULL); 2351 /* 2352 * Action geom. 2353 */ 2354 gp = g_new_geomf(mp, "%s", md->md_name); 2355 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2356 gp->start = g_mirror_start; 2357 gp->spoiled = g_mirror_spoiled; 2358 gp->orphan = g_mirror_orphan; 2359 gp->access = g_mirror_access; 2360 gp->dumpconf = g_mirror_dumpconf; 2361 2362 sc->sc_id = md->md_mid; 2363 sc->sc_slice = md->md_slice; 2364 sc->sc_balance = md->md_balance; 2365 sc->sc_mediasize = md->md_mediasize; 2366 sc->sc_sectorsize = md->md_sectorsize; 2367 sc->sc_ndisks = md->md_all; 2368 sc->sc_flags = md->md_mflags; 2369 sc->sc_bump_syncid = 0; 2370 bioq_init(&sc->sc_queue); 2371 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2372 LIST_INIT(&sc->sc_disks); 2373 TAILQ_INIT(&sc->sc_events); 2374 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2375 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2376 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2377 gp->softc = sc; 2378 sc->sc_geom = gp; 2379 sc->sc_provider = NULL; 2380 /* 2381 * Synchronization geom. 2382 */ 2383 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2384 gp->softc = sc; 2385 gp->spoiled = g_mirror_spoiled; 2386 gp->orphan = g_mirror_orphan; 2387 sc->sc_sync.ds_geom = gp; 2388 sc->sc_sync.ds_block = atomic_load_acq_int(&g_mirror_sync_block_size); 2389 sc->sc_sync.ds_ndisks = 0; 2390 sc->sc_sync.ds_zone = uma_zcreate("gmirror:sync", sc->sc_sync.ds_block, 2391 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2392 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2393 "g_mirror %s", md->md_name); 2394 if (error != 0) { 2395 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2396 sc->sc_name); 2397 uma_zdestroy(sc->sc_sync.ds_zone); 2398 g_destroy_geom(sc->sc_sync.ds_geom); 2399 mtx_destroy(&sc->sc_events_mtx); 2400 mtx_destroy(&sc->sc_queue_mtx); 2401 g_destroy_geom(sc->sc_geom); 2402 free(sc, M_MIRROR); 2403 return (NULL); 2404 } 2405 2406 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2407 2408 /* 2409 * Run timeout. 2410 */ 2411 timeout = atomic_load_acq_int(&g_mirror_timeout); 2412 callout_reset(&sc->sc_callout, timeout * hz, g_mirror_go, sc); 2413 return (sc->sc_geom); 2414} 2415 2416int 2417g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2418{ 2419 struct g_provider *pp; 2420 2421 g_topology_assert(); 2422 2423 if (sc == NULL) 2424 return (ENXIO); 2425 pp = sc->sc_provider; 2426 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2427 if (force) { 2428 G_MIRROR_DEBUG(0, "Device %s is still open, so it " 2429 "can't be definitely removed.", pp->name); 2430 } else { 2431 G_MIRROR_DEBUG(1, 2432 "Device %s is still open (r%dw%de%d).", pp->name, 2433 pp->acr, pp->acw, pp->ace); 2434 return (EBUSY); 2435 } 2436 } 2437 2438 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2439 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2440 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2441 mtx_lock(&sc->sc_queue_mtx); 2442 wakeup(sc); 2443 mtx_unlock(&sc->sc_queue_mtx); 2444 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2445 while (sc->sc_worker != NULL) 2446 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2447 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2448 g_mirror_destroy_device(sc); 2449 free(sc, M_MIRROR); 2450 return (0); 2451} 2452 2453static void 2454g_mirror_taste_orphan(struct g_consumer *cp) 2455{ 2456 2457 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2458 cp->provider->name)); 2459} 2460 2461static struct g_geom * 2462g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2463{ 2464 struct g_mirror_metadata md; 2465 struct g_mirror_softc *sc; 2466 struct g_consumer *cp; 2467 struct g_geom *gp; 2468 int error; 2469 2470 g_topology_assert(); 2471 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2472 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2473 2474 gp = g_new_geomf(mp, "mirror:taste"); 2475 /* 2476 * This orphan function should be never called. 2477 */ 2478 gp->orphan = g_mirror_taste_orphan; 2479 cp = g_new_consumer(gp); 2480 g_attach(cp, pp); 2481 error = g_mirror_read_metadata(cp, &md); 2482 g_detach(cp); 2483 g_destroy_consumer(cp); 2484 g_destroy_geom(gp); 2485 if (error != 0) 2486 return (NULL); 2487 gp = NULL; 2488 2489 if (md.md_version > G_MIRROR_VERSION) { 2490 printf("geom_mirror.ko module is too old to handle %s.\n", 2491 pp->name); 2492 return (NULL); 2493 } 2494 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2495 G_MIRROR_DEBUG(0, 2496 "Device %s: provider %s marked as inactive, skipping.", 2497 md.md_name, pp->name); 2498 return (NULL); 2499 } 2500 if (g_mirror_debug >= 2) 2501 mirror_metadata_dump(&md); 2502 2503 /* 2504 * Let's check if device already exists. 2505 */ 2506 LIST_FOREACH(gp, &mp->geom, geom) { 2507 sc = gp->softc; 2508 if (sc == NULL) 2509 continue; 2510 if (sc->sc_sync.ds_geom == gp) 2511 continue; 2512 if (strcmp(md.md_name, sc->sc_name) != 0) 2513 continue; 2514 if (md.md_mid != sc->sc_id) { 2515 G_MIRROR_DEBUG(0, "Device %s already configured.", 2516 sc->sc_name); 2517 return (NULL); 2518 } 2519 break; 2520 } 2521 if (gp == NULL) { 2522 gp = g_mirror_create(mp, &md); 2523 if (gp == NULL) { 2524 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2525 md.md_name); 2526 return (NULL); 2527 } 2528 sc = gp->softc; 2529 } 2530 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2531 error = g_mirror_add_disk(sc, pp, &md); 2532 if (error != 0) { 2533 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2534 pp->name, gp->name, error); 2535 if (LIST_EMPTY(&sc->sc_disks)) 2536 g_mirror_destroy(sc, 1); 2537 return (NULL); 2538 } 2539 return (gp); 2540} 2541 2542static int 2543g_mirror_destroy_geom(struct gctl_req *req __unused, 2544 struct g_class *mp __unused, struct g_geom *gp) 2545{ 2546 2547 return (g_mirror_destroy(gp->softc, 0)); 2548} 2549 2550static void 2551g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2552 struct g_consumer *cp, struct g_provider *pp) 2553{ 2554 struct g_mirror_softc *sc; 2555 2556 g_topology_assert(); 2557 2558 sc = gp->softc; 2559 if (sc == NULL) 2560 return; 2561 /* Skip synchronization geom. */ 2562 if (gp == sc->sc_sync.ds_geom) 2563 return; 2564 if (pp != NULL) { 2565 /* Nothing here. */ 2566 } else if (cp != NULL) { 2567 struct g_mirror_disk *disk; 2568 2569 disk = cp->private; 2570 if (disk == NULL) 2571 return; 2572 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2573 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2574 sbuf_printf(sb, "%s<Synchronized>", indent); 2575 if (disk->d_sync.ds_offset_done == 0) 2576 sbuf_printf(sb, "0%%"); 2577 else { 2578 sbuf_printf(sb, "%u%%", 2579 (u_int)((disk->d_sync.ds_offset_done * 100) / 2580 sc->sc_provider->mediasize)); 2581 } 2582 sbuf_printf(sb, "</Synchronized>\n"); 2583 } 2584 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2585 disk->d_sync.ds_syncid); 2586 sbuf_printf(sb, "%s<Flags>", indent); 2587 if (disk->d_flags == 0) 2588 sbuf_printf(sb, "NONE"); 2589 else { 2590 int first = 1; 2591 2592 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2593 if (!first) 2594 sbuf_printf(sb, ", "); 2595 else 2596 first = 0; 2597 sbuf_printf(sb, "DIRTY"); 2598 } 2599 if ((disk->d_flags & 2600 G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2601 if (!first) 2602 sbuf_printf(sb, ", "); 2603 else 2604 first = 0; 2605 sbuf_printf(sb, "INACTIVE"); 2606 } 2607 if ((disk->d_flags & 2608 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2609 if (!first) 2610 sbuf_printf(sb, ", "); 2611 else 2612 first = 0; 2613 sbuf_printf(sb, "SYNCHRONIZING"); 2614 } 2615 if ((disk->d_flags & 2616 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2617 if (!first) 2618 sbuf_printf(sb, ", "); 2619 else 2620 first = 0; 2621 sbuf_printf(sb, "FORCE_SYNC"); 2622 } 2623 } 2624 sbuf_printf(sb, "</Flags>\n"); 2625 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2626 disk->d_priority); 2627 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2628 g_mirror_disk_state2str(disk->d_state)); 2629 } else { 2630 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2631 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2632 sbuf_printf(sb, "%s<Flags>", indent); 2633 if (sc->sc_flags == 0) 2634 sbuf_printf(sb, "NONE"); 2635 else { 2636 int first = 1; 2637 2638 if ((sc->sc_flags & 2639 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0) { 2640 if (!first) 2641 sbuf_printf(sb, ", "); 2642 else 2643 first = 0; 2644 sbuf_printf(sb, "NOAUTOSYNC"); 2645 } 2646 } 2647 sbuf_printf(sb, "</Flags>\n"); 2648 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2649 (u_int)sc->sc_slice); 2650 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2651 balance_name(sc->sc_balance)); 2652 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2653 sc->sc_ndisks); 2654 } 2655} 2656 2657DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2658