g_mirror.c revision 133173
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 133173 2004-08-05 14:07:21Z 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 disk->d_sync.ds_offset_done = bp->bio_offset + bp->bio_length; 988 g_destroy_bio(bp); 989 if (disk->d_sync.ds_offset_done == sc->sc_provider->mediasize) { 990 /* 991 * Disk up-to-date, activate it. 992 */ 993 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 994 G_MIRROR_EVENT_DONTWAIT); 995 return; 996 } else if ((disk->d_sync.ds_offset_done % 997 (sc->sc_sync.ds_block * 100)) == 0) { 998 /* 999 * Update offset_done on every 100 blocks. 1000 * XXX: This should be configurable. 1001 */ 1002 g_topology_lock(); 1003 g_mirror_update_metadata(disk); 1004 g_topology_unlock(); 1005 } 1006 return; 1007 default: 1008 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1009 bp->bio_cmd, sc->sc_name)); 1010 break; 1011 } 1012} 1013 1014static void 1015g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1016{ 1017 struct g_mirror_disk *disk; 1018 struct g_consumer *cp; 1019 struct bio *cbp; 1020 1021 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1022 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1023 break; 1024 } 1025 if (disk == NULL) { 1026 if (bp->bio_error == 0) 1027 bp->bio_error = ENXIO; 1028 g_io_deliver(bp, bp->bio_error); 1029 return; 1030 } 1031 cbp = g_clone_bio(bp); 1032 if (cbp == NULL) { 1033 if (bp->bio_error == 0) 1034 bp->bio_error = ENOMEM; 1035 g_io_deliver(bp, bp->bio_error); 1036 return; 1037 } 1038 /* 1039 * Fill in the component buf structure. 1040 */ 1041 cp = disk->d_consumer; 1042 cbp->bio_done = g_mirror_done; 1043 cbp->bio_to = cp->provider; 1044 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1045 KASSERT(cp->acr > 0 && cp->ace > 0, 1046 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1047 cp->acw, cp->ace)); 1048 g_io_request(cbp, cp); 1049} 1050 1051static void 1052g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1053{ 1054 struct g_mirror_disk *disk; 1055 struct g_consumer *cp; 1056 struct bio *cbp; 1057 1058 disk = g_mirror_get_disk(sc); 1059 if (disk == NULL) { 1060 if (bp->bio_error == 0) 1061 bp->bio_error = ENXIO; 1062 g_io_deliver(bp, bp->bio_error); 1063 return; 1064 } 1065 cbp = g_clone_bio(bp); 1066 if (cbp == NULL) { 1067 if (bp->bio_error == 0) 1068 bp->bio_error = ENOMEM; 1069 g_io_deliver(bp, bp->bio_error); 1070 return; 1071 } 1072 /* 1073 * Fill in the component buf structure. 1074 */ 1075 cp = disk->d_consumer; 1076 cbp->bio_done = g_mirror_done; 1077 cbp->bio_to = cp->provider; 1078 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1079 KASSERT(cp->acr > 0 && cp->ace > 0, 1080 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1081 cp->acw, cp->ace)); 1082 g_io_request(cbp, cp); 1083} 1084 1085static void 1086g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1087{ 1088 struct g_mirror_disk *disk, *dp; 1089 struct g_consumer *cp; 1090 struct bio *cbp; 1091 struct bintime curtime; 1092 1093 binuptime(&curtime); 1094 /* 1095 * Find a disk which the smallest load. 1096 */ 1097 disk = NULL; 1098 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1099 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1100 continue; 1101 /* If disk wasn't used for more than 2 sec, use it. */ 1102 if (curtime.sec - dp->d_last_used.sec >= 2) { 1103 disk = dp; 1104 break; 1105 } 1106 if (disk == NULL || 1107 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1108 disk = dp; 1109 } 1110 } 1111 cbp = g_clone_bio(bp); 1112 if (cbp == NULL) { 1113 if (bp->bio_error == 0) 1114 bp->bio_error = ENOMEM; 1115 g_io_deliver(bp, bp->bio_error); 1116 return; 1117 } 1118 /* 1119 * Fill in the component buf structure. 1120 */ 1121 cp = disk->d_consumer; 1122 cbp->bio_done = g_mirror_done; 1123 cbp->bio_to = cp->provider; 1124 binuptime(&disk->d_last_used); 1125 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1126 KASSERT(cp->acr > 0 && cp->ace > 0, 1127 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1128 cp->acw, cp->ace)); 1129 g_io_request(cbp, cp); 1130} 1131 1132static void 1133g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1134{ 1135 struct bio_queue_head queue; 1136 struct g_mirror_disk *disk; 1137 struct g_consumer *cp; 1138 struct bio *cbp; 1139 off_t left, mod, offset, slice; 1140 u_char *data; 1141 u_int ndisks; 1142 1143 if (bp->bio_length <= sc->sc_slice) { 1144 g_mirror_request_round_robin(sc, bp); 1145 return; 1146 } 1147 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1148 slice = bp->bio_length / ndisks; 1149 mod = slice % sc->sc_provider->sectorsize; 1150 if (mod != 0) 1151 slice += sc->sc_provider->sectorsize - mod; 1152 /* 1153 * Allocate all bios before sending any request, so we can 1154 * return ENOMEM in nice and clean way. 1155 */ 1156 left = bp->bio_length; 1157 offset = bp->bio_offset; 1158 data = bp->bio_data; 1159 bioq_init(&queue); 1160 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1161 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1162 continue; 1163 cbp = g_clone_bio(bp); 1164 if (cbp == NULL) { 1165 for (cbp = bioq_first(&queue); cbp != NULL; 1166 cbp = bioq_first(&queue)) { 1167 bioq_remove(&queue, cbp); 1168 g_destroy_bio(cbp); 1169 } 1170 if (bp->bio_error == 0) 1171 bp->bio_error = ENOMEM; 1172 g_io_deliver(bp, bp->bio_error); 1173 return; 1174 } 1175 bioq_insert_tail(&queue, cbp); 1176 cbp->bio_done = g_mirror_done; 1177 cbp->bio_caller1 = disk; 1178 cbp->bio_to = disk->d_consumer->provider; 1179 cbp->bio_offset = offset; 1180 cbp->bio_data = data; 1181 cbp->bio_length = MIN(left, slice); 1182 left -= cbp->bio_length; 1183 if (left == 0) 1184 break; 1185 offset += cbp->bio_length; 1186 data += cbp->bio_length; 1187 } 1188 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1189 bioq_remove(&queue, cbp); 1190 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1191 disk = cbp->bio_caller1; 1192 cbp->bio_caller1 = NULL; 1193 cp = disk->d_consumer; 1194 KASSERT(cp->acr > 0 && cp->ace > 0, 1195 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1196 cp->acr, cp->acw, cp->ace)); 1197 g_io_request(cbp, disk->d_consumer); 1198 } 1199} 1200 1201static void 1202g_mirror_register_request(struct bio *bp) 1203{ 1204 struct g_mirror_softc *sc; 1205 1206 sc = bp->bio_to->geom->softc; 1207 switch (bp->bio_cmd) { 1208 case BIO_READ: 1209 switch (sc->sc_balance) { 1210 case G_MIRROR_BALANCE_LOAD: 1211 g_mirror_request_load(sc, bp); 1212 break; 1213 case G_MIRROR_BALANCE_PREFER: 1214 g_mirror_request_prefer(sc, bp); 1215 break; 1216 case G_MIRROR_BALANCE_ROUND_ROBIN: 1217 g_mirror_request_round_robin(sc, bp); 1218 break; 1219 case G_MIRROR_BALANCE_SPLIT: 1220 g_mirror_request_split(sc, bp); 1221 break; 1222 } 1223 return; 1224 case BIO_WRITE: 1225 case BIO_DELETE: 1226 { 1227 struct g_mirror_disk *disk; 1228 struct bio_queue_head queue; 1229 struct g_consumer *cp; 1230 struct bio *cbp; 1231 1232 /* 1233 * Allocate all bios before sending any request, so we can 1234 * return ENOMEM in nice and clean way. 1235 */ 1236 bioq_init(&queue); 1237 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1238 switch (disk->d_state) { 1239 case G_MIRROR_DISK_STATE_ACTIVE: 1240 break; 1241 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1242 if (bp->bio_offset >= disk->d_sync.ds_offset) 1243 continue; 1244 break; 1245 default: 1246 continue; 1247 } 1248 cbp = g_clone_bio(bp); 1249 if (cbp == NULL) { 1250 for (cbp = bioq_first(&queue); cbp != NULL; 1251 cbp = bioq_first(&queue)) { 1252 bioq_remove(&queue, cbp); 1253 g_destroy_bio(cbp); 1254 } 1255 if (bp->bio_error == 0) 1256 bp->bio_error = ENOMEM; 1257 g_io_deliver(bp, bp->bio_error); 1258 return; 1259 } 1260 bioq_insert_tail(&queue, cbp); 1261 } 1262 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1263 switch (disk->d_state) { 1264 case G_MIRROR_DISK_STATE_ACTIVE: 1265 break; 1266 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1267 if (bp->bio_offset >= disk->d_sync.ds_offset) 1268 continue; 1269 break; 1270 default: 1271 continue; 1272 } 1273 cbp = bioq_first(&queue); 1274 KASSERT(cbp != NULL, ("NULL cbp! (device %s).", 1275 sc->sc_name)); 1276 bioq_remove(&queue, cbp); 1277 cp = disk->d_consumer; 1278 cbp->bio_done = g_mirror_done; 1279 cbp->bio_to = cp->provider; 1280 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1281 KASSERT(cp->acw > 0 && cp->ace > 0, 1282 ("Consumer %s not opened (r%dw%de%d).", 1283 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1284 g_io_request(cbp, cp); 1285 } 1286 /* 1287 * Bump syncid on first write. 1288 */ 1289 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) { 1290 sc->sc_bump_syncid = 0; 1291 g_topology_lock(); 1292 g_mirror_bump_syncid(sc); 1293 g_topology_unlock(); 1294 } 1295 return; 1296 } 1297 default: 1298 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1299 bp->bio_cmd, sc->sc_name)); 1300 break; 1301 } 1302} 1303 1304/* 1305 * Worker thread. 1306 */ 1307static void 1308g_mirror_worker(void *arg) 1309{ 1310 struct g_mirror_softc *sc; 1311 struct g_mirror_disk *disk; 1312 struct g_mirror_event *ep; 1313 struct bio *bp; 1314 u_int nreqs; 1315 1316 sc = arg; 1317 curthread->td_base_pri = PRIBIO; 1318 1319 nreqs = 0; 1320 for (;;) { 1321 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1322 /* 1323 * First take a look at events. 1324 * This is important to handle events before any I/O requests. 1325 */ 1326 ep = g_mirror_event_get(sc); 1327 if (ep != NULL) { 1328 g_topology_lock(); 1329 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1330 /* Update only device status. */ 1331 G_MIRROR_DEBUG(3, 1332 "Running event for device %s.", 1333 sc->sc_name); 1334 ep->e_error = 0; 1335 g_mirror_update_device(sc, 1); 1336 } else { 1337 /* Update disk status. */ 1338 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1339 g_mirror_get_diskname(ep->e_disk)); 1340 ep->e_error = g_mirror_update_disk(ep->e_disk, 1341 ep->e_state); 1342 if (ep->e_error == 0) 1343 g_mirror_update_device(sc, 0); 1344 } 1345 g_topology_unlock(); 1346 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1347 KASSERT(ep->e_error == 0, 1348 ("Error cannot be handled.")); 1349 g_mirror_event_free(ep); 1350 } else { 1351 ep->e_flags |= G_MIRROR_EVENT_DONE; 1352 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1353 ep); 1354 mtx_lock(&sc->sc_events_mtx); 1355 wakeup(ep); 1356 mtx_unlock(&sc->sc_events_mtx); 1357 } 1358 if ((sc->sc_flags & 1359 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1360end: 1361 if ((sc->sc_flags & 1362 G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1363 G_MIRROR_DEBUG(4, "%s: Waking up %p.", 1364 __func__, &sc->sc_worker); 1365 wakeup(&sc->sc_worker); 1366 sc->sc_worker = NULL; 1367 } else { 1368 g_topology_lock(); 1369 g_mirror_destroy_device(sc); 1370 g_topology_unlock(); 1371 free(sc, M_MIRROR); 1372 } 1373 kthread_exit(0); 1374 } 1375 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1376 continue; 1377 } 1378 /* 1379 * Now I/O requests. 1380 */ 1381 /* Get first request from the queue. */ 1382 mtx_lock(&sc->sc_queue_mtx); 1383 bp = bioq_first(&sc->sc_queue); 1384 if (bp == NULL) { 1385 if ((sc->sc_flags & 1386 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1387 mtx_unlock(&sc->sc_queue_mtx); 1388 goto end; 1389 } 1390 } 1391 if (sc->sc_sync.ds_ndisks > 0 && 1392 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1393 mtx_unlock(&sc->sc_queue_mtx); 1394 /* 1395 * It is time for synchronization... 1396 */ 1397 nreqs = 0; 1398 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1399 if (disk->d_state != 1400 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1401 continue; 1402 } 1403 if (disk->d_sync.ds_offset >= 1404 sc->sc_provider->mediasize) { 1405 continue; 1406 } 1407 if (disk->d_sync.ds_offset > 1408 disk->d_sync.ds_offset_done) { 1409 continue; 1410 } 1411 g_mirror_sync_one(disk); 1412 } 1413 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1414 goto sleep; 1415 } 1416 if (bp == NULL) { 1417 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 0); 1418 G_MIRROR_DEBUG(5, "%s: I'm here 3.", __func__); 1419 continue; 1420 } 1421 nreqs++; 1422 bioq_remove(&sc->sc_queue, bp); 1423 mtx_unlock(&sc->sc_queue_mtx); 1424 1425 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1426 g_mirror_regular_request(bp); 1427 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1428 u_int timeout, sps; 1429 1430 g_mirror_sync_request(bp); 1431sleep: 1432 sps = atomic_load_acq_int(&g_mirror_syncs_per_sec); 1433 if (sps == 0) { 1434 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1435 continue; 1436 } 1437 mtx_lock(&sc->sc_queue_mtx); 1438 if (bioq_first(&sc->sc_queue) != NULL) { 1439 mtx_unlock(&sc->sc_queue_mtx); 1440 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1441 continue; 1442 } 1443 timeout = hz / sps; 1444 if (timeout == 0) 1445 timeout = 1; 1446 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w2", 1447 timeout); 1448 } else { 1449 g_mirror_register_request(bp); 1450 } 1451 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1452 } 1453} 1454 1455/* 1456 * Open disk's consumer if needed. 1457 */ 1458static void 1459g_mirror_update_access(struct g_mirror_disk *disk) 1460{ 1461 struct g_provider *pp; 1462 struct g_consumer *cp; 1463 int acr, acw, ace, cpw, error; 1464 1465 g_topology_assert(); 1466 1467 cp = disk->d_consumer; 1468 pp = disk->d_softc->sc_provider; 1469 if (pp == NULL) { 1470 acr = -cp->acr; 1471 acw = -cp->acw; 1472 ace = -cp->ace; 1473 } else { 1474 acr = pp->acr - cp->acr; 1475 acw = pp->acw - cp->acw; 1476 ace = pp->ace - cp->ace; 1477 /* Grab an extra "exclusive" bit. */ 1478 if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) 1479 ace++; 1480 } 1481 if (acr == 0 && acw == 0 && ace == 0) 1482 return; 1483 cpw = cp->acw; 1484 error = g_access(cp, acr, acw, ace); 1485 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, acr, 1486 acw, ace, error); 1487 if (error != 0) { 1488 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1489 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1490 G_MIRROR_EVENT_DONTWAIT); 1491 return; 1492 } 1493 if (cpw == 0 && cp->acw > 0) { 1494 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1495 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1496 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1497 } else if (cpw > 0 && cp->acw == 0) { 1498 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1499 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1500 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1501 } 1502} 1503 1504static void 1505g_mirror_sync_start(struct g_mirror_disk *disk) 1506{ 1507 struct g_mirror_softc *sc; 1508 struct g_consumer *cp; 1509 int error; 1510 1511 g_topology_assert(); 1512 1513 sc = disk->d_softc; 1514 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1515 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1516 sc->sc_state)); 1517 cp = disk->d_consumer; 1518 KASSERT(cp->acr == 0 && cp->acw == 0 && cp->ace == 0, 1519 ("Consumer %s already opened.", cp->provider->name)); 1520 1521 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1522 g_mirror_get_diskname(disk)); 1523 error = g_access(cp, 0, 1, 1); 1524 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, 1, 1525 1, error); 1526 if (error != 0) { 1527 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1528 G_MIRROR_EVENT_DONTWAIT); 1529 return; 1530 } 1531 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1532 KASSERT(disk->d_sync.ds_consumer == NULL, 1533 ("Sync consumer already exists (device=%s, disk=%s).", 1534 sc->sc_name, g_mirror_get_diskname(disk))); 1535 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1536 disk->d_sync.ds_consumer->private = disk; 1537 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1538 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1539 disk->d_softc->sc_name, error)); 1540 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1541 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1542 disk->d_softc->sc_name, error)); 1543 sc->sc_sync.ds_ndisks++; 1544} 1545 1546/* 1547 * Stop synchronization process. 1548 * type: 0 - synchronization finished 1549 * 1 - synchronization stopped 1550 */ 1551static void 1552g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1553{ 1554 struct g_consumer *cp; 1555 1556 g_topology_assert(); 1557 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1558 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1559 g_mirror_disk_state2str(disk->d_state))); 1560 if (disk->d_sync.ds_consumer == NULL) 1561 return; 1562 1563 if (type == 0) { 1564 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1565 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1566 } else /* if (type == 1) */ { 1567 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1568 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1569 } 1570 cp = disk->d_sync.ds_consumer; 1571 g_access(cp, -1, 0, 0); 1572 g_mirror_kill_consumer(disk->d_softc, cp); 1573 disk->d_sync.ds_consumer = NULL; 1574 disk->d_softc->sc_sync.ds_ndisks--; 1575 cp = disk->d_consumer; 1576 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 1577 ("Consumer %s not opened.", cp->provider->name)); 1578 g_access(cp, 0, -1, -1); 1579 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, -1, 1580 -1, 0); 1581 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1582} 1583 1584static void 1585g_mirror_launch_provider(struct g_mirror_softc *sc) 1586{ 1587 struct g_mirror_disk *disk; 1588 struct g_provider *pp; 1589 1590 g_topology_assert(); 1591 1592 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1593 pp->mediasize = sc->sc_mediasize; 1594 pp->sectorsize = sc->sc_sectorsize; 1595 sc->sc_provider = pp; 1596 g_error_provider(pp, 0); 1597 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1598 pp->name); 1599 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1600 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1601 g_mirror_sync_start(disk); 1602 } 1603} 1604 1605static void 1606g_mirror_destroy_provider(struct g_mirror_softc *sc) 1607{ 1608 struct g_mirror_disk *disk; 1609 struct bio *bp; 1610 1611 g_topology_assert(); 1612 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1613 sc->sc_name)); 1614 1615 g_error_provider(sc->sc_provider, ENXIO); 1616 mtx_lock(&sc->sc_queue_mtx); 1617 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1618 bioq_remove(&sc->sc_queue, bp); 1619 g_io_deliver(bp, ENXIO); 1620 } 1621 mtx_unlock(&sc->sc_queue_mtx); 1622 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1623 sc->sc_provider->name); 1624 sc->sc_provider->flags |= G_PF_WITHER; 1625 g_orphan_provider(sc->sc_provider, ENXIO); 1626 sc->sc_provider = NULL; 1627 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1628 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1629 g_mirror_sync_stop(disk, 1); 1630 } 1631} 1632 1633static void 1634g_mirror_go(void *arg) 1635{ 1636 struct g_mirror_softc *sc; 1637 1638 sc = arg; 1639 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1640 g_mirror_event_send(sc, 0, 1641 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1642} 1643 1644static u_int 1645g_mirror_determine_state(struct g_mirror_disk *disk) 1646{ 1647 struct g_mirror_softc *sc; 1648 u_int state; 1649 1650 sc = disk->d_softc; 1651 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1652 if ((disk->d_flags & 1653 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1654 /* Disk does not need synchronization. */ 1655 state = G_MIRROR_DISK_STATE_ACTIVE; 1656 } else { 1657 if ((sc->sc_flags & 1658 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1659 (disk->d_flags & 1660 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1661 /* 1662 * We can start synchronization from 1663 * the stored offset. 1664 */ 1665 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1666 } else { 1667 state = G_MIRROR_DISK_STATE_STALE; 1668 } 1669 } 1670 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1671 /* 1672 * Reset all synchronization data for this disk, 1673 * because if it even was synchronized, it was 1674 * synchronized to disks with different syncid. 1675 */ 1676 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1677 disk->d_sync.ds_offset = 0; 1678 disk->d_sync.ds_offset_done = 0; 1679 disk->d_sync.ds_syncid = sc->sc_syncid; 1680 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1681 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1682 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1683 } else { 1684 state = G_MIRROR_DISK_STATE_STALE; 1685 } 1686 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1687 /* 1688 * Not good, NOT GOOD! 1689 * It means that mirror was started on stale disks 1690 * and more fresh disk just arrive. 1691 * If there were writes, mirror is fucked up, sorry. 1692 * I think the best choice here is don't touch 1693 * this disk and inform the user laudly. 1694 */ 1695 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1696 "disk (%s) arrives!! It will not be connected to the " 1697 "running device.", sc->sc_name, 1698 g_mirror_get_diskname(disk)); 1699 g_mirror_destroy_disk(disk); 1700 state = G_MIRROR_DISK_STATE_NONE; 1701 /* Return immediately, because disk was destroyed. */ 1702 return (state); 1703 } 1704 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1705 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1706 return (state); 1707} 1708 1709/* 1710 * Update device state. 1711 */ 1712static void 1713g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1714{ 1715 struct g_mirror_disk *disk; 1716 u_int state; 1717 1718 g_topology_assert(); 1719 1720 switch (sc->sc_state) { 1721 case G_MIRROR_DEVICE_STATE_STARTING: 1722 { 1723 struct g_mirror_disk *pdisk; 1724 u_int dirty, ndisks, syncid; 1725 1726 KASSERT(sc->sc_provider == NULL, 1727 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1728 /* 1729 * Are we ready? We are, if all disks are connected or 1730 * if we have any disks and 'force' is true. 1731 */ 1732 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1733 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1734 ; 1735 } else if (g_mirror_ndisks(sc, -1) == 0) { 1736 /* 1737 * Disks went down in starting phase, so destroy 1738 * device. 1739 */ 1740 callout_drain(&sc->sc_callout); 1741 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1742 return; 1743 } else { 1744 return; 1745 } 1746 1747 /* 1748 * Activate all disks with the biggest syncid. 1749 */ 1750 if (force) { 1751 /* 1752 * If 'force' is true, we have been called due to 1753 * timeout, so don't bother canceling timeout. 1754 */ 1755 ndisks = 0; 1756 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1757 if ((disk->d_flags & 1758 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1759 ndisks++; 1760 } 1761 } 1762 if (ndisks == 0) { 1763 /* No valid disks found, destroy device. */ 1764 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1765 return; 1766 } 1767 } else { 1768 /* Cancel timeout. */ 1769 callout_drain(&sc->sc_callout); 1770 } 1771 1772 /* 1773 * Find disk with the biggest syncid. 1774 */ 1775 syncid = 0; 1776 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1777 if (disk->d_sync.ds_syncid > syncid) 1778 syncid = disk->d_sync.ds_syncid; 1779 } 1780 1781 /* 1782 * Here we need to look for dirty disks and if all disks 1783 * with the biggest syncid are dirty, we have to choose 1784 * one with the biggest priority and rebuild the rest. 1785 */ 1786 /* 1787 * Find the number of dirty disks with the biggest syncid. 1788 * Find the number of disks with the biggest syncid. 1789 * While here, find a disk with the biggest priority. 1790 */ 1791 dirty = ndisks = 0; 1792 pdisk = NULL; 1793 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1794 if (disk->d_sync.ds_syncid != syncid) 1795 continue; 1796 if ((disk->d_flags & 1797 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1798 continue; 1799 } 1800 ndisks++; 1801 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1802 dirty++; 1803 if (pdisk == NULL || 1804 pdisk->d_priority < disk->d_priority) { 1805 pdisk = disk; 1806 } 1807 } 1808 } 1809 if (dirty == 0) { 1810 /* No dirty disks at all, great. */ 1811 } else if (dirty == ndisks) { 1812 /* 1813 * Force synchronization for all dirty disks except one 1814 * with the biggest priority. 1815 */ 1816 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1817 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1818 "master disk for synchronization.", 1819 g_mirror_get_diskname(pdisk), sc->sc_name); 1820 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1821 if (disk->d_sync.ds_syncid != syncid) 1822 continue; 1823 if ((disk->d_flags & 1824 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1825 continue; 1826 } 1827 KASSERT((disk->d_flags & 1828 G_MIRROR_DISK_FLAG_DIRTY) != 0, 1829 ("Disk %s isn't marked as dirty.", 1830 g_mirror_get_diskname(disk))); 1831 /* Skip the disk with the biggest priority. */ 1832 if (disk == pdisk) 1833 continue; 1834 disk->d_sync.ds_syncid = 0; 1835 } 1836 } else if (dirty < ndisks) { 1837 /* 1838 * Force synchronization for all dirty disks. 1839 * We have some non-dirty disks. 1840 */ 1841 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1842 if (disk->d_sync.ds_syncid != syncid) 1843 continue; 1844 if ((disk->d_flags & 1845 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1846 continue; 1847 } 1848 if ((disk->d_flags & 1849 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1850 continue; 1851 } 1852 disk->d_sync.ds_syncid = 0; 1853 } 1854 } 1855 1856 /* Reset hint. */ 1857 sc->sc_hint = NULL; 1858 sc->sc_syncid = syncid; 1859 if (force) { 1860 /* Remember to bump syncid on first write. */ 1861 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1862 } 1863 state = G_MIRROR_DEVICE_STATE_RUNNING; 1864 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 1865 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 1866 g_mirror_device_state2str(state)); 1867 sc->sc_state = state; 1868 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1869 state = g_mirror_determine_state(disk); 1870 g_mirror_event_send(disk, state, 1871 G_MIRROR_EVENT_DONTWAIT); 1872 if (state == G_MIRROR_DISK_STATE_STALE) { 1873 sc->sc_bump_syncid = 1874 G_MIRROR_BUMP_ON_FIRST_WRITE; 1875 } 1876 } 1877 break; 1878 } 1879 case G_MIRROR_DEVICE_STATE_RUNNING: 1880 /* 1881 * Bump syncid here, if we need to do it immediately. 1882 */ 1883 if (sc->sc_bump_syncid == G_MIRROR_BUMP_IMMEDIATELY) { 1884 sc->sc_bump_syncid = 0; 1885 g_mirror_bump_syncid(sc); 1886 } 1887 if (g_mirror_ndisks(sc, -1) == 0) { 1888 /* 1889 * No disks at all, we need to destroy device. 1890 */ 1891 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1892 } else if (g_mirror_ndisks(sc, 1893 G_MIRROR_DISK_STATE_ACTIVE) == 0 && 1894 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 1895 /* 1896 * No active disks, destroy provider. 1897 */ 1898 if (sc->sc_provider != NULL) 1899 g_mirror_destroy_provider(sc); 1900 } else if (g_mirror_ndisks(sc, 1901 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 1902 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 1903 /* 1904 * We have active disks, launch provider if it doesn't 1905 * exist. 1906 */ 1907 if (sc->sc_provider == NULL) 1908 g_mirror_launch_provider(sc); 1909 } 1910 break; 1911 default: 1912 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 1913 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 1914 break; 1915 } 1916} 1917 1918/* 1919 * Update disk state and device state if needed. 1920 */ 1921#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 1922 "Disk %s state changed from %s to %s (device %s).", \ 1923 g_mirror_get_diskname(disk), \ 1924 g_mirror_disk_state2str(disk->d_state), \ 1925 g_mirror_disk_state2str(state), sc->sc_name) 1926static int 1927g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 1928{ 1929 struct g_mirror_softc *sc; 1930 1931 g_topology_assert(); 1932 1933 sc = disk->d_softc; 1934again: 1935 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 1936 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 1937 g_mirror_disk_state2str(state)); 1938 switch (state) { 1939 case G_MIRROR_DISK_STATE_NEW: 1940 /* 1941 * Possible scenarios: 1942 * 1. New disk arrive. 1943 */ 1944 /* Previous state should be NONE. */ 1945 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 1946 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1947 g_mirror_disk_state2str(disk->d_state))); 1948 DISK_STATE_CHANGED(); 1949 1950 disk->d_state = state; 1951 if (LIST_EMPTY(&sc->sc_disks)) 1952 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 1953 else { 1954 struct g_mirror_disk *dp; 1955 1956 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1957 if (disk->d_priority >= dp->d_priority) { 1958 LIST_INSERT_BEFORE(dp, disk, d_next); 1959 dp = NULL; 1960 break; 1961 } 1962 if (LIST_NEXT(dp, d_next) == NULL) 1963 break; 1964 } 1965 if (dp != NULL) 1966 LIST_INSERT_AFTER(dp, disk, d_next); 1967 } 1968 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 1969 sc->sc_name, g_mirror_get_diskname(disk)); 1970 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 1971 break; 1972 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1973 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 1974 g_mirror_device_state2str(sc->sc_state), 1975 g_mirror_get_diskname(disk), 1976 g_mirror_disk_state2str(disk->d_state))); 1977 state = g_mirror_determine_state(disk); 1978 if (state != G_MIRROR_DISK_STATE_NONE) 1979 goto again; 1980 break; 1981 case G_MIRROR_DISK_STATE_ACTIVE: 1982 /* 1983 * Possible scenarios: 1984 * 1. New disk does not need synchronization. 1985 * 2. Synchronization process finished successfully. 1986 */ 1987 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1988 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 1989 g_mirror_device_state2str(sc->sc_state), 1990 g_mirror_get_diskname(disk), 1991 g_mirror_disk_state2str(disk->d_state))); 1992 /* Previous state should be NEW or SYNCHRONIZING. */ 1993 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 1994 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1995 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1996 g_mirror_disk_state2str(disk->d_state))); 1997 DISK_STATE_CHANGED(); 1998 1999 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2000 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2001 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2002 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2003 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2004 g_mirror_sync_stop(disk, 0); 2005 } 2006 disk->d_state = state; 2007 disk->d_sync.ds_offset = 0; 2008 disk->d_sync.ds_offset_done = 0; 2009 g_mirror_update_access(disk); 2010 g_mirror_update_metadata(disk); 2011 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2012 sc->sc_name, g_mirror_get_diskname(disk)); 2013 break; 2014 case G_MIRROR_DISK_STATE_STALE: 2015 /* 2016 * Possible scenarios: 2017 * 1. Stale disk was connected. 2018 */ 2019 /* Previous state should be NEW. */ 2020 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2021 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2022 g_mirror_disk_state2str(disk->d_state))); 2023 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2024 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2025 g_mirror_device_state2str(sc->sc_state), 2026 g_mirror_get_diskname(disk), 2027 g_mirror_disk_state2str(disk->d_state))); 2028 /* 2029 * STALE state is only possible if device is marked 2030 * NOAUTOSYNC. 2031 */ 2032 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2033 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2034 g_mirror_device_state2str(sc->sc_state), 2035 g_mirror_get_diskname(disk), 2036 g_mirror_disk_state2str(disk->d_state))); 2037 DISK_STATE_CHANGED(); 2038 2039 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2040 disk->d_state = state; 2041 g_mirror_update_metadata(disk); 2042 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2043 sc->sc_name, g_mirror_get_diskname(disk)); 2044 break; 2045 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2046 /* 2047 * Possible scenarios: 2048 * 1. Disk which needs synchronization was connected. 2049 */ 2050 /* Previous state should be NEW. */ 2051 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2052 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2053 g_mirror_disk_state2str(disk->d_state))); 2054 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2055 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2056 g_mirror_device_state2str(sc->sc_state), 2057 g_mirror_get_diskname(disk), 2058 g_mirror_disk_state2str(disk->d_state))); 2059 DISK_STATE_CHANGED(); 2060 2061 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2062 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2063 disk->d_state = state; 2064 if (sc->sc_provider != NULL) { 2065 g_mirror_sync_start(disk); 2066 g_mirror_update_metadata(disk); 2067 } 2068 break; 2069 case G_MIRROR_DISK_STATE_DISCONNECTED: 2070 /* 2071 * Possible scenarios: 2072 * 1. Device wasn't running yet, but disk disappear. 2073 * 2. Disk was active and disapppear. 2074 * 3. Disk disappear during synchronization process. 2075 */ 2076 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2077 /* 2078 * Previous state should be ACTIVE, STALE or 2079 * SYNCHRONIZING. 2080 */ 2081 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2082 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2083 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2084 ("Wrong disk state (%s, %s).", 2085 g_mirror_get_diskname(disk), 2086 g_mirror_disk_state2str(disk->d_state))); 2087 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2088 /* Previous state should be NEW. */ 2089 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2090 ("Wrong disk state (%s, %s).", 2091 g_mirror_get_diskname(disk), 2092 g_mirror_disk_state2str(disk->d_state))); 2093 /* 2094 * Reset bumping syncid if disk disappeared in STARTING 2095 * state. 2096 */ 2097 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) 2098 sc->sc_bump_syncid = 0; 2099#ifdef INVARIANTS 2100 } else { 2101 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2102 sc->sc_name, 2103 g_mirror_device_state2str(sc->sc_state), 2104 g_mirror_get_diskname(disk), 2105 g_mirror_disk_state2str(disk->d_state))); 2106#endif 2107 } 2108 DISK_STATE_CHANGED(); 2109 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2110 sc->sc_name, g_mirror_get_diskname(disk)); 2111 2112 g_mirror_destroy_disk(disk); 2113 break; 2114 case G_MIRROR_DISK_STATE_DESTROY: 2115 { 2116 int error; 2117 2118 error = g_mirror_clear_metadata(disk); 2119 if (error != 0) 2120 return (error); 2121 DISK_STATE_CHANGED(); 2122 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2123 sc->sc_name, g_mirror_get_diskname(disk)); 2124 2125 g_mirror_destroy_disk(disk); 2126 sc->sc_ndisks--; 2127 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2128 g_mirror_update_metadata(disk); 2129 } 2130 break; 2131 } 2132 default: 2133 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2134 break; 2135 } 2136 return (0); 2137} 2138#undef DISK_STATE_CHANGED 2139 2140static int 2141g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2142{ 2143 struct g_provider *pp; 2144 u_char *buf; 2145 int error; 2146 2147 g_topology_assert(); 2148 2149 error = g_access(cp, 1, 0, 0); 2150 if (error != 0) 2151 return (error); 2152 pp = cp->provider; 2153 g_topology_unlock(); 2154 /* Metadata are stored on last sector. */ 2155 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2156 &error); 2157 g_topology_lock(); 2158 if (buf == NULL) { 2159 g_access(cp, -1, 0, 0); 2160 return (error); 2161 } 2162 if (error != 0) { 2163 g_access(cp, -1, 0, 0); 2164 g_free(buf); 2165 return (error); 2166 } 2167 error = g_access(cp, -1, 0, 0); 2168 KASSERT(error == 0, ("Cannot decrease access count for %s.", pp->name)); 2169 2170 /* Decode metadata. */ 2171 error = mirror_metadata_decode(buf, md); 2172 g_free(buf); 2173 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2174 return (EINVAL); 2175 if (error != 0) { 2176 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2177 cp->provider->name); 2178 return (error); 2179 } 2180 2181 return (0); 2182} 2183 2184static int 2185g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2186 struct g_mirror_metadata *md) 2187{ 2188 2189 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2190 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2191 pp->name, md->md_did); 2192 return (EEXIST); 2193 } 2194 if (md->md_all != sc->sc_ndisks) { 2195 G_MIRROR_DEBUG(1, 2196 "Invalid '%s' field on disk %s (device %s), skipping.", 2197 "md_all", pp->name, sc->sc_name); 2198 return (EINVAL); 2199 } 2200 if (md->md_slice != sc->sc_slice) { 2201 G_MIRROR_DEBUG(1, 2202 "Invalid '%s' field on disk %s (device %s), skipping.", 2203 "md_slice", pp->name, sc->sc_name); 2204 return (EINVAL); 2205 } 2206 if (md->md_balance != sc->sc_balance) { 2207 G_MIRROR_DEBUG(1, 2208 "Invalid '%s' field on disk %s (device %s), skipping.", 2209 "md_balance", pp->name, sc->sc_name); 2210 return (EINVAL); 2211 } 2212 if (md->md_mediasize != sc->sc_mediasize) { 2213 G_MIRROR_DEBUG(1, 2214 "Invalid '%s' field on disk %s (device %s), skipping.", 2215 "md_mediasize", pp->name, sc->sc_name); 2216 return (EINVAL); 2217 } 2218 if (sc->sc_mediasize > pp->mediasize) { 2219 G_MIRROR_DEBUG(1, 2220 "Invalid size of disk %s (device %s), skipping.", pp->name, 2221 sc->sc_name); 2222 return (EINVAL); 2223 } 2224 if (md->md_sectorsize != sc->sc_sectorsize) { 2225 G_MIRROR_DEBUG(1, 2226 "Invalid '%s' field on disk %s (device %s), skipping.", 2227 "md_sectorsize", pp->name, sc->sc_name); 2228 return (EINVAL); 2229 } 2230 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2231 G_MIRROR_DEBUG(1, 2232 "Invalid sector size of disk %s (device %s), skipping.", 2233 pp->name, sc->sc_name); 2234 return (EINVAL); 2235 } 2236 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2237 G_MIRROR_DEBUG(1, 2238 "Invalid device flags on disk %s (device %s), skipping.", 2239 pp->name, sc->sc_name); 2240 return (EINVAL); 2241 } 2242 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2243 G_MIRROR_DEBUG(1, 2244 "Invalid disk flags on disk %s (device %s), skipping.", 2245 pp->name, sc->sc_name); 2246 return (EINVAL); 2247 } 2248 return (0); 2249} 2250 2251static int 2252g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2253 struct g_mirror_metadata *md) 2254{ 2255 struct g_mirror_disk *disk; 2256 int error; 2257 2258 g_topology_assert(); 2259 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2260 2261 error = g_mirror_check_metadata(sc, pp, md); 2262 if (error != 0) 2263 return (error); 2264 disk = g_mirror_init_disk(sc, pp, md, &error); 2265 if (disk == NULL) 2266 return (error); 2267 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2268 G_MIRROR_EVENT_WAIT); 2269 return (error); 2270} 2271 2272static int 2273g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2274{ 2275 struct g_mirror_softc *sc; 2276 struct g_mirror_disk *disk; 2277 int dcr, dcw, dce, err, error; 2278 2279 g_topology_assert(); 2280 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2281 acw, ace); 2282 2283 dcr = pp->acr + acr; 2284 dcw = pp->acw + acw; 2285 dce = pp->ace + ace; 2286 2287 /* On first open, grab an extra "exclusive" bit */ 2288 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0) 2289 ace++; 2290 /* ... and let go of it on last close */ 2291 if (dcr == 0 && dcw == 0 && dce == 0) 2292 ace--; 2293 2294 sc = pp->geom->softc; 2295 if (sc == NULL || LIST_EMPTY(&sc->sc_disks)) { 2296 if (acr <= 0 && acw <= 0 && ace <= 0) 2297 return (0); 2298 else 2299 return (ENXIO); 2300 } 2301 error = ENXIO; 2302 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2303 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2304 continue; 2305 err = g_access(disk->d_consumer, acr, acw, ace); 2306 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 2307 g_mirror_get_diskname(disk), acr, acw, ace, err); 2308 if (err == 0) { 2309 /* 2310 * Mark disk as dirty on open and unmark on close. 2311 */ 2312 if (pp->acw == 0 && dcw > 0) { 2313 G_MIRROR_DEBUG(1, 2314 "Disk %s (device %s) marked as dirty.", 2315 g_mirror_get_diskname(disk), sc->sc_name); 2316 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2317 g_mirror_update_metadata(disk); 2318 } else if (pp->acw > 0 && dcw == 0) { 2319 G_MIRROR_DEBUG(1, 2320 "Disk %s (device %s) marked as clean.", 2321 g_mirror_get_diskname(disk), sc->sc_name); 2322 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2323 g_mirror_update_metadata(disk); 2324 } 2325 error = 0; 2326 } else { 2327 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 2328 g_mirror_event_send(disk, 2329 G_MIRROR_DISK_STATE_DISCONNECTED, 2330 G_MIRROR_EVENT_DONTWAIT); 2331 } 2332 } 2333 return (error); 2334} 2335 2336static struct g_geom * 2337g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2338{ 2339 struct g_mirror_softc *sc; 2340 struct g_geom *gp; 2341 int error, timeout; 2342 2343 g_topology_assert(); 2344 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2345 md->md_mid); 2346 2347 /* One disk is minimum. */ 2348 if (md->md_all < 1) 2349 return (NULL); 2350 /* 2351 * Action geom. 2352 */ 2353 gp = g_new_geomf(mp, "%s", md->md_name); 2354 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2355 gp->start = g_mirror_start; 2356 gp->spoiled = g_mirror_spoiled; 2357 gp->orphan = g_mirror_orphan; 2358 gp->access = g_mirror_access; 2359 gp->dumpconf = g_mirror_dumpconf; 2360 2361 sc->sc_id = md->md_mid; 2362 sc->sc_slice = md->md_slice; 2363 sc->sc_balance = md->md_balance; 2364 sc->sc_mediasize = md->md_mediasize; 2365 sc->sc_sectorsize = md->md_sectorsize; 2366 sc->sc_ndisks = md->md_all; 2367 sc->sc_flags = md->md_mflags; 2368 sc->sc_bump_syncid = 0; 2369 bioq_init(&sc->sc_queue); 2370 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2371 LIST_INIT(&sc->sc_disks); 2372 TAILQ_INIT(&sc->sc_events); 2373 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2374 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2375 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2376 gp->softc = sc; 2377 sc->sc_geom = gp; 2378 sc->sc_provider = NULL; 2379 /* 2380 * Synchronization geom. 2381 */ 2382 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2383 gp->softc = sc; 2384 gp->spoiled = g_mirror_spoiled; 2385 gp->orphan = g_mirror_orphan; 2386 sc->sc_sync.ds_geom = gp; 2387 sc->sc_sync.ds_block = atomic_load_acq_int(&g_mirror_sync_block_size); 2388 sc->sc_sync.ds_ndisks = 0; 2389 sc->sc_sync.ds_zone = uma_zcreate("gmirror:sync", sc->sc_sync.ds_block, 2390 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2391 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2392 "g_mirror %s", md->md_name); 2393 if (error != 0) { 2394 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2395 sc->sc_name); 2396 uma_zdestroy(sc->sc_sync.ds_zone); 2397 g_destroy_geom(sc->sc_sync.ds_geom); 2398 mtx_destroy(&sc->sc_events_mtx); 2399 mtx_destroy(&sc->sc_queue_mtx); 2400 g_destroy_geom(sc->sc_geom); 2401 free(sc, M_MIRROR); 2402 return (NULL); 2403 } 2404 2405 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2406 2407 /* 2408 * Run timeout. 2409 */ 2410 timeout = atomic_load_acq_int(&g_mirror_timeout); 2411 callout_reset(&sc->sc_callout, timeout * hz, g_mirror_go, sc); 2412 return (sc->sc_geom); 2413} 2414 2415int 2416g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2417{ 2418 struct g_provider *pp; 2419 2420 g_topology_assert(); 2421 2422 if (sc == NULL) 2423 return (ENXIO); 2424 pp = sc->sc_provider; 2425 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2426 if (force) { 2427 G_MIRROR_DEBUG(0, "Device %s is still open, so it " 2428 "can't be definitely removed.", pp->name); 2429 } else { 2430 G_MIRROR_DEBUG(1, 2431 "Device %s is still open (r%dw%de%d).", pp->name, 2432 pp->acr, pp->acw, pp->ace); 2433 return (EBUSY); 2434 } 2435 } 2436 2437 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2438 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2439 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2440 mtx_lock(&sc->sc_queue_mtx); 2441 wakeup(sc); 2442 mtx_unlock(&sc->sc_queue_mtx); 2443 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2444 while (sc->sc_worker != NULL) 2445 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2446 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2447 g_mirror_destroy_device(sc); 2448 free(sc, M_MIRROR); 2449 return (0); 2450} 2451 2452static void 2453g_mirror_taste_orphan(struct g_consumer *cp) 2454{ 2455 2456 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2457 cp->provider->name)); 2458} 2459 2460static struct g_geom * 2461g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2462{ 2463 struct g_mirror_metadata md; 2464 struct g_mirror_softc *sc; 2465 struct g_consumer *cp; 2466 struct g_geom *gp; 2467 int error; 2468 2469 g_topology_assert(); 2470 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2471 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2472 2473 gp = g_new_geomf(mp, "mirror:taste"); 2474 /* 2475 * This orphan function should be never called. 2476 */ 2477 gp->orphan = g_mirror_taste_orphan; 2478 cp = g_new_consumer(gp); 2479 g_attach(cp, pp); 2480 error = g_mirror_read_metadata(cp, &md); 2481 g_detach(cp); 2482 g_destroy_consumer(cp); 2483 g_destroy_geom(gp); 2484 if (error != 0) 2485 return (NULL); 2486 gp = NULL; 2487 2488 if (md.md_version > G_MIRROR_VERSION) { 2489 printf("geom_mirror.ko module is too old to handle %s.\n", 2490 pp->name); 2491 return (NULL); 2492 } 2493 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2494 G_MIRROR_DEBUG(0, 2495 "Device %s: provider %s marked as inactive, skipping.", 2496 md.md_name, pp->name); 2497 return (NULL); 2498 } 2499 if (g_mirror_debug >= 2) 2500 mirror_metadata_dump(&md); 2501 2502 /* 2503 * Let's check if device already exists. 2504 */ 2505 LIST_FOREACH(gp, &mp->geom, geom) { 2506 sc = gp->softc; 2507 if (sc == NULL) 2508 continue; 2509 if (sc->sc_sync.ds_geom == gp) 2510 continue; 2511 if (strcmp(md.md_name, sc->sc_name) != 0) 2512 continue; 2513 if (md.md_mid != sc->sc_id) { 2514 G_MIRROR_DEBUG(0, "Device %s already configured.", 2515 sc->sc_name); 2516 return (NULL); 2517 } 2518 break; 2519 } 2520 if (gp == NULL) { 2521 gp = g_mirror_create(mp, &md); 2522 if (gp == NULL) { 2523 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2524 md.md_name); 2525 return (NULL); 2526 } 2527 sc = gp->softc; 2528 } 2529 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2530 error = g_mirror_add_disk(sc, pp, &md); 2531 if (error != 0) { 2532 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2533 pp->name, gp->name, error); 2534 if (LIST_EMPTY(&sc->sc_disks)) 2535 g_mirror_destroy(sc, 1); 2536 return (NULL); 2537 } 2538 return (gp); 2539} 2540 2541static int 2542g_mirror_destroy_geom(struct gctl_req *req __unused, 2543 struct g_class *mp __unused, struct g_geom *gp) 2544{ 2545 2546 return (g_mirror_destroy(gp->softc, 0)); 2547} 2548 2549static void 2550g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2551 struct g_consumer *cp, struct g_provider *pp) 2552{ 2553 struct g_mirror_softc *sc; 2554 2555 g_topology_assert(); 2556 2557 sc = gp->softc; 2558 if (sc == NULL) 2559 return; 2560 /* Skip synchronization geom. */ 2561 if (gp == sc->sc_sync.ds_geom) 2562 return; 2563 if (pp != NULL) { 2564 /* Nothing here. */ 2565 } else if (cp != NULL) { 2566 struct g_mirror_disk *disk; 2567 2568 disk = cp->private; 2569 if (disk == NULL) 2570 return; 2571 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2572 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2573 sbuf_printf(sb, "%s<Synchronized>", indent); 2574 if (disk->d_sync.ds_offset_done == 0) 2575 sbuf_printf(sb, "0%%"); 2576 else { 2577 sbuf_printf(sb, "%u%%", 2578 (u_int)((disk->d_sync.ds_offset_done * 100) / 2579 sc->sc_provider->mediasize)); 2580 } 2581 sbuf_printf(sb, "</Synchronized>\n"); 2582 } 2583 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2584 disk->d_sync.ds_syncid); 2585 sbuf_printf(sb, "%s<Flags>", indent); 2586 if (disk->d_flags == 0) 2587 sbuf_printf(sb, "NONE"); 2588 else { 2589 int first = 1; 2590 2591 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2592 if (!first) 2593 sbuf_printf(sb, ", "); 2594 else 2595 first = 0; 2596 sbuf_printf(sb, "DIRTY"); 2597 } 2598 if ((disk->d_flags & 2599 G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2600 if (!first) 2601 sbuf_printf(sb, ", "); 2602 else 2603 first = 0; 2604 sbuf_printf(sb, "INACTIVE"); 2605 } 2606 if ((disk->d_flags & 2607 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2608 if (!first) 2609 sbuf_printf(sb, ", "); 2610 else 2611 first = 0; 2612 sbuf_printf(sb, "SYNCHRONIZING"); 2613 } 2614 if ((disk->d_flags & 2615 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2616 if (!first) 2617 sbuf_printf(sb, ", "); 2618 else 2619 first = 0; 2620 sbuf_printf(sb, "FORCE_SYNC"); 2621 } 2622 } 2623 sbuf_printf(sb, "</Flags>\n"); 2624 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2625 disk->d_priority); 2626 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2627 g_mirror_disk_state2str(disk->d_state)); 2628 } else { 2629 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2630 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2631 sbuf_printf(sb, "%s<Flags>", indent); 2632 if (sc->sc_flags == 0) 2633 sbuf_printf(sb, "NONE"); 2634 else { 2635 int first = 1; 2636 2637 if ((sc->sc_flags & 2638 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0) { 2639 if (!first) 2640 sbuf_printf(sb, ", "); 2641 else 2642 first = 0; 2643 sbuf_printf(sb, "NOAUTOSYNC"); 2644 } 2645 } 2646 sbuf_printf(sb, "</Flags>\n"); 2647 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2648 (u_int)sc->sc_slice); 2649 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2650 balance_name(sc->sc_balance)); 2651 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2652 sc->sc_ndisks); 2653 } 2654} 2655 2656DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2657