g_mirror.c revision 133484
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 133484 2004-08-11 11:10:46Z 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 .version = G_VERSION, 85 .ctlreq = g_mirror_config, 86 .taste = g_mirror_taste, 87 .destroy_geom = g_mirror_destroy_geom 88}; 89 90 91static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 92static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 93static void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force); 94static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 95 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 96static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 97 98 99static const char * 100g_mirror_disk_state2str(int state) 101{ 102 103 switch (state) { 104 case G_MIRROR_DISK_STATE_NONE: 105 return ("NONE"); 106 case G_MIRROR_DISK_STATE_NEW: 107 return ("NEW"); 108 case G_MIRROR_DISK_STATE_ACTIVE: 109 return ("ACTIVE"); 110 case G_MIRROR_DISK_STATE_STALE: 111 return ("STALE"); 112 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 113 return ("SYNCHRONIZING"); 114 case G_MIRROR_DISK_STATE_DISCONNECTED: 115 return ("DISCONNECTED"); 116 case G_MIRROR_DISK_STATE_DESTROY: 117 return ("DESTROY"); 118 default: 119 return ("INVALID"); 120 } 121} 122 123static const char * 124g_mirror_device_state2str(int state) 125{ 126 127 switch (state) { 128 case G_MIRROR_DEVICE_STATE_STARTING: 129 return ("STARTING"); 130 case G_MIRROR_DEVICE_STATE_RUNNING: 131 return ("RUNNING"); 132 default: 133 return ("INVALID"); 134 } 135} 136 137static const char * 138g_mirror_get_diskname(struct g_mirror_disk *disk) 139{ 140 141 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 142 return ("[unknown]"); 143 return (disk->d_name); 144} 145 146/* 147 * --- Events handling functions --- 148 * Events in geom_mirror are used to maintain disks and device status 149 * from one thread to simplify locking. 150 */ 151static void 152g_mirror_event_free(struct g_mirror_event *ep) 153{ 154 155 free(ep, M_MIRROR); 156} 157 158int 159g_mirror_event_send(void *arg, int state, int flags) 160{ 161 struct g_mirror_softc *sc; 162 struct g_mirror_disk *disk; 163 struct g_mirror_event *ep; 164 int error; 165 166 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 167 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 168 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 169 disk = NULL; 170 sc = arg; 171 } else { 172 disk = arg; 173 sc = disk->d_softc; 174 } 175 ep->e_disk = disk; 176 ep->e_state = state; 177 ep->e_flags = flags; 178 ep->e_error = 0; 179 mtx_lock(&sc->sc_events_mtx); 180 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 181 mtx_unlock(&sc->sc_events_mtx); 182 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 183 mtx_lock(&sc->sc_queue_mtx); 184 wakeup(sc); 185 mtx_unlock(&sc->sc_queue_mtx); 186 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 187 return (0); 188 g_topology_assert(); 189 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 190 g_topology_unlock(); 191 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 192 mtx_lock(&sc->sc_events_mtx); 193 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 194 hz * 5); 195 } 196 /* Don't even try to use 'sc' here, because it could be already dead. */ 197 g_topology_lock(); 198 error = ep->e_error; 199 g_mirror_event_free(ep); 200 return (error); 201} 202 203static struct g_mirror_event * 204g_mirror_event_get(struct g_mirror_softc *sc) 205{ 206 struct g_mirror_event *ep; 207 208 mtx_lock(&sc->sc_events_mtx); 209 ep = TAILQ_FIRST(&sc->sc_events); 210 if (ep != NULL) 211 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 212 mtx_unlock(&sc->sc_events_mtx); 213 return (ep); 214} 215 216static void 217g_mirror_event_cancel(struct g_mirror_disk *disk) 218{ 219 struct g_mirror_softc *sc; 220 struct g_mirror_event *ep, *tmpep; 221 222 g_topology_assert(); 223 224 sc = disk->d_softc; 225 mtx_lock(&sc->sc_events_mtx); 226 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 227 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 228 continue; 229 if (ep->e_disk != disk) 230 continue; 231 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 232 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 233 g_mirror_event_free(ep); 234 else { 235 ep->e_error = ECANCELED; 236 wakeup(ep); 237 } 238 } 239 mtx_unlock(&sc->sc_events_mtx); 240} 241 242/* 243 * Return the number of disks in given state. 244 * If state is equal to -1, count all connected disks. 245 */ 246u_int 247g_mirror_ndisks(struct g_mirror_softc *sc, int state) 248{ 249 struct g_mirror_disk *disk; 250 u_int n = 0; 251 252 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 253 if (state == -1 || disk->d_state == state) 254 n++; 255 } 256 return (n); 257} 258 259/* 260 * Find a disk in mirror by its disk ID. 261 */ 262static struct g_mirror_disk * 263g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 264{ 265 struct g_mirror_disk *disk; 266 267 g_topology_assert(); 268 269 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 270 if (disk->d_id == id) 271 return (disk); 272 } 273 return (NULL); 274} 275 276static u_int 277g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 278{ 279 struct bio *bp; 280 u_int nreqs = 0; 281 282 mtx_lock(&sc->sc_queue_mtx); 283 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 284 if (bp->bio_from == cp) 285 nreqs++; 286 } 287 mtx_unlock(&sc->sc_queue_mtx); 288 return (nreqs); 289} 290 291static int 292g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 293{ 294 295 if (cp->nstart != cp->nend) { 296 G_MIRROR_DEBUG(2, 297 "I/O requests for %s exist, can't destroy it now.", 298 cp->provider->name); 299 return (1); 300 } 301 if (g_mirror_nrequests(sc, cp) > 0) { 302 G_MIRROR_DEBUG(2, 303 "I/O requests for %s in queue, can't destroy it now.", 304 cp->provider->name); 305 return (1); 306 } 307 return (0); 308} 309 310static void 311g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 312{ 313 314 g_topology_assert(); 315 316 cp->private = NULL; 317 if (g_mirror_is_busy(sc, cp)) 318 return; 319 G_MIRROR_DEBUG(2, "Consumer %s destroyed.", cp->provider->name); 320 g_detach(cp); 321 g_destroy_consumer(cp); 322} 323 324static int 325g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 326{ 327 int error; 328 329 g_topology_assert(); 330 KASSERT(disk->d_consumer == NULL, 331 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 332 333 disk->d_consumer = g_new_consumer(disk->d_softc->sc_geom); 334 disk->d_consumer->private = disk; 335 error = g_attach(disk->d_consumer, pp); 336 if (error != 0) 337 return (error); 338 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 339 return (0); 340} 341 342static void 343g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 344{ 345 346 g_topology_assert(); 347 348 if (cp == NULL) 349 return; 350 if (cp->provider != NULL) { 351 G_MIRROR_DEBUG(2, "Disk %s disconnected.", cp->provider->name); 352 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) { 353 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 354 cp->provider->name, -cp->acr, -cp->acw, -cp->ace, 355 0); 356 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 357 } 358 g_mirror_kill_consumer(sc, cp); 359 } else { 360 g_destroy_consumer(cp); 361 } 362} 363 364/* 365 * Initialize disk. This means allocate memory, create consumer, attach it 366 * to the provider and open access (r1w1e1) to it. 367 */ 368static struct g_mirror_disk * 369g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 370 struct g_mirror_metadata *md, int *errorp) 371{ 372 struct g_mirror_disk *disk; 373 int error; 374 375 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 376 if (disk == NULL) { 377 error = ENOMEM; 378 goto fail; 379 } 380 disk->d_softc = sc; 381 error = g_mirror_connect_disk(disk, pp); 382 if (error != 0) 383 goto fail; 384 disk->d_id = md->md_did; 385 disk->d_state = G_MIRROR_DISK_STATE_NONE; 386 disk->d_priority = md->md_priority; 387 disk->d_delay.sec = 0; 388 disk->d_delay.frac = 0; 389 binuptime(&disk->d_last_used); 390 disk->d_flags = md->md_dflags; 391 if (md->md_provider[0] != '\0') 392 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 393 disk->d_sync.ds_consumer = NULL; 394 disk->d_sync.ds_offset = md->md_sync_offset; 395 disk->d_sync.ds_offset_done = md->md_sync_offset; 396 disk->d_sync.ds_syncid = md->md_syncid; 397 if (errorp != NULL) 398 *errorp = 0; 399 return (disk); 400fail: 401 if (errorp != NULL) 402 *errorp = error; 403 if (disk != NULL) { 404 g_mirror_disconnect_consumer(sc, disk->d_consumer); 405 free(disk, M_MIRROR); 406 } 407 return (NULL); 408} 409 410static void 411g_mirror_destroy_disk(struct g_mirror_disk *disk) 412{ 413 struct g_mirror_softc *sc; 414 415 g_topology_assert(); 416 417 LIST_REMOVE(disk, d_next); 418 g_mirror_event_cancel(disk); 419 sc = disk->d_softc; 420 if (sc->sc_hint == disk) 421 sc->sc_hint = NULL; 422 switch (disk->d_state) { 423 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 424 g_mirror_sync_stop(disk, 1); 425 /* FALLTHROUGH */ 426 case G_MIRROR_DISK_STATE_NEW: 427 case G_MIRROR_DISK_STATE_STALE: 428 case G_MIRROR_DISK_STATE_ACTIVE: 429 g_mirror_disconnect_consumer(sc, disk->d_consumer); 430 free(disk, M_MIRROR); 431 break; 432 default: 433 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 434 g_mirror_get_diskname(disk), 435 g_mirror_disk_state2str(disk->d_state))); 436 } 437} 438 439static void 440g_mirror_destroy_device(struct g_mirror_softc *sc) 441{ 442 struct g_mirror_disk *disk; 443 struct g_mirror_event *ep; 444 struct g_geom *gp; 445 struct g_consumer *cp, *tmpcp; 446 447 g_topology_assert(); 448 449 gp = sc->sc_geom; 450 if (sc->sc_provider != NULL) 451 g_mirror_destroy_provider(sc); 452 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 453 disk = LIST_FIRST(&sc->sc_disks)) { 454 g_mirror_destroy_disk(disk); 455 } 456 while ((ep = g_mirror_event_get(sc)) != NULL) { 457 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 458 g_mirror_event_free(ep); 459 else { 460 ep->e_error = ECANCELED; 461 ep->e_flags |= G_MIRROR_EVENT_DONE; 462 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 463 mtx_lock(&sc->sc_events_mtx); 464 wakeup(ep); 465 mtx_unlock(&sc->sc_events_mtx); 466 } 467 } 468 callout_drain(&sc->sc_callout); 469 gp->softc = NULL; 470 uma_zdestroy(sc->sc_sync.ds_zone); 471 472 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 473 g_mirror_disconnect_consumer(sc, cp); 474 } 475 sc->sc_sync.ds_geom->softc = NULL; 476 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 477 mtx_destroy(&sc->sc_queue_mtx); 478 mtx_destroy(&sc->sc_events_mtx); 479 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 480 g_wither_geom(gp, ENXIO); 481} 482 483static void 484g_mirror_orphan(struct g_consumer *cp) 485{ 486 struct g_mirror_disk *disk; 487 488 g_topology_assert(); 489 490 disk = cp->private; 491 if (disk == NULL) 492 return; 493 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 494 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 495 G_MIRROR_EVENT_DONTWAIT); 496} 497 498static void 499g_mirror_spoiled(struct g_consumer *cp) 500{ 501 struct g_mirror_disk *disk; 502 503 g_topology_assert(); 504 505 disk = cp->private; 506 if (disk == NULL) 507 return; 508 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 509 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 510 G_MIRROR_EVENT_DONTWAIT); 511} 512 513/* 514 * Function should return the next active disk on the list. 515 * It is possible that it will be the same disk as given. 516 * If there are no active disks on list, NULL is returned. 517 */ 518static __inline struct g_mirror_disk * 519g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 520{ 521 struct g_mirror_disk *dp; 522 523 for (dp = LIST_NEXT(disk, d_next); dp != disk; 524 dp = LIST_NEXT(dp, d_next)) { 525 if (dp == NULL) 526 dp = LIST_FIRST(&sc->sc_disks); 527 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 528 break; 529 } 530 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 531 return (NULL); 532 return (dp); 533} 534 535static struct g_mirror_disk * 536g_mirror_get_disk(struct g_mirror_softc *sc) 537{ 538 struct g_mirror_disk *disk; 539 540 if (sc->sc_hint == NULL) { 541 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 542 if (sc->sc_hint == NULL) 543 return (NULL); 544 } 545 disk = sc->sc_hint; 546 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 547 disk = g_mirror_find_next(sc, disk); 548 if (disk == NULL) 549 return (NULL); 550 } 551 sc->sc_hint = g_mirror_find_next(sc, disk); 552 return (disk); 553} 554 555static int 556g_mirror_clear_metadata(struct g_mirror_disk *disk) 557{ 558 struct g_mirror_softc *sc; 559 struct g_consumer *cp; 560 off_t offset, length; 561 u_char *sector; 562 int close = 0, error = 0; 563 564 g_topology_assert(); 565 566 sc = disk->d_softc; 567 cp = disk->d_consumer; 568 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 569 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 570 length = cp->provider->sectorsize; 571 offset = cp->provider->mediasize - length; 572 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 573 /* 574 * Open consumer if it wasn't opened and remember to close it. 575 */ 576 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 577 error = g_access(cp, 0, 1, 1); 578 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 579 cp->provider->name, 0, 1, 1, error); 580 if (error == 0) 581 close = 1; 582#ifdef INVARIANTS 583 } else { 584 KASSERT(cp->acw > 0 && cp->ace > 0, 585 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 586 cp->acr, cp->acw, cp->ace)); 587#endif 588 } 589 if (error == 0) { 590 g_topology_unlock(); 591 error = g_write_data(cp, offset, sector, length); 592 g_topology_lock(); 593 } 594 free(sector, M_MIRROR); 595 if (close) { 596 g_access(cp, 0, -1, -1); 597 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 598 cp->provider->name, 0, -1, -1, 0); 599 } 600 if (error != 0) { 601 G_MIRROR_DEBUG(0, "Cannot clear metadata on disk %s.", 602 g_mirror_get_diskname(disk)); 603 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 604 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 605 G_MIRROR_EVENT_DONTWAIT); 606 return (error); 607 } 608 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 609 g_mirror_get_diskname(disk)); 610 return (0); 611} 612 613void 614g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 615 struct g_mirror_metadata *md) 616{ 617 618 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 619 md->md_version = G_MIRROR_VERSION; 620 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 621 md->md_mid = sc->sc_id; 622 md->md_all = sc->sc_ndisks; 623 md->md_slice = sc->sc_slice; 624 md->md_balance = sc->sc_balance; 625 md->md_mediasize = sc->sc_mediasize; 626 md->md_sectorsize = sc->sc_sectorsize; 627 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 628 bzero(md->md_provider, sizeof(md->md_provider)); 629 if (disk == NULL) { 630 md->md_did = arc4random(); 631 md->md_priority = 0; 632 md->md_syncid = 0; 633 md->md_dflags = 0; 634 md->md_sync_offset = 0; 635 } else { 636 md->md_did = disk->d_id; 637 md->md_priority = disk->d_priority; 638 md->md_syncid = disk->d_sync.ds_syncid; 639 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 640 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 641 md->md_sync_offset = disk->d_sync.ds_offset_done; 642 else 643 md->md_sync_offset = 0; 644 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 645 strlcpy(md->md_provider, 646 disk->d_consumer->provider->name, 647 sizeof(md->md_provider)); 648 } 649 } 650} 651 652void 653g_mirror_update_metadata(struct g_mirror_disk *disk) 654{ 655 struct g_mirror_softc *sc; 656 struct g_mirror_metadata md; 657 struct g_consumer *cp; 658 off_t offset, length; 659 u_char *sector; 660 int close = 0, error = 0; 661 662 g_topology_assert(); 663 664 sc = disk->d_softc; 665 cp = disk->d_consumer; 666 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 667 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 668 length = cp->provider->sectorsize; 669 offset = cp->provider->mediasize - length; 670 sector = malloc((size_t)length, M_MIRROR, M_WAITOK); 671 /* 672 * Open consumer if it wasn't opened and remember to close it. 673 */ 674 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 675 error = g_access(cp, 0, 1, 1); 676 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 677 cp->provider->name, 0, 1, 1, error); 678 if (error == 0) 679 close = 1; 680#ifdef INVARIANTS 681 } else { 682 KASSERT(cp->acw > 0 && cp->ace > 0, 683 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 684 cp->acr, cp->acw, cp->ace)); 685#endif 686 } 687 if (error == 0) { 688 g_mirror_fill_metadata(sc, disk, &md); 689 mirror_metadata_encode(&md, sector); 690 g_topology_unlock(); 691 error = g_write_data(cp, offset, sector, length); 692 g_topology_lock(); 693 } 694 free(sector, M_MIRROR); 695 if (close) { 696 g_access(cp, 0, -1, -1); 697 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 698 cp->provider->name, 0, -1, -1, 0); 699 } 700 if (error != 0) { 701 G_MIRROR_DEBUG(0, 702 "Cannot update metadata on disk %s (error=%d).", 703 g_mirror_get_diskname(disk), error); 704 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 705 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 706 G_MIRROR_EVENT_DONTWAIT); 707 return; 708 } 709 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 710 g_mirror_get_diskname(disk)); 711} 712 713static void 714g_mirror_bump_syncid(struct g_mirror_softc *sc) 715{ 716 struct g_mirror_disk *disk; 717 718 g_topology_assert(); 719 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 720 ("%s called with no active disks (device=%s).", __func__, 721 sc->sc_name)); 722 723 sc->sc_syncid++; 724 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 725 sc->sc_syncid); 726 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 727 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 728 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 729 disk->d_sync.ds_syncid = sc->sc_syncid; 730 g_mirror_update_metadata(disk); 731 } 732 } 733} 734 735static __inline int 736bintime_cmp(struct bintime *bt1, struct bintime *bt2) 737{ 738 739 if (bt1->sec < bt2->sec) 740 return (-1); 741 else if (bt1->sec > bt2->sec) 742 return (1); 743 if (bt1->frac < bt2->frac) 744 return (-1); 745 else if (bt1->frac > bt2->frac) 746 return (1); 747 return (0); 748} 749 750static void 751g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 752{ 753 754 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 755 return; 756 binuptime(&disk->d_delay); 757 bintime_sub(&disk->d_delay, &bp->bio_t0); 758} 759 760static void 761g_mirror_done(struct bio *bp) 762{ 763 struct g_mirror_softc *sc; 764 765 sc = bp->bio_from->geom->softc; 766 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 767 mtx_lock(&sc->sc_queue_mtx); 768 bioq_disksort(&sc->sc_queue, bp); 769 wakeup(sc); 770 mtx_unlock(&sc->sc_queue_mtx); 771} 772 773static void 774g_mirror_regular_request(struct bio *bp) 775{ 776 struct g_mirror_softc *sc; 777 struct g_mirror_disk *disk; 778 struct bio *pbp; 779 780 g_topology_assert_not(); 781 782 pbp = bp->bio_parent; 783 sc = pbp->bio_to->geom->softc; 784 disk = bp->bio_from->private; 785 if (disk == NULL) { 786 g_topology_lock(); 787 g_mirror_kill_consumer(sc, bp->bio_from); 788 g_topology_unlock(); 789 } else { 790 g_mirror_update_delay(disk, bp); 791 } 792 793 pbp->bio_inbed++; 794 KASSERT(pbp->bio_inbed <= pbp->bio_children, 795 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 796 pbp->bio_children)); 797 if (bp->bio_error == 0 && pbp->bio_error == 0) { 798 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 799 g_destroy_bio(bp); 800 if (pbp->bio_children == pbp->bio_inbed) { 801 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 802 pbp->bio_completed = pbp->bio_length; 803 g_io_deliver(pbp, pbp->bio_error); 804 } 805 return; 806 } else if (bp->bio_error != 0) { 807 if (pbp->bio_error == 0) 808 pbp->bio_error = bp->bio_error; 809 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 810 bp->bio_error); 811 if (disk != NULL) { 812 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 813 g_mirror_event_send(disk, 814 G_MIRROR_DISK_STATE_DISCONNECTED, 815 G_MIRROR_EVENT_DONTWAIT); 816 } 817 switch (pbp->bio_cmd) { 818 case BIO_DELETE: 819 case BIO_WRITE: 820 pbp->bio_inbed--; 821 pbp->bio_children--; 822 break; 823 } 824 } 825 g_destroy_bio(bp); 826 827 switch (pbp->bio_cmd) { 828 case BIO_READ: 829 if (pbp->bio_children == pbp->bio_inbed) { 830 pbp->bio_error = 0; 831 mtx_lock(&sc->sc_queue_mtx); 832 bioq_disksort(&sc->sc_queue, pbp); 833 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 834 wakeup(sc); 835 mtx_unlock(&sc->sc_queue_mtx); 836 } 837 break; 838 case BIO_DELETE: 839 case BIO_WRITE: 840 if (pbp->bio_children == 0) { 841 /* 842 * All requests failed. 843 */ 844 } else if (pbp->bio_inbed < pbp->bio_children) { 845 /* Do nothing. */ 846 break; 847 } else if (pbp->bio_children == pbp->bio_inbed) { 848 /* Some requests succeeded. */ 849 pbp->bio_error = 0; 850 pbp->bio_completed = pbp->bio_length; 851 } 852 g_io_deliver(pbp, pbp->bio_error); 853 break; 854 default: 855 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 856 break; 857 } 858} 859 860static void 861g_mirror_sync_done(struct bio *bp) 862{ 863 struct g_mirror_softc *sc; 864 865 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 866 sc = bp->bio_from->geom->softc; 867 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 868 mtx_lock(&sc->sc_queue_mtx); 869 bioq_disksort(&sc->sc_queue, bp); 870 wakeup(sc); 871 mtx_unlock(&sc->sc_queue_mtx); 872} 873 874static void 875g_mirror_start(struct bio *bp) 876{ 877 struct g_mirror_softc *sc; 878 879 sc = bp->bio_to->geom->softc; 880 /* 881 * If sc == NULL or there are no valid disks, provider's error 882 * should be set and g_mirror_start() should not be called at all. 883 */ 884 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 885 ("Provider's error should be set (error=%d)(mirror=%s).", 886 bp->bio_to->error, bp->bio_to->name)); 887 G_MIRROR_LOGREQ(3, bp, "Request received."); 888 889 switch (bp->bio_cmd) { 890 case BIO_READ: 891 case BIO_WRITE: 892 case BIO_DELETE: 893 break; 894 case BIO_GETATTR: 895 default: 896 g_io_deliver(bp, EOPNOTSUPP); 897 return; 898 } 899 mtx_lock(&sc->sc_queue_mtx); 900 bioq_disksort(&sc->sc_queue, bp); 901 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 902 wakeup(sc); 903 mtx_unlock(&sc->sc_queue_mtx); 904} 905 906/* 907 * Send one synchronization request. 908 */ 909static void 910g_mirror_sync_one(struct g_mirror_disk *disk) 911{ 912 struct g_mirror_softc *sc; 913 struct bio *bp; 914 915 sc = disk->d_softc; 916 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 917 ("Disk %s is not marked for synchronization.", 918 g_mirror_get_diskname(disk))); 919 920 bp = g_new_bio(); 921 if (bp == NULL) 922 return; 923 bp->bio_parent = NULL; 924 bp->bio_cmd = BIO_READ; 925 bp->bio_offset = disk->d_sync.ds_offset; 926 bp->bio_length = MIN(sc->sc_sync.ds_block, 927 sc->sc_mediasize - bp->bio_offset); 928 bp->bio_cflags = 0; 929 bp->bio_done = g_mirror_sync_done; 930 bp->bio_data = uma_zalloc(sc->sc_sync.ds_zone, M_NOWAIT | M_ZERO); 931 if (bp->bio_data == NULL) { 932 g_destroy_bio(bp); 933 return; 934 } 935 disk->d_sync.ds_offset += bp->bio_length; 936 bp->bio_to = sc->sc_provider; 937 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 938 g_io_request(bp, disk->d_sync.ds_consumer); 939} 940 941static void 942g_mirror_sync_request(struct bio *bp) 943{ 944 struct g_mirror_softc *sc; 945 struct g_mirror_disk *disk; 946 947 sc = bp->bio_from->geom->softc; 948 disk = bp->bio_from->private; 949 if (disk == NULL) { 950 g_topology_lock(); 951 g_mirror_kill_consumer(sc, bp->bio_from); 952 g_topology_unlock(); 953 uma_zfree(sc->sc_sync.ds_zone, bp->bio_data); 954 g_destroy_bio(bp); 955 return; 956 } 957 958 /* 959 * Synchronization request. 960 */ 961 switch (bp->bio_cmd) { 962 case BIO_READ: 963 { 964 struct g_consumer *cp; 965 966 if (bp->bio_error != 0) { 967 G_MIRROR_LOGREQ(0, bp, 968 "Synchronization request failed (error=%d).", 969 bp->bio_error); 970 uma_zfree(sc->sc_sync.ds_zone, bp->bio_data); 971 g_destroy_bio(bp); 972 return; 973 } 974 bp->bio_cmd = BIO_WRITE; 975 bp->bio_cflags = 0; 976 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 977 cp = disk->d_consumer; 978 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 979 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 980 cp->acr, cp->acw, cp->ace)); 981 g_io_request(bp, cp); 982 return; 983 } 984 case BIO_WRITE: 985 uma_zfree(sc->sc_sync.ds_zone, bp->bio_data); 986 if (bp->bio_error != 0) { 987 G_MIRROR_LOGREQ(0, bp, 988 "Synchronization request failed (error=%d).", 989 bp->bio_error); 990 g_destroy_bio(bp); 991 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 992 g_mirror_event_send(disk, 993 G_MIRROR_DISK_STATE_DISCONNECTED, 994 G_MIRROR_EVENT_DONTWAIT); 995 return; 996 } 997 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 998 disk->d_sync.ds_offset_done = bp->bio_offset + bp->bio_length; 999 g_destroy_bio(bp); 1000 if (disk->d_sync.ds_offset_done == sc->sc_provider->mediasize) { 1001 /* 1002 * Disk up-to-date, activate it. 1003 */ 1004 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1005 G_MIRROR_EVENT_DONTWAIT); 1006 return; 1007 } else if ((disk->d_sync.ds_offset_done % 1008 (sc->sc_sync.ds_block * 100)) == 0) { 1009 /* 1010 * Update offset_done on every 100 blocks. 1011 * XXX: This should be configurable. 1012 */ 1013 g_topology_lock(); 1014 g_mirror_update_metadata(disk); 1015 g_topology_unlock(); 1016 } 1017 return; 1018 default: 1019 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1020 bp->bio_cmd, sc->sc_name)); 1021 break; 1022 } 1023} 1024 1025static void 1026g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1027{ 1028 struct g_mirror_disk *disk; 1029 struct g_consumer *cp; 1030 struct bio *cbp; 1031 1032 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1033 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1034 break; 1035 } 1036 if (disk == NULL) { 1037 if (bp->bio_error == 0) 1038 bp->bio_error = ENXIO; 1039 g_io_deliver(bp, bp->bio_error); 1040 return; 1041 } 1042 cbp = g_clone_bio(bp); 1043 if (cbp == NULL) { 1044 if (bp->bio_error == 0) 1045 bp->bio_error = ENOMEM; 1046 g_io_deliver(bp, bp->bio_error); 1047 return; 1048 } 1049 /* 1050 * Fill in the component buf structure. 1051 */ 1052 cp = disk->d_consumer; 1053 cbp->bio_done = g_mirror_done; 1054 cbp->bio_to = cp->provider; 1055 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1056 KASSERT(cp->acr > 0 && cp->ace > 0, 1057 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1058 cp->acw, cp->ace)); 1059 g_io_request(cbp, cp); 1060} 1061 1062static void 1063g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1064{ 1065 struct g_mirror_disk *disk; 1066 struct g_consumer *cp; 1067 struct bio *cbp; 1068 1069 disk = g_mirror_get_disk(sc); 1070 if (disk == NULL) { 1071 if (bp->bio_error == 0) 1072 bp->bio_error = ENXIO; 1073 g_io_deliver(bp, bp->bio_error); 1074 return; 1075 } 1076 cbp = g_clone_bio(bp); 1077 if (cbp == NULL) { 1078 if (bp->bio_error == 0) 1079 bp->bio_error = ENOMEM; 1080 g_io_deliver(bp, bp->bio_error); 1081 return; 1082 } 1083 /* 1084 * Fill in the component buf structure. 1085 */ 1086 cp = disk->d_consumer; 1087 cbp->bio_done = g_mirror_done; 1088 cbp->bio_to = cp->provider; 1089 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1090 KASSERT(cp->acr > 0 && cp->ace > 0, 1091 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1092 cp->acw, cp->ace)); 1093 g_io_request(cbp, cp); 1094} 1095 1096static void 1097g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1098{ 1099 struct g_mirror_disk *disk, *dp; 1100 struct g_consumer *cp; 1101 struct bio *cbp; 1102 struct bintime curtime; 1103 1104 binuptime(&curtime); 1105 /* 1106 * Find a disk which the smallest load. 1107 */ 1108 disk = NULL; 1109 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1110 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1111 continue; 1112 /* If disk wasn't used for more than 2 sec, use it. */ 1113 if (curtime.sec - dp->d_last_used.sec >= 2) { 1114 disk = dp; 1115 break; 1116 } 1117 if (disk == NULL || 1118 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1119 disk = dp; 1120 } 1121 } 1122 cbp = g_clone_bio(bp); 1123 if (cbp == NULL) { 1124 if (bp->bio_error == 0) 1125 bp->bio_error = ENOMEM; 1126 g_io_deliver(bp, bp->bio_error); 1127 return; 1128 } 1129 /* 1130 * Fill in the component buf structure. 1131 */ 1132 cp = disk->d_consumer; 1133 cbp->bio_done = g_mirror_done; 1134 cbp->bio_to = cp->provider; 1135 binuptime(&disk->d_last_used); 1136 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1137 KASSERT(cp->acr > 0 && cp->ace > 0, 1138 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1139 cp->acw, cp->ace)); 1140 g_io_request(cbp, cp); 1141} 1142 1143static void 1144g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1145{ 1146 struct bio_queue_head queue; 1147 struct g_mirror_disk *disk; 1148 struct g_consumer *cp; 1149 struct bio *cbp; 1150 off_t left, mod, offset, slice; 1151 u_char *data; 1152 u_int ndisks; 1153 1154 if (bp->bio_length <= sc->sc_slice) { 1155 g_mirror_request_round_robin(sc, bp); 1156 return; 1157 } 1158 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1159 slice = bp->bio_length / ndisks; 1160 mod = slice % sc->sc_provider->sectorsize; 1161 if (mod != 0) 1162 slice += sc->sc_provider->sectorsize - mod; 1163 /* 1164 * Allocate all bios before sending any request, so we can 1165 * return ENOMEM in nice and clean way. 1166 */ 1167 left = bp->bio_length; 1168 offset = bp->bio_offset; 1169 data = bp->bio_data; 1170 bioq_init(&queue); 1171 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1172 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1173 continue; 1174 cbp = g_clone_bio(bp); 1175 if (cbp == NULL) { 1176 for (cbp = bioq_first(&queue); cbp != NULL; 1177 cbp = bioq_first(&queue)) { 1178 bioq_remove(&queue, cbp); 1179 g_destroy_bio(cbp); 1180 } 1181 if (bp->bio_error == 0) 1182 bp->bio_error = ENOMEM; 1183 g_io_deliver(bp, bp->bio_error); 1184 return; 1185 } 1186 bioq_insert_tail(&queue, cbp); 1187 cbp->bio_done = g_mirror_done; 1188 cbp->bio_caller1 = disk; 1189 cbp->bio_to = disk->d_consumer->provider; 1190 cbp->bio_offset = offset; 1191 cbp->bio_data = data; 1192 cbp->bio_length = MIN(left, slice); 1193 left -= cbp->bio_length; 1194 if (left == 0) 1195 break; 1196 offset += cbp->bio_length; 1197 data += cbp->bio_length; 1198 } 1199 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1200 bioq_remove(&queue, cbp); 1201 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1202 disk = cbp->bio_caller1; 1203 cbp->bio_caller1 = NULL; 1204 cp = disk->d_consumer; 1205 KASSERT(cp->acr > 0 && cp->ace > 0, 1206 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1207 cp->acr, cp->acw, cp->ace)); 1208 g_io_request(cbp, disk->d_consumer); 1209 } 1210} 1211 1212static void 1213g_mirror_register_request(struct bio *bp) 1214{ 1215 struct g_mirror_softc *sc; 1216 1217 sc = bp->bio_to->geom->softc; 1218 switch (bp->bio_cmd) { 1219 case BIO_READ: 1220 switch (sc->sc_balance) { 1221 case G_MIRROR_BALANCE_LOAD: 1222 g_mirror_request_load(sc, bp); 1223 break; 1224 case G_MIRROR_BALANCE_PREFER: 1225 g_mirror_request_prefer(sc, bp); 1226 break; 1227 case G_MIRROR_BALANCE_ROUND_ROBIN: 1228 g_mirror_request_round_robin(sc, bp); 1229 break; 1230 case G_MIRROR_BALANCE_SPLIT: 1231 g_mirror_request_split(sc, bp); 1232 break; 1233 } 1234 return; 1235 case BIO_WRITE: 1236 case BIO_DELETE: 1237 { 1238 struct g_mirror_disk *disk; 1239 struct bio_queue_head queue; 1240 struct g_consumer *cp; 1241 struct bio *cbp; 1242 1243 /* 1244 * Allocate all bios before sending any request, so we can 1245 * return ENOMEM in nice and clean way. 1246 */ 1247 bioq_init(&queue); 1248 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1249 switch (disk->d_state) { 1250 case G_MIRROR_DISK_STATE_ACTIVE: 1251 break; 1252 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1253 if (bp->bio_offset >= disk->d_sync.ds_offset) 1254 continue; 1255 break; 1256 default: 1257 continue; 1258 } 1259 cbp = g_clone_bio(bp); 1260 if (cbp == NULL) { 1261 for (cbp = bioq_first(&queue); cbp != NULL; 1262 cbp = bioq_first(&queue)) { 1263 bioq_remove(&queue, cbp); 1264 g_destroy_bio(cbp); 1265 } 1266 if (bp->bio_error == 0) 1267 bp->bio_error = ENOMEM; 1268 g_io_deliver(bp, bp->bio_error); 1269 return; 1270 } 1271 bioq_insert_tail(&queue, cbp); 1272 } 1273 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1274 switch (disk->d_state) { 1275 case G_MIRROR_DISK_STATE_ACTIVE: 1276 break; 1277 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1278 if (bp->bio_offset >= disk->d_sync.ds_offset) 1279 continue; 1280 break; 1281 default: 1282 continue; 1283 } 1284 cbp = bioq_first(&queue); 1285 KASSERT(cbp != NULL, ("NULL cbp! (device %s).", 1286 sc->sc_name)); 1287 bioq_remove(&queue, cbp); 1288 cp = disk->d_consumer; 1289 cbp->bio_done = g_mirror_done; 1290 cbp->bio_to = cp->provider; 1291 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1292 KASSERT(cp->acw > 0 && cp->ace > 0, 1293 ("Consumer %s not opened (r%dw%de%d).", 1294 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1295 g_io_request(cbp, cp); 1296 } 1297 /* 1298 * Bump syncid on first write. 1299 */ 1300 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) { 1301 sc->sc_bump_syncid = 0; 1302 g_topology_lock(); 1303 g_mirror_bump_syncid(sc); 1304 g_topology_unlock(); 1305 } 1306 return; 1307 } 1308 default: 1309 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1310 bp->bio_cmd, sc->sc_name)); 1311 break; 1312 } 1313} 1314 1315static int 1316g_mirror_can_destroy(struct g_mirror_softc *sc) 1317{ 1318 struct g_geom *gp; 1319 struct g_consumer *cp; 1320 1321 g_topology_assert(); 1322 gp = sc->sc_geom; 1323 LIST_FOREACH(cp, &gp->consumer, consumer) { 1324 if (g_mirror_is_busy(sc, cp)) 1325 return (0); 1326 } 1327 gp = sc->sc_sync.ds_geom; 1328 LIST_FOREACH(cp, &gp->consumer, consumer) { 1329 if (g_mirror_is_busy(sc, cp)) 1330 return (0); 1331 } 1332 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1333 sc->sc_name); 1334 return (1); 1335} 1336 1337static int 1338g_mirror_try_destroy(struct g_mirror_softc *sc) 1339{ 1340 1341 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1342 g_topology_lock(); 1343 if (!g_mirror_can_destroy(sc)) { 1344 g_topology_unlock(); 1345 return (0); 1346 } 1347 g_topology_unlock(); 1348 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1349 &sc->sc_worker); 1350 wakeup(&sc->sc_worker); 1351 sc->sc_worker = NULL; 1352 } else { 1353 g_topology_lock(); 1354 if (!g_mirror_can_destroy(sc)) { 1355 g_topology_unlock(); 1356 return (0); 1357 } 1358 g_mirror_destroy_device(sc); 1359 g_topology_unlock(); 1360 free(sc, M_MIRROR); 1361 } 1362 return (1); 1363} 1364 1365/* 1366 * Worker thread. 1367 */ 1368static void 1369g_mirror_worker(void *arg) 1370{ 1371 struct g_mirror_softc *sc; 1372 struct g_mirror_disk *disk; 1373 struct g_mirror_event *ep; 1374 struct bio *bp; 1375 u_int nreqs; 1376 1377 sc = arg; 1378 curthread->td_base_pri = PRIBIO; 1379 1380 nreqs = 0; 1381 for (;;) { 1382 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1383 /* 1384 * First take a look at events. 1385 * This is important to handle events before any I/O requests. 1386 */ 1387 ep = g_mirror_event_get(sc); 1388 if (ep != NULL) { 1389 g_topology_lock(); 1390 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1391 /* Update only device status. */ 1392 G_MIRROR_DEBUG(3, 1393 "Running event for device %s.", 1394 sc->sc_name); 1395 ep->e_error = 0; 1396 g_mirror_update_device(sc, 1); 1397 } else { 1398 /* Update disk status. */ 1399 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1400 g_mirror_get_diskname(ep->e_disk)); 1401 ep->e_error = g_mirror_update_disk(ep->e_disk, 1402 ep->e_state); 1403 if (ep->e_error == 0) 1404 g_mirror_update_device(sc, 0); 1405 } 1406 g_topology_unlock(); 1407 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1408 KASSERT(ep->e_error == 0, 1409 ("Error cannot be handled.")); 1410 g_mirror_event_free(ep); 1411 } else { 1412 ep->e_flags |= G_MIRROR_EVENT_DONE; 1413 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1414 ep); 1415 mtx_lock(&sc->sc_events_mtx); 1416 wakeup(ep); 1417 mtx_unlock(&sc->sc_events_mtx); 1418 } 1419 if ((sc->sc_flags & 1420 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1421 if (g_mirror_try_destroy(sc)) 1422 kthread_exit(0); 1423 } 1424 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1425 continue; 1426 } 1427 /* 1428 * Now I/O requests. 1429 */ 1430 /* Get first request from the queue. */ 1431 mtx_lock(&sc->sc_queue_mtx); 1432 bp = bioq_first(&sc->sc_queue); 1433 if (bp == NULL) { 1434 if ((sc->sc_flags & 1435 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1436 mtx_unlock(&sc->sc_queue_mtx); 1437 if (g_mirror_try_destroy(sc)) 1438 kthread_exit(0); 1439 mtx_lock(&sc->sc_queue_mtx); 1440 } 1441 } 1442 if (sc->sc_sync.ds_ndisks > 0 && 1443 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1444 mtx_unlock(&sc->sc_queue_mtx); 1445 /* 1446 * It is time for synchronization... 1447 */ 1448 nreqs = 0; 1449 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1450 if (disk->d_state != 1451 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1452 continue; 1453 } 1454 if (disk->d_sync.ds_offset >= 1455 sc->sc_provider->mediasize) { 1456 continue; 1457 } 1458 if (disk->d_sync.ds_offset > 1459 disk->d_sync.ds_offset_done) { 1460 continue; 1461 } 1462 g_mirror_sync_one(disk); 1463 } 1464 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1465 goto sleep; 1466 } 1467 if (bp == NULL) { 1468 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 0); 1469 G_MIRROR_DEBUG(5, "%s: I'm here 3.", __func__); 1470 continue; 1471 } 1472 nreqs++; 1473 bioq_remove(&sc->sc_queue, bp); 1474 mtx_unlock(&sc->sc_queue_mtx); 1475 1476 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1477 g_mirror_regular_request(bp); 1478 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1479 u_int timeout, sps; 1480 1481 g_mirror_sync_request(bp); 1482sleep: 1483 sps = atomic_load_acq_int(&g_mirror_syncs_per_sec); 1484 if (sps == 0) { 1485 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1486 continue; 1487 } 1488 mtx_lock(&sc->sc_queue_mtx); 1489 if (bioq_first(&sc->sc_queue) != NULL) { 1490 mtx_unlock(&sc->sc_queue_mtx); 1491 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1492 continue; 1493 } 1494 timeout = hz / sps; 1495 if (timeout == 0) 1496 timeout = 1; 1497 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w2", 1498 timeout); 1499 } else { 1500 g_mirror_register_request(bp); 1501 } 1502 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1503 } 1504} 1505 1506/* 1507 * Open disk's consumer if needed. 1508 */ 1509static void 1510g_mirror_update_access(struct g_mirror_disk *disk) 1511{ 1512 struct g_provider *pp; 1513 struct g_consumer *cp; 1514 int acr, acw, ace, cpw, error; 1515 1516 g_topology_assert(); 1517 1518 cp = disk->d_consumer; 1519 pp = disk->d_softc->sc_provider; 1520 if (pp == NULL) { 1521 acr = -cp->acr; 1522 acw = -cp->acw; 1523 ace = -cp->ace; 1524 } else { 1525 acr = pp->acr - cp->acr; 1526 acw = pp->acw - cp->acw; 1527 ace = pp->ace - cp->ace; 1528 /* Grab an extra "exclusive" bit. */ 1529 if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) 1530 ace++; 1531 } 1532 if (acr == 0 && acw == 0 && ace == 0) 1533 return; 1534 cpw = cp->acw; 1535 error = g_access(cp, acr, acw, ace); 1536 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, acr, 1537 acw, ace, error); 1538 if (error != 0) { 1539 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1540 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1541 G_MIRROR_EVENT_DONTWAIT); 1542 return; 1543 } 1544 if (cpw == 0 && cp->acw > 0) { 1545 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1546 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1547 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1548 } else if (cpw > 0 && cp->acw == 0) { 1549 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1550 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1551 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1552 } 1553} 1554 1555static void 1556g_mirror_sync_start(struct g_mirror_disk *disk) 1557{ 1558 struct g_mirror_softc *sc; 1559 struct g_consumer *cp; 1560 int error; 1561 1562 g_topology_assert(); 1563 1564 sc = disk->d_softc; 1565 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1566 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1567 sc->sc_state)); 1568 cp = disk->d_consumer; 1569 KASSERT(cp->acr == 0 && cp->acw == 0 && cp->ace == 0, 1570 ("Consumer %s already opened.", cp->provider->name)); 1571 1572 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1573 g_mirror_get_diskname(disk)); 1574 error = g_access(cp, 0, 1, 1); 1575 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, 1, 1576 1, error); 1577 if (error != 0) { 1578 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1579 G_MIRROR_EVENT_DONTWAIT); 1580 return; 1581 } 1582 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1583 KASSERT(disk->d_sync.ds_consumer == NULL, 1584 ("Sync consumer already exists (device=%s, disk=%s).", 1585 sc->sc_name, g_mirror_get_diskname(disk))); 1586 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1587 disk->d_sync.ds_consumer->private = disk; 1588 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1589 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1590 disk->d_softc->sc_name, error)); 1591 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1592 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1593 disk->d_softc->sc_name, error)); 1594 sc->sc_sync.ds_ndisks++; 1595} 1596 1597/* 1598 * Stop synchronization process. 1599 * type: 0 - synchronization finished 1600 * 1 - synchronization stopped 1601 */ 1602static void 1603g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1604{ 1605 struct g_consumer *cp; 1606 1607 g_topology_assert(); 1608 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1609 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1610 g_mirror_disk_state2str(disk->d_state))); 1611 if (disk->d_sync.ds_consumer == NULL) 1612 return; 1613 1614 if (type == 0) { 1615 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1616 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1617 } else /* if (type == 1) */ { 1618 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1619 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1620 } 1621 cp = disk->d_sync.ds_consumer; 1622 g_access(cp, -1, 0, 0); 1623 g_mirror_kill_consumer(disk->d_softc, cp); 1624 disk->d_sync.ds_consumer = NULL; 1625 disk->d_softc->sc_sync.ds_ndisks--; 1626 cp = disk->d_consumer; 1627 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 1628 ("Consumer %s not opened.", cp->provider->name)); 1629 g_access(cp, 0, -1, -1); 1630 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, -1, 1631 -1, 0); 1632 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1633} 1634 1635static void 1636g_mirror_launch_provider(struct g_mirror_softc *sc) 1637{ 1638 struct g_mirror_disk *disk; 1639 struct g_provider *pp; 1640 1641 g_topology_assert(); 1642 1643 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1644 pp->mediasize = sc->sc_mediasize; 1645 pp->sectorsize = sc->sc_sectorsize; 1646 sc->sc_provider = pp; 1647 g_error_provider(pp, 0); 1648 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1649 pp->name); 1650 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1651 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1652 g_mirror_sync_start(disk); 1653 } 1654} 1655 1656static void 1657g_mirror_destroy_provider(struct g_mirror_softc *sc) 1658{ 1659 struct g_mirror_disk *disk; 1660 struct bio *bp; 1661 1662 g_topology_assert(); 1663 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1664 sc->sc_name)); 1665 1666 g_error_provider(sc->sc_provider, ENXIO); 1667 mtx_lock(&sc->sc_queue_mtx); 1668 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1669 bioq_remove(&sc->sc_queue, bp); 1670 g_io_deliver(bp, ENXIO); 1671 } 1672 mtx_unlock(&sc->sc_queue_mtx); 1673 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1674 sc->sc_provider->name); 1675 sc->sc_provider->flags |= G_PF_WITHER; 1676 g_orphan_provider(sc->sc_provider, ENXIO); 1677 sc->sc_provider = NULL; 1678 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1679 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1680 g_mirror_sync_stop(disk, 1); 1681 } 1682} 1683 1684static void 1685g_mirror_go(void *arg) 1686{ 1687 struct g_mirror_softc *sc; 1688 1689 sc = arg; 1690 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1691 g_mirror_event_send(sc, 0, 1692 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1693} 1694 1695static u_int 1696g_mirror_determine_state(struct g_mirror_disk *disk) 1697{ 1698 struct g_mirror_softc *sc; 1699 u_int state; 1700 1701 sc = disk->d_softc; 1702 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1703 if ((disk->d_flags & 1704 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1705 /* Disk does not need synchronization. */ 1706 state = G_MIRROR_DISK_STATE_ACTIVE; 1707 } else { 1708 if ((sc->sc_flags & 1709 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1710 (disk->d_flags & 1711 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1712 /* 1713 * We can start synchronization from 1714 * the stored offset. 1715 */ 1716 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1717 } else { 1718 state = G_MIRROR_DISK_STATE_STALE; 1719 } 1720 } 1721 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1722 /* 1723 * Reset all synchronization data for this disk, 1724 * because if it even was synchronized, it was 1725 * synchronized to disks with different syncid. 1726 */ 1727 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1728 disk->d_sync.ds_offset = 0; 1729 disk->d_sync.ds_offset_done = 0; 1730 disk->d_sync.ds_syncid = sc->sc_syncid; 1731 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1732 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1733 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1734 } else { 1735 state = G_MIRROR_DISK_STATE_STALE; 1736 } 1737 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1738 /* 1739 * Not good, NOT GOOD! 1740 * It means that mirror was started on stale disks 1741 * and more fresh disk just arrive. 1742 * If there were writes, mirror is fucked up, sorry. 1743 * I think the best choice here is don't touch 1744 * this disk and inform the user laudly. 1745 */ 1746 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1747 "disk (%s) arrives!! It will not be connected to the " 1748 "running device.", sc->sc_name, 1749 g_mirror_get_diskname(disk)); 1750 g_mirror_destroy_disk(disk); 1751 state = G_MIRROR_DISK_STATE_NONE; 1752 /* Return immediately, because disk was destroyed. */ 1753 return (state); 1754 } 1755 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1756 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1757 return (state); 1758} 1759 1760/* 1761 * Update device state. 1762 */ 1763static void 1764g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1765{ 1766 struct g_mirror_disk *disk; 1767 u_int state; 1768 1769 g_topology_assert(); 1770 1771 switch (sc->sc_state) { 1772 case G_MIRROR_DEVICE_STATE_STARTING: 1773 { 1774 struct g_mirror_disk *pdisk; 1775 u_int dirty, ndisks, syncid; 1776 1777 KASSERT(sc->sc_provider == NULL, 1778 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1779 /* 1780 * Are we ready? We are, if all disks are connected or 1781 * if we have any disks and 'force' is true. 1782 */ 1783 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1784 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1785 ; 1786 } else if (g_mirror_ndisks(sc, -1) == 0) { 1787 /* 1788 * Disks went down in starting phase, so destroy 1789 * device. 1790 */ 1791 callout_drain(&sc->sc_callout); 1792 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1793 return; 1794 } else { 1795 return; 1796 } 1797 1798 /* 1799 * Activate all disks with the biggest syncid. 1800 */ 1801 if (force) { 1802 /* 1803 * If 'force' is true, we have been called due to 1804 * timeout, so don't bother canceling timeout. 1805 */ 1806 ndisks = 0; 1807 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1808 if ((disk->d_flags & 1809 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1810 ndisks++; 1811 } 1812 } 1813 if (ndisks == 0) { 1814 /* No valid disks found, destroy device. */ 1815 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1816 return; 1817 } 1818 } else { 1819 /* Cancel timeout. */ 1820 callout_drain(&sc->sc_callout); 1821 } 1822 1823 /* 1824 * Find disk with the biggest syncid. 1825 */ 1826 syncid = 0; 1827 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1828 if (disk->d_sync.ds_syncid > syncid) 1829 syncid = disk->d_sync.ds_syncid; 1830 } 1831 1832 /* 1833 * Here we need to look for dirty disks and if all disks 1834 * with the biggest syncid are dirty, we have to choose 1835 * one with the biggest priority and rebuild the rest. 1836 */ 1837 /* 1838 * Find the number of dirty disks with the biggest syncid. 1839 * Find the number of disks with the biggest syncid. 1840 * While here, find a disk with the biggest priority. 1841 */ 1842 dirty = ndisks = 0; 1843 pdisk = NULL; 1844 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1845 if (disk->d_sync.ds_syncid != syncid) 1846 continue; 1847 if ((disk->d_flags & 1848 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1849 continue; 1850 } 1851 ndisks++; 1852 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1853 dirty++; 1854 if (pdisk == NULL || 1855 pdisk->d_priority < disk->d_priority) { 1856 pdisk = disk; 1857 } 1858 } 1859 } 1860 if (dirty == 0) { 1861 /* No dirty disks at all, great. */ 1862 } else if (dirty == ndisks) { 1863 /* 1864 * Force synchronization for all dirty disks except one 1865 * with the biggest priority. 1866 */ 1867 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1868 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1869 "master disk for synchronization.", 1870 g_mirror_get_diskname(pdisk), sc->sc_name); 1871 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1872 if (disk->d_sync.ds_syncid != syncid) 1873 continue; 1874 if ((disk->d_flags & 1875 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1876 continue; 1877 } 1878 KASSERT((disk->d_flags & 1879 G_MIRROR_DISK_FLAG_DIRTY) != 0, 1880 ("Disk %s isn't marked as dirty.", 1881 g_mirror_get_diskname(disk))); 1882 /* Skip the disk with the biggest priority. */ 1883 if (disk == pdisk) 1884 continue; 1885 disk->d_sync.ds_syncid = 0; 1886 } 1887 } else if (dirty < ndisks) { 1888 /* 1889 * Force synchronization for all dirty disks. 1890 * We have some non-dirty disks. 1891 */ 1892 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1893 if (disk->d_sync.ds_syncid != syncid) 1894 continue; 1895 if ((disk->d_flags & 1896 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1897 continue; 1898 } 1899 if ((disk->d_flags & 1900 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1901 continue; 1902 } 1903 disk->d_sync.ds_syncid = 0; 1904 } 1905 } 1906 1907 /* Reset hint. */ 1908 sc->sc_hint = NULL; 1909 sc->sc_syncid = syncid; 1910 if (force) { 1911 /* Remember to bump syncid on first write. */ 1912 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1913 } 1914 state = G_MIRROR_DEVICE_STATE_RUNNING; 1915 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 1916 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 1917 g_mirror_device_state2str(state)); 1918 sc->sc_state = state; 1919 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1920 state = g_mirror_determine_state(disk); 1921 g_mirror_event_send(disk, state, 1922 G_MIRROR_EVENT_DONTWAIT); 1923 if (state == G_MIRROR_DISK_STATE_STALE) { 1924 sc->sc_bump_syncid = 1925 G_MIRROR_BUMP_ON_FIRST_WRITE; 1926 } 1927 } 1928 break; 1929 } 1930 case G_MIRROR_DEVICE_STATE_RUNNING: 1931 /* 1932 * Bump syncid here, if we need to do it immediately. 1933 */ 1934 if (sc->sc_bump_syncid == G_MIRROR_BUMP_IMMEDIATELY) { 1935 sc->sc_bump_syncid = 0; 1936 g_mirror_bump_syncid(sc); 1937 } 1938 if (g_mirror_ndisks(sc, -1) == 0) { 1939 /* 1940 * No disks at all, we need to destroy device. 1941 */ 1942 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1943 } else if (g_mirror_ndisks(sc, 1944 G_MIRROR_DISK_STATE_ACTIVE) == 0 && 1945 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 1946 /* 1947 * No active disks, destroy provider. 1948 */ 1949 if (sc->sc_provider != NULL) 1950 g_mirror_destroy_provider(sc); 1951 } else if (g_mirror_ndisks(sc, 1952 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 1953 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 1954 /* 1955 * We have active disks, launch provider if it doesn't 1956 * exist. 1957 */ 1958 if (sc->sc_provider == NULL) 1959 g_mirror_launch_provider(sc); 1960 } 1961 break; 1962 default: 1963 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 1964 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 1965 break; 1966 } 1967} 1968 1969/* 1970 * Update disk state and device state if needed. 1971 */ 1972#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 1973 "Disk %s state changed from %s to %s (device %s).", \ 1974 g_mirror_get_diskname(disk), \ 1975 g_mirror_disk_state2str(disk->d_state), \ 1976 g_mirror_disk_state2str(state), sc->sc_name) 1977static int 1978g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 1979{ 1980 struct g_mirror_softc *sc; 1981 1982 g_topology_assert(); 1983 1984 sc = disk->d_softc; 1985again: 1986 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 1987 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 1988 g_mirror_disk_state2str(state)); 1989 switch (state) { 1990 case G_MIRROR_DISK_STATE_NEW: 1991 /* 1992 * Possible scenarios: 1993 * 1. New disk arrive. 1994 */ 1995 /* Previous state should be NONE. */ 1996 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 1997 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1998 g_mirror_disk_state2str(disk->d_state))); 1999 DISK_STATE_CHANGED(); 2000 2001 disk->d_state = state; 2002 if (LIST_EMPTY(&sc->sc_disks)) 2003 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2004 else { 2005 struct g_mirror_disk *dp; 2006 2007 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2008 if (disk->d_priority >= dp->d_priority) { 2009 LIST_INSERT_BEFORE(dp, disk, d_next); 2010 dp = NULL; 2011 break; 2012 } 2013 if (LIST_NEXT(dp, d_next) == NULL) 2014 break; 2015 } 2016 if (dp != NULL) 2017 LIST_INSERT_AFTER(dp, disk, d_next); 2018 } 2019 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2020 sc->sc_name, g_mirror_get_diskname(disk)); 2021 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2022 break; 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 state = g_mirror_determine_state(disk); 2029 if (state != G_MIRROR_DISK_STATE_NONE) 2030 goto again; 2031 break; 2032 case G_MIRROR_DISK_STATE_ACTIVE: 2033 /* 2034 * Possible scenarios: 2035 * 1. New disk does not need synchronization. 2036 * 2. Synchronization process finished successfully. 2037 */ 2038 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2039 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2040 g_mirror_device_state2str(sc->sc_state), 2041 g_mirror_get_diskname(disk), 2042 g_mirror_disk_state2str(disk->d_state))); 2043 /* Previous state should be NEW or SYNCHRONIZING. */ 2044 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2045 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2046 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2047 g_mirror_disk_state2str(disk->d_state))); 2048 DISK_STATE_CHANGED(); 2049 2050 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2051 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2052 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2053 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2054 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2055 g_mirror_sync_stop(disk, 0); 2056 } 2057 disk->d_state = state; 2058 disk->d_sync.ds_offset = 0; 2059 disk->d_sync.ds_offset_done = 0; 2060 g_mirror_update_access(disk); 2061 g_mirror_update_metadata(disk); 2062 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2063 sc->sc_name, g_mirror_get_diskname(disk)); 2064 break; 2065 case G_MIRROR_DISK_STATE_STALE: 2066 /* 2067 * Possible scenarios: 2068 * 1. Stale disk was connected. 2069 */ 2070 /* Previous state should be NEW. */ 2071 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2072 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2073 g_mirror_disk_state2str(disk->d_state))); 2074 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2075 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2076 g_mirror_device_state2str(sc->sc_state), 2077 g_mirror_get_diskname(disk), 2078 g_mirror_disk_state2str(disk->d_state))); 2079 /* 2080 * STALE state is only possible if device is marked 2081 * NOAUTOSYNC. 2082 */ 2083 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2084 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2085 g_mirror_device_state2str(sc->sc_state), 2086 g_mirror_get_diskname(disk), 2087 g_mirror_disk_state2str(disk->d_state))); 2088 DISK_STATE_CHANGED(); 2089 2090 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2091 disk->d_state = state; 2092 g_mirror_update_metadata(disk); 2093 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2094 sc->sc_name, g_mirror_get_diskname(disk)); 2095 break; 2096 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2097 /* 2098 * Possible scenarios: 2099 * 1. Disk which needs synchronization was connected. 2100 */ 2101 /* Previous state should be NEW. */ 2102 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2103 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2104 g_mirror_disk_state2str(disk->d_state))); 2105 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2106 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2107 g_mirror_device_state2str(sc->sc_state), 2108 g_mirror_get_diskname(disk), 2109 g_mirror_disk_state2str(disk->d_state))); 2110 DISK_STATE_CHANGED(); 2111 2112 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2113 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2114 disk->d_state = state; 2115 if (sc->sc_provider != NULL) { 2116 g_mirror_sync_start(disk); 2117 g_mirror_update_metadata(disk); 2118 } 2119 break; 2120 case G_MIRROR_DISK_STATE_DISCONNECTED: 2121 /* 2122 * Possible scenarios: 2123 * 1. Device wasn't running yet, but disk disappear. 2124 * 2. Disk was active and disapppear. 2125 * 3. Disk disappear during synchronization process. 2126 */ 2127 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2128 /* 2129 * Previous state should be ACTIVE, STALE or 2130 * SYNCHRONIZING. 2131 */ 2132 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2133 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2134 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2135 ("Wrong disk state (%s, %s).", 2136 g_mirror_get_diskname(disk), 2137 g_mirror_disk_state2str(disk->d_state))); 2138 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2139 /* Previous state should be NEW. */ 2140 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2141 ("Wrong disk state (%s, %s).", 2142 g_mirror_get_diskname(disk), 2143 g_mirror_disk_state2str(disk->d_state))); 2144 /* 2145 * Reset bumping syncid if disk disappeared in STARTING 2146 * state. 2147 */ 2148 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) 2149 sc->sc_bump_syncid = 0; 2150#ifdef INVARIANTS 2151 } else { 2152 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2153 sc->sc_name, 2154 g_mirror_device_state2str(sc->sc_state), 2155 g_mirror_get_diskname(disk), 2156 g_mirror_disk_state2str(disk->d_state))); 2157#endif 2158 } 2159 DISK_STATE_CHANGED(); 2160 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2161 sc->sc_name, g_mirror_get_diskname(disk)); 2162 2163 g_mirror_destroy_disk(disk); 2164 break; 2165 case G_MIRROR_DISK_STATE_DESTROY: 2166 { 2167 int error; 2168 2169 error = g_mirror_clear_metadata(disk); 2170 if (error != 0) 2171 return (error); 2172 DISK_STATE_CHANGED(); 2173 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2174 sc->sc_name, g_mirror_get_diskname(disk)); 2175 2176 g_mirror_destroy_disk(disk); 2177 sc->sc_ndisks--; 2178 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2179 g_mirror_update_metadata(disk); 2180 } 2181 break; 2182 } 2183 default: 2184 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2185 break; 2186 } 2187 return (0); 2188} 2189#undef DISK_STATE_CHANGED 2190 2191static int 2192g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2193{ 2194 struct g_provider *pp; 2195 u_char *buf; 2196 int error; 2197 2198 g_topology_assert(); 2199 2200 error = g_access(cp, 1, 0, 0); 2201 if (error != 0) 2202 return (error); 2203 pp = cp->provider; 2204 g_topology_unlock(); 2205 /* Metadata are stored on last sector. */ 2206 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2207 &error); 2208 g_topology_lock(); 2209 if (buf == NULL) { 2210 g_access(cp, -1, 0, 0); 2211 return (error); 2212 } 2213 if (error != 0) { 2214 g_access(cp, -1, 0, 0); 2215 g_free(buf); 2216 return (error); 2217 } 2218 error = g_access(cp, -1, 0, 0); 2219 KASSERT(error == 0, ("Cannot decrease access count for %s.", pp->name)); 2220 2221 /* Decode metadata. */ 2222 error = mirror_metadata_decode(buf, md); 2223 g_free(buf); 2224 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2225 return (EINVAL); 2226 if (error != 0) { 2227 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2228 cp->provider->name); 2229 return (error); 2230 } 2231 2232 return (0); 2233} 2234 2235static int 2236g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2237 struct g_mirror_metadata *md) 2238{ 2239 2240 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2241 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2242 pp->name, md->md_did); 2243 return (EEXIST); 2244 } 2245 if (md->md_all != sc->sc_ndisks) { 2246 G_MIRROR_DEBUG(1, 2247 "Invalid '%s' field on disk %s (device %s), skipping.", 2248 "md_all", pp->name, sc->sc_name); 2249 return (EINVAL); 2250 } 2251 if (md->md_slice != sc->sc_slice) { 2252 G_MIRROR_DEBUG(1, 2253 "Invalid '%s' field on disk %s (device %s), skipping.", 2254 "md_slice", pp->name, sc->sc_name); 2255 return (EINVAL); 2256 } 2257 if (md->md_balance != sc->sc_balance) { 2258 G_MIRROR_DEBUG(1, 2259 "Invalid '%s' field on disk %s (device %s), skipping.", 2260 "md_balance", pp->name, sc->sc_name); 2261 return (EINVAL); 2262 } 2263 if (md->md_mediasize != sc->sc_mediasize) { 2264 G_MIRROR_DEBUG(1, 2265 "Invalid '%s' field on disk %s (device %s), skipping.", 2266 "md_mediasize", pp->name, sc->sc_name); 2267 return (EINVAL); 2268 } 2269 if (sc->sc_mediasize > pp->mediasize) { 2270 G_MIRROR_DEBUG(1, 2271 "Invalid size of disk %s (device %s), skipping.", pp->name, 2272 sc->sc_name); 2273 return (EINVAL); 2274 } 2275 if (md->md_sectorsize != sc->sc_sectorsize) { 2276 G_MIRROR_DEBUG(1, 2277 "Invalid '%s' field on disk %s (device %s), skipping.", 2278 "md_sectorsize", pp->name, sc->sc_name); 2279 return (EINVAL); 2280 } 2281 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2282 G_MIRROR_DEBUG(1, 2283 "Invalid sector size of disk %s (device %s), skipping.", 2284 pp->name, sc->sc_name); 2285 return (EINVAL); 2286 } 2287 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2288 G_MIRROR_DEBUG(1, 2289 "Invalid device flags on disk %s (device %s), skipping.", 2290 pp->name, sc->sc_name); 2291 return (EINVAL); 2292 } 2293 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2294 G_MIRROR_DEBUG(1, 2295 "Invalid disk flags on disk %s (device %s), skipping.", 2296 pp->name, sc->sc_name); 2297 return (EINVAL); 2298 } 2299 return (0); 2300} 2301 2302static int 2303g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2304 struct g_mirror_metadata *md) 2305{ 2306 struct g_mirror_disk *disk; 2307 int error; 2308 2309 g_topology_assert(); 2310 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2311 2312 error = g_mirror_check_metadata(sc, pp, md); 2313 if (error != 0) 2314 return (error); 2315 disk = g_mirror_init_disk(sc, pp, md, &error); 2316 if (disk == NULL) 2317 return (error); 2318 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2319 G_MIRROR_EVENT_WAIT); 2320 return (error); 2321} 2322 2323static int 2324g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2325{ 2326 struct g_mirror_softc *sc; 2327 struct g_mirror_disk *disk; 2328 int dcr, dcw, dce, err, error; 2329 2330 g_topology_assert(); 2331 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2332 acw, ace); 2333 2334 dcr = pp->acr + acr; 2335 dcw = pp->acw + acw; 2336 dce = pp->ace + ace; 2337 2338 /* On first open, grab an extra "exclusive" bit */ 2339 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0) 2340 ace++; 2341 /* ... and let go of it on last close */ 2342 if (dcr == 0 && dcw == 0 && dce == 0) 2343 ace--; 2344 2345 sc = pp->geom->softc; 2346 if (sc == NULL || LIST_EMPTY(&sc->sc_disks)) { 2347 if (acr <= 0 && acw <= 0 && ace <= 0) 2348 return (0); 2349 else 2350 return (ENXIO); 2351 } 2352 error = ENXIO; 2353 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2354 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2355 continue; 2356 err = g_access(disk->d_consumer, acr, acw, ace); 2357 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 2358 g_mirror_get_diskname(disk), acr, acw, ace, err); 2359 if (err == 0) { 2360 /* 2361 * Mark disk as dirty on open and unmark on close. 2362 */ 2363 if (pp->acw == 0 && dcw > 0) { 2364 G_MIRROR_DEBUG(1, 2365 "Disk %s (device %s) marked as dirty.", 2366 g_mirror_get_diskname(disk), sc->sc_name); 2367 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2368 g_mirror_update_metadata(disk); 2369 } else if (pp->acw > 0 && dcw == 0) { 2370 G_MIRROR_DEBUG(1, 2371 "Disk %s (device %s) marked as clean.", 2372 g_mirror_get_diskname(disk), sc->sc_name); 2373 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2374 g_mirror_update_metadata(disk); 2375 } 2376 error = 0; 2377 } else { 2378 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 2379 g_mirror_event_send(disk, 2380 G_MIRROR_DISK_STATE_DISCONNECTED, 2381 G_MIRROR_EVENT_DONTWAIT); 2382 } 2383 } 2384 return (error); 2385} 2386 2387static struct g_geom * 2388g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2389{ 2390 struct g_mirror_softc *sc; 2391 struct g_geom *gp; 2392 int error, timeout; 2393 2394 g_topology_assert(); 2395 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2396 md->md_mid); 2397 2398 /* One disk is minimum. */ 2399 if (md->md_all < 1) 2400 return (NULL); 2401 /* 2402 * Action geom. 2403 */ 2404 gp = g_new_geomf(mp, "%s", md->md_name); 2405 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2406 gp->start = g_mirror_start; 2407 gp->spoiled = g_mirror_spoiled; 2408 gp->orphan = g_mirror_orphan; 2409 gp->access = g_mirror_access; 2410 gp->dumpconf = g_mirror_dumpconf; 2411 2412 sc->sc_id = md->md_mid; 2413 sc->sc_slice = md->md_slice; 2414 sc->sc_balance = md->md_balance; 2415 sc->sc_mediasize = md->md_mediasize; 2416 sc->sc_sectorsize = md->md_sectorsize; 2417 sc->sc_ndisks = md->md_all; 2418 sc->sc_flags = md->md_mflags; 2419 sc->sc_bump_syncid = 0; 2420 bioq_init(&sc->sc_queue); 2421 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2422 LIST_INIT(&sc->sc_disks); 2423 TAILQ_INIT(&sc->sc_events); 2424 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2425 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2426 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2427 gp->softc = sc; 2428 sc->sc_geom = gp; 2429 sc->sc_provider = NULL; 2430 /* 2431 * Synchronization geom. 2432 */ 2433 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2434 gp->softc = sc; 2435 gp->spoiled = g_mirror_spoiled; 2436 gp->orphan = g_mirror_orphan; 2437 sc->sc_sync.ds_geom = gp; 2438 sc->sc_sync.ds_block = atomic_load_acq_int(&g_mirror_sync_block_size); 2439 sc->sc_sync.ds_ndisks = 0; 2440 sc->sc_sync.ds_zone = uma_zcreate("gmirror:sync", sc->sc_sync.ds_block, 2441 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2442 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2443 "g_mirror %s", md->md_name); 2444 if (error != 0) { 2445 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2446 sc->sc_name); 2447 uma_zdestroy(sc->sc_sync.ds_zone); 2448 g_destroy_geom(sc->sc_sync.ds_geom); 2449 mtx_destroy(&sc->sc_events_mtx); 2450 mtx_destroy(&sc->sc_queue_mtx); 2451 g_destroy_geom(sc->sc_geom); 2452 free(sc, M_MIRROR); 2453 return (NULL); 2454 } 2455 2456 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2457 2458 /* 2459 * Run timeout. 2460 */ 2461 timeout = atomic_load_acq_int(&g_mirror_timeout); 2462 callout_reset(&sc->sc_callout, timeout * hz, g_mirror_go, sc); 2463 return (sc->sc_geom); 2464} 2465 2466int 2467g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2468{ 2469 struct g_provider *pp; 2470 2471 g_topology_assert(); 2472 2473 if (sc == NULL) 2474 return (ENXIO); 2475 pp = sc->sc_provider; 2476 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2477 if (force) { 2478 G_MIRROR_DEBUG(0, "Device %s is still open, so it " 2479 "can't be definitely removed.", pp->name); 2480 } else { 2481 G_MIRROR_DEBUG(1, 2482 "Device %s is still open (r%dw%de%d).", pp->name, 2483 pp->acr, pp->acw, pp->ace); 2484 return (EBUSY); 2485 } 2486 } 2487 2488 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2489 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2490 g_topology_unlock(); 2491 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2492 mtx_lock(&sc->sc_queue_mtx); 2493 wakeup(sc); 2494 mtx_unlock(&sc->sc_queue_mtx); 2495 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2496 while (sc->sc_worker != NULL) 2497 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2498 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2499 g_topology_lock(); 2500 g_mirror_destroy_device(sc); 2501 free(sc, M_MIRROR); 2502 return (0); 2503} 2504 2505static void 2506g_mirror_taste_orphan(struct g_consumer *cp) 2507{ 2508 2509 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2510 cp->provider->name)); 2511} 2512 2513static struct g_geom * 2514g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2515{ 2516 struct g_mirror_metadata md; 2517 struct g_mirror_softc *sc; 2518 struct g_consumer *cp; 2519 struct g_geom *gp; 2520 int error; 2521 2522 g_topology_assert(); 2523 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2524 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2525 2526 gp = g_new_geomf(mp, "mirror:taste"); 2527 /* 2528 * This orphan function should be never called. 2529 */ 2530 gp->orphan = g_mirror_taste_orphan; 2531 cp = g_new_consumer(gp); 2532 g_attach(cp, pp); 2533 error = g_mirror_read_metadata(cp, &md); 2534 g_detach(cp); 2535 g_destroy_consumer(cp); 2536 g_destroy_geom(gp); 2537 if (error != 0) 2538 return (NULL); 2539 gp = NULL; 2540 2541 if (md.md_version > G_MIRROR_VERSION) { 2542 printf("geom_mirror.ko module is too old to handle %s.\n", 2543 pp->name); 2544 return (NULL); 2545 } 2546 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2547 return (NULL); 2548 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2549 G_MIRROR_DEBUG(0, 2550 "Device %s: provider %s marked as inactive, skipping.", 2551 md.md_name, pp->name); 2552 return (NULL); 2553 } 2554 if (g_mirror_debug >= 2) 2555 mirror_metadata_dump(&md); 2556 2557 /* 2558 * Let's check if device already exists. 2559 */ 2560 LIST_FOREACH(gp, &mp->geom, geom) { 2561 sc = gp->softc; 2562 if (sc == NULL) 2563 continue; 2564 if (sc->sc_sync.ds_geom == gp) 2565 continue; 2566 if (strcmp(md.md_name, sc->sc_name) != 0) 2567 continue; 2568 if (md.md_mid != sc->sc_id) { 2569 G_MIRROR_DEBUG(0, "Device %s already configured.", 2570 sc->sc_name); 2571 return (NULL); 2572 } 2573 break; 2574 } 2575 if (gp == NULL) { 2576 gp = g_mirror_create(mp, &md); 2577 if (gp == NULL) { 2578 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2579 md.md_name); 2580 return (NULL); 2581 } 2582 sc = gp->softc; 2583 } 2584 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2585 error = g_mirror_add_disk(sc, pp, &md); 2586 if (error != 0) { 2587 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2588 pp->name, gp->name, error); 2589 if (LIST_EMPTY(&sc->sc_disks)) 2590 g_mirror_destroy(sc, 1); 2591 return (NULL); 2592 } 2593 return (gp); 2594} 2595 2596static int 2597g_mirror_destroy_geom(struct gctl_req *req __unused, 2598 struct g_class *mp __unused, struct g_geom *gp) 2599{ 2600 2601 return (g_mirror_destroy(gp->softc, 0)); 2602} 2603 2604static void 2605g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2606 struct g_consumer *cp, struct g_provider *pp) 2607{ 2608 struct g_mirror_softc *sc; 2609 2610 g_topology_assert(); 2611 2612 sc = gp->softc; 2613 if (sc == NULL) 2614 return; 2615 /* Skip synchronization geom. */ 2616 if (gp == sc->sc_sync.ds_geom) 2617 return; 2618 if (pp != NULL) { 2619 /* Nothing here. */ 2620 } else if (cp != NULL) { 2621 struct g_mirror_disk *disk; 2622 2623 disk = cp->private; 2624 if (disk == NULL) 2625 return; 2626 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2627 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2628 sbuf_printf(sb, "%s<Synchronized>", indent); 2629 if (disk->d_sync.ds_offset_done == 0) 2630 sbuf_printf(sb, "0%%"); 2631 else { 2632 sbuf_printf(sb, "%u%%", 2633 (u_int)((disk->d_sync.ds_offset_done * 100) / 2634 sc->sc_provider->mediasize)); 2635 } 2636 sbuf_printf(sb, "</Synchronized>\n"); 2637 } 2638 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2639 disk->d_sync.ds_syncid); 2640 sbuf_printf(sb, "%s<Flags>", indent); 2641 if (disk->d_flags == 0) 2642 sbuf_printf(sb, "NONE"); 2643 else { 2644 int first = 1; 2645 2646#define ADD_FLAG(flag, name) do { \ 2647 if ((disk->d_flags & (flag)) != 0) { \ 2648 if (!first) \ 2649 sbuf_printf(sb, ", "); \ 2650 else \ 2651 first = 0; \ 2652 sbuf_printf(sb, name); \ 2653 } \ 2654} while (0) 2655 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2656 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2657 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2658 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2659 "SYNCHRONIZING"); 2660 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2661#undef ADD_FLAG 2662 } 2663 sbuf_printf(sb, "</Flags>\n"); 2664 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2665 disk->d_priority); 2666 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2667 g_mirror_disk_state2str(disk->d_state)); 2668 } else { 2669 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2670 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2671 sbuf_printf(sb, "%s<Flags>", indent); 2672 if (sc->sc_flags == 0) 2673 sbuf_printf(sb, "NONE"); 2674 else { 2675 int first = 1; 2676 2677#define ADD_FLAG(flag, name) do { \ 2678 if ((sc->sc_flags & (flag)) != 0) { \ 2679 if (!first) \ 2680 sbuf_printf(sb, ", "); \ 2681 else \ 2682 first = 0; \ 2683 sbuf_printf(sb, name); \ 2684 } \ 2685} while (0) 2686 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2687#undef ADD_FLAG 2688 } 2689 sbuf_printf(sb, "</Flags>\n"); 2690 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2691 (u_int)sc->sc_slice); 2692 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2693 balance_name(sc->sc_balance)); 2694 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2695 sc->sc_ndisks); 2696 } 2697} 2698 2699DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2700