g_mirror.c revision 132922
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 132922 2004-07-31 11:22:03Z 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} 400 401/* 402 * Free the disk. 403 */ 404static void 405g_mirror_free_disk(struct g_mirror_disk *disk) 406{ 407 408 g_topology_assert(); 409 410 g_mirror_disconnect_disk(disk); 411 free(disk, M_MIRROR); 412} 413 414static void 415g_mirror_destroy_disk(struct g_mirror_disk *disk) 416{ 417 struct g_mirror_softc *sc; 418 419 g_topology_assert(); 420 421 LIST_REMOVE(disk, d_next); 422 g_mirror_event_cancel(disk); 423 sc = disk->d_softc; 424 if (sc->sc_hint == disk) 425 sc->sc_hint = NULL; 426 switch (disk->d_state) { 427 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 428 g_mirror_sync_stop(disk, 1); 429 /* FALLTHROUGH */ 430 case G_MIRROR_DISK_STATE_NEW: 431 case G_MIRROR_DISK_STATE_STALE: 432 case G_MIRROR_DISK_STATE_ACTIVE: 433 g_mirror_free_disk(disk); 434 break; 435 default: 436 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 437 g_mirror_get_diskname(disk), 438 g_mirror_disk_state2str(disk->d_state))); 439 } 440} 441 442static void 443g_mirror_destroy_device(struct g_mirror_softc *sc) 444{ 445 struct g_mirror_disk *disk; 446 struct g_mirror_event *ep; 447 struct g_geom *gp; 448 struct g_consumer *cp; 449 450 g_topology_assert(); 451 452 gp = sc->sc_geom; 453 if (sc->sc_provider != NULL) 454 g_mirror_destroy_provider(sc); 455 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 456 disk = LIST_FIRST(&sc->sc_disks)) { 457 g_mirror_destroy_disk(disk); 458 } 459 while ((ep = g_mirror_event_get(sc)) != NULL) { 460 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 461 g_mirror_event_free(ep); 462 else { 463 ep->e_error = ECANCELED; 464 ep->e_flags |= G_MIRROR_EVENT_DONE; 465 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 466 mtx_lock(&sc->sc_events_mtx); 467 wakeup(ep); 468 mtx_unlock(&sc->sc_events_mtx); 469 } 470 } 471 callout_drain(&sc->sc_callout); 472 gp->softc = NULL; 473 uma_zdestroy(sc->sc_sync.ds_zone); 474 while ((cp = LIST_FIRST(&sc->sc_sync.ds_geom->consumer)) != NULL) { 475 if (cp->provider != NULL) { 476 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 477 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 478 g_detach(cp); 479 } 480 g_destroy_consumer(cp); 481 } 482 sc->sc_sync.ds_geom->softc = NULL; 483 g_destroy_geom(sc->sc_sync.ds_geom); 484 mtx_destroy(&sc->sc_queue_mtx); 485 mtx_destroy(&sc->sc_events_mtx); 486 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 487 g_wither_geom(gp, ENXIO); 488} 489 490static void 491g_mirror_orphan(struct g_consumer *cp) 492{ 493 struct g_mirror_disk *disk; 494 495 g_topology_assert(); 496 497 disk = cp->private; 498 if (disk == NULL) 499 return; 500 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 501 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 502 G_MIRROR_EVENT_DONTWAIT); 503} 504 505/* 506 * Function should return the next active disk on the list. 507 * It is possible that it will be the same disk as given. 508 * If there are no active disks on list, NULL is returned. 509 */ 510static __inline struct g_mirror_disk * 511g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 512{ 513 struct g_mirror_disk *dp; 514 515 for (dp = LIST_NEXT(disk, d_next); dp != disk; 516 dp = LIST_NEXT(dp, d_next)) { 517 if (dp == NULL) 518 dp = LIST_FIRST(&sc->sc_disks); 519 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 520 break; 521 } 522 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 523 return (NULL); 524 return (dp); 525} 526 527static struct g_mirror_disk * 528g_mirror_get_disk(struct g_mirror_softc *sc) 529{ 530 struct g_mirror_disk *disk; 531 532 if (sc->sc_hint == NULL) { 533 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 534 if (sc->sc_hint == NULL) 535 return (NULL); 536 } 537 disk = sc->sc_hint; 538 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 539 disk = g_mirror_find_next(sc, disk); 540 if (disk == NULL) 541 return (NULL); 542 } 543 sc->sc_hint = g_mirror_find_next(sc, disk); 544 return (disk); 545} 546 547static int 548g_mirror_clear_metadata(struct g_mirror_disk *disk) 549{ 550 struct g_mirror_softc *sc; 551 struct g_consumer *cp; 552 off_t offset, length; 553 u_char *sector; 554 int close = 0, error = 0; 555 556 g_topology_assert(); 557 558 sc = disk->d_softc; 559 cp = disk->d_consumer; 560 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 561 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 562 length = cp->provider->sectorsize; 563 offset = cp->provider->mediasize - length; 564 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 565 /* 566 * Open consumer if it wasn't opened and remember to close it. 567 */ 568 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 569 error = g_access(cp, 0, 1, 1); 570 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 571 cp->provider->name, 0, 1, 1, error); 572 if (error == 0) 573 close = 1; 574#ifdef INVARIANTS 575 } else { 576 KASSERT(cp->acw > 0 && cp->ace > 0, 577 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 578 cp->acr, cp->acw, cp->ace)); 579#endif 580 } 581 if (error == 0) { 582 g_topology_unlock(); 583 error = g_write_data(cp, offset, sector, length); 584 g_topology_lock(); 585 } 586 free(sector, M_MIRROR); 587 if (close) { 588 g_access(cp, 0, -1, -1); 589 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 590 cp->provider->name, 0, -1, -1, 0); 591 } 592 if (error != 0) { 593 G_MIRROR_DEBUG(0, "Cannot clear metadata on disk %s.", 594 g_mirror_get_diskname(disk)); 595 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 596 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 597 G_MIRROR_EVENT_DONTWAIT); 598 return (error); 599 } 600 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 601 g_mirror_get_diskname(disk)); 602 return (0); 603} 604 605void 606g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 607 struct g_mirror_metadata *md) 608{ 609 610 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 611 md->md_version = G_MIRROR_VERSION; 612 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 613 md->md_mid = sc->sc_id; 614 md->md_all = sc->sc_ndisks; 615 md->md_slice = sc->sc_slice; 616 md->md_balance = sc->sc_balance; 617 md->md_mediasize = sc->sc_mediasize; 618 md->md_sectorsize = sc->sc_sectorsize; 619 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 620 if (disk == NULL) { 621 md->md_did = arc4random(); 622 md->md_priority = 0; 623 md->md_syncid = 0; 624 md->md_dflags = 0; 625 md->md_sync_offset = 0; 626 } else { 627 md->md_did = disk->d_id; 628 md->md_priority = disk->d_priority; 629 md->md_syncid = disk->d_sync.ds_syncid; 630 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 631 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 632 md->md_sync_offset = disk->d_sync.ds_offset_done; 633 else 634 md->md_sync_offset = 0; 635 } 636} 637 638void 639g_mirror_update_metadata(struct g_mirror_disk *disk) 640{ 641 struct g_mirror_softc *sc; 642 struct g_mirror_metadata md; 643 struct g_consumer *cp; 644 off_t offset, length; 645 u_char *sector; 646 int close = 0, error = 0; 647 648 g_topology_assert(); 649 650 sc = disk->d_softc; 651 cp = disk->d_consumer; 652 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 653 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 654 length = cp->provider->sectorsize; 655 offset = cp->provider->mediasize - length; 656 sector = malloc((size_t)length, M_MIRROR, M_WAITOK); 657 /* 658 * Open consumer if it wasn't opened and remember to close it. 659 */ 660 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 661 error = g_access(cp, 0, 1, 1); 662 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 663 cp->provider->name, 0, 1, 1, error); 664 if (error == 0) 665 close = 1; 666#ifdef INVARIANTS 667 } else { 668 KASSERT(cp->acw > 0 && cp->ace > 0, 669 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 670 cp->acr, cp->acw, cp->ace)); 671#endif 672 } 673 if (error == 0) { 674 g_mirror_fill_metadata(sc, disk, &md); 675 mirror_metadata_encode(&md, sector); 676 g_topology_unlock(); 677 error = g_write_data(cp, offset, sector, length); 678 g_topology_lock(); 679 } 680 free(sector, M_MIRROR); 681 if (close) { 682 g_access(cp, 0, -1, -1); 683 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 684 cp->provider->name, 0, -1, -1, 0); 685 } 686 if (error != 0) { 687 G_MIRROR_DEBUG(0, 688 "Cannot update metadata on disk %s (error=%d).", 689 g_mirror_get_diskname(disk), error); 690 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 691 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 692 G_MIRROR_EVENT_DONTWAIT); 693 return; 694 } 695 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 696 g_mirror_get_diskname(disk)); 697} 698 699static void 700g_mirror_bump_syncid(struct g_mirror_softc *sc) 701{ 702 struct g_mirror_disk *disk; 703 704 g_topology_assert(); 705 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 706 ("%s called with no active disks (device=%s).", __func__, 707 sc->sc_name)); 708 709 sc->sc_syncid++; 710 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 711 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 712 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 713 disk->d_sync.ds_syncid = sc->sc_syncid; 714 g_mirror_update_metadata(disk); 715 } 716 } 717} 718 719static __inline int 720bintime_cmp(struct bintime *bt1, struct bintime *bt2) 721{ 722 723 if (bt1->sec < bt2->sec) 724 return (-1); 725 else if (bt1->sec > bt2->sec) 726 return (1); 727 if (bt1->frac < bt2->frac) 728 return (-1); 729 else if (bt1->frac > bt2->frac) 730 return (1); 731 return (0); 732} 733 734static void 735g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 736{ 737 738 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 739 return; 740 binuptime(&disk->d_delay); 741 bintime_sub(&disk->d_delay, &bp->bio_t0); 742} 743 744static void 745g_mirror_done(struct bio *bp) 746{ 747 struct g_mirror_softc *sc; 748 749 sc = bp->bio_from->geom->softc; 750 bp->bio_flags = BIO_FLAG1; 751 mtx_lock(&sc->sc_queue_mtx); 752 bioq_disksort(&sc->sc_queue, bp); 753 wakeup(sc); 754 mtx_unlock(&sc->sc_queue_mtx); 755} 756 757static void 758g_mirror_regular_request(struct bio *bp) 759{ 760 struct g_mirror_softc *sc; 761 struct g_mirror_disk *disk; 762 struct bio *pbp; 763 764 g_topology_assert_not(); 765 766 pbp = bp->bio_parent; 767 sc = pbp->bio_to->geom->softc; 768 disk = bp->bio_from->private; 769 if (disk == NULL) { 770 g_topology_lock(); 771 g_mirror_kill_consumer(sc, bp->bio_from); 772 g_topology_unlock(); 773 } else { 774 g_mirror_update_delay(disk, bp); 775 } 776 777 pbp->bio_inbed++; 778 KASSERT(pbp->bio_inbed <= pbp->bio_children, 779 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 780 pbp->bio_children)); 781 if (bp->bio_error == 0 && pbp->bio_error == 0) { 782 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 783 g_destroy_bio(bp); 784 if (pbp->bio_children == pbp->bio_inbed) { 785 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 786 pbp->bio_completed = pbp->bio_length; 787 g_io_deliver(pbp, pbp->bio_error); 788 } 789 return; 790 } else if (bp->bio_error != 0) { 791 if (pbp->bio_error == 0) 792 pbp->bio_error = bp->bio_error; 793 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 794 bp->bio_error); 795 if (disk != NULL) { 796 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 797 g_mirror_event_send(disk, 798 G_MIRROR_DISK_STATE_DISCONNECTED, 799 G_MIRROR_EVENT_DONTWAIT); 800 } 801 switch (pbp->bio_cmd) { 802 case BIO_DELETE: 803 case BIO_WRITE: 804 pbp->bio_inbed--; 805 pbp->bio_children--; 806 break; 807 } 808 } 809 g_destroy_bio(bp); 810 811 switch (pbp->bio_cmd) { 812 case BIO_READ: 813 if (pbp->bio_children == pbp->bio_inbed) { 814 pbp->bio_error = 0; 815 mtx_lock(&sc->sc_queue_mtx); 816 bioq_disksort(&sc->sc_queue, pbp); 817 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 818 wakeup(sc); 819 mtx_unlock(&sc->sc_queue_mtx); 820 } 821 break; 822 case BIO_DELETE: 823 case BIO_WRITE: 824 if (pbp->bio_children == 0) { 825 /* 826 * All requests failed. 827 */ 828 } else if (pbp->bio_inbed < pbp->bio_children) { 829 /* Do nothing. */ 830 break; 831 } else if (pbp->bio_children == pbp->bio_inbed) { 832 /* Some requests succeeded. */ 833 pbp->bio_error = 0; 834 pbp->bio_completed = pbp->bio_length; 835 } 836 g_io_deliver(pbp, pbp->bio_error); 837 break; 838 default: 839 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 840 break; 841 } 842} 843 844static void 845g_mirror_sync_done(struct bio *bp) 846{ 847 struct g_mirror_softc *sc; 848 849 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 850 sc = bp->bio_from->geom->softc; 851 bp->bio_flags = BIO_FLAG2; 852 mtx_lock(&sc->sc_queue_mtx); 853 bioq_disksort(&sc->sc_queue, bp); 854 wakeup(sc); 855 mtx_unlock(&sc->sc_queue_mtx); 856} 857 858static void 859g_mirror_start(struct bio *bp) 860{ 861 struct g_mirror_softc *sc; 862 863 sc = bp->bio_to->geom->softc; 864 /* 865 * If sc == NULL or there are no valid disks, provider's error 866 * should be set and g_mirror_start() should not be called at all. 867 */ 868 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 869 ("Provider's error should be set (error=%d)(mirror=%s).", 870 bp->bio_to->error, bp->bio_to->name)); 871 G_MIRROR_LOGREQ(3, bp, "Request received."); 872 873 switch (bp->bio_cmd) { 874 case BIO_READ: 875 case BIO_WRITE: 876 case BIO_DELETE: 877 break; 878 case BIO_GETATTR: 879 default: 880 g_io_deliver(bp, EOPNOTSUPP); 881 return; 882 } 883 mtx_lock(&sc->sc_queue_mtx); 884 bioq_disksort(&sc->sc_queue, bp); 885 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 886 wakeup(sc); 887 mtx_unlock(&sc->sc_queue_mtx); 888} 889 890/* 891 * Send one synchronization request. 892 */ 893static void 894g_mirror_sync_one(struct g_mirror_disk *disk) 895{ 896 struct g_mirror_softc *sc; 897 struct bio *bp; 898 899 sc = disk->d_softc; 900 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 901 ("Disk %s is not marked for synchronization.", 902 g_mirror_get_diskname(disk))); 903 904 bp = g_new_bio(); 905 if (bp == NULL) 906 return; 907 bp->bio_parent = NULL; 908 bp->bio_cmd = BIO_READ; 909 bp->bio_offset = disk->d_sync.ds_offset; 910 bp->bio_length = MIN(sc->sc_sync.ds_block, 911 sc->sc_mediasize - bp->bio_offset); 912 bp->bio_flags = 0; 913 bp->bio_done = g_mirror_sync_done; 914 bp->bio_data = uma_zalloc(sc->sc_sync.ds_zone, M_NOWAIT | M_ZERO); 915 if (bp->bio_data == NULL) { 916 g_destroy_bio(bp); 917 return; 918 } 919 disk->d_sync.ds_offset += bp->bio_length; 920 bp->bio_to = sc->sc_provider; 921 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 922 g_io_request(bp, disk->d_sync.ds_consumer); 923} 924 925static void 926g_mirror_sync_request(struct bio *bp) 927{ 928 struct g_mirror_softc *sc; 929 struct g_mirror_disk *disk; 930 931 sc = bp->bio_from->geom->softc; 932 disk = bp->bio_from->private; 933 if (disk == NULL) { 934 g_topology_lock(); 935 g_mirror_kill_consumer(sc, bp->bio_from); 936 g_topology_unlock(); 937 uma_zfree(sc->sc_sync.ds_zone, bp->bio_data); 938 g_destroy_bio(bp); 939 return; 940 } 941 942 /* 943 * Synchronization request. 944 */ 945 switch (bp->bio_cmd) { 946 case BIO_READ: 947 { 948 struct g_consumer *cp; 949 950 if (bp->bio_error != 0) { 951 G_MIRROR_LOGREQ(0, bp, 952 "Synchronization request failed (error=%d).", 953 bp->bio_error); 954 uma_zfree(sc->sc_sync.ds_zone, bp->bio_data); 955 g_destroy_bio(bp); 956 return; 957 } 958 bp->bio_cmd = BIO_WRITE; 959 bp->bio_flags = 0; 960 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 961 cp = disk->d_consumer; 962 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 963 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 964 cp->acr, cp->acw, cp->ace)); 965 g_io_request(bp, cp); 966 return; 967 } 968 case BIO_WRITE: 969 uma_zfree(sc->sc_sync.ds_zone, bp->bio_data); 970 if (bp->bio_error != 0) { 971 G_MIRROR_LOGREQ(0, bp, 972 "Synchronization request failed (error=%d).", 973 bp->bio_error); 974 g_destroy_bio(bp); 975 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 976 g_mirror_event_send(disk, 977 G_MIRROR_DISK_STATE_DISCONNECTED, 978 G_MIRROR_EVENT_DONTWAIT); 979 return; 980 } 981 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 982 g_destroy_bio(bp); 983 disk->d_sync.ds_offset_done = bp->bio_offset + bp->bio_length; 984 if (bp->bio_offset + bp->bio_length == 985 sc->sc_provider->mediasize) { 986 /* 987 * Disk up-to-date, activate it. 988 */ 989 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 990 G_MIRROR_EVENT_DONTWAIT); 991 return; 992 } else if ((disk->d_sync.ds_offset_done % 993 (sc->sc_sync.ds_block * 100)) == 0) { 994 /* 995 * Update offset_done on every 100 blocks. 996 * XXX: This should be configurable. 997 */ 998 g_topology_lock(); 999 g_mirror_update_metadata(disk); 1000 g_topology_unlock(); 1001 } 1002 return; 1003 default: 1004 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1005 bp->bio_cmd, sc->sc_name)); 1006 break; 1007 } 1008} 1009 1010static void 1011g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1012{ 1013 struct g_mirror_disk *disk; 1014 struct g_consumer *cp; 1015 struct bio *cbp; 1016 1017 disk = g_mirror_get_disk(sc); 1018 if (disk == NULL) { 1019 if (bp->bio_error == 0) 1020 bp->bio_error = ENXIO; 1021 g_io_deliver(bp, bp->bio_error); 1022 return; 1023 } 1024 cbp = g_clone_bio(bp); 1025 if (cbp == NULL) { 1026 if (bp->bio_error == 0) 1027 bp->bio_error = ENOMEM; 1028 g_io_deliver(bp, bp->bio_error); 1029 return; 1030 } 1031 /* 1032 * Fill in the component buf structure. 1033 */ 1034 cp = disk->d_consumer; 1035 cbp->bio_done = g_mirror_done; 1036 cbp->bio_to = cp->provider; 1037 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1038 KASSERT(cp->acr > 0 && cp->ace > 0, 1039 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1040 cp->acw, cp->ace)); 1041 g_io_request(cbp, cp); 1042} 1043 1044static void 1045g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1046{ 1047 struct g_mirror_disk *disk, *dp; 1048 struct g_consumer *cp; 1049 struct bio *cbp; 1050 struct bintime curtime; 1051 1052 binuptime(&curtime); 1053 /* 1054 * Find a disk which the smallest load. 1055 */ 1056 disk = NULL; 1057 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1058 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1059 continue; 1060 /* If disk wasn't used for more than 2 sec, use it. */ 1061 if (curtime.sec - dp->d_last_used.sec >= 2) { 1062 disk = dp; 1063 break; 1064 } 1065 if (disk == NULL || 1066 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1067 disk = dp; 1068 } 1069 } 1070 cbp = g_clone_bio(bp); 1071 if (cbp == NULL) { 1072 if (bp->bio_error == 0) 1073 bp->bio_error = ENOMEM; 1074 g_io_deliver(bp, bp->bio_error); 1075 return; 1076 } 1077 /* 1078 * Fill in the component buf structure. 1079 */ 1080 cp = disk->d_consumer; 1081 cbp->bio_done = g_mirror_done; 1082 cbp->bio_to = cp->provider; 1083 binuptime(&disk->d_last_used); 1084 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1085 KASSERT(cp->acr > 0 && cp->ace > 0, 1086 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1087 cp->acw, cp->ace)); 1088 g_io_request(cbp, cp); 1089} 1090 1091static void 1092g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1093{ 1094 struct bio_queue_head queue; 1095 struct g_mirror_disk *disk; 1096 struct g_consumer *cp; 1097 struct bio *cbp; 1098 off_t left, mod, offset, slice; 1099 u_char *data; 1100 u_int ndisks; 1101 1102 if (bp->bio_length <= sc->sc_slice) { 1103 g_mirror_request_round_robin(sc, bp); 1104 return; 1105 } 1106 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1107 slice = bp->bio_length / ndisks; 1108 mod = slice % sc->sc_provider->sectorsize; 1109 if (mod != 0) 1110 slice += sc->sc_provider->sectorsize - mod; 1111 /* 1112 * Allocate all bios before sending any request, so we can 1113 * return ENOMEM in nice and clean way. 1114 */ 1115 left = bp->bio_length; 1116 offset = bp->bio_offset; 1117 data = bp->bio_data; 1118 bioq_init(&queue); 1119 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1120 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1121 continue; 1122 cbp = g_clone_bio(bp); 1123 if (cbp == NULL) { 1124 for (cbp = bioq_first(&queue); cbp != NULL; 1125 cbp = bioq_first(&queue)) { 1126 bioq_remove(&queue, cbp); 1127 g_destroy_bio(cbp); 1128 } 1129 if (bp->bio_error == 0) 1130 bp->bio_error = ENOMEM; 1131 g_io_deliver(bp, bp->bio_error); 1132 return; 1133 } 1134 bioq_insert_tail(&queue, cbp); 1135 cbp->bio_done = g_mirror_done; 1136 cbp->bio_caller1 = disk; 1137 cbp->bio_to = disk->d_consumer->provider; 1138 cbp->bio_offset = offset; 1139 cbp->bio_data = data; 1140 cbp->bio_length = MIN(left, slice); 1141 left -= cbp->bio_length; 1142 if (left == 0) 1143 break; 1144 offset += cbp->bio_length; 1145 data += cbp->bio_length; 1146 } 1147 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1148 bioq_remove(&queue, cbp); 1149 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1150 disk = cbp->bio_caller1; 1151 cbp->bio_caller1 = NULL; 1152 cp = disk->d_consumer; 1153 KASSERT(cp->acr > 0 && cp->ace > 0, 1154 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1155 cp->acr, cp->acw, cp->ace)); 1156 g_io_request(cbp, disk->d_consumer); 1157 } 1158} 1159 1160static void 1161g_mirror_register_request(struct bio *bp) 1162{ 1163 struct g_mirror_softc *sc; 1164 1165 sc = bp->bio_to->geom->softc; 1166 switch (bp->bio_cmd) { 1167 case BIO_READ: 1168 switch (sc->sc_balance) { 1169 case G_MIRROR_BALANCE_ROUND_ROBIN: 1170 g_mirror_request_round_robin(sc, bp); 1171 break; 1172 case G_MIRROR_BALANCE_LOAD: 1173 g_mirror_request_load(sc, bp); 1174 break; 1175 case G_MIRROR_BALANCE_SPLIT: 1176 g_mirror_request_split(sc, bp); 1177 break; 1178 } 1179 return; 1180 case BIO_WRITE: 1181 case BIO_DELETE: 1182 { 1183 struct g_mirror_disk *disk; 1184 struct bio_queue_head queue; 1185 struct g_consumer *cp; 1186 struct bio *cbp; 1187 1188 /* 1189 * Allocate all bios before sending any request, so we can 1190 * return ENOMEM in nice and clean way. 1191 */ 1192 bioq_init(&queue); 1193 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1194 switch (disk->d_state) { 1195 case G_MIRROR_DISK_STATE_ACTIVE: 1196 break; 1197 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1198 if (bp->bio_offset >= disk->d_sync.ds_offset) 1199 continue; 1200 break; 1201 default: 1202 continue; 1203 } 1204 cbp = g_clone_bio(bp); 1205 if (cbp == NULL) { 1206 for (cbp = bioq_first(&queue); cbp != NULL; 1207 cbp = bioq_first(&queue)) { 1208 bioq_remove(&queue, cbp); 1209 g_destroy_bio(cbp); 1210 } 1211 if (bp->bio_error == 0) 1212 bp->bio_error = ENOMEM; 1213 g_io_deliver(bp, bp->bio_error); 1214 return; 1215 } 1216 bioq_insert_tail(&queue, cbp); 1217 } 1218 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1219 switch (disk->d_state) { 1220 case G_MIRROR_DISK_STATE_ACTIVE: 1221 break; 1222 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1223 if (bp->bio_offset >= disk->d_sync.ds_offset) 1224 continue; 1225 break; 1226 default: 1227 continue; 1228 } 1229 cbp = bioq_first(&queue); 1230 KASSERT(cbp != NULL, ("NULL cbp! (device %s).", 1231 sc->sc_name)); 1232 bioq_remove(&queue, cbp); 1233 cp = disk->d_consumer; 1234 cbp->bio_done = g_mirror_done; 1235 cbp->bio_to = cp->provider; 1236 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1237 KASSERT(cp->acw > 0 && cp->ace > 0, 1238 ("Consumer %s not opened (r%dw%de%d).", 1239 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1240 g_io_request(cbp, cp); 1241 } 1242 /* 1243 * Bump syncid on first write. 1244 */ 1245 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) { 1246 sc->sc_bump_syncid = 0; 1247 g_topology_lock(); 1248 g_mirror_bump_syncid(sc); 1249 g_topology_unlock(); 1250 } 1251 return; 1252 } 1253 default: 1254 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1255 bp->bio_cmd, sc->sc_name)); 1256 break; 1257 } 1258} 1259 1260/* 1261 * Worker thread. 1262 */ 1263static void 1264g_mirror_worker(void *arg) 1265{ 1266 struct g_mirror_softc *sc; 1267 struct g_mirror_disk *disk; 1268 struct g_mirror_event *ep; 1269 struct bio *bp; 1270 u_int nreqs; 1271 1272 sc = arg; 1273 curthread->td_base_pri = PRIBIO; 1274 1275 nreqs = 0; 1276 for (;;) { 1277 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1278 /* 1279 * First take a look at events. 1280 * This is important to handle events before any I/O requests. 1281 */ 1282 ep = g_mirror_event_get(sc); 1283 if (ep != NULL) { 1284 g_topology_lock(); 1285 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1286 /* Update only device status. */ 1287 G_MIRROR_DEBUG(3, 1288 "Running event for device %s.", 1289 sc->sc_name); 1290 ep->e_error = 0; 1291 g_mirror_update_device(sc, 1); 1292 } else { 1293 /* Update disk status. */ 1294 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1295 g_mirror_get_diskname(ep->e_disk)); 1296 ep->e_error = g_mirror_update_disk(ep->e_disk, 1297 ep->e_state); 1298 if (ep->e_error == 0) 1299 g_mirror_update_device(sc, 0); 1300 } 1301 g_topology_unlock(); 1302 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1303 KASSERT(ep->e_error == 0, 1304 ("Error cannot be handled.")); 1305 g_mirror_event_free(ep); 1306 } else { 1307 ep->e_flags |= G_MIRROR_EVENT_DONE; 1308 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1309 ep); 1310 mtx_lock(&sc->sc_events_mtx); 1311 wakeup(ep); 1312 mtx_unlock(&sc->sc_events_mtx); 1313 } 1314 if ((sc->sc_flags & 1315 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1316end: 1317 if ((sc->sc_flags & 1318 G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1319 G_MIRROR_DEBUG(4, "%s: Waking up %p.", 1320 __func__, &sc->sc_worker); 1321 wakeup(&sc->sc_worker); 1322 sc->sc_worker = NULL; 1323 } else { 1324 g_topology_lock(); 1325 g_mirror_destroy_device(sc); 1326 g_topology_unlock(); 1327 free(sc, M_MIRROR); 1328 } 1329 kthread_exit(0); 1330 } 1331 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1332 continue; 1333 } 1334 /* 1335 * Now I/O requests. 1336 */ 1337 /* Get first request from the queue. */ 1338 mtx_lock(&sc->sc_queue_mtx); 1339 bp = bioq_first(&sc->sc_queue); 1340 if (bp == NULL) { 1341 if ((sc->sc_flags & 1342 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1343 mtx_unlock(&sc->sc_queue_mtx); 1344 goto end; 1345 } 1346 } 1347 if (sc->sc_sync.ds_ndisks > 0 && 1348 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1349 mtx_unlock(&sc->sc_queue_mtx); 1350 /* 1351 * It is time for synchronization... 1352 */ 1353 nreqs = 0; 1354 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1355 if (disk->d_state != 1356 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1357 continue; 1358 } 1359 if (disk->d_sync.ds_offset >= 1360 sc->sc_provider->mediasize) { 1361 continue; 1362 } 1363 if (disk->d_sync.ds_offset > 1364 disk->d_sync.ds_offset_done) { 1365 continue; 1366 } 1367 g_mirror_sync_one(disk); 1368 } 1369 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1370 goto sleep; 1371 } 1372 if (bp == NULL) { 1373 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 0); 1374 G_MIRROR_DEBUG(5, "%s: I'm here 3.", __func__); 1375 continue; 1376 } 1377 nreqs++; 1378 bioq_remove(&sc->sc_queue, bp); 1379 mtx_unlock(&sc->sc_queue_mtx); 1380 1381 if ((bp->bio_flags & BIO_FLAG1) != 0) { 1382 g_mirror_regular_request(bp); 1383 } else if ((bp->bio_flags & BIO_FLAG2) != 0) { 1384 u_int timeout, sps; 1385 1386 g_mirror_sync_request(bp); 1387sleep: 1388 sps = atomic_load_acq_int(&g_mirror_syncs_per_sec); 1389 if (sps == 0) { 1390 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1391 continue; 1392 } 1393 mtx_lock(&sc->sc_queue_mtx); 1394 if (bioq_first(&sc->sc_queue) != NULL) { 1395 mtx_unlock(&sc->sc_queue_mtx); 1396 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1397 continue; 1398 } 1399 timeout = hz / sps; 1400 if (timeout == 0) 1401 timeout = 1; 1402 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w2", 1403 timeout); 1404 } else { 1405 g_mirror_register_request(bp); 1406 } 1407 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1408 } 1409} 1410 1411/* 1412 * Open disk's consumer if needed. 1413 */ 1414static void 1415g_mirror_update_access(struct g_mirror_disk *disk) 1416{ 1417 struct g_provider *pp; 1418 struct g_consumer *cp; 1419 int acr, acw, ace, cpw, error; 1420 1421 g_topology_assert(); 1422 1423 cp = disk->d_consumer; 1424 pp = disk->d_softc->sc_provider; 1425 if (pp == NULL) { 1426 acr = -cp->acr; 1427 acw = -cp->acw; 1428 ace = -cp->ace; 1429 } else { 1430 acr = pp->acr - cp->acr; 1431 acw = pp->acw - cp->acw; 1432 ace = pp->ace - cp->ace; 1433 /* Grab an extra "exclusive" bit. */ 1434 if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) 1435 ace++; 1436 } 1437 if (acr == 0 && acw == 0 && ace == 0) 1438 return; 1439 cpw = cp->acw; 1440 error = g_access(cp, acr, acw, ace); 1441 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, acr, 1442 acw, ace, error); 1443 if (error != 0) { 1444 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1445 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1446 G_MIRROR_EVENT_DONTWAIT); 1447 return; 1448 } 1449 if (cpw == 0 && cp->acw > 0) { 1450 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1451 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1452 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1453 } else if (cpw > 0 && cp->acw == 0) { 1454 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1455 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1456 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1457 } 1458} 1459 1460static void 1461g_mirror_sync_start(struct g_mirror_disk *disk) 1462{ 1463 struct g_mirror_softc *sc; 1464 struct g_consumer *cp; 1465 int error; 1466 1467 g_topology_assert(); 1468 1469 sc = disk->d_softc; 1470 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1471 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1472 sc->sc_state)); 1473 cp = disk->d_consumer; 1474 KASSERT(cp->acr == 0 && cp->acw == 0 && cp->ace == 0, 1475 ("Consumer %s already opened.", cp->provider->name)); 1476 1477 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1478 g_mirror_get_diskname(disk)); 1479 error = g_access(cp, 0, 1, 1); 1480 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, 1, 1481 1, error); 1482 if (error != 0) { 1483 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1484 G_MIRROR_EVENT_DONTWAIT); 1485 return; 1486 } 1487 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1488 KASSERT(disk->d_sync.ds_consumer == NULL, 1489 ("Sync consumer already exists (device=%s, disk=%s).", 1490 sc->sc_name, g_mirror_get_diskname(disk))); 1491 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1492 disk->d_sync.ds_consumer->private = disk; 1493 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1494 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1495 disk->d_softc->sc_name, error)); 1496 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1497 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1498 disk->d_softc->sc_name, error)); 1499 sc->sc_sync.ds_ndisks++; 1500} 1501 1502/* 1503 * Stop synchronization process. 1504 * type: 0 - synchronization finished 1505 * 1 - synchronization stopped 1506 */ 1507static void 1508g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1509{ 1510 struct g_consumer *cp; 1511 1512 g_topology_assert(); 1513 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1514 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1515 g_mirror_disk_state2str(disk->d_state))); 1516 if (disk->d_sync.ds_consumer == NULL) 1517 return; 1518 1519 if (type == 0) { 1520 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1521 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1522 } else /* if (type == 1) */ { 1523 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1524 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1525 } 1526 cp = disk->d_sync.ds_consumer; 1527 g_access(cp, -1, 0, 0); 1528 g_mirror_kill_consumer(disk->d_softc, cp); 1529 disk->d_sync.ds_consumer = NULL; 1530 disk->d_softc->sc_sync.ds_ndisks--; 1531 cp = disk->d_consumer; 1532 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 1533 ("Consumer %s not opened.", cp->provider->name)); 1534 g_access(cp, 0, -1, -1); 1535 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, -1, 1536 -1, 0); 1537 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1538} 1539 1540static void 1541g_mirror_launch_provider(struct g_mirror_softc *sc) 1542{ 1543 struct g_mirror_disk *disk; 1544 struct g_provider *pp; 1545 1546 g_topology_assert(); 1547 1548 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1549 pp->mediasize = sc->sc_mediasize; 1550 pp->sectorsize = sc->sc_sectorsize; 1551 sc->sc_provider = pp; 1552 g_error_provider(pp, 0); 1553 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1554 pp->name); 1555 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1556 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1557 g_mirror_sync_start(disk); 1558 } 1559} 1560 1561static void 1562g_mirror_destroy_provider(struct g_mirror_softc *sc) 1563{ 1564 struct g_mirror_disk *disk; 1565 struct bio *bp; 1566 1567 g_topology_assert(); 1568 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1569 sc->sc_name)); 1570 1571 g_error_provider(sc->sc_provider, ENXIO); 1572 mtx_lock(&sc->sc_queue_mtx); 1573 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1574 bioq_remove(&sc->sc_queue, bp); 1575 g_io_deliver(bp, ENXIO); 1576 } 1577 mtx_unlock(&sc->sc_queue_mtx); 1578 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1579 sc->sc_provider->name); 1580 sc->sc_provider->flags |= G_PF_WITHER; 1581 g_orphan_provider(sc->sc_provider, ENXIO); 1582 sc->sc_provider = NULL; 1583 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1584 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1585 g_mirror_sync_stop(disk, 1); 1586 } 1587} 1588 1589static void 1590g_mirror_go(void *arg) 1591{ 1592 struct g_mirror_softc *sc; 1593 1594 sc = arg; 1595 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1596 g_mirror_event_send(sc, 0, 1597 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1598} 1599 1600static u_int 1601g_mirror_determine_state(struct g_mirror_disk *disk) 1602{ 1603 struct g_mirror_softc *sc; 1604 u_int state; 1605 1606 sc = disk->d_softc; 1607 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1608 if ((disk->d_flags & 1609 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1610 /* Disk does not need synchronization. */ 1611 state = G_MIRROR_DISK_STATE_ACTIVE; 1612 } else { 1613 if ((sc->sc_flags & 1614 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1615 (disk->d_flags & 1616 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1617 /* 1618 * We can start synchronization from 1619 * the stored offset. 1620 */ 1621 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1622 } else { 1623 state = G_MIRROR_DISK_STATE_STALE; 1624 } 1625 } 1626 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1627 /* 1628 * Reset all synchronization data for this disk, 1629 * because if it even was synchronized, it was 1630 * synchronized to disks with different syncid. 1631 */ 1632 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1633 disk->d_sync.ds_offset = 0; 1634 disk->d_sync.ds_offset_done = 0; 1635 disk->d_sync.ds_syncid = sc->sc_syncid; 1636 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1637 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1638 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1639 } else { 1640 state = G_MIRROR_DISK_STATE_STALE; 1641 } 1642 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1643 /* 1644 * Not good, NOT GOOD! 1645 * It means that mirror was started on stale disks 1646 * and more fresh disk just arrive. 1647 * If there were writes, mirror is fucked up, sorry. 1648 * I think the best choice here is don't touch 1649 * this disk and inform the user laudly. 1650 */ 1651 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1652 "disk (%s) arrives!! It will not be connected to the " 1653 "running device.", sc->sc_name, 1654 g_mirror_get_diskname(disk)); 1655 g_mirror_destroy_disk(disk); 1656 state = G_MIRROR_DISK_STATE_NONE; 1657 /* Return immediately, because disk was destroyed. */ 1658 return (state); 1659 } 1660 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1661 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1662 return (state); 1663} 1664 1665/* 1666 * Update device state. 1667 */ 1668static void 1669g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1670{ 1671 struct g_mirror_disk *disk; 1672 u_int state; 1673 1674 g_topology_assert(); 1675 1676 switch (sc->sc_state) { 1677 case G_MIRROR_DEVICE_STATE_STARTING: 1678 { 1679 struct g_mirror_disk *pdisk; 1680 u_int dirty, ndisks, syncid; 1681 1682 KASSERT(sc->sc_provider == NULL, 1683 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1684 /* 1685 * Are we ready? We are, if all disks are connected or 1686 * if we have any disks and 'force' is true. 1687 */ 1688 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1689 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1690 ; 1691 } else if (g_mirror_ndisks(sc, -1) == 0) { 1692 /* 1693 * Disks went down in starting phase, so destroy 1694 * device. 1695 */ 1696 callout_drain(&sc->sc_callout); 1697 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1698 return; 1699 } else { 1700 return; 1701 } 1702 1703 /* 1704 * Activate all disks with the biggest syncid. 1705 */ 1706 if (force) { 1707 /* 1708 * If called with 'force' true, we're called from 1709 * timeout * procedure, so don't bother canceling 1710 * timeout. 1711 */ 1712 ndisks = 0; 1713 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1714 if ((disk->d_flags & 1715 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1716 ndisks++; 1717 } 1718 } 1719 if (ndisks == 0) { 1720 int timeout; 1721 1722 /* No valid disks still, wait some more. */ 1723 timeout = 1724 atomic_load_acq_int(&g_mirror_timeout); 1725 callout_reset(&sc->sc_callout, timeout * hz, 1726 g_mirror_go, sc); 1727 return; 1728 } 1729 } else { 1730 /* Cancel timeout. */ 1731 callout_drain(&sc->sc_callout); 1732 } 1733 1734 /* 1735 * Find disk with the biggest syncid. 1736 */ 1737 syncid = 0; 1738 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1739 if (disk->d_sync.ds_syncid > syncid) 1740 syncid = disk->d_sync.ds_syncid; 1741 } 1742 1743 /* 1744 * Here we need to look for dirty disks and if all disks 1745 * with the biggest syncid are dirty, we have to choose 1746 * one with the biggest priority and rebuild the rest. 1747 */ 1748 /* 1749 * Find the number of dirty disks with the biggest syncid. 1750 * Find the number of disks with the biggest syncid. 1751 * While here, find a disk with the biggest priority. 1752 */ 1753 dirty = ndisks = 0; 1754 pdisk = NULL; 1755 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1756 if (disk->d_sync.ds_syncid != syncid) 1757 continue; 1758 if ((disk->d_flags & 1759 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1760 continue; 1761 } 1762 ndisks++; 1763 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1764 dirty++; 1765 if (pdisk == NULL || 1766 pdisk->d_priority < disk->d_priority) { 1767 pdisk = disk; 1768 } 1769 } 1770 } 1771 if (dirty == 0) { 1772 /* No dirty disks at all, great. */ 1773 } else if (dirty == ndisks) { 1774 /* 1775 * Force synchronization for all dirty disks except one 1776 * with the biggest priority. 1777 */ 1778 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1779 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1780 "master disk for synchronization.", 1781 g_mirror_get_diskname(pdisk), sc->sc_name); 1782 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1783 if (disk->d_sync.ds_syncid != syncid) 1784 continue; 1785 if ((disk->d_flags & 1786 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1787 continue; 1788 } 1789 KASSERT((disk->d_flags & 1790 G_MIRROR_DISK_FLAG_DIRTY) != 0, 1791 ("Disk %s isn't marked as dirty.", 1792 g_mirror_get_diskname(disk))); 1793 /* Skip the disk with the biggest priority. */ 1794 if (disk == pdisk) 1795 continue; 1796 disk->d_sync.ds_syncid = 0; 1797 } 1798 } else if (dirty < ndisks) { 1799 /* 1800 * Force synchronization for all dirty disks. 1801 * We have some non-dirty disks. 1802 */ 1803 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1804 if (disk->d_sync.ds_syncid != syncid) 1805 continue; 1806 if ((disk->d_flags & 1807 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1808 continue; 1809 } 1810 if ((disk->d_flags & 1811 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1812 continue; 1813 } 1814 disk->d_sync.ds_syncid = 0; 1815 } 1816 } 1817 1818 /* Reset hint. */ 1819 sc->sc_hint = NULL; 1820 sc->sc_syncid = syncid; 1821 if (force) { 1822 /* Remember to bump syncid on first write. */ 1823 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1824 } 1825 state = G_MIRROR_DEVICE_STATE_RUNNING; 1826 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 1827 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 1828 g_mirror_device_state2str(state)); 1829 sc->sc_state = state; 1830 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1831 state = g_mirror_determine_state(disk); 1832 g_mirror_event_send(disk, state, 1833 G_MIRROR_EVENT_DONTWAIT); 1834 if (state == G_MIRROR_DISK_STATE_STALE) { 1835 sc->sc_bump_syncid = 1836 G_MIRROR_BUMP_ON_FIRST_WRITE; 1837 } 1838 } 1839 break; 1840 } 1841 case G_MIRROR_DEVICE_STATE_RUNNING: 1842 /* 1843 * Bump syncid here, if we need to do it immediately. 1844 */ 1845 if (sc->sc_bump_syncid == G_MIRROR_BUMP_IMMEDIATELY) { 1846 sc->sc_bump_syncid = 0; 1847 g_mirror_bump_syncid(sc); 1848 } 1849 if (g_mirror_ndisks(sc, -1) == 0) { 1850 /* 1851 * No disks at all, we need to destroy device. 1852 */ 1853 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1854 } else if (g_mirror_ndisks(sc, 1855 G_MIRROR_DISK_STATE_ACTIVE) == 0 && 1856 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 1857 /* 1858 * No active disks, destroy provider. 1859 */ 1860 if (sc->sc_provider != NULL) 1861 g_mirror_destroy_provider(sc); 1862 } else if (g_mirror_ndisks(sc, 1863 G_MIRROR_DISK_STATE_ACTIVE) > 0) { 1864 /* 1865 * We have active disks, launch provider if it doesn't 1866 * exist. 1867 */ 1868 if (sc->sc_provider == NULL) 1869 g_mirror_launch_provider(sc); 1870 } 1871 break; 1872 default: 1873 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 1874 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 1875 break; 1876 } 1877} 1878 1879/* 1880 * Update disk state and device state if needed. 1881 */ 1882#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 1883 "Disk %s state changed from %s to %s (device %s).", \ 1884 g_mirror_get_diskname(disk), \ 1885 g_mirror_disk_state2str(disk->d_state), \ 1886 g_mirror_disk_state2str(state), sc->sc_name) 1887static int 1888g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 1889{ 1890 struct g_mirror_softc *sc; 1891 1892 g_topology_assert(); 1893 1894 sc = disk->d_softc; 1895again: 1896 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 1897 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 1898 g_mirror_disk_state2str(state)); 1899 switch (state) { 1900 case G_MIRROR_DISK_STATE_NEW: 1901 /* 1902 * Possible scenarios: 1903 * 1. New disk arrive. 1904 */ 1905 /* Previous state should be NONE. */ 1906 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 1907 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1908 g_mirror_disk_state2str(disk->d_state))); 1909 DISK_STATE_CHANGED(); 1910 1911 disk->d_state = state; 1912 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 1913 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 1914 sc->sc_name, g_mirror_get_diskname(disk)); 1915 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 1916 break; 1917 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1918 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 1919 g_mirror_device_state2str(sc->sc_state), 1920 g_mirror_get_diskname(disk), 1921 g_mirror_disk_state2str(disk->d_state))); 1922 state = g_mirror_determine_state(disk); 1923 if (state != G_MIRROR_DISK_STATE_NONE) 1924 goto again; 1925 break; 1926 case G_MIRROR_DISK_STATE_ACTIVE: 1927 /* 1928 * Possible scenarios: 1929 * 1. New disk does not need synchronization. 1930 * 2. Synchronization process finished successfully. 1931 */ 1932 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1933 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 1934 g_mirror_device_state2str(sc->sc_state), 1935 g_mirror_get_diskname(disk), 1936 g_mirror_disk_state2str(disk->d_state))); 1937 /* Previous state should be NEW or SYNCHRONIZING. */ 1938 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 1939 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1940 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1941 g_mirror_disk_state2str(disk->d_state))); 1942 DISK_STATE_CHANGED(); 1943 1944 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 1945 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1946 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1947 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1948 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 1949 g_mirror_sync_stop(disk, 0); 1950 } 1951 disk->d_state = state; 1952 disk->d_sync.ds_offset = 0; 1953 disk->d_sync.ds_offset_done = 0; 1954 g_mirror_update_access(disk); 1955 g_mirror_update_metadata(disk); 1956 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 1957 sc->sc_name, g_mirror_get_diskname(disk)); 1958 break; 1959 case G_MIRROR_DISK_STATE_STALE: 1960 /* 1961 * Possible scenarios: 1962 * 1. Stale disk was connected. 1963 */ 1964 /* Previous state should be NEW. */ 1965 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 1966 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1967 g_mirror_disk_state2str(disk->d_state))); 1968 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1969 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 1970 g_mirror_device_state2str(sc->sc_state), 1971 g_mirror_get_diskname(disk), 1972 g_mirror_disk_state2str(disk->d_state))); 1973 /* 1974 * STALE state is only possible if device is marked 1975 * NOAUTOSYNC. 1976 */ 1977 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 1978 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 1979 g_mirror_device_state2str(sc->sc_state), 1980 g_mirror_get_diskname(disk), 1981 g_mirror_disk_state2str(disk->d_state))); 1982 DISK_STATE_CHANGED(); 1983 1984 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1985 disk->d_state = state; 1986 g_mirror_update_metadata(disk); 1987 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 1988 sc->sc_name, g_mirror_get_diskname(disk)); 1989 break; 1990 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1991 /* 1992 * Possible scenarios: 1993 * 1. Disk which needs synchronization was connected. 1994 */ 1995 /* Previous state should be NEW. */ 1996 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 1997 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1998 g_mirror_disk_state2str(disk->d_state))); 1999 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2000 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2001 g_mirror_device_state2str(sc->sc_state), 2002 g_mirror_get_diskname(disk), 2003 g_mirror_disk_state2str(disk->d_state))); 2004 DISK_STATE_CHANGED(); 2005 2006 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2007 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2008 disk->d_state = state; 2009 if (sc->sc_provider != NULL) { 2010 g_mirror_sync_start(disk); 2011 g_mirror_update_metadata(disk); 2012 } 2013 break; 2014 case G_MIRROR_DISK_STATE_DISCONNECTED: 2015 /* 2016 * Possible scenarios: 2017 * 1. Device wasn't running yet, but disk disappear. 2018 * 2. Disk was active and disapppear. 2019 * 3. Disk disappear during synchronization process. 2020 */ 2021 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2022 /* 2023 * Previous state should be ACTIVE, STALE or 2024 * SYNCHRONIZING. 2025 */ 2026 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2027 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2028 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2029 ("Wrong disk state (%s, %s).", 2030 g_mirror_get_diskname(disk), 2031 g_mirror_disk_state2str(disk->d_state))); 2032 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2033 /* Previous state should be NEW. */ 2034 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2035 ("Wrong disk state (%s, %s).", 2036 g_mirror_get_diskname(disk), 2037 g_mirror_disk_state2str(disk->d_state))); 2038 /* 2039 * Reset bumping syncid if disk disappeared in STARTING 2040 * state. 2041 */ 2042 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) 2043 sc->sc_bump_syncid = 0; 2044#ifdef INVARIANTS 2045 } else { 2046 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2047 sc->sc_name, 2048 g_mirror_device_state2str(sc->sc_state), 2049 g_mirror_get_diskname(disk), 2050 g_mirror_disk_state2str(disk->d_state))); 2051#endif 2052 } 2053 DISK_STATE_CHANGED(); 2054 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2055 sc->sc_name, g_mirror_get_diskname(disk)); 2056 2057 g_mirror_destroy_disk(disk); 2058 break; 2059 case G_MIRROR_DISK_STATE_DESTROY: 2060 { 2061 int error; 2062 2063 error = g_mirror_clear_metadata(disk); 2064 if (error != 0) 2065 return (error); 2066 DISK_STATE_CHANGED(); 2067 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2068 sc->sc_name, g_mirror_get_diskname(disk)); 2069 2070 g_mirror_destroy_disk(disk); 2071 sc->sc_ndisks--; 2072 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2073 g_mirror_update_metadata(disk); 2074 } 2075 break; 2076 } 2077 default: 2078 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2079 break; 2080 } 2081 return (0); 2082} 2083#undef DISK_STATE_CHANGED 2084 2085static int 2086g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2087{ 2088 struct g_provider *pp; 2089 u_char *buf; 2090 int error; 2091 2092 g_topology_assert(); 2093 2094 error = g_access(cp, 1, 0, 0); 2095 if (error != 0) 2096 return (error); 2097 pp = cp->provider; 2098 g_topology_unlock(); 2099 /* Metadata are stored on last sector. */ 2100 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2101 &error); 2102 g_topology_lock(); 2103 if (buf == NULL) { 2104 g_access(cp, -1, 0, 0); 2105 return (error); 2106 } 2107 if (error != 0) { 2108 g_access(cp, -1, 0, 0); 2109 g_free(buf); 2110 return (error); 2111 } 2112 error = g_access(cp, -1, 0, 0); 2113 KASSERT(error == 0, ("Cannot decrease access count for %s.", pp->name)); 2114 2115 /* Decode metadata. */ 2116 error = mirror_metadata_decode(buf, md); 2117 g_free(buf); 2118 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2119 return (EINVAL); 2120 if (error != 0) { 2121 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2122 cp->provider->name); 2123 return (error); 2124 } 2125 2126 return (0); 2127} 2128 2129static int 2130g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2131 struct g_mirror_metadata *md) 2132{ 2133 2134 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2135 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2136 pp->name, md->md_did); 2137 return (EEXIST); 2138 } 2139 if (md->md_all != sc->sc_ndisks) { 2140 G_MIRROR_DEBUG(1, 2141 "Invalid '%s' field on disk %s (device %s), skipping.", 2142 "md_all", pp->name, sc->sc_name); 2143 return (EINVAL); 2144 } 2145 if (md->md_slice != sc->sc_slice) { 2146 G_MIRROR_DEBUG(1, 2147 "Invalid '%s' field on disk %s (device %s), skipping.", 2148 "md_slice", pp->name, sc->sc_name); 2149 return (EINVAL); 2150 } 2151 if (md->md_balance != sc->sc_balance) { 2152 G_MIRROR_DEBUG(1, 2153 "Invalid '%s' field on disk %s (device %s), skipping.", 2154 "md_balance", pp->name, sc->sc_name); 2155 return (EINVAL); 2156 } 2157 if (md->md_mediasize != sc->sc_mediasize) { 2158 G_MIRROR_DEBUG(1, 2159 "Invalid '%s' field on disk %s (device %s), skipping.", 2160 "md_mediasize", pp->name, sc->sc_name); 2161 return (EINVAL); 2162 } 2163 if (sc->sc_mediasize > pp->mediasize) { 2164 G_MIRROR_DEBUG(1, 2165 "Invalid size of disk %s (device %s), skipping.", pp->name, 2166 sc->sc_name); 2167 return (EINVAL); 2168 } 2169 if (md->md_sectorsize != sc->sc_sectorsize) { 2170 G_MIRROR_DEBUG(1, 2171 "Invalid '%s' field on disk %s (device %s), skipping.", 2172 "md_sectorsize", pp->name, sc->sc_name); 2173 return (EINVAL); 2174 } 2175 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2176 G_MIRROR_DEBUG(1, 2177 "Invalid sector size of disk %s (device %s), skipping.", 2178 pp->name, sc->sc_name); 2179 return (EINVAL); 2180 } 2181 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2182 G_MIRROR_DEBUG(1, 2183 "Invalid device flags on disk %s (device %s), skipping.", 2184 pp->name, sc->sc_name); 2185 return (EINVAL); 2186 } 2187 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2188 G_MIRROR_DEBUG(1, 2189 "Invalid disk flags on disk %s (device %s), skipping.", 2190 pp->name, sc->sc_name); 2191 return (EINVAL); 2192 } 2193 return (0); 2194} 2195 2196static int 2197g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2198 struct g_mirror_metadata *md) 2199{ 2200 struct g_mirror_disk *disk; 2201 int error; 2202 2203 g_topology_assert(); 2204 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2205 2206 error = g_mirror_check_metadata(sc, pp, md); 2207 if (error != 0) 2208 return (error); 2209 disk = g_mirror_init_disk(sc, pp, md, &error); 2210 if (disk == NULL) 2211 return (error); 2212 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2213 G_MIRROR_EVENT_WAIT); 2214 return (error); 2215} 2216 2217static int 2218g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2219{ 2220 struct g_mirror_softc *sc; 2221 struct g_mirror_disk *disk; 2222 int dcr, dcw, dce, err, error; 2223 2224 g_topology_assert(); 2225 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2226 acw, ace); 2227 2228 dcr = pp->acr + acr; 2229 dcw = pp->acw + acw; 2230 dce = pp->ace + ace; 2231 2232 /* On first open, grab an extra "exclusive" bit */ 2233 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0) 2234 ace++; 2235 /* ... and let go of it on last close */ 2236 if (dcr == 0 && dcw == 0 && dce == 0) 2237 ace--; 2238 2239 sc = pp->geom->softc; 2240 if (sc == NULL || LIST_EMPTY(&sc->sc_disks)) { 2241 if (acr <= 0 && acw <= 0 && ace <= 0) 2242 return (0); 2243 else 2244 return (ENXIO); 2245 } 2246 error = ENXIO; 2247 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2248 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2249 continue; 2250 err = g_access(disk->d_consumer, acr, acw, ace); 2251 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 2252 g_mirror_get_diskname(disk), acr, acw, ace, err); 2253 if (err == 0) { 2254 /* 2255 * Mark disk as dirty on open and unmark on close. 2256 */ 2257 if (pp->acw == 0 && dcw > 0) { 2258 G_MIRROR_DEBUG(1, 2259 "Disk %s (device %s) marked as dirty.", 2260 g_mirror_get_diskname(disk), sc->sc_name); 2261 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2262 g_mirror_update_metadata(disk); 2263 } else if (pp->acw > 0 && dcw == 0) { 2264 G_MIRROR_DEBUG(1, 2265 "Disk %s (device %s) marked as clean.", 2266 g_mirror_get_diskname(disk), sc->sc_name); 2267 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2268 g_mirror_update_metadata(disk); 2269 } 2270 error = 0; 2271 } else { 2272 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 2273 g_mirror_event_send(disk, 2274 G_MIRROR_DISK_STATE_DISCONNECTED, 2275 G_MIRROR_EVENT_DONTWAIT); 2276 } 2277 } 2278 return (error); 2279} 2280 2281static struct g_geom * 2282g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2283{ 2284 struct g_mirror_softc *sc; 2285 struct g_geom *gp; 2286 int error, timeout; 2287 2288 g_topology_assert(); 2289 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2290 md->md_mid); 2291 2292 /* One disk is minimum. */ 2293 if (md->md_all < 1) 2294 return (NULL); 2295 /* 2296 * Action geom. 2297 */ 2298 gp = g_new_geomf(mp, "%s", md->md_name); 2299 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2300 gp->start = g_mirror_start; 2301 gp->spoiled = g_mirror_orphan; 2302 gp->orphan = g_mirror_orphan; 2303 gp->access = g_mirror_access; 2304 gp->dumpconf = g_mirror_dumpconf; 2305 2306 sc->sc_id = md->md_mid; 2307 sc->sc_slice = md->md_slice; 2308 sc->sc_balance = md->md_balance; 2309 sc->sc_mediasize = md->md_mediasize; 2310 sc->sc_sectorsize = md->md_sectorsize; 2311 sc->sc_ndisks = md->md_all; 2312 sc->sc_flags = md->md_mflags; 2313 sc->sc_bump_syncid = 0; 2314 bioq_init(&sc->sc_queue); 2315 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2316 LIST_INIT(&sc->sc_disks); 2317 TAILQ_INIT(&sc->sc_events); 2318 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2319 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2320 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2321 gp->softc = sc; 2322 sc->sc_geom = gp; 2323 sc->sc_provider = NULL; 2324 /* 2325 * Synchronization geom. 2326 */ 2327 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2328 gp->softc = sc; 2329 gp->spoiled = g_mirror_orphan; 2330 gp->orphan = g_mirror_orphan; 2331 sc->sc_sync.ds_geom = gp; 2332 sc->sc_sync.ds_block = atomic_load_acq_int(&g_mirror_sync_block_size); 2333 sc->sc_sync.ds_ndisks = 0; 2334 sc->sc_sync.ds_zone = uma_zcreate("gmirror:sync", sc->sc_sync.ds_block, 2335 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 2336 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2337 "g_mirror %s", md->md_name); 2338 if (error != 0) { 2339 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2340 sc->sc_name); 2341 uma_zdestroy(sc->sc_sync.ds_zone); 2342 g_destroy_geom(sc->sc_sync.ds_geom); 2343 mtx_destroy(&sc->sc_events_mtx); 2344 mtx_destroy(&sc->sc_queue_mtx); 2345 g_destroy_geom(sc->sc_geom); 2346 free(sc, M_MIRROR); 2347 return (NULL); 2348 } 2349 2350 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2351 2352 /* 2353 * Run timeout. 2354 */ 2355 timeout = atomic_load_acq_int(&g_mirror_timeout); 2356 callout_reset(&sc->sc_callout, timeout * hz, g_mirror_go, sc); 2357 return (sc->sc_geom); 2358} 2359 2360int 2361g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2362{ 2363 struct g_provider *pp; 2364 2365 g_topology_assert(); 2366 2367 if (sc == NULL) 2368 return (ENXIO); 2369 pp = sc->sc_provider; 2370 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2371 if (force) { 2372 G_MIRROR_DEBUG(0, "Device %s is still open, so it " 2373 "can't be definitely removed.", pp->name); 2374 } else { 2375 G_MIRROR_DEBUG(1, 2376 "Device %s is still open (r%dw%de%d).", pp->name, 2377 pp->acr, pp->acw, pp->ace); 2378 return (EBUSY); 2379 } 2380 } 2381 2382 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2383 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2384 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2385 mtx_lock(&sc->sc_queue_mtx); 2386 wakeup(sc); 2387 mtx_unlock(&sc->sc_queue_mtx); 2388 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2389 while (sc->sc_worker != NULL) 2390 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2391 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2392 g_mirror_destroy_device(sc); 2393 free(sc, M_MIRROR); 2394 return (0); 2395} 2396 2397static void 2398g_mirror_taste_orphan(struct g_consumer *cp) 2399{ 2400 2401 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2402 cp->provider->name)); 2403} 2404 2405static struct g_geom * 2406g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2407{ 2408 struct g_mirror_metadata md; 2409 struct g_mirror_softc *sc; 2410 struct g_consumer *cp; 2411 struct g_geom *gp; 2412 int error; 2413 2414 g_topology_assert(); 2415 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2416 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2417 2418 gp = g_new_geomf(mp, "mirror:taste"); 2419 /* 2420 * This orphan function should be never called. 2421 */ 2422 gp->orphan = g_mirror_taste_orphan; 2423 cp = g_new_consumer(gp); 2424 g_attach(cp, pp); 2425 error = g_mirror_read_metadata(cp, &md); 2426 g_detach(cp); 2427 g_destroy_consumer(cp); 2428 g_destroy_geom(gp); 2429 if (error != 0) 2430 return (NULL); 2431 gp = NULL; 2432 2433 if (md.md_version > G_MIRROR_VERSION) { 2434 printf("geom_mirror.ko module is too old to handle %s.\n", 2435 pp->name); 2436 return (NULL); 2437 } 2438 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2439 G_MIRROR_DEBUG(0, 2440 "Device %s: provider %s marked as inactive, skipping.", 2441 md.md_name, pp->name); 2442 return (NULL); 2443 } 2444 if (g_mirror_debug >= 2) 2445 mirror_metadata_dump(&md); 2446 2447 /* 2448 * Let's check if device already exists. 2449 */ 2450 LIST_FOREACH(gp, &mp->geom, geom) { 2451 sc = gp->softc; 2452 if (sc == NULL) 2453 continue; 2454 if (sc->sc_sync.ds_geom == gp) 2455 continue; 2456 if (strcmp(md.md_name, sc->sc_name) != 0) 2457 continue; 2458 if (md.md_mid != sc->sc_id) { 2459 G_MIRROR_DEBUG(0, "Device %s already configured.", 2460 sc->sc_name); 2461 return (NULL); 2462 } 2463 break; 2464 } 2465 if (gp == NULL) { 2466 gp = g_mirror_create(mp, &md); 2467 if (gp == NULL) { 2468 G_MIRROR_DEBUG(0, "Cannot create device %s.mirror", 2469 md.md_name); 2470 return (NULL); 2471 } 2472 sc = gp->softc; 2473 } 2474 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2475 error = g_mirror_add_disk(sc, pp, &md); 2476 if (error != 0) { 2477 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2478 pp->name, gp->name, error); 2479 if (LIST_EMPTY(&sc->sc_disks)) 2480 g_mirror_destroy(sc, 1); 2481 return (NULL); 2482 } 2483 return (gp); 2484} 2485 2486static int 2487g_mirror_destroy_geom(struct gctl_req *req __unused, 2488 struct g_class *mp __unused, struct g_geom *gp) 2489{ 2490 2491 return (g_mirror_destroy(gp->softc, 0)); 2492} 2493 2494static void 2495g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2496 struct g_consumer *cp, struct g_provider *pp) 2497{ 2498 struct g_mirror_softc *sc; 2499 2500 g_topology_assert(); 2501 2502 sc = gp->softc; 2503 if (sc == NULL) 2504 return; 2505 /* Skip synchronization geom. */ 2506 if (gp == sc->sc_sync.ds_geom) 2507 return; 2508 if (pp != NULL) { 2509 /* Nothing here. */ 2510 } else if (cp != NULL) { 2511 struct g_mirror_disk *disk; 2512 2513 disk = cp->private; 2514 if (disk == NULL) 2515 return; 2516 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2517 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2518 sbuf_printf(sb, "%s<Synchronized>", indent); 2519 if (disk->d_sync.ds_offset_done == 0) 2520 sbuf_printf(sb, "0%%"); 2521 else { 2522 sbuf_printf(sb, "%u%%", 2523 (u_int)((disk->d_sync.ds_offset_done * 100) / 2524 sc->sc_provider->mediasize)); 2525 } 2526 sbuf_printf(sb, "</Synchronized>\n"); 2527 } 2528 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2529 disk->d_sync.ds_syncid); 2530 sbuf_printf(sb, "%s<Flags>", indent); 2531 if (disk->d_flags == 0) 2532 sbuf_printf(sb, "NONE"); 2533 else { 2534 int first = 1; 2535 2536 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2537 if (!first) 2538 sbuf_printf(sb, ", "); 2539 else 2540 first = 0; 2541 sbuf_printf(sb, "DIRTY"); 2542 } 2543 if ((disk->d_flags & 2544 G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2545 if (!first) 2546 sbuf_printf(sb, ", "); 2547 else 2548 first = 0; 2549 sbuf_printf(sb, "INACTIVE"); 2550 } 2551 if ((disk->d_flags & 2552 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2553 if (!first) 2554 sbuf_printf(sb, ", "); 2555 else 2556 first = 0; 2557 sbuf_printf(sb, "SYNCHRONIZING"); 2558 } 2559 if ((disk->d_flags & 2560 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2561 if (!first) 2562 sbuf_printf(sb, ", "); 2563 else 2564 first = 0; 2565 sbuf_printf(sb, "FORCE_SYNC"); 2566 } 2567 } 2568 sbuf_printf(sb, "</Flags>\n"); 2569 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2570 disk->d_priority); 2571 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2572 g_mirror_disk_state2str(disk->d_state)); 2573 } else { 2574 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2575 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2576 sbuf_printf(sb, "%s<Flags>", indent); 2577 if (sc->sc_flags == 0) 2578 sbuf_printf(sb, "NONE"); 2579 else { 2580 int first = 1; 2581 2582 if ((sc->sc_flags & 2583 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0) { 2584 if (!first) 2585 sbuf_printf(sb, ", "); 2586 else 2587 first = 0; 2588 sbuf_printf(sb, "NOAUTOSYNC"); 2589 } 2590 } 2591 sbuf_printf(sb, "</Flags>\n"); 2592 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2593 (u_int)sc->sc_slice); 2594 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2595 balance_name(sc->sc_balance)); 2596 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2597 sc->sc_ndisks); 2598 } 2599} 2600 2601DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2602