g_mirror.c revision 133114
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 133114 2004-08-04 10:02:06Z pjd $"); 29 30#include <sys/param.h> 31#include <sys/systm.h> 32#include <sys/kernel.h> 33#include <sys/module.h> 34#include <sys/limits.h> 35#include <sys/lock.h> 36#include <sys/mutex.h> 37#include <sys/bio.h> 38#include <sys/sysctl.h> 39#include <sys/malloc.h> 40#include <sys/bitstring.h> 41#include <vm/uma.h> 42#include <machine/atomic.h> 43#include <geom/geom.h> 44#include <sys/proc.h> 45#include <sys/kthread.h> 46#include <geom/mirror/g_mirror.h> 47 48 49static MALLOC_DEFINE(M_MIRROR, "mirror data", "GEOM_MIRROR Data"); 50 51SYSCTL_DECL(_kern_geom); 52SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, "GEOM_MIRROR stuff"); 53u_int g_mirror_debug = 0; 54SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RW, &g_mirror_debug, 0, 55 "Debug level"); 56static u_int g_mirror_sync_block_size = 131072; 57SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_block_size, CTLFLAG_RW, 58 &g_mirror_sync_block_size, 0, "Synchronization block size"); 59static u_int g_mirror_timeout = 8; 60SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RW, &g_mirror_timeout, 61 0, "Time to wait on all mirror components"); 62static u_int g_mirror_reqs_per_sync = 5; 63SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, reqs_per_sync, CTLFLAG_RW, 64 &g_mirror_reqs_per_sync, 0, 65 "Number of regular I/O requests per synchronization request"); 66static u_int g_mirror_syncs_per_sec = 100; 67SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, syncs_per_sec, CTLFLAG_RW, 68 &g_mirror_syncs_per_sec, 0, 69 "Number of synchronizations requests per second"); 70 71#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 72 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 73 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 74 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 75} while (0) 76 77 78static int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp, 79 struct g_geom *gp); 80static g_taste_t g_mirror_taste; 81 82struct g_class g_mirror_class = { 83 .name = G_MIRROR_CLASS_NAME, 84 .ctlreq = g_mirror_config, 85 .taste = g_mirror_taste, 86 .destroy_geom = g_mirror_destroy_geom 87}; 88 89 90static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 91static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 92static void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force); 93static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 94 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 95static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 96 97 98static const char * 99g_mirror_disk_state2str(int state) 100{ 101 102 switch (state) { 103 case G_MIRROR_DISK_STATE_NONE: 104 return ("NONE"); 105 case G_MIRROR_DISK_STATE_NEW: 106 return ("NEW"); 107 case G_MIRROR_DISK_STATE_ACTIVE: 108 return ("ACTIVE"); 109 case G_MIRROR_DISK_STATE_STALE: 110 return ("STALE"); 111 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 112 return ("SYNCHRONIZING"); 113 case G_MIRROR_DISK_STATE_DISCONNECTED: 114 return ("DISCONNECTED"); 115 case G_MIRROR_DISK_STATE_DESTROY: 116 return ("DESTROY"); 117 default: 118 return ("INVALID"); 119 } 120} 121 122static const char * 123g_mirror_device_state2str(int state) 124{ 125 126 switch (state) { 127 case G_MIRROR_DEVICE_STATE_STARTING: 128 return ("STARTING"); 129 case G_MIRROR_DEVICE_STATE_RUNNING: 130 return ("RUNNING"); 131 default: 132 return ("INVALID"); 133 } 134} 135 136static const char * 137g_mirror_get_diskname(struct g_mirror_disk *disk) 138{ 139 140 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 141 return ("[unknown]"); 142 return (disk->d_name); 143} 144 145/* 146 * --- Events handling functions --- 147 * Events in geom_mirror are used to maintain disks and device status 148 * from one thread to simplify locking. 149 */ 150static void 151g_mirror_event_free(struct g_mirror_event *ep) 152{ 153 154 free(ep, M_MIRROR); 155} 156 157int 158g_mirror_event_send(void *arg, int state, int flags) 159{ 160 struct g_mirror_softc *sc; 161 struct g_mirror_disk *disk; 162 struct g_mirror_event *ep; 163 int error; 164 165 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 166 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 167 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 168 disk = NULL; 169 sc = arg; 170 } else { 171 disk = arg; 172 sc = disk->d_softc; 173 } 174 ep->e_disk = disk; 175 ep->e_state = state; 176 ep->e_flags = flags; 177 ep->e_error = 0; 178 mtx_lock(&sc->sc_events_mtx); 179 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 180 mtx_unlock(&sc->sc_events_mtx); 181 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 182 mtx_lock(&sc->sc_queue_mtx); 183 wakeup(sc); 184 mtx_unlock(&sc->sc_queue_mtx); 185 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 186 return (0); 187 g_topology_assert(); 188 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 189 g_topology_unlock(); 190 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 191 mtx_lock(&sc->sc_events_mtx); 192 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 193 hz * 5); 194 } 195 /* Don't even try to use 'sc' here, because it could be already dead. */ 196 g_topology_lock(); 197 error = ep->e_error; 198 g_mirror_event_free(ep); 199 return (error); 200} 201 202static struct g_mirror_event * 203g_mirror_event_get(struct g_mirror_softc *sc) 204{ 205 struct g_mirror_event *ep; 206 207 mtx_lock(&sc->sc_events_mtx); 208 ep = TAILQ_FIRST(&sc->sc_events); 209 if (ep != NULL) 210 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 211 mtx_unlock(&sc->sc_events_mtx); 212 return (ep); 213} 214 215static void 216g_mirror_event_cancel(struct g_mirror_disk *disk) 217{ 218 struct g_mirror_softc *sc; 219 struct g_mirror_event *ep, *tmpep; 220 221 g_topology_assert(); 222 223 sc = disk->d_softc; 224 mtx_lock(&sc->sc_events_mtx); 225 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 226 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 227 continue; 228 if (ep->e_disk != disk) 229 continue; 230 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 231 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 232 g_mirror_event_free(ep); 233 else { 234 ep->e_error = ECANCELED; 235 wakeup(ep); 236 } 237 } 238 mtx_unlock(&sc->sc_events_mtx); 239} 240 241/* 242 * Return the number of disks in given state. 243 * If state is equal to -1, count all connected disks. 244 */ 245u_int 246g_mirror_ndisks(struct g_mirror_softc *sc, int state) 247{ 248 struct g_mirror_disk *disk; 249 u_int n = 0; 250 251 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 252 if (state == -1 || disk->d_state == state) 253 n++; 254 } 255 return (n); 256} 257 258/* 259 * Find a disk in mirror by its disk ID. 260 */ 261static struct g_mirror_disk * 262g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 263{ 264 struct g_mirror_disk *disk; 265 266 g_topology_assert(); 267 268 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 269 if (disk->d_id == id) 270 return (disk); 271 } 272 return (NULL); 273} 274 275static u_int 276g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 277{ 278 struct bio *bp; 279 u_int nreqs = 0; 280 281 mtx_lock(&sc->sc_queue_mtx); 282 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 283 if (bp->bio_from == cp) 284 nreqs++; 285 } 286 mtx_unlock(&sc->sc_queue_mtx); 287 return (nreqs); 288} 289 290static void 291g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 292{ 293 294 g_topology_assert(); 295 296 cp->private = NULL; 297 if (cp->nstart != cp->nend) { 298 G_MIRROR_DEBUG(2, 299 "I/O requests for %s exist, can't destroy it now.", 300 cp->provider->name); 301 return; 302 } 303 if (g_mirror_nrequests(sc, cp) > 0) { 304 G_MIRROR_DEBUG(2, 305 "I/O requests for %s in queue, can't destroy it now.", 306 cp->provider->name); 307 return; 308 } 309 G_MIRROR_DEBUG(2, "Consumer %s destroyed.", cp->provider->name); 310 g_detach(cp); 311 g_destroy_consumer(cp); 312} 313 314static int 315g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 316{ 317 int error; 318 319 g_topology_assert(); 320 KASSERT(disk->d_consumer == NULL, 321 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 322 323 disk->d_consumer = g_new_consumer(disk->d_softc->sc_geom); 324 disk->d_consumer->private = disk; 325 error = g_attach(disk->d_consumer, pp); 326 if (error != 0) 327 return (error); 328 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 329 return (0); 330} 331 332static void 333g_mirror_disconnect_disk(struct g_mirror_disk *disk) 334{ 335 struct g_consumer *cp; 336 337 g_topology_assert(); 338 339 cp = disk->d_consumer; 340 if (cp == NULL) 341 return; 342 if (cp->provider != NULL) { 343 G_MIRROR_DEBUG(2, "Disk %s disconnected.", 344 g_mirror_get_diskname(disk)); 345 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) { 346 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 347 cp->provider->name, -cp->acr, -cp->acw, -cp->ace, 348 0); 349 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 350 } 351 g_mirror_kill_consumer(disk->d_softc, cp); 352 } else { 353 g_destroy_consumer(cp); 354 } 355} 356 357/* 358 * Initialize disk. This means allocate memory, create consumer, attach it 359 * to the provider and open access (r1w1e1) to it. 360 */ 361static struct g_mirror_disk * 362g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 363 struct g_mirror_metadata *md, int *errorp) 364{ 365 struct g_mirror_disk *disk; 366 int error; 367 368 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 369 if (disk == NULL) { 370 error = ENOMEM; 371 goto fail; 372 } 373 disk->d_softc = sc; 374 error = g_mirror_connect_disk(disk, pp); 375 if (error != 0) 376 goto fail; 377 disk->d_id = md->md_did; 378 disk->d_state = G_MIRROR_DISK_STATE_NONE; 379 disk->d_priority = md->md_priority; 380 disk->d_delay.sec = 0; 381 disk->d_delay.frac = 0; 382 binuptime(&disk->d_last_used); 383 disk->d_flags = md->md_dflags; 384 disk->d_sync.ds_consumer = NULL; 385 disk->d_sync.ds_offset = md->md_sync_offset; 386 disk->d_sync.ds_offset_done = md->md_sync_offset; 387 disk->d_sync.ds_syncid = md->md_syncid; 388 if (errorp != NULL) 389 *errorp = 0; 390 return (disk); 391fail: 392 if (errorp != NULL) 393 *errorp = error; 394 if (disk != NULL) { 395 g_mirror_disconnect_disk(disk); 396 free(disk, M_MIRROR); 397 } 398 return (NULL); 399} 400 401static void 402g_mirror_destroy_disk(struct g_mirror_disk *disk) 403{ 404 struct g_mirror_softc *sc; 405 406 g_topology_assert(); 407 408 LIST_REMOVE(disk, d_next); 409 g_mirror_event_cancel(disk); 410 sc = disk->d_softc; 411 if (sc->sc_hint == disk) 412 sc->sc_hint = NULL; 413 switch (disk->d_state) { 414 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 415 g_mirror_sync_stop(disk, 1); 416 /* FALLTHROUGH */ 417 case G_MIRROR_DISK_STATE_NEW: 418 case G_MIRROR_DISK_STATE_STALE: 419 case G_MIRROR_DISK_STATE_ACTIVE: 420 g_mirror_disconnect_disk(disk); 421 free(disk, M_MIRROR); 422 break; 423 default: 424 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 425 g_mirror_get_diskname(disk), 426 g_mirror_disk_state2str(disk->d_state))); 427 } 428} 429 430static void 431g_mirror_destroy_device(struct g_mirror_softc *sc) 432{ 433 struct g_mirror_disk *disk; 434 struct g_mirror_event *ep; 435 struct g_geom *gp; 436 struct g_consumer *cp; 437 438 g_topology_assert(); 439 440 gp = sc->sc_geom; 441 if (sc->sc_provider != NULL) 442 g_mirror_destroy_provider(sc); 443 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 444 disk = LIST_FIRST(&sc->sc_disks)) { 445 g_mirror_destroy_disk(disk); 446 } 447 while ((ep = g_mirror_event_get(sc)) != NULL) { 448 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 449 g_mirror_event_free(ep); 450 else { 451 ep->e_error = ECANCELED; 452 ep->e_flags |= G_MIRROR_EVENT_DONE; 453 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 454 mtx_lock(&sc->sc_events_mtx); 455 wakeup(ep); 456 mtx_unlock(&sc->sc_events_mtx); 457 } 458 } 459 callout_drain(&sc->sc_callout); 460 gp->softc = NULL; 461 uma_zdestroy(sc->sc_sync.ds_zone); 462 while ((cp = LIST_FIRST(&sc->sc_sync.ds_geom->consumer)) != NULL) { 463 if (cp->provider != NULL) { 464 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 465 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 466 g_detach(cp); 467 } 468 g_destroy_consumer(cp); 469 } 470 sc->sc_sync.ds_geom->softc = NULL; 471 g_destroy_geom(sc->sc_sync.ds_geom); 472 mtx_destroy(&sc->sc_queue_mtx); 473 mtx_destroy(&sc->sc_events_mtx); 474 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 475 g_wither_geom(gp, ENXIO); 476} 477 478static void 479g_mirror_orphan(struct g_consumer *cp) 480{ 481 struct g_mirror_disk *disk; 482 483 g_topology_assert(); 484 485 disk = cp->private; 486 if (disk == NULL) 487 return; 488 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 489 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 490 G_MIRROR_EVENT_DONTWAIT); 491} 492 493static void 494g_mirror_spoiled(struct g_consumer *cp) 495{ 496 struct g_mirror_disk *disk; 497 498 g_topology_assert(); 499 500 disk = cp->private; 501 if (disk == NULL) 502 return; 503 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 504 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 505 G_MIRROR_EVENT_DONTWAIT); 506} 507 508/* 509 * Function should return the next active disk on the list. 510 * It is possible that it will be the same disk as given. 511 * If there are no active disks on list, NULL is returned. 512 */ 513static __inline struct g_mirror_disk * 514g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 515{ 516 struct g_mirror_disk *dp; 517 518 for (dp = LIST_NEXT(disk, d_next); dp != disk; 519 dp = LIST_NEXT(dp, d_next)) { 520 if (dp == NULL) 521 dp = LIST_FIRST(&sc->sc_disks); 522 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 523 break; 524 } 525 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 526 return (NULL); 527 return (dp); 528} 529 530static struct g_mirror_disk * 531g_mirror_get_disk(struct g_mirror_softc *sc) 532{ 533 struct g_mirror_disk *disk; 534 535 if (sc->sc_hint == NULL) { 536 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 537 if (sc->sc_hint == NULL) 538 return (NULL); 539 } 540 disk = sc->sc_hint; 541 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 542 disk = g_mirror_find_next(sc, disk); 543 if (disk == NULL) 544 return (NULL); 545 } 546 sc->sc_hint = g_mirror_find_next(sc, disk); 547 return (disk); 548} 549 550static int 551g_mirror_clear_metadata(struct g_mirror_disk *disk) 552{ 553 struct g_mirror_softc *sc; 554 struct g_consumer *cp; 555 off_t offset, length; 556 u_char *sector; 557 int close = 0, error = 0; 558 559 g_topology_assert(); 560 561 sc = disk->d_softc; 562 cp = disk->d_consumer; 563 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 564 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 565 length = cp->provider->sectorsize; 566 offset = cp->provider->mediasize - length; 567 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 568 /* 569 * Open consumer if it wasn't opened and remember to close it. 570 */ 571 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 572 error = g_access(cp, 0, 1, 1); 573 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 574 cp->provider->name, 0, 1, 1, error); 575 if (error == 0) 576 close = 1; 577#ifdef INVARIANTS 578 } else { 579 KASSERT(cp->acw > 0 && cp->ace > 0, 580 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 581 cp->acr, cp->acw, cp->ace)); 582#endif 583 } 584 if (error == 0) { 585 g_topology_unlock(); 586 error = g_write_data(cp, offset, sector, length); 587 g_topology_lock(); 588 } 589 free(sector, M_MIRROR); 590 if (close) { 591 g_access(cp, 0, -1, -1); 592 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 593 cp->provider->name, 0, -1, -1, 0); 594 } 595 if (error != 0) { 596 G_MIRROR_DEBUG(0, "Cannot clear metadata on disk %s.", 597 g_mirror_get_diskname(disk)); 598 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 599 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 600 G_MIRROR_EVENT_DONTWAIT); 601 return (error); 602 } 603 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 604 g_mirror_get_diskname(disk)); 605 return (0); 606} 607 608void 609g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 610 struct g_mirror_metadata *md) 611{ 612 613 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 614 md->md_version = G_MIRROR_VERSION; 615 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 616 md->md_mid = sc->sc_id; 617 md->md_all = sc->sc_ndisks; 618 md->md_slice = sc->sc_slice; 619 md->md_balance = sc->sc_balance; 620 md->md_mediasize = sc->sc_mediasize; 621 md->md_sectorsize = sc->sc_sectorsize; 622 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 623 if (disk == NULL) { 624 md->md_did = arc4random(); 625 md->md_priority = 0; 626 md->md_syncid = 0; 627 md->md_dflags = 0; 628 md->md_sync_offset = 0; 629 } else { 630 md->md_did = disk->d_id; 631 md->md_priority = disk->d_priority; 632 md->md_syncid = disk->d_sync.ds_syncid; 633 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 634 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 635 md->md_sync_offset = disk->d_sync.ds_offset_done; 636 else 637 md->md_sync_offset = 0; 638 } 639} 640 641void 642g_mirror_update_metadata(struct g_mirror_disk *disk) 643{ 644 struct g_mirror_softc *sc; 645 struct g_mirror_metadata md; 646 struct g_consumer *cp; 647 off_t offset, length; 648 u_char *sector; 649 int close = 0, error = 0; 650 651 g_topology_assert(); 652 653 sc = disk->d_softc; 654 cp = disk->d_consumer; 655 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 656 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 657 length = cp->provider->sectorsize; 658 offset = cp->provider->mediasize - length; 659 sector = malloc((size_t)length, M_MIRROR, M_WAITOK); 660 /* 661 * Open consumer if it wasn't opened and remember to close it. 662 */ 663 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 664 error = g_access(cp, 0, 1, 1); 665 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 666 cp->provider->name, 0, 1, 1, error); 667 if (error == 0) 668 close = 1; 669#ifdef INVARIANTS 670 } else { 671 KASSERT(cp->acw > 0 && cp->ace > 0, 672 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 673 cp->acr, cp->acw, cp->ace)); 674#endif 675 } 676 if (error == 0) { 677 g_mirror_fill_metadata(sc, disk, &md); 678 mirror_metadata_encode(&md, sector); 679 g_topology_unlock(); 680 error = g_write_data(cp, offset, sector, length); 681 g_topology_lock(); 682 } 683 free(sector, M_MIRROR); 684 if (close) { 685 g_access(cp, 0, -1, -1); 686 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 687 cp->provider->name, 0, -1, -1, 0); 688 } 689 if (error != 0) { 690 G_MIRROR_DEBUG(0, 691 "Cannot update metadata on disk %s (error=%d).", 692 g_mirror_get_diskname(disk), error); 693 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 694 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 695 G_MIRROR_EVENT_DONTWAIT); 696 return; 697 } 698 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 699 g_mirror_get_diskname(disk)); 700} 701 702static void 703g_mirror_bump_syncid(struct g_mirror_softc *sc) 704{ 705 struct g_mirror_disk *disk; 706 707 g_topology_assert(); 708 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 709 ("%s called with no active disks (device=%s).", __func__, 710 sc->sc_name)); 711 712 sc->sc_syncid++; 713 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 714 sc->sc_syncid); 715 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 716 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 717 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 718 disk->d_sync.ds_syncid = sc->sc_syncid; 719 g_mirror_update_metadata(disk); 720 } 721 } 722} 723 724static __inline int 725bintime_cmp(struct bintime *bt1, struct bintime *bt2) 726{ 727 728 if (bt1->sec < bt2->sec) 729 return (-1); 730 else if (bt1->sec > bt2->sec) 731 return (1); 732 if (bt1->frac < bt2->frac) 733 return (-1); 734 else if (bt1->frac > bt2->frac) 735 return (1); 736 return (0); 737} 738 739static void 740g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 741{ 742 743 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 744 return; 745 binuptime(&disk->d_delay); 746 bintime_sub(&disk->d_delay, &bp->bio_t0); 747} 748 749static void 750g_mirror_done(struct bio *bp) 751{ 752 struct g_mirror_softc *sc; 753 754 sc = bp->bio_from->geom->softc; 755 bp->bio_flags = BIO_FLAG1; 756 mtx_lock(&sc->sc_queue_mtx); 757 bioq_disksort(&sc->sc_queue, bp); 758 wakeup(sc); 759 mtx_unlock(&sc->sc_queue_mtx); 760} 761 762static void 763g_mirror_regular_request(struct bio *bp) 764{ 765 struct g_mirror_softc *sc; 766 struct g_mirror_disk *disk; 767 struct bio *pbp; 768 769 g_topology_assert_not(); 770 771 pbp = bp->bio_parent; 772 sc = pbp->bio_to->geom->softc; 773 disk = bp->bio_from->private; 774 if (disk == NULL) { 775 g_topology_lock(); 776 g_mirror_kill_consumer(sc, bp->bio_from); 777 g_topology_unlock(); 778 } else { 779 g_mirror_update_delay(disk, bp); 780 } 781 782 pbp->bio_inbed++; 783 KASSERT(pbp->bio_inbed <= pbp->bio_children, 784 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 785 pbp->bio_children)); 786 if (bp->bio_error == 0 && pbp->bio_error == 0) { 787 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 788 g_destroy_bio(bp); 789 if (pbp->bio_children == pbp->bio_inbed) { 790 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 791 pbp->bio_completed = pbp->bio_length; 792 g_io_deliver(pbp, pbp->bio_error); 793 } 794 return; 795 } else if (bp->bio_error != 0) { 796 if (pbp->bio_error == 0) 797 pbp->bio_error = bp->bio_error; 798 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 799 bp->bio_error); 800 if (disk != NULL) { 801 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 802 g_mirror_event_send(disk, 803 G_MIRROR_DISK_STATE_DISCONNECTED, 804 G_MIRROR_EVENT_DONTWAIT); 805 } 806 switch (pbp->bio_cmd) { 807 case BIO_DELETE: 808 case BIO_WRITE: 809 pbp->bio_inbed--; 810 pbp->bio_children--; 811 break; 812 } 813 } 814 g_destroy_bio(bp); 815 816 switch (pbp->bio_cmd) { 817 case BIO_READ: 818 if (pbp->bio_children == pbp->bio_inbed) { 819 pbp->bio_error = 0; 820 mtx_lock(&sc->sc_queue_mtx); 821 bioq_disksort(&sc->sc_queue, pbp); 822 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 823 wakeup(sc); 824 mtx_unlock(&sc->sc_queue_mtx); 825 } 826 break; 827 case BIO_DELETE: 828 case BIO_WRITE: 829 if (pbp->bio_children == 0) { 830 /* 831 * All requests failed. 832 */ 833 } else if (pbp->bio_inbed < pbp->bio_children) { 834 /* Do nothing. */ 835 break; 836 } else if (pbp->bio_children == pbp->bio_inbed) { 837 /* Some requests succeeded. */ 838 pbp->bio_error = 0; 839 pbp->bio_completed = pbp->bio_length; 840 } 841 g_io_deliver(pbp, pbp->bio_error); 842 break; 843 default: 844 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 845 break; 846 } 847} 848 849static void 850g_mirror_sync_done(struct bio *bp) 851{ 852 struct g_mirror_softc *sc; 853 854 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 855 sc = bp->bio_from->geom->softc; 856 bp->bio_flags = BIO_FLAG2; 857 mtx_lock(&sc->sc_queue_mtx); 858 bioq_disksort(&sc->sc_queue, bp); 859 wakeup(sc); 860 mtx_unlock(&sc->sc_queue_mtx); 861} 862 863static void 864g_mirror_start(struct bio *bp) 865{ 866 struct g_mirror_softc *sc; 867 868 sc = bp->bio_to->geom->softc; 869 /* 870 * If sc == NULL or there are no valid disks, provider's error 871 * should be set and g_mirror_start() should not be called at all. 872 */ 873 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 874 ("Provider's error should be set (error=%d)(mirror=%s).", 875 bp->bio_to->error, bp->bio_to->name)); 876 G_MIRROR_LOGREQ(3, bp, "Request received."); 877 878 switch (bp->bio_cmd) { 879 case BIO_READ: 880 case BIO_WRITE: 881 case BIO_DELETE: 882 break; 883 case BIO_GETATTR: 884 default: 885 g_io_deliver(bp, EOPNOTSUPP); 886 return; 887 } 888 mtx_lock(&sc->sc_queue_mtx); 889 bioq_disksort(&sc->sc_queue, bp); 890 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 891 wakeup(sc); 892 mtx_unlock(&sc->sc_queue_mtx); 893} 894 895/* 896 * Send one synchronization request. 897 */ 898static void 899g_mirror_sync_one(struct g_mirror_disk *disk) 900{ 901 struct g_mirror_softc *sc; 902 struct bio *bp; 903 904 sc = disk->d_softc; 905 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 906 ("Disk %s is not marked for synchronization.", 907 g_mirror_get_diskname(disk))); 908 909 bp = g_new_bio(); 910 if (bp == NULL) 911 return; 912 bp->bio_parent = NULL; 913 bp->bio_cmd = BIO_READ; 914 bp->bio_offset = disk->d_sync.ds_offset; 915 bp->bio_length = MIN(sc->sc_sync.ds_block, 916 sc->sc_mediasize - bp->bio_offset); 917 bp->bio_flags = 0; 918 bp->bio_done = g_mirror_sync_done; 919 bp->bio_data = uma_zalloc(sc->sc_sync.ds_zone, M_NOWAIT | M_ZERO); 920 if (bp->bio_data == NULL) { 921 g_destroy_bio(bp); 922 return; 923 } 924 disk->d_sync.ds_offset += bp->bio_length; 925 bp->bio_to = sc->sc_provider; 926 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 927 g_io_request(bp, disk->d_sync.ds_consumer); 928} 929 930static void 931g_mirror_sync_request(struct bio *bp) 932{ 933 struct g_mirror_softc *sc; 934 struct g_mirror_disk *disk; 935 936 sc = bp->bio_from->geom->softc; 937 disk = bp->bio_from->private; 938 if (disk == NULL) { 939 g_topology_lock(); 940 g_mirror_kill_consumer(sc, bp->bio_from); 941 g_topology_unlock(); 942 uma_zfree(sc->sc_sync.ds_zone, bp->bio_data); 943 g_destroy_bio(bp); 944 return; 945 } 946 947 /* 948 * Synchronization request. 949 */ 950 switch (bp->bio_cmd) { 951 case BIO_READ: 952 { 953 struct g_consumer *cp; 954 955 if (bp->bio_error != 0) { 956 G_MIRROR_LOGREQ(0, bp, 957 "Synchronization request failed (error=%d).", 958 bp->bio_error); 959 uma_zfree(sc->sc_sync.ds_zone, bp->bio_data); 960 g_destroy_bio(bp); 961 return; 962 } 963 bp->bio_cmd = BIO_WRITE; 964 bp->bio_flags = 0; 965 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 966 cp = disk->d_consumer; 967 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 968 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 969 cp->acr, cp->acw, cp->ace)); 970 g_io_request(bp, cp); 971 return; 972 } 973 case BIO_WRITE: 974 uma_zfree(sc->sc_sync.ds_zone, bp->bio_data); 975 if (bp->bio_error != 0) { 976 G_MIRROR_LOGREQ(0, bp, 977 "Synchronization request failed (error=%d).", 978 bp->bio_error); 979 g_destroy_bio(bp); 980 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 981 g_mirror_event_send(disk, 982 G_MIRROR_DISK_STATE_DISCONNECTED, 983 G_MIRROR_EVENT_DONTWAIT); 984 return; 985 } 986 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 987 g_destroy_bio(bp); 988 disk->d_sync.ds_offset_done = bp->bio_offset + bp->bio_length; 989 if (bp->bio_offset + bp->bio_length == 990 sc->sc_provider->mediasize) { 991 /* 992 * Disk up-to-date, activate it. 993 */ 994 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 995 G_MIRROR_EVENT_DONTWAIT); 996 return; 997 } else if ((disk->d_sync.ds_offset_done % 998 (sc->sc_sync.ds_block * 100)) == 0) { 999 /* 1000 * Update offset_done on every 100 blocks. 1001 * XXX: This should be configurable. 1002 */ 1003 g_topology_lock(); 1004 g_mirror_update_metadata(disk); 1005 g_topology_unlock(); 1006 } 1007 return; 1008 default: 1009 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1010 bp->bio_cmd, sc->sc_name)); 1011 break; 1012 } 1013} 1014 1015static void 1016g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1017{ 1018 struct g_mirror_disk *disk; 1019 struct g_consumer *cp; 1020 struct bio *cbp; 1021 1022 disk = g_mirror_get_disk(sc); 1023 if (disk == NULL) { 1024 if (bp->bio_error == 0) 1025 bp->bio_error = ENXIO; 1026 g_io_deliver(bp, bp->bio_error); 1027 return; 1028 } 1029 cbp = g_clone_bio(bp); 1030 if (cbp == NULL) { 1031 if (bp->bio_error == 0) 1032 bp->bio_error = ENOMEM; 1033 g_io_deliver(bp, bp->bio_error); 1034 return; 1035 } 1036 /* 1037 * Fill in the component buf structure. 1038 */ 1039 cp = disk->d_consumer; 1040 cbp->bio_done = g_mirror_done; 1041 cbp->bio_to = cp->provider; 1042 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1043 KASSERT(cp->acr > 0 && cp->ace > 0, 1044 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1045 cp->acw, cp->ace)); 1046 g_io_request(cbp, cp); 1047} 1048 1049static void 1050g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1051{ 1052 struct g_mirror_disk *disk, *dp; 1053 struct g_consumer *cp; 1054 struct bio *cbp; 1055 struct bintime curtime; 1056 1057 binuptime(&curtime); 1058 /* 1059 * Find a disk which the smallest load. 1060 */ 1061 disk = NULL; 1062 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1063 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1064 continue; 1065 /* If disk wasn't used for more than 2 sec, use it. */ 1066 if (curtime.sec - dp->d_last_used.sec >= 2) { 1067 disk = dp; 1068 break; 1069 } 1070 if (disk == NULL || 1071 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1072 disk = dp; 1073 } 1074 } 1075 cbp = g_clone_bio(bp); 1076 if (cbp == NULL) { 1077 if (bp->bio_error == 0) 1078 bp->bio_error = ENOMEM; 1079 g_io_deliver(bp, bp->bio_error); 1080 return; 1081 } 1082 /* 1083 * Fill in the component buf structure. 1084 */ 1085 cp = disk->d_consumer; 1086 cbp->bio_done = g_mirror_done; 1087 cbp->bio_to = cp->provider; 1088 binuptime(&disk->d_last_used); 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_split(struct g_mirror_softc *sc, struct bio *bp) 1098{ 1099 struct bio_queue_head queue; 1100 struct g_mirror_disk *disk; 1101 struct g_consumer *cp; 1102 struct bio *cbp; 1103 off_t left, mod, offset, slice; 1104 u_char *data; 1105 u_int ndisks; 1106 1107 if (bp->bio_length <= sc->sc_slice) { 1108 g_mirror_request_round_robin(sc, bp); 1109 return; 1110 } 1111 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1112 slice = bp->bio_length / ndisks; 1113 mod = slice % sc->sc_provider->sectorsize; 1114 if (mod != 0) 1115 slice += sc->sc_provider->sectorsize - mod; 1116 /* 1117 * Allocate all bios before sending any request, so we can 1118 * return ENOMEM in nice and clean way. 1119 */ 1120 left = bp->bio_length; 1121 offset = bp->bio_offset; 1122 data = bp->bio_data; 1123 bioq_init(&queue); 1124 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1125 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1126 continue; 1127 cbp = g_clone_bio(bp); 1128 if (cbp == NULL) { 1129 for (cbp = bioq_first(&queue); cbp != NULL; 1130 cbp = bioq_first(&queue)) { 1131 bioq_remove(&queue, cbp); 1132 g_destroy_bio(cbp); 1133 } 1134 if (bp->bio_error == 0) 1135 bp->bio_error = ENOMEM; 1136 g_io_deliver(bp, bp->bio_error); 1137 return; 1138 } 1139 bioq_insert_tail(&queue, cbp); 1140 cbp->bio_done = g_mirror_done; 1141 cbp->bio_caller1 = disk; 1142 cbp->bio_to = disk->d_consumer->provider; 1143 cbp->bio_offset = offset; 1144 cbp->bio_data = data; 1145 cbp->bio_length = MIN(left, slice); 1146 left -= cbp->bio_length; 1147 if (left == 0) 1148 break; 1149 offset += cbp->bio_length; 1150 data += cbp->bio_length; 1151 } 1152 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1153 bioq_remove(&queue, cbp); 1154 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1155 disk = cbp->bio_caller1; 1156 cbp->bio_caller1 = NULL; 1157 cp = disk->d_consumer; 1158 KASSERT(cp->acr > 0 && cp->ace > 0, 1159 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1160 cp->acr, cp->acw, cp->ace)); 1161 g_io_request(cbp, disk->d_consumer); 1162 } 1163} 1164 1165static void 1166g_mirror_register_request(struct bio *bp) 1167{ 1168 struct g_mirror_softc *sc; 1169 1170 sc = bp->bio_to->geom->softc; 1171 switch (bp->bio_cmd) { 1172 case BIO_READ: 1173 switch (sc->sc_balance) { 1174 case G_MIRROR_BALANCE_ROUND_ROBIN: 1175 g_mirror_request_round_robin(sc, bp); 1176 break; 1177 case G_MIRROR_BALANCE_LOAD: 1178 g_mirror_request_load(sc, bp); 1179 break; 1180 case G_MIRROR_BALANCE_SPLIT: 1181 g_mirror_request_split(sc, bp); 1182 break; 1183 } 1184 return; 1185 case BIO_WRITE: 1186 case BIO_DELETE: 1187 { 1188 struct g_mirror_disk *disk; 1189 struct bio_queue_head queue; 1190 struct g_consumer *cp; 1191 struct bio *cbp; 1192 1193 /* 1194 * Allocate all bios before sending any request, so we can 1195 * return ENOMEM in nice and clean way. 1196 */ 1197 bioq_init(&queue); 1198 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1199 switch (disk->d_state) { 1200 case G_MIRROR_DISK_STATE_ACTIVE: 1201 break; 1202 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1203 if (bp->bio_offset >= disk->d_sync.ds_offset) 1204 continue; 1205 break; 1206 default: 1207 continue; 1208 } 1209 cbp = g_clone_bio(bp); 1210 if (cbp == NULL) { 1211 for (cbp = bioq_first(&queue); cbp != NULL; 1212 cbp = bioq_first(&queue)) { 1213 bioq_remove(&queue, cbp); 1214 g_destroy_bio(cbp); 1215 } 1216 if (bp->bio_error == 0) 1217 bp->bio_error = ENOMEM; 1218 g_io_deliver(bp, bp->bio_error); 1219 return; 1220 } 1221 bioq_insert_tail(&queue, cbp); 1222 } 1223 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1224 switch (disk->d_state) { 1225 case G_MIRROR_DISK_STATE_ACTIVE: 1226 break; 1227 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1228 if (bp->bio_offset >= disk->d_sync.ds_offset) 1229 continue; 1230 break; 1231 default: 1232 continue; 1233 } 1234 cbp = bioq_first(&queue); 1235 KASSERT(cbp != NULL, ("NULL cbp! (device %s).", 1236 sc->sc_name)); 1237 bioq_remove(&queue, cbp); 1238 cp = disk->d_consumer; 1239 cbp->bio_done = g_mirror_done; 1240 cbp->bio_to = cp->provider; 1241 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1242 KASSERT(cp->acw > 0 && cp->ace > 0, 1243 ("Consumer %s not opened (r%dw%de%d).", 1244 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1245 g_io_request(cbp, cp); 1246 } 1247 /* 1248 * Bump syncid on first write. 1249 */ 1250 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) { 1251 sc->sc_bump_syncid = 0; 1252 g_topology_lock(); 1253 g_mirror_bump_syncid(sc); 1254 g_topology_unlock(); 1255 } 1256 return; 1257 } 1258 default: 1259 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1260 bp->bio_cmd, sc->sc_name)); 1261 break; 1262 } 1263} 1264 1265/* 1266 * Worker thread. 1267 */ 1268static void 1269g_mirror_worker(void *arg) 1270{ 1271 struct g_mirror_softc *sc; 1272 struct g_mirror_disk *disk; 1273 struct g_mirror_event *ep; 1274 struct bio *bp; 1275 u_int nreqs; 1276 1277 sc = arg; 1278 curthread->td_base_pri = PRIBIO; 1279 1280 nreqs = 0; 1281 for (;;) { 1282 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1283 /* 1284 * First take a look at events. 1285 * This is important to handle events before any I/O requests. 1286 */ 1287 ep = g_mirror_event_get(sc); 1288 if (ep != NULL) { 1289 g_topology_lock(); 1290 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1291 /* Update only device status. */ 1292 G_MIRROR_DEBUG(3, 1293 "Running event for device %s.", 1294 sc->sc_name); 1295 ep->e_error = 0; 1296 g_mirror_update_device(sc, 1); 1297 } else { 1298 /* Update disk status. */ 1299 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1300 g_mirror_get_diskname(ep->e_disk)); 1301 ep->e_error = g_mirror_update_disk(ep->e_disk, 1302 ep->e_state); 1303 if (ep->e_error == 0) 1304 g_mirror_update_device(sc, 0); 1305 } 1306 g_topology_unlock(); 1307 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1308 KASSERT(ep->e_error == 0, 1309 ("Error cannot be handled.")); 1310 g_mirror_event_free(ep); 1311 } else { 1312 ep->e_flags |= G_MIRROR_EVENT_DONE; 1313 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1314 ep); 1315 mtx_lock(&sc->sc_events_mtx); 1316 wakeup(ep); 1317 mtx_unlock(&sc->sc_events_mtx); 1318 } 1319 if ((sc->sc_flags & 1320 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1321end: 1322 if ((sc->sc_flags & 1323 G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1324 G_MIRROR_DEBUG(4, "%s: Waking up %p.", 1325 __func__, &sc->sc_worker); 1326 wakeup(&sc->sc_worker); 1327 sc->sc_worker = NULL; 1328 } else { 1329 g_topology_lock(); 1330 g_mirror_destroy_device(sc); 1331 g_topology_unlock(); 1332 free(sc, M_MIRROR); 1333 } 1334 kthread_exit(0); 1335 } 1336 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1337 continue; 1338 } 1339 /* 1340 * Now I/O requests. 1341 */ 1342 /* Get first request from the queue. */ 1343 mtx_lock(&sc->sc_queue_mtx); 1344 bp = bioq_first(&sc->sc_queue); 1345 if (bp == NULL) { 1346 if ((sc->sc_flags & 1347 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1348 mtx_unlock(&sc->sc_queue_mtx); 1349 goto end; 1350 } 1351 } 1352 if (sc->sc_sync.ds_ndisks > 0 && 1353 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1354 mtx_unlock(&sc->sc_queue_mtx); 1355 /* 1356 * It is time for synchronization... 1357 */ 1358 nreqs = 0; 1359 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1360 if (disk->d_state != 1361 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1362 continue; 1363 } 1364 if (disk->d_sync.ds_offset >= 1365 sc->sc_provider->mediasize) { 1366 continue; 1367 } 1368 if (disk->d_sync.ds_offset > 1369 disk->d_sync.ds_offset_done) { 1370 continue; 1371 } 1372 g_mirror_sync_one(disk); 1373 } 1374 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1375 goto sleep; 1376 } 1377 if (bp == NULL) { 1378 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 0); 1379 G_MIRROR_DEBUG(5, "%s: I'm here 3.", __func__); 1380 continue; 1381 } 1382 nreqs++; 1383 bioq_remove(&sc->sc_queue, bp); 1384 mtx_unlock(&sc->sc_queue_mtx); 1385 1386 if ((bp->bio_flags & BIO_FLAG1) != 0) { 1387 g_mirror_regular_request(bp); 1388 } else if ((bp->bio_flags & BIO_FLAG2) != 0) { 1389 u_int timeout, sps; 1390 1391 g_mirror_sync_request(bp); 1392sleep: 1393 sps = atomic_load_acq_int(&g_mirror_syncs_per_sec); 1394 if (sps == 0) { 1395 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1396 continue; 1397 } 1398 mtx_lock(&sc->sc_queue_mtx); 1399 if (bioq_first(&sc->sc_queue) != NULL) { 1400 mtx_unlock(&sc->sc_queue_mtx); 1401 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1402 continue; 1403 } 1404 timeout = hz / sps; 1405 if (timeout == 0) 1406 timeout = 1; 1407 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w2", 1408 timeout); 1409 } else { 1410 g_mirror_register_request(bp); 1411 } 1412 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1413 } 1414} 1415 1416/* 1417 * Open disk's consumer if needed. 1418 */ 1419static void 1420g_mirror_update_access(struct g_mirror_disk *disk) 1421{ 1422 struct g_provider *pp; 1423 struct g_consumer *cp; 1424 int acr, acw, ace, cpw, error; 1425 1426 g_topology_assert(); 1427 1428 cp = disk->d_consumer; 1429 pp = disk->d_softc->sc_provider; 1430 if (pp == NULL) { 1431 acr = -cp->acr; 1432 acw = -cp->acw; 1433 ace = -cp->ace; 1434 } else { 1435 acr = pp->acr - cp->acr; 1436 acw = pp->acw - cp->acw; 1437 ace = pp->ace - cp->ace; 1438 /* Grab an extra "exclusive" bit. */ 1439 if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) 1440 ace++; 1441 } 1442 if (acr == 0 && acw == 0 && ace == 0) 1443 return; 1444 cpw = cp->acw; 1445 error = g_access(cp, acr, acw, ace); 1446 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, acr, 1447 acw, ace, error); 1448 if (error != 0) { 1449 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1450 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1451 G_MIRROR_EVENT_DONTWAIT); 1452 return; 1453 } 1454 if (cpw == 0 && cp->acw > 0) { 1455 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1456 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1457 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1458 } else if (cpw > 0 && cp->acw == 0) { 1459 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1460 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1461 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1462 } 1463} 1464 1465static void 1466g_mirror_sync_start(struct g_mirror_disk *disk) 1467{ 1468 struct g_mirror_softc *sc; 1469 struct g_consumer *cp; 1470 int error; 1471 1472 g_topology_assert(); 1473 1474 sc = disk->d_softc; 1475 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1476 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1477 sc->sc_state)); 1478 cp = disk->d_consumer; 1479 KASSERT(cp->acr == 0 && cp->acw == 0 && cp->ace == 0, 1480 ("Consumer %s already opened.", cp->provider->name)); 1481 1482 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1483 g_mirror_get_diskname(disk)); 1484 error = g_access(cp, 0, 1, 1); 1485 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, 1, 1486 1, error); 1487 if (error != 0) { 1488 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1489 G_MIRROR_EVENT_DONTWAIT); 1490 return; 1491 } 1492 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1493 KASSERT(disk->d_sync.ds_consumer == NULL, 1494 ("Sync consumer already exists (device=%s, disk=%s).", 1495 sc->sc_name, g_mirror_get_diskname(disk))); 1496 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1497 disk->d_sync.ds_consumer->private = disk; 1498 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1499 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1500 disk->d_softc->sc_name, error)); 1501 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1502 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1503 disk->d_softc->sc_name, error)); 1504 sc->sc_sync.ds_ndisks++; 1505} 1506 1507/* 1508 * Stop synchronization process. 1509 * type: 0 - synchronization finished 1510 * 1 - synchronization stopped 1511 */ 1512static void 1513g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1514{ 1515 struct g_consumer *cp; 1516 1517 g_topology_assert(); 1518 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1519 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1520 g_mirror_disk_state2str(disk->d_state))); 1521 if (disk->d_sync.ds_consumer == NULL) 1522 return; 1523 1524 if (type == 0) { 1525 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1526 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1527 } else /* if (type == 1) */ { 1528 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1529 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1530 } 1531 cp = disk->d_sync.ds_consumer; 1532 g_access(cp, -1, 0, 0); 1533 g_mirror_kill_consumer(disk->d_softc, cp); 1534 disk->d_sync.ds_consumer = NULL; 1535 disk->d_softc->sc_sync.ds_ndisks--; 1536 cp = disk->d_consumer; 1537 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 1538 ("Consumer %s not opened.", cp->provider->name)); 1539 g_access(cp, 0, -1, -1); 1540 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, -1, 1541 -1, 0); 1542 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1543} 1544 1545static void 1546g_mirror_launch_provider(struct g_mirror_softc *sc) 1547{ 1548 struct g_mirror_disk *disk; 1549 struct g_provider *pp; 1550 1551 g_topology_assert(); 1552 1553 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1554 pp->mediasize = sc->sc_mediasize; 1555 pp->sectorsize = sc->sc_sectorsize; 1556 sc->sc_provider = pp; 1557 g_error_provider(pp, 0); 1558 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1559 pp->name); 1560 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1561 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1562 g_mirror_sync_start(disk); 1563 } 1564} 1565 1566static void 1567g_mirror_destroy_provider(struct g_mirror_softc *sc) 1568{ 1569 struct g_mirror_disk *disk; 1570 struct bio *bp; 1571 1572 g_topology_assert(); 1573 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1574 sc->sc_name)); 1575 1576 g_error_provider(sc->sc_provider, ENXIO); 1577 mtx_lock(&sc->sc_queue_mtx); 1578 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1579 bioq_remove(&sc->sc_queue, bp); 1580 g_io_deliver(bp, ENXIO); 1581 } 1582 mtx_unlock(&sc->sc_queue_mtx); 1583 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1584 sc->sc_provider->name); 1585 sc->sc_provider->flags |= G_PF_WITHER; 1586 g_orphan_provider(sc->sc_provider, ENXIO); 1587 sc->sc_provider = NULL; 1588 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1589 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1590 g_mirror_sync_stop(disk, 1); 1591 } 1592} 1593 1594static void 1595g_mirror_go(void *arg) 1596{ 1597 struct g_mirror_softc *sc; 1598 1599 sc = arg; 1600 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1601 g_mirror_event_send(sc, 0, 1602 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1603} 1604 1605static u_int 1606g_mirror_determine_state(struct g_mirror_disk *disk) 1607{ 1608 struct g_mirror_softc *sc; 1609 u_int state; 1610 1611 sc = disk->d_softc; 1612 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1613 if ((disk->d_flags & 1614 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1615 /* Disk does not need synchronization. */ 1616 state = G_MIRROR_DISK_STATE_ACTIVE; 1617 } else { 1618 if ((sc->sc_flags & 1619 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1620 (disk->d_flags & 1621 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1622 /* 1623 * We can start synchronization from 1624 * the stored offset. 1625 */ 1626 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1627 } else { 1628 state = G_MIRROR_DISK_STATE_STALE; 1629 } 1630 } 1631 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1632 /* 1633 * Reset all synchronization data for this disk, 1634 * because if it even was synchronized, it was 1635 * synchronized to disks with different syncid. 1636 */ 1637 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1638 disk->d_sync.ds_offset = 0; 1639 disk->d_sync.ds_offset_done = 0; 1640 disk->d_sync.ds_syncid = sc->sc_syncid; 1641 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1642 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1643 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1644 } else { 1645 state = G_MIRROR_DISK_STATE_STALE; 1646 } 1647 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1648 /* 1649 * Not good, NOT GOOD! 1650 * It means that mirror was started on stale disks 1651 * and more fresh disk just arrive. 1652 * If there were writes, mirror is fucked up, sorry. 1653 * I think the best choice here is don't touch 1654 * this disk and inform the user laudly. 1655 */ 1656 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1657 "disk (%s) arrives!! It will not be connected to the " 1658 "running device.", sc->sc_name, 1659 g_mirror_get_diskname(disk)); 1660 g_mirror_destroy_disk(disk); 1661 state = G_MIRROR_DISK_STATE_NONE; 1662 /* Return immediately, because disk was destroyed. */ 1663 return (state); 1664 } 1665 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1666 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1667 return (state); 1668} 1669 1670/* 1671 * Update device state. 1672 */ 1673static void 1674g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1675{ 1676 struct g_mirror_disk *disk; 1677 u_int state; 1678 1679 g_topology_assert(); 1680 1681 switch (sc->sc_state) { 1682 case G_MIRROR_DEVICE_STATE_STARTING: 1683 { 1684 struct g_mirror_disk *pdisk; 1685 u_int dirty, ndisks, syncid; 1686 1687 KASSERT(sc->sc_provider == NULL, 1688 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1689 /* 1690 * Are we ready? We are, if all disks are connected or 1691 * if we have any disks and 'force' is true. 1692 */ 1693 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1694 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1695 ; 1696 } else if (g_mirror_ndisks(sc, -1) == 0) { 1697 /* 1698 * Disks went down in starting phase, so destroy 1699 * device. 1700 */ 1701 callout_drain(&sc->sc_callout); 1702 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1703 return; 1704 } else { 1705 return; 1706 } 1707 1708 /* 1709 * Activate all disks with the biggest syncid. 1710 */ 1711 if (force) { 1712 /* 1713 * If 'force' is true, we have been called due to 1714 * timeout, so don't bother canceling timeout. 1715 */ 1716 ndisks = 0; 1717 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1718 if ((disk->d_flags & 1719 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1720 ndisks++; 1721 } 1722 } 1723 if (ndisks == 0) { 1724 /* No valid disks found, destroy device. */ 1725 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1726 return; 1727 } 1728 } else { 1729 /* Cancel timeout. */ 1730 callout_drain(&sc->sc_callout); 1731 } 1732 1733 /* 1734 * Find disk with the biggest syncid. 1735 */ 1736 syncid = 0; 1737 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1738 if (disk->d_sync.ds_syncid > syncid) 1739 syncid = disk->d_sync.ds_syncid; 1740 } 1741 1742 /* 1743 * Here we need to look for dirty disks and if all disks 1744 * with the biggest syncid are dirty, we have to choose 1745 * one with the biggest priority and rebuild the rest. 1746 */ 1747 /* 1748 * Find the number of dirty disks with the biggest syncid. 1749 * Find the number of disks with the biggest syncid. 1750 * While here, find a disk with the biggest priority. 1751 */ 1752 dirty = ndisks = 0; 1753 pdisk = NULL; 1754 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1755 if (disk->d_sync.ds_syncid != syncid) 1756 continue; 1757 if ((disk->d_flags & 1758 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1759 continue; 1760 } 1761 ndisks++; 1762 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1763 dirty++; 1764 if (pdisk == NULL || 1765 pdisk->d_priority < disk->d_priority) { 1766 pdisk = disk; 1767 } 1768 } 1769 } 1770 if (dirty == 0) { 1771 /* No dirty disks at all, great. */ 1772 } else if (dirty == ndisks) { 1773 /* 1774 * Force synchronization for all dirty disks except one 1775 * with the biggest priority. 1776 */ 1777 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1778 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1779 "master disk for synchronization.", 1780 g_mirror_get_diskname(pdisk), sc->sc_name); 1781 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1782 if (disk->d_sync.ds_syncid != syncid) 1783 continue; 1784 if ((disk->d_flags & 1785 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1786 continue; 1787 } 1788 KASSERT((disk->d_flags & 1789 G_MIRROR_DISK_FLAG_DIRTY) != 0, 1790 ("Disk %s isn't marked as dirty.", 1791 g_mirror_get_diskname(disk))); 1792 /* Skip the disk with the biggest priority. */ 1793 if (disk == pdisk) 1794 continue; 1795 disk->d_sync.ds_syncid = 0; 1796 } 1797 } else if (dirty < ndisks) { 1798 /* 1799 * Force synchronization for all dirty disks. 1800 * We have some non-dirty disks. 1801 */ 1802 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1803 if (disk->d_sync.ds_syncid != syncid) 1804 continue; 1805 if ((disk->d_flags & 1806 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1807 continue; 1808 } 1809 if ((disk->d_flags & 1810 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1811 continue; 1812 } 1813 disk->d_sync.ds_syncid = 0; 1814 } 1815 } 1816 1817 /* Reset hint. */ 1818 sc->sc_hint = NULL; 1819 sc->sc_syncid = syncid; 1820 if (force) { 1821 /* Remember to bump syncid on first write. */ 1822 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1823 } 1824 state = G_MIRROR_DEVICE_STATE_RUNNING; 1825 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 1826 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 1827 g_mirror_device_state2str(state)); 1828 sc->sc_state = state; 1829 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1830 state = g_mirror_determine_state(disk); 1831 g_mirror_event_send(disk, state, 1832 G_MIRROR_EVENT_DONTWAIT); 1833 if (state == G_MIRROR_DISK_STATE_STALE) { 1834 sc->sc_bump_syncid = 1835 G_MIRROR_BUMP_ON_FIRST_WRITE; 1836 } 1837 } 1838 break; 1839 } 1840 case G_MIRROR_DEVICE_STATE_RUNNING: 1841 /* 1842 * Bump syncid here, if we need to do it immediately. 1843 */ 1844 if (sc->sc_bump_syncid == G_MIRROR_BUMP_IMMEDIATELY) { 1845 sc->sc_bump_syncid = 0; 1846 g_mirror_bump_syncid(sc); 1847 } 1848 if (g_mirror_ndisks(sc, -1) == 0) { 1849 /* 1850 * No disks at all, we need to destroy device. 1851 */ 1852 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1853 } else if (g_mirror_ndisks(sc, 1854 G_MIRROR_DISK_STATE_ACTIVE) == 0 && 1855 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 1856 /* 1857 * No active disks, destroy provider. 1858 */ 1859 if (sc->sc_provider != NULL) 1860 g_mirror_destroy_provider(sc); 1861 } else if (g_mirror_ndisks(sc, 1862 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 1863 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 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_spoiled; 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_spoiled; 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.", 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