g_mirror.c revision 139051
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 139051 2004-12-19 23:12:00Z 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/eventhandler.h> 41#include <vm/uma.h> 42#include <geom/geom.h> 43#include <sys/proc.h> 44#include <sys/kthread.h> 45#include <geom/mirror/g_mirror.h> 46 47 48static MALLOC_DEFINE(M_MIRROR, "mirror data", "GEOM_MIRROR Data"); 49 50SYSCTL_DECL(_kern_geom); 51SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, "GEOM_MIRROR stuff"); 52u_int g_mirror_debug = 0; 53TUNABLE_INT("kern.geom.mirror.debug", &g_mirror_debug); 54SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RW, &g_mirror_debug, 0, 55 "Debug level"); 56static u_int g_mirror_timeout = 4; 57TUNABLE_INT("kern.geom.mirror.timeout", &g_mirror_timeout); 58SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RW, &g_mirror_timeout, 59 0, "Time to wait on all mirror components"); 60static u_int g_mirror_idletime = 5; 61TUNABLE_INT("kern.geom.mirror.idletime", &g_mirror_idletime); 62SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RW, 63 &g_mirror_idletime, 0, "Mark components as clean when idling"); 64static u_int g_mirror_reqs_per_sync = 5; 65SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, reqs_per_sync, CTLFLAG_RW, 66 &g_mirror_reqs_per_sync, 0, 67 "Number of regular I/O requests per synchronization request"); 68static u_int g_mirror_syncs_per_sec = 100; 69SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, syncs_per_sec, CTLFLAG_RW, 70 &g_mirror_syncs_per_sec, 0, 71 "Number of synchronizations requests per second"); 72 73#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 74 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 75 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 76 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 77} while (0) 78 79static eventhandler_tag g_mirror_ehtag = NULL; 80 81static int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp, 82 struct g_geom *gp); 83static g_taste_t g_mirror_taste; 84static void g_mirror_init(struct g_class *mp); 85static void g_mirror_fini(struct g_class *mp); 86 87struct g_class g_mirror_class = { 88 .name = G_MIRROR_CLASS_NAME, 89 .version = G_VERSION, 90 .ctlreq = g_mirror_config, 91 .taste = g_mirror_taste, 92 .destroy_geom = g_mirror_destroy_geom, 93 .init = g_mirror_init, 94 .fini = g_mirror_fini 95}; 96 97 98static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 99static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 100static void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force); 101static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 102 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 103static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 104 105 106static const char * 107g_mirror_disk_state2str(int state) 108{ 109 110 switch (state) { 111 case G_MIRROR_DISK_STATE_NONE: 112 return ("NONE"); 113 case G_MIRROR_DISK_STATE_NEW: 114 return ("NEW"); 115 case G_MIRROR_DISK_STATE_ACTIVE: 116 return ("ACTIVE"); 117 case G_MIRROR_DISK_STATE_STALE: 118 return ("STALE"); 119 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 120 return ("SYNCHRONIZING"); 121 case G_MIRROR_DISK_STATE_DISCONNECTED: 122 return ("DISCONNECTED"); 123 case G_MIRROR_DISK_STATE_DESTROY: 124 return ("DESTROY"); 125 default: 126 return ("INVALID"); 127 } 128} 129 130static const char * 131g_mirror_device_state2str(int state) 132{ 133 134 switch (state) { 135 case G_MIRROR_DEVICE_STATE_STARTING: 136 return ("STARTING"); 137 case G_MIRROR_DEVICE_STATE_RUNNING: 138 return ("RUNNING"); 139 default: 140 return ("INVALID"); 141 } 142} 143 144static const char * 145g_mirror_get_diskname(struct g_mirror_disk *disk) 146{ 147 148 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 149 return ("[unknown]"); 150 return (disk->d_name); 151} 152 153/* 154 * --- Events handling functions --- 155 * Events in geom_mirror are used to maintain disks and device status 156 * from one thread to simplify locking. 157 */ 158static void 159g_mirror_event_free(struct g_mirror_event *ep) 160{ 161 162 free(ep, M_MIRROR); 163} 164 165int 166g_mirror_event_send(void *arg, int state, int flags) 167{ 168 struct g_mirror_softc *sc; 169 struct g_mirror_disk *disk; 170 struct g_mirror_event *ep; 171 int error; 172 173 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 174 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 175 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 176 disk = NULL; 177 sc = arg; 178 } else { 179 disk = arg; 180 sc = disk->d_softc; 181 } 182 ep->e_disk = disk; 183 ep->e_state = state; 184 ep->e_flags = flags; 185 ep->e_error = 0; 186 mtx_lock(&sc->sc_events_mtx); 187 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 188 mtx_unlock(&sc->sc_events_mtx); 189 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 190 mtx_lock(&sc->sc_queue_mtx); 191 wakeup(sc); 192 mtx_unlock(&sc->sc_queue_mtx); 193 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 194 return (0); 195 g_topology_assert(); 196 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 197 g_topology_unlock(); 198 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 199 mtx_lock(&sc->sc_events_mtx); 200 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 201 hz * 5); 202 } 203 /* Don't even try to use 'sc' here, because it could be already dead. */ 204 g_topology_lock(); 205 error = ep->e_error; 206 g_mirror_event_free(ep); 207 return (error); 208} 209 210static struct g_mirror_event * 211g_mirror_event_get(struct g_mirror_softc *sc) 212{ 213 struct g_mirror_event *ep; 214 215 mtx_lock(&sc->sc_events_mtx); 216 ep = TAILQ_FIRST(&sc->sc_events); 217 if (ep != NULL) 218 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 219 mtx_unlock(&sc->sc_events_mtx); 220 return (ep); 221} 222 223static void 224g_mirror_event_cancel(struct g_mirror_disk *disk) 225{ 226 struct g_mirror_softc *sc; 227 struct g_mirror_event *ep, *tmpep; 228 229 g_topology_assert(); 230 231 sc = disk->d_softc; 232 mtx_lock(&sc->sc_events_mtx); 233 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 234 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 235 continue; 236 if (ep->e_disk != disk) 237 continue; 238 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 239 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 240 g_mirror_event_free(ep); 241 else { 242 ep->e_error = ECANCELED; 243 wakeup(ep); 244 } 245 } 246 mtx_unlock(&sc->sc_events_mtx); 247} 248 249/* 250 * Return the number of disks in given state. 251 * If state is equal to -1, count all connected disks. 252 */ 253u_int 254g_mirror_ndisks(struct g_mirror_softc *sc, int state) 255{ 256 struct g_mirror_disk *disk; 257 u_int n = 0; 258 259 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 260 if (state == -1 || disk->d_state == state) 261 n++; 262 } 263 return (n); 264} 265 266/* 267 * Find a disk in mirror by its disk ID. 268 */ 269static struct g_mirror_disk * 270g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 271{ 272 struct g_mirror_disk *disk; 273 274 g_topology_assert(); 275 276 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 277 if (disk->d_id == id) 278 return (disk); 279 } 280 return (NULL); 281} 282 283static u_int 284g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 285{ 286 struct bio *bp; 287 u_int nreqs = 0; 288 289 mtx_lock(&sc->sc_queue_mtx); 290 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 291 if (bp->bio_from == cp) 292 nreqs++; 293 } 294 mtx_unlock(&sc->sc_queue_mtx); 295 return (nreqs); 296} 297 298static int 299g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 300{ 301 302 if (cp->index > 0) { 303 G_MIRROR_DEBUG(2, 304 "I/O requests for %s exist, can't destroy it now.", 305 cp->provider->name); 306 return (1); 307 } 308 if (g_mirror_nrequests(sc, cp) > 0) { 309 G_MIRROR_DEBUG(2, 310 "I/O requests for %s in queue, can't destroy it now.", 311 cp->provider->name); 312 return (1); 313 } 314 return (0); 315} 316 317static void 318g_mirror_destroy_consumer(void *arg, int flags) 319{ 320 struct g_consumer *cp; 321 322 cp = arg; 323 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 324 g_detach(cp); 325 g_destroy_consumer(cp); 326} 327 328static void 329g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 330{ 331 struct g_provider *pp; 332 int retaste_wait; 333 334 g_topology_assert(); 335 336 cp->private = NULL; 337 if (g_mirror_is_busy(sc, cp)) 338 return; 339 pp = cp->provider; 340 retaste_wait = 0; 341 if (cp->acw == 1) { 342 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 343 retaste_wait = 1; 344 } 345 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 346 -cp->acw, -cp->ace, 0); 347 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 348 if (retaste_wait) { 349 /* 350 * After retaste event was send (inside g_access()), we can send 351 * event to detach and destroy consumer. 352 * A class, which has consumer to the given provider connected 353 * will not receive retaste event for the provider. 354 * This is the way how I ignore retaste events when I close 355 * consumers opened for write: I detach and destroy consumer 356 * after retaste event is sent. 357 */ 358 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 359 return; 360 } 361 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 362 g_detach(cp); 363 g_destroy_consumer(cp); 364} 365 366static int 367g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 368{ 369 int error; 370 371 g_topology_assert(); 372 KASSERT(disk->d_consumer == NULL, 373 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 374 375 disk->d_consumer = g_new_consumer(disk->d_softc->sc_geom); 376 disk->d_consumer->private = disk; 377 disk->d_consumer->index = 0; 378 error = g_attach(disk->d_consumer, pp); 379 if (error != 0) 380 return (error); 381 error = g_access(disk->d_consumer, 1, 1, 1); 382 if (error != 0) { 383 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 384 pp->name, error); 385 return (error); 386 } 387 388 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 389 return (0); 390} 391 392static void 393g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 394{ 395 396 g_topology_assert(); 397 398 if (cp == NULL) 399 return; 400 if (cp->provider != NULL) 401 g_mirror_kill_consumer(sc, cp); 402 else 403 g_destroy_consumer(cp); 404} 405 406/* 407 * Initialize disk. This means allocate memory, create consumer, attach it 408 * to the provider and open access (r1w1e1) to it. 409 */ 410static struct g_mirror_disk * 411g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 412 struct g_mirror_metadata *md, int *errorp) 413{ 414 struct g_mirror_disk *disk; 415 int error; 416 417 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 418 if (disk == NULL) { 419 error = ENOMEM; 420 goto fail; 421 } 422 disk->d_softc = sc; 423 error = g_mirror_connect_disk(disk, pp); 424 if (error != 0) 425 goto fail; 426 disk->d_id = md->md_did; 427 disk->d_state = G_MIRROR_DISK_STATE_NONE; 428 disk->d_priority = md->md_priority; 429 disk->d_delay.sec = 0; 430 disk->d_delay.frac = 0; 431 binuptime(&disk->d_last_used); 432 disk->d_flags = md->md_dflags; 433 if (md->md_provider[0] != '\0') 434 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 435 disk->d_sync.ds_consumer = NULL; 436 disk->d_sync.ds_offset = md->md_sync_offset; 437 disk->d_sync.ds_offset_done = md->md_sync_offset; 438 disk->d_sync.ds_resync = -1; 439 disk->d_sync.ds_syncid = md->md_syncid; 440 if (errorp != NULL) 441 *errorp = 0; 442 return (disk); 443fail: 444 if (errorp != NULL) 445 *errorp = error; 446 if (disk != NULL) { 447 g_mirror_disconnect_consumer(sc, disk->d_consumer); 448 free(disk, M_MIRROR); 449 } 450 return (NULL); 451} 452 453static void 454g_mirror_destroy_disk(struct g_mirror_disk *disk) 455{ 456 struct g_mirror_softc *sc; 457 458 g_topology_assert(); 459 460 LIST_REMOVE(disk, d_next); 461 g_mirror_event_cancel(disk); 462 sc = disk->d_softc; 463 if (sc->sc_hint == disk) 464 sc->sc_hint = NULL; 465 switch (disk->d_state) { 466 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 467 g_mirror_sync_stop(disk, 1); 468 /* FALLTHROUGH */ 469 case G_MIRROR_DISK_STATE_NEW: 470 case G_MIRROR_DISK_STATE_STALE: 471 case G_MIRROR_DISK_STATE_ACTIVE: 472 g_mirror_disconnect_consumer(sc, disk->d_consumer); 473 free(disk, M_MIRROR); 474 break; 475 default: 476 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 477 g_mirror_get_diskname(disk), 478 g_mirror_disk_state2str(disk->d_state))); 479 } 480} 481 482static void 483g_mirror_destroy_device(struct g_mirror_softc *sc) 484{ 485 struct g_mirror_disk *disk; 486 struct g_mirror_event *ep; 487 struct g_geom *gp; 488 struct g_consumer *cp, *tmpcp; 489 490 g_topology_assert(); 491 492 gp = sc->sc_geom; 493 if (sc->sc_provider != NULL) 494 g_mirror_destroy_provider(sc); 495 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 496 disk = LIST_FIRST(&sc->sc_disks)) { 497 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 498 g_mirror_update_metadata(disk); 499 g_mirror_destroy_disk(disk); 500 } 501 while ((ep = g_mirror_event_get(sc)) != NULL) { 502 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 503 g_mirror_event_free(ep); 504 else { 505 ep->e_error = ECANCELED; 506 ep->e_flags |= G_MIRROR_EVENT_DONE; 507 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 508 mtx_lock(&sc->sc_events_mtx); 509 wakeup(ep); 510 mtx_unlock(&sc->sc_events_mtx); 511 } 512 } 513 callout_drain(&sc->sc_callout); 514 gp->softc = NULL; 515 516 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 517 g_mirror_disconnect_consumer(sc, cp); 518 } 519 sc->sc_sync.ds_geom->softc = NULL; 520 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 521 mtx_destroy(&sc->sc_queue_mtx); 522 mtx_destroy(&sc->sc_events_mtx); 523 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 524 g_wither_geom(gp, ENXIO); 525} 526 527static void 528g_mirror_orphan(struct g_consumer *cp) 529{ 530 struct g_mirror_disk *disk; 531 532 g_topology_assert(); 533 534 disk = cp->private; 535 if (disk == NULL) 536 return; 537 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 538 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 539 G_MIRROR_EVENT_DONTWAIT); 540} 541 542static void 543g_mirror_spoiled(struct g_consumer *cp) 544{ 545 struct g_mirror_disk *disk; 546 547 g_topology_assert(); 548 549 disk = cp->private; 550 if (disk == NULL) 551 return; 552 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 553 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 554 G_MIRROR_EVENT_DONTWAIT); 555} 556 557/* 558 * Function should return the next active disk on the list. 559 * It is possible that it will be the same disk as given. 560 * If there are no active disks on list, NULL is returned. 561 */ 562static __inline struct g_mirror_disk * 563g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 564{ 565 struct g_mirror_disk *dp; 566 567 for (dp = LIST_NEXT(disk, d_next); dp != disk; 568 dp = LIST_NEXT(dp, d_next)) { 569 if (dp == NULL) 570 dp = LIST_FIRST(&sc->sc_disks); 571 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 572 break; 573 } 574 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 575 return (NULL); 576 return (dp); 577} 578 579static struct g_mirror_disk * 580g_mirror_get_disk(struct g_mirror_softc *sc) 581{ 582 struct g_mirror_disk *disk; 583 584 if (sc->sc_hint == NULL) { 585 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 586 if (sc->sc_hint == NULL) 587 return (NULL); 588 } 589 disk = sc->sc_hint; 590 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 591 disk = g_mirror_find_next(sc, disk); 592 if (disk == NULL) 593 return (NULL); 594 } 595 sc->sc_hint = g_mirror_find_next(sc, disk); 596 return (disk); 597} 598 599static int 600g_mirror_write_metadata(struct g_mirror_disk *disk, 601 struct g_mirror_metadata *md) 602{ 603 struct g_mirror_softc *sc; 604 struct g_consumer *cp; 605 off_t offset, length; 606 u_char *sector; 607 int error = 0; 608 609 g_topology_assert(); 610 611 sc = disk->d_softc; 612 cp = disk->d_consumer; 613 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 614 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 615 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 616 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 617 cp->acw, cp->ace)); 618 length = cp->provider->sectorsize; 619 offset = cp->provider->mediasize - length; 620 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 621 if (md != NULL) 622 mirror_metadata_encode(md, sector); 623 g_topology_unlock(); 624 error = g_write_data(cp, offset, sector, length); 625 g_topology_lock(); 626 free(sector, M_MIRROR); 627 if (error != 0) { 628 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 629 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 630 G_MIRROR_EVENT_DONTWAIT); 631 } 632 return (error); 633} 634 635static int 636g_mirror_clear_metadata(struct g_mirror_disk *disk) 637{ 638 int error; 639 640 g_topology_assert(); 641 error = g_mirror_write_metadata(disk, NULL); 642 if (error == 0) { 643 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 644 g_mirror_get_diskname(disk)); 645 } else { 646 G_MIRROR_DEBUG(0, 647 "Cannot clear metadata on disk %s (error=%d).", 648 g_mirror_get_diskname(disk), error); 649 } 650 return (error); 651} 652 653void 654g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 655 struct g_mirror_metadata *md) 656{ 657 658 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 659 md->md_version = G_MIRROR_VERSION; 660 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 661 md->md_mid = sc->sc_id; 662 md->md_all = sc->sc_ndisks; 663 md->md_slice = sc->sc_slice; 664 md->md_balance = sc->sc_balance; 665 md->md_mediasize = sc->sc_mediasize; 666 md->md_sectorsize = sc->sc_sectorsize; 667 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 668 bzero(md->md_provider, sizeof(md->md_provider)); 669 if (disk == NULL) { 670 md->md_did = arc4random(); 671 md->md_priority = 0; 672 md->md_syncid = 0; 673 md->md_dflags = 0; 674 md->md_sync_offset = 0; 675 } else { 676 md->md_did = disk->d_id; 677 md->md_priority = disk->d_priority; 678 md->md_syncid = disk->d_sync.ds_syncid; 679 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 680 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 681 md->md_sync_offset = disk->d_sync.ds_offset_done; 682 else 683 md->md_sync_offset = 0; 684 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 685 strlcpy(md->md_provider, 686 disk->d_consumer->provider->name, 687 sizeof(md->md_provider)); 688 } 689 } 690} 691 692void 693g_mirror_update_metadata(struct g_mirror_disk *disk) 694{ 695 struct g_mirror_metadata md; 696 int error; 697 698 g_topology_assert(); 699 g_mirror_fill_metadata(disk->d_softc, disk, &md); 700 error = g_mirror_write_metadata(disk, &md); 701 if (error == 0) { 702 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 703 g_mirror_get_diskname(disk)); 704 } else { 705 G_MIRROR_DEBUG(0, 706 "Cannot update metadata on disk %s (error=%d).", 707 g_mirror_get_diskname(disk), error); 708 } 709} 710 711static void 712g_mirror_bump_syncid(struct g_mirror_softc *sc) 713{ 714 struct g_mirror_disk *disk; 715 716 g_topology_assert(); 717 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 718 ("%s called with no active disks (device=%s).", __func__, 719 sc->sc_name)); 720 721 sc->sc_syncid++; 722 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 723 sc->sc_syncid); 724 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 725 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 726 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 727 disk->d_sync.ds_syncid = sc->sc_syncid; 728 g_mirror_update_metadata(disk); 729 } 730 } 731} 732 733static void 734g_mirror_idle(struct g_mirror_softc *sc) 735{ 736 struct g_mirror_disk *disk; 737 738 if (sc->sc_provider == NULL || sc->sc_provider->acw == 0) 739 return; 740 sc->sc_idle = 1; 741 g_topology_lock(); 742 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 743 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 744 continue; 745 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 746 g_mirror_get_diskname(disk), sc->sc_name); 747 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 748 g_mirror_update_metadata(disk); 749 } 750 g_topology_unlock(); 751} 752 753static void 754g_mirror_unidle(struct g_mirror_softc *sc) 755{ 756 struct g_mirror_disk *disk; 757 758 sc->sc_idle = 0; 759 g_topology_lock(); 760 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 761 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 762 continue; 763 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 764 g_mirror_get_diskname(disk), sc->sc_name); 765 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 766 g_mirror_update_metadata(disk); 767 } 768 g_topology_unlock(); 769} 770 771/* 772 * Return 1 if we should check if mirror is idling. 773 */ 774static int 775g_mirror_check_idle(struct g_mirror_softc *sc) 776{ 777 struct g_mirror_disk *disk; 778 779 if (sc->sc_idle) 780 return (0); 781 if (sc->sc_provider != NULL && sc->sc_provider->acw == 0) 782 return (0); 783 /* 784 * Check if there are no in-flight requests. 785 */ 786 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 787 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 788 continue; 789 if (disk->d_consumer->index > 0) 790 return (0); 791 } 792 return (1); 793} 794 795static __inline int 796bintime_cmp(struct bintime *bt1, struct bintime *bt2) 797{ 798 799 if (bt1->sec < bt2->sec) 800 return (-1); 801 else if (bt1->sec > bt2->sec) 802 return (1); 803 if (bt1->frac < bt2->frac) 804 return (-1); 805 else if (bt1->frac > bt2->frac) 806 return (1); 807 return (0); 808} 809 810static void 811g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 812{ 813 814 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 815 return; 816 binuptime(&disk->d_delay); 817 bintime_sub(&disk->d_delay, &bp->bio_t0); 818} 819 820static void 821g_mirror_done(struct bio *bp) 822{ 823 struct g_mirror_softc *sc; 824 825 sc = bp->bio_from->geom->softc; 826 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 827 mtx_lock(&sc->sc_queue_mtx); 828 bioq_disksort(&sc->sc_queue, bp); 829 wakeup(sc); 830 mtx_unlock(&sc->sc_queue_mtx); 831} 832 833static void 834g_mirror_regular_request(struct bio *bp) 835{ 836 struct g_mirror_softc *sc; 837 struct g_mirror_disk *disk; 838 struct bio *pbp; 839 840 g_topology_assert_not(); 841 842 bp->bio_from->index--; 843 pbp = bp->bio_parent; 844 sc = pbp->bio_to->geom->softc; 845 disk = bp->bio_from->private; 846 if (disk == NULL) { 847 g_topology_lock(); 848 g_mirror_kill_consumer(sc, bp->bio_from); 849 g_topology_unlock(); 850 } else { 851 g_mirror_update_delay(disk, bp); 852 } 853 854 pbp->bio_inbed++; 855 KASSERT(pbp->bio_inbed <= pbp->bio_children, 856 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 857 pbp->bio_children)); 858 if (bp->bio_error == 0 && pbp->bio_error == 0) { 859 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 860 g_destroy_bio(bp); 861 if (pbp->bio_children == pbp->bio_inbed) { 862 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 863 pbp->bio_completed = pbp->bio_length; 864 g_io_deliver(pbp, pbp->bio_error); 865 } 866 return; 867 } else if (bp->bio_error != 0) { 868 if (pbp->bio_error == 0) 869 pbp->bio_error = bp->bio_error; 870 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 871 bp->bio_error); 872 if (disk != NULL) { 873 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 874 g_mirror_event_send(disk, 875 G_MIRROR_DISK_STATE_DISCONNECTED, 876 G_MIRROR_EVENT_DONTWAIT); 877 } 878 switch (pbp->bio_cmd) { 879 case BIO_DELETE: 880 case BIO_WRITE: 881 pbp->bio_inbed--; 882 pbp->bio_children--; 883 break; 884 } 885 } 886 g_destroy_bio(bp); 887 888 switch (pbp->bio_cmd) { 889 case BIO_READ: 890 if (pbp->bio_children == pbp->bio_inbed) { 891 pbp->bio_error = 0; 892 mtx_lock(&sc->sc_queue_mtx); 893 bioq_disksort(&sc->sc_queue, pbp); 894 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 895 wakeup(sc); 896 mtx_unlock(&sc->sc_queue_mtx); 897 } 898 break; 899 case BIO_DELETE: 900 case BIO_WRITE: 901 if (pbp->bio_children == 0) { 902 /* 903 * All requests failed. 904 */ 905 } else if (pbp->bio_inbed < pbp->bio_children) { 906 /* Do nothing. */ 907 break; 908 } else if (pbp->bio_children == pbp->bio_inbed) { 909 /* Some requests succeeded. */ 910 pbp->bio_error = 0; 911 pbp->bio_completed = pbp->bio_length; 912 } 913 g_io_deliver(pbp, pbp->bio_error); 914 break; 915 default: 916 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 917 break; 918 } 919} 920 921static void 922g_mirror_sync_done(struct bio *bp) 923{ 924 struct g_mirror_softc *sc; 925 926 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 927 sc = bp->bio_from->geom->softc; 928 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 929 mtx_lock(&sc->sc_queue_mtx); 930 bioq_disksort(&sc->sc_queue, bp); 931 wakeup(sc); 932 mtx_unlock(&sc->sc_queue_mtx); 933} 934 935static void 936g_mirror_start(struct bio *bp) 937{ 938 struct g_mirror_softc *sc; 939 940 sc = bp->bio_to->geom->softc; 941 /* 942 * If sc == NULL or there are no valid disks, provider's error 943 * should be set and g_mirror_start() should not be called at all. 944 */ 945 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 946 ("Provider's error should be set (error=%d)(mirror=%s).", 947 bp->bio_to->error, bp->bio_to->name)); 948 G_MIRROR_LOGREQ(3, bp, "Request received."); 949 950 switch (bp->bio_cmd) { 951 case BIO_READ: 952 case BIO_WRITE: 953 case BIO_DELETE: 954 break; 955 case BIO_GETATTR: 956 default: 957 g_io_deliver(bp, EOPNOTSUPP); 958 return; 959 } 960 mtx_lock(&sc->sc_queue_mtx); 961 bioq_disksort(&sc->sc_queue, bp); 962 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 963 wakeup(sc); 964 mtx_unlock(&sc->sc_queue_mtx); 965} 966 967/* 968 * Send one synchronization request. 969 */ 970static void 971g_mirror_sync_one(struct g_mirror_disk *disk) 972{ 973 struct g_mirror_softc *sc; 974 struct bio *bp; 975 976 sc = disk->d_softc; 977 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 978 ("Disk %s is not marked for synchronization.", 979 g_mirror_get_diskname(disk))); 980 981 bp = g_new_bio(); 982 if (bp == NULL) 983 return; 984 bp->bio_parent = NULL; 985 bp->bio_cmd = BIO_READ; 986 bp->bio_offset = disk->d_sync.ds_offset; 987 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 988 bp->bio_cflags = 0; 989 bp->bio_done = g_mirror_sync_done; 990 bp->bio_data = disk->d_sync.ds_data; 991 if (bp->bio_data == NULL) { 992 g_destroy_bio(bp); 993 return; 994 } 995 disk->d_sync.ds_offset += bp->bio_length; 996 bp->bio_to = sc->sc_provider; 997 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 998 disk->d_sync.ds_consumer->index++; 999 g_io_request(bp, disk->d_sync.ds_consumer); 1000} 1001 1002static void 1003g_mirror_sync_request(struct bio *bp) 1004{ 1005 struct g_mirror_softc *sc; 1006 struct g_mirror_disk *disk; 1007 1008 bp->bio_from->index--; 1009 sc = bp->bio_from->geom->softc; 1010 disk = bp->bio_from->private; 1011 if (disk == NULL) { 1012 g_topology_lock(); 1013 g_mirror_kill_consumer(sc, bp->bio_from); 1014 g_topology_unlock(); 1015 g_destroy_bio(bp); 1016 return; 1017 } 1018 1019 /* 1020 * Synchronization request. 1021 */ 1022 switch (bp->bio_cmd) { 1023 case BIO_READ: 1024 { 1025 struct g_consumer *cp; 1026 1027 if (bp->bio_error != 0) { 1028 G_MIRROR_LOGREQ(0, bp, 1029 "Synchronization request failed (error=%d).", 1030 bp->bio_error); 1031 g_destroy_bio(bp); 1032 return; 1033 } 1034 G_MIRROR_LOGREQ(3, bp, 1035 "Synchronization request half-finished."); 1036 bp->bio_cmd = BIO_WRITE; 1037 bp->bio_cflags = 0; 1038 cp = disk->d_consumer; 1039 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1040 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1041 cp->acr, cp->acw, cp->ace)); 1042 cp->index++; 1043 g_io_request(bp, cp); 1044 return; 1045 } 1046 case BIO_WRITE: 1047 { 1048 struct g_mirror_disk_sync *sync; 1049 1050 if (bp->bio_error != 0) { 1051 G_MIRROR_LOGREQ(0, bp, 1052 "Synchronization request failed (error=%d).", 1053 bp->bio_error); 1054 g_destroy_bio(bp); 1055 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 1056 g_mirror_event_send(disk, 1057 G_MIRROR_DISK_STATE_DISCONNECTED, 1058 G_MIRROR_EVENT_DONTWAIT); 1059 return; 1060 } 1061 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1062 sync = &disk->d_sync; 1063 sync->ds_offset_done = bp->bio_offset + bp->bio_length; 1064 g_destroy_bio(bp); 1065 if (sync->ds_resync != -1) 1066 break; 1067 if (sync->ds_offset_done == sc->sc_provider->mediasize) { 1068 /* 1069 * Disk up-to-date, activate it. 1070 */ 1071 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1072 G_MIRROR_EVENT_DONTWAIT); 1073 return; 1074 } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) { 1075 /* 1076 * Update offset_done on every 100 blocks. 1077 * XXX: This should be configurable. 1078 */ 1079 g_topology_lock(); 1080 g_mirror_update_metadata(disk); 1081 g_topology_unlock(); 1082 } 1083 return; 1084 } 1085 default: 1086 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1087 bp->bio_cmd, sc->sc_name)); 1088 break; 1089 } 1090} 1091 1092static void 1093g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1094{ 1095 struct g_mirror_disk *disk; 1096 struct g_consumer *cp; 1097 struct bio *cbp; 1098 1099 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1100 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1101 break; 1102 } 1103 if (disk == NULL) { 1104 if (bp->bio_error == 0) 1105 bp->bio_error = ENXIO; 1106 g_io_deliver(bp, bp->bio_error); 1107 return; 1108 } 1109 cbp = g_clone_bio(bp); 1110 if (cbp == NULL) { 1111 if (bp->bio_error == 0) 1112 bp->bio_error = ENOMEM; 1113 g_io_deliver(bp, bp->bio_error); 1114 return; 1115 } 1116 /* 1117 * Fill in the component buf structure. 1118 */ 1119 cp = disk->d_consumer; 1120 cbp->bio_done = g_mirror_done; 1121 cbp->bio_to = cp->provider; 1122 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1123 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1124 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1125 cp->acw, cp->ace)); 1126 cp->index++; 1127 g_io_request(cbp, cp); 1128} 1129 1130static void 1131g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1132{ 1133 struct g_mirror_disk *disk; 1134 struct g_consumer *cp; 1135 struct bio *cbp; 1136 1137 disk = g_mirror_get_disk(sc); 1138 if (disk == NULL) { 1139 if (bp->bio_error == 0) 1140 bp->bio_error = ENXIO; 1141 g_io_deliver(bp, bp->bio_error); 1142 return; 1143 } 1144 cbp = g_clone_bio(bp); 1145 if (cbp == NULL) { 1146 if (bp->bio_error == 0) 1147 bp->bio_error = ENOMEM; 1148 g_io_deliver(bp, bp->bio_error); 1149 return; 1150 } 1151 /* 1152 * Fill in the component buf structure. 1153 */ 1154 cp = disk->d_consumer; 1155 cbp->bio_done = g_mirror_done; 1156 cbp->bio_to = cp->provider; 1157 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1158 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1159 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1160 cp->acw, cp->ace)); 1161 cp->index++; 1162 g_io_request(cbp, cp); 1163} 1164 1165static void 1166g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1167{ 1168 struct g_mirror_disk *disk, *dp; 1169 struct g_consumer *cp; 1170 struct bio *cbp; 1171 struct bintime curtime; 1172 1173 binuptime(&curtime); 1174 /* 1175 * Find a disk which the smallest load. 1176 */ 1177 disk = NULL; 1178 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1179 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1180 continue; 1181 /* If disk wasn't used for more than 2 sec, use it. */ 1182 if (curtime.sec - dp->d_last_used.sec >= 2) { 1183 disk = dp; 1184 break; 1185 } 1186 if (disk == NULL || 1187 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1188 disk = dp; 1189 } 1190 } 1191 cbp = g_clone_bio(bp); 1192 if (cbp == NULL) { 1193 if (bp->bio_error == 0) 1194 bp->bio_error = ENOMEM; 1195 g_io_deliver(bp, bp->bio_error); 1196 return; 1197 } 1198 /* 1199 * Fill in the component buf structure. 1200 */ 1201 cp = disk->d_consumer; 1202 cbp->bio_done = g_mirror_done; 1203 cbp->bio_to = cp->provider; 1204 binuptime(&disk->d_last_used); 1205 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1206 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1207 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1208 cp->acw, cp->ace)); 1209 cp->index++; 1210 g_io_request(cbp, cp); 1211} 1212 1213static void 1214g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1215{ 1216 struct bio_queue_head queue; 1217 struct g_mirror_disk *disk; 1218 struct g_consumer *cp; 1219 struct bio *cbp; 1220 off_t left, mod, offset, slice; 1221 u_char *data; 1222 u_int ndisks; 1223 1224 if (bp->bio_length <= sc->sc_slice) { 1225 g_mirror_request_round_robin(sc, bp); 1226 return; 1227 } 1228 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1229 slice = bp->bio_length / ndisks; 1230 mod = slice % sc->sc_provider->sectorsize; 1231 if (mod != 0) 1232 slice += sc->sc_provider->sectorsize - mod; 1233 /* 1234 * Allocate all bios before sending any request, so we can 1235 * return ENOMEM in nice and clean way. 1236 */ 1237 left = bp->bio_length; 1238 offset = bp->bio_offset; 1239 data = bp->bio_data; 1240 bioq_init(&queue); 1241 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1242 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1243 continue; 1244 cbp = g_clone_bio(bp); 1245 if (cbp == NULL) { 1246 for (cbp = bioq_first(&queue); cbp != NULL; 1247 cbp = bioq_first(&queue)) { 1248 bioq_remove(&queue, cbp); 1249 g_destroy_bio(cbp); 1250 } 1251 if (bp->bio_error == 0) 1252 bp->bio_error = ENOMEM; 1253 g_io_deliver(bp, bp->bio_error); 1254 return; 1255 } 1256 bioq_insert_tail(&queue, cbp); 1257 cbp->bio_done = g_mirror_done; 1258 cbp->bio_caller1 = disk; 1259 cbp->bio_to = disk->d_consumer->provider; 1260 cbp->bio_offset = offset; 1261 cbp->bio_data = data; 1262 cbp->bio_length = MIN(left, slice); 1263 left -= cbp->bio_length; 1264 if (left == 0) 1265 break; 1266 offset += cbp->bio_length; 1267 data += cbp->bio_length; 1268 } 1269 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1270 bioq_remove(&queue, cbp); 1271 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1272 disk = cbp->bio_caller1; 1273 cbp->bio_caller1 = NULL; 1274 cp = disk->d_consumer; 1275 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1276 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1277 cp->acr, cp->acw, cp->ace)); 1278 disk->d_consumer->index++; 1279 g_io_request(cbp, disk->d_consumer); 1280 } 1281} 1282 1283static void 1284g_mirror_register_request(struct bio *bp) 1285{ 1286 struct g_mirror_softc *sc; 1287 1288 sc = bp->bio_to->geom->softc; 1289 switch (bp->bio_cmd) { 1290 case BIO_READ: 1291 switch (sc->sc_balance) { 1292 case G_MIRROR_BALANCE_LOAD: 1293 g_mirror_request_load(sc, bp); 1294 break; 1295 case G_MIRROR_BALANCE_PREFER: 1296 g_mirror_request_prefer(sc, bp); 1297 break; 1298 case G_MIRROR_BALANCE_ROUND_ROBIN: 1299 g_mirror_request_round_robin(sc, bp); 1300 break; 1301 case G_MIRROR_BALANCE_SPLIT: 1302 g_mirror_request_split(sc, bp); 1303 break; 1304 } 1305 return; 1306 case BIO_WRITE: 1307 case BIO_DELETE: 1308 { 1309 struct g_mirror_disk *disk; 1310 struct g_mirror_disk_sync *sync; 1311 struct bio_queue_head queue; 1312 struct g_consumer *cp; 1313 struct bio *cbp; 1314 1315 if (sc->sc_idle) 1316 g_mirror_unidle(sc); 1317 /* 1318 * Allocate all bios before sending any request, so we can 1319 * return ENOMEM in nice and clean way. 1320 */ 1321 bioq_init(&queue); 1322 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1323 sync = &disk->d_sync; 1324 switch (disk->d_state) { 1325 case G_MIRROR_DISK_STATE_ACTIVE: 1326 break; 1327 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1328 if (bp->bio_offset >= sync->ds_offset) 1329 continue; 1330 else if (bp->bio_offset + bp->bio_length > 1331 sync->ds_offset_done && 1332 (bp->bio_offset < sync->ds_resync || 1333 sync->ds_resync == -1)) { 1334 sync->ds_resync = bp->bio_offset - 1335 (bp->bio_offset % MAXPHYS); 1336 } 1337 break; 1338 default: 1339 continue; 1340 } 1341 cbp = g_clone_bio(bp); 1342 if (cbp == NULL) { 1343 for (cbp = bioq_first(&queue); cbp != NULL; 1344 cbp = bioq_first(&queue)) { 1345 bioq_remove(&queue, cbp); 1346 g_destroy_bio(cbp); 1347 } 1348 if (bp->bio_error == 0) 1349 bp->bio_error = ENOMEM; 1350 g_io_deliver(bp, bp->bio_error); 1351 return; 1352 } 1353 bioq_insert_tail(&queue, cbp); 1354 cbp->bio_done = g_mirror_done; 1355 cp = disk->d_consumer; 1356 cbp->bio_caller1 = cp; 1357 cbp->bio_to = cp->provider; 1358 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1359 ("Consumer %s not opened (r%dw%de%d).", 1360 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1361 } 1362 for (cbp = bioq_first(&queue); cbp != NULL; 1363 cbp = bioq_first(&queue)) { 1364 bioq_remove(&queue, cbp); 1365 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1366 cp = cbp->bio_caller1; 1367 cbp->bio_caller1 = NULL; 1368 cp->index++; 1369 g_io_request(cbp, cp); 1370 } 1371 /* 1372 * Bump syncid on first write. 1373 */ 1374 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) { 1375 sc->sc_bump_syncid = 0; 1376 g_topology_lock(); 1377 g_mirror_bump_syncid(sc); 1378 g_topology_unlock(); 1379 } 1380 return; 1381 } 1382 default: 1383 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1384 bp->bio_cmd, sc->sc_name)); 1385 break; 1386 } 1387} 1388 1389static int 1390g_mirror_can_destroy(struct g_mirror_softc *sc) 1391{ 1392 struct g_geom *gp; 1393 struct g_consumer *cp; 1394 1395 g_topology_assert(); 1396 gp = sc->sc_geom; 1397 LIST_FOREACH(cp, &gp->consumer, consumer) { 1398 if (g_mirror_is_busy(sc, cp)) 1399 return (0); 1400 } 1401 gp = sc->sc_sync.ds_geom; 1402 LIST_FOREACH(cp, &gp->consumer, consumer) { 1403 if (g_mirror_is_busy(sc, cp)) 1404 return (0); 1405 } 1406 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1407 sc->sc_name); 1408 return (1); 1409} 1410 1411static int 1412g_mirror_try_destroy(struct g_mirror_softc *sc) 1413{ 1414 1415 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1416 g_topology_lock(); 1417 if (!g_mirror_can_destroy(sc)) { 1418 g_topology_unlock(); 1419 return (0); 1420 } 1421 g_topology_unlock(); 1422 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1423 &sc->sc_worker); 1424 wakeup(&sc->sc_worker); 1425 sc->sc_worker = NULL; 1426 } else { 1427 g_topology_lock(); 1428 if (!g_mirror_can_destroy(sc)) { 1429 g_topology_unlock(); 1430 return (0); 1431 } 1432 g_mirror_destroy_device(sc); 1433 g_topology_unlock(); 1434 free(sc, M_MIRROR); 1435 } 1436 return (1); 1437} 1438 1439/* 1440 * Worker thread. 1441 */ 1442static void 1443g_mirror_worker(void *arg) 1444{ 1445 struct g_mirror_softc *sc; 1446 struct g_mirror_disk *disk; 1447 struct g_mirror_disk_sync *sync; 1448 struct g_mirror_event *ep; 1449 struct bio *bp; 1450 u_int nreqs; 1451 1452 sc = arg; 1453 curthread->td_base_pri = PRIBIO; 1454 1455 nreqs = 0; 1456 for (;;) { 1457 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1458 /* 1459 * First take a look at events. 1460 * This is important to handle events before any I/O requests. 1461 */ 1462 ep = g_mirror_event_get(sc); 1463 if (ep != NULL) { 1464 g_topology_lock(); 1465 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1466 /* Update only device status. */ 1467 G_MIRROR_DEBUG(3, 1468 "Running event for device %s.", 1469 sc->sc_name); 1470 ep->e_error = 0; 1471 g_mirror_update_device(sc, 1); 1472 } else { 1473 /* Update disk status. */ 1474 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1475 g_mirror_get_diskname(ep->e_disk)); 1476 ep->e_error = g_mirror_update_disk(ep->e_disk, 1477 ep->e_state); 1478 if (ep->e_error == 0) 1479 g_mirror_update_device(sc, 0); 1480 } 1481 g_topology_unlock(); 1482 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1483 KASSERT(ep->e_error == 0, 1484 ("Error cannot be handled.")); 1485 g_mirror_event_free(ep); 1486 } else { 1487 ep->e_flags |= G_MIRROR_EVENT_DONE; 1488 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1489 ep); 1490 mtx_lock(&sc->sc_events_mtx); 1491 wakeup(ep); 1492 mtx_unlock(&sc->sc_events_mtx); 1493 } 1494 if ((sc->sc_flags & 1495 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1496 if (g_mirror_try_destroy(sc)) 1497 kthread_exit(0); 1498 } 1499 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1500 continue; 1501 } 1502 /* 1503 * Now I/O requests. 1504 */ 1505 /* Get first request from the queue. */ 1506 mtx_lock(&sc->sc_queue_mtx); 1507 bp = bioq_first(&sc->sc_queue); 1508 if (bp == NULL) { 1509 if ((sc->sc_flags & 1510 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1511 mtx_unlock(&sc->sc_queue_mtx); 1512 if (g_mirror_try_destroy(sc)) 1513 kthread_exit(0); 1514 mtx_lock(&sc->sc_queue_mtx); 1515 } 1516 } 1517 if (sc->sc_sync.ds_ndisks > 0 && 1518 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1519 mtx_unlock(&sc->sc_queue_mtx); 1520 /* 1521 * It is time for synchronization... 1522 */ 1523 nreqs = 0; 1524 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1525 if (disk->d_state != 1526 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1527 continue; 1528 } 1529 sync = &disk->d_sync; 1530 if (sync->ds_offset >= 1531 sc->sc_provider->mediasize) { 1532 continue; 1533 } 1534 if (sync->ds_offset > sync->ds_offset_done) 1535 continue; 1536 if (sync->ds_resync != -1) { 1537 sync->ds_offset = sync->ds_resync; 1538 sync->ds_offset_done = sync->ds_resync; 1539 sync->ds_resync = -1; 1540 } 1541 g_mirror_sync_one(disk); 1542 } 1543 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1544 goto sleep; 1545 } 1546 if (bp == NULL) { 1547 if (g_mirror_check_idle(sc)) { 1548 u_int idletime; 1549 1550 idletime = g_mirror_idletime; 1551 if (idletime == 0) 1552 idletime = 1; 1553 idletime *= hz; 1554 if (msleep(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1555 "m:w1", idletime) == EWOULDBLOCK) { 1556 G_MIRROR_DEBUG(5, "%s: I'm here 3.", 1557 __func__); 1558 /* 1559 * No I/O requests in 'idletime' seconds, 1560 * so mark components as clean. 1561 */ 1562 g_mirror_idle(sc); 1563 } 1564 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1565 } else { 1566 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1567 "m:w2", 0); 1568 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1569 } 1570 continue; 1571 } 1572 nreqs++; 1573 bioq_remove(&sc->sc_queue, bp); 1574 mtx_unlock(&sc->sc_queue_mtx); 1575 1576 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1577 g_mirror_regular_request(bp); 1578 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1579 u_int timeout, sps; 1580 1581 g_mirror_sync_request(bp); 1582sleep: 1583 sps = g_mirror_syncs_per_sec; 1584 if (sps == 0) { 1585 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1586 continue; 1587 } 1588 mtx_lock(&sc->sc_queue_mtx); 1589 if (bioq_first(&sc->sc_queue) != NULL) { 1590 mtx_unlock(&sc->sc_queue_mtx); 1591 G_MIRROR_DEBUG(5, "%s: I'm here 7.", __func__); 1592 continue; 1593 } 1594 timeout = hz / sps; 1595 if (timeout == 0) 1596 timeout = 1; 1597 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w3", 1598 timeout); 1599 } else { 1600 g_mirror_register_request(bp); 1601 } 1602 G_MIRROR_DEBUG(5, "%s: I'm here 8.", __func__); 1603 } 1604} 1605 1606/* 1607 * Open disk's consumer if needed. 1608 */ 1609static void 1610g_mirror_update_access(struct g_mirror_disk *disk) 1611{ 1612 struct g_provider *pp; 1613 1614 g_topology_assert(); 1615 1616 pp = disk->d_softc->sc_provider; 1617 if (pp == NULL) 1618 return; 1619 if (pp->acw > 0) { 1620 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1621 G_MIRROR_DEBUG(1, 1622 "Disk %s (device %s) marked as dirty.", 1623 g_mirror_get_diskname(disk), 1624 disk->d_softc->sc_name); 1625 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1626 } 1627 } else if (pp->acw == 0) { 1628 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1629 G_MIRROR_DEBUG(1, 1630 "Disk %s (device %s) marked as clean.", 1631 g_mirror_get_diskname(disk), 1632 disk->d_softc->sc_name); 1633 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1634 } 1635 } 1636} 1637 1638static void 1639g_mirror_sync_start(struct g_mirror_disk *disk) 1640{ 1641 struct g_mirror_softc *sc; 1642 struct g_consumer *cp; 1643 int error; 1644 1645 g_topology_assert(); 1646 1647 sc = disk->d_softc; 1648 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1649 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1650 sc->sc_state)); 1651 cp = disk->d_consumer; 1652 1653 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1654 g_mirror_get_diskname(disk)); 1655 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1656 KASSERT(disk->d_sync.ds_consumer == NULL, 1657 ("Sync consumer already exists (device=%s, disk=%s).", 1658 sc->sc_name, g_mirror_get_diskname(disk))); 1659 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1660 disk->d_sync.ds_consumer->private = disk; 1661 disk->d_sync.ds_consumer->index = 0; 1662 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1663 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1664 disk->d_softc->sc_name, error)); 1665 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1666 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1667 disk->d_softc->sc_name, error)); 1668 disk->d_sync.ds_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1669 sc->sc_sync.ds_ndisks++; 1670} 1671 1672/* 1673 * Stop synchronization process. 1674 * type: 0 - synchronization finished 1675 * 1 - synchronization stopped 1676 */ 1677static void 1678g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1679{ 1680 struct g_consumer *cp; 1681 1682 g_topology_assert(); 1683 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1684 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1685 g_mirror_disk_state2str(disk->d_state))); 1686 if (disk->d_sync.ds_consumer == NULL) 1687 return; 1688 1689 if (type == 0) { 1690 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1691 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1692 } else /* if (type == 1) */ { 1693 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1694 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1695 } 1696 cp = disk->d_sync.ds_consumer; 1697 g_mirror_kill_consumer(disk->d_softc, cp); 1698 free(disk->d_sync.ds_data, M_MIRROR); 1699 disk->d_sync.ds_consumer = NULL; 1700 disk->d_softc->sc_sync.ds_ndisks--; 1701 cp = disk->d_consumer; 1702 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1703} 1704 1705static void 1706g_mirror_launch_provider(struct g_mirror_softc *sc) 1707{ 1708 struct g_mirror_disk *disk; 1709 struct g_provider *pp; 1710 1711 g_topology_assert(); 1712 1713 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1714 pp->mediasize = sc->sc_mediasize; 1715 pp->sectorsize = sc->sc_sectorsize; 1716 sc->sc_provider = pp; 1717 g_error_provider(pp, 0); 1718 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1719 pp->name); 1720 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1721 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1722 g_mirror_sync_start(disk); 1723 } 1724} 1725 1726static void 1727g_mirror_destroy_provider(struct g_mirror_softc *sc) 1728{ 1729 struct g_mirror_disk *disk; 1730 struct bio *bp; 1731 1732 g_topology_assert(); 1733 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1734 sc->sc_name)); 1735 1736 g_error_provider(sc->sc_provider, ENXIO); 1737 mtx_lock(&sc->sc_queue_mtx); 1738 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1739 bioq_remove(&sc->sc_queue, bp); 1740 g_io_deliver(bp, ENXIO); 1741 } 1742 mtx_unlock(&sc->sc_queue_mtx); 1743 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1744 sc->sc_provider->name); 1745 sc->sc_provider->flags |= G_PF_WITHER; 1746 g_orphan_provider(sc->sc_provider, ENXIO); 1747 sc->sc_provider = NULL; 1748 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1749 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1750 g_mirror_sync_stop(disk, 1); 1751 } 1752} 1753 1754static void 1755g_mirror_go(void *arg) 1756{ 1757 struct g_mirror_softc *sc; 1758 1759 sc = arg; 1760 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1761 g_mirror_event_send(sc, 0, 1762 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1763} 1764 1765static u_int 1766g_mirror_determine_state(struct g_mirror_disk *disk) 1767{ 1768 struct g_mirror_softc *sc; 1769 u_int state; 1770 1771 sc = disk->d_softc; 1772 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1773 if ((disk->d_flags & 1774 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1775 /* Disk does not need synchronization. */ 1776 state = G_MIRROR_DISK_STATE_ACTIVE; 1777 } else { 1778 if ((sc->sc_flags & 1779 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1780 (disk->d_flags & 1781 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1782 /* 1783 * We can start synchronization from 1784 * the stored offset. 1785 */ 1786 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1787 } else { 1788 state = G_MIRROR_DISK_STATE_STALE; 1789 } 1790 } 1791 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1792 /* 1793 * Reset all synchronization data for this disk, 1794 * because if it even was synchronized, it was 1795 * synchronized to disks with different syncid. 1796 */ 1797 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1798 disk->d_sync.ds_offset = 0; 1799 disk->d_sync.ds_offset_done = 0; 1800 disk->d_sync.ds_syncid = sc->sc_syncid; 1801 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1802 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1803 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1804 } else { 1805 state = G_MIRROR_DISK_STATE_STALE; 1806 } 1807 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1808 /* 1809 * Not good, NOT GOOD! 1810 * It means that mirror was started on stale disks 1811 * and more fresh disk just arrive. 1812 * If there were writes, mirror is fucked up, sorry. 1813 * I think the best choice here is don't touch 1814 * this disk and inform the user laudly. 1815 */ 1816 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1817 "disk (%s) arrives!! It will not be connected to the " 1818 "running device.", sc->sc_name, 1819 g_mirror_get_diskname(disk)); 1820 g_mirror_destroy_disk(disk); 1821 state = G_MIRROR_DISK_STATE_NONE; 1822 /* Return immediately, because disk was destroyed. */ 1823 return (state); 1824 } 1825 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1826 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1827 return (state); 1828} 1829 1830/* 1831 * Update device state. 1832 */ 1833static void 1834g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1835{ 1836 struct g_mirror_disk *disk; 1837 u_int state; 1838 1839 g_topology_assert(); 1840 1841 switch (sc->sc_state) { 1842 case G_MIRROR_DEVICE_STATE_STARTING: 1843 { 1844 struct g_mirror_disk *pdisk; 1845 u_int dirty, ndisks, syncid; 1846 1847 KASSERT(sc->sc_provider == NULL, 1848 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1849 /* 1850 * Are we ready? We are, if all disks are connected or 1851 * if we have any disks and 'force' is true. 1852 */ 1853 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1854 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1855 ; 1856 } else if (g_mirror_ndisks(sc, -1) == 0) { 1857 /* 1858 * Disks went down in starting phase, so destroy 1859 * device. 1860 */ 1861 callout_drain(&sc->sc_callout); 1862 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1863 return; 1864 } else { 1865 return; 1866 } 1867 1868 /* 1869 * Activate all disks with the biggest syncid. 1870 */ 1871 if (force) { 1872 /* 1873 * If 'force' is true, we have been called due to 1874 * timeout, so don't bother canceling timeout. 1875 */ 1876 ndisks = 0; 1877 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1878 if ((disk->d_flags & 1879 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1880 ndisks++; 1881 } 1882 } 1883 if (ndisks == 0) { 1884 /* No valid disks found, destroy device. */ 1885 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1886 return; 1887 } 1888 } else { 1889 /* Cancel timeout. */ 1890 callout_drain(&sc->sc_callout); 1891 } 1892 1893 /* 1894 * Find disk with the biggest syncid. 1895 */ 1896 syncid = 0; 1897 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1898 if (disk->d_sync.ds_syncid > syncid) 1899 syncid = disk->d_sync.ds_syncid; 1900 } 1901 1902 /* 1903 * Here we need to look for dirty disks and if all disks 1904 * with the biggest syncid are dirty, we have to choose 1905 * one with the biggest priority and rebuild the rest. 1906 */ 1907 /* 1908 * Find the number of dirty disks with the biggest syncid. 1909 * Find the number of disks with the biggest syncid. 1910 * While here, find a disk with the biggest priority. 1911 */ 1912 dirty = ndisks = 0; 1913 pdisk = NULL; 1914 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1915 if (disk->d_sync.ds_syncid != syncid) 1916 continue; 1917 if ((disk->d_flags & 1918 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1919 continue; 1920 } 1921 ndisks++; 1922 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1923 dirty++; 1924 if (pdisk == NULL || 1925 pdisk->d_priority < disk->d_priority) { 1926 pdisk = disk; 1927 } 1928 } 1929 } 1930 if (dirty == 0) { 1931 /* No dirty disks at all, great. */ 1932 } else if (dirty == ndisks) { 1933 /* 1934 * Force synchronization for all dirty disks except one 1935 * with the biggest priority. 1936 */ 1937 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1938 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1939 "master disk for synchronization.", 1940 g_mirror_get_diskname(pdisk), sc->sc_name); 1941 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1942 if (disk->d_sync.ds_syncid != syncid) 1943 continue; 1944 if ((disk->d_flags & 1945 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1946 continue; 1947 } 1948 KASSERT((disk->d_flags & 1949 G_MIRROR_DISK_FLAG_DIRTY) != 0, 1950 ("Disk %s isn't marked as dirty.", 1951 g_mirror_get_diskname(disk))); 1952 /* Skip the disk with the biggest priority. */ 1953 if (disk == pdisk) 1954 continue; 1955 disk->d_sync.ds_syncid = 0; 1956 } 1957 } else if (dirty < ndisks) { 1958 /* 1959 * Force synchronization for all dirty disks. 1960 * We have some non-dirty disks. 1961 */ 1962 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1963 if (disk->d_sync.ds_syncid != syncid) 1964 continue; 1965 if ((disk->d_flags & 1966 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1967 continue; 1968 } 1969 if ((disk->d_flags & 1970 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1971 continue; 1972 } 1973 disk->d_sync.ds_syncid = 0; 1974 } 1975 } 1976 1977 /* Reset hint. */ 1978 sc->sc_hint = NULL; 1979 sc->sc_syncid = syncid; 1980 if (force) { 1981 /* Remember to bump syncid on first write. */ 1982 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1983 } 1984 state = G_MIRROR_DEVICE_STATE_RUNNING; 1985 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 1986 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 1987 g_mirror_device_state2str(state)); 1988 sc->sc_state = state; 1989 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1990 state = g_mirror_determine_state(disk); 1991 g_mirror_event_send(disk, state, 1992 G_MIRROR_EVENT_DONTWAIT); 1993 if (state == G_MIRROR_DISK_STATE_STALE) { 1994 sc->sc_bump_syncid = 1995 G_MIRROR_BUMP_ON_FIRST_WRITE; 1996 } 1997 } 1998 wakeup(&g_mirror_class); 1999 break; 2000 } 2001 case G_MIRROR_DEVICE_STATE_RUNNING: 2002 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2003 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2004 /* 2005 * No active disks or no disks at all, 2006 * so destroy device. 2007 */ 2008 if (sc->sc_provider != NULL) 2009 g_mirror_destroy_provider(sc); 2010 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2011 break; 2012 } else if (g_mirror_ndisks(sc, 2013 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2014 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2015 /* 2016 * We have active disks, launch provider if it doesn't 2017 * exist. 2018 */ 2019 if (sc->sc_provider == NULL) 2020 g_mirror_launch_provider(sc); 2021 } 2022 /* 2023 * Bump syncid here, if we need to do it immediately. 2024 */ 2025 if (sc->sc_bump_syncid == G_MIRROR_BUMP_IMMEDIATELY) { 2026 sc->sc_bump_syncid = 0; 2027 g_mirror_bump_syncid(sc); 2028 } 2029 break; 2030 default: 2031 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2032 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2033 break; 2034 } 2035} 2036 2037/* 2038 * Update disk state and device state if needed. 2039 */ 2040#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2041 "Disk %s state changed from %s to %s (device %s).", \ 2042 g_mirror_get_diskname(disk), \ 2043 g_mirror_disk_state2str(disk->d_state), \ 2044 g_mirror_disk_state2str(state), sc->sc_name) 2045static int 2046g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2047{ 2048 struct g_mirror_softc *sc; 2049 2050 g_topology_assert(); 2051 2052 sc = disk->d_softc; 2053again: 2054 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2055 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2056 g_mirror_disk_state2str(state)); 2057 switch (state) { 2058 case G_MIRROR_DISK_STATE_NEW: 2059 /* 2060 * Possible scenarios: 2061 * 1. New disk arrive. 2062 */ 2063 /* Previous state should be NONE. */ 2064 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2065 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2066 g_mirror_disk_state2str(disk->d_state))); 2067 DISK_STATE_CHANGED(); 2068 2069 disk->d_state = state; 2070 if (LIST_EMPTY(&sc->sc_disks)) 2071 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2072 else { 2073 struct g_mirror_disk *dp; 2074 2075 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2076 if (disk->d_priority >= dp->d_priority) { 2077 LIST_INSERT_BEFORE(dp, disk, d_next); 2078 dp = NULL; 2079 break; 2080 } 2081 if (LIST_NEXT(dp, d_next) == NULL) 2082 break; 2083 } 2084 if (dp != NULL) 2085 LIST_INSERT_AFTER(dp, disk, d_next); 2086 } 2087 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2088 sc->sc_name, g_mirror_get_diskname(disk)); 2089 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2090 break; 2091 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2092 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2093 g_mirror_device_state2str(sc->sc_state), 2094 g_mirror_get_diskname(disk), 2095 g_mirror_disk_state2str(disk->d_state))); 2096 state = g_mirror_determine_state(disk); 2097 if (state != G_MIRROR_DISK_STATE_NONE) 2098 goto again; 2099 break; 2100 case G_MIRROR_DISK_STATE_ACTIVE: 2101 /* 2102 * Possible scenarios: 2103 * 1. New disk does not need synchronization. 2104 * 2. Synchronization process finished successfully. 2105 */ 2106 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2107 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2108 g_mirror_device_state2str(sc->sc_state), 2109 g_mirror_get_diskname(disk), 2110 g_mirror_disk_state2str(disk->d_state))); 2111 /* Previous state should be NEW or SYNCHRONIZING. */ 2112 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2113 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2114 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2115 g_mirror_disk_state2str(disk->d_state))); 2116 DISK_STATE_CHANGED(); 2117 2118 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2119 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2120 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2121 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2122 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2123 g_mirror_sync_stop(disk, 0); 2124 } 2125 disk->d_state = state; 2126 disk->d_sync.ds_offset = 0; 2127 disk->d_sync.ds_offset_done = 0; 2128 g_mirror_update_access(disk); 2129 g_mirror_update_metadata(disk); 2130 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2131 sc->sc_name, g_mirror_get_diskname(disk)); 2132 break; 2133 case G_MIRROR_DISK_STATE_STALE: 2134 /* 2135 * Possible scenarios: 2136 * 1. Stale disk was connected. 2137 */ 2138 /* Previous state should be NEW. */ 2139 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2140 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2141 g_mirror_disk_state2str(disk->d_state))); 2142 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2143 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2144 g_mirror_device_state2str(sc->sc_state), 2145 g_mirror_get_diskname(disk), 2146 g_mirror_disk_state2str(disk->d_state))); 2147 /* 2148 * STALE state is only possible if device is marked 2149 * NOAUTOSYNC. 2150 */ 2151 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2152 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2153 g_mirror_device_state2str(sc->sc_state), 2154 g_mirror_get_diskname(disk), 2155 g_mirror_disk_state2str(disk->d_state))); 2156 DISK_STATE_CHANGED(); 2157 2158 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2159 disk->d_state = state; 2160 g_mirror_update_metadata(disk); 2161 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2162 sc->sc_name, g_mirror_get_diskname(disk)); 2163 break; 2164 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2165 /* 2166 * Possible scenarios: 2167 * 1. Disk which needs synchronization was connected. 2168 */ 2169 /* Previous state should be NEW. */ 2170 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2171 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2172 g_mirror_disk_state2str(disk->d_state))); 2173 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2174 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2175 g_mirror_device_state2str(sc->sc_state), 2176 g_mirror_get_diskname(disk), 2177 g_mirror_disk_state2str(disk->d_state))); 2178 DISK_STATE_CHANGED(); 2179 2180 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2181 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2182 disk->d_state = state; 2183 if (sc->sc_provider != NULL) { 2184 g_mirror_sync_start(disk); 2185 g_mirror_update_metadata(disk); 2186 } 2187 break; 2188 case G_MIRROR_DISK_STATE_DISCONNECTED: 2189 /* 2190 * Possible scenarios: 2191 * 1. Device wasn't running yet, but disk disappear. 2192 * 2. Disk was active and disapppear. 2193 * 3. Disk disappear during synchronization process. 2194 */ 2195 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2196 /* 2197 * Previous state should be ACTIVE, STALE or 2198 * SYNCHRONIZING. 2199 */ 2200 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2201 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2202 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2203 ("Wrong disk state (%s, %s).", 2204 g_mirror_get_diskname(disk), 2205 g_mirror_disk_state2str(disk->d_state))); 2206 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2207 /* Previous state should be NEW. */ 2208 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2209 ("Wrong disk state (%s, %s).", 2210 g_mirror_get_diskname(disk), 2211 g_mirror_disk_state2str(disk->d_state))); 2212 /* 2213 * Reset bumping syncid if disk disappeared in STARTING 2214 * state. 2215 */ 2216 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) 2217 sc->sc_bump_syncid = 0; 2218#ifdef INVARIANTS 2219 } else { 2220 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2221 sc->sc_name, 2222 g_mirror_device_state2str(sc->sc_state), 2223 g_mirror_get_diskname(disk), 2224 g_mirror_disk_state2str(disk->d_state))); 2225#endif 2226 } 2227 DISK_STATE_CHANGED(); 2228 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2229 sc->sc_name, g_mirror_get_diskname(disk)); 2230 2231 g_mirror_destroy_disk(disk); 2232 break; 2233 case G_MIRROR_DISK_STATE_DESTROY: 2234 { 2235 int error; 2236 2237 error = g_mirror_clear_metadata(disk); 2238 if (error != 0) 2239 return (error); 2240 DISK_STATE_CHANGED(); 2241 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2242 sc->sc_name, g_mirror_get_diskname(disk)); 2243 2244 g_mirror_destroy_disk(disk); 2245 sc->sc_ndisks--; 2246 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2247 g_mirror_update_metadata(disk); 2248 } 2249 break; 2250 } 2251 default: 2252 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2253 break; 2254 } 2255 return (0); 2256} 2257#undef DISK_STATE_CHANGED 2258 2259static int 2260g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2261{ 2262 struct g_provider *pp; 2263 u_char *buf; 2264 int error; 2265 2266 g_topology_assert(); 2267 2268 error = g_access(cp, 1, 0, 0); 2269 if (error != 0) 2270 return (error); 2271 pp = cp->provider; 2272 g_topology_unlock(); 2273 /* Metadata are stored on last sector. */ 2274 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2275 &error); 2276 g_topology_lock(); 2277 g_access(cp, -1, 0, 0); 2278 if (error != 0) { 2279 if (buf != NULL) 2280 g_free(buf); 2281 return (error); 2282 } 2283 2284 /* Decode metadata. */ 2285 error = mirror_metadata_decode(buf, md); 2286 g_free(buf); 2287 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2288 return (EINVAL); 2289 if (error != 0) { 2290 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2291 cp->provider->name); 2292 return (error); 2293 } 2294 2295 return (0); 2296} 2297 2298static int 2299g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2300 struct g_mirror_metadata *md) 2301{ 2302 2303 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2304 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2305 pp->name, md->md_did); 2306 return (EEXIST); 2307 } 2308 if (md->md_all != sc->sc_ndisks) { 2309 G_MIRROR_DEBUG(1, 2310 "Invalid '%s' field on disk %s (device %s), skipping.", 2311 "md_all", pp->name, sc->sc_name); 2312 return (EINVAL); 2313 } 2314 if (md->md_slice != sc->sc_slice) { 2315 G_MIRROR_DEBUG(1, 2316 "Invalid '%s' field on disk %s (device %s), skipping.", 2317 "md_slice", pp->name, sc->sc_name); 2318 return (EINVAL); 2319 } 2320 if (md->md_balance != sc->sc_balance) { 2321 G_MIRROR_DEBUG(1, 2322 "Invalid '%s' field on disk %s (device %s), skipping.", 2323 "md_balance", pp->name, sc->sc_name); 2324 return (EINVAL); 2325 } 2326 if (md->md_mediasize != sc->sc_mediasize) { 2327 G_MIRROR_DEBUG(1, 2328 "Invalid '%s' field on disk %s (device %s), skipping.", 2329 "md_mediasize", pp->name, sc->sc_name); 2330 return (EINVAL); 2331 } 2332 if (sc->sc_mediasize > pp->mediasize) { 2333 G_MIRROR_DEBUG(1, 2334 "Invalid size of disk %s (device %s), skipping.", pp->name, 2335 sc->sc_name); 2336 return (EINVAL); 2337 } 2338 if (md->md_sectorsize != sc->sc_sectorsize) { 2339 G_MIRROR_DEBUG(1, 2340 "Invalid '%s' field on disk %s (device %s), skipping.", 2341 "md_sectorsize", pp->name, sc->sc_name); 2342 return (EINVAL); 2343 } 2344 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2345 G_MIRROR_DEBUG(1, 2346 "Invalid sector size of disk %s (device %s), skipping.", 2347 pp->name, sc->sc_name); 2348 return (EINVAL); 2349 } 2350 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2351 G_MIRROR_DEBUG(1, 2352 "Invalid device flags on disk %s (device %s), skipping.", 2353 pp->name, sc->sc_name); 2354 return (EINVAL); 2355 } 2356 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2357 G_MIRROR_DEBUG(1, 2358 "Invalid disk flags on disk %s (device %s), skipping.", 2359 pp->name, sc->sc_name); 2360 return (EINVAL); 2361 } 2362 return (0); 2363} 2364 2365static int 2366g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2367 struct g_mirror_metadata *md) 2368{ 2369 struct g_mirror_disk *disk; 2370 int error; 2371 2372 g_topology_assert(); 2373 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2374 2375 error = g_mirror_check_metadata(sc, pp, md); 2376 if (error != 0) 2377 return (error); 2378 disk = g_mirror_init_disk(sc, pp, md, &error); 2379 if (disk == NULL) 2380 return (error); 2381 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2382 G_MIRROR_EVENT_WAIT); 2383 return (error); 2384} 2385 2386static int 2387g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2388{ 2389 struct g_mirror_softc *sc; 2390 struct g_mirror_disk *disk; 2391 int dcr, dcw, dce; 2392 2393 g_topology_assert(); 2394 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2395 acw, ace); 2396 2397 dcr = pp->acr + acr; 2398 dcw = pp->acw + acw; 2399 dce = pp->ace + ace; 2400 2401 sc = pp->geom->softc; 2402 if (sc == NULL || LIST_EMPTY(&sc->sc_disks) || 2403 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 2404 if (acr <= 0 && acw <= 0 && ace <= 0) 2405 return (0); 2406 else 2407 return (ENXIO); 2408 } 2409 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2410 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2411 continue; 2412 /* 2413 * Mark disk as dirty on open and unmark on close. 2414 */ 2415 if (pp->acw == 0 && dcw > 0) { 2416 G_MIRROR_DEBUG(1, 2417 "Disk %s (device %s) marked as dirty.", 2418 g_mirror_get_diskname(disk), sc->sc_name); 2419 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2420 g_mirror_update_metadata(disk); 2421 } else if (pp->acw > 0 && dcw == 0) { 2422 G_MIRROR_DEBUG(1, 2423 "Disk %s (device %s) marked as clean.", 2424 g_mirror_get_diskname(disk), sc->sc_name); 2425 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2426 g_mirror_update_metadata(disk); 2427 } 2428 } 2429 return (0); 2430} 2431 2432static struct g_geom * 2433g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2434{ 2435 struct g_mirror_softc *sc; 2436 struct g_geom *gp; 2437 int error, timeout; 2438 2439 g_topology_assert(); 2440 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2441 md->md_mid); 2442 2443 /* One disk is minimum. */ 2444 if (md->md_all < 1) 2445 return (NULL); 2446 /* 2447 * Action geom. 2448 */ 2449 gp = g_new_geomf(mp, "%s", md->md_name); 2450 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2451 gp->start = g_mirror_start; 2452 gp->spoiled = g_mirror_spoiled; 2453 gp->orphan = g_mirror_orphan; 2454 gp->access = g_mirror_access; 2455 gp->dumpconf = g_mirror_dumpconf; 2456 2457 sc->sc_id = md->md_mid; 2458 sc->sc_slice = md->md_slice; 2459 sc->sc_balance = md->md_balance; 2460 sc->sc_mediasize = md->md_mediasize; 2461 sc->sc_sectorsize = md->md_sectorsize; 2462 sc->sc_ndisks = md->md_all; 2463 sc->sc_flags = md->md_mflags; 2464 sc->sc_bump_syncid = 0; 2465 sc->sc_idle = 0; 2466 bioq_init(&sc->sc_queue); 2467 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2468 LIST_INIT(&sc->sc_disks); 2469 TAILQ_INIT(&sc->sc_events); 2470 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2471 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2472 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2473 gp->softc = sc; 2474 sc->sc_geom = gp; 2475 sc->sc_provider = NULL; 2476 /* 2477 * Synchronization geom. 2478 */ 2479 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2480 gp->softc = sc; 2481 gp->orphan = g_mirror_orphan; 2482 sc->sc_sync.ds_geom = gp; 2483 sc->sc_sync.ds_ndisks = 0; 2484 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2485 "g_mirror %s", md->md_name); 2486 if (error != 0) { 2487 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2488 sc->sc_name); 2489 g_destroy_geom(sc->sc_sync.ds_geom); 2490 mtx_destroy(&sc->sc_events_mtx); 2491 mtx_destroy(&sc->sc_queue_mtx); 2492 g_destroy_geom(sc->sc_geom); 2493 free(sc, M_MIRROR); 2494 return (NULL); 2495 } 2496 2497 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2498 2499 /* 2500 * Run timeout. 2501 */ 2502 timeout = g_mirror_timeout * hz; 2503 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2504 return (sc->sc_geom); 2505} 2506 2507int 2508g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2509{ 2510 struct g_provider *pp; 2511 2512 g_topology_assert(); 2513 2514 if (sc == NULL) 2515 return (ENXIO); 2516 pp = sc->sc_provider; 2517 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2518 if (force) { 2519 G_MIRROR_DEBUG(0, "Device %s is still open, so it " 2520 "can't be definitely removed.", pp->name); 2521 } else { 2522 G_MIRROR_DEBUG(1, 2523 "Device %s is still open (r%dw%de%d).", pp->name, 2524 pp->acr, pp->acw, pp->ace); 2525 return (EBUSY); 2526 } 2527 } 2528 2529 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2530 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2531 g_topology_unlock(); 2532 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2533 mtx_lock(&sc->sc_queue_mtx); 2534 wakeup(sc); 2535 mtx_unlock(&sc->sc_queue_mtx); 2536 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2537 while (sc->sc_worker != NULL) 2538 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2539 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2540 g_topology_lock(); 2541 g_mirror_destroy_device(sc); 2542 free(sc, M_MIRROR); 2543 return (0); 2544} 2545 2546static void 2547g_mirror_taste_orphan(struct g_consumer *cp) 2548{ 2549 2550 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2551 cp->provider->name)); 2552} 2553 2554static struct g_geom * 2555g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2556{ 2557 struct g_mirror_metadata md; 2558 struct g_mirror_softc *sc; 2559 struct g_consumer *cp; 2560 struct g_geom *gp; 2561 int error; 2562 2563 g_topology_assert(); 2564 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2565 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2566 2567 gp = g_new_geomf(mp, "mirror:taste"); 2568 /* 2569 * This orphan function should be never called. 2570 */ 2571 gp->orphan = g_mirror_taste_orphan; 2572 cp = g_new_consumer(gp); 2573 g_attach(cp, pp); 2574 error = g_mirror_read_metadata(cp, &md); 2575 g_detach(cp); 2576 g_destroy_consumer(cp); 2577 g_destroy_geom(gp); 2578 if (error != 0) 2579 return (NULL); 2580 gp = NULL; 2581 2582 if (md.md_version > G_MIRROR_VERSION) { 2583 printf("geom_mirror.ko module is too old to handle %s.\n", 2584 pp->name); 2585 return (NULL); 2586 } 2587 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2588 return (NULL); 2589 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2590 G_MIRROR_DEBUG(0, 2591 "Device %s: provider %s marked as inactive, skipping.", 2592 md.md_name, pp->name); 2593 return (NULL); 2594 } 2595 if (g_mirror_debug >= 2) 2596 mirror_metadata_dump(&md); 2597 2598 /* 2599 * Let's check if device already exists. 2600 */ 2601 sc = NULL; 2602 LIST_FOREACH(gp, &mp->geom, geom) { 2603 sc = gp->softc; 2604 if (sc == NULL) 2605 continue; 2606 if (sc->sc_sync.ds_geom == gp) 2607 continue; 2608 if (strcmp(md.md_name, sc->sc_name) != 0) 2609 continue; 2610 if (md.md_mid != sc->sc_id) { 2611 G_MIRROR_DEBUG(0, "Device %s already configured.", 2612 sc->sc_name); 2613 return (NULL); 2614 } 2615 break; 2616 } 2617 if (gp == NULL) { 2618 gp = g_mirror_create(mp, &md); 2619 if (gp == NULL) { 2620 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2621 md.md_name); 2622 return (NULL); 2623 } 2624 sc = gp->softc; 2625 } 2626 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2627 error = g_mirror_add_disk(sc, pp, &md); 2628 if (error != 0) { 2629 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2630 pp->name, gp->name, error); 2631 if (LIST_EMPTY(&sc->sc_disks)) 2632 g_mirror_destroy(sc, 1); 2633 return (NULL); 2634 } 2635 return (gp); 2636} 2637 2638static int 2639g_mirror_destroy_geom(struct gctl_req *req __unused, 2640 struct g_class *mp __unused, struct g_geom *gp) 2641{ 2642 2643 return (g_mirror_destroy(gp->softc, 0)); 2644} 2645 2646static void 2647g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2648 struct g_consumer *cp, struct g_provider *pp) 2649{ 2650 struct g_mirror_softc *sc; 2651 2652 g_topology_assert(); 2653 2654 sc = gp->softc; 2655 if (sc == NULL) 2656 return; 2657 /* Skip synchronization geom. */ 2658 if (gp == sc->sc_sync.ds_geom) 2659 return; 2660 if (pp != NULL) { 2661 /* Nothing here. */ 2662 } else if (cp != NULL) { 2663 struct g_mirror_disk *disk; 2664 2665 disk = cp->private; 2666 if (disk == NULL) 2667 return; 2668 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2669 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2670 sbuf_printf(sb, "%s<Synchronized>", indent); 2671 if (disk->d_sync.ds_offset_done == 0) 2672 sbuf_printf(sb, "0%%"); 2673 else { 2674 sbuf_printf(sb, "%u%%", 2675 (u_int)((disk->d_sync.ds_offset_done * 100) / 2676 sc->sc_provider->mediasize)); 2677 } 2678 sbuf_printf(sb, "</Synchronized>\n"); 2679 } 2680 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2681 disk->d_sync.ds_syncid); 2682 sbuf_printf(sb, "%s<Flags>", indent); 2683 if (disk->d_flags == 0) 2684 sbuf_printf(sb, "NONE"); 2685 else { 2686 int first = 1; 2687 2688#define ADD_FLAG(flag, name) do { \ 2689 if ((disk->d_flags & (flag)) != 0) { \ 2690 if (!first) \ 2691 sbuf_printf(sb, ", "); \ 2692 else \ 2693 first = 0; \ 2694 sbuf_printf(sb, name); \ 2695 } \ 2696} while (0) 2697 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2698 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2699 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2700 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2701 "SYNCHRONIZING"); 2702 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2703#undef ADD_FLAG 2704 } 2705 sbuf_printf(sb, "</Flags>\n"); 2706 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2707 disk->d_priority); 2708 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2709 g_mirror_disk_state2str(disk->d_state)); 2710 } else { 2711 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2712 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2713 sbuf_printf(sb, "%s<Flags>", indent); 2714 if (sc->sc_flags == 0) 2715 sbuf_printf(sb, "NONE"); 2716 else { 2717 int first = 1; 2718 2719#define ADD_FLAG(flag, name) do { \ 2720 if ((sc->sc_flags & (flag)) != 0) { \ 2721 if (!first) \ 2722 sbuf_printf(sb, ", "); \ 2723 else \ 2724 first = 0; \ 2725 sbuf_printf(sb, name); \ 2726 } \ 2727} while (0) 2728 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2729#undef ADD_FLAG 2730 } 2731 sbuf_printf(sb, "</Flags>\n"); 2732 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2733 (u_int)sc->sc_slice); 2734 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2735 balance_name(sc->sc_balance)); 2736 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2737 sc->sc_ndisks); 2738 sbuf_printf(sb, "%s<State>", indent); 2739 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2740 sbuf_printf(sb, "%s", "STARTING"); 2741 else if (sc->sc_ndisks == 2742 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 2743 sbuf_printf(sb, "%s", "COMPLETE"); 2744 else 2745 sbuf_printf(sb, "%s", "DEGRADED"); 2746 sbuf_printf(sb, "</State>\n"); 2747 } 2748} 2749 2750static void 2751g_mirror_shutdown(void *arg, int howto) 2752{ 2753 struct g_class *mp; 2754 struct g_geom *gp, *gp2; 2755 2756 mp = arg; 2757 DROP_GIANT(); 2758 g_topology_lock(); 2759 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2760 if (gp->softc == NULL) 2761 continue; 2762 g_mirror_destroy(gp->softc, 1); 2763 } 2764 g_topology_unlock(); 2765 PICKUP_GIANT(); 2766#if 0 2767 tsleep(&gp, PRIBIO, "m:shutdown", hz * 20); 2768#endif 2769} 2770 2771static void 2772g_mirror_init(struct g_class *mp) 2773{ 2774 2775 g_mirror_ehtag = EVENTHANDLER_REGISTER(shutdown_post_sync, 2776 g_mirror_shutdown, mp, SHUTDOWN_PRI_FIRST); 2777 if (g_mirror_ehtag == NULL) 2778 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 2779} 2780 2781static void 2782g_mirror_fini(struct g_class *mp) 2783{ 2784 2785 if (g_mirror_ehtag == NULL) 2786 return; 2787 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_ehtag); 2788} 2789 2790static int 2791g_mirror_can_go(void) 2792{ 2793 struct g_mirror_softc *sc; 2794 struct g_geom *gp; 2795 struct g_provider *pp; 2796 int can_go; 2797 2798 DROP_GIANT(); 2799 can_go = 1; 2800 g_topology_lock(); 2801 LIST_FOREACH(gp, &g_mirror_class.geom, geom) { 2802 sc = gp->softc; 2803 if (sc == NULL) { 2804 can_go = 0; 2805 break; 2806 } 2807 pp = sc->sc_provider; 2808 if (pp == NULL || pp->error != 0) { 2809 can_go = 0; 2810 break; 2811 } 2812 } 2813 g_topology_unlock(); 2814 PICKUP_GIANT(); 2815 return (can_go); 2816} 2817 2818static void 2819g_mirror_rootwait(void) 2820{ 2821 2822 /* 2823 * HACK: Wait for GEOM, because g_mirror_rootwait() can be called, 2824 * HACK: before we get providers for tasting. 2825 */ 2826 tsleep(&g_mirror_class, PRIBIO, "mroot", hz * 3); 2827 /* 2828 * Wait for mirrors in degraded state. 2829 */ 2830 for (;;) { 2831 if (g_mirror_can_go()) 2832 break; 2833 tsleep(&g_mirror_class, PRIBIO, "mroot", hz); 2834 } 2835} 2836 2837SYSINIT(g_mirror_root, SI_SUB_RAID, SI_ORDER_FIRST, g_mirror_rootwait, NULL) 2838 2839DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2840