g_mirror.c revision 155539
157434Smarkm/*- 257434Smarkm * Copyright (c) 2004-2005 Pawel Jakub Dawidek <pjd@FreeBSD.org> 3255460Sdes * All rights reserved. 4255460Sdes * 557434Smarkm * Redistribution and use in source and binary forms, with or without 6197679Sdes * modification, are permitted provided that the following conditions 7158519Sdes * are met: 857434Smarkm * 1. Redistributions of source code must retain the above copyright 9255460Sdes * notice, this list of conditions and the following disclaimer. 10255460Sdes * 2. Redistributions in binary form must reproduce the above copyright 11197679Sdes * notice, this list of conditions and the following disclaimer in the 12255460Sdes * documentation and/or other materials provided with the distribution. 13197679Sdes * 14255460Sdes * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15255460Sdes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16255386Sdes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 1774818Sru * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18255460Sdes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19255460Sdes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20255460Sdes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21255460Sdes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22255460Sdes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23255460Sdes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24255460Sdes * SUCH DAMAGE. 25255460Sdes */ 26255460Sdes 27255460Sdes#include <sys/cdefs.h> 2857434Smarkm__FBSDID("$FreeBSD: head/sys/geom/mirror/g_mirror.c 155539 2006-02-11 14:42:23Z pjd $"); 2957434Smarkm 3074818Sru#include <sys/param.h> 31158529Sdes#include <sys/systm.h> 32158529Sdes#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 <sys/sched.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; 54TUNABLE_INT("kern.geom.mirror.debug", &g_mirror_debug); 55SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RW, &g_mirror_debug, 0, 56 "Debug level"); 57static u_int g_mirror_timeout = 4; 58TUNABLE_INT("kern.geom.mirror.timeout", &g_mirror_timeout); 59SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RW, &g_mirror_timeout, 60 0, "Time to wait on all mirror components"); 61static u_int g_mirror_idletime = 5; 62TUNABLE_INT("kern.geom.mirror.idletime", &g_mirror_idletime); 63SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RW, 64 &g_mirror_idletime, 0, "Mark components as clean when idling"); 65static u_int g_mirror_reqs_per_sync = 5; 66SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, reqs_per_sync, CTLFLAG_RW, 67 &g_mirror_reqs_per_sync, 0, 68 "Number of regular I/O requests per synchronization request"); 69static u_int g_mirror_syncs_per_sec = 1000; 70SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, syncs_per_sec, CTLFLAG_RW, 71 &g_mirror_syncs_per_sec, 0, 72 "Number of synchronizations requests per second"); 73 74#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 75 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 76 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 77 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 78} while (0) 79 80static eventhandler_tag g_mirror_ehtag = NULL; 81 82static int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp, 83 struct g_geom *gp); 84static g_taste_t g_mirror_taste; 85static void g_mirror_init(struct g_class *mp); 86static void g_mirror_fini(struct g_class *mp); 87 88struct g_class g_mirror_class = { 89 .name = G_MIRROR_CLASS_NAME, 90 .version = G_VERSION, 91 .ctlreq = g_mirror_config, 92 .taste = g_mirror_taste, 93 .destroy_geom = g_mirror_destroy_geom, 94 .init = g_mirror_init, 95 .fini = g_mirror_fini 96}; 97 98 99static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 100static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 101static void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force); 102static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 103 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 104static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 105 106 107static const char * 108g_mirror_disk_state2str(int state) 109{ 110 111 switch (state) { 112 case G_MIRROR_DISK_STATE_NONE: 113 return ("NONE"); 114 case G_MIRROR_DISK_STATE_NEW: 115 return ("NEW"); 116 case G_MIRROR_DISK_STATE_ACTIVE: 117 return ("ACTIVE"); 118 case G_MIRROR_DISK_STATE_STALE: 119 return ("STALE"); 120 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 121 return ("SYNCHRONIZING"); 122 case G_MIRROR_DISK_STATE_DISCONNECTED: 123 return ("DISCONNECTED"); 124 case G_MIRROR_DISK_STATE_DESTROY: 125 return ("DESTROY"); 126 default: 127 return ("INVALID"); 128 } 129} 130 131static const char * 132g_mirror_device_state2str(int state) 133{ 134 135 switch (state) { 136 case G_MIRROR_DEVICE_STATE_STARTING: 137 return ("STARTING"); 138 case G_MIRROR_DEVICE_STATE_RUNNING: 139 return ("RUNNING"); 140 default: 141 return ("INVALID"); 142 } 143} 144 145static const char * 146g_mirror_get_diskname(struct g_mirror_disk *disk) 147{ 148 149 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 150 return ("[unknown]"); 151 return (disk->d_name); 152} 153 154/* 155 * --- Events handling functions --- 156 * Events in geom_mirror are used to maintain disks and device status 157 * from one thread to simplify locking. 158 */ 159static void 160g_mirror_event_free(struct g_mirror_event *ep) 161{ 162 163 free(ep, M_MIRROR); 164} 165 166int 167g_mirror_event_send(void *arg, int state, int flags) 168{ 169 struct g_mirror_softc *sc; 170 struct g_mirror_disk *disk; 171 struct g_mirror_event *ep; 172 int error; 173 174 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 175 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 176 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 177 disk = NULL; 178 sc = arg; 179 } else { 180 disk = arg; 181 sc = disk->d_softc; 182 } 183 ep->e_disk = disk; 184 ep->e_state = state; 185 ep->e_flags = flags; 186 ep->e_error = 0; 187 mtx_lock(&sc->sc_events_mtx); 188 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 189 mtx_unlock(&sc->sc_events_mtx); 190 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 191 mtx_lock(&sc->sc_queue_mtx); 192 wakeup(sc); 193 mtx_unlock(&sc->sc_queue_mtx); 194 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 195 return (0); 196 g_topology_assert(); 197 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 198 g_topology_unlock(); 199 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 200 mtx_lock(&sc->sc_events_mtx); 201 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 202 hz * 5); 203 } 204 /* Don't even try to use 'sc' here, because it could be already dead. */ 205 g_topology_lock(); 206 error = ep->e_error; 207 g_mirror_event_free(ep); 208 return (error); 209} 210 211static struct g_mirror_event * 212g_mirror_event_get(struct g_mirror_softc *sc) 213{ 214 struct g_mirror_event *ep; 215 216 mtx_lock(&sc->sc_events_mtx); 217 ep = TAILQ_FIRST(&sc->sc_events); 218 mtx_unlock(&sc->sc_events_mtx); 219 return (ep); 220} 221 222static void 223g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 224{ 225 226 mtx_lock(&sc->sc_events_mtx); 227 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 228 mtx_unlock(&sc->sc_events_mtx); 229} 230 231static void 232g_mirror_event_cancel(struct g_mirror_disk *disk) 233{ 234 struct g_mirror_softc *sc; 235 struct g_mirror_event *ep, *tmpep; 236 237 g_topology_assert(); 238 239 sc = disk->d_softc; 240 mtx_lock(&sc->sc_events_mtx); 241 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 242 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 243 continue; 244 if (ep->e_disk != disk) 245 continue; 246 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 247 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 248 g_mirror_event_free(ep); 249 else { 250 ep->e_error = ECANCELED; 251 wakeup(ep); 252 } 253 } 254 mtx_unlock(&sc->sc_events_mtx); 255} 256 257/* 258 * Return the number of disks in given state. 259 * If state is equal to -1, count all connected disks. 260 */ 261u_int 262g_mirror_ndisks(struct g_mirror_softc *sc, int state) 263{ 264 struct g_mirror_disk *disk; 265 u_int n = 0; 266 267 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 268 if (state == -1 || disk->d_state == state) 269 n++; 270 } 271 return (n); 272} 273 274/* 275 * Find a disk in mirror by its disk ID. 276 */ 277static struct g_mirror_disk * 278g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 279{ 280 struct g_mirror_disk *disk; 281 282 g_topology_assert(); 283 284 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 285 if (disk->d_id == id) 286 return (disk); 287 } 288 return (NULL); 289} 290 291static u_int 292g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 293{ 294 struct bio *bp; 295 u_int nreqs = 0; 296 297 mtx_lock(&sc->sc_queue_mtx); 298 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 299 if (bp->bio_from == cp) 300 nreqs++; 301 } 302 mtx_unlock(&sc->sc_queue_mtx); 303 return (nreqs); 304} 305 306static int 307g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 308{ 309 310 if (cp->index > 0) { 311 G_MIRROR_DEBUG(2, 312 "I/O requests for %s exist, can't destroy it now.", 313 cp->provider->name); 314 return (1); 315 } 316 if (g_mirror_nrequests(sc, cp) > 0) { 317 G_MIRROR_DEBUG(2, 318 "I/O requests for %s in queue, can't destroy it now.", 319 cp->provider->name); 320 return (1); 321 } 322 return (0); 323} 324 325static void 326g_mirror_destroy_consumer(void *arg, int flags __unused) 327{ 328 struct g_consumer *cp; 329 330 cp = arg; 331 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 332 g_detach(cp); 333 g_destroy_consumer(cp); 334} 335 336static void 337g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 338{ 339 struct g_provider *pp; 340 int retaste_wait; 341 342 g_topology_assert(); 343 344 cp->private = NULL; 345 if (g_mirror_is_busy(sc, cp)) 346 return; 347 pp = cp->provider; 348 retaste_wait = 0; 349 if (cp->acw == 1) { 350 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 351 retaste_wait = 1; 352 } 353 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 354 -cp->acw, -cp->ace, 0); 355 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 356 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 357 if (retaste_wait) { 358 /* 359 * After retaste event was send (inside g_access()), we can send 360 * event to detach and destroy consumer. 361 * A class, which has consumer to the given provider connected 362 * will not receive retaste event for the provider. 363 * This is the way how I ignore retaste events when I close 364 * consumers opened for write: I detach and destroy consumer 365 * after retaste event is sent. 366 */ 367 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 368 return; 369 } 370 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 371 g_detach(cp); 372 g_destroy_consumer(cp); 373} 374 375static int 376g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 377{ 378 struct g_consumer *cp; 379 int error; 380 381 g_topology_assert(); 382 KASSERT(disk->d_consumer == NULL, 383 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 384 385 cp = g_new_consumer(disk->d_softc->sc_geom); 386 error = g_attach(cp, pp); 387 if (error != 0) { 388 g_destroy_consumer(cp); 389 return (error); 390 } 391 error = g_access(cp, 1, 1, 1); 392 if (error != 0) { 393 g_detach(cp); 394 g_destroy_consumer(cp); 395 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 396 pp->name, error); 397 return (error); 398 } 399 disk->d_consumer = cp; 400 disk->d_consumer->private = disk; 401 disk->d_consumer->index = 0; 402 403 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 404 return (0); 405} 406 407static void 408g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 409{ 410 411 g_topology_assert(); 412 413 if (cp == NULL) 414 return; 415 if (cp->provider != NULL) 416 g_mirror_kill_consumer(sc, cp); 417 else 418 g_destroy_consumer(cp); 419} 420 421/* 422 * Initialize disk. This means allocate memory, create consumer, attach it 423 * to the provider and open access (r1w1e1) to it. 424 */ 425static struct g_mirror_disk * 426g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 427 struct g_mirror_metadata *md, int *errorp) 428{ 429 struct g_mirror_disk *disk; 430 int error; 431 432 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 433 if (disk == NULL) { 434 error = ENOMEM; 435 goto fail; 436 } 437 disk->d_softc = sc; 438 error = g_mirror_connect_disk(disk, pp); 439 if (error != 0) 440 goto fail; 441 disk->d_id = md->md_did; 442 disk->d_state = G_MIRROR_DISK_STATE_NONE; 443 disk->d_priority = md->md_priority; 444 disk->d_delay.sec = 0; 445 disk->d_delay.frac = 0; 446 binuptime(&disk->d_last_used); 447 disk->d_flags = md->md_dflags; 448 if (md->md_provider[0] != '\0') 449 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 450 disk->d_sync.ds_consumer = NULL; 451 disk->d_sync.ds_offset = md->md_sync_offset; 452 disk->d_sync.ds_offset_done = md->md_sync_offset; 453 disk->d_sync.ds_resync = -1; 454 disk->d_genid = md->md_genid; 455 disk->d_sync.ds_syncid = md->md_syncid; 456 if (errorp != NULL) 457 *errorp = 0; 458 return (disk); 459fail: 460 if (errorp != NULL) 461 *errorp = error; 462 if (disk != NULL) 463 free(disk, M_MIRROR); 464 return (NULL); 465} 466 467static void 468g_mirror_destroy_disk(struct g_mirror_disk *disk) 469{ 470 struct g_mirror_softc *sc; 471 472 g_topology_assert(); 473 474 LIST_REMOVE(disk, d_next); 475 g_mirror_event_cancel(disk); 476 sc = disk->d_softc; 477 if (sc->sc_hint == disk) 478 sc->sc_hint = NULL; 479 switch (disk->d_state) { 480 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 481 g_mirror_sync_stop(disk, 1); 482 /* FALLTHROUGH */ 483 case G_MIRROR_DISK_STATE_NEW: 484 case G_MIRROR_DISK_STATE_STALE: 485 case G_MIRROR_DISK_STATE_ACTIVE: 486 g_mirror_disconnect_consumer(sc, disk->d_consumer); 487 free(disk, M_MIRROR); 488 break; 489 default: 490 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 491 g_mirror_get_diskname(disk), 492 g_mirror_disk_state2str(disk->d_state))); 493 } 494} 495 496static void 497g_mirror_destroy_device(struct g_mirror_softc *sc) 498{ 499 struct g_mirror_disk *disk; 500 struct g_mirror_event *ep; 501 struct g_geom *gp; 502 struct g_consumer *cp, *tmpcp; 503 504 g_topology_assert(); 505 506 gp = sc->sc_geom; 507 if (sc->sc_provider != NULL) 508 g_mirror_destroy_provider(sc); 509 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 510 disk = LIST_FIRST(&sc->sc_disks)) { 511 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 512 g_mirror_update_metadata(disk); 513 g_mirror_destroy_disk(disk); 514 } 515 while ((ep = g_mirror_event_get(sc)) != NULL) { 516 g_mirror_event_remove(sc, ep); 517 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 518 g_mirror_event_free(ep); 519 else { 520 ep->e_error = ECANCELED; 521 ep->e_flags |= G_MIRROR_EVENT_DONE; 522 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 523 mtx_lock(&sc->sc_events_mtx); 524 wakeup(ep); 525 mtx_unlock(&sc->sc_events_mtx); 526 } 527 } 528 callout_drain(&sc->sc_callout); 529 gp->softc = NULL; 530 531 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 532 g_mirror_disconnect_consumer(sc, cp); 533 } 534 sc->sc_sync.ds_geom->softc = NULL; 535 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 536 mtx_destroy(&sc->sc_queue_mtx); 537 mtx_destroy(&sc->sc_events_mtx); 538 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 539 g_wither_geom(gp, ENXIO); 540} 541 542static void 543g_mirror_orphan(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_id |= G_MIRROR_BUMP_SYNCID; 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_id |= G_MIRROR_BUMP_GENID; 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_genid = sc->sc_genid; 666 md->md_mediasize = sc->sc_mediasize; 667 md->md_sectorsize = sc->sc_sectorsize; 668 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 669 bzero(md->md_provider, sizeof(md->md_provider)); 670 if (disk == NULL) { 671 md->md_did = arc4random(); 672 md->md_priority = 0; 673 md->md_syncid = 0; 674 md->md_dflags = 0; 675 md->md_sync_offset = 0; 676 md->md_provsize = 0; 677 } else { 678 md->md_did = disk->d_id; 679 md->md_priority = disk->d_priority; 680 md->md_syncid = disk->d_sync.ds_syncid; 681 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 682 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 683 md->md_sync_offset = disk->d_sync.ds_offset_done; 684 else 685 md->md_sync_offset = 0; 686 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 687 strlcpy(md->md_provider, 688 disk->d_consumer->provider->name, 689 sizeof(md->md_provider)); 690 } 691 md->md_provsize = disk->d_consumer->provider->mediasize; 692 } 693} 694 695void 696g_mirror_update_metadata(struct g_mirror_disk *disk) 697{ 698 struct g_mirror_metadata md; 699 int error; 700 701 g_topology_assert(); 702 g_mirror_fill_metadata(disk->d_softc, disk, &md); 703 error = g_mirror_write_metadata(disk, &md); 704 if (error == 0) { 705 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 706 g_mirror_get_diskname(disk)); 707 } else { 708 G_MIRROR_DEBUG(0, 709 "Cannot update metadata on disk %s (error=%d).", 710 g_mirror_get_diskname(disk), error); 711 } 712} 713 714static void 715g_mirror_bump_syncid(struct g_mirror_softc *sc) 716{ 717 struct g_mirror_disk *disk; 718 719 g_topology_assert(); 720 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 721 ("%s called with no active disks (device=%s).", __func__, 722 sc->sc_name)); 723 724 sc->sc_syncid++; 725 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 726 sc->sc_syncid); 727 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 728 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 729 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 730 disk->d_sync.ds_syncid = sc->sc_syncid; 731 g_mirror_update_metadata(disk); 732 } 733 } 734} 735 736static void 737g_mirror_bump_genid(struct g_mirror_softc *sc) 738{ 739 struct g_mirror_disk *disk; 740 741 g_topology_assert(); 742 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 743 ("%s called with no active disks (device=%s).", __func__, 744 sc->sc_name)); 745 746 sc->sc_genid++; 747 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 748 sc->sc_genid); 749 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 750 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 751 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 752 disk->d_genid = sc->sc_genid; 753 g_mirror_update_metadata(disk); 754 } 755 } 756} 757 758static int 759g_mirror_idle(struct g_mirror_softc *sc, int from_access) 760{ 761 struct g_mirror_disk *disk; 762 int timeout; 763 764 if (sc->sc_provider == NULL) 765 return (0); 766 if (sc->sc_idle) 767 return (0); 768 if (sc->sc_writes > 0) 769 return (0); 770 if (!from_access && sc->sc_provider->acw > 0) { 771 timeout = g_mirror_idletime - (time_second - sc->sc_last_write); 772 if (timeout > 0) 773 return (timeout); 774 } 775 sc->sc_idle = 1; 776 if (!from_access) 777 g_topology_lock(); 778 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 779 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 780 continue; 781 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 782 g_mirror_get_diskname(disk), sc->sc_name); 783 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 784 g_mirror_update_metadata(disk); 785 } 786 if (!from_access) 787 g_topology_unlock(); 788 return (0); 789} 790 791static void 792g_mirror_unidle(struct g_mirror_softc *sc) 793{ 794 struct g_mirror_disk *disk; 795 796 sc->sc_idle = 0; 797 sc->sc_last_write = time_second; 798 g_topology_lock(); 799 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 800 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 801 continue; 802 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 803 g_mirror_get_diskname(disk), sc->sc_name); 804 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 805 g_mirror_update_metadata(disk); 806 } 807 g_topology_unlock(); 808} 809 810static __inline int 811bintime_cmp(struct bintime *bt1, struct bintime *bt2) 812{ 813 814 if (bt1->sec < bt2->sec) 815 return (-1); 816 else if (bt1->sec > bt2->sec) 817 return (1); 818 if (bt1->frac < bt2->frac) 819 return (-1); 820 else if (bt1->frac > bt2->frac) 821 return (1); 822 return (0); 823} 824 825static void 826g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 827{ 828 829 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 830 return; 831 binuptime(&disk->d_delay); 832 bintime_sub(&disk->d_delay, &bp->bio_t0); 833} 834 835static void 836g_mirror_done(struct bio *bp) 837{ 838 struct g_mirror_softc *sc; 839 840 sc = bp->bio_from->geom->softc; 841 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 842 mtx_lock(&sc->sc_queue_mtx); 843 bioq_disksort(&sc->sc_queue, bp); 844 wakeup(sc); 845 mtx_unlock(&sc->sc_queue_mtx); 846} 847 848static void 849g_mirror_regular_request(struct bio *bp) 850{ 851 struct g_mirror_softc *sc; 852 struct g_mirror_disk *disk; 853 struct bio *pbp; 854 855 g_topology_assert_not(); 856 857 pbp = bp->bio_parent; 858 sc = pbp->bio_to->geom->softc; 859 bp->bio_from->index--; 860 if (bp->bio_cmd == BIO_WRITE) 861 sc->sc_writes--; 862 disk = bp->bio_from->private; 863 if (disk == NULL) { 864 g_topology_lock(); 865 g_mirror_kill_consumer(sc, bp->bio_from); 866 g_topology_unlock(); 867 } else { 868 g_mirror_update_delay(disk, bp); 869 } 870 871 pbp->bio_inbed++; 872 KASSERT(pbp->bio_inbed <= pbp->bio_children, 873 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 874 pbp->bio_children)); 875 if (bp->bio_error == 0 && pbp->bio_error == 0) { 876 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 877 g_destroy_bio(bp); 878 if (pbp->bio_children == pbp->bio_inbed) { 879 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 880 pbp->bio_completed = pbp->bio_length; 881 g_io_deliver(pbp, pbp->bio_error); 882 } 883 return; 884 } else if (bp->bio_error != 0) { 885 if (pbp->bio_error == 0) 886 pbp->bio_error = bp->bio_error; 887 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 888 bp->bio_error); 889 if (disk != NULL) { 890 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 891 g_mirror_event_send(disk, 892 G_MIRROR_DISK_STATE_DISCONNECTED, 893 G_MIRROR_EVENT_DONTWAIT); 894 } 895 switch (pbp->bio_cmd) { 896 case BIO_DELETE: 897 case BIO_WRITE: 898 pbp->bio_inbed--; 899 pbp->bio_children--; 900 break; 901 } 902 } 903 g_destroy_bio(bp); 904 905 switch (pbp->bio_cmd) { 906 case BIO_READ: 907 if (pbp->bio_children == pbp->bio_inbed) { 908 pbp->bio_error = 0; 909 mtx_lock(&sc->sc_queue_mtx); 910 bioq_disksort(&sc->sc_queue, pbp); 911 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 912 wakeup(sc); 913 mtx_unlock(&sc->sc_queue_mtx); 914 } 915 break; 916 case BIO_DELETE: 917 case BIO_WRITE: 918 if (pbp->bio_children == 0) { 919 /* 920 * All requests failed. 921 */ 922 } else if (pbp->bio_inbed < pbp->bio_children) { 923 /* Do nothing. */ 924 break; 925 } else if (pbp->bio_children == pbp->bio_inbed) { 926 /* Some requests succeeded. */ 927 pbp->bio_error = 0; 928 pbp->bio_completed = pbp->bio_length; 929 } 930 g_io_deliver(pbp, pbp->bio_error); 931 break; 932 default: 933 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 934 break; 935 } 936} 937 938static void 939g_mirror_sync_done(struct bio *bp) 940{ 941 struct g_mirror_softc *sc; 942 943 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 944 sc = bp->bio_from->geom->softc; 945 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 946 mtx_lock(&sc->sc_queue_mtx); 947 bioq_disksort(&sc->sc_queue, bp); 948 wakeup(sc); 949 mtx_unlock(&sc->sc_queue_mtx); 950} 951 952static void 953g_mirror_start(struct bio *bp) 954{ 955 struct g_mirror_softc *sc; 956 957 sc = bp->bio_to->geom->softc; 958 /* 959 * If sc == NULL or there are no valid disks, provider's error 960 * should be set and g_mirror_start() should not be called at all. 961 */ 962 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 963 ("Provider's error should be set (error=%d)(mirror=%s).", 964 bp->bio_to->error, bp->bio_to->name)); 965 G_MIRROR_LOGREQ(3, bp, "Request received."); 966 967 switch (bp->bio_cmd) { 968 case BIO_READ: 969 case BIO_WRITE: 970 case BIO_DELETE: 971 break; 972 case BIO_GETATTR: 973 default: 974 g_io_deliver(bp, EOPNOTSUPP); 975 return; 976 } 977 mtx_lock(&sc->sc_queue_mtx); 978 bioq_disksort(&sc->sc_queue, bp); 979 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 980 wakeup(sc); 981 mtx_unlock(&sc->sc_queue_mtx); 982} 983 984/* 985 * Send one synchronization request. 986 */ 987static void 988g_mirror_sync_one(struct g_mirror_disk *disk) 989{ 990 struct g_mirror_softc *sc; 991 struct bio *bp; 992 993 sc = disk->d_softc; 994 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 995 ("Disk %s is not marked for synchronization.", 996 g_mirror_get_diskname(disk))); 997 998 bp = g_new_bio(); 999 if (bp == NULL) 1000 return; 1001 bp->bio_parent = NULL; 1002 bp->bio_cmd = BIO_READ; 1003 bp->bio_offset = disk->d_sync.ds_offset; 1004 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1005 bp->bio_cflags = 0; 1006 bp->bio_done = g_mirror_sync_done; 1007 bp->bio_data = disk->d_sync.ds_data; 1008 if (bp->bio_data == NULL) { 1009 g_destroy_bio(bp); 1010 return; 1011 } 1012 disk->d_sync.ds_offset += bp->bio_length; 1013 bp->bio_to = sc->sc_provider; 1014 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1015 disk->d_sync.ds_consumer->index++; 1016 g_io_request(bp, disk->d_sync.ds_consumer); 1017} 1018 1019static void 1020g_mirror_sync_request(struct bio *bp) 1021{ 1022 struct g_mirror_softc *sc; 1023 struct g_mirror_disk *disk; 1024 1025 bp->bio_from->index--; 1026 sc = bp->bio_from->geom->softc; 1027 disk = bp->bio_from->private; 1028 if (disk == NULL) { 1029 g_topology_lock(); 1030 g_mirror_kill_consumer(sc, bp->bio_from); 1031 g_topology_unlock(); 1032 g_destroy_bio(bp); 1033 return; 1034 } 1035 1036 /* 1037 * Synchronization request. 1038 */ 1039 switch (bp->bio_cmd) { 1040 case BIO_READ: 1041 { 1042 struct g_consumer *cp; 1043 1044 if (bp->bio_error != 0) { 1045 G_MIRROR_LOGREQ(0, bp, 1046 "Synchronization request failed (error=%d).", 1047 bp->bio_error); 1048 g_destroy_bio(bp); 1049 return; 1050 } 1051 G_MIRROR_LOGREQ(3, bp, 1052 "Synchronization request half-finished."); 1053 bp->bio_cmd = BIO_WRITE; 1054 bp->bio_cflags = 0; 1055 cp = disk->d_consumer; 1056 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1057 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1058 cp->acr, cp->acw, cp->ace)); 1059 cp->index++; 1060 g_io_request(bp, cp); 1061 return; 1062 } 1063 case BIO_WRITE: 1064 { 1065 struct g_mirror_disk_sync *sync; 1066 1067 if (bp->bio_error != 0) { 1068 G_MIRROR_LOGREQ(0, bp, 1069 "Synchronization request failed (error=%d).", 1070 bp->bio_error); 1071 g_destroy_bio(bp); 1072 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1073 g_mirror_event_send(disk, 1074 G_MIRROR_DISK_STATE_DISCONNECTED, 1075 G_MIRROR_EVENT_DONTWAIT); 1076 return; 1077 } 1078 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1079 sync = &disk->d_sync; 1080 sync->ds_offset_done = bp->bio_offset + bp->bio_length; 1081 g_destroy_bio(bp); 1082 if (sync->ds_resync != -1) 1083 break; 1084 if (sync->ds_offset_done == sc->sc_provider->mediasize) { 1085 /* 1086 * Disk up-to-date, activate it. 1087 */ 1088 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1089 G_MIRROR_EVENT_DONTWAIT); 1090 return; 1091 } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) { 1092 /* 1093 * Update offset_done on every 100 blocks. 1094 * XXX: This should be configurable. 1095 */ 1096 g_topology_lock(); 1097 g_mirror_update_metadata(disk); 1098 g_topology_unlock(); 1099 } 1100 return; 1101 } 1102 default: 1103 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1104 bp->bio_cmd, sc->sc_name)); 1105 break; 1106 } 1107} 1108 1109static void 1110g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1111{ 1112 struct g_mirror_disk *disk; 1113 struct g_consumer *cp; 1114 struct bio *cbp; 1115 1116 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1117 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1118 break; 1119 } 1120 if (disk == NULL) { 1121 if (bp->bio_error == 0) 1122 bp->bio_error = ENXIO; 1123 g_io_deliver(bp, bp->bio_error); 1124 return; 1125 } 1126 cbp = g_clone_bio(bp); 1127 if (cbp == NULL) { 1128 if (bp->bio_error == 0) 1129 bp->bio_error = ENOMEM; 1130 g_io_deliver(bp, bp->bio_error); 1131 return; 1132 } 1133 /* 1134 * Fill in the component buf structure. 1135 */ 1136 cp = disk->d_consumer; 1137 cbp->bio_done = g_mirror_done; 1138 cbp->bio_to = cp->provider; 1139 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1140 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1141 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1142 cp->acw, cp->ace)); 1143 cp->index++; 1144 g_io_request(cbp, cp); 1145} 1146 1147static void 1148g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1149{ 1150 struct g_mirror_disk *disk; 1151 struct g_consumer *cp; 1152 struct bio *cbp; 1153 1154 disk = g_mirror_get_disk(sc); 1155 if (disk == NULL) { 1156 if (bp->bio_error == 0) 1157 bp->bio_error = ENXIO; 1158 g_io_deliver(bp, bp->bio_error); 1159 return; 1160 } 1161 cbp = g_clone_bio(bp); 1162 if (cbp == NULL) { 1163 if (bp->bio_error == 0) 1164 bp->bio_error = ENOMEM; 1165 g_io_deliver(bp, bp->bio_error); 1166 return; 1167 } 1168 /* 1169 * Fill in the component buf structure. 1170 */ 1171 cp = disk->d_consumer; 1172 cbp->bio_done = g_mirror_done; 1173 cbp->bio_to = cp->provider; 1174 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1175 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1176 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1177 cp->acw, cp->ace)); 1178 cp->index++; 1179 g_io_request(cbp, cp); 1180} 1181 1182static void 1183g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1184{ 1185 struct g_mirror_disk *disk, *dp; 1186 struct g_consumer *cp; 1187 struct bio *cbp; 1188 struct bintime curtime; 1189 1190 binuptime(&curtime); 1191 /* 1192 * Find a disk which the smallest load. 1193 */ 1194 disk = NULL; 1195 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1196 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1197 continue; 1198 /* If disk wasn't used for more than 2 sec, use it. */ 1199 if (curtime.sec - dp->d_last_used.sec >= 2) { 1200 disk = dp; 1201 break; 1202 } 1203 if (disk == NULL || 1204 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1205 disk = dp; 1206 } 1207 } 1208 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1209 cbp = g_clone_bio(bp); 1210 if (cbp == NULL) { 1211 if (bp->bio_error == 0) 1212 bp->bio_error = ENOMEM; 1213 g_io_deliver(bp, bp->bio_error); 1214 return; 1215 } 1216 /* 1217 * Fill in the component buf structure. 1218 */ 1219 cp = disk->d_consumer; 1220 cbp->bio_done = g_mirror_done; 1221 cbp->bio_to = cp->provider; 1222 binuptime(&disk->d_last_used); 1223 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1224 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1225 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1226 cp->acw, cp->ace)); 1227 cp->index++; 1228 g_io_request(cbp, cp); 1229} 1230 1231static void 1232g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1233{ 1234 struct bio_queue_head queue; 1235 struct g_mirror_disk *disk; 1236 struct g_consumer *cp; 1237 struct bio *cbp; 1238 off_t left, mod, offset, slice; 1239 u_char *data; 1240 u_int ndisks; 1241 1242 if (bp->bio_length <= sc->sc_slice) { 1243 g_mirror_request_round_robin(sc, bp); 1244 return; 1245 } 1246 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1247 slice = bp->bio_length / ndisks; 1248 mod = slice % sc->sc_provider->sectorsize; 1249 if (mod != 0) 1250 slice += sc->sc_provider->sectorsize - mod; 1251 /* 1252 * Allocate all bios before sending any request, so we can 1253 * return ENOMEM in nice and clean way. 1254 */ 1255 left = bp->bio_length; 1256 offset = bp->bio_offset; 1257 data = bp->bio_data; 1258 bioq_init(&queue); 1259 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1260 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1261 continue; 1262 cbp = g_clone_bio(bp); 1263 if (cbp == NULL) { 1264 for (cbp = bioq_first(&queue); cbp != NULL; 1265 cbp = bioq_first(&queue)) { 1266 bioq_remove(&queue, cbp); 1267 g_destroy_bio(cbp); 1268 } 1269 if (bp->bio_error == 0) 1270 bp->bio_error = ENOMEM; 1271 g_io_deliver(bp, bp->bio_error); 1272 return; 1273 } 1274 bioq_insert_tail(&queue, cbp); 1275 cbp->bio_done = g_mirror_done; 1276 cbp->bio_caller1 = disk; 1277 cbp->bio_to = disk->d_consumer->provider; 1278 cbp->bio_offset = offset; 1279 cbp->bio_data = data; 1280 cbp->bio_length = MIN(left, slice); 1281 left -= cbp->bio_length; 1282 if (left == 0) 1283 break; 1284 offset += cbp->bio_length; 1285 data += cbp->bio_length; 1286 } 1287 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1288 bioq_remove(&queue, cbp); 1289 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1290 disk = cbp->bio_caller1; 1291 cbp->bio_caller1 = NULL; 1292 cp = disk->d_consumer; 1293 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1294 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1295 cp->acr, cp->acw, cp->ace)); 1296 disk->d_consumer->index++; 1297 g_io_request(cbp, disk->d_consumer); 1298 } 1299} 1300 1301static void 1302g_mirror_register_request(struct bio *bp) 1303{ 1304 struct g_mirror_softc *sc; 1305 1306 sc = bp->bio_to->geom->softc; 1307 switch (bp->bio_cmd) { 1308 case BIO_READ: 1309 switch (sc->sc_balance) { 1310 case G_MIRROR_BALANCE_LOAD: 1311 g_mirror_request_load(sc, bp); 1312 break; 1313 case G_MIRROR_BALANCE_PREFER: 1314 g_mirror_request_prefer(sc, bp); 1315 break; 1316 case G_MIRROR_BALANCE_ROUND_ROBIN: 1317 g_mirror_request_round_robin(sc, bp); 1318 break; 1319 case G_MIRROR_BALANCE_SPLIT: 1320 g_mirror_request_split(sc, bp); 1321 break; 1322 } 1323 return; 1324 case BIO_WRITE: 1325 case BIO_DELETE: 1326 { 1327 struct g_mirror_disk *disk; 1328 struct g_mirror_disk_sync *sync; 1329 struct bio_queue_head queue; 1330 struct g_consumer *cp; 1331 struct bio *cbp; 1332 1333 if (sc->sc_idle) 1334 g_mirror_unidle(sc); 1335 else 1336 sc->sc_last_write = time_second; 1337 1338 /* 1339 * Allocate all bios before sending any request, so we can 1340 * return ENOMEM in nice and clean way. 1341 */ 1342 bioq_init(&queue); 1343 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1344 sync = &disk->d_sync; 1345 switch (disk->d_state) { 1346 case G_MIRROR_DISK_STATE_ACTIVE: 1347 break; 1348 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1349 if (bp->bio_offset >= sync->ds_offset) 1350 continue; 1351 else if (bp->bio_offset + bp->bio_length > 1352 sync->ds_offset_done && 1353 (bp->bio_offset < sync->ds_resync || 1354 sync->ds_resync == -1)) { 1355 sync->ds_resync = bp->bio_offset - 1356 (bp->bio_offset % MAXPHYS); 1357 } 1358 break; 1359 default: 1360 continue; 1361 } 1362 cbp = g_clone_bio(bp); 1363 if (cbp == NULL) { 1364 for (cbp = bioq_first(&queue); cbp != NULL; 1365 cbp = bioq_first(&queue)) { 1366 bioq_remove(&queue, cbp); 1367 g_destroy_bio(cbp); 1368 } 1369 if (bp->bio_error == 0) 1370 bp->bio_error = ENOMEM; 1371 g_io_deliver(bp, bp->bio_error); 1372 return; 1373 } 1374 bioq_insert_tail(&queue, cbp); 1375 cbp->bio_done = g_mirror_done; 1376 cp = disk->d_consumer; 1377 cbp->bio_caller1 = cp; 1378 cbp->bio_to = cp->provider; 1379 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1380 ("Consumer %s not opened (r%dw%de%d).", 1381 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1382 } 1383 for (cbp = bioq_first(&queue); cbp != NULL; 1384 cbp = bioq_first(&queue)) { 1385 bioq_remove(&queue, cbp); 1386 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1387 cp = cbp->bio_caller1; 1388 cbp->bio_caller1 = NULL; 1389 cp->index++; 1390 sc->sc_writes++; 1391 g_io_request(cbp, cp); 1392 } 1393 /* 1394 * Bump syncid on first write. 1395 */ 1396 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1397 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1398 g_topology_lock(); 1399 g_mirror_bump_syncid(sc); 1400 g_topology_unlock(); 1401 } 1402 return; 1403 } 1404 default: 1405 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1406 bp->bio_cmd, sc->sc_name)); 1407 break; 1408 } 1409} 1410 1411static int 1412g_mirror_can_destroy(struct g_mirror_softc *sc) 1413{ 1414 struct g_geom *gp; 1415 struct g_consumer *cp; 1416 1417 g_topology_assert(); 1418 gp = sc->sc_geom; 1419 LIST_FOREACH(cp, &gp->consumer, consumer) { 1420 if (g_mirror_is_busy(sc, cp)) 1421 return (0); 1422 } 1423 gp = sc->sc_sync.ds_geom; 1424 LIST_FOREACH(cp, &gp->consumer, consumer) { 1425 if (g_mirror_is_busy(sc, cp)) 1426 return (0); 1427 } 1428 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1429 sc->sc_name); 1430 return (1); 1431} 1432 1433static int 1434g_mirror_try_destroy(struct g_mirror_softc *sc) 1435{ 1436 1437 if (sc->sc_rootmount != NULL) { 1438 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1439 sc->sc_rootmount); 1440 root_mount_rel(sc->sc_rootmount); 1441 sc->sc_rootmount = NULL; 1442 } 1443 g_topology_lock(); 1444 if (!g_mirror_can_destroy(sc)) { 1445 g_topology_unlock(); 1446 return (0); 1447 } 1448 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1449 g_topology_unlock(); 1450 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1451 &sc->sc_worker); 1452 wakeup(&sc->sc_worker); 1453 sc->sc_worker = NULL; 1454 } else { 1455 g_mirror_destroy_device(sc); 1456 g_topology_unlock(); 1457 free(sc, M_MIRROR); 1458 } 1459 return (1); 1460} 1461 1462/* 1463 * Worker thread. 1464 */ 1465static void 1466g_mirror_worker(void *arg) 1467{ 1468 struct g_mirror_softc *sc; 1469 struct g_mirror_disk *disk; 1470 struct g_mirror_disk_sync *sync; 1471 struct g_mirror_event *ep; 1472 struct bio *bp; 1473 u_int nreqs; 1474 int timeout; 1475 1476 sc = arg; 1477 mtx_lock_spin(&sched_lock); 1478 sched_prio(curthread, PRIBIO); 1479 mtx_unlock_spin(&sched_lock); 1480 1481 nreqs = 0; 1482 for (;;) { 1483 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1484 /* 1485 * First take a look at events. 1486 * This is important to handle events before any I/O requests. 1487 */ 1488 ep = g_mirror_event_get(sc); 1489 if (ep != NULL && g_topology_try_lock()) { 1490 g_mirror_event_remove(sc, ep); 1491 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1492 /* Update only device status. */ 1493 G_MIRROR_DEBUG(3, 1494 "Running event for device %s.", 1495 sc->sc_name); 1496 ep->e_error = 0; 1497 g_mirror_update_device(sc, 1); 1498 } else { 1499 /* Update disk status. */ 1500 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1501 g_mirror_get_diskname(ep->e_disk)); 1502 ep->e_error = g_mirror_update_disk(ep->e_disk, 1503 ep->e_state); 1504 if (ep->e_error == 0) 1505 g_mirror_update_device(sc, 0); 1506 } 1507 g_topology_unlock(); 1508 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1509 KASSERT(ep->e_error == 0, 1510 ("Error cannot be handled.")); 1511 g_mirror_event_free(ep); 1512 } else { 1513 ep->e_flags |= G_MIRROR_EVENT_DONE; 1514 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1515 ep); 1516 mtx_lock(&sc->sc_events_mtx); 1517 wakeup(ep); 1518 mtx_unlock(&sc->sc_events_mtx); 1519 } 1520 if ((sc->sc_flags & 1521 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1522 if (g_mirror_try_destroy(sc)) 1523 kthread_exit(0); 1524 } 1525 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1526 continue; 1527 } 1528 /* 1529 * Check if we can mark array as CLEAN and if we can't take 1530 * how much seconds should we wait. 1531 */ 1532 timeout = g_mirror_idle(sc, 0); 1533 /* 1534 * Now I/O requests. 1535 */ 1536 /* Get first request from the queue. */ 1537 mtx_lock(&sc->sc_queue_mtx); 1538 bp = bioq_first(&sc->sc_queue); 1539 if (bp == NULL) { 1540 if (ep != NULL) { 1541 /* 1542 * No I/O requests and topology lock was 1543 * already held? Try again. 1544 */ 1545 mtx_unlock(&sc->sc_queue_mtx); 1546 continue; 1547 } 1548 if ((sc->sc_flags & 1549 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1550 mtx_unlock(&sc->sc_queue_mtx); 1551 if (g_mirror_try_destroy(sc)) 1552 kthread_exit(0); 1553 mtx_lock(&sc->sc_queue_mtx); 1554 } 1555 } 1556 if (sc->sc_sync.ds_ndisks > 0 && 1557 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1558 mtx_unlock(&sc->sc_queue_mtx); 1559 /* 1560 * It is time for synchronization... 1561 */ 1562 nreqs = 0; 1563 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1564 if (disk->d_state != 1565 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1566 continue; 1567 } 1568 sync = &disk->d_sync; 1569 if (sync->ds_offset >= 1570 sc->sc_provider->mediasize) { 1571 continue; 1572 } 1573 if (sync->ds_offset > sync->ds_offset_done) 1574 continue; 1575 if (sync->ds_resync != -1) { 1576 sync->ds_offset = sync->ds_resync; 1577 sync->ds_offset_done = sync->ds_resync; 1578 sync->ds_resync = -1; 1579 } 1580 g_mirror_sync_one(disk); 1581 } 1582 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1583 goto sleep; 1584 } 1585 if (bp == NULL) { 1586 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1587 timeout * hz); 1588 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1589 continue; 1590 } 1591 nreqs++; 1592 bioq_remove(&sc->sc_queue, bp); 1593 mtx_unlock(&sc->sc_queue_mtx); 1594 1595 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1596 g_mirror_regular_request(bp); 1597 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1598 u_int timeout, sps; 1599 1600 g_mirror_sync_request(bp); 1601sleep: 1602 sps = g_mirror_syncs_per_sec; 1603 if (sps == 0) { 1604 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1605 continue; 1606 } 1607 if (ep != NULL) { 1608 /* 1609 * We have some pending events, don't sleep now. 1610 */ 1611 G_MIRROR_DEBUG(5, "%s: I'm here 7.", __func__); 1612 continue; 1613 } 1614 mtx_lock(&sc->sc_queue_mtx); 1615 if (bioq_first(&sc->sc_queue) != NULL) { 1616 mtx_unlock(&sc->sc_queue_mtx); 1617 G_MIRROR_DEBUG(5, "%s: I'm here 8.", __func__); 1618 continue; 1619 } 1620 timeout = hz / sps; 1621 if (timeout == 0) 1622 timeout = 1; 1623 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w3", 1624 timeout); 1625 } else { 1626 g_mirror_register_request(bp); 1627 } 1628 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1629 } 1630} 1631 1632static void 1633g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 1634{ 1635 1636 g_topology_assert(); 1637 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1638 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1639 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1640 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1641 } else if (sc->sc_idle && 1642 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1643 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1644 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1645 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1646 } 1647} 1648 1649static void 1650g_mirror_sync_start(struct g_mirror_disk *disk) 1651{ 1652 struct g_mirror_softc *sc; 1653 int error; 1654 1655 g_topology_assert(); 1656 1657 sc = disk->d_softc; 1658 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1659 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1660 sc->sc_state)); 1661 1662 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1663 g_mirror_get_diskname(disk)); 1664 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1665 KASSERT(disk->d_sync.ds_consumer == NULL, 1666 ("Sync consumer already exists (device=%s, disk=%s).", 1667 sc->sc_name, g_mirror_get_diskname(disk))); 1668 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1669 disk->d_sync.ds_consumer->private = disk; 1670 disk->d_sync.ds_consumer->index = 0; 1671 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1672 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1673 disk->d_softc->sc_name, error)); 1674 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1675 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1676 disk->d_softc->sc_name, error)); 1677 disk->d_sync.ds_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1678 sc->sc_sync.ds_ndisks++; 1679} 1680 1681/* 1682 * Stop synchronization process. 1683 * type: 0 - synchronization finished 1684 * 1 - synchronization stopped 1685 */ 1686static void 1687g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1688{ 1689 1690 g_topology_assert(); 1691 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1692 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1693 g_mirror_disk_state2str(disk->d_state))); 1694 if (disk->d_sync.ds_consumer == NULL) 1695 return; 1696 1697 if (type == 0) { 1698 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1699 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1700 } else /* if (type == 1) */ { 1701 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1702 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1703 } 1704 g_mirror_kill_consumer(disk->d_softc, disk->d_sync.ds_consumer); 1705 free(disk->d_sync.ds_data, M_MIRROR); 1706 disk->d_sync.ds_consumer = NULL; 1707 disk->d_softc->sc_sync.ds_ndisks--; 1708 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1709} 1710 1711static void 1712g_mirror_launch_provider(struct g_mirror_softc *sc) 1713{ 1714 struct g_mirror_disk *disk; 1715 struct g_provider *pp; 1716 1717 g_topology_assert(); 1718 1719 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1720 pp->mediasize = sc->sc_mediasize; 1721 pp->sectorsize = sc->sc_sectorsize; 1722 sc->sc_provider = pp; 1723 g_error_provider(pp, 0); 1724 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1725 pp->name); 1726 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1727 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1728 g_mirror_sync_start(disk); 1729 } 1730} 1731 1732static void 1733g_mirror_destroy_provider(struct g_mirror_softc *sc) 1734{ 1735 struct g_mirror_disk *disk; 1736 struct bio *bp; 1737 1738 g_topology_assert(); 1739 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1740 sc->sc_name)); 1741 1742 g_error_provider(sc->sc_provider, ENXIO); 1743 mtx_lock(&sc->sc_queue_mtx); 1744 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1745 bioq_remove(&sc->sc_queue, bp); 1746 g_io_deliver(bp, ENXIO); 1747 } 1748 mtx_unlock(&sc->sc_queue_mtx); 1749 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1750 sc->sc_provider->name); 1751 sc->sc_provider->flags |= G_PF_WITHER; 1752 g_orphan_provider(sc->sc_provider, ENXIO); 1753 sc->sc_provider = NULL; 1754 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1755 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1756 g_mirror_sync_stop(disk, 1); 1757 } 1758} 1759 1760static void 1761g_mirror_go(void *arg) 1762{ 1763 struct g_mirror_softc *sc; 1764 1765 sc = arg; 1766 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1767 g_mirror_event_send(sc, 0, 1768 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1769} 1770 1771static u_int 1772g_mirror_determine_state(struct g_mirror_disk *disk) 1773{ 1774 struct g_mirror_softc *sc; 1775 u_int state; 1776 1777 sc = disk->d_softc; 1778 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1779 if ((disk->d_flags & 1780 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1781 /* Disk does not need synchronization. */ 1782 state = G_MIRROR_DISK_STATE_ACTIVE; 1783 } else { 1784 if ((sc->sc_flags & 1785 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1786 (disk->d_flags & 1787 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1788 /* 1789 * We can start synchronization from 1790 * the stored offset. 1791 */ 1792 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1793 } else { 1794 state = G_MIRROR_DISK_STATE_STALE; 1795 } 1796 } 1797 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1798 /* 1799 * Reset all synchronization data for this disk, 1800 * because if it even was synchronized, it was 1801 * synchronized to disks with different syncid. 1802 */ 1803 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1804 disk->d_sync.ds_offset = 0; 1805 disk->d_sync.ds_offset_done = 0; 1806 disk->d_sync.ds_syncid = sc->sc_syncid; 1807 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1808 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1809 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1810 } else { 1811 state = G_MIRROR_DISK_STATE_STALE; 1812 } 1813 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1814 /* 1815 * Not good, NOT GOOD! 1816 * It means that mirror was started on stale disks 1817 * and more fresh disk just arrive. 1818 * If there were writes, mirror is fucked up, sorry. 1819 * I think the best choice here is don't touch 1820 * this disk and inform the user laudly. 1821 */ 1822 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1823 "disk (%s) arrives!! It will not be connected to the " 1824 "running device.", sc->sc_name, 1825 g_mirror_get_diskname(disk)); 1826 g_mirror_destroy_disk(disk); 1827 state = G_MIRROR_DISK_STATE_NONE; 1828 /* Return immediately, because disk was destroyed. */ 1829 return (state); 1830 } 1831 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1832 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1833 return (state); 1834} 1835 1836/* 1837 * Update device state. 1838 */ 1839static void 1840g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1841{ 1842 struct g_mirror_disk *disk; 1843 u_int state; 1844 1845 g_topology_assert(); 1846 1847 switch (sc->sc_state) { 1848 case G_MIRROR_DEVICE_STATE_STARTING: 1849 { 1850 struct g_mirror_disk *pdisk, *tdisk; 1851 u_int dirty, ndisks, genid, syncid; 1852 1853 KASSERT(sc->sc_provider == NULL, 1854 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1855 /* 1856 * Are we ready? We are, if all disks are connected or 1857 * if we have any disks and 'force' is true. 1858 */ 1859 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1860 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1861 ; 1862 } else if (g_mirror_ndisks(sc, -1) == 0) { 1863 /* 1864 * Disks went down in starting phase, so destroy 1865 * device. 1866 */ 1867 callout_drain(&sc->sc_callout); 1868 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1869 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1870 sc->sc_rootmount); 1871 root_mount_rel(sc->sc_rootmount); 1872 sc->sc_rootmount = NULL; 1873 return; 1874 } else { 1875 return; 1876 } 1877 1878 /* 1879 * Activate all disks with the biggest syncid. 1880 */ 1881 if (force) { 1882 /* 1883 * If 'force' is true, we have been called due to 1884 * timeout, so don't bother canceling timeout. 1885 */ 1886 ndisks = 0; 1887 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1888 if ((disk->d_flags & 1889 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1890 ndisks++; 1891 } 1892 } 1893 if (ndisks == 0) { 1894 /* No valid disks found, destroy device. */ 1895 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1896 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 1897 __LINE__, sc->sc_rootmount); 1898 root_mount_rel(sc->sc_rootmount); 1899 sc->sc_rootmount = NULL; 1900 return; 1901 } 1902 } else { 1903 /* Cancel timeout. */ 1904 callout_drain(&sc->sc_callout); 1905 } 1906 1907 /* 1908 * Find the biggest genid. 1909 */ 1910 genid = 0; 1911 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1912 if (disk->d_genid > genid) 1913 genid = disk->d_genid; 1914 } 1915 sc->sc_genid = genid; 1916 /* 1917 * Remove all disks without the biggest genid. 1918 */ 1919 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 1920 if (disk->d_genid < genid) { 1921 G_MIRROR_DEBUG(0, 1922 "Component %s (device %s) broken, skipping.", 1923 g_mirror_get_diskname(disk), sc->sc_name); 1924 g_mirror_destroy_disk(disk); 1925 } 1926 } 1927 1928 /* 1929 * Find the biggest syncid. 1930 */ 1931 syncid = 0; 1932 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1933 if (disk->d_sync.ds_syncid > syncid) 1934 syncid = disk->d_sync.ds_syncid; 1935 } 1936 1937 /* 1938 * Here we need to look for dirty disks and if all disks 1939 * with the biggest syncid are dirty, we have to choose 1940 * one with the biggest priority and rebuild the rest. 1941 */ 1942 /* 1943 * Find the number of dirty disks with the biggest syncid. 1944 * Find the number of disks with the biggest syncid. 1945 * While here, find a disk with the biggest priority. 1946 */ 1947 dirty = ndisks = 0; 1948 pdisk = NULL; 1949 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1950 if (disk->d_sync.ds_syncid != syncid) 1951 continue; 1952 if ((disk->d_flags & 1953 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1954 continue; 1955 } 1956 ndisks++; 1957 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1958 dirty++; 1959 if (pdisk == NULL || 1960 pdisk->d_priority < disk->d_priority) { 1961 pdisk = disk; 1962 } 1963 } 1964 } 1965 if (dirty == 0) { 1966 /* No dirty disks at all, great. */ 1967 } else if (dirty == ndisks) { 1968 /* 1969 * Force synchronization for all dirty disks except one 1970 * with the biggest priority. 1971 */ 1972 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1973 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1974 "master disk for synchronization.", 1975 g_mirror_get_diskname(pdisk), sc->sc_name); 1976 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1977 if (disk->d_sync.ds_syncid != syncid) 1978 continue; 1979 if ((disk->d_flags & 1980 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1981 continue; 1982 } 1983 KASSERT((disk->d_flags & 1984 G_MIRROR_DISK_FLAG_DIRTY) != 0, 1985 ("Disk %s isn't marked as dirty.", 1986 g_mirror_get_diskname(disk))); 1987 /* Skip the disk with the biggest priority. */ 1988 if (disk == pdisk) 1989 continue; 1990 disk->d_sync.ds_syncid = 0; 1991 } 1992 } else if (dirty < ndisks) { 1993 /* 1994 * Force synchronization for all dirty disks. 1995 * We have some non-dirty disks. 1996 */ 1997 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1998 if (disk->d_sync.ds_syncid != syncid) 1999 continue; 2000 if ((disk->d_flags & 2001 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2002 continue; 2003 } 2004 if ((disk->d_flags & 2005 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2006 continue; 2007 } 2008 disk->d_sync.ds_syncid = 0; 2009 } 2010 } 2011 2012 /* Reset hint. */ 2013 sc->sc_hint = NULL; 2014 sc->sc_syncid = syncid; 2015 if (force) { 2016 /* Remember to bump syncid on first write. */ 2017 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2018 } 2019 state = G_MIRROR_DEVICE_STATE_RUNNING; 2020 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2021 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2022 g_mirror_device_state2str(state)); 2023 sc->sc_state = state; 2024 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2025 state = g_mirror_determine_state(disk); 2026 g_mirror_event_send(disk, state, 2027 G_MIRROR_EVENT_DONTWAIT); 2028 if (state == G_MIRROR_DISK_STATE_STALE) 2029 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2030 } 2031 break; 2032 } 2033 case G_MIRROR_DEVICE_STATE_RUNNING: 2034 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2035 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2036 /* 2037 * No active disks or no disks at all, 2038 * so destroy device. 2039 */ 2040 if (sc->sc_provider != NULL) 2041 g_mirror_destroy_provider(sc); 2042 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2043 break; 2044 } else if (g_mirror_ndisks(sc, 2045 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2046 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2047 /* 2048 * We have active disks, launch provider if it doesn't 2049 * exist. 2050 */ 2051 if (sc->sc_provider == NULL) 2052 g_mirror_launch_provider(sc); 2053 if (sc->sc_rootmount != NULL) { 2054 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2055 __LINE__, sc->sc_rootmount); 2056 root_mount_rel(sc->sc_rootmount); 2057 sc->sc_rootmount = NULL; 2058 } 2059 } 2060 /* 2061 * Genid should be bumped immediately, so do it here. 2062 */ 2063 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2064 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2065 g_mirror_bump_genid(sc); 2066 } 2067 break; 2068 default: 2069 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2070 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2071 break; 2072 } 2073} 2074 2075/* 2076 * Update disk state and device state if needed. 2077 */ 2078#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2079 "Disk %s state changed from %s to %s (device %s).", \ 2080 g_mirror_get_diskname(disk), \ 2081 g_mirror_disk_state2str(disk->d_state), \ 2082 g_mirror_disk_state2str(state), sc->sc_name) 2083static int 2084g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2085{ 2086 struct g_mirror_softc *sc; 2087 2088 g_topology_assert(); 2089 2090 sc = disk->d_softc; 2091again: 2092 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2093 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2094 g_mirror_disk_state2str(state)); 2095 switch (state) { 2096 case G_MIRROR_DISK_STATE_NEW: 2097 /* 2098 * Possible scenarios: 2099 * 1. New disk arrive. 2100 */ 2101 /* Previous state should be NONE. */ 2102 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2103 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2104 g_mirror_disk_state2str(disk->d_state))); 2105 DISK_STATE_CHANGED(); 2106 2107 disk->d_state = state; 2108 if (LIST_EMPTY(&sc->sc_disks)) 2109 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2110 else { 2111 struct g_mirror_disk *dp; 2112 2113 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2114 if (disk->d_priority >= dp->d_priority) { 2115 LIST_INSERT_BEFORE(dp, disk, d_next); 2116 dp = NULL; 2117 break; 2118 } 2119 if (LIST_NEXT(dp, d_next) == NULL) 2120 break; 2121 } 2122 if (dp != NULL) 2123 LIST_INSERT_AFTER(dp, disk, d_next); 2124 } 2125 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2126 sc->sc_name, g_mirror_get_diskname(disk)); 2127 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2128 break; 2129 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2130 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2131 g_mirror_device_state2str(sc->sc_state), 2132 g_mirror_get_diskname(disk), 2133 g_mirror_disk_state2str(disk->d_state))); 2134 state = g_mirror_determine_state(disk); 2135 if (state != G_MIRROR_DISK_STATE_NONE) 2136 goto again; 2137 break; 2138 case G_MIRROR_DISK_STATE_ACTIVE: 2139 /* 2140 * Possible scenarios: 2141 * 1. New disk does not need synchronization. 2142 * 2. Synchronization process finished successfully. 2143 */ 2144 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2145 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2146 g_mirror_device_state2str(sc->sc_state), 2147 g_mirror_get_diskname(disk), 2148 g_mirror_disk_state2str(disk->d_state))); 2149 /* Previous state should be NEW or SYNCHRONIZING. */ 2150 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2151 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2152 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2153 g_mirror_disk_state2str(disk->d_state))); 2154 DISK_STATE_CHANGED(); 2155 2156 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2157 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2158 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2159 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2160 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2161 g_mirror_sync_stop(disk, 0); 2162 } 2163 disk->d_state = state; 2164 disk->d_sync.ds_offset = 0; 2165 disk->d_sync.ds_offset_done = 0; 2166 g_mirror_update_idle(sc, disk); 2167 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2168 sc->sc_name, g_mirror_get_diskname(disk)); 2169 break; 2170 case G_MIRROR_DISK_STATE_STALE: 2171 /* 2172 * Possible scenarios: 2173 * 1. Stale disk was connected. 2174 */ 2175 /* Previous state should be NEW. */ 2176 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2177 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2178 g_mirror_disk_state2str(disk->d_state))); 2179 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2180 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2181 g_mirror_device_state2str(sc->sc_state), 2182 g_mirror_get_diskname(disk), 2183 g_mirror_disk_state2str(disk->d_state))); 2184 /* 2185 * STALE state is only possible if device is marked 2186 * NOAUTOSYNC. 2187 */ 2188 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2189 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2190 g_mirror_device_state2str(sc->sc_state), 2191 g_mirror_get_diskname(disk), 2192 g_mirror_disk_state2str(disk->d_state))); 2193 DISK_STATE_CHANGED(); 2194 2195 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2196 disk->d_state = state; 2197 g_mirror_update_metadata(disk); 2198 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2199 sc->sc_name, g_mirror_get_diskname(disk)); 2200 break; 2201 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2202 /* 2203 * Possible scenarios: 2204 * 1. Disk which needs synchronization was connected. 2205 */ 2206 /* Previous state should be NEW. */ 2207 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2208 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2209 g_mirror_disk_state2str(disk->d_state))); 2210 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2211 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2212 g_mirror_device_state2str(sc->sc_state), 2213 g_mirror_get_diskname(disk), 2214 g_mirror_disk_state2str(disk->d_state))); 2215 DISK_STATE_CHANGED(); 2216 2217 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2218 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2219 disk->d_state = state; 2220 if (sc->sc_provider != NULL) { 2221 g_mirror_sync_start(disk); 2222 g_mirror_update_metadata(disk); 2223 } 2224 break; 2225 case G_MIRROR_DISK_STATE_DISCONNECTED: 2226 /* 2227 * Possible scenarios: 2228 * 1. Device wasn't running yet, but disk disappear. 2229 * 2. Disk was active and disapppear. 2230 * 3. Disk disappear during synchronization process. 2231 */ 2232 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2233 /* 2234 * Previous state should be ACTIVE, STALE or 2235 * SYNCHRONIZING. 2236 */ 2237 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2238 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2239 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2240 ("Wrong disk state (%s, %s).", 2241 g_mirror_get_diskname(disk), 2242 g_mirror_disk_state2str(disk->d_state))); 2243 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2244 /* Previous state should be NEW. */ 2245 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2246 ("Wrong disk state (%s, %s).", 2247 g_mirror_get_diskname(disk), 2248 g_mirror_disk_state2str(disk->d_state))); 2249 /* 2250 * Reset bumping syncid if disk disappeared in STARTING 2251 * state. 2252 */ 2253 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2254 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2255#ifdef INVARIANTS 2256 } else { 2257 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2258 sc->sc_name, 2259 g_mirror_device_state2str(sc->sc_state), 2260 g_mirror_get_diskname(disk), 2261 g_mirror_disk_state2str(disk->d_state))); 2262#endif 2263 } 2264 DISK_STATE_CHANGED(); 2265 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2266 sc->sc_name, g_mirror_get_diskname(disk)); 2267 2268 g_mirror_destroy_disk(disk); 2269 break; 2270 case G_MIRROR_DISK_STATE_DESTROY: 2271 { 2272 int error; 2273 2274 error = g_mirror_clear_metadata(disk); 2275 if (error != 0) 2276 return (error); 2277 DISK_STATE_CHANGED(); 2278 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2279 sc->sc_name, g_mirror_get_diskname(disk)); 2280 2281 g_mirror_destroy_disk(disk); 2282 sc->sc_ndisks--; 2283 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2284 g_mirror_update_metadata(disk); 2285 } 2286 break; 2287 } 2288 default: 2289 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2290 break; 2291 } 2292 return (0); 2293} 2294#undef DISK_STATE_CHANGED 2295 2296int 2297g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2298{ 2299 struct g_provider *pp; 2300 u_char *buf; 2301 int error; 2302 2303 g_topology_assert(); 2304 2305 error = g_access(cp, 1, 0, 0); 2306 if (error != 0) 2307 return (error); 2308 pp = cp->provider; 2309 g_topology_unlock(); 2310 /* Metadata are stored on last sector. */ 2311 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2312 &error); 2313 g_topology_lock(); 2314 g_access(cp, -1, 0, 0); 2315 if (buf == NULL) { 2316 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2317 cp->provider->name, error); 2318 return (error); 2319 } 2320 2321 /* Decode metadata. */ 2322 error = mirror_metadata_decode(buf, md); 2323 g_free(buf); 2324 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2325 return (EINVAL); 2326 if (md->md_version > G_MIRROR_VERSION) { 2327 G_MIRROR_DEBUG(0, 2328 "Kernel module is too old to handle metadata from %s.", 2329 cp->provider->name); 2330 return (EINVAL); 2331 } 2332 if (error != 0) { 2333 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2334 cp->provider->name); 2335 return (error); 2336 } 2337 2338 return (0); 2339} 2340 2341static int 2342g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2343 struct g_mirror_metadata *md) 2344{ 2345 2346 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2347 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2348 pp->name, md->md_did); 2349 return (EEXIST); 2350 } 2351 if (md->md_all != sc->sc_ndisks) { 2352 G_MIRROR_DEBUG(1, 2353 "Invalid '%s' field on disk %s (device %s), skipping.", 2354 "md_all", pp->name, sc->sc_name); 2355 return (EINVAL); 2356 } 2357 if (md->md_slice != sc->sc_slice) { 2358 G_MIRROR_DEBUG(1, 2359 "Invalid '%s' field on disk %s (device %s), skipping.", 2360 "md_slice", pp->name, sc->sc_name); 2361 return (EINVAL); 2362 } 2363 if (md->md_balance != sc->sc_balance) { 2364 G_MIRROR_DEBUG(1, 2365 "Invalid '%s' field on disk %s (device %s), skipping.", 2366 "md_balance", pp->name, sc->sc_name); 2367 return (EINVAL); 2368 } 2369 if (md->md_mediasize != sc->sc_mediasize) { 2370 G_MIRROR_DEBUG(1, 2371 "Invalid '%s' field on disk %s (device %s), skipping.", 2372 "md_mediasize", pp->name, sc->sc_name); 2373 return (EINVAL); 2374 } 2375 if (sc->sc_mediasize > pp->mediasize) { 2376 G_MIRROR_DEBUG(1, 2377 "Invalid size of disk %s (device %s), skipping.", pp->name, 2378 sc->sc_name); 2379 return (EINVAL); 2380 } 2381 if (md->md_sectorsize != sc->sc_sectorsize) { 2382 G_MIRROR_DEBUG(1, 2383 "Invalid '%s' field on disk %s (device %s), skipping.", 2384 "md_sectorsize", pp->name, sc->sc_name); 2385 return (EINVAL); 2386 } 2387 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2388 G_MIRROR_DEBUG(1, 2389 "Invalid sector size of disk %s (device %s), skipping.", 2390 pp->name, sc->sc_name); 2391 return (EINVAL); 2392 } 2393 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2394 G_MIRROR_DEBUG(1, 2395 "Invalid device flags on disk %s (device %s), skipping.", 2396 pp->name, sc->sc_name); 2397 return (EINVAL); 2398 } 2399 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2400 G_MIRROR_DEBUG(1, 2401 "Invalid disk flags on disk %s (device %s), skipping.", 2402 pp->name, sc->sc_name); 2403 return (EINVAL); 2404 } 2405 return (0); 2406} 2407 2408int 2409g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2410 struct g_mirror_metadata *md) 2411{ 2412 struct g_mirror_disk *disk; 2413 int error; 2414 2415 g_topology_assert(); 2416 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2417 2418 error = g_mirror_check_metadata(sc, pp, md); 2419 if (error != 0) 2420 return (error); 2421 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2422 md->md_genid < sc->sc_genid) { 2423 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2424 pp->name, sc->sc_name); 2425 return (EINVAL); 2426 } 2427 disk = g_mirror_init_disk(sc, pp, md, &error); 2428 if (disk == NULL) 2429 return (error); 2430 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2431 G_MIRROR_EVENT_WAIT); 2432 if (error != 0) 2433 return (error); 2434 if (md->md_version < G_MIRROR_VERSION) { 2435 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2436 pp->name, md->md_version, G_MIRROR_VERSION); 2437 g_mirror_update_metadata(disk); 2438 } 2439 return (0); 2440} 2441 2442static int 2443g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2444{ 2445 struct g_mirror_softc *sc; 2446 int dcr, dcw, dce; 2447 2448 g_topology_assert(); 2449 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2450 acw, ace); 2451 2452 dcr = pp->acr + acr; 2453 dcw = pp->acw + acw; 2454 dce = pp->ace + ace; 2455 2456 sc = pp->geom->softc; 2457 if (sc == NULL || LIST_EMPTY(&sc->sc_disks) || 2458 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 2459 if (acr <= 0 && acw <= 0 && ace <= 0) 2460 return (0); 2461 else 2462 return (ENXIO); 2463 } 2464 if (dcw == 0 && !sc->sc_idle) 2465 g_mirror_idle(sc, 1); 2466 return (0); 2467} 2468 2469static struct g_geom * 2470g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2471{ 2472 struct g_mirror_softc *sc; 2473 struct g_geom *gp; 2474 int error, timeout; 2475 2476 g_topology_assert(); 2477 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2478 md->md_mid); 2479 2480 /* One disk is minimum. */ 2481 if (md->md_all < 1) 2482 return (NULL); 2483 /* 2484 * Action geom. 2485 */ 2486 gp = g_new_geomf(mp, "%s", md->md_name); 2487 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2488 gp->start = g_mirror_start; 2489 gp->orphan = g_mirror_orphan; 2490 gp->access = g_mirror_access; 2491 gp->dumpconf = g_mirror_dumpconf; 2492 2493 sc->sc_id = md->md_mid; 2494 sc->sc_slice = md->md_slice; 2495 sc->sc_balance = md->md_balance; 2496 sc->sc_mediasize = md->md_mediasize; 2497 sc->sc_sectorsize = md->md_sectorsize; 2498 sc->sc_ndisks = md->md_all; 2499 sc->sc_flags = md->md_mflags; 2500 sc->sc_bump_id = 0; 2501 sc->sc_idle = 1; 2502 sc->sc_last_write = time_second; 2503 sc->sc_writes = 0; 2504 bioq_init(&sc->sc_queue); 2505 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2506 LIST_INIT(&sc->sc_disks); 2507 TAILQ_INIT(&sc->sc_events); 2508 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2509 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2510 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2511 gp->softc = sc; 2512 sc->sc_geom = gp; 2513 sc->sc_provider = NULL; 2514 /* 2515 * Synchronization geom. 2516 */ 2517 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2518 gp->softc = sc; 2519 gp->orphan = g_mirror_orphan; 2520 sc->sc_sync.ds_geom = gp; 2521 sc->sc_sync.ds_ndisks = 0; 2522 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2523 "g_mirror %s", md->md_name); 2524 if (error != 0) { 2525 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2526 sc->sc_name); 2527 g_destroy_geom(sc->sc_sync.ds_geom); 2528 mtx_destroy(&sc->sc_events_mtx); 2529 mtx_destroy(&sc->sc_queue_mtx); 2530 g_destroy_geom(sc->sc_geom); 2531 free(sc, M_MIRROR); 2532 return (NULL); 2533 } 2534 2535 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2536 2537 sc->sc_rootmount = root_mount_hold("GMIRROR"); 2538 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 2539 /* 2540 * Run timeout. 2541 */ 2542 timeout = g_mirror_timeout * hz; 2543 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2544 return (sc->sc_geom); 2545} 2546 2547int 2548g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2549{ 2550 struct g_provider *pp; 2551 2552 g_topology_assert(); 2553 2554 if (sc == NULL) 2555 return (ENXIO); 2556 pp = sc->sc_provider; 2557 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2558 if (force) { 2559 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 2560 "can't be definitely removed.", pp->name); 2561 } else { 2562 G_MIRROR_DEBUG(1, 2563 "Device %s is still open (r%dw%de%d).", pp->name, 2564 pp->acr, pp->acw, pp->ace); 2565 return (EBUSY); 2566 } 2567 } 2568 2569 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2570 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2571 g_topology_unlock(); 2572 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2573 mtx_lock(&sc->sc_queue_mtx); 2574 wakeup(sc); 2575 mtx_unlock(&sc->sc_queue_mtx); 2576 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2577 while (sc->sc_worker != NULL) 2578 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2579 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2580 g_topology_lock(); 2581 g_mirror_destroy_device(sc); 2582 free(sc, M_MIRROR); 2583 return (0); 2584} 2585 2586static void 2587g_mirror_taste_orphan(struct g_consumer *cp) 2588{ 2589 2590 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2591 cp->provider->name)); 2592} 2593 2594static struct g_geom * 2595g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2596{ 2597 struct g_mirror_metadata md; 2598 struct g_mirror_softc *sc; 2599 struct g_consumer *cp; 2600 struct g_geom *gp; 2601 int error; 2602 2603 g_topology_assert(); 2604 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2605 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2606 2607 gp = g_new_geomf(mp, "mirror:taste"); 2608 /* 2609 * This orphan function should be never called. 2610 */ 2611 gp->orphan = g_mirror_taste_orphan; 2612 cp = g_new_consumer(gp); 2613 g_attach(cp, pp); 2614 error = g_mirror_read_metadata(cp, &md); 2615 g_detach(cp); 2616 g_destroy_consumer(cp); 2617 g_destroy_geom(gp); 2618 if (error != 0) 2619 return (NULL); 2620 gp = NULL; 2621 2622 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2623 return (NULL); 2624 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 2625 return (NULL); 2626 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2627 G_MIRROR_DEBUG(0, 2628 "Device %s: provider %s marked as inactive, skipping.", 2629 md.md_name, pp->name); 2630 return (NULL); 2631 } 2632 if (g_mirror_debug >= 2) 2633 mirror_metadata_dump(&md); 2634 2635 /* 2636 * Let's check if device already exists. 2637 */ 2638 sc = NULL; 2639 LIST_FOREACH(gp, &mp->geom, geom) { 2640 sc = gp->softc; 2641 if (sc == NULL) 2642 continue; 2643 if (sc->sc_sync.ds_geom == gp) 2644 continue; 2645 if (strcmp(md.md_name, sc->sc_name) != 0) 2646 continue; 2647 if (md.md_mid != sc->sc_id) { 2648 G_MIRROR_DEBUG(0, "Device %s already configured.", 2649 sc->sc_name); 2650 return (NULL); 2651 } 2652 break; 2653 } 2654 if (gp == NULL) { 2655 gp = g_mirror_create(mp, &md); 2656 if (gp == NULL) { 2657 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2658 md.md_name); 2659 return (NULL); 2660 } 2661 sc = gp->softc; 2662 } 2663 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2664 error = g_mirror_add_disk(sc, pp, &md); 2665 if (error != 0) { 2666 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2667 pp->name, gp->name, error); 2668 if (LIST_EMPTY(&sc->sc_disks)) 2669 g_mirror_destroy(sc, 1); 2670 return (NULL); 2671 } 2672 return (gp); 2673} 2674 2675static int 2676g_mirror_destroy_geom(struct gctl_req *req __unused, 2677 struct g_class *mp __unused, struct g_geom *gp) 2678{ 2679 2680 return (g_mirror_destroy(gp->softc, 0)); 2681} 2682 2683static void 2684g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2685 struct g_consumer *cp, struct g_provider *pp) 2686{ 2687 struct g_mirror_softc *sc; 2688 2689 g_topology_assert(); 2690 2691 sc = gp->softc; 2692 if (sc == NULL) 2693 return; 2694 /* Skip synchronization geom. */ 2695 if (gp == sc->sc_sync.ds_geom) 2696 return; 2697 if (pp != NULL) { 2698 /* Nothing here. */ 2699 } else if (cp != NULL) { 2700 struct g_mirror_disk *disk; 2701 2702 disk = cp->private; 2703 if (disk == NULL) 2704 return; 2705 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2706 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2707 sbuf_printf(sb, "%s<Synchronized>", indent); 2708 if (disk->d_sync.ds_offset_done == 0) 2709 sbuf_printf(sb, "0%%"); 2710 else { 2711 sbuf_printf(sb, "%u%%", 2712 (u_int)((disk->d_sync.ds_offset_done * 100) / 2713 sc->sc_provider->mediasize)); 2714 } 2715 sbuf_printf(sb, "</Synchronized>\n"); 2716 } 2717 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2718 disk->d_sync.ds_syncid); 2719 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 2720 disk->d_genid); 2721 sbuf_printf(sb, "%s<Flags>", indent); 2722 if (disk->d_flags == 0) 2723 sbuf_printf(sb, "NONE"); 2724 else { 2725 int first = 1; 2726 2727#define ADD_FLAG(flag, name) do { \ 2728 if ((disk->d_flags & (flag)) != 0) { \ 2729 if (!first) \ 2730 sbuf_printf(sb, ", "); \ 2731 else \ 2732 first = 0; \ 2733 sbuf_printf(sb, name); \ 2734 } \ 2735} while (0) 2736 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2737 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2738 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2739 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2740 "SYNCHRONIZING"); 2741 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2742#undef ADD_FLAG 2743 } 2744 sbuf_printf(sb, "</Flags>\n"); 2745 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2746 disk->d_priority); 2747 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2748 g_mirror_disk_state2str(disk->d_state)); 2749 } else { 2750 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2751 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2752 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 2753 sbuf_printf(sb, "%s<Flags>", indent); 2754 if (sc->sc_flags == 0) 2755 sbuf_printf(sb, "NONE"); 2756 else { 2757 int first = 1; 2758 2759#define ADD_FLAG(flag, name) do { \ 2760 if ((sc->sc_flags & (flag)) != 0) { \ 2761 if (!first) \ 2762 sbuf_printf(sb, ", "); \ 2763 else \ 2764 first = 0; \ 2765 sbuf_printf(sb, name); \ 2766 } \ 2767} while (0) 2768 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2769#undef ADD_FLAG 2770 } 2771 sbuf_printf(sb, "</Flags>\n"); 2772 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2773 (u_int)sc->sc_slice); 2774 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2775 balance_name(sc->sc_balance)); 2776 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2777 sc->sc_ndisks); 2778 sbuf_printf(sb, "%s<State>", indent); 2779 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2780 sbuf_printf(sb, "%s", "STARTING"); 2781 else if (sc->sc_ndisks == 2782 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 2783 sbuf_printf(sb, "%s", "COMPLETE"); 2784 else 2785 sbuf_printf(sb, "%s", "DEGRADED"); 2786 sbuf_printf(sb, "</State>\n"); 2787 } 2788} 2789 2790static void 2791g_mirror_shutdown(void *arg, int howto) 2792{ 2793 struct g_class *mp; 2794 struct g_geom *gp, *gp2; 2795 2796 mp = arg; 2797 DROP_GIANT(); 2798 g_topology_lock(); 2799 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2800 if (gp->softc == NULL) 2801 continue; 2802 g_mirror_destroy(gp->softc, 1); 2803 } 2804 g_topology_unlock(); 2805 PICKUP_GIANT(); 2806#if 0 2807 tsleep(&gp, PRIBIO, "m:shutdown", hz * 20); 2808#endif 2809} 2810 2811static void 2812g_mirror_init(struct g_class *mp) 2813{ 2814 2815 g_mirror_ehtag = EVENTHANDLER_REGISTER(shutdown_post_sync, 2816 g_mirror_shutdown, mp, SHUTDOWN_PRI_FIRST); 2817 if (g_mirror_ehtag == NULL) 2818 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 2819} 2820 2821static void 2822g_mirror_fini(struct g_class *mp) 2823{ 2824 2825 if (g_mirror_ehtag == NULL) 2826 return; 2827 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_ehtag); 2828} 2829 2830DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2831