g_mirror.c revision 139213
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 139213 2004-12-22 23:09:32Z 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 mtx_unlock(&sc->sc_events_mtx); 218 return (ep); 219} 220 221static void 222g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 223{ 224 225 mtx_lock(&sc->sc_events_mtx); 226 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 227 mtx_unlock(&sc->sc_events_mtx); 228} 229 230static void 231g_mirror_event_cancel(struct g_mirror_disk *disk) 232{ 233 struct g_mirror_softc *sc; 234 struct g_mirror_event *ep, *tmpep; 235 236 g_topology_assert(); 237 238 sc = disk->d_softc; 239 mtx_lock(&sc->sc_events_mtx); 240 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 241 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 242 continue; 243 if (ep->e_disk != disk) 244 continue; 245 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 246 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 247 g_mirror_event_free(ep); 248 else { 249 ep->e_error = ECANCELED; 250 wakeup(ep); 251 } 252 } 253 mtx_unlock(&sc->sc_events_mtx); 254} 255 256/* 257 * Return the number of disks in given state. 258 * If state is equal to -1, count all connected disks. 259 */ 260u_int 261g_mirror_ndisks(struct g_mirror_softc *sc, int state) 262{ 263 struct g_mirror_disk *disk; 264 u_int n = 0; 265 266 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 267 if (state == -1 || disk->d_state == state) 268 n++; 269 } 270 return (n); 271} 272 273/* 274 * Find a disk in mirror by its disk ID. 275 */ 276static struct g_mirror_disk * 277g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 278{ 279 struct g_mirror_disk *disk; 280 281 g_topology_assert(); 282 283 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 284 if (disk->d_id == id) 285 return (disk); 286 } 287 return (NULL); 288} 289 290static u_int 291g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 292{ 293 struct bio *bp; 294 u_int nreqs = 0; 295 296 mtx_lock(&sc->sc_queue_mtx); 297 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 298 if (bp->bio_from == cp) 299 nreqs++; 300 } 301 mtx_unlock(&sc->sc_queue_mtx); 302 return (nreqs); 303} 304 305static int 306g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 307{ 308 309 if (cp->index > 0) { 310 G_MIRROR_DEBUG(2, 311 "I/O requests for %s exist, can't destroy it now.", 312 cp->provider->name); 313 return (1); 314 } 315 if (g_mirror_nrequests(sc, cp) > 0) { 316 G_MIRROR_DEBUG(2, 317 "I/O requests for %s in queue, can't destroy it now.", 318 cp->provider->name); 319 return (1); 320 } 321 return (0); 322} 323 324static void 325g_mirror_destroy_consumer(void *arg, int flags __unused) 326{ 327 struct g_consumer *cp; 328 329 cp = arg; 330 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 331 g_detach(cp); 332 g_destroy_consumer(cp); 333} 334 335static void 336g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 337{ 338 struct g_provider *pp; 339 int retaste_wait; 340 341 g_topology_assert(); 342 343 cp->private = NULL; 344 if (g_mirror_is_busy(sc, cp)) 345 return; 346 pp = cp->provider; 347 retaste_wait = 0; 348 if (cp->acw == 1) { 349 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 350 retaste_wait = 1; 351 } 352 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 353 -cp->acw, -cp->ace, 0); 354 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 355 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 356 if (retaste_wait) { 357 /* 358 * After retaste event was send (inside g_access()), we can send 359 * event to detach and destroy consumer. 360 * A class, which has consumer to the given provider connected 361 * will not receive retaste event for the provider. 362 * This is the way how I ignore retaste events when I close 363 * consumers opened for write: I detach and destroy consumer 364 * after retaste event is sent. 365 */ 366 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 367 return; 368 } 369 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 370 g_detach(cp); 371 g_destroy_consumer(cp); 372} 373 374static int 375g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 376{ 377 int error; 378 379 g_topology_assert(); 380 KASSERT(disk->d_consumer == NULL, 381 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 382 383 disk->d_consumer = g_new_consumer(disk->d_softc->sc_geom); 384 disk->d_consumer->private = disk; 385 disk->d_consumer->index = 0; 386 error = g_attach(disk->d_consumer, pp); 387 if (error != 0) 388 return (error); 389 error = g_access(disk->d_consumer, 1, 1, 1); 390 if (error != 0) { 391 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 392 pp->name, error); 393 return (error); 394 } 395 396 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 397 return (0); 398} 399 400static void 401g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 402{ 403 404 g_topology_assert(); 405 406 if (cp == NULL) 407 return; 408 if (cp->provider != NULL) 409 g_mirror_kill_consumer(sc, cp); 410 else 411 g_destroy_consumer(cp); 412} 413 414/* 415 * Initialize disk. This means allocate memory, create consumer, attach it 416 * to the provider and open access (r1w1e1) to it. 417 */ 418static struct g_mirror_disk * 419g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 420 struct g_mirror_metadata *md, int *errorp) 421{ 422 struct g_mirror_disk *disk; 423 int error; 424 425 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 426 if (disk == NULL) { 427 error = ENOMEM; 428 goto fail; 429 } 430 disk->d_softc = sc; 431 error = g_mirror_connect_disk(disk, pp); 432 if (error != 0) 433 goto fail; 434 disk->d_id = md->md_did; 435 disk->d_state = G_MIRROR_DISK_STATE_NONE; 436 disk->d_priority = md->md_priority; 437 disk->d_delay.sec = 0; 438 disk->d_delay.frac = 0; 439 binuptime(&disk->d_last_used); 440 disk->d_flags = md->md_dflags; 441 if (md->md_provider[0] != '\0') 442 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 443 disk->d_sync.ds_consumer = NULL; 444 disk->d_sync.ds_offset = md->md_sync_offset; 445 disk->d_sync.ds_offset_done = md->md_sync_offset; 446 disk->d_sync.ds_resync = -1; 447 disk->d_genid = md->md_genid; 448 disk->d_sync.ds_syncid = md->md_syncid; 449 if (errorp != NULL) 450 *errorp = 0; 451 return (disk); 452fail: 453 if (errorp != NULL) 454 *errorp = error; 455 if (disk != NULL) { 456 g_mirror_disconnect_consumer(sc, disk->d_consumer); 457 free(disk, M_MIRROR); 458 } 459 return (NULL); 460} 461 462static void 463g_mirror_destroy_disk(struct g_mirror_disk *disk) 464{ 465 struct g_mirror_softc *sc; 466 467 g_topology_assert(); 468 469 LIST_REMOVE(disk, d_next); 470 g_mirror_event_cancel(disk); 471 sc = disk->d_softc; 472 if (sc->sc_hint == disk) 473 sc->sc_hint = NULL; 474 switch (disk->d_state) { 475 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 476 g_mirror_sync_stop(disk, 1); 477 /* FALLTHROUGH */ 478 case G_MIRROR_DISK_STATE_NEW: 479 case G_MIRROR_DISK_STATE_STALE: 480 case G_MIRROR_DISK_STATE_ACTIVE: 481 g_mirror_disconnect_consumer(sc, disk->d_consumer); 482 free(disk, M_MIRROR); 483 break; 484 default: 485 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 486 g_mirror_get_diskname(disk), 487 g_mirror_disk_state2str(disk->d_state))); 488 } 489} 490 491static void 492g_mirror_destroy_device(struct g_mirror_softc *sc) 493{ 494 struct g_mirror_disk *disk; 495 struct g_mirror_event *ep; 496 struct g_geom *gp; 497 struct g_consumer *cp, *tmpcp; 498 499 g_topology_assert(); 500 501 gp = sc->sc_geom; 502 if (sc->sc_provider != NULL) 503 g_mirror_destroy_provider(sc); 504 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 505 disk = LIST_FIRST(&sc->sc_disks)) { 506 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 507 g_mirror_update_metadata(disk); 508 g_mirror_destroy_disk(disk); 509 } 510 while ((ep = g_mirror_event_get(sc)) != NULL) { 511 g_mirror_event_remove(sc, ep); 512 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 513 g_mirror_event_free(ep); 514 else { 515 ep->e_error = ECANCELED; 516 ep->e_flags |= G_MIRROR_EVENT_DONE; 517 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 518 mtx_lock(&sc->sc_events_mtx); 519 wakeup(ep); 520 mtx_unlock(&sc->sc_events_mtx); 521 } 522 } 523 callout_drain(&sc->sc_callout); 524 gp->softc = NULL; 525 526 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 527 g_mirror_disconnect_consumer(sc, cp); 528 } 529 sc->sc_sync.ds_geom->softc = NULL; 530 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 531 mtx_destroy(&sc->sc_queue_mtx); 532 mtx_destroy(&sc->sc_events_mtx); 533 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 534 g_wither_geom(gp, ENXIO); 535} 536 537static void 538g_mirror_orphan(struct g_consumer *cp) 539{ 540 struct g_mirror_disk *disk; 541 542 g_topology_assert(); 543 544 disk = cp->private; 545 if (disk == NULL) 546 return; 547 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_OFW; 548 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 549 G_MIRROR_EVENT_DONTWAIT); 550} 551 552static void 553g_mirror_spoiled(struct g_consumer *cp) 554{ 555 struct g_mirror_disk *disk; 556 557 g_topology_assert(); 558 559 disk = cp->private; 560 if (disk == NULL) 561 return; 562 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_IMM; 563 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 564 G_MIRROR_EVENT_DONTWAIT); 565} 566 567/* 568 * Function should return the next active disk on the list. 569 * It is possible that it will be the same disk as given. 570 * If there are no active disks on list, NULL is returned. 571 */ 572static __inline struct g_mirror_disk * 573g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 574{ 575 struct g_mirror_disk *dp; 576 577 for (dp = LIST_NEXT(disk, d_next); dp != disk; 578 dp = LIST_NEXT(dp, d_next)) { 579 if (dp == NULL) 580 dp = LIST_FIRST(&sc->sc_disks); 581 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 582 break; 583 } 584 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 585 return (NULL); 586 return (dp); 587} 588 589static struct g_mirror_disk * 590g_mirror_get_disk(struct g_mirror_softc *sc) 591{ 592 struct g_mirror_disk *disk; 593 594 if (sc->sc_hint == NULL) { 595 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 596 if (sc->sc_hint == NULL) 597 return (NULL); 598 } 599 disk = sc->sc_hint; 600 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 601 disk = g_mirror_find_next(sc, disk); 602 if (disk == NULL) 603 return (NULL); 604 } 605 sc->sc_hint = g_mirror_find_next(sc, disk); 606 return (disk); 607} 608 609static int 610g_mirror_write_metadata(struct g_mirror_disk *disk, 611 struct g_mirror_metadata *md) 612{ 613 struct g_mirror_softc *sc; 614 struct g_consumer *cp; 615 off_t offset, length; 616 u_char *sector; 617 int error = 0; 618 619 g_topology_assert(); 620 621 sc = disk->d_softc; 622 cp = disk->d_consumer; 623 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 624 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 625 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 626 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 627 cp->acw, cp->ace)); 628 length = cp->provider->sectorsize; 629 offset = cp->provider->mediasize - length; 630 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 631 if (md != NULL) 632 mirror_metadata_encode(md, sector); 633 g_topology_unlock(); 634 error = g_write_data(cp, offset, sector, length); 635 g_topology_lock(); 636 free(sector, M_MIRROR); 637 if (error != 0) { 638 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_GENID_IMM; 639 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 640 G_MIRROR_EVENT_DONTWAIT); 641 } 642 return (error); 643} 644 645static int 646g_mirror_clear_metadata(struct g_mirror_disk *disk) 647{ 648 int error; 649 650 g_topology_assert(); 651 error = g_mirror_write_metadata(disk, NULL); 652 if (error == 0) { 653 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 654 g_mirror_get_diskname(disk)); 655 } else { 656 G_MIRROR_DEBUG(0, 657 "Cannot clear metadata on disk %s (error=%d).", 658 g_mirror_get_diskname(disk), error); 659 } 660 return (error); 661} 662 663void 664g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 665 struct g_mirror_metadata *md) 666{ 667 668 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 669 md->md_version = G_MIRROR_VERSION; 670 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 671 md->md_mid = sc->sc_id; 672 md->md_all = sc->sc_ndisks; 673 md->md_slice = sc->sc_slice; 674 md->md_balance = sc->sc_balance; 675 md->md_genid = sc->sc_genid; 676 md->md_mediasize = sc->sc_mediasize; 677 md->md_sectorsize = sc->sc_sectorsize; 678 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 679 bzero(md->md_provider, sizeof(md->md_provider)); 680 if (disk == NULL) { 681 md->md_did = arc4random(); 682 md->md_priority = 0; 683 md->md_syncid = 0; 684 md->md_dflags = 0; 685 md->md_sync_offset = 0; 686 } else { 687 md->md_did = disk->d_id; 688 md->md_priority = disk->d_priority; 689 md->md_syncid = disk->d_sync.ds_syncid; 690 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 691 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 692 md->md_sync_offset = disk->d_sync.ds_offset_done; 693 else 694 md->md_sync_offset = 0; 695 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 696 strlcpy(md->md_provider, 697 disk->d_consumer->provider->name, 698 sizeof(md->md_provider)); 699 } 700 } 701} 702 703void 704g_mirror_update_metadata(struct g_mirror_disk *disk) 705{ 706 struct g_mirror_metadata md; 707 int error; 708 709 g_topology_assert(); 710 g_mirror_fill_metadata(disk->d_softc, disk, &md); 711 error = g_mirror_write_metadata(disk, &md); 712 if (error == 0) { 713 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 714 g_mirror_get_diskname(disk)); 715 } else { 716 G_MIRROR_DEBUG(0, 717 "Cannot update metadata on disk %s (error=%d).", 718 g_mirror_get_diskname(disk), error); 719 } 720} 721 722static void 723g_mirror_bump_syncid(struct g_mirror_softc *sc) 724{ 725 struct g_mirror_disk *disk; 726 727 g_topology_assert(); 728 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 729 ("%s called with no active disks (device=%s).", __func__, 730 sc->sc_name)); 731 732 sc->sc_syncid++; 733 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 734 sc->sc_syncid); 735 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 736 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 737 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 738 disk->d_sync.ds_syncid = sc->sc_syncid; 739 g_mirror_update_metadata(disk); 740 } 741 } 742} 743 744static void 745g_mirror_bump_genid(struct g_mirror_softc *sc) 746{ 747 struct g_mirror_disk *disk; 748 749 g_topology_assert(); 750 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 751 ("%s called with no active disks (device=%s).", __func__, 752 sc->sc_name)); 753 754 sc->sc_genid++; 755 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 756 sc->sc_genid); 757 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 758 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 759 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 760 g_mirror_update_metadata(disk); 761 } 762 } 763} 764 765static void 766g_mirror_idle(struct g_mirror_softc *sc) 767{ 768 struct g_mirror_disk *disk; 769 770 if (sc->sc_provider == NULL || sc->sc_provider->acw == 0) 771 return; 772 sc->sc_idle = 1; 773 g_topology_lock(); 774 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 775 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 776 continue; 777 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 778 g_mirror_get_diskname(disk), sc->sc_name); 779 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 780 g_mirror_update_metadata(disk); 781 } 782 g_topology_unlock(); 783} 784 785static void 786g_mirror_unidle(struct g_mirror_softc *sc) 787{ 788 struct g_mirror_disk *disk; 789 790 sc->sc_idle = 0; 791 g_topology_lock(); 792 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 793 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 794 continue; 795 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 796 g_mirror_get_diskname(disk), sc->sc_name); 797 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 798 g_mirror_update_metadata(disk); 799 } 800 g_topology_unlock(); 801} 802 803/* 804 * Return 1 if we should check if mirror is idling. 805 */ 806static int 807g_mirror_check_idle(struct g_mirror_softc *sc) 808{ 809 struct g_mirror_disk *disk; 810 811 if (sc->sc_idle) 812 return (0); 813 if (sc->sc_provider != NULL && sc->sc_provider->acw == 0) 814 return (0); 815 /* 816 * Check if there are no in-flight requests. 817 */ 818 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 819 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 820 continue; 821 if (disk->d_consumer->index > 0) 822 return (0); 823 } 824 return (1); 825} 826 827static __inline int 828bintime_cmp(struct bintime *bt1, struct bintime *bt2) 829{ 830 831 if (bt1->sec < bt2->sec) 832 return (-1); 833 else if (bt1->sec > bt2->sec) 834 return (1); 835 if (bt1->frac < bt2->frac) 836 return (-1); 837 else if (bt1->frac > bt2->frac) 838 return (1); 839 return (0); 840} 841 842static void 843g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 844{ 845 846 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 847 return; 848 binuptime(&disk->d_delay); 849 bintime_sub(&disk->d_delay, &bp->bio_t0); 850} 851 852static void 853g_mirror_done(struct bio *bp) 854{ 855 struct g_mirror_softc *sc; 856 857 sc = bp->bio_from->geom->softc; 858 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 859 mtx_lock(&sc->sc_queue_mtx); 860 bioq_disksort(&sc->sc_queue, bp); 861 wakeup(sc); 862 mtx_unlock(&sc->sc_queue_mtx); 863} 864 865static void 866g_mirror_regular_request(struct bio *bp) 867{ 868 struct g_mirror_softc *sc; 869 struct g_mirror_disk *disk; 870 struct bio *pbp; 871 872 g_topology_assert_not(); 873 874 bp->bio_from->index--; 875 pbp = bp->bio_parent; 876 sc = pbp->bio_to->geom->softc; 877 disk = bp->bio_from->private; 878 if (disk == NULL) { 879 g_topology_lock(); 880 g_mirror_kill_consumer(sc, bp->bio_from); 881 g_topology_unlock(); 882 } else { 883 g_mirror_update_delay(disk, bp); 884 } 885 886 pbp->bio_inbed++; 887 KASSERT(pbp->bio_inbed <= pbp->bio_children, 888 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 889 pbp->bio_children)); 890 if (bp->bio_error == 0 && pbp->bio_error == 0) { 891 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 892 g_destroy_bio(bp); 893 if (pbp->bio_children == pbp->bio_inbed) { 894 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 895 pbp->bio_completed = pbp->bio_length; 896 g_io_deliver(pbp, pbp->bio_error); 897 } 898 return; 899 } else if (bp->bio_error != 0) { 900 if (pbp->bio_error == 0) 901 pbp->bio_error = bp->bio_error; 902 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 903 bp->bio_error); 904 if (disk != NULL) { 905 sc->sc_bump_id |= G_MIRROR_BUMP_GENID_IMM; 906 g_mirror_event_send(disk, 907 G_MIRROR_DISK_STATE_DISCONNECTED, 908 G_MIRROR_EVENT_DONTWAIT); 909 } 910 switch (pbp->bio_cmd) { 911 case BIO_DELETE: 912 case BIO_WRITE: 913 pbp->bio_inbed--; 914 pbp->bio_children--; 915 break; 916 } 917 } 918 g_destroy_bio(bp); 919 920 switch (pbp->bio_cmd) { 921 case BIO_READ: 922 if (pbp->bio_children == pbp->bio_inbed) { 923 pbp->bio_error = 0; 924 mtx_lock(&sc->sc_queue_mtx); 925 bioq_disksort(&sc->sc_queue, pbp); 926 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 927 wakeup(sc); 928 mtx_unlock(&sc->sc_queue_mtx); 929 } 930 break; 931 case BIO_DELETE: 932 case BIO_WRITE: 933 if (pbp->bio_children == 0) { 934 /* 935 * All requests failed. 936 */ 937 } else if (pbp->bio_inbed < pbp->bio_children) { 938 /* Do nothing. */ 939 break; 940 } else if (pbp->bio_children == pbp->bio_inbed) { 941 /* Some requests succeeded. */ 942 pbp->bio_error = 0; 943 pbp->bio_completed = pbp->bio_length; 944 } 945 g_io_deliver(pbp, pbp->bio_error); 946 break; 947 default: 948 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 949 break; 950 } 951} 952 953static void 954g_mirror_sync_done(struct bio *bp) 955{ 956 struct g_mirror_softc *sc; 957 958 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 959 sc = bp->bio_from->geom->softc; 960 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 961 mtx_lock(&sc->sc_queue_mtx); 962 bioq_disksort(&sc->sc_queue, bp); 963 wakeup(sc); 964 mtx_unlock(&sc->sc_queue_mtx); 965} 966 967static void 968g_mirror_start(struct bio *bp) 969{ 970 struct g_mirror_softc *sc; 971 972 sc = bp->bio_to->geom->softc; 973 /* 974 * If sc == NULL or there are no valid disks, provider's error 975 * should be set and g_mirror_start() should not be called at all. 976 */ 977 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 978 ("Provider's error should be set (error=%d)(mirror=%s).", 979 bp->bio_to->error, bp->bio_to->name)); 980 G_MIRROR_LOGREQ(3, bp, "Request received."); 981 982 switch (bp->bio_cmd) { 983 case BIO_READ: 984 case BIO_WRITE: 985 case BIO_DELETE: 986 break; 987 case BIO_GETATTR: 988 default: 989 g_io_deliver(bp, EOPNOTSUPP); 990 return; 991 } 992 mtx_lock(&sc->sc_queue_mtx); 993 bioq_disksort(&sc->sc_queue, bp); 994 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 995 wakeup(sc); 996 mtx_unlock(&sc->sc_queue_mtx); 997} 998 999/* 1000 * Send one synchronization request. 1001 */ 1002static void 1003g_mirror_sync_one(struct g_mirror_disk *disk) 1004{ 1005 struct g_mirror_softc *sc; 1006 struct bio *bp; 1007 1008 sc = disk->d_softc; 1009 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1010 ("Disk %s is not marked for synchronization.", 1011 g_mirror_get_diskname(disk))); 1012 1013 bp = g_new_bio(); 1014 if (bp == NULL) 1015 return; 1016 bp->bio_parent = NULL; 1017 bp->bio_cmd = BIO_READ; 1018 bp->bio_offset = disk->d_sync.ds_offset; 1019 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1020 bp->bio_cflags = 0; 1021 bp->bio_done = g_mirror_sync_done; 1022 bp->bio_data = disk->d_sync.ds_data; 1023 if (bp->bio_data == NULL) { 1024 g_destroy_bio(bp); 1025 return; 1026 } 1027 disk->d_sync.ds_offset += bp->bio_length; 1028 bp->bio_to = sc->sc_provider; 1029 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1030 disk->d_sync.ds_consumer->index++; 1031 g_io_request(bp, disk->d_sync.ds_consumer); 1032} 1033 1034static void 1035g_mirror_sync_request(struct bio *bp) 1036{ 1037 struct g_mirror_softc *sc; 1038 struct g_mirror_disk *disk; 1039 1040 bp->bio_from->index--; 1041 sc = bp->bio_from->geom->softc; 1042 disk = bp->bio_from->private; 1043 if (disk == NULL) { 1044 g_topology_lock(); 1045 g_mirror_kill_consumer(sc, bp->bio_from); 1046 g_topology_unlock(); 1047 g_destroy_bio(bp); 1048 return; 1049 } 1050 1051 /* 1052 * Synchronization request. 1053 */ 1054 switch (bp->bio_cmd) { 1055 case BIO_READ: 1056 { 1057 struct g_consumer *cp; 1058 1059 if (bp->bio_error != 0) { 1060 G_MIRROR_LOGREQ(0, bp, 1061 "Synchronization request failed (error=%d).", 1062 bp->bio_error); 1063 g_destroy_bio(bp); 1064 return; 1065 } 1066 G_MIRROR_LOGREQ(3, bp, 1067 "Synchronization request half-finished."); 1068 bp->bio_cmd = BIO_WRITE; 1069 bp->bio_cflags = 0; 1070 cp = disk->d_consumer; 1071 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1072 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1073 cp->acr, cp->acw, cp->ace)); 1074 cp->index++; 1075 g_io_request(bp, cp); 1076 return; 1077 } 1078 case BIO_WRITE: 1079 { 1080 struct g_mirror_disk_sync *sync; 1081 1082 if (bp->bio_error != 0) { 1083 G_MIRROR_LOGREQ(0, bp, 1084 "Synchronization request failed (error=%d).", 1085 bp->bio_error); 1086 g_destroy_bio(bp); 1087 sc->sc_bump_id |= G_MIRROR_BUMP_GENID_IMM; 1088 g_mirror_event_send(disk, 1089 G_MIRROR_DISK_STATE_DISCONNECTED, 1090 G_MIRROR_EVENT_DONTWAIT); 1091 return; 1092 } 1093 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1094 sync = &disk->d_sync; 1095 sync->ds_offset_done = bp->bio_offset + bp->bio_length; 1096 g_destroy_bio(bp); 1097 if (sync->ds_resync != -1) 1098 break; 1099 if (sync->ds_offset_done == sc->sc_provider->mediasize) { 1100 /* 1101 * Disk up-to-date, activate it. 1102 */ 1103 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1104 G_MIRROR_EVENT_DONTWAIT); 1105 return; 1106 } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) { 1107 /* 1108 * Update offset_done on every 100 blocks. 1109 * XXX: This should be configurable. 1110 */ 1111 g_topology_lock(); 1112 g_mirror_update_metadata(disk); 1113 g_topology_unlock(); 1114 } 1115 return; 1116 } 1117 default: 1118 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1119 bp->bio_cmd, sc->sc_name)); 1120 break; 1121 } 1122} 1123 1124static void 1125g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1126{ 1127 struct g_mirror_disk *disk; 1128 struct g_consumer *cp; 1129 struct bio *cbp; 1130 1131 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1132 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1133 break; 1134 } 1135 if (disk == NULL) { 1136 if (bp->bio_error == 0) 1137 bp->bio_error = ENXIO; 1138 g_io_deliver(bp, bp->bio_error); 1139 return; 1140 } 1141 cbp = g_clone_bio(bp); 1142 if (cbp == NULL) { 1143 if (bp->bio_error == 0) 1144 bp->bio_error = ENOMEM; 1145 g_io_deliver(bp, bp->bio_error); 1146 return; 1147 } 1148 /* 1149 * Fill in the component buf structure. 1150 */ 1151 cp = disk->d_consumer; 1152 cbp->bio_done = g_mirror_done; 1153 cbp->bio_to = cp->provider; 1154 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1155 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1156 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1157 cp->acw, cp->ace)); 1158 cp->index++; 1159 g_io_request(cbp, cp); 1160} 1161 1162static void 1163g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1164{ 1165 struct g_mirror_disk *disk; 1166 struct g_consumer *cp; 1167 struct bio *cbp; 1168 1169 disk = g_mirror_get_disk(sc); 1170 if (disk == NULL) { 1171 if (bp->bio_error == 0) 1172 bp->bio_error = ENXIO; 1173 g_io_deliver(bp, bp->bio_error); 1174 return; 1175 } 1176 cbp = g_clone_bio(bp); 1177 if (cbp == NULL) { 1178 if (bp->bio_error == 0) 1179 bp->bio_error = ENOMEM; 1180 g_io_deliver(bp, bp->bio_error); 1181 return; 1182 } 1183 /* 1184 * Fill in the component buf structure. 1185 */ 1186 cp = disk->d_consumer; 1187 cbp->bio_done = g_mirror_done; 1188 cbp->bio_to = cp->provider; 1189 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1190 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1191 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1192 cp->acw, cp->ace)); 1193 cp->index++; 1194 g_io_request(cbp, cp); 1195} 1196 1197static void 1198g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1199{ 1200 struct g_mirror_disk *disk, *dp; 1201 struct g_consumer *cp; 1202 struct bio *cbp; 1203 struct bintime curtime; 1204 1205 binuptime(&curtime); 1206 /* 1207 * Find a disk which the smallest load. 1208 */ 1209 disk = NULL; 1210 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1211 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1212 continue; 1213 /* If disk wasn't used for more than 2 sec, use it. */ 1214 if (curtime.sec - dp->d_last_used.sec >= 2) { 1215 disk = dp; 1216 break; 1217 } 1218 if (disk == NULL || 1219 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1220 disk = dp; 1221 } 1222 } 1223 cbp = g_clone_bio(bp); 1224 if (cbp == NULL) { 1225 if (bp->bio_error == 0) 1226 bp->bio_error = ENOMEM; 1227 g_io_deliver(bp, bp->bio_error); 1228 return; 1229 } 1230 /* 1231 * Fill in the component buf structure. 1232 */ 1233 cp = disk->d_consumer; 1234 cbp->bio_done = g_mirror_done; 1235 cbp->bio_to = cp->provider; 1236 binuptime(&disk->d_last_used); 1237 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1238 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1239 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1240 cp->acw, cp->ace)); 1241 cp->index++; 1242 g_io_request(cbp, cp); 1243} 1244 1245static void 1246g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1247{ 1248 struct bio_queue_head queue; 1249 struct g_mirror_disk *disk; 1250 struct g_consumer *cp; 1251 struct bio *cbp; 1252 off_t left, mod, offset, slice; 1253 u_char *data; 1254 u_int ndisks; 1255 1256 if (bp->bio_length <= sc->sc_slice) { 1257 g_mirror_request_round_robin(sc, bp); 1258 return; 1259 } 1260 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1261 slice = bp->bio_length / ndisks; 1262 mod = slice % sc->sc_provider->sectorsize; 1263 if (mod != 0) 1264 slice += sc->sc_provider->sectorsize - mod; 1265 /* 1266 * Allocate all bios before sending any request, so we can 1267 * return ENOMEM in nice and clean way. 1268 */ 1269 left = bp->bio_length; 1270 offset = bp->bio_offset; 1271 data = bp->bio_data; 1272 bioq_init(&queue); 1273 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1274 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1275 continue; 1276 cbp = g_clone_bio(bp); 1277 if (cbp == NULL) { 1278 for (cbp = bioq_first(&queue); cbp != NULL; 1279 cbp = bioq_first(&queue)) { 1280 bioq_remove(&queue, cbp); 1281 g_destroy_bio(cbp); 1282 } 1283 if (bp->bio_error == 0) 1284 bp->bio_error = ENOMEM; 1285 g_io_deliver(bp, bp->bio_error); 1286 return; 1287 } 1288 bioq_insert_tail(&queue, cbp); 1289 cbp->bio_done = g_mirror_done; 1290 cbp->bio_caller1 = disk; 1291 cbp->bio_to = disk->d_consumer->provider; 1292 cbp->bio_offset = offset; 1293 cbp->bio_data = data; 1294 cbp->bio_length = MIN(left, slice); 1295 left -= cbp->bio_length; 1296 if (left == 0) 1297 break; 1298 offset += cbp->bio_length; 1299 data += cbp->bio_length; 1300 } 1301 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1302 bioq_remove(&queue, cbp); 1303 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1304 disk = cbp->bio_caller1; 1305 cbp->bio_caller1 = NULL; 1306 cp = disk->d_consumer; 1307 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1308 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1309 cp->acr, cp->acw, cp->ace)); 1310 disk->d_consumer->index++; 1311 g_io_request(cbp, disk->d_consumer); 1312 } 1313} 1314 1315static void 1316g_mirror_register_request(struct bio *bp) 1317{ 1318 struct g_mirror_softc *sc; 1319 1320 sc = bp->bio_to->geom->softc; 1321 switch (bp->bio_cmd) { 1322 case BIO_READ: 1323 switch (sc->sc_balance) { 1324 case G_MIRROR_BALANCE_LOAD: 1325 g_mirror_request_load(sc, bp); 1326 break; 1327 case G_MIRROR_BALANCE_PREFER: 1328 g_mirror_request_prefer(sc, bp); 1329 break; 1330 case G_MIRROR_BALANCE_ROUND_ROBIN: 1331 g_mirror_request_round_robin(sc, bp); 1332 break; 1333 case G_MIRROR_BALANCE_SPLIT: 1334 g_mirror_request_split(sc, bp); 1335 break; 1336 } 1337 return; 1338 case BIO_WRITE: 1339 case BIO_DELETE: 1340 { 1341 struct g_mirror_disk *disk; 1342 struct g_mirror_disk_sync *sync; 1343 struct bio_queue_head queue; 1344 struct g_consumer *cp; 1345 struct bio *cbp; 1346 1347 if (sc->sc_idle) 1348 g_mirror_unidle(sc); 1349 /* 1350 * Allocate all bios before sending any request, so we can 1351 * return ENOMEM in nice and clean way. 1352 */ 1353 bioq_init(&queue); 1354 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1355 sync = &disk->d_sync; 1356 switch (disk->d_state) { 1357 case G_MIRROR_DISK_STATE_ACTIVE: 1358 break; 1359 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1360 if (bp->bio_offset >= sync->ds_offset) 1361 continue; 1362 else if (bp->bio_offset + bp->bio_length > 1363 sync->ds_offset_done && 1364 (bp->bio_offset < sync->ds_resync || 1365 sync->ds_resync == -1)) { 1366 sync->ds_resync = bp->bio_offset - 1367 (bp->bio_offset % MAXPHYS); 1368 } 1369 break; 1370 default: 1371 continue; 1372 } 1373 cbp = g_clone_bio(bp); 1374 if (cbp == NULL) { 1375 for (cbp = bioq_first(&queue); cbp != NULL; 1376 cbp = bioq_first(&queue)) { 1377 bioq_remove(&queue, cbp); 1378 g_destroy_bio(cbp); 1379 } 1380 if (bp->bio_error == 0) 1381 bp->bio_error = ENOMEM; 1382 g_io_deliver(bp, bp->bio_error); 1383 return; 1384 } 1385 bioq_insert_tail(&queue, cbp); 1386 cbp->bio_done = g_mirror_done; 1387 cp = disk->d_consumer; 1388 cbp->bio_caller1 = cp; 1389 cbp->bio_to = cp->provider; 1390 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1391 ("Consumer %s not opened (r%dw%de%d).", 1392 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1393 } 1394 for (cbp = bioq_first(&queue); cbp != NULL; 1395 cbp = bioq_first(&queue)) { 1396 bioq_remove(&queue, cbp); 1397 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1398 cp = cbp->bio_caller1; 1399 cbp->bio_caller1 = NULL; 1400 cp->index++; 1401 g_io_request(cbp, cp); 1402 } 1403 /* 1404 * Bump syncid on first write. 1405 */ 1406 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_OFW) != 0) { 1407 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1408 g_topology_lock(); 1409 g_mirror_bump_syncid(sc); 1410 g_topology_unlock(); 1411 } 1412 return; 1413 } 1414 default: 1415 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1416 bp->bio_cmd, sc->sc_name)); 1417 break; 1418 } 1419} 1420 1421static int 1422g_mirror_can_destroy(struct g_mirror_softc *sc) 1423{ 1424 struct g_geom *gp; 1425 struct g_consumer *cp; 1426 1427 g_topology_assert(); 1428 gp = sc->sc_geom; 1429 LIST_FOREACH(cp, &gp->consumer, consumer) { 1430 if (g_mirror_is_busy(sc, cp)) 1431 return (0); 1432 } 1433 gp = sc->sc_sync.ds_geom; 1434 LIST_FOREACH(cp, &gp->consumer, consumer) { 1435 if (g_mirror_is_busy(sc, cp)) 1436 return (0); 1437 } 1438 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1439 sc->sc_name); 1440 return (1); 1441} 1442 1443static int 1444g_mirror_try_destroy(struct g_mirror_softc *sc) 1445{ 1446 1447 g_topology_lock(); 1448 if (!g_mirror_can_destroy(sc)) { 1449 g_topology_unlock(); 1450 return (0); 1451 } 1452 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1453 g_topology_unlock(); 1454 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1455 &sc->sc_worker); 1456 wakeup(&sc->sc_worker); 1457 sc->sc_worker = NULL; 1458 } else { 1459 g_mirror_destroy_device(sc); 1460 g_topology_unlock(); 1461 free(sc, M_MIRROR); 1462 } 1463 return (1); 1464} 1465 1466/* 1467 * Worker thread. 1468 */ 1469static void 1470g_mirror_worker(void *arg) 1471{ 1472 struct g_mirror_softc *sc; 1473 struct g_mirror_disk *disk; 1474 struct g_mirror_disk_sync *sync; 1475 struct g_mirror_event *ep; 1476 struct bio *bp; 1477 u_int nreqs; 1478 1479 sc = arg; 1480 curthread->td_base_pri = PRIBIO; 1481 1482 nreqs = 0; 1483 for (;;) { 1484 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1485 /* 1486 * First take a look at events. 1487 * This is important to handle events before any I/O requests. 1488 */ 1489 ep = g_mirror_event_get(sc); 1490 if (ep != NULL && g_topology_try_lock()) { 1491 g_mirror_event_remove(sc, ep); 1492 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1493 /* Update only device status. */ 1494 G_MIRROR_DEBUG(3, 1495 "Running event for device %s.", 1496 sc->sc_name); 1497 ep->e_error = 0; 1498 g_mirror_update_device(sc, 1); 1499 } else { 1500 /* Update disk status. */ 1501 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1502 g_mirror_get_diskname(ep->e_disk)); 1503 ep->e_error = g_mirror_update_disk(ep->e_disk, 1504 ep->e_state); 1505 if (ep->e_error == 0) 1506 g_mirror_update_device(sc, 0); 1507 } 1508 g_topology_unlock(); 1509 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1510 KASSERT(ep->e_error == 0, 1511 ("Error cannot be handled.")); 1512 g_mirror_event_free(ep); 1513 } else { 1514 ep->e_flags |= G_MIRROR_EVENT_DONE; 1515 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1516 ep); 1517 mtx_lock(&sc->sc_events_mtx); 1518 wakeup(ep); 1519 mtx_unlock(&sc->sc_events_mtx); 1520 } 1521 if ((sc->sc_flags & 1522 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1523 if (g_mirror_try_destroy(sc)) 1524 kthread_exit(0); 1525 } 1526 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1527 continue; 1528 } 1529 /* 1530 * Now I/O requests. 1531 */ 1532 /* Get first request from the queue. */ 1533 mtx_lock(&sc->sc_queue_mtx); 1534 bp = bioq_first(&sc->sc_queue); 1535 if (bp == NULL) { 1536 if (ep != NULL) { 1537 /* 1538 * No I/O requests and topology lock was 1539 * already held? Try again. 1540 */ 1541 mtx_unlock(&sc->sc_queue_mtx); 1542 continue; 1543 } 1544 if ((sc->sc_flags & 1545 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1546 mtx_unlock(&sc->sc_queue_mtx); 1547 if (g_mirror_try_destroy(sc)) 1548 kthread_exit(0); 1549 mtx_lock(&sc->sc_queue_mtx); 1550 } 1551 } 1552 if (sc->sc_sync.ds_ndisks > 0 && 1553 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1554 mtx_unlock(&sc->sc_queue_mtx); 1555 /* 1556 * It is time for synchronization... 1557 */ 1558 nreqs = 0; 1559 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1560 if (disk->d_state != 1561 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1562 continue; 1563 } 1564 sync = &disk->d_sync; 1565 if (sync->ds_offset >= 1566 sc->sc_provider->mediasize) { 1567 continue; 1568 } 1569 if (sync->ds_offset > sync->ds_offset_done) 1570 continue; 1571 if (sync->ds_resync != -1) { 1572 sync->ds_offset = sync->ds_resync; 1573 sync->ds_offset_done = sync->ds_resync; 1574 sync->ds_resync = -1; 1575 } 1576 g_mirror_sync_one(disk); 1577 } 1578 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1579 goto sleep; 1580 } 1581 if (bp == NULL) { 1582 if (g_mirror_check_idle(sc)) { 1583 u_int idletime; 1584 1585 idletime = g_mirror_idletime; 1586 if (idletime == 0) 1587 idletime = 1; 1588 idletime *= hz; 1589 if (msleep(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1590 "m:w1", idletime) == EWOULDBLOCK) { 1591 G_MIRROR_DEBUG(5, "%s: I'm here 3.", 1592 __func__); 1593 /* 1594 * No I/O requests in 'idletime' seconds, 1595 * so mark components as clean. 1596 */ 1597 g_mirror_idle(sc); 1598 } 1599 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1600 } else { 1601 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1602 "m:w2", 0); 1603 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1604 } 1605 continue; 1606 } 1607 nreqs++; 1608 bioq_remove(&sc->sc_queue, bp); 1609 mtx_unlock(&sc->sc_queue_mtx); 1610 1611 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1612 g_mirror_regular_request(bp); 1613 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1614 u_int timeout, sps; 1615 1616 g_mirror_sync_request(bp); 1617sleep: 1618 sps = g_mirror_syncs_per_sec; 1619 if (sps == 0) { 1620 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1621 continue; 1622 } 1623 if (ep != NULL) { 1624 /* 1625 * We have some pending events, don't sleep now. 1626 */ 1627 G_MIRROR_DEBUG(5, "%s: I'm here 7.", __func__); 1628 continue; 1629 } 1630 mtx_lock(&sc->sc_queue_mtx); 1631 if (bioq_first(&sc->sc_queue) != NULL) { 1632 mtx_unlock(&sc->sc_queue_mtx); 1633 G_MIRROR_DEBUG(5, "%s: I'm here 8.", __func__); 1634 continue; 1635 } 1636 timeout = hz / sps; 1637 if (timeout == 0) 1638 timeout = 1; 1639 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w3", 1640 timeout); 1641 } else { 1642 g_mirror_register_request(bp); 1643 } 1644 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1645 } 1646} 1647 1648/* 1649 * Open disk's consumer if needed. 1650 */ 1651static void 1652g_mirror_update_access(struct g_mirror_disk *disk) 1653{ 1654 struct g_provider *pp; 1655 1656 g_topology_assert(); 1657 1658 pp = disk->d_softc->sc_provider; 1659 if (pp == NULL) 1660 return; 1661 if (pp->acw > 0) { 1662 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1663 G_MIRROR_DEBUG(1, 1664 "Disk %s (device %s) marked as dirty.", 1665 g_mirror_get_diskname(disk), 1666 disk->d_softc->sc_name); 1667 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1668 } 1669 } else if (pp->acw == 0) { 1670 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1671 G_MIRROR_DEBUG(1, 1672 "Disk %s (device %s) marked as clean.", 1673 g_mirror_get_diskname(disk), 1674 disk->d_softc->sc_name); 1675 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1676 } 1677 } 1678} 1679 1680static void 1681g_mirror_sync_start(struct g_mirror_disk *disk) 1682{ 1683 struct g_mirror_softc *sc; 1684 int error; 1685 1686 g_topology_assert(); 1687 1688 sc = disk->d_softc; 1689 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1690 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1691 sc->sc_state)); 1692 1693 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1694 g_mirror_get_diskname(disk)); 1695 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1696 KASSERT(disk->d_sync.ds_consumer == NULL, 1697 ("Sync consumer already exists (device=%s, disk=%s).", 1698 sc->sc_name, g_mirror_get_diskname(disk))); 1699 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1700 disk->d_sync.ds_consumer->private = disk; 1701 disk->d_sync.ds_consumer->index = 0; 1702 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1703 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1704 disk->d_softc->sc_name, error)); 1705 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1706 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1707 disk->d_softc->sc_name, error)); 1708 disk->d_sync.ds_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1709 sc->sc_sync.ds_ndisks++; 1710} 1711 1712/* 1713 * Stop synchronization process. 1714 * type: 0 - synchronization finished 1715 * 1 - synchronization stopped 1716 */ 1717static void 1718g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1719{ 1720 1721 g_topology_assert(); 1722 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1723 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1724 g_mirror_disk_state2str(disk->d_state))); 1725 if (disk->d_sync.ds_consumer == NULL) 1726 return; 1727 1728 if (type == 0) { 1729 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1730 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1731 } else /* if (type == 1) */ { 1732 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1733 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1734 } 1735 g_mirror_kill_consumer(disk->d_softc, disk->d_sync.ds_consumer); 1736 free(disk->d_sync.ds_data, M_MIRROR); 1737 disk->d_sync.ds_consumer = NULL; 1738 disk->d_softc->sc_sync.ds_ndisks--; 1739 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1740} 1741 1742static void 1743g_mirror_launch_provider(struct g_mirror_softc *sc) 1744{ 1745 struct g_mirror_disk *disk; 1746 struct g_provider *pp; 1747 1748 g_topology_assert(); 1749 1750 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1751 pp->mediasize = sc->sc_mediasize; 1752 pp->sectorsize = sc->sc_sectorsize; 1753 sc->sc_provider = pp; 1754 g_error_provider(pp, 0); 1755 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1756 pp->name); 1757 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1758 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1759 g_mirror_sync_start(disk); 1760 } 1761} 1762 1763static void 1764g_mirror_destroy_provider(struct g_mirror_softc *sc) 1765{ 1766 struct g_mirror_disk *disk; 1767 struct bio *bp; 1768 1769 g_topology_assert(); 1770 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1771 sc->sc_name)); 1772 1773 g_error_provider(sc->sc_provider, ENXIO); 1774 mtx_lock(&sc->sc_queue_mtx); 1775 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1776 bioq_remove(&sc->sc_queue, bp); 1777 g_io_deliver(bp, ENXIO); 1778 } 1779 mtx_unlock(&sc->sc_queue_mtx); 1780 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1781 sc->sc_provider->name); 1782 sc->sc_provider->flags |= G_PF_WITHER; 1783 g_orphan_provider(sc->sc_provider, ENXIO); 1784 sc->sc_provider = NULL; 1785 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1786 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1787 g_mirror_sync_stop(disk, 1); 1788 } 1789} 1790 1791static void 1792g_mirror_go(void *arg) 1793{ 1794 struct g_mirror_softc *sc; 1795 1796 sc = arg; 1797 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1798 g_mirror_event_send(sc, 0, 1799 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1800} 1801 1802static u_int 1803g_mirror_determine_state(struct g_mirror_disk *disk) 1804{ 1805 struct g_mirror_softc *sc; 1806 u_int state; 1807 1808 sc = disk->d_softc; 1809 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1810 if ((disk->d_flags & 1811 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1812 /* Disk does not need synchronization. */ 1813 state = G_MIRROR_DISK_STATE_ACTIVE; 1814 } else { 1815 if ((sc->sc_flags & 1816 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1817 (disk->d_flags & 1818 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1819 /* 1820 * We can start synchronization from 1821 * the stored offset. 1822 */ 1823 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1824 } else { 1825 state = G_MIRROR_DISK_STATE_STALE; 1826 } 1827 } 1828 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1829 /* 1830 * Reset all synchronization data for this disk, 1831 * because if it even was synchronized, it was 1832 * synchronized to disks with different syncid. 1833 */ 1834 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1835 disk->d_sync.ds_offset = 0; 1836 disk->d_sync.ds_offset_done = 0; 1837 disk->d_sync.ds_syncid = sc->sc_syncid; 1838 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1839 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1840 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1841 } else { 1842 state = G_MIRROR_DISK_STATE_STALE; 1843 } 1844 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1845 /* 1846 * Not good, NOT GOOD! 1847 * It means that mirror was started on stale disks 1848 * and more fresh disk just arrive. 1849 * If there were writes, mirror is fucked up, sorry. 1850 * I think the best choice here is don't touch 1851 * this disk and inform the user laudly. 1852 */ 1853 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1854 "disk (%s) arrives!! It will not be connected to the " 1855 "running device.", sc->sc_name, 1856 g_mirror_get_diskname(disk)); 1857 g_mirror_destroy_disk(disk); 1858 state = G_MIRROR_DISK_STATE_NONE; 1859 /* Return immediately, because disk was destroyed. */ 1860 return (state); 1861 } 1862 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1863 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1864 return (state); 1865} 1866 1867/* 1868 * Update device state. 1869 */ 1870static void 1871g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1872{ 1873 struct g_mirror_disk *disk; 1874 u_int state; 1875 1876 g_topology_assert(); 1877 1878 switch (sc->sc_state) { 1879 case G_MIRROR_DEVICE_STATE_STARTING: 1880 { 1881 struct g_mirror_disk *pdisk, *tdisk; 1882 u_int dirty, ndisks, genid, syncid; 1883 1884 KASSERT(sc->sc_provider == NULL, 1885 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1886 /* 1887 * Are we ready? We are, if all disks are connected or 1888 * if we have any disks and 'force' is true. 1889 */ 1890 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1891 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1892 ; 1893 } else if (g_mirror_ndisks(sc, -1) == 0) { 1894 /* 1895 * Disks went down in starting phase, so destroy 1896 * device. 1897 */ 1898 callout_drain(&sc->sc_callout); 1899 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1900 return; 1901 } else { 1902 return; 1903 } 1904 1905 /* 1906 * Activate all disks with the biggest syncid. 1907 */ 1908 if (force) { 1909 /* 1910 * If 'force' is true, we have been called due to 1911 * timeout, so don't bother canceling timeout. 1912 */ 1913 ndisks = 0; 1914 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1915 if ((disk->d_flags & 1916 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1917 ndisks++; 1918 } 1919 } 1920 if (ndisks == 0) { 1921 /* No valid disks found, destroy device. */ 1922 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1923 return; 1924 } 1925 } else { 1926 /* Cancel timeout. */ 1927 callout_drain(&sc->sc_callout); 1928 } 1929 1930 /* 1931 * Find the biggest genid. 1932 */ 1933 genid = 0; 1934 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1935 if (disk->d_genid > genid) 1936 genid = disk->d_genid; 1937 } 1938 sc->sc_genid = genid; 1939 /* 1940 * Remove all disks without the biggest genid. 1941 */ 1942 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 1943 if (disk->d_genid < genid) { 1944 G_MIRROR_DEBUG(0, 1945 "Component %s (device %s) broken, skipping.", 1946 g_mirror_get_diskname(disk), sc->sc_name); 1947 g_mirror_destroy_disk(disk); 1948 } 1949 } 1950 1951 /* 1952 * Find the biggest syncid. 1953 */ 1954 syncid = 0; 1955 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1956 if (disk->d_sync.ds_syncid > syncid) 1957 syncid = disk->d_sync.ds_syncid; 1958 } 1959 1960 /* 1961 * Here we need to look for dirty disks and if all disks 1962 * with the biggest syncid are dirty, we have to choose 1963 * one with the biggest priority and rebuild the rest. 1964 */ 1965 /* 1966 * Find the number of dirty disks with the biggest syncid. 1967 * Find the number of disks with the biggest syncid. 1968 * While here, find a disk with the biggest priority. 1969 */ 1970 dirty = ndisks = 0; 1971 pdisk = NULL; 1972 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1973 if (disk->d_sync.ds_syncid != syncid) 1974 continue; 1975 if ((disk->d_flags & 1976 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1977 continue; 1978 } 1979 ndisks++; 1980 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1981 dirty++; 1982 if (pdisk == NULL || 1983 pdisk->d_priority < disk->d_priority) { 1984 pdisk = disk; 1985 } 1986 } 1987 } 1988 if (dirty == 0) { 1989 /* No dirty disks at all, great. */ 1990 } else if (dirty == ndisks) { 1991 /* 1992 * Force synchronization for all dirty disks except one 1993 * with the biggest priority. 1994 */ 1995 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1996 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1997 "master disk for synchronization.", 1998 g_mirror_get_diskname(pdisk), sc->sc_name); 1999 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2000 if (disk->d_sync.ds_syncid != syncid) 2001 continue; 2002 if ((disk->d_flags & 2003 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2004 continue; 2005 } 2006 KASSERT((disk->d_flags & 2007 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2008 ("Disk %s isn't marked as dirty.", 2009 g_mirror_get_diskname(disk))); 2010 /* Skip the disk with the biggest priority. */ 2011 if (disk == pdisk) 2012 continue; 2013 disk->d_sync.ds_syncid = 0; 2014 } 2015 } else if (dirty < ndisks) { 2016 /* 2017 * Force synchronization for all dirty disks. 2018 * We have some non-dirty disks. 2019 */ 2020 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2021 if (disk->d_sync.ds_syncid != syncid) 2022 continue; 2023 if ((disk->d_flags & 2024 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2025 continue; 2026 } 2027 if ((disk->d_flags & 2028 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2029 continue; 2030 } 2031 disk->d_sync.ds_syncid = 0; 2032 } 2033 } 2034 2035 /* Reset hint. */ 2036 sc->sc_hint = NULL; 2037 sc->sc_syncid = syncid; 2038 if (force) { 2039 /* Remember to bump syncid on first write. */ 2040 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_OFW; 2041 } 2042 state = G_MIRROR_DEVICE_STATE_RUNNING; 2043 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2044 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2045 g_mirror_device_state2str(state)); 2046 sc->sc_state = state; 2047 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2048 state = g_mirror_determine_state(disk); 2049 g_mirror_event_send(disk, state, 2050 G_MIRROR_EVENT_DONTWAIT); 2051 if (state == G_MIRROR_DISK_STATE_STALE) 2052 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_OFW; 2053 } 2054 wakeup(&g_mirror_class); 2055 break; 2056 } 2057 case G_MIRROR_DEVICE_STATE_RUNNING: 2058 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2059 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2060 /* 2061 * No active disks or no disks at all, 2062 * so destroy device. 2063 */ 2064 if (sc->sc_provider != NULL) 2065 g_mirror_destroy_provider(sc); 2066 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2067 break; 2068 } else if (g_mirror_ndisks(sc, 2069 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2070 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2071 /* 2072 * We have active disks, launch provider if it doesn't 2073 * exist. 2074 */ 2075 if (sc->sc_provider == NULL) 2076 g_mirror_launch_provider(sc); 2077 } 2078 /* 2079 * Bump syncid here, if we need to do it immediately. 2080 */ 2081 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_IMM) != 0) { 2082 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2083 g_mirror_bump_syncid(sc); 2084 } 2085 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID_IMM) != 0) { 2086 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2087 g_mirror_bump_genid(sc); 2088 } 2089 break; 2090 default: 2091 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2092 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2093 break; 2094 } 2095} 2096 2097/* 2098 * Update disk state and device state if needed. 2099 */ 2100#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2101 "Disk %s state changed from %s to %s (device %s).", \ 2102 g_mirror_get_diskname(disk), \ 2103 g_mirror_disk_state2str(disk->d_state), \ 2104 g_mirror_disk_state2str(state), sc->sc_name) 2105static int 2106g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2107{ 2108 struct g_mirror_softc *sc; 2109 2110 g_topology_assert(); 2111 2112 sc = disk->d_softc; 2113again: 2114 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2115 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2116 g_mirror_disk_state2str(state)); 2117 switch (state) { 2118 case G_MIRROR_DISK_STATE_NEW: 2119 /* 2120 * Possible scenarios: 2121 * 1. New disk arrive. 2122 */ 2123 /* Previous state should be NONE. */ 2124 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2125 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2126 g_mirror_disk_state2str(disk->d_state))); 2127 DISK_STATE_CHANGED(); 2128 2129 disk->d_state = state; 2130 if (LIST_EMPTY(&sc->sc_disks)) 2131 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2132 else { 2133 struct g_mirror_disk *dp; 2134 2135 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2136 if (disk->d_priority >= dp->d_priority) { 2137 LIST_INSERT_BEFORE(dp, disk, d_next); 2138 dp = NULL; 2139 break; 2140 } 2141 if (LIST_NEXT(dp, d_next) == NULL) 2142 break; 2143 } 2144 if (dp != NULL) 2145 LIST_INSERT_AFTER(dp, disk, d_next); 2146 } 2147 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2148 sc->sc_name, g_mirror_get_diskname(disk)); 2149 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2150 break; 2151 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 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 state = g_mirror_determine_state(disk); 2157 if (state != G_MIRROR_DISK_STATE_NONE) 2158 goto again; 2159 break; 2160 case G_MIRROR_DISK_STATE_ACTIVE: 2161 /* 2162 * Possible scenarios: 2163 * 1. New disk does not need synchronization. 2164 * 2. Synchronization process finished successfully. 2165 */ 2166 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2167 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2168 g_mirror_device_state2str(sc->sc_state), 2169 g_mirror_get_diskname(disk), 2170 g_mirror_disk_state2str(disk->d_state))); 2171 /* Previous state should be NEW or SYNCHRONIZING. */ 2172 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2173 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2174 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2175 g_mirror_disk_state2str(disk->d_state))); 2176 DISK_STATE_CHANGED(); 2177 2178 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2179 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2180 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2181 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2182 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2183 g_mirror_sync_stop(disk, 0); 2184 } 2185 disk->d_state = state; 2186 disk->d_sync.ds_offset = 0; 2187 disk->d_sync.ds_offset_done = 0; 2188 g_mirror_update_access(disk); 2189 g_mirror_update_metadata(disk); 2190 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2191 sc->sc_name, g_mirror_get_diskname(disk)); 2192 break; 2193 case G_MIRROR_DISK_STATE_STALE: 2194 /* 2195 * Possible scenarios: 2196 * 1. Stale disk was connected. 2197 */ 2198 /* Previous state should be NEW. */ 2199 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2200 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2201 g_mirror_disk_state2str(disk->d_state))); 2202 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2203 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2204 g_mirror_device_state2str(sc->sc_state), 2205 g_mirror_get_diskname(disk), 2206 g_mirror_disk_state2str(disk->d_state))); 2207 /* 2208 * STALE state is only possible if device is marked 2209 * NOAUTOSYNC. 2210 */ 2211 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2212 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2213 g_mirror_device_state2str(sc->sc_state), 2214 g_mirror_get_diskname(disk), 2215 g_mirror_disk_state2str(disk->d_state))); 2216 DISK_STATE_CHANGED(); 2217 2218 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2219 disk->d_state = state; 2220 g_mirror_update_metadata(disk); 2221 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2222 sc->sc_name, g_mirror_get_diskname(disk)); 2223 break; 2224 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2225 /* 2226 * Possible scenarios: 2227 * 1. Disk which needs synchronization was connected. 2228 */ 2229 /* Previous state should be NEW. */ 2230 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2231 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2232 g_mirror_disk_state2str(disk->d_state))); 2233 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2234 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2235 g_mirror_device_state2str(sc->sc_state), 2236 g_mirror_get_diskname(disk), 2237 g_mirror_disk_state2str(disk->d_state))); 2238 DISK_STATE_CHANGED(); 2239 2240 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2241 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2242 disk->d_state = state; 2243 if (sc->sc_provider != NULL) { 2244 g_mirror_sync_start(disk); 2245 g_mirror_update_metadata(disk); 2246 } 2247 break; 2248 case G_MIRROR_DISK_STATE_DISCONNECTED: 2249 /* 2250 * Possible scenarios: 2251 * 1. Device wasn't running yet, but disk disappear. 2252 * 2. Disk was active and disapppear. 2253 * 3. Disk disappear during synchronization process. 2254 */ 2255 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2256 /* 2257 * Previous state should be ACTIVE, STALE or 2258 * SYNCHRONIZING. 2259 */ 2260 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2261 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2262 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2263 ("Wrong disk state (%s, %s).", 2264 g_mirror_get_diskname(disk), 2265 g_mirror_disk_state2str(disk->d_state))); 2266 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2267 /* Previous state should be NEW. */ 2268 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2269 ("Wrong disk state (%s, %s).", 2270 g_mirror_get_diskname(disk), 2271 g_mirror_disk_state2str(disk->d_state))); 2272 /* 2273 * Reset bumping syncid if disk disappeared in STARTING 2274 * state. 2275 */ 2276 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_OFW) != 0) 2277 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2278#ifdef INVARIANTS 2279 } else { 2280 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2281 sc->sc_name, 2282 g_mirror_device_state2str(sc->sc_state), 2283 g_mirror_get_diskname(disk), 2284 g_mirror_disk_state2str(disk->d_state))); 2285#endif 2286 } 2287 DISK_STATE_CHANGED(); 2288 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2289 sc->sc_name, g_mirror_get_diskname(disk)); 2290 2291 g_mirror_destroy_disk(disk); 2292 break; 2293 case G_MIRROR_DISK_STATE_DESTROY: 2294 { 2295 int error; 2296 2297 error = g_mirror_clear_metadata(disk); 2298 if (error != 0) 2299 return (error); 2300 DISK_STATE_CHANGED(); 2301 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2302 sc->sc_name, g_mirror_get_diskname(disk)); 2303 2304 g_mirror_destroy_disk(disk); 2305 sc->sc_ndisks--; 2306 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2307 g_mirror_update_metadata(disk); 2308 } 2309 break; 2310 } 2311 default: 2312 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2313 break; 2314 } 2315 return (0); 2316} 2317#undef DISK_STATE_CHANGED 2318 2319static int 2320g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2321{ 2322 struct g_provider *pp; 2323 u_char *buf; 2324 int error; 2325 2326 g_topology_assert(); 2327 2328 error = g_access(cp, 1, 0, 0); 2329 if (error != 0) 2330 return (error); 2331 pp = cp->provider; 2332 g_topology_unlock(); 2333 /* Metadata are stored on last sector. */ 2334 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2335 &error); 2336 g_topology_lock(); 2337 g_access(cp, -1, 0, 0); 2338 if (error != 0) { 2339 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2340 cp->provider->name, error); 2341 if (buf != NULL) 2342 g_free(buf); 2343 return (error); 2344 } 2345 2346 /* Decode metadata. */ 2347 error = mirror_metadata_decode(buf, md); 2348 g_free(buf); 2349 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2350 return (EINVAL); 2351 if (md->md_version > G_MIRROR_VERSION) { 2352 G_MIRROR_DEBUG(0, 2353 "Kernel module is too old to handle metadata from %s.", 2354 cp->provider->name); 2355 return (EINVAL); 2356 } 2357 if (error != 0) { 2358 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2359 cp->provider->name); 2360 return (error); 2361 } 2362 2363 return (0); 2364} 2365 2366static int 2367g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2368 struct g_mirror_metadata *md) 2369{ 2370 2371 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2372 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2373 pp->name, md->md_did); 2374 return (EEXIST); 2375 } 2376 if (md->md_all != sc->sc_ndisks) { 2377 G_MIRROR_DEBUG(1, 2378 "Invalid '%s' field on disk %s (device %s), skipping.", 2379 "md_all", pp->name, sc->sc_name); 2380 return (EINVAL); 2381 } 2382 if (md->md_slice != sc->sc_slice) { 2383 G_MIRROR_DEBUG(1, 2384 "Invalid '%s' field on disk %s (device %s), skipping.", 2385 "md_slice", pp->name, sc->sc_name); 2386 return (EINVAL); 2387 } 2388 if (md->md_balance != sc->sc_balance) { 2389 G_MIRROR_DEBUG(1, 2390 "Invalid '%s' field on disk %s (device %s), skipping.", 2391 "md_balance", pp->name, sc->sc_name); 2392 return (EINVAL); 2393 } 2394 if (md->md_mediasize != sc->sc_mediasize) { 2395 G_MIRROR_DEBUG(1, 2396 "Invalid '%s' field on disk %s (device %s), skipping.", 2397 "md_mediasize", pp->name, sc->sc_name); 2398 return (EINVAL); 2399 } 2400 if (sc->sc_mediasize > pp->mediasize) { 2401 G_MIRROR_DEBUG(1, 2402 "Invalid size of disk %s (device %s), skipping.", pp->name, 2403 sc->sc_name); 2404 return (EINVAL); 2405 } 2406 if (md->md_sectorsize != sc->sc_sectorsize) { 2407 G_MIRROR_DEBUG(1, 2408 "Invalid '%s' field on disk %s (device %s), skipping.", 2409 "md_sectorsize", pp->name, sc->sc_name); 2410 return (EINVAL); 2411 } 2412 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2413 G_MIRROR_DEBUG(1, 2414 "Invalid sector size of disk %s (device %s), skipping.", 2415 pp->name, sc->sc_name); 2416 return (EINVAL); 2417 } 2418 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2419 G_MIRROR_DEBUG(1, 2420 "Invalid device flags on disk %s (device %s), skipping.", 2421 pp->name, sc->sc_name); 2422 return (EINVAL); 2423 } 2424 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2425 G_MIRROR_DEBUG(1, 2426 "Invalid disk flags on disk %s (device %s), skipping.", 2427 pp->name, sc->sc_name); 2428 return (EINVAL); 2429 } 2430 return (0); 2431} 2432 2433static int 2434g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2435 struct g_mirror_metadata *md) 2436{ 2437 struct g_mirror_disk *disk; 2438 int error; 2439 2440 g_topology_assert(); 2441 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2442 2443 error = g_mirror_check_metadata(sc, pp, md); 2444 if (error != 0) 2445 return (error); 2446 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2447 md->md_genid < sc->sc_genid) { 2448 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2449 pp->name, sc->sc_name); 2450 return (EINVAL); 2451 } 2452 disk = g_mirror_init_disk(sc, pp, md, &error); 2453 if (disk == NULL) 2454 return (error); 2455 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2456 G_MIRROR_EVENT_WAIT); 2457 if (error != 0) 2458 return (error); 2459 if (md->md_version < G_MIRROR_VERSION) { 2460 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2461 pp->name, md->md_version, G_MIRROR_VERSION); 2462 g_mirror_update_metadata(disk); 2463 } 2464 return (0); 2465} 2466 2467static int 2468g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2469{ 2470 struct g_mirror_softc *sc; 2471 struct g_mirror_disk *disk; 2472 int dcr, dcw, dce; 2473 2474 g_topology_assert(); 2475 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2476 acw, ace); 2477 2478 dcr = pp->acr + acr; 2479 dcw = pp->acw + acw; 2480 dce = pp->ace + ace; 2481 2482 sc = pp->geom->softc; 2483 if (sc == NULL || LIST_EMPTY(&sc->sc_disks) || 2484 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 2485 if (acr <= 0 && acw <= 0 && ace <= 0) 2486 return (0); 2487 else 2488 return (ENXIO); 2489 } 2490 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2491 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2492 continue; 2493 /* 2494 * Mark disk as dirty on open and unmark on close. 2495 */ 2496 if (pp->acw == 0 && dcw > 0) { 2497 G_MIRROR_DEBUG(1, 2498 "Disk %s (device %s) marked as dirty.", 2499 g_mirror_get_diskname(disk), sc->sc_name); 2500 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2501 g_mirror_update_metadata(disk); 2502 } else if (pp->acw > 0 && dcw == 0) { 2503 G_MIRROR_DEBUG(1, 2504 "Disk %s (device %s) marked as clean.", 2505 g_mirror_get_diskname(disk), sc->sc_name); 2506 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2507 g_mirror_update_metadata(disk); 2508 } 2509 } 2510 return (0); 2511} 2512 2513static struct g_geom * 2514g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2515{ 2516 struct g_mirror_softc *sc; 2517 struct g_geom *gp; 2518 int error, timeout; 2519 2520 g_topology_assert(); 2521 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2522 md->md_mid); 2523 2524 /* One disk is minimum. */ 2525 if (md->md_all < 1) 2526 return (NULL); 2527 /* 2528 * Action geom. 2529 */ 2530 gp = g_new_geomf(mp, "%s", md->md_name); 2531 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2532 gp->start = g_mirror_start; 2533 gp->spoiled = g_mirror_spoiled; 2534 gp->orphan = g_mirror_orphan; 2535 gp->access = g_mirror_access; 2536 gp->dumpconf = g_mirror_dumpconf; 2537 2538 sc->sc_id = md->md_mid; 2539 sc->sc_slice = md->md_slice; 2540 sc->sc_balance = md->md_balance; 2541 sc->sc_mediasize = md->md_mediasize; 2542 sc->sc_sectorsize = md->md_sectorsize; 2543 sc->sc_ndisks = md->md_all; 2544 sc->sc_flags = md->md_mflags; 2545 sc->sc_bump_id = 0; 2546 sc->sc_idle = 0; 2547 bioq_init(&sc->sc_queue); 2548 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2549 LIST_INIT(&sc->sc_disks); 2550 TAILQ_INIT(&sc->sc_events); 2551 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2552 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2553 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2554 gp->softc = sc; 2555 sc->sc_geom = gp; 2556 sc->sc_provider = NULL; 2557 /* 2558 * Synchronization geom. 2559 */ 2560 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2561 gp->softc = sc; 2562 gp->orphan = g_mirror_orphan; 2563 sc->sc_sync.ds_geom = gp; 2564 sc->sc_sync.ds_ndisks = 0; 2565 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2566 "g_mirror %s", md->md_name); 2567 if (error != 0) { 2568 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2569 sc->sc_name); 2570 g_destroy_geom(sc->sc_sync.ds_geom); 2571 mtx_destroy(&sc->sc_events_mtx); 2572 mtx_destroy(&sc->sc_queue_mtx); 2573 g_destroy_geom(sc->sc_geom); 2574 free(sc, M_MIRROR); 2575 return (NULL); 2576 } 2577 2578 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2579 2580 /* 2581 * Run timeout. 2582 */ 2583 timeout = g_mirror_timeout * hz; 2584 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2585 return (sc->sc_geom); 2586} 2587 2588int 2589g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2590{ 2591 struct g_provider *pp; 2592 2593 g_topology_assert(); 2594 2595 if (sc == NULL) 2596 return (ENXIO); 2597 pp = sc->sc_provider; 2598 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2599 if (force) { 2600 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 2601 "can't be definitely removed.", pp->name); 2602 } else { 2603 G_MIRROR_DEBUG(1, 2604 "Device %s is still open (r%dw%de%d).", pp->name, 2605 pp->acr, pp->acw, pp->ace); 2606 return (EBUSY); 2607 } 2608 } 2609 2610 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2611 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2612 g_topology_unlock(); 2613 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2614 mtx_lock(&sc->sc_queue_mtx); 2615 wakeup(sc); 2616 mtx_unlock(&sc->sc_queue_mtx); 2617 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2618 while (sc->sc_worker != NULL) 2619 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2620 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2621 g_topology_lock(); 2622 g_mirror_destroy_device(sc); 2623 free(sc, M_MIRROR); 2624 return (0); 2625} 2626 2627static void 2628g_mirror_taste_orphan(struct g_consumer *cp) 2629{ 2630 2631 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2632 cp->provider->name)); 2633} 2634 2635static struct g_geom * 2636g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2637{ 2638 struct g_mirror_metadata md; 2639 struct g_mirror_softc *sc; 2640 struct g_consumer *cp; 2641 struct g_geom *gp; 2642 int error; 2643 2644 g_topology_assert(); 2645 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2646 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2647 2648 gp = g_new_geomf(mp, "mirror:taste"); 2649 /* 2650 * This orphan function should be never called. 2651 */ 2652 gp->orphan = g_mirror_taste_orphan; 2653 cp = g_new_consumer(gp); 2654 g_attach(cp, pp); 2655 error = g_mirror_read_metadata(cp, &md); 2656 g_detach(cp); 2657 g_destroy_consumer(cp); 2658 g_destroy_geom(gp); 2659 if (error != 0) 2660 return (NULL); 2661 gp = NULL; 2662 2663 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2664 return (NULL); 2665 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2666 G_MIRROR_DEBUG(0, 2667 "Device %s: provider %s marked as inactive, skipping.", 2668 md.md_name, pp->name); 2669 return (NULL); 2670 } 2671 if (g_mirror_debug >= 2) 2672 mirror_metadata_dump(&md); 2673 2674 /* 2675 * Let's check if device already exists. 2676 */ 2677 sc = NULL; 2678 LIST_FOREACH(gp, &mp->geom, geom) { 2679 sc = gp->softc; 2680 if (sc == NULL) 2681 continue; 2682 if (sc->sc_sync.ds_geom == gp) 2683 continue; 2684 if (strcmp(md.md_name, sc->sc_name) != 0) 2685 continue; 2686 if (md.md_mid != sc->sc_id) { 2687 G_MIRROR_DEBUG(0, "Device %s already configured.", 2688 sc->sc_name); 2689 return (NULL); 2690 } 2691 break; 2692 } 2693 if (gp == NULL) { 2694 gp = g_mirror_create(mp, &md); 2695 if (gp == NULL) { 2696 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2697 md.md_name); 2698 return (NULL); 2699 } 2700 sc = gp->softc; 2701 } 2702 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2703 error = g_mirror_add_disk(sc, pp, &md); 2704 if (error != 0) { 2705 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2706 pp->name, gp->name, error); 2707 if (LIST_EMPTY(&sc->sc_disks)) 2708 g_mirror_destroy(sc, 1); 2709 return (NULL); 2710 } 2711 return (gp); 2712} 2713 2714static int 2715g_mirror_destroy_geom(struct gctl_req *req __unused, 2716 struct g_class *mp __unused, struct g_geom *gp) 2717{ 2718 2719 return (g_mirror_destroy(gp->softc, 0)); 2720} 2721 2722static void 2723g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2724 struct g_consumer *cp, struct g_provider *pp) 2725{ 2726 struct g_mirror_softc *sc; 2727 2728 g_topology_assert(); 2729 2730 sc = gp->softc; 2731 if (sc == NULL) 2732 return; 2733 /* Skip synchronization geom. */ 2734 if (gp == sc->sc_sync.ds_geom) 2735 return; 2736 if (pp != NULL) { 2737 /* Nothing here. */ 2738 } else if (cp != NULL) { 2739 struct g_mirror_disk *disk; 2740 2741 disk = cp->private; 2742 if (disk == NULL) 2743 return; 2744 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2745 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2746 sbuf_printf(sb, "%s<Synchronized>", indent); 2747 if (disk->d_sync.ds_offset_done == 0) 2748 sbuf_printf(sb, "0%%"); 2749 else { 2750 sbuf_printf(sb, "%u%%", 2751 (u_int)((disk->d_sync.ds_offset_done * 100) / 2752 sc->sc_provider->mediasize)); 2753 } 2754 sbuf_printf(sb, "</Synchronized>\n"); 2755 } 2756 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2757 disk->d_sync.ds_syncid); 2758 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 2759 disk->d_genid); 2760 sbuf_printf(sb, "%s<Flags>", indent); 2761 if (disk->d_flags == 0) 2762 sbuf_printf(sb, "NONE"); 2763 else { 2764 int first = 1; 2765 2766#define ADD_FLAG(flag, name) do { \ 2767 if ((disk->d_flags & (flag)) != 0) { \ 2768 if (!first) \ 2769 sbuf_printf(sb, ", "); \ 2770 else \ 2771 first = 0; \ 2772 sbuf_printf(sb, name); \ 2773 } \ 2774} while (0) 2775 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2776 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2777 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2778 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2779 "SYNCHRONIZING"); 2780 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2781#undef ADD_FLAG 2782 } 2783 sbuf_printf(sb, "</Flags>\n"); 2784 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2785 disk->d_priority); 2786 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2787 g_mirror_disk_state2str(disk->d_state)); 2788 } else { 2789 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2790 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2791 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 2792 sbuf_printf(sb, "%s<Flags>", indent); 2793 if (sc->sc_flags == 0) 2794 sbuf_printf(sb, "NONE"); 2795 else { 2796 int first = 1; 2797 2798#define ADD_FLAG(flag, name) do { \ 2799 if ((sc->sc_flags & (flag)) != 0) { \ 2800 if (!first) \ 2801 sbuf_printf(sb, ", "); \ 2802 else \ 2803 first = 0; \ 2804 sbuf_printf(sb, name); \ 2805 } \ 2806} while (0) 2807 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2808#undef ADD_FLAG 2809 } 2810 sbuf_printf(sb, "</Flags>\n"); 2811 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2812 (u_int)sc->sc_slice); 2813 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2814 balance_name(sc->sc_balance)); 2815 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2816 sc->sc_ndisks); 2817 sbuf_printf(sb, "%s<State>", indent); 2818 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2819 sbuf_printf(sb, "%s", "STARTING"); 2820 else if (sc->sc_ndisks == 2821 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 2822 sbuf_printf(sb, "%s", "COMPLETE"); 2823 else 2824 sbuf_printf(sb, "%s", "DEGRADED"); 2825 sbuf_printf(sb, "</State>\n"); 2826 } 2827} 2828 2829static void 2830g_mirror_shutdown(void *arg, int howto) 2831{ 2832 struct g_class *mp; 2833 struct g_geom *gp, *gp2; 2834 2835 mp = arg; 2836 DROP_GIANT(); 2837 g_topology_lock(); 2838 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2839 if (gp->softc == NULL) 2840 continue; 2841 g_mirror_destroy(gp->softc, 1); 2842 } 2843 g_topology_unlock(); 2844 PICKUP_GIANT(); 2845#if 0 2846 tsleep(&gp, PRIBIO, "m:shutdown", hz * 20); 2847#endif 2848} 2849 2850static void 2851g_mirror_init(struct g_class *mp) 2852{ 2853 2854 g_mirror_ehtag = EVENTHANDLER_REGISTER(shutdown_post_sync, 2855 g_mirror_shutdown, mp, SHUTDOWN_PRI_FIRST); 2856 if (g_mirror_ehtag == NULL) 2857 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 2858} 2859 2860static void 2861g_mirror_fini(struct g_class *mp) 2862{ 2863 2864 if (g_mirror_ehtag == NULL) 2865 return; 2866 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_ehtag); 2867} 2868 2869static int 2870g_mirror_can_go(void) 2871{ 2872 struct g_mirror_softc *sc; 2873 struct g_geom *gp; 2874 struct g_provider *pp; 2875 int can_go; 2876 2877 DROP_GIANT(); 2878 can_go = 1; 2879 g_topology_lock(); 2880 LIST_FOREACH(gp, &g_mirror_class.geom, geom) { 2881 sc = gp->softc; 2882 if (sc == NULL) { 2883 can_go = 0; 2884 break; 2885 } 2886 pp = sc->sc_provider; 2887 if (pp == NULL || pp->error != 0) { 2888 can_go = 0; 2889 break; 2890 } 2891 } 2892 g_topology_unlock(); 2893 PICKUP_GIANT(); 2894 return (can_go); 2895} 2896 2897static void 2898g_mirror_rootwait(void) 2899{ 2900 2901 /* 2902 * HACK: Wait for GEOM, because g_mirror_rootwait() can be called, 2903 * HACK: before we get providers for tasting. 2904 */ 2905 tsleep(&g_mirror_class, PRIBIO, "mroot", hz * 3); 2906 /* 2907 * Wait for mirrors in degraded state. 2908 */ 2909 for (;;) { 2910 if (g_mirror_can_go()) 2911 break; 2912 tsleep(&g_mirror_class, PRIBIO, "mroot", hz); 2913 } 2914} 2915 2916SYSINIT(g_mirror_root, SI_SUB_RAID, SI_ORDER_FIRST, g_mirror_rootwait, NULL) 2917 2918DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2919