g_mirror.c revision 139246
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 139246 2004-12-23 21:15:15Z 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 disk->d_genid = sc->sc_genid; 761 g_mirror_update_metadata(disk); 762 } 763 } 764} 765 766static void 767g_mirror_idle(struct g_mirror_softc *sc) 768{ 769 struct g_mirror_disk *disk; 770 771 if (sc->sc_provider == NULL || sc->sc_provider->acw == 0) 772 return; 773 sc->sc_idle = 1; 774 g_topology_lock(); 775 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 776 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 777 continue; 778 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 779 g_mirror_get_diskname(disk), sc->sc_name); 780 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 781 g_mirror_update_metadata(disk); 782 } 783 g_topology_unlock(); 784} 785 786static void 787g_mirror_unidle(struct g_mirror_softc *sc) 788{ 789 struct g_mirror_disk *disk; 790 791 sc->sc_idle = 0; 792 g_topology_lock(); 793 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 794 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 795 continue; 796 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 797 g_mirror_get_diskname(disk), sc->sc_name); 798 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 799 g_mirror_update_metadata(disk); 800 } 801 g_topology_unlock(); 802} 803 804/* 805 * Return 1 if we should check if mirror is idling. 806 */ 807static int 808g_mirror_check_idle(struct g_mirror_softc *sc) 809{ 810 struct g_mirror_disk *disk; 811 812 if (sc->sc_idle) 813 return (0); 814 if (sc->sc_provider != NULL && sc->sc_provider->acw == 0) 815 return (0); 816 /* 817 * Check if there are no in-flight requests. 818 */ 819 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 820 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 821 continue; 822 if (disk->d_consumer->index > 0) 823 return (0); 824 } 825 return (1); 826} 827 828static __inline int 829bintime_cmp(struct bintime *bt1, struct bintime *bt2) 830{ 831 832 if (bt1->sec < bt2->sec) 833 return (-1); 834 else if (bt1->sec > bt2->sec) 835 return (1); 836 if (bt1->frac < bt2->frac) 837 return (-1); 838 else if (bt1->frac > bt2->frac) 839 return (1); 840 return (0); 841} 842 843static void 844g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 845{ 846 847 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 848 return; 849 binuptime(&disk->d_delay); 850 bintime_sub(&disk->d_delay, &bp->bio_t0); 851} 852 853static void 854g_mirror_done(struct bio *bp) 855{ 856 struct g_mirror_softc *sc; 857 858 sc = bp->bio_from->geom->softc; 859 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 860 mtx_lock(&sc->sc_queue_mtx); 861 bioq_disksort(&sc->sc_queue, bp); 862 wakeup(sc); 863 mtx_unlock(&sc->sc_queue_mtx); 864} 865 866static void 867g_mirror_regular_request(struct bio *bp) 868{ 869 struct g_mirror_softc *sc; 870 struct g_mirror_disk *disk; 871 struct bio *pbp; 872 873 g_topology_assert_not(); 874 875 bp->bio_from->index--; 876 pbp = bp->bio_parent; 877 sc = pbp->bio_to->geom->softc; 878 disk = bp->bio_from->private; 879 if (disk == NULL) { 880 g_topology_lock(); 881 g_mirror_kill_consumer(sc, bp->bio_from); 882 g_topology_unlock(); 883 } else { 884 g_mirror_update_delay(disk, bp); 885 } 886 887 pbp->bio_inbed++; 888 KASSERT(pbp->bio_inbed <= pbp->bio_children, 889 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 890 pbp->bio_children)); 891 if (bp->bio_error == 0 && pbp->bio_error == 0) { 892 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 893 g_destroy_bio(bp); 894 if (pbp->bio_children == pbp->bio_inbed) { 895 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 896 pbp->bio_completed = pbp->bio_length; 897 g_io_deliver(pbp, pbp->bio_error); 898 } 899 return; 900 } else if (bp->bio_error != 0) { 901 if (pbp->bio_error == 0) 902 pbp->bio_error = bp->bio_error; 903 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 904 bp->bio_error); 905 if (disk != NULL) { 906 sc->sc_bump_id |= G_MIRROR_BUMP_GENID_IMM; 907 g_mirror_event_send(disk, 908 G_MIRROR_DISK_STATE_DISCONNECTED, 909 G_MIRROR_EVENT_DONTWAIT); 910 } 911 switch (pbp->bio_cmd) { 912 case BIO_DELETE: 913 case BIO_WRITE: 914 pbp->bio_inbed--; 915 pbp->bio_children--; 916 break; 917 } 918 } 919 g_destroy_bio(bp); 920 921 switch (pbp->bio_cmd) { 922 case BIO_READ: 923 if (pbp->bio_children == pbp->bio_inbed) { 924 pbp->bio_error = 0; 925 mtx_lock(&sc->sc_queue_mtx); 926 bioq_disksort(&sc->sc_queue, pbp); 927 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 928 wakeup(sc); 929 mtx_unlock(&sc->sc_queue_mtx); 930 } 931 break; 932 case BIO_DELETE: 933 case BIO_WRITE: 934 if (pbp->bio_children == 0) { 935 /* 936 * All requests failed. 937 */ 938 } else if (pbp->bio_inbed < pbp->bio_children) { 939 /* Do nothing. */ 940 break; 941 } else if (pbp->bio_children == pbp->bio_inbed) { 942 /* Some requests succeeded. */ 943 pbp->bio_error = 0; 944 pbp->bio_completed = pbp->bio_length; 945 } 946 g_io_deliver(pbp, pbp->bio_error); 947 break; 948 default: 949 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 950 break; 951 } 952} 953 954static void 955g_mirror_sync_done(struct bio *bp) 956{ 957 struct g_mirror_softc *sc; 958 959 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 960 sc = bp->bio_from->geom->softc; 961 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 962 mtx_lock(&sc->sc_queue_mtx); 963 bioq_disksort(&sc->sc_queue, bp); 964 wakeup(sc); 965 mtx_unlock(&sc->sc_queue_mtx); 966} 967 968static void 969g_mirror_start(struct bio *bp) 970{ 971 struct g_mirror_softc *sc; 972 973 sc = bp->bio_to->geom->softc; 974 /* 975 * If sc == NULL or there are no valid disks, provider's error 976 * should be set and g_mirror_start() should not be called at all. 977 */ 978 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 979 ("Provider's error should be set (error=%d)(mirror=%s).", 980 bp->bio_to->error, bp->bio_to->name)); 981 G_MIRROR_LOGREQ(3, bp, "Request received."); 982 983 switch (bp->bio_cmd) { 984 case BIO_READ: 985 case BIO_WRITE: 986 case BIO_DELETE: 987 break; 988 case BIO_GETATTR: 989 default: 990 g_io_deliver(bp, EOPNOTSUPP); 991 return; 992 } 993 mtx_lock(&sc->sc_queue_mtx); 994 bioq_disksort(&sc->sc_queue, bp); 995 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 996 wakeup(sc); 997 mtx_unlock(&sc->sc_queue_mtx); 998} 999 1000/* 1001 * Send one synchronization request. 1002 */ 1003static void 1004g_mirror_sync_one(struct g_mirror_disk *disk) 1005{ 1006 struct g_mirror_softc *sc; 1007 struct bio *bp; 1008 1009 sc = disk->d_softc; 1010 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1011 ("Disk %s is not marked for synchronization.", 1012 g_mirror_get_diskname(disk))); 1013 1014 bp = g_new_bio(); 1015 if (bp == NULL) 1016 return; 1017 bp->bio_parent = NULL; 1018 bp->bio_cmd = BIO_READ; 1019 bp->bio_offset = disk->d_sync.ds_offset; 1020 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1021 bp->bio_cflags = 0; 1022 bp->bio_done = g_mirror_sync_done; 1023 bp->bio_data = disk->d_sync.ds_data; 1024 if (bp->bio_data == NULL) { 1025 g_destroy_bio(bp); 1026 return; 1027 } 1028 disk->d_sync.ds_offset += bp->bio_length; 1029 bp->bio_to = sc->sc_provider; 1030 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1031 disk->d_sync.ds_consumer->index++; 1032 g_io_request(bp, disk->d_sync.ds_consumer); 1033} 1034 1035static void 1036g_mirror_sync_request(struct bio *bp) 1037{ 1038 struct g_mirror_softc *sc; 1039 struct g_mirror_disk *disk; 1040 1041 bp->bio_from->index--; 1042 sc = bp->bio_from->geom->softc; 1043 disk = bp->bio_from->private; 1044 if (disk == NULL) { 1045 g_topology_lock(); 1046 g_mirror_kill_consumer(sc, bp->bio_from); 1047 g_topology_unlock(); 1048 g_destroy_bio(bp); 1049 return; 1050 } 1051 1052 /* 1053 * Synchronization request. 1054 */ 1055 switch (bp->bio_cmd) { 1056 case BIO_READ: 1057 { 1058 struct g_consumer *cp; 1059 1060 if (bp->bio_error != 0) { 1061 G_MIRROR_LOGREQ(0, bp, 1062 "Synchronization request failed (error=%d).", 1063 bp->bio_error); 1064 g_destroy_bio(bp); 1065 return; 1066 } 1067 G_MIRROR_LOGREQ(3, bp, 1068 "Synchronization request half-finished."); 1069 bp->bio_cmd = BIO_WRITE; 1070 bp->bio_cflags = 0; 1071 cp = disk->d_consumer; 1072 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1073 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1074 cp->acr, cp->acw, cp->ace)); 1075 cp->index++; 1076 g_io_request(bp, cp); 1077 return; 1078 } 1079 case BIO_WRITE: 1080 { 1081 struct g_mirror_disk_sync *sync; 1082 1083 if (bp->bio_error != 0) { 1084 G_MIRROR_LOGREQ(0, bp, 1085 "Synchronization request failed (error=%d).", 1086 bp->bio_error); 1087 g_destroy_bio(bp); 1088 sc->sc_bump_id |= G_MIRROR_BUMP_GENID_IMM; 1089 g_mirror_event_send(disk, 1090 G_MIRROR_DISK_STATE_DISCONNECTED, 1091 G_MIRROR_EVENT_DONTWAIT); 1092 return; 1093 } 1094 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1095 sync = &disk->d_sync; 1096 sync->ds_offset_done = bp->bio_offset + bp->bio_length; 1097 g_destroy_bio(bp); 1098 if (sync->ds_resync != -1) 1099 break; 1100 if (sync->ds_offset_done == sc->sc_provider->mediasize) { 1101 /* 1102 * Disk up-to-date, activate it. 1103 */ 1104 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1105 G_MIRROR_EVENT_DONTWAIT); 1106 return; 1107 } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) { 1108 /* 1109 * Update offset_done on every 100 blocks. 1110 * XXX: This should be configurable. 1111 */ 1112 g_topology_lock(); 1113 g_mirror_update_metadata(disk); 1114 g_topology_unlock(); 1115 } 1116 return; 1117 } 1118 default: 1119 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1120 bp->bio_cmd, sc->sc_name)); 1121 break; 1122 } 1123} 1124 1125static void 1126g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1127{ 1128 struct g_mirror_disk *disk; 1129 struct g_consumer *cp; 1130 struct bio *cbp; 1131 1132 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1133 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1134 break; 1135 } 1136 if (disk == NULL) { 1137 if (bp->bio_error == 0) 1138 bp->bio_error = ENXIO; 1139 g_io_deliver(bp, bp->bio_error); 1140 return; 1141 } 1142 cbp = g_clone_bio(bp); 1143 if (cbp == NULL) { 1144 if (bp->bio_error == 0) 1145 bp->bio_error = ENOMEM; 1146 g_io_deliver(bp, bp->bio_error); 1147 return; 1148 } 1149 /* 1150 * Fill in the component buf structure. 1151 */ 1152 cp = disk->d_consumer; 1153 cbp->bio_done = g_mirror_done; 1154 cbp->bio_to = cp->provider; 1155 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1156 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1157 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1158 cp->acw, cp->ace)); 1159 cp->index++; 1160 g_io_request(cbp, cp); 1161} 1162 1163static void 1164g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1165{ 1166 struct g_mirror_disk *disk; 1167 struct g_consumer *cp; 1168 struct bio *cbp; 1169 1170 disk = g_mirror_get_disk(sc); 1171 if (disk == NULL) { 1172 if (bp->bio_error == 0) 1173 bp->bio_error = ENXIO; 1174 g_io_deliver(bp, bp->bio_error); 1175 return; 1176 } 1177 cbp = g_clone_bio(bp); 1178 if (cbp == NULL) { 1179 if (bp->bio_error == 0) 1180 bp->bio_error = ENOMEM; 1181 g_io_deliver(bp, bp->bio_error); 1182 return; 1183 } 1184 /* 1185 * Fill in the component buf structure. 1186 */ 1187 cp = disk->d_consumer; 1188 cbp->bio_done = g_mirror_done; 1189 cbp->bio_to = cp->provider; 1190 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1191 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1192 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1193 cp->acw, cp->ace)); 1194 cp->index++; 1195 g_io_request(cbp, cp); 1196} 1197 1198static void 1199g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1200{ 1201 struct g_mirror_disk *disk, *dp; 1202 struct g_consumer *cp; 1203 struct bio *cbp; 1204 struct bintime curtime; 1205 1206 binuptime(&curtime); 1207 /* 1208 * Find a disk which the smallest load. 1209 */ 1210 disk = NULL; 1211 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1212 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1213 continue; 1214 /* If disk wasn't used for more than 2 sec, use it. */ 1215 if (curtime.sec - dp->d_last_used.sec >= 2) { 1216 disk = dp; 1217 break; 1218 } 1219 if (disk == NULL || 1220 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1221 disk = dp; 1222 } 1223 } 1224 cbp = g_clone_bio(bp); 1225 if (cbp == NULL) { 1226 if (bp->bio_error == 0) 1227 bp->bio_error = ENOMEM; 1228 g_io_deliver(bp, bp->bio_error); 1229 return; 1230 } 1231 /* 1232 * Fill in the component buf structure. 1233 */ 1234 cp = disk->d_consumer; 1235 cbp->bio_done = g_mirror_done; 1236 cbp->bio_to = cp->provider; 1237 binuptime(&disk->d_last_used); 1238 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1239 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1240 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1241 cp->acw, cp->ace)); 1242 cp->index++; 1243 g_io_request(cbp, cp); 1244} 1245 1246static void 1247g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1248{ 1249 struct bio_queue_head queue; 1250 struct g_mirror_disk *disk; 1251 struct g_consumer *cp; 1252 struct bio *cbp; 1253 off_t left, mod, offset, slice; 1254 u_char *data; 1255 u_int ndisks; 1256 1257 if (bp->bio_length <= sc->sc_slice) { 1258 g_mirror_request_round_robin(sc, bp); 1259 return; 1260 } 1261 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1262 slice = bp->bio_length / ndisks; 1263 mod = slice % sc->sc_provider->sectorsize; 1264 if (mod != 0) 1265 slice += sc->sc_provider->sectorsize - mod; 1266 /* 1267 * Allocate all bios before sending any request, so we can 1268 * return ENOMEM in nice and clean way. 1269 */ 1270 left = bp->bio_length; 1271 offset = bp->bio_offset; 1272 data = bp->bio_data; 1273 bioq_init(&queue); 1274 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1275 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1276 continue; 1277 cbp = g_clone_bio(bp); 1278 if (cbp == NULL) { 1279 for (cbp = bioq_first(&queue); cbp != NULL; 1280 cbp = bioq_first(&queue)) { 1281 bioq_remove(&queue, cbp); 1282 g_destroy_bio(cbp); 1283 } 1284 if (bp->bio_error == 0) 1285 bp->bio_error = ENOMEM; 1286 g_io_deliver(bp, bp->bio_error); 1287 return; 1288 } 1289 bioq_insert_tail(&queue, cbp); 1290 cbp->bio_done = g_mirror_done; 1291 cbp->bio_caller1 = disk; 1292 cbp->bio_to = disk->d_consumer->provider; 1293 cbp->bio_offset = offset; 1294 cbp->bio_data = data; 1295 cbp->bio_length = MIN(left, slice); 1296 left -= cbp->bio_length; 1297 if (left == 0) 1298 break; 1299 offset += cbp->bio_length; 1300 data += cbp->bio_length; 1301 } 1302 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1303 bioq_remove(&queue, cbp); 1304 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1305 disk = cbp->bio_caller1; 1306 cbp->bio_caller1 = NULL; 1307 cp = disk->d_consumer; 1308 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1309 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1310 cp->acr, cp->acw, cp->ace)); 1311 disk->d_consumer->index++; 1312 g_io_request(cbp, disk->d_consumer); 1313 } 1314} 1315 1316static void 1317g_mirror_register_request(struct bio *bp) 1318{ 1319 struct g_mirror_softc *sc; 1320 1321 sc = bp->bio_to->geom->softc; 1322 switch (bp->bio_cmd) { 1323 case BIO_READ: 1324 switch (sc->sc_balance) { 1325 case G_MIRROR_BALANCE_LOAD: 1326 g_mirror_request_load(sc, bp); 1327 break; 1328 case G_MIRROR_BALANCE_PREFER: 1329 g_mirror_request_prefer(sc, bp); 1330 break; 1331 case G_MIRROR_BALANCE_ROUND_ROBIN: 1332 g_mirror_request_round_robin(sc, bp); 1333 break; 1334 case G_MIRROR_BALANCE_SPLIT: 1335 g_mirror_request_split(sc, bp); 1336 break; 1337 } 1338 return; 1339 case BIO_WRITE: 1340 case BIO_DELETE: 1341 { 1342 struct g_mirror_disk *disk; 1343 struct g_mirror_disk_sync *sync; 1344 struct bio_queue_head queue; 1345 struct g_consumer *cp; 1346 struct bio *cbp; 1347 1348 if (sc->sc_idle) 1349 g_mirror_unidle(sc); 1350 /* 1351 * Allocate all bios before sending any request, so we can 1352 * return ENOMEM in nice and clean way. 1353 */ 1354 bioq_init(&queue); 1355 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1356 sync = &disk->d_sync; 1357 switch (disk->d_state) { 1358 case G_MIRROR_DISK_STATE_ACTIVE: 1359 break; 1360 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1361 if (bp->bio_offset >= sync->ds_offset) 1362 continue; 1363 else if (bp->bio_offset + bp->bio_length > 1364 sync->ds_offset_done && 1365 (bp->bio_offset < sync->ds_resync || 1366 sync->ds_resync == -1)) { 1367 sync->ds_resync = bp->bio_offset - 1368 (bp->bio_offset % MAXPHYS); 1369 } 1370 break; 1371 default: 1372 continue; 1373 } 1374 cbp = g_clone_bio(bp); 1375 if (cbp == NULL) { 1376 for (cbp = bioq_first(&queue); cbp != NULL; 1377 cbp = bioq_first(&queue)) { 1378 bioq_remove(&queue, cbp); 1379 g_destroy_bio(cbp); 1380 } 1381 if (bp->bio_error == 0) 1382 bp->bio_error = ENOMEM; 1383 g_io_deliver(bp, bp->bio_error); 1384 return; 1385 } 1386 bioq_insert_tail(&queue, cbp); 1387 cbp->bio_done = g_mirror_done; 1388 cp = disk->d_consumer; 1389 cbp->bio_caller1 = cp; 1390 cbp->bio_to = cp->provider; 1391 KASSERT(cp->acr == 1 && cp->acw == 1 && cp->ace == 1, 1392 ("Consumer %s not opened (r%dw%de%d).", 1393 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1394 } 1395 for (cbp = bioq_first(&queue); cbp != NULL; 1396 cbp = bioq_first(&queue)) { 1397 bioq_remove(&queue, cbp); 1398 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1399 cp = cbp->bio_caller1; 1400 cbp->bio_caller1 = NULL; 1401 cp->index++; 1402 g_io_request(cbp, cp); 1403 } 1404 /* 1405 * Bump syncid on first write. 1406 */ 1407 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_OFW) != 0) { 1408 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1409 g_topology_lock(); 1410 g_mirror_bump_syncid(sc); 1411 g_topology_unlock(); 1412 } 1413 return; 1414 } 1415 default: 1416 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1417 bp->bio_cmd, sc->sc_name)); 1418 break; 1419 } 1420} 1421 1422static int 1423g_mirror_can_destroy(struct g_mirror_softc *sc) 1424{ 1425 struct g_geom *gp; 1426 struct g_consumer *cp; 1427 1428 g_topology_assert(); 1429 gp = sc->sc_geom; 1430 LIST_FOREACH(cp, &gp->consumer, consumer) { 1431 if (g_mirror_is_busy(sc, cp)) 1432 return (0); 1433 } 1434 gp = sc->sc_sync.ds_geom; 1435 LIST_FOREACH(cp, &gp->consumer, consumer) { 1436 if (g_mirror_is_busy(sc, cp)) 1437 return (0); 1438 } 1439 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1440 sc->sc_name); 1441 return (1); 1442} 1443 1444static int 1445g_mirror_try_destroy(struct g_mirror_softc *sc) 1446{ 1447 1448 g_topology_lock(); 1449 if (!g_mirror_can_destroy(sc)) { 1450 g_topology_unlock(); 1451 return (0); 1452 } 1453 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1454 g_topology_unlock(); 1455 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1456 &sc->sc_worker); 1457 wakeup(&sc->sc_worker); 1458 sc->sc_worker = NULL; 1459 } else { 1460 g_mirror_destroy_device(sc); 1461 g_topology_unlock(); 1462 free(sc, M_MIRROR); 1463 } 1464 return (1); 1465} 1466 1467/* 1468 * Worker thread. 1469 */ 1470static void 1471g_mirror_worker(void *arg) 1472{ 1473 struct g_mirror_softc *sc; 1474 struct g_mirror_disk *disk; 1475 struct g_mirror_disk_sync *sync; 1476 struct g_mirror_event *ep; 1477 struct bio *bp; 1478 u_int nreqs; 1479 1480 sc = arg; 1481 curthread->td_base_pri = PRIBIO; 1482 1483 nreqs = 0; 1484 for (;;) { 1485 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1486 /* 1487 * First take a look at events. 1488 * This is important to handle events before any I/O requests. 1489 */ 1490 ep = g_mirror_event_get(sc); 1491 if (ep != NULL && g_topology_try_lock()) { 1492 g_mirror_event_remove(sc, ep); 1493 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1494 /* Update only device status. */ 1495 G_MIRROR_DEBUG(3, 1496 "Running event for device %s.", 1497 sc->sc_name); 1498 ep->e_error = 0; 1499 g_mirror_update_device(sc, 1); 1500 } else { 1501 /* Update disk status. */ 1502 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1503 g_mirror_get_diskname(ep->e_disk)); 1504 ep->e_error = g_mirror_update_disk(ep->e_disk, 1505 ep->e_state); 1506 if (ep->e_error == 0) 1507 g_mirror_update_device(sc, 0); 1508 } 1509 g_topology_unlock(); 1510 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1511 KASSERT(ep->e_error == 0, 1512 ("Error cannot be handled.")); 1513 g_mirror_event_free(ep); 1514 } else { 1515 ep->e_flags |= G_MIRROR_EVENT_DONE; 1516 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1517 ep); 1518 mtx_lock(&sc->sc_events_mtx); 1519 wakeup(ep); 1520 mtx_unlock(&sc->sc_events_mtx); 1521 } 1522 if ((sc->sc_flags & 1523 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1524 if (g_mirror_try_destroy(sc)) 1525 kthread_exit(0); 1526 } 1527 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1528 continue; 1529 } 1530 /* 1531 * Now I/O requests. 1532 */ 1533 /* Get first request from the queue. */ 1534 mtx_lock(&sc->sc_queue_mtx); 1535 bp = bioq_first(&sc->sc_queue); 1536 if (bp == NULL) { 1537 if (ep != NULL) { 1538 /* 1539 * No I/O requests and topology lock was 1540 * already held? Try again. 1541 */ 1542 mtx_unlock(&sc->sc_queue_mtx); 1543 continue; 1544 } 1545 if ((sc->sc_flags & 1546 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1547 mtx_unlock(&sc->sc_queue_mtx); 1548 if (g_mirror_try_destroy(sc)) 1549 kthread_exit(0); 1550 mtx_lock(&sc->sc_queue_mtx); 1551 } 1552 } 1553 if (sc->sc_sync.ds_ndisks > 0 && 1554 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1555 mtx_unlock(&sc->sc_queue_mtx); 1556 /* 1557 * It is time for synchronization... 1558 */ 1559 nreqs = 0; 1560 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1561 if (disk->d_state != 1562 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1563 continue; 1564 } 1565 sync = &disk->d_sync; 1566 if (sync->ds_offset >= 1567 sc->sc_provider->mediasize) { 1568 continue; 1569 } 1570 if (sync->ds_offset > sync->ds_offset_done) 1571 continue; 1572 if (sync->ds_resync != -1) { 1573 sync->ds_offset = sync->ds_resync; 1574 sync->ds_offset_done = sync->ds_resync; 1575 sync->ds_resync = -1; 1576 } 1577 g_mirror_sync_one(disk); 1578 } 1579 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1580 goto sleep; 1581 } 1582 if (bp == NULL) { 1583 if (g_mirror_check_idle(sc)) { 1584 u_int idletime; 1585 1586 idletime = g_mirror_idletime; 1587 if (idletime == 0) 1588 idletime = 1; 1589 idletime *= hz; 1590 if (msleep(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1591 "m:w1", idletime) == EWOULDBLOCK) { 1592 G_MIRROR_DEBUG(5, "%s: I'm here 3.", 1593 __func__); 1594 /* 1595 * No I/O requests in 'idletime' seconds, 1596 * so mark components as clean. 1597 */ 1598 g_mirror_idle(sc); 1599 } 1600 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1601 } else { 1602 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1603 "m:w2", 0); 1604 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1605 } 1606 continue; 1607 } 1608 nreqs++; 1609 bioq_remove(&sc->sc_queue, bp); 1610 mtx_unlock(&sc->sc_queue_mtx); 1611 1612 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1613 g_mirror_regular_request(bp); 1614 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1615 u_int timeout, sps; 1616 1617 g_mirror_sync_request(bp); 1618sleep: 1619 sps = g_mirror_syncs_per_sec; 1620 if (sps == 0) { 1621 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1622 continue; 1623 } 1624 if (ep != NULL) { 1625 /* 1626 * We have some pending events, don't sleep now. 1627 */ 1628 G_MIRROR_DEBUG(5, "%s: I'm here 7.", __func__); 1629 continue; 1630 } 1631 mtx_lock(&sc->sc_queue_mtx); 1632 if (bioq_first(&sc->sc_queue) != NULL) { 1633 mtx_unlock(&sc->sc_queue_mtx); 1634 G_MIRROR_DEBUG(5, "%s: I'm here 8.", __func__); 1635 continue; 1636 } 1637 timeout = hz / sps; 1638 if (timeout == 0) 1639 timeout = 1; 1640 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w3", 1641 timeout); 1642 } else { 1643 g_mirror_register_request(bp); 1644 } 1645 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1646 } 1647} 1648 1649/* 1650 * Open disk's consumer if needed. 1651 */ 1652static void 1653g_mirror_update_access(struct g_mirror_disk *disk) 1654{ 1655 struct g_provider *pp; 1656 1657 g_topology_assert(); 1658 1659 pp = disk->d_softc->sc_provider; 1660 if (pp == NULL) 1661 return; 1662 if (pp->acw > 0) { 1663 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1664 G_MIRROR_DEBUG(1, 1665 "Disk %s (device %s) marked as dirty.", 1666 g_mirror_get_diskname(disk), 1667 disk->d_softc->sc_name); 1668 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1669 } 1670 } else if (pp->acw == 0) { 1671 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1672 G_MIRROR_DEBUG(1, 1673 "Disk %s (device %s) marked as clean.", 1674 g_mirror_get_diskname(disk), 1675 disk->d_softc->sc_name); 1676 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1677 } 1678 } 1679} 1680 1681static void 1682g_mirror_sync_start(struct g_mirror_disk *disk) 1683{ 1684 struct g_mirror_softc *sc; 1685 int error; 1686 1687 g_topology_assert(); 1688 1689 sc = disk->d_softc; 1690 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1691 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1692 sc->sc_state)); 1693 1694 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1695 g_mirror_get_diskname(disk)); 1696 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1697 KASSERT(disk->d_sync.ds_consumer == NULL, 1698 ("Sync consumer already exists (device=%s, disk=%s).", 1699 sc->sc_name, g_mirror_get_diskname(disk))); 1700 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1701 disk->d_sync.ds_consumer->private = disk; 1702 disk->d_sync.ds_consumer->index = 0; 1703 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1704 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1705 disk->d_softc->sc_name, error)); 1706 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1707 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1708 disk->d_softc->sc_name, error)); 1709 disk->d_sync.ds_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1710 sc->sc_sync.ds_ndisks++; 1711} 1712 1713/* 1714 * Stop synchronization process. 1715 * type: 0 - synchronization finished 1716 * 1 - synchronization stopped 1717 */ 1718static void 1719g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1720{ 1721 1722 g_topology_assert(); 1723 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1724 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1725 g_mirror_disk_state2str(disk->d_state))); 1726 if (disk->d_sync.ds_consumer == NULL) 1727 return; 1728 1729 if (type == 0) { 1730 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1731 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1732 } else /* if (type == 1) */ { 1733 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1734 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1735 } 1736 g_mirror_kill_consumer(disk->d_softc, disk->d_sync.ds_consumer); 1737 free(disk->d_sync.ds_data, M_MIRROR); 1738 disk->d_sync.ds_consumer = NULL; 1739 disk->d_softc->sc_sync.ds_ndisks--; 1740 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1741} 1742 1743static void 1744g_mirror_launch_provider(struct g_mirror_softc *sc) 1745{ 1746 struct g_mirror_disk *disk; 1747 struct g_provider *pp; 1748 1749 g_topology_assert(); 1750 1751 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1752 pp->mediasize = sc->sc_mediasize; 1753 pp->sectorsize = sc->sc_sectorsize; 1754 sc->sc_provider = pp; 1755 g_error_provider(pp, 0); 1756 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1757 pp->name); 1758 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1759 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1760 g_mirror_sync_start(disk); 1761 } 1762} 1763 1764static void 1765g_mirror_destroy_provider(struct g_mirror_softc *sc) 1766{ 1767 struct g_mirror_disk *disk; 1768 struct bio *bp; 1769 1770 g_topology_assert(); 1771 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1772 sc->sc_name)); 1773 1774 g_error_provider(sc->sc_provider, ENXIO); 1775 mtx_lock(&sc->sc_queue_mtx); 1776 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1777 bioq_remove(&sc->sc_queue, bp); 1778 g_io_deliver(bp, ENXIO); 1779 } 1780 mtx_unlock(&sc->sc_queue_mtx); 1781 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1782 sc->sc_provider->name); 1783 sc->sc_provider->flags |= G_PF_WITHER; 1784 g_orphan_provider(sc->sc_provider, ENXIO); 1785 sc->sc_provider = NULL; 1786 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1787 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1788 g_mirror_sync_stop(disk, 1); 1789 } 1790} 1791 1792static void 1793g_mirror_go(void *arg) 1794{ 1795 struct g_mirror_softc *sc; 1796 1797 sc = arg; 1798 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1799 g_mirror_event_send(sc, 0, 1800 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1801} 1802 1803static u_int 1804g_mirror_determine_state(struct g_mirror_disk *disk) 1805{ 1806 struct g_mirror_softc *sc; 1807 u_int state; 1808 1809 sc = disk->d_softc; 1810 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1811 if ((disk->d_flags & 1812 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1813 /* Disk does not need synchronization. */ 1814 state = G_MIRROR_DISK_STATE_ACTIVE; 1815 } else { 1816 if ((sc->sc_flags & 1817 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1818 (disk->d_flags & 1819 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1820 /* 1821 * We can start synchronization from 1822 * the stored offset. 1823 */ 1824 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1825 } else { 1826 state = G_MIRROR_DISK_STATE_STALE; 1827 } 1828 } 1829 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1830 /* 1831 * Reset all synchronization data for this disk, 1832 * because if it even was synchronized, it was 1833 * synchronized to disks with different syncid. 1834 */ 1835 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1836 disk->d_sync.ds_offset = 0; 1837 disk->d_sync.ds_offset_done = 0; 1838 disk->d_sync.ds_syncid = sc->sc_syncid; 1839 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1840 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1841 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1842 } else { 1843 state = G_MIRROR_DISK_STATE_STALE; 1844 } 1845 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1846 /* 1847 * Not good, NOT GOOD! 1848 * It means that mirror was started on stale disks 1849 * and more fresh disk just arrive. 1850 * If there were writes, mirror is fucked up, sorry. 1851 * I think the best choice here is don't touch 1852 * this disk and inform the user laudly. 1853 */ 1854 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1855 "disk (%s) arrives!! It will not be connected to the " 1856 "running device.", sc->sc_name, 1857 g_mirror_get_diskname(disk)); 1858 g_mirror_destroy_disk(disk); 1859 state = G_MIRROR_DISK_STATE_NONE; 1860 /* Return immediately, because disk was destroyed. */ 1861 return (state); 1862 } 1863 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1864 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1865 return (state); 1866} 1867 1868/* 1869 * Update device state. 1870 */ 1871static void 1872g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1873{ 1874 struct g_mirror_disk *disk; 1875 u_int state; 1876 1877 g_topology_assert(); 1878 1879 switch (sc->sc_state) { 1880 case G_MIRROR_DEVICE_STATE_STARTING: 1881 { 1882 struct g_mirror_disk *pdisk, *tdisk; 1883 u_int dirty, ndisks, genid, syncid; 1884 1885 KASSERT(sc->sc_provider == NULL, 1886 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1887 /* 1888 * Are we ready? We are, if all disks are connected or 1889 * if we have any disks and 'force' is true. 1890 */ 1891 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1892 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1893 ; 1894 } else if (g_mirror_ndisks(sc, -1) == 0) { 1895 /* 1896 * Disks went down in starting phase, so destroy 1897 * device. 1898 */ 1899 callout_drain(&sc->sc_callout); 1900 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1901 return; 1902 } else { 1903 return; 1904 } 1905 1906 /* 1907 * Activate all disks with the biggest syncid. 1908 */ 1909 if (force) { 1910 /* 1911 * If 'force' is true, we have been called due to 1912 * timeout, so don't bother canceling timeout. 1913 */ 1914 ndisks = 0; 1915 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1916 if ((disk->d_flags & 1917 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1918 ndisks++; 1919 } 1920 } 1921 if (ndisks == 0) { 1922 /* No valid disks found, destroy device. */ 1923 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1924 return; 1925 } 1926 } else { 1927 /* Cancel timeout. */ 1928 callout_drain(&sc->sc_callout); 1929 } 1930 1931 /* 1932 * Find the biggest genid. 1933 */ 1934 genid = 0; 1935 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1936 if (disk->d_genid > genid) 1937 genid = disk->d_genid; 1938 } 1939 sc->sc_genid = genid; 1940 /* 1941 * Remove all disks without the biggest genid. 1942 */ 1943 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 1944 if (disk->d_genid < genid) { 1945 G_MIRROR_DEBUG(0, 1946 "Component %s (device %s) broken, skipping.", 1947 g_mirror_get_diskname(disk), sc->sc_name); 1948 g_mirror_destroy_disk(disk); 1949 } 1950 } 1951 1952 /* 1953 * Find the biggest syncid. 1954 */ 1955 syncid = 0; 1956 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1957 if (disk->d_sync.ds_syncid > syncid) 1958 syncid = disk->d_sync.ds_syncid; 1959 } 1960 1961 /* 1962 * Here we need to look for dirty disks and if all disks 1963 * with the biggest syncid are dirty, we have to choose 1964 * one with the biggest priority and rebuild the rest. 1965 */ 1966 /* 1967 * Find the number of dirty disks with the biggest syncid. 1968 * Find the number of disks with the biggest syncid. 1969 * While here, find a disk with the biggest priority. 1970 */ 1971 dirty = ndisks = 0; 1972 pdisk = NULL; 1973 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1974 if (disk->d_sync.ds_syncid != syncid) 1975 continue; 1976 if ((disk->d_flags & 1977 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1978 continue; 1979 } 1980 ndisks++; 1981 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1982 dirty++; 1983 if (pdisk == NULL || 1984 pdisk->d_priority < disk->d_priority) { 1985 pdisk = disk; 1986 } 1987 } 1988 } 1989 if (dirty == 0) { 1990 /* No dirty disks at all, great. */ 1991 } else if (dirty == ndisks) { 1992 /* 1993 * Force synchronization for all dirty disks except one 1994 * with the biggest priority. 1995 */ 1996 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1997 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1998 "master disk for synchronization.", 1999 g_mirror_get_diskname(pdisk), sc->sc_name); 2000 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2001 if (disk->d_sync.ds_syncid != syncid) 2002 continue; 2003 if ((disk->d_flags & 2004 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2005 continue; 2006 } 2007 KASSERT((disk->d_flags & 2008 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2009 ("Disk %s isn't marked as dirty.", 2010 g_mirror_get_diskname(disk))); 2011 /* Skip the disk with the biggest priority. */ 2012 if (disk == pdisk) 2013 continue; 2014 disk->d_sync.ds_syncid = 0; 2015 } 2016 } else if (dirty < ndisks) { 2017 /* 2018 * Force synchronization for all dirty disks. 2019 * We have some non-dirty disks. 2020 */ 2021 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2022 if (disk->d_sync.ds_syncid != syncid) 2023 continue; 2024 if ((disk->d_flags & 2025 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2026 continue; 2027 } 2028 if ((disk->d_flags & 2029 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2030 continue; 2031 } 2032 disk->d_sync.ds_syncid = 0; 2033 } 2034 } 2035 2036 /* Reset hint. */ 2037 sc->sc_hint = NULL; 2038 sc->sc_syncid = syncid; 2039 if (force) { 2040 /* Remember to bump syncid on first write. */ 2041 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_OFW; 2042 } 2043 state = G_MIRROR_DEVICE_STATE_RUNNING; 2044 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2045 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2046 g_mirror_device_state2str(state)); 2047 sc->sc_state = state; 2048 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2049 state = g_mirror_determine_state(disk); 2050 g_mirror_event_send(disk, state, 2051 G_MIRROR_EVENT_DONTWAIT); 2052 if (state == G_MIRROR_DISK_STATE_STALE) 2053 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_OFW; 2054 } 2055 wakeup(&g_mirror_class); 2056 break; 2057 } 2058 case G_MIRROR_DEVICE_STATE_RUNNING: 2059 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2060 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2061 /* 2062 * No active disks or no disks at all, 2063 * so destroy device. 2064 */ 2065 if (sc->sc_provider != NULL) 2066 g_mirror_destroy_provider(sc); 2067 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2068 break; 2069 } else if (g_mirror_ndisks(sc, 2070 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2071 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2072 /* 2073 * We have active disks, launch provider if it doesn't 2074 * exist. 2075 */ 2076 if (sc->sc_provider == NULL) 2077 g_mirror_launch_provider(sc); 2078 } 2079 /* 2080 * Bump syncid here, if we need to do it immediately. 2081 */ 2082 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_IMM) != 0) { 2083 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2084 g_mirror_bump_syncid(sc); 2085 } 2086 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID_IMM) != 0) { 2087 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2088 g_mirror_bump_genid(sc); 2089 } 2090 break; 2091 default: 2092 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2093 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2094 break; 2095 } 2096} 2097 2098/* 2099 * Update disk state and device state if needed. 2100 */ 2101#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2102 "Disk %s state changed from %s to %s (device %s).", \ 2103 g_mirror_get_diskname(disk), \ 2104 g_mirror_disk_state2str(disk->d_state), \ 2105 g_mirror_disk_state2str(state), sc->sc_name) 2106static int 2107g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2108{ 2109 struct g_mirror_softc *sc; 2110 2111 g_topology_assert(); 2112 2113 sc = disk->d_softc; 2114again: 2115 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2116 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2117 g_mirror_disk_state2str(state)); 2118 switch (state) { 2119 case G_MIRROR_DISK_STATE_NEW: 2120 /* 2121 * Possible scenarios: 2122 * 1. New disk arrive. 2123 */ 2124 /* Previous state should be NONE. */ 2125 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2126 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2127 g_mirror_disk_state2str(disk->d_state))); 2128 DISK_STATE_CHANGED(); 2129 2130 disk->d_state = state; 2131 if (LIST_EMPTY(&sc->sc_disks)) 2132 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2133 else { 2134 struct g_mirror_disk *dp; 2135 2136 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2137 if (disk->d_priority >= dp->d_priority) { 2138 LIST_INSERT_BEFORE(dp, disk, d_next); 2139 dp = NULL; 2140 break; 2141 } 2142 if (LIST_NEXT(dp, d_next) == NULL) 2143 break; 2144 } 2145 if (dp != NULL) 2146 LIST_INSERT_AFTER(dp, disk, d_next); 2147 } 2148 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2149 sc->sc_name, g_mirror_get_diskname(disk)); 2150 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2151 break; 2152 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2153 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2154 g_mirror_device_state2str(sc->sc_state), 2155 g_mirror_get_diskname(disk), 2156 g_mirror_disk_state2str(disk->d_state))); 2157 state = g_mirror_determine_state(disk); 2158 if (state != G_MIRROR_DISK_STATE_NONE) 2159 goto again; 2160 break; 2161 case G_MIRROR_DISK_STATE_ACTIVE: 2162 /* 2163 * Possible scenarios: 2164 * 1. New disk does not need synchronization. 2165 * 2. Synchronization process finished successfully. 2166 */ 2167 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2168 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2169 g_mirror_device_state2str(sc->sc_state), 2170 g_mirror_get_diskname(disk), 2171 g_mirror_disk_state2str(disk->d_state))); 2172 /* Previous state should be NEW or SYNCHRONIZING. */ 2173 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2174 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2175 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2176 g_mirror_disk_state2str(disk->d_state))); 2177 DISK_STATE_CHANGED(); 2178 2179 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2180 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2181 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2182 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2183 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2184 g_mirror_sync_stop(disk, 0); 2185 } 2186 disk->d_state = state; 2187 disk->d_sync.ds_offset = 0; 2188 disk->d_sync.ds_offset_done = 0; 2189 g_mirror_update_access(disk); 2190 g_mirror_update_metadata(disk); 2191 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2192 sc->sc_name, g_mirror_get_diskname(disk)); 2193 break; 2194 case G_MIRROR_DISK_STATE_STALE: 2195 /* 2196 * Possible scenarios: 2197 * 1. Stale disk was connected. 2198 */ 2199 /* Previous state should be NEW. */ 2200 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2201 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2202 g_mirror_disk_state2str(disk->d_state))); 2203 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2204 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2205 g_mirror_device_state2str(sc->sc_state), 2206 g_mirror_get_diskname(disk), 2207 g_mirror_disk_state2str(disk->d_state))); 2208 /* 2209 * STALE state is only possible if device is marked 2210 * NOAUTOSYNC. 2211 */ 2212 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2213 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2214 g_mirror_device_state2str(sc->sc_state), 2215 g_mirror_get_diskname(disk), 2216 g_mirror_disk_state2str(disk->d_state))); 2217 DISK_STATE_CHANGED(); 2218 2219 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2220 disk->d_state = state; 2221 g_mirror_update_metadata(disk); 2222 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2223 sc->sc_name, g_mirror_get_diskname(disk)); 2224 break; 2225 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2226 /* 2227 * Possible scenarios: 2228 * 1. Disk which needs synchronization was connected. 2229 */ 2230 /* Previous state should be NEW. */ 2231 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2232 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2233 g_mirror_disk_state2str(disk->d_state))); 2234 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2235 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2236 g_mirror_device_state2str(sc->sc_state), 2237 g_mirror_get_diskname(disk), 2238 g_mirror_disk_state2str(disk->d_state))); 2239 DISK_STATE_CHANGED(); 2240 2241 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2242 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2243 disk->d_state = state; 2244 if (sc->sc_provider != NULL) { 2245 g_mirror_sync_start(disk); 2246 g_mirror_update_metadata(disk); 2247 } 2248 break; 2249 case G_MIRROR_DISK_STATE_DISCONNECTED: 2250 /* 2251 * Possible scenarios: 2252 * 1. Device wasn't running yet, but disk disappear. 2253 * 2. Disk was active and disapppear. 2254 * 3. Disk disappear during synchronization process. 2255 */ 2256 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2257 /* 2258 * Previous state should be ACTIVE, STALE or 2259 * SYNCHRONIZING. 2260 */ 2261 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2262 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2263 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2264 ("Wrong disk state (%s, %s).", 2265 g_mirror_get_diskname(disk), 2266 g_mirror_disk_state2str(disk->d_state))); 2267 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2268 /* Previous state should be NEW. */ 2269 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2270 ("Wrong disk state (%s, %s).", 2271 g_mirror_get_diskname(disk), 2272 g_mirror_disk_state2str(disk->d_state))); 2273 /* 2274 * Reset bumping syncid if disk disappeared in STARTING 2275 * state. 2276 */ 2277 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_OFW) != 0) 2278 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2279#ifdef INVARIANTS 2280 } else { 2281 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2282 sc->sc_name, 2283 g_mirror_device_state2str(sc->sc_state), 2284 g_mirror_get_diskname(disk), 2285 g_mirror_disk_state2str(disk->d_state))); 2286#endif 2287 } 2288 DISK_STATE_CHANGED(); 2289 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2290 sc->sc_name, g_mirror_get_diskname(disk)); 2291 2292 g_mirror_destroy_disk(disk); 2293 break; 2294 case G_MIRROR_DISK_STATE_DESTROY: 2295 { 2296 int error; 2297 2298 error = g_mirror_clear_metadata(disk); 2299 if (error != 0) 2300 return (error); 2301 DISK_STATE_CHANGED(); 2302 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2303 sc->sc_name, g_mirror_get_diskname(disk)); 2304 2305 g_mirror_destroy_disk(disk); 2306 sc->sc_ndisks--; 2307 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2308 g_mirror_update_metadata(disk); 2309 } 2310 break; 2311 } 2312 default: 2313 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2314 break; 2315 } 2316 return (0); 2317} 2318#undef DISK_STATE_CHANGED 2319 2320static int 2321g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2322{ 2323 struct g_provider *pp; 2324 u_char *buf; 2325 int error; 2326 2327 g_topology_assert(); 2328 2329 error = g_access(cp, 1, 0, 0); 2330 if (error != 0) 2331 return (error); 2332 pp = cp->provider; 2333 g_topology_unlock(); 2334 /* Metadata are stored on last sector. */ 2335 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2336 &error); 2337 g_topology_lock(); 2338 g_access(cp, -1, 0, 0); 2339 if (error != 0) { 2340 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2341 cp->provider->name, error); 2342 if (buf != NULL) 2343 g_free(buf); 2344 return (error); 2345 } 2346 2347 /* Decode metadata. */ 2348 error = mirror_metadata_decode(buf, md); 2349 g_free(buf); 2350 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2351 return (EINVAL); 2352 if (md->md_version > G_MIRROR_VERSION) { 2353 G_MIRROR_DEBUG(0, 2354 "Kernel module is too old to handle metadata from %s.", 2355 cp->provider->name); 2356 return (EINVAL); 2357 } 2358 if (error != 0) { 2359 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2360 cp->provider->name); 2361 return (error); 2362 } 2363 2364 return (0); 2365} 2366 2367static int 2368g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2369 struct g_mirror_metadata *md) 2370{ 2371 2372 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2373 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2374 pp->name, md->md_did); 2375 return (EEXIST); 2376 } 2377 if (md->md_all != sc->sc_ndisks) { 2378 G_MIRROR_DEBUG(1, 2379 "Invalid '%s' field on disk %s (device %s), skipping.", 2380 "md_all", pp->name, sc->sc_name); 2381 return (EINVAL); 2382 } 2383 if (md->md_slice != sc->sc_slice) { 2384 G_MIRROR_DEBUG(1, 2385 "Invalid '%s' field on disk %s (device %s), skipping.", 2386 "md_slice", pp->name, sc->sc_name); 2387 return (EINVAL); 2388 } 2389 if (md->md_balance != sc->sc_balance) { 2390 G_MIRROR_DEBUG(1, 2391 "Invalid '%s' field on disk %s (device %s), skipping.", 2392 "md_balance", pp->name, sc->sc_name); 2393 return (EINVAL); 2394 } 2395 if (md->md_mediasize != sc->sc_mediasize) { 2396 G_MIRROR_DEBUG(1, 2397 "Invalid '%s' field on disk %s (device %s), skipping.", 2398 "md_mediasize", pp->name, sc->sc_name); 2399 return (EINVAL); 2400 } 2401 if (sc->sc_mediasize > pp->mediasize) { 2402 G_MIRROR_DEBUG(1, 2403 "Invalid size of disk %s (device %s), skipping.", pp->name, 2404 sc->sc_name); 2405 return (EINVAL); 2406 } 2407 if (md->md_sectorsize != sc->sc_sectorsize) { 2408 G_MIRROR_DEBUG(1, 2409 "Invalid '%s' field on disk %s (device %s), skipping.", 2410 "md_sectorsize", pp->name, sc->sc_name); 2411 return (EINVAL); 2412 } 2413 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2414 G_MIRROR_DEBUG(1, 2415 "Invalid sector size of disk %s (device %s), skipping.", 2416 pp->name, sc->sc_name); 2417 return (EINVAL); 2418 } 2419 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2420 G_MIRROR_DEBUG(1, 2421 "Invalid device flags on disk %s (device %s), skipping.", 2422 pp->name, sc->sc_name); 2423 return (EINVAL); 2424 } 2425 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2426 G_MIRROR_DEBUG(1, 2427 "Invalid disk flags on disk %s (device %s), skipping.", 2428 pp->name, sc->sc_name); 2429 return (EINVAL); 2430 } 2431 return (0); 2432} 2433 2434static int 2435g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2436 struct g_mirror_metadata *md) 2437{ 2438 struct g_mirror_disk *disk; 2439 int error; 2440 2441 g_topology_assert(); 2442 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2443 2444 error = g_mirror_check_metadata(sc, pp, md); 2445 if (error != 0) 2446 return (error); 2447 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2448 md->md_genid < sc->sc_genid) { 2449 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2450 pp->name, sc->sc_name); 2451 return (EINVAL); 2452 } 2453 disk = g_mirror_init_disk(sc, pp, md, &error); 2454 if (disk == NULL) 2455 return (error); 2456 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2457 G_MIRROR_EVENT_WAIT); 2458 if (error != 0) 2459 return (error); 2460 if (md->md_version < G_MIRROR_VERSION) { 2461 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2462 pp->name, md->md_version, G_MIRROR_VERSION); 2463 g_mirror_update_metadata(disk); 2464 } 2465 return (0); 2466} 2467 2468static int 2469g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2470{ 2471 struct g_mirror_softc *sc; 2472 struct g_mirror_disk *disk; 2473 int dcr, dcw, dce; 2474 2475 g_topology_assert(); 2476 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2477 acw, ace); 2478 2479 dcr = pp->acr + acr; 2480 dcw = pp->acw + acw; 2481 dce = pp->ace + ace; 2482 2483 sc = pp->geom->softc; 2484 if (sc == NULL || LIST_EMPTY(&sc->sc_disks) || 2485 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 2486 if (acr <= 0 && acw <= 0 && ace <= 0) 2487 return (0); 2488 else 2489 return (ENXIO); 2490 } 2491 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2492 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2493 continue; 2494 /* 2495 * Mark disk as dirty on open and unmark on close. 2496 */ 2497 if (pp->acw == 0 && dcw > 0) { 2498 G_MIRROR_DEBUG(1, 2499 "Disk %s (device %s) marked as dirty.", 2500 g_mirror_get_diskname(disk), sc->sc_name); 2501 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2502 g_mirror_update_metadata(disk); 2503 } else if (pp->acw > 0 && dcw == 0) { 2504 G_MIRROR_DEBUG(1, 2505 "Disk %s (device %s) marked as clean.", 2506 g_mirror_get_diskname(disk), sc->sc_name); 2507 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2508 g_mirror_update_metadata(disk); 2509 } 2510 } 2511 return (0); 2512} 2513 2514static struct g_geom * 2515g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2516{ 2517 struct g_mirror_softc *sc; 2518 struct g_geom *gp; 2519 int error, timeout; 2520 2521 g_topology_assert(); 2522 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2523 md->md_mid); 2524 2525 /* One disk is minimum. */ 2526 if (md->md_all < 1) 2527 return (NULL); 2528 /* 2529 * Action geom. 2530 */ 2531 gp = g_new_geomf(mp, "%s", md->md_name); 2532 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2533 gp->start = g_mirror_start; 2534 gp->spoiled = g_mirror_spoiled; 2535 gp->orphan = g_mirror_orphan; 2536 gp->access = g_mirror_access; 2537 gp->dumpconf = g_mirror_dumpconf; 2538 2539 sc->sc_id = md->md_mid; 2540 sc->sc_slice = md->md_slice; 2541 sc->sc_balance = md->md_balance; 2542 sc->sc_mediasize = md->md_mediasize; 2543 sc->sc_sectorsize = md->md_sectorsize; 2544 sc->sc_ndisks = md->md_all; 2545 sc->sc_flags = md->md_mflags; 2546 sc->sc_bump_id = 0; 2547 sc->sc_idle = 0; 2548 bioq_init(&sc->sc_queue); 2549 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2550 LIST_INIT(&sc->sc_disks); 2551 TAILQ_INIT(&sc->sc_events); 2552 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2553 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2554 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2555 gp->softc = sc; 2556 sc->sc_geom = gp; 2557 sc->sc_provider = NULL; 2558 /* 2559 * Synchronization geom. 2560 */ 2561 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2562 gp->softc = sc; 2563 gp->orphan = g_mirror_orphan; 2564 sc->sc_sync.ds_geom = gp; 2565 sc->sc_sync.ds_ndisks = 0; 2566 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2567 "g_mirror %s", md->md_name); 2568 if (error != 0) { 2569 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2570 sc->sc_name); 2571 g_destroy_geom(sc->sc_sync.ds_geom); 2572 mtx_destroy(&sc->sc_events_mtx); 2573 mtx_destroy(&sc->sc_queue_mtx); 2574 g_destroy_geom(sc->sc_geom); 2575 free(sc, M_MIRROR); 2576 return (NULL); 2577 } 2578 2579 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2580 2581 /* 2582 * Run timeout. 2583 */ 2584 timeout = g_mirror_timeout * hz; 2585 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2586 return (sc->sc_geom); 2587} 2588 2589int 2590g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2591{ 2592 struct g_provider *pp; 2593 2594 g_topology_assert(); 2595 2596 if (sc == NULL) 2597 return (ENXIO); 2598 pp = sc->sc_provider; 2599 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2600 if (force) { 2601 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 2602 "can't be definitely removed.", pp->name); 2603 } else { 2604 G_MIRROR_DEBUG(1, 2605 "Device %s is still open (r%dw%de%d).", pp->name, 2606 pp->acr, pp->acw, pp->ace); 2607 return (EBUSY); 2608 } 2609 } 2610 2611 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2612 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2613 g_topology_unlock(); 2614 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2615 mtx_lock(&sc->sc_queue_mtx); 2616 wakeup(sc); 2617 mtx_unlock(&sc->sc_queue_mtx); 2618 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2619 while (sc->sc_worker != NULL) 2620 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2621 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2622 g_topology_lock(); 2623 g_mirror_destroy_device(sc); 2624 free(sc, M_MIRROR); 2625 return (0); 2626} 2627 2628static void 2629g_mirror_taste_orphan(struct g_consumer *cp) 2630{ 2631 2632 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2633 cp->provider->name)); 2634} 2635 2636static struct g_geom * 2637g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2638{ 2639 struct g_mirror_metadata md; 2640 struct g_mirror_softc *sc; 2641 struct g_consumer *cp; 2642 struct g_geom *gp; 2643 int error; 2644 2645 g_topology_assert(); 2646 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2647 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2648 2649 gp = g_new_geomf(mp, "mirror:taste"); 2650 /* 2651 * This orphan function should be never called. 2652 */ 2653 gp->orphan = g_mirror_taste_orphan; 2654 cp = g_new_consumer(gp); 2655 g_attach(cp, pp); 2656 error = g_mirror_read_metadata(cp, &md); 2657 g_detach(cp); 2658 g_destroy_consumer(cp); 2659 g_destroy_geom(gp); 2660 if (error != 0) 2661 return (NULL); 2662 gp = NULL; 2663 2664 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2665 return (NULL); 2666 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2667 G_MIRROR_DEBUG(0, 2668 "Device %s: provider %s marked as inactive, skipping.", 2669 md.md_name, pp->name); 2670 return (NULL); 2671 } 2672 if (g_mirror_debug >= 2) 2673 mirror_metadata_dump(&md); 2674 2675 /* 2676 * Let's check if device already exists. 2677 */ 2678 sc = NULL; 2679 LIST_FOREACH(gp, &mp->geom, geom) { 2680 sc = gp->softc; 2681 if (sc == NULL) 2682 continue; 2683 if (sc->sc_sync.ds_geom == gp) 2684 continue; 2685 if (strcmp(md.md_name, sc->sc_name) != 0) 2686 continue; 2687 if (md.md_mid != sc->sc_id) { 2688 G_MIRROR_DEBUG(0, "Device %s already configured.", 2689 sc->sc_name); 2690 return (NULL); 2691 } 2692 break; 2693 } 2694 if (gp == NULL) { 2695 gp = g_mirror_create(mp, &md); 2696 if (gp == NULL) { 2697 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2698 md.md_name); 2699 return (NULL); 2700 } 2701 sc = gp->softc; 2702 } 2703 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2704 error = g_mirror_add_disk(sc, pp, &md); 2705 if (error != 0) { 2706 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2707 pp->name, gp->name, error); 2708 if (LIST_EMPTY(&sc->sc_disks)) 2709 g_mirror_destroy(sc, 1); 2710 return (NULL); 2711 } 2712 return (gp); 2713} 2714 2715static int 2716g_mirror_destroy_geom(struct gctl_req *req __unused, 2717 struct g_class *mp __unused, struct g_geom *gp) 2718{ 2719 2720 return (g_mirror_destroy(gp->softc, 0)); 2721} 2722 2723static void 2724g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2725 struct g_consumer *cp, struct g_provider *pp) 2726{ 2727 struct g_mirror_softc *sc; 2728 2729 g_topology_assert(); 2730 2731 sc = gp->softc; 2732 if (sc == NULL) 2733 return; 2734 /* Skip synchronization geom. */ 2735 if (gp == sc->sc_sync.ds_geom) 2736 return; 2737 if (pp != NULL) { 2738 /* Nothing here. */ 2739 } else if (cp != NULL) { 2740 struct g_mirror_disk *disk; 2741 2742 disk = cp->private; 2743 if (disk == NULL) 2744 return; 2745 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2746 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2747 sbuf_printf(sb, "%s<Synchronized>", indent); 2748 if (disk->d_sync.ds_offset_done == 0) 2749 sbuf_printf(sb, "0%%"); 2750 else { 2751 sbuf_printf(sb, "%u%%", 2752 (u_int)((disk->d_sync.ds_offset_done * 100) / 2753 sc->sc_provider->mediasize)); 2754 } 2755 sbuf_printf(sb, "</Synchronized>\n"); 2756 } 2757 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2758 disk->d_sync.ds_syncid); 2759 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 2760 disk->d_genid); 2761 sbuf_printf(sb, "%s<Flags>", indent); 2762 if (disk->d_flags == 0) 2763 sbuf_printf(sb, "NONE"); 2764 else { 2765 int first = 1; 2766 2767#define ADD_FLAG(flag, name) do { \ 2768 if ((disk->d_flags & (flag)) != 0) { \ 2769 if (!first) \ 2770 sbuf_printf(sb, ", "); \ 2771 else \ 2772 first = 0; \ 2773 sbuf_printf(sb, name); \ 2774 } \ 2775} while (0) 2776 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2777 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2778 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2779 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2780 "SYNCHRONIZING"); 2781 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2782#undef ADD_FLAG 2783 } 2784 sbuf_printf(sb, "</Flags>\n"); 2785 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2786 disk->d_priority); 2787 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2788 g_mirror_disk_state2str(disk->d_state)); 2789 } else { 2790 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2791 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2792 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 2793 sbuf_printf(sb, "%s<Flags>", indent); 2794 if (sc->sc_flags == 0) 2795 sbuf_printf(sb, "NONE"); 2796 else { 2797 int first = 1; 2798 2799#define ADD_FLAG(flag, name) do { \ 2800 if ((sc->sc_flags & (flag)) != 0) { \ 2801 if (!first) \ 2802 sbuf_printf(sb, ", "); \ 2803 else \ 2804 first = 0; \ 2805 sbuf_printf(sb, name); \ 2806 } \ 2807} while (0) 2808 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2809#undef ADD_FLAG 2810 } 2811 sbuf_printf(sb, "</Flags>\n"); 2812 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2813 (u_int)sc->sc_slice); 2814 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2815 balance_name(sc->sc_balance)); 2816 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2817 sc->sc_ndisks); 2818 sbuf_printf(sb, "%s<State>", indent); 2819 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2820 sbuf_printf(sb, "%s", "STARTING"); 2821 else if (sc->sc_ndisks == 2822 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 2823 sbuf_printf(sb, "%s", "COMPLETE"); 2824 else 2825 sbuf_printf(sb, "%s", "DEGRADED"); 2826 sbuf_printf(sb, "</State>\n"); 2827 } 2828} 2829 2830static void 2831g_mirror_shutdown(void *arg, int howto) 2832{ 2833 struct g_class *mp; 2834 struct g_geom *gp, *gp2; 2835 2836 mp = arg; 2837 DROP_GIANT(); 2838 g_topology_lock(); 2839 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2840 if (gp->softc == NULL) 2841 continue; 2842 g_mirror_destroy(gp->softc, 1); 2843 } 2844 g_topology_unlock(); 2845 PICKUP_GIANT(); 2846#if 0 2847 tsleep(&gp, PRIBIO, "m:shutdown", hz * 20); 2848#endif 2849} 2850 2851static void 2852g_mirror_init(struct g_class *mp) 2853{ 2854 2855 g_mirror_ehtag = EVENTHANDLER_REGISTER(shutdown_post_sync, 2856 g_mirror_shutdown, mp, SHUTDOWN_PRI_FIRST); 2857 if (g_mirror_ehtag == NULL) 2858 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 2859} 2860 2861static void 2862g_mirror_fini(struct g_class *mp) 2863{ 2864 2865 if (g_mirror_ehtag == NULL) 2866 return; 2867 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_ehtag); 2868} 2869 2870static int 2871g_mirror_can_go(void) 2872{ 2873 struct g_mirror_softc *sc; 2874 struct g_geom *gp; 2875 struct g_provider *pp; 2876 int can_go; 2877 2878 DROP_GIANT(); 2879 can_go = 1; 2880 g_topology_lock(); 2881 LIST_FOREACH(gp, &g_mirror_class.geom, geom) { 2882 sc = gp->softc; 2883 if (sc == NULL) { 2884 can_go = 0; 2885 break; 2886 } 2887 pp = sc->sc_provider; 2888 if (pp == NULL || pp->error != 0) { 2889 can_go = 0; 2890 break; 2891 } 2892 } 2893 g_topology_unlock(); 2894 PICKUP_GIANT(); 2895 return (can_go); 2896} 2897 2898static void 2899g_mirror_rootwait(void) 2900{ 2901 2902 /* 2903 * HACK: Wait for GEOM, because g_mirror_rootwait() can be called, 2904 * HACK: before we get providers for tasting. 2905 */ 2906 tsleep(&g_mirror_class, PRIBIO, "mroot", hz * 3); 2907 /* 2908 * Wait for mirrors in degraded state. 2909 */ 2910 for (;;) { 2911 if (g_mirror_can_go()) 2912 break; 2913 tsleep(&g_mirror_class, PRIBIO, "mroot", hz); 2914 } 2915} 2916 2917SYSINIT(g_mirror_root, SI_SUB_RAID, SI_ORDER_FIRST, g_mirror_rootwait, NULL) 2918 2919DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2920