g_mirror.c revision 137253
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 137253 2004-11-05 12:31:32Z 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 <vm/uma.h> 41#include <geom/geom.h> 42#include <sys/proc.h> 43#include <sys/kthread.h> 44#include <geom/mirror/g_mirror.h> 45 46 47static MALLOC_DEFINE(M_MIRROR, "mirror data", "GEOM_MIRROR Data"); 48 49SYSCTL_DECL(_kern_geom); 50SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, "GEOM_MIRROR stuff"); 51u_int g_mirror_debug = 0; 52TUNABLE_INT("kern.geom.mirror.debug", &g_mirror_debug); 53SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RW, &g_mirror_debug, 0, 54 "Debug level"); 55static u_int g_mirror_timeout = 4; 56TUNABLE_INT("kern.geom.mirror.timeout", &g_mirror_timeout); 57SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RW, &g_mirror_timeout, 58 0, "Time to wait on all mirror components"); 59static u_int g_mirror_idletime = 5; 60TUNABLE_INT("kern.geom.mirror.idletime", &g_mirror_idletime); 61SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RW, 62 &g_mirror_idletime, 0, "Mark components as clean when idling"); 63static u_int g_mirror_reqs_per_sync = 5; 64SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, reqs_per_sync, CTLFLAG_RW, 65 &g_mirror_reqs_per_sync, 0, 66 "Number of regular I/O requests per synchronization request"); 67static u_int g_mirror_syncs_per_sec = 100; 68SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, syncs_per_sec, CTLFLAG_RW, 69 &g_mirror_syncs_per_sec, 0, 70 "Number of synchronizations requests per second"); 71 72#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 73 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 74 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 75 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 76} while (0) 77 78 79static int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp, 80 struct g_geom *gp); 81static g_taste_t g_mirror_taste; 82 83struct g_class g_mirror_class = { 84 .name = G_MIRROR_CLASS_NAME, 85 .version = G_VERSION, 86 .ctlreq = g_mirror_config, 87 .taste = g_mirror_taste, 88 .destroy_geom = g_mirror_destroy_geom 89}; 90 91 92static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 93static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 94static void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force); 95static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 96 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 97static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 98 99 100static const char * 101g_mirror_disk_state2str(int state) 102{ 103 104 switch (state) { 105 case G_MIRROR_DISK_STATE_NONE: 106 return ("NONE"); 107 case G_MIRROR_DISK_STATE_NEW: 108 return ("NEW"); 109 case G_MIRROR_DISK_STATE_ACTIVE: 110 return ("ACTIVE"); 111 case G_MIRROR_DISK_STATE_STALE: 112 return ("STALE"); 113 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 114 return ("SYNCHRONIZING"); 115 case G_MIRROR_DISK_STATE_DISCONNECTED: 116 return ("DISCONNECTED"); 117 case G_MIRROR_DISK_STATE_DESTROY: 118 return ("DESTROY"); 119 default: 120 return ("INVALID"); 121 } 122} 123 124static const char * 125g_mirror_device_state2str(int state) 126{ 127 128 switch (state) { 129 case G_MIRROR_DEVICE_STATE_STARTING: 130 return ("STARTING"); 131 case G_MIRROR_DEVICE_STATE_RUNNING: 132 return ("RUNNING"); 133 default: 134 return ("INVALID"); 135 } 136} 137 138static const char * 139g_mirror_get_diskname(struct g_mirror_disk *disk) 140{ 141 142 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 143 return ("[unknown]"); 144 return (disk->d_name); 145} 146 147/* 148 * --- Events handling functions --- 149 * Events in geom_mirror are used to maintain disks and device status 150 * from one thread to simplify locking. 151 */ 152static void 153g_mirror_event_free(struct g_mirror_event *ep) 154{ 155 156 free(ep, M_MIRROR); 157} 158 159int 160g_mirror_event_send(void *arg, int state, int flags) 161{ 162 struct g_mirror_softc *sc; 163 struct g_mirror_disk *disk; 164 struct g_mirror_event *ep; 165 int error; 166 167 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 168 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 169 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 170 disk = NULL; 171 sc = arg; 172 } else { 173 disk = arg; 174 sc = disk->d_softc; 175 } 176 ep->e_disk = disk; 177 ep->e_state = state; 178 ep->e_flags = flags; 179 ep->e_error = 0; 180 mtx_lock(&sc->sc_events_mtx); 181 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 182 mtx_unlock(&sc->sc_events_mtx); 183 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 184 mtx_lock(&sc->sc_queue_mtx); 185 wakeup(sc); 186 mtx_unlock(&sc->sc_queue_mtx); 187 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 188 return (0); 189 g_topology_assert(); 190 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 191 g_topology_unlock(); 192 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 193 mtx_lock(&sc->sc_events_mtx); 194 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 195 hz * 5); 196 } 197 /* Don't even try to use 'sc' here, because it could be already dead. */ 198 g_topology_lock(); 199 error = ep->e_error; 200 g_mirror_event_free(ep); 201 return (error); 202} 203 204static struct g_mirror_event * 205g_mirror_event_get(struct g_mirror_softc *sc) 206{ 207 struct g_mirror_event *ep; 208 209 mtx_lock(&sc->sc_events_mtx); 210 ep = TAILQ_FIRST(&sc->sc_events); 211 if (ep != NULL) 212 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 213 mtx_unlock(&sc->sc_events_mtx); 214 return (ep); 215} 216 217static void 218g_mirror_event_cancel(struct g_mirror_disk *disk) 219{ 220 struct g_mirror_softc *sc; 221 struct g_mirror_event *ep, *tmpep; 222 223 g_topology_assert(); 224 225 sc = disk->d_softc; 226 mtx_lock(&sc->sc_events_mtx); 227 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 228 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 229 continue; 230 if (ep->e_disk != disk) 231 continue; 232 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 233 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 234 g_mirror_event_free(ep); 235 else { 236 ep->e_error = ECANCELED; 237 wakeup(ep); 238 } 239 } 240 mtx_unlock(&sc->sc_events_mtx); 241} 242 243/* 244 * Return the number of disks in given state. 245 * If state is equal to -1, count all connected disks. 246 */ 247u_int 248g_mirror_ndisks(struct g_mirror_softc *sc, int state) 249{ 250 struct g_mirror_disk *disk; 251 u_int n = 0; 252 253 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 254 if (state == -1 || disk->d_state == state) 255 n++; 256 } 257 return (n); 258} 259 260/* 261 * Find a disk in mirror by its disk ID. 262 */ 263static struct g_mirror_disk * 264g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 265{ 266 struct g_mirror_disk *disk; 267 268 g_topology_assert(); 269 270 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 271 if (disk->d_id == id) 272 return (disk); 273 } 274 return (NULL); 275} 276 277static u_int 278g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 279{ 280 struct bio *bp; 281 u_int nreqs = 0; 282 283 mtx_lock(&sc->sc_queue_mtx); 284 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 285 if (bp->bio_from == cp) 286 nreqs++; 287 } 288 mtx_unlock(&sc->sc_queue_mtx); 289 return (nreqs); 290} 291 292static int 293g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 294{ 295 296 if (cp->index > 0) { 297 G_MIRROR_DEBUG(2, 298 "I/O requests for %s exist, can't destroy it now.", 299 cp->provider->name); 300 return (1); 301 } 302 if (g_mirror_nrequests(sc, cp) > 0) { 303 G_MIRROR_DEBUG(2, 304 "I/O requests for %s in queue, can't destroy it now.", 305 cp->provider->name); 306 return (1); 307 } 308 return (0); 309} 310 311static void 312g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 313{ 314 315 g_topology_assert(); 316 317 cp->private = NULL; 318 if (g_mirror_is_busy(sc, cp)) 319 return; 320 G_MIRROR_DEBUG(2, "Consumer %s destroyed.", cp->provider->name); 321 g_detach(cp); 322 g_destroy_consumer(cp); 323} 324 325static int 326g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 327{ 328 int error; 329 330 g_topology_assert(); 331 KASSERT(disk->d_consumer == NULL, 332 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 333 334 disk->d_consumer = g_new_consumer(disk->d_softc->sc_geom); 335 disk->d_consumer->private = disk; 336 disk->d_consumer->index = 0; 337 error = g_attach(disk->d_consumer, pp); 338 if (error != 0) 339 return (error); 340 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 341 return (0); 342} 343 344static void 345g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 346{ 347 348 g_topology_assert(); 349 350 if (cp == NULL) 351 return; 352 if (cp->provider != NULL) { 353 G_MIRROR_DEBUG(2, "Disk %s disconnected.", cp->provider->name); 354 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) { 355 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 356 cp->provider->name, -cp->acr, -cp->acw, -cp->ace, 357 0); 358 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 359 } 360 g_mirror_kill_consumer(sc, cp); 361 } else { 362 g_destroy_consumer(cp); 363 } 364} 365 366/* 367 * Initialize disk. This means allocate memory, create consumer, attach it 368 * to the provider and open access (r1w1e1) to it. 369 */ 370static struct g_mirror_disk * 371g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 372 struct g_mirror_metadata *md, int *errorp) 373{ 374 struct g_mirror_disk *disk; 375 int error; 376 377 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 378 if (disk == NULL) { 379 error = ENOMEM; 380 goto fail; 381 } 382 disk->d_softc = sc; 383 error = g_mirror_connect_disk(disk, pp); 384 if (error != 0) 385 goto fail; 386 disk->d_id = md->md_did; 387 disk->d_state = G_MIRROR_DISK_STATE_NONE; 388 disk->d_priority = md->md_priority; 389 disk->d_delay.sec = 0; 390 disk->d_delay.frac = 0; 391 binuptime(&disk->d_last_used); 392 disk->d_flags = md->md_dflags; 393 if (md->md_provider[0] != '\0') 394 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 395 disk->d_sync.ds_consumer = NULL; 396 disk->d_sync.ds_offset = md->md_sync_offset; 397 disk->d_sync.ds_offset_done = md->md_sync_offset; 398 disk->d_sync.ds_resync = -1; 399 disk->d_sync.ds_syncid = md->md_syncid; 400 if (errorp != NULL) 401 *errorp = 0; 402 return (disk); 403fail: 404 if (errorp != NULL) 405 *errorp = error; 406 if (disk != NULL) { 407 g_mirror_disconnect_consumer(sc, disk->d_consumer); 408 free(disk, M_MIRROR); 409 } 410 return (NULL); 411} 412 413static void 414g_mirror_destroy_disk(struct g_mirror_disk *disk) 415{ 416 struct g_mirror_softc *sc; 417 418 g_topology_assert(); 419 420 LIST_REMOVE(disk, d_next); 421 g_mirror_event_cancel(disk); 422 sc = disk->d_softc; 423 if (sc->sc_hint == disk) 424 sc->sc_hint = NULL; 425 switch (disk->d_state) { 426 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 427 g_mirror_sync_stop(disk, 1); 428 /* FALLTHROUGH */ 429 case G_MIRROR_DISK_STATE_NEW: 430 case G_MIRROR_DISK_STATE_STALE: 431 case G_MIRROR_DISK_STATE_ACTIVE: 432 g_mirror_disconnect_consumer(sc, disk->d_consumer); 433 free(disk, M_MIRROR); 434 break; 435 default: 436 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 437 g_mirror_get_diskname(disk), 438 g_mirror_disk_state2str(disk->d_state))); 439 } 440} 441 442static void 443g_mirror_destroy_device(struct g_mirror_softc *sc) 444{ 445 struct g_mirror_disk *disk; 446 struct g_mirror_event *ep; 447 struct g_geom *gp; 448 struct g_consumer *cp, *tmpcp; 449 450 g_topology_assert(); 451 452 gp = sc->sc_geom; 453 if (sc->sc_provider != NULL) 454 g_mirror_destroy_provider(sc); 455 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 456 disk = LIST_FIRST(&sc->sc_disks)) { 457 g_mirror_destroy_disk(disk); 458 } 459 while ((ep = g_mirror_event_get(sc)) != NULL) { 460 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 461 g_mirror_event_free(ep); 462 else { 463 ep->e_error = ECANCELED; 464 ep->e_flags |= G_MIRROR_EVENT_DONE; 465 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 466 mtx_lock(&sc->sc_events_mtx); 467 wakeup(ep); 468 mtx_unlock(&sc->sc_events_mtx); 469 } 470 } 471 callout_drain(&sc->sc_callout); 472 gp->softc = NULL; 473 474 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 475 g_mirror_disconnect_consumer(sc, cp); 476 } 477 sc->sc_sync.ds_geom->softc = NULL; 478 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 479 mtx_destroy(&sc->sc_queue_mtx); 480 mtx_destroy(&sc->sc_events_mtx); 481 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 482 g_wither_geom(gp, ENXIO); 483} 484 485static void 486g_mirror_orphan(struct g_consumer *cp) 487{ 488 struct g_mirror_disk *disk; 489 490 g_topology_assert(); 491 492 disk = cp->private; 493 if (disk == NULL) 494 return; 495 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 496 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 497 G_MIRROR_EVENT_DONTWAIT); 498} 499 500static void 501g_mirror_spoiled(struct g_consumer *cp) 502{ 503 struct g_mirror_disk *disk; 504 505 g_topology_assert(); 506 507 disk = cp->private; 508 if (disk == NULL) 509 return; 510 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 511 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 512 G_MIRROR_EVENT_DONTWAIT); 513} 514 515/* 516 * Function should return the next active disk on the list. 517 * It is possible that it will be the same disk as given. 518 * If there are no active disks on list, NULL is returned. 519 */ 520static __inline struct g_mirror_disk * 521g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 522{ 523 struct g_mirror_disk *dp; 524 525 for (dp = LIST_NEXT(disk, d_next); dp != disk; 526 dp = LIST_NEXT(dp, d_next)) { 527 if (dp == NULL) 528 dp = LIST_FIRST(&sc->sc_disks); 529 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 530 break; 531 } 532 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 533 return (NULL); 534 return (dp); 535} 536 537static struct g_mirror_disk * 538g_mirror_get_disk(struct g_mirror_softc *sc) 539{ 540 struct g_mirror_disk *disk; 541 542 if (sc->sc_hint == NULL) { 543 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 544 if (sc->sc_hint == NULL) 545 return (NULL); 546 } 547 disk = sc->sc_hint; 548 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 549 disk = g_mirror_find_next(sc, disk); 550 if (disk == NULL) 551 return (NULL); 552 } 553 sc->sc_hint = g_mirror_find_next(sc, disk); 554 return (disk); 555} 556 557static int 558g_mirror_write_metadata(struct g_mirror_disk *disk, 559 struct g_mirror_metadata *md) 560{ 561 struct g_mirror_softc *sc; 562 struct g_consumer *cp; 563 off_t offset, length; 564 u_char *sector; 565 int close = 0, error = 0; 566 567 g_topology_assert(); 568 569 sc = disk->d_softc; 570 cp = disk->d_consumer; 571 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 572 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 573 length = cp->provider->sectorsize; 574 offset = cp->provider->mediasize - length; 575 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 576 /* 577 * Open consumer if it wasn't opened and remember to close it. 578 */ 579 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 580 error = g_access(cp, 0, 1, 1); 581 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 582 cp->provider->name, 0, 1, 1, error); 583 if (error == 0) 584 close = 1; 585#ifdef INVARIANTS 586 } else { 587 KASSERT(cp->acw > 0 && cp->ace > 0, 588 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 589 cp->acr, cp->acw, cp->ace)); 590#endif 591 } 592 if (error == 0) { 593 if (md != NULL) 594 mirror_metadata_encode(md, sector); 595 g_topology_unlock(); 596 error = g_write_data(cp, offset, sector, length); 597 g_topology_lock(); 598 } 599 free(sector, M_MIRROR); 600 if (close) { 601 g_access(cp, 0, -1, -1); 602 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 603 cp->provider->name, 0, -1, -1, 0); 604 } 605 if (error != 0) { 606 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 607 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 608 G_MIRROR_EVENT_DONTWAIT); 609 } 610 return (error); 611} 612 613static int 614g_mirror_clear_metadata(struct g_mirror_disk *disk) 615{ 616 int error; 617 618 g_topology_assert(); 619 error = g_mirror_write_metadata(disk, NULL); 620 if (error == 0) { 621 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 622 g_mirror_get_diskname(disk)); 623 } else { 624 G_MIRROR_DEBUG(0, 625 "Cannot clear metadata on disk %s (error=%d).", 626 g_mirror_get_diskname(disk), error); 627 } 628 return (error); 629} 630 631void 632g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 633 struct g_mirror_metadata *md) 634{ 635 636 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 637 md->md_version = G_MIRROR_VERSION; 638 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 639 md->md_mid = sc->sc_id; 640 md->md_all = sc->sc_ndisks; 641 md->md_slice = sc->sc_slice; 642 md->md_balance = sc->sc_balance; 643 md->md_mediasize = sc->sc_mediasize; 644 md->md_sectorsize = sc->sc_sectorsize; 645 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 646 bzero(md->md_provider, sizeof(md->md_provider)); 647 if (disk == NULL) { 648 md->md_did = arc4random(); 649 md->md_priority = 0; 650 md->md_syncid = 0; 651 md->md_dflags = 0; 652 md->md_sync_offset = 0; 653 } else { 654 md->md_did = disk->d_id; 655 md->md_priority = disk->d_priority; 656 md->md_syncid = disk->d_sync.ds_syncid; 657 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 658 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 659 md->md_sync_offset = disk->d_sync.ds_offset_done; 660 else 661 md->md_sync_offset = 0; 662 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 663 strlcpy(md->md_provider, 664 disk->d_consumer->provider->name, 665 sizeof(md->md_provider)); 666 } 667 } 668} 669 670void 671g_mirror_update_metadata(struct g_mirror_disk *disk) 672{ 673 struct g_mirror_metadata md; 674 int error; 675 676 g_topology_assert(); 677 g_mirror_fill_metadata(disk->d_softc, disk, &md); 678 error = g_mirror_write_metadata(disk, &md); 679 if (error == 0) { 680 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 681 g_mirror_get_diskname(disk)); 682 } else { 683 G_MIRROR_DEBUG(0, 684 "Cannot update metadata on disk %s (error=%d).", 685 g_mirror_get_diskname(disk), error); 686 } 687} 688 689static void 690g_mirror_bump_syncid(struct g_mirror_softc *sc) 691{ 692 struct g_mirror_disk *disk; 693 694 g_topology_assert(); 695 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 696 ("%s called with no active disks (device=%s).", __func__, 697 sc->sc_name)); 698 699 sc->sc_syncid++; 700 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 701 sc->sc_syncid); 702 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 703 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 704 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 705 disk->d_sync.ds_syncid = sc->sc_syncid; 706 g_mirror_update_metadata(disk); 707 } 708 } 709} 710 711static void 712g_mirror_idle(struct g_mirror_softc *sc) 713{ 714 struct g_mirror_disk *disk; 715 716 if (sc->sc_provider == NULL || sc->sc_provider->acw == 0) 717 return; 718 sc->sc_idle = 1; 719 g_topology_lock(); 720 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 721 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 722 continue; 723 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 724 g_mirror_get_diskname(disk), sc->sc_name); 725 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 726 g_mirror_update_metadata(disk); 727 } 728 g_topology_unlock(); 729} 730 731static void 732g_mirror_unidle(struct g_mirror_softc *sc) 733{ 734 struct g_mirror_disk *disk; 735 736 sc->sc_idle = 0; 737 g_topology_lock(); 738 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 739 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 740 continue; 741 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 742 g_mirror_get_diskname(disk), sc->sc_name); 743 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 744 g_mirror_update_metadata(disk); 745 } 746 g_topology_unlock(); 747} 748 749static __inline int 750bintime_cmp(struct bintime *bt1, struct bintime *bt2) 751{ 752 753 if (bt1->sec < bt2->sec) 754 return (-1); 755 else if (bt1->sec > bt2->sec) 756 return (1); 757 if (bt1->frac < bt2->frac) 758 return (-1); 759 else if (bt1->frac > bt2->frac) 760 return (1); 761 return (0); 762} 763 764static void 765g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 766{ 767 768 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 769 return; 770 binuptime(&disk->d_delay); 771 bintime_sub(&disk->d_delay, &bp->bio_t0); 772} 773 774static void 775g_mirror_done(struct bio *bp) 776{ 777 struct g_mirror_softc *sc; 778 779 sc = bp->bio_from->geom->softc; 780 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 781 mtx_lock(&sc->sc_queue_mtx); 782 bioq_disksort(&sc->sc_queue, bp); 783 wakeup(sc); 784 mtx_unlock(&sc->sc_queue_mtx); 785} 786 787static void 788g_mirror_regular_request(struct bio *bp) 789{ 790 struct g_mirror_softc *sc; 791 struct g_mirror_disk *disk; 792 struct bio *pbp; 793 794 g_topology_assert_not(); 795 796 bp->bio_from->index--; 797 pbp = bp->bio_parent; 798 sc = pbp->bio_to->geom->softc; 799 disk = bp->bio_from->private; 800 if (disk == NULL) { 801 g_topology_lock(); 802 g_mirror_kill_consumer(sc, bp->bio_from); 803 g_topology_unlock(); 804 } else { 805 g_mirror_update_delay(disk, bp); 806 } 807 808 pbp->bio_inbed++; 809 KASSERT(pbp->bio_inbed <= pbp->bio_children, 810 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 811 pbp->bio_children)); 812 if (bp->bio_error == 0 && pbp->bio_error == 0) { 813 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 814 g_destroy_bio(bp); 815 if (pbp->bio_children == pbp->bio_inbed) { 816 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 817 pbp->bio_completed = pbp->bio_length; 818 g_io_deliver(pbp, pbp->bio_error); 819 } 820 return; 821 } else if (bp->bio_error != 0) { 822 if (pbp->bio_error == 0) 823 pbp->bio_error = bp->bio_error; 824 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 825 bp->bio_error); 826 if (disk != NULL) { 827 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 828 g_mirror_event_send(disk, 829 G_MIRROR_DISK_STATE_DISCONNECTED, 830 G_MIRROR_EVENT_DONTWAIT); 831 } 832 switch (pbp->bio_cmd) { 833 case BIO_DELETE: 834 case BIO_WRITE: 835 pbp->bio_inbed--; 836 pbp->bio_children--; 837 break; 838 } 839 } 840 g_destroy_bio(bp); 841 842 switch (pbp->bio_cmd) { 843 case BIO_READ: 844 if (pbp->bio_children == pbp->bio_inbed) { 845 pbp->bio_error = 0; 846 mtx_lock(&sc->sc_queue_mtx); 847 bioq_disksort(&sc->sc_queue, pbp); 848 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 849 wakeup(sc); 850 mtx_unlock(&sc->sc_queue_mtx); 851 } 852 break; 853 case BIO_DELETE: 854 case BIO_WRITE: 855 if (pbp->bio_children == 0) { 856 /* 857 * All requests failed. 858 */ 859 } else if (pbp->bio_inbed < pbp->bio_children) { 860 /* Do nothing. */ 861 break; 862 } else if (pbp->bio_children == pbp->bio_inbed) { 863 /* Some requests succeeded. */ 864 pbp->bio_error = 0; 865 pbp->bio_completed = pbp->bio_length; 866 } 867 g_io_deliver(pbp, pbp->bio_error); 868 break; 869 default: 870 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 871 break; 872 } 873} 874 875static void 876g_mirror_sync_done(struct bio *bp) 877{ 878 struct g_mirror_softc *sc; 879 880 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 881 sc = bp->bio_from->geom->softc; 882 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 883 mtx_lock(&sc->sc_queue_mtx); 884 bioq_disksort(&sc->sc_queue, bp); 885 wakeup(sc); 886 mtx_unlock(&sc->sc_queue_mtx); 887} 888 889static void 890g_mirror_start(struct bio *bp) 891{ 892 struct g_mirror_softc *sc; 893 894 sc = bp->bio_to->geom->softc; 895 /* 896 * If sc == NULL or there are no valid disks, provider's error 897 * should be set and g_mirror_start() should not be called at all. 898 */ 899 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 900 ("Provider's error should be set (error=%d)(mirror=%s).", 901 bp->bio_to->error, bp->bio_to->name)); 902 G_MIRROR_LOGREQ(3, bp, "Request received."); 903 904 switch (bp->bio_cmd) { 905 case BIO_READ: 906 case BIO_WRITE: 907 case BIO_DELETE: 908 break; 909 case BIO_GETATTR: 910 default: 911 g_io_deliver(bp, EOPNOTSUPP); 912 return; 913 } 914 mtx_lock(&sc->sc_queue_mtx); 915 bioq_disksort(&sc->sc_queue, bp); 916 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 917 wakeup(sc); 918 mtx_unlock(&sc->sc_queue_mtx); 919} 920 921/* 922 * Send one synchronization request. 923 */ 924static void 925g_mirror_sync_one(struct g_mirror_disk *disk) 926{ 927 struct g_mirror_softc *sc; 928 struct bio *bp; 929 930 sc = disk->d_softc; 931 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 932 ("Disk %s is not marked for synchronization.", 933 g_mirror_get_diskname(disk))); 934 935 bp = g_new_bio(); 936 if (bp == NULL) 937 return; 938 bp->bio_parent = NULL; 939 bp->bio_cmd = BIO_READ; 940 bp->bio_offset = disk->d_sync.ds_offset; 941 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 942 bp->bio_cflags = 0; 943 bp->bio_done = g_mirror_sync_done; 944 bp->bio_data = disk->d_sync.ds_data; 945 if (bp->bio_data == NULL) { 946 g_destroy_bio(bp); 947 return; 948 } 949 disk->d_sync.ds_offset += bp->bio_length; 950 bp->bio_to = sc->sc_provider; 951 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 952 disk->d_sync.ds_consumer->index++; 953 g_io_request(bp, disk->d_sync.ds_consumer); 954} 955 956static void 957g_mirror_sync_request(struct bio *bp) 958{ 959 struct g_mirror_softc *sc; 960 struct g_mirror_disk *disk; 961 962 bp->bio_from->index--; 963 sc = bp->bio_from->geom->softc; 964 disk = bp->bio_from->private; 965 if (disk == NULL) { 966 g_topology_lock(); 967 g_mirror_kill_consumer(sc, bp->bio_from); 968 g_topology_unlock(); 969 g_destroy_bio(bp); 970 return; 971 } 972 973 /* 974 * Synchronization request. 975 */ 976 switch (bp->bio_cmd) { 977 case BIO_READ: 978 { 979 struct g_consumer *cp; 980 981 if (bp->bio_error != 0) { 982 G_MIRROR_LOGREQ(0, bp, 983 "Synchronization request failed (error=%d).", 984 bp->bio_error); 985 g_destroy_bio(bp); 986 return; 987 } 988 G_MIRROR_LOGREQ(3, bp, 989 "Synchronization request half-finished."); 990 bp->bio_cmd = BIO_WRITE; 991 bp->bio_cflags = 0; 992 cp = disk->d_consumer; 993 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 994 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 995 cp->acr, cp->acw, cp->ace)); 996 cp->index++; 997 g_io_request(bp, cp); 998 return; 999 } 1000 case BIO_WRITE: 1001 { 1002 struct g_mirror_disk_sync *sync; 1003 1004 if (bp->bio_error != 0) { 1005 G_MIRROR_LOGREQ(0, bp, 1006 "Synchronization request failed (error=%d).", 1007 bp->bio_error); 1008 g_destroy_bio(bp); 1009 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 1010 g_mirror_event_send(disk, 1011 G_MIRROR_DISK_STATE_DISCONNECTED, 1012 G_MIRROR_EVENT_DONTWAIT); 1013 return; 1014 } 1015 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1016 sync = &disk->d_sync; 1017 sync->ds_offset_done = bp->bio_offset + bp->bio_length; 1018 g_destroy_bio(bp); 1019 if (sync->ds_resync != -1) 1020 break; 1021 if (sync->ds_offset_done == sc->sc_provider->mediasize) { 1022 /* 1023 * Disk up-to-date, activate it. 1024 */ 1025 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1026 G_MIRROR_EVENT_DONTWAIT); 1027 return; 1028 } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) { 1029 /* 1030 * Update offset_done on every 100 blocks. 1031 * XXX: This should be configurable. 1032 */ 1033 g_topology_lock(); 1034 g_mirror_update_metadata(disk); 1035 g_topology_unlock(); 1036 } 1037 return; 1038 } 1039 default: 1040 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1041 bp->bio_cmd, sc->sc_name)); 1042 break; 1043 } 1044} 1045 1046static void 1047g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1048{ 1049 struct g_mirror_disk *disk; 1050 struct g_consumer *cp; 1051 struct bio *cbp; 1052 1053 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1054 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1055 break; 1056 } 1057 if (disk == NULL) { 1058 if (bp->bio_error == 0) 1059 bp->bio_error = ENXIO; 1060 g_io_deliver(bp, bp->bio_error); 1061 return; 1062 } 1063 cbp = g_clone_bio(bp); 1064 if (cbp == NULL) { 1065 if (bp->bio_error == 0) 1066 bp->bio_error = ENOMEM; 1067 g_io_deliver(bp, bp->bio_error); 1068 return; 1069 } 1070 /* 1071 * Fill in the component buf structure. 1072 */ 1073 cp = disk->d_consumer; 1074 cbp->bio_done = g_mirror_done; 1075 cbp->bio_to = cp->provider; 1076 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1077 KASSERT(cp->acr > 0 && cp->ace > 0, 1078 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1079 cp->acw, cp->ace)); 1080 cp->index++; 1081 g_io_request(cbp, cp); 1082} 1083 1084static void 1085g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1086{ 1087 struct g_mirror_disk *disk; 1088 struct g_consumer *cp; 1089 struct bio *cbp; 1090 1091 disk = g_mirror_get_disk(sc); 1092 if (disk == NULL) { 1093 if (bp->bio_error == 0) 1094 bp->bio_error = ENXIO; 1095 g_io_deliver(bp, bp->bio_error); 1096 return; 1097 } 1098 cbp = g_clone_bio(bp); 1099 if (cbp == NULL) { 1100 if (bp->bio_error == 0) 1101 bp->bio_error = ENOMEM; 1102 g_io_deliver(bp, bp->bio_error); 1103 return; 1104 } 1105 /* 1106 * Fill in the component buf structure. 1107 */ 1108 cp = disk->d_consumer; 1109 cbp->bio_done = g_mirror_done; 1110 cbp->bio_to = cp->provider; 1111 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1112 KASSERT(cp->acr > 0 && cp->ace > 0, 1113 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1114 cp->acw, cp->ace)); 1115 cp->index++; 1116 g_io_request(cbp, cp); 1117} 1118 1119static void 1120g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1121{ 1122 struct g_mirror_disk *disk, *dp; 1123 struct g_consumer *cp; 1124 struct bio *cbp; 1125 struct bintime curtime; 1126 1127 binuptime(&curtime); 1128 /* 1129 * Find a disk which the smallest load. 1130 */ 1131 disk = NULL; 1132 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1133 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1134 continue; 1135 /* If disk wasn't used for more than 2 sec, use it. */ 1136 if (curtime.sec - dp->d_last_used.sec >= 2) { 1137 disk = dp; 1138 break; 1139 } 1140 if (disk == NULL || 1141 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1142 disk = dp; 1143 } 1144 } 1145 cbp = g_clone_bio(bp); 1146 if (cbp == NULL) { 1147 if (bp->bio_error == 0) 1148 bp->bio_error = ENOMEM; 1149 g_io_deliver(bp, bp->bio_error); 1150 return; 1151 } 1152 /* 1153 * Fill in the component buf structure. 1154 */ 1155 cp = disk->d_consumer; 1156 cbp->bio_done = g_mirror_done; 1157 cbp->bio_to = cp->provider; 1158 binuptime(&disk->d_last_used); 1159 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1160 KASSERT(cp->acr > 0 && cp->ace > 0, 1161 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1162 cp->acw, cp->ace)); 1163 cp->index++; 1164 g_io_request(cbp, cp); 1165} 1166 1167static void 1168g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1169{ 1170 struct bio_queue_head queue; 1171 struct g_mirror_disk *disk; 1172 struct g_consumer *cp; 1173 struct bio *cbp; 1174 off_t left, mod, offset, slice; 1175 u_char *data; 1176 u_int ndisks; 1177 1178 if (bp->bio_length <= sc->sc_slice) { 1179 g_mirror_request_round_robin(sc, bp); 1180 return; 1181 } 1182 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1183 slice = bp->bio_length / ndisks; 1184 mod = slice % sc->sc_provider->sectorsize; 1185 if (mod != 0) 1186 slice += sc->sc_provider->sectorsize - mod; 1187 /* 1188 * Allocate all bios before sending any request, so we can 1189 * return ENOMEM in nice and clean way. 1190 */ 1191 left = bp->bio_length; 1192 offset = bp->bio_offset; 1193 data = bp->bio_data; 1194 bioq_init(&queue); 1195 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1196 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1197 continue; 1198 cbp = g_clone_bio(bp); 1199 if (cbp == NULL) { 1200 for (cbp = bioq_first(&queue); cbp != NULL; 1201 cbp = bioq_first(&queue)) { 1202 bioq_remove(&queue, cbp); 1203 g_destroy_bio(cbp); 1204 } 1205 if (bp->bio_error == 0) 1206 bp->bio_error = ENOMEM; 1207 g_io_deliver(bp, bp->bio_error); 1208 return; 1209 } 1210 bioq_insert_tail(&queue, cbp); 1211 cbp->bio_done = g_mirror_done; 1212 cbp->bio_caller1 = disk; 1213 cbp->bio_to = disk->d_consumer->provider; 1214 cbp->bio_offset = offset; 1215 cbp->bio_data = data; 1216 cbp->bio_length = MIN(left, slice); 1217 left -= cbp->bio_length; 1218 if (left == 0) 1219 break; 1220 offset += cbp->bio_length; 1221 data += cbp->bio_length; 1222 } 1223 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1224 bioq_remove(&queue, cbp); 1225 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1226 disk = cbp->bio_caller1; 1227 cbp->bio_caller1 = NULL; 1228 cp = disk->d_consumer; 1229 KASSERT(cp->acr > 0 && cp->ace > 0, 1230 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1231 cp->acr, cp->acw, cp->ace)); 1232 disk->d_consumer->index++; 1233 g_io_request(cbp, disk->d_consumer); 1234 } 1235} 1236 1237static void 1238g_mirror_register_request(struct bio *bp) 1239{ 1240 struct g_mirror_softc *sc; 1241 1242 sc = bp->bio_to->geom->softc; 1243 switch (bp->bio_cmd) { 1244 case BIO_READ: 1245 switch (sc->sc_balance) { 1246 case G_MIRROR_BALANCE_LOAD: 1247 g_mirror_request_load(sc, bp); 1248 break; 1249 case G_MIRROR_BALANCE_PREFER: 1250 g_mirror_request_prefer(sc, bp); 1251 break; 1252 case G_MIRROR_BALANCE_ROUND_ROBIN: 1253 g_mirror_request_round_robin(sc, bp); 1254 break; 1255 case G_MIRROR_BALANCE_SPLIT: 1256 g_mirror_request_split(sc, bp); 1257 break; 1258 } 1259 return; 1260 case BIO_WRITE: 1261 case BIO_DELETE: 1262 { 1263 struct g_mirror_disk *disk; 1264 struct g_mirror_disk_sync *sync; 1265 struct bio_queue_head queue; 1266 struct g_consumer *cp; 1267 struct bio *cbp; 1268 1269 if (sc->sc_idle) 1270 g_mirror_unidle(sc); 1271 /* 1272 * Allocate all bios before sending any request, so we can 1273 * return ENOMEM in nice and clean way. 1274 */ 1275 bioq_init(&queue); 1276 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1277 sync = &disk->d_sync; 1278 switch (disk->d_state) { 1279 case G_MIRROR_DISK_STATE_ACTIVE: 1280 break; 1281 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1282 if (bp->bio_offset >= sync->ds_offset) 1283 continue; 1284 else if (bp->bio_offset + bp->bio_length > 1285 sync->ds_offset_done && 1286 (bp->bio_offset < sync->ds_resync || 1287 sync->ds_resync == -1)) { 1288 sync->ds_resync = bp->bio_offset - 1289 (bp->bio_offset % MAXPHYS); 1290 } 1291 break; 1292 default: 1293 continue; 1294 } 1295 cbp = g_clone_bio(bp); 1296 if (cbp == NULL) { 1297 for (cbp = bioq_first(&queue); cbp != NULL; 1298 cbp = bioq_first(&queue)) { 1299 bioq_remove(&queue, cbp); 1300 g_destroy_bio(cbp); 1301 } 1302 if (bp->bio_error == 0) 1303 bp->bio_error = ENOMEM; 1304 g_io_deliver(bp, bp->bio_error); 1305 return; 1306 } 1307 bioq_insert_tail(&queue, cbp); 1308 cbp->bio_done = g_mirror_done; 1309 cp = disk->d_consumer; 1310 cbp->bio_caller1 = cp; 1311 cbp->bio_to = cp->provider; 1312 KASSERT(cp->acw > 0 && cp->ace > 0, 1313 ("Consumer %s not opened (r%dw%de%d).", 1314 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1315 } 1316 for (cbp = bioq_first(&queue); cbp != NULL; 1317 cbp = bioq_first(&queue)) { 1318 bioq_remove(&queue, cbp); 1319 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1320 cp = cbp->bio_caller1; 1321 cbp->bio_caller1 = NULL; 1322 cp->index++; 1323 g_io_request(cbp, cp); 1324 } 1325 /* 1326 * Bump syncid on first write. 1327 */ 1328 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) { 1329 sc->sc_bump_syncid = 0; 1330 g_topology_lock(); 1331 g_mirror_bump_syncid(sc); 1332 g_topology_unlock(); 1333 } 1334 return; 1335 } 1336 default: 1337 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1338 bp->bio_cmd, sc->sc_name)); 1339 break; 1340 } 1341} 1342 1343static int 1344g_mirror_can_destroy(struct g_mirror_softc *sc) 1345{ 1346 struct g_geom *gp; 1347 struct g_consumer *cp; 1348 1349 g_topology_assert(); 1350 gp = sc->sc_geom; 1351 LIST_FOREACH(cp, &gp->consumer, consumer) { 1352 if (g_mirror_is_busy(sc, cp)) 1353 return (0); 1354 } 1355 gp = sc->sc_sync.ds_geom; 1356 LIST_FOREACH(cp, &gp->consumer, consumer) { 1357 if (g_mirror_is_busy(sc, cp)) 1358 return (0); 1359 } 1360 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1361 sc->sc_name); 1362 return (1); 1363} 1364 1365static int 1366g_mirror_try_destroy(struct g_mirror_softc *sc) 1367{ 1368 1369 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1370 g_topology_lock(); 1371 if (!g_mirror_can_destroy(sc)) { 1372 g_topology_unlock(); 1373 return (0); 1374 } 1375 g_topology_unlock(); 1376 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1377 &sc->sc_worker); 1378 wakeup(&sc->sc_worker); 1379 sc->sc_worker = NULL; 1380 } else { 1381 g_topology_lock(); 1382 if (!g_mirror_can_destroy(sc)) { 1383 g_topology_unlock(); 1384 return (0); 1385 } 1386 g_mirror_destroy_device(sc); 1387 g_topology_unlock(); 1388 free(sc, M_MIRROR); 1389 } 1390 return (1); 1391} 1392 1393/* 1394 * Worker thread. 1395 */ 1396static void 1397g_mirror_worker(void *arg) 1398{ 1399 struct g_mirror_softc *sc; 1400 struct g_mirror_disk *disk; 1401 struct g_mirror_disk_sync *sync; 1402 struct g_mirror_event *ep; 1403 struct bio *bp; 1404 u_int nreqs; 1405 1406 sc = arg; 1407 curthread->td_base_pri = PRIBIO; 1408 1409 nreqs = 0; 1410 for (;;) { 1411 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1412 /* 1413 * First take a look at events. 1414 * This is important to handle events before any I/O requests. 1415 */ 1416 ep = g_mirror_event_get(sc); 1417 if (ep != NULL) { 1418 g_topology_lock(); 1419 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1420 /* Update only device status. */ 1421 G_MIRROR_DEBUG(3, 1422 "Running event for device %s.", 1423 sc->sc_name); 1424 ep->e_error = 0; 1425 g_mirror_update_device(sc, 1); 1426 } else { 1427 /* Update disk status. */ 1428 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1429 g_mirror_get_diskname(ep->e_disk)); 1430 ep->e_error = g_mirror_update_disk(ep->e_disk, 1431 ep->e_state); 1432 if (ep->e_error == 0) 1433 g_mirror_update_device(sc, 0); 1434 } 1435 g_topology_unlock(); 1436 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1437 KASSERT(ep->e_error == 0, 1438 ("Error cannot be handled.")); 1439 g_mirror_event_free(ep); 1440 } else { 1441 ep->e_flags |= G_MIRROR_EVENT_DONE; 1442 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1443 ep); 1444 mtx_lock(&sc->sc_events_mtx); 1445 wakeup(ep); 1446 mtx_unlock(&sc->sc_events_mtx); 1447 } 1448 if ((sc->sc_flags & 1449 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1450 if (g_mirror_try_destroy(sc)) 1451 kthread_exit(0); 1452 } 1453 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1454 continue; 1455 } 1456 /* 1457 * Now I/O requests. 1458 */ 1459 /* Get first request from the queue. */ 1460 mtx_lock(&sc->sc_queue_mtx); 1461 bp = bioq_first(&sc->sc_queue); 1462 if (bp == NULL) { 1463 if ((sc->sc_flags & 1464 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1465 mtx_unlock(&sc->sc_queue_mtx); 1466 if (g_mirror_try_destroy(sc)) 1467 kthread_exit(0); 1468 mtx_lock(&sc->sc_queue_mtx); 1469 } 1470 } 1471 if (sc->sc_sync.ds_ndisks > 0 && 1472 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1473 mtx_unlock(&sc->sc_queue_mtx); 1474 /* 1475 * It is time for synchronization... 1476 */ 1477 nreqs = 0; 1478 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1479 if (disk->d_state != 1480 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1481 continue; 1482 } 1483 sync = &disk->d_sync; 1484 if (sync->ds_offset >= 1485 sc->sc_provider->mediasize) { 1486 continue; 1487 } 1488 if (sync->ds_offset > sync->ds_offset_done) 1489 continue; 1490 if (sync->ds_resync != -1) { 1491 sync->ds_offset = sync->ds_resync; 1492 sync->ds_offset_done = sync->ds_resync; 1493 sync->ds_resync = -1; 1494 } 1495 g_mirror_sync_one(disk); 1496 } 1497 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1498 goto sleep; 1499 } 1500 if (bp == NULL) { 1501#define G_MIRROR_IS_IDLE(sc) ((sc)->sc_idle || \ 1502 ((sc)->sc_provider != NULL && \ 1503 (sc)->sc_provider->acw == 0)) 1504 if (G_MIRROR_IS_IDLE(sc)) { 1505 /* 1506 * If we're already in idle state, sleep without 1507 * a timeout. 1508 */ 1509 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1510 "m:w1", 0); 1511 G_MIRROR_DEBUG(5, "%s: I'm here 3.", __func__); 1512 } else { 1513 u_int idletime; 1514 1515 idletime = g_mirror_idletime; 1516 if (idletime == 0) 1517 idletime = 1; 1518 idletime *= hz; 1519 if (msleep(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, 1520 "m:w2", idletime) == EWOULDBLOCK) { 1521 G_MIRROR_DEBUG(5, "%s: I'm here 4.", 1522 __func__); 1523 /* 1524 * No I/O requests in 5 seconds, so mark 1525 * components as clean. 1526 */ 1527 g_mirror_idle(sc); 1528 } 1529 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1530 } 1531 continue; 1532 } 1533 nreqs++; 1534 bioq_remove(&sc->sc_queue, bp); 1535 mtx_unlock(&sc->sc_queue_mtx); 1536 1537 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1538 g_mirror_regular_request(bp); 1539 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1540 u_int timeout, sps; 1541 1542 g_mirror_sync_request(bp); 1543sleep: 1544 sps = g_mirror_syncs_per_sec; 1545 if (sps == 0) { 1546 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1547 continue; 1548 } 1549 mtx_lock(&sc->sc_queue_mtx); 1550 if (bioq_first(&sc->sc_queue) != NULL) { 1551 mtx_unlock(&sc->sc_queue_mtx); 1552 G_MIRROR_DEBUG(5, "%s: I'm here 7.", __func__); 1553 continue; 1554 } 1555 timeout = hz / sps; 1556 if (timeout == 0) 1557 timeout = 1; 1558 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w3", 1559 timeout); 1560 } else { 1561 g_mirror_register_request(bp); 1562 } 1563 G_MIRROR_DEBUG(5, "%s: I'm here 8.", __func__); 1564 } 1565} 1566 1567/* 1568 * Open disk's consumer if needed. 1569 */ 1570static void 1571g_mirror_update_access(struct g_mirror_disk *disk) 1572{ 1573 struct g_provider *pp; 1574 struct g_consumer *cp; 1575 int acr, acw, ace, cpw, error; 1576 1577 g_topology_assert(); 1578 1579 cp = disk->d_consumer; 1580 pp = disk->d_softc->sc_provider; 1581 if (pp == NULL) { 1582 acr = -cp->acr; 1583 acw = -cp->acw; 1584 ace = -cp->ace; 1585 } else { 1586 acr = pp->acr - cp->acr; 1587 acw = pp->acw - cp->acw; 1588 ace = pp->ace - cp->ace; 1589 /* Grab an extra "exclusive" bit. */ 1590 if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) 1591 ace++; 1592 } 1593 if (acr == 0 && acw == 0 && ace == 0) 1594 return; 1595 cpw = cp->acw; 1596 error = g_access(cp, acr, acw, ace); 1597 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, acr, 1598 acw, ace, error); 1599 if (error != 0) { 1600 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1601 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1602 G_MIRROR_EVENT_DONTWAIT); 1603 return; 1604 } 1605 if (cpw == 0 && cp->acw > 0) { 1606 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1607 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1608 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1609 } else if (cpw > 0 && cp->acw == 0) { 1610 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1611 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1612 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1613 } 1614} 1615 1616static void 1617g_mirror_sync_start(struct g_mirror_disk *disk) 1618{ 1619 struct g_mirror_softc *sc; 1620 struct g_consumer *cp; 1621 int error; 1622 1623 g_topology_assert(); 1624 1625 sc = disk->d_softc; 1626 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1627 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1628 sc->sc_state)); 1629 cp = disk->d_consumer; 1630 KASSERT(cp->acr == 0 && cp->acw == 0 && cp->ace == 0, 1631 ("Consumer %s already opened.", cp->provider->name)); 1632 1633 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1634 g_mirror_get_diskname(disk)); 1635 error = g_access(cp, 0, 1, 1); 1636 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, 1, 1637 1, error); 1638 if (error != 0) { 1639 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1640 G_MIRROR_EVENT_DONTWAIT); 1641 return; 1642 } 1643 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1644 KASSERT(disk->d_sync.ds_consumer == NULL, 1645 ("Sync consumer already exists (device=%s, disk=%s).", 1646 sc->sc_name, g_mirror_get_diskname(disk))); 1647 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1648 disk->d_sync.ds_consumer->private = disk; 1649 disk->d_sync.ds_consumer->index = 0; 1650 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1651 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1652 disk->d_softc->sc_name, error)); 1653 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1654 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1655 disk->d_softc->sc_name, error)); 1656 disk->d_sync.ds_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1657 sc->sc_sync.ds_ndisks++; 1658} 1659 1660/* 1661 * Stop synchronization process. 1662 * type: 0 - synchronization finished 1663 * 1 - synchronization stopped 1664 */ 1665static void 1666g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1667{ 1668 struct g_consumer *cp; 1669 1670 g_topology_assert(); 1671 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1672 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1673 g_mirror_disk_state2str(disk->d_state))); 1674 if (disk->d_sync.ds_consumer == NULL) 1675 return; 1676 1677 if (type == 0) { 1678 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1679 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1680 } else /* if (type == 1) */ { 1681 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1682 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1683 } 1684 cp = disk->d_sync.ds_consumer; 1685 g_access(cp, -1, 0, 0); 1686 g_mirror_kill_consumer(disk->d_softc, cp); 1687 free(disk->d_sync.ds_data, M_MIRROR); 1688 disk->d_sync.ds_consumer = NULL; 1689 disk->d_softc->sc_sync.ds_ndisks--; 1690 cp = disk->d_consumer; 1691 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 1692 ("Consumer %s not opened.", cp->provider->name)); 1693 g_access(cp, 0, -1, -1); 1694 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, -1, 1695 -1, 0); 1696 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1697} 1698 1699static void 1700g_mirror_launch_provider(struct g_mirror_softc *sc) 1701{ 1702 struct g_mirror_disk *disk; 1703 struct g_provider *pp; 1704 1705 g_topology_assert(); 1706 1707 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1708 pp->mediasize = sc->sc_mediasize; 1709 pp->sectorsize = sc->sc_sectorsize; 1710 sc->sc_provider = pp; 1711 g_error_provider(pp, 0); 1712 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1713 pp->name); 1714 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1715 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1716 g_mirror_sync_start(disk); 1717 } 1718} 1719 1720static void 1721g_mirror_destroy_provider(struct g_mirror_softc *sc) 1722{ 1723 struct g_mirror_disk *disk; 1724 struct bio *bp; 1725 1726 g_topology_assert(); 1727 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1728 sc->sc_name)); 1729 1730 g_error_provider(sc->sc_provider, ENXIO); 1731 mtx_lock(&sc->sc_queue_mtx); 1732 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1733 bioq_remove(&sc->sc_queue, bp); 1734 g_io_deliver(bp, ENXIO); 1735 } 1736 mtx_unlock(&sc->sc_queue_mtx); 1737 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1738 sc->sc_provider->name); 1739 sc->sc_provider->flags |= G_PF_WITHER; 1740 g_orphan_provider(sc->sc_provider, ENXIO); 1741 sc->sc_provider = NULL; 1742 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1743 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1744 g_mirror_sync_stop(disk, 1); 1745 } 1746} 1747 1748static void 1749g_mirror_go(void *arg) 1750{ 1751 struct g_mirror_softc *sc; 1752 1753 sc = arg; 1754 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1755 g_mirror_event_send(sc, 0, 1756 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1757} 1758 1759static u_int 1760g_mirror_determine_state(struct g_mirror_disk *disk) 1761{ 1762 struct g_mirror_softc *sc; 1763 u_int state; 1764 1765 sc = disk->d_softc; 1766 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1767 if ((disk->d_flags & 1768 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1769 /* Disk does not need synchronization. */ 1770 state = G_MIRROR_DISK_STATE_ACTIVE; 1771 } else { 1772 if ((sc->sc_flags & 1773 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1774 (disk->d_flags & 1775 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1776 /* 1777 * We can start synchronization from 1778 * the stored offset. 1779 */ 1780 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1781 } else { 1782 state = G_MIRROR_DISK_STATE_STALE; 1783 } 1784 } 1785 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1786 /* 1787 * Reset all synchronization data for this disk, 1788 * because if it even was synchronized, it was 1789 * synchronized to disks with different syncid. 1790 */ 1791 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1792 disk->d_sync.ds_offset = 0; 1793 disk->d_sync.ds_offset_done = 0; 1794 disk->d_sync.ds_syncid = sc->sc_syncid; 1795 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1796 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1797 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1798 } else { 1799 state = G_MIRROR_DISK_STATE_STALE; 1800 } 1801 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1802 /* 1803 * Not good, NOT GOOD! 1804 * It means that mirror was started on stale disks 1805 * and more fresh disk just arrive. 1806 * If there were writes, mirror is fucked up, sorry. 1807 * I think the best choice here is don't touch 1808 * this disk and inform the user laudly. 1809 */ 1810 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1811 "disk (%s) arrives!! It will not be connected to the " 1812 "running device.", sc->sc_name, 1813 g_mirror_get_diskname(disk)); 1814 g_mirror_destroy_disk(disk); 1815 state = G_MIRROR_DISK_STATE_NONE; 1816 /* Return immediately, because disk was destroyed. */ 1817 return (state); 1818 } 1819 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1820 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1821 return (state); 1822} 1823 1824/* 1825 * Update device state. 1826 */ 1827static void 1828g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1829{ 1830 struct g_mirror_disk *disk; 1831 u_int state; 1832 1833 g_topology_assert(); 1834 1835 switch (sc->sc_state) { 1836 case G_MIRROR_DEVICE_STATE_STARTING: 1837 { 1838 struct g_mirror_disk *pdisk; 1839 u_int dirty, ndisks, syncid; 1840 1841 KASSERT(sc->sc_provider == NULL, 1842 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1843 /* 1844 * Are we ready? We are, if all disks are connected or 1845 * if we have any disks and 'force' is true. 1846 */ 1847 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1848 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1849 ; 1850 } else if (g_mirror_ndisks(sc, -1) == 0) { 1851 /* 1852 * Disks went down in starting phase, so destroy 1853 * device. 1854 */ 1855 callout_drain(&sc->sc_callout); 1856 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1857 return; 1858 } else { 1859 return; 1860 } 1861 1862 /* 1863 * Activate all disks with the biggest syncid. 1864 */ 1865 if (force) { 1866 /* 1867 * If 'force' is true, we have been called due to 1868 * timeout, so don't bother canceling timeout. 1869 */ 1870 ndisks = 0; 1871 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1872 if ((disk->d_flags & 1873 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1874 ndisks++; 1875 } 1876 } 1877 if (ndisks == 0) { 1878 /* No valid disks found, destroy device. */ 1879 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1880 return; 1881 } 1882 } else { 1883 /* Cancel timeout. */ 1884 callout_drain(&sc->sc_callout); 1885 } 1886 1887 /* 1888 * Find disk with the biggest syncid. 1889 */ 1890 syncid = 0; 1891 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1892 if (disk->d_sync.ds_syncid > syncid) 1893 syncid = disk->d_sync.ds_syncid; 1894 } 1895 1896 /* 1897 * Here we need to look for dirty disks and if all disks 1898 * with the biggest syncid are dirty, we have to choose 1899 * one with the biggest priority and rebuild the rest. 1900 */ 1901 /* 1902 * Find the number of dirty disks with the biggest syncid. 1903 * Find the number of disks with the biggest syncid. 1904 * While here, find a disk with the biggest priority. 1905 */ 1906 dirty = ndisks = 0; 1907 pdisk = NULL; 1908 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1909 if (disk->d_sync.ds_syncid != syncid) 1910 continue; 1911 if ((disk->d_flags & 1912 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1913 continue; 1914 } 1915 ndisks++; 1916 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1917 dirty++; 1918 if (pdisk == NULL || 1919 pdisk->d_priority < disk->d_priority) { 1920 pdisk = disk; 1921 } 1922 } 1923 } 1924 if (dirty == 0) { 1925 /* No dirty disks at all, great. */ 1926 } else if (dirty == ndisks) { 1927 /* 1928 * Force synchronization for all dirty disks except one 1929 * with the biggest priority. 1930 */ 1931 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1932 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1933 "master disk for synchronization.", 1934 g_mirror_get_diskname(pdisk), sc->sc_name); 1935 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1936 if (disk->d_sync.ds_syncid != syncid) 1937 continue; 1938 if ((disk->d_flags & 1939 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1940 continue; 1941 } 1942 KASSERT((disk->d_flags & 1943 G_MIRROR_DISK_FLAG_DIRTY) != 0, 1944 ("Disk %s isn't marked as dirty.", 1945 g_mirror_get_diskname(disk))); 1946 /* Skip the disk with the biggest priority. */ 1947 if (disk == pdisk) 1948 continue; 1949 disk->d_sync.ds_syncid = 0; 1950 } 1951 } else if (dirty < ndisks) { 1952 /* 1953 * Force synchronization for all dirty disks. 1954 * We have some non-dirty disks. 1955 */ 1956 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1957 if (disk->d_sync.ds_syncid != syncid) 1958 continue; 1959 if ((disk->d_flags & 1960 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1961 continue; 1962 } 1963 if ((disk->d_flags & 1964 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1965 continue; 1966 } 1967 disk->d_sync.ds_syncid = 0; 1968 } 1969 } 1970 1971 /* Reset hint. */ 1972 sc->sc_hint = NULL; 1973 sc->sc_syncid = syncid; 1974 if (force) { 1975 /* Remember to bump syncid on first write. */ 1976 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1977 } 1978 state = G_MIRROR_DEVICE_STATE_RUNNING; 1979 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 1980 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 1981 g_mirror_device_state2str(state)); 1982 sc->sc_state = state; 1983 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1984 state = g_mirror_determine_state(disk); 1985 g_mirror_event_send(disk, state, 1986 G_MIRROR_EVENT_DONTWAIT); 1987 if (state == G_MIRROR_DISK_STATE_STALE) { 1988 sc->sc_bump_syncid = 1989 G_MIRROR_BUMP_ON_FIRST_WRITE; 1990 } 1991 } 1992 wakeup(&g_mirror_class); 1993 break; 1994 } 1995 case G_MIRROR_DEVICE_STATE_RUNNING: 1996 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 1997 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 1998 /* 1999 * No active disks or no disks at all, 2000 * so destroy device. 2001 */ 2002 if (sc->sc_provider != NULL) 2003 g_mirror_destroy_provider(sc); 2004 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2005 break; 2006 } else if (g_mirror_ndisks(sc, 2007 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2008 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2009 /* 2010 * We have active disks, launch provider if it doesn't 2011 * exist. 2012 */ 2013 if (sc->sc_provider == NULL) 2014 g_mirror_launch_provider(sc); 2015 } 2016 /* 2017 * Bump syncid here, if we need to do it immediately. 2018 */ 2019 if (sc->sc_bump_syncid == G_MIRROR_BUMP_IMMEDIATELY) { 2020 sc->sc_bump_syncid = 0; 2021 g_mirror_bump_syncid(sc); 2022 } 2023 break; 2024 default: 2025 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2026 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2027 break; 2028 } 2029} 2030 2031/* 2032 * Update disk state and device state if needed. 2033 */ 2034#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2035 "Disk %s state changed from %s to %s (device %s).", \ 2036 g_mirror_get_diskname(disk), \ 2037 g_mirror_disk_state2str(disk->d_state), \ 2038 g_mirror_disk_state2str(state), sc->sc_name) 2039static int 2040g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2041{ 2042 struct g_mirror_softc *sc; 2043 2044 g_topology_assert(); 2045 2046 sc = disk->d_softc; 2047again: 2048 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2049 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2050 g_mirror_disk_state2str(state)); 2051 switch (state) { 2052 case G_MIRROR_DISK_STATE_NEW: 2053 /* 2054 * Possible scenarios: 2055 * 1. New disk arrive. 2056 */ 2057 /* Previous state should be NONE. */ 2058 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2059 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2060 g_mirror_disk_state2str(disk->d_state))); 2061 DISK_STATE_CHANGED(); 2062 2063 disk->d_state = state; 2064 if (LIST_EMPTY(&sc->sc_disks)) 2065 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2066 else { 2067 struct g_mirror_disk *dp; 2068 2069 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2070 if (disk->d_priority >= dp->d_priority) { 2071 LIST_INSERT_BEFORE(dp, disk, d_next); 2072 dp = NULL; 2073 break; 2074 } 2075 if (LIST_NEXT(dp, d_next) == NULL) 2076 break; 2077 } 2078 if (dp != NULL) 2079 LIST_INSERT_AFTER(dp, disk, d_next); 2080 } 2081 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2082 sc->sc_name, g_mirror_get_diskname(disk)); 2083 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2084 break; 2085 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2086 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2087 g_mirror_device_state2str(sc->sc_state), 2088 g_mirror_get_diskname(disk), 2089 g_mirror_disk_state2str(disk->d_state))); 2090 state = g_mirror_determine_state(disk); 2091 if (state != G_MIRROR_DISK_STATE_NONE) 2092 goto again; 2093 break; 2094 case G_MIRROR_DISK_STATE_ACTIVE: 2095 /* 2096 * Possible scenarios: 2097 * 1. New disk does not need synchronization. 2098 * 2. Synchronization process finished successfully. 2099 */ 2100 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2101 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2102 g_mirror_device_state2str(sc->sc_state), 2103 g_mirror_get_diskname(disk), 2104 g_mirror_disk_state2str(disk->d_state))); 2105 /* Previous state should be NEW or SYNCHRONIZING. */ 2106 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2107 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2108 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2109 g_mirror_disk_state2str(disk->d_state))); 2110 DISK_STATE_CHANGED(); 2111 2112 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2113 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2114 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2115 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2116 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2117 g_mirror_sync_stop(disk, 0); 2118 } 2119 disk->d_state = state; 2120 disk->d_sync.ds_offset = 0; 2121 disk->d_sync.ds_offset_done = 0; 2122 g_mirror_update_access(disk); 2123 g_mirror_update_metadata(disk); 2124 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2125 sc->sc_name, g_mirror_get_diskname(disk)); 2126 break; 2127 case G_MIRROR_DISK_STATE_STALE: 2128 /* 2129 * Possible scenarios: 2130 * 1. Stale disk was connected. 2131 */ 2132 /* Previous state should be NEW. */ 2133 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2134 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2135 g_mirror_disk_state2str(disk->d_state))); 2136 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2137 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2138 g_mirror_device_state2str(sc->sc_state), 2139 g_mirror_get_diskname(disk), 2140 g_mirror_disk_state2str(disk->d_state))); 2141 /* 2142 * STALE state is only possible if device is marked 2143 * NOAUTOSYNC. 2144 */ 2145 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2146 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2147 g_mirror_device_state2str(sc->sc_state), 2148 g_mirror_get_diskname(disk), 2149 g_mirror_disk_state2str(disk->d_state))); 2150 DISK_STATE_CHANGED(); 2151 2152 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2153 disk->d_state = state; 2154 g_mirror_update_metadata(disk); 2155 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2156 sc->sc_name, g_mirror_get_diskname(disk)); 2157 break; 2158 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2159 /* 2160 * Possible scenarios: 2161 * 1. Disk which needs synchronization was connected. 2162 */ 2163 /* Previous state should be NEW. */ 2164 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2165 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2166 g_mirror_disk_state2str(disk->d_state))); 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 DISK_STATE_CHANGED(); 2173 2174 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2175 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2176 disk->d_state = state; 2177 if (sc->sc_provider != NULL) { 2178 g_mirror_sync_start(disk); 2179 g_mirror_update_metadata(disk); 2180 } 2181 break; 2182 case G_MIRROR_DISK_STATE_DISCONNECTED: 2183 /* 2184 * Possible scenarios: 2185 * 1. Device wasn't running yet, but disk disappear. 2186 * 2. Disk was active and disapppear. 2187 * 3. Disk disappear during synchronization process. 2188 */ 2189 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2190 /* 2191 * Previous state should be ACTIVE, STALE or 2192 * SYNCHRONIZING. 2193 */ 2194 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2195 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2196 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2197 ("Wrong disk state (%s, %s).", 2198 g_mirror_get_diskname(disk), 2199 g_mirror_disk_state2str(disk->d_state))); 2200 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2201 /* Previous state should be NEW. */ 2202 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2203 ("Wrong disk state (%s, %s).", 2204 g_mirror_get_diskname(disk), 2205 g_mirror_disk_state2str(disk->d_state))); 2206 /* 2207 * Reset bumping syncid if disk disappeared in STARTING 2208 * state. 2209 */ 2210 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) 2211 sc->sc_bump_syncid = 0; 2212#ifdef INVARIANTS 2213 } else { 2214 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2215 sc->sc_name, 2216 g_mirror_device_state2str(sc->sc_state), 2217 g_mirror_get_diskname(disk), 2218 g_mirror_disk_state2str(disk->d_state))); 2219#endif 2220 } 2221 DISK_STATE_CHANGED(); 2222 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2223 sc->sc_name, g_mirror_get_diskname(disk)); 2224 2225 g_mirror_destroy_disk(disk); 2226 break; 2227 case G_MIRROR_DISK_STATE_DESTROY: 2228 { 2229 int error; 2230 2231 error = g_mirror_clear_metadata(disk); 2232 if (error != 0) 2233 return (error); 2234 DISK_STATE_CHANGED(); 2235 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2236 sc->sc_name, g_mirror_get_diskname(disk)); 2237 2238 g_mirror_destroy_disk(disk); 2239 sc->sc_ndisks--; 2240 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2241 g_mirror_update_metadata(disk); 2242 } 2243 break; 2244 } 2245 default: 2246 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2247 break; 2248 } 2249 return (0); 2250} 2251#undef DISK_STATE_CHANGED 2252 2253static int 2254g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2255{ 2256 struct g_provider *pp; 2257 u_char *buf; 2258 int error; 2259 2260 g_topology_assert(); 2261 2262 error = g_access(cp, 1, 0, 0); 2263 if (error != 0) 2264 return (error); 2265 pp = cp->provider; 2266 g_topology_unlock(); 2267 /* Metadata are stored on last sector. */ 2268 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2269 &error); 2270 g_topology_lock(); 2271 if (buf == NULL) { 2272 g_access(cp, -1, 0, 0); 2273 return (error); 2274 } 2275 if (error != 0) { 2276 g_access(cp, -1, 0, 0); 2277 g_free(buf); 2278 return (error); 2279 } 2280 error = g_access(cp, -1, 0, 0); 2281 KASSERT(error == 0, ("Cannot decrease access count for %s.", pp->name)); 2282 2283 /* Decode metadata. */ 2284 error = mirror_metadata_decode(buf, md); 2285 g_free(buf); 2286 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2287 return (EINVAL); 2288 if (error != 0) { 2289 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2290 cp->provider->name); 2291 return (error); 2292 } 2293 2294 return (0); 2295} 2296 2297static int 2298g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2299 struct g_mirror_metadata *md) 2300{ 2301 2302 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2303 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2304 pp->name, md->md_did); 2305 return (EEXIST); 2306 } 2307 if (md->md_all != sc->sc_ndisks) { 2308 G_MIRROR_DEBUG(1, 2309 "Invalid '%s' field on disk %s (device %s), skipping.", 2310 "md_all", pp->name, sc->sc_name); 2311 return (EINVAL); 2312 } 2313 if (md->md_slice != sc->sc_slice) { 2314 G_MIRROR_DEBUG(1, 2315 "Invalid '%s' field on disk %s (device %s), skipping.", 2316 "md_slice", pp->name, sc->sc_name); 2317 return (EINVAL); 2318 } 2319 if (md->md_balance != sc->sc_balance) { 2320 G_MIRROR_DEBUG(1, 2321 "Invalid '%s' field on disk %s (device %s), skipping.", 2322 "md_balance", pp->name, sc->sc_name); 2323 return (EINVAL); 2324 } 2325 if (md->md_mediasize != sc->sc_mediasize) { 2326 G_MIRROR_DEBUG(1, 2327 "Invalid '%s' field on disk %s (device %s), skipping.", 2328 "md_mediasize", pp->name, sc->sc_name); 2329 return (EINVAL); 2330 } 2331 if (sc->sc_mediasize > pp->mediasize) { 2332 G_MIRROR_DEBUG(1, 2333 "Invalid size of disk %s (device %s), skipping.", pp->name, 2334 sc->sc_name); 2335 return (EINVAL); 2336 } 2337 if (md->md_sectorsize != sc->sc_sectorsize) { 2338 G_MIRROR_DEBUG(1, 2339 "Invalid '%s' field on disk %s (device %s), skipping.", 2340 "md_sectorsize", pp->name, sc->sc_name); 2341 return (EINVAL); 2342 } 2343 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2344 G_MIRROR_DEBUG(1, 2345 "Invalid sector size of disk %s (device %s), skipping.", 2346 pp->name, sc->sc_name); 2347 return (EINVAL); 2348 } 2349 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2350 G_MIRROR_DEBUG(1, 2351 "Invalid device flags on disk %s (device %s), skipping.", 2352 pp->name, sc->sc_name); 2353 return (EINVAL); 2354 } 2355 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2356 G_MIRROR_DEBUG(1, 2357 "Invalid disk flags on disk %s (device %s), skipping.", 2358 pp->name, sc->sc_name); 2359 return (EINVAL); 2360 } 2361 return (0); 2362} 2363 2364static int 2365g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2366 struct g_mirror_metadata *md) 2367{ 2368 struct g_mirror_disk *disk; 2369 int error; 2370 2371 g_topology_assert(); 2372 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2373 2374 error = g_mirror_check_metadata(sc, pp, md); 2375 if (error != 0) 2376 return (error); 2377 disk = g_mirror_init_disk(sc, pp, md, &error); 2378 if (disk == NULL) 2379 return (error); 2380 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2381 G_MIRROR_EVENT_WAIT); 2382 return (error); 2383} 2384 2385static int 2386g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2387{ 2388 struct g_mirror_softc *sc; 2389 struct g_mirror_disk *disk; 2390 int dcr, dcw, dce, err, error; 2391 2392 g_topology_assert(); 2393 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2394 acw, ace); 2395 2396 dcr = pp->acr + acr; 2397 dcw = pp->acw + acw; 2398 dce = pp->ace + ace; 2399 2400 /* On first open, grab an extra "exclusive" bit */ 2401 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0) 2402 ace++; 2403 /* ... and let go of it on last close */ 2404 if (dcr == 0 && dcw == 0 && dce == 0) 2405 ace--; 2406 2407 sc = pp->geom->softc; 2408 if (sc == NULL || LIST_EMPTY(&sc->sc_disks)) { 2409 if (acr <= 0 && acw <= 0 && ace <= 0) 2410 return (0); 2411 else 2412 return (ENXIO); 2413 } 2414 error = ENXIO; 2415 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2416 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2417 continue; 2418 err = g_access(disk->d_consumer, acr, acw, ace); 2419 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 2420 g_mirror_get_diskname(disk), acr, acw, ace, err); 2421 if (err == 0) { 2422 /* 2423 * Mark disk as dirty on open and unmark on close. 2424 */ 2425 if (pp->acw == 0 && dcw > 0) { 2426 G_MIRROR_DEBUG(1, 2427 "Disk %s (device %s) marked as dirty.", 2428 g_mirror_get_diskname(disk), sc->sc_name); 2429 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2430 g_mirror_update_metadata(disk); 2431 } else if (pp->acw > 0 && dcw == 0) { 2432 G_MIRROR_DEBUG(1, 2433 "Disk %s (device %s) marked as clean.", 2434 g_mirror_get_diskname(disk), sc->sc_name); 2435 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2436 g_mirror_update_metadata(disk); 2437 } 2438 error = 0; 2439 } else { 2440 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 2441 g_mirror_event_send(disk, 2442 G_MIRROR_DISK_STATE_DISCONNECTED, 2443 G_MIRROR_EVENT_DONTWAIT); 2444 } 2445 } 2446 /* 2447 * Be sure to return 0 for negativate access requests. 2448 * In case of some HW problems, it is possible that we don't have 2449 * any active disk here, so loop above will be no-op and error will 2450 * be ENXIO. 2451 */ 2452 if (error != 0 && acr <= 0 && acw <= 0 && ace <= 0) 2453 error = 0; 2454 return (error); 2455} 2456 2457static struct g_geom * 2458g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2459{ 2460 struct g_mirror_softc *sc; 2461 struct g_geom *gp; 2462 int error, timeout; 2463 2464 g_topology_assert(); 2465 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2466 md->md_mid); 2467 2468 /* One disk is minimum. */ 2469 if (md->md_all < 1) 2470 return (NULL); 2471 /* 2472 * Action geom. 2473 */ 2474 gp = g_new_geomf(mp, "%s", md->md_name); 2475 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2476 gp->start = g_mirror_start; 2477 gp->spoiled = g_mirror_spoiled; 2478 gp->orphan = g_mirror_orphan; 2479 gp->access = g_mirror_access; 2480 gp->dumpconf = g_mirror_dumpconf; 2481 2482 sc->sc_id = md->md_mid; 2483 sc->sc_slice = md->md_slice; 2484 sc->sc_balance = md->md_balance; 2485 sc->sc_mediasize = md->md_mediasize; 2486 sc->sc_sectorsize = md->md_sectorsize; 2487 sc->sc_ndisks = md->md_all; 2488 sc->sc_flags = md->md_mflags; 2489 sc->sc_bump_syncid = 0; 2490 sc->sc_idle = 0; 2491 bioq_init(&sc->sc_queue); 2492 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2493 LIST_INIT(&sc->sc_disks); 2494 TAILQ_INIT(&sc->sc_events); 2495 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2496 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2497 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2498 gp->softc = sc; 2499 sc->sc_geom = gp; 2500 sc->sc_provider = NULL; 2501 /* 2502 * Synchronization geom. 2503 */ 2504 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2505 gp->softc = sc; 2506 gp->orphan = g_mirror_orphan; 2507 sc->sc_sync.ds_geom = gp; 2508 sc->sc_sync.ds_ndisks = 0; 2509 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2510 "g_mirror %s", md->md_name); 2511 if (error != 0) { 2512 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2513 sc->sc_name); 2514 g_destroy_geom(sc->sc_sync.ds_geom); 2515 mtx_destroy(&sc->sc_events_mtx); 2516 mtx_destroy(&sc->sc_queue_mtx); 2517 g_destroy_geom(sc->sc_geom); 2518 free(sc, M_MIRROR); 2519 return (NULL); 2520 } 2521 2522 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2523 2524 /* 2525 * Run timeout. 2526 */ 2527 timeout = g_mirror_timeout * hz; 2528 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 2529 return (sc->sc_geom); 2530} 2531 2532int 2533g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2534{ 2535 struct g_provider *pp; 2536 2537 g_topology_assert(); 2538 2539 if (sc == NULL) 2540 return (ENXIO); 2541 pp = sc->sc_provider; 2542 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2543 if (force) { 2544 G_MIRROR_DEBUG(0, "Device %s is still open, so it " 2545 "can't be definitely removed.", pp->name); 2546 } else { 2547 G_MIRROR_DEBUG(1, 2548 "Device %s is still open (r%dw%de%d).", pp->name, 2549 pp->acr, pp->acw, pp->ace); 2550 return (EBUSY); 2551 } 2552 } 2553 2554 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2555 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2556 g_topology_unlock(); 2557 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2558 mtx_lock(&sc->sc_queue_mtx); 2559 wakeup(sc); 2560 mtx_unlock(&sc->sc_queue_mtx); 2561 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2562 while (sc->sc_worker != NULL) 2563 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2564 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2565 g_topology_lock(); 2566 g_mirror_destroy_device(sc); 2567 free(sc, M_MIRROR); 2568 return (0); 2569} 2570 2571static void 2572g_mirror_taste_orphan(struct g_consumer *cp) 2573{ 2574 2575 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2576 cp->provider->name)); 2577} 2578 2579static struct g_geom * 2580g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2581{ 2582 struct g_mirror_metadata md; 2583 struct g_mirror_softc *sc; 2584 struct g_consumer *cp; 2585 struct g_geom *gp; 2586 int error; 2587 2588 g_topology_assert(); 2589 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2590 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2591 2592 gp = g_new_geomf(mp, "mirror:taste"); 2593 /* 2594 * This orphan function should be never called. 2595 */ 2596 gp->orphan = g_mirror_taste_orphan; 2597 cp = g_new_consumer(gp); 2598 g_attach(cp, pp); 2599 error = g_mirror_read_metadata(cp, &md); 2600 g_detach(cp); 2601 g_destroy_consumer(cp); 2602 g_destroy_geom(gp); 2603 if (error != 0) 2604 return (NULL); 2605 gp = NULL; 2606 2607 if (md.md_version > G_MIRROR_VERSION) { 2608 printf("geom_mirror.ko module is too old to handle %s.\n", 2609 pp->name); 2610 return (NULL); 2611 } 2612 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2613 return (NULL); 2614 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2615 G_MIRROR_DEBUG(0, 2616 "Device %s: provider %s marked as inactive, skipping.", 2617 md.md_name, pp->name); 2618 return (NULL); 2619 } 2620 if (g_mirror_debug >= 2) 2621 mirror_metadata_dump(&md); 2622 2623 /* 2624 * Let's check if device already exists. 2625 */ 2626 sc = NULL; 2627 LIST_FOREACH(gp, &mp->geom, geom) { 2628 sc = gp->softc; 2629 if (sc == NULL) 2630 continue; 2631 if (sc->sc_sync.ds_geom == gp) 2632 continue; 2633 if (strcmp(md.md_name, sc->sc_name) != 0) 2634 continue; 2635 if (md.md_mid != sc->sc_id) { 2636 G_MIRROR_DEBUG(0, "Device %s already configured.", 2637 sc->sc_name); 2638 return (NULL); 2639 } 2640 break; 2641 } 2642 if (gp == NULL) { 2643 gp = g_mirror_create(mp, &md); 2644 if (gp == NULL) { 2645 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2646 md.md_name); 2647 return (NULL); 2648 } 2649 sc = gp->softc; 2650 } 2651 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2652 error = g_mirror_add_disk(sc, pp, &md); 2653 if (error != 0) { 2654 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2655 pp->name, gp->name, error); 2656 if (LIST_EMPTY(&sc->sc_disks)) 2657 g_mirror_destroy(sc, 1); 2658 return (NULL); 2659 } 2660 return (gp); 2661} 2662 2663static int 2664g_mirror_destroy_geom(struct gctl_req *req __unused, 2665 struct g_class *mp __unused, struct g_geom *gp) 2666{ 2667 2668 return (g_mirror_destroy(gp->softc, 0)); 2669} 2670 2671static void 2672g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2673 struct g_consumer *cp, struct g_provider *pp) 2674{ 2675 struct g_mirror_softc *sc; 2676 2677 g_topology_assert(); 2678 2679 sc = gp->softc; 2680 if (sc == NULL) 2681 return; 2682 /* Skip synchronization geom. */ 2683 if (gp == sc->sc_sync.ds_geom) 2684 return; 2685 if (pp != NULL) { 2686 /* Nothing here. */ 2687 } else if (cp != NULL) { 2688 struct g_mirror_disk *disk; 2689 2690 disk = cp->private; 2691 if (disk == NULL) 2692 return; 2693 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2694 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2695 sbuf_printf(sb, "%s<Synchronized>", indent); 2696 if (disk->d_sync.ds_offset_done == 0) 2697 sbuf_printf(sb, "0%%"); 2698 else { 2699 sbuf_printf(sb, "%u%%", 2700 (u_int)((disk->d_sync.ds_offset_done * 100) / 2701 sc->sc_provider->mediasize)); 2702 } 2703 sbuf_printf(sb, "</Synchronized>\n"); 2704 } 2705 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2706 disk->d_sync.ds_syncid); 2707 sbuf_printf(sb, "%s<Flags>", indent); 2708 if (disk->d_flags == 0) 2709 sbuf_printf(sb, "NONE"); 2710 else { 2711 int first = 1; 2712 2713#define ADD_FLAG(flag, name) do { \ 2714 if ((disk->d_flags & (flag)) != 0) { \ 2715 if (!first) \ 2716 sbuf_printf(sb, ", "); \ 2717 else \ 2718 first = 0; \ 2719 sbuf_printf(sb, name); \ 2720 } \ 2721} while (0) 2722 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2723 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2724 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2725 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2726 "SYNCHRONIZING"); 2727 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2728#undef ADD_FLAG 2729 } 2730 sbuf_printf(sb, "</Flags>\n"); 2731 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2732 disk->d_priority); 2733 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2734 g_mirror_disk_state2str(disk->d_state)); 2735 } else { 2736 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2737 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2738 sbuf_printf(sb, "%s<Flags>", indent); 2739 if (sc->sc_flags == 0) 2740 sbuf_printf(sb, "NONE"); 2741 else { 2742 int first = 1; 2743 2744#define ADD_FLAG(flag, name) do { \ 2745 if ((sc->sc_flags & (flag)) != 0) { \ 2746 if (!first) \ 2747 sbuf_printf(sb, ", "); \ 2748 else \ 2749 first = 0; \ 2750 sbuf_printf(sb, name); \ 2751 } \ 2752} while (0) 2753 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2754#undef ADD_FLAG 2755 } 2756 sbuf_printf(sb, "</Flags>\n"); 2757 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2758 (u_int)sc->sc_slice); 2759 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2760 balance_name(sc->sc_balance)); 2761 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2762 sc->sc_ndisks); 2763 sbuf_printf(sb, "%s<State>", indent); 2764 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2765 sbuf_printf(sb, "%s", "STARTING"); 2766 else if (sc->sc_ndisks == 2767 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 2768 sbuf_printf(sb, "%s", "COMPLETE"); 2769 else 2770 sbuf_printf(sb, "%s", "DEGRADED"); 2771 sbuf_printf(sb, "</State>\n"); 2772 } 2773} 2774 2775static int 2776g_mirror_can_go(void) 2777{ 2778 struct g_mirror_softc *sc; 2779 struct g_geom *gp; 2780 struct g_provider *pp; 2781 int can_go; 2782 2783 DROP_GIANT(); 2784 can_go = 1; 2785 g_topology_lock(); 2786 LIST_FOREACH(gp, &g_mirror_class.geom, geom) { 2787 sc = gp->softc; 2788 if (sc == NULL) { 2789 can_go = 0; 2790 break; 2791 } 2792 pp = sc->sc_provider; 2793 if (pp == NULL || pp->error != 0) { 2794 can_go = 0; 2795 break; 2796 } 2797 } 2798 g_topology_unlock(); 2799 PICKUP_GIANT(); 2800 return (can_go); 2801} 2802 2803static void 2804g_mirror_rootwait(void) 2805{ 2806 2807 /* 2808 * HACK: Wait for GEOM, because g_mirror_rootwait() can be called, 2809 * HACK: before we get providers for tasting. 2810 */ 2811 tsleep(&g_mirror_class, PRIBIO, "mroot", hz * 3); 2812 /* 2813 * Wait for mirrors in degraded state. 2814 */ 2815 for (;;) { 2816 if (g_mirror_can_go()) 2817 break; 2818 tsleep(&g_mirror_class, PRIBIO, "mroot", hz); 2819 } 2820} 2821 2822SYSINIT(g_mirror_root, SI_SUB_RAID, SI_ORDER_FIRST, g_mirror_rootwait, NULL) 2823 2824DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2825