1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2004-2006 Pawel Jakub Dawidek <pjd@FreeBSD.org> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include <sys/param.h> 33 #include <sys/systm.h> 34 #include <sys/fail.h> 35 #include <sys/kernel.h> 36 #include <sys/module.h> 37 #include <sys/limits.h> 38 #include <sys/lock.h> 39 #include <sys/mutex.h> 40 #include <sys/bio.h> 41 #include <sys/sbuf.h> 42 #include <sys/sysctl.h> 43 #include <sys/malloc.h> 44 #include <sys/eventhandler.h> 45 #include <vm/uma.h> 46 #include <geom/geom.h> 47 #include <sys/proc.h> 48 #include <sys/kthread.h> 49 #include <sys/sched.h> 50 #include <geom/mirror/g_mirror.h> 51 52 FEATURE(geom_mirror, "GEOM mirroring support"); 53 54 static MALLOC_DEFINE(M_MIRROR, "mirror_data", "GEOM_MIRROR Data"); 55 56 SYSCTL_DECL(_kern_geom); 57 static SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, 58 "GEOM_MIRROR stuff"); 59 int g_mirror_debug = 0; 60 SYSCTL_INT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RWTUN, &g_mirror_debug, 0, 61 "Debug level"); 62 static u_int g_mirror_timeout = 4; 63 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RWTUN, &g_mirror_timeout, 64 0, "Time to wait on all mirror components"); 65 static u_int g_mirror_idletime = 5; 66 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RWTUN, 67 &g_mirror_idletime, 0, "Mark components as clean when idling"); 68 static u_int g_mirror_disconnect_on_failure = 1; 69 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN, 70 &g_mirror_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 71 static u_int g_mirror_syncreqs = 2; 72 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_requests, CTLFLAG_RDTUN, 73 &g_mirror_syncreqs, 0, "Parallel synchronization I/O requests."); 74 static u_int g_mirror_sync_period = 5; 75 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_update_period, CTLFLAG_RWTUN, 76 &g_mirror_sync_period, 0, 77 "Metadata update period during synchroniztion, in seconds"); 78 79 #define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 80 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 81 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 82 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 83 } while (0) 84 85 static eventhandler_tag g_mirror_post_sync = NULL; 86 static int g_mirror_shutdown = 0; 87 88 static g_ctl_destroy_geom_t g_mirror_destroy_geom; 89 static g_taste_t g_mirror_taste; 90 static g_init_t g_mirror_init; 91 static g_fini_t g_mirror_fini; 92 static g_provgone_t g_mirror_providergone; 93 static g_resize_t g_mirror_resize; 94 95 struct g_class g_mirror_class = { 96 .name = G_MIRROR_CLASS_NAME, 97 .version = G_VERSION, 98 .ctlreq = g_mirror_config, 99 .taste = g_mirror_taste, 100 .destroy_geom = g_mirror_destroy_geom, 101 .init = g_mirror_init, 102 .fini = g_mirror_fini, 103 .providergone = g_mirror_providergone, 104 .resize = g_mirror_resize 105 }; 106 107 108 static void g_mirror_destroy_provider(struct g_mirror_softc *sc); 109 static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 110 static void g_mirror_update_device(struct g_mirror_softc *sc, bool force); 111 static void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 112 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 113 static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 114 static void g_mirror_register_request(struct bio *bp); 115 static void g_mirror_sync_release(struct g_mirror_softc *sc); 116 117 118 static const char * 119 g_mirror_disk_state2str(int state) 120 { 121 122 switch (state) { 123 case G_MIRROR_DISK_STATE_NONE: 124 return ("NONE"); 125 case G_MIRROR_DISK_STATE_NEW: 126 return ("NEW"); 127 case G_MIRROR_DISK_STATE_ACTIVE: 128 return ("ACTIVE"); 129 case G_MIRROR_DISK_STATE_STALE: 130 return ("STALE"); 131 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 132 return ("SYNCHRONIZING"); 133 case G_MIRROR_DISK_STATE_DISCONNECTED: 134 return ("DISCONNECTED"); 135 case G_MIRROR_DISK_STATE_DESTROY: 136 return ("DESTROY"); 137 default: 138 return ("INVALID"); 139 } 140 } 141 142 static const char * 143 g_mirror_device_state2str(int state) 144 { 145 146 switch (state) { 147 case G_MIRROR_DEVICE_STATE_STARTING: 148 return ("STARTING"); 149 case G_MIRROR_DEVICE_STATE_RUNNING: 150 return ("RUNNING"); 151 default: 152 return ("INVALID"); 153 } 154 } 155 156 static const char * 157 g_mirror_get_diskname(struct g_mirror_disk *disk) 158 { 159 160 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 161 return ("[unknown]"); 162 return (disk->d_name); 163 } 164 165 /* 166 * --- Events handling functions --- 167 * Events in geom_mirror are used to maintain disks and device status 168 * from one thread to simplify locking. 169 */ 170 static void 171 g_mirror_event_free(struct g_mirror_event *ep) 172 { 173 174 free(ep, M_MIRROR); 175 } 176 177 int 178 g_mirror_event_send(void *arg, int state, int flags) 179 { 180 struct g_mirror_softc *sc; 181 struct g_mirror_disk *disk; 182 struct g_mirror_event *ep; 183 int error; 184 185 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 186 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 187 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 188 disk = NULL; 189 sc = arg; 190 } else { 191 disk = arg; 192 sc = disk->d_softc; 193 } 194 ep->e_disk = disk; 195 ep->e_state = state; 196 ep->e_flags = flags; 197 ep->e_error = 0; 198 mtx_lock(&sc->sc_events_mtx); 199 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 200 mtx_unlock(&sc->sc_events_mtx); 201 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 202 mtx_lock(&sc->sc_queue_mtx); 203 wakeup(sc); 204 mtx_unlock(&sc->sc_queue_mtx); 205 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 206 return (0); 207 sx_assert(&sc->sc_lock, SX_XLOCKED); 208 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 209 sx_xunlock(&sc->sc_lock); 210 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 211 mtx_lock(&sc->sc_events_mtx); 212 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 213 hz * 5); 214 } 215 error = ep->e_error; 216 g_mirror_event_free(ep); 217 sx_xlock(&sc->sc_lock); 218 return (error); 219 } 220 221 static struct g_mirror_event * 222 g_mirror_event_get(struct g_mirror_softc *sc) 223 { 224 struct g_mirror_event *ep; 225 226 mtx_lock(&sc->sc_events_mtx); 227 ep = TAILQ_FIRST(&sc->sc_events); 228 mtx_unlock(&sc->sc_events_mtx); 229 return (ep); 230 } 231 232 static void 233 g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 234 { 235 236 mtx_lock(&sc->sc_events_mtx); 237 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 238 mtx_unlock(&sc->sc_events_mtx); 239 } 240 241 static void 242 g_mirror_event_cancel(struct g_mirror_disk *disk) 243 { 244 struct g_mirror_softc *sc; 245 struct g_mirror_event *ep, *tmpep; 246 247 sc = disk->d_softc; 248 sx_assert(&sc->sc_lock, SX_XLOCKED); 249 250 mtx_lock(&sc->sc_events_mtx); 251 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 252 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 253 continue; 254 if (ep->e_disk != disk) 255 continue; 256 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 257 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 258 g_mirror_event_free(ep); 259 else { 260 ep->e_error = ECANCELED; 261 wakeup(ep); 262 } 263 } 264 mtx_unlock(&sc->sc_events_mtx); 265 } 266 267 /* 268 * Return the number of disks in given state. 269 * If state is equal to -1, count all connected disks. 270 */ 271 u_int 272 g_mirror_ndisks(struct g_mirror_softc *sc, int state) 273 { 274 struct g_mirror_disk *disk; 275 u_int n = 0; 276 277 sx_assert(&sc->sc_lock, SX_LOCKED); 278 279 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 280 if (state == -1 || disk->d_state == state) 281 n++; 282 } 283 return (n); 284 } 285 286 /* 287 * Find a disk in mirror by its disk ID. 288 */ 289 static struct g_mirror_disk * 290 g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 291 { 292 struct g_mirror_disk *disk; 293 294 sx_assert(&sc->sc_lock, SX_XLOCKED); 295 296 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 297 if (disk->d_id == id) 298 return (disk); 299 } 300 return (NULL); 301 } 302 303 static u_int 304 g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 305 { 306 struct bio *bp; 307 u_int nreqs = 0; 308 309 mtx_lock(&sc->sc_queue_mtx); 310 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 311 if (bp->bio_from == cp) 312 nreqs++; 313 } 314 mtx_unlock(&sc->sc_queue_mtx); 315 return (nreqs); 316 } 317 318 static int 319 g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 320 { 321 322 if (cp->index > 0) { 323 G_MIRROR_DEBUG(2, 324 "I/O requests for %s exist, can't destroy it now.", 325 cp->provider->name); 326 return (1); 327 } 328 if (g_mirror_nrequests(sc, cp) > 0) { 329 G_MIRROR_DEBUG(2, 330 "I/O requests for %s in queue, can't destroy it now.", 331 cp->provider->name); 332 return (1); 333 } 334 return (0); 335 } 336 337 static void 338 g_mirror_destroy_consumer(void *arg, int flags __unused) 339 { 340 struct g_consumer *cp; 341 342 g_topology_assert(); 343 344 cp = arg; 345 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 346 g_detach(cp); 347 g_destroy_consumer(cp); 348 } 349 350 static void 351 g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 352 { 353 struct g_provider *pp; 354 int retaste_wait; 355 356 g_topology_assert(); 357 358 cp->private = NULL; 359 if (g_mirror_is_busy(sc, cp)) 360 return; 361 pp = cp->provider; 362 retaste_wait = 0; 363 if (cp->acw == 1) { 364 if ((pp->geom->flags & G_GEOM_WITHER) == 0) 365 retaste_wait = 1; 366 } 367 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 368 -cp->acw, -cp->ace, 0); 369 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 370 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 371 if (retaste_wait) { 372 /* 373 * After retaste event was send (inside g_access()), we can send 374 * event to detach and destroy consumer. 375 * A class, which has consumer to the given provider connected 376 * will not receive retaste event for the provider. 377 * This is the way how I ignore retaste events when I close 378 * consumers opened for write: I detach and destroy consumer 379 * after retaste event is sent. 380 */ 381 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 382 return; 383 } 384 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 385 g_detach(cp); 386 g_destroy_consumer(cp); 387 } 388 389 static int 390 g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 391 { 392 struct g_consumer *cp; 393 int error; 394 395 g_topology_assert_not(); 396 KASSERT(disk->d_consumer == NULL, 397 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 398 399 g_topology_lock(); 400 cp = g_new_consumer(disk->d_softc->sc_geom); 401 cp->flags |= G_CF_DIRECT_RECEIVE; 402 error = g_attach(cp, pp); 403 if (error != 0) { 404 g_destroy_consumer(cp); 405 g_topology_unlock(); 406 return (error); 407 } 408 error = g_access(cp, 1, 1, 1); 409 if (error != 0) { 410 g_detach(cp); 411 g_destroy_consumer(cp); 412 g_topology_unlock(); 413 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 414 pp->name, error); 415 return (error); 416 } 417 g_topology_unlock(); 418 disk->d_consumer = cp; 419 disk->d_consumer->private = disk; 420 disk->d_consumer->index = 0; 421 422 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 423 return (0); 424 } 425 426 static void 427 g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 428 { 429 430 g_topology_assert(); 431 432 if (cp == NULL) 433 return; 434 if (cp->provider != NULL) 435 g_mirror_kill_consumer(sc, cp); 436 else 437 g_destroy_consumer(cp); 438 } 439 440 /* 441 * Initialize disk. This means allocate memory, create consumer, attach it 442 * to the provider and open access (r1w1e1) to it. 443 */ 444 static struct g_mirror_disk * 445 g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 446 struct g_mirror_metadata *md, int *errorp) 447 { 448 struct g_mirror_disk *disk; 449 int i, error; 450 451 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 452 if (disk == NULL) { 453 error = ENOMEM; 454 goto fail; 455 } 456 disk->d_softc = sc; 457 error = g_mirror_connect_disk(disk, pp); 458 if (error != 0) 459 goto fail; 460 disk->d_id = md->md_did; 461 disk->d_state = G_MIRROR_DISK_STATE_NONE; 462 disk->d_priority = md->md_priority; 463 disk->d_flags = md->md_dflags; 464 error = g_getattr("GEOM::candelete", disk->d_consumer, &i); 465 if (error == 0 && i != 0) 466 disk->d_flags |= G_MIRROR_DISK_FLAG_CANDELETE; 467 if (md->md_provider[0] != '\0') 468 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 469 disk->d_sync.ds_consumer = NULL; 470 disk->d_sync.ds_offset = md->md_sync_offset; 471 disk->d_sync.ds_offset_done = md->md_sync_offset; 472 disk->d_sync.ds_update_ts = time_uptime; 473 disk->d_genid = md->md_genid; 474 disk->d_sync.ds_syncid = md->md_syncid; 475 if (errorp != NULL) 476 *errorp = 0; 477 return (disk); 478 fail: 479 if (errorp != NULL) 480 *errorp = error; 481 if (disk != NULL) 482 free(disk, M_MIRROR); 483 return (NULL); 484 } 485 486 static void 487 g_mirror_destroy_disk(struct g_mirror_disk *disk) 488 { 489 struct g_mirror_softc *sc; 490 491 g_topology_assert_not(); 492 sc = disk->d_softc; 493 sx_assert(&sc->sc_lock, SX_XLOCKED); 494 495 LIST_REMOVE(disk, d_next); 496 g_mirror_event_cancel(disk); 497 if (sc->sc_hint == disk) 498 sc->sc_hint = NULL; 499 switch (disk->d_state) { 500 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 501 g_mirror_sync_stop(disk, 1); 502 /* FALLTHROUGH */ 503 case G_MIRROR_DISK_STATE_NEW: 504 case G_MIRROR_DISK_STATE_STALE: 505 case G_MIRROR_DISK_STATE_ACTIVE: 506 g_topology_lock(); 507 g_mirror_disconnect_consumer(sc, disk->d_consumer); 508 g_topology_unlock(); 509 free(disk, M_MIRROR); 510 break; 511 default: 512 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 513 g_mirror_get_diskname(disk), 514 g_mirror_disk_state2str(disk->d_state))); 515 } 516 } 517 518 static void 519 g_mirror_free_device(struct g_mirror_softc *sc) 520 { 521 522 mtx_destroy(&sc->sc_queue_mtx); 523 mtx_destroy(&sc->sc_events_mtx); 524 mtx_destroy(&sc->sc_done_mtx); 525 sx_destroy(&sc->sc_lock); 526 free(sc, M_MIRROR); 527 } 528 529 static void 530 g_mirror_providergone(struct g_provider *pp) 531 { 532 struct g_mirror_softc *sc = pp->private; 533 534 if ((--sc->sc_refcnt) == 0) 535 g_mirror_free_device(sc); 536 } 537 538 static void 539 g_mirror_destroy_device(struct g_mirror_softc *sc) 540 { 541 struct g_mirror_disk *disk; 542 struct g_mirror_event *ep; 543 struct g_geom *gp; 544 struct g_consumer *cp, *tmpcp; 545 546 g_topology_assert_not(); 547 sx_assert(&sc->sc_lock, SX_XLOCKED); 548 549 gp = sc->sc_geom; 550 if (sc->sc_provider != NULL) 551 g_mirror_destroy_provider(sc); 552 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 553 disk = LIST_FIRST(&sc->sc_disks)) { 554 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 555 g_mirror_update_metadata(disk); 556 g_mirror_destroy_disk(disk); 557 } 558 while ((ep = g_mirror_event_get(sc)) != NULL) { 559 g_mirror_event_remove(sc, ep); 560 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 561 g_mirror_event_free(ep); 562 else { 563 ep->e_error = ECANCELED; 564 ep->e_flags |= G_MIRROR_EVENT_DONE; 565 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 566 mtx_lock(&sc->sc_events_mtx); 567 wakeup(ep); 568 mtx_unlock(&sc->sc_events_mtx); 569 } 570 } 571 callout_drain(&sc->sc_callout); 572 573 g_topology_lock(); 574 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 575 g_mirror_disconnect_consumer(sc, cp); 576 } 577 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 578 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 579 g_wither_geom(gp, ENXIO); 580 sx_xunlock(&sc->sc_lock); 581 if ((--sc->sc_refcnt) == 0) 582 g_mirror_free_device(sc); 583 g_topology_unlock(); 584 } 585 586 static void 587 g_mirror_orphan(struct g_consumer *cp) 588 { 589 struct g_mirror_disk *disk; 590 591 g_topology_assert(); 592 593 disk = cp->private; 594 if (disk == NULL) 595 return; 596 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 597 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 598 G_MIRROR_EVENT_DONTWAIT); 599 } 600 601 /* 602 * Function should return the next active disk on the list. 603 * It is possible that it will be the same disk as given. 604 * If there are no active disks on list, NULL is returned. 605 */ 606 static __inline struct g_mirror_disk * 607 g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 608 { 609 struct g_mirror_disk *dp; 610 611 for (dp = LIST_NEXT(disk, d_next); dp != disk; 612 dp = LIST_NEXT(dp, d_next)) { 613 if (dp == NULL) 614 dp = LIST_FIRST(&sc->sc_disks); 615 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 616 break; 617 } 618 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 619 return (NULL); 620 return (dp); 621 } 622 623 static struct g_mirror_disk * 624 g_mirror_get_disk(struct g_mirror_softc *sc) 625 { 626 struct g_mirror_disk *disk; 627 628 if (sc->sc_hint == NULL) { 629 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 630 if (sc->sc_hint == NULL) 631 return (NULL); 632 } 633 disk = sc->sc_hint; 634 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 635 disk = g_mirror_find_next(sc, disk); 636 if (disk == NULL) 637 return (NULL); 638 } 639 sc->sc_hint = g_mirror_find_next(sc, disk); 640 return (disk); 641 } 642 643 static int 644 g_mirror_write_metadata(struct g_mirror_disk *disk, 645 struct g_mirror_metadata *md) 646 { 647 struct g_mirror_softc *sc; 648 struct g_consumer *cp; 649 off_t offset, length; 650 u_char *sector; 651 int error = 0; 652 653 g_topology_assert_not(); 654 sc = disk->d_softc; 655 sx_assert(&sc->sc_lock, SX_LOCKED); 656 657 cp = disk->d_consumer; 658 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 659 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 660 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 661 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 662 cp->acw, cp->ace)); 663 length = cp->provider->sectorsize; 664 offset = cp->provider->mediasize - length; 665 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 666 if (md != NULL && 667 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) { 668 /* 669 * Handle the case, when the size of parent provider reduced. 670 */ 671 if (offset < md->md_mediasize) 672 error = ENOSPC; 673 else 674 mirror_metadata_encode(md, sector); 675 } 676 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_metadata_write, error); 677 if (error == 0) 678 error = g_write_data(cp, offset, sector, length); 679 free(sector, M_MIRROR); 680 if (error != 0) { 681 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 682 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 683 G_MIRROR_DEBUG(0, "Cannot write metadata on %s " 684 "(device=%s, error=%d).", 685 g_mirror_get_diskname(disk), sc->sc_name, error); 686 } else { 687 G_MIRROR_DEBUG(1, "Cannot write metadata on %s " 688 "(device=%s, error=%d).", 689 g_mirror_get_diskname(disk), sc->sc_name, error); 690 } 691 if (g_mirror_disconnect_on_failure && 692 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 693 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 694 g_mirror_event_send(disk, 695 G_MIRROR_DISK_STATE_DISCONNECTED, 696 G_MIRROR_EVENT_DONTWAIT); 697 } 698 } 699 return (error); 700 } 701 702 static int 703 g_mirror_clear_metadata(struct g_mirror_disk *disk) 704 { 705 int error; 706 707 g_topology_assert_not(); 708 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 709 710 if (disk->d_softc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 711 return (0); 712 error = g_mirror_write_metadata(disk, NULL); 713 if (error == 0) { 714 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 715 g_mirror_get_diskname(disk)); 716 } else { 717 G_MIRROR_DEBUG(0, 718 "Cannot clear metadata on disk %s (error=%d).", 719 g_mirror_get_diskname(disk), error); 720 } 721 return (error); 722 } 723 724 void 725 g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 726 struct g_mirror_metadata *md) 727 { 728 729 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 730 md->md_version = G_MIRROR_VERSION; 731 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 732 md->md_mid = sc->sc_id; 733 md->md_all = sc->sc_ndisks; 734 md->md_slice = sc->sc_slice; 735 md->md_balance = sc->sc_balance; 736 md->md_genid = sc->sc_genid; 737 md->md_mediasize = sc->sc_mediasize; 738 md->md_sectorsize = sc->sc_sectorsize; 739 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 740 bzero(md->md_provider, sizeof(md->md_provider)); 741 if (disk == NULL) { 742 md->md_did = arc4random(); 743 md->md_priority = 0; 744 md->md_syncid = 0; 745 md->md_dflags = 0; 746 md->md_sync_offset = 0; 747 md->md_provsize = 0; 748 } else { 749 md->md_did = disk->d_id; 750 md->md_priority = disk->d_priority; 751 md->md_syncid = disk->d_sync.ds_syncid; 752 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 753 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 754 md->md_sync_offset = disk->d_sync.ds_offset_done; 755 else 756 md->md_sync_offset = 0; 757 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 758 strlcpy(md->md_provider, 759 disk->d_consumer->provider->name, 760 sizeof(md->md_provider)); 761 } 762 md->md_provsize = disk->d_consumer->provider->mediasize; 763 } 764 } 765 766 void 767 g_mirror_update_metadata(struct g_mirror_disk *disk) 768 { 769 struct g_mirror_softc *sc; 770 struct g_mirror_metadata md; 771 int error; 772 773 g_topology_assert_not(); 774 sc = disk->d_softc; 775 sx_assert(&sc->sc_lock, SX_LOCKED); 776 777 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 778 return; 779 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) 780 g_mirror_fill_metadata(sc, disk, &md); 781 error = g_mirror_write_metadata(disk, &md); 782 if (error == 0) { 783 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 784 g_mirror_get_diskname(disk)); 785 } else { 786 G_MIRROR_DEBUG(0, 787 "Cannot update metadata on disk %s (error=%d).", 788 g_mirror_get_diskname(disk), error); 789 } 790 } 791 792 static void 793 g_mirror_bump_syncid(struct g_mirror_softc *sc) 794 { 795 struct g_mirror_disk *disk; 796 797 g_topology_assert_not(); 798 sx_assert(&sc->sc_lock, SX_XLOCKED); 799 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 800 ("%s called with no active disks (device=%s).", __func__, 801 sc->sc_name)); 802 803 sc->sc_syncid++; 804 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 805 sc->sc_syncid); 806 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 807 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 808 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 809 disk->d_sync.ds_syncid = sc->sc_syncid; 810 g_mirror_update_metadata(disk); 811 } 812 } 813 } 814 815 static void 816 g_mirror_bump_genid(struct g_mirror_softc *sc) 817 { 818 struct g_mirror_disk *disk; 819 820 g_topology_assert_not(); 821 sx_assert(&sc->sc_lock, SX_XLOCKED); 822 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 823 ("%s called with no active disks (device=%s).", __func__, 824 sc->sc_name)); 825 826 sc->sc_genid++; 827 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 828 sc->sc_genid); 829 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 830 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 831 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 832 disk->d_genid = sc->sc_genid; 833 g_mirror_update_metadata(disk); 834 } 835 } 836 } 837 838 static int 839 g_mirror_idle(struct g_mirror_softc *sc, int acw) 840 { 841 struct g_mirror_disk *disk; 842 int timeout; 843 844 g_topology_assert_not(); 845 sx_assert(&sc->sc_lock, SX_XLOCKED); 846 847 if (sc->sc_provider == NULL) 848 return (0); 849 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 850 return (0); 851 if (sc->sc_idle) 852 return (0); 853 if (sc->sc_writes > 0) 854 return (0); 855 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 856 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); 857 if (!g_mirror_shutdown && timeout > 0) 858 return (timeout); 859 } 860 sc->sc_idle = 1; 861 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 862 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 863 continue; 864 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 865 g_mirror_get_diskname(disk), sc->sc_name); 866 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 867 g_mirror_update_metadata(disk); 868 } 869 return (0); 870 } 871 872 static void 873 g_mirror_unidle(struct g_mirror_softc *sc) 874 { 875 struct g_mirror_disk *disk; 876 877 g_topology_assert_not(); 878 sx_assert(&sc->sc_lock, SX_XLOCKED); 879 880 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 881 return; 882 sc->sc_idle = 0; 883 sc->sc_last_write = time_uptime; 884 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 885 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 886 continue; 887 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 888 g_mirror_get_diskname(disk), sc->sc_name); 889 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 890 g_mirror_update_metadata(disk); 891 } 892 } 893 894 static void 895 g_mirror_flush_done(struct bio *bp) 896 { 897 struct g_mirror_softc *sc; 898 struct bio *pbp; 899 900 pbp = bp->bio_parent; 901 sc = pbp->bio_to->private; 902 mtx_lock(&sc->sc_done_mtx); 903 if (pbp->bio_error == 0) 904 pbp->bio_error = bp->bio_error; 905 pbp->bio_completed += bp->bio_completed; 906 pbp->bio_inbed++; 907 if (pbp->bio_children == pbp->bio_inbed) { 908 mtx_unlock(&sc->sc_done_mtx); 909 g_io_deliver(pbp, pbp->bio_error); 910 } else 911 mtx_unlock(&sc->sc_done_mtx); 912 g_destroy_bio(bp); 913 } 914 915 static void 916 g_mirror_done(struct bio *bp) 917 { 918 struct g_mirror_softc *sc; 919 920 sc = bp->bio_from->geom->softc; 921 bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR; 922 mtx_lock(&sc->sc_queue_mtx); 923 bioq_insert_tail(&sc->sc_queue, bp); 924 mtx_unlock(&sc->sc_queue_mtx); 925 wakeup(sc); 926 } 927 928 static void 929 g_mirror_regular_request(struct bio *bp) 930 { 931 struct g_mirror_softc *sc; 932 struct g_mirror_disk *disk; 933 struct bio *pbp; 934 935 g_topology_assert_not(); 936 937 pbp = bp->bio_parent; 938 sc = pbp->bio_to->private; 939 bp->bio_from->index--; 940 if (bp->bio_cmd == BIO_WRITE) 941 sc->sc_writes--; 942 disk = bp->bio_from->private; 943 if (disk == NULL) { 944 g_topology_lock(); 945 g_mirror_kill_consumer(sc, bp->bio_from); 946 g_topology_unlock(); 947 } 948 949 if (bp->bio_cmd == BIO_READ) 950 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_read, 951 bp->bio_error); 952 else if (bp->bio_cmd == BIO_WRITE) 953 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_write, 954 bp->bio_error); 955 956 pbp->bio_inbed++; 957 KASSERT(pbp->bio_inbed <= pbp->bio_children, 958 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 959 pbp->bio_children)); 960 if (bp->bio_error == 0 && pbp->bio_error == 0) { 961 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 962 g_destroy_bio(bp); 963 if (pbp->bio_children == pbp->bio_inbed) { 964 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 965 pbp->bio_completed = pbp->bio_length; 966 if (pbp->bio_cmd == BIO_WRITE || 967 pbp->bio_cmd == BIO_DELETE) { 968 bioq_remove(&sc->sc_inflight, pbp); 969 /* Release delayed sync requests if possible. */ 970 g_mirror_sync_release(sc); 971 } 972 g_io_deliver(pbp, pbp->bio_error); 973 } 974 return; 975 } else if (bp->bio_error != 0) { 976 if (pbp->bio_error == 0) 977 pbp->bio_error = bp->bio_error; 978 if (disk != NULL) { 979 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 980 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 981 G_MIRROR_LOGREQ(0, bp, 982 "Request failed (error=%d).", 983 bp->bio_error); 984 } else { 985 G_MIRROR_LOGREQ(1, bp, 986 "Request failed (error=%d).", 987 bp->bio_error); 988 } 989 if (g_mirror_disconnect_on_failure && 990 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) 991 { 992 if (bp->bio_error == ENXIO && 993 bp->bio_cmd == BIO_READ) 994 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 995 else if (bp->bio_error == ENXIO) 996 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID_NOW; 997 else 998 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 999 g_mirror_event_send(disk, 1000 G_MIRROR_DISK_STATE_DISCONNECTED, 1001 G_MIRROR_EVENT_DONTWAIT); 1002 } 1003 } 1004 switch (pbp->bio_cmd) { 1005 case BIO_DELETE: 1006 case BIO_WRITE: 1007 pbp->bio_inbed--; 1008 pbp->bio_children--; 1009 break; 1010 } 1011 } 1012 g_destroy_bio(bp); 1013 1014 switch (pbp->bio_cmd) { 1015 case BIO_READ: 1016 if (pbp->bio_inbed < pbp->bio_children) 1017 break; 1018 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 1019 g_io_deliver(pbp, pbp->bio_error); 1020 else { 1021 pbp->bio_error = 0; 1022 mtx_lock(&sc->sc_queue_mtx); 1023 bioq_insert_tail(&sc->sc_queue, pbp); 1024 mtx_unlock(&sc->sc_queue_mtx); 1025 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1026 wakeup(sc); 1027 } 1028 break; 1029 case BIO_DELETE: 1030 case BIO_WRITE: 1031 if (pbp->bio_children == 0) { 1032 /* 1033 * All requests failed. 1034 */ 1035 } else if (pbp->bio_inbed < pbp->bio_children) { 1036 /* Do nothing. */ 1037 break; 1038 } else if (pbp->bio_children == pbp->bio_inbed) { 1039 /* Some requests succeeded. */ 1040 pbp->bio_error = 0; 1041 pbp->bio_completed = pbp->bio_length; 1042 } 1043 bioq_remove(&sc->sc_inflight, pbp); 1044 /* Release delayed sync requests if possible. */ 1045 g_mirror_sync_release(sc); 1046 g_io_deliver(pbp, pbp->bio_error); 1047 break; 1048 default: 1049 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 1050 break; 1051 } 1052 } 1053 1054 static void 1055 g_mirror_sync_done(struct bio *bp) 1056 { 1057 struct g_mirror_softc *sc; 1058 1059 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1060 sc = bp->bio_from->geom->softc; 1061 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; 1062 mtx_lock(&sc->sc_queue_mtx); 1063 bioq_insert_tail(&sc->sc_queue, bp); 1064 mtx_unlock(&sc->sc_queue_mtx); 1065 wakeup(sc); 1066 } 1067 1068 static void 1069 g_mirror_candelete(struct bio *bp) 1070 { 1071 struct g_mirror_softc *sc; 1072 struct g_mirror_disk *disk; 1073 int *val; 1074 1075 sc = bp->bio_to->private; 1076 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1077 if (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) 1078 break; 1079 } 1080 val = (int *)bp->bio_data; 1081 *val = (disk != NULL); 1082 g_io_deliver(bp, 0); 1083 } 1084 1085 static void 1086 g_mirror_kernel_dump(struct bio *bp) 1087 { 1088 struct g_mirror_softc *sc; 1089 struct g_mirror_disk *disk; 1090 struct bio *cbp; 1091 struct g_kerneldump *gkd; 1092 1093 /* 1094 * We configure dumping to the first component, because this component 1095 * will be used for reading with 'prefer' balance algorithm. 1096 * If the component with the highest priority is currently disconnected 1097 * we will not be able to read the dump after the reboot if it will be 1098 * connected and synchronized later. Can we do something better? 1099 */ 1100 sc = bp->bio_to->private; 1101 disk = LIST_FIRST(&sc->sc_disks); 1102 1103 gkd = (struct g_kerneldump *)bp->bio_data; 1104 if (gkd->length > bp->bio_to->mediasize) 1105 gkd->length = bp->bio_to->mediasize; 1106 cbp = g_clone_bio(bp); 1107 if (cbp == NULL) { 1108 g_io_deliver(bp, ENOMEM); 1109 return; 1110 } 1111 cbp->bio_done = g_std_done; 1112 g_io_request(cbp, disk->d_consumer); 1113 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1114 g_mirror_get_diskname(disk)); 1115 } 1116 1117 static void 1118 g_mirror_flush(struct g_mirror_softc *sc, struct bio *bp) 1119 { 1120 struct bio_queue_head queue; 1121 struct g_mirror_disk *disk; 1122 struct g_consumer *cp; 1123 struct bio *cbp; 1124 1125 bioq_init(&queue); 1126 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1127 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1128 continue; 1129 cbp = g_clone_bio(bp); 1130 if (cbp == NULL) { 1131 while ((cbp = bioq_takefirst(&queue)) != NULL) 1132 g_destroy_bio(cbp); 1133 if (bp->bio_error == 0) 1134 bp->bio_error = ENOMEM; 1135 g_io_deliver(bp, bp->bio_error); 1136 return; 1137 } 1138 bioq_insert_tail(&queue, cbp); 1139 cbp->bio_done = g_mirror_flush_done; 1140 cbp->bio_caller1 = disk; 1141 cbp->bio_to = disk->d_consumer->provider; 1142 } 1143 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1144 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1145 disk = cbp->bio_caller1; 1146 cbp->bio_caller1 = NULL; 1147 cp = disk->d_consumer; 1148 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1149 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1150 cp->acr, cp->acw, cp->ace)); 1151 g_io_request(cbp, disk->d_consumer); 1152 } 1153 } 1154 1155 static void 1156 g_mirror_start(struct bio *bp) 1157 { 1158 struct g_mirror_softc *sc; 1159 1160 sc = bp->bio_to->private; 1161 /* 1162 * If sc == NULL or there are no valid disks, provider's error 1163 * should be set and g_mirror_start() should not be called at all. 1164 */ 1165 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1166 ("Provider's error should be set (error=%d)(mirror=%s).", 1167 bp->bio_to->error, bp->bio_to->name)); 1168 G_MIRROR_LOGREQ(3, bp, "Request received."); 1169 1170 switch (bp->bio_cmd) { 1171 case BIO_READ: 1172 case BIO_WRITE: 1173 case BIO_DELETE: 1174 break; 1175 case BIO_FLUSH: 1176 g_mirror_flush(sc, bp); 1177 return; 1178 case BIO_GETATTR: 1179 if (!strcmp(bp->bio_attribute, "GEOM::candelete")) { 1180 g_mirror_candelete(bp); 1181 return; 1182 } else if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1183 g_mirror_kernel_dump(bp); 1184 return; 1185 } 1186 /* FALLTHROUGH */ 1187 default: 1188 g_io_deliver(bp, EOPNOTSUPP); 1189 return; 1190 } 1191 mtx_lock(&sc->sc_queue_mtx); 1192 if (bp->bio_to->error != 0) { 1193 mtx_unlock(&sc->sc_queue_mtx); 1194 g_io_deliver(bp, bp->bio_to->error); 1195 return; 1196 } 1197 bioq_insert_tail(&sc->sc_queue, bp); 1198 mtx_unlock(&sc->sc_queue_mtx); 1199 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1200 wakeup(sc); 1201 } 1202 1203 /* 1204 * Return TRUE if the given request is colliding with a in-progress 1205 * synchronization request. 1206 */ 1207 static int 1208 g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1209 { 1210 struct g_mirror_disk *disk; 1211 struct bio *sbp; 1212 off_t rstart, rend, sstart, send; 1213 u_int i; 1214 1215 if (sc->sc_sync.ds_ndisks == 0) 1216 return (0); 1217 rstart = bp->bio_offset; 1218 rend = bp->bio_offset + bp->bio_length; 1219 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1220 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1221 continue; 1222 for (i = 0; i < g_mirror_syncreqs; i++) { 1223 sbp = disk->d_sync.ds_bios[i]; 1224 if (sbp == NULL) 1225 continue; 1226 sstart = sbp->bio_offset; 1227 send = sbp->bio_offset + sbp->bio_length; 1228 if (rend > sstart && rstart < send) 1229 return (1); 1230 } 1231 } 1232 return (0); 1233 } 1234 1235 /* 1236 * Return TRUE if the given sync request is colliding with a in-progress regular 1237 * request. 1238 */ 1239 static int 1240 g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1241 { 1242 off_t rstart, rend, sstart, send; 1243 struct bio *bp; 1244 1245 if (sc->sc_sync.ds_ndisks == 0) 1246 return (0); 1247 sstart = sbp->bio_offset; 1248 send = sbp->bio_offset + sbp->bio_length; 1249 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { 1250 rstart = bp->bio_offset; 1251 rend = bp->bio_offset + bp->bio_length; 1252 if (rend > sstart && rstart < send) 1253 return (1); 1254 } 1255 return (0); 1256 } 1257 1258 /* 1259 * Puts request onto delayed queue. 1260 */ 1261 static void 1262 g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1263 { 1264 1265 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1266 bioq_insert_head(&sc->sc_regular_delayed, bp); 1267 } 1268 1269 /* 1270 * Puts synchronization request onto delayed queue. 1271 */ 1272 static void 1273 g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1274 { 1275 1276 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1277 bioq_insert_tail(&sc->sc_sync_delayed, bp); 1278 } 1279 1280 /* 1281 * Releases delayed regular requests which don't collide anymore with sync 1282 * requests. 1283 */ 1284 static void 1285 g_mirror_regular_release(struct g_mirror_softc *sc) 1286 { 1287 struct bio *bp, *bp2; 1288 1289 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { 1290 if (g_mirror_sync_collision(sc, bp)) 1291 continue; 1292 bioq_remove(&sc->sc_regular_delayed, bp); 1293 G_MIRROR_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1294 mtx_lock(&sc->sc_queue_mtx); 1295 bioq_insert_head(&sc->sc_queue, bp); 1296 mtx_unlock(&sc->sc_queue_mtx); 1297 } 1298 } 1299 1300 /* 1301 * Releases delayed sync requests which don't collide anymore with regular 1302 * requests. 1303 */ 1304 static void 1305 g_mirror_sync_release(struct g_mirror_softc *sc) 1306 { 1307 struct bio *bp, *bp2; 1308 1309 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { 1310 if (g_mirror_regular_collision(sc, bp)) 1311 continue; 1312 bioq_remove(&sc->sc_sync_delayed, bp); 1313 G_MIRROR_LOGREQ(2, bp, 1314 "Releasing delayed synchronization request."); 1315 g_io_request(bp, bp->bio_from); 1316 } 1317 } 1318 1319 /* 1320 * Free a synchronization request and clear its slot in the array. 1321 */ 1322 static void 1323 g_mirror_sync_request_free(struct g_mirror_disk *disk, struct bio *bp) 1324 { 1325 int idx; 1326 1327 if (disk != NULL && disk->d_sync.ds_bios != NULL) { 1328 idx = (int)(uintptr_t)bp->bio_caller1; 1329 KASSERT(disk->d_sync.ds_bios[idx] == bp, 1330 ("unexpected sync BIO at %p:%d", disk, idx)); 1331 disk->d_sync.ds_bios[idx] = NULL; 1332 } 1333 free(bp->bio_data, M_MIRROR); 1334 g_destroy_bio(bp); 1335 } 1336 1337 /* 1338 * Handle synchronization requests. 1339 * Every synchronization request is two-steps process: first, READ request is 1340 * send to active provider and then WRITE request (with read data) to the provider 1341 * being synchronized. When WRITE is finished, new synchronization request is 1342 * send. 1343 */ 1344 static void 1345 g_mirror_sync_request(struct bio *bp) 1346 { 1347 struct g_mirror_softc *sc; 1348 struct g_mirror_disk *disk; 1349 struct g_mirror_disk_sync *sync; 1350 1351 bp->bio_from->index--; 1352 sc = bp->bio_from->geom->softc; 1353 disk = bp->bio_from->private; 1354 if (disk == NULL) { 1355 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1356 g_topology_lock(); 1357 g_mirror_kill_consumer(sc, bp->bio_from); 1358 g_topology_unlock(); 1359 g_mirror_sync_request_free(NULL, bp); 1360 sx_xlock(&sc->sc_lock); 1361 return; 1362 } 1363 1364 /* 1365 * Synchronization request. 1366 */ 1367 switch (bp->bio_cmd) { 1368 case BIO_READ: 1369 { 1370 struct g_consumer *cp; 1371 1372 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_read, 1373 bp->bio_error); 1374 1375 if (bp->bio_error != 0) { 1376 G_MIRROR_LOGREQ(0, bp, 1377 "Synchronization request failed (error=%d).", 1378 bp->bio_error); 1379 g_mirror_sync_request_free(disk, bp); 1380 return; 1381 } 1382 G_MIRROR_LOGREQ(3, bp, 1383 "Synchronization request half-finished."); 1384 bp->bio_cmd = BIO_WRITE; 1385 bp->bio_cflags = 0; 1386 cp = disk->d_consumer; 1387 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1388 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1389 cp->acr, cp->acw, cp->ace)); 1390 cp->index++; 1391 g_io_request(bp, cp); 1392 return; 1393 } 1394 case BIO_WRITE: 1395 { 1396 off_t offset; 1397 void *data; 1398 int i, idx; 1399 1400 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_write, 1401 bp->bio_error); 1402 1403 if (bp->bio_error != 0) { 1404 G_MIRROR_LOGREQ(0, bp, 1405 "Synchronization request failed (error=%d).", 1406 bp->bio_error); 1407 g_mirror_sync_request_free(disk, bp); 1408 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1409 g_mirror_event_send(disk, 1410 G_MIRROR_DISK_STATE_DISCONNECTED, 1411 G_MIRROR_EVENT_DONTWAIT); 1412 return; 1413 } 1414 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1415 sync = &disk->d_sync; 1416 if (sync->ds_offset >= sc->sc_mediasize || 1417 sync->ds_consumer == NULL || 1418 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1419 /* Don't send more synchronization requests. */ 1420 sync->ds_inflight--; 1421 g_mirror_sync_request_free(disk, bp); 1422 if (sync->ds_inflight > 0) 1423 return; 1424 if (sync->ds_consumer == NULL || 1425 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1426 return; 1427 } 1428 /* Disk up-to-date, activate it. */ 1429 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1430 G_MIRROR_EVENT_DONTWAIT); 1431 return; 1432 } 1433 1434 /* Send next synchronization request. */ 1435 data = bp->bio_data; 1436 idx = (int)(uintptr_t)bp->bio_caller1; 1437 g_reset_bio(bp); 1438 bp->bio_cmd = BIO_READ; 1439 bp->bio_offset = sync->ds_offset; 1440 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1441 sync->ds_offset += bp->bio_length; 1442 bp->bio_done = g_mirror_sync_done; 1443 bp->bio_data = data; 1444 bp->bio_from = sync->ds_consumer; 1445 bp->bio_to = sc->sc_provider; 1446 bp->bio_caller1 = (void *)(uintptr_t)idx; 1447 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1448 sync->ds_consumer->index++; 1449 /* 1450 * Delay the request if it is colliding with a regular request. 1451 */ 1452 if (g_mirror_regular_collision(sc, bp)) 1453 g_mirror_sync_delay(sc, bp); 1454 else 1455 g_io_request(bp, sync->ds_consumer); 1456 1457 /* Release delayed requests if possible. */ 1458 g_mirror_regular_release(sc); 1459 1460 /* Find the smallest offset */ 1461 offset = sc->sc_mediasize; 1462 for (i = 0; i < g_mirror_syncreqs; i++) { 1463 bp = sync->ds_bios[i]; 1464 if (bp != NULL && bp->bio_offset < offset) 1465 offset = bp->bio_offset; 1466 } 1467 if (g_mirror_sync_period > 0 && 1468 time_uptime - sync->ds_update_ts > g_mirror_sync_period) { 1469 sync->ds_offset_done = offset; 1470 g_mirror_update_metadata(disk); 1471 sync->ds_update_ts = time_uptime; 1472 } 1473 return; 1474 } 1475 default: 1476 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1477 bp->bio_cmd, sc->sc_name)); 1478 break; 1479 } 1480 } 1481 1482 static void 1483 g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1484 { 1485 struct g_mirror_disk *disk; 1486 struct g_consumer *cp; 1487 struct bio *cbp; 1488 1489 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1490 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1491 break; 1492 } 1493 if (disk == NULL) { 1494 if (bp->bio_error == 0) 1495 bp->bio_error = ENXIO; 1496 g_io_deliver(bp, bp->bio_error); 1497 return; 1498 } 1499 cbp = g_clone_bio(bp); 1500 if (cbp == NULL) { 1501 if (bp->bio_error == 0) 1502 bp->bio_error = ENOMEM; 1503 g_io_deliver(bp, bp->bio_error); 1504 return; 1505 } 1506 /* 1507 * Fill in the component buf structure. 1508 */ 1509 cp = disk->d_consumer; 1510 cbp->bio_done = g_mirror_done; 1511 cbp->bio_to = cp->provider; 1512 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1513 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1514 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1515 cp->acw, cp->ace)); 1516 cp->index++; 1517 g_io_request(cbp, cp); 1518 } 1519 1520 static void 1521 g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1522 { 1523 struct g_mirror_disk *disk; 1524 struct g_consumer *cp; 1525 struct bio *cbp; 1526 1527 disk = g_mirror_get_disk(sc); 1528 if (disk == NULL) { 1529 if (bp->bio_error == 0) 1530 bp->bio_error = ENXIO; 1531 g_io_deliver(bp, bp->bio_error); 1532 return; 1533 } 1534 cbp = g_clone_bio(bp); 1535 if (cbp == NULL) { 1536 if (bp->bio_error == 0) 1537 bp->bio_error = ENOMEM; 1538 g_io_deliver(bp, bp->bio_error); 1539 return; 1540 } 1541 /* 1542 * Fill in the component buf structure. 1543 */ 1544 cp = disk->d_consumer; 1545 cbp->bio_done = g_mirror_done; 1546 cbp->bio_to = cp->provider; 1547 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1548 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1549 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1550 cp->acw, cp->ace)); 1551 cp->index++; 1552 g_io_request(cbp, cp); 1553 } 1554 1555 #define TRACK_SIZE (1 * 1024 * 1024) 1556 #define LOAD_SCALE 256 1557 #define ABS(x) (((x) >= 0) ? (x) : (-(x))) 1558 1559 static void 1560 g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1561 { 1562 struct g_mirror_disk *disk, *dp; 1563 struct g_consumer *cp; 1564 struct bio *cbp; 1565 int prio, best; 1566 1567 /* Find a disk with the smallest load. */ 1568 disk = NULL; 1569 best = INT_MAX; 1570 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1571 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1572 continue; 1573 prio = dp->load; 1574 /* If disk head is precisely in position - highly prefer it. */ 1575 if (dp->d_last_offset == bp->bio_offset) 1576 prio -= 2 * LOAD_SCALE; 1577 else 1578 /* If disk head is close to position - prefer it. */ 1579 if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE) 1580 prio -= 1 * LOAD_SCALE; 1581 if (prio <= best) { 1582 disk = dp; 1583 best = prio; 1584 } 1585 } 1586 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1587 cbp = g_clone_bio(bp); 1588 if (cbp == NULL) { 1589 if (bp->bio_error == 0) 1590 bp->bio_error = ENOMEM; 1591 g_io_deliver(bp, bp->bio_error); 1592 return; 1593 } 1594 /* 1595 * Fill in the component buf structure. 1596 */ 1597 cp = disk->d_consumer; 1598 cbp->bio_done = g_mirror_done; 1599 cbp->bio_to = cp->provider; 1600 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1601 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1602 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1603 cp->acw, cp->ace)); 1604 cp->index++; 1605 /* Remember last head position */ 1606 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1607 /* Update loads. */ 1608 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1609 dp->load = (dp->d_consumer->index * LOAD_SCALE + 1610 dp->load * 7) / 8; 1611 } 1612 g_io_request(cbp, cp); 1613 } 1614 1615 static void 1616 g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1617 { 1618 struct bio_queue_head queue; 1619 struct g_mirror_disk *disk; 1620 struct g_consumer *cp; 1621 struct bio *cbp; 1622 off_t left, mod, offset, slice; 1623 u_char *data; 1624 u_int ndisks; 1625 1626 if (bp->bio_length <= sc->sc_slice) { 1627 g_mirror_request_round_robin(sc, bp); 1628 return; 1629 } 1630 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1631 slice = bp->bio_length / ndisks; 1632 mod = slice % sc->sc_provider->sectorsize; 1633 if (mod != 0) 1634 slice += sc->sc_provider->sectorsize - mod; 1635 /* 1636 * Allocate all bios before sending any request, so we can 1637 * return ENOMEM in nice and clean way. 1638 */ 1639 left = bp->bio_length; 1640 offset = bp->bio_offset; 1641 data = bp->bio_data; 1642 bioq_init(&queue); 1643 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1644 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1645 continue; 1646 cbp = g_clone_bio(bp); 1647 if (cbp == NULL) { 1648 while ((cbp = bioq_takefirst(&queue)) != NULL) 1649 g_destroy_bio(cbp); 1650 if (bp->bio_error == 0) 1651 bp->bio_error = ENOMEM; 1652 g_io_deliver(bp, bp->bio_error); 1653 return; 1654 } 1655 bioq_insert_tail(&queue, cbp); 1656 cbp->bio_done = g_mirror_done; 1657 cbp->bio_caller1 = disk; 1658 cbp->bio_to = disk->d_consumer->provider; 1659 cbp->bio_offset = offset; 1660 cbp->bio_data = data; 1661 cbp->bio_length = MIN(left, slice); 1662 left -= cbp->bio_length; 1663 if (left == 0) 1664 break; 1665 offset += cbp->bio_length; 1666 data += cbp->bio_length; 1667 } 1668 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1669 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1670 disk = cbp->bio_caller1; 1671 cbp->bio_caller1 = NULL; 1672 cp = disk->d_consumer; 1673 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1674 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1675 cp->acr, cp->acw, cp->ace)); 1676 disk->d_consumer->index++; 1677 g_io_request(cbp, disk->d_consumer); 1678 } 1679 } 1680 1681 static void 1682 g_mirror_register_request(struct bio *bp) 1683 { 1684 struct g_mirror_softc *sc; 1685 1686 sc = bp->bio_to->private; 1687 switch (bp->bio_cmd) { 1688 case BIO_READ: 1689 switch (sc->sc_balance) { 1690 case G_MIRROR_BALANCE_LOAD: 1691 g_mirror_request_load(sc, bp); 1692 break; 1693 case G_MIRROR_BALANCE_PREFER: 1694 g_mirror_request_prefer(sc, bp); 1695 break; 1696 case G_MIRROR_BALANCE_ROUND_ROBIN: 1697 g_mirror_request_round_robin(sc, bp); 1698 break; 1699 case G_MIRROR_BALANCE_SPLIT: 1700 g_mirror_request_split(sc, bp); 1701 break; 1702 } 1703 return; 1704 case BIO_WRITE: 1705 case BIO_DELETE: 1706 { 1707 struct g_mirror_disk *disk; 1708 struct g_mirror_disk_sync *sync; 1709 struct bio_queue_head queue; 1710 struct g_consumer *cp; 1711 struct bio *cbp; 1712 1713 /* 1714 * Delay the request if it is colliding with a synchronization 1715 * request. 1716 */ 1717 if (g_mirror_sync_collision(sc, bp)) { 1718 g_mirror_regular_delay(sc, bp); 1719 return; 1720 } 1721 1722 if (sc->sc_idle) 1723 g_mirror_unidle(sc); 1724 else 1725 sc->sc_last_write = time_uptime; 1726 1727 /* 1728 * Bump syncid on first write. 1729 */ 1730 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1731 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1732 g_mirror_bump_syncid(sc); 1733 } 1734 1735 /* 1736 * Allocate all bios before sending any request, so we can 1737 * return ENOMEM in nice and clean way. 1738 */ 1739 bioq_init(&queue); 1740 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1741 sync = &disk->d_sync; 1742 switch (disk->d_state) { 1743 case G_MIRROR_DISK_STATE_ACTIVE: 1744 break; 1745 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1746 if (bp->bio_offset >= sync->ds_offset) 1747 continue; 1748 break; 1749 default: 1750 continue; 1751 } 1752 if (bp->bio_cmd == BIO_DELETE && 1753 (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0) 1754 continue; 1755 cbp = g_clone_bio(bp); 1756 if (cbp == NULL) { 1757 while ((cbp = bioq_takefirst(&queue)) != NULL) 1758 g_destroy_bio(cbp); 1759 if (bp->bio_error == 0) 1760 bp->bio_error = ENOMEM; 1761 g_io_deliver(bp, bp->bio_error); 1762 return; 1763 } 1764 bioq_insert_tail(&queue, cbp); 1765 cbp->bio_done = g_mirror_done; 1766 cp = disk->d_consumer; 1767 cbp->bio_caller1 = cp; 1768 cbp->bio_to = cp->provider; 1769 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1770 ("Consumer %s not opened (r%dw%de%d).", 1771 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1772 } 1773 if (bioq_first(&queue) == NULL) { 1774 g_io_deliver(bp, EOPNOTSUPP); 1775 return; 1776 } 1777 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1778 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1779 cp = cbp->bio_caller1; 1780 cbp->bio_caller1 = NULL; 1781 cp->index++; 1782 sc->sc_writes++; 1783 g_io_request(cbp, cp); 1784 } 1785 /* 1786 * Put request onto inflight queue, so we can check if new 1787 * synchronization requests don't collide with it. 1788 */ 1789 bioq_insert_tail(&sc->sc_inflight, bp); 1790 return; 1791 } 1792 default: 1793 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1794 bp->bio_cmd, sc->sc_name)); 1795 break; 1796 } 1797 } 1798 1799 static int 1800 g_mirror_can_destroy(struct g_mirror_softc *sc) 1801 { 1802 struct g_geom *gp; 1803 struct g_consumer *cp; 1804 1805 g_topology_assert(); 1806 gp = sc->sc_geom; 1807 if (gp->softc == NULL) 1808 return (1); 1809 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0) 1810 return (0); 1811 LIST_FOREACH(cp, &gp->consumer, consumer) { 1812 if (g_mirror_is_busy(sc, cp)) 1813 return (0); 1814 } 1815 gp = sc->sc_sync.ds_geom; 1816 LIST_FOREACH(cp, &gp->consumer, consumer) { 1817 if (g_mirror_is_busy(sc, cp)) 1818 return (0); 1819 } 1820 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1821 sc->sc_name); 1822 return (1); 1823 } 1824 1825 static int 1826 g_mirror_try_destroy(struct g_mirror_softc *sc) 1827 { 1828 1829 if (sc->sc_rootmount != NULL) { 1830 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1831 sc->sc_rootmount); 1832 root_mount_rel(sc->sc_rootmount); 1833 sc->sc_rootmount = NULL; 1834 } 1835 g_topology_lock(); 1836 if (!g_mirror_can_destroy(sc)) { 1837 g_topology_unlock(); 1838 return (0); 1839 } 1840 sc->sc_geom->softc = NULL; 1841 sc->sc_sync.ds_geom->softc = NULL; 1842 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DRAIN) != 0) { 1843 g_topology_unlock(); 1844 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1845 &sc->sc_worker); 1846 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1847 sx_xunlock(&sc->sc_lock); 1848 wakeup(&sc->sc_worker); 1849 sc->sc_worker = NULL; 1850 } else { 1851 g_topology_unlock(); 1852 g_mirror_destroy_device(sc); 1853 } 1854 return (1); 1855 } 1856 1857 /* 1858 * Worker thread. 1859 */ 1860 static void 1861 g_mirror_worker(void *arg) 1862 { 1863 struct g_mirror_softc *sc; 1864 struct g_mirror_event *ep; 1865 struct bio *bp; 1866 int timeout; 1867 1868 sc = arg; 1869 thread_lock(curthread); 1870 sched_prio(curthread, PRIBIO); 1871 thread_unlock(curthread); 1872 1873 sx_xlock(&sc->sc_lock); 1874 for (;;) { 1875 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1876 /* 1877 * First take a look at events. 1878 * This is important to handle events before any I/O requests. 1879 */ 1880 ep = g_mirror_event_get(sc); 1881 if (ep != NULL) { 1882 g_mirror_event_remove(sc, ep); 1883 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1884 /* Update only device status. */ 1885 G_MIRROR_DEBUG(3, 1886 "Running event for device %s.", 1887 sc->sc_name); 1888 ep->e_error = 0; 1889 g_mirror_update_device(sc, true); 1890 } else { 1891 /* Update disk status. */ 1892 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1893 g_mirror_get_diskname(ep->e_disk)); 1894 ep->e_error = g_mirror_update_disk(ep->e_disk, 1895 ep->e_state); 1896 if (ep->e_error == 0) 1897 g_mirror_update_device(sc, false); 1898 } 1899 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1900 KASSERT(ep->e_error == 0, 1901 ("Error cannot be handled.")); 1902 g_mirror_event_free(ep); 1903 } else { 1904 ep->e_flags |= G_MIRROR_EVENT_DONE; 1905 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1906 ep); 1907 mtx_lock(&sc->sc_events_mtx); 1908 wakeup(ep); 1909 mtx_unlock(&sc->sc_events_mtx); 1910 } 1911 if ((sc->sc_flags & 1912 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1913 if (g_mirror_try_destroy(sc)) { 1914 curthread->td_pflags &= ~TDP_GEOM; 1915 G_MIRROR_DEBUG(1, "Thread exiting."); 1916 kproc_exit(0); 1917 } 1918 } 1919 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1920 continue; 1921 } 1922 /* 1923 * Check if we can mark array as CLEAN and if we can't take 1924 * how much seconds should we wait. 1925 */ 1926 timeout = g_mirror_idle(sc, -1); 1927 /* 1928 * Now I/O requests. 1929 */ 1930 /* Get first request from the queue. */ 1931 mtx_lock(&sc->sc_queue_mtx); 1932 bp = bioq_takefirst(&sc->sc_queue); 1933 if (bp == NULL) { 1934 if ((sc->sc_flags & 1935 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1936 mtx_unlock(&sc->sc_queue_mtx); 1937 if (g_mirror_try_destroy(sc)) { 1938 curthread->td_pflags &= ~TDP_GEOM; 1939 G_MIRROR_DEBUG(1, "Thread exiting."); 1940 kproc_exit(0); 1941 } 1942 mtx_lock(&sc->sc_queue_mtx); 1943 if (bioq_first(&sc->sc_queue) != NULL) { 1944 mtx_unlock(&sc->sc_queue_mtx); 1945 continue; 1946 } 1947 } 1948 sx_xunlock(&sc->sc_lock); 1949 /* 1950 * XXX: We can miss an event here, because an event 1951 * can be added without sx-device-lock and without 1952 * mtx-queue-lock. Maybe I should just stop using 1953 * dedicated mutex for events synchronization and 1954 * stick with the queue lock? 1955 * The event will hang here until next I/O request 1956 * or next event is received. 1957 */ 1958 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1959 timeout * hz); 1960 sx_xlock(&sc->sc_lock); 1961 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1962 continue; 1963 } 1964 mtx_unlock(&sc->sc_queue_mtx); 1965 1966 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 1967 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1968 g_mirror_sync_request(bp); /* READ */ 1969 } else if (bp->bio_to != sc->sc_provider) { 1970 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 1971 g_mirror_regular_request(bp); 1972 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 1973 g_mirror_sync_request(bp); /* WRITE */ 1974 else { 1975 KASSERT(0, 1976 ("Invalid request cflags=0x%hx to=%s.", 1977 bp->bio_cflags, bp->bio_to->name)); 1978 } 1979 } else { 1980 g_mirror_register_request(bp); 1981 } 1982 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1983 } 1984 } 1985 1986 static void 1987 g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 1988 { 1989 1990 sx_assert(&sc->sc_lock, SX_LOCKED); 1991 1992 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 1993 return; 1994 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1995 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as dirty.", 1996 g_mirror_get_diskname(disk), sc->sc_name); 1997 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1998 } else if (sc->sc_idle && 1999 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2000 G_MIRROR_DEBUG(2, "Disk %s (device %s) marked as clean.", 2001 g_mirror_get_diskname(disk), sc->sc_name); 2002 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2003 } 2004 } 2005 2006 static void 2007 g_mirror_sync_start(struct g_mirror_disk *disk) 2008 { 2009 struct g_mirror_softc *sc; 2010 struct g_consumer *cp; 2011 struct bio *bp; 2012 int error, i; 2013 2014 g_topology_assert_not(); 2015 sc = disk->d_softc; 2016 sx_assert(&sc->sc_lock, SX_LOCKED); 2017 2018 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2019 ("Disk %s is not marked for synchronization.", 2020 g_mirror_get_diskname(disk))); 2021 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2022 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 2023 sc->sc_state)); 2024 2025 sx_xunlock(&sc->sc_lock); 2026 g_topology_lock(); 2027 cp = g_new_consumer(sc->sc_sync.ds_geom); 2028 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 2029 error = g_attach(cp, sc->sc_provider); 2030 KASSERT(error == 0, 2031 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 2032 error = g_access(cp, 1, 0, 0); 2033 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 2034 g_topology_unlock(); 2035 sx_xlock(&sc->sc_lock); 2036 2037 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 2038 g_mirror_get_diskname(disk)); 2039 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0) 2040 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2041 KASSERT(disk->d_sync.ds_consumer == NULL, 2042 ("Sync consumer already exists (device=%s, disk=%s).", 2043 sc->sc_name, g_mirror_get_diskname(disk))); 2044 2045 disk->d_sync.ds_consumer = cp; 2046 disk->d_sync.ds_consumer->private = disk; 2047 disk->d_sync.ds_consumer->index = 0; 2048 2049 /* 2050 * Allocate memory for synchronization bios and initialize them. 2051 */ 2052 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 2053 M_MIRROR, M_WAITOK); 2054 for (i = 0; i < g_mirror_syncreqs; i++) { 2055 bp = g_alloc_bio(); 2056 disk->d_sync.ds_bios[i] = bp; 2057 bp->bio_parent = NULL; 2058 bp->bio_cmd = BIO_READ; 2059 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 2060 bp->bio_cflags = 0; 2061 bp->bio_offset = disk->d_sync.ds_offset; 2062 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 2063 disk->d_sync.ds_offset += bp->bio_length; 2064 bp->bio_done = g_mirror_sync_done; 2065 bp->bio_from = disk->d_sync.ds_consumer; 2066 bp->bio_to = sc->sc_provider; 2067 bp->bio_caller1 = (void *)(uintptr_t)i; 2068 } 2069 2070 /* Increase the number of disks in SYNCHRONIZING state. */ 2071 sc->sc_sync.ds_ndisks++; 2072 /* Set the number of in-flight synchronization requests. */ 2073 disk->d_sync.ds_inflight = g_mirror_syncreqs; 2074 2075 /* 2076 * Fire off first synchronization requests. 2077 */ 2078 for (i = 0; i < g_mirror_syncreqs; i++) { 2079 bp = disk->d_sync.ds_bios[i]; 2080 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 2081 disk->d_sync.ds_consumer->index++; 2082 /* 2083 * Delay the request if it is colliding with a regular request. 2084 */ 2085 if (g_mirror_regular_collision(sc, bp)) 2086 g_mirror_sync_delay(sc, bp); 2087 else 2088 g_io_request(bp, disk->d_sync.ds_consumer); 2089 } 2090 } 2091 2092 /* 2093 * Stop synchronization process. 2094 * type: 0 - synchronization finished 2095 * 1 - synchronization stopped 2096 */ 2097 static void 2098 g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 2099 { 2100 struct g_mirror_softc *sc; 2101 struct g_consumer *cp; 2102 2103 g_topology_assert_not(); 2104 sc = disk->d_softc; 2105 sx_assert(&sc->sc_lock, SX_LOCKED); 2106 2107 KASSERT(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 if (disk->d_sync.ds_consumer == NULL) 2111 return; 2112 2113 if (type == 0) { 2114 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2115 sc->sc_name, g_mirror_get_diskname(disk)); 2116 } else /* if (type == 1) */ { 2117 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2118 sc->sc_name, g_mirror_get_diskname(disk)); 2119 } 2120 g_mirror_regular_release(sc); 2121 free(disk->d_sync.ds_bios, M_MIRROR); 2122 disk->d_sync.ds_bios = NULL; 2123 cp = disk->d_sync.ds_consumer; 2124 disk->d_sync.ds_consumer = NULL; 2125 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2126 sc->sc_sync.ds_ndisks--; 2127 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2128 g_topology_lock(); 2129 g_mirror_kill_consumer(sc, cp); 2130 g_topology_unlock(); 2131 sx_xlock(&sc->sc_lock); 2132 } 2133 2134 static void 2135 g_mirror_launch_provider(struct g_mirror_softc *sc) 2136 { 2137 struct g_mirror_disk *disk; 2138 struct g_provider *pp, *dp; 2139 2140 sx_assert(&sc->sc_lock, SX_LOCKED); 2141 2142 g_topology_lock(); 2143 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 2144 pp->flags |= G_PF_DIRECT_RECEIVE; 2145 pp->mediasize = sc->sc_mediasize; 2146 pp->sectorsize = sc->sc_sectorsize; 2147 pp->stripesize = 0; 2148 pp->stripeoffset = 0; 2149 2150 /* Splitting of unmapped BIO's could work but isn't implemented now */ 2151 if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT) 2152 pp->flags |= G_PF_ACCEPT_UNMAPPED; 2153 2154 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2155 if (disk->d_consumer && disk->d_consumer->provider) { 2156 dp = disk->d_consumer->provider; 2157 if (dp->stripesize > pp->stripesize) { 2158 pp->stripesize = dp->stripesize; 2159 pp->stripeoffset = dp->stripeoffset; 2160 } 2161 /* A provider underneath us doesn't support unmapped */ 2162 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { 2163 G_MIRROR_DEBUG(0, "Cancelling unmapped " 2164 "because of %s.", dp->name); 2165 pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 2166 } 2167 } 2168 } 2169 pp->private = sc; 2170 sc->sc_refcnt++; 2171 sc->sc_provider = pp; 2172 g_error_provider(pp, 0); 2173 g_topology_unlock(); 2174 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2175 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 2176 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2177 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2178 g_mirror_sync_start(disk); 2179 } 2180 } 2181 2182 static void 2183 g_mirror_destroy_provider(struct g_mirror_softc *sc) 2184 { 2185 struct g_mirror_disk *disk; 2186 struct bio *bp; 2187 2188 g_topology_assert_not(); 2189 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2190 sc->sc_name)); 2191 2192 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2193 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2194 g_mirror_sync_stop(disk, 1); 2195 } 2196 2197 g_topology_lock(); 2198 g_error_provider(sc->sc_provider, ENXIO); 2199 mtx_lock(&sc->sc_queue_mtx); 2200 while ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) { 2201 /* 2202 * Abort any pending I/O that wasn't generated by us. 2203 * Synchronization requests and requests destined for individual 2204 * mirror components can be destroyed immediately. 2205 */ 2206 if (bp->bio_to == sc->sc_provider && 2207 bp->bio_from->geom != sc->sc_sync.ds_geom) { 2208 g_io_deliver(bp, ENXIO); 2209 } else { 2210 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2211 free(bp->bio_data, M_MIRROR); 2212 g_destroy_bio(bp); 2213 } 2214 } 2215 mtx_unlock(&sc->sc_queue_mtx); 2216 g_wither_provider(sc->sc_provider, ENXIO); 2217 sc->sc_provider = NULL; 2218 G_MIRROR_DEBUG(0, "Device %s: provider destroyed.", sc->sc_name); 2219 g_topology_unlock(); 2220 } 2221 2222 static void 2223 g_mirror_go(void *arg) 2224 { 2225 struct g_mirror_softc *sc; 2226 2227 sc = arg; 2228 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2229 g_mirror_event_send(sc, 0, 2230 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2231 } 2232 2233 static u_int 2234 g_mirror_determine_state(struct g_mirror_disk *disk) 2235 { 2236 struct g_mirror_softc *sc; 2237 u_int state; 2238 2239 sc = disk->d_softc; 2240 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2241 if ((disk->d_flags & 2242 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0 && 2243 (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 || 2244 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0)) { 2245 /* Disk does not need synchronization. */ 2246 state = G_MIRROR_DISK_STATE_ACTIVE; 2247 } else { 2248 if ((sc->sc_flags & 2249 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2250 (disk->d_flags & 2251 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2252 /* 2253 * We can start synchronization from 2254 * the stored offset. 2255 */ 2256 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2257 } else { 2258 state = G_MIRROR_DISK_STATE_STALE; 2259 } 2260 } 2261 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2262 /* 2263 * Reset all synchronization data for this disk, 2264 * because if it even was synchronized, it was 2265 * synchronized to disks with different syncid. 2266 */ 2267 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2268 disk->d_sync.ds_offset = 0; 2269 disk->d_sync.ds_offset_done = 0; 2270 disk->d_sync.ds_syncid = sc->sc_syncid; 2271 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2272 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2273 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2274 } else { 2275 state = G_MIRROR_DISK_STATE_STALE; 2276 } 2277 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2278 /* 2279 * Not good, NOT GOOD! 2280 * It means that mirror was started on stale disks 2281 * and more fresh disk just arrive. 2282 * If there were writes, mirror is broken, sorry. 2283 * I think the best choice here is don't touch 2284 * this disk and inform the user loudly. 2285 */ 2286 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2287 "disk (%s) arrives!! It will not be connected to the " 2288 "running device.", sc->sc_name, 2289 g_mirror_get_diskname(disk)); 2290 g_mirror_destroy_disk(disk); 2291 state = G_MIRROR_DISK_STATE_NONE; 2292 /* Return immediately, because disk was destroyed. */ 2293 return (state); 2294 } 2295 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2296 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2297 return (state); 2298 } 2299 2300 /* 2301 * Update device state. 2302 */ 2303 static void 2304 g_mirror_update_device(struct g_mirror_softc *sc, bool force) 2305 { 2306 struct g_mirror_disk *disk; 2307 u_int state; 2308 2309 sx_assert(&sc->sc_lock, SX_XLOCKED); 2310 2311 switch (sc->sc_state) { 2312 case G_MIRROR_DEVICE_STATE_STARTING: 2313 { 2314 struct g_mirror_disk *pdisk, *tdisk; 2315 u_int dirty, ndisks, genid, syncid; 2316 bool broken; 2317 2318 KASSERT(sc->sc_provider == NULL, 2319 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2320 /* 2321 * Are we ready? We are, if all disks are connected or 2322 * if we have any disks and 'force' is true. 2323 */ 2324 ndisks = g_mirror_ndisks(sc, -1); 2325 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2326 ; 2327 } else if (ndisks == 0) { 2328 /* 2329 * Disks went down in starting phase, so destroy 2330 * device. 2331 */ 2332 callout_drain(&sc->sc_callout); 2333 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2334 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2335 sc->sc_rootmount); 2336 root_mount_rel(sc->sc_rootmount); 2337 sc->sc_rootmount = NULL; 2338 return; 2339 } else { 2340 return; 2341 } 2342 2343 /* 2344 * Activate all disks with the biggest syncid. 2345 */ 2346 if (force) { 2347 /* 2348 * If 'force' is true, we have been called due to 2349 * timeout, so don't bother canceling timeout. 2350 */ 2351 ndisks = 0; 2352 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2353 if ((disk->d_flags & 2354 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2355 ndisks++; 2356 } 2357 } 2358 if (ndisks == 0) { 2359 /* No valid disks found, destroy device. */ 2360 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2361 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2362 __LINE__, sc->sc_rootmount); 2363 root_mount_rel(sc->sc_rootmount); 2364 sc->sc_rootmount = NULL; 2365 return; 2366 } 2367 } else { 2368 /* Cancel timeout. */ 2369 callout_drain(&sc->sc_callout); 2370 } 2371 2372 /* 2373 * Find the biggest genid. 2374 */ 2375 genid = 0; 2376 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2377 if (disk->d_genid > genid) 2378 genid = disk->d_genid; 2379 } 2380 sc->sc_genid = genid; 2381 /* 2382 * Remove all disks without the biggest genid. 2383 */ 2384 broken = false; 2385 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2386 if (disk->d_genid < genid) { 2387 G_MIRROR_DEBUG(0, 2388 "Component %s (device %s) broken, skipping.", 2389 g_mirror_get_diskname(disk), sc->sc_name); 2390 g_mirror_destroy_disk(disk); 2391 /* 2392 * Bump the syncid in case we discover a healthy 2393 * replacement disk after starting the mirror. 2394 */ 2395 broken = true; 2396 } 2397 } 2398 2399 /* 2400 * Find the biggest syncid. 2401 */ 2402 syncid = 0; 2403 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2404 if (disk->d_sync.ds_syncid > syncid) 2405 syncid = disk->d_sync.ds_syncid; 2406 } 2407 2408 /* 2409 * Here we need to look for dirty disks and if all disks 2410 * with the biggest syncid are dirty, we have to choose 2411 * one with the biggest priority and rebuild the rest. 2412 */ 2413 /* 2414 * Find the number of dirty disks with the biggest syncid. 2415 * Find the number of disks with the biggest syncid. 2416 * While here, find a disk with the biggest priority. 2417 */ 2418 dirty = ndisks = 0; 2419 pdisk = NULL; 2420 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2421 if (disk->d_sync.ds_syncid != syncid) 2422 continue; 2423 if ((disk->d_flags & 2424 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2425 continue; 2426 } 2427 ndisks++; 2428 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2429 dirty++; 2430 if (pdisk == NULL || 2431 pdisk->d_priority < disk->d_priority) { 2432 pdisk = disk; 2433 } 2434 } 2435 } 2436 if (dirty == 0) { 2437 /* No dirty disks at all, great. */ 2438 } else if (dirty == ndisks) { 2439 /* 2440 * Force synchronization for all dirty disks except one 2441 * with the biggest priority. 2442 */ 2443 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2444 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2445 "master disk for synchronization.", 2446 g_mirror_get_diskname(pdisk), sc->sc_name); 2447 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2448 if (disk->d_sync.ds_syncid != syncid) 2449 continue; 2450 if ((disk->d_flags & 2451 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2452 continue; 2453 } 2454 KASSERT((disk->d_flags & 2455 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2456 ("Disk %s isn't marked as dirty.", 2457 g_mirror_get_diskname(disk))); 2458 /* Skip the disk with the biggest priority. */ 2459 if (disk == pdisk) 2460 continue; 2461 disk->d_sync.ds_syncid = 0; 2462 } 2463 } else if (dirty < ndisks) { 2464 /* 2465 * Force synchronization for all dirty disks. 2466 * We have some non-dirty disks. 2467 */ 2468 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2469 if (disk->d_sync.ds_syncid != syncid) 2470 continue; 2471 if ((disk->d_flags & 2472 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2473 continue; 2474 } 2475 if ((disk->d_flags & 2476 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2477 continue; 2478 } 2479 disk->d_sync.ds_syncid = 0; 2480 } 2481 } 2482 2483 /* Reset hint. */ 2484 sc->sc_hint = NULL; 2485 sc->sc_syncid = syncid; 2486 if (force || broken) { 2487 /* Remember to bump syncid on first write. */ 2488 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2489 } 2490 state = G_MIRROR_DEVICE_STATE_RUNNING; 2491 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2492 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2493 g_mirror_device_state2str(state)); 2494 sc->sc_state = state; 2495 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2496 state = g_mirror_determine_state(disk); 2497 g_mirror_event_send(disk, state, 2498 G_MIRROR_EVENT_DONTWAIT); 2499 if (state == G_MIRROR_DISK_STATE_STALE) 2500 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2501 } 2502 break; 2503 } 2504 case G_MIRROR_DEVICE_STATE_RUNNING: 2505 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2506 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2507 /* 2508 * No usable disks, so destroy the device. 2509 */ 2510 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2511 break; 2512 } else if (g_mirror_ndisks(sc, 2513 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2514 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2515 /* 2516 * We have active disks, launch provider if it doesn't 2517 * exist. 2518 */ 2519 if (sc->sc_provider == NULL) 2520 g_mirror_launch_provider(sc); 2521 if (sc->sc_rootmount != NULL) { 2522 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2523 __LINE__, sc->sc_rootmount); 2524 root_mount_rel(sc->sc_rootmount); 2525 sc->sc_rootmount = NULL; 2526 } 2527 } 2528 /* 2529 * Genid should be bumped immediately, so do it here. 2530 */ 2531 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2532 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2533 g_mirror_bump_genid(sc); 2534 } 2535 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID_NOW) != 0) { 2536 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID_NOW; 2537 g_mirror_bump_syncid(sc); 2538 } 2539 break; 2540 default: 2541 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2542 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2543 break; 2544 } 2545 } 2546 2547 /* 2548 * Update disk state and device state if needed. 2549 */ 2550 #define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2551 "Disk %s state changed from %s to %s (device %s).", \ 2552 g_mirror_get_diskname(disk), \ 2553 g_mirror_disk_state2str(disk->d_state), \ 2554 g_mirror_disk_state2str(state), sc->sc_name) 2555 static int 2556 g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2557 { 2558 struct g_mirror_softc *sc; 2559 2560 sc = disk->d_softc; 2561 sx_assert(&sc->sc_lock, SX_XLOCKED); 2562 2563 again: 2564 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2565 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2566 g_mirror_disk_state2str(state)); 2567 switch (state) { 2568 case G_MIRROR_DISK_STATE_NEW: 2569 /* 2570 * Possible scenarios: 2571 * 1. New disk arrive. 2572 */ 2573 /* Previous state should be NONE. */ 2574 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2575 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2576 g_mirror_disk_state2str(disk->d_state))); 2577 DISK_STATE_CHANGED(); 2578 2579 disk->d_state = state; 2580 if (LIST_EMPTY(&sc->sc_disks)) 2581 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2582 else { 2583 struct g_mirror_disk *dp; 2584 2585 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2586 if (disk->d_priority >= dp->d_priority) { 2587 LIST_INSERT_BEFORE(dp, disk, d_next); 2588 dp = NULL; 2589 break; 2590 } 2591 if (LIST_NEXT(dp, d_next) == NULL) 2592 break; 2593 } 2594 if (dp != NULL) 2595 LIST_INSERT_AFTER(dp, disk, d_next); 2596 } 2597 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2598 sc->sc_name, g_mirror_get_diskname(disk)); 2599 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2600 break; 2601 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2602 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2603 g_mirror_device_state2str(sc->sc_state), 2604 g_mirror_get_diskname(disk), 2605 g_mirror_disk_state2str(disk->d_state))); 2606 state = g_mirror_determine_state(disk); 2607 if (state != G_MIRROR_DISK_STATE_NONE) 2608 goto again; 2609 break; 2610 case G_MIRROR_DISK_STATE_ACTIVE: 2611 /* 2612 * Possible scenarios: 2613 * 1. New disk does not need synchronization. 2614 * 2. Synchronization process finished successfully. 2615 */ 2616 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2617 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2618 g_mirror_device_state2str(sc->sc_state), 2619 g_mirror_get_diskname(disk), 2620 g_mirror_disk_state2str(disk->d_state))); 2621 /* Previous state should be NEW or SYNCHRONIZING. */ 2622 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2623 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2624 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2625 g_mirror_disk_state2str(disk->d_state))); 2626 DISK_STATE_CHANGED(); 2627 2628 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2629 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2630 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2631 g_mirror_sync_stop(disk, 0); 2632 } 2633 disk->d_state = state; 2634 disk->d_sync.ds_offset = 0; 2635 disk->d_sync.ds_offset_done = 0; 2636 g_mirror_update_idle(sc, disk); 2637 g_mirror_update_metadata(disk); 2638 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2639 sc->sc_name, g_mirror_get_diskname(disk)); 2640 break; 2641 case G_MIRROR_DISK_STATE_STALE: 2642 /* 2643 * Possible scenarios: 2644 * 1. Stale disk was connected. 2645 */ 2646 /* Previous state should be NEW. */ 2647 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2648 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2649 g_mirror_disk_state2str(disk->d_state))); 2650 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2651 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2652 g_mirror_device_state2str(sc->sc_state), 2653 g_mirror_get_diskname(disk), 2654 g_mirror_disk_state2str(disk->d_state))); 2655 /* 2656 * STALE state is only possible if device is marked 2657 * NOAUTOSYNC. 2658 */ 2659 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2660 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2661 g_mirror_device_state2str(sc->sc_state), 2662 g_mirror_get_diskname(disk), 2663 g_mirror_disk_state2str(disk->d_state))); 2664 DISK_STATE_CHANGED(); 2665 2666 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2667 disk->d_state = state; 2668 g_mirror_update_metadata(disk); 2669 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2670 sc->sc_name, g_mirror_get_diskname(disk)); 2671 break; 2672 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2673 /* 2674 * Possible scenarios: 2675 * 1. Disk which needs synchronization was connected. 2676 */ 2677 /* Previous state should be NEW. */ 2678 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2679 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2680 g_mirror_disk_state2str(disk->d_state))); 2681 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2682 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2683 g_mirror_device_state2str(sc->sc_state), 2684 g_mirror_get_diskname(disk), 2685 g_mirror_disk_state2str(disk->d_state))); 2686 DISK_STATE_CHANGED(); 2687 2688 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2689 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2690 disk->d_state = state; 2691 if (sc->sc_provider != NULL) { 2692 g_mirror_sync_start(disk); 2693 g_mirror_update_metadata(disk); 2694 } 2695 break; 2696 case G_MIRROR_DISK_STATE_DISCONNECTED: 2697 /* 2698 * Possible scenarios: 2699 * 1. Device wasn't running yet, but disk disappear. 2700 * 2. Disk was active and disapppear. 2701 * 3. Disk disappear during synchronization process. 2702 */ 2703 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2704 /* 2705 * Previous state should be ACTIVE, STALE or 2706 * SYNCHRONIZING. 2707 */ 2708 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2709 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2710 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2711 ("Wrong disk state (%s, %s).", 2712 g_mirror_get_diskname(disk), 2713 g_mirror_disk_state2str(disk->d_state))); 2714 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2715 /* Previous state should be NEW. */ 2716 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2717 ("Wrong disk state (%s, %s).", 2718 g_mirror_get_diskname(disk), 2719 g_mirror_disk_state2str(disk->d_state))); 2720 /* 2721 * Reset bumping syncid if disk disappeared in STARTING 2722 * state. 2723 */ 2724 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2725 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2726 #ifdef INVARIANTS 2727 } else { 2728 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2729 sc->sc_name, 2730 g_mirror_device_state2str(sc->sc_state), 2731 g_mirror_get_diskname(disk), 2732 g_mirror_disk_state2str(disk->d_state))); 2733 #endif 2734 } 2735 DISK_STATE_CHANGED(); 2736 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2737 sc->sc_name, g_mirror_get_diskname(disk)); 2738 2739 g_mirror_destroy_disk(disk); 2740 break; 2741 case G_MIRROR_DISK_STATE_DESTROY: 2742 { 2743 int error; 2744 2745 error = g_mirror_clear_metadata(disk); 2746 if (error != 0) { 2747 G_MIRROR_DEBUG(0, 2748 "Device %s: failed to clear metadata on %s: %d.", 2749 sc->sc_name, g_mirror_get_diskname(disk), error); 2750 break; 2751 } 2752 DISK_STATE_CHANGED(); 2753 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2754 sc->sc_name, g_mirror_get_diskname(disk)); 2755 2756 g_mirror_destroy_disk(disk); 2757 sc->sc_ndisks--; 2758 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2759 g_mirror_update_metadata(disk); 2760 } 2761 break; 2762 } 2763 default: 2764 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2765 break; 2766 } 2767 return (0); 2768 } 2769 #undef DISK_STATE_CHANGED 2770 2771 int 2772 g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2773 { 2774 struct g_provider *pp; 2775 u_char *buf; 2776 int error; 2777 2778 g_topology_assert(); 2779 2780 error = g_access(cp, 1, 0, 0); 2781 if (error != 0) 2782 return (error); 2783 pp = cp->provider; 2784 g_topology_unlock(); 2785 /* Metadata are stored on last sector. */ 2786 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2787 &error); 2788 g_topology_lock(); 2789 g_access(cp, -1, 0, 0); 2790 if (buf == NULL) { 2791 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2792 cp->provider->name, error); 2793 return (error); 2794 } 2795 2796 /* Decode metadata. */ 2797 error = mirror_metadata_decode(buf, md); 2798 g_free(buf); 2799 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2800 return (EINVAL); 2801 if (md->md_version > G_MIRROR_VERSION) { 2802 G_MIRROR_DEBUG(0, 2803 "Kernel module is too old to handle metadata from %s.", 2804 cp->provider->name); 2805 return (EINVAL); 2806 } 2807 if (error != 0) { 2808 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2809 cp->provider->name); 2810 return (error); 2811 } 2812 2813 return (0); 2814 } 2815 2816 static int 2817 g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2818 struct g_mirror_metadata *md) 2819 { 2820 2821 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2822 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2823 pp->name, md->md_did); 2824 return (EEXIST); 2825 } 2826 if (md->md_all != sc->sc_ndisks) { 2827 G_MIRROR_DEBUG(1, 2828 "Invalid '%s' field on disk %s (device %s), skipping.", 2829 "md_all", pp->name, sc->sc_name); 2830 return (EINVAL); 2831 } 2832 if (md->md_slice != sc->sc_slice) { 2833 G_MIRROR_DEBUG(1, 2834 "Invalid '%s' field on disk %s (device %s), skipping.", 2835 "md_slice", pp->name, sc->sc_name); 2836 return (EINVAL); 2837 } 2838 if (md->md_balance != sc->sc_balance) { 2839 G_MIRROR_DEBUG(1, 2840 "Invalid '%s' field on disk %s (device %s), skipping.", 2841 "md_balance", pp->name, sc->sc_name); 2842 return (EINVAL); 2843 } 2844 #if 0 2845 if (md->md_mediasize != sc->sc_mediasize) { 2846 G_MIRROR_DEBUG(1, 2847 "Invalid '%s' field on disk %s (device %s), skipping.", 2848 "md_mediasize", pp->name, sc->sc_name); 2849 return (EINVAL); 2850 } 2851 #endif 2852 if (sc->sc_mediasize > pp->mediasize) { 2853 G_MIRROR_DEBUG(1, 2854 "Invalid size of disk %s (device %s), skipping.", pp->name, 2855 sc->sc_name); 2856 return (EINVAL); 2857 } 2858 if (md->md_sectorsize != sc->sc_sectorsize) { 2859 G_MIRROR_DEBUG(1, 2860 "Invalid '%s' field on disk %s (device %s), skipping.", 2861 "md_sectorsize", pp->name, sc->sc_name); 2862 return (EINVAL); 2863 } 2864 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2865 G_MIRROR_DEBUG(1, 2866 "Invalid sector size of disk %s (device %s), skipping.", 2867 pp->name, sc->sc_name); 2868 return (EINVAL); 2869 } 2870 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2871 G_MIRROR_DEBUG(1, 2872 "Invalid device flags on disk %s (device %s), skipping.", 2873 pp->name, sc->sc_name); 2874 return (EINVAL); 2875 } 2876 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2877 G_MIRROR_DEBUG(1, 2878 "Invalid disk flags on disk %s (device %s), skipping.", 2879 pp->name, sc->sc_name); 2880 return (EINVAL); 2881 } 2882 return (0); 2883 } 2884 2885 int 2886 g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2887 struct g_mirror_metadata *md) 2888 { 2889 struct g_mirror_disk *disk; 2890 int error; 2891 2892 g_topology_assert_not(); 2893 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2894 2895 error = g_mirror_check_metadata(sc, pp, md); 2896 if (error != 0) 2897 return (error); 2898 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2899 md->md_genid < sc->sc_genid) { 2900 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2901 pp->name, sc->sc_name); 2902 return (EINVAL); 2903 } 2904 disk = g_mirror_init_disk(sc, pp, md, &error); 2905 if (disk == NULL) 2906 return (error); 2907 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2908 G_MIRROR_EVENT_WAIT); 2909 if (error != 0) 2910 return (error); 2911 if (md->md_version < G_MIRROR_VERSION) { 2912 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2913 pp->name, md->md_version, G_MIRROR_VERSION); 2914 g_mirror_update_metadata(disk); 2915 } 2916 return (0); 2917 } 2918 2919 static void 2920 g_mirror_destroy_delayed(void *arg, int flag) 2921 { 2922 struct g_mirror_softc *sc; 2923 int error; 2924 2925 if (flag == EV_CANCEL) { 2926 G_MIRROR_DEBUG(1, "Destroying canceled."); 2927 return; 2928 } 2929 sc = arg; 2930 g_topology_unlock(); 2931 sx_xlock(&sc->sc_lock); 2932 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2933 ("DESTROY flag set on %s.", sc->sc_name)); 2934 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0, 2935 ("CLOSEWAIT flag not set on %s.", sc->sc_name)); 2936 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 2937 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 2938 if (error != 0) { 2939 G_MIRROR_DEBUG(0, "Cannot destroy %s (error=%d).", 2940 sc->sc_name, error); 2941 sx_xunlock(&sc->sc_lock); 2942 } 2943 g_topology_lock(); 2944 } 2945 2946 static int 2947 g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2948 { 2949 struct g_mirror_softc *sc; 2950 int error = 0; 2951 2952 g_topology_assert(); 2953 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2954 acw, ace); 2955 2956 sc = pp->private; 2957 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 2958 2959 g_topology_unlock(); 2960 sx_xlock(&sc->sc_lock); 2961 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 2962 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 || 2963 LIST_EMPTY(&sc->sc_disks)) { 2964 if (acr > 0 || acw > 0 || ace > 0) 2965 error = ENXIO; 2966 goto end; 2967 } 2968 sc->sc_provider_open += acr + acw + ace; 2969 if (pp->acw + acw == 0) 2970 g_mirror_idle(sc, 0); 2971 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_CLOSEWAIT) != 0 && 2972 sc->sc_provider_open == 0) 2973 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, sc, NULL); 2974 end: 2975 sx_xunlock(&sc->sc_lock); 2976 g_topology_lock(); 2977 return (error); 2978 } 2979 2980 struct g_geom * 2981 g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md, 2982 u_int type) 2983 { 2984 struct g_mirror_softc *sc; 2985 struct g_geom *gp; 2986 int error, timeout; 2987 2988 g_topology_assert(); 2989 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2990 md->md_mid); 2991 2992 /* One disk is minimum. */ 2993 if (md->md_all < 1) 2994 return (NULL); 2995 /* 2996 * Action geom. 2997 */ 2998 gp = g_new_geomf(mp, "%s", md->md_name); 2999 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 3000 gp->start = g_mirror_start; 3001 gp->orphan = g_mirror_orphan; 3002 gp->access = g_mirror_access; 3003 gp->dumpconf = g_mirror_dumpconf; 3004 3005 sc->sc_type = type; 3006 sc->sc_id = md->md_mid; 3007 sc->sc_slice = md->md_slice; 3008 sc->sc_balance = md->md_balance; 3009 sc->sc_mediasize = md->md_mediasize; 3010 sc->sc_sectorsize = md->md_sectorsize; 3011 sc->sc_ndisks = md->md_all; 3012 sc->sc_flags = md->md_mflags; 3013 sc->sc_bump_id = 0; 3014 sc->sc_idle = 1; 3015 sc->sc_last_write = time_uptime; 3016 sc->sc_writes = 0; 3017 sc->sc_refcnt = 1; 3018 sx_init(&sc->sc_lock, "gmirror:lock"); 3019 bioq_init(&sc->sc_queue); 3020 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 3021 bioq_init(&sc->sc_regular_delayed); 3022 bioq_init(&sc->sc_inflight); 3023 bioq_init(&sc->sc_sync_delayed); 3024 LIST_INIT(&sc->sc_disks); 3025 TAILQ_INIT(&sc->sc_events); 3026 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 3027 callout_init(&sc->sc_callout, 1); 3028 mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF); 3029 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 3030 gp->softc = sc; 3031 sc->sc_geom = gp; 3032 sc->sc_provider = NULL; 3033 sc->sc_provider_open = 0; 3034 /* 3035 * Synchronization geom. 3036 */ 3037 gp = g_new_geomf(mp, "%s.sync", md->md_name); 3038 gp->softc = sc; 3039 gp->orphan = g_mirror_orphan; 3040 sc->sc_sync.ds_geom = gp; 3041 sc->sc_sync.ds_ndisks = 0; 3042 error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 3043 "g_mirror %s", md->md_name); 3044 if (error != 0) { 3045 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 3046 sc->sc_name); 3047 g_destroy_geom(sc->sc_sync.ds_geom); 3048 g_destroy_geom(sc->sc_geom); 3049 g_mirror_free_device(sc); 3050 return (NULL); 3051 } 3052 3053 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 3054 sc->sc_name, sc->sc_ndisks, sc->sc_id); 3055 3056 sc->sc_rootmount = root_mount_hold("GMIRROR"); 3057 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 3058 /* 3059 * Run timeout. 3060 */ 3061 timeout = g_mirror_timeout * hz; 3062 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 3063 return (sc->sc_geom); 3064 } 3065 3066 int 3067 g_mirror_destroy(struct g_mirror_softc *sc, int how) 3068 { 3069 struct g_mirror_disk *disk; 3070 3071 g_topology_assert_not(); 3072 sx_assert(&sc->sc_lock, SX_XLOCKED); 3073 3074 if (sc->sc_provider_open != 0) { 3075 switch (how) { 3076 case G_MIRROR_DESTROY_SOFT: 3077 G_MIRROR_DEBUG(1, 3078 "Device %s is still open (%d).", sc->sc_name, 3079 sc->sc_provider_open); 3080 return (EBUSY); 3081 case G_MIRROR_DESTROY_DELAYED: 3082 G_MIRROR_DEBUG(1, 3083 "Device %s will be destroyed on last close.", 3084 sc->sc_name); 3085 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 3086 if (disk->d_state == 3087 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3088 g_mirror_sync_stop(disk, 1); 3089 } 3090 } 3091 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_CLOSEWAIT; 3092 return (EBUSY); 3093 case G_MIRROR_DESTROY_HARD: 3094 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 3095 "can't be definitely removed.", sc->sc_name); 3096 } 3097 } 3098 3099 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3100 sx_xunlock(&sc->sc_lock); 3101 return (0); 3102 } 3103 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 3104 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DRAIN; 3105 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3106 sx_xunlock(&sc->sc_lock); 3107 mtx_lock(&sc->sc_queue_mtx); 3108 wakeup(sc); 3109 mtx_unlock(&sc->sc_queue_mtx); 3110 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3111 while (sc->sc_worker != NULL) 3112 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 3113 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3114 sx_xlock(&sc->sc_lock); 3115 g_mirror_destroy_device(sc); 3116 return (0); 3117 } 3118 3119 static void 3120 g_mirror_taste_orphan(struct g_consumer *cp) 3121 { 3122 3123 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3124 cp->provider->name)); 3125 } 3126 3127 static struct g_geom * 3128 g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3129 { 3130 struct g_mirror_metadata md; 3131 struct g_mirror_softc *sc; 3132 struct g_consumer *cp; 3133 struct g_geom *gp; 3134 int error; 3135 3136 g_topology_assert(); 3137 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3138 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 3139 3140 gp = g_new_geomf(mp, "mirror:taste"); 3141 /* 3142 * This orphan function should be never called. 3143 */ 3144 gp->orphan = g_mirror_taste_orphan; 3145 cp = g_new_consumer(gp); 3146 g_attach(cp, pp); 3147 error = g_mirror_read_metadata(cp, &md); 3148 g_detach(cp); 3149 g_destroy_consumer(cp); 3150 g_destroy_geom(gp); 3151 if (error != 0) 3152 return (NULL); 3153 gp = NULL; 3154 3155 if (md.md_provider[0] != '\0' && 3156 !g_compare_names(md.md_provider, pp->name)) 3157 return (NULL); 3158 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3159 return (NULL); 3160 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 3161 G_MIRROR_DEBUG(0, 3162 "Device %s: provider %s marked as inactive, skipping.", 3163 md.md_name, pp->name); 3164 return (NULL); 3165 } 3166 if (g_mirror_debug >= 2) 3167 mirror_metadata_dump(&md); 3168 3169 /* 3170 * Let's check if device already exists. 3171 */ 3172 sc = NULL; 3173 LIST_FOREACH(gp, &mp->geom, geom) { 3174 sc = gp->softc; 3175 if (sc == NULL) 3176 continue; 3177 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 3178 continue; 3179 if (sc->sc_sync.ds_geom == gp) 3180 continue; 3181 if (strcmp(md.md_name, sc->sc_name) != 0) 3182 continue; 3183 if (md.md_mid != sc->sc_id) { 3184 G_MIRROR_DEBUG(0, "Device %s already configured.", 3185 sc->sc_name); 3186 return (NULL); 3187 } 3188 break; 3189 } 3190 if (gp == NULL) { 3191 gp = g_mirror_create(mp, &md, G_MIRROR_TYPE_AUTOMATIC); 3192 if (gp == NULL) { 3193 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3194 md.md_name); 3195 return (NULL); 3196 } 3197 sc = gp->softc; 3198 } 3199 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3200 g_topology_unlock(); 3201 sx_xlock(&sc->sc_lock); 3202 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; 3203 error = g_mirror_add_disk(sc, pp, &md); 3204 if (error != 0) { 3205 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3206 pp->name, gp->name, error); 3207 if (LIST_EMPTY(&sc->sc_disks)) { 3208 g_cancel_event(sc); 3209 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3210 g_topology_lock(); 3211 return (NULL); 3212 } 3213 gp = NULL; 3214 } 3215 sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; 3216 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3217 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3218 g_topology_lock(); 3219 return (NULL); 3220 } 3221 sx_xunlock(&sc->sc_lock); 3222 g_topology_lock(); 3223 return (gp); 3224 } 3225 3226 static void 3227 g_mirror_resize(struct g_consumer *cp) 3228 { 3229 struct g_mirror_disk *disk; 3230 3231 g_topology_assert(); 3232 g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name); 3233 3234 disk = cp->private; 3235 if (disk == NULL) 3236 return; 3237 g_topology_unlock(); 3238 g_mirror_update_metadata(disk); 3239 g_topology_lock(); 3240 } 3241 3242 static int 3243 g_mirror_destroy_geom(struct gctl_req *req __unused, 3244 struct g_class *mp __unused, struct g_geom *gp) 3245 { 3246 struct g_mirror_softc *sc; 3247 int error; 3248 3249 g_topology_unlock(); 3250 sc = gp->softc; 3251 sx_xlock(&sc->sc_lock); 3252 g_cancel_event(sc); 3253 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3254 if (error != 0) 3255 sx_xunlock(&sc->sc_lock); 3256 g_topology_lock(); 3257 return (error); 3258 } 3259 3260 static void 3261 g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3262 struct g_consumer *cp, struct g_provider *pp) 3263 { 3264 struct g_mirror_softc *sc; 3265 3266 g_topology_assert(); 3267 3268 sc = gp->softc; 3269 if (sc == NULL) 3270 return; 3271 /* Skip synchronization geom. */ 3272 if (gp == sc->sc_sync.ds_geom) 3273 return; 3274 if (pp != NULL) { 3275 /* Nothing here. */ 3276 } else if (cp != NULL) { 3277 struct g_mirror_disk *disk; 3278 3279 disk = cp->private; 3280 if (disk == NULL) 3281 return; 3282 g_topology_unlock(); 3283 sx_xlock(&sc->sc_lock); 3284 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3285 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3286 sbuf_printf(sb, "%s<Synchronized>", indent); 3287 if (disk->d_sync.ds_offset == 0) 3288 sbuf_printf(sb, "0%%"); 3289 else { 3290 sbuf_printf(sb, "%u%%", 3291 (u_int)((disk->d_sync.ds_offset * 100) / 3292 sc->sc_provider->mediasize)); 3293 } 3294 sbuf_printf(sb, "</Synchronized>\n"); 3295 if (disk->d_sync.ds_offset > 0) { 3296 sbuf_printf(sb, "%s<BytesSynced>%jd" 3297 "</BytesSynced>\n", indent, 3298 (intmax_t)disk->d_sync.ds_offset); 3299 } 3300 } 3301 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3302 disk->d_sync.ds_syncid); 3303 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3304 disk->d_genid); 3305 sbuf_printf(sb, "%s<Flags>", indent); 3306 if (disk->d_flags == 0) 3307 sbuf_printf(sb, "NONE"); 3308 else { 3309 int first = 1; 3310 3311 #define ADD_FLAG(flag, name) do { \ 3312 if ((disk->d_flags & (flag)) != 0) { \ 3313 if (!first) \ 3314 sbuf_printf(sb, ", "); \ 3315 else \ 3316 first = 0; \ 3317 sbuf_printf(sb, name); \ 3318 } \ 3319 } while (0) 3320 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3321 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3322 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3323 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3324 "SYNCHRONIZING"); 3325 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3326 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3327 #undef ADD_FLAG 3328 } 3329 sbuf_printf(sb, "</Flags>\n"); 3330 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3331 disk->d_priority); 3332 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3333 g_mirror_disk_state2str(disk->d_state)); 3334 sx_xunlock(&sc->sc_lock); 3335 g_topology_lock(); 3336 } else { 3337 g_topology_unlock(); 3338 sx_xlock(&sc->sc_lock); 3339 sbuf_printf(sb, "%s<Type>", indent); 3340 switch (sc->sc_type) { 3341 case G_MIRROR_TYPE_AUTOMATIC: 3342 sbuf_printf(sb, "AUTOMATIC"); 3343 break; 3344 case G_MIRROR_TYPE_MANUAL: 3345 sbuf_printf(sb, "MANUAL"); 3346 break; 3347 default: 3348 sbuf_printf(sb, "UNKNOWN"); 3349 break; 3350 } 3351 sbuf_printf(sb, "</Type>\n"); 3352 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3353 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3354 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3355 sbuf_printf(sb, "%s<Flags>", indent); 3356 if (sc->sc_flags == 0) 3357 sbuf_printf(sb, "NONE"); 3358 else { 3359 int first = 1; 3360 3361 #define ADD_FLAG(flag, name) do { \ 3362 if ((sc->sc_flags & (flag)) != 0) { \ 3363 if (!first) \ 3364 sbuf_printf(sb, ", "); \ 3365 else \ 3366 first = 0; \ 3367 sbuf_printf(sb, name); \ 3368 } \ 3369 } while (0) 3370 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3371 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3372 #undef ADD_FLAG 3373 } 3374 sbuf_printf(sb, "</Flags>\n"); 3375 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3376 (u_int)sc->sc_slice); 3377 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3378 balance_name(sc->sc_balance)); 3379 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3380 sc->sc_ndisks); 3381 sbuf_printf(sb, "%s<State>", indent); 3382 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3383 sbuf_printf(sb, "%s", "STARTING"); 3384 else if (sc->sc_ndisks == 3385 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3386 sbuf_printf(sb, "%s", "COMPLETE"); 3387 else 3388 sbuf_printf(sb, "%s", "DEGRADED"); 3389 sbuf_printf(sb, "</State>\n"); 3390 sx_xunlock(&sc->sc_lock); 3391 g_topology_lock(); 3392 } 3393 } 3394 3395 static void 3396 g_mirror_shutdown_post_sync(void *arg, int howto) 3397 { 3398 struct g_class *mp; 3399 struct g_geom *gp, *gp2; 3400 struct g_mirror_softc *sc; 3401 int error; 3402 3403 if (panicstr != NULL) 3404 return; 3405 3406 mp = arg; 3407 g_topology_lock(); 3408 g_mirror_shutdown = 1; 3409 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3410 if ((sc = gp->softc) == NULL) 3411 continue; 3412 /* Skip synchronization geom. */ 3413 if (gp == sc->sc_sync.ds_geom) 3414 continue; 3415 g_topology_unlock(); 3416 sx_xlock(&sc->sc_lock); 3417 g_mirror_idle(sc, -1); 3418 g_cancel_event(sc); 3419 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3420 if (error != 0) 3421 sx_xunlock(&sc->sc_lock); 3422 g_topology_lock(); 3423 } 3424 g_topology_unlock(); 3425 } 3426 3427 static void 3428 g_mirror_init(struct g_class *mp) 3429 { 3430 3431 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3432 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3433 if (g_mirror_post_sync == NULL) 3434 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3435 } 3436 3437 static void 3438 g_mirror_fini(struct g_class *mp) 3439 { 3440 3441 if (g_mirror_post_sync != NULL) 3442 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); 3443 } 3444 3445 DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3446