g_mirror.c revision 136191
190792Sgshapiro/*- 2261363Sgshapiro * Copyright (c) 2004 Pawel Jakub Dawidek <pjd@FreeBSD.org> 390792Sgshapiro * All rights reserved. 490792Sgshapiro * 590792Sgshapiro * Redistribution and use in source and binary forms, with or without 690792Sgshapiro * modification, are permitted provided that the following conditions 790792Sgshapiro * are met: 890792Sgshapiro * 1. Redistributions of source code must retain the above copyright 9261363Sgshapiro * notice, this list of conditions and the following disclaimer. 1090792Sgshapiro * 2. Redistributions in binary form must reproduce the above copyright 1190792Sgshapiro * notice, this list of conditions and the following disclaimer in the 1290792Sgshapiro * documentation and/or other materials provided with the distribution. 1390792Sgshapiro * 1490792Sgshapiro * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 1590792Sgshapiro * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 1690792Sgshapiro * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 1790792Sgshapiro * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 1890792Sgshapiro * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 1990792Sgshapiro * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2090792Sgshapiro * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2190792Sgshapiro * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2290792Sgshapiro * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 2390792Sgshapiro * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 2490792Sgshapiro * SUCH DAMAGE. 2590792Sgshapiro */ 2690792Sgshapiro 2790792Sgshapiro#include <sys/cdefs.h> 2890792Sgshapiro__FBSDID("$FreeBSD: head/sys/geom/mirror/g_mirror.c 136191 2004-10-06 14:15:47Z pjd $"); 2990792Sgshapiro 3090792Sgshapiro#include <sys/param.h> 3190792Sgshapiro#include <sys/systm.h> 3290792Sgshapiro#include <sys/kernel.h> 3390792Sgshapiro#include <sys/module.h> 3490792Sgshapiro#include <sys/limits.h> 3590792Sgshapiro#include <sys/lock.h> 3690792Sgshapiro#include <sys/mutex.h> 3790792Sgshapiro#include <sys/bio.h> 3890792Sgshapiro#include <sys/sysctl.h> 3990792Sgshapiro#include <sys/malloc.h> 4090792Sgshapiro#include <sys/bitstring.h> 4190792Sgshapiro#include <vm/uma.h> 4290792Sgshapiro#include <machine/atomic.h> 4390792Sgshapiro#include <geom/geom.h> 4490792Sgshapiro#include <sys/proc.h> 4590792Sgshapiro#include <sys/kthread.h> 4694334Sgshapiro#include <geom/mirror/g_mirror.h> 4794334Sgshapiro 4894334Sgshapiro 4990792Sgshapirostatic MALLOC_DEFINE(M_MIRROR, "mirror data", "GEOM_MIRROR Data"); 5090792Sgshapiro 5190792SgshapiroSYSCTL_DECL(_kern_geom); 5290792SgshapiroSYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, "GEOM_MIRROR stuff"); 5390792Sgshapirou_int g_mirror_debug = 0; 5490792SgshapiroTUNABLE_INT("kern.geom.mirror.debug", &g_mirror_debug); 5590792SgshapiroSYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RW, &g_mirror_debug, 0, 5690792Sgshapiro "Debug level"); 5790792Sgshapirostatic u_int g_mirror_timeout = 4; 5890792SgshapiroTUNABLE_INT("kern.geom.mirror.timeout", &g_mirror_timeout); 5990792SgshapiroSYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RW, &g_mirror_timeout, 6090792Sgshapiro 0, "Time to wait on all mirror components"); 6190792Sgshapirostatic u_int g_mirror_reqs_per_sync = 5; 6290792SgshapiroSYSCTL_UINT(_kern_geom_mirror, OID_AUTO, reqs_per_sync, CTLFLAG_RW, 6390792Sgshapiro &g_mirror_reqs_per_sync, 0, 6490792Sgshapiro "Number of regular I/O requests per synchronization request"); 6590792Sgshapirostatic u_int g_mirror_syncs_per_sec = 100; 6690792SgshapiroSYSCTL_UINT(_kern_geom_mirror, OID_AUTO, syncs_per_sec, CTLFLAG_RW, 6790792Sgshapiro &g_mirror_syncs_per_sec, 0, 6890792Sgshapiro "Number of synchronizations requests per second"); 6990792Sgshapiro 70125820Sgshapiro#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 7190792Sgshapiro G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 7290792Sgshapiro msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 73125820Sgshapiro G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 7490792Sgshapiro} while (0) 7590792Sgshapiro 7690792Sgshapiro 77125820Sgshapirostatic int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp, 7890792Sgshapiro struct g_geom *gp); 7990792Sgshapirostatic g_taste_t g_mirror_taste; 8090792Sgshapiro 8190792Sgshapirostruct g_class g_mirror_class = { 8290792Sgshapiro .name = G_MIRROR_CLASS_NAME, 8390792Sgshapiro .version = G_VERSION, 8490792Sgshapiro .ctlreq = g_mirror_config, 8590792Sgshapiro .taste = g_mirror_taste, 8690792Sgshapiro .destroy_geom = g_mirror_destroy_geom 8790792Sgshapiro}; 8890792Sgshapiro 8990792Sgshapiro 9090792Sgshapirostatic void g_mirror_destroy_provider(struct g_mirror_softc *sc); 9190792Sgshapirostatic int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 9290792Sgshapirostatic void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force); 9390792Sgshapirostatic void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 9490792Sgshapiro struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 9590792Sgshapirostatic void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 9690792Sgshapiro 9790792Sgshapiro 9890792Sgshapirostatic const char * 9990792Sgshapirog_mirror_disk_state2str(int state) 10090792Sgshapiro{ 10190792Sgshapiro 10290792Sgshapiro switch (state) { 10390792Sgshapiro case G_MIRROR_DISK_STATE_NONE: 10490792Sgshapiro return ("NONE"); 10590792Sgshapiro case G_MIRROR_DISK_STATE_NEW: 10690792Sgshapiro return ("NEW"); 10790792Sgshapiro case G_MIRROR_DISK_STATE_ACTIVE: 10890792Sgshapiro return ("ACTIVE"); 10990792Sgshapiro case G_MIRROR_DISK_STATE_STALE: 11090792Sgshapiro return ("STALE"); 11190792Sgshapiro case G_MIRROR_DISK_STATE_SYNCHRONIZING: 11290792Sgshapiro return ("SYNCHRONIZING"); 11390792Sgshapiro case G_MIRROR_DISK_STATE_DISCONNECTED: 11490792Sgshapiro return ("DISCONNECTED"); 11590792Sgshapiro case G_MIRROR_DISK_STATE_DESTROY: 11690792Sgshapiro return ("DESTROY"); 11790792Sgshapiro default: 11890792Sgshapiro return ("INVALID"); 11990792Sgshapiro } 12090792Sgshapiro} 12190792Sgshapiro 12290792Sgshapirostatic const char * 12390792Sgshapirog_mirror_device_state2str(int state) 12490792Sgshapiro{ 12590792Sgshapiro 12690792Sgshapiro switch (state) { 12790792Sgshapiro case G_MIRROR_DEVICE_STATE_STARTING: 12890792Sgshapiro return ("STARTING"); 12990792Sgshapiro case G_MIRROR_DEVICE_STATE_RUNNING: 13090792Sgshapiro return ("RUNNING"); 13190792Sgshapiro default: 13290792Sgshapiro return ("INVALID"); 13390792Sgshapiro } 13490792Sgshapiro} 13590792Sgshapiro 13690792Sgshapirostatic const char * 13790792Sgshapirog_mirror_get_diskname(struct g_mirror_disk *disk) 13890792Sgshapiro{ 13990792Sgshapiro 14090792Sgshapiro if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 14190792Sgshapiro return ("[unknown]"); 14290792Sgshapiro return (disk->d_name); 14390792Sgshapiro} 14490792Sgshapiro 14590792Sgshapiro/* 14690792Sgshapiro * --- Events handling functions --- 14790792Sgshapiro * Events in geom_mirror are used to maintain disks and device status 148 * from one thread to simplify locking. 149 */ 150static void 151g_mirror_event_free(struct g_mirror_event *ep) 152{ 153 154 free(ep, M_MIRROR); 155} 156 157int 158g_mirror_event_send(void *arg, int state, int flags) 159{ 160 struct g_mirror_softc *sc; 161 struct g_mirror_disk *disk; 162 struct g_mirror_event *ep; 163 int error; 164 165 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 166 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 167 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 168 disk = NULL; 169 sc = arg; 170 } else { 171 disk = arg; 172 sc = disk->d_softc; 173 } 174 ep->e_disk = disk; 175 ep->e_state = state; 176 ep->e_flags = flags; 177 ep->e_error = 0; 178 mtx_lock(&sc->sc_events_mtx); 179 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 180 mtx_unlock(&sc->sc_events_mtx); 181 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 182 mtx_lock(&sc->sc_queue_mtx); 183 wakeup(sc); 184 mtx_unlock(&sc->sc_queue_mtx); 185 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 186 return (0); 187 g_topology_assert(); 188 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 189 g_topology_unlock(); 190 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 191 mtx_lock(&sc->sc_events_mtx); 192 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 193 hz * 5); 194 } 195 /* Don't even try to use 'sc' here, because it could be already dead. */ 196 g_topology_lock(); 197 error = ep->e_error; 198 g_mirror_event_free(ep); 199 return (error); 200} 201 202static struct g_mirror_event * 203g_mirror_event_get(struct g_mirror_softc *sc) 204{ 205 struct g_mirror_event *ep; 206 207 mtx_lock(&sc->sc_events_mtx); 208 ep = TAILQ_FIRST(&sc->sc_events); 209 if (ep != NULL) 210 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 211 mtx_unlock(&sc->sc_events_mtx); 212 return (ep); 213} 214 215static void 216g_mirror_event_cancel(struct g_mirror_disk *disk) 217{ 218 struct g_mirror_softc *sc; 219 struct g_mirror_event *ep, *tmpep; 220 221 g_topology_assert(); 222 223 sc = disk->d_softc; 224 mtx_lock(&sc->sc_events_mtx); 225 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 226 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 227 continue; 228 if (ep->e_disk != disk) 229 continue; 230 TAILQ_REMOVE(&sc->sc_events, ep, e_next); 231 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 232 g_mirror_event_free(ep); 233 else { 234 ep->e_error = ECANCELED; 235 wakeup(ep); 236 } 237 } 238 mtx_unlock(&sc->sc_events_mtx); 239} 240 241/* 242 * Return the number of disks in given state. 243 * If state is equal to -1, count all connected disks. 244 */ 245u_int 246g_mirror_ndisks(struct g_mirror_softc *sc, int state) 247{ 248 struct g_mirror_disk *disk; 249 u_int n = 0; 250 251 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 252 if (state == -1 || disk->d_state == state) 253 n++; 254 } 255 return (n); 256} 257 258/* 259 * Find a disk in mirror by its disk ID. 260 */ 261static struct g_mirror_disk * 262g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 263{ 264 struct g_mirror_disk *disk; 265 266 g_topology_assert(); 267 268 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 269 if (disk->d_id == id) 270 return (disk); 271 } 272 return (NULL); 273} 274 275static u_int 276g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 277{ 278 struct bio *bp; 279 u_int nreqs = 0; 280 281 mtx_lock(&sc->sc_queue_mtx); 282 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 283 if (bp->bio_from == cp) 284 nreqs++; 285 } 286 mtx_unlock(&sc->sc_queue_mtx); 287 return (nreqs); 288} 289 290static int 291g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 292{ 293 294 if (cp->nstart != cp->nend) { 295 G_MIRROR_DEBUG(2, 296 "I/O requests for %s exist, can't destroy it now.", 297 cp->provider->name); 298 return (1); 299 } 300 if (g_mirror_nrequests(sc, cp) > 0) { 301 G_MIRROR_DEBUG(2, 302 "I/O requests for %s in queue, can't destroy it now.", 303 cp->provider->name); 304 return (1); 305 } 306 return (0); 307} 308 309static void 310g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 311{ 312 313 g_topology_assert(); 314 315 cp->private = NULL; 316 if (g_mirror_is_busy(sc, cp)) 317 return; 318 G_MIRROR_DEBUG(2, "Consumer %s destroyed.", cp->provider->name); 319 g_detach(cp); 320 g_destroy_consumer(cp); 321} 322 323static int 324g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 325{ 326 int error; 327 328 g_topology_assert(); 329 KASSERT(disk->d_consumer == NULL, 330 ("Disk already connected (device %s).", disk->d_softc->sc_name)); 331 332 disk->d_consumer = g_new_consumer(disk->d_softc->sc_geom); 333 disk->d_consumer->private = disk; 334 error = g_attach(disk->d_consumer, pp); 335 if (error != 0) 336 return (error); 337 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 338 return (0); 339} 340 341static void 342g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 343{ 344 345 g_topology_assert(); 346 347 if (cp == NULL) 348 return; 349 if (cp->provider != NULL) { 350 G_MIRROR_DEBUG(2, "Disk %s disconnected.", cp->provider->name); 351 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) { 352 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 353 cp->provider->name, -cp->acr, -cp->acw, -cp->ace, 354 0); 355 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 356 } 357 g_mirror_kill_consumer(sc, cp); 358 } else { 359 g_destroy_consumer(cp); 360 } 361} 362 363/* 364 * Initialize disk. This means allocate memory, create consumer, attach it 365 * to the provider and open access (r1w1e1) to it. 366 */ 367static struct g_mirror_disk * 368g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 369 struct g_mirror_metadata *md, int *errorp) 370{ 371 struct g_mirror_disk *disk; 372 int error; 373 374 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 375 if (disk == NULL) { 376 error = ENOMEM; 377 goto fail; 378 } 379 disk->d_softc = sc; 380 error = g_mirror_connect_disk(disk, pp); 381 if (error != 0) 382 goto fail; 383 disk->d_id = md->md_did; 384 disk->d_state = G_MIRROR_DISK_STATE_NONE; 385 disk->d_priority = md->md_priority; 386 disk->d_delay.sec = 0; 387 disk->d_delay.frac = 0; 388 binuptime(&disk->d_last_used); 389 disk->d_flags = md->md_dflags; 390 if (md->md_provider[0] != '\0') 391 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 392 disk->d_sync.ds_consumer = NULL; 393 disk->d_sync.ds_offset = md->md_sync_offset; 394 disk->d_sync.ds_offset_done = md->md_sync_offset; 395 disk->d_sync.ds_resync = -1; 396 disk->d_sync.ds_syncid = md->md_syncid; 397 if (errorp != NULL) 398 *errorp = 0; 399 return (disk); 400fail: 401 if (errorp != NULL) 402 *errorp = error; 403 if (disk != NULL) { 404 g_mirror_disconnect_consumer(sc, disk->d_consumer); 405 free(disk, M_MIRROR); 406 } 407 return (NULL); 408} 409 410static void 411g_mirror_destroy_disk(struct g_mirror_disk *disk) 412{ 413 struct g_mirror_softc *sc; 414 415 g_topology_assert(); 416 417 LIST_REMOVE(disk, d_next); 418 g_mirror_event_cancel(disk); 419 sc = disk->d_softc; 420 if (sc->sc_hint == disk) 421 sc->sc_hint = NULL; 422 switch (disk->d_state) { 423 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 424 g_mirror_sync_stop(disk, 1); 425 /* FALLTHROUGH */ 426 case G_MIRROR_DISK_STATE_NEW: 427 case G_MIRROR_DISK_STATE_STALE: 428 case G_MIRROR_DISK_STATE_ACTIVE: 429 g_mirror_disconnect_consumer(sc, disk->d_consumer); 430 free(disk, M_MIRROR); 431 break; 432 default: 433 KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 434 g_mirror_get_diskname(disk), 435 g_mirror_disk_state2str(disk->d_state))); 436 } 437} 438 439static void 440g_mirror_destroy_device(struct g_mirror_softc *sc) 441{ 442 struct g_mirror_disk *disk; 443 struct g_mirror_event *ep; 444 struct g_geom *gp; 445 struct g_consumer *cp, *tmpcp; 446 447 g_topology_assert(); 448 449 gp = sc->sc_geom; 450 if (sc->sc_provider != NULL) 451 g_mirror_destroy_provider(sc); 452 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 453 disk = LIST_FIRST(&sc->sc_disks)) { 454 g_mirror_destroy_disk(disk); 455 } 456 while ((ep = g_mirror_event_get(sc)) != NULL) { 457 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 458 g_mirror_event_free(ep); 459 else { 460 ep->e_error = ECANCELED; 461 ep->e_flags |= G_MIRROR_EVENT_DONE; 462 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 463 mtx_lock(&sc->sc_events_mtx); 464 wakeup(ep); 465 mtx_unlock(&sc->sc_events_mtx); 466 } 467 } 468 callout_drain(&sc->sc_callout); 469 gp->softc = NULL; 470 471 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 472 g_mirror_disconnect_consumer(sc, cp); 473 } 474 sc->sc_sync.ds_geom->softc = NULL; 475 g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 476 mtx_destroy(&sc->sc_queue_mtx); 477 mtx_destroy(&sc->sc_events_mtx); 478 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 479 g_wither_geom(gp, ENXIO); 480} 481 482static void 483g_mirror_orphan(struct g_consumer *cp) 484{ 485 struct g_mirror_disk *disk; 486 487 g_topology_assert(); 488 489 disk = cp->private; 490 if (disk == NULL) 491 return; 492 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 493 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 494 G_MIRROR_EVENT_DONTWAIT); 495} 496 497static void 498g_mirror_spoiled(struct g_consumer *cp) 499{ 500 struct g_mirror_disk *disk; 501 502 g_topology_assert(); 503 504 disk = cp->private; 505 if (disk == NULL) 506 return; 507 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 508 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 509 G_MIRROR_EVENT_DONTWAIT); 510} 511 512/* 513 * Function should return the next active disk on the list. 514 * It is possible that it will be the same disk as given. 515 * If there are no active disks on list, NULL is returned. 516 */ 517static __inline struct g_mirror_disk * 518g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 519{ 520 struct g_mirror_disk *dp; 521 522 for (dp = LIST_NEXT(disk, d_next); dp != disk; 523 dp = LIST_NEXT(dp, d_next)) { 524 if (dp == NULL) 525 dp = LIST_FIRST(&sc->sc_disks); 526 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 527 break; 528 } 529 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 530 return (NULL); 531 return (dp); 532} 533 534static struct g_mirror_disk * 535g_mirror_get_disk(struct g_mirror_softc *sc) 536{ 537 struct g_mirror_disk *disk; 538 539 if (sc->sc_hint == NULL) { 540 sc->sc_hint = LIST_FIRST(&sc->sc_disks); 541 if (sc->sc_hint == NULL) 542 return (NULL); 543 } 544 disk = sc->sc_hint; 545 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 546 disk = g_mirror_find_next(sc, disk); 547 if (disk == NULL) 548 return (NULL); 549 } 550 sc->sc_hint = g_mirror_find_next(sc, disk); 551 return (disk); 552} 553 554static int 555g_mirror_write_metadata(struct g_mirror_disk *disk, 556 struct g_mirror_metadata *md) 557{ 558 struct g_mirror_softc *sc; 559 struct g_consumer *cp; 560 off_t offset, length; 561 u_char *sector; 562 int close = 0, error = 0; 563 564 g_topology_assert(); 565 566 sc = disk->d_softc; 567 cp = disk->d_consumer; 568 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 569 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 570 length = cp->provider->sectorsize; 571 offset = cp->provider->mediasize - length; 572 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 573 /* 574 * Open consumer if it wasn't opened and remember to close it. 575 */ 576 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 577 error = g_access(cp, 0, 1, 1); 578 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 579 cp->provider->name, 0, 1, 1, error); 580 if (error == 0) 581 close = 1; 582#ifdef INVARIANTS 583 } else { 584 KASSERT(cp->acw > 0 && cp->ace > 0, 585 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 586 cp->acr, cp->acw, cp->ace)); 587#endif 588 } 589 if (error == 0) { 590 if (md != NULL) 591 mirror_metadata_encode(md, sector); 592 g_topology_unlock(); 593 error = g_write_data(cp, offset, sector, length); 594 g_topology_lock(); 595 } 596 free(sector, M_MIRROR); 597 if (close) { 598 g_access(cp, 0, -1, -1); 599 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 600 cp->provider->name, 0, -1, -1, 0); 601 } 602 if (error != 0) { 603 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 604 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 605 G_MIRROR_EVENT_DONTWAIT); 606 } 607 return (error); 608} 609 610static int 611g_mirror_clear_metadata(struct g_mirror_disk *disk) 612{ 613 int error; 614 615 g_topology_assert(); 616 error = g_mirror_write_metadata(disk, NULL); 617 if (error == 0) { 618 G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 619 g_mirror_get_diskname(disk)); 620 } else { 621 G_MIRROR_DEBUG(0, 622 "Cannot clear metadata on disk %s (error=%d).", 623 g_mirror_get_diskname(disk), error); 624 } 625 return (error); 626} 627 628void 629g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 630 struct g_mirror_metadata *md) 631{ 632 633 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 634 md->md_version = G_MIRROR_VERSION; 635 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 636 md->md_mid = sc->sc_id; 637 md->md_all = sc->sc_ndisks; 638 md->md_slice = sc->sc_slice; 639 md->md_balance = sc->sc_balance; 640 md->md_mediasize = sc->sc_mediasize; 641 md->md_sectorsize = sc->sc_sectorsize; 642 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 643 bzero(md->md_provider, sizeof(md->md_provider)); 644 if (disk == NULL) { 645 md->md_did = arc4random(); 646 md->md_priority = 0; 647 md->md_syncid = 0; 648 md->md_dflags = 0; 649 md->md_sync_offset = 0; 650 } else { 651 md->md_did = disk->d_id; 652 md->md_priority = disk->d_priority; 653 md->md_syncid = disk->d_sync.ds_syncid; 654 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 655 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 656 md->md_sync_offset = disk->d_sync.ds_offset_done; 657 else 658 md->md_sync_offset = 0; 659 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 660 strlcpy(md->md_provider, 661 disk->d_consumer->provider->name, 662 sizeof(md->md_provider)); 663 } 664 } 665} 666 667void 668g_mirror_update_metadata(struct g_mirror_disk *disk) 669{ 670 struct g_mirror_metadata md; 671 int error; 672 673 g_topology_assert(); 674 g_mirror_fill_metadata(disk->d_softc, disk, &md); 675 error = g_mirror_write_metadata(disk, &md); 676 if (error == 0) { 677 G_MIRROR_DEBUG(2, "Metadata on %s updated.", 678 g_mirror_get_diskname(disk)); 679 } else { 680 G_MIRROR_DEBUG(0, 681 "Cannot update metadata on disk %s (error=%d).", 682 g_mirror_get_diskname(disk), error); 683 } 684} 685 686static void 687g_mirror_bump_syncid(struct g_mirror_softc *sc) 688{ 689 struct g_mirror_disk *disk; 690 691 g_topology_assert(); 692 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 693 ("%s called with no active disks (device=%s).", __func__, 694 sc->sc_name)); 695 696 sc->sc_syncid++; 697 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 698 sc->sc_syncid); 699 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 700 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 701 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 702 disk->d_sync.ds_syncid = sc->sc_syncid; 703 g_mirror_update_metadata(disk); 704 } 705 } 706} 707 708static __inline int 709bintime_cmp(struct bintime *bt1, struct bintime *bt2) 710{ 711 712 if (bt1->sec < bt2->sec) 713 return (-1); 714 else if (bt1->sec > bt2->sec) 715 return (1); 716 if (bt1->frac < bt2->frac) 717 return (-1); 718 else if (bt1->frac > bt2->frac) 719 return (1); 720 return (0); 721} 722 723static void 724g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 725{ 726 727 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 728 return; 729 binuptime(&disk->d_delay); 730 bintime_sub(&disk->d_delay, &bp->bio_t0); 731} 732 733static void 734g_mirror_done(struct bio *bp) 735{ 736 struct g_mirror_softc *sc; 737 738 sc = bp->bio_from->geom->softc; 739 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 740 mtx_lock(&sc->sc_queue_mtx); 741 bioq_disksort(&sc->sc_queue, bp); 742 wakeup(sc); 743 mtx_unlock(&sc->sc_queue_mtx); 744} 745 746static void 747g_mirror_regular_request(struct bio *bp) 748{ 749 struct g_mirror_softc *sc; 750 struct g_mirror_disk *disk; 751 struct bio *pbp; 752 753 g_topology_assert_not(); 754 755 pbp = bp->bio_parent; 756 sc = pbp->bio_to->geom->softc; 757 disk = bp->bio_from->private; 758 if (disk == NULL) { 759 g_topology_lock(); 760 g_mirror_kill_consumer(sc, bp->bio_from); 761 g_topology_unlock(); 762 } else { 763 g_mirror_update_delay(disk, bp); 764 } 765 766 pbp->bio_inbed++; 767 KASSERT(pbp->bio_inbed <= pbp->bio_children, 768 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 769 pbp->bio_children)); 770 if (bp->bio_error == 0 && pbp->bio_error == 0) { 771 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 772 g_destroy_bio(bp); 773 if (pbp->bio_children == pbp->bio_inbed) { 774 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 775 pbp->bio_completed = pbp->bio_length; 776 g_io_deliver(pbp, pbp->bio_error); 777 } 778 return; 779 } else if (bp->bio_error != 0) { 780 if (pbp->bio_error == 0) 781 pbp->bio_error = bp->bio_error; 782 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 783 bp->bio_error); 784 if (disk != NULL) { 785 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 786 g_mirror_event_send(disk, 787 G_MIRROR_DISK_STATE_DISCONNECTED, 788 G_MIRROR_EVENT_DONTWAIT); 789 } 790 switch (pbp->bio_cmd) { 791 case BIO_DELETE: 792 case BIO_WRITE: 793 pbp->bio_inbed--; 794 pbp->bio_children--; 795 break; 796 } 797 } 798 g_destroy_bio(bp); 799 800 switch (pbp->bio_cmd) { 801 case BIO_READ: 802 if (pbp->bio_children == pbp->bio_inbed) { 803 pbp->bio_error = 0; 804 mtx_lock(&sc->sc_queue_mtx); 805 bioq_disksort(&sc->sc_queue, pbp); 806 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 807 wakeup(sc); 808 mtx_unlock(&sc->sc_queue_mtx); 809 } 810 break; 811 case BIO_DELETE: 812 case BIO_WRITE: 813 if (pbp->bio_children == 0) { 814 /* 815 * All requests failed. 816 */ 817 } else if (pbp->bio_inbed < pbp->bio_children) { 818 /* Do nothing. */ 819 break; 820 } else if (pbp->bio_children == pbp->bio_inbed) { 821 /* Some requests succeeded. */ 822 pbp->bio_error = 0; 823 pbp->bio_completed = pbp->bio_length; 824 } 825 g_io_deliver(pbp, pbp->bio_error); 826 break; 827 default: 828 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 829 break; 830 } 831} 832 833static void 834g_mirror_sync_done(struct bio *bp) 835{ 836 struct g_mirror_softc *sc; 837 838 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 839 sc = bp->bio_from->geom->softc; 840 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 841 mtx_lock(&sc->sc_queue_mtx); 842 bioq_disksort(&sc->sc_queue, bp); 843 wakeup(sc); 844 mtx_unlock(&sc->sc_queue_mtx); 845} 846 847static void 848g_mirror_start(struct bio *bp) 849{ 850 struct g_mirror_softc *sc; 851 852 sc = bp->bio_to->geom->softc; 853 /* 854 * If sc == NULL or there are no valid disks, provider's error 855 * should be set and g_mirror_start() should not be called at all. 856 */ 857 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 858 ("Provider's error should be set (error=%d)(mirror=%s).", 859 bp->bio_to->error, bp->bio_to->name)); 860 G_MIRROR_LOGREQ(3, bp, "Request received."); 861 862 switch (bp->bio_cmd) { 863 case BIO_READ: 864 case BIO_WRITE: 865 case BIO_DELETE: 866 break; 867 case BIO_GETATTR: 868 default: 869 g_io_deliver(bp, EOPNOTSUPP); 870 return; 871 } 872 mtx_lock(&sc->sc_queue_mtx); 873 bioq_disksort(&sc->sc_queue, bp); 874 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 875 wakeup(sc); 876 mtx_unlock(&sc->sc_queue_mtx); 877} 878 879/* 880 * Send one synchronization request. 881 */ 882static void 883g_mirror_sync_one(struct g_mirror_disk *disk) 884{ 885 struct g_mirror_softc *sc; 886 struct bio *bp; 887 888 sc = disk->d_softc; 889 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 890 ("Disk %s is not marked for synchronization.", 891 g_mirror_get_diskname(disk))); 892 893 bp = g_new_bio(); 894 if (bp == NULL) 895 return; 896 bp->bio_parent = NULL; 897 bp->bio_cmd = BIO_READ; 898 bp->bio_offset = disk->d_sync.ds_offset; 899 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 900 bp->bio_cflags = 0; 901 bp->bio_done = g_mirror_sync_done; 902 bp->bio_data = disk->d_sync.ds_data; 903 if (bp->bio_data == NULL) { 904 g_destroy_bio(bp); 905 return; 906 } 907 disk->d_sync.ds_offset += bp->bio_length; 908 bp->bio_to = sc->sc_provider; 909 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 910 g_io_request(bp, disk->d_sync.ds_consumer); 911} 912 913static void 914g_mirror_sync_request(struct bio *bp) 915{ 916 struct g_mirror_softc *sc; 917 struct g_mirror_disk *disk; 918 919 sc = bp->bio_from->geom->softc; 920 disk = bp->bio_from->private; 921 if (disk == NULL) { 922 g_topology_lock(); 923 g_mirror_kill_consumer(sc, bp->bio_from); 924 g_topology_unlock(); 925 g_destroy_bio(bp); 926 return; 927 } 928 929 /* 930 * Synchronization request. 931 */ 932 switch (bp->bio_cmd) { 933 case BIO_READ: 934 { 935 struct g_consumer *cp; 936 937 if (bp->bio_error != 0) { 938 G_MIRROR_LOGREQ(0, bp, 939 "Synchronization request failed (error=%d).", 940 bp->bio_error); 941 g_destroy_bio(bp); 942 return; 943 } 944 bp->bio_cmd = BIO_WRITE; 945 bp->bio_cflags = 0; 946 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 947 cp = disk->d_consumer; 948 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 949 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 950 cp->acr, cp->acw, cp->ace)); 951 g_io_request(bp, cp); 952 return; 953 } 954 case BIO_WRITE: 955 { 956 struct g_mirror_disk_sync *sync; 957 958 if (bp->bio_error != 0) { 959 G_MIRROR_LOGREQ(0, bp, 960 "Synchronization request failed (error=%d).", 961 bp->bio_error); 962 g_destroy_bio(bp); 963 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 964 g_mirror_event_send(disk, 965 G_MIRROR_DISK_STATE_DISCONNECTED, 966 G_MIRROR_EVENT_DONTWAIT); 967 return; 968 } 969 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 970 sync = &disk->d_sync; 971 sync->ds_offset_done = bp->bio_offset + bp->bio_length; 972 g_destroy_bio(bp); 973 if (sync->ds_resync != -1) 974 break; 975 if (sync->ds_offset_done == sc->sc_provider->mediasize) { 976 /* 977 * Disk up-to-date, activate it. 978 */ 979 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 980 G_MIRROR_EVENT_DONTWAIT); 981 return; 982 } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) { 983 /* 984 * Update offset_done on every 100 blocks. 985 * XXX: This should be configurable. 986 */ 987 g_topology_lock(); 988 g_mirror_update_metadata(disk); 989 g_topology_unlock(); 990 } 991 return; 992 } 993 default: 994 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 995 bp->bio_cmd, sc->sc_name)); 996 break; 997 } 998} 999 1000static void 1001g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1002{ 1003 struct g_mirror_disk *disk; 1004 struct g_consumer *cp; 1005 struct bio *cbp; 1006 1007 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1008 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1009 break; 1010 } 1011 if (disk == NULL) { 1012 if (bp->bio_error == 0) 1013 bp->bio_error = ENXIO; 1014 g_io_deliver(bp, bp->bio_error); 1015 return; 1016 } 1017 cbp = g_clone_bio(bp); 1018 if (cbp == NULL) { 1019 if (bp->bio_error == 0) 1020 bp->bio_error = ENOMEM; 1021 g_io_deliver(bp, bp->bio_error); 1022 return; 1023 } 1024 /* 1025 * Fill in the component buf structure. 1026 */ 1027 cp = disk->d_consumer; 1028 cbp->bio_done = g_mirror_done; 1029 cbp->bio_to = cp->provider; 1030 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1031 KASSERT(cp->acr > 0 && cp->ace > 0, 1032 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1033 cp->acw, cp->ace)); 1034 g_io_request(cbp, cp); 1035} 1036 1037static void 1038g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1039{ 1040 struct g_mirror_disk *disk; 1041 struct g_consumer *cp; 1042 struct bio *cbp; 1043 1044 disk = g_mirror_get_disk(sc); 1045 if (disk == NULL) { 1046 if (bp->bio_error == 0) 1047 bp->bio_error = ENXIO; 1048 g_io_deliver(bp, bp->bio_error); 1049 return; 1050 } 1051 cbp = g_clone_bio(bp); 1052 if (cbp == NULL) { 1053 if (bp->bio_error == 0) 1054 bp->bio_error = ENOMEM; 1055 g_io_deliver(bp, bp->bio_error); 1056 return; 1057 } 1058 /* 1059 * Fill in the component buf structure. 1060 */ 1061 cp = disk->d_consumer; 1062 cbp->bio_done = g_mirror_done; 1063 cbp->bio_to = cp->provider; 1064 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1065 KASSERT(cp->acr > 0 && cp->ace > 0, 1066 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1067 cp->acw, cp->ace)); 1068 g_io_request(cbp, cp); 1069} 1070 1071static void 1072g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1073{ 1074 struct g_mirror_disk *disk, *dp; 1075 struct g_consumer *cp; 1076 struct bio *cbp; 1077 struct bintime curtime; 1078 1079 binuptime(&curtime); 1080 /* 1081 * Find a disk which the smallest load. 1082 */ 1083 disk = NULL; 1084 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1085 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1086 continue; 1087 /* If disk wasn't used for more than 2 sec, use it. */ 1088 if (curtime.sec - dp->d_last_used.sec >= 2) { 1089 disk = dp; 1090 break; 1091 } 1092 if (disk == NULL || 1093 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 1094 disk = dp; 1095 } 1096 } 1097 cbp = g_clone_bio(bp); 1098 if (cbp == NULL) { 1099 if (bp->bio_error == 0) 1100 bp->bio_error = ENOMEM; 1101 g_io_deliver(bp, bp->bio_error); 1102 return; 1103 } 1104 /* 1105 * Fill in the component buf structure. 1106 */ 1107 cp = disk->d_consumer; 1108 cbp->bio_done = g_mirror_done; 1109 cbp->bio_to = cp->provider; 1110 binuptime(&disk->d_last_used); 1111 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1112 KASSERT(cp->acr > 0 && cp->ace > 0, 1113 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1114 cp->acw, cp->ace)); 1115 g_io_request(cbp, cp); 1116} 1117 1118static void 1119g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1120{ 1121 struct bio_queue_head queue; 1122 struct g_mirror_disk *disk; 1123 struct g_consumer *cp; 1124 struct bio *cbp; 1125 off_t left, mod, offset, slice; 1126 u_char *data; 1127 u_int ndisks; 1128 1129 if (bp->bio_length <= sc->sc_slice) { 1130 g_mirror_request_round_robin(sc, bp); 1131 return; 1132 } 1133 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1134 slice = bp->bio_length / ndisks; 1135 mod = slice % sc->sc_provider->sectorsize; 1136 if (mod != 0) 1137 slice += sc->sc_provider->sectorsize - mod; 1138 /* 1139 * Allocate all bios before sending any request, so we can 1140 * return ENOMEM in nice and clean way. 1141 */ 1142 left = bp->bio_length; 1143 offset = bp->bio_offset; 1144 data = bp->bio_data; 1145 bioq_init(&queue); 1146 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1147 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1148 continue; 1149 cbp = g_clone_bio(bp); 1150 if (cbp == NULL) { 1151 for (cbp = bioq_first(&queue); cbp != NULL; 1152 cbp = bioq_first(&queue)) { 1153 bioq_remove(&queue, cbp); 1154 g_destroy_bio(cbp); 1155 } 1156 if (bp->bio_error == 0) 1157 bp->bio_error = ENOMEM; 1158 g_io_deliver(bp, bp->bio_error); 1159 return; 1160 } 1161 bioq_insert_tail(&queue, cbp); 1162 cbp->bio_done = g_mirror_done; 1163 cbp->bio_caller1 = disk; 1164 cbp->bio_to = disk->d_consumer->provider; 1165 cbp->bio_offset = offset; 1166 cbp->bio_data = data; 1167 cbp->bio_length = MIN(left, slice); 1168 left -= cbp->bio_length; 1169 if (left == 0) 1170 break; 1171 offset += cbp->bio_length; 1172 data += cbp->bio_length; 1173 } 1174 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1175 bioq_remove(&queue, cbp); 1176 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1177 disk = cbp->bio_caller1; 1178 cbp->bio_caller1 = NULL; 1179 cp = disk->d_consumer; 1180 KASSERT(cp->acr > 0 && cp->ace > 0, 1181 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1182 cp->acr, cp->acw, cp->ace)); 1183 g_io_request(cbp, disk->d_consumer); 1184 } 1185} 1186 1187static void 1188g_mirror_register_request(struct bio *bp) 1189{ 1190 struct g_mirror_softc *sc; 1191 1192 sc = bp->bio_to->geom->softc; 1193 switch (bp->bio_cmd) { 1194 case BIO_READ: 1195 switch (sc->sc_balance) { 1196 case G_MIRROR_BALANCE_LOAD: 1197 g_mirror_request_load(sc, bp); 1198 break; 1199 case G_MIRROR_BALANCE_PREFER: 1200 g_mirror_request_prefer(sc, bp); 1201 break; 1202 case G_MIRROR_BALANCE_ROUND_ROBIN: 1203 g_mirror_request_round_robin(sc, bp); 1204 break; 1205 case G_MIRROR_BALANCE_SPLIT: 1206 g_mirror_request_split(sc, bp); 1207 break; 1208 } 1209 return; 1210 case BIO_WRITE: 1211 case BIO_DELETE: 1212 { 1213 struct g_mirror_disk *disk; 1214 struct g_mirror_disk_sync *sync; 1215 struct bio_queue_head queue; 1216 struct g_consumer *cp; 1217 struct bio *cbp; 1218 1219 /* 1220 * Allocate all bios before sending any request, so we can 1221 * return ENOMEM in nice and clean way. 1222 */ 1223 bioq_init(&queue); 1224 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1225 sync = &disk->d_sync; 1226 switch (disk->d_state) { 1227 case G_MIRROR_DISK_STATE_ACTIVE: 1228 break; 1229 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1230 if (bp->bio_offset >= sync->ds_offset) 1231 continue; 1232 else if (bp->bio_offset + bp->bio_length > 1233 sync->ds_offset_done && 1234 (bp->bio_offset < sync->ds_resync || 1235 sync->ds_resync == -1)) { 1236 sync->ds_resync = bp->bio_offset - 1237 (bp->bio_offset % MAXPHYS); 1238 } 1239 break; 1240 default: 1241 continue; 1242 } 1243 cbp = g_clone_bio(bp); 1244 if (cbp == NULL) { 1245 for (cbp = bioq_first(&queue); cbp != NULL; 1246 cbp = bioq_first(&queue)) { 1247 bioq_remove(&queue, cbp); 1248 g_destroy_bio(cbp); 1249 } 1250 if (bp->bio_error == 0) 1251 bp->bio_error = ENOMEM; 1252 g_io_deliver(bp, bp->bio_error); 1253 return; 1254 } 1255 bioq_insert_tail(&queue, cbp); 1256 cbp->bio_done = g_mirror_done; 1257 cp = disk->d_consumer; 1258 cbp->bio_caller1 = cp; 1259 cbp->bio_to = cp->provider; 1260 KASSERT(cp->acw > 0 && cp->ace > 0, 1261 ("Consumer %s not opened (r%dw%de%d).", 1262 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1263 } 1264 for (cbp = bioq_first(&queue); cbp != NULL; 1265 cbp = bioq_first(&queue)) { 1266 bioq_remove(&queue, cbp); 1267 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1268 cp = cbp->bio_caller1; 1269 cbp->bio_caller1 = NULL; 1270 g_io_request(cbp, cp); 1271 } 1272 /* 1273 * Bump syncid on first write. 1274 */ 1275 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) { 1276 sc->sc_bump_syncid = 0; 1277 g_topology_lock(); 1278 g_mirror_bump_syncid(sc); 1279 g_topology_unlock(); 1280 } 1281 return; 1282 } 1283 default: 1284 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1285 bp->bio_cmd, sc->sc_name)); 1286 break; 1287 } 1288} 1289 1290static int 1291g_mirror_can_destroy(struct g_mirror_softc *sc) 1292{ 1293 struct g_geom *gp; 1294 struct g_consumer *cp; 1295 1296 g_topology_assert(); 1297 gp = sc->sc_geom; 1298 LIST_FOREACH(cp, &gp->consumer, consumer) { 1299 if (g_mirror_is_busy(sc, cp)) 1300 return (0); 1301 } 1302 gp = sc->sc_sync.ds_geom; 1303 LIST_FOREACH(cp, &gp->consumer, consumer) { 1304 if (g_mirror_is_busy(sc, cp)) 1305 return (0); 1306 } 1307 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1308 sc->sc_name); 1309 return (1); 1310} 1311 1312static int 1313g_mirror_try_destroy(struct g_mirror_softc *sc) 1314{ 1315 1316 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1317 g_topology_lock(); 1318 if (!g_mirror_can_destroy(sc)) { 1319 g_topology_unlock(); 1320 return (0); 1321 } 1322 g_topology_unlock(); 1323 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1324 &sc->sc_worker); 1325 wakeup(&sc->sc_worker); 1326 sc->sc_worker = NULL; 1327 } else { 1328 g_topology_lock(); 1329 if (!g_mirror_can_destroy(sc)) { 1330 g_topology_unlock(); 1331 return (0); 1332 } 1333 g_mirror_destroy_device(sc); 1334 g_topology_unlock(); 1335 free(sc, M_MIRROR); 1336 } 1337 return (1); 1338} 1339 1340/* 1341 * Worker thread. 1342 */ 1343static void 1344g_mirror_worker(void *arg) 1345{ 1346 struct g_mirror_softc *sc; 1347 struct g_mirror_disk *disk; 1348 struct g_mirror_disk_sync *sync; 1349 struct g_mirror_event *ep; 1350 struct bio *bp; 1351 u_int nreqs; 1352 1353 sc = arg; 1354 curthread->td_base_pri = PRIBIO; 1355 1356 nreqs = 0; 1357 for (;;) { 1358 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1359 /* 1360 * First take a look at events. 1361 * This is important to handle events before any I/O requests. 1362 */ 1363 ep = g_mirror_event_get(sc); 1364 if (ep != NULL) { 1365 g_topology_lock(); 1366 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1367 /* Update only device status. */ 1368 G_MIRROR_DEBUG(3, 1369 "Running event for device %s.", 1370 sc->sc_name); 1371 ep->e_error = 0; 1372 g_mirror_update_device(sc, 1); 1373 } else { 1374 /* Update disk status. */ 1375 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1376 g_mirror_get_diskname(ep->e_disk)); 1377 ep->e_error = g_mirror_update_disk(ep->e_disk, 1378 ep->e_state); 1379 if (ep->e_error == 0) 1380 g_mirror_update_device(sc, 0); 1381 } 1382 g_topology_unlock(); 1383 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1384 KASSERT(ep->e_error == 0, 1385 ("Error cannot be handled.")); 1386 g_mirror_event_free(ep); 1387 } else { 1388 ep->e_flags |= G_MIRROR_EVENT_DONE; 1389 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1390 ep); 1391 mtx_lock(&sc->sc_events_mtx); 1392 wakeup(ep); 1393 mtx_unlock(&sc->sc_events_mtx); 1394 } 1395 if ((sc->sc_flags & 1396 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1397 if (g_mirror_try_destroy(sc)) 1398 kthread_exit(0); 1399 } 1400 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1401 continue; 1402 } 1403 /* 1404 * Now I/O requests. 1405 */ 1406 /* Get first request from the queue. */ 1407 mtx_lock(&sc->sc_queue_mtx); 1408 bp = bioq_first(&sc->sc_queue); 1409 if (bp == NULL) { 1410 if ((sc->sc_flags & 1411 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1412 mtx_unlock(&sc->sc_queue_mtx); 1413 if (g_mirror_try_destroy(sc)) 1414 kthread_exit(0); 1415 mtx_lock(&sc->sc_queue_mtx); 1416 } 1417 } 1418 if (sc->sc_sync.ds_ndisks > 0 && 1419 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 1420 mtx_unlock(&sc->sc_queue_mtx); 1421 /* 1422 * It is time for synchronization... 1423 */ 1424 nreqs = 0; 1425 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1426 if (disk->d_state != 1427 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 1428 continue; 1429 } 1430 sync = &disk->d_sync; 1431 if (sync->ds_offset >= 1432 sc->sc_provider->mediasize) { 1433 continue; 1434 } 1435 if (sync->ds_offset > sync->ds_offset_done) 1436 continue; 1437 if (sync->ds_resync != -1) { 1438 sync->ds_offset = sync->ds_resync; 1439 sync->ds_offset_done = sync->ds_resync; 1440 sync->ds_resync = -1; 1441 } 1442 g_mirror_sync_one(disk); 1443 } 1444 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 1445 goto sleep; 1446 } 1447 if (bp == NULL) { 1448 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 0); 1449 G_MIRROR_DEBUG(5, "%s: I'm here 3.", __func__); 1450 continue; 1451 } 1452 nreqs++; 1453 bioq_remove(&sc->sc_queue, bp); 1454 mtx_unlock(&sc->sc_queue_mtx); 1455 1456 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 1457 g_mirror_regular_request(bp); 1458 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1459 u_int timeout, sps; 1460 1461 g_mirror_sync_request(bp); 1462sleep: 1463 sps = atomic_load_acq_int(&g_mirror_syncs_per_sec); 1464 if (sps == 0) { 1465 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 1466 continue; 1467 } 1468 mtx_lock(&sc->sc_queue_mtx); 1469 if (bioq_first(&sc->sc_queue) != NULL) { 1470 mtx_unlock(&sc->sc_queue_mtx); 1471 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1472 continue; 1473 } 1474 timeout = hz / sps; 1475 if (timeout == 0) 1476 timeout = 1; 1477 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w2", 1478 timeout); 1479 } else { 1480 g_mirror_register_request(bp); 1481 } 1482 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 1483 } 1484} 1485 1486/* 1487 * Open disk's consumer if needed. 1488 */ 1489static void 1490g_mirror_update_access(struct g_mirror_disk *disk) 1491{ 1492 struct g_provider *pp; 1493 struct g_consumer *cp; 1494 int acr, acw, ace, cpw, error; 1495 1496 g_topology_assert(); 1497 1498 cp = disk->d_consumer; 1499 pp = disk->d_softc->sc_provider; 1500 if (pp == NULL) { 1501 acr = -cp->acr; 1502 acw = -cp->acw; 1503 ace = -cp->ace; 1504 } else { 1505 acr = pp->acr - cp->acr; 1506 acw = pp->acw - cp->acw; 1507 ace = pp->ace - cp->ace; 1508 /* Grab an extra "exclusive" bit. */ 1509 if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) 1510 ace++; 1511 } 1512 if (acr == 0 && acw == 0 && ace == 0) 1513 return; 1514 cpw = cp->acw; 1515 error = g_access(cp, acr, acw, ace); 1516 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, acr, 1517 acw, ace, error); 1518 if (error != 0) { 1519 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1520 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1521 G_MIRROR_EVENT_DONTWAIT); 1522 return; 1523 } 1524 if (cpw == 0 && cp->acw > 0) { 1525 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1526 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1527 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1528 } else if (cpw > 0 && cp->acw == 0) { 1529 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1530 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1531 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1532 } 1533} 1534 1535static void 1536g_mirror_sync_start(struct g_mirror_disk *disk) 1537{ 1538 struct g_mirror_softc *sc; 1539 struct g_consumer *cp; 1540 int error; 1541 1542 g_topology_assert(); 1543 1544 sc = disk->d_softc; 1545 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1546 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1547 sc->sc_state)); 1548 cp = disk->d_consumer; 1549 KASSERT(cp->acr == 0 && cp->acw == 0 && cp->ace == 0, 1550 ("Consumer %s already opened.", cp->provider->name)); 1551 1552 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1553 g_mirror_get_diskname(disk)); 1554 error = g_access(cp, 0, 1, 1); 1555 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, 1, 1556 1, error); 1557 if (error != 0) { 1558 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1559 G_MIRROR_EVENT_DONTWAIT); 1560 return; 1561 } 1562 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1563 KASSERT(disk->d_sync.ds_consumer == NULL, 1564 ("Sync consumer already exists (device=%s, disk=%s).", 1565 sc->sc_name, g_mirror_get_diskname(disk))); 1566 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1567 disk->d_sync.ds_consumer->private = disk; 1568 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1569 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1570 disk->d_softc->sc_name, error)); 1571 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1572 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1573 disk->d_softc->sc_name, error)); 1574 disk->d_sync.ds_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1575 sc->sc_sync.ds_ndisks++; 1576} 1577 1578/* 1579 * Stop synchronization process. 1580 * type: 0 - synchronization finished 1581 * 1 - synchronization stopped 1582 */ 1583static void 1584g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1585{ 1586 struct g_consumer *cp; 1587 1588 g_topology_assert(); 1589 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1590 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1591 g_mirror_disk_state2str(disk->d_state))); 1592 if (disk->d_sync.ds_consumer == NULL) 1593 return; 1594 1595 if (type == 0) { 1596 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1597 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1598 } else /* if (type == 1) */ { 1599 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1600 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1601 } 1602 cp = disk->d_sync.ds_consumer; 1603 g_access(cp, -1, 0, 0); 1604 g_mirror_kill_consumer(disk->d_softc, cp); 1605 free(disk->d_sync.ds_data, M_MIRROR); 1606 disk->d_sync.ds_consumer = NULL; 1607 disk->d_softc->sc_sync.ds_ndisks--; 1608 cp = disk->d_consumer; 1609 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 1610 ("Consumer %s not opened.", cp->provider->name)); 1611 g_access(cp, 0, -1, -1); 1612 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, -1, 1613 -1, 0); 1614 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1615} 1616 1617static void 1618g_mirror_launch_provider(struct g_mirror_softc *sc) 1619{ 1620 struct g_mirror_disk *disk; 1621 struct g_provider *pp; 1622 1623 g_topology_assert(); 1624 1625 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1626 pp->mediasize = sc->sc_mediasize; 1627 pp->sectorsize = sc->sc_sectorsize; 1628 sc->sc_provider = pp; 1629 g_error_provider(pp, 0); 1630 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1631 pp->name); 1632 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1633 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1634 g_mirror_sync_start(disk); 1635 } 1636} 1637 1638static void 1639g_mirror_destroy_provider(struct g_mirror_softc *sc) 1640{ 1641 struct g_mirror_disk *disk; 1642 struct bio *bp; 1643 1644 g_topology_assert(); 1645 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1646 sc->sc_name)); 1647 1648 g_error_provider(sc->sc_provider, ENXIO); 1649 mtx_lock(&sc->sc_queue_mtx); 1650 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1651 bioq_remove(&sc->sc_queue, bp); 1652 g_io_deliver(bp, ENXIO); 1653 } 1654 mtx_unlock(&sc->sc_queue_mtx); 1655 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1656 sc->sc_provider->name); 1657 sc->sc_provider->flags |= G_PF_WITHER; 1658 g_orphan_provider(sc->sc_provider, ENXIO); 1659 sc->sc_provider = NULL; 1660 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1661 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1662 g_mirror_sync_stop(disk, 1); 1663 } 1664} 1665 1666static void 1667g_mirror_go(void *arg) 1668{ 1669 struct g_mirror_softc *sc; 1670 1671 sc = arg; 1672 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1673 g_mirror_event_send(sc, 0, 1674 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1675} 1676 1677static u_int 1678g_mirror_determine_state(struct g_mirror_disk *disk) 1679{ 1680 struct g_mirror_softc *sc; 1681 u_int state; 1682 1683 sc = disk->d_softc; 1684 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1685 if ((disk->d_flags & 1686 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1687 /* Disk does not need synchronization. */ 1688 state = G_MIRROR_DISK_STATE_ACTIVE; 1689 } else { 1690 if ((sc->sc_flags & 1691 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1692 (disk->d_flags & 1693 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1694 /* 1695 * We can start synchronization from 1696 * the stored offset. 1697 */ 1698 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1699 } else { 1700 state = G_MIRROR_DISK_STATE_STALE; 1701 } 1702 } 1703 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1704 /* 1705 * Reset all synchronization data for this disk, 1706 * because if it even was synchronized, it was 1707 * synchronized to disks with different syncid. 1708 */ 1709 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1710 disk->d_sync.ds_offset = 0; 1711 disk->d_sync.ds_offset_done = 0; 1712 disk->d_sync.ds_syncid = sc->sc_syncid; 1713 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1714 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1715 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1716 } else { 1717 state = G_MIRROR_DISK_STATE_STALE; 1718 } 1719 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1720 /* 1721 * Not good, NOT GOOD! 1722 * It means that mirror was started on stale disks 1723 * and more fresh disk just arrive. 1724 * If there were writes, mirror is fucked up, sorry. 1725 * I think the best choice here is don't touch 1726 * this disk and inform the user laudly. 1727 */ 1728 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1729 "disk (%s) arrives!! It will not be connected to the " 1730 "running device.", sc->sc_name, 1731 g_mirror_get_diskname(disk)); 1732 g_mirror_destroy_disk(disk); 1733 state = G_MIRROR_DISK_STATE_NONE; 1734 /* Return immediately, because disk was destroyed. */ 1735 return (state); 1736 } 1737 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1738 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1739 return (state); 1740} 1741 1742/* 1743 * Update device state. 1744 */ 1745static void 1746g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1747{ 1748 struct g_mirror_disk *disk; 1749 u_int state; 1750 1751 g_topology_assert(); 1752 1753 switch (sc->sc_state) { 1754 case G_MIRROR_DEVICE_STATE_STARTING: 1755 { 1756 struct g_mirror_disk *pdisk; 1757 u_int dirty, ndisks, syncid; 1758 1759 KASSERT(sc->sc_provider == NULL, 1760 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1761 /* 1762 * Are we ready? We are, if all disks are connected or 1763 * if we have any disks and 'force' is true. 1764 */ 1765 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1766 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1767 ; 1768 } else if (g_mirror_ndisks(sc, -1) == 0) { 1769 /* 1770 * Disks went down in starting phase, so destroy 1771 * device. 1772 */ 1773 callout_drain(&sc->sc_callout); 1774 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1775 return; 1776 } else { 1777 return; 1778 } 1779 1780 /* 1781 * Activate all disks with the biggest syncid. 1782 */ 1783 if (force) { 1784 /* 1785 * If 'force' is true, we have been called due to 1786 * timeout, so don't bother canceling timeout. 1787 */ 1788 ndisks = 0; 1789 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1790 if ((disk->d_flags & 1791 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1792 ndisks++; 1793 } 1794 } 1795 if (ndisks == 0) { 1796 /* No valid disks found, destroy device. */ 1797 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1798 return; 1799 } 1800 } else { 1801 /* Cancel timeout. */ 1802 callout_drain(&sc->sc_callout); 1803 } 1804 1805 /* 1806 * Find disk with the biggest syncid. 1807 */ 1808 syncid = 0; 1809 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1810 if (disk->d_sync.ds_syncid > syncid) 1811 syncid = disk->d_sync.ds_syncid; 1812 } 1813 1814 /* 1815 * Here we need to look for dirty disks and if all disks 1816 * with the biggest syncid are dirty, we have to choose 1817 * one with the biggest priority and rebuild the rest. 1818 */ 1819 /* 1820 * Find the number of dirty disks with the biggest syncid. 1821 * Find the number of disks with the biggest syncid. 1822 * While here, find a disk with the biggest priority. 1823 */ 1824 dirty = ndisks = 0; 1825 pdisk = NULL; 1826 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1827 if (disk->d_sync.ds_syncid != syncid) 1828 continue; 1829 if ((disk->d_flags & 1830 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1831 continue; 1832 } 1833 ndisks++; 1834 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1835 dirty++; 1836 if (pdisk == NULL || 1837 pdisk->d_priority < disk->d_priority) { 1838 pdisk = disk; 1839 } 1840 } 1841 } 1842 if (dirty == 0) { 1843 /* No dirty disks at all, great. */ 1844 } else if (dirty == ndisks) { 1845 /* 1846 * Force synchronization for all dirty disks except one 1847 * with the biggest priority. 1848 */ 1849 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1850 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1851 "master disk for synchronization.", 1852 g_mirror_get_diskname(pdisk), sc->sc_name); 1853 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1854 if (disk->d_sync.ds_syncid != syncid) 1855 continue; 1856 if ((disk->d_flags & 1857 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1858 continue; 1859 } 1860 KASSERT((disk->d_flags & 1861 G_MIRROR_DISK_FLAG_DIRTY) != 0, 1862 ("Disk %s isn't marked as dirty.", 1863 g_mirror_get_diskname(disk))); 1864 /* Skip the disk with the biggest priority. */ 1865 if (disk == pdisk) 1866 continue; 1867 disk->d_sync.ds_syncid = 0; 1868 } 1869 } else if (dirty < ndisks) { 1870 /* 1871 * Force synchronization for all dirty disks. 1872 * We have some non-dirty disks. 1873 */ 1874 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1875 if (disk->d_sync.ds_syncid != syncid) 1876 continue; 1877 if ((disk->d_flags & 1878 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1879 continue; 1880 } 1881 if ((disk->d_flags & 1882 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1883 continue; 1884 } 1885 disk->d_sync.ds_syncid = 0; 1886 } 1887 } 1888 1889 /* Reset hint. */ 1890 sc->sc_hint = NULL; 1891 sc->sc_syncid = syncid; 1892 if (force) { 1893 /* Remember to bump syncid on first write. */ 1894 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1895 } 1896 state = G_MIRROR_DEVICE_STATE_RUNNING; 1897 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 1898 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 1899 g_mirror_device_state2str(state)); 1900 sc->sc_state = state; 1901 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1902 state = g_mirror_determine_state(disk); 1903 g_mirror_event_send(disk, state, 1904 G_MIRROR_EVENT_DONTWAIT); 1905 if (state == G_MIRROR_DISK_STATE_STALE) { 1906 sc->sc_bump_syncid = 1907 G_MIRROR_BUMP_ON_FIRST_WRITE; 1908 } 1909 } 1910 wakeup(&g_mirror_class); 1911 break; 1912 } 1913 case G_MIRROR_DEVICE_STATE_RUNNING: 1914 if (g_mirror_ndisks(sc, -1) == 0) { 1915 /* 1916 * No disks at all, we need to destroy device. 1917 */ 1918 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1919 break; 1920 } else if (g_mirror_ndisks(sc, 1921 G_MIRROR_DISK_STATE_ACTIVE) == 0 && 1922 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 1923 /* 1924 * No active disks, destroy provider. 1925 */ 1926 if (sc->sc_provider != NULL) 1927 g_mirror_destroy_provider(sc); 1928 break; 1929 } else if (g_mirror_ndisks(sc, 1930 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 1931 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 1932 /* 1933 * We have active disks, launch provider if it doesn't 1934 * exist. 1935 */ 1936 if (sc->sc_provider == NULL) 1937 g_mirror_launch_provider(sc); 1938 } 1939 /* 1940 * Bump syncid here, if we need to do it immediately. 1941 */ 1942 if (sc->sc_bump_syncid == G_MIRROR_BUMP_IMMEDIATELY) { 1943 sc->sc_bump_syncid = 0; 1944 g_mirror_bump_syncid(sc); 1945 } 1946 break; 1947 default: 1948 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 1949 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 1950 break; 1951 } 1952} 1953 1954/* 1955 * Update disk state and device state if needed. 1956 */ 1957#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 1958 "Disk %s state changed from %s to %s (device %s).", \ 1959 g_mirror_get_diskname(disk), \ 1960 g_mirror_disk_state2str(disk->d_state), \ 1961 g_mirror_disk_state2str(state), sc->sc_name) 1962static int 1963g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 1964{ 1965 struct g_mirror_softc *sc; 1966 1967 g_topology_assert(); 1968 1969 sc = disk->d_softc; 1970again: 1971 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 1972 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 1973 g_mirror_disk_state2str(state)); 1974 switch (state) { 1975 case G_MIRROR_DISK_STATE_NEW: 1976 /* 1977 * Possible scenarios: 1978 * 1. New disk arrive. 1979 */ 1980 /* Previous state should be NONE. */ 1981 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 1982 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1983 g_mirror_disk_state2str(disk->d_state))); 1984 DISK_STATE_CHANGED(); 1985 1986 disk->d_state = state; 1987 if (LIST_EMPTY(&sc->sc_disks)) 1988 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 1989 else { 1990 struct g_mirror_disk *dp; 1991 1992 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1993 if (disk->d_priority >= dp->d_priority) { 1994 LIST_INSERT_BEFORE(dp, disk, d_next); 1995 dp = NULL; 1996 break; 1997 } 1998 if (LIST_NEXT(dp, d_next) == NULL) 1999 break; 2000 } 2001 if (dp != NULL) 2002 LIST_INSERT_AFTER(dp, disk, d_next); 2003 } 2004 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2005 sc->sc_name, g_mirror_get_diskname(disk)); 2006 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2007 break; 2008 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2009 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2010 g_mirror_device_state2str(sc->sc_state), 2011 g_mirror_get_diskname(disk), 2012 g_mirror_disk_state2str(disk->d_state))); 2013 state = g_mirror_determine_state(disk); 2014 if (state != G_MIRROR_DISK_STATE_NONE) 2015 goto again; 2016 break; 2017 case G_MIRROR_DISK_STATE_ACTIVE: 2018 /* 2019 * Possible scenarios: 2020 * 1. New disk does not need synchronization. 2021 * 2. Synchronization process finished successfully. 2022 */ 2023 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2024 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2025 g_mirror_device_state2str(sc->sc_state), 2026 g_mirror_get_diskname(disk), 2027 g_mirror_disk_state2str(disk->d_state))); 2028 /* Previous state should be NEW or SYNCHRONIZING. */ 2029 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2030 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2031 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2032 g_mirror_disk_state2str(disk->d_state))); 2033 DISK_STATE_CHANGED(); 2034 2035 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2036 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2037 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2038 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2039 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2040 g_mirror_sync_stop(disk, 0); 2041 } 2042 disk->d_state = state; 2043 disk->d_sync.ds_offset = 0; 2044 disk->d_sync.ds_offset_done = 0; 2045 g_mirror_update_access(disk); 2046 g_mirror_update_metadata(disk); 2047 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2048 sc->sc_name, g_mirror_get_diskname(disk)); 2049 break; 2050 case G_MIRROR_DISK_STATE_STALE: 2051 /* 2052 * Possible scenarios: 2053 * 1. Stale disk was connected. 2054 */ 2055 /* Previous state should be NEW. */ 2056 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2057 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2058 g_mirror_disk_state2str(disk->d_state))); 2059 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2060 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2061 g_mirror_device_state2str(sc->sc_state), 2062 g_mirror_get_diskname(disk), 2063 g_mirror_disk_state2str(disk->d_state))); 2064 /* 2065 * STALE state is only possible if device is marked 2066 * NOAUTOSYNC. 2067 */ 2068 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2069 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2070 g_mirror_device_state2str(sc->sc_state), 2071 g_mirror_get_diskname(disk), 2072 g_mirror_disk_state2str(disk->d_state))); 2073 DISK_STATE_CHANGED(); 2074 2075 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2076 disk->d_state = state; 2077 g_mirror_update_metadata(disk); 2078 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2079 sc->sc_name, g_mirror_get_diskname(disk)); 2080 break; 2081 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2082 /* 2083 * Possible scenarios: 2084 * 1. Disk which needs synchronization was connected. 2085 */ 2086 /* Previous state should be NEW. */ 2087 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2088 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2089 g_mirror_disk_state2str(disk->d_state))); 2090 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2091 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2092 g_mirror_device_state2str(sc->sc_state), 2093 g_mirror_get_diskname(disk), 2094 g_mirror_disk_state2str(disk->d_state))); 2095 DISK_STATE_CHANGED(); 2096 2097 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2098 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2099 disk->d_state = state; 2100 if (sc->sc_provider != NULL) { 2101 g_mirror_sync_start(disk); 2102 g_mirror_update_metadata(disk); 2103 } 2104 break; 2105 case G_MIRROR_DISK_STATE_DISCONNECTED: 2106 /* 2107 * Possible scenarios: 2108 * 1. Device wasn't running yet, but disk disappear. 2109 * 2. Disk was active and disapppear. 2110 * 3. Disk disappear during synchronization process. 2111 */ 2112 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2113 /* 2114 * Previous state should be ACTIVE, STALE or 2115 * SYNCHRONIZING. 2116 */ 2117 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2118 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2119 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2120 ("Wrong disk state (%s, %s).", 2121 g_mirror_get_diskname(disk), 2122 g_mirror_disk_state2str(disk->d_state))); 2123 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2124 /* Previous state should be NEW. */ 2125 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2126 ("Wrong disk state (%s, %s).", 2127 g_mirror_get_diskname(disk), 2128 g_mirror_disk_state2str(disk->d_state))); 2129 /* 2130 * Reset bumping syncid if disk disappeared in STARTING 2131 * state. 2132 */ 2133 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) 2134 sc->sc_bump_syncid = 0; 2135#ifdef INVARIANTS 2136 } else { 2137 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2138 sc->sc_name, 2139 g_mirror_device_state2str(sc->sc_state), 2140 g_mirror_get_diskname(disk), 2141 g_mirror_disk_state2str(disk->d_state))); 2142#endif 2143 } 2144 DISK_STATE_CHANGED(); 2145 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2146 sc->sc_name, g_mirror_get_diskname(disk)); 2147 2148 g_mirror_destroy_disk(disk); 2149 break; 2150 case G_MIRROR_DISK_STATE_DESTROY: 2151 { 2152 int error; 2153 2154 error = g_mirror_clear_metadata(disk); 2155 if (error != 0) 2156 return (error); 2157 DISK_STATE_CHANGED(); 2158 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2159 sc->sc_name, g_mirror_get_diskname(disk)); 2160 2161 g_mirror_destroy_disk(disk); 2162 sc->sc_ndisks--; 2163 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2164 g_mirror_update_metadata(disk); 2165 } 2166 break; 2167 } 2168 default: 2169 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2170 break; 2171 } 2172 return (0); 2173} 2174#undef DISK_STATE_CHANGED 2175 2176static int 2177g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2178{ 2179 struct g_provider *pp; 2180 u_char *buf; 2181 int error; 2182 2183 g_topology_assert(); 2184 2185 error = g_access(cp, 1, 0, 0); 2186 if (error != 0) 2187 return (error); 2188 pp = cp->provider; 2189 g_topology_unlock(); 2190 /* Metadata are stored on last sector. */ 2191 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2192 &error); 2193 g_topology_lock(); 2194 if (buf == NULL) { 2195 g_access(cp, -1, 0, 0); 2196 return (error); 2197 } 2198 if (error != 0) { 2199 g_access(cp, -1, 0, 0); 2200 g_free(buf); 2201 return (error); 2202 } 2203 error = g_access(cp, -1, 0, 0); 2204 KASSERT(error == 0, ("Cannot decrease access count for %s.", pp->name)); 2205 2206 /* Decode metadata. */ 2207 error = mirror_metadata_decode(buf, md); 2208 g_free(buf); 2209 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2210 return (EINVAL); 2211 if (error != 0) { 2212 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2213 cp->provider->name); 2214 return (error); 2215 } 2216 2217 return (0); 2218} 2219 2220static int 2221g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2222 struct g_mirror_metadata *md) 2223{ 2224 2225 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2226 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2227 pp->name, md->md_did); 2228 return (EEXIST); 2229 } 2230 if (md->md_all != sc->sc_ndisks) { 2231 G_MIRROR_DEBUG(1, 2232 "Invalid '%s' field on disk %s (device %s), skipping.", 2233 "md_all", pp->name, sc->sc_name); 2234 return (EINVAL); 2235 } 2236 if (md->md_slice != sc->sc_slice) { 2237 G_MIRROR_DEBUG(1, 2238 "Invalid '%s' field on disk %s (device %s), skipping.", 2239 "md_slice", pp->name, sc->sc_name); 2240 return (EINVAL); 2241 } 2242 if (md->md_balance != sc->sc_balance) { 2243 G_MIRROR_DEBUG(1, 2244 "Invalid '%s' field on disk %s (device %s), skipping.", 2245 "md_balance", pp->name, sc->sc_name); 2246 return (EINVAL); 2247 } 2248 if (md->md_mediasize != sc->sc_mediasize) { 2249 G_MIRROR_DEBUG(1, 2250 "Invalid '%s' field on disk %s (device %s), skipping.", 2251 "md_mediasize", pp->name, sc->sc_name); 2252 return (EINVAL); 2253 } 2254 if (sc->sc_mediasize > pp->mediasize) { 2255 G_MIRROR_DEBUG(1, 2256 "Invalid size of disk %s (device %s), skipping.", pp->name, 2257 sc->sc_name); 2258 return (EINVAL); 2259 } 2260 if (md->md_sectorsize != sc->sc_sectorsize) { 2261 G_MIRROR_DEBUG(1, 2262 "Invalid '%s' field on disk %s (device %s), skipping.", 2263 "md_sectorsize", pp->name, sc->sc_name); 2264 return (EINVAL); 2265 } 2266 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2267 G_MIRROR_DEBUG(1, 2268 "Invalid sector size of disk %s (device %s), skipping.", 2269 pp->name, sc->sc_name); 2270 return (EINVAL); 2271 } 2272 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2273 G_MIRROR_DEBUG(1, 2274 "Invalid device flags on disk %s (device %s), skipping.", 2275 pp->name, sc->sc_name); 2276 return (EINVAL); 2277 } 2278 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2279 G_MIRROR_DEBUG(1, 2280 "Invalid disk flags on disk %s (device %s), skipping.", 2281 pp->name, sc->sc_name); 2282 return (EINVAL); 2283 } 2284 return (0); 2285} 2286 2287static int 2288g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2289 struct g_mirror_metadata *md) 2290{ 2291 struct g_mirror_disk *disk; 2292 int error; 2293 2294 g_topology_assert(); 2295 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2296 2297 error = g_mirror_check_metadata(sc, pp, md); 2298 if (error != 0) 2299 return (error); 2300 disk = g_mirror_init_disk(sc, pp, md, &error); 2301 if (disk == NULL) 2302 return (error); 2303 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2304 G_MIRROR_EVENT_WAIT); 2305 return (error); 2306} 2307 2308static int 2309g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2310{ 2311 struct g_mirror_softc *sc; 2312 struct g_mirror_disk *disk; 2313 int dcr, dcw, dce, err, error; 2314 2315 g_topology_assert(); 2316 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2317 acw, ace); 2318 2319 dcr = pp->acr + acr; 2320 dcw = pp->acw + acw; 2321 dce = pp->ace + ace; 2322 2323 /* On first open, grab an extra "exclusive" bit */ 2324 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0) 2325 ace++; 2326 /* ... and let go of it on last close */ 2327 if (dcr == 0 && dcw == 0 && dce == 0) 2328 ace--; 2329 2330 sc = pp->geom->softc; 2331 if (sc == NULL || LIST_EMPTY(&sc->sc_disks)) { 2332 if (acr <= 0 && acw <= 0 && ace <= 0) 2333 return (0); 2334 else 2335 return (ENXIO); 2336 } 2337 error = ENXIO; 2338 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2339 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2340 continue; 2341 err = g_access(disk->d_consumer, acr, acw, ace); 2342 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 2343 g_mirror_get_diskname(disk), acr, acw, ace, err); 2344 if (err == 0) { 2345 /* 2346 * Mark disk as dirty on open and unmark on close. 2347 */ 2348 if (pp->acw == 0 && dcw > 0) { 2349 G_MIRROR_DEBUG(1, 2350 "Disk %s (device %s) marked as dirty.", 2351 g_mirror_get_diskname(disk), sc->sc_name); 2352 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2353 g_mirror_update_metadata(disk); 2354 } else if (pp->acw > 0 && dcw == 0) { 2355 G_MIRROR_DEBUG(1, 2356 "Disk %s (device %s) marked as clean.", 2357 g_mirror_get_diskname(disk), sc->sc_name); 2358 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2359 g_mirror_update_metadata(disk); 2360 } 2361 error = 0; 2362 } else { 2363 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 2364 g_mirror_event_send(disk, 2365 G_MIRROR_DISK_STATE_DISCONNECTED, 2366 G_MIRROR_EVENT_DONTWAIT); 2367 } 2368 } 2369 return (error); 2370} 2371 2372static struct g_geom * 2373g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2374{ 2375 struct g_mirror_softc *sc; 2376 struct g_geom *gp; 2377 int error, timeout; 2378 2379 g_topology_assert(); 2380 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2381 md->md_mid); 2382 2383 /* One disk is minimum. */ 2384 if (md->md_all < 1) 2385 return (NULL); 2386 /* 2387 * Action geom. 2388 */ 2389 gp = g_new_geomf(mp, "%s", md->md_name); 2390 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2391 gp->start = g_mirror_start; 2392 gp->spoiled = g_mirror_spoiled; 2393 gp->orphan = g_mirror_orphan; 2394 gp->access = g_mirror_access; 2395 gp->dumpconf = g_mirror_dumpconf; 2396 2397 sc->sc_id = md->md_mid; 2398 sc->sc_slice = md->md_slice; 2399 sc->sc_balance = md->md_balance; 2400 sc->sc_mediasize = md->md_mediasize; 2401 sc->sc_sectorsize = md->md_sectorsize; 2402 sc->sc_ndisks = md->md_all; 2403 sc->sc_flags = md->md_mflags; 2404 sc->sc_bump_syncid = 0; 2405 bioq_init(&sc->sc_queue); 2406 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2407 LIST_INIT(&sc->sc_disks); 2408 TAILQ_INIT(&sc->sc_events); 2409 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2410 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2411 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2412 gp->softc = sc; 2413 sc->sc_geom = gp; 2414 sc->sc_provider = NULL; 2415 /* 2416 * Synchronization geom. 2417 */ 2418 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2419 gp->softc = sc; 2420 gp->orphan = g_mirror_orphan; 2421 sc->sc_sync.ds_geom = gp; 2422 sc->sc_sync.ds_ndisks = 0; 2423 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2424 "g_mirror %s", md->md_name); 2425 if (error != 0) { 2426 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2427 sc->sc_name); 2428 g_destroy_geom(sc->sc_sync.ds_geom); 2429 mtx_destroy(&sc->sc_events_mtx); 2430 mtx_destroy(&sc->sc_queue_mtx); 2431 g_destroy_geom(sc->sc_geom); 2432 free(sc, M_MIRROR); 2433 return (NULL); 2434 } 2435 2436 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2437 2438 /* 2439 * Run timeout. 2440 */ 2441 timeout = atomic_load_acq_int(&g_mirror_timeout); 2442 callout_reset(&sc->sc_callout, timeout * hz, g_mirror_go, sc); 2443 return (sc->sc_geom); 2444} 2445 2446int 2447g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2448{ 2449 struct g_provider *pp; 2450 2451 g_topology_assert(); 2452 2453 if (sc == NULL) 2454 return (ENXIO); 2455 pp = sc->sc_provider; 2456 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2457 if (force) { 2458 G_MIRROR_DEBUG(0, "Device %s is still open, so it " 2459 "can't be definitely removed.", pp->name); 2460 } else { 2461 G_MIRROR_DEBUG(1, 2462 "Device %s is still open (r%dw%de%d).", pp->name, 2463 pp->acr, pp->acw, pp->ace); 2464 return (EBUSY); 2465 } 2466 } 2467 2468 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2469 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2470 g_topology_unlock(); 2471 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2472 mtx_lock(&sc->sc_queue_mtx); 2473 wakeup(sc); 2474 mtx_unlock(&sc->sc_queue_mtx); 2475 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2476 while (sc->sc_worker != NULL) 2477 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2478 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2479 g_topology_lock(); 2480 g_mirror_destroy_device(sc); 2481 free(sc, M_MIRROR); 2482 return (0); 2483} 2484 2485static void 2486g_mirror_taste_orphan(struct g_consumer *cp) 2487{ 2488 2489 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2490 cp->provider->name)); 2491} 2492 2493static struct g_geom * 2494g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2495{ 2496 struct g_mirror_metadata md; 2497 struct g_mirror_softc *sc; 2498 struct g_consumer *cp; 2499 struct g_geom *gp; 2500 int error; 2501 2502 g_topology_assert(); 2503 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2504 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2505 2506 gp = g_new_geomf(mp, "mirror:taste"); 2507 /* 2508 * This orphan function should be never called. 2509 */ 2510 gp->orphan = g_mirror_taste_orphan; 2511 cp = g_new_consumer(gp); 2512 g_attach(cp, pp); 2513 error = g_mirror_read_metadata(cp, &md); 2514 g_detach(cp); 2515 g_destroy_consumer(cp); 2516 g_destroy_geom(gp); 2517 if (error != 0) 2518 return (NULL); 2519 gp = NULL; 2520 2521 if (md.md_version > G_MIRROR_VERSION) { 2522 printf("geom_mirror.ko module is too old to handle %s.\n", 2523 pp->name); 2524 return (NULL); 2525 } 2526 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2527 return (NULL); 2528 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2529 G_MIRROR_DEBUG(0, 2530 "Device %s: provider %s marked as inactive, skipping.", 2531 md.md_name, pp->name); 2532 return (NULL); 2533 } 2534 if (g_mirror_debug >= 2) 2535 mirror_metadata_dump(&md); 2536 2537 /* 2538 * Let's check if device already exists. 2539 */ 2540 sc = NULL; 2541 LIST_FOREACH(gp, &mp->geom, geom) { 2542 sc = gp->softc; 2543 if (sc == NULL) 2544 continue; 2545 if (sc->sc_sync.ds_geom == gp) 2546 continue; 2547 if (strcmp(md.md_name, sc->sc_name) != 0) 2548 continue; 2549 if (md.md_mid != sc->sc_id) { 2550 G_MIRROR_DEBUG(0, "Device %s already configured.", 2551 sc->sc_name); 2552 return (NULL); 2553 } 2554 break; 2555 } 2556 if (gp == NULL) { 2557 gp = g_mirror_create(mp, &md); 2558 if (gp == NULL) { 2559 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2560 md.md_name); 2561 return (NULL); 2562 } 2563 sc = gp->softc; 2564 } 2565 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2566 error = g_mirror_add_disk(sc, pp, &md); 2567 if (error != 0) { 2568 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2569 pp->name, gp->name, error); 2570 if (LIST_EMPTY(&sc->sc_disks)) 2571 g_mirror_destroy(sc, 1); 2572 return (NULL); 2573 } 2574 return (gp); 2575} 2576 2577static int 2578g_mirror_destroy_geom(struct gctl_req *req __unused, 2579 struct g_class *mp __unused, struct g_geom *gp) 2580{ 2581 2582 return (g_mirror_destroy(gp->softc, 0)); 2583} 2584 2585static void 2586g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2587 struct g_consumer *cp, struct g_provider *pp) 2588{ 2589 struct g_mirror_softc *sc; 2590 2591 g_topology_assert(); 2592 2593 sc = gp->softc; 2594 if (sc == NULL) 2595 return; 2596 /* Skip synchronization geom. */ 2597 if (gp == sc->sc_sync.ds_geom) 2598 return; 2599 if (pp != NULL) { 2600 /* Nothing here. */ 2601 } else if (cp != NULL) { 2602 struct g_mirror_disk *disk; 2603 2604 disk = cp->private; 2605 if (disk == NULL) 2606 return; 2607 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2608 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2609 sbuf_printf(sb, "%s<Synchronized>", indent); 2610 if (disk->d_sync.ds_offset_done == 0) 2611 sbuf_printf(sb, "0%%"); 2612 else { 2613 sbuf_printf(sb, "%u%%", 2614 (u_int)((disk->d_sync.ds_offset_done * 100) / 2615 sc->sc_provider->mediasize)); 2616 } 2617 sbuf_printf(sb, "</Synchronized>\n"); 2618 } 2619 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2620 disk->d_sync.ds_syncid); 2621 sbuf_printf(sb, "%s<Flags>", indent); 2622 if (disk->d_flags == 0) 2623 sbuf_printf(sb, "NONE"); 2624 else { 2625 int first = 1; 2626 2627#define ADD_FLAG(flag, name) do { \ 2628 if ((disk->d_flags & (flag)) != 0) { \ 2629 if (!first) \ 2630 sbuf_printf(sb, ", "); \ 2631 else \ 2632 first = 0; \ 2633 sbuf_printf(sb, name); \ 2634 } \ 2635} while (0) 2636 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2637 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2638 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2639 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2640 "SYNCHRONIZING"); 2641 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2642#undef ADD_FLAG 2643 } 2644 sbuf_printf(sb, "</Flags>\n"); 2645 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2646 disk->d_priority); 2647 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2648 g_mirror_disk_state2str(disk->d_state)); 2649 } else { 2650 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2651 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2652 sbuf_printf(sb, "%s<Flags>", indent); 2653 if (sc->sc_flags == 0) 2654 sbuf_printf(sb, "NONE"); 2655 else { 2656 int first = 1; 2657 2658#define ADD_FLAG(flag, name) do { \ 2659 if ((sc->sc_flags & (flag)) != 0) { \ 2660 if (!first) \ 2661 sbuf_printf(sb, ", "); \ 2662 else \ 2663 first = 0; \ 2664 sbuf_printf(sb, name); \ 2665 } \ 2666} while (0) 2667 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2668#undef ADD_FLAG 2669 } 2670 sbuf_printf(sb, "</Flags>\n"); 2671 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2672 (u_int)sc->sc_slice); 2673 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2674 balance_name(sc->sc_balance)); 2675 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2676 sc->sc_ndisks); 2677 sbuf_printf(sb, "%s<State>", indent); 2678 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2679 sbuf_printf(sb, "%s", "STARTING"); 2680 else if (sc->sc_ndisks == 2681 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 2682 sbuf_printf(sb, "%s", "COMPLETE"); 2683 else 2684 sbuf_printf(sb, "%s", "DEGRADED"); 2685 sbuf_printf(sb, "</State>\n"); 2686 } 2687} 2688 2689static int 2690g_mirror_can_go(void) 2691{ 2692 struct g_mirror_softc *sc; 2693 struct g_geom *gp; 2694 struct g_provider *pp; 2695 int can_go; 2696 2697 DROP_GIANT(); 2698 can_go = 1; 2699 g_topology_lock(); 2700 LIST_FOREACH(gp, &g_mirror_class.geom, geom) { 2701 sc = gp->softc; 2702 if (sc == NULL) 2703 continue; 2704 pp = sc->sc_provider; 2705 if (pp == NULL || pp->error != 0) { 2706 can_go = 0; 2707 break; 2708 } 2709 } 2710 g_topology_unlock(); 2711 PICKUP_GIANT(); 2712 return (can_go); 2713} 2714 2715static void 2716g_mirror_rootwait(void) 2717{ 2718 2719 /* 2720 * Wait for mirrors in degraded state. 2721 */ 2722 for (;;) { 2723 if (g_mirror_can_go()) 2724 break; 2725 tsleep(&g_mirror_class, PRIBIO, "mroot", hz); 2726 } 2727} 2728 2729SYSINIT(g_mirror_root, SI_SUB_RAID, SI_ORDER_FIRST, g_mirror_rootwait, NULL) 2730 2731DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2732