g_mirror.c revision 324402
143166Snsouch/*- 243166Snsouch * Copyright (c) 2004-2006 Pawel Jakub Dawidek <pjd@FreeBSD.org> 343166Snsouch * All rights reserved. 443166Snsouch * 543166Snsouch * Redistribution and use in source and binary forms, with or without 643166Snsouch * modification, are permitted provided that the following conditions 743166Snsouch * are met: 843166Snsouch * 1. Redistributions of source code must retain the above copyright 943166Snsouch * notice, this list of conditions and the following disclaimer. 1043166Snsouch * 2. Redistributions in binary form must reproduce the above copyright 1143166Snsouch * notice, this list of conditions and the following disclaimer in the 1243166Snsouch * documentation and/or other materials provided with the distribution. 1343166Snsouch * 1443166Snsouch * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 1543166Snsouch * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 1643166Snsouch * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 1743166Snsouch * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 1843166Snsouch * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 1943166Snsouch * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2043166Snsouch * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2143166Snsouch * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2243166Snsouch * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 2343166Snsouch * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 2443166Snsouch * SUCH DAMAGE. 2543166Snsouch */ 2643166Snsouch 27116192Sobrien#include <sys/cdefs.h> 28116192Sobrien__FBSDID("$FreeBSD: stable/11/sys/geom/mirror/g_mirror.c 324402 2017-10-07 23:06:49Z ngie $"); 29116192Sobrien 3043166Snsouch#include <sys/param.h> 3143166Snsouch#include <sys/systm.h> 32165951Sjhb#include <sys/fail.h> 3343166Snsouch#include <sys/kernel.h> 34165951Sjhb#include <sys/module.h> 3543166Snsouch#include <sys/limits.h> 36165951Sjhb#include <sys/lock.h> 3746651Speter#include <sys/mutex.h> 38165951Sjhb#include <sys/bio.h> 3943166Snsouch#include <sys/sbuf.h> 4043166Snsouch#include <sys/sysctl.h> 4143166Snsouch#include <sys/malloc.h> 4243166Snsouch#include <sys/eventhandler.h> 43119288Simp#include <vm/uma.h> 44119288Simp#include <geom/geom.h> 4543166Snsouch#include <sys/proc.h> 4643166Snsouch#include <sys/kthread.h> 4743166Snsouch#include <sys/sched.h> 4843166Snsouch#include <geom/mirror/g_mirror.h> 49165951Sjhb 50165951SjhbFEATURE(geom_mirror, "GEOM mirroring support"); 51165951Sjhb 52165951Sjhbstatic MALLOC_DEFINE(M_MIRROR, "mirror_data", "GEOM_MIRROR Data"); 53165951Sjhb 54165951SjhbSYSCTL_DECL(_kern_geom); 55165951Sjhbstatic SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, 56165951Sjhb "GEOM_MIRROR stuff"); 5743166Snsouchu_int g_mirror_debug = 0; 58162289SjhbSYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RWTUN, &g_mirror_debug, 0, 59165951Sjhb "Debug level"); 60165951Sjhbstatic u_int g_mirror_timeout = 4; 61165951SjhbSYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RWTUN, &g_mirror_timeout, 62165951Sjhb 0, "Time to wait on all mirror components"); 6343166Snsouchstatic u_int g_mirror_idletime = 5; 6443166SnsouchSYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RWTUN, 65165951Sjhb &g_mirror_idletime, 0, "Mark components as clean when idling"); 66165951Sjhbstatic u_int g_mirror_disconnect_on_failure = 1; 67165951SjhbSYSCTL_UINT(_kern_geom_mirror, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN, 68165951Sjhb &g_mirror_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 69162234Sjhbstatic u_int g_mirror_syncreqs = 2; 7043166SnsouchSYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_requests, CTLFLAG_RDTUN, 7143166Snsouch &g_mirror_syncreqs, 0, "Parallel synchronization I/O requests."); 7243166Snsouch 7343166Snsouch#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 7443166Snsouch G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 7543166Snsouch msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 7643166Snsouch G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 7743166Snsouch} while (0) 7843166Snsouch 79162234Sjhbstatic eventhandler_tag g_mirror_post_sync = NULL; 80165951Sjhbstatic int g_mirror_shutdown = 0; 81165951Sjhb 82165951Sjhbstatic g_ctl_destroy_geom_t g_mirror_destroy_geom; 83165951Sjhbstatic g_taste_t g_mirror_taste; 84165951Sjhbstatic g_init_t g_mirror_init; 85162289Sjhbstatic g_fini_t g_mirror_fini; 86165951Sjhbstatic g_provgone_t g_mirror_providergone; 87165951Sjhbstatic g_resize_t g_mirror_resize; 88165951Sjhb 8943166Snsouchstruct g_class g_mirror_class = { 90165951Sjhb .name = G_MIRROR_CLASS_NAME, 91165951Sjhb .version = G_VERSION, 92165951Sjhb .ctlreq = g_mirror_config, 93165951Sjhb .taste = g_mirror_taste, 94165951Sjhb .destroy_geom = g_mirror_destroy_geom, 95165951Sjhb .init = g_mirror_init, 96165951Sjhb .fini = g_mirror_fini, 97165951Sjhb .providergone = g_mirror_providergone, 98165951Sjhb .resize = g_mirror_resize 99165951Sjhb}; 100165951Sjhb 101165951Sjhb 10243166Snsouchstatic void g_mirror_destroy_provider(struct g_mirror_softc *sc); 103165951Sjhbstatic int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 104165951Sjhbstatic void g_mirror_update_device(struct g_mirror_softc *sc, bool force); 105162289Sjhbstatic void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 106165951Sjhb struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 107165951Sjhbstatic void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 108165951Sjhbstatic void g_mirror_register_request(struct bio *bp); 109165951Sjhbstatic void g_mirror_sync_release(struct g_mirror_softc *sc); 110165951Sjhb 111165951Sjhb 112162289Sjhbstatic const char * 113178972Sjhbg_mirror_disk_state2str(int state) 114178972Sjhb{ 115165951Sjhb 11643166Snsouch switch (state) { 117165951Sjhb case G_MIRROR_DISK_STATE_NONE: 118165951Sjhb return ("NONE"); 119165951Sjhb case G_MIRROR_DISK_STATE_NEW: 120165951Sjhb return ("NEW"); 121165951Sjhb case G_MIRROR_DISK_STATE_ACTIVE: 122165951Sjhb return ("ACTIVE"); 123165951Sjhb case G_MIRROR_DISK_STATE_STALE: 124165951Sjhb return ("STALE"); 12543166Snsouch case G_MIRROR_DISK_STATE_SYNCHRONIZING: 126165951Sjhb return ("SYNCHRONIZING"); 127165951Sjhb case G_MIRROR_DISK_STATE_DISCONNECTED: 128165951Sjhb return ("DISCONNECTED"); 129165951Sjhb case G_MIRROR_DISK_STATE_DESTROY: 130165951Sjhb return ("DESTROY"); 131165951Sjhb default: 132165951Sjhb return ("INVALID"); 133165951Sjhb } 134165951Sjhb} 135165951Sjhb 136165951Sjhbstatic const char * 137165951Sjhbg_mirror_device_state2str(int state) 138165951Sjhb{ 139165951Sjhb 140165951Sjhb switch (state) { 141165951Sjhb case G_MIRROR_DEVICE_STATE_STARTING: 142165951Sjhb return ("STARTING"); 143165951Sjhb case G_MIRROR_DEVICE_STATE_RUNNING: 144168870Sjhb return ("RUNNING"); 14543166Snsouch default: 146165951Sjhb return ("INVALID"); 147165951Sjhb } 148165951Sjhb} 149165951Sjhb 150165951Sjhbstatic const char * 15146651Speterg_mirror_get_diskname(struct g_mirror_disk *disk) 152165951Sjhb{ 153165951Sjhb 154165951Sjhb if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 155165951Sjhb return ("[unknown]"); 156165951Sjhb return (disk->d_name); 157165951Sjhb} 158165951Sjhb 159165951Sjhb/* 160165951Sjhb * --- Events handling functions --- 161165951Sjhb * Events in geom_mirror are used to maintain disks and device status 16243166Snsouch * from one thread to simplify locking. 163179622Sjhb */ 164179622Sjhbstatic void 165165951Sjhbg_mirror_event_free(struct g_mirror_event *ep) 166165951Sjhb{ 167165951Sjhb 168165951Sjhb free(ep, M_MIRROR); 169162289Sjhb} 170165951Sjhb 171165951Sjhbint 172165951Sjhbg_mirror_event_send(void *arg, int state, int flags) 173165951Sjhb{ 174165951Sjhb struct g_mirror_softc *sc; 175165951Sjhb struct g_mirror_disk *disk; 176165951Sjhb struct g_mirror_event *ep; 177165951Sjhb int error; 178165951Sjhb 179162289Sjhb ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 180165951Sjhb G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 181165951Sjhb if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 182165951Sjhb disk = NULL; 183165951Sjhb sc = arg; 184165951Sjhb } else { 18543166Snsouch disk = arg; 186165951Sjhb sc = disk->d_softc; 187165951Sjhb } 188165951Sjhb ep->e_disk = disk; 189165951Sjhb ep->e_state = state; 190165951Sjhb ep->e_flags = flags; 191162289Sjhb ep->e_error = 0; 192165951Sjhb mtx_lock(&sc->sc_events_mtx); 19343166Snsouch TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 194162289Sjhb mtx_unlock(&sc->sc_events_mtx); 195165951Sjhb G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 196162289Sjhb mtx_lock(&sc->sc_queue_mtx); 197165951Sjhb wakeup(sc); 198165951Sjhb mtx_unlock(&sc->sc_queue_mtx); 199165951Sjhb if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 200162289Sjhb return (0); 201165951Sjhb sx_assert(&sc->sc_lock, SX_XLOCKED); 202165951Sjhb G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 203165951Sjhb sx_xunlock(&sc->sc_lock); 204165951Sjhb while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 205165951Sjhb mtx_lock(&sc->sc_events_mtx); 206165951Sjhb MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 207165951Sjhb hz * 5); 208165951Sjhb } 209165951Sjhb error = ep->e_error; 210165951Sjhb g_mirror_event_free(ep); 211165951Sjhb sx_xlock(&sc->sc_lock); 21243166Snsouch return (error); 213165951Sjhb} 214165951Sjhb 215165951Sjhbstatic struct g_mirror_event * 21643166Snsouchg_mirror_event_get(struct g_mirror_softc *sc) 217165951Sjhb{ 218162289Sjhb struct g_mirror_event *ep; 219165951Sjhb 220165951Sjhb mtx_lock(&sc->sc_events_mtx); 221165951Sjhb ep = TAILQ_FIRST(&sc->sc_events); 222165951Sjhb mtx_unlock(&sc->sc_events_mtx); 22343166Snsouch return (ep); 22443166Snsouch} 225162289Sjhb 226162234Sjhbstatic void 22743166Snsouchg_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep) 22843166Snsouch{ 229162289Sjhb 23043166Snsouch mtx_lock(&sc->sc_events_mtx); 23143166Snsouch TAILQ_REMOVE(&sc->sc_events, ep, e_next); 23243166Snsouch mtx_unlock(&sc->sc_events_mtx); 23343166Snsouch} 23443166Snsouch 23543166Snsouchstatic void 23643166Snsouchg_mirror_event_cancel(struct g_mirror_disk *disk) 23743166Snsouch{ 238162289Sjhb struct g_mirror_softc *sc; 23943166Snsouch struct g_mirror_event *ep, *tmpep; 24043166Snsouch 241162289Sjhb sc = disk->d_softc; 242162289Sjhb sx_assert(&sc->sc_lock, SX_XLOCKED); 243162289Sjhb 244165951Sjhb mtx_lock(&sc->sc_events_mtx); 245162289Sjhb TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 246162289Sjhb if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 247165951Sjhb continue; 248165951Sjhb if (ep->e_disk != disk) 24943166Snsouch continue; 250165951Sjhb TAILQ_REMOVE(&sc->sc_events, ep, e_next); 25143166Snsouch if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 252162289Sjhb g_mirror_event_free(ep); 253165951Sjhb else { 254165951Sjhb ep->e_error = ECANCELED; 255162289Sjhb wakeup(ep); 256162289Sjhb } 25743166Snsouch } 258165951Sjhb mtx_unlock(&sc->sc_events_mtx); 25943166Snsouch} 260162289Sjhb 261165951Sjhb/* 262165951Sjhb * Return the number of disks in given state. 263165951Sjhb * If state is equal to -1, count all connected disks. 264162289Sjhb */ 26543166Snsouchu_int 26643166Snsouchg_mirror_ndisks(struct g_mirror_softc *sc, int state) 26743166Snsouch{ 268165951Sjhb struct g_mirror_disk *disk; 26943166Snsouch u_int n = 0; 270165951Sjhb 271162289Sjhb sx_assert(&sc->sc_lock, SX_LOCKED); 272165951Sjhb 273162289Sjhb LIST_FOREACH(disk, &sc->sc_disks, d_next) { 274162289Sjhb if (state == -1 || disk->d_state == state) 275162289Sjhb n++; 276162289Sjhb } 277162289Sjhb return (n); 278162289Sjhb} 279165951Sjhb 280165951Sjhb/* 281162289Sjhb * Find a disk in mirror by its disk ID. 282162289Sjhb */ 283162289Sjhbstatic struct g_mirror_disk * 284162289Sjhbg_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 28549064Snsouch{ 286162289Sjhb struct g_mirror_disk *disk; 28743166Snsouch 288162289Sjhb sx_assert(&sc->sc_lock, SX_XLOCKED); 28943166Snsouch 290162289Sjhb LIST_FOREACH(disk, &sc->sc_disks, d_next) { 29143166Snsouch if (disk->d_id == id) 292165951Sjhb return (disk); 29343166Snsouch } 294165951Sjhb return (NULL); 295162289Sjhb} 296165951Sjhb 297162289Sjhbstatic u_int 298165951Sjhbg_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 299165951Sjhb{ 300165951Sjhb struct bio *bp; 301165951Sjhb u_int nreqs = 0; 302162289Sjhb 30343166Snsouch mtx_lock(&sc->sc_queue_mtx); 304162289Sjhb TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 305162289Sjhb if (bp->bio_from == cp) 306165951Sjhb nreqs++; 307162289Sjhb } 308162289Sjhb mtx_unlock(&sc->sc_queue_mtx); 309165951Sjhb return (nreqs); 31043166Snsouch} 31143166Snsouch 312162289Sjhbstatic int 313165951Sjhbg_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 31443166Snsouch{ 31543166Snsouch 31643288Sdillon if (cp->index > 0) { 31743288Sdillon G_MIRROR_DEBUG(2, 318165951Sjhb "I/O requests for %s exist, can't destroy it now.", 31943288Sdillon cp->provider->name); 32043288Sdillon return (1); 321162289Sjhb } 322165951Sjhb if (g_mirror_nrequests(sc, cp) > 0) { 32343166Snsouch G_MIRROR_DEBUG(2, 324165951Sjhb "I/O requests for %s in queue, can't destroy it now.", 325162289Sjhb cp->provider->name); 32643166Snsouch return (1); 32743166Snsouch } 328162289Sjhb return (0); 329162289Sjhb} 33043166Snsouch 331165951Sjhbstatic void 332165951Sjhbg_mirror_destroy_consumer(void *arg, int flags __unused) 333165951Sjhb{ 334162289Sjhb struct g_consumer *cp; 335165951Sjhb 336165951Sjhb g_topology_assert(); 337165951Sjhb 338165951Sjhb cp = arg; 339165951Sjhb G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 340165951Sjhb g_detach(cp); 341165951Sjhb g_destroy_consumer(cp); 342165951Sjhb} 343165951Sjhb 34443166Snsouchstatic void 345162289Sjhbg_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 346165951Sjhb{ 347162289Sjhb struct g_provider *pp; 34843166Snsouch int retaste_wait; 34943166Snsouch 350162289Sjhb g_topology_assert(); 35143166Snsouch 35243166Snsouch cp->private = NULL; 353165951Sjhb if (g_mirror_is_busy(sc, cp)) 35443166Snsouch return; 35543166Snsouch pp = cp->provider; 356162289Sjhb retaste_wait = 0; 357165951Sjhb if (cp->acw == 1) { 358165951Sjhb if ((pp->geom->flags & G_GEOM_WITHER) == 0) 359162289Sjhb retaste_wait = 1; 36043166Snsouch } 361162289Sjhb G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr, 362162289Sjhb -cp->acw, -cp->ace, 0); 363162289Sjhb if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 364189882Savg g_access(cp, -cp->acr, -cp->acw, -cp->ace); 365162289Sjhb if (retaste_wait) { 366165951Sjhb /* 36743166Snsouch * After retaste event was send (inside g_access()), we can send 36843166Snsouch * event to detach and destroy consumer. 369165951Sjhb * A class, which has consumer to the given provider connected 370189882Savg * will not receive retaste event for the provider. 371165951Sjhb * This is the way how I ignore retaste events when I close 372165951Sjhb * consumers opened for write: I detach and destroy consumer 373165951Sjhb * after retaste event is sent. 374165951Sjhb */ 375165951Sjhb g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL); 376165951Sjhb return; 377165951Sjhb } 378165951Sjhb G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name); 379165951Sjhb g_detach(cp); 380189882Savg g_destroy_consumer(cp); 381189882Savg} 382189882Savg 383189882Savgstatic int 384165951Sjhbg_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 385165951Sjhb{ 386165951Sjhb struct g_consumer *cp; 387162289Sjhb int error; 388162289Sjhb 389162289Sjhb g_topology_assert_not(); 390162289Sjhb KASSERT(disk->d_consumer == NULL, 391162289Sjhb ("Disk already connected (device %s).", disk->d_softc->sc_name)); 392162289Sjhb 393162289Sjhb g_topology_lock(); 39443166Snsouch cp = g_new_consumer(disk->d_softc->sc_geom); 395162289Sjhb cp->flags |= G_CF_DIRECT_RECEIVE; 396165951Sjhb error = g_attach(cp, pp); 397162289Sjhb if (error != 0) { 398165951Sjhb g_destroy_consumer(cp); 39943166Snsouch g_topology_unlock(); 400165951Sjhb return (error); 401165951Sjhb } 402165951Sjhb error = g_access(cp, 1, 1, 1); 403162289Sjhb if (error != 0) { 404165951Sjhb g_detach(cp); 405162289Sjhb g_destroy_consumer(cp); 406162289Sjhb g_topology_unlock(); 407162289Sjhb G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).", 408165951Sjhb pp->name, error); 409162289Sjhb return (error); 410165951Sjhb } 411162289Sjhb g_topology_unlock(); 412162289Sjhb disk->d_consumer = cp; 413189882Savg disk->d_consumer->private = disk; 414162289Sjhb disk->d_consumer->index = 0; 415178972Sjhb 416162289Sjhb G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 41743166Snsouch return (0); 41843166Snsouch} 419162289Sjhb 420165951Sjhbstatic void 421162289Sjhbg_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 422165951Sjhb{ 423165951Sjhb 424165951Sjhb g_topology_assert(); 42543166Snsouch 426162289Sjhb if (cp == NULL) 42743166Snsouch return; 428162289Sjhb if (cp->provider != NULL) 42943166Snsouch g_mirror_kill_consumer(sc, cp); 430162289Sjhb else 431165951Sjhb g_destroy_consumer(cp); 432162289Sjhb} 433165951Sjhb 434162289Sjhb/* 435165951Sjhb * Initialize disk. This means allocate memory, create consumer, attach it 436165951Sjhb * to the provider and open access (r1w1e1) to it. 437165951Sjhb */ 438162289Sjhbstatic struct g_mirror_disk * 439165951Sjhbg_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 440162289Sjhb struct g_mirror_metadata *md, int *errorp) 441172667Sjhb{ 442165951Sjhb struct g_mirror_disk *disk; 443165951Sjhb int i, error; 444162289Sjhb 445189882Savg disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 446162289Sjhb if (disk == NULL) { 447165951Sjhb error = ENOMEM; 44843166Snsouch goto fail; 449165951Sjhb } 450162289Sjhb disk->d_softc = sc; 45143166Snsouch error = g_mirror_connect_disk(disk, pp); 452162289Sjhb if (error != 0) 45343166Snsouch goto fail; 45443166Snsouch disk->d_id = md->md_did; 455162289Sjhb disk->d_state = G_MIRROR_DISK_STATE_NONE; 456162289Sjhb disk->d_priority = md->md_priority; 457162289Sjhb disk->d_flags = md->md_dflags; 458162289Sjhb error = g_getattr("GEOM::candelete", disk->d_consumer, &i); 459162289Sjhb if (error == 0 && i != 0) 460165951Sjhb disk->d_flags |= G_MIRROR_DISK_FLAG_CANDELETE; 461165951Sjhb if (md->md_provider[0] != '\0') 462165951Sjhb disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 463165951Sjhb disk->d_sync.ds_consumer = NULL; 464165951Sjhb disk->d_sync.ds_offset = md->md_sync_offset; 46543166Snsouch disk->d_sync.ds_offset_done = md->md_sync_offset; 46643166Snsouch disk->d_genid = md->md_genid; 46743166Snsouch disk->d_sync.ds_syncid = md->md_syncid; 46843166Snsouch if (errorp != NULL) 46943166Snsouch *errorp = 0; 470162289Sjhb return (disk); 471165951Sjhbfail: 472162289Sjhb if (errorp != NULL) 473162289Sjhb *errorp = error; 474162289Sjhb if (disk != NULL) 475162289Sjhb free(disk, M_MIRROR); 476162289Sjhb return (NULL); 477162289Sjhb} 478162289Sjhb 479162289Sjhbstatic void 48043166Snsouchg_mirror_destroy_disk(struct g_mirror_disk *disk) 481162289Sjhb{ 482162289Sjhb struct g_mirror_softc *sc; 483162289Sjhb 484162289Sjhb g_topology_assert_not(); 485165951Sjhb sc = disk->d_softc; 486162289Sjhb sx_assert(&sc->sc_lock, SX_XLOCKED); 487165951Sjhb 488165951Sjhb LIST_REMOVE(disk, d_next); 489165951Sjhb g_mirror_event_cancel(disk); 490165951Sjhb if (sc->sc_hint == disk) 491165951Sjhb sc->sc_hint = NULL; 492165951Sjhb switch (disk->d_state) { 493162289Sjhb case G_MIRROR_DISK_STATE_SYNCHRONIZING: 494165951Sjhb g_mirror_sync_stop(disk, 1); 495165951Sjhb /* FALLTHROUGH */ 496165951Sjhb case G_MIRROR_DISK_STATE_NEW: 497165951Sjhb case G_MIRROR_DISK_STATE_STALE: 498162289Sjhb case G_MIRROR_DISK_STATE_ACTIVE: 49943166Snsouch g_topology_lock(); 50043166Snsouch g_mirror_disconnect_consumer(sc, disk->d_consumer); 50143166Snsouch g_topology_unlock(); 50243166Snsouch free(disk, M_MIRROR); 50343166Snsouch break; 504162289Sjhb default: 505162289Sjhb KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 506162289Sjhb g_mirror_get_diskname(disk), 507165951Sjhb g_mirror_disk_state2str(disk->d_state))); 508165951Sjhb } 509165951Sjhb} 510165951Sjhb 511165951Sjhbstatic void 512162289Sjhbg_mirror_free_device(struct g_mirror_softc *sc) 513165951Sjhb{ 514165951Sjhb 515165951Sjhb mtx_destroy(&sc->sc_queue_mtx); 516165951Sjhb mtx_destroy(&sc->sc_events_mtx); 517165951Sjhb mtx_destroy(&sc->sc_done_mtx); 518162289Sjhb sx_destroy(&sc->sc_lock); 51943166Snsouch free(sc, M_MIRROR); 520162289Sjhb} 52143166Snsouch 52243166Snsouchstatic void 52343166Snsouchg_mirror_providergone(struct g_provider *pp) 524162289Sjhb{ 525162289Sjhb struct g_mirror_softc *sc = pp->private; 526162289Sjhb 527165951Sjhb if ((--sc->sc_refcnt) == 0) 528165951Sjhb g_mirror_free_device(sc); 529165951Sjhb} 530165951Sjhb 531165951Sjhbstatic void 532165951Sjhbg_mirror_destroy_device(struct g_mirror_softc *sc) 533165951Sjhb{ 534165951Sjhb struct g_mirror_disk *disk; 535165951Sjhb struct g_mirror_event *ep; 536165951Sjhb struct g_geom *gp; 53743166Snsouch struct g_consumer *cp, *tmpcp; 538165951Sjhb 539165951Sjhb g_topology_assert_not(); 540165951Sjhb sx_assert(&sc->sc_lock, SX_XLOCKED); 541165951Sjhb 542165951Sjhb gp = sc->sc_geom; 54343166Snsouch if (sc->sc_provider != NULL) 544165951Sjhb g_mirror_destroy_provider(sc); 54543166Snsouch for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 546162289Sjhb disk = LIST_FIRST(&sc->sc_disks)) { 547165951Sjhb disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 548162289Sjhb g_mirror_update_metadata(disk); 54943166Snsouch g_mirror_destroy_disk(disk); 550162289Sjhb } 55143166Snsouch while ((ep = g_mirror_event_get(sc)) != NULL) { 55243166Snsouch g_mirror_event_remove(sc, ep); 55343166Snsouch if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 554162289Sjhb g_mirror_event_free(ep); 555162289Sjhb else { 556162289Sjhb ep->e_error = ECANCELED; 557165951Sjhb ep->e_flags |= G_MIRROR_EVENT_DONE; 558165951Sjhb G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 559165951Sjhb mtx_lock(&sc->sc_events_mtx); 560165951Sjhb wakeup(ep); 561165951Sjhb mtx_unlock(&sc->sc_events_mtx); 562162289Sjhb } 563165951Sjhb } 564165951Sjhb callout_drain(&sc->sc_callout); 565165951Sjhb 566165951Sjhb g_topology_lock(); 567165951Sjhb LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 568165951Sjhb g_mirror_disconnect_consumer(sc, cp); 569162289Sjhb } 57043166Snsouch g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 571162289Sjhb G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 57243166Snsouch g_wither_geom(gp, ENXIO); 57343166Snsouch sx_xunlock(&sc->sc_lock); 57443166Snsouch if ((--sc->sc_refcnt) == 0) 575162289Sjhb g_mirror_free_device(sc); 576162289Sjhb g_topology_unlock(); 577162289Sjhb} 578165951Sjhb 579165951Sjhbstatic void 580165951Sjhbg_mirror_orphan(struct g_consumer *cp) 581165951Sjhb{ 582165951Sjhb struct g_mirror_disk *disk; 583162289Sjhb 584165951Sjhb g_topology_assert(); 585165951Sjhb 586165951Sjhb disk = cp->private; 587165951Sjhb if (disk == NULL) 588165951Sjhb return; 589165951Sjhb disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 590165951Sjhb g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 591162289Sjhb G_MIRROR_EVENT_DONTWAIT); 59243166Snsouch} 59343166Snsouch 59443166Snsouch/* 59543166Snsouch * Function should return the next active disk on the list. 59643166Snsouch * It is possible that it will be the same disk as given. 597162289Sjhb * If there are no active disks on list, NULL is returned. 598162289Sjhb */ 599162289Sjhbstatic __inline struct g_mirror_disk * 600165951Sjhbg_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 601165951Sjhb{ 602165951Sjhb struct g_mirror_disk *dp; 603165951Sjhb 604165951Sjhb for (dp = LIST_NEXT(disk, d_next); dp != disk; 605162289Sjhb dp = LIST_NEXT(dp, d_next)) { 606165951Sjhb if (dp == NULL) 607165951Sjhb dp = LIST_FIRST(&sc->sc_disks); 608165951Sjhb if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 609165951Sjhb break; 610165951Sjhb } 611165951Sjhb if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 612165951Sjhb return (NULL); 613162289Sjhb return (dp); 61443166Snsouch} 615165951Sjhb 61643166Snsouchstatic struct g_mirror_disk * 61743166Snsouchg_mirror_get_disk(struct g_mirror_softc *sc) 61843166Snsouch{ 619162289Sjhb struct g_mirror_disk *disk; 620162289Sjhb 621162289Sjhb if (sc->sc_hint == NULL) { 622165951Sjhb sc->sc_hint = LIST_FIRST(&sc->sc_disks); 623165951Sjhb if (sc->sc_hint == NULL) 624165951Sjhb return (NULL); 625165951Sjhb } 626165951Sjhb disk = sc->sc_hint; 627162289Sjhb if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 628165951Sjhb disk = g_mirror_find_next(sc, disk); 629165951Sjhb if (disk == NULL) 630165951Sjhb return (NULL); 631165951Sjhb } 632165951Sjhb sc->sc_hint = g_mirror_find_next(sc, disk); 633165951Sjhb return (disk); 634165951Sjhb} 635165951Sjhb 636165951Sjhbstatic int 637162289Sjhbg_mirror_write_metadata(struct g_mirror_disk *disk, 63843166Snsouch struct g_mirror_metadata *md) 639162289Sjhb{ 64043166Snsouch struct g_mirror_softc *sc; 64143166Snsouch struct g_consumer *cp; 64243166Snsouch off_t offset, length; 64343166Snsouch u_char *sector; 64443166Snsouch int error = 0; 64543166Snsouch 64643166Snsouch g_topology_assert_not(); 64743166Snsouch sc = disk->d_softc; 64843166Snsouch sx_assert(&sc->sc_lock, SX_LOCKED); 649162289Sjhb 650162289Sjhb cp = disk->d_consumer; 651162289Sjhb KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 652165951Sjhb KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 653165951Sjhb KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 654165951Sjhb ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr, 655165951Sjhb cp->acw, cp->ace)); 656165951Sjhb length = cp->provider->sectorsize; 657162289Sjhb offset = cp->provider->mediasize - length; 658165951Sjhb sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 659165951Sjhb if (md != NULL && 660165951Sjhb (sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) { 661165951Sjhb /* 662165951Sjhb * Handle the case, when the size of parent provider reduced. 663165951Sjhb */ 664165951Sjhb if (offset < md->md_mediasize) 665165951Sjhb error = ENOSPC; 666165951Sjhb else 667162289Sjhb mirror_metadata_encode(md, sector); 668165951Sjhb } 669162289Sjhb KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_metadata_write, error); 67043166Snsouch if (error == 0) 671165951Sjhb error = g_write_data(cp, offset, sector, length); 67243166Snsouch free(sector, M_MIRROR); 67343166Snsouch if (error != 0) { 674162289Sjhb if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 67543166Snsouch disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 67643166Snsouch G_MIRROR_DEBUG(0, "Cannot write metadata on %s " 67743166Snsouch "(device=%s, error=%d).", 678162289Sjhb g_mirror_get_diskname(disk), sc->sc_name, error); 679162289Sjhb } else { 680162289Sjhb G_MIRROR_DEBUG(1, "Cannot write metadata on %s " 681162289Sjhb "(device=%s, error=%d).", 682165951Sjhb g_mirror_get_diskname(disk), sc->sc_name, error); 683162289Sjhb } 684165951Sjhb if (g_mirror_disconnect_on_failure && 685165951Sjhb g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) { 686165951Sjhb sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 687165951Sjhb g_mirror_event_send(disk, 688165951Sjhb G_MIRROR_DISK_STATE_DISCONNECTED, 689162289Sjhb G_MIRROR_EVENT_DONTWAIT); 690165951Sjhb } 691165951Sjhb } 692165951Sjhb return (error); 693165951Sjhb} 694165951Sjhb 695165951Sjhbstatic int 696165951Sjhbg_mirror_clear_metadata(struct g_mirror_disk *disk) 697165951Sjhb{ 698165951Sjhb int error; 699165951Sjhb 700165951Sjhb g_topology_assert_not(); 701165951Sjhb sx_assert(&disk->d_softc->sc_lock, SX_LOCKED); 702162289Sjhb 70343166Snsouch if (disk->d_softc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 70443166Snsouch return (0); 70543166Snsouch error = g_mirror_write_metadata(disk, NULL); 706162234Sjhb if (error == 0) { 70743166Snsouch G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 708162289Sjhb g_mirror_get_diskname(disk)); 709162289Sjhb } else { 710162234Sjhb G_MIRROR_DEBUG(0, 711162289Sjhb "Cannot clear metadata on disk %s (error=%d).", 712162289Sjhb g_mirror_get_diskname(disk), error); 713165951Sjhb } 714162289Sjhb return (error); 715165951Sjhb} 716165951Sjhb 717165951Sjhbvoid 718165951Sjhbg_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 719165951Sjhb struct g_mirror_metadata *md) 720165951Sjhb{ 721165951Sjhb 722165951Sjhb strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 723165951Sjhb md->md_version = G_MIRROR_VERSION; 724165951Sjhb strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 725165951Sjhb md->md_mid = sc->sc_id; 726165951Sjhb md->md_all = sc->sc_ndisks; 727165951Sjhb md->md_slice = sc->sc_slice; 728165951Sjhb md->md_balance = sc->sc_balance; 729165951Sjhb md->md_genid = sc->sc_genid; 730165951Sjhb md->md_mediasize = sc->sc_mediasize; 731165951Sjhb md->md_sectorsize = sc->sc_sectorsize; 732165951Sjhb md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 733165951Sjhb bzero(md->md_provider, sizeof(md->md_provider)); 734165951Sjhb if (disk == NULL) { 735165951Sjhb md->md_did = arc4random(); 736165951Sjhb md->md_priority = 0; 73743166Snsouch md->md_syncid = 0; 738165951Sjhb md->md_dflags = 0; 739165951Sjhb md->md_sync_offset = 0; 740165951Sjhb md->md_provsize = 0; 74143166Snsouch } else { 742165951Sjhb md->md_did = disk->d_id; 743162289Sjhb md->md_priority = disk->d_priority; 74443166Snsouch md->md_syncid = disk->d_sync.ds_syncid; 74543166Snsouch md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 746165951Sjhb if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 74743166Snsouch md->md_sync_offset = disk->d_sync.ds_offset_done; 748165951Sjhb else 749165951Sjhb md->md_sync_offset = 0; 750165951Sjhb if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 751165951Sjhb strlcpy(md->md_provider, 752165951Sjhb disk->d_consumer->provider->name, 75346651Speter sizeof(md->md_provider)); 754165951Sjhb } 755165951Sjhb md->md_provsize = disk->d_consumer->provider->mediasize; 75643166Snsouch } 757165951Sjhb} 758165951Sjhb 759165951Sjhbvoid 760165951Sjhbg_mirror_update_metadata(struct g_mirror_disk *disk) 761165951Sjhb{ 762165951Sjhb struct g_mirror_softc *sc; 763165951Sjhb struct g_mirror_metadata md; 764165951Sjhb int error; 765165951Sjhb 766165951Sjhb g_topology_assert_not(); 767165951Sjhb sc = disk->d_softc; 768165951Sjhb sx_assert(&sc->sc_lock, SX_LOCKED); 769162289Sjhb 770165951Sjhb if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 771165951Sjhb return; 77243166Snsouch if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) 773165951Sjhb g_mirror_fill_metadata(sc, disk, &md); 774165951Sjhb error = g_mirror_write_metadata(disk, &md); 775165951Sjhb if (error == 0) { 776165951Sjhb G_MIRROR_DEBUG(2, "Metadata on %s updated.", 777165951Sjhb g_mirror_get_diskname(disk)); 778162289Sjhb } else { 779165951Sjhb G_MIRROR_DEBUG(0, 780162289Sjhb "Cannot update metadata on disk %s (error=%d).", 781165951Sjhb g_mirror_get_diskname(disk), error); 782165951Sjhb } 783} 784 785static void 786g_mirror_bump_syncid(struct g_mirror_softc *sc) 787{ 788 struct g_mirror_disk *disk; 789 790 g_topology_assert_not(); 791 sx_assert(&sc->sc_lock, SX_XLOCKED); 792 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 793 ("%s called with no active disks (device=%s).", __func__, 794 sc->sc_name)); 795 796 sc->sc_syncid++; 797 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 798 sc->sc_syncid); 799 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 800 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 801 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 802 disk->d_sync.ds_syncid = sc->sc_syncid; 803 g_mirror_update_metadata(disk); 804 } 805 } 806} 807 808static void 809g_mirror_bump_genid(struct g_mirror_softc *sc) 810{ 811 struct g_mirror_disk *disk; 812 813 g_topology_assert_not(); 814 sx_assert(&sc->sc_lock, SX_XLOCKED); 815 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 816 ("%s called with no active disks (device=%s).", __func__, 817 sc->sc_name)); 818 819 sc->sc_genid++; 820 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name, 821 sc->sc_genid); 822 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 823 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 824 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 825 disk->d_genid = sc->sc_genid; 826 g_mirror_update_metadata(disk); 827 } 828 } 829} 830 831static int 832g_mirror_idle(struct g_mirror_softc *sc, int acw) 833{ 834 struct g_mirror_disk *disk; 835 int timeout; 836 837 g_topology_assert_not(); 838 sx_assert(&sc->sc_lock, SX_XLOCKED); 839 840 if (sc->sc_provider == NULL) 841 return (0); 842 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 843 return (0); 844 if (sc->sc_idle) 845 return (0); 846 if (sc->sc_writes > 0) 847 return (0); 848 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) { 849 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write); 850 if (!g_mirror_shutdown && timeout > 0) 851 return (timeout); 852 } 853 sc->sc_idle = 1; 854 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 855 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 856 continue; 857 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 858 g_mirror_get_diskname(disk), sc->sc_name); 859 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 860 g_mirror_update_metadata(disk); 861 } 862 return (0); 863} 864 865static void 866g_mirror_unidle(struct g_mirror_softc *sc) 867{ 868 struct g_mirror_disk *disk; 869 870 g_topology_assert_not(); 871 sx_assert(&sc->sc_lock, SX_XLOCKED); 872 873 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 874 return; 875 sc->sc_idle = 0; 876 sc->sc_last_write = time_uptime; 877 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 878 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 879 continue; 880 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 881 g_mirror_get_diskname(disk), sc->sc_name); 882 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 883 g_mirror_update_metadata(disk); 884 } 885} 886 887static void 888g_mirror_flush_done(struct bio *bp) 889{ 890 struct g_mirror_softc *sc; 891 struct bio *pbp; 892 893 pbp = bp->bio_parent; 894 sc = pbp->bio_to->private; 895 mtx_lock(&sc->sc_done_mtx); 896 if (pbp->bio_error == 0) 897 pbp->bio_error = bp->bio_error; 898 pbp->bio_completed += bp->bio_completed; 899 pbp->bio_inbed++; 900 if (pbp->bio_children == pbp->bio_inbed) { 901 mtx_unlock(&sc->sc_done_mtx); 902 g_io_deliver(pbp, pbp->bio_error); 903 } else 904 mtx_unlock(&sc->sc_done_mtx); 905 g_destroy_bio(bp); 906} 907 908static void 909g_mirror_done(struct bio *bp) 910{ 911 struct g_mirror_softc *sc; 912 913 sc = bp->bio_from->geom->softc; 914 bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR; 915 mtx_lock(&sc->sc_queue_mtx); 916 bioq_insert_tail(&sc->sc_queue, bp); 917 mtx_unlock(&sc->sc_queue_mtx); 918 wakeup(sc); 919} 920 921static void 922g_mirror_regular_request(struct bio *bp) 923{ 924 struct g_mirror_softc *sc; 925 struct g_mirror_disk *disk; 926 struct bio *pbp; 927 928 g_topology_assert_not(); 929 930 pbp = bp->bio_parent; 931 sc = pbp->bio_to->private; 932 bp->bio_from->index--; 933 if (bp->bio_cmd == BIO_WRITE) 934 sc->sc_writes--; 935 disk = bp->bio_from->private; 936 if (disk == NULL) { 937 g_topology_lock(); 938 g_mirror_kill_consumer(sc, bp->bio_from); 939 g_topology_unlock(); 940 } 941 942 if (bp->bio_cmd == BIO_READ) 943 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_read, 944 bp->bio_error); 945 else if (bp->bio_cmd == BIO_WRITE) 946 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_regular_request_write, 947 bp->bio_error); 948 949 pbp->bio_inbed++; 950 KASSERT(pbp->bio_inbed <= pbp->bio_children, 951 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 952 pbp->bio_children)); 953 if (bp->bio_error == 0 && pbp->bio_error == 0) { 954 G_MIRROR_LOGREQ(3, bp, "Request delivered."); 955 g_destroy_bio(bp); 956 if (pbp->bio_children == pbp->bio_inbed) { 957 G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 958 pbp->bio_completed = pbp->bio_length; 959 if (pbp->bio_cmd == BIO_WRITE || 960 pbp->bio_cmd == BIO_DELETE) { 961 bioq_remove(&sc->sc_inflight, pbp); 962 /* Release delayed sync requests if possible. */ 963 g_mirror_sync_release(sc); 964 } 965 g_io_deliver(pbp, pbp->bio_error); 966 } 967 return; 968 } else if (bp->bio_error != 0) { 969 if (pbp->bio_error == 0) 970 pbp->bio_error = bp->bio_error; 971 if (disk != NULL) { 972 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) { 973 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN; 974 G_MIRROR_LOGREQ(0, bp, 975 "Request failed (error=%d).", 976 bp->bio_error); 977 } else { 978 G_MIRROR_LOGREQ(1, bp, 979 "Request failed (error=%d).", 980 bp->bio_error); 981 } 982 if (g_mirror_disconnect_on_failure && 983 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) 984 { 985 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 986 g_mirror_event_send(disk, 987 G_MIRROR_DISK_STATE_DISCONNECTED, 988 G_MIRROR_EVENT_DONTWAIT); 989 } 990 } 991 switch (pbp->bio_cmd) { 992 case BIO_DELETE: 993 case BIO_WRITE: 994 pbp->bio_inbed--; 995 pbp->bio_children--; 996 break; 997 } 998 } 999 g_destroy_bio(bp); 1000 1001 switch (pbp->bio_cmd) { 1002 case BIO_READ: 1003 if (pbp->bio_inbed < pbp->bio_children) 1004 break; 1005 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1) 1006 g_io_deliver(pbp, pbp->bio_error); 1007 else { 1008 pbp->bio_error = 0; 1009 mtx_lock(&sc->sc_queue_mtx); 1010 bioq_insert_tail(&sc->sc_queue, pbp); 1011 mtx_unlock(&sc->sc_queue_mtx); 1012 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1013 wakeup(sc); 1014 } 1015 break; 1016 case BIO_DELETE: 1017 case BIO_WRITE: 1018 if (pbp->bio_children == 0) { 1019 /* 1020 * All requests failed. 1021 */ 1022 } else if (pbp->bio_inbed < pbp->bio_children) { 1023 /* Do nothing. */ 1024 break; 1025 } else if (pbp->bio_children == pbp->bio_inbed) { 1026 /* Some requests succeeded. */ 1027 pbp->bio_error = 0; 1028 pbp->bio_completed = pbp->bio_length; 1029 } 1030 bioq_remove(&sc->sc_inflight, pbp); 1031 /* Release delayed sync requests if possible. */ 1032 g_mirror_sync_release(sc); 1033 g_io_deliver(pbp, pbp->bio_error); 1034 break; 1035 default: 1036 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 1037 break; 1038 } 1039} 1040 1041static void 1042g_mirror_sync_done(struct bio *bp) 1043{ 1044 struct g_mirror_softc *sc; 1045 1046 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 1047 sc = bp->bio_from->geom->softc; 1048 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC; 1049 mtx_lock(&sc->sc_queue_mtx); 1050 bioq_insert_tail(&sc->sc_queue, bp); 1051 mtx_unlock(&sc->sc_queue_mtx); 1052 wakeup(sc); 1053} 1054 1055static void 1056g_mirror_candelete(struct bio *bp) 1057{ 1058 struct g_mirror_softc *sc; 1059 struct g_mirror_disk *disk; 1060 int *val; 1061 1062 sc = bp->bio_to->private; 1063 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1064 if (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) 1065 break; 1066 } 1067 val = (int *)bp->bio_data; 1068 *val = (disk != NULL); 1069 g_io_deliver(bp, 0); 1070} 1071 1072static void 1073g_mirror_kernel_dump(struct bio *bp) 1074{ 1075 struct g_mirror_softc *sc; 1076 struct g_mirror_disk *disk; 1077 struct bio *cbp; 1078 struct g_kerneldump *gkd; 1079 1080 /* 1081 * We configure dumping to the first component, because this component 1082 * will be used for reading with 'prefer' balance algorithm. 1083 * If the component with the highest priority is currently disconnected 1084 * we will not be able to read the dump after the reboot if it will be 1085 * connected and synchronized later. Can we do something better? 1086 */ 1087 sc = bp->bio_to->private; 1088 disk = LIST_FIRST(&sc->sc_disks); 1089 1090 gkd = (struct g_kerneldump *)bp->bio_data; 1091 if (gkd->length > bp->bio_to->mediasize) 1092 gkd->length = bp->bio_to->mediasize; 1093 cbp = g_clone_bio(bp); 1094 if (cbp == NULL) { 1095 g_io_deliver(bp, ENOMEM); 1096 return; 1097 } 1098 cbp->bio_done = g_std_done; 1099 g_io_request(cbp, disk->d_consumer); 1100 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.", 1101 g_mirror_get_diskname(disk)); 1102} 1103 1104static void 1105g_mirror_flush(struct g_mirror_softc *sc, struct bio *bp) 1106{ 1107 struct bio_queue_head queue; 1108 struct g_mirror_disk *disk; 1109 struct g_consumer *cp; 1110 struct bio *cbp; 1111 1112 bioq_init(&queue); 1113 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1114 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1115 continue; 1116 cbp = g_clone_bio(bp); 1117 if (cbp == NULL) { 1118 while ((cbp = bioq_takefirst(&queue)) != NULL) 1119 g_destroy_bio(cbp); 1120 if (bp->bio_error == 0) 1121 bp->bio_error = ENOMEM; 1122 g_io_deliver(bp, bp->bio_error); 1123 return; 1124 } 1125 bioq_insert_tail(&queue, cbp); 1126 cbp->bio_done = g_mirror_flush_done; 1127 cbp->bio_caller1 = disk; 1128 cbp->bio_to = disk->d_consumer->provider; 1129 } 1130 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1131 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1132 disk = cbp->bio_caller1; 1133 cbp->bio_caller1 = NULL; 1134 cp = disk->d_consumer; 1135 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1136 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1137 cp->acr, cp->acw, cp->ace)); 1138 g_io_request(cbp, disk->d_consumer); 1139 } 1140} 1141 1142static void 1143g_mirror_start(struct bio *bp) 1144{ 1145 struct g_mirror_softc *sc; 1146 1147 sc = bp->bio_to->private; 1148 /* 1149 * If sc == NULL or there are no valid disks, provider's error 1150 * should be set and g_mirror_start() should not be called at all. 1151 */ 1152 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1153 ("Provider's error should be set (error=%d)(mirror=%s).", 1154 bp->bio_to->error, bp->bio_to->name)); 1155 G_MIRROR_LOGREQ(3, bp, "Request received."); 1156 1157 switch (bp->bio_cmd) { 1158 case BIO_READ: 1159 case BIO_WRITE: 1160 case BIO_DELETE: 1161 break; 1162 case BIO_FLUSH: 1163 g_mirror_flush(sc, bp); 1164 return; 1165 case BIO_GETATTR: 1166 if (!strcmp(bp->bio_attribute, "GEOM::candelete")) { 1167 g_mirror_candelete(bp); 1168 return; 1169 } else if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) { 1170 g_mirror_kernel_dump(bp); 1171 return; 1172 } 1173 /* FALLTHROUGH */ 1174 default: 1175 g_io_deliver(bp, EOPNOTSUPP); 1176 return; 1177 } 1178 mtx_lock(&sc->sc_queue_mtx); 1179 bioq_insert_tail(&sc->sc_queue, bp); 1180 mtx_unlock(&sc->sc_queue_mtx); 1181 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 1182 wakeup(sc); 1183} 1184 1185/* 1186 * Return TRUE if the given request is colliding with a in-progress 1187 * synchronization request. 1188 */ 1189static int 1190g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp) 1191{ 1192 struct g_mirror_disk *disk; 1193 struct bio *sbp; 1194 off_t rstart, rend, sstart, send; 1195 u_int i; 1196 1197 if (sc->sc_sync.ds_ndisks == 0) 1198 return (0); 1199 rstart = bp->bio_offset; 1200 rend = bp->bio_offset + bp->bio_length; 1201 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1202 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING) 1203 continue; 1204 for (i = 0; i < g_mirror_syncreqs; i++) { 1205 sbp = disk->d_sync.ds_bios[i]; 1206 if (sbp == NULL) 1207 continue; 1208 sstart = sbp->bio_offset; 1209 send = sbp->bio_offset + sbp->bio_length; 1210 if (rend > sstart && rstart < send) 1211 return (1); 1212 } 1213 } 1214 return (0); 1215} 1216 1217/* 1218 * Return TRUE if the given sync request is colliding with a in-progress regular 1219 * request. 1220 */ 1221static int 1222g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp) 1223{ 1224 off_t rstart, rend, sstart, send; 1225 struct bio *bp; 1226 1227 if (sc->sc_sync.ds_ndisks == 0) 1228 return (0); 1229 sstart = sbp->bio_offset; 1230 send = sbp->bio_offset + sbp->bio_length; 1231 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) { 1232 rstart = bp->bio_offset; 1233 rend = bp->bio_offset + bp->bio_length; 1234 if (rend > sstart && rstart < send) 1235 return (1); 1236 } 1237 return (0); 1238} 1239 1240/* 1241 * Puts request onto delayed queue. 1242 */ 1243static void 1244g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp) 1245{ 1246 1247 G_MIRROR_LOGREQ(2, bp, "Delaying request."); 1248 bioq_insert_head(&sc->sc_regular_delayed, bp); 1249} 1250 1251/* 1252 * Puts synchronization request onto delayed queue. 1253 */ 1254static void 1255g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp) 1256{ 1257 1258 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request."); 1259 bioq_insert_tail(&sc->sc_sync_delayed, bp); 1260} 1261 1262/* 1263 * Releases delayed regular requests which don't collide anymore with sync 1264 * requests. 1265 */ 1266static void 1267g_mirror_regular_release(struct g_mirror_softc *sc) 1268{ 1269 struct bio *bp, *bp2; 1270 1271 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) { 1272 if (g_mirror_sync_collision(sc, bp)) 1273 continue; 1274 bioq_remove(&sc->sc_regular_delayed, bp); 1275 G_MIRROR_LOGREQ(2, bp, "Releasing delayed request (%p).", bp); 1276 mtx_lock(&sc->sc_queue_mtx); 1277 bioq_insert_head(&sc->sc_queue, bp); 1278#if 0 1279 /* 1280 * wakeup() is not needed, because this function is called from 1281 * the worker thread. 1282 */ 1283 wakeup(&sc->sc_queue); 1284#endif 1285 mtx_unlock(&sc->sc_queue_mtx); 1286 } 1287} 1288 1289/* 1290 * Releases delayed sync requests which don't collide anymore with regular 1291 * requests. 1292 */ 1293static void 1294g_mirror_sync_release(struct g_mirror_softc *sc) 1295{ 1296 struct bio *bp, *bp2; 1297 1298 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) { 1299 if (g_mirror_regular_collision(sc, bp)) 1300 continue; 1301 bioq_remove(&sc->sc_sync_delayed, bp); 1302 G_MIRROR_LOGREQ(2, bp, 1303 "Releasing delayed synchronization request."); 1304 g_io_request(bp, bp->bio_from); 1305 } 1306} 1307 1308/* 1309 * Handle synchronization requests. 1310 * Every synchronization request is two-steps process: first, READ request is 1311 * send to active provider and then WRITE request (with read data) to the provider 1312 * being synchronized. When WRITE is finished, new synchronization request is 1313 * send. 1314 */ 1315static void 1316g_mirror_sync_request(struct bio *bp) 1317{ 1318 struct g_mirror_softc *sc; 1319 struct g_mirror_disk *disk; 1320 1321 bp->bio_from->index--; 1322 sc = bp->bio_from->geom->softc; 1323 disk = bp->bio_from->private; 1324 if (disk == NULL) { 1325 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 1326 g_topology_lock(); 1327 g_mirror_kill_consumer(sc, bp->bio_from); 1328 g_topology_unlock(); 1329 free(bp->bio_data, M_MIRROR); 1330 g_destroy_bio(bp); 1331 sx_xlock(&sc->sc_lock); 1332 return; 1333 } 1334 1335 /* 1336 * Synchronization request. 1337 */ 1338 switch (bp->bio_cmd) { 1339 case BIO_READ: 1340 { 1341 struct g_consumer *cp; 1342 1343 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_read, 1344 bp->bio_error); 1345 1346 if (bp->bio_error != 0) { 1347 G_MIRROR_LOGREQ(0, bp, 1348 "Synchronization request failed (error=%d).", 1349 bp->bio_error); 1350 g_destroy_bio(bp); 1351 return; 1352 } 1353 G_MIRROR_LOGREQ(3, bp, 1354 "Synchronization request half-finished."); 1355 bp->bio_cmd = BIO_WRITE; 1356 bp->bio_cflags = 0; 1357 cp = disk->d_consumer; 1358 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1359 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1360 cp->acr, cp->acw, cp->ace)); 1361 cp->index++; 1362 g_io_request(bp, cp); 1363 return; 1364 } 1365 case BIO_WRITE: 1366 { 1367 struct g_mirror_disk_sync *sync; 1368 off_t offset; 1369 void *data; 1370 int i; 1371 1372 KFAIL_POINT_ERROR(DEBUG_FP, g_mirror_sync_request_write, 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_destroy_bio(bp); 1380 sc->sc_bump_id |= G_MIRROR_BUMP_GENID; 1381 g_mirror_event_send(disk, 1382 G_MIRROR_DISK_STATE_DISCONNECTED, 1383 G_MIRROR_EVENT_DONTWAIT); 1384 return; 1385 } 1386 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 1387 sync = &disk->d_sync; 1388 if (sync->ds_offset >= sc->sc_mediasize || 1389 sync->ds_consumer == NULL || 1390 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1391 /* Don't send more synchronization requests. */ 1392 sync->ds_inflight--; 1393 if (sync->ds_bios != NULL) { 1394 i = (int)(uintptr_t)bp->bio_caller1; 1395 sync->ds_bios[i] = NULL; 1396 } 1397 free(bp->bio_data, M_MIRROR); 1398 g_destroy_bio(bp); 1399 if (sync->ds_inflight > 0) 1400 return; 1401 if (sync->ds_consumer == NULL || 1402 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1403 return; 1404 } 1405 /* Disk up-to-date, activate it. */ 1406 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 1407 G_MIRROR_EVENT_DONTWAIT); 1408 return; 1409 } 1410 1411 /* Send next synchronization request. */ 1412 data = bp->bio_data; 1413 g_reset_bio(bp); 1414 bp->bio_cmd = BIO_READ; 1415 bp->bio_offset = sync->ds_offset; 1416 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 1417 sync->ds_offset += bp->bio_length; 1418 bp->bio_done = g_mirror_sync_done; 1419 bp->bio_data = data; 1420 bp->bio_from = sync->ds_consumer; 1421 bp->bio_to = sc->sc_provider; 1422 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 1423 sync->ds_consumer->index++; 1424 /* 1425 * Delay the request if it is colliding with a regular request. 1426 */ 1427 if (g_mirror_regular_collision(sc, bp)) 1428 g_mirror_sync_delay(sc, bp); 1429 else 1430 g_io_request(bp, sync->ds_consumer); 1431 1432 /* Release delayed requests if possible. */ 1433 g_mirror_regular_release(sc); 1434 1435 /* Find the smallest offset */ 1436 offset = sc->sc_mediasize; 1437 for (i = 0; i < g_mirror_syncreqs; i++) { 1438 bp = sync->ds_bios[i]; 1439 if (bp->bio_offset < offset) 1440 offset = bp->bio_offset; 1441 } 1442 if (sync->ds_offset_done + (MAXPHYS * 100) < offset) { 1443 /* Update offset_done on every 100 blocks. */ 1444 sync->ds_offset_done = offset; 1445 g_mirror_update_metadata(disk); 1446 } 1447 return; 1448 } 1449 default: 1450 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1451 bp->bio_cmd, sc->sc_name)); 1452 break; 1453 } 1454} 1455 1456static void 1457g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 1458{ 1459 struct g_mirror_disk *disk; 1460 struct g_consumer *cp; 1461 struct bio *cbp; 1462 1463 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1464 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 1465 break; 1466 } 1467 if (disk == NULL) { 1468 if (bp->bio_error == 0) 1469 bp->bio_error = ENXIO; 1470 g_io_deliver(bp, bp->bio_error); 1471 return; 1472 } 1473 cbp = g_clone_bio(bp); 1474 if (cbp == NULL) { 1475 if (bp->bio_error == 0) 1476 bp->bio_error = ENOMEM; 1477 g_io_deliver(bp, bp->bio_error); 1478 return; 1479 } 1480 /* 1481 * Fill in the component buf structure. 1482 */ 1483 cp = disk->d_consumer; 1484 cbp->bio_done = g_mirror_done; 1485 cbp->bio_to = cp->provider; 1486 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1487 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1488 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1489 cp->acw, cp->ace)); 1490 cp->index++; 1491 g_io_request(cbp, cp); 1492} 1493 1494static void 1495g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 1496{ 1497 struct g_mirror_disk *disk; 1498 struct g_consumer *cp; 1499 struct bio *cbp; 1500 1501 disk = g_mirror_get_disk(sc); 1502 if (disk == NULL) { 1503 if (bp->bio_error == 0) 1504 bp->bio_error = ENXIO; 1505 g_io_deliver(bp, bp->bio_error); 1506 return; 1507 } 1508 cbp = g_clone_bio(bp); 1509 if (cbp == NULL) { 1510 if (bp->bio_error == 0) 1511 bp->bio_error = ENOMEM; 1512 g_io_deliver(bp, bp->bio_error); 1513 return; 1514 } 1515 /* 1516 * Fill in the component buf structure. 1517 */ 1518 cp = disk->d_consumer; 1519 cbp->bio_done = g_mirror_done; 1520 cbp->bio_to = cp->provider; 1521 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1522 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1523 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1524 cp->acw, cp->ace)); 1525 cp->index++; 1526 g_io_request(cbp, cp); 1527} 1528 1529#define TRACK_SIZE (1 * 1024 * 1024) 1530#define LOAD_SCALE 256 1531#define ABS(x) (((x) >= 0) ? (x) : (-(x))) 1532 1533static void 1534g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 1535{ 1536 struct g_mirror_disk *disk, *dp; 1537 struct g_consumer *cp; 1538 struct bio *cbp; 1539 int prio, best; 1540 1541 /* Find a disk with the smallest load. */ 1542 disk = NULL; 1543 best = INT_MAX; 1544 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1545 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1546 continue; 1547 prio = dp->load; 1548 /* If disk head is precisely in position - highly prefer it. */ 1549 if (dp->d_last_offset == bp->bio_offset) 1550 prio -= 2 * LOAD_SCALE; 1551 else 1552 /* If disk head is close to position - prefer it. */ 1553 if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE) 1554 prio -= 1 * LOAD_SCALE; 1555 if (prio <= best) { 1556 disk = dp; 1557 best = prio; 1558 } 1559 } 1560 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name)); 1561 cbp = g_clone_bio(bp); 1562 if (cbp == NULL) { 1563 if (bp->bio_error == 0) 1564 bp->bio_error = ENOMEM; 1565 g_io_deliver(bp, bp->bio_error); 1566 return; 1567 } 1568 /* 1569 * Fill in the component buf structure. 1570 */ 1571 cp = disk->d_consumer; 1572 cbp->bio_done = g_mirror_done; 1573 cbp->bio_to = cp->provider; 1574 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1575 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1576 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1577 cp->acw, cp->ace)); 1578 cp->index++; 1579 /* Remember last head position */ 1580 disk->d_last_offset = bp->bio_offset + bp->bio_length; 1581 /* Update loads. */ 1582 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1583 dp->load = (dp->d_consumer->index * LOAD_SCALE + 1584 dp->load * 7) / 8; 1585 } 1586 g_io_request(cbp, cp); 1587} 1588 1589static void 1590g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 1591{ 1592 struct bio_queue_head queue; 1593 struct g_mirror_disk *disk; 1594 struct g_consumer *cp; 1595 struct bio *cbp; 1596 off_t left, mod, offset, slice; 1597 u_char *data; 1598 u_int ndisks; 1599 1600 if (bp->bio_length <= sc->sc_slice) { 1601 g_mirror_request_round_robin(sc, bp); 1602 return; 1603 } 1604 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 1605 slice = bp->bio_length / ndisks; 1606 mod = slice % sc->sc_provider->sectorsize; 1607 if (mod != 0) 1608 slice += sc->sc_provider->sectorsize - mod; 1609 /* 1610 * Allocate all bios before sending any request, so we can 1611 * return ENOMEM in nice and clean way. 1612 */ 1613 left = bp->bio_length; 1614 offset = bp->bio_offset; 1615 data = bp->bio_data; 1616 bioq_init(&queue); 1617 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1618 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 1619 continue; 1620 cbp = g_clone_bio(bp); 1621 if (cbp == NULL) { 1622 while ((cbp = bioq_takefirst(&queue)) != NULL) 1623 g_destroy_bio(cbp); 1624 if (bp->bio_error == 0) 1625 bp->bio_error = ENOMEM; 1626 g_io_deliver(bp, bp->bio_error); 1627 return; 1628 } 1629 bioq_insert_tail(&queue, cbp); 1630 cbp->bio_done = g_mirror_done; 1631 cbp->bio_caller1 = disk; 1632 cbp->bio_to = disk->d_consumer->provider; 1633 cbp->bio_offset = offset; 1634 cbp->bio_data = data; 1635 cbp->bio_length = MIN(left, slice); 1636 left -= cbp->bio_length; 1637 if (left == 0) 1638 break; 1639 offset += cbp->bio_length; 1640 data += cbp->bio_length; 1641 } 1642 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1643 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1644 disk = cbp->bio_caller1; 1645 cbp->bio_caller1 = NULL; 1646 cp = disk->d_consumer; 1647 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1648 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 1649 cp->acr, cp->acw, cp->ace)); 1650 disk->d_consumer->index++; 1651 g_io_request(cbp, disk->d_consumer); 1652 } 1653} 1654 1655static void 1656g_mirror_register_request(struct bio *bp) 1657{ 1658 struct g_mirror_softc *sc; 1659 1660 sc = bp->bio_to->private; 1661 switch (bp->bio_cmd) { 1662 case BIO_READ: 1663 switch (sc->sc_balance) { 1664 case G_MIRROR_BALANCE_LOAD: 1665 g_mirror_request_load(sc, bp); 1666 break; 1667 case G_MIRROR_BALANCE_PREFER: 1668 g_mirror_request_prefer(sc, bp); 1669 break; 1670 case G_MIRROR_BALANCE_ROUND_ROBIN: 1671 g_mirror_request_round_robin(sc, bp); 1672 break; 1673 case G_MIRROR_BALANCE_SPLIT: 1674 g_mirror_request_split(sc, bp); 1675 break; 1676 } 1677 return; 1678 case BIO_WRITE: 1679 case BIO_DELETE: 1680 { 1681 struct g_mirror_disk *disk; 1682 struct g_mirror_disk_sync *sync; 1683 struct bio_queue_head queue; 1684 struct g_consumer *cp; 1685 struct bio *cbp; 1686 1687 /* 1688 * Delay the request if it is colliding with a synchronization 1689 * request. 1690 */ 1691 if (g_mirror_sync_collision(sc, bp)) { 1692 g_mirror_regular_delay(sc, bp); 1693 return; 1694 } 1695 1696 if (sc->sc_idle) 1697 g_mirror_unidle(sc); 1698 else 1699 sc->sc_last_write = time_uptime; 1700 1701 /* 1702 * Allocate all bios before sending any request, so we can 1703 * return ENOMEM in nice and clean way. 1704 */ 1705 bioq_init(&queue); 1706 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1707 sync = &disk->d_sync; 1708 switch (disk->d_state) { 1709 case G_MIRROR_DISK_STATE_ACTIVE: 1710 break; 1711 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1712 if (bp->bio_offset >= sync->ds_offset) 1713 continue; 1714 break; 1715 default: 1716 continue; 1717 } 1718 if (bp->bio_cmd == BIO_DELETE && 1719 (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0) 1720 continue; 1721 cbp = g_clone_bio(bp); 1722 if (cbp == NULL) { 1723 while ((cbp = bioq_takefirst(&queue)) != NULL) 1724 g_destroy_bio(cbp); 1725 if (bp->bio_error == 0) 1726 bp->bio_error = ENOMEM; 1727 g_io_deliver(bp, bp->bio_error); 1728 return; 1729 } 1730 bioq_insert_tail(&queue, cbp); 1731 cbp->bio_done = g_mirror_done; 1732 cp = disk->d_consumer; 1733 cbp->bio_caller1 = cp; 1734 cbp->bio_to = cp->provider; 1735 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1, 1736 ("Consumer %s not opened (r%dw%de%d).", 1737 cp->provider->name, cp->acr, cp->acw, cp->ace)); 1738 } 1739 if (bioq_first(&queue) == NULL) { 1740 g_io_deliver(bp, EOPNOTSUPP); 1741 return; 1742 } 1743 while ((cbp = bioq_takefirst(&queue)) != NULL) { 1744 G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1745 cp = cbp->bio_caller1; 1746 cbp->bio_caller1 = NULL; 1747 cp->index++; 1748 sc->sc_writes++; 1749 g_io_request(cbp, cp); 1750 } 1751 /* 1752 * Put request onto inflight queue, so we can check if new 1753 * synchronization requests don't collide with it. 1754 */ 1755 bioq_insert_tail(&sc->sc_inflight, bp); 1756 /* 1757 * Bump syncid on first write. 1758 */ 1759 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) { 1760 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 1761 g_mirror_bump_syncid(sc); 1762 } 1763 return; 1764 } 1765 default: 1766 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 1767 bp->bio_cmd, sc->sc_name)); 1768 break; 1769 } 1770} 1771 1772static int 1773g_mirror_can_destroy(struct g_mirror_softc *sc) 1774{ 1775 struct g_geom *gp; 1776 struct g_consumer *cp; 1777 1778 g_topology_assert(); 1779 gp = sc->sc_geom; 1780 if (gp->softc == NULL) 1781 return (1); 1782 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0) 1783 return (0); 1784 LIST_FOREACH(cp, &gp->consumer, consumer) { 1785 if (g_mirror_is_busy(sc, cp)) 1786 return (0); 1787 } 1788 gp = sc->sc_sync.ds_geom; 1789 LIST_FOREACH(cp, &gp->consumer, consumer) { 1790 if (g_mirror_is_busy(sc, cp)) 1791 return (0); 1792 } 1793 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 1794 sc->sc_name); 1795 return (1); 1796} 1797 1798static int 1799g_mirror_try_destroy(struct g_mirror_softc *sc) 1800{ 1801 1802 if (sc->sc_rootmount != NULL) { 1803 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 1804 sc->sc_rootmount); 1805 root_mount_rel(sc->sc_rootmount); 1806 sc->sc_rootmount = NULL; 1807 } 1808 g_topology_lock(); 1809 if (!g_mirror_can_destroy(sc)) { 1810 g_topology_unlock(); 1811 return (0); 1812 } 1813 sc->sc_geom->softc = NULL; 1814 sc->sc_sync.ds_geom->softc = NULL; 1815 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1816 g_topology_unlock(); 1817 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1818 &sc->sc_worker); 1819 /* Unlock sc_lock here, as it can be destroyed after wakeup. */ 1820 sx_xunlock(&sc->sc_lock); 1821 wakeup(&sc->sc_worker); 1822 sc->sc_worker = NULL; 1823 } else { 1824 g_topology_unlock(); 1825 g_mirror_destroy_device(sc); 1826 } 1827 return (1); 1828} 1829 1830/* 1831 * Worker thread. 1832 */ 1833static void 1834g_mirror_worker(void *arg) 1835{ 1836 struct g_mirror_softc *sc; 1837 struct g_mirror_event *ep; 1838 struct bio *bp; 1839 int timeout; 1840 1841 sc = arg; 1842 thread_lock(curthread); 1843 sched_prio(curthread, PRIBIO); 1844 thread_unlock(curthread); 1845 1846 sx_xlock(&sc->sc_lock); 1847 for (;;) { 1848 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 1849 /* 1850 * First take a look at events. 1851 * This is important to handle events before any I/O requests. 1852 */ 1853 ep = g_mirror_event_get(sc); 1854 if (ep != NULL) { 1855 g_mirror_event_remove(sc, ep); 1856 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 1857 /* Update only device status. */ 1858 G_MIRROR_DEBUG(3, 1859 "Running event for device %s.", 1860 sc->sc_name); 1861 ep->e_error = 0; 1862 g_mirror_update_device(sc, true); 1863 } else { 1864 /* Update disk status. */ 1865 G_MIRROR_DEBUG(3, "Running event for disk %s.", 1866 g_mirror_get_diskname(ep->e_disk)); 1867 ep->e_error = g_mirror_update_disk(ep->e_disk, 1868 ep->e_state); 1869 if (ep->e_error == 0) 1870 g_mirror_update_device(sc, false); 1871 } 1872 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 1873 KASSERT(ep->e_error == 0, 1874 ("Error cannot be handled.")); 1875 g_mirror_event_free(ep); 1876 } else { 1877 ep->e_flags |= G_MIRROR_EVENT_DONE; 1878 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 1879 ep); 1880 mtx_lock(&sc->sc_events_mtx); 1881 wakeup(ep); 1882 mtx_unlock(&sc->sc_events_mtx); 1883 } 1884 if ((sc->sc_flags & 1885 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1886 if (g_mirror_try_destroy(sc)) { 1887 curthread->td_pflags &= ~TDP_GEOM; 1888 G_MIRROR_DEBUG(1, "Thread exiting."); 1889 kproc_exit(0); 1890 } 1891 } 1892 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 1893 continue; 1894 } 1895 /* 1896 * Check if we can mark array as CLEAN and if we can't take 1897 * how much seconds should we wait. 1898 */ 1899 timeout = g_mirror_idle(sc, -1); 1900 /* 1901 * Now I/O requests. 1902 */ 1903 /* Get first request from the queue. */ 1904 mtx_lock(&sc->sc_queue_mtx); 1905 bp = bioq_takefirst(&sc->sc_queue); 1906 if (bp == NULL) { 1907 if ((sc->sc_flags & 1908 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 1909 mtx_unlock(&sc->sc_queue_mtx); 1910 if (g_mirror_try_destroy(sc)) { 1911 curthread->td_pflags &= ~TDP_GEOM; 1912 G_MIRROR_DEBUG(1, "Thread exiting."); 1913 kproc_exit(0); 1914 } 1915 mtx_lock(&sc->sc_queue_mtx); 1916 if (bioq_first(&sc->sc_queue) != NULL) { 1917 mtx_unlock(&sc->sc_queue_mtx); 1918 continue; 1919 } 1920 } 1921 sx_xunlock(&sc->sc_lock); 1922 /* 1923 * XXX: We can miss an event here, because an event 1924 * can be added without sx-device-lock and without 1925 * mtx-queue-lock. Maybe I should just stop using 1926 * dedicated mutex for events synchronization and 1927 * stick with the queue lock? 1928 * The event will hang here until next I/O request 1929 * or next event is received. 1930 */ 1931 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 1932 timeout * hz); 1933 sx_xlock(&sc->sc_lock); 1934 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 1935 continue; 1936 } 1937 mtx_unlock(&sc->sc_queue_mtx); 1938 1939 if (bp->bio_from->geom == sc->sc_sync.ds_geom && 1940 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 1941 g_mirror_sync_request(bp); /* READ */ 1942 } else if (bp->bio_to != sc->sc_provider) { 1943 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) 1944 g_mirror_regular_request(bp); 1945 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 1946 g_mirror_sync_request(bp); /* WRITE */ 1947 else { 1948 KASSERT(0, 1949 ("Invalid request cflags=0x%hx to=%s.", 1950 bp->bio_cflags, bp->bio_to->name)); 1951 } 1952 } else { 1953 g_mirror_register_request(bp); 1954 } 1955 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__); 1956 } 1957} 1958 1959static void 1960g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 1961{ 1962 1963 sx_assert(&sc->sc_lock, SX_LOCKED); 1964 1965 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0) 1966 return; 1967 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1968 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1969 g_mirror_get_diskname(disk), sc->sc_name); 1970 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1971 } else if (sc->sc_idle && 1972 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1973 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1974 g_mirror_get_diskname(disk), sc->sc_name); 1975 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1976 } 1977} 1978 1979static void 1980g_mirror_sync_start(struct g_mirror_disk *disk) 1981{ 1982 struct g_mirror_softc *sc; 1983 struct g_consumer *cp; 1984 struct bio *bp; 1985 int error, i; 1986 1987 g_topology_assert_not(); 1988 sc = disk->d_softc; 1989 sx_assert(&sc->sc_lock, SX_LOCKED); 1990 1991 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1992 ("Disk %s is not marked for synchronization.", 1993 g_mirror_get_diskname(disk))); 1994 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1995 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1996 sc->sc_state)); 1997 1998 sx_xunlock(&sc->sc_lock); 1999 g_topology_lock(); 2000 cp = g_new_consumer(sc->sc_sync.ds_geom); 2001 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 2002 error = g_attach(cp, sc->sc_provider); 2003 KASSERT(error == 0, 2004 ("Cannot attach to %s (error=%d).", sc->sc_name, error)); 2005 error = g_access(cp, 1, 0, 0); 2006 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error)); 2007 g_topology_unlock(); 2008 sx_xlock(&sc->sc_lock); 2009 2010 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 2011 g_mirror_get_diskname(disk)); 2012 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0) 2013 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2014 KASSERT(disk->d_sync.ds_consumer == NULL, 2015 ("Sync consumer already exists (device=%s, disk=%s).", 2016 sc->sc_name, g_mirror_get_diskname(disk))); 2017 2018 disk->d_sync.ds_consumer = cp; 2019 disk->d_sync.ds_consumer->private = disk; 2020 disk->d_sync.ds_consumer->index = 0; 2021 2022 /* 2023 * Allocate memory for synchronization bios and initialize them. 2024 */ 2025 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs, 2026 M_MIRROR, M_WAITOK); 2027 for (i = 0; i < g_mirror_syncreqs; i++) { 2028 bp = g_alloc_bio(); 2029 disk->d_sync.ds_bios[i] = bp; 2030 bp->bio_parent = NULL; 2031 bp->bio_cmd = BIO_READ; 2032 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 2033 bp->bio_cflags = 0; 2034 bp->bio_offset = disk->d_sync.ds_offset; 2035 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 2036 disk->d_sync.ds_offset += bp->bio_length; 2037 bp->bio_done = g_mirror_sync_done; 2038 bp->bio_from = disk->d_sync.ds_consumer; 2039 bp->bio_to = sc->sc_provider; 2040 bp->bio_caller1 = (void *)(uintptr_t)i; 2041 } 2042 2043 /* Increase the number of disks in SYNCHRONIZING state. */ 2044 sc->sc_sync.ds_ndisks++; 2045 /* Set the number of in-flight synchronization requests. */ 2046 disk->d_sync.ds_inflight = g_mirror_syncreqs; 2047 2048 /* 2049 * Fire off first synchronization requests. 2050 */ 2051 for (i = 0; i < g_mirror_syncreqs; i++) { 2052 bp = disk->d_sync.ds_bios[i]; 2053 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 2054 disk->d_sync.ds_consumer->index++; 2055 /* 2056 * Delay the request if it is colliding with a regular request. 2057 */ 2058 if (g_mirror_regular_collision(sc, bp)) 2059 g_mirror_sync_delay(sc, bp); 2060 else 2061 g_io_request(bp, disk->d_sync.ds_consumer); 2062 } 2063} 2064 2065/* 2066 * Stop synchronization process. 2067 * type: 0 - synchronization finished 2068 * 1 - synchronization stopped 2069 */ 2070static void 2071g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 2072{ 2073 struct g_mirror_softc *sc; 2074 struct g_consumer *cp; 2075 2076 g_topology_assert_not(); 2077 sc = disk->d_softc; 2078 sx_assert(&sc->sc_lock, SX_LOCKED); 2079 2080 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2081 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2082 g_mirror_disk_state2str(disk->d_state))); 2083 if (disk->d_sync.ds_consumer == NULL) 2084 return; 2085 2086 if (type == 0) { 2087 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 2088 sc->sc_name, g_mirror_get_diskname(disk)); 2089 } else /* if (type == 1) */ { 2090 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 2091 sc->sc_name, g_mirror_get_diskname(disk)); 2092 } 2093 free(disk->d_sync.ds_bios, M_MIRROR); 2094 disk->d_sync.ds_bios = NULL; 2095 cp = disk->d_sync.ds_consumer; 2096 disk->d_sync.ds_consumer = NULL; 2097 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2098 sc->sc_sync.ds_ndisks--; 2099 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */ 2100 g_topology_lock(); 2101 g_mirror_kill_consumer(sc, cp); 2102 g_topology_unlock(); 2103 sx_xlock(&sc->sc_lock); 2104} 2105 2106static void 2107g_mirror_launch_provider(struct g_mirror_softc *sc) 2108{ 2109 struct g_mirror_disk *disk; 2110 struct g_provider *pp, *dp; 2111 2112 sx_assert(&sc->sc_lock, SX_LOCKED); 2113 2114 g_topology_lock(); 2115 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 2116 pp->flags |= G_PF_DIRECT_RECEIVE; 2117 pp->mediasize = sc->sc_mediasize; 2118 pp->sectorsize = sc->sc_sectorsize; 2119 pp->stripesize = 0; 2120 pp->stripeoffset = 0; 2121 2122 /* Splitting of unmapped BIO's could work but isn't implemented now */ 2123 if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT) 2124 pp->flags |= G_PF_ACCEPT_UNMAPPED; 2125 2126 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2127 if (disk->d_consumer && disk->d_consumer->provider) { 2128 dp = disk->d_consumer->provider; 2129 if (dp->stripesize > pp->stripesize) { 2130 pp->stripesize = dp->stripesize; 2131 pp->stripeoffset = dp->stripeoffset; 2132 } 2133 /* A provider underneath us doesn't support unmapped */ 2134 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) { 2135 G_MIRROR_DEBUG(0, "Cancelling unmapped " 2136 "because of %s.", dp->name); 2137 pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 2138 } 2139 } 2140 } 2141 pp->private = sc; 2142 sc->sc_refcnt++; 2143 sc->sc_provider = pp; 2144 g_error_provider(pp, 0); 2145 g_topology_unlock(); 2146 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name, 2147 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks); 2148 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2149 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2150 g_mirror_sync_start(disk); 2151 } 2152} 2153 2154static void 2155g_mirror_destroy_provider(struct g_mirror_softc *sc) 2156{ 2157 struct g_mirror_disk *disk; 2158 struct bio *bp; 2159 2160 g_topology_assert_not(); 2161 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 2162 sc->sc_name)); 2163 2164 g_topology_lock(); 2165 g_error_provider(sc->sc_provider, ENXIO); 2166 mtx_lock(&sc->sc_queue_mtx); 2167 while ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) { 2168 /* 2169 * Abort any pending I/O that wasn't generated by us. 2170 * Synchronization requests and requests destined for individual 2171 * mirror components can be destroyed immediately. 2172 */ 2173 if (bp->bio_to == sc->sc_provider && 2174 bp->bio_from->geom != sc->sc_sync.ds_geom) { 2175 g_io_deliver(bp, ENXIO); 2176 } else { 2177 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) 2178 free(bp->bio_data, M_MIRROR); 2179 g_destroy_bio(bp); 2180 } 2181 } 2182 mtx_unlock(&sc->sc_queue_mtx); 2183 g_wither_provider(sc->sc_provider, ENXIO); 2184 sc->sc_provider = NULL; 2185 G_MIRROR_DEBUG(0, "Device %s: provider destroyed.", sc->sc_name); 2186 g_topology_unlock(); 2187 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2188 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 2189 g_mirror_sync_stop(disk, 1); 2190 } 2191} 2192 2193static void 2194g_mirror_go(void *arg) 2195{ 2196 struct g_mirror_softc *sc; 2197 2198 sc = arg; 2199 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 2200 g_mirror_event_send(sc, 0, 2201 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 2202} 2203 2204static u_int 2205g_mirror_determine_state(struct g_mirror_disk *disk) 2206{ 2207 struct g_mirror_softc *sc; 2208 u_int state; 2209 2210 sc = disk->d_softc; 2211 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 2212 if ((disk->d_flags & 2213 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0 && 2214 (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 || 2215 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0)) { 2216 /* Disk does not need synchronization. */ 2217 state = G_MIRROR_DISK_STATE_ACTIVE; 2218 } else { 2219 if ((sc->sc_flags & 2220 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2221 (disk->d_flags & 2222 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2223 /* 2224 * We can start synchronization from 2225 * the stored offset. 2226 */ 2227 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2228 } else { 2229 state = G_MIRROR_DISK_STATE_STALE; 2230 } 2231 } 2232 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 2233 /* 2234 * Reset all synchronization data for this disk, 2235 * because if it even was synchronized, it was 2236 * synchronized to disks with different syncid. 2237 */ 2238 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2239 disk->d_sync.ds_offset = 0; 2240 disk->d_sync.ds_offset_done = 0; 2241 disk->d_sync.ds_syncid = sc->sc_syncid; 2242 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 2243 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 2244 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 2245 } else { 2246 state = G_MIRROR_DISK_STATE_STALE; 2247 } 2248 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 2249 /* 2250 * Not good, NOT GOOD! 2251 * It means that mirror was started on stale disks 2252 * and more fresh disk just arrive. 2253 * If there were writes, mirror is broken, sorry. 2254 * I think the best choice here is don't touch 2255 * this disk and inform the user loudly. 2256 */ 2257 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 2258 "disk (%s) arrives!! It will not be connected to the " 2259 "running device.", sc->sc_name, 2260 g_mirror_get_diskname(disk)); 2261 g_mirror_destroy_disk(disk); 2262 state = G_MIRROR_DISK_STATE_NONE; 2263 /* Return immediately, because disk was destroyed. */ 2264 return (state); 2265 } 2266 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 2267 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 2268 return (state); 2269} 2270 2271/* 2272 * Update device state. 2273 */ 2274static void 2275g_mirror_update_device(struct g_mirror_softc *sc, bool force) 2276{ 2277 struct g_mirror_disk *disk; 2278 u_int state; 2279 2280 sx_assert(&sc->sc_lock, SX_XLOCKED); 2281 2282 switch (sc->sc_state) { 2283 case G_MIRROR_DEVICE_STATE_STARTING: 2284 { 2285 struct g_mirror_disk *pdisk, *tdisk; 2286 u_int dirty, ndisks, genid, syncid; 2287 bool broken; 2288 2289 KASSERT(sc->sc_provider == NULL, 2290 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 2291 /* 2292 * Are we ready? We are, if all disks are connected or 2293 * if we have any disks and 'force' is true. 2294 */ 2295 ndisks = g_mirror_ndisks(sc, -1); 2296 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) { 2297 ; 2298 } else if (ndisks == 0) { 2299 /* 2300 * Disks went down in starting phase, so destroy 2301 * device. 2302 */ 2303 callout_drain(&sc->sc_callout); 2304 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2305 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__, 2306 sc->sc_rootmount); 2307 root_mount_rel(sc->sc_rootmount); 2308 sc->sc_rootmount = NULL; 2309 return; 2310 } else { 2311 return; 2312 } 2313 2314 /* 2315 * Activate all disks with the biggest syncid. 2316 */ 2317 if (force) { 2318 /* 2319 * If 'force' is true, we have been called due to 2320 * timeout, so don't bother canceling timeout. 2321 */ 2322 ndisks = 0; 2323 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2324 if ((disk->d_flags & 2325 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 2326 ndisks++; 2327 } 2328 } 2329 if (ndisks == 0) { 2330 /* No valid disks found, destroy device. */ 2331 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2332 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2333 __LINE__, sc->sc_rootmount); 2334 root_mount_rel(sc->sc_rootmount); 2335 sc->sc_rootmount = NULL; 2336 return; 2337 } 2338 } else { 2339 /* Cancel timeout. */ 2340 callout_drain(&sc->sc_callout); 2341 } 2342 2343 /* 2344 * Find the biggest genid. 2345 */ 2346 genid = 0; 2347 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2348 if (disk->d_genid > genid) 2349 genid = disk->d_genid; 2350 } 2351 sc->sc_genid = genid; 2352 /* 2353 * Remove all disks without the biggest genid. 2354 */ 2355 broken = false; 2356 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { 2357 if (disk->d_genid < genid) { 2358 G_MIRROR_DEBUG(0, 2359 "Component %s (device %s) broken, skipping.", 2360 g_mirror_get_diskname(disk), sc->sc_name); 2361 g_mirror_destroy_disk(disk); 2362 /* 2363 * Bump the syncid in case we discover a healthy 2364 * replacement disk after starting the mirror. 2365 */ 2366 broken = true; 2367 } 2368 } 2369 2370 /* 2371 * Find the biggest syncid. 2372 */ 2373 syncid = 0; 2374 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2375 if (disk->d_sync.ds_syncid > syncid) 2376 syncid = disk->d_sync.ds_syncid; 2377 } 2378 2379 /* 2380 * Here we need to look for dirty disks and if all disks 2381 * with the biggest syncid are dirty, we have to choose 2382 * one with the biggest priority and rebuild the rest. 2383 */ 2384 /* 2385 * Find the number of dirty disks with the biggest syncid. 2386 * Find the number of disks with the biggest syncid. 2387 * While here, find a disk with the biggest priority. 2388 */ 2389 dirty = ndisks = 0; 2390 pdisk = NULL; 2391 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2392 if (disk->d_sync.ds_syncid != syncid) 2393 continue; 2394 if ((disk->d_flags & 2395 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2396 continue; 2397 } 2398 ndisks++; 2399 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 2400 dirty++; 2401 if (pdisk == NULL || 2402 pdisk->d_priority < disk->d_priority) { 2403 pdisk = disk; 2404 } 2405 } 2406 } 2407 if (dirty == 0) { 2408 /* No dirty disks at all, great. */ 2409 } else if (dirty == ndisks) { 2410 /* 2411 * Force synchronization for all dirty disks except one 2412 * with the biggest priority. 2413 */ 2414 KASSERT(pdisk != NULL, ("pdisk == NULL")); 2415 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 2416 "master disk for synchronization.", 2417 g_mirror_get_diskname(pdisk), sc->sc_name); 2418 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2419 if (disk->d_sync.ds_syncid != syncid) 2420 continue; 2421 if ((disk->d_flags & 2422 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2423 continue; 2424 } 2425 KASSERT((disk->d_flags & 2426 G_MIRROR_DISK_FLAG_DIRTY) != 0, 2427 ("Disk %s isn't marked as dirty.", 2428 g_mirror_get_diskname(disk))); 2429 /* Skip the disk with the biggest priority. */ 2430 if (disk == pdisk) 2431 continue; 2432 disk->d_sync.ds_syncid = 0; 2433 } 2434 } else if (dirty < ndisks) { 2435 /* 2436 * Force synchronization for all dirty disks. 2437 * We have some non-dirty disks. 2438 */ 2439 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2440 if (disk->d_sync.ds_syncid != syncid) 2441 continue; 2442 if ((disk->d_flags & 2443 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 2444 continue; 2445 } 2446 if ((disk->d_flags & 2447 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 2448 continue; 2449 } 2450 disk->d_sync.ds_syncid = 0; 2451 } 2452 } 2453 2454 /* Reset hint. */ 2455 sc->sc_hint = NULL; 2456 sc->sc_syncid = syncid; 2457 if (force || broken) { 2458 /* Remember to bump syncid on first write. */ 2459 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2460 } 2461 state = G_MIRROR_DEVICE_STATE_RUNNING; 2462 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 2463 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 2464 g_mirror_device_state2str(state)); 2465 sc->sc_state = state; 2466 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2467 state = g_mirror_determine_state(disk); 2468 g_mirror_event_send(disk, state, 2469 G_MIRROR_EVENT_DONTWAIT); 2470 if (state == G_MIRROR_DISK_STATE_STALE) 2471 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID; 2472 } 2473 break; 2474 } 2475 case G_MIRROR_DEVICE_STATE_RUNNING: 2476 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 && 2477 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2478 /* 2479 * No active disks or no disks at all, 2480 * so destroy device. 2481 */ 2482 if (sc->sc_provider != NULL) 2483 g_mirror_destroy_provider(sc); 2484 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2485 break; 2486 } else if (g_mirror_ndisks(sc, 2487 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 2488 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 2489 /* 2490 * We have active disks, launch provider if it doesn't 2491 * exist. 2492 */ 2493 if (sc->sc_provider == NULL) 2494 g_mirror_launch_provider(sc); 2495 if (sc->sc_rootmount != NULL) { 2496 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", 2497 __LINE__, sc->sc_rootmount); 2498 root_mount_rel(sc->sc_rootmount); 2499 sc->sc_rootmount = NULL; 2500 } 2501 } 2502 /* 2503 * Genid should be bumped immediately, so do it here. 2504 */ 2505 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) { 2506 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID; 2507 g_mirror_bump_genid(sc); 2508 } 2509 break; 2510 default: 2511 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 2512 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 2513 break; 2514 } 2515} 2516 2517/* 2518 * Update disk state and device state if needed. 2519 */ 2520#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 2521 "Disk %s state changed from %s to %s (device %s).", \ 2522 g_mirror_get_diskname(disk), \ 2523 g_mirror_disk_state2str(disk->d_state), \ 2524 g_mirror_disk_state2str(state), sc->sc_name) 2525static int 2526g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 2527{ 2528 struct g_mirror_softc *sc; 2529 2530 sc = disk->d_softc; 2531 sx_assert(&sc->sc_lock, SX_XLOCKED); 2532 2533again: 2534 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 2535 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 2536 g_mirror_disk_state2str(state)); 2537 switch (state) { 2538 case G_MIRROR_DISK_STATE_NEW: 2539 /* 2540 * Possible scenarios: 2541 * 1. New disk arrive. 2542 */ 2543 /* Previous state should be NONE. */ 2544 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 2545 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2546 g_mirror_disk_state2str(disk->d_state))); 2547 DISK_STATE_CHANGED(); 2548 2549 disk->d_state = state; 2550 if (LIST_EMPTY(&sc->sc_disks)) 2551 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 2552 else { 2553 struct g_mirror_disk *dp; 2554 2555 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 2556 if (disk->d_priority >= dp->d_priority) { 2557 LIST_INSERT_BEFORE(dp, disk, d_next); 2558 dp = NULL; 2559 break; 2560 } 2561 if (LIST_NEXT(dp, d_next) == NULL) 2562 break; 2563 } 2564 if (dp != NULL) 2565 LIST_INSERT_AFTER(dp, disk, d_next); 2566 } 2567 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.", 2568 sc->sc_name, g_mirror_get_diskname(disk)); 2569 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2570 break; 2571 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2572 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2573 g_mirror_device_state2str(sc->sc_state), 2574 g_mirror_get_diskname(disk), 2575 g_mirror_disk_state2str(disk->d_state))); 2576 state = g_mirror_determine_state(disk); 2577 if (state != G_MIRROR_DISK_STATE_NONE) 2578 goto again; 2579 break; 2580 case G_MIRROR_DISK_STATE_ACTIVE: 2581 /* 2582 * Possible scenarios: 2583 * 1. New disk does not need synchronization. 2584 * 2. Synchronization process finished successfully. 2585 */ 2586 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2587 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2588 g_mirror_device_state2str(sc->sc_state), 2589 g_mirror_get_diskname(disk), 2590 g_mirror_disk_state2str(disk->d_state))); 2591 /* Previous state should be NEW or SYNCHRONIZING. */ 2592 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2593 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2594 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2595 g_mirror_disk_state2str(disk->d_state))); 2596 DISK_STATE_CHANGED(); 2597 2598 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2599 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2600 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2601 g_mirror_sync_stop(disk, 0); 2602 } 2603 disk->d_state = state; 2604 disk->d_sync.ds_offset = 0; 2605 disk->d_sync.ds_offset_done = 0; 2606 g_mirror_update_idle(sc, disk); 2607 g_mirror_update_metadata(disk); 2608 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.", 2609 sc->sc_name, g_mirror_get_diskname(disk)); 2610 break; 2611 case G_MIRROR_DISK_STATE_STALE: 2612 /* 2613 * Possible scenarios: 2614 * 1. Stale disk was connected. 2615 */ 2616 /* Previous state should be NEW. */ 2617 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2618 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2619 g_mirror_disk_state2str(disk->d_state))); 2620 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2621 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2622 g_mirror_device_state2str(sc->sc_state), 2623 g_mirror_get_diskname(disk), 2624 g_mirror_disk_state2str(disk->d_state))); 2625 /* 2626 * STALE state is only possible if device is marked 2627 * NOAUTOSYNC. 2628 */ 2629 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2630 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2631 g_mirror_device_state2str(sc->sc_state), 2632 g_mirror_get_diskname(disk), 2633 g_mirror_disk_state2str(disk->d_state))); 2634 DISK_STATE_CHANGED(); 2635 2636 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2637 disk->d_state = state; 2638 g_mirror_update_metadata(disk); 2639 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2640 sc->sc_name, g_mirror_get_diskname(disk)); 2641 break; 2642 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2643 /* 2644 * Possible scenarios: 2645 * 1. Disk which needs synchronization was connected. 2646 */ 2647 /* Previous state should be NEW. */ 2648 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2649 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2650 g_mirror_disk_state2str(disk->d_state))); 2651 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2652 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2653 g_mirror_device_state2str(sc->sc_state), 2654 g_mirror_get_diskname(disk), 2655 g_mirror_disk_state2str(disk->d_state))); 2656 DISK_STATE_CHANGED(); 2657 2658 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2659 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2660 disk->d_state = state; 2661 if (sc->sc_provider != NULL) { 2662 g_mirror_sync_start(disk); 2663 g_mirror_update_metadata(disk); 2664 } 2665 break; 2666 case G_MIRROR_DISK_STATE_DISCONNECTED: 2667 /* 2668 * Possible scenarios: 2669 * 1. Device wasn't running yet, but disk disappear. 2670 * 2. Disk was active and disapppear. 2671 * 3. Disk disappear during synchronization process. 2672 */ 2673 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2674 /* 2675 * Previous state should be ACTIVE, STALE or 2676 * SYNCHRONIZING. 2677 */ 2678 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2679 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2680 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2681 ("Wrong disk state (%s, %s).", 2682 g_mirror_get_diskname(disk), 2683 g_mirror_disk_state2str(disk->d_state))); 2684 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2685 /* Previous state should be NEW. */ 2686 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2687 ("Wrong disk state (%s, %s).", 2688 g_mirror_get_diskname(disk), 2689 g_mirror_disk_state2str(disk->d_state))); 2690 /* 2691 * Reset bumping syncid if disk disappeared in STARTING 2692 * state. 2693 */ 2694 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) 2695 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID; 2696#ifdef INVARIANTS 2697 } else { 2698 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2699 sc->sc_name, 2700 g_mirror_device_state2str(sc->sc_state), 2701 g_mirror_get_diskname(disk), 2702 g_mirror_disk_state2str(disk->d_state))); 2703#endif 2704 } 2705 DISK_STATE_CHANGED(); 2706 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2707 sc->sc_name, g_mirror_get_diskname(disk)); 2708 2709 g_mirror_destroy_disk(disk); 2710 break; 2711 case G_MIRROR_DISK_STATE_DESTROY: 2712 { 2713 int error; 2714 2715 error = g_mirror_clear_metadata(disk); 2716 if (error != 0) { 2717 G_MIRROR_DEBUG(0, 2718 "Device %s: failed to clear metadata on %s: %d.", 2719 sc->sc_name, g_mirror_get_diskname(disk), error); 2720 break; 2721 } 2722 DISK_STATE_CHANGED(); 2723 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2724 sc->sc_name, g_mirror_get_diskname(disk)); 2725 2726 g_mirror_destroy_disk(disk); 2727 sc->sc_ndisks--; 2728 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2729 g_mirror_update_metadata(disk); 2730 } 2731 break; 2732 } 2733 default: 2734 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2735 break; 2736 } 2737 return (0); 2738} 2739#undef DISK_STATE_CHANGED 2740 2741int 2742g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2743{ 2744 struct g_provider *pp; 2745 u_char *buf; 2746 int error; 2747 2748 g_topology_assert(); 2749 2750 error = g_access(cp, 1, 0, 0); 2751 if (error != 0) 2752 return (error); 2753 pp = cp->provider; 2754 g_topology_unlock(); 2755 /* Metadata are stored on last sector. */ 2756 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2757 &error); 2758 g_topology_lock(); 2759 g_access(cp, -1, 0, 0); 2760 if (buf == NULL) { 2761 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).", 2762 cp->provider->name, error); 2763 return (error); 2764 } 2765 2766 /* Decode metadata. */ 2767 error = mirror_metadata_decode(buf, md); 2768 g_free(buf); 2769 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2770 return (EINVAL); 2771 if (md->md_version > G_MIRROR_VERSION) { 2772 G_MIRROR_DEBUG(0, 2773 "Kernel module is too old to handle metadata from %s.", 2774 cp->provider->name); 2775 return (EINVAL); 2776 } 2777 if (error != 0) { 2778 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2779 cp->provider->name); 2780 return (error); 2781 } 2782 2783 return (0); 2784} 2785 2786static int 2787g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2788 struct g_mirror_metadata *md) 2789{ 2790 2791 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2792 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2793 pp->name, md->md_did); 2794 return (EEXIST); 2795 } 2796 if (md->md_all != sc->sc_ndisks) { 2797 G_MIRROR_DEBUG(1, 2798 "Invalid '%s' field on disk %s (device %s), skipping.", 2799 "md_all", pp->name, sc->sc_name); 2800 return (EINVAL); 2801 } 2802 if (md->md_slice != sc->sc_slice) { 2803 G_MIRROR_DEBUG(1, 2804 "Invalid '%s' field on disk %s (device %s), skipping.", 2805 "md_slice", pp->name, sc->sc_name); 2806 return (EINVAL); 2807 } 2808 if (md->md_balance != sc->sc_balance) { 2809 G_MIRROR_DEBUG(1, 2810 "Invalid '%s' field on disk %s (device %s), skipping.", 2811 "md_balance", pp->name, sc->sc_name); 2812 return (EINVAL); 2813 } 2814#if 0 2815 if (md->md_mediasize != sc->sc_mediasize) { 2816 G_MIRROR_DEBUG(1, 2817 "Invalid '%s' field on disk %s (device %s), skipping.", 2818 "md_mediasize", pp->name, sc->sc_name); 2819 return (EINVAL); 2820 } 2821#endif 2822 if (sc->sc_mediasize > pp->mediasize) { 2823 G_MIRROR_DEBUG(1, 2824 "Invalid size of disk %s (device %s), skipping.", pp->name, 2825 sc->sc_name); 2826 return (EINVAL); 2827 } 2828 if (md->md_sectorsize != sc->sc_sectorsize) { 2829 G_MIRROR_DEBUG(1, 2830 "Invalid '%s' field on disk %s (device %s), skipping.", 2831 "md_sectorsize", pp->name, sc->sc_name); 2832 return (EINVAL); 2833 } 2834 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2835 G_MIRROR_DEBUG(1, 2836 "Invalid sector size of disk %s (device %s), skipping.", 2837 pp->name, sc->sc_name); 2838 return (EINVAL); 2839 } 2840 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2841 G_MIRROR_DEBUG(1, 2842 "Invalid device flags on disk %s (device %s), skipping.", 2843 pp->name, sc->sc_name); 2844 return (EINVAL); 2845 } 2846 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2847 G_MIRROR_DEBUG(1, 2848 "Invalid disk flags on disk %s (device %s), skipping.", 2849 pp->name, sc->sc_name); 2850 return (EINVAL); 2851 } 2852 return (0); 2853} 2854 2855int 2856g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2857 struct g_mirror_metadata *md) 2858{ 2859 struct g_mirror_disk *disk; 2860 int error; 2861 2862 g_topology_assert_not(); 2863 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2864 2865 error = g_mirror_check_metadata(sc, pp, md); 2866 if (error != 0) 2867 return (error); 2868 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING && 2869 md->md_genid < sc->sc_genid) { 2870 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.", 2871 pp->name, sc->sc_name); 2872 return (EINVAL); 2873 } 2874 disk = g_mirror_init_disk(sc, pp, md, &error); 2875 if (disk == NULL) 2876 return (error); 2877 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2878 G_MIRROR_EVENT_WAIT); 2879 if (error != 0) 2880 return (error); 2881 if (md->md_version < G_MIRROR_VERSION) { 2882 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).", 2883 pp->name, md->md_version, G_MIRROR_VERSION); 2884 g_mirror_update_metadata(disk); 2885 } 2886 return (0); 2887} 2888 2889static void 2890g_mirror_destroy_delayed(void *arg, int flag) 2891{ 2892 struct g_mirror_softc *sc; 2893 int error; 2894 2895 if (flag == EV_CANCEL) { 2896 G_MIRROR_DEBUG(1, "Destroying canceled."); 2897 return; 2898 } 2899 sc = arg; 2900 g_topology_unlock(); 2901 sx_xlock(&sc->sc_lock); 2902 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0, 2903 ("DESTROY flag set on %s.", sc->sc_name)); 2904 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0, 2905 ("DESTROYING flag not set on %s.", sc->sc_name)); 2906 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name); 2907 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT); 2908 if (error != 0) { 2909 G_MIRROR_DEBUG(0, "Cannot destroy %s (error=%d).", 2910 sc->sc_name, error); 2911 sx_xunlock(&sc->sc_lock); 2912 } 2913 g_topology_lock(); 2914} 2915 2916static int 2917g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2918{ 2919 struct g_mirror_softc *sc; 2920 int error = 0; 2921 2922 g_topology_assert(); 2923 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2924 acw, ace); 2925 2926 sc = pp->private; 2927 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 2928 2929 g_topology_unlock(); 2930 sx_xlock(&sc->sc_lock); 2931 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 || 2932 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0 || 2933 LIST_EMPTY(&sc->sc_disks)) { 2934 if (acr > 0 || acw > 0 || ace > 0) 2935 error = ENXIO; 2936 goto end; 2937 } 2938 sc->sc_provider_open += acr + acw + ace; 2939 if (pp->acw + acw == 0) 2940 g_mirror_idle(sc, 0); 2941 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0 && 2942 sc->sc_provider_open == 0) 2943 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK, sc, NULL); 2944end: 2945 sx_xunlock(&sc->sc_lock); 2946 g_topology_lock(); 2947 return (error); 2948} 2949 2950struct g_geom * 2951g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md, 2952 u_int type) 2953{ 2954 struct g_mirror_softc *sc; 2955 struct g_geom *gp; 2956 int error, timeout; 2957 2958 g_topology_assert(); 2959 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2960 md->md_mid); 2961 2962 /* One disk is minimum. */ 2963 if (md->md_all < 1) 2964 return (NULL); 2965 /* 2966 * Action geom. 2967 */ 2968 gp = g_new_geomf(mp, "%s", md->md_name); 2969 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2970 gp->start = g_mirror_start; 2971 gp->orphan = g_mirror_orphan; 2972 gp->access = g_mirror_access; 2973 gp->dumpconf = g_mirror_dumpconf; 2974 2975 sc->sc_type = type; 2976 sc->sc_id = md->md_mid; 2977 sc->sc_slice = md->md_slice; 2978 sc->sc_balance = md->md_balance; 2979 sc->sc_mediasize = md->md_mediasize; 2980 sc->sc_sectorsize = md->md_sectorsize; 2981 sc->sc_ndisks = md->md_all; 2982 sc->sc_flags = md->md_mflags; 2983 sc->sc_bump_id = 0; 2984 sc->sc_idle = 1; 2985 sc->sc_last_write = time_uptime; 2986 sc->sc_writes = 0; 2987 sc->sc_refcnt = 1; 2988 sx_init(&sc->sc_lock, "gmirror:lock"); 2989 bioq_init(&sc->sc_queue); 2990 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2991 bioq_init(&sc->sc_regular_delayed); 2992 bioq_init(&sc->sc_inflight); 2993 bioq_init(&sc->sc_sync_delayed); 2994 LIST_INIT(&sc->sc_disks); 2995 TAILQ_INIT(&sc->sc_events); 2996 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2997 callout_init(&sc->sc_callout, 1); 2998 mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF); 2999 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 3000 gp->softc = sc; 3001 sc->sc_geom = gp; 3002 sc->sc_provider = NULL; 3003 sc->sc_provider_open = 0; 3004 /* 3005 * Synchronization geom. 3006 */ 3007 gp = g_new_geomf(mp, "%s.sync", md->md_name); 3008 gp->softc = sc; 3009 gp->orphan = g_mirror_orphan; 3010 sc->sc_sync.ds_geom = gp; 3011 sc->sc_sync.ds_ndisks = 0; 3012 error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 3013 "g_mirror %s", md->md_name); 3014 if (error != 0) { 3015 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 3016 sc->sc_name); 3017 g_destroy_geom(sc->sc_sync.ds_geom); 3018 g_destroy_geom(sc->sc_geom); 3019 g_mirror_free_device(sc); 3020 return (NULL); 3021 } 3022 3023 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).", 3024 sc->sc_name, sc->sc_ndisks, sc->sc_id); 3025 3026 sc->sc_rootmount = root_mount_hold("GMIRROR"); 3027 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount); 3028 /* 3029 * Run timeout. 3030 */ 3031 timeout = g_mirror_timeout * hz; 3032 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc); 3033 return (sc->sc_geom); 3034} 3035 3036int 3037g_mirror_destroy(struct g_mirror_softc *sc, int how) 3038{ 3039 struct g_mirror_disk *disk; 3040 3041 g_topology_assert_not(); 3042 sx_assert(&sc->sc_lock, SX_XLOCKED); 3043 3044 if (sc->sc_provider_open != 0) { 3045 switch (how) { 3046 case G_MIRROR_DESTROY_SOFT: 3047 G_MIRROR_DEBUG(1, 3048 "Device %s is still open (%d).", sc->sc_name, 3049 sc->sc_provider_open); 3050 return (EBUSY); 3051 case G_MIRROR_DESTROY_DELAYED: 3052 G_MIRROR_DEBUG(1, 3053 "Device %s will be destroyed on last close.", 3054 sc->sc_name); 3055 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 3056 if (disk->d_state == 3057 G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3058 g_mirror_sync_stop(disk, 1); 3059 } 3060 } 3061 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROYING; 3062 return (EBUSY); 3063 case G_MIRROR_DESTROY_HARD: 3064 G_MIRROR_DEBUG(1, "Device %s is still open, so it " 3065 "can't be definitely removed.", sc->sc_name); 3066 } 3067 } 3068 3069 g_topology_lock(); 3070 if (sc->sc_geom->softc == NULL) { 3071 g_topology_unlock(); 3072 return (0); 3073 } 3074 sc->sc_geom->softc = NULL; 3075 sc->sc_sync.ds_geom->softc = NULL; 3076 g_topology_unlock(); 3077 3078 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 3079 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 3080 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 3081 sx_xunlock(&sc->sc_lock); 3082 mtx_lock(&sc->sc_queue_mtx); 3083 wakeup(sc); 3084 mtx_unlock(&sc->sc_queue_mtx); 3085 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 3086 while (sc->sc_worker != NULL) 3087 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 3088 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 3089 sx_xlock(&sc->sc_lock); 3090 g_mirror_destroy_device(sc); 3091 return (0); 3092} 3093 3094static void 3095g_mirror_taste_orphan(struct g_consumer *cp) 3096{ 3097 3098 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 3099 cp->provider->name)); 3100} 3101 3102static struct g_geom * 3103g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 3104{ 3105 struct g_mirror_metadata md; 3106 struct g_mirror_softc *sc; 3107 struct g_consumer *cp; 3108 struct g_geom *gp; 3109 int error; 3110 3111 g_topology_assert(); 3112 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 3113 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 3114 3115 gp = g_new_geomf(mp, "mirror:taste"); 3116 /* 3117 * This orphan function should be never called. 3118 */ 3119 gp->orphan = g_mirror_taste_orphan; 3120 cp = g_new_consumer(gp); 3121 g_attach(cp, pp); 3122 error = g_mirror_read_metadata(cp, &md); 3123 g_detach(cp); 3124 g_destroy_consumer(cp); 3125 g_destroy_geom(gp); 3126 if (error != 0) 3127 return (NULL); 3128 gp = NULL; 3129 3130 if (md.md_provider[0] != '\0' && 3131 !g_compare_names(md.md_provider, pp->name)) 3132 return (NULL); 3133 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) 3134 return (NULL); 3135 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 3136 G_MIRROR_DEBUG(0, 3137 "Device %s: provider %s marked as inactive, skipping.", 3138 md.md_name, pp->name); 3139 return (NULL); 3140 } 3141 if (g_mirror_debug >= 2) 3142 mirror_metadata_dump(&md); 3143 3144 /* 3145 * Let's check if device already exists. 3146 */ 3147 sc = NULL; 3148 LIST_FOREACH(gp, &mp->geom, geom) { 3149 sc = gp->softc; 3150 if (sc == NULL) 3151 continue; 3152 if (sc->sc_type != G_MIRROR_TYPE_AUTOMATIC) 3153 continue; 3154 if (sc->sc_sync.ds_geom == gp) 3155 continue; 3156 if (strcmp(md.md_name, sc->sc_name) != 0) 3157 continue; 3158 if (md.md_mid != sc->sc_id) { 3159 G_MIRROR_DEBUG(0, "Device %s already configured.", 3160 sc->sc_name); 3161 return (NULL); 3162 } 3163 break; 3164 } 3165 if (gp == NULL) { 3166 gp = g_mirror_create(mp, &md, G_MIRROR_TYPE_AUTOMATIC); 3167 if (gp == NULL) { 3168 G_MIRROR_DEBUG(0, "Cannot create device %s.", 3169 md.md_name); 3170 return (NULL); 3171 } 3172 sc = gp->softc; 3173 } 3174 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 3175 g_topology_unlock(); 3176 sx_xlock(&sc->sc_lock); 3177 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING; 3178 error = g_mirror_add_disk(sc, pp, &md); 3179 if (error != 0) { 3180 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 3181 pp->name, gp->name, error); 3182 if (LIST_EMPTY(&sc->sc_disks)) { 3183 g_cancel_event(sc); 3184 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3185 g_topology_lock(); 3186 return (NULL); 3187 } 3188 gp = NULL; 3189 } 3190 sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING; 3191 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 3192 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD); 3193 g_topology_lock(); 3194 return (NULL); 3195 } 3196 sx_xunlock(&sc->sc_lock); 3197 g_topology_lock(); 3198 return (gp); 3199} 3200 3201static void 3202g_mirror_resize(struct g_consumer *cp) 3203{ 3204 struct g_mirror_disk *disk; 3205 3206 g_topology_assert(); 3207 g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name); 3208 3209 disk = cp->private; 3210 if (disk == NULL) 3211 return; 3212 g_topology_unlock(); 3213 g_mirror_update_metadata(disk); 3214 g_topology_lock(); 3215} 3216 3217static int 3218g_mirror_destroy_geom(struct gctl_req *req __unused, 3219 struct g_class *mp __unused, struct g_geom *gp) 3220{ 3221 struct g_mirror_softc *sc; 3222 int error; 3223 3224 g_topology_unlock(); 3225 sc = gp->softc; 3226 sx_xlock(&sc->sc_lock); 3227 g_cancel_event(sc); 3228 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT); 3229 if (error != 0) 3230 sx_xunlock(&sc->sc_lock); 3231 g_topology_lock(); 3232 return (error); 3233} 3234 3235static void 3236g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 3237 struct g_consumer *cp, struct g_provider *pp) 3238{ 3239 struct g_mirror_softc *sc; 3240 3241 g_topology_assert(); 3242 3243 sc = gp->softc; 3244 if (sc == NULL) 3245 return; 3246 /* Skip synchronization geom. */ 3247 if (gp == sc->sc_sync.ds_geom) 3248 return; 3249 if (pp != NULL) { 3250 /* Nothing here. */ 3251 } else if (cp != NULL) { 3252 struct g_mirror_disk *disk; 3253 3254 disk = cp->private; 3255 if (disk == NULL) 3256 return; 3257 g_topology_unlock(); 3258 sx_xlock(&sc->sc_lock); 3259 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 3260 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 3261 sbuf_printf(sb, "%s<Synchronized>", indent); 3262 if (disk->d_sync.ds_offset == 0) 3263 sbuf_printf(sb, "0%%"); 3264 else { 3265 sbuf_printf(sb, "%u%%", 3266 (u_int)((disk->d_sync.ds_offset * 100) / 3267 sc->sc_provider->mediasize)); 3268 } 3269 sbuf_printf(sb, "</Synchronized>\n"); 3270 if (disk->d_sync.ds_offset > 0) { 3271 sbuf_printf(sb, "%s<BytesSynced>%jd" 3272 "</BytesSynced>\n", indent, 3273 (intmax_t)disk->d_sync.ds_offset); 3274 } 3275 } 3276 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 3277 disk->d_sync.ds_syncid); 3278 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, 3279 disk->d_genid); 3280 sbuf_printf(sb, "%s<Flags>", indent); 3281 if (disk->d_flags == 0) 3282 sbuf_printf(sb, "NONE"); 3283 else { 3284 int first = 1; 3285 3286#define ADD_FLAG(flag, name) do { \ 3287 if ((disk->d_flags & (flag)) != 0) { \ 3288 if (!first) \ 3289 sbuf_printf(sb, ", "); \ 3290 else \ 3291 first = 0; \ 3292 sbuf_printf(sb, name); \ 3293 } \ 3294} while (0) 3295 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 3296 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 3297 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 3298 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 3299 "SYNCHRONIZING"); 3300 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 3301 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN"); 3302#undef ADD_FLAG 3303 } 3304 sbuf_printf(sb, "</Flags>\n"); 3305 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 3306 disk->d_priority); 3307 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 3308 g_mirror_disk_state2str(disk->d_state)); 3309 sx_xunlock(&sc->sc_lock); 3310 g_topology_lock(); 3311 } else { 3312 g_topology_unlock(); 3313 sx_xlock(&sc->sc_lock); 3314 sbuf_printf(sb, "%s<Type>", indent); 3315 switch (sc->sc_type) { 3316 case G_MIRROR_TYPE_AUTOMATIC: 3317 sbuf_printf(sb, "AUTOMATIC"); 3318 break; 3319 case G_MIRROR_TYPE_MANUAL: 3320 sbuf_printf(sb, "MANUAL"); 3321 break; 3322 default: 3323 sbuf_printf(sb, "UNKNOWN"); 3324 break; 3325 } 3326 sbuf_printf(sb, "</Type>\n"); 3327 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 3328 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 3329 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid); 3330 sbuf_printf(sb, "%s<Flags>", indent); 3331 if (sc->sc_flags == 0) 3332 sbuf_printf(sb, "NONE"); 3333 else { 3334 int first = 1; 3335 3336#define ADD_FLAG(flag, name) do { \ 3337 if ((sc->sc_flags & (flag)) != 0) { \ 3338 if (!first) \ 3339 sbuf_printf(sb, ", "); \ 3340 else \ 3341 first = 0; \ 3342 sbuf_printf(sb, name); \ 3343 } \ 3344} while (0) 3345 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC"); 3346 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 3347#undef ADD_FLAG 3348 } 3349 sbuf_printf(sb, "</Flags>\n"); 3350 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 3351 (u_int)sc->sc_slice); 3352 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 3353 balance_name(sc->sc_balance)); 3354 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 3355 sc->sc_ndisks); 3356 sbuf_printf(sb, "%s<State>", indent); 3357 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 3358 sbuf_printf(sb, "%s", "STARTING"); 3359 else if (sc->sc_ndisks == 3360 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 3361 sbuf_printf(sb, "%s", "COMPLETE"); 3362 else 3363 sbuf_printf(sb, "%s", "DEGRADED"); 3364 sbuf_printf(sb, "</State>\n"); 3365 sx_xunlock(&sc->sc_lock); 3366 g_topology_lock(); 3367 } 3368} 3369 3370static void 3371g_mirror_shutdown_post_sync(void *arg, int howto) 3372{ 3373 struct g_class *mp; 3374 struct g_geom *gp, *gp2; 3375 struct g_mirror_softc *sc; 3376 int error; 3377 3378 if (panicstr != NULL) 3379 return; 3380 3381 mp = arg; 3382 g_topology_lock(); 3383 g_mirror_shutdown = 1; 3384 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 3385 if ((sc = gp->softc) == NULL) 3386 continue; 3387 /* Skip synchronization geom. */ 3388 if (gp == sc->sc_sync.ds_geom) 3389 continue; 3390 g_topology_unlock(); 3391 sx_xlock(&sc->sc_lock); 3392 g_mirror_idle(sc, -1); 3393 g_cancel_event(sc); 3394 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED); 3395 if (error != 0) 3396 sx_xunlock(&sc->sc_lock); 3397 g_topology_lock(); 3398 } 3399 g_topology_unlock(); 3400} 3401 3402static void 3403g_mirror_init(struct g_class *mp) 3404{ 3405 3406 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 3407 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 3408 if (g_mirror_post_sync == NULL) 3409 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event."); 3410} 3411 3412static void 3413g_mirror_fini(struct g_class *mp) 3414{ 3415 3416 if (g_mirror_post_sync != NULL) 3417 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync); 3418} 3419 3420DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 3421