g_mirror.c revision 136504
121308Sache/*- 221308Sache * Copyright (c) 2004 Pawel Jakub Dawidek <pjd@FreeBSD.org> 321308Sache * All rights reserved. 421308Sache * 521308Sache * Redistribution and use in source and binary forms, with or without 621308Sache * modification, are permitted provided that the following conditions 721308Sache * are met: 821308Sache * 1. Redistributions of source code must retain the above copyright 921308Sache * notice, this list of conditions and the following disclaimer. 1021308Sache * 2. Redistributions in binary form must reproduce the above copyright 1158310Sache * notice, this list of conditions and the following disclaimer in the 1221308Sache * documentation and/or other materials provided with the distribution. 1321308Sache * 1421308Sache * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 1521308Sache * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 1621308Sache * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 1721308Sache * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 1821308Sache * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 1921308Sache * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2021308Sache * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2121308Sache * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2258310Sache * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 2321308Sache * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 2421308Sache * SUCH DAMAGE. 2521308Sache */ 2621308Sache 2721308Sache#include <sys/cdefs.h> 2821308Sache__FBSDID("$FreeBSD: head/sys/geom/mirror/g_mirror.c 136504 2004-10-14 07:55:29Z pjd $"); 2921308Sache 3021308Sache#include <sys/param.h> 3121308Sache#include <sys/systm.h> 3221308Sache#include <sys/kernel.h> 3321308Sache#include <sys/module.h> 3421308Sache#include <sys/limits.h> 3521308Sache#include <sys/lock.h> 3621308Sache#include <sys/mutex.h> 3721308Sache#include <sys/bio.h> 3821308Sache#include <sys/sysctl.h> 3921308Sache#include <sys/malloc.h> 4021308Sache#include <sys/bitstring.h> 4121308Sache#include <vm/uma.h> 4221308Sache#include <machine/atomic.h> 4321308Sache#include <geom/geom.h> 4421308Sache#include <sys/proc.h> 4521308Sache#include <sys/kthread.h> 4621308Sache#include <geom/mirror/g_mirror.h> 4721308Sache 4821308Sache 4921308Sachestatic MALLOC_DEFINE(M_MIRROR, "mirror data", "GEOM_MIRROR Data"); 5021308Sache 5121308SacheSYSCTL_DECL(_kern_geom); 5221308SacheSYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, "GEOM_MIRROR stuff"); 5321308Sacheu_int g_mirror_debug = 0; 54119610SacheTUNABLE_INT("kern.geom.mirror.debug", &g_mirror_debug); 55119610SacheSYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RW, &g_mirror_debug, 0, 5621308Sache "Debug level"); 5721308Sachestatic u_int g_mirror_timeout = 4; 5821308SacheTUNABLE_INT("kern.geom.mirror.timeout", &g_mirror_timeout); 5958310SacheSYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RW, &g_mirror_timeout, 6058310Sache 0, "Time to wait on all mirror components"); 6158310Sachestatic u_int g_mirror_reqs_per_sync = 5; 6221308SacheSYSCTL_UINT(_kern_geom_mirror, OID_AUTO, reqs_per_sync, CTLFLAG_RW, 6321308Sache &g_mirror_reqs_per_sync, 0, 6421308Sache "Number of regular I/O requests per synchronization request"); 6521308Sachestatic u_int g_mirror_syncs_per_sec = 100; 6621308SacheSYSCTL_UINT(_kern_geom_mirror, OID_AUTO, syncs_per_sec, CTLFLAG_RW, 6721308Sache &g_mirror_syncs_per_sec, 0, 6821308Sache "Number of synchronizations requests per second"); 6921308Sache 70119610Sache#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \ 7121308Sache G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 7221308Sache msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 7321308Sache G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 7421308Sache} while (0) 7521308Sache 7621308Sache 7721308Sachestatic int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp, 7821308Sache struct g_geom *gp); 7921308Sachestatic g_taste_t g_mirror_taste; 8021308Sache 8121308Sachestruct g_class g_mirror_class = { 8221308Sache .name = G_MIRROR_CLASS_NAME, 8321308Sache .version = G_VERSION, 8421308Sache .ctlreq = g_mirror_config, 8521308Sache .taste = g_mirror_taste, 8621308Sache .destroy_geom = g_mirror_destroy_geom 8721308Sache}; 8821308Sache 8921308Sache 90119610Sachestatic void g_mirror_destroy_provider(struct g_mirror_softc *sc); 91119610Sachestatic int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state); 92119610Sachestatic void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force); 9321308Sachestatic void g_mirror_dumpconf(struct sbuf *sb, const char *indent, 94119610Sache struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 9521308Sachestatic void g_mirror_sync_stop(struct g_mirror_disk *disk, int type); 9621308Sache 9721308Sache 9821308Sachestatic const char * 9921308Sacheg_mirror_disk_state2str(int state) 10021308Sache{ 10121308Sache 10275406Sache switch (state) { 10321308Sache case G_MIRROR_DISK_STATE_NONE: 10421308Sache return ("NONE"); 105119610Sache case G_MIRROR_DISK_STATE_NEW: 10621308Sache return ("NEW"); 107119610Sache case G_MIRROR_DISK_STATE_ACTIVE: 108119610Sache return ("ACTIVE"); 109119610Sache case G_MIRROR_DISK_STATE_STALE: 11021308Sache return ("STALE"); 11121308Sache case G_MIRROR_DISK_STATE_SYNCHRONIZING: 11221308Sache return ("SYNCHRONIZING"); 11321308Sache case G_MIRROR_DISK_STATE_DISCONNECTED: 11421308Sache return ("DISCONNECTED"); 11521308Sache case G_MIRROR_DISK_STATE_DESTROY: 11621308Sache return ("DESTROY"); 11721308Sache default: 11821308Sache return ("INVALID"); 11921308Sache } 12021308Sache} 12121308Sache 12221308Sachestatic const char * 12321308Sacheg_mirror_device_state2str(int state) 12421308Sache{ 12521308Sache 12621308Sache switch (state) { 12721308Sache case G_MIRROR_DEVICE_STATE_STARTING: 12821308Sache return ("STARTING"); 12921308Sache case G_MIRROR_DEVICE_STATE_RUNNING: 13021308Sache return ("RUNNING"); 13121308Sache default: 13221308Sache return ("INVALID"); 13321308Sache } 13421308Sache} 13521308Sache 13621308Sachestatic const char * 13721308Sacheg_mirror_get_diskname(struct g_mirror_disk *disk) 13821308Sache{ 13921308Sache 14021308Sache if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 14121308Sache return ("[unknown]"); 14221308Sache return (disk->d_name); 14321308Sache} 14421308Sache 14521308Sache/* 14621308Sache * --- Events handling functions --- 14721308Sache * Events in geom_mirror are used to maintain disks and device status 14821308Sache * from one thread to simplify locking. 14921308Sache */ 15021308Sachestatic void 15121308Sacheg_mirror_event_free(struct g_mirror_event *ep) 15221308Sache{ 15321308Sache 15421308Sache free(ep, M_MIRROR); 15521308Sache} 15621308Sache 15721308Sacheint 15821308Sacheg_mirror_event_send(void *arg, int state, int flags) 15921308Sache{ 16021308Sache struct g_mirror_softc *sc; 16121308Sache struct g_mirror_disk *disk; 16221308Sache struct g_mirror_event *ep; 163119610Sache int error; 164119610Sache 16521308Sache ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK); 16621308Sache G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep); 16721308Sache if ((flags & G_MIRROR_EVENT_DEVICE) != 0) { 16821308Sache disk = NULL; 16921308Sache sc = arg; 17021308Sache } else { 171119610Sache disk = arg; 17221308Sache sc = disk->d_softc; 17321308Sache } 17421308Sache ep->e_disk = disk; 17521308Sache ep->e_state = state; 17621308Sache ep->e_flags = flags; 17721308Sache ep->e_error = 0; 17821308Sache mtx_lock(&sc->sc_events_mtx); 17921308Sache TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 18021308Sache mtx_unlock(&sc->sc_events_mtx); 18121308Sache G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 18221308Sache mtx_lock(&sc->sc_queue_mtx); 183119610Sache wakeup(sc); 18421308Sache mtx_unlock(&sc->sc_queue_mtx); 18521308Sache if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0) 186119610Sache return (0); 18721308Sache g_topology_assert(); 18821308Sache G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep); 18921308Sache g_topology_unlock(); 19021308Sache while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) { 19121308Sache mtx_lock(&sc->sc_events_mtx); 19221308Sache MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event", 19321308Sache hz * 5); 19421308Sache } 19521308Sache /* Don't even try to use 'sc' here, because it could be already dead. */ 19621308Sache g_topology_lock(); 19721308Sache error = ep->e_error; 19821308Sache g_mirror_event_free(ep); 19921308Sache return (error); 20021308Sache} 20121308Sache 20221308Sachestatic struct g_mirror_event * 20321308Sacheg_mirror_event_get(struct g_mirror_softc *sc) 20421308Sache{ 20521308Sache struct g_mirror_event *ep; 20621308Sache 20721308Sache mtx_lock(&sc->sc_events_mtx); 20821308Sache ep = TAILQ_FIRST(&sc->sc_events); 20921308Sache if (ep != NULL) 21021308Sache TAILQ_REMOVE(&sc->sc_events, ep, e_next); 21121308Sache mtx_unlock(&sc->sc_events_mtx); 21221308Sache return (ep); 21321308Sache} 21421308Sache 21521308Sachestatic void 21621308Sacheg_mirror_event_cancel(struct g_mirror_disk *disk) 21721308Sache{ 21821308Sache struct g_mirror_softc *sc; 21921308Sache struct g_mirror_event *ep, *tmpep; 22021308Sache 22121308Sache g_topology_assert(); 22221308Sache 22321308Sache sc = disk->d_softc; 22421308Sache mtx_lock(&sc->sc_events_mtx); 22521308Sache TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 22621308Sache if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) 22721308Sache continue; 22821308Sache if (ep->e_disk != disk) 22921308Sache continue; 23021308Sache TAILQ_REMOVE(&sc->sc_events, ep, e_next); 23121308Sache if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 23221308Sache g_mirror_event_free(ep); 23321308Sache else { 23421308Sache ep->e_error = ECANCELED; 23521308Sache wakeup(ep); 23621308Sache } 23721308Sache } 23821308Sache mtx_unlock(&sc->sc_events_mtx); 23921308Sache} 24021308Sache 24121308Sache/* 24221308Sache * Return the number of disks in given state. 24321308Sache * If state is equal to -1, count all connected disks. 24421308Sache */ 24521308Sacheu_int 24621308Sacheg_mirror_ndisks(struct g_mirror_softc *sc, int state) 24721308Sache{ 24821308Sache struct g_mirror_disk *disk; 24921308Sache u_int n = 0; 25021308Sache 25121308Sache LIST_FOREACH(disk, &sc->sc_disks, d_next) { 25221308Sache if (state == -1 || disk->d_state == state) 25321308Sache n++; 25421308Sache } 25521308Sache return (n); 25621308Sache} 25721308Sache 25821308Sache/* 25921308Sache * Find a disk in mirror by its disk ID. 26021308Sache */ 26121308Sachestatic struct g_mirror_disk * 26221308Sacheg_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id) 26321308Sache{ 26421308Sache struct g_mirror_disk *disk; 26521308Sache 26621308Sache g_topology_assert(); 26721308Sache 26821308Sache LIST_FOREACH(disk, &sc->sc_disks, d_next) { 26921308Sache if (disk->d_id == id) 27021308Sache return (disk); 27121308Sache } 27275406Sache return (NULL); 27321308Sache} 27421308Sache 27521308Sachestatic u_int 27621308Sacheg_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp) 27721308Sache{ 27821308Sache struct bio *bp; 27921308Sache u_int nreqs = 0; 28021308Sache 28121308Sache mtx_lock(&sc->sc_queue_mtx); 28221308Sache TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 28321308Sache if (bp->bio_from == cp) 28421308Sache nreqs++; 28521308Sache } 28621308Sache mtx_unlock(&sc->sc_queue_mtx); 28721308Sache return (nreqs); 28821308Sache} 28921308Sache 29021308Sachestatic int 29121308Sacheg_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp) 29221308Sache{ 29321308Sache 29421308Sache if (cp->nstart != cp->nend) { 29521308Sache G_MIRROR_DEBUG(2, 29621308Sache "I/O requests for %s exist, can't destroy it now.", 29721308Sache cp->provider->name); 29821308Sache return (1); 29921308Sache } 30021308Sache if (g_mirror_nrequests(sc, cp) > 0) { 30121308Sache G_MIRROR_DEBUG(2, 30221308Sache "I/O requests for %s in queue, can't destroy it now.", 30321308Sache cp->provider->name); 30421308Sache return (1); 30521308Sache } 30621308Sache return (0); 30721308Sache} 30821308Sache 30921308Sachestatic void 31021308Sacheg_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 31121308Sache{ 31221308Sache 31321308Sache g_topology_assert(); 31421308Sache 31521308Sache cp->private = NULL; 31621308Sache if (g_mirror_is_busy(sc, cp)) 31721308Sache return; 31821308Sache G_MIRROR_DEBUG(2, "Consumer %s destroyed.", cp->provider->name); 31921308Sache g_detach(cp); 32021308Sache g_destroy_consumer(cp); 32121308Sache} 32221308Sache 32321308Sachestatic int 32421308Sacheg_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp) 32521308Sache{ 32621308Sache int error; 32721308Sache 32875406Sache g_topology_assert(); 32921308Sache KASSERT(disk->d_consumer == NULL, 33021308Sache ("Disk already connected (device %s).", disk->d_softc->sc_name)); 33121308Sache 33221308Sache disk->d_consumer = g_new_consumer(disk->d_softc->sc_geom); 33347558Sache disk->d_consumer->private = disk; 33421308Sache error = g_attach(disk->d_consumer, pp); 33547558Sache if (error != 0) 33621308Sache return (error); 33721308Sache G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk)); 33821308Sache return (0); 33921308Sache} 34021308Sache 34121308Sachestatic void 34221308Sacheg_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp) 34321308Sache{ 34421308Sache 34521308Sache g_topology_assert(); 34621308Sache 34721308Sache if (cp == NULL) 34821308Sache return; 34921308Sache if (cp->provider != NULL) { 35075406Sache G_MIRROR_DEBUG(2, "Disk %s disconnected.", cp->provider->name); 35121308Sache if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) { 35221308Sache G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 35321308Sache cp->provider->name, -cp->acr, -cp->acw, -cp->ace, 35421308Sache 0); 35547558Sache g_access(cp, -cp->acr, -cp->acw, -cp->ace); 35621308Sache } 35747558Sache g_mirror_kill_consumer(sc, cp); 35821308Sache } else { 35921308Sache g_destroy_consumer(cp); 36021308Sache } 36121308Sache} 36221308Sache 36321308Sache/* 36421308Sache * Initialize disk. This means allocate memory, create consumer, attach it 36521308Sache * to the provider and open access (r1w1e1) to it. 36621308Sache */ 36721308Sachestatic struct g_mirror_disk * 36875406Sacheg_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp, 36921308Sache struct g_mirror_metadata *md, int *errorp) 37021308Sache{ 37121308Sache struct g_mirror_disk *disk; 37221308Sache int error; 37347558Sache 37421308Sache disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO); 37547558Sache if (disk == NULL) { 37621308Sache error = ENOMEM; 37721308Sache goto fail; 37821308Sache } 37921308Sache disk->d_softc = sc; 38021308Sache error = g_mirror_connect_disk(disk, pp); 38147558Sache if (error != 0) 38247558Sache goto fail; 38321308Sache disk->d_id = md->md_did; 38421308Sache disk->d_state = G_MIRROR_DISK_STATE_NONE; 38521308Sache disk->d_priority = md->md_priority; 38621308Sache disk->d_delay.sec = 0; 38721308Sache disk->d_delay.frac = 0; 38821308Sache binuptime(&disk->d_last_used); 38921308Sache disk->d_flags = md->md_dflags; 39021308Sache if (md->md_provider[0] != '\0') 39121308Sache disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED; 39221308Sache disk->d_sync.ds_consumer = NULL; 39321308Sache disk->d_sync.ds_offset = md->md_sync_offset; 39421308Sache disk->d_sync.ds_offset_done = md->md_sync_offset; 39521308Sache disk->d_sync.ds_resync = -1; 39621308Sache disk->d_sync.ds_syncid = md->md_syncid; 39721308Sache if (errorp != NULL) 39847558Sache *errorp = 0; 39947558Sache return (disk); 40021308Sachefail: 40121308Sache if (errorp != NULL) 40221308Sache *errorp = error; 40321308Sache if (disk != NULL) { 40421308Sache g_mirror_disconnect_consumer(sc, disk->d_consumer); 40521308Sache free(disk, M_MIRROR); 40621308Sache } 40721308Sache return (NULL); 40821308Sache} 40921308Sache 41021308Sachestatic void 41121308Sacheg_mirror_destroy_disk(struct g_mirror_disk *disk) 41221308Sache{ 41321308Sache struct g_mirror_softc *sc; 41421308Sache 41521308Sache g_topology_assert(); 41621308Sache 41721308Sache LIST_REMOVE(disk, d_next); 41821308Sache g_mirror_event_cancel(disk); 41921308Sache sc = disk->d_softc; 42021308Sache if (sc->sc_hint == disk) 42121308Sache sc->sc_hint = NULL; 42247558Sache switch (disk->d_state) { 42347558Sache case G_MIRROR_DISK_STATE_SYNCHRONIZING: 42421308Sache g_mirror_sync_stop(disk, 1); 42521308Sache /* FALLTHROUGH */ 42621308Sache case G_MIRROR_DISK_STATE_NEW: 42721308Sache case G_MIRROR_DISK_STATE_STALE: 42821308Sache case G_MIRROR_DISK_STATE_ACTIVE: 42921308Sache g_mirror_disconnect_consumer(sc, disk->d_consumer); 43021308Sache free(disk, M_MIRROR); 43121308Sache break; 43221308Sache default: 43321308Sache KASSERT(0 == 1, ("Wrong disk state (%s, %s).", 43421308Sache g_mirror_get_diskname(disk), 43521308Sache g_mirror_disk_state2str(disk->d_state))); 43621308Sache } 43721308Sache} 43821308Sache 43921308Sachestatic void 44021308Sacheg_mirror_destroy_device(struct g_mirror_softc *sc) 44121308Sache{ 44221308Sache struct g_mirror_disk *disk; 44321308Sache struct g_mirror_event *ep; 44421308Sache struct g_geom *gp; 44521308Sache struct g_consumer *cp, *tmpcp; 44621308Sache 44721308Sache g_topology_assert(); 44821308Sache 44921308Sache gp = sc->sc_geom; 45021308Sache if (sc->sc_provider != NULL) 45121308Sache g_mirror_destroy_provider(sc); 45247558Sache for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL; 45347558Sache disk = LIST_FIRST(&sc->sc_disks)) { 45421308Sache g_mirror_destroy_disk(disk); 45521308Sache } 45621308Sache while ((ep = g_mirror_event_get(sc)) != NULL) { 45721308Sache if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) 458119610Sache g_mirror_event_free(ep); 45921308Sache else { 460119610Sache ep->e_error = ECANCELED; 46121308Sache ep->e_flags |= G_MIRROR_EVENT_DONE; 46221308Sache G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep); 46321308Sache mtx_lock(&sc->sc_events_mtx); 46421308Sache wakeup(ep); 465119610Sache mtx_unlock(&sc->sc_events_mtx); 46621308Sache } 46721308Sache } 46821308Sache callout_drain(&sc->sc_callout); 46921308Sache gp->softc = NULL; 47021308Sache 47121308Sache LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) { 47221308Sache g_mirror_disconnect_consumer(sc, cp); 47321308Sache } 47421308Sache sc->sc_sync.ds_geom->softc = NULL; 47521308Sache g_wither_geom(sc->sc_sync.ds_geom, ENXIO); 47621308Sache mtx_destroy(&sc->sc_queue_mtx); 47721308Sache mtx_destroy(&sc->sc_events_mtx); 47847558Sache G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name); 47947558Sache g_wither_geom(gp, ENXIO); 48021308Sache} 48121308Sache 48221308Sachestatic void 48321308Sacheg_mirror_orphan(struct g_consumer *cp) 48421308Sache{ 48521308Sache struct g_mirror_disk *disk; 48621308Sache 48721308Sache g_topology_assert(); 48821308Sache 48921308Sache disk = cp->private; 49021308Sache if (disk == NULL) 49121308Sache return; 49221308Sache disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 49321308Sache g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 49421308Sache G_MIRROR_EVENT_DONTWAIT); 495119610Sache} 496119610Sache 497119610Sachestatic void 49821308Sacheg_mirror_spoiled(struct g_consumer *cp) 49921308Sache{ 50021308Sache struct g_mirror_disk *disk; 50121308Sache 50221308Sache g_topology_assert(); 50321308Sache 50421308Sache disk = cp->private; 505119610Sache if (disk == NULL) 506119610Sache return; 50721308Sache disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 508119610Sache g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 50921308Sache G_MIRROR_EVENT_DONTWAIT); 51021308Sache} 51121308Sache 51221308Sache/* 51321308Sache * Function should return the next active disk on the list. 51421308Sache * It is possible that it will be the same disk as given. 51521308Sache * If there are no active disks on list, NULL is returned. 51621308Sache */ 51747558Sachestatic __inline struct g_mirror_disk * 51847558Sacheg_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk) 51921308Sache{ 52021308Sache struct g_mirror_disk *dp; 52121308Sache 52221308Sache for (dp = LIST_NEXT(disk, d_next); dp != disk; 52321308Sache dp = LIST_NEXT(dp, d_next)) { 52421308Sache if (dp == NULL) 52521308Sache dp = LIST_FIRST(&sc->sc_disks); 52621308Sache if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE) 52721308Sache break; 52821308Sache } 52921308Sache if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 530119610Sache return (NULL); 531119610Sache return (dp); 53221308Sache} 533119610Sache 53421308Sachestatic struct g_mirror_disk * 53521308Sacheg_mirror_get_disk(struct g_mirror_softc *sc) 53621308Sache{ 53721308Sache struct g_mirror_disk *disk; 53821308Sache 53921308Sache if (sc->sc_hint == NULL) { 54021308Sache sc->sc_hint = LIST_FIRST(&sc->sc_disks); 54121308Sache if (sc->sc_hint == NULL) 54221308Sache return (NULL); 54321308Sache } 54421308Sache disk = sc->sc_hint; 54521308Sache if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) { 54621308Sache disk = g_mirror_find_next(sc, disk); 54721308Sache if (disk == NULL) 54821308Sache return (NULL); 54921308Sache } 55021308Sache sc->sc_hint = g_mirror_find_next(sc, disk); 55121308Sache return (disk); 55221308Sache} 55321308Sache 55421308Sachestatic int 555119610Sacheg_mirror_write_metadata(struct g_mirror_disk *disk, 556119610Sache struct g_mirror_metadata *md) 557119610Sache{ 558119610Sache struct g_mirror_softc *sc; 559119610Sache struct g_consumer *cp; 560119610Sache off_t offset, length; 561119610Sache u_char *sector; 562119610Sache int close = 0, error = 0; 563119610Sache 564119610Sache g_topology_assert(); 565119610Sache 56621308Sache sc = disk->d_softc; 56721308Sache cp = disk->d_consumer; 56821308Sache KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name)); 56921308Sache KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name)); 57021308Sache length = cp->provider->sectorsize; 57121308Sache offset = cp->provider->mediasize - length; 57221308Sache sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO); 57321308Sache /* 57421308Sache * Open consumer if it wasn't opened and remember to close it. 57521308Sache */ 57621308Sache if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) { 57721308Sache error = g_access(cp, 0, 1, 1); 57821308Sache G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 57921308Sache cp->provider->name, 0, 1, 1, error); 58021308Sache if (error == 0) 58121308Sache close = 1; 58221308Sache#ifdef INVARIANTS 58321308Sache } else { 58421308Sache KASSERT(cp->acw > 0 && cp->ace > 0, 58521308Sache ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 58621308Sache cp->acr, cp->acw, cp->ace)); 58721308Sache#endif 58821308Sache } 58921308Sache if (error == 0) { 59021308Sache if (md != NULL) 59121308Sache mirror_metadata_encode(md, sector); 59221308Sache g_topology_unlock(); 59321308Sache error = g_write_data(cp, offset, sector, length); 59421308Sache g_topology_lock(); 59521308Sache } 59621308Sache free(sector, M_MIRROR); 59721308Sache if (close) { 59821308Sache g_access(cp, 0, -1, -1); 59921308Sache G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 60021308Sache cp->provider->name, 0, -1, -1, 0); 60121308Sache } 60221308Sache if (error != 0) { 60321308Sache disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 60421308Sache g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 60521308Sache G_MIRROR_EVENT_DONTWAIT); 60635486Sache } 60735486Sache return (error); 60835486Sache} 60935486Sache 61035486Sachestatic int 61135486Sacheg_mirror_clear_metadata(struct g_mirror_disk *disk) 61235486Sache{ 61321308Sache int error; 61421308Sache 61521308Sache g_topology_assert(); 61621308Sache error = g_mirror_write_metadata(disk, NULL); 61721308Sache if (error == 0) { 61821308Sache G_MIRROR_DEBUG(2, "Metadata on %s cleared.", 619119610Sache g_mirror_get_diskname(disk)); 62021308Sache } else { 62121308Sache G_MIRROR_DEBUG(0, 62221308Sache "Cannot clear metadata on disk %s (error=%d).", 62321308Sache g_mirror_get_diskname(disk), error); 62421308Sache } 62521308Sache return (error); 62621308Sache} 62721308Sache 62821308Sachevoid 62921308Sacheg_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk, 630119610Sache struct g_mirror_metadata *md) 63121308Sache{ 63221308Sache 63321308Sache strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic)); 63421308Sache md->md_version = G_MIRROR_VERSION; 63521308Sache strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name)); 63621308Sache md->md_mid = sc->sc_id; 63721308Sache md->md_all = sc->sc_ndisks; 63821308Sache md->md_slice = sc->sc_slice; 63921308Sache md->md_balance = sc->sc_balance; 64021308Sache md->md_mediasize = sc->sc_mediasize; 64121308Sache md->md_sectorsize = sc->sc_sectorsize; 64221308Sache md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK); 64321308Sache bzero(md->md_provider, sizeof(md->md_provider)); 64421308Sache if (disk == NULL) { 64521308Sache md->md_did = arc4random(); 64621308Sache md->md_priority = 0; 64721308Sache md->md_syncid = 0; 64821308Sache md->md_dflags = 0; 64921308Sache md->md_sync_offset = 0; 65021308Sache } else { 65121308Sache md->md_did = disk->d_id; 65221308Sache md->md_priority = disk->d_priority; 65321308Sache md->md_syncid = disk->d_sync.ds_syncid; 65421308Sache md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK); 65521308Sache if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 65621308Sache md->md_sync_offset = disk->d_sync.ds_offset_done; 65721308Sache else 65821308Sache md->md_sync_offset = 0; 659119610Sache if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) { 66021308Sache strlcpy(md->md_provider, 66121308Sache disk->d_consumer->provider->name, 66221308Sache sizeof(md->md_provider)); 66321308Sache } 66421308Sache } 66521308Sache} 66621308Sache 66721308Sachevoid 66821308Sacheg_mirror_update_metadata(struct g_mirror_disk *disk) 66921308Sache{ 67021308Sache struct g_mirror_metadata md; 67121308Sache int error; 67221308Sache 67321308Sache g_topology_assert(); 67421308Sache g_mirror_fill_metadata(disk->d_softc, disk, &md); 67521308Sache error = g_mirror_write_metadata(disk, &md); 676119610Sache if (error == 0) { 677119610Sache G_MIRROR_DEBUG(2, "Metadata on %s updated.", 678119610Sache g_mirror_get_diskname(disk)); 679119610Sache } else { 680119610Sache G_MIRROR_DEBUG(0, 681119610Sache "Cannot update metadata on disk %s (error=%d).", 682119610Sache g_mirror_get_diskname(disk), error); 683119610Sache } 684119610Sache} 685119610Sache 686119610Sachestatic void 687119610Sacheg_mirror_bump_syncid(struct g_mirror_softc *sc) 688119610Sache{ 689119610Sache struct g_mirror_disk *disk; 690119610Sache 691119610Sache g_topology_assert(); 692119610Sache KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0, 693119610Sache ("%s called with no active disks (device=%s).", __func__, 694119610Sache sc->sc_name)); 695119610Sache 696119610Sache sc->sc_syncid++; 697119610Sache G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name, 698119610Sache sc->sc_syncid); 699119610Sache LIST_FOREACH(disk, &sc->sc_disks, d_next) { 700119610Sache if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 701119610Sache disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 702119610Sache disk->d_sync.ds_syncid = sc->sc_syncid; 703119610Sache g_mirror_update_metadata(disk); 704119610Sache } 705119610Sache } 706119610Sache} 707119610Sache 708119610Sachestatic __inline int 709119610Sachebintime_cmp(struct bintime *bt1, struct bintime *bt2) 710119610Sache{ 711119610Sache 712119610Sache if (bt1->sec < bt2->sec) 713119610Sache return (-1); 714119610Sache else if (bt1->sec > bt2->sec) 715119610Sache return (1); 716119610Sache if (bt1->frac < bt2->frac) 717119610Sache return (-1); 718119610Sache else if (bt1->frac > bt2->frac) 719119610Sache return (1); 720119610Sache return (0); 72121308Sache} 72221308Sache 72321308Sachestatic void 72421308Sacheg_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp) 72521308Sache{ 72621308Sache 72721308Sache if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD) 72821308Sache return; 72921308Sache binuptime(&disk->d_delay); 73021308Sache bintime_sub(&disk->d_delay, &bp->bio_t0); 731119610Sache} 732119610Sache 733119610Sachestatic void 734119610Sacheg_mirror_done(struct bio *bp) 735119610Sache{ 73621308Sache struct g_mirror_softc *sc; 73721308Sache 73821308Sache sc = bp->bio_from->geom->softc; 73921308Sache bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR; 74021308Sache mtx_lock(&sc->sc_queue_mtx); 74121308Sache bioq_disksort(&sc->sc_queue, bp); 74221308Sache wakeup(sc); 74321308Sache mtx_unlock(&sc->sc_queue_mtx); 74421308Sache} 745119610Sache 74621308Sachestatic void 74721308Sacheg_mirror_regular_request(struct bio *bp) 74821308Sache{ 74921308Sache struct g_mirror_softc *sc; 75021308Sache struct g_mirror_disk *disk; 75121308Sache struct bio *pbp; 75221308Sache 75321308Sache g_topology_assert_not(); 754119610Sache 75521308Sache pbp = bp->bio_parent; 75621308Sache sc = pbp->bio_to->geom->softc; 75721308Sache disk = bp->bio_from->private; 75821308Sache if (disk == NULL) { 759119610Sache g_topology_lock(); 76021308Sache g_mirror_kill_consumer(sc, bp->bio_from); 76121308Sache g_topology_unlock(); 76221308Sache } else { 76321308Sache g_mirror_update_delay(disk, bp); 76421308Sache } 76521308Sache 76621308Sache pbp->bio_inbed++; 76721308Sache KASSERT(pbp->bio_inbed <= pbp->bio_children, 76821308Sache ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed, 769119610Sache pbp->bio_children)); 77021308Sache if (bp->bio_error == 0 && pbp->bio_error == 0) { 77147558Sache G_MIRROR_LOGREQ(3, bp, "Request delivered."); 772119610Sache g_destroy_bio(bp); 77321308Sache if (pbp->bio_children == pbp->bio_inbed) { 77421308Sache G_MIRROR_LOGREQ(3, pbp, "Request delivered."); 77521308Sache pbp->bio_completed = pbp->bio_length; 77621308Sache g_io_deliver(pbp, pbp->bio_error); 77721308Sache } 77821308Sache return; 77921308Sache } else if (bp->bio_error != 0) { 780119610Sache if (pbp->bio_error == 0) 781119610Sache pbp->bio_error = bp->bio_error; 782119610Sache G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).", 783119610Sache bp->bio_error); 784119610Sache if (disk != NULL) { 785119610Sache sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 78621308Sache g_mirror_event_send(disk, 78721308Sache G_MIRROR_DISK_STATE_DISCONNECTED, 78821308Sache G_MIRROR_EVENT_DONTWAIT); 78921308Sache } 79021308Sache switch (pbp->bio_cmd) { 79121308Sache case BIO_DELETE: 79221308Sache case BIO_WRITE: 79321308Sache pbp->bio_inbed--; 79421308Sache pbp->bio_children--; 79521308Sache break; 79621308Sache } 79721308Sache } 79821308Sache g_destroy_bio(bp); 79921308Sache 80021308Sache switch (pbp->bio_cmd) { 80121308Sache case BIO_READ: 80221308Sache if (pbp->bio_children == pbp->bio_inbed) { 80321308Sache pbp->bio_error = 0; 80421308Sache mtx_lock(&sc->sc_queue_mtx); 80521308Sache bioq_disksort(&sc->sc_queue, pbp); 80621308Sache G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 80721308Sache wakeup(sc); 80875406Sache mtx_unlock(&sc->sc_queue_mtx); 80921308Sache } 81075406Sache break; 81121308Sache case BIO_DELETE: 81221308Sache case BIO_WRITE: 81321308Sache if (pbp->bio_children == 0) { 81421308Sache /* 81521308Sache * All requests failed. 81621308Sache */ 81721308Sache } else if (pbp->bio_inbed < pbp->bio_children) { 81821308Sache /* Do nothing. */ 81921308Sache break; 82021308Sache } else if (pbp->bio_children == pbp->bio_inbed) { 82175406Sache /* Some requests succeeded. */ 82221308Sache pbp->bio_error = 0; 82375406Sache pbp->bio_completed = pbp->bio_length; 82421308Sache } 82521308Sache g_io_deliver(pbp, pbp->bio_error); 82621308Sache break; 82721308Sache default: 82821308Sache KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd)); 82921308Sache break; 83021308Sache } 83121308Sache} 83221308Sache 83321308Sachestatic void 83421308Sacheg_mirror_sync_done(struct bio *bp) 83521308Sache{ 83621308Sache struct g_mirror_softc *sc; 83721308Sache 83821308Sache G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered."); 83921308Sache sc = bp->bio_from->geom->softc; 84021308Sache bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC; 84121308Sache mtx_lock(&sc->sc_queue_mtx); 84221308Sache bioq_disksort(&sc->sc_queue, bp); 84321308Sache wakeup(sc); 84421308Sache mtx_unlock(&sc->sc_queue_mtx); 84521308Sache} 84621308Sache 84721308Sachestatic void 84821308Sacheg_mirror_start(struct bio *bp) 84921308Sache{ 85021308Sache struct g_mirror_softc *sc; 85121308Sache 85221308Sache sc = bp->bio_to->geom->softc; 85321308Sache /* 85421308Sache * If sc == NULL or there are no valid disks, provider's error 85521308Sache * should be set and g_mirror_start() should not be called at all. 85621308Sache */ 85721308Sache KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 85821308Sache ("Provider's error should be set (error=%d)(mirror=%s).", 85921308Sache bp->bio_to->error, bp->bio_to->name)); 86021308Sache G_MIRROR_LOGREQ(3, bp, "Request received."); 86121308Sache 86221308Sache switch (bp->bio_cmd) { 86321308Sache case BIO_READ: 86421308Sache case BIO_WRITE: 86521308Sache case BIO_DELETE: 86621308Sache break; 86721308Sache case BIO_GETATTR: 86821308Sache default: 86921308Sache g_io_deliver(bp, EOPNOTSUPP); 87021308Sache return; 87121308Sache } 87221308Sache mtx_lock(&sc->sc_queue_mtx); 87321308Sache bioq_disksort(&sc->sc_queue, bp); 87421308Sache G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 87521308Sache wakeup(sc); 87621308Sache mtx_unlock(&sc->sc_queue_mtx); 87721308Sache} 87821308Sache 87921308Sache/* 88021308Sache * Send one synchronization request. 88121308Sache */ 88221308Sachestatic void 88321308Sacheg_mirror_sync_one(struct g_mirror_disk *disk) 88421308Sache{ 88521308Sache struct g_mirror_softc *sc; 88621308Sache struct bio *bp; 887119610Sache 88821308Sache sc = disk->d_softc; 88921308Sache KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 89021308Sache ("Disk %s is not marked for synchronization.", 89121308Sache g_mirror_get_diskname(disk))); 89221308Sache 89375406Sache bp = g_new_bio(); 89475406Sache if (bp == NULL) 89521308Sache return; 89621308Sache bp->bio_parent = NULL; 89721308Sache bp->bio_cmd = BIO_READ; 89821308Sache bp->bio_offset = disk->d_sync.ds_offset; 89921308Sache bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset); 90075406Sache bp->bio_cflags = 0; 90121308Sache bp->bio_done = g_mirror_sync_done; 90221308Sache bp->bio_data = disk->d_sync.ds_data; 90375406Sache if (bp->bio_data == NULL) { 90475406Sache g_destroy_bio(bp); 90575406Sache return; 90675406Sache } 90775406Sache disk->d_sync.ds_offset += bp->bio_length; 90875406Sache bp->bio_to = sc->sc_provider; 90975406Sache G_MIRROR_LOGREQ(3, bp, "Sending synchronization request."); 91075406Sache g_io_request(bp, disk->d_sync.ds_consumer); 911119610Sache} 91275406Sache 91321308Sachestatic void 91475406Sacheg_mirror_sync_request(struct bio *bp) 91521308Sache{ 916119610Sache struct g_mirror_softc *sc; 91721308Sache struct g_mirror_disk *disk; 91821308Sache 91921308Sache sc = bp->bio_from->geom->softc; 92021308Sache disk = bp->bio_from->private; 92121308Sache if (disk == NULL) { 92221308Sache g_topology_lock(); 92321308Sache g_mirror_kill_consumer(sc, bp->bio_from); 92421308Sache g_topology_unlock(); 92521308Sache g_destroy_bio(bp); 92621308Sache return; 92721308Sache } 92821308Sache 92921308Sache /* 93021308Sache * Synchronization request. 93121308Sache */ 93221308Sache switch (bp->bio_cmd) { 93321308Sache case BIO_READ: 93421308Sache { 93521308Sache struct g_consumer *cp; 93621308Sache 93721308Sache if (bp->bio_error != 0) { 93821308Sache G_MIRROR_LOGREQ(0, bp, 93975406Sache "Synchronization request failed (error=%d).", 94075406Sache bp->bio_error); 94121308Sache g_destroy_bio(bp); 94221308Sache return; 94321308Sache } 94421308Sache bp->bio_cmd = BIO_WRITE; 94521308Sache bp->bio_cflags = 0; 94621308Sache G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 94721308Sache cp = disk->d_consumer; 94821308Sache KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 94921308Sache ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 95021308Sache cp->acr, cp->acw, cp->ace)); 95121308Sache g_io_request(bp, cp); 95221308Sache return; 95321308Sache } 95421308Sache case BIO_WRITE: 95521308Sache { 95621308Sache struct g_mirror_disk_sync *sync; 95775406Sache 95821308Sache if (bp->bio_error != 0) { 95921308Sache G_MIRROR_LOGREQ(0, bp, 96021308Sache "Synchronization request failed (error=%d).", 96121308Sache bp->bio_error); 96221308Sache g_destroy_bio(bp); 96321308Sache sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY; 96421308Sache g_mirror_event_send(disk, 96521308Sache G_MIRROR_DISK_STATE_DISCONNECTED, 96621308Sache G_MIRROR_EVENT_DONTWAIT); 96721308Sache return; 96821308Sache } 96921308Sache G_MIRROR_LOGREQ(3, bp, "Synchronization request finished."); 97021308Sache sync = &disk->d_sync; 97121308Sache sync->ds_offset_done = bp->bio_offset + bp->bio_length; 97221308Sache g_destroy_bio(bp); 97321308Sache if (sync->ds_resync != -1) 97421308Sache break; 97521308Sache if (sync->ds_offset_done == sc->sc_provider->mediasize) { 97621308Sache /* 97721308Sache * Disk up-to-date, activate it. 97821308Sache */ 97921308Sache g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE, 98021308Sache G_MIRROR_EVENT_DONTWAIT); 98121308Sache return; 98221308Sache } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) { 98321308Sache /* 98421308Sache * Update offset_done on every 100 blocks. 98575406Sache * XXX: This should be configurable. 98621308Sache */ 98721308Sache g_topology_lock(); 98821308Sache g_mirror_update_metadata(disk); 98921308Sache g_topology_unlock(); 99021308Sache } 99121308Sache return; 99221308Sache } 99321308Sache default: 99421308Sache KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 99521308Sache bp->bio_cmd, sc->sc_name)); 99621308Sache break; 99721308Sache } 99821308Sache} 99921308Sache 100021308Sachestatic void 100121308Sacheg_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp) 100221308Sache{ 100321308Sache struct g_mirror_disk *disk; 100421308Sache struct g_consumer *cp; 100521308Sache struct bio *cbp; 100621308Sache 100721308Sache LIST_FOREACH(disk, &sc->sc_disks, d_next) { 100821308Sache if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE) 100921308Sache break; 101021308Sache } 101121308Sache if (disk == NULL) { 101221308Sache if (bp->bio_error == 0) 101321308Sache bp->bio_error = ENXIO; 101421308Sache g_io_deliver(bp, bp->bio_error); 101521308Sache return; 101621308Sache } 101721308Sache cbp = g_clone_bio(bp); 101821308Sache if (cbp == NULL) { 101921308Sache if (bp->bio_error == 0) 102021308Sache bp->bio_error = ENOMEM; 102121308Sache g_io_deliver(bp, bp->bio_error); 102221308Sache return; 102321308Sache } 102421308Sache /* 102521308Sache * Fill in the component buf structure. 102621308Sache */ 102721308Sache cp = disk->d_consumer; 102821308Sache cbp->bio_done = g_mirror_done; 102921308Sache cbp->bio_to = cp->provider; 103021308Sache G_MIRROR_LOGREQ(3, cbp, "Sending request."); 103121308Sache KASSERT(cp->acr > 0 && cp->ace > 0, 103221308Sache ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 103321308Sache cp->acw, cp->ace)); 103421308Sache g_io_request(cbp, cp); 103575406Sache} 103621308Sache 103721308Sachestatic void 103821308Sacheg_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp) 103921308Sache{ 104021308Sache struct g_mirror_disk *disk; 104121308Sache struct g_consumer *cp; 104221308Sache struct bio *cbp; 104321308Sache 104421308Sache disk = g_mirror_get_disk(sc); 104521308Sache if (disk == NULL) { 104621308Sache if (bp->bio_error == 0) 104721308Sache bp->bio_error = ENXIO; 104821308Sache g_io_deliver(bp, bp->bio_error); 104921308Sache return; 105021308Sache } 105121308Sache cbp = g_clone_bio(bp); 105221308Sache if (cbp == NULL) { 105321308Sache if (bp->bio_error == 0) 105421308Sache bp->bio_error = ENOMEM; 105521308Sache g_io_deliver(bp, bp->bio_error); 105621308Sache return; 105721308Sache } 105821308Sache /* 105921308Sache * Fill in the component buf structure. 106021308Sache */ 106175406Sache cp = disk->d_consumer; 106221308Sache cbp->bio_done = g_mirror_done; 106321308Sache cbp->bio_to = cp->provider; 106421308Sache G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1065119610Sache KASSERT(cp->acr > 0 && cp->ace > 0, 1066119610Sache ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1067119610Sache cp->acw, cp->ace)); 1068119610Sache g_io_request(cbp, cp); 106921308Sache} 107021308Sache 107121308Sachestatic void 107221308Sacheg_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp) 107321308Sache{ 107421308Sache struct g_mirror_disk *disk, *dp; 107521308Sache struct g_consumer *cp; 1076119610Sache struct bio *cbp; 107721308Sache struct bintime curtime; 107821308Sache 107921308Sache binuptime(&curtime); 108021308Sache /* 108121308Sache * Find a disk which the smallest load. 108221308Sache */ 108321308Sache disk = NULL; 108421308Sache LIST_FOREACH(dp, &sc->sc_disks, d_next) { 108521308Sache if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE) 108621308Sache continue; 108721308Sache /* If disk wasn't used for more than 2 sec, use it. */ 108821308Sache if (curtime.sec - dp->d_last_used.sec >= 2) { 108921308Sache disk = dp; 109021308Sache break; 109121308Sache } 109221308Sache if (disk == NULL || 109321308Sache bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) { 109421308Sache disk = dp; 109521308Sache } 109621308Sache } 109721308Sache cbp = g_clone_bio(bp); 109821308Sache if (cbp == NULL) { 109921308Sache if (bp->bio_error == 0) 110021308Sache bp->bio_error = ENOMEM; 1101119610Sache g_io_deliver(bp, bp->bio_error); 1102119610Sache return; 1103119610Sache } 1104119610Sache /* 110521308Sache * Fill in the component buf structure. 1106119610Sache */ 110721308Sache cp = disk->d_consumer; 110821308Sache cbp->bio_done = g_mirror_done; 110921308Sache cbp->bio_to = cp->provider; 111021308Sache binuptime(&disk->d_last_used); 111121308Sache G_MIRROR_LOGREQ(3, cbp, "Sending request."); 111221308Sache KASSERT(cp->acr > 0 && cp->ace > 0, 111321308Sache ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr, 1114119610Sache cp->acw, cp->ace)); 1115119610Sache g_io_request(cbp, cp); 1116119610Sache} 111721308Sache 1118119610Sachestatic void 111921308Sacheg_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp) 112075406Sache{ 1121119610Sache struct bio_queue_head queue; 1122119610Sache struct g_mirror_disk *disk; 1123119610Sache struct g_consumer *cp; 1124119610Sache struct bio *cbp; 112575406Sache off_t left, mod, offset, slice; 112675406Sache u_char *data; 112775406Sache u_int ndisks; 1128119610Sache 112975406Sache if (bp->bio_length <= sc->sc_slice) { 113021308Sache g_mirror_request_round_robin(sc, bp); 113121308Sache return; 113221308Sache } 113321308Sache ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE); 113421308Sache slice = bp->bio_length / ndisks; 113521308Sache mod = slice % sc->sc_provider->sectorsize; 113621308Sache if (mod != 0) 113721308Sache slice += sc->sc_provider->sectorsize - mod; 113821308Sache /* 113921308Sache * Allocate all bios before sending any request, so we can 114021308Sache * return ENOMEM in nice and clean way. 114121308Sache */ 114221308Sache left = bp->bio_length; 114321308Sache offset = bp->bio_offset; 114421308Sache data = bp->bio_data; 114521308Sache bioq_init(&queue); 114621308Sache LIST_FOREACH(disk, &sc->sc_disks, d_next) { 114721308Sache if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 114821308Sache continue; 114921308Sache cbp = g_clone_bio(bp); 115021308Sache if (cbp == NULL) { 1151119610Sache for (cbp = bioq_first(&queue); cbp != NULL; 1152119610Sache cbp = bioq_first(&queue)) { 1153119610Sache bioq_remove(&queue, cbp); 115421308Sache g_destroy_bio(cbp); 1155119610Sache } 115621308Sache if (bp->bio_error == 0) 115721308Sache bp->bio_error = ENOMEM; 115821308Sache g_io_deliver(bp, bp->bio_error); 115921308Sache return; 116021308Sache } 116121308Sache bioq_insert_tail(&queue, cbp); 116221308Sache cbp->bio_done = g_mirror_done; 1163119610Sache cbp->bio_caller1 = disk; 116421308Sache cbp->bio_to = disk->d_consumer->provider; 116521308Sache cbp->bio_offset = offset; 116621308Sache cbp->bio_data = data; 116721308Sache cbp->bio_length = MIN(left, slice); 1168119610Sache left -= cbp->bio_length; 1169119610Sache if (left == 0) 1170119610Sache break; 1171119610Sache offset += cbp->bio_length; 1172119610Sache data += cbp->bio_length; 1173119610Sache } 1174119610Sache for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) { 1175119610Sache bioq_remove(&queue, cbp); 1176119610Sache G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1177119610Sache disk = cbp->bio_caller1; 1178119610Sache cbp->bio_caller1 = NULL; 1179119610Sache cp = disk->d_consumer; 1180119610Sache KASSERT(cp->acr > 0 && cp->ace > 0, 1181119610Sache ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, 118221308Sache cp->acr, cp->acw, cp->ace)); 118321308Sache g_io_request(cbp, disk->d_consumer); 118421308Sache } 118521308Sache} 118675406Sache 118721308Sachestatic void 118821308Sacheg_mirror_register_request(struct bio *bp) 118921308Sache{ 119021308Sache struct g_mirror_softc *sc; 119121308Sache 119221308Sache sc = bp->bio_to->geom->softc; 119321308Sache switch (bp->bio_cmd) { 119421308Sache case BIO_READ: 119521308Sache switch (sc->sc_balance) { 119621308Sache case G_MIRROR_BALANCE_LOAD: 1197119610Sache g_mirror_request_load(sc, bp); 1198119610Sache break; 1199119610Sache case G_MIRROR_BALANCE_PREFER: 1200119610Sache g_mirror_request_prefer(sc, bp); 120121308Sache break; 1202119610Sache case G_MIRROR_BALANCE_ROUND_ROBIN: 1203119610Sache g_mirror_request_round_robin(sc, bp); 1204119610Sache break; 1205119610Sache case G_MIRROR_BALANCE_SPLIT: 1206119610Sache g_mirror_request_split(sc, bp); 1207119610Sache break; 120821308Sache } 120921308Sache return; 121021308Sache case BIO_WRITE: 121121308Sache case BIO_DELETE: 121221308Sache { 121321308Sache struct g_mirror_disk *disk; 121421308Sache struct g_mirror_disk_sync *sync; 121521308Sache struct bio_queue_head queue; 121675406Sache struct g_consumer *cp; 121721308Sache struct bio *cbp; 121821308Sache 121921308Sache /* 122021308Sache * Allocate all bios before sending any request, so we can 122121308Sache * return ENOMEM in nice and clean way. 122221308Sache */ 122321308Sache bioq_init(&queue); 122421308Sache LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1225119610Sache sync = &disk->d_sync; 1226119610Sache switch (disk->d_state) { 1227119610Sache case G_MIRROR_DISK_STATE_ACTIVE: 1228119610Sache break; 1229119610Sache case G_MIRROR_DISK_STATE_SYNCHRONIZING: 1230119610Sache if (bp->bio_offset >= sync->ds_offset) 123121308Sache continue; 123221308Sache else if (bp->bio_offset + bp->bio_length > 123321308Sache sync->ds_offset_done && 123421308Sache (bp->bio_offset < sync->ds_resync || 123521308Sache sync->ds_resync == -1)) { 123621308Sache sync->ds_resync = bp->bio_offset - 123721308Sache (bp->bio_offset % MAXPHYS); 123821308Sache } 123921308Sache break; 124075406Sache default: 124121308Sache continue; 124221308Sache } 124321308Sache cbp = g_clone_bio(bp); 124421308Sache if (cbp == NULL) { 124521308Sache for (cbp = bioq_first(&queue); cbp != NULL; 124621308Sache cbp = bioq_first(&queue)) { 124721308Sache bioq_remove(&queue, cbp); 124821308Sache g_destroy_bio(cbp); 124921308Sache } 125021308Sache if (bp->bio_error == 0) 125121308Sache bp->bio_error = ENOMEM; 125221308Sache g_io_deliver(bp, bp->bio_error); 125321308Sache return; 125421308Sache } 125521308Sache bioq_insert_tail(&queue, cbp); 125621308Sache cbp->bio_done = g_mirror_done; 125721308Sache cp = disk->d_consumer; 125821308Sache cbp->bio_caller1 = cp; 125921308Sache cbp->bio_to = cp->provider; 126021308Sache KASSERT(cp->acw > 0 && cp->ace > 0, 126121308Sache ("Consumer %s not opened (r%dw%de%d).", 126221308Sache cp->provider->name, cp->acr, cp->acw, cp->ace)); 126321308Sache } 126421308Sache for (cbp = bioq_first(&queue); cbp != NULL; 1265119610Sache cbp = bioq_first(&queue)) { 1266119610Sache bioq_remove(&queue, cbp); 1267119610Sache G_MIRROR_LOGREQ(3, cbp, "Sending request."); 1268119610Sache cp = cbp->bio_caller1; 1269119610Sache cbp->bio_caller1 = NULL; 127021308Sache g_io_request(cbp, cp); 127121308Sache } 127221308Sache /* 127321308Sache * Bump syncid on first write. 127421308Sache */ 127521308Sache if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) { 127621308Sache sc->sc_bump_syncid = 0; 127721308Sache g_topology_lock(); 127821308Sache g_mirror_bump_syncid(sc); 127975406Sache g_topology_unlock(); 128075406Sache } 128175406Sache return; 128275406Sache } 128375406Sache default: 128421308Sache KASSERT(1 == 0, ("Invalid command here: %u (device=%s)", 128521308Sache bp->bio_cmd, sc->sc_name)); 128621308Sache break; 128721308Sache } 128821308Sache} 128921308Sache 129021308Sachestatic int 129121308Sacheg_mirror_can_destroy(struct g_mirror_softc *sc) 129221308Sache{ 1293119610Sache struct g_geom *gp; 1294119610Sache struct g_consumer *cp; 1295119610Sache 1296119610Sache g_topology_assert(); 1297119610Sache gp = sc->sc_geom; 1298119610Sache LIST_FOREACH(cp, &gp->consumer, consumer) { 1299119610Sache if (g_mirror_is_busy(sc, cp)) 1300119610Sache return (0); 1301119610Sache } 1302119610Sache gp = sc->sc_sync.ds_geom; 1303119610Sache LIST_FOREACH(cp, &gp->consumer, consumer) { 130421308Sache if (g_mirror_is_busy(sc, cp)) 1305119610Sache return (0); 130621308Sache } 130721308Sache G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.", 130821308Sache sc->sc_name); 130921308Sache return (1); 131021308Sache} 131121308Sache 131221308Sachestatic int 131321308Sacheg_mirror_try_destroy(struct g_mirror_softc *sc) 131421308Sache{ 131521308Sache 1316119610Sache if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) { 1317119610Sache g_topology_lock(); 1318119610Sache if (!g_mirror_can_destroy(sc)) { 131921308Sache g_topology_unlock(); 1320119610Sache return (0); 132121308Sache } 132221308Sache g_topology_unlock(); 132321308Sache G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 132421308Sache &sc->sc_worker); 132521308Sache wakeup(&sc->sc_worker); 132621308Sache sc->sc_worker = NULL; 132721308Sache } else { 132821308Sache g_topology_lock(); 132921308Sache if (!g_mirror_can_destroy(sc)) { 133021308Sache g_topology_unlock(); 133121308Sache return (0); 133221308Sache } 1333119610Sache g_mirror_destroy_device(sc); 133421308Sache g_topology_unlock(); 1335119610Sache free(sc, M_MIRROR); 1336119610Sache } 1337119610Sache return (1); 133821308Sache} 133921308Sache 134021308Sache/* 134121308Sache * Worker thread. 134221308Sache */ 134321308Sachestatic void 134421308Sacheg_mirror_worker(void *arg) 134521308Sache{ 134621308Sache struct g_mirror_softc *sc; 134721308Sache struct g_mirror_disk *disk; 134821308Sache struct g_mirror_disk_sync *sync; 134921308Sache struct g_mirror_event *ep; 135021308Sache struct bio *bp; 135121308Sache u_int nreqs; 135275406Sache 135321308Sache sc = arg; 135421308Sache curthread->td_base_pri = PRIBIO; 135521308Sache 135621308Sache nreqs = 0; 135721308Sache for (;;) { 135821308Sache G_MIRROR_DEBUG(5, "%s: Let's see...", __func__); 135921308Sache /* 136021308Sache * First take a look at events. 1361119610Sache * This is important to handle events before any I/O requests. 136221308Sache */ 136321308Sache ep = g_mirror_event_get(sc); 136421308Sache if (ep != NULL) { 136521308Sache g_topology_lock(); 136621308Sache if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) { 136721308Sache /* Update only device status. */ 136821308Sache G_MIRROR_DEBUG(3, 136921308Sache "Running event for device %s.", 137021308Sache sc->sc_name); 137121308Sache ep->e_error = 0; 137221308Sache g_mirror_update_device(sc, 1); 137321308Sache } else { 137421308Sache /* Update disk status. */ 137521308Sache G_MIRROR_DEBUG(3, "Running event for disk %s.", 137621308Sache g_mirror_get_diskname(ep->e_disk)); 137721308Sache ep->e_error = g_mirror_update_disk(ep->e_disk, 137821308Sache ep->e_state); 137921308Sache if (ep->e_error == 0) 138021308Sache g_mirror_update_device(sc, 0); 138121308Sache } 138221308Sache g_topology_unlock(); 138321308Sache if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) { 138421308Sache KASSERT(ep->e_error == 0, 138521308Sache ("Error cannot be handled.")); 138621308Sache g_mirror_event_free(ep); 138721308Sache } else { 138821308Sache ep->e_flags |= G_MIRROR_EVENT_DONE; 138921308Sache G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, 139021308Sache ep); 139121308Sache mtx_lock(&sc->sc_events_mtx); 139221308Sache wakeup(ep); 139321308Sache mtx_unlock(&sc->sc_events_mtx); 139421308Sache } 139521308Sache if ((sc->sc_flags & 139621308Sache G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 139721308Sache if (g_mirror_try_destroy(sc)) 139821308Sache kthread_exit(0); 139921308Sache } 140021308Sache G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__); 140121308Sache continue; 140221308Sache } 140321308Sache /* 140421308Sache * Now I/O requests. 140521308Sache */ 140621308Sache /* Get first request from the queue. */ 140721308Sache mtx_lock(&sc->sc_queue_mtx); 140821308Sache bp = bioq_first(&sc->sc_queue); 140921308Sache if (bp == NULL) { 141021308Sache if ((sc->sc_flags & 141121308Sache G_MIRROR_DEVICE_FLAG_DESTROY) != 0) { 141221308Sache mtx_unlock(&sc->sc_queue_mtx); 141321308Sache if (g_mirror_try_destroy(sc)) 141421308Sache kthread_exit(0); 141521308Sache mtx_lock(&sc->sc_queue_mtx); 141621308Sache } 141721308Sache } 141821308Sache if (sc->sc_sync.ds_ndisks > 0 && 141921308Sache (bp == NULL || nreqs > g_mirror_reqs_per_sync)) { 142021308Sache mtx_unlock(&sc->sc_queue_mtx); 142121308Sache /* 142275406Sache * It is time for synchronization... 142321308Sache */ 142421308Sache nreqs = 0; 142521308Sache LIST_FOREACH(disk, &sc->sc_disks, d_next) { 142621308Sache if (disk->d_state != 142721308Sache G_MIRROR_DISK_STATE_SYNCHRONIZING) { 142821308Sache continue; 142921308Sache } 143021308Sache sync = &disk->d_sync; 143121308Sache if (sync->ds_offset >= 143221308Sache sc->sc_provider->mediasize) { 143321308Sache continue; 143421308Sache } 143521308Sache if (sync->ds_offset > sync->ds_offset_done) 143621308Sache continue; 143721308Sache if (sync->ds_resync != -1) { 143821308Sache sync->ds_offset = sync->ds_resync; 143921308Sache sync->ds_offset_done = sync->ds_resync; 144075406Sache sync->ds_resync = -1; 144121308Sache } 144275406Sache g_mirror_sync_one(disk); 144375406Sache } 1444119610Sache G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__); 144521308Sache goto sleep; 144675406Sache } 144721308Sache if (bp == NULL) { 144821308Sache MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1", 0); 144921308Sache G_MIRROR_DEBUG(5, "%s: I'm here 3.", __func__); 145021308Sache continue; 145121308Sache } 145221308Sache nreqs++; 145321308Sache bioq_remove(&sc->sc_queue, bp); 145421308Sache mtx_unlock(&sc->sc_queue_mtx); 145521308Sache 145621308Sache if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) { 145721308Sache g_mirror_regular_request(bp); 145821308Sache } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) { 145921308Sache u_int timeout, sps; 146075406Sache 146121308Sache g_mirror_sync_request(bp); 146275406Sachesleep: 146375406Sache sps = atomic_load_acq_int(&g_mirror_syncs_per_sec); 146421308Sache if (sps == 0) { 146521308Sache G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__); 146621308Sache continue; 146721308Sache } 146821308Sache mtx_lock(&sc->sc_queue_mtx); 1469119610Sache if (bioq_first(&sc->sc_queue) != NULL) { 147021308Sache mtx_unlock(&sc->sc_queue_mtx); 147175406Sache G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__); 147221308Sache continue; 147321308Sache } 147421308Sache timeout = hz / sps; 147521308Sache if (timeout == 0) 147621308Sache timeout = 1; 147721308Sache MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w2", 147875406Sache timeout); 147921308Sache } else { 148021308Sache g_mirror_register_request(bp); 148121308Sache } 148221308Sache G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__); 148321308Sache } 148421308Sache} 148521308Sache 1486/* 1487 * Open disk's consumer if needed. 1488 */ 1489static void 1490g_mirror_update_access(struct g_mirror_disk *disk) 1491{ 1492 struct g_provider *pp; 1493 struct g_consumer *cp; 1494 int acr, acw, ace, cpw, error; 1495 1496 g_topology_assert(); 1497 1498 cp = disk->d_consumer; 1499 pp = disk->d_softc->sc_provider; 1500 if (pp == NULL) { 1501 acr = -cp->acr; 1502 acw = -cp->acw; 1503 ace = -cp->ace; 1504 } else { 1505 acr = pp->acr - cp->acr; 1506 acw = pp->acw - cp->acw; 1507 ace = pp->ace - cp->ace; 1508 /* Grab an extra "exclusive" bit. */ 1509 if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) 1510 ace++; 1511 } 1512 if (acr == 0 && acw == 0 && ace == 0) 1513 return; 1514 cpw = cp->acw; 1515 error = g_access(cp, acr, acw, ace); 1516 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, acr, 1517 acw, ace, error); 1518 if (error != 0) { 1519 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1520 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1521 G_MIRROR_EVENT_DONTWAIT); 1522 return; 1523 } 1524 if (cpw == 0 && cp->acw > 0) { 1525 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.", 1526 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1527 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1528 } else if (cpw > 0 && cp->acw == 0) { 1529 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.", 1530 g_mirror_get_diskname(disk), disk->d_softc->sc_name); 1531 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1532 } 1533} 1534 1535static void 1536g_mirror_sync_start(struct g_mirror_disk *disk) 1537{ 1538 struct g_mirror_softc *sc; 1539 struct g_consumer *cp; 1540 int error; 1541 1542 g_topology_assert(); 1543 1544 sc = disk->d_softc; 1545 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 1546 ("Device not in RUNNING state (%s, %u).", sc->sc_name, 1547 sc->sc_state)); 1548 cp = disk->d_consumer; 1549 KASSERT(cp->acr == 0 && cp->acw == 0 && cp->ace == 0, 1550 ("Consumer %s already opened.", cp->provider->name)); 1551 1552 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name, 1553 g_mirror_get_diskname(disk)); 1554 error = g_access(cp, 0, 1, 1); 1555 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, 1, 1556 1, error); 1557 if (error != 0) { 1558 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED, 1559 G_MIRROR_EVENT_DONTWAIT); 1560 return; 1561 } 1562 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 1563 KASSERT(disk->d_sync.ds_consumer == NULL, 1564 ("Sync consumer already exists (device=%s, disk=%s).", 1565 sc->sc_name, g_mirror_get_diskname(disk))); 1566 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom); 1567 disk->d_sync.ds_consumer->private = disk; 1568 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider); 1569 KASSERT(error == 0, ("Cannot attach to %s (error=%d).", 1570 disk->d_softc->sc_name, error)); 1571 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0); 1572 KASSERT(error == 0, ("Cannot open %s (error=%d).", 1573 disk->d_softc->sc_name, error)); 1574 disk->d_sync.ds_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK); 1575 sc->sc_sync.ds_ndisks++; 1576} 1577 1578/* 1579 * Stop synchronization process. 1580 * type: 0 - synchronization finished 1581 * 1 - synchronization stopped 1582 */ 1583static void 1584g_mirror_sync_stop(struct g_mirror_disk *disk, int type) 1585{ 1586 struct g_consumer *cp; 1587 1588 g_topology_assert(); 1589 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 1590 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1591 g_mirror_disk_state2str(disk->d_state))); 1592 if (disk->d_sync.ds_consumer == NULL) 1593 return; 1594 1595 if (type == 0) { 1596 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.", 1597 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1598 } else /* if (type == 1) */ { 1599 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.", 1600 disk->d_softc->sc_name, g_mirror_get_diskname(disk)); 1601 } 1602 cp = disk->d_sync.ds_consumer; 1603 g_access(cp, -1, 0, 0); 1604 g_mirror_kill_consumer(disk->d_softc, cp); 1605 free(disk->d_sync.ds_data, M_MIRROR); 1606 disk->d_sync.ds_consumer = NULL; 1607 disk->d_softc->sc_sync.ds_ndisks--; 1608 cp = disk->d_consumer; 1609 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1, 1610 ("Consumer %s not opened.", cp->provider->name)); 1611 g_access(cp, 0, -1, -1); 1612 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, -1, 1613 -1, 0); 1614 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 1615} 1616 1617static void 1618g_mirror_launch_provider(struct g_mirror_softc *sc) 1619{ 1620 struct g_mirror_disk *disk; 1621 struct g_provider *pp; 1622 1623 g_topology_assert(); 1624 1625 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name); 1626 pp->mediasize = sc->sc_mediasize; 1627 pp->sectorsize = sc->sc_sectorsize; 1628 sc->sc_provider = pp; 1629 g_error_provider(pp, 0); 1630 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name, 1631 pp->name); 1632 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1633 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1634 g_mirror_sync_start(disk); 1635 } 1636} 1637 1638static void 1639g_mirror_destroy_provider(struct g_mirror_softc *sc) 1640{ 1641 struct g_mirror_disk *disk; 1642 struct bio *bp; 1643 1644 g_topology_assert(); 1645 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).", 1646 sc->sc_name)); 1647 1648 g_error_provider(sc->sc_provider, ENXIO); 1649 mtx_lock(&sc->sc_queue_mtx); 1650 while ((bp = bioq_first(&sc->sc_queue)) != NULL) { 1651 bioq_remove(&sc->sc_queue, bp); 1652 g_io_deliver(bp, ENXIO); 1653 } 1654 mtx_unlock(&sc->sc_queue_mtx); 1655 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name, 1656 sc->sc_provider->name); 1657 sc->sc_provider->flags |= G_PF_WITHER; 1658 g_orphan_provider(sc->sc_provider, ENXIO); 1659 sc->sc_provider = NULL; 1660 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1661 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) 1662 g_mirror_sync_stop(disk, 1); 1663 } 1664} 1665 1666static void 1667g_mirror_go(void *arg) 1668{ 1669 struct g_mirror_softc *sc; 1670 1671 sc = arg; 1672 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name); 1673 g_mirror_event_send(sc, 0, 1674 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE); 1675} 1676 1677static u_int 1678g_mirror_determine_state(struct g_mirror_disk *disk) 1679{ 1680 struct g_mirror_softc *sc; 1681 u_int state; 1682 1683 sc = disk->d_softc; 1684 if (sc->sc_syncid == disk->d_sync.ds_syncid) { 1685 if ((disk->d_flags & 1686 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1687 /* Disk does not need synchronization. */ 1688 state = G_MIRROR_DISK_STATE_ACTIVE; 1689 } else { 1690 if ((sc->sc_flags & 1691 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1692 (disk->d_flags & 1693 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1694 /* 1695 * We can start synchronization from 1696 * the stored offset. 1697 */ 1698 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1699 } else { 1700 state = G_MIRROR_DISK_STATE_STALE; 1701 } 1702 } 1703 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) { 1704 /* 1705 * Reset all synchronization data for this disk, 1706 * because if it even was synchronized, it was 1707 * synchronized to disks with different syncid. 1708 */ 1709 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING; 1710 disk->d_sync.ds_offset = 0; 1711 disk->d_sync.ds_offset_done = 0; 1712 disk->d_sync.ds_syncid = sc->sc_syncid; 1713 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 || 1714 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) { 1715 state = G_MIRROR_DISK_STATE_SYNCHRONIZING; 1716 } else { 1717 state = G_MIRROR_DISK_STATE_STALE; 1718 } 1719 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ { 1720 /* 1721 * Not good, NOT GOOD! 1722 * It means that mirror was started on stale disks 1723 * and more fresh disk just arrive. 1724 * If there were writes, mirror is fucked up, sorry. 1725 * I think the best choice here is don't touch 1726 * this disk and inform the user laudly. 1727 */ 1728 G_MIRROR_DEBUG(0, "Device %s was started before the freshest " 1729 "disk (%s) arrives!! It will not be connected to the " 1730 "running device.", sc->sc_name, 1731 g_mirror_get_diskname(disk)); 1732 g_mirror_destroy_disk(disk); 1733 state = G_MIRROR_DISK_STATE_NONE; 1734 /* Return immediately, because disk was destroyed. */ 1735 return (state); 1736 } 1737 G_MIRROR_DEBUG(3, "State for %s disk: %s.", 1738 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state)); 1739 return (state); 1740} 1741 1742/* 1743 * Update device state. 1744 */ 1745static void 1746g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force) 1747{ 1748 struct g_mirror_disk *disk; 1749 u_int state; 1750 1751 g_topology_assert(); 1752 1753 switch (sc->sc_state) { 1754 case G_MIRROR_DEVICE_STATE_STARTING: 1755 { 1756 struct g_mirror_disk *pdisk; 1757 u_int dirty, ndisks, syncid; 1758 1759 KASSERT(sc->sc_provider == NULL, 1760 ("Non-NULL provider in STARTING state (%s).", sc->sc_name)); 1761 /* 1762 * Are we ready? We are, if all disks are connected or 1763 * if we have any disks and 'force' is true. 1764 */ 1765 if ((force && g_mirror_ndisks(sc, -1) > 0) || 1766 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) { 1767 ; 1768 } else if (g_mirror_ndisks(sc, -1) == 0) { 1769 /* 1770 * Disks went down in starting phase, so destroy 1771 * device. 1772 */ 1773 callout_drain(&sc->sc_callout); 1774 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1775 return; 1776 } else { 1777 return; 1778 } 1779 1780 /* 1781 * Activate all disks with the biggest syncid. 1782 */ 1783 if (force) { 1784 /* 1785 * If 'force' is true, we have been called due to 1786 * timeout, so don't bother canceling timeout. 1787 */ 1788 ndisks = 0; 1789 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1790 if ((disk->d_flags & 1791 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) { 1792 ndisks++; 1793 } 1794 } 1795 if (ndisks == 0) { 1796 /* No valid disks found, destroy device. */ 1797 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1798 return; 1799 } 1800 } else { 1801 /* Cancel timeout. */ 1802 callout_drain(&sc->sc_callout); 1803 } 1804 1805 /* 1806 * Find disk with the biggest syncid. 1807 */ 1808 syncid = 0; 1809 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1810 if (disk->d_sync.ds_syncid > syncid) 1811 syncid = disk->d_sync.ds_syncid; 1812 } 1813 1814 /* 1815 * Here we need to look for dirty disks and if all disks 1816 * with the biggest syncid are dirty, we have to choose 1817 * one with the biggest priority and rebuild the rest. 1818 */ 1819 /* 1820 * Find the number of dirty disks with the biggest syncid. 1821 * Find the number of disks with the biggest syncid. 1822 * While here, find a disk with the biggest priority. 1823 */ 1824 dirty = ndisks = 0; 1825 pdisk = NULL; 1826 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1827 if (disk->d_sync.ds_syncid != syncid) 1828 continue; 1829 if ((disk->d_flags & 1830 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1831 continue; 1832 } 1833 ndisks++; 1834 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) { 1835 dirty++; 1836 if (pdisk == NULL || 1837 pdisk->d_priority < disk->d_priority) { 1838 pdisk = disk; 1839 } 1840 } 1841 } 1842 if (dirty == 0) { 1843 /* No dirty disks at all, great. */ 1844 } else if (dirty == ndisks) { 1845 /* 1846 * Force synchronization for all dirty disks except one 1847 * with the biggest priority. 1848 */ 1849 KASSERT(pdisk != NULL, ("pdisk == NULL")); 1850 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a " 1851 "master disk for synchronization.", 1852 g_mirror_get_diskname(pdisk), sc->sc_name); 1853 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1854 if (disk->d_sync.ds_syncid != syncid) 1855 continue; 1856 if ((disk->d_flags & 1857 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1858 continue; 1859 } 1860 KASSERT((disk->d_flags & 1861 G_MIRROR_DISK_FLAG_DIRTY) != 0, 1862 ("Disk %s isn't marked as dirty.", 1863 g_mirror_get_diskname(disk))); 1864 /* Skip the disk with the biggest priority. */ 1865 if (disk == pdisk) 1866 continue; 1867 disk->d_sync.ds_syncid = 0; 1868 } 1869 } else if (dirty < ndisks) { 1870 /* 1871 * Force synchronization for all dirty disks. 1872 * We have some non-dirty disks. 1873 */ 1874 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1875 if (disk->d_sync.ds_syncid != syncid) 1876 continue; 1877 if ((disk->d_flags & 1878 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) { 1879 continue; 1880 } 1881 if ((disk->d_flags & 1882 G_MIRROR_DISK_FLAG_DIRTY) == 0) { 1883 continue; 1884 } 1885 disk->d_sync.ds_syncid = 0; 1886 } 1887 } 1888 1889 /* Reset hint. */ 1890 sc->sc_hint = NULL; 1891 sc->sc_syncid = syncid; 1892 if (force) { 1893 /* Remember to bump syncid on first write. */ 1894 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 1895 } 1896 state = G_MIRROR_DEVICE_STATE_RUNNING; 1897 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.", 1898 sc->sc_name, g_mirror_device_state2str(sc->sc_state), 1899 g_mirror_device_state2str(state)); 1900 sc->sc_state = state; 1901 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 1902 state = g_mirror_determine_state(disk); 1903 g_mirror_event_send(disk, state, 1904 G_MIRROR_EVENT_DONTWAIT); 1905 if (state == G_MIRROR_DISK_STATE_STALE) { 1906 sc->sc_bump_syncid = 1907 G_MIRROR_BUMP_ON_FIRST_WRITE; 1908 } 1909 } 1910 wakeup(&g_mirror_class); 1911 break; 1912 } 1913 case G_MIRROR_DEVICE_STATE_RUNNING: 1914 if (g_mirror_ndisks(sc, -1) == 0) { 1915 /* 1916 * No disks at all, we need to destroy device. 1917 */ 1918 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 1919 break; 1920 } else if (g_mirror_ndisks(sc, 1921 G_MIRROR_DISK_STATE_ACTIVE) == 0 && 1922 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 1923 /* 1924 * No active disks, destroy provider. 1925 */ 1926 if (sc->sc_provider != NULL) 1927 g_mirror_destroy_provider(sc); 1928 break; 1929 } else if (g_mirror_ndisks(sc, 1930 G_MIRROR_DISK_STATE_ACTIVE) > 0 && 1931 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) { 1932 /* 1933 * We have active disks, launch provider if it doesn't 1934 * exist. 1935 */ 1936 if (sc->sc_provider == NULL) 1937 g_mirror_launch_provider(sc); 1938 } 1939 /* 1940 * Bump syncid here, if we need to do it immediately. 1941 */ 1942 if (sc->sc_bump_syncid == G_MIRROR_BUMP_IMMEDIATELY) { 1943 sc->sc_bump_syncid = 0; 1944 g_mirror_bump_syncid(sc); 1945 } 1946 break; 1947 default: 1948 KASSERT(1 == 0, ("Wrong device state (%s, %s).", 1949 sc->sc_name, g_mirror_device_state2str(sc->sc_state))); 1950 break; 1951 } 1952} 1953 1954/* 1955 * Update disk state and device state if needed. 1956 */ 1957#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \ 1958 "Disk %s state changed from %s to %s (device %s).", \ 1959 g_mirror_get_diskname(disk), \ 1960 g_mirror_disk_state2str(disk->d_state), \ 1961 g_mirror_disk_state2str(state), sc->sc_name) 1962static int 1963g_mirror_update_disk(struct g_mirror_disk *disk, u_int state) 1964{ 1965 struct g_mirror_softc *sc; 1966 1967 g_topology_assert(); 1968 1969 sc = disk->d_softc; 1970again: 1971 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.", 1972 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state), 1973 g_mirror_disk_state2str(state)); 1974 switch (state) { 1975 case G_MIRROR_DISK_STATE_NEW: 1976 /* 1977 * Possible scenarios: 1978 * 1. New disk arrive. 1979 */ 1980 /* Previous state should be NONE. */ 1981 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE, 1982 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 1983 g_mirror_disk_state2str(disk->d_state))); 1984 DISK_STATE_CHANGED(); 1985 1986 disk->d_state = state; 1987 if (LIST_EMPTY(&sc->sc_disks)) 1988 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next); 1989 else { 1990 struct g_mirror_disk *dp; 1991 1992 LIST_FOREACH(dp, &sc->sc_disks, d_next) { 1993 if (disk->d_priority >= dp->d_priority) { 1994 LIST_INSERT_BEFORE(dp, disk, d_next); 1995 dp = NULL; 1996 break; 1997 } 1998 if (LIST_NEXT(dp, d_next) == NULL) 1999 break; 2000 } 2001 if (dp != NULL) 2002 LIST_INSERT_AFTER(dp, disk, d_next); 2003 } 2004 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.", 2005 sc->sc_name, g_mirror_get_diskname(disk)); 2006 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2007 break; 2008 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2009 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2010 g_mirror_device_state2str(sc->sc_state), 2011 g_mirror_get_diskname(disk), 2012 g_mirror_disk_state2str(disk->d_state))); 2013 state = g_mirror_determine_state(disk); 2014 if (state != G_MIRROR_DISK_STATE_NONE) 2015 goto again; 2016 break; 2017 case G_MIRROR_DISK_STATE_ACTIVE: 2018 /* 2019 * Possible scenarios: 2020 * 1. New disk does not need synchronization. 2021 * 2. Synchronization process finished successfully. 2022 */ 2023 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2024 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2025 g_mirror_device_state2str(sc->sc_state), 2026 g_mirror_get_diskname(disk), 2027 g_mirror_disk_state2str(disk->d_state))); 2028 /* Previous state should be NEW or SYNCHRONIZING. */ 2029 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW || 2030 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2031 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2032 g_mirror_disk_state2str(disk->d_state))); 2033 DISK_STATE_CHANGED(); 2034 2035 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2036 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2037 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2038 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING; 2039 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC; 2040 g_mirror_sync_stop(disk, 0); 2041 } 2042 disk->d_state = state; 2043 disk->d_sync.ds_offset = 0; 2044 disk->d_sync.ds_offset_done = 0; 2045 g_mirror_update_access(disk); 2046 g_mirror_update_metadata(disk); 2047 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.", 2048 sc->sc_name, g_mirror_get_diskname(disk)); 2049 break; 2050 case G_MIRROR_DISK_STATE_STALE: 2051 /* 2052 * Possible scenarios: 2053 * 1. Stale disk was connected. 2054 */ 2055 /* Previous state should be NEW. */ 2056 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2057 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2058 g_mirror_disk_state2str(disk->d_state))); 2059 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2060 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2061 g_mirror_device_state2str(sc->sc_state), 2062 g_mirror_get_diskname(disk), 2063 g_mirror_disk_state2str(disk->d_state))); 2064 /* 2065 * STALE state is only possible if device is marked 2066 * NOAUTOSYNC. 2067 */ 2068 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0, 2069 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2070 g_mirror_device_state2str(sc->sc_state), 2071 g_mirror_get_diskname(disk), 2072 g_mirror_disk_state2str(disk->d_state))); 2073 DISK_STATE_CHANGED(); 2074 2075 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2076 disk->d_state = state; 2077 g_mirror_update_metadata(disk); 2078 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.", 2079 sc->sc_name, g_mirror_get_diskname(disk)); 2080 break; 2081 case G_MIRROR_DISK_STATE_SYNCHRONIZING: 2082 /* 2083 * Possible scenarios: 2084 * 1. Disk which needs synchronization was connected. 2085 */ 2086 /* Previous state should be NEW. */ 2087 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2088 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk), 2089 g_mirror_disk_state2str(disk->d_state))); 2090 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING, 2091 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name, 2092 g_mirror_device_state2str(sc->sc_state), 2093 g_mirror_get_diskname(disk), 2094 g_mirror_disk_state2str(disk->d_state))); 2095 DISK_STATE_CHANGED(); 2096 2097 if (disk->d_state == G_MIRROR_DISK_STATE_NEW) 2098 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2099 disk->d_state = state; 2100 if (sc->sc_provider != NULL) { 2101 g_mirror_sync_start(disk); 2102 g_mirror_update_metadata(disk); 2103 } 2104 break; 2105 case G_MIRROR_DISK_STATE_DISCONNECTED: 2106 /* 2107 * Possible scenarios: 2108 * 1. Device wasn't running yet, but disk disappear. 2109 * 2. Disk was active and disapppear. 2110 * 3. Disk disappear during synchronization process. 2111 */ 2112 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) { 2113 /* 2114 * Previous state should be ACTIVE, STALE or 2115 * SYNCHRONIZING. 2116 */ 2117 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE || 2118 disk->d_state == G_MIRROR_DISK_STATE_STALE || 2119 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING, 2120 ("Wrong disk state (%s, %s).", 2121 g_mirror_get_diskname(disk), 2122 g_mirror_disk_state2str(disk->d_state))); 2123 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) { 2124 /* Previous state should be NEW. */ 2125 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW, 2126 ("Wrong disk state (%s, %s).", 2127 g_mirror_get_diskname(disk), 2128 g_mirror_disk_state2str(disk->d_state))); 2129 /* 2130 * Reset bumping syncid if disk disappeared in STARTING 2131 * state. 2132 */ 2133 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) 2134 sc->sc_bump_syncid = 0; 2135#ifdef INVARIANTS 2136 } else { 2137 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).", 2138 sc->sc_name, 2139 g_mirror_device_state2str(sc->sc_state), 2140 g_mirror_get_diskname(disk), 2141 g_mirror_disk_state2str(disk->d_state))); 2142#endif 2143 } 2144 DISK_STATE_CHANGED(); 2145 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.", 2146 sc->sc_name, g_mirror_get_diskname(disk)); 2147 2148 g_mirror_destroy_disk(disk); 2149 break; 2150 case G_MIRROR_DISK_STATE_DESTROY: 2151 { 2152 int error; 2153 2154 error = g_mirror_clear_metadata(disk); 2155 if (error != 0) 2156 return (error); 2157 DISK_STATE_CHANGED(); 2158 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", 2159 sc->sc_name, g_mirror_get_diskname(disk)); 2160 2161 g_mirror_destroy_disk(disk); 2162 sc->sc_ndisks--; 2163 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2164 g_mirror_update_metadata(disk); 2165 } 2166 break; 2167 } 2168 default: 2169 KASSERT(1 == 0, ("Unknown state (%u).", state)); 2170 break; 2171 } 2172 return (0); 2173} 2174#undef DISK_STATE_CHANGED 2175 2176static int 2177g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md) 2178{ 2179 struct g_provider *pp; 2180 u_char *buf; 2181 int error; 2182 2183 g_topology_assert(); 2184 2185 error = g_access(cp, 1, 0, 0); 2186 if (error != 0) 2187 return (error); 2188 pp = cp->provider; 2189 g_topology_unlock(); 2190 /* Metadata are stored on last sector. */ 2191 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, 2192 &error); 2193 g_topology_lock(); 2194 if (buf == NULL) { 2195 g_access(cp, -1, 0, 0); 2196 return (error); 2197 } 2198 if (error != 0) { 2199 g_access(cp, -1, 0, 0); 2200 g_free(buf); 2201 return (error); 2202 } 2203 error = g_access(cp, -1, 0, 0); 2204 KASSERT(error == 0, ("Cannot decrease access count for %s.", pp->name)); 2205 2206 /* Decode metadata. */ 2207 error = mirror_metadata_decode(buf, md); 2208 g_free(buf); 2209 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0) 2210 return (EINVAL); 2211 if (error != 0) { 2212 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.", 2213 cp->provider->name); 2214 return (error); 2215 } 2216 2217 return (0); 2218} 2219 2220static int 2221g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp, 2222 struct g_mirror_metadata *md) 2223{ 2224 2225 if (g_mirror_id2disk(sc, md->md_did) != NULL) { 2226 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.", 2227 pp->name, md->md_did); 2228 return (EEXIST); 2229 } 2230 if (md->md_all != sc->sc_ndisks) { 2231 G_MIRROR_DEBUG(1, 2232 "Invalid '%s' field on disk %s (device %s), skipping.", 2233 "md_all", pp->name, sc->sc_name); 2234 return (EINVAL); 2235 } 2236 if (md->md_slice != sc->sc_slice) { 2237 G_MIRROR_DEBUG(1, 2238 "Invalid '%s' field on disk %s (device %s), skipping.", 2239 "md_slice", pp->name, sc->sc_name); 2240 return (EINVAL); 2241 } 2242 if (md->md_balance != sc->sc_balance) { 2243 G_MIRROR_DEBUG(1, 2244 "Invalid '%s' field on disk %s (device %s), skipping.", 2245 "md_balance", pp->name, sc->sc_name); 2246 return (EINVAL); 2247 } 2248 if (md->md_mediasize != sc->sc_mediasize) { 2249 G_MIRROR_DEBUG(1, 2250 "Invalid '%s' field on disk %s (device %s), skipping.", 2251 "md_mediasize", pp->name, sc->sc_name); 2252 return (EINVAL); 2253 } 2254 if (sc->sc_mediasize > pp->mediasize) { 2255 G_MIRROR_DEBUG(1, 2256 "Invalid size of disk %s (device %s), skipping.", pp->name, 2257 sc->sc_name); 2258 return (EINVAL); 2259 } 2260 if (md->md_sectorsize != sc->sc_sectorsize) { 2261 G_MIRROR_DEBUG(1, 2262 "Invalid '%s' field on disk %s (device %s), skipping.", 2263 "md_sectorsize", pp->name, sc->sc_name); 2264 return (EINVAL); 2265 } 2266 if ((sc->sc_sectorsize % pp->sectorsize) != 0) { 2267 G_MIRROR_DEBUG(1, 2268 "Invalid sector size of disk %s (device %s), skipping.", 2269 pp->name, sc->sc_name); 2270 return (EINVAL); 2271 } 2272 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) { 2273 G_MIRROR_DEBUG(1, 2274 "Invalid device flags on disk %s (device %s), skipping.", 2275 pp->name, sc->sc_name); 2276 return (EINVAL); 2277 } 2278 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) { 2279 G_MIRROR_DEBUG(1, 2280 "Invalid disk flags on disk %s (device %s), skipping.", 2281 pp->name, sc->sc_name); 2282 return (EINVAL); 2283 } 2284 return (0); 2285} 2286 2287static int 2288g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp, 2289 struct g_mirror_metadata *md) 2290{ 2291 struct g_mirror_disk *disk; 2292 int error; 2293 2294 g_topology_assert(); 2295 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name); 2296 2297 error = g_mirror_check_metadata(sc, pp, md); 2298 if (error != 0) 2299 return (error); 2300 disk = g_mirror_init_disk(sc, pp, md, &error); 2301 if (disk == NULL) 2302 return (error); 2303 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW, 2304 G_MIRROR_EVENT_WAIT); 2305 return (error); 2306} 2307 2308static int 2309g_mirror_access(struct g_provider *pp, int acr, int acw, int ace) 2310{ 2311 struct g_mirror_softc *sc; 2312 struct g_mirror_disk *disk; 2313 int dcr, dcw, dce, err, error; 2314 2315 g_topology_assert(); 2316 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr, 2317 acw, ace); 2318 2319 dcr = pp->acr + acr; 2320 dcw = pp->acw + acw; 2321 dce = pp->ace + ace; 2322 2323 /* On first open, grab an extra "exclusive" bit */ 2324 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0) 2325 ace++; 2326 /* ... and let go of it on last close */ 2327 if (dcr == 0 && dcw == 0 && dce == 0) 2328 ace--; 2329 2330 sc = pp->geom->softc; 2331 if (sc == NULL || LIST_EMPTY(&sc->sc_disks)) { 2332 if (acr <= 0 && acw <= 0 && ace <= 0) 2333 return (0); 2334 else 2335 return (ENXIO); 2336 } 2337 error = ENXIO; 2338 LIST_FOREACH(disk, &sc->sc_disks, d_next) { 2339 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) 2340 continue; 2341 err = g_access(disk->d_consumer, acr, acw, ace); 2342 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", 2343 g_mirror_get_diskname(disk), acr, acw, ace, err); 2344 if (err == 0) { 2345 /* 2346 * Mark disk as dirty on open and unmark on close. 2347 */ 2348 if (pp->acw == 0 && dcw > 0) { 2349 G_MIRROR_DEBUG(1, 2350 "Disk %s (device %s) marked as dirty.", 2351 g_mirror_get_diskname(disk), sc->sc_name); 2352 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY; 2353 g_mirror_update_metadata(disk); 2354 } else if (pp->acw > 0 && dcw == 0) { 2355 G_MIRROR_DEBUG(1, 2356 "Disk %s (device %s) marked as clean.", 2357 g_mirror_get_diskname(disk), sc->sc_name); 2358 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY; 2359 g_mirror_update_metadata(disk); 2360 } 2361 error = 0; 2362 } else { 2363 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE; 2364 g_mirror_event_send(disk, 2365 G_MIRROR_DISK_STATE_DISCONNECTED, 2366 G_MIRROR_EVENT_DONTWAIT); 2367 } 2368 } 2369 /* 2370 * Be sure to return 0 for negativate access requests. 2371 * In case of some HW problems, it is possible that we don't have 2372 * any active disk here, so loop above will be no-op and error will 2373 * be ENXIO. 2374 */ 2375 if (error != 0 && acr <= 0 && acw <= 0 && ace <= 0) 2376 error = 0; 2377 return (error); 2378} 2379 2380static struct g_geom * 2381g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md) 2382{ 2383 struct g_mirror_softc *sc; 2384 struct g_geom *gp; 2385 int error, timeout; 2386 2387 g_topology_assert(); 2388 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name, 2389 md->md_mid); 2390 2391 /* One disk is minimum. */ 2392 if (md->md_all < 1) 2393 return (NULL); 2394 /* 2395 * Action geom. 2396 */ 2397 gp = g_new_geomf(mp, "%s", md->md_name); 2398 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO); 2399 gp->start = g_mirror_start; 2400 gp->spoiled = g_mirror_spoiled; 2401 gp->orphan = g_mirror_orphan; 2402 gp->access = g_mirror_access; 2403 gp->dumpconf = g_mirror_dumpconf; 2404 2405 sc->sc_id = md->md_mid; 2406 sc->sc_slice = md->md_slice; 2407 sc->sc_balance = md->md_balance; 2408 sc->sc_mediasize = md->md_mediasize; 2409 sc->sc_sectorsize = md->md_sectorsize; 2410 sc->sc_ndisks = md->md_all; 2411 sc->sc_flags = md->md_mflags; 2412 sc->sc_bump_syncid = 0; 2413 bioq_init(&sc->sc_queue); 2414 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 2415 LIST_INIT(&sc->sc_disks); 2416 TAILQ_INIT(&sc->sc_events); 2417 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF); 2418 callout_init(&sc->sc_callout, CALLOUT_MPSAFE); 2419 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING; 2420 gp->softc = sc; 2421 sc->sc_geom = gp; 2422 sc->sc_provider = NULL; 2423 /* 2424 * Synchronization geom. 2425 */ 2426 gp = g_new_geomf(mp, "%s.sync", md->md_name); 2427 gp->softc = sc; 2428 gp->orphan = g_mirror_orphan; 2429 sc->sc_sync.ds_geom = gp; 2430 sc->sc_sync.ds_ndisks = 0; 2431 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0, 2432 "g_mirror %s", md->md_name); 2433 if (error != 0) { 2434 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.", 2435 sc->sc_name); 2436 g_destroy_geom(sc->sc_sync.ds_geom); 2437 mtx_destroy(&sc->sc_events_mtx); 2438 mtx_destroy(&sc->sc_queue_mtx); 2439 g_destroy_geom(sc->sc_geom); 2440 free(sc, M_MIRROR); 2441 return (NULL); 2442 } 2443 2444 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id); 2445 2446 /* 2447 * Run timeout. 2448 */ 2449 timeout = atomic_load_acq_int(&g_mirror_timeout); 2450 callout_reset(&sc->sc_callout, timeout * hz, g_mirror_go, sc); 2451 return (sc->sc_geom); 2452} 2453 2454int 2455g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force) 2456{ 2457 struct g_provider *pp; 2458 2459 g_topology_assert(); 2460 2461 if (sc == NULL) 2462 return (ENXIO); 2463 pp = sc->sc_provider; 2464 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) { 2465 if (force) { 2466 G_MIRROR_DEBUG(0, "Device %s is still open, so it " 2467 "can't be definitely removed.", pp->name); 2468 } else { 2469 G_MIRROR_DEBUG(1, 2470 "Device %s is still open (r%dw%de%d).", pp->name, 2471 pp->acr, pp->acw, pp->ace); 2472 return (EBUSY); 2473 } 2474 } 2475 2476 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY; 2477 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT; 2478 g_topology_unlock(); 2479 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc); 2480 mtx_lock(&sc->sc_queue_mtx); 2481 wakeup(sc); 2482 mtx_unlock(&sc->sc_queue_mtx); 2483 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker); 2484 while (sc->sc_worker != NULL) 2485 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5); 2486 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker); 2487 g_topology_lock(); 2488 g_mirror_destroy_device(sc); 2489 free(sc, M_MIRROR); 2490 return (0); 2491} 2492 2493static void 2494g_mirror_taste_orphan(struct g_consumer *cp) 2495{ 2496 2497 KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2498 cp->provider->name)); 2499} 2500 2501static struct g_geom * 2502g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2503{ 2504 struct g_mirror_metadata md; 2505 struct g_mirror_softc *sc; 2506 struct g_consumer *cp; 2507 struct g_geom *gp; 2508 int error; 2509 2510 g_topology_assert(); 2511 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2512 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name); 2513 2514 gp = g_new_geomf(mp, "mirror:taste"); 2515 /* 2516 * This orphan function should be never called. 2517 */ 2518 gp->orphan = g_mirror_taste_orphan; 2519 cp = g_new_consumer(gp); 2520 g_attach(cp, pp); 2521 error = g_mirror_read_metadata(cp, &md); 2522 g_detach(cp); 2523 g_destroy_consumer(cp); 2524 g_destroy_geom(gp); 2525 if (error != 0) 2526 return (NULL); 2527 gp = NULL; 2528 2529 if (md.md_version > G_MIRROR_VERSION) { 2530 printf("geom_mirror.ko module is too old to handle %s.\n", 2531 pp->name); 2532 return (NULL); 2533 } 2534 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) 2535 return (NULL); 2536 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) { 2537 G_MIRROR_DEBUG(0, 2538 "Device %s: provider %s marked as inactive, skipping.", 2539 md.md_name, pp->name); 2540 return (NULL); 2541 } 2542 if (g_mirror_debug >= 2) 2543 mirror_metadata_dump(&md); 2544 2545 /* 2546 * Let's check if device already exists. 2547 */ 2548 sc = NULL; 2549 LIST_FOREACH(gp, &mp->geom, geom) { 2550 sc = gp->softc; 2551 if (sc == NULL) 2552 continue; 2553 if (sc->sc_sync.ds_geom == gp) 2554 continue; 2555 if (strcmp(md.md_name, sc->sc_name) != 0) 2556 continue; 2557 if (md.md_mid != sc->sc_id) { 2558 G_MIRROR_DEBUG(0, "Device %s already configured.", 2559 sc->sc_name); 2560 return (NULL); 2561 } 2562 break; 2563 } 2564 if (gp == NULL) { 2565 gp = g_mirror_create(mp, &md); 2566 if (gp == NULL) { 2567 G_MIRROR_DEBUG(0, "Cannot create device %s.", 2568 md.md_name); 2569 return (NULL); 2570 } 2571 sc = gp->softc; 2572 } 2573 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name); 2574 error = g_mirror_add_disk(sc, pp, &md); 2575 if (error != 0) { 2576 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).", 2577 pp->name, gp->name, error); 2578 if (LIST_EMPTY(&sc->sc_disks)) 2579 g_mirror_destroy(sc, 1); 2580 return (NULL); 2581 } 2582 return (gp); 2583} 2584 2585static int 2586g_mirror_destroy_geom(struct gctl_req *req __unused, 2587 struct g_class *mp __unused, struct g_geom *gp) 2588{ 2589 2590 return (g_mirror_destroy(gp->softc, 0)); 2591} 2592 2593static void 2594g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2595 struct g_consumer *cp, struct g_provider *pp) 2596{ 2597 struct g_mirror_softc *sc; 2598 2599 g_topology_assert(); 2600 2601 sc = gp->softc; 2602 if (sc == NULL) 2603 return; 2604 /* Skip synchronization geom. */ 2605 if (gp == sc->sc_sync.ds_geom) 2606 return; 2607 if (pp != NULL) { 2608 /* Nothing here. */ 2609 } else if (cp != NULL) { 2610 struct g_mirror_disk *disk; 2611 2612 disk = cp->private; 2613 if (disk == NULL) 2614 return; 2615 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id); 2616 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) { 2617 sbuf_printf(sb, "%s<Synchronized>", indent); 2618 if (disk->d_sync.ds_offset_done == 0) 2619 sbuf_printf(sb, "0%%"); 2620 else { 2621 sbuf_printf(sb, "%u%%", 2622 (u_int)((disk->d_sync.ds_offset_done * 100) / 2623 sc->sc_provider->mediasize)); 2624 } 2625 sbuf_printf(sb, "</Synchronized>\n"); 2626 } 2627 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, 2628 disk->d_sync.ds_syncid); 2629 sbuf_printf(sb, "%s<Flags>", indent); 2630 if (disk->d_flags == 0) 2631 sbuf_printf(sb, "NONE"); 2632 else { 2633 int first = 1; 2634 2635#define ADD_FLAG(flag, name) do { \ 2636 if ((disk->d_flags & (flag)) != 0) { \ 2637 if (!first) \ 2638 sbuf_printf(sb, ", "); \ 2639 else \ 2640 first = 0; \ 2641 sbuf_printf(sb, name); \ 2642 } \ 2643} while (0) 2644 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY"); 2645 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED"); 2646 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE"); 2647 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING, 2648 "SYNCHRONIZING"); 2649 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC"); 2650#undef ADD_FLAG 2651 } 2652 sbuf_printf(sb, "</Flags>\n"); 2653 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent, 2654 disk->d_priority); 2655 sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2656 g_mirror_disk_state2str(disk->d_state)); 2657 } else { 2658 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id); 2659 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid); 2660 sbuf_printf(sb, "%s<Flags>", indent); 2661 if (sc->sc_flags == 0) 2662 sbuf_printf(sb, "NONE"); 2663 else { 2664 int first = 1; 2665 2666#define ADD_FLAG(flag, name) do { \ 2667 if ((sc->sc_flags & (flag)) != 0) { \ 2668 if (!first) \ 2669 sbuf_printf(sb, ", "); \ 2670 else \ 2671 first = 0; \ 2672 sbuf_printf(sb, name); \ 2673 } \ 2674} while (0) 2675 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC"); 2676#undef ADD_FLAG 2677 } 2678 sbuf_printf(sb, "</Flags>\n"); 2679 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent, 2680 (u_int)sc->sc_slice); 2681 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent, 2682 balance_name(sc->sc_balance)); 2683 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2684 sc->sc_ndisks); 2685 sbuf_printf(sb, "%s<State>", indent); 2686 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) 2687 sbuf_printf(sb, "%s", "STARTING"); 2688 else if (sc->sc_ndisks == 2689 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE)) 2690 sbuf_printf(sb, "%s", "COMPLETE"); 2691 else 2692 sbuf_printf(sb, "%s", "DEGRADED"); 2693 sbuf_printf(sb, "</State>\n"); 2694 } 2695} 2696 2697static int 2698g_mirror_can_go(void) 2699{ 2700 struct g_mirror_softc *sc; 2701 struct g_geom *gp; 2702 struct g_provider *pp; 2703 int can_go; 2704 2705 DROP_GIANT(); 2706 can_go = 1; 2707 g_topology_lock(); 2708 LIST_FOREACH(gp, &g_mirror_class.geom, geom) { 2709 sc = gp->softc; 2710 if (sc == NULL) { 2711 can_go = 0; 2712 break; 2713 } 2714 pp = sc->sc_provider; 2715 if (pp == NULL || pp->error != 0) { 2716 can_go = 0; 2717 break; 2718 } 2719 } 2720 g_topology_unlock(); 2721 PICKUP_GIANT(); 2722 return (can_go); 2723} 2724 2725static void 2726g_mirror_rootwait(void) 2727{ 2728 2729 /* 2730 * HACK: Wait for GEOM, because g_mirror_rootwait() can be called, 2731 * HACK: before we get providers for tasting. 2732 */ 2733 tsleep(&g_mirror_class, PRIBIO, "mroot", hz * 3); 2734 /* 2735 * Wait for mirrors in degraded state. 2736 */ 2737 for (;;) { 2738 if (g_mirror_can_go()) 2739 break; 2740 tsleep(&g_mirror_class, PRIBIO, "mroot", hz); 2741 } 2742} 2743 2744SYSINIT(g_mirror_root, SI_SUB_RAID, SI_ORDER_FIRST, g_mirror_rootwait, NULL) 2745 2746DECLARE_GEOM_CLASS(g_mirror_class, g_mirror); 2747