2 * Copyright (c) 2004-2006 Pawel Jakub Dawidek <pjd@FreeBSD.org>
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/kernel.h>
33 #include <sys/module.h>
34 #include <sys/limits.h>
36 #include <sys/mutex.h>
39 #include <sys/sysctl.h>
40 #include <sys/malloc.h>
41 #include <sys/eventhandler.h>
43 #include <geom/geom.h>
45 #include <sys/kthread.h>
46 #include <sys/sched.h>
47 #include <geom/mirror/g_mirror.h>
49 FEATURE(geom_mirror, "GEOM mirroring support");
51 static MALLOC_DEFINE(M_MIRROR, "mirror_data", "GEOM_MIRROR Data");
53 SYSCTL_DECL(_kern_geom);
54 static SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0,
56 u_int g_mirror_debug = 0;
57 TUNABLE_INT("kern.geom.mirror.debug", &g_mirror_debug);
58 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RW, &g_mirror_debug, 0,
60 static u_int g_mirror_timeout = 4;
61 TUNABLE_INT("kern.geom.mirror.timeout", &g_mirror_timeout);
62 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RW, &g_mirror_timeout,
63 0, "Time to wait on all mirror components");
64 static u_int g_mirror_idletime = 5;
65 TUNABLE_INT("kern.geom.mirror.idletime", &g_mirror_idletime);
66 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RW,
67 &g_mirror_idletime, 0, "Mark components as clean when idling");
68 static u_int g_mirror_disconnect_on_failure = 1;
69 TUNABLE_INT("kern.geom.mirror.disconnect_on_failure",
70 &g_mirror_disconnect_on_failure);
71 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, disconnect_on_failure, CTLFLAG_RW,
72 &g_mirror_disconnect_on_failure, 0, "Disconnect component on I/O failure.");
73 static u_int g_mirror_syncreqs = 2;
74 TUNABLE_INT("kern.geom.mirror.sync_requests", &g_mirror_syncreqs);
75 SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, sync_requests, CTLFLAG_RDTUN,
76 &g_mirror_syncreqs, 0, "Parallel synchronization I/O requests.");
78 #define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \
79 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \
80 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \
81 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \
84 static eventhandler_tag g_mirror_post_sync = NULL;
85 static int g_mirror_shutdown = 0;
87 static int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp,
89 static g_taste_t g_mirror_taste;
90 static g_resize_t g_mirror_resize;
91 static void g_mirror_init(struct g_class *mp);
92 static void g_mirror_fini(struct g_class *mp);
94 struct g_class g_mirror_class = {
95 .name = G_MIRROR_CLASS_NAME,
97 .ctlreq = g_mirror_config,
98 .taste = g_mirror_taste,
99 .destroy_geom = g_mirror_destroy_geom,
100 .init = g_mirror_init,
101 .fini = g_mirror_fini,
102 .resize = g_mirror_resize
106 static void g_mirror_destroy_provider(struct g_mirror_softc *sc);
107 static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state);
108 static void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force);
109 static void g_mirror_dumpconf(struct sbuf *sb, const char *indent,
110 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp);
111 static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type);
112 static void g_mirror_register_request(struct bio *bp);
113 static void g_mirror_sync_release(struct g_mirror_softc *sc);
117 g_mirror_disk_state2str(int state)
121 case G_MIRROR_DISK_STATE_NONE:
123 case G_MIRROR_DISK_STATE_NEW:
125 case G_MIRROR_DISK_STATE_ACTIVE:
127 case G_MIRROR_DISK_STATE_STALE:
129 case G_MIRROR_DISK_STATE_SYNCHRONIZING:
130 return ("SYNCHRONIZING");
131 case G_MIRROR_DISK_STATE_DISCONNECTED:
132 return ("DISCONNECTED");
133 case G_MIRROR_DISK_STATE_DESTROY:
141 g_mirror_device_state2str(int state)
145 case G_MIRROR_DEVICE_STATE_STARTING:
147 case G_MIRROR_DEVICE_STATE_RUNNING:
155 g_mirror_get_diskname(struct g_mirror_disk *disk)
158 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL)
159 return ("[unknown]");
160 return (disk->d_name);
164 * --- Events handling functions ---
165 * Events in geom_mirror are used to maintain disks and device status
166 * from one thread to simplify locking.
169 g_mirror_event_free(struct g_mirror_event *ep)
176 g_mirror_event_send(void *arg, int state, int flags)
178 struct g_mirror_softc *sc;
179 struct g_mirror_disk *disk;
180 struct g_mirror_event *ep;
183 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK);
184 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep);
185 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) {
196 mtx_lock(&sc->sc_events_mtx);
197 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next);
198 mtx_unlock(&sc->sc_events_mtx);
199 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc);
200 mtx_lock(&sc->sc_queue_mtx);
202 mtx_unlock(&sc->sc_queue_mtx);
203 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0)
205 sx_assert(&sc->sc_lock, SX_XLOCKED);
206 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep);
207 sx_xunlock(&sc->sc_lock);
208 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) {
209 mtx_lock(&sc->sc_events_mtx);
210 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event",
214 g_mirror_event_free(ep);
215 sx_xlock(&sc->sc_lock);
219 static struct g_mirror_event *
220 g_mirror_event_get(struct g_mirror_softc *sc)
222 struct g_mirror_event *ep;
224 mtx_lock(&sc->sc_events_mtx);
225 ep = TAILQ_FIRST(&sc->sc_events);
226 mtx_unlock(&sc->sc_events_mtx);
231 g_mirror_event_remove(struct g_mirror_softc *sc, struct g_mirror_event *ep)
234 mtx_lock(&sc->sc_events_mtx);
235 TAILQ_REMOVE(&sc->sc_events, ep, e_next);
236 mtx_unlock(&sc->sc_events_mtx);
240 g_mirror_event_cancel(struct g_mirror_disk *disk)
242 struct g_mirror_softc *sc;
243 struct g_mirror_event *ep, *tmpep;
246 sx_assert(&sc->sc_lock, SX_XLOCKED);
248 mtx_lock(&sc->sc_events_mtx);
249 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) {
250 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0)
252 if (ep->e_disk != disk)
254 TAILQ_REMOVE(&sc->sc_events, ep, e_next);
255 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0)
256 g_mirror_event_free(ep);
258 ep->e_error = ECANCELED;
262 mtx_unlock(&sc->sc_events_mtx);
266 * Return the number of disks in given state.
267 * If state is equal to -1, count all connected disks.
270 g_mirror_ndisks(struct g_mirror_softc *sc, int state)
272 struct g_mirror_disk *disk;
275 sx_assert(&sc->sc_lock, SX_LOCKED);
277 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
278 if (state == -1 || disk->d_state == state)
285 * Find a disk in mirror by its disk ID.
287 static struct g_mirror_disk *
288 g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id)
290 struct g_mirror_disk *disk;
292 sx_assert(&sc->sc_lock, SX_XLOCKED);
294 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
295 if (disk->d_id == id)
302 g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp)
307 mtx_lock(&sc->sc_queue_mtx);
308 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
309 if (bp->bio_from == cp)
312 mtx_unlock(&sc->sc_queue_mtx);
317 g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp)
322 "I/O requests for %s exist, can't destroy it now.",
326 if (g_mirror_nrequests(sc, cp) > 0) {
328 "I/O requests for %s in queue, can't destroy it now.",
336 g_mirror_destroy_consumer(void *arg, int flags __unused)
338 struct g_consumer *cp;
343 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", cp->provider->name);
345 g_destroy_consumer(cp);
349 g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp)
351 struct g_provider *pp;
357 if (g_mirror_is_busy(sc, cp))
362 if ((pp->geom->flags & G_GEOM_WITHER) == 0)
365 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr,
366 -cp->acw, -cp->ace, 0);
367 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
368 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
371 * After retaste event was send (inside g_access()), we can send
372 * event to detach and destroy consumer.
373 * A class, which has consumer to the given provider connected
374 * will not receive retaste event for the provider.
375 * This is the way how I ignore retaste events when I close
376 * consumers opened for write: I detach and destroy consumer
377 * after retaste event is sent.
379 g_post_event(g_mirror_destroy_consumer, cp, M_WAITOK, NULL);
382 G_MIRROR_DEBUG(1, "Consumer %s destroyed.", pp->name);
384 g_destroy_consumer(cp);
388 g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp)
390 struct g_consumer *cp;
393 g_topology_assert_not();
394 KASSERT(disk->d_consumer == NULL,
395 ("Disk already connected (device %s).", disk->d_softc->sc_name));
398 cp = g_new_consumer(disk->d_softc->sc_geom);
399 cp->flags |= G_CF_DIRECT_RECEIVE;
400 error = g_attach(cp, pp);
402 g_destroy_consumer(cp);
406 error = g_access(cp, 1, 1, 1);
409 g_destroy_consumer(cp);
411 G_MIRROR_DEBUG(0, "Cannot open consumer %s (error=%d).",
416 disk->d_consumer = cp;
417 disk->d_consumer->private = disk;
418 disk->d_consumer->index = 0;
420 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk));
425 g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp)
432 if (cp->provider != NULL)
433 g_mirror_kill_consumer(sc, cp);
435 g_destroy_consumer(cp);
439 * Initialize disk. This means allocate memory, create consumer, attach it
440 * to the provider and open access (r1w1e1) to it.
442 static struct g_mirror_disk *
443 g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp,
444 struct g_mirror_metadata *md, int *errorp)
446 struct g_mirror_disk *disk;
449 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO);
455 error = g_mirror_connect_disk(disk, pp);
458 disk->d_id = md->md_did;
459 disk->d_state = G_MIRROR_DISK_STATE_NONE;
460 disk->d_priority = md->md_priority;
461 disk->d_flags = md->md_dflags;
462 error = g_getattr("GEOM::candelete", disk->d_consumer, &i);
463 if (error == 0 && i != 0)
464 disk->d_flags |= G_MIRROR_DISK_FLAG_CANDELETE;
465 if (md->md_provider[0] != '\0')
466 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED;
467 disk->d_sync.ds_consumer = NULL;
468 disk->d_sync.ds_offset = md->md_sync_offset;
469 disk->d_sync.ds_offset_done = md->md_sync_offset;
470 disk->d_genid = md->md_genid;
471 disk->d_sync.ds_syncid = md->md_syncid;
479 free(disk, M_MIRROR);
484 g_mirror_destroy_disk(struct g_mirror_disk *disk)
486 struct g_mirror_softc *sc;
488 g_topology_assert_not();
490 sx_assert(&sc->sc_lock, SX_XLOCKED);
492 LIST_REMOVE(disk, d_next);
493 g_mirror_event_cancel(disk);
494 if (sc->sc_hint == disk)
496 switch (disk->d_state) {
497 case G_MIRROR_DISK_STATE_SYNCHRONIZING:
498 g_mirror_sync_stop(disk, 1);
500 case G_MIRROR_DISK_STATE_NEW:
501 case G_MIRROR_DISK_STATE_STALE:
502 case G_MIRROR_DISK_STATE_ACTIVE:
504 g_mirror_disconnect_consumer(sc, disk->d_consumer);
506 free(disk, M_MIRROR);
509 KASSERT(0 == 1, ("Wrong disk state (%s, %s).",
510 g_mirror_get_diskname(disk),
511 g_mirror_disk_state2str(disk->d_state)));
516 g_mirror_destroy_device(struct g_mirror_softc *sc)
518 struct g_mirror_disk *disk;
519 struct g_mirror_event *ep;
521 struct g_consumer *cp, *tmpcp;
523 g_topology_assert_not();
524 sx_assert(&sc->sc_lock, SX_XLOCKED);
527 if (sc->sc_provider != NULL)
528 g_mirror_destroy_provider(sc);
529 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL;
530 disk = LIST_FIRST(&sc->sc_disks)) {
531 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY;
532 g_mirror_update_metadata(disk);
533 g_mirror_destroy_disk(disk);
535 while ((ep = g_mirror_event_get(sc)) != NULL) {
536 g_mirror_event_remove(sc, ep);
537 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0)
538 g_mirror_event_free(ep);
540 ep->e_error = ECANCELED;
541 ep->e_flags |= G_MIRROR_EVENT_DONE;
542 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep);
543 mtx_lock(&sc->sc_events_mtx);
545 mtx_unlock(&sc->sc_events_mtx);
548 callout_drain(&sc->sc_callout);
551 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) {
552 g_mirror_disconnect_consumer(sc, cp);
554 g_wither_geom(sc->sc_sync.ds_geom, ENXIO);
555 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name);
556 g_wither_geom(gp, ENXIO);
558 mtx_destroy(&sc->sc_queue_mtx);
559 mtx_destroy(&sc->sc_events_mtx);
560 mtx_destroy(&sc->sc_done_mtx);
561 sx_xunlock(&sc->sc_lock);
562 sx_destroy(&sc->sc_lock);
566 g_mirror_orphan(struct g_consumer *cp)
568 struct g_mirror_disk *disk;
575 disk->d_softc->sc_bump_id |= G_MIRROR_BUMP_SYNCID;
576 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED,
577 G_MIRROR_EVENT_DONTWAIT);
581 * Function should return the next active disk on the list.
582 * It is possible that it will be the same disk as given.
583 * If there are no active disks on list, NULL is returned.
585 static __inline struct g_mirror_disk *
586 g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk)
588 struct g_mirror_disk *dp;
590 for (dp = LIST_NEXT(disk, d_next); dp != disk;
591 dp = LIST_NEXT(dp, d_next)) {
593 dp = LIST_FIRST(&sc->sc_disks);
594 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE)
597 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE)
602 static struct g_mirror_disk *
603 g_mirror_get_disk(struct g_mirror_softc *sc)
605 struct g_mirror_disk *disk;
607 if (sc->sc_hint == NULL) {
608 sc->sc_hint = LIST_FIRST(&sc->sc_disks);
609 if (sc->sc_hint == NULL)
613 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) {
614 disk = g_mirror_find_next(sc, disk);
618 sc->sc_hint = g_mirror_find_next(sc, disk);
623 g_mirror_write_metadata(struct g_mirror_disk *disk,
624 struct g_mirror_metadata *md)
626 struct g_mirror_softc *sc;
627 struct g_consumer *cp;
628 off_t offset, length;
632 g_topology_assert_not();
634 sx_assert(&sc->sc_lock, SX_LOCKED);
636 cp = disk->d_consumer;
637 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name));
638 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name));
639 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
640 ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr,
642 length = cp->provider->sectorsize;
643 offset = cp->provider->mediasize - length;
644 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO);
646 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0) {
648 * Handle the case, when the size of parent provider reduced.
650 if (offset < md->md_mediasize)
653 mirror_metadata_encode(md, sector);
656 error = g_write_data(cp, offset, sector, length);
657 free(sector, M_MIRROR);
659 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) {
660 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN;
661 G_MIRROR_DEBUG(0, "Cannot write metadata on %s "
662 "(device=%s, error=%d).",
663 g_mirror_get_diskname(disk), sc->sc_name, error);
665 G_MIRROR_DEBUG(1, "Cannot write metadata on %s "
666 "(device=%s, error=%d).",
667 g_mirror_get_diskname(disk), sc->sc_name, error);
669 if (g_mirror_disconnect_on_failure &&
670 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1) {
671 sc->sc_bump_id |= G_MIRROR_BUMP_GENID;
672 g_mirror_event_send(disk,
673 G_MIRROR_DISK_STATE_DISCONNECTED,
674 G_MIRROR_EVENT_DONTWAIT);
681 g_mirror_clear_metadata(struct g_mirror_disk *disk)
685 g_topology_assert_not();
686 sx_assert(&disk->d_softc->sc_lock, SX_LOCKED);
688 error = g_mirror_write_metadata(disk, NULL);
690 G_MIRROR_DEBUG(2, "Metadata on %s cleared.",
691 g_mirror_get_diskname(disk));
694 "Cannot clear metadata on disk %s (error=%d).",
695 g_mirror_get_diskname(disk), error);
701 g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk,
702 struct g_mirror_metadata *md)
705 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic));
706 md->md_version = G_MIRROR_VERSION;
707 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name));
708 md->md_mid = sc->sc_id;
709 md->md_all = sc->sc_ndisks;
710 md->md_slice = sc->sc_slice;
711 md->md_balance = sc->sc_balance;
712 md->md_genid = sc->sc_genid;
713 md->md_mediasize = sc->sc_mediasize;
714 md->md_sectorsize = sc->sc_sectorsize;
715 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK);
716 bzero(md->md_provider, sizeof(md->md_provider));
718 md->md_did = arc4random();
722 md->md_sync_offset = 0;
725 md->md_did = disk->d_id;
726 md->md_priority = disk->d_priority;
727 md->md_syncid = disk->d_sync.ds_syncid;
728 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK);
729 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING)
730 md->md_sync_offset = disk->d_sync.ds_offset_done;
732 md->md_sync_offset = 0;
733 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) {
734 strlcpy(md->md_provider,
735 disk->d_consumer->provider->name,
736 sizeof(md->md_provider));
738 md->md_provsize = disk->d_consumer->provider->mediasize;
743 g_mirror_update_metadata(struct g_mirror_disk *disk)
745 struct g_mirror_softc *sc;
746 struct g_mirror_metadata md;
749 g_topology_assert_not();
751 sx_assert(&sc->sc_lock, SX_LOCKED);
753 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WIPE) == 0)
754 g_mirror_fill_metadata(sc, disk, &md);
755 error = g_mirror_write_metadata(disk, &md);
757 G_MIRROR_DEBUG(2, "Metadata on %s updated.",
758 g_mirror_get_diskname(disk));
761 "Cannot update metadata on disk %s (error=%d).",
762 g_mirror_get_diskname(disk), error);
767 g_mirror_bump_syncid(struct g_mirror_softc *sc)
769 struct g_mirror_disk *disk;
771 g_topology_assert_not();
772 sx_assert(&sc->sc_lock, SX_XLOCKED);
773 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0,
774 ("%s called with no active disks (device=%s).", __func__,
778 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name,
780 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
781 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE ||
782 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) {
783 disk->d_sync.ds_syncid = sc->sc_syncid;
784 g_mirror_update_metadata(disk);
790 g_mirror_bump_genid(struct g_mirror_softc *sc)
792 struct g_mirror_disk *disk;
794 g_topology_assert_not();
795 sx_assert(&sc->sc_lock, SX_XLOCKED);
796 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0,
797 ("%s called with no active disks (device=%s).", __func__,
801 G_MIRROR_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name,
803 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
804 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE ||
805 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) {
806 disk->d_genid = sc->sc_genid;
807 g_mirror_update_metadata(disk);
813 g_mirror_idle(struct g_mirror_softc *sc, int acw)
815 struct g_mirror_disk *disk;
818 g_topology_assert_not();
819 sx_assert(&sc->sc_lock, SX_XLOCKED);
821 if (sc->sc_provider == NULL)
823 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0)
827 if (sc->sc_writes > 0)
829 if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) {
830 timeout = g_mirror_idletime - (time_uptime - sc->sc_last_write);
831 if (!g_mirror_shutdown && timeout > 0)
835 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
836 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE)
838 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.",
839 g_mirror_get_diskname(disk), sc->sc_name);
840 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY;
841 g_mirror_update_metadata(disk);
847 g_mirror_unidle(struct g_mirror_softc *sc)
849 struct g_mirror_disk *disk;
851 g_topology_assert_not();
852 sx_assert(&sc->sc_lock, SX_XLOCKED);
854 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0)
857 sc->sc_last_write = time_uptime;
858 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
859 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE)
861 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.",
862 g_mirror_get_diskname(disk), sc->sc_name);
863 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY;
864 g_mirror_update_metadata(disk);
869 g_mirror_flush_done(struct bio *bp)
871 struct g_mirror_softc *sc;
874 pbp = bp->bio_parent;
875 sc = pbp->bio_to->geom->softc;
876 mtx_lock(&sc->sc_done_mtx);
877 if (pbp->bio_error == 0)
878 pbp->bio_error = bp->bio_error;
879 pbp->bio_completed += bp->bio_completed;
881 if (pbp->bio_children == pbp->bio_inbed) {
882 mtx_unlock(&sc->sc_done_mtx);
883 g_io_deliver(pbp, pbp->bio_error);
885 mtx_unlock(&sc->sc_done_mtx);
890 g_mirror_done(struct bio *bp)
892 struct g_mirror_softc *sc;
894 sc = bp->bio_from->geom->softc;
895 bp->bio_cflags = G_MIRROR_BIO_FLAG_REGULAR;
896 mtx_lock(&sc->sc_queue_mtx);
897 bioq_disksort(&sc->sc_queue, bp);
898 mtx_unlock(&sc->sc_queue_mtx);
903 g_mirror_regular_request(struct bio *bp)
905 struct g_mirror_softc *sc;
906 struct g_mirror_disk *disk;
909 g_topology_assert_not();
911 pbp = bp->bio_parent;
912 sc = pbp->bio_to->geom->softc;
913 bp->bio_from->index--;
914 if (bp->bio_cmd == BIO_WRITE)
916 disk = bp->bio_from->private;
919 g_mirror_kill_consumer(sc, bp->bio_from);
924 KASSERT(pbp->bio_inbed <= pbp->bio_children,
925 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed,
927 if (bp->bio_error == 0 && pbp->bio_error == 0) {
928 G_MIRROR_LOGREQ(3, bp, "Request delivered.");
930 if (pbp->bio_children == pbp->bio_inbed) {
931 G_MIRROR_LOGREQ(3, pbp, "Request delivered.");
932 pbp->bio_completed = pbp->bio_length;
933 if (pbp->bio_cmd == BIO_WRITE ||
934 pbp->bio_cmd == BIO_DELETE) {
935 bioq_remove(&sc->sc_inflight, pbp);
936 /* Release delayed sync requests if possible. */
937 g_mirror_sync_release(sc);
939 g_io_deliver(pbp, pbp->bio_error);
942 } else if (bp->bio_error != 0) {
943 if (pbp->bio_error == 0)
944 pbp->bio_error = bp->bio_error;
946 if ((disk->d_flags & G_MIRROR_DISK_FLAG_BROKEN) == 0) {
947 disk->d_flags |= G_MIRROR_DISK_FLAG_BROKEN;
948 G_MIRROR_LOGREQ(0, bp,
949 "Request failed (error=%d).",
952 G_MIRROR_LOGREQ(1, bp,
953 "Request failed (error=%d).",
956 if (g_mirror_disconnect_on_failure &&
957 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 1)
959 sc->sc_bump_id |= G_MIRROR_BUMP_GENID;
960 g_mirror_event_send(disk,
961 G_MIRROR_DISK_STATE_DISCONNECTED,
962 G_MIRROR_EVENT_DONTWAIT);
965 switch (pbp->bio_cmd) {
975 switch (pbp->bio_cmd) {
977 if (pbp->bio_inbed < pbp->bio_children)
979 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 1)
980 g_io_deliver(pbp, pbp->bio_error);
983 mtx_lock(&sc->sc_queue_mtx);
984 bioq_disksort(&sc->sc_queue, pbp);
985 mtx_unlock(&sc->sc_queue_mtx);
986 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc);
992 if (pbp->bio_children == 0) {
994 * All requests failed.
996 } else if (pbp->bio_inbed < pbp->bio_children) {
999 } else if (pbp->bio_children == pbp->bio_inbed) {
1000 /* Some requests succeeded. */
1002 pbp->bio_completed = pbp->bio_length;
1004 bioq_remove(&sc->sc_inflight, pbp);
1005 /* Release delayed sync requests if possible. */
1006 g_mirror_sync_release(sc);
1007 g_io_deliver(pbp, pbp->bio_error);
1010 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd));
1016 g_mirror_sync_done(struct bio *bp)
1018 struct g_mirror_softc *sc;
1020 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered.");
1021 sc = bp->bio_from->geom->softc;
1022 bp->bio_cflags = G_MIRROR_BIO_FLAG_SYNC;
1023 mtx_lock(&sc->sc_queue_mtx);
1024 bioq_disksort(&sc->sc_queue, bp);
1025 mtx_unlock(&sc->sc_queue_mtx);
1030 g_mirror_kernel_dump(struct bio *bp)
1032 struct g_mirror_softc *sc;
1033 struct g_mirror_disk *disk;
1035 struct g_kerneldump *gkd;
1038 * We configure dumping to the first component, because this component
1039 * will be used for reading with 'prefer' balance algorithm.
1040 * If the component with the higest priority is currently disconnected
1041 * we will not be able to read the dump after the reboot if it will be
1042 * connected and synchronized later. Can we do something better?
1044 sc = bp->bio_to->geom->softc;
1045 disk = LIST_FIRST(&sc->sc_disks);
1047 gkd = (struct g_kerneldump *)bp->bio_data;
1048 if (gkd->length > bp->bio_to->mediasize)
1049 gkd->length = bp->bio_to->mediasize;
1050 cbp = g_clone_bio(bp);
1052 g_io_deliver(bp, ENOMEM);
1055 cbp->bio_done = g_std_done;
1056 g_io_request(cbp, disk->d_consumer);
1057 G_MIRROR_DEBUG(1, "Kernel dump will go to %s.",
1058 g_mirror_get_diskname(disk));
1062 g_mirror_flush(struct g_mirror_softc *sc, struct bio *bp)
1064 struct bio_queue_head queue;
1065 struct g_mirror_disk *disk;
1066 struct g_consumer *cp;
1070 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
1071 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE)
1073 cbp = g_clone_bio(bp);
1075 while ((cbp = bioq_takefirst(&queue)) != NULL)
1077 if (bp->bio_error == 0)
1078 bp->bio_error = ENOMEM;
1079 g_io_deliver(bp, bp->bio_error);
1082 bioq_insert_tail(&queue, cbp);
1083 cbp->bio_done = g_mirror_flush_done;
1084 cbp->bio_caller1 = disk;
1085 cbp->bio_to = disk->d_consumer->provider;
1087 while ((cbp = bioq_takefirst(&queue)) != NULL) {
1088 G_MIRROR_LOGREQ(3, cbp, "Sending request.");
1089 disk = cbp->bio_caller1;
1090 cbp->bio_caller1 = NULL;
1091 cp = disk->d_consumer;
1092 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
1093 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name,
1094 cp->acr, cp->acw, cp->ace));
1095 g_io_request(cbp, disk->d_consumer);
1100 g_mirror_start(struct bio *bp)
1102 struct g_mirror_softc *sc;
1104 sc = bp->bio_to->geom->softc;
1106 * If sc == NULL or there are no valid disks, provider's error
1107 * should be set and g_mirror_start() should not be called at all.
1109 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING,
1110 ("Provider's error should be set (error=%d)(mirror=%s).",
1111 bp->bio_to->error, bp->bio_to->name));
1112 G_MIRROR_LOGREQ(3, bp, "Request received.");
1114 switch (bp->bio_cmd) {
1120 g_mirror_flush(sc, bp);
1123 if (g_handleattr_int(bp, "GEOM::candelete", 1))
1125 else if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) {
1126 g_mirror_kernel_dump(bp);
1131 g_io_deliver(bp, EOPNOTSUPP);
1134 mtx_lock(&sc->sc_queue_mtx);
1135 bioq_disksort(&sc->sc_queue, bp);
1136 mtx_unlock(&sc->sc_queue_mtx);
1137 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc);
1142 * Return TRUE if the given request is colliding with a in-progress
1143 * synchronization request.
1146 g_mirror_sync_collision(struct g_mirror_softc *sc, struct bio *bp)
1148 struct g_mirror_disk *disk;
1150 off_t rstart, rend, sstart, send;
1153 if (sc->sc_sync.ds_ndisks == 0)
1155 rstart = bp->bio_offset;
1156 rend = bp->bio_offset + bp->bio_length;
1157 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
1158 if (disk->d_state != G_MIRROR_DISK_STATE_SYNCHRONIZING)
1160 for (i = 0; i < g_mirror_syncreqs; i++) {
1161 sbp = disk->d_sync.ds_bios[i];
1164 sstart = sbp->bio_offset;
1165 send = sbp->bio_offset + sbp->bio_length;
1166 if (rend > sstart && rstart < send)
1174 * Return TRUE if the given sync request is colliding with a in-progress regular
1178 g_mirror_regular_collision(struct g_mirror_softc *sc, struct bio *sbp)
1180 off_t rstart, rend, sstart, send;
1183 if (sc->sc_sync.ds_ndisks == 0)
1185 sstart = sbp->bio_offset;
1186 send = sbp->bio_offset + sbp->bio_length;
1187 TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) {
1188 rstart = bp->bio_offset;
1189 rend = bp->bio_offset + bp->bio_length;
1190 if (rend > sstart && rstart < send)
1197 * Puts request onto delayed queue.
1200 g_mirror_regular_delay(struct g_mirror_softc *sc, struct bio *bp)
1203 G_MIRROR_LOGREQ(2, bp, "Delaying request.");
1204 bioq_insert_head(&sc->sc_regular_delayed, bp);
1208 * Puts synchronization request onto delayed queue.
1211 g_mirror_sync_delay(struct g_mirror_softc *sc, struct bio *bp)
1214 G_MIRROR_LOGREQ(2, bp, "Delaying synchronization request.");
1215 bioq_insert_tail(&sc->sc_sync_delayed, bp);
1219 * Releases delayed regular requests which don't collide anymore with sync
1223 g_mirror_regular_release(struct g_mirror_softc *sc)
1225 struct bio *bp, *bp2;
1227 TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) {
1228 if (g_mirror_sync_collision(sc, bp))
1230 bioq_remove(&sc->sc_regular_delayed, bp);
1231 G_MIRROR_LOGREQ(2, bp, "Releasing delayed request (%p).", bp);
1232 mtx_lock(&sc->sc_queue_mtx);
1233 bioq_insert_head(&sc->sc_queue, bp);
1236 * wakeup() is not needed, because this function is called from
1237 * the worker thread.
1239 wakeup(&sc->sc_queue);
1241 mtx_unlock(&sc->sc_queue_mtx);
1246 * Releases delayed sync requests which don't collide anymore with regular
1250 g_mirror_sync_release(struct g_mirror_softc *sc)
1252 struct bio *bp, *bp2;
1254 TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) {
1255 if (g_mirror_regular_collision(sc, bp))
1257 bioq_remove(&sc->sc_sync_delayed, bp);
1258 G_MIRROR_LOGREQ(2, bp,
1259 "Releasing delayed synchronization request.");
1260 g_io_request(bp, bp->bio_from);
1265 * Handle synchronization requests.
1266 * Every synchronization request is two-steps process: first, READ request is
1267 * send to active provider and then WRITE request (with read data) to the provider
1268 * beeing synchronized. When WRITE is finished, new synchronization request is
1272 g_mirror_sync_request(struct bio *bp)
1274 struct g_mirror_softc *sc;
1275 struct g_mirror_disk *disk;
1277 bp->bio_from->index--;
1278 sc = bp->bio_from->geom->softc;
1279 disk = bp->bio_from->private;
1281 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */
1283 g_mirror_kill_consumer(sc, bp->bio_from);
1284 g_topology_unlock();
1285 free(bp->bio_data, M_MIRROR);
1287 sx_xlock(&sc->sc_lock);
1292 * Synchronization request.
1294 switch (bp->bio_cmd) {
1297 struct g_consumer *cp;
1299 if (bp->bio_error != 0) {
1300 G_MIRROR_LOGREQ(0, bp,
1301 "Synchronization request failed (error=%d).",
1306 G_MIRROR_LOGREQ(3, bp,
1307 "Synchronization request half-finished.");
1308 bp->bio_cmd = BIO_WRITE;
1310 cp = disk->d_consumer;
1311 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
1312 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name,
1313 cp->acr, cp->acw, cp->ace));
1315 g_io_request(bp, cp);
1320 struct g_mirror_disk_sync *sync;
1325 if (bp->bio_error != 0) {
1326 G_MIRROR_LOGREQ(0, bp,
1327 "Synchronization request failed (error=%d).",
1330 sc->sc_bump_id |= G_MIRROR_BUMP_GENID;
1331 g_mirror_event_send(disk,
1332 G_MIRROR_DISK_STATE_DISCONNECTED,
1333 G_MIRROR_EVENT_DONTWAIT);
1336 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished.");
1337 sync = &disk->d_sync;
1338 if (sync->ds_offset >= sc->sc_mediasize ||
1339 sync->ds_consumer == NULL ||
1340 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) {
1341 /* Don't send more synchronization requests. */
1342 sync->ds_inflight--;
1343 if (sync->ds_bios != NULL) {
1344 i = (int)(uintptr_t)bp->bio_caller1;
1345 sync->ds_bios[i] = NULL;
1347 free(bp->bio_data, M_MIRROR);
1349 if (sync->ds_inflight > 0)
1351 if (sync->ds_consumer == NULL ||
1352 (sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) {
1355 /* Disk up-to-date, activate it. */
1356 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE,
1357 G_MIRROR_EVENT_DONTWAIT);
1361 /* Send next synchronization request. */
1362 data = bp->bio_data;
1363 bzero(bp, sizeof(*bp));
1364 bp->bio_cmd = BIO_READ;
1365 bp->bio_offset = sync->ds_offset;
1366 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset);
1367 sync->ds_offset += bp->bio_length;
1368 bp->bio_done = g_mirror_sync_done;
1369 bp->bio_data = data;
1370 bp->bio_from = sync->ds_consumer;
1371 bp->bio_to = sc->sc_provider;
1372 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request.");
1373 sync->ds_consumer->index++;
1375 * Delay the request if it is colliding with a regular request.
1377 if (g_mirror_regular_collision(sc, bp))
1378 g_mirror_sync_delay(sc, bp);
1380 g_io_request(bp, sync->ds_consumer);
1382 /* Release delayed requests if possible. */
1383 g_mirror_regular_release(sc);
1385 /* Find the smallest offset */
1386 offset = sc->sc_mediasize;
1387 for (i = 0; i < g_mirror_syncreqs; i++) {
1388 bp = sync->ds_bios[i];
1389 if (bp->bio_offset < offset)
1390 offset = bp->bio_offset;
1392 if (sync->ds_offset_done + (MAXPHYS * 100) < offset) {
1393 /* Update offset_done on every 100 blocks. */
1394 sync->ds_offset_done = offset;
1395 g_mirror_update_metadata(disk);
1400 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)",
1401 bp->bio_cmd, sc->sc_name));
1407 g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp)
1409 struct g_mirror_disk *disk;
1410 struct g_consumer *cp;
1413 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
1414 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE)
1418 if (bp->bio_error == 0)
1419 bp->bio_error = ENXIO;
1420 g_io_deliver(bp, bp->bio_error);
1423 cbp = g_clone_bio(bp);
1425 if (bp->bio_error == 0)
1426 bp->bio_error = ENOMEM;
1427 g_io_deliver(bp, bp->bio_error);
1431 * Fill in the component buf structure.
1433 cp = disk->d_consumer;
1434 cbp->bio_done = g_mirror_done;
1435 cbp->bio_to = cp->provider;
1436 G_MIRROR_LOGREQ(3, cbp, "Sending request.");
1437 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
1438 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr,
1441 g_io_request(cbp, cp);
1445 g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp)
1447 struct g_mirror_disk *disk;
1448 struct g_consumer *cp;
1451 disk = g_mirror_get_disk(sc);
1453 if (bp->bio_error == 0)
1454 bp->bio_error = ENXIO;
1455 g_io_deliver(bp, bp->bio_error);
1458 cbp = g_clone_bio(bp);
1460 if (bp->bio_error == 0)
1461 bp->bio_error = ENOMEM;
1462 g_io_deliver(bp, bp->bio_error);
1466 * Fill in the component buf structure.
1468 cp = disk->d_consumer;
1469 cbp->bio_done = g_mirror_done;
1470 cbp->bio_to = cp->provider;
1471 G_MIRROR_LOGREQ(3, cbp, "Sending request.");
1472 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
1473 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr,
1476 g_io_request(cbp, cp);
1479 #define TRACK_SIZE (1 * 1024 * 1024)
1480 #define LOAD_SCALE 256
1481 #define ABS(x) (((x) >= 0) ? (x) : (-(x)))
1484 g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp)
1486 struct g_mirror_disk *disk, *dp;
1487 struct g_consumer *cp;
1491 /* Find a disk with the smallest load. */
1494 LIST_FOREACH(dp, &sc->sc_disks, d_next) {
1495 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE)
1498 /* If disk head is precisely in position - highly prefer it. */
1499 if (dp->d_last_offset == bp->bio_offset)
1500 prio -= 2 * LOAD_SCALE;
1502 /* If disk head is close to position - prefer it. */
1503 if (ABS(dp->d_last_offset - bp->bio_offset) < TRACK_SIZE)
1504 prio -= 1 * LOAD_SCALE;
1510 KASSERT(disk != NULL, ("NULL disk for %s.", sc->sc_name));
1511 cbp = g_clone_bio(bp);
1513 if (bp->bio_error == 0)
1514 bp->bio_error = ENOMEM;
1515 g_io_deliver(bp, bp->bio_error);
1519 * Fill in the component buf structure.
1521 cp = disk->d_consumer;
1522 cbp->bio_done = g_mirror_done;
1523 cbp->bio_to = cp->provider;
1524 G_MIRROR_LOGREQ(3, cbp, "Sending request.");
1525 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
1526 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr,
1529 /* Remember last head position */
1530 disk->d_last_offset = bp->bio_offset + bp->bio_length;
1532 LIST_FOREACH(dp, &sc->sc_disks, d_next) {
1533 dp->load = (dp->d_consumer->index * LOAD_SCALE +
1536 g_io_request(cbp, cp);
1540 g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp)
1542 struct bio_queue_head queue;
1543 struct g_mirror_disk *disk;
1544 struct g_consumer *cp;
1546 off_t left, mod, offset, slice;
1550 if (bp->bio_length <= sc->sc_slice) {
1551 g_mirror_request_round_robin(sc, bp);
1554 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE);
1555 slice = bp->bio_length / ndisks;
1556 mod = slice % sc->sc_provider->sectorsize;
1558 slice += sc->sc_provider->sectorsize - mod;
1560 * Allocate all bios before sending any request, so we can
1561 * return ENOMEM in nice and clean way.
1563 left = bp->bio_length;
1564 offset = bp->bio_offset;
1565 data = bp->bio_data;
1567 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
1568 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE)
1570 cbp = g_clone_bio(bp);
1572 while ((cbp = bioq_takefirst(&queue)) != NULL)
1573 bioq_remove(&queue, cbp);
1574 if (bp->bio_error == 0)
1575 bp->bio_error = ENOMEM;
1576 g_io_deliver(bp, bp->bio_error);
1579 bioq_insert_tail(&queue, cbp);
1580 cbp->bio_done = g_mirror_done;
1581 cbp->bio_caller1 = disk;
1582 cbp->bio_to = disk->d_consumer->provider;
1583 cbp->bio_offset = offset;
1584 cbp->bio_data = data;
1585 cbp->bio_length = MIN(left, slice);
1586 left -= cbp->bio_length;
1589 offset += cbp->bio_length;
1590 data += cbp->bio_length;
1592 while ((cbp = bioq_takefirst(&queue)) != NULL) {
1593 G_MIRROR_LOGREQ(3, cbp, "Sending request.");
1594 disk = cbp->bio_caller1;
1595 cbp->bio_caller1 = NULL;
1596 cp = disk->d_consumer;
1597 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
1598 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name,
1599 cp->acr, cp->acw, cp->ace));
1600 disk->d_consumer->index++;
1601 g_io_request(cbp, disk->d_consumer);
1606 g_mirror_register_request(struct bio *bp)
1608 struct g_mirror_softc *sc;
1610 sc = bp->bio_to->geom->softc;
1611 switch (bp->bio_cmd) {
1613 switch (sc->sc_balance) {
1614 case G_MIRROR_BALANCE_LOAD:
1615 g_mirror_request_load(sc, bp);
1617 case G_MIRROR_BALANCE_PREFER:
1618 g_mirror_request_prefer(sc, bp);
1620 case G_MIRROR_BALANCE_ROUND_ROBIN:
1621 g_mirror_request_round_robin(sc, bp);
1623 case G_MIRROR_BALANCE_SPLIT:
1624 g_mirror_request_split(sc, bp);
1631 struct g_mirror_disk *disk;
1632 struct g_mirror_disk_sync *sync;
1633 struct bio_queue_head queue;
1634 struct g_consumer *cp;
1638 * Delay the request if it is colliding with a synchronization
1641 if (g_mirror_sync_collision(sc, bp)) {
1642 g_mirror_regular_delay(sc, bp);
1647 g_mirror_unidle(sc);
1649 sc->sc_last_write = time_uptime;
1652 * Allocate all bios before sending any request, so we can
1653 * return ENOMEM in nice and clean way.
1656 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
1657 sync = &disk->d_sync;
1658 switch (disk->d_state) {
1659 case G_MIRROR_DISK_STATE_ACTIVE:
1661 case G_MIRROR_DISK_STATE_SYNCHRONIZING:
1662 if (bp->bio_offset >= sync->ds_offset)
1668 if (bp->bio_cmd == BIO_DELETE &&
1669 (disk->d_flags & G_MIRROR_DISK_FLAG_CANDELETE) == 0)
1671 cbp = g_clone_bio(bp);
1673 while ((cbp = bioq_takefirst(&queue)) != NULL)
1675 if (bp->bio_error == 0)
1676 bp->bio_error = ENOMEM;
1677 g_io_deliver(bp, bp->bio_error);
1680 bioq_insert_tail(&queue, cbp);
1681 cbp->bio_done = g_mirror_done;
1682 cp = disk->d_consumer;
1683 cbp->bio_caller1 = cp;
1684 cbp->bio_to = cp->provider;
1685 KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
1686 ("Consumer %s not opened (r%dw%de%d).",
1687 cp->provider->name, cp->acr, cp->acw, cp->ace));
1689 while ((cbp = bioq_takefirst(&queue)) != NULL) {
1690 G_MIRROR_LOGREQ(3, cbp, "Sending request.");
1691 cp = cbp->bio_caller1;
1692 cbp->bio_caller1 = NULL;
1695 g_io_request(cbp, cp);
1698 * Put request onto inflight queue, so we can check if new
1699 * synchronization requests don't collide with it.
1701 bioq_insert_tail(&sc->sc_inflight, bp);
1703 * Bump syncid on first write.
1705 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0) {
1706 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID;
1707 g_mirror_bump_syncid(sc);
1712 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)",
1713 bp->bio_cmd, sc->sc_name));
1719 g_mirror_can_destroy(struct g_mirror_softc *sc)
1722 struct g_consumer *cp;
1724 g_topology_assert();
1726 if (gp->softc == NULL)
1728 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_TASTING) != 0)
1730 LIST_FOREACH(cp, &gp->consumer, consumer) {
1731 if (g_mirror_is_busy(sc, cp))
1734 gp = sc->sc_sync.ds_geom;
1735 LIST_FOREACH(cp, &gp->consumer, consumer) {
1736 if (g_mirror_is_busy(sc, cp))
1739 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.",
1745 g_mirror_try_destroy(struct g_mirror_softc *sc)
1748 if (sc->sc_rootmount != NULL) {
1749 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__,
1751 root_mount_rel(sc->sc_rootmount);
1752 sc->sc_rootmount = NULL;
1755 if (!g_mirror_can_destroy(sc)) {
1756 g_topology_unlock();
1759 sc->sc_geom->softc = NULL;
1760 sc->sc_sync.ds_geom->softc = NULL;
1761 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) {
1762 g_topology_unlock();
1763 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__,
1765 /* Unlock sc_lock here, as it can be destroyed after wakeup. */
1766 sx_xunlock(&sc->sc_lock);
1767 wakeup(&sc->sc_worker);
1768 sc->sc_worker = NULL;
1770 g_topology_unlock();
1771 g_mirror_destroy_device(sc);
1781 g_mirror_worker(void *arg)
1783 struct g_mirror_softc *sc;
1784 struct g_mirror_event *ep;
1789 thread_lock(curthread);
1790 sched_prio(curthread, PRIBIO);
1791 thread_unlock(curthread);
1793 sx_xlock(&sc->sc_lock);
1795 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__);
1797 * First take a look at events.
1798 * This is important to handle events before any I/O requests.
1800 ep = g_mirror_event_get(sc);
1802 g_mirror_event_remove(sc, ep);
1803 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) {
1804 /* Update only device status. */
1806 "Running event for device %s.",
1809 g_mirror_update_device(sc, 1);
1811 /* Update disk status. */
1812 G_MIRROR_DEBUG(3, "Running event for disk %s.",
1813 g_mirror_get_diskname(ep->e_disk));
1814 ep->e_error = g_mirror_update_disk(ep->e_disk,
1816 if (ep->e_error == 0)
1817 g_mirror_update_device(sc, 0);
1819 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) {
1820 KASSERT(ep->e_error == 0,
1821 ("Error cannot be handled."));
1822 g_mirror_event_free(ep);
1824 ep->e_flags |= G_MIRROR_EVENT_DONE;
1825 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__,
1827 mtx_lock(&sc->sc_events_mtx);
1829 mtx_unlock(&sc->sc_events_mtx);
1832 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) {
1833 if (g_mirror_try_destroy(sc)) {
1834 curthread->td_pflags &= ~TDP_GEOM;
1835 G_MIRROR_DEBUG(1, "Thread exiting.");
1839 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__);
1843 * Check if we can mark array as CLEAN and if we can't take
1844 * how much seconds should we wait.
1846 timeout = g_mirror_idle(sc, -1);
1850 /* Get first request from the queue. */
1851 mtx_lock(&sc->sc_queue_mtx);
1852 bp = bioq_first(&sc->sc_queue);
1855 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) {
1856 mtx_unlock(&sc->sc_queue_mtx);
1857 if (g_mirror_try_destroy(sc)) {
1858 curthread->td_pflags &= ~TDP_GEOM;
1859 G_MIRROR_DEBUG(1, "Thread exiting.");
1862 mtx_lock(&sc->sc_queue_mtx);
1864 sx_xunlock(&sc->sc_lock);
1866 * XXX: We can miss an event here, because an event
1867 * can be added without sx-device-lock and without
1868 * mtx-queue-lock. Maybe I should just stop using
1869 * dedicated mutex for events synchronization and
1870 * stick with the queue lock?
1871 * The event will hang here until next I/O request
1872 * or next event is received.
1874 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w1",
1876 sx_xlock(&sc->sc_lock);
1877 G_MIRROR_DEBUG(5, "%s: I'm here 4.", __func__);
1880 bioq_remove(&sc->sc_queue, bp);
1881 mtx_unlock(&sc->sc_queue_mtx);
1883 if (bp->bio_from->geom == sc->sc_sync.ds_geom &&
1884 (bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) {
1885 g_mirror_sync_request(bp); /* READ */
1886 } else if (bp->bio_to != sc->sc_provider) {
1887 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0)
1888 g_mirror_regular_request(bp);
1889 else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0)
1890 g_mirror_sync_request(bp); /* WRITE */
1893 ("Invalid request cflags=0x%hhx to=%s.",
1894 bp->bio_cflags, bp->bio_to->name));
1897 g_mirror_register_request(bp);
1899 G_MIRROR_DEBUG(5, "%s: I'm here 9.", __func__);
1904 g_mirror_update_idle(struct g_mirror_softc *sc, struct g_mirror_disk *disk)
1907 sx_assert(&sc->sc_lock, SX_LOCKED);
1909 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) != 0)
1911 if (!sc->sc_idle && (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) {
1912 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.",
1913 g_mirror_get_diskname(disk), sc->sc_name);
1914 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY;
1915 } else if (sc->sc_idle &&
1916 (disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) {
1917 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.",
1918 g_mirror_get_diskname(disk), sc->sc_name);
1919 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY;
1924 g_mirror_sync_start(struct g_mirror_disk *disk)
1926 struct g_mirror_softc *sc;
1927 struct g_consumer *cp;
1931 g_topology_assert_not();
1933 sx_assert(&sc->sc_lock, SX_LOCKED);
1935 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING,
1936 ("Disk %s is not marked for synchronization.",
1937 g_mirror_get_diskname(disk)));
1938 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING,
1939 ("Device not in RUNNING state (%s, %u).", sc->sc_name,
1942 sx_xunlock(&sc->sc_lock);
1944 cp = g_new_consumer(sc->sc_sync.ds_geom);
1945 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
1946 error = g_attach(cp, sc->sc_provider);
1948 ("Cannot attach to %s (error=%d).", sc->sc_name, error));
1949 error = g_access(cp, 1, 0, 0);
1950 KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error));
1951 g_topology_unlock();
1952 sx_xlock(&sc->sc_lock);
1954 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name,
1955 g_mirror_get_diskname(disk));
1956 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOFAILSYNC) == 0)
1957 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY;
1958 KASSERT(disk->d_sync.ds_consumer == NULL,
1959 ("Sync consumer already exists (device=%s, disk=%s).",
1960 sc->sc_name, g_mirror_get_diskname(disk)));
1962 disk->d_sync.ds_consumer = cp;
1963 disk->d_sync.ds_consumer->private = disk;
1964 disk->d_sync.ds_consumer->index = 0;
1967 * Allocate memory for synchronization bios and initialize them.
1969 disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_mirror_syncreqs,
1970 M_MIRROR, M_WAITOK);
1971 for (i = 0; i < g_mirror_syncreqs; i++) {
1973 disk->d_sync.ds_bios[i] = bp;
1974 bp->bio_parent = NULL;
1975 bp->bio_cmd = BIO_READ;
1976 bp->bio_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK);
1978 bp->bio_offset = disk->d_sync.ds_offset;
1979 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset);
1980 disk->d_sync.ds_offset += bp->bio_length;
1981 bp->bio_done = g_mirror_sync_done;
1982 bp->bio_from = disk->d_sync.ds_consumer;
1983 bp->bio_to = sc->sc_provider;
1984 bp->bio_caller1 = (void *)(uintptr_t)i;
1987 /* Increase the number of disks in SYNCHRONIZING state. */
1988 sc->sc_sync.ds_ndisks++;
1989 /* Set the number of in-flight synchronization requests. */
1990 disk->d_sync.ds_inflight = g_mirror_syncreqs;
1993 * Fire off first synchronization requests.
1995 for (i = 0; i < g_mirror_syncreqs; i++) {
1996 bp = disk->d_sync.ds_bios[i];
1997 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request.");
1998 disk->d_sync.ds_consumer->index++;
2000 * Delay the request if it is colliding with a regular request.
2002 if (g_mirror_regular_collision(sc, bp))
2003 g_mirror_sync_delay(sc, bp);
2005 g_io_request(bp, disk->d_sync.ds_consumer);
2010 * Stop synchronization process.
2011 * type: 0 - synchronization finished
2012 * 1 - synchronization stopped
2015 g_mirror_sync_stop(struct g_mirror_disk *disk, int type)
2017 struct g_mirror_softc *sc;
2018 struct g_consumer *cp;
2020 g_topology_assert_not();
2022 sx_assert(&sc->sc_lock, SX_LOCKED);
2024 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING,
2025 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk),
2026 g_mirror_disk_state2str(disk->d_state)));
2027 if (disk->d_sync.ds_consumer == NULL)
2031 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.",
2032 sc->sc_name, g_mirror_get_diskname(disk));
2033 } else /* if (type == 1) */ {
2034 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.",
2035 sc->sc_name, g_mirror_get_diskname(disk));
2037 free(disk->d_sync.ds_bios, M_MIRROR);
2038 disk->d_sync.ds_bios = NULL;
2039 cp = disk->d_sync.ds_consumer;
2040 disk->d_sync.ds_consumer = NULL;
2041 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY;
2042 sc->sc_sync.ds_ndisks--;
2043 sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */
2045 g_mirror_kill_consumer(sc, cp);
2046 g_topology_unlock();
2047 sx_xlock(&sc->sc_lock);
2051 g_mirror_launch_provider(struct g_mirror_softc *sc)
2053 struct g_mirror_disk *disk;
2054 struct g_provider *pp, *dp;
2056 sx_assert(&sc->sc_lock, SX_LOCKED);
2059 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name);
2060 pp->flags |= G_PF_DIRECT_RECEIVE;
2061 pp->mediasize = sc->sc_mediasize;
2062 pp->sectorsize = sc->sc_sectorsize;
2064 pp->stripeoffset = 0;
2066 /* Splitting of unmapped BIO's could work but isn't implemented now */
2067 if (sc->sc_balance != G_MIRROR_BALANCE_SPLIT)
2068 pp->flags |= G_PF_ACCEPT_UNMAPPED;
2070 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
2071 if (disk->d_consumer && disk->d_consumer->provider) {
2072 dp = disk->d_consumer->provider;
2073 if (dp->stripesize > pp->stripesize) {
2074 pp->stripesize = dp->stripesize;
2075 pp->stripeoffset = dp->stripeoffset;
2077 /* A provider underneath us doesn't support unmapped */
2078 if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) {
2079 G_MIRROR_DEBUG(0, "Cancelling unmapped "
2080 "because of %s.", dp->name);
2081 pp->flags &= ~G_PF_ACCEPT_UNMAPPED;
2085 sc->sc_provider = pp;
2086 g_error_provider(pp, 0);
2087 g_topology_unlock();
2088 G_MIRROR_DEBUG(0, "Device %s launched (%u/%u).", pp->name,
2089 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE), sc->sc_ndisks);
2090 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
2091 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING)
2092 g_mirror_sync_start(disk);
2097 g_mirror_destroy_provider(struct g_mirror_softc *sc)
2099 struct g_mirror_disk *disk;
2102 g_topology_assert_not();
2103 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).",
2107 g_error_provider(sc->sc_provider, ENXIO);
2108 mtx_lock(&sc->sc_queue_mtx);
2109 while ((bp = bioq_takefirst(&sc->sc_queue)) != NULL)
2110 g_io_deliver(bp, ENXIO);
2111 mtx_unlock(&sc->sc_queue_mtx);
2112 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name,
2113 sc->sc_provider->name);
2114 sc->sc_provider->flags |= G_PF_WITHER;
2115 g_orphan_provider(sc->sc_provider, ENXIO);
2116 g_topology_unlock();
2117 sc->sc_provider = NULL;
2118 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
2119 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING)
2120 g_mirror_sync_stop(disk, 1);
2125 g_mirror_go(void *arg)
2127 struct g_mirror_softc *sc;
2130 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name);
2131 g_mirror_event_send(sc, 0,
2132 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE);
2136 g_mirror_determine_state(struct g_mirror_disk *disk)
2138 struct g_mirror_softc *sc;
2142 if (sc->sc_syncid == disk->d_sync.ds_syncid) {
2143 if ((disk->d_flags &
2144 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) {
2145 /* Disk does not need synchronization. */
2146 state = G_MIRROR_DISK_STATE_ACTIVE;
2149 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 ||
2151 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) {
2153 * We can start synchronization from
2154 * the stored offset.
2156 state = G_MIRROR_DISK_STATE_SYNCHRONIZING;
2158 state = G_MIRROR_DISK_STATE_STALE;
2161 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) {
2163 * Reset all synchronization data for this disk,
2164 * because if it even was synchronized, it was
2165 * synchronized to disks with different syncid.
2167 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING;
2168 disk->d_sync.ds_offset = 0;
2169 disk->d_sync.ds_offset_done = 0;
2170 disk->d_sync.ds_syncid = sc->sc_syncid;
2171 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 ||
2172 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) {
2173 state = G_MIRROR_DISK_STATE_SYNCHRONIZING;
2175 state = G_MIRROR_DISK_STATE_STALE;
2177 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ {
2179 * Not good, NOT GOOD!
2180 * It means that mirror was started on stale disks
2181 * and more fresh disk just arrive.
2182 * If there were writes, mirror is broken, sorry.
2183 * I think the best choice here is don't touch
2184 * this disk and inform the user loudly.
2186 G_MIRROR_DEBUG(0, "Device %s was started before the freshest "
2187 "disk (%s) arrives!! It will not be connected to the "
2188 "running device.", sc->sc_name,
2189 g_mirror_get_diskname(disk));
2190 g_mirror_destroy_disk(disk);
2191 state = G_MIRROR_DISK_STATE_NONE;
2192 /* Return immediately, because disk was destroyed. */
2195 G_MIRROR_DEBUG(3, "State for %s disk: %s.",
2196 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state));
2201 * Update device state.
2204 g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force)
2206 struct g_mirror_disk *disk;
2209 sx_assert(&sc->sc_lock, SX_XLOCKED);
2211 switch (sc->sc_state) {
2212 case G_MIRROR_DEVICE_STATE_STARTING:
2214 struct g_mirror_disk *pdisk, *tdisk;
2215 u_int dirty, ndisks, genid, syncid;
2217 KASSERT(sc->sc_provider == NULL,
2218 ("Non-NULL provider in STARTING state (%s).", sc->sc_name));
2220 * Are we ready? We are, if all disks are connected or
2221 * if we have any disks and 'force' is true.
2223 ndisks = g_mirror_ndisks(sc, -1);
2224 if (sc->sc_ndisks == ndisks || (force && ndisks > 0)) {
2226 } else if (ndisks == 0) {
2228 * Disks went down in starting phase, so destroy
2231 callout_drain(&sc->sc_callout);
2232 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY;
2233 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p", __LINE__,
2235 root_mount_rel(sc->sc_rootmount);
2236 sc->sc_rootmount = NULL;
2243 * Activate all disks with the biggest syncid.
2247 * If 'force' is true, we have been called due to
2248 * timeout, so don't bother canceling timeout.
2251 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
2252 if ((disk->d_flags &
2253 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) {
2258 /* No valid disks found, destroy device. */
2259 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY;
2260 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p",
2261 __LINE__, sc->sc_rootmount);
2262 root_mount_rel(sc->sc_rootmount);
2263 sc->sc_rootmount = NULL;
2267 /* Cancel timeout. */
2268 callout_drain(&sc->sc_callout);
2272 * Find the biggest genid.
2275 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
2276 if (disk->d_genid > genid)
2277 genid = disk->d_genid;
2279 sc->sc_genid = genid;
2281 * Remove all disks without the biggest genid.
2283 LIST_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) {
2284 if (disk->d_genid < genid) {
2286 "Component %s (device %s) broken, skipping.",
2287 g_mirror_get_diskname(disk), sc->sc_name);
2288 g_mirror_destroy_disk(disk);
2293 * Find the biggest syncid.
2296 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
2297 if (disk->d_sync.ds_syncid > syncid)
2298 syncid = disk->d_sync.ds_syncid;
2302 * Here we need to look for dirty disks and if all disks
2303 * with the biggest syncid are dirty, we have to choose
2304 * one with the biggest priority and rebuild the rest.
2307 * Find the number of dirty disks with the biggest syncid.
2308 * Find the number of disks with the biggest syncid.
2309 * While here, find a disk with the biggest priority.
2313 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
2314 if (disk->d_sync.ds_syncid != syncid)
2316 if ((disk->d_flags &
2317 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) {
2321 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) {
2323 if (pdisk == NULL ||
2324 pdisk->d_priority < disk->d_priority) {
2330 /* No dirty disks at all, great. */
2331 } else if (dirty == ndisks) {
2333 * Force synchronization for all dirty disks except one
2334 * with the biggest priority.
2336 KASSERT(pdisk != NULL, ("pdisk == NULL"));
2337 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a "
2338 "master disk for synchronization.",
2339 g_mirror_get_diskname(pdisk), sc->sc_name);
2340 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
2341 if (disk->d_sync.ds_syncid != syncid)
2343 if ((disk->d_flags &
2344 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) {
2347 KASSERT((disk->d_flags &
2348 G_MIRROR_DISK_FLAG_DIRTY) != 0,
2349 ("Disk %s isn't marked as dirty.",
2350 g_mirror_get_diskname(disk)));
2351 /* Skip the disk with the biggest priority. */
2354 disk->d_sync.ds_syncid = 0;
2356 } else if (dirty < ndisks) {
2358 * Force synchronization for all dirty disks.
2359 * We have some non-dirty disks.
2361 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
2362 if (disk->d_sync.ds_syncid != syncid)
2364 if ((disk->d_flags &
2365 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) {
2368 if ((disk->d_flags &
2369 G_MIRROR_DISK_FLAG_DIRTY) == 0) {
2372 disk->d_sync.ds_syncid = 0;
2378 sc->sc_syncid = syncid;
2380 /* Remember to bump syncid on first write. */
2381 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID;
2383 state = G_MIRROR_DEVICE_STATE_RUNNING;
2384 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.",
2385 sc->sc_name, g_mirror_device_state2str(sc->sc_state),
2386 g_mirror_device_state2str(state));
2387 sc->sc_state = state;
2388 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
2389 state = g_mirror_determine_state(disk);
2390 g_mirror_event_send(disk, state,
2391 G_MIRROR_EVENT_DONTWAIT);
2392 if (state == G_MIRROR_DISK_STATE_STALE)
2393 sc->sc_bump_id |= G_MIRROR_BUMP_SYNCID;
2397 case G_MIRROR_DEVICE_STATE_RUNNING:
2398 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 &&
2399 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) {
2401 * No active disks or no disks at all,
2402 * so destroy device.
2404 if (sc->sc_provider != NULL)
2405 g_mirror_destroy_provider(sc);
2406 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY;
2408 } else if (g_mirror_ndisks(sc,
2409 G_MIRROR_DISK_STATE_ACTIVE) > 0 &&
2410 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) {
2412 * We have active disks, launch provider if it doesn't
2415 if (sc->sc_provider == NULL)
2416 g_mirror_launch_provider(sc);
2417 if (sc->sc_rootmount != NULL) {
2418 G_MIRROR_DEBUG(1, "root_mount_rel[%u] %p",
2419 __LINE__, sc->sc_rootmount);
2420 root_mount_rel(sc->sc_rootmount);
2421 sc->sc_rootmount = NULL;
2425 * Genid should be bumped immediately, so do it here.
2427 if ((sc->sc_bump_id & G_MIRROR_BUMP_GENID) != 0) {
2428 sc->sc_bump_id &= ~G_MIRROR_BUMP_GENID;
2429 g_mirror_bump_genid(sc);
2433 KASSERT(1 == 0, ("Wrong device state (%s, %s).",
2434 sc->sc_name, g_mirror_device_state2str(sc->sc_state)));
2440 * Update disk state and device state if needed.
2442 #define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \
2443 "Disk %s state changed from %s to %s (device %s).", \
2444 g_mirror_get_diskname(disk), \
2445 g_mirror_disk_state2str(disk->d_state), \
2446 g_mirror_disk_state2str(state), sc->sc_name)
2448 g_mirror_update_disk(struct g_mirror_disk *disk, u_int state)
2450 struct g_mirror_softc *sc;
2453 sx_assert(&sc->sc_lock, SX_XLOCKED);
2456 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.",
2457 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state),
2458 g_mirror_disk_state2str(state));
2460 case G_MIRROR_DISK_STATE_NEW:
2462 * Possible scenarios:
2463 * 1. New disk arrive.
2465 /* Previous state should be NONE. */
2466 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE,
2467 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk),
2468 g_mirror_disk_state2str(disk->d_state)));
2469 DISK_STATE_CHANGED();
2471 disk->d_state = state;
2472 if (LIST_EMPTY(&sc->sc_disks))
2473 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next);
2475 struct g_mirror_disk *dp;
2477 LIST_FOREACH(dp, &sc->sc_disks, d_next) {
2478 if (disk->d_priority >= dp->d_priority) {
2479 LIST_INSERT_BEFORE(dp, disk, d_next);
2483 if (LIST_NEXT(dp, d_next) == NULL)
2487 LIST_INSERT_AFTER(dp, disk, d_next);
2489 G_MIRROR_DEBUG(1, "Device %s: provider %s detected.",
2490 sc->sc_name, g_mirror_get_diskname(disk));
2491 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING)
2493 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING,
2494 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2495 g_mirror_device_state2str(sc->sc_state),
2496 g_mirror_get_diskname(disk),
2497 g_mirror_disk_state2str(disk->d_state)));
2498 state = g_mirror_determine_state(disk);
2499 if (state != G_MIRROR_DISK_STATE_NONE)
2502 case G_MIRROR_DISK_STATE_ACTIVE:
2504 * Possible scenarios:
2505 * 1. New disk does not need synchronization.
2506 * 2. Synchronization process finished successfully.
2508 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING,
2509 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2510 g_mirror_device_state2str(sc->sc_state),
2511 g_mirror_get_diskname(disk),
2512 g_mirror_disk_state2str(disk->d_state)));
2513 /* Previous state should be NEW or SYNCHRONIZING. */
2514 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW ||
2515 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING,
2516 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk),
2517 g_mirror_disk_state2str(disk->d_state)));
2518 DISK_STATE_CHANGED();
2520 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) {
2521 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING;
2522 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC;
2523 g_mirror_sync_stop(disk, 0);
2525 disk->d_state = state;
2526 disk->d_sync.ds_offset = 0;
2527 disk->d_sync.ds_offset_done = 0;
2528 g_mirror_update_idle(sc, disk);
2529 g_mirror_update_metadata(disk);
2530 G_MIRROR_DEBUG(1, "Device %s: provider %s activated.",
2531 sc->sc_name, g_mirror_get_diskname(disk));
2533 case G_MIRROR_DISK_STATE_STALE:
2535 * Possible scenarios:
2536 * 1. Stale disk was connected.
2538 /* Previous state should be NEW. */
2539 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW,
2540 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk),
2541 g_mirror_disk_state2str(disk->d_state)));
2542 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING,
2543 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2544 g_mirror_device_state2str(sc->sc_state),
2545 g_mirror_get_diskname(disk),
2546 g_mirror_disk_state2str(disk->d_state)));
2548 * STALE state is only possible if device is marked
2551 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0,
2552 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2553 g_mirror_device_state2str(sc->sc_state),
2554 g_mirror_get_diskname(disk),
2555 g_mirror_disk_state2str(disk->d_state)));
2556 DISK_STATE_CHANGED();
2558 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY;
2559 disk->d_state = state;
2560 g_mirror_update_metadata(disk);
2561 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.",
2562 sc->sc_name, g_mirror_get_diskname(disk));
2564 case G_MIRROR_DISK_STATE_SYNCHRONIZING:
2566 * Possible scenarios:
2567 * 1. Disk which needs synchronization was connected.
2569 /* Previous state should be NEW. */
2570 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW,
2571 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk),
2572 g_mirror_disk_state2str(disk->d_state)));
2573 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING,
2574 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2575 g_mirror_device_state2str(sc->sc_state),
2576 g_mirror_get_diskname(disk),
2577 g_mirror_disk_state2str(disk->d_state)));
2578 DISK_STATE_CHANGED();
2580 if (disk->d_state == G_MIRROR_DISK_STATE_NEW)
2581 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY;
2582 disk->d_state = state;
2583 if (sc->sc_provider != NULL) {
2584 g_mirror_sync_start(disk);
2585 g_mirror_update_metadata(disk);
2588 case G_MIRROR_DISK_STATE_DISCONNECTED:
2590 * Possible scenarios:
2591 * 1. Device wasn't running yet, but disk disappear.
2592 * 2. Disk was active and disapppear.
2593 * 3. Disk disappear during synchronization process.
2595 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) {
2597 * Previous state should be ACTIVE, STALE or
2600 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE ||
2601 disk->d_state == G_MIRROR_DISK_STATE_STALE ||
2602 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING,
2603 ("Wrong disk state (%s, %s).",
2604 g_mirror_get_diskname(disk),
2605 g_mirror_disk_state2str(disk->d_state)));
2606 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) {
2607 /* Previous state should be NEW. */
2608 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW,
2609 ("Wrong disk state (%s, %s).",
2610 g_mirror_get_diskname(disk),
2611 g_mirror_disk_state2str(disk->d_state)));
2613 * Reset bumping syncid if disk disappeared in STARTING
2616 if ((sc->sc_bump_id & G_MIRROR_BUMP_SYNCID) != 0)
2617 sc->sc_bump_id &= ~G_MIRROR_BUMP_SYNCID;
2620 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).",
2622 g_mirror_device_state2str(sc->sc_state),
2623 g_mirror_get_diskname(disk),
2624 g_mirror_disk_state2str(disk->d_state)));
2627 DISK_STATE_CHANGED();
2628 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.",
2629 sc->sc_name, g_mirror_get_diskname(disk));
2631 g_mirror_destroy_disk(disk);
2633 case G_MIRROR_DISK_STATE_DESTROY:
2637 error = g_mirror_clear_metadata(disk);
2640 DISK_STATE_CHANGED();
2641 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.",
2642 sc->sc_name, g_mirror_get_diskname(disk));
2644 g_mirror_destroy_disk(disk);
2646 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
2647 g_mirror_update_metadata(disk);
2652 KASSERT(1 == 0, ("Unknown state (%u).", state));
2657 #undef DISK_STATE_CHANGED
2660 g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md)
2662 struct g_provider *pp;
2666 g_topology_assert();
2668 error = g_access(cp, 1, 0, 0);
2672 g_topology_unlock();
2673 /* Metadata are stored on last sector. */
2674 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize,
2677 g_access(cp, -1, 0, 0);
2679 G_MIRROR_DEBUG(1, "Cannot read metadata from %s (error=%d).",
2680 cp->provider->name, error);
2684 /* Decode metadata. */
2685 error = mirror_metadata_decode(buf, md);
2687 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0)
2689 if (md->md_version > G_MIRROR_VERSION) {
2691 "Kernel module is too old to handle metadata from %s.",
2692 cp->provider->name);
2696 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.",
2697 cp->provider->name);
2705 g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp,
2706 struct g_mirror_metadata *md)
2709 if (g_mirror_id2disk(sc, md->md_did) != NULL) {
2710 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.",
2711 pp->name, md->md_did);
2714 if (md->md_all != sc->sc_ndisks) {
2716 "Invalid '%s' field on disk %s (device %s), skipping.",
2717 "md_all", pp->name, sc->sc_name);
2720 if (md->md_slice != sc->sc_slice) {
2722 "Invalid '%s' field on disk %s (device %s), skipping.",
2723 "md_slice", pp->name, sc->sc_name);
2726 if (md->md_balance != sc->sc_balance) {
2728 "Invalid '%s' field on disk %s (device %s), skipping.",
2729 "md_balance", pp->name, sc->sc_name);
2733 if (md->md_mediasize != sc->sc_mediasize) {
2735 "Invalid '%s' field on disk %s (device %s), skipping.",
2736 "md_mediasize", pp->name, sc->sc_name);
2740 if (sc->sc_mediasize > pp->mediasize) {
2742 "Invalid size of disk %s (device %s), skipping.", pp->name,
2746 if (md->md_sectorsize != sc->sc_sectorsize) {
2748 "Invalid '%s' field on disk %s (device %s), skipping.",
2749 "md_sectorsize", pp->name, sc->sc_name);
2752 if ((sc->sc_sectorsize % pp->sectorsize) != 0) {
2754 "Invalid sector size of disk %s (device %s), skipping.",
2755 pp->name, sc->sc_name);
2758 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) {
2760 "Invalid device flags on disk %s (device %s), skipping.",
2761 pp->name, sc->sc_name);
2764 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) {
2766 "Invalid disk flags on disk %s (device %s), skipping.",
2767 pp->name, sc->sc_name);
2774 g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp,
2775 struct g_mirror_metadata *md)
2777 struct g_mirror_disk *disk;
2780 g_topology_assert_not();
2781 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name);
2783 error = g_mirror_check_metadata(sc, pp, md);
2786 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING &&
2787 md->md_genid < sc->sc_genid) {
2788 G_MIRROR_DEBUG(0, "Component %s (device %s) broken, skipping.",
2789 pp->name, sc->sc_name);
2792 disk = g_mirror_init_disk(sc, pp, md, &error);
2795 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW,
2796 G_MIRROR_EVENT_WAIT);
2799 if (md->md_version < G_MIRROR_VERSION) {
2800 G_MIRROR_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).",
2801 pp->name, md->md_version, G_MIRROR_VERSION);
2802 g_mirror_update_metadata(disk);
2808 g_mirror_destroy_delayed(void *arg, int flag)
2810 struct g_mirror_softc *sc;
2813 if (flag == EV_CANCEL) {
2814 G_MIRROR_DEBUG(1, "Destroying canceled.");
2818 g_topology_unlock();
2819 sx_xlock(&sc->sc_lock);
2820 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) == 0,
2821 ("DESTROY flag set on %s.", sc->sc_name));
2822 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0,
2823 ("DESTROYING flag not set on %s.", sc->sc_name));
2824 G_MIRROR_DEBUG(1, "Destroying %s (delayed).", sc->sc_name);
2825 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_SOFT);
2827 G_MIRROR_DEBUG(0, "Cannot destroy %s.", sc->sc_name);
2828 sx_xunlock(&sc->sc_lock);
2834 g_mirror_access(struct g_provider *pp, int acr, int acw, int ace)
2836 struct g_mirror_softc *sc;
2837 int dcr, dcw, dce, error = 0;
2839 g_topology_assert();
2840 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr,
2843 sc = pp->geom->softc;
2844 if (sc == NULL && acr <= 0 && acw <= 0 && ace <= 0)
2846 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name));
2848 dcr = pp->acr + acr;
2849 dcw = pp->acw + acw;
2850 dce = pp->ace + ace;
2852 g_topology_unlock();
2853 sx_xlock(&sc->sc_lock);
2854 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0 ||
2855 LIST_EMPTY(&sc->sc_disks)) {
2856 if (acr > 0 || acw > 0 || ace > 0)
2861 g_mirror_idle(sc, dcw);
2862 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROYING) != 0) {
2863 if (acr > 0 || acw > 0 || ace > 0) {
2867 if (dcr == 0 && dcw == 0 && dce == 0) {
2868 g_post_event(g_mirror_destroy_delayed, sc, M_WAITOK,
2873 sx_xunlock(&sc->sc_lock);
2878 static struct g_geom *
2879 g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md)
2881 struct g_mirror_softc *sc;
2885 g_topology_assert();
2886 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name,
2889 /* One disk is minimum. */
2895 gp = g_new_geomf(mp, "%s", md->md_name);
2896 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO);
2897 gp->start = g_mirror_start;
2898 gp->orphan = g_mirror_orphan;
2899 gp->access = g_mirror_access;
2900 gp->dumpconf = g_mirror_dumpconf;
2902 sc->sc_id = md->md_mid;
2903 sc->sc_slice = md->md_slice;
2904 sc->sc_balance = md->md_balance;
2905 sc->sc_mediasize = md->md_mediasize;
2906 sc->sc_sectorsize = md->md_sectorsize;
2907 sc->sc_ndisks = md->md_all;
2908 sc->sc_flags = md->md_mflags;
2911 sc->sc_last_write = time_uptime;
2913 sx_init(&sc->sc_lock, "gmirror:lock");
2914 bioq_init(&sc->sc_queue);
2915 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF);
2916 bioq_init(&sc->sc_regular_delayed);
2917 bioq_init(&sc->sc_inflight);
2918 bioq_init(&sc->sc_sync_delayed);
2919 LIST_INIT(&sc->sc_disks);
2920 TAILQ_INIT(&sc->sc_events);
2921 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF);
2922 callout_init(&sc->sc_callout, CALLOUT_MPSAFE);
2923 mtx_init(&sc->sc_done_mtx, "gmirror:done", NULL, MTX_DEF);
2924 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING;
2927 sc->sc_provider = NULL;
2929 * Synchronization geom.
2931 gp = g_new_geomf(mp, "%s.sync", md->md_name);
2933 gp->orphan = g_mirror_orphan;
2934 sc->sc_sync.ds_geom = gp;
2935 sc->sc_sync.ds_ndisks = 0;
2936 error = kproc_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0,
2937 "g_mirror %s", md->md_name);
2939 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.",
2941 g_destroy_geom(sc->sc_sync.ds_geom);
2942 mtx_destroy(&sc->sc_done_mtx);
2943 mtx_destroy(&sc->sc_events_mtx);
2944 mtx_destroy(&sc->sc_queue_mtx);
2945 sx_destroy(&sc->sc_lock);
2946 g_destroy_geom(sc->sc_geom);
2951 G_MIRROR_DEBUG(1, "Device %s created (%u components, id=%u).",
2952 sc->sc_name, sc->sc_ndisks, sc->sc_id);
2954 sc->sc_rootmount = root_mount_hold("GMIRROR");
2955 G_MIRROR_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount);
2959 timeout = g_mirror_timeout * hz;
2960 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc);
2961 return (sc->sc_geom);
2965 g_mirror_destroy(struct g_mirror_softc *sc, int how)
2967 struct g_mirror_disk *disk;
2968 struct g_provider *pp;
2970 g_topology_assert_not();
2973 sx_assert(&sc->sc_lock, SX_XLOCKED);
2975 pp = sc->sc_provider;
2976 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
2978 case G_MIRROR_DESTROY_SOFT:
2980 "Device %s is still open (r%dw%de%d).", pp->name,
2981 pp->acr, pp->acw, pp->ace);
2983 case G_MIRROR_DESTROY_DELAYED:
2985 "Device %s will be destroyed on last close.",
2987 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
2988 if (disk->d_state ==
2989 G_MIRROR_DISK_STATE_SYNCHRONIZING) {
2990 g_mirror_sync_stop(disk, 1);
2993 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROYING;
2995 case G_MIRROR_DESTROY_HARD:
2996 G_MIRROR_DEBUG(1, "Device %s is still open, so it "
2997 "can't be definitely removed.", pp->name);
3002 if (sc->sc_geom->softc == NULL) {
3003 g_topology_unlock();
3006 sc->sc_geom->softc = NULL;
3007 sc->sc_sync.ds_geom->softc = NULL;
3008 g_topology_unlock();
3010 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY;
3011 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT;
3012 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc);
3013 sx_xunlock(&sc->sc_lock);
3014 mtx_lock(&sc->sc_queue_mtx);
3016 mtx_unlock(&sc->sc_queue_mtx);
3017 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker);
3018 while (sc->sc_worker != NULL)
3019 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5);
3020 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker);
3021 sx_xlock(&sc->sc_lock);
3022 g_mirror_destroy_device(sc);
3028 g_mirror_taste_orphan(struct g_consumer *cp)
3031 KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
3032 cp->provider->name));
3035 static struct g_geom *
3036 g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
3038 struct g_mirror_metadata md;
3039 struct g_mirror_softc *sc;
3040 struct g_consumer *cp;
3044 g_topology_assert();
3045 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
3046 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name);
3048 gp = g_new_geomf(mp, "mirror:taste");
3050 * This orphan function should be never called.
3052 gp->orphan = g_mirror_taste_orphan;
3053 cp = g_new_consumer(gp);
3055 error = g_mirror_read_metadata(cp, &md);
3057 g_destroy_consumer(cp);
3063 if (md.md_provider[0] != '\0' &&
3064 !g_compare_names(md.md_provider, pp->name))
3066 if (md.md_provsize != 0 && md.md_provsize != pp->mediasize)
3068 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) {
3070 "Device %s: provider %s marked as inactive, skipping.",
3071 md.md_name, pp->name);
3074 if (g_mirror_debug >= 2)
3075 mirror_metadata_dump(&md);
3078 * Let's check if device already exists.
3081 LIST_FOREACH(gp, &mp->geom, geom) {
3085 if (sc->sc_sync.ds_geom == gp)
3087 if (strcmp(md.md_name, sc->sc_name) != 0)
3089 if (md.md_mid != sc->sc_id) {
3090 G_MIRROR_DEBUG(0, "Device %s already configured.",
3097 gp = g_mirror_create(mp, &md);
3099 G_MIRROR_DEBUG(0, "Cannot create device %s.",
3105 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name);
3106 g_topology_unlock();
3107 sx_xlock(&sc->sc_lock);
3108 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_TASTING;
3109 error = g_mirror_add_disk(sc, pp, &md);
3111 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).",
3112 pp->name, gp->name, error);
3113 if (LIST_EMPTY(&sc->sc_disks)) {
3115 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD);
3121 sc->sc_flags &= ~G_MIRROR_DEVICE_FLAG_TASTING;
3122 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_DESTROY) != 0) {
3123 g_mirror_destroy(sc, G_MIRROR_DESTROY_HARD);
3127 sx_xunlock(&sc->sc_lock);
3133 g_mirror_resize(struct g_consumer *cp)
3135 struct g_mirror_disk *disk;
3137 g_topology_assert();
3138 g_trace(G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name);
3143 g_topology_unlock();
3144 g_mirror_update_metadata(disk);
3149 g_mirror_destroy_geom(struct gctl_req *req __unused,
3150 struct g_class *mp __unused, struct g_geom *gp)
3152 struct g_mirror_softc *sc;
3155 g_topology_unlock();
3157 sx_xlock(&sc->sc_lock);
3159 error = g_mirror_destroy(gp->softc, G_MIRROR_DESTROY_SOFT);
3161 sx_xunlock(&sc->sc_lock);
3167 g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
3168 struct g_consumer *cp, struct g_provider *pp)
3170 struct g_mirror_softc *sc;
3172 g_topology_assert();
3177 /* Skip synchronization geom. */
3178 if (gp == sc->sc_sync.ds_geom)
3182 } else if (cp != NULL) {
3183 struct g_mirror_disk *disk;
3188 g_topology_unlock();
3189 sx_xlock(&sc->sc_lock);
3190 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id);
3191 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) {
3192 sbuf_printf(sb, "%s<Synchronized>", indent);
3193 if (disk->d_sync.ds_offset == 0)
3194 sbuf_printf(sb, "0%%");
3196 sbuf_printf(sb, "%u%%",
3197 (u_int)((disk->d_sync.ds_offset * 100) /
3198 sc->sc_provider->mediasize));
3200 sbuf_printf(sb, "</Synchronized>\n");
3201 if (disk->d_sync.ds_offset > 0) {
3202 sbuf_printf(sb, "%s<BytesSynced>%jd"
3203 "</BytesSynced>\n", indent,
3204 (intmax_t)disk->d_sync.ds_offset);
3207 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent,
3208 disk->d_sync.ds_syncid);
3209 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent,
3211 sbuf_printf(sb, "%s<Flags>", indent);
3212 if (disk->d_flags == 0)
3213 sbuf_printf(sb, "NONE");
3217 #define ADD_FLAG(flag, name) do { \
3218 if ((disk->d_flags & (flag)) != 0) { \
3220 sbuf_printf(sb, ", "); \
3223 sbuf_printf(sb, name); \
3226 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY");
3227 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED");
3228 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE");
3229 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING,
3231 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC");
3232 ADD_FLAG(G_MIRROR_DISK_FLAG_BROKEN, "BROKEN");
3235 sbuf_printf(sb, "</Flags>\n");
3236 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent,
3238 sbuf_printf(sb, "%s<State>%s</State>\n", indent,
3239 g_mirror_disk_state2str(disk->d_state));
3240 sx_xunlock(&sc->sc_lock);
3243 g_topology_unlock();
3244 sx_xlock(&sc->sc_lock);
3245 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id);
3246 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid);
3247 sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid);
3248 sbuf_printf(sb, "%s<Flags>", indent);
3249 if (sc->sc_flags == 0)
3250 sbuf_printf(sb, "NONE");
3254 #define ADD_FLAG(flag, name) do { \
3255 if ((sc->sc_flags & (flag)) != 0) { \
3257 sbuf_printf(sb, ", "); \
3260 sbuf_printf(sb, name); \
3263 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC");
3264 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC");
3267 sbuf_printf(sb, "</Flags>\n");
3268 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent,
3269 (u_int)sc->sc_slice);
3270 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent,
3271 balance_name(sc->sc_balance));
3272 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent,
3274 sbuf_printf(sb, "%s<State>", indent);
3275 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING)
3276 sbuf_printf(sb, "%s", "STARTING");
3277 else if (sc->sc_ndisks ==
3278 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE))
3279 sbuf_printf(sb, "%s", "COMPLETE");
3281 sbuf_printf(sb, "%s", "DEGRADED");
3282 sbuf_printf(sb, "</State>\n");
3283 sx_xunlock(&sc->sc_lock);
3289 g_mirror_shutdown_post_sync(void *arg, int howto)
3292 struct g_geom *gp, *gp2;
3293 struct g_mirror_softc *sc;
3299 g_mirror_shutdown = 1;
3300 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
3301 if ((sc = gp->softc) == NULL)
3303 /* Skip synchronization geom. */
3304 if (gp == sc->sc_sync.ds_geom)
3306 g_topology_unlock();
3307 sx_xlock(&sc->sc_lock);
3308 g_mirror_idle(sc, -1);
3310 error = g_mirror_destroy(sc, G_MIRROR_DESTROY_DELAYED);
3312 sx_xunlock(&sc->sc_lock);
3315 g_topology_unlock();
3320 g_mirror_init(struct g_class *mp)
3323 g_mirror_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync,
3324 g_mirror_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST);
3325 if (g_mirror_post_sync == NULL)
3326 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event.");
3330 g_mirror_fini(struct g_class *mp)
3333 if (g_mirror_post_sync != NULL)
3334 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_post_sync);
3337 DECLARE_GEOM_CLASS(g_mirror_class, g_mirror);