1/*-
2 * Copyright (c) 2004 Pawel Jakub Dawidek <pjd@FreeBSD.org>
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
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.
13 *
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
24 * SUCH DAMAGE.
25 */
26
27#include <sys/cdefs.h>
28__FBSDID("$FreeBSD: head/sys/geom/mirror/g_mirror.c 137254 2004-11-05 12:35:21Z pjd $");
29
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>
35#include <sys/lock.h>
36#include <sys/mutex.h>
37#include <sys/bio.h>
38#include <sys/sysctl.h>
39#include <sys/malloc.h>
40#include <sys/eventhandler.h>
41#include <vm/uma.h>
42#include <geom/geom.h>
43#include <sys/proc.h>
44#include <sys/kthread.h>
45#include <geom/mirror/g_mirror.h>
46
47
48static MALLOC_DEFINE(M_MIRROR, "mirror data", "GEOM_MIRROR Data");
49
50SYSCTL_DECL(_kern_geom);
51SYSCTL_NODE(_kern_geom, OID_AUTO, mirror, CTLFLAG_RW, 0, "GEOM_MIRROR stuff");
52u_int g_mirror_debug = 0;
53TUNABLE_INT("kern.geom.mirror.debug", &g_mirror_debug);
54SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, debug, CTLFLAG_RW, &g_mirror_debug, 0,
55 "Debug level");
56static u_int g_mirror_timeout = 4;
57TUNABLE_INT("kern.geom.mirror.timeout", &g_mirror_timeout);
58SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, timeout, CTLFLAG_RW, &g_mirror_timeout,
59 0, "Time to wait on all mirror components");
60static u_int g_mirror_idletime = 5;
61TUNABLE_INT("kern.geom.mirror.idletime", &g_mirror_idletime);
62SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, idletime, CTLFLAG_RW,
63 &g_mirror_idletime, 0, "Mark components as clean when idling");
64static u_int g_mirror_reqs_per_sync = 5;
65SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, reqs_per_sync, CTLFLAG_RW,
66 &g_mirror_reqs_per_sync, 0,
67 "Number of regular I/O requests per synchronization request");
68static u_int g_mirror_syncs_per_sec = 100;
69SYSCTL_UINT(_kern_geom_mirror, OID_AUTO, syncs_per_sec, CTLFLAG_RW,
70 &g_mirror_syncs_per_sec, 0,
71 "Number of synchronizations requests per second");
72
73#define MSLEEP(ident, mtx, priority, wmesg, timeout) do { \
74 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \
75 msleep((ident), (mtx), (priority), (wmesg), (timeout)); \
76 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \
77} while (0)
78
79static eventhandler_tag g_mirror_ehtag = NULL;
80
81static int g_mirror_destroy_geom(struct gctl_req *req, struct g_class *mp,
82 struct g_geom *gp);
83static g_taste_t g_mirror_taste;
84static void g_mirror_init(struct g_class *mp);
85static void g_mirror_fini(struct g_class *mp);
86
87struct g_class g_mirror_class = {
88 .name = G_MIRROR_CLASS_NAME,
89 .version = G_VERSION,
90 .ctlreq = g_mirror_config,
91 .taste = g_mirror_taste,
92 .destroy_geom = g_mirror_destroy_geom,
93 .init = g_mirror_init,
94 .fini = g_mirror_fini
95};
96
97
98static void g_mirror_destroy_provider(struct g_mirror_softc *sc);
99static int g_mirror_update_disk(struct g_mirror_disk *disk, u_int state);
100static void g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force);
101static void g_mirror_dumpconf(struct sbuf *sb, const char *indent,
102 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp);
103static void g_mirror_sync_stop(struct g_mirror_disk *disk, int type);
104
105
106static const char *
107g_mirror_disk_state2str(int state)
108{
109
110 switch (state) {
111 case G_MIRROR_DISK_STATE_NONE:
112 return ("NONE");
113 case G_MIRROR_DISK_STATE_NEW:
114 return ("NEW");
115 case G_MIRROR_DISK_STATE_ACTIVE:
116 return ("ACTIVE");
117 case G_MIRROR_DISK_STATE_STALE:
118 return ("STALE");
119 case G_MIRROR_DISK_STATE_SYNCHRONIZING:
120 return ("SYNCHRONIZING");
121 case G_MIRROR_DISK_STATE_DISCONNECTED:
122 return ("DISCONNECTED");
123 case G_MIRROR_DISK_STATE_DESTROY:
124 return ("DESTROY");
125 default:
126 return ("INVALID");
127 }
128}
129
130static const char *
131g_mirror_device_state2str(int state)
132{
133
134 switch (state) {
135 case G_MIRROR_DEVICE_STATE_STARTING:
136 return ("STARTING");
137 case G_MIRROR_DEVICE_STATE_RUNNING:
138 return ("RUNNING");
139 default:
140 return ("INVALID");
141 }
142}
143
144static const char *
145g_mirror_get_diskname(struct g_mirror_disk *disk)
146{
147
148 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL)
149 return ("[unknown]");
150 return (disk->d_name);
151}
152
153/*
154 * --- Events handling functions ---
155 * Events in geom_mirror are used to maintain disks and device status
156 * from one thread to simplify locking.
157 */
158static void
159g_mirror_event_free(struct g_mirror_event *ep)
160{
161
162 free(ep, M_MIRROR);
163}
164
165int
166g_mirror_event_send(void *arg, int state, int flags)
167{
168 struct g_mirror_softc *sc;
169 struct g_mirror_disk *disk;
170 struct g_mirror_event *ep;
171 int error;
172
173 ep = malloc(sizeof(*ep), M_MIRROR, M_WAITOK);
174 G_MIRROR_DEBUG(4, "%s: Sending event %p.", __func__, ep);
175 if ((flags & G_MIRROR_EVENT_DEVICE) != 0) {
176 disk = NULL;
177 sc = arg;
178 } else {
179 disk = arg;
180 sc = disk->d_softc;
181 }
182 ep->e_disk = disk;
183 ep->e_state = state;
184 ep->e_flags = flags;
185 ep->e_error = 0;
186 mtx_lock(&sc->sc_events_mtx);
187 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next);
188 mtx_unlock(&sc->sc_events_mtx);
189 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc);
190 mtx_lock(&sc->sc_queue_mtx);
191 wakeup(sc);
192 mtx_unlock(&sc->sc_queue_mtx);
193 if ((flags & G_MIRROR_EVENT_DONTWAIT) != 0)
194 return (0);
195 g_topology_assert();
196 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, ep);
197 g_topology_unlock();
198 while ((ep->e_flags & G_MIRROR_EVENT_DONE) == 0) {
199 mtx_lock(&sc->sc_events_mtx);
200 MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "m:event",
201 hz * 5);
202 }
203 /* Don't even try to use 'sc' here, because it could be already dead. */
204 g_topology_lock();
205 error = ep->e_error;
206 g_mirror_event_free(ep);
207 return (error);
208}
209
210static struct g_mirror_event *
211g_mirror_event_get(struct g_mirror_softc *sc)
212{
213 struct g_mirror_event *ep;
214
215 mtx_lock(&sc->sc_events_mtx);
216 ep = TAILQ_FIRST(&sc->sc_events);
217 if (ep != NULL)
218 TAILQ_REMOVE(&sc->sc_events, ep, e_next);
219 mtx_unlock(&sc->sc_events_mtx);
220 return (ep);
221}
222
223static void
224g_mirror_event_cancel(struct g_mirror_disk *disk)
225{
226 struct g_mirror_softc *sc;
227 struct g_mirror_event *ep, *tmpep;
228
229 g_topology_assert();
230
231 sc = disk->d_softc;
232 mtx_lock(&sc->sc_events_mtx);
233 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) {
234 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0)
235 continue;
236 if (ep->e_disk != disk)
237 continue;
238 TAILQ_REMOVE(&sc->sc_events, ep, e_next);
239 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0)
240 g_mirror_event_free(ep);
241 else {
242 ep->e_error = ECANCELED;
243 wakeup(ep);
244 }
245 }
246 mtx_unlock(&sc->sc_events_mtx);
247}
248
249/*
250 * Return the number of disks in given state.
251 * If state is equal to -1, count all connected disks.
252 */
253u_int
254g_mirror_ndisks(struct g_mirror_softc *sc, int state)
255{
256 struct g_mirror_disk *disk;
257 u_int n = 0;
258
259 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
260 if (state == -1 || disk->d_state == state)
261 n++;
262 }
263 return (n);
264}
265
266/*
267 * Find a disk in mirror by its disk ID.
268 */
269static struct g_mirror_disk *
270g_mirror_id2disk(struct g_mirror_softc *sc, uint32_t id)
271{
272 struct g_mirror_disk *disk;
273
274 g_topology_assert();
275
276 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
277 if (disk->d_id == id)
278 return (disk);
279 }
280 return (NULL);
281}
282
283static u_int
284g_mirror_nrequests(struct g_mirror_softc *sc, struct g_consumer *cp)
285{
286 struct bio *bp;
287 u_int nreqs = 0;
288
289 mtx_lock(&sc->sc_queue_mtx);
290 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
291 if (bp->bio_from == cp)
292 nreqs++;
293 }
294 mtx_unlock(&sc->sc_queue_mtx);
295 return (nreqs);
296}
297
298static int
299g_mirror_is_busy(struct g_mirror_softc *sc, struct g_consumer *cp)
300{
301
302 if (cp->index > 0) {
303 G_MIRROR_DEBUG(2,
304 "I/O requests for %s exist, can't destroy it now.",
305 cp->provider->name);
306 return (1);
307 }
308 if (g_mirror_nrequests(sc, cp) > 0) {
309 G_MIRROR_DEBUG(2,
310 "I/O requests for %s in queue, can't destroy it now.",
311 cp->provider->name);
312 return (1);
313 }
314 return (0);
315}
316
317static void
318g_mirror_kill_consumer(struct g_mirror_softc *sc, struct g_consumer *cp)
319{
320
321 g_topology_assert();
322
323 cp->private = NULL;
324 if (g_mirror_is_busy(sc, cp))
325 return;
326 G_MIRROR_DEBUG(2, "Consumer %s destroyed.", cp->provider->name);
327 g_detach(cp);
328 g_destroy_consumer(cp);
329}
330
331static int
332g_mirror_connect_disk(struct g_mirror_disk *disk, struct g_provider *pp)
333{
334 int error;
335
336 g_topology_assert();
337 KASSERT(disk->d_consumer == NULL,
338 ("Disk already connected (device %s).", disk->d_softc->sc_name));
339
340 disk->d_consumer = g_new_consumer(disk->d_softc->sc_geom);
341 disk->d_consumer->private = disk;
342 disk->d_consumer->index = 0;
343 error = g_attach(disk->d_consumer, pp);
344 if (error != 0)
345 return (error);
346 G_MIRROR_DEBUG(2, "Disk %s connected.", g_mirror_get_diskname(disk));
347 return (0);
348}
349
350static void
351g_mirror_disconnect_consumer(struct g_mirror_softc *sc, struct g_consumer *cp)
352{
353
354 g_topology_assert();
355
356 if (cp == NULL)
357 return;
358 if (cp->provider != NULL) {
359 G_MIRROR_DEBUG(2, "Disk %s disconnected.", cp->provider->name);
360 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) {
361 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d",
362 cp->provider->name, -cp->acr, -cp->acw, -cp->ace,
363 0);
364 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
365 }
366 g_mirror_kill_consumer(sc, cp);
367 } else {
368 g_destroy_consumer(cp);
369 }
370}
371
372/*
373 * Initialize disk. This means allocate memory, create consumer, attach it
374 * to the provider and open access (r1w1e1) to it.
375 */
376static struct g_mirror_disk *
377g_mirror_init_disk(struct g_mirror_softc *sc, struct g_provider *pp,
378 struct g_mirror_metadata *md, int *errorp)
379{
380 struct g_mirror_disk *disk;
381 int error;
382
383 disk = malloc(sizeof(*disk), M_MIRROR, M_NOWAIT | M_ZERO);
384 if (disk == NULL) {
385 error = ENOMEM;
386 goto fail;
387 }
388 disk->d_softc = sc;
389 error = g_mirror_connect_disk(disk, pp);
390 if (error != 0)
391 goto fail;
392 disk->d_id = md->md_did;
393 disk->d_state = G_MIRROR_DISK_STATE_NONE;
394 disk->d_priority = md->md_priority;
395 disk->d_delay.sec = 0;
396 disk->d_delay.frac = 0;
397 binuptime(&disk->d_last_used);
398 disk->d_flags = md->md_dflags;
399 if (md->md_provider[0] != '\0')
400 disk->d_flags |= G_MIRROR_DISK_FLAG_HARDCODED;
401 disk->d_sync.ds_consumer = NULL;
402 disk->d_sync.ds_offset = md->md_sync_offset;
403 disk->d_sync.ds_offset_done = md->md_sync_offset;
404 disk->d_sync.ds_resync = -1;
405 disk->d_sync.ds_syncid = md->md_syncid;
406 if (errorp != NULL)
407 *errorp = 0;
408 return (disk);
409fail:
410 if (errorp != NULL)
411 *errorp = error;
412 if (disk != NULL) {
413 g_mirror_disconnect_consumer(sc, disk->d_consumer);
414 free(disk, M_MIRROR);
415 }
416 return (NULL);
417}
418
419static void
420g_mirror_destroy_disk(struct g_mirror_disk *disk)
421{
422 struct g_mirror_softc *sc;
423
424 g_topology_assert();
425
426 LIST_REMOVE(disk, d_next);
427 g_mirror_event_cancel(disk);
428 sc = disk->d_softc;
429 if (sc->sc_hint == disk)
430 sc->sc_hint = NULL;
431 switch (disk->d_state) {
432 case G_MIRROR_DISK_STATE_SYNCHRONIZING:
433 g_mirror_sync_stop(disk, 1);
434 /* FALLTHROUGH */
435 case G_MIRROR_DISK_STATE_NEW:
436 case G_MIRROR_DISK_STATE_STALE:
437 case G_MIRROR_DISK_STATE_ACTIVE:
438 g_mirror_disconnect_consumer(sc, disk->d_consumer);
439 free(disk, M_MIRROR);
440 break;
441 default:
442 KASSERT(0 == 1, ("Wrong disk state (%s, %s).",
443 g_mirror_get_diskname(disk),
444 g_mirror_disk_state2str(disk->d_state)));
445 }
446}
447
448static void
449g_mirror_destroy_device(struct g_mirror_softc *sc)
450{
451 struct g_mirror_disk *disk;
452 struct g_mirror_event *ep;
453 struct g_geom *gp;
454 struct g_consumer *cp, *tmpcp;
455
456 g_topology_assert();
457
458 gp = sc->sc_geom;
459 if (sc->sc_provider != NULL)
460 g_mirror_destroy_provider(sc);
461 for (disk = LIST_FIRST(&sc->sc_disks); disk != NULL;
462 disk = LIST_FIRST(&sc->sc_disks)) {
463 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY;
464 g_mirror_update_metadata(disk);
465 g_mirror_destroy_disk(disk);
466 }
467 while ((ep = g_mirror_event_get(sc)) != NULL) {
468 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0)
469 g_mirror_event_free(ep);
470 else {
471 ep->e_error = ECANCELED;
472 ep->e_flags |= G_MIRROR_EVENT_DONE;
473 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, ep);
474 mtx_lock(&sc->sc_events_mtx);
475 wakeup(ep);
476 mtx_unlock(&sc->sc_events_mtx);
477 }
478 }
479 callout_drain(&sc->sc_callout);
480 gp->softc = NULL;
481
482 LIST_FOREACH_SAFE(cp, &sc->sc_sync.ds_geom->consumer, consumer, tmpcp) {
483 g_mirror_disconnect_consumer(sc, cp);
484 }
485 sc->sc_sync.ds_geom->softc = NULL;
486 g_wither_geom(sc->sc_sync.ds_geom, ENXIO);
487 mtx_destroy(&sc->sc_queue_mtx);
488 mtx_destroy(&sc->sc_events_mtx);
489 G_MIRROR_DEBUG(0, "Device %s destroyed.", gp->name);
490 g_wither_geom(gp, ENXIO);
491}
492
493static void
494g_mirror_orphan(struct g_consumer *cp)
495{
496 struct g_mirror_disk *disk;
497
498 g_topology_assert();
499
500 disk = cp->private;
501 if (disk == NULL)
502 return;
503 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE;
504 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED,
505 G_MIRROR_EVENT_DONTWAIT);
506}
507
508static void
509g_mirror_spoiled(struct g_consumer *cp)
510{
511 struct g_mirror_disk *disk;
512
513 g_topology_assert();
514
515 disk = cp->private;
516 if (disk == NULL)
517 return;
518 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY;
519 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED,
520 G_MIRROR_EVENT_DONTWAIT);
521}
522
523/*
524 * Function should return the next active disk on the list.
525 * It is possible that it will be the same disk as given.
526 * If there are no active disks on list, NULL is returned.
527 */
528static __inline struct g_mirror_disk *
529g_mirror_find_next(struct g_mirror_softc *sc, struct g_mirror_disk *disk)
530{
531 struct g_mirror_disk *dp;
532
533 for (dp = LIST_NEXT(disk, d_next); dp != disk;
534 dp = LIST_NEXT(dp, d_next)) {
535 if (dp == NULL)
536 dp = LIST_FIRST(&sc->sc_disks);
537 if (dp->d_state == G_MIRROR_DISK_STATE_ACTIVE)
538 break;
539 }
540 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE)
541 return (NULL);
542 return (dp);
543}
544
545static struct g_mirror_disk *
546g_mirror_get_disk(struct g_mirror_softc *sc)
547{
548 struct g_mirror_disk *disk;
549
550 if (sc->sc_hint == NULL) {
551 sc->sc_hint = LIST_FIRST(&sc->sc_disks);
552 if (sc->sc_hint == NULL)
553 return (NULL);
554 }
555 disk = sc->sc_hint;
556 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE) {
557 disk = g_mirror_find_next(sc, disk);
558 if (disk == NULL)
559 return (NULL);
560 }
561 sc->sc_hint = g_mirror_find_next(sc, disk);
562 return (disk);
563}
564
565static int
566g_mirror_write_metadata(struct g_mirror_disk *disk,
567 struct g_mirror_metadata *md)
568{
569 struct g_mirror_softc *sc;
570 struct g_consumer *cp;
571 off_t offset, length;
572 u_char *sector;
573 int close = 0, error = 0;
574
575 g_topology_assert();
576
577 sc = disk->d_softc;
578 cp = disk->d_consumer;
579 KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name));
580 KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name));
581 length = cp->provider->sectorsize;
582 offset = cp->provider->mediasize - length;
583 sector = malloc((size_t)length, M_MIRROR, M_WAITOK | M_ZERO);
584 /*
585 * Open consumer if it wasn't opened and remember to close it.
586 */
587 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) == 0) {
588 error = g_access(cp, 0, 1, 1);
589 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d",
590 cp->provider->name, 0, 1, 1, error);
591 if (error == 0)
592 close = 1;
593#ifdef INVARIANTS
594 } else {
595 KASSERT(cp->acw > 0 && cp->ace > 0,
596 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name,
597 cp->acr, cp->acw, cp->ace));
598#endif
599 }
600 if (error == 0) {
601 if (md != NULL)
602 mirror_metadata_encode(md, sector);
603 g_topology_unlock();
604 error = g_write_data(cp, offset, sector, length);
605 g_topology_lock();
606 }
607 free(sector, M_MIRROR);
608 if (close) {
609 g_access(cp, 0, -1, -1);
610 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d",
611 cp->provider->name, 0, -1, -1, 0);
612 }
613 if (error != 0) {
614 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY;
615 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED,
616 G_MIRROR_EVENT_DONTWAIT);
617 }
618 return (error);
619}
620
621static int
622g_mirror_clear_metadata(struct g_mirror_disk *disk)
623{
624 int error;
625
626 g_topology_assert();
627 error = g_mirror_write_metadata(disk, NULL);
628 if (error == 0) {
629 G_MIRROR_DEBUG(2, "Metadata on %s cleared.",
630 g_mirror_get_diskname(disk));
631 } else {
632 G_MIRROR_DEBUG(0,
633 "Cannot clear metadata on disk %s (error=%d).",
634 g_mirror_get_diskname(disk), error);
635 }
636 return (error);
637}
638
639void
640g_mirror_fill_metadata(struct g_mirror_softc *sc, struct g_mirror_disk *disk,
641 struct g_mirror_metadata *md)
642{
643
644 strlcpy(md->md_magic, G_MIRROR_MAGIC, sizeof(md->md_magic));
645 md->md_version = G_MIRROR_VERSION;
646 strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name));
647 md->md_mid = sc->sc_id;
648 md->md_all = sc->sc_ndisks;
649 md->md_slice = sc->sc_slice;
650 md->md_balance = sc->sc_balance;
651 md->md_mediasize = sc->sc_mediasize;
652 md->md_sectorsize = sc->sc_sectorsize;
653 md->md_mflags = (sc->sc_flags & G_MIRROR_DEVICE_FLAG_MASK);
654 bzero(md->md_provider, sizeof(md->md_provider));
655 if (disk == NULL) {
656 md->md_did = arc4random();
657 md->md_priority = 0;
658 md->md_syncid = 0;
659 md->md_dflags = 0;
660 md->md_sync_offset = 0;
661 } else {
662 md->md_did = disk->d_id;
663 md->md_priority = disk->d_priority;
664 md->md_syncid = disk->d_sync.ds_syncid;
665 md->md_dflags = (disk->d_flags & G_MIRROR_DISK_FLAG_MASK);
666 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING)
667 md->md_sync_offset = disk->d_sync.ds_offset_done;
668 else
669 md->md_sync_offset = 0;
670 if ((disk->d_flags & G_MIRROR_DISK_FLAG_HARDCODED) != 0) {
671 strlcpy(md->md_provider,
672 disk->d_consumer->provider->name,
673 sizeof(md->md_provider));
674 }
675 }
676}
677
678void
679g_mirror_update_metadata(struct g_mirror_disk *disk)
680{
681 struct g_mirror_metadata md;
682 int error;
683
684 g_topology_assert();
685 g_mirror_fill_metadata(disk->d_softc, disk, &md);
686 error = g_mirror_write_metadata(disk, &md);
687 if (error == 0) {
688 G_MIRROR_DEBUG(2, "Metadata on %s updated.",
689 g_mirror_get_diskname(disk));
690 } else {
691 G_MIRROR_DEBUG(0,
692 "Cannot update metadata on disk %s (error=%d).",
693 g_mirror_get_diskname(disk), error);
694 }
695}
696
697static void
698g_mirror_bump_syncid(struct g_mirror_softc *sc)
699{
700 struct g_mirror_disk *disk;
701
702 g_topology_assert();
703 KASSERT(g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) > 0,
704 ("%s called with no active disks (device=%s).", __func__,
705 sc->sc_name));
706
707 sc->sc_syncid++;
708 G_MIRROR_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name,
709 sc->sc_syncid);
710 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
711 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE ||
712 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) {
713 disk->d_sync.ds_syncid = sc->sc_syncid;
714 g_mirror_update_metadata(disk);
715 }
716 }
717}
718
719static void
720g_mirror_idle(struct g_mirror_softc *sc)
721{
722 struct g_mirror_disk *disk;
723
724 if (sc->sc_provider == NULL || sc->sc_provider->acw == 0)
725 return;
726 sc->sc_idle = 1;
727 g_topology_lock();
728 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
729 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE)
730 continue;
731 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.",
732 g_mirror_get_diskname(disk), sc->sc_name);
733 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY;
734 g_mirror_update_metadata(disk);
735 }
736 g_topology_unlock();
737}
738
739static void
740g_mirror_unidle(struct g_mirror_softc *sc)
741{
742 struct g_mirror_disk *disk;
743
744 sc->sc_idle = 0;
745 g_topology_lock();
746 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
747 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE)
748 continue;
749 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.",
750 g_mirror_get_diskname(disk), sc->sc_name);
751 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY;
752 g_mirror_update_metadata(disk);
753 }
754 g_topology_unlock();
755}
756
757static __inline int
758bintime_cmp(struct bintime *bt1, struct bintime *bt2)
759{
760
761 if (bt1->sec < bt2->sec)
762 return (-1);
763 else if (bt1->sec > bt2->sec)
764 return (1);
765 if (bt1->frac < bt2->frac)
766 return (-1);
767 else if (bt1->frac > bt2->frac)
768 return (1);
769 return (0);
770}
771
772static void
773g_mirror_update_delay(struct g_mirror_disk *disk, struct bio *bp)
774{
775
776 if (disk->d_softc->sc_balance != G_MIRROR_BALANCE_LOAD)
777 return;
778 binuptime(&disk->d_delay);
779 bintime_sub(&disk->d_delay, &bp->bio_t0);
780}
781
782static void
783g_mirror_done(struct bio *bp)
784{
785 struct g_mirror_softc *sc;
786
787 sc = bp->bio_from->geom->softc;
788 bp->bio_cflags |= G_MIRROR_BIO_FLAG_REGULAR;
789 mtx_lock(&sc->sc_queue_mtx);
790 bioq_disksort(&sc->sc_queue, bp);
791 wakeup(sc);
792 mtx_unlock(&sc->sc_queue_mtx);
793}
794
795static void
796g_mirror_regular_request(struct bio *bp)
797{
798 struct g_mirror_softc *sc;
799 struct g_mirror_disk *disk;
800 struct bio *pbp;
801
802 g_topology_assert_not();
803
804 bp->bio_from->index--;
805 pbp = bp->bio_parent;
806 sc = pbp->bio_to->geom->softc;
807 disk = bp->bio_from->private;
808 if (disk == NULL) {
809 g_topology_lock();
810 g_mirror_kill_consumer(sc, bp->bio_from);
811 g_topology_unlock();
812 } else {
813 g_mirror_update_delay(disk, bp);
814 }
815
816 pbp->bio_inbed++;
817 KASSERT(pbp->bio_inbed <= pbp->bio_children,
818 ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed,
819 pbp->bio_children));
820 if (bp->bio_error == 0 && pbp->bio_error == 0) {
821 G_MIRROR_LOGREQ(3, bp, "Request delivered.");
822 g_destroy_bio(bp);
823 if (pbp->bio_children == pbp->bio_inbed) {
824 G_MIRROR_LOGREQ(3, pbp, "Request delivered.");
825 pbp->bio_completed = pbp->bio_length;
826 g_io_deliver(pbp, pbp->bio_error);
827 }
828 return;
829 } else if (bp->bio_error != 0) {
830 if (pbp->bio_error == 0)
831 pbp->bio_error = bp->bio_error;
832 G_MIRROR_LOGREQ(0, bp, "Request failed (error=%d).",
833 bp->bio_error);
834 if (disk != NULL) {
835 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY;
836 g_mirror_event_send(disk,
837 G_MIRROR_DISK_STATE_DISCONNECTED,
838 G_MIRROR_EVENT_DONTWAIT);
839 }
840 switch (pbp->bio_cmd) {
841 case BIO_DELETE:
842 case BIO_WRITE:
843 pbp->bio_inbed--;
844 pbp->bio_children--;
845 break;
846 }
847 }
848 g_destroy_bio(bp);
849
850 switch (pbp->bio_cmd) {
851 case BIO_READ:
852 if (pbp->bio_children == pbp->bio_inbed) {
853 pbp->bio_error = 0;
854 mtx_lock(&sc->sc_queue_mtx);
855 bioq_disksort(&sc->sc_queue, pbp);
856 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc);
857 wakeup(sc);
858 mtx_unlock(&sc->sc_queue_mtx);
859 }
860 break;
861 case BIO_DELETE:
862 case BIO_WRITE:
863 if (pbp->bio_children == 0) {
864 /*
865 * All requests failed.
866 */
867 } else if (pbp->bio_inbed < pbp->bio_children) {
868 /* Do nothing. */
869 break;
870 } else if (pbp->bio_children == pbp->bio_inbed) {
871 /* Some requests succeeded. */
872 pbp->bio_error = 0;
873 pbp->bio_completed = pbp->bio_length;
874 }
875 g_io_deliver(pbp, pbp->bio_error);
876 break;
877 default:
878 KASSERT(1 == 0, ("Invalid request: %u.", pbp->bio_cmd));
879 break;
880 }
881}
882
883static void
884g_mirror_sync_done(struct bio *bp)
885{
886 struct g_mirror_softc *sc;
887
888 G_MIRROR_LOGREQ(3, bp, "Synchronization request delivered.");
889 sc = bp->bio_from->geom->softc;
890 bp->bio_cflags |= G_MIRROR_BIO_FLAG_SYNC;
891 mtx_lock(&sc->sc_queue_mtx);
892 bioq_disksort(&sc->sc_queue, bp);
893 wakeup(sc);
894 mtx_unlock(&sc->sc_queue_mtx);
895}
896
897static void
898g_mirror_start(struct bio *bp)
899{
900 struct g_mirror_softc *sc;
901
902 sc = bp->bio_to->geom->softc;
903 /*
904 * If sc == NULL or there are no valid disks, provider's error
905 * should be set and g_mirror_start() should not be called at all.
906 */
907 KASSERT(sc != NULL && sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING,
908 ("Provider's error should be set (error=%d)(mirror=%s).",
909 bp->bio_to->error, bp->bio_to->name));
910 G_MIRROR_LOGREQ(3, bp, "Request received.");
911
912 switch (bp->bio_cmd) {
913 case BIO_READ:
914 case BIO_WRITE:
915 case BIO_DELETE:
916 break;
917 case BIO_GETATTR:
918 default:
919 g_io_deliver(bp, EOPNOTSUPP);
920 return;
921 }
922 mtx_lock(&sc->sc_queue_mtx);
923 bioq_disksort(&sc->sc_queue, bp);
924 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc);
925 wakeup(sc);
926 mtx_unlock(&sc->sc_queue_mtx);
927}
928
929/*
930 * Send one synchronization request.
931 */
932static void
933g_mirror_sync_one(struct g_mirror_disk *disk)
934{
935 struct g_mirror_softc *sc;
936 struct bio *bp;
937
938 sc = disk->d_softc;
939 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING,
940 ("Disk %s is not marked for synchronization.",
941 g_mirror_get_diskname(disk)));
942
943 bp = g_new_bio();
944 if (bp == NULL)
945 return;
946 bp->bio_parent = NULL;
947 bp->bio_cmd = BIO_READ;
948 bp->bio_offset = disk->d_sync.ds_offset;
949 bp->bio_length = MIN(MAXPHYS, sc->sc_mediasize - bp->bio_offset);
950 bp->bio_cflags = 0;
951 bp->bio_done = g_mirror_sync_done;
952 bp->bio_data = disk->d_sync.ds_data;
953 if (bp->bio_data == NULL) {
954 g_destroy_bio(bp);
955 return;
956 }
957 disk->d_sync.ds_offset += bp->bio_length;
958 bp->bio_to = sc->sc_provider;
959 G_MIRROR_LOGREQ(3, bp, "Sending synchronization request.");
960 disk->d_sync.ds_consumer->index++;
961 g_io_request(bp, disk->d_sync.ds_consumer);
962}
963
964static void
965g_mirror_sync_request(struct bio *bp)
966{
967 struct g_mirror_softc *sc;
968 struct g_mirror_disk *disk;
969
970 bp->bio_from->index--;
971 sc = bp->bio_from->geom->softc;
972 disk = bp->bio_from->private;
973 if (disk == NULL) {
974 g_topology_lock();
975 g_mirror_kill_consumer(sc, bp->bio_from);
976 g_topology_unlock();
977 g_destroy_bio(bp);
978 return;
979 }
980
981 /*
982 * Synchronization request.
983 */
984 switch (bp->bio_cmd) {
985 case BIO_READ:
986 {
987 struct g_consumer *cp;
988
989 if (bp->bio_error != 0) {
990 G_MIRROR_LOGREQ(0, bp,
991 "Synchronization request failed (error=%d).",
992 bp->bio_error);
993 g_destroy_bio(bp);
994 return;
995 }
996 G_MIRROR_LOGREQ(3, bp,
997 "Synchronization request half-finished.");
998 bp->bio_cmd = BIO_WRITE;
999 bp->bio_cflags = 0;
1000 cp = disk->d_consumer;
1001 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1,
1002 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name,
1003 cp->acr, cp->acw, cp->ace));
1004 cp->index++;
1005 g_io_request(bp, cp);
1006 return;
1007 }
1008 case BIO_WRITE:
1009 {
1010 struct g_mirror_disk_sync *sync;
1011
1012 if (bp->bio_error != 0) {
1013 G_MIRROR_LOGREQ(0, bp,
1014 "Synchronization request failed (error=%d).",
1015 bp->bio_error);
1016 g_destroy_bio(bp);
1017 sc->sc_bump_syncid = G_MIRROR_BUMP_IMMEDIATELY;
1018 g_mirror_event_send(disk,
1019 G_MIRROR_DISK_STATE_DISCONNECTED,
1020 G_MIRROR_EVENT_DONTWAIT);
1021 return;
1022 }
1023 G_MIRROR_LOGREQ(3, bp, "Synchronization request finished.");
1024 sync = &disk->d_sync;
1025 sync->ds_offset_done = bp->bio_offset + bp->bio_length;
1026 g_destroy_bio(bp);
1027 if (sync->ds_resync != -1)
1028 break;
1029 if (sync->ds_offset_done == sc->sc_provider->mediasize) {
1030 /*
1031 * Disk up-to-date, activate it.
1032 */
1033 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_ACTIVE,
1034 G_MIRROR_EVENT_DONTWAIT);
1035 return;
1036 } else if (sync->ds_offset_done % (MAXPHYS * 100) == 0) {
1037 /*
1038 * Update offset_done on every 100 blocks.
1039 * XXX: This should be configurable.
1040 */
1041 g_topology_lock();
1042 g_mirror_update_metadata(disk);
1043 g_topology_unlock();
1044 }
1045 return;
1046 }
1047 default:
1048 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)",
1049 bp->bio_cmd, sc->sc_name));
1050 break;
1051 }
1052}
1053
1054static void
1055g_mirror_request_prefer(struct g_mirror_softc *sc, struct bio *bp)
1056{
1057 struct g_mirror_disk *disk;
1058 struct g_consumer *cp;
1059 struct bio *cbp;
1060
1061 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
1062 if (disk->d_state == G_MIRROR_DISK_STATE_ACTIVE)
1063 break;
1064 }
1065 if (disk == NULL) {
1066 if (bp->bio_error == 0)
1067 bp->bio_error = ENXIO;
1068 g_io_deliver(bp, bp->bio_error);
1069 return;
1070 }
1071 cbp = g_clone_bio(bp);
1072 if (cbp == NULL) {
1073 if (bp->bio_error == 0)
1074 bp->bio_error = ENOMEM;
1075 g_io_deliver(bp, bp->bio_error);
1076 return;
1077 }
1078 /*
1079 * Fill in the component buf structure.
1080 */
1081 cp = disk->d_consumer;
1082 cbp->bio_done = g_mirror_done;
1083 cbp->bio_to = cp->provider;
1084 G_MIRROR_LOGREQ(3, cbp, "Sending request.");
1085 KASSERT(cp->acr > 0 && cp->ace > 0,
1086 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr,
1087 cp->acw, cp->ace));
1088 cp->index++;
1089 g_io_request(cbp, cp);
1090}
1091
1092static void
1093g_mirror_request_round_robin(struct g_mirror_softc *sc, struct bio *bp)
1094{
1095 struct g_mirror_disk *disk;
1096 struct g_consumer *cp;
1097 struct bio *cbp;
1098
1099 disk = g_mirror_get_disk(sc);
1100 if (disk == NULL) {
1101 if (bp->bio_error == 0)
1102 bp->bio_error = ENXIO;
1103 g_io_deliver(bp, bp->bio_error);
1104 return;
1105 }
1106 cbp = g_clone_bio(bp);
1107 if (cbp == NULL) {
1108 if (bp->bio_error == 0)
1109 bp->bio_error = ENOMEM;
1110 g_io_deliver(bp, bp->bio_error);
1111 return;
1112 }
1113 /*
1114 * Fill in the component buf structure.
1115 */
1116 cp = disk->d_consumer;
1117 cbp->bio_done = g_mirror_done;
1118 cbp->bio_to = cp->provider;
1119 G_MIRROR_LOGREQ(3, cbp, "Sending request.");
1120 KASSERT(cp->acr > 0 && cp->ace > 0,
1121 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr,
1122 cp->acw, cp->ace));
1123 cp->index++;
1124 g_io_request(cbp, cp);
1125}
1126
1127static void
1128g_mirror_request_load(struct g_mirror_softc *sc, struct bio *bp)
1129{
1130 struct g_mirror_disk *disk, *dp;
1131 struct g_consumer *cp;
1132 struct bio *cbp;
1133 struct bintime curtime;
1134
1135 binuptime(&curtime);
1136 /*
1137 * Find a disk which the smallest load.
1138 */
1139 disk = NULL;
1140 LIST_FOREACH(dp, &sc->sc_disks, d_next) {
1141 if (dp->d_state != G_MIRROR_DISK_STATE_ACTIVE)
1142 continue;
1143 /* If disk wasn't used for more than 2 sec, use it. */
1144 if (curtime.sec - dp->d_last_used.sec >= 2) {
1145 disk = dp;
1146 break;
1147 }
1148 if (disk == NULL ||
1149 bintime_cmp(&dp->d_delay, &disk->d_delay) < 0) {
1150 disk = dp;
1151 }
1152 }
1153 cbp = g_clone_bio(bp);
1154 if (cbp == NULL) {
1155 if (bp->bio_error == 0)
1156 bp->bio_error = ENOMEM;
1157 g_io_deliver(bp, bp->bio_error);
1158 return;
1159 }
1160 /*
1161 * Fill in the component buf structure.
1162 */
1163 cp = disk->d_consumer;
1164 cbp->bio_done = g_mirror_done;
1165 cbp->bio_to = cp->provider;
1166 binuptime(&disk->d_last_used);
1167 G_MIRROR_LOGREQ(3, cbp, "Sending request.");
1168 KASSERT(cp->acr > 0 && cp->ace > 0,
1169 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name, cp->acr,
1170 cp->acw, cp->ace));
1171 cp->index++;
1172 g_io_request(cbp, cp);
1173}
1174
1175static void
1176g_mirror_request_split(struct g_mirror_softc *sc, struct bio *bp)
1177{
1178 struct bio_queue_head queue;
1179 struct g_mirror_disk *disk;
1180 struct g_consumer *cp;
1181 struct bio *cbp;
1182 off_t left, mod, offset, slice;
1183 u_char *data;
1184 u_int ndisks;
1185
1186 if (bp->bio_length <= sc->sc_slice) {
1187 g_mirror_request_round_robin(sc, bp);
1188 return;
1189 }
1190 ndisks = g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE);
1191 slice = bp->bio_length / ndisks;
1192 mod = slice % sc->sc_provider->sectorsize;
1193 if (mod != 0)
1194 slice += sc->sc_provider->sectorsize - mod;
1195 /*
1196 * Allocate all bios before sending any request, so we can
1197 * return ENOMEM in nice and clean way.
1198 */
1199 left = bp->bio_length;
1200 offset = bp->bio_offset;
1201 data = bp->bio_data;
1202 bioq_init(&queue);
1203 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
1204 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE)
1205 continue;
1206 cbp = g_clone_bio(bp);
1207 if (cbp == NULL) {
1208 for (cbp = bioq_first(&queue); cbp != NULL;
1209 cbp = bioq_first(&queue)) {
1210 bioq_remove(&queue, cbp);
1211 g_destroy_bio(cbp);
1212 }
1213 if (bp->bio_error == 0)
1214 bp->bio_error = ENOMEM;
1215 g_io_deliver(bp, bp->bio_error);
1216 return;
1217 }
1218 bioq_insert_tail(&queue, cbp);
1219 cbp->bio_done = g_mirror_done;
1220 cbp->bio_caller1 = disk;
1221 cbp->bio_to = disk->d_consumer->provider;
1222 cbp->bio_offset = offset;
1223 cbp->bio_data = data;
1224 cbp->bio_length = MIN(left, slice);
1225 left -= cbp->bio_length;
1226 if (left == 0)
1227 break;
1228 offset += cbp->bio_length;
1229 data += cbp->bio_length;
1230 }
1231 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) {
1232 bioq_remove(&queue, cbp);
1233 G_MIRROR_LOGREQ(3, cbp, "Sending request.");
1234 disk = cbp->bio_caller1;
1235 cbp->bio_caller1 = NULL;
1236 cp = disk->d_consumer;
1237 KASSERT(cp->acr > 0 && cp->ace > 0,
1238 ("Consumer %s not opened (r%dw%de%d).", cp->provider->name,
1239 cp->acr, cp->acw, cp->ace));
1240 disk->d_consumer->index++;
1241 g_io_request(cbp, disk->d_consumer);
1242 }
1243}
1244
1245static void
1246g_mirror_register_request(struct bio *bp)
1247{
1248 struct g_mirror_softc *sc;
1249
1250 sc = bp->bio_to->geom->softc;
1251 switch (bp->bio_cmd) {
1252 case BIO_READ:
1253 switch (sc->sc_balance) {
1254 case G_MIRROR_BALANCE_LOAD:
1255 g_mirror_request_load(sc, bp);
1256 break;
1257 case G_MIRROR_BALANCE_PREFER:
1258 g_mirror_request_prefer(sc, bp);
1259 break;
1260 case G_MIRROR_BALANCE_ROUND_ROBIN:
1261 g_mirror_request_round_robin(sc, bp);
1262 break;
1263 case G_MIRROR_BALANCE_SPLIT:
1264 g_mirror_request_split(sc, bp);
1265 break;
1266 }
1267 return;
1268 case BIO_WRITE:
1269 case BIO_DELETE:
1270 {
1271 struct g_mirror_disk *disk;
1272 struct g_mirror_disk_sync *sync;
1273 struct bio_queue_head queue;
1274 struct g_consumer *cp;
1275 struct bio *cbp;
1276
1277 if (sc->sc_idle)
1278 g_mirror_unidle(sc);
1279 /*
1280 * Allocate all bios before sending any request, so we can
1281 * return ENOMEM in nice and clean way.
1282 */
1283 bioq_init(&queue);
1284 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
1285 sync = &disk->d_sync;
1286 switch (disk->d_state) {
1287 case G_MIRROR_DISK_STATE_ACTIVE:
1288 break;
1289 case G_MIRROR_DISK_STATE_SYNCHRONIZING:
1290 if (bp->bio_offset >= sync->ds_offset)
1291 continue;
1292 else if (bp->bio_offset + bp->bio_length >
1293 sync->ds_offset_done &&
1294 (bp->bio_offset < sync->ds_resync ||
1295 sync->ds_resync == -1)) {
1296 sync->ds_resync = bp->bio_offset -
1297 (bp->bio_offset % MAXPHYS);
1298 }
1299 break;
1300 default:
1301 continue;
1302 }
1303 cbp = g_clone_bio(bp);
1304 if (cbp == NULL) {
1305 for (cbp = bioq_first(&queue); cbp != NULL;
1306 cbp = bioq_first(&queue)) {
1307 bioq_remove(&queue, cbp);
1308 g_destroy_bio(cbp);
1309 }
1310 if (bp->bio_error == 0)
1311 bp->bio_error = ENOMEM;
1312 g_io_deliver(bp, bp->bio_error);
1313 return;
1314 }
1315 bioq_insert_tail(&queue, cbp);
1316 cbp->bio_done = g_mirror_done;
1317 cp = disk->d_consumer;
1318 cbp->bio_caller1 = cp;
1319 cbp->bio_to = cp->provider;
1320 KASSERT(cp->acw > 0 && cp->ace > 0,
1321 ("Consumer %s not opened (r%dw%de%d).",
1322 cp->provider->name, cp->acr, cp->acw, cp->ace));
1323 }
1324 for (cbp = bioq_first(&queue); cbp != NULL;
1325 cbp = bioq_first(&queue)) {
1326 bioq_remove(&queue, cbp);
1327 G_MIRROR_LOGREQ(3, cbp, "Sending request.");
1328 cp = cbp->bio_caller1;
1329 cbp->bio_caller1 = NULL;
1330 cp->index++;
1331 g_io_request(cbp, cp);
1332 }
1333 /*
1334 * Bump syncid on first write.
1335 */
1336 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE) {
1337 sc->sc_bump_syncid = 0;
1338 g_topology_lock();
1339 g_mirror_bump_syncid(sc);
1340 g_topology_unlock();
1341 }
1342 return;
1343 }
1344 default:
1345 KASSERT(1 == 0, ("Invalid command here: %u (device=%s)",
1346 bp->bio_cmd, sc->sc_name));
1347 break;
1348 }
1349}
1350
1351static int
1352g_mirror_can_destroy(struct g_mirror_softc *sc)
1353{
1354 struct g_geom *gp;
1355 struct g_consumer *cp;
1356
1357 g_topology_assert();
1358 gp = sc->sc_geom;
1359 LIST_FOREACH(cp, &gp->consumer, consumer) {
1360 if (g_mirror_is_busy(sc, cp))
1361 return (0);
1362 }
1363 gp = sc->sc_sync.ds_geom;
1364 LIST_FOREACH(cp, &gp->consumer, consumer) {
1365 if (g_mirror_is_busy(sc, cp))
1366 return (0);
1367 }
1368 G_MIRROR_DEBUG(2, "No I/O requests for %s, it can be destroyed.",
1369 sc->sc_name);
1370 return (1);
1371}
1372
1373static int
1374g_mirror_try_destroy(struct g_mirror_softc *sc)
1375{
1376
1377 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_WAIT) != 0) {
1378 g_topology_lock();
1379 if (!g_mirror_can_destroy(sc)) {
1380 g_topology_unlock();
1381 return (0);
1382 }
1383 g_topology_unlock();
1384 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__,
1385 &sc->sc_worker);
1386 wakeup(&sc->sc_worker);
1387 sc->sc_worker = NULL;
1388 } else {
1389 g_topology_lock();
1390 if (!g_mirror_can_destroy(sc)) {
1391 g_topology_unlock();
1392 return (0);
1393 }
1394 g_mirror_destroy_device(sc);
1395 g_topology_unlock();
1396 free(sc, M_MIRROR);
1397 }
1398 return (1);
1399}
1400
1401/*
1402 * Worker thread.
1403 */
1404static void
1405g_mirror_worker(void *arg)
1406{
1407 struct g_mirror_softc *sc;
1408 struct g_mirror_disk *disk;
1409 struct g_mirror_disk_sync *sync;
1410 struct g_mirror_event *ep;
1411 struct bio *bp;
1412 u_int nreqs;
1413
1414 sc = arg;
1415 curthread->td_base_pri = PRIBIO;
1416
1417 nreqs = 0;
1418 for (;;) {
1419 G_MIRROR_DEBUG(5, "%s: Let's see...", __func__);
1420 /*
1421 * First take a look at events.
1422 * This is important to handle events before any I/O requests.
1423 */
1424 ep = g_mirror_event_get(sc);
1425 if (ep != NULL) {
1426 g_topology_lock();
1427 if ((ep->e_flags & G_MIRROR_EVENT_DEVICE) != 0) {
1428 /* Update only device status. */
1429 G_MIRROR_DEBUG(3,
1430 "Running event for device %s.",
1431 sc->sc_name);
1432 ep->e_error = 0;
1433 g_mirror_update_device(sc, 1);
1434 } else {
1435 /* Update disk status. */
1436 G_MIRROR_DEBUG(3, "Running event for disk %s.",
1437 g_mirror_get_diskname(ep->e_disk));
1438 ep->e_error = g_mirror_update_disk(ep->e_disk,
1439 ep->e_state);
1440 if (ep->e_error == 0)
1441 g_mirror_update_device(sc, 0);
1442 }
1443 g_topology_unlock();
1444 if ((ep->e_flags & G_MIRROR_EVENT_DONTWAIT) != 0) {
1445 KASSERT(ep->e_error == 0,
1446 ("Error cannot be handled."));
1447 g_mirror_event_free(ep);
1448 } else {
1449 ep->e_flags |= G_MIRROR_EVENT_DONE;
1450 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__,
1451 ep);
1452 mtx_lock(&sc->sc_events_mtx);
1453 wakeup(ep);
1454 mtx_unlock(&sc->sc_events_mtx);
1455 }
1456 if ((sc->sc_flags &
1457 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) {
1458 if (g_mirror_try_destroy(sc))
1459 kthread_exit(0);
1460 }
1461 G_MIRROR_DEBUG(5, "%s: I'm here 1.", __func__);
1462 continue;
1463 }
1464 /*
1465 * Now I/O requests.
1466 */
1467 /* Get first request from the queue. */
1468 mtx_lock(&sc->sc_queue_mtx);
1469 bp = bioq_first(&sc->sc_queue);
1470 if (bp == NULL) {
1471 if ((sc->sc_flags &
1472 G_MIRROR_DEVICE_FLAG_DESTROY) != 0) {
1473 mtx_unlock(&sc->sc_queue_mtx);
1474 if (g_mirror_try_destroy(sc))
1475 kthread_exit(0);
1476 mtx_lock(&sc->sc_queue_mtx);
1477 }
1478 }
1479 if (sc->sc_sync.ds_ndisks > 0 &&
1480 (bp == NULL || nreqs > g_mirror_reqs_per_sync)) {
1481 mtx_unlock(&sc->sc_queue_mtx);
1482 /*
1483 * It is time for synchronization...
1484 */
1485 nreqs = 0;
1486 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
1487 if (disk->d_state !=
1488 G_MIRROR_DISK_STATE_SYNCHRONIZING) {
1489 continue;
1490 }
1491 sync = &disk->d_sync;
1492 if (sync->ds_offset >=
1493 sc->sc_provider->mediasize) {
1494 continue;
1495 }
1496 if (sync->ds_offset > sync->ds_offset_done)
1497 continue;
1498 if (sync->ds_resync != -1) {
1499 sync->ds_offset = sync->ds_resync;
1500 sync->ds_offset_done = sync->ds_resync;
1501 sync->ds_resync = -1;
1502 }
1503 g_mirror_sync_one(disk);
1504 }
1505 G_MIRROR_DEBUG(5, "%s: I'm here 2.", __func__);
1506 goto sleep;
1507 }
1508 if (bp == NULL) {
1509#define G_MIRROR_IS_IDLE(sc) ((sc)->sc_idle || \
1510 ((sc)->sc_provider != NULL && \
1511 (sc)->sc_provider->acw == 0))
1512 if (G_MIRROR_IS_IDLE(sc)) {
1513 /*
1514 * If we're already in idle state, sleep without
1515 * a timeout.
1516 */
1517 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP,
1518 "m:w1", 0);
1519 G_MIRROR_DEBUG(5, "%s: I'm here 3.", __func__);
1520 } else {
1521 u_int idletime;
1522
1523 idletime = g_mirror_idletime;
1524 if (idletime == 0)
1525 idletime = 1;
1526 idletime *= hz;
1527 if (msleep(sc, &sc->sc_queue_mtx, PRIBIO | PDROP,
1528 "m:w2", idletime) == EWOULDBLOCK) {
1529 G_MIRROR_DEBUG(5, "%s: I'm here 4.",
1530 __func__);
1531 /*
1532 * No I/O requests in 5 seconds, so mark
1533 * components as clean.
1534 */
1535 g_mirror_idle(sc);
1536 }
1537 G_MIRROR_DEBUG(5, "%s: I'm here 5.", __func__);
1538 }
1539 continue;
1540 }
1541 nreqs++;
1542 bioq_remove(&sc->sc_queue, bp);
1543 mtx_unlock(&sc->sc_queue_mtx);
1544
1545 if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_REGULAR) != 0) {
1546 g_mirror_regular_request(bp);
1547 } else if ((bp->bio_cflags & G_MIRROR_BIO_FLAG_SYNC) != 0) {
1548 u_int timeout, sps;
1549
1550 g_mirror_sync_request(bp);
1551sleep:
1552 sps = g_mirror_syncs_per_sec;
1553 if (sps == 0) {
1554 G_MIRROR_DEBUG(5, "%s: I'm here 6.", __func__);
1555 continue;
1556 }
1557 mtx_lock(&sc->sc_queue_mtx);
1558 if (bioq_first(&sc->sc_queue) != NULL) {
1559 mtx_unlock(&sc->sc_queue_mtx);
1560 G_MIRROR_DEBUG(5, "%s: I'm here 7.", __func__);
1561 continue;
1562 }
1563 timeout = hz / sps;
1564 if (timeout == 0)
1565 timeout = 1;
1566 MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:w3",
1567 timeout);
1568 } else {
1569 g_mirror_register_request(bp);
1570 }
1571 G_MIRROR_DEBUG(5, "%s: I'm here 8.", __func__);
1572 }
1573}
1574
1575/*
1576 * Open disk's consumer if needed.
1577 */
1578static void
1579g_mirror_update_access(struct g_mirror_disk *disk)
1580{
1581 struct g_provider *pp;
1582 struct g_consumer *cp;
1583 int acr, acw, ace, cpw, error;
1584
1585 g_topology_assert();
1586
1587 cp = disk->d_consumer;
1588 pp = disk->d_softc->sc_provider;
1589 if (pp == NULL) {
1590 acr = -cp->acr;
1591 acw = -cp->acw;
1592 ace = -cp->ace;
1593 } else {
1594 acr = pp->acr - cp->acr;
1595 acw = pp->acw - cp->acw;
1596 ace = pp->ace - cp->ace;
1597 /* Grab an extra "exclusive" bit. */
1598 if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0)
1599 ace++;
1600 }
1601 if (acr == 0 && acw == 0 && ace == 0)
1602 return;
1603 cpw = cp->acw;
1604 error = g_access(cp, acr, acw, ace);
1605 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, acr,
1606 acw, ace, error);
1607 if (error != 0) {
1608 disk->d_softc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE;
1609 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED,
1610 G_MIRROR_EVENT_DONTWAIT);
1611 return;
1612 }
1613 if (cpw == 0 && cp->acw > 0) {
1614 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as dirty.",
1615 g_mirror_get_diskname(disk), disk->d_softc->sc_name);
1616 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY;
1617 } else if (cpw > 0 && cp->acw == 0) {
1618 G_MIRROR_DEBUG(1, "Disk %s (device %s) marked as clean.",
1619 g_mirror_get_diskname(disk), disk->d_softc->sc_name);
1620 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY;
1621 }
1622}
1623
1624static void
1625g_mirror_sync_start(struct g_mirror_disk *disk)
1626{
1627 struct g_mirror_softc *sc;
1628 struct g_consumer *cp;
1629 int error;
1630
1631 g_topology_assert();
1632
1633 sc = disk->d_softc;
1634 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING,
1635 ("Device not in RUNNING state (%s, %u).", sc->sc_name,
1636 sc->sc_state));
1637 cp = disk->d_consumer;
1638 KASSERT(cp->acr == 0 && cp->acw == 0 && cp->ace == 0,
1639 ("Consumer %s already opened.", cp->provider->name));
1640
1641 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name,
1642 g_mirror_get_diskname(disk));
1643 error = g_access(cp, 0, 1, 1);
1644 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, 1,
1645 1, error);
1646 if (error != 0) {
1647 g_mirror_event_send(disk, G_MIRROR_DISK_STATE_DISCONNECTED,
1648 G_MIRROR_EVENT_DONTWAIT);
1649 return;
1650 }
1651 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY;
1652 KASSERT(disk->d_sync.ds_consumer == NULL,
1653 ("Sync consumer already exists (device=%s, disk=%s).",
1654 sc->sc_name, g_mirror_get_diskname(disk)));
1655 disk->d_sync.ds_consumer = g_new_consumer(sc->sc_sync.ds_geom);
1656 disk->d_sync.ds_consumer->private = disk;
1657 disk->d_sync.ds_consumer->index = 0;
1658 error = g_attach(disk->d_sync.ds_consumer, disk->d_softc->sc_provider);
1659 KASSERT(error == 0, ("Cannot attach to %s (error=%d).",
1660 disk->d_softc->sc_name, error));
1661 error = g_access(disk->d_sync.ds_consumer, 1, 0, 0);
1662 KASSERT(error == 0, ("Cannot open %s (error=%d).",
1663 disk->d_softc->sc_name, error));
1664 disk->d_sync.ds_data = malloc(MAXPHYS, M_MIRROR, M_WAITOK);
1665 sc->sc_sync.ds_ndisks++;
1666}
1667
1668/*
1669 * Stop synchronization process.
1670 * type: 0 - synchronization finished
1671 * 1 - synchronization stopped
1672 */
1673static void
1674g_mirror_sync_stop(struct g_mirror_disk *disk, int type)
1675{
1676 struct g_consumer *cp;
1677
1678 g_topology_assert();
1679 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING,
1680 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk),
1681 g_mirror_disk_state2str(disk->d_state)));
1682 if (disk->d_sync.ds_consumer == NULL)
1683 return;
1684
1685 if (type == 0) {
1686 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s finished.",
1687 disk->d_softc->sc_name, g_mirror_get_diskname(disk));
1688 } else /* if (type == 1) */ {
1689 G_MIRROR_DEBUG(0, "Device %s: rebuilding provider %s stopped.",
1690 disk->d_softc->sc_name, g_mirror_get_diskname(disk));
1691 }
1692 cp = disk->d_sync.ds_consumer;
1693 g_access(cp, -1, 0, 0);
1694 g_mirror_kill_consumer(disk->d_softc, cp);
1695 free(disk->d_sync.ds_data, M_MIRROR);
1696 disk->d_sync.ds_consumer = NULL;
1697 disk->d_softc->sc_sync.ds_ndisks--;
1698 cp = disk->d_consumer;
1699 KASSERT(cp->acr == 0 && cp->acw == 1 && cp->ace == 1,
1700 ("Consumer %s not opened.", cp->provider->name));
1701 g_access(cp, 0, -1, -1);
1702 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d", cp->provider->name, 0, -1,
1703 -1, 0);
1704 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY;
1705}
1706
1707static void
1708g_mirror_launch_provider(struct g_mirror_softc *sc)
1709{
1710 struct g_mirror_disk *disk;
1711 struct g_provider *pp;
1712
1713 g_topology_assert();
1714
1715 pp = g_new_providerf(sc->sc_geom, "mirror/%s", sc->sc_name);
1716 pp->mediasize = sc->sc_mediasize;
1717 pp->sectorsize = sc->sc_sectorsize;
1718 sc->sc_provider = pp;
1719 g_error_provider(pp, 0);
1720 G_MIRROR_DEBUG(0, "Device %s: provider %s launched.", sc->sc_name,
1721 pp->name);
1722 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
1723 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING)
1724 g_mirror_sync_start(disk);
1725 }
1726}
1727
1728static void
1729g_mirror_destroy_provider(struct g_mirror_softc *sc)
1730{
1731 struct g_mirror_disk *disk;
1732 struct bio *bp;
1733
1734 g_topology_assert();
1735 KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).",
1736 sc->sc_name));
1737
1738 g_error_provider(sc->sc_provider, ENXIO);
1739 mtx_lock(&sc->sc_queue_mtx);
1740 while ((bp = bioq_first(&sc->sc_queue)) != NULL) {
1741 bioq_remove(&sc->sc_queue, bp);
1742 g_io_deliver(bp, ENXIO);
1743 }
1744 mtx_unlock(&sc->sc_queue_mtx);
1745 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name,
1746 sc->sc_provider->name);
1747 sc->sc_provider->flags |= G_PF_WITHER;
1748 g_orphan_provider(sc->sc_provider, ENXIO);
1749 sc->sc_provider = NULL;
1750 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
1751 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING)
1752 g_mirror_sync_stop(disk, 1);
1753 }
1754}
1755
1756static void
1757g_mirror_go(void *arg)
1758{
1759 struct g_mirror_softc *sc;
1760
1761 sc = arg;
1762 G_MIRROR_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name);
1763 g_mirror_event_send(sc, 0,
1764 G_MIRROR_EVENT_DONTWAIT | G_MIRROR_EVENT_DEVICE);
1765}
1766
1767static u_int
1768g_mirror_determine_state(struct g_mirror_disk *disk)
1769{
1770 struct g_mirror_softc *sc;
1771 u_int state;
1772
1773 sc = disk->d_softc;
1774 if (sc->sc_syncid == disk->d_sync.ds_syncid) {
1775 if ((disk->d_flags &
1776 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) {
1777 /* Disk does not need synchronization. */
1778 state = G_MIRROR_DISK_STATE_ACTIVE;
1779 } else {
1780 if ((sc->sc_flags &
1781 G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 ||
1782 (disk->d_flags &
1783 G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) {
1784 /*
1785 * We can start synchronization from
1786 * the stored offset.
1787 */
1788 state = G_MIRROR_DISK_STATE_SYNCHRONIZING;
1789 } else {
1790 state = G_MIRROR_DISK_STATE_STALE;
1791 }
1792 }
1793 } else if (disk->d_sync.ds_syncid < sc->sc_syncid) {
1794 /*
1795 * Reset all synchronization data for this disk,
1796 * because if it even was synchronized, it was
1797 * synchronized to disks with different syncid.
1798 */
1799 disk->d_flags |= G_MIRROR_DISK_FLAG_SYNCHRONIZING;
1800 disk->d_sync.ds_offset = 0;
1801 disk->d_sync.ds_offset_done = 0;
1802 disk->d_sync.ds_syncid = sc->sc_syncid;
1803 if ((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) == 0 ||
1804 (disk->d_flags & G_MIRROR_DISK_FLAG_FORCE_SYNC) != 0) {
1805 state = G_MIRROR_DISK_STATE_SYNCHRONIZING;
1806 } else {
1807 state = G_MIRROR_DISK_STATE_STALE;
1808 }
1809 } else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ {
1810 /*
1811 * Not good, NOT GOOD!
1812 * It means that mirror was started on stale disks
1813 * and more fresh disk just arrive.
1814 * If there were writes, mirror is fucked up, sorry.
1815 * I think the best choice here is don't touch
1816 * this disk and inform the user laudly.
1817 */
1818 G_MIRROR_DEBUG(0, "Device %s was started before the freshest "
1819 "disk (%s) arrives!! It will not be connected to the "
1820 "running device.", sc->sc_name,
1821 g_mirror_get_diskname(disk));
1822 g_mirror_destroy_disk(disk);
1823 state = G_MIRROR_DISK_STATE_NONE;
1824 /* Return immediately, because disk was destroyed. */
1825 return (state);
1826 }
1827 G_MIRROR_DEBUG(3, "State for %s disk: %s.",
1828 g_mirror_get_diskname(disk), g_mirror_disk_state2str(state));
1829 return (state);
1830}
1831
1832/*
1833 * Update device state.
1834 */
1835static void
1836g_mirror_update_device(struct g_mirror_softc *sc, boolean_t force)
1837{
1838 struct g_mirror_disk *disk;
1839 u_int state;
1840
1841 g_topology_assert();
1842
1843 switch (sc->sc_state) {
1844 case G_MIRROR_DEVICE_STATE_STARTING:
1845 {
1846 struct g_mirror_disk *pdisk;
1847 u_int dirty, ndisks, syncid;
1848
1849 KASSERT(sc->sc_provider == NULL,
1850 ("Non-NULL provider in STARTING state (%s).", sc->sc_name));
1851 /*
1852 * Are we ready? We are, if all disks are connected or
1853 * if we have any disks and 'force' is true.
1854 */
1855 if ((force && g_mirror_ndisks(sc, -1) > 0) ||
1856 sc->sc_ndisks == g_mirror_ndisks(sc, -1)) {
1857 ;
1858 } else if (g_mirror_ndisks(sc, -1) == 0) {
1859 /*
1860 * Disks went down in starting phase, so destroy
1861 * device.
1862 */
1863 callout_drain(&sc->sc_callout);
1864 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY;
1865 return;
1866 } else {
1867 return;
1868 }
1869
1870 /*
1871 * Activate all disks with the biggest syncid.
1872 */
1873 if (force) {
1874 /*
1875 * If 'force' is true, we have been called due to
1876 * timeout, so don't bother canceling timeout.
1877 */
1878 ndisks = 0;
1879 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
1880 if ((disk->d_flags &
1881 G_MIRROR_DISK_FLAG_SYNCHRONIZING) == 0) {
1882 ndisks++;
1883 }
1884 }
1885 if (ndisks == 0) {
1886 /* No valid disks found, destroy device. */
1887 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY;
1888 return;
1889 }
1890 } else {
1891 /* Cancel timeout. */
1892 callout_drain(&sc->sc_callout);
1893 }
1894
1895 /*
1896 * Find disk with the biggest syncid.
1897 */
1898 syncid = 0;
1899 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
1900 if (disk->d_sync.ds_syncid > syncid)
1901 syncid = disk->d_sync.ds_syncid;
1902 }
1903
1904 /*
1905 * Here we need to look for dirty disks and if all disks
1906 * with the biggest syncid are dirty, we have to choose
1907 * one with the biggest priority and rebuild the rest.
1908 */
1909 /*
1910 * Find the number of dirty disks with the biggest syncid.
1911 * Find the number of disks with the biggest syncid.
1912 * While here, find a disk with the biggest priority.
1913 */
1914 dirty = ndisks = 0;
1915 pdisk = NULL;
1916 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
1917 if (disk->d_sync.ds_syncid != syncid)
1918 continue;
1919 if ((disk->d_flags &
1920 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) {
1921 continue;
1922 }
1923 ndisks++;
1924 if ((disk->d_flags & G_MIRROR_DISK_FLAG_DIRTY) != 0) {
1925 dirty++;
1926 if (pdisk == NULL ||
1927 pdisk->d_priority < disk->d_priority) {
1928 pdisk = disk;
1929 }
1930 }
1931 }
1932 if (dirty == 0) {
1933 /* No dirty disks at all, great. */
1934 } else if (dirty == ndisks) {
1935 /*
1936 * Force synchronization for all dirty disks except one
1937 * with the biggest priority.
1938 */
1939 KASSERT(pdisk != NULL, ("pdisk == NULL"));
1940 G_MIRROR_DEBUG(1, "Using disk %s (device %s) as a "
1941 "master disk for synchronization.",
1942 g_mirror_get_diskname(pdisk), sc->sc_name);
1943 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
1944 if (disk->d_sync.ds_syncid != syncid)
1945 continue;
1946 if ((disk->d_flags &
1947 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) {
1948 continue;
1949 }
1950 KASSERT((disk->d_flags &
1951 G_MIRROR_DISK_FLAG_DIRTY) != 0,
1952 ("Disk %s isn't marked as dirty.",
1953 g_mirror_get_diskname(disk)));
1954 /* Skip the disk with the biggest priority. */
1955 if (disk == pdisk)
1956 continue;
1957 disk->d_sync.ds_syncid = 0;
1958 }
1959 } else if (dirty < ndisks) {
1960 /*
1961 * Force synchronization for all dirty disks.
1962 * We have some non-dirty disks.
1963 */
1964 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
1965 if (disk->d_sync.ds_syncid != syncid)
1966 continue;
1967 if ((disk->d_flags &
1968 G_MIRROR_DISK_FLAG_SYNCHRONIZING) != 0) {
1969 continue;
1970 }
1971 if ((disk->d_flags &
1972 G_MIRROR_DISK_FLAG_DIRTY) == 0) {
1973 continue;
1974 }
1975 disk->d_sync.ds_syncid = 0;
1976 }
1977 }
1978
1979 /* Reset hint. */
1980 sc->sc_hint = NULL;
1981 sc->sc_syncid = syncid;
1982 if (force) {
1983 /* Remember to bump syncid on first write. */
1984 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE;
1985 }
1986 state = G_MIRROR_DEVICE_STATE_RUNNING;
1987 G_MIRROR_DEBUG(1, "Device %s state changed from %s to %s.",
1988 sc->sc_name, g_mirror_device_state2str(sc->sc_state),
1989 g_mirror_device_state2str(state));
1990 sc->sc_state = state;
1991 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
1992 state = g_mirror_determine_state(disk);
1993 g_mirror_event_send(disk, state,
1994 G_MIRROR_EVENT_DONTWAIT);
1995 if (state == G_MIRROR_DISK_STATE_STALE) {
1996 sc->sc_bump_syncid =
1997 G_MIRROR_BUMP_ON_FIRST_WRITE;
1998 }
1999 }
2000 wakeup(&g_mirror_class);
2001 break;
2002 }
2003 case G_MIRROR_DEVICE_STATE_RUNNING:
2004 if (g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE) == 0 &&
2005 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) {
2006 /*
2007 * No active disks or no disks at all,
2008 * so destroy device.
2009 */
2010 if (sc->sc_provider != NULL)
2011 g_mirror_destroy_provider(sc);
2012 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY;
2013 break;
2014 } else if (g_mirror_ndisks(sc,
2015 G_MIRROR_DISK_STATE_ACTIVE) > 0 &&
2016 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_NEW) == 0) {
2017 /*
2018 * We have active disks, launch provider if it doesn't
2019 * exist.
2020 */
2021 if (sc->sc_provider == NULL)
2022 g_mirror_launch_provider(sc);
2023 }
2024 /*
2025 * Bump syncid here, if we need to do it immediately.
2026 */
2027 if (sc->sc_bump_syncid == G_MIRROR_BUMP_IMMEDIATELY) {
2028 sc->sc_bump_syncid = 0;
2029 g_mirror_bump_syncid(sc);
2030 }
2031 break;
2032 default:
2033 KASSERT(1 == 0, ("Wrong device state (%s, %s).",
2034 sc->sc_name, g_mirror_device_state2str(sc->sc_state)));
2035 break;
2036 }
2037}
2038
2039/*
2040 * Update disk state and device state if needed.
2041 */
2042#define DISK_STATE_CHANGED() G_MIRROR_DEBUG(1, \
2043 "Disk %s state changed from %s to %s (device %s).", \
2044 g_mirror_get_diskname(disk), \
2045 g_mirror_disk_state2str(disk->d_state), \
2046 g_mirror_disk_state2str(state), sc->sc_name)
2047static int
2048g_mirror_update_disk(struct g_mirror_disk *disk, u_int state)
2049{
2050 struct g_mirror_softc *sc;
2051
2052 g_topology_assert();
2053
2054 sc = disk->d_softc;
2055again:
2056 G_MIRROR_DEBUG(3, "Changing disk %s state from %s to %s.",
2057 g_mirror_get_diskname(disk), g_mirror_disk_state2str(disk->d_state),
2058 g_mirror_disk_state2str(state));
2059 switch (state) {
2060 case G_MIRROR_DISK_STATE_NEW:
2061 /*
2062 * Possible scenarios:
2063 * 1. New disk arrive.
2064 */
2065 /* Previous state should be NONE. */
2066 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NONE,
2067 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk),
2068 g_mirror_disk_state2str(disk->d_state)));
2069 DISK_STATE_CHANGED();
2070
2071 disk->d_state = state;
2072 if (LIST_EMPTY(&sc->sc_disks))
2073 LIST_INSERT_HEAD(&sc->sc_disks, disk, d_next);
2074 else {
2075 struct g_mirror_disk *dp;
2076
2077 LIST_FOREACH(dp, &sc->sc_disks, d_next) {
2078 if (disk->d_priority >= dp->d_priority) {
2079 LIST_INSERT_BEFORE(dp, disk, d_next);
2080 dp = NULL;
2081 break;
2082 }
2083 if (LIST_NEXT(dp, d_next) == NULL)
2084 break;
2085 }
2086 if (dp != NULL)
2087 LIST_INSERT_AFTER(dp, disk, d_next);
2088 }
2089 G_MIRROR_DEBUG(0, "Device %s: provider %s detected.",
2090 sc->sc_name, g_mirror_get_diskname(disk));
2091 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING)
2092 break;
2093 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING,
2094 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2095 g_mirror_device_state2str(sc->sc_state),
2096 g_mirror_get_diskname(disk),
2097 g_mirror_disk_state2str(disk->d_state)));
2098 state = g_mirror_determine_state(disk);
2099 if (state != G_MIRROR_DISK_STATE_NONE)
2100 goto again;
2101 break;
2102 case G_MIRROR_DISK_STATE_ACTIVE:
2103 /*
2104 * Possible scenarios:
2105 * 1. New disk does not need synchronization.
2106 * 2. Synchronization process finished successfully.
2107 */
2108 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING,
2109 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2110 g_mirror_device_state2str(sc->sc_state),
2111 g_mirror_get_diskname(disk),
2112 g_mirror_disk_state2str(disk->d_state)));
2113 /* Previous state should be NEW or SYNCHRONIZING. */
2114 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW ||
2115 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING,
2116 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk),
2117 g_mirror_disk_state2str(disk->d_state)));
2118 DISK_STATE_CHANGED();
2119
2120 if (disk->d_state == G_MIRROR_DISK_STATE_NEW)
2121 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY;
2122 else if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) {
2123 disk->d_flags &= ~G_MIRROR_DISK_FLAG_SYNCHRONIZING;
2124 disk->d_flags &= ~G_MIRROR_DISK_FLAG_FORCE_SYNC;
2125 g_mirror_sync_stop(disk, 0);
2126 }
2127 disk->d_state = state;
2128 disk->d_sync.ds_offset = 0;
2129 disk->d_sync.ds_offset_done = 0;
2130 g_mirror_update_access(disk);
2131 g_mirror_update_metadata(disk);
2132 G_MIRROR_DEBUG(0, "Device %s: provider %s activated.",
2133 sc->sc_name, g_mirror_get_diskname(disk));
2134 break;
2135 case G_MIRROR_DISK_STATE_STALE:
2136 /*
2137 * Possible scenarios:
2138 * 1. Stale disk was connected.
2139 */
2140 /* Previous state should be NEW. */
2141 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW,
2142 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk),
2143 g_mirror_disk_state2str(disk->d_state)));
2144 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING,
2145 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2146 g_mirror_device_state2str(sc->sc_state),
2147 g_mirror_get_diskname(disk),
2148 g_mirror_disk_state2str(disk->d_state)));
2149 /*
2150 * STALE state is only possible if device is marked
2151 * NOAUTOSYNC.
2152 */
2153 KASSERT((sc->sc_flags & G_MIRROR_DEVICE_FLAG_NOAUTOSYNC) != 0,
2154 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2155 g_mirror_device_state2str(sc->sc_state),
2156 g_mirror_get_diskname(disk),
2157 g_mirror_disk_state2str(disk->d_state)));
2158 DISK_STATE_CHANGED();
2159
2160 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY;
2161 disk->d_state = state;
2162 g_mirror_update_metadata(disk);
2163 G_MIRROR_DEBUG(0, "Device %s: provider %s is stale.",
2164 sc->sc_name, g_mirror_get_diskname(disk));
2165 break;
2166 case G_MIRROR_DISK_STATE_SYNCHRONIZING:
2167 /*
2168 * Possible scenarios:
2169 * 1. Disk which needs synchronization was connected.
2170 */
2171 /* Previous state should be NEW. */
2172 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW,
2173 ("Wrong disk state (%s, %s).", g_mirror_get_diskname(disk),
2174 g_mirror_disk_state2str(disk->d_state)));
2175 KASSERT(sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING,
2176 ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2177 g_mirror_device_state2str(sc->sc_state),
2178 g_mirror_get_diskname(disk),
2179 g_mirror_disk_state2str(disk->d_state)));
2180 DISK_STATE_CHANGED();
2181
2182 if (disk->d_state == G_MIRROR_DISK_STATE_NEW)
2183 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY;
2184 disk->d_state = state;
2185 if (sc->sc_provider != NULL) {
2186 g_mirror_sync_start(disk);
2187 g_mirror_update_metadata(disk);
2188 }
2189 break;
2190 case G_MIRROR_DISK_STATE_DISCONNECTED:
2191 /*
2192 * Possible scenarios:
2193 * 1. Device wasn't running yet, but disk disappear.
2194 * 2. Disk was active and disapppear.
2195 * 3. Disk disappear during synchronization process.
2196 */
2197 if (sc->sc_state == G_MIRROR_DEVICE_STATE_RUNNING) {
2198 /*
2199 * Previous state should be ACTIVE, STALE or
2200 * SYNCHRONIZING.
2201 */
2202 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_ACTIVE ||
2203 disk->d_state == G_MIRROR_DISK_STATE_STALE ||
2204 disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING,
2205 ("Wrong disk state (%s, %s).",
2206 g_mirror_get_diskname(disk),
2207 g_mirror_disk_state2str(disk->d_state)));
2208 } else if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING) {
2209 /* Previous state should be NEW. */
2210 KASSERT(disk->d_state == G_MIRROR_DISK_STATE_NEW,
2211 ("Wrong disk state (%s, %s).",
2212 g_mirror_get_diskname(disk),
2213 g_mirror_disk_state2str(disk->d_state)));
2214 /*
2215 * Reset bumping syncid if disk disappeared in STARTING
2216 * state.
2217 */
2218 if (sc->sc_bump_syncid == G_MIRROR_BUMP_ON_FIRST_WRITE)
2219 sc->sc_bump_syncid = 0;
2220#ifdef INVARIANTS
2221 } else {
2222 KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).",
2223 sc->sc_name,
2224 g_mirror_device_state2str(sc->sc_state),
2225 g_mirror_get_diskname(disk),
2226 g_mirror_disk_state2str(disk->d_state)));
2227#endif
2228 }
2229 DISK_STATE_CHANGED();
2230 G_MIRROR_DEBUG(0, "Device %s: provider %s disconnected.",
2231 sc->sc_name, g_mirror_get_diskname(disk));
2232
2233 g_mirror_destroy_disk(disk);
2234 break;
2235 case G_MIRROR_DISK_STATE_DESTROY:
2236 {
2237 int error;
2238
2239 error = g_mirror_clear_metadata(disk);
2240 if (error != 0)
2241 return (error);
2242 DISK_STATE_CHANGED();
2243 G_MIRROR_DEBUG(0, "Device %s: provider %s destroyed.",
2244 sc->sc_name, g_mirror_get_diskname(disk));
2245
2246 g_mirror_destroy_disk(disk);
2247 sc->sc_ndisks--;
2248 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
2249 g_mirror_update_metadata(disk);
2250 }
2251 break;
2252 }
2253 default:
2254 KASSERT(1 == 0, ("Unknown state (%u).", state));
2255 break;
2256 }
2257 return (0);
2258}
2259#undef DISK_STATE_CHANGED
2260
2261static int
2262g_mirror_read_metadata(struct g_consumer *cp, struct g_mirror_metadata *md)
2263{
2264 struct g_provider *pp;
2265 u_char *buf;
2266 int error;
2267
2268 g_topology_assert();
2269
2270 error = g_access(cp, 1, 0, 0);
2271 if (error != 0)
2272 return (error);
2273 pp = cp->provider;
2274 g_topology_unlock();
2275 /* Metadata are stored on last sector. */
2276 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize,
2277 &error);
2278 g_topology_lock();
2279 if (buf == NULL) {
2280 g_access(cp, -1, 0, 0);
2281 return (error);
2282 }
2283 if (error != 0) {
2284 g_access(cp, -1, 0, 0);
2285 g_free(buf);
2286 return (error);
2287 }
2288 error = g_access(cp, -1, 0, 0);
2289 KASSERT(error == 0, ("Cannot decrease access count for %s.", pp->name));
2290
2291 /* Decode metadata. */
2292 error = mirror_metadata_decode(buf, md);
2293 g_free(buf);
2294 if (strcmp(md->md_magic, G_MIRROR_MAGIC) != 0)
2295 return (EINVAL);
2296 if (error != 0) {
2297 G_MIRROR_DEBUG(1, "MD5 metadata hash mismatch for provider %s.",
2298 cp->provider->name);
2299 return (error);
2300 }
2301
2302 return (0);
2303}
2304
2305static int
2306g_mirror_check_metadata(struct g_mirror_softc *sc, struct g_provider *pp,
2307 struct g_mirror_metadata *md)
2308{
2309
2310 if (g_mirror_id2disk(sc, md->md_did) != NULL) {
2311 G_MIRROR_DEBUG(1, "Disk %s (id=%u) already exists, skipping.",
2312 pp->name, md->md_did);
2313 return (EEXIST);
2314 }
2315 if (md->md_all != sc->sc_ndisks) {
2316 G_MIRROR_DEBUG(1,
2317 "Invalid '%s' field on disk %s (device %s), skipping.",
2318 "md_all", pp->name, sc->sc_name);
2319 return (EINVAL);
2320 }
2321 if (md->md_slice != sc->sc_slice) {
2322 G_MIRROR_DEBUG(1,
2323 "Invalid '%s' field on disk %s (device %s), skipping.",
2324 "md_slice", pp->name, sc->sc_name);
2325 return (EINVAL);
2326 }
2327 if (md->md_balance != sc->sc_balance) {
2328 G_MIRROR_DEBUG(1,
2329 "Invalid '%s' field on disk %s (device %s), skipping.",
2330 "md_balance", pp->name, sc->sc_name);
2331 return (EINVAL);
2332 }
2333 if (md->md_mediasize != sc->sc_mediasize) {
2334 G_MIRROR_DEBUG(1,
2335 "Invalid '%s' field on disk %s (device %s), skipping.",
2336 "md_mediasize", pp->name, sc->sc_name);
2337 return (EINVAL);
2338 }
2339 if (sc->sc_mediasize > pp->mediasize) {
2340 G_MIRROR_DEBUG(1,
2341 "Invalid size of disk %s (device %s), skipping.", pp->name,
2342 sc->sc_name);
2343 return (EINVAL);
2344 }
2345 if (md->md_sectorsize != sc->sc_sectorsize) {
2346 G_MIRROR_DEBUG(1,
2347 "Invalid '%s' field on disk %s (device %s), skipping.",
2348 "md_sectorsize", pp->name, sc->sc_name);
2349 return (EINVAL);
2350 }
2351 if ((sc->sc_sectorsize % pp->sectorsize) != 0) {
2352 G_MIRROR_DEBUG(1,
2353 "Invalid sector size of disk %s (device %s), skipping.",
2354 pp->name, sc->sc_name);
2355 return (EINVAL);
2356 }
2357 if ((md->md_mflags & ~G_MIRROR_DEVICE_FLAG_MASK) != 0) {
2358 G_MIRROR_DEBUG(1,
2359 "Invalid device flags on disk %s (device %s), skipping.",
2360 pp->name, sc->sc_name);
2361 return (EINVAL);
2362 }
2363 if ((md->md_dflags & ~G_MIRROR_DISK_FLAG_MASK) != 0) {
2364 G_MIRROR_DEBUG(1,
2365 "Invalid disk flags on disk %s (device %s), skipping.",
2366 pp->name, sc->sc_name);
2367 return (EINVAL);
2368 }
2369 return (0);
2370}
2371
2372static int
2373g_mirror_add_disk(struct g_mirror_softc *sc, struct g_provider *pp,
2374 struct g_mirror_metadata *md)
2375{
2376 struct g_mirror_disk *disk;
2377 int error;
2378
2379 g_topology_assert();
2380 G_MIRROR_DEBUG(2, "Adding disk %s.", pp->name);
2381
2382 error = g_mirror_check_metadata(sc, pp, md);
2383 if (error != 0)
2384 return (error);
2385 disk = g_mirror_init_disk(sc, pp, md, &error);
2386 if (disk == NULL)
2387 return (error);
2388 error = g_mirror_event_send(disk, G_MIRROR_DISK_STATE_NEW,
2389 G_MIRROR_EVENT_WAIT);
2390 return (error);
2391}
2392
2393static int
2394g_mirror_access(struct g_provider *pp, int acr, int acw, int ace)
2395{
2396 struct g_mirror_softc *sc;
2397 struct g_mirror_disk *disk;
2398 int dcr, dcw, dce, err, error;
2399
2400 g_topology_assert();
2401 G_MIRROR_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr,
2402 acw, ace);
2403
2404 dcr = pp->acr + acr;
2405 dcw = pp->acw + acw;
2406 dce = pp->ace + ace;
2407
2408 /* On first open, grab an extra "exclusive" bit */
2409 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0)
2410 ace++;
2411 /* ... and let go of it on last close */
2412 if (dcr == 0 && dcw == 0 && dce == 0)
2413 ace--;
2414
2415 sc = pp->geom->softc;
2416 if (sc == NULL || LIST_EMPTY(&sc->sc_disks)) {
2417 if (acr <= 0 && acw <= 0 && ace <= 0)
2418 return (0);
2419 else
2420 return (ENXIO);
2421 }
2422 error = ENXIO;
2423 LIST_FOREACH(disk, &sc->sc_disks, d_next) {
2424 if (disk->d_state != G_MIRROR_DISK_STATE_ACTIVE)
2425 continue;
2426 err = g_access(disk->d_consumer, acr, acw, ace);
2427 G_MIRROR_DEBUG(2, "Access %s r%dw%de%d = %d",
2428 g_mirror_get_diskname(disk), acr, acw, ace, err);
2429 if (err == 0) {
2430 /*
2431 * Mark disk as dirty on open and unmark on close.
2432 */
2433 if (pp->acw == 0 && dcw > 0) {
2434 G_MIRROR_DEBUG(1,
2435 "Disk %s (device %s) marked as dirty.",
2436 g_mirror_get_diskname(disk), sc->sc_name);
2437 disk->d_flags |= G_MIRROR_DISK_FLAG_DIRTY;
2438 g_mirror_update_metadata(disk);
2439 } else if (pp->acw > 0 && dcw == 0) {
2440 G_MIRROR_DEBUG(1,
2441 "Disk %s (device %s) marked as clean.",
2442 g_mirror_get_diskname(disk), sc->sc_name);
2443 disk->d_flags &= ~G_MIRROR_DISK_FLAG_DIRTY;
2444 g_mirror_update_metadata(disk);
2445 }
2446 error = 0;
2447 } else {
2448 sc->sc_bump_syncid = G_MIRROR_BUMP_ON_FIRST_WRITE;
2449 g_mirror_event_send(disk,
2450 G_MIRROR_DISK_STATE_DISCONNECTED,
2451 G_MIRROR_EVENT_DONTWAIT);
2452 }
2453 }
2454 /*
2455 * Be sure to return 0 for negativate access requests.
2456 * In case of some HW problems, it is possible that we don't have
2457 * any active disk here, so loop above will be no-op and error will
2458 * be ENXIO.
2459 */
2460 if (error != 0 && acr <= 0 && acw <= 0 && ace <= 0)
2461 error = 0;
2462 return (error);
2463}
2464
2465static struct g_geom *
2466g_mirror_create(struct g_class *mp, const struct g_mirror_metadata *md)
2467{
2468 struct g_mirror_softc *sc;
2469 struct g_geom *gp;
2470 int error, timeout;
2471
2472 g_topology_assert();
2473 G_MIRROR_DEBUG(1, "Creating device %s (id=%u).", md->md_name,
2474 md->md_mid);
2475
2476 /* One disk is minimum. */
2477 if (md->md_all < 1)
2478 return (NULL);
2479 /*
2480 * Action geom.
2481 */
2482 gp = g_new_geomf(mp, "%s", md->md_name);
2483 sc = malloc(sizeof(*sc), M_MIRROR, M_WAITOK | M_ZERO);
2484 gp->start = g_mirror_start;
2485 gp->spoiled = g_mirror_spoiled;
2486 gp->orphan = g_mirror_orphan;
2487 gp->access = g_mirror_access;
2488 gp->dumpconf = g_mirror_dumpconf;
2489
2490 sc->sc_id = md->md_mid;
2491 sc->sc_slice = md->md_slice;
2492 sc->sc_balance = md->md_balance;
2493 sc->sc_mediasize = md->md_mediasize;
2494 sc->sc_sectorsize = md->md_sectorsize;
2495 sc->sc_ndisks = md->md_all;
2496 sc->sc_flags = md->md_mflags;
2497 sc->sc_bump_syncid = 0;
2498 sc->sc_idle = 0;
2499 bioq_init(&sc->sc_queue);
2500 mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF);
2501 LIST_INIT(&sc->sc_disks);
2502 TAILQ_INIT(&sc->sc_events);
2503 mtx_init(&sc->sc_events_mtx, "gmirror:events", NULL, MTX_DEF);
2504 callout_init(&sc->sc_callout, CALLOUT_MPSAFE);
2505 sc->sc_state = G_MIRROR_DEVICE_STATE_STARTING;
2506 gp->softc = sc;
2507 sc->sc_geom = gp;
2508 sc->sc_provider = NULL;
2509 /*
2510 * Synchronization geom.
2511 */
2512 gp = g_new_geomf(mp, "%s.sync", md->md_name);
2513 gp->softc = sc;
2514 gp->orphan = g_mirror_orphan;
2515 sc->sc_sync.ds_geom = gp;
2516 sc->sc_sync.ds_ndisks = 0;
2517 error = kthread_create(g_mirror_worker, sc, &sc->sc_worker, 0, 0,
2518 "g_mirror %s", md->md_name);
2519 if (error != 0) {
2520 G_MIRROR_DEBUG(1, "Cannot create kernel thread for %s.",
2521 sc->sc_name);
2522 g_destroy_geom(sc->sc_sync.ds_geom);
2523 mtx_destroy(&sc->sc_events_mtx);
2524 mtx_destroy(&sc->sc_queue_mtx);
2525 g_destroy_geom(sc->sc_geom);
2526 free(sc, M_MIRROR);
2527 return (NULL);
2528 }
2529
2530 G_MIRROR_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id);
2531
2532 /*
2533 * Run timeout.
2534 */
2535 timeout = g_mirror_timeout * hz;
2536 callout_reset(&sc->sc_callout, timeout, g_mirror_go, sc);
2537 return (sc->sc_geom);
2538}
2539
2540int
2541g_mirror_destroy(struct g_mirror_softc *sc, boolean_t force)
2542{
2543 struct g_provider *pp;
2544
2545 g_topology_assert();
2546
2547 if (sc == NULL)
2548 return (ENXIO);
2549 pp = sc->sc_provider;
2550 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
2551 if (force) {
2552 G_MIRROR_DEBUG(0, "Device %s is still open, so it "
2553 "can't be definitely removed.", pp->name);
2554 } else {
2555 G_MIRROR_DEBUG(1,
2556 "Device %s is still open (r%dw%de%d).", pp->name,
2557 pp->acr, pp->acw, pp->ace);
2558 return (EBUSY);
2559 }
2560 }
2561
2562 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_DESTROY;
2563 sc->sc_flags |= G_MIRROR_DEVICE_FLAG_WAIT;
2564 g_topology_unlock();
2565 G_MIRROR_DEBUG(4, "%s: Waking up %p.", __func__, sc);
2566 mtx_lock(&sc->sc_queue_mtx);
2567 wakeup(sc);
2568 mtx_unlock(&sc->sc_queue_mtx);
2569 G_MIRROR_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker);
2570 while (sc->sc_worker != NULL)
2571 tsleep(&sc->sc_worker, PRIBIO, "m:destroy", hz / 5);
2572 G_MIRROR_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker);
2573 g_topology_lock();
2574 g_mirror_destroy_device(sc);
2575 free(sc, M_MIRROR);
2576 return (0);
2577}
2578
2579static void
2580g_mirror_taste_orphan(struct g_consumer *cp)
2581{
2582
2583 KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
2584 cp->provider->name));
2585}
2586
2587static struct g_geom *
2588g_mirror_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
2589{
2590 struct g_mirror_metadata md;
2591 struct g_mirror_softc *sc;
2592 struct g_consumer *cp;
2593 struct g_geom *gp;
2594 int error;
2595
2596 g_topology_assert();
2597 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
2598 G_MIRROR_DEBUG(2, "Tasting %s.", pp->name);
2599
2600 gp = g_new_geomf(mp, "mirror:taste");
2601 /*
2602 * This orphan function should be never called.
2603 */
2604 gp->orphan = g_mirror_taste_orphan;
2605 cp = g_new_consumer(gp);
2606 g_attach(cp, pp);
2607 error = g_mirror_read_metadata(cp, &md);
2608 g_detach(cp);
2609 g_destroy_consumer(cp);
2610 g_destroy_geom(gp);
2611 if (error != 0)
2612 return (NULL);
2613 gp = NULL;
2614
2615 if (md.md_version > G_MIRROR_VERSION) {
2616 printf("geom_mirror.ko module is too old to handle %s.\n",
2617 pp->name);
2618 return (NULL);
2619 }
2620 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0)
2621 return (NULL);
2622 if ((md.md_dflags & G_MIRROR_DISK_FLAG_INACTIVE) != 0) {
2623 G_MIRROR_DEBUG(0,
2624 "Device %s: provider %s marked as inactive, skipping.",
2625 md.md_name, pp->name);
2626 return (NULL);
2627 }
2628 if (g_mirror_debug >= 2)
2629 mirror_metadata_dump(&md);
2630
2631 /*
2632 * Let's check if device already exists.
2633 */
2634 sc = NULL;
2635 LIST_FOREACH(gp, &mp->geom, geom) {
2636 sc = gp->softc;
2637 if (sc == NULL)
2638 continue;
2639 if (sc->sc_sync.ds_geom == gp)
2640 continue;
2641 if (strcmp(md.md_name, sc->sc_name) != 0)
2642 continue;
2643 if (md.md_mid != sc->sc_id) {
2644 G_MIRROR_DEBUG(0, "Device %s already configured.",
2645 sc->sc_name);
2646 return (NULL);
2647 }
2648 break;
2649 }
2650 if (gp == NULL) {
2651 gp = g_mirror_create(mp, &md);
2652 if (gp == NULL) {
2653 G_MIRROR_DEBUG(0, "Cannot create device %s.",
2654 md.md_name);
2655 return (NULL);
2656 }
2657 sc = gp->softc;
2658 }
2659 G_MIRROR_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name);
2660 error = g_mirror_add_disk(sc, pp, &md);
2661 if (error != 0) {
2662 G_MIRROR_DEBUG(0, "Cannot add disk %s to %s (error=%d).",
2663 pp->name, gp->name, error);
2664 if (LIST_EMPTY(&sc->sc_disks))
2665 g_mirror_destroy(sc, 1);
2666 return (NULL);
2667 }
2668 return (gp);
2669}
2670
2671static int
2672g_mirror_destroy_geom(struct gctl_req *req __unused,
2673 struct g_class *mp __unused, struct g_geom *gp)
2674{
2675
2676 return (g_mirror_destroy(gp->softc, 0));
2677}
2678
2679static void
2680g_mirror_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2681 struct g_consumer *cp, struct g_provider *pp)
2682{
2683 struct g_mirror_softc *sc;
2684
2685 g_topology_assert();
2686
2687 sc = gp->softc;
2688 if (sc == NULL)
2689 return;
2690 /* Skip synchronization geom. */
2691 if (gp == sc->sc_sync.ds_geom)
2692 return;
2693 if (pp != NULL) {
2694 /* Nothing here. */
2695 } else if (cp != NULL) {
2696 struct g_mirror_disk *disk;
2697
2698 disk = cp->private;
2699 if (disk == NULL)
2700 return;
2701 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)disk->d_id);
2702 if (disk->d_state == G_MIRROR_DISK_STATE_SYNCHRONIZING) {
2703 sbuf_printf(sb, "%s<Synchronized>", indent);
2704 if (disk->d_sync.ds_offset_done == 0)
2705 sbuf_printf(sb, "0%%");
2706 else {
2707 sbuf_printf(sb, "%u%%",
2708 (u_int)((disk->d_sync.ds_offset_done * 100) /
2709 sc->sc_provider->mediasize));
2710 }
2711 sbuf_printf(sb, "</Synchronized>\n");
2712 }
2713 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent,
2714 disk->d_sync.ds_syncid);
2715 sbuf_printf(sb, "%s<Flags>", indent);
2716 if (disk->d_flags == 0)
2717 sbuf_printf(sb, "NONE");
2718 else {
2719 int first = 1;
2720
2721#define ADD_FLAG(flag, name) do { \
2722 if ((disk->d_flags & (flag)) != 0) { \
2723 if (!first) \
2724 sbuf_printf(sb, ", "); \
2725 else \
2726 first = 0; \
2727 sbuf_printf(sb, name); \
2728 } \
2729} while (0)
2730 ADD_FLAG(G_MIRROR_DISK_FLAG_DIRTY, "DIRTY");
2731 ADD_FLAG(G_MIRROR_DISK_FLAG_HARDCODED, "HARDCODED");
2732 ADD_FLAG(G_MIRROR_DISK_FLAG_INACTIVE, "INACTIVE");
2733 ADD_FLAG(G_MIRROR_DISK_FLAG_SYNCHRONIZING,
2734 "SYNCHRONIZING");
2735 ADD_FLAG(G_MIRROR_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC");
2736#undef ADD_FLAG
2737 }
2738 sbuf_printf(sb, "</Flags>\n");
2739 sbuf_printf(sb, "%s<Priority>%u</Priority>\n", indent,
2740 disk->d_priority);
2741 sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2742 g_mirror_disk_state2str(disk->d_state));
2743 } else {
2744 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id);
2745 sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid);
2746 sbuf_printf(sb, "%s<Flags>", indent);
2747 if (sc->sc_flags == 0)
2748 sbuf_printf(sb, "NONE");
2749 else {
2750 int first = 1;
2751
2752#define ADD_FLAG(flag, name) do { \
2753 if ((sc->sc_flags & (flag)) != 0) { \
2754 if (!first) \
2755 sbuf_printf(sb, ", "); \
2756 else \
2757 first = 0; \
2758 sbuf_printf(sb, name); \
2759 } \
2760} while (0)
2761 ADD_FLAG(G_MIRROR_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC");
2762#undef ADD_FLAG
2763 }
2764 sbuf_printf(sb, "</Flags>\n");
2765 sbuf_printf(sb, "%s<Slice>%u</Slice>\n", indent,
2766 (u_int)sc->sc_slice);
2767 sbuf_printf(sb, "%s<Balance>%s</Balance>\n", indent,
2768 balance_name(sc->sc_balance));
2769 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent,
2770 sc->sc_ndisks);
2771 sbuf_printf(sb, "%s<State>", indent);
2772 if (sc->sc_state == G_MIRROR_DEVICE_STATE_STARTING)
2773 sbuf_printf(sb, "%s", "STARTING");
2774 else if (sc->sc_ndisks ==
2775 g_mirror_ndisks(sc, G_MIRROR_DISK_STATE_ACTIVE))
2776 sbuf_printf(sb, "%s", "COMPLETE");
2777 else
2778 sbuf_printf(sb, "%s", "DEGRADED");
2779 sbuf_printf(sb, "</State>\n");
2780 }
2781}
2782
2783static void
2784g_mirror_shutdown(void *arg, int howto)
2785{
2786 struct g_class *mp;
2787 struct g_geom *gp, *gp2;
2788
2789 mp = arg;
2790 g_topology_lock();
2791 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
2792 if (gp->softc == NULL)
2793 continue;
2794 g_mirror_destroy(gp->softc, 1);
2795 }
2796 g_topology_unlock();
2797#if 0
2798 tsleep(&gp, PRIBIO, "m:shutdown", hz * 20);
2799#endif
2800}
2801
2802static void
2803g_mirror_init(struct g_class *mp)
2804{
2805
2806 g_mirror_ehtag = EVENTHANDLER_REGISTER(shutdown_post_sync,
2807 g_mirror_shutdown, mp, SHUTDOWN_PRI_FIRST);
2808 if (g_mirror_ehtag == NULL)
2809 G_MIRROR_DEBUG(0, "Warning! Cannot register shutdown event.");
2810}
2811
2812static void
2813g_mirror_fini(struct g_class *mp)
2814{
2815
2816 if (g_mirror_ehtag == NULL)
2817 return;
2818 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_mirror_ehtag);
2819}
2820
2821static int
2822g_mirror_can_go(void)
2823{
2824 struct g_mirror_softc *sc;
2825 struct g_geom *gp;
2826 struct g_provider *pp;
2827 int can_go;
2828
2829 DROP_GIANT();
2830 can_go = 1;
2831 g_topology_lock();
2832 LIST_FOREACH(gp, &g_mirror_class.geom, geom) {
2833 sc = gp->softc;
2834 if (sc == NULL) {
2835 can_go = 0;
2836 break;
2837 }
2838 pp = sc->sc_provider;
2839 if (pp == NULL || pp->error != 0) {
2840 can_go = 0;
2841 break;
2842 }
2843 }
2844 g_topology_unlock();
2845 PICKUP_GIANT();
2846 return (can_go);
2847}
2848
2849static void
2850g_mirror_rootwait(void)
2851{
2852
2853 /*
2854 * HACK: Wait for GEOM, because g_mirror_rootwait() can be called,
2855 * HACK: before we get providers for tasting.
2856 */
2857 tsleep(&g_mirror_class, PRIBIO, "mroot", hz * 3);
2858 /*
2859 * Wait for mirrors in degraded state.
2860 */
2861 for (;;) {
2862 if (g_mirror_can_go())
2863 break;
2864 tsleep(&g_mirror_class, PRIBIO, "mroot", hz);
2865 }
2866}
2867
2868SYSINIT(g_mirror_root, SI_SUB_RAID, SI_ORDER_FIRST, g_mirror_rootwait, NULL)
2869
2870DECLARE_GEOM_CLASS(g_mirror_class, g_mirror);