xref: /freebsd/sys/geom/vinum/geom_vinum.c (revision f4b37ed0f8b307b1f3f0f630ca725d68f1dff30d)
1 /*-
2  *  Copyright (c) 2004, 2007 Lukas Ertl
3  *  Copyright (c) 2007, 2009 Ulf Lilleengen
4  *  All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  *
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/bio.h>
34 #include <sys/kernel.h>
35 #include <sys/kthread.h>
36 #include <sys/lock.h>
37 #include <sys/malloc.h>
38 #include <sys/module.h>
39 #include <sys/mutex.h>
40 #include <sys/sbuf.h>
41 #include <sys/sysctl.h>
42 #include <sys/systm.h>
43 
44 #include <geom/geom.h>
45 #include <geom/vinum/geom_vinum_var.h>
46 #include <geom/vinum/geom_vinum.h>
47 #include <geom/vinum/geom_vinum_raid5.h>
48 
49 SYSCTL_DECL(_kern_geom);
50 static SYSCTL_NODE(_kern_geom, OID_AUTO, vinum, CTLFLAG_RW, 0,
51     "GEOM_VINUM stuff");
52 u_int g_vinum_debug = 0;
53 SYSCTL_UINT(_kern_geom_vinum, OID_AUTO, debug, CTLFLAG_RWTUN, &g_vinum_debug, 0,
54     "Debug level");
55 
56 static int	gv_create(struct g_geom *, struct gctl_req *);
57 static void	gv_attach(struct gv_softc *, struct gctl_req *);
58 static void	gv_detach(struct gv_softc *, struct gctl_req *);
59 static void	gv_parityop(struct gv_softc *, struct gctl_req *);
60 
61 
62 static void
63 gv_orphan(struct g_consumer *cp)
64 {
65 	struct g_geom *gp;
66 	struct gv_softc *sc;
67 	struct gv_drive *d;
68 
69 	g_topology_assert();
70 
71 	KASSERT(cp != NULL, ("gv_orphan: null cp"));
72 	gp = cp->geom;
73 	KASSERT(gp != NULL, ("gv_orphan: null gp"));
74 	sc = gp->softc;
75 	KASSERT(sc != NULL, ("gv_orphan: null sc"));
76 	d = cp->private;
77 	KASSERT(d != NULL, ("gv_orphan: null d"));
78 
79 	g_trace(G_T_TOPOLOGY, "gv_orphan(%s)", gp->name);
80 
81 	gv_post_event(sc, GV_EVENT_DRIVE_LOST, d, NULL, 0, 0);
82 }
83 
84 void
85 gv_start(struct bio *bp)
86 {
87 	struct g_geom *gp;
88 	struct gv_softc *sc;
89 
90 	gp = bp->bio_to->geom;
91 	sc = gp->softc;
92 
93 	switch (bp->bio_cmd) {
94 	case BIO_READ:
95 	case BIO_WRITE:
96 	case BIO_DELETE:
97 		break;
98 	case BIO_GETATTR:
99 	default:
100 		g_io_deliver(bp, EOPNOTSUPP);
101 		return;
102 	}
103 	mtx_lock(&sc->bqueue_mtx);
104 	bioq_disksort(sc->bqueue_down, bp);
105 	wakeup(sc);
106 	mtx_unlock(&sc->bqueue_mtx);
107 }
108 
109 void
110 gv_done(struct bio *bp)
111 {
112 	struct g_geom *gp;
113 	struct gv_softc *sc;
114 
115 	KASSERT(bp != NULL, ("NULL bp"));
116 
117 	gp = bp->bio_from->geom;
118 	sc = gp->softc;
119 
120 	mtx_lock(&sc->bqueue_mtx);
121 	bioq_disksort(sc->bqueue_up, bp);
122 	wakeup(sc);
123 	mtx_unlock(&sc->bqueue_mtx);
124 }
125 
126 int
127 gv_access(struct g_provider *pp, int dr, int dw, int de)
128 {
129 	struct g_geom *gp;
130 	struct gv_softc *sc;
131 	struct gv_drive *d, *d2;
132 	int error;
133 
134 	gp = pp->geom;
135 	sc = gp->softc;
136 	/*
137 	 * We want to modify the read count with the write count in case we have
138 	 * plexes in a RAID-5 organization.
139 	 */
140 	dr += dw;
141 
142 	LIST_FOREACH(d, &sc->drives, drive) {
143 		if (d->consumer == NULL)
144 			continue;
145 		error = g_access(d->consumer, dr, dw, de);
146 		if (error) {
147 			LIST_FOREACH(d2, &sc->drives, drive) {
148 				if (d == d2)
149 					break;
150 				g_access(d2->consumer, -dr, -dw, -de);
151 			}
152 			G_VINUM_DEBUG(0, "g_access '%s' failed: %d", d->name,
153 			    error);
154 			return (error);
155 		}
156 	}
157 	return (0);
158 }
159 
160 static void
161 gv_init(struct g_class *mp)
162 {
163 	struct g_geom *gp;
164 	struct gv_softc *sc;
165 
166 	g_trace(G_T_TOPOLOGY, "gv_init(%p)", mp);
167 
168 	gp = g_new_geomf(mp, "VINUM");
169 	gp->spoiled = gv_orphan;
170 	gp->orphan = gv_orphan;
171 	gp->access = gv_access;
172 	gp->start = gv_start;
173 	gp->softc = g_malloc(sizeof(struct gv_softc), M_WAITOK | M_ZERO);
174 	sc = gp->softc;
175 	sc->geom = gp;
176 	sc->bqueue_down = g_malloc(sizeof(struct bio_queue_head),
177 	    M_WAITOK | M_ZERO);
178 	sc->bqueue_up = g_malloc(sizeof(struct bio_queue_head),
179 	    M_WAITOK | M_ZERO);
180 	bioq_init(sc->bqueue_down);
181 	bioq_init(sc->bqueue_up);
182 	LIST_INIT(&sc->drives);
183 	LIST_INIT(&sc->subdisks);
184 	LIST_INIT(&sc->plexes);
185 	LIST_INIT(&sc->volumes);
186 	TAILQ_INIT(&sc->equeue);
187 	mtx_init(&sc->config_mtx, "gv_config", NULL, MTX_DEF);
188 	mtx_init(&sc->equeue_mtx, "gv_equeue", NULL, MTX_DEF);
189 	mtx_init(&sc->bqueue_mtx, "gv_bqueue", NULL, MTX_DEF);
190 	kproc_create(gv_worker, sc, &sc->worker, 0, 0, "gv_worker");
191 }
192 
193 static int
194 gv_unload(struct gctl_req *req, struct g_class *mp, struct g_geom *gp)
195 {
196 	struct gv_softc *sc;
197 
198 	g_trace(G_T_TOPOLOGY, "gv_unload(%p)", mp);
199 
200 	g_topology_assert();
201 	sc = gp->softc;
202 
203 	if (sc != NULL) {
204 		gv_worker_exit(sc);
205 		gp->softc = NULL;
206 		g_wither_geom(gp, ENXIO);
207 	}
208 
209 	return (0);
210 }
211 
212 /* Handle userland request of attaching object. */
213 static void
214 gv_attach(struct gv_softc *sc, struct gctl_req *req)
215 {
216 	struct gv_volume *v;
217 	struct gv_plex *p;
218 	struct gv_sd *s;
219 	off_t *offset;
220 	int *rename, type_child, type_parent;
221 	char *child, *parent;
222 
223 	child = gctl_get_param(req, "child", NULL);
224 	if (child == NULL) {
225 		gctl_error(req, "no child given");
226 		return;
227 	}
228 	parent = gctl_get_param(req, "parent", NULL);
229 	if (parent == NULL) {
230 		gctl_error(req, "no parent given");
231 		return;
232 	}
233 	offset = gctl_get_paraml(req, "offset", sizeof(*offset));
234 	if (offset == NULL) {
235 		gctl_error(req, "no offset given");
236 		return;
237 	}
238 	rename = gctl_get_paraml(req, "rename", sizeof(*rename));
239 	if (rename == NULL) {
240 		gctl_error(req, "no rename flag given");
241 		return;
242 	}
243 
244 	type_child = gv_object_type(sc, child);
245 	type_parent = gv_object_type(sc, parent);
246 
247 	switch (type_child) {
248 	case GV_TYPE_PLEX:
249 		if (type_parent != GV_TYPE_VOL) {
250 			gctl_error(req, "no such volume to attach to");
251 			return;
252 		}
253 		v = gv_find_vol(sc, parent);
254 		p = gv_find_plex(sc, child);
255 		gv_post_event(sc, GV_EVENT_ATTACH_PLEX, p, v, *offset, *rename);
256 		break;
257 	case GV_TYPE_SD:
258 		if (type_parent != GV_TYPE_PLEX) {
259 			gctl_error(req, "no such plex to attach to");
260 			return;
261 		}
262 		p = gv_find_plex(sc, parent);
263 		s = gv_find_sd(sc, child);
264 		gv_post_event(sc, GV_EVENT_ATTACH_SD, s, p, *offset, *rename);
265 		break;
266 	default:
267 		gctl_error(req, "invalid child type");
268 		break;
269 	}
270 }
271 
272 /* Handle userland request of detaching object. */
273 static void
274 gv_detach(struct gv_softc *sc, struct gctl_req *req)
275 {
276 	struct gv_plex *p;
277 	struct gv_sd *s;
278 	int *flags, type;
279 	char *object;
280 
281 	object = gctl_get_param(req, "object", NULL);
282 	if (object == NULL) {
283 		gctl_error(req, "no argument given");
284 		return;
285 	}
286 
287 	flags = gctl_get_paraml(req, "flags", sizeof(*flags));
288 	type = gv_object_type(sc, object);
289 	switch (type) {
290 	case GV_TYPE_PLEX:
291 		p = gv_find_plex(sc, object);
292 		gv_post_event(sc, GV_EVENT_DETACH_PLEX, p, NULL, *flags, 0);
293 		break;
294 	case GV_TYPE_SD:
295 		s = gv_find_sd(sc, object);
296 		gv_post_event(sc, GV_EVENT_DETACH_SD, s, NULL, *flags, 0);
297 		break;
298 	default:
299 		gctl_error(req, "invalid object type");
300 		break;
301 	}
302 }
303 
304 /* Handle userland requests for creating new objects. */
305 static int
306 gv_create(struct g_geom *gp, struct gctl_req *req)
307 {
308 	struct gv_softc *sc;
309 	struct gv_drive *d, *d2;
310 	struct gv_plex *p, *p2;
311 	struct gv_sd *s, *s2;
312 	struct gv_volume *v, *v2;
313 	struct g_provider *pp;
314 	int error, i, *drives, *flags, *plexes, *subdisks, *volumes;
315 	char buf[20];
316 
317 	g_topology_assert();
318 
319 	sc = gp->softc;
320 
321 	/* Find out how many of each object have been passed in. */
322 	volumes = gctl_get_paraml(req, "volumes", sizeof(*volumes));
323 	plexes = gctl_get_paraml(req, "plexes", sizeof(*plexes));
324 	subdisks = gctl_get_paraml(req, "subdisks", sizeof(*subdisks));
325 	drives = gctl_get_paraml(req, "drives", sizeof(*drives));
326 	if (volumes == NULL || plexes == NULL || subdisks == NULL ||
327 	    drives == NULL) {
328 		gctl_error(req, "number of objects not given");
329 		return (-1);
330 	}
331 	flags = gctl_get_paraml(req, "flags", sizeof(*flags));
332 	if (flags == NULL) {
333 		gctl_error(req, "flags not given");
334 		return (-1);
335 	}
336 
337 	/* First, handle drive definitions ... */
338 	for (i = 0; i < *drives; i++) {
339 		snprintf(buf, sizeof(buf), "drive%d", i);
340 		d2 = gctl_get_paraml(req, buf, sizeof(*d2));
341 		if (d2 == NULL) {
342 			gctl_error(req, "no drive definition given");
343 			return (-1);
344 		}
345 		/*
346 		 * Make sure that the device specified in the drive config is
347 		 * an active GEOM provider.
348 		 */
349 		pp = g_provider_by_name(d2->device);
350 		if (pp == NULL) {
351 			gctl_error(req, "%s: device not found", d2->device);
352 			goto error;
353 		}
354 		if (gv_find_drive(sc, d2->name) != NULL) {
355 			/* Ignore error. */
356 			if (*flags & GV_FLAG_F)
357 				continue;
358 			gctl_error(req, "drive '%s' already exists", d2->name);
359 			goto error;
360 		}
361 		if (gv_find_drive_device(sc, d2->device) != NULL) {
362 			gctl_error(req, "device '%s' already configured in "
363 			    "gvinum", d2->device);
364 			goto error;
365 		}
366 
367 
368 		d = g_malloc(sizeof(*d), M_WAITOK | M_ZERO);
369 		bcopy(d2, d, sizeof(*d));
370 
371 		gv_post_event(sc, GV_EVENT_CREATE_DRIVE, d, NULL, 0, 0);
372 	}
373 
374 	/* ... then volume definitions ... */
375 	for (i = 0; i < *volumes; i++) {
376 		error = 0;
377 		snprintf(buf, sizeof(buf), "volume%d", i);
378 		v2 = gctl_get_paraml(req, buf, sizeof(*v2));
379 		if (v2 == NULL) {
380 			gctl_error(req, "no volume definition given");
381 			return (-1);
382 		}
383 		if (gv_find_vol(sc, v2->name) != NULL) {
384 			/* Ignore error. */
385 			if (*flags & GV_FLAG_F)
386 				continue;
387 			gctl_error(req, "volume '%s' already exists", v2->name);
388 			goto error;
389 		}
390 
391 		v = g_malloc(sizeof(*v), M_WAITOK | M_ZERO);
392 		bcopy(v2, v, sizeof(*v));
393 
394 		gv_post_event(sc, GV_EVENT_CREATE_VOLUME, v, NULL, 0, 0);
395 	}
396 
397 	/* ... then plex definitions ... */
398 	for (i = 0; i < *plexes; i++) {
399 		error = 0;
400 		snprintf(buf, sizeof(buf), "plex%d", i);
401 		p2 = gctl_get_paraml(req, buf, sizeof(*p2));
402 		if (p2 == NULL) {
403 			gctl_error(req, "no plex definition given");
404 			return (-1);
405 		}
406 		if (gv_find_plex(sc, p2->name) != NULL) {
407 			/* Ignore error. */
408 			if (*flags & GV_FLAG_F)
409 				continue;
410 			gctl_error(req, "plex '%s' already exists", p2->name);
411 			goto error;
412 		}
413 
414 		p = g_malloc(sizeof(*p), M_WAITOK | M_ZERO);
415 		bcopy(p2, p, sizeof(*p));
416 
417 		gv_post_event(sc, GV_EVENT_CREATE_PLEX, p, NULL, 0, 0);
418 	}
419 
420 	/* ... and, finally, subdisk definitions. */
421 	for (i = 0; i < *subdisks; i++) {
422 		error = 0;
423 		snprintf(buf, sizeof(buf), "sd%d", i);
424 		s2 = gctl_get_paraml(req, buf, sizeof(*s2));
425 		if (s2 == NULL) {
426 			gctl_error(req, "no subdisk definition given");
427 			return (-1);
428 		}
429 		if (gv_find_sd(sc, s2->name) != NULL) {
430 			/* Ignore error. */
431 			if (*flags & GV_FLAG_F)
432 				continue;
433 			gctl_error(req, "sd '%s' already exists", s2->name);
434 			goto error;
435 		}
436 
437 		s = g_malloc(sizeof(*s), M_WAITOK | M_ZERO);
438 		bcopy(s2, s, sizeof(*s));
439 
440 		gv_post_event(sc, GV_EVENT_CREATE_SD, s, NULL, 0, 0);
441 	}
442 
443 error:
444 	gv_post_event(sc, GV_EVENT_SETUP_OBJECTS, sc, NULL, 0, 0);
445 	gv_post_event(sc, GV_EVENT_SAVE_CONFIG, sc, NULL, 0, 0);
446 
447 	return (0);
448 }
449 
450 static void
451 gv_config(struct gctl_req *req, struct g_class *mp, char const *verb)
452 {
453 	struct g_geom *gp;
454 	struct gv_softc *sc;
455 	struct sbuf *sb;
456 	char *comment;
457 
458 	g_topology_assert();
459 
460 	gp = LIST_FIRST(&mp->geom);
461 	sc = gp->softc;
462 
463 	if (!strcmp(verb, "attach")) {
464 		gv_attach(sc, req);
465 
466 	} else if (!strcmp(verb, "concat")) {
467 		gv_concat(gp, req);
468 
469 	} else if (!strcmp(verb, "detach")) {
470 		gv_detach(sc, req);
471 
472 	} else if (!strcmp(verb, "list")) {
473 		gv_list(gp, req);
474 
475 	/* Save our configuration back to disk. */
476 	} else if (!strcmp(verb, "saveconfig")) {
477 		gv_post_event(sc, GV_EVENT_SAVE_CONFIG, sc, NULL, 0, 0);
478 
479 	/* Return configuration in string form. */
480 	} else if (!strcmp(verb, "getconfig")) {
481 		comment = gctl_get_param(req, "comment", NULL);
482 		if (comment == NULL) {
483 			gctl_error(req, "no comment parameter given");
484 			return;
485 		}
486 		sb = sbuf_new(NULL, NULL, GV_CFG_LEN, SBUF_FIXEDLEN);
487 		gv_format_config(sc, sb, 0, comment);
488 		sbuf_finish(sb);
489 		gctl_set_param(req, "config", sbuf_data(sb), sbuf_len(sb) + 1);
490 		sbuf_delete(sb);
491 
492 	} else if (!strcmp(verb, "create")) {
493 		gv_create(gp, req);
494 
495 	} else if (!strcmp(verb, "mirror")) {
496 		gv_mirror(gp, req);
497 
498 	} else if (!strcmp(verb, "move")) {
499 		gv_move(gp, req);
500 
501 	} else if (!strcmp(verb, "raid5")) {
502 		gv_raid5(gp, req);
503 
504 	} else if (!strcmp(verb, "rebuildparity") ||
505 	    !strcmp(verb, "checkparity")) {
506 		gv_parityop(sc, req);
507 
508 	} else if (!strcmp(verb, "remove")) {
509 		gv_remove(gp, req);
510 
511 	} else if (!strcmp(verb, "rename")) {
512 		gv_rename(gp, req);
513 
514 	} else if (!strcmp(verb, "resetconfig")) {
515 		gv_post_event(sc, GV_EVENT_RESET_CONFIG, sc, NULL, 0, 0);
516 
517 	} else if (!strcmp(verb, "start")) {
518 		gv_start_obj(gp, req);
519 
520 	} else if (!strcmp(verb, "stripe")) {
521 		gv_stripe(gp, req);
522 
523 	} else if (!strcmp(verb, "setstate")) {
524 		gv_setstate(gp, req);
525 	} else
526 		gctl_error(req, "Unknown verb parameter");
527 }
528 
529 static void
530 gv_parityop(struct gv_softc *sc, struct gctl_req *req)
531 {
532 	struct gv_plex *p;
533 	int *flags, *rebuild, type;
534 	char *plex;
535 
536 	plex = gctl_get_param(req, "plex", NULL);
537 	if (plex == NULL) {
538 		gctl_error(req, "no plex given");
539 		return;
540 	}
541 
542 	flags = gctl_get_paraml(req, "flags", sizeof(*flags));
543 	if (flags == NULL) {
544 		gctl_error(req, "no flags given");
545 		return;
546 	}
547 
548 	rebuild = gctl_get_paraml(req, "rebuild", sizeof(*rebuild));
549 	if (rebuild == NULL) {
550 		gctl_error(req, "no operation given");
551 		return;
552 	}
553 
554 	type = gv_object_type(sc, plex);
555 	if (type != GV_TYPE_PLEX) {
556 		gctl_error(req, "'%s' is not a plex", plex);
557 		return;
558 	}
559 	p = gv_find_plex(sc, plex);
560 
561 	if (p->state != GV_PLEX_UP) {
562 		gctl_error(req, "plex %s is not completely accessible",
563 		    p->name);
564 		return;
565 	}
566 
567 	if (p->org != GV_PLEX_RAID5) {
568 		gctl_error(req, "plex %s is not a RAID5 plex", p->name);
569 		return;
570 	}
571 
572 	/* Put it in the event queue. */
573 	/* XXX: The state of the plex might have changed when this event is
574 	 * picked up ... We should perhaps check this afterwards. */
575 	if (*rebuild)
576 		gv_post_event(sc, GV_EVENT_PARITY_REBUILD, p, NULL, 0, 0);
577 	else
578 		gv_post_event(sc, GV_EVENT_PARITY_CHECK, p, NULL, 0, 0);
579 }
580 
581 
582 static struct g_geom *
583 gv_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
584 {
585 	struct g_geom *gp;
586 	struct g_consumer *cp;
587 	struct gv_softc *sc;
588 	struct gv_hdr vhdr;
589 	int error;
590 
591  	g_topology_assert();
592 	g_trace(G_T_TOPOLOGY, "gv_taste(%s, %s)", mp->name, pp->name);
593 
594 	gp = LIST_FIRST(&mp->geom);
595 	if (gp == NULL) {
596 		G_VINUM_DEBUG(0, "error: tasting, but not initialized?");
597 		return (NULL);
598 	}
599 	sc = gp->softc;
600 
601 	cp = g_new_consumer(gp);
602 	if (g_attach(cp, pp) != 0) {
603 		g_destroy_consumer(cp);
604 		return (NULL);
605 	}
606 	if (g_access(cp, 1, 0, 0) != 0) {
607 		g_detach(cp);
608 		g_destroy_consumer(cp);
609 		return (NULL);
610 	}
611 	g_topology_unlock();
612 
613 	error = gv_read_header(cp, &vhdr);
614 
615 	g_topology_lock();
616 	g_access(cp, -1, 0, 0);
617 	g_detach(cp);
618 	g_destroy_consumer(cp);
619 
620 	/* Check if what we've been given is a valid vinum drive. */
621 	if (!error)
622 		gv_post_event(sc, GV_EVENT_DRIVE_TASTED, pp, NULL, 0, 0);
623 
624 	return (NULL);
625 }
626 
627 void
628 gv_worker(void *arg)
629 {
630 	struct g_provider *pp;
631 	struct gv_softc *sc;
632 	struct gv_event *ev;
633 	struct gv_volume *v;
634 	struct gv_plex *p;
635 	struct gv_sd *s;
636 	struct gv_drive *d;
637 	struct bio *bp;
638 	int newstate, flags, err, rename;
639 	char *newname;
640 	off_t offset;
641 
642 	sc = arg;
643 	KASSERT(sc != NULL, ("NULL sc"));
644 	for (;;) {
645 		/* Look at the events first... */
646 		ev = gv_get_event(sc);
647 		if (ev != NULL) {
648 			gv_remove_event(sc, ev);
649 
650 			switch (ev->type) {
651 			case GV_EVENT_DRIVE_TASTED:
652 				G_VINUM_DEBUG(2, "event 'drive tasted'");
653 				pp = ev->arg1;
654 				gv_drive_tasted(sc, pp);
655 				break;
656 
657 			case GV_EVENT_DRIVE_LOST:
658 				G_VINUM_DEBUG(2, "event 'drive lost'");
659 				d = ev->arg1;
660 				gv_drive_lost(sc, d);
661 				break;
662 
663 			case GV_EVENT_CREATE_DRIVE:
664 				G_VINUM_DEBUG(2, "event 'create drive'");
665 				d = ev->arg1;
666 				gv_create_drive(sc, d);
667 				break;
668 
669 			case GV_EVENT_CREATE_VOLUME:
670 				G_VINUM_DEBUG(2, "event 'create volume'");
671 				v = ev->arg1;
672 				gv_create_volume(sc, v);
673 				break;
674 
675 			case GV_EVENT_CREATE_PLEX:
676 				G_VINUM_DEBUG(2, "event 'create plex'");
677 				p = ev->arg1;
678 				gv_create_plex(sc, p);
679 				break;
680 
681 			case GV_EVENT_CREATE_SD:
682 				G_VINUM_DEBUG(2, "event 'create sd'");
683 				s = ev->arg1;
684 				gv_create_sd(sc, s);
685 				break;
686 
687 			case GV_EVENT_RM_DRIVE:
688 				G_VINUM_DEBUG(2, "event 'remove drive'");
689 				d = ev->arg1;
690 				flags = ev->arg3;
691 				gv_rm_drive(sc, d, flags);
692 				/*gv_setup_objects(sc);*/
693 				break;
694 
695 			case GV_EVENT_RM_VOLUME:
696 				G_VINUM_DEBUG(2, "event 'remove volume'");
697 				v = ev->arg1;
698 				gv_rm_vol(sc, v);
699 				/*gv_setup_objects(sc);*/
700 				break;
701 
702 			case GV_EVENT_RM_PLEX:
703 				G_VINUM_DEBUG(2, "event 'remove plex'");
704 				p = ev->arg1;
705 				gv_rm_plex(sc, p);
706 				/*gv_setup_objects(sc);*/
707 				break;
708 
709 			case GV_EVENT_RM_SD:
710 				G_VINUM_DEBUG(2, "event 'remove sd'");
711 				s = ev->arg1;
712 				gv_rm_sd(sc, s);
713 				/*gv_setup_objects(sc);*/
714 				break;
715 
716 			case GV_EVENT_SAVE_CONFIG:
717 				G_VINUM_DEBUG(2, "event 'save config'");
718 				gv_save_config(sc);
719 				break;
720 
721 			case GV_EVENT_SET_SD_STATE:
722 				G_VINUM_DEBUG(2, "event 'setstate sd'");
723 				s = ev->arg1;
724 				newstate = ev->arg3;
725 				flags = ev->arg4;
726 				err = gv_set_sd_state(s, newstate, flags);
727 				if (err)
728 					G_VINUM_DEBUG(0, "error setting subdisk"
729 					    " state: error code %d", err);
730 				break;
731 
732 			case GV_EVENT_SET_DRIVE_STATE:
733 				G_VINUM_DEBUG(2, "event 'setstate drive'");
734 				d = ev->arg1;
735 				newstate = ev->arg3;
736 				flags = ev->arg4;
737 				err = gv_set_drive_state(d, newstate, flags);
738 				if (err)
739 					G_VINUM_DEBUG(0, "error setting drive "
740 					    "state: error code %d", err);
741 				break;
742 
743 			case GV_EVENT_SET_VOL_STATE:
744 				G_VINUM_DEBUG(2, "event 'setstate volume'");
745 				v = ev->arg1;
746 				newstate = ev->arg3;
747 				flags = ev->arg4;
748 				err = gv_set_vol_state(v, newstate, flags);
749 				if (err)
750 					G_VINUM_DEBUG(0, "error setting volume "
751 					    "state: error code %d", err);
752 				break;
753 
754 			case GV_EVENT_SET_PLEX_STATE:
755 				G_VINUM_DEBUG(2, "event 'setstate plex'");
756 				p = ev->arg1;
757 				newstate = ev->arg3;
758 				flags = ev->arg4;
759 				err = gv_set_plex_state(p, newstate, flags);
760 				if (err)
761 					G_VINUM_DEBUG(0, "error setting plex "
762 					    "state: error code %d", err);
763 				break;
764 
765 			case GV_EVENT_SETUP_OBJECTS:
766 				G_VINUM_DEBUG(2, "event 'setup objects'");
767 				gv_setup_objects(sc);
768 				break;
769 
770 			case GV_EVENT_RESET_CONFIG:
771 				G_VINUM_DEBUG(2, "event 'resetconfig'");
772 				err = gv_resetconfig(sc);
773 				if (err)
774 					G_VINUM_DEBUG(0, "error resetting "
775 					    "config: error code %d", err);
776 				break;
777 
778 			case GV_EVENT_PARITY_REBUILD:
779 				/*
780 				 * Start the rebuild. The gv_plex_done will
781 				 * handle issuing of the remaining rebuild bio's
782 				 * until it's finished.
783 				 */
784 				G_VINUM_DEBUG(2, "event 'rebuild'");
785 				p = ev->arg1;
786 				if (p->state != GV_PLEX_UP) {
787 					G_VINUM_DEBUG(0, "plex %s is not "
788 					    "completely accessible", p->name);
789 					break;
790 				}
791 				if (p->flags & GV_PLEX_SYNCING ||
792 				    p->flags & GV_PLEX_REBUILDING ||
793 				    p->flags & GV_PLEX_GROWING) {
794 					G_VINUM_DEBUG(0, "plex %s is busy with "
795 					    "syncing or parity build", p->name);
796 					break;
797 				}
798 				p->synced = 0;
799 				p->flags |= GV_PLEX_REBUILDING;
800 				g_topology_assert_not();
801 				g_topology_lock();
802 				err = gv_access(p->vol_sc->provider, 1, 1, 0);
803 				if (err) {
804 					G_VINUM_DEBUG(0, "unable to access "
805 					    "provider");
806 					break;
807 				}
808 				g_topology_unlock();
809 				gv_parity_request(p, GV_BIO_CHECK |
810 				    GV_BIO_PARITY, 0);
811 				break;
812 
813 			case GV_EVENT_PARITY_CHECK:
814 				/* Start parity check. */
815 				G_VINUM_DEBUG(2, "event 'check'");
816 				p = ev->arg1;
817 				if (p->state != GV_PLEX_UP) {
818 					G_VINUM_DEBUG(0, "plex %s is not "
819 					    "completely accessible", p->name);
820 					break;
821 				}
822 				if (p->flags & GV_PLEX_SYNCING ||
823 				    p->flags & GV_PLEX_REBUILDING ||
824 				    p->flags & GV_PLEX_GROWING) {
825 					G_VINUM_DEBUG(0, "plex %s is busy with "
826 					    "syncing or parity build", p->name);
827 					break;
828 				}
829 				p->synced = 0;
830 				g_topology_assert_not();
831 				g_topology_lock();
832 				err = gv_access(p->vol_sc->provider, 1, 1, 0);
833 				if (err) {
834 					G_VINUM_DEBUG(0, "unable to access "
835 					    "provider");
836 					break;
837 				}
838 				g_topology_unlock();
839 				gv_parity_request(p, GV_BIO_CHECK, 0);
840 				break;
841 
842 			case GV_EVENT_START_PLEX:
843 				G_VINUM_DEBUG(2, "event 'start' plex");
844 				p = ev->arg1;
845 				gv_start_plex(p);
846 				break;
847 
848 			case GV_EVENT_START_VOLUME:
849 				G_VINUM_DEBUG(2, "event 'start' volume");
850 				v = ev->arg1;
851 				gv_start_vol(v);
852 				break;
853 
854 			case GV_EVENT_ATTACH_PLEX:
855 				G_VINUM_DEBUG(2, "event 'attach' plex");
856 				p = ev->arg1;
857 				v = ev->arg2;
858 				rename = ev->arg4;
859 				err = gv_attach_plex(p, v, rename);
860 				if (err)
861 					G_VINUM_DEBUG(0, "error attaching %s to"
862 					    " %s: error code %d", p->name,
863 					    v->name, err);
864 				break;
865 
866 			case GV_EVENT_ATTACH_SD:
867 				G_VINUM_DEBUG(2, "event 'attach' sd");
868 				s = ev->arg1;
869 				p = ev->arg2;
870 				offset = ev->arg3;
871 				rename = ev->arg4;
872 				err = gv_attach_sd(s, p, offset, rename);
873 				if (err)
874 					G_VINUM_DEBUG(0, "error attaching %s to"
875 					    " %s: error code %d", s->name,
876 					    p->name, err);
877 				break;
878 
879 			case GV_EVENT_DETACH_PLEX:
880 				G_VINUM_DEBUG(2, "event 'detach' plex");
881 				p = ev->arg1;
882 				flags = ev->arg3;
883 				err = gv_detach_plex(p, flags);
884 				if (err)
885 					G_VINUM_DEBUG(0, "error detaching %s: "
886 					    "error code %d", p->name, err);
887 				break;
888 
889 			case GV_EVENT_DETACH_SD:
890 				G_VINUM_DEBUG(2, "event 'detach' sd");
891 				s = ev->arg1;
892 				flags = ev->arg3;
893 				err = gv_detach_sd(s, flags);
894 				if (err)
895 					G_VINUM_DEBUG(0, "error detaching %s: "
896 					    "error code %d", s->name, err);
897 				break;
898 
899 			case GV_EVENT_RENAME_VOL:
900 				G_VINUM_DEBUG(2, "event 'rename' volume");
901 				v = ev->arg1;
902 				newname = ev->arg2;
903 				flags = ev->arg3;
904 				err = gv_rename_vol(sc, v, newname, flags);
905 				if (err)
906 					G_VINUM_DEBUG(0, "error renaming %s to "
907 					    "%s: error code %d", v->name,
908 					    newname, err);
909 				g_free(newname);
910 				/* Destroy and recreate the provider if we can. */
911 				if (gv_provider_is_open(v->provider)) {
912 					G_VINUM_DEBUG(0, "unable to rename "
913 					    "provider to %s: provider in use",
914 					    v->name);
915 					break;
916 				}
917 				g_topology_lock();
918 				g_wither_provider(v->provider, ENOENT);
919 				g_topology_unlock();
920 				v->provider = NULL;
921 				gv_post_event(sc, GV_EVENT_SETUP_OBJECTS, sc,
922 				    NULL, 0, 0);
923 				break;
924 
925 			case GV_EVENT_RENAME_PLEX:
926 				G_VINUM_DEBUG(2, "event 'rename' plex");
927 				p = ev->arg1;
928 				newname = ev->arg2;
929 				flags = ev->arg3;
930 				err = gv_rename_plex(sc, p, newname, flags);
931 				if (err)
932 					G_VINUM_DEBUG(0, "error renaming %s to "
933 					    "%s: error code %d", p->name,
934 					    newname, err);
935 				g_free(newname);
936 				break;
937 
938 			case GV_EVENT_RENAME_SD:
939 				G_VINUM_DEBUG(2, "event 'rename' sd");
940 				s = ev->arg1;
941 				newname = ev->arg2;
942 				flags = ev->arg3;
943 				err = gv_rename_sd(sc, s, newname, flags);
944 				if (err)
945 					G_VINUM_DEBUG(0, "error renaming %s to "
946 					    "%s: error code %d", s->name,
947 					    newname, err);
948 				g_free(newname);
949 				break;
950 
951 			case GV_EVENT_RENAME_DRIVE:
952 				G_VINUM_DEBUG(2, "event 'rename' drive");
953 				d = ev->arg1;
954 				newname = ev->arg2;
955 				flags = ev->arg3;
956 				err = gv_rename_drive(sc, d, newname, flags);
957 				if (err)
958 					G_VINUM_DEBUG(0, "error renaming %s to "
959 					    "%s: error code %d", d->name,
960 					    newname, err);
961 				g_free(newname);
962 				break;
963 
964 			case GV_EVENT_MOVE_SD:
965 				G_VINUM_DEBUG(2, "event 'move' sd");
966 				s = ev->arg1;
967 				d = ev->arg2;
968 				flags = ev->arg3;
969 				err = gv_move_sd(sc, s, d, flags);
970 				if (err)
971 					G_VINUM_DEBUG(0, "error moving %s to "
972 					    "%s: error code %d", s->name,
973 					    d->name, err);
974 				break;
975 
976 			case GV_EVENT_THREAD_EXIT:
977 				G_VINUM_DEBUG(2, "event 'thread exit'");
978 				g_free(ev);
979 				mtx_lock(&sc->equeue_mtx);
980 				mtx_lock(&sc->bqueue_mtx);
981 				gv_cleanup(sc);
982 				mtx_destroy(&sc->bqueue_mtx);
983 				mtx_destroy(&sc->equeue_mtx);
984 				g_free(sc->bqueue_down);
985 				g_free(sc->bqueue_up);
986 				g_free(sc);
987 				kproc_exit(0);
988 				/* NOTREACHED */
989 
990 			default:
991 				G_VINUM_DEBUG(1, "unknown event %d", ev->type);
992 			}
993 
994 			g_free(ev);
995 			continue;
996 		}
997 
998 		/* ... then do I/O processing. */
999 		mtx_lock(&sc->bqueue_mtx);
1000 		/* First do new requests. */
1001 		bp = bioq_takefirst(sc->bqueue_down);
1002 		if (bp != NULL) {
1003 			mtx_unlock(&sc->bqueue_mtx);
1004 			/* A bio that interfered with another bio. */
1005 			if (bp->bio_pflags & GV_BIO_ONHOLD) {
1006 				s = bp->bio_caller1;
1007 				p = s->plex_sc;
1008 				/* Is it still locked out? */
1009 				if (gv_stripe_active(p, bp)) {
1010 					/* Park the bio on the waiting queue. */
1011 					bioq_disksort(p->wqueue, bp);
1012 				} else {
1013 					bp->bio_pflags &= ~GV_BIO_ONHOLD;
1014 					g_io_request(bp, s->drive_sc->consumer);
1015 				}
1016 			/* A special request requireing special handling. */
1017 			} else if (bp->bio_pflags & GV_BIO_INTERNAL) {
1018 				p = bp->bio_caller1;
1019 				gv_plex_start(p, bp);
1020 			} else {
1021 				gv_volume_start(sc, bp);
1022 			}
1023 			mtx_lock(&sc->bqueue_mtx);
1024 		}
1025 		/* Then do completed requests. */
1026 		bp = bioq_takefirst(sc->bqueue_up);
1027 		if (bp == NULL) {
1028 			msleep(sc, &sc->bqueue_mtx, PRIBIO, "-", hz/10);
1029 			mtx_unlock(&sc->bqueue_mtx);
1030 			continue;
1031 		}
1032 		mtx_unlock(&sc->bqueue_mtx);
1033 		gv_bio_done(sc, bp);
1034 	}
1035 }
1036 
1037 #define	VINUM_CLASS_NAME "VINUM"
1038 
1039 static struct g_class g_vinum_class	= {
1040 	.name = VINUM_CLASS_NAME,
1041 	.version = G_VERSION,
1042 	.init = gv_init,
1043 	.taste = gv_taste,
1044 	.ctlreq = gv_config,
1045 	.destroy_geom = gv_unload,
1046 };
1047 
1048 DECLARE_GEOM_CLASS(g_vinum_class, g_vinum);
1049