xref: /freebsd/sys/geom/vinum/geom_vinum.c (revision 02e9120893770924227138ba49df1edb3896112a)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  *  Copyright (c) 2004, 2007 Lukas Ertl
5  *  Copyright (c) 2007, 2009 Ulf Lilleengen
6  *  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  */
30 
31 #include <sys/param.h>
32 #include <sys/bio.h>
33 #include <sys/kernel.h>
34 #include <sys/kthread.h>
35 #include <sys/lock.h>
36 #include <sys/malloc.h>
37 #include <sys/module.h>
38 #include <sys/mutex.h>
39 #include <sys/sbuf.h>
40 #include <sys/sysctl.h>
41 #include <sys/systm.h>
42 
43 #include <geom/geom.h>
44 #include <geom/geom_dbg.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 | CTLFLAG_MPSAFE, 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 static void
62 gv_orphan(struct g_consumer *cp)
63 {
64 	struct g_geom *gp;
65 	struct gv_softc *sc;
66 	struct gv_drive *d;
67 
68 	g_topology_assert();
69 
70 	KASSERT(cp != NULL, ("gv_orphan: null cp"));
71 	gp = cp->geom;
72 	KASSERT(gp != NULL, ("gv_orphan: null gp"));
73 	sc = gp->softc;
74 	KASSERT(sc != NULL, ("gv_orphan: null sc"));
75 	d = cp->private;
76 	KASSERT(d != NULL, ("gv_orphan: null d"));
77 
78 	g_trace(G_T_TOPOLOGY, "gv_orphan(%s)", gp->name);
79 
80 	gv_post_event(sc, GV_EVENT_DRIVE_LOST, d, NULL, 0, 0);
81 }
82 
83 void
84 gv_start(struct bio *bp)
85 {
86 	struct g_geom *gp;
87 	struct gv_softc *sc;
88 
89 	gp = bp->bio_to->geom;
90 	sc = gp->softc;
91 
92 	switch (bp->bio_cmd) {
93 	case BIO_READ:
94 	case BIO_WRITE:
95 	case BIO_DELETE:
96 		break;
97 	case BIO_GETATTR:
98 	default:
99 		g_io_deliver(bp, EOPNOTSUPP);
100 		return;
101 	}
102 	mtx_lock(&sc->bqueue_mtx);
103 	bioq_disksort(sc->bqueue_down, bp);
104 	wakeup(sc);
105 	mtx_unlock(&sc->bqueue_mtx);
106 }
107 
108 void
109 gv_done(struct bio *bp)
110 {
111 	struct g_geom *gp;
112 	struct gv_softc *sc;
113 
114 	KASSERT(bp != NULL, ("NULL bp"));
115 
116 	gp = bp->bio_from->geom;
117 	sc = gp->softc;
118 
119 	mtx_lock(&sc->bqueue_mtx);
120 	bioq_disksort(sc->bqueue_up, bp);
121 	wakeup(sc);
122 	mtx_unlock(&sc->bqueue_mtx);
123 }
124 
125 int
126 gv_access(struct g_provider *pp, int dr, int dw, int de)
127 {
128 	struct g_geom *gp;
129 	struct gv_softc *sc;
130 	struct gv_drive *d, *d2;
131 	int error;
132 
133 	gp = pp->geom;
134 	sc = gp->softc;
135 	/*
136 	 * We want to modify the read count with the write count in case we have
137 	 * plexes in a RAID-5 organization.
138 	 */
139 	dr += dw;
140 
141 	LIST_FOREACH(d, &sc->drives, drive) {
142 		if (d->consumer == NULL)
143 			continue;
144 		error = g_access(d->consumer, dr, dw, de);
145 		if (error) {
146 			LIST_FOREACH(d2, &sc->drives, drive) {
147 				if (d == d2)
148 					break;
149 				g_access(d2->consumer, -dr, -dw, -de);
150 			}
151 			G_VINUM_DEBUG(0, "g_access '%s' failed: %d", d->name,
152 			    error);
153 			return (error);
154 		}
155 	}
156 	return (0);
157 }
158 
159 static void
160 gv_init(struct g_class *mp)
161 {
162 	struct g_geom *gp;
163 	struct gv_softc *sc;
164 
165 	g_trace(G_T_TOPOLOGY, "gv_init(%p)", mp);
166 
167 	gp = g_new_geomf(mp, "VINUM");
168 	gp->spoiled = gv_orphan;
169 	gp->orphan = gv_orphan;
170 	gp->access = gv_access;
171 	gp->start = gv_start;
172 	gp->softc = g_malloc(sizeof(struct gv_softc), M_WAITOK | M_ZERO);
173 	sc = gp->softc;
174 	sc->geom = gp;
175 	sc->bqueue_down = g_malloc(sizeof(struct bio_queue_head),
176 	    M_WAITOK | M_ZERO);
177 	sc->bqueue_up = g_malloc(sizeof(struct bio_queue_head),
178 	    M_WAITOK | M_ZERO);
179 	bioq_init(sc->bqueue_down);
180 	bioq_init(sc->bqueue_up);
181 	LIST_INIT(&sc->drives);
182 	LIST_INIT(&sc->subdisks);
183 	LIST_INIT(&sc->plexes);
184 	LIST_INIT(&sc->volumes);
185 	TAILQ_INIT(&sc->equeue);
186 	mtx_init(&sc->config_mtx, "gv_config", NULL, MTX_DEF);
187 	mtx_init(&sc->equeue_mtx, "gv_equeue", NULL, MTX_DEF);
188 	mtx_init(&sc->bqueue_mtx, "gv_bqueue", NULL, MTX_DEF);
189 	kproc_create(gv_worker, sc, &sc->worker, 0, 0, "gv_worker");
190 }
191 
192 static int
193 gv_unload(struct gctl_req *req, struct g_class *mp, struct g_geom *gp)
194 {
195 	struct gv_softc *sc;
196 
197 	g_trace(G_T_TOPOLOGY, "gv_unload(%p)", mp);
198 
199 	g_topology_assert();
200 	sc = gp->softc;
201 
202 	if (sc != NULL) {
203 		gv_worker_exit(sc);
204 		gp->softc = NULL;
205 		g_wither_geom(gp, ENXIO);
206 	}
207 
208 	return (0);
209 }
210 
211 /* Handle userland request of attaching object. */
212 static void
213 gv_attach(struct gv_softc *sc, struct gctl_req *req)
214 {
215 	struct gv_volume *v;
216 	struct gv_plex *p;
217 	struct gv_sd *s;
218 	off_t *offset;
219 	int *rename, type_child, type_parent;
220 	char *child, *parent;
221 
222 	child = gctl_get_param(req, "child", NULL);
223 	if (child == NULL) {
224 		gctl_error(req, "no child given");
225 		return;
226 	}
227 	parent = gctl_get_param(req, "parent", NULL);
228 	if (parent == NULL) {
229 		gctl_error(req, "no parent given");
230 		return;
231 	}
232 	offset = gctl_get_paraml(req, "offset", sizeof(*offset));
233 	if (offset == NULL) {
234 		gctl_error(req, "no offset given");
235 		return;
236 	}
237 	rename = gctl_get_paraml(req, "rename", sizeof(*rename));
238 	if (rename == NULL) {
239 		gctl_error(req, "no rename flag given");
240 		return;
241 	}
242 
243 	type_child = gv_object_type(sc, child);
244 	type_parent = gv_object_type(sc, parent);
245 
246 	switch (type_child) {
247 	case GV_TYPE_PLEX:
248 		if (type_parent != GV_TYPE_VOL) {
249 			gctl_error(req, "no such volume to attach to");
250 			return;
251 		}
252 		v = gv_find_vol(sc, parent);
253 		p = gv_find_plex(sc, child);
254 		gv_post_event(sc, GV_EVENT_ATTACH_PLEX, p, v, *offset, *rename);
255 		break;
256 	case GV_TYPE_SD:
257 		if (type_parent != GV_TYPE_PLEX) {
258 			gctl_error(req, "no such plex to attach to");
259 			return;
260 		}
261 		p = gv_find_plex(sc, parent);
262 		s = gv_find_sd(sc, child);
263 		gv_post_event(sc, GV_EVENT_ATTACH_SD, s, p, *offset, *rename);
264 		break;
265 	default:
266 		gctl_error(req, "invalid child type");
267 		break;
268 	}
269 }
270 
271 /* Handle userland request of detaching object. */
272 static void
273 gv_detach(struct gv_softc *sc, struct gctl_req *req)
274 {
275 	struct gv_plex *p;
276 	struct gv_sd *s;
277 	int *flags, type;
278 	char *object;
279 
280 	object = gctl_get_param(req, "object", NULL);
281 	if (object == NULL) {
282 		gctl_error(req, "no argument given");
283 		return;
284 	}
285 
286 	flags = gctl_get_paraml(req, "flags", sizeof(*flags));
287 	type = gv_object_type(sc, object);
288 	switch (type) {
289 	case GV_TYPE_PLEX:
290 		p = gv_find_plex(sc, object);
291 		gv_post_event(sc, GV_EVENT_DETACH_PLEX, p, NULL, *flags, 0);
292 		break;
293 	case GV_TYPE_SD:
294 		s = gv_find_sd(sc, object);
295 		gv_post_event(sc, GV_EVENT_DETACH_SD, s, NULL, *flags, 0);
296 		break;
297 	default:
298 		gctl_error(req, "invalid object type");
299 		break;
300 	}
301 }
302 
303 /* Handle userland requests for creating new objects. */
304 static int
305 gv_create(struct g_geom *gp, struct gctl_req *req)
306 {
307 	struct gv_softc *sc;
308 	struct gv_drive *d, *d2;
309 	struct gv_plex *p, *p2;
310 	struct gv_sd *s, *s2;
311 	struct gv_volume *v, *v2;
312 	struct g_provider *pp;
313 	int i, *drives, *flags, *plexes, *subdisks, *volumes;
314 	char buf[20];
315 
316 	g_topology_assert();
317 
318 	sc = gp->softc;
319 
320 	/* Find out how many of each object have been passed in. */
321 	volumes = gctl_get_paraml(req, "volumes", sizeof(*volumes));
322 	plexes = gctl_get_paraml(req, "plexes", sizeof(*plexes));
323 	subdisks = gctl_get_paraml(req, "subdisks", sizeof(*subdisks));
324 	drives = gctl_get_paraml(req, "drives", sizeof(*drives));
325 	if (volumes == NULL || plexes == NULL || subdisks == NULL ||
326 	    drives == NULL) {
327 		gctl_error(req, "number of objects not given");
328 		return (-1);
329 	}
330 	flags = gctl_get_paraml(req, "flags", sizeof(*flags));
331 	if (flags == NULL) {
332 		gctl_error(req, "flags not given");
333 		return (-1);
334 	}
335 
336 	/* First, handle drive definitions ... */
337 	for (i = 0; i < *drives; i++) {
338 		snprintf(buf, sizeof(buf), "drive%d", i);
339 		d2 = gctl_get_paraml(req, buf, sizeof(*d2));
340 		if (d2 == NULL) {
341 			gctl_error(req, "no drive definition given");
342 			return (-1);
343 		}
344 		/*
345 		 * Make sure that the device specified in the drive config is
346 		 * an active GEOM provider.
347 		 */
348 		pp = g_provider_by_name(d2->device);
349 		if (pp == NULL) {
350 			gctl_error(req, "%s: device not found", d2->device);
351 			goto error;
352 		}
353 		if (gv_find_drive(sc, d2->name) != NULL) {
354 			/* Ignore error. */
355 			if (*flags & GV_FLAG_F)
356 				continue;
357 			gctl_error(req, "drive '%s' already exists", d2->name);
358 			goto error;
359 		}
360 		if (gv_find_drive_device(sc, d2->device) != NULL) {
361 			gctl_error(req, "device '%s' already configured in "
362 			    "gvinum", d2->device);
363 			goto error;
364 		}
365 
366 		d = g_malloc(sizeof(*d), M_WAITOK | M_ZERO);
367 		bcopy(d2, d, sizeof(*d));
368 
369 		gv_post_event(sc, GV_EVENT_CREATE_DRIVE, d, NULL, 0, 0);
370 	}
371 
372 	/* ... then volume definitions ... */
373 	for (i = 0; i < *volumes; i++) {
374 		snprintf(buf, sizeof(buf), "volume%d", i);
375 		v2 = gctl_get_paraml(req, buf, sizeof(*v2));
376 		if (v2 == NULL) {
377 			gctl_error(req, "no volume definition given");
378 			return (-1);
379 		}
380 		if (gv_find_vol(sc, v2->name) != NULL) {
381 			/* Ignore error. */
382 			if (*flags & GV_FLAG_F)
383 				continue;
384 			gctl_error(req, "volume '%s' already exists", v2->name);
385 			goto error;
386 		}
387 
388 		v = g_malloc(sizeof(*v), M_WAITOK | M_ZERO);
389 		bcopy(v2, v, sizeof(*v));
390 
391 		gv_post_event(sc, GV_EVENT_CREATE_VOLUME, v, NULL, 0, 0);
392 	}
393 
394 	/* ... then plex definitions ... */
395 	for (i = 0; i < *plexes; i++) {
396 		snprintf(buf, sizeof(buf), "plex%d", i);
397 		p2 = gctl_get_paraml(req, buf, sizeof(*p2));
398 		if (p2 == NULL) {
399 			gctl_error(req, "no plex definition given");
400 			return (-1);
401 		}
402 		if (gv_find_plex(sc, p2->name) != NULL) {
403 			/* Ignore error. */
404 			if (*flags & GV_FLAG_F)
405 				continue;
406 			gctl_error(req, "plex '%s' already exists", p2->name);
407 			goto error;
408 		}
409 
410 		p = g_malloc(sizeof(*p), M_WAITOK | M_ZERO);
411 		bcopy(p2, p, sizeof(*p));
412 
413 		gv_post_event(sc, GV_EVENT_CREATE_PLEX, p, NULL, 0, 0);
414 	}
415 
416 	/* ... and, finally, subdisk definitions. */
417 	for (i = 0; i < *subdisks; i++) {
418 		snprintf(buf, sizeof(buf), "sd%d", i);
419 		s2 = gctl_get_paraml(req, buf, sizeof(*s2));
420 		if (s2 == NULL) {
421 			gctl_error(req, "no subdisk definition given");
422 			return (-1);
423 		}
424 		if (gv_find_sd(sc, s2->name) != NULL) {
425 			/* Ignore error. */
426 			if (*flags & GV_FLAG_F)
427 				continue;
428 			gctl_error(req, "sd '%s' already exists", s2->name);
429 			goto error;
430 		}
431 
432 		s = g_malloc(sizeof(*s), M_WAITOK | M_ZERO);
433 		bcopy(s2, s, sizeof(*s));
434 
435 		gv_post_event(sc, GV_EVENT_CREATE_SD, s, NULL, 0, 0);
436 	}
437 
438 error:
439 	gv_post_event(sc, GV_EVENT_SETUP_OBJECTS, sc, NULL, 0, 0);
440 	gv_post_event(sc, GV_EVENT_SAVE_CONFIG, sc, NULL, 0, 0);
441 
442 	return (0);
443 }
444 
445 static void
446 gv_config(struct gctl_req *req, struct g_class *mp, char const *verb)
447 {
448 	struct g_geom *gp;
449 	struct gv_softc *sc;
450 	struct sbuf *sb;
451 	char *comment;
452 
453 	g_topology_assert();
454 
455 	gp = LIST_FIRST(&mp->geom);
456 	sc = gp->softc;
457 
458 	if (!strcmp(verb, "attach")) {
459 		gv_attach(sc, req);
460 
461 	} else if (!strcmp(verb, "concat")) {
462 		gv_concat(gp, req);
463 
464 	} else if (!strcmp(verb, "detach")) {
465 		gv_detach(sc, req);
466 
467 	} else if (!strcmp(verb, "list")) {
468 		gv_list(gp, req);
469 
470 	/* Save our configuration back to disk. */
471 	} else if (!strcmp(verb, "saveconfig")) {
472 		gv_post_event(sc, GV_EVENT_SAVE_CONFIG, sc, NULL, 0, 0);
473 
474 	/* Return configuration in string form. */
475 	} else if (!strcmp(verb, "getconfig")) {
476 		comment = gctl_get_param(req, "comment", NULL);
477 		if (comment == NULL) {
478 			gctl_error(req, "no comment parameter given");
479 			return;
480 		}
481 		sb = sbuf_new(NULL, NULL, GV_CFG_LEN, SBUF_FIXEDLEN);
482 		gv_format_config(sc, sb, 0, comment);
483 		sbuf_finish(sb);
484 		gctl_set_param(req, "config", sbuf_data(sb), sbuf_len(sb) + 1);
485 		sbuf_delete(sb);
486 
487 	} else if (!strcmp(verb, "create")) {
488 		gv_create(gp, req);
489 
490 	} else if (!strcmp(verb, "mirror")) {
491 		gv_mirror(gp, req);
492 
493 	} else if (!strcmp(verb, "move")) {
494 		gv_move(gp, req);
495 
496 	} else if (!strcmp(verb, "raid5")) {
497 		gv_raid5(gp, req);
498 
499 	} else if (!strcmp(verb, "rebuildparity") ||
500 	    !strcmp(verb, "checkparity")) {
501 		gv_parityop(sc, req);
502 
503 	} else if (!strcmp(verb, "remove")) {
504 		gv_remove(gp, req);
505 
506 	} else if (!strcmp(verb, "rename")) {
507 		gv_rename(gp, req);
508 
509 	} else if (!strcmp(verb, "resetconfig")) {
510 		gv_post_event(sc, GV_EVENT_RESET_CONFIG, sc, NULL, 0, 0);
511 
512 	} else if (!strcmp(verb, "start")) {
513 		gv_start_obj(gp, req);
514 
515 	} else if (!strcmp(verb, "stripe")) {
516 		gv_stripe(gp, req);
517 
518 	} else if (!strcmp(verb, "setstate")) {
519 		gv_setstate(gp, req);
520 	} else
521 		gctl_error(req, "Unknown verb parameter");
522 }
523 
524 static void
525 gv_parityop(struct gv_softc *sc, struct gctl_req *req)
526 {
527 	struct gv_plex *p;
528 	int *flags, *rebuild, type;
529 	char *plex;
530 
531 	plex = gctl_get_param(req, "plex", NULL);
532 	if (plex == NULL) {
533 		gctl_error(req, "no plex given");
534 		return;
535 	}
536 
537 	flags = gctl_get_paraml(req, "flags", sizeof(*flags));
538 	if (flags == NULL) {
539 		gctl_error(req, "no flags given");
540 		return;
541 	}
542 
543 	rebuild = gctl_get_paraml(req, "rebuild", sizeof(*rebuild));
544 	if (rebuild == NULL) {
545 		gctl_error(req, "no operation given");
546 		return;
547 	}
548 
549 	type = gv_object_type(sc, plex);
550 	if (type != GV_TYPE_PLEX) {
551 		gctl_error(req, "'%s' is not a plex", plex);
552 		return;
553 	}
554 	p = gv_find_plex(sc, plex);
555 
556 	if (p->state != GV_PLEX_UP) {
557 		gctl_error(req, "plex %s is not completely accessible",
558 		    p->name);
559 		return;
560 	}
561 
562 	if (p->org != GV_PLEX_RAID5) {
563 		gctl_error(req, "plex %s is not a RAID5 plex", p->name);
564 		return;
565 	}
566 
567 	/* Put it in the event queue. */
568 	/* XXX: The state of the plex might have changed when this event is
569 	 * picked up ... We should perhaps check this afterwards. */
570 	if (*rebuild)
571 		gv_post_event(sc, GV_EVENT_PARITY_REBUILD, p, NULL, 0, 0);
572 	else
573 		gv_post_event(sc, GV_EVENT_PARITY_CHECK, p, NULL, 0, 0);
574 }
575 
576 static struct g_geom *
577 gv_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
578 {
579 	struct g_geom *gp;
580 	struct g_consumer *cp;
581 	struct gv_softc *sc;
582 	struct gv_hdr vhdr;
583 	int error;
584 
585  	g_topology_assert();
586 	g_trace(G_T_TOPOLOGY, "gv_taste(%s, %s)", mp->name, pp->name);
587 
588 	gp = LIST_FIRST(&mp->geom);
589 	if (gp == NULL) {
590 		G_VINUM_DEBUG(0, "error: tasting, but not initialized?");
591 		return (NULL);
592 	}
593 	sc = gp->softc;
594 
595 	cp = g_new_consumer(gp);
596 	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
597 	if (g_attach(cp, pp) != 0) {
598 		g_destroy_consumer(cp);
599 		return (NULL);
600 	}
601 	if (g_access(cp, 1, 0, 0) != 0) {
602 		g_detach(cp);
603 		g_destroy_consumer(cp);
604 		return (NULL);
605 	}
606 	g_topology_unlock();
607 
608 	error = gv_read_header(cp, &vhdr);
609 
610 	g_topology_lock();
611 	g_access(cp, -1, 0, 0);
612 	g_detach(cp);
613 	g_destroy_consumer(cp);
614 
615 	/* Check if what we've been given is a valid vinum drive. */
616 	if (!error)
617 		gv_post_event(sc, GV_EVENT_DRIVE_TASTED, pp, NULL, 0, 0);
618 
619 	return (NULL);
620 }
621 
622 void
623 gv_worker(void *arg)
624 {
625 	struct g_provider *pp;
626 	struct gv_softc *sc;
627 	struct gv_event *ev;
628 	struct gv_volume *v;
629 	struct gv_plex *p;
630 	struct gv_sd *s;
631 	struct gv_drive *d;
632 	struct bio *bp;
633 	int newstate, flags, err, rename;
634 	char *newname;
635 	off_t offset;
636 
637 	sc = arg;
638 	KASSERT(sc != NULL, ("NULL sc"));
639 	for (;;) {
640 		/* Look at the events first... */
641 		ev = gv_get_event(sc);
642 		if (ev != NULL) {
643 			gv_remove_event(sc, ev);
644 
645 			switch (ev->type) {
646 			case GV_EVENT_DRIVE_TASTED:
647 				G_VINUM_DEBUG(2, "event 'drive tasted'");
648 				pp = ev->arg1;
649 				gv_drive_tasted(sc, pp);
650 				break;
651 
652 			case GV_EVENT_DRIVE_LOST:
653 				G_VINUM_DEBUG(2, "event 'drive lost'");
654 				d = ev->arg1;
655 				gv_drive_lost(sc, d);
656 				break;
657 
658 			case GV_EVENT_CREATE_DRIVE:
659 				G_VINUM_DEBUG(2, "event 'create drive'");
660 				d = ev->arg1;
661 				gv_create_drive(sc, d);
662 				break;
663 
664 			case GV_EVENT_CREATE_VOLUME:
665 				G_VINUM_DEBUG(2, "event 'create volume'");
666 				v = ev->arg1;
667 				gv_create_volume(sc, v);
668 				break;
669 
670 			case GV_EVENT_CREATE_PLEX:
671 				G_VINUM_DEBUG(2, "event 'create plex'");
672 				p = ev->arg1;
673 				gv_create_plex(sc, p);
674 				break;
675 
676 			case GV_EVENT_CREATE_SD:
677 				G_VINUM_DEBUG(2, "event 'create sd'");
678 				s = ev->arg1;
679 				gv_create_sd(sc, s);
680 				break;
681 
682 			case GV_EVENT_RM_DRIVE:
683 				G_VINUM_DEBUG(2, "event 'remove drive'");
684 				d = ev->arg1;
685 				flags = ev->arg3;
686 				gv_rm_drive(sc, d, flags);
687 				/*gv_setup_objects(sc);*/
688 				break;
689 
690 			case GV_EVENT_RM_VOLUME:
691 				G_VINUM_DEBUG(2, "event 'remove volume'");
692 				v = ev->arg1;
693 				gv_rm_vol(sc, v);
694 				/*gv_setup_objects(sc);*/
695 				break;
696 
697 			case GV_EVENT_RM_PLEX:
698 				G_VINUM_DEBUG(2, "event 'remove plex'");
699 				p = ev->arg1;
700 				gv_rm_plex(sc, p);
701 				/*gv_setup_objects(sc);*/
702 				break;
703 
704 			case GV_EVENT_RM_SD:
705 				G_VINUM_DEBUG(2, "event 'remove sd'");
706 				s = ev->arg1;
707 				gv_rm_sd(sc, s);
708 				/*gv_setup_objects(sc);*/
709 				break;
710 
711 			case GV_EVENT_SAVE_CONFIG:
712 				G_VINUM_DEBUG(2, "event 'save config'");
713 				gv_save_config(sc);
714 				break;
715 
716 			case GV_EVENT_SET_SD_STATE:
717 				G_VINUM_DEBUG(2, "event 'setstate sd'");
718 				s = ev->arg1;
719 				newstate = ev->arg3;
720 				flags = ev->arg4;
721 				err = gv_set_sd_state(s, newstate, flags);
722 				if (err)
723 					G_VINUM_DEBUG(0, "error setting subdisk"
724 					    " state: error code %d", err);
725 				break;
726 
727 			case GV_EVENT_SET_DRIVE_STATE:
728 				G_VINUM_DEBUG(2, "event 'setstate drive'");
729 				d = ev->arg1;
730 				newstate = ev->arg3;
731 				flags = ev->arg4;
732 				err = gv_set_drive_state(d, newstate, flags);
733 				if (err)
734 					G_VINUM_DEBUG(0, "error setting drive "
735 					    "state: error code %d", err);
736 				break;
737 
738 			case GV_EVENT_SET_VOL_STATE:
739 				G_VINUM_DEBUG(2, "event 'setstate volume'");
740 				v = ev->arg1;
741 				newstate = ev->arg3;
742 				flags = ev->arg4;
743 				err = gv_set_vol_state(v, newstate, flags);
744 				if (err)
745 					G_VINUM_DEBUG(0, "error setting volume "
746 					    "state: error code %d", err);
747 				break;
748 
749 			case GV_EVENT_SET_PLEX_STATE:
750 				G_VINUM_DEBUG(2, "event 'setstate plex'");
751 				p = ev->arg1;
752 				newstate = ev->arg3;
753 				flags = ev->arg4;
754 				err = gv_set_plex_state(p, newstate, flags);
755 				if (err)
756 					G_VINUM_DEBUG(0, "error setting plex "
757 					    "state: error code %d", err);
758 				break;
759 
760 			case GV_EVENT_SETUP_OBJECTS:
761 				G_VINUM_DEBUG(2, "event 'setup objects'");
762 				gv_setup_objects(sc);
763 				break;
764 
765 			case GV_EVENT_RESET_CONFIG:
766 				G_VINUM_DEBUG(2, "event 'resetconfig'");
767 				err = gv_resetconfig(sc);
768 				if (err)
769 					G_VINUM_DEBUG(0, "error resetting "
770 					    "config: error code %d", err);
771 				break;
772 
773 			case GV_EVENT_PARITY_REBUILD:
774 				/*
775 				 * Start the rebuild. The gv_plex_done will
776 				 * handle issuing of the remaining rebuild bio's
777 				 * until it's finished.
778 				 */
779 				G_VINUM_DEBUG(2, "event 'rebuild'");
780 				p = ev->arg1;
781 				if (p->state != GV_PLEX_UP) {
782 					G_VINUM_DEBUG(0, "plex %s is not "
783 					    "completely accessible", p->name);
784 					break;
785 				}
786 				if (p->flags & GV_PLEX_SYNCING ||
787 				    p->flags & GV_PLEX_REBUILDING ||
788 				    p->flags & GV_PLEX_GROWING) {
789 					G_VINUM_DEBUG(0, "plex %s is busy with "
790 					    "syncing or parity build", p->name);
791 					break;
792 				}
793 				p->synced = 0;
794 				p->flags |= GV_PLEX_REBUILDING;
795 				g_topology_assert_not();
796 				g_topology_lock();
797 				err = gv_access(p->vol_sc->provider, 1, 1, 0);
798 				if (err) {
799 					G_VINUM_DEBUG(0, "unable to access "
800 					    "provider");
801 					break;
802 				}
803 				g_topology_unlock();
804 				gv_parity_request(p, GV_BIO_CHECK |
805 				    GV_BIO_PARITY, 0);
806 				break;
807 
808 			case GV_EVENT_PARITY_CHECK:
809 				/* Start parity check. */
810 				G_VINUM_DEBUG(2, "event 'check'");
811 				p = ev->arg1;
812 				if (p->state != GV_PLEX_UP) {
813 					G_VINUM_DEBUG(0, "plex %s is not "
814 					    "completely accessible", p->name);
815 					break;
816 				}
817 				if (p->flags & GV_PLEX_SYNCING ||
818 				    p->flags & GV_PLEX_REBUILDING ||
819 				    p->flags & GV_PLEX_GROWING) {
820 					G_VINUM_DEBUG(0, "plex %s is busy with "
821 					    "syncing or parity build", p->name);
822 					break;
823 				}
824 				p->synced = 0;
825 				g_topology_assert_not();
826 				g_topology_lock();
827 				err = gv_access(p->vol_sc->provider, 1, 1, 0);
828 				if (err) {
829 					G_VINUM_DEBUG(0, "unable to access "
830 					    "provider");
831 					break;
832 				}
833 				g_topology_unlock();
834 				gv_parity_request(p, GV_BIO_CHECK, 0);
835 				break;
836 
837 			case GV_EVENT_START_PLEX:
838 				G_VINUM_DEBUG(2, "event 'start' plex");
839 				p = ev->arg1;
840 				gv_start_plex(p);
841 				break;
842 
843 			case GV_EVENT_START_VOLUME:
844 				G_VINUM_DEBUG(2, "event 'start' volume");
845 				v = ev->arg1;
846 				gv_start_vol(v);
847 				break;
848 
849 			case GV_EVENT_ATTACH_PLEX:
850 				G_VINUM_DEBUG(2, "event 'attach' plex");
851 				p = ev->arg1;
852 				v = ev->arg2;
853 				rename = ev->arg4;
854 				err = gv_attach_plex(p, v, rename);
855 				if (err)
856 					G_VINUM_DEBUG(0, "error attaching %s to"
857 					    " %s: error code %d", p->name,
858 					    v->name, err);
859 				break;
860 
861 			case GV_EVENT_ATTACH_SD:
862 				G_VINUM_DEBUG(2, "event 'attach' sd");
863 				s = ev->arg1;
864 				p = ev->arg2;
865 				offset = ev->arg3;
866 				rename = ev->arg4;
867 				err = gv_attach_sd(s, p, offset, rename);
868 				if (err)
869 					G_VINUM_DEBUG(0, "error attaching %s to"
870 					    " %s: error code %d", s->name,
871 					    p->name, err);
872 				break;
873 
874 			case GV_EVENT_DETACH_PLEX:
875 				G_VINUM_DEBUG(2, "event 'detach' plex");
876 				p = ev->arg1;
877 				flags = ev->arg3;
878 				err = gv_detach_plex(p, flags);
879 				if (err)
880 					G_VINUM_DEBUG(0, "error detaching %s: "
881 					    "error code %d", p->name, err);
882 				break;
883 
884 			case GV_EVENT_DETACH_SD:
885 				G_VINUM_DEBUG(2, "event 'detach' sd");
886 				s = ev->arg1;
887 				flags = ev->arg3;
888 				err = gv_detach_sd(s, flags);
889 				if (err)
890 					G_VINUM_DEBUG(0, "error detaching %s: "
891 					    "error code %d", s->name, err);
892 				break;
893 
894 			case GV_EVENT_RENAME_VOL:
895 				G_VINUM_DEBUG(2, "event 'rename' volume");
896 				v = ev->arg1;
897 				newname = ev->arg2;
898 				flags = ev->arg3;
899 				err = gv_rename_vol(sc, v, newname, flags);
900 				if (err)
901 					G_VINUM_DEBUG(0, "error renaming %s to "
902 					    "%s: error code %d", v->name,
903 					    newname, err);
904 				g_free(newname);
905 				/* Destroy and recreate the provider if we can. */
906 				if (gv_provider_is_open(v->provider)) {
907 					G_VINUM_DEBUG(0, "unable to rename "
908 					    "provider to %s: provider in use",
909 					    v->name);
910 					break;
911 				}
912 				g_topology_lock();
913 				g_wither_provider(v->provider, ENOENT);
914 				g_topology_unlock();
915 				v->provider = NULL;
916 				gv_post_event(sc, GV_EVENT_SETUP_OBJECTS, sc,
917 				    NULL, 0, 0);
918 				break;
919 
920 			case GV_EVENT_RENAME_PLEX:
921 				G_VINUM_DEBUG(2, "event 'rename' plex");
922 				p = ev->arg1;
923 				newname = ev->arg2;
924 				flags = ev->arg3;
925 				err = gv_rename_plex(sc, p, newname, flags);
926 				if (err)
927 					G_VINUM_DEBUG(0, "error renaming %s to "
928 					    "%s: error code %d", p->name,
929 					    newname, err);
930 				g_free(newname);
931 				break;
932 
933 			case GV_EVENT_RENAME_SD:
934 				G_VINUM_DEBUG(2, "event 'rename' sd");
935 				s = ev->arg1;
936 				newname = ev->arg2;
937 				flags = ev->arg3;
938 				err = gv_rename_sd(sc, s, newname, flags);
939 				if (err)
940 					G_VINUM_DEBUG(0, "error renaming %s to "
941 					    "%s: error code %d", s->name,
942 					    newname, err);
943 				g_free(newname);
944 				break;
945 
946 			case GV_EVENT_RENAME_DRIVE:
947 				G_VINUM_DEBUG(2, "event 'rename' drive");
948 				d = ev->arg1;
949 				newname = ev->arg2;
950 				flags = ev->arg3;
951 				err = gv_rename_drive(sc, d, newname, flags);
952 				if (err)
953 					G_VINUM_DEBUG(0, "error renaming %s to "
954 					    "%s: error code %d", d->name,
955 					    newname, err);
956 				g_free(newname);
957 				break;
958 
959 			case GV_EVENT_MOVE_SD:
960 				G_VINUM_DEBUG(2, "event 'move' sd");
961 				s = ev->arg1;
962 				d = ev->arg2;
963 				flags = ev->arg3;
964 				err = gv_move_sd(sc, s, d, flags);
965 				if (err)
966 					G_VINUM_DEBUG(0, "error moving %s to "
967 					    "%s: error code %d", s->name,
968 					    d->name, err);
969 				break;
970 
971 			case GV_EVENT_THREAD_EXIT:
972 				G_VINUM_DEBUG(2, "event 'thread exit'");
973 				g_free(ev);
974 				mtx_lock(&sc->equeue_mtx);
975 				mtx_lock(&sc->bqueue_mtx);
976 				gv_cleanup(sc);
977 				mtx_destroy(&sc->bqueue_mtx);
978 				mtx_destroy(&sc->equeue_mtx);
979 				g_free(sc->bqueue_down);
980 				g_free(sc->bqueue_up);
981 				g_free(sc);
982 				kproc_exit(0);
983 				/* NOTREACHED */
984 
985 			default:
986 				G_VINUM_DEBUG(1, "unknown event %d", ev->type);
987 			}
988 
989 			g_free(ev);
990 			continue;
991 		}
992 
993 		/* ... then do I/O processing. */
994 		mtx_lock(&sc->bqueue_mtx);
995 		/* First do new requests. */
996 		bp = bioq_takefirst(sc->bqueue_down);
997 		if (bp != NULL) {
998 			mtx_unlock(&sc->bqueue_mtx);
999 			/* A bio that interfered with another bio. */
1000 			if (bp->bio_pflags & GV_BIO_ONHOLD) {
1001 				s = bp->bio_caller1;
1002 				p = s->plex_sc;
1003 				/* Is it still locked out? */
1004 				if (gv_stripe_active(p, bp)) {
1005 					/* Park the bio on the waiting queue. */
1006 					bioq_disksort(p->wqueue, bp);
1007 				} else {
1008 					bp->bio_pflags &= ~GV_BIO_ONHOLD;
1009 					g_io_request(bp, s->drive_sc->consumer);
1010 				}
1011 			/* A special request requireing special handling. */
1012 			} else if (bp->bio_pflags & GV_BIO_INTERNAL) {
1013 				p = bp->bio_caller1;
1014 				gv_plex_start(p, bp);
1015 			} else {
1016 				gv_volume_start(sc, bp);
1017 			}
1018 			mtx_lock(&sc->bqueue_mtx);
1019 		}
1020 		/* Then do completed requests. */
1021 		bp = bioq_takefirst(sc->bqueue_up);
1022 		if (bp == NULL) {
1023 			msleep(sc, &sc->bqueue_mtx, PRIBIO, "-", hz/10);
1024 			mtx_unlock(&sc->bqueue_mtx);
1025 			continue;
1026 		}
1027 		mtx_unlock(&sc->bqueue_mtx);
1028 		gv_bio_done(sc, bp);
1029 	}
1030 }
1031 
1032 #define	VINUM_CLASS_NAME "VINUM"
1033 
1034 static struct g_class g_vinum_class	= {
1035 	.name = VINUM_CLASS_NAME,
1036 	.version = G_VERSION,
1037 	.init = gv_init,
1038 	.taste = gv_taste,
1039 	.ctlreq = gv_config,
1040 	.destroy_geom = gv_unload,
1041 };
1042 
1043 DECLARE_GEOM_CLASS(g_vinum_class, g_vinum);
1044 MODULE_VERSION(geom_vinum, 0);
1045