xref: /freebsd/sys/geom/gate/g_gate.c (revision c6ec7d31830ab1c80edae95ad5e4b9dba10c47ac)
1 /*-
2  * Copyright (c) 2004-2006 Pawel Jakub Dawidek <pjd@FreeBSD.org>
3  * Copyright (c) 2009-2010 The FreeBSD Foundation
4  * All rights reserved.
5  *
6  * Portions of this software were developed by Pawel Jakub Dawidek
7  * under sponsorship from the FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/bio.h>
37 #include <sys/conf.h>
38 #include <sys/kernel.h>
39 #include <sys/kthread.h>
40 #include <sys/fcntl.h>
41 #include <sys/linker.h>
42 #include <sys/lock.h>
43 #include <sys/malloc.h>
44 #include <sys/mutex.h>
45 #include <sys/proc.h>
46 #include <sys/limits.h>
47 #include <sys/queue.h>
48 #include <sys/sbuf.h>
49 #include <sys/sysctl.h>
50 #include <sys/signalvar.h>
51 #include <sys/time.h>
52 #include <machine/atomic.h>
53 
54 #include <geom/geom.h>
55 #include <geom/gate/g_gate.h>
56 
57 FEATURE(geom_gate, "GEOM Gate module");
58 
59 static MALLOC_DEFINE(M_GATE, "gg_data", "GEOM Gate Data");
60 
61 SYSCTL_DECL(_kern_geom);
62 static SYSCTL_NODE(_kern_geom, OID_AUTO, gate, CTLFLAG_RW, 0,
63     "GEOM_GATE configuration");
64 static int g_gate_debug = 0;
65 TUNABLE_INT("kern.geom.gate.debug", &g_gate_debug);
66 SYSCTL_INT(_kern_geom_gate, OID_AUTO, debug, CTLFLAG_RW, &g_gate_debug, 0,
67     "Debug level");
68 static u_int g_gate_maxunits = 256;
69 TUNABLE_INT("kern.geom.gate.maxunits", &g_gate_maxunits);
70 SYSCTL_UINT(_kern_geom_gate, OID_AUTO, maxunits, CTLFLAG_RDTUN,
71     &g_gate_maxunits, 0, "Maximum number of ggate devices");
72 
73 struct g_class g_gate_class = {
74 	.name = G_GATE_CLASS_NAME,
75 	.version = G_VERSION,
76 };
77 
78 static struct cdev *status_dev;
79 static d_ioctl_t g_gate_ioctl;
80 static struct cdevsw g_gate_cdevsw = {
81 	.d_version =	D_VERSION,
82 	.d_ioctl =	g_gate_ioctl,
83 	.d_name =	G_GATE_CTL_NAME
84 };
85 
86 
87 static struct g_gate_softc **g_gate_units;
88 static u_int g_gate_nunits;
89 static struct mtx g_gate_units_lock;
90 
91 static int
92 g_gate_destroy(struct g_gate_softc *sc, boolean_t force)
93 {
94 	struct g_provider *pp;
95 	struct g_consumer *cp;
96 	struct g_geom *gp;
97 	struct bio *bp;
98 
99 	g_topology_assert();
100 	mtx_assert(&g_gate_units_lock, MA_OWNED);
101 	pp = sc->sc_provider;
102 	if (!force && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
103 		mtx_unlock(&g_gate_units_lock);
104 		return (EBUSY);
105 	}
106 	mtx_unlock(&g_gate_units_lock);
107 	mtx_lock(&sc->sc_queue_mtx);
108 	if ((sc->sc_flags & G_GATE_FLAG_DESTROY) == 0)
109 		sc->sc_flags |= G_GATE_FLAG_DESTROY;
110 	wakeup(sc);
111 	mtx_unlock(&sc->sc_queue_mtx);
112 	gp = pp->geom;
113 	pp->flags |= G_PF_WITHER;
114 	g_orphan_provider(pp, ENXIO);
115 	callout_drain(&sc->sc_callout);
116 	mtx_lock(&sc->sc_queue_mtx);
117 	while ((bp = bioq_first(&sc->sc_inqueue)) != NULL) {
118 		bioq_remove(&sc->sc_inqueue, bp);
119 		sc->sc_queue_count--;
120 		G_GATE_LOGREQ(1, bp, "Request canceled.");
121 		g_io_deliver(bp, ENXIO);
122 	}
123 	while ((bp = bioq_first(&sc->sc_outqueue)) != NULL) {
124 		bioq_remove(&sc->sc_outqueue, bp);
125 		sc->sc_queue_count--;
126 		G_GATE_LOGREQ(1, bp, "Request canceled.");
127 		g_io_deliver(bp, ENXIO);
128 	}
129 	mtx_unlock(&sc->sc_queue_mtx);
130 	g_topology_unlock();
131 	mtx_lock(&g_gate_units_lock);
132 	/* One reference is ours. */
133 	sc->sc_ref--;
134 	while (sc->sc_ref > 0)
135 		msleep(&sc->sc_ref, &g_gate_units_lock, 0, "gg:destroy", 0);
136 	g_gate_units[sc->sc_unit] = NULL;
137 	KASSERT(g_gate_nunits > 0, ("negative g_gate_nunits?"));
138 	g_gate_nunits--;
139 	mtx_unlock(&g_gate_units_lock);
140 	mtx_destroy(&sc->sc_queue_mtx);
141 	g_topology_lock();
142 	if ((cp = sc->sc_readcons) != NULL) {
143 		sc->sc_readcons = NULL;
144 		(void)g_access(cp, -1, 0, 0);
145 		g_detach(cp);
146 		g_destroy_consumer(cp);
147 	}
148 	G_GATE_DEBUG(1, "Device %s destroyed.", gp->name);
149 	gp->softc = NULL;
150 	g_wither_geom(gp, ENXIO);
151 	sc->sc_provider = NULL;
152 	free(sc, M_GATE);
153 	return (0);
154 }
155 
156 static int
157 g_gate_access(struct g_provider *pp, int dr, int dw, int de)
158 {
159 	struct g_gate_softc *sc;
160 
161 	if (dr <= 0 && dw <= 0 && de <= 0)
162 		return (0);
163 	sc = pp->geom->softc;
164 	if (sc == NULL || (sc->sc_flags & G_GATE_FLAG_DESTROY) != 0)
165 		return (ENXIO);
166 	/* XXX: Hack to allow read-only mounts. */
167 #if 0
168 	if ((sc->sc_flags & G_GATE_FLAG_READONLY) != 0 && dw > 0)
169 		return (EPERM);
170 #endif
171 	if ((sc->sc_flags & G_GATE_FLAG_WRITEONLY) != 0 && dr > 0)
172 		return (EPERM);
173 	return (0);
174 }
175 
176 static void
177 g_gate_queue_io(struct bio *bp)
178 {
179 	struct g_gate_softc *sc;
180 
181 	sc = bp->bio_to->geom->softc;
182 	if (sc == NULL || (sc->sc_flags & G_GATE_FLAG_DESTROY) != 0) {
183 		g_io_deliver(bp, ENXIO);
184 		return;
185 	}
186 
187 	mtx_lock(&sc->sc_queue_mtx);
188 
189 	if (sc->sc_queue_size > 0 && sc->sc_queue_count > sc->sc_queue_size) {
190 		mtx_unlock(&sc->sc_queue_mtx);
191 		G_GATE_LOGREQ(1, bp, "Queue full, request canceled.");
192 		g_io_deliver(bp, ENOMEM);
193 		return;
194 	}
195 
196 	bp->bio_driver1 = (void *)sc->sc_seq;
197 	sc->sc_seq++;
198 	sc->sc_queue_count++;
199 
200 	bioq_insert_tail(&sc->sc_inqueue, bp);
201 	wakeup(sc);
202 
203 	mtx_unlock(&sc->sc_queue_mtx);
204 }
205 
206 static void
207 g_gate_done(struct bio *cbp)
208 {
209 	struct bio *pbp;
210 
211 	pbp = cbp->bio_parent;
212 	if (cbp->bio_error == 0) {
213 		pbp->bio_completed = cbp->bio_completed;
214 		g_destroy_bio(cbp);
215 		pbp->bio_inbed++;
216 		g_io_deliver(pbp, 0);
217 	} else {
218 		/* If direct read failed, pass it through userland daemon. */
219 		g_destroy_bio(cbp);
220 		pbp->bio_children--;
221 		g_gate_queue_io(pbp);
222 	}
223 }
224 
225 static void
226 g_gate_start(struct bio *pbp)
227 {
228 	struct g_gate_softc *sc;
229 
230 	sc = pbp->bio_to->geom->softc;
231 	if (sc == NULL || (sc->sc_flags & G_GATE_FLAG_DESTROY) != 0) {
232 		g_io_deliver(pbp, ENXIO);
233 		return;
234 	}
235 	G_GATE_LOGREQ(2, pbp, "Request received.");
236 	switch (pbp->bio_cmd) {
237 	case BIO_READ:
238 		if (sc->sc_readcons != NULL) {
239 			struct bio *cbp;
240 
241 			cbp = g_clone_bio(pbp);
242 			if (cbp == NULL) {
243 				g_io_deliver(pbp, ENOMEM);
244 				return;
245 			}
246 			cbp->bio_done = g_gate_done;
247 			cbp->bio_offset = pbp->bio_offset + sc->sc_readoffset;
248 			cbp->bio_data = pbp->bio_data;
249 			cbp->bio_length = pbp->bio_length;
250 			cbp->bio_to = sc->sc_readcons->provider;
251 			g_io_request(cbp, sc->sc_readcons);
252 			return;
253 		}
254 		break;
255 	case BIO_DELETE:
256 	case BIO_WRITE:
257 	case BIO_FLUSH:
258 		/* XXX: Hack to allow read-only mounts. */
259 		if ((sc->sc_flags & G_GATE_FLAG_READONLY) != 0) {
260 			g_io_deliver(pbp, EPERM);
261 			return;
262 		}
263 		break;
264 	case BIO_GETATTR:
265 	default:
266 		G_GATE_LOGREQ(2, pbp, "Ignoring request.");
267 		g_io_deliver(pbp, EOPNOTSUPP);
268 		return;
269 	}
270 
271 	g_gate_queue_io(pbp);
272 }
273 
274 static struct g_gate_softc *
275 g_gate_hold(int unit, const char *name)
276 {
277 	struct g_gate_softc *sc = NULL;
278 
279 	mtx_lock(&g_gate_units_lock);
280 	if (unit >= 0 && unit < g_gate_maxunits)
281 		sc = g_gate_units[unit];
282 	else if (unit == G_GATE_NAME_GIVEN) {
283 		KASSERT(name != NULL, ("name is NULL"));
284 		for (unit = 0; unit < g_gate_maxunits; unit++) {
285 			if (g_gate_units[unit] == NULL)
286 				continue;
287 			if (strcmp(name,
288 			    g_gate_units[unit]->sc_provider->name) != 0) {
289 				continue;
290 			}
291 			sc = g_gate_units[unit];
292 			break;
293 		}
294 	}
295 	if (sc != NULL)
296 		sc->sc_ref++;
297 	mtx_unlock(&g_gate_units_lock);
298 	return (sc);
299 }
300 
301 static void
302 g_gate_release(struct g_gate_softc *sc)
303 {
304 
305 	g_topology_assert_not();
306 	mtx_lock(&g_gate_units_lock);
307 	sc->sc_ref--;
308 	KASSERT(sc->sc_ref >= 0, ("Negative sc_ref for %s.", sc->sc_name));
309 	if (sc->sc_ref == 0 && (sc->sc_flags & G_GATE_FLAG_DESTROY) != 0)
310 		wakeup(&sc->sc_ref);
311 	mtx_unlock(&g_gate_units_lock);
312 }
313 
314 static int
315 g_gate_getunit(int unit, int *errorp)
316 {
317 
318 	mtx_assert(&g_gate_units_lock, MA_OWNED);
319 	if (unit >= 0) {
320 		if (unit >= g_gate_maxunits)
321 			*errorp = EINVAL;
322 		else if (g_gate_units[unit] == NULL)
323 			return (unit);
324 		else
325 			*errorp = EEXIST;
326 	} else {
327 		for (unit = 0; unit < g_gate_maxunits; unit++) {
328 			if (g_gate_units[unit] == NULL)
329 				return (unit);
330 		}
331 		*errorp = ENFILE;
332 	}
333 	return (-1);
334 }
335 
336 static void
337 g_gate_guard(void *arg)
338 {
339 	struct g_gate_softc *sc;
340 	struct bintime curtime;
341 	struct bio *bp, *bp2;
342 
343 	sc = arg;
344 	binuptime(&curtime);
345 	g_gate_hold(sc->sc_unit, NULL);
346 	mtx_lock(&sc->sc_queue_mtx);
347 	TAILQ_FOREACH_SAFE(bp, &sc->sc_inqueue.queue, bio_queue, bp2) {
348 		if (curtime.sec - bp->bio_t0.sec < 5)
349 			continue;
350 		bioq_remove(&sc->sc_inqueue, bp);
351 		sc->sc_queue_count--;
352 		G_GATE_LOGREQ(1, bp, "Request timeout.");
353 		g_io_deliver(bp, EIO);
354 	}
355 	TAILQ_FOREACH_SAFE(bp, &sc->sc_outqueue.queue, bio_queue, bp2) {
356 		if (curtime.sec - bp->bio_t0.sec < 5)
357 			continue;
358 		bioq_remove(&sc->sc_outqueue, bp);
359 		sc->sc_queue_count--;
360 		G_GATE_LOGREQ(1, bp, "Request timeout.");
361 		g_io_deliver(bp, EIO);
362 	}
363 	mtx_unlock(&sc->sc_queue_mtx);
364 	if ((sc->sc_flags & G_GATE_FLAG_DESTROY) == 0) {
365 		callout_reset(&sc->sc_callout, sc->sc_timeout * hz,
366 		    g_gate_guard, sc);
367 	}
368 	g_gate_release(sc);
369 }
370 
371 static void
372 g_gate_orphan(struct g_consumer *cp)
373 {
374 	struct g_gate_softc *sc;
375 	struct g_geom *gp;
376 
377 	g_topology_assert();
378 	gp = cp->geom;
379 	sc = gp->softc;
380 	if (sc == NULL)
381 		return;
382 	KASSERT(cp == sc->sc_readcons, ("cp=%p sc_readcons=%p", cp,
383 	    sc->sc_readcons));
384 	sc->sc_readcons = NULL;
385 	G_GATE_DEBUG(1, "Destroying read consumer on provider %s orphan.",
386 	    cp->provider->name);
387 	(void)g_access(cp, -1, 0, 0);
388 	g_detach(cp);
389 	g_destroy_consumer(cp);
390 }
391 
392 static void
393 g_gate_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
394     struct g_consumer *cp, struct g_provider *pp)
395 {
396 	struct g_gate_softc *sc;
397 
398 	sc = gp->softc;
399 	if (sc == NULL || pp != NULL || cp != NULL)
400 		return;
401 	sc = g_gate_hold(sc->sc_unit, NULL);
402 	if (sc == NULL)
403 		return;
404 	if ((sc->sc_flags & G_GATE_FLAG_READONLY) != 0) {
405 		sbuf_printf(sb, "%s<access>%s</access>\n", indent, "read-only");
406 	} else if ((sc->sc_flags & G_GATE_FLAG_WRITEONLY) != 0) {
407 		sbuf_printf(sb, "%s<access>%s</access>\n", indent,
408 		    "write-only");
409 	} else {
410 		sbuf_printf(sb, "%s<access>%s</access>\n", indent,
411 		    "read-write");
412 	}
413 	if (sc->sc_readcons != NULL) {
414 		sbuf_printf(sb, "%s<read_offset>%jd</read_offset>\n",
415 		    indent, (intmax_t)sc->sc_readoffset);
416 		sbuf_printf(sb, "%s<read_provider>%s</read_provider>\n",
417 		    indent, sc->sc_readcons->provider->name);
418 	}
419 	sbuf_printf(sb, "%s<timeout>%u</timeout>\n", indent, sc->sc_timeout);
420 	sbuf_printf(sb, "%s<info>%s</info>\n", indent, sc->sc_info);
421 	sbuf_printf(sb, "%s<queue_count>%u</queue_count>\n", indent,
422 	    sc->sc_queue_count);
423 	sbuf_printf(sb, "%s<queue_size>%u</queue_size>\n", indent,
424 	    sc->sc_queue_size);
425 	sbuf_printf(sb, "%s<ref>%u</ref>\n", indent, sc->sc_ref);
426 	sbuf_printf(sb, "%s<unit>%d</unit>\n", indent, sc->sc_unit);
427 	g_topology_unlock();
428 	g_gate_release(sc);
429 	g_topology_lock();
430 }
431 
432 static int
433 g_gate_create(struct g_gate_ctl_create *ggio)
434 {
435 	struct g_gate_softc *sc;
436 	struct g_geom *gp;
437 	struct g_provider *pp, *ropp;
438 	struct g_consumer *cp;
439 	char name[NAME_MAX];
440 	int error = 0, unit;
441 
442 	if (ggio->gctl_mediasize <= 0) {
443 		G_GATE_DEBUG(1, "Invalid media size.");
444 		return (EINVAL);
445 	}
446 	if (ggio->gctl_sectorsize <= 0) {
447 		G_GATE_DEBUG(1, "Invalid sector size.");
448 		return (EINVAL);
449 	}
450 	if (!powerof2(ggio->gctl_sectorsize)) {
451 		G_GATE_DEBUG(1, "Invalid sector size.");
452 		return (EINVAL);
453 	}
454 	if ((ggio->gctl_mediasize % ggio->gctl_sectorsize) != 0) {
455 		G_GATE_DEBUG(1, "Invalid media size.");
456 		return (EINVAL);
457 	}
458 	if ((ggio->gctl_flags & G_GATE_FLAG_READONLY) != 0 &&
459 	    (ggio->gctl_flags & G_GATE_FLAG_WRITEONLY) != 0) {
460 		G_GATE_DEBUG(1, "Invalid flags.");
461 		return (EINVAL);
462 	}
463 	if (ggio->gctl_unit != G_GATE_UNIT_AUTO &&
464 	    ggio->gctl_unit != G_GATE_NAME_GIVEN &&
465 	    ggio->gctl_unit < 0) {
466 		G_GATE_DEBUG(1, "Invalid unit number.");
467 		return (EINVAL);
468 	}
469 	if (ggio->gctl_unit == G_GATE_NAME_GIVEN &&
470 	    ggio->gctl_name[0] == '\0') {
471 		G_GATE_DEBUG(1, "No device name.");
472 		return (EINVAL);
473 	}
474 
475 	sc = malloc(sizeof(*sc), M_GATE, M_WAITOK | M_ZERO);
476 	sc->sc_flags = (ggio->gctl_flags & G_GATE_USERFLAGS);
477 	strlcpy(sc->sc_info, ggio->gctl_info, sizeof(sc->sc_info));
478 	sc->sc_seq = 1;
479 	bioq_init(&sc->sc_inqueue);
480 	bioq_init(&sc->sc_outqueue);
481 	mtx_init(&sc->sc_queue_mtx, "gg:queue", NULL, MTX_DEF);
482 	sc->sc_queue_count = 0;
483 	sc->sc_queue_size = ggio->gctl_maxcount;
484 	if (sc->sc_queue_size > G_GATE_MAX_QUEUE_SIZE)
485 		sc->sc_queue_size = G_GATE_MAX_QUEUE_SIZE;
486 	sc->sc_timeout = ggio->gctl_timeout;
487 	callout_init(&sc->sc_callout, CALLOUT_MPSAFE);
488 
489 	mtx_lock(&g_gate_units_lock);
490 	sc->sc_unit = g_gate_getunit(ggio->gctl_unit, &error);
491 	if (sc->sc_unit < 0)
492 		goto fail1;
493 	if (ggio->gctl_unit == G_GATE_NAME_GIVEN)
494 		snprintf(name, sizeof(name), "%s", ggio->gctl_name);
495 	else {
496 		snprintf(name, sizeof(name), "%s%d", G_GATE_PROVIDER_NAME,
497 		    sc->sc_unit);
498 	}
499 	/* Check for name collision. */
500 	for (unit = 0; unit < g_gate_maxunits; unit++) {
501 		if (g_gate_units[unit] == NULL)
502 			continue;
503 		if (strcmp(name, g_gate_units[unit]->sc_name) != 0)
504 			continue;
505 		error = EEXIST;
506 		goto fail1;
507 	}
508 	sc->sc_name = name;
509 	g_gate_units[sc->sc_unit] = sc;
510 	g_gate_nunits++;
511 	mtx_unlock(&g_gate_units_lock);
512 
513 	g_topology_lock();
514 
515 	if (ggio->gctl_readprov[0] == '\0') {
516 		ropp = NULL;
517 	} else {
518 		ropp = g_provider_by_name(ggio->gctl_readprov);
519 		if (ropp == NULL) {
520 			G_GATE_DEBUG(1, "Provider %s doesn't exist.",
521 			    ggio->gctl_readprov);
522 			error = EINVAL;
523 			goto fail2;
524 		}
525 		if ((ggio->gctl_readoffset % ggio->gctl_sectorsize) != 0) {
526 			G_GATE_DEBUG(1, "Invalid read offset.");
527 			error = EINVAL;
528 			goto fail2;
529 		}
530 		if (ggio->gctl_mediasize + ggio->gctl_readoffset >
531 		    ropp->mediasize) {
532 			G_GATE_DEBUG(1, "Invalid read offset or media size.");
533 			error = EINVAL;
534 			goto fail2;
535 		}
536 	}
537 
538 	gp = g_new_geomf(&g_gate_class, "%s", name);
539 	gp->start = g_gate_start;
540 	gp->access = g_gate_access;
541 	gp->orphan = g_gate_orphan;
542 	gp->dumpconf = g_gate_dumpconf;
543 	gp->softc = sc;
544 
545 	if (ropp != NULL) {
546 		cp = g_new_consumer(gp);
547 		error = g_attach(cp, ropp);
548 		if (error != 0) {
549 			G_GATE_DEBUG(1, "Unable to attach to %s.", ropp->name);
550 			goto fail3;
551 		}
552 		error = g_access(cp, 1, 0, 0);
553 		if (error != 0) {
554 			G_GATE_DEBUG(1, "Unable to access %s.", ropp->name);
555 			g_detach(cp);
556 			goto fail3;
557 		}
558 		sc->sc_readcons = cp;
559 		sc->sc_readoffset = ggio->gctl_readoffset;
560 	}
561 
562 	ggio->gctl_unit = sc->sc_unit;
563 
564 	pp = g_new_providerf(gp, "%s", name);
565 	pp->mediasize = ggio->gctl_mediasize;
566 	pp->sectorsize = ggio->gctl_sectorsize;
567 	sc->sc_provider = pp;
568 	g_error_provider(pp, 0);
569 
570 	g_topology_unlock();
571 	mtx_lock(&g_gate_units_lock);
572 	sc->sc_name = sc->sc_provider->name;
573 	mtx_unlock(&g_gate_units_lock);
574 	G_GATE_DEBUG(1, "Device %s created.", gp->name);
575 
576 	if (sc->sc_timeout > 0) {
577 		callout_reset(&sc->sc_callout, sc->sc_timeout * hz,
578 		    g_gate_guard, sc);
579 	}
580 	return (0);
581 fail3:
582 	g_destroy_consumer(cp);
583 	g_destroy_geom(gp);
584 fail2:
585 	g_topology_unlock();
586 	mtx_lock(&g_gate_units_lock);
587 	g_gate_units[sc->sc_unit] = NULL;
588 	KASSERT(g_gate_nunits > 0, ("negative g_gate_nunits?"));
589 	g_gate_nunits--;
590 fail1:
591 	mtx_unlock(&g_gate_units_lock);
592 	mtx_destroy(&sc->sc_queue_mtx);
593 	free(sc, M_GATE);
594 	return (error);
595 }
596 
597 static int
598 g_gate_modify(struct g_gate_softc *sc, struct g_gate_ctl_modify *ggio)
599 {
600 	struct g_provider *pp;
601 	struct g_consumer *cp;
602 	int error;
603 
604 	if ((ggio->gctl_modify & GG_MODIFY_MEDIASIZE) != 0) {
605 		if (ggio->gctl_mediasize <= 0) {
606 			G_GATE_DEBUG(1, "Invalid media size.");
607 			return (EINVAL);
608 		}
609 		pp = sc->sc_provider;
610 		if ((ggio->gctl_mediasize % pp->sectorsize) != 0) {
611 			G_GATE_DEBUG(1, "Invalid media size.");
612 			return (EINVAL);
613 		}
614 		/* TODO */
615 		return (EOPNOTSUPP);
616 	}
617 
618 	if ((ggio->gctl_modify & GG_MODIFY_INFO) != 0)
619 		(void)strlcpy(sc->sc_info, ggio->gctl_info, sizeof(sc->sc_info));
620 
621 	cp = NULL;
622 
623 	if ((ggio->gctl_modify & GG_MODIFY_READPROV) != 0) {
624 		g_topology_lock();
625 		if (sc->sc_readcons != NULL) {
626 			cp = sc->sc_readcons;
627 			sc->sc_readcons = NULL;
628 			(void)g_access(cp, -1, 0, 0);
629 			g_detach(cp);
630 			g_destroy_consumer(cp);
631 		}
632 		if (ggio->gctl_readprov[0] != '\0') {
633 			pp = g_provider_by_name(ggio->gctl_readprov);
634 			if (pp == NULL) {
635 				g_topology_unlock();
636 				G_GATE_DEBUG(1, "Provider %s doesn't exist.",
637 				    ggio->gctl_readprov);
638 				return (EINVAL);
639 			}
640 			cp = g_new_consumer(sc->sc_provider->geom);
641 			error = g_attach(cp, pp);
642 			if (error != 0) {
643 				G_GATE_DEBUG(1, "Unable to attach to %s.",
644 				    pp->name);
645 			} else {
646 				error = g_access(cp, 1, 0, 0);
647 				if (error != 0) {
648 					G_GATE_DEBUG(1, "Unable to access %s.",
649 					    pp->name);
650 					g_detach(cp);
651 				}
652 			}
653 			if (error != 0) {
654 				g_destroy_consumer(cp);
655 				g_topology_unlock();
656 				return (error);
657 			}
658 		}
659 	} else {
660 		cp = sc->sc_readcons;
661 	}
662 
663 	if ((ggio->gctl_modify & GG_MODIFY_READOFFSET) != 0) {
664 		if (cp == NULL) {
665 			G_GATE_DEBUG(1, "No read provider.");
666 			return (EINVAL);
667 		}
668 		pp = sc->sc_provider;
669 		if ((ggio->gctl_readoffset % pp->sectorsize) != 0) {
670 			G_GATE_DEBUG(1, "Invalid read offset.");
671 			return (EINVAL);
672 		}
673 		if (pp->mediasize + ggio->gctl_readoffset >
674 		    cp->provider->mediasize) {
675 			G_GATE_DEBUG(1, "Invalid read offset or media size.");
676 			return (EINVAL);
677 		}
678 		sc->sc_readoffset = ggio->gctl_readoffset;
679 	}
680 
681 	if ((ggio->gctl_modify & GG_MODIFY_READPROV) != 0) {
682 		sc->sc_readcons = cp;
683 		g_topology_unlock();
684 	}
685 
686 	return (0);
687 }
688 
689 #define	G_GATE_CHECK_VERSION(ggio)	do {				\
690 	if ((ggio)->gctl_version != G_GATE_VERSION) {			\
691 		printf("Version mismatch %d != %d.\n",			\
692 		    ggio->gctl_version, G_GATE_VERSION);		\
693 		return (EINVAL);					\
694 	}								\
695 } while (0)
696 static int
697 g_gate_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td)
698 {
699 	struct g_gate_softc *sc;
700 	struct bio *bp;
701 	int error = 0;
702 
703 	G_GATE_DEBUG(4, "ioctl(%s, %lx, %p, %x, %p)", devtoname(dev), cmd, addr,
704 	    flags, td);
705 
706 	switch (cmd) {
707 	case G_GATE_CMD_CREATE:
708 	    {
709 		struct g_gate_ctl_create *ggio = (void *)addr;
710 
711 		G_GATE_CHECK_VERSION(ggio);
712 		error = g_gate_create(ggio);
713 		/*
714 		 * Reset TDP_GEOM flag.
715 		 * There are pending events for sure, because we just created
716 		 * new provider and other classes want to taste it, but we
717 		 * cannot answer on I/O requests until we're here.
718 		 */
719 		td->td_pflags &= ~TDP_GEOM;
720 		return (error);
721 	    }
722 	case G_GATE_CMD_MODIFY:
723 	    {
724 		struct g_gate_ctl_modify *ggio = (void *)addr;
725 
726 		G_GATE_CHECK_VERSION(ggio);
727 		sc = g_gate_hold(ggio->gctl_unit, NULL);
728 		if (sc == NULL)
729 			return (ENXIO);
730 		error = g_gate_modify(sc, ggio);
731 		g_gate_release(sc);
732 		return (error);
733 	    }
734 	case G_GATE_CMD_DESTROY:
735 	    {
736 		struct g_gate_ctl_destroy *ggio = (void *)addr;
737 
738 		G_GATE_CHECK_VERSION(ggio);
739 		sc = g_gate_hold(ggio->gctl_unit, ggio->gctl_name);
740 		if (sc == NULL)
741 			return (ENXIO);
742 		g_topology_lock();
743 		mtx_lock(&g_gate_units_lock);
744 		error = g_gate_destroy(sc, ggio->gctl_force);
745 		g_topology_unlock();
746 		if (error != 0)
747 			g_gate_release(sc);
748 		return (error);
749 	    }
750 	case G_GATE_CMD_CANCEL:
751 	    {
752 		struct g_gate_ctl_cancel *ggio = (void *)addr;
753 		struct bio *tbp, *lbp;
754 
755 		G_GATE_CHECK_VERSION(ggio);
756 		sc = g_gate_hold(ggio->gctl_unit, ggio->gctl_name);
757 		if (sc == NULL)
758 			return (ENXIO);
759 		lbp = NULL;
760 		mtx_lock(&sc->sc_queue_mtx);
761 		TAILQ_FOREACH_SAFE(bp, &sc->sc_outqueue.queue, bio_queue, tbp) {
762 			if (ggio->gctl_seq == 0 ||
763 			    ggio->gctl_seq == (uintptr_t)bp->bio_driver1) {
764 				G_GATE_LOGREQ(1, bp, "Request canceled.");
765 				bioq_remove(&sc->sc_outqueue, bp);
766 				/*
767 				 * Be sure to put requests back onto incoming
768 				 * queue in the proper order.
769 				 */
770 				if (lbp == NULL)
771 					bioq_insert_head(&sc->sc_inqueue, bp);
772 				else {
773 					TAILQ_INSERT_AFTER(&sc->sc_inqueue.queue,
774 					    lbp, bp, bio_queue);
775 				}
776 				lbp = bp;
777 				/*
778 				 * If only one request was canceled, leave now.
779 				 */
780 				if (ggio->gctl_seq != 0)
781 					break;
782 			}
783 		}
784 		if (ggio->gctl_unit == G_GATE_NAME_GIVEN)
785 			ggio->gctl_unit = sc->sc_unit;
786 		mtx_unlock(&sc->sc_queue_mtx);
787 		g_gate_release(sc);
788 		return (error);
789 	    }
790 	case G_GATE_CMD_START:
791 	    {
792 		struct g_gate_ctl_io *ggio = (void *)addr;
793 
794 		G_GATE_CHECK_VERSION(ggio);
795 		sc = g_gate_hold(ggio->gctl_unit, NULL);
796 		if (sc == NULL)
797 			return (ENXIO);
798 		error = 0;
799 		for (;;) {
800 			mtx_lock(&sc->sc_queue_mtx);
801 			bp = bioq_first(&sc->sc_inqueue);
802 			if (bp != NULL)
803 				break;
804 			if ((sc->sc_flags & G_GATE_FLAG_DESTROY) != 0) {
805 				ggio->gctl_error = ECANCELED;
806 				mtx_unlock(&sc->sc_queue_mtx);
807 				goto start_end;
808 			}
809 			if (msleep(sc, &sc->sc_queue_mtx,
810 			    PPAUSE | PDROP | PCATCH, "ggwait", 0) != 0) {
811 				ggio->gctl_error = ECANCELED;
812 				goto start_end;
813 			}
814 		}
815 		ggio->gctl_cmd = bp->bio_cmd;
816 		if ((bp->bio_cmd == BIO_DELETE || bp->bio_cmd == BIO_WRITE) &&
817 		    bp->bio_length > ggio->gctl_length) {
818 			mtx_unlock(&sc->sc_queue_mtx);
819 			ggio->gctl_length = bp->bio_length;
820 			ggio->gctl_error = ENOMEM;
821 			goto start_end;
822 		}
823 		bioq_remove(&sc->sc_inqueue, bp);
824 		bioq_insert_tail(&sc->sc_outqueue, bp);
825 		mtx_unlock(&sc->sc_queue_mtx);
826 
827 		ggio->gctl_seq = (uintptr_t)bp->bio_driver1;
828 		ggio->gctl_offset = bp->bio_offset;
829 		ggio->gctl_length = bp->bio_length;
830 
831 		switch (bp->bio_cmd) {
832 		case BIO_READ:
833 		case BIO_DELETE:
834 		case BIO_FLUSH:
835 			break;
836 		case BIO_WRITE:
837 			error = copyout(bp->bio_data, ggio->gctl_data,
838 			    bp->bio_length);
839 			if (error != 0) {
840 				mtx_lock(&sc->sc_queue_mtx);
841 				bioq_remove(&sc->sc_outqueue, bp);
842 				bioq_insert_head(&sc->sc_inqueue, bp);
843 				mtx_unlock(&sc->sc_queue_mtx);
844 				goto start_end;
845 			}
846 			break;
847 		}
848 start_end:
849 		g_gate_release(sc);
850 		return (error);
851 	    }
852 	case G_GATE_CMD_DONE:
853 	    {
854 		struct g_gate_ctl_io *ggio = (void *)addr;
855 
856 		G_GATE_CHECK_VERSION(ggio);
857 		sc = g_gate_hold(ggio->gctl_unit, NULL);
858 		if (sc == NULL)
859 			return (ENOENT);
860 		error = 0;
861 		mtx_lock(&sc->sc_queue_mtx);
862 		TAILQ_FOREACH(bp, &sc->sc_outqueue.queue, bio_queue) {
863 			if (ggio->gctl_seq == (uintptr_t)bp->bio_driver1)
864 				break;
865 		}
866 		if (bp != NULL) {
867 			bioq_remove(&sc->sc_outqueue, bp);
868 			sc->sc_queue_count--;
869 		}
870 		mtx_unlock(&sc->sc_queue_mtx);
871 		if (bp == NULL) {
872 			/*
873 			 * Request was probably canceled.
874 			 */
875 			goto done_end;
876 		}
877 		if (ggio->gctl_error == EAGAIN) {
878 			bp->bio_error = 0;
879 			G_GATE_LOGREQ(1, bp, "Request desisted.");
880 			mtx_lock(&sc->sc_queue_mtx);
881 			sc->sc_queue_count++;
882 			bioq_insert_head(&sc->sc_inqueue, bp);
883 			wakeup(sc);
884 			mtx_unlock(&sc->sc_queue_mtx);
885 		} else {
886 			bp->bio_error = ggio->gctl_error;
887 			if (bp->bio_error == 0) {
888 				bp->bio_completed = bp->bio_length;
889 				switch (bp->bio_cmd) {
890 				case BIO_READ:
891 					error = copyin(ggio->gctl_data,
892 					    bp->bio_data, bp->bio_length);
893 					if (error != 0)
894 						bp->bio_error = error;
895 					break;
896 				case BIO_DELETE:
897 				case BIO_WRITE:
898 				case BIO_FLUSH:
899 					break;
900 				}
901 			}
902 			G_GATE_LOGREQ(2, bp, "Request done.");
903 			g_io_deliver(bp, bp->bio_error);
904 		}
905 done_end:
906 		g_gate_release(sc);
907 		return (error);
908 	    }
909 	}
910 	return (ENOIOCTL);
911 }
912 
913 static void
914 g_gate_device(void)
915 {
916 
917 	status_dev = make_dev(&g_gate_cdevsw, 0x0, UID_ROOT, GID_WHEEL, 0600,
918 	    G_GATE_CTL_NAME);
919 }
920 
921 static int
922 g_gate_modevent(module_t mod, int type, void *data)
923 {
924 	int error = 0;
925 
926 	switch (type) {
927 	case MOD_LOAD:
928 		mtx_init(&g_gate_units_lock, "gg_units_lock", NULL, MTX_DEF);
929 		g_gate_units = malloc(g_gate_maxunits * sizeof(g_gate_units[0]),
930 		    M_GATE, M_WAITOK | M_ZERO);
931 		g_gate_nunits = 0;
932 		g_gate_device();
933 		break;
934 	case MOD_UNLOAD:
935 		mtx_lock(&g_gate_units_lock);
936 		if (g_gate_nunits > 0) {
937 			mtx_unlock(&g_gate_units_lock);
938 			error = EBUSY;
939 			break;
940 		}
941 		mtx_unlock(&g_gate_units_lock);
942 		mtx_destroy(&g_gate_units_lock);
943 		if (status_dev != 0)
944 			destroy_dev(status_dev);
945 		free(g_gate_units, M_GATE);
946 		break;
947 	default:
948 		return (EOPNOTSUPP);
949 		break;
950 	}
951 
952 	return (error);
953 }
954 static moduledata_t g_gate_module = {
955 	G_GATE_MOD_NAME,
956 	g_gate_modevent,
957 	NULL
958 };
959 DECLARE_MODULE(geom_gate, g_gate_module, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
960 DECLARE_GEOM_CLASS(g_gate_class, g_gate);
961