xref: /freebsd/sys/geom/gate/g_gate.c (revision 6be3386466ab79a84b48429ae66244f21526d3df)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2004-2006 Pawel Jakub Dawidek <pjd@FreeBSD.org>
5  * Copyright (c) 2009-2010 The FreeBSD Foundation
6  * All rights reserved.
7  *
8  * Portions of this software were developed by Pawel Jakub Dawidek
9  * under sponsorship from the FreeBSD Foundation.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/bio.h>
39 #include <sys/conf.h>
40 #include <sys/kernel.h>
41 #include <sys/kthread.h>
42 #include <sys/fcntl.h>
43 #include <sys/linker.h>
44 #include <sys/lock.h>
45 #include <sys/malloc.h>
46 #include <sys/mutex.h>
47 #include <sys/proc.h>
48 #include <sys/limits.h>
49 #include <sys/queue.h>
50 #include <sys/sbuf.h>
51 #include <sys/sysctl.h>
52 #include <sys/signalvar.h>
53 #include <sys/time.h>
54 #include <machine/atomic.h>
55 
56 #include <geom/geom.h>
57 #include <geom/geom_dbg.h>
58 #include <geom/gate/g_gate.h>
59 
60 FEATURE(geom_gate, "GEOM Gate module");
61 
62 static MALLOC_DEFINE(M_GATE, "gg_data", "GEOM Gate Data");
63 
64 SYSCTL_DECL(_kern_geom);
65 static SYSCTL_NODE(_kern_geom, OID_AUTO, gate, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
66     "GEOM_GATE configuration");
67 static int g_gate_debug = 0;
68 SYSCTL_INT(_kern_geom_gate, OID_AUTO, debug, CTLFLAG_RWTUN, &g_gate_debug, 0,
69     "Debug level");
70 static u_int g_gate_maxunits = 256;
71 SYSCTL_UINT(_kern_geom_gate, OID_AUTO, maxunits, CTLFLAG_RDTUN,
72     &g_gate_maxunits, 0, "Maximum number of ggate devices");
73 
74 struct g_class g_gate_class = {
75 	.name = G_GATE_CLASS_NAME,
76 	.version = G_VERSION,
77 };
78 
79 static struct cdev *status_dev;
80 static d_ioctl_t g_gate_ioctl;
81 static struct cdevsw g_gate_cdevsw = {
82 	.d_version =	D_VERSION,
83 	.d_ioctl =	g_gate_ioctl,
84 	.d_name =	G_GATE_CTL_NAME
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 void
92 g_gate_detach(void *arg, int flags __unused)
93 {
94 	struct g_consumer *cp = arg;
95 
96 	g_topology_assert();
97 	G_GATE_DEBUG(1, "Destroying read consumer on provider %s orphan.",
98 	    cp->provider->name);
99 	(void)g_access(cp, -1, 0, 0);
100 	g_detach(cp);
101 	g_destroy_consumer(cp);
102 }
103 
104 static int
105 g_gate_destroy(struct g_gate_softc *sc, boolean_t force)
106 {
107 	struct bio_queue_head queue;
108 	struct g_provider *pp;
109 	struct g_consumer *cp;
110 	struct g_geom *gp;
111 	struct bio *bp;
112 
113 	g_topology_assert();
114 	mtx_assert(&g_gate_units_lock, MA_OWNED);
115 	pp = sc->sc_provider;
116 	if (!force && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
117 		mtx_unlock(&g_gate_units_lock);
118 		return (EBUSY);
119 	}
120 	mtx_unlock(&g_gate_units_lock);
121 	mtx_lock(&sc->sc_queue_mtx);
122 	if ((sc->sc_flags & G_GATE_FLAG_DESTROY) == 0)
123 		sc->sc_flags |= G_GATE_FLAG_DESTROY;
124 	wakeup(sc);
125 	mtx_unlock(&sc->sc_queue_mtx);
126 	gp = pp->geom;
127 	g_wither_provider(pp, ENXIO);
128 	callout_drain(&sc->sc_callout);
129 	bioq_init(&queue);
130 	mtx_lock(&sc->sc_queue_mtx);
131 	while ((bp = bioq_takefirst(&sc->sc_inqueue)) != NULL) {
132 		sc->sc_queue_count--;
133 		bioq_insert_tail(&queue, bp);
134 	}
135 	while ((bp = bioq_takefirst(&sc->sc_outqueue)) != NULL) {
136 		sc->sc_queue_count--;
137 		bioq_insert_tail(&queue, bp);
138 	}
139 	mtx_unlock(&sc->sc_queue_mtx);
140 	g_topology_unlock();
141 	while ((bp = bioq_takefirst(&queue)) != NULL) {
142 		G_GATE_LOGREQ(1, bp, "Request canceled.");
143 		g_io_deliver(bp, ENXIO);
144 	}
145 	mtx_lock(&g_gate_units_lock);
146 	/* One reference is ours. */
147 	sc->sc_ref--;
148 	while (sc->sc_ref > 0)
149 		msleep(&sc->sc_ref, &g_gate_units_lock, 0, "gg:destroy", 0);
150 	g_gate_units[sc->sc_unit] = NULL;
151 	KASSERT(g_gate_nunits > 0, ("negative g_gate_nunits?"));
152 	g_gate_nunits--;
153 	mtx_unlock(&g_gate_units_lock);
154 	mtx_destroy(&sc->sc_queue_mtx);
155 	mtx_destroy(&sc->sc_read_mtx);
156 	g_topology_lock();
157 	if ((cp = sc->sc_readcons) != NULL) {
158 		sc->sc_readcons = NULL;
159 		(void)g_access(cp, -1, 0, 0);
160 		g_detach(cp);
161 		g_destroy_consumer(cp);
162 	}
163 	G_GATE_DEBUG(1, "Device %s destroyed.", gp->name);
164 	gp->softc = NULL;
165 	g_wither_geom(gp, ENXIO);
166 	sc->sc_provider = NULL;
167 	free(sc, M_GATE);
168 	return (0);
169 }
170 
171 static int
172 g_gate_access(struct g_provider *pp, int dr, int dw, int de)
173 {
174 	struct g_gate_softc *sc;
175 
176 	if (dr <= 0 && dw <= 0 && de <= 0)
177 		return (0);
178 	sc = pp->geom->softc;
179 	if (sc == NULL || (sc->sc_flags & G_GATE_FLAG_DESTROY) != 0)
180 		return (ENXIO);
181 	/* XXX: Hack to allow read-only mounts. */
182 #if 0
183 	if ((sc->sc_flags & G_GATE_FLAG_READONLY) != 0 && dw > 0)
184 		return (EPERM);
185 #endif
186 	if ((sc->sc_flags & G_GATE_FLAG_WRITEONLY) != 0 && dr > 0)
187 		return (EPERM);
188 	return (0);
189 }
190 
191 static void
192 g_gate_queue_io(struct bio *bp)
193 {
194 	struct g_gate_softc *sc;
195 
196 	sc = bp->bio_to->geom->softc;
197 	if (sc == NULL || (sc->sc_flags & G_GATE_FLAG_DESTROY) != 0) {
198 		g_io_deliver(bp, ENXIO);
199 		return;
200 	}
201 
202 	mtx_lock(&sc->sc_queue_mtx);
203 
204 	if (sc->sc_queue_size > 0 && sc->sc_queue_count > sc->sc_queue_size) {
205 		mtx_unlock(&sc->sc_queue_mtx);
206 		G_GATE_LOGREQ(1, bp, "Queue full, request canceled.");
207 		g_io_deliver(bp, ENOMEM);
208 		return;
209 	}
210 
211 	bp->bio_driver1 = (void *)sc->sc_seq;
212 	sc->sc_seq++;
213 	sc->sc_queue_count++;
214 
215 	bioq_insert_tail(&sc->sc_inqueue, bp);
216 	wakeup(sc);
217 
218 	mtx_unlock(&sc->sc_queue_mtx);
219 }
220 
221 static void
222 g_gate_done(struct bio *cbp)
223 {
224 	struct g_gate_softc *sc;
225 	struct bio *pbp;
226 	struct g_consumer *cp;
227 
228 	cp = cbp->bio_from;
229 	pbp = cbp->bio_parent;
230 	if (cbp->bio_error == 0) {
231 		pbp->bio_completed = cbp->bio_completed;
232 		g_destroy_bio(cbp);
233 		pbp->bio_inbed++;
234 		g_io_deliver(pbp, 0);
235 	} else {
236 		/* If direct read failed, pass it through userland daemon. */
237 		g_destroy_bio(cbp);
238 		pbp->bio_children--;
239 		g_gate_queue_io(pbp);
240 	}
241 
242 	sc = cp->geom->softc;
243 	mtx_lock(&sc->sc_read_mtx);
244 	if (--cp->index == 0 && sc->sc_readcons != cp)
245 		g_post_event(g_gate_detach, cp, M_NOWAIT, NULL);
246 	mtx_unlock(&sc->sc_read_mtx);
247 }
248 
249 static void
250 g_gate_start(struct bio *pbp)
251 {
252 	struct g_gate_softc *sc;
253 	struct g_consumer *cp;
254 	struct bio *cbp;
255 
256 	sc = pbp->bio_to->geom->softc;
257 	if (sc == NULL || (sc->sc_flags & G_GATE_FLAG_DESTROY) != 0) {
258 		g_io_deliver(pbp, ENXIO);
259 		return;
260 	}
261 	G_GATE_LOGREQ(2, pbp, "Request received.");
262 	switch (pbp->bio_cmd) {
263 	case BIO_READ:
264 		if (sc->sc_readcons == NULL)
265 			break;
266 		cbp = g_clone_bio(pbp);
267 		if (cbp == NULL) {
268 			g_io_deliver(pbp, ENOMEM);
269 			return;
270 		}
271 		mtx_lock(&sc->sc_read_mtx);
272 		if ((cp = sc->sc_readcons) == NULL) {
273 			mtx_unlock(&sc->sc_read_mtx);
274 			g_destroy_bio(cbp);
275 			pbp->bio_children--;
276 			break;
277 		}
278 		cp->index++;
279 		cbp->bio_offset = pbp->bio_offset + sc->sc_readoffset;
280 		mtx_unlock(&sc->sc_read_mtx);
281 		cbp->bio_done = g_gate_done;
282 		g_io_request(cbp, cp);
283 		return;
284 	case BIO_DELETE:
285 	case BIO_WRITE:
286 	case BIO_FLUSH:
287 	case BIO_SPEEDUP:
288 		/* XXX: Hack to allow read-only mounts. */
289 		if ((sc->sc_flags & G_GATE_FLAG_READONLY) != 0) {
290 			g_io_deliver(pbp, EPERM);
291 			return;
292 		}
293 		break;
294 	case BIO_GETATTR:
295 	default:
296 		G_GATE_LOGREQ(2, pbp, "Ignoring request.");
297 		g_io_deliver(pbp, EOPNOTSUPP);
298 		return;
299 	}
300 
301 	g_gate_queue_io(pbp);
302 }
303 
304 static struct g_gate_softc *
305 g_gate_hold(int unit, const char *name)
306 {
307 	struct g_gate_softc *sc = NULL;
308 
309 	mtx_lock(&g_gate_units_lock);
310 	if (unit >= 0 && unit < g_gate_maxunits)
311 		sc = g_gate_units[unit];
312 	else if (unit == G_GATE_NAME_GIVEN) {
313 		KASSERT(name != NULL, ("name is NULL"));
314 		for (unit = 0; unit < g_gate_maxunits; unit++) {
315 			if (g_gate_units[unit] == NULL)
316 				continue;
317 			if (strcmp(name,
318 			    g_gate_units[unit]->sc_provider->name) != 0) {
319 				continue;
320 			}
321 			sc = g_gate_units[unit];
322 			break;
323 		}
324 	}
325 	if (sc != NULL)
326 		sc->sc_ref++;
327 	mtx_unlock(&g_gate_units_lock);
328 	return (sc);
329 }
330 
331 static void
332 g_gate_release(struct g_gate_softc *sc)
333 {
334 
335 	g_topology_assert_not();
336 	mtx_lock(&g_gate_units_lock);
337 	sc->sc_ref--;
338 	KASSERT(sc->sc_ref >= 0, ("Negative sc_ref for %s.", sc->sc_name));
339 	if (sc->sc_ref == 0 && (sc->sc_flags & G_GATE_FLAG_DESTROY) != 0)
340 		wakeup(&sc->sc_ref);
341 	mtx_unlock(&g_gate_units_lock);
342 }
343 
344 static int
345 g_gate_getunit(int unit, int *errorp)
346 {
347 
348 	mtx_assert(&g_gate_units_lock, MA_OWNED);
349 	if (unit >= 0) {
350 		if (unit >= g_gate_maxunits)
351 			*errorp = EINVAL;
352 		else if (g_gate_units[unit] == NULL)
353 			return (unit);
354 		else
355 			*errorp = EEXIST;
356 	} else {
357 		for (unit = 0; unit < g_gate_maxunits; unit++) {
358 			if (g_gate_units[unit] == NULL)
359 				return (unit);
360 		}
361 		*errorp = ENFILE;
362 	}
363 	return (-1);
364 }
365 
366 static void
367 g_gate_guard(void *arg)
368 {
369 	struct bio_queue_head queue;
370 	struct g_gate_softc *sc;
371 	struct bintime curtime;
372 	struct bio *bp, *bp2;
373 
374 	sc = arg;
375 	binuptime(&curtime);
376 	g_gate_hold(sc->sc_unit, NULL);
377 	bioq_init(&queue);
378 	mtx_lock(&sc->sc_queue_mtx);
379 	TAILQ_FOREACH_SAFE(bp, &sc->sc_inqueue.queue, bio_queue, bp2) {
380 		if (curtime.sec - bp->bio_t0.sec < 5)
381 			continue;
382 		bioq_remove(&sc->sc_inqueue, bp);
383 		sc->sc_queue_count--;
384 		bioq_insert_tail(&queue, bp);
385 	}
386 	TAILQ_FOREACH_SAFE(bp, &sc->sc_outqueue.queue, bio_queue, bp2) {
387 		if (curtime.sec - bp->bio_t0.sec < 5)
388 			continue;
389 		bioq_remove(&sc->sc_outqueue, bp);
390 		sc->sc_queue_count--;
391 		bioq_insert_tail(&queue, bp);
392 	}
393 	mtx_unlock(&sc->sc_queue_mtx);
394 	while ((bp = bioq_takefirst(&queue)) != NULL) {
395 		G_GATE_LOGREQ(1, bp, "Request timeout.");
396 		g_io_deliver(bp, EIO);
397 	}
398 	if ((sc->sc_flags & G_GATE_FLAG_DESTROY) == 0) {
399 		callout_reset(&sc->sc_callout, sc->sc_timeout * hz,
400 		    g_gate_guard, sc);
401 	}
402 	g_gate_release(sc);
403 }
404 
405 static void
406 g_gate_orphan(struct g_consumer *cp)
407 {
408 	struct g_gate_softc *sc;
409 	struct g_geom *gp;
410 	int done;
411 
412 	g_topology_assert();
413 	gp = cp->geom;
414 	sc = gp->softc;
415 	mtx_lock(&sc->sc_read_mtx);
416 	if (sc->sc_readcons == cp)
417 		sc->sc_readcons = NULL;
418 	done = (cp->index == 0);
419 	mtx_unlock(&sc->sc_read_mtx);
420 	if (done)
421 		g_gate_detach(cp, 0);
422 }
423 
424 static void
425 g_gate_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
426     struct g_consumer *cp, struct g_provider *pp)
427 {
428 	struct g_gate_softc *sc;
429 
430 	sc = gp->softc;
431 	if (sc == NULL || pp != NULL || cp != NULL)
432 		return;
433 	sc = g_gate_hold(sc->sc_unit, NULL);
434 	if (sc == NULL)
435 		return;
436 	if ((sc->sc_flags & G_GATE_FLAG_READONLY) != 0) {
437 		sbuf_printf(sb, "%s<access>%s</access>\n", indent, "read-only");
438 	} else if ((sc->sc_flags & G_GATE_FLAG_WRITEONLY) != 0) {
439 		sbuf_printf(sb, "%s<access>%s</access>\n", indent,
440 		    "write-only");
441 	} else {
442 		sbuf_printf(sb, "%s<access>%s</access>\n", indent,
443 		    "read-write");
444 	}
445 	if (sc->sc_readcons != NULL) {
446 		sbuf_printf(sb, "%s<read_offset>%jd</read_offset>\n",
447 		    indent, (intmax_t)sc->sc_readoffset);
448 		sbuf_printf(sb, "%s<read_provider>%s</read_provider>\n",
449 		    indent, sc->sc_readcons->provider->name);
450 	}
451 	sbuf_printf(sb, "%s<timeout>%u</timeout>\n", indent, sc->sc_timeout);
452 	sbuf_printf(sb, "%s<info>%s</info>\n", indent, sc->sc_info);
453 	sbuf_printf(sb, "%s<queue_count>%u</queue_count>\n", indent,
454 	    sc->sc_queue_count);
455 	sbuf_printf(sb, "%s<queue_size>%u</queue_size>\n", indent,
456 	    sc->sc_queue_size);
457 	sbuf_printf(sb, "%s<ref>%u</ref>\n", indent, sc->sc_ref);
458 	sbuf_printf(sb, "%s<unit>%d</unit>\n", indent, sc->sc_unit);
459 	g_topology_unlock();
460 	g_gate_release(sc);
461 	g_topology_lock();
462 }
463 
464 static int
465 g_gate_create(struct g_gate_ctl_create *ggio)
466 {
467 	struct g_gate_softc *sc;
468 	struct g_geom *gp;
469 	struct g_provider *pp, *ropp;
470 	struct g_consumer *cp;
471 	char name[NAME_MAX];
472 	int error = 0, unit;
473 
474 	if (ggio->gctl_mediasize <= 0) {
475 		G_GATE_DEBUG(1, "Invalid media size.");
476 		return (EINVAL);
477 	}
478 	if (ggio->gctl_sectorsize <= 0) {
479 		G_GATE_DEBUG(1, "Invalid sector size.");
480 		return (EINVAL);
481 	}
482 	if (!powerof2(ggio->gctl_sectorsize)) {
483 		G_GATE_DEBUG(1, "Invalid sector size.");
484 		return (EINVAL);
485 	}
486 	if ((ggio->gctl_mediasize % ggio->gctl_sectorsize) != 0) {
487 		G_GATE_DEBUG(1, "Invalid media size.");
488 		return (EINVAL);
489 	}
490 	if ((ggio->gctl_flags & G_GATE_FLAG_READONLY) != 0 &&
491 	    (ggio->gctl_flags & G_GATE_FLAG_WRITEONLY) != 0) {
492 		G_GATE_DEBUG(1, "Invalid flags.");
493 		return (EINVAL);
494 	}
495 	if (ggio->gctl_unit != G_GATE_UNIT_AUTO &&
496 	    ggio->gctl_unit != G_GATE_NAME_GIVEN &&
497 	    ggio->gctl_unit < 0) {
498 		G_GATE_DEBUG(1, "Invalid unit number.");
499 		return (EINVAL);
500 	}
501 	if (ggio->gctl_unit == G_GATE_NAME_GIVEN &&
502 	    ggio->gctl_name[0] == '\0') {
503 		G_GATE_DEBUG(1, "No device name.");
504 		return (EINVAL);
505 	}
506 
507 	sc = malloc(sizeof(*sc), M_GATE, M_WAITOK | M_ZERO);
508 	sc->sc_flags = (ggio->gctl_flags & G_GATE_USERFLAGS);
509 	strlcpy(sc->sc_info, ggio->gctl_info, sizeof(sc->sc_info));
510 	sc->sc_seq = 1;
511 	bioq_init(&sc->sc_inqueue);
512 	bioq_init(&sc->sc_outqueue);
513 	mtx_init(&sc->sc_queue_mtx, "gg:queue", NULL, MTX_DEF);
514 	mtx_init(&sc->sc_read_mtx, "gg:read", NULL, MTX_DEF);
515 	sc->sc_queue_count = 0;
516 	sc->sc_queue_size = ggio->gctl_maxcount;
517 	if (sc->sc_queue_size > G_GATE_MAX_QUEUE_SIZE)
518 		sc->sc_queue_size = G_GATE_MAX_QUEUE_SIZE;
519 	sc->sc_timeout = ggio->gctl_timeout;
520 	callout_init(&sc->sc_callout, 1);
521 
522 	mtx_lock(&g_gate_units_lock);
523 	sc->sc_unit = g_gate_getunit(ggio->gctl_unit, &error);
524 	if (sc->sc_unit < 0)
525 		goto fail1;
526 	if (ggio->gctl_unit == G_GATE_NAME_GIVEN)
527 		snprintf(name, sizeof(name), "%s", ggio->gctl_name);
528 	else {
529 		snprintf(name, sizeof(name), "%s%d", G_GATE_PROVIDER_NAME,
530 		    sc->sc_unit);
531 	}
532 	/* Check for name collision. */
533 	for (unit = 0; unit < g_gate_maxunits; unit++) {
534 		if (g_gate_units[unit] == NULL)
535 			continue;
536 		if (strcmp(name, g_gate_units[unit]->sc_name) != 0)
537 			continue;
538 		error = EEXIST;
539 		goto fail1;
540 	}
541 	sc->sc_name = name;
542 	g_gate_units[sc->sc_unit] = sc;
543 	g_gate_nunits++;
544 	mtx_unlock(&g_gate_units_lock);
545 
546 	g_topology_lock();
547 
548 	if (ggio->gctl_readprov[0] == '\0') {
549 		ropp = NULL;
550 	} else {
551 		ropp = g_provider_by_name(ggio->gctl_readprov);
552 		if (ropp == NULL) {
553 			G_GATE_DEBUG(1, "Provider %s doesn't exist.",
554 			    ggio->gctl_readprov);
555 			error = EINVAL;
556 			goto fail2;
557 		}
558 		if ((ggio->gctl_readoffset % ggio->gctl_sectorsize) != 0) {
559 			G_GATE_DEBUG(1, "Invalid read offset.");
560 			error = EINVAL;
561 			goto fail2;
562 		}
563 		if (ggio->gctl_mediasize + ggio->gctl_readoffset >
564 		    ropp->mediasize) {
565 			G_GATE_DEBUG(1, "Invalid read offset or media size.");
566 			error = EINVAL;
567 			goto fail2;
568 		}
569 	}
570 
571 	gp = g_new_geomf(&g_gate_class, "%s", name);
572 	gp->start = g_gate_start;
573 	gp->access = g_gate_access;
574 	gp->orphan = g_gate_orphan;
575 	gp->dumpconf = g_gate_dumpconf;
576 	gp->softc = sc;
577 
578 	if (ropp != NULL) {
579 		cp = g_new_consumer(gp);
580 		cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
581 		error = g_attach(cp, ropp);
582 		if (error != 0) {
583 			G_GATE_DEBUG(1, "Unable to attach to %s.", ropp->name);
584 			goto fail3;
585 		}
586 		error = g_access(cp, 1, 0, 0);
587 		if (error != 0) {
588 			G_GATE_DEBUG(1, "Unable to access %s.", ropp->name);
589 			g_detach(cp);
590 			goto fail3;
591 		}
592 		sc->sc_readcons = cp;
593 		sc->sc_readoffset = ggio->gctl_readoffset;
594 	}
595 
596 	ggio->gctl_unit = sc->sc_unit;
597 
598 	pp = g_new_providerf(gp, "%s", name);
599 	pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
600 	pp->mediasize = ggio->gctl_mediasize;
601 	pp->sectorsize = ggio->gctl_sectorsize;
602 	sc->sc_provider = pp;
603 	g_error_provider(pp, 0);
604 
605 	g_topology_unlock();
606 	mtx_lock(&g_gate_units_lock);
607 	sc->sc_name = sc->sc_provider->name;
608 	mtx_unlock(&g_gate_units_lock);
609 	G_GATE_DEBUG(1, "Device %s created.", gp->name);
610 
611 	if (sc->sc_timeout > 0) {
612 		callout_reset(&sc->sc_callout, sc->sc_timeout * hz,
613 		    g_gate_guard, sc);
614 	}
615 	return (0);
616 fail3:
617 	g_destroy_consumer(cp);
618 	g_destroy_geom(gp);
619 fail2:
620 	g_topology_unlock();
621 	mtx_lock(&g_gate_units_lock);
622 	g_gate_units[sc->sc_unit] = NULL;
623 	KASSERT(g_gate_nunits > 0, ("negative g_gate_nunits?"));
624 	g_gate_nunits--;
625 fail1:
626 	mtx_unlock(&g_gate_units_lock);
627 	mtx_destroy(&sc->sc_queue_mtx);
628 	mtx_destroy(&sc->sc_read_mtx);
629 	free(sc, M_GATE);
630 	return (error);
631 }
632 
633 static int
634 g_gate_modify(struct g_gate_softc *sc, struct g_gate_ctl_modify *ggio)
635 {
636 	struct g_provider *pp;
637 	struct g_consumer *cp;
638 	int done, error;
639 
640 	if ((ggio->gctl_modify & GG_MODIFY_MEDIASIZE) != 0) {
641 		if (ggio->gctl_mediasize <= 0) {
642 			G_GATE_DEBUG(1, "Invalid media size.");
643 			return (EINVAL);
644 		}
645 		pp = sc->sc_provider;
646 		if ((ggio->gctl_mediasize % pp->sectorsize) != 0) {
647 			G_GATE_DEBUG(1, "Invalid media size.");
648 			return (EINVAL);
649 		}
650 		g_resize_provider(pp, ggio->gctl_mediasize);
651 		return (0);
652 	}
653 
654 	if ((ggio->gctl_modify & GG_MODIFY_INFO) != 0)
655 		(void)strlcpy(sc->sc_info, ggio->gctl_info, sizeof(sc->sc_info));
656 
657 	cp = NULL;
658 
659 	if ((ggio->gctl_modify & GG_MODIFY_READPROV) != 0) {
660 		g_topology_lock();
661 		mtx_lock(&sc->sc_read_mtx);
662 		if ((cp = sc->sc_readcons) != NULL) {
663 			sc->sc_readcons = NULL;
664 			done = (cp->index == 0);
665 			mtx_unlock(&sc->sc_read_mtx);
666 			if (done)
667 				g_gate_detach(cp, 0);
668 		} else
669 			mtx_unlock(&sc->sc_read_mtx);
670 		if (ggio->gctl_readprov[0] != '\0') {
671 			pp = g_provider_by_name(ggio->gctl_readprov);
672 			if (pp == NULL) {
673 				g_topology_unlock();
674 				G_GATE_DEBUG(1, "Provider %s doesn't exist.",
675 				    ggio->gctl_readprov);
676 				return (EINVAL);
677 			}
678 			cp = g_new_consumer(sc->sc_provider->geom);
679 			cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
680 			error = g_attach(cp, pp);
681 			if (error != 0) {
682 				G_GATE_DEBUG(1, "Unable to attach to %s.",
683 				    pp->name);
684 			} else {
685 				error = g_access(cp, 1, 0, 0);
686 				if (error != 0) {
687 					G_GATE_DEBUG(1, "Unable to access %s.",
688 					    pp->name);
689 					g_detach(cp);
690 				}
691 			}
692 			if (error != 0) {
693 				g_destroy_consumer(cp);
694 				g_topology_unlock();
695 				return (error);
696 			}
697 		}
698 	} else {
699 		cp = sc->sc_readcons;
700 	}
701 
702 	if ((ggio->gctl_modify & GG_MODIFY_READOFFSET) != 0) {
703 		if (cp == NULL) {
704 			G_GATE_DEBUG(1, "No read provider.");
705 			return (EINVAL);
706 		}
707 		pp = sc->sc_provider;
708 		if ((ggio->gctl_readoffset % pp->sectorsize) != 0) {
709 			G_GATE_DEBUG(1, "Invalid read offset.");
710 			return (EINVAL);
711 		}
712 		if (pp->mediasize + ggio->gctl_readoffset >
713 		    cp->provider->mediasize) {
714 			G_GATE_DEBUG(1, "Invalid read offset or media size.");
715 			return (EINVAL);
716 		}
717 		sc->sc_readoffset = ggio->gctl_readoffset;
718 	}
719 
720 	if ((ggio->gctl_modify & GG_MODIFY_READPROV) != 0) {
721 		sc->sc_readcons = cp;
722 		g_topology_unlock();
723 	}
724 
725 	return (0);
726 }
727 
728 #define	G_GATE_CHECK_VERSION(ggio)	do {				\
729 	if ((ggio)->gctl_version != G_GATE_VERSION) {			\
730 		printf("Version mismatch %d != %d.\n",			\
731 		    ggio->gctl_version, G_GATE_VERSION);		\
732 		return (EINVAL);					\
733 	}								\
734 } while (0)
735 static int
736 g_gate_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td)
737 {
738 	struct g_gate_softc *sc;
739 	struct bio *bp;
740 	int error = 0;
741 
742 	G_GATE_DEBUG(4, "ioctl(%s, %lx, %p, %x, %p)", devtoname(dev), cmd, addr,
743 	    flags, td);
744 
745 	switch (cmd) {
746 	case G_GATE_CMD_CREATE:
747 	    {
748 		struct g_gate_ctl_create *ggio = (void *)addr;
749 
750 		G_GATE_CHECK_VERSION(ggio);
751 		error = g_gate_create(ggio);
752 		/*
753 		 * Reset TDP_GEOM flag.
754 		 * There are pending events for sure, because we just created
755 		 * new provider and other classes want to taste it, but we
756 		 * cannot answer on I/O requests until we're here.
757 		 */
758 		td->td_pflags &= ~TDP_GEOM;
759 		return (error);
760 	    }
761 	case G_GATE_CMD_MODIFY:
762 	    {
763 		struct g_gate_ctl_modify *ggio = (void *)addr;
764 
765 		G_GATE_CHECK_VERSION(ggio);
766 		sc = g_gate_hold(ggio->gctl_unit, NULL);
767 		if (sc == NULL)
768 			return (ENXIO);
769 		error = g_gate_modify(sc, ggio);
770 		g_gate_release(sc);
771 		return (error);
772 	    }
773 	case G_GATE_CMD_DESTROY:
774 	    {
775 		struct g_gate_ctl_destroy *ggio = (void *)addr;
776 
777 		G_GATE_CHECK_VERSION(ggio);
778 		sc = g_gate_hold(ggio->gctl_unit, ggio->gctl_name);
779 		if (sc == NULL)
780 			return (ENXIO);
781 		g_topology_lock();
782 		mtx_lock(&g_gate_units_lock);
783 		error = g_gate_destroy(sc, ggio->gctl_force);
784 		g_topology_unlock();
785 		if (error != 0)
786 			g_gate_release(sc);
787 		return (error);
788 	    }
789 	case G_GATE_CMD_CANCEL:
790 	    {
791 		struct g_gate_ctl_cancel *ggio = (void *)addr;
792 		struct bio *tbp, *lbp;
793 
794 		G_GATE_CHECK_VERSION(ggio);
795 		sc = g_gate_hold(ggio->gctl_unit, ggio->gctl_name);
796 		if (sc == NULL)
797 			return (ENXIO);
798 		lbp = NULL;
799 		mtx_lock(&sc->sc_queue_mtx);
800 		TAILQ_FOREACH_SAFE(bp, &sc->sc_outqueue.queue, bio_queue, tbp) {
801 			if (ggio->gctl_seq == 0 ||
802 			    ggio->gctl_seq == (uintptr_t)bp->bio_driver1) {
803 				G_GATE_LOGREQ(1, bp, "Request canceled.");
804 				bioq_remove(&sc->sc_outqueue, bp);
805 				/*
806 				 * Be sure to put requests back onto incoming
807 				 * queue in the proper order.
808 				 */
809 				if (lbp == NULL)
810 					bioq_insert_head(&sc->sc_inqueue, bp);
811 				else {
812 					TAILQ_INSERT_AFTER(&sc->sc_inqueue.queue,
813 					    lbp, bp, bio_queue);
814 				}
815 				lbp = bp;
816 				/*
817 				 * If only one request was canceled, leave now.
818 				 */
819 				if (ggio->gctl_seq != 0)
820 					break;
821 			}
822 		}
823 		if (ggio->gctl_unit == G_GATE_NAME_GIVEN)
824 			ggio->gctl_unit = sc->sc_unit;
825 		mtx_unlock(&sc->sc_queue_mtx);
826 		g_gate_release(sc);
827 		return (error);
828 	    }
829 	case G_GATE_CMD_START:
830 	    {
831 		struct g_gate_ctl_io *ggio = (void *)addr;
832 
833 		G_GATE_CHECK_VERSION(ggio);
834 		sc = g_gate_hold(ggio->gctl_unit, NULL);
835 		if (sc == NULL)
836 			return (ENXIO);
837 		error = 0;
838 		for (;;) {
839 			mtx_lock(&sc->sc_queue_mtx);
840 			bp = bioq_first(&sc->sc_inqueue);
841 			if (bp != NULL)
842 				break;
843 			if ((sc->sc_flags & G_GATE_FLAG_DESTROY) != 0) {
844 				ggio->gctl_error = ECANCELED;
845 				mtx_unlock(&sc->sc_queue_mtx);
846 				goto start_end;
847 			}
848 			if (msleep(sc, &sc->sc_queue_mtx,
849 			    PPAUSE | PDROP | PCATCH, "ggwait", 0) != 0) {
850 				ggio->gctl_error = ECANCELED;
851 				goto start_end;
852 			}
853 		}
854 		ggio->gctl_cmd = bp->bio_cmd;
855 		if (bp->bio_cmd == BIO_WRITE &&
856 		    bp->bio_length > ggio->gctl_length) {
857 			mtx_unlock(&sc->sc_queue_mtx);
858 			ggio->gctl_length = bp->bio_length;
859 			ggio->gctl_error = ENOMEM;
860 			goto start_end;
861 		}
862 		bioq_remove(&sc->sc_inqueue, bp);
863 		bioq_insert_tail(&sc->sc_outqueue, bp);
864 		mtx_unlock(&sc->sc_queue_mtx);
865 
866 		ggio->gctl_seq = (uintptr_t)bp->bio_driver1;
867 		ggio->gctl_offset = bp->bio_offset;
868 		ggio->gctl_length = bp->bio_length;
869 
870 		switch (bp->bio_cmd) {
871 		case BIO_READ:
872 		case BIO_DELETE:
873 		case BIO_FLUSH:
874 		case BIO_SPEEDUP:
875 			break;
876 		case BIO_WRITE:
877 			error = copyout(bp->bio_data, ggio->gctl_data,
878 			    bp->bio_length);
879 			if (error != 0) {
880 				mtx_lock(&sc->sc_queue_mtx);
881 				bioq_remove(&sc->sc_outqueue, bp);
882 				bioq_insert_head(&sc->sc_inqueue, bp);
883 				mtx_unlock(&sc->sc_queue_mtx);
884 				goto start_end;
885 			}
886 			break;
887 		}
888 start_end:
889 		g_gate_release(sc);
890 		return (error);
891 	    }
892 	case G_GATE_CMD_DONE:
893 	    {
894 		struct g_gate_ctl_io *ggio = (void *)addr;
895 
896 		G_GATE_CHECK_VERSION(ggio);
897 		sc = g_gate_hold(ggio->gctl_unit, NULL);
898 		if (sc == NULL)
899 			return (ENOENT);
900 		error = 0;
901 		mtx_lock(&sc->sc_queue_mtx);
902 		TAILQ_FOREACH(bp, &sc->sc_outqueue.queue, bio_queue) {
903 			if (ggio->gctl_seq == (uintptr_t)bp->bio_driver1)
904 				break;
905 		}
906 		if (bp != NULL) {
907 			bioq_remove(&sc->sc_outqueue, bp);
908 			sc->sc_queue_count--;
909 		}
910 		mtx_unlock(&sc->sc_queue_mtx);
911 		if (bp == NULL) {
912 			/*
913 			 * Request was probably canceled.
914 			 */
915 			goto done_end;
916 		}
917 		if (ggio->gctl_error == EAGAIN) {
918 			bp->bio_error = 0;
919 			G_GATE_LOGREQ(1, bp, "Request desisted.");
920 			mtx_lock(&sc->sc_queue_mtx);
921 			sc->sc_queue_count++;
922 			bioq_insert_head(&sc->sc_inqueue, bp);
923 			wakeup(sc);
924 			mtx_unlock(&sc->sc_queue_mtx);
925 		} else {
926 			bp->bio_error = ggio->gctl_error;
927 			if (bp->bio_error == 0) {
928 				bp->bio_completed = bp->bio_length;
929 				switch (bp->bio_cmd) {
930 				case BIO_READ:
931 					error = copyin(ggio->gctl_data,
932 					    bp->bio_data, bp->bio_length);
933 					if (error != 0)
934 						bp->bio_error = error;
935 					break;
936 				case BIO_DELETE:
937 				case BIO_WRITE:
938 				case BIO_FLUSH:
939 				case BIO_SPEEDUP:
940 					break;
941 				}
942 			}
943 			G_GATE_LOGREQ(2, bp, "Request done.");
944 			g_io_deliver(bp, bp->bio_error);
945 		}
946 done_end:
947 		g_gate_release(sc);
948 		return (error);
949 	    }
950 	}
951 	return (ENOIOCTL);
952 }
953 
954 static void
955 g_gate_device(void)
956 {
957 
958 	status_dev = make_dev(&g_gate_cdevsw, 0x0, UID_ROOT, GID_WHEEL, 0600,
959 	    G_GATE_CTL_NAME);
960 }
961 
962 static int
963 g_gate_modevent(module_t mod, int type, void *data)
964 {
965 	int error = 0;
966 
967 	switch (type) {
968 	case MOD_LOAD:
969 		mtx_init(&g_gate_units_lock, "gg_units_lock", NULL, MTX_DEF);
970 		g_gate_units = malloc(g_gate_maxunits * sizeof(g_gate_units[0]),
971 		    M_GATE, M_WAITOK | M_ZERO);
972 		g_gate_nunits = 0;
973 		g_gate_device();
974 		break;
975 	case MOD_UNLOAD:
976 		mtx_lock(&g_gate_units_lock);
977 		if (g_gate_nunits > 0) {
978 			mtx_unlock(&g_gate_units_lock);
979 			error = EBUSY;
980 			break;
981 		}
982 		mtx_unlock(&g_gate_units_lock);
983 		mtx_destroy(&g_gate_units_lock);
984 		if (status_dev != NULL)
985 			destroy_dev(status_dev);
986 		free(g_gate_units, M_GATE);
987 		break;
988 	default:
989 		return (EOPNOTSUPP);
990 		break;
991 	}
992 
993 	return (error);
994 }
995 static moduledata_t g_gate_module = {
996 	G_GATE_MOD_NAME,
997 	g_gate_modevent,
998 	NULL
999 };
1000 DECLARE_MODULE(geom_gate, g_gate_module, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
1001 DECLARE_GEOM_CLASS(g_gate_class, g_gate);
1002 MODULE_VERSION(geom_gate, 0);
1003