xref: /freebsd/sys/geom/gate/g_gate.c (revision 81ad6265)
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 + 1];
472 	char readprov[NAME_MAX + 1];
473 	int error = 0, unit;
474 
475 	if (ggio->gctl_mediasize <= 0) {
476 		G_GATE_DEBUG(1, "Invalid media size.");
477 		return (EINVAL);
478 	}
479 	if (ggio->gctl_sectorsize <= 0) {
480 		G_GATE_DEBUG(1, "Invalid sector size.");
481 		return (EINVAL);
482 	}
483 	if (!powerof2(ggio->gctl_sectorsize)) {
484 		G_GATE_DEBUG(1, "Invalid sector size.");
485 		return (EINVAL);
486 	}
487 	if ((ggio->gctl_mediasize % ggio->gctl_sectorsize) != 0) {
488 		G_GATE_DEBUG(1, "Invalid media size.");
489 		return (EINVAL);
490 	}
491 	if ((ggio->gctl_flags & G_GATE_FLAG_READONLY) != 0 &&
492 	    (ggio->gctl_flags & G_GATE_FLAG_WRITEONLY) != 0) {
493 		G_GATE_DEBUG(1, "Invalid flags.");
494 		return (EINVAL);
495 	}
496 	if (ggio->gctl_unit != G_GATE_UNIT_AUTO &&
497 	    ggio->gctl_unit != G_GATE_NAME_GIVEN &&
498 	    ggio->gctl_unit < 0) {
499 		G_GATE_DEBUG(1, "Invalid unit number.");
500 		return (EINVAL);
501 	}
502 	if (ggio->gctl_unit == G_GATE_NAME_GIVEN &&
503 	    ggio->gctl_name[0] == '\0') {
504 		G_GATE_DEBUG(1, "No device name.");
505 		return (EINVAL);
506 	}
507 
508 	sc = malloc(sizeof(*sc), M_GATE, M_WAITOK | M_ZERO);
509 	sc->sc_flags = (ggio->gctl_flags & G_GATE_USERFLAGS);
510 	memset(sc->sc_info, 0, sizeof(sc->sc_info));
511 	strncpy(sc->sc_info, ggio->gctl_info,
512 	    MIN(sizeof(sc->sc_info) - 1, sizeof(ggio->gctl_info)));
513 	sc->sc_seq = 1;
514 	bioq_init(&sc->sc_inqueue);
515 	bioq_init(&sc->sc_outqueue);
516 	mtx_init(&sc->sc_queue_mtx, "gg:queue", NULL, MTX_DEF);
517 	mtx_init(&sc->sc_read_mtx, "gg:read", NULL, MTX_DEF);
518 	sc->sc_queue_count = 0;
519 	sc->sc_queue_size = ggio->gctl_maxcount;
520 	if (sc->sc_queue_size > G_GATE_MAX_QUEUE_SIZE)
521 		sc->sc_queue_size = G_GATE_MAX_QUEUE_SIZE;
522 	sc->sc_timeout = ggio->gctl_timeout;
523 	callout_init(&sc->sc_callout, 1);
524 
525 	mtx_lock(&g_gate_units_lock);
526 	sc->sc_unit = g_gate_getunit(ggio->gctl_unit, &error);
527 	if (sc->sc_unit < 0)
528 		goto fail1;
529 	if (ggio->gctl_unit == G_GATE_NAME_GIVEN) {
530 		memset(name, 0, sizeof(name));
531 		strncpy(name, ggio->gctl_name,
532 		    MIN(sizeof(name) - 1, sizeof(ggio->gctl_name)));
533 	} else {
534 		snprintf(name, sizeof(name), "%s%d", G_GATE_PROVIDER_NAME,
535 		    sc->sc_unit);
536 	}
537 	/* Check for name collision. */
538 	for (unit = 0; unit < g_gate_maxunits; unit++) {
539 		if (g_gate_units[unit] == NULL)
540 			continue;
541 		if (strcmp(name, g_gate_units[unit]->sc_name) != 0)
542 			continue;
543 		error = EEXIST;
544 		goto fail1;
545 	}
546 	// local stack buffer 'name' assigned here temporarily only.
547 	// the real provider name is assigned below.
548 	sc->sc_name = name;
549 	g_gate_units[sc->sc_unit] = sc;
550 	g_gate_nunits++;
551 	mtx_unlock(&g_gate_units_lock);
552 
553 	g_topology_lock();
554 
555 	if (ggio->gctl_readprov[0] == '\0') {
556 		ropp = NULL;
557 	} else {
558 		memset(readprov, 0, sizeof(readprov));
559 		strncpy(readprov, ggio->gctl_readprov,
560 		    MIN(sizeof(readprov) - 1, sizeof(ggio->gctl_readprov)));
561 		ropp = g_provider_by_name(readprov);
562 		if (ropp == NULL) {
563 			G_GATE_DEBUG(1, "Provider %s doesn't exist.", readprov);
564 			error = EINVAL;
565 			goto fail2;
566 		}
567 		if ((ggio->gctl_readoffset % ggio->gctl_sectorsize) != 0) {
568 			G_GATE_DEBUG(1, "Invalid read offset.");
569 			error = EINVAL;
570 			goto fail2;
571 		}
572 		if (ggio->gctl_mediasize + ggio->gctl_readoffset >
573 		    ropp->mediasize) {
574 			G_GATE_DEBUG(1, "Invalid read offset or media size.");
575 			error = EINVAL;
576 			goto fail2;
577 		}
578 	}
579 
580 	gp = g_new_geomf(&g_gate_class, "%s", name);
581 	gp->start = g_gate_start;
582 	gp->access = g_gate_access;
583 	gp->orphan = g_gate_orphan;
584 	gp->dumpconf = g_gate_dumpconf;
585 	gp->softc = sc;
586 
587 	if (ropp != NULL) {
588 		cp = g_new_consumer(gp);
589 		cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
590 		error = g_attach(cp, ropp);
591 		if (error != 0) {
592 			G_GATE_DEBUG(1, "Unable to attach to %s.", ropp->name);
593 			goto fail3;
594 		}
595 		error = g_access(cp, 1, 0, 0);
596 		if (error != 0) {
597 			G_GATE_DEBUG(1, "Unable to access %s.", ropp->name);
598 			g_detach(cp);
599 			goto fail3;
600 		}
601 		sc->sc_readcons = cp;
602 		sc->sc_readoffset = ggio->gctl_readoffset;
603 	}
604 
605 	ggio->gctl_unit = sc->sc_unit;
606 
607 	pp = g_new_providerf(gp, "%s", name);
608 	pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
609 	pp->mediasize = ggio->gctl_mediasize;
610 	pp->sectorsize = ggio->gctl_sectorsize;
611 	sc->sc_provider = pp;
612 	g_error_provider(pp, 0);
613 
614 	g_topology_unlock();
615 	mtx_lock(&g_gate_units_lock);
616 	sc->sc_name = sc->sc_provider->name;
617 	mtx_unlock(&g_gate_units_lock);
618 	G_GATE_DEBUG(1, "Device %s created.", gp->name);
619 
620 	if (sc->sc_timeout > 0) {
621 		callout_reset(&sc->sc_callout, sc->sc_timeout * hz,
622 		    g_gate_guard, sc);
623 	}
624 	return (0);
625 fail3:
626 	g_destroy_consumer(cp);
627 	g_destroy_geom(gp);
628 fail2:
629 	g_topology_unlock();
630 	mtx_lock(&g_gate_units_lock);
631 	g_gate_units[sc->sc_unit] = NULL;
632 	KASSERT(g_gate_nunits > 0, ("negative g_gate_nunits?"));
633 	g_gate_nunits--;
634 fail1:
635 	mtx_unlock(&g_gate_units_lock);
636 	mtx_destroy(&sc->sc_queue_mtx);
637 	mtx_destroy(&sc->sc_read_mtx);
638 	free(sc, M_GATE);
639 	return (error);
640 }
641 
642 static int
643 g_gate_modify(struct g_gate_softc *sc, struct g_gate_ctl_modify *ggio)
644 {
645 	char readprov[NAME_MAX + 1];
646 	struct g_provider *pp;
647 	struct g_consumer *cp;
648 	int done, error;
649 
650 	if ((ggio->gctl_modify & GG_MODIFY_MEDIASIZE) != 0) {
651 		if (ggio->gctl_mediasize <= 0) {
652 			G_GATE_DEBUG(1, "Invalid media size.");
653 			return (EINVAL);
654 		}
655 		pp = sc->sc_provider;
656 		if ((ggio->gctl_mediasize % pp->sectorsize) != 0) {
657 			G_GATE_DEBUG(1, "Invalid media size.");
658 			return (EINVAL);
659 		}
660 		g_resize_provider(pp, ggio->gctl_mediasize);
661 		return (0);
662 	}
663 
664 	if ((ggio->gctl_modify & GG_MODIFY_INFO) != 0) {
665 		memset(sc->sc_info, 0, sizeof(sc->sc_info));
666 		strncpy(sc->sc_info, ggio->gctl_info,
667 		    MIN(sizeof(sc->sc_info) - 1, sizeof(ggio->gctl_info)));
668 	}
669 	cp = NULL;
670 
671 	if ((ggio->gctl_modify & GG_MODIFY_READPROV) != 0) {
672 		g_topology_lock();
673 		mtx_lock(&sc->sc_read_mtx);
674 		if ((cp = sc->sc_readcons) != NULL) {
675 			sc->sc_readcons = NULL;
676 			done = (cp->index == 0);
677 			mtx_unlock(&sc->sc_read_mtx);
678 			if (done)
679 				g_gate_detach(cp, 0);
680 		} else
681 			mtx_unlock(&sc->sc_read_mtx);
682 		if (ggio->gctl_readprov[0] != '\0') {
683 			memset(readprov, 0, sizeof(readprov));
684 			strncpy(readprov, ggio->gctl_readprov,
685 			    MIN(sizeof(readprov) - 1,
686 			    sizeof(ggio->gctl_readprov)));
687 			pp = g_provider_by_name(readprov);
688 			if (pp == NULL) {
689 				g_topology_unlock();
690 				G_GATE_DEBUG(1, "Provider %s doesn't exist.",
691 				    readprov);
692 				return (EINVAL);
693 			}
694 			cp = g_new_consumer(sc->sc_provider->geom);
695 			cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
696 			error = g_attach(cp, pp);
697 			if (error != 0) {
698 				G_GATE_DEBUG(1, "Unable to attach to %s.",
699 				    pp->name);
700 			} else {
701 				error = g_access(cp, 1, 0, 0);
702 				if (error != 0) {
703 					G_GATE_DEBUG(1, "Unable to access %s.",
704 					    pp->name);
705 					g_detach(cp);
706 				}
707 			}
708 			if (error != 0) {
709 				g_destroy_consumer(cp);
710 				g_topology_unlock();
711 				return (error);
712 			}
713 		}
714 	} else {
715 		cp = sc->sc_readcons;
716 	}
717 
718 	if ((ggio->gctl_modify & GG_MODIFY_READOFFSET) != 0) {
719 		if (cp == NULL) {
720 			G_GATE_DEBUG(1, "No read provider.");
721 			return (EINVAL);
722 		}
723 		pp = sc->sc_provider;
724 		if ((ggio->gctl_readoffset % pp->sectorsize) != 0) {
725 			G_GATE_DEBUG(1, "Invalid read offset.");
726 			return (EINVAL);
727 		}
728 		if (pp->mediasize + ggio->gctl_readoffset >
729 		    cp->provider->mediasize) {
730 			G_GATE_DEBUG(1, "Invalid read offset or media size.");
731 			return (EINVAL);
732 		}
733 		sc->sc_readoffset = ggio->gctl_readoffset;
734 	}
735 
736 	if ((ggio->gctl_modify & GG_MODIFY_READPROV) != 0) {
737 		sc->sc_readcons = cp;
738 		g_topology_unlock();
739 	}
740 
741 	return (0);
742 }
743 
744 #define	G_GATE_CHECK_VERSION(ggio)	do {				\
745 	if ((ggio)->gctl_version != G_GATE_VERSION) {			\
746 		printf("Version mismatch %d != %d.\n",			\
747 		    ggio->gctl_version, G_GATE_VERSION);		\
748 		return (EINVAL);					\
749 	}								\
750 } while (0)
751 static int
752 g_gate_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td)
753 {
754 	struct g_gate_softc *sc;
755 	struct bio *bp;
756 	int error = 0;
757 
758 	G_GATE_DEBUG(4, "ioctl(%s, %lx, %p, %x, %p)", devtoname(dev), cmd, addr,
759 	    flags, td);
760 
761 	switch (cmd) {
762 	case G_GATE_CMD_CREATE:
763 	    {
764 		struct g_gate_ctl_create *ggio = (void *)addr;
765 
766 		G_GATE_CHECK_VERSION(ggio);
767 		error = g_gate_create(ggio);
768 		/*
769 		 * Reset TDP_GEOM flag.
770 		 * There are pending events for sure, because we just created
771 		 * new provider and other classes want to taste it, but we
772 		 * cannot answer on I/O requests until we're here.
773 		 */
774 		td->td_pflags &= ~TDP_GEOM;
775 		return (error);
776 	    }
777 	case G_GATE_CMD_MODIFY:
778 	    {
779 		struct g_gate_ctl_modify *ggio = (void *)addr;
780 
781 		G_GATE_CHECK_VERSION(ggio);
782 		sc = g_gate_hold(ggio->gctl_unit, NULL);
783 		if (sc == NULL)
784 			return (ENXIO);
785 		error = g_gate_modify(sc, ggio);
786 		g_gate_release(sc);
787 		return (error);
788 	    }
789 	case G_GATE_CMD_DESTROY:
790 	    {
791 		struct g_gate_ctl_destroy *ggio = (void *)addr;
792 
793 		G_GATE_CHECK_VERSION(ggio);
794 		sc = g_gate_hold(ggio->gctl_unit, ggio->gctl_name);
795 		if (sc == NULL)
796 			return (ENXIO);
797 		g_topology_lock();
798 		mtx_lock(&g_gate_units_lock);
799 		error = g_gate_destroy(sc, ggio->gctl_force);
800 		g_topology_unlock();
801 		if (error != 0)
802 			g_gate_release(sc);
803 		return (error);
804 	    }
805 	case G_GATE_CMD_CANCEL:
806 	    {
807 		struct g_gate_ctl_cancel *ggio = (void *)addr;
808 		struct bio *tbp, *lbp;
809 
810 		G_GATE_CHECK_VERSION(ggio);
811 		sc = g_gate_hold(ggio->gctl_unit, ggio->gctl_name);
812 		if (sc == NULL)
813 			return (ENXIO);
814 		lbp = NULL;
815 		mtx_lock(&sc->sc_queue_mtx);
816 		TAILQ_FOREACH_SAFE(bp, &sc->sc_outqueue.queue, bio_queue, tbp) {
817 			if (ggio->gctl_seq == 0 ||
818 			    ggio->gctl_seq == (uintptr_t)bp->bio_driver1) {
819 				G_GATE_LOGREQ(1, bp, "Request canceled.");
820 				bioq_remove(&sc->sc_outqueue, bp);
821 				/*
822 				 * Be sure to put requests back onto incoming
823 				 * queue in the proper order.
824 				 */
825 				if (lbp == NULL)
826 					bioq_insert_head(&sc->sc_inqueue, bp);
827 				else {
828 					TAILQ_INSERT_AFTER(&sc->sc_inqueue.queue,
829 					    lbp, bp, bio_queue);
830 				}
831 				lbp = bp;
832 				/*
833 				 * If only one request was canceled, leave now.
834 				 */
835 				if (ggio->gctl_seq != 0)
836 					break;
837 			}
838 		}
839 		if (ggio->gctl_unit == G_GATE_NAME_GIVEN)
840 			ggio->gctl_unit = sc->sc_unit;
841 		mtx_unlock(&sc->sc_queue_mtx);
842 		g_gate_release(sc);
843 		return (error);
844 	    }
845 	case G_GATE_CMD_START:
846 	    {
847 		struct g_gate_ctl_io *ggio = (void *)addr;
848 
849 		G_GATE_CHECK_VERSION(ggio);
850 		sc = g_gate_hold(ggio->gctl_unit, NULL);
851 		if (sc == NULL)
852 			return (ENXIO);
853 		error = 0;
854 		for (;;) {
855 			mtx_lock(&sc->sc_queue_mtx);
856 			bp = bioq_first(&sc->sc_inqueue);
857 			if (bp != NULL)
858 				break;
859 			if ((sc->sc_flags & G_GATE_FLAG_DESTROY) != 0) {
860 				ggio->gctl_error = ECANCELED;
861 				mtx_unlock(&sc->sc_queue_mtx);
862 				goto start_end;
863 			}
864 			error = msleep(sc, &sc->sc_queue_mtx,
865 				PPAUSE | PDROP | PCATCH, "ggwait", 0);
866 			if (error != 0)
867 				goto start_end;
868 		}
869 		ggio->gctl_cmd = bp->bio_cmd;
870 		if (bp->bio_cmd == BIO_WRITE &&
871 		    bp->bio_length > ggio->gctl_length) {
872 			mtx_unlock(&sc->sc_queue_mtx);
873 			ggio->gctl_length = bp->bio_length;
874 			ggio->gctl_error = ENOMEM;
875 			goto start_end;
876 		}
877 		bioq_remove(&sc->sc_inqueue, bp);
878 		bioq_insert_tail(&sc->sc_outqueue, bp);
879 		mtx_unlock(&sc->sc_queue_mtx);
880 
881 		ggio->gctl_seq = (uintptr_t)bp->bio_driver1;
882 		ggio->gctl_offset = bp->bio_offset;
883 		ggio->gctl_length = bp->bio_length;
884 
885 		switch (bp->bio_cmd) {
886 		case BIO_READ:
887 		case BIO_DELETE:
888 		case BIO_FLUSH:
889 		case BIO_SPEEDUP:
890 			break;
891 		case BIO_WRITE:
892 			error = copyout(bp->bio_data, ggio->gctl_data,
893 			    bp->bio_length);
894 			if (error != 0) {
895 				mtx_lock(&sc->sc_queue_mtx);
896 				bioq_remove(&sc->sc_outqueue, bp);
897 				bioq_insert_head(&sc->sc_inqueue, bp);
898 				mtx_unlock(&sc->sc_queue_mtx);
899 				goto start_end;
900 			}
901 			break;
902 		}
903 start_end:
904 		g_gate_release(sc);
905 		return (error);
906 	    }
907 	case G_GATE_CMD_DONE:
908 	    {
909 		struct g_gate_ctl_io *ggio = (void *)addr;
910 
911 		G_GATE_CHECK_VERSION(ggio);
912 		sc = g_gate_hold(ggio->gctl_unit, NULL);
913 		if (sc == NULL)
914 			return (ENOENT);
915 		error = 0;
916 		mtx_lock(&sc->sc_queue_mtx);
917 		TAILQ_FOREACH(bp, &sc->sc_outqueue.queue, bio_queue) {
918 			if (ggio->gctl_seq == (uintptr_t)bp->bio_driver1)
919 				break;
920 		}
921 		if (bp != NULL) {
922 			bioq_remove(&sc->sc_outqueue, bp);
923 			sc->sc_queue_count--;
924 		}
925 		mtx_unlock(&sc->sc_queue_mtx);
926 		if (bp == NULL) {
927 			/*
928 			 * Request was probably canceled.
929 			 */
930 			goto done_end;
931 		}
932 		if (ggio->gctl_error == EAGAIN) {
933 			bp->bio_error = 0;
934 			G_GATE_LOGREQ(1, bp, "Request desisted.");
935 			mtx_lock(&sc->sc_queue_mtx);
936 			sc->sc_queue_count++;
937 			bioq_insert_head(&sc->sc_inqueue, bp);
938 			wakeup(sc);
939 			mtx_unlock(&sc->sc_queue_mtx);
940 		} else {
941 			bp->bio_error = ggio->gctl_error;
942 			if (bp->bio_error == 0) {
943 				bp->bio_completed = bp->bio_length;
944 				switch (bp->bio_cmd) {
945 				case BIO_READ:
946 					error = copyin(ggio->gctl_data,
947 					    bp->bio_data, bp->bio_length);
948 					if (error != 0)
949 						bp->bio_error = error;
950 					break;
951 				case BIO_DELETE:
952 				case BIO_WRITE:
953 				case BIO_FLUSH:
954 				case BIO_SPEEDUP:
955 					break;
956 				}
957 			}
958 			G_GATE_LOGREQ(2, bp, "Request done.");
959 			g_io_deliver(bp, bp->bio_error);
960 		}
961 done_end:
962 		g_gate_release(sc);
963 		return (error);
964 	    }
965 	}
966 	return (ENOIOCTL);
967 }
968 
969 static void
970 g_gate_device(void)
971 {
972 
973 	status_dev = make_dev(&g_gate_cdevsw, 0x0, UID_ROOT, GID_WHEEL, 0600,
974 	    G_GATE_CTL_NAME);
975 }
976 
977 static int
978 g_gate_modevent(module_t mod, int type, void *data)
979 {
980 	int error = 0;
981 
982 	switch (type) {
983 	case MOD_LOAD:
984 		mtx_init(&g_gate_units_lock, "gg_units_lock", NULL, MTX_DEF);
985 		g_gate_units = malloc(g_gate_maxunits * sizeof(g_gate_units[0]),
986 		    M_GATE, M_WAITOK | M_ZERO);
987 		g_gate_nunits = 0;
988 		g_gate_device();
989 		break;
990 	case MOD_UNLOAD:
991 		mtx_lock(&g_gate_units_lock);
992 		if (g_gate_nunits > 0) {
993 			mtx_unlock(&g_gate_units_lock);
994 			error = EBUSY;
995 			break;
996 		}
997 		mtx_unlock(&g_gate_units_lock);
998 		mtx_destroy(&g_gate_units_lock);
999 		if (status_dev != NULL)
1000 			destroy_dev(status_dev);
1001 		free(g_gate_units, M_GATE);
1002 		break;
1003 	default:
1004 		return (EOPNOTSUPP);
1005 		break;
1006 	}
1007 
1008 	return (error);
1009 }
1010 static moduledata_t g_gate_module = {
1011 	G_GATE_MOD_NAME,
1012 	g_gate_modevent,
1013 	NULL
1014 };
1015 DECLARE_MODULE(geom_gate, g_gate_module, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
1016 DECLARE_GEOM_CLASS(g_gate_class, g_gate);
1017 MODULE_VERSION(geom_gate, 0);
1018