xref: /freebsd/sys/geom/gate/g_gate.c (revision f126890ac5386406dadf7c4cfa9566cbb56537c5)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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/param.h>
34 #include <sys/systm.h>
35 #include <sys/bio.h>
36 #include <sys/conf.h>
37 #include <sys/kernel.h>
38 #include <sys/kthread.h>
39 #include <sys/fcntl.h>
40 #include <sys/linker.h>
41 #include <sys/lock.h>
42 #include <sys/malloc.h>
43 #include <sys/mutex.h>
44 #include <sys/proc.h>
45 #include <sys/limits.h>
46 #include <sys/queue.h>
47 #include <sys/sbuf.h>
48 #include <sys/sysctl.h>
49 #include <sys/signalvar.h>
50 #include <sys/time.h>
51 #include <machine/atomic.h>
52 
53 #include <geom/geom.h>
54 #include <geom/geom_dbg.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 | CTLFLAG_MPSAFE, 0,
63     "GEOM_GATE configuration");
64 static int g_gate_debug = 0;
65 SYSCTL_INT(_kern_geom_gate, OID_AUTO, debug, CTLFLAG_RWTUN, &g_gate_debug, 0,
66     "Debug level");
67 static u_int g_gate_maxunits = 256;
68 SYSCTL_UINT(_kern_geom_gate, OID_AUTO, maxunits, CTLFLAG_RDTUN,
69     &g_gate_maxunits, 0, "Maximum number of ggate devices");
70 
71 struct g_class g_gate_class = {
72 	.name = G_GATE_CLASS_NAME,
73 	.version = G_VERSION,
74 };
75 
76 static struct cdev *status_dev;
77 static d_ioctl_t g_gate_ioctl;
78 static struct cdevsw g_gate_cdevsw = {
79 	.d_version =	D_VERSION,
80 	.d_ioctl =	g_gate_ioctl,
81 	.d_name =	G_GATE_CTL_NAME
82 };
83 
84 static struct g_gate_softc **g_gate_units;
85 static u_int g_gate_nunits;
86 static struct mtx g_gate_units_lock;
87 
88 static void
89 g_gate_detach(void *arg, int flags __unused)
90 {
91 	struct g_consumer *cp = arg;
92 
93 	g_topology_assert();
94 	G_GATE_DEBUG(1, "Destroying read consumer on provider %s orphan.",
95 	    cp->provider->name);
96 	(void)g_access(cp, -1, 0, 0);
97 	g_detach(cp);
98 	g_destroy_consumer(cp);
99 }
100 
101 static int
102 g_gate_destroy(struct g_gate_softc *sc, boolean_t force)
103 {
104 	struct bio_queue_head queue;
105 	struct g_provider *pp;
106 	struct g_consumer *cp;
107 	struct g_geom *gp;
108 	struct bio *bp;
109 
110 	g_topology_assert();
111 	mtx_assert(&g_gate_units_lock, MA_OWNED);
112 	pp = sc->sc_provider;
113 	if (!force && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
114 		mtx_unlock(&g_gate_units_lock);
115 		return (EBUSY);
116 	}
117 	mtx_unlock(&g_gate_units_lock);
118 	mtx_lock(&sc->sc_queue_mtx);
119 	if ((sc->sc_flags & G_GATE_FLAG_DESTROY) == 0)
120 		sc->sc_flags |= G_GATE_FLAG_DESTROY;
121 	wakeup(sc);
122 	mtx_unlock(&sc->sc_queue_mtx);
123 	gp = pp->geom;
124 	g_wither_provider(pp, ENXIO);
125 	callout_drain(&sc->sc_callout);
126 	bioq_init(&queue);
127 	mtx_lock(&sc->sc_queue_mtx);
128 	while ((bp = bioq_takefirst(&sc->sc_inqueue)) != NULL) {
129 		sc->sc_queue_count--;
130 		bioq_insert_tail(&queue, bp);
131 	}
132 	while ((bp = bioq_takefirst(&sc->sc_outqueue)) != NULL) {
133 		sc->sc_queue_count--;
134 		bioq_insert_tail(&queue, bp);
135 	}
136 	mtx_unlock(&sc->sc_queue_mtx);
137 	g_topology_unlock();
138 	while ((bp = bioq_takefirst(&queue)) != NULL) {
139 		G_GATE_LOGREQ(1, bp, "Request canceled.");
140 		g_io_deliver(bp, ENXIO);
141 	}
142 	mtx_lock(&g_gate_units_lock);
143 	/* One reference is ours. */
144 	sc->sc_ref--;
145 	while (sc->sc_ref > 0)
146 		msleep(&sc->sc_ref, &g_gate_units_lock, 0, "gg:destroy", 0);
147 	g_gate_units[sc->sc_unit] = NULL;
148 	KASSERT(g_gate_nunits > 0, ("negative g_gate_nunits?"));
149 	g_gate_nunits--;
150 	mtx_unlock(&g_gate_units_lock);
151 	mtx_destroy(&sc->sc_queue_mtx);
152 	mtx_destroy(&sc->sc_read_mtx);
153 	g_topology_lock();
154 	if ((cp = sc->sc_readcons) != NULL) {
155 		sc->sc_readcons = NULL;
156 		(void)g_access(cp, -1, 0, 0);
157 		g_detach(cp);
158 		g_destroy_consumer(cp);
159 	}
160 	G_GATE_DEBUG(1, "Device %s destroyed.", gp->name);
161 	gp->softc = NULL;
162 	g_wither_geom(gp, ENXIO);
163 	sc->sc_provider = NULL;
164 	free(sc, M_GATE);
165 	return (0);
166 }
167 
168 static int
169 g_gate_access(struct g_provider *pp, int dr, int dw, int de)
170 {
171 	struct g_gate_softc *sc;
172 
173 	if (dr <= 0 && dw <= 0 && de <= 0)
174 		return (0);
175 	sc = pp->geom->softc;
176 	if (sc == NULL || (sc->sc_flags & G_GATE_FLAG_DESTROY) != 0)
177 		return (ENXIO);
178 	/* XXX: Hack to allow read-only mounts. */
179 #if 0
180 	if ((sc->sc_flags & G_GATE_FLAG_READONLY) != 0 && dw > 0)
181 		return (EPERM);
182 #endif
183 	if ((sc->sc_flags & G_GATE_FLAG_WRITEONLY) != 0 && dr > 0)
184 		return (EPERM);
185 	return (0);
186 }
187 
188 static void
189 g_gate_queue_io(struct bio *bp)
190 {
191 	struct g_gate_softc *sc;
192 
193 	sc = bp->bio_to->geom->softc;
194 	if (sc == NULL || (sc->sc_flags & G_GATE_FLAG_DESTROY) != 0) {
195 		g_io_deliver(bp, ENXIO);
196 		return;
197 	}
198 
199 	mtx_lock(&sc->sc_queue_mtx);
200 
201 	if (sc->sc_queue_size > 0 && sc->sc_queue_count > sc->sc_queue_size) {
202 		mtx_unlock(&sc->sc_queue_mtx);
203 		G_GATE_LOGREQ(1, bp, "Queue full, request canceled.");
204 		g_io_deliver(bp, ENOMEM);
205 		return;
206 	}
207 
208 	bp->bio_driver1 = (void *)sc->sc_seq;
209 	sc->sc_seq++;
210 	sc->sc_queue_count++;
211 
212 	bioq_insert_tail(&sc->sc_inqueue, bp);
213 	wakeup(sc);
214 
215 	mtx_unlock(&sc->sc_queue_mtx);
216 }
217 
218 static void
219 g_gate_done(struct bio *cbp)
220 {
221 	struct g_gate_softc *sc;
222 	struct bio *pbp;
223 	struct g_consumer *cp;
224 
225 	cp = cbp->bio_from;
226 	pbp = cbp->bio_parent;
227 	if (cbp->bio_error == 0) {
228 		pbp->bio_completed = cbp->bio_completed;
229 		g_destroy_bio(cbp);
230 		pbp->bio_inbed++;
231 		g_io_deliver(pbp, 0);
232 	} else {
233 		/* If direct read failed, pass it through userland daemon. */
234 		g_destroy_bio(cbp);
235 		pbp->bio_children--;
236 		g_gate_queue_io(pbp);
237 	}
238 
239 	sc = cp->geom->softc;
240 	mtx_lock(&sc->sc_read_mtx);
241 	if (--cp->index == 0 && sc->sc_readcons != cp)
242 		g_post_event(g_gate_detach, cp, M_NOWAIT, NULL);
243 	mtx_unlock(&sc->sc_read_mtx);
244 }
245 
246 static void
247 g_gate_start(struct bio *pbp)
248 {
249 	struct g_gate_softc *sc;
250 	struct g_consumer *cp;
251 	struct bio *cbp;
252 
253 	sc = pbp->bio_to->geom->softc;
254 	if (sc == NULL || (sc->sc_flags & G_GATE_FLAG_DESTROY) != 0) {
255 		g_io_deliver(pbp, ENXIO);
256 		return;
257 	}
258 	G_GATE_LOGREQ(2, pbp, "Request received.");
259 	switch (pbp->bio_cmd) {
260 	case BIO_READ:
261 		if (sc->sc_readcons == NULL)
262 			break;
263 		cbp = g_clone_bio(pbp);
264 		if (cbp == NULL) {
265 			g_io_deliver(pbp, ENOMEM);
266 			return;
267 		}
268 		mtx_lock(&sc->sc_read_mtx);
269 		if ((cp = sc->sc_readcons) == NULL) {
270 			mtx_unlock(&sc->sc_read_mtx);
271 			g_destroy_bio(cbp);
272 			pbp->bio_children--;
273 			break;
274 		}
275 		cp->index++;
276 		cbp->bio_offset = pbp->bio_offset + sc->sc_readoffset;
277 		mtx_unlock(&sc->sc_read_mtx);
278 		cbp->bio_done = g_gate_done;
279 		g_io_request(cbp, cp);
280 		return;
281 	case BIO_DELETE:
282 	case BIO_WRITE:
283 	case BIO_FLUSH:
284 	case BIO_SPEEDUP:
285 		/* XXX: Hack to allow read-only mounts. */
286 		if ((sc->sc_flags & G_GATE_FLAG_READONLY) != 0) {
287 			g_io_deliver(pbp, EPERM);
288 			return;
289 		}
290 		break;
291 	case BIO_GETATTR:
292 	default:
293 		G_GATE_LOGREQ(2, pbp, "Ignoring request.");
294 		g_io_deliver(pbp, EOPNOTSUPP);
295 		return;
296 	}
297 
298 	g_gate_queue_io(pbp);
299 }
300 
301 static struct g_gate_softc *
302 g_gate_hold(int unit, const char *name)
303 {
304 	struct g_gate_softc *sc = NULL;
305 
306 	mtx_lock(&g_gate_units_lock);
307 	if (unit >= 0 && unit < g_gate_maxunits)
308 		sc = g_gate_units[unit];
309 	else if (unit == G_GATE_NAME_GIVEN) {
310 		KASSERT(name != NULL, ("name is NULL"));
311 		for (unit = 0; unit < g_gate_maxunits; unit++) {
312 			if (g_gate_units[unit] == NULL)
313 				continue;
314 			if (strcmp(name,
315 			    g_gate_units[unit]->sc_provider->name) != 0) {
316 				continue;
317 			}
318 			sc = g_gate_units[unit];
319 			break;
320 		}
321 	}
322 	if (sc != NULL)
323 		sc->sc_ref++;
324 	mtx_unlock(&g_gate_units_lock);
325 	return (sc);
326 }
327 
328 static void
329 g_gate_release(struct g_gate_softc *sc)
330 {
331 
332 	g_topology_assert_not();
333 	mtx_lock(&g_gate_units_lock);
334 	sc->sc_ref--;
335 	KASSERT(sc->sc_ref >= 0, ("Negative sc_ref for %s.", sc->sc_name));
336 	if (sc->sc_ref == 0 && (sc->sc_flags & G_GATE_FLAG_DESTROY) != 0)
337 		wakeup(&sc->sc_ref);
338 	mtx_unlock(&g_gate_units_lock);
339 }
340 
341 static int
342 g_gate_getunit(int unit, int *errorp)
343 {
344 
345 	mtx_assert(&g_gate_units_lock, MA_OWNED);
346 	if (unit >= 0) {
347 		if (unit >= g_gate_maxunits)
348 			*errorp = EINVAL;
349 		else if (g_gate_units[unit] == NULL)
350 			return (unit);
351 		else
352 			*errorp = EEXIST;
353 	} else {
354 		for (unit = 0; unit < g_gate_maxunits; unit++) {
355 			if (g_gate_units[unit] == NULL)
356 				return (unit);
357 		}
358 		*errorp = ENFILE;
359 	}
360 	return (-1);
361 }
362 
363 static void
364 g_gate_guard(void *arg)
365 {
366 	struct bio_queue_head queue;
367 	struct g_gate_softc *sc;
368 	struct bintime curtime;
369 	struct bio *bp, *bp2;
370 
371 	sc = arg;
372 	binuptime(&curtime);
373 	g_gate_hold(sc->sc_unit, NULL);
374 	bioq_init(&queue);
375 	mtx_lock(&sc->sc_queue_mtx);
376 	TAILQ_FOREACH_SAFE(bp, &sc->sc_inqueue.queue, bio_queue, bp2) {
377 		if (curtime.sec - bp->bio_t0.sec < 5)
378 			continue;
379 		bioq_remove(&sc->sc_inqueue, bp);
380 		sc->sc_queue_count--;
381 		bioq_insert_tail(&queue, bp);
382 	}
383 	TAILQ_FOREACH_SAFE(bp, &sc->sc_outqueue.queue, bio_queue, bp2) {
384 		if (curtime.sec - bp->bio_t0.sec < 5)
385 			continue;
386 		bioq_remove(&sc->sc_outqueue, bp);
387 		sc->sc_queue_count--;
388 		bioq_insert_tail(&queue, bp);
389 	}
390 	mtx_unlock(&sc->sc_queue_mtx);
391 	while ((bp = bioq_takefirst(&queue)) != NULL) {
392 		G_GATE_LOGREQ(1, bp, "Request timeout.");
393 		g_io_deliver(bp, EIO);
394 	}
395 	if ((sc->sc_flags & G_GATE_FLAG_DESTROY) == 0) {
396 		callout_reset(&sc->sc_callout, sc->sc_timeout * hz,
397 		    g_gate_guard, sc);
398 	}
399 	g_gate_release(sc);
400 }
401 
402 static void
403 g_gate_orphan(struct g_consumer *cp)
404 {
405 	struct g_gate_softc *sc;
406 	struct g_geom *gp;
407 	int done;
408 
409 	g_topology_assert();
410 	gp = cp->geom;
411 	sc = gp->softc;
412 	mtx_lock(&sc->sc_read_mtx);
413 	if (sc->sc_readcons == cp)
414 		sc->sc_readcons = NULL;
415 	done = (cp->index == 0);
416 	mtx_unlock(&sc->sc_read_mtx);
417 	if (done)
418 		g_gate_detach(cp, 0);
419 }
420 
421 static void
422 g_gate_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
423     struct g_consumer *cp, struct g_provider *pp)
424 {
425 	struct g_gate_softc *sc;
426 
427 	sc = gp->softc;
428 	if (sc == NULL || pp != NULL || cp != NULL)
429 		return;
430 	sc = g_gate_hold(sc->sc_unit, NULL);
431 	if (sc == NULL)
432 		return;
433 	if ((sc->sc_flags & G_GATE_FLAG_READONLY) != 0) {
434 		sbuf_printf(sb, "%s<access>%s</access>\n", indent, "read-only");
435 	} else if ((sc->sc_flags & G_GATE_FLAG_WRITEONLY) != 0) {
436 		sbuf_printf(sb, "%s<access>%s</access>\n", indent,
437 		    "write-only");
438 	} else {
439 		sbuf_printf(sb, "%s<access>%s</access>\n", indent,
440 		    "read-write");
441 	}
442 	if (sc->sc_readcons != NULL) {
443 		sbuf_printf(sb, "%s<read_offset>%jd</read_offset>\n",
444 		    indent, (intmax_t)sc->sc_readoffset);
445 		sbuf_printf(sb, "%s<read_provider>%s</read_provider>\n",
446 		    indent, sc->sc_readcons->provider->name);
447 	}
448 	sbuf_printf(sb, "%s<timeout>%u</timeout>\n", indent, sc->sc_timeout);
449 	sbuf_printf(sb, "%s<info>%s</info>\n", indent, sc->sc_info);
450 	sbuf_printf(sb, "%s<queue_count>%u</queue_count>\n", indent,
451 	    sc->sc_queue_count);
452 	sbuf_printf(sb, "%s<queue_size>%u</queue_size>\n", indent,
453 	    sc->sc_queue_size);
454 	sbuf_printf(sb, "%s<ref>%u</ref>\n", indent, sc->sc_ref);
455 	sbuf_printf(sb, "%s<unit>%d</unit>\n", indent, sc->sc_unit);
456 	g_topology_unlock();
457 	g_gate_release(sc);
458 	g_topology_lock();
459 }
460 
461 static int
462 g_gate_create(struct g_gate_ctl_create *ggio)
463 {
464 	struct g_gate_softc *sc;
465 	struct g_geom *gp;
466 	struct g_provider *pp, *ropp;
467 	struct g_consumer *cp;
468 	char name[NAME_MAX + 1];
469 	char readprov[NAME_MAX + 1];
470 	int error = 0, unit;
471 
472 	if (ggio->gctl_mediasize <= 0) {
473 		G_GATE_DEBUG(1, "Invalid media size.");
474 		return (EINVAL);
475 	}
476 	if (ggio->gctl_sectorsize <= 0) {
477 		G_GATE_DEBUG(1, "Invalid sector size.");
478 		return (EINVAL);
479 	}
480 	if (!powerof2(ggio->gctl_sectorsize)) {
481 		G_GATE_DEBUG(1, "Invalid sector size.");
482 		return (EINVAL);
483 	}
484 	if ((ggio->gctl_mediasize % ggio->gctl_sectorsize) != 0) {
485 		G_GATE_DEBUG(1, "Invalid media size.");
486 		return (EINVAL);
487 	}
488 	if ((ggio->gctl_flags & G_GATE_FLAG_READONLY) != 0 &&
489 	    (ggio->gctl_flags & G_GATE_FLAG_WRITEONLY) != 0) {
490 		G_GATE_DEBUG(1, "Invalid flags.");
491 		return (EINVAL);
492 	}
493 	if (ggio->gctl_unit != G_GATE_UNIT_AUTO &&
494 	    ggio->gctl_unit != G_GATE_NAME_GIVEN &&
495 	    ggio->gctl_unit < 0) {
496 		G_GATE_DEBUG(1, "Invalid unit number.");
497 		return (EINVAL);
498 	}
499 	if (ggio->gctl_unit == G_GATE_NAME_GIVEN &&
500 	    ggio->gctl_name[0] == '\0') {
501 		G_GATE_DEBUG(1, "No device name.");
502 		return (EINVAL);
503 	}
504 
505 	sc = malloc(sizeof(*sc), M_GATE, M_WAITOK | M_ZERO);
506 	sc->sc_flags = (ggio->gctl_flags & G_GATE_USERFLAGS);
507 	memset(sc->sc_info, 0, sizeof(sc->sc_info));
508 	strncpy(sc->sc_info, ggio->gctl_info,
509 	    MIN(sizeof(sc->sc_info) - 1, sizeof(ggio->gctl_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 		memset(name, 0, sizeof(name));
528 		strncpy(name, ggio->gctl_name,
529 		    MIN(sizeof(name) - 1, sizeof(ggio->gctl_name)));
530 	} else {
531 		snprintf(name, sizeof(name), "%s%d", G_GATE_PROVIDER_NAME,
532 		    sc->sc_unit);
533 	}
534 	/* Check for name collision. */
535 	for (unit = 0; unit < g_gate_maxunits; unit++) {
536 		if (g_gate_units[unit] == NULL)
537 			continue;
538 		if (strcmp(name, g_gate_units[unit]->sc_name) != 0)
539 			continue;
540 		error = EEXIST;
541 		goto fail1;
542 	}
543 	// local stack buffer 'name' assigned here temporarily only.
544 	// the real provider name is assigned below.
545 	sc->sc_name = name;
546 	g_gate_units[sc->sc_unit] = sc;
547 	g_gate_nunits++;
548 	mtx_unlock(&g_gate_units_lock);
549 
550 	g_topology_lock();
551 
552 	if (ggio->gctl_readprov[0] == '\0') {
553 		ropp = NULL;
554 	} else {
555 		memset(readprov, 0, sizeof(readprov));
556 		strncpy(readprov, ggio->gctl_readprov,
557 		    MIN(sizeof(readprov) - 1, sizeof(ggio->gctl_readprov)));
558 		ropp = g_provider_by_name(readprov);
559 		if (ropp == NULL) {
560 			G_GATE_DEBUG(1, "Provider %s doesn't exist.", readprov);
561 			error = EINVAL;
562 			goto fail2;
563 		}
564 		if ((ggio->gctl_readoffset % ggio->gctl_sectorsize) != 0) {
565 			G_GATE_DEBUG(1, "Invalid read offset.");
566 			error = EINVAL;
567 			goto fail2;
568 		}
569 		if (ggio->gctl_mediasize + ggio->gctl_readoffset >
570 		    ropp->mediasize) {
571 			G_GATE_DEBUG(1, "Invalid read offset or media size.");
572 			error = EINVAL;
573 			goto fail2;
574 		}
575 	}
576 
577 	gp = g_new_geomf(&g_gate_class, "%s", name);
578 	gp->start = g_gate_start;
579 	gp->access = g_gate_access;
580 	gp->orphan = g_gate_orphan;
581 	gp->dumpconf = g_gate_dumpconf;
582 	gp->softc = sc;
583 
584 	if (ropp != NULL) {
585 		cp = g_new_consumer(gp);
586 		cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
587 		error = g_attach(cp, ropp);
588 		if (error != 0) {
589 			G_GATE_DEBUG(1, "Unable to attach to %s.", ropp->name);
590 			goto fail3;
591 		}
592 		error = g_access(cp, 1, 0, 0);
593 		if (error != 0) {
594 			G_GATE_DEBUG(1, "Unable to access %s.", ropp->name);
595 			g_detach(cp);
596 			goto fail3;
597 		}
598 		sc->sc_readcons = cp;
599 		sc->sc_readoffset = ggio->gctl_readoffset;
600 	}
601 
602 	ggio->gctl_unit = sc->sc_unit;
603 
604 	pp = g_new_providerf(gp, "%s", name);
605 	pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
606 	pp->mediasize = ggio->gctl_mediasize;
607 	pp->sectorsize = ggio->gctl_sectorsize;
608 	sc->sc_provider = pp;
609 	g_error_provider(pp, 0);
610 
611 	g_topology_unlock();
612 	mtx_lock(&g_gate_units_lock);
613 	sc->sc_name = sc->sc_provider->name;
614 	mtx_unlock(&g_gate_units_lock);
615 	G_GATE_DEBUG(1, "Device %s created.", gp->name);
616 
617 	if (sc->sc_timeout > 0) {
618 		callout_reset(&sc->sc_callout, sc->sc_timeout * hz,
619 		    g_gate_guard, sc);
620 	}
621 	return (0);
622 fail3:
623 	g_destroy_consumer(cp);
624 	g_destroy_geom(gp);
625 fail2:
626 	g_topology_unlock();
627 	mtx_lock(&g_gate_units_lock);
628 	g_gate_units[sc->sc_unit] = NULL;
629 	KASSERT(g_gate_nunits > 0, ("negative g_gate_nunits?"));
630 	g_gate_nunits--;
631 fail1:
632 	mtx_unlock(&g_gate_units_lock);
633 	mtx_destroy(&sc->sc_queue_mtx);
634 	mtx_destroy(&sc->sc_read_mtx);
635 	free(sc, M_GATE);
636 	return (error);
637 }
638 
639 static int
640 g_gate_modify(struct g_gate_softc *sc, struct g_gate_ctl_modify *ggio)
641 {
642 	char readprov[NAME_MAX + 1];
643 	struct g_provider *pp;
644 	struct g_consumer *cp;
645 	int done, error;
646 
647 	if ((ggio->gctl_modify & GG_MODIFY_MEDIASIZE) != 0) {
648 		if (ggio->gctl_mediasize <= 0) {
649 			G_GATE_DEBUG(1, "Invalid media size.");
650 			return (EINVAL);
651 		}
652 		pp = sc->sc_provider;
653 		if ((ggio->gctl_mediasize % pp->sectorsize) != 0) {
654 			G_GATE_DEBUG(1, "Invalid media size.");
655 			return (EINVAL);
656 		}
657 		g_resize_provider(pp, ggio->gctl_mediasize);
658 		return (0);
659 	}
660 
661 	if ((ggio->gctl_modify & GG_MODIFY_INFO) != 0) {
662 		memset(sc->sc_info, 0, sizeof(sc->sc_info));
663 		strncpy(sc->sc_info, ggio->gctl_info,
664 		    MIN(sizeof(sc->sc_info) - 1, sizeof(ggio->gctl_info)));
665 	}
666 	cp = NULL;
667 
668 	if ((ggio->gctl_modify & GG_MODIFY_READPROV) != 0) {
669 		g_topology_lock();
670 		mtx_lock(&sc->sc_read_mtx);
671 		if ((cp = sc->sc_readcons) != NULL) {
672 			sc->sc_readcons = NULL;
673 			done = (cp->index == 0);
674 			mtx_unlock(&sc->sc_read_mtx);
675 			if (done)
676 				g_gate_detach(cp, 0);
677 		} else
678 			mtx_unlock(&sc->sc_read_mtx);
679 		if (ggio->gctl_readprov[0] != '\0') {
680 			memset(readprov, 0, sizeof(readprov));
681 			strncpy(readprov, ggio->gctl_readprov,
682 			    MIN(sizeof(readprov) - 1,
683 			    sizeof(ggio->gctl_readprov)));
684 			pp = g_provider_by_name(readprov);
685 			if (pp == NULL) {
686 				g_topology_unlock();
687 				G_GATE_DEBUG(1, "Provider %s doesn't exist.",
688 				    readprov);
689 				return (EINVAL);
690 			}
691 			cp = g_new_consumer(sc->sc_provider->geom);
692 			cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
693 			error = g_attach(cp, pp);
694 			if (error != 0) {
695 				G_GATE_DEBUG(1, "Unable to attach to %s.",
696 				    pp->name);
697 			} else {
698 				error = g_access(cp, 1, 0, 0);
699 				if (error != 0) {
700 					G_GATE_DEBUG(1, "Unable to access %s.",
701 					    pp->name);
702 					g_detach(cp);
703 				}
704 			}
705 			if (error != 0) {
706 				g_destroy_consumer(cp);
707 				g_topology_unlock();
708 				return (error);
709 			}
710 		}
711 	} else {
712 		cp = sc->sc_readcons;
713 	}
714 
715 	if ((ggio->gctl_modify & GG_MODIFY_READOFFSET) != 0) {
716 		if (cp == NULL) {
717 			G_GATE_DEBUG(1, "No read provider.");
718 			return (EINVAL);
719 		}
720 		pp = sc->sc_provider;
721 		if ((ggio->gctl_readoffset % pp->sectorsize) != 0) {
722 			G_GATE_DEBUG(1, "Invalid read offset.");
723 			return (EINVAL);
724 		}
725 		if (pp->mediasize + ggio->gctl_readoffset >
726 		    cp->provider->mediasize) {
727 			G_GATE_DEBUG(1, "Invalid read offset or media size.");
728 			return (EINVAL);
729 		}
730 		sc->sc_readoffset = ggio->gctl_readoffset;
731 	}
732 
733 	if ((ggio->gctl_modify & GG_MODIFY_READPROV) != 0) {
734 		sc->sc_readcons = cp;
735 		g_topology_unlock();
736 	}
737 
738 	return (0);
739 }
740 
741 #define	G_GATE_CHECK_VERSION(ggio)	do {				\
742 	if ((ggio)->gctl_version != G_GATE_VERSION) {			\
743 		printf("Version mismatch %d != %d.\n",			\
744 		    ggio->gctl_version, G_GATE_VERSION);		\
745 		return (EINVAL);					\
746 	}								\
747 } while (0)
748 static int
749 g_gate_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td)
750 {
751 	struct g_gate_softc *sc;
752 	struct bio *bp;
753 	int error = 0;
754 
755 	G_GATE_DEBUG(4, "ioctl(%s, %lx, %p, %x, %p)", devtoname(dev), cmd, addr,
756 	    flags, td);
757 
758 	switch (cmd) {
759 	case G_GATE_CMD_CREATE:
760 	    {
761 		struct g_gate_ctl_create *ggio = (void *)addr;
762 
763 		G_GATE_CHECK_VERSION(ggio);
764 		error = g_gate_create(ggio);
765 		/*
766 		 * Reset TDP_GEOM flag.
767 		 * There are pending events for sure, because we just created
768 		 * new provider and other classes want to taste it, but we
769 		 * cannot answer on I/O requests until we're here.
770 		 */
771 		td->td_pflags &= ~TDP_GEOM;
772 		return (error);
773 	    }
774 	case G_GATE_CMD_MODIFY:
775 	    {
776 		struct g_gate_ctl_modify *ggio = (void *)addr;
777 
778 		G_GATE_CHECK_VERSION(ggio);
779 		sc = g_gate_hold(ggio->gctl_unit, NULL);
780 		if (sc == NULL)
781 			return (ENXIO);
782 		error = g_gate_modify(sc, ggio);
783 		g_gate_release(sc);
784 		return (error);
785 	    }
786 	case G_GATE_CMD_DESTROY:
787 	    {
788 		struct g_gate_ctl_destroy *ggio = (void *)addr;
789 
790 		G_GATE_CHECK_VERSION(ggio);
791 		sc = g_gate_hold(ggio->gctl_unit, ggio->gctl_name);
792 		if (sc == NULL)
793 			return (ENXIO);
794 		g_topology_lock();
795 		mtx_lock(&g_gate_units_lock);
796 		error = g_gate_destroy(sc, ggio->gctl_force);
797 		g_topology_unlock();
798 		if (error != 0)
799 			g_gate_release(sc);
800 		return (error);
801 	    }
802 	case G_GATE_CMD_CANCEL:
803 	    {
804 		struct g_gate_ctl_cancel *ggio = (void *)addr;
805 		struct bio *tbp, *lbp;
806 
807 		G_GATE_CHECK_VERSION(ggio);
808 		sc = g_gate_hold(ggio->gctl_unit, ggio->gctl_name);
809 		if (sc == NULL)
810 			return (ENXIO);
811 		lbp = NULL;
812 		mtx_lock(&sc->sc_queue_mtx);
813 		TAILQ_FOREACH_SAFE(bp, &sc->sc_outqueue.queue, bio_queue, tbp) {
814 			if (ggio->gctl_seq == 0 ||
815 			    ggio->gctl_seq == (uintptr_t)bp->bio_driver1) {
816 				G_GATE_LOGREQ(1, bp, "Request canceled.");
817 				bioq_remove(&sc->sc_outqueue, bp);
818 				/*
819 				 * Be sure to put requests back onto incoming
820 				 * queue in the proper order.
821 				 */
822 				if (lbp == NULL)
823 					bioq_insert_head(&sc->sc_inqueue, bp);
824 				else {
825 					TAILQ_INSERT_AFTER(&sc->sc_inqueue.queue,
826 					    lbp, bp, bio_queue);
827 				}
828 				lbp = bp;
829 				/*
830 				 * If only one request was canceled, leave now.
831 				 */
832 				if (ggio->gctl_seq != 0)
833 					break;
834 			}
835 		}
836 		if (ggio->gctl_unit == G_GATE_NAME_GIVEN)
837 			ggio->gctl_unit = sc->sc_unit;
838 		mtx_unlock(&sc->sc_queue_mtx);
839 		g_gate_release(sc);
840 		return (error);
841 	    }
842 	case G_GATE_CMD_START:
843 	    {
844 		struct g_gate_ctl_io *ggio = (void *)addr;
845 
846 		G_GATE_CHECK_VERSION(ggio);
847 		sc = g_gate_hold(ggio->gctl_unit, NULL);
848 		if (sc == NULL)
849 			return (ENXIO);
850 		error = 0;
851 		for (;;) {
852 			mtx_lock(&sc->sc_queue_mtx);
853 			bp = bioq_first(&sc->sc_inqueue);
854 			if (bp != NULL)
855 				break;
856 			if ((sc->sc_flags & G_GATE_FLAG_DESTROY) != 0) {
857 				ggio->gctl_error = ECANCELED;
858 				mtx_unlock(&sc->sc_queue_mtx);
859 				goto start_end;
860 			}
861 			error = msleep(sc, &sc->sc_queue_mtx,
862 				PPAUSE | PDROP | PCATCH, "ggwait", 0);
863 			if (error != 0)
864 				goto start_end;
865 		}
866 		ggio->gctl_cmd = bp->bio_cmd;
867 		if (bp->bio_cmd == BIO_WRITE &&
868 		    bp->bio_length > ggio->gctl_length) {
869 			mtx_unlock(&sc->sc_queue_mtx);
870 			ggio->gctl_length = bp->bio_length;
871 			ggio->gctl_error = ENOMEM;
872 			goto start_end;
873 		}
874 		bioq_remove(&sc->sc_inqueue, bp);
875 		bioq_insert_tail(&sc->sc_outqueue, bp);
876 		mtx_unlock(&sc->sc_queue_mtx);
877 
878 		ggio->gctl_seq = (uintptr_t)bp->bio_driver1;
879 		ggio->gctl_offset = bp->bio_offset;
880 		ggio->gctl_length = bp->bio_length;
881 
882 		switch (bp->bio_cmd) {
883 		case BIO_READ:
884 		case BIO_DELETE:
885 		case BIO_FLUSH:
886 		case BIO_SPEEDUP:
887 			break;
888 		case BIO_WRITE:
889 			error = copyout(bp->bio_data, ggio->gctl_data,
890 			    bp->bio_length);
891 			if (error != 0) {
892 				mtx_lock(&sc->sc_queue_mtx);
893 				bioq_remove(&sc->sc_outqueue, bp);
894 				bioq_insert_head(&sc->sc_inqueue, bp);
895 				mtx_unlock(&sc->sc_queue_mtx);
896 				goto start_end;
897 			}
898 			break;
899 		}
900 start_end:
901 		g_gate_release(sc);
902 		return (error);
903 	    }
904 	case G_GATE_CMD_DONE:
905 	    {
906 		struct g_gate_ctl_io *ggio = (void *)addr;
907 
908 		G_GATE_CHECK_VERSION(ggio);
909 		sc = g_gate_hold(ggio->gctl_unit, NULL);
910 		if (sc == NULL)
911 			return (ENOENT);
912 		error = 0;
913 		mtx_lock(&sc->sc_queue_mtx);
914 		TAILQ_FOREACH(bp, &sc->sc_outqueue.queue, bio_queue) {
915 			if (ggio->gctl_seq == (uintptr_t)bp->bio_driver1)
916 				break;
917 		}
918 		if (bp != NULL) {
919 			bioq_remove(&sc->sc_outqueue, bp);
920 			sc->sc_queue_count--;
921 		}
922 		mtx_unlock(&sc->sc_queue_mtx);
923 		if (bp == NULL) {
924 			/*
925 			 * Request was probably canceled.
926 			 */
927 			goto done_end;
928 		}
929 		if (ggio->gctl_error == EAGAIN) {
930 			bp->bio_error = 0;
931 			G_GATE_LOGREQ(1, bp, "Request desisted.");
932 			mtx_lock(&sc->sc_queue_mtx);
933 			sc->sc_queue_count++;
934 			bioq_insert_head(&sc->sc_inqueue, bp);
935 			wakeup(sc);
936 			mtx_unlock(&sc->sc_queue_mtx);
937 		} else {
938 			bp->bio_error = ggio->gctl_error;
939 			if (bp->bio_error == 0) {
940 				bp->bio_completed = bp->bio_length;
941 				switch (bp->bio_cmd) {
942 				case BIO_READ:
943 					error = copyin(ggio->gctl_data,
944 					    bp->bio_data, bp->bio_length);
945 					if (error != 0)
946 						bp->bio_error = error;
947 					break;
948 				case BIO_DELETE:
949 				case BIO_WRITE:
950 				case BIO_FLUSH:
951 				case BIO_SPEEDUP:
952 					break;
953 				}
954 			}
955 			G_GATE_LOGREQ(2, bp, "Request done.");
956 			g_io_deliver(bp, bp->bio_error);
957 		}
958 done_end:
959 		g_gate_release(sc);
960 		return (error);
961 	    }
962 	}
963 	return (ENOIOCTL);
964 }
965 
966 static void
967 g_gate_device(void)
968 {
969 
970 	status_dev = make_dev(&g_gate_cdevsw, 0x0, UID_ROOT, GID_WHEEL, 0600,
971 	    G_GATE_CTL_NAME);
972 }
973 
974 static int
975 g_gate_modevent(module_t mod, int type, void *data)
976 {
977 	int error = 0;
978 
979 	switch (type) {
980 	case MOD_LOAD:
981 		mtx_init(&g_gate_units_lock, "gg_units_lock", NULL, MTX_DEF);
982 		g_gate_units = malloc(g_gate_maxunits * sizeof(g_gate_units[0]),
983 		    M_GATE, M_WAITOK | M_ZERO);
984 		g_gate_nunits = 0;
985 		g_gate_device();
986 		break;
987 	case MOD_UNLOAD:
988 		mtx_lock(&g_gate_units_lock);
989 		if (g_gate_nunits > 0) {
990 			mtx_unlock(&g_gate_units_lock);
991 			error = EBUSY;
992 			break;
993 		}
994 		mtx_unlock(&g_gate_units_lock);
995 		mtx_destroy(&g_gate_units_lock);
996 		if (status_dev != NULL)
997 			destroy_dev(status_dev);
998 		free(g_gate_units, M_GATE);
999 		break;
1000 	default:
1001 		return (EOPNOTSUPP);
1002 		break;
1003 	}
1004 
1005 	return (error);
1006 }
1007 static moduledata_t g_gate_module = {
1008 	G_GATE_MOD_NAME,
1009 	g_gate_modevent,
1010 	NULL
1011 };
1012 DECLARE_MODULE(geom_gate, g_gate_module, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
1013 DECLARE_GEOM_CLASS(g_gate_class, g_gate);
1014 MODULE_VERSION(geom_gate, 0);
1015