xref: /freebsd/sys/geom/geom_dev.c (revision f39bffc62c1395bde25d152c7f68fdf7cbaab414)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2002 Poul-Henning Kamp
5  * Copyright (c) 2002 Networks Associates Technology, Inc.
6  * All rights reserved.
7  *
8  * This software was developed for the FreeBSD Project by Poul-Henning Kamp
9  * and NAI Labs, the Security Research Division of Network Associates, Inc.
10  * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
11  * DARPA CHATS research program.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. The names of the authors may not be used to endorse or promote
22  *    products derived from this software without specific prior written
23  *    permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/malloc.h>
44 #include <sys/kernel.h>
45 #include <sys/conf.h>
46 #include <sys/ctype.h>
47 #include <sys/bio.h>
48 #include <sys/bus.h>
49 #include <sys/lock.h>
50 #include <sys/mutex.h>
51 #include <sys/proc.h>
52 #include <sys/errno.h>
53 #include <sys/time.h>
54 #include <sys/disk.h>
55 #include <sys/fcntl.h>
56 #include <sys/limits.h>
57 #include <sys/sysctl.h>
58 #include <geom/geom.h>
59 #include <geom/geom_int.h>
60 #include <machine/stdarg.h>
61 
62 struct g_dev_softc {
63 	struct mtx	 sc_mtx;
64 	struct cdev	*sc_dev;
65 	struct cdev	*sc_alias;
66 	int		 sc_open;
67 	int		 sc_active;
68 };
69 
70 static d_open_t		g_dev_open;
71 static d_close_t	g_dev_close;
72 static d_strategy_t	g_dev_strategy;
73 static d_ioctl_t	g_dev_ioctl;
74 
75 static struct cdevsw g_dev_cdevsw = {
76 	.d_version =	D_VERSION,
77 	.d_open =	g_dev_open,
78 	.d_close =	g_dev_close,
79 	.d_read =	physread,
80 	.d_write =	physwrite,
81 	.d_ioctl =	g_dev_ioctl,
82 	.d_strategy =	g_dev_strategy,
83 	.d_name =	"g_dev",
84 	.d_flags =	D_DISK | D_TRACKCLOSE,
85 };
86 
87 static g_init_t g_dev_init;
88 static g_fini_t g_dev_fini;
89 static g_taste_t g_dev_taste;
90 static g_orphan_t g_dev_orphan;
91 static g_attrchanged_t g_dev_attrchanged;
92 
93 static struct g_class g_dev_class	= {
94 	.name = "DEV",
95 	.version = G_VERSION,
96 	.init = g_dev_init,
97 	.fini = g_dev_fini,
98 	.taste = g_dev_taste,
99 	.orphan = g_dev_orphan,
100 	.attrchanged = g_dev_attrchanged
101 };
102 
103 /*
104  * We target 262144 (8 x 32768) sectors by default as this significantly
105  * increases the throughput on commonly used SSD's with a marginal
106  * increase in non-interruptible request latency.
107  */
108 static uint64_t g_dev_del_max_sectors = 262144;
109 SYSCTL_DECL(_kern_geom);
110 SYSCTL_NODE(_kern_geom, OID_AUTO, dev, CTLFLAG_RW, 0, "GEOM_DEV stuff");
111 SYSCTL_QUAD(_kern_geom_dev, OID_AUTO, delete_max_sectors, CTLFLAG_RW,
112     &g_dev_del_max_sectors, 0, "Maximum number of sectors in a single "
113     "delete request sent to the provider. Larger requests are chunked "
114     "so they can be interrupted. (0 = disable chunking)");
115 
116 static char *dumpdev = NULL;
117 static void
118 g_dev_init(struct g_class *mp)
119 {
120 
121 	dumpdev = kern_getenv("dumpdev");
122 }
123 
124 static void
125 g_dev_fini(struct g_class *mp)
126 {
127 
128 	freeenv(dumpdev);
129 	dumpdev = NULL;
130 }
131 
132 static int
133 g_dev_setdumpdev(struct cdev *dev, struct diocskerneldump_arg *kda,
134     struct thread *td)
135 {
136 	struct g_kerneldump kd;
137 	struct g_consumer *cp;
138 	int error, len;
139 
140 	if (dev == NULL || kda == NULL)
141 		return (clear_dumper(td));
142 
143 	cp = dev->si_drv2;
144 	len = sizeof(kd);
145 	memset(&kd, 0, len);
146 	kd.offset = 0;
147 	kd.length = OFF_MAX;
148 	error = g_io_getattr("GEOM::kerneldump", cp, &len, &kd);
149 	if (error != 0)
150 		return (error);
151 
152 	error = set_dumper(&kd.di, devtoname(dev), td, kda->kda_compression,
153 	    kda->kda_encryption, kda->kda_key, kda->kda_encryptedkeysize,
154 	    kda->kda_encryptedkey);
155 	if (error == 0)
156 		dev->si_flags |= SI_DUMPDEV;
157 
158 	return (error);
159 }
160 
161 static int
162 init_dumpdev(struct cdev *dev)
163 {
164 	struct diocskerneldump_arg kda;
165 	struct g_consumer *cp;
166 	const char *devprefix = "/dev/", *devname;
167 	int error;
168 	size_t len;
169 
170 	bzero(&kda, sizeof(kda));
171 	kda.kda_enable = 1;
172 
173 	if (dumpdev == NULL)
174 		return (0);
175 
176 	len = strlen(devprefix);
177 	devname = devtoname(dev);
178 	if (strcmp(devname, dumpdev) != 0 &&
179 	   (strncmp(dumpdev, devprefix, len) != 0 ||
180 	    strcmp(devname, dumpdev + len) != 0))
181 		return (0);
182 
183 	cp = (struct g_consumer *)dev->si_drv2;
184 	error = g_access(cp, 1, 0, 0);
185 	if (error != 0)
186 		return (error);
187 
188 	error = g_dev_setdumpdev(dev, &kda, curthread);
189 	if (error == 0) {
190 		freeenv(dumpdev);
191 		dumpdev = NULL;
192 	}
193 
194 	(void)g_access(cp, -1, 0, 0);
195 
196 	return (error);
197 }
198 
199 static void
200 g_dev_destroy(void *arg, int flags __unused)
201 {
202 	struct g_consumer *cp;
203 	struct g_geom *gp;
204 	struct g_dev_softc *sc;
205 	char buf[SPECNAMELEN + 6];
206 
207 	g_topology_assert();
208 	cp = arg;
209 	gp = cp->geom;
210 	sc = cp->private;
211 	g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name);
212 	snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
213 	devctl_notify_f("GEOM", "DEV", "DESTROY", buf, M_WAITOK);
214 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
215 		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
216 	g_detach(cp);
217 	g_destroy_consumer(cp);
218 	g_destroy_geom(gp);
219 	mtx_destroy(&sc->sc_mtx);
220 	g_free(sc);
221 }
222 
223 void
224 g_dev_print(void)
225 {
226 	struct g_geom *gp;
227 	char const *p = "";
228 
229 	LIST_FOREACH(gp, &g_dev_class.geom, geom) {
230 		printf("%s%s", p, gp->name);
231 		p = " ";
232 	}
233 	printf("\n");
234 }
235 
236 static void
237 g_dev_set_physpath(struct g_consumer *cp)
238 {
239 	struct g_dev_softc *sc;
240 	char *physpath;
241 	int error, physpath_len;
242 
243 	if (g_access(cp, 1, 0, 0) != 0)
244 		return;
245 
246 	sc = cp->private;
247 	physpath_len = MAXPATHLEN;
248 	physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
249 	error = g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath);
250 	g_access(cp, -1, 0, 0);
251 	if (error == 0 && strlen(physpath) != 0) {
252 		struct cdev *dev, *old_alias_dev;
253 		struct cdev **alias_devp;
254 
255 		dev = sc->sc_dev;
256 		old_alias_dev = sc->sc_alias;
257 		alias_devp = (struct cdev **)&sc->sc_alias;
258 		make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp, dev,
259 		    old_alias_dev, physpath);
260 	} else if (sc->sc_alias) {
261 		destroy_dev((struct cdev *)sc->sc_alias);
262 		sc->sc_alias = NULL;
263 	}
264 	g_free(physpath);
265 }
266 
267 static void
268 g_dev_set_media(struct g_consumer *cp)
269 {
270 	struct g_dev_softc *sc;
271 	struct cdev *dev;
272 	char buf[SPECNAMELEN + 6];
273 
274 	sc = cp->private;
275 	dev = sc->sc_dev;
276 	snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
277 	devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
278 	devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
279 	dev = sc->sc_alias;
280 	if (dev != NULL) {
281 		snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
282 		devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
283 		devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
284 	}
285 }
286 
287 static void
288 g_dev_attrchanged(struct g_consumer *cp, const char *attr)
289 {
290 
291 	if (strcmp(attr, "GEOM::media") == 0) {
292 		g_dev_set_media(cp);
293 		return;
294 	}
295 
296 	if (strcmp(attr, "GEOM::physpath") == 0) {
297 		g_dev_set_physpath(cp);
298 		return;
299 	}
300 }
301 
302 struct g_provider *
303 g_dev_getprovider(struct cdev *dev)
304 {
305 	struct g_consumer *cp;
306 
307 	g_topology_assert();
308 	if (dev == NULL)
309 		return (NULL);
310 	if (dev->si_devsw != &g_dev_cdevsw)
311 		return (NULL);
312 	cp = dev->si_drv2;
313 	return (cp->provider);
314 }
315 
316 static struct g_geom *
317 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
318 {
319 	struct g_geom *gp;
320 	struct g_geom_alias *gap;
321 	struct g_consumer *cp;
322 	struct g_dev_softc *sc;
323 	int error;
324 	struct cdev *dev, *adev;
325 	char buf[SPECNAMELEN + 6];
326 
327 	g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
328 	g_topology_assert();
329 	gp = g_new_geomf(mp, "%s", pp->name);
330 	sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
331 	mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF);
332 	cp = g_new_consumer(gp);
333 	cp->private = sc;
334 	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
335 	error = g_attach(cp, pp);
336 	KASSERT(error == 0,
337 	    ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
338 	error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev,
339 	    &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name);
340 	if (error != 0) {
341 		printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
342 		    __func__, gp->name, error);
343 		g_detach(cp);
344 		g_destroy_consumer(cp);
345 		g_destroy_geom(gp);
346 		mtx_destroy(&sc->sc_mtx);
347 		g_free(sc);
348 		return (NULL);
349 	}
350 	dev->si_flags |= SI_UNMAPPED;
351 	sc->sc_dev = dev;
352 
353 	dev->si_iosize_max = MAXPHYS;
354 	dev->si_drv2 = cp;
355 	error = init_dumpdev(dev);
356 	if (error != 0)
357 		printf("%s: init_dumpdev() failed (gp->name=%s, error=%d)\n",
358 		    __func__, gp->name, error);
359 
360 	g_dev_attrchanged(cp, "GEOM::physpath");
361 	snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
362 	devctl_notify_f("GEOM", "DEV", "CREATE", buf, M_WAITOK);
363 	/*
364 	 * Now add all the aliases for this drive
365 	 */
366 	LIST_FOREACH(gap, &pp->geom->aliases, ga_next) {
367 		error = make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &adev, dev,
368 		    "%s", gap->ga_alias);
369 		if (error) {
370 			printf("%s: make_dev_alias_p() failed (name=%s, error=%d)\n",
371 			    __func__, gap->ga_alias, error);
372 			continue;
373 		}
374 		snprintf(buf, sizeof(buf), "cdev=%s", gap->ga_alias);
375 		devctl_notify_f("GEOM", "DEV", "CREATE", buf, M_WAITOK);
376 	}
377 
378 	return (gp);
379 }
380 
381 static int
382 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
383 {
384 	struct g_consumer *cp;
385 	struct g_dev_softc *sc;
386 	int error, r, w, e;
387 
388 	cp = dev->si_drv2;
389 	if (cp == NULL)
390 		return (ENXIO);		/* g_dev_taste() not done yet */
391 	g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
392 	    cp->geom->name, flags, fmt, td);
393 
394 	r = flags & FREAD ? 1 : 0;
395 	w = flags & FWRITE ? 1 : 0;
396 #ifdef notyet
397 	e = flags & O_EXCL ? 1 : 0;
398 #else
399 	e = 0;
400 #endif
401 
402 	/*
403 	 * This happens on attempt to open a device node with O_EXEC.
404 	 */
405 	if (r + w + e == 0)
406 		return (EINVAL);
407 
408 	if (w) {
409 		/*
410 		 * When running in very secure mode, do not allow
411 		 * opens for writing of any disks.
412 		 */
413 		error = securelevel_ge(td->td_ucred, 2);
414 		if (error)
415 			return (error);
416 	}
417 	g_topology_lock();
418 	error = g_access(cp, r, w, e);
419 	g_topology_unlock();
420 	if (error == 0) {
421 		sc = cp->private;
422 		mtx_lock(&sc->sc_mtx);
423 		if (sc->sc_open == 0 && sc->sc_active != 0)
424 			wakeup(&sc->sc_active);
425 		sc->sc_open += r + w + e;
426 		mtx_unlock(&sc->sc_mtx);
427 	}
428 	return (error);
429 }
430 
431 static int
432 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
433 {
434 	struct g_consumer *cp;
435 	struct g_dev_softc *sc;
436 	int error, r, w, e;
437 
438 	cp = dev->si_drv2;
439 	if (cp == NULL)
440 		return (ENXIO);
441 	g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
442 	    cp->geom->name, flags, fmt, td);
443 
444 	r = flags & FREAD ? -1 : 0;
445 	w = flags & FWRITE ? -1 : 0;
446 #ifdef notyet
447 	e = flags & O_EXCL ? -1 : 0;
448 #else
449 	e = 0;
450 #endif
451 
452 	/*
453 	 * The vgonel(9) - caused by eg. forced unmount of devfs - calls
454 	 * VOP_CLOSE(9) on devfs vnode without any FREAD or FWRITE flags,
455 	 * which would result in zero deltas, which in turn would cause
456 	 * panic in g_access(9).
457 	 *
458 	 * Note that we cannot zero the counters (ie. do "r = cp->acr"
459 	 * etc) instead, because the consumer might be opened in another
460 	 * devfs instance.
461 	 */
462 	if (r + w + e == 0)
463 		return (EINVAL);
464 
465 	sc = cp->private;
466 	mtx_lock(&sc->sc_mtx);
467 	sc->sc_open += r + w + e;
468 	while (sc->sc_open == 0 && sc->sc_active != 0)
469 		msleep(&sc->sc_active, &sc->sc_mtx, 0, "PRIBIO", 0);
470 	mtx_unlock(&sc->sc_mtx);
471 	g_topology_lock();
472 	error = g_access(cp, r, w, e);
473 	g_topology_unlock();
474 	return (error);
475 }
476 
477 /*
478  * XXX: Until we have unmessed the ioctl situation, there is a race against
479  * XXX: a concurrent orphanization.  We cannot close it by holding topology
480  * XXX: since that would prevent us from doing our job, and stalling events
481  * XXX: will break (actually: stall) the BSD disklabel hacks.
482  */
483 static int
484 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
485 {
486 	struct g_consumer *cp;
487 	struct g_provider *pp;
488 	off_t offset, length, chunk, odd;
489 	int i, error;
490 
491 	cp = dev->si_drv2;
492 	pp = cp->provider;
493 
494 	error = 0;
495 	KASSERT(cp->acr || cp->acw,
496 	    ("Consumer with zero access count in g_dev_ioctl"));
497 
498 	i = IOCPARM_LEN(cmd);
499 	switch (cmd) {
500 	case DIOCGSECTORSIZE:
501 		*(u_int *)data = cp->provider->sectorsize;
502 		if (*(u_int *)data == 0)
503 			error = ENOENT;
504 		break;
505 	case DIOCGMEDIASIZE:
506 		*(off_t *)data = cp->provider->mediasize;
507 		if (*(off_t *)data == 0)
508 			error = ENOENT;
509 		break;
510 	case DIOCGFWSECTORS:
511 		error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
512 		if (error == 0 && *(u_int *)data == 0)
513 			error = ENOENT;
514 		break;
515 	case DIOCGFWHEADS:
516 		error = g_io_getattr("GEOM::fwheads", cp, &i, data);
517 		if (error == 0 && *(u_int *)data == 0)
518 			error = ENOENT;
519 		break;
520 	case DIOCGFRONTSTUFF:
521 		error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
522 		break;
523 #ifdef COMPAT_FREEBSD11
524 	case DIOCSKERNELDUMP_FREEBSD11:
525 	    {
526 		struct diocskerneldump_arg kda;
527 
528 		bzero(&kda, sizeof(kda));
529 		kda.kda_encryption = KERNELDUMP_ENC_NONE;
530 		kda.kda_enable = (uint8_t)*(u_int *)data;
531 		if (kda.kda_enable == 0)
532 			error = g_dev_setdumpdev(NULL, NULL, td);
533 		else
534 			error = g_dev_setdumpdev(dev, &kda, td);
535 		break;
536 	    }
537 #endif
538 	case DIOCSKERNELDUMP:
539 	    {
540 		struct diocskerneldump_arg *kda;
541 		uint8_t *encryptedkey;
542 
543 		kda = (struct diocskerneldump_arg *)data;
544 		if (kda->kda_enable == 0) {
545 			error = g_dev_setdumpdev(NULL, NULL, td);
546 			break;
547 		}
548 
549 		if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
550 			if (kda->kda_encryptedkeysize <= 0 ||
551 			    kda->kda_encryptedkeysize >
552 			    KERNELDUMP_ENCKEY_MAX_SIZE) {
553 				return (EINVAL);
554 			}
555 			encryptedkey = malloc(kda->kda_encryptedkeysize, M_TEMP,
556 			    M_WAITOK);
557 			error = copyin(kda->kda_encryptedkey, encryptedkey,
558 			    kda->kda_encryptedkeysize);
559 		} else {
560 			encryptedkey = NULL;
561 		}
562 		if (error == 0) {
563 			kda->kda_encryptedkey = encryptedkey;
564 			error = g_dev_setdumpdev(dev, kda, td);
565 		}
566 		if (encryptedkey != NULL) {
567 			explicit_bzero(encryptedkey, kda->kda_encryptedkeysize);
568 			free(encryptedkey, M_TEMP);
569 		}
570 		explicit_bzero(kda, sizeof(*kda));
571 		break;
572 	    }
573 	case DIOCGFLUSH:
574 		error = g_io_flush(cp);
575 		break;
576 	case DIOCGDELETE:
577 		offset = ((off_t *)data)[0];
578 		length = ((off_t *)data)[1];
579 		if ((offset % cp->provider->sectorsize) != 0 ||
580 		    (length % cp->provider->sectorsize) != 0 || length <= 0) {
581 			printf("%s: offset=%jd length=%jd\n", __func__, offset,
582 			    length);
583 			error = EINVAL;
584 			break;
585 		}
586 		while (length > 0) {
587 			chunk = length;
588 			if (g_dev_del_max_sectors != 0 && chunk >
589 			    g_dev_del_max_sectors * cp->provider->sectorsize) {
590 				chunk = g_dev_del_max_sectors *
591 				    cp->provider->sectorsize;
592 				if (cp->provider->stripesize > 0) {
593 					odd = (offset + chunk +
594 					    cp->provider->stripeoffset) %
595 					    cp->provider->stripesize;
596 					if (chunk > odd)
597 						chunk -= odd;
598 				}
599 			}
600 			error = g_delete_data(cp, offset, chunk);
601 			length -= chunk;
602 			offset += chunk;
603 			if (error)
604 				break;
605 			/*
606 			 * Since the request size can be large, the service
607 			 * time can be is likewise.  We make this ioctl
608 			 * interruptible by checking for signals for each bio.
609 			 */
610 			if (SIGPENDING(td))
611 				break;
612 		}
613 		break;
614 	case DIOCGIDENT:
615 		error = g_io_getattr("GEOM::ident", cp, &i, data);
616 		break;
617 	case DIOCGPROVIDERNAME:
618 		if (pp == NULL)
619 			return (ENOENT);
620 		strlcpy(data, pp->name, i);
621 		break;
622 	case DIOCGSTRIPESIZE:
623 		*(off_t *)data = cp->provider->stripesize;
624 		break;
625 	case DIOCGSTRIPEOFFSET:
626 		*(off_t *)data = cp->provider->stripeoffset;
627 		break;
628 	case DIOCGPHYSPATH:
629 		error = g_io_getattr("GEOM::physpath", cp, &i, data);
630 		if (error == 0 && *(char *)data == '\0')
631 			error = ENOENT;
632 		break;
633 	case DIOCGATTR: {
634 		struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
635 
636 		if (arg->len > sizeof(arg->value)) {
637 			error = EINVAL;
638 			break;
639 		}
640 		error = g_io_getattr(arg->name, cp, &arg->len, &arg->value);
641 		break;
642 	}
643 	case DIOCZONECMD: {
644 		struct disk_zone_args *zone_args =(struct disk_zone_args *)data;
645 		struct disk_zone_rep_entry *new_entries, *old_entries;
646 		struct disk_zone_report *rep;
647 		size_t alloc_size;
648 
649 		old_entries = NULL;
650 		new_entries = NULL;
651 		rep = NULL;
652 		alloc_size = 0;
653 
654 		if (zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES) {
655 
656 			rep = &zone_args->zone_params.report;
657 			alloc_size = rep->entries_allocated *
658 			    sizeof(struct disk_zone_rep_entry);
659 			if (alloc_size != 0)
660 				new_entries = g_malloc(alloc_size,
661 				    M_WAITOK| M_ZERO);
662 			old_entries = rep->entries;
663 			rep->entries = new_entries;
664 		}
665 		error = g_io_zonecmd(zone_args, cp);
666 		if ((zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES)
667 		 && (alloc_size != 0)
668 		 && (error == 0)) {
669 			error = copyout(new_entries, old_entries, alloc_size);
670 		}
671 		if ((old_entries != NULL)
672 		 && (rep != NULL))
673 			rep->entries = old_entries;
674 
675 		if (new_entries != NULL)
676 			g_free(new_entries);
677 		break;
678 	}
679 	default:
680 		if (cp->provider->geom->ioctl != NULL) {
681 			error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td);
682 		} else {
683 			error = ENOIOCTL;
684 		}
685 	}
686 
687 	return (error);
688 }
689 
690 static void
691 g_dev_done(struct bio *bp2)
692 {
693 	struct g_consumer *cp;
694 	struct g_dev_softc *sc;
695 	struct bio *bp;
696 	int destroy;
697 
698 	cp = bp2->bio_from;
699 	sc = cp->private;
700 	bp = bp2->bio_parent;
701 	bp->bio_error = bp2->bio_error;
702 	bp->bio_completed = bp2->bio_completed;
703 	bp->bio_resid = bp->bio_length - bp2->bio_completed;
704 	if (bp2->bio_cmd == BIO_ZONE)
705 		bcopy(&bp2->bio_zone, &bp->bio_zone, sizeof(bp->bio_zone));
706 
707 	if (bp2->bio_error != 0) {
708 		g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
709 		    bp2, bp2->bio_error);
710 		bp->bio_flags |= BIO_ERROR;
711 	} else {
712 		g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
713 		    bp2, bp, bp2->bio_resid, (intmax_t)bp2->bio_completed);
714 	}
715 	g_destroy_bio(bp2);
716 	destroy = 0;
717 	mtx_lock(&sc->sc_mtx);
718 	if ((--sc->sc_active) == 0) {
719 		if (sc->sc_open == 0)
720 			wakeup(&sc->sc_active);
721 		if (sc->sc_dev == NULL)
722 			destroy = 1;
723 	}
724 	mtx_unlock(&sc->sc_mtx);
725 	if (destroy)
726 		g_post_event(g_dev_destroy, cp, M_NOWAIT, NULL);
727 	biodone(bp);
728 }
729 
730 static void
731 g_dev_strategy(struct bio *bp)
732 {
733 	struct g_consumer *cp;
734 	struct bio *bp2;
735 	struct cdev *dev;
736 	struct g_dev_softc *sc;
737 
738 	KASSERT(bp->bio_cmd == BIO_READ ||
739 	        bp->bio_cmd == BIO_WRITE ||
740 	        bp->bio_cmd == BIO_DELETE ||
741 		bp->bio_cmd == BIO_FLUSH ||
742 		bp->bio_cmd == BIO_ZONE,
743 		("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
744 	dev = bp->bio_dev;
745 	cp = dev->si_drv2;
746 	sc = cp->private;
747 	KASSERT(cp->acr || cp->acw,
748 	    ("Consumer with zero access count in g_dev_strategy"));
749 	biotrack(bp, __func__);
750 #ifdef INVARIANTS
751 	if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
752 	    (bp->bio_bcount % cp->provider->sectorsize) != 0) {
753 		bp->bio_resid = bp->bio_bcount;
754 		biofinish(bp, NULL, EINVAL);
755 		return;
756 	}
757 #endif
758 	mtx_lock(&sc->sc_mtx);
759 	KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
760 	sc->sc_active++;
761 	mtx_unlock(&sc->sc_mtx);
762 
763 	for (;;) {
764 		/*
765 		 * XXX: This is not an ideal solution, but I believe it to
766 		 * XXX: deadlock safely, all things considered.
767 		 */
768 		bp2 = g_clone_bio(bp);
769 		if (bp2 != NULL)
770 			break;
771 		pause("gdstrat", hz / 10);
772 	}
773 	KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
774 	bp2->bio_done = g_dev_done;
775 	g_trace(G_T_BIO,
776 	    "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
777 	    bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
778 	    bp2->bio_data, bp2->bio_cmd);
779 	g_io_request(bp2, cp);
780 	KASSERT(cp->acr || cp->acw,
781 	    ("g_dev_strategy raced with g_dev_close and lost"));
782 
783 }
784 
785 /*
786  * g_dev_callback()
787  *
788  * Called by devfs when asynchronous device destruction is completed.
789  * - Mark that we have no attached device any more.
790  * - If there are no outstanding requests, schedule geom destruction.
791  *   Otherwise destruction will be scheduled later by g_dev_done().
792  */
793 
794 static void
795 g_dev_callback(void *arg)
796 {
797 	struct g_consumer *cp;
798 	struct g_dev_softc *sc;
799 	int destroy;
800 
801 	cp = arg;
802 	sc = cp->private;
803 	g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
804 
805 	mtx_lock(&sc->sc_mtx);
806 	sc->sc_dev = NULL;
807 	sc->sc_alias = NULL;
808 	destroy = (sc->sc_active == 0);
809 	mtx_unlock(&sc->sc_mtx);
810 	if (destroy)
811 		g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
812 }
813 
814 /*
815  * g_dev_orphan()
816  *
817  * Called from below when the provider orphaned us.
818  * - Clear any dump settings.
819  * - Request asynchronous device destruction to prevent any more requests
820  *   from coming in.  The provider is already marked with an error, so
821  *   anything which comes in the interim will be returned immediately.
822  */
823 
824 static void
825 g_dev_orphan(struct g_consumer *cp)
826 {
827 	struct cdev *dev;
828 	struct g_dev_softc *sc;
829 
830 	g_topology_assert();
831 	sc = cp->private;
832 	dev = sc->sc_dev;
833 	g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
834 
835 	/* Reset any dump-area set on this device */
836 	if (dev->si_flags & SI_DUMPDEV)
837 		(void)clear_dumper(curthread);
838 
839 	/* Destroy the struct cdev *so we get no more requests */
840 	destroy_dev_sched_cb(dev, g_dev_callback, cp);
841 }
842 
843 DECLARE_GEOM_CLASS(g_dev_class, g_dev);
844