xref: /freebsd/sys/geom/geom_dev.c (revision 40427cca7a9ae77b095936fb1954417c290cfb17)
1 /*-
2  * Copyright (c) 2002 Poul-Henning Kamp
3  * Copyright (c) 2002 Networks Associates Technology, Inc.
4  * All rights reserved.
5  *
6  * This software was developed for the FreeBSD Project by Poul-Henning Kamp
7  * and NAI Labs, the Security Research Division of Network Associates, Inc.
8  * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
9  * DARPA CHATS research program.
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  * 3. The names of the authors may not be used to endorse or promote
20  *    products derived from this software without specific prior written
21  *    permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include "opt_compat.h"
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 (set_dumper(NULL, NULL, td, 0, NULL, 0, NULL));
142 
143 	cp = dev->si_drv2;
144 	len = sizeof(kd);
145 	kd.offset = 0;
146 	kd.length = OFF_MAX;
147 	error = g_io_getattr("GEOM::kerneldump", cp, &len, &kd);
148 	if (error != 0)
149 		return (error);
150 
151 	error = set_dumper(&kd.di, devtoname(dev), td, kda->kda_encryption,
152 	    kda->kda_key, kda->kda_encryptedkeysize, kda->kda_encryptedkey);
153 	if (error == 0)
154 		dev->si_flags |= SI_DUMPDEV;
155 
156 	return (error);
157 }
158 
159 static int
160 init_dumpdev(struct cdev *dev)
161 {
162 	struct diocskerneldump_arg kda;
163 	struct g_consumer *cp;
164 	const char *devprefix = "/dev/", *devname;
165 	int error;
166 	size_t len;
167 
168 	bzero(&kda, sizeof(kda));
169 	kda.kda_enable = 1;
170 
171 	if (dumpdev == NULL)
172 		return (0);
173 
174 	len = strlen(devprefix);
175 	devname = devtoname(dev);
176 	if (strcmp(devname, dumpdev) != 0 &&
177 	   (strncmp(dumpdev, devprefix, len) != 0 ||
178 	    strcmp(devname, dumpdev + len) != 0))
179 		return (0);
180 
181 	cp = (struct g_consumer *)dev->si_drv2;
182 	error = g_access(cp, 1, 0, 0);
183 	if (error != 0)
184 		return (error);
185 
186 	error = g_dev_setdumpdev(dev, &kda, curthread);
187 	if (error == 0) {
188 		freeenv(dumpdev);
189 		dumpdev = NULL;
190 	}
191 
192 	(void)g_access(cp, -1, 0, 0);
193 
194 	return (error);
195 }
196 
197 static void
198 g_dev_destroy(void *arg, int flags __unused)
199 {
200 	struct g_consumer *cp;
201 	struct g_geom *gp;
202 	struct g_dev_softc *sc;
203 	char buf[SPECNAMELEN + 6];
204 
205 	g_topology_assert();
206 	cp = arg;
207 	gp = cp->geom;
208 	sc = cp->private;
209 	g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name);
210 	snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
211 	devctl_notify_f("GEOM", "DEV", "DESTROY", buf, M_WAITOK);
212 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
213 		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
214 	g_detach(cp);
215 	g_destroy_consumer(cp);
216 	g_destroy_geom(gp);
217 	mtx_destroy(&sc->sc_mtx);
218 	g_free(sc);
219 }
220 
221 void
222 g_dev_print(void)
223 {
224 	struct g_geom *gp;
225 	char const *p = "";
226 
227 	LIST_FOREACH(gp, &g_dev_class.geom, geom) {
228 		printf("%s%s", p, gp->name);
229 		p = " ";
230 	}
231 	printf("\n");
232 }
233 
234 static void
235 g_dev_set_physpath(struct g_consumer *cp)
236 {
237 	struct g_dev_softc *sc;
238 	char *physpath;
239 	int error, physpath_len;
240 
241 	if (g_access(cp, 1, 0, 0) != 0)
242 		return;
243 
244 	sc = cp->private;
245 	physpath_len = MAXPATHLEN;
246 	physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
247 	error = g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath);
248 	g_access(cp, -1, 0, 0);
249 	if (error == 0 && strlen(physpath) != 0) {
250 		struct cdev *dev, *old_alias_dev;
251 		struct cdev **alias_devp;
252 
253 		dev = sc->sc_dev;
254 		old_alias_dev = sc->sc_alias;
255 		alias_devp = (struct cdev **)&sc->sc_alias;
256 		make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp, dev,
257 		    old_alias_dev, physpath);
258 	} else if (sc->sc_alias) {
259 		destroy_dev((struct cdev *)sc->sc_alias);
260 		sc->sc_alias = NULL;
261 	}
262 	g_free(physpath);
263 }
264 
265 static void
266 g_dev_set_media(struct g_consumer *cp)
267 {
268 	struct g_dev_softc *sc;
269 	struct cdev *dev;
270 	char buf[SPECNAMELEN + 6];
271 
272 	sc = cp->private;
273 	dev = sc->sc_dev;
274 	snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
275 	devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
276 	devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
277 	dev = sc->sc_alias;
278 	if (dev != NULL) {
279 		snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
280 		devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
281 		devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
282 	}
283 }
284 
285 static void
286 g_dev_attrchanged(struct g_consumer *cp, const char *attr)
287 {
288 
289 	if (strcmp(attr, "GEOM::media") == 0) {
290 		g_dev_set_media(cp);
291 		return;
292 	}
293 
294 	if (strcmp(attr, "GEOM::physpath") == 0) {
295 		g_dev_set_physpath(cp);
296 		return;
297 	}
298 }
299 
300 struct g_provider *
301 g_dev_getprovider(struct cdev *dev)
302 {
303 	struct g_consumer *cp;
304 
305 	g_topology_assert();
306 	if (dev == NULL)
307 		return (NULL);
308 	if (dev->si_devsw != &g_dev_cdevsw)
309 		return (NULL);
310 	cp = dev->si_drv2;
311 	return (cp->provider);
312 }
313 
314 static struct g_geom *
315 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
316 {
317 	struct g_geom *gp;
318 	struct g_geom_alias *gap;
319 	struct g_consumer *cp;
320 	struct g_dev_softc *sc;
321 	int error;
322 	struct cdev *dev, *adev;
323 	char buf[SPECNAMELEN + 6];
324 
325 	g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
326 	g_topology_assert();
327 	gp = g_new_geomf(mp, "%s", pp->name);
328 	sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
329 	mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF);
330 	cp = g_new_consumer(gp);
331 	cp->private = sc;
332 	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
333 	error = g_attach(cp, pp);
334 	KASSERT(error == 0,
335 	    ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
336 	error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev,
337 	    &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name);
338 	if (error != 0) {
339 		printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
340 		    __func__, gp->name, error);
341 		g_detach(cp);
342 		g_destroy_consumer(cp);
343 		g_destroy_geom(gp);
344 		mtx_destroy(&sc->sc_mtx);
345 		g_free(sc);
346 		return (NULL);
347 	}
348 	dev->si_flags |= SI_UNMAPPED;
349 	sc->sc_dev = dev;
350 
351 	dev->si_iosize_max = MAXPHYS;
352 	dev->si_drv2 = cp;
353 	error = init_dumpdev(dev);
354 	if (error != 0)
355 		printf("%s: init_dumpdev() failed (gp->name=%s, error=%d)\n",
356 		    __func__, gp->name, error);
357 
358 	g_dev_attrchanged(cp, "GEOM::physpath");
359 	snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
360 	devctl_notify_f("GEOM", "DEV", "CREATE", buf, M_WAITOK);
361 	/*
362 	 * Now add all the aliases for this drive
363 	 */
364 	LIST_FOREACH(gap, &pp->geom->aliases, ga_next) {
365 		error = make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &adev, dev,
366 		    "%s", gap->ga_alias);
367 		if (error) {
368 			printf("%s: make_dev_alias_p() failed (name=%s, error=%d)\n",
369 			    __func__, gap->ga_alias, error);
370 			continue;
371 		}
372 		snprintf(buf, sizeof(buf), "cdev=%s", gap->ga_alias);
373 		devctl_notify_f("GEOM", "DEV", "CREATE", buf, M_WAITOK);
374 	}
375 
376 	return (gp);
377 }
378 
379 static int
380 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
381 {
382 	struct g_consumer *cp;
383 	struct g_dev_softc *sc;
384 	int error, r, w, e;
385 
386 	cp = dev->si_drv2;
387 	if (cp == NULL)
388 		return (ENXIO);		/* g_dev_taste() not done yet */
389 	g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
390 	    cp->geom->name, flags, fmt, td);
391 
392 	r = flags & FREAD ? 1 : 0;
393 	w = flags & FWRITE ? 1 : 0;
394 #ifdef notyet
395 	e = flags & O_EXCL ? 1 : 0;
396 #else
397 	e = 0;
398 #endif
399 
400 	/*
401 	 * This happens on attempt to open a device node with O_EXEC.
402 	 */
403 	if (r + w + e == 0)
404 		return (EINVAL);
405 
406 	if (w) {
407 		/*
408 		 * When running in very secure mode, do not allow
409 		 * opens for writing of any disks.
410 		 */
411 		error = securelevel_ge(td->td_ucred, 2);
412 		if (error)
413 			return (error);
414 	}
415 	g_topology_lock();
416 	error = g_access(cp, r, w, e);
417 	g_topology_unlock();
418 	if (error == 0) {
419 		sc = cp->private;
420 		mtx_lock(&sc->sc_mtx);
421 		if (sc->sc_open == 0 && sc->sc_active != 0)
422 			wakeup(&sc->sc_active);
423 		sc->sc_open += r + w + e;
424 		mtx_unlock(&sc->sc_mtx);
425 	}
426 	return (error);
427 }
428 
429 static int
430 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
431 {
432 	struct g_consumer *cp;
433 	struct g_dev_softc *sc;
434 	int error, r, w, e;
435 
436 	cp = dev->si_drv2;
437 	if (cp == NULL)
438 		return (ENXIO);
439 	g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
440 	    cp->geom->name, flags, fmt, td);
441 
442 	r = flags & FREAD ? -1 : 0;
443 	w = flags & FWRITE ? -1 : 0;
444 #ifdef notyet
445 	e = flags & O_EXCL ? -1 : 0;
446 #else
447 	e = 0;
448 #endif
449 
450 	/*
451 	 * The vgonel(9) - caused by eg. forced unmount of devfs - calls
452 	 * VOP_CLOSE(9) on devfs vnode without any FREAD or FWRITE flags,
453 	 * which would result in zero deltas, which in turn would cause
454 	 * panic in g_access(9).
455 	 *
456 	 * Note that we cannot zero the counters (ie. do "r = cp->acr"
457 	 * etc) instead, because the consumer might be opened in another
458 	 * devfs instance.
459 	 */
460 	if (r + w + e == 0)
461 		return (EINVAL);
462 
463 	sc = cp->private;
464 	mtx_lock(&sc->sc_mtx);
465 	sc->sc_open += r + w + e;
466 	while (sc->sc_open == 0 && sc->sc_active != 0)
467 		msleep(&sc->sc_active, &sc->sc_mtx, 0, "PRIBIO", 0);
468 	mtx_unlock(&sc->sc_mtx);
469 	g_topology_lock();
470 	error = g_access(cp, r, w, e);
471 	g_topology_unlock();
472 	return (error);
473 }
474 
475 /*
476  * XXX: Until we have unmessed the ioctl situation, there is a race against
477  * XXX: a concurrent orphanization.  We cannot close it by holding topology
478  * XXX: since that would prevent us from doing our job, and stalling events
479  * XXX: will break (actually: stall) the BSD disklabel hacks.
480  */
481 static int
482 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
483 {
484 	struct g_consumer *cp;
485 	struct g_provider *pp;
486 	off_t offset, length, chunk, odd;
487 	int i, error;
488 
489 	cp = dev->si_drv2;
490 	pp = cp->provider;
491 
492 	error = 0;
493 	KASSERT(cp->acr || cp->acw,
494 	    ("Consumer with zero access count in g_dev_ioctl"));
495 
496 	i = IOCPARM_LEN(cmd);
497 	switch (cmd) {
498 	case DIOCGSECTORSIZE:
499 		*(u_int *)data = cp->provider->sectorsize;
500 		if (*(u_int *)data == 0)
501 			error = ENOENT;
502 		break;
503 	case DIOCGMEDIASIZE:
504 		*(off_t *)data = cp->provider->mediasize;
505 		if (*(off_t *)data == 0)
506 			error = ENOENT;
507 		break;
508 	case DIOCGFWSECTORS:
509 		error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
510 		if (error == 0 && *(u_int *)data == 0)
511 			error = ENOENT;
512 		break;
513 	case DIOCGFWHEADS:
514 		error = g_io_getattr("GEOM::fwheads", cp, &i, data);
515 		if (error == 0 && *(u_int *)data == 0)
516 			error = ENOENT;
517 		break;
518 	case DIOCGFRONTSTUFF:
519 		error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
520 		break;
521 #ifdef COMPAT_FREEBSD11
522 	case DIOCSKERNELDUMP_FREEBSD11:
523 	    {
524 		struct diocskerneldump_arg kda;
525 
526 		bzero(&kda, sizeof(kda));
527 		kda.kda_encryption = KERNELDUMP_ENC_NONE;
528 		kda.kda_enable = (uint8_t)*(u_int *)data;
529 		if (kda.kda_enable == 0)
530 			error = g_dev_setdumpdev(NULL, NULL, td);
531 		else
532 			error = g_dev_setdumpdev(dev, &kda, td);
533 		break;
534 	    }
535 #endif
536 	case DIOCSKERNELDUMP:
537 	    {
538 		struct diocskerneldump_arg *kda;
539 		uint8_t *encryptedkey;
540 
541 		kda = (struct diocskerneldump_arg *)data;
542 		if (kda->kda_enable == 0) {
543 			error = g_dev_setdumpdev(NULL, NULL, td);
544 			break;
545 		}
546 
547 		if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
548 			if (kda->kda_encryptedkeysize <= 0 ||
549 			    kda->kda_encryptedkeysize >
550 			    KERNELDUMP_ENCKEY_MAX_SIZE) {
551 				return (EINVAL);
552 			}
553 			encryptedkey = malloc(kda->kda_encryptedkeysize, M_TEMP,
554 			    M_WAITOK);
555 			error = copyin(kda->kda_encryptedkey, encryptedkey,
556 			    kda->kda_encryptedkeysize);
557 		} else {
558 			encryptedkey = NULL;
559 		}
560 		if (error == 0) {
561 			kda->kda_encryptedkey = encryptedkey;
562 			error = g_dev_setdumpdev(dev, kda, td);
563 		}
564 		if (encryptedkey != NULL) {
565 			explicit_bzero(encryptedkey, kda->kda_encryptedkeysize);
566 			free(encryptedkey, M_TEMP);
567 		}
568 		explicit_bzero(kda, sizeof(*kda));
569 		break;
570 	    }
571 	case DIOCGFLUSH:
572 		error = g_io_flush(cp);
573 		break;
574 	case DIOCGDELETE:
575 		offset = ((off_t *)data)[0];
576 		length = ((off_t *)data)[1];
577 		if ((offset % cp->provider->sectorsize) != 0 ||
578 		    (length % cp->provider->sectorsize) != 0 || length <= 0) {
579 			printf("%s: offset=%jd length=%jd\n", __func__, offset,
580 			    length);
581 			error = EINVAL;
582 			break;
583 		}
584 		while (length > 0) {
585 			chunk = length;
586 			if (g_dev_del_max_sectors != 0 && chunk >
587 			    g_dev_del_max_sectors * cp->provider->sectorsize) {
588 				chunk = g_dev_del_max_sectors *
589 				    cp->provider->sectorsize;
590 				if (cp->provider->stripesize > 0) {
591 					odd = (offset + chunk +
592 					    cp->provider->stripeoffset) %
593 					    cp->provider->stripesize;
594 					if (chunk > odd)
595 						chunk -= odd;
596 				}
597 			}
598 			error = g_delete_data(cp, offset, chunk);
599 			length -= chunk;
600 			offset += chunk;
601 			if (error)
602 				break;
603 			/*
604 			 * Since the request size can be large, the service
605 			 * time can be is likewise.  We make this ioctl
606 			 * interruptible by checking for signals for each bio.
607 			 */
608 			if (SIGPENDING(td))
609 				break;
610 		}
611 		break;
612 	case DIOCGIDENT:
613 		error = g_io_getattr("GEOM::ident", cp, &i, data);
614 		break;
615 	case DIOCGPROVIDERNAME:
616 		if (pp == NULL)
617 			return (ENOENT);
618 		strlcpy(data, pp->name, i);
619 		break;
620 	case DIOCGSTRIPESIZE:
621 		*(off_t *)data = cp->provider->stripesize;
622 		break;
623 	case DIOCGSTRIPEOFFSET:
624 		*(off_t *)data = cp->provider->stripeoffset;
625 		break;
626 	case DIOCGPHYSPATH:
627 		error = g_io_getattr("GEOM::physpath", cp, &i, data);
628 		if (error == 0 && *(char *)data == '\0')
629 			error = ENOENT;
630 		break;
631 	case DIOCGATTR: {
632 		struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
633 
634 		if (arg->len > sizeof(arg->value)) {
635 			error = EINVAL;
636 			break;
637 		}
638 		error = g_io_getattr(arg->name, cp, &arg->len, &arg->value);
639 		break;
640 	}
641 	case DIOCZONECMD: {
642 		struct disk_zone_args *zone_args =(struct disk_zone_args *)data;
643 		struct disk_zone_rep_entry *new_entries, *old_entries;
644 		struct disk_zone_report *rep;
645 		size_t alloc_size;
646 
647 		old_entries = NULL;
648 		new_entries = NULL;
649 		rep = NULL;
650 		alloc_size = 0;
651 
652 		if (zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES) {
653 
654 			rep = &zone_args->zone_params.report;
655 			alloc_size = rep->entries_allocated *
656 			    sizeof(struct disk_zone_rep_entry);
657 			if (alloc_size != 0)
658 				new_entries = g_malloc(alloc_size,
659 				    M_WAITOK| M_ZERO);
660 			old_entries = rep->entries;
661 			rep->entries = new_entries;
662 		}
663 		error = g_io_zonecmd(zone_args, cp);
664 		if ((zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES)
665 		 && (alloc_size != 0)
666 		 && (error == 0)) {
667 			error = copyout(new_entries, old_entries, alloc_size);
668 		}
669 		if ((old_entries != NULL)
670 		 && (rep != NULL))
671 			rep->entries = old_entries;
672 
673 		if (new_entries != NULL)
674 			g_free(new_entries);
675 		break;
676 	}
677 	default:
678 		if (cp->provider->geom->ioctl != NULL) {
679 			error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td);
680 		} else {
681 			error = ENOIOCTL;
682 		}
683 	}
684 
685 	return (error);
686 }
687 
688 static void
689 g_dev_done(struct bio *bp2)
690 {
691 	struct g_consumer *cp;
692 	struct g_dev_softc *sc;
693 	struct bio *bp;
694 	int destroy;
695 
696 	cp = bp2->bio_from;
697 	sc = cp->private;
698 	bp = bp2->bio_parent;
699 	bp->bio_error = bp2->bio_error;
700 	bp->bio_completed = bp2->bio_completed;
701 	bp->bio_resid = bp->bio_length - bp2->bio_completed;
702 	if (bp2->bio_cmd == BIO_ZONE)
703 		bcopy(&bp2->bio_zone, &bp->bio_zone, sizeof(bp->bio_zone));
704 
705 	if (bp2->bio_error != 0) {
706 		g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
707 		    bp2, bp2->bio_error);
708 		bp->bio_flags |= BIO_ERROR;
709 	} else {
710 		g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
711 		    bp2, bp, bp2->bio_resid, (intmax_t)bp2->bio_completed);
712 	}
713 	g_destroy_bio(bp2);
714 	destroy = 0;
715 	mtx_lock(&sc->sc_mtx);
716 	if ((--sc->sc_active) == 0) {
717 		if (sc->sc_open == 0)
718 			wakeup(&sc->sc_active);
719 		if (sc->sc_dev == NULL)
720 			destroy = 1;
721 	}
722 	mtx_unlock(&sc->sc_mtx);
723 	if (destroy)
724 		g_post_event(g_dev_destroy, cp, M_NOWAIT, NULL);
725 	biodone(bp);
726 }
727 
728 static void
729 g_dev_strategy(struct bio *bp)
730 {
731 	struct g_consumer *cp;
732 	struct bio *bp2;
733 	struct cdev *dev;
734 	struct g_dev_softc *sc;
735 
736 	KASSERT(bp->bio_cmd == BIO_READ ||
737 	        bp->bio_cmd == BIO_WRITE ||
738 	        bp->bio_cmd == BIO_DELETE ||
739 		bp->bio_cmd == BIO_FLUSH ||
740 		bp->bio_cmd == BIO_ZONE,
741 		("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
742 	dev = bp->bio_dev;
743 	cp = dev->si_drv2;
744 	sc = cp->private;
745 	KASSERT(cp->acr || cp->acw,
746 	    ("Consumer with zero access count in g_dev_strategy"));
747 	biotrack(bp, __func__);
748 #ifdef INVARIANTS
749 	if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
750 	    (bp->bio_bcount % cp->provider->sectorsize) != 0) {
751 		bp->bio_resid = bp->bio_bcount;
752 		biofinish(bp, NULL, EINVAL);
753 		return;
754 	}
755 #endif
756 	mtx_lock(&sc->sc_mtx);
757 	KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
758 	sc->sc_active++;
759 	mtx_unlock(&sc->sc_mtx);
760 
761 	for (;;) {
762 		/*
763 		 * XXX: This is not an ideal solution, but I believe it to
764 		 * XXX: deadlock safely, all things considered.
765 		 */
766 		bp2 = g_clone_bio(bp);
767 		if (bp2 != NULL)
768 			break;
769 		pause("gdstrat", hz / 10);
770 	}
771 	KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
772 	bp2->bio_done = g_dev_done;
773 	g_trace(G_T_BIO,
774 	    "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
775 	    bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
776 	    bp2->bio_data, bp2->bio_cmd);
777 	g_io_request(bp2, cp);
778 	KASSERT(cp->acr || cp->acw,
779 	    ("g_dev_strategy raced with g_dev_close and lost"));
780 
781 }
782 
783 /*
784  * g_dev_callback()
785  *
786  * Called by devfs when asynchronous device destruction is completed.
787  * - Mark that we have no attached device any more.
788  * - If there are no outstanding requests, schedule geom destruction.
789  *   Otherwise destruction will be scheduled later by g_dev_done().
790  */
791 
792 static void
793 g_dev_callback(void *arg)
794 {
795 	struct g_consumer *cp;
796 	struct g_dev_softc *sc;
797 	int destroy;
798 
799 	cp = arg;
800 	sc = cp->private;
801 	g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
802 
803 	mtx_lock(&sc->sc_mtx);
804 	sc->sc_dev = NULL;
805 	sc->sc_alias = NULL;
806 	destroy = (sc->sc_active == 0);
807 	mtx_unlock(&sc->sc_mtx);
808 	if (destroy)
809 		g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
810 }
811 
812 /*
813  * g_dev_orphan()
814  *
815  * Called from below when the provider orphaned us.
816  * - Clear any dump settings.
817  * - Request asynchronous device destruction to prevent any more requests
818  *   from coming in.  The provider is already marked with an error, so
819  *   anything which comes in the interim will be returned immediately.
820  */
821 
822 static void
823 g_dev_orphan(struct g_consumer *cp)
824 {
825 	struct cdev *dev;
826 	struct g_dev_softc *sc;
827 
828 	g_topology_assert();
829 	sc = cp->private;
830 	dev = sc->sc_dev;
831 	g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
832 
833 	/* Reset any dump-area set on this device */
834 	if (dev->si_flags & SI_DUMPDEV)
835 		(void)set_dumper(NULL, NULL, curthread, 0, NULL, 0, NULL);
836 
837 	/* Destroy the struct cdev *so we get no more requests */
838 	destroy_dev_sched_cb(dev, g_dev_callback, cp);
839 }
840 
841 DECLARE_GEOM_CLASS(g_dev_class, g_dev);
842