xref: /freebsd/sys/geom/geom_dev.c (revision 076ad2f836d5f49dc1375f1677335a48fe0d4b82)
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_consumer *cp;
319 	struct g_dev_softc *sc;
320 	int error;
321 	struct cdev *dev;
322 	char buf[SPECNAMELEN + 6];
323 
324 	g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
325 	g_topology_assert();
326 	gp = g_new_geomf(mp, "%s", pp->name);
327 	sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
328 	mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF);
329 	cp = g_new_consumer(gp);
330 	cp->private = sc;
331 	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
332 	error = g_attach(cp, pp);
333 	KASSERT(error == 0,
334 	    ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
335 	error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev,
336 	    &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name);
337 	if (error != 0) {
338 		printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
339 		    __func__, gp->name, error);
340 		g_detach(cp);
341 		g_destroy_consumer(cp);
342 		g_destroy_geom(gp);
343 		mtx_destroy(&sc->sc_mtx);
344 		g_free(sc);
345 		return (NULL);
346 	}
347 	dev->si_flags |= SI_UNMAPPED;
348 	sc->sc_dev = dev;
349 
350 	dev->si_iosize_max = MAXPHYS;
351 	dev->si_drv2 = cp;
352 	error = init_dumpdev(dev);
353 	if (error != 0)
354 		printf("%s: init_dumpdev() failed (gp->name=%s, error=%d)\n",
355 		    __func__, gp->name, error);
356 
357 	g_dev_attrchanged(cp, "GEOM::physpath");
358 	snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
359 	devctl_notify_f("GEOM", "DEV", "CREATE", buf, M_WAITOK);
360 
361 	return (gp);
362 }
363 
364 static int
365 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
366 {
367 	struct g_consumer *cp;
368 	struct g_dev_softc *sc;
369 	int error, r, w, e;
370 
371 	cp = dev->si_drv2;
372 	if (cp == NULL)
373 		return (ENXIO);		/* g_dev_taste() not done yet */
374 	g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
375 	    cp->geom->name, flags, fmt, td);
376 
377 	r = flags & FREAD ? 1 : 0;
378 	w = flags & FWRITE ? 1 : 0;
379 #ifdef notyet
380 	e = flags & O_EXCL ? 1 : 0;
381 #else
382 	e = 0;
383 #endif
384 
385 	/*
386 	 * This happens on attempt to open a device node with O_EXEC.
387 	 */
388 	if (r + w + e == 0)
389 		return (EINVAL);
390 
391 	if (w) {
392 		/*
393 		 * When running in very secure mode, do not allow
394 		 * opens for writing of any disks.
395 		 */
396 		error = securelevel_ge(td->td_ucred, 2);
397 		if (error)
398 			return (error);
399 	}
400 	g_topology_lock();
401 	error = g_access(cp, r, w, e);
402 	g_topology_unlock();
403 	if (error == 0) {
404 		sc = cp->private;
405 		mtx_lock(&sc->sc_mtx);
406 		if (sc->sc_open == 0 && sc->sc_active != 0)
407 			wakeup(&sc->sc_active);
408 		sc->sc_open += r + w + e;
409 		mtx_unlock(&sc->sc_mtx);
410 	}
411 	return (error);
412 }
413 
414 static int
415 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
416 {
417 	struct g_consumer *cp;
418 	struct g_dev_softc *sc;
419 	int error, r, w, e;
420 
421 	cp = dev->si_drv2;
422 	if (cp == NULL)
423 		return (ENXIO);
424 	g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
425 	    cp->geom->name, flags, fmt, td);
426 
427 	r = flags & FREAD ? -1 : 0;
428 	w = flags & FWRITE ? -1 : 0;
429 #ifdef notyet
430 	e = flags & O_EXCL ? -1 : 0;
431 #else
432 	e = 0;
433 #endif
434 
435 	/*
436 	 * The vgonel(9) - caused by eg. forced unmount of devfs - calls
437 	 * VOP_CLOSE(9) on devfs vnode without any FREAD or FWRITE flags,
438 	 * which would result in zero deltas, which in turn would cause
439 	 * panic in g_access(9).
440 	 *
441 	 * Note that we cannot zero the counters (ie. do "r = cp->acr"
442 	 * etc) instead, because the consumer might be opened in another
443 	 * devfs instance.
444 	 */
445 	if (r + w + e == 0)
446 		return (EINVAL);
447 
448 	sc = cp->private;
449 	mtx_lock(&sc->sc_mtx);
450 	sc->sc_open += r + w + e;
451 	while (sc->sc_open == 0 && sc->sc_active != 0)
452 		msleep(&sc->sc_active, &sc->sc_mtx, 0, "PRIBIO", 0);
453 	mtx_unlock(&sc->sc_mtx);
454 	g_topology_lock();
455 	error = g_access(cp, r, w, e);
456 	g_topology_unlock();
457 	return (error);
458 }
459 
460 /*
461  * XXX: Until we have unmessed the ioctl situation, there is a race against
462  * XXX: a concurrent orphanization.  We cannot close it by holding topology
463  * XXX: since that would prevent us from doing our job, and stalling events
464  * XXX: will break (actually: stall) the BSD disklabel hacks.
465  */
466 static int
467 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
468 {
469 	struct g_consumer *cp;
470 	struct g_provider *pp;
471 	off_t offset, length, chunk;
472 	int i, error;
473 
474 	cp = dev->si_drv2;
475 	pp = cp->provider;
476 
477 	error = 0;
478 	KASSERT(cp->acr || cp->acw,
479 	    ("Consumer with zero access count in g_dev_ioctl"));
480 
481 	i = IOCPARM_LEN(cmd);
482 	switch (cmd) {
483 	case DIOCGSECTORSIZE:
484 		*(u_int *)data = cp->provider->sectorsize;
485 		if (*(u_int *)data == 0)
486 			error = ENOENT;
487 		break;
488 	case DIOCGMEDIASIZE:
489 		*(off_t *)data = cp->provider->mediasize;
490 		if (*(off_t *)data == 0)
491 			error = ENOENT;
492 		break;
493 	case DIOCGFWSECTORS:
494 		error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
495 		if (error == 0 && *(u_int *)data == 0)
496 			error = ENOENT;
497 		break;
498 	case DIOCGFWHEADS:
499 		error = g_io_getattr("GEOM::fwheads", cp, &i, data);
500 		if (error == 0 && *(u_int *)data == 0)
501 			error = ENOENT;
502 		break;
503 	case DIOCGFRONTSTUFF:
504 		error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
505 		break;
506 #ifdef COMPAT_FREEBSD11
507 	case DIOCSKERNELDUMP_FREEBSD11:
508 	    {
509 		struct diocskerneldump_arg kda;
510 
511 		bzero(&kda, sizeof(kda));
512 		kda.kda_encryption = KERNELDUMP_ENC_NONE;
513 		kda.kda_enable = (uint8_t)*(u_int *)data;
514 		if (kda.kda_enable == 0)
515 			error = g_dev_setdumpdev(NULL, NULL, td);
516 		else
517 			error = g_dev_setdumpdev(dev, &kda, td);
518 		break;
519 	    }
520 #endif
521 	case DIOCSKERNELDUMP:
522 	    {
523 		struct diocskerneldump_arg *kda;
524 		uint8_t *encryptedkey;
525 
526 		kda = (struct diocskerneldump_arg *)data;
527 		if (kda->kda_enable == 0) {
528 			error = g_dev_setdumpdev(NULL, NULL, td);
529 			break;
530 		}
531 
532 		if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
533 			if (kda->kda_encryptedkeysize <= 0 ||
534 			    kda->kda_encryptedkeysize >
535 			    KERNELDUMP_ENCKEY_MAX_SIZE) {
536 				return (EINVAL);
537 			}
538 			encryptedkey = malloc(kda->kda_encryptedkeysize, M_TEMP,
539 			    M_WAITOK);
540 			error = copyin(kda->kda_encryptedkey, encryptedkey,
541 			    kda->kda_encryptedkeysize);
542 		} else {
543 			encryptedkey = NULL;
544 		}
545 		if (error == 0) {
546 			kda->kda_encryptedkey = encryptedkey;
547 			error = g_dev_setdumpdev(dev, kda, td);
548 		}
549 		if (encryptedkey != NULL) {
550 			explicit_bzero(encryptedkey, kda->kda_encryptedkeysize);
551 			free(encryptedkey, M_TEMP);
552 		}
553 		explicit_bzero(kda, sizeof(*kda));
554 		break;
555 	    }
556 	case DIOCGFLUSH:
557 		error = g_io_flush(cp);
558 		break;
559 	case DIOCGDELETE:
560 		offset = ((off_t *)data)[0];
561 		length = ((off_t *)data)[1];
562 		if ((offset % cp->provider->sectorsize) != 0 ||
563 		    (length % cp->provider->sectorsize) != 0 || length <= 0) {
564 			printf("%s: offset=%jd length=%jd\n", __func__, offset,
565 			    length);
566 			error = EINVAL;
567 			break;
568 		}
569 		while (length > 0) {
570 			chunk = length;
571 			if (g_dev_del_max_sectors != 0 && chunk >
572 			    g_dev_del_max_sectors * cp->provider->sectorsize) {
573 				chunk = g_dev_del_max_sectors *
574 				    cp->provider->sectorsize;
575 			}
576 			error = g_delete_data(cp, offset, chunk);
577 			length -= chunk;
578 			offset += chunk;
579 			if (error)
580 				break;
581 			/*
582 			 * Since the request size can be large, the service
583 			 * time can be is likewise.  We make this ioctl
584 			 * interruptible by checking for signals for each bio.
585 			 */
586 			if (SIGPENDING(td))
587 				break;
588 		}
589 		break;
590 	case DIOCGIDENT:
591 		error = g_io_getattr("GEOM::ident", cp, &i, data);
592 		break;
593 	case DIOCGPROVIDERNAME:
594 		if (pp == NULL)
595 			return (ENOENT);
596 		strlcpy(data, pp->name, i);
597 		break;
598 	case DIOCGSTRIPESIZE:
599 		*(off_t *)data = cp->provider->stripesize;
600 		break;
601 	case DIOCGSTRIPEOFFSET:
602 		*(off_t *)data = cp->provider->stripeoffset;
603 		break;
604 	case DIOCGPHYSPATH:
605 		error = g_io_getattr("GEOM::physpath", cp, &i, data);
606 		if (error == 0 && *(char *)data == '\0')
607 			error = ENOENT;
608 		break;
609 	case DIOCGATTR: {
610 		struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
611 
612 		if (arg->len > sizeof(arg->value)) {
613 			error = EINVAL;
614 			break;
615 		}
616 		error = g_io_getattr(arg->name, cp, &arg->len, &arg->value);
617 		break;
618 	}
619 	case DIOCZONECMD: {
620 		struct disk_zone_args *zone_args =(struct disk_zone_args *)data;
621 		struct disk_zone_rep_entry *new_entries, *old_entries;
622 		struct disk_zone_report *rep;
623 		size_t alloc_size;
624 
625 		old_entries = NULL;
626 		new_entries = NULL;
627 		rep = NULL;
628 		alloc_size = 0;
629 
630 		if (zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES) {
631 
632 			rep = &zone_args->zone_params.report;
633 			alloc_size = rep->entries_allocated *
634 			    sizeof(struct disk_zone_rep_entry);
635 			if (alloc_size != 0)
636 				new_entries = g_malloc(alloc_size,
637 				    M_WAITOK| M_ZERO);
638 			old_entries = rep->entries;
639 			rep->entries = new_entries;
640 		}
641 		error = g_io_zonecmd(zone_args, cp);
642 		if ((zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES)
643 		 && (alloc_size != 0)
644 		 && (error == 0)) {
645 			error = copyout(new_entries, old_entries, alloc_size);
646 		}
647 		if ((old_entries != NULL)
648 		 && (rep != NULL))
649 			rep->entries = old_entries;
650 
651 		if (new_entries != NULL)
652 			g_free(new_entries);
653 		break;
654 	}
655 	default:
656 		if (cp->provider->geom->ioctl != NULL) {
657 			error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td);
658 		} else {
659 			error = ENOIOCTL;
660 		}
661 	}
662 
663 	return (error);
664 }
665 
666 static void
667 g_dev_done(struct bio *bp2)
668 {
669 	struct g_consumer *cp;
670 	struct g_dev_softc *sc;
671 	struct bio *bp;
672 	int destroy;
673 
674 	cp = bp2->bio_from;
675 	sc = cp->private;
676 	bp = bp2->bio_parent;
677 	bp->bio_error = bp2->bio_error;
678 	bp->bio_completed = bp2->bio_completed;
679 	bp->bio_resid = bp->bio_length - bp2->bio_completed;
680 	if (bp2->bio_cmd == BIO_ZONE)
681 		bcopy(&bp2->bio_zone, &bp->bio_zone, sizeof(bp->bio_zone));
682 
683 	if (bp2->bio_error != 0) {
684 		g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
685 		    bp2, bp2->bio_error);
686 		bp->bio_flags |= BIO_ERROR;
687 	} else {
688 		g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
689 		    bp2, bp, bp2->bio_resid, (intmax_t)bp2->bio_completed);
690 	}
691 	g_destroy_bio(bp2);
692 	destroy = 0;
693 	mtx_lock(&sc->sc_mtx);
694 	if ((--sc->sc_active) == 0) {
695 		if (sc->sc_open == 0)
696 			wakeup(&sc->sc_active);
697 		if (sc->sc_dev == NULL)
698 			destroy = 1;
699 	}
700 	mtx_unlock(&sc->sc_mtx);
701 	if (destroy)
702 		g_post_event(g_dev_destroy, cp, M_NOWAIT, NULL);
703 	biodone(bp);
704 }
705 
706 static void
707 g_dev_strategy(struct bio *bp)
708 {
709 	struct g_consumer *cp;
710 	struct bio *bp2;
711 	struct cdev *dev;
712 	struct g_dev_softc *sc;
713 
714 	KASSERT(bp->bio_cmd == BIO_READ ||
715 	        bp->bio_cmd == BIO_WRITE ||
716 	        bp->bio_cmd == BIO_DELETE ||
717 		bp->bio_cmd == BIO_FLUSH ||
718 		bp->bio_cmd == BIO_ZONE,
719 		("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
720 	dev = bp->bio_dev;
721 	cp = dev->si_drv2;
722 	sc = cp->private;
723 	KASSERT(cp->acr || cp->acw,
724 	    ("Consumer with zero access count in g_dev_strategy"));
725 	biotrack(bp, __func__);
726 #ifdef INVARIANTS
727 	if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
728 	    (bp->bio_bcount % cp->provider->sectorsize) != 0) {
729 		bp->bio_resid = bp->bio_bcount;
730 		biofinish(bp, NULL, EINVAL);
731 		return;
732 	}
733 #endif
734 	mtx_lock(&sc->sc_mtx);
735 	KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
736 	sc->sc_active++;
737 	mtx_unlock(&sc->sc_mtx);
738 
739 	for (;;) {
740 		/*
741 		 * XXX: This is not an ideal solution, but I believe it to
742 		 * XXX: deadlock safely, all things considered.
743 		 */
744 		bp2 = g_clone_bio(bp);
745 		if (bp2 != NULL)
746 			break;
747 		pause("gdstrat", hz / 10);
748 	}
749 	KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
750 	bp2->bio_done = g_dev_done;
751 	g_trace(G_T_BIO,
752 	    "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
753 	    bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
754 	    bp2->bio_data, bp2->bio_cmd);
755 	g_io_request(bp2, cp);
756 	KASSERT(cp->acr || cp->acw,
757 	    ("g_dev_strategy raced with g_dev_close and lost"));
758 
759 }
760 
761 /*
762  * g_dev_callback()
763  *
764  * Called by devfs when asynchronous device destruction is completed.
765  * - Mark that we have no attached device any more.
766  * - If there are no outstanding requests, schedule geom destruction.
767  *   Otherwise destruction will be scheduled later by g_dev_done().
768  */
769 
770 static void
771 g_dev_callback(void *arg)
772 {
773 	struct g_consumer *cp;
774 	struct g_dev_softc *sc;
775 	int destroy;
776 
777 	cp = arg;
778 	sc = cp->private;
779 	g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
780 
781 	mtx_lock(&sc->sc_mtx);
782 	sc->sc_dev = NULL;
783 	sc->sc_alias = NULL;
784 	destroy = (sc->sc_active == 0);
785 	mtx_unlock(&sc->sc_mtx);
786 	if (destroy)
787 		g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
788 }
789 
790 /*
791  * g_dev_orphan()
792  *
793  * Called from below when the provider orphaned us.
794  * - Clear any dump settings.
795  * - Request asynchronous device destruction to prevent any more requests
796  *   from coming in.  The provider is already marked with an error, so
797  *   anything which comes in the interim will be returned immediately.
798  */
799 
800 static void
801 g_dev_orphan(struct g_consumer *cp)
802 {
803 	struct cdev *dev;
804 	struct g_dev_softc *sc;
805 
806 	g_topology_assert();
807 	sc = cp->private;
808 	dev = sc->sc_dev;
809 	g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
810 
811 	/* Reset any dump-area set on this device */
812 	if (dev->si_flags & SI_DUMPDEV)
813 		(void)set_dumper(NULL, NULL, curthread, 0, NULL, 0, NULL);
814 
815 	/* Destroy the struct cdev *so we get no more requests */
816 	destroy_dev_sched_cb(dev, g_dev_callback, cp);
817 }
818 
819 DECLARE_GEOM_CLASS(g_dev_class, g_dev);
820