xref: /freebsd/sys/geom/geom_dev.c (revision f7c32ed617858bcd22f8d1b03199099d50125721)
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/devctl.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 	u_int		 sc_active;
68 #define	SC_A_DESTROY	(1 << 31)
69 #define	SC_A_OPEN	(1 << 30)
70 #define	SC_A_ACTIVE	(SC_A_OPEN - 1)
71 };
72 
73 static d_open_t		g_dev_open;
74 static d_close_t	g_dev_close;
75 static d_strategy_t	g_dev_strategy;
76 static d_ioctl_t	g_dev_ioctl;
77 
78 static struct cdevsw g_dev_cdevsw = {
79 	.d_version =	D_VERSION,
80 	.d_open =	g_dev_open,
81 	.d_close =	g_dev_close,
82 	.d_read =	physread,
83 	.d_write =	physwrite,
84 	.d_ioctl =	g_dev_ioctl,
85 	.d_strategy =	g_dev_strategy,
86 	.d_name =	"g_dev",
87 	.d_flags =	D_DISK | D_TRACKCLOSE,
88 };
89 
90 static g_init_t g_dev_init;
91 static g_fini_t g_dev_fini;
92 static g_taste_t g_dev_taste;
93 static g_orphan_t g_dev_orphan;
94 static g_attrchanged_t g_dev_attrchanged;
95 static g_resize_t g_dev_resize;
96 
97 static struct g_class g_dev_class	= {
98 	.name = "DEV",
99 	.version = G_VERSION,
100 	.init = g_dev_init,
101 	.fini = g_dev_fini,
102 	.taste = g_dev_taste,
103 	.orphan = g_dev_orphan,
104 	.attrchanged = g_dev_attrchanged,
105 	.resize = g_dev_resize
106 };
107 
108 /*
109  * We target 262144 (8 x 32768) sectors by default as this significantly
110  * increases the throughput on commonly used SSD's with a marginal
111  * increase in non-interruptible request latency.
112  */
113 static uint64_t g_dev_del_max_sectors = 262144;
114 SYSCTL_DECL(_kern_geom);
115 SYSCTL_NODE(_kern_geom, OID_AUTO, dev, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
116     "GEOM_DEV stuff");
117 SYSCTL_QUAD(_kern_geom_dev, OID_AUTO, delete_max_sectors, CTLFLAG_RW,
118     &g_dev_del_max_sectors, 0, "Maximum number of sectors in a single "
119     "delete request sent to the provider. Larger requests are chunked "
120     "so they can be interrupted. (0 = disable chunking)");
121 
122 static char *dumpdev = NULL;
123 static void
124 g_dev_init(struct g_class *mp)
125 {
126 
127 	dumpdev = kern_getenv("dumpdev");
128 }
129 
130 static void
131 g_dev_fini(struct g_class *mp)
132 {
133 
134 	freeenv(dumpdev);
135 	dumpdev = NULL;
136 }
137 
138 static int
139 g_dev_setdumpdev(struct cdev *dev, struct diocskerneldump_arg *kda)
140 {
141 	struct g_kerneldump kd;
142 	struct g_consumer *cp;
143 	int error, len;
144 
145 	MPASS(dev != NULL && kda != NULL);
146 	MPASS(kda->kda_index != KDA_REMOVE);
147 
148 	cp = dev->si_drv2;
149 	len = sizeof(kd);
150 	memset(&kd, 0, len);
151 	kd.offset = 0;
152 	kd.length = OFF_MAX;
153 	error = g_io_getattr("GEOM::kerneldump", cp, &len, &kd);
154 	if (error != 0)
155 		return (error);
156 
157 	error = dumper_insert(&kd.di, devtoname(dev), kda);
158 	if (error == 0)
159 		dev->si_flags |= SI_DUMPDEV;
160 
161 	return (error);
162 }
163 
164 static int
165 init_dumpdev(struct cdev *dev)
166 {
167 	struct diocskerneldump_arg kda;
168 	struct g_consumer *cp;
169 	const char *devprefix = _PATH_DEV, *devname;
170 	int error;
171 	size_t len;
172 
173 	bzero(&kda, sizeof(kda));
174 	kda.kda_index = KDA_APPEND;
175 
176 	if (dumpdev == NULL)
177 		return (0);
178 
179 	len = strlen(devprefix);
180 	devname = devtoname(dev);
181 	if (strcmp(devname, dumpdev) != 0 &&
182 	   (strncmp(dumpdev, devprefix, len) != 0 ||
183 	    strcmp(devname, dumpdev + len) != 0))
184 		return (0);
185 
186 	cp = (struct g_consumer *)dev->si_drv2;
187 	error = g_access(cp, 1, 0, 0);
188 	if (error != 0)
189 		return (error);
190 
191 	error = g_dev_setdumpdev(dev, &kda);
192 	if (error == 0) {
193 		freeenv(dumpdev);
194 		dumpdev = NULL;
195 	}
196 
197 	(void)g_access(cp, -1, 0, 0);
198 
199 	return (error);
200 }
201 
202 static void
203 g_dev_destroy(void *arg, int flags __unused)
204 {
205 	struct g_consumer *cp;
206 	struct g_geom *gp;
207 	struct g_dev_softc *sc;
208 	char buf[SPECNAMELEN + 6];
209 
210 	g_topology_assert();
211 	cp = arg;
212 	gp = cp->geom;
213 	sc = cp->private;
214 	g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name);
215 	snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
216 	devctl_notify("GEOM", "DEV", "DESTROY", buf);
217 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
218 		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
219 	g_detach(cp);
220 	g_destroy_consumer(cp);
221 	g_destroy_geom(gp);
222 	mtx_destroy(&sc->sc_mtx);
223 	g_free(sc);
224 }
225 
226 void
227 g_dev_print(void)
228 {
229 	struct g_geom *gp;
230 	char const *p = "";
231 
232 	LIST_FOREACH(gp, &g_dev_class.geom, geom) {
233 		printf("%s%s", p, gp->name);
234 		p = " ";
235 	}
236 	printf("\n");
237 }
238 
239 static void
240 g_dev_set_physpath(struct g_consumer *cp)
241 {
242 	struct g_dev_softc *sc;
243 	char *physpath;
244 	int error, physpath_len;
245 
246 	if (g_access(cp, 1, 0, 0) != 0)
247 		return;
248 
249 	sc = cp->private;
250 	physpath_len = MAXPATHLEN;
251 	physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
252 	error = g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath);
253 	g_access(cp, -1, 0, 0);
254 	if (error == 0 && strlen(physpath) != 0) {
255 		struct cdev *dev, *old_alias_dev;
256 		struct cdev **alias_devp;
257 
258 		dev = sc->sc_dev;
259 		old_alias_dev = sc->sc_alias;
260 		alias_devp = (struct cdev **)&sc->sc_alias;
261 		make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp, dev,
262 		    old_alias_dev, physpath);
263 	} else if (sc->sc_alias) {
264 		destroy_dev((struct cdev *)sc->sc_alias);
265 		sc->sc_alias = NULL;
266 	}
267 	g_free(physpath);
268 }
269 
270 static void
271 g_dev_set_media(struct g_consumer *cp)
272 {
273 	struct g_dev_softc *sc;
274 	struct cdev *dev;
275 	char buf[SPECNAMELEN + 6];
276 
277 	sc = cp->private;
278 	dev = sc->sc_dev;
279 	snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
280 	devctl_notify("DEVFS", "CDEV", "MEDIACHANGE", buf);
281 	devctl_notify("GEOM", "DEV", "MEDIACHANGE", buf);
282 	dev = sc->sc_alias;
283 	if (dev != NULL) {
284 		snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
285 		devctl_notify("DEVFS", "CDEV", "MEDIACHANGE", buf);
286 		devctl_notify("GEOM", "DEV", "MEDIACHANGE", buf);
287 	}
288 }
289 
290 static void
291 g_dev_attrchanged(struct g_consumer *cp, const char *attr)
292 {
293 
294 	if (strcmp(attr, "GEOM::media") == 0) {
295 		g_dev_set_media(cp);
296 		return;
297 	}
298 
299 	if (strcmp(attr, "GEOM::physpath") == 0) {
300 		g_dev_set_physpath(cp);
301 		return;
302 	}
303 }
304 
305 static void
306 g_dev_resize(struct g_consumer *cp)
307 {
308 	char buf[SPECNAMELEN + 6];
309 
310 	snprintf(buf, sizeof(buf), "cdev=%s", cp->provider->name);
311 	devctl_notify("GEOM", "DEV", "SIZECHANGE", buf);
312 }
313 
314 struct g_provider *
315 g_dev_getprovider(struct cdev *dev)
316 {
317 	struct g_consumer *cp;
318 
319 	g_topology_assert();
320 	if (dev == NULL)
321 		return (NULL);
322 	if (dev->si_devsw != &g_dev_cdevsw)
323 		return (NULL);
324 	cp = dev->si_drv2;
325 	return (cp->provider);
326 }
327 
328 static struct g_geom *
329 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
330 {
331 	struct g_geom *gp;
332 	struct g_geom_alias *gap;
333 	struct g_consumer *cp;
334 	struct g_dev_softc *sc;
335 	int error;
336 	struct cdev *dev, *adev;
337 	char buf[SPECNAMELEN + 6];
338 	struct make_dev_args args;
339 
340 	g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
341 	g_topology_assert();
342 	gp = g_new_geomf(mp, "%s", pp->name);
343 	sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
344 	mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF);
345 	cp = g_new_consumer(gp);
346 	cp->private = sc;
347 	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
348 	error = g_attach(cp, pp);
349 	if (error != 0) {
350 		printf("%s: g_dev_taste(%s) failed to g_attach, error=%d\n",
351 		    __func__, pp->name, error);
352 		g_destroy_consumer(cp);
353 		g_destroy_geom(gp);
354 		mtx_destroy(&sc->sc_mtx);
355 		g_free(sc);
356 		return (NULL);
357 	}
358 	make_dev_args_init(&args);
359 	args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
360 	args.mda_devsw = &g_dev_cdevsw;
361 	args.mda_cr = NULL;
362 	args.mda_uid = UID_ROOT;
363 	args.mda_gid = GID_OPERATOR;
364 	args.mda_mode = 0640;
365 	args.mda_si_drv1 = sc;
366 	args.mda_si_drv2 = cp;
367 	error = make_dev_s(&args, &sc->sc_dev, "%s", gp->name);
368 	if (error != 0) {
369 		printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
370 		    __func__, gp->name, error);
371 		g_detach(cp);
372 		g_destroy_consumer(cp);
373 		g_destroy_geom(gp);
374 		mtx_destroy(&sc->sc_mtx);
375 		g_free(sc);
376 		return (NULL);
377 	}
378 	dev = sc->sc_dev;
379 	dev->si_flags |= SI_UNMAPPED;
380 	dev->si_iosize_max = maxphys;
381 	error = init_dumpdev(dev);
382 	if (error != 0)
383 		printf("%s: init_dumpdev() failed (gp->name=%s, error=%d)\n",
384 		    __func__, gp->name, error);
385 
386 	g_dev_attrchanged(cp, "GEOM::physpath");
387 	snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
388 	devctl_notify("GEOM", "DEV", "CREATE", buf);
389 	/*
390 	 * Now add all the aliases for this drive
391 	 */
392 	LIST_FOREACH(gap, &pp->aliases, ga_next) {
393 		error = make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &adev, dev,
394 		    "%s", gap->ga_alias);
395 		if (error) {
396 			printf("%s: make_dev_alias_p() failed (name=%s, error=%d)\n",
397 			    __func__, gap->ga_alias, error);
398 			continue;
399 		}
400 		snprintf(buf, sizeof(buf), "cdev=%s", gap->ga_alias);
401 		devctl_notify("GEOM", "DEV", "CREATE", buf);
402 	}
403 
404 	return (gp);
405 }
406 
407 static int
408 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
409 {
410 	struct g_consumer *cp;
411 	struct g_dev_softc *sc;
412 	int error, r, w, e;
413 
414 	cp = dev->si_drv2;
415 	g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
416 	    cp->geom->name, flags, fmt, td);
417 
418 	r = flags & FREAD ? 1 : 0;
419 	w = flags & FWRITE ? 1 : 0;
420 #ifdef notyet
421 	e = flags & O_EXCL ? 1 : 0;
422 #else
423 	e = 0;
424 #endif
425 
426 	/*
427 	 * This happens on attempt to open a device node with O_EXEC.
428 	 */
429 	if (r + w + e == 0)
430 		return (EINVAL);
431 
432 	if (w) {
433 		/*
434 		 * When running in very secure mode, do not allow
435 		 * opens for writing of any disks.
436 		 */
437 		error = securelevel_ge(td->td_ucred, 2);
438 		if (error)
439 			return (error);
440 	}
441 	g_topology_lock();
442 	error = g_access(cp, r, w, e);
443 	g_topology_unlock();
444 	if (error == 0) {
445 		sc = dev->si_drv1;
446 		mtx_lock(&sc->sc_mtx);
447 		if (sc->sc_open == 0 && (sc->sc_active & SC_A_ACTIVE) != 0)
448 			wakeup(&sc->sc_active);
449 		sc->sc_open += r + w + e;
450 		if (sc->sc_open == 0)
451 			atomic_clear_int(&sc->sc_active, SC_A_OPEN);
452 		else
453 			atomic_set_int(&sc->sc_active, SC_A_OPEN);
454 		mtx_unlock(&sc->sc_mtx);
455 	}
456 	return (error);
457 }
458 
459 static int
460 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
461 {
462 	struct g_consumer *cp;
463 	struct g_dev_softc *sc;
464 	int error, r, w, e;
465 
466 	cp = dev->si_drv2;
467 	g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
468 	    cp->geom->name, flags, fmt, td);
469 
470 	r = flags & FREAD ? -1 : 0;
471 	w = flags & FWRITE ? -1 : 0;
472 #ifdef notyet
473 	e = flags & O_EXCL ? -1 : 0;
474 #else
475 	e = 0;
476 #endif
477 
478 	/*
479 	 * The vgonel(9) - caused by eg. forced unmount of devfs - calls
480 	 * VOP_CLOSE(9) on devfs vnode without any FREAD or FWRITE flags,
481 	 * which would result in zero deltas, which in turn would cause
482 	 * panic in g_access(9).
483 	 *
484 	 * Note that we cannot zero the counters (ie. do "r = cp->acr"
485 	 * etc) instead, because the consumer might be opened in another
486 	 * devfs instance.
487 	 */
488 	if (r + w + e == 0)
489 		return (EINVAL);
490 
491 	sc = dev->si_drv1;
492 	mtx_lock(&sc->sc_mtx);
493 	sc->sc_open += r + w + e;
494 	if (sc->sc_open == 0)
495 		atomic_clear_int(&sc->sc_active, SC_A_OPEN);
496 	else
497 		atomic_set_int(&sc->sc_active, SC_A_OPEN);
498 	while (sc->sc_open == 0 && (sc->sc_active & SC_A_ACTIVE) != 0)
499 		msleep(&sc->sc_active, &sc->sc_mtx, 0, "g_dev_close", hz / 10);
500 	mtx_unlock(&sc->sc_mtx);
501 	g_topology_lock();
502 	error = g_access(cp, r, w, e);
503 	g_topology_unlock();
504 	return (error);
505 }
506 
507 static int
508 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
509 {
510 	struct g_consumer *cp;
511 	struct g_provider *pp;
512 	off_t offset, length, chunk, odd;
513 	int i, error;
514 #ifdef COMPAT_FREEBSD12
515 	struct diocskerneldump_arg kda_copy;
516 #endif
517 
518 	cp = dev->si_drv2;
519 	pp = cp->provider;
520 
521 	/* If consumer or provider is dying, don't disturb. */
522 	if (cp->flags & G_CF_ORPHAN)
523 		return (ENXIO);
524 	if (pp->error)
525 		return (pp->error);
526 
527 	error = 0;
528 	KASSERT(cp->acr || cp->acw,
529 	    ("Consumer with zero access count in g_dev_ioctl"));
530 
531 	i = IOCPARM_LEN(cmd);
532 	switch (cmd) {
533 	case DIOCGSECTORSIZE:
534 		*(u_int *)data = pp->sectorsize;
535 		if (*(u_int *)data == 0)
536 			error = ENOENT;
537 		break;
538 	case DIOCGMEDIASIZE:
539 		*(off_t *)data = pp->mediasize;
540 		if (*(off_t *)data == 0)
541 			error = ENOENT;
542 		break;
543 	case DIOCGFWSECTORS:
544 		error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
545 		if (error == 0 && *(u_int *)data == 0)
546 			error = ENOENT;
547 		break;
548 	case DIOCGFWHEADS:
549 		error = g_io_getattr("GEOM::fwheads", cp, &i, data);
550 		if (error == 0 && *(u_int *)data == 0)
551 			error = ENOENT;
552 		break;
553 #ifdef COMPAT_FREEBSD11
554 	case DIOCSKERNELDUMP_FREEBSD11:
555 	    {
556 		struct diocskerneldump_arg kda;
557 
558 		gone_in(13, "FreeBSD 11.x ABI compat");
559 
560 		bzero(&kda, sizeof(kda));
561 		kda.kda_encryption = KERNELDUMP_ENC_NONE;
562 		kda.kda_index = (*(u_int *)data ? 0 : KDA_REMOVE_ALL);
563 		if (kda.kda_index == KDA_REMOVE_ALL)
564 			error = dumper_remove(devtoname(dev), &kda);
565 		else
566 			error = g_dev_setdumpdev(dev, &kda);
567 		break;
568 	    }
569 #endif
570 #ifdef COMPAT_FREEBSD12
571 	case DIOCSKERNELDUMP_FREEBSD12:
572 	    {
573 		struct diocskerneldump_arg_freebsd12 *kda12;
574 
575 		gone_in(14, "FreeBSD 12.x ABI compat");
576 
577 		kda12 = (void *)data;
578 		memcpy(&kda_copy, kda12, sizeof(kda_copy));
579 		kda_copy.kda_index = (kda12->kda12_enable ?
580 		    0 : KDA_REMOVE_ALL);
581 
582 		explicit_bzero(kda12, sizeof(*kda12));
583 		/* Kludge to pass kda_copy to kda in fallthrough. */
584 		data = (void *)&kda_copy;
585 	    }
586 	    /* FALLTHROUGH */
587 #endif
588 	case DIOCSKERNELDUMP:
589 	    {
590 		struct diocskerneldump_arg *kda;
591 		uint8_t *encryptedkey;
592 
593 		kda = (struct diocskerneldump_arg *)data;
594 		if (kda->kda_index == KDA_REMOVE_ALL ||
595 		    kda->kda_index == KDA_REMOVE_DEV ||
596 		    kda->kda_index == KDA_REMOVE) {
597 			error = dumper_remove(devtoname(dev), kda);
598 			explicit_bzero(kda, sizeof(*kda));
599 			break;
600 		}
601 
602 		if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
603 			if (kda->kda_encryptedkeysize == 0 ||
604 			    kda->kda_encryptedkeysize >
605 			    KERNELDUMP_ENCKEY_MAX_SIZE) {
606 				explicit_bzero(kda, sizeof(*kda));
607 				return (EINVAL);
608 			}
609 			encryptedkey = malloc(kda->kda_encryptedkeysize, M_TEMP,
610 			    M_WAITOK);
611 			error = copyin(kda->kda_encryptedkey, encryptedkey,
612 			    kda->kda_encryptedkeysize);
613 		} else {
614 			encryptedkey = NULL;
615 		}
616 		if (error == 0) {
617 			kda->kda_encryptedkey = encryptedkey;
618 			error = g_dev_setdumpdev(dev, kda);
619 		}
620 		zfree(encryptedkey, M_TEMP);
621 		explicit_bzero(kda, sizeof(*kda));
622 		break;
623 	    }
624 	case DIOCGFLUSH:
625 		error = g_io_flush(cp);
626 		break;
627 	case DIOCGDELETE:
628 		offset = ((off_t *)data)[0];
629 		length = ((off_t *)data)[1];
630 		if ((offset % pp->sectorsize) != 0 ||
631 		    (length % pp->sectorsize) != 0 || length <= 0) {
632 			printf("%s: offset=%jd length=%jd\n", __func__, offset,
633 			    length);
634 			error = EINVAL;
635 			break;
636 		}
637 		if ((pp->mediasize > 0) && (offset >= pp->mediasize)) {
638 			/*
639 			 * Catch out-of-bounds requests here. The problem is
640 			 * that due to historical GEOM I/O implementation
641 			 * peculatities, g_delete_data() would always return
642 			 * success for requests starting just the next byte
643 			 * after providers media boundary. Condition check on
644 			 * non-zero media size, since that condition would
645 			 * (most likely) cause ENXIO instead.
646 			 */
647 			error = EIO;
648 			break;
649 		}
650 		while (length > 0) {
651 			chunk = length;
652 			if (g_dev_del_max_sectors != 0 &&
653 			    chunk > g_dev_del_max_sectors * pp->sectorsize) {
654 				chunk = g_dev_del_max_sectors * pp->sectorsize;
655 				if (pp->stripesize > 0) {
656 					odd = (offset + chunk +
657 					    pp->stripeoffset) % pp->stripesize;
658 					if (chunk > odd)
659 						chunk -= odd;
660 				}
661 			}
662 			error = g_delete_data(cp, offset, chunk);
663 			length -= chunk;
664 			offset += chunk;
665 			if (error)
666 				break;
667 			/*
668 			 * Since the request size can be large, the service
669 			 * time can be is likewise.  We make this ioctl
670 			 * interruptible by checking for signals for each bio.
671 			 */
672 			if (SIGPENDING(td))
673 				break;
674 		}
675 		break;
676 	case DIOCGIDENT:
677 		error = g_io_getattr("GEOM::ident", cp, &i, data);
678 		break;
679 	case DIOCGPROVIDERNAME:
680 		strlcpy(data, pp->name, i);
681 		break;
682 	case DIOCGSTRIPESIZE:
683 		*(off_t *)data = pp->stripesize;
684 		break;
685 	case DIOCGSTRIPEOFFSET:
686 		*(off_t *)data = pp->stripeoffset;
687 		break;
688 	case DIOCGPHYSPATH:
689 		error = g_io_getattr("GEOM::physpath", cp, &i, data);
690 		if (error == 0 && *(char *)data == '\0')
691 			error = ENOENT;
692 		break;
693 	case DIOCGATTR: {
694 		struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
695 
696 		if (arg->len > sizeof(arg->value)) {
697 			error = EINVAL;
698 			break;
699 		}
700 		error = g_io_getattr(arg->name, cp, &arg->len, &arg->value);
701 		break;
702 	}
703 	case DIOCZONECMD: {
704 		struct disk_zone_args *zone_args =(struct disk_zone_args *)data;
705 		struct disk_zone_rep_entry *new_entries, *old_entries;
706 		struct disk_zone_report *rep;
707 		size_t alloc_size;
708 
709 		old_entries = NULL;
710 		new_entries = NULL;
711 		rep = NULL;
712 		alloc_size = 0;
713 
714 		if (zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES) {
715 			rep = &zone_args->zone_params.report;
716 #define	MAXENTRIES	(maxphys / sizeof(struct disk_zone_rep_entry))
717 			if (rep->entries_allocated > MAXENTRIES)
718 				rep->entries_allocated = MAXENTRIES;
719 			alloc_size = rep->entries_allocated *
720 			    sizeof(struct disk_zone_rep_entry);
721 			if (alloc_size != 0)
722 				new_entries = g_malloc(alloc_size,
723 				    M_WAITOK | M_ZERO);
724 			old_entries = rep->entries;
725 			rep->entries = new_entries;
726 		}
727 		error = g_io_zonecmd(zone_args, cp);
728 		if (zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES &&
729 		    alloc_size != 0 && error == 0)
730 			error = copyout(new_entries, old_entries, alloc_size);
731 		if (old_entries != NULL && rep != NULL)
732 			rep->entries = old_entries;
733 		if (new_entries != NULL)
734 			g_free(new_entries);
735 		break;
736 	}
737 	default:
738 		if (pp->geom->ioctl != NULL) {
739 			error = pp->geom->ioctl(pp, cmd, data, fflag, td);
740 		} else {
741 			error = ENOIOCTL;
742 		}
743 	}
744 
745 	return (error);
746 }
747 
748 static void
749 g_dev_done(struct bio *bp2)
750 {
751 	struct g_consumer *cp;
752 	struct g_dev_softc *sc;
753 	struct bio *bp;
754 	int active;
755 
756 	cp = bp2->bio_from;
757 	sc = cp->private;
758 	bp = bp2->bio_parent;
759 	bp->bio_error = bp2->bio_error;
760 	bp->bio_completed = bp2->bio_completed;
761 	bp->bio_resid = bp->bio_length - bp2->bio_completed;
762 	if (bp2->bio_cmd == BIO_ZONE)
763 		bcopy(&bp2->bio_zone, &bp->bio_zone, sizeof(bp->bio_zone));
764 
765 	if (bp2->bio_error != 0) {
766 		g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
767 		    bp2, bp2->bio_error);
768 		bp->bio_flags |= BIO_ERROR;
769 	} else {
770 		g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
771 		    bp2, bp, bp2->bio_resid, (intmax_t)bp2->bio_completed);
772 	}
773 	g_destroy_bio(bp2);
774 	active = atomic_fetchadd_int(&sc->sc_active, -1) - 1;
775 	if ((active & SC_A_ACTIVE) == 0) {
776 		if ((active & SC_A_OPEN) == 0)
777 			wakeup(&sc->sc_active);
778 		if (active & SC_A_DESTROY)
779 			g_post_event(g_dev_destroy, cp, M_NOWAIT, NULL);
780 	}
781 	biodone(bp);
782 }
783 
784 static void
785 g_dev_strategy(struct bio *bp)
786 {
787 	struct g_consumer *cp;
788 	struct bio *bp2;
789 	struct cdev *dev;
790 	struct g_dev_softc *sc;
791 
792 	KASSERT(bp->bio_cmd == BIO_READ ||
793 	        bp->bio_cmd == BIO_WRITE ||
794 	        bp->bio_cmd == BIO_DELETE ||
795 		bp->bio_cmd == BIO_FLUSH ||
796 		bp->bio_cmd == BIO_ZONE,
797 		("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
798 	dev = bp->bio_dev;
799 	cp = dev->si_drv2;
800 	KASSERT(cp->acr || cp->acw,
801 	    ("Consumer with zero access count in g_dev_strategy"));
802 	biotrack(bp, __func__);
803 #ifdef INVARIANTS
804 	if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
805 	    (bp->bio_bcount % cp->provider->sectorsize) != 0) {
806 		bp->bio_resid = bp->bio_bcount;
807 		biofinish(bp, NULL, EINVAL);
808 		return;
809 	}
810 #endif
811 	sc = dev->si_drv1;
812 	KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
813 	atomic_add_int(&sc->sc_active, 1);
814 
815 	for (;;) {
816 		/*
817 		 * XXX: This is not an ideal solution, but I believe it to
818 		 * XXX: deadlock safely, all things considered.
819 		 */
820 		bp2 = g_clone_bio(bp);
821 		if (bp2 != NULL)
822 			break;
823 		pause("gdstrat", hz / 10);
824 	}
825 	KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
826 	bp2->bio_done = g_dev_done;
827 	g_trace(G_T_BIO,
828 	    "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
829 	    bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
830 	    bp2->bio_data, bp2->bio_cmd);
831 	g_io_request(bp2, cp);
832 	KASSERT(cp->acr || cp->acw,
833 	    ("g_dev_strategy raced with g_dev_close and lost"));
834 
835 }
836 
837 /*
838  * g_dev_callback()
839  *
840  * Called by devfs when asynchronous device destruction is completed.
841  * - Mark that we have no attached device any more.
842  * - If there are no outstanding requests, schedule geom destruction.
843  *   Otherwise destruction will be scheduled later by g_dev_done().
844  */
845 
846 static void
847 g_dev_callback(void *arg)
848 {
849 	struct g_consumer *cp;
850 	struct g_dev_softc *sc;
851 	int active;
852 
853 	cp = arg;
854 	sc = cp->private;
855 	g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
856 
857 	sc->sc_dev = NULL;
858 	sc->sc_alias = NULL;
859 	active = atomic_fetchadd_int(&sc->sc_active, SC_A_DESTROY);
860 	if ((active & SC_A_ACTIVE) == 0)
861 		g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
862 }
863 
864 /*
865  * g_dev_orphan()
866  *
867  * Called from below when the provider orphaned us.
868  * - Clear any dump settings.
869  * - Request asynchronous device destruction to prevent any more requests
870  *   from coming in.  The provider is already marked with an error, so
871  *   anything which comes in the interim will be returned immediately.
872  */
873 
874 static void
875 g_dev_orphan(struct g_consumer *cp)
876 {
877 	struct cdev *dev;
878 	struct g_dev_softc *sc;
879 
880 	g_topology_assert();
881 	sc = cp->private;
882 	dev = sc->sc_dev;
883 	g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
884 
885 	/* Reset any dump-area set on this device */
886 	if (dev->si_flags & SI_DUMPDEV) {
887 		struct diocskerneldump_arg kda;
888 
889 		bzero(&kda, sizeof(kda));
890 		kda.kda_index = KDA_REMOVE_DEV;
891 		(void)dumper_remove(devtoname(dev), &kda);
892 	}
893 
894 	/* Destroy the struct cdev *so we get no more requests */
895 	delist_dev(dev);
896 	destroy_dev_sched_cb(dev, g_dev_callback, cp);
897 }
898 
899 DECLARE_GEOM_CLASS(g_dev_class, g_dev);
900