xref: /freebsd/sys/geom/geom_dev.c (revision 9e5787d2284e187abb5b654d924394a65772e004)
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_f("GEOM", "DEV", "DESTROY", buf, M_WAITOK);
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_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
281 	devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
282 	dev = sc->sc_alias;
283 	if (dev != NULL) {
284 		snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
285 		devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
286 		devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
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_f("GEOM", "DEV", "SIZECHANGE", buf, M_WAITOK);
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 	KASSERT(error == 0,
350 	    ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
351 
352 	make_dev_args_init(&args);
353 	args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
354 	args.mda_devsw = &g_dev_cdevsw;
355 	args.mda_cr = NULL;
356 	args.mda_uid = UID_ROOT;
357 	args.mda_gid = GID_OPERATOR;
358 	args.mda_mode = 0640;
359 	args.mda_si_drv1 = sc;
360 	args.mda_si_drv2 = cp;
361 	error = make_dev_s(&args, &sc->sc_dev, "%s", gp->name);
362 	if (error != 0) {
363 		printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
364 		    __func__, gp->name, error);
365 		g_detach(cp);
366 		g_destroy_consumer(cp);
367 		g_destroy_geom(gp);
368 		mtx_destroy(&sc->sc_mtx);
369 		g_free(sc);
370 		return (NULL);
371 	}
372 	dev = sc->sc_dev;
373 	dev->si_flags |= SI_UNMAPPED;
374 	dev->si_iosize_max = MAXPHYS;
375 	error = init_dumpdev(dev);
376 	if (error != 0)
377 		printf("%s: init_dumpdev() failed (gp->name=%s, error=%d)\n",
378 		    __func__, gp->name, error);
379 
380 	g_dev_attrchanged(cp, "GEOM::physpath");
381 	snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
382 	devctl_notify_f("GEOM", "DEV", "CREATE", buf, M_WAITOK);
383 	/*
384 	 * Now add all the aliases for this drive
385 	 */
386 	LIST_FOREACH(gap, &pp->aliases, ga_next) {
387 		error = make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &adev, dev,
388 		    "%s", gap->ga_alias);
389 		if (error) {
390 			printf("%s: make_dev_alias_p() failed (name=%s, error=%d)\n",
391 			    __func__, gap->ga_alias, error);
392 			continue;
393 		}
394 		snprintf(buf, sizeof(buf), "cdev=%s", gap->ga_alias);
395 		devctl_notify_f("GEOM", "DEV", "CREATE", buf, M_WAITOK);
396 	}
397 
398 	return (gp);
399 }
400 
401 static int
402 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
403 {
404 	struct g_consumer *cp;
405 	struct g_dev_softc *sc;
406 	int error, r, w, e;
407 
408 	cp = dev->si_drv2;
409 	g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
410 	    cp->geom->name, flags, fmt, td);
411 
412 	r = flags & FREAD ? 1 : 0;
413 	w = flags & FWRITE ? 1 : 0;
414 #ifdef notyet
415 	e = flags & O_EXCL ? 1 : 0;
416 #else
417 	e = 0;
418 #endif
419 
420 	/*
421 	 * This happens on attempt to open a device node with O_EXEC.
422 	 */
423 	if (r + w + e == 0)
424 		return (EINVAL);
425 
426 	if (w) {
427 		/*
428 		 * When running in very secure mode, do not allow
429 		 * opens for writing of any disks.
430 		 */
431 		error = securelevel_ge(td->td_ucred, 2);
432 		if (error)
433 			return (error);
434 	}
435 	g_topology_lock();
436 	error = g_access(cp, r, w, e);
437 	g_topology_unlock();
438 	if (error == 0) {
439 		sc = dev->si_drv1;
440 		mtx_lock(&sc->sc_mtx);
441 		if (sc->sc_open == 0 && (sc->sc_active & SC_A_ACTIVE) != 0)
442 			wakeup(&sc->sc_active);
443 		sc->sc_open += r + w + e;
444 		if (sc->sc_open == 0)
445 			atomic_clear_int(&sc->sc_active, SC_A_OPEN);
446 		else
447 			atomic_set_int(&sc->sc_active, SC_A_OPEN);
448 		mtx_unlock(&sc->sc_mtx);
449 	}
450 	return (error);
451 }
452 
453 static int
454 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
455 {
456 	struct g_consumer *cp;
457 	struct g_dev_softc *sc;
458 	int error, r, w, e;
459 
460 	cp = dev->si_drv2;
461 	g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
462 	    cp->geom->name, flags, fmt, td);
463 
464 	r = flags & FREAD ? -1 : 0;
465 	w = flags & FWRITE ? -1 : 0;
466 #ifdef notyet
467 	e = flags & O_EXCL ? -1 : 0;
468 #else
469 	e = 0;
470 #endif
471 
472 	/*
473 	 * The vgonel(9) - caused by eg. forced unmount of devfs - calls
474 	 * VOP_CLOSE(9) on devfs vnode without any FREAD or FWRITE flags,
475 	 * which would result in zero deltas, which in turn would cause
476 	 * panic in g_access(9).
477 	 *
478 	 * Note that we cannot zero the counters (ie. do "r = cp->acr"
479 	 * etc) instead, because the consumer might be opened in another
480 	 * devfs instance.
481 	 */
482 	if (r + w + e == 0)
483 		return (EINVAL);
484 
485 	sc = dev->si_drv1;
486 	mtx_lock(&sc->sc_mtx);
487 	sc->sc_open += r + w + e;
488 	if (sc->sc_open == 0)
489 		atomic_clear_int(&sc->sc_active, SC_A_OPEN);
490 	else
491 		atomic_set_int(&sc->sc_active, SC_A_OPEN);
492 	while (sc->sc_open == 0 && (sc->sc_active & SC_A_ACTIVE) != 0)
493 		msleep(&sc->sc_active, &sc->sc_mtx, 0, "g_dev_close", hz / 10);
494 	mtx_unlock(&sc->sc_mtx);
495 	g_topology_lock();
496 	error = g_access(cp, r, w, e);
497 	g_topology_unlock();
498 	return (error);
499 }
500 
501 static int
502 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
503 {
504 	struct g_consumer *cp;
505 	struct g_provider *pp;
506 	off_t offset, length, chunk, odd;
507 	int i, error;
508 #ifdef COMPAT_FREEBSD12
509 	struct diocskerneldump_arg kda_copy;
510 #endif
511 
512 	cp = dev->si_drv2;
513 	pp = cp->provider;
514 
515 	/* If consumer or provider is dying, don't disturb. */
516 	if (cp->flags & G_CF_ORPHAN)
517 		return (ENXIO);
518 	if (pp->error)
519 		return (pp->error);
520 
521 	error = 0;
522 	KASSERT(cp->acr || cp->acw,
523 	    ("Consumer with zero access count in g_dev_ioctl"));
524 
525 	i = IOCPARM_LEN(cmd);
526 	switch (cmd) {
527 	case DIOCGSECTORSIZE:
528 		*(u_int *)data = pp->sectorsize;
529 		if (*(u_int *)data == 0)
530 			error = ENOENT;
531 		break;
532 	case DIOCGMEDIASIZE:
533 		*(off_t *)data = pp->mediasize;
534 		if (*(off_t *)data == 0)
535 			error = ENOENT;
536 		break;
537 	case DIOCGFWSECTORS:
538 		error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
539 		if (error == 0 && *(u_int *)data == 0)
540 			error = ENOENT;
541 		break;
542 	case DIOCGFWHEADS:
543 		error = g_io_getattr("GEOM::fwheads", cp, &i, data);
544 		if (error == 0 && *(u_int *)data == 0)
545 			error = ENOENT;
546 		break;
547 	case DIOCGFRONTSTUFF:
548 		error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
549 		break;
550 #ifdef COMPAT_FREEBSD11
551 	case DIOCSKERNELDUMP_FREEBSD11:
552 	    {
553 		struct diocskerneldump_arg kda;
554 
555 		gone_in(13, "FreeBSD 11.x ABI compat");
556 
557 		bzero(&kda, sizeof(kda));
558 		kda.kda_encryption = KERNELDUMP_ENC_NONE;
559 		kda.kda_index = (*(u_int *)data ? 0 : KDA_REMOVE_ALL);
560 		if (kda.kda_index == KDA_REMOVE_ALL)
561 			error = dumper_remove(devtoname(dev), &kda);
562 		else
563 			error = g_dev_setdumpdev(dev, &kda);
564 		break;
565 	    }
566 #endif
567 #ifdef COMPAT_FREEBSD12
568 	case DIOCSKERNELDUMP_FREEBSD12:
569 	    {
570 		struct diocskerneldump_arg_freebsd12 *kda12;
571 
572 		gone_in(14, "FreeBSD 12.x ABI compat");
573 
574 		kda12 = (void *)data;
575 		memcpy(&kda_copy, kda12, sizeof(kda_copy));
576 		kda_copy.kda_index = (kda12->kda12_enable ?
577 		    0 : KDA_REMOVE_ALL);
578 
579 		explicit_bzero(kda12, sizeof(*kda12));
580 		/* Kludge to pass kda_copy to kda in fallthrough. */
581 		data = (void *)&kda_copy;
582 	    }
583 	    /* FALLTHROUGH */
584 #endif
585 	case DIOCSKERNELDUMP:
586 	    {
587 		struct diocskerneldump_arg *kda;
588 		uint8_t *encryptedkey;
589 
590 		kda = (struct diocskerneldump_arg *)data;
591 		if (kda->kda_index == KDA_REMOVE_ALL ||
592 		    kda->kda_index == KDA_REMOVE_DEV ||
593 		    kda->kda_index == KDA_REMOVE) {
594 			error = dumper_remove(devtoname(dev), kda);
595 			explicit_bzero(kda, sizeof(*kda));
596 			break;
597 		}
598 
599 		if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
600 			if (kda->kda_encryptedkeysize == 0 ||
601 			    kda->kda_encryptedkeysize >
602 			    KERNELDUMP_ENCKEY_MAX_SIZE) {
603 				explicit_bzero(kda, sizeof(*kda));
604 				return (EINVAL);
605 			}
606 			encryptedkey = malloc(kda->kda_encryptedkeysize, M_TEMP,
607 			    M_WAITOK);
608 			error = copyin(kda->kda_encryptedkey, encryptedkey,
609 			    kda->kda_encryptedkeysize);
610 		} else {
611 			encryptedkey = NULL;
612 		}
613 		if (error == 0) {
614 			kda->kda_encryptedkey = encryptedkey;
615 			error = g_dev_setdumpdev(dev, kda);
616 		}
617 		zfree(encryptedkey, M_TEMP);
618 		explicit_bzero(kda, sizeof(*kda));
619 		break;
620 	    }
621 	case DIOCGFLUSH:
622 		error = g_io_flush(cp);
623 		break;
624 	case DIOCGDELETE:
625 		offset = ((off_t *)data)[0];
626 		length = ((off_t *)data)[1];
627 		if ((offset % pp->sectorsize) != 0 ||
628 		    (length % pp->sectorsize) != 0 || length <= 0) {
629 			printf("%s: offset=%jd length=%jd\n", __func__, offset,
630 			    length);
631 			error = EINVAL;
632 			break;
633 		}
634 		if ((pp->mediasize > 0) && (offset >= pp->mediasize)) {
635 			/*
636 			 * Catch out-of-bounds requests here. The problem is
637 			 * that due to historical GEOM I/O implementation
638 			 * peculatities, g_delete_data() would always return
639 			 * success for requests starting just the next byte
640 			 * after providers media boundary. Condition check on
641 			 * non-zero media size, since that condition would
642 			 * (most likely) cause ENXIO instead.
643 			 */
644 			error = EIO;
645 			break;
646 		}
647 		while (length > 0) {
648 			chunk = length;
649 			if (g_dev_del_max_sectors != 0 &&
650 			    chunk > g_dev_del_max_sectors * pp->sectorsize) {
651 				chunk = g_dev_del_max_sectors * pp->sectorsize;
652 				if (pp->stripesize > 0) {
653 					odd = (offset + chunk +
654 					    pp->stripeoffset) % pp->stripesize;
655 					if (chunk > odd)
656 						chunk -= odd;
657 				}
658 			}
659 			error = g_delete_data(cp, offset, chunk);
660 			length -= chunk;
661 			offset += chunk;
662 			if (error)
663 				break;
664 			/*
665 			 * Since the request size can be large, the service
666 			 * time can be is likewise.  We make this ioctl
667 			 * interruptible by checking for signals for each bio.
668 			 */
669 			if (SIGPENDING(td))
670 				break;
671 		}
672 		break;
673 	case DIOCGIDENT:
674 		error = g_io_getattr("GEOM::ident", cp, &i, data);
675 		break;
676 	case DIOCGPROVIDERNAME:
677 		strlcpy(data, pp->name, i);
678 		break;
679 	case DIOCGSTRIPESIZE:
680 		*(off_t *)data = pp->stripesize;
681 		break;
682 	case DIOCGSTRIPEOFFSET:
683 		*(off_t *)data = pp->stripeoffset;
684 		break;
685 	case DIOCGPHYSPATH:
686 		error = g_io_getattr("GEOM::physpath", cp, &i, data);
687 		if (error == 0 && *(char *)data == '\0')
688 			error = ENOENT;
689 		break;
690 	case DIOCGATTR: {
691 		struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
692 
693 		if (arg->len > sizeof(arg->value)) {
694 			error = EINVAL;
695 			break;
696 		}
697 		error = g_io_getattr(arg->name, cp, &arg->len, &arg->value);
698 		break;
699 	}
700 	case DIOCZONECMD: {
701 		struct disk_zone_args *zone_args =(struct disk_zone_args *)data;
702 		struct disk_zone_rep_entry *new_entries, *old_entries;
703 		struct disk_zone_report *rep;
704 		size_t alloc_size;
705 
706 		old_entries = NULL;
707 		new_entries = NULL;
708 		rep = NULL;
709 		alloc_size = 0;
710 
711 		if (zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES) {
712 			rep = &zone_args->zone_params.report;
713 #define	MAXENTRIES	(MAXPHYS / sizeof(struct disk_zone_rep_entry))
714 			if (rep->entries_allocated > MAXENTRIES)
715 				rep->entries_allocated = MAXENTRIES;
716 			alloc_size = rep->entries_allocated *
717 			    sizeof(struct disk_zone_rep_entry);
718 			if (alloc_size != 0)
719 				new_entries = g_malloc(alloc_size,
720 				    M_WAITOK| M_ZERO);
721 			old_entries = rep->entries;
722 			rep->entries = new_entries;
723 		}
724 		error = g_io_zonecmd(zone_args, cp);
725 		if (zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES &&
726 		    alloc_size != 0 && error == 0)
727 			error = copyout(new_entries, old_entries, alloc_size);
728 		if (old_entries != NULL && rep != NULL)
729 			rep->entries = old_entries;
730 		if (new_entries != NULL)
731 			g_free(new_entries);
732 		break;
733 	}
734 	default:
735 		if (pp->geom->ioctl != NULL) {
736 			error = pp->geom->ioctl(pp, cmd, data, fflag, td);
737 		} else {
738 			error = ENOIOCTL;
739 		}
740 	}
741 
742 	return (error);
743 }
744 
745 static void
746 g_dev_done(struct bio *bp2)
747 {
748 	struct g_consumer *cp;
749 	struct g_dev_softc *sc;
750 	struct bio *bp;
751 	int active;
752 
753 	cp = bp2->bio_from;
754 	sc = cp->private;
755 	bp = bp2->bio_parent;
756 	bp->bio_error = bp2->bio_error;
757 	bp->bio_completed = bp2->bio_completed;
758 	bp->bio_resid = bp->bio_length - bp2->bio_completed;
759 	if (bp2->bio_cmd == BIO_ZONE)
760 		bcopy(&bp2->bio_zone, &bp->bio_zone, sizeof(bp->bio_zone));
761 
762 	if (bp2->bio_error != 0) {
763 		g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
764 		    bp2, bp2->bio_error);
765 		bp->bio_flags |= BIO_ERROR;
766 	} else {
767 		g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
768 		    bp2, bp, bp2->bio_resid, (intmax_t)bp2->bio_completed);
769 	}
770 	g_destroy_bio(bp2);
771 	active = atomic_fetchadd_int(&sc->sc_active, -1) - 1;
772 	if ((active & SC_A_ACTIVE) == 0) {
773 		if ((active & SC_A_OPEN) == 0)
774 			wakeup(&sc->sc_active);
775 		if (active & SC_A_DESTROY)
776 			g_post_event(g_dev_destroy, cp, M_NOWAIT, NULL);
777 	}
778 	biodone(bp);
779 }
780 
781 static void
782 g_dev_strategy(struct bio *bp)
783 {
784 	struct g_consumer *cp;
785 	struct bio *bp2;
786 	struct cdev *dev;
787 	struct g_dev_softc *sc;
788 
789 	KASSERT(bp->bio_cmd == BIO_READ ||
790 	        bp->bio_cmd == BIO_WRITE ||
791 	        bp->bio_cmd == BIO_DELETE ||
792 		bp->bio_cmd == BIO_FLUSH ||
793 		bp->bio_cmd == BIO_ZONE,
794 		("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
795 	dev = bp->bio_dev;
796 	cp = dev->si_drv2;
797 	KASSERT(cp->acr || cp->acw,
798 	    ("Consumer with zero access count in g_dev_strategy"));
799 	biotrack(bp, __func__);
800 #ifdef INVARIANTS
801 	if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
802 	    (bp->bio_bcount % cp->provider->sectorsize) != 0) {
803 		bp->bio_resid = bp->bio_bcount;
804 		biofinish(bp, NULL, EINVAL);
805 		return;
806 	}
807 #endif
808 	sc = dev->si_drv1;
809 	KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
810 	atomic_add_int(&sc->sc_active, 1);
811 
812 	for (;;) {
813 		/*
814 		 * XXX: This is not an ideal solution, but I believe it to
815 		 * XXX: deadlock safely, all things considered.
816 		 */
817 		bp2 = g_clone_bio(bp);
818 		if (bp2 != NULL)
819 			break;
820 		pause("gdstrat", hz / 10);
821 	}
822 	KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
823 	bp2->bio_done = g_dev_done;
824 	g_trace(G_T_BIO,
825 	    "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
826 	    bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
827 	    bp2->bio_data, bp2->bio_cmd);
828 	g_io_request(bp2, cp);
829 	KASSERT(cp->acr || cp->acw,
830 	    ("g_dev_strategy raced with g_dev_close and lost"));
831 
832 }
833 
834 /*
835  * g_dev_callback()
836  *
837  * Called by devfs when asynchronous device destruction is completed.
838  * - Mark that we have no attached device any more.
839  * - If there are no outstanding requests, schedule geom destruction.
840  *   Otherwise destruction will be scheduled later by g_dev_done().
841  */
842 
843 static void
844 g_dev_callback(void *arg)
845 {
846 	struct g_consumer *cp;
847 	struct g_dev_softc *sc;
848 	int active;
849 
850 	cp = arg;
851 	sc = cp->private;
852 	g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
853 
854 	sc->sc_dev = NULL;
855 	sc->sc_alias = NULL;
856 	active = atomic_fetchadd_int(&sc->sc_active, SC_A_DESTROY);
857 	if ((active & SC_A_ACTIVE) == 0)
858 		g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
859 }
860 
861 /*
862  * g_dev_orphan()
863  *
864  * Called from below when the provider orphaned us.
865  * - Clear any dump settings.
866  * - Request asynchronous device destruction to prevent any more requests
867  *   from coming in.  The provider is already marked with an error, so
868  *   anything which comes in the interim will be returned immediately.
869  */
870 
871 static void
872 g_dev_orphan(struct g_consumer *cp)
873 {
874 	struct cdev *dev;
875 	struct g_dev_softc *sc;
876 
877 	g_topology_assert();
878 	sc = cp->private;
879 	dev = sc->sc_dev;
880 	g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
881 
882 	/* Reset any dump-area set on this device */
883 	if (dev->si_flags & SI_DUMPDEV) {
884 		struct diocskerneldump_arg kda;
885 
886 		bzero(&kda, sizeof(kda));
887 		kda.kda_index = KDA_REMOVE_DEV;
888 		(void)dumper_remove(devtoname(dev), &kda);
889 	}
890 
891 	/* Destroy the struct cdev *so we get no more requests */
892 	delist_dev(dev);
893 	destroy_dev_sched_cb(dev, g_dev_callback, cp);
894 }
895 
896 DECLARE_GEOM_CLASS(g_dev_class, g_dev);
897