xref: /freebsd/sys/geom/geom_dev.c (revision 4c1a82cea504df7a79f5bd8f7d0a41cacccff16e)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2002 Poul-Henning Kamp
5  * Copyright (c) 2002 Networks Associates Technology, Inc.
6  * All rights reserved.
7  *
8  * This software was developed for the FreeBSD Project by Poul-Henning Kamp
9  * and NAI Labs, the Security Research Division of Network Associates, Inc.
10  * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
11  * DARPA CHATS research program.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. The names of the authors may not be used to endorse or promote
22  *    products derived from this software without specific prior written
23  *    permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/malloc.h>
44 #include <sys/kernel.h>
45 #include <sys/conf.h>
46 #include <sys/ctype.h>
47 #include <sys/bio.h>
48 #include <sys/bus.h>
49 #include <sys/lock.h>
50 #include <sys/mutex.h>
51 #include <sys/proc.h>
52 #include <sys/errno.h>
53 #include <sys/time.h>
54 #include <sys/disk.h>
55 #include <sys/fcntl.h>
56 #include <sys/limits.h>
57 #include <sys/sysctl.h>
58 #include <geom/geom.h>
59 #include <geom/geom_int.h>
60 #include <machine/stdarg.h>
61 
62 struct g_dev_softc {
63 	struct mtx	 sc_mtx;
64 	struct cdev	*sc_dev;
65 	struct cdev	*sc_alias;
66 	int		 sc_open;
67 	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, 0, "GEOM_DEV stuff");
116 SYSCTL_QUAD(_kern_geom_dev, OID_AUTO, delete_max_sectors, CTLFLAG_RW,
117     &g_dev_del_max_sectors, 0, "Maximum number of sectors in a single "
118     "delete request sent to the provider. Larger requests are chunked "
119     "so they can be interrupted. (0 = disable chunking)");
120 
121 static char *dumpdev = NULL;
122 static void
123 g_dev_init(struct g_class *mp)
124 {
125 
126 	dumpdev = kern_getenv("dumpdev");
127 }
128 
129 static void
130 g_dev_fini(struct g_class *mp)
131 {
132 
133 	freeenv(dumpdev);
134 	dumpdev = NULL;
135 }
136 
137 static int
138 g_dev_setdumpdev(struct cdev *dev, struct diocskerneldump_arg *kda)
139 {
140 	struct g_kerneldump kd;
141 	struct g_consumer *cp;
142 	int error, len;
143 
144 	MPASS(dev != NULL && kda != NULL);
145 	MPASS(kda->kda_index != KDA_REMOVE);
146 
147 	cp = dev->si_drv2;
148 	len = sizeof(kd);
149 	memset(&kd, 0, len);
150 	kd.offset = 0;
151 	kd.length = OFF_MAX;
152 	error = g_io_getattr("GEOM::kerneldump", cp, &len, &kd);
153 	if (error != 0)
154 		return (error);
155 
156 	error = dumper_insert(&kd.di, devtoname(dev), kda);
157 	if (error == 0)
158 		dev->si_flags |= SI_DUMPDEV;
159 
160 	return (error);
161 }
162 
163 static int
164 init_dumpdev(struct cdev *dev)
165 {
166 	struct diocskerneldump_arg kda;
167 	struct g_consumer *cp;
168 	const char *devprefix = "/dev/", *devname;
169 	int error;
170 	size_t len;
171 
172 	bzero(&kda, sizeof(kda));
173 	kda.kda_index = KDA_APPEND;
174 
175 	if (dumpdev == NULL)
176 		return (0);
177 
178 	len = strlen(devprefix);
179 	devname = devtoname(dev);
180 	if (strcmp(devname, dumpdev) != 0 &&
181 	   (strncmp(dumpdev, devprefix, len) != 0 ||
182 	    strcmp(devname, dumpdev + len) != 0))
183 		return (0);
184 
185 	cp = (struct g_consumer *)dev->si_drv2;
186 	error = g_access(cp, 1, 0, 0);
187 	if (error != 0)
188 		return (error);
189 
190 	error = g_dev_setdumpdev(dev, &kda);
191 	if (error == 0) {
192 		freeenv(dumpdev);
193 		dumpdev = NULL;
194 	}
195 
196 	(void)g_access(cp, -1, 0, 0);
197 
198 	return (error);
199 }
200 
201 static void
202 g_dev_destroy(void *arg, int flags __unused)
203 {
204 	struct g_consumer *cp;
205 	struct g_geom *gp;
206 	struct g_dev_softc *sc;
207 	char buf[SPECNAMELEN + 6];
208 
209 	g_topology_assert();
210 	cp = arg;
211 	gp = cp->geom;
212 	sc = cp->private;
213 	g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name);
214 	snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
215 	devctl_notify_f("GEOM", "DEV", "DESTROY", buf, M_WAITOK);
216 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
217 		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
218 	g_detach(cp);
219 	g_destroy_consumer(cp);
220 	g_destroy_geom(gp);
221 	mtx_destroy(&sc->sc_mtx);
222 	g_free(sc);
223 }
224 
225 void
226 g_dev_print(void)
227 {
228 	struct g_geom *gp;
229 	char const *p = "";
230 
231 	LIST_FOREACH(gp, &g_dev_class.geom, geom) {
232 		printf("%s%s", p, gp->name);
233 		p = " ";
234 	}
235 	printf("\n");
236 }
237 
238 static void
239 g_dev_set_physpath(struct g_consumer *cp)
240 {
241 	struct g_dev_softc *sc;
242 	char *physpath;
243 	int error, physpath_len;
244 
245 	if (g_access(cp, 1, 0, 0) != 0)
246 		return;
247 
248 	sc = cp->private;
249 	physpath_len = MAXPATHLEN;
250 	physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
251 	error = g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath);
252 	g_access(cp, -1, 0, 0);
253 	if (error == 0 && strlen(physpath) != 0) {
254 		struct cdev *dev, *old_alias_dev;
255 		struct cdev **alias_devp;
256 
257 		dev = sc->sc_dev;
258 		old_alias_dev = sc->sc_alias;
259 		alias_devp = (struct cdev **)&sc->sc_alias;
260 		make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp, dev,
261 		    old_alias_dev, physpath);
262 	} else if (sc->sc_alias) {
263 		destroy_dev((struct cdev *)sc->sc_alias);
264 		sc->sc_alias = NULL;
265 	}
266 	g_free(physpath);
267 }
268 
269 static void
270 g_dev_set_media(struct g_consumer *cp)
271 {
272 	struct g_dev_softc *sc;
273 	struct cdev *dev;
274 	char buf[SPECNAMELEN + 6];
275 
276 	sc = cp->private;
277 	dev = sc->sc_dev;
278 	snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
279 	devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
280 	devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
281 	dev = sc->sc_alias;
282 	if (dev != NULL) {
283 		snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
284 		devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
285 		devctl_notify_f("GEOM", "DEV", "MEDIACHANGE", buf, M_WAITOK);
286 	}
287 }
288 
289 static void
290 g_dev_attrchanged(struct g_consumer *cp, const char *attr)
291 {
292 
293 	if (strcmp(attr, "GEOM::media") == 0) {
294 		g_dev_set_media(cp);
295 		return;
296 	}
297 
298 	if (strcmp(attr, "GEOM::physpath") == 0) {
299 		g_dev_set_physpath(cp);
300 		return;
301 	}
302 }
303 
304 static void
305 g_dev_resize(struct g_consumer *cp)
306 {
307 	char buf[SPECNAMELEN + 6];
308 
309 	snprintf(buf, sizeof(buf), "cdev=%s", cp->provider->name);
310 	devctl_notify_f("GEOM", "DEV", "SIZECHANGE", buf, M_WAITOK);
311 }
312 
313 struct g_provider *
314 g_dev_getprovider(struct cdev *dev)
315 {
316 	struct g_consumer *cp;
317 
318 	g_topology_assert();
319 	if (dev == NULL)
320 		return (NULL);
321 	if (dev->si_devsw != &g_dev_cdevsw)
322 		return (NULL);
323 	cp = dev->si_drv2;
324 	return (cp->provider);
325 }
326 
327 static struct g_geom *
328 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
329 {
330 	struct g_geom *gp;
331 	struct g_geom_alias *gap;
332 	struct g_consumer *cp;
333 	struct g_dev_softc *sc;
334 	int error;
335 	struct cdev *dev, *adev;
336 	char buf[SPECNAMELEN + 6];
337 
338 	g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
339 	g_topology_assert();
340 	gp = g_new_geomf(mp, "%s", pp->name);
341 	sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
342 	mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF);
343 	cp = g_new_consumer(gp);
344 	cp->private = sc;
345 	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
346 	error = g_attach(cp, pp);
347 	KASSERT(error == 0,
348 	    ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
349 	error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev,
350 	    &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name);
351 	if (error != 0) {
352 		printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
353 		    __func__, gp->name, error);
354 		g_detach(cp);
355 		g_destroy_consumer(cp);
356 		g_destroy_geom(gp);
357 		mtx_destroy(&sc->sc_mtx);
358 		g_free(sc);
359 		return (NULL);
360 	}
361 	dev->si_flags |= SI_UNMAPPED;
362 	sc->sc_dev = dev;
363 
364 	dev->si_iosize_max = MAXPHYS;
365 	dev->si_drv2 = cp;
366 	error = init_dumpdev(dev);
367 	if (error != 0)
368 		printf("%s: init_dumpdev() failed (gp->name=%s, error=%d)\n",
369 		    __func__, gp->name, error);
370 
371 	g_dev_attrchanged(cp, "GEOM::physpath");
372 	snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
373 	devctl_notify_f("GEOM", "DEV", "CREATE", buf, M_WAITOK);
374 	/*
375 	 * Now add all the aliases for this drive
376 	 */
377 	LIST_FOREACH(gap, &pp->geom->aliases, ga_next) {
378 		error = make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &adev, dev,
379 		    "%s", gap->ga_alias);
380 		if (error) {
381 			printf("%s: make_dev_alias_p() failed (name=%s, error=%d)\n",
382 			    __func__, gap->ga_alias, error);
383 			continue;
384 		}
385 		snprintf(buf, sizeof(buf), "cdev=%s", gap->ga_alias);
386 		devctl_notify_f("GEOM", "DEV", "CREATE", buf, M_WAITOK);
387 	}
388 
389 	return (gp);
390 }
391 
392 static int
393 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
394 {
395 	struct g_consumer *cp;
396 	struct g_dev_softc *sc;
397 	int error, r, w, e;
398 
399 	cp = dev->si_drv2;
400 	if (cp == NULL)
401 		return (ENXIO);		/* g_dev_taste() not done yet */
402 	g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
403 	    cp->geom->name, flags, fmt, td);
404 
405 	r = flags & FREAD ? 1 : 0;
406 	w = flags & FWRITE ? 1 : 0;
407 #ifdef notyet
408 	e = flags & O_EXCL ? 1 : 0;
409 #else
410 	e = 0;
411 #endif
412 
413 	/*
414 	 * This happens on attempt to open a device node with O_EXEC.
415 	 */
416 	if (r + w + e == 0)
417 		return (EINVAL);
418 
419 	if (w) {
420 		/*
421 		 * When running in very secure mode, do not allow
422 		 * opens for writing of any disks.
423 		 */
424 		error = securelevel_ge(td->td_ucred, 2);
425 		if (error)
426 			return (error);
427 	}
428 	g_topology_lock();
429 	error = g_access(cp, r, w, e);
430 	g_topology_unlock();
431 	if (error == 0) {
432 		sc = cp->private;
433 		mtx_lock(&sc->sc_mtx);
434 		if (sc->sc_open == 0 && (sc->sc_active & SC_A_ACTIVE) != 0)
435 			wakeup(&sc->sc_active);
436 		sc->sc_open += r + w + e;
437 		if (sc->sc_open == 0)
438 			atomic_clear_int(&sc->sc_active, SC_A_OPEN);
439 		else
440 			atomic_set_int(&sc->sc_active, SC_A_OPEN);
441 		mtx_unlock(&sc->sc_mtx);
442 	}
443 	return (error);
444 }
445 
446 static int
447 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
448 {
449 	struct g_consumer *cp;
450 	struct g_dev_softc *sc;
451 	int error, r, w, e;
452 
453 	cp = dev->si_drv2;
454 	if (cp == NULL)
455 		return (ENXIO);
456 	g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
457 	    cp->geom->name, flags, fmt, td);
458 
459 	r = flags & FREAD ? -1 : 0;
460 	w = flags & FWRITE ? -1 : 0;
461 #ifdef notyet
462 	e = flags & O_EXCL ? -1 : 0;
463 #else
464 	e = 0;
465 #endif
466 
467 	/*
468 	 * The vgonel(9) - caused by eg. forced unmount of devfs - calls
469 	 * VOP_CLOSE(9) on devfs vnode without any FREAD or FWRITE flags,
470 	 * which would result in zero deltas, which in turn would cause
471 	 * panic in g_access(9).
472 	 *
473 	 * Note that we cannot zero the counters (ie. do "r = cp->acr"
474 	 * etc) instead, because the consumer might be opened in another
475 	 * devfs instance.
476 	 */
477 	if (r + w + e == 0)
478 		return (EINVAL);
479 
480 	sc = cp->private;
481 	mtx_lock(&sc->sc_mtx);
482 	sc->sc_open += r + w + e;
483 	if (sc->sc_open == 0)
484 		atomic_clear_int(&sc->sc_active, SC_A_OPEN);
485 	else
486 		atomic_set_int(&sc->sc_active, SC_A_OPEN);
487 	while (sc->sc_open == 0 && (sc->sc_active & SC_A_ACTIVE) != 0)
488 		msleep(&sc->sc_active, &sc->sc_mtx, 0, "g_dev_close", hz / 10);
489 	mtx_unlock(&sc->sc_mtx);
490 	g_topology_lock();
491 	error = g_access(cp, r, w, e);
492 	g_topology_unlock();
493 	return (error);
494 }
495 
496 /*
497  * XXX: Until we have unmessed the ioctl situation, there is a race against
498  * XXX: a concurrent orphanization.  We cannot close it by holding topology
499  * XXX: since that would prevent us from doing our job, and stalling events
500  * XXX: will break (actually: stall) the BSD disklabel hacks.
501  */
502 static int
503 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
504 {
505 	struct g_consumer *cp;
506 	struct g_provider *pp;
507 	off_t offset, length, chunk, odd;
508 	int i, error;
509 #ifdef COMPAT_FREEBSD12
510 	struct diocskerneldump_arg kda_copy;
511 #endif
512 
513 	cp = dev->si_drv2;
514 	pp = cp->provider;
515 
516 	error = 0;
517 	KASSERT(cp->acr || cp->acw,
518 	    ("Consumer with zero access count in g_dev_ioctl"));
519 
520 	i = IOCPARM_LEN(cmd);
521 	switch (cmd) {
522 	case DIOCGSECTORSIZE:
523 		*(u_int *)data = cp->provider->sectorsize;
524 		if (*(u_int *)data == 0)
525 			error = ENOENT;
526 		break;
527 	case DIOCGMEDIASIZE:
528 		*(off_t *)data = cp->provider->mediasize;
529 		if (*(off_t *)data == 0)
530 			error = ENOENT;
531 		break;
532 	case DIOCGFWSECTORS:
533 		error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
534 		if (error == 0 && *(u_int *)data == 0)
535 			error = ENOENT;
536 		break;
537 	case DIOCGFWHEADS:
538 		error = g_io_getattr("GEOM::fwheads", cp, &i, data);
539 		if (error == 0 && *(u_int *)data == 0)
540 			error = ENOENT;
541 		break;
542 	case DIOCGFRONTSTUFF:
543 		error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
544 		break;
545 #ifdef COMPAT_FREEBSD11
546 	case DIOCSKERNELDUMP_FREEBSD11:
547 	    {
548 		struct diocskerneldump_arg kda;
549 
550 		gone_in(13, "FreeBSD 11.x ABI compat");
551 
552 		bzero(&kda, sizeof(kda));
553 		kda.kda_encryption = KERNELDUMP_ENC_NONE;
554 		kda.kda_index = (*(u_int *)data ? 0 : KDA_REMOVE_ALL);
555 		if (kda.kda_index == KDA_REMOVE_ALL)
556 			error = dumper_remove(devtoname(dev), &kda);
557 		else
558 			error = g_dev_setdumpdev(dev, &kda);
559 		break;
560 	    }
561 #endif
562 #ifdef COMPAT_FREEBSD12
563 	case DIOCSKERNELDUMP_FREEBSD12:
564 	    {
565 		struct diocskerneldump_arg_freebsd12 *kda12;
566 
567 		gone_in(14, "FreeBSD 12.x ABI compat");
568 
569 		kda12 = (void *)data;
570 		memcpy(&kda_copy, kda12, sizeof(kda_copy));
571 		kda_copy.kda_index = (kda12->kda12_enable ?
572 		    0 : KDA_REMOVE_ALL);
573 
574 		explicit_bzero(kda12, sizeof(*kda12));
575 		/* Kludge to pass kda_copy to kda in fallthrough. */
576 		data = (void *)&kda_copy;
577 	    }
578 	    /* FALLTHROUGH */
579 #endif
580 	case DIOCSKERNELDUMP:
581 	    {
582 		struct diocskerneldump_arg *kda;
583 		uint8_t *encryptedkey;
584 
585 		kda = (struct diocskerneldump_arg *)data;
586 		if (kda->kda_index == KDA_REMOVE_ALL ||
587 		    kda->kda_index == KDA_REMOVE_DEV ||
588 		    kda->kda_index == KDA_REMOVE) {
589 			error = dumper_remove(devtoname(dev), kda);
590 			explicit_bzero(kda, sizeof(*kda));
591 			break;
592 		}
593 
594 		if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
595 			if (kda->kda_encryptedkeysize == 0 ||
596 			    kda->kda_encryptedkeysize >
597 			    KERNELDUMP_ENCKEY_MAX_SIZE) {
598 				explicit_bzero(kda, sizeof(*kda));
599 				return (EINVAL);
600 			}
601 			encryptedkey = malloc(kda->kda_encryptedkeysize, M_TEMP,
602 			    M_WAITOK);
603 			error = copyin(kda->kda_encryptedkey, encryptedkey,
604 			    kda->kda_encryptedkeysize);
605 		} else {
606 			encryptedkey = NULL;
607 		}
608 		if (error == 0) {
609 			kda->kda_encryptedkey = encryptedkey;
610 			error = g_dev_setdumpdev(dev, kda);
611 		}
612 		if (encryptedkey != NULL) {
613 			explicit_bzero(encryptedkey, kda->kda_encryptedkeysize);
614 			free(encryptedkey, M_TEMP);
615 		}
616 		explicit_bzero(kda, sizeof(*kda));
617 		break;
618 	    }
619 	case DIOCGFLUSH:
620 		error = g_io_flush(cp);
621 		break;
622 	case DIOCGDELETE:
623 		offset = ((off_t *)data)[0];
624 		length = ((off_t *)data)[1];
625 		if ((offset % cp->provider->sectorsize) != 0 ||
626 		    (length % cp->provider->sectorsize) != 0 || length <= 0) {
627 			printf("%s: offset=%jd length=%jd\n", __func__, offset,
628 			    length);
629 			error = EINVAL;
630 			break;
631 		}
632 		if ((cp->provider->mediasize > 0) &&
633 		    (offset >= cp->provider->mediasize)) {
634 			/*
635 			 * Catch out-of-bounds requests here. The problem is
636 			 * that due to historical GEOM I/O implementation
637 			 * peculatities, g_delete_data() would always return
638 			 * success for requests starting just the next byte
639 			 * after providers media boundary. Condition check on
640 			 * non-zero media size, since that condition would
641 			 * (most likely) cause ENXIO instead.
642 			 */
643 			error = EIO;
644 			break;
645 		}
646 		while (length > 0) {
647 			chunk = length;
648 			if (g_dev_del_max_sectors != 0 && chunk >
649 			    g_dev_del_max_sectors * cp->provider->sectorsize) {
650 				chunk = g_dev_del_max_sectors *
651 				    cp->provider->sectorsize;
652 				if (cp->provider->stripesize > 0) {
653 					odd = (offset + chunk +
654 					    cp->provider->stripeoffset) %
655 					    cp->provider->stripesize;
656 					if (chunk > odd)
657 						chunk -= odd;
658 				}
659 			}
660 			error = g_delete_data(cp, offset, chunk);
661 			length -= chunk;
662 			offset += chunk;
663 			if (error)
664 				break;
665 			/*
666 			 * Since the request size can be large, the service
667 			 * time can be is likewise.  We make this ioctl
668 			 * interruptible by checking for signals for each bio.
669 			 */
670 			if (SIGPENDING(td))
671 				break;
672 		}
673 		break;
674 	case DIOCGIDENT:
675 		error = g_io_getattr("GEOM::ident", cp, &i, data);
676 		break;
677 	case DIOCGPROVIDERNAME:
678 		if (pp == NULL)
679 			return (ENOENT);
680 		strlcpy(data, pp->name, i);
681 		break;
682 	case DIOCGSTRIPESIZE:
683 		*(off_t *)data = cp->provider->stripesize;
684 		break;
685 	case DIOCGSTRIPEOFFSET:
686 		*(off_t *)data = cp->provider->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 (cp->provider->geom->ioctl != NULL) {
739 			error = cp->provider->geom->ioctl(cp->provider, 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 	sc = cp->private;
801 	KASSERT(cp->acr || cp->acw,
802 	    ("Consumer with zero access count in g_dev_strategy"));
803 	biotrack(bp, __func__);
804 #ifdef INVARIANTS
805 	if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
806 	    (bp->bio_bcount % cp->provider->sectorsize) != 0) {
807 		bp->bio_resid = bp->bio_bcount;
808 		biofinish(bp, NULL, EINVAL);
809 		return;
810 	}
811 #endif
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