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: stable/12/sys/geom/geom_dev.c 372423 2022-08-20 02:57:40Z git2svn $");
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 
96 static struct g_class g_dev_class	= {
97 	.name = "DEV",
98 	.version = G_VERSION,
99 	.init = g_dev_init,
100 	.fini = g_dev_fini,
101 	.taste = g_dev_taste,
102 	.orphan = g_dev_orphan,
103 	.attrchanged = g_dev_attrchanged
104 };
105 
106 /*
107  * We target 262144 (8 x 32768) sectors by default as this significantly
108  * increases the throughput on commonly used SSD's with a marginal
109  * increase in non-interruptible request latency.
110  */
111 static uint64_t g_dev_del_max_sectors = 262144;
112 SYSCTL_DECL(_kern_geom);
113 SYSCTL_NODE(_kern_geom, OID_AUTO, dev, CTLFLAG_RW, 0, "GEOM_DEV stuff");
114 SYSCTL_QUAD(_kern_geom_dev, OID_AUTO, delete_max_sectors, CTLFLAG_RW,
115     &g_dev_del_max_sectors, 0, "Maximum number of sectors in a single "
116     "delete request sent to the provider. Larger requests are chunked "
117     "so they can be interrupted. (0 = disable chunking)");
118 
119 static char *dumpdev = NULL;
120 static void
g_dev_init(struct g_class * mp)121 g_dev_init(struct g_class *mp)
122 {
123 
124 	dumpdev = kern_getenv("dumpdev");
125 }
126 
127 static void
g_dev_fini(struct g_class * mp)128 g_dev_fini(struct g_class *mp)
129 {
130 
131 	freeenv(dumpdev);
132 	dumpdev = NULL;
133 }
134 
135 static int
g_dev_setdumpdev(struct cdev * dev,struct diocskerneldump_arg * kda,struct thread * td)136 g_dev_setdumpdev(struct cdev *dev, struct diocskerneldump_arg *kda,
137     struct thread *td)
138 {
139 	struct g_kerneldump kd;
140 	struct g_consumer *cp;
141 	int error, len;
142 
143 	if (dev == NULL || kda == NULL)
144 		return (clear_dumper(td));
145 
146 	cp = dev->si_drv2;
147 	len = sizeof(kd);
148 	memset(&kd, 0, len);
149 	kd.offset = 0;
150 	kd.length = OFF_MAX;
151 	error = g_io_getattr("GEOM::kerneldump", cp, &len, &kd);
152 	if (error != 0)
153 		return (error);
154 
155 	error = set_dumper(&kd.di, devtoname(dev), td, kda->kda_compression,
156 	    kda->kda_encryption, kda->kda_key, kda->kda_encryptedkeysize,
157 	    kda->kda_encryptedkey);
158 	if (error == 0)
159 		dev->si_flags |= SI_DUMPDEV;
160 
161 	return (error);
162 }
163 
164 static int
init_dumpdev(struct cdev * dev)165 init_dumpdev(struct cdev *dev)
166 {
167 	struct diocskerneldump_arg kda;
168 	struct g_consumer *cp;
169 	const char *devprefix = "/dev/", *devname;
170 	int error;
171 	size_t len;
172 
173 	bzero(&kda, sizeof(kda));
174 	kda.kda_enable = 1;
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, curthread);
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
g_dev_destroy(void * arg,int flags __unused)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
g_dev_print(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
g_dev_set_physpath(struct g_consumer * cp)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 | MAKEDEV_CHECKNAME,
262 		    alias_devp, dev, 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
g_dev_set_media(struct g_consumer * cp)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
g_dev_attrchanged(struct g_consumer * cp,const char * attr)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 struct g_provider *
g_dev_getprovider(struct cdev * dev)306 g_dev_getprovider(struct cdev *dev)
307 {
308 	struct g_consumer *cp;
309 
310 	g_topology_assert();
311 	if (dev == NULL)
312 		return (NULL);
313 	if (dev->si_devsw != &g_dev_cdevsw)
314 		return (NULL);
315 	cp = dev->si_drv2;
316 	return (cp->provider);
317 }
318 
319 static struct g_geom *
g_dev_taste(struct g_class * mp,struct g_provider * pp,int insist __unused)320 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
321 {
322 	struct g_geom *gp;
323 	struct g_geom_alias *gap;
324 	struct g_consumer *cp;
325 	struct g_dev_softc *sc;
326 	int error;
327 	struct cdev *dev, *adev;
328 	char buf[SPECNAMELEN + 6];
329 	struct make_dev_args args;
330 
331 	g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
332 	g_topology_assert();
333 	gp = g_new_geomf(mp, "%s", pp->name);
334 	sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
335 	mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF);
336 	cp = g_new_consumer(gp);
337 	cp->private = sc;
338 	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
339 	error = g_attach(cp, pp);
340 	KASSERT(error == 0,
341 	    ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
342 
343 	make_dev_args_init(&args);
344 	args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
345 	args.mda_devsw = &g_dev_cdevsw;
346 	args.mda_cr = NULL;
347 	args.mda_uid = UID_ROOT;
348 	args.mda_gid = GID_OPERATOR;
349 	args.mda_mode = 0640;
350 	args.mda_si_drv1 = sc;
351 	args.mda_si_drv2 = cp;
352 	error = make_dev_s(&args, &sc->sc_dev, "%s", gp->name);
353 	if (error != 0) {
354 		printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
355 		    __func__, gp->name, error);
356 		g_detach(cp);
357 		g_destroy_consumer(cp);
358 		g_destroy_geom(gp);
359 		mtx_destroy(&sc->sc_mtx);
360 		g_free(sc);
361 		return (NULL);
362 	}
363 	dev = sc->sc_dev;
364 	dev->si_flags |= SI_UNMAPPED;
365 	dev->si_iosize_max = MAXPHYS;
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
g_dev_open(struct cdev * dev,int flags,int fmt,struct thread * td)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 	g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
401 	    cp->geom->name, flags, fmt, td);
402 
403 	r = flags & FREAD ? 1 : 0;
404 	w = flags & FWRITE ? 1 : 0;
405 #ifdef notyet
406 	e = flags & O_EXCL ? 1 : 0;
407 #else
408 	e = 0;
409 #endif
410 
411 	/*
412 	 * This happens on attempt to open a device node with O_EXEC.
413 	 */
414 	if (r + w + e == 0)
415 		return (EINVAL);
416 
417 	if (w) {
418 		/*
419 		 * When running in very secure mode, do not allow
420 		 * opens for writing of any disks.
421 		 */
422 		error = securelevel_ge(td->td_ucred, 2);
423 		if (error)
424 			return (error);
425 	}
426 	g_topology_lock();
427 	error = g_access(cp, r, w, e);
428 	g_topology_unlock();
429 	if (error == 0) {
430 		sc = dev->si_drv1;
431 		mtx_lock(&sc->sc_mtx);
432 		if (sc->sc_open == 0 && (sc->sc_active & SC_A_ACTIVE) != 0)
433 			wakeup(&sc->sc_active);
434 		sc->sc_open += r + w + e;
435 		if (sc->sc_open == 0)
436 			atomic_clear_int(&sc->sc_active, SC_A_OPEN);
437 		else
438 			atomic_set_int(&sc->sc_active, SC_A_OPEN);
439 		mtx_unlock(&sc->sc_mtx);
440 	}
441 	return (error);
442 }
443 
444 static int
g_dev_close(struct cdev * dev,int flags,int fmt,struct thread * td)445 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
446 {
447 	struct g_consumer *cp;
448 	struct g_dev_softc *sc;
449 	int error, r, w, e;
450 
451 	cp = dev->si_drv2;
452 	g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
453 	    cp->geom->name, flags, fmt, td);
454 
455 	r = flags & FREAD ? -1 : 0;
456 	w = flags & FWRITE ? -1 : 0;
457 #ifdef notyet
458 	e = flags & O_EXCL ? -1 : 0;
459 #else
460 	e = 0;
461 #endif
462 
463 	/*
464 	 * The vgonel(9) - caused by eg. forced unmount of devfs - calls
465 	 * VOP_CLOSE(9) on devfs vnode without any FREAD or FWRITE flags,
466 	 * which would result in zero deltas, which in turn would cause
467 	 * panic in g_access(9).
468 	 *
469 	 * Note that we cannot zero the counters (ie. do "r = cp->acr"
470 	 * etc) instead, because the consumer might be opened in another
471 	 * devfs instance.
472 	 */
473 	if (r + w + e == 0)
474 		return (EINVAL);
475 
476 	sc = dev->si_drv1;
477 	mtx_lock(&sc->sc_mtx);
478 	sc->sc_open += r + w + e;
479 	if (sc->sc_open == 0)
480 		atomic_clear_int(&sc->sc_active, SC_A_OPEN);
481 	else
482 		atomic_set_int(&sc->sc_active, SC_A_OPEN);
483 	while (sc->sc_open == 0 && (sc->sc_active & SC_A_ACTIVE) != 0)
484 		msleep(&sc->sc_active, &sc->sc_mtx, 0, "g_dev_close", hz / 10);
485 	mtx_unlock(&sc->sc_mtx);
486 	g_topology_lock();
487 	error = g_access(cp, r, w, e);
488 	g_topology_unlock();
489 	return (error);
490 }
491 
492 static int
g_dev_ioctl(struct cdev * dev,u_long cmd,caddr_t data,int fflag,struct thread * td)493 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
494 {
495 	struct g_consumer *cp;
496 	struct g_provider *pp;
497 	off_t offset, length, chunk, odd;
498 	int i, error;
499 
500 	cp = dev->si_drv2;
501 	pp = cp->provider;
502 
503 	/* If consumer or provider is dying, don't disturb. */
504 	if (cp->flags & G_CF_ORPHAN)
505 		return (ENXIO);
506 	if (pp->error)
507 		return (pp->error);
508 
509 	error = 0;
510 	KASSERT(cp->acr || cp->acw,
511 	    ("Consumer with zero access count in g_dev_ioctl"));
512 
513 	i = IOCPARM_LEN(cmd);
514 	switch (cmd) {
515 	case DIOCGSECTORSIZE:
516 		*(u_int *)data = pp->sectorsize;
517 		if (*(u_int *)data == 0)
518 			error = ENOENT;
519 		break;
520 	case DIOCGMEDIASIZE:
521 		*(off_t *)data = pp->mediasize;
522 		if (*(off_t *)data == 0)
523 			error = ENOENT;
524 		break;
525 	case DIOCGFWSECTORS:
526 		error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
527 		if (error == 0 && *(u_int *)data == 0)
528 			error = ENOENT;
529 		break;
530 	case DIOCGFWHEADS:
531 		error = g_io_getattr("GEOM::fwheads", cp, &i, data);
532 		if (error == 0 && *(u_int *)data == 0)
533 			error = ENOENT;
534 		break;
535 	case DIOCGFRONTSTUFF:
536 		error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
537 		break;
538 #ifdef COMPAT_FREEBSD11
539 	case DIOCSKERNELDUMP_FREEBSD11:
540 	    {
541 		struct diocskerneldump_arg kda;
542 
543 		bzero(&kda, sizeof(kda));
544 		kda.kda_encryption = KERNELDUMP_ENC_NONE;
545 		kda.kda_enable = (uint8_t)*(u_int *)data;
546 		if (kda.kda_enable == 0)
547 			error = g_dev_setdumpdev(NULL, NULL, td);
548 		else
549 			error = g_dev_setdumpdev(dev, &kda, td);
550 		break;
551 	    }
552 #endif
553 	case DIOCSKERNELDUMP:
554 	    {
555 		struct diocskerneldump_arg *kda;
556 		uint8_t *encryptedkey;
557 
558 		kda = (struct diocskerneldump_arg *)data;
559 		if (kda->kda_enable == 0) {
560 			error = g_dev_setdumpdev(NULL, NULL, td);
561 			break;
562 		}
563 
564 		if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
565 			if (kda->kda_encryptedkeysize <= 0 ||
566 			    kda->kda_encryptedkeysize >
567 			    KERNELDUMP_ENCKEY_MAX_SIZE) {
568 				return (EINVAL);
569 			}
570 			encryptedkey = malloc(kda->kda_encryptedkeysize, M_TEMP,
571 			    M_WAITOK);
572 			error = copyin(kda->kda_encryptedkey, encryptedkey,
573 			    kda->kda_encryptedkeysize);
574 		} else {
575 			encryptedkey = NULL;
576 		}
577 		if (error == 0) {
578 			kda->kda_encryptedkey = encryptedkey;
579 			error = g_dev_setdumpdev(dev, kda, td);
580 		}
581 		if (encryptedkey != NULL) {
582 			explicit_bzero(encryptedkey, kda->kda_encryptedkeysize);
583 			free(encryptedkey, M_TEMP);
584 		}
585 		explicit_bzero(kda, sizeof(*kda));
586 		break;
587 	    }
588 	case DIOCGFLUSH:
589 		error = g_io_flush(cp);
590 		break;
591 	case DIOCGDELETE:
592 		offset = ((off_t *)data)[0];
593 		length = ((off_t *)data)[1];
594 		if ((offset % pp->sectorsize) != 0 ||
595 		    (length % pp->sectorsize) != 0 || length <= 0) {
596 			printf("%s: offset=%jd length=%jd\n", __func__, offset,
597 			    length);
598 			error = EINVAL;
599 			break;
600 		}
601 		while (length > 0) {
602 			chunk = length;
603 			if (g_dev_del_max_sectors != 0 &&
604 			    chunk > g_dev_del_max_sectors * pp->sectorsize) {
605 				chunk = g_dev_del_max_sectors * pp->sectorsize;
606 				if (pp->stripesize > 0) {
607 					odd = (offset + chunk +
608 					    pp->stripeoffset) % pp->stripesize;
609 					if (chunk > odd)
610 						chunk -= odd;
611 				}
612 			}
613 			error = g_delete_data(cp, offset, chunk);
614 			length -= chunk;
615 			offset += chunk;
616 			if (error)
617 				break;
618 			/*
619 			 * Since the request size can be large, the service
620 			 * time can be is likewise.  We make this ioctl
621 			 * interruptible by checking for signals for each bio.
622 			 */
623 			if (SIGPENDING(td))
624 				break;
625 		}
626 		break;
627 	case DIOCGIDENT:
628 		error = g_io_getattr("GEOM::ident", cp, &i, data);
629 		break;
630 	case DIOCGPROVIDERNAME:
631 		strlcpy(data, pp->name, i);
632 		break;
633 	case DIOCGSTRIPESIZE:
634 		*(off_t *)data = pp->stripesize;
635 		break;
636 	case DIOCGSTRIPEOFFSET:
637 		*(off_t *)data = pp->stripeoffset;
638 		break;
639 	case DIOCGPHYSPATH:
640 		error = g_io_getattr("GEOM::physpath", cp, &i, data);
641 		if (error == 0 && *(char *)data == '\0')
642 			error = ENOENT;
643 		break;
644 	case DIOCGATTR: {
645 		struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
646 
647 		if (arg->len > sizeof(arg->value)) {
648 			error = EINVAL;
649 			break;
650 		}
651 		error = g_io_getattr(arg->name, cp, &arg->len, &arg->value);
652 		break;
653 	}
654 	case DIOCZONECMD: {
655 		struct disk_zone_args *zone_args =(struct disk_zone_args *)data;
656 		struct disk_zone_rep_entry *new_entries, *old_entries;
657 		struct disk_zone_report *rep;
658 		size_t alloc_size;
659 
660 		old_entries = NULL;
661 		new_entries = NULL;
662 		rep = NULL;
663 		alloc_size = 0;
664 
665 		if (zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES) {
666 			rep = &zone_args->zone_params.report;
667 #define	MAXENTRIES	(MAXPHYS / sizeof(struct disk_zone_rep_entry))
668 			if (rep->entries_allocated > MAXENTRIES)
669 				rep->entries_allocated = MAXENTRIES;
670 			alloc_size = rep->entries_allocated *
671 			    sizeof(struct disk_zone_rep_entry);
672 			if (alloc_size != 0)
673 				new_entries = g_malloc(alloc_size,
674 				    M_WAITOK| M_ZERO);
675 			old_entries = rep->entries;
676 			rep->entries = new_entries;
677 		}
678 		error = g_io_zonecmd(zone_args, cp);
679 		if (zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES &&
680 		    alloc_size != 0 && error == 0)
681 			error = copyout(new_entries, old_entries, alloc_size);
682 		if (old_entries != NULL && rep != NULL)
683 			rep->entries = old_entries;
684 		if (new_entries != NULL)
685 			g_free(new_entries);
686 		break;
687 	}
688 	default:
689 		if (pp->geom->ioctl != NULL) {
690 			error = pp->geom->ioctl(pp, cmd, data, fflag, td);
691 		} else {
692 			error = ENOIOCTL;
693 		}
694 	}
695 
696 	return (error);
697 }
698 
699 static void
g_dev_done(struct bio * bp2)700 g_dev_done(struct bio *bp2)
701 {
702 	struct g_consumer *cp;
703 	struct g_dev_softc *sc;
704 	struct bio *bp;
705 	int active;
706 
707 	cp = bp2->bio_from;
708 	sc = cp->private;
709 	bp = bp2->bio_parent;
710 	bp->bio_error = bp2->bio_error;
711 	bp->bio_completed = bp2->bio_completed;
712 	bp->bio_resid = bp->bio_length - bp2->bio_completed;
713 	if (bp2->bio_cmd == BIO_ZONE)
714 		bcopy(&bp2->bio_zone, &bp->bio_zone, sizeof(bp->bio_zone));
715 
716 	if (bp2->bio_error != 0) {
717 		g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
718 		    bp2, bp2->bio_error);
719 		bp->bio_flags |= BIO_ERROR;
720 	} else {
721 		g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
722 		    bp2, bp, bp2->bio_resid, (intmax_t)bp2->bio_completed);
723 	}
724 	g_destroy_bio(bp2);
725 	active = atomic_fetchadd_int(&sc->sc_active, -1) - 1;
726 	if ((active & SC_A_ACTIVE) == 0) {
727 		if ((active & SC_A_OPEN) == 0)
728 			wakeup(&sc->sc_active);
729 		if (active & SC_A_DESTROY)
730 			g_post_event(g_dev_destroy, cp, M_NOWAIT, NULL);
731 	}
732 	biodone(bp);
733 }
734 
735 static void
g_dev_strategy(struct bio * bp)736 g_dev_strategy(struct bio *bp)
737 {
738 	struct g_consumer *cp;
739 	struct bio *bp2;
740 	struct cdev *dev;
741 	struct g_dev_softc *sc;
742 
743 	KASSERT(bp->bio_cmd == BIO_READ ||
744 	        bp->bio_cmd == BIO_WRITE ||
745 	        bp->bio_cmd == BIO_DELETE ||
746 		bp->bio_cmd == BIO_FLUSH ||
747 		bp->bio_cmd == BIO_ZONE,
748 		("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
749 	dev = bp->bio_dev;
750 	cp = dev->si_drv2;
751 	KASSERT(cp->acr || cp->acw,
752 	    ("Consumer with zero access count in g_dev_strategy"));
753 	biotrack(bp, __func__);
754 #ifdef INVARIANTS
755 	if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
756 	    (bp->bio_bcount % cp->provider->sectorsize) != 0) {
757 		bp->bio_resid = bp->bio_bcount;
758 		biofinish(bp, NULL, EINVAL);
759 		return;
760 	}
761 #endif
762 	sc = dev->si_drv1;
763 	KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
764 	atomic_add_int(&sc->sc_active, 1);
765 
766 	for (;;) {
767 		/*
768 		 * XXX: This is not an ideal solution, but I believe it to
769 		 * XXX: deadlock safely, all things considered.
770 		 */
771 		bp2 = g_clone_bio(bp);
772 		if (bp2 != NULL)
773 			break;
774 		pause("gdstrat", hz / 10);
775 	}
776 	KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
777 	bp2->bio_done = g_dev_done;
778 	g_trace(G_T_BIO,
779 	    "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
780 	    bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
781 	    bp2->bio_data, bp2->bio_cmd);
782 	g_io_request(bp2, cp);
783 	KASSERT(cp->acr || cp->acw,
784 	    ("g_dev_strategy raced with g_dev_close and lost"));
785 
786 }
787 
788 /*
789  * g_dev_callback()
790  *
791  * Called by devfs when asynchronous device destruction is completed.
792  * - Mark that we have no attached device any more.
793  * - If there are no outstanding requests, schedule geom destruction.
794  *   Otherwise destruction will be scheduled later by g_dev_done().
795  */
796 
797 static void
g_dev_callback(void * arg)798 g_dev_callback(void *arg)
799 {
800 	struct g_consumer *cp;
801 	struct g_dev_softc *sc;
802 	int active;
803 
804 	cp = arg;
805 	sc = cp->private;
806 	g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
807 
808 	sc->sc_dev = NULL;
809 	sc->sc_alias = NULL;
810 	active = atomic_fetchadd_int(&sc->sc_active, SC_A_DESTROY);
811 	if ((active & SC_A_ACTIVE) == 0)
812 		g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
813 }
814 
815 /*
816  * g_dev_orphan()
817  *
818  * Called from below when the provider orphaned us.
819  * - Clear any dump settings.
820  * - Request asynchronous device destruction to prevent any more requests
821  *   from coming in.  The provider is already marked with an error, so
822  *   anything which comes in the interim will be returned immediately.
823  */
824 
825 static void
g_dev_orphan(struct g_consumer * cp)826 g_dev_orphan(struct g_consumer *cp)
827 {
828 	struct cdev *dev;
829 	struct g_dev_softc *sc;
830 
831 	g_topology_assert();
832 	sc = cp->private;
833 	dev = sc->sc_dev;
834 	g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
835 
836 	/* Reset any dump-area set on this device */
837 	if (dev->si_flags & SI_DUMPDEV)
838 		(void)clear_dumper(curthread);
839 
840 	/* Destroy the struct cdev *so we get no more requests */
841 	delist_dev(dev);
842 	destroy_dev_sched_cb(dev, g_dev_callback, cp);
843 }
844 
845 DECLARE_GEOM_CLASS(g_dev_class, g_dev);
846