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/param.h>
39#include <sys/systm.h>
40#include <sys/malloc.h>
41#include <sys/kernel.h>
42#include <sys/conf.h>
43#include <sys/ctype.h>
44#include <sys/bio.h>
45#include <sys/devctl.h>
46#include <sys/lock.h>
47#include <sys/mutex.h>
48#include <sys/proc.h>
49#include <sys/errno.h>
50#include <sys/time.h>
51#include <sys/disk.h>
52#include <sys/fcntl.h>
53#include <sys/limits.h>
54#include <sys/selinfo.h>
55#include <sys/sysctl.h>
56#include <geom/geom.h>
57#include <geom/geom_int.h>
58#include <machine/stdarg.h>
59
60struct g_dev_softc {
61	struct mtx	 sc_mtx;
62	struct cdev	*sc_dev;
63	struct cdev	*sc_alias;
64	int		 sc_open;
65	u_int		 sc_active;
66	struct selinfo	 sc_selinfo;
67#define	SC_A_DESTROY	(1 << 31)
68#define	SC_A_OPEN	(1 << 30)
69#define	SC_A_ACTIVE	(SC_A_OPEN - 1)
70};
71
72static d_open_t		g_dev_open;
73static d_close_t	g_dev_close;
74static d_strategy_t	g_dev_strategy;
75static d_ioctl_t	g_dev_ioctl;
76static d_kqfilter_t	g_dev_kqfilter;
77
78static void		gdev_filter_detach(struct knote *kn);
79static int		gdev_filter_vnode(struct knote *kn, long hint);
80
81static struct filterops gdev_filterops_vnode = {
82	.f_isfd = 1,
83	.f_detach = gdev_filter_detach,
84	.f_event = gdev_filter_vnode,
85};
86
87static struct cdevsw g_dev_cdevsw = {
88	.d_version =	D_VERSION,
89	.d_open =	g_dev_open,
90	.d_close =	g_dev_close,
91	.d_read =	physread,
92	.d_write =	physwrite,
93	.d_ioctl =	g_dev_ioctl,
94	.d_strategy =	g_dev_strategy,
95	.d_name =	"g_dev",
96	.d_flags =	D_DISK | D_TRACKCLOSE,
97	.d_kqfilter =	g_dev_kqfilter,
98};
99
100static g_init_t g_dev_init;
101static g_fini_t g_dev_fini;
102static g_taste_t g_dev_taste;
103static g_orphan_t g_dev_orphan;
104static g_attrchanged_t g_dev_attrchanged;
105static g_resize_t g_dev_resize;
106
107static struct g_class g_dev_class	= {
108	.name = "DEV",
109	.version = G_VERSION,
110	.init = g_dev_init,
111	.fini = g_dev_fini,
112	.taste = g_dev_taste,
113	.orphan = g_dev_orphan,
114	.attrchanged = g_dev_attrchanged,
115	.resize = g_dev_resize
116};
117
118/*
119 * We target 262144 (8 x 32768) sectors by default as this significantly
120 * increases the throughput on commonly used SSD's with a marginal
121 * increase in non-interruptible request latency.
122 */
123static uint64_t g_dev_del_max_sectors = 262144;
124SYSCTL_DECL(_kern_geom);
125SYSCTL_NODE(_kern_geom, OID_AUTO, dev, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
126    "GEOM_DEV stuff");
127SYSCTL_QUAD(_kern_geom_dev, OID_AUTO, delete_max_sectors, CTLFLAG_RW,
128    &g_dev_del_max_sectors, 0, "Maximum number of sectors in a single "
129    "delete request sent to the provider. Larger requests are chunked "
130    "so they can be interrupted. (0 = disable chunking)");
131
132static char *dumpdev = NULL;
133static void
134g_dev_init(struct g_class *mp)
135{
136
137	dumpdev = kern_getenv("dumpdev");
138}
139
140static void
141g_dev_fini(struct g_class *mp)
142{
143
144	freeenv(dumpdev);
145	dumpdev = NULL;
146}
147
148static int
149g_dev_setdumpdev(struct cdev *dev, struct diocskerneldump_arg *kda)
150{
151	struct g_kerneldump kd;
152	struct g_consumer *cp;
153	int error, len;
154
155	MPASS(dev != NULL && kda != NULL);
156	MPASS(kda->kda_index != KDA_REMOVE);
157
158	cp = dev->si_drv2;
159	len = sizeof(kd);
160	memset(&kd, 0, len);
161	kd.offset = 0;
162	kd.length = OFF_MAX;
163	error = g_io_getattr("GEOM::kerneldump", cp, &len, &kd);
164	if (error != 0)
165		return (error);
166
167	error = dumper_insert(&kd.di, devtoname(dev), kda);
168	if (error == 0)
169		dev->si_flags |= SI_DUMPDEV;
170
171	return (error);
172}
173
174static int
175init_dumpdev(struct cdev *dev)
176{
177	struct diocskerneldump_arg kda;
178	struct g_consumer *cp;
179	const char *devprefix = _PATH_DEV, *devname;
180	int error;
181	size_t len;
182
183	bzero(&kda, sizeof(kda));
184	kda.kda_index = KDA_APPEND;
185
186	if (dumpdev == NULL)
187		return (0);
188
189	len = strlen(devprefix);
190	devname = devtoname(dev);
191	if (strcmp(devname, dumpdev) != 0 &&
192	   (strncmp(dumpdev, devprefix, len) != 0 ||
193	    strcmp(devname, dumpdev + len) != 0))
194		return (0);
195
196	cp = (struct g_consumer *)dev->si_drv2;
197	error = g_access(cp, 1, 0, 0);
198	if (error != 0)
199		return (error);
200
201	error = g_dev_setdumpdev(dev, &kda);
202	if (error == 0) {
203		freeenv(dumpdev);
204		dumpdev = NULL;
205	}
206
207	(void)g_access(cp, -1, 0, 0);
208
209	return (error);
210}
211
212static void
213g_dev_destroy(void *arg, int flags __unused)
214{
215	struct g_consumer *cp;
216	struct g_geom *gp;
217	struct g_dev_softc *sc;
218	char buf[SPECNAMELEN + 6];
219
220	g_topology_assert();
221	cp = arg;
222	gp = cp->geom;
223	sc = cp->private;
224	g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name);
225	snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
226	devctl_notify("GEOM", "DEV", "DESTROY", buf);
227	knlist_clear(&sc->sc_selinfo.si_note, 0);
228	knlist_destroy(&sc->sc_selinfo.si_note);
229	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
230		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
231	g_detach(cp);
232	g_destroy_consumer(cp);
233	g_destroy_geom(gp);
234	mtx_destroy(&sc->sc_mtx);
235	g_free(sc);
236}
237
238void
239g_dev_print(void)
240{
241	struct g_geom *gp;
242	char const *p = "";
243
244	LIST_FOREACH(gp, &g_dev_class.geom, geom) {
245		printf("%s%s", p, gp->name);
246		p = " ";
247	}
248	printf("\n");
249}
250
251static void
252g_dev_set_physpath(struct g_consumer *cp)
253{
254	struct g_dev_softc *sc;
255	char *physpath;
256	int error, physpath_len;
257
258	if (g_access(cp, 1, 0, 0) != 0)
259		return;
260
261	sc = cp->private;
262	physpath_len = MAXPATHLEN;
263	physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
264	error = g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath);
265	g_access(cp, -1, 0, 0);
266	if (error == 0 && strlen(physpath) != 0) {
267		struct cdev *dev, *old_alias_dev;
268		struct cdev **alias_devp;
269
270		dev = sc->sc_dev;
271		old_alias_dev = sc->sc_alias;
272		alias_devp = (struct cdev **)&sc->sc_alias;
273		make_dev_physpath_alias(MAKEDEV_WAITOK | MAKEDEV_CHECKNAME,
274		    alias_devp, dev, old_alias_dev, physpath);
275	} else if (sc->sc_alias) {
276		destroy_dev((struct cdev *)sc->sc_alias);
277		sc->sc_alias = NULL;
278	}
279	g_free(physpath);
280}
281
282static void
283g_dev_set_media(struct g_consumer *cp)
284{
285	struct g_dev_softc *sc;
286	struct cdev *dev;
287	char buf[SPECNAMELEN + 6];
288
289	sc = cp->private;
290	dev = sc->sc_dev;
291	snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
292	devctl_notify("DEVFS", "CDEV", "MEDIACHANGE", buf);
293	devctl_notify("GEOM", "DEV", "MEDIACHANGE", buf);
294	dev = sc->sc_alias;
295	if (dev != NULL) {
296		snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
297		devctl_notify("DEVFS", "CDEV", "MEDIACHANGE", buf);
298		devctl_notify("GEOM", "DEV", "MEDIACHANGE", buf);
299	}
300}
301
302static void
303g_dev_attrchanged(struct g_consumer *cp, const char *attr)
304{
305
306	if (strcmp(attr, "GEOM::media") == 0) {
307		g_dev_set_media(cp);
308		return;
309	}
310
311	if (strcmp(attr, "GEOM::physpath") == 0) {
312		g_dev_set_physpath(cp);
313		return;
314	}
315}
316
317static void
318g_dev_resize(struct g_consumer *cp)
319{
320	struct g_dev_softc *sc;
321	char buf[SPECNAMELEN + 6];
322
323	sc = cp->private;
324	KNOTE_UNLOCKED(&sc->sc_selinfo.si_note, NOTE_ATTRIB);
325
326	snprintf(buf, sizeof(buf), "cdev=%s", cp->provider->name);
327	devctl_notify("GEOM", "DEV", "SIZECHANGE", buf);
328}
329
330struct g_provider *
331g_dev_getprovider(struct cdev *dev)
332{
333	struct g_consumer *cp;
334
335	g_topology_assert();
336	if (dev == NULL)
337		return (NULL);
338	if (dev->si_devsw != &g_dev_cdevsw)
339		return (NULL);
340	cp = dev->si_drv2;
341	return (cp->provider);
342}
343
344static struct g_geom *
345g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
346{
347	struct g_geom *gp;
348	struct g_geom_alias *gap;
349	struct g_consumer *cp;
350	struct g_dev_softc *sc;
351	int error;
352	struct cdev *dev, *adev;
353	char buf[SPECNAMELEN + 6];
354	struct make_dev_args args;
355
356	g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
357	g_topology_assert();
358	gp = g_new_geomf(mp, "%s", pp->name);
359	sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
360	mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF);
361	cp = g_new_consumer(gp);
362	cp->private = sc;
363	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
364	error = g_attach(cp, pp);
365	if (error != 0) {
366		printf("%s: g_dev_taste(%s) failed to g_attach, error=%d\n",
367		    __func__, pp->name, error);
368		g_destroy_consumer(cp);
369		g_destroy_geom(gp);
370		mtx_destroy(&sc->sc_mtx);
371		g_free(sc);
372		return (NULL);
373	}
374	make_dev_args_init(&args);
375	args.mda_flags = MAKEDEV_CHECKNAME | MAKEDEV_WAITOK;
376	args.mda_devsw = &g_dev_cdevsw;
377	args.mda_cr = NULL;
378	args.mda_uid = UID_ROOT;
379	args.mda_gid = GID_OPERATOR;
380	args.mda_mode = 0640;
381	args.mda_si_drv1 = sc;
382	args.mda_si_drv2 = cp;
383	error = make_dev_s(&args, &sc->sc_dev, "%s", gp->name);
384	if (error != 0) {
385		printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
386		    __func__, gp->name, error);
387		g_detach(cp);
388		g_destroy_consumer(cp);
389		g_destroy_geom(gp);
390		mtx_destroy(&sc->sc_mtx);
391		g_free(sc);
392		return (NULL);
393	}
394	dev = sc->sc_dev;
395	dev->si_flags |= SI_UNMAPPED;
396	dev->si_iosize_max = maxphys;
397	knlist_init_mtx(&sc->sc_selinfo.si_note, &sc->sc_mtx);
398	error = init_dumpdev(dev);
399	if (error != 0)
400		printf("%s: init_dumpdev() failed (gp->name=%s, error=%d)\n",
401		    __func__, gp->name, error);
402
403	g_dev_attrchanged(cp, "GEOM::physpath");
404	snprintf(buf, sizeof(buf), "cdev=%s", gp->name);
405	devctl_notify("GEOM", "DEV", "CREATE", buf);
406	/*
407	 * Now add all the aliases for this drive
408	 */
409	LIST_FOREACH(gap, &pp->aliases, ga_next) {
410		error = make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &adev, dev,
411		    "%s", gap->ga_alias);
412		if (error) {
413			printf("%s: make_dev_alias_p() failed (name=%s, error=%d)\n",
414			    __func__, gap->ga_alias, error);
415			continue;
416		}
417		snprintf(buf, sizeof(buf), "cdev=%s", gap->ga_alias);
418		devctl_notify("GEOM", "DEV", "CREATE", buf);
419	}
420
421	return (gp);
422}
423
424static int
425g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
426{
427	struct g_consumer *cp;
428	struct g_dev_softc *sc;
429	int error, r, w, e;
430
431	cp = dev->si_drv2;
432	g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
433	    cp->geom->name, flags, fmt, td);
434
435	r = flags & FREAD ? 1 : 0;
436	w = flags & FWRITE ? 1 : 0;
437#ifdef notyet
438	e = flags & O_EXCL ? 1 : 0;
439#else
440	e = 0;
441#endif
442
443	/*
444	 * This happens on attempt to open a device node with O_EXEC.
445	 */
446	if (r + w + e == 0)
447		return (EINVAL);
448
449	if (w) {
450		/*
451		 * When running in very secure mode, do not allow
452		 * opens for writing of any disks.
453		 */
454		error = securelevel_ge(td->td_ucred, 2);
455		if (error)
456			return (error);
457	}
458	g_topology_lock();
459	error = g_access(cp, r, w, e);
460	g_topology_unlock();
461	if (error == 0) {
462		sc = dev->si_drv1;
463		mtx_lock(&sc->sc_mtx);
464		if (sc->sc_open == 0 && (sc->sc_active & SC_A_ACTIVE) != 0)
465			wakeup(&sc->sc_active);
466		sc->sc_open += r + w + e;
467		if (sc->sc_open == 0)
468			atomic_clear_int(&sc->sc_active, SC_A_OPEN);
469		else
470			atomic_set_int(&sc->sc_active, SC_A_OPEN);
471		KNOTE_LOCKED(&sc->sc_selinfo.si_note, NOTE_OPEN);
472		mtx_unlock(&sc->sc_mtx);
473	}
474	return (error);
475}
476
477static int
478g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
479{
480	struct g_consumer *cp;
481	struct g_dev_softc *sc;
482	int error, r, w, e;
483
484	cp = dev->si_drv2;
485	g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
486	    cp->geom->name, flags, fmt, td);
487
488	r = flags & FREAD ? -1 : 0;
489	w = flags & FWRITE ? -1 : 0;
490#ifdef notyet
491	e = flags & O_EXCL ? -1 : 0;
492#else
493	e = 0;
494#endif
495
496	/*
497	 * The vgonel(9) - caused by eg. forced unmount of devfs - calls
498	 * VOP_CLOSE(9) on devfs vnode without any FREAD or FWRITE flags,
499	 * which would result in zero deltas, which in turn would cause
500	 * panic in g_access(9).
501	 *
502	 * Note that we cannot zero the counters (ie. do "r = cp->acr"
503	 * etc) instead, because the consumer might be opened in another
504	 * devfs instance.
505	 */
506	if (r + w + e == 0)
507		return (EINVAL);
508
509	sc = dev->si_drv1;
510	mtx_lock(&sc->sc_mtx);
511	sc->sc_open += r + w + e;
512	if (sc->sc_open == 0)
513		atomic_clear_int(&sc->sc_active, SC_A_OPEN);
514	else
515		atomic_set_int(&sc->sc_active, SC_A_OPEN);
516	while (sc->sc_open == 0 && (sc->sc_active & SC_A_ACTIVE) != 0)
517		msleep(&sc->sc_active, &sc->sc_mtx, 0, "g_dev_close", hz / 10);
518	KNOTE_LOCKED(&sc->sc_selinfo.si_note, NOTE_CLOSE | (w ? NOTE_CLOSE_WRITE : 0));
519	mtx_unlock(&sc->sc_mtx);
520	g_topology_lock();
521	error = g_access(cp, r, w, e);
522	g_topology_unlock();
523	return (error);
524}
525
526static int
527g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
528{
529	struct g_consumer *cp;
530	struct g_provider *pp;
531	off_t offset, length, chunk, odd;
532	int i, error;
533
534	cp = dev->si_drv2;
535	pp = cp->provider;
536
537	/* If consumer or provider is dying, don't disturb. */
538	if (cp->flags & G_CF_ORPHAN)
539		return (ENXIO);
540	if (pp->error)
541		return (pp->error);
542
543	error = 0;
544	KASSERT(cp->acr || cp->acw,
545	    ("Consumer with zero access count in g_dev_ioctl"));
546
547	i = IOCPARM_LEN(cmd);
548	switch (cmd) {
549	case DIOCGSECTORSIZE:
550		*(u_int *)data = pp->sectorsize;
551		if (*(u_int *)data == 0)
552			error = ENOENT;
553		break;
554	case DIOCGMEDIASIZE:
555		*(off_t *)data = pp->mediasize;
556		if (*(off_t *)data == 0)
557			error = ENOENT;
558		break;
559	case DIOCGFWSECTORS:
560		error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
561		if (error == 0 && *(u_int *)data == 0)
562			error = ENOENT;
563		break;
564	case DIOCGFWHEADS:
565		error = g_io_getattr("GEOM::fwheads", cp, &i, data);
566		if (error == 0 && *(u_int *)data == 0)
567			error = ENOENT;
568		break;
569	case DIOCSKERNELDUMP:
570	    {
571		struct diocskerneldump_arg *kda;
572		uint8_t *encryptedkey;
573
574		kda = (struct diocskerneldump_arg *)data;
575		if (kda->kda_index == KDA_REMOVE_ALL ||
576		    kda->kda_index == KDA_REMOVE_DEV ||
577		    kda->kda_index == KDA_REMOVE) {
578			error = dumper_remove(devtoname(dev), kda);
579			explicit_bzero(kda, sizeof(*kda));
580			break;
581		}
582
583		if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
584			if (kda->kda_encryptedkeysize == 0 ||
585			    kda->kda_encryptedkeysize >
586			    KERNELDUMP_ENCKEY_MAX_SIZE) {
587				explicit_bzero(kda, sizeof(*kda));
588				return (EINVAL);
589			}
590			encryptedkey = malloc(kda->kda_encryptedkeysize, M_TEMP,
591			    M_WAITOK);
592			error = copyin(kda->kda_encryptedkey, encryptedkey,
593			    kda->kda_encryptedkeysize);
594		} else {
595			encryptedkey = NULL;
596		}
597		if (error == 0) {
598			kda->kda_encryptedkey = encryptedkey;
599			error = g_dev_setdumpdev(dev, kda);
600		}
601		zfree(encryptedkey, M_TEMP);
602		explicit_bzero(kda, sizeof(*kda));
603		break;
604	    }
605	case DIOCGFLUSH:
606		error = g_io_flush(cp);
607		break;
608	case DIOCGDELETE:
609		offset = ((off_t *)data)[0];
610		length = ((off_t *)data)[1];
611		if ((offset % pp->sectorsize) != 0 ||
612		    (length % pp->sectorsize) != 0 || length <= 0) {
613			printf("%s: offset=%jd length=%jd\n", __func__, offset,
614			    length);
615			error = EINVAL;
616			break;
617		}
618		while (length > 0) {
619			chunk = length;
620			if (g_dev_del_max_sectors != 0 &&
621			    chunk > g_dev_del_max_sectors * pp->sectorsize) {
622				chunk = g_dev_del_max_sectors * pp->sectorsize;
623				if (pp->stripesize > 0) {
624					odd = (offset + chunk +
625					    pp->stripeoffset) % pp->stripesize;
626					if (chunk > odd)
627						chunk -= odd;
628				}
629			}
630			error = g_delete_data(cp, offset, chunk);
631			length -= chunk;
632			offset += chunk;
633			if (error)
634				break;
635			/*
636			 * Since the request size can be large, the service
637			 * time can be is likewise.  We make this ioctl
638			 * interruptible by checking for signals for each bio.
639			 */
640			if (SIGPENDING(td))
641				break;
642		}
643		break;
644	case DIOCGIDENT:
645		error = g_io_getattr("GEOM::ident", cp, &i, data);
646		break;
647	case DIOCGPROVIDERNAME:
648		strlcpy(data, pp->name, i);
649		break;
650	case DIOCGSTRIPESIZE:
651		*(off_t *)data = pp->stripesize;
652		break;
653	case DIOCGSTRIPEOFFSET:
654		*(off_t *)data = pp->stripeoffset;
655		break;
656	case DIOCGPHYSPATH:
657		error = g_io_getattr("GEOM::physpath", cp, &i, data);
658		if (error == 0 && *(char *)data == '\0')
659			error = ENOENT;
660		break;
661	case DIOCGATTR: {
662		struct diocgattr_arg *arg = (struct diocgattr_arg *)data;
663
664		if (arg->len > sizeof(arg->value)) {
665			error = EINVAL;
666			break;
667		}
668		error = g_io_getattr(arg->name, cp, &arg->len, &arg->value);
669		break;
670	}
671	case DIOCZONECMD: {
672		struct disk_zone_args *zone_args =(struct disk_zone_args *)data;
673		struct disk_zone_rep_entry *new_entries, *old_entries;
674		struct disk_zone_report *rep;
675		size_t alloc_size;
676
677		old_entries = NULL;
678		new_entries = NULL;
679		rep = NULL;
680		alloc_size = 0;
681
682		if (zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES) {
683			rep = &zone_args->zone_params.report;
684#define	MAXENTRIES	(maxphys / sizeof(struct disk_zone_rep_entry))
685			if (rep->entries_allocated > MAXENTRIES)
686				rep->entries_allocated = MAXENTRIES;
687			alloc_size = rep->entries_allocated *
688			    sizeof(struct disk_zone_rep_entry);
689			if (alloc_size != 0)
690				new_entries = g_malloc(alloc_size,
691				    M_WAITOK | M_ZERO);
692			old_entries = rep->entries;
693			rep->entries = new_entries;
694		}
695		error = g_io_zonecmd(zone_args, cp);
696		if (zone_args->zone_cmd == DISK_ZONE_REPORT_ZONES &&
697		    alloc_size != 0 && error == 0)
698			error = copyout(new_entries, old_entries, alloc_size);
699		if (old_entries != NULL && rep != NULL)
700			rep->entries = old_entries;
701		g_free(new_entries);
702		break;
703	}
704	default:
705		if (pp->geom->ioctl != NULL) {
706			error = pp->geom->ioctl(pp, cmd, data, fflag, td);
707		} else {
708			error = ENOIOCTL;
709		}
710	}
711
712	return (error);
713}
714
715static void
716g_dev_done(struct bio *bp2)
717{
718	struct g_consumer *cp;
719	struct g_dev_softc *sc;
720	struct bio *bp;
721	int active;
722
723	cp = bp2->bio_from;
724	sc = cp->private;
725	bp = bp2->bio_parent;
726	bp->bio_error = bp2->bio_error;
727	bp->bio_completed = bp2->bio_completed;
728	bp->bio_resid = bp->bio_length - bp2->bio_completed;
729	if (bp2->bio_cmd == BIO_ZONE)
730		bcopy(&bp2->bio_zone, &bp->bio_zone, sizeof(bp->bio_zone));
731
732	if (bp2->bio_error != 0) {
733		g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
734		    bp2, bp2->bio_error);
735		bp->bio_flags |= BIO_ERROR;
736	} else {
737		if (bp->bio_cmd == BIO_READ)
738			KNOTE_UNLOCKED(&sc->sc_selinfo.si_note, NOTE_READ);
739		if (bp->bio_cmd == BIO_WRITE)
740			KNOTE_UNLOCKED(&sc->sc_selinfo.si_note, NOTE_WRITE);
741		g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
742		    bp2, bp, bp2->bio_resid, (intmax_t)bp2->bio_completed);
743	}
744	g_destroy_bio(bp2);
745	active = atomic_fetchadd_int(&sc->sc_active, -1) - 1;
746	if ((active & SC_A_ACTIVE) == 0) {
747		if ((active & SC_A_OPEN) == 0)
748			wakeup(&sc->sc_active);
749		if (active & SC_A_DESTROY)
750			g_post_event(g_dev_destroy, cp, M_NOWAIT, NULL);
751	}
752	biodone(bp);
753}
754
755static void
756g_dev_strategy(struct bio *bp)
757{
758	struct g_consumer *cp;
759	struct bio *bp2;
760	struct cdev *dev;
761	struct g_dev_softc *sc;
762
763	KASSERT(bp->bio_cmd == BIO_READ ||
764	        bp->bio_cmd == BIO_WRITE ||
765	        bp->bio_cmd == BIO_DELETE ||
766		bp->bio_cmd == BIO_FLUSH ||
767		bp->bio_cmd == BIO_ZONE,
768		("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
769	dev = bp->bio_dev;
770	cp = dev->si_drv2;
771	KASSERT(cp->acr || cp->acw,
772	    ("Consumer with zero access count in g_dev_strategy"));
773	biotrack(bp, __func__);
774#ifdef INVARIANTS
775	if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
776	    (bp->bio_bcount % cp->provider->sectorsize) != 0) {
777		bp->bio_resid = bp->bio_bcount;
778		biofinish(bp, NULL, EINVAL);
779		return;
780	}
781#endif
782	sc = dev->si_drv1;
783	KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
784	atomic_add_int(&sc->sc_active, 1);
785
786	for (;;) {
787		/*
788		 * XXX: This is not an ideal solution, but I believe it to
789		 * XXX: deadlock safely, all things considered.
790		 */
791		bp2 = g_clone_bio(bp);
792		if (bp2 != NULL)
793			break;
794		pause("gdstrat", hz / 10);
795	}
796	KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
797	bp2->bio_done = g_dev_done;
798	g_trace(G_T_BIO,
799	    "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
800	    bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
801	    bp2->bio_data, bp2->bio_cmd);
802	g_io_request(bp2, cp);
803	KASSERT(cp->acr || cp->acw,
804	    ("g_dev_strategy raced with g_dev_close and lost"));
805
806}
807
808/*
809 * g_dev_callback()
810 *
811 * Called by devfs when asynchronous device destruction is completed.
812 * - Mark that we have no attached device any more.
813 * - If there are no outstanding requests, schedule geom destruction.
814 *   Otherwise destruction will be scheduled later by g_dev_done().
815 */
816
817static void
818g_dev_callback(void *arg)
819{
820	struct g_consumer *cp;
821	struct g_dev_softc *sc;
822	int active;
823
824	cp = arg;
825	sc = cp->private;
826	g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
827
828	sc->sc_dev = NULL;
829	sc->sc_alias = NULL;
830	active = atomic_fetchadd_int(&sc->sc_active, SC_A_DESTROY);
831	if ((active & SC_A_ACTIVE) == 0)
832		g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
833}
834
835/*
836 * g_dev_orphan()
837 *
838 * Called from below when the provider orphaned us.
839 * - Clear any dump settings.
840 * - Request asynchronous device destruction to prevent any more requests
841 *   from coming in.  The provider is already marked with an error, so
842 *   anything which comes in the interim will be returned immediately.
843 */
844
845static void
846g_dev_orphan(struct g_consumer *cp)
847{
848	struct cdev *dev;
849	struct g_dev_softc *sc;
850
851	g_topology_assert();
852	sc = cp->private;
853	dev = sc->sc_dev;
854	g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
855
856	/* Reset any dump-area set on this device */
857	if (dev->si_flags & SI_DUMPDEV) {
858		struct diocskerneldump_arg kda;
859
860		bzero(&kda, sizeof(kda));
861		kda.kda_index = KDA_REMOVE_DEV;
862		(void)dumper_remove(devtoname(dev), &kda);
863	}
864
865	/* Destroy the struct cdev *so we get no more requests */
866	delist_dev(dev);
867	destroy_dev_sched_cb(dev, g_dev_callback, cp);
868}
869
870static void
871gdev_filter_detach(struct knote *kn)
872{
873	struct g_dev_softc *sc;
874
875	sc = kn->kn_hook;
876
877	knlist_remove(&sc->sc_selinfo.si_note, kn, 0);
878}
879
880static int
881gdev_filter_vnode(struct knote *kn, long hint)
882{
883	kn->kn_fflags |= kn->kn_sfflags & hint;
884
885	return (kn->kn_fflags != 0);
886}
887
888static int
889g_dev_kqfilter(struct cdev *dev, struct knote *kn)
890{
891	struct g_dev_softc *sc;
892
893	sc = dev->si_drv1;
894
895	if (kn->kn_filter != EVFILT_VNODE)
896		return (EINVAL);
897
898#define SUPPORTED_EVENTS (NOTE_ATTRIB | NOTE_OPEN | NOTE_CLOSE | \
899    NOTE_CLOSE_WRITE | NOTE_READ | NOTE_WRITE)
900	if (kn->kn_sfflags & ~SUPPORTED_EVENTS)
901		return (EOPNOTSUPP);
902
903	kn->kn_fop = &gdev_filterops_vnode;
904	kn->kn_hook = sc;
905	knlist_add(&sc->sc_selinfo.si_note, kn, 0);
906
907	return (0);
908}
909
910DECLARE_GEOM_CLASS(g_dev_class, g_dev);
911