xref: /freebsd/sys/geom/geom_dev.c (revision 1669d8afc64812c8d2d1d147ae1fd42ff441e1b1)
1 /*-
2  * Copyright (c) 2002 Poul-Henning Kamp
3  * Copyright (c) 2002 Networks Associates Technology, Inc.
4  * All rights reserved.
5  *
6  * This software was developed for the FreeBSD Project by Poul-Henning Kamp
7  * and NAI Labs, the Security Research Division of Network Associates, Inc.
8  * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
9  * DARPA CHATS research program.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. The names of the authors may not be used to endorse or promote
20  *    products derived from this software without specific prior written
21  *    permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/malloc.h>
42 #include <sys/kernel.h>
43 #include <sys/conf.h>
44 #include <sys/bio.h>
45 #include <sys/lock.h>
46 #include <sys/mutex.h>
47 #include <sys/proc.h>
48 #include <sys/errno.h>
49 #include <sys/time.h>
50 #include <sys/disk.h>
51 #include <sys/fcntl.h>
52 #include <sys/limits.h>
53 #include <geom/geom.h>
54 #include <geom/geom_int.h>
55 
56 static d_open_t		g_dev_open;
57 static d_close_t	g_dev_close;
58 static d_strategy_t	g_dev_strategy;
59 static d_ioctl_t	g_dev_ioctl;
60 
61 static struct cdevsw g_dev_cdevsw = {
62 	.d_version =	D_VERSION,
63 	.d_open =	g_dev_open,
64 	.d_close =	g_dev_close,
65 	.d_read =	physread,
66 	.d_write =	physwrite,
67 	.d_ioctl =	g_dev_ioctl,
68 	.d_strategy =	g_dev_strategy,
69 	.d_name =	"g_dev",
70 	.d_flags =	D_DISK | D_TRACKCLOSE,
71 };
72 
73 static g_taste_t g_dev_taste;
74 static g_orphan_t g_dev_orphan;
75 static g_init_t		g_dev_init;
76 
77 static struct g_class g_dev_class	= {
78 	.name = "DEV",
79 	.version = G_VERSION,
80 	.taste = g_dev_taste,
81 	.orphan = g_dev_orphan,
82 	.init = g_dev_init,
83 };
84 
85 static struct unrhdr *unithdr;	/* Locked by topology */
86 
87 static void
88 g_dev_init(struct g_class *mp)
89 {
90 
91 	unithdr = new_unrhdr(0, minor2unit(MAXMINOR), NULL);
92 }
93 
94 void
95 g_dev_print(void)
96 {
97 	struct g_geom *gp;
98 	char const *p = "";
99 
100 	LIST_FOREACH(gp, &g_dev_class.geom, geom) {
101 		printf("%s%s", p, gp->name);
102 		p = " ";
103 	}
104 	printf("\n");
105 }
106 
107 struct g_provider *
108 g_dev_getprovider(struct cdev *dev)
109 {
110 	struct g_consumer *cp;
111 
112 	g_topology_assert();
113 	if (dev == NULL)
114 		return (NULL);
115 	if (dev->si_devsw != &g_dev_cdevsw)
116 		return (NULL);
117 	cp = dev->si_drv2;
118 	return (cp->provider);
119 }
120 
121 
122 static struct g_geom *
123 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
124 {
125 	struct g_geom *gp;
126 	struct g_consumer *cp;
127 	int error;
128 	struct cdev *dev;
129 	u_int unit;
130 
131 	g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
132 	g_topology_assert();
133 	LIST_FOREACH(cp, &pp->consumers, consumers)
134 		if (cp->geom->class == mp)
135 			return (NULL);
136 	gp = g_new_geomf(mp, pp->name);
137 	cp = g_new_consumer(gp);
138 	error = g_attach(cp, pp);
139 	KASSERT(error == 0,
140 	    ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
141 	unit = alloc_unr(unithdr);
142 	dev = make_dev(&g_dev_cdevsw, unit2minor(unit),
143 	    UID_ROOT, GID_OPERATOR, 0640, gp->name);
144 	if (pp->flags & G_PF_CANDELETE)
145 		dev->si_flags |= SI_CANDELETE;
146 	dev->si_iosize_max = MAXPHYS;
147 	gp->softc = dev;
148 	dev->si_drv1 = gp;
149 	dev->si_drv2 = cp;
150 	return (gp);
151 }
152 
153 static int
154 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
155 {
156 	struct g_geom *gp;
157 	struct g_consumer *cp;
158 	int error, r, w, e;
159 
160 	gp = dev->si_drv1;
161 	cp = dev->si_drv2;
162 	if (gp == NULL || cp == NULL || gp->softc != dev)
163 		return(ENXIO);		/* g_dev_taste() not done yet */
164 
165 	g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
166 	    gp->name, flags, fmt, td);
167 
168 	r = flags & FREAD ? 1 : 0;
169 	w = flags & FWRITE ? 1 : 0;
170 #ifdef notyet
171 	e = flags & O_EXCL ? 1 : 0;
172 #else
173 	e = 0;
174 #endif
175 	if (w) {
176 		/*
177 		 * When running in very secure mode, do not allow
178 		 * opens for writing of any disks.
179 		 */
180 		error = securelevel_ge(td->td_ucred, 2);
181 		if (error)
182 			return (error);
183 	}
184 	g_topology_lock();
185 	if (dev->si_devsw == NULL)
186 		error = ENXIO;		/* We were orphaned */
187 	else
188 		error = g_access(cp, r, w, e);
189 	g_topology_unlock();
190 	return(error);
191 }
192 
193 static int
194 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
195 {
196 	struct g_geom *gp;
197 	struct g_consumer *cp;
198 	int error, r, w, e, i;
199 
200 	gp = dev->si_drv1;
201 	cp = dev->si_drv2;
202 	if (gp == NULL || cp == NULL)
203 		return(ENXIO);
204 	g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
205 	    gp->name, flags, fmt, td);
206 	r = flags & FREAD ? -1 : 0;
207 	w = flags & FWRITE ? -1 : 0;
208 #ifdef notyet
209 	e = flags & O_EXCL ? -1 : 0;
210 #else
211 	e = 0;
212 #endif
213 	g_topology_lock();
214 	if (dev->si_devsw == NULL)
215 		error = ENXIO;		/* We were orphaned */
216 	else
217 		error = g_access(cp, r, w, e);
218 	for (i = 0; i < 10 * hz;) {
219 		if (cp->acr != 0 || cp->acw != 0)
220 			break;
221  		if (cp->nstart == cp->nend)
222 			break;
223 		pause("gdevwclose", hz / 10);
224 		i += hz / 10;
225 	}
226 	if (cp->acr == 0 && cp->acw == 0 && cp->nstart != cp->nend) {
227 		printf("WARNING: Final close of geom_dev(%s) %s %s\n",
228 		    gp->name,
229 		    "still has outstanding I/O after 10 seconds.",
230 		    "Completing close anyway, panic may happen later.");
231 	}
232 	g_topology_unlock();
233 	return (error);
234 }
235 
236 /*
237  * XXX: Until we have unmessed the ioctl situation, there is a race against
238  * XXX: a concurrent orphanization.  We cannot close it by holding topology
239  * XXX: since that would prevent us from doing our job, and stalling events
240  * XXX: will break (actually: stall) the BSD disklabel hacks.
241  */
242 static int
243 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
244 {
245 	struct g_geom *gp;
246 	struct g_consumer *cp;
247 	struct g_kerneldump kd;
248 	off_t offset, length, chunk;
249 	int i, error;
250 	u_int u;
251 
252 	gp = dev->si_drv1;
253 	cp = dev->si_drv2;
254 
255 	error = 0;
256 	KASSERT(cp->acr || cp->acw,
257 	    ("Consumer with zero access count in g_dev_ioctl"));
258 
259 	i = IOCPARM_LEN(cmd);
260 	switch (cmd) {
261 	case DIOCGSECTORSIZE:
262 		*(u_int *)data = cp->provider->sectorsize;
263 		if (*(u_int *)data == 0)
264 			error = ENOENT;
265 		break;
266 	case DIOCGMEDIASIZE:
267 		*(off_t *)data = cp->provider->mediasize;
268 		if (*(off_t *)data == 0)
269 			error = ENOENT;
270 		break;
271 	case DIOCGFWSECTORS:
272 		error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
273 		if (error == 0 && *(u_int *)data == 0)
274 			error = ENOENT;
275 		break;
276 	case DIOCGFWHEADS:
277 		error = g_io_getattr("GEOM::fwheads", cp, &i, data);
278 		if (error == 0 && *(u_int *)data == 0)
279 			error = ENOENT;
280 		break;
281 	case DIOCGFRONTSTUFF:
282 		error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
283 		break;
284 	case DIOCSKERNELDUMP:
285 		u = *((u_int *)data);
286 		if (!u) {
287 			set_dumper(NULL);
288 			error = 0;
289 			break;
290 		}
291 		kd.offset = 0;
292 		kd.length = OFF_MAX;
293 		i = sizeof kd;
294 		error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd);
295 		if (!error)
296 			dev->si_flags |= SI_DUMPDEV;
297 		break;
298 	case DIOCGFLUSH:
299 		error = g_io_flush(cp);
300 		break;
301 	case DIOCGDELETE:
302 		offset = ((off_t *)data)[0];
303 		length = ((off_t *)data)[1];
304 		if ((offset % cp->provider->sectorsize) != 0 ||
305 		    (length % cp->provider->sectorsize) != 0 || length <= 0) {
306 			printf("%s: offset=%jd length=%jd\n", __func__, offset,
307 			    length);
308 			error = EINVAL;
309 			break;
310 		}
311 		while (length > 0) {
312 			chunk = length;
313 			if (chunk > 1024 * cp->provider->sectorsize)
314 				chunk = 1024 * cp->provider->sectorsize;
315 			error = g_delete_data(cp, offset, chunk);
316 			length -= chunk;
317 			offset += chunk;
318 			if (error)
319 				break;
320 			/*
321 			 * Since the request size is unbounded, the service
322 			 * time is likewise.  We make this ioctl interruptible
323 			 * by checking for signals for each bio.
324 			 */
325 			if (SIGPENDING(td))
326 				break;
327 		}
328 		break;
329 	case DIOCGIDENT:
330 		error = g_io_getattr("GEOM::ident", cp, &i, data);
331 		break;
332 
333 	default:
334 		if (cp->provider->geom->ioctl != NULL) {
335 			error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td);
336 		} else {
337 			error = ENOIOCTL;
338 		}
339 	}
340 
341 	return (error);
342 }
343 
344 static void
345 g_dev_done(struct bio *bp2)
346 {
347 	struct bio *bp;
348 
349 	bp = bp2->bio_parent;
350 	bp->bio_error = bp2->bio_error;
351 	if (bp->bio_error != 0) {
352 		g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
353 		    bp2, bp->bio_error);
354 		bp->bio_flags |= BIO_ERROR;
355 	} else {
356 		g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
357 		    bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed);
358 	}
359 	bp->bio_resid = bp->bio_length - bp2->bio_completed;
360 	bp->bio_completed = bp2->bio_completed;
361 	g_destroy_bio(bp2);
362 	biodone(bp);
363 }
364 
365 static void
366 g_dev_strategy(struct bio *bp)
367 {
368 	struct g_consumer *cp;
369 	struct bio *bp2;
370 	struct cdev *dev;
371 
372 	KASSERT(bp->bio_cmd == BIO_READ ||
373 	        bp->bio_cmd == BIO_WRITE ||
374 	        bp->bio_cmd == BIO_DELETE,
375 		("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
376 	dev = bp->bio_dev;
377 	cp = dev->si_drv2;
378 	KASSERT(cp->acr || cp->acw,
379 	    ("Consumer with zero access count in g_dev_strategy"));
380 
381 	if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
382 	    (bp->bio_bcount % cp->provider->sectorsize) != 0) {
383 		bp->bio_resid = bp->bio_bcount;
384 		biofinish(bp, NULL, EINVAL);
385 		return;
386 	}
387 
388 	for (;;) {
389 		/*
390 		 * XXX: This is not an ideal solution, but I belive it to
391 		 * XXX: deadlock safe, all things considered.
392 		 */
393 		bp2 = g_clone_bio(bp);
394 		if (bp2 != NULL)
395 			break;
396 		pause("gdstrat", hz / 10);
397 	}
398 	KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
399 	bp2->bio_done = g_dev_done;
400 	g_trace(G_T_BIO,
401 	    "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
402 	    bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
403 	    bp2->bio_data, bp2->bio_cmd);
404 	g_io_request(bp2, cp);
405 	KASSERT(cp->acr || cp->acw,
406 	    ("g_dev_strategy raced with g_dev_close and lost"));
407 
408 }
409 
410 /*
411  * g_dev_orphan()
412  *
413  * Called from below when the provider orphaned us.
414  * - Clear any dump settings.
415  * - Destroy the struct cdev *to prevent any more request from coming in.  The
416  *   provider is already marked with an error, so anything which comes in
417  *   in the interrim will be returned immediately.
418  * - Wait for any outstanding I/O to finish.
419  * - Set our access counts to zero, whatever they were.
420  * - Detach and self-destruct.
421  */
422 
423 static void
424 g_dev_orphan(struct g_consumer *cp)
425 {
426 	struct g_geom *gp;
427 	struct cdev *dev;
428 	u_int unit;
429 
430 	g_topology_assert();
431 	gp = cp->geom;
432 	dev = gp->softc;
433 	g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, gp->name);
434 
435 	/* Reset any dump-area set on this device */
436 	if (dev->si_flags & SI_DUMPDEV)
437 		set_dumper(NULL);
438 
439 	/* Destroy the struct cdev *so we get no more requests */
440 	unit = dev2unit(dev);
441 	destroy_dev(dev);
442 	free_unr(unithdr, unit);
443 
444 	/* Wait for the cows to come home */
445 	while (cp->nstart != cp->nend)
446 		pause("gdevorphan", hz / 10);
447 
448 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
449 		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
450 
451 	g_detach(cp);
452 	g_destroy_consumer(cp);
453 	g_destroy_geom(gp);
454 }
455 
456 DECLARE_GEOM_CLASS(g_dev_class, g_dev);
457