geom_dev.c revision 133318 
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: head/sys/geom/geom_dev.c 133318 2004-08-08 07:57:53Z phk $");
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
56static d_open_t		g_dev_open;
57static d_close_t	g_dev_close;
58static d_strategy_t	g_dev_strategy;
59static d_ioctl_t	g_dev_ioctl;
60
61static 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_maj =	GEOM_MAJOR,
71	.d_flags =	D_DISK | D_TRACKCLOSE,
72};
73
74static g_taste_t g_dev_taste;
75static g_orphan_t g_dev_orphan;
76
77static struct g_class g_dev_class	= {
78	.name = "DEV",
79	.version = G_VERSION,
80	.taste = g_dev_taste,
81	.orphan = g_dev_orphan,
82};
83
84void
85g_dev_print(void)
86{
87	struct g_geom *gp;
88	char const *p = "";
89
90	LIST_FOREACH(gp, &g_dev_class.geom, geom) {
91		printf("%s%s", p, gp->name);
92		p = " ";
93	}
94	printf("\n");
95}
96
97struct g_provider *
98g_dev_getprovider(struct cdev *dev)
99{
100	struct g_consumer *cp;
101
102	if (dev == NULL)
103		return (NULL);
104	if (devsw(dev) != &g_dev_cdevsw)
105		return (NULL);
106	cp = dev->si_drv2;
107	return (cp->provider);
108}
109
110
111static struct g_geom *
112g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
113{
114	struct g_geom *gp;
115	struct g_consumer *cp;
116	static int unit = GEOM_MINOR_PROVIDERS;
117	int error;
118	struct cdev *dev;
119
120	g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
121	g_topology_assert();
122	LIST_FOREACH(cp, &pp->consumers, consumers)
123		if (cp->geom->class == mp)
124			return (NULL);
125	gp = g_new_geomf(mp, pp->name);
126	cp = g_new_consumer(gp);
127	error = g_attach(cp, pp);
128	KASSERT(error == 0,
129	    ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
130	/*
131	 * XXX: I'm not 100% sure we can call make_dev(9) without Giant
132	 * yet.  Once we can, we don't need to drop topology here either.
133	 */
134	g_topology_unlock();
135	mtx_lock(&Giant);
136	dev = make_dev(&g_dev_cdevsw, unit2minor(unit++),
137	    UID_ROOT, GID_OPERATOR, 0640, gp->name);
138	if (pp->flags & G_PF_CANDELETE)
139		dev->si_flags |= SI_CANDELETE;
140	mtx_unlock(&Giant);
141	g_topology_lock();
142	dev->si_iosize_max = MAXPHYS;
143	dev->si_stripesize = pp->stripesize;
144	dev->si_stripeoffset = pp->stripeoffset;
145	gp->softc = dev;
146	dev->si_drv1 = gp;
147	dev->si_drv2 = cp;
148	return (gp);
149}
150
151static int
152g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
153{
154	struct g_geom *gp;
155	struct g_consumer *cp;
156	int error, r, w, e;
157
158	gp = dev->si_drv1;
159	cp = dev->si_drv2;
160	if (gp == NULL || cp == NULL || gp->softc != dev)
161		return(ENXIO);		/* g_dev_taste() not done yet */
162
163	g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
164	    gp->name, flags, fmt, td);
165
166	r = flags & FREAD ? 1 : 0;
167	w = flags & FWRITE ? 1 : 0;
168#ifdef notyet
169	e = flags & O_EXCL ? 1 : 0;
170#else
171	e = 0;
172#endif
173	if (w) {
174		/*
175		 * When running in very secure mode, do not allow
176		 * opens for writing of any disks.
177		 */
178		error = securelevel_ge(td->td_ucred, 2);
179		if (error)
180			return (error);
181	}
182	g_topology_lock();
183	if (dev->si_devsw == NULL)
184		error = ENXIO;		/* We were orphaned */
185	else
186		error = g_access(cp, r, w, e);
187	g_topology_unlock();
188	g_waitidle();
189	if (!error)
190		dev->si_bsize_phys = cp->provider->sectorsize;
191	return(error);
192}
193
194static int
195g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
196{
197	struct g_geom *gp;
198	struct g_consumer *cp;
199	int error, r, w, e, i;
200
201	gp = dev->si_drv1;
202	cp = dev->si_drv2;
203	if (gp == NULL || cp == NULL)
204		return(ENXIO);
205	g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
206	    gp->name, flags, fmt, td);
207	r = flags & FREAD ? -1 : 0;
208	w = flags & FWRITE ? -1 : 0;
209#ifdef notyet
210	e = flags & O_EXCL ? -1 : 0;
211#else
212	e = 0;
213#endif
214	g_topology_lock();
215	if (dev->si_devsw == NULL)
216		error = ENXIO;		/* We were orphaned */
217	else
218		error = g_access(cp, r, w, e);
219	for (i = 0; i < 10 * hz;) {
220		if (cp->acr != 0 || cp->acw != 0)
221			break;
222 		if (cp->nstart == cp->nend)
223			break;
224		tsleep(&i, PRIBIO, "gdevwclose", hz / 10);
225		i += hz / 10;
226	}
227	if (cp->acr == 0 && cp->acw == 0 && cp->nstart != cp->nend) {
228		printf("WARNING: Final close of geom_dev(%s) %s %s\n",
229		    gp->name,
230		    "still has outstanding I/O after 10 seconds.",
231		    "Completing close anyway, panic may happen later.");
232	}
233	g_topology_unlock();
234	g_waitidle();
235	return (error);
236}
237
238/*
239 * XXX: Until we have unmessed the ioctl situation, there is a race against
240 * XXX: a concurrent orphanization.  We cannot close it by holding topology
241 * XXX: since that would prevent us from doing our job, and stalling events
242 * XXX: will break (actually: stall) the BSD disklabel hacks.
243 */
244static int
245g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
246{
247	struct g_geom *gp;
248	struct g_consumer *cp;
249	struct g_kerneldump kd;
250	int i, error;
251	u_int u;
252
253	gp = dev->si_drv1;
254	cp = dev->si_drv2;
255
256	error = 0;
257	KASSERT(cp->acr || cp->acw,
258	    ("Consumer with zero access count in g_dev_ioctl"));
259
260	i = IOCPARM_LEN(cmd);
261	switch (cmd) {
262	case DIOCGSECTORSIZE:
263		*(u_int *)data = cp->provider->sectorsize;
264		if (*(u_int *)data == 0)
265			error = ENOENT;
266		break;
267	case DIOCGMEDIASIZE:
268		*(off_t *)data = cp->provider->mediasize;
269		if (*(off_t *)data == 0)
270			error = ENOENT;
271		break;
272	case DIOCGFWSECTORS:
273		error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
274		if (error == 0 && *(u_int *)data == 0)
275			error = ENOENT;
276		break;
277	case DIOCGFWHEADS:
278		error = g_io_getattr("GEOM::fwheads", cp, &i, data);
279		if (error == 0 && *(u_int *)data == 0)
280			error = ENOENT;
281		break;
282	case DIOCGFRONTSTUFF:
283		error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
284		break;
285	case DIOCSKERNELDUMP:
286		u = *((u_int *)data);
287		if (!u) {
288			set_dumper(NULL);
289			error = 0;
290			break;
291		}
292		kd.offset = 0;
293		kd.length = OFF_MAX;
294		i = sizeof kd;
295		error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd);
296		if (!error)
297			dev->si_flags |= SI_DUMPDEV;
298		break;
299
300	default:
301		if (cp->provider->geom->ioctl != NULL) {
302			error = cp->provider->geom->ioctl(cp->provider, cmd, data, td);
303		} else {
304			error = ENOIOCTL;
305		}
306	}
307
308	g_waitidle();
309	return (error);
310}
311
312static void
313g_dev_done(struct bio *bp2)
314{
315	struct bio *bp;
316
317	bp = bp2->bio_parent;
318	bp->bio_error = bp2->bio_error;
319	if (bp->bio_error != 0) {
320		g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
321		    bp2, bp->bio_error);
322		bp->bio_flags |= BIO_ERROR;
323	} else {
324		g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
325		    bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed);
326	}
327	bp->bio_resid = bp->bio_bcount - bp2->bio_completed;
328	g_destroy_bio(bp2);
329	biodone(bp);
330}
331
332static void
333g_dev_strategy(struct bio *bp)
334{
335	struct g_consumer *cp;
336	struct bio *bp2;
337	struct cdev *dev;
338
339	KASSERT(bp->bio_cmd == BIO_READ ||
340	        bp->bio_cmd == BIO_WRITE ||
341	        bp->bio_cmd == BIO_DELETE,
342		("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
343	dev = bp->bio_dev;
344	cp = dev->si_drv2;
345	KASSERT(cp->acr || cp->acw,
346	    ("Consumer with zero access count in g_dev_strategy"));
347
348	for (;;) {
349		/*
350		 * XXX: This is not an ideal solution, but I belive it to
351		 * XXX: deadlock safe, all things considered.
352		 */
353		bp2 = g_clone_bio(bp);
354		if (bp2 != NULL)
355			break;
356		tsleep(&bp, PRIBIO, "gdstrat", hz / 10);
357	}
358	KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
359	bp2->bio_length = (off_t)bp->bio_bcount;
360	bp2->bio_done = g_dev_done;
361	g_trace(G_T_BIO,
362	    "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
363	    bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
364	    bp2->bio_data, bp2->bio_cmd);
365	g_io_request(bp2, cp);
366	KASSERT(cp->acr || cp->acw,
367	    ("g_dev_strategy raced with g_dev_close and lost"));
368
369}
370
371/*
372 * g_dev_orphan()
373 *
374 * Called from below when the provider orphaned us.
375 * - Clear any dump settings.
376 * - Destroy the struct cdev *to prevent any more request from coming in.  The
377 *   provider is already marked with an error, so anything which comes in
378 *   in the interrim will be returned immediately.
379 * - Wait for any outstanding I/O to finish.
380 * - Set our access counts to zero, whatever they were.
381 * - Detach and self-destruct.
382 */
383
384static void
385g_dev_orphan(struct g_consumer *cp)
386{
387	struct g_geom *gp;
388	struct cdev *dev;
389
390	g_topology_assert();
391	gp = cp->geom;
392	dev = gp->softc;
393	g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, gp->name);
394
395	/* Reset any dump-area set on this device */
396	if (dev->si_flags & SI_DUMPDEV)
397		set_dumper(NULL);
398
399	/* Destroy the struct cdev *so we get no more requests */
400	destroy_dev(dev);
401
402	/* Wait for the cows to come home */
403	while (cp->nstart != cp->nend)
404		msleep(&dev, NULL, PRIBIO, "gdevorphan", hz / 10);
405
406	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
407		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
408
409	g_detach(cp);
410	g_destroy_consumer(cp);
411	g_destroy_geom(gp);
412}
413
414DECLARE_GEOM_CLASS(g_dev_class, g_dev);
415