kern_conf.c revision 120529 
1/*-
2 * Copyright (c) 1999-2002 Poul-Henning Kamp
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 */
26
27#include <sys/cdefs.h>
28__FBSDID("$FreeBSD: head/sys/kern/kern_conf.c 120529 2003-09-27 21:50:00Z phk $");
29
30#include <sys/param.h>
31#include <sys/kernel.h>
32#include <sys/systm.h>
33#include <sys/bio.h>
34#include <sys/lock.h>
35#include <sys/mutex.h>
36#include <sys/sysctl.h>
37#include <sys/module.h>
38#include <sys/malloc.h>
39#include <sys/conf.h>
40#include <sys/vnode.h>
41#include <sys/queue.h>
42#include <sys/poll.h>
43#include <sys/ctype.h>
44#include <machine/stdarg.h>
45
46static MALLOC_DEFINE(M_DEVT, "dev_t", "dev_t storage");
47
48/* Built at compile time from sys/conf/majors */
49extern unsigned char reserved_majors[256];
50
51/*
52 * This is the number of hash-buckets.  Experiements with 'real-life'
53 * udev_t's show that a prime halfway between two powers of two works
54 * best.
55 */
56#define DEVT_HASH 83
57
58/* The number of dev_t's we can create before malloc(9) kick in.  */
59#define DEVT_STASH 50
60
61static struct cdev devt_stash[DEVT_STASH];
62
63static LIST_HEAD(, cdev) dev_hash[DEVT_HASH];
64
65static LIST_HEAD(, cdev) dev_free;
66
67static int ready_for_devs;
68
69static int free_devt;
70SYSCTL_INT(_debug, OID_AUTO, free_devt, CTLFLAG_RW, &free_devt, 0, "");
71
72int
73nullop(void)
74{
75
76	return (0);
77}
78
79int
80eopnotsupp(void)
81{
82
83	return (EOPNOTSUPP);
84}
85
86static int
87enxio(void)
88{
89	return (ENXIO);
90}
91
92static int
93enodev(void)
94{
95	return (ENODEV);
96}
97
98/* Define a dead_cdevsw for use when devices leave unexpectedly. */
99
100#define dead_open	(d_open_t *)enxio
101#define dead_close	(d_close_t *)enxio
102#define dead_read	(d_read_t *)enxio
103#define dead_write	(d_write_t *)enxio
104#define dead_ioctl	(d_ioctl_t *)enxio
105#define dead_poll	(d_poll_t *)enodev
106#define dead_mmap	(d_mmap_t *)enodev
107
108static void
109dead_strategy(struct bio *bp)
110{
111
112	biofinish(bp, NULL, ENXIO);
113}
114
115#define dead_dump	(dumper_t *)enxio
116#define dead_kqfilter	(d_kqfilter_t *)enxio
117
118static struct cdevsw dead_cdevsw = {
119	.d_open =	dead_open,
120	.d_close =	dead_close,
121	.d_read =	dead_read,
122	.d_write =	dead_write,
123	.d_ioctl =	dead_ioctl,
124	.d_poll =	dead_poll,
125	.d_mmap =	dead_mmap,
126	.d_strategy =	dead_strategy,
127	.d_name =	"dead",
128	.d_maj =	255,
129	.d_dump =	dead_dump,
130	.d_kqfilter =	dead_kqfilter
131};
132
133/* Default methods if driver does not specify method */
134
135#define null_open	(d_open_t *)nullop
136#define null_close	(d_close_t *)nullop
137#define no_read		(d_read_t *)enodev
138#define no_write	(d_write_t *)enodev
139#define no_ioctl	(d_ioctl_t *)enodev
140#define no_mmap		(d_mmap_t *)enodev
141
142static int
143no_kqfilter(dev_t dev __unused, struct knote *kn __unused)
144{
145
146	return (1);
147}
148
149static void
150no_strategy(struct bio *bp)
151{
152
153	biofinish(bp, NULL, ENODEV);
154}
155
156static int
157no_poll(dev_t dev __unused, int events, struct thread *td __unused)
158{
159	/*
160	 * Return true for read/write.  If the user asked for something
161	 * special, return POLLNVAL, so that clients have a way of
162	 * determining reliably whether or not the extended
163	 * functionality is present without hard-coding knowledge
164	 * of specific filesystem implementations.
165	 * Stay in sync with vop_nopoll().
166	 */
167	if (events & ~POLLSTANDARD)
168		return (POLLNVAL);
169
170	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
171}
172
173#define no_dump		(dumper_t *)enodev
174
175struct cdevsw *
176devsw(dev_t dev)
177{
178	if (dev->si_devsw)
179		return (dev->si_devsw);
180	return (&dead_cdevsw);
181}
182
183/*
184 * dev_t and u_dev_t primitives
185 */
186
187int
188major(dev_t x)
189{
190	if (x == NODEV)
191		return NOUDEV;
192	return((x->si_udev >> 8) & 0xff);
193}
194
195int
196minor(dev_t x)
197{
198	if (x == NODEV)
199		return NOUDEV;
200	return(x->si_udev & 0xffff00ff);
201}
202
203int
204dev2unit(dev_t x)
205{
206	int i;
207
208	if (x == NODEV)
209		return NOUDEV;
210	i = minor(x);
211	return ((i & 0xff) | (i >> 8));
212}
213
214int
215unit2minor(int unit)
216{
217
218	KASSERT(unit <= 0xffffff, ("Invalid unit (%d) in unit2minor", unit));
219	return ((unit & 0xff) | ((unit << 8) & ~0xffff));
220}
221
222static dev_t
223allocdev(void)
224{
225	static int stashed;
226	struct cdev *si;
227
228	if (LIST_FIRST(&dev_free)) {
229		si = LIST_FIRST(&dev_free);
230		LIST_REMOVE(si, si_hash);
231	} else if (stashed >= DEVT_STASH) {
232		MALLOC(si, struct cdev *, sizeof(*si), M_DEVT,
233		    M_USE_RESERVE | M_ZERO | M_WAITOK);
234	} else {
235		si = devt_stash + stashed++;
236		bzero(si, sizeof *si);
237		si->si_flags |= SI_STASHED;
238	}
239	si->__si_namebuf[0] = '\0';
240	si->si_name = si->__si_namebuf;
241	LIST_INIT(&si->si_children);
242	TAILQ_INIT(&si->si_snapshots);
243	return (si);
244}
245
246dev_t
247makedev(int x, int y)
248{
249	struct cdev *si;
250	udev_t	udev;
251	int hash;
252
253	if (x == umajor(NOUDEV) && y == uminor(NOUDEV))
254		panic("makedev of NOUDEV");
255	udev = (x << 8) | y;
256	hash = udev % DEVT_HASH;
257	LIST_FOREACH(si, &dev_hash[hash], si_hash) {
258		if (si->si_udev == udev)
259			return (si);
260	}
261	si = allocdev();
262	si->si_udev = udev;
263	LIST_INSERT_HEAD(&dev_hash[hash], si, si_hash);
264        return (si);
265}
266
267void
268freedev(dev_t dev)
269{
270
271	if (!free_devt)
272		return;
273	if (SLIST_FIRST(&dev->si_hlist))
274		return;
275	if (dev->si_devsw || dev->si_drv1 || dev->si_drv2)
276		return;
277	LIST_REMOVE(dev, si_hash);
278	if (dev->si_flags & SI_STASHED) {
279		bzero(dev, sizeof(*dev));
280		dev->si_flags |= SI_STASHED;
281		LIST_INSERT_HEAD(&dev_free, dev, si_hash);
282	} else {
283		FREE(dev, M_DEVT);
284	}
285}
286
287udev_t
288dev2udev(dev_t x)
289{
290	if (x == NODEV)
291		return NOUDEV;
292	return (x->si_udev);
293}
294
295dev_t
296udev2dev(udev_t x, int b)
297{
298
299	if (x == NOUDEV)
300		return (NODEV);
301	switch (b) {
302		case 0:
303			return makedev(umajor(x), uminor(x));
304		case 1:
305			return (NODEV);
306		default:
307			Debugger("udev2dev(...,X)");
308			return NODEV;
309	}
310}
311
312int
313uminor(udev_t dev)
314{
315	return(dev & 0xffff00ff);
316}
317
318int
319umajor(udev_t dev)
320{
321	return((dev & 0xff00) >> 8);
322}
323
324udev_t
325makeudev(int x, int y)
326{
327        return ((x << 8) | y);
328}
329
330dev_t
331make_dev(struct cdevsw *devsw, int minor, uid_t uid, gid_t gid, int perms, const char *fmt, ...)
332{
333	dev_t	dev;
334	va_list ap;
335	int i;
336
337	KASSERT((minor & ~0xffff00ff) == 0,
338	    ("Invalid minor (0x%x) in make_dev", minor));
339
340	if (devsw->d_open == NULL)	devsw->d_open = null_open;
341	if (devsw->d_close == NULL)	devsw->d_close = null_close;
342	if (devsw->d_read == NULL)	devsw->d_read = no_read;
343	if (devsw->d_write == NULL)	devsw->d_write = no_write;
344	if (devsw->d_ioctl == NULL)	devsw->d_ioctl = no_ioctl;
345	if (devsw->d_poll == NULL)	devsw->d_poll = no_poll;
346	if (devsw->d_mmap == NULL)	devsw->d_mmap = no_mmap;
347	if (devsw->d_strategy == NULL)	devsw->d_strategy = no_strategy;
348	if (devsw->d_dump == NULL)	devsw->d_dump = no_dump;
349	if (devsw->d_kqfilter == NULL)	devsw->d_kqfilter = no_kqfilter;
350
351	if (devsw->d_maj == MAJOR_AUTO) {
352		for (i = NUMCDEVSW - 1; i > 0; i--)
353			if (reserved_majors[i] != i)
354				break;
355		KASSERT(i > 0, ("Out of major numbers (%s)", devsw->d_name));
356		devsw->d_maj = i;
357		reserved_majors[i] = i;
358	} else {
359		if (devsw->d_maj == 256)	/* XXX: tty_cons.c is magic */
360			devsw->d_maj = 0;
361		KASSERT(devsw->d_maj >= 0 && devsw->d_maj < 256,
362		    ("Invalid major (%d) in make_dev", devsw->d_maj));
363		if (reserved_majors[devsw->d_maj] != devsw->d_maj) {
364			printf("WARNING: driver \"%s\" used %s %d\n",
365			    devsw->d_name, "unreserved major device number",
366			    devsw->d_maj);
367			reserved_majors[devsw->d_maj] = devsw->d_maj;
368		}
369	}
370
371	if (!ready_for_devs) {
372		printf("WARNING: Driver mistake: make_dev(%s) called before SI_SUB_DRIVERS\n",
373		       fmt);
374		/* XXX panic here once drivers are cleaned up */
375	}
376
377	dev = makedev(devsw->d_maj, minor);
378	if (dev->si_flags & SI_CHEAPCLONE &&
379	    dev->si_flags & SI_NAMED &&
380	    dev->si_devsw == devsw) {
381		/*
382		 * This is allowed as it removes races and generally
383		 * simplifies cloning devices.
384		 */
385		return (dev);
386	}
387	if (dev->si_flags & SI_NAMED) {
388		printf( "WARNING: Driver mistake: repeat make_dev(\"%s\")\n",
389		    dev->si_name);
390		panic("don't do that");
391	}
392	va_start(ap, fmt);
393	i = vsnrprintf(dev->__si_namebuf, sizeof dev->__si_namebuf, 32, fmt, ap);
394	if (i > (sizeof dev->__si_namebuf - 1)) {
395		printf("WARNING: Device name truncated! (%s)",
396		    dev->__si_namebuf);
397	}
398	va_end(ap);
399	dev->si_devsw = devsw;
400	dev->si_uid = uid;
401	dev->si_gid = gid;
402	dev->si_mode = perms;
403	dev->si_flags |= SI_NAMED;
404
405	devfs_create(dev);
406	return (dev);
407}
408
409int
410dev_named(dev_t pdev, const char *name)
411{
412	dev_t cdev;
413
414	if (strcmp(devtoname(pdev), name) == 0)
415		return (1);
416	LIST_FOREACH(cdev, &pdev->si_children, si_siblings)
417		if (strcmp(devtoname(cdev), name) == 0)
418			return (1);
419	return (0);
420}
421
422void
423dev_depends(dev_t pdev, dev_t cdev)
424{
425
426	cdev->si_parent = pdev;
427	cdev->si_flags |= SI_CHILD;
428	LIST_INSERT_HEAD(&pdev->si_children, cdev, si_siblings);
429}
430
431dev_t
432make_dev_alias(dev_t pdev, const char *fmt, ...)
433{
434	dev_t	dev;
435	va_list ap;
436	int i;
437
438	dev = allocdev();
439	dev->si_flags |= SI_ALIAS;
440	dev->si_flags |= SI_NAMED;
441	dev_depends(pdev, dev);
442	va_start(ap, fmt);
443	i = vsnrprintf(dev->__si_namebuf, sizeof dev->__si_namebuf, 32, fmt, ap);
444	if (i > (sizeof dev->__si_namebuf - 1)) {
445		printf("WARNING: Device name truncated! (%s)",
446		    dev->__si_namebuf);
447	}
448	va_end(ap);
449
450	devfs_create(dev);
451	return (dev);
452}
453
454void
455revoke_and_destroy_dev(dev_t dev)
456{
457	struct vnode *vp;
458
459	GIANT_REQUIRED;
460
461	vp = SLIST_FIRST(&dev->si_hlist);
462	if (vp != NULL)
463		VOP_REVOKE(vp, REVOKEALL);
464	destroy_dev(dev);
465}
466
467void
468destroy_dev(dev_t dev)
469{
470
471	if (!(dev->si_flags & SI_NAMED)) {
472		printf( "WARNING: Driver mistake: destroy_dev on %d/%d\n",
473		    major(dev), minor(dev));
474		panic("don't do that");
475	}
476
477	devfs_destroy(dev);
478	if (dev->si_flags & SI_CHILD) {
479		LIST_REMOVE(dev, si_siblings);
480		dev->si_flags &= ~SI_CHILD;
481	}
482	while (!LIST_EMPTY(&dev->si_children))
483		destroy_dev(LIST_FIRST(&dev->si_children));
484	dev->si_drv1 = 0;
485	dev->si_drv2 = 0;
486	dev->si_devsw = 0;
487	bzero(&dev->__si_u, sizeof(dev->__si_u));
488	dev->si_flags &= ~SI_NAMED;
489	dev->si_flags &= ~SI_ALIAS;
490	freedev(dev);
491}
492
493const char *
494devtoname(dev_t dev)
495{
496	char *p;
497	int mynor;
498
499	if (dev->si_name[0] == '#' || dev->si_name[0] == '\0') {
500		p = dev->si_name;
501		if (devsw(dev))
502			sprintf(p, "#%s/", devsw(dev)->d_name);
503		else
504			sprintf(p, "#%d/", major(dev));
505		p += strlen(p);
506		mynor = minor(dev);
507		if (mynor < 0 || mynor > 255)
508			sprintf(p, "%#x", (u_int)mynor);
509		else
510			sprintf(p, "%d", mynor);
511	}
512	return (dev->si_name);
513}
514
515int
516dev_stdclone(char *name, char **namep, const char *stem, int *unit)
517{
518	int u, i;
519
520	i = strlen(stem);
521	if (bcmp(stem, name, i) != 0)
522		return (0);
523	if (!isdigit(name[i]))
524		return (0);
525	u = 0;
526	if (name[i] == '0' && isdigit(name[i+1]))
527		return (0);
528	while (isdigit(name[i])) {
529		u *= 10;
530		u += name[i++] - '0';
531	}
532	if (u > 0xffffff)
533		return (0);
534	*unit = u;
535	if (namep)
536		*namep = &name[i];
537	if (name[i])
538		return (2);
539	return (1);
540}
541
542/*
543 * Helper sysctl for devname(3).  We're given a {u}dev_t and return
544 * the name, if any, registered by the device driver.
545 */
546static int
547sysctl_devname(SYSCTL_HANDLER_ARGS)
548{
549	int error;
550	udev_t ud;
551	dev_t dev;
552
553	error = SYSCTL_IN(req, &ud, sizeof (ud));
554	if (error)
555		return (error);
556	if (ud == NOUDEV)
557		return(EINVAL);
558	dev = makedev(umajor(ud), uminor(ud));
559	if (dev->si_name[0] == '\0')
560		error = ENOENT;
561	else
562		error = SYSCTL_OUT(req, dev->si_name, strlen(dev->si_name) + 1);
563	freedev(dev);
564	return (error);
565}
566
567SYSCTL_PROC(_kern, OID_AUTO, devname, CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_ANYBODY,
568	NULL, 0, sysctl_devname, "", "devname(3) handler");
569
570/*
571 * Set ready_for_devs; prior to this point, device creation is not allowed.
572 */
573static void
574dev_set_ready(void *junk)
575{
576	ready_for_devs = 1;
577}
578
579SYSINIT(dev_ready, SI_SUB_DEVFS, SI_ORDER_FIRST, dev_set_ready, NULL);
580