kern_conf.c revision 179248 
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 179248 2008-05-23 16:38:38Z kib $");
29
30#include <sys/param.h>
31#include <sys/kernel.h>
32#include <sys/systm.h>
33#include <sys/bus.h>
34#include <sys/bio.h>
35#include <sys/lock.h>
36#include <sys/mutex.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/sx.h>
44#include <sys/ctype.h>
45#include <sys/tty.h>
46#include <sys/ucred.h>
47#include <sys/taskqueue.h>
48#include <machine/stdarg.h>
49
50#include <fs/devfs/devfs_int.h>
51
52static MALLOC_DEFINE(M_DEVT, "cdev", "cdev storage");
53
54struct mtx devmtx;
55static void destroy_devl(struct cdev *dev);
56static int destroy_dev_sched_cbl(struct cdev *dev,
57    void (*cb)(void *), void *arg);
58static struct cdev *make_dev_credv(int flags,
59    struct cdevsw *devsw, int minornr,
60    struct ucred *cr, uid_t uid, gid_t gid, int mode, const char *fmt,
61    va_list ap);
62
63static struct cdev_priv_list cdevp_free_list =
64    TAILQ_HEAD_INITIALIZER(cdevp_free_list);
65static SLIST_HEAD(free_cdevsw, cdevsw) cdevsw_gt_post_list =
66    SLIST_HEAD_INITIALIZER();
67
68void
69dev_lock(void)
70{
71
72	mtx_lock(&devmtx);
73}
74
75/*
76 * Free all the memory collected while the cdev mutex was
77 * locked. Since devmtx is after the system map mutex, free() cannot
78 * be called immediately and is postponed until cdev mutex can be
79 * dropped.
80 */
81static void
82dev_unlock_and_free(void)
83{
84	struct cdev_priv_list cdp_free;
85	struct free_cdevsw csw_free;
86	struct cdev_priv *cdp;
87	struct cdevsw *csw;
88
89	mtx_assert(&devmtx, MA_OWNED);
90
91	/*
92	 * Make the local copy of the list heads while the dev_mtx is
93	 * held. Free it later.
94	 */
95	TAILQ_INIT(&cdp_free);
96	TAILQ_CONCAT(&cdp_free, &cdevp_free_list, cdp_list);
97	csw_free = cdevsw_gt_post_list;
98	SLIST_INIT(&cdevsw_gt_post_list);
99
100	mtx_unlock(&devmtx);
101
102	while ((cdp = TAILQ_FIRST(&cdp_free)) != NULL) {
103		TAILQ_REMOVE(&cdp_free, cdp, cdp_list);
104		devfs_free(&cdp->cdp_c);
105	}
106	while ((csw = SLIST_FIRST(&csw_free)) != NULL) {
107		SLIST_REMOVE_HEAD(&csw_free, d_postfree_list);
108		free(csw, M_DEVT);
109	}
110}
111
112static void
113dev_free_devlocked(struct cdev *cdev)
114{
115	struct cdev_priv *cdp;
116
117	mtx_assert(&devmtx, MA_OWNED);
118	cdp = cdev->si_priv;
119	TAILQ_INSERT_HEAD(&cdevp_free_list, cdp, cdp_list);
120}
121
122static void
123cdevsw_free_devlocked(struct cdevsw *csw)
124{
125
126	mtx_assert(&devmtx, MA_OWNED);
127	SLIST_INSERT_HEAD(&cdevsw_gt_post_list, csw, d_postfree_list);
128}
129
130void
131dev_unlock(void)
132{
133
134	mtx_unlock(&devmtx);
135}
136
137void
138dev_ref(struct cdev *dev)
139{
140
141	mtx_assert(&devmtx, MA_NOTOWNED);
142	mtx_lock(&devmtx);
143	dev->si_refcount++;
144	mtx_unlock(&devmtx);
145}
146
147void
148dev_refl(struct cdev *dev)
149{
150
151	mtx_assert(&devmtx, MA_OWNED);
152	dev->si_refcount++;
153}
154
155void
156dev_rel(struct cdev *dev)
157{
158	int flag = 0;
159
160	mtx_assert(&devmtx, MA_NOTOWNED);
161	dev_lock();
162	dev->si_refcount--;
163	KASSERT(dev->si_refcount >= 0,
164	    ("dev_rel(%s) gave negative count", devtoname(dev)));
165#if 0
166	if (dev->si_usecount == 0 &&
167	    (dev->si_flags & SI_CHEAPCLONE) && (dev->si_flags & SI_NAMED))
168		;
169	else
170#endif
171	if (dev->si_devsw == NULL && dev->si_refcount == 0) {
172		LIST_REMOVE(dev, si_list);
173		flag = 1;
174	}
175	dev_unlock();
176	if (flag)
177		devfs_free(dev);
178}
179
180struct cdevsw *
181dev_refthread(struct cdev *dev)
182{
183	struct cdevsw *csw;
184	struct cdev_priv *cdp;
185
186	mtx_assert(&devmtx, MA_NOTOWNED);
187	dev_lock();
188	csw = dev->si_devsw;
189	if (csw != NULL) {
190		cdp = dev->si_priv;
191		if ((cdp->cdp_flags & CDP_SCHED_DTR) == 0)
192			dev->si_threadcount++;
193		else
194			csw = NULL;
195	}
196	dev_unlock();
197	return (csw);
198}
199
200struct cdevsw *
201devvn_refthread(struct vnode *vp, struct cdev **devp)
202{
203	struct cdevsw *csw;
204	struct cdev_priv *cdp;
205
206	mtx_assert(&devmtx, MA_NOTOWNED);
207	csw = NULL;
208	dev_lock();
209	*devp = vp->v_rdev;
210	if (*devp != NULL) {
211		cdp = (*devp)->si_priv;
212		if ((cdp->cdp_flags & CDP_SCHED_DTR) == 0) {
213			csw = (*devp)->si_devsw;
214			if (csw != NULL)
215				(*devp)->si_threadcount++;
216		}
217	}
218	dev_unlock();
219	return (csw);
220}
221
222void
223dev_relthread(struct cdev *dev)
224{
225
226	mtx_assert(&devmtx, MA_NOTOWNED);
227	dev_lock();
228	KASSERT(dev->si_threadcount > 0,
229	    ("%s threadcount is wrong", dev->si_name));
230	dev->si_threadcount--;
231	dev_unlock();
232}
233
234int
235nullop(void)
236{
237
238	return (0);
239}
240
241int
242eopnotsupp(void)
243{
244
245	return (EOPNOTSUPP);
246}
247
248static int
249enxio(void)
250{
251	return (ENXIO);
252}
253
254static int
255enodev(void)
256{
257	return (ENODEV);
258}
259
260/* Define a dead_cdevsw for use when devices leave unexpectedly. */
261
262#define dead_open	(d_open_t *)enxio
263#define dead_close	(d_close_t *)enxio
264#define dead_read	(d_read_t *)enxio
265#define dead_write	(d_write_t *)enxio
266#define dead_ioctl	(d_ioctl_t *)enxio
267#define dead_poll	(d_poll_t *)enodev
268#define dead_mmap	(d_mmap_t *)enodev
269
270static void
271dead_strategy(struct bio *bp)
272{
273
274	biofinish(bp, NULL, ENXIO);
275}
276
277#define dead_dump	(dumper_t *)enxio
278#define dead_kqfilter	(d_kqfilter_t *)enxio
279
280static struct cdevsw dead_cdevsw = {
281	.d_version =	D_VERSION,
282	.d_flags =	D_NEEDGIANT, /* XXX: does dead_strategy need this ? */
283	.d_open =	dead_open,
284	.d_close =	dead_close,
285	.d_read =	dead_read,
286	.d_write =	dead_write,
287	.d_ioctl =	dead_ioctl,
288	.d_poll =	dead_poll,
289	.d_mmap =	dead_mmap,
290	.d_strategy =	dead_strategy,
291	.d_name =	"dead",
292	.d_dump =	dead_dump,
293	.d_kqfilter =	dead_kqfilter
294};
295
296/* Default methods if driver does not specify method */
297
298#define null_open	(d_open_t *)nullop
299#define null_close	(d_close_t *)nullop
300#define no_read		(d_read_t *)enodev
301#define no_write	(d_write_t *)enodev
302#define no_ioctl	(d_ioctl_t *)enodev
303#define no_mmap		(d_mmap_t *)enodev
304#define no_kqfilter	(d_kqfilter_t *)enodev
305
306static void
307no_strategy(struct bio *bp)
308{
309
310	biofinish(bp, NULL, ENODEV);
311}
312
313static int
314no_poll(struct cdev *dev __unused, int events, struct thread *td __unused)
315{
316	/*
317	 * Return true for read/write.  If the user asked for something
318	 * special, return POLLNVAL, so that clients have a way of
319	 * determining reliably whether or not the extended
320	 * functionality is present without hard-coding knowledge
321	 * of specific filesystem implementations.
322	 * Stay in sync with vop_nopoll().
323	 */
324	if (events & ~POLLSTANDARD)
325		return (POLLNVAL);
326
327	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
328}
329
330#define no_dump		(dumper_t *)enodev
331
332static int
333giant_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
334{
335	struct cdevsw *dsw;
336	int retval;
337
338	dsw = dev_refthread(dev);
339	if (dsw == NULL)
340		return (ENXIO);
341	mtx_lock(&Giant);
342	retval = dsw->d_gianttrick->d_open(dev, oflags, devtype, td);
343	mtx_unlock(&Giant);
344	dev_relthread(dev);
345	return (retval);
346}
347
348static int
349giant_fdopen(struct cdev *dev, int oflags, struct thread *td, struct file *fp)
350{
351	struct cdevsw *dsw;
352	int retval;
353
354	dsw = dev_refthread(dev);
355	if (dsw == NULL)
356		return (ENXIO);
357	mtx_lock(&Giant);
358	retval = dsw->d_gianttrick->d_fdopen(dev, oflags, td, fp);
359	mtx_unlock(&Giant);
360	dev_relthread(dev);
361	return (retval);
362}
363
364static int
365giant_close(struct cdev *dev, int fflag, int devtype, struct thread *td)
366{
367	struct cdevsw *dsw;
368	int retval;
369
370	dsw = dev_refthread(dev);
371	if (dsw == NULL)
372		return (ENXIO);
373	mtx_lock(&Giant);
374	retval = dsw->d_gianttrick->d_close(dev, fflag, devtype, td);
375	mtx_unlock(&Giant);
376	dev_relthread(dev);
377	return (retval);
378}
379
380static void
381giant_strategy(struct bio *bp)
382{
383	struct cdevsw *dsw;
384	struct cdev *dev;
385
386	dev = bp->bio_dev;
387	dsw = dev_refthread(dev);
388	if (dsw == NULL) {
389		biofinish(bp, NULL, ENXIO);
390		return;
391	}
392	mtx_lock(&Giant);
393	dsw->d_gianttrick->d_strategy(bp);
394	mtx_unlock(&Giant);
395	dev_relthread(dev);
396}
397
398static int
399giant_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
400{
401	struct cdevsw *dsw;
402	int retval;
403
404	dsw = dev_refthread(dev);
405	if (dsw == NULL)
406		return (ENXIO);
407	mtx_lock(&Giant);
408	retval = dsw->d_gianttrick->d_ioctl(dev, cmd, data, fflag, td);
409	mtx_unlock(&Giant);
410	dev_relthread(dev);
411	return (retval);
412}
413
414static int
415giant_read(struct cdev *dev, struct uio *uio, int ioflag)
416{
417	struct cdevsw *dsw;
418	int retval;
419
420	dsw = dev_refthread(dev);
421	if (dsw == NULL)
422		return (ENXIO);
423	mtx_lock(&Giant);
424	retval = dsw->d_gianttrick->d_read(dev, uio, ioflag);
425	mtx_unlock(&Giant);
426	dev_relthread(dev);
427	return (retval);
428}
429
430static int
431giant_write(struct cdev *dev, struct uio *uio, int ioflag)
432{
433	struct cdevsw *dsw;
434	int retval;
435
436	dsw = dev_refthread(dev);
437	if (dsw == NULL)
438		return (ENXIO);
439	mtx_lock(&Giant);
440	retval = dsw->d_gianttrick->d_write(dev, uio, ioflag);
441	mtx_unlock(&Giant);
442	dev_relthread(dev);
443	return (retval);
444}
445
446static int
447giant_poll(struct cdev *dev, int events, struct thread *td)
448{
449	struct cdevsw *dsw;
450	int retval;
451
452	dsw = dev_refthread(dev);
453	if (dsw == NULL)
454		return (ENXIO);
455	mtx_lock(&Giant);
456	retval = dsw->d_gianttrick->d_poll(dev, events, td);
457	mtx_unlock(&Giant);
458	dev_relthread(dev);
459	return (retval);
460}
461
462static int
463giant_kqfilter(struct cdev *dev, struct knote *kn)
464{
465	struct cdevsw *dsw;
466	int retval;
467
468	dsw = dev_refthread(dev);
469	if (dsw == NULL)
470		return (ENXIO);
471	mtx_lock(&Giant);
472	retval = dsw->d_gianttrick->d_kqfilter(dev, kn);
473	mtx_unlock(&Giant);
474	dev_relthread(dev);
475	return (retval);
476}
477
478static int
479giant_mmap(struct cdev *dev, vm_offset_t offset, vm_paddr_t *paddr, int nprot)
480{
481	struct cdevsw *dsw;
482	int retval;
483
484	dsw = dev_refthread(dev);
485	if (dsw == NULL)
486		return (ENXIO);
487	mtx_lock(&Giant);
488	retval = dsw->d_gianttrick->d_mmap(dev, offset, paddr, nprot);
489	mtx_unlock(&Giant);
490	dev_relthread(dev);
491	return (retval);
492}
493
494
495/*
496 * struct cdev * and u_dev_t primitives
497 */
498
499int
500minor(struct cdev *x)
501{
502	if (x == NULL)
503		return NODEV;
504	return(x->si_drv0 & MAXMINOR);
505}
506
507int
508dev2unit(struct cdev *x)
509{
510
511	if (x == NULL)
512		return NODEV;
513	return (minor2unit(minor(x)));
514}
515
516u_int
517minor2unit(u_int _minor)
518{
519
520	KASSERT((_minor & ~MAXMINOR) == 0, ("Illegal minor %x", _minor));
521	return ((_minor & 0xff) | ((_minor >> 8) & 0xffff00));
522}
523
524int
525unit2minor(int unit)
526{
527
528	KASSERT(unit <= 0xffffff, ("Invalid unit (%d) in unit2minor", unit));
529	return ((unit & 0xff) | ((unit << 8) & ~0xffff));
530}
531
532static void
533notify(struct cdev *dev, const char *ev)
534{
535	static const char prefix[] = "cdev=";
536	char *data;
537	int namelen;
538
539	if (cold)
540		return;
541	namelen = strlen(dev->si_name);
542	data = malloc(namelen + sizeof(prefix), M_TEMP, M_WAITOK);
543	memcpy(data, prefix, sizeof(prefix) - 1);
544	memcpy(data + sizeof(prefix) - 1, dev->si_name, namelen + 1);
545	devctl_notify("DEVFS", "CDEV", ev, data);
546	free(data, M_TEMP);
547}
548
549static void
550notify_create(struct cdev *dev)
551{
552
553	notify(dev, "CREATE");
554}
555
556static void
557notify_destroy(struct cdev *dev)
558{
559
560	notify(dev, "DESTROY");
561}
562
563static struct cdev *
564newdev(struct cdevsw *csw, int y, struct cdev *si)
565{
566	struct cdev *si2;
567	dev_t	udev;
568
569	mtx_assert(&devmtx, MA_OWNED);
570	udev = y;
571	LIST_FOREACH(si2, &csw->d_devs, si_list) {
572		if (si2->si_drv0 == udev) {
573			dev_free_devlocked(si);
574			return (si2);
575		}
576	}
577	si->si_drv0 = udev;
578	si->si_devsw = csw;
579	LIST_INSERT_HEAD(&csw->d_devs, si, si_list);
580	return (si);
581}
582
583int
584uminor(dev_t dev)
585{
586	return (dev & MAXMINOR);
587}
588
589int
590umajor(dev_t dev)
591{
592	return ((dev & ~MAXMINOR) >> 8);
593}
594
595static void
596fini_cdevsw(struct cdevsw *devsw)
597{
598	struct cdevsw *gt;
599
600	if (devsw->d_gianttrick != NULL) {
601		gt = devsw->d_gianttrick;
602		memcpy(devsw, gt, sizeof *devsw);
603		cdevsw_free_devlocked(gt);
604		devsw->d_gianttrick = NULL;
605	}
606	devsw->d_flags &= ~D_INIT;
607}
608
609static void
610prep_cdevsw(struct cdevsw *devsw)
611{
612	struct cdevsw *dsw2;
613
614	mtx_assert(&devmtx, MA_OWNED);
615	if (devsw->d_flags & D_INIT)
616		return;
617	if (devsw->d_flags & D_NEEDGIANT) {
618		dev_unlock();
619		dsw2 = malloc(sizeof *dsw2, M_DEVT, M_WAITOK);
620		dev_lock();
621	} else
622		dsw2 = NULL;
623	if (devsw->d_flags & D_INIT) {
624		if (dsw2 != NULL)
625			cdevsw_free_devlocked(dsw2);
626		return;
627	}
628
629	if (devsw->d_version != D_VERSION_01) {
630		printf(
631		    "WARNING: Device driver \"%s\" has wrong version %s\n",
632		    devsw->d_name == NULL ? "???" : devsw->d_name,
633		    "and is disabled.  Recompile KLD module.");
634		devsw->d_open = dead_open;
635		devsw->d_close = dead_close;
636		devsw->d_read = dead_read;
637		devsw->d_write = dead_write;
638		devsw->d_ioctl = dead_ioctl;
639		devsw->d_poll = dead_poll;
640		devsw->d_mmap = dead_mmap;
641		devsw->d_strategy = dead_strategy;
642		devsw->d_dump = dead_dump;
643		devsw->d_kqfilter = dead_kqfilter;
644	}
645
646	if (devsw->d_flags & D_TTY) {
647		if (devsw->d_ioctl == NULL)	devsw->d_ioctl = ttyioctl;
648		if (devsw->d_read == NULL)	devsw->d_read = ttyread;
649		if (devsw->d_write == NULL)	devsw->d_write = ttywrite;
650		if (devsw->d_kqfilter == NULL)	devsw->d_kqfilter = ttykqfilter;
651		if (devsw->d_poll == NULL)	devsw->d_poll = ttypoll;
652	}
653
654	if (devsw->d_flags & D_NEEDGIANT) {
655		if (devsw->d_gianttrick == NULL) {
656			memcpy(dsw2, devsw, sizeof *dsw2);
657			devsw->d_gianttrick = dsw2;
658			dsw2 = NULL;
659		}
660	}
661
662#define FIXUP(member, noop, giant) 				\
663	do {							\
664		if (devsw->member == NULL) {			\
665			devsw->member = noop;			\
666		} else if (devsw->d_flags & D_NEEDGIANT)	\
667			devsw->member = giant;			\
668		}						\
669	while (0)
670
671	FIXUP(d_open,		null_open,	giant_open);
672	FIXUP(d_fdopen,		NULL,		giant_fdopen);
673	FIXUP(d_close,		null_close,	giant_close);
674	FIXUP(d_read,		no_read,	giant_read);
675	FIXUP(d_write,		no_write,	giant_write);
676	FIXUP(d_ioctl,		no_ioctl,	giant_ioctl);
677	FIXUP(d_poll,		no_poll,	giant_poll);
678	FIXUP(d_mmap,		no_mmap,	giant_mmap);
679	FIXUP(d_strategy,	no_strategy,	giant_strategy);
680	FIXUP(d_kqfilter,	no_kqfilter,	giant_kqfilter);
681
682	if (devsw->d_dump == NULL)	devsw->d_dump = no_dump;
683
684	LIST_INIT(&devsw->d_devs);
685
686	devsw->d_flags |= D_INIT;
687
688	if (dsw2 != NULL)
689		cdevsw_free_devlocked(dsw2);
690}
691
692struct cdev *
693make_dev_credv(int flags, struct cdevsw *devsw, int minornr,
694    struct ucred *cr, uid_t uid,
695    gid_t gid, int mode, const char *fmt, va_list ap)
696{
697	struct cdev *dev;
698	int i;
699
700	KASSERT((minornr & ~MAXMINOR) == 0,
701	    ("Invalid minor (0x%x) in make_dev", minornr));
702
703	dev = devfs_alloc();
704	dev_lock();
705	prep_cdevsw(devsw);
706	dev = newdev(devsw, minornr, dev);
707	if (flags & MAKEDEV_REF)
708		dev_refl(dev);
709	if (dev->si_flags & SI_CHEAPCLONE &&
710	    dev->si_flags & SI_NAMED) {
711		/*
712		 * This is allowed as it removes races and generally
713		 * simplifies cloning devices.
714		 * XXX: still ??
715		 */
716		dev_unlock_and_free();
717		return (dev);
718	}
719	KASSERT(!(dev->si_flags & SI_NAMED),
720	    ("make_dev() by driver %s on pre-existing device (min=%x, name=%s)",
721	    devsw->d_name, minor(dev), devtoname(dev)));
722
723	i = vsnrprintf(dev->__si_namebuf, sizeof dev->__si_namebuf, 32, fmt, ap);
724	if (i > (sizeof dev->__si_namebuf - 1)) {
725		printf("WARNING: Device name truncated! (%s)\n",
726		    dev->__si_namebuf);
727	}
728
729	dev->si_flags |= SI_NAMED;
730#ifdef MAC
731	if (cr != NULL)
732		dev->si_cred = crhold(cr);
733	else
734#endif
735		dev->si_cred = NULL;
736	dev->si_uid = uid;
737	dev->si_gid = gid;
738	dev->si_mode = mode;
739
740	devfs_create(dev);
741	clean_unrhdrl(devfs_inos);
742	dev_unlock_and_free();
743
744	notify_create(dev);
745
746	return (dev);
747}
748
749struct cdev *
750make_dev(struct cdevsw *devsw, int minornr, uid_t uid, gid_t gid, int mode,
751    const char *fmt, ...)
752{
753	struct cdev *dev;
754	va_list ap;
755
756	va_start(ap, fmt);
757	dev = make_dev_credv(0, devsw, minornr, NULL, uid, gid, mode, fmt, ap);
758	va_end(ap);
759	return (dev);
760}
761
762struct cdev *
763make_dev_cred(struct cdevsw *devsw, int minornr, struct ucred *cr, uid_t uid,
764    gid_t gid, int mode, const char *fmt, ...)
765{
766	struct cdev *dev;
767	va_list ap;
768
769	va_start(ap, fmt);
770	dev = make_dev_credv(0, devsw, minornr, cr, uid, gid, mode, fmt, ap);
771	va_end(ap);
772
773	return (dev);
774}
775
776struct cdev *
777make_dev_credf(int flags, struct cdevsw *devsw, int minornr,
778    struct ucred *cr, uid_t uid,
779    gid_t gid, int mode, const char *fmt, ...)
780{
781	struct cdev *dev;
782	va_list ap;
783
784	va_start(ap, fmt);
785	dev = make_dev_credv(flags, devsw, minornr, cr, uid, gid, mode,
786	    fmt, ap);
787	va_end(ap);
788
789	return (dev);
790}
791
792static void
793dev_dependsl(struct cdev *pdev, struct cdev *cdev)
794{
795
796	cdev->si_parent = pdev;
797	cdev->si_flags |= SI_CHILD;
798	LIST_INSERT_HEAD(&pdev->si_children, cdev, si_siblings);
799}
800
801
802void
803dev_depends(struct cdev *pdev, struct cdev *cdev)
804{
805
806	dev_lock();
807	dev_dependsl(pdev, cdev);
808	dev_unlock();
809}
810
811struct cdev *
812make_dev_alias(struct cdev *pdev, const char *fmt, ...)
813{
814	struct cdev *dev;
815	va_list ap;
816	int i;
817
818	dev = devfs_alloc();
819	dev_lock();
820	dev->si_flags |= SI_ALIAS;
821	dev->si_flags |= SI_NAMED;
822	va_start(ap, fmt);
823	i = vsnrprintf(dev->__si_namebuf, sizeof dev->__si_namebuf, 32, fmt, ap);
824	if (i > (sizeof dev->__si_namebuf - 1)) {
825		printf("WARNING: Device name truncated! (%s)\n",
826		    dev->__si_namebuf);
827	}
828	va_end(ap);
829
830	devfs_create(dev);
831	clean_unrhdrl(devfs_inos);
832	dev_unlock();
833	dev_depends(pdev, dev);
834
835	notify_create(dev);
836
837	return (dev);
838}
839
840static void
841destroy_devl(struct cdev *dev)
842{
843	struct cdevsw *csw;
844	struct cdev_privdata *p, *p1;
845
846	mtx_assert(&devmtx, MA_OWNED);
847	KASSERT(dev->si_flags & SI_NAMED,
848	    ("WARNING: Driver mistake: destroy_dev on %d\n", minor(dev)));
849
850	devfs_destroy(dev);
851
852	/* Remove name marking */
853	dev->si_flags &= ~SI_NAMED;
854
855	/* If we are a child, remove us from the parents list */
856	if (dev->si_flags & SI_CHILD) {
857		LIST_REMOVE(dev, si_siblings);
858		dev->si_flags &= ~SI_CHILD;
859	}
860
861	/* Kill our children */
862	while (!LIST_EMPTY(&dev->si_children))
863		destroy_devl(LIST_FIRST(&dev->si_children));
864
865	/* Remove from clone list */
866	if (dev->si_flags & SI_CLONELIST) {
867		LIST_REMOVE(dev, si_clone);
868		dev->si_flags &= ~SI_CLONELIST;
869	}
870
871	dev->si_refcount++;	/* Avoid race with dev_rel() */
872	csw = dev->si_devsw;
873	dev->si_devsw = NULL;	/* already NULL for SI_ALIAS */
874	while (csw != NULL && csw->d_purge != NULL && dev->si_threadcount) {
875		csw->d_purge(dev);
876		msleep(csw, &devmtx, PRIBIO, "devprg", hz/10);
877		if (dev->si_threadcount)
878			printf("Still %lu threads in %s\n",
879			    dev->si_threadcount, devtoname(dev));
880	}
881	while (dev->si_threadcount != 0) {
882		/* Use unique dummy wait ident */
883		msleep(&csw, &devmtx, PRIBIO, "devdrn", hz / 10);
884	}
885
886	dev_unlock();
887	notify_destroy(dev);
888	mtx_lock(&cdevpriv_mtx);
889	LIST_FOREACH_SAFE(p, &dev->si_priv->cdp_fdpriv, cdpd_list, p1) {
890		devfs_destroy_cdevpriv(p);
891		mtx_lock(&cdevpriv_mtx);
892	}
893	mtx_unlock(&cdevpriv_mtx);
894	dev_lock();
895
896	dev->si_drv1 = 0;
897	dev->si_drv2 = 0;
898	bzero(&dev->__si_u, sizeof(dev->__si_u));
899
900	if (!(dev->si_flags & SI_ALIAS)) {
901		/* Remove from cdevsw list */
902		LIST_REMOVE(dev, si_list);
903
904		/* If cdevsw has no more struct cdev *'s, clean it */
905		if (LIST_EMPTY(&csw->d_devs)) {
906			fini_cdevsw(csw);
907			wakeup(&csw->d_devs);
908		}
909	}
910	dev->si_flags &= ~SI_ALIAS;
911	dev->si_refcount--;	/* Avoid race with dev_rel() */
912
913	if (dev->si_refcount > 0) {
914		LIST_INSERT_HEAD(&dead_cdevsw.d_devs, dev, si_list);
915	} else {
916		dev_free_devlocked(dev);
917	}
918}
919
920void
921destroy_dev(struct cdev *dev)
922{
923
924	dev_lock();
925	destroy_devl(dev);
926	dev_unlock_and_free();
927}
928
929const char *
930devtoname(struct cdev *dev)
931{
932	char *p;
933	struct cdevsw *csw;
934	int mynor;
935
936	if (dev->si_name[0] == '#' || dev->si_name[0] == '\0') {
937		p = dev->si_name;
938		csw = dev_refthread(dev);
939		if (csw != NULL) {
940			sprintf(p, "(%s)", csw->d_name);
941			dev_relthread(dev);
942		}
943		p += strlen(p);
944		mynor = minor(dev);
945		if (mynor < 0 || mynor > 255)
946			sprintf(p, "/%#x", (u_int)mynor);
947		else
948			sprintf(p, "/%d", mynor);
949	}
950	return (dev->si_name);
951}
952
953int
954dev_stdclone(char *name, char **namep, const char *stem, int *unit)
955{
956	int u, i;
957
958	i = strlen(stem);
959	if (bcmp(stem, name, i) != 0)
960		return (0);
961	if (!isdigit(name[i]))
962		return (0);
963	u = 0;
964	if (name[i] == '0' && isdigit(name[i+1]))
965		return (0);
966	while (isdigit(name[i])) {
967		u *= 10;
968		u += name[i++] - '0';
969	}
970	if (u > 0xffffff)
971		return (0);
972	*unit = u;
973	if (namep)
974		*namep = &name[i];
975	if (name[i])
976		return (2);
977	return (1);
978}
979
980/*
981 * Helper functions for cloning device drivers.
982 *
983 * The objective here is to make it unnecessary for the device drivers to
984 * use rman or similar to manage their unit number space.  Due to the way
985 * we do "on-demand" devices, using rman or other "private" methods
986 * will be very tricky to lock down properly once we lock down this file.
987 *
988 * Instead we give the drivers these routines which puts the struct cdev *'s
989 * that are to be managed on their own list, and gives the driver the ability
990 * to ask for the first free unit number or a given specified unit number.
991 *
992 * In addition these routines support paired devices (pty, nmdm and similar)
993 * by respecting a number of "flag" bits in the minor number.
994 *
995 */
996
997struct clonedevs {
998	LIST_HEAD(,cdev)	head;
999};
1000
1001void
1002clone_setup(struct clonedevs **cdp)
1003{
1004
1005	*cdp = malloc(sizeof **cdp, M_DEVBUF, M_WAITOK | M_ZERO);
1006	LIST_INIT(&(*cdp)->head);
1007}
1008
1009int
1010clone_create(struct clonedevs **cdp, struct cdevsw *csw, int *up, struct cdev **dp, int extra)
1011{
1012	struct clonedevs *cd;
1013	struct cdev *dev, *ndev, *dl, *de;
1014	int unit, low, u;
1015
1016	KASSERT(*cdp != NULL,
1017	    ("clone_setup() not called in driver \"%s\"", csw->d_name));
1018	KASSERT(!(extra & CLONE_UNITMASK),
1019	    ("Illegal extra bits (0x%x) in clone_create", extra));
1020	KASSERT(*up <= CLONE_UNITMASK,
1021	    ("Too high unit (0x%x) in clone_create", *up));
1022
1023
1024	/*
1025	 * Search the list for a lot of things in one go:
1026	 *   A preexisting match is returned immediately.
1027	 *   The lowest free unit number if we are passed -1, and the place
1028	 *	 in the list where we should insert that new element.
1029	 *   The place to insert a specified unit number, if applicable
1030	 *       the end of the list.
1031	 */
1032	unit = *up;
1033	ndev = devfs_alloc();
1034	dev_lock();
1035	prep_cdevsw(csw);
1036	low = extra;
1037	de = dl = NULL;
1038	cd = *cdp;
1039	LIST_FOREACH(dev, &cd->head, si_clone) {
1040		KASSERT(dev->si_flags & SI_CLONELIST,
1041		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
1042		u = dev2unit(dev);
1043		if (u == (unit | extra)) {
1044			*dp = dev;
1045			dev_unlock();
1046			devfs_free(ndev);
1047			return (0);
1048		}
1049		if (unit == -1 && u == low) {
1050			low++;
1051			de = dev;
1052			continue;
1053		} else if (u < (unit | extra)) {
1054			de = dev;
1055			continue;
1056		} else if (u > (unit | extra)) {
1057			dl = dev;
1058			break;
1059		}
1060	}
1061	if (unit == -1)
1062		unit = low & CLONE_UNITMASK;
1063	dev = newdev(csw, unit2minor(unit | extra), ndev);
1064	if (dev->si_flags & SI_CLONELIST) {
1065		printf("dev %p (%s) is on clonelist\n", dev, dev->si_name);
1066		printf("unit=%d, low=%d, extra=0x%x\n", unit, low, extra);
1067		LIST_FOREACH(dev, &cd->head, si_clone) {
1068			printf("\t%p %s\n", dev, dev->si_name);
1069		}
1070		panic("foo");
1071	}
1072	KASSERT(!(dev->si_flags & SI_CLONELIST),
1073	    ("Dev %p(%s) should not be on clonelist", dev, dev->si_name));
1074	if (dl != NULL)
1075		LIST_INSERT_BEFORE(dl, dev, si_clone);
1076	else if (de != NULL)
1077		LIST_INSERT_AFTER(de, dev, si_clone);
1078	else
1079		LIST_INSERT_HEAD(&cd->head, dev, si_clone);
1080	dev->si_flags |= SI_CLONELIST;
1081	*up = unit;
1082	dev_unlock_and_free();
1083	return (1);
1084}
1085
1086/*
1087 * Kill everything still on the list.  The driver should already have
1088 * disposed of any softc hung of the struct cdev *'s at this time.
1089 */
1090void
1091clone_cleanup(struct clonedevs **cdp)
1092{
1093	struct cdev *dev;
1094	struct cdev_priv *cp;
1095	struct clonedevs *cd;
1096
1097	cd = *cdp;
1098	if (cd == NULL)
1099		return;
1100	dev_lock();
1101	while (!LIST_EMPTY(&cd->head)) {
1102		dev = LIST_FIRST(&cd->head);
1103		LIST_REMOVE(dev, si_clone);
1104		KASSERT(dev->si_flags & SI_CLONELIST,
1105		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
1106		dev->si_flags &= ~SI_CLONELIST;
1107		cp = dev->si_priv;
1108		if (!(cp->cdp_flags & CDP_SCHED_DTR)) {
1109			cp->cdp_flags |= CDP_SCHED_DTR;
1110			KASSERT(dev->si_flags & SI_NAMED,
1111				("Driver has goofed in cloning underways udev %x", dev->si_drv0));
1112			destroy_devl(dev);
1113		}
1114	}
1115	dev_unlock_and_free();
1116	free(cd, M_DEVBUF);
1117	*cdp = NULL;
1118}
1119
1120static TAILQ_HEAD(, cdev_priv) dev_ddtr =
1121	TAILQ_HEAD_INITIALIZER(dev_ddtr);
1122static struct task dev_dtr_task;
1123
1124static void
1125destroy_dev_tq(void *ctx, int pending)
1126{
1127	struct cdev_priv *cp;
1128	struct cdev *dev;
1129	void (*cb)(void *);
1130	void *cb_arg;
1131
1132	dev_lock();
1133	while (!TAILQ_EMPTY(&dev_ddtr)) {
1134		cp = TAILQ_FIRST(&dev_ddtr);
1135		dev = &cp->cdp_c;
1136		KASSERT(cp->cdp_flags & CDP_SCHED_DTR,
1137		    ("cdev %p in dev_destroy_tq without CDP_SCHED_DTR", cp));
1138		TAILQ_REMOVE(&dev_ddtr, cp, cdp_dtr_list);
1139		cb = cp->cdp_dtr_cb;
1140		cb_arg = cp->cdp_dtr_cb_arg;
1141		destroy_devl(dev);
1142		dev_unlock_and_free();
1143		dev_rel(dev);
1144		if (cb != NULL)
1145			cb(cb_arg);
1146		dev_lock();
1147	}
1148	dev_unlock();
1149}
1150
1151/*
1152 * devmtx shall be locked on entry. devmtx will be unlocked after
1153 * function return.
1154 */
1155static int
1156destroy_dev_sched_cbl(struct cdev *dev, void (*cb)(void *), void *arg)
1157{
1158	struct cdev_priv *cp;
1159
1160	mtx_assert(&devmtx, MA_OWNED);
1161	cp = dev->si_priv;
1162	if (cp->cdp_flags & CDP_SCHED_DTR) {
1163		dev_unlock();
1164		return (0);
1165	}
1166	dev_refl(dev);
1167	cp->cdp_flags |= CDP_SCHED_DTR;
1168	cp->cdp_dtr_cb = cb;
1169	cp->cdp_dtr_cb_arg = arg;
1170	TAILQ_INSERT_TAIL(&dev_ddtr, cp, cdp_dtr_list);
1171	dev_unlock();
1172	taskqueue_enqueue(taskqueue_swi_giant, &dev_dtr_task);
1173	return (1);
1174}
1175
1176int
1177destroy_dev_sched_cb(struct cdev *dev, void (*cb)(void *), void *arg)
1178{
1179	dev_lock();
1180	return (destroy_dev_sched_cbl(dev, cb, arg));
1181}
1182
1183int
1184destroy_dev_sched(struct cdev *dev)
1185{
1186	return (destroy_dev_sched_cb(dev, NULL, NULL));
1187}
1188
1189void
1190destroy_dev_drain(struct cdevsw *csw)
1191{
1192
1193	dev_lock();
1194	while (!LIST_EMPTY(&csw->d_devs)) {
1195		msleep(&csw->d_devs, &devmtx, PRIBIO, "devscd", hz/10);
1196	}
1197	dev_unlock();
1198}
1199
1200void
1201drain_dev_clone_events(void)
1202{
1203
1204	sx_xlock(&clone_drain_lock);
1205	sx_xunlock(&clone_drain_lock);
1206}
1207
1208static void
1209devdtr_init(void *dummy __unused)
1210{
1211
1212	TASK_INIT(&dev_dtr_task, 0, destroy_dev_tq, NULL);
1213}
1214
1215SYSINIT(devdtr, SI_SUB_DEVFS, SI_ORDER_SECOND, devdtr_init, NULL);
1216