kern_conf.c revision 181905 
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 181905 2008-08-20 08:31:58Z ed $");
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/ucred.h>
46#include <sys/taskqueue.h>
47#include <machine/stdarg.h>
48
49#include <fs/devfs/devfs_int.h>
50
51static MALLOC_DEFINE(M_DEVT, "cdev", "cdev storage");
52
53struct mtx devmtx;
54static void destroy_devl(struct cdev *dev);
55static int destroy_dev_sched_cbl(struct cdev *dev,
56    void (*cb)(void *), void *arg);
57static struct cdev *make_dev_credv(int flags,
58    struct cdevsw *devsw, int minornr,
59    struct ucred *cr, uid_t uid, gid_t gid, int mode, const char *fmt,
60    va_list ap);
61
62static struct cdev_priv_list cdevp_free_list =
63    TAILQ_HEAD_INITIALIZER(cdevp_free_list);
64static SLIST_HEAD(free_cdevsw, cdevsw) cdevsw_gt_post_list =
65    SLIST_HEAD_INITIALIZER();
66
67void
68dev_lock(void)
69{
70
71	mtx_lock(&devmtx);
72}
73
74/*
75 * Free all the memory collected while the cdev mutex was
76 * locked. Since devmtx is after the system map mutex, free() cannot
77 * be called immediately and is postponed until cdev mutex can be
78 * dropped.
79 */
80static void
81dev_unlock_and_free(void)
82{
83	struct cdev_priv_list cdp_free;
84	struct free_cdevsw csw_free;
85	struct cdev_priv *cdp;
86	struct cdevsw *csw;
87
88	mtx_assert(&devmtx, MA_OWNED);
89
90	/*
91	 * Make the local copy of the list heads while the dev_mtx is
92	 * held. Free it later.
93	 */
94	TAILQ_INIT(&cdp_free);
95	TAILQ_CONCAT(&cdp_free, &cdevp_free_list, cdp_list);
96	csw_free = cdevsw_gt_post_list;
97	SLIST_INIT(&cdevsw_gt_post_list);
98
99	mtx_unlock(&devmtx);
100
101	while ((cdp = TAILQ_FIRST(&cdp_free)) != NULL) {
102		TAILQ_REMOVE(&cdp_free, cdp, cdp_list);
103		devfs_free(&cdp->cdp_c);
104	}
105	while ((csw = SLIST_FIRST(&csw_free)) != NULL) {
106		SLIST_REMOVE_HEAD(&csw_free, d_postfree_list);
107		free(csw, M_DEVT);
108	}
109}
110
111static void
112dev_free_devlocked(struct cdev *cdev)
113{
114	struct cdev_priv *cdp;
115
116	mtx_assert(&devmtx, MA_OWNED);
117	cdp = cdev2priv(cdev);
118	TAILQ_INSERT_HEAD(&cdevp_free_list, cdp, cdp_list);
119}
120
121static void
122cdevsw_free_devlocked(struct cdevsw *csw)
123{
124
125	mtx_assert(&devmtx, MA_OWNED);
126	SLIST_INSERT_HEAD(&cdevsw_gt_post_list, csw, d_postfree_list);
127}
128
129void
130dev_unlock(void)
131{
132
133	mtx_unlock(&devmtx);
134}
135
136void
137dev_ref(struct cdev *dev)
138{
139
140	mtx_assert(&devmtx, MA_NOTOWNED);
141	mtx_lock(&devmtx);
142	dev->si_refcount++;
143	mtx_unlock(&devmtx);
144}
145
146void
147dev_refl(struct cdev *dev)
148{
149
150	mtx_assert(&devmtx, MA_OWNED);
151	dev->si_refcount++;
152}
153
154void
155dev_rel(struct cdev *dev)
156{
157	int flag = 0;
158
159	mtx_assert(&devmtx, MA_NOTOWNED);
160	dev_lock();
161	dev->si_refcount--;
162	KASSERT(dev->si_refcount >= 0,
163	    ("dev_rel(%s) gave negative count", devtoname(dev)));
164#if 0
165	if (dev->si_usecount == 0 &&
166	    (dev->si_flags & SI_CHEAPCLONE) && (dev->si_flags & SI_NAMED))
167		;
168	else
169#endif
170	if (dev->si_devsw == NULL && dev->si_refcount == 0) {
171		LIST_REMOVE(dev, si_list);
172		flag = 1;
173	}
174	dev_unlock();
175	if (flag)
176		devfs_free(dev);
177}
178
179struct cdevsw *
180dev_refthread(struct cdev *dev)
181{
182	struct cdevsw *csw;
183	struct cdev_priv *cdp;
184
185	mtx_assert(&devmtx, MA_NOTOWNED);
186	dev_lock();
187	csw = dev->si_devsw;
188	if (csw != NULL) {
189		cdp = cdev2priv(dev);
190		if ((cdp->cdp_flags & CDP_SCHED_DTR) == 0)
191			dev->si_threadcount++;
192		else
193			csw = NULL;
194	}
195	dev_unlock();
196	return (csw);
197}
198
199struct cdevsw *
200devvn_refthread(struct vnode *vp, struct cdev **devp)
201{
202	struct cdevsw *csw;
203	struct cdev_priv *cdp;
204
205	mtx_assert(&devmtx, MA_NOTOWNED);
206	csw = NULL;
207	dev_lock();
208	*devp = vp->v_rdev;
209	if (*devp != NULL) {
210		cdp = cdev2priv(*devp);
211		if ((cdp->cdp_flags & CDP_SCHED_DTR) == 0) {
212			csw = (*devp)->si_devsw;
213			if (csw != NULL)
214				(*devp)->si_threadcount++;
215		}
216	}
217	dev_unlock();
218	return (csw);
219}
220
221void
222dev_relthread(struct cdev *dev)
223{
224
225	mtx_assert(&devmtx, MA_NOTOWNED);
226	dev_lock();
227	KASSERT(dev->si_threadcount > 0,
228	    ("%s threadcount is wrong", dev->si_name));
229	dev->si_threadcount--;
230	dev_unlock();
231}
232
233int
234nullop(void)
235{
236
237	return (0);
238}
239
240int
241eopnotsupp(void)
242{
243
244	return (EOPNOTSUPP);
245}
246
247static int
248enxio(void)
249{
250	return (ENXIO);
251}
252
253static int
254enodev(void)
255{
256	return (ENODEV);
257}
258
259/* Define a dead_cdevsw for use when devices leave unexpectedly. */
260
261#define dead_open	(d_open_t *)enxio
262#define dead_close	(d_close_t *)enxio
263#define dead_read	(d_read_t *)enxio
264#define dead_write	(d_write_t *)enxio
265#define dead_ioctl	(d_ioctl_t *)enxio
266#define dead_poll	(d_poll_t *)enodev
267#define dead_mmap	(d_mmap_t *)enodev
268
269static void
270dead_strategy(struct bio *bp)
271{
272
273	biofinish(bp, NULL, ENXIO);
274}
275
276#define dead_dump	(dumper_t *)enxio
277#define dead_kqfilter	(d_kqfilter_t *)enxio
278
279static struct cdevsw dead_cdevsw = {
280	.d_version =	D_VERSION,
281	.d_flags =	D_NEEDGIANT, /* XXX: does dead_strategy need this ? */
282	.d_open =	dead_open,
283	.d_close =	dead_close,
284	.d_read =	dead_read,
285	.d_write =	dead_write,
286	.d_ioctl =	dead_ioctl,
287	.d_poll =	dead_poll,
288	.d_mmap =	dead_mmap,
289	.d_strategy =	dead_strategy,
290	.d_name =	"dead",
291	.d_dump =	dead_dump,
292	.d_kqfilter =	dead_kqfilter
293};
294
295/* Default methods if driver does not specify method */
296
297#define null_open	(d_open_t *)nullop
298#define null_close	(d_close_t *)nullop
299#define no_read		(d_read_t *)enodev
300#define no_write	(d_write_t *)enodev
301#define no_ioctl	(d_ioctl_t *)enodev
302#define no_mmap		(d_mmap_t *)enodev
303#define no_kqfilter	(d_kqfilter_t *)enodev
304
305static void
306no_strategy(struct bio *bp)
307{
308
309	biofinish(bp, NULL, ENODEV);
310}
311
312static int
313no_poll(struct cdev *dev __unused, int events, struct thread *td __unused)
314{
315	/*
316	 * Return true for read/write.  If the user asked for something
317	 * special, return POLLNVAL, so that clients have a way of
318	 * determining reliably whether or not the extended
319	 * functionality is present without hard-coding knowledge
320	 * of specific filesystem implementations.
321	 * Stay in sync with vop_nopoll().
322	 */
323	if (events & ~POLLSTANDARD)
324		return (POLLNVAL);
325
326	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
327}
328
329#define no_dump		(dumper_t *)enodev
330
331static int
332giant_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
333{
334	struct cdevsw *dsw;
335	int retval;
336
337	dsw = dev_refthread(dev);
338	if (dsw == NULL)
339		return (ENXIO);
340	mtx_lock(&Giant);
341	retval = dsw->d_gianttrick->d_open(dev, oflags, devtype, td);
342	mtx_unlock(&Giant);
343	dev_relthread(dev);
344	return (retval);
345}
346
347static int
348giant_fdopen(struct cdev *dev, int oflags, struct thread *td, struct file *fp)
349{
350	struct cdevsw *dsw;
351	int retval;
352
353	dsw = dev_refthread(dev);
354	if (dsw == NULL)
355		return (ENXIO);
356	mtx_lock(&Giant);
357	retval = dsw->d_gianttrick->d_fdopen(dev, oflags, td, fp);
358	mtx_unlock(&Giant);
359	dev_relthread(dev);
360	return (retval);
361}
362
363static int
364giant_close(struct cdev *dev, int fflag, int devtype, struct thread *td)
365{
366	struct cdevsw *dsw;
367	int retval;
368
369	dsw = dev_refthread(dev);
370	if (dsw == NULL)
371		return (ENXIO);
372	mtx_lock(&Giant);
373	retval = dsw->d_gianttrick->d_close(dev, fflag, devtype, td);
374	mtx_unlock(&Giant);
375	dev_relthread(dev);
376	return (retval);
377}
378
379static void
380giant_strategy(struct bio *bp)
381{
382	struct cdevsw *dsw;
383	struct cdev *dev;
384
385	dev = bp->bio_dev;
386	dsw = dev_refthread(dev);
387	if (dsw == NULL) {
388		biofinish(bp, NULL, ENXIO);
389		return;
390	}
391	mtx_lock(&Giant);
392	dsw->d_gianttrick->d_strategy(bp);
393	mtx_unlock(&Giant);
394	dev_relthread(dev);
395}
396
397static int
398giant_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
399{
400	struct cdevsw *dsw;
401	int retval;
402
403	dsw = dev_refthread(dev);
404	if (dsw == NULL)
405		return (ENXIO);
406	mtx_lock(&Giant);
407	retval = dsw->d_gianttrick->d_ioctl(dev, cmd, data, fflag, td);
408	mtx_unlock(&Giant);
409	dev_relthread(dev);
410	return (retval);
411}
412
413static int
414giant_read(struct cdev *dev, struct uio *uio, int ioflag)
415{
416	struct cdevsw *dsw;
417	int retval;
418
419	dsw = dev_refthread(dev);
420	if (dsw == NULL)
421		return (ENXIO);
422	mtx_lock(&Giant);
423	retval = dsw->d_gianttrick->d_read(dev, uio, ioflag);
424	mtx_unlock(&Giant);
425	dev_relthread(dev);
426	return (retval);
427}
428
429static int
430giant_write(struct cdev *dev, struct uio *uio, int ioflag)
431{
432	struct cdevsw *dsw;
433	int retval;
434
435	dsw = dev_refthread(dev);
436	if (dsw == NULL)
437		return (ENXIO);
438	mtx_lock(&Giant);
439	retval = dsw->d_gianttrick->d_write(dev, uio, ioflag);
440	mtx_unlock(&Giant);
441	dev_relthread(dev);
442	return (retval);
443}
444
445static int
446giant_poll(struct cdev *dev, int events, struct thread *td)
447{
448	struct cdevsw *dsw;
449	int retval;
450
451	dsw = dev_refthread(dev);
452	if (dsw == NULL)
453		return (ENXIO);
454	mtx_lock(&Giant);
455	retval = dsw->d_gianttrick->d_poll(dev, events, td);
456	mtx_unlock(&Giant);
457	dev_relthread(dev);
458	return (retval);
459}
460
461static int
462giant_kqfilter(struct cdev *dev, struct knote *kn)
463{
464	struct cdevsw *dsw;
465	int retval;
466
467	dsw = dev_refthread(dev);
468	if (dsw == NULL)
469		return (ENXIO);
470	mtx_lock(&Giant);
471	retval = dsw->d_gianttrick->d_kqfilter(dev, kn);
472	mtx_unlock(&Giant);
473	dev_relthread(dev);
474	return (retval);
475}
476
477static int
478giant_mmap(struct cdev *dev, vm_offset_t offset, vm_paddr_t *paddr, int nprot)
479{
480	struct cdevsw *dsw;
481	int retval;
482
483	dsw = dev_refthread(dev);
484	if (dsw == NULL)
485		return (ENXIO);
486	mtx_lock(&Giant);
487	retval = dsw->d_gianttrick->d_mmap(dev, offset, paddr, nprot);
488	mtx_unlock(&Giant);
489	dev_relthread(dev);
490	return (retval);
491}
492
493
494static void
495notify(struct cdev *dev, const char *ev)
496{
497	static const char prefix[] = "cdev=";
498	char *data;
499	int namelen;
500
501	if (cold)
502		return;
503	namelen = strlen(dev->si_name);
504	data = malloc(namelen + sizeof(prefix), M_TEMP, M_WAITOK);
505	memcpy(data, prefix, sizeof(prefix) - 1);
506	memcpy(data + sizeof(prefix) - 1, dev->si_name, namelen + 1);
507	devctl_notify("DEVFS", "CDEV", ev, data);
508	free(data, M_TEMP);
509}
510
511static void
512notify_create(struct cdev *dev)
513{
514
515	notify(dev, "CREATE");
516}
517
518static void
519notify_destroy(struct cdev *dev)
520{
521
522	notify(dev, "DESTROY");
523}
524
525static struct cdev *
526newdev(struct cdevsw *csw, int y, struct cdev *si)
527{
528	struct cdev *si2;
529	dev_t	udev;
530
531	mtx_assert(&devmtx, MA_OWNED);
532	udev = y;
533	if (csw->d_flags & D_NEEDMINOR) {
534		/* We may want to return an existing device */
535		LIST_FOREACH(si2, &csw->d_devs, si_list) {
536			if (si2->si_drv0 == udev) {
537				dev_free_devlocked(si);
538				return (si2);
539			}
540		}
541	}
542	si->si_drv0 = udev;
543	si->si_devsw = csw;
544	LIST_INSERT_HEAD(&csw->d_devs, si, si_list);
545	return (si);
546}
547
548#define UMINORMASK	0xffff00ffU
549
550int
551uminor(dev_t dev)
552{
553	return (dev & UMINORMASK);
554}
555
556int
557umajor(dev_t dev)
558{
559	return ((dev & ~UMINORMASK) >> 8);
560}
561
562static void
563fini_cdevsw(struct cdevsw *devsw)
564{
565	struct cdevsw *gt;
566
567	if (devsw->d_gianttrick != NULL) {
568		gt = devsw->d_gianttrick;
569		memcpy(devsw, gt, sizeof *devsw);
570		cdevsw_free_devlocked(gt);
571		devsw->d_gianttrick = NULL;
572	}
573	devsw->d_flags &= ~D_INIT;
574}
575
576static void
577prep_cdevsw(struct cdevsw *devsw)
578{
579	struct cdevsw *dsw2;
580
581	mtx_assert(&devmtx, MA_OWNED);
582	if (devsw->d_flags & D_INIT)
583		return;
584	if (devsw->d_flags & D_NEEDGIANT) {
585		dev_unlock();
586		dsw2 = malloc(sizeof *dsw2, M_DEVT, M_WAITOK);
587		dev_lock();
588	} else
589		dsw2 = NULL;
590	if (devsw->d_flags & D_INIT) {
591		if (dsw2 != NULL)
592			cdevsw_free_devlocked(dsw2);
593		return;
594	}
595
596	if (devsw->d_version != D_VERSION_01) {
597		printf(
598		    "WARNING: Device driver \"%s\" has wrong version %s\n",
599		    devsw->d_name == NULL ? "???" : devsw->d_name,
600		    "and is disabled.  Recompile KLD module.");
601		devsw->d_open = dead_open;
602		devsw->d_close = dead_close;
603		devsw->d_read = dead_read;
604		devsw->d_write = dead_write;
605		devsw->d_ioctl = dead_ioctl;
606		devsw->d_poll = dead_poll;
607		devsw->d_mmap = dead_mmap;
608		devsw->d_strategy = dead_strategy;
609		devsw->d_dump = dead_dump;
610		devsw->d_kqfilter = dead_kqfilter;
611	}
612
613	if (devsw->d_flags & D_NEEDGIANT) {
614		if (devsw->d_gianttrick == NULL) {
615			memcpy(dsw2, devsw, sizeof *dsw2);
616			devsw->d_gianttrick = dsw2;
617			dsw2 = NULL;
618		}
619	}
620
621#define FIXUP(member, noop, giant) 				\
622	do {							\
623		if (devsw->member == NULL) {			\
624			devsw->member = noop;			\
625		} else if (devsw->d_flags & D_NEEDGIANT)	\
626			devsw->member = giant;			\
627		}						\
628	while (0)
629
630	FIXUP(d_open,		null_open,	giant_open);
631	FIXUP(d_fdopen,		NULL,		giant_fdopen);
632	FIXUP(d_close,		null_close,	giant_close);
633	FIXUP(d_read,		no_read,	giant_read);
634	FIXUP(d_write,		no_write,	giant_write);
635	FIXUP(d_ioctl,		no_ioctl,	giant_ioctl);
636	FIXUP(d_poll,		no_poll,	giant_poll);
637	FIXUP(d_mmap,		no_mmap,	giant_mmap);
638	FIXUP(d_strategy,	no_strategy,	giant_strategy);
639	FIXUP(d_kqfilter,	no_kqfilter,	giant_kqfilter);
640
641	if (devsw->d_dump == NULL)	devsw->d_dump = no_dump;
642
643	LIST_INIT(&devsw->d_devs);
644
645	devsw->d_flags |= D_INIT;
646
647	if (dsw2 != NULL)
648		cdevsw_free_devlocked(dsw2);
649}
650
651struct cdev *
652make_dev_credv(int flags, struct cdevsw *devsw, int minornr,
653    struct ucred *cr, uid_t uid,
654    gid_t gid, int mode, const char *fmt, va_list ap)
655{
656	struct cdev *dev;
657	int i;
658
659	dev = devfs_alloc();
660	dev_lock();
661	prep_cdevsw(devsw);
662	dev = newdev(devsw, minornr, dev);
663	if (flags & MAKEDEV_REF)
664		dev_refl(dev);
665	if (dev->si_flags & SI_CHEAPCLONE &&
666	    dev->si_flags & SI_NAMED) {
667		/*
668		 * This is allowed as it removes races and generally
669		 * simplifies cloning devices.
670		 * XXX: still ??
671		 */
672		dev_unlock_and_free();
673		return (dev);
674	}
675	KASSERT(!(dev->si_flags & SI_NAMED),
676	    ("make_dev() by driver %s on pre-existing device (min=%x, name=%s)",
677	    devsw->d_name, minor(dev), devtoname(dev)));
678
679	i = vsnrprintf(dev->__si_namebuf, sizeof dev->__si_namebuf, 32, fmt, ap);
680	if (i > (sizeof dev->__si_namebuf - 1)) {
681		printf("WARNING: Device name truncated! (%s)\n",
682		    dev->__si_namebuf);
683	}
684
685	dev->si_flags |= SI_NAMED;
686	if (cr != NULL)
687		dev->si_cred = crhold(cr);
688	else
689		dev->si_cred = NULL;
690	dev->si_uid = uid;
691	dev->si_gid = gid;
692	dev->si_mode = mode;
693
694	devfs_create(dev);
695	clean_unrhdrl(devfs_inos);
696	dev_unlock_and_free();
697
698	notify_create(dev);
699
700	return (dev);
701}
702
703struct cdev *
704make_dev(struct cdevsw *devsw, int minornr, uid_t uid, gid_t gid, int mode,
705    const char *fmt, ...)
706{
707	struct cdev *dev;
708	va_list ap;
709
710	va_start(ap, fmt);
711	dev = make_dev_credv(0, devsw, minornr, NULL, uid, gid, mode, fmt, ap);
712	va_end(ap);
713	return (dev);
714}
715
716struct cdev *
717make_dev_cred(struct cdevsw *devsw, int minornr, struct ucred *cr, uid_t uid,
718    gid_t gid, int mode, const char *fmt, ...)
719{
720	struct cdev *dev;
721	va_list ap;
722
723	va_start(ap, fmt);
724	dev = make_dev_credv(0, devsw, minornr, cr, uid, gid, mode, fmt, ap);
725	va_end(ap);
726
727	return (dev);
728}
729
730struct cdev *
731make_dev_credf(int flags, struct cdevsw *devsw, int minornr,
732    struct ucred *cr, uid_t uid,
733    gid_t gid, int mode, const char *fmt, ...)
734{
735	struct cdev *dev;
736	va_list ap;
737
738	va_start(ap, fmt);
739	dev = make_dev_credv(flags, devsw, minornr, cr, uid, gid, mode,
740	    fmt, ap);
741	va_end(ap);
742
743	return (dev);
744}
745
746static void
747dev_dependsl(struct cdev *pdev, struct cdev *cdev)
748{
749
750	cdev->si_parent = pdev;
751	cdev->si_flags |= SI_CHILD;
752	LIST_INSERT_HEAD(&pdev->si_children, cdev, si_siblings);
753}
754
755
756void
757dev_depends(struct cdev *pdev, struct cdev *cdev)
758{
759
760	dev_lock();
761	dev_dependsl(pdev, cdev);
762	dev_unlock();
763}
764
765struct cdev *
766make_dev_alias(struct cdev *pdev, const char *fmt, ...)
767{
768	struct cdev *dev;
769	va_list ap;
770	int i;
771
772	KASSERT(pdev != NULL, ("NULL pdev"));
773	dev = devfs_alloc();
774	dev_lock();
775	dev->si_flags |= SI_ALIAS;
776	dev->si_flags |= SI_NAMED;
777	va_start(ap, fmt);
778	i = vsnrprintf(dev->__si_namebuf, sizeof dev->__si_namebuf, 32, fmt, ap);
779	if (i > (sizeof dev->__si_namebuf - 1)) {
780		printf("WARNING: Device name truncated! (%s)\n",
781		    dev->__si_namebuf);
782	}
783	va_end(ap);
784
785	devfs_create(dev);
786	dev_dependsl(pdev, dev);
787	clean_unrhdrl(devfs_inos);
788	dev_unlock();
789
790	notify_create(dev);
791
792	return (dev);
793}
794
795static void
796destroy_devl(struct cdev *dev)
797{
798	struct cdevsw *csw;
799	struct cdev_privdata *p, *p1;
800
801	mtx_assert(&devmtx, MA_OWNED);
802	KASSERT(dev->si_flags & SI_NAMED,
803	    ("WARNING: Driver mistake: destroy_dev on %d\n", minor(dev)));
804
805	devfs_destroy(dev);
806
807	/* Remove name marking */
808	dev->si_flags &= ~SI_NAMED;
809
810	/* If we are a child, remove us from the parents list */
811	if (dev->si_flags & SI_CHILD) {
812		LIST_REMOVE(dev, si_siblings);
813		dev->si_flags &= ~SI_CHILD;
814	}
815
816	/* Kill our children */
817	while (!LIST_EMPTY(&dev->si_children))
818		destroy_devl(LIST_FIRST(&dev->si_children));
819
820	/* Remove from clone list */
821	if (dev->si_flags & SI_CLONELIST) {
822		LIST_REMOVE(dev, si_clone);
823		dev->si_flags &= ~SI_CLONELIST;
824	}
825
826	dev->si_refcount++;	/* Avoid race with dev_rel() */
827	csw = dev->si_devsw;
828	dev->si_devsw = NULL;	/* already NULL for SI_ALIAS */
829	while (csw != NULL && csw->d_purge != NULL && dev->si_threadcount) {
830		csw->d_purge(dev);
831		msleep(csw, &devmtx, PRIBIO, "devprg", hz/10);
832		if (dev->si_threadcount)
833			printf("Still %lu threads in %s\n",
834			    dev->si_threadcount, devtoname(dev));
835	}
836	while (dev->si_threadcount != 0) {
837		/* Use unique dummy wait ident */
838		msleep(&csw, &devmtx, PRIBIO, "devdrn", hz / 10);
839	}
840
841	dev_unlock();
842	notify_destroy(dev);
843	mtx_lock(&cdevpriv_mtx);
844	LIST_FOREACH_SAFE(p, &cdev2priv(dev)->cdp_fdpriv, cdpd_list, p1) {
845		devfs_destroy_cdevpriv(p);
846		mtx_lock(&cdevpriv_mtx);
847	}
848	mtx_unlock(&cdevpriv_mtx);
849	dev_lock();
850
851	dev->si_drv1 = 0;
852	dev->si_drv2 = 0;
853	bzero(&dev->__si_u, sizeof(dev->__si_u));
854
855	if (!(dev->si_flags & SI_ALIAS)) {
856		/* Remove from cdevsw list */
857		LIST_REMOVE(dev, si_list);
858
859		/* If cdevsw has no more struct cdev *'s, clean it */
860		if (LIST_EMPTY(&csw->d_devs)) {
861			fini_cdevsw(csw);
862			wakeup(&csw->d_devs);
863		}
864	}
865	dev->si_flags &= ~SI_ALIAS;
866	dev->si_refcount--;	/* Avoid race with dev_rel() */
867
868	if (dev->si_refcount > 0) {
869		LIST_INSERT_HEAD(&dead_cdevsw.d_devs, dev, si_list);
870	} else {
871		dev_free_devlocked(dev);
872	}
873}
874
875void
876destroy_dev(struct cdev *dev)
877{
878
879	dev_lock();
880	destroy_devl(dev);
881	dev_unlock_and_free();
882}
883
884const char *
885devtoname(struct cdev *dev)
886{
887	char *p;
888	struct cdevsw *csw;
889	int mynor;
890
891	if (dev->si_name[0] == '#' || dev->si_name[0] == '\0') {
892		p = dev->si_name;
893		csw = dev_refthread(dev);
894		if (csw != NULL) {
895			sprintf(p, "(%s)", csw->d_name);
896			dev_relthread(dev);
897		}
898		p += strlen(p);
899		mynor = minor(dev);
900		if (mynor < 0 || mynor > 255)
901			sprintf(p, "/%#x", (u_int)mynor);
902		else
903			sprintf(p, "/%d", mynor);
904	}
905	return (dev->si_name);
906}
907
908int
909dev_stdclone(char *name, char **namep, const char *stem, int *unit)
910{
911	int u, i;
912
913	i = strlen(stem);
914	if (bcmp(stem, name, i) != 0)
915		return (0);
916	if (!isdigit(name[i]))
917		return (0);
918	u = 0;
919	if (name[i] == '0' && isdigit(name[i+1]))
920		return (0);
921	while (isdigit(name[i])) {
922		u *= 10;
923		u += name[i++] - '0';
924	}
925	if (u > 0xffffff)
926		return (0);
927	*unit = u;
928	if (namep)
929		*namep = &name[i];
930	if (name[i])
931		return (2);
932	return (1);
933}
934
935/*
936 * Helper functions for cloning device drivers.
937 *
938 * The objective here is to make it unnecessary for the device drivers to
939 * use rman or similar to manage their unit number space.  Due to the way
940 * we do "on-demand" devices, using rman or other "private" methods
941 * will be very tricky to lock down properly once we lock down this file.
942 *
943 * Instead we give the drivers these routines which puts the struct cdev *'s
944 * that are to be managed on their own list, and gives the driver the ability
945 * to ask for the first free unit number or a given specified unit number.
946 *
947 * In addition these routines support paired devices (pty, nmdm and similar)
948 * by respecting a number of "flag" bits in the minor number.
949 *
950 */
951
952struct clonedevs {
953	LIST_HEAD(,cdev)	head;
954};
955
956void
957clone_setup(struct clonedevs **cdp)
958{
959
960	*cdp = malloc(sizeof **cdp, M_DEVBUF, M_WAITOK | M_ZERO);
961	LIST_INIT(&(*cdp)->head);
962}
963
964int
965clone_create(struct clonedevs **cdp, struct cdevsw *csw, int *up, struct cdev **dp, int extra)
966{
967	struct clonedevs *cd;
968	struct cdev *dev, *ndev, *dl, *de;
969	int unit, low, u;
970
971	KASSERT(*cdp != NULL,
972	    ("clone_setup() not called in driver \"%s\"", csw->d_name));
973	KASSERT(!(extra & CLONE_UNITMASK),
974	    ("Illegal extra bits (0x%x) in clone_create", extra));
975	KASSERT(*up <= CLONE_UNITMASK,
976	    ("Too high unit (0x%x) in clone_create", *up));
977	KASSERT(csw->d_flags & D_NEEDMINOR,
978	    ("clone_create() on cdevsw without minor numbers"));
979
980
981	/*
982	 * Search the list for a lot of things in one go:
983	 *   A preexisting match is returned immediately.
984	 *   The lowest free unit number if we are passed -1, and the place
985	 *	 in the list where we should insert that new element.
986	 *   The place to insert a specified unit number, if applicable
987	 *       the end of the list.
988	 */
989	unit = *up;
990	ndev = devfs_alloc();
991	dev_lock();
992	prep_cdevsw(csw);
993	low = extra;
994	de = dl = NULL;
995	cd = *cdp;
996	LIST_FOREACH(dev, &cd->head, si_clone) {
997		KASSERT(dev->si_flags & SI_CLONELIST,
998		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
999		u = dev2unit(dev);
1000		if (u == (unit | extra)) {
1001			*dp = dev;
1002			dev_unlock();
1003			devfs_free(ndev);
1004			return (0);
1005		}
1006		if (unit == -1 && u == low) {
1007			low++;
1008			de = dev;
1009			continue;
1010		} else if (u < (unit | extra)) {
1011			de = dev;
1012			continue;
1013		} else if (u > (unit | extra)) {
1014			dl = dev;
1015			break;
1016		}
1017	}
1018	if (unit == -1)
1019		unit = low & CLONE_UNITMASK;
1020	dev = newdev(csw, unit2minor(unit | extra), ndev);
1021	if (dev->si_flags & SI_CLONELIST) {
1022		printf("dev %p (%s) is on clonelist\n", dev, dev->si_name);
1023		printf("unit=%d, low=%d, extra=0x%x\n", unit, low, extra);
1024		LIST_FOREACH(dev, &cd->head, si_clone) {
1025			printf("\t%p %s\n", dev, dev->si_name);
1026		}
1027		panic("foo");
1028	}
1029	KASSERT(!(dev->si_flags & SI_CLONELIST),
1030	    ("Dev %p(%s) should not be on clonelist", dev, dev->si_name));
1031	if (dl != NULL)
1032		LIST_INSERT_BEFORE(dl, dev, si_clone);
1033	else if (de != NULL)
1034		LIST_INSERT_AFTER(de, dev, si_clone);
1035	else
1036		LIST_INSERT_HEAD(&cd->head, dev, si_clone);
1037	dev->si_flags |= SI_CLONELIST;
1038	*up = unit;
1039	dev_unlock_and_free();
1040	return (1);
1041}
1042
1043/*
1044 * Kill everything still on the list.  The driver should already have
1045 * disposed of any softc hung of the struct cdev *'s at this time.
1046 */
1047void
1048clone_cleanup(struct clonedevs **cdp)
1049{
1050	struct cdev *dev;
1051	struct cdev_priv *cp;
1052	struct clonedevs *cd;
1053
1054	cd = *cdp;
1055	if (cd == NULL)
1056		return;
1057	dev_lock();
1058	while (!LIST_EMPTY(&cd->head)) {
1059		dev = LIST_FIRST(&cd->head);
1060		LIST_REMOVE(dev, si_clone);
1061		KASSERT(dev->si_flags & SI_CLONELIST,
1062		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
1063		dev->si_flags &= ~SI_CLONELIST;
1064		cp = cdev2priv(dev);
1065		if (!(cp->cdp_flags & CDP_SCHED_DTR)) {
1066			cp->cdp_flags |= CDP_SCHED_DTR;
1067			KASSERT(dev->si_flags & SI_NAMED,
1068				("Driver has goofed in cloning underways udev %x", dev->si_drv0));
1069			destroy_devl(dev);
1070		}
1071	}
1072	dev_unlock_and_free();
1073	free(cd, M_DEVBUF);
1074	*cdp = NULL;
1075}
1076
1077static TAILQ_HEAD(, cdev_priv) dev_ddtr =
1078	TAILQ_HEAD_INITIALIZER(dev_ddtr);
1079static struct task dev_dtr_task;
1080
1081static void
1082destroy_dev_tq(void *ctx, int pending)
1083{
1084	struct cdev_priv *cp;
1085	struct cdev *dev;
1086	void (*cb)(void *);
1087	void *cb_arg;
1088
1089	dev_lock();
1090	while (!TAILQ_EMPTY(&dev_ddtr)) {
1091		cp = TAILQ_FIRST(&dev_ddtr);
1092		dev = &cp->cdp_c;
1093		KASSERT(cp->cdp_flags & CDP_SCHED_DTR,
1094		    ("cdev %p in dev_destroy_tq without CDP_SCHED_DTR", cp));
1095		TAILQ_REMOVE(&dev_ddtr, cp, cdp_dtr_list);
1096		cb = cp->cdp_dtr_cb;
1097		cb_arg = cp->cdp_dtr_cb_arg;
1098		destroy_devl(dev);
1099		dev_unlock_and_free();
1100		dev_rel(dev);
1101		if (cb != NULL)
1102			cb(cb_arg);
1103		dev_lock();
1104	}
1105	dev_unlock();
1106}
1107
1108/*
1109 * devmtx shall be locked on entry. devmtx will be unlocked after
1110 * function return.
1111 */
1112static int
1113destroy_dev_sched_cbl(struct cdev *dev, void (*cb)(void *), void *arg)
1114{
1115	struct cdev_priv *cp;
1116
1117	mtx_assert(&devmtx, MA_OWNED);
1118	cp = cdev2priv(dev);
1119	if (cp->cdp_flags & CDP_SCHED_DTR) {
1120		dev_unlock();
1121		return (0);
1122	}
1123	dev_refl(dev);
1124	cp->cdp_flags |= CDP_SCHED_DTR;
1125	cp->cdp_dtr_cb = cb;
1126	cp->cdp_dtr_cb_arg = arg;
1127	TAILQ_INSERT_TAIL(&dev_ddtr, cp, cdp_dtr_list);
1128	dev_unlock();
1129	taskqueue_enqueue(taskqueue_swi_giant, &dev_dtr_task);
1130	return (1);
1131}
1132
1133int
1134destroy_dev_sched_cb(struct cdev *dev, void (*cb)(void *), void *arg)
1135{
1136	dev_lock();
1137	return (destroy_dev_sched_cbl(dev, cb, arg));
1138}
1139
1140int
1141destroy_dev_sched(struct cdev *dev)
1142{
1143	return (destroy_dev_sched_cb(dev, NULL, NULL));
1144}
1145
1146void
1147destroy_dev_drain(struct cdevsw *csw)
1148{
1149
1150	dev_lock();
1151	while (!LIST_EMPTY(&csw->d_devs)) {
1152		msleep(&csw->d_devs, &devmtx, PRIBIO, "devscd", hz/10);
1153	}
1154	dev_unlock();
1155}
1156
1157void
1158drain_dev_clone_events(void)
1159{
1160
1161	sx_xlock(&clone_drain_lock);
1162	sx_xunlock(&clone_drain_lock);
1163}
1164
1165static void
1166devdtr_init(void *dummy __unused)
1167{
1168
1169	TASK_INIT(&dev_dtr_task, 0, destroy_dev_tq, NULL);
1170}
1171
1172SYSINIT(devdtr, SI_SUB_DEVFS, SI_ORDER_SECOND, devdtr_init, NULL);
1173