xref: /freebsd/sys/kern/kern_conf.c (revision 10f0bcab61ef441cb5af32fb706688d8cbd55dc0)
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("$FreeBSD$");
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/module.h>
37 #include <sys/malloc.h>
38 #include <sys/conf.h>
39 #include <sys/vnode.h>
40 #include <sys/queue.h>
41 #include <sys/poll.h>
42 #include <sys/sx.h>
43 #include <sys/ctype.h>
44 #include <sys/tty.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 
51 static MALLOC_DEFINE(M_DEVT, "cdev", "cdev storage");
52 
53 struct mtx devmtx;
54 static void destroy_devl(struct cdev *dev);
55 static int destroy_dev_sched_cbl(struct cdev *dev,
56     void (*cb)(void *), void *arg);
57 static 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 
62 static struct cdev_priv_list cdevp_free_list =
63     TAILQ_HEAD_INITIALIZER(cdevp_free_list);
64 static SLIST_HEAD(free_cdevsw, cdevsw) cdevsw_gt_post_list =
65     SLIST_HEAD_INITIALIZER();
66 
67 void
68 dev_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  */
80 static void
81 dev_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 
111 static void
112 dev_free_devlocked(struct cdev *cdev)
113 {
114 	struct cdev_priv *cdp;
115 
116 	mtx_assert(&devmtx, MA_OWNED);
117 	cdp = cdev->si_priv;
118 	TAILQ_INSERT_HEAD(&cdevp_free_list, cdp, cdp_list);
119 }
120 
121 static void
122 cdevsw_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 
129 void
130 dev_unlock(void)
131 {
132 
133 	mtx_unlock(&devmtx);
134 }
135 
136 void
137 dev_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 
146 void
147 dev_refl(struct cdev *dev)
148 {
149 
150 	mtx_assert(&devmtx, MA_OWNED);
151 	dev->si_refcount++;
152 }
153 
154 void
155 dev_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 
179 struct cdevsw *
180 dev_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 = dev->si_priv;
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 
199 struct cdevsw *
200 devvn_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 = (*devp)->si_priv;
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 
221 void
222 dev_relthread(struct cdev *dev)
223 {
224 
225 	mtx_assert(&devmtx, MA_NOTOWNED);
226 	dev_lock();
227 	dev->si_threadcount--;
228 	dev_unlock();
229 }
230 
231 int
232 nullop(void)
233 {
234 
235 	return (0);
236 }
237 
238 int
239 eopnotsupp(void)
240 {
241 
242 	return (EOPNOTSUPP);
243 }
244 
245 static int
246 enxio(void)
247 {
248 	return (ENXIO);
249 }
250 
251 static int
252 enodev(void)
253 {
254 	return (ENODEV);
255 }
256 
257 /* Define a dead_cdevsw for use when devices leave unexpectedly. */
258 
259 #define dead_open	(d_open_t *)enxio
260 #define dead_close	(d_close_t *)enxio
261 #define dead_read	(d_read_t *)enxio
262 #define dead_write	(d_write_t *)enxio
263 #define dead_ioctl	(d_ioctl_t *)enxio
264 #define dead_poll	(d_poll_t *)enodev
265 #define dead_mmap	(d_mmap_t *)enodev
266 
267 static void
268 dead_strategy(struct bio *bp)
269 {
270 
271 	biofinish(bp, NULL, ENXIO);
272 }
273 
274 #define dead_dump	(dumper_t *)enxio
275 #define dead_kqfilter	(d_kqfilter_t *)enxio
276 
277 static struct cdevsw dead_cdevsw = {
278 	.d_version =	D_VERSION,
279 	.d_flags =	D_NEEDGIANT, /* XXX: does dead_strategy need this ? */
280 	.d_open =	dead_open,
281 	.d_close =	dead_close,
282 	.d_read =	dead_read,
283 	.d_write =	dead_write,
284 	.d_ioctl =	dead_ioctl,
285 	.d_poll =	dead_poll,
286 	.d_mmap =	dead_mmap,
287 	.d_strategy =	dead_strategy,
288 	.d_name =	"dead",
289 	.d_dump =	dead_dump,
290 	.d_kqfilter =	dead_kqfilter
291 };
292 
293 /* Default methods if driver does not specify method */
294 
295 #define null_open	(d_open_t *)nullop
296 #define null_close	(d_close_t *)nullop
297 #define no_read		(d_read_t *)enodev
298 #define no_write	(d_write_t *)enodev
299 #define no_ioctl	(d_ioctl_t *)enodev
300 #define no_mmap		(d_mmap_t *)enodev
301 #define no_kqfilter	(d_kqfilter_t *)enodev
302 
303 static void
304 no_strategy(struct bio *bp)
305 {
306 
307 	biofinish(bp, NULL, ENODEV);
308 }
309 
310 static int
311 no_poll(struct cdev *dev __unused, int events, struct thread *td __unused)
312 {
313 	/*
314 	 * Return true for read/write.  If the user asked for something
315 	 * special, return POLLNVAL, so that clients have a way of
316 	 * determining reliably whether or not the extended
317 	 * functionality is present without hard-coding knowledge
318 	 * of specific filesystem implementations.
319 	 * Stay in sync with vop_nopoll().
320 	 */
321 	if (events & ~POLLSTANDARD)
322 		return (POLLNVAL);
323 
324 	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
325 }
326 
327 #define no_dump		(dumper_t *)enodev
328 
329 static int
330 giant_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
331 {
332 	struct cdevsw *dsw;
333 	int retval;
334 
335 	dsw = dev_refthread(dev);
336 	if (dsw == NULL)
337 		return (ENXIO);
338 	mtx_lock(&Giant);
339 	retval = dsw->d_gianttrick->d_open(dev, oflags, devtype, td);
340 	mtx_unlock(&Giant);
341 	dev_relthread(dev);
342 	return (retval);
343 }
344 
345 static int
346 giant_fdopen(struct cdev *dev, int oflags, struct thread *td, struct file *fp)
347 {
348 	struct cdevsw *dsw;
349 	int retval;
350 
351 	dsw = dev_refthread(dev);
352 	if (dsw == NULL)
353 		return (ENXIO);
354 	mtx_lock(&Giant);
355 	retval = dsw->d_gianttrick->d_fdopen(dev, oflags, td, fp);
356 	mtx_unlock(&Giant);
357 	dev_relthread(dev);
358 	return (retval);
359 }
360 
361 static int
362 giant_close(struct cdev *dev, int fflag, int devtype, struct thread *td)
363 {
364 	struct cdevsw *dsw;
365 	int retval;
366 
367 	dsw = dev_refthread(dev);
368 	if (dsw == NULL)
369 		return (ENXIO);
370 	mtx_lock(&Giant);
371 	retval = dsw->d_gianttrick->d_close(dev, fflag, devtype, td);
372 	mtx_unlock(&Giant);
373 	dev_relthread(dev);
374 	return (retval);
375 }
376 
377 static void
378 giant_strategy(struct bio *bp)
379 {
380 	struct cdevsw *dsw;
381 	struct cdev *dev;
382 
383 	dev = bp->bio_dev;
384 	dsw = dev_refthread(dev);
385 	if (dsw == NULL) {
386 		biofinish(bp, NULL, ENXIO);
387 		return;
388 	}
389 	mtx_lock(&Giant);
390 	dsw->d_gianttrick->d_strategy(bp);
391 	mtx_unlock(&Giant);
392 	dev_relthread(dev);
393 }
394 
395 static int
396 giant_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
397 {
398 	struct cdevsw *dsw;
399 	int retval;
400 
401 	dsw = dev_refthread(dev);
402 	if (dsw == NULL)
403 		return (ENXIO);
404 	mtx_lock(&Giant);
405 	retval = dev->si_devsw->d_gianttrick->
406 	    d_ioctl(dev, cmd, data, fflag, td);
407 	mtx_unlock(&Giant);
408 	dev_relthread(dev);
409 	return (retval);
410 }
411 
412 static int
413 giant_read(struct cdev *dev, struct uio *uio, int ioflag)
414 {
415 	struct cdevsw *dsw;
416 	int retval;
417 
418 	dsw = dev_refthread(dev);
419 	if (dsw == NULL)
420 		return (ENXIO);
421 	mtx_lock(&Giant);
422 	retval = dev->si_devsw->d_gianttrick->
423 	    d_read(dev, uio, ioflag);
424 	mtx_unlock(&Giant);
425 	dev_relthread(dev);
426 	return (retval);
427 }
428 
429 static int
430 giant_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 
445 static int
446 giant_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 
461 static int
462 giant_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 
477 static int
478 giant_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 
494 /*
495  * struct cdev * and u_dev_t primitives
496  */
497 
498 int
499 minor(struct cdev *x)
500 {
501 	if (x == NULL)
502 		return NODEV;
503 	return(x->si_drv0 & MAXMINOR);
504 }
505 
506 int
507 dev2unit(struct cdev *x)
508 {
509 
510 	if (x == NULL)
511 		return NODEV;
512 	return (minor2unit(minor(x)));
513 }
514 
515 u_int
516 minor2unit(u_int _minor)
517 {
518 
519 	KASSERT((_minor & ~MAXMINOR) == 0, ("Illegal minor %x", _minor));
520 	return ((_minor & 0xff) | ((_minor >> 8) & 0xffff00));
521 }
522 
523 int
524 unit2minor(int unit)
525 {
526 
527 	KASSERT(unit <= 0xffffff, ("Invalid unit (%d) in unit2minor", unit));
528 	return ((unit & 0xff) | ((unit << 8) & ~0xffff));
529 }
530 
531 static struct cdev *
532 newdev(struct cdevsw *csw, int y, struct cdev *si)
533 {
534 	struct cdev *si2;
535 	dev_t	udev;
536 
537 	mtx_assert(&devmtx, MA_OWNED);
538 	udev = y;
539 	LIST_FOREACH(si2, &csw->d_devs, si_list) {
540 		if (si2->si_drv0 == udev) {
541 			dev_free_devlocked(si);
542 			return (si2);
543 		}
544 	}
545 	si->si_drv0 = udev;
546 	si->si_devsw = csw;
547 	LIST_INSERT_HEAD(&csw->d_devs, si, si_list);
548 	return (si);
549 }
550 
551 int
552 uminor(dev_t dev)
553 {
554 	return (dev & MAXMINOR);
555 }
556 
557 int
558 umajor(dev_t dev)
559 {
560 	return ((dev & ~MAXMINOR) >> 8);
561 }
562 
563 static void
564 fini_cdevsw(struct cdevsw *devsw)
565 {
566 	struct cdevsw *gt;
567 
568 	if (devsw->d_gianttrick != NULL) {
569 		gt = devsw->d_gianttrick;
570 		memcpy(devsw, gt, sizeof *devsw);
571 		cdevsw_free_devlocked(gt);
572 		devsw->d_gianttrick = NULL;
573 	}
574 	devsw->d_flags &= ~D_INIT;
575 }
576 
577 static void
578 prep_cdevsw(struct cdevsw *devsw)
579 {
580 	struct cdevsw *dsw2;
581 
582 	mtx_assert(&devmtx, MA_OWNED);
583 	if (devsw->d_flags & D_INIT)
584 		return;
585 	if (devsw->d_flags & D_NEEDGIANT) {
586 		dev_unlock();
587 		dsw2 = malloc(sizeof *dsw2, M_DEVT, M_WAITOK);
588 		dev_lock();
589 	} else
590 		dsw2 = NULL;
591 	if (devsw->d_flags & D_INIT) {
592 		if (dsw2 != NULL)
593 			cdevsw_free_devlocked(dsw2);
594 		return;
595 	}
596 
597 	if (devsw->d_version != D_VERSION_01) {
598 		printf(
599 		    "WARNING: Device driver \"%s\" has wrong version %s\n",
600 		    devsw->d_name == NULL ? "???" : devsw->d_name,
601 		    "and is disabled.  Recompile KLD module.");
602 		devsw->d_open = dead_open;
603 		devsw->d_close = dead_close;
604 		devsw->d_read = dead_read;
605 		devsw->d_write = dead_write;
606 		devsw->d_ioctl = dead_ioctl;
607 		devsw->d_poll = dead_poll;
608 		devsw->d_mmap = dead_mmap;
609 		devsw->d_strategy = dead_strategy;
610 		devsw->d_dump = dead_dump;
611 		devsw->d_kqfilter = dead_kqfilter;
612 	}
613 
614 	if (devsw->d_flags & D_TTY) {
615 		if (devsw->d_ioctl == NULL)	devsw->d_ioctl = ttyioctl;
616 		if (devsw->d_read == NULL)	devsw->d_read = ttyread;
617 		if (devsw->d_write == NULL)	devsw->d_write = ttywrite;
618 		if (devsw->d_kqfilter == NULL)	devsw->d_kqfilter = ttykqfilter;
619 		if (devsw->d_poll == NULL)	devsw->d_poll = ttypoll;
620 	}
621 
622 	if (devsw->d_flags & D_NEEDGIANT) {
623 		if (devsw->d_gianttrick == NULL) {
624 			memcpy(dsw2, devsw, sizeof *dsw2);
625 			devsw->d_gianttrick = dsw2;
626 			dsw2 = NULL;
627 		}
628 	}
629 
630 #define FIXUP(member, noop, giant) 				\
631 	do {							\
632 		if (devsw->member == NULL) {			\
633 			devsw->member = noop;			\
634 		} else if (devsw->d_flags & D_NEEDGIANT)	\
635 			devsw->member = giant;			\
636 		}						\
637 	while (0)
638 
639 	FIXUP(d_open,		null_open,	giant_open);
640 	FIXUP(d_fdopen,		NULL,		giant_fdopen);
641 	FIXUP(d_close,		null_close,	giant_close);
642 	FIXUP(d_read,		no_read,	giant_read);
643 	FIXUP(d_write,		no_write,	giant_write);
644 	FIXUP(d_ioctl,		no_ioctl,	giant_ioctl);
645 	FIXUP(d_poll,		no_poll,	giant_poll);
646 	FIXUP(d_mmap,		no_mmap,	giant_mmap);
647 	FIXUP(d_strategy,	no_strategy,	giant_strategy);
648 	FIXUP(d_kqfilter,	no_kqfilter,	giant_kqfilter);
649 
650 	if (devsw->d_dump == NULL)	devsw->d_dump = no_dump;
651 
652 	LIST_INIT(&devsw->d_devs);
653 
654 	devsw->d_flags |= D_INIT;
655 
656 	if (dsw2 != NULL)
657 		cdevsw_free_devlocked(dsw2);
658 }
659 
660 struct cdev *
661 make_dev_credv(int flags, struct cdevsw *devsw, int minornr,
662     struct ucred *cr, uid_t uid,
663     gid_t gid, int mode, const char *fmt, va_list ap)
664 {
665 	struct cdev *dev;
666 	int i;
667 
668 	KASSERT((minornr & ~MAXMINOR) == 0,
669 	    ("Invalid minor (0x%x) in make_dev", minornr));
670 
671 	dev = devfs_alloc();
672 	dev_lock();
673 	prep_cdevsw(devsw);
674 	dev = newdev(devsw, minornr, dev);
675 	if (flags & MAKEDEV_REF)
676 		dev_refl(dev);
677 	if (dev->si_flags & SI_CHEAPCLONE &&
678 	    dev->si_flags & SI_NAMED) {
679 		/*
680 		 * This is allowed as it removes races and generally
681 		 * simplifies cloning devices.
682 		 * XXX: still ??
683 		 */
684 		dev_unlock_and_free();
685 		return (dev);
686 	}
687 	KASSERT(!(dev->si_flags & SI_NAMED),
688 	    ("make_dev() by driver %s on pre-existing device (min=%x, name=%s)",
689 	    devsw->d_name, minor(dev), devtoname(dev)));
690 
691 	i = vsnrprintf(dev->__si_namebuf, sizeof dev->__si_namebuf, 32, fmt, ap);
692 	if (i > (sizeof dev->__si_namebuf - 1)) {
693 		printf("WARNING: Device name truncated! (%s)\n",
694 		    dev->__si_namebuf);
695 	}
696 
697 	dev->si_flags |= SI_NAMED;
698 #ifdef MAC
699 	if (cr != NULL)
700 		dev->si_cred = crhold(cr);
701 	else
702 #endif
703 		dev->si_cred = NULL;
704 	dev->si_uid = uid;
705 	dev->si_gid = gid;
706 	dev->si_mode = mode;
707 
708 	devfs_create(dev);
709 	clean_unrhdrl(devfs_inos);
710 	dev_unlock_and_free();
711 	return (dev);
712 }
713 
714 struct cdev *
715 make_dev(struct cdevsw *devsw, int minornr, uid_t uid, gid_t gid, int mode,
716     const char *fmt, ...)
717 {
718 	struct cdev *dev;
719 	va_list ap;
720 
721 	va_start(ap, fmt);
722 	dev = make_dev_credv(0, devsw, minornr, NULL, uid, gid, mode, fmt, ap);
723 	va_end(ap);
724 	return (dev);
725 }
726 
727 struct cdev *
728 make_dev_cred(struct cdevsw *devsw, int minornr, struct ucred *cr, uid_t uid,
729     gid_t gid, int mode, const char *fmt, ...)
730 {
731 	struct cdev *dev;
732 	va_list ap;
733 
734 	va_start(ap, fmt);
735 	dev = make_dev_credv(0, devsw, minornr, cr, uid, gid, mode, fmt, ap);
736 	va_end(ap);
737 
738 	return (dev);
739 }
740 
741 struct cdev *
742 make_dev_credf(int flags, struct cdevsw *devsw, int minornr,
743     struct ucred *cr, uid_t uid,
744     gid_t gid, int mode, const char *fmt, ...)
745 {
746 	struct cdev *dev;
747 	va_list ap;
748 
749 	va_start(ap, fmt);
750 	dev = make_dev_credv(flags, devsw, minornr, cr, uid, gid, mode,
751 	    fmt, ap);
752 	va_end(ap);
753 
754 	return (dev);
755 }
756 
757 static void
758 dev_dependsl(struct cdev *pdev, struct cdev *cdev)
759 {
760 
761 	cdev->si_parent = pdev;
762 	cdev->si_flags |= SI_CHILD;
763 	LIST_INSERT_HEAD(&pdev->si_children, cdev, si_siblings);
764 }
765 
766 
767 void
768 dev_depends(struct cdev *pdev, struct cdev *cdev)
769 {
770 
771 	dev_lock();
772 	dev_dependsl(pdev, cdev);
773 	dev_unlock();
774 }
775 
776 struct cdev *
777 make_dev_alias(struct cdev *pdev, const char *fmt, ...)
778 {
779 	struct cdev *dev;
780 	va_list ap;
781 	int i;
782 
783 	dev = devfs_alloc();
784 	dev_lock();
785 	dev->si_flags |= SI_ALIAS;
786 	dev->si_flags |= SI_NAMED;
787 	va_start(ap, fmt);
788 	i = vsnrprintf(dev->__si_namebuf, sizeof dev->__si_namebuf, 32, fmt, ap);
789 	if (i > (sizeof dev->__si_namebuf - 1)) {
790 		printf("WARNING: Device name truncated! (%s)\n",
791 		    dev->__si_namebuf);
792 	}
793 	va_end(ap);
794 
795 	devfs_create(dev);
796 	clean_unrhdrl(devfs_inos);
797 	dev_unlock();
798 	dev_depends(pdev, dev);
799 	return (dev);
800 }
801 
802 static void
803 destroy_devl(struct cdev *dev)
804 {
805 	struct cdevsw *csw;
806 
807 	mtx_assert(&devmtx, MA_OWNED);
808 	KASSERT(dev->si_flags & SI_NAMED,
809 	    ("WARNING: Driver mistake: destroy_dev on %d\n", minor(dev)));
810 
811 	devfs_destroy(dev);
812 
813 	/* Remove name marking */
814 	dev->si_flags &= ~SI_NAMED;
815 
816 	/* If we are a child, remove us from the parents list */
817 	if (dev->si_flags & SI_CHILD) {
818 		LIST_REMOVE(dev, si_siblings);
819 		dev->si_flags &= ~SI_CHILD;
820 	}
821 
822 	/* Kill our children */
823 	while (!LIST_EMPTY(&dev->si_children))
824 		destroy_devl(LIST_FIRST(&dev->si_children));
825 
826 	/* Remove from clone list */
827 	if (dev->si_flags & SI_CLONELIST) {
828 		LIST_REMOVE(dev, si_clone);
829 		dev->si_flags &= ~SI_CLONELIST;
830 	}
831 
832 	dev->si_refcount++;	/* Avoid race with dev_rel() */
833 	csw = dev->si_devsw;
834 	dev->si_devsw = NULL;	/* already NULL for SI_ALIAS */
835 	while (csw != NULL && csw->d_purge != NULL && dev->si_threadcount) {
836 		csw->d_purge(dev);
837 		msleep(csw, &devmtx, PRIBIO, "devprg", hz/10);
838 		if (dev->si_threadcount)
839 			printf("Still %lu threads in %s\n",
840 			    dev->si_threadcount, devtoname(dev));
841 	}
842 	while (dev->si_threadcount != 0) {
843 		/* Use unique dummy wait ident */
844 		msleep(&csw, &devmtx, PRIBIO, "devdrn", hz / 10);
845 	}
846 
847 	dev->si_drv1 = 0;
848 	dev->si_drv2 = 0;
849 	bzero(&dev->__si_u, sizeof(dev->__si_u));
850 
851 	if (!(dev->si_flags & SI_ALIAS)) {
852 		/* Remove from cdevsw list */
853 		LIST_REMOVE(dev, si_list);
854 
855 		/* If cdevsw has no more struct cdev *'s, clean it */
856 		if (LIST_EMPTY(&csw->d_devs)) {
857 			fini_cdevsw(csw);
858 			wakeup(&csw->d_devs);
859 		}
860 	}
861 	dev->si_flags &= ~SI_ALIAS;
862 	dev->si_refcount--;	/* Avoid race with dev_rel() */
863 
864 	if (dev->si_refcount > 0) {
865 		LIST_INSERT_HEAD(&dead_cdevsw.d_devs, dev, si_list);
866 	} else {
867 		dev_free_devlocked(dev);
868 	}
869 }
870 
871 void
872 destroy_dev(struct cdev *dev)
873 {
874 
875 	dev_lock();
876 	destroy_devl(dev);
877 	dev_unlock_and_free();
878 }
879 
880 const char *
881 devtoname(struct cdev *dev)
882 {
883 	char *p;
884 	struct cdevsw *csw;
885 	int mynor;
886 
887 	if (dev->si_name[0] == '#' || dev->si_name[0] == '\0') {
888 		p = dev->si_name;
889 		csw = dev_refthread(dev);
890 		if (csw != NULL) {
891 			sprintf(p, "(%s)", csw->d_name);
892 			dev_relthread(dev);
893 		}
894 		p += strlen(p);
895 		mynor = minor(dev);
896 		if (mynor < 0 || mynor > 255)
897 			sprintf(p, "/%#x", (u_int)mynor);
898 		else
899 			sprintf(p, "/%d", mynor);
900 	}
901 	return (dev->si_name);
902 }
903 
904 int
905 dev_stdclone(char *name, char **namep, const char *stem, int *unit)
906 {
907 	int u, i;
908 
909 	i = strlen(stem);
910 	if (bcmp(stem, name, i) != 0)
911 		return (0);
912 	if (!isdigit(name[i]))
913 		return (0);
914 	u = 0;
915 	if (name[i] == '0' && isdigit(name[i+1]))
916 		return (0);
917 	while (isdigit(name[i])) {
918 		u *= 10;
919 		u += name[i++] - '0';
920 	}
921 	if (u > 0xffffff)
922 		return (0);
923 	*unit = u;
924 	if (namep)
925 		*namep = &name[i];
926 	if (name[i])
927 		return (2);
928 	return (1);
929 }
930 
931 /*
932  * Helper functions for cloning device drivers.
933  *
934  * The objective here is to make it unnecessary for the device drivers to
935  * use rman or similar to manage their unit number space.  Due to the way
936  * we do "on-demand" devices, using rman or other "private" methods
937  * will be very tricky to lock down properly once we lock down this file.
938  *
939  * Instead we give the drivers these routines which puts the struct cdev *'s
940  * that are to be managed on their own list, and gives the driver the ability
941  * to ask for the first free unit number or a given specified unit number.
942  *
943  * In addition these routines support paired devices (pty, nmdm and similar)
944  * by respecting a number of "flag" bits in the minor number.
945  *
946  */
947 
948 struct clonedevs {
949 	LIST_HEAD(,cdev)	head;
950 };
951 
952 void
953 clone_setup(struct clonedevs **cdp)
954 {
955 
956 	*cdp = malloc(sizeof **cdp, M_DEVBUF, M_WAITOK | M_ZERO);
957 	LIST_INIT(&(*cdp)->head);
958 }
959 
960 int
961 clone_create(struct clonedevs **cdp, struct cdevsw *csw, int *up, struct cdev **dp, int extra)
962 {
963 	struct clonedevs *cd;
964 	struct cdev *dev, *ndev, *dl, *de;
965 	int unit, low, u;
966 
967 	KASSERT(*cdp != NULL,
968 	    ("clone_setup() not called in driver \"%s\"", csw->d_name));
969 	KASSERT(!(extra & CLONE_UNITMASK),
970 	    ("Illegal extra bits (0x%x) in clone_create", extra));
971 	KASSERT(*up <= CLONE_UNITMASK,
972 	    ("Too high unit (0x%x) in clone_create", *up));
973 
974 
975 	/*
976 	 * Search the list for a lot of things in one go:
977 	 *   A preexisting match is returned immediately.
978 	 *   The lowest free unit number if we are passed -1, and the place
979 	 *	 in the list where we should insert that new element.
980 	 *   The place to insert a specified unit number, if applicable
981 	 *       the end of the list.
982 	 */
983 	unit = *up;
984 	ndev = devfs_alloc();
985 	dev_lock();
986 	prep_cdevsw(csw);
987 	low = extra;
988 	de = dl = NULL;
989 	cd = *cdp;
990 	LIST_FOREACH(dev, &cd->head, si_clone) {
991 		KASSERT(dev->si_flags & SI_CLONELIST,
992 		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
993 		u = dev2unit(dev);
994 		if (u == (unit | extra)) {
995 			*dp = dev;
996 			dev_unlock();
997 			devfs_free(ndev);
998 			return (0);
999 		}
1000 		if (unit == -1 && u == low) {
1001 			low++;
1002 			de = dev;
1003 			continue;
1004 		} else if (u < (unit | extra)) {
1005 			de = dev;
1006 			continue;
1007 		} else if (u > (unit | extra)) {
1008 			dl = dev;
1009 			break;
1010 		}
1011 	}
1012 	if (unit == -1)
1013 		unit = low & CLONE_UNITMASK;
1014 	dev = newdev(csw, unit2minor(unit | extra), ndev);
1015 	if (dev->si_flags & SI_CLONELIST) {
1016 		printf("dev %p (%s) is on clonelist\n", dev, dev->si_name);
1017 		printf("unit=%d, low=%d, extra=0x%x\n", unit, low, extra);
1018 		LIST_FOREACH(dev, &cd->head, si_clone) {
1019 			printf("\t%p %s\n", dev, dev->si_name);
1020 		}
1021 		panic("foo");
1022 	}
1023 	KASSERT(!(dev->si_flags & SI_CLONELIST),
1024 	    ("Dev %p(%s) should not be on clonelist", dev, dev->si_name));
1025 	if (dl != NULL)
1026 		LIST_INSERT_BEFORE(dl, dev, si_clone);
1027 	else if (de != NULL)
1028 		LIST_INSERT_AFTER(de, dev, si_clone);
1029 	else
1030 		LIST_INSERT_HEAD(&cd->head, dev, si_clone);
1031 	dev->si_flags |= SI_CLONELIST;
1032 	*up = unit;
1033 	dev_unlock_and_free();
1034 	return (1);
1035 }
1036 
1037 /*
1038  * Kill everything still on the list.  The driver should already have
1039  * disposed of any softc hung of the struct cdev *'s at this time.
1040  */
1041 void
1042 clone_cleanup(struct clonedevs **cdp)
1043 {
1044 	struct cdev *dev;
1045 	struct cdev_priv *cp;
1046 	struct clonedevs *cd;
1047 
1048 	cd = *cdp;
1049 	if (cd == NULL)
1050 		return;
1051 	dev_lock();
1052 	while (!LIST_EMPTY(&cd->head)) {
1053 		dev = LIST_FIRST(&cd->head);
1054 		LIST_REMOVE(dev, si_clone);
1055 		KASSERT(dev->si_flags & SI_CLONELIST,
1056 		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
1057 		dev->si_flags &= ~SI_CLONELIST;
1058 		cp = dev->si_priv;
1059 		if (!(cp->cdp_flags & CDP_SCHED_DTR)) {
1060 			cp->cdp_flags |= CDP_SCHED_DTR;
1061 			KASSERT(dev->si_flags & SI_NAMED,
1062 				("Driver has goofed in cloning underways udev %x", dev->si_drv0));
1063 			destroy_devl(dev);
1064 		}
1065 	}
1066 	dev_unlock_and_free();
1067 	free(cd, M_DEVBUF);
1068 	*cdp = NULL;
1069 }
1070 
1071 static TAILQ_HEAD(, cdev_priv) dev_ddtr =
1072 	TAILQ_HEAD_INITIALIZER(dev_ddtr);
1073 static struct task dev_dtr_task;
1074 
1075 static void
1076 destroy_dev_tq(void *ctx, int pending)
1077 {
1078 	struct cdev_priv *cp;
1079 	struct cdev *dev;
1080 	void (*cb)(void *);
1081 	void *cb_arg;
1082 
1083 	dev_lock();
1084 	while (!TAILQ_EMPTY(&dev_ddtr)) {
1085 		cp = TAILQ_FIRST(&dev_ddtr);
1086 		dev = &cp->cdp_c;
1087 		KASSERT(cp->cdp_flags & CDP_SCHED_DTR,
1088 		    ("cdev %p in dev_destroy_tq without CDP_SCHED_DTR", cp));
1089 		TAILQ_REMOVE(&dev_ddtr, cp, cdp_dtr_list);
1090 		cb = cp->cdp_dtr_cb;
1091 		cb_arg = cp->cdp_dtr_cb_arg;
1092 		destroy_devl(dev);
1093 		dev_unlock_and_free();
1094 		dev_rel(dev);
1095 		if (cb != NULL)
1096 			cb(cb_arg);
1097 		dev_lock();
1098 	}
1099 	dev_unlock();
1100 }
1101 
1102 /*
1103  * devmtx shall be locked on entry. devmtx will be unlocked after
1104  * function return.
1105  */
1106 static int
1107 destroy_dev_sched_cbl(struct cdev *dev, void (*cb)(void *), void *arg)
1108 {
1109 	struct cdev_priv *cp;
1110 
1111 	mtx_assert(&devmtx, MA_OWNED);
1112 	cp = dev->si_priv;
1113 	if (cp->cdp_flags & CDP_SCHED_DTR) {
1114 		dev_unlock();
1115 		return (0);
1116 	}
1117 	dev_refl(dev);
1118 	cp->cdp_flags |= CDP_SCHED_DTR;
1119 	cp->cdp_dtr_cb = cb;
1120 	cp->cdp_dtr_cb_arg = arg;
1121 	TAILQ_INSERT_TAIL(&dev_ddtr, cp, cdp_dtr_list);
1122 	dev_unlock();
1123 	taskqueue_enqueue(taskqueue_swi_giant, &dev_dtr_task);
1124 	return (1);
1125 }
1126 
1127 int
1128 destroy_dev_sched_cb(struct cdev *dev, void (*cb)(void *), void *arg)
1129 {
1130 	dev_lock();
1131 	return (destroy_dev_sched_cbl(dev, cb, arg));
1132 }
1133 
1134 int
1135 destroy_dev_sched(struct cdev *dev)
1136 {
1137 	return (destroy_dev_sched_cb(dev, NULL, NULL));
1138 }
1139 
1140 void
1141 destroy_dev_drain(struct cdevsw *csw)
1142 {
1143 
1144 	dev_lock();
1145 	while (!LIST_EMPTY(&csw->d_devs)) {
1146 		msleep(&csw->d_devs, &devmtx, PRIBIO, "devscd", hz/10);
1147 	}
1148 	dev_unlock();
1149 }
1150 
1151 void
1152 drain_dev_clone_events(void)
1153 {
1154 
1155 	sx_xlock(&clone_drain_lock);
1156 	sx_xunlock(&clone_drain_lock);
1157 }
1158 
1159 static void
1160 devdtr_init(void *dummy __unused)
1161 {
1162 
1163 	TASK_INIT(&dev_dtr_task, 0, destroy_dev_tq, NULL);
1164 }
1165 
1166 SYSINIT(devdtr, SI_SUB_DEVFS, SI_ORDER_SECOND, devdtr_init, NULL);
1167