xref: /freebsd/sys/kern/kern_conf.c (revision 2be1a816b9ff69588e55be0a84cbe2a31efc0f2f)
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 = dsw->d_gianttrick->d_ioctl(dev, cmd, data, fflag, td);
406 	mtx_unlock(&Giant);
407 	dev_relthread(dev);
408 	return (retval);
409 }
410 
411 static int
412 giant_read(struct cdev *dev, struct uio *uio, int ioflag)
413 {
414 	struct cdevsw *dsw;
415 	int retval;
416 
417 	dsw = dev_refthread(dev);
418 	if (dsw == NULL)
419 		return (ENXIO);
420 	mtx_lock(&Giant);
421 	retval = dsw->d_gianttrick->d_read(dev, uio, ioflag);
422 	mtx_unlock(&Giant);
423 	dev_relthread(dev);
424 	return (retval);
425 }
426 
427 static int
428 giant_write(struct cdev *dev, struct uio *uio, int ioflag)
429 {
430 	struct cdevsw *dsw;
431 	int retval;
432 
433 	dsw = dev_refthread(dev);
434 	if (dsw == NULL)
435 		return (ENXIO);
436 	mtx_lock(&Giant);
437 	retval = dsw->d_gianttrick->d_write(dev, uio, ioflag);
438 	mtx_unlock(&Giant);
439 	dev_relthread(dev);
440 	return (retval);
441 }
442 
443 static int
444 giant_poll(struct cdev *dev, int events, struct thread *td)
445 {
446 	struct cdevsw *dsw;
447 	int retval;
448 
449 	dsw = dev_refthread(dev);
450 	if (dsw == NULL)
451 		return (ENXIO);
452 	mtx_lock(&Giant);
453 	retval = dsw->d_gianttrick->d_poll(dev, events, td);
454 	mtx_unlock(&Giant);
455 	dev_relthread(dev);
456 	return (retval);
457 }
458 
459 static int
460 giant_kqfilter(struct cdev *dev, struct knote *kn)
461 {
462 	struct cdevsw *dsw;
463 	int retval;
464 
465 	dsw = dev_refthread(dev);
466 	if (dsw == NULL)
467 		return (ENXIO);
468 	mtx_lock(&Giant);
469 	retval = dsw->d_gianttrick->d_kqfilter(dev, kn);
470 	mtx_unlock(&Giant);
471 	dev_relthread(dev);
472 	return (retval);
473 }
474 
475 static int
476 giant_mmap(struct cdev *dev, vm_offset_t offset, vm_paddr_t *paddr, int nprot)
477 {
478 	struct cdevsw *dsw;
479 	int retval;
480 
481 	dsw = dev_refthread(dev);
482 	if (dsw == NULL)
483 		return (ENXIO);
484 	mtx_lock(&Giant);
485 	retval = dsw->d_gianttrick->d_mmap(dev, offset, paddr, nprot);
486 	mtx_unlock(&Giant);
487 	dev_relthread(dev);
488 	return (retval);
489 }
490 
491 
492 /*
493  * struct cdev * and u_dev_t primitives
494  */
495 
496 int
497 minor(struct cdev *x)
498 {
499 	if (x == NULL)
500 		return NODEV;
501 	return(x->si_drv0 & MAXMINOR);
502 }
503 
504 int
505 dev2unit(struct cdev *x)
506 {
507 
508 	if (x == NULL)
509 		return NODEV;
510 	return (minor2unit(minor(x)));
511 }
512 
513 u_int
514 minor2unit(u_int _minor)
515 {
516 
517 	KASSERT((_minor & ~MAXMINOR) == 0, ("Illegal minor %x", _minor));
518 	return ((_minor & 0xff) | ((_minor >> 8) & 0xffff00));
519 }
520 
521 int
522 unit2minor(int unit)
523 {
524 
525 	KASSERT(unit <= 0xffffff, ("Invalid unit (%d) in unit2minor", unit));
526 	return ((unit & 0xff) | ((unit << 8) & ~0xffff));
527 }
528 
529 static struct cdev *
530 newdev(struct cdevsw *csw, int y, struct cdev *si)
531 {
532 	struct cdev *si2;
533 	dev_t	udev;
534 
535 	mtx_assert(&devmtx, MA_OWNED);
536 	udev = y;
537 	LIST_FOREACH(si2, &csw->d_devs, si_list) {
538 		if (si2->si_drv0 == udev) {
539 			dev_free_devlocked(si);
540 			return (si2);
541 		}
542 	}
543 	si->si_drv0 = udev;
544 	si->si_devsw = csw;
545 	LIST_INSERT_HEAD(&csw->d_devs, si, si_list);
546 	return (si);
547 }
548 
549 int
550 uminor(dev_t dev)
551 {
552 	return (dev & MAXMINOR);
553 }
554 
555 int
556 umajor(dev_t dev)
557 {
558 	return ((dev & ~MAXMINOR) >> 8);
559 }
560 
561 static void
562 fini_cdevsw(struct cdevsw *devsw)
563 {
564 	struct cdevsw *gt;
565 
566 	if (devsw->d_gianttrick != NULL) {
567 		gt = devsw->d_gianttrick;
568 		memcpy(devsw, gt, sizeof *devsw);
569 		cdevsw_free_devlocked(gt);
570 		devsw->d_gianttrick = NULL;
571 	}
572 	devsw->d_flags &= ~D_INIT;
573 }
574 
575 static void
576 prep_cdevsw(struct cdevsw *devsw)
577 {
578 	struct cdevsw *dsw2;
579 
580 	mtx_assert(&devmtx, MA_OWNED);
581 	if (devsw->d_flags & D_INIT)
582 		return;
583 	if (devsw->d_flags & D_NEEDGIANT) {
584 		dev_unlock();
585 		dsw2 = malloc(sizeof *dsw2, M_DEVT, M_WAITOK);
586 		dev_lock();
587 	} else
588 		dsw2 = NULL;
589 	if (devsw->d_flags & D_INIT) {
590 		if (dsw2 != NULL)
591 			cdevsw_free_devlocked(dsw2);
592 		return;
593 	}
594 
595 	if (devsw->d_version != D_VERSION_01) {
596 		printf(
597 		    "WARNING: Device driver \"%s\" has wrong version %s\n",
598 		    devsw->d_name == NULL ? "???" : devsw->d_name,
599 		    "and is disabled.  Recompile KLD module.");
600 		devsw->d_open = dead_open;
601 		devsw->d_close = dead_close;
602 		devsw->d_read = dead_read;
603 		devsw->d_write = dead_write;
604 		devsw->d_ioctl = dead_ioctl;
605 		devsw->d_poll = dead_poll;
606 		devsw->d_mmap = dead_mmap;
607 		devsw->d_strategy = dead_strategy;
608 		devsw->d_dump = dead_dump;
609 		devsw->d_kqfilter = dead_kqfilter;
610 	}
611 
612 	if (devsw->d_flags & D_TTY) {
613 		if (devsw->d_ioctl == NULL)	devsw->d_ioctl = ttyioctl;
614 		if (devsw->d_read == NULL)	devsw->d_read = ttyread;
615 		if (devsw->d_write == NULL)	devsw->d_write = ttywrite;
616 		if (devsw->d_kqfilter == NULL)	devsw->d_kqfilter = ttykqfilter;
617 		if (devsw->d_poll == NULL)	devsw->d_poll = ttypoll;
618 	}
619 
620 	if (devsw->d_flags & D_NEEDGIANT) {
621 		if (devsw->d_gianttrick == NULL) {
622 			memcpy(dsw2, devsw, sizeof *dsw2);
623 			devsw->d_gianttrick = dsw2;
624 			dsw2 = NULL;
625 		}
626 	}
627 
628 #define FIXUP(member, noop, giant) 				\
629 	do {							\
630 		if (devsw->member == NULL) {			\
631 			devsw->member = noop;			\
632 		} else if (devsw->d_flags & D_NEEDGIANT)	\
633 			devsw->member = giant;			\
634 		}						\
635 	while (0)
636 
637 	FIXUP(d_open,		null_open,	giant_open);
638 	FIXUP(d_fdopen,		NULL,		giant_fdopen);
639 	FIXUP(d_close,		null_close,	giant_close);
640 	FIXUP(d_read,		no_read,	giant_read);
641 	FIXUP(d_write,		no_write,	giant_write);
642 	FIXUP(d_ioctl,		no_ioctl,	giant_ioctl);
643 	FIXUP(d_poll,		no_poll,	giant_poll);
644 	FIXUP(d_mmap,		no_mmap,	giant_mmap);
645 	FIXUP(d_strategy,	no_strategy,	giant_strategy);
646 	FIXUP(d_kqfilter,	no_kqfilter,	giant_kqfilter);
647 
648 	if (devsw->d_dump == NULL)	devsw->d_dump = no_dump;
649 
650 	LIST_INIT(&devsw->d_devs);
651 
652 	devsw->d_flags |= D_INIT;
653 
654 	if (dsw2 != NULL)
655 		cdevsw_free_devlocked(dsw2);
656 }
657 
658 struct cdev *
659 make_dev_credv(int flags, struct cdevsw *devsw, int minornr,
660     struct ucred *cr, uid_t uid,
661     gid_t gid, int mode, const char *fmt, va_list ap)
662 {
663 	struct cdev *dev;
664 	int i;
665 
666 	KASSERT((minornr & ~MAXMINOR) == 0,
667 	    ("Invalid minor (0x%x) in make_dev", minornr));
668 
669 	dev = devfs_alloc();
670 	dev_lock();
671 	prep_cdevsw(devsw);
672 	dev = newdev(devsw, minornr, dev);
673 	if (flags & MAKEDEV_REF)
674 		dev_refl(dev);
675 	if (dev->si_flags & SI_CHEAPCLONE &&
676 	    dev->si_flags & SI_NAMED) {
677 		/*
678 		 * This is allowed as it removes races and generally
679 		 * simplifies cloning devices.
680 		 * XXX: still ??
681 		 */
682 		dev_unlock_and_free();
683 		return (dev);
684 	}
685 	KASSERT(!(dev->si_flags & SI_NAMED),
686 	    ("make_dev() by driver %s on pre-existing device (min=%x, name=%s)",
687 	    devsw->d_name, minor(dev), devtoname(dev)));
688 
689 	i = vsnrprintf(dev->__si_namebuf, sizeof dev->__si_namebuf, 32, fmt, ap);
690 	if (i > (sizeof dev->__si_namebuf - 1)) {
691 		printf("WARNING: Device name truncated! (%s)\n",
692 		    dev->__si_namebuf);
693 	}
694 
695 	dev->si_flags |= SI_NAMED;
696 #ifdef MAC
697 	if (cr != NULL)
698 		dev->si_cred = crhold(cr);
699 	else
700 #endif
701 		dev->si_cred = NULL;
702 	dev->si_uid = uid;
703 	dev->si_gid = gid;
704 	dev->si_mode = mode;
705 
706 	devfs_create(dev);
707 	clean_unrhdrl(devfs_inos);
708 	dev_unlock_and_free();
709 	return (dev);
710 }
711 
712 struct cdev *
713 make_dev(struct cdevsw *devsw, int minornr, uid_t uid, gid_t gid, int mode,
714     const char *fmt, ...)
715 {
716 	struct cdev *dev;
717 	va_list ap;
718 
719 	va_start(ap, fmt);
720 	dev = make_dev_credv(0, devsw, minornr, NULL, uid, gid, mode, fmt, ap);
721 	va_end(ap);
722 	return (dev);
723 }
724 
725 struct cdev *
726 make_dev_cred(struct cdevsw *devsw, int minornr, struct ucred *cr, uid_t uid,
727     gid_t gid, int mode, const char *fmt, ...)
728 {
729 	struct cdev *dev;
730 	va_list ap;
731 
732 	va_start(ap, fmt);
733 	dev = make_dev_credv(0, devsw, minornr, cr, uid, gid, mode, fmt, ap);
734 	va_end(ap);
735 
736 	return (dev);
737 }
738 
739 struct cdev *
740 make_dev_credf(int flags, struct cdevsw *devsw, int minornr,
741     struct ucred *cr, uid_t uid,
742     gid_t gid, int mode, const char *fmt, ...)
743 {
744 	struct cdev *dev;
745 	va_list ap;
746 
747 	va_start(ap, fmt);
748 	dev = make_dev_credv(flags, devsw, minornr, cr, uid, gid, mode,
749 	    fmt, ap);
750 	va_end(ap);
751 
752 	return (dev);
753 }
754 
755 static void
756 dev_dependsl(struct cdev *pdev, struct cdev *cdev)
757 {
758 
759 	cdev->si_parent = pdev;
760 	cdev->si_flags |= SI_CHILD;
761 	LIST_INSERT_HEAD(&pdev->si_children, cdev, si_siblings);
762 }
763 
764 
765 void
766 dev_depends(struct cdev *pdev, struct cdev *cdev)
767 {
768 
769 	dev_lock();
770 	dev_dependsl(pdev, cdev);
771 	dev_unlock();
772 }
773 
774 struct cdev *
775 make_dev_alias(struct cdev *pdev, const char *fmt, ...)
776 {
777 	struct cdev *dev;
778 	va_list ap;
779 	int i;
780 
781 	dev = devfs_alloc();
782 	dev_lock();
783 	dev->si_flags |= SI_ALIAS;
784 	dev->si_flags |= SI_NAMED;
785 	va_start(ap, fmt);
786 	i = vsnrprintf(dev->__si_namebuf, sizeof dev->__si_namebuf, 32, fmt, ap);
787 	if (i > (sizeof dev->__si_namebuf - 1)) {
788 		printf("WARNING: Device name truncated! (%s)\n",
789 		    dev->__si_namebuf);
790 	}
791 	va_end(ap);
792 
793 	devfs_create(dev);
794 	clean_unrhdrl(devfs_inos);
795 	dev_unlock();
796 	dev_depends(pdev, dev);
797 	return (dev);
798 }
799 
800 static void
801 destroy_devl(struct cdev *dev)
802 {
803 	struct cdevsw *csw;
804 
805 	mtx_assert(&devmtx, MA_OWNED);
806 	KASSERT(dev->si_flags & SI_NAMED,
807 	    ("WARNING: Driver mistake: destroy_dev on %d\n", minor(dev)));
808 
809 	devfs_destroy(dev);
810 
811 	/* Remove name marking */
812 	dev->si_flags &= ~SI_NAMED;
813 
814 	/* If we are a child, remove us from the parents list */
815 	if (dev->si_flags & SI_CHILD) {
816 		LIST_REMOVE(dev, si_siblings);
817 		dev->si_flags &= ~SI_CHILD;
818 	}
819 
820 	/* Kill our children */
821 	while (!LIST_EMPTY(&dev->si_children))
822 		destroy_devl(LIST_FIRST(&dev->si_children));
823 
824 	/* Remove from clone list */
825 	if (dev->si_flags & SI_CLONELIST) {
826 		LIST_REMOVE(dev, si_clone);
827 		dev->si_flags &= ~SI_CLONELIST;
828 	}
829 
830 	dev->si_refcount++;	/* Avoid race with dev_rel() */
831 	csw = dev->si_devsw;
832 	dev->si_devsw = NULL;	/* already NULL for SI_ALIAS */
833 	while (csw != NULL && csw->d_purge != NULL && dev->si_threadcount) {
834 		csw->d_purge(dev);
835 		msleep(csw, &devmtx, PRIBIO, "devprg", hz/10);
836 		if (dev->si_threadcount)
837 			printf("Still %lu threads in %s\n",
838 			    dev->si_threadcount, devtoname(dev));
839 	}
840 	while (dev->si_threadcount != 0) {
841 		/* Use unique dummy wait ident */
842 		msleep(&csw, &devmtx, PRIBIO, "devdrn", hz / 10);
843 	}
844 
845 	dev->si_drv1 = 0;
846 	dev->si_drv2 = 0;
847 	bzero(&dev->__si_u, sizeof(dev->__si_u));
848 
849 	if (!(dev->si_flags & SI_ALIAS)) {
850 		/* Remove from cdevsw list */
851 		LIST_REMOVE(dev, si_list);
852 
853 		/* If cdevsw has no more struct cdev *'s, clean it */
854 		if (LIST_EMPTY(&csw->d_devs)) {
855 			fini_cdevsw(csw);
856 			wakeup(&csw->d_devs);
857 		}
858 	}
859 	dev->si_flags &= ~SI_ALIAS;
860 	dev->si_refcount--;	/* Avoid race with dev_rel() */
861 
862 	if (dev->si_refcount > 0) {
863 		LIST_INSERT_HEAD(&dead_cdevsw.d_devs, dev, si_list);
864 	} else {
865 		dev_free_devlocked(dev);
866 	}
867 }
868 
869 void
870 destroy_dev(struct cdev *dev)
871 {
872 
873 	dev_lock();
874 	destroy_devl(dev);
875 	dev_unlock_and_free();
876 }
877 
878 const char *
879 devtoname(struct cdev *dev)
880 {
881 	char *p;
882 	struct cdevsw *csw;
883 	int mynor;
884 
885 	if (dev->si_name[0] == '#' || dev->si_name[0] == '\0') {
886 		p = dev->si_name;
887 		csw = dev_refthread(dev);
888 		if (csw != NULL) {
889 			sprintf(p, "(%s)", csw->d_name);
890 			dev_relthread(dev);
891 		}
892 		p += strlen(p);
893 		mynor = minor(dev);
894 		if (mynor < 0 || mynor > 255)
895 			sprintf(p, "/%#x", (u_int)mynor);
896 		else
897 			sprintf(p, "/%d", mynor);
898 	}
899 	return (dev->si_name);
900 }
901 
902 int
903 dev_stdclone(char *name, char **namep, const char *stem, int *unit)
904 {
905 	int u, i;
906 
907 	i = strlen(stem);
908 	if (bcmp(stem, name, i) != 0)
909 		return (0);
910 	if (!isdigit(name[i]))
911 		return (0);
912 	u = 0;
913 	if (name[i] == '0' && isdigit(name[i+1]))
914 		return (0);
915 	while (isdigit(name[i])) {
916 		u *= 10;
917 		u += name[i++] - '0';
918 	}
919 	if (u > 0xffffff)
920 		return (0);
921 	*unit = u;
922 	if (namep)
923 		*namep = &name[i];
924 	if (name[i])
925 		return (2);
926 	return (1);
927 }
928 
929 /*
930  * Helper functions for cloning device drivers.
931  *
932  * The objective here is to make it unnecessary for the device drivers to
933  * use rman or similar to manage their unit number space.  Due to the way
934  * we do "on-demand" devices, using rman or other "private" methods
935  * will be very tricky to lock down properly once we lock down this file.
936  *
937  * Instead we give the drivers these routines which puts the struct cdev *'s
938  * that are to be managed on their own list, and gives the driver the ability
939  * to ask for the first free unit number or a given specified unit number.
940  *
941  * In addition these routines support paired devices (pty, nmdm and similar)
942  * by respecting a number of "flag" bits in the minor number.
943  *
944  */
945 
946 struct clonedevs {
947 	LIST_HEAD(,cdev)	head;
948 };
949 
950 void
951 clone_setup(struct clonedevs **cdp)
952 {
953 
954 	*cdp = malloc(sizeof **cdp, M_DEVBUF, M_WAITOK | M_ZERO);
955 	LIST_INIT(&(*cdp)->head);
956 }
957 
958 int
959 clone_create(struct clonedevs **cdp, struct cdevsw *csw, int *up, struct cdev **dp, int extra)
960 {
961 	struct clonedevs *cd;
962 	struct cdev *dev, *ndev, *dl, *de;
963 	int unit, low, u;
964 
965 	KASSERT(*cdp != NULL,
966 	    ("clone_setup() not called in driver \"%s\"", csw->d_name));
967 	KASSERT(!(extra & CLONE_UNITMASK),
968 	    ("Illegal extra bits (0x%x) in clone_create", extra));
969 	KASSERT(*up <= CLONE_UNITMASK,
970 	    ("Too high unit (0x%x) in clone_create", *up));
971 
972 
973 	/*
974 	 * Search the list for a lot of things in one go:
975 	 *   A preexisting match is returned immediately.
976 	 *   The lowest free unit number if we are passed -1, and the place
977 	 *	 in the list where we should insert that new element.
978 	 *   The place to insert a specified unit number, if applicable
979 	 *       the end of the list.
980 	 */
981 	unit = *up;
982 	ndev = devfs_alloc();
983 	dev_lock();
984 	prep_cdevsw(csw);
985 	low = extra;
986 	de = dl = NULL;
987 	cd = *cdp;
988 	LIST_FOREACH(dev, &cd->head, si_clone) {
989 		KASSERT(dev->si_flags & SI_CLONELIST,
990 		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
991 		u = dev2unit(dev);
992 		if (u == (unit | extra)) {
993 			*dp = dev;
994 			dev_unlock();
995 			devfs_free(ndev);
996 			return (0);
997 		}
998 		if (unit == -1 && u == low) {
999 			low++;
1000 			de = dev;
1001 			continue;
1002 		} else if (u < (unit | extra)) {
1003 			de = dev;
1004 			continue;
1005 		} else if (u > (unit | extra)) {
1006 			dl = dev;
1007 			break;
1008 		}
1009 	}
1010 	if (unit == -1)
1011 		unit = low & CLONE_UNITMASK;
1012 	dev = newdev(csw, unit2minor(unit | extra), ndev);
1013 	if (dev->si_flags & SI_CLONELIST) {
1014 		printf("dev %p (%s) is on clonelist\n", dev, dev->si_name);
1015 		printf("unit=%d, low=%d, extra=0x%x\n", unit, low, extra);
1016 		LIST_FOREACH(dev, &cd->head, si_clone) {
1017 			printf("\t%p %s\n", dev, dev->si_name);
1018 		}
1019 		panic("foo");
1020 	}
1021 	KASSERT(!(dev->si_flags & SI_CLONELIST),
1022 	    ("Dev %p(%s) should not be on clonelist", dev, dev->si_name));
1023 	if (dl != NULL)
1024 		LIST_INSERT_BEFORE(dl, dev, si_clone);
1025 	else if (de != NULL)
1026 		LIST_INSERT_AFTER(de, dev, si_clone);
1027 	else
1028 		LIST_INSERT_HEAD(&cd->head, dev, si_clone);
1029 	dev->si_flags |= SI_CLONELIST;
1030 	*up = unit;
1031 	dev_unlock_and_free();
1032 	return (1);
1033 }
1034 
1035 /*
1036  * Kill everything still on the list.  The driver should already have
1037  * disposed of any softc hung of the struct cdev *'s at this time.
1038  */
1039 void
1040 clone_cleanup(struct clonedevs **cdp)
1041 {
1042 	struct cdev *dev;
1043 	struct cdev_priv *cp;
1044 	struct clonedevs *cd;
1045 
1046 	cd = *cdp;
1047 	if (cd == NULL)
1048 		return;
1049 	dev_lock();
1050 	while (!LIST_EMPTY(&cd->head)) {
1051 		dev = LIST_FIRST(&cd->head);
1052 		LIST_REMOVE(dev, si_clone);
1053 		KASSERT(dev->si_flags & SI_CLONELIST,
1054 		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
1055 		dev->si_flags &= ~SI_CLONELIST;
1056 		cp = dev->si_priv;
1057 		if (!(cp->cdp_flags & CDP_SCHED_DTR)) {
1058 			cp->cdp_flags |= CDP_SCHED_DTR;
1059 			KASSERT(dev->si_flags & SI_NAMED,
1060 				("Driver has goofed in cloning underways udev %x", dev->si_drv0));
1061 			destroy_devl(dev);
1062 		}
1063 	}
1064 	dev_unlock_and_free();
1065 	free(cd, M_DEVBUF);
1066 	*cdp = NULL;
1067 }
1068 
1069 static TAILQ_HEAD(, cdev_priv) dev_ddtr =
1070 	TAILQ_HEAD_INITIALIZER(dev_ddtr);
1071 static struct task dev_dtr_task;
1072 
1073 static void
1074 destroy_dev_tq(void *ctx, int pending)
1075 {
1076 	struct cdev_priv *cp;
1077 	struct cdev *dev;
1078 	void (*cb)(void *);
1079 	void *cb_arg;
1080 
1081 	dev_lock();
1082 	while (!TAILQ_EMPTY(&dev_ddtr)) {
1083 		cp = TAILQ_FIRST(&dev_ddtr);
1084 		dev = &cp->cdp_c;
1085 		KASSERT(cp->cdp_flags & CDP_SCHED_DTR,
1086 		    ("cdev %p in dev_destroy_tq without CDP_SCHED_DTR", cp));
1087 		TAILQ_REMOVE(&dev_ddtr, cp, cdp_dtr_list);
1088 		cb = cp->cdp_dtr_cb;
1089 		cb_arg = cp->cdp_dtr_cb_arg;
1090 		destroy_devl(dev);
1091 		dev_unlock_and_free();
1092 		dev_rel(dev);
1093 		if (cb != NULL)
1094 			cb(cb_arg);
1095 		dev_lock();
1096 	}
1097 	dev_unlock();
1098 }
1099 
1100 /*
1101  * devmtx shall be locked on entry. devmtx will be unlocked after
1102  * function return.
1103  */
1104 static int
1105 destroy_dev_sched_cbl(struct cdev *dev, void (*cb)(void *), void *arg)
1106 {
1107 	struct cdev_priv *cp;
1108 
1109 	mtx_assert(&devmtx, MA_OWNED);
1110 	cp = dev->si_priv;
1111 	if (cp->cdp_flags & CDP_SCHED_DTR) {
1112 		dev_unlock();
1113 		return (0);
1114 	}
1115 	dev_refl(dev);
1116 	cp->cdp_flags |= CDP_SCHED_DTR;
1117 	cp->cdp_dtr_cb = cb;
1118 	cp->cdp_dtr_cb_arg = arg;
1119 	TAILQ_INSERT_TAIL(&dev_ddtr, cp, cdp_dtr_list);
1120 	dev_unlock();
1121 	taskqueue_enqueue(taskqueue_swi_giant, &dev_dtr_task);
1122 	return (1);
1123 }
1124 
1125 int
1126 destroy_dev_sched_cb(struct cdev *dev, void (*cb)(void *), void *arg)
1127 {
1128 	dev_lock();
1129 	return (destroy_dev_sched_cbl(dev, cb, arg));
1130 }
1131 
1132 int
1133 destroy_dev_sched(struct cdev *dev)
1134 {
1135 	return (destroy_dev_sched_cb(dev, NULL, NULL));
1136 }
1137 
1138 void
1139 destroy_dev_drain(struct cdevsw *csw)
1140 {
1141 
1142 	dev_lock();
1143 	while (!LIST_EMPTY(&csw->d_devs)) {
1144 		msleep(&csw->d_devs, &devmtx, PRIBIO, "devscd", hz/10);
1145 	}
1146 	dev_unlock();
1147 }
1148 
1149 void
1150 drain_dev_clone_events(void)
1151 {
1152 
1153 	sx_xlock(&clone_drain_lock);
1154 	sx_xunlock(&clone_drain_lock);
1155 }
1156 
1157 static void
1158 devdtr_init(void *dummy __unused)
1159 {
1160 
1161 	TASK_INIT(&dev_dtr_task, 0, destroy_dev_tq, NULL);
1162 }
1163 
1164 SYSINIT(devdtr, SI_SUB_DEVFS, SI_ORDER_SECOND, devdtr_init, NULL);
1165