xref: /freebsd/sys/kern/kern_conf.c (revision b2db760808f74bb53c232900091c9da801ebbfcc)
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/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 #include <vm/vm.h>
51 
52 static MALLOC_DEFINE(M_DEVT, "cdev", "cdev storage");
53 
54 struct mtx devmtx;
55 static void destroy_devl(struct cdev *dev);
56 static int destroy_dev_sched_cbl(struct cdev *dev,
57     void (*cb)(void *), void *arg);
58 static int make_dev_credv(int flags, struct cdev **dres, struct cdevsw *devsw,
59     int unit, 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(cdevsw_gt_post_list);
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 = cdev2priv(cdev);
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, int *ref)
181 {
182 	struct cdevsw *csw;
183 	struct cdev_priv *cdp;
184 
185 	mtx_assert(&devmtx, MA_NOTOWNED);
186 	if ((dev->si_flags & SI_ETERNAL) != 0) {
187 		*ref = 0;
188 		return (dev->si_devsw);
189 	}
190 	dev_lock();
191 	csw = dev->si_devsw;
192 	if (csw != NULL) {
193 		cdp = cdev2priv(dev);
194 		if ((cdp->cdp_flags & CDP_SCHED_DTR) == 0)
195 			dev->si_threadcount++;
196 		else
197 			csw = NULL;
198 	}
199 	dev_unlock();
200 	*ref = 1;
201 	return (csw);
202 }
203 
204 struct cdevsw *
205 devvn_refthread(struct vnode *vp, struct cdev **devp, int *ref)
206 {
207 	struct cdevsw *csw;
208 	struct cdev_priv *cdp;
209 	struct cdev *dev;
210 
211 	mtx_assert(&devmtx, MA_NOTOWNED);
212 	if ((vp->v_vflag & VV_ETERNALDEV) != 0) {
213 		dev = vp->v_rdev;
214 		if (dev == NULL)
215 			return (NULL);
216 		KASSERT((dev->si_flags & SI_ETERNAL) != 0,
217 		    ("Not eternal cdev"));
218 		*ref = 0;
219 		csw = dev->si_devsw;
220 		KASSERT(csw != NULL, ("Eternal cdev is destroyed"));
221 		*devp = dev;
222 		return (csw);
223 	}
224 
225 	csw = NULL;
226 	dev_lock();
227 	dev = vp->v_rdev;
228 	if (dev == NULL) {
229 		dev_unlock();
230 		return (NULL);
231 	}
232 	cdp = cdev2priv(dev);
233 	if ((cdp->cdp_flags & CDP_SCHED_DTR) == 0) {
234 		csw = dev->si_devsw;
235 		if (csw != NULL)
236 			dev->si_threadcount++;
237 	}
238 	dev_unlock();
239 	if (csw != NULL) {
240 		*devp = dev;
241 		*ref = 1;
242 	}
243 	return (csw);
244 }
245 
246 void
247 dev_relthread(struct cdev *dev, int ref)
248 {
249 
250 	mtx_assert(&devmtx, MA_NOTOWNED);
251 	if (!ref)
252 		return;
253 	dev_lock();
254 	KASSERT(dev->si_threadcount > 0,
255 	    ("%s threadcount is wrong", dev->si_name));
256 	dev->si_threadcount--;
257 	dev_unlock();
258 }
259 
260 int
261 nullop(void)
262 {
263 
264 	return (0);
265 }
266 
267 int
268 eopnotsupp(void)
269 {
270 
271 	return (EOPNOTSUPP);
272 }
273 
274 static int
275 enxio(void)
276 {
277 	return (ENXIO);
278 }
279 
280 static int
281 enodev(void)
282 {
283 	return (ENODEV);
284 }
285 
286 /* Define a dead_cdevsw for use when devices leave unexpectedly. */
287 
288 #define dead_open	(d_open_t *)enxio
289 #define dead_close	(d_close_t *)enxio
290 #define dead_read	(d_read_t *)enxio
291 #define dead_write	(d_write_t *)enxio
292 #define dead_ioctl	(d_ioctl_t *)enxio
293 #define dead_poll	(d_poll_t *)enodev
294 #define dead_mmap	(d_mmap_t *)enodev
295 
296 static void
297 dead_strategy(struct bio *bp)
298 {
299 
300 	biofinish(bp, NULL, ENXIO);
301 }
302 
303 #define dead_dump	(dumper_t *)enxio
304 #define dead_kqfilter	(d_kqfilter_t *)enxio
305 #define dead_mmap_single (d_mmap_single_t *)enodev
306 
307 static struct cdevsw dead_cdevsw = {
308 	.d_version =	D_VERSION,
309 	.d_flags =	D_NEEDGIANT, /* XXX: does dead_strategy need this ? */
310 	.d_open =	dead_open,
311 	.d_close =	dead_close,
312 	.d_read =	dead_read,
313 	.d_write =	dead_write,
314 	.d_ioctl =	dead_ioctl,
315 	.d_poll =	dead_poll,
316 	.d_mmap =	dead_mmap,
317 	.d_strategy =	dead_strategy,
318 	.d_name =	"dead",
319 	.d_dump =	dead_dump,
320 	.d_kqfilter =	dead_kqfilter,
321 	.d_mmap_single = dead_mmap_single
322 };
323 
324 /* Default methods if driver does not specify method */
325 
326 #define null_open	(d_open_t *)nullop
327 #define null_close	(d_close_t *)nullop
328 #define no_read		(d_read_t *)enodev
329 #define no_write	(d_write_t *)enodev
330 #define no_ioctl	(d_ioctl_t *)enodev
331 #define no_mmap		(d_mmap_t *)enodev
332 #define no_kqfilter	(d_kqfilter_t *)enodev
333 #define no_mmap_single	(d_mmap_single_t *)enodev
334 
335 static void
336 no_strategy(struct bio *bp)
337 {
338 
339 	biofinish(bp, NULL, ENODEV);
340 }
341 
342 static int
343 no_poll(struct cdev *dev __unused, int events, struct thread *td __unused)
344 {
345 
346 	return (poll_no_poll(events));
347 }
348 
349 #define no_dump		(dumper_t *)enodev
350 
351 static int
352 giant_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
353 {
354 	struct cdevsw *dsw;
355 	int ref, retval;
356 
357 	dsw = dev_refthread(dev, &ref);
358 	if (dsw == NULL)
359 		return (ENXIO);
360 	mtx_lock(&Giant);
361 	retval = dsw->d_gianttrick->d_open(dev, oflags, devtype, td);
362 	mtx_unlock(&Giant);
363 	dev_relthread(dev, ref);
364 	return (retval);
365 }
366 
367 static int
368 giant_fdopen(struct cdev *dev, int oflags, struct thread *td, struct file *fp)
369 {
370 	struct cdevsw *dsw;
371 	int ref, retval;
372 
373 	dsw = dev_refthread(dev, &ref);
374 	if (dsw == NULL)
375 		return (ENXIO);
376 	mtx_lock(&Giant);
377 	retval = dsw->d_gianttrick->d_fdopen(dev, oflags, td, fp);
378 	mtx_unlock(&Giant);
379 	dev_relthread(dev, ref);
380 	return (retval);
381 }
382 
383 static int
384 giant_close(struct cdev *dev, int fflag, int devtype, struct thread *td)
385 {
386 	struct cdevsw *dsw;
387 	int ref, retval;
388 
389 	dsw = dev_refthread(dev, &ref);
390 	if (dsw == NULL)
391 		return (ENXIO);
392 	mtx_lock(&Giant);
393 	retval = dsw->d_gianttrick->d_close(dev, fflag, devtype, td);
394 	mtx_unlock(&Giant);
395 	dev_relthread(dev, ref);
396 	return (retval);
397 }
398 
399 static void
400 giant_strategy(struct bio *bp)
401 {
402 	struct cdevsw *dsw;
403 	struct cdev *dev;
404 	int ref;
405 
406 	dev = bp->bio_dev;
407 	dsw = dev_refthread(dev, &ref);
408 	if (dsw == NULL) {
409 		biofinish(bp, NULL, ENXIO);
410 		return;
411 	}
412 	mtx_lock(&Giant);
413 	dsw->d_gianttrick->d_strategy(bp);
414 	mtx_unlock(&Giant);
415 	dev_relthread(dev, ref);
416 }
417 
418 static int
419 giant_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
420 {
421 	struct cdevsw *dsw;
422 	int ref, retval;
423 
424 	dsw = dev_refthread(dev, &ref);
425 	if (dsw == NULL)
426 		return (ENXIO);
427 	mtx_lock(&Giant);
428 	retval = dsw->d_gianttrick->d_ioctl(dev, cmd, data, fflag, td);
429 	mtx_unlock(&Giant);
430 	dev_relthread(dev, ref);
431 	return (retval);
432 }
433 
434 static int
435 giant_read(struct cdev *dev, struct uio *uio, int ioflag)
436 {
437 	struct cdevsw *dsw;
438 	int ref, retval;
439 
440 	dsw = dev_refthread(dev, &ref);
441 	if (dsw == NULL)
442 		return (ENXIO);
443 	mtx_lock(&Giant);
444 	retval = dsw->d_gianttrick->d_read(dev, uio, ioflag);
445 	mtx_unlock(&Giant);
446 	dev_relthread(dev, ref);
447 	return (retval);
448 }
449 
450 static int
451 giant_write(struct cdev *dev, struct uio *uio, int ioflag)
452 {
453 	struct cdevsw *dsw;
454 	int ref, retval;
455 
456 	dsw = dev_refthread(dev, &ref);
457 	if (dsw == NULL)
458 		return (ENXIO);
459 	mtx_lock(&Giant);
460 	retval = dsw->d_gianttrick->d_write(dev, uio, ioflag);
461 	mtx_unlock(&Giant);
462 	dev_relthread(dev, ref);
463 	return (retval);
464 }
465 
466 static int
467 giant_poll(struct cdev *dev, int events, struct thread *td)
468 {
469 	struct cdevsw *dsw;
470 	int ref, retval;
471 
472 	dsw = dev_refthread(dev, &ref);
473 	if (dsw == NULL)
474 		return (ENXIO);
475 	mtx_lock(&Giant);
476 	retval = dsw->d_gianttrick->d_poll(dev, events, td);
477 	mtx_unlock(&Giant);
478 	dev_relthread(dev, ref);
479 	return (retval);
480 }
481 
482 static int
483 giant_kqfilter(struct cdev *dev, struct knote *kn)
484 {
485 	struct cdevsw *dsw;
486 	int ref, retval;
487 
488 	dsw = dev_refthread(dev, &ref);
489 	if (dsw == NULL)
490 		return (ENXIO);
491 	mtx_lock(&Giant);
492 	retval = dsw->d_gianttrick->d_kqfilter(dev, kn);
493 	mtx_unlock(&Giant);
494 	dev_relthread(dev, ref);
495 	return (retval);
496 }
497 
498 static int
499 giant_mmap(struct cdev *dev, vm_ooffset_t offset, vm_paddr_t *paddr, int nprot,
500     vm_memattr_t *memattr)
501 {
502 	struct cdevsw *dsw;
503 	int ref, retval;
504 
505 	dsw = dev_refthread(dev, &ref);
506 	if (dsw == NULL)
507 		return (ENXIO);
508 	mtx_lock(&Giant);
509 	retval = dsw->d_gianttrick->d_mmap(dev, offset, paddr, nprot,
510 	    memattr);
511 	mtx_unlock(&Giant);
512 	dev_relthread(dev, ref);
513 	return (retval);
514 }
515 
516 static int
517 giant_mmap_single(struct cdev *dev, vm_ooffset_t *offset, vm_size_t size,
518     vm_object_t *object, int nprot)
519 {
520 	struct cdevsw *dsw;
521 	int ref, retval;
522 
523 	dsw = dev_refthread(dev, &ref);
524 	if (dsw == NULL)
525 		return (ENXIO);
526 	mtx_lock(&Giant);
527 	retval = dsw->d_gianttrick->d_mmap_single(dev, offset, size, object,
528 	    nprot);
529 	mtx_unlock(&Giant);
530 	dev_relthread(dev, ref);
531 	return (retval);
532 }
533 
534 static void
535 notify(struct cdev *dev, const char *ev, int flags)
536 {
537 	static const char prefix[] = "cdev=";
538 	char *data;
539 	int namelen, mflags;
540 
541 	if (cold)
542 		return;
543 	mflags = (flags & MAKEDEV_NOWAIT) ? M_NOWAIT : M_WAITOK;
544 	namelen = strlen(dev->si_name);
545 	data = malloc(namelen + sizeof(prefix), M_TEMP, mflags);
546 	if (data == NULL)
547 		return;
548 	memcpy(data, prefix, sizeof(prefix) - 1);
549 	memcpy(data + sizeof(prefix) - 1, dev->si_name, namelen + 1);
550 	devctl_notify_f("DEVFS", "CDEV", ev, data, mflags);
551 	free(data, M_TEMP);
552 }
553 
554 static void
555 notify_create(struct cdev *dev, int flags)
556 {
557 
558 	notify(dev, "CREATE", flags);
559 }
560 
561 static void
562 notify_destroy(struct cdev *dev)
563 {
564 
565 	notify(dev, "DESTROY", MAKEDEV_WAITOK);
566 }
567 
568 static struct cdev *
569 newdev(struct cdevsw *csw, int unit, struct cdev *si)
570 {
571 	struct cdev *si2;
572 
573 	mtx_assert(&devmtx, MA_OWNED);
574 	if (csw->d_flags & D_NEEDMINOR) {
575 		/* We may want to return an existing device */
576 		LIST_FOREACH(si2, &csw->d_devs, si_list) {
577 			if (dev2unit(si2) == unit) {
578 				dev_free_devlocked(si);
579 				return (si2);
580 			}
581 		}
582 	}
583 	si->si_drv0 = unit;
584 	si->si_devsw = csw;
585 	LIST_INSERT_HEAD(&csw->d_devs, si, si_list);
586 	return (si);
587 }
588 
589 static void
590 fini_cdevsw(struct cdevsw *devsw)
591 {
592 	struct cdevsw *gt;
593 
594 	if (devsw->d_gianttrick != NULL) {
595 		gt = devsw->d_gianttrick;
596 		memcpy(devsw, gt, sizeof *devsw);
597 		cdevsw_free_devlocked(gt);
598 		devsw->d_gianttrick = NULL;
599 	}
600 	devsw->d_flags &= ~D_INIT;
601 }
602 
603 static int
604 prep_cdevsw(struct cdevsw *devsw, int flags)
605 {
606 	struct cdevsw *dsw2;
607 
608 	mtx_assert(&devmtx, MA_OWNED);
609 	if (devsw->d_flags & D_INIT)
610 		return (0);
611 	if (devsw->d_flags & D_NEEDGIANT) {
612 		dev_unlock();
613 		dsw2 = malloc(sizeof *dsw2, M_DEVT,
614 		     (flags & MAKEDEV_NOWAIT) ? M_NOWAIT : M_WAITOK);
615 		dev_lock();
616 		if (dsw2 == NULL && !(devsw->d_flags & D_INIT))
617 			return (ENOMEM);
618 	} else
619 		dsw2 = NULL;
620 	if (devsw->d_flags & D_INIT) {
621 		if (dsw2 != NULL)
622 			cdevsw_free_devlocked(dsw2);
623 		return (0);
624 	}
625 
626 	if (devsw->d_version != D_VERSION_03) {
627 		printf(
628 		    "WARNING: Device driver \"%s\" has wrong version %s\n",
629 		    devsw->d_name == NULL ? "???" : devsw->d_name,
630 		    "and is disabled.  Recompile KLD module.");
631 		devsw->d_open = dead_open;
632 		devsw->d_close = dead_close;
633 		devsw->d_read = dead_read;
634 		devsw->d_write = dead_write;
635 		devsw->d_ioctl = dead_ioctl;
636 		devsw->d_poll = dead_poll;
637 		devsw->d_mmap = dead_mmap;
638 		devsw->d_mmap_single = dead_mmap_single;
639 		devsw->d_strategy = dead_strategy;
640 		devsw->d_dump = dead_dump;
641 		devsw->d_kqfilter = dead_kqfilter;
642 	}
643 
644 	if (devsw->d_flags & D_NEEDGIANT) {
645 		if (devsw->d_gianttrick == NULL) {
646 			memcpy(dsw2, devsw, sizeof *dsw2);
647 			devsw->d_gianttrick = dsw2;
648 			dsw2 = NULL;
649 		}
650 	}
651 
652 #define FIXUP(member, noop, giant) 				\
653 	do {							\
654 		if (devsw->member == NULL) {			\
655 			devsw->member = noop;			\
656 		} else if (devsw->d_flags & D_NEEDGIANT)	\
657 			devsw->member = giant;			\
658 		}						\
659 	while (0)
660 
661 	FIXUP(d_open,		null_open,	giant_open);
662 	FIXUP(d_fdopen,		NULL,		giant_fdopen);
663 	FIXUP(d_close,		null_close,	giant_close);
664 	FIXUP(d_read,		no_read,	giant_read);
665 	FIXUP(d_write,		no_write,	giant_write);
666 	FIXUP(d_ioctl,		no_ioctl,	giant_ioctl);
667 	FIXUP(d_poll,		no_poll,	giant_poll);
668 	FIXUP(d_mmap,		no_mmap,	giant_mmap);
669 	FIXUP(d_strategy,	no_strategy,	giant_strategy);
670 	FIXUP(d_kqfilter,	no_kqfilter,	giant_kqfilter);
671 	FIXUP(d_mmap_single,	no_mmap_single,	giant_mmap_single);
672 
673 	if (devsw->d_dump == NULL)	devsw->d_dump = no_dump;
674 
675 	LIST_INIT(&devsw->d_devs);
676 
677 	devsw->d_flags |= D_INIT;
678 
679 	if (dsw2 != NULL)
680 		cdevsw_free_devlocked(dsw2);
681 	return (0);
682 }
683 
684 static int
685 make_dev_credv(int flags, struct cdev **dres, struct cdevsw *devsw, int unit,
686     struct ucred *cr, uid_t uid, gid_t gid, int mode, const char *fmt,
687     va_list ap)
688 {
689 	struct cdev *dev;
690 	int i, res;
691 
692 	KASSERT((flags & MAKEDEV_WAITOK) == 0 || (flags & MAKEDEV_NOWAIT) == 0,
693 	    ("make_dev_credv: both WAITOK and NOWAIT specified"));
694 	dev = devfs_alloc(flags);
695 	if (dev == NULL)
696 		return (ENOMEM);
697 	dev_lock();
698 	res = prep_cdevsw(devsw, flags);
699 	if (res != 0) {
700 		dev_unlock();
701 		devfs_free(dev);
702 		return (res);
703 	}
704 	dev = newdev(devsw, unit, dev);
705 	if (flags & MAKEDEV_REF)
706 		dev_refl(dev);
707 	if (flags & MAKEDEV_ETERNAL)
708 		dev->si_flags |= SI_ETERNAL;
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 		*dres = dev;
718 		return (0);
719 	}
720 	KASSERT(!(dev->si_flags & SI_NAMED),
721 	    ("make_dev() by driver %s on pre-existing device (min=%x, name=%s)",
722 	    devsw->d_name, dev2unit(dev), devtoname(dev)));
723 
724 	i = vsnrprintf(dev->__si_namebuf, sizeof dev->__si_namebuf, 32, fmt, ap);
725 	if (i > (sizeof dev->__si_namebuf - 1)) {
726 		printf("WARNING: Device name truncated! (%s)\n",
727 		    dev->__si_namebuf);
728 	}
729 
730 	dev->si_flags |= SI_NAMED;
731 	if (cr != NULL)
732 		dev->si_cred = crhold(cr);
733 	dev->si_uid = uid;
734 	dev->si_gid = gid;
735 	dev->si_mode = mode;
736 
737 	devfs_create(dev);
738 	clean_unrhdrl(devfs_inos);
739 	dev_unlock_and_free();
740 
741 	notify_create(dev, flags);
742 
743 	*dres = dev;
744 	return (0);
745 }
746 
747 struct cdev *
748 make_dev(struct cdevsw *devsw, int unit, uid_t uid, gid_t gid, int mode,
749     const char *fmt, ...)
750 {
751 	struct cdev *dev;
752 	va_list ap;
753 	int res;
754 
755 	va_start(ap, fmt);
756 	res = make_dev_credv(0, &dev, devsw, unit, NULL, uid, gid, mode, fmt,
757 	    ap);
758 	va_end(ap);
759 	KASSERT(res == 0 && dev != NULL, ("make_dev: failed make_dev_credv"));
760 	return (dev);
761 }
762 
763 struct cdev *
764 make_dev_cred(struct cdevsw *devsw, int unit, struct ucred *cr, uid_t uid,
765     gid_t gid, int mode, const char *fmt, ...)
766 {
767 	struct cdev *dev;
768 	va_list ap;
769 	int res;
770 
771 	va_start(ap, fmt);
772 	res = make_dev_credv(0, &dev, devsw, unit, cr, uid, gid, mode, fmt, ap);
773 	va_end(ap);
774 
775 	KASSERT(res == 0 && dev != NULL,
776 	    ("make_dev_cred: failed make_dev_credv"));
777 	return (dev);
778 }
779 
780 struct cdev *
781 make_dev_credf(int flags, struct cdevsw *devsw, int unit, struct ucred *cr,
782     uid_t uid, gid_t gid, int mode, const char *fmt, ...)
783 {
784 	struct cdev *dev;
785 	va_list ap;
786 	int res;
787 
788 	va_start(ap, fmt);
789 	res = make_dev_credv(flags, &dev, devsw, unit, cr, uid, gid, mode,
790 	    fmt, ap);
791 	va_end(ap);
792 
793 	KASSERT((flags & MAKEDEV_NOWAIT) != 0 || res == 0,
794 	    ("make_dev_credf: failed make_dev_credv"));
795 	return (res == 0 ? dev : NULL);
796 }
797 
798 int
799 make_dev_p(int flags, struct cdev **cdev, struct cdevsw *devsw,
800     struct ucred *cr, uid_t uid, gid_t gid, int mode, const char *fmt, ...)
801 {
802 	va_list ap;
803 	int res;
804 
805 	va_start(ap, fmt);
806 	res = make_dev_credv(flags, cdev, devsw, 0, cr, uid, gid, mode,
807 	    fmt, ap);
808 	va_end(ap);
809 
810 	KASSERT((flags & MAKEDEV_NOWAIT) != 0 || res == 0,
811 	    ("make_dev_p: failed make_dev_credv"));
812 	return (res);
813 }
814 
815 static void
816 dev_dependsl(struct cdev *pdev, struct cdev *cdev)
817 {
818 
819 	cdev->si_parent = pdev;
820 	cdev->si_flags |= SI_CHILD;
821 	LIST_INSERT_HEAD(&pdev->si_children, cdev, si_siblings);
822 }
823 
824 
825 void
826 dev_depends(struct cdev *pdev, struct cdev *cdev)
827 {
828 
829 	dev_lock();
830 	dev_dependsl(pdev, cdev);
831 	dev_unlock();
832 }
833 
834 struct cdev *
835 make_dev_alias(struct cdev *pdev, const char *fmt, ...)
836 {
837 	struct cdev *dev;
838 	va_list ap;
839 	int i;
840 
841 	KASSERT(pdev != NULL, ("NULL pdev"));
842 	dev = devfs_alloc(MAKEDEV_WAITOK);
843 	dev_lock();
844 	dev->si_flags |= SI_ALIAS;
845 	dev->si_flags |= SI_NAMED;
846 	va_start(ap, fmt);
847 	i = vsnrprintf(dev->__si_namebuf, sizeof dev->__si_namebuf, 32, fmt, ap);
848 	if (i > (sizeof dev->__si_namebuf - 1)) {
849 		printf("WARNING: Device name truncated! (%s)\n",
850 		    dev->__si_namebuf);
851 	}
852 	va_end(ap);
853 
854 	devfs_create(dev);
855 	dev_dependsl(pdev, dev);
856 	clean_unrhdrl(devfs_inos);
857 	dev_unlock();
858 
859 	notify_create(dev, MAKEDEV_WAITOK);
860 
861 	return (dev);
862 }
863 
864 static void
865 destroy_devl(struct cdev *dev)
866 {
867 	struct cdevsw *csw;
868 	struct cdev_privdata *p, *p1;
869 
870 	mtx_assert(&devmtx, MA_OWNED);
871 	KASSERT(dev->si_flags & SI_NAMED,
872 	    ("WARNING: Driver mistake: destroy_dev on %d\n", dev2unit(dev)));
873 	KASSERT((dev->si_flags & SI_ETERNAL) == 0,
874 	    ("WARNING: Driver mistake: destroy_dev on eternal %d\n",
875 	     dev2unit(dev)));
876 
877 	devfs_destroy(dev);
878 
879 	/* Remove name marking */
880 	dev->si_flags &= ~SI_NAMED;
881 
882 	/* If we are a child, remove us from the parents list */
883 	if (dev->si_flags & SI_CHILD) {
884 		LIST_REMOVE(dev, si_siblings);
885 		dev->si_flags &= ~SI_CHILD;
886 	}
887 
888 	/* Kill our children */
889 	while (!LIST_EMPTY(&dev->si_children))
890 		destroy_devl(LIST_FIRST(&dev->si_children));
891 
892 	/* Remove from clone list */
893 	if (dev->si_flags & SI_CLONELIST) {
894 		LIST_REMOVE(dev, si_clone);
895 		dev->si_flags &= ~SI_CLONELIST;
896 	}
897 
898 	dev->si_refcount++;	/* Avoid race with dev_rel() */
899 	csw = dev->si_devsw;
900 	dev->si_devsw = NULL;	/* already NULL for SI_ALIAS */
901 	while (csw != NULL && csw->d_purge != NULL && dev->si_threadcount) {
902 		csw->d_purge(dev);
903 		msleep(csw, &devmtx, PRIBIO, "devprg", hz/10);
904 		if (dev->si_threadcount)
905 			printf("Still %lu threads in %s\n",
906 			    dev->si_threadcount, devtoname(dev));
907 	}
908 	while (dev->si_threadcount != 0) {
909 		/* Use unique dummy wait ident */
910 		msleep(&csw, &devmtx, PRIBIO, "devdrn", hz / 10);
911 	}
912 
913 	dev_unlock();
914 	notify_destroy(dev);
915 	mtx_lock(&cdevpriv_mtx);
916 	LIST_FOREACH_SAFE(p, &cdev2priv(dev)->cdp_fdpriv, cdpd_list, p1) {
917 		devfs_destroy_cdevpriv(p);
918 		mtx_lock(&cdevpriv_mtx);
919 	}
920 	mtx_unlock(&cdevpriv_mtx);
921 	dev_lock();
922 
923 	dev->si_drv1 = 0;
924 	dev->si_drv2 = 0;
925 	bzero(&dev->__si_u, sizeof(dev->__si_u));
926 
927 	if (!(dev->si_flags & SI_ALIAS)) {
928 		/* Remove from cdevsw list */
929 		LIST_REMOVE(dev, si_list);
930 
931 		/* If cdevsw has no more struct cdev *'s, clean it */
932 		if (LIST_EMPTY(&csw->d_devs)) {
933 			fini_cdevsw(csw);
934 			wakeup(&csw->d_devs);
935 		}
936 	}
937 	dev->si_flags &= ~SI_ALIAS;
938 	dev->si_refcount--;	/* Avoid race with dev_rel() */
939 
940 	if (dev->si_refcount > 0) {
941 		LIST_INSERT_HEAD(&dead_cdevsw.d_devs, dev, si_list);
942 	} else {
943 		dev_free_devlocked(dev);
944 	}
945 }
946 
947 void
948 destroy_dev(struct cdev *dev)
949 {
950 
951 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "destroy_dev");
952 	dev_lock();
953 	destroy_devl(dev);
954 	dev_unlock_and_free();
955 }
956 
957 const char *
958 devtoname(struct cdev *dev)
959 {
960 
961 	return (dev->si_name);
962 }
963 
964 int
965 dev_stdclone(char *name, char **namep, const char *stem, int *unit)
966 {
967 	int u, i;
968 
969 	i = strlen(stem);
970 	if (bcmp(stem, name, i) != 0)
971 		return (0);
972 	if (!isdigit(name[i]))
973 		return (0);
974 	u = 0;
975 	if (name[i] == '0' && isdigit(name[i+1]))
976 		return (0);
977 	while (isdigit(name[i])) {
978 		u *= 10;
979 		u += name[i++] - '0';
980 	}
981 	if (u > 0xffffff)
982 		return (0);
983 	*unit = u;
984 	if (namep)
985 		*namep = &name[i];
986 	if (name[i])
987 		return (2);
988 	return (1);
989 }
990 
991 /*
992  * Helper functions for cloning device drivers.
993  *
994  * The objective here is to make it unnecessary for the device drivers to
995  * use rman or similar to manage their unit number space.  Due to the way
996  * we do "on-demand" devices, using rman or other "private" methods
997  * will be very tricky to lock down properly once we lock down this file.
998  *
999  * Instead we give the drivers these routines which puts the struct cdev *'s
1000  * that are to be managed on their own list, and gives the driver the ability
1001  * to ask for the first free unit number or a given specified unit number.
1002  *
1003  * In addition these routines support paired devices (pty, nmdm and similar)
1004  * by respecting a number of "flag" bits in the minor number.
1005  *
1006  */
1007 
1008 struct clonedevs {
1009 	LIST_HEAD(,cdev)	head;
1010 };
1011 
1012 void
1013 clone_setup(struct clonedevs **cdp)
1014 {
1015 
1016 	*cdp = malloc(sizeof **cdp, M_DEVBUF, M_WAITOK | M_ZERO);
1017 	LIST_INIT(&(*cdp)->head);
1018 }
1019 
1020 int
1021 clone_create(struct clonedevs **cdp, struct cdevsw *csw, int *up,
1022     struct cdev **dp, int extra)
1023 {
1024 	struct clonedevs *cd;
1025 	struct cdev *dev, *ndev, *dl, *de;
1026 	int unit, low, u;
1027 
1028 	KASSERT(*cdp != NULL,
1029 	    ("clone_setup() not called in driver \"%s\"", csw->d_name));
1030 	KASSERT(!(extra & CLONE_UNITMASK),
1031 	    ("Illegal extra bits (0x%x) in clone_create", extra));
1032 	KASSERT(*up <= CLONE_UNITMASK,
1033 	    ("Too high unit (0x%x) in clone_create", *up));
1034 	KASSERT(csw->d_flags & D_NEEDMINOR,
1035 	    ("clone_create() on cdevsw without minor numbers"));
1036 
1037 
1038 	/*
1039 	 * Search the list for a lot of things in one go:
1040 	 *   A preexisting match is returned immediately.
1041 	 *   The lowest free unit number if we are passed -1, and the place
1042 	 *	 in the list where we should insert that new element.
1043 	 *   The place to insert a specified unit number, if applicable
1044 	 *       the end of the list.
1045 	 */
1046 	unit = *up;
1047 	ndev = devfs_alloc(MAKEDEV_WAITOK);
1048 	dev_lock();
1049 	prep_cdevsw(csw, MAKEDEV_WAITOK);
1050 	low = extra;
1051 	de = dl = NULL;
1052 	cd = *cdp;
1053 	LIST_FOREACH(dev, &cd->head, si_clone) {
1054 		KASSERT(dev->si_flags & SI_CLONELIST,
1055 		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
1056 		u = dev2unit(dev);
1057 		if (u == (unit | extra)) {
1058 			*dp = dev;
1059 			dev_unlock();
1060 			devfs_free(ndev);
1061 			return (0);
1062 		}
1063 		if (unit == -1 && u == low) {
1064 			low++;
1065 			de = dev;
1066 			continue;
1067 		} else if (u < (unit | extra)) {
1068 			de = dev;
1069 			continue;
1070 		} else if (u > (unit | extra)) {
1071 			dl = dev;
1072 			break;
1073 		}
1074 	}
1075 	if (unit == -1)
1076 		unit = low & CLONE_UNITMASK;
1077 	dev = newdev(csw, unit | extra, ndev);
1078 	if (dev->si_flags & SI_CLONELIST) {
1079 		printf("dev %p (%s) is on clonelist\n", dev, dev->si_name);
1080 		printf("unit=%d, low=%d, extra=0x%x\n", unit, low, extra);
1081 		LIST_FOREACH(dev, &cd->head, si_clone) {
1082 			printf("\t%p %s\n", dev, dev->si_name);
1083 		}
1084 		panic("foo");
1085 	}
1086 	KASSERT(!(dev->si_flags & SI_CLONELIST),
1087 	    ("Dev %p(%s) should not be on clonelist", dev, dev->si_name));
1088 	if (dl != NULL)
1089 		LIST_INSERT_BEFORE(dl, dev, si_clone);
1090 	else if (de != NULL)
1091 		LIST_INSERT_AFTER(de, dev, si_clone);
1092 	else
1093 		LIST_INSERT_HEAD(&cd->head, dev, si_clone);
1094 	dev->si_flags |= SI_CLONELIST;
1095 	*up = unit;
1096 	dev_unlock_and_free();
1097 	return (1);
1098 }
1099 
1100 /*
1101  * Kill everything still on the list.  The driver should already have
1102  * disposed of any softc hung of the struct cdev *'s at this time.
1103  */
1104 void
1105 clone_cleanup(struct clonedevs **cdp)
1106 {
1107 	struct cdev *dev;
1108 	struct cdev_priv *cp;
1109 	struct clonedevs *cd;
1110 
1111 	cd = *cdp;
1112 	if (cd == NULL)
1113 		return;
1114 	dev_lock();
1115 	while (!LIST_EMPTY(&cd->head)) {
1116 		dev = LIST_FIRST(&cd->head);
1117 		LIST_REMOVE(dev, si_clone);
1118 		KASSERT(dev->si_flags & SI_CLONELIST,
1119 		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
1120 		dev->si_flags &= ~SI_CLONELIST;
1121 		cp = cdev2priv(dev);
1122 		if (!(cp->cdp_flags & CDP_SCHED_DTR)) {
1123 			cp->cdp_flags |= CDP_SCHED_DTR;
1124 			KASSERT(dev->si_flags & SI_NAMED,
1125 				("Driver has goofed in cloning underways udev %x unit %x", dev2udev(dev), dev2unit(dev)));
1126 			destroy_devl(dev);
1127 		}
1128 	}
1129 	dev_unlock_and_free();
1130 	free(cd, M_DEVBUF);
1131 	*cdp = NULL;
1132 }
1133 
1134 static TAILQ_HEAD(, cdev_priv) dev_ddtr =
1135 	TAILQ_HEAD_INITIALIZER(dev_ddtr);
1136 static struct task dev_dtr_task;
1137 
1138 static void
1139 destroy_dev_tq(void *ctx, int pending)
1140 {
1141 	struct cdev_priv *cp;
1142 	struct cdev *dev;
1143 	void (*cb)(void *);
1144 	void *cb_arg;
1145 
1146 	dev_lock();
1147 	while (!TAILQ_EMPTY(&dev_ddtr)) {
1148 		cp = TAILQ_FIRST(&dev_ddtr);
1149 		dev = &cp->cdp_c;
1150 		KASSERT(cp->cdp_flags & CDP_SCHED_DTR,
1151 		    ("cdev %p in dev_destroy_tq without CDP_SCHED_DTR", cp));
1152 		TAILQ_REMOVE(&dev_ddtr, cp, cdp_dtr_list);
1153 		cb = cp->cdp_dtr_cb;
1154 		cb_arg = cp->cdp_dtr_cb_arg;
1155 		destroy_devl(dev);
1156 		dev_unlock_and_free();
1157 		dev_rel(dev);
1158 		if (cb != NULL)
1159 			cb(cb_arg);
1160 		dev_lock();
1161 	}
1162 	dev_unlock();
1163 }
1164 
1165 /*
1166  * devmtx shall be locked on entry. devmtx will be unlocked after
1167  * function return.
1168  */
1169 static int
1170 destroy_dev_sched_cbl(struct cdev *dev, void (*cb)(void *), void *arg)
1171 {
1172 	struct cdev_priv *cp;
1173 
1174 	mtx_assert(&devmtx, MA_OWNED);
1175 	cp = cdev2priv(dev);
1176 	if (cp->cdp_flags & CDP_SCHED_DTR) {
1177 		dev_unlock();
1178 		return (0);
1179 	}
1180 	dev_refl(dev);
1181 	cp->cdp_flags |= CDP_SCHED_DTR;
1182 	cp->cdp_dtr_cb = cb;
1183 	cp->cdp_dtr_cb_arg = arg;
1184 	TAILQ_INSERT_TAIL(&dev_ddtr, cp, cdp_dtr_list);
1185 	dev_unlock();
1186 	taskqueue_enqueue(taskqueue_swi_giant, &dev_dtr_task);
1187 	return (1);
1188 }
1189 
1190 int
1191 destroy_dev_sched_cb(struct cdev *dev, void (*cb)(void *), void *arg)
1192 {
1193 
1194 	dev_lock();
1195 	return (destroy_dev_sched_cbl(dev, cb, arg));
1196 }
1197 
1198 int
1199 destroy_dev_sched(struct cdev *dev)
1200 {
1201 
1202 	return (destroy_dev_sched_cb(dev, NULL, NULL));
1203 }
1204 
1205 void
1206 destroy_dev_drain(struct cdevsw *csw)
1207 {
1208 
1209 	dev_lock();
1210 	while (!LIST_EMPTY(&csw->d_devs)) {
1211 		msleep(&csw->d_devs, &devmtx, PRIBIO, "devscd", hz/10);
1212 	}
1213 	dev_unlock();
1214 }
1215 
1216 void
1217 drain_dev_clone_events(void)
1218 {
1219 
1220 	sx_xlock(&clone_drain_lock);
1221 	sx_xunlock(&clone_drain_lock);
1222 }
1223 
1224 static void
1225 devdtr_init(void *dummy __unused)
1226 {
1227 
1228 	TASK_INIT(&dev_dtr_task, 0, destroy_dev_tq, NULL);
1229 }
1230 
1231 SYSINIT(devdtr, SI_SUB_DEVFS, SI_ORDER_SECOND, devdtr_init, NULL);
1232 
1233 #include "opt_ddb.h"
1234 #ifdef DDB
1235 #include <sys/kernel.h>
1236 
1237 #include <ddb/ddb.h>
1238 
1239 DB_SHOW_COMMAND(cdev, db_show_cdev)
1240 {
1241 	struct cdev_priv *cdp;
1242 	struct cdev *dev;
1243 	u_int flags;
1244 	char buf[512];
1245 
1246 	if (!have_addr) {
1247 		TAILQ_FOREACH(cdp, &cdevp_list, cdp_list) {
1248 			dev = &cdp->cdp_c;
1249 			db_printf("%s %p\n", dev->si_name, dev);
1250 			if (db_pager_quit)
1251 				break;
1252 		}
1253 		return;
1254 	}
1255 
1256 	dev = (struct cdev *)addr;
1257 	cdp = cdev2priv(dev);
1258 	db_printf("dev %s ref %d use %ld thr %ld inuse %u fdpriv %p\n",
1259 	    dev->si_name, dev->si_refcount, dev->si_usecount,
1260 	    dev->si_threadcount, cdp->cdp_inuse, cdp->cdp_fdpriv.lh_first);
1261 	db_printf("devsw %p si_drv0 %d si_drv1 %p si_drv2 %p\n",
1262 	    dev->si_devsw, dev->si_drv0, dev->si_drv1, dev->si_drv2);
1263 	flags = dev->si_flags;
1264 #define	SI_FLAG(flag)	do {						\
1265 	if (flags & (flag)) {						\
1266 		if (buf[0] != '\0')					\
1267 			strlcat(buf, ", ", sizeof(buf));		\
1268 		strlcat(buf, (#flag) + 3, sizeof(buf));			\
1269 		flags &= ~(flag);					\
1270 	}								\
1271 } while (0)
1272 	buf[0] = '\0';
1273 	SI_FLAG(SI_ETERNAL);
1274 	SI_FLAG(SI_ALIAS);
1275 	SI_FLAG(SI_NAMED);
1276 	SI_FLAG(SI_CHEAPCLONE);
1277 	SI_FLAG(SI_CHILD);
1278 	SI_FLAG(SI_DEVOPEN);
1279 	SI_FLAG(SI_CONSOPEN);
1280 	SI_FLAG(SI_DUMPDEV);
1281 	SI_FLAG(SI_CANDELETE);
1282 	SI_FLAG(SI_CLONELIST);
1283 	db_printf("si_flags %s\n", buf);
1284 
1285 	flags = cdp->cdp_flags;
1286 #define	CDP_FLAG(flag)	do {						\
1287 	if (flags & (flag)) {						\
1288 		if (buf[0] != '\0')					\
1289 			strlcat(buf, ", ", sizeof(buf));		\
1290 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
1291 		flags &= ~(flag);					\
1292 	}								\
1293 } while (0)
1294 	buf[0] = '\0';
1295 	CDP_FLAG(CDP_ACTIVE);
1296 	CDP_FLAG(CDP_SCHED_DTR);
1297 	db_printf("cdp_flags %s\n", buf);
1298 }
1299 #endif
1300