xref: /freebsd/sys/kern/kern_conf.c (revision 0b3105a37d7adcadcb720112fed4dc4e8040be99)
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 void destroy_dev_tq(void *ctx, int pending);
59 static int make_dev_credv(int flags, struct cdev **dres, struct cdevsw *devsw,
60     int unit, struct ucred *cr, uid_t uid, gid_t gid, int mode, const char *fmt,
61     va_list ap);
62 
63 static struct cdev_priv_list cdevp_free_list =
64     TAILQ_HEAD_INITIALIZER(cdevp_free_list);
65 static SLIST_HEAD(free_cdevsw, cdevsw) cdevsw_gt_post_list =
66     SLIST_HEAD_INITIALIZER(cdevsw_gt_post_list);
67 
68 void
69 dev_lock(void)
70 {
71 
72 	mtx_lock(&devmtx);
73 }
74 
75 /*
76  * Free all the memory collected while the cdev mutex was
77  * locked. Since devmtx is after the system map mutex, free() cannot
78  * be called immediately and is postponed until cdev mutex can be
79  * dropped.
80  */
81 static void
82 dev_unlock_and_free(void)
83 {
84 	struct cdev_priv_list cdp_free;
85 	struct free_cdevsw csw_free;
86 	struct cdev_priv *cdp;
87 	struct cdevsw *csw;
88 
89 	mtx_assert(&devmtx, MA_OWNED);
90 
91 	/*
92 	 * Make the local copy of the list heads while the dev_mtx is
93 	 * held. Free it later.
94 	 */
95 	TAILQ_INIT(&cdp_free);
96 	TAILQ_CONCAT(&cdp_free, &cdevp_free_list, cdp_list);
97 	csw_free = cdevsw_gt_post_list;
98 	SLIST_INIT(&cdevsw_gt_post_list);
99 
100 	mtx_unlock(&devmtx);
101 
102 	while ((cdp = TAILQ_FIRST(&cdp_free)) != NULL) {
103 		TAILQ_REMOVE(&cdp_free, cdp, cdp_list);
104 		devfs_free(&cdp->cdp_c);
105 	}
106 	while ((csw = SLIST_FIRST(&csw_free)) != NULL) {
107 		SLIST_REMOVE_HEAD(&csw_free, d_postfree_list);
108 		free(csw, M_DEVT);
109 	}
110 }
111 
112 static void
113 dev_free_devlocked(struct cdev *cdev)
114 {
115 	struct cdev_priv *cdp;
116 
117 	mtx_assert(&devmtx, MA_OWNED);
118 	cdp = cdev2priv(cdev);
119 	KASSERT((cdp->cdp_flags & CDP_UNREF_DTR) == 0,
120 	    ("destroy_dev() was not called after delist_dev(%p)", cdev));
121 	TAILQ_INSERT_HEAD(&cdevp_free_list, cdp, cdp_list);
122 }
123 
124 static void
125 cdevsw_free_devlocked(struct cdevsw *csw)
126 {
127 
128 	mtx_assert(&devmtx, MA_OWNED);
129 	SLIST_INSERT_HEAD(&cdevsw_gt_post_list, csw, d_postfree_list);
130 }
131 
132 void
133 dev_unlock(void)
134 {
135 
136 	mtx_unlock(&devmtx);
137 }
138 
139 void
140 dev_ref(struct cdev *dev)
141 {
142 
143 	mtx_assert(&devmtx, MA_NOTOWNED);
144 	mtx_lock(&devmtx);
145 	dev->si_refcount++;
146 	mtx_unlock(&devmtx);
147 }
148 
149 void
150 dev_refl(struct cdev *dev)
151 {
152 
153 	mtx_assert(&devmtx, MA_OWNED);
154 	dev->si_refcount++;
155 }
156 
157 void
158 dev_rel(struct cdev *dev)
159 {
160 	int flag = 0;
161 
162 	mtx_assert(&devmtx, MA_NOTOWNED);
163 	dev_lock();
164 	dev->si_refcount--;
165 	KASSERT(dev->si_refcount >= 0,
166 	    ("dev_rel(%s) gave negative count", devtoname(dev)));
167 #if 0
168 	if (dev->si_usecount == 0 &&
169 	    (dev->si_flags & SI_CHEAPCLONE) && (dev->si_flags & SI_NAMED))
170 		;
171 	else
172 #endif
173 	if (dev->si_devsw == NULL && dev->si_refcount == 0) {
174 		LIST_REMOVE(dev, si_list);
175 		flag = 1;
176 	}
177 	dev_unlock();
178 	if (flag)
179 		devfs_free(dev);
180 }
181 
182 struct cdevsw *
183 dev_refthread(struct cdev *dev, int *ref)
184 {
185 	struct cdevsw *csw;
186 	struct cdev_priv *cdp;
187 
188 	mtx_assert(&devmtx, MA_NOTOWNED);
189 	if ((dev->si_flags & SI_ETERNAL) != 0) {
190 		*ref = 0;
191 		return (dev->si_devsw);
192 	}
193 	dev_lock();
194 	csw = dev->si_devsw;
195 	if (csw != NULL) {
196 		cdp = cdev2priv(dev);
197 		if ((cdp->cdp_flags & CDP_SCHED_DTR) == 0)
198 			atomic_add_long(&dev->si_threadcount, 1);
199 		else
200 			csw = NULL;
201 	}
202 	dev_unlock();
203 	*ref = 1;
204 	return (csw);
205 }
206 
207 struct cdevsw *
208 devvn_refthread(struct vnode *vp, struct cdev **devp, int *ref)
209 {
210 	struct cdevsw *csw;
211 	struct cdev_priv *cdp;
212 	struct cdev *dev;
213 
214 	mtx_assert(&devmtx, MA_NOTOWNED);
215 	if ((vp->v_vflag & VV_ETERNALDEV) != 0) {
216 		dev = vp->v_rdev;
217 		if (dev == NULL)
218 			return (NULL);
219 		KASSERT((dev->si_flags & SI_ETERNAL) != 0,
220 		    ("Not eternal cdev"));
221 		*ref = 0;
222 		csw = dev->si_devsw;
223 		KASSERT(csw != NULL, ("Eternal cdev is destroyed"));
224 		*devp = dev;
225 		return (csw);
226 	}
227 
228 	csw = NULL;
229 	dev_lock();
230 	dev = vp->v_rdev;
231 	if (dev == NULL) {
232 		dev_unlock();
233 		return (NULL);
234 	}
235 	cdp = cdev2priv(dev);
236 	if ((cdp->cdp_flags & CDP_SCHED_DTR) == 0) {
237 		csw = dev->si_devsw;
238 		if (csw != NULL)
239 			atomic_add_long(&dev->si_threadcount, 1);
240 	}
241 	dev_unlock();
242 	if (csw != NULL) {
243 		*devp = dev;
244 		*ref = 1;
245 	}
246 	return (csw);
247 }
248 
249 void
250 dev_relthread(struct cdev *dev, int ref)
251 {
252 
253 	mtx_assert(&devmtx, MA_NOTOWNED);
254 	if (!ref)
255 		return;
256 	KASSERT(dev->si_threadcount > 0,
257 	    ("%s threadcount is wrong", dev->si_name));
258 	atomic_subtract_rel_long(&dev->si_threadcount, 1);
259 }
260 
261 int
262 nullop(void)
263 {
264 
265 	return (0);
266 }
267 
268 int
269 eopnotsupp(void)
270 {
271 
272 	return (EOPNOTSUPP);
273 }
274 
275 static int
276 enxio(void)
277 {
278 	return (ENXIO);
279 }
280 
281 static int
282 enodev(void)
283 {
284 	return (ENODEV);
285 }
286 
287 /* Define a dead_cdevsw for use when devices leave unexpectedly. */
288 
289 #define dead_open	(d_open_t *)enxio
290 #define dead_close	(d_close_t *)enxio
291 #define dead_read	(d_read_t *)enxio
292 #define dead_write	(d_write_t *)enxio
293 #define dead_ioctl	(d_ioctl_t *)enxio
294 #define dead_poll	(d_poll_t *)enodev
295 #define dead_mmap	(d_mmap_t *)enodev
296 
297 static void
298 dead_strategy(struct bio *bp)
299 {
300 
301 	biofinish(bp, NULL, ENXIO);
302 }
303 
304 #define dead_dump	(dumper_t *)enxio
305 #define dead_kqfilter	(d_kqfilter_t *)enxio
306 #define dead_mmap_single (d_mmap_single_t *)enodev
307 
308 static struct cdevsw dead_cdevsw = {
309 	.d_version =	D_VERSION,
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 make_dev_args *args, struct cdev *si)
570 {
571 	struct cdev *si2;
572 	struct cdevsw *csw;
573 
574 	mtx_assert(&devmtx, MA_OWNED);
575 	csw = args->mda_devsw;
576 	if (csw->d_flags & D_NEEDMINOR) {
577 		/* We may want to return an existing device */
578 		LIST_FOREACH(si2, &csw->d_devs, si_list) {
579 			if (dev2unit(si2) == args->mda_unit) {
580 				dev_free_devlocked(si);
581 				return (si2);
582 			}
583 		}
584 	}
585 	si->si_drv0 = args->mda_unit;
586 	si->si_devsw = csw;
587 	si->si_drv1 = args->mda_si_drv1;
588 	si->si_drv2 = args->mda_si_drv2;
589 	LIST_INSERT_HEAD(&csw->d_devs, si, si_list);
590 	return (si);
591 }
592 
593 static void
594 fini_cdevsw(struct cdevsw *devsw)
595 {
596 	struct cdevsw *gt;
597 
598 	if (devsw->d_gianttrick != NULL) {
599 		gt = devsw->d_gianttrick;
600 		memcpy(devsw, gt, sizeof *devsw);
601 		cdevsw_free_devlocked(gt);
602 		devsw->d_gianttrick = NULL;
603 	}
604 	devsw->d_flags &= ~D_INIT;
605 }
606 
607 static int
608 prep_cdevsw(struct cdevsw *devsw, int flags)
609 {
610 	struct cdevsw *dsw2;
611 
612 	mtx_assert(&devmtx, MA_OWNED);
613 	if (devsw->d_flags & D_INIT)
614 		return (0);
615 	if (devsw->d_flags & D_NEEDGIANT) {
616 		dev_unlock();
617 		dsw2 = malloc(sizeof *dsw2, M_DEVT,
618 		     (flags & MAKEDEV_NOWAIT) ? M_NOWAIT : M_WAITOK);
619 		dev_lock();
620 		if (dsw2 == NULL && !(devsw->d_flags & D_INIT))
621 			return (ENOMEM);
622 	} else
623 		dsw2 = NULL;
624 	if (devsw->d_flags & D_INIT) {
625 		if (dsw2 != NULL)
626 			cdevsw_free_devlocked(dsw2);
627 		return (0);
628 	}
629 
630 	if (devsw->d_version != D_VERSION_03) {
631 		printf(
632 		    "WARNING: Device driver \"%s\" has wrong version %s\n",
633 		    devsw->d_name == NULL ? "???" : devsw->d_name,
634 		    "and is disabled.  Recompile KLD module.");
635 		devsw->d_open = dead_open;
636 		devsw->d_close = dead_close;
637 		devsw->d_read = dead_read;
638 		devsw->d_write = dead_write;
639 		devsw->d_ioctl = dead_ioctl;
640 		devsw->d_poll = dead_poll;
641 		devsw->d_mmap = dead_mmap;
642 		devsw->d_mmap_single = dead_mmap_single;
643 		devsw->d_strategy = dead_strategy;
644 		devsw->d_dump = dead_dump;
645 		devsw->d_kqfilter = dead_kqfilter;
646 	}
647 
648 	if (devsw->d_flags & D_NEEDGIANT) {
649 		if (devsw->d_gianttrick == NULL) {
650 			memcpy(dsw2, devsw, sizeof *dsw2);
651 			devsw->d_gianttrick = dsw2;
652 			dsw2 = NULL;
653 		}
654 	}
655 
656 #define FIXUP(member, noop, giant) 				\
657 	do {							\
658 		if (devsw->member == NULL) {			\
659 			devsw->member = noop;			\
660 		} else if (devsw->d_flags & D_NEEDGIANT)	\
661 			devsw->member = giant;			\
662 		}						\
663 	while (0)
664 
665 	FIXUP(d_open,		null_open,	giant_open);
666 	FIXUP(d_fdopen,		NULL,		giant_fdopen);
667 	FIXUP(d_close,		null_close,	giant_close);
668 	FIXUP(d_read,		no_read,	giant_read);
669 	FIXUP(d_write,		no_write,	giant_write);
670 	FIXUP(d_ioctl,		no_ioctl,	giant_ioctl);
671 	FIXUP(d_poll,		no_poll,	giant_poll);
672 	FIXUP(d_mmap,		no_mmap,	giant_mmap);
673 	FIXUP(d_strategy,	no_strategy,	giant_strategy);
674 	FIXUP(d_kqfilter,	no_kqfilter,	giant_kqfilter);
675 	FIXUP(d_mmap_single,	no_mmap_single,	giant_mmap_single);
676 
677 	if (devsw->d_dump == NULL)	devsw->d_dump = no_dump;
678 
679 	LIST_INIT(&devsw->d_devs);
680 
681 	devsw->d_flags |= D_INIT;
682 
683 	if (dsw2 != NULL)
684 		cdevsw_free_devlocked(dsw2);
685 	return (0);
686 }
687 
688 static int
689 prep_devname(struct cdev *dev, const char *fmt, va_list ap)
690 {
691 	int len;
692 	char *from, *q, *s, *to;
693 
694 	mtx_assert(&devmtx, MA_OWNED);
695 
696 	len = vsnrprintf(dev->si_name, sizeof(dev->si_name), 32, fmt, ap);
697 	if (len > sizeof(dev->si_name) - 1)
698 		return (ENAMETOOLONG);
699 
700 	/* Strip leading slashes. */
701 	for (from = dev->si_name; *from == '/'; from++)
702 		;
703 
704 	for (to = dev->si_name; *from != '\0'; from++, to++) {
705 		/*
706 		 * Spaces and double quotation marks cause
707 		 * problems for the devctl(4) protocol.
708 		 * Reject names containing those characters.
709 		 */
710 		if (isspace(*from) || *from == '"')
711 			return (EINVAL);
712 		/* Treat multiple sequential slashes as single. */
713 		while (from[0] == '/' && from[1] == '/')
714 			from++;
715 		/* Trailing slash is considered invalid. */
716 		if (from[0] == '/' && from[1] == '\0')
717 			return (EINVAL);
718 		*to = *from;
719 	}
720 	*to = '\0';
721 
722 	if (dev->si_name[0] == '\0')
723 		return (EINVAL);
724 
725 	/* Disallow "." and ".." components. */
726 	for (s = dev->si_name;;) {
727 		for (q = s; *q != '/' && *q != '\0'; q++)
728 			;
729 		if (q - s == 1 && s[0] == '.')
730 			return (EINVAL);
731 		if (q - s == 2 && s[0] == '.' && s[1] == '.')
732 			return (EINVAL);
733 		if (*q != '/')
734 			break;
735 		s = q + 1;
736 	}
737 
738 	if (devfs_dev_exists(dev->si_name) != 0)
739 		return (EEXIST);
740 
741 	return (0);
742 }
743 
744 void
745 make_dev_args_init_impl(struct make_dev_args *args, size_t sz)
746 {
747 
748 	bzero(args, sz);
749 	args->mda_size = sz;
750 }
751 
752 static int
753 make_dev_sv(struct make_dev_args *args1, struct cdev **dres,
754     const char *fmt, va_list ap)
755 {
756 	struct cdev *dev, *dev_new;
757 	struct make_dev_args args;
758 	int res;
759 
760 	bzero(&args, sizeof(args));
761 	if (sizeof(args) < args1->mda_size)
762 		return (EINVAL);
763 	bcopy(args1, &args, args1->mda_size);
764 	KASSERT((args.mda_flags & MAKEDEV_WAITOK) == 0 ||
765 	    (args.mda_flags & MAKEDEV_NOWAIT) == 0,
766 	    ("make_dev_sv: both WAITOK and NOWAIT specified"));
767 	dev_new = devfs_alloc(args.mda_flags);
768 	if (dev_new == NULL)
769 		return (ENOMEM);
770 	dev_lock();
771 	res = prep_cdevsw(args.mda_devsw, args.mda_flags);
772 	if (res != 0) {
773 		dev_unlock();
774 		devfs_free(dev_new);
775 		return (res);
776 	}
777 	dev = newdev(&args, dev_new);
778 	if ((dev->si_flags & SI_NAMED) == 0) {
779 		res = prep_devname(dev, fmt, ap);
780 		if (res != 0) {
781 			if ((args.mda_flags & MAKEDEV_CHECKNAME) == 0) {
782 				panic(
783 			"make_dev_sv: bad si_name (error=%d, si_name=%s)",
784 				    res, dev->si_name);
785 			}
786 			if (dev == dev_new) {
787 				LIST_REMOVE(dev, si_list);
788 				dev_unlock();
789 				devfs_free(dev);
790 			} else
791 				dev_unlock();
792 			return (res);
793 		}
794 	}
795 	if ((args.mda_flags & MAKEDEV_REF) != 0)
796 		dev_refl(dev);
797 	if ((args.mda_flags & MAKEDEV_ETERNAL) != 0)
798 		dev->si_flags |= SI_ETERNAL;
799 	if (dev->si_flags & SI_CHEAPCLONE &&
800 	    dev->si_flags & SI_NAMED) {
801 		/*
802 		 * This is allowed as it removes races and generally
803 		 * simplifies cloning devices.
804 		 * XXX: still ??
805 		 */
806 		dev_unlock_and_free();
807 		*dres = dev;
808 		return (0);
809 	}
810 	KASSERT(!(dev->si_flags & SI_NAMED),
811 	    ("make_dev() by driver %s on pre-existing device (min=%x, name=%s)",
812 	    args.mda_devsw->d_name, dev2unit(dev), devtoname(dev)));
813 	dev->si_flags |= SI_NAMED;
814 	if (args.mda_cr != NULL)
815 		dev->si_cred = crhold(args.mda_cr);
816 	dev->si_uid = args.mda_uid;
817 	dev->si_gid = args.mda_gid;
818 	dev->si_mode = args.mda_mode;
819 
820 	devfs_create(dev);
821 	clean_unrhdrl(devfs_inos);
822 	dev_unlock_and_free();
823 
824 	notify_create(dev, args.mda_flags);
825 
826 	*dres = dev;
827 	return (0);
828 }
829 
830 int
831 make_dev_s(struct make_dev_args *args, struct cdev **dres,
832     const char *fmt, ...)
833 {
834 	va_list ap;
835 	int res;
836 
837 	va_start(ap, fmt);
838 	res = make_dev_sv(args, dres, fmt, ap);
839 	va_end(ap);
840 	return (res);
841 }
842 
843 static int
844 make_dev_credv(int flags, struct cdev **dres, struct cdevsw *devsw, int unit,
845     struct ucred *cr, uid_t uid, gid_t gid, int mode, const char *fmt,
846     va_list ap)
847 {
848 	struct make_dev_args args;
849 
850 	make_dev_args_init(&args);
851 	args.mda_flags = flags;
852 	args.mda_devsw = devsw;
853 	args.mda_cr = cr;
854 	args.mda_uid = uid;
855 	args.mda_gid = gid;
856 	args.mda_mode = mode;
857 	args.mda_unit = unit;
858 	return (make_dev_sv(&args, dres, fmt, ap));
859 }
860 
861 struct cdev *
862 make_dev(struct cdevsw *devsw, int unit, uid_t uid, gid_t gid, int mode,
863     const char *fmt, ...)
864 {
865 	struct cdev *dev;
866 	va_list ap;
867 	int res;
868 
869 	va_start(ap, fmt);
870 	res = make_dev_credv(0, &dev, devsw, unit, NULL, uid, gid, mode, fmt,
871 	    ap);
872 	va_end(ap);
873 	KASSERT(res == 0 && dev != NULL,
874 	    ("make_dev: failed make_dev_credv (error=%d)", res));
875 	return (dev);
876 }
877 
878 struct cdev *
879 make_dev_cred(struct cdevsw *devsw, int unit, struct ucred *cr, uid_t uid,
880     gid_t gid, int mode, const char *fmt, ...)
881 {
882 	struct cdev *dev;
883 	va_list ap;
884 	int res;
885 
886 	va_start(ap, fmt);
887 	res = make_dev_credv(0, &dev, devsw, unit, cr, uid, gid, mode, fmt, ap);
888 	va_end(ap);
889 
890 	KASSERT(res == 0 && dev != NULL,
891 	    ("make_dev_cred: failed make_dev_credv (error=%d)", res));
892 	return (dev);
893 }
894 
895 struct cdev *
896 make_dev_credf(int flags, struct cdevsw *devsw, int unit, struct ucred *cr,
897     uid_t uid, gid_t gid, int mode, const char *fmt, ...)
898 {
899 	struct cdev *dev;
900 	va_list ap;
901 	int res;
902 
903 	va_start(ap, fmt);
904 	res = make_dev_credv(flags, &dev, devsw, unit, cr, uid, gid, mode,
905 	    fmt, ap);
906 	va_end(ap);
907 
908 	KASSERT(((flags & MAKEDEV_NOWAIT) != 0 && res == ENOMEM) ||
909 	    ((flags & MAKEDEV_CHECKNAME) != 0 && res != ENOMEM) || res == 0,
910 	    ("make_dev_credf: failed make_dev_credv (error=%d)", res));
911 	return (res == 0 ? dev : NULL);
912 }
913 
914 int
915 make_dev_p(int flags, struct cdev **cdev, struct cdevsw *devsw,
916     struct ucred *cr, uid_t uid, gid_t gid, int mode, const char *fmt, ...)
917 {
918 	va_list ap;
919 	int res;
920 
921 	va_start(ap, fmt);
922 	res = make_dev_credv(flags, cdev, devsw, 0, cr, uid, gid, mode,
923 	    fmt, ap);
924 	va_end(ap);
925 
926 	KASSERT(((flags & MAKEDEV_NOWAIT) != 0 && res == ENOMEM) ||
927 	    ((flags & MAKEDEV_CHECKNAME) != 0 && res != ENOMEM) || res == 0,
928 	    ("make_dev_p: failed make_dev_credv (error=%d)", res));
929 	return (res);
930 }
931 
932 static void
933 dev_dependsl(struct cdev *pdev, struct cdev *cdev)
934 {
935 
936 	cdev->si_parent = pdev;
937 	cdev->si_flags |= SI_CHILD;
938 	LIST_INSERT_HEAD(&pdev->si_children, cdev, si_siblings);
939 }
940 
941 
942 void
943 dev_depends(struct cdev *pdev, struct cdev *cdev)
944 {
945 
946 	dev_lock();
947 	dev_dependsl(pdev, cdev);
948 	dev_unlock();
949 }
950 
951 static int
952 make_dev_alias_v(int flags, struct cdev **cdev, struct cdev *pdev,
953     const char *fmt, va_list ap)
954 {
955 	struct cdev *dev;
956 	int error;
957 
958 	KASSERT(pdev != NULL, ("make_dev_alias_v: pdev is NULL"));
959 	KASSERT((flags & MAKEDEV_WAITOK) == 0 || (flags & MAKEDEV_NOWAIT) == 0,
960 	    ("make_dev_alias_v: both WAITOK and NOWAIT specified"));
961 	KASSERT((flags & ~(MAKEDEV_WAITOK | MAKEDEV_NOWAIT |
962 	    MAKEDEV_CHECKNAME)) == 0,
963 	    ("make_dev_alias_v: invalid flags specified (flags=%02x)", flags));
964 
965 	dev = devfs_alloc(flags);
966 	if (dev == NULL)
967 		return (ENOMEM);
968 	dev_lock();
969 	dev->si_flags |= SI_ALIAS;
970 	error = prep_devname(dev, fmt, ap);
971 	if (error != 0) {
972 		if ((flags & MAKEDEV_CHECKNAME) == 0) {
973 			panic("make_dev_alias_v: bad si_name "
974 			    "(error=%d, si_name=%s)", error, dev->si_name);
975 		}
976 		dev_unlock();
977 		devfs_free(dev);
978 		return (error);
979 	}
980 	dev->si_flags |= SI_NAMED;
981 	devfs_create(dev);
982 	dev_dependsl(pdev, dev);
983 	clean_unrhdrl(devfs_inos);
984 	dev_unlock();
985 
986 	notify_create(dev, flags);
987 	*cdev = dev;
988 
989 	return (0);
990 }
991 
992 struct cdev *
993 make_dev_alias(struct cdev *pdev, const char *fmt, ...)
994 {
995 	struct cdev *dev;
996 	va_list ap;
997 	int res;
998 
999 	va_start(ap, fmt);
1000 	res = make_dev_alias_v(MAKEDEV_WAITOK, &dev, pdev, fmt, ap);
1001 	va_end(ap);
1002 
1003 	KASSERT(res == 0 && dev != NULL,
1004 	    ("make_dev_alias: failed make_dev_alias_v (error=%d)", res));
1005 	return (dev);
1006 }
1007 
1008 int
1009 make_dev_alias_p(int flags, struct cdev **cdev, struct cdev *pdev,
1010     const char *fmt, ...)
1011 {
1012 	va_list ap;
1013 	int res;
1014 
1015 	va_start(ap, fmt);
1016 	res = make_dev_alias_v(flags, cdev, pdev, fmt, ap);
1017 	va_end(ap);
1018 	return (res);
1019 }
1020 
1021 int
1022 make_dev_physpath_alias(int flags, struct cdev **cdev, struct cdev *pdev,
1023     struct cdev *old_alias, const char *physpath)
1024 {
1025 	char *devfspath;
1026 	int physpath_len;
1027 	int max_parentpath_len;
1028 	int parentpath_len;
1029 	int devfspathbuf_len;
1030 	int mflags;
1031 	int ret;
1032 
1033 	*cdev = NULL;
1034 	devfspath = NULL;
1035 	physpath_len = strlen(physpath);
1036 	ret = EINVAL;
1037 	if (physpath_len == 0)
1038 		goto out;
1039 
1040 	if (strncmp("id1,", physpath, 4) == 0) {
1041 		physpath += 4;
1042 		physpath_len -= 4;
1043 		if (physpath_len == 0)
1044 			goto out;
1045 	}
1046 
1047 	max_parentpath_len = SPECNAMELEN - physpath_len - /*/*/1;
1048 	parentpath_len = strlen(pdev->si_name);
1049 	if (max_parentpath_len < parentpath_len) {
1050 		if (bootverbose)
1051 			printf("WARNING: Unable to alias %s "
1052 			    "to %s/%s - path too long\n",
1053 			    pdev->si_name, physpath, pdev->si_name);
1054 		ret = ENAMETOOLONG;
1055 		goto out;
1056 	}
1057 
1058 	mflags = (flags & MAKEDEV_NOWAIT) ? M_NOWAIT : M_WAITOK;
1059 	devfspathbuf_len = physpath_len + /*/*/1 + parentpath_len + /*NUL*/1;
1060 	devfspath = malloc(devfspathbuf_len, M_DEVBUF, mflags);
1061 	if (devfspath == NULL) {
1062 		ret = ENOMEM;
1063 		goto out;
1064 	}
1065 
1066 	sprintf(devfspath, "%s/%s", physpath, pdev->si_name);
1067 	if (old_alias != NULL && strcmp(old_alias->si_name, devfspath) == 0) {
1068 		/* Retain the existing alias. */
1069 		*cdev = old_alias;
1070 		old_alias = NULL;
1071 		ret = 0;
1072 	} else {
1073 		ret = make_dev_alias_p(flags, cdev, pdev, "%s", devfspath);
1074 	}
1075 out:
1076 	if (old_alias != NULL)
1077 		destroy_dev(old_alias);
1078 	if (devfspath != NULL)
1079 		free(devfspath, M_DEVBUF);
1080 	return (ret);
1081 }
1082 
1083 static void
1084 destroy_devl(struct cdev *dev)
1085 {
1086 	struct cdevsw *csw;
1087 	struct cdev_privdata *p;
1088 	struct cdev_priv *cdp;
1089 
1090 	mtx_assert(&devmtx, MA_OWNED);
1091 	KASSERT(dev->si_flags & SI_NAMED,
1092 	    ("WARNING: Driver mistake: destroy_dev on %d\n", dev2unit(dev)));
1093 	KASSERT((dev->si_flags & SI_ETERNAL) == 0,
1094 	    ("WARNING: Driver mistake: destroy_dev on eternal %d\n",
1095 	     dev2unit(dev)));
1096 
1097 	cdp = cdev2priv(dev);
1098 	if ((cdp->cdp_flags & CDP_UNREF_DTR) == 0) {
1099 		/*
1100 		 * Avoid race with dev_rel(), e.g. from the populate
1101 		 * loop.  If CDP_UNREF_DTR flag is set, the reference
1102 		 * to be dropped at the end of destroy_devl() was
1103 		 * already taken by delist_dev_locked().
1104 		 */
1105 		dev_refl(dev);
1106 
1107 		devfs_destroy(dev);
1108 	}
1109 
1110 	/* Remove name marking */
1111 	dev->si_flags &= ~SI_NAMED;
1112 
1113 	/* If we are a child, remove us from the parents list */
1114 	if (dev->si_flags & SI_CHILD) {
1115 		LIST_REMOVE(dev, si_siblings);
1116 		dev->si_flags &= ~SI_CHILD;
1117 	}
1118 
1119 	/* Kill our children */
1120 	while (!LIST_EMPTY(&dev->si_children))
1121 		destroy_devl(LIST_FIRST(&dev->si_children));
1122 
1123 	/* Remove from clone list */
1124 	if (dev->si_flags & SI_CLONELIST) {
1125 		LIST_REMOVE(dev, si_clone);
1126 		dev->si_flags &= ~SI_CLONELIST;
1127 	}
1128 
1129 	csw = dev->si_devsw;
1130 	dev->si_devsw = NULL;	/* already NULL for SI_ALIAS */
1131 	while (csw != NULL && csw->d_purge != NULL && dev->si_threadcount) {
1132 		csw->d_purge(dev);
1133 		msleep(csw, &devmtx, PRIBIO, "devprg", hz/10);
1134 		if (dev->si_threadcount)
1135 			printf("Still %lu threads in %s\n",
1136 			    dev->si_threadcount, devtoname(dev));
1137 	}
1138 	while (dev->si_threadcount != 0) {
1139 		/* Use unique dummy wait ident */
1140 		msleep(&csw, &devmtx, PRIBIO, "devdrn", hz / 10);
1141 	}
1142 
1143 	dev_unlock();
1144 	if ((cdp->cdp_flags & CDP_UNREF_DTR) == 0) {
1145 		/* avoid out of order notify events */
1146 		notify_destroy(dev);
1147 	}
1148 	mtx_lock(&cdevpriv_mtx);
1149 	while ((p = LIST_FIRST(&cdp->cdp_fdpriv)) != NULL) {
1150 		devfs_destroy_cdevpriv(p);
1151 		mtx_lock(&cdevpriv_mtx);
1152 	}
1153 	mtx_unlock(&cdevpriv_mtx);
1154 	dev_lock();
1155 
1156 	dev->si_drv1 = 0;
1157 	dev->si_drv2 = 0;
1158 	bzero(&dev->__si_u, sizeof(dev->__si_u));
1159 
1160 	if (!(dev->si_flags & SI_ALIAS)) {
1161 		/* Remove from cdevsw list */
1162 		LIST_REMOVE(dev, si_list);
1163 
1164 		/* If cdevsw has no more struct cdev *'s, clean it */
1165 		if (LIST_EMPTY(&csw->d_devs)) {
1166 			fini_cdevsw(csw);
1167 			wakeup(&csw->d_devs);
1168 		}
1169 	}
1170 	dev->si_flags &= ~SI_ALIAS;
1171 	cdp->cdp_flags &= ~CDP_UNREF_DTR;
1172 	dev->si_refcount--;
1173 
1174 	if (dev->si_refcount > 0)
1175 		LIST_INSERT_HEAD(&dead_cdevsw.d_devs, dev, si_list);
1176 	else
1177 		dev_free_devlocked(dev);
1178 }
1179 
1180 static void
1181 delist_dev_locked(struct cdev *dev)
1182 {
1183 	struct cdev_priv *cdp;
1184 	struct cdev *child;
1185 
1186 	mtx_assert(&devmtx, MA_OWNED);
1187 	cdp = cdev2priv(dev);
1188 	if ((cdp->cdp_flags & CDP_UNREF_DTR) != 0)
1189 		return;
1190 	cdp->cdp_flags |= CDP_UNREF_DTR;
1191 	dev_refl(dev);
1192 	devfs_destroy(dev);
1193 	LIST_FOREACH(child, &dev->si_children, si_siblings)
1194 		delist_dev_locked(child);
1195 	dev_unlock();
1196 	/* ensure the destroy event is queued in order */
1197 	notify_destroy(dev);
1198 	dev_lock();
1199 }
1200 
1201 /*
1202  * This function will delist a character device and its children from
1203  * the directory listing and create a destroy event without waiting
1204  * for all character device references to go away. At some later point
1205  * destroy_dev() must be called to complete the character device
1206  * destruction. After calling this function the character device name
1207  * can instantly be re-used.
1208  */
1209 void
1210 delist_dev(struct cdev *dev)
1211 {
1212 
1213 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "delist_dev");
1214 	dev_lock();
1215 	delist_dev_locked(dev);
1216 	dev_unlock();
1217 }
1218 
1219 void
1220 destroy_dev(struct cdev *dev)
1221 {
1222 
1223 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "destroy_dev");
1224 	dev_lock();
1225 	destroy_devl(dev);
1226 	dev_unlock_and_free();
1227 }
1228 
1229 const char *
1230 devtoname(struct cdev *dev)
1231 {
1232 
1233 	return (dev->si_name);
1234 }
1235 
1236 int
1237 dev_stdclone(char *name, char **namep, const char *stem, int *unit)
1238 {
1239 	int u, i;
1240 
1241 	i = strlen(stem);
1242 	if (bcmp(stem, name, i) != 0)
1243 		return (0);
1244 	if (!isdigit(name[i]))
1245 		return (0);
1246 	u = 0;
1247 	if (name[i] == '0' && isdigit(name[i+1]))
1248 		return (0);
1249 	while (isdigit(name[i])) {
1250 		u *= 10;
1251 		u += name[i++] - '0';
1252 	}
1253 	if (u > 0xffffff)
1254 		return (0);
1255 	*unit = u;
1256 	if (namep)
1257 		*namep = &name[i];
1258 	if (name[i])
1259 		return (2);
1260 	return (1);
1261 }
1262 
1263 /*
1264  * Helper functions for cloning device drivers.
1265  *
1266  * The objective here is to make it unnecessary for the device drivers to
1267  * use rman or similar to manage their unit number space.  Due to the way
1268  * we do "on-demand" devices, using rman or other "private" methods
1269  * will be very tricky to lock down properly once we lock down this file.
1270  *
1271  * Instead we give the drivers these routines which puts the struct cdev *'s
1272  * that are to be managed on their own list, and gives the driver the ability
1273  * to ask for the first free unit number or a given specified unit number.
1274  *
1275  * In addition these routines support paired devices (pty, nmdm and similar)
1276  * by respecting a number of "flag" bits in the minor number.
1277  *
1278  */
1279 
1280 struct clonedevs {
1281 	LIST_HEAD(,cdev)	head;
1282 };
1283 
1284 void
1285 clone_setup(struct clonedevs **cdp)
1286 {
1287 
1288 	*cdp = malloc(sizeof **cdp, M_DEVBUF, M_WAITOK | M_ZERO);
1289 	LIST_INIT(&(*cdp)->head);
1290 }
1291 
1292 int
1293 clone_create(struct clonedevs **cdp, struct cdevsw *csw, int *up,
1294     struct cdev **dp, int extra)
1295 {
1296 	struct clonedevs *cd;
1297 	struct cdev *dev, *ndev, *dl, *de;
1298 	struct make_dev_args args;
1299 	int unit, low, u;
1300 
1301 	KASSERT(*cdp != NULL,
1302 	    ("clone_setup() not called in driver \"%s\"", csw->d_name));
1303 	KASSERT(!(extra & CLONE_UNITMASK),
1304 	    ("Illegal extra bits (0x%x) in clone_create", extra));
1305 	KASSERT(*up <= CLONE_UNITMASK,
1306 	    ("Too high unit (0x%x) in clone_create", *up));
1307 	KASSERT(csw->d_flags & D_NEEDMINOR,
1308 	    ("clone_create() on cdevsw without minor numbers"));
1309 
1310 
1311 	/*
1312 	 * Search the list for a lot of things in one go:
1313 	 *   A preexisting match is returned immediately.
1314 	 *   The lowest free unit number if we are passed -1, and the place
1315 	 *	 in the list where we should insert that new element.
1316 	 *   The place to insert a specified unit number, if applicable
1317 	 *       the end of the list.
1318 	 */
1319 	unit = *up;
1320 	ndev = devfs_alloc(MAKEDEV_WAITOK);
1321 	dev_lock();
1322 	prep_cdevsw(csw, MAKEDEV_WAITOK);
1323 	low = extra;
1324 	de = dl = NULL;
1325 	cd = *cdp;
1326 	LIST_FOREACH(dev, &cd->head, si_clone) {
1327 		KASSERT(dev->si_flags & SI_CLONELIST,
1328 		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
1329 		u = dev2unit(dev);
1330 		if (u == (unit | extra)) {
1331 			*dp = dev;
1332 			dev_unlock();
1333 			devfs_free(ndev);
1334 			return (0);
1335 		}
1336 		if (unit == -1 && u == low) {
1337 			low++;
1338 			de = dev;
1339 			continue;
1340 		} else if (u < (unit | extra)) {
1341 			de = dev;
1342 			continue;
1343 		} else if (u > (unit | extra)) {
1344 			dl = dev;
1345 			break;
1346 		}
1347 	}
1348 	if (unit == -1)
1349 		unit = low & CLONE_UNITMASK;
1350 	make_dev_args_init(&args);
1351 	args.mda_unit = unit | extra;
1352 	args.mda_devsw = csw;
1353 	dev = newdev(&args, ndev);
1354 	if (dev->si_flags & SI_CLONELIST) {
1355 		printf("dev %p (%s) is on clonelist\n", dev, dev->si_name);
1356 		printf("unit=%d, low=%d, extra=0x%x\n", unit, low, extra);
1357 		LIST_FOREACH(dev, &cd->head, si_clone) {
1358 			printf("\t%p %s\n", dev, dev->si_name);
1359 		}
1360 		panic("foo");
1361 	}
1362 	KASSERT(!(dev->si_flags & SI_CLONELIST),
1363 	    ("Dev %p(%s) should not be on clonelist", dev, dev->si_name));
1364 	if (dl != NULL)
1365 		LIST_INSERT_BEFORE(dl, dev, si_clone);
1366 	else if (de != NULL)
1367 		LIST_INSERT_AFTER(de, dev, si_clone);
1368 	else
1369 		LIST_INSERT_HEAD(&cd->head, dev, si_clone);
1370 	dev->si_flags |= SI_CLONELIST;
1371 	*up = unit;
1372 	dev_unlock_and_free();
1373 	return (1);
1374 }
1375 
1376 /*
1377  * Kill everything still on the list.  The driver should already have
1378  * disposed of any softc hung of the struct cdev *'s at this time.
1379  */
1380 void
1381 clone_cleanup(struct clonedevs **cdp)
1382 {
1383 	struct cdev *dev;
1384 	struct cdev_priv *cp;
1385 	struct clonedevs *cd;
1386 
1387 	cd = *cdp;
1388 	if (cd == NULL)
1389 		return;
1390 	dev_lock();
1391 	while (!LIST_EMPTY(&cd->head)) {
1392 		dev = LIST_FIRST(&cd->head);
1393 		LIST_REMOVE(dev, si_clone);
1394 		KASSERT(dev->si_flags & SI_CLONELIST,
1395 		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
1396 		dev->si_flags &= ~SI_CLONELIST;
1397 		cp = cdev2priv(dev);
1398 		if (!(cp->cdp_flags & CDP_SCHED_DTR)) {
1399 			cp->cdp_flags |= CDP_SCHED_DTR;
1400 			KASSERT(dev->si_flags & SI_NAMED,
1401 				("Driver has goofed in cloning underways udev %jx unit %x",
1402 				(uintmax_t)dev2udev(dev), dev2unit(dev)));
1403 			destroy_devl(dev);
1404 		}
1405 	}
1406 	dev_unlock_and_free();
1407 	free(cd, M_DEVBUF);
1408 	*cdp = NULL;
1409 }
1410 
1411 static TAILQ_HEAD(, cdev_priv) dev_ddtr =
1412 	TAILQ_HEAD_INITIALIZER(dev_ddtr);
1413 static struct task dev_dtr_task = TASK_INITIALIZER(0, destroy_dev_tq, NULL);
1414 
1415 static void
1416 destroy_dev_tq(void *ctx, int pending)
1417 {
1418 	struct cdev_priv *cp;
1419 	struct cdev *dev;
1420 	void (*cb)(void *);
1421 	void *cb_arg;
1422 
1423 	dev_lock();
1424 	while (!TAILQ_EMPTY(&dev_ddtr)) {
1425 		cp = TAILQ_FIRST(&dev_ddtr);
1426 		dev = &cp->cdp_c;
1427 		KASSERT(cp->cdp_flags & CDP_SCHED_DTR,
1428 		    ("cdev %p in dev_destroy_tq without CDP_SCHED_DTR", cp));
1429 		TAILQ_REMOVE(&dev_ddtr, cp, cdp_dtr_list);
1430 		cb = cp->cdp_dtr_cb;
1431 		cb_arg = cp->cdp_dtr_cb_arg;
1432 		destroy_devl(dev);
1433 		dev_unlock_and_free();
1434 		dev_rel(dev);
1435 		if (cb != NULL)
1436 			cb(cb_arg);
1437 		dev_lock();
1438 	}
1439 	dev_unlock();
1440 }
1441 
1442 /*
1443  * devmtx shall be locked on entry. devmtx will be unlocked after
1444  * function return.
1445  */
1446 static int
1447 destroy_dev_sched_cbl(struct cdev *dev, void (*cb)(void *), void *arg)
1448 {
1449 	struct cdev_priv *cp;
1450 
1451 	mtx_assert(&devmtx, MA_OWNED);
1452 	cp = cdev2priv(dev);
1453 	if (cp->cdp_flags & CDP_SCHED_DTR) {
1454 		dev_unlock();
1455 		return (0);
1456 	}
1457 	dev_refl(dev);
1458 	cp->cdp_flags |= CDP_SCHED_DTR;
1459 	cp->cdp_dtr_cb = cb;
1460 	cp->cdp_dtr_cb_arg = arg;
1461 	TAILQ_INSERT_TAIL(&dev_ddtr, cp, cdp_dtr_list);
1462 	dev_unlock();
1463 	taskqueue_enqueue(taskqueue_swi_giant, &dev_dtr_task);
1464 	return (1);
1465 }
1466 
1467 int
1468 destroy_dev_sched_cb(struct cdev *dev, void (*cb)(void *), void *arg)
1469 {
1470 
1471 	dev_lock();
1472 	return (destroy_dev_sched_cbl(dev, cb, arg));
1473 }
1474 
1475 int
1476 destroy_dev_sched(struct cdev *dev)
1477 {
1478 
1479 	return (destroy_dev_sched_cb(dev, NULL, NULL));
1480 }
1481 
1482 void
1483 destroy_dev_drain(struct cdevsw *csw)
1484 {
1485 
1486 	dev_lock();
1487 	while (!LIST_EMPTY(&csw->d_devs)) {
1488 		msleep(&csw->d_devs, &devmtx, PRIBIO, "devscd", hz/10);
1489 	}
1490 	dev_unlock();
1491 }
1492 
1493 void
1494 drain_dev_clone_events(void)
1495 {
1496 
1497 	sx_xlock(&clone_drain_lock);
1498 	sx_xunlock(&clone_drain_lock);
1499 }
1500 
1501 #include "opt_ddb.h"
1502 #ifdef DDB
1503 #include <sys/kernel.h>
1504 
1505 #include <ddb/ddb.h>
1506 
1507 DB_SHOW_COMMAND(cdev, db_show_cdev)
1508 {
1509 	struct cdev_priv *cdp;
1510 	struct cdev *dev;
1511 	u_int flags;
1512 	char buf[512];
1513 
1514 	if (!have_addr) {
1515 		TAILQ_FOREACH(cdp, &cdevp_list, cdp_list) {
1516 			dev = &cdp->cdp_c;
1517 			db_printf("%s %p\n", dev->si_name, dev);
1518 			if (db_pager_quit)
1519 				break;
1520 		}
1521 		return;
1522 	}
1523 
1524 	dev = (struct cdev *)addr;
1525 	cdp = cdev2priv(dev);
1526 	db_printf("dev %s ref %d use %ld thr %ld inuse %u fdpriv %p\n",
1527 	    dev->si_name, dev->si_refcount, dev->si_usecount,
1528 	    dev->si_threadcount, cdp->cdp_inuse, cdp->cdp_fdpriv.lh_first);
1529 	db_printf("devsw %p si_drv0 %d si_drv1 %p si_drv2 %p\n",
1530 	    dev->si_devsw, dev->si_drv0, dev->si_drv1, dev->si_drv2);
1531 	flags = dev->si_flags;
1532 #define	SI_FLAG(flag)	do {						\
1533 	if (flags & (flag)) {						\
1534 		if (buf[0] != '\0')					\
1535 			strlcat(buf, ", ", sizeof(buf));		\
1536 		strlcat(buf, (#flag) + 3, sizeof(buf));			\
1537 		flags &= ~(flag);					\
1538 	}								\
1539 } while (0)
1540 	buf[0] = '\0';
1541 	SI_FLAG(SI_ETERNAL);
1542 	SI_FLAG(SI_ALIAS);
1543 	SI_FLAG(SI_NAMED);
1544 	SI_FLAG(SI_CHEAPCLONE);
1545 	SI_FLAG(SI_CHILD);
1546 	SI_FLAG(SI_DUMPDEV);
1547 	SI_FLAG(SI_CLONELIST);
1548 	db_printf("si_flags %s\n", buf);
1549 
1550 	flags = cdp->cdp_flags;
1551 #define	CDP_FLAG(flag)	do {						\
1552 	if (flags & (flag)) {						\
1553 		if (buf[0] != '\0')					\
1554 			strlcat(buf, ", ", sizeof(buf));		\
1555 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
1556 		flags &= ~(flag);					\
1557 	}								\
1558 } while (0)
1559 	buf[0] = '\0';
1560 	CDP_FLAG(CDP_ACTIVE);
1561 	CDP_FLAG(CDP_SCHED_DTR);
1562 	db_printf("cdp_flags %s\n", buf);
1563 }
1564 #endif
1565