xref: /freebsd/sys/kern/kern_conf.c (revision aa64588d28258aef88cc33b8043112e8856948d0)
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 struct cdev *make_dev_credv(int flags,
59     struct cdevsw *devsw, int unit,
60     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 	TAILQ_INSERT_HEAD(&cdevp_free_list, cdp, cdp_list);
120 }
121 
122 static void
123 cdevsw_free_devlocked(struct cdevsw *csw)
124 {
125 
126 	mtx_assert(&devmtx, MA_OWNED);
127 	SLIST_INSERT_HEAD(&cdevsw_gt_post_list, csw, d_postfree_list);
128 }
129 
130 void
131 dev_unlock(void)
132 {
133 
134 	mtx_unlock(&devmtx);
135 }
136 
137 void
138 dev_ref(struct cdev *dev)
139 {
140 
141 	mtx_assert(&devmtx, MA_NOTOWNED);
142 	mtx_lock(&devmtx);
143 	dev->si_refcount++;
144 	mtx_unlock(&devmtx);
145 }
146 
147 void
148 dev_refl(struct cdev *dev)
149 {
150 
151 	mtx_assert(&devmtx, MA_OWNED);
152 	dev->si_refcount++;
153 }
154 
155 void
156 dev_rel(struct cdev *dev)
157 {
158 	int flag = 0;
159 
160 	mtx_assert(&devmtx, MA_NOTOWNED);
161 	dev_lock();
162 	dev->si_refcount--;
163 	KASSERT(dev->si_refcount >= 0,
164 	    ("dev_rel(%s) gave negative count", devtoname(dev)));
165 #if 0
166 	if (dev->si_usecount == 0 &&
167 	    (dev->si_flags & SI_CHEAPCLONE) && (dev->si_flags & SI_NAMED))
168 		;
169 	else
170 #endif
171 	if (dev->si_devsw == NULL && dev->si_refcount == 0) {
172 		LIST_REMOVE(dev, si_list);
173 		flag = 1;
174 	}
175 	dev_unlock();
176 	if (flag)
177 		devfs_free(dev);
178 }
179 
180 struct cdevsw *
181 dev_refthread(struct cdev *dev)
182 {
183 	struct cdevsw *csw;
184 	struct cdev_priv *cdp;
185 
186 	mtx_assert(&devmtx, MA_NOTOWNED);
187 	dev_lock();
188 	csw = dev->si_devsw;
189 	if (csw != NULL) {
190 		cdp = cdev2priv(dev);
191 		if ((cdp->cdp_flags & CDP_SCHED_DTR) == 0)
192 			dev->si_threadcount++;
193 		else
194 			csw = NULL;
195 	}
196 	dev_unlock();
197 	return (csw);
198 }
199 
200 struct cdevsw *
201 devvn_refthread(struct vnode *vp, struct cdev **devp)
202 {
203 	struct cdevsw *csw;
204 	struct cdev_priv *cdp;
205 
206 	mtx_assert(&devmtx, MA_NOTOWNED);
207 	csw = NULL;
208 	dev_lock();
209 	*devp = vp->v_rdev;
210 	if (*devp != NULL) {
211 		cdp = cdev2priv(*devp);
212 		if ((cdp->cdp_flags & CDP_SCHED_DTR) == 0) {
213 			csw = (*devp)->si_devsw;
214 			if (csw != NULL)
215 				(*devp)->si_threadcount++;
216 		}
217 	}
218 	dev_unlock();
219 	return (csw);
220 }
221 
222 void
223 dev_relthread(struct cdev *dev)
224 {
225 
226 	mtx_assert(&devmtx, MA_NOTOWNED);
227 	dev_lock();
228 	KASSERT(dev->si_threadcount > 0,
229 	    ("%s threadcount is wrong", dev->si_name));
230 	dev->si_threadcount--;
231 	dev_unlock();
232 }
233 
234 int
235 nullop(void)
236 {
237 
238 	return (0);
239 }
240 
241 int
242 eopnotsupp(void)
243 {
244 
245 	return (EOPNOTSUPP);
246 }
247 
248 static int
249 enxio(void)
250 {
251 	return (ENXIO);
252 }
253 
254 static int
255 enodev(void)
256 {
257 	return (ENODEV);
258 }
259 
260 /* Define a dead_cdevsw for use when devices leave unexpectedly. */
261 
262 #define dead_open	(d_open_t *)enxio
263 #define dead_close	(d_close_t *)enxio
264 #define dead_read	(d_read_t *)enxio
265 #define dead_write	(d_write_t *)enxio
266 #define dead_ioctl	(d_ioctl_t *)enxio
267 #define dead_poll	(d_poll_t *)enodev
268 #define dead_mmap	(d_mmap_t *)enodev
269 
270 static void
271 dead_strategy(struct bio *bp)
272 {
273 
274 	biofinish(bp, NULL, ENXIO);
275 }
276 
277 #define dead_dump	(dumper_t *)enxio
278 #define dead_kqfilter	(d_kqfilter_t *)enxio
279 #define dead_mmap_single (d_mmap_single_t *)enodev
280 
281 static struct cdevsw dead_cdevsw = {
282 	.d_version =	D_VERSION,
283 	.d_flags =	D_NEEDGIANT, /* XXX: does dead_strategy need this ? */
284 	.d_open =	dead_open,
285 	.d_close =	dead_close,
286 	.d_read =	dead_read,
287 	.d_write =	dead_write,
288 	.d_ioctl =	dead_ioctl,
289 	.d_poll =	dead_poll,
290 	.d_mmap =	dead_mmap,
291 	.d_strategy =	dead_strategy,
292 	.d_name =	"dead",
293 	.d_dump =	dead_dump,
294 	.d_kqfilter =	dead_kqfilter,
295 	.d_mmap_single = dead_mmap_single
296 };
297 
298 /* Default methods if driver does not specify method */
299 
300 #define null_open	(d_open_t *)nullop
301 #define null_close	(d_close_t *)nullop
302 #define no_read		(d_read_t *)enodev
303 #define no_write	(d_write_t *)enodev
304 #define no_ioctl	(d_ioctl_t *)enodev
305 #define no_mmap		(d_mmap_t *)enodev
306 #define no_kqfilter	(d_kqfilter_t *)enodev
307 #define no_mmap_single	(d_mmap_single_t *)enodev
308 
309 static void
310 no_strategy(struct bio *bp)
311 {
312 
313 	biofinish(bp, NULL, ENODEV);
314 }
315 
316 static int
317 no_poll(struct cdev *dev __unused, int events, struct thread *td __unused)
318 {
319 
320 	return (poll_no_poll(events));
321 }
322 
323 #define no_dump		(dumper_t *)enodev
324 
325 static int
326 giant_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
327 {
328 	struct cdevsw *dsw;
329 	int retval;
330 
331 	dsw = dev_refthread(dev);
332 	if (dsw == NULL)
333 		return (ENXIO);
334 	mtx_lock(&Giant);
335 	retval = dsw->d_gianttrick->d_open(dev, oflags, devtype, td);
336 	mtx_unlock(&Giant);
337 	dev_relthread(dev);
338 	return (retval);
339 }
340 
341 static int
342 giant_fdopen(struct cdev *dev, int oflags, struct thread *td, struct file *fp)
343 {
344 	struct cdevsw *dsw;
345 	int retval;
346 
347 	dsw = dev_refthread(dev);
348 	if (dsw == NULL)
349 		return (ENXIO);
350 	mtx_lock(&Giant);
351 	retval = dsw->d_gianttrick->d_fdopen(dev, oflags, td, fp);
352 	mtx_unlock(&Giant);
353 	dev_relthread(dev);
354 	return (retval);
355 }
356 
357 static int
358 giant_close(struct cdev *dev, int fflag, int devtype, struct thread *td)
359 {
360 	struct cdevsw *dsw;
361 	int retval;
362 
363 	dsw = dev_refthread(dev);
364 	if (dsw == NULL)
365 		return (ENXIO);
366 	mtx_lock(&Giant);
367 	retval = dsw->d_gianttrick->d_close(dev, fflag, devtype, td);
368 	mtx_unlock(&Giant);
369 	dev_relthread(dev);
370 	return (retval);
371 }
372 
373 static void
374 giant_strategy(struct bio *bp)
375 {
376 	struct cdevsw *dsw;
377 	struct cdev *dev;
378 
379 	dev = bp->bio_dev;
380 	dsw = dev_refthread(dev);
381 	if (dsw == NULL) {
382 		biofinish(bp, NULL, ENXIO);
383 		return;
384 	}
385 	mtx_lock(&Giant);
386 	dsw->d_gianttrick->d_strategy(bp);
387 	mtx_unlock(&Giant);
388 	dev_relthread(dev);
389 }
390 
391 static int
392 giant_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
393 {
394 	struct cdevsw *dsw;
395 	int retval;
396 
397 	dsw = dev_refthread(dev);
398 	if (dsw == NULL)
399 		return (ENXIO);
400 	mtx_lock(&Giant);
401 	retval = dsw->d_gianttrick->d_ioctl(dev, cmd, data, fflag, td);
402 	mtx_unlock(&Giant);
403 	dev_relthread(dev);
404 	return (retval);
405 }
406 
407 static int
408 giant_read(struct cdev *dev, struct uio *uio, int ioflag)
409 {
410 	struct cdevsw *dsw;
411 	int retval;
412 
413 	dsw = dev_refthread(dev);
414 	if (dsw == NULL)
415 		return (ENXIO);
416 	mtx_lock(&Giant);
417 	retval = dsw->d_gianttrick->d_read(dev, uio, ioflag);
418 	mtx_unlock(&Giant);
419 	dev_relthread(dev);
420 	return (retval);
421 }
422 
423 static int
424 giant_write(struct cdev *dev, struct uio *uio, int ioflag)
425 {
426 	struct cdevsw *dsw;
427 	int retval;
428 
429 	dsw = dev_refthread(dev);
430 	if (dsw == NULL)
431 		return (ENXIO);
432 	mtx_lock(&Giant);
433 	retval = dsw->d_gianttrick->d_write(dev, uio, ioflag);
434 	mtx_unlock(&Giant);
435 	dev_relthread(dev);
436 	return (retval);
437 }
438 
439 static int
440 giant_poll(struct cdev *dev, int events, struct thread *td)
441 {
442 	struct cdevsw *dsw;
443 	int retval;
444 
445 	dsw = dev_refthread(dev);
446 	if (dsw == NULL)
447 		return (ENXIO);
448 	mtx_lock(&Giant);
449 	retval = dsw->d_gianttrick->d_poll(dev, events, td);
450 	mtx_unlock(&Giant);
451 	dev_relthread(dev);
452 	return (retval);
453 }
454 
455 static int
456 giant_kqfilter(struct cdev *dev, struct knote *kn)
457 {
458 	struct cdevsw *dsw;
459 	int retval;
460 
461 	dsw = dev_refthread(dev);
462 	if (dsw == NULL)
463 		return (ENXIO);
464 	mtx_lock(&Giant);
465 	retval = dsw->d_gianttrick->d_kqfilter(dev, kn);
466 	mtx_unlock(&Giant);
467 	dev_relthread(dev);
468 	return (retval);
469 }
470 
471 static int
472 giant_mmap(struct cdev *dev, vm_ooffset_t offset, vm_paddr_t *paddr, int nprot,
473     vm_memattr_t *memattr)
474 {
475 	struct cdevsw *dsw;
476 	int retval;
477 
478 	dsw = dev_refthread(dev);
479 	if (dsw == NULL)
480 		return (ENXIO);
481 	mtx_lock(&Giant);
482 	retval = dsw->d_gianttrick->d_mmap(dev, offset, paddr, nprot,
483 	    memattr);
484 	mtx_unlock(&Giant);
485 	dev_relthread(dev);
486 	return (retval);
487 }
488 
489 static int
490 giant_mmap_single(struct cdev *dev, vm_ooffset_t *offset, vm_size_t size,
491     vm_object_t *object, int nprot)
492 {
493 	struct cdevsw *dsw;
494 	int retval;
495 
496 	dsw = dev_refthread(dev);
497 	if (dsw == NULL)
498 		return (ENXIO);
499 	mtx_lock(&Giant);
500 	retval = dsw->d_gianttrick->d_mmap_single(dev, offset, size, object,
501 	    nprot);
502 	mtx_unlock(&Giant);
503 	dev_relthread(dev);
504 	return (retval);
505 }
506 
507 static void
508 notify(struct cdev *dev, const char *ev, int flags)
509 {
510 	static const char prefix[] = "cdev=";
511 	char *data;
512 	int namelen;
513 
514 	if (cold)
515 		return;
516 	namelen = strlen(dev->si_name);
517 	data = malloc(namelen + sizeof(prefix), M_TEMP,
518 	     (flags & MAKEDEV_NOWAIT) ? M_NOWAIT : M_WAITOK);
519 	if (data == NULL)
520 		return;
521 	memcpy(data, prefix, sizeof(prefix) - 1);
522 	memcpy(data + sizeof(prefix) - 1, dev->si_name, namelen + 1);
523 	devctl_notify("DEVFS", "CDEV", ev, data);
524 	free(data, M_TEMP);
525 }
526 
527 static void
528 notify_create(struct cdev *dev, int flags)
529 {
530 
531 	notify(dev, "CREATE", flags);
532 }
533 
534 static void
535 notify_destroy(struct cdev *dev)
536 {
537 
538 	notify(dev, "DESTROY", MAKEDEV_WAITOK);
539 }
540 
541 static struct cdev *
542 newdev(struct cdevsw *csw, int unit, struct cdev *si)
543 {
544 	struct cdev *si2;
545 
546 	mtx_assert(&devmtx, MA_OWNED);
547 	if (csw->d_flags & D_NEEDMINOR) {
548 		/* We may want to return an existing device */
549 		LIST_FOREACH(si2, &csw->d_devs, si_list) {
550 			if (dev2unit(si2) == unit) {
551 				dev_free_devlocked(si);
552 				return (si2);
553 			}
554 		}
555 	}
556 	si->si_drv0 = unit;
557 	si->si_devsw = csw;
558 	LIST_INSERT_HEAD(&csw->d_devs, si, si_list);
559 	return (si);
560 }
561 
562 static void
563 fini_cdevsw(struct cdevsw *devsw)
564 {
565 	struct cdevsw *gt;
566 
567 	if (devsw->d_gianttrick != NULL) {
568 		gt = devsw->d_gianttrick;
569 		memcpy(devsw, gt, sizeof *devsw);
570 		cdevsw_free_devlocked(gt);
571 		devsw->d_gianttrick = NULL;
572 	}
573 	devsw->d_flags &= ~D_INIT;
574 }
575 
576 static int
577 prep_cdevsw(struct cdevsw *devsw, int flags)
578 {
579 	struct cdevsw *dsw2;
580 
581 	mtx_assert(&devmtx, MA_OWNED);
582 	if (devsw->d_flags & D_INIT)
583 		return (1);
584 	if (devsw->d_flags & D_NEEDGIANT) {
585 		dev_unlock();
586 		dsw2 = malloc(sizeof *dsw2, M_DEVT,
587 		     (flags & MAKEDEV_NOWAIT) ? M_NOWAIT : M_WAITOK);
588 		dev_lock();
589 		if (dsw2 == NULL && !(devsw->d_flags & D_INIT))
590 			return (0);
591 	} else
592 		dsw2 = NULL;
593 	if (devsw->d_flags & D_INIT) {
594 		if (dsw2 != NULL)
595 			cdevsw_free_devlocked(dsw2);
596 		return (1);
597 	}
598 
599 	if (devsw->d_version != D_VERSION_03) {
600 		printf(
601 		    "WARNING: Device driver \"%s\" has wrong version %s\n",
602 		    devsw->d_name == NULL ? "???" : devsw->d_name,
603 		    "and is disabled.  Recompile KLD module.");
604 		devsw->d_open = dead_open;
605 		devsw->d_close = dead_close;
606 		devsw->d_read = dead_read;
607 		devsw->d_write = dead_write;
608 		devsw->d_ioctl = dead_ioctl;
609 		devsw->d_poll = dead_poll;
610 		devsw->d_mmap = dead_mmap;
611 		devsw->d_mmap_single = dead_mmap_single;
612 		devsw->d_strategy = dead_strategy;
613 		devsw->d_dump = dead_dump;
614 		devsw->d_kqfilter = dead_kqfilter;
615 	}
616 
617 	if (devsw->d_flags & D_NEEDGIANT) {
618 		if (devsw->d_gianttrick == NULL) {
619 			memcpy(dsw2, devsw, sizeof *dsw2);
620 			devsw->d_gianttrick = dsw2;
621 			dsw2 = NULL;
622 		}
623 	}
624 
625 #define FIXUP(member, noop, giant) 				\
626 	do {							\
627 		if (devsw->member == NULL) {			\
628 			devsw->member = noop;			\
629 		} else if (devsw->d_flags & D_NEEDGIANT)	\
630 			devsw->member = giant;			\
631 		}						\
632 	while (0)
633 
634 	FIXUP(d_open,		null_open,	giant_open);
635 	FIXUP(d_fdopen,		NULL,		giant_fdopen);
636 	FIXUP(d_close,		null_close,	giant_close);
637 	FIXUP(d_read,		no_read,	giant_read);
638 	FIXUP(d_write,		no_write,	giant_write);
639 	FIXUP(d_ioctl,		no_ioctl,	giant_ioctl);
640 	FIXUP(d_poll,		no_poll,	giant_poll);
641 	FIXUP(d_mmap,		no_mmap,	giant_mmap);
642 	FIXUP(d_strategy,	no_strategy,	giant_strategy);
643 	FIXUP(d_kqfilter,	no_kqfilter,	giant_kqfilter);
644 	FIXUP(d_mmap_single,	no_mmap_single,	giant_mmap_single);
645 
646 	if (devsw->d_dump == NULL)	devsw->d_dump = no_dump;
647 
648 	LIST_INIT(&devsw->d_devs);
649 
650 	devsw->d_flags |= D_INIT;
651 
652 	if (dsw2 != NULL)
653 		cdevsw_free_devlocked(dsw2);
654 	return (1);
655 }
656 
657 static struct cdev *
658 make_dev_credv(int flags, struct cdevsw *devsw, int unit,
659     struct ucred *cr, uid_t uid,
660     gid_t gid, int mode, const char *fmt, va_list ap)
661 {
662 	struct cdev *dev;
663 	int i;
664 
665 	dev = devfs_alloc(flags);
666 	if (dev == NULL)
667 		return (NULL);
668 	dev_lock();
669 	if (!prep_cdevsw(devsw, flags)) {
670 		dev_unlock();
671 		devfs_free(dev);
672 		return (NULL);
673 	}
674 	dev = newdev(devsw, unit, dev);
675 	if (flags & MAKEDEV_REF)
676 		dev_refl(dev);
677 	if (dev->si_flags & SI_CHEAPCLONE &&
678 	    dev->si_flags & SI_NAMED) {
679 		/*
680 		 * This is allowed as it removes races and generally
681 		 * simplifies cloning devices.
682 		 * XXX: still ??
683 		 */
684 		dev_unlock_and_free();
685 		return (dev);
686 	}
687 	KASSERT(!(dev->si_flags & SI_NAMED),
688 	    ("make_dev() by driver %s on pre-existing device (min=%x, name=%s)",
689 	    devsw->d_name, dev2unit(dev), devtoname(dev)));
690 
691 	i = vsnrprintf(dev->__si_namebuf, sizeof dev->__si_namebuf, 32, fmt, ap);
692 	if (i > (sizeof dev->__si_namebuf - 1)) {
693 		printf("WARNING: Device name truncated! (%s)\n",
694 		    dev->__si_namebuf);
695 	}
696 
697 	dev->si_flags |= SI_NAMED;
698 	if (cr != NULL)
699 		dev->si_cred = crhold(cr);
700 	dev->si_uid = uid;
701 	dev->si_gid = gid;
702 	dev->si_mode = mode;
703 
704 	devfs_create(dev);
705 	clean_unrhdrl(devfs_inos);
706 	dev_unlock_and_free();
707 
708 	notify_create(dev, flags);
709 
710 	return (dev);
711 }
712 
713 struct cdev *
714 make_dev(struct cdevsw *devsw, int unit, uid_t uid, gid_t gid, int mode,
715     const char *fmt, ...)
716 {
717 	struct cdev *dev;
718 	va_list ap;
719 
720 	va_start(ap, fmt);
721 	dev = make_dev_credv(0, devsw, unit, NULL, uid, gid, mode, fmt, ap);
722 	va_end(ap);
723 	return (dev);
724 }
725 
726 struct cdev *
727 make_dev_cred(struct cdevsw *devsw, int unit, struct ucred *cr, uid_t uid,
728     gid_t gid, int mode, const char *fmt, ...)
729 {
730 	struct cdev *dev;
731 	va_list ap;
732 
733 	va_start(ap, fmt);
734 	dev = make_dev_credv(0, devsw, unit, cr, uid, gid, mode, fmt, ap);
735 	va_end(ap);
736 
737 	return (dev);
738 }
739 
740 struct cdev *
741 make_dev_credf(int flags, struct cdevsw *devsw, int unit,
742     struct ucred *cr, uid_t uid,
743     gid_t gid, int mode, const char *fmt, ...)
744 {
745 	struct cdev *dev;
746 	va_list ap;
747 
748 	va_start(ap, fmt);
749 	dev = make_dev_credv(flags, devsw, unit, cr, uid, gid, mode,
750 	    fmt, ap);
751 	va_end(ap);
752 
753 	return (dev);
754 }
755 
756 static void
757 dev_dependsl(struct cdev *pdev, struct cdev *cdev)
758 {
759 
760 	cdev->si_parent = pdev;
761 	cdev->si_flags |= SI_CHILD;
762 	LIST_INSERT_HEAD(&pdev->si_children, cdev, si_siblings);
763 }
764 
765 
766 void
767 dev_depends(struct cdev *pdev, struct cdev *cdev)
768 {
769 
770 	dev_lock();
771 	dev_dependsl(pdev, cdev);
772 	dev_unlock();
773 }
774 
775 struct cdev *
776 make_dev_alias(struct cdev *pdev, const char *fmt, ...)
777 {
778 	struct cdev *dev;
779 	va_list ap;
780 	int i;
781 
782 	KASSERT(pdev != NULL, ("NULL pdev"));
783 	dev = devfs_alloc(MAKEDEV_WAITOK);
784 	dev_lock();
785 	dev->si_flags |= SI_ALIAS;
786 	dev->si_flags |= SI_NAMED;
787 	va_start(ap, fmt);
788 	i = vsnrprintf(dev->__si_namebuf, sizeof dev->__si_namebuf, 32, fmt, ap);
789 	if (i > (sizeof dev->__si_namebuf - 1)) {
790 		printf("WARNING: Device name truncated! (%s)\n",
791 		    dev->__si_namebuf);
792 	}
793 	va_end(ap);
794 
795 	devfs_create(dev);
796 	dev_dependsl(pdev, dev);
797 	clean_unrhdrl(devfs_inos);
798 	dev_unlock();
799 
800 	notify_create(dev, MAKEDEV_WAITOK);
801 
802 	return (dev);
803 }
804 
805 static void
806 destroy_devl(struct cdev *dev)
807 {
808 	struct cdevsw *csw;
809 	struct cdev_privdata *p, *p1;
810 
811 	mtx_assert(&devmtx, MA_OWNED);
812 	KASSERT(dev->si_flags & SI_NAMED,
813 	    ("WARNING: Driver mistake: destroy_dev on %d\n", dev2unit(dev)));
814 
815 	devfs_destroy(dev);
816 
817 	/* Remove name marking */
818 	dev->si_flags &= ~SI_NAMED;
819 
820 	/* If we are a child, remove us from the parents list */
821 	if (dev->si_flags & SI_CHILD) {
822 		LIST_REMOVE(dev, si_siblings);
823 		dev->si_flags &= ~SI_CHILD;
824 	}
825 
826 	/* Kill our children */
827 	while (!LIST_EMPTY(&dev->si_children))
828 		destroy_devl(LIST_FIRST(&dev->si_children));
829 
830 	/* Remove from clone list */
831 	if (dev->si_flags & SI_CLONELIST) {
832 		LIST_REMOVE(dev, si_clone);
833 		dev->si_flags &= ~SI_CLONELIST;
834 	}
835 
836 	dev->si_refcount++;	/* Avoid race with dev_rel() */
837 	csw = dev->si_devsw;
838 	dev->si_devsw = NULL;	/* already NULL for SI_ALIAS */
839 	while (csw != NULL && csw->d_purge != NULL && dev->si_threadcount) {
840 		csw->d_purge(dev);
841 		msleep(csw, &devmtx, PRIBIO, "devprg", hz/10);
842 		if (dev->si_threadcount)
843 			printf("Still %lu threads in %s\n",
844 			    dev->si_threadcount, devtoname(dev));
845 	}
846 	while (dev->si_threadcount != 0) {
847 		/* Use unique dummy wait ident */
848 		msleep(&csw, &devmtx, PRIBIO, "devdrn", hz / 10);
849 	}
850 
851 	dev_unlock();
852 	notify_destroy(dev);
853 	mtx_lock(&cdevpriv_mtx);
854 	LIST_FOREACH_SAFE(p, &cdev2priv(dev)->cdp_fdpriv, cdpd_list, p1) {
855 		devfs_destroy_cdevpriv(p);
856 		mtx_lock(&cdevpriv_mtx);
857 	}
858 	mtx_unlock(&cdevpriv_mtx);
859 	dev_lock();
860 
861 	dev->si_drv1 = 0;
862 	dev->si_drv2 = 0;
863 	bzero(&dev->__si_u, sizeof(dev->__si_u));
864 
865 	if (!(dev->si_flags & SI_ALIAS)) {
866 		/* Remove from cdevsw list */
867 		LIST_REMOVE(dev, si_list);
868 
869 		/* If cdevsw has no more struct cdev *'s, clean it */
870 		if (LIST_EMPTY(&csw->d_devs)) {
871 			fini_cdevsw(csw);
872 			wakeup(&csw->d_devs);
873 		}
874 	}
875 	dev->si_flags &= ~SI_ALIAS;
876 	dev->si_refcount--;	/* Avoid race with dev_rel() */
877 
878 	if (dev->si_refcount > 0) {
879 		LIST_INSERT_HEAD(&dead_cdevsw.d_devs, dev, si_list);
880 	} else {
881 		dev_free_devlocked(dev);
882 	}
883 }
884 
885 void
886 destroy_dev(struct cdev *dev)
887 {
888 
889 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "destroy_dev");
890 	dev_lock();
891 	destroy_devl(dev);
892 	dev_unlock_and_free();
893 }
894 
895 const char *
896 devtoname(struct cdev *dev)
897 {
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,
960     struct cdev **dp, int extra)
961 {
962 	struct clonedevs *cd;
963 	struct cdev *dev, *ndev, *dl, *de;
964 	int unit, low, u;
965 
966 	KASSERT(*cdp != NULL,
967 	    ("clone_setup() not called in driver \"%s\"", csw->d_name));
968 	KASSERT(!(extra & CLONE_UNITMASK),
969 	    ("Illegal extra bits (0x%x) in clone_create", extra));
970 	KASSERT(*up <= CLONE_UNITMASK,
971 	    ("Too high unit (0x%x) in clone_create", *up));
972 	KASSERT(csw->d_flags & D_NEEDMINOR,
973 	    ("clone_create() on cdevsw without minor numbers"));
974 
975 
976 	/*
977 	 * Search the list for a lot of things in one go:
978 	 *   A preexisting match is returned immediately.
979 	 *   The lowest free unit number if we are passed -1, and the place
980 	 *	 in the list where we should insert that new element.
981 	 *   The place to insert a specified unit number, if applicable
982 	 *       the end of the list.
983 	 */
984 	unit = *up;
985 	ndev = devfs_alloc(MAKEDEV_WAITOK);
986 	dev_lock();
987 	prep_cdevsw(csw, MAKEDEV_WAITOK);
988 	low = extra;
989 	de = dl = NULL;
990 	cd = *cdp;
991 	LIST_FOREACH(dev, &cd->head, si_clone) {
992 		KASSERT(dev->si_flags & SI_CLONELIST,
993 		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
994 		u = dev2unit(dev);
995 		if (u == (unit | extra)) {
996 			*dp = dev;
997 			dev_unlock();
998 			devfs_free(ndev);
999 			return (0);
1000 		}
1001 		if (unit == -1 && u == low) {
1002 			low++;
1003 			de = dev;
1004 			continue;
1005 		} else if (u < (unit | extra)) {
1006 			de = dev;
1007 			continue;
1008 		} else if (u > (unit | extra)) {
1009 			dl = dev;
1010 			break;
1011 		}
1012 	}
1013 	if (unit == -1)
1014 		unit = low & CLONE_UNITMASK;
1015 	dev = newdev(csw, unit | extra, ndev);
1016 	if (dev->si_flags & SI_CLONELIST) {
1017 		printf("dev %p (%s) is on clonelist\n", dev, dev->si_name);
1018 		printf("unit=%d, low=%d, extra=0x%x\n", unit, low, extra);
1019 		LIST_FOREACH(dev, &cd->head, si_clone) {
1020 			printf("\t%p %s\n", dev, dev->si_name);
1021 		}
1022 		panic("foo");
1023 	}
1024 	KASSERT(!(dev->si_flags & SI_CLONELIST),
1025 	    ("Dev %p(%s) should not be on clonelist", dev, dev->si_name));
1026 	if (dl != NULL)
1027 		LIST_INSERT_BEFORE(dl, dev, si_clone);
1028 	else if (de != NULL)
1029 		LIST_INSERT_AFTER(de, dev, si_clone);
1030 	else
1031 		LIST_INSERT_HEAD(&cd->head, dev, si_clone);
1032 	dev->si_flags |= SI_CLONELIST;
1033 	*up = unit;
1034 	dev_unlock_and_free();
1035 	return (1);
1036 }
1037 
1038 /*
1039  * Kill everything still on the list.  The driver should already have
1040  * disposed of any softc hung of the struct cdev *'s at this time.
1041  */
1042 void
1043 clone_cleanup(struct clonedevs **cdp)
1044 {
1045 	struct cdev *dev;
1046 	struct cdev_priv *cp;
1047 	struct clonedevs *cd;
1048 
1049 	cd = *cdp;
1050 	if (cd == NULL)
1051 		return;
1052 	dev_lock();
1053 	while (!LIST_EMPTY(&cd->head)) {
1054 		dev = LIST_FIRST(&cd->head);
1055 		LIST_REMOVE(dev, si_clone);
1056 		KASSERT(dev->si_flags & SI_CLONELIST,
1057 		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
1058 		dev->si_flags &= ~SI_CLONELIST;
1059 		cp = cdev2priv(dev);
1060 		if (!(cp->cdp_flags & CDP_SCHED_DTR)) {
1061 			cp->cdp_flags |= CDP_SCHED_DTR;
1062 			KASSERT(dev->si_flags & SI_NAMED,
1063 				("Driver has goofed in cloning underways udev %x unit %x", dev2udev(dev), dev2unit(dev)));
1064 			destroy_devl(dev);
1065 		}
1066 	}
1067 	dev_unlock_and_free();
1068 	free(cd, M_DEVBUF);
1069 	*cdp = NULL;
1070 }
1071 
1072 static TAILQ_HEAD(, cdev_priv) dev_ddtr =
1073 	TAILQ_HEAD_INITIALIZER(dev_ddtr);
1074 static struct task dev_dtr_task;
1075 
1076 static void
1077 destroy_dev_tq(void *ctx, int pending)
1078 {
1079 	struct cdev_priv *cp;
1080 	struct cdev *dev;
1081 	void (*cb)(void *);
1082 	void *cb_arg;
1083 
1084 	dev_lock();
1085 	while (!TAILQ_EMPTY(&dev_ddtr)) {
1086 		cp = TAILQ_FIRST(&dev_ddtr);
1087 		dev = &cp->cdp_c;
1088 		KASSERT(cp->cdp_flags & CDP_SCHED_DTR,
1089 		    ("cdev %p in dev_destroy_tq without CDP_SCHED_DTR", cp));
1090 		TAILQ_REMOVE(&dev_ddtr, cp, cdp_dtr_list);
1091 		cb = cp->cdp_dtr_cb;
1092 		cb_arg = cp->cdp_dtr_cb_arg;
1093 		destroy_devl(dev);
1094 		dev_unlock_and_free();
1095 		dev_rel(dev);
1096 		if (cb != NULL)
1097 			cb(cb_arg);
1098 		dev_lock();
1099 	}
1100 	dev_unlock();
1101 }
1102 
1103 /*
1104  * devmtx shall be locked on entry. devmtx will be unlocked after
1105  * function return.
1106  */
1107 static int
1108 destroy_dev_sched_cbl(struct cdev *dev, void (*cb)(void *), void *arg)
1109 {
1110 	struct cdev_priv *cp;
1111 
1112 	mtx_assert(&devmtx, MA_OWNED);
1113 	cp = cdev2priv(dev);
1114 	if (cp->cdp_flags & CDP_SCHED_DTR) {
1115 		dev_unlock();
1116 		return (0);
1117 	}
1118 	dev_refl(dev);
1119 	cp->cdp_flags |= CDP_SCHED_DTR;
1120 	cp->cdp_dtr_cb = cb;
1121 	cp->cdp_dtr_cb_arg = arg;
1122 	TAILQ_INSERT_TAIL(&dev_ddtr, cp, cdp_dtr_list);
1123 	dev_unlock();
1124 	taskqueue_enqueue(taskqueue_swi_giant, &dev_dtr_task);
1125 	return (1);
1126 }
1127 
1128 int
1129 destroy_dev_sched_cb(struct cdev *dev, void (*cb)(void *), void *arg)
1130 {
1131 
1132 	dev_lock();
1133 	return (destroy_dev_sched_cbl(dev, cb, arg));
1134 }
1135 
1136 int
1137 destroy_dev_sched(struct cdev *dev)
1138 {
1139 
1140 	return (destroy_dev_sched_cb(dev, NULL, NULL));
1141 }
1142 
1143 void
1144 destroy_dev_drain(struct cdevsw *csw)
1145 {
1146 
1147 	dev_lock();
1148 	while (!LIST_EMPTY(&csw->d_devs)) {
1149 		msleep(&csw->d_devs, &devmtx, PRIBIO, "devscd", hz/10);
1150 	}
1151 	dev_unlock();
1152 }
1153 
1154 void
1155 drain_dev_clone_events(void)
1156 {
1157 
1158 	sx_xlock(&clone_drain_lock);
1159 	sx_xunlock(&clone_drain_lock);
1160 }
1161 
1162 static void
1163 devdtr_init(void *dummy __unused)
1164 {
1165 
1166 	TASK_INIT(&dev_dtr_task, 0, destroy_dev_tq, NULL);
1167 }
1168 
1169 SYSINIT(devdtr, SI_SUB_DEVFS, SI_ORDER_SECOND, devdtr_init, NULL);
1170