xref: /freebsd/sys/kern/kern_conf.c (revision cacdd70cc751fb68dec4b86c5e5b8c969b6e26ef)
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/tty.h>
46 #include <sys/ucred.h>
47 #include <sys/taskqueue.h>
48 #include <machine/stdarg.h>
49 
50 #include <fs/devfs/devfs_int.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 minornr,
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();
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 
280 static struct cdevsw dead_cdevsw = {
281 	.d_version =	D_VERSION,
282 	.d_flags =	D_NEEDGIANT, /* XXX: does dead_strategy need this ? */
283 	.d_open =	dead_open,
284 	.d_close =	dead_close,
285 	.d_read =	dead_read,
286 	.d_write =	dead_write,
287 	.d_ioctl =	dead_ioctl,
288 	.d_poll =	dead_poll,
289 	.d_mmap =	dead_mmap,
290 	.d_strategy =	dead_strategy,
291 	.d_name =	"dead",
292 	.d_dump =	dead_dump,
293 	.d_kqfilter =	dead_kqfilter
294 };
295 
296 /* Default methods if driver does not specify method */
297 
298 #define null_open	(d_open_t *)nullop
299 #define null_close	(d_close_t *)nullop
300 #define no_read		(d_read_t *)enodev
301 #define no_write	(d_write_t *)enodev
302 #define no_ioctl	(d_ioctl_t *)enodev
303 #define no_mmap		(d_mmap_t *)enodev
304 #define no_kqfilter	(d_kqfilter_t *)enodev
305 
306 static void
307 no_strategy(struct bio *bp)
308 {
309 
310 	biofinish(bp, NULL, ENODEV);
311 }
312 
313 static int
314 no_poll(struct cdev *dev __unused, int events, struct thread *td __unused)
315 {
316 	/*
317 	 * Return true for read/write.  If the user asked for something
318 	 * special, return POLLNVAL, so that clients have a way of
319 	 * determining reliably whether or not the extended
320 	 * functionality is present without hard-coding knowledge
321 	 * of specific filesystem implementations.
322 	 * Stay in sync with vop_nopoll().
323 	 */
324 	if (events & ~POLLSTANDARD)
325 		return (POLLNVAL);
326 
327 	return (events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
328 }
329 
330 #define no_dump		(dumper_t *)enodev
331 
332 static int
333 giant_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
334 {
335 	struct cdevsw *dsw;
336 	int retval;
337 
338 	dsw = dev_refthread(dev);
339 	if (dsw == NULL)
340 		return (ENXIO);
341 	mtx_lock(&Giant);
342 	retval = dsw->d_gianttrick->d_open(dev, oflags, devtype, td);
343 	mtx_unlock(&Giant);
344 	dev_relthread(dev);
345 	return (retval);
346 }
347 
348 static int
349 giant_fdopen(struct cdev *dev, int oflags, struct thread *td, struct file *fp)
350 {
351 	struct cdevsw *dsw;
352 	int retval;
353 
354 	dsw = dev_refthread(dev);
355 	if (dsw == NULL)
356 		return (ENXIO);
357 	mtx_lock(&Giant);
358 	retval = dsw->d_gianttrick->d_fdopen(dev, oflags, td, fp);
359 	mtx_unlock(&Giant);
360 	dev_relthread(dev);
361 	return (retval);
362 }
363 
364 static int
365 giant_close(struct cdev *dev, int fflag, int devtype, struct thread *td)
366 {
367 	struct cdevsw *dsw;
368 	int retval;
369 
370 	dsw = dev_refthread(dev);
371 	if (dsw == NULL)
372 		return (ENXIO);
373 	mtx_lock(&Giant);
374 	retval = dsw->d_gianttrick->d_close(dev, fflag, devtype, td);
375 	mtx_unlock(&Giant);
376 	dev_relthread(dev);
377 	return (retval);
378 }
379 
380 static void
381 giant_strategy(struct bio *bp)
382 {
383 	struct cdevsw *dsw;
384 	struct cdev *dev;
385 
386 	dev = bp->bio_dev;
387 	dsw = dev_refthread(dev);
388 	if (dsw == NULL) {
389 		biofinish(bp, NULL, ENXIO);
390 		return;
391 	}
392 	mtx_lock(&Giant);
393 	dsw->d_gianttrick->d_strategy(bp);
394 	mtx_unlock(&Giant);
395 	dev_relthread(dev);
396 }
397 
398 static int
399 giant_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
400 {
401 	struct cdevsw *dsw;
402 	int retval;
403 
404 	dsw = dev_refthread(dev);
405 	if (dsw == NULL)
406 		return (ENXIO);
407 	mtx_lock(&Giant);
408 	retval = dsw->d_gianttrick->d_ioctl(dev, cmd, data, fflag, td);
409 	mtx_unlock(&Giant);
410 	dev_relthread(dev);
411 	return (retval);
412 }
413 
414 static int
415 giant_read(struct cdev *dev, struct uio *uio, int ioflag)
416 {
417 	struct cdevsw *dsw;
418 	int retval;
419 
420 	dsw = dev_refthread(dev);
421 	if (dsw == NULL)
422 		return (ENXIO);
423 	mtx_lock(&Giant);
424 	retval = dsw->d_gianttrick->d_read(dev, uio, ioflag);
425 	mtx_unlock(&Giant);
426 	dev_relthread(dev);
427 	return (retval);
428 }
429 
430 static int
431 giant_write(struct cdev *dev, struct uio *uio, int ioflag)
432 {
433 	struct cdevsw *dsw;
434 	int retval;
435 
436 	dsw = dev_refthread(dev);
437 	if (dsw == NULL)
438 		return (ENXIO);
439 	mtx_lock(&Giant);
440 	retval = dsw->d_gianttrick->d_write(dev, uio, ioflag);
441 	mtx_unlock(&Giant);
442 	dev_relthread(dev);
443 	return (retval);
444 }
445 
446 static int
447 giant_poll(struct cdev *dev, int events, struct thread *td)
448 {
449 	struct cdevsw *dsw;
450 	int retval;
451 
452 	dsw = dev_refthread(dev);
453 	if (dsw == NULL)
454 		return (ENXIO);
455 	mtx_lock(&Giant);
456 	retval = dsw->d_gianttrick->d_poll(dev, events, td);
457 	mtx_unlock(&Giant);
458 	dev_relthread(dev);
459 	return (retval);
460 }
461 
462 static int
463 giant_kqfilter(struct cdev *dev, struct knote *kn)
464 {
465 	struct cdevsw *dsw;
466 	int retval;
467 
468 	dsw = dev_refthread(dev);
469 	if (dsw == NULL)
470 		return (ENXIO);
471 	mtx_lock(&Giant);
472 	retval = dsw->d_gianttrick->d_kqfilter(dev, kn);
473 	mtx_unlock(&Giant);
474 	dev_relthread(dev);
475 	return (retval);
476 }
477 
478 static int
479 giant_mmap(struct cdev *dev, vm_offset_t offset, vm_paddr_t *paddr, int nprot)
480 {
481 	struct cdevsw *dsw;
482 	int retval;
483 
484 	dsw = dev_refthread(dev);
485 	if (dsw == NULL)
486 		return (ENXIO);
487 	mtx_lock(&Giant);
488 	retval = dsw->d_gianttrick->d_mmap(dev, offset, paddr, nprot);
489 	mtx_unlock(&Giant);
490 	dev_relthread(dev);
491 	return (retval);
492 }
493 
494 
495 static void
496 notify(struct cdev *dev, const char *ev)
497 {
498 	static const char prefix[] = "cdev=";
499 	char *data;
500 	int namelen;
501 
502 	if (cold)
503 		return;
504 	namelen = strlen(dev->si_name);
505 	data = malloc(namelen + sizeof(prefix), M_TEMP, M_WAITOK);
506 	memcpy(data, prefix, sizeof(prefix) - 1);
507 	memcpy(data + sizeof(prefix) - 1, dev->si_name, namelen + 1);
508 	devctl_notify("DEVFS", "CDEV", ev, data);
509 	free(data, M_TEMP);
510 }
511 
512 static void
513 notify_create(struct cdev *dev)
514 {
515 
516 	notify(dev, "CREATE");
517 }
518 
519 static void
520 notify_destroy(struct cdev *dev)
521 {
522 
523 	notify(dev, "DESTROY");
524 }
525 
526 static struct cdev *
527 newdev(struct cdevsw *csw, int y, struct cdev *si)
528 {
529 	struct cdev *si2;
530 	dev_t	udev;
531 
532 	mtx_assert(&devmtx, MA_OWNED);
533 	udev = y;
534 	if (csw->d_flags & D_NEEDMINOR) {
535 		/* We may want to return an existing device */
536 		LIST_FOREACH(si2, &csw->d_devs, si_list) {
537 			if (si2->si_drv0 == udev) {
538 				dev_free_devlocked(si);
539 				return (si2);
540 			}
541 		}
542 	}
543 	si->si_drv0 = udev;
544 	si->si_devsw = csw;
545 	LIST_INSERT_HEAD(&csw->d_devs, si, si_list);
546 	return (si);
547 }
548 
549 #define UMINORMASK	0xffff00ffU
550 
551 int
552 uminor(dev_t dev)
553 {
554 	return (dev & UMINORMASK);
555 }
556 
557 int
558 umajor(dev_t dev)
559 {
560 	return ((dev & ~UMINORMASK) >> 8);
561 }
562 
563 static void
564 fini_cdevsw(struct cdevsw *devsw)
565 {
566 	struct cdevsw *gt;
567 
568 	if (devsw->d_gianttrick != NULL) {
569 		gt = devsw->d_gianttrick;
570 		memcpy(devsw, gt, sizeof *devsw);
571 		cdevsw_free_devlocked(gt);
572 		devsw->d_gianttrick = NULL;
573 	}
574 	devsw->d_flags &= ~D_INIT;
575 }
576 
577 static void
578 prep_cdevsw(struct cdevsw *devsw)
579 {
580 	struct cdevsw *dsw2;
581 
582 	mtx_assert(&devmtx, MA_OWNED);
583 	if (devsw->d_flags & D_INIT)
584 		return;
585 	if (devsw->d_flags & D_NEEDGIANT) {
586 		dev_unlock();
587 		dsw2 = malloc(sizeof *dsw2, M_DEVT, M_WAITOK);
588 		dev_lock();
589 	} else
590 		dsw2 = NULL;
591 	if (devsw->d_flags & D_INIT) {
592 		if (dsw2 != NULL)
593 			cdevsw_free_devlocked(dsw2);
594 		return;
595 	}
596 
597 	if (devsw->d_version != D_VERSION_01) {
598 		printf(
599 		    "WARNING: Device driver \"%s\" has wrong version %s\n",
600 		    devsw->d_name == NULL ? "???" : devsw->d_name,
601 		    "and is disabled.  Recompile KLD module.");
602 		devsw->d_open = dead_open;
603 		devsw->d_close = dead_close;
604 		devsw->d_read = dead_read;
605 		devsw->d_write = dead_write;
606 		devsw->d_ioctl = dead_ioctl;
607 		devsw->d_poll = dead_poll;
608 		devsw->d_mmap = dead_mmap;
609 		devsw->d_strategy = dead_strategy;
610 		devsw->d_dump = dead_dump;
611 		devsw->d_kqfilter = dead_kqfilter;
612 	}
613 
614 	if (devsw->d_flags & D_TTY) {
615 		if (devsw->d_ioctl == NULL)	devsw->d_ioctl = ttyioctl;
616 		if (devsw->d_read == NULL)	devsw->d_read = ttyread;
617 		if (devsw->d_write == NULL)	devsw->d_write = ttywrite;
618 		if (devsw->d_kqfilter == NULL)	devsw->d_kqfilter = ttykqfilter;
619 		if (devsw->d_poll == NULL)	devsw->d_poll = ttypoll;
620 	}
621 
622 	if (devsw->d_flags & D_NEEDGIANT) {
623 		if (devsw->d_gianttrick == NULL) {
624 			memcpy(dsw2, devsw, sizeof *dsw2);
625 			devsw->d_gianttrick = dsw2;
626 			dsw2 = NULL;
627 		}
628 	}
629 
630 #define FIXUP(member, noop, giant) 				\
631 	do {							\
632 		if (devsw->member == NULL) {			\
633 			devsw->member = noop;			\
634 		} else if (devsw->d_flags & D_NEEDGIANT)	\
635 			devsw->member = giant;			\
636 		}						\
637 	while (0)
638 
639 	FIXUP(d_open,		null_open,	giant_open);
640 	FIXUP(d_fdopen,		NULL,		giant_fdopen);
641 	FIXUP(d_close,		null_close,	giant_close);
642 	FIXUP(d_read,		no_read,	giant_read);
643 	FIXUP(d_write,		no_write,	giant_write);
644 	FIXUP(d_ioctl,		no_ioctl,	giant_ioctl);
645 	FIXUP(d_poll,		no_poll,	giant_poll);
646 	FIXUP(d_mmap,		no_mmap,	giant_mmap);
647 	FIXUP(d_strategy,	no_strategy,	giant_strategy);
648 	FIXUP(d_kqfilter,	no_kqfilter,	giant_kqfilter);
649 
650 	if (devsw->d_dump == NULL)	devsw->d_dump = no_dump;
651 
652 	LIST_INIT(&devsw->d_devs);
653 
654 	devsw->d_flags |= D_INIT;
655 
656 	if (dsw2 != NULL)
657 		cdevsw_free_devlocked(dsw2);
658 }
659 
660 struct cdev *
661 make_dev_credv(int flags, struct cdevsw *devsw, int minornr,
662     struct ucred *cr, uid_t uid,
663     gid_t gid, int mode, const char *fmt, va_list ap)
664 {
665 	struct cdev *dev;
666 	int i;
667 
668 	dev = devfs_alloc();
669 	dev_lock();
670 	prep_cdevsw(devsw);
671 	dev = newdev(devsw, minornr, dev);
672 	if (flags & MAKEDEV_REF)
673 		dev_refl(dev);
674 	if (dev->si_flags & SI_CHEAPCLONE &&
675 	    dev->si_flags & SI_NAMED) {
676 		/*
677 		 * This is allowed as it removes races and generally
678 		 * simplifies cloning devices.
679 		 * XXX: still ??
680 		 */
681 		dev_unlock_and_free();
682 		return (dev);
683 	}
684 	KASSERT(!(dev->si_flags & SI_NAMED),
685 	    ("make_dev() by driver %s on pre-existing device (min=%x, name=%s)",
686 	    devsw->d_name, minor(dev), devtoname(dev)));
687 
688 	i = vsnrprintf(dev->__si_namebuf, sizeof dev->__si_namebuf, 32, fmt, ap);
689 	if (i > (sizeof dev->__si_namebuf - 1)) {
690 		printf("WARNING: Device name truncated! (%s)\n",
691 		    dev->__si_namebuf);
692 	}
693 
694 	dev->si_flags |= SI_NAMED;
695 #ifdef MAC
696 	if (cr != NULL)
697 		dev->si_cred = crhold(cr);
698 	else
699 #endif
700 		dev->si_cred = NULL;
701 	dev->si_uid = uid;
702 	dev->si_gid = gid;
703 	dev->si_mode = mode;
704 
705 	devfs_create(dev);
706 	clean_unrhdrl(devfs_inos);
707 	dev_unlock_and_free();
708 
709 	notify_create(dev);
710 
711 	return (dev);
712 }
713 
714 struct cdev *
715 make_dev(struct cdevsw *devsw, int minornr, uid_t uid, gid_t gid, int mode,
716     const char *fmt, ...)
717 {
718 	struct cdev *dev;
719 	va_list ap;
720 
721 	va_start(ap, fmt);
722 	dev = make_dev_credv(0, devsw, minornr, NULL, uid, gid, mode, fmt, ap);
723 	va_end(ap);
724 	return (dev);
725 }
726 
727 struct cdev *
728 make_dev_cred(struct cdevsw *devsw, int minornr, struct ucred *cr, uid_t uid,
729     gid_t gid, int mode, const char *fmt, ...)
730 {
731 	struct cdev *dev;
732 	va_list ap;
733 
734 	va_start(ap, fmt);
735 	dev = make_dev_credv(0, devsw, minornr, cr, uid, gid, mode, fmt, ap);
736 	va_end(ap);
737 
738 	return (dev);
739 }
740 
741 struct cdev *
742 make_dev_credf(int flags, struct cdevsw *devsw, int minornr,
743     struct ucred *cr, uid_t uid,
744     gid_t gid, int mode, const char *fmt, ...)
745 {
746 	struct cdev *dev;
747 	va_list ap;
748 
749 	va_start(ap, fmt);
750 	dev = make_dev_credv(flags, devsw, minornr, cr, uid, gid, mode,
751 	    fmt, ap);
752 	va_end(ap);
753 
754 	return (dev);
755 }
756 
757 static void
758 dev_dependsl(struct cdev *pdev, struct cdev *cdev)
759 {
760 
761 	cdev->si_parent = pdev;
762 	cdev->si_flags |= SI_CHILD;
763 	LIST_INSERT_HEAD(&pdev->si_children, cdev, si_siblings);
764 }
765 
766 
767 void
768 dev_depends(struct cdev *pdev, struct cdev *cdev)
769 {
770 
771 	dev_lock();
772 	dev_dependsl(pdev, cdev);
773 	dev_unlock();
774 }
775 
776 struct cdev *
777 make_dev_alias(struct cdev *pdev, const char *fmt, ...)
778 {
779 	struct cdev *dev;
780 	va_list ap;
781 	int i;
782 
783 	KASSERT(pdev != NULL, ("NULL pdev"));
784 	dev = devfs_alloc();
785 	dev_lock();
786 	dev->si_flags |= SI_ALIAS;
787 	dev->si_flags |= SI_NAMED;
788 	va_start(ap, fmt);
789 	i = vsnrprintf(dev->__si_namebuf, sizeof dev->__si_namebuf, 32, fmt, ap);
790 	if (i > (sizeof dev->__si_namebuf - 1)) {
791 		printf("WARNING: Device name truncated! (%s)\n",
792 		    dev->__si_namebuf);
793 	}
794 	va_end(ap);
795 
796 	devfs_create(dev);
797 	dev_dependsl(pdev, dev);
798 	clean_unrhdrl(devfs_inos);
799 	dev_unlock();
800 
801 	notify_create(dev);
802 
803 	return (dev);
804 }
805 
806 static void
807 destroy_devl(struct cdev *dev)
808 {
809 	struct cdevsw *csw;
810 	struct cdev_privdata *p, *p1;
811 
812 	mtx_assert(&devmtx, MA_OWNED);
813 	KASSERT(dev->si_flags & SI_NAMED,
814 	    ("WARNING: Driver mistake: destroy_dev on %d\n", minor(dev)));
815 
816 	devfs_destroy(dev);
817 
818 	/* Remove name marking */
819 	dev->si_flags &= ~SI_NAMED;
820 
821 	/* If we are a child, remove us from the parents list */
822 	if (dev->si_flags & SI_CHILD) {
823 		LIST_REMOVE(dev, si_siblings);
824 		dev->si_flags &= ~SI_CHILD;
825 	}
826 
827 	/* Kill our children */
828 	while (!LIST_EMPTY(&dev->si_children))
829 		destroy_devl(LIST_FIRST(&dev->si_children));
830 
831 	/* Remove from clone list */
832 	if (dev->si_flags & SI_CLONELIST) {
833 		LIST_REMOVE(dev, si_clone);
834 		dev->si_flags &= ~SI_CLONELIST;
835 	}
836 
837 	dev->si_refcount++;	/* Avoid race with dev_rel() */
838 	csw = dev->si_devsw;
839 	dev->si_devsw = NULL;	/* already NULL for SI_ALIAS */
840 	while (csw != NULL && csw->d_purge != NULL && dev->si_threadcount) {
841 		csw->d_purge(dev);
842 		msleep(csw, &devmtx, PRIBIO, "devprg", hz/10);
843 		if (dev->si_threadcount)
844 			printf("Still %lu threads in %s\n",
845 			    dev->si_threadcount, devtoname(dev));
846 	}
847 	while (dev->si_threadcount != 0) {
848 		/* Use unique dummy wait ident */
849 		msleep(&csw, &devmtx, PRIBIO, "devdrn", hz / 10);
850 	}
851 
852 	dev_unlock();
853 	notify_destroy(dev);
854 	mtx_lock(&cdevpriv_mtx);
855 	LIST_FOREACH_SAFE(p, &cdev2priv(dev)->cdp_fdpriv, cdpd_list, p1) {
856 		devfs_destroy_cdevpriv(p);
857 		mtx_lock(&cdevpriv_mtx);
858 	}
859 	mtx_unlock(&cdevpriv_mtx);
860 	dev_lock();
861 
862 	dev->si_drv1 = 0;
863 	dev->si_drv2 = 0;
864 	bzero(&dev->__si_u, sizeof(dev->__si_u));
865 
866 	if (!(dev->si_flags & SI_ALIAS)) {
867 		/* Remove from cdevsw list */
868 		LIST_REMOVE(dev, si_list);
869 
870 		/* If cdevsw has no more struct cdev *'s, clean it */
871 		if (LIST_EMPTY(&csw->d_devs)) {
872 			fini_cdevsw(csw);
873 			wakeup(&csw->d_devs);
874 		}
875 	}
876 	dev->si_flags &= ~SI_ALIAS;
877 	dev->si_refcount--;	/* Avoid race with dev_rel() */
878 
879 	if (dev->si_refcount > 0) {
880 		LIST_INSERT_HEAD(&dead_cdevsw.d_devs, dev, si_list);
881 	} else {
882 		dev_free_devlocked(dev);
883 	}
884 }
885 
886 void
887 destroy_dev(struct cdev *dev)
888 {
889 
890 	dev_lock();
891 	destroy_devl(dev);
892 	dev_unlock_and_free();
893 }
894 
895 const char *
896 devtoname(struct cdev *dev)
897 {
898 	char *p;
899 	struct cdevsw *csw;
900 	int mynor;
901 
902 	if (dev->si_name[0] == '#' || dev->si_name[0] == '\0') {
903 		p = dev->si_name;
904 		csw = dev_refthread(dev);
905 		if (csw != NULL) {
906 			sprintf(p, "(%s)", csw->d_name);
907 			dev_relthread(dev);
908 		}
909 		p += strlen(p);
910 		mynor = minor(dev);
911 		if (mynor < 0 || mynor > 255)
912 			sprintf(p, "/%#x", (u_int)mynor);
913 		else
914 			sprintf(p, "/%d", mynor);
915 	}
916 	return (dev->si_name);
917 }
918 
919 int
920 dev_stdclone(char *name, char **namep, const char *stem, int *unit)
921 {
922 	int u, i;
923 
924 	i = strlen(stem);
925 	if (bcmp(stem, name, i) != 0)
926 		return (0);
927 	if (!isdigit(name[i]))
928 		return (0);
929 	u = 0;
930 	if (name[i] == '0' && isdigit(name[i+1]))
931 		return (0);
932 	while (isdigit(name[i])) {
933 		u *= 10;
934 		u += name[i++] - '0';
935 	}
936 	if (u > 0xffffff)
937 		return (0);
938 	*unit = u;
939 	if (namep)
940 		*namep = &name[i];
941 	if (name[i])
942 		return (2);
943 	return (1);
944 }
945 
946 /*
947  * Helper functions for cloning device drivers.
948  *
949  * The objective here is to make it unnecessary for the device drivers to
950  * use rman or similar to manage their unit number space.  Due to the way
951  * we do "on-demand" devices, using rman or other "private" methods
952  * will be very tricky to lock down properly once we lock down this file.
953  *
954  * Instead we give the drivers these routines which puts the struct cdev *'s
955  * that are to be managed on their own list, and gives the driver the ability
956  * to ask for the first free unit number or a given specified unit number.
957  *
958  * In addition these routines support paired devices (pty, nmdm and similar)
959  * by respecting a number of "flag" bits in the minor number.
960  *
961  */
962 
963 struct clonedevs {
964 	LIST_HEAD(,cdev)	head;
965 };
966 
967 void
968 clone_setup(struct clonedevs **cdp)
969 {
970 
971 	*cdp = malloc(sizeof **cdp, M_DEVBUF, M_WAITOK | M_ZERO);
972 	LIST_INIT(&(*cdp)->head);
973 }
974 
975 int
976 clone_create(struct clonedevs **cdp, struct cdevsw *csw, int *up, struct cdev **dp, int extra)
977 {
978 	struct clonedevs *cd;
979 	struct cdev *dev, *ndev, *dl, *de;
980 	int unit, low, u;
981 
982 	KASSERT(*cdp != NULL,
983 	    ("clone_setup() not called in driver \"%s\"", csw->d_name));
984 	KASSERT(!(extra & CLONE_UNITMASK),
985 	    ("Illegal extra bits (0x%x) in clone_create", extra));
986 	KASSERT(*up <= CLONE_UNITMASK,
987 	    ("Too high unit (0x%x) in clone_create", *up));
988 	KASSERT(csw->d_flags & D_NEEDMINOR,
989 	    ("clone_create() on cdevsw without minor numbers"));
990 
991 
992 	/*
993 	 * Search the list for a lot of things in one go:
994 	 *   A preexisting match is returned immediately.
995 	 *   The lowest free unit number if we are passed -1, and the place
996 	 *	 in the list where we should insert that new element.
997 	 *   The place to insert a specified unit number, if applicable
998 	 *       the end of the list.
999 	 */
1000 	unit = *up;
1001 	ndev = devfs_alloc();
1002 	dev_lock();
1003 	prep_cdevsw(csw);
1004 	low = extra;
1005 	de = dl = NULL;
1006 	cd = *cdp;
1007 	LIST_FOREACH(dev, &cd->head, si_clone) {
1008 		KASSERT(dev->si_flags & SI_CLONELIST,
1009 		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
1010 		u = dev2unit(dev);
1011 		if (u == (unit | extra)) {
1012 			*dp = dev;
1013 			dev_unlock();
1014 			devfs_free(ndev);
1015 			return (0);
1016 		}
1017 		if (unit == -1 && u == low) {
1018 			low++;
1019 			de = dev;
1020 			continue;
1021 		} else if (u < (unit | extra)) {
1022 			de = dev;
1023 			continue;
1024 		} else if (u > (unit | extra)) {
1025 			dl = dev;
1026 			break;
1027 		}
1028 	}
1029 	if (unit == -1)
1030 		unit = low & CLONE_UNITMASK;
1031 	dev = newdev(csw, unit2minor(unit | extra), ndev);
1032 	if (dev->si_flags & SI_CLONELIST) {
1033 		printf("dev %p (%s) is on clonelist\n", dev, dev->si_name);
1034 		printf("unit=%d, low=%d, extra=0x%x\n", unit, low, extra);
1035 		LIST_FOREACH(dev, &cd->head, si_clone) {
1036 			printf("\t%p %s\n", dev, dev->si_name);
1037 		}
1038 		panic("foo");
1039 	}
1040 	KASSERT(!(dev->si_flags & SI_CLONELIST),
1041 	    ("Dev %p(%s) should not be on clonelist", dev, dev->si_name));
1042 	if (dl != NULL)
1043 		LIST_INSERT_BEFORE(dl, dev, si_clone);
1044 	else if (de != NULL)
1045 		LIST_INSERT_AFTER(de, dev, si_clone);
1046 	else
1047 		LIST_INSERT_HEAD(&cd->head, dev, si_clone);
1048 	dev->si_flags |= SI_CLONELIST;
1049 	*up = unit;
1050 	dev_unlock_and_free();
1051 	return (1);
1052 }
1053 
1054 /*
1055  * Kill everything still on the list.  The driver should already have
1056  * disposed of any softc hung of the struct cdev *'s at this time.
1057  */
1058 void
1059 clone_cleanup(struct clonedevs **cdp)
1060 {
1061 	struct cdev *dev;
1062 	struct cdev_priv *cp;
1063 	struct clonedevs *cd;
1064 
1065 	cd = *cdp;
1066 	if (cd == NULL)
1067 		return;
1068 	dev_lock();
1069 	while (!LIST_EMPTY(&cd->head)) {
1070 		dev = LIST_FIRST(&cd->head);
1071 		LIST_REMOVE(dev, si_clone);
1072 		KASSERT(dev->si_flags & SI_CLONELIST,
1073 		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
1074 		dev->si_flags &= ~SI_CLONELIST;
1075 		cp = cdev2priv(dev);
1076 		if (!(cp->cdp_flags & CDP_SCHED_DTR)) {
1077 			cp->cdp_flags |= CDP_SCHED_DTR;
1078 			KASSERT(dev->si_flags & SI_NAMED,
1079 				("Driver has goofed in cloning underways udev %x", dev->si_drv0));
1080 			destroy_devl(dev);
1081 		}
1082 	}
1083 	dev_unlock_and_free();
1084 	free(cd, M_DEVBUF);
1085 	*cdp = NULL;
1086 }
1087 
1088 static TAILQ_HEAD(, cdev_priv) dev_ddtr =
1089 	TAILQ_HEAD_INITIALIZER(dev_ddtr);
1090 static struct task dev_dtr_task;
1091 
1092 static void
1093 destroy_dev_tq(void *ctx, int pending)
1094 {
1095 	struct cdev_priv *cp;
1096 	struct cdev *dev;
1097 	void (*cb)(void *);
1098 	void *cb_arg;
1099 
1100 	dev_lock();
1101 	while (!TAILQ_EMPTY(&dev_ddtr)) {
1102 		cp = TAILQ_FIRST(&dev_ddtr);
1103 		dev = &cp->cdp_c;
1104 		KASSERT(cp->cdp_flags & CDP_SCHED_DTR,
1105 		    ("cdev %p in dev_destroy_tq without CDP_SCHED_DTR", cp));
1106 		TAILQ_REMOVE(&dev_ddtr, cp, cdp_dtr_list);
1107 		cb = cp->cdp_dtr_cb;
1108 		cb_arg = cp->cdp_dtr_cb_arg;
1109 		destroy_devl(dev);
1110 		dev_unlock_and_free();
1111 		dev_rel(dev);
1112 		if (cb != NULL)
1113 			cb(cb_arg);
1114 		dev_lock();
1115 	}
1116 	dev_unlock();
1117 }
1118 
1119 /*
1120  * devmtx shall be locked on entry. devmtx will be unlocked after
1121  * function return.
1122  */
1123 static int
1124 destroy_dev_sched_cbl(struct cdev *dev, void (*cb)(void *), void *arg)
1125 {
1126 	struct cdev_priv *cp;
1127 
1128 	mtx_assert(&devmtx, MA_OWNED);
1129 	cp = cdev2priv(dev);
1130 	if (cp->cdp_flags & CDP_SCHED_DTR) {
1131 		dev_unlock();
1132 		return (0);
1133 	}
1134 	dev_refl(dev);
1135 	cp->cdp_flags |= CDP_SCHED_DTR;
1136 	cp->cdp_dtr_cb = cb;
1137 	cp->cdp_dtr_cb_arg = arg;
1138 	TAILQ_INSERT_TAIL(&dev_ddtr, cp, cdp_dtr_list);
1139 	dev_unlock();
1140 	taskqueue_enqueue(taskqueue_swi_giant, &dev_dtr_task);
1141 	return (1);
1142 }
1143 
1144 int
1145 destroy_dev_sched_cb(struct cdev *dev, void (*cb)(void *), void *arg)
1146 {
1147 	dev_lock();
1148 	return (destroy_dev_sched_cbl(dev, cb, arg));
1149 }
1150 
1151 int
1152 destroy_dev_sched(struct cdev *dev)
1153 {
1154 	return (destroy_dev_sched_cb(dev, NULL, NULL));
1155 }
1156 
1157 void
1158 destroy_dev_drain(struct cdevsw *csw)
1159 {
1160 
1161 	dev_lock();
1162 	while (!LIST_EMPTY(&csw->d_devs)) {
1163 		msleep(&csw->d_devs, &devmtx, PRIBIO, "devscd", hz/10);
1164 	}
1165 	dev_unlock();
1166 }
1167 
1168 void
1169 drain_dev_clone_events(void)
1170 {
1171 
1172 	sx_xlock(&clone_drain_lock);
1173 	sx_xunlock(&clone_drain_lock);
1174 }
1175 
1176 static void
1177 devdtr_init(void *dummy __unused)
1178 {
1179 
1180 	TASK_INIT(&dev_dtr_task, 0, destroy_dev_tq, NULL);
1181 }
1182 
1183 SYSINIT(devdtr, SI_SUB_DEVFS, SI_ORDER_SECOND, devdtr_init, NULL);
1184