xref: /freebsd/sys/kern/kern_conf.c (revision 690b7ea081790eef2c890f63a4fe7e195cf51df0)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1999-2002 Poul-Henning Kamp
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/kernel.h>
34 #include <sys/systm.h>
35 #include <sys/bio.h>
36 #include <sys/devctl.h>
37 #include <sys/lock.h>
38 #include <sys/mutex.h>
39 #include <sys/module.h>
40 #include <sys/malloc.h>
41 #include <sys/conf.h>
42 #include <sys/vnode.h>
43 #include <sys/queue.h>
44 #include <sys/poll.h>
45 #include <sys/sx.h>
46 #include <sys/ctype.h>
47 #include <sys/ucred.h>
48 #include <sys/taskqueue.h>
49 #include <machine/stdarg.h>
50 
51 #include <fs/devfs/devfs_int.h>
52 #include <vm/vm.h>
53 
54 static MALLOC_DEFINE(M_DEVT, "cdev", "cdev storage");
55 
56 struct mtx devmtx;
57 static void destroy_devl(struct cdev *dev);
58 static int destroy_dev_sched_cbl(struct cdev *dev,
59     void (*cb)(void *), void *arg);
60 static void destroy_dev_tq(void *ctx, int pending);
61 static void destroy_dev_tq_giant(void *ctx, int pending);
62 static int make_dev_credv(int flags, struct cdev **dres, struct cdevsw *devsw,
63     int unit, struct ucred *cr, uid_t uid, gid_t gid, int mode, const char *fmt,
64     va_list ap);
65 
66 static struct cdev_priv_list cdevp_free_list =
67     TAILQ_HEAD_INITIALIZER(cdevp_free_list);
68 static SLIST_HEAD(free_cdevsw, cdevsw) cdevsw_gt_post_list =
69     SLIST_HEAD_INITIALIZER(cdevsw_gt_post_list);
70 
71 void
72 dev_lock(void)
73 {
74 
75 	mtx_lock(&devmtx);
76 }
77 
78 /*
79  * Free all the memory collected while the cdev mutex was
80  * locked. Since devmtx is after the system map mutex, free() cannot
81  * be called immediately and is postponed until cdev mutex can be
82  * dropped.
83  */
84 static void
85 dev_unlock_and_free(void)
86 {
87 	struct cdev_priv_list cdp_free;
88 	struct free_cdevsw csw_free;
89 	struct cdev_priv *cdp;
90 	struct cdevsw *csw;
91 
92 	dev_lock_assert_locked();
93 
94 	/*
95 	 * Make the local copy of the list heads while the dev_mtx is
96 	 * held. Free it later.
97 	 */
98 	TAILQ_INIT(&cdp_free);
99 	TAILQ_CONCAT(&cdp_free, &cdevp_free_list, cdp_list);
100 	csw_free = cdevsw_gt_post_list;
101 	SLIST_INIT(&cdevsw_gt_post_list);
102 
103 	mtx_unlock(&devmtx);
104 
105 	while ((cdp = TAILQ_FIRST(&cdp_free)) != NULL) {
106 		TAILQ_REMOVE(&cdp_free, cdp, cdp_list);
107 		devfs_free(&cdp->cdp_c);
108 	}
109 	while ((csw = SLIST_FIRST(&csw_free)) != NULL) {
110 		SLIST_REMOVE_HEAD(&csw_free, d_postfree_list);
111 		free(csw, M_DEVT);
112 	}
113 }
114 
115 static void
116 dev_free_devlocked(struct cdev *cdev)
117 {
118 	struct cdev_priv *cdp;
119 
120 	dev_lock_assert_locked();
121 	cdp = cdev2priv(cdev);
122 	KASSERT((cdp->cdp_flags & CDP_UNREF_DTR) == 0,
123 	    ("destroy_dev() was not called after delist_dev(%p)", cdev));
124 	TAILQ_INSERT_HEAD(&cdevp_free_list, cdp, cdp_list);
125 }
126 
127 static void
128 cdevsw_free_devlocked(struct cdevsw *csw)
129 {
130 
131 	dev_lock_assert_locked();
132 	SLIST_INSERT_HEAD(&cdevsw_gt_post_list, csw, d_postfree_list);
133 }
134 
135 void
136 dev_unlock(void)
137 {
138 
139 	mtx_unlock(&devmtx);
140 }
141 
142 void
143 dev_ref(struct cdev *dev)
144 {
145 
146 	dev_lock_assert_unlocked();
147 	mtx_lock(&devmtx);
148 	dev->si_refcount++;
149 	mtx_unlock(&devmtx);
150 }
151 
152 void
153 dev_refl(struct cdev *dev)
154 {
155 
156 	dev_lock_assert_locked();
157 	dev->si_refcount++;
158 }
159 
160 void
161 dev_rel(struct cdev *dev)
162 {
163 	int flag = 0;
164 
165 	dev_lock_assert_unlocked();
166 	dev_lock();
167 	dev->si_refcount--;
168 	KASSERT(dev->si_refcount >= 0,
169 	    ("dev_rel(%s) gave negative count", devtoname(dev)));
170 	if (dev->si_devsw == NULL && dev->si_refcount == 0) {
171 		LIST_REMOVE(dev, si_list);
172 		flag = 1;
173 	}
174 	dev_unlock();
175 	if (flag)
176 		devfs_free(dev);
177 }
178 
179 struct cdevsw *
180 dev_refthread(struct cdev *dev, int *ref)
181 {
182 	struct cdevsw *csw;
183 	struct cdev_priv *cdp;
184 
185 	dev_lock_assert_unlocked();
186 	if ((dev->si_flags & SI_ETERNAL) != 0) {
187 		*ref = 0;
188 		return (dev->si_devsw);
189 	}
190 	cdp = cdev2priv(dev);
191 	mtx_lock(&cdp->cdp_threadlock);
192 	csw = dev->si_devsw;
193 	if (csw != NULL) {
194 		if ((cdp->cdp_flags & CDP_SCHED_DTR) == 0)
195 			atomic_add_long(&dev->si_threadcount, 1);
196 		else
197 			csw = NULL;
198 	}
199 	mtx_unlock(&cdp->cdp_threadlock);
200 	if (csw != NULL)
201 		*ref = 1;
202 	return (csw);
203 }
204 
205 struct cdevsw *
206 devvn_refthread(struct vnode *vp, struct cdev **devp, int *ref)
207 {
208 	struct cdevsw *csw;
209 	struct cdev_priv *cdp;
210 	struct cdev *dev;
211 
212 	dev_lock_assert_unlocked();
213 	if ((vp->v_vflag & VV_ETERNALDEV) != 0) {
214 		dev = vp->v_rdev;
215 		if (dev == NULL)
216 			return (NULL);
217 		KASSERT((dev->si_flags & SI_ETERNAL) != 0,
218 		    ("Not eternal cdev"));
219 		*ref = 0;
220 		csw = dev->si_devsw;
221 		KASSERT(csw != NULL, ("Eternal cdev is destroyed"));
222 		*devp = dev;
223 		return (csw);
224 	}
225 
226 	csw = NULL;
227 	VI_LOCK(vp);
228 	dev = vp->v_rdev;
229 	if (dev == NULL) {
230 		VI_UNLOCK(vp);
231 		return (NULL);
232 	}
233 	cdp = cdev2priv(dev);
234 	mtx_lock(&cdp->cdp_threadlock);
235 	if ((cdp->cdp_flags & CDP_SCHED_DTR) == 0) {
236 		csw = dev->si_devsw;
237 		if (csw != NULL)
238 			atomic_add_long(&dev->si_threadcount, 1);
239 	}
240 	mtx_unlock(&cdp->cdp_threadlock);
241 	VI_UNLOCK(vp);
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 	dev_lock_assert_unlocked();
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("DEVFS", "CDEV", ev, data);
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 	dev_lock_assert_locked();
575 	csw = args->mda_devsw;
576 	si2 = NULL;
577 	if (csw->d_flags & D_NEEDMINOR) {
578 		/* We may want to return an existing device */
579 		LIST_FOREACH(si2, &csw->d_devs, si_list) {
580 			if (dev2unit(si2) == args->mda_unit) {
581 				dev_free_devlocked(si);
582 				si = si2;
583 				break;
584 			}
585 		}
586 
587 		/*
588 		 * If we're returning an existing device, we should make sure
589 		 * it isn't already initialized.  This would have been caught
590 		 * in consumers anyways, but it's good to catch such a case
591 		 * early.  We still need to complete initialization of the
592 		 * device, and we'll use whatever make_dev_args were passed in
593 		 * to do so.
594 		 */
595 		KASSERT(si2 == NULL || (si2->si_flags & SI_NAMED) == 0,
596 		    ("make_dev() by driver %s on pre-existing device (min=%x, name=%s)",
597 		    args->mda_devsw->d_name, dev2unit(si2), devtoname(si2)));
598 	}
599 	si->si_drv0 = args->mda_unit;
600 	si->si_drv1 = args->mda_si_drv1;
601 	si->si_drv2 = args->mda_si_drv2;
602 	/* Only push to csw->d_devs if it's not a cloned device. */
603 	if (si2 == NULL) {
604 		si->si_devsw = csw;
605 		LIST_INSERT_HEAD(&csw->d_devs, si, si_list);
606 	} else {
607 		KASSERT(si->si_devsw == csw,
608 		    ("%s: inconsistent devsw between clone_create() and make_dev()",
609 		    __func__));
610 	}
611 	return (si);
612 }
613 
614 static void
615 fini_cdevsw(struct cdevsw *devsw)
616 {
617 	struct cdevsw *gt;
618 
619 	if (devsw->d_gianttrick != NULL) {
620 		gt = devsw->d_gianttrick;
621 		memcpy(devsw, gt, sizeof *devsw);
622 		cdevsw_free_devlocked(gt);
623 		devsw->d_gianttrick = NULL;
624 	}
625 	devsw->d_flags &= ~D_INIT;
626 }
627 
628 static int
629 prep_cdevsw(struct cdevsw *devsw, int flags)
630 {
631 	struct cdevsw *dsw2;
632 
633 	dev_lock_assert_locked();
634 	if (devsw->d_flags & D_INIT)
635 		return (0);
636 	if (devsw->d_flags & D_NEEDGIANT) {
637 		dev_unlock();
638 		dsw2 = malloc(sizeof *dsw2, M_DEVT,
639 		     (flags & MAKEDEV_NOWAIT) ? M_NOWAIT : M_WAITOK);
640 		dev_lock();
641 		if (dsw2 == NULL && !(devsw->d_flags & D_INIT))
642 			return (ENOMEM);
643 	} else
644 		dsw2 = NULL;
645 	if (devsw->d_flags & D_INIT) {
646 		if (dsw2 != NULL)
647 			cdevsw_free_devlocked(dsw2);
648 		return (0);
649 	}
650 
651 	if (devsw->d_version != D_VERSION_04) {
652 		printf(
653 		    "WARNING: Device driver \"%s\" has wrong version %s\n",
654 		    devsw->d_name == NULL ? "???" : devsw->d_name,
655 		    "and is disabled.  Recompile KLD module.");
656 		devsw->d_open = dead_open;
657 		devsw->d_close = dead_close;
658 		devsw->d_read = dead_read;
659 		devsw->d_write = dead_write;
660 		devsw->d_ioctl = dead_ioctl;
661 		devsw->d_poll = dead_poll;
662 		devsw->d_mmap = dead_mmap;
663 		devsw->d_mmap_single = dead_mmap_single;
664 		devsw->d_strategy = dead_strategy;
665 		devsw->d_dump = dead_dump;
666 		devsw->d_kqfilter = dead_kqfilter;
667 	}
668 
669 	if ((devsw->d_flags & D_NEEDGIANT) != 0) {
670 		if ((devsw->d_flags & D_GIANTOK) == 0) {
671 			printf(
672 			    "WARNING: Device \"%s\" is Giant locked and may be "
673 			    "deleted before FreeBSD 14.0.\n",
674 			    devsw->d_name == NULL ? "???" : devsw->d_name);
675 		}
676 		if (devsw->d_gianttrick == NULL) {
677 			memcpy(dsw2, devsw, sizeof *dsw2);
678 			devsw->d_gianttrick = dsw2;
679 			dsw2 = NULL;
680 		}
681 	}
682 
683 #define FIXUP(member, noop, giant) 				\
684 	do {							\
685 		if (devsw->member == NULL) {			\
686 			devsw->member = noop;			\
687 		} else if (devsw->d_flags & D_NEEDGIANT)	\
688 			devsw->member = giant;			\
689 		}						\
690 	while (0)
691 
692 	FIXUP(d_open,		null_open,	giant_open);
693 	FIXUP(d_fdopen,		NULL,		giant_fdopen);
694 	FIXUP(d_close,		null_close,	giant_close);
695 	FIXUP(d_read,		no_read,	giant_read);
696 	FIXUP(d_write,		no_write,	giant_write);
697 	FIXUP(d_ioctl,		no_ioctl,	giant_ioctl);
698 	FIXUP(d_poll,		no_poll,	giant_poll);
699 	FIXUP(d_mmap,		no_mmap,	giant_mmap);
700 	FIXUP(d_strategy,	no_strategy,	giant_strategy);
701 	FIXUP(d_kqfilter,	no_kqfilter,	giant_kqfilter);
702 	FIXUP(d_mmap_single,	no_mmap_single,	giant_mmap_single);
703 
704 	if (devsw->d_dump == NULL)	devsw->d_dump = no_dump;
705 
706 	LIST_INIT(&devsw->d_devs);
707 
708 	devsw->d_flags |= D_INIT;
709 
710 	if (dsw2 != NULL)
711 		cdevsw_free_devlocked(dsw2);
712 	return (0);
713 }
714 
715 static int
716 prep_devname(struct cdev *dev, const char *fmt, va_list ap)
717 {
718 	int len;
719 	char *from, *q, *s, *to;
720 
721 	dev_lock_assert_locked();
722 
723 	len = vsnrprintf(dev->si_name, sizeof(dev->si_name), 32, fmt, ap);
724 	if (len > sizeof(dev->si_name) - 1)
725 		return (ENAMETOOLONG);
726 
727 	/* Strip leading slashes. */
728 	for (from = dev->si_name; *from == '/'; from++)
729 		;
730 
731 	for (to = dev->si_name; *from != '\0'; from++, to++) {
732 		/*
733 		 * Spaces and double quotation marks cause
734 		 * problems for the devctl(4) protocol.
735 		 * Reject names containing those characters.
736 		 */
737 		if (isspace(*from) || *from == '"')
738 			return (EINVAL);
739 		/* Treat multiple sequential slashes as single. */
740 		while (from[0] == '/' && from[1] == '/')
741 			from++;
742 		/* Trailing slash is considered invalid. */
743 		if (from[0] == '/' && from[1] == '\0')
744 			return (EINVAL);
745 		*to = *from;
746 	}
747 	*to = '\0';
748 
749 	if (dev->si_name[0] == '\0')
750 		return (EINVAL);
751 
752 	/* Disallow "." and ".." components. */
753 	for (s = dev->si_name;;) {
754 		for (q = s; *q != '/' && *q != '\0'; q++)
755 			;
756 		if (q - s == 1 && s[0] == '.')
757 			return (EINVAL);
758 		if (q - s == 2 && s[0] == '.' && s[1] == '.')
759 			return (EINVAL);
760 		if (*q != '/')
761 			break;
762 		s = q + 1;
763 	}
764 
765 	if (devfs_dev_exists(dev->si_name) != 0)
766 		return (EEXIST);
767 
768 	return (0);
769 }
770 
771 void
772 make_dev_args_init_impl(struct make_dev_args *args, size_t sz)
773 {
774 
775 	bzero(args, sz);
776 	args->mda_size = sz;
777 }
778 
779 static int
780 make_dev_sv(struct make_dev_args *args1, struct cdev **dres,
781     const char *fmt, va_list ap)
782 {
783 	struct cdev *dev, *dev_new;
784 	struct make_dev_args args;
785 	int res;
786 
787 	bzero(&args, sizeof(args));
788 	if (sizeof(args) < args1->mda_size)
789 		return (EINVAL);
790 	bcopy(args1, &args, args1->mda_size);
791 	KASSERT((args.mda_flags & MAKEDEV_WAITOK) == 0 ||
792 	    (args.mda_flags & MAKEDEV_NOWAIT) == 0,
793 	    ("make_dev_sv: both WAITOK and NOWAIT specified"));
794 	dev_new = devfs_alloc(args.mda_flags);
795 	if (dev_new == NULL)
796 		return (ENOMEM);
797 	dev_lock();
798 	res = prep_cdevsw(args.mda_devsw, args.mda_flags);
799 	if (res != 0) {
800 		dev_unlock();
801 		devfs_free(dev_new);
802 		return (res);
803 	}
804 	dev = newdev(&args, dev_new);
805 	if ((dev->si_flags & SI_NAMED) == 0) {
806 		res = prep_devname(dev, fmt, ap);
807 		if (res != 0) {
808 			if ((args.mda_flags & MAKEDEV_CHECKNAME) == 0) {
809 				panic(
810 			"make_dev_sv: bad si_name (error=%d, si_name=%s)",
811 				    res, dev->si_name);
812 			}
813 			if (dev == dev_new) {
814 				LIST_REMOVE(dev, si_list);
815 				dev_unlock();
816 				devfs_free(dev);
817 			} else
818 				dev_unlock();
819 			return (res);
820 		}
821 	}
822 	if ((args.mda_flags & MAKEDEV_REF) != 0)
823 		dev_refl(dev);
824 	if ((args.mda_flags & MAKEDEV_ETERNAL) != 0)
825 		dev->si_flags |= SI_ETERNAL;
826 	KASSERT(!(dev->si_flags & SI_NAMED),
827 	    ("make_dev() by driver %s on pre-existing device (min=%x, name=%s)",
828 	    args.mda_devsw->d_name, dev2unit(dev), devtoname(dev)));
829 	dev->si_flags |= SI_NAMED;
830 	if (args.mda_cr != NULL)
831 		dev->si_cred = crhold(args.mda_cr);
832 	dev->si_uid = args.mda_uid;
833 	dev->si_gid = args.mda_gid;
834 	dev->si_mode = args.mda_mode;
835 
836 	devfs_create(dev);
837 	clean_unrhdrl(devfs_inos);
838 	dev_unlock_and_free();
839 
840 	notify_create(dev, args.mda_flags);
841 
842 	*dres = dev;
843 	return (0);
844 }
845 
846 int
847 make_dev_s(struct make_dev_args *args, struct cdev **dres,
848     const char *fmt, ...)
849 {
850 	va_list ap;
851 	int res;
852 
853 	va_start(ap, fmt);
854 	res = make_dev_sv(args, dres, fmt, ap);
855 	va_end(ap);
856 	return (res);
857 }
858 
859 static int
860 make_dev_credv(int flags, struct cdev **dres, struct cdevsw *devsw, int unit,
861     struct ucred *cr, uid_t uid, gid_t gid, int mode, const char *fmt,
862     va_list ap)
863 {
864 	struct make_dev_args args;
865 
866 	make_dev_args_init(&args);
867 	args.mda_flags = flags;
868 	args.mda_devsw = devsw;
869 	args.mda_cr = cr;
870 	args.mda_uid = uid;
871 	args.mda_gid = gid;
872 	args.mda_mode = mode;
873 	args.mda_unit = unit;
874 	return (make_dev_sv(&args, dres, fmt, ap));
875 }
876 
877 struct cdev *
878 make_dev(struct cdevsw *devsw, int unit, uid_t uid, gid_t gid, int mode,
879     const char *fmt, ...)
880 {
881 	struct cdev *dev;
882 	va_list ap;
883 	int res __unused;
884 
885 	va_start(ap, fmt);
886 	res = make_dev_credv(0, &dev, devsw, unit, NULL, uid, gid, mode, fmt,
887 		      ap);
888 	va_end(ap);
889 	KASSERT(res == 0 && dev != NULL,
890 	    ("make_dev: failed make_dev_credv (error=%d)", res));
891 	return (dev);
892 }
893 
894 struct cdev *
895 make_dev_cred(struct cdevsw *devsw, int unit, struct ucred *cr, uid_t uid,
896     gid_t gid, int mode, const char *fmt, ...)
897 {
898 	struct cdev *dev;
899 	va_list ap;
900 	int res __unused;
901 
902 	va_start(ap, fmt);
903 	res = make_dev_credv(0, &dev, devsw, unit, cr, uid, gid, mode, fmt, ap);
904 	va_end(ap);
905 
906 	KASSERT(res == 0 && dev != NULL,
907 	    ("make_dev_cred: failed make_dev_credv (error=%d)", res));
908 	return (dev);
909 }
910 
911 struct cdev *
912 make_dev_credf(int flags, struct cdevsw *devsw, int unit, struct ucred *cr,
913     uid_t uid, gid_t gid, int mode, const char *fmt, ...)
914 {
915 	struct cdev *dev;
916 	va_list ap;
917 	int res;
918 
919 	va_start(ap, fmt);
920 	res = make_dev_credv(flags, &dev, devsw, unit, cr, uid, gid, mode,
921 	    fmt, ap);
922 	va_end(ap);
923 
924 	KASSERT(((flags & MAKEDEV_NOWAIT) != 0 && res == ENOMEM) ||
925 	    ((flags & MAKEDEV_CHECKNAME) != 0 && res != ENOMEM) || res == 0,
926 	    ("make_dev_credf: failed make_dev_credv (error=%d)", res));
927 	return (res == 0 ? dev : NULL);
928 }
929 
930 int
931 make_dev_p(int flags, struct cdev **cdev, struct cdevsw *devsw,
932     struct ucred *cr, uid_t uid, gid_t gid, int mode, const char *fmt, ...)
933 {
934 	va_list ap;
935 	int res;
936 
937 	va_start(ap, fmt);
938 	res = make_dev_credv(flags, cdev, devsw, 0, cr, uid, gid, mode,
939 	    fmt, ap);
940 	va_end(ap);
941 
942 	KASSERT(((flags & MAKEDEV_NOWAIT) != 0 && res == ENOMEM) ||
943 	    ((flags & MAKEDEV_CHECKNAME) != 0 && res != ENOMEM) || res == 0,
944 	    ("make_dev_p: failed make_dev_credv (error=%d)", res));
945 	return (res);
946 }
947 
948 static void
949 dev_dependsl(struct cdev *pdev, struct cdev *cdev)
950 {
951 
952 	cdev->si_parent = pdev;
953 	cdev->si_flags |= SI_CHILD;
954 	LIST_INSERT_HEAD(&pdev->si_children, cdev, si_siblings);
955 }
956 
957 void
958 dev_depends(struct cdev *pdev, struct cdev *cdev)
959 {
960 
961 	dev_lock();
962 	dev_dependsl(pdev, cdev);
963 	dev_unlock();
964 }
965 
966 static int
967 make_dev_alias_v(int flags, struct cdev **cdev, struct cdev *pdev,
968     const char *fmt, va_list ap)
969 {
970 	struct cdev *dev;
971 	int error;
972 
973 	KASSERT(pdev != NULL, ("make_dev_alias_v: pdev is NULL"));
974 	KASSERT((flags & MAKEDEV_WAITOK) == 0 || (flags & MAKEDEV_NOWAIT) == 0,
975 	    ("make_dev_alias_v: both WAITOK and NOWAIT specified"));
976 	KASSERT((flags & ~(MAKEDEV_WAITOK | MAKEDEV_NOWAIT |
977 	    MAKEDEV_CHECKNAME)) == 0,
978 	    ("make_dev_alias_v: invalid flags specified (flags=%02x)", flags));
979 
980 	dev = devfs_alloc(flags);
981 	if (dev == NULL)
982 		return (ENOMEM);
983 	dev_lock();
984 	dev->si_flags |= SI_ALIAS;
985 	error = prep_devname(dev, fmt, ap);
986 	if (error != 0) {
987 		if ((flags & MAKEDEV_CHECKNAME) == 0) {
988 			panic("make_dev_alias_v: bad si_name "
989 			    "(error=%d, si_name=%s)", error, dev->si_name);
990 		}
991 		dev_unlock();
992 		devfs_free(dev);
993 		return (error);
994 	}
995 	dev->si_flags |= SI_NAMED;
996 	devfs_create(dev);
997 	dev_dependsl(pdev, dev);
998 	clean_unrhdrl(devfs_inos);
999 	dev_unlock();
1000 
1001 	notify_create(dev, flags);
1002 	*cdev = dev;
1003 
1004 	return (0);
1005 }
1006 
1007 struct cdev *
1008 make_dev_alias(struct cdev *pdev, const char *fmt, ...)
1009 {
1010 	struct cdev *dev;
1011 	va_list ap;
1012 	int res __unused;
1013 
1014 	va_start(ap, fmt);
1015 	res = make_dev_alias_v(MAKEDEV_WAITOK, &dev, pdev, fmt, ap);
1016 	va_end(ap);
1017 
1018 	KASSERT(res == 0 && dev != NULL,
1019 	    ("make_dev_alias: failed make_dev_alias_v (error=%d)", res));
1020 	return (dev);
1021 }
1022 
1023 int
1024 make_dev_alias_p(int flags, struct cdev **cdev, struct cdev *pdev,
1025     const char *fmt, ...)
1026 {
1027 	va_list ap;
1028 	int res;
1029 
1030 	va_start(ap, fmt);
1031 	res = make_dev_alias_v(flags, cdev, pdev, fmt, ap);
1032 	va_end(ap);
1033 	return (res);
1034 }
1035 
1036 int
1037 make_dev_physpath_alias(int flags, struct cdev **cdev, struct cdev *pdev,
1038     struct cdev *old_alias, const char *physpath)
1039 {
1040 	char *devfspath;
1041 	int physpath_len;
1042 	int max_parentpath_len;
1043 	int parentpath_len;
1044 	int devfspathbuf_len;
1045 	int mflags;
1046 	int ret;
1047 
1048 	*cdev = NULL;
1049 	devfspath = NULL;
1050 	physpath_len = strlen(physpath);
1051 	ret = EINVAL;
1052 	if (physpath_len == 0)
1053 		goto out;
1054 
1055 	if (strncmp("id1,", physpath, 4) == 0) {
1056 		physpath += 4;
1057 		physpath_len -= 4;
1058 		if (physpath_len == 0)
1059 			goto out;
1060 	}
1061 
1062 	max_parentpath_len = SPECNAMELEN - physpath_len - /*/*/1;
1063 	parentpath_len = strlen(pdev->si_name);
1064 	if (max_parentpath_len < parentpath_len) {
1065 		if (bootverbose)
1066 			printf("WARNING: Unable to alias %s "
1067 			    "to %s/%s - path too long\n",
1068 			    pdev->si_name, physpath, pdev->si_name);
1069 		ret = ENAMETOOLONG;
1070 		goto out;
1071 	}
1072 
1073 	mflags = (flags & MAKEDEV_NOWAIT) ? M_NOWAIT : M_WAITOK;
1074 	devfspathbuf_len = physpath_len + /*/*/1 + parentpath_len + /*NUL*/1;
1075 	devfspath = malloc(devfspathbuf_len, M_DEVBUF, mflags);
1076 	if (devfspath == NULL) {
1077 		ret = ENOMEM;
1078 		goto out;
1079 	}
1080 
1081 	sprintf(devfspath, "%s/%s", physpath, pdev->si_name);
1082 	if (old_alias != NULL && strcmp(old_alias->si_name, devfspath) == 0) {
1083 		/* Retain the existing alias. */
1084 		*cdev = old_alias;
1085 		old_alias = NULL;
1086 		ret = 0;
1087 	} else {
1088 		ret = make_dev_alias_p(flags, cdev, pdev, "%s", devfspath);
1089 	}
1090 out:
1091 	if (old_alias != NULL)
1092 		destroy_dev(old_alias);
1093 	if (devfspath != NULL)
1094 		free(devfspath, M_DEVBUF);
1095 	return (ret);
1096 }
1097 
1098 static void
1099 destroy_devl(struct cdev *dev)
1100 {
1101 	struct cdevsw *csw;
1102 	struct cdev_privdata *p;
1103 	struct cdev_priv *cdp;
1104 
1105 	dev_lock_assert_locked();
1106 	KASSERT(dev->si_flags & SI_NAMED,
1107 	    ("WARNING: Driver mistake: destroy_dev on %d\n", dev2unit(dev)));
1108 	KASSERT((dev->si_flags & SI_ETERNAL) == 0,
1109 	    ("WARNING: Driver mistake: destroy_dev on eternal %d\n",
1110 	     dev2unit(dev)));
1111 
1112 	cdp = cdev2priv(dev);
1113 	if ((cdp->cdp_flags & CDP_UNREF_DTR) == 0) {
1114 		/*
1115 		 * Avoid race with dev_rel(), e.g. from the populate
1116 		 * loop.  If CDP_UNREF_DTR flag is set, the reference
1117 		 * to be dropped at the end of destroy_devl() was
1118 		 * already taken by delist_dev_locked().
1119 		 */
1120 		dev_refl(dev);
1121 
1122 		devfs_destroy(dev);
1123 	}
1124 
1125 	/* Remove name marking */
1126 	dev->si_flags &= ~SI_NAMED;
1127 
1128 	/* If we are a child, remove us from the parents list */
1129 	if (dev->si_flags & SI_CHILD) {
1130 		LIST_REMOVE(dev, si_siblings);
1131 		dev->si_flags &= ~SI_CHILD;
1132 	}
1133 
1134 	/* Kill our children */
1135 	while (!LIST_EMPTY(&dev->si_children))
1136 		destroy_devl(LIST_FIRST(&dev->si_children));
1137 
1138 	/* Remove from clone list */
1139 	if (dev->si_flags & SI_CLONELIST) {
1140 		LIST_REMOVE(dev, si_clone);
1141 		dev->si_flags &= ~SI_CLONELIST;
1142 	}
1143 
1144 	mtx_lock(&cdp->cdp_threadlock);
1145 	csw = dev->si_devsw;
1146 	dev->si_devsw = NULL;	/* already NULL for SI_ALIAS */
1147 	while (csw != NULL && csw->d_purge != NULL && dev->si_threadcount) {
1148 		csw->d_purge(dev);
1149 		mtx_unlock(&cdp->cdp_threadlock);
1150 		msleep(csw, &devmtx, PRIBIO, "devprg", hz/10);
1151 		mtx_lock(&cdp->cdp_threadlock);
1152 		if (dev->si_threadcount)
1153 			printf("Still %lu threads in %s\n",
1154 			    dev->si_threadcount, devtoname(dev));
1155 	}
1156 	while (dev->si_threadcount != 0) {
1157 		/* Use unique dummy wait ident */
1158 		mtx_unlock(&cdp->cdp_threadlock);
1159 		msleep(&csw, &devmtx, PRIBIO, "devdrn", hz / 10);
1160 		mtx_lock(&cdp->cdp_threadlock);
1161 	}
1162 
1163 	mtx_unlock(&cdp->cdp_threadlock);
1164 	dev_unlock();
1165 	if ((cdp->cdp_flags & CDP_UNREF_DTR) == 0) {
1166 		/* avoid out of order notify events */
1167 		notify_destroy(dev);
1168 	}
1169 	mtx_lock(&cdevpriv_mtx);
1170 	while ((p = LIST_FIRST(&cdp->cdp_fdpriv)) != NULL) {
1171 		devfs_destroy_cdevpriv(p);
1172 		mtx_lock(&cdevpriv_mtx);
1173 	}
1174 	mtx_unlock(&cdevpriv_mtx);
1175 	dev_lock();
1176 
1177 	dev->si_drv1 = 0;
1178 	dev->si_drv2 = 0;
1179 
1180 	if (!(dev->si_flags & SI_ALIAS)) {
1181 		/* Remove from cdevsw list */
1182 		LIST_REMOVE(dev, si_list);
1183 
1184 		/* If cdevsw has no more struct cdev *'s, clean it */
1185 		if (LIST_EMPTY(&csw->d_devs)) {
1186 			fini_cdevsw(csw);
1187 			wakeup(&csw->d_devs);
1188 		}
1189 	}
1190 	dev->si_flags &= ~SI_ALIAS;
1191 	cdp->cdp_flags &= ~CDP_UNREF_DTR;
1192 	dev->si_refcount--;
1193 
1194 	if (dev->si_refcount > 0)
1195 		LIST_INSERT_HEAD(&dead_cdevsw.d_devs, dev, si_list);
1196 	else
1197 		dev_free_devlocked(dev);
1198 }
1199 
1200 static void
1201 delist_dev_locked(struct cdev *dev)
1202 {
1203 	struct cdev_priv *cdp;
1204 	struct cdev *child;
1205 
1206 	dev_lock_assert_locked();
1207 	cdp = cdev2priv(dev);
1208 	if ((cdp->cdp_flags & CDP_UNREF_DTR) != 0)
1209 		return;
1210 	cdp->cdp_flags |= CDP_UNREF_DTR;
1211 	dev_refl(dev);
1212 	devfs_destroy(dev);
1213 	LIST_FOREACH(child, &dev->si_children, si_siblings)
1214 		delist_dev_locked(child);
1215 	dev_unlock();
1216 	/* ensure the destroy event is queued in order */
1217 	notify_destroy(dev);
1218 	dev_lock();
1219 }
1220 
1221 /*
1222  * This function will delist a character device and its children from
1223  * the directory listing and create a destroy event without waiting
1224  * for all character device references to go away. At some later point
1225  * destroy_dev() must be called to complete the character device
1226  * destruction. After calling this function the character device name
1227  * can instantly be re-used.
1228  */
1229 void
1230 delist_dev(struct cdev *dev)
1231 {
1232 
1233 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "delist_dev");
1234 	dev_lock();
1235 	delist_dev_locked(dev);
1236 	dev_unlock();
1237 }
1238 
1239 void
1240 destroy_dev(struct cdev *dev)
1241 {
1242 
1243 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "destroy_dev");
1244 	dev_lock();
1245 	destroy_devl(dev);
1246 	dev_unlock_and_free();
1247 }
1248 
1249 const char *
1250 devtoname(struct cdev *dev)
1251 {
1252 
1253 	return (dev->si_name);
1254 }
1255 
1256 int
1257 dev_stdclone(char *name, char **namep, const char *stem, int *unit)
1258 {
1259 	int u, i;
1260 
1261 	i = strlen(stem);
1262 	if (strncmp(stem, name, i) != 0)
1263 		return (0);
1264 	if (!isdigit(name[i]))
1265 		return (0);
1266 	u = 0;
1267 	if (name[i] == '0' && isdigit(name[i+1]))
1268 		return (0);
1269 	while (isdigit(name[i])) {
1270 		u *= 10;
1271 		u += name[i++] - '0';
1272 	}
1273 	if (u > 0xffffff)
1274 		return (0);
1275 	*unit = u;
1276 	if (namep)
1277 		*namep = &name[i];
1278 	if (name[i])
1279 		return (2);
1280 	return (1);
1281 }
1282 
1283 /*
1284  * Helper functions for cloning device drivers.
1285  *
1286  * The objective here is to make it unnecessary for the device drivers to
1287  * use rman or similar to manage their unit number space.  Due to the way
1288  * we do "on-demand" devices, using rman or other "private" methods
1289  * will be very tricky to lock down properly once we lock down this file.
1290  *
1291  * Instead we give the drivers these routines which puts the struct cdev *'s
1292  * that are to be managed on their own list, and gives the driver the ability
1293  * to ask for the first free unit number or a given specified unit number.
1294  *
1295  * In addition these routines support paired devices (pty, nmdm and similar)
1296  * by respecting a number of "flag" bits in the minor number.
1297  *
1298  */
1299 
1300 struct clonedevs {
1301 	LIST_HEAD(,cdev)	head;
1302 };
1303 
1304 void
1305 clone_setup(struct clonedevs **cdp)
1306 {
1307 
1308 	*cdp = malloc(sizeof **cdp, M_DEVBUF, M_WAITOK | M_ZERO);
1309 	LIST_INIT(&(*cdp)->head);
1310 }
1311 
1312 int
1313 clone_create(struct clonedevs **cdp, struct cdevsw *csw, int *up,
1314     struct cdev **dp, int extra)
1315 {
1316 	struct clonedevs *cd;
1317 	struct cdev *dev, *ndev, *dl, *de;
1318 	struct make_dev_args args;
1319 	int unit, low, u;
1320 
1321 	KASSERT(*cdp != NULL,
1322 	    ("clone_setup() not called in driver \"%s\"", csw->d_name));
1323 	KASSERT(!(extra & CLONE_UNITMASK),
1324 	    ("Illegal extra bits (0x%x) in clone_create", extra));
1325 	KASSERT(*up <= CLONE_UNITMASK,
1326 	    ("Too high unit (0x%x) in clone_create", *up));
1327 	KASSERT(csw->d_flags & D_NEEDMINOR,
1328 	    ("clone_create() on cdevsw without minor numbers"));
1329 
1330 	/*
1331 	 * Search the list for a lot of things in one go:
1332 	 *   A preexisting match is returned immediately.
1333 	 *   The lowest free unit number if we are passed -1, and the place
1334 	 *	 in the list where we should insert that new element.
1335 	 *   The place to insert a specified unit number, if applicable
1336 	 *       the end of the list.
1337 	 */
1338 	unit = *up;
1339 	ndev = devfs_alloc(MAKEDEV_WAITOK);
1340 	dev_lock();
1341 	prep_cdevsw(csw, MAKEDEV_WAITOK);
1342 	low = extra;
1343 	de = dl = NULL;
1344 	cd = *cdp;
1345 	LIST_FOREACH(dev, &cd->head, si_clone) {
1346 		KASSERT(dev->si_flags & SI_CLONELIST,
1347 		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
1348 		u = dev2unit(dev);
1349 		if (u == (unit | extra)) {
1350 			*dp = dev;
1351 			dev_unlock();
1352 			devfs_free(ndev);
1353 			return (0);
1354 		}
1355 		if (unit == -1 && u == low) {
1356 			low++;
1357 			de = dev;
1358 			continue;
1359 		} else if (u < (unit | extra)) {
1360 			de = dev;
1361 			continue;
1362 		} else if (u > (unit | extra)) {
1363 			dl = dev;
1364 			break;
1365 		}
1366 	}
1367 	if (unit == -1)
1368 		unit = low & CLONE_UNITMASK;
1369 	make_dev_args_init(&args);
1370 	args.mda_unit = unit | extra;
1371 	args.mda_devsw = csw;
1372 	dev = newdev(&args, ndev);
1373 	if (dev->si_flags & SI_CLONELIST) {
1374 		printf("dev %p (%s) is on clonelist\n", dev, dev->si_name);
1375 		printf("unit=%d, low=%d, extra=0x%x\n", unit, low, extra);
1376 		LIST_FOREACH(dev, &cd->head, si_clone) {
1377 			printf("\t%p %s\n", dev, dev->si_name);
1378 		}
1379 		panic("foo");
1380 	}
1381 	KASSERT(!(dev->si_flags & SI_CLONELIST),
1382 	    ("Dev %p(%s) should not be on clonelist", dev, dev->si_name));
1383 	if (dl != NULL)
1384 		LIST_INSERT_BEFORE(dl, dev, si_clone);
1385 	else if (de != NULL)
1386 		LIST_INSERT_AFTER(de, dev, si_clone);
1387 	else
1388 		LIST_INSERT_HEAD(&cd->head, dev, si_clone);
1389 	dev->si_flags |= SI_CLONELIST;
1390 	*up = unit;
1391 	dev_unlock_and_free();
1392 	return (1);
1393 }
1394 
1395 /*
1396  * Kill everything still on the list.  The driver should already have
1397  * disposed of any softc hung of the struct cdev *'s at this time.
1398  */
1399 void
1400 clone_cleanup(struct clonedevs **cdp)
1401 {
1402 	struct cdev *dev;
1403 	struct cdev_priv *cp;
1404 	struct clonedevs *cd;
1405 
1406 	cd = *cdp;
1407 	if (cd == NULL)
1408 		return;
1409 	dev_lock();
1410 	while (!LIST_EMPTY(&cd->head)) {
1411 		dev = LIST_FIRST(&cd->head);
1412 		LIST_REMOVE(dev, si_clone);
1413 		KASSERT(dev->si_flags & SI_CLONELIST,
1414 		    ("Dev %p(%s) should be on clonelist", dev, dev->si_name));
1415 		dev->si_flags &= ~SI_CLONELIST;
1416 		cp = cdev2priv(dev);
1417 		if (!(cp->cdp_flags & CDP_SCHED_DTR)) {
1418 			cp->cdp_flags |= CDP_SCHED_DTR;
1419 			KASSERT(dev->si_flags & SI_NAMED,
1420 				("Driver has goofed in cloning underways udev %jx unit %x",
1421 				(uintmax_t)dev2udev(dev), dev2unit(dev)));
1422 			destroy_devl(dev);
1423 		}
1424 	}
1425 	dev_unlock_and_free();
1426 	free(cd, M_DEVBUF);
1427 	*cdp = NULL;
1428 }
1429 
1430 static TAILQ_HEAD(, cdev_priv) dev_ddtr =
1431 	TAILQ_HEAD_INITIALIZER(dev_ddtr);
1432 static TAILQ_HEAD(, cdev_priv) dev_ddtr_giant =
1433 	TAILQ_HEAD_INITIALIZER(dev_ddtr_giant);
1434 static struct task dev_dtr_task = TASK_INITIALIZER(0, destroy_dev_tq, &dev_ddtr);
1435 static struct task dev_dtr_task_giant = TASK_INITIALIZER(0, destroy_dev_tq_giant,
1436     &dev_ddtr_giant);
1437 
1438 static void
1439 destroy_dev_tq(void *ctx, int pending)
1440 {
1441 	TAILQ_HEAD(, cdev_priv) *ddtr = ctx;
1442 	struct cdev_priv *cp;
1443 	struct cdev *dev;
1444 	void (*cb)(void *);
1445 	void *cb_arg;
1446 
1447 	dev_lock();
1448 	while (!TAILQ_EMPTY(ddtr)) {
1449 		cp = TAILQ_FIRST(ddtr);
1450 		dev = &cp->cdp_c;
1451 		KASSERT(cp->cdp_flags & CDP_SCHED_DTR,
1452 		    ("cdev %p in dev_destroy_tq without CDP_SCHED_DTR", cp));
1453 		TAILQ_REMOVE(ddtr, cp, cdp_dtr_list);
1454 		cb = cp->cdp_dtr_cb;
1455 		cb_arg = cp->cdp_dtr_cb_arg;
1456 		destroy_devl(dev);
1457 		dev_unlock_and_free();
1458 		dev_rel(dev);
1459 		if (cb != NULL)
1460 			cb(cb_arg);
1461 		dev_lock();
1462 	}
1463 	dev_unlock();
1464 }
1465 
1466 static void
1467 destroy_dev_tq_giant(void *ctx, int pending)
1468 {
1469 	mtx_lock(&Giant);
1470 	destroy_dev_tq(ctx, pending);
1471 	mtx_unlock(&Giant);
1472 }
1473 
1474 /*
1475  * devmtx shall be locked on entry. devmtx will be unlocked after
1476  * function return.
1477  */
1478 static int
1479 destroy_dev_sched_cbl(struct cdev *dev, void (*cb)(void *), void *arg)
1480 {
1481 	struct cdev_priv *cp;
1482 	bool need_giant;
1483 
1484 	dev_lock_assert_locked();
1485 	cp = cdev2priv(dev);
1486 	if (cp->cdp_flags & CDP_SCHED_DTR) {
1487 		dev_unlock();
1488 		return (0);
1489 	}
1490 	dev_refl(dev);
1491 	cp->cdp_flags |= CDP_SCHED_DTR;
1492 	cp->cdp_dtr_cb = cb;
1493 	cp->cdp_dtr_cb_arg = arg;
1494 	need_giant = (dev->si_devsw->d_flags & D_NEEDGIANT) != 0;
1495 	if (need_giant)
1496 		TAILQ_INSERT_TAIL(&dev_ddtr_giant, cp, cdp_dtr_list);
1497 	else
1498 		TAILQ_INSERT_TAIL(&dev_ddtr, cp, cdp_dtr_list);
1499 	dev_unlock();
1500 	if (need_giant)
1501 		taskqueue_enqueue(taskqueue_thread, &dev_dtr_task_giant);
1502 	else
1503 		taskqueue_enqueue(taskqueue_thread, &dev_dtr_task);
1504 	return (1);
1505 }
1506 
1507 int
1508 destroy_dev_sched_cb(struct cdev *dev, void (*cb)(void *), void *arg)
1509 {
1510 
1511 	dev_lock();
1512 	return (destroy_dev_sched_cbl(dev, cb, arg));
1513 }
1514 
1515 int
1516 destroy_dev_sched(struct cdev *dev)
1517 {
1518 
1519 	return (destroy_dev_sched_cb(dev, NULL, NULL));
1520 }
1521 
1522 void
1523 destroy_dev_drain(struct cdevsw *csw)
1524 {
1525 
1526 	dev_lock();
1527 	while (!LIST_EMPTY(&csw->d_devs)) {
1528 		msleep(&csw->d_devs, &devmtx, PRIBIO, "devscd", hz/10);
1529 	}
1530 	dev_unlock();
1531 }
1532 
1533 void
1534 drain_dev_clone_events(void)
1535 {
1536 
1537 	sx_xlock(&clone_drain_lock);
1538 	sx_xunlock(&clone_drain_lock);
1539 }
1540 
1541 #include "opt_ddb.h"
1542 #ifdef DDB
1543 #include <sys/kernel.h>
1544 
1545 #include <ddb/ddb.h>
1546 
1547 DB_SHOW_COMMAND(cdev, db_show_cdev)
1548 {
1549 	struct cdev_priv *cdp;
1550 	struct cdev *dev;
1551 	u_int flags;
1552 	char buf[512];
1553 
1554 	if (!have_addr) {
1555 		TAILQ_FOREACH(cdp, &cdevp_list, cdp_list) {
1556 			dev = &cdp->cdp_c;
1557 			db_printf("%s %p\n", dev->si_name, dev);
1558 			if (db_pager_quit)
1559 				break;
1560 		}
1561 		return;
1562 	}
1563 
1564 	dev = (struct cdev *)addr;
1565 	cdp = cdev2priv(dev);
1566 	db_printf("dev %s ref %d use %ld thr %ld inuse %u fdpriv %p\n",
1567 	    dev->si_name, dev->si_refcount, dev->si_usecount,
1568 	    dev->si_threadcount, cdp->cdp_inuse, cdp->cdp_fdpriv.lh_first);
1569 	db_printf("devsw %p si_drv0 %d si_drv1 %p si_drv2 %p\n",
1570 	    dev->si_devsw, dev->si_drv0, dev->si_drv1, dev->si_drv2);
1571 	flags = dev->si_flags;
1572 #define	SI_FLAG(flag)	do {						\
1573 	if (flags & (flag)) {						\
1574 		if (buf[0] != '\0')					\
1575 			strlcat(buf, ", ", sizeof(buf));		\
1576 		strlcat(buf, (#flag) + 3, sizeof(buf));			\
1577 		flags &= ~(flag);					\
1578 	}								\
1579 } while (0)
1580 	buf[0] = '\0';
1581 	SI_FLAG(SI_ETERNAL);
1582 	SI_FLAG(SI_ALIAS);
1583 	SI_FLAG(SI_NAMED);
1584 	SI_FLAG(SI_CHILD);
1585 	SI_FLAG(SI_DUMPDEV);
1586 	SI_FLAG(SI_CLONELIST);
1587 	db_printf("si_flags %s\n", buf);
1588 
1589 	flags = cdp->cdp_flags;
1590 #define	CDP_FLAG(flag)	do {						\
1591 	if (flags & (flag)) {						\
1592 		if (buf[0] != '\0')					\
1593 			strlcat(buf, ", ", sizeof(buf));		\
1594 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
1595 		flags &= ~(flag);					\
1596 	}								\
1597 } while (0)
1598 	buf[0] = '\0';
1599 	CDP_FLAG(CDP_ACTIVE);
1600 	CDP_FLAG(CDP_SCHED_DTR);
1601 	db_printf("cdp_flags %s\n", buf);
1602 }
1603 #endif
1604