xref: /freebsd/sys/fs/devfs/devfs_vnops.c (revision 5ca8e32633c4ffbbcd6762e5888b6a4ba0708c6c)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2000-2004
5  *	Poul-Henning Kamp.  All rights reserved.
6  * Copyright (c) 1989, 1992-1993, 1995
7  *	The Regents of the University of California.  All rights reserved.
8  *
9  * This code is derived from software donated to Berkeley by
10  * Jan-Simon Pendry.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vnops.c 1.43
33  */
34 
35 /*
36  * TODO:
37  *	mkdir: want it ?
38  */
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/conf.h>
43 #include <sys/dirent.h>
44 #include <sys/eventhandler.h>
45 #include <sys/fcntl.h>
46 #include <sys/file.h>
47 #include <sys/filedesc.h>
48 #include <sys/filio.h>
49 #include <sys/jail.h>
50 #include <sys/kernel.h>
51 #include <sys/limits.h>
52 #include <sys/lock.h>
53 #include <sys/malloc.h>
54 #include <sys/mman.h>
55 #include <sys/mount.h>
56 #include <sys/namei.h>
57 #include <sys/priv.h>
58 #include <sys/proc.h>
59 #include <sys/stat.h>
60 #include <sys/sx.h>
61 #include <sys/sysctl.h>
62 #include <sys/time.h>
63 #include <sys/ttycom.h>
64 #include <sys/unistd.h>
65 #include <sys/vnode.h>
66 
67 static struct vop_vector devfs_vnodeops;
68 static struct vop_vector devfs_specops;
69 static struct fileops devfs_ops_f;
70 
71 #include <fs/devfs/devfs.h>
72 #include <fs/devfs/devfs_int.h>
73 
74 #include <security/mac/mac_framework.h>
75 
76 #include <vm/vm.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_object.h>
79 
80 static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data");
81 
82 struct mtx	devfs_de_interlock;
83 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF);
84 struct mtx	cdevpriv_mtx;
85 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
86 
87 SYSCTL_DECL(_vfs_devfs);
88 
89 static int devfs_dotimes;
90 SYSCTL_INT(_vfs_devfs, OID_AUTO, dotimes, CTLFLAG_RW,
91     &devfs_dotimes, 0, "Update timestamps on DEVFS with default precision");
92 
93 /*
94  * Update devfs node timestamp.  Note that updates are unlocked and
95  * stat(2) could see partially updated times.
96  */
97 static void
98 devfs_timestamp(struct timespec *tsp)
99 {
100 	time_t ts;
101 
102 	if (devfs_dotimes) {
103 		vfs_timestamp(tsp);
104 	} else {
105 		ts = time_second;
106 		if (tsp->tv_sec != ts) {
107 			tsp->tv_sec = ts;
108 			tsp->tv_nsec = 0;
109 		}
110 	}
111 }
112 
113 static int
114 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp,
115     int *ref)
116 {
117 	*dswp = devvn_refthread(fp->f_vnode, devp, ref);
118 	if (*dswp == NULL || *devp != fp->f_data) {
119 		if (*dswp != NULL)
120 			dev_relthread(*devp, *ref);
121 		return (ENXIO);
122 	}
123 	KASSERT((*devp)->si_refcount > 0,
124 	    ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
125 	if (*dswp == NULL)
126 		return (ENXIO);
127 	curthread->td_fpop = fp;
128 	return (0);
129 }
130 
131 int
132 devfs_get_cdevpriv(void **datap)
133 {
134 	struct file *fp;
135 	struct cdev_privdata *p;
136 	int error;
137 
138 	fp = curthread->td_fpop;
139 	if (fp == NULL)
140 		return (EBADF);
141 	p = fp->f_cdevpriv;
142 	if (p != NULL) {
143 		error = 0;
144 		*datap = p->cdpd_data;
145 	} else
146 		error = ENOENT;
147 	return (error);
148 }
149 
150 int
151 devfs_set_cdevpriv(void *priv, d_priv_dtor_t *priv_dtr)
152 {
153 	struct file *fp;
154 	struct cdev_priv *cdp;
155 	struct cdev_privdata *p;
156 	int error;
157 
158 	fp = curthread->td_fpop;
159 	if (fp == NULL)
160 		return (ENOENT);
161 	cdp = cdev2priv((struct cdev *)fp->f_data);
162 	p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
163 	p->cdpd_data = priv;
164 	p->cdpd_dtr = priv_dtr;
165 	p->cdpd_fp = fp;
166 	mtx_lock(&cdevpriv_mtx);
167 	if (fp->f_cdevpriv == NULL) {
168 		LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
169 		fp->f_cdevpriv = p;
170 		mtx_unlock(&cdevpriv_mtx);
171 		error = 0;
172 	} else {
173 		mtx_unlock(&cdevpriv_mtx);
174 		free(p, M_CDEVPDATA);
175 		error = EBUSY;
176 	}
177 	return (error);
178 }
179 
180 int
181 devfs_foreach_cdevpriv(struct cdev *dev, int (*cb)(void *data, void *arg),
182     void *arg)
183 {
184 	struct cdev_priv *cdp;
185 	struct cdev_privdata *p;
186 	int error;
187 
188 	cdp = cdev2priv(dev);
189 	error = 0;
190 	mtx_lock(&cdevpriv_mtx);
191 	LIST_FOREACH(p, &cdp->cdp_fdpriv, cdpd_list) {
192 		error = cb(p->cdpd_data, arg);
193 		if (error != 0)
194 			break;
195 	}
196 	mtx_unlock(&cdevpriv_mtx);
197 	return (error);
198 }
199 
200 void
201 devfs_destroy_cdevpriv(struct cdev_privdata *p)
202 {
203 
204 	mtx_assert(&cdevpriv_mtx, MA_OWNED);
205 	KASSERT(p->cdpd_fp->f_cdevpriv == p,
206 	    ("devfs_destoy_cdevpriv %p != %p", p->cdpd_fp->f_cdevpriv, p));
207 	p->cdpd_fp->f_cdevpriv = NULL;
208 	LIST_REMOVE(p, cdpd_list);
209 	mtx_unlock(&cdevpriv_mtx);
210 	(p->cdpd_dtr)(p->cdpd_data);
211 	free(p, M_CDEVPDATA);
212 }
213 
214 static void
215 devfs_fpdrop(struct file *fp)
216 {
217 	struct cdev_privdata *p;
218 
219 	mtx_lock(&cdevpriv_mtx);
220 	if ((p = fp->f_cdevpriv) == NULL) {
221 		mtx_unlock(&cdevpriv_mtx);
222 		return;
223 	}
224 	devfs_destroy_cdevpriv(p);
225 }
226 
227 void
228 devfs_clear_cdevpriv(void)
229 {
230 	struct file *fp;
231 
232 	fp = curthread->td_fpop;
233 	if (fp == NULL)
234 		return;
235 	devfs_fpdrop(fp);
236 }
237 
238 static void
239 devfs_usecount_add(struct vnode *vp)
240 {
241 	struct devfs_dirent *de;
242 	struct cdev *dev;
243 
244 	mtx_lock(&devfs_de_interlock);
245 	VI_LOCK(vp);
246 	VNPASS(vp->v_type == VCHR || vp->v_type == VBAD, vp);
247 	if (VN_IS_DOOMED(vp)) {
248 		goto out_unlock;
249 	}
250 
251 	de = vp->v_data;
252 	dev = vp->v_rdev;
253 	MPASS(de != NULL);
254 	MPASS(dev != NULL);
255 	dev->si_usecount++;
256 	de->de_usecount++;
257 out_unlock:
258 	VI_UNLOCK(vp);
259 	mtx_unlock(&devfs_de_interlock);
260 }
261 
262 static void
263 devfs_usecount_subl(struct vnode *vp)
264 {
265 	struct devfs_dirent *de;
266 	struct cdev *dev;
267 
268 	mtx_assert(&devfs_de_interlock, MA_OWNED);
269 	ASSERT_VI_LOCKED(vp, __func__);
270 	VNPASS(vp->v_type == VCHR || vp->v_type == VBAD, vp);
271 
272 	de = vp->v_data;
273 	dev = vp->v_rdev;
274 	if (de == NULL)
275 		return;
276 	if (dev == NULL) {
277 		MPASS(de->de_usecount == 0);
278 		return;
279 	}
280 	if (dev->si_usecount < de->de_usecount)
281 		panic("%s: si_usecount underflow for dev %p "
282 		    "(has %ld, dirent has %d)\n",
283 		    __func__, dev, dev->si_usecount, de->de_usecount);
284 	if (VN_IS_DOOMED(vp)) {
285 		dev->si_usecount -= de->de_usecount;
286 		de->de_usecount = 0;
287 	} else {
288 		if (de->de_usecount == 0)
289 			panic("%s: de_usecount underflow for dev %p\n",
290 			    __func__, dev);
291 		dev->si_usecount--;
292 		de->de_usecount--;
293 	}
294 }
295 
296 static void
297 devfs_usecount_sub(struct vnode *vp)
298 {
299 
300 	mtx_lock(&devfs_de_interlock);
301 	VI_LOCK(vp);
302 	devfs_usecount_subl(vp);
303 	VI_UNLOCK(vp);
304 	mtx_unlock(&devfs_de_interlock);
305 }
306 
307 static int
308 devfs_usecountl(struct vnode *vp)
309 {
310 
311 	VNPASS(vp->v_type == VCHR, vp);
312 	mtx_assert(&devfs_de_interlock, MA_OWNED);
313 	ASSERT_VI_LOCKED(vp, __func__);
314 	return (vp->v_rdev->si_usecount);
315 }
316 
317 int
318 devfs_usecount(struct vnode *vp)
319 {
320 	int count;
321 
322 	VNPASS(vp->v_type == VCHR, vp);
323 	mtx_lock(&devfs_de_interlock);
324 	VI_LOCK(vp);
325 	count = devfs_usecountl(vp);
326 	VI_UNLOCK(vp);
327 	mtx_unlock(&devfs_de_interlock);
328 	return (count);
329 }
330 
331 void
332 devfs_ctty_ref(struct vnode *vp)
333 {
334 
335 	vrefact(vp);
336 	devfs_usecount_add(vp);
337 }
338 
339 void
340 devfs_ctty_unref(struct vnode *vp)
341 {
342 
343 	devfs_usecount_sub(vp);
344 	vrele(vp);
345 }
346 
347 /*
348  * On success devfs_populate_vp() returns with dmp->dm_lock held.
349  */
350 static int
351 devfs_populate_vp(struct vnode *vp)
352 {
353 	struct devfs_dirent *de;
354 	struct devfs_mount *dmp;
355 	int locked;
356 
357 	ASSERT_VOP_LOCKED(vp, "devfs_populate_vp");
358 
359 	dmp = VFSTODEVFS(vp->v_mount);
360 	if (!devfs_populate_needed(dmp)) {
361 		sx_xlock(&dmp->dm_lock);
362 		goto out_nopopulate;
363 	}
364 
365 	locked = VOP_ISLOCKED(vp);
366 
367 	sx_xlock(&dmp->dm_lock);
368 	DEVFS_DMP_HOLD(dmp);
369 
370 	/* Can't call devfs_populate() with the vnode lock held. */
371 	VOP_UNLOCK(vp);
372 	devfs_populate(dmp);
373 
374 	sx_xunlock(&dmp->dm_lock);
375 	vn_lock(vp, locked | LK_RETRY);
376 	sx_xlock(&dmp->dm_lock);
377 	if (DEVFS_DMP_DROP(dmp)) {
378 		sx_xunlock(&dmp->dm_lock);
379 		devfs_unmount_final(dmp);
380 		return (ERESTART);
381 	}
382 out_nopopulate:
383 	if (VN_IS_DOOMED(vp)) {
384 		sx_xunlock(&dmp->dm_lock);
385 		return (ERESTART);
386 	}
387 	de = vp->v_data;
388 	KASSERT(de != NULL,
389 	    ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed"));
390 	if ((de->de_flags & DE_DOOMED) != 0) {
391 		sx_xunlock(&dmp->dm_lock);
392 		return (ERESTART);
393 	}
394 
395 	return (0);
396 }
397 
398 static int
399 devfs_vptocnp(struct vop_vptocnp_args *ap)
400 {
401 	struct vnode *vp = ap->a_vp;
402 	struct vnode **dvp = ap->a_vpp;
403 	struct devfs_mount *dmp;
404 	char *buf = ap->a_buf;
405 	size_t *buflen = ap->a_buflen;
406 	struct devfs_dirent *dd, *de;
407 	int i, error;
408 
409 	dmp = VFSTODEVFS(vp->v_mount);
410 
411 	error = devfs_populate_vp(vp);
412 	if (error != 0)
413 		return (error);
414 
415 	if (vp->v_type != VCHR && vp->v_type != VDIR) {
416 		error = ENOENT;
417 		goto finished;
418 	}
419 
420 	dd = vp->v_data;
421 	if (vp->v_type == VDIR && dd == dmp->dm_rootdir) {
422 		*dvp = vp;
423 		vref(*dvp);
424 		goto finished;
425 	}
426 
427 	i = *buflen;
428 	i -= dd->de_dirent->d_namlen;
429 	if (i < 0) {
430 		error = ENOMEM;
431 		goto finished;
432 	}
433 	bcopy(dd->de_dirent->d_name, buf + i, dd->de_dirent->d_namlen);
434 	*buflen = i;
435 	de = devfs_parent_dirent(dd);
436 	if (de == NULL) {
437 		error = ENOENT;
438 		goto finished;
439 	}
440 	mtx_lock(&devfs_de_interlock);
441 	*dvp = de->de_vnode;
442 	if (*dvp != NULL) {
443 		VI_LOCK(*dvp);
444 		mtx_unlock(&devfs_de_interlock);
445 		vholdl(*dvp);
446 		VI_UNLOCK(*dvp);
447 		vref(*dvp);
448 		vdrop(*dvp);
449 	} else {
450 		mtx_unlock(&devfs_de_interlock);
451 		error = ENOENT;
452 	}
453 finished:
454 	sx_xunlock(&dmp->dm_lock);
455 	return (error);
456 }
457 
458 /*
459  * Construct the fully qualified path name relative to the mountpoint.
460  * If a NULL cnp is provided, no '/' is appended to the resulting path.
461  */
462 char *
463 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd,
464     struct componentname *cnp)
465 {
466 	int i;
467 	struct devfs_dirent *de;
468 
469 	sx_assert(&dmp->dm_lock, SA_LOCKED);
470 
471 	i = SPECNAMELEN;
472 	buf[i] = '\0';
473 	if (cnp != NULL)
474 		i -= cnp->cn_namelen;
475 	if (i < 0)
476 		 return (NULL);
477 	if (cnp != NULL)
478 		bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
479 	de = dd;
480 	while (de != dmp->dm_rootdir) {
481 		if (cnp != NULL || i < SPECNAMELEN) {
482 			i--;
483 			if (i < 0)
484 				 return (NULL);
485 			buf[i] = '/';
486 		}
487 		i -= de->de_dirent->d_namlen;
488 		if (i < 0)
489 			 return (NULL);
490 		bcopy(de->de_dirent->d_name, buf + i,
491 		    de->de_dirent->d_namlen);
492 		de = devfs_parent_dirent(de);
493 		if (de == NULL)
494 			return (NULL);
495 	}
496 	return (buf + i);
497 }
498 
499 static int
500 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
501 	struct devfs_dirent *de)
502 {
503 	int not_found;
504 
505 	not_found = 0;
506 	if (de->de_flags & DE_DOOMED)
507 		not_found = 1;
508 	if (DEVFS_DE_DROP(de)) {
509 		KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
510 		devfs_dirent_free(de);
511 	}
512 	if (DEVFS_DMP_DROP(dmp)) {
513 		KASSERT(not_found == 1,
514 			("DEVFS mount struct freed before dirent"));
515 		not_found = 2;
516 		sx_xunlock(&dmp->dm_lock);
517 		devfs_unmount_final(dmp);
518 	}
519 	if (not_found == 1 || (drop_dm_lock && not_found != 2))
520 		sx_unlock(&dmp->dm_lock);
521 	return (not_found);
522 }
523 
524 /*
525  * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
526  * it on return.
527  */
528 int
529 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode,
530     struct vnode **vpp)
531 {
532 	int error;
533 	struct vnode *vp;
534 	struct cdev *dev;
535 	struct devfs_mount *dmp;
536 	struct cdevsw *dsw;
537 	enum vgetstate vs;
538 
539 	dmp = VFSTODEVFS(mp);
540 	if (de->de_flags & DE_DOOMED) {
541 		sx_xunlock(&dmp->dm_lock);
542 		return (ENOENT);
543 	}
544 loop:
545 	DEVFS_DE_HOLD(de);
546 	DEVFS_DMP_HOLD(dmp);
547 	mtx_lock(&devfs_de_interlock);
548 	vp = de->de_vnode;
549 	if (vp != NULL) {
550 		vs = vget_prep(vp);
551 		mtx_unlock(&devfs_de_interlock);
552 		sx_xunlock(&dmp->dm_lock);
553 		vget_finish(vp, lockmode | LK_RETRY, vs);
554 		sx_xlock(&dmp->dm_lock);
555 		if (devfs_allocv_drop_refs(0, dmp, de)) {
556 			vput(vp);
557 			return (ENOENT);
558 		}
559 		else if (VN_IS_DOOMED(vp)) {
560 			mtx_lock(&devfs_de_interlock);
561 			if (de->de_vnode == vp) {
562 				de->de_vnode = NULL;
563 				vp->v_data = NULL;
564 			}
565 			mtx_unlock(&devfs_de_interlock);
566 			vput(vp);
567 			goto loop;
568 		}
569 		sx_xunlock(&dmp->dm_lock);
570 		*vpp = vp;
571 		return (0);
572 	}
573 	mtx_unlock(&devfs_de_interlock);
574 	if (de->de_dirent->d_type == DT_CHR) {
575 		if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
576 			devfs_allocv_drop_refs(1, dmp, de);
577 			return (ENOENT);
578 		}
579 		dev = &de->de_cdp->cdp_c;
580 	} else {
581 		dev = NULL;
582 	}
583 	error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
584 	if (error != 0) {
585 		devfs_allocv_drop_refs(1, dmp, de);
586 		printf("devfs_allocv: failed to allocate new vnode\n");
587 		return (error);
588 	}
589 
590 	if (de->de_dirent->d_type == DT_CHR) {
591 		vp->v_type = VCHR;
592 		VI_LOCK(vp);
593 		dev_lock();
594 		dev_refl(dev);
595 		/* XXX: v_rdev should be protect by vnode lock */
596 		vp->v_rdev = dev;
597 		VNPASS(vp->v_usecount == 1, vp);
598 		/* Special casing of ttys for deadfs.  Probably redundant. */
599 		dsw = dev->si_devsw;
600 		if (dsw != NULL && (dsw->d_flags & D_TTY) != 0)
601 			vp->v_vflag |= VV_ISTTY;
602 		dev_unlock();
603 		VI_UNLOCK(vp);
604 		if ((dev->si_flags & SI_ETERNAL) != 0)
605 			vp->v_vflag |= VV_ETERNALDEV;
606 		vp->v_op = &devfs_specops;
607 	} else if (de->de_dirent->d_type == DT_DIR) {
608 		vp->v_type = VDIR;
609 	} else if (de->de_dirent->d_type == DT_LNK) {
610 		vp->v_type = VLNK;
611 	} else {
612 		vp->v_type = VBAD;
613 	}
614 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
615 	VN_LOCK_ASHARE(vp);
616 	mtx_lock(&devfs_de_interlock);
617 	vp->v_data = de;
618 	de->de_vnode = vp;
619 	mtx_unlock(&devfs_de_interlock);
620 	error = insmntque1(vp, mp);
621 	if (error != 0) {
622 		mtx_lock(&devfs_de_interlock);
623 		vp->v_data = NULL;
624 		de->de_vnode = NULL;
625 		mtx_unlock(&devfs_de_interlock);
626 		vgone(vp);
627 		vput(vp);
628 		(void) devfs_allocv_drop_refs(1, dmp, de);
629 		return (error);
630 	}
631 	if (devfs_allocv_drop_refs(0, dmp, de)) {
632 		vgone(vp);
633 		vput(vp);
634 		return (ENOENT);
635 	}
636 #ifdef MAC
637 	mac_devfs_vnode_associate(mp, de, vp);
638 #endif
639 	sx_xunlock(&dmp->dm_lock);
640 	vn_set_state(vp, VSTATE_CONSTRUCTED);
641 	*vpp = vp;
642 	return (0);
643 }
644 
645 static int
646 devfs_access(struct vop_access_args *ap)
647 {
648 	struct vnode *vp = ap->a_vp;
649 	struct devfs_dirent *de;
650 	struct proc *p;
651 	int error;
652 
653 	de = vp->v_data;
654 	if (vp->v_type == VDIR)
655 		de = de->de_dir;
656 
657 	error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
658 	    ap->a_accmode, ap->a_cred);
659 	if (error == 0)
660 		return (0);
661 	if (error != EACCES)
662 		return (error);
663 	p = ap->a_td->td_proc;
664 	/* We do, however, allow access to the controlling terminal */
665 	PROC_LOCK(p);
666 	if (!(p->p_flag & P_CONTROLT)) {
667 		PROC_UNLOCK(p);
668 		return (error);
669 	}
670 	if (p->p_session->s_ttydp == de->de_cdp)
671 		error = 0;
672 	PROC_UNLOCK(p);
673 	return (error);
674 }
675 
676 _Static_assert(((FMASK | FCNTLFLAGS) & (FLASTCLOSE | FREVOKE)) == 0,
677     "devfs-only flag reuse failed");
678 
679 static int
680 devfs_close(struct vop_close_args *ap)
681 {
682 	struct vnode *vp = ap->a_vp, *oldvp;
683 	struct thread *td = ap->a_td;
684 	struct proc *p;
685 	struct cdev *dev = vp->v_rdev;
686 	struct cdevsw *dsw;
687 	struct devfs_dirent *de = vp->v_data;
688 	int dflags, error, ref, vp_locked;
689 
690 	/*
691 	 * XXX: Don't call d_close() if we were called because of
692 	 * XXX: insmntque() failure.
693 	 */
694 	if (vp->v_data == NULL)
695 		return (0);
696 
697 	/*
698 	 * Hack: a tty device that is a controlling terminal
699 	 * has a reference from the session structure.
700 	 * We cannot easily tell that a character device is
701 	 * a controlling terminal, unless it is the closing
702 	 * process' controlling terminal.  In that case,
703 	 * if the reference count is 2 (this last descriptor
704 	 * plus the session), release the reference from the session.
705 	 */
706 	if (de->de_usecount == 2 && td != NULL) {
707 		p = td->td_proc;
708 		PROC_LOCK(p);
709 		if (vp == p->p_session->s_ttyvp) {
710 			PROC_UNLOCK(p);
711 			oldvp = NULL;
712 			sx_xlock(&proctree_lock);
713 			if (vp == p->p_session->s_ttyvp) {
714 				SESS_LOCK(p->p_session);
715 				mtx_lock(&devfs_de_interlock);
716 				VI_LOCK(vp);
717 				if (devfs_usecountl(vp) == 2 && !VN_IS_DOOMED(vp)) {
718 					p->p_session->s_ttyvp = NULL;
719 					p->p_session->s_ttydp = NULL;
720 					oldvp = vp;
721 				}
722 				VI_UNLOCK(vp);
723 				mtx_unlock(&devfs_de_interlock);
724 				SESS_UNLOCK(p->p_session);
725 			}
726 			sx_xunlock(&proctree_lock);
727 			if (oldvp != NULL)
728 				devfs_ctty_unref(oldvp);
729 		} else
730 			PROC_UNLOCK(p);
731 	}
732 	/*
733 	 * We do not want to really close the device if it
734 	 * is still in use unless we are trying to close it
735 	 * forcibly. Since every use (buffer, vnode, swap, cmap)
736 	 * holds a reference to the vnode, and because we mark
737 	 * any other vnodes that alias this device, when the
738 	 * sum of the reference counts on all the aliased
739 	 * vnodes descends to one, we are on last close.
740 	 */
741 	dsw = dev_refthread(dev, &ref);
742 	if (dsw == NULL)
743 		return (ENXIO);
744 	dflags = 0;
745 	mtx_lock(&devfs_de_interlock);
746 	VI_LOCK(vp);
747 	if (devfs_usecountl(vp) == 1)
748 		dflags |= FLASTCLOSE;
749 	devfs_usecount_subl(vp);
750 	mtx_unlock(&devfs_de_interlock);
751 	if (VN_IS_DOOMED(vp)) {
752 		/* Forced close. */
753 		dflags |= FREVOKE | FNONBLOCK;
754 	} else if (dsw->d_flags & D_TRACKCLOSE) {
755 		/* Keep device updated on status. */
756 	} else if ((dflags & FLASTCLOSE) == 0) {
757 		VI_UNLOCK(vp);
758 		dev_relthread(dev, ref);
759 		return (0);
760 	}
761 	vholdnz(vp);
762 	VI_UNLOCK(vp);
763 	vp_locked = VOP_ISLOCKED(vp);
764 	VOP_UNLOCK(vp);
765 	KASSERT(dev->si_refcount > 0,
766 	    ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
767 	error = dsw->d_close(dev, ap->a_fflag | dflags, S_IFCHR, td);
768 	dev_relthread(dev, ref);
769 	vn_lock(vp, vp_locked | LK_RETRY);
770 	vdrop(vp);
771 	return (error);
772 }
773 
774 static int
775 devfs_close_f(struct file *fp, struct thread *td)
776 {
777 	int error;
778 	struct file *fpop;
779 
780 	/*
781 	 * NB: td may be NULL if this descriptor is closed due to
782 	 * garbage collection from a closed UNIX domain socket.
783 	 */
784 	fpop = curthread->td_fpop;
785 	curthread->td_fpop = fp;
786 	error = vnops.fo_close(fp, td);
787 	curthread->td_fpop = fpop;
788 
789 	/*
790 	 * The f_cdevpriv cannot be assigned non-NULL value while we
791 	 * are destroying the file.
792 	 */
793 	if (fp->f_cdevpriv != NULL)
794 		devfs_fpdrop(fp);
795 	return (error);
796 }
797 
798 static int
799 devfs_getattr(struct vop_getattr_args *ap)
800 {
801 	struct vnode *vp = ap->a_vp;
802 	struct vattr *vap = ap->a_vap;
803 	struct devfs_dirent *de;
804 	struct devfs_mount *dmp;
805 	struct cdev *dev;
806 	struct timeval boottime;
807 	int error;
808 
809 	error = devfs_populate_vp(vp);
810 	if (error != 0)
811 		return (error);
812 
813 	dmp = VFSTODEVFS(vp->v_mount);
814 	sx_xunlock(&dmp->dm_lock);
815 
816 	de = vp->v_data;
817 	KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
818 	if (vp->v_type == VDIR) {
819 		de = de->de_dir;
820 		KASSERT(de != NULL,
821 		    ("Null dir dirent in devfs_getattr vp=%p", vp));
822 	}
823 	vap->va_uid = de->de_uid;
824 	vap->va_gid = de->de_gid;
825 	vap->va_mode = de->de_mode;
826 	if (vp->v_type == VLNK)
827 		vap->va_size = strlen(de->de_symlink);
828 	else if (vp->v_type == VDIR)
829 		vap->va_size = vap->va_bytes = DEV_BSIZE;
830 	else
831 		vap->va_size = 0;
832 	if (vp->v_type != VDIR)
833 		vap->va_bytes = 0;
834 	vap->va_blocksize = DEV_BSIZE;
835 	vap->va_type = vp->v_type;
836 
837 	getboottime(&boottime);
838 #define fix(aa)							\
839 	do {							\
840 		if ((aa).tv_sec <= 3600) {			\
841 			(aa).tv_sec = boottime.tv_sec;		\
842 			(aa).tv_nsec = boottime.tv_usec * 1000; \
843 		}						\
844 	} while (0)
845 
846 	if (vp->v_type != VCHR)  {
847 		fix(de->de_atime);
848 		vap->va_atime = de->de_atime;
849 		fix(de->de_mtime);
850 		vap->va_mtime = de->de_mtime;
851 		fix(de->de_ctime);
852 		vap->va_ctime = de->de_ctime;
853 	} else {
854 		dev = vp->v_rdev;
855 		fix(dev->si_atime);
856 		vap->va_atime = dev->si_atime;
857 		fix(dev->si_mtime);
858 		vap->va_mtime = dev->si_mtime;
859 		fix(dev->si_ctime);
860 		vap->va_ctime = dev->si_ctime;
861 
862 		vap->va_rdev = cdev2priv(dev)->cdp_inode;
863 	}
864 	vap->va_gen = 0;
865 	vap->va_flags = 0;
866 	vap->va_filerev = 0;
867 	vap->va_nlink = de->de_links;
868 	vap->va_fileid = de->de_inode;
869 
870 	return (error);
871 }
872 
873 /* ARGSUSED */
874 static int
875 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
876 {
877 	struct file *fpop;
878 	int error;
879 
880 	fpop = td->td_fpop;
881 	td->td_fpop = fp;
882 	error = vnops.fo_ioctl(fp, com, data, cred, td);
883 	td->td_fpop = fpop;
884 	return (error);
885 }
886 
887 void *
888 fiodgname_buf_get_ptr(void *fgnp, u_long com)
889 {
890 	union {
891 		struct fiodgname_arg	fgn;
892 #ifdef COMPAT_FREEBSD32
893 		struct fiodgname_arg32	fgn32;
894 #endif
895 	} *fgnup;
896 
897 	fgnup = fgnp;
898 	switch (com) {
899 	case FIODGNAME:
900 		return (fgnup->fgn.buf);
901 #ifdef COMPAT_FREEBSD32
902 	case FIODGNAME_32:
903 		return ((void *)(uintptr_t)fgnup->fgn32.buf);
904 #endif
905 	default:
906 		panic("Unhandled ioctl command %ld", com);
907 	}
908 }
909 
910 static int
911 devfs_ioctl(struct vop_ioctl_args *ap)
912 {
913 	struct fiodgname_arg *fgn;
914 	struct vnode *vpold, *vp;
915 	struct cdevsw *dsw;
916 	struct thread *td;
917 	struct session *sess;
918 	struct cdev *dev;
919 	int error, ref, i;
920 	const char *p;
921 	u_long com;
922 
923 	vp = ap->a_vp;
924 	com = ap->a_command;
925 	td = ap->a_td;
926 
927 	dsw = devvn_refthread(vp, &dev, &ref);
928 	if (dsw == NULL)
929 		return (ENXIO);
930 	KASSERT(dev->si_refcount > 0,
931 	    ("devfs: un-referenced struct cdev *(%s)", devtoname(dev)));
932 
933 	switch (com) {
934 	case FIODTYPE:
935 		*(int *)ap->a_data = dsw->d_flags & D_TYPEMASK;
936 		error = 0;
937 		break;
938 	case FIODGNAME:
939 #ifdef	COMPAT_FREEBSD32
940 	case FIODGNAME_32:
941 #endif
942 		fgn = ap->a_data;
943 		p = devtoname(dev);
944 		i = strlen(p) + 1;
945 		if (i > fgn->len)
946 			error = EINVAL;
947 		else
948 			error = copyout(p, fiodgname_buf_get_ptr(fgn, com), i);
949 		break;
950 	default:
951 		error = dsw->d_ioctl(dev, com, ap->a_data, ap->a_fflag, td);
952 	}
953 
954 	dev_relthread(dev, ref);
955 	if (error == ENOIOCTL)
956 		error = ENOTTY;
957 
958 	if (error == 0 && com == TIOCSCTTY) {
959 		/*
960 		 * Do nothing if reassigning same control tty, or if the
961 		 * control tty has already disappeared.  If it disappeared,
962 		 * it's because we were racing with TIOCNOTTY.  TIOCNOTTY
963 		 * already took care of releasing the old vnode and we have
964 		 * nothing left to do.
965 		 */
966 		sx_slock(&proctree_lock);
967 		sess = td->td_proc->p_session;
968 		if (sess->s_ttyvp == vp || sess->s_ttyp == NULL) {
969 			sx_sunlock(&proctree_lock);
970 			return (0);
971 		}
972 
973 		devfs_ctty_ref(vp);
974 		SESS_LOCK(sess);
975 		vpold = sess->s_ttyvp;
976 		sess->s_ttyvp = vp;
977 		sess->s_ttydp = cdev2priv(dev);
978 		SESS_UNLOCK(sess);
979 
980 		sx_sunlock(&proctree_lock);
981 
982 		/* Get rid of reference to old control tty */
983 		if (vpold)
984 			devfs_ctty_unref(vpold);
985 	}
986 	return (error);
987 }
988 
989 /* ARGSUSED */
990 static int
991 devfs_kqfilter_f(struct file *fp, struct knote *kn)
992 {
993 	struct cdev *dev;
994 	struct cdevsw *dsw;
995 	int error, ref;
996 	struct file *fpop;
997 	struct thread *td;
998 
999 	td = curthread;
1000 	fpop = td->td_fpop;
1001 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1002 	if (error)
1003 		return (error);
1004 	error = dsw->d_kqfilter(dev, kn);
1005 	td->td_fpop = fpop;
1006 	dev_relthread(dev, ref);
1007 	return (error);
1008 }
1009 
1010 static inline int
1011 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
1012 {
1013 	struct cdev_priv *cdp;
1014 	struct ucred *dcr;
1015 	struct proc *p;
1016 	int error;
1017 
1018 	cdp = de->de_cdp;
1019 	if (cdp == NULL)
1020 		return (0);
1021 	dcr = cdp->cdp_c.si_cred;
1022 	if (dcr == NULL)
1023 		return (0);
1024 
1025 	error = prison_check(td->td_ucred, dcr);
1026 	if (error == 0)
1027 		return (0);
1028 	/* We do, however, allow access to the controlling terminal */
1029 	p = td->td_proc;
1030 	PROC_LOCK(p);
1031 	if (!(p->p_flag & P_CONTROLT)) {
1032 		PROC_UNLOCK(p);
1033 		return (error);
1034 	}
1035 	if (p->p_session->s_ttydp == cdp)
1036 		error = 0;
1037 	PROC_UNLOCK(p);
1038 	return (error);
1039 }
1040 
1041 static int
1042 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
1043 {
1044 	struct componentname *cnp;
1045 	struct vnode *dvp, **vpp;
1046 	struct thread *td;
1047 	struct devfs_dirent *de, *dd;
1048 	struct devfs_dirent **dde;
1049 	struct devfs_mount *dmp;
1050 	struct mount *mp;
1051 	struct cdev *cdev;
1052 	int error, flags, nameiop, dvplocked;
1053 	char specname[SPECNAMELEN + 1], *pname;
1054 
1055 	td = curthread;
1056 	cnp = ap->a_cnp;
1057 	vpp = ap->a_vpp;
1058 	dvp = ap->a_dvp;
1059 	pname = cnp->cn_nameptr;
1060 	flags = cnp->cn_flags;
1061 	nameiop = cnp->cn_nameiop;
1062 	mp = dvp->v_mount;
1063 	dmp = VFSTODEVFS(mp);
1064 	dd = dvp->v_data;
1065 	*vpp = NULLVP;
1066 
1067 	if ((flags & ISLASTCN) && nameiop == RENAME)
1068 		return (EOPNOTSUPP);
1069 
1070 	if (dvp->v_type != VDIR)
1071 		return (ENOTDIR);
1072 
1073 	if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
1074 		return (EIO);
1075 
1076 	error = vn_dir_check_exec(dvp, cnp);
1077 	if (error != 0)
1078 		return (error);
1079 
1080 	if (cnp->cn_namelen == 1 && *pname == '.') {
1081 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
1082 			return (EINVAL);
1083 		*vpp = dvp;
1084 		VREF(dvp);
1085 		return (0);
1086 	}
1087 
1088 	if (flags & ISDOTDOT) {
1089 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
1090 			return (EINVAL);
1091 		de = devfs_parent_dirent(dd);
1092 		if (de == NULL)
1093 			return (ENOENT);
1094 		dvplocked = VOP_ISLOCKED(dvp);
1095 		VOP_UNLOCK(dvp);
1096 		error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK,
1097 		    vpp);
1098 		*dm_unlock = 0;
1099 		vn_lock(dvp, dvplocked | LK_RETRY);
1100 		return (error);
1101 	}
1102 
1103 	dd = dvp->v_data;
1104 	de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
1105 	while (de == NULL) {	/* While(...) so we can use break */
1106 
1107 		if (nameiop == DELETE)
1108 			return (ENOENT);
1109 
1110 		/*
1111 		 * OK, we didn't have an entry for the name we were asked for
1112 		 * so we try to see if anybody can create it on demand.
1113 		 */
1114 		pname = devfs_fqpn(specname, dmp, dd, cnp);
1115 		if (pname == NULL)
1116 			break;
1117 
1118 		cdev = NULL;
1119 		DEVFS_DMP_HOLD(dmp);
1120 		sx_xunlock(&dmp->dm_lock);
1121 		EVENTHANDLER_INVOKE(dev_clone,
1122 		    td->td_ucred, pname, strlen(pname), &cdev);
1123 
1124 		if (cdev == NULL)
1125 			sx_xlock(&dmp->dm_lock);
1126 		else if (devfs_populate_vp(dvp) != 0) {
1127 			*dm_unlock = 0;
1128 			sx_xlock(&dmp->dm_lock);
1129 			if (DEVFS_DMP_DROP(dmp)) {
1130 				sx_xunlock(&dmp->dm_lock);
1131 				devfs_unmount_final(dmp);
1132 			} else
1133 				sx_xunlock(&dmp->dm_lock);
1134 			dev_rel(cdev);
1135 			return (ENOENT);
1136 		}
1137 		if (DEVFS_DMP_DROP(dmp)) {
1138 			*dm_unlock = 0;
1139 			sx_xunlock(&dmp->dm_lock);
1140 			devfs_unmount_final(dmp);
1141 			if (cdev != NULL)
1142 				dev_rel(cdev);
1143 			return (ENOENT);
1144 		}
1145 
1146 		if (cdev == NULL)
1147 			break;
1148 
1149 		dev_lock();
1150 		dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
1151 		if (dde != NULL && *dde != NULL)
1152 			de = *dde;
1153 		dev_unlock();
1154 		dev_rel(cdev);
1155 		break;
1156 	}
1157 
1158 	if (de == NULL || de->de_flags & DE_WHITEOUT) {
1159 		if ((nameiop == CREATE || nameiop == RENAME) &&
1160 		    (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
1161 			return (EJUSTRETURN);
1162 		}
1163 		return (ENOENT);
1164 	}
1165 
1166 	if (devfs_prison_check(de, td))
1167 		return (ENOENT);
1168 
1169 	if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
1170 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
1171 		if (error)
1172 			return (error);
1173 		if (*vpp == dvp) {
1174 			VREF(dvp);
1175 			*vpp = dvp;
1176 			return (0);
1177 		}
1178 	}
1179 	error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, vpp);
1180 	*dm_unlock = 0;
1181 	return (error);
1182 }
1183 
1184 static int
1185 devfs_lookup(struct vop_lookup_args *ap)
1186 {
1187 	int j;
1188 	struct devfs_mount *dmp;
1189 	int dm_unlock;
1190 
1191 	if (devfs_populate_vp(ap->a_dvp) != 0)
1192 		return (ENOTDIR);
1193 
1194 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1195 	dm_unlock = 1;
1196 	j = devfs_lookupx(ap, &dm_unlock);
1197 	if (dm_unlock == 1)
1198 		sx_xunlock(&dmp->dm_lock);
1199 	return (j);
1200 }
1201 
1202 static int
1203 devfs_mknod(struct vop_mknod_args *ap)
1204 {
1205 	struct componentname *cnp;
1206 	struct vnode *dvp, **vpp;
1207 	struct devfs_dirent *dd, *de;
1208 	struct devfs_mount *dmp;
1209 	int error;
1210 
1211 	/*
1212 	 * The only type of node we should be creating here is a
1213 	 * character device, for anything else return EOPNOTSUPP.
1214 	 */
1215 	if (ap->a_vap->va_type != VCHR)
1216 		return (EOPNOTSUPP);
1217 	dvp = ap->a_dvp;
1218 	dmp = VFSTODEVFS(dvp->v_mount);
1219 
1220 	cnp = ap->a_cnp;
1221 	vpp = ap->a_vpp;
1222 	dd = dvp->v_data;
1223 
1224 	error = ENOENT;
1225 	sx_xlock(&dmp->dm_lock);
1226 	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1227 		if (cnp->cn_namelen != de->de_dirent->d_namlen)
1228 			continue;
1229 		if (de->de_dirent->d_type == DT_CHR &&
1230 		    (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0)
1231 			continue;
1232 		if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1233 		    de->de_dirent->d_namlen) != 0)
1234 			continue;
1235 		if (de->de_flags & DE_WHITEOUT)
1236 			break;
1237 		goto notfound;
1238 	}
1239 	if (de == NULL)
1240 		goto notfound;
1241 	de->de_flags &= ~DE_WHITEOUT;
1242 	error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1243 	return (error);
1244 notfound:
1245 	sx_xunlock(&dmp->dm_lock);
1246 	return (error);
1247 }
1248 
1249 /* ARGSUSED */
1250 static int
1251 devfs_open(struct vop_open_args *ap)
1252 {
1253 	struct thread *td = ap->a_td;
1254 	struct vnode *vp = ap->a_vp;
1255 	struct cdev *dev = vp->v_rdev;
1256 	struct file *fp = ap->a_fp;
1257 	int error, ref, vlocked;
1258 	struct cdevsw *dsw;
1259 	struct file *fpop;
1260 
1261 	if (vp->v_type == VBLK)
1262 		return (ENXIO);
1263 
1264 	if (dev == NULL)
1265 		return (ENXIO);
1266 
1267 	/* Make this field valid before any I/O in d_open. */
1268 	if (dev->si_iosize_max == 0)
1269 		dev->si_iosize_max = DFLTPHYS;
1270 
1271 	dsw = dev_refthread(dev, &ref);
1272 	if (dsw == NULL)
1273 		return (ENXIO);
1274 	if (fp == NULL && dsw->d_fdopen != NULL) {
1275 		dev_relthread(dev, ref);
1276 		return (ENXIO);
1277 	}
1278 
1279 	if (vp->v_type == VCHR)
1280 		devfs_usecount_add(vp);
1281 
1282 	vlocked = VOP_ISLOCKED(vp);
1283 	VOP_UNLOCK(vp);
1284 
1285 	fpop = td->td_fpop;
1286 	td->td_fpop = fp;
1287 	if (fp != NULL) {
1288 		fp->f_data = dev;
1289 		fp->f_vnode = vp;
1290 	}
1291 	if (dsw->d_fdopen != NULL)
1292 		error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1293 	else
1294 		error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1295 	/* Clean up any cdevpriv upon error. */
1296 	if (error != 0)
1297 		devfs_clear_cdevpriv();
1298 	td->td_fpop = fpop;
1299 
1300 	vn_lock(vp, vlocked | LK_RETRY);
1301 	if (error != 0 && vp->v_type == VCHR)
1302 		devfs_usecount_sub(vp);
1303 
1304 	dev_relthread(dev, ref);
1305 	if (error != 0) {
1306 		if (error == ERESTART)
1307 			error = EINTR;
1308 		return (error);
1309 	}
1310 
1311 #if 0	/* /dev/console */
1312 	KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1313 #else
1314 	if (fp == NULL)
1315 		return (error);
1316 #endif
1317 	if (fp->f_ops == &badfileops)
1318 		finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1319 	return (error);
1320 }
1321 
1322 static int
1323 devfs_pathconf(struct vop_pathconf_args *ap)
1324 {
1325 
1326 	switch (ap->a_name) {
1327 	case _PC_FILESIZEBITS:
1328 		*ap->a_retval = 64;
1329 		return (0);
1330 	case _PC_NAME_MAX:
1331 		*ap->a_retval = NAME_MAX;
1332 		return (0);
1333 	case _PC_LINK_MAX:
1334 		*ap->a_retval = INT_MAX;
1335 		return (0);
1336 	case _PC_SYMLINK_MAX:
1337 		*ap->a_retval = MAXPATHLEN;
1338 		return (0);
1339 	case _PC_MAX_CANON:
1340 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1341 			*ap->a_retval = MAX_CANON;
1342 			return (0);
1343 		}
1344 		return (EINVAL);
1345 	case _PC_MAX_INPUT:
1346 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1347 			*ap->a_retval = MAX_INPUT;
1348 			return (0);
1349 		}
1350 		return (EINVAL);
1351 	case _PC_VDISABLE:
1352 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1353 			*ap->a_retval = _POSIX_VDISABLE;
1354 			return (0);
1355 		}
1356 		return (EINVAL);
1357 	case _PC_MAC_PRESENT:
1358 #ifdef MAC
1359 		/*
1360 		 * If MAC is enabled, devfs automatically supports
1361 		 * trivial non-persistent label storage.
1362 		 */
1363 		*ap->a_retval = 1;
1364 #else
1365 		*ap->a_retval = 0;
1366 #endif
1367 		return (0);
1368 	case _PC_CHOWN_RESTRICTED:
1369 		*ap->a_retval = 1;
1370 		return (0);
1371 	default:
1372 		return (vop_stdpathconf(ap));
1373 	}
1374 	/* NOTREACHED */
1375 }
1376 
1377 /* ARGSUSED */
1378 static int
1379 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1380 {
1381 	struct cdev *dev;
1382 	struct cdevsw *dsw;
1383 	int error, ref;
1384 	struct file *fpop;
1385 
1386 	fpop = td->td_fpop;
1387 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1388 	if (error != 0) {
1389 		error = vnops.fo_poll(fp, events, cred, td);
1390 		return (error);
1391 	}
1392 	error = dsw->d_poll(dev, events, td);
1393 	td->td_fpop = fpop;
1394 	dev_relthread(dev, ref);
1395 	return(error);
1396 }
1397 
1398 /*
1399  * Print out the contents of a special device vnode.
1400  */
1401 static int
1402 devfs_print(struct vop_print_args *ap)
1403 {
1404 
1405 	printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1406 	return (0);
1407 }
1408 
1409 static int
1410 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
1411     int flags, struct thread *td)
1412 {
1413 	struct cdev *dev;
1414 	int ioflag, error, ref;
1415 	ssize_t resid;
1416 	struct cdevsw *dsw;
1417 	struct file *fpop;
1418 
1419 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1420 		return (EINVAL);
1421 	fpop = td->td_fpop;
1422 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1423 	if (error != 0) {
1424 		error = vnops.fo_read(fp, uio, cred, flags, td);
1425 		return (error);
1426 	}
1427 	resid = uio->uio_resid;
1428 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1429 	if (ioflag & O_DIRECT)
1430 		ioflag |= IO_DIRECT;
1431 
1432 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1433 	error = dsw->d_read(dev, uio, ioflag);
1434 	if (uio->uio_resid != resid || (error == 0 && resid != 0))
1435 		devfs_timestamp(&dev->si_atime);
1436 	td->td_fpop = fpop;
1437 	dev_relthread(dev, ref);
1438 
1439 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF_R);
1440 	return (error);
1441 }
1442 
1443 static int
1444 devfs_readdir(struct vop_readdir_args *ap)
1445 {
1446 	int error;
1447 	struct uio *uio;
1448 	struct dirent *dp;
1449 	struct devfs_dirent *dd;
1450 	struct devfs_dirent *de;
1451 	struct devfs_mount *dmp;
1452 	off_t off;
1453 	int *tmp_ncookies = NULL;
1454 
1455 	if (ap->a_vp->v_type != VDIR)
1456 		return (ENOTDIR);
1457 
1458 	uio = ap->a_uio;
1459 	if (uio->uio_offset < 0)
1460 		return (EINVAL);
1461 
1462 	/*
1463 	 * XXX: This is a temporary hack to get around this filesystem not
1464 	 * supporting cookies. We store the location of the ncookies pointer
1465 	 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1466 	 * and set the number of cookies to 0. We then set the pointer to
1467 	 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1468 	 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1469 	 * pointer to its original location before returning to the caller.
1470 	 */
1471 	if (ap->a_ncookies != NULL) {
1472 		tmp_ncookies = ap->a_ncookies;
1473 		*ap->a_ncookies = 0;
1474 		ap->a_ncookies = NULL;
1475 	}
1476 
1477 	dmp = VFSTODEVFS(ap->a_vp->v_mount);
1478 	if (devfs_populate_vp(ap->a_vp) != 0) {
1479 		if (tmp_ncookies != NULL)
1480 			ap->a_ncookies = tmp_ncookies;
1481 		return (EIO);
1482 	}
1483 	error = 0;
1484 	de = ap->a_vp->v_data;
1485 	off = 0;
1486 	TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1487 		KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1488 		if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1489 			continue;
1490 		if (devfs_prison_check(dd, uio->uio_td))
1491 			continue;
1492 		if (dd->de_dirent->d_type == DT_DIR)
1493 			de = dd->de_dir;
1494 		else
1495 			de = dd;
1496 		dp = dd->de_dirent;
1497 		MPASS(dp->d_reclen == GENERIC_DIRSIZ(dp));
1498 		if (dp->d_reclen > uio->uio_resid)
1499 			break;
1500 		dp->d_fileno = de->de_inode;
1501 		/* NOTE: d_off is the offset for the *next* entry. */
1502 		dp->d_off = off + dp->d_reclen;
1503 		if (off >= uio->uio_offset) {
1504 			error = vfs_read_dirent(ap, dp, off);
1505 			if (error)
1506 				break;
1507 		}
1508 		off += dp->d_reclen;
1509 	}
1510 	sx_xunlock(&dmp->dm_lock);
1511 	uio->uio_offset = off;
1512 
1513 	/*
1514 	 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1515 	 * place.
1516 	 */
1517 	if (tmp_ncookies != NULL)
1518 		ap->a_ncookies = tmp_ncookies;
1519 
1520 	return (error);
1521 }
1522 
1523 static int
1524 devfs_readlink(struct vop_readlink_args *ap)
1525 {
1526 	struct devfs_dirent *de;
1527 
1528 	de = ap->a_vp->v_data;
1529 	return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1530 }
1531 
1532 static void
1533 devfs_reclaiml(struct vnode *vp)
1534 {
1535 	struct devfs_dirent *de;
1536 
1537 	mtx_assert(&devfs_de_interlock, MA_OWNED);
1538 	de = vp->v_data;
1539 	if (de != NULL) {
1540 		MPASS(de->de_usecount == 0);
1541 		de->de_vnode = NULL;
1542 		vp->v_data = NULL;
1543 	}
1544 }
1545 
1546 static int
1547 devfs_reclaim(struct vop_reclaim_args *ap)
1548 {
1549 	struct vnode *vp;
1550 
1551 	vp = ap->a_vp;
1552 	mtx_lock(&devfs_de_interlock);
1553 	devfs_reclaiml(vp);
1554 	mtx_unlock(&devfs_de_interlock);
1555 	return (0);
1556 }
1557 
1558 static int
1559 devfs_reclaim_vchr(struct vop_reclaim_args *ap)
1560 {
1561 	struct vnode *vp;
1562 	struct cdev *dev;
1563 
1564 	vp = ap->a_vp;
1565 	MPASS(vp->v_type == VCHR);
1566 
1567 	mtx_lock(&devfs_de_interlock);
1568 	VI_LOCK(vp);
1569 	devfs_usecount_subl(vp);
1570 	devfs_reclaiml(vp);
1571 	mtx_unlock(&devfs_de_interlock);
1572 	dev_lock();
1573 	dev = vp->v_rdev;
1574 	vp->v_rdev = NULL;
1575 	dev_unlock();
1576 	VI_UNLOCK(vp);
1577 	if (dev != NULL)
1578 		dev_rel(dev);
1579 	return (0);
1580 }
1581 
1582 static int
1583 devfs_remove(struct vop_remove_args *ap)
1584 {
1585 	struct vnode *dvp = ap->a_dvp;
1586 	struct vnode *vp = ap->a_vp;
1587 	struct devfs_dirent *dd;
1588 	struct devfs_dirent *de, *de_covered;
1589 	struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1590 
1591 	ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1592 	ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1593 
1594 	sx_xlock(&dmp->dm_lock);
1595 	dd = ap->a_dvp->v_data;
1596 	de = vp->v_data;
1597 	if (de->de_cdp == NULL) {
1598 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1599 		if (de->de_dirent->d_type == DT_LNK) {
1600 			de_covered = devfs_find(dd, de->de_dirent->d_name,
1601 			    de->de_dirent->d_namlen, 0);
1602 			if (de_covered != NULL)
1603 				de_covered->de_flags &= ~DE_COVERED;
1604 		}
1605 		/* We need to unlock dvp because devfs_delete() may lock it. */
1606 		VOP_UNLOCK(vp);
1607 		if (dvp != vp)
1608 			VOP_UNLOCK(dvp);
1609 		devfs_delete(dmp, de, 0);
1610 		sx_xunlock(&dmp->dm_lock);
1611 		if (dvp != vp)
1612 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1613 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1614 	} else {
1615 		de->de_flags |= DE_WHITEOUT;
1616 		sx_xunlock(&dmp->dm_lock);
1617 	}
1618 	return (0);
1619 }
1620 
1621 /*
1622  * Revoke is called on a tty when a terminal session ends.  The vnode
1623  * is orphaned by setting v_op to deadfs so we need to let go of it
1624  * as well so that we create a new one next time around.
1625  *
1626  */
1627 static int
1628 devfs_revoke(struct vop_revoke_args *ap)
1629 {
1630 	struct vnode *vp = ap->a_vp, *vp2;
1631 	struct cdev *dev;
1632 	struct cdev_priv *cdp;
1633 	struct devfs_dirent *de;
1634 	enum vgetstate vs;
1635 	u_int i;
1636 
1637 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1638 
1639 	dev = vp->v_rdev;
1640 	cdp = cdev2priv(dev);
1641 
1642 	dev_lock();
1643 	cdp->cdp_inuse++;
1644 	dev_unlock();
1645 
1646 	vhold(vp);
1647 	vgone(vp);
1648 	vdrop(vp);
1649 
1650 	VOP_UNLOCK(vp);
1651  loop:
1652 	for (;;) {
1653 		mtx_lock(&devfs_de_interlock);
1654 		dev_lock();
1655 		vp2 = NULL;
1656 		for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1657 			de = cdp->cdp_dirents[i];
1658 			if (de == NULL)
1659 				continue;
1660 
1661 			vp2 = de->de_vnode;
1662 			if (vp2 != NULL) {
1663 				dev_unlock();
1664 				vs = vget_prep(vp2);
1665 				mtx_unlock(&devfs_de_interlock);
1666 				if (vget_finish(vp2, LK_EXCLUSIVE, vs) != 0)
1667 					goto loop;
1668 				vhold(vp2);
1669 				vgone(vp2);
1670 				vdrop(vp2);
1671 				vput(vp2);
1672 				break;
1673 			}
1674 		}
1675 		if (vp2 != NULL) {
1676 			continue;
1677 		}
1678 		dev_unlock();
1679 		mtx_unlock(&devfs_de_interlock);
1680 		break;
1681 	}
1682 	dev_lock();
1683 	cdp->cdp_inuse--;
1684 	if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1685 		KASSERT((cdp->cdp_flags & CDP_ON_ACTIVE_LIST) != 0,
1686 		    ("%s: cdp %p (%s) not on active list",
1687 		    __func__, cdp, dev->si_name));
1688 		cdp->cdp_flags &= ~CDP_ON_ACTIVE_LIST;
1689 		TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1690 		dev_unlock();
1691 		dev_rel(&cdp->cdp_c);
1692 	} else
1693 		dev_unlock();
1694 
1695 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1696 	return (0);
1697 }
1698 
1699 static int
1700 devfs_rioctl(struct vop_ioctl_args *ap)
1701 {
1702 	struct vnode *vp;
1703 	struct devfs_mount *dmp;
1704 	int error;
1705 
1706 	vp = ap->a_vp;
1707 	vn_lock(vp, LK_SHARED | LK_RETRY);
1708 	if (VN_IS_DOOMED(vp)) {
1709 		VOP_UNLOCK(vp);
1710 		return (EBADF);
1711 	}
1712 	dmp = VFSTODEVFS(vp->v_mount);
1713 	sx_xlock(&dmp->dm_lock);
1714 	VOP_UNLOCK(vp);
1715 	DEVFS_DMP_HOLD(dmp);
1716 	devfs_populate(dmp);
1717 	if (DEVFS_DMP_DROP(dmp)) {
1718 		sx_xunlock(&dmp->dm_lock);
1719 		devfs_unmount_final(dmp);
1720 		return (ENOENT);
1721 	}
1722 	error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1723 	sx_xunlock(&dmp->dm_lock);
1724 	return (error);
1725 }
1726 
1727 static int
1728 devfs_rread(struct vop_read_args *ap)
1729 {
1730 
1731 	if (ap->a_vp->v_type != VDIR)
1732 		return (EINVAL);
1733 	return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1734 }
1735 
1736 static int
1737 devfs_setattr(struct vop_setattr_args *ap)
1738 {
1739 	struct devfs_dirent *de;
1740 	struct vattr *vap;
1741 	struct vnode *vp;
1742 	struct thread *td;
1743 	int c, error;
1744 	uid_t uid;
1745 	gid_t gid;
1746 
1747 	vap = ap->a_vap;
1748 	vp = ap->a_vp;
1749 	td = curthread;
1750 	if ((vap->va_type != VNON) ||
1751 	    (vap->va_nlink != VNOVAL) ||
1752 	    (vap->va_fsid != VNOVAL) ||
1753 	    (vap->va_fileid != VNOVAL) ||
1754 	    (vap->va_blocksize != VNOVAL) ||
1755 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1756 	    (vap->va_rdev != VNOVAL) ||
1757 	    ((int)vap->va_bytes != VNOVAL) ||
1758 	    (vap->va_gen != VNOVAL)) {
1759 		return (EINVAL);
1760 	}
1761 
1762 	error = devfs_populate_vp(vp);
1763 	if (error != 0)
1764 		return (error);
1765 
1766 	de = vp->v_data;
1767 	if (vp->v_type == VDIR)
1768 		de = de->de_dir;
1769 
1770 	c = 0;
1771 	if (vap->va_uid == (uid_t)VNOVAL)
1772 		uid = de->de_uid;
1773 	else
1774 		uid = vap->va_uid;
1775 	if (vap->va_gid == (gid_t)VNOVAL)
1776 		gid = de->de_gid;
1777 	else
1778 		gid = vap->va_gid;
1779 	if (uid != de->de_uid || gid != de->de_gid) {
1780 		if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1781 		    (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1782 			error = priv_check(td, PRIV_VFS_CHOWN);
1783 			if (error != 0)
1784 				goto ret;
1785 		}
1786 		de->de_uid = uid;
1787 		de->de_gid = gid;
1788 		c = 1;
1789 	}
1790 
1791 	if (vap->va_mode != (mode_t)VNOVAL) {
1792 		if (ap->a_cred->cr_uid != de->de_uid) {
1793 			error = priv_check(td, PRIV_VFS_ADMIN);
1794 			if (error != 0)
1795 				goto ret;
1796 		}
1797 		de->de_mode = vap->va_mode;
1798 		c = 1;
1799 	}
1800 
1801 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1802 		error = vn_utimes_perm(vp, vap, ap->a_cred, td);
1803 		if (error != 0)
1804 			goto ret;
1805 		if (vap->va_atime.tv_sec != VNOVAL) {
1806 			if (vp->v_type == VCHR)
1807 				vp->v_rdev->si_atime = vap->va_atime;
1808 			else
1809 				de->de_atime = vap->va_atime;
1810 		}
1811 		if (vap->va_mtime.tv_sec != VNOVAL) {
1812 			if (vp->v_type == VCHR)
1813 				vp->v_rdev->si_mtime = vap->va_mtime;
1814 			else
1815 				de->de_mtime = vap->va_mtime;
1816 		}
1817 		c = 1;
1818 	}
1819 
1820 	if (c) {
1821 		if (vp->v_type == VCHR)
1822 			vfs_timestamp(&vp->v_rdev->si_ctime);
1823 		else
1824 			vfs_timestamp(&de->de_mtime);
1825 	}
1826 
1827 ret:
1828 	sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock);
1829 	return (error);
1830 }
1831 
1832 #ifdef MAC
1833 static int
1834 devfs_setlabel(struct vop_setlabel_args *ap)
1835 {
1836 	struct vnode *vp;
1837 	struct devfs_dirent *de;
1838 
1839 	vp = ap->a_vp;
1840 	de = vp->v_data;
1841 
1842 	mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1843 	mac_devfs_update(vp->v_mount, de, vp);
1844 
1845 	return (0);
1846 }
1847 #endif
1848 
1849 static int
1850 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred)
1851 {
1852 
1853 	return (vnops.fo_stat(fp, sb, cred));
1854 }
1855 
1856 static int
1857 devfs_symlink(struct vop_symlink_args *ap)
1858 {
1859 	int i, error;
1860 	struct devfs_dirent *dd;
1861 	struct devfs_dirent *de, *de_covered, *de_dotdot;
1862 	struct devfs_mount *dmp;
1863 
1864 	error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1865 	if (error)
1866 		return(error);
1867 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1868 	if (devfs_populate_vp(ap->a_dvp) != 0)
1869 		return (ENOENT);
1870 
1871 	dd = ap->a_dvp->v_data;
1872 	de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1873 	de->de_flags = DE_USER;
1874 	de->de_uid = 0;
1875 	de->de_gid = 0;
1876 	de->de_mode = 0755;
1877 	de->de_inode = alloc_unr(devfs_inos);
1878 	de->de_dir = dd;
1879 	de->de_dirent->d_type = DT_LNK;
1880 	i = strlen(ap->a_target) + 1;
1881 	de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1882 	bcopy(ap->a_target, de->de_symlink, i);
1883 #ifdef MAC
1884 	mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1885 #endif
1886 	de_covered = devfs_find(dd, de->de_dirent->d_name,
1887 	    de->de_dirent->d_namlen, 0);
1888 	if (de_covered != NULL) {
1889 		if ((de_covered->de_flags & DE_USER) != 0) {
1890 			devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1891 			sx_xunlock(&dmp->dm_lock);
1892 			return (EEXIST);
1893 		}
1894 		KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1895 		    ("devfs_symlink: entry %p already covered", de_covered));
1896 		de_covered->de_flags |= DE_COVERED;
1897 	}
1898 
1899 	de_dotdot = TAILQ_FIRST(&dd->de_dlist);		/* "." */
1900 	de_dotdot = TAILQ_NEXT(de_dotdot, de_list);	/* ".." */
1901 	TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1902 	devfs_dir_ref_de(dmp, dd);
1903 	devfs_rules_apply(dmp, de);
1904 
1905 	return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1906 }
1907 
1908 static int
1909 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1910 {
1911 
1912 	return (vnops.fo_truncate(fp, length, cred, td));
1913 }
1914 
1915 static int
1916 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
1917     int flags, struct thread *td)
1918 {
1919 	struct cdev *dev;
1920 	int error, ioflag, ref;
1921 	ssize_t resid;
1922 	struct cdevsw *dsw;
1923 	struct file *fpop;
1924 
1925 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1926 		return (EINVAL);
1927 	fpop = td->td_fpop;
1928 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1929 	if (error != 0) {
1930 		error = vnops.fo_write(fp, uio, cred, flags, td);
1931 		return (error);
1932 	}
1933 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1934 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1935 	if (ioflag & O_DIRECT)
1936 		ioflag |= IO_DIRECT;
1937 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1938 
1939 	resid = uio->uio_resid;
1940 
1941 	error = dsw->d_write(dev, uio, ioflag);
1942 	if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1943 		devfs_timestamp(&dev->si_ctime);
1944 		dev->si_mtime = dev->si_ctime;
1945 	}
1946 	td->td_fpop = fpop;
1947 	dev_relthread(dev, ref);
1948 
1949 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF_W);
1950 	return (error);
1951 }
1952 
1953 static int
1954 devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1955     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1956     struct thread *td)
1957 {
1958 	struct cdev *dev;
1959 	struct cdevsw *dsw;
1960 	struct mount *mp;
1961 	struct vnode *vp;
1962 	struct file *fpop;
1963 	vm_object_t object;
1964 	vm_prot_t maxprot;
1965 	int error, ref;
1966 
1967 	vp = fp->f_vnode;
1968 
1969 	/*
1970 	 * Ensure that file and memory protections are
1971 	 * compatible.
1972 	 */
1973 	mp = vp->v_mount;
1974 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1975 		maxprot = VM_PROT_NONE;
1976 		if ((prot & VM_PROT_EXECUTE) != 0)
1977 			return (EACCES);
1978 	} else
1979 		maxprot = VM_PROT_EXECUTE;
1980 	if ((fp->f_flag & FREAD) != 0)
1981 		maxprot |= VM_PROT_READ;
1982 	else if ((prot & VM_PROT_READ) != 0)
1983 		return (EACCES);
1984 
1985 	/*
1986 	 * If we are sharing potential changes via MAP_SHARED and we
1987 	 * are trying to get write permission although we opened it
1988 	 * without asking for it, bail out.
1989 	 *
1990 	 * Note that most character devices always share mappings.
1991 	 * The one exception is that D_MMAP_ANON devices
1992 	 * (i.e. /dev/zero) permit private writable mappings.
1993 	 *
1994 	 * Rely on vm_mmap_cdev() to fail invalid MAP_PRIVATE requests
1995 	 * as well as updating maxprot to permit writing for
1996 	 * D_MMAP_ANON devices rather than doing that here.
1997 	 */
1998 	if ((flags & MAP_SHARED) != 0) {
1999 		if ((fp->f_flag & FWRITE) != 0)
2000 			maxprot |= VM_PROT_WRITE;
2001 		else if ((prot & VM_PROT_WRITE) != 0)
2002 			return (EACCES);
2003 	}
2004 	maxprot &= cap_maxprot;
2005 
2006 	fpop = td->td_fpop;
2007 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
2008 	if (error != 0)
2009 		return (error);
2010 
2011 	error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, dev, dsw, &foff,
2012 	    &object);
2013 	td->td_fpop = fpop;
2014 	dev_relthread(dev, ref);
2015 	if (error != 0)
2016 		return (error);
2017 
2018 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2019 	    foff, FALSE, td);
2020 	if (error != 0)
2021 		vm_object_deallocate(object);
2022 	return (error);
2023 }
2024 
2025 dev_t
2026 dev2udev(struct cdev *x)
2027 {
2028 	if (x == NULL)
2029 		return (NODEV);
2030 	return (cdev2priv(x)->cdp_inode);
2031 }
2032 
2033 static int
2034 devfs_cmp_f(struct file *fp1, struct file *fp2, struct thread *td)
2035 {
2036 	if (fp2->f_type != DTYPE_VNODE || fp2->f_ops != &devfs_ops_f)
2037 		return (3);
2038 	return (kcmp_cmp((uintptr_t)fp1->f_data, (uintptr_t)fp2->f_data));
2039 }
2040 
2041 static struct fileops devfs_ops_f = {
2042 	.fo_read =	devfs_read_f,
2043 	.fo_write =	devfs_write_f,
2044 	.fo_truncate =	devfs_truncate_f,
2045 	.fo_ioctl =	devfs_ioctl_f,
2046 	.fo_poll =	devfs_poll_f,
2047 	.fo_kqfilter =	devfs_kqfilter_f,
2048 	.fo_stat =	devfs_stat_f,
2049 	.fo_close =	devfs_close_f,
2050 	.fo_chmod =	vn_chmod,
2051 	.fo_chown =	vn_chown,
2052 	.fo_sendfile =	vn_sendfile,
2053 	.fo_seek =	vn_seek,
2054 	.fo_fill_kinfo = vn_fill_kinfo,
2055 	.fo_mmap =	devfs_mmap_f,
2056 	.fo_cmp =	devfs_cmp_f,
2057 	.fo_flags =	DFLAG_PASSABLE | DFLAG_SEEKABLE
2058 };
2059 
2060 /* Vops for non-CHR vnodes in /dev. */
2061 static struct vop_vector devfs_vnodeops = {
2062 	.vop_default =		&default_vnodeops,
2063 
2064 	.vop_access =		devfs_access,
2065 	.vop_getattr =		devfs_getattr,
2066 	.vop_ioctl =		devfs_rioctl,
2067 	.vop_lookup =		devfs_lookup,
2068 	.vop_mknod =		devfs_mknod,
2069 	.vop_pathconf =		devfs_pathconf,
2070 	.vop_read =		devfs_rread,
2071 	.vop_readdir =		devfs_readdir,
2072 	.vop_readlink =		devfs_readlink,
2073 	.vop_reclaim =		devfs_reclaim,
2074 	.vop_remove =		devfs_remove,
2075 	.vop_revoke =		devfs_revoke,
2076 	.vop_setattr =		devfs_setattr,
2077 #ifdef MAC
2078 	.vop_setlabel =		devfs_setlabel,
2079 #endif
2080 	.vop_symlink =		devfs_symlink,
2081 	.vop_vptocnp =		devfs_vptocnp,
2082 	.vop_lock1 =		vop_lock,
2083 	.vop_unlock =		vop_unlock,
2084 	.vop_islocked =		vop_islocked,
2085 	.vop_add_writecount =	vop_stdadd_writecount_nomsync,
2086 };
2087 VFS_VOP_VECTOR_REGISTER(devfs_vnodeops);
2088 
2089 /* Vops for VCHR vnodes in /dev. */
2090 static struct vop_vector devfs_specops = {
2091 	.vop_default =		&default_vnodeops,
2092 
2093 	.vop_access =		devfs_access,
2094 	.vop_bmap =		VOP_PANIC,
2095 	.vop_close =		devfs_close,
2096 	.vop_create =		VOP_PANIC,
2097 	.vop_fsync =		vop_stdfsync,
2098 	.vop_getattr =		devfs_getattr,
2099 	.vop_ioctl =		devfs_ioctl,
2100 	.vop_link =		VOP_PANIC,
2101 	.vop_mkdir =		VOP_PANIC,
2102 	.vop_mknod =		VOP_PANIC,
2103 	.vop_open =		devfs_open,
2104 	.vop_pathconf =		devfs_pathconf,
2105 	.vop_poll =		dead_poll,
2106 	.vop_print =		devfs_print,
2107 	.vop_read =		dead_read,
2108 	.vop_readdir =		VOP_PANIC,
2109 	.vop_readlink =		VOP_PANIC,
2110 	.vop_reallocblks =	VOP_PANIC,
2111 	.vop_reclaim =		devfs_reclaim_vchr,
2112 	.vop_remove =		devfs_remove,
2113 	.vop_rename =		VOP_PANIC,
2114 	.vop_revoke =		devfs_revoke,
2115 	.vop_rmdir =		VOP_PANIC,
2116 	.vop_setattr =		devfs_setattr,
2117 #ifdef MAC
2118 	.vop_setlabel =		devfs_setlabel,
2119 #endif
2120 	.vop_strategy =		VOP_PANIC,
2121 	.vop_symlink =		VOP_PANIC,
2122 	.vop_vptocnp =		devfs_vptocnp,
2123 	.vop_write =		dead_write,
2124 	.vop_lock1 =		vop_lock,
2125 	.vop_unlock =		vop_unlock,
2126 	.vop_islocked =		vop_islocked,
2127 	.vop_add_writecount =	vop_stdadd_writecount_nomsync,
2128 };
2129 VFS_VOP_VECTOR_REGISTER(devfs_specops);
2130 
2131 /*
2132  * Our calling convention to the device drivers used to be that we passed
2133  * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
2134  * flags instead since that's what open(), close() and ioctl() takes and
2135  * we don't really want vnode.h in device drivers.
2136  * We solved the source compatibility by redefining some vnode flags to
2137  * be the same as the fcntl ones and by sending down the bitwise OR of
2138  * the respective fcntl/vnode flags.  These CTASSERTS make sure nobody
2139  * pulls the rug out under this.
2140  */
2141 CTASSERT(O_NONBLOCK == IO_NDELAY);
2142 CTASSERT(O_FSYNC == IO_SYNC);
2143