xref: /freebsd-14.2/sys/fs/devfs/devfs_vnops.c (revision aef45041)
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  *
33  *	@(#)kernfs_vnops.c	8.15 (Berkeley) 5/21/95
34  * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vnops.c 1.43
35  */
36 
37 /*
38  * TODO:
39  *	mkdir: want it ?
40  */
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/conf.h>
45 #include <sys/dirent.h>
46 #include <sys/eventhandler.h>
47 #include <sys/fcntl.h>
48 #include <sys/file.h>
49 #include <sys/filedesc.h>
50 #include <sys/filio.h>
51 #include <sys/jail.h>
52 #include <sys/kernel.h>
53 #include <sys/limits.h>
54 #include <sys/lock.h>
55 #include <sys/malloc.h>
56 #include <sys/mman.h>
57 #include <sys/mount.h>
58 #include <sys/namei.h>
59 #include <sys/priv.h>
60 #include <sys/proc.h>
61 #include <sys/stat.h>
62 #include <sys/sx.h>
63 #include <sys/sysctl.h>
64 #include <sys/time.h>
65 #include <sys/ttycom.h>
66 #include <sys/unistd.h>
67 #include <sys/vnode.h>
68 
69 static struct vop_vector devfs_vnodeops;
70 static struct vop_vector devfs_specops;
71 static struct fileops devfs_ops_f;
72 
73 #include <fs/devfs/devfs.h>
74 #include <fs/devfs/devfs_int.h>
75 
76 #include <security/mac/mac_framework.h>
77 
78 #include <vm/vm.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_object.h>
81 
82 static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data");
83 
84 struct mtx	devfs_de_interlock;
85 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF);
86 struct mtx	cdevpriv_mtx;
87 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
88 
89 SYSCTL_DECL(_vfs_devfs);
90 
91 static int devfs_dotimes;
92 SYSCTL_INT(_vfs_devfs, OID_AUTO, dotimes, CTLFLAG_RW,
93     &devfs_dotimes, 0, "Update timestamps on DEVFS with default precision");
94 
95 /*
96  * Update devfs node timestamp.  Note that updates are unlocked and
97  * stat(2) could see partially updated times.
98  */
99 static void
100 devfs_timestamp(struct timespec *tsp)
101 {
102 	time_t ts;
103 
104 	if (devfs_dotimes) {
105 		vfs_timestamp(tsp);
106 	} else {
107 		ts = time_second;
108 		if (tsp->tv_sec != ts) {
109 			tsp->tv_sec = ts;
110 			tsp->tv_nsec = 0;
111 		}
112 	}
113 }
114 
115 static int
116 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp,
117     int *ref)
118 {
119 	*dswp = devvn_refthread(fp->f_vnode, devp, ref);
120 	if (*dswp == NULL || *devp != fp->f_data) {
121 		if (*dswp != NULL)
122 			dev_relthread(*devp, *ref);
123 		return (ENXIO);
124 	}
125 	KASSERT((*devp)->si_refcount > 0,
126 	    ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
127 	if (*dswp == NULL)
128 		return (ENXIO);
129 	curthread->td_fpop = fp;
130 	return (0);
131 }
132 
133 int
134 devfs_get_cdevpriv(void **datap)
135 {
136 	struct file *fp;
137 	struct cdev_privdata *p;
138 	int error;
139 
140 	fp = curthread->td_fpop;
141 	if (fp == NULL)
142 		return (EBADF);
143 	p = fp->f_cdevpriv;
144 	if (p != NULL) {
145 		error = 0;
146 		*datap = p->cdpd_data;
147 	} else
148 		error = ENOENT;
149 	return (error);
150 }
151 
152 int
153 devfs_set_cdevpriv(void *priv, d_priv_dtor_t *priv_dtr)
154 {
155 	struct file *fp;
156 	struct cdev_priv *cdp;
157 	struct cdev_privdata *p;
158 	int error;
159 
160 	fp = curthread->td_fpop;
161 	if (fp == NULL)
162 		return (ENOENT);
163 	cdp = cdev2priv((struct cdev *)fp->f_data);
164 	p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
165 	p->cdpd_data = priv;
166 	p->cdpd_dtr = priv_dtr;
167 	p->cdpd_fp = fp;
168 	mtx_lock(&cdevpriv_mtx);
169 	if (fp->f_cdevpriv == NULL) {
170 		LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
171 		fp->f_cdevpriv = p;
172 		mtx_unlock(&cdevpriv_mtx);
173 		error = 0;
174 	} else {
175 		mtx_unlock(&cdevpriv_mtx);
176 		free(p, M_CDEVPDATA);
177 		error = EBUSY;
178 	}
179 	return (error);
180 }
181 
182 int
183 devfs_foreach_cdevpriv(struct cdev *dev, int (*cb)(void *data, void *arg),
184     void *arg)
185 {
186 	struct cdev_priv *cdp;
187 	struct cdev_privdata *p;
188 	int error;
189 
190 	cdp = cdev2priv(dev);
191 	error = 0;
192 	mtx_lock(&cdevpriv_mtx);
193 	LIST_FOREACH(p, &cdp->cdp_fdpriv, cdpd_list) {
194 		error = cb(p->cdpd_data, arg);
195 		if (error != 0)
196 			break;
197 	}
198 	mtx_unlock(&cdevpriv_mtx);
199 	return (error);
200 }
201 
202 void
203 devfs_destroy_cdevpriv(struct cdev_privdata *p)
204 {
205 
206 	mtx_assert(&cdevpriv_mtx, MA_OWNED);
207 	KASSERT(p->cdpd_fp->f_cdevpriv == p,
208 	    ("devfs_destoy_cdevpriv %p != %p", p->cdpd_fp->f_cdevpriv, p));
209 	p->cdpd_fp->f_cdevpriv = NULL;
210 	LIST_REMOVE(p, cdpd_list);
211 	mtx_unlock(&cdevpriv_mtx);
212 	(p->cdpd_dtr)(p->cdpd_data);
213 	free(p, M_CDEVPDATA);
214 }
215 
216 static void
217 devfs_fpdrop(struct file *fp)
218 {
219 	struct cdev_privdata *p;
220 
221 	mtx_lock(&cdevpriv_mtx);
222 	if ((p = fp->f_cdevpriv) == NULL) {
223 		mtx_unlock(&cdevpriv_mtx);
224 		return;
225 	}
226 	devfs_destroy_cdevpriv(p);
227 }
228 
229 void
230 devfs_clear_cdevpriv(void)
231 {
232 	struct file *fp;
233 
234 	fp = curthread->td_fpop;
235 	if (fp == NULL)
236 		return;
237 	devfs_fpdrop(fp);
238 }
239 
240 static void
241 devfs_usecount_add(struct vnode *vp)
242 {
243 	struct devfs_dirent *de;
244 	struct cdev *dev;
245 
246 	mtx_lock(&devfs_de_interlock);
247 	VI_LOCK(vp);
248 	VNPASS(vp->v_type == VCHR || vp->v_type == VBAD, vp);
249 	if (VN_IS_DOOMED(vp)) {
250 		goto out_unlock;
251 	}
252 
253 	de = vp->v_data;
254 	dev = vp->v_rdev;
255 	MPASS(de != NULL);
256 	MPASS(dev != NULL);
257 	dev->si_usecount++;
258 	de->de_usecount++;
259 out_unlock:
260 	VI_UNLOCK(vp);
261 	mtx_unlock(&devfs_de_interlock);
262 }
263 
264 static void
265 devfs_usecount_subl(struct vnode *vp)
266 {
267 	struct devfs_dirent *de;
268 	struct cdev *dev;
269 
270 	mtx_assert(&devfs_de_interlock, MA_OWNED);
271 	ASSERT_VI_LOCKED(vp, __func__);
272 	VNPASS(vp->v_type == VCHR || vp->v_type == VBAD, vp);
273 
274 	de = vp->v_data;
275 	dev = vp->v_rdev;
276 	if (de == NULL)
277 		return;
278 	if (dev == NULL) {
279 		MPASS(de->de_usecount == 0);
280 		return;
281 	}
282 	if (dev->si_usecount < de->de_usecount)
283 		panic("%s: si_usecount underflow for dev %p "
284 		    "(has %ld, dirent has %d)\n",
285 		    __func__, dev, dev->si_usecount, de->de_usecount);
286 	if (VN_IS_DOOMED(vp)) {
287 		dev->si_usecount -= de->de_usecount;
288 		de->de_usecount = 0;
289 	} else {
290 		if (de->de_usecount == 0)
291 			panic("%s: de_usecount underflow for dev %p\n",
292 			    __func__, dev);
293 		dev->si_usecount--;
294 		de->de_usecount--;
295 	}
296 }
297 
298 static void
299 devfs_usecount_sub(struct vnode *vp)
300 {
301 
302 	mtx_lock(&devfs_de_interlock);
303 	VI_LOCK(vp);
304 	devfs_usecount_subl(vp);
305 	VI_UNLOCK(vp);
306 	mtx_unlock(&devfs_de_interlock);
307 }
308 
309 static int
310 devfs_usecountl(struct vnode *vp)
311 {
312 
313 	VNPASS(vp->v_type == VCHR, vp);
314 	mtx_assert(&devfs_de_interlock, MA_OWNED);
315 	ASSERT_VI_LOCKED(vp, __func__);
316 	return (vp->v_rdev->si_usecount);
317 }
318 
319 int
320 devfs_usecount(struct vnode *vp)
321 {
322 	int count;
323 
324 	VNPASS(vp->v_type == VCHR, vp);
325 	mtx_lock(&devfs_de_interlock);
326 	VI_LOCK(vp);
327 	count = devfs_usecountl(vp);
328 	VI_UNLOCK(vp);
329 	mtx_unlock(&devfs_de_interlock);
330 	return (count);
331 }
332 
333 void
334 devfs_ctty_ref(struct vnode *vp)
335 {
336 
337 	vrefact(vp);
338 	devfs_usecount_add(vp);
339 }
340 
341 void
342 devfs_ctty_unref(struct vnode *vp)
343 {
344 
345 	devfs_usecount_sub(vp);
346 	vrele(vp);
347 }
348 
349 /*
350  * On success devfs_populate_vp() returns with dmp->dm_lock held.
351  */
352 static int
353 devfs_populate_vp(struct vnode *vp)
354 {
355 	struct devfs_dirent *de;
356 	struct devfs_mount *dmp;
357 	int locked;
358 
359 	ASSERT_VOP_LOCKED(vp, "devfs_populate_vp");
360 
361 	dmp = VFSTODEVFS(vp->v_mount);
362 	if (!devfs_populate_needed(dmp)) {
363 		sx_xlock(&dmp->dm_lock);
364 		goto out_nopopulate;
365 	}
366 
367 	locked = VOP_ISLOCKED(vp);
368 
369 	sx_xlock(&dmp->dm_lock);
370 	DEVFS_DMP_HOLD(dmp);
371 
372 	/* Can't call devfs_populate() with the vnode lock held. */
373 	VOP_UNLOCK(vp);
374 	devfs_populate(dmp);
375 
376 	sx_xunlock(&dmp->dm_lock);
377 	vn_lock(vp, locked | LK_RETRY);
378 	sx_xlock(&dmp->dm_lock);
379 	if (DEVFS_DMP_DROP(dmp)) {
380 		sx_xunlock(&dmp->dm_lock);
381 		devfs_unmount_final(dmp);
382 		return (ERESTART);
383 	}
384 out_nopopulate:
385 	if (VN_IS_DOOMED(vp)) {
386 		sx_xunlock(&dmp->dm_lock);
387 		return (ERESTART);
388 	}
389 	de = vp->v_data;
390 	KASSERT(de != NULL,
391 	    ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed"));
392 	if ((de->de_flags & DE_DOOMED) != 0) {
393 		sx_xunlock(&dmp->dm_lock);
394 		return (ERESTART);
395 	}
396 
397 	return (0);
398 }
399 
400 static int
401 devfs_vptocnp(struct vop_vptocnp_args *ap)
402 {
403 	struct vnode *vp = ap->a_vp;
404 	struct vnode **dvp = ap->a_vpp;
405 	struct devfs_mount *dmp;
406 	char *buf = ap->a_buf;
407 	size_t *buflen = ap->a_buflen;
408 	struct devfs_dirent *dd, *de;
409 	int i, error;
410 
411 	dmp = VFSTODEVFS(vp->v_mount);
412 
413 	error = devfs_populate_vp(vp);
414 	if (error != 0)
415 		return (error);
416 
417 	if (vp->v_type != VCHR && vp->v_type != VDIR) {
418 		error = ENOENT;
419 		goto finished;
420 	}
421 
422 	dd = vp->v_data;
423 	if (vp->v_type == VDIR && dd == dmp->dm_rootdir) {
424 		*dvp = vp;
425 		vref(*dvp);
426 		goto finished;
427 	}
428 
429 	i = *buflen;
430 	i -= dd->de_dirent->d_namlen;
431 	if (i < 0) {
432 		error = ENOMEM;
433 		goto finished;
434 	}
435 	bcopy(dd->de_dirent->d_name, buf + i, dd->de_dirent->d_namlen);
436 	*buflen = i;
437 	de = devfs_parent_dirent(dd);
438 	if (de == NULL) {
439 		error = ENOENT;
440 		goto finished;
441 	}
442 	mtx_lock(&devfs_de_interlock);
443 	*dvp = de->de_vnode;
444 	if (*dvp != NULL) {
445 		VI_LOCK(*dvp);
446 		mtx_unlock(&devfs_de_interlock);
447 		vholdl(*dvp);
448 		VI_UNLOCK(*dvp);
449 		vref(*dvp);
450 		vdrop(*dvp);
451 	} else {
452 		mtx_unlock(&devfs_de_interlock);
453 		error = ENOENT;
454 	}
455 finished:
456 	sx_xunlock(&dmp->dm_lock);
457 	return (error);
458 }
459 
460 /*
461  * Construct the fully qualified path name relative to the mountpoint.
462  * If a NULL cnp is provided, no '/' is appended to the resulting path.
463  */
464 char *
465 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd,
466     struct componentname *cnp)
467 {
468 	int i;
469 	struct devfs_dirent *de;
470 
471 	sx_assert(&dmp->dm_lock, SA_LOCKED);
472 
473 	i = SPECNAMELEN;
474 	buf[i] = '\0';
475 	if (cnp != NULL)
476 		i -= cnp->cn_namelen;
477 	if (i < 0)
478 		 return (NULL);
479 	if (cnp != NULL)
480 		bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
481 	de = dd;
482 	while (de != dmp->dm_rootdir) {
483 		if (cnp != NULL || i < SPECNAMELEN) {
484 			i--;
485 			if (i < 0)
486 				 return (NULL);
487 			buf[i] = '/';
488 		}
489 		i -= de->de_dirent->d_namlen;
490 		if (i < 0)
491 			 return (NULL);
492 		bcopy(de->de_dirent->d_name, buf + i,
493 		    de->de_dirent->d_namlen);
494 		de = devfs_parent_dirent(de);
495 		if (de == NULL)
496 			return (NULL);
497 	}
498 	return (buf + i);
499 }
500 
501 static int
502 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
503 	struct devfs_dirent *de)
504 {
505 	int not_found;
506 
507 	not_found = 0;
508 	if (de->de_flags & DE_DOOMED)
509 		not_found = 1;
510 	if (DEVFS_DE_DROP(de)) {
511 		KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
512 		devfs_dirent_free(de);
513 	}
514 	if (DEVFS_DMP_DROP(dmp)) {
515 		KASSERT(not_found == 1,
516 			("DEVFS mount struct freed before dirent"));
517 		not_found = 2;
518 		sx_xunlock(&dmp->dm_lock);
519 		devfs_unmount_final(dmp);
520 	}
521 	if (not_found == 1 || (drop_dm_lock && not_found != 2))
522 		sx_unlock(&dmp->dm_lock);
523 	return (not_found);
524 }
525 
526 /*
527  * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
528  * it on return.
529  */
530 int
531 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode,
532     struct vnode **vpp)
533 {
534 	int error;
535 	struct vnode *vp;
536 	struct cdev *dev;
537 	struct devfs_mount *dmp;
538 	struct cdevsw *dsw;
539 	enum vgetstate vs;
540 
541 	dmp = VFSTODEVFS(mp);
542 	if (de->de_flags & DE_DOOMED) {
543 		sx_xunlock(&dmp->dm_lock);
544 		return (ENOENT);
545 	}
546 loop:
547 	DEVFS_DE_HOLD(de);
548 	DEVFS_DMP_HOLD(dmp);
549 	mtx_lock(&devfs_de_interlock);
550 	vp = de->de_vnode;
551 	if (vp != NULL) {
552 		vs = vget_prep(vp);
553 		mtx_unlock(&devfs_de_interlock);
554 		sx_xunlock(&dmp->dm_lock);
555 		vget_finish(vp, lockmode | LK_RETRY, vs);
556 		sx_xlock(&dmp->dm_lock);
557 		if (devfs_allocv_drop_refs(0, dmp, de)) {
558 			vput(vp);
559 			return (ENOENT);
560 		}
561 		else if (VN_IS_DOOMED(vp)) {
562 			mtx_lock(&devfs_de_interlock);
563 			if (de->de_vnode == vp) {
564 				de->de_vnode = NULL;
565 				vp->v_data = NULL;
566 			}
567 			mtx_unlock(&devfs_de_interlock);
568 			vput(vp);
569 			goto loop;
570 		}
571 		sx_xunlock(&dmp->dm_lock);
572 		*vpp = vp;
573 		return (0);
574 	}
575 	mtx_unlock(&devfs_de_interlock);
576 	if (de->de_dirent->d_type == DT_CHR) {
577 		if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
578 			devfs_allocv_drop_refs(1, dmp, de);
579 			return (ENOENT);
580 		}
581 		dev = &de->de_cdp->cdp_c;
582 	} else {
583 		dev = NULL;
584 	}
585 	error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
586 	if (error != 0) {
587 		devfs_allocv_drop_refs(1, dmp, de);
588 		printf("devfs_allocv: failed to allocate new vnode\n");
589 		return (error);
590 	}
591 
592 	if (de->de_dirent->d_type == DT_CHR) {
593 		vp->v_type = VCHR;
594 		VI_LOCK(vp);
595 		dev_lock();
596 		dev_refl(dev);
597 		/* XXX: v_rdev should be protect by vnode lock */
598 		vp->v_rdev = dev;
599 		VNPASS(vp->v_usecount == 1, vp);
600 		/* Special casing of ttys for deadfs.  Probably redundant. */
601 		dsw = dev->si_devsw;
602 		if (dsw != NULL && (dsw->d_flags & D_TTY) != 0)
603 			vp->v_vflag |= VV_ISTTY;
604 		dev_unlock();
605 		VI_UNLOCK(vp);
606 		if ((dev->si_flags & SI_ETERNAL) != 0)
607 			vp->v_vflag |= VV_ETERNALDEV;
608 		vp->v_op = &devfs_specops;
609 	} else if (de->de_dirent->d_type == DT_DIR) {
610 		vp->v_type = VDIR;
611 	} else if (de->de_dirent->d_type == DT_LNK) {
612 		vp->v_type = VLNK;
613 	} else {
614 		vp->v_type = VBAD;
615 	}
616 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
617 	VN_LOCK_ASHARE(vp);
618 	mtx_lock(&devfs_de_interlock);
619 	vp->v_data = de;
620 	de->de_vnode = vp;
621 	mtx_unlock(&devfs_de_interlock);
622 	error = insmntque1(vp, mp);
623 	if (error != 0) {
624 		mtx_lock(&devfs_de_interlock);
625 		vp->v_data = NULL;
626 		de->de_vnode = NULL;
627 		mtx_unlock(&devfs_de_interlock);
628 		vgone(vp);
629 		vput(vp);
630 		(void) devfs_allocv_drop_refs(1, dmp, de);
631 		return (error);
632 	}
633 	if (devfs_allocv_drop_refs(0, dmp, de)) {
634 		vgone(vp);
635 		vput(vp);
636 		return (ENOENT);
637 	}
638 #ifdef MAC
639 	mac_devfs_vnode_associate(mp, de, vp);
640 #endif
641 	sx_xunlock(&dmp->dm_lock);
642 	vn_set_state(vp, VSTATE_CONSTRUCTED);
643 	*vpp = vp;
644 	return (0);
645 }
646 
647 static int
648 devfs_access(struct vop_access_args *ap)
649 {
650 	struct vnode *vp = ap->a_vp;
651 	struct devfs_dirent *de;
652 	struct proc *p;
653 	int error;
654 
655 	de = vp->v_data;
656 	if (vp->v_type == VDIR)
657 		de = de->de_dir;
658 
659 	error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
660 	    ap->a_accmode, ap->a_cred);
661 	if (error == 0)
662 		return (0);
663 	if (error != EACCES)
664 		return (error);
665 	p = ap->a_td->td_proc;
666 	/* We do, however, allow access to the controlling terminal */
667 	PROC_LOCK(p);
668 	if (!(p->p_flag & P_CONTROLT)) {
669 		PROC_UNLOCK(p);
670 		return (error);
671 	}
672 	if (p->p_session->s_ttydp == de->de_cdp)
673 		error = 0;
674 	PROC_UNLOCK(p);
675 	return (error);
676 }
677 
678 _Static_assert(((FMASK | FCNTLFLAGS) & (FLASTCLOSE | FREVOKE)) == 0,
679     "devfs-only flag reuse failed");
680 
681 static int
682 devfs_close(struct vop_close_args *ap)
683 {
684 	struct vnode *vp = ap->a_vp, *oldvp;
685 	struct thread *td = ap->a_td;
686 	struct proc *p;
687 	struct cdev *dev = vp->v_rdev;
688 	struct cdevsw *dsw;
689 	struct devfs_dirent *de = vp->v_data;
690 	int dflags, error, ref, vp_locked;
691 
692 	/*
693 	 * XXX: Don't call d_close() if we were called because of
694 	 * XXX: insmntque() failure.
695 	 */
696 	if (vp->v_data == NULL)
697 		return (0);
698 
699 	/*
700 	 * Hack: a tty device that is a controlling terminal
701 	 * has a reference from the session structure.
702 	 * We cannot easily tell that a character device is
703 	 * a controlling terminal, unless it is the closing
704 	 * process' controlling terminal.  In that case,
705 	 * if the reference count is 2 (this last descriptor
706 	 * plus the session), release the reference from the session.
707 	 */
708 	if (de->de_usecount == 2 && td != NULL) {
709 		p = td->td_proc;
710 		PROC_LOCK(p);
711 		if (vp == p->p_session->s_ttyvp) {
712 			PROC_UNLOCK(p);
713 			oldvp = NULL;
714 			sx_xlock(&proctree_lock);
715 			if (vp == p->p_session->s_ttyvp) {
716 				SESS_LOCK(p->p_session);
717 				mtx_lock(&devfs_de_interlock);
718 				VI_LOCK(vp);
719 				if (devfs_usecountl(vp) == 2 && !VN_IS_DOOMED(vp)) {
720 					p->p_session->s_ttyvp = NULL;
721 					p->p_session->s_ttydp = NULL;
722 					oldvp = vp;
723 				}
724 				VI_UNLOCK(vp);
725 				mtx_unlock(&devfs_de_interlock);
726 				SESS_UNLOCK(p->p_session);
727 			}
728 			sx_xunlock(&proctree_lock);
729 			if (oldvp != NULL)
730 				devfs_ctty_unref(oldvp);
731 		} else
732 			PROC_UNLOCK(p);
733 	}
734 	/*
735 	 * We do not want to really close the device if it
736 	 * is still in use unless we are trying to close it
737 	 * forcibly. Since every use (buffer, vnode, swap, cmap)
738 	 * holds a reference to the vnode, and because we mark
739 	 * any other vnodes that alias this device, when the
740 	 * sum of the reference counts on all the aliased
741 	 * vnodes descends to one, we are on last close.
742 	 */
743 	dsw = dev_refthread(dev, &ref);
744 	if (dsw == NULL)
745 		return (ENXIO);
746 	dflags = 0;
747 	mtx_lock(&devfs_de_interlock);
748 	VI_LOCK(vp);
749 	if (devfs_usecountl(vp) == 1)
750 		dflags |= FLASTCLOSE;
751 	devfs_usecount_subl(vp);
752 	mtx_unlock(&devfs_de_interlock);
753 	if (VN_IS_DOOMED(vp)) {
754 		/* Forced close. */
755 		dflags |= FREVOKE | FNONBLOCK;
756 	} else if (dsw->d_flags & D_TRACKCLOSE) {
757 		/* Keep device updated on status. */
758 	} else if ((dflags & FLASTCLOSE) == 0) {
759 		VI_UNLOCK(vp);
760 		dev_relthread(dev, ref);
761 		return (0);
762 	}
763 	vholdnz(vp);
764 	VI_UNLOCK(vp);
765 	vp_locked = VOP_ISLOCKED(vp);
766 	VOP_UNLOCK(vp);
767 	KASSERT(dev->si_refcount > 0,
768 	    ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
769 	error = dsw->d_close(dev, ap->a_fflag | dflags, S_IFCHR, td);
770 	dev_relthread(dev, ref);
771 	vn_lock(vp, vp_locked | LK_RETRY);
772 	vdrop(vp);
773 	return (error);
774 }
775 
776 static int
777 devfs_close_f(struct file *fp, struct thread *td)
778 {
779 	int error;
780 	struct file *fpop;
781 
782 	/*
783 	 * NB: td may be NULL if this descriptor is closed due to
784 	 * garbage collection from a closed UNIX domain socket.
785 	 */
786 	fpop = curthread->td_fpop;
787 	curthread->td_fpop = fp;
788 	error = vnops.fo_close(fp, td);
789 	curthread->td_fpop = fpop;
790 
791 	/*
792 	 * The f_cdevpriv cannot be assigned non-NULL value while we
793 	 * are destroying the file.
794 	 */
795 	if (fp->f_cdevpriv != NULL)
796 		devfs_fpdrop(fp);
797 	return (error);
798 }
799 
800 static int
801 devfs_getattr(struct vop_getattr_args *ap)
802 {
803 	struct vnode *vp = ap->a_vp;
804 	struct vattr *vap = ap->a_vap;
805 	struct devfs_dirent *de;
806 	struct devfs_mount *dmp;
807 	struct cdev *dev;
808 	struct timeval boottime;
809 	int error;
810 
811 	error = devfs_populate_vp(vp);
812 	if (error != 0)
813 		return (error);
814 
815 	dmp = VFSTODEVFS(vp->v_mount);
816 	sx_xunlock(&dmp->dm_lock);
817 
818 	de = vp->v_data;
819 	KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
820 	if (vp->v_type == VDIR) {
821 		de = de->de_dir;
822 		KASSERT(de != NULL,
823 		    ("Null dir dirent in devfs_getattr vp=%p", vp));
824 	}
825 	vap->va_uid = de->de_uid;
826 	vap->va_gid = de->de_gid;
827 	vap->va_mode = de->de_mode;
828 	if (vp->v_type == VLNK)
829 		vap->va_size = strlen(de->de_symlink);
830 	else if (vp->v_type == VDIR)
831 		vap->va_size = vap->va_bytes = DEV_BSIZE;
832 	else
833 		vap->va_size = 0;
834 	if (vp->v_type != VDIR)
835 		vap->va_bytes = 0;
836 	vap->va_blocksize = DEV_BSIZE;
837 	vap->va_type = vp->v_type;
838 
839 	getboottime(&boottime);
840 #define fix(aa)							\
841 	do {							\
842 		if ((aa).tv_sec <= 3600) {			\
843 			(aa).tv_sec = boottime.tv_sec;		\
844 			(aa).tv_nsec = boottime.tv_usec * 1000; \
845 		}						\
846 	} while (0)
847 
848 	if (vp->v_type != VCHR)  {
849 		fix(de->de_atime);
850 		vap->va_atime = de->de_atime;
851 		fix(de->de_mtime);
852 		vap->va_mtime = de->de_mtime;
853 		fix(de->de_ctime);
854 		vap->va_ctime = de->de_ctime;
855 	} else {
856 		dev = vp->v_rdev;
857 		fix(dev->si_atime);
858 		vap->va_atime = dev->si_atime;
859 		fix(dev->si_mtime);
860 		vap->va_mtime = dev->si_mtime;
861 		fix(dev->si_ctime);
862 		vap->va_ctime = dev->si_ctime;
863 
864 		vap->va_rdev = cdev2priv(dev)->cdp_inode;
865 	}
866 	vap->va_gen = 0;
867 	vap->va_flags = 0;
868 	vap->va_filerev = 0;
869 	vap->va_nlink = de->de_links;
870 	vap->va_fileid = de->de_inode;
871 
872 	return (error);
873 }
874 
875 /* ARGSUSED */
876 static int
877 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
878 {
879 	struct file *fpop;
880 	int error;
881 
882 	fpop = td->td_fpop;
883 	td->td_fpop = fp;
884 	error = vnops.fo_ioctl(fp, com, data, cred, td);
885 	td->td_fpop = fpop;
886 	return (error);
887 }
888 
889 void *
890 fiodgname_buf_get_ptr(void *fgnp, u_long com)
891 {
892 	union {
893 		struct fiodgname_arg	fgn;
894 #ifdef COMPAT_FREEBSD32
895 		struct fiodgname_arg32	fgn32;
896 #endif
897 	} *fgnup;
898 
899 	fgnup = fgnp;
900 	switch (com) {
901 	case FIODGNAME:
902 		return (fgnup->fgn.buf);
903 #ifdef COMPAT_FREEBSD32
904 	case FIODGNAME_32:
905 		return ((void *)(uintptr_t)fgnup->fgn32.buf);
906 #endif
907 	default:
908 		panic("Unhandled ioctl command %ld", com);
909 	}
910 }
911 
912 static int
913 devfs_ioctl(struct vop_ioctl_args *ap)
914 {
915 	struct fiodgname_arg *fgn;
916 	struct vnode *vpold, *vp;
917 	struct cdevsw *dsw;
918 	struct thread *td;
919 	struct session *sess;
920 	struct cdev *dev;
921 	int error, ref, i;
922 	const char *p;
923 	u_long com;
924 
925 	vp = ap->a_vp;
926 	com = ap->a_command;
927 	td = ap->a_td;
928 
929 	dsw = devvn_refthread(vp, &dev, &ref);
930 	if (dsw == NULL)
931 		return (ENXIO);
932 	KASSERT(dev->si_refcount > 0,
933 	    ("devfs: un-referenced struct cdev *(%s)", devtoname(dev)));
934 
935 	switch (com) {
936 	case FIODTYPE:
937 		*(int *)ap->a_data = dsw->d_flags & D_TYPEMASK;
938 		error = 0;
939 		break;
940 	case FIODGNAME:
941 #ifdef	COMPAT_FREEBSD32
942 	case FIODGNAME_32:
943 #endif
944 		fgn = ap->a_data;
945 		p = devtoname(dev);
946 		i = strlen(p) + 1;
947 		if (i > fgn->len)
948 			error = EINVAL;
949 		else
950 			error = copyout(p, fiodgname_buf_get_ptr(fgn, com), i);
951 		break;
952 	default:
953 		error = dsw->d_ioctl(dev, com, ap->a_data, ap->a_fflag, td);
954 	}
955 
956 	dev_relthread(dev, ref);
957 	if (error == ENOIOCTL)
958 		error = ENOTTY;
959 
960 	if (error == 0 && com == TIOCSCTTY) {
961 		/*
962 		 * Do nothing if reassigning same control tty, or if the
963 		 * control tty has already disappeared.  If it disappeared,
964 		 * it's because we were racing with TIOCNOTTY.  TIOCNOTTY
965 		 * already took care of releasing the old vnode and we have
966 		 * nothing left to do.
967 		 */
968 		sx_slock(&proctree_lock);
969 		sess = td->td_proc->p_session;
970 		if (sess->s_ttyvp == vp || sess->s_ttyp == NULL) {
971 			sx_sunlock(&proctree_lock);
972 			return (0);
973 		}
974 
975 		devfs_ctty_ref(vp);
976 		SESS_LOCK(sess);
977 		vpold = sess->s_ttyvp;
978 		sess->s_ttyvp = vp;
979 		sess->s_ttydp = cdev2priv(dev);
980 		SESS_UNLOCK(sess);
981 
982 		sx_sunlock(&proctree_lock);
983 
984 		/* Get rid of reference to old control tty */
985 		if (vpold)
986 			devfs_ctty_unref(vpold);
987 	}
988 	return (error);
989 }
990 
991 /* ARGSUSED */
992 static int
993 devfs_kqfilter_f(struct file *fp, struct knote *kn)
994 {
995 	struct cdev *dev;
996 	struct cdevsw *dsw;
997 	int error, ref;
998 	struct file *fpop;
999 	struct thread *td;
1000 
1001 	td = curthread;
1002 	fpop = td->td_fpop;
1003 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1004 	if (error)
1005 		return (error);
1006 	error = dsw->d_kqfilter(dev, kn);
1007 	td->td_fpop = fpop;
1008 	dev_relthread(dev, ref);
1009 	return (error);
1010 }
1011 
1012 static inline int
1013 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
1014 {
1015 	struct cdev_priv *cdp;
1016 	struct ucred *dcr;
1017 	struct proc *p;
1018 	int error;
1019 
1020 	cdp = de->de_cdp;
1021 	if (cdp == NULL)
1022 		return (0);
1023 	dcr = cdp->cdp_c.si_cred;
1024 	if (dcr == NULL)
1025 		return (0);
1026 
1027 	error = prison_check(td->td_ucred, dcr);
1028 	if (error == 0)
1029 		return (0);
1030 	/* We do, however, allow access to the controlling terminal */
1031 	p = td->td_proc;
1032 	PROC_LOCK(p);
1033 	if (!(p->p_flag & P_CONTROLT)) {
1034 		PROC_UNLOCK(p);
1035 		return (error);
1036 	}
1037 	if (p->p_session->s_ttydp == cdp)
1038 		error = 0;
1039 	PROC_UNLOCK(p);
1040 	return (error);
1041 }
1042 
1043 static int
1044 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
1045 {
1046 	struct componentname *cnp;
1047 	struct vnode *dvp, **vpp;
1048 	struct thread *td;
1049 	struct devfs_dirent *de, *dd;
1050 	struct devfs_dirent **dde;
1051 	struct devfs_mount *dmp;
1052 	struct mount *mp;
1053 	struct cdev *cdev;
1054 	int error, flags, nameiop, dvplocked;
1055 	char specname[SPECNAMELEN + 1], *pname;
1056 
1057 	td = curthread;
1058 	cnp = ap->a_cnp;
1059 	vpp = ap->a_vpp;
1060 	dvp = ap->a_dvp;
1061 	pname = cnp->cn_nameptr;
1062 	flags = cnp->cn_flags;
1063 	nameiop = cnp->cn_nameiop;
1064 	mp = dvp->v_mount;
1065 	dmp = VFSTODEVFS(mp);
1066 	dd = dvp->v_data;
1067 	*vpp = NULLVP;
1068 
1069 	if ((flags & ISLASTCN) && nameiop == RENAME)
1070 		return (EOPNOTSUPP);
1071 
1072 	if (dvp->v_type != VDIR)
1073 		return (ENOTDIR);
1074 
1075 	if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
1076 		return (EIO);
1077 
1078 	error = vn_dir_check_exec(dvp, cnp);
1079 	if (error != 0)
1080 		return (error);
1081 
1082 	if (cnp->cn_namelen == 1 && *pname == '.') {
1083 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
1084 			return (EINVAL);
1085 		*vpp = dvp;
1086 		VREF(dvp);
1087 		return (0);
1088 	}
1089 
1090 	if (flags & ISDOTDOT) {
1091 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
1092 			return (EINVAL);
1093 		de = devfs_parent_dirent(dd);
1094 		if (de == NULL)
1095 			return (ENOENT);
1096 		dvplocked = VOP_ISLOCKED(dvp);
1097 		VOP_UNLOCK(dvp);
1098 		error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK,
1099 		    vpp);
1100 		*dm_unlock = 0;
1101 		vn_lock(dvp, dvplocked | LK_RETRY);
1102 		return (error);
1103 	}
1104 
1105 	dd = dvp->v_data;
1106 	de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
1107 	while (de == NULL) {	/* While(...) so we can use break */
1108 
1109 		if (nameiop == DELETE)
1110 			return (ENOENT);
1111 
1112 		/*
1113 		 * OK, we didn't have an entry for the name we were asked for
1114 		 * so we try to see if anybody can create it on demand.
1115 		 */
1116 		pname = devfs_fqpn(specname, dmp, dd, cnp);
1117 		if (pname == NULL)
1118 			break;
1119 
1120 		cdev = NULL;
1121 		DEVFS_DMP_HOLD(dmp);
1122 		sx_xunlock(&dmp->dm_lock);
1123 		EVENTHANDLER_INVOKE(dev_clone,
1124 		    td->td_ucred, pname, strlen(pname), &cdev);
1125 
1126 		if (cdev == NULL)
1127 			sx_xlock(&dmp->dm_lock);
1128 		else if (devfs_populate_vp(dvp) != 0) {
1129 			*dm_unlock = 0;
1130 			sx_xlock(&dmp->dm_lock);
1131 			if (DEVFS_DMP_DROP(dmp)) {
1132 				sx_xunlock(&dmp->dm_lock);
1133 				devfs_unmount_final(dmp);
1134 			} else
1135 				sx_xunlock(&dmp->dm_lock);
1136 			dev_rel(cdev);
1137 			return (ENOENT);
1138 		}
1139 		if (DEVFS_DMP_DROP(dmp)) {
1140 			*dm_unlock = 0;
1141 			sx_xunlock(&dmp->dm_lock);
1142 			devfs_unmount_final(dmp);
1143 			if (cdev != NULL)
1144 				dev_rel(cdev);
1145 			return (ENOENT);
1146 		}
1147 
1148 		if (cdev == NULL)
1149 			break;
1150 
1151 		dev_lock();
1152 		dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
1153 		if (dde != NULL && *dde != NULL)
1154 			de = *dde;
1155 		dev_unlock();
1156 		dev_rel(cdev);
1157 		break;
1158 	}
1159 
1160 	if (de == NULL || de->de_flags & DE_WHITEOUT) {
1161 		if ((nameiop == CREATE || nameiop == RENAME) &&
1162 		    (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
1163 			return (EJUSTRETURN);
1164 		}
1165 		return (ENOENT);
1166 	}
1167 
1168 	if (devfs_prison_check(de, td))
1169 		return (ENOENT);
1170 
1171 	if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
1172 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
1173 		if (error)
1174 			return (error);
1175 		if (*vpp == dvp) {
1176 			VREF(dvp);
1177 			*vpp = dvp;
1178 			return (0);
1179 		}
1180 	}
1181 	error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, vpp);
1182 	*dm_unlock = 0;
1183 	return (error);
1184 }
1185 
1186 static int
1187 devfs_lookup(struct vop_lookup_args *ap)
1188 {
1189 	int j;
1190 	struct devfs_mount *dmp;
1191 	int dm_unlock;
1192 
1193 	if (devfs_populate_vp(ap->a_dvp) != 0)
1194 		return (ENOTDIR);
1195 
1196 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1197 	dm_unlock = 1;
1198 	j = devfs_lookupx(ap, &dm_unlock);
1199 	if (dm_unlock == 1)
1200 		sx_xunlock(&dmp->dm_lock);
1201 	return (j);
1202 }
1203 
1204 static int
1205 devfs_mknod(struct vop_mknod_args *ap)
1206 {
1207 	struct componentname *cnp;
1208 	struct vnode *dvp, **vpp;
1209 	struct devfs_dirent *dd, *de;
1210 	struct devfs_mount *dmp;
1211 	int error;
1212 
1213 	/*
1214 	 * The only type of node we should be creating here is a
1215 	 * character device, for anything else return EOPNOTSUPP.
1216 	 */
1217 	if (ap->a_vap->va_type != VCHR)
1218 		return (EOPNOTSUPP);
1219 	dvp = ap->a_dvp;
1220 	dmp = VFSTODEVFS(dvp->v_mount);
1221 
1222 	cnp = ap->a_cnp;
1223 	vpp = ap->a_vpp;
1224 	dd = dvp->v_data;
1225 
1226 	error = ENOENT;
1227 	sx_xlock(&dmp->dm_lock);
1228 	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1229 		if (cnp->cn_namelen != de->de_dirent->d_namlen)
1230 			continue;
1231 		if (de->de_dirent->d_type == DT_CHR &&
1232 		    (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0)
1233 			continue;
1234 		if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1235 		    de->de_dirent->d_namlen) != 0)
1236 			continue;
1237 		if (de->de_flags & DE_WHITEOUT)
1238 			break;
1239 		goto notfound;
1240 	}
1241 	if (de == NULL)
1242 		goto notfound;
1243 	de->de_flags &= ~DE_WHITEOUT;
1244 	error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1245 	return (error);
1246 notfound:
1247 	sx_xunlock(&dmp->dm_lock);
1248 	return (error);
1249 }
1250 
1251 /* ARGSUSED */
1252 static int
1253 devfs_open(struct vop_open_args *ap)
1254 {
1255 	struct thread *td = ap->a_td;
1256 	struct vnode *vp = ap->a_vp;
1257 	struct cdev *dev = vp->v_rdev;
1258 	struct file *fp = ap->a_fp;
1259 	int error, ref, vlocked;
1260 	struct cdevsw *dsw;
1261 	struct file *fpop;
1262 
1263 	if (vp->v_type == VBLK)
1264 		return (ENXIO);
1265 
1266 	if (dev == NULL)
1267 		return (ENXIO);
1268 
1269 	/* Make this field valid before any I/O in d_open. */
1270 	if (dev->si_iosize_max == 0)
1271 		dev->si_iosize_max = DFLTPHYS;
1272 
1273 	dsw = dev_refthread(dev, &ref);
1274 	if (dsw == NULL)
1275 		return (ENXIO);
1276 	if (fp == NULL && dsw->d_fdopen != NULL) {
1277 		dev_relthread(dev, ref);
1278 		return (ENXIO);
1279 	}
1280 
1281 	if (vp->v_type == VCHR)
1282 		devfs_usecount_add(vp);
1283 
1284 	vlocked = VOP_ISLOCKED(vp);
1285 	VOP_UNLOCK(vp);
1286 
1287 	fpop = td->td_fpop;
1288 	td->td_fpop = fp;
1289 	if (fp != NULL) {
1290 		fp->f_data = dev;
1291 		fp->f_vnode = vp;
1292 	}
1293 	if (dsw->d_fdopen != NULL)
1294 		error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1295 	else
1296 		error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1297 	/* Clean up any cdevpriv upon error. */
1298 	if (error != 0)
1299 		devfs_clear_cdevpriv();
1300 	td->td_fpop = fpop;
1301 
1302 	vn_lock(vp, vlocked | LK_RETRY);
1303 	if (error != 0 && vp->v_type == VCHR)
1304 		devfs_usecount_sub(vp);
1305 
1306 	dev_relthread(dev, ref);
1307 	if (error != 0) {
1308 		if (error == ERESTART)
1309 			error = EINTR;
1310 		return (error);
1311 	}
1312 
1313 #if 0	/* /dev/console */
1314 	KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1315 #else
1316 	if (fp == NULL)
1317 		return (error);
1318 #endif
1319 	if (fp->f_ops == &badfileops)
1320 		finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1321 	return (error);
1322 }
1323 
1324 static int
1325 devfs_pathconf(struct vop_pathconf_args *ap)
1326 {
1327 
1328 	switch (ap->a_name) {
1329 	case _PC_FILESIZEBITS:
1330 		*ap->a_retval = 64;
1331 		return (0);
1332 	case _PC_NAME_MAX:
1333 		*ap->a_retval = NAME_MAX;
1334 		return (0);
1335 	case _PC_LINK_MAX:
1336 		*ap->a_retval = INT_MAX;
1337 		return (0);
1338 	case _PC_SYMLINK_MAX:
1339 		*ap->a_retval = MAXPATHLEN;
1340 		return (0);
1341 	case _PC_MAX_CANON:
1342 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1343 			*ap->a_retval = MAX_CANON;
1344 			return (0);
1345 		}
1346 		return (EINVAL);
1347 	case _PC_MAX_INPUT:
1348 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1349 			*ap->a_retval = MAX_INPUT;
1350 			return (0);
1351 		}
1352 		return (EINVAL);
1353 	case _PC_VDISABLE:
1354 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1355 			*ap->a_retval = _POSIX_VDISABLE;
1356 			return (0);
1357 		}
1358 		return (EINVAL);
1359 	case _PC_MAC_PRESENT:
1360 #ifdef MAC
1361 		/*
1362 		 * If MAC is enabled, devfs automatically supports
1363 		 * trivial non-persistent label storage.
1364 		 */
1365 		*ap->a_retval = 1;
1366 #else
1367 		*ap->a_retval = 0;
1368 #endif
1369 		return (0);
1370 	case _PC_CHOWN_RESTRICTED:
1371 		*ap->a_retval = 1;
1372 		return (0);
1373 	default:
1374 		return (vop_stdpathconf(ap));
1375 	}
1376 	/* NOTREACHED */
1377 }
1378 
1379 /* ARGSUSED */
1380 static int
1381 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1382 {
1383 	struct cdev *dev;
1384 	struct cdevsw *dsw;
1385 	int error, ref;
1386 	struct file *fpop;
1387 
1388 	fpop = td->td_fpop;
1389 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1390 	if (error != 0) {
1391 		error = vnops.fo_poll(fp, events, cred, td);
1392 		return (error);
1393 	}
1394 	error = dsw->d_poll(dev, events, td);
1395 	td->td_fpop = fpop;
1396 	dev_relthread(dev, ref);
1397 	return(error);
1398 }
1399 
1400 /*
1401  * Print out the contents of a special device vnode.
1402  */
1403 static int
1404 devfs_print(struct vop_print_args *ap)
1405 {
1406 
1407 	printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1408 	return (0);
1409 }
1410 
1411 static int
1412 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
1413     int flags, struct thread *td)
1414 {
1415 	struct cdev *dev;
1416 	int ioflag, error, ref;
1417 	ssize_t resid;
1418 	struct cdevsw *dsw;
1419 	struct file *fpop;
1420 
1421 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1422 		return (EINVAL);
1423 	fpop = td->td_fpop;
1424 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1425 	if (error != 0) {
1426 		error = vnops.fo_read(fp, uio, cred, flags, td);
1427 		return (error);
1428 	}
1429 	resid = uio->uio_resid;
1430 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1431 	if (ioflag & O_DIRECT)
1432 		ioflag |= IO_DIRECT;
1433 
1434 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1435 	error = dsw->d_read(dev, uio, ioflag);
1436 	if (uio->uio_resid != resid || (error == 0 && resid != 0))
1437 		devfs_timestamp(&dev->si_atime);
1438 	td->td_fpop = fpop;
1439 	dev_relthread(dev, ref);
1440 
1441 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF_R);
1442 	return (error);
1443 }
1444 
1445 static int
1446 devfs_readdir(struct vop_readdir_args *ap)
1447 {
1448 	int error;
1449 	struct uio *uio;
1450 	struct dirent *dp;
1451 	struct devfs_dirent *dd;
1452 	struct devfs_dirent *de;
1453 	struct devfs_mount *dmp;
1454 	off_t off;
1455 	int *tmp_ncookies = NULL;
1456 
1457 	if (ap->a_vp->v_type != VDIR)
1458 		return (ENOTDIR);
1459 
1460 	uio = ap->a_uio;
1461 	if (uio->uio_offset < 0)
1462 		return (EINVAL);
1463 
1464 	/*
1465 	 * XXX: This is a temporary hack to get around this filesystem not
1466 	 * supporting cookies. We store the location of the ncookies pointer
1467 	 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1468 	 * and set the number of cookies to 0. We then set the pointer to
1469 	 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1470 	 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1471 	 * pointer to its original location before returning to the caller.
1472 	 */
1473 	if (ap->a_ncookies != NULL) {
1474 		tmp_ncookies = ap->a_ncookies;
1475 		*ap->a_ncookies = 0;
1476 		ap->a_ncookies = NULL;
1477 	}
1478 
1479 	dmp = VFSTODEVFS(ap->a_vp->v_mount);
1480 	if (devfs_populate_vp(ap->a_vp) != 0) {
1481 		if (tmp_ncookies != NULL)
1482 			ap->a_ncookies = tmp_ncookies;
1483 		return (EIO);
1484 	}
1485 	error = 0;
1486 	de = ap->a_vp->v_data;
1487 	off = 0;
1488 	TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1489 		KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1490 		if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1491 			continue;
1492 		if (devfs_prison_check(dd, uio->uio_td))
1493 			continue;
1494 		if (dd->de_dirent->d_type == DT_DIR)
1495 			de = dd->de_dir;
1496 		else
1497 			de = dd;
1498 		dp = dd->de_dirent;
1499 		MPASS(dp->d_reclen == GENERIC_DIRSIZ(dp));
1500 		if (dp->d_reclen > uio->uio_resid)
1501 			break;
1502 		dp->d_fileno = de->de_inode;
1503 		/* NOTE: d_off is the offset for the *next* entry. */
1504 		dp->d_off = off + dp->d_reclen;
1505 		if (off >= uio->uio_offset) {
1506 			error = vfs_read_dirent(ap, dp, off);
1507 			if (error)
1508 				break;
1509 		}
1510 		off += dp->d_reclen;
1511 	}
1512 	sx_xunlock(&dmp->dm_lock);
1513 	uio->uio_offset = off;
1514 
1515 	/*
1516 	 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1517 	 * place.
1518 	 */
1519 	if (tmp_ncookies != NULL)
1520 		ap->a_ncookies = tmp_ncookies;
1521 
1522 	return (error);
1523 }
1524 
1525 static int
1526 devfs_readlink(struct vop_readlink_args *ap)
1527 {
1528 	struct devfs_dirent *de;
1529 
1530 	de = ap->a_vp->v_data;
1531 	return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1532 }
1533 
1534 static void
1535 devfs_reclaiml(struct vnode *vp)
1536 {
1537 	struct devfs_dirent *de;
1538 
1539 	mtx_assert(&devfs_de_interlock, MA_OWNED);
1540 	de = vp->v_data;
1541 	if (de != NULL) {
1542 		MPASS(de->de_usecount == 0);
1543 		de->de_vnode = NULL;
1544 		vp->v_data = NULL;
1545 	}
1546 }
1547 
1548 static int
1549 devfs_reclaim(struct vop_reclaim_args *ap)
1550 {
1551 	struct vnode *vp;
1552 
1553 	vp = ap->a_vp;
1554 	mtx_lock(&devfs_de_interlock);
1555 	devfs_reclaiml(vp);
1556 	mtx_unlock(&devfs_de_interlock);
1557 	return (0);
1558 }
1559 
1560 static int
1561 devfs_reclaim_vchr(struct vop_reclaim_args *ap)
1562 {
1563 	struct vnode *vp;
1564 	struct cdev *dev;
1565 
1566 	vp = ap->a_vp;
1567 	MPASS(vp->v_type == VCHR);
1568 
1569 	mtx_lock(&devfs_de_interlock);
1570 	VI_LOCK(vp);
1571 	devfs_usecount_subl(vp);
1572 	devfs_reclaiml(vp);
1573 	mtx_unlock(&devfs_de_interlock);
1574 	dev_lock();
1575 	dev = vp->v_rdev;
1576 	vp->v_rdev = NULL;
1577 	dev_unlock();
1578 	VI_UNLOCK(vp);
1579 	if (dev != NULL)
1580 		dev_rel(dev);
1581 	return (0);
1582 }
1583 
1584 static int
1585 devfs_remove(struct vop_remove_args *ap)
1586 {
1587 	struct vnode *dvp = ap->a_dvp;
1588 	struct vnode *vp = ap->a_vp;
1589 	struct devfs_dirent *dd;
1590 	struct devfs_dirent *de, *de_covered;
1591 	struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1592 
1593 	ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1594 	ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1595 
1596 	sx_xlock(&dmp->dm_lock);
1597 	dd = ap->a_dvp->v_data;
1598 	de = vp->v_data;
1599 	if (de->de_cdp == NULL) {
1600 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1601 		if (de->de_dirent->d_type == DT_LNK) {
1602 			de_covered = devfs_find(dd, de->de_dirent->d_name,
1603 			    de->de_dirent->d_namlen, 0);
1604 			if (de_covered != NULL)
1605 				de_covered->de_flags &= ~DE_COVERED;
1606 		}
1607 		/* We need to unlock dvp because devfs_delete() may lock it. */
1608 		VOP_UNLOCK(vp);
1609 		if (dvp != vp)
1610 			VOP_UNLOCK(dvp);
1611 		devfs_delete(dmp, de, 0);
1612 		sx_xunlock(&dmp->dm_lock);
1613 		if (dvp != vp)
1614 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1615 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1616 	} else {
1617 		de->de_flags |= DE_WHITEOUT;
1618 		sx_xunlock(&dmp->dm_lock);
1619 	}
1620 	return (0);
1621 }
1622 
1623 /*
1624  * Revoke is called on a tty when a terminal session ends.  The vnode
1625  * is orphaned by setting v_op to deadfs so we need to let go of it
1626  * as well so that we create a new one next time around.
1627  *
1628  */
1629 static int
1630 devfs_revoke(struct vop_revoke_args *ap)
1631 {
1632 	struct vnode *vp = ap->a_vp, *vp2;
1633 	struct cdev *dev;
1634 	struct cdev_priv *cdp;
1635 	struct devfs_dirent *de;
1636 	enum vgetstate vs;
1637 	u_int i;
1638 
1639 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1640 
1641 	dev = vp->v_rdev;
1642 	cdp = cdev2priv(dev);
1643 
1644 	dev_lock();
1645 	cdp->cdp_inuse++;
1646 	dev_unlock();
1647 
1648 	vhold(vp);
1649 	vgone(vp);
1650 	vdrop(vp);
1651 
1652 	VOP_UNLOCK(vp);
1653  loop:
1654 	for (;;) {
1655 		mtx_lock(&devfs_de_interlock);
1656 		dev_lock();
1657 		vp2 = NULL;
1658 		for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1659 			de = cdp->cdp_dirents[i];
1660 			if (de == NULL)
1661 				continue;
1662 
1663 			vp2 = de->de_vnode;
1664 			if (vp2 != NULL) {
1665 				dev_unlock();
1666 				vs = vget_prep(vp2);
1667 				mtx_unlock(&devfs_de_interlock);
1668 				if (vget_finish(vp2, LK_EXCLUSIVE, vs) != 0)
1669 					goto loop;
1670 				vhold(vp2);
1671 				vgone(vp2);
1672 				vdrop(vp2);
1673 				vput(vp2);
1674 				break;
1675 			}
1676 		}
1677 		if (vp2 != NULL) {
1678 			continue;
1679 		}
1680 		dev_unlock();
1681 		mtx_unlock(&devfs_de_interlock);
1682 		break;
1683 	}
1684 	dev_lock();
1685 	cdp->cdp_inuse--;
1686 	if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1687 		KASSERT((cdp->cdp_flags & CDP_ON_ACTIVE_LIST) != 0,
1688 		    ("%s: cdp %p (%s) not on active list",
1689 		    __func__, cdp, dev->si_name));
1690 		cdp->cdp_flags &= ~CDP_ON_ACTIVE_LIST;
1691 		TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1692 		dev_unlock();
1693 		dev_rel(&cdp->cdp_c);
1694 	} else
1695 		dev_unlock();
1696 
1697 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1698 	return (0);
1699 }
1700 
1701 static int
1702 devfs_rioctl(struct vop_ioctl_args *ap)
1703 {
1704 	struct vnode *vp;
1705 	struct devfs_mount *dmp;
1706 	int error;
1707 
1708 	vp = ap->a_vp;
1709 	vn_lock(vp, LK_SHARED | LK_RETRY);
1710 	if (VN_IS_DOOMED(vp)) {
1711 		VOP_UNLOCK(vp);
1712 		return (EBADF);
1713 	}
1714 	dmp = VFSTODEVFS(vp->v_mount);
1715 	sx_xlock(&dmp->dm_lock);
1716 	VOP_UNLOCK(vp);
1717 	DEVFS_DMP_HOLD(dmp);
1718 	devfs_populate(dmp);
1719 	if (DEVFS_DMP_DROP(dmp)) {
1720 		sx_xunlock(&dmp->dm_lock);
1721 		devfs_unmount_final(dmp);
1722 		return (ENOENT);
1723 	}
1724 	error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1725 	sx_xunlock(&dmp->dm_lock);
1726 	return (error);
1727 }
1728 
1729 static int
1730 devfs_rread(struct vop_read_args *ap)
1731 {
1732 
1733 	if (ap->a_vp->v_type != VDIR)
1734 		return (EINVAL);
1735 	return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1736 }
1737 
1738 static int
1739 devfs_setattr(struct vop_setattr_args *ap)
1740 {
1741 	struct devfs_dirent *de;
1742 	struct vattr *vap;
1743 	struct vnode *vp;
1744 	struct thread *td;
1745 	int c, error;
1746 	uid_t uid;
1747 	gid_t gid;
1748 
1749 	vap = ap->a_vap;
1750 	vp = ap->a_vp;
1751 	td = curthread;
1752 	if ((vap->va_type != VNON) ||
1753 	    (vap->va_nlink != VNOVAL) ||
1754 	    (vap->va_fsid != VNOVAL) ||
1755 	    (vap->va_fileid != VNOVAL) ||
1756 	    (vap->va_blocksize != VNOVAL) ||
1757 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1758 	    (vap->va_rdev != VNOVAL) ||
1759 	    ((int)vap->va_bytes != VNOVAL) ||
1760 	    (vap->va_gen != VNOVAL)) {
1761 		return (EINVAL);
1762 	}
1763 
1764 	error = devfs_populate_vp(vp);
1765 	if (error != 0)
1766 		return (error);
1767 
1768 	de = vp->v_data;
1769 	if (vp->v_type == VDIR)
1770 		de = de->de_dir;
1771 
1772 	c = 0;
1773 	if (vap->va_uid == (uid_t)VNOVAL)
1774 		uid = de->de_uid;
1775 	else
1776 		uid = vap->va_uid;
1777 	if (vap->va_gid == (gid_t)VNOVAL)
1778 		gid = de->de_gid;
1779 	else
1780 		gid = vap->va_gid;
1781 	if (uid != de->de_uid || gid != de->de_gid) {
1782 		if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1783 		    (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1784 			error = priv_check(td, PRIV_VFS_CHOWN);
1785 			if (error != 0)
1786 				goto ret;
1787 		}
1788 		de->de_uid = uid;
1789 		de->de_gid = gid;
1790 		c = 1;
1791 	}
1792 
1793 	if (vap->va_mode != (mode_t)VNOVAL) {
1794 		if (ap->a_cred->cr_uid != de->de_uid) {
1795 			error = priv_check(td, PRIV_VFS_ADMIN);
1796 			if (error != 0)
1797 				goto ret;
1798 		}
1799 		de->de_mode = vap->va_mode;
1800 		c = 1;
1801 	}
1802 
1803 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1804 		error = vn_utimes_perm(vp, vap, ap->a_cred, td);
1805 		if (error != 0)
1806 			goto ret;
1807 		if (vap->va_atime.tv_sec != VNOVAL) {
1808 			if (vp->v_type == VCHR)
1809 				vp->v_rdev->si_atime = vap->va_atime;
1810 			else
1811 				de->de_atime = vap->va_atime;
1812 		}
1813 		if (vap->va_mtime.tv_sec != VNOVAL) {
1814 			if (vp->v_type == VCHR)
1815 				vp->v_rdev->si_mtime = vap->va_mtime;
1816 			else
1817 				de->de_mtime = vap->va_mtime;
1818 		}
1819 		c = 1;
1820 	}
1821 
1822 	if (c) {
1823 		if (vp->v_type == VCHR)
1824 			vfs_timestamp(&vp->v_rdev->si_ctime);
1825 		else
1826 			vfs_timestamp(&de->de_mtime);
1827 	}
1828 
1829 ret:
1830 	sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock);
1831 	return (error);
1832 }
1833 
1834 #ifdef MAC
1835 static int
1836 devfs_setlabel(struct vop_setlabel_args *ap)
1837 {
1838 	struct vnode *vp;
1839 	struct devfs_dirent *de;
1840 
1841 	vp = ap->a_vp;
1842 	de = vp->v_data;
1843 
1844 	mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1845 	mac_devfs_update(vp->v_mount, de, vp);
1846 
1847 	return (0);
1848 }
1849 #endif
1850 
1851 static int
1852 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred)
1853 {
1854 
1855 	return (vnops.fo_stat(fp, sb, cred));
1856 }
1857 
1858 static int
1859 devfs_symlink(struct vop_symlink_args *ap)
1860 {
1861 	int i, error;
1862 	struct devfs_dirent *dd;
1863 	struct devfs_dirent *de, *de_covered, *de_dotdot;
1864 	struct devfs_mount *dmp;
1865 
1866 	error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1867 	if (error)
1868 		return(error);
1869 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1870 	if (devfs_populate_vp(ap->a_dvp) != 0)
1871 		return (ENOENT);
1872 
1873 	dd = ap->a_dvp->v_data;
1874 	de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1875 	de->de_flags = DE_USER;
1876 	de->de_uid = 0;
1877 	de->de_gid = 0;
1878 	de->de_mode = 0755;
1879 	de->de_inode = alloc_unr(devfs_inos);
1880 	de->de_dir = dd;
1881 	de->de_dirent->d_type = DT_LNK;
1882 	i = strlen(ap->a_target) + 1;
1883 	de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1884 	bcopy(ap->a_target, de->de_symlink, i);
1885 #ifdef MAC
1886 	mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1887 #endif
1888 	de_covered = devfs_find(dd, de->de_dirent->d_name,
1889 	    de->de_dirent->d_namlen, 0);
1890 	if (de_covered != NULL) {
1891 		if ((de_covered->de_flags & DE_USER) != 0) {
1892 			devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1893 			sx_xunlock(&dmp->dm_lock);
1894 			return (EEXIST);
1895 		}
1896 		KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1897 		    ("devfs_symlink: entry %p already covered", de_covered));
1898 		de_covered->de_flags |= DE_COVERED;
1899 	}
1900 
1901 	de_dotdot = TAILQ_FIRST(&dd->de_dlist);		/* "." */
1902 	de_dotdot = TAILQ_NEXT(de_dotdot, de_list);	/* ".." */
1903 	TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1904 	devfs_dir_ref_de(dmp, dd);
1905 	devfs_rules_apply(dmp, de);
1906 
1907 	return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1908 }
1909 
1910 static int
1911 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1912 {
1913 
1914 	return (vnops.fo_truncate(fp, length, cred, td));
1915 }
1916 
1917 static int
1918 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
1919     int flags, struct thread *td)
1920 {
1921 	struct cdev *dev;
1922 	int error, ioflag, ref;
1923 	ssize_t resid;
1924 	struct cdevsw *dsw;
1925 	struct file *fpop;
1926 
1927 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1928 		return (EINVAL);
1929 	fpop = td->td_fpop;
1930 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1931 	if (error != 0) {
1932 		error = vnops.fo_write(fp, uio, cred, flags, td);
1933 		return (error);
1934 	}
1935 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1936 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1937 	if (ioflag & O_DIRECT)
1938 		ioflag |= IO_DIRECT;
1939 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1940 
1941 	resid = uio->uio_resid;
1942 
1943 	error = dsw->d_write(dev, uio, ioflag);
1944 	if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1945 		devfs_timestamp(&dev->si_ctime);
1946 		dev->si_mtime = dev->si_ctime;
1947 	}
1948 	td->td_fpop = fpop;
1949 	dev_relthread(dev, ref);
1950 
1951 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF_W);
1952 	return (error);
1953 }
1954 
1955 static int
1956 devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1957     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1958     struct thread *td)
1959 {
1960 	struct cdev *dev;
1961 	struct cdevsw *dsw;
1962 	struct mount *mp;
1963 	struct vnode *vp;
1964 	struct file *fpop;
1965 	vm_object_t object;
1966 	vm_prot_t maxprot;
1967 	int error, ref;
1968 
1969 	vp = fp->f_vnode;
1970 
1971 	/*
1972 	 * Ensure that file and memory protections are
1973 	 * compatible.
1974 	 */
1975 	mp = vp->v_mount;
1976 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1977 		maxprot = VM_PROT_NONE;
1978 		if ((prot & VM_PROT_EXECUTE) != 0)
1979 			return (EACCES);
1980 	} else
1981 		maxprot = VM_PROT_EXECUTE;
1982 	if ((fp->f_flag & FREAD) != 0)
1983 		maxprot |= VM_PROT_READ;
1984 	else if ((prot & VM_PROT_READ) != 0)
1985 		return (EACCES);
1986 
1987 	/*
1988 	 * If we are sharing potential changes via MAP_SHARED and we
1989 	 * are trying to get write permission although we opened it
1990 	 * without asking for it, bail out.
1991 	 *
1992 	 * Note that most character devices always share mappings.
1993 	 * The one exception is that D_MMAP_ANON devices
1994 	 * (i.e. /dev/zero) permit private writable mappings.
1995 	 *
1996 	 * Rely on vm_mmap_cdev() to fail invalid MAP_PRIVATE requests
1997 	 * as well as updating maxprot to permit writing for
1998 	 * D_MMAP_ANON devices rather than doing that here.
1999 	 */
2000 	if ((flags & MAP_SHARED) != 0) {
2001 		if ((fp->f_flag & FWRITE) != 0)
2002 			maxprot |= VM_PROT_WRITE;
2003 		else if ((prot & VM_PROT_WRITE) != 0)
2004 			return (EACCES);
2005 	}
2006 	maxprot &= cap_maxprot;
2007 
2008 	fpop = td->td_fpop;
2009 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
2010 	if (error != 0)
2011 		return (error);
2012 
2013 	error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, dev, dsw, &foff,
2014 	    &object);
2015 	td->td_fpop = fpop;
2016 	dev_relthread(dev, ref);
2017 	if (error != 0)
2018 		return (error);
2019 
2020 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2021 	    foff, FALSE, td);
2022 	if (error != 0)
2023 		vm_object_deallocate(object);
2024 	return (error);
2025 }
2026 
2027 dev_t
2028 dev2udev(struct cdev *x)
2029 {
2030 	if (x == NULL)
2031 		return (NODEV);
2032 	return (cdev2priv(x)->cdp_inode);
2033 }
2034 
2035 static int
2036 devfs_cmp_f(struct file *fp1, struct file *fp2, struct thread *td)
2037 {
2038 	if (fp2->f_type != DTYPE_VNODE || fp2->f_ops != &devfs_ops_f)
2039 		return (3);
2040 	return (kcmp_cmp((uintptr_t)fp1->f_data, (uintptr_t)fp2->f_data));
2041 }
2042 
2043 static struct fileops devfs_ops_f = {
2044 	.fo_read =	devfs_read_f,
2045 	.fo_write =	devfs_write_f,
2046 	.fo_truncate =	devfs_truncate_f,
2047 	.fo_ioctl =	devfs_ioctl_f,
2048 	.fo_poll =	devfs_poll_f,
2049 	.fo_kqfilter =	devfs_kqfilter_f,
2050 	.fo_stat =	devfs_stat_f,
2051 	.fo_close =	devfs_close_f,
2052 	.fo_chmod =	vn_chmod,
2053 	.fo_chown =	vn_chown,
2054 	.fo_sendfile =	vn_sendfile,
2055 	.fo_seek =	vn_seek,
2056 	.fo_fill_kinfo = vn_fill_kinfo,
2057 	.fo_mmap =	devfs_mmap_f,
2058 	.fo_cmp =	devfs_cmp_f,
2059 	.fo_flags =	DFLAG_PASSABLE | DFLAG_SEEKABLE
2060 };
2061 
2062 /* Vops for non-CHR vnodes in /dev. */
2063 static struct vop_vector devfs_vnodeops = {
2064 	.vop_default =		&default_vnodeops,
2065 
2066 	.vop_access =		devfs_access,
2067 	.vop_getattr =		devfs_getattr,
2068 	.vop_ioctl =		devfs_rioctl,
2069 	.vop_lookup =		devfs_lookup,
2070 	.vop_mknod =		devfs_mknod,
2071 	.vop_pathconf =		devfs_pathconf,
2072 	.vop_read =		devfs_rread,
2073 	.vop_readdir =		devfs_readdir,
2074 	.vop_readlink =		devfs_readlink,
2075 	.vop_reclaim =		devfs_reclaim,
2076 	.vop_remove =		devfs_remove,
2077 	.vop_revoke =		devfs_revoke,
2078 	.vop_setattr =		devfs_setattr,
2079 #ifdef MAC
2080 	.vop_setlabel =		devfs_setlabel,
2081 #endif
2082 	.vop_symlink =		devfs_symlink,
2083 	.vop_vptocnp =		devfs_vptocnp,
2084 	.vop_lock1 =		vop_lock,
2085 	.vop_unlock =		vop_unlock,
2086 	.vop_islocked =		vop_islocked,
2087 	.vop_add_writecount =	vop_stdadd_writecount_nomsync,
2088 };
2089 VFS_VOP_VECTOR_REGISTER(devfs_vnodeops);
2090 
2091 /* Vops for VCHR vnodes in /dev. */
2092 static struct vop_vector devfs_specops = {
2093 	.vop_default =		&default_vnodeops,
2094 
2095 	.vop_access =		devfs_access,
2096 	.vop_bmap =		VOP_PANIC,
2097 	.vop_close =		devfs_close,
2098 	.vop_create =		VOP_PANIC,
2099 	.vop_fsync =		vop_stdfsync,
2100 	.vop_getattr =		devfs_getattr,
2101 	.vop_ioctl =		devfs_ioctl,
2102 	.vop_link =		VOP_PANIC,
2103 	.vop_mkdir =		VOP_PANIC,
2104 	.vop_mknod =		VOP_PANIC,
2105 	.vop_open =		devfs_open,
2106 	.vop_pathconf =		devfs_pathconf,
2107 	.vop_poll =		dead_poll,
2108 	.vop_print =		devfs_print,
2109 	.vop_read =		dead_read,
2110 	.vop_readdir =		VOP_PANIC,
2111 	.vop_readlink =		VOP_PANIC,
2112 	.vop_reallocblks =	VOP_PANIC,
2113 	.vop_reclaim =		devfs_reclaim_vchr,
2114 	.vop_remove =		devfs_remove,
2115 	.vop_rename =		VOP_PANIC,
2116 	.vop_revoke =		devfs_revoke,
2117 	.vop_rmdir =		VOP_PANIC,
2118 	.vop_setattr =		devfs_setattr,
2119 #ifdef MAC
2120 	.vop_setlabel =		devfs_setlabel,
2121 #endif
2122 	.vop_strategy =		VOP_PANIC,
2123 	.vop_symlink =		VOP_PANIC,
2124 	.vop_vptocnp =		devfs_vptocnp,
2125 	.vop_write =		dead_write,
2126 	.vop_lock1 =		vop_lock,
2127 	.vop_unlock =		vop_unlock,
2128 	.vop_islocked =		vop_islocked,
2129 	.vop_add_writecount =	vop_stdadd_writecount_nomsync,
2130 };
2131 VFS_VOP_VECTOR_REGISTER(devfs_specops);
2132 
2133 /*
2134  * Our calling convention to the device drivers used to be that we passed
2135  * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
2136  * flags instead since that's what open(), close() and ioctl() takes and
2137  * we don't really want vnode.h in device drivers.
2138  * We solved the source compatibility by redefining some vnode flags to
2139  * be the same as the fcntl ones and by sending down the bitwise OR of
2140  * the respective fcntl/vnode flags.  These CTASSERTS make sure nobody
2141  * pulls the rug out under this.
2142  */
2143 CTASSERT(O_NONBLOCK == IO_NDELAY);
2144 CTASSERT(O_FSYNC == IO_SYNC);
2145