xref: /freebsd/sys/fs/devfs/devfs_vnops.c (revision 3494f7c0)
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 void
181 devfs_destroy_cdevpriv(struct cdev_privdata *p)
182 {
183 
184 	mtx_assert(&cdevpriv_mtx, MA_OWNED);
185 	KASSERT(p->cdpd_fp->f_cdevpriv == p,
186 	    ("devfs_destoy_cdevpriv %p != %p", p->cdpd_fp->f_cdevpriv, p));
187 	p->cdpd_fp->f_cdevpriv = NULL;
188 	LIST_REMOVE(p, cdpd_list);
189 	mtx_unlock(&cdevpriv_mtx);
190 	(p->cdpd_dtr)(p->cdpd_data);
191 	free(p, M_CDEVPDATA);
192 }
193 
194 static void
195 devfs_fpdrop(struct file *fp)
196 {
197 	struct cdev_privdata *p;
198 
199 	mtx_lock(&cdevpriv_mtx);
200 	if ((p = fp->f_cdevpriv) == NULL) {
201 		mtx_unlock(&cdevpriv_mtx);
202 		return;
203 	}
204 	devfs_destroy_cdevpriv(p);
205 }
206 
207 void
208 devfs_clear_cdevpriv(void)
209 {
210 	struct file *fp;
211 
212 	fp = curthread->td_fpop;
213 	if (fp == NULL)
214 		return;
215 	devfs_fpdrop(fp);
216 }
217 
218 static void
219 devfs_usecount_add(struct vnode *vp)
220 {
221 	struct devfs_dirent *de;
222 	struct cdev *dev;
223 
224 	mtx_lock(&devfs_de_interlock);
225 	VI_LOCK(vp);
226 	VNPASS(vp->v_type == VCHR || vp->v_type == VBAD, vp);
227 	if (VN_IS_DOOMED(vp)) {
228 		goto out_unlock;
229 	}
230 
231 	de = vp->v_data;
232 	dev = vp->v_rdev;
233 	MPASS(de != NULL);
234 	MPASS(dev != NULL);
235 	dev->si_usecount++;
236 	de->de_usecount++;
237 out_unlock:
238 	VI_UNLOCK(vp);
239 	mtx_unlock(&devfs_de_interlock);
240 }
241 
242 static void
243 devfs_usecount_subl(struct vnode *vp)
244 {
245 	struct devfs_dirent *de;
246 	struct cdev *dev;
247 
248 	mtx_assert(&devfs_de_interlock, MA_OWNED);
249 	ASSERT_VI_LOCKED(vp, __func__);
250 	VNPASS(vp->v_type == VCHR || vp->v_type == VBAD, vp);
251 
252 	de = vp->v_data;
253 	dev = vp->v_rdev;
254 	if (de == NULL)
255 		return;
256 	if (dev == NULL) {
257 		MPASS(de->de_usecount == 0);
258 		return;
259 	}
260 	if (dev->si_usecount < de->de_usecount)
261 		panic("%s: si_usecount underflow for dev %p "
262 		    "(has %ld, dirent has %d)\n",
263 		    __func__, dev, dev->si_usecount, de->de_usecount);
264 	if (VN_IS_DOOMED(vp)) {
265 		dev->si_usecount -= de->de_usecount;
266 		de->de_usecount = 0;
267 	} else {
268 		if (de->de_usecount == 0)
269 			panic("%s: de_usecount underflow for dev %p\n",
270 			    __func__, dev);
271 		dev->si_usecount--;
272 		de->de_usecount--;
273 	}
274 }
275 
276 static void
277 devfs_usecount_sub(struct vnode *vp)
278 {
279 
280 	mtx_lock(&devfs_de_interlock);
281 	VI_LOCK(vp);
282 	devfs_usecount_subl(vp);
283 	VI_UNLOCK(vp);
284 	mtx_unlock(&devfs_de_interlock);
285 }
286 
287 static int
288 devfs_usecountl(struct vnode *vp)
289 {
290 
291 	VNPASS(vp->v_type == VCHR, vp);
292 	mtx_assert(&devfs_de_interlock, MA_OWNED);
293 	ASSERT_VI_LOCKED(vp, __func__);
294 	return (vp->v_rdev->si_usecount);
295 }
296 
297 int
298 devfs_usecount(struct vnode *vp)
299 {
300 	int count;
301 
302 	VNPASS(vp->v_type == VCHR, vp);
303 	mtx_lock(&devfs_de_interlock);
304 	VI_LOCK(vp);
305 	count = devfs_usecountl(vp);
306 	VI_UNLOCK(vp);
307 	mtx_unlock(&devfs_de_interlock);
308 	return (count);
309 }
310 
311 void
312 devfs_ctty_ref(struct vnode *vp)
313 {
314 
315 	vrefact(vp);
316 	devfs_usecount_add(vp);
317 }
318 
319 void
320 devfs_ctty_unref(struct vnode *vp)
321 {
322 
323 	devfs_usecount_sub(vp);
324 	vrele(vp);
325 }
326 
327 /*
328  * On success devfs_populate_vp() returns with dmp->dm_lock held.
329  */
330 static int
331 devfs_populate_vp(struct vnode *vp)
332 {
333 	struct devfs_dirent *de;
334 	struct devfs_mount *dmp;
335 	int locked;
336 
337 	ASSERT_VOP_LOCKED(vp, "devfs_populate_vp");
338 
339 	dmp = VFSTODEVFS(vp->v_mount);
340 	if (!devfs_populate_needed(dmp)) {
341 		sx_xlock(&dmp->dm_lock);
342 		goto out_nopopulate;
343 	}
344 
345 	locked = VOP_ISLOCKED(vp);
346 
347 	sx_xlock(&dmp->dm_lock);
348 	DEVFS_DMP_HOLD(dmp);
349 
350 	/* Can't call devfs_populate() with the vnode lock held. */
351 	VOP_UNLOCK(vp);
352 	devfs_populate(dmp);
353 
354 	sx_xunlock(&dmp->dm_lock);
355 	vn_lock(vp, locked | LK_RETRY);
356 	sx_xlock(&dmp->dm_lock);
357 	if (DEVFS_DMP_DROP(dmp)) {
358 		sx_xunlock(&dmp->dm_lock);
359 		devfs_unmount_final(dmp);
360 		return (ERESTART);
361 	}
362 out_nopopulate:
363 	if (VN_IS_DOOMED(vp)) {
364 		sx_xunlock(&dmp->dm_lock);
365 		return (ERESTART);
366 	}
367 	de = vp->v_data;
368 	KASSERT(de != NULL,
369 	    ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed"));
370 	if ((de->de_flags & DE_DOOMED) != 0) {
371 		sx_xunlock(&dmp->dm_lock);
372 		return (ERESTART);
373 	}
374 
375 	return (0);
376 }
377 
378 static int
379 devfs_vptocnp(struct vop_vptocnp_args *ap)
380 {
381 	struct vnode *vp = ap->a_vp;
382 	struct vnode **dvp = ap->a_vpp;
383 	struct devfs_mount *dmp;
384 	char *buf = ap->a_buf;
385 	size_t *buflen = ap->a_buflen;
386 	struct devfs_dirent *dd, *de;
387 	int i, error;
388 
389 	dmp = VFSTODEVFS(vp->v_mount);
390 
391 	error = devfs_populate_vp(vp);
392 	if (error != 0)
393 		return (error);
394 
395 	if (vp->v_type != VCHR && vp->v_type != VDIR) {
396 		error = ENOENT;
397 		goto finished;
398 	}
399 
400 	dd = vp->v_data;
401 	if (vp->v_type == VDIR && dd == dmp->dm_rootdir) {
402 		*dvp = vp;
403 		vref(*dvp);
404 		goto finished;
405 	}
406 
407 	i = *buflen;
408 	i -= dd->de_dirent->d_namlen;
409 	if (i < 0) {
410 		error = ENOMEM;
411 		goto finished;
412 	}
413 	bcopy(dd->de_dirent->d_name, buf + i, dd->de_dirent->d_namlen);
414 	*buflen = i;
415 	de = devfs_parent_dirent(dd);
416 	if (de == NULL) {
417 		error = ENOENT;
418 		goto finished;
419 	}
420 	mtx_lock(&devfs_de_interlock);
421 	*dvp = de->de_vnode;
422 	if (*dvp != NULL) {
423 		VI_LOCK(*dvp);
424 		mtx_unlock(&devfs_de_interlock);
425 		vholdl(*dvp);
426 		VI_UNLOCK(*dvp);
427 		vref(*dvp);
428 		vdrop(*dvp);
429 	} else {
430 		mtx_unlock(&devfs_de_interlock);
431 		error = ENOENT;
432 	}
433 finished:
434 	sx_xunlock(&dmp->dm_lock);
435 	return (error);
436 }
437 
438 /*
439  * Construct the fully qualified path name relative to the mountpoint.
440  * If a NULL cnp is provided, no '/' is appended to the resulting path.
441  */
442 char *
443 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd,
444     struct componentname *cnp)
445 {
446 	int i;
447 	struct devfs_dirent *de;
448 
449 	sx_assert(&dmp->dm_lock, SA_LOCKED);
450 
451 	i = SPECNAMELEN;
452 	buf[i] = '\0';
453 	if (cnp != NULL)
454 		i -= cnp->cn_namelen;
455 	if (i < 0)
456 		 return (NULL);
457 	if (cnp != NULL)
458 		bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
459 	de = dd;
460 	while (de != dmp->dm_rootdir) {
461 		if (cnp != NULL || i < SPECNAMELEN) {
462 			i--;
463 			if (i < 0)
464 				 return (NULL);
465 			buf[i] = '/';
466 		}
467 		i -= de->de_dirent->d_namlen;
468 		if (i < 0)
469 			 return (NULL);
470 		bcopy(de->de_dirent->d_name, buf + i,
471 		    de->de_dirent->d_namlen);
472 		de = devfs_parent_dirent(de);
473 		if (de == NULL)
474 			return (NULL);
475 	}
476 	return (buf + i);
477 }
478 
479 static int
480 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
481 	struct devfs_dirent *de)
482 {
483 	int not_found;
484 
485 	not_found = 0;
486 	if (de->de_flags & DE_DOOMED)
487 		not_found = 1;
488 	if (DEVFS_DE_DROP(de)) {
489 		KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
490 		devfs_dirent_free(de);
491 	}
492 	if (DEVFS_DMP_DROP(dmp)) {
493 		KASSERT(not_found == 1,
494 			("DEVFS mount struct freed before dirent"));
495 		not_found = 2;
496 		sx_xunlock(&dmp->dm_lock);
497 		devfs_unmount_final(dmp);
498 	}
499 	if (not_found == 1 || (drop_dm_lock && not_found != 2))
500 		sx_unlock(&dmp->dm_lock);
501 	return (not_found);
502 }
503 
504 /*
505  * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
506  * it on return.
507  */
508 int
509 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode,
510     struct vnode **vpp)
511 {
512 	int error;
513 	struct vnode *vp;
514 	struct cdev *dev;
515 	struct devfs_mount *dmp;
516 	struct cdevsw *dsw;
517 	enum vgetstate vs;
518 
519 	dmp = VFSTODEVFS(mp);
520 	if (de->de_flags & DE_DOOMED) {
521 		sx_xunlock(&dmp->dm_lock);
522 		return (ENOENT);
523 	}
524 loop:
525 	DEVFS_DE_HOLD(de);
526 	DEVFS_DMP_HOLD(dmp);
527 	mtx_lock(&devfs_de_interlock);
528 	vp = de->de_vnode;
529 	if (vp != NULL) {
530 		vs = vget_prep(vp);
531 		mtx_unlock(&devfs_de_interlock);
532 		sx_xunlock(&dmp->dm_lock);
533 		vget_finish(vp, lockmode | LK_RETRY, vs);
534 		sx_xlock(&dmp->dm_lock);
535 		if (devfs_allocv_drop_refs(0, dmp, de)) {
536 			vput(vp);
537 			return (ENOENT);
538 		}
539 		else if (VN_IS_DOOMED(vp)) {
540 			mtx_lock(&devfs_de_interlock);
541 			if (de->de_vnode == vp) {
542 				de->de_vnode = NULL;
543 				vp->v_data = NULL;
544 			}
545 			mtx_unlock(&devfs_de_interlock);
546 			vput(vp);
547 			goto loop;
548 		}
549 		sx_xunlock(&dmp->dm_lock);
550 		*vpp = vp;
551 		return (0);
552 	}
553 	mtx_unlock(&devfs_de_interlock);
554 	if (de->de_dirent->d_type == DT_CHR) {
555 		if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
556 			devfs_allocv_drop_refs(1, dmp, de);
557 			return (ENOENT);
558 		}
559 		dev = &de->de_cdp->cdp_c;
560 	} else {
561 		dev = NULL;
562 	}
563 	error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
564 	if (error != 0) {
565 		devfs_allocv_drop_refs(1, dmp, de);
566 		printf("devfs_allocv: failed to allocate new vnode\n");
567 		return (error);
568 	}
569 
570 	if (de->de_dirent->d_type == DT_CHR) {
571 		vp->v_type = VCHR;
572 		VI_LOCK(vp);
573 		dev_lock();
574 		dev_refl(dev);
575 		/* XXX: v_rdev should be protect by vnode lock */
576 		vp->v_rdev = dev;
577 		VNPASS(vp->v_usecount == 1, vp);
578 		/* Special casing of ttys for deadfs.  Probably redundant. */
579 		dsw = dev->si_devsw;
580 		if (dsw != NULL && (dsw->d_flags & D_TTY) != 0)
581 			vp->v_vflag |= VV_ISTTY;
582 		dev_unlock();
583 		VI_UNLOCK(vp);
584 		if ((dev->si_flags & SI_ETERNAL) != 0)
585 			vp->v_vflag |= VV_ETERNALDEV;
586 		vp->v_op = &devfs_specops;
587 	} else if (de->de_dirent->d_type == DT_DIR) {
588 		vp->v_type = VDIR;
589 	} else if (de->de_dirent->d_type == DT_LNK) {
590 		vp->v_type = VLNK;
591 	} else {
592 		vp->v_type = VBAD;
593 	}
594 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
595 	VN_LOCK_ASHARE(vp);
596 	mtx_lock(&devfs_de_interlock);
597 	vp->v_data = de;
598 	de->de_vnode = vp;
599 	mtx_unlock(&devfs_de_interlock);
600 	error = insmntque1(vp, mp);
601 	if (error != 0) {
602 		mtx_lock(&devfs_de_interlock);
603 		vp->v_data = NULL;
604 		de->de_vnode = NULL;
605 		mtx_unlock(&devfs_de_interlock);
606 		vgone(vp);
607 		vput(vp);
608 		(void) devfs_allocv_drop_refs(1, dmp, de);
609 		return (error);
610 	}
611 	if (devfs_allocv_drop_refs(0, dmp, de)) {
612 		vgone(vp);
613 		vput(vp);
614 		return (ENOENT);
615 	}
616 #ifdef MAC
617 	mac_devfs_vnode_associate(mp, de, vp);
618 #endif
619 	sx_xunlock(&dmp->dm_lock);
620 	vn_set_state(vp, VSTATE_CONSTRUCTED);
621 	*vpp = vp;
622 	return (0);
623 }
624 
625 static int
626 devfs_access(struct vop_access_args *ap)
627 {
628 	struct vnode *vp = ap->a_vp;
629 	struct devfs_dirent *de;
630 	struct proc *p;
631 	int error;
632 
633 	de = vp->v_data;
634 	if (vp->v_type == VDIR)
635 		de = de->de_dir;
636 
637 	error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
638 	    ap->a_accmode, ap->a_cred);
639 	if (error == 0)
640 		return (0);
641 	if (error != EACCES)
642 		return (error);
643 	p = ap->a_td->td_proc;
644 	/* We do, however, allow access to the controlling terminal */
645 	PROC_LOCK(p);
646 	if (!(p->p_flag & P_CONTROLT)) {
647 		PROC_UNLOCK(p);
648 		return (error);
649 	}
650 	if (p->p_session->s_ttydp == de->de_cdp)
651 		error = 0;
652 	PROC_UNLOCK(p);
653 	return (error);
654 }
655 
656 _Static_assert(((FMASK | FCNTLFLAGS) & (FLASTCLOSE | FREVOKE)) == 0,
657     "devfs-only flag reuse failed");
658 
659 static int
660 devfs_close(struct vop_close_args *ap)
661 {
662 	struct vnode *vp = ap->a_vp, *oldvp;
663 	struct thread *td = ap->a_td;
664 	struct proc *p;
665 	struct cdev *dev = vp->v_rdev;
666 	struct cdevsw *dsw;
667 	struct devfs_dirent *de = vp->v_data;
668 	int dflags, error, ref, vp_locked;
669 
670 	/*
671 	 * XXX: Don't call d_close() if we were called because of
672 	 * XXX: insmntque() failure.
673 	 */
674 	if (vp->v_data == NULL)
675 		return (0);
676 
677 	/*
678 	 * Hack: a tty device that is a controlling terminal
679 	 * has a reference from the session structure.
680 	 * We cannot easily tell that a character device is
681 	 * a controlling terminal, unless it is the closing
682 	 * process' controlling terminal.  In that case,
683 	 * if the reference count is 2 (this last descriptor
684 	 * plus the session), release the reference from the session.
685 	 */
686 	if (de->de_usecount == 2 && td != NULL) {
687 		p = td->td_proc;
688 		PROC_LOCK(p);
689 		if (vp == p->p_session->s_ttyvp) {
690 			PROC_UNLOCK(p);
691 			oldvp = NULL;
692 			sx_xlock(&proctree_lock);
693 			if (vp == p->p_session->s_ttyvp) {
694 				SESS_LOCK(p->p_session);
695 				mtx_lock(&devfs_de_interlock);
696 				VI_LOCK(vp);
697 				if (devfs_usecountl(vp) == 2 && !VN_IS_DOOMED(vp)) {
698 					p->p_session->s_ttyvp = NULL;
699 					p->p_session->s_ttydp = NULL;
700 					oldvp = vp;
701 				}
702 				VI_UNLOCK(vp);
703 				mtx_unlock(&devfs_de_interlock);
704 				SESS_UNLOCK(p->p_session);
705 			}
706 			sx_xunlock(&proctree_lock);
707 			if (oldvp != NULL)
708 				devfs_ctty_unref(oldvp);
709 		} else
710 			PROC_UNLOCK(p);
711 	}
712 	/*
713 	 * We do not want to really close the device if it
714 	 * is still in use unless we are trying to close it
715 	 * forcibly. Since every use (buffer, vnode, swap, cmap)
716 	 * holds a reference to the vnode, and because we mark
717 	 * any other vnodes that alias this device, when the
718 	 * sum of the reference counts on all the aliased
719 	 * vnodes descends to one, we are on last close.
720 	 */
721 	dsw = dev_refthread(dev, &ref);
722 	if (dsw == NULL)
723 		return (ENXIO);
724 	dflags = 0;
725 	mtx_lock(&devfs_de_interlock);
726 	VI_LOCK(vp);
727 	if (devfs_usecountl(vp) == 1)
728 		dflags |= FLASTCLOSE;
729 	devfs_usecount_subl(vp);
730 	mtx_unlock(&devfs_de_interlock);
731 	if (VN_IS_DOOMED(vp)) {
732 		/* Forced close. */
733 		dflags |= FREVOKE | FNONBLOCK;
734 	} else if (dsw->d_flags & D_TRACKCLOSE) {
735 		/* Keep device updated on status. */
736 	} else if ((dflags & FLASTCLOSE) == 0) {
737 		VI_UNLOCK(vp);
738 		dev_relthread(dev, ref);
739 		return (0);
740 	}
741 	vholdnz(vp);
742 	VI_UNLOCK(vp);
743 	vp_locked = VOP_ISLOCKED(vp);
744 	VOP_UNLOCK(vp);
745 	KASSERT(dev->si_refcount > 0,
746 	    ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
747 	error = dsw->d_close(dev, ap->a_fflag | dflags, S_IFCHR, td);
748 	dev_relthread(dev, ref);
749 	vn_lock(vp, vp_locked | LK_RETRY);
750 	vdrop(vp);
751 	return (error);
752 }
753 
754 static int
755 devfs_close_f(struct file *fp, struct thread *td)
756 {
757 	int error;
758 	struct file *fpop;
759 
760 	/*
761 	 * NB: td may be NULL if this descriptor is closed due to
762 	 * garbage collection from a closed UNIX domain socket.
763 	 */
764 	fpop = curthread->td_fpop;
765 	curthread->td_fpop = fp;
766 	error = vnops.fo_close(fp, td);
767 	curthread->td_fpop = fpop;
768 
769 	/*
770 	 * The f_cdevpriv cannot be assigned non-NULL value while we
771 	 * are destroying the file.
772 	 */
773 	if (fp->f_cdevpriv != NULL)
774 		devfs_fpdrop(fp);
775 	return (error);
776 }
777 
778 static int
779 devfs_getattr(struct vop_getattr_args *ap)
780 {
781 	struct vnode *vp = ap->a_vp;
782 	struct vattr *vap = ap->a_vap;
783 	struct devfs_dirent *de;
784 	struct devfs_mount *dmp;
785 	struct cdev *dev;
786 	struct timeval boottime;
787 	int error;
788 
789 	error = devfs_populate_vp(vp);
790 	if (error != 0)
791 		return (error);
792 
793 	dmp = VFSTODEVFS(vp->v_mount);
794 	sx_xunlock(&dmp->dm_lock);
795 
796 	de = vp->v_data;
797 	KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
798 	if (vp->v_type == VDIR) {
799 		de = de->de_dir;
800 		KASSERT(de != NULL,
801 		    ("Null dir dirent in devfs_getattr vp=%p", vp));
802 	}
803 	vap->va_uid = de->de_uid;
804 	vap->va_gid = de->de_gid;
805 	vap->va_mode = de->de_mode;
806 	if (vp->v_type == VLNK)
807 		vap->va_size = strlen(de->de_symlink);
808 	else if (vp->v_type == VDIR)
809 		vap->va_size = vap->va_bytes = DEV_BSIZE;
810 	else
811 		vap->va_size = 0;
812 	if (vp->v_type != VDIR)
813 		vap->va_bytes = 0;
814 	vap->va_blocksize = DEV_BSIZE;
815 	vap->va_type = vp->v_type;
816 
817 	getboottime(&boottime);
818 #define fix(aa)							\
819 	do {							\
820 		if ((aa).tv_sec <= 3600) {			\
821 			(aa).tv_sec = boottime.tv_sec;		\
822 			(aa).tv_nsec = boottime.tv_usec * 1000; \
823 		}						\
824 	} while (0)
825 
826 	if (vp->v_type != VCHR)  {
827 		fix(de->de_atime);
828 		vap->va_atime = de->de_atime;
829 		fix(de->de_mtime);
830 		vap->va_mtime = de->de_mtime;
831 		fix(de->de_ctime);
832 		vap->va_ctime = de->de_ctime;
833 	} else {
834 		dev = vp->v_rdev;
835 		fix(dev->si_atime);
836 		vap->va_atime = dev->si_atime;
837 		fix(dev->si_mtime);
838 		vap->va_mtime = dev->si_mtime;
839 		fix(dev->si_ctime);
840 		vap->va_ctime = dev->si_ctime;
841 
842 		vap->va_rdev = cdev2priv(dev)->cdp_inode;
843 	}
844 	vap->va_gen = 0;
845 	vap->va_flags = 0;
846 	vap->va_filerev = 0;
847 	vap->va_nlink = de->de_links;
848 	vap->va_fileid = de->de_inode;
849 
850 	return (error);
851 }
852 
853 /* ARGSUSED */
854 static int
855 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
856 {
857 	struct file *fpop;
858 	int error;
859 
860 	fpop = td->td_fpop;
861 	td->td_fpop = fp;
862 	error = vnops.fo_ioctl(fp, com, data, cred, td);
863 	td->td_fpop = fpop;
864 	return (error);
865 }
866 
867 void *
868 fiodgname_buf_get_ptr(void *fgnp, u_long com)
869 {
870 	union {
871 		struct fiodgname_arg	fgn;
872 #ifdef COMPAT_FREEBSD32
873 		struct fiodgname_arg32	fgn32;
874 #endif
875 	} *fgnup;
876 
877 	fgnup = fgnp;
878 	switch (com) {
879 	case FIODGNAME:
880 		return (fgnup->fgn.buf);
881 #ifdef COMPAT_FREEBSD32
882 	case FIODGNAME_32:
883 		return ((void *)(uintptr_t)fgnup->fgn32.buf);
884 #endif
885 	default:
886 		panic("Unhandled ioctl command %ld", com);
887 	}
888 }
889 
890 static int
891 devfs_ioctl(struct vop_ioctl_args *ap)
892 {
893 	struct fiodgname_arg *fgn;
894 	struct vnode *vpold, *vp;
895 	struct cdevsw *dsw;
896 	struct thread *td;
897 	struct session *sess;
898 	struct cdev *dev;
899 	int error, ref, i;
900 	const char *p;
901 	u_long com;
902 
903 	vp = ap->a_vp;
904 	com = ap->a_command;
905 	td = ap->a_td;
906 
907 	dsw = devvn_refthread(vp, &dev, &ref);
908 	if (dsw == NULL)
909 		return (ENXIO);
910 	KASSERT(dev->si_refcount > 0,
911 	    ("devfs: un-referenced struct cdev *(%s)", devtoname(dev)));
912 
913 	switch (com) {
914 	case FIODTYPE:
915 		*(int *)ap->a_data = dsw->d_flags & D_TYPEMASK;
916 		error = 0;
917 		break;
918 	case FIODGNAME:
919 #ifdef	COMPAT_FREEBSD32
920 	case FIODGNAME_32:
921 #endif
922 		fgn = ap->a_data;
923 		p = devtoname(dev);
924 		i = strlen(p) + 1;
925 		if (i > fgn->len)
926 			error = EINVAL;
927 		else
928 			error = copyout(p, fiodgname_buf_get_ptr(fgn, com), i);
929 		break;
930 	default:
931 		error = dsw->d_ioctl(dev, com, ap->a_data, ap->a_fflag, td);
932 	}
933 
934 	dev_relthread(dev, ref);
935 	if (error == ENOIOCTL)
936 		error = ENOTTY;
937 
938 	if (error == 0 && com == TIOCSCTTY) {
939 		/*
940 		 * Do nothing if reassigning same control tty, or if the
941 		 * control tty has already disappeared.  If it disappeared,
942 		 * it's because we were racing with TIOCNOTTY.  TIOCNOTTY
943 		 * already took care of releasing the old vnode and we have
944 		 * nothing left to do.
945 		 */
946 		sx_slock(&proctree_lock);
947 		sess = td->td_proc->p_session;
948 		if (sess->s_ttyvp == vp || sess->s_ttyp == NULL) {
949 			sx_sunlock(&proctree_lock);
950 			return (0);
951 		}
952 
953 		devfs_ctty_ref(vp);
954 		SESS_LOCK(sess);
955 		vpold = sess->s_ttyvp;
956 		sess->s_ttyvp = vp;
957 		sess->s_ttydp = cdev2priv(dev);
958 		SESS_UNLOCK(sess);
959 
960 		sx_sunlock(&proctree_lock);
961 
962 		/* Get rid of reference to old control tty */
963 		if (vpold)
964 			devfs_ctty_unref(vpold);
965 	}
966 	return (error);
967 }
968 
969 /* ARGSUSED */
970 static int
971 devfs_kqfilter_f(struct file *fp, struct knote *kn)
972 {
973 	struct cdev *dev;
974 	struct cdevsw *dsw;
975 	int error, ref;
976 	struct file *fpop;
977 	struct thread *td;
978 
979 	td = curthread;
980 	fpop = td->td_fpop;
981 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
982 	if (error)
983 		return (error);
984 	error = dsw->d_kqfilter(dev, kn);
985 	td->td_fpop = fpop;
986 	dev_relthread(dev, ref);
987 	return (error);
988 }
989 
990 static inline int
991 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
992 {
993 	struct cdev_priv *cdp;
994 	struct ucred *dcr;
995 	struct proc *p;
996 	int error;
997 
998 	cdp = de->de_cdp;
999 	if (cdp == NULL)
1000 		return (0);
1001 	dcr = cdp->cdp_c.si_cred;
1002 	if (dcr == NULL)
1003 		return (0);
1004 
1005 	error = prison_check(td->td_ucred, dcr);
1006 	if (error == 0)
1007 		return (0);
1008 	/* We do, however, allow access to the controlling terminal */
1009 	p = td->td_proc;
1010 	PROC_LOCK(p);
1011 	if (!(p->p_flag & P_CONTROLT)) {
1012 		PROC_UNLOCK(p);
1013 		return (error);
1014 	}
1015 	if (p->p_session->s_ttydp == cdp)
1016 		error = 0;
1017 	PROC_UNLOCK(p);
1018 	return (error);
1019 }
1020 
1021 static int
1022 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
1023 {
1024 	struct componentname *cnp;
1025 	struct vnode *dvp, **vpp;
1026 	struct thread *td;
1027 	struct devfs_dirent *de, *dd;
1028 	struct devfs_dirent **dde;
1029 	struct devfs_mount *dmp;
1030 	struct mount *mp;
1031 	struct cdev *cdev;
1032 	int error, flags, nameiop, dvplocked;
1033 	char specname[SPECNAMELEN + 1], *pname;
1034 
1035 	td = curthread;
1036 	cnp = ap->a_cnp;
1037 	vpp = ap->a_vpp;
1038 	dvp = ap->a_dvp;
1039 	pname = cnp->cn_nameptr;
1040 	flags = cnp->cn_flags;
1041 	nameiop = cnp->cn_nameiop;
1042 	mp = dvp->v_mount;
1043 	dmp = VFSTODEVFS(mp);
1044 	dd = dvp->v_data;
1045 	*vpp = NULLVP;
1046 
1047 	if ((flags & ISLASTCN) && nameiop == RENAME)
1048 		return (EOPNOTSUPP);
1049 
1050 	if (dvp->v_type != VDIR)
1051 		return (ENOTDIR);
1052 
1053 	if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
1054 		return (EIO);
1055 
1056 	error = vn_dir_check_exec(dvp, cnp);
1057 	if (error != 0)
1058 		return (error);
1059 
1060 	if (cnp->cn_namelen == 1 && *pname == '.') {
1061 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
1062 			return (EINVAL);
1063 		*vpp = dvp;
1064 		VREF(dvp);
1065 		return (0);
1066 	}
1067 
1068 	if (flags & ISDOTDOT) {
1069 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
1070 			return (EINVAL);
1071 		de = devfs_parent_dirent(dd);
1072 		if (de == NULL)
1073 			return (ENOENT);
1074 		dvplocked = VOP_ISLOCKED(dvp);
1075 		VOP_UNLOCK(dvp);
1076 		error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK,
1077 		    vpp);
1078 		*dm_unlock = 0;
1079 		vn_lock(dvp, dvplocked | LK_RETRY);
1080 		return (error);
1081 	}
1082 
1083 	dd = dvp->v_data;
1084 	de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
1085 	while (de == NULL) {	/* While(...) so we can use break */
1086 
1087 		if (nameiop == DELETE)
1088 			return (ENOENT);
1089 
1090 		/*
1091 		 * OK, we didn't have an entry for the name we were asked for
1092 		 * so we try to see if anybody can create it on demand.
1093 		 */
1094 		pname = devfs_fqpn(specname, dmp, dd, cnp);
1095 		if (pname == NULL)
1096 			break;
1097 
1098 		cdev = NULL;
1099 		DEVFS_DMP_HOLD(dmp);
1100 		sx_xunlock(&dmp->dm_lock);
1101 		EVENTHANDLER_INVOKE(dev_clone,
1102 		    td->td_ucred, pname, strlen(pname), &cdev);
1103 
1104 		if (cdev == NULL)
1105 			sx_xlock(&dmp->dm_lock);
1106 		else if (devfs_populate_vp(dvp) != 0) {
1107 			*dm_unlock = 0;
1108 			sx_xlock(&dmp->dm_lock);
1109 			if (DEVFS_DMP_DROP(dmp)) {
1110 				sx_xunlock(&dmp->dm_lock);
1111 				devfs_unmount_final(dmp);
1112 			} else
1113 				sx_xunlock(&dmp->dm_lock);
1114 			dev_rel(cdev);
1115 			return (ENOENT);
1116 		}
1117 		if (DEVFS_DMP_DROP(dmp)) {
1118 			*dm_unlock = 0;
1119 			sx_xunlock(&dmp->dm_lock);
1120 			devfs_unmount_final(dmp);
1121 			if (cdev != NULL)
1122 				dev_rel(cdev);
1123 			return (ENOENT);
1124 		}
1125 
1126 		if (cdev == NULL)
1127 			break;
1128 
1129 		dev_lock();
1130 		dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
1131 		if (dde != NULL && *dde != NULL)
1132 			de = *dde;
1133 		dev_unlock();
1134 		dev_rel(cdev);
1135 		break;
1136 	}
1137 
1138 	if (de == NULL || de->de_flags & DE_WHITEOUT) {
1139 		if ((nameiop == CREATE || nameiop == RENAME) &&
1140 		    (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
1141 			return (EJUSTRETURN);
1142 		}
1143 		return (ENOENT);
1144 	}
1145 
1146 	if (devfs_prison_check(de, td))
1147 		return (ENOENT);
1148 
1149 	if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
1150 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
1151 		if (error)
1152 			return (error);
1153 		if (*vpp == dvp) {
1154 			VREF(dvp);
1155 			*vpp = dvp;
1156 			return (0);
1157 		}
1158 	}
1159 	error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, vpp);
1160 	*dm_unlock = 0;
1161 	return (error);
1162 }
1163 
1164 static int
1165 devfs_lookup(struct vop_lookup_args *ap)
1166 {
1167 	int j;
1168 	struct devfs_mount *dmp;
1169 	int dm_unlock;
1170 
1171 	if (devfs_populate_vp(ap->a_dvp) != 0)
1172 		return (ENOTDIR);
1173 
1174 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1175 	dm_unlock = 1;
1176 	j = devfs_lookupx(ap, &dm_unlock);
1177 	if (dm_unlock == 1)
1178 		sx_xunlock(&dmp->dm_lock);
1179 	return (j);
1180 }
1181 
1182 static int
1183 devfs_mknod(struct vop_mknod_args *ap)
1184 {
1185 	struct componentname *cnp;
1186 	struct vnode *dvp, **vpp;
1187 	struct devfs_dirent *dd, *de;
1188 	struct devfs_mount *dmp;
1189 	int error;
1190 
1191 	/*
1192 	 * The only type of node we should be creating here is a
1193 	 * character device, for anything else return EOPNOTSUPP.
1194 	 */
1195 	if (ap->a_vap->va_type != VCHR)
1196 		return (EOPNOTSUPP);
1197 	dvp = ap->a_dvp;
1198 	dmp = VFSTODEVFS(dvp->v_mount);
1199 
1200 	cnp = ap->a_cnp;
1201 	vpp = ap->a_vpp;
1202 	dd = dvp->v_data;
1203 
1204 	error = ENOENT;
1205 	sx_xlock(&dmp->dm_lock);
1206 	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1207 		if (cnp->cn_namelen != de->de_dirent->d_namlen)
1208 			continue;
1209 		if (de->de_dirent->d_type == DT_CHR &&
1210 		    (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0)
1211 			continue;
1212 		if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1213 		    de->de_dirent->d_namlen) != 0)
1214 			continue;
1215 		if (de->de_flags & DE_WHITEOUT)
1216 			break;
1217 		goto notfound;
1218 	}
1219 	if (de == NULL)
1220 		goto notfound;
1221 	de->de_flags &= ~DE_WHITEOUT;
1222 	error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1223 	return (error);
1224 notfound:
1225 	sx_xunlock(&dmp->dm_lock);
1226 	return (error);
1227 }
1228 
1229 /* ARGSUSED */
1230 static int
1231 devfs_open(struct vop_open_args *ap)
1232 {
1233 	struct thread *td = ap->a_td;
1234 	struct vnode *vp = ap->a_vp;
1235 	struct cdev *dev = vp->v_rdev;
1236 	struct file *fp = ap->a_fp;
1237 	int error, ref, vlocked;
1238 	struct cdevsw *dsw;
1239 	struct file *fpop;
1240 
1241 	if (vp->v_type == VBLK)
1242 		return (ENXIO);
1243 
1244 	if (dev == NULL)
1245 		return (ENXIO);
1246 
1247 	/* Make this field valid before any I/O in d_open. */
1248 	if (dev->si_iosize_max == 0)
1249 		dev->si_iosize_max = DFLTPHYS;
1250 
1251 	dsw = dev_refthread(dev, &ref);
1252 	if (dsw == NULL)
1253 		return (ENXIO);
1254 	if (fp == NULL && dsw->d_fdopen != NULL) {
1255 		dev_relthread(dev, ref);
1256 		return (ENXIO);
1257 	}
1258 
1259 	if (vp->v_type == VCHR)
1260 		devfs_usecount_add(vp);
1261 
1262 	vlocked = VOP_ISLOCKED(vp);
1263 	VOP_UNLOCK(vp);
1264 
1265 	fpop = td->td_fpop;
1266 	td->td_fpop = fp;
1267 	if (fp != NULL) {
1268 		fp->f_data = dev;
1269 		fp->f_vnode = vp;
1270 	}
1271 	if (dsw->d_fdopen != NULL)
1272 		error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1273 	else
1274 		error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1275 	/* Clean up any cdevpriv upon error. */
1276 	if (error != 0)
1277 		devfs_clear_cdevpriv();
1278 	td->td_fpop = fpop;
1279 
1280 	vn_lock(vp, vlocked | LK_RETRY);
1281 	if (error != 0 && vp->v_type == VCHR)
1282 		devfs_usecount_sub(vp);
1283 
1284 	dev_relthread(dev, ref);
1285 	if (error != 0) {
1286 		if (error == ERESTART)
1287 			error = EINTR;
1288 		return (error);
1289 	}
1290 
1291 #if 0	/* /dev/console */
1292 	KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1293 #else
1294 	if (fp == NULL)
1295 		return (error);
1296 #endif
1297 	if (fp->f_ops == &badfileops)
1298 		finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1299 	return (error);
1300 }
1301 
1302 static int
1303 devfs_pathconf(struct vop_pathconf_args *ap)
1304 {
1305 
1306 	switch (ap->a_name) {
1307 	case _PC_FILESIZEBITS:
1308 		*ap->a_retval = 64;
1309 		return (0);
1310 	case _PC_NAME_MAX:
1311 		*ap->a_retval = NAME_MAX;
1312 		return (0);
1313 	case _PC_LINK_MAX:
1314 		*ap->a_retval = INT_MAX;
1315 		return (0);
1316 	case _PC_SYMLINK_MAX:
1317 		*ap->a_retval = MAXPATHLEN;
1318 		return (0);
1319 	case _PC_MAX_CANON:
1320 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1321 			*ap->a_retval = MAX_CANON;
1322 			return (0);
1323 		}
1324 		return (EINVAL);
1325 	case _PC_MAX_INPUT:
1326 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1327 			*ap->a_retval = MAX_INPUT;
1328 			return (0);
1329 		}
1330 		return (EINVAL);
1331 	case _PC_VDISABLE:
1332 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1333 			*ap->a_retval = _POSIX_VDISABLE;
1334 			return (0);
1335 		}
1336 		return (EINVAL);
1337 	case _PC_MAC_PRESENT:
1338 #ifdef MAC
1339 		/*
1340 		 * If MAC is enabled, devfs automatically supports
1341 		 * trivial non-persistent label storage.
1342 		 */
1343 		*ap->a_retval = 1;
1344 #else
1345 		*ap->a_retval = 0;
1346 #endif
1347 		return (0);
1348 	case _PC_CHOWN_RESTRICTED:
1349 		*ap->a_retval = 1;
1350 		return (0);
1351 	default:
1352 		return (vop_stdpathconf(ap));
1353 	}
1354 	/* NOTREACHED */
1355 }
1356 
1357 /* ARGSUSED */
1358 static int
1359 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1360 {
1361 	struct cdev *dev;
1362 	struct cdevsw *dsw;
1363 	int error, ref;
1364 	struct file *fpop;
1365 
1366 	fpop = td->td_fpop;
1367 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1368 	if (error != 0) {
1369 		error = vnops.fo_poll(fp, events, cred, td);
1370 		return (error);
1371 	}
1372 	error = dsw->d_poll(dev, events, td);
1373 	td->td_fpop = fpop;
1374 	dev_relthread(dev, ref);
1375 	return(error);
1376 }
1377 
1378 /*
1379  * Print out the contents of a special device vnode.
1380  */
1381 static int
1382 devfs_print(struct vop_print_args *ap)
1383 {
1384 
1385 	printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1386 	return (0);
1387 }
1388 
1389 static int
1390 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
1391     int flags, struct thread *td)
1392 {
1393 	struct cdev *dev;
1394 	int ioflag, error, ref;
1395 	ssize_t resid;
1396 	struct cdevsw *dsw;
1397 	struct file *fpop;
1398 
1399 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1400 		return (EINVAL);
1401 	fpop = td->td_fpop;
1402 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1403 	if (error != 0) {
1404 		error = vnops.fo_read(fp, uio, cred, flags, td);
1405 		return (error);
1406 	}
1407 	resid = uio->uio_resid;
1408 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1409 	if (ioflag & O_DIRECT)
1410 		ioflag |= IO_DIRECT;
1411 
1412 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1413 	error = dsw->d_read(dev, uio, ioflag);
1414 	if (uio->uio_resid != resid || (error == 0 && resid != 0))
1415 		devfs_timestamp(&dev->si_atime);
1416 	td->td_fpop = fpop;
1417 	dev_relthread(dev, ref);
1418 
1419 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF_R);
1420 	return (error);
1421 }
1422 
1423 static int
1424 devfs_readdir(struct vop_readdir_args *ap)
1425 {
1426 	int error;
1427 	struct uio *uio;
1428 	struct dirent *dp;
1429 	struct devfs_dirent *dd;
1430 	struct devfs_dirent *de;
1431 	struct devfs_mount *dmp;
1432 	off_t off;
1433 	int *tmp_ncookies = NULL;
1434 
1435 	if (ap->a_vp->v_type != VDIR)
1436 		return (ENOTDIR);
1437 
1438 	uio = ap->a_uio;
1439 	if (uio->uio_offset < 0)
1440 		return (EINVAL);
1441 
1442 	/*
1443 	 * XXX: This is a temporary hack to get around this filesystem not
1444 	 * supporting cookies. We store the location of the ncookies pointer
1445 	 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1446 	 * and set the number of cookies to 0. We then set the pointer to
1447 	 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1448 	 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1449 	 * pointer to its original location before returning to the caller.
1450 	 */
1451 	if (ap->a_ncookies != NULL) {
1452 		tmp_ncookies = ap->a_ncookies;
1453 		*ap->a_ncookies = 0;
1454 		ap->a_ncookies = NULL;
1455 	}
1456 
1457 	dmp = VFSTODEVFS(ap->a_vp->v_mount);
1458 	if (devfs_populate_vp(ap->a_vp) != 0) {
1459 		if (tmp_ncookies != NULL)
1460 			ap->a_ncookies = tmp_ncookies;
1461 		return (EIO);
1462 	}
1463 	error = 0;
1464 	de = ap->a_vp->v_data;
1465 	off = 0;
1466 	TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1467 		KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1468 		if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1469 			continue;
1470 		if (devfs_prison_check(dd, uio->uio_td))
1471 			continue;
1472 		if (dd->de_dirent->d_type == DT_DIR)
1473 			de = dd->de_dir;
1474 		else
1475 			de = dd;
1476 		dp = dd->de_dirent;
1477 		MPASS(dp->d_reclen == GENERIC_DIRSIZ(dp));
1478 		if (dp->d_reclen > uio->uio_resid)
1479 			break;
1480 		dp->d_fileno = de->de_inode;
1481 		/* NOTE: d_off is the offset for the *next* entry. */
1482 		dp->d_off = off + dp->d_reclen;
1483 		if (off >= uio->uio_offset) {
1484 			error = vfs_read_dirent(ap, dp, off);
1485 			if (error)
1486 				break;
1487 		}
1488 		off += dp->d_reclen;
1489 	}
1490 	sx_xunlock(&dmp->dm_lock);
1491 	uio->uio_offset = off;
1492 
1493 	/*
1494 	 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1495 	 * place.
1496 	 */
1497 	if (tmp_ncookies != NULL)
1498 		ap->a_ncookies = tmp_ncookies;
1499 
1500 	return (error);
1501 }
1502 
1503 static int
1504 devfs_readlink(struct vop_readlink_args *ap)
1505 {
1506 	struct devfs_dirent *de;
1507 
1508 	de = ap->a_vp->v_data;
1509 	return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1510 }
1511 
1512 static void
1513 devfs_reclaiml(struct vnode *vp)
1514 {
1515 	struct devfs_dirent *de;
1516 
1517 	mtx_assert(&devfs_de_interlock, MA_OWNED);
1518 	de = vp->v_data;
1519 	if (de != NULL) {
1520 		MPASS(de->de_usecount == 0);
1521 		de->de_vnode = NULL;
1522 		vp->v_data = NULL;
1523 	}
1524 }
1525 
1526 static int
1527 devfs_reclaim(struct vop_reclaim_args *ap)
1528 {
1529 	struct vnode *vp;
1530 
1531 	vp = ap->a_vp;
1532 	mtx_lock(&devfs_de_interlock);
1533 	devfs_reclaiml(vp);
1534 	mtx_unlock(&devfs_de_interlock);
1535 	return (0);
1536 }
1537 
1538 static int
1539 devfs_reclaim_vchr(struct vop_reclaim_args *ap)
1540 {
1541 	struct vnode *vp;
1542 	struct cdev *dev;
1543 
1544 	vp = ap->a_vp;
1545 	MPASS(vp->v_type == VCHR);
1546 
1547 	mtx_lock(&devfs_de_interlock);
1548 	VI_LOCK(vp);
1549 	devfs_usecount_subl(vp);
1550 	devfs_reclaiml(vp);
1551 	mtx_unlock(&devfs_de_interlock);
1552 	dev_lock();
1553 	dev = vp->v_rdev;
1554 	vp->v_rdev = NULL;
1555 	dev_unlock();
1556 	VI_UNLOCK(vp);
1557 	if (dev != NULL)
1558 		dev_rel(dev);
1559 	return (0);
1560 }
1561 
1562 static int
1563 devfs_remove(struct vop_remove_args *ap)
1564 {
1565 	struct vnode *dvp = ap->a_dvp;
1566 	struct vnode *vp = ap->a_vp;
1567 	struct devfs_dirent *dd;
1568 	struct devfs_dirent *de, *de_covered;
1569 	struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1570 
1571 	ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1572 	ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1573 
1574 	sx_xlock(&dmp->dm_lock);
1575 	dd = ap->a_dvp->v_data;
1576 	de = vp->v_data;
1577 	if (de->de_cdp == NULL) {
1578 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1579 		if (de->de_dirent->d_type == DT_LNK) {
1580 			de_covered = devfs_find(dd, de->de_dirent->d_name,
1581 			    de->de_dirent->d_namlen, 0);
1582 			if (de_covered != NULL)
1583 				de_covered->de_flags &= ~DE_COVERED;
1584 		}
1585 		/* We need to unlock dvp because devfs_delete() may lock it. */
1586 		VOP_UNLOCK(vp);
1587 		if (dvp != vp)
1588 			VOP_UNLOCK(dvp);
1589 		devfs_delete(dmp, de, 0);
1590 		sx_xunlock(&dmp->dm_lock);
1591 		if (dvp != vp)
1592 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1593 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1594 	} else {
1595 		de->de_flags |= DE_WHITEOUT;
1596 		sx_xunlock(&dmp->dm_lock);
1597 	}
1598 	return (0);
1599 }
1600 
1601 /*
1602  * Revoke is called on a tty when a terminal session ends.  The vnode
1603  * is orphaned by setting v_op to deadfs so we need to let go of it
1604  * as well so that we create a new one next time around.
1605  *
1606  */
1607 static int
1608 devfs_revoke(struct vop_revoke_args *ap)
1609 {
1610 	struct vnode *vp = ap->a_vp, *vp2;
1611 	struct cdev *dev;
1612 	struct cdev_priv *cdp;
1613 	struct devfs_dirent *de;
1614 	enum vgetstate vs;
1615 	u_int i;
1616 
1617 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1618 
1619 	dev = vp->v_rdev;
1620 	cdp = cdev2priv(dev);
1621 
1622 	dev_lock();
1623 	cdp->cdp_inuse++;
1624 	dev_unlock();
1625 
1626 	vhold(vp);
1627 	vgone(vp);
1628 	vdrop(vp);
1629 
1630 	VOP_UNLOCK(vp);
1631  loop:
1632 	for (;;) {
1633 		mtx_lock(&devfs_de_interlock);
1634 		dev_lock();
1635 		vp2 = NULL;
1636 		for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1637 			de = cdp->cdp_dirents[i];
1638 			if (de == NULL)
1639 				continue;
1640 
1641 			vp2 = de->de_vnode;
1642 			if (vp2 != NULL) {
1643 				dev_unlock();
1644 				vs = vget_prep(vp2);
1645 				mtx_unlock(&devfs_de_interlock);
1646 				if (vget_finish(vp2, LK_EXCLUSIVE, vs) != 0)
1647 					goto loop;
1648 				vhold(vp2);
1649 				vgone(vp2);
1650 				vdrop(vp2);
1651 				vput(vp2);
1652 				break;
1653 			}
1654 		}
1655 		if (vp2 != NULL) {
1656 			continue;
1657 		}
1658 		dev_unlock();
1659 		mtx_unlock(&devfs_de_interlock);
1660 		break;
1661 	}
1662 	dev_lock();
1663 	cdp->cdp_inuse--;
1664 	if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1665 		KASSERT((cdp->cdp_flags & CDP_ON_ACTIVE_LIST) != 0,
1666 		    ("%s: cdp %p (%s) not on active list",
1667 		    __func__, cdp, dev->si_name));
1668 		cdp->cdp_flags &= ~CDP_ON_ACTIVE_LIST;
1669 		TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1670 		dev_unlock();
1671 		dev_rel(&cdp->cdp_c);
1672 	} else
1673 		dev_unlock();
1674 
1675 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1676 	return (0);
1677 }
1678 
1679 static int
1680 devfs_rioctl(struct vop_ioctl_args *ap)
1681 {
1682 	struct vnode *vp;
1683 	struct devfs_mount *dmp;
1684 	int error;
1685 
1686 	vp = ap->a_vp;
1687 	vn_lock(vp, LK_SHARED | LK_RETRY);
1688 	if (VN_IS_DOOMED(vp)) {
1689 		VOP_UNLOCK(vp);
1690 		return (EBADF);
1691 	}
1692 	dmp = VFSTODEVFS(vp->v_mount);
1693 	sx_xlock(&dmp->dm_lock);
1694 	VOP_UNLOCK(vp);
1695 	DEVFS_DMP_HOLD(dmp);
1696 	devfs_populate(dmp);
1697 	if (DEVFS_DMP_DROP(dmp)) {
1698 		sx_xunlock(&dmp->dm_lock);
1699 		devfs_unmount_final(dmp);
1700 		return (ENOENT);
1701 	}
1702 	error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1703 	sx_xunlock(&dmp->dm_lock);
1704 	return (error);
1705 }
1706 
1707 static int
1708 devfs_rread(struct vop_read_args *ap)
1709 {
1710 
1711 	if (ap->a_vp->v_type != VDIR)
1712 		return (EINVAL);
1713 	return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1714 }
1715 
1716 static int
1717 devfs_setattr(struct vop_setattr_args *ap)
1718 {
1719 	struct devfs_dirent *de;
1720 	struct vattr *vap;
1721 	struct vnode *vp;
1722 	struct thread *td;
1723 	int c, error;
1724 	uid_t uid;
1725 	gid_t gid;
1726 
1727 	vap = ap->a_vap;
1728 	vp = ap->a_vp;
1729 	td = curthread;
1730 	if ((vap->va_type != VNON) ||
1731 	    (vap->va_nlink != VNOVAL) ||
1732 	    (vap->va_fsid != VNOVAL) ||
1733 	    (vap->va_fileid != VNOVAL) ||
1734 	    (vap->va_blocksize != VNOVAL) ||
1735 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1736 	    (vap->va_rdev != VNOVAL) ||
1737 	    ((int)vap->va_bytes != VNOVAL) ||
1738 	    (vap->va_gen != VNOVAL)) {
1739 		return (EINVAL);
1740 	}
1741 
1742 	error = devfs_populate_vp(vp);
1743 	if (error != 0)
1744 		return (error);
1745 
1746 	de = vp->v_data;
1747 	if (vp->v_type == VDIR)
1748 		de = de->de_dir;
1749 
1750 	c = 0;
1751 	if (vap->va_uid == (uid_t)VNOVAL)
1752 		uid = de->de_uid;
1753 	else
1754 		uid = vap->va_uid;
1755 	if (vap->va_gid == (gid_t)VNOVAL)
1756 		gid = de->de_gid;
1757 	else
1758 		gid = vap->va_gid;
1759 	if (uid != de->de_uid || gid != de->de_gid) {
1760 		if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1761 		    (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1762 			error = priv_check(td, PRIV_VFS_CHOWN);
1763 			if (error != 0)
1764 				goto ret;
1765 		}
1766 		de->de_uid = uid;
1767 		de->de_gid = gid;
1768 		c = 1;
1769 	}
1770 
1771 	if (vap->va_mode != (mode_t)VNOVAL) {
1772 		if (ap->a_cred->cr_uid != de->de_uid) {
1773 			error = priv_check(td, PRIV_VFS_ADMIN);
1774 			if (error != 0)
1775 				goto ret;
1776 		}
1777 		de->de_mode = vap->va_mode;
1778 		c = 1;
1779 	}
1780 
1781 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1782 		error = vn_utimes_perm(vp, vap, ap->a_cred, td);
1783 		if (error != 0)
1784 			goto ret;
1785 		if (vap->va_atime.tv_sec != VNOVAL) {
1786 			if (vp->v_type == VCHR)
1787 				vp->v_rdev->si_atime = vap->va_atime;
1788 			else
1789 				de->de_atime = vap->va_atime;
1790 		}
1791 		if (vap->va_mtime.tv_sec != VNOVAL) {
1792 			if (vp->v_type == VCHR)
1793 				vp->v_rdev->si_mtime = vap->va_mtime;
1794 			else
1795 				de->de_mtime = vap->va_mtime;
1796 		}
1797 		c = 1;
1798 	}
1799 
1800 	if (c) {
1801 		if (vp->v_type == VCHR)
1802 			vfs_timestamp(&vp->v_rdev->si_ctime);
1803 		else
1804 			vfs_timestamp(&de->de_mtime);
1805 	}
1806 
1807 ret:
1808 	sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock);
1809 	return (error);
1810 }
1811 
1812 #ifdef MAC
1813 static int
1814 devfs_setlabel(struct vop_setlabel_args *ap)
1815 {
1816 	struct vnode *vp;
1817 	struct devfs_dirent *de;
1818 
1819 	vp = ap->a_vp;
1820 	de = vp->v_data;
1821 
1822 	mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1823 	mac_devfs_update(vp->v_mount, de, vp);
1824 
1825 	return (0);
1826 }
1827 #endif
1828 
1829 static int
1830 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred)
1831 {
1832 
1833 	return (vnops.fo_stat(fp, sb, cred));
1834 }
1835 
1836 static int
1837 devfs_symlink(struct vop_symlink_args *ap)
1838 {
1839 	int i, error;
1840 	struct devfs_dirent *dd;
1841 	struct devfs_dirent *de, *de_covered, *de_dotdot;
1842 	struct devfs_mount *dmp;
1843 
1844 	error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1845 	if (error)
1846 		return(error);
1847 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1848 	if (devfs_populate_vp(ap->a_dvp) != 0)
1849 		return (ENOENT);
1850 
1851 	dd = ap->a_dvp->v_data;
1852 	de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1853 	de->de_flags = DE_USER;
1854 	de->de_uid = 0;
1855 	de->de_gid = 0;
1856 	de->de_mode = 0755;
1857 	de->de_inode = alloc_unr(devfs_inos);
1858 	de->de_dir = dd;
1859 	de->de_dirent->d_type = DT_LNK;
1860 	i = strlen(ap->a_target) + 1;
1861 	de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1862 	bcopy(ap->a_target, de->de_symlink, i);
1863 #ifdef MAC
1864 	mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1865 #endif
1866 	de_covered = devfs_find(dd, de->de_dirent->d_name,
1867 	    de->de_dirent->d_namlen, 0);
1868 	if (de_covered != NULL) {
1869 		if ((de_covered->de_flags & DE_USER) != 0) {
1870 			devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1871 			sx_xunlock(&dmp->dm_lock);
1872 			return (EEXIST);
1873 		}
1874 		KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1875 		    ("devfs_symlink: entry %p already covered", de_covered));
1876 		de_covered->de_flags |= DE_COVERED;
1877 	}
1878 
1879 	de_dotdot = TAILQ_FIRST(&dd->de_dlist);		/* "." */
1880 	de_dotdot = TAILQ_NEXT(de_dotdot, de_list);	/* ".." */
1881 	TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1882 	devfs_dir_ref_de(dmp, dd);
1883 	devfs_rules_apply(dmp, de);
1884 
1885 	return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1886 }
1887 
1888 static int
1889 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1890 {
1891 
1892 	return (vnops.fo_truncate(fp, length, cred, td));
1893 }
1894 
1895 static int
1896 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
1897     int flags, struct thread *td)
1898 {
1899 	struct cdev *dev;
1900 	int error, ioflag, ref;
1901 	ssize_t resid;
1902 	struct cdevsw *dsw;
1903 	struct file *fpop;
1904 
1905 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1906 		return (EINVAL);
1907 	fpop = td->td_fpop;
1908 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1909 	if (error != 0) {
1910 		error = vnops.fo_write(fp, uio, cred, flags, td);
1911 		return (error);
1912 	}
1913 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1914 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1915 	if (ioflag & O_DIRECT)
1916 		ioflag |= IO_DIRECT;
1917 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1918 
1919 	resid = uio->uio_resid;
1920 
1921 	error = dsw->d_write(dev, uio, ioflag);
1922 	if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1923 		devfs_timestamp(&dev->si_ctime);
1924 		dev->si_mtime = dev->si_ctime;
1925 	}
1926 	td->td_fpop = fpop;
1927 	dev_relthread(dev, ref);
1928 
1929 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF_W);
1930 	return (error);
1931 }
1932 
1933 static int
1934 devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1935     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1936     struct thread *td)
1937 {
1938 	struct cdev *dev;
1939 	struct cdevsw *dsw;
1940 	struct mount *mp;
1941 	struct vnode *vp;
1942 	struct file *fpop;
1943 	vm_object_t object;
1944 	vm_prot_t maxprot;
1945 	int error, ref;
1946 
1947 	vp = fp->f_vnode;
1948 
1949 	/*
1950 	 * Ensure that file and memory protections are
1951 	 * compatible.
1952 	 */
1953 	mp = vp->v_mount;
1954 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1955 		maxprot = VM_PROT_NONE;
1956 		if ((prot & VM_PROT_EXECUTE) != 0)
1957 			return (EACCES);
1958 	} else
1959 		maxprot = VM_PROT_EXECUTE;
1960 	if ((fp->f_flag & FREAD) != 0)
1961 		maxprot |= VM_PROT_READ;
1962 	else if ((prot & VM_PROT_READ) != 0)
1963 		return (EACCES);
1964 
1965 	/*
1966 	 * If we are sharing potential changes via MAP_SHARED and we
1967 	 * are trying to get write permission although we opened it
1968 	 * without asking for it, bail out.
1969 	 *
1970 	 * Note that most character devices always share mappings.
1971 	 * The one exception is that D_MMAP_ANON devices
1972 	 * (i.e. /dev/zero) permit private writable mappings.
1973 	 *
1974 	 * Rely on vm_mmap_cdev() to fail invalid MAP_PRIVATE requests
1975 	 * as well as updating maxprot to permit writing for
1976 	 * D_MMAP_ANON devices rather than doing that here.
1977 	 */
1978 	if ((flags & MAP_SHARED) != 0) {
1979 		if ((fp->f_flag & FWRITE) != 0)
1980 			maxprot |= VM_PROT_WRITE;
1981 		else if ((prot & VM_PROT_WRITE) != 0)
1982 			return (EACCES);
1983 	}
1984 	maxprot &= cap_maxprot;
1985 
1986 	fpop = td->td_fpop;
1987 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1988 	if (error != 0)
1989 		return (error);
1990 
1991 	error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, dev, dsw, &foff,
1992 	    &object);
1993 	td->td_fpop = fpop;
1994 	dev_relthread(dev, ref);
1995 	if (error != 0)
1996 		return (error);
1997 
1998 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1999 	    foff, FALSE, td);
2000 	if (error != 0)
2001 		vm_object_deallocate(object);
2002 	return (error);
2003 }
2004 
2005 dev_t
2006 dev2udev(struct cdev *x)
2007 {
2008 	if (x == NULL)
2009 		return (NODEV);
2010 	return (cdev2priv(x)->cdp_inode);
2011 }
2012 
2013 static struct fileops devfs_ops_f = {
2014 	.fo_read =	devfs_read_f,
2015 	.fo_write =	devfs_write_f,
2016 	.fo_truncate =	devfs_truncate_f,
2017 	.fo_ioctl =	devfs_ioctl_f,
2018 	.fo_poll =	devfs_poll_f,
2019 	.fo_kqfilter =	devfs_kqfilter_f,
2020 	.fo_stat =	devfs_stat_f,
2021 	.fo_close =	devfs_close_f,
2022 	.fo_chmod =	vn_chmod,
2023 	.fo_chown =	vn_chown,
2024 	.fo_sendfile =	vn_sendfile,
2025 	.fo_seek =	vn_seek,
2026 	.fo_fill_kinfo = vn_fill_kinfo,
2027 	.fo_mmap =	devfs_mmap_f,
2028 	.fo_flags =	DFLAG_PASSABLE | DFLAG_SEEKABLE
2029 };
2030 
2031 /* Vops for non-CHR vnodes in /dev. */
2032 static struct vop_vector devfs_vnodeops = {
2033 	.vop_default =		&default_vnodeops,
2034 
2035 	.vop_access =		devfs_access,
2036 	.vop_getattr =		devfs_getattr,
2037 	.vop_ioctl =		devfs_rioctl,
2038 	.vop_lookup =		devfs_lookup,
2039 	.vop_mknod =		devfs_mknod,
2040 	.vop_pathconf =		devfs_pathconf,
2041 	.vop_read =		devfs_rread,
2042 	.vop_readdir =		devfs_readdir,
2043 	.vop_readlink =		devfs_readlink,
2044 	.vop_reclaim =		devfs_reclaim,
2045 	.vop_remove =		devfs_remove,
2046 	.vop_revoke =		devfs_revoke,
2047 	.vop_setattr =		devfs_setattr,
2048 #ifdef MAC
2049 	.vop_setlabel =		devfs_setlabel,
2050 #endif
2051 	.vop_symlink =		devfs_symlink,
2052 	.vop_vptocnp =		devfs_vptocnp,
2053 	.vop_lock1 =		vop_lock,
2054 	.vop_unlock =		vop_unlock,
2055 	.vop_islocked =		vop_islocked,
2056 	.vop_add_writecount =	vop_stdadd_writecount_nomsync,
2057 };
2058 VFS_VOP_VECTOR_REGISTER(devfs_vnodeops);
2059 
2060 /* Vops for VCHR vnodes in /dev. */
2061 static struct vop_vector devfs_specops = {
2062 	.vop_default =		&default_vnodeops,
2063 
2064 	.vop_access =		devfs_access,
2065 	.vop_bmap =		VOP_PANIC,
2066 	.vop_close =		devfs_close,
2067 	.vop_create =		VOP_PANIC,
2068 	.vop_fsync =		vop_stdfsync,
2069 	.vop_getattr =		devfs_getattr,
2070 	.vop_ioctl =		devfs_ioctl,
2071 	.vop_link =		VOP_PANIC,
2072 	.vop_mkdir =		VOP_PANIC,
2073 	.vop_mknod =		VOP_PANIC,
2074 	.vop_open =		devfs_open,
2075 	.vop_pathconf =		devfs_pathconf,
2076 	.vop_poll =		dead_poll,
2077 	.vop_print =		devfs_print,
2078 	.vop_read =		dead_read,
2079 	.vop_readdir =		VOP_PANIC,
2080 	.vop_readlink =		VOP_PANIC,
2081 	.vop_reallocblks =	VOP_PANIC,
2082 	.vop_reclaim =		devfs_reclaim_vchr,
2083 	.vop_remove =		devfs_remove,
2084 	.vop_rename =		VOP_PANIC,
2085 	.vop_revoke =		devfs_revoke,
2086 	.vop_rmdir =		VOP_PANIC,
2087 	.vop_setattr =		devfs_setattr,
2088 #ifdef MAC
2089 	.vop_setlabel =		devfs_setlabel,
2090 #endif
2091 	.vop_strategy =		VOP_PANIC,
2092 	.vop_symlink =		VOP_PANIC,
2093 	.vop_vptocnp =		devfs_vptocnp,
2094 	.vop_write =		dead_write,
2095 	.vop_lock1 =		vop_lock,
2096 	.vop_unlock =		vop_unlock,
2097 	.vop_islocked =		vop_islocked,
2098 	.vop_add_writecount =	vop_stdadd_writecount_nomsync,
2099 };
2100 VFS_VOP_VECTOR_REGISTER(devfs_specops);
2101 
2102 /*
2103  * Our calling convention to the device drivers used to be that we passed
2104  * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
2105  * flags instead since that's what open(), close() and ioctl() takes and
2106  * we don't really want vnode.h in device drivers.
2107  * We solved the source compatibility by redefining some vnode flags to
2108  * be the same as the fcntl ones and by sending down the bitwise OR of
2109  * the respective fcntl/vnode flags.  These CTASSERTS make sure nobody
2110  * pulls the rug out under this.
2111  */
2112 CTASSERT(O_NONBLOCK == IO_NDELAY);
2113 CTASSERT(O_FSYNC == IO_SYNC);
2114