xref: /freebsd/sys/fs/devfs/devfs_vnops.c (revision 6829dae12bb055451fa467da4589c43bd03b1e64)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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  * $FreeBSD$
37  */
38 
39 /*
40  * TODO:
41  *	mkdir: want it ?
42  */
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/conf.h>
47 #include <sys/dirent.h>
48 #include <sys/fcntl.h>
49 #include <sys/file.h>
50 #include <sys/filedesc.h>
51 #include <sys/filio.h>
52 #include <sys/jail.h>
53 #include <sys/kernel.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 sx	clone_drain_lock;
87 SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock");
88 struct mtx	cdevpriv_mtx;
89 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
90 
91 SYSCTL_DECL(_vfs_devfs);
92 
93 static int devfs_dotimes;
94 SYSCTL_INT(_vfs_devfs, OID_AUTO, dotimes, CTLFLAG_RW,
95     &devfs_dotimes, 0, "Update timestamps on DEVFS with default precision");
96 
97 /*
98  * Update devfs node timestamp.  Note that updates are unlocked and
99  * stat(2) could see partially updated times.
100  */
101 static void
102 devfs_timestamp(struct timespec *tsp)
103 {
104 	time_t ts;
105 
106 	if (devfs_dotimes) {
107 		vfs_timestamp(tsp);
108 	} else {
109 		ts = time_second;
110 		if (tsp->tv_sec != ts) {
111 			tsp->tv_sec = ts;
112 			tsp->tv_nsec = 0;
113 		}
114 	}
115 }
116 
117 static int
118 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp,
119     int *ref)
120 {
121 
122 	*dswp = devvn_refthread(fp->f_vnode, devp, ref);
123 	if (*devp != fp->f_data) {
124 		if (*dswp != NULL)
125 			dev_relthread(*devp, *ref);
126 		return (ENXIO);
127 	}
128 	KASSERT((*devp)->si_refcount > 0,
129 	    ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
130 	if (*dswp == NULL)
131 		return (ENXIO);
132 	curthread->td_fpop = fp;
133 	return (0);
134 }
135 
136 int
137 devfs_get_cdevpriv(void **datap)
138 {
139 	struct file *fp;
140 	struct cdev_privdata *p;
141 	int error;
142 
143 	fp = curthread->td_fpop;
144 	if (fp == NULL)
145 		return (EBADF);
146 	p = fp->f_cdevpriv;
147 	if (p != NULL) {
148 		error = 0;
149 		*datap = p->cdpd_data;
150 	} else
151 		error = ENOENT;
152 	return (error);
153 }
154 
155 int
156 devfs_set_cdevpriv(void *priv, d_priv_dtor_t *priv_dtr)
157 {
158 	struct file *fp;
159 	struct cdev_priv *cdp;
160 	struct cdev_privdata *p;
161 	int error;
162 
163 	fp = curthread->td_fpop;
164 	if (fp == NULL)
165 		return (ENOENT);
166 	cdp = cdev2priv((struct cdev *)fp->f_data);
167 	p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
168 	p->cdpd_data = priv;
169 	p->cdpd_dtr = priv_dtr;
170 	p->cdpd_fp = fp;
171 	mtx_lock(&cdevpriv_mtx);
172 	if (fp->f_cdevpriv == NULL) {
173 		LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
174 		fp->f_cdevpriv = p;
175 		mtx_unlock(&cdevpriv_mtx);
176 		error = 0;
177 	} else {
178 		mtx_unlock(&cdevpriv_mtx);
179 		free(p, M_CDEVPDATA);
180 		error = EBUSY;
181 	}
182 	return (error);
183 }
184 
185 void
186 devfs_destroy_cdevpriv(struct cdev_privdata *p)
187 {
188 
189 	mtx_assert(&cdevpriv_mtx, MA_OWNED);
190 	KASSERT(p->cdpd_fp->f_cdevpriv == p,
191 	    ("devfs_destoy_cdevpriv %p != %p", p->cdpd_fp->f_cdevpriv, p));
192 	p->cdpd_fp->f_cdevpriv = NULL;
193 	LIST_REMOVE(p, cdpd_list);
194 	mtx_unlock(&cdevpriv_mtx);
195 	(p->cdpd_dtr)(p->cdpd_data);
196 	free(p, M_CDEVPDATA);
197 }
198 
199 static void
200 devfs_fpdrop(struct file *fp)
201 {
202 	struct cdev_privdata *p;
203 
204 	mtx_lock(&cdevpriv_mtx);
205 	if ((p = fp->f_cdevpriv) == NULL) {
206 		mtx_unlock(&cdevpriv_mtx);
207 		return;
208 	}
209 	devfs_destroy_cdevpriv(p);
210 }
211 
212 void
213 devfs_clear_cdevpriv(void)
214 {
215 	struct file *fp;
216 
217 	fp = curthread->td_fpop;
218 	if (fp == NULL)
219 		return;
220 	devfs_fpdrop(fp);
221 }
222 
223 /*
224  * On success devfs_populate_vp() returns with dmp->dm_lock held.
225  */
226 static int
227 devfs_populate_vp(struct vnode *vp)
228 {
229 	struct devfs_dirent *de;
230 	struct devfs_mount *dmp;
231 	int locked;
232 
233 	ASSERT_VOP_LOCKED(vp, "devfs_populate_vp");
234 
235 	dmp = VFSTODEVFS(vp->v_mount);
236 	locked = VOP_ISLOCKED(vp);
237 
238 	sx_xlock(&dmp->dm_lock);
239 	DEVFS_DMP_HOLD(dmp);
240 
241 	/* Can't call devfs_populate() with the vnode lock held. */
242 	VOP_UNLOCK(vp, 0);
243 	devfs_populate(dmp);
244 
245 	sx_xunlock(&dmp->dm_lock);
246 	vn_lock(vp, locked | LK_RETRY);
247 	sx_xlock(&dmp->dm_lock);
248 	if (DEVFS_DMP_DROP(dmp)) {
249 		sx_xunlock(&dmp->dm_lock);
250 		devfs_unmount_final(dmp);
251 		return (ERESTART);
252 	}
253 	if ((vp->v_iflag & VI_DOOMED) != 0) {
254 		sx_xunlock(&dmp->dm_lock);
255 		return (ERESTART);
256 	}
257 	de = vp->v_data;
258 	KASSERT(de != NULL,
259 	    ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed"));
260 	if ((de->de_flags & DE_DOOMED) != 0) {
261 		sx_xunlock(&dmp->dm_lock);
262 		return (ERESTART);
263 	}
264 
265 	return (0);
266 }
267 
268 static int
269 devfs_vptocnp(struct vop_vptocnp_args *ap)
270 {
271 	struct vnode *vp = ap->a_vp;
272 	struct vnode **dvp = ap->a_vpp;
273 	struct devfs_mount *dmp;
274 	char *buf = ap->a_buf;
275 	int *buflen = ap->a_buflen;
276 	struct devfs_dirent *dd, *de;
277 	int i, error;
278 
279 	dmp = VFSTODEVFS(vp->v_mount);
280 
281 	error = devfs_populate_vp(vp);
282 	if (error != 0)
283 		return (error);
284 
285 	i = *buflen;
286 	dd = vp->v_data;
287 
288 	if (vp->v_type == VCHR) {
289 		i -= strlen(dd->de_cdp->cdp_c.si_name);
290 		if (i < 0) {
291 			error = ENOMEM;
292 			goto finished;
293 		}
294 		bcopy(dd->de_cdp->cdp_c.si_name, buf + i,
295 		    strlen(dd->de_cdp->cdp_c.si_name));
296 		de = dd->de_dir;
297 	} else if (vp->v_type == VDIR) {
298 		if (dd == dmp->dm_rootdir) {
299 			*dvp = vp;
300 			vref(*dvp);
301 			goto finished;
302 		}
303 		i -= dd->de_dirent->d_namlen;
304 		if (i < 0) {
305 			error = ENOMEM;
306 			goto finished;
307 		}
308 		bcopy(dd->de_dirent->d_name, buf + i,
309 		    dd->de_dirent->d_namlen);
310 		de = dd;
311 	} else {
312 		error = ENOENT;
313 		goto finished;
314 	}
315 	*buflen = i;
316 	de = devfs_parent_dirent(de);
317 	if (de == NULL) {
318 		error = ENOENT;
319 		goto finished;
320 	}
321 	mtx_lock(&devfs_de_interlock);
322 	*dvp = de->de_vnode;
323 	if (*dvp != NULL) {
324 		VI_LOCK(*dvp);
325 		mtx_unlock(&devfs_de_interlock);
326 		vholdl(*dvp);
327 		VI_UNLOCK(*dvp);
328 		vref(*dvp);
329 		vdrop(*dvp);
330 	} else {
331 		mtx_unlock(&devfs_de_interlock);
332 		error = ENOENT;
333 	}
334 finished:
335 	sx_xunlock(&dmp->dm_lock);
336 	return (error);
337 }
338 
339 /*
340  * Construct the fully qualified path name relative to the mountpoint.
341  * If a NULL cnp is provided, no '/' is appended to the resulting path.
342  */
343 char *
344 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd,
345     struct componentname *cnp)
346 {
347 	int i;
348 	struct devfs_dirent *de;
349 
350 	sx_assert(&dmp->dm_lock, SA_LOCKED);
351 
352 	i = SPECNAMELEN;
353 	buf[i] = '\0';
354 	if (cnp != NULL)
355 		i -= cnp->cn_namelen;
356 	if (i < 0)
357 		 return (NULL);
358 	if (cnp != NULL)
359 		bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
360 	de = dd;
361 	while (de != dmp->dm_rootdir) {
362 		if (cnp != NULL || i < SPECNAMELEN) {
363 			i--;
364 			if (i < 0)
365 				 return (NULL);
366 			buf[i] = '/';
367 		}
368 		i -= de->de_dirent->d_namlen;
369 		if (i < 0)
370 			 return (NULL);
371 		bcopy(de->de_dirent->d_name, buf + i,
372 		    de->de_dirent->d_namlen);
373 		de = devfs_parent_dirent(de);
374 		if (de == NULL)
375 			return (NULL);
376 	}
377 	return (buf + i);
378 }
379 
380 static int
381 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
382 	struct devfs_dirent *de)
383 {
384 	int not_found;
385 
386 	not_found = 0;
387 	if (de->de_flags & DE_DOOMED)
388 		not_found = 1;
389 	if (DEVFS_DE_DROP(de)) {
390 		KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
391 		devfs_dirent_free(de);
392 	}
393 	if (DEVFS_DMP_DROP(dmp)) {
394 		KASSERT(not_found == 1,
395 			("DEVFS mount struct freed before dirent"));
396 		not_found = 2;
397 		sx_xunlock(&dmp->dm_lock);
398 		devfs_unmount_final(dmp);
399 	}
400 	if (not_found == 1 || (drop_dm_lock && not_found != 2))
401 		sx_unlock(&dmp->dm_lock);
402 	return (not_found);
403 }
404 
405 static void
406 devfs_insmntque_dtr(struct vnode *vp, void *arg)
407 {
408 	struct devfs_dirent *de;
409 
410 	de = (struct devfs_dirent *)arg;
411 	mtx_lock(&devfs_de_interlock);
412 	vp->v_data = NULL;
413 	de->de_vnode = NULL;
414 	mtx_unlock(&devfs_de_interlock);
415 	vgone(vp);
416 	vput(vp);
417 }
418 
419 /*
420  * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
421  * it on return.
422  */
423 int
424 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode,
425     struct vnode **vpp)
426 {
427 	int error;
428 	struct vnode *vp;
429 	struct cdev *dev;
430 	struct devfs_mount *dmp;
431 	struct cdevsw *dsw;
432 
433 	dmp = VFSTODEVFS(mp);
434 	if (de->de_flags & DE_DOOMED) {
435 		sx_xunlock(&dmp->dm_lock);
436 		return (ENOENT);
437 	}
438 loop:
439 	DEVFS_DE_HOLD(de);
440 	DEVFS_DMP_HOLD(dmp);
441 	mtx_lock(&devfs_de_interlock);
442 	vp = de->de_vnode;
443 	if (vp != NULL) {
444 		VI_LOCK(vp);
445 		mtx_unlock(&devfs_de_interlock);
446 		sx_xunlock(&dmp->dm_lock);
447 		vget(vp, lockmode | LK_INTERLOCK | LK_RETRY, curthread);
448 		sx_xlock(&dmp->dm_lock);
449 		if (devfs_allocv_drop_refs(0, dmp, de)) {
450 			vput(vp);
451 			return (ENOENT);
452 		}
453 		else if ((vp->v_iflag & VI_DOOMED) != 0) {
454 			mtx_lock(&devfs_de_interlock);
455 			if (de->de_vnode == vp) {
456 				de->de_vnode = NULL;
457 				vp->v_data = NULL;
458 			}
459 			mtx_unlock(&devfs_de_interlock);
460 			vput(vp);
461 			goto loop;
462 		}
463 		sx_xunlock(&dmp->dm_lock);
464 		*vpp = vp;
465 		return (0);
466 	}
467 	mtx_unlock(&devfs_de_interlock);
468 	if (de->de_dirent->d_type == DT_CHR) {
469 		if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
470 			devfs_allocv_drop_refs(1, dmp, de);
471 			return (ENOENT);
472 		}
473 		dev = &de->de_cdp->cdp_c;
474 	} else {
475 		dev = NULL;
476 	}
477 	error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
478 	if (error != 0) {
479 		devfs_allocv_drop_refs(1, dmp, de);
480 		printf("devfs_allocv: failed to allocate new vnode\n");
481 		return (error);
482 	}
483 
484 	if (de->de_dirent->d_type == DT_CHR) {
485 		vp->v_type = VCHR;
486 		VI_LOCK(vp);
487 		dev_lock();
488 		dev_refl(dev);
489 		/* XXX: v_rdev should be protect by vnode lock */
490 		vp->v_rdev = dev;
491 		KASSERT(vp->v_usecount == 1,
492 		    ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount));
493 		dev->si_usecount += vp->v_usecount;
494 		/* Special casing of ttys for deadfs.  Probably redundant. */
495 		dsw = dev->si_devsw;
496 		if (dsw != NULL && (dsw->d_flags & D_TTY) != 0)
497 			vp->v_vflag |= VV_ISTTY;
498 		dev_unlock();
499 		VI_UNLOCK(vp);
500 		if ((dev->si_flags & SI_ETERNAL) != 0)
501 			vp->v_vflag |= VV_ETERNALDEV;
502 		vp->v_op = &devfs_specops;
503 	} else if (de->de_dirent->d_type == DT_DIR) {
504 		vp->v_type = VDIR;
505 	} else if (de->de_dirent->d_type == DT_LNK) {
506 		vp->v_type = VLNK;
507 	} else {
508 		vp->v_type = VBAD;
509 	}
510 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
511 	VN_LOCK_ASHARE(vp);
512 	mtx_lock(&devfs_de_interlock);
513 	vp->v_data = de;
514 	de->de_vnode = vp;
515 	mtx_unlock(&devfs_de_interlock);
516 	error = insmntque1(vp, mp, devfs_insmntque_dtr, de);
517 	if (error != 0) {
518 		(void) devfs_allocv_drop_refs(1, dmp, de);
519 		return (error);
520 	}
521 	if (devfs_allocv_drop_refs(0, dmp, de)) {
522 		vput(vp);
523 		return (ENOENT);
524 	}
525 #ifdef MAC
526 	mac_devfs_vnode_associate(mp, de, vp);
527 #endif
528 	sx_xunlock(&dmp->dm_lock);
529 	*vpp = vp;
530 	return (0);
531 }
532 
533 static int
534 devfs_access(struct vop_access_args *ap)
535 {
536 	struct vnode *vp = ap->a_vp;
537 	struct devfs_dirent *de;
538 	struct proc *p;
539 	int error;
540 
541 	de = vp->v_data;
542 	if (vp->v_type == VDIR)
543 		de = de->de_dir;
544 
545 	error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
546 	    ap->a_accmode, ap->a_cred, NULL);
547 	if (error == 0)
548 		return (0);
549 	if (error != EACCES)
550 		return (error);
551 	p = ap->a_td->td_proc;
552 	/* We do, however, allow access to the controlling terminal */
553 	PROC_LOCK(p);
554 	if (!(p->p_flag & P_CONTROLT)) {
555 		PROC_UNLOCK(p);
556 		return (error);
557 	}
558 	if (p->p_session->s_ttydp == de->de_cdp)
559 		error = 0;
560 	PROC_UNLOCK(p);
561 	return (error);
562 }
563 
564 _Static_assert(((FMASK | FCNTLFLAGS) & (FLASTCLOSE | FREVOKE)) == 0,
565     "devfs-only flag reuse failed");
566 
567 static int
568 devfs_close(struct vop_close_args *ap)
569 {
570 	struct vnode *vp = ap->a_vp, *oldvp;
571 	struct thread *td = ap->a_td;
572 	struct proc *p;
573 	struct cdev *dev = vp->v_rdev;
574 	struct cdevsw *dsw;
575 	int dflags, error, ref, vp_locked;
576 
577 	/*
578 	 * XXX: Don't call d_close() if we were called because of
579 	 * XXX: insmntque1() failure.
580 	 */
581 	if (vp->v_data == NULL)
582 		return (0);
583 
584 	/*
585 	 * Hack: a tty device that is a controlling terminal
586 	 * has a reference from the session structure.
587 	 * We cannot easily tell that a character device is
588 	 * a controlling terminal, unless it is the closing
589 	 * process' controlling terminal.  In that case,
590 	 * if the reference count is 2 (this last descriptor
591 	 * plus the session), release the reference from the session.
592 	 */
593 	if (td != NULL) {
594 		p = td->td_proc;
595 		PROC_LOCK(p);
596 		if (vp == p->p_session->s_ttyvp) {
597 			PROC_UNLOCK(p);
598 			oldvp = NULL;
599 			sx_xlock(&proctree_lock);
600 			if (vp == p->p_session->s_ttyvp) {
601 				SESS_LOCK(p->p_session);
602 				VI_LOCK(vp);
603 				if (count_dev(dev) == 2 &&
604 				    (vp->v_iflag & VI_DOOMED) == 0) {
605 					p->p_session->s_ttyvp = NULL;
606 					p->p_session->s_ttydp = NULL;
607 					oldvp = vp;
608 				}
609 				VI_UNLOCK(vp);
610 				SESS_UNLOCK(p->p_session);
611 			}
612 			sx_xunlock(&proctree_lock);
613 			if (oldvp != NULL)
614 				vrele(oldvp);
615 		} else
616 			PROC_UNLOCK(p);
617 	}
618 	/*
619 	 * We do not want to really close the device if it
620 	 * is still in use unless we are trying to close it
621 	 * forcibly. Since every use (buffer, vnode, swap, cmap)
622 	 * holds a reference to the vnode, and because we mark
623 	 * any other vnodes that alias this device, when the
624 	 * sum of the reference counts on all the aliased
625 	 * vnodes descends to one, we are on last close.
626 	 */
627 	dsw = dev_refthread(dev, &ref);
628 	if (dsw == NULL)
629 		return (ENXIO);
630 	dflags = 0;
631 	VI_LOCK(vp);
632 	if (vp->v_iflag & VI_DOOMED) {
633 		/* Forced close. */
634 		dflags |= FREVOKE | FNONBLOCK;
635 	} else if (dsw->d_flags & D_TRACKCLOSE) {
636 		/* Keep device updated on status. */
637 	} else if (count_dev(dev) > 1) {
638 		VI_UNLOCK(vp);
639 		dev_relthread(dev, ref);
640 		return (0);
641 	}
642 	if (count_dev(dev) == 1)
643 		dflags |= FLASTCLOSE;
644 	vholdl(vp);
645 	VI_UNLOCK(vp);
646 	vp_locked = VOP_ISLOCKED(vp);
647 	VOP_UNLOCK(vp, 0);
648 	KASSERT(dev->si_refcount > 0,
649 	    ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
650 	error = dsw->d_close(dev, ap->a_fflag | dflags, S_IFCHR, td);
651 	dev_relthread(dev, ref);
652 	vn_lock(vp, vp_locked | LK_RETRY);
653 	vdrop(vp);
654 	return (error);
655 }
656 
657 static int
658 devfs_close_f(struct file *fp, struct thread *td)
659 {
660 	int error;
661 	struct file *fpop;
662 
663 	/*
664 	 * NB: td may be NULL if this descriptor is closed due to
665 	 * garbage collection from a closed UNIX domain socket.
666 	 */
667 	fpop = curthread->td_fpop;
668 	curthread->td_fpop = fp;
669 	error = vnops.fo_close(fp, td);
670 	curthread->td_fpop = fpop;
671 
672 	/*
673 	 * The f_cdevpriv cannot be assigned non-NULL value while we
674 	 * are destroying the file.
675 	 */
676 	if (fp->f_cdevpriv != NULL)
677 		devfs_fpdrop(fp);
678 	return (error);
679 }
680 
681 static int
682 devfs_getattr(struct vop_getattr_args *ap)
683 {
684 	struct vnode *vp = ap->a_vp;
685 	struct vattr *vap = ap->a_vap;
686 	struct devfs_dirent *de;
687 	struct devfs_mount *dmp;
688 	struct cdev *dev;
689 	struct timeval boottime;
690 	int error;
691 
692 	error = devfs_populate_vp(vp);
693 	if (error != 0)
694 		return (error);
695 
696 	dmp = VFSTODEVFS(vp->v_mount);
697 	sx_xunlock(&dmp->dm_lock);
698 
699 	de = vp->v_data;
700 	KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
701 	if (vp->v_type == VDIR) {
702 		de = de->de_dir;
703 		KASSERT(de != NULL,
704 		    ("Null dir dirent in devfs_getattr vp=%p", vp));
705 	}
706 	vap->va_uid = de->de_uid;
707 	vap->va_gid = de->de_gid;
708 	vap->va_mode = de->de_mode;
709 	if (vp->v_type == VLNK)
710 		vap->va_size = strlen(de->de_symlink);
711 	else if (vp->v_type == VDIR)
712 		vap->va_size = vap->va_bytes = DEV_BSIZE;
713 	else
714 		vap->va_size = 0;
715 	if (vp->v_type != VDIR)
716 		vap->va_bytes = 0;
717 	vap->va_blocksize = DEV_BSIZE;
718 	vap->va_type = vp->v_type;
719 
720 	getboottime(&boottime);
721 #define fix(aa)							\
722 	do {							\
723 		if ((aa).tv_sec <= 3600) {			\
724 			(aa).tv_sec = boottime.tv_sec;		\
725 			(aa).tv_nsec = boottime.tv_usec * 1000; \
726 		}						\
727 	} while (0)
728 
729 	if (vp->v_type != VCHR)  {
730 		fix(de->de_atime);
731 		vap->va_atime = de->de_atime;
732 		fix(de->de_mtime);
733 		vap->va_mtime = de->de_mtime;
734 		fix(de->de_ctime);
735 		vap->va_ctime = de->de_ctime;
736 	} else {
737 		dev = vp->v_rdev;
738 		fix(dev->si_atime);
739 		vap->va_atime = dev->si_atime;
740 		fix(dev->si_mtime);
741 		vap->va_mtime = dev->si_mtime;
742 		fix(dev->si_ctime);
743 		vap->va_ctime = dev->si_ctime;
744 
745 		vap->va_rdev = cdev2priv(dev)->cdp_inode;
746 	}
747 	vap->va_gen = 0;
748 	vap->va_flags = 0;
749 	vap->va_filerev = 0;
750 	vap->va_nlink = de->de_links;
751 	vap->va_fileid = de->de_inode;
752 
753 	return (error);
754 }
755 
756 /* ARGSUSED */
757 static int
758 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
759 {
760 	struct file *fpop;
761 	int error;
762 
763 	fpop = td->td_fpop;
764 	td->td_fpop = fp;
765 	error = vnops.fo_ioctl(fp, com, data, cred, td);
766 	td->td_fpop = fpop;
767 	return (error);
768 }
769 
770 void *
771 fiodgname_buf_get_ptr(void *fgnp, u_long com)
772 {
773 	union {
774 		struct fiodgname_arg	fgn;
775 #ifdef COMPAT_FREEBSD32
776 		struct fiodgname_arg32	fgn32;
777 #endif
778 	} *fgnup;
779 
780 	fgnup = fgnp;
781 	switch (com) {
782 	case FIODGNAME:
783 		return (fgnup->fgn.buf);
784 #ifdef COMPAT_FREEBSD32
785 	case FIODGNAME_32:
786 		return ((void *)(uintptr_t)fgnup->fgn32.buf);
787 #endif
788 	default:
789 		panic("Unhandled ioctl command %ld", com);
790 	}
791 }
792 
793 static int
794 devfs_ioctl(struct vop_ioctl_args *ap)
795 {
796 	struct fiodgname_arg *fgn;
797 	struct vnode *vpold, *vp;
798 	struct cdevsw *dsw;
799 	struct thread *td;
800 	struct cdev *dev;
801 	int error, ref, i;
802 	const char *p;
803 	u_long com;
804 
805 	vp = ap->a_vp;
806 	com = ap->a_command;
807 	td = ap->a_td;
808 
809 	dsw = devvn_refthread(vp, &dev, &ref);
810 	if (dsw == NULL)
811 		return (ENXIO);
812 	KASSERT(dev->si_refcount > 0,
813 	    ("devfs: un-referenced struct cdev *(%s)", devtoname(dev)));
814 
815 	switch (com) {
816 	case FIODTYPE:
817 		*(int *)ap->a_data = dsw->d_flags & D_TYPEMASK;
818 		error = 0;
819 		break;
820 	case FIODGNAME:
821 #ifdef	COMPAT_FREEBSD32
822 	case FIODGNAME_32:
823 #endif
824 		fgn = ap->a_data;
825 		p = devtoname(dev);
826 		i = strlen(p) + 1;
827 		if (i > fgn->len)
828 			error = EINVAL;
829 		else
830 			error = copyout(p, fiodgname_buf_get_ptr(fgn, com), i);
831 		break;
832 	default:
833 		error = dsw->d_ioctl(dev, com, ap->a_data, ap->a_fflag, td);
834 	}
835 
836 	dev_relthread(dev, ref);
837 	if (error == ENOIOCTL)
838 		error = ENOTTY;
839 
840 	if (error == 0 && com == TIOCSCTTY) {
841 		/* Do nothing if reassigning same control tty */
842 		sx_slock(&proctree_lock);
843 		if (td->td_proc->p_session->s_ttyvp == vp) {
844 			sx_sunlock(&proctree_lock);
845 			return (0);
846 		}
847 
848 		vpold = td->td_proc->p_session->s_ttyvp;
849 		VREF(vp);
850 		SESS_LOCK(td->td_proc->p_session);
851 		td->td_proc->p_session->s_ttyvp = vp;
852 		td->td_proc->p_session->s_ttydp = cdev2priv(dev);
853 		SESS_UNLOCK(td->td_proc->p_session);
854 
855 		sx_sunlock(&proctree_lock);
856 
857 		/* Get rid of reference to old control tty */
858 		if (vpold)
859 			vrele(vpold);
860 	}
861 	return (error);
862 }
863 
864 /* ARGSUSED */
865 static int
866 devfs_kqfilter_f(struct file *fp, struct knote *kn)
867 {
868 	struct cdev *dev;
869 	struct cdevsw *dsw;
870 	int error, ref;
871 	struct file *fpop;
872 	struct thread *td;
873 
874 	td = curthread;
875 	fpop = td->td_fpop;
876 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
877 	if (error)
878 		return (error);
879 	error = dsw->d_kqfilter(dev, kn);
880 	td->td_fpop = fpop;
881 	dev_relthread(dev, ref);
882 	return (error);
883 }
884 
885 static inline int
886 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
887 {
888 	struct cdev_priv *cdp;
889 	struct ucred *dcr;
890 	struct proc *p;
891 	int error;
892 
893 	cdp = de->de_cdp;
894 	if (cdp == NULL)
895 		return (0);
896 	dcr = cdp->cdp_c.si_cred;
897 	if (dcr == NULL)
898 		return (0);
899 
900 	error = prison_check(td->td_ucred, dcr);
901 	if (error == 0)
902 		return (0);
903 	/* We do, however, allow access to the controlling terminal */
904 	p = td->td_proc;
905 	PROC_LOCK(p);
906 	if (!(p->p_flag & P_CONTROLT)) {
907 		PROC_UNLOCK(p);
908 		return (error);
909 	}
910 	if (p->p_session->s_ttydp == cdp)
911 		error = 0;
912 	PROC_UNLOCK(p);
913 	return (error);
914 }
915 
916 static int
917 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
918 {
919 	struct componentname *cnp;
920 	struct vnode *dvp, **vpp;
921 	struct thread *td;
922 	struct devfs_dirent *de, *dd;
923 	struct devfs_dirent **dde;
924 	struct devfs_mount *dmp;
925 	struct mount *mp;
926 	struct cdev *cdev;
927 	int error, flags, nameiop, dvplocked;
928 	char specname[SPECNAMELEN + 1], *pname;
929 
930 	cnp = ap->a_cnp;
931 	vpp = ap->a_vpp;
932 	dvp = ap->a_dvp;
933 	pname = cnp->cn_nameptr;
934 	td = cnp->cn_thread;
935 	flags = cnp->cn_flags;
936 	nameiop = cnp->cn_nameiop;
937 	mp = dvp->v_mount;
938 	dmp = VFSTODEVFS(mp);
939 	dd = dvp->v_data;
940 	*vpp = NULLVP;
941 
942 	if ((flags & ISLASTCN) && nameiop == RENAME)
943 		return (EOPNOTSUPP);
944 
945 	if (dvp->v_type != VDIR)
946 		return (ENOTDIR);
947 
948 	if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
949 		return (EIO);
950 
951 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
952 	if (error)
953 		return (error);
954 
955 	if (cnp->cn_namelen == 1 && *pname == '.') {
956 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
957 			return (EINVAL);
958 		*vpp = dvp;
959 		VREF(dvp);
960 		return (0);
961 	}
962 
963 	if (flags & ISDOTDOT) {
964 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
965 			return (EINVAL);
966 		de = devfs_parent_dirent(dd);
967 		if (de == NULL)
968 			return (ENOENT);
969 		dvplocked = VOP_ISLOCKED(dvp);
970 		VOP_UNLOCK(dvp, 0);
971 		error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK,
972 		    vpp);
973 		*dm_unlock = 0;
974 		vn_lock(dvp, dvplocked | LK_RETRY);
975 		return (error);
976 	}
977 
978 	dd = dvp->v_data;
979 	de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
980 	while (de == NULL) {	/* While(...) so we can use break */
981 
982 		if (nameiop == DELETE)
983 			return (ENOENT);
984 
985 		/*
986 		 * OK, we didn't have an entry for the name we were asked for
987 		 * so we try to see if anybody can create it on demand.
988 		 */
989 		pname = devfs_fqpn(specname, dmp, dd, cnp);
990 		if (pname == NULL)
991 			break;
992 
993 		cdev = NULL;
994 		DEVFS_DMP_HOLD(dmp);
995 		sx_xunlock(&dmp->dm_lock);
996 		sx_slock(&clone_drain_lock);
997 		EVENTHANDLER_INVOKE(dev_clone,
998 		    td->td_ucred, pname, strlen(pname), &cdev);
999 		sx_sunlock(&clone_drain_lock);
1000 
1001 		if (cdev == NULL)
1002 			sx_xlock(&dmp->dm_lock);
1003 		else if (devfs_populate_vp(dvp) != 0) {
1004 			*dm_unlock = 0;
1005 			sx_xlock(&dmp->dm_lock);
1006 			if (DEVFS_DMP_DROP(dmp)) {
1007 				sx_xunlock(&dmp->dm_lock);
1008 				devfs_unmount_final(dmp);
1009 			} else
1010 				sx_xunlock(&dmp->dm_lock);
1011 			dev_rel(cdev);
1012 			return (ENOENT);
1013 		}
1014 		if (DEVFS_DMP_DROP(dmp)) {
1015 			*dm_unlock = 0;
1016 			sx_xunlock(&dmp->dm_lock);
1017 			devfs_unmount_final(dmp);
1018 			if (cdev != NULL)
1019 				dev_rel(cdev);
1020 			return (ENOENT);
1021 		}
1022 
1023 		if (cdev == NULL)
1024 			break;
1025 
1026 		dev_lock();
1027 		dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
1028 		if (dde != NULL && *dde != NULL)
1029 			de = *dde;
1030 		dev_unlock();
1031 		dev_rel(cdev);
1032 		break;
1033 	}
1034 
1035 	if (de == NULL || de->de_flags & DE_WHITEOUT) {
1036 		if ((nameiop == CREATE || nameiop == RENAME) &&
1037 		    (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
1038 			cnp->cn_flags |= SAVENAME;
1039 			return (EJUSTRETURN);
1040 		}
1041 		return (ENOENT);
1042 	}
1043 
1044 	if (devfs_prison_check(de, td))
1045 		return (ENOENT);
1046 
1047 	if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
1048 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
1049 		if (error)
1050 			return (error);
1051 		if (*vpp == dvp) {
1052 			VREF(dvp);
1053 			*vpp = dvp;
1054 			return (0);
1055 		}
1056 	}
1057 	error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, vpp);
1058 	*dm_unlock = 0;
1059 	return (error);
1060 }
1061 
1062 static int
1063 devfs_lookup(struct vop_lookup_args *ap)
1064 {
1065 	int j;
1066 	struct devfs_mount *dmp;
1067 	int dm_unlock;
1068 
1069 	if (devfs_populate_vp(ap->a_dvp) != 0)
1070 		return (ENOTDIR);
1071 
1072 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1073 	dm_unlock = 1;
1074 	j = devfs_lookupx(ap, &dm_unlock);
1075 	if (dm_unlock == 1)
1076 		sx_xunlock(&dmp->dm_lock);
1077 	return (j);
1078 }
1079 
1080 static int
1081 devfs_mknod(struct vop_mknod_args *ap)
1082 {
1083 	struct componentname *cnp;
1084 	struct vnode *dvp, **vpp;
1085 	struct devfs_dirent *dd, *de;
1086 	struct devfs_mount *dmp;
1087 	int error;
1088 
1089 	/*
1090 	 * The only type of node we should be creating here is a
1091 	 * character device, for anything else return EOPNOTSUPP.
1092 	 */
1093 	if (ap->a_vap->va_type != VCHR)
1094 		return (EOPNOTSUPP);
1095 	dvp = ap->a_dvp;
1096 	dmp = VFSTODEVFS(dvp->v_mount);
1097 
1098 	cnp = ap->a_cnp;
1099 	vpp = ap->a_vpp;
1100 	dd = dvp->v_data;
1101 
1102 	error = ENOENT;
1103 	sx_xlock(&dmp->dm_lock);
1104 	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1105 		if (cnp->cn_namelen != de->de_dirent->d_namlen)
1106 			continue;
1107 		if (de->de_dirent->d_type == DT_CHR &&
1108 		    (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0)
1109 			continue;
1110 		if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1111 		    de->de_dirent->d_namlen) != 0)
1112 			continue;
1113 		if (de->de_flags & DE_WHITEOUT)
1114 			break;
1115 		goto notfound;
1116 	}
1117 	if (de == NULL)
1118 		goto notfound;
1119 	de->de_flags &= ~DE_WHITEOUT;
1120 	error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1121 	return (error);
1122 notfound:
1123 	sx_xunlock(&dmp->dm_lock);
1124 	return (error);
1125 }
1126 
1127 /* ARGSUSED */
1128 static int
1129 devfs_open(struct vop_open_args *ap)
1130 {
1131 	struct thread *td = ap->a_td;
1132 	struct vnode *vp = ap->a_vp;
1133 	struct cdev *dev = vp->v_rdev;
1134 	struct file *fp = ap->a_fp;
1135 	int error, ref, vlocked;
1136 	struct cdevsw *dsw;
1137 	struct file *fpop;
1138 	struct mtx *mtxp;
1139 
1140 	if (vp->v_type == VBLK)
1141 		return (ENXIO);
1142 
1143 	if (dev == NULL)
1144 		return (ENXIO);
1145 
1146 	/* Make this field valid before any I/O in d_open. */
1147 	if (dev->si_iosize_max == 0)
1148 		dev->si_iosize_max = DFLTPHYS;
1149 
1150 	dsw = dev_refthread(dev, &ref);
1151 	if (dsw == NULL)
1152 		return (ENXIO);
1153 	if (fp == NULL && dsw->d_fdopen != NULL) {
1154 		dev_relthread(dev, ref);
1155 		return (ENXIO);
1156 	}
1157 
1158 	vlocked = VOP_ISLOCKED(vp);
1159 	VOP_UNLOCK(vp, 0);
1160 
1161 	fpop = td->td_fpop;
1162 	td->td_fpop = fp;
1163 	if (fp != NULL) {
1164 		fp->f_data = dev;
1165 		fp->f_vnode = vp;
1166 	}
1167 	if (dsw->d_fdopen != NULL)
1168 		error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1169 	else
1170 		error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1171 	/* Clean up any cdevpriv upon error. */
1172 	if (error != 0)
1173 		devfs_clear_cdevpriv();
1174 	td->td_fpop = fpop;
1175 
1176 	vn_lock(vp, vlocked | LK_RETRY);
1177 	dev_relthread(dev, ref);
1178 	if (error != 0) {
1179 		if (error == ERESTART)
1180 			error = EINTR;
1181 		return (error);
1182 	}
1183 
1184 #if 0	/* /dev/console */
1185 	KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1186 #else
1187 	if (fp == NULL)
1188 		return (error);
1189 #endif
1190 	if (fp->f_ops == &badfileops)
1191 		finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1192 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
1193 
1194 	/*
1195 	 * Hint to the dofilewrite() to not force the buffer draining
1196 	 * on the writer to the file.  Most likely, the write would
1197 	 * not need normal buffers.
1198 	 */
1199 	mtx_lock(mtxp);
1200 	fp->f_vnread_flags |= FDEVFS_VNODE;
1201 	mtx_unlock(mtxp);
1202 	return (error);
1203 }
1204 
1205 static int
1206 devfs_pathconf(struct vop_pathconf_args *ap)
1207 {
1208 
1209 	switch (ap->a_name) {
1210 	case _PC_FILESIZEBITS:
1211 		*ap->a_retval = 64;
1212 		return (0);
1213 	case _PC_NAME_MAX:
1214 		*ap->a_retval = NAME_MAX;
1215 		return (0);
1216 	case _PC_LINK_MAX:
1217 		*ap->a_retval = INT_MAX;
1218 		return (0);
1219 	case _PC_SYMLINK_MAX:
1220 		*ap->a_retval = MAXPATHLEN;
1221 		return (0);
1222 	case _PC_MAX_CANON:
1223 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1224 			*ap->a_retval = MAX_CANON;
1225 			return (0);
1226 		}
1227 		return (EINVAL);
1228 	case _PC_MAX_INPUT:
1229 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1230 			*ap->a_retval = MAX_INPUT;
1231 			return (0);
1232 		}
1233 		return (EINVAL);
1234 	case _PC_VDISABLE:
1235 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1236 			*ap->a_retval = _POSIX_VDISABLE;
1237 			return (0);
1238 		}
1239 		return (EINVAL);
1240 	case _PC_MAC_PRESENT:
1241 #ifdef MAC
1242 		/*
1243 		 * If MAC is enabled, devfs automatically supports
1244 		 * trivial non-persistant label storage.
1245 		 */
1246 		*ap->a_retval = 1;
1247 #else
1248 		*ap->a_retval = 0;
1249 #endif
1250 		return (0);
1251 	case _PC_CHOWN_RESTRICTED:
1252 		*ap->a_retval = 1;
1253 		return (0);
1254 	default:
1255 		return (vop_stdpathconf(ap));
1256 	}
1257 	/* NOTREACHED */
1258 }
1259 
1260 /* ARGSUSED */
1261 static int
1262 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1263 {
1264 	struct cdev *dev;
1265 	struct cdevsw *dsw;
1266 	int error, ref;
1267 	struct file *fpop;
1268 
1269 	fpop = td->td_fpop;
1270 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1271 	if (error != 0) {
1272 		error = vnops.fo_poll(fp, events, cred, td);
1273 		return (error);
1274 	}
1275 	error = dsw->d_poll(dev, events, td);
1276 	td->td_fpop = fpop;
1277 	dev_relthread(dev, ref);
1278 	return(error);
1279 }
1280 
1281 /*
1282  * Print out the contents of a special device vnode.
1283  */
1284 static int
1285 devfs_print(struct vop_print_args *ap)
1286 {
1287 
1288 	printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1289 	return (0);
1290 }
1291 
1292 static int
1293 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
1294     int flags, struct thread *td)
1295 {
1296 	struct cdev *dev;
1297 	int ioflag, error, ref;
1298 	ssize_t resid;
1299 	struct cdevsw *dsw;
1300 	struct file *fpop;
1301 
1302 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1303 		return (EINVAL);
1304 	fpop = td->td_fpop;
1305 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1306 	if (error != 0) {
1307 		error = vnops.fo_read(fp, uio, cred, flags, td);
1308 		return (error);
1309 	}
1310 	resid = uio->uio_resid;
1311 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1312 	if (ioflag & O_DIRECT)
1313 		ioflag |= IO_DIRECT;
1314 
1315 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1316 	error = dsw->d_read(dev, uio, ioflag);
1317 	if (uio->uio_resid != resid || (error == 0 && resid != 0))
1318 		devfs_timestamp(&dev->si_atime);
1319 	td->td_fpop = fpop;
1320 	dev_relthread(dev, ref);
1321 
1322 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1323 	return (error);
1324 }
1325 
1326 static int
1327 devfs_readdir(struct vop_readdir_args *ap)
1328 {
1329 	int error;
1330 	struct uio *uio;
1331 	struct dirent *dp;
1332 	struct devfs_dirent *dd;
1333 	struct devfs_dirent *de;
1334 	struct devfs_mount *dmp;
1335 	off_t off;
1336 	int *tmp_ncookies = NULL;
1337 
1338 	if (ap->a_vp->v_type != VDIR)
1339 		return (ENOTDIR);
1340 
1341 	uio = ap->a_uio;
1342 	if (uio->uio_offset < 0)
1343 		return (EINVAL);
1344 
1345 	/*
1346 	 * XXX: This is a temporary hack to get around this filesystem not
1347 	 * supporting cookies. We store the location of the ncookies pointer
1348 	 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1349 	 * and set the number of cookies to 0. We then set the pointer to
1350 	 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1351 	 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1352 	 * pointer to its original location before returning to the caller.
1353 	 */
1354 	if (ap->a_ncookies != NULL) {
1355 		tmp_ncookies = ap->a_ncookies;
1356 		*ap->a_ncookies = 0;
1357 		ap->a_ncookies = NULL;
1358 	}
1359 
1360 	dmp = VFSTODEVFS(ap->a_vp->v_mount);
1361 	if (devfs_populate_vp(ap->a_vp) != 0) {
1362 		if (tmp_ncookies != NULL)
1363 			ap->a_ncookies = tmp_ncookies;
1364 		return (EIO);
1365 	}
1366 	error = 0;
1367 	de = ap->a_vp->v_data;
1368 	off = 0;
1369 	TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1370 		KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1371 		if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1372 			continue;
1373 		if (devfs_prison_check(dd, uio->uio_td))
1374 			continue;
1375 		if (dd->de_dirent->d_type == DT_DIR)
1376 			de = dd->de_dir;
1377 		else
1378 			de = dd;
1379 		dp = dd->de_dirent;
1380 		MPASS(dp->d_reclen == GENERIC_DIRSIZ(dp));
1381 		if (dp->d_reclen > uio->uio_resid)
1382 			break;
1383 		dp->d_fileno = de->de_inode;
1384 		/* NOTE: d_off is the offset for the *next* entry. */
1385 		dp->d_off = off + dp->d_reclen;
1386 		if (off >= uio->uio_offset) {
1387 			error = vfs_read_dirent(ap, dp, off);
1388 			if (error)
1389 				break;
1390 		}
1391 		off += dp->d_reclen;
1392 	}
1393 	sx_xunlock(&dmp->dm_lock);
1394 	uio->uio_offset = off;
1395 
1396 	/*
1397 	 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1398 	 * place.
1399 	 */
1400 	if (tmp_ncookies != NULL)
1401 		ap->a_ncookies = tmp_ncookies;
1402 
1403 	return (error);
1404 }
1405 
1406 static int
1407 devfs_readlink(struct vop_readlink_args *ap)
1408 {
1409 	struct devfs_dirent *de;
1410 
1411 	de = ap->a_vp->v_data;
1412 	return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1413 }
1414 
1415 static int
1416 devfs_reclaim(struct vop_reclaim_args *ap)
1417 {
1418 	struct vnode *vp;
1419 	struct devfs_dirent *de;
1420 
1421 	vp = ap->a_vp;
1422 	mtx_lock(&devfs_de_interlock);
1423 	de = vp->v_data;
1424 	if (de != NULL) {
1425 		de->de_vnode = NULL;
1426 		vp->v_data = NULL;
1427 	}
1428 	mtx_unlock(&devfs_de_interlock);
1429 	vnode_destroy_vobject(vp);
1430 	return (0);
1431 }
1432 
1433 static int
1434 devfs_reclaim_vchr(struct vop_reclaim_args *ap)
1435 {
1436 	struct vnode *vp;
1437 	struct cdev *dev;
1438 
1439 	vp = ap->a_vp;
1440 	MPASS(vp->v_type == VCHR);
1441 
1442 	devfs_reclaim(ap);
1443 
1444 	VI_LOCK(vp);
1445 	dev_lock();
1446 	dev = vp->v_rdev;
1447 	vp->v_rdev = NULL;
1448 	if (dev != NULL)
1449 		dev->si_usecount -= vp->v_usecount;
1450 	dev_unlock();
1451 	VI_UNLOCK(vp);
1452 	if (dev != NULL)
1453 		dev_rel(dev);
1454 	return (0);
1455 }
1456 
1457 static int
1458 devfs_remove(struct vop_remove_args *ap)
1459 {
1460 	struct vnode *dvp = ap->a_dvp;
1461 	struct vnode *vp = ap->a_vp;
1462 	struct devfs_dirent *dd;
1463 	struct devfs_dirent *de, *de_covered;
1464 	struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1465 
1466 	ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1467 	ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1468 
1469 	sx_xlock(&dmp->dm_lock);
1470 	dd = ap->a_dvp->v_data;
1471 	de = vp->v_data;
1472 	if (de->de_cdp == NULL) {
1473 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1474 		if (de->de_dirent->d_type == DT_LNK) {
1475 			de_covered = devfs_find(dd, de->de_dirent->d_name,
1476 			    de->de_dirent->d_namlen, 0);
1477 			if (de_covered != NULL)
1478 				de_covered->de_flags &= ~DE_COVERED;
1479 		}
1480 		/* We need to unlock dvp because devfs_delete() may lock it. */
1481 		VOP_UNLOCK(vp, 0);
1482 		if (dvp != vp)
1483 			VOP_UNLOCK(dvp, 0);
1484 		devfs_delete(dmp, de, 0);
1485 		sx_xunlock(&dmp->dm_lock);
1486 		if (dvp != vp)
1487 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1488 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1489 	} else {
1490 		de->de_flags |= DE_WHITEOUT;
1491 		sx_xunlock(&dmp->dm_lock);
1492 	}
1493 	return (0);
1494 }
1495 
1496 /*
1497  * Revoke is called on a tty when a terminal session ends.  The vnode
1498  * is orphaned by setting v_op to deadfs so we need to let go of it
1499  * as well so that we create a new one next time around.
1500  *
1501  */
1502 static int
1503 devfs_revoke(struct vop_revoke_args *ap)
1504 {
1505 	struct vnode *vp = ap->a_vp, *vp2;
1506 	struct cdev *dev;
1507 	struct cdev_priv *cdp;
1508 	struct devfs_dirent *de;
1509 	u_int i;
1510 
1511 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1512 
1513 	dev = vp->v_rdev;
1514 	cdp = cdev2priv(dev);
1515 
1516 	dev_lock();
1517 	cdp->cdp_inuse++;
1518 	dev_unlock();
1519 
1520 	vhold(vp);
1521 	vgone(vp);
1522 	vdrop(vp);
1523 
1524 	VOP_UNLOCK(vp,0);
1525  loop:
1526 	for (;;) {
1527 		mtx_lock(&devfs_de_interlock);
1528 		dev_lock();
1529 		vp2 = NULL;
1530 		for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1531 			de = cdp->cdp_dirents[i];
1532 			if (de == NULL)
1533 				continue;
1534 
1535 			vp2 = de->de_vnode;
1536 			if (vp2 != NULL) {
1537 				dev_unlock();
1538 				VI_LOCK(vp2);
1539 				mtx_unlock(&devfs_de_interlock);
1540 				if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
1541 				    curthread))
1542 					goto loop;
1543 				vhold(vp2);
1544 				vgone(vp2);
1545 				vdrop(vp2);
1546 				vput(vp2);
1547 				break;
1548 			}
1549 		}
1550 		if (vp2 != NULL) {
1551 			continue;
1552 		}
1553 		dev_unlock();
1554 		mtx_unlock(&devfs_de_interlock);
1555 		break;
1556 	}
1557 	dev_lock();
1558 	cdp->cdp_inuse--;
1559 	if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1560 		TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1561 		dev_unlock();
1562 		dev_rel(&cdp->cdp_c);
1563 	} else
1564 		dev_unlock();
1565 
1566 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1567 	return (0);
1568 }
1569 
1570 static int
1571 devfs_rioctl(struct vop_ioctl_args *ap)
1572 {
1573 	struct vnode *vp;
1574 	struct devfs_mount *dmp;
1575 	int error;
1576 
1577 	vp = ap->a_vp;
1578 	vn_lock(vp, LK_SHARED | LK_RETRY);
1579 	if (vp->v_iflag & VI_DOOMED) {
1580 		VOP_UNLOCK(vp, 0);
1581 		return (EBADF);
1582 	}
1583 	dmp = VFSTODEVFS(vp->v_mount);
1584 	sx_xlock(&dmp->dm_lock);
1585 	VOP_UNLOCK(vp, 0);
1586 	DEVFS_DMP_HOLD(dmp);
1587 	devfs_populate(dmp);
1588 	if (DEVFS_DMP_DROP(dmp)) {
1589 		sx_xunlock(&dmp->dm_lock);
1590 		devfs_unmount_final(dmp);
1591 		return (ENOENT);
1592 	}
1593 	error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1594 	sx_xunlock(&dmp->dm_lock);
1595 	return (error);
1596 }
1597 
1598 static int
1599 devfs_rread(struct vop_read_args *ap)
1600 {
1601 
1602 	if (ap->a_vp->v_type != VDIR)
1603 		return (EINVAL);
1604 	return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1605 }
1606 
1607 static int
1608 devfs_setattr(struct vop_setattr_args *ap)
1609 {
1610 	struct devfs_dirent *de;
1611 	struct vattr *vap;
1612 	struct vnode *vp;
1613 	struct thread *td;
1614 	int c, error;
1615 	uid_t uid;
1616 	gid_t gid;
1617 
1618 	vap = ap->a_vap;
1619 	vp = ap->a_vp;
1620 	td = curthread;
1621 	if ((vap->va_type != VNON) ||
1622 	    (vap->va_nlink != VNOVAL) ||
1623 	    (vap->va_fsid != VNOVAL) ||
1624 	    (vap->va_fileid != VNOVAL) ||
1625 	    (vap->va_blocksize != VNOVAL) ||
1626 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1627 	    (vap->va_rdev != VNOVAL) ||
1628 	    ((int)vap->va_bytes != VNOVAL) ||
1629 	    (vap->va_gen != VNOVAL)) {
1630 		return (EINVAL);
1631 	}
1632 
1633 	error = devfs_populate_vp(vp);
1634 	if (error != 0)
1635 		return (error);
1636 
1637 	de = vp->v_data;
1638 	if (vp->v_type == VDIR)
1639 		de = de->de_dir;
1640 
1641 	c = 0;
1642 	if (vap->va_uid == (uid_t)VNOVAL)
1643 		uid = de->de_uid;
1644 	else
1645 		uid = vap->va_uid;
1646 	if (vap->va_gid == (gid_t)VNOVAL)
1647 		gid = de->de_gid;
1648 	else
1649 		gid = vap->va_gid;
1650 	if (uid != de->de_uid || gid != de->de_gid) {
1651 		if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1652 		    (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1653 			error = priv_check(td, PRIV_VFS_CHOWN);
1654 			if (error != 0)
1655 				goto ret;
1656 		}
1657 		de->de_uid = uid;
1658 		de->de_gid = gid;
1659 		c = 1;
1660 	}
1661 
1662 	if (vap->va_mode != (mode_t)VNOVAL) {
1663 		if (ap->a_cred->cr_uid != de->de_uid) {
1664 			error = priv_check(td, PRIV_VFS_ADMIN);
1665 			if (error != 0)
1666 				goto ret;
1667 		}
1668 		de->de_mode = vap->va_mode;
1669 		c = 1;
1670 	}
1671 
1672 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1673 		error = vn_utimes_perm(vp, vap, ap->a_cred, td);
1674 		if (error != 0)
1675 			goto ret;
1676 		if (vap->va_atime.tv_sec != VNOVAL) {
1677 			if (vp->v_type == VCHR)
1678 				vp->v_rdev->si_atime = vap->va_atime;
1679 			else
1680 				de->de_atime = vap->va_atime;
1681 		}
1682 		if (vap->va_mtime.tv_sec != VNOVAL) {
1683 			if (vp->v_type == VCHR)
1684 				vp->v_rdev->si_mtime = vap->va_mtime;
1685 			else
1686 				de->de_mtime = vap->va_mtime;
1687 		}
1688 		c = 1;
1689 	}
1690 
1691 	if (c) {
1692 		if (vp->v_type == VCHR)
1693 			vfs_timestamp(&vp->v_rdev->si_ctime);
1694 		else
1695 			vfs_timestamp(&de->de_mtime);
1696 	}
1697 
1698 ret:
1699 	sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock);
1700 	return (error);
1701 }
1702 
1703 #ifdef MAC
1704 static int
1705 devfs_setlabel(struct vop_setlabel_args *ap)
1706 {
1707 	struct vnode *vp;
1708 	struct devfs_dirent *de;
1709 
1710 	vp = ap->a_vp;
1711 	de = vp->v_data;
1712 
1713 	mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1714 	mac_devfs_update(vp->v_mount, de, vp);
1715 
1716 	return (0);
1717 }
1718 #endif
1719 
1720 static int
1721 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1722 {
1723 
1724 	return (vnops.fo_stat(fp, sb, cred, td));
1725 }
1726 
1727 static int
1728 devfs_symlink(struct vop_symlink_args *ap)
1729 {
1730 	int i, error;
1731 	struct devfs_dirent *dd;
1732 	struct devfs_dirent *de, *de_covered, *de_dotdot;
1733 	struct devfs_mount *dmp;
1734 
1735 	error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1736 	if (error)
1737 		return(error);
1738 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1739 	if (devfs_populate_vp(ap->a_dvp) != 0)
1740 		return (ENOENT);
1741 
1742 	dd = ap->a_dvp->v_data;
1743 	de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1744 	de->de_flags = DE_USER;
1745 	de->de_uid = 0;
1746 	de->de_gid = 0;
1747 	de->de_mode = 0755;
1748 	de->de_inode = alloc_unr(devfs_inos);
1749 	de->de_dir = dd;
1750 	de->de_dirent->d_type = DT_LNK;
1751 	i = strlen(ap->a_target) + 1;
1752 	de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1753 	bcopy(ap->a_target, de->de_symlink, i);
1754 #ifdef MAC
1755 	mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1756 #endif
1757 	de_covered = devfs_find(dd, de->de_dirent->d_name,
1758 	    de->de_dirent->d_namlen, 0);
1759 	if (de_covered != NULL) {
1760 		if ((de_covered->de_flags & DE_USER) != 0) {
1761 			devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1762 			sx_xunlock(&dmp->dm_lock);
1763 			return (EEXIST);
1764 		}
1765 		KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1766 		    ("devfs_symlink: entry %p already covered", de_covered));
1767 		de_covered->de_flags |= DE_COVERED;
1768 	}
1769 
1770 	de_dotdot = TAILQ_FIRST(&dd->de_dlist);		/* "." */
1771 	de_dotdot = TAILQ_NEXT(de_dotdot, de_list);	/* ".." */
1772 	TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1773 	devfs_dir_ref_de(dmp, dd);
1774 	devfs_rules_apply(dmp, de);
1775 
1776 	return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1777 }
1778 
1779 static int
1780 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1781 {
1782 
1783 	return (vnops.fo_truncate(fp, length, cred, td));
1784 }
1785 
1786 static int
1787 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
1788     int flags, struct thread *td)
1789 {
1790 	struct cdev *dev;
1791 	int error, ioflag, ref;
1792 	ssize_t resid;
1793 	struct cdevsw *dsw;
1794 	struct file *fpop;
1795 
1796 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1797 		return (EINVAL);
1798 	fpop = td->td_fpop;
1799 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1800 	if (error != 0) {
1801 		error = vnops.fo_write(fp, uio, cred, flags, td);
1802 		return (error);
1803 	}
1804 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1805 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1806 	if (ioflag & O_DIRECT)
1807 		ioflag |= IO_DIRECT;
1808 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1809 
1810 	resid = uio->uio_resid;
1811 
1812 	error = dsw->d_write(dev, uio, ioflag);
1813 	if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1814 		devfs_timestamp(&dev->si_ctime);
1815 		dev->si_mtime = dev->si_ctime;
1816 	}
1817 	td->td_fpop = fpop;
1818 	dev_relthread(dev, ref);
1819 
1820 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1821 	return (error);
1822 }
1823 
1824 static int
1825 devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1826     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1827     struct thread *td)
1828 {
1829 	struct cdev *dev;
1830 	struct cdevsw *dsw;
1831 	struct mount *mp;
1832 	struct vnode *vp;
1833 	struct file *fpop;
1834 	vm_object_t object;
1835 	vm_prot_t maxprot;
1836 	int error, ref;
1837 
1838 	vp = fp->f_vnode;
1839 
1840 	/*
1841 	 * Ensure that file and memory protections are
1842 	 * compatible.
1843 	 */
1844 	mp = vp->v_mount;
1845 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1846 		maxprot = VM_PROT_NONE;
1847 		if ((prot & VM_PROT_EXECUTE) != 0)
1848 			return (EACCES);
1849 	} else
1850 		maxprot = VM_PROT_EXECUTE;
1851 	if ((fp->f_flag & FREAD) != 0)
1852 		maxprot |= VM_PROT_READ;
1853 	else if ((prot & VM_PROT_READ) != 0)
1854 		return (EACCES);
1855 
1856 	/*
1857 	 * If we are sharing potential changes via MAP_SHARED and we
1858 	 * are trying to get write permission although we opened it
1859 	 * without asking for it, bail out.
1860 	 *
1861 	 * Note that most character devices always share mappings.
1862 	 * The one exception is that D_MMAP_ANON devices
1863 	 * (i.e. /dev/zero) permit private writable mappings.
1864 	 *
1865 	 * Rely on vm_mmap_cdev() to fail invalid MAP_PRIVATE requests
1866 	 * as well as updating maxprot to permit writing for
1867 	 * D_MMAP_ANON devices rather than doing that here.
1868 	 */
1869 	if ((flags & MAP_SHARED) != 0) {
1870 		if ((fp->f_flag & FWRITE) != 0)
1871 			maxprot |= VM_PROT_WRITE;
1872 		else if ((prot & VM_PROT_WRITE) != 0)
1873 			return (EACCES);
1874 	}
1875 	maxprot &= cap_maxprot;
1876 
1877 	fpop = td->td_fpop;
1878 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1879 	if (error != 0)
1880 		return (error);
1881 
1882 	error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, dev, dsw, &foff,
1883 	    &object);
1884 	td->td_fpop = fpop;
1885 	dev_relthread(dev, ref);
1886 	if (error != 0)
1887 		return (error);
1888 
1889 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1890 	    foff, FALSE, td);
1891 	if (error != 0)
1892 		vm_object_deallocate(object);
1893 	return (error);
1894 }
1895 
1896 dev_t
1897 dev2udev(struct cdev *x)
1898 {
1899 	if (x == NULL)
1900 		return (NODEV);
1901 	return (cdev2priv(x)->cdp_inode);
1902 }
1903 
1904 static struct fileops devfs_ops_f = {
1905 	.fo_read =	devfs_read_f,
1906 	.fo_write =	devfs_write_f,
1907 	.fo_truncate =	devfs_truncate_f,
1908 	.fo_ioctl =	devfs_ioctl_f,
1909 	.fo_poll =	devfs_poll_f,
1910 	.fo_kqfilter =	devfs_kqfilter_f,
1911 	.fo_stat =	devfs_stat_f,
1912 	.fo_close =	devfs_close_f,
1913 	.fo_chmod =	vn_chmod,
1914 	.fo_chown =	vn_chown,
1915 	.fo_sendfile =	vn_sendfile,
1916 	.fo_seek =	vn_seek,
1917 	.fo_fill_kinfo = vn_fill_kinfo,
1918 	.fo_mmap =	devfs_mmap_f,
1919 	.fo_flags =	DFLAG_PASSABLE | DFLAG_SEEKABLE
1920 };
1921 
1922 /* Vops for non-CHR vnodes in /dev. */
1923 static struct vop_vector devfs_vnodeops = {
1924 	.vop_default =		&default_vnodeops,
1925 
1926 	.vop_access =		devfs_access,
1927 	.vop_getattr =		devfs_getattr,
1928 	.vop_ioctl =		devfs_rioctl,
1929 	.vop_lookup =		devfs_lookup,
1930 	.vop_mknod =		devfs_mknod,
1931 	.vop_pathconf =		devfs_pathconf,
1932 	.vop_read =		devfs_rread,
1933 	.vop_readdir =		devfs_readdir,
1934 	.vop_readlink =		devfs_readlink,
1935 	.vop_reclaim =		devfs_reclaim,
1936 	.vop_remove =		devfs_remove,
1937 	.vop_revoke =		devfs_revoke,
1938 	.vop_setattr =		devfs_setattr,
1939 #ifdef MAC
1940 	.vop_setlabel =		devfs_setlabel,
1941 #endif
1942 	.vop_symlink =		devfs_symlink,
1943 	.vop_vptocnp =		devfs_vptocnp,
1944 };
1945 
1946 /* Vops for VCHR vnodes in /dev. */
1947 static struct vop_vector devfs_specops = {
1948 	.vop_default =		&default_vnodeops,
1949 
1950 	.vop_access =		devfs_access,
1951 	.vop_bmap =		VOP_PANIC,
1952 	.vop_close =		devfs_close,
1953 	.vop_create =		VOP_PANIC,
1954 	.vop_fsync =		vop_stdfsync,
1955 	.vop_getattr =		devfs_getattr,
1956 	.vop_ioctl =		devfs_ioctl,
1957 	.vop_link =		VOP_PANIC,
1958 	.vop_mkdir =		VOP_PANIC,
1959 	.vop_mknod =		VOP_PANIC,
1960 	.vop_open =		devfs_open,
1961 	.vop_pathconf =		devfs_pathconf,
1962 	.vop_poll =		dead_poll,
1963 	.vop_print =		devfs_print,
1964 	.vop_read =		dead_read,
1965 	.vop_readdir =		VOP_PANIC,
1966 	.vop_readlink =		VOP_PANIC,
1967 	.vop_reallocblks =	VOP_PANIC,
1968 	.vop_reclaim =		devfs_reclaim_vchr,
1969 	.vop_remove =		devfs_remove,
1970 	.vop_rename =		VOP_PANIC,
1971 	.vop_revoke =		devfs_revoke,
1972 	.vop_rmdir =		VOP_PANIC,
1973 	.vop_setattr =		devfs_setattr,
1974 #ifdef MAC
1975 	.vop_setlabel =		devfs_setlabel,
1976 #endif
1977 	.vop_strategy =		VOP_PANIC,
1978 	.vop_symlink =		VOP_PANIC,
1979 	.vop_vptocnp =		devfs_vptocnp,
1980 	.vop_write =		dead_write,
1981 };
1982 
1983 /*
1984  * Our calling convention to the device drivers used to be that we passed
1985  * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
1986  * flags instead since that's what open(), close() and ioctl() takes and
1987  * we don't really want vnode.h in device drivers.
1988  * We solved the source compatibility by redefining some vnode flags to
1989  * be the same as the fcntl ones and by sending down the bitwise OR of
1990  * the respective fcntl/vnode flags.  These CTASSERTS make sure nobody
1991  * pulls the rug out under this.
1992  */
1993 CTASSERT(O_NONBLOCK == IO_NDELAY);
1994 CTASSERT(O_FSYNC == IO_SYNC);
1995