xref: /freebsd/sys/fs/devfs/devfs_vnops.c (revision 1f474190)
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/eventhandler.h>
49 #include <sys/fcntl.h>
50 #include <sys/file.h>
51 #include <sys/filedesc.h>
52 #include <sys/filio.h>
53 #include <sys/jail.h>
54 #include <sys/kernel.h>
55 #include <sys/limits.h>
56 #include <sys/lock.h>
57 #include <sys/malloc.h>
58 #include <sys/mman.h>
59 #include <sys/mount.h>
60 #include <sys/namei.h>
61 #include <sys/priv.h>
62 #include <sys/proc.h>
63 #include <sys/stat.h>
64 #include <sys/sx.h>
65 #include <sys/sysctl.h>
66 #include <sys/time.h>
67 #include <sys/ttycom.h>
68 #include <sys/unistd.h>
69 #include <sys/vnode.h>
70 
71 static struct vop_vector devfs_vnodeops;
72 static struct vop_vector devfs_specops;
73 static struct fileops devfs_ops_f;
74 
75 #include <fs/devfs/devfs.h>
76 #include <fs/devfs/devfs_int.h>
77 
78 #include <security/mac/mac_framework.h>
79 
80 #include <vm/vm.h>
81 #include <vm/vm_extern.h>
82 #include <vm/vm_object.h>
83 
84 static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data");
85 
86 struct mtx	devfs_de_interlock;
87 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF);
88 struct sx	clone_drain_lock;
89 SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock");
90 struct mtx	cdevpriv_mtx;
91 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
92 
93 SYSCTL_DECL(_vfs_devfs);
94 
95 static int devfs_dotimes;
96 SYSCTL_INT(_vfs_devfs, OID_AUTO, dotimes, CTLFLAG_RW,
97     &devfs_dotimes, 0, "Update timestamps on DEVFS with default precision");
98 
99 /*
100  * Update devfs node timestamp.  Note that updates are unlocked and
101  * stat(2) could see partially updated times.
102  */
103 static void
104 devfs_timestamp(struct timespec *tsp)
105 {
106 	time_t ts;
107 
108 	if (devfs_dotimes) {
109 		vfs_timestamp(tsp);
110 	} else {
111 		ts = time_second;
112 		if (tsp->tv_sec != ts) {
113 			tsp->tv_sec = ts;
114 			tsp->tv_nsec = 0;
115 		}
116 	}
117 }
118 
119 static int
120 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp,
121     int *ref)
122 {
123 
124 	*dswp = devvn_refthread(fp->f_vnode, devp, ref);
125 	if (*devp != fp->f_data) {
126 		if (*dswp != NULL)
127 			dev_relthread(*devp, *ref);
128 		return (ENXIO);
129 	}
130 	KASSERT((*devp)->si_refcount > 0,
131 	    ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
132 	if (*dswp == NULL)
133 		return (ENXIO);
134 	curthread->td_fpop = fp;
135 	return (0);
136 }
137 
138 int
139 devfs_get_cdevpriv(void **datap)
140 {
141 	struct file *fp;
142 	struct cdev_privdata *p;
143 	int error;
144 
145 	fp = curthread->td_fpop;
146 	if (fp == NULL)
147 		return (EBADF);
148 	p = fp->f_cdevpriv;
149 	if (p != NULL) {
150 		error = 0;
151 		*datap = p->cdpd_data;
152 	} else
153 		error = ENOENT;
154 	return (error);
155 }
156 
157 int
158 devfs_set_cdevpriv(void *priv, d_priv_dtor_t *priv_dtr)
159 {
160 	struct file *fp;
161 	struct cdev_priv *cdp;
162 	struct cdev_privdata *p;
163 	int error;
164 
165 	fp = curthread->td_fpop;
166 	if (fp == NULL)
167 		return (ENOENT);
168 	cdp = cdev2priv((struct cdev *)fp->f_data);
169 	p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
170 	p->cdpd_data = priv;
171 	p->cdpd_dtr = priv_dtr;
172 	p->cdpd_fp = fp;
173 	mtx_lock(&cdevpriv_mtx);
174 	if (fp->f_cdevpriv == NULL) {
175 		LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
176 		fp->f_cdevpriv = p;
177 		mtx_unlock(&cdevpriv_mtx);
178 		error = 0;
179 	} else {
180 		mtx_unlock(&cdevpriv_mtx);
181 		free(p, M_CDEVPDATA);
182 		error = EBUSY;
183 	}
184 	return (error);
185 }
186 
187 void
188 devfs_destroy_cdevpriv(struct cdev_privdata *p)
189 {
190 
191 	mtx_assert(&cdevpriv_mtx, MA_OWNED);
192 	KASSERT(p->cdpd_fp->f_cdevpriv == p,
193 	    ("devfs_destoy_cdevpriv %p != %p", p->cdpd_fp->f_cdevpriv, p));
194 	p->cdpd_fp->f_cdevpriv = NULL;
195 	LIST_REMOVE(p, cdpd_list);
196 	mtx_unlock(&cdevpriv_mtx);
197 	(p->cdpd_dtr)(p->cdpd_data);
198 	free(p, M_CDEVPDATA);
199 }
200 
201 static void
202 devfs_fpdrop(struct file *fp)
203 {
204 	struct cdev_privdata *p;
205 
206 	mtx_lock(&cdevpriv_mtx);
207 	if ((p = fp->f_cdevpriv) == NULL) {
208 		mtx_unlock(&cdevpriv_mtx);
209 		return;
210 	}
211 	devfs_destroy_cdevpriv(p);
212 }
213 
214 void
215 devfs_clear_cdevpriv(void)
216 {
217 	struct file *fp;
218 
219 	fp = curthread->td_fpop;
220 	if (fp == NULL)
221 		return;
222 	devfs_fpdrop(fp);
223 }
224 
225 static void
226 devfs_usecount_add(struct vnode *vp)
227 {
228 	struct devfs_dirent *de;
229 	struct cdev *dev;
230 
231 	mtx_lock(&devfs_de_interlock);
232 	VI_LOCK(vp);
233 	VNPASS(vp->v_type == VCHR || vp->v_type == VBAD, vp);
234 	if (VN_IS_DOOMED(vp)) {
235 		goto out_unlock;
236 	}
237 
238 	de = vp->v_data;
239 	dev = vp->v_rdev;
240 	MPASS(de != NULL);
241 	MPASS(dev != NULL);
242 	dev->si_usecount++;
243 	de->de_usecount++;
244 out_unlock:
245 	VI_UNLOCK(vp);
246 	mtx_unlock(&devfs_de_interlock);
247 }
248 
249 static void
250 devfs_usecount_subl(struct vnode *vp)
251 {
252 	struct devfs_dirent *de;
253 	struct cdev *dev;
254 
255 	mtx_assert(&devfs_de_interlock, MA_OWNED);
256 	ASSERT_VI_LOCKED(vp, __func__);
257 	VNPASS(vp->v_type == VCHR || vp->v_type == VBAD, vp);
258 
259 	de = vp->v_data;
260 	dev = vp->v_rdev;
261 	if (de == NULL)
262 		return;
263 	if (dev == NULL) {
264 		MPASS(de->de_usecount == 0);
265 		return;
266 	}
267 	if (dev->si_usecount < de->de_usecount)
268 		panic("%s: si_usecount underflow for dev %p "
269 		    "(has %ld, dirent has %d)\n",
270 		    __func__, dev, dev->si_usecount, de->de_usecount);
271 	if (VN_IS_DOOMED(vp)) {
272 		dev->si_usecount -= de->de_usecount;
273 		de->de_usecount = 0;
274 	} else {
275 		if (de->de_usecount == 0)
276 			panic("%s: de_usecount underflow for dev %p\n",
277 			    __func__, dev);
278 		dev->si_usecount--;
279 		de->de_usecount--;
280 	}
281 }
282 
283 static void
284 devfs_usecount_sub(struct vnode *vp)
285 {
286 
287 	mtx_lock(&devfs_de_interlock);
288 	VI_LOCK(vp);
289 	devfs_usecount_subl(vp);
290 	VI_UNLOCK(vp);
291 	mtx_unlock(&devfs_de_interlock);
292 }
293 
294 static int
295 devfs_usecountl(struct vnode *vp)
296 {
297 
298 	VNPASS(vp->v_type == VCHR, vp);
299 	mtx_assert(&devfs_de_interlock, MA_OWNED);
300 	ASSERT_VI_LOCKED(vp, __func__);
301 	return (vp->v_rdev->si_usecount);
302 }
303 
304 int
305 devfs_usecount(struct vnode *vp)
306 {
307 	int count;
308 
309 	VNPASS(vp->v_type == VCHR, vp);
310 	mtx_lock(&devfs_de_interlock);
311 	VI_LOCK(vp);
312 	count = devfs_usecountl(vp);
313 	VI_UNLOCK(vp);
314 	mtx_unlock(&devfs_de_interlock);
315 	return (count);
316 }
317 
318 void
319 devfs_ctty_ref(struct vnode *vp)
320 {
321 
322 	vrefact(vp);
323 	devfs_usecount_add(vp);
324 }
325 
326 void
327 devfs_ctty_unref(struct vnode *vp)
328 {
329 
330 	devfs_usecount_sub(vp);
331 	vrele(vp);
332 }
333 
334 /*
335  * On success devfs_populate_vp() returns with dmp->dm_lock held.
336  */
337 static int
338 devfs_populate_vp(struct vnode *vp)
339 {
340 	struct devfs_dirent *de;
341 	struct devfs_mount *dmp;
342 	int locked;
343 
344 	ASSERT_VOP_LOCKED(vp, "devfs_populate_vp");
345 
346 	dmp = VFSTODEVFS(vp->v_mount);
347 	if (!devfs_populate_needed(dmp)) {
348 		sx_xlock(&dmp->dm_lock);
349 		goto out_nopopulate;
350 	}
351 
352 	locked = VOP_ISLOCKED(vp);
353 
354 	sx_xlock(&dmp->dm_lock);
355 	DEVFS_DMP_HOLD(dmp);
356 
357 	/* Can't call devfs_populate() with the vnode lock held. */
358 	VOP_UNLOCK(vp);
359 	devfs_populate(dmp);
360 
361 	sx_xunlock(&dmp->dm_lock);
362 	vn_lock(vp, locked | LK_RETRY);
363 	sx_xlock(&dmp->dm_lock);
364 	if (DEVFS_DMP_DROP(dmp)) {
365 		sx_xunlock(&dmp->dm_lock);
366 		devfs_unmount_final(dmp);
367 		return (ERESTART);
368 	}
369 out_nopopulate:
370 	if (VN_IS_DOOMED(vp)) {
371 		sx_xunlock(&dmp->dm_lock);
372 		return (ERESTART);
373 	}
374 	de = vp->v_data;
375 	KASSERT(de != NULL,
376 	    ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed"));
377 	if ((de->de_flags & DE_DOOMED) != 0) {
378 		sx_xunlock(&dmp->dm_lock);
379 		return (ERESTART);
380 	}
381 
382 	return (0);
383 }
384 
385 static int
386 devfs_vptocnp(struct vop_vptocnp_args *ap)
387 {
388 	struct vnode *vp = ap->a_vp;
389 	struct vnode **dvp = ap->a_vpp;
390 	struct devfs_mount *dmp;
391 	char *buf = ap->a_buf;
392 	size_t *buflen = ap->a_buflen;
393 	struct devfs_dirent *dd, *de;
394 	int i, error;
395 
396 	dmp = VFSTODEVFS(vp->v_mount);
397 
398 	error = devfs_populate_vp(vp);
399 	if (error != 0)
400 		return (error);
401 
402 	if (vp->v_type != VCHR && vp->v_type != VDIR) {
403 		error = ENOENT;
404 		goto finished;
405 	}
406 
407 	dd = vp->v_data;
408 	if (vp->v_type == VDIR && dd == dmp->dm_rootdir) {
409 		*dvp = vp;
410 		vref(*dvp);
411 		goto finished;
412 	}
413 
414 	i = *buflen;
415 	i -= dd->de_dirent->d_namlen;
416 	if (i < 0) {
417 		error = ENOMEM;
418 		goto finished;
419 	}
420 	bcopy(dd->de_dirent->d_name, buf + i, dd->de_dirent->d_namlen);
421 	*buflen = i;
422 	de = devfs_parent_dirent(dd);
423 	if (de == NULL) {
424 		error = ENOENT;
425 		goto finished;
426 	}
427 	mtx_lock(&devfs_de_interlock);
428 	*dvp = de->de_vnode;
429 	if (*dvp != NULL) {
430 		VI_LOCK(*dvp);
431 		mtx_unlock(&devfs_de_interlock);
432 		vholdl(*dvp);
433 		VI_UNLOCK(*dvp);
434 		vref(*dvp);
435 		vdrop(*dvp);
436 	} else {
437 		mtx_unlock(&devfs_de_interlock);
438 		error = ENOENT;
439 	}
440 finished:
441 	sx_xunlock(&dmp->dm_lock);
442 	return (error);
443 }
444 
445 /*
446  * Construct the fully qualified path name relative to the mountpoint.
447  * If a NULL cnp is provided, no '/' is appended to the resulting path.
448  */
449 char *
450 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd,
451     struct componentname *cnp)
452 {
453 	int i;
454 	struct devfs_dirent *de;
455 
456 	sx_assert(&dmp->dm_lock, SA_LOCKED);
457 
458 	i = SPECNAMELEN;
459 	buf[i] = '\0';
460 	if (cnp != NULL)
461 		i -= cnp->cn_namelen;
462 	if (i < 0)
463 		 return (NULL);
464 	if (cnp != NULL)
465 		bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
466 	de = dd;
467 	while (de != dmp->dm_rootdir) {
468 		if (cnp != NULL || i < SPECNAMELEN) {
469 			i--;
470 			if (i < 0)
471 				 return (NULL);
472 			buf[i] = '/';
473 		}
474 		i -= de->de_dirent->d_namlen;
475 		if (i < 0)
476 			 return (NULL);
477 		bcopy(de->de_dirent->d_name, buf + i,
478 		    de->de_dirent->d_namlen);
479 		de = devfs_parent_dirent(de);
480 		if (de == NULL)
481 			return (NULL);
482 	}
483 	return (buf + i);
484 }
485 
486 static int
487 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
488 	struct devfs_dirent *de)
489 {
490 	int not_found;
491 
492 	not_found = 0;
493 	if (de->de_flags & DE_DOOMED)
494 		not_found = 1;
495 	if (DEVFS_DE_DROP(de)) {
496 		KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
497 		devfs_dirent_free(de);
498 	}
499 	if (DEVFS_DMP_DROP(dmp)) {
500 		KASSERT(not_found == 1,
501 			("DEVFS mount struct freed before dirent"));
502 		not_found = 2;
503 		sx_xunlock(&dmp->dm_lock);
504 		devfs_unmount_final(dmp);
505 	}
506 	if (not_found == 1 || (drop_dm_lock && not_found != 2))
507 		sx_unlock(&dmp->dm_lock);
508 	return (not_found);
509 }
510 
511 static void
512 devfs_insmntque_dtr(struct vnode *vp, void *arg)
513 {
514 	struct devfs_dirent *de;
515 
516 	de = (struct devfs_dirent *)arg;
517 	mtx_lock(&devfs_de_interlock);
518 	vp->v_data = NULL;
519 	de->de_vnode = NULL;
520 	mtx_unlock(&devfs_de_interlock);
521 	vgone(vp);
522 	vput(vp);
523 }
524 
525 /*
526  * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
527  * it on return.
528  */
529 int
530 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode,
531     struct vnode **vpp)
532 {
533 	int error;
534 	struct vnode *vp;
535 	struct cdev *dev;
536 	struct devfs_mount *dmp;
537 	struct cdevsw *dsw;
538 	enum vgetstate vs;
539 
540 	dmp = VFSTODEVFS(mp);
541 	if (de->de_flags & DE_DOOMED) {
542 		sx_xunlock(&dmp->dm_lock);
543 		return (ENOENT);
544 	}
545 loop:
546 	DEVFS_DE_HOLD(de);
547 	DEVFS_DMP_HOLD(dmp);
548 	mtx_lock(&devfs_de_interlock);
549 	vp = de->de_vnode;
550 	if (vp != NULL) {
551 		vs = vget_prep(vp);
552 		mtx_unlock(&devfs_de_interlock);
553 		sx_xunlock(&dmp->dm_lock);
554 		vget_finish(vp, lockmode | LK_RETRY, vs);
555 		sx_xlock(&dmp->dm_lock);
556 		if (devfs_allocv_drop_refs(0, dmp, de)) {
557 			vput(vp);
558 			return (ENOENT);
559 		}
560 		else if (VN_IS_DOOMED(vp)) {
561 			mtx_lock(&devfs_de_interlock);
562 			if (de->de_vnode == vp) {
563 				de->de_vnode = NULL;
564 				vp->v_data = NULL;
565 			}
566 			mtx_unlock(&devfs_de_interlock);
567 			vput(vp);
568 			goto loop;
569 		}
570 		sx_xunlock(&dmp->dm_lock);
571 		*vpp = vp;
572 		return (0);
573 	}
574 	mtx_unlock(&devfs_de_interlock);
575 	if (de->de_dirent->d_type == DT_CHR) {
576 		if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
577 			devfs_allocv_drop_refs(1, dmp, de);
578 			return (ENOENT);
579 		}
580 		dev = &de->de_cdp->cdp_c;
581 	} else {
582 		dev = NULL;
583 	}
584 	error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
585 	if (error != 0) {
586 		devfs_allocv_drop_refs(1, dmp, de);
587 		printf("devfs_allocv: failed to allocate new vnode\n");
588 		return (error);
589 	}
590 
591 	if (de->de_dirent->d_type == DT_CHR) {
592 		vp->v_type = VCHR;
593 		VI_LOCK(vp);
594 		dev_lock();
595 		dev_refl(dev);
596 		/* XXX: v_rdev should be protect by vnode lock */
597 		vp->v_rdev = dev;
598 		VNPASS(vp->v_usecount == 1, vp);
599 		/* Special casing of ttys for deadfs.  Probably redundant. */
600 		dsw = dev->si_devsw;
601 		if (dsw != NULL && (dsw->d_flags & D_TTY) != 0)
602 			vp->v_vflag |= VV_ISTTY;
603 		dev_unlock();
604 		VI_UNLOCK(vp);
605 		if ((dev->si_flags & SI_ETERNAL) != 0)
606 			vp->v_vflag |= VV_ETERNALDEV;
607 		vp->v_op = &devfs_specops;
608 	} else if (de->de_dirent->d_type == DT_DIR) {
609 		vp->v_type = VDIR;
610 	} else if (de->de_dirent->d_type == DT_LNK) {
611 		vp->v_type = VLNK;
612 	} else {
613 		vp->v_type = VBAD;
614 	}
615 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
616 	VN_LOCK_ASHARE(vp);
617 	mtx_lock(&devfs_de_interlock);
618 	vp->v_data = de;
619 	de->de_vnode = vp;
620 	mtx_unlock(&devfs_de_interlock);
621 	error = insmntque1(vp, mp, devfs_insmntque_dtr, de);
622 	if (error != 0) {
623 		(void) devfs_allocv_drop_refs(1, dmp, de);
624 		return (error);
625 	}
626 	if (devfs_allocv_drop_refs(0, dmp, de)) {
627 		vput(vp);
628 		return (ENOENT);
629 	}
630 #ifdef MAC
631 	mac_devfs_vnode_associate(mp, de, vp);
632 #endif
633 	sx_xunlock(&dmp->dm_lock);
634 	*vpp = vp;
635 	return (0);
636 }
637 
638 static int
639 devfs_access(struct vop_access_args *ap)
640 {
641 	struct vnode *vp = ap->a_vp;
642 	struct devfs_dirent *de;
643 	struct proc *p;
644 	int error;
645 
646 	de = vp->v_data;
647 	if (vp->v_type == VDIR)
648 		de = de->de_dir;
649 
650 	error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
651 	    ap->a_accmode, ap->a_cred);
652 	if (error == 0)
653 		return (0);
654 	if (error != EACCES)
655 		return (error);
656 	p = ap->a_td->td_proc;
657 	/* We do, however, allow access to the controlling terminal */
658 	PROC_LOCK(p);
659 	if (!(p->p_flag & P_CONTROLT)) {
660 		PROC_UNLOCK(p);
661 		return (error);
662 	}
663 	if (p->p_session->s_ttydp == de->de_cdp)
664 		error = 0;
665 	PROC_UNLOCK(p);
666 	return (error);
667 }
668 
669 _Static_assert(((FMASK | FCNTLFLAGS) & (FLASTCLOSE | FREVOKE)) == 0,
670     "devfs-only flag reuse failed");
671 
672 static int
673 devfs_close(struct vop_close_args *ap)
674 {
675 	struct vnode *vp = ap->a_vp, *oldvp;
676 	struct thread *td = ap->a_td;
677 	struct proc *p;
678 	struct cdev *dev = vp->v_rdev;
679 	struct cdevsw *dsw;
680 	struct devfs_dirent *de = vp->v_data;
681 	int dflags, error, ref, vp_locked;
682 
683 	/*
684 	 * XXX: Don't call d_close() if we were called because of
685 	 * XXX: insmntque1() failure.
686 	 */
687 	if (vp->v_data == NULL)
688 		return (0);
689 
690 	/*
691 	 * Hack: a tty device that is a controlling terminal
692 	 * has a reference from the session structure.
693 	 * We cannot easily tell that a character device is
694 	 * a controlling terminal, unless it is the closing
695 	 * process' controlling terminal.  In that case,
696 	 * if the reference count is 2 (this last descriptor
697 	 * plus the session), release the reference from the session.
698 	 */
699 	if (de->de_usecount == 2 && td != NULL) {
700 		p = td->td_proc;
701 		PROC_LOCK(p);
702 		if (vp == p->p_session->s_ttyvp) {
703 			PROC_UNLOCK(p);
704 			oldvp = NULL;
705 			sx_xlock(&proctree_lock);
706 			if (vp == p->p_session->s_ttyvp) {
707 				SESS_LOCK(p->p_session);
708 				mtx_lock(&devfs_de_interlock);
709 				VI_LOCK(vp);
710 				if (devfs_usecountl(vp) == 2 && !VN_IS_DOOMED(vp)) {
711 					p->p_session->s_ttyvp = NULL;
712 					p->p_session->s_ttydp = NULL;
713 					oldvp = vp;
714 				}
715 				VI_UNLOCK(vp);
716 				mtx_unlock(&devfs_de_interlock);
717 				SESS_UNLOCK(p->p_session);
718 			}
719 			sx_xunlock(&proctree_lock);
720 			if (oldvp != NULL)
721 				devfs_ctty_unref(oldvp);
722 		} else
723 			PROC_UNLOCK(p);
724 	}
725 	/*
726 	 * We do not want to really close the device if it
727 	 * is still in use unless we are trying to close it
728 	 * forcibly. Since every use (buffer, vnode, swap, cmap)
729 	 * holds a reference to the vnode, and because we mark
730 	 * any other vnodes that alias this device, when the
731 	 * sum of the reference counts on all the aliased
732 	 * vnodes descends to one, we are on last close.
733 	 */
734 	dsw = dev_refthread(dev, &ref);
735 	if (dsw == NULL)
736 		return (ENXIO);
737 	dflags = 0;
738 	mtx_lock(&devfs_de_interlock);
739 	VI_LOCK(vp);
740 	if (devfs_usecountl(vp) == 1)
741 		dflags |= FLASTCLOSE;
742 	devfs_usecount_subl(vp);
743 	mtx_unlock(&devfs_de_interlock);
744 	if (VN_IS_DOOMED(vp)) {
745 		/* Forced close. */
746 		dflags |= FREVOKE | FNONBLOCK;
747 	} else if (dsw->d_flags & D_TRACKCLOSE) {
748 		/* Keep device updated on status. */
749 	} else if ((dflags & FLASTCLOSE) == 0) {
750 		VI_UNLOCK(vp);
751 		dev_relthread(dev, ref);
752 		return (0);
753 	}
754 	vholdnz(vp);
755 	VI_UNLOCK(vp);
756 	vp_locked = VOP_ISLOCKED(vp);
757 	VOP_UNLOCK(vp);
758 	KASSERT(dev->si_refcount > 0,
759 	    ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
760 	error = dsw->d_close(dev, ap->a_fflag | dflags, S_IFCHR, td);
761 	dev_relthread(dev, ref);
762 	vn_lock(vp, vp_locked | LK_RETRY);
763 	vdrop(vp);
764 	return (error);
765 }
766 
767 static int
768 devfs_close_f(struct file *fp, struct thread *td)
769 {
770 	int error;
771 	struct file *fpop;
772 
773 	/*
774 	 * NB: td may be NULL if this descriptor is closed due to
775 	 * garbage collection from a closed UNIX domain socket.
776 	 */
777 	fpop = curthread->td_fpop;
778 	curthread->td_fpop = fp;
779 	error = vnops.fo_close(fp, td);
780 	curthread->td_fpop = fpop;
781 
782 	/*
783 	 * The f_cdevpriv cannot be assigned non-NULL value while we
784 	 * are destroying the file.
785 	 */
786 	if (fp->f_cdevpriv != NULL)
787 		devfs_fpdrop(fp);
788 	return (error);
789 }
790 
791 static int
792 devfs_getattr(struct vop_getattr_args *ap)
793 {
794 	struct vnode *vp = ap->a_vp;
795 	struct vattr *vap = ap->a_vap;
796 	struct devfs_dirent *de;
797 	struct devfs_mount *dmp;
798 	struct cdev *dev;
799 	struct timeval boottime;
800 	int error;
801 
802 	error = devfs_populate_vp(vp);
803 	if (error != 0)
804 		return (error);
805 
806 	dmp = VFSTODEVFS(vp->v_mount);
807 	sx_xunlock(&dmp->dm_lock);
808 
809 	de = vp->v_data;
810 	KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
811 	if (vp->v_type == VDIR) {
812 		de = de->de_dir;
813 		KASSERT(de != NULL,
814 		    ("Null dir dirent in devfs_getattr vp=%p", vp));
815 	}
816 	vap->va_uid = de->de_uid;
817 	vap->va_gid = de->de_gid;
818 	vap->va_mode = de->de_mode;
819 	if (vp->v_type == VLNK)
820 		vap->va_size = strlen(de->de_symlink);
821 	else if (vp->v_type == VDIR)
822 		vap->va_size = vap->va_bytes = DEV_BSIZE;
823 	else
824 		vap->va_size = 0;
825 	if (vp->v_type != VDIR)
826 		vap->va_bytes = 0;
827 	vap->va_blocksize = DEV_BSIZE;
828 	vap->va_type = vp->v_type;
829 
830 	getboottime(&boottime);
831 #define fix(aa)							\
832 	do {							\
833 		if ((aa).tv_sec <= 3600) {			\
834 			(aa).tv_sec = boottime.tv_sec;		\
835 			(aa).tv_nsec = boottime.tv_usec * 1000; \
836 		}						\
837 	} while (0)
838 
839 	if (vp->v_type != VCHR)  {
840 		fix(de->de_atime);
841 		vap->va_atime = de->de_atime;
842 		fix(de->de_mtime);
843 		vap->va_mtime = de->de_mtime;
844 		fix(de->de_ctime);
845 		vap->va_ctime = de->de_ctime;
846 	} else {
847 		dev = vp->v_rdev;
848 		fix(dev->si_atime);
849 		vap->va_atime = dev->si_atime;
850 		fix(dev->si_mtime);
851 		vap->va_mtime = dev->si_mtime;
852 		fix(dev->si_ctime);
853 		vap->va_ctime = dev->si_ctime;
854 
855 		vap->va_rdev = cdev2priv(dev)->cdp_inode;
856 	}
857 	vap->va_gen = 0;
858 	vap->va_flags = 0;
859 	vap->va_filerev = 0;
860 	vap->va_nlink = de->de_links;
861 	vap->va_fileid = de->de_inode;
862 
863 	return (error);
864 }
865 
866 /* ARGSUSED */
867 static int
868 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
869 {
870 	struct file *fpop;
871 	int error;
872 
873 	fpop = td->td_fpop;
874 	td->td_fpop = fp;
875 	error = vnops.fo_ioctl(fp, com, data, cred, td);
876 	td->td_fpop = fpop;
877 	return (error);
878 }
879 
880 void *
881 fiodgname_buf_get_ptr(void *fgnp, u_long com)
882 {
883 	union {
884 		struct fiodgname_arg	fgn;
885 #ifdef COMPAT_FREEBSD32
886 		struct fiodgname_arg32	fgn32;
887 #endif
888 	} *fgnup;
889 
890 	fgnup = fgnp;
891 	switch (com) {
892 	case FIODGNAME:
893 		return (fgnup->fgn.buf);
894 #ifdef COMPAT_FREEBSD32
895 	case FIODGNAME_32:
896 		return ((void *)(uintptr_t)fgnup->fgn32.buf);
897 #endif
898 	default:
899 		panic("Unhandled ioctl command %ld", com);
900 	}
901 }
902 
903 static int
904 devfs_ioctl(struct vop_ioctl_args *ap)
905 {
906 	struct fiodgname_arg *fgn;
907 	struct vnode *vpold, *vp;
908 	struct cdevsw *dsw;
909 	struct thread *td;
910 	struct session *sess;
911 	struct cdev *dev;
912 	int error, ref, i;
913 	const char *p;
914 	u_long com;
915 
916 	vp = ap->a_vp;
917 	com = ap->a_command;
918 	td = ap->a_td;
919 
920 	dsw = devvn_refthread(vp, &dev, &ref);
921 	if (dsw == NULL)
922 		return (ENXIO);
923 	KASSERT(dev->si_refcount > 0,
924 	    ("devfs: un-referenced struct cdev *(%s)", devtoname(dev)));
925 
926 	switch (com) {
927 	case FIODTYPE:
928 		*(int *)ap->a_data = dsw->d_flags & D_TYPEMASK;
929 		error = 0;
930 		break;
931 	case FIODGNAME:
932 #ifdef	COMPAT_FREEBSD32
933 	case FIODGNAME_32:
934 #endif
935 		fgn = ap->a_data;
936 		p = devtoname(dev);
937 		i = strlen(p) + 1;
938 		if (i > fgn->len)
939 			error = EINVAL;
940 		else
941 			error = copyout(p, fiodgname_buf_get_ptr(fgn, com), i);
942 		break;
943 	default:
944 		error = dsw->d_ioctl(dev, com, ap->a_data, ap->a_fflag, td);
945 	}
946 
947 	dev_relthread(dev, ref);
948 	if (error == ENOIOCTL)
949 		error = ENOTTY;
950 
951 	if (error == 0 && com == TIOCSCTTY) {
952 		/*
953 		 * Do nothing if reassigning same control tty, or if the
954 		 * control tty has already disappeared.  If it disappeared,
955 		 * it's because we were racing with TIOCNOTTY.  TIOCNOTTY
956 		 * already took care of releasing the old vnode and we have
957 		 * nothing left to do.
958 		 */
959 		sx_slock(&proctree_lock);
960 		sess = td->td_proc->p_session;
961 		if (sess->s_ttyvp == vp || sess->s_ttyp == NULL) {
962 			sx_sunlock(&proctree_lock);
963 			return (0);
964 		}
965 
966 		devfs_ctty_ref(vp);
967 		SESS_LOCK(sess);
968 		vpold = sess->s_ttyvp;
969 		sess->s_ttyvp = vp;
970 		sess->s_ttydp = cdev2priv(dev);
971 		SESS_UNLOCK(sess);
972 
973 		sx_sunlock(&proctree_lock);
974 
975 		/* Get rid of reference to old control tty */
976 		if (vpold)
977 			vrele(vpold);
978 	}
979 	return (error);
980 }
981 
982 /* ARGSUSED */
983 static int
984 devfs_kqfilter_f(struct file *fp, struct knote *kn)
985 {
986 	struct cdev *dev;
987 	struct cdevsw *dsw;
988 	int error, ref;
989 	struct file *fpop;
990 	struct thread *td;
991 
992 	td = curthread;
993 	fpop = td->td_fpop;
994 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
995 	if (error)
996 		return (error);
997 	error = dsw->d_kqfilter(dev, kn);
998 	td->td_fpop = fpop;
999 	dev_relthread(dev, ref);
1000 	return (error);
1001 }
1002 
1003 static inline int
1004 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
1005 {
1006 	struct cdev_priv *cdp;
1007 	struct ucred *dcr;
1008 	struct proc *p;
1009 	int error;
1010 
1011 	cdp = de->de_cdp;
1012 	if (cdp == NULL)
1013 		return (0);
1014 	dcr = cdp->cdp_c.si_cred;
1015 	if (dcr == NULL)
1016 		return (0);
1017 
1018 	error = prison_check(td->td_ucred, dcr);
1019 	if (error == 0)
1020 		return (0);
1021 	/* We do, however, allow access to the controlling terminal */
1022 	p = td->td_proc;
1023 	PROC_LOCK(p);
1024 	if (!(p->p_flag & P_CONTROLT)) {
1025 		PROC_UNLOCK(p);
1026 		return (error);
1027 	}
1028 	if (p->p_session->s_ttydp == cdp)
1029 		error = 0;
1030 	PROC_UNLOCK(p);
1031 	return (error);
1032 }
1033 
1034 static int
1035 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
1036 {
1037 	struct componentname *cnp;
1038 	struct vnode *dvp, **vpp;
1039 	struct thread *td;
1040 	struct devfs_dirent *de, *dd;
1041 	struct devfs_dirent **dde;
1042 	struct devfs_mount *dmp;
1043 	struct mount *mp;
1044 	struct cdev *cdev;
1045 	int error, flags, nameiop, dvplocked;
1046 	char specname[SPECNAMELEN + 1], *pname;
1047 
1048 	cnp = ap->a_cnp;
1049 	vpp = ap->a_vpp;
1050 	dvp = ap->a_dvp;
1051 	pname = cnp->cn_nameptr;
1052 	td = cnp->cn_thread;
1053 	flags = cnp->cn_flags;
1054 	nameiop = cnp->cn_nameiop;
1055 	mp = dvp->v_mount;
1056 	dmp = VFSTODEVFS(mp);
1057 	dd = dvp->v_data;
1058 	*vpp = NULLVP;
1059 
1060 	if ((flags & ISLASTCN) && nameiop == RENAME)
1061 		return (EOPNOTSUPP);
1062 
1063 	if (dvp->v_type != VDIR)
1064 		return (ENOTDIR);
1065 
1066 	if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
1067 		return (EIO);
1068 
1069 	error = vn_dir_check_exec(dvp, cnp);
1070 	if (error != 0)
1071 		return (error);
1072 
1073 	if (cnp->cn_namelen == 1 && *pname == '.') {
1074 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
1075 			return (EINVAL);
1076 		*vpp = dvp;
1077 		VREF(dvp);
1078 		return (0);
1079 	}
1080 
1081 	if (flags & ISDOTDOT) {
1082 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
1083 			return (EINVAL);
1084 		de = devfs_parent_dirent(dd);
1085 		if (de == NULL)
1086 			return (ENOENT);
1087 		dvplocked = VOP_ISLOCKED(dvp);
1088 		VOP_UNLOCK(dvp);
1089 		error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK,
1090 		    vpp);
1091 		*dm_unlock = 0;
1092 		vn_lock(dvp, dvplocked | LK_RETRY);
1093 		return (error);
1094 	}
1095 
1096 	dd = dvp->v_data;
1097 	de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
1098 	while (de == NULL) {	/* While(...) so we can use break */
1099 
1100 		if (nameiop == DELETE)
1101 			return (ENOENT);
1102 
1103 		/*
1104 		 * OK, we didn't have an entry for the name we were asked for
1105 		 * so we try to see if anybody can create it on demand.
1106 		 */
1107 		pname = devfs_fqpn(specname, dmp, dd, cnp);
1108 		if (pname == NULL)
1109 			break;
1110 
1111 		cdev = NULL;
1112 		DEVFS_DMP_HOLD(dmp);
1113 		sx_xunlock(&dmp->dm_lock);
1114 		sx_slock(&clone_drain_lock);
1115 		EVENTHANDLER_INVOKE(dev_clone,
1116 		    td->td_ucred, pname, strlen(pname), &cdev);
1117 		sx_sunlock(&clone_drain_lock);
1118 
1119 		if (cdev == NULL)
1120 			sx_xlock(&dmp->dm_lock);
1121 		else if (devfs_populate_vp(dvp) != 0) {
1122 			*dm_unlock = 0;
1123 			sx_xlock(&dmp->dm_lock);
1124 			if (DEVFS_DMP_DROP(dmp)) {
1125 				sx_xunlock(&dmp->dm_lock);
1126 				devfs_unmount_final(dmp);
1127 			} else
1128 				sx_xunlock(&dmp->dm_lock);
1129 			dev_rel(cdev);
1130 			return (ENOENT);
1131 		}
1132 		if (DEVFS_DMP_DROP(dmp)) {
1133 			*dm_unlock = 0;
1134 			sx_xunlock(&dmp->dm_lock);
1135 			devfs_unmount_final(dmp);
1136 			if (cdev != NULL)
1137 				dev_rel(cdev);
1138 			return (ENOENT);
1139 		}
1140 
1141 		if (cdev == NULL)
1142 			break;
1143 
1144 		dev_lock();
1145 		dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
1146 		if (dde != NULL && *dde != NULL)
1147 			de = *dde;
1148 		dev_unlock();
1149 		dev_rel(cdev);
1150 		break;
1151 	}
1152 
1153 	if (de == NULL || de->de_flags & DE_WHITEOUT) {
1154 		if ((nameiop == CREATE || nameiop == RENAME) &&
1155 		    (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
1156 			cnp->cn_flags |= SAVENAME;
1157 			return (EJUSTRETURN);
1158 		}
1159 		return (ENOENT);
1160 	}
1161 
1162 	if (devfs_prison_check(de, td))
1163 		return (ENOENT);
1164 
1165 	if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
1166 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
1167 		if (error)
1168 			return (error);
1169 		if (*vpp == dvp) {
1170 			VREF(dvp);
1171 			*vpp = dvp;
1172 			return (0);
1173 		}
1174 	}
1175 	error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, vpp);
1176 	*dm_unlock = 0;
1177 	return (error);
1178 }
1179 
1180 static int
1181 devfs_lookup(struct vop_lookup_args *ap)
1182 {
1183 	int j;
1184 	struct devfs_mount *dmp;
1185 	int dm_unlock;
1186 
1187 	if (devfs_populate_vp(ap->a_dvp) != 0)
1188 		return (ENOTDIR);
1189 
1190 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1191 	dm_unlock = 1;
1192 	j = devfs_lookupx(ap, &dm_unlock);
1193 	if (dm_unlock == 1)
1194 		sx_xunlock(&dmp->dm_lock);
1195 	return (j);
1196 }
1197 
1198 static int
1199 devfs_mknod(struct vop_mknod_args *ap)
1200 {
1201 	struct componentname *cnp;
1202 	struct vnode *dvp, **vpp;
1203 	struct devfs_dirent *dd, *de;
1204 	struct devfs_mount *dmp;
1205 	int error;
1206 
1207 	/*
1208 	 * The only type of node we should be creating here is a
1209 	 * character device, for anything else return EOPNOTSUPP.
1210 	 */
1211 	if (ap->a_vap->va_type != VCHR)
1212 		return (EOPNOTSUPP);
1213 	dvp = ap->a_dvp;
1214 	dmp = VFSTODEVFS(dvp->v_mount);
1215 
1216 	cnp = ap->a_cnp;
1217 	vpp = ap->a_vpp;
1218 	dd = dvp->v_data;
1219 
1220 	error = ENOENT;
1221 	sx_xlock(&dmp->dm_lock);
1222 	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1223 		if (cnp->cn_namelen != de->de_dirent->d_namlen)
1224 			continue;
1225 		if (de->de_dirent->d_type == DT_CHR &&
1226 		    (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0)
1227 			continue;
1228 		if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1229 		    de->de_dirent->d_namlen) != 0)
1230 			continue;
1231 		if (de->de_flags & DE_WHITEOUT)
1232 			break;
1233 		goto notfound;
1234 	}
1235 	if (de == NULL)
1236 		goto notfound;
1237 	de->de_flags &= ~DE_WHITEOUT;
1238 	error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1239 	return (error);
1240 notfound:
1241 	sx_xunlock(&dmp->dm_lock);
1242 	return (error);
1243 }
1244 
1245 /* ARGSUSED */
1246 static int
1247 devfs_open(struct vop_open_args *ap)
1248 {
1249 	struct thread *td = ap->a_td;
1250 	struct vnode *vp = ap->a_vp;
1251 	struct cdev *dev = vp->v_rdev;
1252 	struct file *fp = ap->a_fp;
1253 	int error, ref, vlocked;
1254 	struct cdevsw *dsw;
1255 	struct file *fpop;
1256 
1257 	if (vp->v_type == VBLK)
1258 		return (ENXIO);
1259 
1260 	if (dev == NULL)
1261 		return (ENXIO);
1262 
1263 	/* Make this field valid before any I/O in d_open. */
1264 	if (dev->si_iosize_max == 0)
1265 		dev->si_iosize_max = DFLTPHYS;
1266 
1267 	dsw = dev_refthread(dev, &ref);
1268 	if (dsw == NULL)
1269 		return (ENXIO);
1270 	if (fp == NULL && dsw->d_fdopen != NULL) {
1271 		dev_relthread(dev, ref);
1272 		return (ENXIO);
1273 	}
1274 
1275 	if (vp->v_type == VCHR)
1276 		devfs_usecount_add(vp);
1277 
1278 	vlocked = VOP_ISLOCKED(vp);
1279 	VOP_UNLOCK(vp);
1280 
1281 	fpop = td->td_fpop;
1282 	td->td_fpop = fp;
1283 	if (fp != NULL) {
1284 		fp->f_data = dev;
1285 		fp->f_vnode = vp;
1286 	}
1287 	if (dsw->d_fdopen != NULL)
1288 		error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1289 	else
1290 		error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1291 	/* Clean up any cdevpriv upon error. */
1292 	if (error != 0)
1293 		devfs_clear_cdevpriv();
1294 	td->td_fpop = fpop;
1295 
1296 	vn_lock(vp, vlocked | LK_RETRY);
1297 	if (error != 0 && vp->v_type == VCHR)
1298 		devfs_usecount_sub(vp);
1299 
1300 	dev_relthread(dev, ref);
1301 	if (error != 0) {
1302 		if (error == ERESTART)
1303 			error = EINTR;
1304 		return (error);
1305 	}
1306 
1307 #if 0	/* /dev/console */
1308 	KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1309 #else
1310 	if (fp == NULL)
1311 		return (error);
1312 #endif
1313 	if (fp->f_ops == &badfileops)
1314 		finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1315 	return (error);
1316 }
1317 
1318 static int
1319 devfs_pathconf(struct vop_pathconf_args *ap)
1320 {
1321 
1322 	switch (ap->a_name) {
1323 	case _PC_FILESIZEBITS:
1324 		*ap->a_retval = 64;
1325 		return (0);
1326 	case _PC_NAME_MAX:
1327 		*ap->a_retval = NAME_MAX;
1328 		return (0);
1329 	case _PC_LINK_MAX:
1330 		*ap->a_retval = INT_MAX;
1331 		return (0);
1332 	case _PC_SYMLINK_MAX:
1333 		*ap->a_retval = MAXPATHLEN;
1334 		return (0);
1335 	case _PC_MAX_CANON:
1336 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1337 			*ap->a_retval = MAX_CANON;
1338 			return (0);
1339 		}
1340 		return (EINVAL);
1341 	case _PC_MAX_INPUT:
1342 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1343 			*ap->a_retval = MAX_INPUT;
1344 			return (0);
1345 		}
1346 		return (EINVAL);
1347 	case _PC_VDISABLE:
1348 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1349 			*ap->a_retval = _POSIX_VDISABLE;
1350 			return (0);
1351 		}
1352 		return (EINVAL);
1353 	case _PC_MAC_PRESENT:
1354 #ifdef MAC
1355 		/*
1356 		 * If MAC is enabled, devfs automatically supports
1357 		 * trivial non-persistant label storage.
1358 		 */
1359 		*ap->a_retval = 1;
1360 #else
1361 		*ap->a_retval = 0;
1362 #endif
1363 		return (0);
1364 	case _PC_CHOWN_RESTRICTED:
1365 		*ap->a_retval = 1;
1366 		return (0);
1367 	default:
1368 		return (vop_stdpathconf(ap));
1369 	}
1370 	/* NOTREACHED */
1371 }
1372 
1373 /* ARGSUSED */
1374 static int
1375 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1376 {
1377 	struct cdev *dev;
1378 	struct cdevsw *dsw;
1379 	int error, ref;
1380 	struct file *fpop;
1381 
1382 	fpop = td->td_fpop;
1383 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1384 	if (error != 0) {
1385 		error = vnops.fo_poll(fp, events, cred, td);
1386 		return (error);
1387 	}
1388 	error = dsw->d_poll(dev, events, td);
1389 	td->td_fpop = fpop;
1390 	dev_relthread(dev, ref);
1391 	return(error);
1392 }
1393 
1394 /*
1395  * Print out the contents of a special device vnode.
1396  */
1397 static int
1398 devfs_print(struct vop_print_args *ap)
1399 {
1400 
1401 	printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1402 	return (0);
1403 }
1404 
1405 static int
1406 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
1407     int flags, struct thread *td)
1408 {
1409 	struct cdev *dev;
1410 	int ioflag, error, ref;
1411 	ssize_t resid;
1412 	struct cdevsw *dsw;
1413 	struct file *fpop;
1414 
1415 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1416 		return (EINVAL);
1417 	fpop = td->td_fpop;
1418 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1419 	if (error != 0) {
1420 		error = vnops.fo_read(fp, uio, cred, flags, td);
1421 		return (error);
1422 	}
1423 	resid = uio->uio_resid;
1424 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1425 	if (ioflag & O_DIRECT)
1426 		ioflag |= IO_DIRECT;
1427 
1428 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1429 	error = dsw->d_read(dev, uio, ioflag);
1430 	if (uio->uio_resid != resid || (error == 0 && resid != 0))
1431 		devfs_timestamp(&dev->si_atime);
1432 	td->td_fpop = fpop;
1433 	dev_relthread(dev, ref);
1434 
1435 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF_R);
1436 	return (error);
1437 }
1438 
1439 static int
1440 devfs_readdir(struct vop_readdir_args *ap)
1441 {
1442 	int error;
1443 	struct uio *uio;
1444 	struct dirent *dp;
1445 	struct devfs_dirent *dd;
1446 	struct devfs_dirent *de;
1447 	struct devfs_mount *dmp;
1448 	off_t off;
1449 	int *tmp_ncookies = NULL;
1450 
1451 	if (ap->a_vp->v_type != VDIR)
1452 		return (ENOTDIR);
1453 
1454 	uio = ap->a_uio;
1455 	if (uio->uio_offset < 0)
1456 		return (EINVAL);
1457 
1458 	/*
1459 	 * XXX: This is a temporary hack to get around this filesystem not
1460 	 * supporting cookies. We store the location of the ncookies pointer
1461 	 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1462 	 * and set the number of cookies to 0. We then set the pointer to
1463 	 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1464 	 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1465 	 * pointer to its original location before returning to the caller.
1466 	 */
1467 	if (ap->a_ncookies != NULL) {
1468 		tmp_ncookies = ap->a_ncookies;
1469 		*ap->a_ncookies = 0;
1470 		ap->a_ncookies = NULL;
1471 	}
1472 
1473 	dmp = VFSTODEVFS(ap->a_vp->v_mount);
1474 	if (devfs_populate_vp(ap->a_vp) != 0) {
1475 		if (tmp_ncookies != NULL)
1476 			ap->a_ncookies = tmp_ncookies;
1477 		return (EIO);
1478 	}
1479 	error = 0;
1480 	de = ap->a_vp->v_data;
1481 	off = 0;
1482 	TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1483 		KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1484 		if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1485 			continue;
1486 		if (devfs_prison_check(dd, uio->uio_td))
1487 			continue;
1488 		if (dd->de_dirent->d_type == DT_DIR)
1489 			de = dd->de_dir;
1490 		else
1491 			de = dd;
1492 		dp = dd->de_dirent;
1493 		MPASS(dp->d_reclen == GENERIC_DIRSIZ(dp));
1494 		if (dp->d_reclen > uio->uio_resid)
1495 			break;
1496 		dp->d_fileno = de->de_inode;
1497 		/* NOTE: d_off is the offset for the *next* entry. */
1498 		dp->d_off = off + dp->d_reclen;
1499 		if (off >= uio->uio_offset) {
1500 			error = vfs_read_dirent(ap, dp, off);
1501 			if (error)
1502 				break;
1503 		}
1504 		off += dp->d_reclen;
1505 	}
1506 	sx_xunlock(&dmp->dm_lock);
1507 	uio->uio_offset = off;
1508 
1509 	/*
1510 	 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1511 	 * place.
1512 	 */
1513 	if (tmp_ncookies != NULL)
1514 		ap->a_ncookies = tmp_ncookies;
1515 
1516 	return (error);
1517 }
1518 
1519 static int
1520 devfs_readlink(struct vop_readlink_args *ap)
1521 {
1522 	struct devfs_dirent *de;
1523 
1524 	de = ap->a_vp->v_data;
1525 	return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1526 }
1527 
1528 static void
1529 devfs_reclaiml(struct vnode *vp)
1530 {
1531 	struct devfs_dirent *de;
1532 
1533 	mtx_assert(&devfs_de_interlock, MA_OWNED);
1534 	de = vp->v_data;
1535 	if (de != NULL) {
1536 		MPASS(de->de_usecount == 0);
1537 		de->de_vnode = NULL;
1538 		vp->v_data = NULL;
1539 	}
1540 }
1541 
1542 static int
1543 devfs_reclaim(struct vop_reclaim_args *ap)
1544 {
1545 	struct vnode *vp;
1546 
1547 	vp = ap->a_vp;
1548 	mtx_lock(&devfs_de_interlock);
1549 	devfs_reclaiml(vp);
1550 	mtx_unlock(&devfs_de_interlock);
1551 	return (0);
1552 }
1553 
1554 static int
1555 devfs_reclaim_vchr(struct vop_reclaim_args *ap)
1556 {
1557 	struct vnode *vp;
1558 	struct cdev *dev;
1559 
1560 	vp = ap->a_vp;
1561 	MPASS(vp->v_type == VCHR);
1562 
1563 	mtx_lock(&devfs_de_interlock);
1564 	VI_LOCK(vp);
1565 	devfs_usecount_subl(vp);
1566 	devfs_reclaiml(vp);
1567 	mtx_unlock(&devfs_de_interlock);
1568 	dev_lock();
1569 	dev = vp->v_rdev;
1570 	vp->v_rdev = NULL;
1571 	dev_unlock();
1572 	VI_UNLOCK(vp);
1573 	if (dev != NULL)
1574 		dev_rel(dev);
1575 	return (0);
1576 }
1577 
1578 static int
1579 devfs_remove(struct vop_remove_args *ap)
1580 {
1581 	struct vnode *dvp = ap->a_dvp;
1582 	struct vnode *vp = ap->a_vp;
1583 	struct devfs_dirent *dd;
1584 	struct devfs_dirent *de, *de_covered;
1585 	struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1586 
1587 	ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1588 	ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1589 
1590 	sx_xlock(&dmp->dm_lock);
1591 	dd = ap->a_dvp->v_data;
1592 	de = vp->v_data;
1593 	if (de->de_cdp == NULL) {
1594 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1595 		if (de->de_dirent->d_type == DT_LNK) {
1596 			de_covered = devfs_find(dd, de->de_dirent->d_name,
1597 			    de->de_dirent->d_namlen, 0);
1598 			if (de_covered != NULL)
1599 				de_covered->de_flags &= ~DE_COVERED;
1600 		}
1601 		/* We need to unlock dvp because devfs_delete() may lock it. */
1602 		VOP_UNLOCK(vp);
1603 		if (dvp != vp)
1604 			VOP_UNLOCK(dvp);
1605 		devfs_delete(dmp, de, 0);
1606 		sx_xunlock(&dmp->dm_lock);
1607 		if (dvp != vp)
1608 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1609 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1610 	} else {
1611 		de->de_flags |= DE_WHITEOUT;
1612 		sx_xunlock(&dmp->dm_lock);
1613 	}
1614 	return (0);
1615 }
1616 
1617 /*
1618  * Revoke is called on a tty when a terminal session ends.  The vnode
1619  * is orphaned by setting v_op to deadfs so we need to let go of it
1620  * as well so that we create a new one next time around.
1621  *
1622  */
1623 static int
1624 devfs_revoke(struct vop_revoke_args *ap)
1625 {
1626 	struct vnode *vp = ap->a_vp, *vp2;
1627 	struct cdev *dev;
1628 	struct cdev_priv *cdp;
1629 	struct devfs_dirent *de;
1630 	enum vgetstate vs;
1631 	u_int i;
1632 
1633 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1634 
1635 	dev = vp->v_rdev;
1636 	cdp = cdev2priv(dev);
1637 
1638 	dev_lock();
1639 	cdp->cdp_inuse++;
1640 	dev_unlock();
1641 
1642 	vhold(vp);
1643 	vgone(vp);
1644 	vdrop(vp);
1645 
1646 	VOP_UNLOCK(vp);
1647  loop:
1648 	for (;;) {
1649 		mtx_lock(&devfs_de_interlock);
1650 		dev_lock();
1651 		vp2 = NULL;
1652 		for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1653 			de = cdp->cdp_dirents[i];
1654 			if (de == NULL)
1655 				continue;
1656 
1657 			vp2 = de->de_vnode;
1658 			if (vp2 != NULL) {
1659 				dev_unlock();
1660 				vs = vget_prep(vp2);
1661 				mtx_unlock(&devfs_de_interlock);
1662 				if (vget_finish(vp2, LK_EXCLUSIVE, vs) != 0)
1663 					goto loop;
1664 				vhold(vp2);
1665 				vgone(vp2);
1666 				vdrop(vp2);
1667 				vput(vp2);
1668 				break;
1669 			}
1670 		}
1671 		if (vp2 != NULL) {
1672 			continue;
1673 		}
1674 		dev_unlock();
1675 		mtx_unlock(&devfs_de_interlock);
1676 		break;
1677 	}
1678 	dev_lock();
1679 	cdp->cdp_inuse--;
1680 	if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1681 		TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1682 		dev_unlock();
1683 		dev_rel(&cdp->cdp_c);
1684 	} else
1685 		dev_unlock();
1686 
1687 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1688 	return (0);
1689 }
1690 
1691 static int
1692 devfs_rioctl(struct vop_ioctl_args *ap)
1693 {
1694 	struct vnode *vp;
1695 	struct devfs_mount *dmp;
1696 	int error;
1697 
1698 	vp = ap->a_vp;
1699 	vn_lock(vp, LK_SHARED | LK_RETRY);
1700 	if (VN_IS_DOOMED(vp)) {
1701 		VOP_UNLOCK(vp);
1702 		return (EBADF);
1703 	}
1704 	dmp = VFSTODEVFS(vp->v_mount);
1705 	sx_xlock(&dmp->dm_lock);
1706 	VOP_UNLOCK(vp);
1707 	DEVFS_DMP_HOLD(dmp);
1708 	devfs_populate(dmp);
1709 	if (DEVFS_DMP_DROP(dmp)) {
1710 		sx_xunlock(&dmp->dm_lock);
1711 		devfs_unmount_final(dmp);
1712 		return (ENOENT);
1713 	}
1714 	error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1715 	sx_xunlock(&dmp->dm_lock);
1716 	return (error);
1717 }
1718 
1719 static int
1720 devfs_rread(struct vop_read_args *ap)
1721 {
1722 
1723 	if (ap->a_vp->v_type != VDIR)
1724 		return (EINVAL);
1725 	return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1726 }
1727 
1728 static int
1729 devfs_setattr(struct vop_setattr_args *ap)
1730 {
1731 	struct devfs_dirent *de;
1732 	struct vattr *vap;
1733 	struct vnode *vp;
1734 	struct thread *td;
1735 	int c, error;
1736 	uid_t uid;
1737 	gid_t gid;
1738 
1739 	vap = ap->a_vap;
1740 	vp = ap->a_vp;
1741 	td = curthread;
1742 	if ((vap->va_type != VNON) ||
1743 	    (vap->va_nlink != VNOVAL) ||
1744 	    (vap->va_fsid != VNOVAL) ||
1745 	    (vap->va_fileid != VNOVAL) ||
1746 	    (vap->va_blocksize != VNOVAL) ||
1747 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1748 	    (vap->va_rdev != VNOVAL) ||
1749 	    ((int)vap->va_bytes != VNOVAL) ||
1750 	    (vap->va_gen != VNOVAL)) {
1751 		return (EINVAL);
1752 	}
1753 
1754 	error = devfs_populate_vp(vp);
1755 	if (error != 0)
1756 		return (error);
1757 
1758 	de = vp->v_data;
1759 	if (vp->v_type == VDIR)
1760 		de = de->de_dir;
1761 
1762 	c = 0;
1763 	if (vap->va_uid == (uid_t)VNOVAL)
1764 		uid = de->de_uid;
1765 	else
1766 		uid = vap->va_uid;
1767 	if (vap->va_gid == (gid_t)VNOVAL)
1768 		gid = de->de_gid;
1769 	else
1770 		gid = vap->va_gid;
1771 	if (uid != de->de_uid || gid != de->de_gid) {
1772 		if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1773 		    (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1774 			error = priv_check(td, PRIV_VFS_CHOWN);
1775 			if (error != 0)
1776 				goto ret;
1777 		}
1778 		de->de_uid = uid;
1779 		de->de_gid = gid;
1780 		c = 1;
1781 	}
1782 
1783 	if (vap->va_mode != (mode_t)VNOVAL) {
1784 		if (ap->a_cred->cr_uid != de->de_uid) {
1785 			error = priv_check(td, PRIV_VFS_ADMIN);
1786 			if (error != 0)
1787 				goto ret;
1788 		}
1789 		de->de_mode = vap->va_mode;
1790 		c = 1;
1791 	}
1792 
1793 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1794 		error = vn_utimes_perm(vp, vap, ap->a_cred, td);
1795 		if (error != 0)
1796 			goto ret;
1797 		if (vap->va_atime.tv_sec != VNOVAL) {
1798 			if (vp->v_type == VCHR)
1799 				vp->v_rdev->si_atime = vap->va_atime;
1800 			else
1801 				de->de_atime = vap->va_atime;
1802 		}
1803 		if (vap->va_mtime.tv_sec != VNOVAL) {
1804 			if (vp->v_type == VCHR)
1805 				vp->v_rdev->si_mtime = vap->va_mtime;
1806 			else
1807 				de->de_mtime = vap->va_mtime;
1808 		}
1809 		c = 1;
1810 	}
1811 
1812 	if (c) {
1813 		if (vp->v_type == VCHR)
1814 			vfs_timestamp(&vp->v_rdev->si_ctime);
1815 		else
1816 			vfs_timestamp(&de->de_mtime);
1817 	}
1818 
1819 ret:
1820 	sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock);
1821 	return (error);
1822 }
1823 
1824 #ifdef MAC
1825 static int
1826 devfs_setlabel(struct vop_setlabel_args *ap)
1827 {
1828 	struct vnode *vp;
1829 	struct devfs_dirent *de;
1830 
1831 	vp = ap->a_vp;
1832 	de = vp->v_data;
1833 
1834 	mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1835 	mac_devfs_update(vp->v_mount, de, vp);
1836 
1837 	return (0);
1838 }
1839 #endif
1840 
1841 static int
1842 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1843 {
1844 
1845 	return (vnops.fo_stat(fp, sb, cred, td));
1846 }
1847 
1848 static int
1849 devfs_symlink(struct vop_symlink_args *ap)
1850 {
1851 	int i, error;
1852 	struct devfs_dirent *dd;
1853 	struct devfs_dirent *de, *de_covered, *de_dotdot;
1854 	struct devfs_mount *dmp;
1855 
1856 	error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1857 	if (error)
1858 		return(error);
1859 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1860 	if (devfs_populate_vp(ap->a_dvp) != 0)
1861 		return (ENOENT);
1862 
1863 	dd = ap->a_dvp->v_data;
1864 	de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1865 	de->de_flags = DE_USER;
1866 	de->de_uid = 0;
1867 	de->de_gid = 0;
1868 	de->de_mode = 0755;
1869 	de->de_inode = alloc_unr(devfs_inos);
1870 	de->de_dir = dd;
1871 	de->de_dirent->d_type = DT_LNK;
1872 	i = strlen(ap->a_target) + 1;
1873 	de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1874 	bcopy(ap->a_target, de->de_symlink, i);
1875 #ifdef MAC
1876 	mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1877 #endif
1878 	de_covered = devfs_find(dd, de->de_dirent->d_name,
1879 	    de->de_dirent->d_namlen, 0);
1880 	if (de_covered != NULL) {
1881 		if ((de_covered->de_flags & DE_USER) != 0) {
1882 			devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1883 			sx_xunlock(&dmp->dm_lock);
1884 			return (EEXIST);
1885 		}
1886 		KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1887 		    ("devfs_symlink: entry %p already covered", de_covered));
1888 		de_covered->de_flags |= DE_COVERED;
1889 	}
1890 
1891 	de_dotdot = TAILQ_FIRST(&dd->de_dlist);		/* "." */
1892 	de_dotdot = TAILQ_NEXT(de_dotdot, de_list);	/* ".." */
1893 	TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1894 	devfs_dir_ref_de(dmp, dd);
1895 	devfs_rules_apply(dmp, de);
1896 
1897 	return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1898 }
1899 
1900 static int
1901 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1902 {
1903 
1904 	return (vnops.fo_truncate(fp, length, cred, td));
1905 }
1906 
1907 static int
1908 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
1909     int flags, struct thread *td)
1910 {
1911 	struct cdev *dev;
1912 	int error, ioflag, ref;
1913 	ssize_t resid;
1914 	struct cdevsw *dsw;
1915 	struct file *fpop;
1916 
1917 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1918 		return (EINVAL);
1919 	fpop = td->td_fpop;
1920 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1921 	if (error != 0) {
1922 		error = vnops.fo_write(fp, uio, cred, flags, td);
1923 		return (error);
1924 	}
1925 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1926 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1927 	if (ioflag & O_DIRECT)
1928 		ioflag |= IO_DIRECT;
1929 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1930 
1931 	resid = uio->uio_resid;
1932 
1933 	error = dsw->d_write(dev, uio, ioflag);
1934 	if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1935 		devfs_timestamp(&dev->si_ctime);
1936 		dev->si_mtime = dev->si_ctime;
1937 	}
1938 	td->td_fpop = fpop;
1939 	dev_relthread(dev, ref);
1940 
1941 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF_W);
1942 	return (error);
1943 }
1944 
1945 static int
1946 devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1947     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1948     struct thread *td)
1949 {
1950 	struct cdev *dev;
1951 	struct cdevsw *dsw;
1952 	struct mount *mp;
1953 	struct vnode *vp;
1954 	struct file *fpop;
1955 	vm_object_t object;
1956 	vm_prot_t maxprot;
1957 	int error, ref;
1958 
1959 	vp = fp->f_vnode;
1960 
1961 	/*
1962 	 * Ensure that file and memory protections are
1963 	 * compatible.
1964 	 */
1965 	mp = vp->v_mount;
1966 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1967 		maxprot = VM_PROT_NONE;
1968 		if ((prot & VM_PROT_EXECUTE) != 0)
1969 			return (EACCES);
1970 	} else
1971 		maxprot = VM_PROT_EXECUTE;
1972 	if ((fp->f_flag & FREAD) != 0)
1973 		maxprot |= VM_PROT_READ;
1974 	else if ((prot & VM_PROT_READ) != 0)
1975 		return (EACCES);
1976 
1977 	/*
1978 	 * If we are sharing potential changes via MAP_SHARED and we
1979 	 * are trying to get write permission although we opened it
1980 	 * without asking for it, bail out.
1981 	 *
1982 	 * Note that most character devices always share mappings.
1983 	 * The one exception is that D_MMAP_ANON devices
1984 	 * (i.e. /dev/zero) permit private writable mappings.
1985 	 *
1986 	 * Rely on vm_mmap_cdev() to fail invalid MAP_PRIVATE requests
1987 	 * as well as updating maxprot to permit writing for
1988 	 * D_MMAP_ANON devices rather than doing that here.
1989 	 */
1990 	if ((flags & MAP_SHARED) != 0) {
1991 		if ((fp->f_flag & FWRITE) != 0)
1992 			maxprot |= VM_PROT_WRITE;
1993 		else if ((prot & VM_PROT_WRITE) != 0)
1994 			return (EACCES);
1995 	}
1996 	maxprot &= cap_maxprot;
1997 
1998 	fpop = td->td_fpop;
1999 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
2000 	if (error != 0)
2001 		return (error);
2002 
2003 	error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, dev, dsw, &foff,
2004 	    &object);
2005 	td->td_fpop = fpop;
2006 	dev_relthread(dev, ref);
2007 	if (error != 0)
2008 		return (error);
2009 
2010 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2011 	    foff, FALSE, td);
2012 	if (error != 0)
2013 		vm_object_deallocate(object);
2014 	return (error);
2015 }
2016 
2017 dev_t
2018 dev2udev(struct cdev *x)
2019 {
2020 	if (x == NULL)
2021 		return (NODEV);
2022 	return (cdev2priv(x)->cdp_inode);
2023 }
2024 
2025 static struct fileops devfs_ops_f = {
2026 	.fo_read =	devfs_read_f,
2027 	.fo_write =	devfs_write_f,
2028 	.fo_truncate =	devfs_truncate_f,
2029 	.fo_ioctl =	devfs_ioctl_f,
2030 	.fo_poll =	devfs_poll_f,
2031 	.fo_kqfilter =	devfs_kqfilter_f,
2032 	.fo_stat =	devfs_stat_f,
2033 	.fo_close =	devfs_close_f,
2034 	.fo_chmod =	vn_chmod,
2035 	.fo_chown =	vn_chown,
2036 	.fo_sendfile =	vn_sendfile,
2037 	.fo_seek =	vn_seek,
2038 	.fo_fill_kinfo = vn_fill_kinfo,
2039 	.fo_mmap =	devfs_mmap_f,
2040 	.fo_flags =	DFLAG_PASSABLE | DFLAG_SEEKABLE
2041 };
2042 
2043 /* Vops for non-CHR vnodes in /dev. */
2044 static struct vop_vector devfs_vnodeops = {
2045 	.vop_default =		&default_vnodeops,
2046 
2047 	.vop_access =		devfs_access,
2048 	.vop_getattr =		devfs_getattr,
2049 	.vop_ioctl =		devfs_rioctl,
2050 	.vop_lookup =		devfs_lookup,
2051 	.vop_mknod =		devfs_mknod,
2052 	.vop_pathconf =		devfs_pathconf,
2053 	.vop_read =		devfs_rread,
2054 	.vop_readdir =		devfs_readdir,
2055 	.vop_readlink =		devfs_readlink,
2056 	.vop_reclaim =		devfs_reclaim,
2057 	.vop_remove =		devfs_remove,
2058 	.vop_revoke =		devfs_revoke,
2059 	.vop_setattr =		devfs_setattr,
2060 #ifdef MAC
2061 	.vop_setlabel =		devfs_setlabel,
2062 #endif
2063 	.vop_symlink =		devfs_symlink,
2064 	.vop_vptocnp =		devfs_vptocnp,
2065 	.vop_lock1 =		vop_lock,
2066 	.vop_unlock =		vop_unlock,
2067 	.vop_islocked =		vop_islocked,
2068 };
2069 VFS_VOP_VECTOR_REGISTER(devfs_vnodeops);
2070 
2071 /* Vops for VCHR vnodes in /dev. */
2072 static struct vop_vector devfs_specops = {
2073 	.vop_default =		&default_vnodeops,
2074 
2075 	.vop_access =		devfs_access,
2076 	.vop_bmap =		VOP_PANIC,
2077 	.vop_close =		devfs_close,
2078 	.vop_create =		VOP_PANIC,
2079 	.vop_fsync =		vop_stdfsync,
2080 	.vop_getattr =		devfs_getattr,
2081 	.vop_ioctl =		devfs_ioctl,
2082 	.vop_link =		VOP_PANIC,
2083 	.vop_mkdir =		VOP_PANIC,
2084 	.vop_mknod =		VOP_PANIC,
2085 	.vop_open =		devfs_open,
2086 	.vop_pathconf =		devfs_pathconf,
2087 	.vop_poll =		dead_poll,
2088 	.vop_print =		devfs_print,
2089 	.vop_read =		dead_read,
2090 	.vop_readdir =		VOP_PANIC,
2091 	.vop_readlink =		VOP_PANIC,
2092 	.vop_reallocblks =	VOP_PANIC,
2093 	.vop_reclaim =		devfs_reclaim_vchr,
2094 	.vop_remove =		devfs_remove,
2095 	.vop_rename =		VOP_PANIC,
2096 	.vop_revoke =		devfs_revoke,
2097 	.vop_rmdir =		VOP_PANIC,
2098 	.vop_setattr =		devfs_setattr,
2099 #ifdef MAC
2100 	.vop_setlabel =		devfs_setlabel,
2101 #endif
2102 	.vop_strategy =		VOP_PANIC,
2103 	.vop_symlink =		VOP_PANIC,
2104 	.vop_vptocnp =		devfs_vptocnp,
2105 	.vop_write =		dead_write,
2106 	.vop_lock1 =		vop_lock,
2107 	.vop_unlock =		vop_unlock,
2108 	.vop_islocked =		vop_islocked,
2109 };
2110 VFS_VOP_VECTOR_REGISTER(devfs_specops);
2111 
2112 /*
2113  * Our calling convention to the device drivers used to be that we passed
2114  * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
2115  * flags instead since that's what open(), close() and ioctl() takes and
2116  * we don't really want vnode.h in device drivers.
2117  * We solved the source compatibility by redefining some vnode flags to
2118  * be the same as the fcntl ones and by sending down the bitwise OR of
2119  * the respective fcntl/vnode flags.  These CTASSERTS make sure nobody
2120  * pulls the rug out under this.
2121  */
2122 CTASSERT(O_NONBLOCK == IO_NDELAY);
2123 CTASSERT(O_FSYNC == IO_SYNC);
2124