xref: /freebsd/sys/fs/devfs/devfs_vnops.c (revision 788ca347b816afd83b2885e0c79aeeb88649b2ab)
1 /*-
2  * Copyright (c) 2000-2004
3  *	Poul-Henning Kamp.  All rights reserved.
4  * Copyright (c) 1989, 1992-1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software donated to Berkeley by
8  * Jan-Simon Pendry.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)kernfs_vnops.c	8.15 (Berkeley) 5/21/95
32  * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vnops.c 1.43
33  *
34  * $FreeBSD$
35  */
36 
37 /*
38  * TODO:
39  *	mkdir: want it ?
40  */
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/conf.h>
45 #include <sys/dirent.h>
46 #include <sys/fcntl.h>
47 #include <sys/file.h>
48 #include <sys/filedesc.h>
49 #include <sys/filio.h>
50 #include <sys/jail.h>
51 #include <sys/kernel.h>
52 #include <sys/lock.h>
53 #include <sys/malloc.h>
54 #include <sys/mount.h>
55 #include <sys/namei.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/stat.h>
59 #include <sys/sx.h>
60 #include <sys/sysctl.h>
61 #include <sys/time.h>
62 #include <sys/ttycom.h>
63 #include <sys/unistd.h>
64 #include <sys/vnode.h>
65 
66 static struct vop_vector devfs_vnodeops;
67 static struct fileops devfs_ops_f;
68 
69 #include <fs/devfs/devfs.h>
70 #include <fs/devfs/devfs_int.h>
71 
72 #include <security/mac/mac_framework.h>
73 
74 static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data");
75 
76 struct mtx	devfs_de_interlock;
77 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF);
78 struct sx	clone_drain_lock;
79 SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock");
80 struct mtx	cdevpriv_mtx;
81 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
82 
83 SYSCTL_DECL(_vfs_devfs);
84 
85 static int devfs_dotimes;
86 SYSCTL_INT(_vfs_devfs, OID_AUTO, dotimes, CTLFLAG_RW,
87     &devfs_dotimes, 0, "Update timestamps on DEVFS with default precision");
88 
89 /*
90  * Update devfs node timestamp.  Note that updates are unlocked and
91  * stat(2) could see partially updated times.
92  */
93 static void
94 devfs_timestamp(struct timespec *tsp)
95 {
96 	time_t ts;
97 
98 	if (devfs_dotimes) {
99 		vfs_timestamp(tsp);
100 	} else {
101 		ts = time_second;
102 		if (tsp->tv_sec != ts) {
103 			tsp->tv_sec = ts;
104 			tsp->tv_nsec = 0;
105 		}
106 	}
107 }
108 
109 static int
110 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp,
111     int *ref)
112 {
113 
114 	*dswp = devvn_refthread(fp->f_vnode, devp, ref);
115 	if (*devp != fp->f_data) {
116 		if (*dswp != NULL)
117 			dev_relthread(*devp, *ref);
118 		return (ENXIO);
119 	}
120 	KASSERT((*devp)->si_refcount > 0,
121 	    ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
122 	if (*dswp == NULL)
123 		return (ENXIO);
124 	curthread->td_fpop = fp;
125 	return (0);
126 }
127 
128 int
129 devfs_get_cdevpriv(void **datap)
130 {
131 	struct file *fp;
132 	struct cdev_privdata *p;
133 	int error;
134 
135 	fp = curthread->td_fpop;
136 	if (fp == NULL)
137 		return (EBADF);
138 	p = fp->f_cdevpriv;
139 	if (p != NULL) {
140 		error = 0;
141 		*datap = p->cdpd_data;
142 	} else
143 		error = ENOENT;
144 	return (error);
145 }
146 
147 int
148 devfs_set_cdevpriv(void *priv, cdevpriv_dtr_t priv_dtr)
149 {
150 	struct file *fp;
151 	struct cdev_priv *cdp;
152 	struct cdev_privdata *p;
153 	int error;
154 
155 	fp = curthread->td_fpop;
156 	if (fp == NULL)
157 		return (ENOENT);
158 	cdp = cdev2priv((struct cdev *)fp->f_data);
159 	p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
160 	p->cdpd_data = priv;
161 	p->cdpd_dtr = priv_dtr;
162 	p->cdpd_fp = fp;
163 	mtx_lock(&cdevpriv_mtx);
164 	if (fp->f_cdevpriv == NULL) {
165 		LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
166 		fp->f_cdevpriv = p;
167 		mtx_unlock(&cdevpriv_mtx);
168 		error = 0;
169 	} else {
170 		mtx_unlock(&cdevpriv_mtx);
171 		free(p, M_CDEVPDATA);
172 		error = EBUSY;
173 	}
174 	return (error);
175 }
176 
177 void
178 devfs_destroy_cdevpriv(struct cdev_privdata *p)
179 {
180 
181 	mtx_assert(&cdevpriv_mtx, MA_OWNED);
182 	p->cdpd_fp->f_cdevpriv = NULL;
183 	LIST_REMOVE(p, cdpd_list);
184 	mtx_unlock(&cdevpriv_mtx);
185 	(p->cdpd_dtr)(p->cdpd_data);
186 	free(p, M_CDEVPDATA);
187 }
188 
189 void
190 devfs_fpdrop(struct file *fp)
191 {
192 	struct cdev_privdata *p;
193 
194 	mtx_lock(&cdevpriv_mtx);
195 	if ((p = fp->f_cdevpriv) == NULL) {
196 		mtx_unlock(&cdevpriv_mtx);
197 		return;
198 	}
199 	devfs_destroy_cdevpriv(p);
200 }
201 
202 void
203 devfs_clear_cdevpriv(void)
204 {
205 	struct file *fp;
206 
207 	fp = curthread->td_fpop;
208 	if (fp == NULL)
209 		return;
210 	devfs_fpdrop(fp);
211 }
212 
213 /*
214  * On success devfs_populate_vp() returns with dmp->dm_lock held.
215  */
216 static int
217 devfs_populate_vp(struct vnode *vp)
218 {
219 	struct devfs_dirent *de;
220 	struct devfs_mount *dmp;
221 	int locked;
222 
223 	ASSERT_VOP_LOCKED(vp, "devfs_populate_vp");
224 
225 	dmp = VFSTODEVFS(vp->v_mount);
226 	locked = VOP_ISLOCKED(vp);
227 
228 	sx_xlock(&dmp->dm_lock);
229 	DEVFS_DMP_HOLD(dmp);
230 
231 	/* Can't call devfs_populate() with the vnode lock held. */
232 	VOP_UNLOCK(vp, 0);
233 	devfs_populate(dmp);
234 
235 	sx_xunlock(&dmp->dm_lock);
236 	vn_lock(vp, locked | LK_RETRY);
237 	sx_xlock(&dmp->dm_lock);
238 	if (DEVFS_DMP_DROP(dmp)) {
239 		sx_xunlock(&dmp->dm_lock);
240 		devfs_unmount_final(dmp);
241 		return (EBADF);
242 	}
243 	if ((vp->v_iflag & VI_DOOMED) != 0) {
244 		sx_xunlock(&dmp->dm_lock);
245 		return (EBADF);
246 	}
247 	de = vp->v_data;
248 	KASSERT(de != NULL,
249 	    ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed"));
250 	if ((de->de_flags & DE_DOOMED) != 0) {
251 		sx_xunlock(&dmp->dm_lock);
252 		return (EBADF);
253 	}
254 
255 	return (0);
256 }
257 
258 static int
259 devfs_vptocnp(struct vop_vptocnp_args *ap)
260 {
261 	struct vnode *vp = ap->a_vp;
262 	struct vnode **dvp = ap->a_vpp;
263 	struct devfs_mount *dmp;
264 	char *buf = ap->a_buf;
265 	int *buflen = ap->a_buflen;
266 	struct devfs_dirent *dd, *de;
267 	int i, error;
268 
269 	dmp = VFSTODEVFS(vp->v_mount);
270 
271 	error = devfs_populate_vp(vp);
272 	if (error != 0)
273 		return (error);
274 
275 	i = *buflen;
276 	dd = vp->v_data;
277 
278 	if (vp->v_type == VCHR) {
279 		i -= strlen(dd->de_cdp->cdp_c.si_name);
280 		if (i < 0) {
281 			error = ENOMEM;
282 			goto finished;
283 		}
284 		bcopy(dd->de_cdp->cdp_c.si_name, buf + i,
285 		    strlen(dd->de_cdp->cdp_c.si_name));
286 		de = dd->de_dir;
287 	} else if (vp->v_type == VDIR) {
288 		if (dd == dmp->dm_rootdir) {
289 			*dvp = vp;
290 			vref(*dvp);
291 			goto finished;
292 		}
293 		i -= dd->de_dirent->d_namlen;
294 		if (i < 0) {
295 			error = ENOMEM;
296 			goto finished;
297 		}
298 		bcopy(dd->de_dirent->d_name, buf + i,
299 		    dd->de_dirent->d_namlen);
300 		de = dd;
301 	} else {
302 		error = ENOENT;
303 		goto finished;
304 	}
305 	*buflen = i;
306 	de = devfs_parent_dirent(de);
307 	if (de == NULL) {
308 		error = ENOENT;
309 		goto finished;
310 	}
311 	mtx_lock(&devfs_de_interlock);
312 	*dvp = de->de_vnode;
313 	if (*dvp != NULL) {
314 		VI_LOCK(*dvp);
315 		mtx_unlock(&devfs_de_interlock);
316 		vholdl(*dvp);
317 		VI_UNLOCK(*dvp);
318 		vref(*dvp);
319 		vdrop(*dvp);
320 	} else {
321 		mtx_unlock(&devfs_de_interlock);
322 		error = ENOENT;
323 	}
324 finished:
325 	sx_xunlock(&dmp->dm_lock);
326 	return (error);
327 }
328 
329 /*
330  * Construct the fully qualified path name relative to the mountpoint.
331  * If a NULL cnp is provided, no '/' is appended to the resulting path.
332  */
333 char *
334 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd,
335     struct componentname *cnp)
336 {
337 	int i;
338 	struct devfs_dirent *de;
339 
340 	sx_assert(&dmp->dm_lock, SA_LOCKED);
341 
342 	i = SPECNAMELEN;
343 	buf[i] = '\0';
344 	if (cnp != NULL)
345 		i -= cnp->cn_namelen;
346 	if (i < 0)
347 		 return (NULL);
348 	if (cnp != NULL)
349 		bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
350 	de = dd;
351 	while (de != dmp->dm_rootdir) {
352 		if (cnp != NULL || i < SPECNAMELEN) {
353 			i--;
354 			if (i < 0)
355 				 return (NULL);
356 			buf[i] = '/';
357 		}
358 		i -= de->de_dirent->d_namlen;
359 		if (i < 0)
360 			 return (NULL);
361 		bcopy(de->de_dirent->d_name, buf + i,
362 		    de->de_dirent->d_namlen);
363 		de = devfs_parent_dirent(de);
364 		if (de == NULL)
365 			return (NULL);
366 	}
367 	return (buf + i);
368 }
369 
370 static int
371 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
372 	struct devfs_dirent *de)
373 {
374 	int not_found;
375 
376 	not_found = 0;
377 	if (de->de_flags & DE_DOOMED)
378 		not_found = 1;
379 	if (DEVFS_DE_DROP(de)) {
380 		KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
381 		devfs_dirent_free(de);
382 	}
383 	if (DEVFS_DMP_DROP(dmp)) {
384 		KASSERT(not_found == 1,
385 			("DEVFS mount struct freed before dirent"));
386 		not_found = 2;
387 		sx_xunlock(&dmp->dm_lock);
388 		devfs_unmount_final(dmp);
389 	}
390 	if (not_found == 1 || (drop_dm_lock && not_found != 2))
391 		sx_unlock(&dmp->dm_lock);
392 	return (not_found);
393 }
394 
395 static void
396 devfs_insmntque_dtr(struct vnode *vp, void *arg)
397 {
398 	struct devfs_dirent *de;
399 
400 	de = (struct devfs_dirent *)arg;
401 	mtx_lock(&devfs_de_interlock);
402 	vp->v_data = NULL;
403 	de->de_vnode = NULL;
404 	mtx_unlock(&devfs_de_interlock);
405 	vgone(vp);
406 	vput(vp);
407 }
408 
409 /*
410  * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
411  * it on return.
412  */
413 int
414 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode,
415     struct vnode **vpp)
416 {
417 	int error;
418 	struct vnode *vp;
419 	struct cdev *dev;
420 	struct devfs_mount *dmp;
421 	struct cdevsw *dsw;
422 
423 	dmp = VFSTODEVFS(mp);
424 	if (de->de_flags & DE_DOOMED) {
425 		sx_xunlock(&dmp->dm_lock);
426 		return (ENOENT);
427 	}
428 loop:
429 	DEVFS_DE_HOLD(de);
430 	DEVFS_DMP_HOLD(dmp);
431 	mtx_lock(&devfs_de_interlock);
432 	vp = de->de_vnode;
433 	if (vp != NULL) {
434 		VI_LOCK(vp);
435 		mtx_unlock(&devfs_de_interlock);
436 		sx_xunlock(&dmp->dm_lock);
437 		vget(vp, lockmode | LK_INTERLOCK | LK_RETRY, curthread);
438 		sx_xlock(&dmp->dm_lock);
439 		if (devfs_allocv_drop_refs(0, dmp, de)) {
440 			vput(vp);
441 			return (ENOENT);
442 		}
443 		else if ((vp->v_iflag & VI_DOOMED) != 0) {
444 			mtx_lock(&devfs_de_interlock);
445 			if (de->de_vnode == vp) {
446 				de->de_vnode = NULL;
447 				vp->v_data = NULL;
448 			}
449 			mtx_unlock(&devfs_de_interlock);
450 			vput(vp);
451 			goto loop;
452 		}
453 		sx_xunlock(&dmp->dm_lock);
454 		*vpp = vp;
455 		return (0);
456 	}
457 	mtx_unlock(&devfs_de_interlock);
458 	if (de->de_dirent->d_type == DT_CHR) {
459 		if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
460 			devfs_allocv_drop_refs(1, dmp, de);
461 			return (ENOENT);
462 		}
463 		dev = &de->de_cdp->cdp_c;
464 	} else {
465 		dev = NULL;
466 	}
467 	error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
468 	if (error != 0) {
469 		devfs_allocv_drop_refs(1, dmp, de);
470 		printf("devfs_allocv: failed to allocate new vnode\n");
471 		return (error);
472 	}
473 
474 	if (de->de_dirent->d_type == DT_CHR) {
475 		vp->v_type = VCHR;
476 		VI_LOCK(vp);
477 		dev_lock();
478 		dev_refl(dev);
479 		/* XXX: v_rdev should be protect by vnode lock */
480 		vp->v_rdev = dev;
481 		KASSERT(vp->v_usecount == 1,
482 		    ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount));
483 		dev->si_usecount += vp->v_usecount;
484 		/* Special casing of ttys for deadfs.  Probably redundant. */
485 		dsw = dev->si_devsw;
486 		if (dsw != NULL && (dsw->d_flags & D_TTY) != 0)
487 			vp->v_vflag |= VV_ISTTY;
488 		dev_unlock();
489 		VI_UNLOCK(vp);
490 		if ((dev->si_flags & SI_ETERNAL) != 0)
491 			vp->v_vflag |= VV_ETERNALDEV;
492 		vp->v_op = &devfs_specops;
493 	} else if (de->de_dirent->d_type == DT_DIR) {
494 		vp->v_type = VDIR;
495 	} else if (de->de_dirent->d_type == DT_LNK) {
496 		vp->v_type = VLNK;
497 	} else {
498 		vp->v_type = VBAD;
499 	}
500 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
501 	VN_LOCK_ASHARE(vp);
502 	mtx_lock(&devfs_de_interlock);
503 	vp->v_data = de;
504 	de->de_vnode = vp;
505 	mtx_unlock(&devfs_de_interlock);
506 	error = insmntque1(vp, mp, devfs_insmntque_dtr, de);
507 	if (error != 0) {
508 		(void) devfs_allocv_drop_refs(1, dmp, de);
509 		return (error);
510 	}
511 	if (devfs_allocv_drop_refs(0, dmp, de)) {
512 		vput(vp);
513 		return (ENOENT);
514 	}
515 #ifdef MAC
516 	mac_devfs_vnode_associate(mp, de, vp);
517 #endif
518 	sx_xunlock(&dmp->dm_lock);
519 	*vpp = vp;
520 	return (0);
521 }
522 
523 static int
524 devfs_access(struct vop_access_args *ap)
525 {
526 	struct vnode *vp = ap->a_vp;
527 	struct devfs_dirent *de;
528 	struct proc *p;
529 	int error;
530 
531 	de = vp->v_data;
532 	if (vp->v_type == VDIR)
533 		de = de->de_dir;
534 
535 	error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
536 	    ap->a_accmode, ap->a_cred, NULL);
537 	if (error == 0)
538 		return (0);
539 	if (error != EACCES)
540 		return (error);
541 	p = ap->a_td->td_proc;
542 	/* We do, however, allow access to the controlling terminal */
543 	PROC_LOCK(p);
544 	if (!(p->p_flag & P_CONTROLT)) {
545 		PROC_UNLOCK(p);
546 		return (error);
547 	}
548 	if (p->p_session->s_ttydp == de->de_cdp)
549 		error = 0;
550 	PROC_UNLOCK(p);
551 	return (error);
552 }
553 
554 /* ARGSUSED */
555 static int
556 devfs_close(struct vop_close_args *ap)
557 {
558 	struct vnode *vp = ap->a_vp, *oldvp;
559 	struct thread *td = ap->a_td;
560 	struct proc *p;
561 	struct cdev *dev = vp->v_rdev;
562 	struct cdevsw *dsw;
563 	int vp_locked, error, ref;
564 
565 	/*
566 	 * XXX: Don't call d_close() if we were called because of
567 	 * XXX: insmntque1() failure.
568 	 */
569 	if (vp->v_data == NULL)
570 		return (0);
571 
572 	/*
573 	 * Hack: a tty device that is a controlling terminal
574 	 * has a reference from the session structure.
575 	 * We cannot easily tell that a character device is
576 	 * a controlling terminal, unless it is the closing
577 	 * process' controlling terminal.  In that case,
578 	 * if the reference count is 2 (this last descriptor
579 	 * plus the session), release the reference from the session.
580 	 */
581 	if (td != NULL) {
582 		p = td->td_proc;
583 		PROC_LOCK(p);
584 		if (vp == p->p_session->s_ttyvp) {
585 			PROC_UNLOCK(p);
586 			oldvp = NULL;
587 			sx_xlock(&proctree_lock);
588 			if (vp == p->p_session->s_ttyvp) {
589 				SESS_LOCK(p->p_session);
590 				VI_LOCK(vp);
591 				if (count_dev(dev) == 2 &&
592 				    (vp->v_iflag & VI_DOOMED) == 0) {
593 					p->p_session->s_ttyvp = NULL;
594 					p->p_session->s_ttydp = NULL;
595 					oldvp = vp;
596 				}
597 				VI_UNLOCK(vp);
598 				SESS_UNLOCK(p->p_session);
599 			}
600 			sx_xunlock(&proctree_lock);
601 			if (oldvp != NULL)
602 				vrele(oldvp);
603 		} else
604 			PROC_UNLOCK(p);
605 	}
606 	/*
607 	 * We do not want to really close the device if it
608 	 * is still in use unless we are trying to close it
609 	 * forcibly. Since every use (buffer, vnode, swap, cmap)
610 	 * holds a reference to the vnode, and because we mark
611 	 * any other vnodes that alias this device, when the
612 	 * sum of the reference counts on all the aliased
613 	 * vnodes descends to one, we are on last close.
614 	 */
615 	dsw = dev_refthread(dev, &ref);
616 	if (dsw == NULL)
617 		return (ENXIO);
618 	VI_LOCK(vp);
619 	if (vp->v_iflag & VI_DOOMED) {
620 		/* Forced close. */
621 	} else if (dsw->d_flags & D_TRACKCLOSE) {
622 		/* Keep device updated on status. */
623 	} else if (count_dev(dev) > 1) {
624 		VI_UNLOCK(vp);
625 		dev_relthread(dev, ref);
626 		return (0);
627 	}
628 	vholdl(vp);
629 	VI_UNLOCK(vp);
630 	vp_locked = VOP_ISLOCKED(vp);
631 	VOP_UNLOCK(vp, 0);
632 	KASSERT(dev->si_refcount > 0,
633 	    ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
634 	error = dsw->d_close(dev, ap->a_fflag, S_IFCHR, td);
635 	dev_relthread(dev, ref);
636 	vn_lock(vp, vp_locked | LK_RETRY);
637 	vdrop(vp);
638 	return (error);
639 }
640 
641 static int
642 devfs_close_f(struct file *fp, struct thread *td)
643 {
644 	int error;
645 	struct file *fpop;
646 
647 	/*
648 	 * NB: td may be NULL if this descriptor is closed due to
649 	 * garbage collection from a closed UNIX domain socket.
650 	 */
651 	fpop = curthread->td_fpop;
652 	curthread->td_fpop = fp;
653 	error = vnops.fo_close(fp, td);
654 	curthread->td_fpop = fpop;
655 
656 	/*
657 	 * The f_cdevpriv cannot be assigned non-NULL value while we
658 	 * are destroying the file.
659 	 */
660 	if (fp->f_cdevpriv != NULL)
661 		devfs_fpdrop(fp);
662 	return (error);
663 }
664 
665 static int
666 devfs_fsync(struct vop_fsync_args *ap)
667 {
668 	int error;
669 	struct bufobj *bo;
670 	struct devfs_dirent *de;
671 
672 	if (!vn_isdisk(ap->a_vp, &error)) {
673 		bo = &ap->a_vp->v_bufobj;
674 		de = ap->a_vp->v_data;
675 		if (error == ENXIO && bo->bo_dirty.bv_cnt > 0) {
676 			printf("Device %s went missing before all of the data "
677 			    "could be written to it; expect data loss.\n",
678 			    de->de_dirent->d_name);
679 
680 			error = vop_stdfsync(ap);
681 			if (bo->bo_dirty.bv_cnt != 0 || error != 0)
682 				panic("devfs_fsync: vop_stdfsync failed.");
683 		}
684 
685 		return (0);
686 	}
687 
688 	return (vop_stdfsync(ap));
689 }
690 
691 static int
692 devfs_getattr(struct vop_getattr_args *ap)
693 {
694 	struct vnode *vp = ap->a_vp;
695 	struct vattr *vap = ap->a_vap;
696 	int error;
697 	struct devfs_dirent *de;
698 	struct devfs_mount *dmp;
699 	struct cdev *dev;
700 
701 	error = devfs_populate_vp(vp);
702 	if (error != 0)
703 		return (error);
704 
705 	dmp = VFSTODEVFS(vp->v_mount);
706 	sx_xunlock(&dmp->dm_lock);
707 
708 	de = vp->v_data;
709 	KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
710 	if (vp->v_type == VDIR) {
711 		de = de->de_dir;
712 		KASSERT(de != NULL,
713 		    ("Null dir dirent in devfs_getattr vp=%p", vp));
714 	}
715 	vap->va_uid = de->de_uid;
716 	vap->va_gid = de->de_gid;
717 	vap->va_mode = de->de_mode;
718 	if (vp->v_type == VLNK)
719 		vap->va_size = strlen(de->de_symlink);
720 	else if (vp->v_type == VDIR)
721 		vap->va_size = vap->va_bytes = DEV_BSIZE;
722 	else
723 		vap->va_size = 0;
724 	if (vp->v_type != VDIR)
725 		vap->va_bytes = 0;
726 	vap->va_blocksize = DEV_BSIZE;
727 	vap->va_type = vp->v_type;
728 
729 #define fix(aa)							\
730 	do {							\
731 		if ((aa).tv_sec <= 3600) {			\
732 			(aa).tv_sec = boottime.tv_sec;		\
733 			(aa).tv_nsec = boottime.tv_usec * 1000; \
734 		}						\
735 	} while (0)
736 
737 	if (vp->v_type != VCHR)  {
738 		fix(de->de_atime);
739 		vap->va_atime = de->de_atime;
740 		fix(de->de_mtime);
741 		vap->va_mtime = de->de_mtime;
742 		fix(de->de_ctime);
743 		vap->va_ctime = de->de_ctime;
744 	} else {
745 		dev = vp->v_rdev;
746 		fix(dev->si_atime);
747 		vap->va_atime = dev->si_atime;
748 		fix(dev->si_mtime);
749 		vap->va_mtime = dev->si_mtime;
750 		fix(dev->si_ctime);
751 		vap->va_ctime = dev->si_ctime;
752 
753 		vap->va_rdev = cdev2priv(dev)->cdp_inode;
754 	}
755 	vap->va_gen = 0;
756 	vap->va_flags = 0;
757 	vap->va_filerev = 0;
758 	vap->va_nlink = de->de_links;
759 	vap->va_fileid = de->de_inode;
760 
761 	return (error);
762 }
763 
764 /* ARGSUSED */
765 static int
766 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
767 {
768 	struct cdev *dev;
769 	struct cdevsw *dsw;
770 	struct vnode *vp;
771 	struct vnode *vpold;
772 	int error, i, ref;
773 	const char *p;
774 	struct fiodgname_arg *fgn;
775 	struct file *fpop;
776 
777 	fpop = td->td_fpop;
778 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
779 	if (error != 0) {
780 		error = vnops.fo_ioctl(fp, com, data, cred, td);
781 		return (error);
782 	}
783 
784 	if (com == FIODTYPE) {
785 		*(int *)data = dsw->d_flags & D_TYPEMASK;
786 		td->td_fpop = fpop;
787 		dev_relthread(dev, ref);
788 		return (0);
789 	} else if (com == FIODGNAME) {
790 		fgn = data;
791 		p = devtoname(dev);
792 		i = strlen(p) + 1;
793 		if (i > fgn->len)
794 			error = EINVAL;
795 		else
796 			error = copyout(p, fgn->buf, i);
797 		td->td_fpop = fpop;
798 		dev_relthread(dev, ref);
799 		return (error);
800 	}
801 	error = dsw->d_ioctl(dev, com, data, fp->f_flag, td);
802 	td->td_fpop = NULL;
803 	dev_relthread(dev, ref);
804 	if (error == ENOIOCTL)
805 		error = ENOTTY;
806 	if (error == 0 && com == TIOCSCTTY) {
807 		vp = fp->f_vnode;
808 
809 		/* Do nothing if reassigning same control tty */
810 		sx_slock(&proctree_lock);
811 		if (td->td_proc->p_session->s_ttyvp == vp) {
812 			sx_sunlock(&proctree_lock);
813 			return (0);
814 		}
815 
816 		vpold = td->td_proc->p_session->s_ttyvp;
817 		VREF(vp);
818 		SESS_LOCK(td->td_proc->p_session);
819 		td->td_proc->p_session->s_ttyvp = vp;
820 		td->td_proc->p_session->s_ttydp = cdev2priv(dev);
821 		SESS_UNLOCK(td->td_proc->p_session);
822 
823 		sx_sunlock(&proctree_lock);
824 
825 		/* Get rid of reference to old control tty */
826 		if (vpold)
827 			vrele(vpold);
828 	}
829 	return (error);
830 }
831 
832 /* ARGSUSED */
833 static int
834 devfs_kqfilter_f(struct file *fp, struct knote *kn)
835 {
836 	struct cdev *dev;
837 	struct cdevsw *dsw;
838 	int error, ref;
839 	struct file *fpop;
840 	struct thread *td;
841 
842 	td = curthread;
843 	fpop = td->td_fpop;
844 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
845 	if (error)
846 		return (error);
847 	error = dsw->d_kqfilter(dev, kn);
848 	td->td_fpop = fpop;
849 	dev_relthread(dev, ref);
850 	return (error);
851 }
852 
853 static inline int
854 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
855 {
856 	struct cdev_priv *cdp;
857 	struct ucred *dcr;
858 	struct proc *p;
859 	int error;
860 
861 	cdp = de->de_cdp;
862 	if (cdp == NULL)
863 		return (0);
864 	dcr = cdp->cdp_c.si_cred;
865 	if (dcr == NULL)
866 		return (0);
867 
868 	error = prison_check(td->td_ucred, dcr);
869 	if (error == 0)
870 		return (0);
871 	/* We do, however, allow access to the controlling terminal */
872 	p = td->td_proc;
873 	PROC_LOCK(p);
874 	if (!(p->p_flag & P_CONTROLT)) {
875 		PROC_UNLOCK(p);
876 		return (error);
877 	}
878 	if (p->p_session->s_ttydp == cdp)
879 		error = 0;
880 	PROC_UNLOCK(p);
881 	return (error);
882 }
883 
884 static int
885 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
886 {
887 	struct componentname *cnp;
888 	struct vnode *dvp, **vpp;
889 	struct thread *td;
890 	struct devfs_dirent *de, *dd;
891 	struct devfs_dirent **dde;
892 	struct devfs_mount *dmp;
893 	struct cdev *cdev;
894 	int error, flags, nameiop, dvplocked;
895 	char specname[SPECNAMELEN + 1], *pname;
896 
897 	cnp = ap->a_cnp;
898 	vpp = ap->a_vpp;
899 	dvp = ap->a_dvp;
900 	pname = cnp->cn_nameptr;
901 	td = cnp->cn_thread;
902 	flags = cnp->cn_flags;
903 	nameiop = cnp->cn_nameiop;
904 	dmp = VFSTODEVFS(dvp->v_mount);
905 	dd = dvp->v_data;
906 	*vpp = NULLVP;
907 
908 	if ((flags & ISLASTCN) && nameiop == RENAME)
909 		return (EOPNOTSUPP);
910 
911 	if (dvp->v_type != VDIR)
912 		return (ENOTDIR);
913 
914 	if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
915 		return (EIO);
916 
917 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
918 	if (error)
919 		return (error);
920 
921 	if (cnp->cn_namelen == 1 && *pname == '.') {
922 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
923 			return (EINVAL);
924 		*vpp = dvp;
925 		VREF(dvp);
926 		return (0);
927 	}
928 
929 	if (flags & ISDOTDOT) {
930 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
931 			return (EINVAL);
932 		de = devfs_parent_dirent(dd);
933 		if (de == NULL)
934 			return (ENOENT);
935 		dvplocked = VOP_ISLOCKED(dvp);
936 		VOP_UNLOCK(dvp, 0);
937 		error = devfs_allocv(de, dvp->v_mount,
938 		    cnp->cn_lkflags & LK_TYPE_MASK, vpp);
939 		*dm_unlock = 0;
940 		vn_lock(dvp, dvplocked | LK_RETRY);
941 		return (error);
942 	}
943 
944 	dd = dvp->v_data;
945 	de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
946 	while (de == NULL) {	/* While(...) so we can use break */
947 
948 		if (nameiop == DELETE)
949 			return (ENOENT);
950 
951 		/*
952 		 * OK, we didn't have an entry for the name we were asked for
953 		 * so we try to see if anybody can create it on demand.
954 		 */
955 		pname = devfs_fqpn(specname, dmp, dd, cnp);
956 		if (pname == NULL)
957 			break;
958 
959 		cdev = NULL;
960 		DEVFS_DMP_HOLD(dmp);
961 		sx_xunlock(&dmp->dm_lock);
962 		sx_slock(&clone_drain_lock);
963 		EVENTHANDLER_INVOKE(dev_clone,
964 		    td->td_ucred, pname, strlen(pname), &cdev);
965 		sx_sunlock(&clone_drain_lock);
966 
967 		if (cdev == NULL)
968 			sx_xlock(&dmp->dm_lock);
969 		else if (devfs_populate_vp(dvp) != 0) {
970 			*dm_unlock = 0;
971 			sx_xlock(&dmp->dm_lock);
972 			if (DEVFS_DMP_DROP(dmp)) {
973 				sx_xunlock(&dmp->dm_lock);
974 				devfs_unmount_final(dmp);
975 			} else
976 				sx_xunlock(&dmp->dm_lock);
977 			dev_rel(cdev);
978 			return (ENOENT);
979 		}
980 		if (DEVFS_DMP_DROP(dmp)) {
981 			*dm_unlock = 0;
982 			sx_xunlock(&dmp->dm_lock);
983 			devfs_unmount_final(dmp);
984 			if (cdev != NULL)
985 				dev_rel(cdev);
986 			return (ENOENT);
987 		}
988 
989 		if (cdev == NULL)
990 			break;
991 
992 		dev_lock();
993 		dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
994 		if (dde != NULL && *dde != NULL)
995 			de = *dde;
996 		dev_unlock();
997 		dev_rel(cdev);
998 		break;
999 	}
1000 
1001 	if (de == NULL || de->de_flags & DE_WHITEOUT) {
1002 		if ((nameiop == CREATE || nameiop == RENAME) &&
1003 		    (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
1004 			cnp->cn_flags |= SAVENAME;
1005 			return (EJUSTRETURN);
1006 		}
1007 		return (ENOENT);
1008 	}
1009 
1010 	if (devfs_prison_check(de, td))
1011 		return (ENOENT);
1012 
1013 	if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
1014 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
1015 		if (error)
1016 			return (error);
1017 		if (*vpp == dvp) {
1018 			VREF(dvp);
1019 			*vpp = dvp;
1020 			return (0);
1021 		}
1022 	}
1023 	error = devfs_allocv(de, dvp->v_mount, cnp->cn_lkflags & LK_TYPE_MASK,
1024 	    vpp);
1025 	*dm_unlock = 0;
1026 	return (error);
1027 }
1028 
1029 static int
1030 devfs_lookup(struct vop_lookup_args *ap)
1031 {
1032 	int j;
1033 	struct devfs_mount *dmp;
1034 	int dm_unlock;
1035 
1036 	if (devfs_populate_vp(ap->a_dvp) != 0)
1037 		return (ENOTDIR);
1038 
1039 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1040 	dm_unlock = 1;
1041 	j = devfs_lookupx(ap, &dm_unlock);
1042 	if (dm_unlock == 1)
1043 		sx_xunlock(&dmp->dm_lock);
1044 	return (j);
1045 }
1046 
1047 static int
1048 devfs_mknod(struct vop_mknod_args *ap)
1049 {
1050 	struct componentname *cnp;
1051 	struct vnode *dvp, **vpp;
1052 	struct devfs_dirent *dd, *de;
1053 	struct devfs_mount *dmp;
1054 	int error;
1055 
1056 	/*
1057 	 * The only type of node we should be creating here is a
1058 	 * character device, for anything else return EOPNOTSUPP.
1059 	 */
1060 	if (ap->a_vap->va_type != VCHR)
1061 		return (EOPNOTSUPP);
1062 	dvp = ap->a_dvp;
1063 	dmp = VFSTODEVFS(dvp->v_mount);
1064 
1065 	cnp = ap->a_cnp;
1066 	vpp = ap->a_vpp;
1067 	dd = dvp->v_data;
1068 
1069 	error = ENOENT;
1070 	sx_xlock(&dmp->dm_lock);
1071 	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1072 		if (cnp->cn_namelen != de->de_dirent->d_namlen)
1073 			continue;
1074 		if (de->de_dirent->d_type == DT_CHR &&
1075 		    (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0)
1076 			continue;
1077 		if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1078 		    de->de_dirent->d_namlen) != 0)
1079 			continue;
1080 		if (de->de_flags & DE_WHITEOUT)
1081 			break;
1082 		goto notfound;
1083 	}
1084 	if (de == NULL)
1085 		goto notfound;
1086 	de->de_flags &= ~DE_WHITEOUT;
1087 	error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1088 	return (error);
1089 notfound:
1090 	sx_xunlock(&dmp->dm_lock);
1091 	return (error);
1092 }
1093 
1094 /* ARGSUSED */
1095 static int
1096 devfs_open(struct vop_open_args *ap)
1097 {
1098 	struct thread *td = ap->a_td;
1099 	struct vnode *vp = ap->a_vp;
1100 	struct cdev *dev = vp->v_rdev;
1101 	struct file *fp = ap->a_fp;
1102 	int error, ref, vlocked;
1103 	struct cdevsw *dsw;
1104 	struct file *fpop;
1105 	struct mtx *mtxp;
1106 
1107 	if (vp->v_type == VBLK)
1108 		return (ENXIO);
1109 
1110 	if (dev == NULL)
1111 		return (ENXIO);
1112 
1113 	/* Make this field valid before any I/O in d_open. */
1114 	if (dev->si_iosize_max == 0)
1115 		dev->si_iosize_max = DFLTPHYS;
1116 
1117 	dsw = dev_refthread(dev, &ref);
1118 	if (dsw == NULL)
1119 		return (ENXIO);
1120 	if (fp == NULL && dsw->d_fdopen != NULL) {
1121 		dev_relthread(dev, ref);
1122 		return (ENXIO);
1123 	}
1124 
1125 	vlocked = VOP_ISLOCKED(vp);
1126 	VOP_UNLOCK(vp, 0);
1127 
1128 	fpop = td->td_fpop;
1129 	td->td_fpop = fp;
1130 	if (fp != NULL) {
1131 		fp->f_data = dev;
1132 		fp->f_vnode = vp;
1133 	}
1134 	if (dsw->d_fdopen != NULL)
1135 		error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1136 	else
1137 		error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1138 	/* cleanup any cdevpriv upon error */
1139 	if (error != 0)
1140 		devfs_clear_cdevpriv();
1141 	td->td_fpop = fpop;
1142 
1143 	vn_lock(vp, vlocked | LK_RETRY);
1144 	dev_relthread(dev, ref);
1145 	if (error != 0) {
1146 		if (error == ERESTART)
1147 			error = EINTR;
1148 		return (error);
1149 	}
1150 
1151 #if 0	/* /dev/console */
1152 	KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1153 #else
1154 	if (fp == NULL)
1155 		return (error);
1156 #endif
1157 	if (fp->f_ops == &badfileops)
1158 		finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1159 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
1160 
1161 	/*
1162 	 * Hint to the dofilewrite() to not force the buffer draining
1163 	 * on the writer to the file.  Most likely, the write would
1164 	 * not need normal buffers.
1165 	 */
1166 	mtx_lock(mtxp);
1167 	fp->f_vnread_flags |= FDEVFS_VNODE;
1168 	mtx_unlock(mtxp);
1169 	return (error);
1170 }
1171 
1172 static int
1173 devfs_pathconf(struct vop_pathconf_args *ap)
1174 {
1175 
1176 	switch (ap->a_name) {
1177 	case _PC_MAC_PRESENT:
1178 #ifdef MAC
1179 		/*
1180 		 * If MAC is enabled, devfs automatically supports
1181 		 * trivial non-persistant label storage.
1182 		 */
1183 		*ap->a_retval = 1;
1184 #else
1185 		*ap->a_retval = 0;
1186 #endif
1187 		return (0);
1188 	default:
1189 		return (vop_stdpathconf(ap));
1190 	}
1191 	/* NOTREACHED */
1192 }
1193 
1194 /* ARGSUSED */
1195 static int
1196 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1197 {
1198 	struct cdev *dev;
1199 	struct cdevsw *dsw;
1200 	int error, ref;
1201 	struct file *fpop;
1202 
1203 	fpop = td->td_fpop;
1204 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1205 	if (error != 0) {
1206 		error = vnops.fo_poll(fp, events, cred, td);
1207 		return (error);
1208 	}
1209 	error = dsw->d_poll(dev, events, td);
1210 	td->td_fpop = fpop;
1211 	dev_relthread(dev, ref);
1212 	return(error);
1213 }
1214 
1215 /*
1216  * Print out the contents of a special device vnode.
1217  */
1218 static int
1219 devfs_print(struct vop_print_args *ap)
1220 {
1221 
1222 	printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1223 	return (0);
1224 }
1225 
1226 static int
1227 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
1228     int flags, struct thread *td)
1229 {
1230 	struct cdev *dev;
1231 	int ioflag, error, ref;
1232 	ssize_t resid;
1233 	struct cdevsw *dsw;
1234 	struct file *fpop;
1235 
1236 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1237 		return (EINVAL);
1238 	fpop = td->td_fpop;
1239 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1240 	if (error != 0) {
1241 		error = vnops.fo_read(fp, uio, cred, flags, td);
1242 		return (error);
1243 	}
1244 	resid = uio->uio_resid;
1245 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1246 	if (ioflag & O_DIRECT)
1247 		ioflag |= IO_DIRECT;
1248 
1249 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1250 	error = dsw->d_read(dev, uio, ioflag);
1251 	if (uio->uio_resid != resid || (error == 0 && resid != 0))
1252 		devfs_timestamp(&dev->si_atime);
1253 	td->td_fpop = fpop;
1254 	dev_relthread(dev, ref);
1255 
1256 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1257 	return (error);
1258 }
1259 
1260 static int
1261 devfs_readdir(struct vop_readdir_args *ap)
1262 {
1263 	int error;
1264 	struct uio *uio;
1265 	struct dirent *dp;
1266 	struct devfs_dirent *dd;
1267 	struct devfs_dirent *de;
1268 	struct devfs_mount *dmp;
1269 	off_t off;
1270 	int *tmp_ncookies = NULL;
1271 
1272 	if (ap->a_vp->v_type != VDIR)
1273 		return (ENOTDIR);
1274 
1275 	uio = ap->a_uio;
1276 	if (uio->uio_offset < 0)
1277 		return (EINVAL);
1278 
1279 	/*
1280 	 * XXX: This is a temporary hack to get around this filesystem not
1281 	 * supporting cookies. We store the location of the ncookies pointer
1282 	 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1283 	 * and set the number of cookies to 0. We then set the pointer to
1284 	 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1285 	 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1286 	 * pointer to its original location before returning to the caller.
1287 	 */
1288 	if (ap->a_ncookies != NULL) {
1289 		tmp_ncookies = ap->a_ncookies;
1290 		*ap->a_ncookies = 0;
1291 		ap->a_ncookies = NULL;
1292 	}
1293 
1294 	dmp = VFSTODEVFS(ap->a_vp->v_mount);
1295 	if (devfs_populate_vp(ap->a_vp) != 0) {
1296 		if (tmp_ncookies != NULL)
1297 			ap->a_ncookies = tmp_ncookies;
1298 		return (EIO);
1299 	}
1300 	error = 0;
1301 	de = ap->a_vp->v_data;
1302 	off = 0;
1303 	TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1304 		KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1305 		if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1306 			continue;
1307 		if (devfs_prison_check(dd, uio->uio_td))
1308 			continue;
1309 		if (dd->de_dirent->d_type == DT_DIR)
1310 			de = dd->de_dir;
1311 		else
1312 			de = dd;
1313 		dp = dd->de_dirent;
1314 		if (dp->d_reclen > uio->uio_resid)
1315 			break;
1316 		dp->d_fileno = de->de_inode;
1317 		if (off >= uio->uio_offset) {
1318 			error = vfs_read_dirent(ap, dp, off);
1319 			if (error)
1320 				break;
1321 		}
1322 		off += dp->d_reclen;
1323 	}
1324 	sx_xunlock(&dmp->dm_lock);
1325 	uio->uio_offset = off;
1326 
1327 	/*
1328 	 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1329 	 * place.
1330 	 */
1331 	if (tmp_ncookies != NULL)
1332 		ap->a_ncookies = tmp_ncookies;
1333 
1334 	return (error);
1335 }
1336 
1337 static int
1338 devfs_readlink(struct vop_readlink_args *ap)
1339 {
1340 	struct devfs_dirent *de;
1341 
1342 	de = ap->a_vp->v_data;
1343 	return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1344 }
1345 
1346 static int
1347 devfs_reclaim(struct vop_reclaim_args *ap)
1348 {
1349 	struct vnode *vp = ap->a_vp;
1350 	struct devfs_dirent *de;
1351 	struct cdev *dev;
1352 
1353 	mtx_lock(&devfs_de_interlock);
1354 	de = vp->v_data;
1355 	if (de != NULL) {
1356 		de->de_vnode = NULL;
1357 		vp->v_data = NULL;
1358 	}
1359 	mtx_unlock(&devfs_de_interlock);
1360 
1361 	vnode_destroy_vobject(vp);
1362 
1363 	VI_LOCK(vp);
1364 	dev_lock();
1365 	dev = vp->v_rdev;
1366 	vp->v_rdev = NULL;
1367 
1368 	if (dev == NULL) {
1369 		dev_unlock();
1370 		VI_UNLOCK(vp);
1371 		return (0);
1372 	}
1373 
1374 	dev->si_usecount -= vp->v_usecount;
1375 	dev_unlock();
1376 	VI_UNLOCK(vp);
1377 	dev_rel(dev);
1378 	return (0);
1379 }
1380 
1381 static int
1382 devfs_remove(struct vop_remove_args *ap)
1383 {
1384 	struct vnode *dvp = ap->a_dvp;
1385 	struct vnode *vp = ap->a_vp;
1386 	struct devfs_dirent *dd;
1387 	struct devfs_dirent *de, *de_covered;
1388 	struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1389 
1390 	ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1391 	ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1392 
1393 	sx_xlock(&dmp->dm_lock);
1394 	dd = ap->a_dvp->v_data;
1395 	de = vp->v_data;
1396 	if (de->de_cdp == NULL) {
1397 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1398 		if (de->de_dirent->d_type == DT_LNK) {
1399 			de_covered = devfs_find(dd, de->de_dirent->d_name,
1400 			    de->de_dirent->d_namlen, 0);
1401 			if (de_covered != NULL)
1402 				de_covered->de_flags &= ~DE_COVERED;
1403 		}
1404 		/* We need to unlock dvp because devfs_delete() may lock it. */
1405 		VOP_UNLOCK(vp, 0);
1406 		if (dvp != vp)
1407 			VOP_UNLOCK(dvp, 0);
1408 		devfs_delete(dmp, de, 0);
1409 		sx_xunlock(&dmp->dm_lock);
1410 		if (dvp != vp)
1411 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1412 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1413 	} else {
1414 		de->de_flags |= DE_WHITEOUT;
1415 		sx_xunlock(&dmp->dm_lock);
1416 	}
1417 	return (0);
1418 }
1419 
1420 /*
1421  * Revoke is called on a tty when a terminal session ends.  The vnode
1422  * is orphaned by setting v_op to deadfs so we need to let go of it
1423  * as well so that we create a new one next time around.
1424  *
1425  */
1426 static int
1427 devfs_revoke(struct vop_revoke_args *ap)
1428 {
1429 	struct vnode *vp = ap->a_vp, *vp2;
1430 	struct cdev *dev;
1431 	struct cdev_priv *cdp;
1432 	struct devfs_dirent *de;
1433 	int i;
1434 
1435 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1436 
1437 	dev = vp->v_rdev;
1438 	cdp = cdev2priv(dev);
1439 
1440 	dev_lock();
1441 	cdp->cdp_inuse++;
1442 	dev_unlock();
1443 
1444 	vhold(vp);
1445 	vgone(vp);
1446 	vdrop(vp);
1447 
1448 	VOP_UNLOCK(vp,0);
1449  loop:
1450 	for (;;) {
1451 		mtx_lock(&devfs_de_interlock);
1452 		dev_lock();
1453 		vp2 = NULL;
1454 		for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1455 			de = cdp->cdp_dirents[i];
1456 			if (de == NULL)
1457 				continue;
1458 
1459 			vp2 = de->de_vnode;
1460 			if (vp2 != NULL) {
1461 				dev_unlock();
1462 				VI_LOCK(vp2);
1463 				mtx_unlock(&devfs_de_interlock);
1464 				if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
1465 				    curthread))
1466 					goto loop;
1467 				vhold(vp2);
1468 				vgone(vp2);
1469 				vdrop(vp2);
1470 				vput(vp2);
1471 				break;
1472 			}
1473 		}
1474 		if (vp2 != NULL) {
1475 			continue;
1476 		}
1477 		dev_unlock();
1478 		mtx_unlock(&devfs_de_interlock);
1479 		break;
1480 	}
1481 	dev_lock();
1482 	cdp->cdp_inuse--;
1483 	if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1484 		TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1485 		dev_unlock();
1486 		dev_rel(&cdp->cdp_c);
1487 	} else
1488 		dev_unlock();
1489 
1490 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1491 	return (0);
1492 }
1493 
1494 static int
1495 devfs_rioctl(struct vop_ioctl_args *ap)
1496 {
1497 	struct vnode *vp;
1498 	struct devfs_mount *dmp;
1499 	int error;
1500 
1501 	vp = ap->a_vp;
1502 	vn_lock(vp, LK_SHARED | LK_RETRY);
1503 	if (vp->v_iflag & VI_DOOMED) {
1504 		VOP_UNLOCK(vp, 0);
1505 		return (EBADF);
1506 	}
1507 	dmp = VFSTODEVFS(vp->v_mount);
1508 	sx_xlock(&dmp->dm_lock);
1509 	VOP_UNLOCK(vp, 0);
1510 	DEVFS_DMP_HOLD(dmp);
1511 	devfs_populate(dmp);
1512 	if (DEVFS_DMP_DROP(dmp)) {
1513 		sx_xunlock(&dmp->dm_lock);
1514 		devfs_unmount_final(dmp);
1515 		return (ENOENT);
1516 	}
1517 	error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1518 	sx_xunlock(&dmp->dm_lock);
1519 	return (error);
1520 }
1521 
1522 static int
1523 devfs_rread(struct vop_read_args *ap)
1524 {
1525 
1526 	if (ap->a_vp->v_type != VDIR)
1527 		return (EINVAL);
1528 	return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1529 }
1530 
1531 static int
1532 devfs_setattr(struct vop_setattr_args *ap)
1533 {
1534 	struct devfs_dirent *de;
1535 	struct vattr *vap;
1536 	struct vnode *vp;
1537 	struct thread *td;
1538 	int c, error;
1539 	uid_t uid;
1540 	gid_t gid;
1541 
1542 	vap = ap->a_vap;
1543 	vp = ap->a_vp;
1544 	td = curthread;
1545 	if ((vap->va_type != VNON) ||
1546 	    (vap->va_nlink != VNOVAL) ||
1547 	    (vap->va_fsid != VNOVAL) ||
1548 	    (vap->va_fileid != VNOVAL) ||
1549 	    (vap->va_blocksize != VNOVAL) ||
1550 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1551 	    (vap->va_rdev != VNOVAL) ||
1552 	    ((int)vap->va_bytes != VNOVAL) ||
1553 	    (vap->va_gen != VNOVAL)) {
1554 		return (EINVAL);
1555 	}
1556 
1557 	de = vp->v_data;
1558 	if (vp->v_type == VDIR)
1559 		de = de->de_dir;
1560 
1561 	error = c = 0;
1562 	if (vap->va_uid == (uid_t)VNOVAL)
1563 		uid = de->de_uid;
1564 	else
1565 		uid = vap->va_uid;
1566 	if (vap->va_gid == (gid_t)VNOVAL)
1567 		gid = de->de_gid;
1568 	else
1569 		gid = vap->va_gid;
1570 	if (uid != de->de_uid || gid != de->de_gid) {
1571 		if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1572 		    (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1573 			error = priv_check(td, PRIV_VFS_CHOWN);
1574 			if (error)
1575 				return (error);
1576 		}
1577 		de->de_uid = uid;
1578 		de->de_gid = gid;
1579 		c = 1;
1580 	}
1581 
1582 	if (vap->va_mode != (mode_t)VNOVAL) {
1583 		if (ap->a_cred->cr_uid != de->de_uid) {
1584 			error = priv_check(td, PRIV_VFS_ADMIN);
1585 			if (error)
1586 				return (error);
1587 		}
1588 		de->de_mode = vap->va_mode;
1589 		c = 1;
1590 	}
1591 
1592 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1593 		error = vn_utimes_perm(vp, vap, ap->a_cred, td);
1594 		if (error != 0)
1595 			return (error);
1596 		if (vap->va_atime.tv_sec != VNOVAL) {
1597 			if (vp->v_type == VCHR)
1598 				vp->v_rdev->si_atime = vap->va_atime;
1599 			else
1600 				de->de_atime = vap->va_atime;
1601 		}
1602 		if (vap->va_mtime.tv_sec != VNOVAL) {
1603 			if (vp->v_type == VCHR)
1604 				vp->v_rdev->si_mtime = vap->va_mtime;
1605 			else
1606 				de->de_mtime = vap->va_mtime;
1607 		}
1608 		c = 1;
1609 	}
1610 
1611 	if (c) {
1612 		if (vp->v_type == VCHR)
1613 			vfs_timestamp(&vp->v_rdev->si_ctime);
1614 		else
1615 			vfs_timestamp(&de->de_mtime);
1616 	}
1617 	return (0);
1618 }
1619 
1620 #ifdef MAC
1621 static int
1622 devfs_setlabel(struct vop_setlabel_args *ap)
1623 {
1624 	struct vnode *vp;
1625 	struct devfs_dirent *de;
1626 
1627 	vp = ap->a_vp;
1628 	de = vp->v_data;
1629 
1630 	mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1631 	mac_devfs_update(vp->v_mount, de, vp);
1632 
1633 	return (0);
1634 }
1635 #endif
1636 
1637 static int
1638 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1639 {
1640 
1641 	return (vnops.fo_stat(fp, sb, cred, td));
1642 }
1643 
1644 static int
1645 devfs_symlink(struct vop_symlink_args *ap)
1646 {
1647 	int i, error;
1648 	struct devfs_dirent *dd;
1649 	struct devfs_dirent *de, *de_covered, *de_dotdot;
1650 	struct devfs_mount *dmp;
1651 
1652 	error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1653 	if (error)
1654 		return(error);
1655 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1656 	if (devfs_populate_vp(ap->a_dvp) != 0)
1657 		return (ENOENT);
1658 
1659 	dd = ap->a_dvp->v_data;
1660 	de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1661 	de->de_flags = DE_USER;
1662 	de->de_uid = 0;
1663 	de->de_gid = 0;
1664 	de->de_mode = 0755;
1665 	de->de_inode = alloc_unr(devfs_inos);
1666 	de->de_dir = dd;
1667 	de->de_dirent->d_type = DT_LNK;
1668 	i = strlen(ap->a_target) + 1;
1669 	de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1670 	bcopy(ap->a_target, de->de_symlink, i);
1671 #ifdef MAC
1672 	mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1673 #endif
1674 	de_covered = devfs_find(dd, de->de_dirent->d_name,
1675 	    de->de_dirent->d_namlen, 0);
1676 	if (de_covered != NULL) {
1677 		if ((de_covered->de_flags & DE_USER) != 0) {
1678 			devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1679 			sx_xunlock(&dmp->dm_lock);
1680 			return (EEXIST);
1681 		}
1682 		KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1683 		    ("devfs_symlink: entry %p already covered", de_covered));
1684 		de_covered->de_flags |= DE_COVERED;
1685 	}
1686 
1687 	de_dotdot = TAILQ_FIRST(&dd->de_dlist);		/* "." */
1688 	de_dotdot = TAILQ_NEXT(de_dotdot, de_list);	/* ".." */
1689 	TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1690 	devfs_dir_ref_de(dmp, dd);
1691 	devfs_rules_apply(dmp, de);
1692 
1693 	return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1694 }
1695 
1696 static int
1697 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1698 {
1699 
1700 	return (vnops.fo_truncate(fp, length, cred, td));
1701 }
1702 
1703 static int
1704 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
1705     int flags, struct thread *td)
1706 {
1707 	struct cdev *dev;
1708 	int error, ioflag, ref;
1709 	ssize_t resid;
1710 	struct cdevsw *dsw;
1711 	struct file *fpop;
1712 
1713 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1714 		return (EINVAL);
1715 	fpop = td->td_fpop;
1716 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1717 	if (error != 0) {
1718 		error = vnops.fo_write(fp, uio, cred, flags, td);
1719 		return (error);
1720 	}
1721 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1722 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1723 	if (ioflag & O_DIRECT)
1724 		ioflag |= IO_DIRECT;
1725 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1726 
1727 	resid = uio->uio_resid;
1728 
1729 	error = dsw->d_write(dev, uio, ioflag);
1730 	if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1731 		devfs_timestamp(&dev->si_ctime);
1732 		dev->si_mtime = dev->si_ctime;
1733 	}
1734 	td->td_fpop = fpop;
1735 	dev_relthread(dev, ref);
1736 
1737 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1738 	return (error);
1739 }
1740 
1741 dev_t
1742 dev2udev(struct cdev *x)
1743 {
1744 	if (x == NULL)
1745 		return (NODEV);
1746 	return (cdev2priv(x)->cdp_inode);
1747 }
1748 
1749 static struct fileops devfs_ops_f = {
1750 	.fo_read =	devfs_read_f,
1751 	.fo_write =	devfs_write_f,
1752 	.fo_truncate =	devfs_truncate_f,
1753 	.fo_ioctl =	devfs_ioctl_f,
1754 	.fo_poll =	devfs_poll_f,
1755 	.fo_kqfilter =	devfs_kqfilter_f,
1756 	.fo_stat =	devfs_stat_f,
1757 	.fo_close =	devfs_close_f,
1758 	.fo_chmod =	vn_chmod,
1759 	.fo_chown =	vn_chown,
1760 	.fo_sendfile =	vn_sendfile,
1761 	.fo_seek =	vn_seek,
1762 	.fo_fill_kinfo = vn_fill_kinfo,
1763 	.fo_flags =	DFLAG_PASSABLE | DFLAG_SEEKABLE
1764 };
1765 
1766 static struct vop_vector devfs_vnodeops = {
1767 	.vop_default =		&default_vnodeops,
1768 
1769 	.vop_access =		devfs_access,
1770 	.vop_getattr =		devfs_getattr,
1771 	.vop_ioctl =		devfs_rioctl,
1772 	.vop_lookup =		devfs_lookup,
1773 	.vop_mknod =		devfs_mknod,
1774 	.vop_pathconf =		devfs_pathconf,
1775 	.vop_read =		devfs_rread,
1776 	.vop_readdir =		devfs_readdir,
1777 	.vop_readlink =		devfs_readlink,
1778 	.vop_reclaim =		devfs_reclaim,
1779 	.vop_remove =		devfs_remove,
1780 	.vop_revoke =		devfs_revoke,
1781 	.vop_setattr =		devfs_setattr,
1782 #ifdef MAC
1783 	.vop_setlabel =		devfs_setlabel,
1784 #endif
1785 	.vop_symlink =		devfs_symlink,
1786 	.vop_vptocnp =		devfs_vptocnp,
1787 };
1788 
1789 struct vop_vector devfs_specops = {
1790 	.vop_default =		&default_vnodeops,
1791 
1792 	.vop_access =		devfs_access,
1793 	.vop_bmap =		VOP_PANIC,
1794 	.vop_close =		devfs_close,
1795 	.vop_create =		VOP_PANIC,
1796 	.vop_fsync =		devfs_fsync,
1797 	.vop_getattr =		devfs_getattr,
1798 	.vop_link =		VOP_PANIC,
1799 	.vop_mkdir =		VOP_PANIC,
1800 	.vop_mknod =		VOP_PANIC,
1801 	.vop_open =		devfs_open,
1802 	.vop_pathconf =		devfs_pathconf,
1803 	.vop_poll =		dead_poll,
1804 	.vop_print =		devfs_print,
1805 	.vop_read =		dead_read,
1806 	.vop_readdir =		VOP_PANIC,
1807 	.vop_readlink =		VOP_PANIC,
1808 	.vop_reallocblks =	VOP_PANIC,
1809 	.vop_reclaim =		devfs_reclaim,
1810 	.vop_remove =		devfs_remove,
1811 	.vop_rename =		VOP_PANIC,
1812 	.vop_revoke =		devfs_revoke,
1813 	.vop_rmdir =		VOP_PANIC,
1814 	.vop_setattr =		devfs_setattr,
1815 #ifdef MAC
1816 	.vop_setlabel =		devfs_setlabel,
1817 #endif
1818 	.vop_strategy =		VOP_PANIC,
1819 	.vop_symlink =		VOP_PANIC,
1820 	.vop_vptocnp =		devfs_vptocnp,
1821 	.vop_write =		dead_write,
1822 };
1823 
1824 /*
1825  * Our calling convention to the device drivers used to be that we passed
1826  * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
1827  * flags instead since that's what open(), close() and ioctl() takes and
1828  * we don't really want vnode.h in device drivers.
1829  * We solved the source compatibility by redefining some vnode flags to
1830  * be the same as the fcntl ones and by sending down the bitwise OR of
1831  * the respective fcntl/vnode flags.  These CTASSERTS make sure nobody
1832  * pulls the rug out under this.
1833  */
1834 CTASSERT(O_NONBLOCK == IO_NDELAY);
1835 CTASSERT(O_FSYNC == IO_SYNC);
1836